Math-Net.

Ru
All Russian mathematical portal

V. G. Drinfeld, Elliptic modules, Mathematics of the USSR-

Sbornik, 1974, Volume 23, Issue 4, 561–592
DOI: 10.1070/SM1974v023n04ABEH001731

Use of the all-Russian mathematical portal Math-Net.Ru implies that you have read and
agreed to these terms of use
http://www.mathnet.ru/eng/agreement

Download details:
IP: 173.165.13.181
April 15, 2024, 22:13:43
Mat. Sbornik Math. USSR Sbornik
Tom 94 (136) (1974), No. 4 Vol. 23 (1974), No. 4

ELLIPTIC MODULES
UDC 519.49
V. G. DRINFEL'D

Abstract. The notion of elliptic module is introduced, generalizing the concept

of an elliptic curve, and an analog of the theory of elliptic and modular curves is con-
structed. Here the role of the group GL(2,Q) is played by GL(2, k), where A; is a func-
tion field. A theorem on the coincidence of L-functions of modular curves and Jacquet-
Langlands L-functions corresponding to k is proved.
Bibliography: 14 items.

Introduction

A) Statement of the problem and main result. Let k be a global field, <*> a place
of k, d a natural number, and k the completion of k at oo.
Definition. The triple {k, oo, d) is called admissible if a) all places of k, except
perhaps oo, are nonarchimedean, and b) d < [k : k ].
Admissible triples are of the following types:
1) k — Q, oo is the archimedean norm, and d— 1;
2) k = Q, °° is the archimedean norm, and d- 2;
3) k is an imaginary quadratic extension of Q, <*> is the archimedean norm, and
d= 1;
4) k is a function field, and oo and d are arbitrary.
The goal of this paper is the generalization of three classical theorems (con-
nected with the first three types of admissible triples): 1) the Kronecker-Weber
theorem, 2) the Eichler-Shimura theorem on (^-functions of modular curves, and 3)
the fundamental theorem on complex multiplication. This generalization is connected
with the fourth type of admissible triple.
Let (k, oo, d) be an admissible triple. We introduce the following notation: A
is the ring of elements of k which are integral at all places except oo; U is the ring
of adeles of k\ and v-, is the ring of adeles without the oo component. We formulate
the classical theorems 1)—3) in a form suitable for generalization. First we introduce
some necessary definitions,
Let (&,·οο, d) be an admissible triple of type 1)—3), and let Κ be a field over A
(i.e. there is given a homomorphism i: A -* K).

AMS (MOS) subject classifications (1970). Primary 1OD25, 12A65; Secondary 14K22.
Copyright Ο 1976, American Mathematical Society
561
562 V. G. DRINFEL D

Definition. An elliptic Α-module of rank d over Κ is the following: in case 1), a

homogeneous torus over K; in.case 2), an elliptic curve over K; and in case 3), an
elliptic curve X over Κ together with a homomorphism φ: A -* End X such that i =
D ° φ, where D: End X -» Κ is a differential.
In the same way we introduce the concept of an elliptic Λ-module of rank d over
S, where S is a scheme over A.
Definition. Let 5 be a scheme over k, I C A a nonzero ideal, and X an elliptic
/l-module of rank d over S. A structure of level I on X is an isomorphism (of A-
modules over S) (I~ /A) x 5 ^ X /? where Xf CX is the annihilator of 7.
If / is sufficiently small, then the functor that associates to the scheme S over
k the set of isomorphism classes of elliptic i4-modules of rank d over S with struc-
ture of level / is representable by a smooth (d — l)-dimensional manifold Νj. We set
Ν = VimJNJ. We can define in a natural way an action of the group GL(d, 21,) on N.

Theorem 1. Let (k, <*>, d) be a triple of type 1) or 3). Then Ν is the spectrum of
the maximal abelian extension of k, which is completely split over oo. The action of
21, coincides with the action from class field theory.

Theorem 2. Let (k, oo, d) be a triple of type 2), let Ν be a smooth compactifica-
tion of N, and let I be a prime number. Then Η {Ν, Q^)~ φ . Vί <g> W^., where Ui is
an irreducible representation of GL(2, 21,) over Q, and W. is a representation of
Gal(Q/Q) in the space Q? . Let U. =&)U?, where U? is an irreducible representa-
p
tion of GL(2, Qp), and let V/ . be the restriction of W. on GaKQ^/Q^).
1) If p ^ / and U? is a representation of class 1 (cf. [l]), then W? is unramified
and L(s, W?) = L(s — %, U?). Here U? is the representation contragredient to U?, and
the L-functions are taken in the sense of Serre and Jacquet-Langlands,
2) φ ? . U. <8>-C ~ H o m G L ( 2 R
)(I1, AQ), where II is a representation of GL.{2, R)
in the space of functions on Pj(R), factored out by the constants, and AQ is the
space of parabolic forms on GL(2, 21) in the sense of [10],

In this paper the concept of elliptic module for triples of type 4) is introduced,
modular varieties are constructed, and analogs of Theorems 1 and 2 are proved for d = 1, 2.
B) Outline of the paper. The concept of elliptic module is introduced in §2. In
§3 an analytic theorem is proved on the uniformization of an elliptic module over k
with the help of a lattice in ^ o In S»5 a universal family of elliptic A-modules of
rank d is constructed. The proof of smoothness of the modular varieties uses
properties of formal modules (cf. §§1 and 4) (a formal module is the analog of a
formal group). Also in ^ 5 , a congruence relation of the form [14] is proved. In §6 the
modular varieties are constructed analytically (as factors of some domain Ω by a
discrete group). The domain Ω is closely connected with the Bruhat-Tits complex
of the group GL(d, k ) . In §7 elliptic modules over complete discrete normed fields
are studied. In §8 an analog of Theorem 1 is studied. In §9 the compactifications of

Editor's note. In English the common term is cusp form.

 ELLIPTIC MODULES 563

7. In §§10 and 11
the analog of Theorem 2 is proved.
This article was written under the influence of the paper [14] and also of conver-
sations with I. I. Pjateckil-Sapiro. However, modular varieties have been constructed
algebraically (as in [6]). The analytic description of modular varieties is based on
ideas from [3] (although the congruence subgroups of GL{2, A) are not Schottky groups).
The author expresses his deep gratitude to Ju. I. Manin and I. I. Pjateckil-Sapiro
for their valuable remarks and attention to this paper.

§1. Formal modules

A) Definitions and notation. A formal group over a ring Β is, by definition, a
formal series F £ β[[χ, y]] such that F(x, y) = F{y, x), F(x, 0) = x, and F(x, F{y, z)) =
F(F(x, y), z). A homomorphism from a formal group F into a formal group G is a
series β 6 B[[x]] such that /3(F(x, y)) = G(/3(x), β(γ)). For any formal group F over Β
there is a canonical homomorphism D: End F -> Β that sends the endomorphism φ to
the number φ (0).
Example. F(x, y) = χ + y; this group is called additive.
Let β have characteristic p. Every endomorphism of the additive group over Β
is defined by a series S°^0£>.x^ . Henceforth we shall identify the element b€ Β with
the endomorphism of multiplication by b. The Frobenius endomorphism (corresponding
to the series x^) will be denoted by τ. Thus the ring of endomorphisms of the additive
group consists of " s e r i e s " ^fTbr1 with the commutation rule rb — b^r. We denote
this ring by B{{r|}.
Let 0 be a ring, β an algebra over 0, and y: Ο -> Β the natural homomorphism.
Definition. A formal 0-module over β is a pair (F, /), where F is a formal group
over B, and / is a homomorphism from 0 to End F such that D ° f = γ.
Example. F(x, y) = χ + y and / (x) = ax for a€ 0 (we write / instead of f{a)).
This module is called additive.
The germ of a formal 0-module over β mod deg n is a pair (F, /), where F £
β[[λτ, y]]/(*, y)n and / e β[[χ]]/(χ"), and all relations between F and / which arise
from the definition of a formal module are satisfied mod deg n.
We consider the following functor from the category of O-algebras to the category
of sets: β —ι > set of formal 0-modules over β. We can clearly represent this functor
by some algebra Λ^. (The generators of AQ are the "indeterminate coefficients"
of the series F and / , and the relations between them are those which are required
for (F, /) to be a formal 0-module.)
AQ has a natural gradation. It is easy to check that the set of germs of formal
0-modules over B mod deg(ra + 1) is canonically isomorphic to the set of 0-module
homomorphisms φ ^ Γ ρ Λ ^ ^ Β such that ip(ab) = ψ{α)φ{^ and ψ(ΐ) = 1. Elements of
the form ab, where a€ AQ, b £ A o, deg a > 1 and deg b > 1, generate a homogeneous
ideal. We denote by AQ the factor of AQ by this ideal.

