Professional Documents
Culture Documents
Springer
New York
Berlin
Heidelberg
Barcelona
Budapest
Hong Kong
London
Milan
Paris
Singapore
Tokyo
Haruzo Hida
Springer
Haruzo Hida
Mathematics Department
UCLA
Box 951555
Los Angeles, California 90095
9 87 6 5 4 32 I SPIN 10950739
springeronline.com
Preface
In the early years of the 1980s, while I was visiting the Institute for Ad-
vanced Study (lAS) at Princeton as a postdoctoral member, I got a fascinating
view, studying congruence modulo a prime among elliptic modular forms, that
an automorphic L-function of a given algebraic group G should have a canon-
ical p-adic counterpart of several variables. I immediately decided to find out
the reason behind this phenomenon and to develop the theory of ordinary
p-adic automorphic forms, allocating 10 to 15 years from that point, putting
off the intended arithmetic study of Shimura varieties via L-functions and
Eisenstein series (for which I visited lAS). Although it took more than 15
years, we now know (at least conjecturally) the exact number of variables for
a given G, and it has been shown that this is a universal phenomenon valid
for holomorphic automorphic forms on Shimura varieties and also for more
general (nonholomorphic) cohomological automorphic forms on automorphic
manifolds (in a markedly different way).
When I was asked to give a series of lectures in the Automorphic Semester
in the year 2000 at the Emile Borel Center (Centre Emile Borel) at the
Poincare Institute in Paris, I chose to give an exposition of the theory of
p-adic (ordinary) families of such automorphic forms p-adic analytically de-
pending on their weights, and this book is the outgrowth of the lectures given
there.
In addition to an exposition of p-adic deformation theory of automor-
phic forms, I decided to give a comprehensive account of the construction
of Shimura varieties (carrying a canonical family of abelian varieties) and
Shimura's reciprocity laws characterizing them. Although this part is essen-
tially expository, some new results of the author are included:
1. a (partially) new proof of some reciprocity laws of Shimura varieties;
2. irreducibility of the Igusa tower over unitary Shimura varieties;
3. a construction of invertible sheaves on Hilbert modular varieties that gives
optimal p-integrality on Hilbert modular forms.
It is my hope that this book is useful to graduate students and researchers
entering this area of research.
The author would like to thank the organizers of the Automorphic Semester
at the Emile Borel Center for their invitation and also wishes to thank many
people including his former and present graduate students who helped him
to write correct English and mathematics. The author acknowledges partial
support from the National Science Foundation (through the research grant:
DMS 9988043 and DMS 0244401).
1 Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Automorphic Forms on Classical Groups. . . . . . . . . . . . . . . . . . . 5
1.2 p-Adic Interpolation of Automorphic Forms. . . . . . . . . . . . . . . . . 8
1.3 p-Adic Automorphic L-functions .......................... 12
1.4 Galois Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 13
1.5 Plan of the Book. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 13
1.6 Notation............................................... 15
3 Modular Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 67
3.1 Basics of Elliptic Curves over a Scheme .................... 67
3.1.1 Definition of Elliptic Curves ............... . . . . . . . .. 68
3.1.2 Cartier Divisors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 68
3.1.3 Picard Schemes ........ . . . . . . . . . . . . . . . . . . . . . . . . . .. 69
3.1.4 Invariant Differentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 70
3.1.5 Classification Functors. . . . . . . . . . . . . . . . . . . . . . . . . . . .. 70
3.1.6 Cartier Duality ................................... 71
3.2 Moduli of Elliptic Curves and the Igusa Tower. . . . . . . . . . . . .. 72
3.2.1 Moduli of Levell over Z[il ......................... 72
3.2.2 Moduli of Pr,(N) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.2.3 Action of G m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 75
3.2.4 Compactification.................................. 77
3.2.5 Moduli of r(N)-Level Structure. . . . . . . . . . . . . . . . . . . .. 79
3.2.6 Hasse Invariant ................................... 80
3.2.7 IgusaCurves ...................................... 81
3.2.8 Irreducibility of Igusa Curves ...... . . . . . . . . . . . . . . . .. 82
3.2.9 p-Adic Elliptic Modular Forms. . . . . . . . . . . . . . . . . . . . .. 85
3.3 p-Ordinary Elliptic Modular Forms . . . . . . . . . . . . . . . . . . . . . . .. 86
3.3.1 Axiomatic Treatment .............................. 86
3.3.2 Bounding the p-Ordinary Rank ..................... 89
3.3.3 p-Ordinary Projector .............................. 90
3.3.4 Families of p-Ordinary Modular Forms . . . . . . . . . . . . . .. 90
3.4 Elliptic A-Adic Forms and p-Adic L-functions . . . . . . . . . . . . . .. 92
3.4.1 Generality of A-Adic Forms ......................... 92
3.4.2 Some p-Adic L-Functions ........................... 94
Introduction
The two major trends in number theory, automorphic and arithmetic, were
initiated hand in hand in the mid nineteenth century by Kronecker and Kum-
mer. One of Kronecker's legacies is the theory of arithmetic elliptic modular
functions (and modular forms) having well-determined algebraic values (up
to a specific transcendental factor: "period") at special points on modular
curves. Shimura varieties (including modular curves) were principally studied
by Shimura and Deligne in the later part of the last century. In particular, the
theory gives a foundation of the rationality of automorphic L-values, because
Shimura varieties often supply enough rationality for us to be able to pinpoint
the transcendental factors of specific L-values (the automorphic periods).
On the other hand, the arithmetic side of number theory heavily influ-
enced by Kummer's work is represented by Iwasawa's theory which interprets
the algebraic part of the L-value as the size of a naturally given group (or
module) defined by purely arithmetic means. The origin of this thought goes
back to the class number formula of Dirichlet, Kummer, and Dedekind, which
gives the size of the ideal class group of a number field as a special value
of an abelian L-function (up to a power of the period 27fi of the exponen-
tial function). The main conjecture in Iwasawa's theory predicts precisely the
same effect, replacing the class group by the so-called "Selmer group" and
its "Iwasawa module" (attached to automorphic Galois representations) and
abelian L-values by automorphic p-adic L-functions (see, for example, [MFG]
Chapter 5).
Although these two trends seemed rather disparate until recently, the strik-
ing idea of Wiles (which has proven two of the most fundamental conjectures
in number theory) has spanned a bridge between the two concepts. Indeed, it
gives a way to find an automorphic form from a purely arithmetically given ob-
ject, for example, a rational elliptic curve whose L-function is directly related
to an elliptic modular Hecke L-function via the Shimura-Taniyama conjecture
(now proven by Wiles and others; see [Wi2] and [BrCDT]).
This type of intricate relation between arithmetic and geometry can also
be found in classical reciprocity laws. There are two aspects of the Artin reci-
pro city law in class field theory. One is representation theoretic, for example,
for the ring A(oo) of finite adeles,
via the identity of L-functions. Here Qab is the maximal abelian extension
of Q (so, it is a composite of all Galois extensions of Q inside C with an
abelian Galois group) and Q~ is the group of positive rational numbers (see
Theorem 2.12). Another geometric reciprocity can be formulated as
and ring-theoretic techniques are used to determine X (see [ICF] Chapter 13).
General Tannakian duality produces from the category of all (pure) mo-
tives an enormous group, the so-called motivic Galois group (see [HMS] Chap-
ters II through IV and [MTV]). However, for Iwasawa-theoretic applications,
the full motivic Galois group appears to be hopelessly large to adapt. If one
wants to create something like Iwasawa's theory in a non-abelian setting, we
therefore need a scheme whose automorphism group has an identification with
G(A(oo))/Z(Q) for a reductive algebraic group G. If G = GL(2)/IQ, the tower
1 Introduction 3
V/lQiabof modular curves has Aut(V/IQi) identified with GL 2 (A( CXJ l)/Z(Q) as
Shimura proved (see Theorem 2.43). The (geometric) decomposition group of
(p) is given by B(Qp) x SL 2 (AJp CXJ l)/{±1} for a Borel subgroup B (see Corol-
lary 4.22). We have been studying various arithmetically constructed modules
over the Hecke algebra of ~!;2;~(7)\
GL2 Z U
p zp
x, relative to the unipotent subgroup
1Qi+
U(71 p ) c B(71 p ) (removing the multiplicative inertia factor from the decom-
position group). Often these are modules given by cohomology groups (either
algebrogeometric or topological). Such study has yielded a p-adic deforma-
tion theory of p-ordinary automorphic forms (see Chapter 8), and it would
be important to study the decomposition group at p of a given Shimura vari-
ety, which is close to the automorphism group of the mod p Shimura variety.
We describe quite a few automorphism groups related to Shimura varieties of
PEL type coming from symplectic and unitary groups (following Shimura's
method and also using slightly more algebraic methods). In particular, the
study of the p-decomposition subgroup provides us with a simple proof of the
irreducibility of the Igusa tower over the mod p Shimura variety (see Sec-
tion 8.4), which supplies us with one of the most useful tools in the p-adic
study of automorphic forms: the q-expansion principle.
Besides determination of the automorphism groups of Shimura varieties,
we present a concrete construction of p-integral models of Shimura varieties
(following works of Shimura, Crothendieck, Mumford, Deligne, and Kottwitz)
and the p-adic deformation theory of p-ordinary automorphic forms, in partic-
ular, the theory of analytic families of p-ordinary (cohomological) automorphic
forms on reductive groups.
We are particularly interested in Shimura varieties that are canonically
moduli varieties of abelian schemes with extra structures. Since abelian
schemes are very specific projective subschemes of a projective space, we can
study abelian schemes through classification theory of projective sub schemes
in a given projective space. The classification was initiated by W.-L. Chow and
was completed by Crothendieck as the theory of Hilbert schemes. In the late
1950s (e.g., [59b] in [CPS] I), Shimura used the Chow coordinate associated
with each projective variety in his determination of the field of moduli of a
given family of abelian varieties (which was later incorporated as an essential
ingredient in his construction of canonical models of automorphic varieties;
see [AAF] 2.8 and Chapter II), and in [CIT] Mumford used Hilbert schemes to
study the integral moduli of abelian schemes, which can be used to construct
Shimura varieties. We give a down-to-earth exposition of the construction of
the Hilbert schemes and the Mumford integral moduli in Chapter 6.
Assuming basic knowledge of abelian schemes (which can be found in
[ABV]) , toroidal compactifications of the moduli (in [DAV]) , and class field
theory, we then give an exposition of the construction of the Shimura varieties
and their models over p-adic integer rings. The Shimura variety is character-
ized by its local and/or global recipmcity laws. The global reciprocity law gives
a geometric expression of the automorphism group of the function field of the
4 1 Introduction
Shimura variety and essentially determines the variety, because the Shimura
variety is often a minimal model of the function field. The local reciprocity
law determines via class field theory the move of special points, called CM
points, under Galois action. The CM points are those points carrying abelian
varieties with complex multiplication (also often called abelian varieties of
CM type) and are Zariski dense in the variety, thereby providing rationality of
the variety. An abelian variety has complex multiplication if its endomorphism
algebra contains a semi-simple commutative algebra of degree over Ql twice
the dimension of the variety (see [ACM] 5.2 and [ABV] Section 19). The first
principal purpose of this book is to supply the base of the construction of
the (p-integral) Shimura variety in as elementary a manner as possible (in-
cluding an exposition of the local and global reciprocity laws), because it is a
pity that no other books (so far published) give a detailed scheme-theoretic
construction of the variety now so important in every aspect of algebraic and
geometric number theory. Shimura's books [AAF] and [EPE] give an elemen-
tary exposition on rationality of unitary and Siegel Shimura varieties and of
automorphic forms and L-functions on them and will be indispensable for
graduate students and researchers who want to enter this area of research.
After going through our exposition of Shimura varieties and p-adic auto-
morphic forms on them, natural subjects of our exposition are
1. p-adic deformation theory of p-ordinary automorphic forms;
2. p-adic automorphic L-functions;
3. p-adic Galois representations;
4. the Iwasawa-theoretic significance of the p-adic L-functions.
In this book, we mainly touch upon the first topic with some examples of
p-adic L-functions, leaving other topics to future treatments. Thus the second
purpose of this book is to introduce integrality of automorphic forms on the
varieties and hence to supply a foundation in order for researchers to pursue
p-adic analytic study of automorphic forms and automorphic L-values.
Here is a brief history of p-adic modular and automorphic forms. The the-
ory of p-adic elliptic modular forms was initiated by Serre in [Se3] taking the
p-adic limit of q-expansions of elliptic modular forms. The geometric signif-
icance of Serre's definition was revealed by Deligne and Katz (e.g., see [K],
[Kl], and [K3]). Serre observed that the Eisenstein series of a given level Np
has Fourier coefficients p-adically analytically depending on its weight k, and
through this observation, a construction of p-adic zeta functions of totally
real fields was given in [Se3], which was later brought to completion in [DR]
through a more geometric formulation. A systematic study of elliptic cuspi-
dal families of Hecke eigenforms p-adic analytically depending on weight was
started by the author of this book in [H86a] and [H86b] and has been devel-
oped into a p-adic deformation theory of p-ordinary automorphic forms on
Shimura varieties and classical groups (e.g., [H02], [H98], and [TiU]). More
recently, Mazur, Gouvea, and Coleman (e.g., [GM], [Co], and [Wa]) extended
the theory to non-p-ordinary elliptic modular forms with positive slope. In par-
1.1 Automorphic Forms on Classical Groups 5
ticular, Mazur and Gouvea conjectured, and Coleman has proven, that there
always exists a p-adic analytic family of elliptic modular forms containing a
given modular form with a positive slope, although only p-ordinary families
seem to have a precise relation to classical objects (typical in Iwasawa's the-
ory; see (3) in Section 1.2), because only for p-ordinary automorphic forms
is p-adic analyticity of their coefficients described by Iwasawa's functions be-
longing to the Iwasawa algebra (and its finite flat extensions). So p-ordinary
automorphic forms enjoy stronger p-adic analyticity than nonordinary forms.
The deformation theory has found numerous profound applications by many
different researchers. For example, it was used as an essential ingredient of a
construction of two-dimensional Galois representations from Hilbert modular
forms [Wi], of a proof of Iwasawa's conjecture in [Wil], of the proof of Fermat's
last theorem ([Wi2 ], [TaW], and see also [MFG] Chapter 3), of the proof of
the Shimura-Taniyama conjecture ([Wi2], [BrCDT] and see also [GME] Chap-
ter 5), and of the proof of the Artin conjecture for many nonsoluble icosahedral
Galois representations (e.g., [BuT] and [BuDST]). More recently, the linear
growth (relative to the degree of the base totally real fields) of the number
of variables of the p-ordinary family of Hilbert modular forms was used by
Skinner and Wiles ([SkW]) to prove modularity of residually reducible (but
irreducible) two-dimensional p-adic Galois representations.
Since our second principal goal is to give an introductory exposition of p-
adic automorphic forms, in the following sections of this introductory chapter,
we give an overview of such automorphic forms (covering all four topics),
putting off to the following chapter an introductory discussion on reciprocity
laws (including a brief history).
( c~s(e)
sm(e)
- sin(e)) =
cos(e)
(*cd*) H (* *) = (ci + d)k.
P cd
In this case, Jp(g, z) = (cz + d)k. This goes as follows. Split GL 2 (lR) = PCo
for P made of upper triangular matrices with 1 at the lower right corner. For
z = x + iy, define pz = (6 ~). Then for 9 E SL 2 (lR), write gpz = Pg(z)u with
u E Co, and we have p(u) = p(p;(~)gpz) = (cz + d)k by computation. Indeed,
Jp(g, z) sends (v,Pz) to (uv,Pg(z)) ~ (v, gpz) = (v,Pg(z)u).
One can view the complexification C as a real algebraic group; let Te be
a maximal real torus of C. With any character K of Te, we can associate a
rational representation p = p~ : C H GL(N) of C with representation space
Lc(K; C) of dimension N. Let V(K) = Lc(K; C). For a congruence discrete
subgroup
r c G(Q), a holomorphic automorphic form of (coherent) weight K is a
holomorphic function J : H H V(K) satisfying J([(z)) = J p ([, z)J(z) for all
'Y E r (with some additional growth condition if r\ H is not compact). Again
the space of holomorphic automorphic forms is trivial unless the weight K is
dominant (with respect to B chosen appropriately).
Often the complex manifold r\ H is canonically algebraizable, yielding
an algebraic variety (or a scheme) X r , called canonical models or Shimura
vaT'ieties, defined over a valuation ring W in a number field with residual
characteristic p. At the same time, we can algebraize the vector bundle V(K)
associated wit~ V(K). Thus we can often have a coherent sheaf i±JK on Xr
giving rise to V(K) after extending the scalar to C. The global sections of
HO(Xr,i±JKR) for a W-algebra R are called R-integral automorphic forms of
weight K. Note that Te is isomorphic to T, because they are maximal tori in
the same group G 1 . Thus we identify T and Te (with a compatible choice of
Borel subgroups B and Be = B n C). On X r , we may regard the r-module
LC(K; R) as a locally constant sheaf associating, with an open subset U c X r ,
sections over U of the covering space L(K; R) = r\(H x LC(K; R)) over X r .
Here the quotient r\(H x LC(K; R)) is taken through the diagonal action.
Thus each dominant weight K E X(T) gives two spaces of automorphic forms:
HO(Xr,i±JK) Y Hd(Xr,Lc(K*;C))
by a generalized "Eichler-Shimura map" which is supposed to be equivari-
ant under Hecke operators. If such equivariance holds, we say that the two
modules, the source and the image, are equivalent as Heeke modules. In the
example of GL(2)/Q, we have r c SL 2(71), K E X(T) = X(CG m ) = 7l, and
K* = K - 2 with
8 1 Introduction
when the weights r;, vary continuously in the p-adic analytic group D =
Homtop_gp(T(Zp), Q;) (fixing an algebraic closure Qp of Qp). Here to describe
our idea of how to interpolate automorphic forms, we write W for the p-adic
completion of W.
On these two families of spaces, there is a natural action of Hecke op-
erators; so, we also want this interpolation to take into account the Hecke
operators. To carry out the p-adic interpolation, we need a good p-integral
structure on the group G; so, we hereafter assume that G is defined over
Z(p) = Q n Zp as an affine group scheme and that G 1 is isomorphic over Zp to
the restriction of the scalar to Zp of a classical Chevalley group over the p-adic
integer ring Op of an unramified finite semi-simple extension of Qp; so, the
Borel subgroup B and its maximal torus T is well defined over Z(p), and T is
split over Op. Thus, if Op = Zp, T(Zp) ~ (Z;r for the rank r of Glover Qp,
and D is (isomorphic to) a disjoint union of finitely many copies of the p-adic
(open) unit disk. In general, T(Zp) is a product of a torsion-free multiplicative
group rT isomorphic to the additive group Z; and a finite group. Then the
Iwasawa algebra W[[T(Zp)]] (which is the projective limit of the group alge-
bras {W[T(Z/pmZ)]}m) is a free module of finite rank over W[[rT]], which is
in turn (noncanonically) isomorphic to the power series ring W[[T1 , ..• , Tn]]
of n variables. What we now do is:
(1) (Universality) Construct a (big) compact module V over W[[T(Zp)]]
such that the r;,-eigenspace V[r;,] contains canonically, as W[[T(Zp)]]-
submodules, Hd(Xr, L(r;,; W)) in the topological case, and HO(Xr / w , ~K)
in the coherent case, respectively;
(2) (Hecke operators) Establish a natural action of Hecke operators on V, and
show the inclusion in (1) is Hecke equivariant;
(3) (Precise control) Find an appropriate W[[T(Zp)]]-submodule X c V of
cofinite type ({o} W -dual is of finite type) such that X is stable under
Hecke operators and X[r;,] (if r;, » 0) is canonically isomorphic to a well-
described subspace of automorphic forms of weight r;, as Hecke modules.
1.2 p-Adic Interpolation of Automorphic Forms 9
L
00
We can deduce from the irreducibility of the Igusa tower that V is the p-adic
completion of
L i().,(T(n))(uK. -
00
writing i = ioo 0 i; 1.
In general, for sufficiently positive weight II: of T, we have Hecke eigenvalues
).,(t)(II:) occurring on the coherent cohomology with coefficients in '<d..K. on the
Shimura variety. We thus have a family {).,( II:)} K. of systems of eigenvalues
(or equivalently, a family of holomorphic automorphic representations). We
give ourselves a rational representation r of the Langlands dual LG of G
and consider the family of automorphic L-functions {L(s, ).,(11:), r)}K.. Suppose
that L(m, ).,(11:), r) for (sufficiently many) integers m (depending on 11:) has
rationality up to a transcendental factor or a period D(m, 11:, r). Then natural
questions we would ask are:
(Q2) Is it possible to interpolate p-adically the value Lh(~,~~~)) ? Is it pos-
sible to find L)..,r(U) E lI[[U]] such that the p-adically analytic function
Galois deformation rings (with a specific universal property; see [FI], [SkW],
and [MFG]).
Some of the papers and preprints of mine related to these subjects can be
downloaded from my web site: www.math.ucla.edu{hida.
1.6 Notation
In this book, we continue to use the notation utilized in this introduction. Here
we describe some of the notation which is used without much explanation. The
symbol Zp denotes the p-adic integer ring inside Qp, and the symbol Z(p) is
used to indicate the valuation ring Zp n Q. A sub field E of ij is called a
number field (often assuming dimlQl E < 00 tacitly). For a number field E,
DE denotes the integer ring of E, DE,p = DE 0z Zp c Ep = E 01Ql Qp and
DE,(p) = DE 0z Z(p) c E. The symbol W is exclusively used to indicate a
valuation ring inside ij with residual characteristic p, and often W is supposed
to be unramified over Z(p). The p-adic completion ~ n W /pnW is denoted by
W, and we write Wm = W/pmw = W/pmw.
The symbol A is the adele ring of Q, and for a vector space of a number
field E, we write VA = V(A) for V 01Ql A. For a subset E of the set of rational
places, we set A(E) = {x E Alx v = 0 for VEE} and ZE = TIpEEZp, where p
runs over finite places in E. We then define Z(E) = ZE n Q. If E = {oo}, {p},
or {p, oo} for a prime p, we write A (00), A (p) or A (poo) for A (E) accordingly. If
E is a set of finite places, we write A(Eoo) for A(EU{oo}). We identify A(E)X
with {x E A x Ixv = 1 for vEE} in an obvious way. The maximal compact
subring of A(oo) is denoted by Z, which is identified with the profinite ring
c
TIp Zp = ~ NZ/NZ. For any Z-module c of finite type, we write for c 0z Z.
An algebraic group T (defined over a subring R of ij) is called a torus
if its scalar extension T jij = T 0R ij is isomorphic to a product G~
of copies of the multiplicative group G m . The character group X*(T) =
HOmalg_gp(Tjij, G mjij ) is simply denoted by X(T), and elements of X(T) are
often called weights of T.
The symbol G is exclusively used for a reductive linear algebraic group
defined over Q. Then G 1 is its derived group of G (as an algebraic group
defined over Q), and Z is the center of G. We use the symbol PG = Gad for
G/Z as an algebraic group over Q. For the real points G(JR) of G, we regard
it as a Lie group, and G(JR)+ is the identity-connected component of the
Lie group G(JR). The symbol B often indicates a standard (upper triangular)
Borel subgroup inside G, and the letters U and N are used to indicate the
unipotent radical of B. The letter U is used when N is used to indicate an
integer (and N is used when U is used to denote a unitary group). The torus
B/U is denoted by Te, and the maximal torus of G 1 induced by Te is denoted
by T.
2
Since reciprocity laws play an important role in this book, we now describe
several reciprocity laws appearing in number theory in terms of automorphism
groups of a field fixing a given prime or a given point. Only this chapter
contains (elementary) exercises.
To fix our idea, we describe here briefly a model reciprocity law. For a
given field ji, we consider the field automorphism group Aut(ji) equipped with
the Krull topology (which is described in Section 2.3). As is clear from the
introduction, a geometric global reciprocity law gives a canonical description
of Aut(ji) by the adelic points of an algebraic (reductive) group G/iQ (modulo
rational center Z(Ql)): Aut(ji) '---+ G~~») typically. If a local ring V with
quotient field ji is given, we could define a decomposition subgroup D of
Aut(ji) (equipped with the Krull topology) by
D = {O' E Aut(ji) IO'(V) = V} ,
and the local reciprocity law for V describes D via a well-defined algebraic
subgroup H of G. Often D is given by the image of adele points, H(A(oo)) (or
Qlp-points, H(Qlp)), of H in G~~»). This description may involve a subtle use
of class field theory, and class field theory supplies us with one of the simplest
examples of the reciprocity laws. The global reciprocity law is a particular
case of the local one (taking V = ji). To have such reciprocity laws, the field
ji cannot be finitely generated over its prime subfield except for the case where
G is finite. So for an algebraic group G of positive dimension, the reciprocity
laws get more geometric flavor, and ji is then realized as the function field of
an infinite tower V of (geometrically irreducible) algebraic varieties defined
over a tower of finite extensions over the prime field. Then if we have local
reciprocity laws covering sufficiently many points of the tower V, the laws
characterize a global model V of ji (a model of a field K is an algebraic variety
whose function field is isomorphic to K in a canonical way). Thus to have a
coherent system of local reciprocity laws is almost equivalent to having a well-
defined global model of a given function field (of infinite type). Similarly, if V
In this section, we sketch the classical reciprocity laws starting with the his-
toric quadratic reciprocity invented by Euler (and proved by Gauss), and
ending with a reciprocity law for a single rational elliptic curve. Here, primes
p and q are always distinct odd primes.
Since ::z:2 == n mod p has a solution if and only if n E (IF;)2 for lFp = ZjpZ,
n H (~) gives an identification of IF; j(lF;)2 with {±1}; in particular, n H
where q is the kernel of the projection of R onto the first factor IFq of IF q EB IFq.
The nontrivial automorphism a ofQ[v'P*) interchanges the roots of X 2 _p*
and interchanges q and q'; so, q' = a(q). Identifying {±1} with Aut(it) =
Gal(Q[v'P*)/Q), (z;i)
gives the generator of the subgroup of Aut(it) fixing a
prime factor of (q).
If (q) = q remains prime, dimlFq R/(q) = 2, and a(q) = q. The stabilizer of
q is the entire Galois group, and (( p;)) = {±1} ~ Aut(it) gives the stabilizer.
In summary, identifying Aut(it) with {±1}, the subgroup generated by (z;i)
gives the stabilizer of a prime ideal qlq in R (in other words, the stabilizer of
the q-adic valuation ring V of it). The stabilizer is called the decomposition
subgroup of q. The information of the decomposition group of q is equivalent
to knowing how the prime (q) splits in R.
element in Aut(lLp) (and [Q[lLp] : Q] = IlF; I), we see that Gal(Q[lLpl/Q) ~ IF;
by XP' We write ¢q E Gal(Q[lLp]/Q) with Xp(¢q) = q; so, ¢q(() = (q.
Again we ask how a prime (q) decomposes in the ring R = Z[lLp], Pick a
prime ideal qlq in R; we find that R/q is a finite extension of lFq; so, it is of the
form lFqf for f = dimlFq R/q. The Galois group Gal(lFqf /lFq) is a cyclic group
of order f generated by a canonical generator F taking x E IF qf to x q E IF qf ;
thus, the automorphism F of lFqf is induced by ¢q. In other words,
• The decomposition group of qlq is given by (¢q) ~ (q) elF;, and
• q is of order f in IF; {=} (q) = qa(q)··· a 9 - 1 (q) in R for the integer
g = [Q[lLp] : Ql/ f,
So, roughly, the complex analytic function exp : C ---+ C x contains all the
information of the reciprocity law, and exp(2ni;;) gives a canonical generator
( of the field Q[lLp] and behaves nicely under its Galois conjugation.
The function exp gives rise to the following exact sequence,
and thus we evaluated the function exp at the fraction 2ni;; E ;;2niZ/2niZ.
The period 2ni of the exponential function "exp" is important to ensure that
the value of "exp" on 2niQ is algebraic, and we have 2ni = I, flt
for the unit
circle 'Y which generates the fundamental group nl(C X ) = Hl(CX,Z),
Hilbert asked in his twelfth problem (of his famous lecture in 1900 at
the International Congress of Mathematicians held in Paris; [HI]), for a given
Galois extension (actually an abelian extension in his original setting),
Is there any complex analytic function which describes fully the reciprocity law
of the extension?
2.1 Sketch of Classical Reciprocity Laws 21
7fl(E(C)) = H 1 (E(C),Z) = a,
induced by du for the variable u of C, we have [2 = J'Y w for the generator, of
just as before. Indeed, for an lR-base (Wl,W2)
of C, we can think of EL(C) = CI L for L = ZWI + ZW2 replacing a by L.
Weierstrass studied analysis and geometry of Riemann surfaces (trans-
forming Riemann's marvelous but rather intuitive ideas into a rigorous math-
ematics). In particular, for the Riemann surface E(C) of genus 1, he created
the following function well-defined over E(C),
x(u) = P(u) = u 2 +
1 L (1 1)
(u _ £)2 - £2 '
£EL-{O}
7fcot(7fz) 1 {II}
= - +L
Z
-- + -- .
00
z+n z-n
n=1
1 g2 2 g3 4 2 '"""'
x(u)=P(u)=u2+20u +2S u + ... , I
y(u)=P(u)=-u3+~anun,
n::::l
where g2 = g2(L) = 60 LNO £-4 and g3 = g3(L) = 140L£;",0£-6. The con-
stants g2 and g3 are actually complex analytic functions of W = (WI, W2).
Canceling the poles, we consider 'P = y2 - 4x 3 + g2X + g3 which is holo-
morphic everywhere on a compact Riemann surface E(C); so, it has to be
constant. The function 'P has to be identically 0, because 'P has no constant
term. Thus u r-+ E(u) = (u 3x(u), u 3y(u), u 3) E p2 embeds the Riemann
surface into the projective space of dimension 2, whose image is an alge-
braic curve (called an elliptic curve) defined by the homogeneous equation:
22 2 Geometric Reciprocity Laws
For almost all prime ideals IiJ of Q(j (a)), 92 and 93 are liJ-integral with L1( a) t=-
O mod 1iJ, and the equation modulo IiJ gives rise to an elliptic curve Ea defined
over a finite field IF. The relative Frobenius map F : (x, y) H (x P, yP) sends
Ea to another elliptic curve E~p) defined by y2 = 4x 3 - 9gx - 9~ E IF[x, y].
Suppose that (p) = IiJ n Z is split into pp in the imaginary quadratic field M.
Then the group of p-torsion points EalP] is the direct sum of the group of p-
torsion points and that ofp: EalP] = Ea[P]EBEa[P]. As is well known (cf. [GME]
2.9.1), we can choose p uniquely so that Ea[P] ~ R/p having p distinct points.
Since F : Ea --+ E~p) is a homomorphism of groups that is a zero map on
the tangent space, it induces a group isomorphism EaOF) ~ E~p\W) (though
not a scheme isomorphism), and hence E~p) ~ Ep-l a, because p : Ea --+ Ea
is an endomorphism (see [GME] 2.9.1) which factors Ea ~ E~p) ~ Ea with
Ker(V) = E~p)[p] = p-Ia/a. This shows that j(a)<Pp = j(p-1a). Since j(a)
characterizes the isomorphism class of Ea (over Q), we know that the j(a)s
indexed by a E elM for the ideal class group elM of M are all distinct. Since
in each class of elM, we can find split p outside a given finite set of primes
of M (the Chebotarev density), we find that {j(a)}aECI M are all conjugates
of each other over M; so, they span the Hilbert class field H/M. This proves
the above theorem (and finiteness of elM).
By the theorem (or its proof), Gal(H/M) ~ elM. This is the explicit class
field theory for the imaginary quadratic field M. This type of result is gener-
alized by Shimura, Taniyama, and Weil to eM fields, using a quotient Cd / L
for a lattice L of higher rank (theory of abelian varieties with complex multi-
plication; see Theorem 4.19 in the text, [IAT] Chapter 5, and [ACM]). A CM
field means a totally imaginary quadratic extension of a totally real field.
24 2 Geometric Reciprocity Laws
Proof. Let f.1N be the group of Nth roots of unity. Then e induces an isomor-
phism L : 7L/ N7L ~ f.1N of groups by m H e (N). Since 0"( (N) is another prim-
itive Nth root for any 0" E Gal(ij/Q) (Q is the algebraic closure of Q in C), 0"
induces an automorphism of the field Q[(N] because 0"( (N) is again a power of
(N. Thus Q[(N]/Q is a Galois extension. Since Aut(7L/N7L) = (7L/N7LV (mul-
tiplicative group) by ¢ H ¢(1), we find X = XN : Gal(Q[(N]/Q) -+ (7L/N7L) X •
If 0" E Ker(x), 0" fixes all Nth roots of unity; so, X is injective. Its p-component
Xpn has to be surjective, since
are the only ideals of V I(p), because (aJXJ + aJ+IXJ+1 + ... + ap_lXp-1) =
(X)J if aJ Ie 0 in IFp. Since p = tli(O) = TIaEGal(IQ[(pl/lQ) cr(w), we find that
(w) ~ (p). Then by the homomorphism theorem applied to V --» V I (p), only
ideals between (p) and (w) are (w)m for m = 1,2, ... ,p - 1. By induction on
n, one can show that
are the only ideals of V I (pn). For a given ideal a C V, a n Z(p) = (pC) because
Z(p) is a discrete valuation ring (DVR). Thus a = (w)m for O:S; m:S; jl(p-1).
This shows that V is a DVR with valuation w given by w(x) = n ~ (x) =
(w)n. The same argument as above works well for V = Z(p) [X]j(Ppn (X)) for
w = (pn - 1, since Ppn(X + 1) == Xpn-l(p-l) mod p. D
0= v(l) = v( -alx-1 - a2x- 2 - ... - anx- n ) :2: min(v(aJ ) + jv(x- 1)) > 0,
J
Let q be a prime different from p. We now look into Vq = Z(q) [X]I (Ppn (X)) =
Z(q)[(pn]. We look into <J5pn(X) = (Ppn(X) mod q) E IFq[X]. Then VqlqVq =
IFq[X]/(<J5pn(X)) and dimlFq VqlqVq = rankz(q) Vq = deg(ppn(X)). Since we
have pn-l (p - 1) distinct primitive pnth roots of unity in IFq, <J5pn (X) does not
have multiple roots, and we see Vq I qVq ~ IF qh EB IF qh EB ... EB IF qig . The image
of X in IF qIJ is a primitive pnth root a = a) of unity in IF qIJ , and IF qIJ = IF q[a].
Thus f = h = 12 = ... = f9 is the minimal exponent so that IF qI contains a
primitive pnth root of unity. Since IF qxI is made up of (q! -l)th roots of unity,
f is the minimal exponent such that pnlq! - 1. The ideal q) = Ker(7r)) for
each projection 7r) : Vq --» VqlqVq --» IFqIJ is a prime of Vq, and we have
II q)
9
(q) = and fg = [VqlqVq : IFq] = deg(ppn) =pn-l(p_l).
)=1
We have proven:
Theorem 2.6 (Cyclotomic Reciprocity) If q # p is a prime, then for
each prime factor q in Z(q) [(pn L the decomposition subgroup of q in G =
Gal(Ql[(pn]/Ql): Dq = {O' E GIO'(q) = q} is given by the cyclic subgroup (cPq)
generated by the Frobenius element cPq with Xpn (cPq) = q. In particular, (q) =
TIcrEG/D q O'(q) and Vq/q "" lFqf for the order f of q in (Z/pnzy.
n =n
Vq
q
Vq = (nqZ(q))l + (nqz(q))( + ... + (nqz(q))(pn-l(p-l)-l = Z[(pn],
where q runs over all prime ideals of 0 = Z[(pn] and q runs over all rational
primes. This shows that Z[(pn] is the integer ring of Ql[(pn]; in other words,
Z[(pn] is the integral closure of Z in Ql[(pn]. Thus Oq = Vq for a prime ideal q
of 0, and we can restate the reciprocity law as follows:
Corollary 2.7 The integer ring 0 of Ql[(pn ] is generated by (pn over Z. For a
prime q # p and a prime factor qIq, the decomposition group D q is independent
28 2 Geometric Reciprocity Laws
of the choice of q and generated by the Probenius element ¢q with Xpn (¢q) =
q. We have (q) = Il:rEG/D q tT(q) and G/q = lFqJ for the order f of q in
(71., / pn 71.,) x. The prime p is totally ramified in G, and the prime factor p of p is
principal and is generated by (pn - 1. In particular, Ip = Dp = Gal(Q[(pn ]/Q)
for the inertia subgroup Ip C Dp.
Recall that p and q are distinct primes, and put N = pmqn. Then we have a
commutative diagram
XN 1 (2.1)
(71.,/N71.,) X ------------+) (71.,/ M71.,) x .
(x mod N)t-+(x mod M)
Thus the composite (inside ij) of Q[(N] for all positive integers N is actually
a union QCYc = UN Q[(N].
Taking the projective limit, we have
(2.2)
We compute Z= ~ N (71.,/ N71.,). We recall that the p-adic integer ring 71.,p =
~n(71.,/pn71.,) is formed by the formal series Ln>o cnpn with 0 ~ C n ~ P - 1,
2.2 Cyclotomic Reciprocity Laws and Adeles 29
which is the valuation ring (and is the completion of Z(p) with respect to the
norm Ixlp = p-v(x)). Since any integer can be expanded as above by p-adic
expansion, Z C Zp, and this inclusion is compatible with Z(p) C Zp. Writing
the prime decomposition of N as N = I1 pl N pe(p), we find by the Chinese
remainder theorem
where p runs over all prime numbers in the last two products. We put the
p-adic topology on Zp so that a system of neighborhoods of x E Zp is given
by {x + pnZp}n::::O. This makes Zp (resp. Z;) a compact profinite ring (resp.
group). We equip Z and ZX with the product topology of each component
(Zp and Z;). Then Z (resp. ZX) is a compact profinite ring (resp. group).
We now want to prove: Gal(Qcyc/Q) ~ AX /QxJR.~ = GLl(A(oo))/Q~ for
the adele ring A using the above expression of Gal(Qcyc /Q). The adele ring A
is a rather complicated ring, but we find it very useful later, and it contains
all arithmetic information of the field of rational numbers Q.
Let us recall the definition of A. We consider the product ring JR. x I1p Qp,
where p runs over all positive prime numbers, and Qp is the field of fractions
of Zp, and Q C Qp, which is the completion of Q under the p-adic norm I Ip
as above. Each p-adic number x E Qp has an expansion Ln»-oo cnpn (en = 0
if n is very negative). Since Z; is made up of L~=o cnpn with p t Co, Qp =
Un pnz; U {O}, where n runs over all integers. Regard Q inside JR. x I1p Qp
p
by sending ~ E Q to (~, C·· ,~, ... ) E JR. x I1p Qp. Let A be the subring of
JR. x I1p Qp generated by Q, Z, and R We often write x = (x oo , ... , x p , .•• )
for an element x E JR. x I1p Qp; so, Xoo E R
Proposition 2.9 We have
and A = (JR. x Z) + Q. Here "almost all p" means "except for finitely many
p."
Proof. The ring A' is a subring of JR. x I1p Qp- For ~ = ~ E Q (a, b E Z), if p t b,
we have ~ E Zp. Thus ~ E A', and A C A'. If x E A', we expand xp = Ln cnpn,
and define [xp] = Ln<o cnpn, which is a fraction with a p-power denominator
and is called the fraction p-part of xp. Then [xp] = 0 for almost all p, and
[x] = Lp[Xp] E Q. Then we look into xp - [x] = xp - [xp] - L#p[Xq] E Qp.
Since the denominator of [Xq] is prime to p, we find [Xq] E Zp. Thus xp - [x] =
xp - [xp] - Lqh[xq] E Zp, and hence x - [x] E Z inside A'. In particular,
x = (x - [xl) + [x] E (JR. x Z) + Q c A. This shows the last identity. 0
30 2 Geometric Reciprocity Laws
Writing the finite part A(oo) = A n TIp Qp, we thus have A(oo) = Z 0z Q.
malized so that it takes p E Qp c A to p(poo) = (p,p, ... ,p, l,p, ... ,p) E Zx.
Proof. We see that ~ E QX n (ZX x ~~) means that ~ = ~oo > 0 and the
numerator and the denominator of ~ = ~p is prime to p for all p; so, ~ = l.
Thus we conclude QX n (ZX x ~~) = {I}. By the isomorphism theorem:
and the map [(x) is called the Artin reciprocity map and sometimes written
as [(x) = [x, Q]. The reciprocity map of the class field theory induces a lo-
cally compact (actually compact profinite) topology on the Galois group. This
topology can be defined in a way intrinsic to Galois' theory, as we show in the
following section.
2.3 A Generalization of Galois Theory 31
Exercises
For our later use, we gather here some results from a generalization of Galois'
theory given in [IAT] 6.3.
We study the Galois theory when dimK L is infinite for a Galois extension
L/ K. First we recall Galois' fundamental theorem when L/ K is finite.
Theorem 2.13 (E. Galois) Suppose that L / K is a finite Galois extension.
Taking an intermediate field L / M / K, L / M is a Galois extension and
(1) There is one- to- one onto correspondence
finite degree with coefficients in K, there are only finitely many conjugates of
~. Thus the Galois closure K[~]gal of K[~] over K is a finite Galois extension
of K, and we get from Theorem 2.13 (2),
Proof. If M / K is a finite extension, for any finite Galois extension L' with
L :J L' :J M, we can extend (J E Gal(M/K) to T E Gal(L'/K) (assertion
(3) of Theorem 2.13). Since L is a union of finite Galois extensions of L',
(J E Gal(L' / K) extends to T E Gal(L/ K). If M / K is an infinite extension,
writing M as a union of finite Galois extensions M' / K and applying the
above fact to M' / K, we find that any (J E Gal( M / K) can be extended to
T E Gal(L/ K). Thus (J r-)- (JIM is a surjective homomorphism of Gal(L/ K)
onto Gal(M/K). Its kernel is given by Gal(L/M) which is normal. 0
Basically the same proof as above yields the following result which looks
stronger than the lemma.
Corollary 2.15 Let M be an intermediate extension of L/ K. Then (J r-)- (JIM
induces an isomorphism Gal(L/K)/Gal(L/M) ~ HomK(M, L), where the set
HomK (M, L) is the collection of K -linear field homomorphisms of Minto L.
The proof of this corollary is left to the reader as an exercise (Exercise 1).
We now give a topology on the group Gal(L/ K) in the following way.
Define a fundamental system of open neighborhoods of the identity by the
collection of subgroups of the form Gal(L/M) for a finite extension M/ K. If
H = Gal(L/M) for a finite extension M/K, the Galois closure Mgal is still a
finite Galois extension of K; so, we find a normal open subgroup Gal(L/Mgal)
inside any open subgroup H. Thus we may define the system to be the set
U of all normal subgroups N = Gal( L / M) for finite Galois extensions M / K
inside L. For each (J. E Gal(L/ K), we define the system of neighborhoods of
(J to be (JU = tiC! = {(IN = N(JIN E U}. By this, if II satisfies the axiom of a
fundamental system of open neighborhoods of a point, G becomes a topolog-
ical group, that is, G is a group with a topology for which the multiplication
(a, b) r-)- ab and inverse a r-)- a- 1 are continuous. The axiom is checked by
Lemma 2.16 We have
(1) For (J i- 1, there exists N E U with (J (j. N (so, G is a Hausdorff group);
2.3 A Generalization of Galois Theory 33
(2) N, H E 11 =} N nH E 11.
(1) Every open subgroup of Gal(L/ K) is closed, and open subgroups corre-
spond to finite extensions LH / K.
(2) Let {Nt hEJ for an index set I be a collection of closed subgroups of the
Galois group Gal(L/K). Writing Nt = Gal(L/Kt ), for the composite M
n
of all {KdtEJ, we have tEJ Nt = Gal(L/M).
(3) If {KthEJ is a collection of intermediate fields of L/ K, then the group
Gal(L/ ntEJ K t ) is the closure of the subgroup in Gal(L/ K) generated by
Gal(L/ K t ) for all i.
(4) For 0- E Gal(L/K), we have Gal(L/o-(M)) = 0-. Gal(L/M)o--l.
(5) If M/K is a Galois extension inside L, then Gal(L/M) is a normal closed
subgroup and Gal(L/K)/Gal(L/M) ~ Gal(M/K) induced by 0- f-t o-IM is
an isomorphism of topological groups.
Proof. Let 9J1 be the set of all intermediate fields of L/ K and Sj be the set
of all closed subgroups of Gal(L/ K). For any M E 9J1, Gal(L/M) is a closed
subgroup of Gal(L/K) by Corollary 2.18; so, M f-t Gal(L/M) defines a map
9J1 -t Sj. We define a reverse map Sj -t 9J1 by H f-t LH.
We first prove that Gal(L/ LH) = H. Since Gal(L/ LH) C Gal(L/ K)
is the collection of all automorphisms of L fixing LH, we confirm that
H C Gal( L / LH). Conversely, for each finite Galois extension M / LH inside
L, H acts on M nontrivially if M -=J LH, and the image under 0- f-t o-IM
in Gal(M/L H ) is a subgroup H' of Gal(M/L H ) and LH = MH'. By Theo-
rem 2.13 (1), H' = Gal(M/ L H ). In other words, for each 0- E Gal(L/ LH),
we find hM E H such that hMIM = o-IM. We consider the infinite set
E = {h M } M C H with M running through all finite Galois extensions M / LH.
Since H is a closed subgroup of a compact group Gal(L/ K), H itself is com-
pact. Then E has a unique accumulation point h. Then for each finite Galois
extension M/LH, o-IM = hiM. Since L = UMM for finite Galois extensions
of M/LH, we find that h = 0- and H = Gal(L/LH).
We now prove LGal(L/M) = M. By Corollary 2.18, H = Gal(L/M) E Sj.
By definition, M C LH. Supposing that LH -=J M, we try to get a con-
tradiction. Pick ~ E LH - M. Then M[~l/M is a finite extension. Thus
M[~lgal C L because L/K is normal. Since ~ tf. M, Gal(M[~lgal/M) -=J {I}.
Pick 0- E Gal(M[~lgal /M) with o-(~) -=J ~. Then by Lemma 2.14, we find
7 E Gal(L/M) = H such that 7(0 = o-(~) -=J ~. This is wrong since o-(~)
has to be ~ because ~ E LH. Thus we find that LGa1(L/M) = M, and the
correspondence is one-to-one and onto.
For an open subgroup H, Gal(L/ K)/ H is discrete and compact; so,
it is finite (see Exercise 4). Since H is the kernel of the continuous map
Gal(L/ K) --;; Gal(L/ K)/ H, H is closed. By H = Gal(L/ LH), we see from
Lemma 2.14 and Corollary 2.15 that Gal(L/K)/H = HomK(LH,L), and
hence LH / K is a finite extension by Corollary 2.15. This shows (1).
Assertion (2) follows from the fact: 0- E Gal(L/ K) is the identity on every
K t if and only if 0- is the identity on the composite of all Ki (Exercise 2).
2.3 A Gellerali;oation of Galois Theory 35
Let Gal(L/ n'EI K,) = H'. Then by definition, H' is a closed subgroup
containing all H, = Gal(L/ K,), and it contains the closure H of the subgroup
generated by H, for all i. Since H is a closed subgroup (so in Sj) containing H"
we see LH C K, and hence LIi c n'EI K,. Thus H ~ H' = Gal(L/ n'EI Ki).
This shows H = H'.
Assertion (4) is plain from the definition of the Galois group, and the last
assertion follows from Lemma 2.14. 0
Proof. Pick a compact subgroup H of Aut(L/ K). Then for each ~ E L, the
topological group Aut(L/K(O) is an open subgroup of Aut(L/K). Thus H' =
H n Gal(L/K(~)) is an open subgroup of H. In particular, H/H' is compact
and discrete; so, it is finite (Exercise 4). Since f(X) = TIhEH/W(X -h(O) has
coefficients in L H, ~ is algebraic over LH. Since h(~) = h/(O <=? hH' = hIH',
we find that the roots of f(X) are all distinct; so LH (0/ LH is a separable
extension. Since h(O E L for all h E H, all the roots of f(X) are in L. The
Galois closure M of LH (0 over LH is a Galois extension of LH inside L.
Since L = U~EL LH (0, we find that L/ LH is a Galois extension. Then by
Theorem 2.19, we find that H = Gal(L/ L H ). Starting from M E SJJ1, we find
that Gal(L/M) E Sj and M = LGaJ(L/M) by Theorem 2.19. 0
Exercises
the function at the given point. Associating the order at a point with mero-
morphic functions gives a valuation of the function field specific to the point.
An idea of how to algebraize a Riemann surface is to consider the set of all val-
uations of its function field trivial over the base field and to replace coordinate
neighborhoods by corresponding valuations. We recall here a formal definition
of discrete valuation rings (DVR) inside an algebraic function field .it over K.
A discrete valuation v of .it trivial on K is a surjective map v : .it -+ Z U {oo}
satisfying the following four conditions:
(VO) v(KX) = 0 (triviality over the base field);
(V1) v(f) = 00 B j = 0;
(V2) v(f + g) ?: min(v(f), v(g)) for all j, g E .it;
(V3) v(f g) = v(f) + v(g) for all j, g E .it.
Here we agree as a convention to have a + 00 = 00 and 00 > a for all a E Z.
By the above properties, Vv = {j E .itlv(f) ?: O} is a subring of .it, and either
x E .it or ~ belongs to V v . In particular, the field of fractions of Vv is equal to
.it. The ring Vv is called a discrete valuation ring, and m = {j E Vv Iv(f) ?: I}
is a unique maximal ideal of Vv (so, Vv is local). The field Vv/m is called the
residue field of the valuation v and is a finite extension of the base field K.
Moreover, every ideal of Vv (except for the zero ideal) is a power of m, and
mn = {j E .itlv(f) ?: n}. See [CRT] Chapter 4 for more about valuation rings.
Our goal in this subsection is to create a space (a geometric object) from
the purely algebraic notion of algebraic function fields .it. The object is called
the Zariski-Riemann space and is the collection of all valuations of .it trivial
over the base field K. The space is an algebraic replacement of the associated
Riemann surface for K = C.
Example 2.22 We start with the simplest Riemann surface: the Riemann
sphere P = P 1 = C u { 00 }. The meromorphic function field of the sphere P is
isomorphic to the rational function field .it = C( x ). Then the polynomial ring
qx] corresponds to the Riemann sphere P with coordinate x in the following
sense. The space P is covered by two coordinate neighborhoods Uo and U=
identical to C, Uo is centered at 0 with complex coordinate x, and the other
U= is centered at 00 with coordinate Xl = ~. Then the field of meromorphic
functions of P is given by C(x), and the polynomial ring qx] is the ring of
functions with the only possible pole at 00. Each meromorphic function ¢ on
Uo finite at 0 has its Taylor expansion in the coordinate x: ¢(x) = ~~=o anx n
whose radius of convergence is positive. If ¢ is not finite at 0, for some pos-
itive m, xm¢(x) is finite at 0, and ¢(x) therefore has its Laurent expansion
~n>_manxn. We can define a valuation Vo on C(x) by assigning the expo-
nent of the leading term of the Laurent expansion to a given meromorphic
function ¢ E C(x). The valuation vo(¢) is just the zero order of the function
¢ holomorphic at O. For any other point a E C = Uo (resp. a = 00 E U oo ),
we can take ta = x - a (resp. too = x- 1 ) as a coordinate around a, and we
can think of the valuation Va giving the order of zero at a. In differential ge-
ometry or in complex analysis, local behavior of functions reflected by Taylor
38 2 Geometric Reciprocity Laws
Example 2.23 Consider the function field Jt = C(x)[y] defined by the equa-
tion y2 = x(x - l)(x - A) for A E C different from 0 and 1. We consider the
square root y(x) = )x(x - l)(x - A) on Uo. This function has two values
on P vanishing at 0,1, A and has a pole at 00; in other words, its inverse
vanishes at 00. Thus the locus of the point (x, y, 1) E p 2 is a two-sheet cov-
ering of P that ramifies at 0,1, A, 00. Around x = 0, if a point circles around
(x, y) = (0,0), for two values of ±y, we have one value of x. In other words,
cutting a line segment [0,1] and [A, 00] from two copies of P and gluing the
two corresponding segments, we get the donut-shape Riemann surface R, on
which the functions y and x both have single values. The field of meromorphic
functions over R is given by C(x)[y]. Moreover, we can embed R into p 2 by
PH (x(P), y(P)), which satisfies the given equation y2 = x(x -l)(x - A). In
other words,
Theorem 2.24 For a given algebraic function field .R defined over a field
K, there exists a unique smooth projective curve V defined over K such that
.R "'" K(V). We also have a canonical one-to-one onto correspondence:
V(K) "'" {discrete valuations of.R trivial over K X with residue field K}.
If .R' I.R is an extension of algebraic function fields defined over K, the corre-
sponding smooth projective algebraic curve gives rise to a covering 7r : VI --+ V
40 2 Geometric Reciprocity Laws
The set {Vlvaluation rings of K(x) trivial on K} is called the set of closed
points of P. They are associated with a maximal ideal of K[x] or K[X-l].
Of course, if V is associated with (t(x)) and t(x) -=I- x, then the same V is
associated with x-deg(t)t(x) E K[x-l] because vp(x) = 0 if t(x) -=I- x. Thus
we find
{closed points of PI K }
_ { . .
V(K) -
. Iv(K X ) = 0 and v(~X)
v.dlscrete valuatIOn with residue field K
= Z} .
Any prime ideal P of K[x] is decomposed into a product of primes P l , ... ,Pg
in R. If K is algebraically closed, then there are 9 distinct points over a given
point P = (x - a) of P for a E K. If K is not algebraically closed, K(P,)
for some i may be a nontrivial finite extension of K. In other words, in the
algebraic closure K of K, we have at most [K(Pi ) : K] (geometric) points
coming out of P, corresponding to embeddings of K(Pi ) into K.
If furthermore, R = K[x, y] for another element y E ~, we have an equation
f(X, Y) = yg + al (X)yg-l + ... + ag(X) satisfied by (x, y). So for a given
point P = (x - a) E P(K), the point (a, f3) E K2 satisfying f(a, f3) = 0 gives
rise to a point P, over P in p2. In this way, P, f-t (a, f3) supplies us with
a projective embedding of V(K) into p 2(K). The image is the projective
algebraic curve defined by the equation f(X, Y) = 0 in p 2 (or more precisely,
using the homogeneous coordinate (X, Y, Z), it is defined by zg f( -i,~) = 0).
Thus except for finitely many points, there are really 9 distinct points in V(K)
over a given point P E P(K). More generally, if R = K[x, Yl,"" Ym], then
P f-t (x, Yl,'" ,Ym) gives an embedding of V into the (m + I)-dimensional
projective space pm+l.
42 2 Geometric Reciprocity Laws
If we take a proper subring K[x, y] <;; R such that .it = K(x, y), the ring
K[x, y] is not integrally closed. In other words, for P E V(K), x and y generate
a sub ring V~ of the valuation ring. Since mp n K[x] = (t(x)) for an irreducible
polynomial t(x), mp C V~ for some e:::; g (since pg C (t(x))Vp).
Let J(X, Y) be the equation of (x, y). We can think of the projective
algebraic curve C C p2 defined by this equation. Then if K is algebraically
closed, a = (x mod P) E K(P) = K and (3 = (y mod P) E K gives a point
(a,(3) of C in p2. Even if V does not ramify, in C(K), P' = (a,fJ) has extra
ramification, and V~ is not a valuation ring.
We have a natural morphism V --+ C --+ P taking P f--t (a, (3) f--t a. In this
sense, V is the largest projective algebraic curve (over P with coordinate x)
giving rise to the algebraic function field .it.
For a projective algebraic curve C, we consider its function field K( C) and
V~ made up of functions in K(C) finite at P E C. If V~ is a valuation ring, we
call P a smooth point. The projective algebraic curve C giving rise to a given
algebraic function field .it is called a model of .it. Among models of .it, there is a
unique model smooth everywhere, which is called the smooth or non-singular
model of .it. If .it' /.it is a finite extension, by the above construction, we have
a covering map V' --+ V as defined in the theorem. 0
For each point P E V(K) for V as in the theorem, J E .it has an expansion
J(t) = I:n»-oo ant n with an E K(P) for a generator t of PVp . An element
t E .it giving rise to the generator t of P is called a uniJormizer at P.
Corollary 2.25 Let the notation and assumption be as in the theorem. Then
the morphism n : V' --+ V is a polynomial map oj the projective coordinates.
Proof. Let R (resp. R') be the integral closure of K[x] in .it (resp . .it').
Choose generators so that R = K[X,Yl, ... ,Ym] and R' = K[x,y~, ... ,y~].
Since R c R', we find x = Jo(x,y~, ... ,y~) and y, = J,(x,y~, ... ,y~). Then
n((x, y~, ... , y~)) = (x, Yl, ... , Ym), as desired. 0
Since any field embedding a : .it --+ .it' brings .it into a subfield a (.it) ,
we have a morphism n : V' --+ Vo- of projective algebraic curves V' and Vo-
corresponding to .it' and a(.it) as long as a leaves the field of definition stable
(even if a is nontrivial on K). If .it is defined by J(X, Y1 , ... , Ym ), then a(.it)
is obviously defined by the polynomial a(f)(X, Y1 , ... , Y m ) obtained from J
by applying a to the coefficients in K of J(X, Y1 , ... , Ym). Thus Vo- as above
is actually given by the conjugate a(V) (i.e., a(V)(K) = a(V(K)) for any
choice of extension of a to K). We have proven:
Corollary 2.26 Let VI K and V; K be smooth projective curves. Then we
have a canonical isomorphism HomK(V, V') ~ HomK(K(V'), K(V)) given
by K(V') :3 ¢ f--t ¢ 0 J Jor a morphism J : V --+ V' oj projective curves, where
HomK(V, V') is the collection oj morphisms oj projective curves defined over
K and HomK(K(V'), K(V)) is the set oj all field homomorphisms over K.
2.4 Algebraic Curves over a Field 4:3
Exercises
(2.8)
2.4.3 Divisors
if P VI,
E
if P E VI/I,
We define the divisor group Div(V) = Div(Jt) by the free abelian group of
all formal finite sums D = LPEV epP, where ep = 0 except for finitely many
P E V. We define the degree of a divisor D = LPEV epP by deg(D) = Lp ep.
We have a homomorphism deg : Div(V) -+ £, and we define Divo(V)
Ker(deg).
Proposition 2.27 If f E Jt, then deg(div(J)) = 0; so, div(J) E Divo(V).
as desired. 0
2.4.4 Differentials
d(Ln p
dt cnt )dt p = ('""""' n l
6 cn nt p - ) dt p.
p n
In particular, df does not have the term t~' If we change coordinate tp into
t'p, the expression of df will change according to the chain rule:
rise to a vector space over it. We see gdf / g' dl' = .if if,
E it, because f and
l' satisfy the polynomial relation F(f,I') = 0 over K. In other words, if
F(X, Y) = I;2,J a2,JX2YJ, then taking a formal derivative of F(f, 1') = 0, we
have
0= dF(f, 1') = L
(ia 2,Jr- 1I'J df + ja 2,)r 1')-1 dl') ,
2,)
and df and dl' are linearly dependent over it. Thus this vector space over it
is one-dimensional.
We define v p (gdf) = v p (g dfd~t:)). Note here that v p (~~;) = 0 for two
parameters tp and t'p at P E V. By the chain rule, d'tp = ; /p ~~'p,
p we have
dfp )
v p ( g dt = vp (gd f
dt'p dt'p)
dt p = vp (gd f
dt'p ) + vp ( dt
dt'pp ) = vp (dJ
g dt'p ) .
Since the last expression of the above formula is the value vp(gdf) computed
with respect to the coordinate t'p, the value vp(gdf) is well-defined indepen-
dently of the choice of the parameter tp around P. In particular, we have
gdf = df (t p ) ) dtp
( g~ = (""
~' cntp
n ) dtp.
n~vp(gdf)
w = ( L cntp) dtp.
n~vp(w)
since the Laurent series expansion of ~r does not involve the term rl which
is not a derivative of a power of t. This finishes the proof.
Since ft./K(x) is a separable finite extension for a nonconstant x E ft., we
have the trace map TrstjK(x) ; ft. --+ K(x) for any nonconstant function x.
We study how the residue map behaves under field extensions using the trace
map. The inclusion K(x) y ft. is induced by the projection x ; V --+ pl. For
a prime ideal p of K[x], we consider X-l(p) = {P}; thus, pR = ITppe p for
the integral closure R of K[x] in ft..
By the Chinese remainder theorem, R/pR ~ ITp R/ pep and R/pn R ~
ITp R/ pnep. Thus the completion of ft. with respect to p is given by fi.p =
ITp:x(P)oo=p ft.p. In particular, we have Trstj K(x) (f) = Lpx(p)oo=p Trstpj K(x)p (f)
for f E ft., because TrstjK(x)(f) is Tr(p(f)) for the matrix p(f) E Mn(K(x))
given by (fgl, fg2,"" jgd) = (gl,"" gd)p(f) for a base g] of ft./ K(x).
In ft.p = K[[tlJ (t = tp), we find t e = utp (e = ep) for a unit power series
u = Co + Cl t + C2t2 + ... with Co i- O. We can take an eth root v of u in
K[[t]] because K is algebraically closed. Changing t by vt, we may assume
that K(x)p = K((t ep )) and tp = teo Then we realize that K((t)) is a Galois
extension of K(x)p = K(W)) with the Galois group isomorphic to the group
of eth roots of unity {Le. Each ( E {Le acts on t by t t-+ (t. Then
TrstpjK(x)pt n = ( L
(EfJ.e
en) t n {~tn
=
e f n,
if eln.
48 2 Geometric Reciprocity Laws
if f3 = a and j = 1,
if f3 = 00 and j = 1, (2.10)
otherwise.
We now treat the general case; so, K is an arbitrary field inside its algebraic
closure K. We suppose that K is algebraically closed in .it. The composite
K.it ~ .it ®K K is an algebraic function field over K. We write V (resp.
V) for the Zariski-Riemann space of.it = K(V) (resp. K.it = K(V». Then
we have a natural inclusion V(K) '-+ V(K). Note here that V(K) = {P E
VIK(P) = K}. If K(P) "2 K, then there are [K(P) : K] points corresponding
to P. In other words, for points P l , ... ,Pd with d = [K(P} : Kl in V(K),
PJ n.it = P {:} vPJ 1.1\ = Vp or, equivalently, Pj corresponds to each field
embedding K(P) '-+ K. We define Resp(w) = Lj ResPJ (w), where w is the
differential of K.it corresponding to w = gdl. Since ResPJ (w) for different j
are conjugates of each other, we find Resp(w) E K. Similarly, we send P to
the sum L J Pj (trace of PJ ) in Div(V) and in this way, we embed Div(V) into
Div(V). This embedding sends P(V) into P(V), and we have deg on Div(V)
by pulling back deg defined on Div(V). In this way, everything we stated (in
2.4 Algebraic Curves over a Field 49
particular, the residue theorem, Theorem 2.29) is valid over the general field
K (not necessarily algebraically closed).
Similarly to L(div(w)), we define for a general divisor D,
where for two divisors D = I:p epP and D' = I:p e'pP, we write D :::: D' if
ep :::: e'p for all P. Plainly L(D) is a K-vector space.
Lemma 2.30 The dimension dimK L(D) is finite. In particular, we have
L(O) = K and dimK L(O) = 1.
Proof. If f is nonconstant, div(f) is nontrivial (in V) and deg( div(f)) = o.
Thus L(O) c K n ~ = K (by the definition of algebraic function field, the
integral closure in ~ of the base field is itself). For sufficiently positive D', we
see L(D) c L(D'). Writing D' = I:p e'pP and expanding f E L(D') into a
Laurent series of tp at P E D', we have e'p linear forms taking f E L(D') to
the coefficient of t pj for 0 < j ~ e'p. If all such linear forms vanish at f, we
see f E L(O) and hence f is a constant. Thus dimK L(D') < 00, which proves
the desired assertion. 0
Since f . L(D) = L(D - div(f)), the dimension dimK L(D) depends only
on the linear equivalence class of D. For the linear equivalence class f2 of
differential divisors, we have a well-defined number dimK L(f2 - D).
We introduce the following Riemann-Roch theorem (see [ENT] Chap-
ter VI, [FAN] 7.2 and [GME] 2.1.3 for different proofs).
Theorem 2.31 We have for all DE Div(V),
Exercises
stA = st + V c II stp,
PEV
(2.12)
Exercises
1. Give a detailed proof of (2.11).
2. Prove that stA is locally compact if and only if K is a finite field.
3. Prove that V(D) + V(D') = V([D, D'D for the least common multiple
[D,D'] of D and D'.
4. Prove that V(D) n V(D') = V((D,D')) for the greatest common divisor
(D,D') of D and D'.
2.5 Elliptic Curves over a Field 51
For divisors m[O], we study L(m[O]). Since 0[0] is the trivial divisor 0, we find
L(O[O]) = K. By definition, [2EjK is one-dimensional over K. Since deg(w) =
2g - 2 = 0 for a differential w on E, if w has a zero, then w has to have a pole.
If w is holomorphic nonzero, it vanishes nowhere. In other words, nowhere-
vanishing differentials are all nonzero multiples of w. In particular, div(w) = O.
Recall the linear equivalence class of div(w l ) for all meromorphic differentials
Wi on E. Then [2 is the linear equivalence class of all meromorphic functions
in R, and L([2 - D) ~ L( -D). If D > 0, then dimK L( -D) = 0 (Exercise 1).
We fix a nowhere-vanishing differential w. If f E L([O]), then f has pos-
sibly only one simple pole at O. Thus Reso(fw) = LPEE Resp(fw) = 0 by
Theorem 2.29. The function f is in L(O) and is constant, and dimK L([O]) =
dimK L(O) = l. By the Riemann-Roch theorem, we find
Exercises
We now embed ElK into the two-dimensional projective space PjK using a
base of L(3[0]) and determine the equation of the image in Pj K" We first
consider L(n[O]) which has dimension n if n > O. We have L([O]) = K and
L(2[0]) = K1 + Kx. Since x has to have a pole of order 2 at 0, we may
normalize x so that x = T- 2(1 + higher terms) in K[[T]]. Here x is unique up
to translation: x ~ x+a with a E K. Then L(3[0]) = K1+Kx+Ky. We may
then normalize y so that y = -T-3(1+higher terms) (following the tradition,
we later rewrite y for 2y; thus, the normalization will be y = -2T- 3(1 +
higher terms) at the end). Then y is unique up to the affine transformation:
y ~ y + ax + b (a, bE K).
Proposition 2.33 Suppose that the characteristic of the base field K is dif-
ferent from 2 and 3. Then for a given pair (E,w) of an elliptic curve E and
a nowhere-vanishing differential w both defined over K, we can find a unique
base (1, x, y) of L(3[0]) such that E is embedded into Pj K by (1, x, y) whose
image is defined by the affine equation
(2.14)
L(4[0]) =K + Kx + Ky + Kx 2 ,
L(5[0]) = K + Kx + Ky + Kx 2 + Kxy and
L(6[0]) = K + Kx + Ky + Kx 2 + Kxy + Kx 3
=K + K x + K y + K x 2 + K xy + K y2,
(2.15)
Exercises
1. Show that dx/y does not vanish at any point on E.
2. Show that if L1 = 0, the curve defined by y2 = 4x 3 - g2X - g3 is not smooth
at the multiple root a of 4x 3 - g2X - g3 = 0.
54 2 Geometric Reciprocity Laws
where [.J indicates the set of isomorphism classes of the objects inside the
bracket and Spec(R)(K) for a ring R is the set of all algebra homomorphisms:
R -t K. The last isomorphism sends (92,93) to the algebra homomorphism ¢
with ¢(X) = g2 and ¢(Y) = 93.
We now classify elliptic curves E eliminating the contribution of the dif-
ferential from the pair (E,w). If ip : E "=' E' for (E,w) and (E',w'), we
have ip*w' = AW with A E K X , because ip*w' is another nowhere-vanishing
differential. Therefore we study KX-orbit: (E,w) mod K X under the action
of A E K X given by (E,W)jK ~ (E,AW)jK, computing the dependence of
9J(E,AW) (j = 2,3) on A for a given pair (E,W)jK' Let T be the parameter
adapted to w. Then AT is adapted to Aw. We see
E )=(-2+T1/J(T)) (EA)=(-2+higherterms)=A_3( )
y ( ,w T3 =>y, w (AT)3 yE,w.
(2.16)
Thus we have
2.5 Elliptic Curves over a Field 55
Theorem 2.34 If two elliptic curves ElK and ElK are isomorphic, then
choosing nowhere-vanishing differentials wIEand wI E" we have gJ (E', w') =
). -2J gJ (E, w) for). E K x. The constant). is given by <p*w' = ).w.
for a twelfth root ). of LJ. (E, w) / LJ. (E' , w'). Note that the twelfth root ). may
not be in K if K is not algebraically closed.
Conversely, for a given j tic {O, I}, the elliptic curve defined by y2 = 4x 3 -
gx - 9 for 9 = J2~Jl has J-invariant 12 3j. If j = 0 or 1, we can take the
following elliptic curve with J = 0 or 12 3 . If J = 0, then y2 = 4x 3 - 1 and if
J = 12 3 , then y2 = 4x 3 - x. Thus we have
Corollary 2.35 If K is algebraically closed, then J(E) = J(E') ¢} E ~ E'
for two elliptic curves over K . Moreover, for any field K, there exists an
elliptic curve E with a given J (E) E K.
Exercises
l. Prove that gJ(E',w') = ).-2]9J(E,w) for suitable wand w' and a suitable
twelfth root). of LJ.(E,w)/LJ.(E',w') if J(E) = J(E').
2. Explain what happens if J(E) = J(E') but E 'f'- E' over a field Knot
necessarily algebraically closed.
{
X(P)X + y(P)Y + Z = 0,
x(Q)X + y(Q)Y + Z = O.
56 2 Geometric Reciprocity Laws
Here (1, x, y) is a base of L(3[0]). Then for a nontrivial solution (a, b, c) of the
above system, we define 9 = ax+by+c. We assume that Q 1= -Po Then b 1= 0
(because b = 0 =} Q = - P as is clear from the definition of - P). Thus 9 has
a pole of order 3 at O. Since 9 is nonconstant with two zeros P, Q, we find a
unique point R such that div(g) = P+Q+R-3[0]. Then we define P+Q to
be - R. By Abel's theorem we prove in the following subsection, this addition
gives rise to an abelian group structure on E. When Q = - P, we just define
P+Q = O.
Here is a geometric interpretation of the above definition of the addition.
Embed E into p2 by 1, x, y. Then we draw a line aX + bY + cZ = 0 passing
through P, Q in p2, where (X, Y, Z) is the homogeneous coordinate of p2.
Then we consider the function ¢ : E -+ pI given by the equation ¢(X, Y, Z) =
(aX + bY + cZ)/Z, which is an element in .R = K(E). We see that ¢ vanishes
at two points P, Q, and the line intersects with E at a unique point Ragain.
Then div(¢) = P+Q+R-3[0]. In particular, the coordinate of R is a rational
function of the coordinates of P and Q; so, + : E x E -+ E is a morphism
of algebraic geometry (in other words, it is induced from a field embedding
.R '-t .R ® K .R). Similarly, P H -Pis an automorphism of the elliptic curve
(therefore of the algebraic function field .R).
After embedding E into p2, for any field automorphism (J" of K, we can
apply (J" to the coordinate of E. Then we get a new elliptic curve EeJ. If E
is defined, for example, by F(X, Y) = :L",J ai,JXiYJ = 0, then E eJ is defined
by FeJ(X, Y) = :L",j (J"(ai,J)Xiyj = O. Since everything we have proved for E
shifts to EeJ, the morphism + : E x E H E will be sent to + : EeJ x EeJ -+ EeJ.
If E is defined over a smaller field k c k = K, the rational functions of the
coordinates of E giving rise to + : Ex E -+ E are therefore invariant under (J";
so, it is a rational function with coefficients in k. Thus what we have said so
far is valid for any elliptic curve defined over any perfect field (not necessarily
over an algebraically closed field).
Exercises
we have defined in the previous subsection is induced from the group structure
of the Jacobian Jac(E).
Proof. By the same argument (as in the last part of the previous subsection)
conjugating by automorphisms of K, everything we prove over an algebraically
closed field K will be valid over any subfield k such that K j k is algebraic.
Thus we may assume that K is algebraically closed. We first remark that
Indeed, if 0 i= f E L(D) implies that div(f) + D ;:::: 0; so, taking the degree,
we have deg(D) ;:::: - deg(div(f)) = O. This shows
function field .R = K(E) into itself. The degree [.R: N*.R] = deg(N) is finite,
because .R is finitely generated over K and of transcendental degree 1. Thus
we find that N : E -+ E is surjective for all positive integer N. In particular,
E(K) is an N-divisible group if K is algebraically closed, and we have proven
this fact as long as N is prime to the characteristic of K.
A naive question is: What is deg(N) = [.R: N*(.R)]? The following theorem
gives us that information and slightly more.
Theorem 2.37 Suppose that K is algebraically closed. Then deg(N) = N 2.
Moreover, if N is outside the characteristic of K, we have E[N](K) ~
(71,1 N71,)2 as abelian groups, and if p > 0 is the characteristic of the field
K, then E lP] is either trivial or isomorphic to 71,1 p71,.
We prove this theorem in the following subsection when K is a subfield of C.
See [GME] 2.6 for more general cases.
In the rest of this subsection, we explore consequences of the theorem.
Start with an arbitrary field K and write K for its algebraic closure. Take an
elliptic curve E defined over K with function field .R. We write the composite
.RK as ~, which gives rise to an elliptic curve E over K. As the algebraic curve
inside pm (the integer m can be taken to be 2 as we have seen), E is just
defined by equations giving relations among generators of.R over K. Since ~
has the same generators (as .R) over K, they give rise to the same algebraic
curve inside pm. For any extension F 1K inside K, we define E (F) to be the
set of points of E with coordinates in F. Thus E(F) is the subset of E(K)
fixed by Gal(K 1F).
Take two points P, Q in E(K). Since P + Q is determined by the third
intersection with E of the line passing through P and Q, if we conjugate the
coordinate by a E Gal(K1K), the resulting image goes to the third intersec-
tion with E of the line passing through O'(P) and O'(Q) (when P = Q, the line
is the line tangent to P E E(K)). This shows that a(P) + a(Q) = O'(P + Q),
and conjugation by Galois automorphism is compatible with the addition of
E. In particular, O'(N . P) = N . O'(P), so the Galois action preserves E[N]
(since 0'(0) = 0 because 0 E E(K)). In other words, a : E[N] -+ E[N] is
an injective group homomorphism. Since E[N] is a finite group, a induces an
automorphism of the group E[N].
Let P, Q be a base of E[N] over 71,IN71,. Then O'(P), O'(Q) is another base.
Write p( a) the base-change matrix:
( a(p))
a(Q)
= (ap+b Q )
cP+dQ
= (acdb) (P)
Q
Exercises
1. Prove that p(T, P(S, R)) = p(p(T, S), R) in the power series ring of three
variables T, S, Rover K.
2. Prove p(U)p(T) = p(UT).
U = P '()
YL () u = -2
u3 - 2 '~
" (u _1 £)3 = -2u -3 + ... ,
€EL-{O}
where
and
This shows
L t ~ [(E )
a
IE:
an elliptic curve over C, ]
,W Ie w: a nowhere-vanishing differential '
where the straight brackets [ ] indicate the set of all isomorphism classes of
the objects inside.
We now make the space Lat more explicit. Two complex numbers WI, W2
span a lattice if and only if Im(wdw2) -=J. O. Let Sj = {z E q Im(z) > O}.
Interchanging WI and W2 if necessary, we may assume that Im(wdw2) > O. So
we get a natural isomorphism of complex manifolds via (:;;; ) f--t (W2' wd W2):
62 2 Geometric Reciprocity Laws
Since v and v' span the same lattice L if and only if v' = av for a E SL 2(Z),
we see that Lat ~ SL 2(Z)\B. The action of a = (~~) E SL 2(Z) on B is given
on <ex x S) by a(u,z) = (cu+d,a(z)) for a(z) = ~;:~.
We now consider the variables g2 and g3 and the J-invariant J defined
below Theorem 2.34 as a function on the upper half complex plane S). In
particular, by Corollary 2.35, J satisfies J(-y(z)) = J(z) for,,( E SL2(Z),
Also we see easily that xu().w) = ).-2xu(w) for nonzero scalar). E <ex. We
call a meromorphic function f on B a modular form of weight k on r(N)
if f satisfies f(-yw) = f(w) for all "( E SL 2(Z) with "( - 12 E NM2(Z) and
f().w) = ).-kf(w) for)' E <ex. In particular, if k = 0, they are called modular
functions on r(N). Thus fu is a modular function on r(N), and Xu is a
modular form of weight 2 on r(N). We often let r(N) denote the subgroup
of SL 2 (Z) made of matrices congruent to b modulo N. We have the following
exact sequence,
The surjectivity of the last "mod N" map is nontrivial (Exercise 1).
Since B = S) x <ex, one may regard a modular form as a meromorphic
function on the upper half complex plane S) by putting f(z) = f (21fi (1' )).
Since a (n = (cz + d) (,t)) for a = (~~), we find that f : S) ---+ <e is a
modular form of weight k on r(N) if the following conditions are satisfied,
(G1) f(a(z)) = f(z)(cz + d)k for all a = (~~) E r(N).
If f is a modular form on r(N), f(z+N) = f(z); so, f is a function on S)jNZ,
where NZ acts by translation. By the variable change, q = qN = expe;;Z),
2.6 Elliptic Modular Function Field 63
we may identify SJ/NZ with the open unit disk {q E rcllql < I} punctured at
o (0 corresponds to 00), and we may regard f as a function of q defined over
the punctured disk. It has a Laurent expansion around 0: f(q) = Ln anqn,
which is called the q-expansion of f. We assume that f(q) is finite tailed; that
is, there exists N > 0 such that an = 0 as long as n < -N.
We can compute explicitly the q-expansion of g2, g3 and Ll for q = ql:
12g 2 = 1 + 240 f {L
n=l O<dln
d 3 } qn
-6393=1-504f{ L d 5 }qn
(2.20)
n=l O<dln
II (1 -
00
Ll = q qn)24.
n=l
Gal(Q(J)(E[N])/Q(J)) C GL 2(Z/NZ).
By the above proof, q gives a parameter at CXl of Y(N) defined over IQ((N);
so, we have
This shows
Corollary 2.41 The curve Y(N)/c is actually defined over IQ[(N], and
We consider the union fi = UN fiN (note here that fiN C fiM if NIM); so, fi
is a field.
Corollary 2.42 We have Gal(fi/IQ(J» ~ GL 2 (Z)/{±I} and fi n <C = IQcyc.
Moreover, each matrix 9 E G L2 (Z) acts on IQcyc by the action of det(g) under
the identification Gal (lQcyc /IQ) = ZX .
For a E M 2 (Z) with det(a) > 0, we can always write by elementary divisor
theory, a = "( (7i ~ h' with "(, "(' E S L2 (Z). We write 8 = (7i ~). Since
the effect of f H f ° 8 on q-expansion is q H qm/n, the action f H f ° a
preserves the coefficient field IQ[(N]. Since 8- 1 r(N)8 :J r(mnN) (Exercise 2),
fiN 08 C fimnN. Since "(, "(' E SL 2 (Z) preserve fiN, we see that fiNoa C fimnN
and hence f H f ° a is an automorphism T(a) of it. For any a E GL 2(1Q)+,
its integer multiple is in M 2 (Z). Thus the linear transformation Z H a(z) is
induced by an integer matrix; so, T(a) E Aut(fi) is well defined on fi; that is,
GL 2(1Q)+ acts on fi by f H f ° a. We can check (e.g., [IAT] 6.4 or [MFG]
Corollary 3.3):
The action of GL 2 (A)+ on yt is as given above. The difficult part of the proof
is the surjectivity of T for which we refer the reader to [I AT] Theorem 6.23.
We later give a sketch of a proof in a more general setting, Theorem 4.14.
For each open compact subgroup S of GL 2 (Ah/(YGL 2 (lR)+, we have an
algebraic function field yts, which gives rise to a projective curve Vs defined
over the fixed subfield ks of IQcyc by det(S) c A x /lQx lR~ = Gal(lQcyc /IQ).
This curve is the modular curve of level S, and Vs(C) = rs\SJ (forgetting
the points above 00 E P1(J)). Here rs = S n SL 2 (1Q). In particular, Y(N)
corresponds to {u E G L2 (2) lu == 12 mod N}, and we have the following
identity,
[(E, cp : (7L/N7L)2 ~ E[N])IK!eN(cp(l, 0), cp(O, 1)) = (N 1 ~ Y(N)(K),
where K is any field extension of IQ[(N] for a specific primitive Nth root of
unity (N, and [.] indicates the set of isomorphism classes of the pairs (E, cp).
We have (E, cp) I K ~ (E', cp') I K if there exists an isomorphism f : E --+ E'
defined over K such that f 0 cp = cp'.
Since A = A ( 00 ) x lR, we have GL 2 ( / x
A)+ (A(oo))
IQ GL 2 (lR) + = GL hlQl) ,writ-
ing Z(IQ) for the center of GL 2 (1Q) isomorphic to IQx. Consider T giving the
isomorphism
GL 2 (A (00))
T: Z(IQ) ~ Aut(yt).
The fixed field yts of T( S) is an algebraic function field; so, it is the function
field of a unique smooth projective curve Vs defined over ks = yts n IQ
(which is the fixed field of &(det(S)) for &in Theorem 2.12). Therefore the
isomorphism T in Theorem 2.43 gives rise to a tower of algebraic curves {Vs} s
GL (A(oo))
defined over ks indexed by open compact subgroups S c ~(IQI) . The tower
{Vs} s is called Shimura's canonical model of the tower of modular curves (or
simply just the tower of modular curves), which classifies elliptic curves with
additional structures and is the simplest example of the Shimura varieties
we study in the rest of the book. We write Ys (C) for the image of r s \SJ
f
in Vs ( e ) for rs = PGL 2 ( IQ ) n S in
GL (A(oo))
~(IQI) . Then Vs - Ys is a set 0
finite points (called cusps of Vs), and Ys is an open algebraic curve (a quasi-
projective curve) defined over k s . In other words, VS/ks is the unique smooth
compactification of Y Slks '
We may regard Vs as defined over IQ forgetting the requirement of the field
of definition to be the algebraic closure of IQ in the function field ks (Vs). By
the strong approximation theorem (i.e., the density of SL 2 (1Q) in SL 2 (A(00));
e.g., [MFG] 3.1.2), GL 2 (1Q)\GL 2 (A(00))/Z(IQ)S is in bijection with Gal(ks/IQ)
by 9 H &( det(g)). We write 9 S for the conjugate 9 . S· g-l. Since ks Q91Q1 e =
ITaEGal(ks/lQl) e by k Q9 x --+ (a(k)x)a, we have
which implies
U VS/C = U
aEGal(ks/QI) EGL ("")\ GL2(A(oo» /S
9 2 '" Z(Q)
(2.22)
U VoS/c(C),
(2.23)
Exercises
In this chapter, we translate the result in the latter part of the previous
chapter into the language of schemes, and at the same time, we sketch a proof
of the vertical control theorem in the p-ordinary case of elliptic modular forms.
There are several different proofs of the vertical control theorem:
(1) Through the moduli theory of elliptic curves; this is what we show (Ann.
Sci. Ec. Norm. Sup. 19 (1986) and [GME] Chapter 3);
(2) Through the study of topological cohomology groups and Jacobians of
modular curves. This way has the advantage of simultaneously producing
at the same time Galois representations into GL 2 (IT), where IT is a quotient
of the universal p-ordinary Hecke algebra. The ring IT could be large and
may be free of finite type over W[[X]] (Inventiones 85 (1986));
(3) Through the theory of p-adic Eisenstein measures and p-adic Rankin con-
volution theory. This method was found by A. Wiles in [Wi] and is pre-
sented in the elliptic modular case in my book [LFE] in Chapter 7;
(4) As an application of the identification of Hecke algebras and universal
Galois deformation rings at many different weights (done by Wiles and
Taylor). This method is presented in my book [MFG] 5.3.5.
We follow the first method. Chapter 3 of the book [GME] contains a more
down-to-earth description of the proof. This first method seems most effec-
tive when we try to generalize the VCT to more general reductive groups
(as long as the group admits Shimura varieties). We generalize this proof to
automorphic forms on more general Shimura varieties (see Sections 4.1, 4.2,
and 8.3).
Studying curves G over a field k in Section 2.4, we have analyzed its set
G(K) of K-points for a field extension K/k. The main idea behind the theory
in Section 2.4 is to associate with each point P E G(K) a discrete valuation
of the function field K(G). However, to study a curve G over a base ring
B, obviously we need to study G(R) directly for an arbitrary B-algebra R.
The main point of scheme theory is to consider G as a covariant functor
R I-t G(R) from the category B-ALG (also written as ALG/ B ) of B-algebras
into the category SETS of sets. We have many interesting functors F : B-
ALG -+ SETS. For example, we may associate with R the set [E/Rl of all
isomorphism classes of elliptic curves defined over R (or Spec(R)). This is
a covariant functor since a B-algebra homomorphism R ~ R' induces the
base change E/R I-t (E ®R R') which is an elliptic curve defined over R'. We
can extend this functor to a contravariant functor from the category B-SG H
(also written as SGH/ B ) of schemes over B to SETS in an obvious manner.
The functor is contravariant, because R I-t Spec(R) reverses the direction of
morphisms. If the functor R I-t [E/Rl is isomorphic to a functor R I-t 9)1(R)
for a B-scheme 9)1, we could use algebraic geometry to study the functor (and
at the same time we may use the functoriality to study the geometry of the
moduli scheme 9)1). Finding 9)1 is a classification problem of elliptic curves.
Of course, we may add some extra structures ¢ to E to classify pairs (E, ¢ ),
and if successful, the moduli 9)1' of the functor R I-t [(E, ¢)/Rl constitutes a
scheme over 9)1 by the morphism (E, ¢) I-t E. We study this type of problem
in this and the following sections. In this section, we summarize necessary
facts on elliptic curves defined over a scheme.
J(D) C OE is S-£l.at (thus locally free). We define C(D) = J(D)-l and put
deg(D) = deg(C(D)) = ranks f*(OD). In particular, the 0 section gives rise
0#
to a divisor [0] of degree 1 given by 0[0] ~ Os. We then consider J(m[O]) =
J([o])m and C(m[O]) = J(m[O])-l for m E Z. The line bundle C(m[O]) can be
regarded as the sheaf of meromorphic functions on E smooth outside 0 and
with possible singularity at 0 having a pole of order ::; m.
If S = Spec(k) for an algebraically closed field k, k-rational effective divi-
sors can be identified with positive linear combinations of points on E(k). We
have deg(Lp mp[P]) = Lp mp. We can thus think of the group Div(Ejk)
of all formal linear combinations (including negative coefficients) of points
on E. Then deg : Div(Ejk) -+ Z is a well-defined homomorphism given by
the above formula. In particular, for any divisor D E Div(Ejk), we have
C(D) = C(D+) 0 C(D_)-l writing D = D+ - D_ for effective divisors D+
and D_, and we can verify deg(C(D)) = deg(D). By Abel's theorem (Theo-
rem 2.36), any invertible sheaf C on E/k is isomorphic to C(D) for a divisor
D, and we define deg(C) = deg(C(D)) if E is defined over an algebraically
closed field k.
For any scheme X, we define Pic(X) as the set of all isomorphism classes
of invertible sheaves on X. The association X H Pic(X) is a contravariant
functor by the pullback of invertible sheaves, and Pic(X) is actually a group
by tensor product.
Let E .4 S be an elliptic curve. We define, for each S-scheme ¢ : T -+ S,
We can extend the degree map to deg : PicE/s(T) -+ Z 7r o(T) for the set of
connected components 7ro(T). Indeed, for any algebraically closed field k and
a geometric point s : Spec(k) Y T, the fiber E(s) = E xs,¢os s = ET XT S
is an elliptic curve over the field k and deg(C) = deg(C(s)) for the pull back
C( s) at s, which is well-defined independently of the choice of s in a connected
component. We embed Z in Z 7r o(T) diagonally, and we define
is a spectrum of a field, the valuation ring Vs' associated with a closed point
s' E C' is given by the stalk Dc, ,s'; so, this definition of deg( ¢) is consistent
with the one given for algebraic curves over a field just after Corollary 2.26.
Thus Adeg(¢) ¢*Oc is an invertible sheaf on C'. If ¢ : E --+ E' is an S-
morphism of elliptic curves, by our convention, ¢ takes 0 E to 0 E' and, hence,
at the side of the Picard scheme, it is just L --+ Adeg(¢) ¢*L; so, obviously ¢
is a homomorphism of group functors. In particular, a morphism of elliptic
curves (taking zero to zero) is a homomorphism of group schemes.
via (X ~ Y) --+ (¢(T): X(T) --+ Y(T)) given by ¢(T)(T -Lr X) = ¢of. This
is intuitively clear because an algebraic variety is just a function associating to
each ring R its R- integral points X (R) = X (Spec( R)). The verification of this
is left to the reader as an exercise (the inverse is given by HomcTF(X, Y) :3
F H F(X)(Ix) where F(X) : X(X) --+ Y(X) = Homs(X, Y); see, for exam-
ple, [MFG] 4.1.3 or [GME] 1.4.3).
This fact can be used as follows. Let N be a positive integer. Since E(T)
is a group, x H N· x gives a functorial map N(T) : E(T) --+ E(T); so, an
endomorphism of elliptic curves N : E --+ E. We define its kernel E[N]
E XE,N.O S:
E[N] ----+ E
1 IN
S ----+ E.
o
It is clear that E[N](T) = Ker(N(T)). It is known that deg N = N 2 (e.g.,
[GME] Theorem 2.6.2) and if N is invertible over S, E[N](k) ~ (Z/NZ)2 for
all algebraically closed fields k with Spec(k) --+ S.
3.1 Basics of Elliptic Curves over a Scheme 71
so, the structure sheaf of Zj NZ is free of finite rank N. Such a group scheme
is called a locally iree group scheme (of rank N).
There is another example of locally free group schemes: starting with the
multiplicative group G m (as a group functor Gm(R) = R X and as an affine
scheme G m = Spec(Z[t, el])), we consider the kernel JLN of N (as a group
functor JLN(R) = {( E RI(N = 1} and as a scheme JLN = Spec(Z[t]j(tN -1)) =
Spec(Z[(ZjNZ)])). Then JLN is a locally free group scheme of rank N. If
N> 1, it is nonisomorphic to ZjNZ, since for any prime p, JLp(lFp) = {1} but
(ZjpZ)(lFp) = ZjpZ. We consider a version of the functor P~dN) defined as
follows,
Pr,CN)(R) = [(E,¢N: JLN '-+ E[N],w)/R]'
The two functors Pr,CN) and PhCN) are isomorphic by Cartier duality (see
[GME]l. 7): If G is a locally free group scheme of rank N over S, there exists a
group scheme G/ s such that G(T) = HomT(GT,G m/ T ) = HomT(GT'~N/T)'
where G m/ S = G m x Sand JLN/S = JLN x S over Spec(Z). We have G s=! G
in an obvious manner, and 'i:jiiZ = JLN by ((m) = (m for ( E JLN(R) and
mE (ZjNZ)(R).
Let E -4
S be an elliptic curve. The section 0 : S -+ E induces a section
of 1* : Pic(S) -+ Pic(E); so, we have a splitting:
Let h, = i, 07r; we have 7r*£lv, = h;:lOv, for V. = 7r- 1 (U,). Let P E Ker(7r);
then
h, 0 P = ii 07r 0 P = i, oOE' = i1 oOE' = h j 0 P.
--
This implies that the h, 0 Ps glue to give a global section hoP
Gm(T), getting a homomorphism Ker(7r*) -+ Ker(7r), which can be easily
verified to be an isomorphism (because twice this operation yields an identity
E r(T, O~) =
Comparing the leading term of T-6, one sees that the seven sections
in the space HO(E, £(6[0])) have to be linearly dependent and satisfy the
following relation,
We can kill in a unique way the terms involving xy and y by a variable change
y H Y + ax + b. Indeed, by the variable change y H Y - TX - a23 , we get a
simplified equation:
y2 = x 3 + b2x2 + b4 x + b6 .
Again a variable change x H x- pt simplifies the equation to
y2 = x 3 + C2 X + C3.
Since £(3[0]) is very ample (deg(£(3[0])) = 3 ~ 2g + 1), by finally making a
variable change 2y H Y (so now the T-expansion of y begins with -2T- 3 ),
we get a unique equation from (E,W)/R:
74 3 Modular Curves
as long as 6 is invertible in R.
Consider (E, P, w) / R for a point P E E[f] (R) of order f for a prime f. We have
a unique rp E M1(R) such that
p
'PE T
Spec(R) --+ E
The group scheme G m acts on the functor PrdN) in the following way,
76 3 Modular Curves
for ), E Gm(R). This induces (by the functoriality described in Section 3.l.5)
an action of G m on MrdN) and hence on Rr,(N).
Here is a general fact on the action of G m (see [GME] l.6.5). Let X be
a B-module for a base ring B. Regard X as a functor from B-ALG to the
category of B-modules B- MOD by X (R) = X (SJ B R. If a group scheme G / B
has an action coming from the following morphism of functors, G x X -+ X
(which induces the set-theoretic action G(R) x X(R) -+ X(R) for each B-
algebra R), we call X a schematic representation of G. It is known that if X
has a schematic action of G m / B , then
such that X[K:] = {x E XI)'· x = A"x}; that is, X[K:] is the eigenspace for the
character Gm(B) -+ B X taking Z E Gm(B) = B X to z".
The action of G m / B on MrdN) gives rise to a schematic action on RrdN)
(because it was defined by functorial action; see Section 3.l.5). Thus we can
split
RrdN)/R = EB R,,(rl(N); R),
"EZ
where on f E R,,(r1(N); R), G m acts by the character -K:.
Since fER" (rl (N); B) is a functorial morphism,
for A(n) = EBJ AnJ (see [GME] Lemma l.3.1). Since Ll- 1 ERe RrdN) , the
graded ring RrdN) has a unit of degree 12, and hence, we have
3.2 Moduli of Elliptic Curves and the Igusa Tower 77
defined over Z[6~], which is locally free of finite rank over Ml = Proj(R) =
pl(J) ~ {oo}. For all geometric points Spec(k) of Spec(Z[6~])' we have
Er,(N)(k) = [(E,¢N)/k] = Yl(N)(k). The above assertion holds for any
Z[6~ ]-algebras R in place of algebraically closed fields k if N ;::: 4.
Here a "geometric point" means that k is an algebraically closed field. This
theorem over Z[iI] for an unspecified M with NIM was proved by Shimura
in the late 1950s (see [CPS] I [57a]) , and the result over Z[~] was proved by
Igusa in [I] soon after Shimura's work.
Returning to the classical setting, we define an arithmetic subgroup r 1 (N)
of SL 2 (rfJ) by
Since SL 2 (Z) acts on the upper half complex plane 5) discretely by z >--+ ~;:~,
we can make a quotient Riemann surface r 1 (N)\5) (see [IAT] Chapter 1). By
the association: z >--+ (Ez(rc) = C/(Zz+Z), P z = ~ E Ez(rc)), it is well known
that r 1 (N) \5) classifies all elliptic curves with a point P of order N over C
(cf. [IAT] Chapter 4 and [GME] 2.4); so, we conclude
3.2.4 Compactification
Let Gr,(N)(Z[6~]) be the integral closure of the ring G(Z) in the graded
r
ring R r , (N) /7/.,[ 1v l' To see that G r , (N) (Z[ 6~]) is a graded ring, we write for
the nontrivial homogeneous projection of highest degree of r E R r ,(N)/7/.,[61vl'
78 3 Modular Curves
. 1
Tate(q) = ProJ (Z[[q]][6"][X, Y, Z]j(Zy2 - 4X 3 + g2(q)XZ 2 + g3(q)Z3)),
which extends the universal curve over pI (J) - {oo} to pI (J) locally at the
cusp 00.
Since Tate(q)(R[[q]]) :J (R[[qW)/qZ (see [GME] Theorem 2.5.1), we may
regard Tate(q) as the algebraization of the formal quotient Gm/Z[[q)]/qZ of the
formal multiplicative group Gm ; so, it has a canonical level structure
account, which can be found in [AME] Chapter 10 and [GME] 3.1.1. Thus
Xl (N) is smooth at the cusps, and moreover f E Ck(rl (N); R) is a function
of (E,¢N,w) satisfying (GO-2) and
(G3) f(Tate(q),¢N,w) E R[(N][[ql/NlJ for any choice of ¢N and w.
Since rl(N)\(Sj U Pl(Q)) is a smooth compact Riemann surface and is the
normalization of pl(J) in Yl(N)(C) ([IAT] Chapter 1), we conclude
The space Ck(rl (N); C) is the classical space of modular forms on r l (N) of
weight k. Since Tate(q)(R[[q]]) ::J (R[[qW)jl~, we may consider Tate(q) to be
a "quotient" of Gm/Z[[q)] j qZ; so, it has a canonical level structure ¢'tr : J1N y
G m ...... Tate(q) and a canonical differential Wean induced by !if identifying
G m = Spec(Z[t, e l ]). In particular,
00
This is because:
1. If Ejiffp is ordinary, then E[P] ~ jJ..p x (ZlpZ) over lFp;
2. jJ..p = Spec(lFp[t]/(t P - 1)) shares the tangent space with G mjIFp ' because
they are both of dimension 1 infinitesimally;
3. p2 = P up to units in the supersingular case.
The zero locus of a nonzero section of a line bundle is an effective divisor;
t
hence, for any algebraically closed field k of characteristic p for p N, the
points in Xl(N)(k) carrying supersingular elliptic curves are finitely many
and algebraic over lFp .
Define [(q) = ((2~p)Ep-1. By the Von Staut theorem, the q-expansion [(q) of
[ is congruent to 1 modulo p; so, [ mod p coincides with H. Let (E, ¢N) jM
be the genus 1 semi-stable curve (completed by appropriate Tate curves at
the cusps) over M = X 1 (N)jw. Let
Mm = Xl (N)jW= = X1(N) Xw Wm .
where P is a point of order pQ. For a given test triple (E, P, ¢N )/R, we have
a unique commutative diagram (up to isomorphisms)
E ~ E
1 1
Spec(R) -----+ YI (N)
and t(P) E (E[pQ]et - E[pQ-I]et) (R). The association (E, P, ¢N) / R rl t(P)
gives rise to the following isomorphism,
Since Tm,Q has the same topological space as TI,Q, we may assume m = 1
in order to prove the irreducibility of Tm,Q; namely, we prove the following
theorem.
Theorem 3.3 (Igusa) Let p be a prime outside Nand iFp be an algebraic
closure oJlF p • Then the Jgusa curve TQ/ JFp := TI,Q over Xl (N) /JFp is irreducible
Jar all LV.
This theorem was first shown by Igusa in [1] using the fact that the inertia
group at each supersingular point of Xl (N)/JF p in Gal(TQjXI(N)) is equal to
the full group (7l,jpQ7l,) x. We prove this theorem assuming p > 3, since we
3.2 Moduli of Elliptic Curves and the Igusa Tower 83
have constructed modular curves only over Z[6~]. Our proof is new and can
be generalized easily to more general Shimura varieties (see Theorem 4.21,
Theorem 6.27, and Corollary 8.17).
Proof. We write Ta = Spec(Va) for an etale extension of Vo over lFp . For two
integers 0 < NIN' prime to p, Ta over X 1 (N') is a finite covering of Ta over
Xl (N), the irreducibility of the Igusa curve over Xl (N') implies that of the
Igusa curve over Xl (N). Thus we may assume that N is large so that Xl (N) is
smooth over Z[6~]. We may construct the Igusa tower T~ over S' = X(N)[i]
in exactly the same manner as Ta/X,(N) , replacing X 1 (N) by X(N). Again
Ta is a finite quotient of T~; so, we only need to prove the irreducibility of
T~. By Zariski's connectedness theorem ([ALG] III.11.5), the irreducibility
of the projective curve X(N) over Q combined with f*OX(N) = Z[(N, 6~]
for f : X(N) --+ Spec(Z[(N' 6~]) implies the irreducibility of Slover lFp (an
algebraic closure of lFp). We extend the scalar to lFp and again write Vo for
V1 ,0 ®IF p IFp. Then, Vo is an integral domain over IFp. Since T~ / S~ is etale, T~ is
a disjoint union of irreducible curves over S~/lFp. By the existence of the Tate
curve, the cusp 00 gives a point of T~, and there exists a unique irreducible
component Ca = Cr(N),a of T~ containing 00. The action of Z; on the level
p-structure permutes irreducible components of T~; so, all other irreducible
components are isomorphic to Ca. In this way, we get a tower of irreducible
curves
... --+ Ca --+ ... --+ Co = S~.
Since (Z/paz) x acts transitively on points of T~ over a given closed point of
S~, C a / S~ is a Galois covering with Galois group
Galois group H surjects down to Gal(K: oo / K:o) by the restriction map; so, we
need to show that Aut (K: oo / K:o) ::) Z;, where K:oo = U", K:",. Let V", be the
valuation ring of v'" in K(X(Np"')). By definition, we have K:", = V",/I,fJ", for
the maximal ideall,fJ", of V",. Note that Qj = Gal(Jt/Q(J)) = GL 2 (Z)/{±1} by
Theorem 2.40. By the description of the action of T( a) on Jt in Section 2.6, we
find that the action of (~ a~l ) with a E Z;is induced by the action ¢p I--t ¢p 0
(~ a~l) on level r(poo)-structures ¢p. By our choice, C", contains the infinity
cusp where the Tate curve Tate(q) /Z[[ql/ p "lJ has the level structure ¢can : !1p" x
Z/p"'Z -+ Tate(q)[p"'] given by ((, m) I--t ((qm/ p") E Gm(W[[ql/p"lJ)/qz. Thus
the action of a diagonal matrix (~ a~l ) brings the etale part ¢~~n = ¢canlz/p"z
to ¢~~n 0 a for a E Z;, which induces the action of a E Z;
on Voo = U", V",.
Hence we need to prove that D / I ::) Z; by a I--t (~ a~l ) for the decomposition
group D C Qj and the inertia group I C Qj of 1,fJ00/1,fJ0. In other words, we need
to show that for a diagonal matrix a E SL 2(Qp), T(a) preserves the valuation
v:Jt-+QU{oo}.
Let Dp be the decomposition subgroup in Aut(Jt) of v; that is,
Corollary 3.4 Let £ IF be a line bundle realized in the function field of TalFp .
Let s be a cusp and U be a Zariski open neighborhood of s E TalFp' If the q-
expansion at s of a section f E HO(U, 12) vanishes, then f vanishes on U.
The space Vr,cN) is the space of p-adic modular forms on rl(N). By taking
the Cartier dual of 7l.,/pa7l., '-+ E(pa], we may regard f E Vm,a as a function
of (E, ¢p : /-Lp'" '-+ E(pa], ¢N) satisfying the following conditions similar to
(Gl-3) for p-adically complete W-algebras R = ~nR/pnR.
(Gpl) (E,¢p,¢N)IR ~ (E',¢~,¢'tv)IR =? f(E,¢p,¢N) = f(E',¢~,¢'tv) E R;
(G p 2) If p : R --+ R' is a continuous W-algebra homomorphism, then
(G p 3) For all level N -structures ¢ N of type r 1 (N) on the Tate curve Tate (q),
3.3.1 Axiomatic'Ireatment
Let r,dk = O(k) for the embedding of X1(N) = PrO~dN)) into the pro-
jective space. Thus r,dk is the quasi-coherent sheaf G rdN) (k) associated with
the graded piece GrdN)(k) ([GME] 1.3.3). Then r,dk = r,d®k. Computing the
genus of Xl(N), the Riemann-Roch theorem tells us that r,dk is very ample if
k 2': 3 (see [GME] Proposition 2.1.4 and Theorem 3.1.2). Therefore r,dk is the
pull back of O( k) of the target projective space.
Let (E,cPN'w) be the universal elliptic curve over Yl(N). For each triple
(E, ¢N, w) defined over R (called a test object), we have a unique morphism [ :
Spec(R) -t Y1(N) such that [*(E,cPN'w) = (E,¢N,W). For each section J E
HO(Y1(N),r,dk), we define [*J = J(E,¢N,W)w®k. The function (E,¢N,W)-t
J(E,¢N,W) satisfies (GO-2). The condition (G3) ensures that J extends to
Xl (N). This shows
Vm,CXJ[k] = HO(Sm,r"i)
and V[k] = HO(SCXJ/w,i:'d. k I8i 1I'p) = HO(Soo/w,i:'d. k ) I8i 1I'p, (3.2)
where 1I'p = r!Jp/Zp. The last identity follows from the affineness of So. When
we generalize this point to classical groups bigger than SL(2)/Q, the affineness
is no longer true; so, we need to work out this point (the base change property)
carefully: how to bring 1I'p inside the cohomology group. In the elliptic modular
case, simply by the affineness of So, (3.2) shows that V[k] is p-divisible, and
its direct summand eV[k] is also p-divisible.
We consider the following condition:
(F) corank w eV[k] = rankw Hom(eV[k], 1I'p) is finite for an integer k.
In practice, this condition is often proven by showing
where
The left-hand side (LHS) of (3.3) is p-divisible, since So is affine. The right-
hand side (RHS) is of finite corank since Xl(N) is projective. Thus eV[k] is
p-divisible of finite corank.
Decompose Z; = r T x Ll for a p-profinite group r T and a prime-to-p
finite group Ll. For simplicity suppose that p > 2. Then rT is isomorphic to
Zp and for its generator " we have W[[rTlJ ~ W[[XlJ = A via, r l 1 + X
(i.e.,,S ~ (1 + X)S = L~o G)XJ), and W[[Z;lJ = A[Ll]. Let vord be the
Pontryagin dual module of eV.
See [FAN] Chapters 1 and 3 for generality on topological groups (in-
cluding Pontryagin duality and profinite groups). My books, [MFG] Chap-
ter 2 and [LFE] Chapter 8, also have some short exposition on this sub-
ject. The Pontryagin dual for p-torsion modules X (equipped with discrete
topology) is given by X* Hom con t(X,1I'p) for 1I'p = r!Jp/Zp with dis-
crete topology. Writing X = ~'EIX, for finite submodules X" we have
88 3 Modular Curves
In the middle equality, we have assumed (3.3). Here the subscript or super-
script ord indicates the image of e.
Write Z = Hom(Ll, WX). Note here that each element X E Z is induced
by (infinitely many) positive k E X(T). Decompose vo rd by the character of
Ll as follows,
Vo rd = EB vord[X]·
xELl
3.3 p-Ordinary Elliptic Modular Forms 89
where s = s(X) = corankweV[k]. If (F) holds for one k, it holds for all
K, E X(T) inducing X, and 7r has to be an isomorphism for the following reason:
The number s is the minimum number of generators of v ord [X]0 A,I< W over W.
We know that this module is W-free, because its dual V[K,] is p-divisible; so, it
is free ofrank s. The morphism 7r induces an isomorphism modulo (1+X) - "(I<
for all K, inducing X. Then Ker(7r) C nl< Ker(7r mod (1 + X - "(1<)) = 0, and
we get
Theorem 3.6 Suppose that (F) holds for one k. Then v ord is A-free of finite
rank, and if (3.3) holds for K, E X(T) with K,I~ = kl~, then
The sum Lc in the above definition has to be understood as a trace from R'
to R (see (4.18)) if R is not a Q-algebra (this subtle point is discussed in detail
later in Section 8.3.1). From the definition, as long as the prime p is invertible
in the base ring R, we can verify the conditions (GO-3) (or (G p l-3)) for
fIU(p). Computing the q-expansion, we know a(n, fIU(p)) = a(np, f). Then
by the q-expansion principle, the operator U(p) is well-defined p-integrally
over W even if the definition of the operator involves division by the prime p.
Since £ == 1 mod p, we confirm (C). Let
= EB Ck(r1(N); R).
00
Cr,(N)(R)
k=O
if k ?: 3,
if k = 2.
Similarly, if we write vg~~p for the subspace of cusp forms in v ord and write
v
Vc':s~ for its Pontryagin dual (i.e., the cuspidal quotient of ord ), the above
v
result is still valid replacing ord and cord
k v
by cusp
ord and Bard
k'
r
Recall the decomposition T(7!..p) = 7!..; = T x L1 for a finite group L1 and
rT ~ 7!..p- We have chosen a generator "I E r T so that 7!..p :3 s H "Is E r T
induces the isomorphism 7!..p ~ r T and A = W[[rTJJ ~ W[[X]] by "I +-+ 1 + X.
Let a( n) : v ord --+ 1I'p be the linear map associating with f its coefficient
of qn in the q-expansion; so, a(n) is in the dual v ord . We now consider
3.3 p-Ordinary Elliptic Modular Forms 91
L
00
By the theory of primitive forms ([MFM] Theorem 4.6.17; see also [H85] Sec-
tion 3), this inclusion is surjective if k ::::: 3. We leave the verification of asser-
tion (3) to the reader. 0
For a given element tJ5 E C(X; A), we get a family of p-ordinary elliptic modular
forms {tJ5( -yk - 1) h>2 whose q-expansion coefficients depend (p-adic) analyt-
ically on the weight k E Homcont(T(Zp),Z;). Each element tJ5 E C(X;A)
is called a p-ordinary A -adic form, and {tJ5(rk - 1) h?:2 is called the p-adic
analytic family of modular forms associated with a A-adic form tJ5.
There is a version of this for ro(N) and cusp forms. For k ::::: 2, we have
1. F(ro(N), x; A) is A-free of finite rank;
2. F(ro(N), X; A) 0A,k W ~ Fkrd(ro(Np), xw-k; W);
3. the above identification is induced by tJ5 f--} Ln a( n, tJ5) ("(k -1 )qn E W[[q]] ,
where X is a character modulo Np and F =C and S (see [GME] Chapter 3
for a proof).