Proposition 1.1. Let η > 2. Then Λ^~ (as an 0-module) can be defined by gen-
erators a and h(a) (for all a £ 0) and the relations
5 6 4
V. G. DRINFEL'D

/ „ -, jh(a), if η is not a power of a prime number

k
[fl(a)op, if n
= p,

h(a + b) — h(a) — h(b) = aCn(a, b), (2)

ah(b)±bnh(a) = h(a,b). (3)
n n n
(Here Cn(x, y) = (x + y) - x - y if η is not a power of a prime number, and
n n n k
Cn{x, y
) = (U + y) - * - y )/p if η = p ).
" —11
—
Proof. Clearly, A o can be defined by the generators h{a) and c. (0 < i < n)
with the relations
a (x + y + Φ {x, y)) + h (a) (x + y)n = ax-\-ayJrh (a) xn^-h (a) yn -\- Φ (αχ, ay),
ax + h {a) xn-\-bx-\-h (b) xn + Φ (ax, bx) = (a + 6) * + Λ (a f 6) JC",
h (ab) xn = a(bx + h (b) xn) -+- Λ (a) (&x)",
(^, y) = ®{y, χ), Φ(ί/, ^) + Φ(*, y + ζ) = Φ(χ, y) + 0>(x + y, ζ)
(where Φ(χ, y) = Σ " cixlyn~ l
). It is well known that from the last two relations fol-
lows Φ(χ, y) = OLC (X, y) (cf. [7]). D
B) The case when 0 is a field.

Proposition 1.2. 1) // 0 is a field, then every formal 0-module is isomorphic to

an additive module.
2) // 0 is infinite, then there exists a unique isomorphism with an additive
module, whose derivative at zero equals 1. In this case Λ η ^ O[cj, c 2 , . . . ] , where
deg c. = i.

Proof. The case of characteristic zero is well known (cf. [7]). Let char 0 = p,
let Β be an algebra over 0, and let (F, f) be a formal 0-module over B. Then multi-
plication by p in F is zero; therefore (cf. [7]) F is isomorphic to an additive group.
We can assume that F is additive. Let /: Ο -» β{{ΓΠ have the form f(a) = a + φ(α)τ
(mod τ ). We shall prove that there exists an s e β such that (1 + ST )f(a)(l + sr^)~
= a (mod τ ) for all a£ O. Indeed, since / is a homomorphism then φ(α + b) = φ(α) +
<t>(jb) and φ(α ο b) = αφ(Β) + bpk<f>(a), and so (bpk - bU(a) = (apk - ώ)ώ(δ). If 0 C F ,,
then the existence and uniqueness of 5 is clear (s = φ(α)/(αρ — a), where a€ 0, a£ F ,).
Ρ
If Ο C F ,2 , then ώ is differentiation; therefore φ = 0. π
ρ'
C) Computation of AQ , when Ο is the ring of integers of a local nonarchimedean
field. From now on, Ο will be the ring of integers of a local nonarchimedean field K,
π will be a prime element of O, p the characteristic of Ο/(π) and q the order of

Proposition 1.3. A Q ^ 0. More precisely:

1) // η is not a power of q, then h(a) = (an - a)u and aC (x, y) = u[(x + y)n - xn - yn],
where u is a generator of A Q ~ .
2) If η - q , then h(a) = (an — a)u/n and a = pu/π, where u is a generator of

Proof. 1) If η is not a power of p, then h(a) can be expressed through CL by

 ELLIPTIC MODULES 565

using formula (1). If η is a power of p, but not a power of q, then there exists a€ Ο
n n n
such that a - afi (π): from (3) it follows that (a - a)h{b) = {b - b)h(a); α can be
determined from (2),
2) Let β be a power of q. There exists an epimorphism AQ -> 0 sending h{a) to
n
{a — α)/π, and «.(-> //π. It remains to prove that A!V~ is generated by h(n). Let
Μ = Λ^~ /{/>("•)!. If x e AQ~ , then we denote by χ the image of χ in M. Since
n
^ = n h(b), then £(ττ-&) = 0 for any be 0. In particular, ~h(p) = 0. But
/ «— 1 — —
Ψ — l ) a , and so α = 0. Then Μ is an 0/(?7)-module, and h: 0/(π) -» Μ is differen-
tiation. Therefore h — 0, i.e. Μ = 0. D
Proposition 1.4. AQ ^ 0 [ g p g2, .. . ] , deg g f = z.

Proof. It follows from Proposition 1.3 that there exists an epimorphism

O[gp g2, ...]-> Λ ο consistent with the gradation. Proposition 1.2 implies that AQ ®
Κ ~ K [ c j , C 2 , . . .] and deg c. = i. Therefore the epimorphism we have constructed is
an isomorphism, Π

Corollary. 1) Every germ of a formal 0-module arises from a formal O-module.

2) // Β -* C is an epimorphism of O-algebras, then every formal 0-module over
C arises from an 0-module over B.

Proposition 1.5. 1) Let (F, /) and (G, g) be formal 0-modules over B, and let
(F, /) Ξ (G, g) mod deg n.
a) // η is not a power of g, then
F(x, y)==G{x, y)+v[{x+y)n—xn—yn] mod deg ( M + 1 ) ,
n n
fa(x) =ga(x) +v(a —a)x mod deg(n+l), v£B;
b) // η is a power of q, then

F (x, y) = G(x, y) + h-£- Cn (x, y) mod deg (n -j- 1),

π
α" — αxn mod deg (n + 1 ) , Λ 6 Β.
fa {χ) = ga (χ) 4- h -—± u

π
2) Let
n
x] ], φ (χ) =x—vx mod deg (n+ 1),

Then

F(x, y)=G{x, y)+v[(x + y)n—xn—yn] mod deg

fa(x)=ga(x)+v(an—a)xnmodaeg (n+1).
Proof. Statement 1 is a reformulation of Proposition 1.3, and statement 2 can be
checked directly. D

Corollary. The formal 0-module (F, f) is isomorphic to an additive module if and

only if the coefficients of f are divisible by π.

D) Classification of formal 0-modules over fields of "finite characteristic."

566 V. G. DRINFEL D

A homomorphism of formal O-modules is a homomorphism of formal groups that commutes

with the action of O.
Let £ be a field over 0/(77·) and let φ be a homomorphism of formal O-modules
ρ
over E. It is known [7] that if φ ^ 0, then φ(χ) - ψ{χ ) and ψ'(0) ^ 0. Since φ com-
mutes with the action of 0, it follows that log q\k. The number k/log q is called
the height of φ; the height of the zero homomorphism is 00. The height of a formal 0-
module is the height of the endomorphism of multiplication by π.
Remark. If Ο is the integral closure of Ο in a finite extension of Κ and η =
[θ : θ ] , then every formal θ'-module is also an 0-module, and its O-height is n times
its θ'-height.

Proposition 1.6. 1) There exist modules of arbitrary height.

2) There exist nonzero homomorphisms only between modules of the same height.
3) A formal 0-module of height h is isomorphic to the additive module mod deg q .

Proof. 1) We consider a homomorphism λ: Λ ο sa O[gv g2, . . . ] - > £ • such that

X(g /, ) ^ 0 and A(g) = 0 for i < q — 1. To this corresponds a formal 0-module over
z
q - 1
Ε of height h.
2) This follows from the fact that the height of the composition of homomorphisms
equals the sum of their heights.
3) This follows from Proposition 1.5. •
Proposition 1.7. 1) All formal O-modules of height h < °o over a separably closed
field Ε are isomorphic.
2) The ring of endomorphisms of such a module is isomorphic to the ring of inte-
gers of a central division algebra over Κ with invariant \/h.

Proof. A formal 0-module (F, /) will be called normal if the following conditions
are satisfied:
q qh
= x
2) F e F ,[[x, y]], fa e F , [[x] for a e O,
3) F(x, y) Ξ χ + y mod deg q ; fa(x) Ξ ax mod deg q .
a) Every formal 0-module over E of height h is isomorphic to a normal one. In-
deed, as in [7], by means of a change of variables we will have /_(*) = x ?
. Then con-
ditions 1) and 2) will be satisfied. By Proposition 1.6 we perform a change of vari-
ables with coefficients from F , , after which condition 3) is satisfied (and, as before,
q

conditions 1) and 2)).

b) With the help of Proposition 1.5 it can be shown that between any two normal
formal O-modules of height h there exists an isomorphism which is the identity
mod deg{q + 1).
c) Let L be the ring of endomorphisms of a normal formal 0-module of height h.
Clearly, Ο C L, L contains no divisors of zero, and L is complete in the 77-adic
topology. It follows from b) that every germ of an endomorphism of our module
mod deg qh with coefficients from F h
arises from an endomorphism. Therefore
 ELLIPTIC MODULES 567

it follows that dim L/nL = h2, and that the center of L/nL coincides with 0/{π).
Therefore L <8> Κ is a central division algebra of dimension h . The height is a norm
on L <8> K, and so L is a maximal order in L ® K. It follows from the relation τ ρ a
= aqr[°Bpqthat the invariant of L <8> Κ equals I/A. D

§2. Elliptic modules (algebraic approach)

A) Definitions and notation. Let β be a ring of characteristic p. We denote by τ
the endomorphism of the (algebraic) additive group over Β that sends t to t (just as
in §1). We shall identify the element b€ Β with the endomorphism of multiplication
by b. Every endomorphism of the (algebraic) additive group over Β has the form
Σ ί & τ ζ , and, moreover, rb - bpr. We denote the ring of such "polynomials" by B{T\.
We have two homomorphisms: c. Β -» Β\τ\, e(b) = b, and D: B{T\ -* B, D(SQ b.r1) = bQ.
The following notation will be used throughout this paper, k is a global
field of characteristic p; °o is a fixed place of &; kv is the completion of k
corresponding to the place ν; \ | is the normed absolute value corresponding
to ν (or its continuation to a finite extension k ); we write | | for | 1^; A =
ixe^l x! < 1 for ν Φ οο\· if f € Spec A, then A is the completion of Λ at
Let Κ be a field over A (i.e. there is defined i: A -» K). We call the place
i (Spec K) £ Spec A (i.e. the ideal Ker i) the "characteristic" (notation: " c h a r " ) .
Thus i is an imbedding if and only if Κ has general "characteristic."
Definition. An elliptic Α-module over Κ is a homomorphism φ: A -> K{r| such
that i = D ο φ and φ ^ ( ο/,
Β) β<Ζ72& and places of finite order. We define a mapping deg: Κ\τ\ -> Ζ in the
following way: deg £Q α τ ζ = p n for α ^ 0; deg 0 = 0.