92 3 Modular Curves
We describe how to view A-adic forms as p-adic modular forms defined over
A. Once this is done, we can evaluate A-adic forms at elliptic curves, which
gives us a convenient method of constructing and analyzing p-adic L-functions.
Then, we give a short account of the A-adic Eisenstein series and examples
of A-adic L-functions. All the arguments presented here can be generalized to
the Hilbert modular case (see Chapter 4), the Siegel-Hilbert modular case,
and the quasi-split unitary case (see Chapter 8).
For simplicity, we assume that p > 2 and only consider the A-adic forms
of level poo; thus, we have N = 1, and rT = 1 + pZp is torsion-free. Let
A = Zp[[TlJ. In the previous section, we have introduced the space G(X; A) of
p-ordinary A-adic forms, which is a free A-module of finite rank with
for all k ;::: 2. Here k : A --+ Zp is the evaluation at "(k - 1 of the power series.
If we identify A with the Iwasawa algebra Zp[[rTll by sending the generator
"( of rT to 1 + T, k is induced by the character rT :3 z H zk E 71.,;.
We write G(A) for the A-module made of formal q-expansions
such that tP("(k-1) E V[kl for infinitely many k. We have EBx G(X; A) c G(A),
where X runs over (actually even) powers of Teichmiiller characters.
We now consider the space of p-adic modular forms VI A over A of level pDQ.
In other words, we make a base change Tm,nlZ p to Tm,nlA = Tm,nlZ p xZp A
and consider p-adic modular forms over A. The functions in VIA = V®zpA
classify couples (E, ¢ : J.lP= Y E[POO]) I R defined over p-adic A-algebras R.
This J E VIA is a functorial rule assigning the value J(E,¢) E R for each
couple (E, ¢)/R as above.
This space has two A-module structures: one coming from the base ring A
and another coming from the action of Gal(Tm,oo/Sm) = 71.,; by the diamond
operators (z). Let v : rT --+ AX be the universal character given by v(z) =
[zl E rT. Then we can define
G(A) = {J E VIAIJI(z) = v(z)J Vz E rT}' (3.5)
Each q> E G(A) has a q-expansion at 00: q>(T, q) = Ln>o a(n; q>)(T)qn. By
definition, we have a natural map VI A 0 A,s Zp --+ Vlzp for each s : A --+ Zp
taking tP(T) to q>("(s - 1) for s E Zp. Here the tensor product is taken using
3.4 Elliptic il-Adic Forms and p-Adic L-functions 93
E9 G(X; A) ~ e(G(A))
x
wean) = cJ>(E, ¢ Kl
f( E ,w ) = f(E,[lk - 1) EWe ffl'l
(EQ1)
[lk "",.
p p
In this subsection, we describe very briefly some p-adic L-functions whose con-
struction stems from the theory of A-adic forms. More general and thorough
discussion can be found in [LFE] Chapters 7 and 10 and [SGL].
For simplicity, we assume that p > 2 and only consider the A-adic Eisen-
stein series of level poo. Let us fix an even power X = w a of the Teichmiiller
character. For simplicity, we choose a¢O mod p - 1; so, X is nontrivial. We
consider the Kubota-Leopoldt p-adic L-function -~Lp(1- s, X) = ao("'(8 - 1)
("'( = 1 + p) with ao E 71,p[[T]] (e.g., [LFE] Sections 3.5 and 4.4). We have, for
positive integers k,
~E (E ) = L(k, Ak)
2 k ,w [2k'
00
Thus !Jt(rm - l)Ex'l/J-' (r-m(1 + T) - 1) E G(X; Zp[[T]]) and hence the p-adic
L-value Lp(rm _l"k - 1) is the coefficient of w(rm - l)Ek- m in tf>(rk - 1)
for a suitable Eisenstein series Ek-m of weight k - m. As is shown by Shimura
[Sh7], this coefficient can be computed by the Rankin product value
over a base category Co (cf. [SCA] VI) means that we have a specified covariant
functor (fiber functor) C ---t Co. An object of AF is a triple (A, A, L)/S, where
(rm1) L = LA : 0 y End(A/s) is an embedding of algebras taking the
identity to the identity;
(rm2) A is an O-linear symmetric isogeny A : A ---t t A induced by an ample
line bundle fiber-by-fiber geometrically (see [GIT] 6.2). Here A is called
symmetric if A = t A. Such a A is called a polarization of A;
(rm3) The image of LA is stable under the Rosati involution on the endo-
morphism algebra End(A) I8Iz Q: a H a* = A-Iota 0 A;
(rm4) As 0 I8Iz Os-modules, we have an isomorphism: Lie(A) ~ 0 I8Iz Os
(B 1f* (.flA/ s) ~ ll-1 I8Iz Os with 1f : A ---t S for the absolute different II
of F) locally under the Zariski topology of S, where the sheaf Lie(A)
of Lie algebras of A (i.e., the direct image of the tangent bundle over
A/ s ) is an O-module by the action induced from L.
The fiber functor is given by AF 3 (A, A, L)/S H S E SCH.
Here is a more detailed description of the above four conditions. Either by
localizing at a rational prime or by extending the scalar to an etale extension,
any projective O-module becomes free; so, we often pretend that the isomor-
phism in (rm4) is global, although Lie(A) is required to be only a locally
free module over (0 I8Iz Os) of rank 1. Since an O-module free of finite rank
is determined by the characteristic polynomials in Os[t] of multiplication by
a E 0, (rm4) is also equivalent to
(det) The characteristic polynomial of each a E 0 on Lie(A) over Os is
given by the image ofI1a(t-o-(a)) E ;E[t] in Os[t], where (j runs over
all embeddings of F into ij.
A morphism f : (A, A, L)/S ---t (A', A', L')/s of AF is an O-linear morphism
f : A/ s ---t A is of abelian schemes over S with A = 1* A' := t fOA' 0 f.
A polarization is an O-linear isogeny A = AL : A ---t t A induced etale
locally by a symmetric line bundle L / A; that is, we have (-1) * L ~ Land
AL(X) = T;(L) 181 L- 1 , where Tx(Y) = x + y. This definition is equivalent to
Mumford's definition ([CIT] 6.3) requiring that A be induced by a symmetric
line bundle LS/As for each geometric point s E S (see [DAV] 1.1.6). Indeed, for
the universal Poincare bundle P on A Xs t A/A, 2A is associated with (1 x A)*P
globally over S (see [GIT] Proposition 6.10).
The morphism A is an isogeny if and only if L is ample (cf. [ABV] Section
6). We have AUg)U = AL + AU, and the monoid of Hom(A, t A) generated by
polarizations forms a cone P(A). If A : A ---t t A is a polarization, Ker(A) is
given by A[c 1 ] for an integral ideal c- 1 i- 0, because Ker(A) is self-dual under
Cartier duality. Then A induces t A ~ A 181 c for the following reason. Tensoring
A over 0 with the exact sequence 0 ---t 0 ---t c ---t c/O ---t 0, by the divisibility
of A, we get another exact sequence 0 ---t Torl(c/O, A) ---t A ---t A 1810 c ---t 0
of fppf abelian sheaves (see Section 8.2.1 for fppf abelian sheaves). Since
o is a Dedekind domain, we have c/O ~ O/c-1, A[c- 1 ] = Torl(c/O,A)
100 4 Hilbert Modular Varieties
Since A[PjO ~ /L~ over k(s) for 9 = dimAs, it is easy to see that the above
morphism induces tAs[Pjet 0lF p k(s) ~ HO(As, DAB/k(s)). By duality and po-
larization, we get As [pjet 0lF p k(s) ~ Lie(As). This shows that
The fact (4.1) shows that Lie(A)/s and Os 0 T(A[POOjet) for the locally
constant sheaf T(A[poojet)/s = ~nA[pnjet have the same stalk everywhere
if Aj S is ordinary at every point over Sand p is locally nilpotent over S, and
hence they are isomorphic.
Corollary 4.2 If p is locally nilpotent over S and an abelian scheme A/ s is
ordinary, we have a canonical isomorphism Lie(A)/s ~ Os 0z T(A[poojet).
If furthermore, pn = 0 over S, we have Lie(A)/s ~ Os 0z A[pnjet, where
A[pnjet is the maximal etale quotient of A[pnj.
102 4 Hilbert Modular Varieties
pairing eN : A[N] x t A[N] -+ /IN (the so-called Weil pairing; see (PI) in 8.2.3)
composed with the polarization A : A -+ t A agrees via ¢ with the natural
pairing \-, .) N : CO- 1 ®7/,/lN )ffi(O®7/,N- 1 ZjZ) -+ /IN given by (a ® (, b ® m) r-+
e (TrF/IQ!(ab)) em, where e(x) = exp(2nix). We often simply write (A, A) for
(A, A, ~ : 0 Y End(A)), dropping ~ as we agree to have ~ always in this
chapter. A morphism f : (A, A, ¢) / S -+ (A', A', ¢') / S is induced by a morphism
f : (A, A) -+ (A', A') in AF with the additional requirement that ¢' = f 0 ¢.
We fix a polarization ideal c, and assume that A is a c-polarization. Then
for a given c, the Hilbert-Blumenthal scheme 9J1( c, r(N)) /7/,[11J represents the
following functor Cr(N) = Cc,r(N) : Z[k]-SCH -+ SETS given by
where as before H is the set of isomorphism classes of the objects * inside the
brackets. When N 2: 3, as we show in Remark 6.21, this functor is rep-
resentable by a quasi-projective scheme over Z[k] (which is smooth over
Z[C, Nd(~')C] for 0 < CEQ with (C) = Q n c, as we show later in Sec-
tion 7.1), and for small N ::; 2, we have at least a coarse moduli scheme (cf.
[GME] 2.3.2).
Let 91 be a nonzero ideal of O. Abusing the notation, we write /l'Yl/7/, for
the locally free group scheme (of finite rank) given by the 91-torsion points
of G m ®7/, 0- 1 : /l'Yl(R) = {x E Gm(R) ®7/, ()-II91x = a}. If 91 is generated by
a positive integer N, we have /l(N) = /IN ®7/, ()-1. We consider the moduli
problem classifying test objects (A, A, i : /l'Yl Y A[N]) / s of level rl (91) in
place of the level r(N)-structure. The classifying functor cr{('Yl) = cc,r{('Yl) :
Z[k]-SCH -+ SETS for an integer N with 91 n Z = (N) is given by
where i'Yl : /l'Yl Y A[N] is the inclusion of locally free group schemes. If
each test object (A, A, i'Yl) / s is rigid without nontrivial automorphism, the
functor is representable by a quasi-projective scheme 9J1( c, rl (91)) /7/,[ 11 J' which
is smooth over Z[Nd(F)]. Thus the functor cr,J('Yl) is representable if 91 is deep
enough. The coarse moduli 9J1( c, rl (91)) /7/,[ 11 J exist for all 91.
Let O~ = r;n Ox. We may ease the polarization condition, that is,
replace a polarization A by a polarization class 0 ~ A = {EA lEE 0 ~ }. Easing
the polarization requirement allows us to change morphisms of our category
from O-linear isomorphisms to O-linear isomorphisms up to unit factors (we
later study categories up to a larger class of isogenies in Section 4.2). For a
given O~A of A, we say that O~A is defined over S if we find an ample line
bundle, locally under Zariski topology on S, giving rise to an element in O~A
(cf. [ABV] Section 8). Consider the following functor Crd'Yl) : SCH/7/,[I/NJ -+
SETS (of level rl(91)) for an integer N with 91 n Z = (N) given by
via ¢ when we deal with level r(N)-structure. The differential w can be recov-
ered by L: A(<C) = (CI/.L: so that w = L*du, where u = (UU)UEI is the variable
on (CI. Conversely, for a given test object (A, A) in AF with c-polarization A,
is a lattice in (CI, and the polarization A : t A ~ A i3) c induces .L: 1\ .L: ~ c* (see
(Pl-3) in Section 8.2.3).
Using this equivalence, we can relate our geometric definition of Hilbert
modular forms with the classical analytic definition. Define 3 by the product
of I copies of the upper half complex plane SJ. By (~~) H(~: :7-f~~:) u'
we have G 1(JR.)/Ci ~ 3 for n
= S02(F i3)Q JR.) and i = (0, ... ,0) E 3
(which is the origin 0 of 3). We regard 3 C F i3)Q (C = (CI made up of
z = (zu )UEI with a totally positive imaginary part.
We identify /1N with N- 1'lL/'lL by exp(21l'im) +-t (m mod'lL). This iden-
tification induces /1(N) = /1N i3) ()-1 ~ (N- 1 'lL/'lL) i3) ()-1 = (N)* /0*, which
further induces /1m ~ 1)1*/0*. Choose two ideals a and b prime to 1)1 with
ab- 1 = c. For each z E 3, we define
with iz : /1m = 1)1* /0* --t (CI /.L: z given by iz(a mod 0*) = (-21l'0a +.L:z)
and ¢m,z : 1)1* /0* X 1)1-1/0 --t (CI / C given by, for J 1 = (~ (/ ),
then 3 is identified with the set X+ of all conjugates under G(JR.)+ of the group
homomorphism ho : C x 3 a + bi r-t C!\ ~) E G (JR.) +. The identification is
given by 3 3 g(i) r-t 9 . hog- l E X+ (i = (R, ... , R) E 3). For each
h E X+ (corresponding to Z E 3), LJlI. 3 v r-t h( a + bi)v gives the complex
structure we want. For the alternating pairing (x, y) = tx . Jly of V = F2
(column vectors), Hh(X, y) = (x, h(i)y) gives a positive definite Hermitian
form on LJlI. = VJlI. so that (-,.) gives rise to the Riemann form on LJlI./ L =
C 1 / L z . The natural action v r-t av of a E G(JR.)+ on LJlI. induces the right
action: z r-t a-l(z) on 3. Indeed, we have
for the adjoint involution i, since a' = III.taJI = det(a)a- l . In other words,
the natural left action of G on V induces the right action z r-t a' (z) = a ~ I (z)
on 3. This is the explanation of why we need to have (a, b) r-t -a + bz.
If a = (~~) E G(Qlh, the multiplication by (-cz + a) E Fr!j induces an
isogeny
I xc-+( -cz+a)x I
(C /La.-1(z),A,¢'Jl,cc1(z)) ) (C /C,det(a)A,¢'Jl,z oa). (4.2)
This shows that the action of a E G(QI)+ sends the complex point of the
projective limit SJJ1(C) = 0.!l N SJJ1(c,T(N))(C) represented by z E 3 to that of
a-l(z). The above isogeny is an isomorphism if and only if a E r(m; a, b) for
the arithmetic subgroup r(m; a, b) of G(QI) defined below.
For four fractional ideals a2 ] C F, we write symbolically (~;; ~;;) for the 0-
lattice in M2 (F) made up of matrices whose (i, j)-entry is in a2 ] for i, j = 1,2.
Define three types of congruence subgroups of G(QI)+ by
rl(m'ab)={(ab)E(
, , c d 0
'Jlabi1 (abl*)lad-bcEO
0
x
+, a-1Em},
rll(m; a, b) = rl(m; a, b) n SL 2 (F), (4.3)
r(m; a, b) = {( ~~) E rl(m; a, b* E m(ab)*}.
Since det(rl(m; a, b)) = 1, the condition a-I Em for rl(m; a, b) automati-
cally implies d - 1 E m.
For the moment, we deal with rl(m)-structures. We let 9 = (g,,) E G(JR.)+
act on 3 by linear fractional transformation of g" on each component z". It
is easy to verify by (4.2), similarly to Theorem 2.39, that
(Ism) (Lz, Az , iz) ~ (Lw, Aw, iw) -¢=? W = r(z) for r E rl(m; a, b).
Thus the complex points of the coarse moduli scheme SJJ1( c; rl (m)) are given
by rl(m; a, b)\3. However, it looks odd that we have an extra dependence on
the pair (a, b) not just on c and m. Actually, as long as ab- l = c, the complex
manifolds rl (m; a, b) \3 are isomorphic for all choices of (a, b). To see this
fact, let us analyze the equivalence classes of cusps of r(O; 0, C~l). Cusps of
4.1 Hilbert-Blumenthal Moduli 107
3 = SjI are given by pI (F) = F U {oo}. Recall that G(Q)+ is the subgroup
of G(Q) made up of matrices with totally positive determinants.
The groups G 1 (Q) and G(Q)+ act on Pl(F) by linear fractional trans-
formation (or equivalently by matrix multiplication on homogeneous coor-
dinates). The stabilizer of the infinity cusp is the subgroup BeG of up-
per triangular matrices. Thus Pl(F) ~ G 1 (Q)/B 1 (Q) = G(Q)+/B(Q)+
for B(Q)+ = B(Q) n G(Q)+ and Bl = B n G 1 . For an arithmetic sub-
group r c G 1 (Q), its closure l' in G 1(A,.(00)) is an open compact sub-
group, and r = l' n G 1 (Q). Two cusps sand s' are equivalent under r
if ,(s) = s' for some, E r. Thus, by the strong approximation theorem,
G 1 (Q) is dense in G 1 (A(00)), and the equivalent classes of cusps are in bi-
jection with r\G 1 (Q)/B 1 (Q) = 1'\G 1 (A(00))/B 1 (Q). We now make this set
explicit for r = r(O; 0, C 1 ). For 9 E G 1 (Q), we write £(g) for the left
column vector of g. Then writing £(g) = t(a, c), define ilc(g) = (cell + aO)
as a fractional ideal of F. Then £(g)b = £(gb) for b = (g:) E B 1 (Q). If
we choose another base, ilc(g) = c'ell + a'O, we find, E r(O; 0, e- 1 ) with
, . £(g) = t(a', c') = £hg). Thus by 9 H ilc(g), we find an isomorphism
r(0;0,c 1 )\G 1 (Q)/B 1 (Q) ~ elF for the class group elF of F. In other
words, the cusps of r( 0; 0, e- 1 ) are in bijection with the class group of F. We
can extend ilc to G 1 (A (00)) by ilc (g) = (cev + aO) n F, which gives rise to the
identity 1'(0; 0, e- 1 )\G 1 (A (00))/ Bl (Q) ~ elF. In particular, ilc (g)o = ilc (ga)
for a E BdA(oo)) with riO n F = o. By the Iwasawa decomposition (e.g.,
[EPEl Lemma 9.2), we have G(A(oo)) = 1'(0; 0, e- 1 )B 1 (A(00)); so, each cusp
of r(0;0,e- 1 ) is represented by b E B 1 (A(00)). Thus we have an association
of cusps s to ideals of F given by choosing b E B 1 (A (00)) representing sand
sending s to the ideal class of il c (b).
There is another description of cusps of r( 0; 0, e- 1 ). The set of pairs
(0, b) with ob- 1 = e is in bijection with the set of cusps of r( 0; 0, e- 1 ) in the
following way. Since ob- 1 is fixed to be e, we find (0, b) = (0, oe- 1) = 0(0, c 1 ).
Thus a standard choice of a cusp is (0, C 1 ), which we call the infinity cusp
of 9)1(e, 1), and 0 corresponds to a cusp s if ilc(b) = 0 for b E B 1 (A(00))
representing s.
For one r of these level structures rl(91), n(91), and r(N), the moduli
space 9)1 ( e, r) has a finite number of cusps (depending on the choice of a level
r-structure on the Tate AVRM Tatea,b(q) we define in Section 4.1.5) over
a given cusp s represented by (0, b). When rl (91) e r e r 1 (91), we have
a canonical level structure i can ,')1 as we specify in the following subsection.
When r = r(N), we have a canonical level structure ¢can,N over an extension
ring Z[d(i)N]((N- 1 0b)) of the canonical ring of definition Z[d(i)Nl((ob)) of
Tatea,b(q). We simply write (0, b) for the cusp of 9)1(e, r) with this canonical
choice of the level structure. Write
108 4 Hilhert Modular Varieties
For 'Y E S L2 (F) with 'Y ( abo ) = (~'1 ), 'Y brings the cusp corresponding to the
pair (0, b) to the standard cusp (0, c- 1). In particular, we have
Thus we find rl (c, sn) \3 ~ rl (sn; 0, b) \3 for the above choice of 'Y. For each
'Y
ideal 0, (0, oc- 1) gives another cusp. The two cusps (0, oe 1) and (5, sc- 1)
°
are equivalent under r1 (c, sn) if = as for an element a E F X with a == 1
mod sn in F~, where F'Jl = TIiI'Jl FI for prime factors [ of sn.
Returning to the coarse moduli 9J1(c,rl(sn)) over Z[d(f)N] (snnz = (N))
for the functor [c,rf('Jl) : Z[d(f)N ]-SCH --7 SETS,
Put
CEClt CECltCl)'l)
(4.7)
G(Q)\G(A)jSt(>J1)Z(JR)Co = U VJ1(c,rl(>J1))(<C),
CECltCl)'l)
G(Q)\G(A)jS(N)Z(JR)Co = U VJ1(c,r(N))(<C),
CECltCl)'l)
at the cusp (a, b, ¢N). The minimal compactification M* (c, r(N)) of the open
modulus space fJJt(c,r(N)) if it exists is characterized by the property that
it is covered by any smooth compactification of fJJt( c, r(N)) (having a divisor
of normal crossing at the cusps) so that the covering map induces an isomor-
phism in the interior fJJt(c, r(N)). Thus we convince ourselves that the formal
stalk of the minimal compactification at the cusp s = (a, b, ¢N) is given by
HO(T(71.)(N), Rs(N)), where
Rs(N) = {ao + ~
t.EN-1abnc
at.qt.lat. E 7l.[~,J1,N]}
on which f. E T(71.)(N) acts by
Here the series ao + Lt. at.qt. is a formal series whose coefficients at. could be
nonzero for all ~ E (C n N- 1 ab)
U {O}. Thus Rs (N) is the completion of the
monoid ring Rs(N) of the semi-group (N- 1 ab) n C under the adic topology
of the augmentation ideal.
Let a V be the dual cone a V = {x E FooITrF/IQi(xa) :::: O}. Then C =
ncr a v. To make our conviction feasible, we try to describe the complicated
ring Rs(N) using more reasonable rings. We consider the monoid ring Rcr(N)
of a V n N- 1 ab. Thus
but the sum ao + I;~ a~q~ is supposed to be finite (so, a~ = 0 for almost
all O. For each (J as above, by (PC3), each cone (J is generated by a part
of a Z-basis of N (ab) *, and hence (J v is generated by a part of a Z- basis
of N- I ab. Thus we may assume that (N (ab) *) n (J is generated over Z by
t I , ... , tr (0 < T :; [F: Q] = g). We have a base 6, ... ,c,g of N-Iab so that
Tr(t,c,)) = 6,) for 1 :; i :; j :; T and Tr(t,c,r+k) = 0 for k > 0, and each
C, EN-lab n (Jv can be uniquely written as C, = I;, mic', with mj E Z and
m) :::: 0 if j :; T. Writing T) = q~J, we find
(4.10)
modulo m fiber M(c, r(N)) 0Z[J.LN] IF' for the maximal ideal meW. By the
projectivity and smoothness of MN over Z [Nd(F) ' f.1N], MN/F is irreducible
(Zariski's connectedness theorem [ALG] III.11.3).
We use the same cone decomposition C for all N. Then we get a finite
covering, MNI := Mc(c, r(N')) -» Mc(c, r(N)) for all multiples N' of N. As
long as N' is prime to p, this covering is tamely ramified at the cuspidal divi-
sors I:s Ds (over Z(p)), and the closure f(o; a, b)(p) in G1(A(poo)) naturally
acts on MN'. Taking the limit M2P) = ~pfNIMNI (N' running over multiples
of N prime to p), we have Gal(M(p) /MN / F ) = feN; a, b)(p). However, the full
group G1(A,(poo)) may not act on M(p).
Let O~)+ = O~) n F':, where O(p) = 0 0z Z(p). Among simplicial cones
in C/T(Z)(N), r-dimensional faces are in bijection with generic points of the
(g - r)th stratum. This is due to the following reason (see [DAV] IV.2.5). If
a cone a is spanned by n linearly independent elements WI, ... , wn in F':,
its dual cone a V is given by QVl + ... + QVg - n + Q+Wl + ... + Q+wn with
TrF/Q(v,w J ) = O. Note that
N (ab ) * = { ex E F I (6 ~) E f( N; a, b) } .
By (PC3), the choices of VI,"" v g - n and WI, ... , Wn can be made to be
generators over Z of N-l(ab). This point that w J is a part of a Z-base of
N-l(ab) assures us of the smoothness of M N . By this fact, w J is prime to p
if pf N; in other words, Wj E O~)+'
We have Ru(N) =Z [Nd(F),f.1N] [T"SJ][T,-l] with T, = qV, and Sj = qW J •
The inertia group is a module under conjugation over DN,1]n/IN,1]n for the
decomposition subgroup D N,1]n C Gal(M(p) /M N ) of TJn. If we incorporate the
data of this Galois action, we have
Theorem 4.7 Let the notation and the assumption be as above. In partic-
ular, p is a prime unramified in F /CQl. Let TJn be a generic point of the nth
stratum of the cuspidal divisor Ds of M N/ fF • Then the p-tame inertia group
I N,1]n C Gal(M(p) / M N ) of TJn is given by I N,1]n ~ N( ab)~ (>9z Z(p) (1)) as mod-
ules over the decomposition group D N,1]n) where Z(p)(l) = ~pfNJ..lN as a
Galois module.
The Galois module structure of the inertia group as in the theorem follows
from Kummer's theory described in [TFGl Section 2.
Consider the ring R[[(ab)+ll ((ab)+ = ab n F.:) made up of all formal. series:
aO + L a~q~ with a~ E R
~E(ab)+
for a given base ring R. Take a subset qub+ = {q~l~ E (ab)+} of R[[(ab)+ll
stable under multiplication, and. write R{ ab} for the localization of R[[( ab)+ II
by the set qub+. Then Z[-b,J..lN]{N- 1 ab} is the localization of the formal
stalk Rs (N) at the cusp 5 = (a, b, ¢ N) of the minimal compactification Miv.
116 4 Hilbert Modular Varieties
1
Hom(ab, Gm(Z[ N' /LN]{ab}))
1
~ Homo(b, Hom(a, Gm(Z[N' /LN]{ab} )))
because (x'rJ jyt.)n = q'rJt. jqt.'rJ = 1. Since we may identify the complex
points of the Tate AVRM Tatea,b (q)(q Iq=exp(Tr(z)) with (CX)I je(Lz) (L z =
21fH(bz + a*)) for the variable z E 3, where e : C I -+ (CX)I given by
(u.,.).,. H (exp(u.,.)), the polarization Az induces the above pairing for any z.
Thus the polarization of the universal AVRM over m(c, r(N)) induces an
isomorphism ({J : t Tatea,b(q) ~ Tateb,a(q) so that the Cartier duality pairing
t Tatea,b(q)[M] x Tatea,b (q)[M] -+ J-tM (in (Pl-3) in the proof of Theorem 8.9)
composed with ({J coincides with the above pairing (-, .) M for all M. Then writ-
ing c = ab- 1 , the natural isomorphism
Hereafter we often make the base change of our moduli spaces m, M and M*
of type r = r(N), rl (91), and r 1 (91) to the valuation ring Wand regard these
schemes as defined over W. Assume that W 0 0 = WI by w 0 a H wO"(a);
so, W contains all conjugates of 0 over Z, and p is unramified in 0 jZ. Further
assume that the polarization ideal c is prime to the residual characteristic p
of W. We fix a cone decomposition C as in (PCl-4) once and for all and write
g/M for the universal semi-AVRM with respect to C.
A little more generally, we start with a W-scheme S carrying a semi-
abelian scheme 9 with real multiplication by O. Thus over an open dense
subscheme So c S, 9 Xs So is an object of A F , and over the closed subscheme
Z = S - So, 9 is a smooth multiplicative group scheme of finite type whose
connected component is given by (Gm 0 0 (which is canonically isomorphic to
a more standard (Gm 0,,-1 because of unramifiedness of pin FjQ). Writing
the structure homomorphism of 9 as 1f : 9 -+ S, we have a vector bundle
118 4 Hilbert Modular Varieties
fo!. = 7r*[2gIS over S. The pullback action of the endomorphism 0 makes !,;Z.
into a schematic TiS-module for T = Reso/zGm. Since p is unramified in
F/«J, identifying X(T) with Z[JJ, we find fo!. = EBaEI!,;Z.a, where fo!.a is the
a-eigenspace of T. Then we define fo!.k = 18J(fo!.a)®ku for k E Z[JJ, which is an
invertible sheaf over S. The sheaf EBkEZ[I] fo!.k of graded Os-algebras is the
affine ring of the scheme M gIs = Specs(S) (see [GME] 1.5.4) representing
the following functor PgI S : S-SCH --+ SETS given by
Now we return to (Jis given by (JIM for the modulus 9Jt and its toroidal
compactification Mover Z[Nd(F) , J-lN]' By a result of Moret-Bailly ([DAV]
Proposition V.2.1), the graded algebra OM = EBJ2:oHO(M,fo!.JI) is of finite
type over Z[Nd(F) ' J-lN]' Recall the Koecher principle (4.10): HO(M,!,;Z.JI) =
HO (9Jt, fo!.J I) for all j E Z if F -I- «J (the case of F = «J is already treated). Thus
4.1 Hilbert-Blumenthal Moduli 119
--
line bundle over T m,n/ R defined at 00.
The sheaf ~oo/soo = OSoo ®zp ~nQ[Pn]o is isomorphic to the dual ~/M of
7r*Lie(Q/M) for 7r : Q --+ M, because Lie(Q[Pn]sJ = Lie(Q/srn) (see Corol-
lary 4.2). In other words, ~/M = Hom(7r*Lie(Q/M), OM) = 7rJ1Q/M is the
algebraization of the formal sheaf ~oo on Soo (which is uniquely determined
by Sl independently of the choice of E). We define ~k C ~k by the invertible
subsheaf of ~k made up of sections of ~k vanishing over the cuspidal divisor
'Dcusp = 7r- 1(M - 9J1) on M. We call ~k the sheaf of cusp forms of weight k.
M = Mc(c, r(N)) for all N > 0 (using this C) and the minimal compact-
ification M* = M* (c, r(N)) of 9Jl( c,r(N)).
Fix a positive integer N prime to p. We put S = M[il and S* = M*[il
for a lift E of the Hasse invariant as in (E). Recall Mm = M 0w Wm, Sm =
S0wWm, and S:n = S*0wWm for Wm = Wjpmw = Wjpmw. The proof of
the vertical control theorem we have given for elliptic modular forms extends
to our Hilbert modular case almost intact except for the following hypothesis
(see the second identity in (3.2)),
(Hp1)
for the sheaf ~k C ~k of cusp forms of weight k. In the elliptic modular case,
we have proven this resorting to the affineness of the open curve Sm there. In
the present case of F =1= Q, the scheme Sm is not affine; so, we need to verify
this. This is a subtle point, and this fact is only valid for the cuspidal sheaf
~k not for ~k, as we show.
Write 1f: S -+ S*. Since by definition HO(S,£) = H O(S*,1f*(£)) for any
sheaf £ / sand S* is affine, we need to verify
-- 1
1f*(~/R)x ~ HO(T(Z)(N),R[[N(abhoJ])
where (ab»o = {t; E ablt;» O}U{O} = Cnab, and f T(Z)(N) in (PC2) acts
E
(e.g., [CGP] III.6.6). This is the case for any R[U]-submodule of R[[*"(abho]l
made up of q-expansions without the constant term. Then we get
(4.12)
H~rd(S'~k) ~ H~rd(M'~k)'
where H~rd(S, 7) = eHO(S, 7) and H~rd(M, 7) = eO HO(M, 7).
Assertion (3) in the theorem is (the Pontryagin dual version of) the "precise
control" described in Section 1.2.
Proof. As we remarked, assertion (1) follows from a more general result which
is given in Chapter 8 (see Corollary 8.4) through the theory of false modular
forms. We do not prove the assertion (1) here. It is known that
From this, the assertions (2), (3) and (4) follow by the same argument as in
the elliptic modular case (see Section 3.3). The argument giving this exact
bound (4.14) is rather involved; so, we allocate the exposition to its own sec-
tion, Section 4.3 (see Theorem 4.37). Here we just prove the boundedness of
dimH~rd(M'~k) (independent of k) assuming the existence of a Hecke equiv-
ariant inclusion map G k (c, T; q '-t Hg (T, L( k*; q) (the generalized Eichler-
Shimura map for k* = k - 21; see Theorems 4.36 and 5.13) for the arithmetic
subgroup T = T(c, N) of SL 2(F) = GI(Q) giving 9)1(c, T(N))(q = T\3. By
the definition (4.3), T c SL2(O(p)) for O(p) = Op n F. Here g = [F : Q],
but actually, the following argument is valid for any integer g ::::: O. Note that
Gk(C, T; W) 0w C = Gk(c,T; q and Gk(c, T; W) = Gk(C, T; W) 0w W (flat
base change; see [GME] Lemma 1.10.2), because these are spaces of global
sections of the invertible sheaf ~k. By the long exact sequence attached to the
short one: L(k*;W) -4 L(k*;W) ---7> L(k*;lF) for IF = W/pW ([MFG] Corol-
lary 4.28), we find that Hg(T, L(k*; W)) 0w IF injects into Hg(T, L(k*; IF));
so, we need to bound dimIFH~rd(T,L(k*;lF)). We may identify L(k*;R) with
the space of R-integral polynomials in (Xo-, Yo- )o-EI homogeneous of degree
k; = ko- - 2 for each pair (Xo-, Yo-). The action of l' E M2(Op) on L(k*; W) is
given by <P((Xo-, Yo-)t(J(1')'), where xx' = det(x). Let
gives a morphism of r'-modules L(k*; IF) ~ IF, where IF is the constant r'-
module with coefficients in IF. Similarly, IF ~ L(k*;lF) given by IF '3 art ayk*
gives a morphism of r' -modules. We get two morphisms of cohomology groups:
for the restriction Resr/T' and the transfer map Trr/T' (cf. [MFG] 4.3.1),
we find that Trr/T' 0 IT 0 t 0 Resr/T' = T(p). This shows that to ReSr/rf is
injective over H;'rd(r, L(k*; IF)). Since Hg(r', IF) is finite-dimensional (because
9J1(c, r(N)) is homologically equivalent to a finite simplicial complex; e.g.,
[H79] Section 3), the dimension of H;'rd(r',L(k*;lF)) is bounded by that of
Hg(r',lF) independently of k. 0
Recall open compact subgroups I\(91), rl(91), and r(91) in G(Z) = GL 2 (O)
defined in (4.5). As long as the moduli problem of type rl (91) is repre-
sentable by a quasi-projective smooth scheme 9J1( c, rl (91)) /Z(p) , for its smooth
projective toroidal compactification M (c, rl (91)) /Z(p) , we get the same as-
sertions as in Theorem 4.9 replacing the pair (9J1(c, r(N)), M(c, r(N))) by
(9J1(c, rl(91)), M(c, rl(91))) for weights k with k E (71.,·1+271.,[1]) (i.e., ku == kT
mod 2 for all (1, T : F '--+ Q). This parity condition is necessary to define
f f-t fk/2 for totally positive units f E T(Z) (since fI = 1 for such units).
We only need the representability of the rlU m 91)-moduli problem for a suf-
ficiently large m and a prime ( to have a result similar to Theorem 4.9 as we
describe this fact now.
To formulate the result, we consider the moduli problem of type r for r =
rl(91) and r 1 (91), and accordingly, when representable, we define D~uSP, V,
V, V~~~p*, and V~~~p with the level r-structure. Even if the modulus problem
is not representable, we have the coarse moduli scheme 9J1( c, r) /W by making
a geometric quotient (cf. [GME] Proposition 1.8.4) of 9J1( c, r(N)) (over W)
124 4 Hilbert Modular Varieties
Proof. We deduce the theorem from Theorem 4.9. First suppose that W con-
tains a primitive Nth root of unity (. For the differental idele d E FJ: with
d(D) = 1 and dO = D, we identify ()-l/N()-l = ()-l ® Z/NZ with O/NO by
x H dx. By m H (m, we have an isomorphism O/NO ~ ()-l®f.LN well-defined
over W. Then 9J1(c, r(N)) over W is the moduli scheme classifying test objects
(A, A, ¢: (O/NO)2 ~ A[N])/R with (¢(l, 0), ¢(O, l))N = (for W-algebras R.
Thus we have a natural action of, E SL 2(O/NO) on 9J1(c,r(N)) induced by
the functorial action on the test objects (A, A, ¢) H (A, A, ¢ 0 I)' The coarse
moduli scheme 9J1(c,rl(N))/w of rhN)-type is given by the geometric quo-
tient 9J1( c, r(N)) /U (cf. [GME] Proposition 1.8.4) by the constant subgroup of
order prime to p, U = {(51) lu EO/NO} c SL 2(O/NO). By the definition
of the quotient, we have
of level rl (1)1) does not have any automorphism other than the identity, be-
cause then the quotient process making 9J1( c, rl (1)1)) out of 9J1( c, r(N)) /w at
the functor side and the moduli side matches. Suppose 1)1 C;; 0, and choose a
prime factor [of 1)1. By the positivity of Rosati involution ([ABV] Section 21),
the automorphism group of a test object (A,),) is finite. By classification of
the endomorphism algebra of a given abelian variety (cf. [ABV] Application
I in Section 21), the size 1Aut(A, )')1 is bounded independently of the test
object. Thus for a sufficiently large power [m, 9J1( c, rl ([ml)1)) is representable
over W, because then each test object of level rl ([ml)1) does not have any
automorphism other than the identity map. Thus the theorem is valid for
level rl ([ml)1). Note that 9J1( c, rl ([ml)1)) is a faithfully fiat finite covering of
9J1( c, rl (1)1)) of degree d prime to p (actually d is a power of N ([)). Since
the restriction composed with the trace map is the multiplication by the de-
gree d, we can realize all objects D~USP(c, rl(I)1)), V(c, rl(I)1)), V(c, rl(I)1)),
vg~~p*(c, rl(I)1)), and vg~~p(c, rl(I)1)) as a W[[T(Zp)]]-direct summand of the
corresponding objects of level rl ([ml)1). By this, we conclude the desired as-
sertions if 1)1 C;; O.
We now suppose that 1)1 = O. Since p is unramified in F/Q and p 2: 5,
Gal(F[p,p]/ F) S=: (Z/pZ) x by sending the Frobenius element at a prime [t p to
N ([) mod p (class field theory). Thus we find a prime [ of F such that N ([)
mod p generates the cyclic group (Z/pZY. In particular p t (N([) ± 1). Then
the modulus variety 9J1( c, r([)) for f is a faithfully fiat finite covering of degree
prime to p over 9J1( c, r(l) ) (the degree is a factor of N ([) (N ([) -1) (N([)2 -1)).
Then by the same argument as above, we get the result for 1)1 = O. D
We now interpret the result obtained in the previous section in terms of auto-
morphic forms on PC = Reso/zPCL(2). For simplicity, we hereafter assume
that k is even, since the general case has been explained in [SGL] Chapter 2.
We consider a test object (A,O~)"i)/R of level r 1 (1)1). For any ideal 3
prime to Np ((N) = I)1nZ), we make a quotient AI = A/A[3] = A@03- 1 ;
thus, A = A I @3. Then AI[I)1] S=: A[I)1] canonically; so, i induces a level r 1 (1)1)-
structure i l into AI. Let P(A) = {A E Hom(A, tA)lt), = ),} and P+(A) C P(A)
be the subset made of polarizations. Then we have an isomorphism), : P S=: c
of O-modules taking P+ onto the subset c+ of totally positive elements of c.
Dualizing the exact sequence 0 --+ A[3] --+ A --+ AI --+ 0, we get another exact
sequence 0 --+ tAl [3] --+ tAl --+ t A --+ 0, because tAl [3] is the Cartier dual of A[3]
(see the proof of Theorem 8.9). This shows tAl S=: t A @3 S=: A @ C3 S=: AI @ C3 2 ;
so, ), induces ),1 : P(AI)+ S=: (q 2 h. Thus (A, O~)" i) f--+ (AI, O~),I, i/) induces
(3) : V(C3 2,rl(I)1)) S=: V(c,r1 (1)1)) by fl(3)(A,0~)"i) = f(AI,O~),I,i/). The
operator (3) is called the diamond operator. We identify V (c, r 1 (1)1)) and
V(C3 2,rl(I)1)) (resp. V(C,rl(I)1)) and V(C3 2 ,rl(I)1))) by (3), and write it as
Vc (resp. V c) whose ordinary and cuspidal part is written as Vc~s~,c (resp.
126 4 Hilbert Modular Varieties
V~~:p,J. Thus these spaces only depend on the strict ideal class of c modulo
square ideal classes. We then define
(4.15)
where c runs over strict equivalence classes of ideals modulo square classes;
thus, it runs over the group cltl(Clt)2, where Clt is the strict ideal class
group. Recall that
u
CEC l il(C l i)2
where c runs over a complete representative set for cltl( Clt)2 in FI; Foo =
F 0Q1 JR.; PGLt(Foo ) is the identity-connected component of PGL 2(Foo ),
and 1\(1)1) is the image of i\(I)1) in PGL 2(Fl oo )). Thus we may regard
V~~:p*(i\(I)1)) as the W-dual of the space ofp-adic cusp forms oflevel i\(I)1)
on PGL 2(FA.). We now repeat the vertical control theorem in the setting of
PGL(2).
Theorem 4.11 Let 1)1 be an integral ideal prime to p with 1)1 n Z = (N). Sup-
pose that p is prime to N d(F)c and p 2: 5. Then we have for T = Reso/zGm,
(1) The dualV~~:p*(i\(I)1)) ofv~~:p(h(I)1)) is a projective W[[T(Zp)]]-module
of finite type;
(2) If k 2: 31, V~~:p*(i\(I)1)) 0W[[T(Zp)]],k W is isomorphic to the direct sum
EBcEc l il(C l i)2 Homw(H~rd(M(c, r 1 (1)1))'~k)' W),
More generally, identifying /l-N 129 0* with O/(N) (for N prime to p), we
can let g E GL 2(0/(N)) act on level r(N)-structure ¢N : (0/(N))2 ~ A[N]
by ¢N I--t ¢N 0 g. Thus for a given subgroup r c GL 2(0), we can think of the
orbit ¢ = ¢N 0 r and can think of the moduli problem classifying (A, A, ¢).
We call this level structure ¢ of type r. We write 9J1(c, r) for the coarse
moduli scheme for this modulus problem of type r with the c-polarization
requirement. We again define the module V~~~p (c, r) of p-adic modular forms
for this modulus problem as before.
Since the stabilizer of the standard subgroup (0/91 EB 0) of (0/91)2 is
given by the subgroup fa (91) of G L2 (0), the level ro (91)-structure introduced
above can be considered to be the level structure of type ro(91). Any level
r o(91)-structure (even if we require isomorphisms to keep polarization) has
automorphisms ±1; so, it cannot be representable without adding an extra
structure. Nevertheless, we obtain the following theorem in exactly the same
manner as in the proof of Theorem 4.1l.
Theorem 4.12 Let the notation be as above. Suppose that 91 is an integral
ideal prime to p with 91 n Z = (N) and that p 2': 5 is prime to Nd(F)c. Then
we have for T = Reso/zGm,
(1) The dual V~~~p*(jh(91)) ofV~~~p(.fo(91)) is a projective W[[T(Zp)]]-module
of finite type;
(2) If k 2': 31, V~~~p*(fo(91)) ®W[[T(Zp)]],k W is isomorphic to the direct sum
EBcEc 1tl(C 1t)2 Homw(H~rd(M(c, ro(91))'~k)' W), where M(c, ro(91)) is
the smooth toroidal compactification of 9J1( c, ro (91)).
where G' runs over all etale cyclic subgroups of order qT disjoint from G.
Since 7r*A = 7r 0 A 0 t7r is a cqT -polarization, the modular form f has to
be defined for abelian varieties with cqT-polarization. When q t 1)1 and N(q)
is invertible in R, we define the action of the Heeke operator T(l, qT) on
f E Gk(cqT, rf(I)1); R) by
where G' runs over all etale cyclic subgroups of order qT. We can check that
fIU(qT) and fIT(l, qT) stay in Gk(C, rf(I)1); R) if f E Gk(cqT, rf(I)1); R).
4.1 Hilbert-Blumenthal Moduli 129
Now we allow the ring R in which N(q) may not be invertible. First assume
that R [N(q)] is flat over R. Then by the flat base change theorem ([ALG]
Proposition III.9.3), we have
Gk(c,r;R/pmR) = HO (9J1(c,r)'~~kjP"'R)
2:' HO (9J1(c, r)'~~k) ®R R/pm R = Gk(c, r; R) ®R R/pm R
p'"
follows from the long exact sequence from the short one R -=----t R - R/pm R.
Thus if the operators U(qT) and T(l, qT) are R-integral, they are well-defined
over R/pm R. Later we give a definition of these operators as an algebraic cor-
respondence, which works well over a more general ring R (see Section 4.2.5).
We now deal with p-adic modular forms defined over a p-adic ring R clas-
sifying test objects (A,>..,i p : J-Lpoo ®()-l y A[poo],¢)jR' (for R-algebras R')
with prime-to-p level structure ¢ of type r. When q is prime to p, N(q) is
invertible in a p-adic ring; so, we define fIU(qT) (for qllJl) and fIT(l, qT) (for
q f 1Jl) replacing the differential (1T*)-lw by the level pOO-structure 1T 0 ip in
the above definitions (4.16) and (4.17).
To define U(p) (pip) acting on p-adic modular forms, let SoojW be the
formal completion of 9J1(c, r)[~] along its modulo p fiber for the lift of the
Hasse invariant as in (E). Let (X,.x, ¢)j'JJt(c,r) be the universal test object
(supposing the classification functor of prime-to-p level ris representable).
We pick an etale cyclic subgroup C' c X[p]jSoo of order p. As we show in
Section 8.3.1 (particularly, the argument below (8.25)), C' can be defined
only over a locally free covering S'oo/ Soo ofrank N(p), and S'oo/ Soo is radiciel
(purely inseparable over the generic point of Soo). By the universality, we have
a unique morphism Spec( R) Y:.t Soo c 9J1( c, r) with cpo (X, .x, ¢) 2:' (A, >.., ¢).
130 4 Hilbert Modular Varieties
Then Spec(R') = cp*S,=, and R' is an R-algebra (locally free of rank N(p))
with trace map Tr : R' --+ R (see [K] 3.11), and G' = cp*C' gives an etale cyclic
subgroup of AIR' of order p. Here the trace Tr(x) is given by the coefficient
(times -1) of the characteristic polynomial over R of multiplication by x on
R'. Since A/G' is defined over R', the operator N(p)U(p) for a prime ideal
pip can be defined by
fiN(p)U(p)(A, A, i p , ¢) = Tr(f(A/G', 7f*A, 7f 0 i p , 7f 0 ¢)). (4.18)
If R[~] is flat over R, the trace Tr(f(A/G', 7f*A, 7f 0 ip, 7f 0 ¢)) E R is just
the sum I: c ' f(A/G', 7f*A, 7f 0 i p , 7f 0 ¢) over all cyclic subgroups G' of order
p generically different from the connected component A[P]jR of A[P]IR; so,
(4.18) is compatible with (4.16). The divisibility of the operator N(p)U(p)
by N(p) can be checked by computing either the q-expansion of fiN(p)U(p)
(in Section 4.2.9) or its expansion at a Serre-Tate deformation space (as we
do later in 8.3.1). The representability of the level r modulus problem is not
essential in defining U(p), because we can define it at any level r if we once
have the operator well-defined for a representable r n rl ([m) moduli problem.
There is a shortcut to get the operator U (I') acting on p-adic modular
forms if R = W. By the earlier argument for classical modular forms, we have
a well-defined W-integral operator U(p) on D~uSP C v cusp in (4.13) (for any
e
weight > 0) given by (4.16). Then by the p-adic density of D t SP in vcusp
(Theorems 4.9 and 4.10), the operator U(p) extends to p-adic modular forms,
which of course coincides with the one given by (4.18).
We have from the definition U(qr) = U(qt. If n = (~) for a ~ E P';, then
A H ~-l A gives a bijection between c-polarizations and nc-polarizations; so,
it induces an identification Gk(c, rl(91); R) ~ Gk(nc, rl(91); R) by the asso-
ciation (A,A) H (A,~-lA). Thus we may regard U(O = TIq U(qe(q») for the
prime decomposition n = TIqqe(q) as an endomorphism of Gk(c,rl(91);R).
This really depends on the choice of ~ (because the identification is given
by A H ~-l A), though U(~) is well-defined, up to a unit scalar multi-
ple, independently of the choice of ~. Similarly, we can define T(l,~) by
T(l,~) = TIq T(l, qe(q»). Thus we have a well-defined U(p) as an endomor-
phism of Gk(c, rl(91); R). Similarly, we have T(p) acting on Gk(C, rl(91); R).
In the above definition, we can replace rl (91) by r 1 (91). Since the polar-
ization of a level r 1 (91) test object is specified only modulo multiplication by
O~, U(~) and T(l,~) are well-defined independently of the generator ~ of n;
thus, we may write U(n) and T(l, n) for U(~) and T(1,~), respectively.
Note that the space of modular forms Gk(c, 91; R) depends only on the
strict ideal class of c. So we define
for the space of p-adic modular forms V(c, 91; R) over R. The Hecke op-
erators U(q) and T(l, qr) permute the components Gk(C, 91; R) and give
4.2 Hilbert Modular Shimura Varieties 131
endomorph isms of G k (T1('01); R) (or V(T1 ('01); R)). The diamond operator
(3): G k (c,'01;R) -t G k (C3 2,'01;R) (resp. V(c,'01;R) -t V(q2,'01;R)) intro-
duced in the previous section for ideals 3 prime to p'01 also acts on the above
spaces (permuting ideal classes c -t C3 2 ). This is the easiest way of extending
the definition of Hecke operators to GL(2) and is based on the first identity
of (4.7). There is another way of extending the definition of Hecke operators
to G L(2), using the open compact subgroup st ('01) in (4.6) in place of Sl ('01).
Then the approximation theorem yields
U 9Jt(c,r~('01))(q = G(Q)\G(A)lsi('01)Z(~)Co.
CECZt ('Jl)
The above naive definition of the Hecke operator acting on Gdc, rl('01); R)
gives rise to the Hecke operator on
c'rve c'rvC
are equivariant under the Hecke operator action. Here in the above sum, c'
runs over the classes in Cit ('01) equivalent to c in Cit.
Since Clt('01) = (Fl oo )) x IF!; det(Tl('01)), we may define a quotient Cl r
of Cit ('01) by det(r) for the more general subgroup r c GL 2(O) containing
Tl ('01). Then we define for a p-adic W-algebra R,
in E~(S), since we have c;'ij = 'ij because c; E K. Thus the choice of A also does
not matter in the target E~, and the above bijection is indeed induced by an
isomorphism of functors if det(K) n O~ c (K n Z(Z))2.
If we take L = i)-I EEl 0 C V, then LI\L ~ 0* by (a, b) 1\ (a', b' ) H a'b - ab' .
For this choice, the polarization ideal is trivial; that is, c = O. If further, we
4.2 Hilbert Modular Shimura Varieties 137
identify f-LN with 71.1 N71. by choosing a primitive Nth root ( = (N of unity in
Q[f-LN], having Tj is equivalent to having
Take the open compact subgroup S(N) in (4.6), a conjugate of r(N), stabi-
lizing the lattice L = ()-l EEl O. Over Q[f-LN], the component of ShS(N)(G, X)
corresponding to [~(N),O (which corresponds in turn to G(Q)S(N)G(JR)+ in
(4.7)) represents the now familiar functor:
The choice (gives rise to the identification Q[f-LN] = Q[Xl!(<PN(X)) with Q[(]
for the cyclotomic polynomial <PN(X) in 2.2.1, and an automorphism 0' E
Gal(Q[(]/Q) changes the identification by ( H (CT, whose action is induced by
¢N H ¢NO( 8 ?) for a unit c E ZX such that (0' = (c. In other words, the action
of (8n E G(A,(oo)) on Sh(G,X)/Q; brings 9J1(O,r(N))/Q;[(] c ShS(N)(G,X)
to its O'-conjugate 9J1(O,r(N));Q;[(] in ShS(N)(G,X)/Q;'
To accommodate all other components G (Q) ( 8?) S (N) c G (A (00)) is to
change the lattice L to Lc (i. (8 ?)-l) n V. Thus we have
=
which implies
Note that the identity over Q is wrong, because, over Q, the group scheme
(f-LN ® ()-l) X (01 NO) is not isomorphic to the constant group (()-l IN()-l) x
(OINO). In any case, by the Galois action on 9J1(c,r(N))/Q;[I-'N] described
just below (4.22), we can descend the right-hand side to the base field Q
to obtain the model ShS(N) (G, X) /Q;, because 9J1( c, r(N)) is quasi-projective
as we already observed (see [GME] Example 1.11.1 for the descent). Thus if
F = Q, Sh S(N)(GL(2),X) is irreducible and is isomorphic to the scheme
9J1(1, r(N)) ®Q; Q[f-LN 1 regarded as defined over Q, but not geometrically
irreducible (whose geometrically irreducible components are isomorphic to
Y(N) = 9J1(1, r(N)) over ij). For F i= Q, ShS(N)(G, X) may not even be ir-
reducible over Q, and the picture is not very straightforward (since it involves
an explicit description of the action of G(A(oo)) on the set 7ro(Sh(G, X) /Iji) of
connected components).
Take a sufficiently large integer N so that S(N) C K. Then the coarse
moduli scheme representing [~ is given by the quotient ShS(N)(G,X)IK.
138 4 I-Iilbert Modular Varieties
This means that to classify test objects, we now allow only isogenies with
degree prime to all primes in E (called "prime-to- E isogenies" ), and the degree
of the polarization A is supposed to be also prime to E. Two polarizations are
equivalent if A = aA' = ,AI 0 a for a totally positive a prime to E.
Fix a prime pEE and an a-lattice LeV = F2 with A(L /\ L) = c*, and
assume self Op-duality of Lp = L 0z Zp under the a-Hermitian alternating
pairing A : V /\ V ~ F. Consider test objects (A,);",fj(E))/S' Here 7](E) :
V(A(1.7°o)) = V 0Ql A(ECX!) ~ V(E)(A) = T(A) 0z A(ECXJ) and A E );" are
supposed to satisfy the following requirement, T(E)(A) /\ T(E)(A) -"-2.t F1ECX!)
is proportional to A : V /\ V ~ F up to scalars in (F 0 A(ECXJ))X. We write
the K(E)-orbit of 7](E) as 'Fj(E). Then we consider the following functor from
Z(l:)-schemes into SETS.
(4.25)
(4.26)
specified (which does not determine the isomorphism class of L if the class
group of F is nontrivial). This problem is more acute at p because over Z(p),
Tp(A) does not determine Lp. Indeed the p-adic Tate module of an abelian
scheme of characteristic p has less rank than its characteristic 0 counterpart.
The self-duality at p of L has to be imposed to overcome this point (see the
argument just above Remark 7.4). Also we need the density of the derived
group G 1 (Q) in G 1 (A.c oo )) (the strong approximation theorem) in order to
know that geometrically irreducible components of She:) are indexed by the
class group Cl(K): 7ro(ShC:}ij) ~ Cl(K). For Shimura varieties of a more
general pair (G, X), these points have to be studied carefully (see Remark 7.4).