Proposition 2.1. a) 0 is an imbedding.

b) There exists d > 0 such that deg φ(α) = \a\ for a 6 A.

Proof, a) If Ker φ ^ 0, then Ker φ would be a maximal ideal (since Κ\τ\ has no
divisors of zero), and so Im φ would be a field, i.e. Im φ C e(K) and so φ = e ο i.
b) Clearly
deg φ {ab) = deg φ (α) · deg φ (b), deg(p(a + 6) e$:max(deg<p(a), degcp(6)),
deg (£(a) = 0 if and only if a = 0, deg 0(a) > 1 for a ^ 0, and deg 0(a) > 1 for some
a 6 A. Therefore deg ο φ extends to a nontrivial absolute value on k, which cannot
correspond to a finite place. D
Definition, d is called the rank of the elliptic Α-module φ.
Example. Let ,4 = F^L*]. Then φ\? = e oz'| F . The representation of φ is
equivalent to the representation of φ(χ) ε Κ{τ\. In place of φ(χ) one can take any ele-
ment of the form (χ) + Σ ? = 1 α.τ °&pq, where d > 1, ad^0, a. € K. The rank of such a
module equals d.
The representation of an elliptic Λ-module over Κ changes any K-algebra into an
Λ-module. Let Κ be the algebraic closure of K.

Proposition 2.2. Let Κ be a divisible Α-module. If a£ A, a ^ 0, then the number

568 V. G. DRINFEL'D

of points of order a in Κ does not exceed \a\d', where d is the rank of the elliptic
Α-module. Equality holds if and only if i(a) ^ 0. The torsion submodule in Κ is
isomorphic to Θ ^ 5 ρ ε α Λ (Kv/Av)'v, where j v < d for ν £ " c h a r " K.

Proof. Every divisible torsion Α-module is isomorphic to φ „ ΛΚ /Α Ϋν,

where the fv are found by counting the number of places of order a. D

Corollary. The rank of an elliptic A-.module is a natural number.

Remark. Since Κ\τ\ C K{lr|}, every elliptic Α-module over Κ defines a formal A-
module over K. If Κ has general characteristic, then every formal Α-module over Κ
uniquely extends to a formal ^-module. If " c h a r " Κ = ν € Spec A, then every formal
Α-module over Κ is uniquely extended from an A -module. The height of the formal
A -module corresponding to the elliptic Α-module is equal to d — f .
C) Isogenies. Let φ: Α -> Κ\τ\ and φ: A -» K{r\ be elliptic Α-modules over K.
A homomorphism from φ to φ is an element α e Κ{τ\ such that αφ{α) = ψ(α)α for α ε Α. Α
nonzero homomorphism is called an isogeny.
Remark. By comparing powers, it follows that isogenies exist only between
modules of the same rank.
Every homomorphism of elliptic Α-modules is also a hor^omorphism of their addi-
tive groups, and so one can consider the kernel of a homomorphism of elliptic modules.
The kernel of an isogeny is a finite /4-invariant group subscheme in the additive
group.

Proposition 2.3. Let φ be an elliptic Α-module. For a finite group subscheme

Η of the additive group, invariant with .respect to A, to be the kernel of an isogeny
from φ to some other module, it is necessary and sufficient that
a) if Κ has general "characteristic", then Η must be reduced;
b) if ν - " c h a r " Κ € Spec A and q is the order of the residue field of v, then
ft = Spec K[t]/(tq ) , where H, is the connected component of H.

Proof. An additive group is obtained by factoring the additive group by H. Let

u€ K{r\ be a homomorphism whose kernel is H. Since Η is invariant with respect
to A, there exists a unique homomorphism φ: Α -> Κ\τ\ such that uφ(a) = i/^(a)w for
a€ A. Clearly φ ^ e ο ζ, and ϋ{φ{α)) = [i{a)]n, where η is the order of HloQ. Therefore,
for φ to be an elliptic module, it is necessary and sufficient that i(an) = i{a) for
a€ A. D

Corollary. Any isogeny can be multiplied by another isogeny to an endomorphism

which is multiplication by α ε A, a ^ 0.

Proposition 2.4. Let X and Υ be elliptic Α-modules of rank d over K. Then

Hom(X, Y) is a protective Α-module of finite dimension {not exceeding d ) . If ν € Spec A,
v£ "char" K, then the homomorphism Hom(X, Y) <S>^ Ay -» Hom^ (TvX, ΤνΥ) isinjec-
tive {here Τ Χ is the v-component of the torsion submodule of K), and its cokernel is
torsiori'free.

Proof (cf. [2]). 1) Hom(X, Y) is a torsion-free Λ-module. If u, u< ε Hom(X, Y), then
 ELLIPTIC MODULES 569

deg(u + w) < max(deg u, deg w) and deg(au) = \a\ deg u, deg u > 1 for u ^ 0. There-
fore, if V C Hom(X, Y) <g>. k is a finite-dimensional subspace, then V f] Hom(X, Y) is
a module of finite type.
2) Let a e A such that \a\ < 1 and |β | = 1 for w ^ v, °o. Then the homomor-
k k
phism Hom(X, Y)/(a ) -» Horn . (Τ Χ, Τ Υ)/(Λ*) is injective, and so limHom(X, Y)/{a )
c^Hom^ (TvX, TvY). On the other hand, Hom(X, Υ) ®Λ Ay -» lim Hom(X, Y)/(/) is a
monomorphism. D

Corollary. Let X be an elliptic Α-module of rank d over K. Then End X is a

projective module, dim End X < d , a?z<s? End X ®^ & zs a division ring. If Κ has
general "characteristic", then End X is commutative and dim End X < d.

§3. Elliptic modules (analytic approach)

Let L be a finite extension of k^, and let Ls be the separable closure of L.
A lattice over L is a finitely-generated discrete /1-submodule in L s , invariant with
respect to Ga\{Ls/L). Let Γ\ and Γ- be lattices over L of dimension d. A mor-
phism from Γ . into Γ is a number <x £ L such that α.Γ\ C Γ\. Composition of raor-
phisms is defined by multiplication of numbers.

Proposition 3.1. The category of elliptic modules of rank d over L is isomorphic

to the category of lattices of dimension d over L.

Proof. 1) Let L be a lattice of dimension d over L. We set

(This product clearly converges uniformly in every circle, since / is an entire func-
tion.) First we prove the following lemma.

Lemma. Let Ε be a field, char Ε = p, and let Δ be a finite subgroup of E. Let

g(z) = Π α € Δ (.ζ - α ) . Then g{z + w) = g{z) + g{w).

Proof. Clearly g(z + w) — g(z) — g(u>) = 0 for z£ Δ or w e A . Therefore g(z)g(u>)

divides g{z + w) — g{z) — g{w). Comparing powers, we obtain g(z + w) =
g(z) + g(w). D
Since Γ ie the union of an increasing sequence of finite subgroups, we see that
f(z + u>) = f{z) + f{w). Clearly / induces a group isomorphism L/T<^ L. Since Γ is an
Α-module, the structure of an Λ-module p a s s e s over from L/T to L. If a £ A and a ^ 0
then f{az) and D o ^ j - ) r (f(z) - /(/3)) are analytic functions of ζ with the same di-
visors. Therefore f{az) = Ρ (f(z)), where Ρ is a polynomial of degree \a\ .
2) Let φ be an elliptic Λ-rnodule of rank d over L. There is a corresponding
formal ^-module over L (cf. the remark at the end of §2B). According to Proposition
1.2, there exists a unique / = 1 + Σ ~ b .τ1 £ L{{r|i such that fa = (fi{a)f for a £ A. We
shall prove that the formal homomorphism / is an analytic homomorphism. Let a£ A,
\a\ > 1 and φ(α) = a + Σ^ a.r1. Let i > s. From the identity fa - 0(a)/ we obtain
570 V. G. DRINFEL'D

Let c . ^ l b ^ " 1
. Then \a \ c . < m a x L < . , s ( \ a . \ p ~ l c . _ ) . L e t 1/\a\ < θ < 1, a n d c . < Q -
m a x ^ . ^ c._. for sufficiently large i. Therefore c{ * 0. Let Γ C L be the kernel of /.
s s
Clearly Γ C L , Γ is invariant with respect to Gal(L /L), Γ is an A-module, and
(ΐ/α)Γ/Γ s* (A/(o)) . Since Γ is discrete, Γ is a lattice over L of dimension d.
3) Let Τl and Γ 2 be lattices of dimension d over L, α e L, and αΓΛ C Γ .
Let /j and f2 be the entire functions constructed in 1) from the lattices Γ. and Γ .
The function f2(dz) is invariant with respect to Γ r Repeating the discussion from
step 1), we get / 2 (az) = Pif^z)), where Ρ is a polynomial. Ρ defines a homomor-
phism of elliptic Λ-modules. On the other hand, let the polynomial Ρ define a homo-
morphism of elliptic ^-modules. Then Ρ ο f is a formal homomorphism from an addi-
tive ^-module into a formal ^-module corresponding to the second elliptic /l-module. It
follows from Proposition 1.2 that P(f.(z)) = f AOLZ) for a uniquely defined a e L ,
Clearly, aXj C Γ 2 > D

Corollary. For any A and d there exist elliptic Α-modules of rank d over k^-

§4. Universal deformations of formal modules

A) Deformations of zero level (cf. [13]). We shall use the same notation as in
§1C. Let Onr be a maximal unramified extension of O, and let Onr be the comple-
tion of Onr. We consider the category C whose objects are complete local Onr-
algebras whose residue fields are isomorphic to Onr/(n)· The morphisms of C are
local homomorphisms of O^-algebras.
Let (G, g) be a formal 0-module over Onr/(n), and let R e C. A deformation of
the module (G, g) with basis R is a formal 0-module (F, /) over R whose reduction
modulo a maximal ideal is (G, g).

Proposition 4.1. Let (F, f) and (F , f ) be deformations of the modules (G, g)

and (G , g ) with basis R. Let φ: (F, /) -* (F , / ) be a homomorphism inducing the
zero homomorphism (G, g)-> {G', g'\ If the height of {G, g) is finite, then 0 = 0 .

Proof. Let m C R be a maximal ideal. It suffices to consider the case when

m r+1
= 0 and φ = 0 mod mr, r > 1. Then F ? (^U), φ(γ)) = φ(χ) + φ(γ), f'a(<f>(x)) =
r
αφ(χ), and so φ(Ρ(χ, y)) = φ(χ) + φ(γ) and <£(/α(*)) = αφ(χ). Let I: m -* R/m be a
linear function, φ' = 1(φ)· Then φ is a homomorphism from(G, g) into an additive
0-module, and so φ - 0. Π
Deformations ( F ^ / j ) and (F2,f?) of the module {G, g) are called isomorphic
if there exists an isomorphism ( F j , /j) ~ ( F 2 , / 2 ) inducing the identity automor-
phism on (G, g). (If (G, g) has finite height, then such an isomorphism is unique.)

Proposition 4.2. Let (G, g) fee a formal 0-module over Onr of finite height h.
The functor that associates to R e C the set of deformations of the module (G, g) up
to isomorphism is represented by the algebra Onr[[t ^, . . . , t^_ j]j.

Proof. Let O[g g . .] = AQ -> Onr/(n) be the homomorphism corresponding

to (G, g). We can assume that under this homomorphism g. ~> 0 for i < q - 1. Let
 ELLIPTIC MODULES 571

(F°, /°) be a deformation of (G, g) with basis 0nr[[t p . . , , tf)_l\] such that the
corresponding homoraorphism A Q -»· 0nr[[t . . ., t,_A\ sends g. into /. for 1 < i <
h l
h - 1 and g. into zero for / < q - 1, ] Φ q - 1. We shall show that (F°, /°) is a uni-
versal deformation.
Let Μ be a vector space over 0ητ/\π)· A 2-dimensional cocycle of the module
(G, g) with coefficients in Μ is a set |Δ e M[[x, y]], δ £ Λ1[[Ϊ]] for «e 0} such that

Δ (y, z) + A (*, G (*/, ζ)) = Δ (χ, y) + Δ (G (*, */), 2), Δ (χ, y) = Δ (ι/, χ),
δ α (x) f ό β (y) + Δ (g fl (χ), g a (//)) = αΔ (χ, y) + δ α (G (*, //)),
δ α (χ) + 6b (χ) -f Δ (^α (χ), g b (χ)) = 6 a+fe (χ), abb (χ) -4- δ α (gb (χ)) = ό Ω 6 (χ).

A coboundary of the series ι/ί e Λ1[[χ]] is a cocycle (Δ, δ), where

Δ ( Χ , y)= yf>(G(x,y)) —\p(x) — ψ(//), δ β (*) =^>(ga{x)) —ayjp(x).
Let Re C, mC R a maximal ideal, mT = 0, r > 1, and (F, /) a deformation of
the module (G, g) with basis £.

Lemma. 1) The exists a one-to-one correspondence between formal 0-modules

(F'j /') over R that are congruent to (F, /) modulo mr and 2-dimensional cocycles of
the module (G, g) with coefficients in mr, To the cocycle (Δ, δ) corresponds the
module ( F \ / ' ) , where F'\x, y) = F{F(x, y), Δ(χ, y)), f'a(x) = F(/fl(x), S f l W).
2) Two cocycles with coefficients in mr are cohomologous if and only if the
corresponding deformations are isomorphic.

Proof. If (F , / ) = (F , / ) mod mr, then the isomorphism of the deformations

(F , / ) and (F , / ) is the identity mod mr (by Proposition 4.1). The remaining
assertions can be verified directly. D
Let φ: 0nr[[tv . . . , th_ χ ]] -> R be a homomorphism such that 0(F°, f°) =
(F, /) mod mTo For the proof of the proposition it suffices to show the existence and
uniqueness of a homomorphism φ: 0nr[[t . . . , t,_ .]] -> R such that φ = φ mod mr
and ^ + ( F 0 , / ° ) « ( F , /). Let y^ Ο ^ Ή ^ , . . . , ί^_ χ ]] -» R be a homomorphism such that
0 θ ρ = φ(ίζ·) + € ? e· £ ffz^. Then the difference between the cocycles corresponding to
<^(F°, f°) and iA+(F0, f°) has the form Σ ^ " 1 e .(A., δ.), where (Δ., δ.) is a cocycle
with coefficients in Onr/(π) (which depends neither on (• nor on r). Then

(Δ(, δί)=ϋ mod degq\ (Au δ ( ) Ξ^Ο mod d e g ( ? ! + 1).

It remains to show that the classes (Δ., δ ) form a basis for the cohomology with
coefficients in Onr/(n)· This follows from the following two assertions (the first is
clear, and the second was essentially proved in §1B).
a) The coboundary xn is congruent to

{(x -\- i/)n — xn — yn, (an — a) xn} mod deg (n + 1),

The coboundary xq is congruent to

[hi JL Ci+h (x, y), hi aQ ~a xiUh\ mod deg (q^h + 1), ht φ 0.

572 V. G. DRINFEL'D

b) Let (Δ, δ) be a cocycle, and let (Δ, δ) = 0 mod deg n. If η is not a power of q,
then

(Δ, ό) = {ν [(χ + y)n — xn — yn, υ (an — a) xn) mod deg (n + 1).

If η is a power of q, then

n
(Δ, b)={h'-2-Cn (x, y), h · — — x \ mod deg (n + 1). D

B) Deformations of arbitrary level. Let Re C, and let m C R be a maximal ideal.

The assignment of a formal 0-module ( F , /) over R turns m into an 0-module. Let the
reduction of ( F , /) modulo m have finite height h. Let η £ Ζ, « > 0.
Definition. A structure of level η on a formal 0-module ( F , /) is an 0-module
homomorphism φ:((ϊ/πη)θ/θ) -> ra such that / (x) is divisible by

Π ( * - Φ (<*))·

For η > I it follows that fn(x) and Π α ε ( ( 1 / )o/o)h ^ x ~ < M ct )) divide each other.
In the case R = 0 / ( 7 7 ) there exists exactly one structure of level n. Let (G, g) be
a formal 0-module over Οητ/(π) of finite height />. A deformation of the module (G, g)
with structure of level η will be called a deformation of level n.

Proposition 4.3. 1) The functor that associates to R e C the set of deformations

of level η of the module (G, g) up to isomorphism is represented by some ring D .
2) D is a regular ring. Let η > 1, and let e. (i = 1, . . . , h) be a basis for
((ΐ/πη)θ/θ) as an 0/(77)"-module. The images of e. in D under the universal de-
formation of level η form a system of local parameters.
3) Let m < n. The homomorphism D -> D is finite and flat.

Proof, a) Let ( F , /) be a universal deformation with basis DQ — C)nr[[t ^ ..., t, _ j]].

Let 0 < r < h. Consider the functor Φ Γ that associates to each D Q -algebra Re C the
set of homomorphisms φ from ((I/77) O/O)r into a maximal ideal of R such that /_(#)
is divisible by

Π ^-φ(α)).

Lemma. Φ is represented by a ring Lf having the following properties:

1) L is a regular ring. Let e. (z = 1, . . . , r) be a basis for {(l/n)O/O)r. Then
the images of e . in L and also t , . . . , t,_, form a system of local parameters.
2) The homomorphism L _ , -> L is finite and flat.

Proof. For r = 0 the lemma i s true. Suppose that r > 1 and that the lemma has
1
been proved for Φ^, y Let φτ_ y ((l/n)O/O)r~ -> L r _ 1 be the homomorphism men-
tioned in the definition of Φ _ ,. We set θ• = φ _ j ( e ) ( l < i < r — l) and
 ELLIPTIC MODULES 573

Π
ae&o/op
Let L = L .[[Θ Y\/g(6 ). We define a homomorphism

φΛ: ( 7

so that the restriction of φ to the first summand coincides with φ _ , and the restric-
tion of φ to the second summand sends I/77 into Θ . Clearly, L is finite and flat
over L _ , ; furthermore,

and so L is regular, and θ,, . . . , θ , t , . . . , t,_ , form a system of local parameters.