If pEE is unramified in F/Q (and K(E) is sufficiently small), She:) is
smooth over Z(p) by the infinitesimal criterion of smoothness (e.g., [NMDl
Proposition 2.2.6); that is, we can show that any characteristic p test object
lifts to characteristic 0 infinitesimally. To explain this, let R be a Z(p)-algebra
with a nilpotent ideal I C R containing a power of p. Put Ro = R/ I. We
fi6
want to show the existence of a lifting of a test object (Ao, );"0, E ))/Ro to R.
The abelian variety Ao lifts to an abelian scheme A/R (with A @R Ro ~ Ao)
by the deformation theory of Grothendieck-Mumford (cf. [CBTl V.1.6, [GITl
Section 6.3, [DAVlI.3, and also Theorem 8.8 in the text and the remark after
the theorem). Since the degree of polarization is prime to p (here we use the
fact that we can choose a representative c prime to p in a given class in Cl(K)),
A also lifts because we may assume that Ao : Ao ---+ tAo is etale (and hence
t A = A/C for an etale subgroup C C A lifting Ker(Ao); see [ECHlI.3.12). As
for the level structure ry6 E ), it is prime to p and hence etale over Ro. Then it
extends uniquely to a level structure ry(E) : v1 E ) ~ VeE) (A) over R. By the
deformation theory of Barsotti-Tate groups (see [CBTl V.1.6), using (det),
we can find a deformation A/R of AO/Ro with an embedding 0 '---+ End(A/R)
compatible with 0 '---+ End(Ao/Ro) (see Theorem 8.8 and the remark after the
theorem).
When E consists of a single prime p, we write t:f) as t:JI), She:) as Sh~),
and so on.
9)1' ~ Sh(G,X)
·1 1n
Spec(QCYC) ------+ Spec(Q) .
for a compatible system of Nth roots of unity so that (m = (;:/m (if min) gives
one geometrically irreducible component, which is the image of X+ x 1 in the
expression (4.21) over <C. This component 9)1 is called the neutral component
of the Shimura variety ([D2] Definition 2.1.5).
Let c = (8 ~) for c E F;c oo ). Since r( c, N) = (cS(N)c- I ) n G I (Q) if
cO n F = c, we find that
(4.28)
is the image of 9)1 under the right action of c; so, 9)1c gives another geomet-
rically irreducible component of Sh( G, X) /Qcyc. In other words, QCYc is the
field of definition of the component 9)1c in the sense that QCYc is the algebraic
closure of Q in the function field of 9)1c over Q.
By the description (4.21) of complex points of the Shimura variety, the
group G(A(oo)) acts transitively on 7ro(Sh(G, X) /ij). Thus, up to conjugation
by G(A(oo)), the automorphism group of a geometrically irreducible compo-
nent is independent of the choice of the component.
We now start writing down explicitly the group that gives an open sub-
group of Aut(9)1/Q) with finite index. As before we let 9 E G(A(oo)) act on
Sh(G, X)/Q by
(A,::\,1]) H (A,::\, 1] 0 g),
which gives a right action of G(A(oo)) on Sh(G, X). We write 7(g) for the
automorphism of Sh(G,X)/Q induced by 9 E G(A(oo)). For each point x =
(A,::\,1]) of the Shimura variety Sh(G, X), we can associate a lattice L =
1]-1 (T(A)) C F1c oo )l and a choice of a base W = (WI, W2) of this lattice
4.2 Hilbert Modular Shimura Varieties 143
(E)
We write 9J1~~) for the image of 9J1 in Sh(17) (G, X). We define the localization
of Z at E by Z(17) = Q n Z17 inside A (00) for Z17 = ITpE17 Zp. As a convention,
we write Z(0) = Q and Z0 = A(oo). We define
cad is the algebraic group quotient of C modulo the center Z, but cad(k)
may not equal C(k)/Z(k) depending on the field k). For the quotient map
ad: C -+ cad, we define C(QI)+ = ad-1(cad(QI)+). Often C(QI)+ = C(QI)+,
but there are examples where C(QI)+ ;> C(QI)+ (like some unitary groups).
Here we use the symbol cad in place of PC in order to facilitate an easy
reference to the papers quoted above ([D2] and [MiS]).
We write Z(QI) for the topological closure of the center Z(QI) of C(QI) in
C(A(oo)). Let the adjoint group cad(QI) act on C(A(oo))/Z(QI) by conjugation
and consider the semi-direct product
on which the product law is given by (g,h)(g',h') = (g. ad(h)(g'), hh'), where
ad(h)(g') = hg'h- 1. We now bring the group C(QI)+ into the semi-direct
G(A(oo») d
product Z(IQl) )<I ca (QI)+ by"( H h-1,adh)). Then the image is a nor-
G(A(oo») d
mal subgroup of Z(IQl) )<I ca (QI)+. We define the amalgamated product
G(A(oo») d G(A(oo») d
Z(IQl) *G(IQl)+ ca (QI)+ to be the quotient of Z(IQl) )<I ca (QI)+ by the im-
. . G(A(oo») d +
age of C(QI)+. We wnte [g, adh)] for the class m Z(IQl) *G(IQl)+ ca (QI)
G(A(oo») d
represented by (g,adh)) E Z(IQl) )<I ca (QI)+. Since
-,
is well-defined, we define a new group [ (C, X) by the inverse image of the
closed subgroup A x FX F:"+/ px F:"+ of F; / FX F:"+ in the amalgamated
G(A(oo») d
product Z(IQl) *G(IQl)+ ca (QI)+.
We may axiomatize the above construction as follows (cf. [D2] 2.0.1). Take
three groups Q, r, and L1. We assume to have a homomorphism r 3 "( H -;y
onto a subgroup r
of Q and to have the following commutative diagram of
group homomorphisms.
(4.29)
Q -----+ Aut(Q).
ad
Here Aut(Q) is the automorphism group of the group Q and ad(g)(x) =gxg- 1
for 9 E Q. We often write 'Ph) for 'P(-;Y) and assume that ,(0) (0 E L1) preserves
r as a whole. We suppose the following two compatibility conditions,
(a) '('Ph)) = ad(-;Y) for all "( E r (commutativity of (4.29));
4.2 Hilbert Modular Shimura Varieties 145
(g, b) (,},-1 , cp(,)) (g, b) -1 = (g. r( b) (,},-1 )r( b· cp(,) )(r( b- 1)(g-I)), ad( b) (cp(,))).
(4.30)
By this isomorphism, the amalgamated product 9 *r..:1 acts on 7'\9, and the
action of [g,b] E 9 *r..:1 on the class [x] in 7'\9 (which is sent to [x, 1] E
..:1\(9 *r ..:1)) is given by
[x] . [g, b] = [x, l][g, b] = [xg, b] = [1, b][r(b)-I(xg), 1] = [r(b)-I(xg), 1].
(4.32)
with 'P given by "( H adh) and "( H "i given by "( H h mod Z(Q)), we find
(4.33)
by g +-t [g,l].
Take the geometrically irreducible component 9]1/QCYC c Sh( G, X) /Qcyc
containing the image of X+ x 1, and define Aut(9]1/Q) by the scheme auto-
morphism group. The following fact has been shown in [Sh5] II 6.5 and [Mt]
Theorem 2 (see also [MiS] 4.6 and 4.13).
Theorem 4.14 Let E be a proper subset of rational primes (E could be
empty). The stabilizer in G(A(Eoo))/Z(7L(E)) of each geometrically irreducible
component of Sh(E)(G,X)/Q is given by [(E)(G,X). When E = 0, the
right action of (g, adh)) E [(G, X) h E G(Q)+) on [x, g'l is given by
[X,g'] H b-1(x), (glg)ad h )], where (glg)ad h ) = ,,(-l(g'gh. Moreover for the
function field jt(E) of 9]1)~), Aut(9]1)~)) = Aut(jt(E)), and we have a canon-
ically splitting exact sequence of topological groups:
and noting X+ = 3. Each [z, g] carries the triple (Az, Az, 7]z 0 g), where
Az(C) = CI/(O* + Oz) and 7]z (b) = bz - a identifying T(A z) = 0* + Oz.
To see this, we note an easy fact that the map: [z,g] H (A z,AZ)7]z og) sends
(3 x G(A(oo))) surjectively onto [(0)(C). Thus we need to check
(A,-l(z),;\,-l(z),7],-l(Z) og) ~ (A z,;\z,7]z 0l'(oo)g) in A~) for I' E G(Ql)+
SA r r S
where O(E)+ = O(E) n F';. Then the strong approximation theorem (e.g.,
[MFG] Theorem 3.2) shows that g(E)(G, X) = G(Z(E»)+ . SA(Z(E)X). As seen
in (4.2), for Z E 3, the test object (£z,det(a)Az,<PN,z 0 a(oo») is prime-to-
E isogenous to (£"'-'(z), A",-,(z), <p"'-'(z») for an element a E G(Z(E»)+ =
G(Z(E») n G(JR)+. Thus a(oo) (a E G(Z(E»)+) sends a point in 9)1(E)(C) rep-
resented by Z E 3 to that represented by a-1(z), and hence the image of
G(Z(E»)+ in '[(G,X) preserves 9)1(E), in particular G(Z(E»)+ c g(E)(G,X)
preserves the component 9)1(E). We conclude from the continuity that r :
,[(E)(G,X) '-+ Aut(9)1)~»). Since G(QE) = B(ZE)G(ZE) for the upper trian-
gular Borel subgroup BeG, it is easy to see that the normalizer of G (ZE)
is itself. Thus any element g E G(A(oo») preserving Sh(G, X)/G(ZE) has its
E-component gE in G(ZE), and ,[(E) (G, X) gives the stabilizer of 9)1(E) in
G(A(Eoo) G(Z XA(Eoo)
( ) = E () • Up to this point, the proof follows Shimura's
Z Z(E) Z(<Q!)nZ(ZExA Eoo )
argument (see, e.g., [AAF] Section 8).
From now on, we argue slightly differently from [Sh5] II, [Mt], and [MiS].
Step 3: Let ~ = (X,~, </» be the universal test object on 9)1(E). If the pullback
O'*X = X x!m,O' 9)1 for 0' E Aut(9)1(E») is induced by an isogeny a : X ---+ O'*X,
we find O'*</> = a 0 </> = </> 0 g for g E G(A(E») and hence r(g) 00'= 1, which
basically finishes the proof.
Towards this goal, we introduce preparatory propositions. We start with
Proposition 4.15 Let 0' E Aut(jt(E»). Let U C 9)1(E) be an open dense
subscheme on which 0' E Aut(jt(E») induces an isomorphism U ~ O'(U). Let
150 4 Hilbert Modular Varieties
K be an open compact subgroup of [(E) (G, X), and write UK for the image of
U in S h (E) / K. Let T) be the generic point of UK and Xry be the generic fiber of
the universal abelian scheme X K over 9J1 K . Then the two abelian varieties Xry
and Xu(ry) (for the generic point 0"( T)) of O"(UK)) are isogenous over!fJ( T), 0"( T)))
(which is the composite in .R(E) of the two function fields of UK and O"(UK )).
The proof uses many facts from the theory of algebraic fundamental groups
1fl (X,x) of schemes X, which is summarized in Section 4.4. Here we use
freely without explanation the terminology concerning fundamental groups,
referring the reader to Section 4.4.
Proof. Open compact subgroups K of [(E) (G, X) give a fundamental system
of open neighborhoods of the identity element of Aut(.R(E)) under the Krull
topology. Let 9J1 K = 9J1/ K whose function field is the fixed field of K in .R(E).
Then K = Gal (9J1/9J1 K ) for the stabilizer of 9J1K. If we choose K sufficiently
small depending on 0" E Aut(9J1/IQJII-'N])' we have UK = O"KO"-I. This follows
from the Galois theory in Section 2.3.2. We write UK (resp. Uu K) for the
image of U (resp. of O"(U)) in 9J1 K (resp. in 9J1uK).
By the argument preceding this proposition, the action of the group
G(A(E)) on the set 1fo(Sh~~)) of connected components factors through the
determinant map of G, and the image of G l (A(E)) in Aut(Sh(E)) is con-
tained in the stabilizer Aut (9J1j1QJ) and hence in the field automorphism
group Aut(.R(E)). Let G l (A(E)) be the image of G l (A(E)) in Aut(.R(E)) (so
0 1 (A (E)) is isomorphic to the quotient of G l (A (E)) by the center of G l (Z(E))).
We take a sufficiently small open compact subgroup S of geE) (G, X). We
write SI = G l (A(E)) n Sand S\ (resp. B) for the image of SI (resp. S)
in 0 1 (A (E)) (resp. in E( G, X)). Shrinking S if necessary, we may assume
that 9J1/9J1s is etale, that the geometric Galois group Gal(9J1 /ij/9J1 s /ij) ~
Gal(9J1 /IQJ(E) /9J1 S /IQJ(E)) = B 1, that S = TIe Se with Se = S n G(!fJe) for primes
f! Ef. IJ, and that uBI = O"B I O"-1 C G l (A(E)).
The Galois group Gal(!fJ(E) /!fJ) acts on the underlying topological space
of Sh(E) / S. Thus we have a finite extension ks of !fJ inside !fJ(E) such that
Gal(Q/ ks) gives the stabilizer of 9J1 s in Sh(E) / S. Since the action of geE)
on the set of geometrically irreducible components 1fo(Sh~~)) factors through
Gal(!fJ(E) /!fJ) via the determinant map composed with the Artin reciprocity
map (Step 2), the field ks is given by the fixed field of det(S) in !fJ(E). The
variety 9J1 s is defined over k s . Thus K = kS(T),O"(T))) is a field of finite type
over !fJ; so, we can apply Theorem 4.13 to Xry and Xu(ry) defined over K. We
fix an algebraic closure K of K and take a geometric point Yj E 9J1 s (K) over T).
Let Xs be the universal abelian scheme over Us. Write X7] = Xry ®1QJ(ry) K for
the geometric fiber of Xs at Tj.
Note that 1fl (Xr;/K' 07]) for the origin 07] of the abelian variety X7] is canon-
ically isomorphic to the Tate module T(Xr;/K) (see [ABV] Section 18) and
that f.Ox = 09]1, because X l-t 9J1 is geometrically connected (by the
4.2 Hilbert Modular Shimura Varieties 151
The action of 1fl (Ua S/ks' d'fj)) on T(E) (XO'(ri)) factors through Gal(U /Ua s) =
aS. We fix a path in Uu S from 1J(fj) to 'fj (as in Remark 4.41) and lift it to a
path from dO'i)) to O'i) in Xas, which induces isomorphisms ([SGA] V.7):
?1 1? 0'·1
'-+ ---;>
°
the polarization) give rise to the identity coset of Aut(F) v: PG(A(Eoo)) in
Aut(GI(A(Eoo))) (only depending on i : Y End(Xa(7))) with ambiguity
modulo right multiplication by Aut(F)), we conclude O"F E Aut(F). 0
Here is the last step.
Step 4: Replacing 0" by O"O"F I for O"F E Aut(F) c Aut(J't), we may assume
that O"F = id F for the starting 0" E Aut(R(El). Thus we may assume that
g(O")S = OO7)S for an isogeny 007) : X7) -+ Xa(7)). Writing the level structure
on X7) as ¢~E), we find that 007) 0 ¢~E) = ¢~~~) 0 as-I for as E G(A(E)). In
other words, as-I(O"(TJ)) = 71, which implies that as = 0" on the Zariski open
dense subset Us of 9Jt s and, hence, they are equal on the entire 9Jt s . Thus 0"
comes from a scheme automorphism of 9Jt, and 0" is induced by a = lims as E
G(A(Eoo))/Z(7l,(E)). Since [(E) is the stabilizer in G(A(Eoo))/Z(7l,(E)) of9Jt(E)
by Step 2, we conclude that a E [(El. 0
mains irreducible (for the residue field IF C fp of 7l,i~)), and M;:) is the
toroidal compactification of 9Jt;;) = 9Jt(E) 0 Z (E) IF. This shows that 9Jt;;)
(p)
·
w hose Image .
III
("(E)
G
(G , X)'IS gIven
. by
A mod p version of the above proof of Theorem 4.14 gives (see [H03b] for a
detailed proof):
Theorem 4.17 Let the notation be as above. Suppose that pEE and that p is
unramified in F /Q. Write ~~E) for the function field of 'JJt(E) over the residue
field IF of Z~~). Then T in Theorem 4.14 induces the following isomorphism,
( a)'(ZZ)=(px(Q)-l(Z)px(Q)-l(z))(CZ+d
PX l I I I
_0
cz+d °
)=(zz)(cz+d _0 )
1 1 cz+d , °
where px(a)' = (~d) E GLi(Foo) = G(lR.)+. This shows that r.p(cz + d) =a
in Mx for r.p = ioo IM x ' that Px is the regular representation given by
4.2 Hilbert Modular Shimura Varieties 155
Px (a) '(ZZ)~(zz)(aO)
l l ~ 1 10C; ,
and that Mx = F[z]. The abelian variety Xx over x has an analytic uni-
formization Xx(CC) = Fc/.c z (Fre = CI) for an O-lattice £Z = bz + cr* c Mx.
Thus End~(Xx) """) Mx and we may identify V(Xx) with Mx ®1Ql A(CXJ).
For the moment, simply write x for [x, 1]. Then the level structure 7]x of
Xx is given by V(A(CXJ)) '3 (a, b) f---t -a + bz E Mx ®1Ql A(CXJ). In particular,
a 0 7]x = 7]x 0 Px (a). Since det(px (a)) = aa e E F for a E Mx for com-
plex conjugation c generating Gal(Mx/F), we find px(a) E Q(G,X). Let
(X,X,1])ISh(G,X) be the universal triple, and f : Sh(G,X) ---+ Sh(G,X)
be an automorphism defined over Q fixing x. Thus f fixes the geometri-
cally irreducible component containing x E Sh(G, X). By Theorem 4.14,
we find g E Q(G, X) such that f = T(g). By the universality of X, we
~,
have a triple (y, A, 1]') ISh(G,X) and an F-linear isogeny a : Y ---+ j*X giv-
ing an isomorphism (y, X', 7]') ~ (1*X, j*X, j*1]) in the category A~, where
j*X = X xSh(G,X),f Sh(G, X) and j*X and j*1] are the pullback of X and 1].
On the other hand, by the definition of the action of T(g), we know that
(1*X, j*X, j*1]) ~ (X, X, 1] 0 g). Thus the isogeny a induces an endomor-
phism ax E End~(Xx), and ax O1]x = 1]x 0 px(a x ) = 1]x 0 g. This shows
that Mx = End~(Xx) and g = px(a x ). In other words, the stabilizer of [x, 1]
in Q(G,X) is given by the image Tx(Q) of M;.
By the de Rham comparison isomorphism combined with the Hodge de-
composition (e.g., [ABV] Chapter 1), we have an Mx-linear isomorphism:
the center of Gal(QcM /Q) and hence is unique (see [IAT] Proposition 5.11
and [ACM] 18.2), we see easily that M' is again a CM field. Thus for any
a E Aut(C/M'), the isomorphism class of AO" over C only depends on a. If
,X : t A 2=' A 0 c, then ,XO" : tAO" 2=' AO" 0 c; so, if (A,'x, ¢N) E £:,r(N) (C), its
conjugate is also in £:,r(N)(C)' Since PicA/K is well-defined over K if A is
defined over a subfield K c C, Aut(C/ K) permutes the component group
NS(A)(K) of PiCA/K' Since the Neron~Severi group NS(A)(C) is in bijec-
tion with the set of Riemann forms inside Homz(a I\z a, Z) (see, e.g., [ABV]
Section 2), NS(A)(C) is a discrete torsion-free group with count ably many
elements. Since A is projective, it is defined over a field K generated over Q by
finitely many elements (giving the coefficients of the homogeneous equations
of A). Thus if x E NS(A)(C) is transcendental over K, its conjugates un-
der Aut(C/K) are uncountable, contradicting the countability of NS(A)(C).
Thus a polarization ,x is always defined over a finite extension of K. Therefore
the number of isomorphism classes of (A,'x, ¢N) over C is countably many,
because that of A is bounded by the class number of M times the number of
CM types. This shows the number of conjugates under Aut(C) of the point
x E ry)1'(c,r(N))(C) carrying (A,'x,¢N) is count ably many; so, x has coor-
dinates in a finite extension Mil / M'. Choosing N sufficiently large so that
£:,l(N) is represented by ry)1'(c,r(N)), we find that (A,'x,¢N) is defined over
Mil; in particular, (A,,X) is defined over Mil.
We let G(Q) act on the column vector space V = F2 through the matrix
multiplication. The action of Tx via p~ makes VA a free module over the adele
ring MA of rank 1. Then the subspace Vx = V 01Ql C on which hx acts by its
restriction /1x = hx IG", x I is preserved by multiplication by M x , yielding an
isomorphism class P of representations of M. In this way, we rediscover the
CM type P, because Vx gives rise to the Lie algebra of A(C) as we put the
complex structure on VIR by h x . Since the isomorphism class P is determined
by its diagonal entries cp, : Mx '---+ C, we may identify P with a formal sum
L, cp, or the set {CPI,"" cPg} as before. Thus the field M' is the minimal
field of definition of /1x : <G m -+ G (the half of h x ). We define a morphism
rx : T~ = ResM' /1Ql<G m -+ Tx by
rx: T x' =
R eSM' /1Ql <G m --'--+
f.1x R
eSM' /1Ql T x ------+
Norm T
x'
Since a morphism of the torus ReSM' /1Ql<G m into another torus of the form
ResK/IQl<G m for a number field K can always be realized by x M det op'(x) for
a representation P' of the algebra M' on a vector space over K, we can write
rx = det oP'. Then P' is a CM type of M', whose explicit form is given in
[ACM] Section 8. The pair (M',p') is called the reflex CM type of (M,p).
For each s E T~(A(oo)) = M'~(OOll we have the Artin reciprocity image
[s,M'] E Gal(M~b/M'), where M~b is the maximal abelian extension of M'.
Since Tx(IR) is the stabilizer of x, we have [x,,),g] = b~I(X),g] = [x,g] for
')' E Tx(Q), and the map: [x,g] M [x,px{rx(s))g] = [x,g . p~(rx(s))] only
depends on the Artin symbol [s, M'] E Gal(M,ab /M') by class field theory.
Hi8 4 Hilbert Modular Varieties
The test object at the right-hand side is sitting over [x,gp~(rx(s-l))], and
the test object at the left-hand side is sitting over u([x,g]). Thus the two
geometric points CJ([x, g]) and [x, gp~(rx(s-l ))] = [x, px(rx(s-l ))g] are equal
in Sh(G,X)(C): u([x,g]) = [x,gp~(rx(s-l))] = [x,px(rx(s-l))g]. In particu-
lar, the point [x, g] is fixed by Aut(C/ M,ab) by the above expression, and we
have [x, g] E Sh(G, X)(M,ab). Then we may replace u in the above displayed
formula by [s, M'l, which gives the last identity in the corollary. 0
Theorem 4.21 (Ribet) Let p be a prime outside N. Then the Jgusa variety
T~ over 9Jt( c, r) jIFp (and hence TO'.) is irreducible for all CY, where r is any
one of r(N), rl(sn), and rl(sn).
4.2 Hilbert Modular Shimura Varieties 159
This result was first shown by Ribet [RiJ (and [DR]) using the description (due
to Deligne [D]) of p-ordinary AVRMs and density of ordinary points. We give
a proof assuming that p is un ramified in F jQ. Our proof is a generalization
of the proof given for the elliptic Igusa tower (Theorem 3.3), and uses the
description of the automorphism group of the Hilbert modular automorphic
function field (Theorem 4.14) and the existence of a smooth projective toroidal
compactification of Hilbert modular varieties ([Raj and [C]). Another proof of
irreducibility has been proposed in [C] along the lines of the proof in the Siegel
modular case of Chai-Faltings ([DAV] V.7). The point of Chai's argument is
that the monodromy group at the cusp of the Igusa tower is isomorphic to
the p-adic closure of the totally positive unit group 0 ~, and therefore, if the
closure is full in the norm-I-subgroup Ker(N : 0; --* Z;), the irreducibility
follows. The Leopoldt conjecture predicts that the closure is of finite index in
the norm-I-subgroup, but it is easy to create an example (even for a quadratic
F) such that the index is nontrivial; so, Chai's proof works only in limited cases
(see [DT] Section 12 for more details of this point). This type of phenomenon
always occurs when we deal with G of the form ResFIQG o with F ::J Q for
unitary or symplectic groups Go over F and, hence, the computation of the
cuspidal monodromy does not yield irreducibility of the Igusa tower in general.
Proof. Let N > 0 be an integer prime to p, and write S'N = 9J1(c, r(N))[i-J/w
for a lift E of the Hasse invariant; so, E E HO(M'N,!,!,J..a(p-I)I) for the mini-
mal compactification M'N of 9J1( c, r(N)) with E == Ha mod p for the Hasse
invariant H in Section 4.1.6. Recall that S'N is a geometrically irreducible
quasi-projective scheme over W (Le., S'N ®w K and S'N ®w lF p are both geo-
metrically irreducible). We write S'N = M'N[i-J.
Let TN,a be the Igusa tower over S'lv IF (IF = IFp = WI = W j p W) clas-
sifying quadruples (A,A,<PN,i p) for ip: /Lpn ®()-I y A. Taking a toroidal
compactification MN of 9J1(c, r(N)) and its ordinary locus SN = MN[i-L we
can define the Igusa tower TN, a over SN in exactly the same way as TN,a in
(4.11). Since TN, a is etale over SN, the local structure around the etale neigh-
borhood of a cuspidal divisor of TN, a is the same as that of SN; so, we can
blow down TN,a to get the minimal compactification T'N,a of TN,a over S'N. In
other words, T'N,a is the normalization of S'N in TN,a. We used in (4.11) the
etale dual Q[1Pf to construct TN, a over the toroidal compactification SN. We
could have used the maximal etale quotient X[pa]et of the universal abelian
scheme X over 9J1(c, r(N)) to get TN,a C TN, a over S'N (removing the cuspidal
divisors), since ~ S:' X[pa]et over S'N by the polarization. By Koecher's
principle, we have HO(TN a,I,!:d.JI) = HO(T'N a,I,!:d.JI). Thus we do not really need
T'N,a in the sequel, altho~gh we use the symbol T'N,a in order to have well-
defined cusps on the variety. We write Ta = ~ NTN,a and T~ = ~ NTlv,a
for? = * and 0, where N runs over all integers prime to p. Similarly we define
S = ~NSN and S7 = ~NSJv.
160 4 Hilbert Modular Varieties
(4.36)
We write Vp for the decomposition group in r(£(G, X)) C Aut(.Jt) for this
valuation v. As seen in the proof of Theorem 3.3, Vp is a closed subgroup of
r(£(G,X)), and we need to prove that Vp/lp contains 0; = T(Zp) for the
inertia subgroup Ip.
Since the quadruples (A, >.., ¢N, ip : /l-p'" ® i)-I '--+ A) are classified over W
by Sh(G, X)/(S(N) n Sf(p"')), it appears that we only need to look at the
decomposition group V~ of v in Aut(9Jl u /IQCYc) (9Jlu = 9Jl c /Up /IQCYc) for
4.2 Hilbert Modular Shimura Varieties 161
Up = n
pfN,a
(S(N) n Si(pa)) = {(b n Iu E Op}.
Res
Aut(Coo ; SO) --+ Aut(SO)
tu tu
y (G 1(A(poo))/{±1}) X H --» G 1(A(poo))/{±1},
where Aut(Coo ; SO) is the group of automorphisms of Coo that induce au-
tomorphisms of So. This commutative diagram shows that the action of
g E G(A(poo)) preserves the irreducible component Coo c T~.
The stability of Coo under G(A(poo)) can also be proven as follows. By
construction, the image CN,a of Coo in TN for an integer N prime to p is a
subscheme of X[ N pal et (for the universal abelian scheme X / S, ), and the action
of g with gp = 1 only concerns the prime-to-p part that is canonically a factor
of X[Npa]et; so, the irreducible component containing 00 is kept under T(g).
Therefore g E G (A (poo)) sends the cusp 00 to another cusp s on the same
irreducible component Coo. This shows that T(g) for g E G 1(A(poo))/{±1} c
[(G,X) preserves v.
There is one more way to show the stability of Coo under G(A(poo)).
Looking at the Tate abelian variety Tateo,c-' (q) = Gm ® ()-l/qO with
level structure cPcan : (J1N ® ()-l) X (a/NO) ~ Tateo,c-1(q)[N] sending
(( ® a,m) to (( ® a)qm/N, we easily see that T (0 ~)-l with (o~) E SL 2(F)
sends q-expansion L~ a~q~ of a modular form to L~ a~ exp(27riTr(5J))q~a/d;
so, the image of B(Q) n G(Q)+ (for the upper triangular Borel subgroup
B(Q) c G(Q)) in [(G, X) preserves v. For the unipotent radical U of B, we
find T(U(A(oo))) C V p, because Vp is closed and U(Q) is dense in U(A(oo)).
Since Dt SP in Theorem 4.9 for the variety 9J1(c, r(N)) is p-adically dense in
vcusP, we find that vt~p = D~uSP ®w IF. We can extend (using q-expansion of
*
elements in VCUSP) the valuation v to D~uSP and f E j{U(Zp) can be written as
for g, h E D~usp for a sufficiently large integer N prime to p. Since g and h
162 4 Hilbert Modular Varieties
are defined over 9J1( c, F(N)) for N prime to p, their valuations are unchanged
under the action of Gl(A(poo»); so, the valuation v on jtU(Zp) is preserved by
Gl(A(poo»). As we have already seen, 7(U(Zp)) preserves v on jt, Gl(A(poo»)
preserves v on jt, and hence it preserves Coo.
By the stability of v under 7( G l (A (poo»)), the closure of 7(G l (A (oo»)B(Ql))
in Aut(jt) is in Dp. Note that the projection of GdA(oo»)B(Ql) into Gl(Qlp)
contains B(Ql) C G l (Qlp) , which is dense in B(Qlp) (p-adic density of F in
Fp). Thus 'Dp contains the image under 7 of
Since T(Zp) normalizes Up, the decomposition group 'D~ of v in Aut(9J1u /QlCYc)
contains the image of T(Zp) isomorphic to 0;.
As we already remarked, the action of a E T(Zp) on 9J1 u is induced by
the action on test objects (A, A, TJ(p) , ip) f---t (A, A, TJ(p) , ip 0 a). This shows that
Cal(Coo/Co) = 0;,and hence Tn is irreducible over Fp. 0
Proof. We have already shown that 7(B(Ql)) c'Dp and 7(G l (A(poo»)) C Dp.
Since 7 ( 6~) with tEA x acts through the Artin reciprocity on the q-
expansion coefficients at the infinity cusp, we find that 7 ( 6~) with t E Ql;
is in 'Dp. These elements topologically generate the subgroup indicated in the
corollary. Conversely, if 7(g) E'Dp for 9 E Q(G, X), since 7(g) acts by det(g)
on Qlcyc via the Artin reciprocity, we find that det(g) E Ql; FX F~+. Thus
modifying 9 by an element in B(Qlp), we may assume that gp E GL2(Op).
We consider the universal abelian scheme X/So. We then have the projection
map 7r : X[POO] -+ X[poo]et. Since the irreducible component C n is the gen-
erator of X[pn]et, looking at the infinity cusp taking the geometric fiber of
Tateo,c-l (q)[pOO] over Ql, 9 has to preserve fJpoc 0()-1 in Tateo,c-l (q)[pOO]; so,
gp E B(Qlp). This shows the desired identity. 0
We write ShK for ShK(G, X). For any 9 E G(A(oo)) and open compact sub-
groups K, K' c G(A(oo)), we have two projections PI : ShK9nK' -+ Sh K ,
given by ([x, h] mod Kg n K') H ([x, h] mod K') and Pg : ShKBnK' -+ ShK
given by ([x, h] mod Kg n K') H ([x, hg-l] mod K), where Kg = g-1 Kg
and gK' = gK' g-l. The image of Pg x PI in Sh K xQ Sh K , is an algebraic
correspondence (KgK') C Sh K xQ Sh K , which only depends on the double
coset K gK' as easily verified, using the fact Sh K = Shl K. We now assume
(sl) We have a vector bundle £/Sh (either an algebrogeometric locally free
sheaf of finite rank on the algebraic variety Sh/ Q or a locally constant
sheaf on the topological space Sh(C)) which descends to a sheaf £K/ShK
of the same type for sufficiently small K so that 7r* £ K = £ for the
projection 7r : Sh -+ Sh K ;
(s2) There exists an open semi-group ,1 C G(A(oo)) such that ,1 acts on
£/Sh via g* making the following diagram commutative,
1 1
Sh - - - t Sh,
g-1
where the bottom arrow is given by the action of g-1 E ,1-1 on Sh and
the top arrow is the pullback left action (because the action of G(A(oo))
on Sh is a right action). For an open compact subgroup K c ,1, we
require that the quotient £1 K under the above action k* of k E K give
rise to £K/ShK; so, we have HO(Sh K , 12 K ) = HO(K,HO(Sh,£));
(s3) We have the trace map on the cohomology groups
two conditions (sl) and (s3) are satisfied by the quotient C/ K for C = rr* Cs
as long as K c S. As we show in the following discussion, the invertible
sheaf ~i:- = ~'ShK and the locally constant sheaf L( K,; R) K = L( K,; R) / ShK
presented in the introduction satisfy these properties.
Hereafter in this section, we limit ourselves to coherent sheaves CK (topo-
logical sheaves are dealt with in Section 4.3.3). Under (sl-2), for subgroups
K and K' of Ll, we define the Heeke operator [KgK'] : H·(ShK,C K ) --+
H·(ShK"CK') associated with the correspondence (KgK') by
and put LlI)1 = (So(SJt)D . So(SJt)) n GL2(FI)1). These semi-groups are char-
acterized by the fact that 8 . U ( 01)1) 8 -1 ::) U ( 01)1) for the upper triangular
unipotent subgroup U c G (so D and LlI)1 are called expanding semi-groups
at SJt). The involution 9 H det(g)g-l = g' brings an expanding semi-group to
the opposite shrinking one. We define Llo(SJt) = LlI)1 x Ll(I)1).
the weight K, of Tc (not just the weight k of T) and the character c. This
involves a fair amount of technicality. However, once we succeed, we will have
a coherent line bundle with eigensections (of all Hecke operators) associated
with motives of the Hodge type directly related to the weight K" whose Galois
representations have determinant character dependent only on c and K,. Thus
the reward for our endeavor is worth going through this technicality. We fix
a base ring B which is often one of the following rings: the valuation ring W
of residual characteristic p > 0, its residue field IF, lFp, Z(p), Q, or C (in any
case, B is an algebra over the ones specified above). Thus the line bundle and
its section are supposed to be defined over a B-algebra R.
We now start our work. We have a natural map 9J1'(c, r(N)) -+ ShS(N)
bringing (A, A, ¢N) to (A, >:, 7]) so that 'f) mod S(N) gives rise to ¢N. On
9J1'(c,r(N)), we have a well-defined invertible sheaf ':':l..k. Since ShS(N) is the
union of its geometrically irreducible components 9J1'(c, r(N)), we have well-
defined ':':l..~(N) for N ~ 3 (and its cuspidal counterpart ':':l..k,S(N)) on ShS(N),
which after taking pullback to Sh, gives rise to a line bundle ':':l..k defined over
Sh. This sheaf satisfies (sl) and (s3), but the resulting sheaf ':':l..k and the Hecke
operators acting on it would depend on our choice of the action of d; so, we
need to clarify this to have a clear picture of the modular forms on Sh and
the Hecke operator action.
Let us explain this process of defining ':':l..k more carefully. We first choose
a complete representative set {c} of strict ideal classes of F. We choose c
prime to p. Once c is chosen, we can always specify a c-polarization A in
the polarization class >:. The endomorphism t(a) E End(A) (a E F) acts on
A by t(a)* A = a 2 A for a test object (A, A, 'f)(p)). Thus even if we take a c-
polarization A, for a unit f E Ox, (A, t(f)* A, 'f)(p) a f) is another test object
with c-polarization representing the same point of Sh(p).
The problem in defining the action g* on ':':l..7Sh is that each point of Sh~)
corresponds to an isogeny class (not a single abelian variety); so, we have
inherent ambiguity coming from suppleness of isogeny. In other words, the
point here is to associate without ambiguity a section w k of ':':l..k with each
section w of HO(A, [lA/S) for a given (A, A, 'f)(p)) E EX') (S) only depending on
the point of Sh representing the test object.
We can do this job of specifying w k functorially by (connected) component
by component; so, we may assume that>: determines a strict ideal class of
c. Then we choose a representative A with the given polarization ideal c.
Let L = ()-l ED c 1 C V and A : L 1\0 L ~ c* be induced by the pairing
((a, a'), (b, b')) = a'b - ab'. We have a canonical alternating pairing (-, -).>- :
V(p) (A) x V(p) (t A) -+ FA.'poc) , which via A and the level structure 'f) induces
a linear map A(p) : ,£(p) I\(j(p) £Cp) ~ (?,)(p). We consider a rigidity condition:
(4.38)
4.2 Hilbert Modular Shimura Varieties 167
global sections over Sh~) are made up of global sections over Sh~; fixed
by K. Similarly, if we can define a good action of Ta(Zp) x G(A,.(p»)+ on
I£.k over the Igusa tower 19I Sh(p) , we can descend I£.~ I 9 to I£.} for K with
Sl(pn)p C Kp c G(Zp).
For the moment, we assume that the base ring B is W. To define the desired
W-integral action of ..1, starting with a K 1-test object (A, A, 'i'j(p) 0 g-l), we
define (Yg , Ag , 'i'jg (p») E £~Kl for 9 K1 = gK1g- 1 equivalent to (A, A, 1](p) 0 g-l)
uniquely in the same manner as above, replacing 1](p) by 1](p) 0 g-l, since we
have the left (inverted) action g-l of 9 in (s2). We need to assume that g-l
for 9 E ..1 has an associated etale isogeny a g : (A, A, 1](p) 0 g-l) -t (Yg , Ag , 1]~p»)
for all p-ordinary test objects A = (A, A, 1](p»). Here we say A p-ordinary if
A is ordinary at p. Also we mean by an "etale isogeny" a g : A -t Yg an
isogeny given by /3 0 a- 1 for etale finite morphisms A' ~ A and A' ~ Yg .
Let ..1 o(pool)'l) = nr
..10 (prl)'l). If 9 E ..10 (pool)'l) , then a g is an etale isogeny.
Thus we assume that ..1- 1 is contained in
(4.39)
where ip : /Lp 0 D- 1 y AlP] is a p-ordinary level structure, and S runs over all
W-schemes. Strictly speaking, the isogeny a g can be defined over a faithfully
flat covering S' / S, which could be inseparable (see Section 4.4.2 for separa-
bility of a morphism of schemes).
1 1
where the bottom arrow is given by [x,h] H [X,hg-l].
If B is a Q-algebra, we can ignore the condition related to v = p (in
particular ip) in (4.39), because ordinarity and etaleness in (4.39) are always
satisfied in this case. Taking now B = Q and .::1 = G(A)+, by the same process
as above, we can define the action of G(A)+ on i,<d.k over the Shimura variety
Sh/ Q . If B is a W-algebra, similarly, we have a well-defined left action of
G (A (poo)) on differentials over S h;% (and I 9/w ).
The problem we now encounter is that the action of .::1 we have so far
defined may not factor through G(A)+ /Z(Q). In other words, the descent to
Sh K we desired to have is for an open compact subgroup K of G(7Lp x A(p))+,
but so far, we described it for K inside G(7L p) x G 1 (A (poo)) x G(JR)+. To make
this wider descent allowing all K(p) in G(A(p))+, we want to have the action
factoring through K/(K n Z(Q)) in order to make the quotient of the sheaf
over Sh by K. We have an isogeny E : (A,C 2A,1],EW) -7 (A,A,1] 0 E,W) for
E E Ox; so, E . W = EW. Thus the problem is to somehow modify the unit
action on differentials w@k H EkW@k for E E K n Z(Q) C Ox. This problem
can be solved if the base ring is complete; so, we assume that B = e or
W = ~nW/pnw. We write v for the place p or ()() (of Q), and use the
symbol B to indicate the base ring depending on our choice of v; so, B = e if
v = ()() and B = W if v = p.
To state the solution (found first in [H91] Section 3 and later developed and
brought into the present form in Fujiwara [Fl]), recall the diagonal torus T c
of G; so, Tc ~ T2. We choose a character wE 7L[I] = X(T) and a continuous
character C : Tc(i) -7 W X such that c+(E)-lE k - 2w = 1 for all units E, where
C+ : Z(.~) -7 W X (for the center Z of G) is given by c+(z) = cl(Z)c2(Z) writing
c(diag[a,d]) = cl(a)c2(d). Define C : T(Z) -7 W X by c(z) = C2 1 (Z)cl(Z).
We assume that c+ extends to a Hecke character c+ : Z(A)/Z(Q) -7 ex with
c+(x oo ) = x;:-k. Here W is equipped with the discrete topology; so, cJ is
of finite order, although c+ extended to Z(A) could be of infinite order. In
particular, C factors through (0 /91)X for an ideal 91.
We fix a prime element Wq for each prime ideal q. We suppose that wp = p
for pip (because pis unramified in F/Q). Sending x E F qX to (u, w~) E O~ xw~
if x = uw~, we have the identification Fqx = O~ x w~. Hence T(A(oo)) =
T(Z) x (EBq w~). Recall that v = p or ()() is the fixed place of Q. We now
extend our Neben character c = (cl,c2,c+) to T(A(oo)) as follows.
(exO) For each wE 7L[I], we extend the character xp H x"; oJT(7Lp) to T(Qp)
trivially on EBp Ip w~. Thus p-2! = 1.
170 4 Hilbert Modular Varieties
for 6 = (~ ~) E LlO ( (v) n 91) U LlO ( (v) n 91) ~. Here (v) n 91 = (p) n 91 if
v = p and (v) n 91 = 91 if v = 00.
(ex3) Since the character Z(A(oo))/Z(Q) ::1 z f-+ Iz(oo) 1.z2z;;-21 c+(z) and c~
on So((v) n 91) coincide on So((v) n 91) n Z(A(oo)), we may extend
c~lso((v)n'l1) to cs = cs: Z(A)So((v)n91) = Z(A)Sl((V)n91) -+ B X by
cs(zs) = Iz(oo)I.z2z;;-2Ic+(z)c~(s) for z E Z(A) and s E Sl((V) n 91).
If we need to emphasize the dependence on v, we add the superscript v, as in
cs for Cs. Note here that cs depends on v but the central character c+ and
cJ (j = 1,2) do not. Also cl(wq) could be nontrivial, but c2(wq) = l.
By computation, we have
We define new actions w®k f-+ gw,c,vw®k of 9 E Z(A)So(91) and w®k f-+ g*w®k
in (s2) of 9 E Llo(91) (or 9 E Llo(91)~) by
The first action is a left action, and the second is considered to be a right
pullback action associated with Sh ::1 x f-+ xg~l E Sh for 9 E Llo(91). These
two actions coincide on So(91) n Z(Q) = Z(Z) and commute each other, and
hence the action of [KgK'] with respect to the second action is well-defined
on the cohomology group of the coherent sheaf on Sh K obtained from f!d.7Sh
by the descent via the first action. For the second action to be valid, one
has to assume that a g is an etale isogeny (which always holds if v = (0).
e
Then, writing ~(voo) for ~(oo) or poo ) accordingly as v = 00 or p, we have
~w,c,vw®k = 1~(voo)I.z2~;;-2Ic+(~(CXJ))~1~~2w~kw®k = w®k (for ~ E O(v)+) by
the product formula 1~(vCXJ)ll~;I = 1, and we are done. Here O(v)+ = 00,)+ if
v = p and O(v)+ = F"; = 00. The action of K(p) is well-defined over W,
if v
because det(g)~ = 1 if gv = l. We have K(p) / Kl ~. K/(Kl X Kp), and the
action of K(p) / Kl is independent of v. Using the action w®k f-+ gw,c,vw®k, as
4.2 Hilbert Modular Shimura Varietie:; 171
The results of (1) and (2) in Theorem 4.24 give us a natural integral
structure over W of the p-adic sheaf "l..:/w and the Archimedean sheaf "l..:/c
over Sh~) (as long as they are well-defined). In other words, we have
172 4 Hilbert Modular Varieties
(4.43)
at the same time. When E restricted to T(Z) is the identity character 1 and
E+ is a power of the modulus character I . lA, we drop E from the notation,
because then K determines E. As we observed already, when k = mI (m E Z),
we had a well-defined ~:;"{,, and we can easily verify that ~/0 = ~:/w for
K = (0, (m - 1)1) and E2 = 1 and E+(X) = Ixli- m . In some of the earlier
works of the author (e.g., [SGL] and [MFG]) , Ll(sn)" and E(X L ) are used in
place of Llo (sn) , and also the roles of K1 and K2 are interchanged (particularly
in the Hodge type of the motive associated to cusp forms of weight K below).
In other words, our formulation is cohomological (and hence Hecke operators
act from the right, because cohomology functors are contravariant), while the
formulation in [SGL] is homological; so, the Hecke operators act from the left.
The two formulations are related by [KgK'] B [(KgK')"l, and the role of K1
and K2 is interchanged in the homological formulation (thUS (m, n) = (K2' Kd
under the notation in [SGL]).
This double-digit weight K has a geometric meaning that the motive M f
associated with a Hilbert modular Hecke eigenform f E HO(Sh K , ~K'E:,K/d
constructed by Blasius and Rogawski ([BlR] for k > 2, [H81] and [Bl] for
k = 2) has Hodge weight (K2,0-,K1,0-) and (K1,0-,K2,0-) at the Archimedean
place (J of F, and by writing ~al for the €-adic Galois character associated
with the Hecke character E+, ~alN£ with the €-adic cyclotomic character N£
gives the determinant of the Galois representation of f (see [BlRl, [Tal, and
[MFG] 5.6.1 for a description of Galois representations and Section 4.2.7 in
the text for ~al).
Our definition of ES in (ex3) looks technical; in particular, the extra factor
Iz(voo)liz;I in (4.41) appears to be strange, but it is placed in order to kill
the factor I det(z)IA in (4.37) (and (4.44)) for the central element z E Z(A).
This is why the central character is given by E+.
When v = p, by our definition of w : T(Qp) -+ W X , g* is well-defined
for all g in the semi-group in (4.39). Still the multiplication by I det(g)IA in
the definition of [KgK'] may not be well-defined in characteristic p (because
Ip(oo) IA is not p-integral). We justify this later by means of the q-expansion
principle.
We now relate the new automorphic definition of Hecke operators to the
geometric one already given in Section 4.1.10. Let K1 = K n (G 1(Qv) x
G(A(v))). Decompose K1g. K,l = Uh K 1h for g E Llo((v) n sn), which yields
KgK' = Uh K . h. Then in the definition (4.37) of the operator [KgK'l,
the factor 'IrK' /(KgnK') ° [g] in down-to-earth terms is to average the image
h*(fw 0k ) over the coset representative h. Strictly speaking, this statement is
only valid when the degree of the isogeny cx g is invertible over the base scheme
S of the test object; otherwise, we need to stick to 'IrK' /(KgnK') because
ShKgnK-/ShK' (or IgKg nK , /Ig K') could be inseparable. If K· h = K· h', we
have (T/oh- 1 )oK = (T/Oh,-l)oK as K-orbits. Thus we have a well-defined value
4.2 Hilbert Modular Shimura Varieties 173
f(A,"X, 'r/ 0 h- 1, w) for modular forms f of level K, because (A, "X, 'r/ 0 h- 1, w)
is a level K test object if so is (A, "X, fj, w). We suppose that 9 E <10(1).
We recall our construction briefly: for a test object (A,A,'r/)/s, the subgroup
C = 'r/(h- 1LIL) for L = 'r/-1(T(A)) is a subgroup of A defined over S. Let
Yh = AIC, and write ah : A --+ Yh for the projection. We thus have (Yh, Ah, 'r/h)
with 'r/h = ah 0 'r/ 0 h- 1 and Ah induced by ah from A. Write x E Sh(S) for
the point represented by (A, "X, 'r/). Then xh- 1 is represented by (Yh , Ah, 'r/h).
Pick Wx E HO(A, [lA/S). Then Wxh-1 = h- 1 . Wx for the corresponding section
over xh- 1 is given by ah*w X' For a global section s E HO(ShK,!,l) given by
s(x) = f(A,"X, fj, wx)w~k of ~l with a modular form f, we have
fl(z)v(A,"X, 'ii, wx)w~k = I det(z) IAcS(Z) (f(A, "X, fj 0 z-l, WXZ-l )w~k)
= c~(z)z;2I f(Yz , "x z , fjz, (a;)-lwx)w~k. (4.45)
If we take the shrinking semi-group <1o(SJt)' in place of our choice <1 o(SJt)
here, the Hecke operator is induced by the right action of g' : ShK --+ Sh Kg"
and we need to replace the covering (A, "X, 'r/, WXh-1) ~ (Yh , "X h , 'r/h, (a h)-lwx )
by another one (Y~,,"X~','r/~,,(a~,)*wx) Q~') (A,"X,'r/,w x ) for the operator
[(KgK')'] in (4.44) (or simply to make the following variable change, 9 H g-"
as was done in [H88], page 307). We have made our choice of the cohomolog-
ical action, because it is directly related to the classical Hecke operator T( q)
already defined in Section 4.1.10.
We assume that Si(SJt) eKe So (SJt). If we consider a test object
(A, "X, fj) sitting over a point of S'K,oo = ShK[~], we have the connected com-
ponent Co(p) of AlP], and all finite subgroups C of A with C n Co(p) =
C XA Co(p) = {O} (scheme-theoretically) are etale. Thus if the test object
is p-ordinary and sitting over TK,m,a. for a > 0, we may normalize L to
be given by 0* EB ,-1 C V with 'r/(L) = T(A), and we may assume (by
174 4 Hilbert Modular Varieties
the irreducibility of TK,m,Q proven in Theorem 4.21) that 7](1'* jO*) = Co(p)
at the characteristic 0 fiber of A. For any prime factor 1', taking the un i-
formizer wp = p of Op, we find KO(p)l (6 ~p) KO(p)l for Ko(pr) = Knih(pr)
is decomposed into Uu mod p KO(p)lhu for hu = (6';:p). Thus all etale
cyclic subgroups of order I' in A are given by 7] (h;/ LjL), and the def-
inition of the Hecke operator U(p) in (4.16) is identical to the operator
[Ko(p) (6 ~p) Ko(p)]oo = [Ko(p) (6 ~p) Ko(p)]p given by (4.44) (because
n
W;"" = 1 by (exO)). We write Tv(l, y) for [K (6 K]v, which really depends
on y E (A(oo))X (as we already remarked). We have Tp(l, y) = y;""Too(1, y).
When Sf((p) n SJ1) eKe So((p) n SJ1) and y(p'Jl) = 1, we write Uv(y) for
Tv (1, y). If YP = 1, these operators do not depend on v; so, we drop v from
our symbols if the choice v is clear from the context. Then we have
(4.46)
Note here that W';"" E W Xby the definition of W';"" in (exO) for any other
generator w~ of pOp, and hence
When a prime qlSJ1, we have a specific subgroup Co(q) = 7]((cq)* jc*), and
recalling the uniformizer Wq of Oq, we find
The operator Tv(l, y) with YP = 1 does not involve y;"" and lyplA =
I det(diag[l, y])IA; so, it is integral over W. We record what we have proven.
Corollary 4.25 Let the notation and the assumption be as in Theorem 4.24
(2). Suppose that c = C21cl factors through (OjSJ1) X • Then we have W-
integral line bundles !;d.""E: C !;d.~ satisfying the three axioms (sl-3) with an
action of Llo((p) n SJ1) in (s2) specified in (4.42) for the coefficient rings W
and <C. These line bundles descend to Sh~jw for K C Ll o(SJ1) maximal at p
as long as Sh(p) jSh~) is etale. The Hecke operators Tp(l, y) = Too(l, y) with
yp = 1 are W -integral.
In Theorem 4.28, we describe when W-integrality for Tp(l, y) and Up(y) with
nontrivial YP holds.
4.2 Hilbert Modular Shimura Varieties 175
We have constructed ~:,K for a closed subgroup K of ..:1 = ..:1 o((p) n 5)1). If
K(p) contains the center Z(A(pOO)) , the center acts on the sheaf ~ K by the
character c~, and we may regard ~ as a line bundle defined over th~ Shimura
variety Sh(PG, X) of the group PG = Reso/zPGL(2), because the central
action is specified. In this subsection, we construct a slightly larger family of
sheaves on Sh(PG, X) containing all ~: so far constructed.
We interpret the sheaf~: as a sheaf over Sh(PG,X). For a E Z(A(oo)),
writing 0 for F n aO, we find 7(A 129 0) = 7(A) @z 0 and 1]-l(7(A 129 0)) =
a1]-l (7(A)) for a test object (A,.x, 1]). Tensoring A with 0 c F :J 0, two
maps A 129 0 -+ A 129 F +-- A induce an isogeny a E Hom~(A 129 0, A). We
get (A, X, 1] 0 a) ~ (A 129 0, a*X, a- l 0 1]). This central action brings the
c-polarization to co 2 -polarization a*X as seen in Section 4.1.9.
Since the Shimura variety Sh(PG, X) is given by Sh(G, X)/Z(A(oo)), we
identify (A,X,1]oa,w) = (A@o,a*X,a-lo1],a*w) with (A,X,1]). If~ de-
scends to Sh(PG, X), the center Z(A(oo)) has to act trivially on ~~; so, we
find that k - 2w = 0 (hence k E 2Z[I]) and that 10+ is trivial. This means
K, = (I - w, w). We write the sheaf ~F-w,w) as ~ for w E Z[I]. We regard
we may regard M~)} as a g-torsor over the variety Sh~(A(OO») (G, X) for
g = (Z(Zp x A(poo))/Z(Z(p))) x T. Here Z(Zp x A(poo))/Z(Z(p)) is a
constant group. Take the vector bundle [2 associated with this torsor over
Sh~(A(OO»)(G,X). For each character (X,2w) : g -+ B X sending (z,x) to
X(z)x 2w (x E T and z E (Z(Zp x A(poo))/Z(Z(p)))), we can cut out a
line bundle ~~ [X]/B given by the sheaf of eigensections under character
g 3 (z,x) f--t X- l (z)x- 2w . In other words, we let the center Z(A(poo)) act
on ~:/Sh(P)(G,X) by X and descend the sheaf ~:/Sh(P)(G,X) to ~[X]/B over
Sh(p) (PG, X) by this new action.
Since Z(A(oo)) :J Z(Q) n Z(Zp x A(poo)) = Z(Z(p)) c Z(Zp x A(poo)), we
have
176 4 Hilbert Modular Varieties
Thus the characters of Z(ZpxA(P=)) are in bijection with continuous even char-
Z(z(p))
'" (0,
~c '61
0/'
'f/ = fid cI<+(w'/2,w'/2)
rv
®'1fJ ;
. 1ar 1y,
partlcu I< 0/'
~c @ 'f/ = ~cW[]
X 'f e+ IS
1
., 1.