It remains to prove that L represents Φ_. It is enough to show that / (x) is divisible
by

Π (χ — qv(a)).

Indeed, / (x) is divisible by χ - φτ(α) for a. e ((l/V)O/O) r . Since L y is regular, it

remains to show that φ is injective. Indeed, if 0 ( Σ ' α. e ) = 0, then Ιία.θ^ belongs
to the square of the maximal ideal of L , and so ex. € (77). π
Setting r = h, we obtain assertions 1)—3) for η — I.
b) Suppose that η > 1 and that statements 1) and 2) about £) have been proved.
η ;
Let e. (1 <i <h) be a basis for ((ΐ/τ7 )θ/θ) \ and let b{ be the image of e. in D^.
Clearly
Dn+1 = D n [[i/lt . · · , ί/η]]/(/π (ί/ι) — & ! , . . . , /« (I/ft) — bh).
Therefore D +
, is regular, (y,, . . . , yΛ is a system of local parameters in D +
,,
and the homomorphism D -* D + ,
is finite and flat. D

Proposition 4.4. Lei (F, /) be a formal 0-module over R e C with structure φ

of level n. Let Ρ C ((l/nn)O/O)h be a submodule. Then

H^SplR [[*]]/ JJ (x - φ (α)) CZ Spf R [[x])

aeP
is an O-invariant group subscheme, and the factor C = F/H is a formal 0-module. If

is an imbedding, then the corresponding homomorphism from ({l/nm)O/O) into the

maximal ideal of R is a structure of level m.

Proof. It is enough to consider the case R - D (cf. Proposition 4.3). Clearly Η

is the minimal closed subscheme in Spf #[[x]] containing Spf /?[[x]]/(x — φ{<χ)) for
all ae P. Since F(</>(a), φ(β)) = φ(α + β) and Spf fl[[x]]/(x ~ <jAa + β)) is contained in Η
574 V. G. DRINFEL'D

6
for α, β Ρ, it follows that Η is invariant under addition. Similarly it is proved that
Η is invariant under the action of 0. Since φ(α) φ 0 for α, ^ 0, the homomorphism
from F to G induces a nonzero tangent mapping. Since D has no divisors of zero,
m
G is a formal 0-module. Since the homomorphism from ({l/n )O/O) into the maximal
ideal D is injective, and D is regular, it follows that this homomorphism is a struc-
ture of level m. Π
C) Deformations of divisible modules, In this section, a formal group will be a
group object in the category of formal schemes. (For example, a discrete group is a
formal group.) Let R e C. A divisible 0-module over R is a formal group F over R
together with a homomorphism /: 0 -> End F such that F, is a formal 0-module, and

(/ < oo). (For R = Onr/(n), the sequence 0 -> F l o C -» F -> F/F[oC -* 0 splits.)
Let the reduction of F, modulo the maximal ideal have finite height h. A struc-
ture of level η on the divisible module (F, /) is a homomorphism ((\/πη)θ/θ)} -*
Mor(Spf R, F), including a structure of level η on F, and an epimorphism
h
i i

The concept of a deformation of level η is introduced as for formal modules. Propo-

sition 4.1 is easily generalized to the case of divisible modules.

Proposition 4.5. Let {G, g) be a divisible 0-module over Onr/(n) with structure
}
of level η such that G'j has height h, and G/G, ~ {K/O) .
Let n£ Ζ, η > 0. The functor that associates to R 6 C the set of deformations of
level η of the module {G, g) with basis R up to isomorphism is represented by the
ring Ε ^ D [[d., . . . , d.]], where D is defined as in Proposition 4.3· In particular,
Ε is a regular ring of dimension j + h, and EQ is smooth over Onr. If m < n, then
the homomorphism Ε -> F is finite and flat.

Proof. Let R e C, and let ( F , /) be a deformation of level η of the formal module

G\ . Clearly its extensions by a deformation of the divisible module G are classified
by the elements of Εχρ(Γ, Mor(Spf R, F)), where Γ is a factor of G/G l o c by elements
of order πη. If Μ is a module over 0, complete in the 77-adic topology, then Μ =
Exp(/C/O, Λ1). Therefore, fixing an isomorphism Γ « {K/O)J, we can identify
Εχρ(Γ, Mor(Spf#, F))
j
with [Mor(Spf R, F)] . α
Remark. Let θ' be the integral closure of Ο in a finite extension of K. By Onr
nr
we mean the maximal unramified extension of 0 in Ο . Let (G, g) be a divisible
θ'-module such that G, has finite height, and let Ε be the basis of a universal
deformation (of level zero) of (G, g). We can consider (G, g) as an 0-module; let E be
the basis of a universal deformation of the 0-module (G, g). Clearly the group of auto-
morphisms of (G, g) acts on F, and, in particular, 0 acts on F. It is easy to show
that if Γ C O ' is a subgroup which generates θ' as an 0-module, then Spf E' - (Spf F) .
 ELLIPTIC MODULES 575

§ 5 . Modular manifolds

A) Endomorphisms of the additive group.

Proposition 5.1. Let Β be a ring of characteristic p with Spec Β connected. Let
ι
fv f2e B\T\, f. = l.lQ α..τ , άχ > 0, ad . invertible for j= 1, 2. Let h £ B\T\ and hf\ =•. f2h.
1) // dl 4 dv then h = Q.
1
2) // d. = dj and h ^ 0, then h has the form "Σ-J h.r , and h, is invertible.

Proof. It suffices to consider the case when β is a local Artinian ring. We pro-
ceed by induction on the length of B.
1) Assume that the coefficients of h are nilpotent. (This is also true if d. 4 d~).
We shall show that h = 0. Let Μ C Β be an ideal such that m2 = 0 and h = Σ^Ατ*,
h. £ m. Then hf'j = f-h = « 0 2 ^· Since d, > 0, we have h - 0.
2) Let d — d^- d and h = Σ!? h .rz; let m C Β be an ideal, m = 0, h Em. Equa-
n
ting the coefficients for τ , we obtain h - 0. π

Proposition 5.2. Let Β be a ring, char Β - p. Let / = Σ^ αΛι e Β\τ\, and let
d > 0, with a, invertible and a. nilpotent for i > d. Then there exists a unique ele~
7
ment of the form 1 + Σ ^ α τ £ Β\τ\ such that the a. are nilpotent and

has degree d.

Proof. The uniqueness follows from Proposition 5· 1.

Let m C Β be an ideal, m = 0, a. € m for i > d, and η > d. Then the degree of
"1

is less than n. From this follows the existence. D

B) Construction of modular schemes.
Definition. Let S be a scheme over A. An elliptic Α-module over S of rank d is
a line bundle L over S together with a homomorphism φ: A -> End L (where End L
is the ring of endomorphisms of L as a group scheme over S) such that 1) for
any a £ Λ the differential of φ{α) is multiplication by a; and 2) for any field Κ and
morphism Spec Κ -> S, the corresponding homomorphism /I -» KJH is an elliptic module
of rank d in the sense of § 1 . A homomorphism of elliptic modules is a homomorphism
of group schemes over S that agrees with the action of A.
Remark. An elliptic module over S is called standard if for every a & A the
endomorphism φ{α) has the form Σ . * * ' bs\ where b{ £ H°{S, Ll~p). According to
Proposition 5.2, every elliptic module is isomorphic to a standard module, and every
isomorphism of a standard module is linear.
Let I C A be an ideal. We denote by V(/) the set of simple ideals containing /.
Let X be an elliptic module over S of rank d; let / φ 0, and let X{ C X be the
annihilator of /. Clearly X, is a finite flat group scheme over S. If the image of S
in Spec A does not intersect V(/), then Xj is etale over S.
576 V. G. DRINFEL D

Definition. A structure of level I on X is a homomorphism of Α-modules φ:

U~ /A) H> Mor(S, X) such that, for any m £ V{l), X as a divisor coincides with the
sum of the divisors φ(α), a e m~ /A.
Remark. If the image of S in Spec A does not intersect V(l), then a structure of
level / is an isomorphism (/" /A) xS^X-

Proposition 5.3. Let I C A be an ideal such that / ^ 0 and V(l) contains more
than one element. The functor that associates to the scheme S over A the set of
elliptic Α-modules of rank d with structure of level I up to isomorphism is represented
by a scheme M. of finite type over Α..

Proof. Let m £ V(l). It is enough to prove that the restriction of our functor to
the category of schemes over Spec A - m is representable. Indeed, if 5 is a scheme
over Spec A - m and X is an elliptic Α-module of rank d over S with structure of
level /, then a choice of nonzero elements {m~ /A) defines a trivialization of the
bundle X. D
C) Deformations of elliptic modules. Let V £ Spec A and Ο = A . The definition
of the category C was given in §4. Let X be an elliptic module of rank d over Anr/v
with structure of level vn.
We consider the functor that associates to R £ C the set of deformations of level
vn of the module X with basis R up to isomorphism. This functor can be represented
in the following way: Let / C A be nonzero ideal, V(/) ^ ν, I ^ A. We lift in any way
the structure of level vn on X to a structure of level lvn. Let y be the corresponding
point in Mj n <g> Anr, and F its completion at a local ring. Then F represents our
functor.
Let R £ C, and let Υ be an elliptic module over R. Then Υ = lim Υ vTl is a
—*
divisible A -module. A structure of level vn on Υ defines a structure of level η on Υ
Thus there is a homomorphism Ε -» F , where Ε was defined in §4C).