. tnvla
(4.49)
Since K contains all the information of the center, the operator (z) (z E
Z(ACpoo»)) acts on GI«K, E; R) through the scalar multiplication by E+(Z)
(see (4.55)).
For K with Sf(SJ1) eKe SO(s:n:) , we define the Hecke algebra hl«K,E;R)
with coefficients in R by the R-subalgebra of End R (SI< (K, E; R)) in the same
manner as above.
In the following subsection, we show that the operators Too(l, w q ) and
Uoo(wq) are W-integral for almost all cases, at least if K1 + K2 ~ 1 and K is
dominant; see (hl-6) for more details.
More generally, for a subgroup K with (Sl(pT) n Sf(s:n:)) eKe SO(pTSJ1)
with r > 0, we define for a profinite W-algebra R,
(h3) R is p-torsion-free.
Suppose that SUI)1) e K e So(I)1). Then the operators (z)CXJ for z E
Z(A(CXJ)), [KgK]= for g E Ll o(l)1) with gp = I, and TCXJ (I,p) are R-integral
under one of the following conditions.
(h4) We have R = H/pmH for a W-algebra H, the Hecke operators are well-
defined for G",(K,c;H) and S",(K,c;H), and Lc!~ is very ample on the
minimal compactification Sh*x:/13 of Sh K / 13 ;
(h5) R = W or Wm = W /pmw, ""1 + ""2 2' 0, and st(l)1) eKe So(I)1);
(h6) R is a Q-algebra.
Under (h5), if we further assume ""1 + ""2 2' I, we have TCXJ(I, w-p) == Up(w- p )
mod p in h", (K n So ((p)), c; W), where Up (w-p) is defined on the bigger space
G",(K n So(p), c; W) than G",(K, c; W) on which Too(l, w-p) is defined.
Proof. We first treat the case where (h4) is valid. Since Lc!~ is very ample, it
is generated by global sections, and G",(K, c; H) 013 R = G",(K, c; R). Since
G",(K, c; H) is stable under the Hecke operators, G",(K, c; R) is also kept by the
Hecke operators. Once we know that G", is stable under the Hecke operators,
S'" is also stable, because the cusps are permuted by the action of JL E Ll.
To see the assertion under (hI), we consider the normalization I g K of the
minimal compactification M'K of sh~lp)xG(7Lp) in Ig K · Let S'K be the formal
completion of M'K [i 1 along the modulo p special fiber for a lift E of the
Hasse invariant. Then we put Ig'K = Ig K XM K S'K. Then S'K and Ig'K are
affine schemes, and I g'K is etale over S'K. This is because the Igusa tower
Igkm over the smooth toroidal compactification Sh~,?p)XG(7Lp) is etale and the
infinitesimal structure of the projection I gkm ---+ I g'K around each cusp of I g'K
is isomorphic (infinitesimally) to that of the projection Sh~,?p)XG(7Lp) ---+ M'K.
Note that Lc!~ for "" E (Z· 1)2 and Lc!c: '" extend to Ig'K and H°(Ig'K, £) =
H°(Igkm , £) by the Koecher principle' for £ = Lc!~ and Lc!c:,,,,. Since Ig'K is
affine, we find FK(K, c; H) 013 R = F",(K, c; R) for F = G and S. Then we
conclude the assertion in the same way as in the case of (h4).
The operators may not be well-defined, because of the factor I det (g) IA in
(4.37), (4.44), and (4.45) (see also (4.16) and (4.17)). If R is a Q-algebra, this
does not cause any problem; so, the assertion under (h6) follows.
To show the assertion under (h2), (h3), or (h5), we apply the q-expansion
principle as described at the beginning of this subsection and Corollary 4.23
to verify that fl [K gK] and fl (z) are R-integral as long as f is R-integral. We
first compute the q-expansion of fl (3) for f E Sk( c, rl (1)1); R) and a fractional
O-ideal3 prime to p. Since we see from (ex3) applied to z E Z(A(PCXJ)), the
factor I det(z)IA = Izli in front of the summation symbol Lh is canceled out
by the factor Izl.,;:2 in (4.41), we only need to write explicitly Tatea,b(q) 03- 1
for 3 = zOnF in order to compute the q-expansion at the cusp (0, b) of fl(3).
Since Yz in (4.45) is given by A03- 1 = A/A[3], for an integral idele z with
3 = F n zO, we find
180 4 Hilbert Modular Varieties
Tt ,o.._1_Gm00*/qJ-lbc,;Gm0(JO)* T ()
a en ,b()
q '<YJ - /-LJ - qJ Ib = ate'n,,-lb q,
0 0
which implies
(4.53)
sending (x, y) /\ (x', y') to x'y - xy', and 'T]can is the projective limit of the
level T(N)-structure
dt
= f(Tateo,c-l (q), a- 2 A can , 'T]can 0 a-I, a-IT)
k -2 dt
=a f(Tateo,c-l (q), a A can , 'T]can, T)'
4.2 Hilbert Modular Shimura Varieties 181
and the factor N(O'.)-2 comes from the shift by I det(O'.(voo))IA in (4.45) of the
action of (O'.(voo))v' For general r\" the action of O'.w,c,v is trivial,
For the sheaf ~ with general r\, and c, since we have the central character c+,
for more general Z E Z(A(oo)) with Zv = 1, we find from (4.45),
(4.55)
This is the reason why we call c+ the central character. Also the two identities
(4.54) and (4.55) for Z = O'.(v) coincide.
Although we do not need this, we now insert the computation of the
effect of U(p) for I,!dmI on the q-expansion, using (4.17), because I,!dmI ex-
tends naively to Sh K giving rise to ~ for r\, = (0, (m - 1)1), C2 = 1, and
c+(x) = Ixll-m. In this special case, (4.16) yields the correct result. By def-
inition, Tateo,c-,(q)[p] is isomorphic to /-lp x q(pC)-I/C- ' . The nonconnected
subgroups over Ware given by Cc, = {(mQmlm E 0/p} for ( E ILp and
a generator Q of qpc- ' /c- l . Thus we see Tateo,c-' (q) /W[/LP][[q~ll~E,-1 ICc, =
Tate(pc)-I,o((q)/W[/Lp][[q<lleEP_I,_I' Writing ap-Ic-l (~, f) for the q-expansion
coefficient of f E GmI(rl (cp, l)1); R) at the cusp (p-1 c-1, 0), we find the
q-expansion coefficient of fIU(p) at the cusp (c-1, 0) is given by
By the orthogonality relation of characters ( I--t (~ of /-lp, we find that only the
coefficient of q~ for ~ E c- 1 (not for ~ E p-1 C-1 - C 1 ) survives and is equal
to ap-Iel (~, f). Thus, the division by N(p) is canceled if R is p-torsion-
free, and the operator U(p) is well-defined on GmI(rl (cp, l)1); R) even for
R with nontrivial p-torsion if GmI (rl(cp,l)1);R) = GmI (rl(cp,l)1);W) ®w
R. In particular, if m » 0 so that I,!dmI is very ample, U (p) is well-defined
over R. Since the ordinary locus S;,. of the minimal compactification M;" =
M* (c, rl (l)1)) is affine, by Koecher's principle, we have
HO(S'/x,/R,l,!dmI) = HO(S~/R,l,!dmI)
= HO(S~/w,l,!dmI) ®w R = HO(S'/x,/w,l,!dmI) ®w R.
Thus the Heeke operator U (p) is well-defined on the space of p-adic modular
forms with coefficients in any p-adically complete W-algebras R.
182 4 Hilbert Modular Varieties
(4.56)
In general, we have
(4.57)
Now we generalize the above formula to 1.Jd.~ using (4.44). Recall the prime
element Wq we have chosen in (exO) so that E2(wq) = w~' = 1. Then by
(4.41), we find that E~ (6 -d:q ) = 1. If w = Wq, then by (4.44), the definition
of Uv (wq) is identical to (4.16). Then the same computation for the sheaf 1.Jd.~
yields
(4.59)
For a prime ideal q outside (v) n 91, we now perform a similar computation
for the sheaf ~ on which .do(91) acts. In this setting, Too (1, 'CDq) = Tp(l, 'CDq);
so, we may assume that v = 00. We implicitly suppose all order q subgroups
of test objects are etale (though after computing q-expansion, we will know in
many cases that the formula is valid even if it is not the case). The central el-
ement z E Z(A) acts by the character Z H Iz(00)1A: 2z;2Ic:;'(z(00)) (see (4.41)).
By the decomposition Kq (6 ~q) Kq = Kq (~q U (Uu n mod q Kq (6;q)),
the latter half has the same effect as c~ (6 ;q ) U ('CD q)
for c~ (6 ;q) =
C2('CDq) = 1; so, we need to compute the effect of the first coset Kq (~q n.
Write h for (~q n. Then h = 'CDqh' for h' = (6 w~l). By (4.41), c~(h') = 1.
The central action of 'CDq gives rise to the multiplication by N(q)2c+('CDq),
because f=d.;,x is the quotient of f=d.k by the action w®k H gw,E,vw®k. Thus
I det(h)IAN(q)2c+('CDq)c~(h') is integral if /';;1 + /';;2 2: O. The effect of h' can
be computed by the etale isogeny,
- dt - ,-I dt
(Tateo,qc-l (q), Acan , 'T/can, t) '" (Tateo,c' (q), Acan , 'T/can 0 h 't)'
we get, if q t 91 is a prime,
(4.60)
184 4 Hilbert Modular Varieties
because Wq,v = 1 if q is prime to (v) n 91. The sheaves id k and id~ have the
same pullback map g* for 9 E Llo(91) only when k = mI and (El' E2' E=;O) =
(l'I Am , 1, 1·1~-2). In this special case, it is easy to see that the formulas (4.60)
and (4.59) match.
We assume (h5). Since we have [KgK]oo = [KgK]p if gp = 1, the assertion
for [KgK] with gp = 1 follows from the argument for (h2). Thus we only need
to take care of Too(l, wp) for pip. We see that N(p)E+(wp) is p-integral as
long as 1>:1 + 1>:2 2' 0 and that N (p )E+ (wp) == 0 mod p as long as 1>:1 + 1>:2 2'
I, because E+ is the finite part of an arithmetic Hecke character of infinity
type -1>:1 - 1>:2 + I. The operator Too (1, wp) is originally defined only over 1Ql.
However by the above formula, if 1>:1 +1>:2 2' 0, it sends the W-integral subspace
GK(K, E; W) into itselffor K with S£(91) eKe So(91). Since N(p)E+(wp) ==
o mod p if 1>:1 + 1>:2 2' I, Too(l,wp) == Up(wp) mod p by the q-expansion
principle (Corollary 4.23 (1)). Thus for K with S£(91) e K e So (91), if
GK(K, E; W) ®w R = GI«K, E; R) and 1>:1 + 1>:2 2' 0 (under unramifiedness of p
in F/IQl), the operator Too(l,wp) is actually well-defined on GI«K,E;R), and
Too(l, wp) is equal to Up(wp) if 1>:1 + 1>:2 2' I and pR = O. D
We record what we have seen for p-adic modular forms in the above proof.
Corollary 4.29 Let T be a group with rl (91) eTc Fa(91). For each p-adic
W -algebra R, the space V(T; R) of p-adic modular forms defined in (4.20)
is stable under the Heeke operators Tp(l, y) with YP = 1, (Y)p for yEO n
(Fipoo)y and Up(Y) for y E Fpx n G p.
By the universality of V (c, T; W) (which classifies (A, X, ¢;, ip) / R for p-adic
W-algebras R), we find V(c, T; R) = V(c, T; W)0wR; so, the assertion for R
follows from the same assertion for W.
(4.61)
where the last identity follows from K:1 = I - w. Note that E 1 ( Ep 'l1) = E 1 ( E) by
(ex1) and that E+(E) = 1 for a unit E E O~. This shows
(4.62)
(4.63)
The definition (4.63) does not depend on the choice of c. Indeed, taking an-
other c' and c' = F n c'O with y = u'(c', by Tateo,e-l(q) ~ Tateo,e,-l(q)
given via t r-+ t ro for w = ~-l( (which pulls back '¥ on Tateo,e'-l (q) to
('b
w,¥ on Tateo,c-l (q)), we have ac- 1 (~; fe I (g ~)) = ae,-l ((; fe' I ~)), be-
cause uu,-l = w- l which compensates the action of t r-+ t W on the level
structure. By definition, if f E Vcusp(Fl(91),c;B)[Kl,cd, then ap(uy;f) =
U:;;t;;l Cl (u)ap(y; f) for u E TeL). Thus a p is well-defined independently of the
choice of c, and for an integral ideal n prime to p and the conductor Itl of Cl,
choosing a finite idele n so that nO = n, ap(nCpctlly; f) is well-defined inde-
pendently of the choice of n. We write ap(yn; f) = ap(nCpctlly; f). We extend
the function a p outside integral ideles by defining the value to be O.
The above definition of ap(Y, f) extends without modification to p-adic
modular forms f in Vcusp (Fl (91), C; B) regarding Vcusp (Fl (91); B) as a subspace
188 4 Hilbert Modular Varieties
of the direct sum EBcEClt Vcusp ( C, r 1(1)1); B), because we have a naturally de-
fined action of (t ?) for u E T(i) on such p-adic modular forms f given by
fl (t ?) (A, A, il)!, ip) = 101 (u)f(A, A, il)! 0 u;r/, ip 0 U;1). By Corollary 4.23, the
p-adic modular form is determined by the function a p on integral ideles. We
have now proved
{f E Vcusp (Fa (1)1) , 10; B) lap ( uy, 1) = 101 (U)U;'" ap(Y, 1) \fu E T(Z) } .
We compute the effect of Tp('Wq) := Tp(l, 'Wq) for a prime q t 1)1 on the
coefficients a p using (4.60). We have, if q pl)1, t
ap(Y, fITp('W q)) =acl (~; fcITp('Wq)1 (t?))
=aq-lc-l (~; fqcl (t?)) (4.64)
+ N(q)c+('Wq)aqC-l (~; fq-1cl (t?)) .
Note that y = u~c implies 'Wqy = u~c'Wq and 'W q 1y = u~C'Wq1. Then replacing
C (resp. c) by c'Wq (resp. C'Wq1) in the first (resp. the second) term of the
last equation of (4.64) and applying the definition (4.63) of ap('Wqy, 1) and
ap ('W q1 y, 1), we find, for all primes q outside pl)1,
The same formula is valid for Too ('Wq) := Too(l, 'Wq) including qlp for our cho-
sen prime element 'Wq (because c1('Wq) = 1); otherwise, the formula depends
on v = p, =. Similarly, we have
( 4.66)
We now introduce the nearly ordinary Hecke algebra and prove the duality
between the algebra and the space of cusp forms.
We take limits e = lim n -+ oo up (p)n! and eo = lim n -+ oo Too (p)n! in the alge-
bra h",(K, 10; R) defined in Section 4.2.8 for a profinite W-algebra R, whenever
the Hecke operator Up(p) = Uoo(p) or Too(p) is in h",(K,c;R). We define for
K with Si(IJt) eKe So(IJt),
if pllJt,
if IJt prime to p,
(4.68)
if pllJt,
if IJt prime to p.
The algebra h~·ord(K, 10; R) is called the nearly ordinary Hecke algebra
on G of level K with Neben (or weight) character (,,-,c) and with coeffi-
cients in R. We split T(Zp) = r x ..:1(p) for the maximal p-profinite sub-
group r and the maximal prime-to-p torsion (finite) group ..:1(p). Since p
is unramified in F/Q, r is torsion-free. For each p-profinite W-algebra R,
any projective R[[T(Zp)]]-module of finite type is R[[r]]-free of finite rank.
We then define hn.ord(lJt, 10; W[[T(Zp)]]) for the W[[T(Zp)]]-subalgebra of
End(V~~~p(.fo(IJt), 10; R)) generated over W[[T(Zp)]] by the operators Tp(Y)
and (YIp for all integral ideles y.
Theorem 4.30 Let K be a subgroup ofG(A(oo)) with Sl(pr) nSi(lJt) eKe
So(prlJt) for IJt prime to p. By the pairing (t, f) = a p(l, fit) for modular forms
f and Hecke operators t, we have the following perfect duality.
(1) Suppose that K is maximal at p and that "- is a dominant weight of Tc
(<;=;. "-2 ::::: "-i). If one of the conditions (h4-6) in Theorem 4.28 is met
depending on the level, we have h",(K,c;R) ~ HomR(S",(K,c;R),R) and
S",(K, 10; R) ~ HomR(h",(K, 10; R), R) for an algebra R finite over W;
(2) If r > 0 and one of the conditions (hl-3) is met,
h~·ord(K, 10; R) ~ HomR(S~·ord(K, 10; R), R)
and s~·ord(K, 10; R) ~ HomR(h~·ord(K, c; R), R)
for a p-profinite algebra R finite over W;
(3) hn.ord(lJt, 10; W[[T(Zp)]]) ~ Homw(~':8~(.fO(IJt), c; W), W)
and ~':8~(.fO(IJt), 10; W) ~ Homw(hn.ord(lJt, 10; W[[T(Zp)]]), W);
(4) hn.ord(lJt, 10; W[[T(Zp)]]) ~ HomR(V~~~p*(.fo(IJt), 10; W), R)
and V~~~p*(.fo(IJt), 10; W) ~ HomR(hn.ord(lJt, 10; W[[T(Zp)]]), R)
for R = w[[r]].
Proof. We give a sketch of a proof. The Hecke algebras and the spaces of
modular forms involved in (1) and (2) are R-modules of finite rank. This
follows from Proposition 4.27 for (1) and Theorem 4.9 for (2).
190 4 Hilbert Modular Varieties
Since the proof of the assertions (1) and (2) is the same and (4) follows
p=
from (3), we prove only (1) and (3). By the long exact sequence of ~""E:/w -'---+
~I<,E:/w --» ~""E:/w=, S",(K,E:;W) Q9 Wm (= HO(ShK/W'~""E:/w) Q9w W m)
injects into S",(K,E:;Wm) (= HO(ShK/W=,~""E:/w=)); so, the q-expansion
principle is valid for the finite module SI«K,E:;W) Q9 W m . For the pair-
ing ( , ) : (h",(K,E:; W) Q9 W m ) x (S",(K, E:; W) Q9w W m ) -+ W m , we have
(U(Y<n)T(l, y(<n)), f) = ap(Y; f) by (4.67), and hence, by the q-expansion prin-
ciple, f = 0 if (h",(K,E:; W) Q9w W m , f) = O. By the perfectness of the duality
of finite modules, we thus have a surjective morphism
which is called the universal nearly p-ordinary Hecke algebra. In the class
of characters {s+xh we have the Teichmiiller lift So+ of s+ mod mw. By
construction, the above algebra only depends on the Teichmiiller lift So
of s mod mw of weight KO = (0,1); so, we write the above algebra as
hnord(sn, So; W[[T(Zp) x rz]]).
For an idele y, we have a well-defined ideal class y E Clj;(pOO)
FJ: /FX(O(pl)XF:;'+. We write [y] for the image of y under Z --+> rz. We
have the following consequence of Corollary 4.31 and the discussion in Sec-
tion 4.2.7.
Corollary 4.32 Suppose that p f 2Nd(F) and that p ::::: 5. Then we have
(1) The Hecke algebra hnord(sn, So; W[[T(Zp) x r z ]]) is a W[[T(Zp) x rzll-
projective module of finite type;
192 4 Hilbert Modular Varieties
(2) When "'2 - "'1 :2: 2I for", E X(To) and cITG(Z) = COITG(Z) with c+ == co+
mod ffiW on Z(A(poo)), for every arithmetic Hecke character X of rz with
oo-type 2v, we have a canonical isomorphism
which sends Tp(Y) ® [y] and Up(Y) ® [y] in h n . ord (l)1, C; W[[T(Zp) x r z ]]) to
Tp(Y) and Up(Y) in h~~(:,v)(I)1, c ® X; W). If we replace 1)1 by pl)1 in the
right-hand side of the above identity, it is valid under the milder condition
"'2 - "'1 :2: I.
Let us recall the definition of the adelic Hilbert modular forms and their Hecke
ring of level 1)1 for an integral ideal 1)1 of F (cf. [SGL] Sections 2.2-4). Let
To = G~/o be the diagonal torus of GL(2)/o; so, To = Reso/zTo. Recall the
characters c = (C1' C2, c+) as in Section 4.2.5. In particular, c+ : Z(A)/Z(Q) -t
ex is an arithmetic Hecke character such that c+(z) = c1(Z)c2(Z) and
c+(x oo ) = X-(1<1+1<2)+l. Suppose that the conductor of cJ (j = 1,2) is a
factor of an integral ideal 1)1' and that the conductor of c- = c;-lc1 is a factor
4.3 Rank of p-Ordinary Cohomology Groups 193
(4.72)
(4.73)
4.3 Rank of p-Ordinary Cohomology Groups 195
(4.74)
u,t
for finitely many u E u(i) and t E Tc(A((X))) with det(t) = det(y) (see [IAT]
Chapter 3 or [MFG] 3.1.6). Since E~ is well-defined over L'l0(1J1) and coincides
with Z H E+(z'l1)E2(det(z('l1))) on Z(A((X))) n L'lo (1J1) , we may extend it to the
subgroup of G(A((X))) generated by Z(A((X)))L'lo(lJ1) (which contains L'l0(1J1)')
so that ~(z6) = E+(z'l1)E2(det(z('l1)))E~(6) for 6 E L'l0(1J1) and z E Z(A(oo)).
We then have
u,t h
(4.75)
The second identity (*) follows from det(t) = det(y), (ut)' = det(t)(ut)-l and
multiplicativity of ~, and by this, the sum is independent of the choice of u and
t as long as det( t) = det(y). It is easy to verify that the operator defined by
(4.75) preserves the space G" (1J1, E; q and 5" (1J1, E; q by confirming (SAl-3)
for fl[5 0 (IJ1)y'50 (1J1)].
The above description (4.75) of the Hecke operator [50 (lJ1)y'50 (1J1)] looks
different from the earlier definition in Sections 4.2.5 and 4.2.6 (especially
(4.44)), because in the earlier definition, we used x H X· h- 1 (x E 5h(G, X))
instead of 9 H 9 . ut (g E G(A)). By applying the involution ", we can take
h = (ut)'. We claim that the two definitions are the same (for the choice
v = 00 in (4.44)). Since the central characters are equal, the two actions coin-
cide for each central element z E Z (A (00)). Thus in order to show that the two
operators coincide for all [5 0 (IJ1)y' . 5 0(1J1)] with y E L'lo (1J1) , we may assume
that y = (6 ~q). For such y, we have t = y' or t = Y in (4.74). If t = y', u
runs over a complete representative set for U ( 0 / q) in U ( 0 q ), and if t = y, we
may take u = 1. We take the decomposition
for Su E 5 0(1J1) and a finite idele c' with c' 5 n F = c/ , where 'Iy' = (~n
and'lu = (61) with v E c running through the congruence classes in c/qc.
Similarly we have
196 4 Hilbert Modular Varieties
X~-I(Z)
ry
and XTyL- I JU- I ( Z ) (u E U(O/q)) give all the quotients of Xz by cyclic
subgroups of order q. Indeed for, = (~ ~) E ro( c, I)1h;l ro( c', 1)1), we have a
natural inclusion £z c (cz + d)£,(z) = L, (f) and an exact sequence
(CZ+d)-1
0--+ (cz + d)£,(z)/ Lz --+ Xz ) X,(z) --+ 0
for (cz + d)£,(z)/ £z ~ O/q. Thus the two sets of abelian varieties,
coincide, where Yh is as in (4.44) made from A = X z . One can also check that
this bijection extends to include the level structures, the polarizations and
the differentials in (4.44) for KgK' = 50 (l)1)y50(1)1). Thus the new action of
[50 (l)1)y'50(1)1)] for y E Ll o (l)1) coincides with the one [50(I)1)y50(1)1)] defined
in (4.44). By linearity, this action of double cosets extends to the ring action
of the double coset ring R(50 (1)1) , Ll o(I)1)).
Each member f of 5,,(1)1,10; C) has a Fourier expansion of the following
form ([MFG] Theorem 3.10 and [SGL] Sections 2.3-4),
for a prime q f 1)1, we can directly verify by computation (see [MFG] Corol-
lary 3.11) the following Archimedean counterpart of (4.65),
if q f 1)1,
if q 11)1.
(4.77)
4.3 Rank of p-Ordinary Cohomology Groups 197
By definition, writing fy for f9 in (S2) for 9 = (y<;) ~), we see easily that
(4.78)
Then by the q-expansion principle, we find S,.,(SJ1, E; R) = HO(Shso('Jl) ' lJd.",c/ R)'
and hence the definition is compatible with the earlier one. We restate (4.43)
slightly more generally for a W-algebra R C Ql with ioo : ij '---+ C:
We recall the embedding ip : ij '---+ ijp. Then for any Qlp-algebra R, we find,
consistently with (4.43),
(4.79)
As we have already seen in Section 4.2.5, these spaces have geometric mean-
ing as the space of R-integral global sections of a line bundle of the Hilbert
modular variety of level SJ1, that is, S,,(SJ1, E; R) ~ HO(ShJ;~('Jl),lJd.""c/R) whose
198 4 Hilhert Modular Varieties
right-hand side has meaning even for W-algebras not in ijpj so, we define
8 K (I)1, cj R) for any p-adically complete W-algebra by the right-hand side.
We hereafter assume
in order to ensure the stability of the spaces of R-integral modular forms under
Hecke operators (see Theorem 4.28).
D lR ~ {M2(lR) if (J E JD (4.82)
0F,a - lHI if(JEJ-JD=JD,
where lHI is the Hamilton quaternion algebra over R Finite places dividing
d(D) and Archimedean places in JD are called the ramified places for D IF . We
can create such a quaternion algebra as follows: Choose a quadratic extension
K of F so that Kq = K 0F Fq is a field for each prime qld(D) and that
Ka = K0F,alR = <C for (J E JD. Find 6 E F such that 6 E NKqIFq(KqX) if
and only if the place q is outside d(D) and JD. We find such a 6 if and only if
the number of ramified places of D is even (see [BNT] IX.3 and XIII.6), and
D ~ {(-8bP d'p) E M 2(K)la,b E K}, where p is the generator of Gal(K/F).
By our choice of 6, we find 0 =I- x = (-8bP d'p) E Dq with N(x) = aa P +6bbP = 0
if and only if the place q is outside JD and d(D). Thus Dq is a division algebra
if and only if q divides d(D) or in JD. The map N : D ---+ F and the trace
of the matrix (-8bP d'p) are called the reduced norm and the reduced trace
Tr: D ---+ F of D, respectively. By our construction, D '-+ M 2(K) induces an
isomorphism D 0F K ~ M2(K) which sends b 0 k to kb.
We fix once and for all an extension of (J : F '-+ Q to (J : F ~ Q for an
algebraic closure F / F. We take a quadratic extension K/ F inside F so that
K 0F,a lR ~ lR x lR as F-algebras for (J E JD , K 0F Fp ~ Fp x Fp for primes
pip and D 0F K ~ M2(K). We can always choose such a quadratic extension
K as long as p t d(D). These conditions imply K 0F lR ~ <C for (J E JD.
We identify D 0F K with M 2 (K) by the above isomorphism. We fix max-
imal O-orders OD and OK of D and K, respectively, and we suppose that
OD 00 OK C M 2(OK). Here an O-order OH cHin a finite-dimensional
semi-simple F-algebra H is an O-subalgebra of H such that for every x E OH,
the commutative subalgebra O[x] cHis an O-module of finite type and
OH 0z Q = H. If OH is an O-order, its conjugate aOHa- 1 for a E H is
another O-order. There are finitely many nonisomorphic maximal O-orders
inside H up to conjugation.
For each prime ideal [prime to d(D), we fix an isomorphism OD,[ ~ M 2(O[)
so that for the p-adic place pip induced by ip 0 (J, this isomorphism coincides
with the one, OD '-+ M 2(OK) tpM) M2(Op). By means of this isomorphism,
we identify D[ with M2(F[). For an integral ideal 1)10 of F prime to d(D),
putting 1)1 = 1)10 d(D), we define
- - D -1 -D
where OD = OD ®z:2:, and 0<)10 = I1!1<)1o Ot. We set So (91) = 8 To (91)8 for
8 = ( d(d 6D)) ~) with a finite idele d such that dO nF = D. Similarly we define
Llcf(91) c D ®<Ql A(oo) so that it is the product of local components Ll t which
coincide with the local components of .1 0 (91) as long as [f d(D) and Ll t = OD,t
if [ld(D). Again we can think of the double coset ring R(Sf(91), Llcf(91)) ~
R(Sf(91), Llcf(91Y) (XL = N(x)x- 1 for the reduced norm map N : D --t
F). We have T(t) = sf (91) (~{ nSf (91) and T(t, t) = sf (91)wtSf (91) in
R(Sf(91), Llcf(91)L) for [ f d(D), because the local component Llcf (91) t at [
of Llcf(91) is identical to Ll o(91)r. For [ld(D), we take at E OD,t so that its
reduced norm generates [Ot. Then we define T(t) = -Sf(91)atSf(91) for
[ld(D), and we have
The above isomorphism brings T(t) and T(t, t) to the corresponding elements
in the right-hand side. As an operator on the space of automorphic forms,
T(t, t) induces the central action (t) of wt·
For a given ring R, we consider the following module L(",*; R) of the mul-
tiplicative semi-group M2(R). Let ",* = ("'1 + 1, "'2) and put n = "'2 - "'1 - 1 E
:2:[1], which is the restriction of ",* E X(Tc) to T c G 1 , and we con-
firm (",IT)* = k - 21 = n. The correspondence '" --t ",* is the one for
G = ResF/<QlGL(2) we discussed in Section 1.1 (see Section 5.3.2 for more
details of this fact). We suppose that n ~ 0 (i.e., na ~ 0 for all (J E 1), and
we consider polynomials with coefficients in R of (Xa, Yo- )aEI homogeneous of
degree na for each pair (X a, Ya). The collection of all such polynomials forms
an R-free module L(",*; R) of rank I1a(n a + 1).
As before, we write v for the fixed place p or 00; so, the base ring B is
W if v = p and CC if v = 00. Suppose that R is a B-algebra. Then iv((J(ov))
(which we write simply (J(ov)) for 0 E GD(A) can be regarded as an element in
M 2 (R). Take a Neben character c as in (exl-4) of Section 4.2.6 with cIT(z)
factoring through (Oj91 o)x. We define cJ (j = 1,2) as in (ex1) and extend
C to Llcf(91) by c15(o) = c2(N(0))C(a) if 0<)10 = (~~). Since 8 --t c15(8)
(4.86)
for 9 E GD(lR) and Z E 3D. We write f1,*,D for the projection of f1,* to £::[J D].
Then, if JD T 0, we define S~ (sn, 10; q to be the space of functions J on
GD(A) with values in L CiD(f1,*,D,q satisfying the following conditions.
(SB1) We have the following automorphy
When JD = J (s- JD = 0), the variety yD(S) is a finite set of points; so,
the condition (SB2) is empty. However, writing S~ (sn, 10; q for the space of
functions satisfying (SB1) in this definite case, we need to take S~ (sn, 10; q
to be the following quotient: S~ (sn, 10; q = S~ (sn, 10; q I Jv(sn, 10; q, where
Jv(sn, 10; q is the subspace made up of functions in S~ (sn, 10; q factoring
through the reduced norm map N : GD(A) -+ T(A). If f1,2 ~ f1,1 T J or 10
is nontrivial for some x E GD(A) with N(x) = 1, Jv(sn, 10; q = 0; so, no
202 4 Hilbert Modular Varieties
modification is necessary. Decomposing sf? (sn)yL sf? (sn) for y E f1{5 (sn) into
Uh hSf? (sn), we shall define the action of R(Sf? (sn), f1{5 (sn)) on S~ (sn, E; q
by
(4.87)
h h
Theorem 4.34 Suppose sn = snod(D) for an integral ideal sno prime to d(D).
Identify R(So(sn), f1o(sn)) and R(Sf?(sn), f1{5(sn)) as in (4.84). Then we have
an R(So(sn), f1o(sn))-linear embedding S~(sn, E; q y S,,:(sn, E; q for all dom-
inant regular"" with ""? - ""p :.0- I D. The image of this embedding only depends
on d(D) and is made up of cusp forms in S,,(sn, E; q new at all primes qld(D).
In particular, if d(D) = 1, the above morphism is a surjective isomorphism.
Since known proofs of this theorem require harmonic analysis and a good
knowledge of representation theory of adele groups (especially the theta cor-
respondence), the proof is beyond the scope of this rather algebraic book; so,
we quote some references where one can find a proof. The above formulation
of the theorem was given in [H81] Section 2 and [H88] Theorem 2.1, where
one can find an exposition of how to deduce this result from the original result
of Jacquet-Langlands (whose exposition can be found in [AAG] Section 10).
We hope to give an exposition of this result in a forthcoming book (through
a line closer to Shimizu's treatment).
where the right-hand side is the direct sum of the group cohomology of the r x-
module L(/'i,*c; R). Even if rs,x does not act freely on the module L(/'i,*c; R),
r
we still have Y (S) ~ Ux x \JjI D for finitely many x with xp = 1, and we
define the left-hand side of (4.88) by the right-hand side of (4.88).
If S c S', we have the trace map Trsl IS (i.e., the transfer map in group
cohomology; see [MFG] 4.3.1) and the restriction map Ressl IS:
We can always choose a multiple 91' = 91q (by a prime q) of 91 so that O,x (91') r
acts freely on JjID.
We have a natural action of the ring R( sf (91), L1~ (91)) on the cohomology
group Hq (YOD (91), L(/'i,*c; R)). The definition of the action is quite similar to
the one (4.37) given for the coherent cohomology groups HO(ShK,id.~), and
the operator induced by the double coset SgS for S = Sf (91) is given by
Here [8] : L(/'i,*c; R) --+ L(/'i,*c; R) is induced from the action of 8 on L(/'i,*c; R).
If q = 0, decomposing Sf(91) (g n Sf(91) = U"" wSf(91), we have
(4.90)
(4.91)
in (S~(p91) (Po" n
We assume that R = Wand v = p. We have pg right cosets (g = [F : Q])
S~(p91)) /S~(p91), and each coset acts on f E Iv(91,c;R)
by f(x) H f(xp p), regarding f as a function of T(/li/ oo »). Since factually
factors through a finite quotient of T(A(oo»), for a high power U(p)n = U(pn)
of Up(p), we have fIU(pn) = png f; so, Iv (91, 10; R) is killed by the p-ordinary
projector e. Thus we have
(4.92)
where the superscript or the subscript n.ord indicates the image of the projec-
tor e = limn-+oo U(p)n!. Since T(p) == U(p) mod p if K,2 - K,1 ::::: I and [K,] ::::: 1,
the same assertion holds for 91 prime to p:
(4.94)
dime SJI D. Then by the compactness of YoD un), the cup product pairing in-
duces ([H94] (5.3)), a nondegenerate pairing
Thus we obtain from Proposition 4.33 and Th80rem 4.36 the following result.
Proposition 4.35 Let V c Hq (YoD ('J1), L(K,*c; W ®z Q)) be as above Jor
q = lID I ::; 1. Suppose F oj Q. Let V* be the dual W -lattice oj V in the
cohomology group Hq (YOD ('J1), L(f:Zc 1; W ®z Q)) under the Poincare duality
pairing ( , ) as above. Then we have an h" ('J1, c; W) -linear map J : V ®w V* --+
S,,('J1,c;W) defined by the q-expansion: J(v ® w) = Lt;»o(vITp(~y),w)qt;,
where we regard the space V ®w V* as an h,,('J1, c; W)-module through the left
Jactor V.
A similar fact for the matrix coefficients of T(y) in place of (vIT(y),w) has
been proven in [Sh8] II, Theorem 3.1 by analytic means without using the
Jacquet-Langlands correspondence.
where S~ (91, t'; q is made up of complex conjugates of elements in S;z (91, t'; q
with complex conjugate character t'.
for 91 prime to p.
If the assertion of the theorem holds for 91 containing a sufficiently large
integer N, the assertion holds for all 91, because by the theory of primitive
forms [MFM] 4.6 and [H88] Section 3 (or the strong multiplicity one theorem
[AAG] Sections 4 and 10), one can recover the dimension of each space with
lower level. Here we give a proof under the assumptions either that p ?: 5
and pis unramified in FjQ or that YoD(91) is smooth and the sheaf L(/'i,*C:; R)
is well-defined over YoD (91). The latter assumptions are valid for 91 having a
sufficiently large integer N.
Proof. We choose an open compact subgroup K C So(91) so that yD(K) is
a smooth manifold and that the sheaf L( Ii* c:; R) is well-defined over Y D (K).
We write Y = yD(K) and Y' = yD(K n So(p)). When Yl(91) is a smooth
manifold and L(Ii*C:; R) is well-defined over Yl(91), we take K = Sf(91). By
the same argument as in the proof of Theorem 4.9, the evaluation at (1,0)
of polynomials in L(Ii*C:; IF) gives rise to a morphism L(/'i,*C:; IF) ---+ L('KZ, IF)
of sheaves on Y', where K, = (0,1) and E is as above. Composing with the
restriction map, again by the same argument as in the proof of Theorem 4.9,
we get the following isomorphism, if /'i,2 - /'i,1 ?: 21,
If 1i2 - 1i1 > I, we only have H~.ord(Y" L(Ii*C:; IF)) S:! H~.ord(Y" L(K,E; IF)). If
1i2-1i1 = I, L(Ii*C:; IF) = L(K,E; IF) (although L(Ii*C:; W) might be different from
L(K,E; W)), and we have H~.ord(Y" L(Ii*C:; IF)) S:! H~.ord(Y', L(K,E; IF)). Thus we
need to choose K so that the dimension dim S;Z,n.ord(91, C:; q only depends
on dimJF H~.ord (Y, L( Ii* c:; IF)). We specify K later.
4.3 Rank of p-Ordinary Cohomology Groups 207
Thus we need to prove the constancy of rankw H~.ord (Yl (1J1), L( /'\:* c; W))
and that H~.ord(Yl(IJ1), L(/'\:*c; W)) is W-free. Take a prime q plJ1 such that
p f (N(r)2 - 1). As we have already remarked in the proof of Theorem 4.10,
there exist infinitely many such primes [ if p ;::: 5 and p is unramified in F /Q.
We take K = So(lJ1) n SHr). Then HQ(Yl(IJ1), L(/'\:*c; R)) is canonically a
direct summand (as Hecke modules) of Hq(y, L(/'\:*c; R)) by the restriction
map for all W-algebras R, because the restriction followed by the transfer
is a multiplication by an integer factor of [So(IJ1),K]I(N(r)2 -1) (cf. [MFG]
Proposition 4.21), which is invertible in R. Thus we only need to prove these
facts for H~.ord(Y' L(/'\:*c; W)).
We may further choose [so that the image ['x of Tx = x . K X-I n G D(Q) in
GDd(Q) = GD(Q)/Z(Q) is torsion-free for all x E GD(A(oo)). Indeed, if'Y E Tx
is torsion in ['x, Fb] is a quadratic extension of F with 'YN = E E Ox for the
order N of the image of'Y in ['x' Write M for the extension of F obtained
by adding Vu for all u E Ox. Since [Ox: (OX)2] = 29, [M : F] S 29. In
M, ( = 'YJE- I has finite order N; so, ( is a root of unity in Mb]. Since
E == 1 mod [, we may assume that ( == 1 mod..c for a prime factor ..c of [
in Mb]. This shows either ( = 1 or £IN for the residual characteristic £ of L
This never happens if £ is sufficiently large, because N is bounded. If ( = 1,
then'Y has two eigenvalues ±yiE; so, yiE == -yiE == 1 mod [, which happens
only if £ = 2. Thus we have plenty of choice of [ as above. Under the above
208 4 Hilbert Modular Varieties
invariant under GL2(Op); so, P"(X,, , Y,,) does not involve a monomial of Y",
that is, P,,(O, Y,,) = O. This shows the desired vanishing. In this proof of
H~.ord(Y' L("'*E; R)) = 0, we have used the strong approximation theorem
to show that T(pn) and U(pn) for a suitable n > 0 preserve each connected
component of Y; so, obviously the argument does not work if ID = 0 where
the strong approximation theorem for G D ,l(A(CXJ)) (for the derived subgroup
G D,l = Ker(N: G D --+ T) c GD) does not hold.
We find H~.ord(L("'*E; W)) ®w Wm = H~.ord(L("'*E; W m )) by the vanish-
ing of H~.ord(L("'*E; W)). Again by the exact sequence
We give a brief exposition of how one can axiomatize Galois theory (the idea
is due to Grothendieck, and the main reference is [SGA] VA). We assume a
minimal knowledge of category theory and terminology which one can find in
[HAL] Chapter II, [GME] Sections 4-5, and [MFG] Section 4. This subsection
interprets the result in Section 2.3.1 in terms of category theory.
We recall some terminology in category theory. Let C be a category. A
morphism X ~ Y in this category is called an epimorphism if for any third
object Z and morphisms Y J!+ Z, 9 0 f = h 0 f {:} 9 = h (so f is a
h
right unit). In other words, for any other object Z in C, the pullback map
Homc(Y, Z) ~ Homc(X, Z) is injective if f is an epimorphism. Similarly a
morphism X ~ Y in this category is called a monomorphism if for any third
object Z and morphisms Z l4 X, fog = f 0 h {:} 9 = h (so f is a left unit).
h
Let SETS be the category of sets. Recall that a covariant (resp. contravari-
ant) functor F : C --+ SETS is called representable if there is an object Fin C
such that we have a system of functorial bijections ¢x : F(X) --+ Homc(F, X)
(resp. ¢x : F(X) --+ Homc(X, F)). Here the word "functorial" means that
f. 0 ¢x = ¢y 0 F(f) in the covariant case and 1* 0 ¢y = ¢x 0 F(f) in the
contravariant case for any morphism X ~ Y in C, where f* (g) = fog for
210 4 Hilbert Modular Varieties
(4.96)
where r runs over all normal subgroups of finite index in 7ri OP (X, x); in other
words, the algebraic fundamental group 7r1(X,X) is the profinite completion
of the topological fundamental group 7ri OP (X, x). If we pick another point
x' E X(IC) and choose a path IJ from x to x', we find 7riOP (X, x) ~ 7ri OP (X, x')
by T f-t IJTIJ-I, and IJ induces a unique isomorphism of functors Fx ~ Fx"
If we have a morphism f : X -t Y, the pullback 1*T = X Xy T for
an etale covering T/ y is an etale covering of X, because a pullback of a
faithful flat morphism is faithfully flat (see [ECR] Corollary I.3.6). If X and
Yare connected, obviously we have Fx 0 1* ~ FfCx) for any geometric point
x E X(Sl), because an etale covering of a connected scheme X is determined
by its fiber at any given geometric fiber ([ECR] I.3.12 and [SGA] I.5.5). Then
1* is an exact functor from ET/ y into ET/ x .
Suppose now that X and Yare connected locally Noetherian. Then
Fx 0 1* gives a covariant functor from ET/ y -t SETS. By definition,
7r1(X,X) = Aut(Fx) , and each IJ E 7r1(X,X) induces an automorphism of
Fx 0 1* ~ FfCx). Thus Aut(Fx 0 1*) = Aut(Ff(x)) = 7r1 (Y, f(x)), and we
have a homomorphism of profinite groups, 7r(f) : 7r1 (X, x) -t 7r1 (Y, f(x)). In
other words, the association (X, x) f-t 7r1 (X, x) is a covariant functor from the
category of pointed schemes into the category of profinite groups.
Using this, we compare fundamental groups of the fibers of X 1-t
Y and
those of X and Y. We always suppose that X and Yare connected locally
Noetherian and that f is proper, surjective and separable with f*(Ox) = Oy
(i.e., each geometric fiber of f is connected; see [ALG] III.11.3 and [SGA]
X.2). A scheme over a field k is called separable if for any field extension Kjk,
its base change to K is a reduced scheme (no nontrivial nilpotents in the
structure sheaf). A morphism X 1-t Y is called separable if f is flat and the
4.4 Appendix: Fundamental Groups 217
fiber Xy = X Q9y Spec(k(y)) is separable for all points y with residue field
key) (see [SGA] X.I).
We choose a point y in the image of f in Y and a geometric point 'fj over y
so that its field keY) is an algebraic closure of key). Take a point x in X with
f(x) = y and a geometric point x over x so that k(x) is an algebraic closure
of k(x) and f(x) = 'fj. Then we take a geometric point a E Xy = X Q9k(y) key)·
We writei: X y "-+ X for the inclusion morphism. By the functoriality of 7r1,
we have the following sequence,
(4.97)
We gather here group-theoretic results we have used in the proof of the global
reciprocity law. Here we use the notation introduced in the proof of Theo-
rem 4.14. In particular, E is a proper subset of the set of rational primes.
In the proof, we have used the following facts for G 1 = ResF/IQJSL(2) and
PG = ResF/IQJ(PGL(2)).
(8) For open compact subgroups S,08 c G1(A,(17oo)), if a : S ~ IYS is an
isomorphism of groups, replacing S by an open subgroup, a is induced by
g(a) E Aut (Fl 17oo ) jA,(17oo)) ~ PG(A,(17oo)): IYs = g(a)sg(a)-l.
the case of classical groups (including SL(2)), this can be proven easily (see
the argument after Proposition 4.47).
8ince the assertions (82-3) can be proven for more general linear algebraic
groups G 1 smooth over the localization Z(pe) C QI at two primes p and fI, we
introduce some terminology in order to prove them for more general groups
including SL(2). For a semi-simple group Glover a field k, we take a maximal
split torus TSP C G 1 defined over k (so TSP is an algebraic subgroup of G 1
maximal among those isomorphic to G~), and take a maximal torus T con-
taining TSP. The group G 1 is split if TSP = T and is quasi-split if the derived
group of the centralizer Z(TSP) of TSP in G 1 is trivial (so, Z(TSP) = T; see
[Ttl). The only thing we need to prove (81-3) is the fact that the structure
of quasi-split groups is determined by its root system and a small amount
of additional data (see [Ttl); particularly, we need the fact that the group is
generated (integrally) by unipotent elements.
For a finite set of rational primes E', we write Z(17') C QI for the localization
outside E'; so, Z(17') = Z [~Ip E E']' We suppose that G 1 / 7L (E') is a (smooth)
4.4 Appendix: Fundamental Groups 219
semi-simple classical group; so, G 1/1QJ and G 1/ lFp for all p 9' E' are semi-simple.
Since the conditions (S2-3) concern only large primes, we may assume
(QS) G 1 / Zp is smooth quasi-split for all primes p outside E' of rational
primes (i.e., all fibers: G 1/ lFp and G 1/lQJp are quasi-split for all p 9' E').
We now prove the assertion (S3). Let ¢ : G 1(Zp) --+ G 1(Ze) be a homomor-
phism. Since G 1(Zp) is quasi split, G 1(Zp) is generated by unipotent elements
(see Lemma 4.46), and its unipotent radical U is generated by an additive
subgroup Ua corresponding to a simple root D'.
If G 1 = SL(n)/z, for example, we may assume that Ua is made of
diag[l J , (6 Y), 1n - J - 2 l with u E Zp for an index j (with 1 :s: j :s: n).
In general, U and Ua are p-profinite. We consider the normalizer N(Ua )
and the centralizer Z(Ua ) of Ua in G 1 . Then by conjugation, N(Ua)jZ(Ua )
acts on Ua . Since ¢ is a group homomorphism, the quotient N(Ua)jZ(Ua )
keeps acting on the image ¢(Ua ) in G 1(Ze) through conjugation by elements in
¢(N(Ua )). If p oJ e, every element of ¢(Ua ) is semi-simple (because unipotent
radical of G 1 (Ze) is e-profinite). Thus the centralizer (resp. the normalizer) of
¢(Ue) is given by Z(Ze) (resp. N(Ze)) for a reductive subgroup Z (resp. N)
of G 1 . Then N(Ze)jZ(Ze) is a finite subgroup of the Weyl group WI of G 1 /([
which is independent of e.
For example, if G 1 = SL(n)/z, and if ¢(Ua ) is made of diagonal matri-
ces diag[(11ml' (21m2' ... , (r 1m r l for generically distinct (J, Z is given by the
subgroup
SL(n) n (GL(ml) x GL(m2) x ... x GL(m r )),
where GL(md x GL(m2) x ... x GL(m r ) is embedded in GL(n) diagonally.
The quotient N j Z in this case is isomorphic to the subgroup of permutation
matrices preserving Z.
If ¢(Ua ) is nontrivial, the image of N(Ua)jZ(Ua ) in Aut(¢(Ua )) grows at
least on the order of pas p grows. In the above example of G 1 = SL(n)/z,
if ¢(Ua ) ~ Zjpmz for m > 0, all elements in Aut(¢(Ua )) ~ (Zjpmz)x come
from N(Ua)jZ(Ua ) ~ Z;. This is impossible if p » IWll. Thus ¢(Ua ) = l.
Since G 1 (Zp) for large enough p is generated by U a for all simple roots D', ¢
has to be trivial for p large enough. 0
Now we prepare several lemmas and propositions in order to prove (Sl-2)
for SL(2). Let K be a number field, and consider V = K2. We consider the
A-linear endomorphism algebra EndA(VA). Let Z(A) be the center of the K A -
linear endomorphism algebra EndKA(VA) C EndA(VA). Thus KA ~ Z(A) by
sending f E KA to the scalar multiplication by f.
Lemma 4.45 The centralizer in EndA(VA) of UA = {( 6~) Ib E K A }
n
and an
element (;j E M2(K) with 1 oJ r E K X is given by Z(A).
M 2 (KA ), we have ~ (~~) = (~i~j ;i~~)· Let N(A) = {(g 8) Ib E A} for a ring
A. We consider the centralizer
Lemma 4.46 Let K be afield. Then SL 2(K) and PSL 2(K) are generated by
unipotent elements. If IKI > 2, the commutator [GL 2(K), GL 2(K)] is equal
to SL 2(K), and [PGL 2(K), PGL 2(K)] = PSL 2(K).
where i(x)(g) = xgx~l. We write B (resp. U) for the upper triangular Borel
subgroup (resp. the upper triangular unipotent subgroup) of G L2 (K). Their
image in PGL 2(K) is denoted by Band U. Let U' be a subgroup of GLn(K)
isomorphic to the additive group K; so, we have a : K '?" U'. Consider the
image a(l) of 1 E K in U'. Replacing K by a finite extension containing an
eigenvalue 0' of a(l), let Ve, C V be the eigenspace of a(l) with eigenvalue 0'.
Then a(~) acts on Va: and a(~)m = a(l) = 0' E End(Va:). Thus we have an
algebra homomorphism K[x]j(x m -0') -+ EndK(Va:) for all 0 < m E Z. If K is
a p-adic local field, nm nm
(KX )m! = {I}. By using this, we find (Kx)m = {I}
for a global field K. Thus if K is a non-Archimedean local field or a global
field, we find that 0' has to be l. Thus U' is made up of commuting unipotent
elements; so, by conjugation, we can bring U' onto U.
Since O"(U) for 0" E Aut(PGL 2(K)) is isomorphic to K in PGL 2(K),
O"(U) is again a unipotent subgroup of PGL 2(K). Since B is the normalizer
of U, again O"(B) is the normalizer of O"(U); so, O"(B) is a Borel subgroup.
We find g E GL 2(K) such that dB) = gBg~l. Thus we may assume that
0" fixes B. Applying the same argument to U, we may assume that 0" fixes
U. Let 0" (6 i) = (6 ~). Then changing 0" by i (a:~' ?) 00", we may assume
that 0" (6 i) = (6 i ). Then 0" induces an automorphism (f of T = B jU. Fix
an isomorphism t : K X '?" T and u : K '?" U. Since 0" sends the center to
the center, we find t(a)u(b)t(a)~l = u(ab). Then applying 0" to the above
formula, we get (f(a)O"(b) = dab) for a E K X and b E K. Taking b = 1,
we find (f(a) = O"(a). Thus O"(ab) = O"(a)O"(b), and 0" : K -+ K is a field
automorphism. Modifying 0" by an element of Aut(K), we may assume that 0"
induces the identity map on B. Thus 0" induces an automorphism of pl(K) =
G L 2 (K) j B fixing CXJ and commuting with all affine transformations z f-7 az+b
of pl(K) for a E K X and b E K. Then 0" is the identity map on pl(K) and,
hence, on GL2(K). We find Out(PGL 2(K)) 2" Aut(K) and Aut(PGL 2(K)) =
Aut(K) ~ PGL2(K).
222 4 Hilbert Modular Varieties
(J((a + b)2) = a-(a 2 + 2ab + b2) = (J(a)2 + 2(J(ab) + (J(b)2 and a-((a + b)2) =
O'(a + b)2 = (J(a + b)2 = a-(a)2 + 2(J(a)(J(b) + (J(b)2. Thus if K has char-
acteristic different from 2, we find (J(ab) = (J(a)(J(b); so, (J E Aut(K).
Thus for a character E : K Xj(KX)2 -+ {±1}, we have O'(x) = E(xk(x);
so, a-(t(a)) = t(O'(a)) = ±tk(a)) = t((J(a)) in PSL 2(K). This shows
Aut(PSL 2 (K)) = Aut(K) ~ PGL 2(K) = Aut(PGL 2(K)).
We now study Aut(SL 2(K)). Since physically SL 2(K) = PSL 2 (K) for a
field of characteristic 2, we may assume that K has a characteristic different
from 2. We have an exact sequence
Here is a direct proof of (Sl) for SL 2 (K) for a finite extension K of Qp,
although the proof in [GAN] is a highly polished and generalized vers.ion
of this argument. The exponential exp(X) = Ln>o -;,~ and the logarithm
10g(1 + X) = Ln>l (-1 )n+l ~n converge p-adically on a sufficiently small
neighborhood of 0 of M 2 (K) and 1 of GL 2 (K), respectively. If (J : S ~ S' is
an isomorphism of open compact subgroups of SL 2 (K), by shrinking Sand
S', we may assume that the log induces an isomorphism log: S ~ 10g(S)
and S' ~ 10g(S'). Then (J induces an isomorphism of the Lie algebra (J :
10g(S) ~ 10g(S'), which extends to an automorphism of the Lie algebra.s[2(K)
of SL 2 (K) by scalar extension. Since (J takes the maximal abelian Lie algebra
N(K) ~ K of .s[2(K) into its conjugate, changing (J by (J composed with a
conjugation by an element of GL 2(K), we may assume that (J takes N(K)
onto N(K). Identifying N(K) with K by sending each matrix in N(K) to its
4.4 Appendix: Fundamental Groups 223
Our argument proving Propositions 4.47 and 4.48 can be easily generalized
to a semi-simple algebra K of finite dimension over Ql or Qlp. We only state
the result leaving the proof to attentive readers.
Proposition 4.49 Assume that K is a semi-simple commutative algebra of
finite dimension over Ql or Qlp (or the direct sum of such). Then the auto-
morphism groups of the groups PGL 2(K), PSL 2(K), and SL 2(K) are given
by Aut(K) I>< PGL2(K). If K is a semi-simple Qlp-algebra, we again have
Aut(PGL 2(0)) = Aut(PSL 2(0)) = Aut(SL 2(0)) = Aut(O) I>< PGL 2(0),
Let G/z(p) for Z(p) = Qnzp be a connected reductive group split over Zp (see
Section 4.4.3 for the definition of split algebraic groups). We first prove semi-
simplicity of the commutative Hecke algebra acting on the nearly ordinary co-
homology group H~.ord(X(S), L) inside the interior cohomology H!q(X(S), L)
for a modular variety X(S) associated with an arbitrary p-power level open
compact subgroup S of G(A 00), where H? is the image of the compactly sup-
ported cohomology group inside the standard cohomology group. Here the
locally constant or coherent sheaf L of Q-vector spaces on X(S) is associ-
ated with a rational representation of G twisted by a finite-order character.
After dealing with topological cohomology groups, we relate as Hecke mod-
ules the topological and the coherent cohomology groups via the generalized
Eichler-Shimura map (which shows the semi-simplicity of the Hecke algebra
acting on the coherent cohomology and topological cohomology). Although
we have assumed that G is split over Zp, the argument works equally well for
a connected reductive G smooth quasi-split over Zp-
We write B (resp. N) for a fixed Borel subgroup of G (resp. the unipotent
radical N of B). We have the torus To = BIN. In this chapter, the character
U is often used to denote a unitary group.
on which we let G(Qp) act by f(x)g = f(gx). Here the word "smooth" meanS
that for each f E Ind~ V', we find an open compact subgroup S such that
f(kx) = f(x) for all k E S. Thus Ind~ V'is automatically smooth.
Since the smooth induction preserves admissibility ([BeZ] 2.3), V =
Ind~ V' has composition series {O} C VI C V2 C ... c V stable under
the action of G, and hence its semi-simplification (Ind~ V')SS = EBJ Vy+1/Vy
is well-defined as an admissible G(Qp)-module. The beauty of the theory of
admissible representations is its purely algebraic nature; so, we do not need to
assume any analytic assumptions; in particular, our representations are often
5.1 Semi-Simplicity of Heeke Algebras 227
not unitary as is clear from our main result (Theorem 5.3) in this section.
Hereafter all representations of G, B, and Tc are supposed to be admissible.
The following results are due to Jacquet and Bernstein-Zelevinsky and are
well known (see [BeZ]),
1. (Frobenius reciprocity; [BeZ] 1.9) HomB(VB, V') ~ Homc(V, Ind~ V');
2. If 1[" is absolutely irreducible, then dimK VB :s; IWI, where W is the Weyl
group of Tc in G (Bernstein-Zelevinsky; [BeZ] 2.9);
3. If 1[" is absolutely irreducible and VB =f. 0, then we have a surjective linear
map Ind~ A -t> V of G(Qp)-modules for a character A : Tc(Qp) --+ K X
(Jacquet; [BeZ] 2.4-5);
4. (Ind~ >:) ss ~ (Ind~ AW) ss for w E W (Bernstein-Zelevinsky; [BeZ] 2.9),
°
We say a representation 1[" is regularly induced (at p) if for each irreducible
component V of 1[";..!!e have a nontrivial >:-eigenspace VB [>:] =f. for a character
A of Tc(Qp) and Aw(t) (w E W) are all distinct.
X~X= u X~u= u
uEN\f.Nf.-l
Xu~. (5.3)
By this fact, the double coset algebras generated additively over Z by double
cosets of X in .1x have the following homomorphic relations as algebras:
via N~N H B~B H S~S for ~ E .1N' These algebras are commutative:
T(~)T(TJ) = for T(~) = N~N and ~,TJ E D. See [IAT] Chapter 3 and
T(~TJ)
[MFG] 3.1.6 for this type of results dealing with GL(2). Here we give a proof
of these facts for G = GL(n). The same argument works well for classical
Chevalley groups quasi-split over Zp replacing aj below by the image of p
under simple coroot a positive with respect to D (i.e., a(p) ED).
Lemma 5.2 Let the notation and the assumption be as above. Assume that
G = G L( n) /zp' The algebras R(S, .1s) for Iwahori subgroups S are com-
mutative, and if S :::) B(Zp), they are all isomorphic to the polynomial ring
Z[h, ... , tn, t;;-l] (with tn inverted) fort J = SajS with a J = diag[ln-J,p1J]' If
S:::) N(Zp), we have R(S, .1s) ~ Z[TS][tl,"" tn, t;;-l] for the quotient group
Ts = Tc(Zp)SjS ofTc(Zp), where Z[Ts] is the group algebra ofTs .
Proof. For simplicity, we write B for B(Zp). For ~ E D, we consider the double
coset B~B. Decompose B = U1)E.S'(f.)(~-l B~ n B)TJ. Multiplying by ~-l B~
from the left, we get ~-l B~B = U1)E.S'(f.) ~-l B~TJ {:} B~B = U1)E.S'(f.) B~TJ. If
~ = diag[pa 1 , ••. ,pan], we have
5.1 Semi-Simplicity of Hecke Algebras 229
and hence we get the identity deg(B~B(B) = deg(B~(B). Since the ajs give
independent generators of D jTc(Zp), the monoid algebra Z[D jTc(Zp)] is
isomorphic to a polynomial ring with n variables Z[a1' ... ,an, a;;-l] with an
inverted (because an E Z(Qip) is invertible in Z(Qip) C D). The association
a J r-+ T( a J ) = Baj B therefore induces a surjective algebra homomorphism
Z[DjTc(Zp)] ---+ R(B,41 B ), which can be easily seen to be an isomorphism.
Replacing DjTc(Zp) by Dj(Tc(Zp) n S) = Ts X (DjTc(Zp)) in the above
argument, the same proof works well for any S with So(r) => S => Sl(r) and
yields R(S, 41s) ~ Z[Ts][DjTc(Zp)] ~ Z[TS][t1, ... , tn, t;;-l], where Z[Ts] is
embedded into R(S, 41s) by sending t E Ts to StS. 0
vIT(O = vl[N~N] = L
uE.;-lN.;\N
V7f(~u) = 1
';N.;-l
V7f(u)7f(Odu, (5.4)
We now fix a finite extension Kover Qp. Let W be the p-adic integer ring
of K. We write S = So(r) (r > 0). Recall the Iwahori decomposition S =
N'Tc(7l,p)N. We consider the space C(R) = {¢: SjN(7l,p) ---+ R} made up of
continuous functions for R = Wand K. Here the continuity is with respect
to the p-adic topology of the source and the target. We would like to make
C a left L1;;l-module for the opposite semi-group L1;;1 of L1p = L1s. This
L1;;l-module C can be considered as the p-adic continuous induction of the
right regular representation of T c (7l,p) on the space of p-adically continuous
functions on Tc(7l,p) with values in R (thus, this is the third induction process
we encounter in this section). For that, we first define a left action of L1p on
Ys = SjN(7l,p). Since S acts on Ys = SjN(7l,p) from the left, we only need to
5.1 Semi-Simplicity of Heeke Algebras 231
where Oe/N is the structure sheaf of the scheme G/N, and we regard its
global section ¢ as a function on G(rfJp)/N(rfJp) with values in K. We let G
act on L(K; K) = Lc(K; K) by g¢(y) = ¢(g~ly). Then L(K; K) = ind~ K~l
(following the normalization of induction as in [RAG] 1.3.3), which is the
induction in the category of scheme theoretic representations (i.e., rational
representations). We call K dominant if L(K; K) I- O. We write a matrix form
of this representation as PI< = pC: : G r--+ GL(L(K; K)).
All dominant weights in X(Te) ~ zr form a cone
does not show up for coherent sheaves for G = SL(2), because Lc(K; K)
is one-dimensional due to the fact that C = T for G = GL(2). A similar
remark applies to Lc(K; K) when G = G 1 = ResF/Q1SL(2). However, in the
Hilbert modular case of G = ResF/Q1GL(2), the induction of K* = (K1' K2) E
X(Tc) for the diagonal torus Tc of G is isomorphic to the tensor product
det"'! 0L(K2 - K1 - I; K) for the restriction K2 - K1 - I of K to T = Tc n G 1
(G 1 = ResF/Q1SL(2)). The p-integral action created through this process is
the action on LC(K2 - K1 - I; K) removing the "det" factor.
(5.9)
(5.10)
This follows from the fact that the normalized action of ~-1 E Ll;l is given by
1i(~)-lc(~)p",(~)-l, while as an element of G(A(oo»), ~-1 acts via c(~)p",(~)-l.
Suppose that G admits Shimura variety Shs of PEL type defined over
its reflex field E (see Chapter 7 for generality of Shimura varieties), whose
complex points give X(S). When Ii = 0 (the identity character), the action
of the Hecke operator is motivic and is (canonically) induced by the Hecke
correspondence T(~) c Shs x Shs, and in this case, Tp(~) = T(~) is the
action of the correspondence on the Betti cohomology of X(S). If Ii > 0,
we may relate cohomology groups of the sheaf L(lic; K) as a part of the
cohomology group with constant coefficients of a certain self-product Z of
copies of the universal abelian scheme over X(S). The Hecke operator then
has an interpretation as an isogeny action on the universal abelian scheme,
and it can be regarded as the action induced by the Hecke correspondence
in Z x Z. The action of Tp(~) and T(~) uses different action of Ll;l. This
action of Ll;l determines the part of the cohomology group over Z identified
with the cohomology group over X(S) with locally constant (but nonconstant)
coefficients. Thus the motivic realization of the two operators T(~) and Tp(~)
could be actually different, and the operator Tp(~) may not even have any
motivic realization (as in the Hilbert modular case of nonparallel weight). For
example, in Scholl's construction [Sc] of the Grothendieck motive associated
with an elliptic Hecke eigenform J, if one changes the action of congruence
subgroup r of SL 2 (7l,) by a power of determinant character, the physical sheaf
over Xr(C) obtained is the same, but its rational structure (including the
234 5 Generalized Eichler-Shimura Map
Galois action) is different. In the elliptic modular case (treated in [Sc]), we can
construct the motive associated with the standard p-adic Galois representation
PI of f and its Tate twists PI(m) as the etale realization of motives directly
realized over a self-product of the universal elliptic curve.
For a Hilbert-Hecke eigenform f, we could twist PI locally at each p-adic
place by a power of the p-adic cyclotomic character, but this twist may not
extend to a global twist because the exponent of the cyclotomic character
depends on the p-adic place. In particular, if f is of nonparallel weight, the
process of defining Tp(~) corresponds to untwisting PI to reach a p-ordinary
Galois representation at each p-adic place p, which cannot be performed glob-
ally; so, the operator Tp(~) may not have a motivic interpretation.
Unfortunately, for a Hilbert-Heeke eigenform f for F =I- Q, the motive of
f realized from the Hilbert modular variety does not produce a rank 2 motive
whose etale realization gives rise to the two-dimensional Galois representation
of f (see [BlRJ, [TaJ, and also [MFG] 5.6.1 for a description of such Galois
representations). Therefore to obtain such a rank 2 motive, we need to re-
sort to some tricky construction either moving to Shimura curves treated in
Section 4.3 (see [H81]) or extracting a rank 2 motive from a rank 4 motive
(which can be constructed from some Shimura surfaces; see [Bl]). Despite such
an effort, this problem is not completely solved yet.
We take ~ E D such that N(Qp) = UJ ~J N(Zp)~-J. The product e =
rL,EE ~n (or its power ILEE ~~a with en > 0) has this property. As seen at
the beginning of Section 3.3, the p-adic (or profinite) limit e = limn-+oo Tp(e)n!
exists as an operator on Hq(X(S), L(K,E; R)) for R = Wand K (and is in-
dependent of the choice of e), because Hq(X(S), L(K,E; R)) is an R-module
of finite type. Thus the limit e exists as an endomorphism of .cq(K,E; R)N for
R = Wand K. It is easy to see, if Sp :J N,
(5.11)
The equality holds if and only if 7r is nearly p-ordinary (in this case, automat-
ically VB =I- 0 and Ind~ >: --» 7rp because V N ~ VB EB V(B)N as R-modules).
.5.1 Semi-Simplicity of Hecke Algebras 235
For the moment, suppose that G(Qp) = GLn(Qp) and factor the character
A as A(diag[tl, ... , t n ]) = rr=l A,(t,). Define the Hecke polynomial (at p) by
H,,(X) = TI~=l (1- A,(p)X), and write L1N for the Newton polygon of H11:(X)
(e.g., [NAZ]). Define the Hodge polygon L1H of 7r to be the Newton polygon
of TI~l (1 - (r;,cp),(p)X). Then the above inequality implies L1N :::: L1H and
the two extreme ends of the two polygons match.
We return to a general group G and assume that 7r is nearly p-ordinary.
By definition, fN(Qp) ¢(u)du = oB(b) fN(Qp) ¢(b-1xb)du. This shows that
(5.12)
(5.13)
Since the dominant r;, is nonnegative with respect to E, r;,p > 0; that is, /'i,p is
in the interior of the Weyl chamber of E. This shows that if wi-I,
(5.14)
First suppose that Rad(V) = 0, which is equivalent to det <P =I- O. Then we
define the unitary group of rp by
Hereafter we assume that rp is nondegenerate (i.e., det <P =I- 0). There is
another realization of a given Hermitian form if it is isotropic. A Hermitian
form is called anisotropic if rp( v, v) = 0 implies v = O. If rp is not anisotropic,
238 5 Generalized Eichler-Shimura Map
Since Tr( <p( w, vd) =I 0, we can solve this equation and define v~ = w + XVI
for the solution X. If dim J = 1, we set J' = CCv~, which proves the claim.
Suppose now that dim J > 1. We take a subspace J 1 of J so that dim J 1 =
dim J - 1. Since J is totally isotropic, J 1 is totally isotropic. By the induction
assumption, we find J{ so that <p induces the duality between J{ and J 1 .
Then any V E J{ n J 1 is orthogonal to every w E J 1 ; so, J{ n J 1 = O. We put
X = Jf + J 1 , on which <p is nondegenerate. Thus V = X EEl X.l. Let J o be the
projection of J to X.l, which is totally isotropic and of dimension 1. We find
inside X.l a subspace J~ dual to J o under <p, and J' = J{ EEl J~ does the job.
Theorem 5.6 We have the following assertions.
(1) If J is a totally isotropic subspace of V, we can find a totally isotropic
subspace J' of V such that <p induces a CC-antilinear isomorphism i : J' ~
Homc(J, CC). Here "CC-antilinearity" means that i(Av) = >:i(v) for A E cc.
(2) If J is a totally isotropic subspace of V, we can find a base v = {VI, ... , Vd}
of V such that
(a) v, for i <S j span J;
(b) <p(v" Vn-J+k) = i5,k for 0 <S k <S j;
(c) <p(v"vk)=Oifj<k<sd-j andi<Sj ori>d-j.
(3) If J is a totally isotropic subspace of V, we can find a base v = {V1' ... , Vd}
of V such that the V,s for i <S j span J, and 1>v = (1~ t 18 ) .
(4) If 1> is isotropic, then we find an integer j > 0 and X E G L d(CC) such that
X1>X* = ( ~ ~ 10)
1) 0 0
for an anisotropic !Jr.
Proposition 5.7 Let V(m,n) = {z E ce~lz*z < In} and X(z) = (~':' 1:)
faT' z E ce~. Then V(m, n) x GLm(C) x GLn(C) ~ 1: by (z, g, h) ~
X(z)(g~).
X *Im,n X = (a*a-c*ca*b-c*d)
bOa_doc bOb-dod
°)
= (AO-B
for positive definite A and B. Thus a*a > c*c ::" 0, because x*c*cx =
(cx)*(cx) ::" O. Thus if au = 0 for u E cem, then u*(a*a)u = 0 and hence
u = O. This shows the linear map cem -+ cem given by u ~ au is injec-
tive and hence invertible. That is, det a i- O. Similarly, b* b < d* d implies that
det d i- O. Then we define z = bd- 1 . The fact d* (d-*b* -ca- 1 )a = b*a-d*c = 0
tells us z* = (bd- 1 )* = ca- 1 . Thus X = (~':' tn) (o~) = X(z) (o~). Since
0> b*b - d*d = d-*(1- z*z)d, we have z*z < In ¢} 1m > zz* because the
signature of Im,n is (m, n). This shows that z E V(m, n), and the map is sur-
jective. Suppose (~':' tn ) = X(z) = X(z') (0 ~) = C;c z~d). Then c = d = 1
and hence z = z', which shows the injectivity. 0
aX(z) = X(a(z)) °
(h(OO,Z) ° )
J(OO,z)
X((aj3)(z))
°
(h(OO{3,z) ° ) = aj3X(z)
J(oo{3,z)
= aX(j3(z)) (h({3,Z)
° ° )
J({3,z)
= X(a(j3(z))) (~h(-OO,-;;{3'-(Z~))h;-;(-;;-{3,--"'Z)
° °
J(oo,{3(z))J({3,z)
)
for a, 13 E Um,n, we conclude
Let 0 be the zero matrix in D(m, n). Then a(O) =0 implies that
aIm,n -- I m,n °
(h(a,O) ° ),
J(a,O)
(5.21)
cp= ( ° °-'In)
° =(O<P)
°° °
06
d n 1[1
Write Y E ~ as
Y = ( adebe)
9 h J
f =
(VI fC)
V2
W J
(5.23)
for z = (~) E 3.
Proposition 5.9 Let <p be as above. Then ~ ~ 3x GLm(C) x GLn(C) by
B(z)(o~) H (z,a,b).
Proof. The argument before the proposition shows the surjectivity of the map
3 x GLm(C) x GLn(C) --7 ~. Suppose B(z) = B(z') (8 ~). Writing = e
( a ~,e) we have
In 0 '
242 5 Generalized Eichler-Shiroura Map
Corollary 5.10 Choosing T E GLd(C) with TcpT* = Im,n (by Theorem 5.5),
the map T : 3 -t V(m, n) given by TB(z) = X(T(z)) (h(~,Z) J(~'Z») with
h(T, z) E GLm(C) and j(T, z) E GLn(C) is a holomorphic isomorphism.
Writing i = ('k) E 3, we have U'I'(JR)jCj ~ 3 by g H g(i), where C j = {g E
U'I'(JR)lg(i) = i}. Furthermore, if we choose T so that T(i) = 0 (possible by
transitivity), a H TaT- I induces an isomorphism Cj ~ Um(JR) x Un(JR).
Proof. By Propositions 5.7 and 5.9, z H T(z) is a well-defined isomorphism.
By definition, a H TaT- I is a surjective isomorphism of Lie groups from
U'I'(JR) onto Um,n' In particular, we have T(a(z)) = TaT-IT(z) by defini-
tion. If we can prove the transitivity of the action of U'I'(JR) on 3, we get
another proof of the transitivity of the action of Um,n on V(m, n). To show
the transitivity under U'I'(JR), we define ~ : 3 -t S+,(C) and TJ : 3 -t S+.(C) by
By definition, we have
We want to find a with a(i) = z for a given z. We assume that the last n rows
of a are (0,0, d) with d E GLn(C). Then aB(w) = B(a(w)) (0 ~) because the
lower right corner of B(w) is In. Thus f,L(a, w) = d. Since B(i)*cp-I B(i) =
°
(6 -'?2)' we have TJ(i) = 2 and d*TJ(z)d = 2. Since TJ(z) > by the definition of
3, we can find a E GLr(C) such that i(x* -x) _y*()-Iy = TJ(z) = 2aa*. Then
for z = t(tx, ty) E 3,
We define
Writing r.p = Tv and taking 1= 'Lr;'=-+::+l CvJ , we see easily that p(z) C Pj (lR).
Thus the above proof of Corollary 5.10 shows
Corollary 5.11 We have U'P(JR) = Pj(lR)C for any maximal totally isotropic
subspace I c V and any maximal compact subgroup C of U'P(lR).
The fact that U'P (lR) = Pj (lR)C holds for any totally isotropic subspace J.
Proof. The assertion follows from Corollary 5.10 for C = C j . Since any other C
is a conjugate of C j , the assertion is clear. This fact follows from the topological
fact that any compact subgroup has a fixed point on U'P(lR)jC j = 3. 0
We now compute the measure on 3 invariant under the action of U'P(lR). For
that, we first compute several formulas:
(5.31)
B(z) (
00)
,e-1 l
10 = (-''7(Z)
0
y'
-,e y . x)
-1 01 0 0 1
that
t'l9(a, zd(a(z) - a(zl))/L(a, z) = z - Zl, (5.35)
/L(a, zd*ry(a(zd, a(z))/L(a, z) = ry(Zl' z), (5.36)
'I9(a, zl)*~(a(zd, a(z))'I9(a, z) = ~(Zl' z). (5.37)
To compute the Jacobian matrix of z t-+ a(z), in (5.35), replace Zl by
z+dz. We get a*dz = t'l9(a,z)-ldz/L(a,z)-l. Writing
r+t r
dz = /\ /\ dZhk /\ azhk,
h=lk=l
we have
(5.38)
246 5 Generalized Eichler-Shimura Map
Such a map p(O) with up(O) 0 p~,(J(U))-l = p(O) exists because the group
GL(m) x GL(n) is identified with a subgroup of GLd(C) = U(d)(C) (d =
m + n); thus, it corresponds to the inclusion
ff"") = PI<'
L C (K, * ; IL- C
'---+
. dOL(m+n)
In P
C
PI<' ff"") = PI<'
= L 0 (K, * ; IL- 0
of Um,n, and we have Pw", (z)(x) = (\; ) x for x E em. Again we verify that
gpw", (z)(x) = Pw", (z)(h(g, z)x), and Pw", (z) is a polynomial in z. For general
K" Lc(K,*;'C) (resp. Lo(K,*; C)) is a quotient of Lc(w n ; C)0 t ® Lc(w m ; C)0 8
(resp. Lo(wn ; C)0 t ® Lo(wm ; C)0 8 ). The general PI<-/1-(z) is a constant multi-
ple of the projected image of the tensor product of copies of Pw, (z) and hence
is a polynomial in z.
For a congruence subgroup r c G(lR) , each global section of ~I< is a
holomorphic function J : V --+ Lc(K,; C) satisfying J(-y(z)) = JI<(-Y, z)J(z)
for JI<(-Y, z) = PI< (J(-y, z)). We define for J E HO(Xr,~7d a holomorphic
differential with values in Lo(K,*; C) by w(f) = PI<' (z)(f)dz. Note that here
Lc(K,;C) = Lc(K,*;C)®Lc(/Lm,n;C) and Lc(/Lm,n;C) is one-dimensional; so,
we can identify Lc(K,; C) with Lc(K,*; C) canonically as vector spaces and,
thus, the above definition is consistent. We verify that a*w(f) = p~, (a)w(f).
Theorem 5.13 Let r be a discrete subgroup oj U(m, n)(lR) with the compact
quotient space Xr = r\V(m, n). Suppose that K, 2: /Lm,n as above, and define
K,*= K, - /Lm,n' Then the association J f--t [w(f)] E Hq(X r , L(K,*; C)) Jor
q = dime V induces an embedding HO(Xr,~I<) '---+ Hq(X r , Lo(K,*; C)), where
[w(f)] is the de Rham cohomology class oj w(f).
When r is a noncocompact congruence subgroup of a global unitary group
G IrQ, the same assertion is valid replacing ~I< by its cuspidal subsheaf ~I<'
5.3 The Eichler-Shimura Map 247
Regard f.-L as a character of the maximal split torus TGSp (2g) of GSp(2g) by
the isomorphism TGL(g) X G m ~ TGSp (2g) given by (t, t') I-t diag[t, tiel].
For a given character K, > 0 of the maximal split torus TGL(g), the auto-
morphic factor SP2g(lR) x fJ g => (g, z) I-t p",(J(g, z)) with J(,,(, z) = cz + d for
9 = (~~) gives rise to a vector bundle i!,2.'" on fJg. Now suppose that K, is a
character of TGSp (2g) , and use the same symbol to denote its restriction to
T Sp (2g) = TGL(g). Suppose also that K, ;::: f.-L. Let K,* = K, - f.-L and consider the
GSp(2g)-module L(K,*; C) on which t E TGSp (2g) acts by ¢(g) I-t ¢(gt). We
want to have a polynomial function p: fJ g -t HomdLdK,*;C),L(K,*;C)) in
coordinates of z E fJ g such that p(g(z))op"'-J1-(J(g, z)) = gp(z) (g E SP2g(lR)).
We show the existence and uniqueness (up to a scalar multiple) of the map.
If we change 9 by gu for u E C for the stabilizer C of i = i1 g (i = A) of
fJ g in SP2g(lR), then the expected formula tells us
248 5 Generalized Eichler-Shimura Map
group of Um,n(lR). In the former case, we call G unitary (or of type A) and
in the latter case, we call G symplectic (or of type C). Again by the Kunneth
formula, to describe the equivariance, we may assume 1171 = 1; so, we drop
the subscript 0". We write /-L for /-Lm,n given in (5.38) when G is unitary and
for /-L in (5.39) when G is symplectic.
We have normalized the Hecke operator on the topological Betti co-
homology group taking the action of ~ E .dB normalized as jf,f(~-1) =
fi:-1(Op~(C1). Note that PK = p---:®x = PXK for an algebraic character
X : G or C -+ G m .
We normalize again in the same way the action on Id.K taking the action of
P;?(C 1) = fi:-1(~)p~(~-1) in addition to the division by /-L(O. Let 1I' = Tp(O
and also write coset representatives as~; that is, T(O = Ut; N(7Lp)~. Recalling
fi:* = fi:/-L-1, we have
w(f)I1I'B = 2)fi:*(O)-lp~*(C1)p(~(z))(f(~(z)))d(~(z))
t;
= 2: p(z) ((fi:/-L-1 (~)) -1 PK-I" (J(~, z)) -1) f(~(z) )/-L(~) -ldz
t;
Here we have added the subscripts Band C in order to emphasize the space on
which the operator acts; that is, B implies the topological Betti cohomology
Hq(X(U),Lc(fi:*;C)) and C indicates the coherent cohomology HO(Shu,Id. K )
(which is a part of the de Rham cohomology). In short, the extra modification
of the action of the Hecke operator T(O by the character /-L on the coherent
cohomology is absorbed by d(~(z)) = /-L(O-ldz in the topological cohomology.
Hence the normalization of Hecke operators at p is identical on the left-hand
side and the right-hand side of the Eichler-Shimura map. This is why we do
not have such a factor in (5.8). We also remark that in the case of GL(2), the
determinant factor I det(g)IA in the definition (4.37) of the Hecke operator on
the coherent cohomology is the value /-L(g) -1 as above. Thus we do not have
this factor in (4.89).
6
Moduli Schemes
We describe the theory of the Hilbert scheme that classifies all closed S-fiat
subschemes of a given projective variety XIs' This is a generalization of the
earlier theory of Chow coordinates that classifies cycles on a projective variety.
The theory is due to Grothendieck and the main source of our exposition is
his Expose 221 in [FGA]. One can find an exposition in [NMD] 8.2 on this
subject under milder assumptions.
where £ is the Os-dual of £. The last equality in (6.1) follows from the defini-
tion of f: Homos' (£s" Os') = r(S', £s,). Thus V(£) gives the vector bundle
associated with the dual sheaf f.
We consider a contravariant functor S' t-+ Homos' (£s' , F s ') for S' -4 S,
where £ and F are two locally free sheaves of finite rank. Note that
Then (6.1) tells us that the above functor is represented by V(£ 00 s :F):
(6.3)
Then GL(£)/s is a group scheme (cf. [GME] 1.6) representing the functor
S' t-+ Autos' (£s,)· In particular, we write GL(n)/s = GL(O'S) and we have
6.1.2 Grassmannians
with exact rows. For each morphism f : S' ----t S, the pullback j*7f : OSI ---»
j* F gives the contravariant functoriality. The stabilizer of 7f : ----t as as
can
be identified with the maximal parabolic subgroup
functorially. We soon verify this fact in down-to-earth terms after stating the
representability as a theorem (Theorem 6.1). Of course, if m = n - lor 1, we
have Grasson,m = pn-l.
We can generalize this construction slightly. Let S be a scheme, and take
a locally free sheaf [IS of rank n. Then, for each S-scheme S' ~ S, we define
a contravariant functor from S-SCH to SETS by
Grass O N+l I = pN; SO, by enlarging the dimension of the projective space,
S '
we confirm that Grass£,1 is projective. We reduce in the following proof of
projectivity of Grass£,m to this case of m = 1 by taking Plucker's coordinates
(described in the proof).
The scheme Grass£,m gives a geometric quotient of GL(n) by P when
£ = Os in the sense of Mumford (see [GIT] 0.6 and [GME] 1.8.3).
Proof We repeat a proof given in [EGA] 1.9.7. By the argument preceding the
theorem, we may assume that £ = Os. We choose a base I = {VI, V2, ... , v n }
of Os over Os and consider a subset J of order m of I. We then define a
subset GJ(S') c Grass£,m(S') by
1 1<>
X ----7 Y.
(3
Then we see
element in Endosl (OS;); so, we have det( w 0 L5 ,J') E G a (S'). The association
1
Here the subscript B indicates a split Borel subgroup of GL(n), since Flag is
represented by a projective scheme FlagB = GL(n)/ B if [ = Os as we show
below (Theorem 6.2). Since [ ~ Os locally, covering S by open subschemes
6.1 Hilbert Schemes 2.57
o over which [ is free, we can glue schemes over the open Os representing
Flag B/ o to get a scheme representing Flag over S. The association ('ll'] : [s' -+
[]) r-+ ('ll'])] induces a functorial map,
Proof. By the argument preceding the theorem, we only need to prove the
represent ability assuming [ = Os. Write
Here we understand that Ker( 'll'o) = [, and j runs over all integers between 0
and n - l. If [ ~ Os and S is affine, writing 1 = ('ll'], cP]) for the standard flag
'll'] : Os -+ O~-], projecting column vectors down to lower n - j coordinates,
the upper unipotent subgroup U of G L( n) = Autos (Os) is the stabilizer of
1. Therefore Flag u is represented by Flagu/s = GL(n)jU. In general, cover S
by open affine schemes S, = Spec(8,) so that we have an isomorphism [Is, ~
Os,. On S'] = S, n S]' by the universality, Flagu / s, x S S'] is canonically
isomorphic to Flagu/s J Xs Sq. Thus these schemes glue each other, giving
rise to a scheme Flagu/s representing Flag u ' Obviously Flagu/s is a T-torsor
for the maximal split torus T c G L( n). Here the action of Ton Flagu is given
n
258 6 Moduli Schemes
Let W : Flagu -t FlagB be the projection (Wj, rPJ) ~ (Wj). Then for a
character K E X(T), we define a sheaf £1<.(0) = HO(w-1(0),OFlagu[K]) for
each open subset 0 c FlagB' Then £1<. is a locally free sheaf on FlagB' Since
f : FlagB -t S is proper fiat over S, we find that f*£1<. (which we again write
£1<.) is a locally free sheaf on S. In this way, we can associate a K-power £1<.
of the original locally free sheaf £ for each weight K E X (T) with respect to
(B, T), which is non-zero if and only if K is a dominant weight of GL(n) with
respect to the pair (B, T).
Quot)',! X/S(SI)
= [w: rP'XF -+> MjM is a coherent OXsl-module fiat over OSI ] .
The isomorphism between the ws are defined in the same manner as in the
case of the functor Grass. Here we do not assume that :F is Os-fiat.
6.1 Hilbert Schemes 259
For simplicity, we always assume that schemes S' (and also S) are Noethe-
rian. Each M E QuotF/X/S(S') has its Hilbert polynomial PM, and obviously
for g : SI! -+ S', g* M has the same Hilbert polynomial. Thus we can split the
functor as
Quot F / X / S = UQuot;/x/s'
P
(i) FJ+k/AkK' for K' = Ker(7["') is locally Os,-free of rank P(j + k) for all
k ::::: o. Here we consider AkK' in A . F;
(ii) Define a graded module K'* by A· K'. We require the associated sheaf
~*
K' = K'/x , on Xs' to satisfy (b) and the quotient M' = F/K' to satisfy
(a) (these (a) and (b) are open conditions: [ABV] Section 5 Corollary or
[ALG] Theorem III.12.8).
For a graded A-module M = EBk M k , putting M(t) = EBk;:::t M k , we have
M(t) ~ M as already remarked. Then the condition (i) ensures that fS',*M'
is locally Os,-free, and the image of Q (S') is characterized by (i) and (ii).
-J
Hilb~/s(S')
= {closed subschemes of X S ' flat over S' with Hilbert polynomial P} .
This scheme is called the Hilbert scheme of X for the polynomial P. There is
a direct way (as indicated in [GIT] 0.5.iii) of constructing Hilb~/S for general
X c pn assuming its existence for X = Pis as proved above. We give details
of this. Let X C Pis be a projective scheme. We write H = Hilb~nlS and
let 1T : Z --+ H be the universal projective scheme over H with i : Z '-+ PiH'
In other words, for any closed subscheme Z C Pis, for S' E SCH IS with
262 6 Moduli Schemes
11 11
Z ----+
c
We define X H = X Xpn/S p/nH and Zx Z Xpn/H X H = X Xpn/S Z as a
subscheme of p/w We define Hilh~/s to be the image of Zx in H, that
is, the topological image with reduced scheme structure. Since Zx -t H is
projective, the image Hilb~/ S is a well-defined closed subscheme of Hilb~n / s·
Then, if Z/SI is a closed sub scheme of X/S', </>* Z = Z c X, and hence
</>*(Zx) = Z with </> : S' -t H factoring through Hilb x / s . Thus we have
Let Y/ s -4
XIS be a morphism of S-schemes. Let Secy/x/s SCH/s-t
SETS be contravariant functors given by
1 1
H=H
such that for any S-scheme S' and a closed subscheme V y YS' flat over S',
we have a unique morphism <pv : S' -+ Hover S such that the pullback of
the above square by <Pv is identical to
1
S'
1
Sf.
Corollary 6.7 Let the notation and the assumption be as in the theorem.
The functor HornS,s' (X, Y)/s is representable by an S-scheme HS's' (X, Y)/s
whose connected components are quasi-projective over S.
264 6 Moduli Schemes
Proof. Consider the functorial map Homs'(Xs' , Y s ') :3 ¢ I---t ¢(ss') E Y(S')
which induces a morphism (J : Hom(X, Y)/s -+ Y by Yoneda's lemma
(see [GME] Lemma 1.4.1). Then the functor Homs,s' (X, Y)/s is again rep-
resentable by a scheme HS,s' (X, Y)/s = H(X, Y)/s xY,u,s' S over S. 0
Corollary 6.8 Let the notation and the assumption be as in the theorem. For
a section s : S '---+ X, consider the following functor,
The functor SCH/ s :3 S' I---t E"s,(X s ') E SETS is representable by a scheme
x/
E s over S whose connected components are quasi-projective over S.
locally free group scheme of finite rank (see [GME] Theorem 4.1.17 (2)); in
this case, we call ¢ an isogeny. Thus ¢ is finite and hence affine (cf. [ALG]
Exercise II.5.17). This implies ker(¢) = Spec(R) for a locally free Os-algebra
R. We define the degree deg(¢) of ¢ by the rankos R. If dims Ker(¢) > 0, we
simply put deg( ¢) = o.
If the connected component E c E~I s contains an isogeny, the degree
is independent of the point of E. To see this, we consider AE = A Xs E.
Then ¢ induces ¢E E End(AEIE)' we have deg(¢) = deg(¢E), and deg(¢E)
is constant on the connected component of E; so, it is constant on E. For an
abelian variety over a field, the number of isogenies with a given positive degree
is finite (positivity of the Rosati involution: see [ABV] Section 21). Thus E is
projective and quasi-finite; so, E is finite over S ([GME] Proposition 1.9.11).
We can generalize this argument to a strongly quasi-projective semi-
abelian scheme. A scheme X -4 S is strongly projective if X Isis finitely
presented and there exists a locally free sheaf Els of a constant finite rank
and we have a closed immersion X '-+ P(E) over S. If XIS has an open
immersion into a strongly projective scheme XIs' we call XIS strongly quasi-
projective. An abelian scheme A ls equipped with a polarization A: A -+ t A is
strongly projective (over any base S), because for a locally ample line bundle
L, Ll = L0 2 is globally ample as remarked in Section 4.1.1 and L~3 is very
ample (e.g., [ABV] Section 17).
As before, we suppose that S is Noetherian and any scheme XIS we con-
sider is of finite type over S. Then automatically XIS is finitely presented. If
XIS is projective, then by definition (cf. [EGA] II.5.5), we have a coherent
sheaf Els and a closed immersion of X into P(E). Covering E by O~+l, we
have a closed immersion P(E) '-+ Pis. Thus XIS is strongly projective under
our finiteness condition. Also if XIS is quasi-projective, it is strongly quasi-
projective under our finiteness assumption. Thus the results stated in [NMD]
Chapter 8 under some strong (quasi) projectivity assumptions hold under the
corresponding projectivity assumptions in our setting.
A semi-abelian scheme 91s is a smooth separated group scheme with ge-
ometrically connected fiber such that each geometric fiber is an extension of
an abelian variety by a torus. The toric rank may depend on the fiber. If the
abelian scheme has polarization, 91s is quasi-projective. We assume that 91s
is a quasi-projective semi-abelian scheme which is an abelian scheme over a
dense open subscheme 0 of S. The semi-abelian scheme over the toroidal com-
pactification of the Mumford moduli extending the universal abelian scheme
(see Theorem 6.20) constructed in [DAV] VI.1 is (strongly) quasi-projective
(because its abelian variety quotient has polarization and the toroidal com-
pactification is Noetherian). Then HilbQls exists as a quasi-projective scheme
(we only proved this when 9/ S is projective; see [NMD] Theorem 8 in Sec-
tion 8.2). Thus the scheme EgiS exists. It is known that any homomorphism of
semi-abelian schemes 90 -+ 90 defined over an open dense sub scheme 0 c S
extends uniquely to 91 s -+ 9j s (endomorphisms are kept under specializa-
266 6 Moduli Schemes
tion; a theorem of M. Raynaud, [NMD] 7.4-5 and [DAV] 1.2.7). Thus if 9/s
is an abelian scheme over a dense open subscheme 0 c S, we have a unique
extension of the scheme Eg % over 0 to a scheme End g / s C Eg/ s over S
which represents the functor T I---t End(QT IT) for the group endomorphism
algebra End(QT IT). Applying the valuative criterion using this rigidity of en-
domorphisms, we find that End g / s has connected components each projective
over S. This shows the following fact.
Corollary 6.9 Let 9/s be a quasi-projective semi-abelian scheme over a
Noetherian base S. Suppose that 9 is an abelian scheme over an open dense
subscheme of S. Then the functor T I---t End T (9T/T) is represented by a
scheme End g / s C Eg/ s over S. Each connected component of End g / s is
projective over S. If the connected component contains an isogeny, it is finite
over S. Here EndT(QT/T) denotes endomorphisms of 9 compatible with group
structure on 9.
This result enables us to prove
Corollary 6.10 Let S be a Noetherian scheme, and 9/s be a quasi-projective
semi-abelian scheme which is an abelian scheme over a dense open subscheme
of S. Let D be a finite-dimensional simple algebra over Q and 0 = OD C D
be a subalgebra such that OD is of finite type over Z and D = OD Q9z Q.
Then the functor SIs I---t [(Qs"e E Homalg(OD,End(Qs'/s,»)le(lD) = idg]
is representable by a scheme Eg/ s over S, and each connected component of
Eg/ s is finite over S. If S is a scheme over a discrete valuation ring of residual
characteristic p > 0, each connected component of Eg/ s is unramified over S
in the sense of [SGA] 1.3. The algebra homomorphism e : OD --+ End(Qs' /s')
as above is in fact an embedding.
Since the property (det) is compatible with base change, the above functor is
a well-defined subfunctor of the functor in Corollary 6.10.
Supposing that Q/s is an abelian scheme with a symmetric polarization
A: Q --+ tQ (t A = A), we introduce a sub functor of (6.6): We have the Rosati
involution a H a* := A-iota 0 A of End!Q(Q/s) = End(Q/s) I8lz Q. We
suppose having an involution p of D and consider the following subfunctor
HO(P;R\ 0(1)) ~ R n ,
the same fact for a* (0 (1)) shows that C is trivial and m = 1. Thus a induces
an automorphism of 0(1), which is an element of GLn(R). 0
6.2 Quotients by PGL(n) 271
For each a J , GL(n + 1) acts on the column vector t(Xcio. J), xio. J), ... , X~o.J))
by matrix multiplication; thus, Do.b 0 x) = detb)Do.(x) for 'Y E GL(n +1).
Thus Do. is invariant under GL(n + 1) up to units.
Proposition 6.14 Let {Sl, S2, ... ,Sv} be a collection of subsets of the set
{O, 1, ... ,n} such that
(1) SJ and U~:; S, have only one common number p,(j) for all 2 ::; j ::; v,
and we define p,(I) to be the least integer in Sl;
(2) {O,I, ... ,n}=U;=lSJ.
Define an affine open subscheme U c P by
°°
(a) DI i= for 1 = (0,1, ... ,n);
(b) D I (,) i= for all i E Sk and for all k = 1,2, ... , v, where
R_
77[ (n+1) , ... , Xn(n+l)., Xo(n+2) , ... , Xn(n+2)..
a. Xo
(n+v)
, ... , Xo
(n+v)]
, ... , Xn
- ((n+k)
x, - 11·Z E S k, k - 12
- " ... )
,v .
272 6 Moduli Schemes
To make this legitimate, we count the number of relations: x~n+k) -1 for each
i E UjSj = {0,1,2, ... ,n}. Hi = M(k), the index i produces two equations
for k = 2, ... , v, and the number of equations is (n + 1) + (v - 1) = n + v.
The above spectrum is a linear subspace in (n + 1) v-dimensional affine space
defined by n + v independent linear equations; so, it is isomorphic to A nv-n.
We then define an embedding ~ : Anv-n y U as follows. For a graded
algebra A = EBk Ak and a homogeneous element a E A+ = EBk>oA k , we write
°
D+(a) = Spee(A[~]o) c Proj(A), where the subscript indicates the degree°
component. Writing Pi : p -+ pn for the ith projection, the composite
~, = Pi 0 ~ which has values in D+(Y) for Y E {X6 i ), ... , X~i)} is determined
by an algebra homomorphism ~i : Z[X6') /Y, ... , X~i) /Y] -+ R.
When i ::; n, we take Y given by Y = X?), and ~i is constant taking
X)') / X,(i) to the Kronecker symbol Dij. When i > n, then we have i - n = k for
some k = 1,2, ... , v, and we choose i, : Spee(R) -+ pn sending the coordinate
(Xo(n+l) , ... , Xn(n+l) ; Xo(n+2) , ... , Xn(n+2) ; ... ; Xo(n+v) , ... , Xn(n+v)) 0 f S pee (R) to
the homogeneous coordinate (x6i ), ... , x~)) of pn. This is legitimate because
for j E Sk, we have x~,) = 1 in Spee(R). This defines a closed immersion of
Spee(R) into U, and ~i(Opn(l)) is the structure sheaf R of Spee(R).
Since Do is invariant (up to the scalar) under the action of GL(n+ 1), U is
stable under PGL(n + 1), and we can extend ~ to PGL(n + 1) x Spee(R) 4 U
by ~(g, x) = g(~(x)).
To define the inverse ¢ of ~, for each given x E U, we need to show that
we can find a unique 9 E PGL(n + 1) so that for i ::; n, Pi(g(X)) is the origin
o in D+(X,(')) = An with homogeneous coordinate (0, ... ,0, to, ... , 0). The
i >- e(i) >- e2(i) >- e3(i) >- ... >- ek(i) = M(l) for some k.
It is easy to check that cPR has values in Spec(R). Since cPp(x) and cPR(X)
contain all the information necessary to compute the coordinates of x, it is an
immersion. By a simple combinatorial computation, basically by definition,
we check that cP 0 " is the identity map of PGL(n + 1) x Spec(R); so, we
conclude that" is a surjective isomorphism. 0
-III no + 1
->- -.
m+1 - n+1
Then we see P - Ust = U1 , n 0 V(I, no), where
Since the matrix X made of homogeneous coordinates of xC,) for x E V(I, no)
has rank at most no, all the (no + 2) x (no + 2)-minors of X have determinant
O. Thus V(I, no) is a closed subscheme and, hence, Ust is an open subscheme.
Proposition 6.16 A geometric point x E Ust if and only if there are integers
N ~ No > 0 and monomials Po, ... , Pm of Da as in Definition 6.13 such that,
for 0::; i ::; m, P, E r(p,.er! ® ... ® .et;'-No ® ... ® .e!;/.) and P,(x) 7"= o.
We repeat the proof of Mumford ([CIT] Proposition 3.6).
Suppose {X(')hEI spans a linear subspace L of pn. Then Da(x) of- 0 implies
the maximum number of Q, in I has to be dim L; so, (6.9) holds (in addition
to Da(x) of- 0) if and only if
£(1) = 0 and £(d(Da)) :2': 0 for all Q with Da(x) of- O. (6.10)
Once this is proven, we just take I = {f3 + 1, ... ,m} and J = {O, 1, ... ,f3},
and the assertion follows.
We prove that the above expression of £ is possible if (6.10) holds. We may
assume that Da(x) of- 0 and £(Da) = L, Aa, is minimal; so, we may assume
that x(O), ... , X ((3) -1) span a linear subspace of dimension j - 1 but X ((3) ) does
not lie in this linear subspace for all j = 1,2, ... ,n. Thus x(O), ... ,x((3n) span
entire pn, and hence D(3(x) of- 0 for 13 = (0,131, ... ,f3n), and
C = {(AD, AI,·· ., Am) E zm+1 I AD :s; Al :s; ... :s; Am and LA, = O} ,
Corollary 6.18 Let the notation be as in Propositions 6.14 and 6.16. Suppose
that x E P = (pn)m+l (for m > n) is stable. Then there exists a sequence of
subsets {51, 52, ... , 5 v } for v = m - n satisfying the conditions in Proposi-
tion 6.14 such that x E unust , where U is the open subscheme defined by the
sequence as in the proposition. Moreover, there is an open neighborhood Uo of
x in Un Ust defined by P o:J. 0 for a monomial P in the Das that is a section
of (.co ®.c 1 ® ... ® .cm)M for some integer M > O.
Proof. Choose N > No > 0 and choose Po, ... , Pm as in Proposition 6.16. Let
Q = Q9D (x)#O Do.. Then P = QNo ® p;o ® ... ® P:;,= for a suitable rJ is a
Q
We then have a smooth geometrically irreducible scheme 9Jt;{N over 7L[(N, d~]
representing the functor classifying triples (A, A, ¢N) satisfying (Al-3) and
(A4'). We apply Corollary 6.11 to the universal abelian scheme XI'JJlJ' and
d,N
Let (A, ¢N, A)ls E Ad,N(S). We consider the invertible sheaf LIs given by
the direct image of L~(A)3, which is very ample, because .c 3 is very ample
if .c is ample (see [ABV] Section 17) and ample ness of L~(A) follows from
A(L~(A)) = 2A as we already remarked. Let S(L) be the symmetric algebra
of L (as in Section 6.1.1) and write P(L) for the projective bundle Proj(S(L))
which is a projective scheme over S locally isomorphic to Pis' A linear rigidi-
jication is an isomorphism L : P (L) ~ Pis. Thus L is determined by the choice
of a base of L up to scalar multiplication. Since the very ample sheaf L~(A)3
on an abelian scheme satisfies ([ABV] Section 16)
• L = f*(L~(A)3) is locally free of finite rank 6nd, and
• R2f*(L~(A)3) = 0 if i > 0,
6.3 Mumford Moduli 279
the formation of f*(y1()..)3) as above commutes with base change (cf. [ALG]
III.12.11). Thus the association 5 >--+ A~,N(5) = [(A,¢N,)..,L)/5] is a well-
defined contravariant functor. The embedding J : A I 5 '--+ P (L) ~ P I5
deter-
mines the sheaf V:1()..)3 = I*Op(L)(l), which in turn determines).. because
A(L) = 6).. (PicA15 FA is torsion-free; see [ABV] Section 19 Corollary 2). Hav-
ing ¢ N is equivalent to having 2n (linearly independent) sections CTJ = ¢ N (e J )
of A over 5 for the standard base {e1, ... , e2n} of (71,/ N71,j2n. We write CTO = 0
for the identity section 0 of A. We record here what we have seen.
Proposition 6.22 The data (A, ¢N,).., L) are completely determined by the
embedding data (1 : A I 5 '--+ PIS' CTo, CT1, ... , CT2n).
In other words, defining another functor 1-Ld,N(5) : 71,[*"]-5CH --+ SETS by
we have an isomorphism of functors: A~,N ~ 1-Ld,N. Here CT1, ... , CT2n are sup-
posed to give 2n linearly independent elements of A [N]( 5) .
For simplicity, we just write P for Pillli" We write So for 5pec(71,[*"]). We con-
sider the functor Hilb~ associating with each 5 the set of closed subschemes
of PI 5 flat over 5 with Hilbert polynomial P. As we have already seen, this
functor is represented by a scheme H = Hilb~. Write Z --+ H for the uni-
versal flat family inside PI H with Hilbert polynomial P. For each subscheme
V c PI 5 flat over 5 having Hilbert polynomial P, we have a unique morphism
h: 5 --+ H such that V is given by 5 XH Z C P 15 over 5.
By the generalized Riemann-Roch theorem ([ABV] Section 16), the Hilbert
polynomial of (A, L) (or of J(A)) is given by P(T) = 6n d . Tn. Thus the image
J(A) induces a unique So-morphism h : 5 --+ H such that J(A) = 5 XH Z as
subschemes of P.
k
Thus Hk classifies all flat closed subschemes with Hilbert polynomial P having
k sections over 5. The projection p: Z(k) = Z XH Hk --+ Hk (p(zo X z) = z)
gives the universal scheme and has tautological k sections
This shows that 1-Ld,N c Hk for k = 2n + l. For simplicity, write Ho for Hk.
Since "smoothness" is an open condition (because it is local; in other words,
280 6 Moduli Schemes
6.3.5 Conclusion
Proof. We need to check the inequality of Definition 6.15 for the point fN(A).
Since the total number of the points is M = N 2n , we only need to prove
for any hyperplane H c pm. Write (Y, Z) for the intersection number of
complementary cycles Z and Y in pm ([ALG] 1.7 and Appendix A). Choose
a I-dimensional cycle "( = ¢(A) XP'" L for a linear subspace L c pm of
co dimension n - 1. We suppose that "( contains the origin of ¢(A). Since
multiplication [N] by N acts on divisors by N 2 ([ABV] Section 6, Corollary 3),
it also acts on the I-cycles by N 2 (by the Albanese-Picard duality; [ABV]
Section 23); so, [N]*"( = N 2"( up to numerical equivalence. Here we use the
fact that char(k) f N. Writing h = ¢-1(H) = ¢(A) Xp'" H, we have
Out of 0, T1, ... , T2n, we create the set of full sections 81, ... ,8M : H5 '-+ Z5
through linear combinations of T J • We embed H5 into (pm)M by sending
x E H5 to (81(X), ... ,8M(X)) E (pm)M. Since the image is contained in the
open set of stable points Ust C (pm)M fiber-by-fiber; so, it is contained in
Ust . Since Ust is a PGL(m + 1)-torsor over its geometric quotient (Z,7r) in
Theorem 6.19, we find the geometric quotient of H5 by PGL(m + 1) as the
image 7r(H5) of H5 in Z. Since we know the Hilbert polynomial of Z5,
we know the degree deg(Z5) (cf. [ALG] 1.7.6). Thus if N is sufficiently large
as in Proposition 6.23, the quotient VJ1d,N = PGL(m + 1)\H5 represents the
functor Ad,N over Z[-b] (see [GIT] Section 3 in Chapter 7 for more details).
Suppose now that N ?: 3. We choose two distinct large primes p and q
outside dN so that Ad,pN and Ad,qN are representable by the above argument,
respectively, over Z[p~] and Z[q~]' The finite (constant) group
[ABV] Theorem 5 in Section 21), the action of the finite group GL 2n (Z/pZ) is
free. If we can cover Ad,pN by affine open subschemes stable under the action,
we can again make the geometric quotient
which represents Ad,N over Z[p~] (see [SGA] V.l, [ABV] Section 12, or [GME]
1.8.3). In order to give a sketch of a proof of finding an affine open cover of
9J1 := 9J1d,pN stable under the action, we consider the universal abelian scheme
X ~ 9J1 and ~ = det(7r*.ftx / 9J1 ). The line bundle ~ is ample (cf. [CSM] V.3, or
see the following subsection for another proof using a compactification of the
e
moduli). By the existence of nontrivial sections invariant under GL 2n (Z/pZ)
(given by theta constants) of ~®k for sufficiently large k, 9J1[~] is affine and
stable under GL 2n (Z/pZ). Since abelian schemes are embedded into a projec-
tive space by algebraic theta functions giving rise to theta constants of level
N if N ~ 3 (see [CSM] Appendix II), moving around e, 9J1[~] covers 9J1. Thus
we have the geometric quotient 9J1 d,N/Z[ p\v 1 as above.
By the universality, the two schemes 9J1 d,N/Z[p\vl and 9J1 d,N/Z[q\vl thus ob-
tained glue together uniquely over Z[p:N] giving rise to the scheme 9J1 d,N/Z[-b 1
representing the functor Ad,N over Z[ -tv]. Even if N ::::; 3, the above process
yields the coarse moduli scheme 9J1 d,N over Z[ -tv] (cf. [CSM] Theorem 1.4, or
[GME] 2.3.2 in the elliptic modular case).
Once represent ability of Ad,N is shown, as long as Aut((A,¢N,A)/iB) =
{I} for any prime p f dN, the formal completion of 9J1 d,N at the closed point
x E 9J1 d,N(iFp) carrying (A,¢N,A)/iF p is isomorphic to a power series ring
W[[t 2J ]h:'02:'OJ:,,;n for the ring W of Witt vectors with coefficients in iFp . This
follows from the deformation theory of abelian schemes (see [DAV] I.3 for a
summary, [CBT] Chapter V for details, and also Section Section 8.2 partic-
ularly 8.2.4 in the text). Thus 9J1d,N for N ~ 3 is smooth over Z[d~]. This
finishes the proof of Theorem 6.20.
via the trace pairing (a, b) >--+ Tr F/Q1 (ab), and F = a Q9z R Indeed we started
(X)
Then we have
HomS-alg-gp(T, G m) = HomS_alg_gp(Y 0z G m , G m)
~ (X*)* 0 HOmS-alg-gp(Gm, G m ) = X.
For any given extension over an 8-scheme 8',
Note that HomS_alg_gp(T, G m) = XIs and that the dual abelian scheme tA/ s
is equal to Ext1_alg_gp(A, G m). See [CSM] Remark 2.1.4, group extensions of
A by G m (not just scheme extensions) correspond to Gm-torsors (hence line
bundles) in the identity-connected component of PiCA/So Thus we have an
association G H ¢c E Homz(X, tA).