Proposition 5.4. This homomorphism Ε -> F is an isomorphism.

Proof, a) If R £ C and Υ is an elliptic module over R, Υ = lim Υ , then a struc-
ture of level 72 on Υ defines a structure of level vn on Y. Thus it suffices to prove
the proposition for η = 0.
b) We reduce the proof to the case A = F Ax], ν = (χ). Let χ £ ν. We may con-
sider an elliptic Α-module as an F [x]-module, and a divisible A^-module as an F [[xJ]-«
module. Let En0 be a basis of the universal deformation of the Fjlx]}'-module
ρ
X, and
F' a basis of the universal deformation of the F [x]-module X. The diagram
Ρ F
t t
E'o-+F'o
is commutative.
Let A be the localization of A by the set of elements of A which are relatively
prime to x. We define an action of A* on F' . Let a £ Α Π A , and let Υ be a deforma-
 ELLIPTIC MODULES 577

tion of the elliptic F [x]-module X with R 6 C. Then there exists a unique, up to

isomorphism, deformation Υ with basis R for which the endomorphism X -> X extends
to a homomorphism V -> Υ'. (Υ' is obtained in the following way: Let be F [x],
b(0)^0, a\b. Since Y,fc. is etale over Spec R, the subscheme X, . C X,^. uniquely
extends to a subscheme Η C Y,^, etale and finite over Spec R. Let Υ' = Υ/Η.) Thus
the subgroup Λ Π Λ acts on Fn', and Λ PlF [χ] acts trivially. This allows us to de-
fine an action of Αχ on FQ. It follows from Proposition 5.1 that Spf FQ = (Spf F'Q) X.
On the other hand (cf. §4C)), the group Αχ C A* acts on E'Q, and Spf EQ =· (Spf E'Q) X.
i t · * . ψ .

Thus the homomorphism £"„ -» F n is consistent with the action of Λ . Since the bottom
row of the commutative diagram is an isomorphism, so is the upper row.
c) It remains to clear up the case η = 0, A = F [χ], ν = (χ). Suppose the elliptic
F. [x]-module X is defined by a homomorphism F [x] -> F j r j , where χ -» Σ. " α .τ1 + τ ,
with a, 41 0· Clearly, F Q = F S i x , α ι , . · · , a J _ Jl» and the universal deformation of the
elliptic module X has the form χ I—*x + Σ. ~\ a.r1 + r . Since En & Fn> i f suffices to
r
ζ= 1 ι U 0

show that a morphism Spf F Q -> Spf £„ induces a monomorphism of tangent s p a c e s . In

e
other words, we must prove that if β,, . . . , |6> ι F and the deformation
d-l (3—1

with basis F [x]/(f ) induces the trivial deformation of the divisible module X, then
βί — 0 for 1 < i < h — 1. Indeed, it follows from the triviality of the deformation of the
formal module X, that β- = 0 for 1 < ι' < h — 1. Let r € F be a point of order χ of
the module X. Then it follows from the triviality of the deformation X that
d-i

2 (at+h*)rPl+rPd=0.
i=h

and so Σ .", β.r^ = 0. Since r can assume p different values, it follows that
β{ = 0 for h < i < d - 1. D

Corollary. Suppose the conditions o'f Proposition 5.3 are satisfied. Then M. is a
smooth d-dimensional manifold. The morphism Μ j -> Spec A is smooth over Spec A —
V(I). If J C /, then the morphism Md, -> M^ is finite and flat.

D) Group actions. We set Μ = lim M.. Let 51 be the ring of adeles of k, and let
21, be the ring of adeles without the component at oo. (Thus 21 = 21 χ ^ and /4 =
lim Λ//.)
We define an action of the group GL(d, 2I.)/&* on Μ . Let 5 be a scheme over A,
and let X be an elliptic module over S of rank d together with a homomorphism φ:
{k/A) -> Mor(S, X) such that for any nonzero ideal / C A the restriction of φ to
(/ ~ /A) is a structure of level /. Let g 6 GL(d, 21 ) be a matrix with coefficients in
A We can regard g as an endomorphism of {k/A) · Its kernel P i s finite. It follows
from Proposition 4.4 that the divisor Η C P, which is equal to the sum of the divisors
φ(α), a £ Ρ, is an ,4-invariant group subscheme, and X/H is an elliptic A-module.
We define φχ: (k/A)d -> Mor(S, X/H) SO the diagram
578 V. G. DRINFEL D

(k/A)d ^ Mor(5, X)
4 I
(kfA)d ^ Mor(S, X/H)
is commutative.
It follows from Proposition 4.4 that for any / the restriction of φι to (l~l/A) is
a structure of level /. We obtain a left action on Μ of the subgroup of matrices in
GL(d, u.) with coefficients in A. Since the subgroup of nonzero elements of A acts
trivially on Μ , we obtain an action of GL(d, 21 )/k .
Let / C A be an ideal satisfying the conditions of Proposition 5.3, and let U, be
the kernel of the homomorphism GL(d, A) -» GL(d, A/l). Then tof = Uj\Md. Indeed, if
J C /, then the morphism

Mj X spec A (Spec /I — 1/ (/)) -> Mj χ S p e c χ (Spec A - V (J))

is a bundle with structure group U./U,. It follows from the normality of Λ1, that M, =

E) Congruence relations. Recall the construction of the induced ring [4]. Let G
be a topological group, Η C G a closed subgroup, and Β a discrete ring on which the
group Η acts continuously on the right. Consider the ring C of continuous functions
/: G -> Β such that f{gh) = /(g) · h for g e G and Λ € f/. This is called the induced
ring; G acts on it by the formula f(g) · (g ) = f(gg ). We shall say that Spec C is in-
duced by the scheme Spec B.
Let ν e Spec A, and let Λ1, . be the fiber of Μ over v. To each point of M.
there corresponds an elliptic /^-module, to which in turn corresponds a formal A -
module. Let WCM, , be the set of points for which the corresponding formal A -
module has height 1.

Proposition 5.5. 1) W is an open, affine, everywhere dense GL(d, 2ΙΛ-invariant

subset.
2) Let β C GL(d, k ) be a group of matrices (a.) such that a-,-0 for i > 1.
Let B' be the preimage of Β in GL{d, 21 ). Then the GL(d, ^i,)-scheme W induced
by any Β'°scheme W has the following properties'.
a) The matrices (« z --) e Β C B' for which \an\v~ 1 and the lower right corner
(i, j > l) coincides with the identity matrix acts trivially on Wrgd .
b) Let π be a simple element of A . The matrix

1.
^ I
acts on W , like the Frobenius of the field Α /(π).
red ' ' ν

Proof. Statement 1) is clear (the density of W follows from Proposition 5.4).

2) To each point w & W corresponds an elliptic module X and a homomorphism
(k /A ) -» X whose kernel is isomorphic to k /A . This defines a GL{d, 21 ,)-invariant
mapping of the set W -> GL{d, 21 ) / β ' . Let / C A be an ideal, 7 ^ 0 . The mapping
 ELLIPTIC MODULES 579

U,\W -> GL{d, 5Ι,)/β' is clearly continuous. Let W. be the preimage in UAW of the
image of the identity in U}\GL(d, %f)/B'. We set W° = lim Wj. Then W° is invariant
with respect to β', and W is induced by the scheme W . Every morphism W d~»W d
is uniquely defined by its action on the set of places. Statements a) and b) follow
from this. D

§6. Uniformization of modular manifolds

A) Three analogs of the upper half-plane. Let Κ be a local nonarchimedean field,
Ο C Κ its ring of integers, and π € 0 a prime element. Let d be a natural number.
1) "Analytic" analog. Let Pi/" be projective space, considered as a rigid
analytic space. The group GL{d, K) acts on Ρ „ " according to the formula
(g; (2i, . · · . Zdi) -> (Zi> · · · » Zd) · g""1·
Let Ω be a set of points of P K ~ not lying in any hyperplane defined over K. The
set Ω is a GL{d, K)-invariant subset of P^~ .
2) "Homogeneous" analog [5]. Let Sd = GL(rf, K)/GL{d, 0). The group GLU, K)
acts on the left on Κ , and so the elements of S can be interpreted as similarity
classes of free J-dimensional O-submodules in Κ . We introduce a metric ρ on S .
Let Μ C Kd, Ν C K rf , Μ W Ν & 0d, MD Ν D Λ , Ν £ Μ and Ν 7> nk~ lM. Let \M\
and |N} be the corresponding elements of S . We set p{{M \, \N\) = k.
A simplex is a subset Δ C S that satisfies one of the following equivalent con-
ditions: a) p{x, y) < 1 for x, y £ A; b) there exist submodules Λ1. C Κ (l < i < k)
s u c h t h a t M . t t O d , M.D M . . . f o r i < k , M , D n M . , a n d Δ = { { M . } , . . . , {M,\\. Thus
S is a simplicial scheme of codimension d — 1. The corresponding polyhedron is de-
noted 5 (R), and the set of points of S (R) with rational barycentric coordinates is
denoted Sd(Q).
3) "Topological" analog [9]. We fix a norm | | on K, and use the same symbol
to denote the extension of this norm to K. We set q = |π~ |. We call the norm ν on
Κ integral (rational) if for every χ £ Κ , χ ^ 0, we have log v(x) £ Ζ (respectively
log v{x) e Q). It can be shown that the set of norms on Κ up to similarity can be
identified with S (R); the class of rational norms corresponds to 5 (Q) and the class
of integral norms corresponds to 5 . Here is the construction. Let Κ 3 Μ , D Μ Ί D
Ο Μ, D πΜ j , α,ρ . . . , α , > 0 , Σ , α. = 1. Let ν • be the norm on K corresponding
to Μ.. To the point of S (R) with coordinates ( a . , . . . , a . ) corresponds
p the c l a s s
z
of norms ν = m a x 1< . ,{ z i d
ν \. The metric ρ on 5 extends to a metric on SS (R)
\
1<.<,{^
<

which we shall also denote by p: If v, and Vj a f e norms on Κ , then

Ρ ({νχ}, {va}) £• log, {sup ^ - . s l l P *\ .