On the other hand, we take the universal extension
1 1
Homz(X, tA) --------t tAo
Pulling back the above exact sequence 0 ---+ <G m ---+ G ---+ A ---+ 0 over tA by
this square, we get 0 ---+ Homz(X, <G m ) = T ---+ G ---+ A ---+ 0 over Homz(X, tA)
such that for any given X E X, the morphism Homz(X, tA) ---+ tA given by
¢ H ¢(X) gives rise (by push-forward) to an extension
sending ¢ : X~ -+ tx to ¢(p). Recall the Poincare line bundle P, that is, the
universal line bundle on X x tx /tx such that for each S-valued point S -'=t tx,
a*P E PicO(X/ s ) is the line bundle represented by a. Let px = P - {O} be
the Gm-torsor associated with P.
We then consider
follows from the symmetry of P and the fact that S(Xu) ~ Pic(Zu) given by
n f-7 c( n )*P is a group homomorphism. Since Eu is the product of the pullback
of P itself by c x c, c( n) *P has a tautological section (or trivialization) T over
Eu. This trivializes the biextension associated with P and hence gives rise to
a homomorphism i : Xu ~ Gu lifting the tautological one c = tc : Xu ~
tx = X on Zu over Eu (see [CSM] II, (2.3.3) for the details and [Mu] for
biextensions) .
We form the contraction-product Eu = Eu x Ea Eu = (Eu xZa Eu)/Eu
under the diagonal action. This product has the following universality.
1. We have the following commutative diagram
Tn
Eu xZa Eu ~ ~u
Eu
21
XZa Eu ~
I'
~u,
Eu XZa Eu ~ F
'I I'
Eu XZa Eu ~ ~u,
I-'
EB EB H
00 00
The last identity (*) follows from Koecher's principle (Theorem 6.24 (4)) if
n > 1. Choosing the toroidal compactification data well, MN is projective
([DAV] V.2.5) so that this graded algebra Qj is finitely generated over Z[-kJ,
and by the first equality, the graded algebra is normal. Thus we may define Qj N
290 6 Moduli Schemes
m=-oo
Remark 6.25 Let the notation and the assumption be as in Remark 6.21.
Carrying out the toroidal compactification process for SJJCfN to obtain a
smooth compactification Mi,'rv
over Z[(N, d~], we apply C~rollary 6.11 to
the semi-abelian scheme Q/MJI . Then, automatically, we get a toroidal com-
d,N
pactification of the moduli SJJC( c, r(N)) with respect to the data induced from
the toroidal compactification data of S*(O) (as long as they are compatible).
Here i)-l is embedded into S* (0) regarding ~ E i)-l as a symmetric bilinear
form on 0 x 0 by (x, y) r--t TrF/IQ(~XY). An important compatibility of the
toroidal compactification data is that the polyhedral decomposition of positive
semi-definite C(OIR) C S*(OIR) has to induce that on F::'+ C i)-l Q9z lR = Foo
by taking intersection with the image of i)-l.
In this section, we study the Shimura variety associated with G = GSp(2n) /IQ'
where, for a ring R,
-ci
for I n = (l~ n ) with 0 < n E Z. Consider a homomorphism ho : S/IR :=
ResCjIRG m --t GSp(2n)/IR of algebraic groups given by S(lR) = ex 3 a + bi r--t
(~i/ln ~:i~) E GSP2n(lR). Let X be the collection of all conjugates of ho under
GSP2n (lR), and write X+ for the connected component of X containing h o .
Then we have X ~ SJn U SSn by sending ghOg-l to g( R . In) E SJn, where
6.4 Siegel Modular Variety 291
Therefore the construction of the Shimura variety for (G, X) follows from the
existence theorem (Theorem 6.20) of the moduli variety over Q of polarized
abelian varieties (proven by Mumford [GIT] and Shimura [Sh3]).
In this section, we give in this section an exposition on its global reciprocity
law and the irreducibility of the Igusa tower over Sh(p)(G, X).
We consider the Q-vector space V = Q2n with alternating form (x, y) = txJny.
Thus G(R) = GSP2n(R) acts on VR = V ®Q R by matrix multiplica-
tion preserving the alternating form up to scalar multiplication. For each
(A,A)ls E A 1,1(S), choosing a geometric point s = Spec(k(s)) of each con-
nected component of S, we consider its physical Tate module
Ts(A) = ~NA[N](k(s)),
We consider level structures TJ : VA(oo) ~ Vs(A) and TJ(p) : VA(poc) ~ V}p) (A).
An element 9 E G(A) acts on TJ (and on TJ(p)) by TJ H TJ 0 g. For a closed
subgroup K c G(A (00)), the level K-structure fj = TJ 0 K is defined over S,
if 0' 0 TJ = TJ 0 ks(u) with ks(O') E K for all 0' E 1f1(S,S) for each chosen
s. Changing s to another point s' in the same connected component gives
isomorphisms (called paths) Ls,s': 1f1(S,S) ~ 1f1(S,S') and LL,: Ts(A) ~
Ts, (A) which satisfy the compatibility condition (see Section 4.4 in the text
292 6 Moduli Schemes
L/NL ~ A[N] (compatible with the polarization form (-, .)). Thus £~0) is
represented by Sh(G,X)/Q = ~NOO1~/Q'
In order to construct a p-integral model of the Shimura variety, we impose
(6.11) by substituting ",(p) for", and require that X be a polarization class of a
principal polarization up to a prime-to-p scalar multiple. Then we can consider
the following smooth model Sh(p)(G,X)/zCP) = ~pfNOO1~/zCP)' whose quo-
tient Sh~) = Sh(p)(G, X)/ K by a closed subgroup K with K = G(Zp) x K(p)
(K(p) c G(A(p))) represents the following functor £jI) : Z(p)-SCH -+ SETS,
(6.13)
where fj indicates the K(p) orbit ",(p) 0 K(p). Here the isomorphism classes are
taken in the category of abelian schemes up to prime-to-p isogeny; so, we have
an isomorphism (A, X, fj(p)) ~ (A', X' , fj' (p)) if we have an isogeny cp : A -+ A'
with p f I Ker(cp)I = deg(cp), cp*X' = X, and fj'(p) = cp 0 fj(p) up to a prime-to-p
scalar multiple.
(1) On Q(17), g E G(A(17oo)) acts by the Artin reciprocity map in Theorem 2.12
composed with the similitude norm v: G(A(17oo)) ---+ AX;
(2) For f E jt(17) regarded as a function on X+, we have T(a)(f) = f 0 a- 1
for a E G(Z(17))+;
(3) The map T is an isomorphism of topological groups with respect to the
adele topology on l(17)(G,X) and the Krull topology on Aut(jt(17)).
The action of T(a) looks different from the one in Theorem 2.43 and actually,
our T(X) here is the inverse of T(X) in Theorem 2.43 in order to make the
construction of T compatible with our definition of the action on the level
structure TJ H TJ 0 g (see (6.14) below).
Recall the valuation ring Zi~) (with residue field IF) in Q(17) of the p-adic
place ip : Q(17) y Qp. If PEE, we have a modulo p version of this theorem
determining Aut(e~17)) for the function field e~17) of 9J1j;) = 9J1(17) Q9Z(E) IF.
(p)
Since it is similar to Theorem 4.17, we leave the formulation of the result to
the reader (see [H03b]).
We repeat here the argument given for the Hilbert modular version (The-
orem 4.14) based on an argument in [AAF] Section 8 and the theorem of
Faltings and Zarhin (Theorem 4.13). For the original proof of Shimura (and
Miyake), we refer the reader to [Sh5] II 6.5 and [Mt].
Proof. Since the proof is the same for any choice of E, we assume that E = (/)
for simplicity, and we throw all superscripts and subscripts E away from our
notation. By our convention, Z(0) = Q. We first check that T(a- 1 ) (f) = f 0 a
by studying the complex points 9J1j{,(cC) of 9J1j{,.
For a given (A, A) E £1(cC), we choose a base {Wl, ... ,wn } of HO(A, DA/c) ,
and consider the lattice .c A = {U-y Wj) J 11' E Ho (A, Z)} in en. By the Poincare
duality and the principal polarization A, we may assume that
.cA = .c z = t L . In ( 1: )
for Z E -Dn (see [AAF] 4.4), regarding L to be a column vector space. Then
A(cC) ~ en j.c z as complex manifolds, and A is induced by the alternating form
1: )/\
Az : .c z /\ .c z ---+ Z sending tx . I n ( ty . I n ( {n ) to x . I n ty E Z. We write
Az for the abelian variety with Az(cC) ~ enj.c z . Define the main involution ~
of G by a· = J;;l . hJn for a E G(R). Note that (ax, y) = (x, a·y) for the
alternating form (x, y) = tx . Jny on L.
If we have an isogeny ¢ : A(cC) ~ en j.c z ---+ en j.c w ~ Y(cC) preserving
the polarization class, and identifying en with the Lie algebra of the abelian
varieties, we have a e-linear map if> : en ---+ en lifting ¢ with if>(Q.c z ) = Q.c w
preserving the polarization form. In other words, we find a E GL 2n (Q) such
that
if> (e . I n (1: ))
= e· a . I n U':' ) j (a·, z)
for all eEL, where a· = J;;l . hJn = v(a)a- 1 and j(g, z) = cz + dE GLn(cC)
for g = (~ ~). Since a preserves polarization up to positive rational numbers,
v(a) > 0, and hence a E G(Q)+. This shows that w = a(z) for a E G(Q)+.
6.4 Siegel Modular Variety 295
we have T(a(oo»)([z, 1]) = [z, a(oo)] = [a-I(z), 1], writing [z, g] for the complex
point of Sh(G, X)(C) represented by (z x g) E nn x G(A.(oo») as in (4.21).
This shows that a E G(Ql)+ c G(A.(oo») acts on f E ~ by f H f 0 a-I.
By the above expression of the complex point of 9J1, we find that T
brings G I (A.(oo»)/{±I} injectively into Aut(9J1/IQI)' We consider the eN,z-
pairing eN,z : Az[N] x Az[N] ----t /LN· Since eN,z(¢N,z(X), ¢N,Ay)) = (]:,y)
for x, y E LIN L, we find that eN,z(ax, ay) = (~(n)(x,y) for a E G(Z). Then in
the same manner as in the proof of Theorem 4.14 Step 2, we create a section
SA : ZX '3 C H (C~n IOn) E G(A.(oo») of V so that the action of T(SA(C)) on the
q-expansion at 00 factors through the Galois action of the Artin symbol [c, Ql]
on the coefficients of the q-expansion. Thus we get the following commutative
diagram with exact rows (and the section SA of v),
f(q) = 2:~2':O a(~, f)q~ with a(~, f) E 8, where ~ runs over positive semi-
definite n x n matrices with coefficients in Q. The minimal compactification
of >JJi.!v/Q is given by Proj(r8f(N)). Thus any element f in the function field
jt = QCYC(>JJi.) for >JJi. = ~ N>JJi.!v Q[J.!N 1 can be written as J with n, d E
r8k for a suitable integer k > 0, w~ere r8k is the kth graded component of
r8. The minimal compactification is projective and can be embedded into a
projective space using theta series, which has integral q-expansion coefficients
(cf. [CSM] Appendix II and [AAF] Section 6). Thus for a given f E jt or
more generally f in the field of fractions Frac( r8) of r8, we can take nand
d with q-expansion coefficients bounded under the p-adic valuation ordp of
W. Thus, the valuation v(n) = Inf~ ordp(a(~, n)) is a well-defined element in
Q U {oo}. Similarly v(d) E Q is also well-defined. Then we define the valuation
v: Frac(r8) -+ Q U {oo} by v(f) = v(n) - v(d).
We have an isomorphism r : £(G, X) ~ Aut(jt) = Aut (>JJi./Q) of
topological groups (Theorem 6.26). Since the restriction of r(x) to QCYc is
given by the action of the similitude norm v(x) E AX /QxlR.~ composed
with the Artin reciprocity (Theorem 2.12), r induces another isomorphism
G 1 (A(00))/{±1} ~ Aut(>JJi./QCYc) = Aut(jt/QCYC).
We write Vp for the decomposition group in Aut(>JJi./ Q) for this valuation
v. As seen in the proof of Theorem 3.3, Vp is a closed subgroup of r(£(G, X)),
and we prove that Vp/Ip contains GLn(7l,p) for the inertia subgroup Ip.
The scheme S* is geometrically irreducible, that is, S*0wQ and S*0wiFp
are both irreducible. We then construct T/:. = ~ KTJ(,a and T; = ~ KTj(,a'
where K c G(Jr,(oo)) runs over all open compact subgroups maximal at p.
Then T~/ S is an etale covering, and we take the geometrically irreducible
component C':x, of T~ containing the infinity cusp, and put C~ = C':x, n T~.
Since the action of GLn(7l,p) for the diagonal subgroup GL(n) of G 1 commutes
with the action of G 1 (A(poo)) and coincides with the action of GLn(7l,p) on
T':x,/ So, we have a commutative diagram with exact rows,
(CO SO)
A u too; ~
-----, Aut(SO)
tu tu
'-+ (G 1 (A (poo)))/ {±1} x H -# G 1 (A (poo))/ {±1},
is dense in the space of p-adic module forms of level poo. From this, we con-
clude that v restricted to j{U(Zp) is preserved by G1 (A. (poo))U(Zp)/ {±1 }U(Zp),
which is enough to show the stability of von j{ under G 1(A.(poo))/{±1} (be-
cause T(U(Zp)) C T(U(A.(oo)) C Dp).
The closure of the image of G1(A.(poo))P(Q) in £(G, X) is in Dp. Note
that the projection of G1(A.(poo))P(Q) into G1(Qp) contains P(Q) C G1(Qp),
which is dense in P(Qp). Thus Dp contains the image under T of
IK~\Gl(ZP)/DI ~ IK~)\Gl(Zp)/K~)1
Then G 1 (Zp) acts on.e m transitively. We write Vm for the element of .em gen-
erated by the first n standard basis of (Z/pmz)2n. Then K~) C K(l) is the
302 6 Moduli Schemes
stabilizer of V'rn in .em. Identifying .em with the subgroups of A[P1nj of a generic
p-ordinary member A/ w of the universal abelian scheme A/ 9J1 (p) , 9J1 m = 9J1K=
classifies isogenies cp : A --t A/V whose kernel is free of rank n over 7l/pm71
generically. On the special fiber of characteristic p of A, this kernel V = Ker( cp)
degenerates into a 7l/pm 71-submodule of A[pm](lF) ~ (71/p m71)n. This image is
determined up to isomorphisms by the intersection VnA[pmjo (Qp) for the con-
nected component A[pmjo over W, which also determines the p-isogeny type of
cp. In [Shlj Theorem 5, it is proven (under some assumptions later removed in
[DAV] VIl.4) that each irreducible component of 9J1 m / IF is determined by the
characteristic 0 intersection VnA[pm]o(Qp). We only need a weaker assertion
that the p-isogeny type only depends on components. Indeed, the p-isogeny
type is locally constant on S h K = /F)1I] by the Serre-Tate deformation theory
(see Theorem 8.9 and the argument around (8.25)). Thus IK,\; \G 1 (71 p)/DI 2>
l.em/rvl, where V rv V' if V n (A[pm]o(Q p)) = V' n (A[pm]o(Qp)). It is easy to
show that the number l.em/rvl is equal to IK,\;\G 1 (71 p )/K'\;) I, which shows
the desired assertion (and also the bijection between .e m / rv and irreducible
components of 9J1 K=/IF). We refer the reader to the book of Faltings and
Chai [DAV] VIl.4 for more detailed information on the mod p correspondence
[G 1 (71 p)O:pG 1 (71 p)]. 0
7
Shimura Varieties
We fix a left D-module V of finite type and assume that we have a nonde-
generate alternating form ( , ) : V x V -+ CQl such that (bv, w) = (v, b*w) for
all bED. Write Vp = V Q91I] CQlp and V00 = V Q91I] R We also assume having an
OD-submodule LeV of finite type such that
(L1) L Q9z CQl = V;
(L2) (,) induces Homz p (Lp, Zp) ~ L p, where Lp = L Q9z Zp-
Put C = EndD(V), which is a semi-simple CQl-algebra with involution denoted
by i (which is often not positive) given by (cv, w) = (v, c'w). Then we define
algebraic CQl-groups G, GU, and U by
As before, we write G I for the derived group of G (and GU), and assume
(sc) G I is simply connected with noncompact G I (JR.).
By this assumption, we have the strong approximation theorem valid for G I ,
that is, GI(Q) is dense in GI(A(oo)) (see [Knl). This type of group covers all
the cases of Shimura varieties we have studied so far. If we take D to be a
totally real field F with the identity involution *, V = F2, L = 0 EB c*, and
((x,x'), (y,y')) = xy' - yx', we have C = M 2(F) with involution x f---t x< =
det(x)x- 1 and therefore GU = ResF/i(JIGL(2) and G 1 = U = ResF/i(JISL(2);
so, we are in the Hilbert modular case. If we take D = Q with the identity
involution, V = Q2g and L = 7/.,2g with (x, y) = tx · Jgy, we have C = M2g (Q)
with involution x f---t Jgtx· J 9 1 and therefore G = GSp(2g) and G I = U =
Sp(2g); so, we are in the Siegel modular case. If * is of the second kind, U is
not equal to the derived group G I = Guder = Gder, and G I is often written as
SU (a special unitary group). We have the inclusion G I = SU cUe G c GU
and exact sequences of fppf-sheaves (see Section 8.2.1 for fppf-sheaves),
1 -+ U -+ G ~ G m -+ 1, 1 -+ U -+ GU ~ ResFo/i(JIGm -+ 1,
1 -+ SU -+ U ~ ResFo/i(JI Ker(Norm : ResF/FoGm -+ Gm/ Fo ) -+ 1,
det x II If' If'
and 1 -+ SU -+ GU ------+ ResF/i(JIllJTm x ResFo/i(JIllJTm, (7.2)
The connected components X(j and X(ju containing ho of these two spaces
are identical.
Since h : C -+ Coo is an lR-algebra homomorphism, we can split VIC =
V ®IQI C into the direct sum of eigenspaces VIC = VI E9 V2 so that h(z) acts on
VI (resp. V2 ) through multiplication by z (resp. z); thereby, we get a complex
vector space structure on Voo by the projection Voo ~ VI. Since h(C) C Coo,
h(z) commutes with the action of D; so, ~ is stable under the action of
DIC = D ®IQI C. We get the representation PI : D '-+ Endc(Vd. We define E
for the subfield of C fixed by {o- E Aut(C)lpl ~ pd. If h'(Z) = g. h(z)g-I for
g E GU(lR), hi induces a similar decomposition VIC = V{ E9 V;, and g induces
a D-linear isomorphism between VI and V{; so, E is independent of the choice
of hi in the GU(lR)-conjugacy class of h. This field E is called the reflex field
of (G, X) (and is a canonical field of definition of our canonical models of
the Shimura variety). By the positivity (pos), the quotient complex torus
Voo/L = VdL has a Riemann form induced by (-, .). The theta functions with
respect to the associated Hermitian form (.,.) in (pos) gives global sections
of an ample line bundle (e.g., [ABV] Chapter I) on Vd L and hence embed
VI / L into a projective space over C. The embedded image is the analytic
space associated with an abelian variety A h / IC • Multiplication by b E OD on
Vd L induces an embedding i : D '-+ EndlQl(Ah/d. Since (bx, y) = (x, b*y),
i(b) H i(b*) is induced by the Rosati involution of the polarization of the
Riemann form (', .). The representation PI is given by the action of D on the
tangent space Lie(Ah) = VI at the origin of A h . Since Ah is projective, the
field of definition of the abelian variety Ah is a field of finite type over Q. By
Corollary 6.11, the couple (Ah'" : D '-+ EndlQl(Ah)) is defined over a field of
finite type over Q. By definition, the field E is the field of rationality of the
representation of D on Lie(Ah); so, the field of definition of (Ah' ,,) always
contains this field E. It would then be natural to expect that the moduli
variety of triples (A, A, ,,) for an abelian variety A with D-linear isomorphism
Lie(A) ~ VI is defined over E.
Since the isomorphism class of PI is determined by Tr(PI) (see [MFG]
Proposition 2.9), E is generated over Q by Tr(pI(b)) for all bED. The field E
is a subfield of the Galois closure pgal of P /Q, and E is a finite extension of Q.
Moreover, under (unr), p is unramified in E/Q. We write OE for the integer
ring of E. Let Z(p) = Zp n Q and put OD,(p) = 0 ®z Z(p), O(p) = 0 ®z Z(p),
and V = OE ®z Z(p). The ring V is a semi-local Dedekind domain with
quotient field E in which p is unramified by (unr).
Let K(p) be an open compact subgroup of G(A,(poo)). We study the clas-
sification problem of the following quadruples (A,A,i,17(P))/s: A is a (pro-
jective) abelian scheme over a base S, t A = Pic~/ s (A) is the dual abelian
scheme of A, A : A -+ t A is an isogeny with degree prime to p (prime-to-p
isogeny) fiber-by-fiber geometrically induced from an ample divisor (polariza-
tion), i : OD,(p) '-+ End~(P) (A) = Ends(A) ®z Z(p) is a Z(p)-algebra embed-
ding (taking 1 to the identity of A) with A 0 i(a*) = 4(a) 0 A for all a E OD,
7.1 PEL Moduli Varieties 307
Z(p)} of )..' to the class:\" of )..: ¢*:\"' = :\". In this case, we write A ~ A'. We
write A ~ A' if the isogeny is an isomorphism of abelian schemes; that is,
deg(¢) = l.
We take the fibered category C = Cv,Y of the quadruples (A,)..,i,'i](P))ls
over the category V-SC H of V-schemes and define
E(p) (S) = E};) (S) = {£lIS = (A, :\", i, 'i](p)) IS I A satisfies (det) below} I ~ .
Here the determinant condition (det) is given as follows. Recall V = 0 E ®z
Z(p) for the integer ring OE of E. As already remarked, V is a semi-local
Dedekind domain with quotient field E. The condition (unr) guarantees that
ov ® V 1m is semi-simple for all maximal ideals m c V. We choose a Z(p)-base
{aJh:S:J:S:r of OV,(p) and consider a homogeneous polynomial
If two abelian schemes A and A' over S are isogenous by a prime-to-p isogeny
over S, we write A ~ A', abusing our notation slightly. If K(p) is sufficiently
small so that Auts(A, A, i, ry(p)) = {lA}, the prime-to-p isogeny giving the
isomorphism A ~ A' in the definition of [(p) can be taken (as explained in
Section 4.2.1) to be an isomorphism by changing A' in the isogeny class under
"~" (and insisting 7](p)(Dp)) = T(A) @z Z(p); d. [D1] 4.10). Therefore we
have []I)(S) S:! [K(S) for another functor [K : SCHlv -+ SETS given by
= { ¢ E Isoms(A,A)
'I ¢ t¢
0
0
i
A' 0 ¢ = CA with c E Z(), }
= i' 0 ¢ and ry'(P) =¢0
P
ry(p)
. (7.4)
Then we claim that the fibered category Cover V-SC H is an algebraic stack
for any given K(p) and is an algebraic space if Autc/s (A) = {lA} for all
objects A of CIS. See [DM] and [DAV] 1.4.6-9 for the definitions of algebraic
310 7 Shirnura Varieties
stacks, but this statement is basically equivalent to the fact that £jf) has a
coarse moduli scheme over V (compatible with base change). Indeed, if K(p)
is sufficiently small, £jf) is representable by a quasi-projective scheme over V
as we show.
Forgetting D-linearity of 7)(p) and restricting i to Q, we have a functor
from CD,v into the fibered category CIQJ,v of (A,.A, i, fj(p)) for D = Q for a
suitable choice of an open compact subgroup GSp(2d)(A(PCXJ)) (d = dim A =
~ dimlQJ V). As we have shown in Theorem 6.20, this fibered category is proven
to be an algebraic stack and is representable by a quasi-projective scheme
M = MIQJ/v by Mumford (see also [CIT], [Sh3], [CSM], and [DAV]).
We now supplement the above outline with details. We show that we
can replace ~ by ~ in the definition of the functor £);), imposing an ad-
ditional condition. Let A/s = (A,.A,i,fj(p))/s. Then 7)(p) induces V(A(PCXJ)) ~
V(PCXJ)(A) and, therefore, we have 7)(p) ([)p)) c V(PCXJ)(A) for [)p) = L0z7!,(p).
If 7)(p)([)p)) c T(p)(A), the cokernel is an etale group subscheme C c A/ s
locally free over S of rank prime to p. Make the quotient abelian scheme
A' = AIC over S (see [ABV] Section 12 and [CME] Proposition l.8.4) , and
then we have a prime-to-p isogeny cp : A -+ A' with Ker(cp) = C. We then
consider A' = (A', .A', i', 7)'(p))/s given by .A = ¢ 0,Ai 0 cp, i'(a) = cp 0 i(a) 0 cp-1,
and 7)'(p) = cp 0 7)(p). Then A' satisfies 7)'(p) (j}p)) = T(p) (A).
If 7)(p) ([)p)) ~ T(p) (A), we can find a prime- to-p isogeny A' -+ A such that
cpo 7)'(p) = 7)(p),.A' = ¢O.A 0 cp, i'(a) = cp-1 0 i(a) 0 cp and 7)'(p) ([)p)) = T(p)(A).
This fact follows from the canonical identification T(A) = 11"1 (A, 0) for the
origin 0 of A (see [ABV] Section 18).
If neither 7)(p) (j}p)) C T(p) (A) nor 7)(p) (j}p)) ~ T(p)(A), we can find two
prime- to-p isogenies A LA" and A' ~ A" for A" = (A",.A", i", 7)" (p)) / s so
that 7)"(P) ([)p)) = 7)(p) (j}p)) n T(p) (A") and 7)'(p) (j}p)) = T(p) (A'). Thus we
can always find in the prime-to-p isogeny class of a given A/ s , a quadruple
d;s with 7)'(p) (j}p)) = T(p)(A').
If cp : A/ s -+ A;s is a prime-to-p isogeny with 7)(p) (j}p)) = T(p)(A) and
7)'(p) (Dp)) = T(p)(A'), then deg(cp) = 1 and A/s ~ A;s by cp. Thus insisting
7)(p)([)p)) = T(p)(A), we can replace ~ by ~ in order to define the functor
£jf). In other words,
The functor defined in this way can be proven to be representable (as now
show) by a V-scheme Sh~jV"
r(pOO) (N) x rp. We only need to show that CD,v is relatively representable
over the Mumford moduli MIQ given by Theorem 6.20. In order to emphasize
the dependence on (D, V, (-, .)), we write £fi for the functor £K with respect
to K and D. Write 2d = dimlQ(V). Then d = dims A for Als E £fi(S),
and d is independent of the choice of AI s by (det). For simplicity, we as-
sume that the polarization pairing ( , ) in (L2) gives the self-duality of L.
We then identify with GSP2d(Z) the similitude group of ( , ) acting on L. In
this way, we let GSp2d(R) act on V(R) = V 01Q R. Write Ko for the open
compact subgroup of GSp2d(A.(oo») maximal among those preserving Land
the level N structures (the group Ko was written as r(N) in Section 6.4.1).
Then Ko n G(A.(oo») = K and K = r(N) c G(A.(oo»). As described in Theo-
rem 6.20, £Ko is representable by a quasi-projective scheme M = MIQ = wtN
defined over Z(p). Let A = (A, A, r;(P»)IM be the universal quadruple over
M with the universal abelian scheme A E A. We consider the functor from
M-SCH into SETS:
TIS H [(AT,AT,i: OD '-+ End T (A T ))li(ID) = idA, Xii 0 ti 0 AT = i 0 *].
This functor is representable by a scheme MDIM (Corollary 6.11). Since the
level structure r;(p) on A gives rise to a level structure fj(p) of (AT, i), we
have a triple A = (AT,i,fj(P»)IT. Thus £fi is a subfunctor of the above
functor. Again by the rigidity of endomorphisms under specialization (de-
scribed in Section 6.1.6), £fi is represented by a closed subscheme Sh~jv
of MD whose connected components are (each) finite over MIQ. We show
that Sh~) is of finite type over MIQ (so it is projective and finite over MIQ).
Take a geometric point x E MIQ; suppose that we have i : OD '-+ EndA
as above, which gives rise to a geometric point y E Sh~). For a given T,
if T is connected, End(AT) 0;z Q is a semi-simple algebra of finite dimen-
sion with positive involution (see [ABV] IV). Thus the number of embedding
D '-+ End(AT) 0;z Q is finite up to inner automorphisms. Moreover, the num-
ber is bounded by a constant only depending on the dimension of A, that is
d, because dim(EndT(A T ) 0;z Q) is bounded by 4d 2 (e.g., [GME] Theorem
4.1.19). If one changes i by an inner automorphism induced by ex E End(AT)
and if we suppose that (AT,exiex-l,).,fj(p») is still an element of £fi(T), it is
easy to see, by the condition that r/ p) (Ltp») = T(p) (A) combined with (LI-2),
that ex has to be an automorphism of AT. Since automorphisms of an abelian
variety preserving a given polarization are only finitely many by the positivity
of the polarization ([ABV] Theorem 5 in Section 21), there are only finitely
many possibilities of ·having i : OD '-+ End(AT) that give rise to an ele-
ment of £fi(T). Thus Sh~jv -t MIQIV is quasi-finite. Then the projectivity of
each connected component of Sh~) over MIQ implies the finiteness of the map
Sh~) -t MIQ. If K is small, the morphism is unramified and finite. Actually,
one can show that the morphism Sh~) -t MIQ is a closed immersion (over Q)
if N is sufficiently large (cf. [D 1] 1.15; see Section 8.4.2 in the text).
312 7 Shimura Varieties
Ei:) (S) = {A/ s = (A, "X, i, iJ) / s I A satisfies (det) and (pol) below} / ~ .
Here ~ is up to isogeny (not up to prime-to-p-isogeny). Let hA : § --+
Autc(H1 (A, JR)) for § = ReslC/IRGm be the homomorphism inducing the Hodge
decomposition of H 1 (A,Q (see just above (pol) in Section 4.2.1). Note that
Vs(A) = H 1 (A,A(00)) = H 1 (A,Q) ®Q A(oo), and recall the condition (pol)
adjusted to our setting:
(pol) There exists an D-linear isomorphism I : V ~ HI (A, Q) such that
1- 10 hA 0 fE Xc, (f ® 1A,(oo») E (7]0 K) and EA (f(x), f(y)) = a(x, y)
up to a E QX,
where EA is the Riemann form on H 1 (A,Q) (see [ABV] Section 1). This
condition really determines the group G as a similitude group of EA under
the identification by f.
7.1 PEL Moduli Varieties 313
In the same way, we can prove the represent ability of the version E<jJ)
of the functor EAf) replacing in the definition of EAf) prime-to-p-isogenies by
isogenies. The new functor E<jJ) therefore gives rise to the Shimura variety
Sh(G, X) defined over the reflex field E. Recall the warning we made after
Theorem 7.1: the condition (pol) is not imposed in the definition of EAf); so,
EAf) restricted to E-algebras may be a disjoint union of E<jJ) for a different
choice of an alternating form on V which is locally equivalent to (-, -). This
subtle point is studied in Section 7.1.5. See Theorem 7.7 for the description
of the variety Sh(G, X).
Depending on the data at the cusps governing toroidal compactifications,
a semi-abelian scheme 9/ MQ is constructed in [DAV] (see Section 6.3.6). Then
even in the quasi-projective (but nonprojective) case, a similar argument using
Corollary 6.11 (applied to 9/ MQ in place of A/ MQ ) gives a scheme projective
over the smooth toroidal compactification M IQl of the Mumford moduli MIQl.
Since the endomorphism algebra of an abelian variety A/k for an algebraically
closed field k (after tensoring Q) is semi-simple, there are only finitely many
possibilities of embedding D into Endk(A) ®z Q up to conjugation. Thus
the morphism M D -+ M IQl has finite geometric fiber everywhere; that is,
the morphism is quasi-finite. Since the scheme M D is proper over M IQl (see
Corollary 6.11), it has to be finite. Thus writing ~ = f*J2 A / MD for f : A -+
MD and defining det(~) by its maximal exterior product, we can define a
graded algebra:
00
n=O
Moreover, as seen in Section 6.3.6, MQ = Proj(IBl§) and hence we have the
minimal compactification of Sh K defined by Sh'K = Proj(IB}§) finite over
M Q.
The smoothness of M D /V around the cusps for a well-chosen cuspidal
datum was shown by Fujiwara for G and GU of types A and C ([F]). If we
choose the cuspidal data for GSp(2d) and G so that they are compatible
(in other words, so that the pullback of the semi-abelian scheme over M IQl is
the semi-abelian scheme over M D associated with the cuspidal data for G),
this guarantees that the q-expansion parameter is well defined over V and
projectivity of M D over V (if the level is prime to p), because it is finite over
projective MIQl.
Even if K is not very small, we always have a coarse moduli scheme ShY::)
representing the functor Ell over V. The above arguments all work well. We
write Sh K for a toroidal compactification of Sh K and Sh'K for the minimal
compactification. Since the natural morphisms
Sh'K -+ MQ and Sh K -+ MIQl
are quasi-finite and projective, they are finite. Let V be the image of Sh'K
in M Q. Then V = Proj(IB'K) for a graded algebra IB'K which is the quotient
314 7 Shimura Varieties
Here <Bfj is the integral closure of <B'K in the total quotient ring of <B'K. This
follows from the fact that Sh'K is smooth outside cusps, and at the cusps, if K
is sufficiently small, M D is finite over M Q (and normal over V). The graded
algebra <Bfj is the graded algebra of automorphic forms on G if dim Sh K > 1.
{¢ ¢
/S
= E Hom(A, A') ® Z P
()¢
I;;; 0
0
>.' 0 = >. 0 i(a) with a E vt+} , (7.5)
i = i' 0 ¢ and Tj'(p) = ¢ 0 Tj(p)
where vt+ indicates the group of totally positive units in Vo. We then consider
the functor
£~(P) (S) = {AlsiA satisfies (det)} / ~,
where ~ indicates isomorphism classes in Ag~s'
7.1 PEL Moduli Varieties 31G
We may allow all isogenies as morphisms to define the category A (0) (with
full level structure) and define a functor from E-SCH to SETS by
We can also show that 2>' is globally induced by a line bundle on A rela-
tively ample over S as follows: picking one symmetric polarization >., the pull-
back LL1(>.) by 1 x >. of the universal line bundle over A Xs t A/A (the Poincare
bundle) is always ample and is equal to 2>' (see [GIT] Proposition 6.10); so, in
the class "X, we can always find a polarization globally over S associated with
a line bundle on A.
7.1.4 Classification of G
where lliI = lR + lRi + lRj + lRk is the algebra of Hamilton quaternions, 10 is the
set of all field embeddings of Fo into lR, Jk = ( 10k -Jk) and Is,t = (~ -~t) are
as before, and x f-t X is either complex conjugation or quaternion conjugation.
There are type-B (orthogonal) groups admitting Shimura varieties, but
they are non-PEL type except for the case when the signature is either (2,2)
or (2,1) (and hence the groups are isogenous to the product of inner forms
of PGL(2)). In Case D, the simple connectedness (sc) fails, but the argument
proving Theorem 7.1 still (almost) works assuming p > 2 (see [Ko]), and we
get Sh)~. When K(p) is sufficiently small, Sh K is smooth over V. This follows
from the fact that the deformation ring of a quadruple (A, >., i, r;(p)) is always
formally smooth (see [LaR] 6.2, also Theorem 8.8 in the text and the remark
after the theorem). If C = EndD(V) is a division algebra ({o} D is a division
algebra and rank D V = 1), as we have already seen, Sh~jv is projective.
Since the stabilizer Co of the fixed ho E X in G(lR) is a maximal compact
subgroup modulo center by (pos), the connected component of X is isomorphic
to the symmetric domain H = G(lR)/Co . An explicit form of H as a classical
bounded matrix domain is given in [Sh] (see also [ACM] Chapter VI for the
7.1 PEL Moduli Varieties 317
In the definition of the functor [jJ'), the structure of the Tate module
V(p)(A) = H1(A,Q) ®IQI A(poo) is specified modulo K for a test object
(A, X, TJ(p») over a characteristic 0 field inside C However, this information
may not uniquely determine the D-Hermitian structure of Hl (A, Q) which
supplies us with the exact information of the algebraic group giving rise to
the Shimura variety; so, the variety Sh~) representing the functor [jJ') may
be a disjoint union of the reduction modulo p of characteristic 0 Shimura va-
rieties associated with different groups locally isomorphic to each other (but
not globally). We study this point in this section to relate Sh(p) to the char-
acteristic 0 fiber of the Shimura variety.
A (D, *)- Hermitian form (.,.) is a symmetric bilinear form on V with
values in Fo such that (bx, y) = (x, b*y) for all x, y E V. The given alternating
form (-,.) : V x V --+ Q can be uniquely lifted to an Fo-linear alternating
form C·)o: V x V --+ Fo so that (.,.) = TrFo/IQI((-,·)o). Pick (A,X,7J(P»)/k E
[f2(P) (k) for an extension field k/ E inside C. Similarly, the Riemann form
EA on Hl (A(r.c) , Q) with values in Q can be uniquely lifted to an Fo-linear
alternating form (., .)" so that TrFo/lQI( (., .».) = E A (-, .). Then the test object
(A, X, 7J(p») /k supplies us with the following information.
The D-module VA := H1(A(r.c),Q) with Fo-linear Riemann form C·)" :
VA X VA --+ Fo and an isomorphism TJ(p) : V ®IQI A(poo) ~ VA ®IQI A(poo)
sending C .) 0 to (-, .)" up to the constant in 0;' (p)"
The Shimura variety S h( GU, X) / E is defined for the group GU; so, our ques-
tion is if the above data are sufficient to determine GU (i.e., to verify (pol)
using information only from defining data of [f2 (p) ). There are three points to
check:
(1) (VIR, C ·)0) ~ (VA ®IQI JR;., (-, .),,), and for any two i-homomorphisms C Y
Coo are conjugates under G(JR;.);
(2) (Vp, C ·)0) ~ (VA ®IQI Qp, C·),,);
(3) the Hasse principle for the alternating form C ·)0.
First we prove (1) following [Ko], Section 2. Recall that 10 is the set of all real
embeddings of Fo (regarded as the set of all Archimedean places of Fo). For
any semi-simple F-algebra B and a B-module M, we put Ba = B ®Fo,a JR;.
and Ma = M ®Fo,a JR;.; so, BIR = B ®IQI JR;. = ITaEl o Ba and MIR = ITaEl o Ma.
~1R 7 Shimura Varieties
Each Dcr has positive involution induced by *, which we still denote by *. Now
a (D, *)-Hermitian form on Vcr is a symmetric lR-bilinear form with (bv, w) =
(v,b*w) for all v,w E Vcr' We start with
Lemma 7.3 Let (.,.) and (.,.)' be two positive-definite (D, *)-Hermitian
forms on Vcr. Then there exists c E C; such that (cv, cw) = (v, w)' for all
V,WEVcr .
Proof. The Hermitian form (-,.) (resp. (', .)') induces an identification ¢ : Vcr ~
Vcr:= HomlR(Vcr,lR) (resp. ¢': Vcr ~ Vcr). We regard Vcr as a Dcr-module by
f H fob*. Since ¢(bv)(w) = (bv, w) = (v, b*w) = ¢(b*w), we find Cl = ¢-lo¢'
is an element in AutD<7(Vcr) = C;. We find (C1V,W) = ¢(cv)(w) = ¢(¢-l 0
¢'(v))( w) = (v, w)'. Since the two forms are (D, *)-Hermitian, we find Cl = ct.
Consider the set C+ = {c E Ccrld = c and c > O}, where c > 0 {:? (cv,w) is
positive-definite. Any positive-definite symmetric inner products of a finite-
dimensional real vector space are equivalent over lR (Sylvester's theorem),
C+ is therefore isomorphic to a positive subcone of the space of positive-
definite symmetric matrices of EndlR(Vcr). In particular, log: C+ ~ 10g(C+)
and exp : log( C+) ~ C+ are well defined by the standard power series expan-
sions. Therefore, we have c = JC1 = exp(log(cd2)) in C+, which gives the
desired element satisfying (cv,cw) = (c 2 v,w) = (C1V,W) = (v,w)'. 0
Hq(Q, GU) -+ Hq(lR., GU) x TIe Hq(Qe, GU) for q = 1 (the Hasse principle
for Hq(Q, GU)).
For groups of types Band D, G I is not simply connected; so, often the
Hasse principle fails (cf. [GCH] III.4.7 and [Kn1]). Thus we assume one of the
following two conditions which guarantees the condition (sc).
(B1) The involution * on D is a positive involution of the second kind
(Case A);
(B2) The involution * is of the first kind, and D~ = D 0Q lR. ~ Md(F~) for
F~ = F 0Q lR. (Case C).
vanishing of Hl(Q, T F ) follows from Hilbert's Theorem 90. This shows that
H1(Q, ZGU) ~ Hl(Q, Td ~ Fox jNF/Fo(FX) for which the Hasse principle
holds (the norm theorem of class field theory for cyclic extensions). Thus the
Hasse principle holds for Hl(Q, GU) as desired.
Now assume that r is odd. Let Zl ~ /-lr be the center of G 1 . Then we
have an exact sequence of groups 1 -+ Zl(Q) -+ Z(Q) -+ ZGu(Q) -+ 1,
because over Q, GU ~ (GL(r) x Gm)Io. Since H2(Q, /-lr) is the kernel of
multiplication by r of the Brauer group of Q (Kummer's theory), the Hasse
principle holds for H 2(Q,/-lr) (class field theory; see [BNT] XIII.3). For
the same reason, Hasse's principle holds for H2(Q, TF)' Since Z = T F , by
Hilbert's Theorem 90, Hl(Q, Z) = O. Then by the short exact sequence,
o = HI(Q, Z) -+ Hl(Q, ZGU) -+ H2(Q, /-lr) -+ H2(Q, Z), we conclude the
Hasse principle for H1(Q, ZGU) and hence for H1(Q, GU). This shows (3) for
320 7 Shimura Varieties
GU. The Hasse principle sometimes fails for G of type A (but in a rather
harmless way; see Remark 7.4 and [Ko] Section 7).
Now we prove (2). Let Oc = {x E ClxL = L}. We need to prove that
if L~ is a self-dual OD,p-Iattice of Vp = V @IQI <Qlp under (-, ·)0, then we have
an element g E GU(<Qlp) such that gLp = L~. We can do this at each place
p of Fo above p. By (unr) , OD,p = TI'lJlp Md(0'lJ). This shows that Oc,p =
TI'lJlp Mr(O'lJ). Thus we may assume that d = 1 (replacing L by EL ~ og for
the *-invariant idempotent E = diag[l, 0, ... ,0] E OD,p). Then the existence
of g E U(<Qlp) c GU(<Qlp) as above follows from [EPEl, Lemma 5.9 (note here
any self-dual lattice of Vp is maximal in the sense of [EPEl, Chapter I). There
is another proof in [Kol, Section 7 reducing the result to a theorem of S. Lang
on the vanishing of HI(IFq, H) for a connected algebraic group H over a finite
field IF q. When the group G is of type D, G may not be connected; so, the
proof of (2) for such groups may also fail.
Since the proof in the case of groups of type C is easy, let us give more
details of Case C. In this case, Lp = 0; for an even integer r. Choose hE C so
that h'Lp = L~. Then the (x, y) f-+ (h'x, h'y)o induces a new skew Hermitian
form (-'·h on Lp; so, we need to find h E O~ such that (hx, hyh = (x, y)o.
Take Xl, YI E Lp with (Xl, YI)O a unit u in Op. Changing YI by U-IYI, we may
assume that (Xl, YI)O = 1. Then Lp = (OpXI + 0pYI) EEl (OpXI + OpYI).l. By
induction on rankLp, we find that (-, ·)0 is equivalent to (x, y) f-+ xJr / 2 t y on
0;. Similarly, (-'·h is equivalent to (x, y) f-+ xJr / 2 t y on 0;, and we find the
desired h E Oc with (hx, hYh = (x, y)o.
We first formulate Shimura varieties as in Deligne [D1] and [D2J, and in many
places, our exposition follows Milne's expository article [Mil, Chapter II. We
consider a reductive group G /iIJ. with G(IR)+ ICo isomorphic to a Hermitian
symmetric domain X+, We write the adjoint group of G as Gad = G I Z in
this section (which was also written as PG before). We can identify X+ with
a connected component of the set X of the G(IR)-conjugacy class of a fixed
homomorphism (of real algebraic groups) ho : § /IR y G /IR whose stabilizer
(under conjugation) is Co. Each maximal compact subgroup of G1(IR) can be
given as the centralizer of such a homomorphism. In other words, we have
the origin 0 E X+ and the corresponding homomorphism ho : § y G with
centralizer Co in G(IR). We write x E X when we consider x geometrically
as a point of the symmetric domain of G and write hx : § Y G /IR when we
consider x E X as a homomorphism.
We start with such a pair (G, X) and require (G, X) to satisfy the following
four axioms.
(D1) For each x E X the Hodge structure on Lie(G) defined by the ho-
momorphism hx is of type {(-1,1),(O,O),(1,-1)}. In other words,
Ad(hx(z)) (z E (Cx = §(IR)) has only eigenvalues z-lz, 1, and zZ-l.
(D2) For each x E X, 9 f-t hx(i)ghx(i)-l is a Cartan involution of the adjoint
group Gj~ = PG /IR; so, the centralizer of hx in G 1(IR) is a maximal
compact subgroup. In other words, the involution gives the twisted
complex conjugation (which we referred to in the introduction) making
the centralizer of hx the subgroup of real points of a real algebraic
group.
(D3) The adjoint group Gj& has no factor defined over Q whose real points
form a compact group (this condition excludes a case we treated in
Section 4.3: G = G D = ResF/ilJ.Dx for a definite quaternion algebra D
over a totally real field F).
322 7 Shimura Varieties
on V" = V ®F,,, JR: (d = dimF V). We suppose that dimF V 2' 3. We write
Is for the places of F where q has signature (d - 2,2) or (2, d - 2). We
put Ir = I - Is. Then G = ResF/rQGpin(q) for the special Clifford group
Gpin(q)/F for q. By definition,
(7.8)
[Mi] pages 323-324). We write [x, g] for the complex point of the Shimura
variety represented by the pair x E X and g E G(A.(oo)). When G is given by
GU of type A or C in (7.1), GU(Q)/GU(QJ+ acts on 7ro(X) transitively; so,
we can rewrite (7.8) as
b- 1 a(x), (agg')'Y] = [a,,,(-I(x), a' (gg')'Y] = b- 1 (x), (gg')'Y] = [x, g](g' ~ "()
and [x,g]b- 1 ~ adb)) = b- 1 (x),(n- 1 ),] = b- 1 (x),"(-lg]
for a, "( E G(Q).
The condition. (rc2) for G = ReSFjIQlGL(2) (for a totally real field F = Fa)
is the local reciprocity law in Section 4.2.3, but in this general case, it is a
requirement for Sh(G, X)jE to be the Shimura variety of (G, X). When (G, X)
comes from the algebra DjF and D-module V as in Section 7.1, special points
are those points on Sh(G, X) carrying abelian varieties of CM-type, and (rc2)
follows from the main theorem of complex multiplication (Theorem 4.19) as
we have proven (rc2) (i.e., Corollary 4.20) for the Hilbert modular variety.
By (7.8), we find [x, ')'g] = [,,(-lX, g] hE G(Q)). For,), E Tx(Q), ')'(x) = x
if x is special with respect to Tx. Thus [x,')'rx(7)a] = ["(-1(x),rx(7)a] =
[x,rx(7)a] is well defined without ambiguity. Since the set of special points is
Zariski dense (actually, it is dense also in complex topology; see [D1] 5.1). In
all three cases (A, B, C), the existence of the canonical models is known (see
[Sh5], lSi], [D1] 4.21,6.4, and [D2] 2.7.20). More generally, if the pair (G, X)
has a symplectic embedding into (GSp(2N), nN U SJN) as in [D2] 2.3.2-3,
the complex proanalytic space Me(G, X) admits a unique canonical model
326 7 Shimura Varieties
Sh(G, X)/E(G,X) (see [D2] 2.7.10). This type of Shimura variety is called a
Shimura variety of abelian type. See [Mi] II for a discussion covering a more
general type of Shimura variety.
and write 7r07r(H) for the group of connected components of 7r(H). We put
7f07r(H)= 7r07r(H)/7ro(H(lR)+), where H(lR)+ is the inverse image of the iden-
tity component Had(lR)+ of Had(lR). We consider the following exact sequence
(7.9)
for a torus ZG (called the cocenter of G) defined over Q. Recall /Lx = hx IlG m xl:
(Gm/IC -+ G /IC. For each x EX, we look at the composite /Lx = X 0 /Lx : (Gm -+
ZG. Since {/Lx}xEX belongs to one conjugacy class X, /Lx is independent of
x and is defined over E = E(G, X) by the definition of the reflex field E. We
G Res(!J.x)
can thus define A: ResE/Q(Gm -+ Z by the composite of ResE/Q(Gm )
ResE/QZG and ResE/QZG ~ ZG. We define a homomorphism
following [MiS], page 929. Here we identify 7ro(Er / EX) with Gal(Eab / E)
sending a prime element w E F x c Fpx at a prime ideal p to the geo-
metric Frobenius at p (thus compatible with the Artin reciprocity law; see
Theorem 2.12). This homomorphism was defined by Deligne in [D2] (2.6.2.1)
as 7roNE /QqM (using an intricate but explicit argument in terms of tor-
sors without assuming (SC), [D2] 2.4-6). By (SC), the strong approxima-
tion theorem affirms (see [D1] Theorem 2.4) that 7r07r(G) ~ 7r07r(ZG) =
7ro(ZG(A)/ZG(Q)) via X. The.homomorphism A gives rise to a homomor-
phism of 7r07r(ResE/Q(Gm) = 7ro(Er / EX) into 7r07r(ZG) = 7ro(ZG (A)/ZG (Q)).
Thus we have
(4) For each special point z E X+ and f E it, f(z) E E~b if f is defined at
z, where X+ is the connected component of X containing o. Moreover,
if f is defined at special z E X+, then T(rz(p))(f) is defined at z and
p(f(z)) = T(rz(p))(f(z));
(5) The map T defines a topological isomorphism of Q(G,X)/Z(Q)G(ffi.)+
onto an open subgroup of Aut (it/ E). For every K E Z, we have T(K) =
Aut (it/kK (Sh K )). Here Aut(it/ E) is equipped with the Krull topology as
described in Section 2.3.
This theorem has been proven by Shimura in [Sh5], II in Case C. After some
work by students of Shimura, the general case in the theorem is covered by
Milne-Shih [MiS], Theorem 4.13. Our formulation here follows [MiS] and is dif-
ferent from Shimura's formulation in which he uses T(X- 1 ) for T in the present
notation to make Galois action as a right action. We followed Shimura's for-
mulation in Theorem 2.43, and this point is obvious from T(a) (f) = f 0 a in
Theorem 2.43 which is the inverse of T(a) in (3) above. Note here that the
group T(9( G, X)) is not just an open subgroup of the field automorphisms of
it but also the subgroup of scheme automorphisms of Sh( G, X) over E.
The cokernel of T also can be described explicitly as a subgroup of
Aut(F/Q) in Cases A, B, and C, and in these cases, one can extend the
group Q( G, X) to describe explicitly the full automorphism group Aut (it/ E)
(see [Sh5] II, [MiS] 4.7 and 4.14, and [H03b]). In the proof of [MiS], one of
the key ingredients of the description of the full group is the determination
of Aut(Cit/CC) by Miyake [Mt], and in [H03b], the theorem of Faltings and
Zarhin (Theorem 4.13) is used in place of Miyake's result.
8
We first describe a formal theory of false automorphic forms and find a set
of conditions for nearly ordinary p-adic cusp forms to have vertical control.
After this, we describe deformation theory by Serre-Tate of abelian varieties
in order to prove the divisibility by the character value fJ(O in (5.40) of the
Hecke operator associated with an element ~ of the expanding semi-group
D in Section 5.1.2. By this, we have well-defined p-integral Hecke operators
Tp(O on coherent cohomology groups and therefore well-defined nearly p-
ordinary automorphic forms. Verifying the set of conditions on sections of
automorphic vector bundles of a given (projective) Shimura variety of PEL
type, we prove the vertical control theorem for automorphic forms on the
Shimura varieties. We also state a similar result for quasi-projective Shimura
varieties of quasi-split unitary groups without a detailed proof. At the end, we
prove the irreducibility of the Igusa tower over the ordinary locus of symplectic
and unitary Shimura varieties, reducing the problem to the Siegel modular
case. By means of the Serre-Tate deformation coordinates, even when the
Shimura variety does not have a cusp, we are able to define an analogue of
the p-adic valuation v (in (4.36)) of its arithmetic automorphic function field;
so, the irreducibility can be proven even for projective symplectic and unitary
Shimura varieties of PEL type.
other hand, sections in HO(M, 1,Jd."') are called true or "classical" automorphic
forms. The theory is a formal generalization of the work of Deligne-Katz in
the elliptic modular case (see [Kl] Appendix III).
The main purpose of this section is threefold:
1. Approximate p-adic automorphic forms by finite sums of classical forms;
2. Define the p-ordinary projectors e and eo;
3. Find a set of (axiomatic) conditions that guarantees the VCT.
Since 7/J : (Z/pnz)g ~ P n is determined by (7/J( el), ... ,7/J( en)) for the standard
base (el, ... ,en ) of (Z/pnz)g, we find that Isoms",,((Z/pnZ)g,Pn ) is repre-
sented by the open subs cherne of Pi( given by
Because Pi(/Sm is etale, Tm,n/s"" is etale. Since each geometric fiber of Tm,n
over Sm is isomorphic to GLg(Z/prZ) everywhere, it is faithfully flat and
finite. Therefore Tm,n is affine over Sm. We define Vm,n = HO(Tm,n, OT"",n)'
The group GLg(Z/pnZ) acts on Tm,n by 7/J f-t x7/J for x E GLg(Z/pnZ)
without fixed points, and we have T m,n/ rn,n' ~ T m,n' for all n' :S n, where
8.1 True and False Automorphic Forms 331
Then we have a tower Vm,o C Vm,l C ... C Vm,n with Vm,o = HO(Sm, Os,,J.
We put Vm,oo = Un Vm,n and Tm,oo = ~nTm,n. Similar to (4.13), we put
(8.2)
(8.3)
For a given ring A or a sheaf of rings A over a scheme, we look at the flag
variety F/A = FlagB/A of all maximal flags in Ag defined in Section 6.l.3. We
write BeG L(g) for the upper triangular Borel subgroup. Let N be its unipo-
tent radical, and put T = BIN for the torus. Note that T ~ BIN ~ t Bit N
for the Borel subgroup t B opposite to B and its unipotent radical tN. Then
F ~ GL(g)IB. We define H/A = FlagN/A = GL(g)IN. Write 1 for the origin
of H represented by the coset N. Each schematic representation of T is com-
pletely reducible (see the following description of schematic representations
and [RAG] I.2 and [GME] 1.6.5), and we have HO(H, OH.) = ffil<EX(T) RA[Ii:]
for the space R A [Ii:] of weight Ii:. Here
The middle equality is given by the Frobenius reciprocity (e.g., [MFG] 2.1.6).
The GL(g)-module RA[I1:]* is generated by a B-stable line of weight 11:, and
any schematic GL(g)-module generated by a B-stable line of weight 11: is a
surjective homomorphic image of RA[I1:]*.
By this universality, taking M = A(O), we know that the coinvariant space
RA[I1:]N := Ho(N, RA[I1:]) is a free A-module of rank 1 on which T acts by -11:,
and there is a unique N-invariant linear form ecan : RA[I1:] ---+ A (up to an
A-unit multiple), which generates (RA[I1:]*)N. We can normalize ecan so that
it is the evaluation of ¢ E RA[I1:] at the origin 1 E GL(g)jN. Then we have a
tautological embedding RA[I1:] '---+ r(GL(g)jN, OCL(g)/N) given by
HO(I1:) = Ind?;(g) A(-I1:) = {J: GL(g);tN ---+ A1If(xt) = r"'f(x) 'Vt E T}.
Let ~m be the locally free sheaf P Q9zp Os", = Pm Q9zp Os", on Sm of rank g.
We define a coherent sheaf ~m,n on Tm,n by (7r:n,nPm) Q9z OT""n (~m = ~m,O
for Tm,o = Sm). On Tm,n with n ::::: m, we have the universal isomorphism
Wean = lean Q9 id : ~m J
n ~ O~m,n
HO(Sm'~':r') =
{J E HO(Tm,m, R v""", [K:])lf(X, h'lj;) = p(h)f(X, 'Ij;) for Vh E GLg(Zp)} .
334 8 Ordinary p-Adic Automorphic Forms
a value 'P(f)(X, 7jJ) E HO(X, Rox [K:]) such that 'P(f)(X, h'lj;) = p(h)'P(f) (X, 7jJ)
for all h E GLg(Zjpmz) and 'P(f)(Y, ¢*7jJ) = ¢*('P(f)(X,7jJ)) for a mor-
phism Y ~ X of Sm-schemes. Similarly, 'P E Vm,n is a functorial rule as-
signing (X, 7jJ) a value 'P(X,7jJn : Pn ~ (Zjpnz)g) E HO(X, Ox) such that
'P(Y,¢*7jJn) = ¢*('P(X,7jJn)) for a morphism Y ~ X of Sm-schemes. Thus we
have a natural map of HO(Tm,m, RTm,m [K:]) into Vm,m associated with each
linear form £ E RVm,m [K:]*. The map associates f E HO(Tm,m, RTm,m [K:]) with
a rule: (X,7jJ) H £('P(f) (X, 7jJ)), which is a matrix coefficient of 'P(f)(X, 7jJ).
We let h E GLg(Zp) act on test objects by (X,7jJ) H (X, h7jJ). In this
way, we identify GLg(Zp) with a subgroup of Aut(Tm,oo/s,J. For the Borel
subgroup B c GL(g), we put T!,n for the quotient Tm,nj B(Zjpnz). Thus
V! n = HO(T! n' OTB ) is made of a functorial rule (X,7jJn) H 'P(X, 'lj;n) E
HO'(X, Ox) such that~(X,b7jJ) = 'P(X,'lj;) for all7jJ and bE B(Zp). We define
V:'n similarly and T;;;,n for the unipotent subgroup NcB. Thus we have
given by L J = 2:;=1 (ZjpnZ)e,. Then 7jJ~ln gives a (full) filtration fil = fil"'n
of Pn , and all full filtrations Pn = P~g) :) p~g-l) :) ... :) pta) = {O} of
Pn are given in this way. Since the stabilizer of In is B(Zjpnz), we may re-
gard 'P E V!,n as a functorial rule assigning a value 'P(X, filn) E HO(X, Ox)
to a test object (X, fil n ). To describe V:'n in this way, we need to bring
in an isomorphism of graded modules ¢n : gr(fil n ) ~ EB;=l (Zjpnz) induc-
ing ¢~) : p(J) j p(J-1) ~ (Zjpnz). In other words, T;;;,n classifies triples
8.1 True and False Automorphic Forms 335
(X, fUn, ¢n). Since we pulled back the filtration In by 'l/Jn, h E GLg(71p)
acts on (X, fil,pn' ¢n) by fil n H 'I/J-1h-1'I/Jfil n (h'I/Jn)*ln and ¢n H
'I/J-1h-1'I/J¢n'
We consider Rv""", [Ii] inside V:'m[li], which is the homomorphic image of
HO(8m,!!:!..':;,) under f H eean 0 <p(f). Thus we have a natural map
(8.4)
where V:'m[fi;] is the fi;-eigenspace of the right action of T. The above map is
injective if m = 00. Define
R'm = (8.5)
I<EX+ (T),I<»O
We assume that the formal sheaf !!:!..= is algebraizable; so, we have a locally
free sheaf !!:!..;s of finite rank such that!!:!.. 0w Wm = !!:!..m for all m. The
sheaf!!:!.. is determined uniquely if it exists. From !!:!.., we can create !!:!..;s as
7r*(')FlagN(~)[fi;] for 7r: FlagN --t 8 (see Section 6.1.3). The global HO(8,!!:!..) in-
jects into ~mHO(8m'!!:!..m)' We define a graded module R' = EBI< HO(8,!!:!..I<)
which is embedded into R'oo = ~mR~. We call an element of HO(8,!!:!..;s)
a false automorphic form of weight fi;, and a true (or classical) automorphic
form is a section in HO (M,!!:!..;M) for a compactification M ::) 8 of 8 we specify
later.
We suppose now that for all fi; » 0, the short exact sequence
(Hpl)
This condition is satisfied when 8 m is affine (see [ALG] 111.3.5). As we show
in Theorem 8.3, (Hpl) implies
(Hp2)
From (Hpl), we have
J3(m)(L f,J
1<»0
= {(X/TTn'Tn''lj;) r-+ Llean(W~an(JI«X''lj;)))}
I<
for fl< E HO(Sm'~':n). Here, the image of J3(m) actually falls in V:'n because
lean 0 PI«u) = lean for all u E N(Zp), and w~an(Jl<) E Rk,Tn[K]. By construc-
tion, J3(n) mod pm = J3(m) for all n > m. Taking the projective limit, we
have 13(00) : R'oc, -t VN = \i!!!m
V:'oo'
Since Sm is flat over Wm = Wjpmw and T!/',n = Tm,njN(Zjpnz) is etale
over Sm, V:'oo is flat over W m . Therefore, VN is a W-flat GLg(Zp)-module.
Remark 8.1 Flatness of HO(M,~I<) is subtle. Let S = 9J1(c, r(N)) as [iJ
defined in (r(N)) of Section 4.1.2 for a lift E of the Hasse invariant. Assume
that p is prime to Nand unramified in F jr[}.. We take Pn to be the maximal
etale quotient of X[pn] for the universal AVRM f : X -t Sm. By Corollary 4.2,
we have that ~ ~ f*[2x/s. From this, the locally free sheaf ~ extends to the
locally free sheaf ~/M = f*[2.X/M on the smooth toroidal compactification M
canonically because the universal abelian scheme extends to a semi-abelian
scheme X over the toroidal compactification as described in Sections 4.1.4
and 6.3.6. As we have seen in Remark 4.8, HO(M/wTn'~7M) is not Wm-free
for 1 < m < 00 if k is trivial on T(lFp), but nontrivial on T(Zjpmz). However,
flatness of ~7M implies W-freeness of HO(M/w'~7M)'
Since B normalizes N, we can think of the action of T = B j N on V Nand
the K-eigenspace VN[K] of VN. By definition, 13 = 13(00) induces
Proposition 8.2 Suppose (Hpl) for S. The above map 131< is an injection.
Proof. Since Tm,n is faithfully flat and etale over Sm, we may make a
base change T::',n/S Tn to T::',n/S Tn xSTn Tm,m, and hence we may suppose
that P is constant. Then V: 00 is made of locally constant functions on
GLg(Zp)jN(Zp) with values i~ Vm,o. By taking the limit, V N is the space
C(GLg(Zp)jN(Zp), Voo,o) of continuous functions on GLg(Zp)jN(Zp) with
values in V00,0 = ~ m Vm,O. Then H O(S, ~I<) is inside the limit of global sec-
tions of \i!!!mRVTn,o[K], which injects into C(GLg(Zp)jN(Zp), Voo,O) [K]. This
shows the assertion. D
0TN
1.n
= OS1 0w W[(GL(g)jN)(71jp n71)] = OS1 [(GL(g)jN) (71jp n71)].
This shows vtoo = .cC((GL(g)jN)(71p), VI,o), where .cC((GL(g)jN) (71 p), VI,o)
is the space of locally constant functions on the p-adic space (GL(g)jN)(71p)
with values in HO(SI, OS1) = VI,o. Writing V!,o as a union of W-free modules
X of finite rank, we have
wEe;
By further raising the power, that is, replacing E by EP=, we may assume
that E mod p = at. Then by definition,
We now show that ,B(R[i]) n VN is dense in VN. Pick LI< fl< E pmvN
for fl< E HO(8,~I<). We need to approximate f = fl< modulo pm+lv N by an
element in HO(M,~I< 0 det(~)k). This section f E HO(8,~I<) can be written
as f == gdE£ mod pm+l for g£ E HO(M,~I< 0 det(~;d). Then for k > e,
we have f == geEk-£ j Ek mod pm+!. We may assume that k = pmt(p - 1).
Then as a function of (X, 'I/J), Ek mod pm+l is a constant. Thus f == geEk-£
mod pm+l Roo. This shows the density of ,B(R [i]) V N in VN.n
Corollary 8.4 Suppose the following conditions for 8 in addition to (Hp1).
(1) We have ScM for a proper fiat scheme M/ w such that 8 m C Mm =
M 0w W m is Zariski dense for all m;
(2) ~ extends to a locally free vector bundle on M of rank g;
(3) There exist an integer t > 0 and a section E E HO(M, det(~)t(p-l)) such
that E mod w is the constant section 1 generating det(~l)t(p-l) ~ OSI;
(4) M - 8 is the zero locus of the section E.
Let diag[X l , ... , X J ] denote the diagonal matrix with diagonal blocks Xl to
X J from the top. We recall the subsemi-group of GL(g) given by
where fil n : Pn / X = P~g) :::) p~g-l) :::) ... :::) p~O) = {O} with 7Ljpn 7L 't;,
p~J) j p~J-l) for j = 1, ... , g. If P is constant, the space classifying the
above test objects over 8 m is given by T!/",n = Tm,njN(7Ljp n7L). Simi-
larly, the classifying space of couples (X, filn) over 8 m is given by T! n =
Tm,nj B(7Ljp n7L). On test objects over Tm,n, we have a natural actio'n of
h E GLn(7Lp) given by (X, 'lj;) H (X, h'lj;). Writing fil n = 'lj;-lln, we then
see that 'lj;-lh-1l n = 'lj;-lln = h· fil n . Thus the Galois action on filtra-
tions is given by h· fil = ('lj;-lh-l'lj;)P~J) and h· ¢ = ('lj;-lh-1'lj;)¢, where
342 8 Ordinary p-Adic Automorphic Forms
'lj; : Pn ~ (Zjpnz)g such that 'lj;* (In, id) = (filn' ¢n) for the standard identi-
fication id : gr(ln) ~ (Zjpnz)g. Thus these test objects are always invariant
under N(Zp). The new test objects (X, filn, ¢n) are useful in defining an
isogeny action of IS E L1 and in constructing the idempotent eGL. We assume
(d1) IS induces an isomorphism IS* filnjlj.x ~ fil n/ x compatible with the
action of the semi-group on fil n (this holds if IS mod p is radiciel),
where the action of IS E L1 on fil n is the multiplication by IS up to
scalars (as we specify later);
(d2) h· X = X if hE GLn(Zp).
Let diag[xl, ... , x]] denote the diagonal matrix whose diagonal blocks are
given by square matrices Xl to X] from the top. We put a] = diag[l g _],p1)].
We consider the double coset ring R(h,n, L1~) (n = 1,2, ... ,00 and? = B, N)
made of Z-linear combinations of double cosets h,nlSh,n for IS E L1~.
As seen in Lemma 5.2 or its proof, once a decomposition N(Zp)a]N(Zp) =
U) N(Zp)lS) is chosen, h,na)h,n for any nand? is decomposed in the same
way, h,na)h,n = U) h,nlS). In other words, we have for a = I1;:i a),
By this fact, R(h,n, L1~) ~ Z[h, ... , tg] (a polynomial ring). See Lemma 5.2
for a proof, though the semi-group D here is a proper subsemi-group of
the expanding semi-group there, strictly speaking. For IS E L1f, the action
p(lS- l ) is well-defined on RA[/'t:] for any p-adic ring A, because p(lS- l )¢(y) =
¢(lSy) for y E GL(g)jN. Decompose h,nlSh,n = U) h,nlS) and regarding
f E HO(T:;',n,f;d.K) as a function of test objects X/ T ", we define
,00 ,
where ISX = (IS· X, IS(IS* filn)). Strictly speaking, the sum above is actually
heuristic, because if the action of IS on S is wildly ramified, that is, radiciel
(i.e" purely inseparable in characteristic p), we need to replace the sum by
the trace as already described (4.18) in Section 4.1.10 (for Hilbert modular
forms). We clarify this point in Section 8.3 more carefully; so, for the moment,
we content ourselves with this heuristic description.
Although we have not yet specified the action of IS E L1~ on IS* filn' if it
is done, then the operator is well defined independently of the choice of IS)
because for U E h,n we have
Furthermore, by (8.9),
Y(F)
n jT -
- {(P(i) )
njT
IFn -_ peg)
n ,
pet) p(,-l) p~,) rv 7L (
n ::J n
. )}
, p~'-l) = pn7L 1::; z ::; 9 ,
Y(Pn)jT = {(P~;T' cp,)I(p~')) E y(Pn)jT, CPt : P~7L ~ ;'~)1) (1::; i::; g)}.
After a finite etale extension to T / Sm, the spaces y(Pn ) and Y(Pn ) get isomor-
phic to y((7L/pn7L)g) xSrn T and Y((7L/pn7L)g) xSrn T, respectively. Writing the
standard base of Las e1, .. " eg, we define I = 0:=;=1
Ae" CPt = id) E Y(Ag),
which we call the origin. We may let GLg(A) act on Y(Ag) and y(Ag) by
x((L,), ¢,) = (xL" x 0 CPt). Then GLg(A)/N(A) ~ Y(Ag) by xN(A) H xl.
Now we assume A to be a p-adic ring. We then define
Similarly, we define YnjT(Pnf) for n ::; n' ::; 00. We note that Yn(7Lg) ~
IN,n/N(7Lp) via xl +-+ x and similarly Yn(7Lg) ~ IB,n/N(7Lp). So we have the
conjugate action of L1 on these spaces introduced in Section 5.1.3.
We now write down explicitly the conjugate action of the semi-group L1
on Yn(Ag). Since y(7Lg) = y(Qg) (because y = FlagB is projective), the group
GLg(Qp) acts naturally on y(7Lg). This action is described as follows. Take
x E GLg(Qp). Then x(L,) = (xL, Q9 Qp n L(7Lp)) E y(7Lg). We write x(L,) =
(x· L,), that is, X· Li = xL, Q9 Qp n L(7Lp). We claim
Lemma 8.6 Let d = diag[pe1, ... ,pe g ] E DCL(g). For (L"cp,) E Yn(7Lg),
[d]i : L,/ L,-l -+ b· Ldb· L,-l given by (e mod L,-l) H (p-e'de mod L,-d
for eEL, is a surjective isomorphism.
Proof. The 7Lp-module L, is generated by L,-l and a vector e whose first i - 1
entries vanish. Then de is divisible by pe, because its bottom 9 - i entries are
divisible by pe'+l because of eg 2: ... 2: e,+l 2: e,. From this, the linear
map (d), : (QpL,/QpL,_l) -+ d(QpL,)/d(QpL,-d induced by d is divisible
by pe, on L,/ L,-l C (QpL,/QpL,-d. By definition, we have d· L, = L" and
p-e'(d), : LdL'-l -+ L,/L,_l = d· L,/d· L,-l is a surjective isomorphism.
o
344 8 Ordinary p-Adic Automorphic Forms
and define the idempotent eCL by eCL = lim n --+ oo t(p)n! whenever it is well-
defined. As we have seen (5.10), K:(a)t(p) = t(p) on HO(S,~K), because on ~K
we used the action of L1 coming from schematic induction.
forms (sections over M). Later we will verify the requirement for automorphic
vector bundles on Shimura varieties of PEL type.
Since in this general situation Sm is not supposed to classify anything, we
cannot define Hecke operators acting on HO(Sm, ~"') in this generality. Any-
way, we suppose a Hecke operator T(p) acting on HO(Sm,~"') and HO(Mm , ~"')
such that T(p) == t(p) on HO(SI'~"') if K , » O. We define eo = limn-too T(p)n!.
Further suppose that there exist two projectors ee acting on V:'oo and ee on
HO(Mm'~"') for K, » 0 (depending on the reductive group G) such that
eeeeL = ee, eeeo = ee, and ee == ee mod p; that is, they are equal to each
other on ecHO(SI'~"')' In addition to the above conditions, writing K for the
field of fractions of W, we suppose the following two conditions,
(C) ec(E· f) = E· (ecf) for I E HO(SI'~"') for E in (E) of Section 8.1.3,
(F) dimK eeHO(M/K'~'" Q9 det(~)'K) is bounded independently of k.
Let 71,72' ... be a sequence of linearly independent elements in ecHO(SI, ~"')
over WI' Since HO(S/w,~'lw) Q9w WI = HO(SI'~"') (Hpl), we can lift 7t
to Ii E HO (S /w, ~'lw) so that 7t = (It mod p). Then for any given in-
teger L > 0, we can find a sufficiently large integer m such that Em It E
HO (M/w, ~'lw ). Since multiplication by E is an isomorphism on SI (by def-
inition of SI = Ml [i]), by (C) and ec == ee mod p, the mod p sections
{(ee(Em Ii) mod P)}.=I, ... ,L are linearly independent over WI by Nakayama's
lemma; so, {ee(EmIt)}.=I, ... ,L are linearly independent over W. This implies
Theorem 8.7 (Drinfeld) Let G and H be abelian fppf -sheaves over B-LR
and I be as above, and suppose that JV+l = O. Let Go and Ho be the restriction
of G and H to B / I -LR. Suppose
(i) G is p-divisible;
(ii) H is formal;
(iii) H(R) -+ H(R/J) is surjective for any nilpotent idealJ (i.e., H is formally
smooth).
Then we have the following four assertions:
(1) Hom13_Gp(G, H) and Hom13/J_Gp(G o, Ho) is p-torsion-free, where the sym-
bol ''Homx_Gp'' stands for the homomorphisms of abelian fppf-sheaves
over X-LR;
(2) The natural map, the so-called "reduction mod I" from Hom13_Gp( G, H)
into Hom13/J_Gp(G o,Ho) is injective;
(3) For fo E Hom13/J_Gp(G o, H o), we have a unique P E Hom13_Gp(G, H) such
that P mod I = NV fo. We write jJv f for P in Hom13_Gp(G, H) ®z Q;
(4) In order that f = t;J E Hom13_Gp (G, H), it is necessary and sufficient
that jJv f kills G[NV].
Proof. The first assertion follows from p-divisibility, because if pf(x) = 0 for
all x, taking y with py = x, we find f(x) = pf(y) = 0 and hence f = O.
We have an exact sequence 0 -+ HJ -+ H -+ Ho -+ 0; so, we have another
exact sequence
Theorem 8.8 (Serre-Tate) The above functor A/B --+ Def(B, Bo) is an
equivalence of categories.
Proof. By Drinfeld's theorem applied to A[POO] and A (both abelian fppf-
sheaf), the functor is fully faithful.
It is known that we can lift Ao to an abelian scheme Y over B. This follows
from the deformation theory of Grothendieck (e.g., [CBT] 2.8.1). Here is a
heuristic argument to find Y/ B under the assumption that Bo is a finite field.
In this case, by a theorem of Tate (see [T1], [D], and [ABV] Section 22), Ao
has complex multiplication by a semi-simple commutative algebra M (with
2 . dim Ao = dimlQ M) whose simple components are all CM fields with its
complex conjugation induced by a Rosati involution. By the theory of abelian
varieties with complex multiplication, Ao can be lifted to an abelian scheme
Y over B with complex multiplication (this is a special case of Theorem 7.1
taking D = V = M). Thus we have an isomorphism a6P ) : Yo[POO] --+ Ao[poo],
and we have a unique lifting (by Drinfeld's theorem) f = NVa(p) : Y[POO] --+ 9
of NVa6P). Since the special fiber is an isogeny having inverse (afr))-l, f is
an isogeny, whose (quasi) inverse is the lift of N V (a6P))-1. Thus Ker(f) is a
finite flat group subscheme of Y. The geometric quotient of Y by a finite flat
group subscheme exists (see [ABV] Section 12) and is an abelian scheme over
B. Then dividing Y by Ker(f), we get the desired A/B E A/B. 0
Suppose Bo is a field of characteristic p. As described in [CBT], IV.2.5 and
V.lo4, we can attach a crystal (]]])B(90) ::) Fl) to the data (Ao, g, c:) formed by
a B-free module ]]])B(90) and a submodule Fl c ]]])B(90) with B-free quotient
]]])B(90)/F 1 such that ]]])B(90) @B Bo = ]]])Bo(90) is the Dieudonne module of
go and Fl @B Bo gives Lie(9o) c ]]])Bo(90)' Then we have
End(A/B) = {a E EndB(Ao/Bo)la*(Fl) C Fl},
regarding End(A/B) C End(Ao/Bo) by Drinfeld's theorem. By this fact com-
bined with a version of the above theorem ([CBT] V.lo1O), one can verify
(using (det) and D-linearity of the polarization Ao) that (A o,Ao,iO,1](p)) E
[(p)(Bo) for the functor [(p) in Theorem 7.1 can be lifted to (A, A, i, 1](p)) E
[(p)(B) (as shown in [LaR] 6.2 and [ZiJ).
8.2 Deformation Theory of Serre and Tate 349
(8.16)
for n ;::: no, which can be rewritten as the following exact sequence:
0-+ Hom(TAo[pOOjet,Gm)[pnj-+ A[pnj ~ Ao[pnjet -+ O. (8.17)
Here we call f : (A, AA, LA) --+ (Y, Ay, Ly) an isomorphism if f : (A, LA) ~
(Y, Ly) and tf 0 Ay 0 f = AA. Note that by the Drinfeld theorem, End(A/ R)
is torsion-free and, hence, End(A/R) Y EndlQl(A/R) = End(A/R) ®z 1Ql. We
write 00* = AOI 0 h 0 Ao for a E End(Ao) ®z 1Ql. Since End(A/R) C End(Ao)
again by Drinfeld's theorem, the involution keeps EndlQl(A/R) stable (because
on EndlQl(A/R)' it is given by 00* = A-Iota 0 A). The involution a --+ 00* is
known to be positive (see [ABV] Section 21). The polarization Ao induces an
isomorphism Ao : A [p=] et ~ tA [p=] et. We identify TAo [p=] et and TAo [p=] et
by Ao. Then the involution a f-t h is replaced by the Rosati involution *.
Then it is clear from the previous theorem that
(8.18)
whose right-hand side is written as Ti'(A). From (det), in the same manner
as in the proof of Proposition 4.1, we see
c;. Lie(A/wo) ~ c;. J;,°(A) @zp Wo as W o @z O-modules, (8.19)
m = rankop c;'Tp°(A) and n = rank o l' c;·Ti' (A), where Tp°(A) = ~JA[pJJO(QI).
The existence of p-ordinary A does not impose any restriction on the signature
(m, n). 0
8.2 Deformation Theory of Serre and Tate 353
Remark 8.11 The converse of Lemma 8.10 should be true. Here are some
examples of how to create an ordinary couple (A,).)/w on Sh(p)(G,X) for a
fairly general class of data (D, V, (.,.)) as in Section 7.1.1.
1. Suppose that (p) = pp with P i= P in F, V = F r (row vector space) and
(x, y) = TrF/'Q(xoHty) for a diagonal matrix H E GLr(Q) and 0 E F
with J = -0. We further suppose that 0 has a positive imaginary part
under the identity embedding id of F into C. We take an ordinary ellip-
tic curve E with complex multiplication by 0 over a discrete valuation
ring W with finite residue field IF. We may assume that L = L + EEl L-
so that H is positive definite on L + and negative definite on L -. Define
A = (L+ Q90,id E) x (L- Q9o,c E), where writing E(C) = Cjioo(O), we
regard E as an O-module by sending a E 0 to multiplication by ioo(a)
on Cjioo(O), and L- Q90,c E is the tensor product with L- twisting the
action of 0 on E by c (i.e., a E 0 is sent to multiplication by ioo (c( a)) on
E(C) = Cjioo(O)). Then Tp°(A) = (TO(E) Q90p L+) EEl CP(E) Q90p L-).
Write H+ (resp. H-) for the restriction of H to L+ and L -. We have
H = diag[H+,H-J. Put IHI = diag[H+, -H-J (a positive majorant of H
in the sense of Siegel). If we identify L = (L+ Q90,id 0) EEl (L- Q90,c 0),
(£ Q9 a,l' Q9 a') := TrF/'Q((oaa'C) . (£IHI· tr)) (a, a' E 0 and £,£' E L)
coincides with (-,.) on the original L. We take a polarization on E giv-
ing an alternating form (a, a') = TrF/'Q(oaa'C). Then the tensor prod-
uct of the polarization on E and IHI on L gives the desired polariza-
tion ). which induces the given alternating form (-,.) = TrF/'Q(oH) on
T(A) = (z+Q90,id0) EEl (Z-Q90,cO) = Z. The triple (A,).,7](p))/w,
choosing a level structure 7](p) extending W suitably, gives a desired ordi-
nary couple on Sh(p) (GU(m, n), X)(W) for the signature (m, n) of H. We
can generalize this construction to the case where D is a division algebra
and V = Dr in an obvious way.
2. Suppose that D is a division algebra, V = D m EEl Dm (row vectors), and
((XI,X2), (YI,Y2)) = TrD/'Q(XltY2 - X2tyi), and L = L' EEl L' for L' = 0D
under (-, .). Define another pairing ( , ) : L' x L' ---+ Q by (x, y) =
TrD/'Q(xty*). We take an elliptic curve E;w with complex multiplica-
tion by the integer ring 0' of another imaginary quadratic field F' where
p splits (F' can be F if P splits in F). We define A = E' Q9z L'. Take the
self-dual polarization on E' and identify HI (E', Z) with Z2 so that the in-
tersection product is given by ((Xl, X2), (YI, Y2)) = XIY2-X2YI on Z2. Then
the tensor product (, )' Q9 (, ) on L = L' EEl L' = Z2 Q9z L' = HI(A,Z)
gives the original pairing (., .). We take the corresponding polarization
). on A = E' Q9z L'. Then (A,).) satisfies the desired requirements for
U(m,m).
Hereafter we assume that
(ord) sMp)(G, X) (lF p) has a point carrying an ordinary abelian scheme.
354 8 Ordinary p-Adic Automorphic Forms
(8.20)
(8.21 )
(8.22)
The results as above (8.20) through (8.22) show the liftability of an or-
dinary point of Sh(p)(G, X)(lF) to characteristic 0 (as used in the proof of
Theorem 7.1).
We prove the VeT for projective Shimura varieties of PEL type. So we re-
turn to the notation introduced in Section 7.1. For simplicity, we give a full
exposition under the following four assumptions.
(u1) F = (Q[R] is an imaginary quadratic field with discriminant -5;
(u2) D / F i=- F is a division algebra with positive involution of the second
kind;
(u3) V = D; so, GU((Q) c ex for the opposite algebra e = tD of D;
(u4) The prime P splits into two distinct primes p and jJ in OF.
We write 0 = OF in this section.
Under (u1), we have G = GU; so, we have an advantage that any state-
ments already given exclusively for G or GU are both valid for our G = GU.
Since F /(Q is a Galois extension, the reflex field E = E( G, X) is a sub field of
F; so, OE,p = Zp by (u4). By (u3), Sh(p)(G, X) is projective over V. To have
the ordinary locus, the condition (ord) in the previous section is always in
force (which can be easily verified as in Remark 8.11 in many cases). It is not
difficult to generalize our result to more general projective Shimura varieties
of PEL type, although the endeavor certainly complicates the notation (see
[H02] for general treatment including quasi-projective Shimura varieties). At
the end of this section, we touch briefly upon the case where P is inert. By
our simplifying assumption, the reflex field for (GU, X) is either (Q or F.
Recall some of the notation introduced in Section Section 7.1 in our special
case. As in 7.1.1, let G = GU be a unitary similitude group defined with
respect to a central division algebra Dover F with a positive involution *.
Fix a prime p, and take V = D and a left OD-ideal L with L071Zp = OD071Zp.
Pick an element with s E D X with s* = -s, and define an alternating form
(x, y) = TrD/Q(xsy*) for x, Y E V, where TrD : D -+ (Q is the reduced trace
composed with TrF/Q : F -+ (Q. Then we see (bx, y) = TrD/Q(bxsy*) =
TrD/Q(xsy*b) = TrD/Q(xs(b*y)*) = (x, bOy) for x, y E V and bED.
Suppose that L satisfies (Ll-2) in Section 7.1.1 for this alternating form;
so, s E O;;,p. As before, we writeL: e -+ e for the involution on e given by
(ex, y) = (x, e"y); then, d = se* S-l. Recall
356 8 Ordinary p-Adic Automorphic Forms
fl(tJ(p) x tm(p))(A,A,i,rj(p),fillI,jil e)
=L p~(a-l )f(Aa, A, i, rj(p) , a(Jil lI , file)). (8.23)
if j > n,
if j -:; n,
if j > n,
(8.24)
if j -:; n.
the image of a in the Levi quotient of P. Indeed, taking a base (x,), of 1PA~t
and (Yk)k of 'TpA~t so that the matrix of the isogeny is exactly a J , the effect
on Q = (Qk,l) = (log(e(xk, yL))) is given by
(8.25)
and S1 has the degree of pure inseparability given by the value in (8.24).
Here is the justification of our argument. Write the multiplicative vari-
able on S as an m x n variable matrix t = (tk,l) (so Qk,l = 109(tk,I)). The
conjugation (10 i:) r-+ a J (10 i: )
a;1 induces Qk,l r-+ p-lQk,1 for some in-
dices (k, l). We split the set of indices (k, l) into a disjoint union I U J of two
subsets so that the conjugation by a J induces Qk,l r-+ p-lQk,1 if and only if
(k, l) E I. The covering S~] is given by Spf(8s[t~:nk,l)EI). Thus a formal
function on S has expansion L€ a€t€ E W~k}] ZIUJ. Writing
for ~ E
~(I) for the I-part of the index ~, a formal function f on S~] has expansion
f = L€ a€t€(I)/pH(J), and we have
The divisibility of T(~) by /1(~) for /1 in (5.40) for the symplectic group
GSp(2n) in the correspondence ring of Sh(p) (GSp(2n), X) /lFp has been proven
in a different manner in [DAV] VIl.4 (and also there is another proof of
Lemma 8.12 via the q-expansion principle in this case: [H02]).
eo = lim (T(p)t!,
n--+oo
(8.27)
We keep the notation introduced in the previous section. Let N be the upper
unipotent subgroup of GLm(Op) x GLn(Op). Applying the theory expounded
in Section 8.1 to the present setting, we can define the space of p-adic modular
forms V on Soo (which is the formal completion of S along Sd. We simply
write V for V N in this subsection. Thus V = ~ i (~k,kl V/tkl) for Vl,k,k ' =
HO(Tf,k,k l , OTe.k.kl )N. We also define V = ~i (~k,kl v'ii'j"kl). We now state
the final result on vertical control in this book.
8.3 Vertical Control Theorem 361
Theorem 8.13 Let the notation be as above. Let (GU, V, D, F,p = pp) be as
in (ul-4) at the beginning of this section. Identify 7Lp = Op naturally. Suppose
that M jzp = Sh~)(GU,X)jZp for an open compact subgroup K c GU(A(=»)
maximal at p is proper smooth over 7Lp and that S jZp = M [j; 1 for a lift E
of the Hasse invariant. Let T be the diagonal torus of G Lm (Op) x G Ln (Op)
and N be the upper unipotent radical of G Lm (Op) x G Ln (Op). Regard T as
a diagonal torus of the unitary group U(Qp). We say K, E X(T) is positive if
K, is dominant with respect to the upper triangular Borel subgroup of UjZp =
GL(m + n) in the sense described in Section 8.1.2. We write X+(T) for the
set of positive weights in X (T).
(1) There exists a canonical inclusion for R = 7Lp and Qp/7Lp,
f3: EB HO(Mjzp'f:d.'" 0z p R) Y V 0z p R.
"'EX+(T)
(5) Let V;rd be the Pontryagin dual 7Lp[[T(7L p)ll-module of eV (which is iso-
morphic to the 7Lp-dual of eV). Then V;rd is 7Lp[[TTll-free of finite rank,
where TT is the maximal p-profinite subgroup of T(7Lp);
(6) If K, E X+(T), V;rd 0zp[[T(Zp)1l,,,, 7Lp ~ Homzp(H~rd(S=,f:d."'),7Lp).
base field is not equal to Ql, we can easily find an example that the cokernel
of the inclusion eHO(S/w,~'Jw) ®w IF Y eHO(S/F,~'JF) is a nonzero finite
module for K, densely populated in X(T). Thus to obtain the result similar to
the above theorem in the nonprojective case, we need to add some assumption,
such as replacing ~" by the cuspidal sheaf~" (although for U (n, n) /IQ and
GSp(2n)/1Q it seems the result is true even for ~"). Here we state a precise
result without proof only for the quasi-split unitary group G = GU(n, n) for
an inert prime p of F (when p splits in F, the result is basically included in
Theorem 8.13 in the sense that we need to replace~" by the cuspidal sheaf~"
and V and V by their cuspidal counterparts Vcusp and V cusp ). Thus D = F,
V = F 2n = Fn E9 Fn, and ((Xl, X2), (Yl, Y2)) = TrF/IQ(Xl tY2 - X2tYl). The
lattice L can be given so that Lp = 0; E9 0;. The balanced maximal parabolic
subgroup P = {( 0: :)} of the unitary group U = U(n, n) has the reductive
part isomorphic to ResF/IQ(GL(n)) via ResF/IQ(GL(n)) 3 a r--+ diag[a, ta-lJ E
P. We consider the diagonal torus T C ResF/IQ(GL(n)) and the unipotent
radical N of the upper triangular Borel subgroup of ResF/IQ(GL(n)). Write
No for the unipotent radical of the upper triangular Borel subgroup of G; so,
N = No/Nn,n for the unipotent radical Nn,n of P. In the same way as in the
Hilbert modular case (see Section 4.1. 7), we define the space of cuspidal p-adic
modular form VCU8P := Vc;;:sp (resp. its p-divisible counterpart: VCU8P := V~sp)
for the Igusa tower over Sh<t) for an open compact subgroup K of G(AJ=))
maximal at p. In this case, the single GLn(Op) acts on the tower (in place of
GLn(Op) x GLn(Op) in the case where p splits).
Theorem 8.14 Let G /IQ be the quasi-split unitary similitude group of signa-
ture (n, n). Suppose that p is inert in the imaginary quadratic field F. Let the
notation be as in Theorem 8.13. Let M/ op be the toroidal compactification of
Sh<:)(GU,X)/op, and put S = M[1;;J for a lift E of the Hasse invariant.
(1) We have an inclusion (3 : EB"EX+(T) HO(M/op'~" ®zp R) Y V ®zp R for
R = 'lLp and Qlp/'lLp with the identity ofq-expansion: (3(j)(q) = Rcan(j)(q).
(2) Im((3cu8P)[~J n V is dense in Vcusp , where (3cusp is the restriction of (3 to
EB"EX+(T) HO(M/op'~,,).
(3) Write U(p) = TIl::oJ::on u J (p) for the standard Hecke operators at p of level
p=, where, for a J = diag[lJ,p1n-J,p21j,p1n-JJ E G(Qlp),
uJ(p) = fLn,n(aJ)-l[No('lLp)aJNo('lLp)J.
Then u J(p) is well-defined over Op if K, ~ fLn,n. Define the ordinary pro-
jector eo = limn-t= U(p)n! on the cuspidal subspace Vcusp of V. Then
H~rd(S,~" ®zp Qlp/'lLp) is a p-divisible module with finite co rank.
(4) If K, is sufficiently regular, H~rd(S,~" ®zp Qlp/'lLp) is isomorphic to a direct
summand H~rd(M,~" ®zp Qlp/'lLp) of HO(M,~" ®zp Qlp/'lLp).
(5) Let V;rd be the Pontryagin dual Op[[T('lLp)JJ-module of eo (Vcusp). Then
V;rd is Op[[rTJJ-free of finite rank, where rT is the maximal p-profinite
subgroup ofT('lLp).
8.4 Irreducibility of Igusa Towers 363
The proof for the cuspidal subs heaves in the nonprojective cases is basically
the same as in the projective case except for an extra effort necessary to verify
(Hp1), because in the nonprojective case, the ordinary locus 5 m of the smooth
toroidal compactification is no longer affine. This can be done via q-expansion
at all cusps of the Igusa tower (in the same manner as in the Hilbert modular
case, Section 4.1.7). As we show in the following section, the Igusa tower T~
over a geometrically irreducible component 5'1 of 51 for general unitary groups
may not be geometrically irreducible, although each irreducible component of
T~ contains a cusp. So, the naive use of q-expansion only at one cusp (often
00) in the elliptic modular case (done in Section 4.2.9) to show the integrality
of the Hecke operator no longer works in general, but we need to use one cusp
at each component. In any case, as we have done in the split case, one may use
the Serre-Tate deformation coordinates in place of the q-expansion in order
to prove the integrality of U (p) in the general case. See [H02] for details of
the arguments in the nonprojective case.
Hereafter we use the symbols (G, Xc) and (GU, Xcu) exclusively for those
defined in Section 7.1.3 (particularly, (7.1)). Write (GSp(2g)/'Q' X So' SJ g USJg )
for the pair described in (C) in Example 7.6. For simplicity, we write
such that i* A CSp (2g) ~ Ac. Our construction of Shc in Sections 7.1.2 and
7.1.3 tells us the finiteness and unramifiedness of this morphism (Corol-
lary 7.2).
Since hx E Xc gives rise to the complex structure on VlR = V ®1Ql JR and
the fiber A of Ac at x has expression A(C) = VlR / L as a complex torus, the
morphism i is described as hx --+ if; 0 hx at the level of the symmetric space for
an embedding if; : G /lR Y GSp(2g) /lR, where G is the subgroup of GU with
rational similitude (defined in Section 7.1.1). Recall that X6 = X6u for which
we write X+. The natural morphism ic : Shc --+ Shcu with icAcu ~ Ac
identifies the neutral component of S hc and that of S hcu. This also follows
from the fact T(Q(GU, Xcu)) = E(G, Xc) (for E(G, Xc) in [MiS] 4.2) inside
because the left-hand side of the above inclusion is the closure of the image
of G(A =)
( 1
in
CU(A,(=l)
1 . Thus Q(GU, Xcu) acts on the neutral component
Zeu(1Ql
of Shc. The embedding if; is obtained by regarding a E G(Q) as a Q-linear
automorphism if; (a) of V preserving the symplectic form (-,.) in (L2) of Sec-
tion 7.1.1; so, if; takes G(Q) into GSp(2g). Thus if; is defined over Q. In other
words, if; is induced by C = EndD(V) y EndlQl(V), and if;(g)(i(x)) = i(g(x))
for x E X+ and 9 E GU(JR)+.
Using the description of complex points Shc(q in (7.7), it is easy to
verify that i is an embedding S hc (q y S h CSp (2g) (q of complex manifolds.
Thus i : Shc --+ Sh CSp (2g) is a closed immersion ([D1] 5.4) defined over E =
E(G, Xc) = E(GU, Xcu) compatible with reciprocity laws in Theorem 7.7
(4). In particular the reflex field E and the constant field QCYc of the arithmetic
function field J for (GSp(2g), X) are linearly disjoint over k = QCYc n E; so, we
366 8 Ordinary p-Adic Automorphic Forms
Recall that the p-integral model She!} is defined over V = W n E. The mor-
phism i induces a finite unramified morphism of the p-integral models
.. Sh(p) Sh(p)
Z• C/V -+ CSp(2g)/V'
Since ¢ brings the connected component G(JR)+ into GSP2g (JR)+ , i brings the
geometrically irreducible component (the neutral component) 9J1g) = 9J1gb
of Shg) containing the image of X+ x 1 into the neutral component 9J1(p) =
~pfM9J1ir of sh~1p(29)/Q'
Supposing the condition (ord) requiring that the neutral component
9J1g) @v JF has dense p-ordinary locus Sl, we look at the stalk Os"x at a
closed point x E Sl(lFp ) in the neutral component of Sl' The formal comple-
tion Sl,x = Sp!(Os"x) along x is the formal deformation space of an ordinary
abelian variety Ax sitting on x. By the description of the deformation space
in Sections 8.2.4 and 8.2.5 (see the following subsection for more details of
this point), we conclude that i induces an embedding of Sl,X into the formal
completion along i(x) of the Siegel modular variety. Since i is unramified over
sh~1p(29) (see Corollary 7.2) and i is generically a closed immersion, i above
is a closed immersion over V of 9J1g) into the ordinary locus of 9J1W for the
residue field JF = lFp of W, although we only need the fact that i is a closed
immersion on an open dense sub scheme of Shg) faithfully flat over V. In any
case, the function field of the geometrically irreducible component of the open
dense subscheme coincides with the one of the entire neutral component, over
E or over K" where we write K, for the residue field of V.
Pick x E Sp2g(A(oo)) with xp E GSP2g(Zp). Then the scheme automor-
phism T(X) E Aut(Shcsp (2g)) induces an automorphism of the p-integral
model sh~1p(29)' which is in turn induced by an isogeny
a(x) : A CSp (2g) -+ T(X)* A CSp (2g) such that a(x) 0 TJ(p) = TJ(p) 0 x
For each special point x E X+ and the reciprocity map rx : Gal(E~b / Ex) -+
Tx(A)/Tx(Q), i(x) E X+ is a special point for 9Jl, and by definition in Sec-
tion 7.2.2, we find ri(x) = ¢ 0 r x . Then we confirm from the local reciprocity
law (rc1-2) that
for each f E mcu and 0- E Gal(E~b / Ex). Since Tx(A) generates GU(A)
(moving around all special points x E X+), from the density of special points
in 9Jlcu, we confirm (8.28). Therefore i induces a well-defined homomorphism
¢: r(g(GU, X)) -+ Aut(JE/E).
From (8.28), we find Do. ::::J r(¢(g(GU, X))). Therefore, r(¢(g)) preserves
the prime ideal D c m cu , and for each 0- E Dv n Do., we have o-(p + q) c
p + q, which shows that 0- as an automorphism of it preserves mcu. We get
8.4 Irreducibility of Igusa Towers 369
Recall the lattice L of V (we have chosen as above) satisfying the conditions
(Ll-2) in Section 7.1 for GU and GSp(2g) at the same time. We have the
maximal compact subgroup GU(Zp) = {x E GU(Qp)lxLp = Lp} of GU(Qp)
and G(Zp) = GU(Zp) n G(Qp) of G(Qp), and we may assume that G(Zp) =
¢-1(GSP2g(Zp)) (choosing a suitable Z-base of L). If a prime factor Po of P
in 0 0 = OFo splits into pp in 0, we have the Po factor of GU(Zp) isomorphic
to G Ln (Op) x 0; for a suitable n. If Po Ip is a prime factor of P in 0 0 giving
a prime p = PoO of 0, the po-factor of GU(Zp) is isomorphic to GUo(Oo,p)
(see Lemma 8.10), where GUo is the quasi-split unitary group defined over
0 0 ®z Z(p) given by
(8.30)
(ord) The Hasse invariant of Acu does not vanish identically on 9J1~~/IF'
370 8 Ordinary p-Adic Automorphic Forms
(8.31)
pEE
for p-adic W-algebras R, where we have written L"j; = (Lp/ L~rd) (6)0 017
and L"j;e = (Lp/ L~rd) (6)0 0Ee. Here 017 = I1 PE E Op. Regarding Ax as
a member A 2 (x) of the universal family over the (geometrically irreducible)
Siegel modular variety 9J1;~, by (8.22), the formal completion 5 along x in
ShK'(GSp(2g), X)(W) is given by
where the symbol Sym 2 indicates the symmetric square tensor over Zp
and L~t = Lp/ L~rd. The morphism i : 9J12?) --+ 9J1(p) restricted to Scu
is the embedding induced by the projection Sym2(L~t) ---» cL"j; (6)0 cL"j;e,
which associates a symmetric form iJt(x, y) = 'IjJ(x, y) + 'IjJ(y, x) E S(R) with
each OD-bilinear form 'IjJ : -y;et(A) x -y;et(A) --+ Gm(R) in Scu(R). Note
that the 0 D-linear form 'IjJ is determined by its restriction to the subspace
I1 PE E (c· -y;;et(A) xc' -y;;~t(A)); so, 'IjJ --+ iJt gives rise to the closed immersion
of Scu into S. The restriction f H flscu for f E 0 s gives rise to a surjective
ring homomorphism 7r : 0 s ---» 0Scu with Ker(7r) = D0s' which shows that
the morphism i is a closed immersion over V of the ordinary locus of Sh2?)
into that of the Siegel modular variety (as claimed in the previous subsection).
Writing 0Scu = W[[t]] for variables t = (tp"])PEE",] for 1 ~ i ~ m(p) and
1 ~ j ~ m(pC), we can make v = vcu explicit. Write R(P) for the function field
of the geometrically irreducible component 9J12?ir of Sh2?ir. Recall the subring
'2.J of the Siegel modular function field ~ and its prime ideal q (vanishing on
372 8 Ordinary p-Adic Automorphic Forms
the image of the unitary Shimura variety in the Siegel modular variety) in
Theorem 8.15. Expanding f E (Q1jq) n jt(p) c 0Sau into I:~ a~t~, we have
a valuation veu(f) = Inf~ ordp(a~). Then Veu over jt is an extension of the
above Veu to jt(p) (whose residue field gives rise to the function field of an
irreducible component of the Igusa tower Too).
Recall the function field ~(p) of the generic fiber of the geometrically ir-
reducible component 9J1(p) of sh~1p(29)' Similarly as above, i(x) in the Igusa
tower over sh~1p(29) (W) gives a point in the irreducible component con-
taining the infinity cusp (as we have already proven the irreducibility in
this case: Theorem 6.27). Thus by Theorem 6.28, the original v defined for
~(p) in the proof of Theorem 6.27 can be rewritten in the same manner as
VeSp(2g)(f) = Inf~ ordp(c~), where we have expanded f E ~(p) into I:~ c~T~
for the variables T with Os = W[[T]]. This is because we can measure the
divisibility by p of an algebraic function on a geometrically irreducible smooth
scheme over W by the expansion at any closed point of the scheme with re-
spect to the local coordinate over W. In our case, one point is given by the
cusp 00 and another given by i(x) (since i(x) and 00 are sitting on the same
geometrically irreducible component of the special fiber). In particular, choos-
ing T suitably, Beu is a complete intersection defined by T' = 0 for a suitable
subset T' C T of variables in T. The generic transversality of 9J1~b/w and
9J1;~ inside 9J1;% claimed in the previous subsection follows from this ex-
pression. By this fact, the transversality is indeed valid in the entire ordinary
locus of 9J1;%. This also shows that, choosing the extension of Veu defined
above for jt(p) to jt compatibly with v on ~, the localization of Q1jq at meu
is the valuation ring of the extension Veu to Jt. In other words, the decom-
position group D for Veu is given by the stabilizer Deu,v of meu studied in
Theorem 8.15.
The symplectic case is similar. In this case, we have by (8.22),
where !l runs over all p-adic places of the base (totally real) field F and
the symbol Sym 2 indicates the symmetric square tensor over G p • The same
argument as above proves that the decomposition group for Veu is given by
the stabilizer of !leu in Theorem 8.15. We get for groups of type A and C:
Theorem 8.16 Suppose (unr) and (ord). Then the group D is the image
(under T) of a conjugate in Q(GU, X) under Aut(9J1eu) of
Proof. This directly follows from Theorem 6.28 and Theorem 8.15, because
the valuation we created, vcu, is given by v = VCSp(2g) on J induced from the
q-expansion by the choice of our embedding ¢ : GU --+ GSp(2g). For general
Vcu not necessarily related to Siegel modular q-expansions, its decomposition
group in Aut(J't) is the conjugate of the one specifically chosen. Since the image
of Q(G, X) in Aut(J't) is a normal subgroup of finite index, the conjugated
image of P remains in that of Q(G, X). 0
We consider the universal abelian scheme Act S over Sl = shgb [-k 1Q9w IF
for the Hasse invariant H. Then we recall the tgusa tower over Sl as follows.
We write Act[pm]/~I for the maximal etale quotient of Act [pm]. Then we
define a constant group scheme C m by
in Case A
in Case C.
By the above result, in the symplectic case, the q-expansion principle holds
as usual (as already stated for the Hilbert modular and the Siegel modular
cases). In the unitary case, to determine a mod p (or p-adic) automorphic form
by q-expansion, one needs an expansion at one cusp for each geometrically
irreducible component of the Shimura variety before the reduction modulo p.
We leave for the reader a precise formulation of this fact.
References
Books
[AAF] G. Shimura, Arithmetic2ty in the Theory of Automorphic Forms, Math-
ematical Surveys and Monographs 82, American Mathematical Society,
Providence, RI, 2000.
[AAG] S. S. Gelbart, Automorphic Forms on Adele Groups, Annals of Math.
Studies 83, Princeton University Press, Princeton, NJ, 1975.
[AAQ] M.-F. Vigneras, Anthmetique des algebres de quaternions, Lecture Notes
in Mathematics 800, Springer, New York, 1980.
[ABV] D. Mumford, Abelian Varieties, TIFR Studies in Mathematics, Oxford
University Press, New York, 1994.
[ACM] G. Shimura, Abelian Varieties w2th Complex Multiplication and Modular
Functions, Princeton University Press, Princeton, NJ, 1998.
[AGD] A. Borel and G. D. Mostow, editors, Algebraic Groups and Discontinuous
Subgroups, Proc. Symp. Pure Math. IX, American Mathematical Society,
Providence, RI, 1966.
[ALF] K. Iwasawa, Algebraic FUnctions, Translation from the 1973 Japanese
edition by Goro Kato. Translations of Mathematical Monographs, 118,
American Mathematical Society, Providence, RI, 1993.
[ALG] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics 52,
Springer, New York, 1977.
[AME] N. M. Katz and B. Mazur, Arithmetic Moduli of Elliptic Curves, Annals
of Math. Studies 108, Princeton University Press, Princeton, NJ, 1985.
[ARE] J.-P. Serre, Abelian £-Ad2c Representations and Elliphc Curves, Ben-
jamin, New York, 1968.
[ARG] G. Cornell and J. H. Silverman, editors, Arithmetic Geometry, Springer,
New York, 1986.
[BALl N. Bourbaki, Algebre, Hermann, Paris, 1958.
[BCM] N. Bourbaki, Algbre Commutative, Hermann, Paris, 1961-1965.
[BLI] N. Bourbaki, Groupes et Algebres de L2e, Hermann, Paris, 1972-1985.
[BNT] A. Weil, Basic Number Theory, Springer, New York, 1974.
[CBT] W. Messing, The Crystals Associated to Barsoth-Tate Groups; With Ap-
plications to Abelwn Schemes, Lecture Notes in Mathematics 264, New
York, Springer, 1972.
376 References
[Ko] R. Kottwitz, Points on Shimura varieties over finite fields, J. Amer. Math.
Soc. 5 (1992), 373-444.
[L] M. J. Larsen, Arithmetic compactification of some Shimura surfaces,
CRM Publications 13 (1992), 31-45.
[LaR] R. P. Langlands and M. Rapoport, Shimuravarietiiten und Gerben, J.
Reine. Angew. Math. 378 (1987), 113-220.
[MaM] Y. Matsushima and S. Murakami, On vector bundle valued harmonic
forms on symmetric spaces, Ann. of Math. 78 (1963), 365-416.
[Mi] J. Milne, Canonical models of (mixed) Shimura varieties and automor-
phic vector bundles, Perspective Math. 10 (1990), 283-414.
[Mil] J. Milne, The points on a Shimura variety modulo a prime of good re-
duction, CRM Publications 13 (1992) ([ZFP]), 151-253
[Mi2] J. Milne, Shimura varieties and motives, Proc. Symp. Pure Math. 55
(1994) ([MTV]) Part 2, 447-523.
[MiS] J. Milne and K-y. Shih, Automorphism groups of Shimura varieties and
reciprocity laws, Amer. J. Math. 103 (1981), 911-935.
[Mt] T. Miyake, On automorphism groups of the fields of automorphic func-
tions, Ann. of Math. 95 (1972), 243-252.
[Mu] D. Mumford, Biextensions of formal groups, in Algebraic Geometry, Ox-
ford Univ. Press, Oxford (1969), 307-322.
[Mu1] D. Mumford, An analytic construction of degenerating abelian varieties
over complete rings, Compositio Math. 24 (1972), 239-272.
[P] A. A. Panchishkin, On the Siegel-Eisenstein measure and its applica-
tions, in "Proceedings of the Conference on p-adic Aspects of the Theory
of Automorphic Representations (Jerusalem, 1998)". Israel J. Math. 120
(2000), part B, 467-509.
[Ra] M. Rapoport, Compactifications de l'espace de modules de Hilbert-
Blumenthal, Compo Math. 36 (1978), 255-335.
[Ri] K. A. Ribet, P-adic interpolation via Hilbert modular forms, Proc.
Symp. Pure Math. 29 (1975), 581-592
[SBT] 1. Shepherd-Barron and R. Taylor, Mod 2 and mod 5 icosahedral repre-
sentations, J. AMS 10 (1997), 283-298.
[Sc] A. J. Scholl, Motives for modular forms, Inventiones Math. 100 (1990),
419-430.
[Se] J.-P, Serre, Geometrie algebrique et geometrie analytique, Annales Inst.
Fourier 6 (1955), 1-42 (CEuvres I 402-443, No. 32).
[Se1] J.-P. Serre, Le probleme des groups de congruence pour SL2, Ann. of
Math. 92 (1970), 487-527 (CEuvres II 487-527, No. 86).
[Se2] J.-P, Serre, Proprietes galoisiennes des points d'ordre fini des courbes
elliptiques, Inventiones Math. 15 (1972), 259-331 (CEuvres III 1-73, No.
94).
[Se3] J.-P. Serre, Formes modulaires et fonctions zeta p-adiques, Lecture notes
in Math. 350 (1973), 191-268 (CEuvres III 95-172, No. 97).
[Se4] J.-P. Serre, Quelques applications du theoreme de densite de Chebotarev,
Publ. IRES. 54 (1981), 323-401 (CEuvres III 563-641, No. 125).
[Sh] G. Shimura, On analytic families of polarized abelian varieties and auto-
morphic functions, Ann. of Math. 78 (1963), 149-192 ([63b] in [CPS] I).
[Sh1] G. Shimura, On modular correspondences for Sp(N, £::) and their con-
gruence relations, Proc. NAS 49 (1963), 824-828 ([63d] in [CPS] I).
References 381