I ^d v2 Kd vx )
If (^j, . . . , zj) e Ω , then the function (a., . . . , α^) μ-> (Σ. α.2:.| is a rational
norm on /C . This defines a GL{d, K)-invariant mapping λ: Ω -> S (Q). (It is easy
to check that λ is surjective.)

Proposition 6.1. 1) Let χ j , . . . , x^ e S and c e Q. Lei

580 V. G. DRINFEL'D

Then Xc is an open affine subset of Pz~ . // c. < c_, then X is in the interior
of X C .
2
2) Ω is an admissible open subset of Ρ J~ .

Proof. L e t η be a natural number, η > c. For each i ( l < i < k), let ν• be a norm
on Κ whose c l a s s i s x{t let C\ = {y e Κ \v{{y) = l\, and let Ρ . C C. be a finite
net. To each pair («, b),where a = (flj, . . . , aj) € Κ and b = ( 6 , , . . . , 6 ,) 6 K^,
there corresponds a rational function r? on P ^ ~ :

/"& (2^> . . · , 2^) =

Σ »* '
Let W C P^~ be the intersection of the domains of definition of the functions rf,
where a, b € Ρ., 1 < i < k. It i s e a s y to check that the s e t of functions T\.r\ (where
Up v^ Ρ ρ induces a closed imbedding of W into an affine s p a c e . Clearly Xc i s the
preimage of the polydisc of radius qc under this mapping. From this, 1) follows.
To deduce 2) from 1), it is sufficient to prove that if β i s a T a t e algebra over K,
and φ: Max Β -» P ^ ~ i s a morphism such that Im φ C Ω , then Im (λ ο φ) i s a
bounded subset of S (Q). Indeed, let φ be defined by functions t.£ Β (1 < ι: < d),
| i . | < 1. L e t

yp ( a l t . . . , ad) = inf
β
Σ·
t= l

Then ^ i s a continuous function on Κ , and ^ vanishes only at zero. L e t

'v(av .. ., α,*) = max \aAt

and let {v| be the corresponding element of S , and e = inf^,, ._ j»/>(a). Clearly
Im(A ο φ) i s contained in a ball with center \v\ and radius log f. D
Let S be the barycentric subdivision of S ( i . e . the points of S are the sim-
p l i c e s of S ; if Δ ρ . . . , Δ are simplices of S such that Aj 3 Δ 2 ^ •·· 3 Δ ^ ,
then {Aj, . · . , Δ^} i s a simplex of S .

Proposition 6.2. Let c € Q, 0 < c < 1. For eacA simplex Δ C S of codimension

k - 1,

/or
4* *€Δ 4*

/£ = λ ~ Η ν ^ ) . T/>e 5e/s (7^ generate a GL(d, K)-invariant admissible affine covering

of Ω^ with nerve ld. If cχ < c2, then t / J 1 C U&2.

Proof. We shall show that the s e t s V& form a covering of S (Q) with nerve 5 .
( T h i s i s sufficient by Proposition 6.1.)
L e t Φ C 5"^ be a simplex of dimension d - 1, and let Φ(0) be the corresponding
 ELLIPTIC MODULES 581

closed subset of S (Q). If χ e <£(Q) has barycentric coordinates (a^, . . . , aj), then it
is easy to check that the distance from χ to the zth point of Φ equals 1 - a.. If
V£H<I>(Q)^0, then AC<P. Indeed, if y ε Δ, ζ £ Φ and x€V£fl<E>(Q), then p(x, y)
< 1, p(z, x) < 1, and so p{y, z) < 2. Consequently Δ υ Φ is a simplex, and Δ C Φ. Let
Δ χ C Φ and Δ 2 C Φ be simplices of dimensions kl - 1 and k2 - 1, k2<kv with Δ- 75 Δ-.
Let the ;th point of Φ lie in Δ t - Δ 2 · If a point of Φ(Q) with coordinates (a ..., a )
lies in Vj Π ^ , then

.^ 3—
3 — cc νν ^ ,,
^ 1+c
a
a.· > and 2 j i > ! 4**
4*
This verifies the condition Σ,α. . = 1.
Denote by Λ1^ the set of the first k points of the simplex Φ. Clearly a point of
Φ(ζ)) with coordinates ( a ^ . . . , a r f ), where a. = (3 - c)/4z for i > 1, will lie in
Ofc_l V^ . It remains to show that the sets V^ cover S (Q). Let x£ Φ(Q) have
coordinates ( a p . . . , α^), α χ > a 2 > · · · > a r f . There exists « > 1 such that α >
(3 - c)/An and a. < (3 ~ c)/4 n for i > n. Then χ ε VCM . a
B) Factorization of rigid analytic spaces by the action of a discrete group. Let Κ
be a field which is complete with respect to a nonarchimedean absolute value.

Proposition 6.3. Let Β be a Tate algebra over K, and let the finite group G act
on B, Then Β is a Tate algebra and Β is finite over Β .

Proof. We represent β as a factor of the algebra of convergent power series R.

Let C be a tensor product of copies of R which correspond to elements of G. Then
G acts on C, and there exists a G-invariant epimorphism C -> B. It is easy to check
that C is a Tate algebra and C is finite over C . Therefore Β is finite over C ,
and Β is finite over C . It follows from this that β is a Tate algebra and β is
finite over β . D

Proposition 6.4. Let B^ and B2 be Tate algebras, and Max B2 C Max β 1 an

open affine subset. Let the finite group G act on By with Max B2 invariant with re-
spect to G. Then the morphism Max B2 -» Max Β l is an open imbedding. If Max Β 2 €
Max Bp then Max B^ € Max B^.

Proof. Clearly the morphism Max B2 -> Max Bl induces a one-to-one correspon-
dence between sets of points and isomorphisms of complete local rings. By a basis
theorem [8] this morphism is an open imbedding. If Max B2 € Max β 1 ? then there
exists a closed imbedding φ: Max Bl -> Max K\t v . . . , tf\ such that 0(Max B2) is
contained in the set |/.| < 1 - e, e > 0. Let s.. be the /th symmetric function in trans-
lations 0*(/z) by elements of G. The functions s.. define a finite morphism from
Max Β | into the identity polydisc which sends Max fi2 into the polydisc of radius 1 — e. •
If β is a Tate algebra under the action of a finite group G, we shall write
G\Max Β in place of Max BG. Let X be an affine analytic space under the action of
the finite group G, let Υ = d\x, and let Ζ be a separable space. Let φ: Χ -* Ζ be
a morphism invariant under the action of G. Then there exists a unique morphism φ:
Υ -> Ζ such that φ = φ ° π, where π: Χ -* Υ is the projection. Indeed, let Ζ = {JZ{
582 V. G. DRINFEL'D

be an admissible affine covering. Then the sets X. = φ~ (Ζ.) are affine and G-
invariant, and a finite number of sets X. cover X. We set Y. = G\X• = 7τ{Χ). The
sets Yi form an admissible covering Y. It remains to check that for each i there
exists an identity morphism φ: Υ . -> Ζ consistent with φ.
Definition. Let X be a separable rigid analytic space. The action of a group Γ
on X is called discrete if there exists a set / and an action of Γ on / and an ad-
missible affine covering of X by sets X. (z € /) for which the following conditions are
satisfied:
1) y (X.) = Xy{i) for γ € Γ and i e I.
2) Let i € I and I~\ - {γ ^ Γ\γ(ϊ) = i\. Then the group ΓΛ is finite.
3) If γ t Γ., then Χ.Π Xy{{) = 0 . If i e I and jel, then X ( y ) H X y ( . ) = 0 for
all but finitely many γ € Γ.
4) Let i € l . Then the covering of (J X ... by the sets X y /x is admissible.
In the situation described by the definition, let Υ = Γ\Χ, and let Υ • be the image
of X. in Y. It is easy to check that in a unique way Υ can be made into a separable
analytic space, and the mapping X -> Υ into a morphism such that the covering of Υ by
the sets Υ • is admissible and affine, and Υ coincides with Γ.\Χ. as a space. If φ
is a Γ-invariant morphism from X into the separable space Z, it is easy to check that
there exists a unique morphism φ: Υ -> Ζ consistent with φ. It follows from Proposi-
tion 6.2 that the discrete subgroup Γ C GL(d, K)/K , where Κ is a local nonarchi-
medea,n field, acts discretely on Ω .
C) Uniformization of m o d u l a r m a n i f o l d s . Inthis subsection Κ = k . I f the
i d e a l / C A s a t i s f i e s t h e c o n d i t i o n s o f P r o p o s i t i o n 5 . 3 ,w e s e t ) R . = { M , 0 k )
r
I I oo an
Indeed, as is clear from the proof of Proposition 5.3, the manifold %^ makes sense for
any nonzero ideal /, / φ.-A.
Let Ω , = GL{d, k)\{Qd χ GL{d, O . (Here GL(d, U.) is considered as a discrete
set.) We introduce a left action of the group GL{d, 2I,)/fC on Ω ,. Also, Ω is an
open-closed subset of Ω . In the same way we define S , S (R), etc.
A projective Λ-module Ρ of dimension d together with a norm ν on Ρ ® ^ k^
will be called a metrized Α-module of dimension d. Metrized Α-modules ( P j , u,) and
(P_, v2) w i H t»e called similar if there exists an isomorphism /: Ρ ^* Ρ _ such that ν,
and / ( i O are proportional. Let I C A be an ideal, Ι Φ Α, Ι Φ 0. A structure of level
I on a ^-dimensional metrized /I-module (P, v) is an isomorphism ι/r: (/~ /A) ~
/~ P/P. Clearly the set of similarity classes of ^-dimensional metrized /1-modules
with structure of level / can be identified with U\S (R).
Let (P, v) be a metrized Λ-module. We define a function ν on Ρ <8> k^/P by the
formula i/(«) = inf i^(f). Let x, y e (/" V^)rf> * ^ 0, V ^ 0· We set μ =
ΰ(φ(χ))/ϊ'(φ{γ)). Then μχ can be considered as a function on Uj\S (R).
Proposition 6.5. The subset Η C Uj\S (R) zs bounded if and only if the (unctions
μχ !„ are bounded for x, y e (l~ /A) , x^O, y^O.

Proof. The necessity is clear. We shall prove sufficiency. Let (P, v) be a

metrized ^-dimensional .-4-module, and let e,, . . . , e^ be a basis for /" P/P. Let
e. be an element in the class of e. which is smallest in norm, and let u be an element
 ELLIPTIC MODULES 583

of / lP which is smallest in norm. Let μχ <c for all χ and y. Then e! /\u\ < cv
where Cj depends only on c. We define a norm ν on Ρ ® k^ by the formula ν (Sfl.e.)
= max. |Λ.|. Then ν (u) > 1/c^ By Minkowski's lemma, the order of /" Ρ/1l.__lAei is
bounded by a constant depending only on c. On the other hand, it follows from Min-
kowski's lemma that p(\v\, \v \) is bounded by a constant depending only on c. (For
the definition of p, see §6A). D

Proposition 6.6. Let I C A be an ideal, / ^ 0, / φ A. Then \J^d = 5Hf. 77>zs

identification is consistent with the projection Μ, -> Mj for J C /, and also with the
action of the group GL(d, 21,).

Proof, a) In §3 we essentially obtained the identity ^/(&„<,) = ί^\Ω U^) con-

sistent with the projections and the action of GL{d, 51,). (That is, a point of Ω with
coordinates (z,, . . . , z,) corresponds to the lattice Σ , Az • with the natural structure
of level /.) It is easy to check that the mapping Ω (k^) -* 5Π/(^^,) induces a morphism
Ω -» Mt. Since this morphism is invariant with respect to U\, there is a corresponding
morphism φ: Uj\&d -* Jllf.
b) Let x& U,\Ω , y = φ(χ), and let 0 and 0 be the completions of the local
rings of the manifolds Uj\Q, and ^Kj at the points χ and y. We shall prove that φ :
0 -> 0 is an isomorphism. Let m CO be a maximal ideal, and let κ be a natural
y χ r y y >
number. There is an elliptic /4-module X with structure of level / over the ring
Ο /mn. Let / € Ο /Μ"[[ζ]] be a formal isomorphism from the additive /^-module into
X (cf. Proposition 1.2). Just as in §3, it can be verified that / is an entire function.
Let Onr be a maximal unramified extension of 0 , and let Γ be the set of zeros of
y y>
/ in the ring Onr/mn. Then Γ is an Λ-module, the homomorphism Γ -» Onr/m is
injective, and its image is a lattice. In this way we obtain a homomorphism Ο -*
Onr/mn that is invariant with respect to G a l ( 0 n r / 0 ). Thus there is constructed a
homomorphism φ: Οχ -> Ο such that φ*ψ - id. Since %l and ί/^ΧΩ are smooth
manifolds of the same dimension (cf. Proposition 5-4), it follows that φ is an isomor-
phism.
Lemma. Let Σ C Mj be an affine open subset. Then the image of Σ in U\S (Q)
is bounded.
Proof. It suffices to consider the case / = (β) with \β\ > 1. Let x, y e (l~ /A) ,
x ^ 0, y φ 0> let / χ be the function on ?llj equal to the quotient of the images of χ
and y in the universal elliptic Α-module; and let c = suPj,- s u P y l / x I* ^° e a c n e ^ e "
ment of Σ ^ ^ ) there corresponds a (^-dimensional lattice Γ C k together with an
2
isomorphism φ: (Γ1/Α)άι*Γ Γ / Γ . Let f(z) = ζ Π α ε Γ ( ΐ - z/a). Let φ (χ) be the
smallest (in modulus) element in the class of φ(χ), and let u be the smallest (in
modulus) element of Γ, u ^ 0. Then

β
1 d
Therefore \φ(χ)/φ{γ)\ < c for x, y £ (/" /A) , x ^ 0, y ^ 0. It remains only to apply
Proposition 6.5. •
584 V. G. DRINFELD

It follows from the lemma that φ~ (Σ) can be covered by a finite number of affine
sets. It follows from the fundamental theorem of [8] that φ is an isomorphism. D

§ 7 . Tate uniformization
Let 0 be a complete discrete normed ring over A, m CO a maximal ideal, Κ the
field of fractions of 0, and Ks the separable closure of K. In this section | j denotes
the norm on Κ and not on k .
Let φ: A -> K\r\ be an elliptic module of rank d. We shall say that φ has stable
reduction if there exists an elliptic Α-module φ': Α -» Κ\τ\ such that φ' α» φ, φ (A) C
0\T\ , and the reduction of φ modulo m is an elliptic Α-module (i.e. there exists an
ae A such that the degree of the reduction of φ'(a) is greater than 1). Clearly the
rank of the reduction φ is not greater than d. In case of equality we shall say that φ
has good reduction. Then φ has good reduction if and only if 9!» is obtained from some
elliptic A-module over Ο by an extension of the ring of scalars.

Proposition 7.1. Every elliptic Α-module over Κ has potentially stable reduction.

Proof. Let ν be an (additive) valuation in K. We set

Let y ρ . . . , yk be generators of the ring A, and let r = inf 1<: <k ω{φ(γ •))· If K' is a
finite extension of Κ such that r · e(K.'/K) e Ζ (where e(K'/Κ) is the index of ramifi-
cation), then φ has stable reduction over Κ . D
Remark. For modules of rank 1 the concepts of stable and good reduction coincide.
Let X be an elliptic Α-module over K. By a lattice in X we mean aprojective
Gal (Ks/K)-invariant Λ-submodule of finite type Γ C X{KS) such that a finite number
of elements of Γ are contained in every disc.

Proposition 7.2. The isomorphism classes of elliptic modules of rank d over Κ

are in one-to-one correspondence with the isomorphism classes of pairs (Χ, Γ), where
X is an elliptic Α-module over Κ of rank d (d, < d) with potentially good reduction,
and Γ is a lattice in X of dimension d — d..

Proof. 1) Just as in §3, one can construct an elliptic Α-module of rank d for the
pair (Χ, Γ).
2) Let φ: A --» 0\τ\ be an elliptic /4-module of rank d over K, and let the reduc-
tion of φ modulo m be an elliptic module of rank d.. The existence and uniqueness
of a pair (φ, u) follows from Proposition 5.2, where φ: A -* 0{r] is an elliptic A-
module over Ο of rank d, ,

for a€ A, a. £m, a.-*0.

Lemma, u is an analytic homomorphism.

Proof. It follows from the relation φ{α)η - uif/(a) that there exist k and s such that
 ELLIPTIC MODULES 585

p p
& > 1 , IfliI < max (I a / + 1 1 , . . . , \ai+k\, \at.x \ , ..., |a t -_ s | )
for i > s. Let c. = sup > j \a.\. Since β.-»0, we have c. < m a x 1 < . < j c j . _ . for i > s.
Let d. = c? . Then β1. < max ,„, ., d. ., d. < 1 for ζ > s and so d. -* 0. D
ζ ζ ζ — 1<7 55 1 ι —] ι ι
s
Let Γ be the kernel of the homomorphism u. Clearly, Γ C K . If γ e Γ, γ ^ 0,
then |y| > 1. Let «£/4, |β| > 1, be such that the image of a in 0 is not zero. We
set f - φ(α). Clearly /" (Γ) /Γ =* [Α/(β)] , and the kernel of the homomorphism
hie to U / ( a ) ] r f l . Therefore Γ//(Γ) e* [A/(a)]d~dl. If
ζ € Ks, \z\ > 1, then \f{z)\ = \z\ °° . It follows from this that Γ is a lattice of dimen-
sion d — d y
3) Thus we have obtained a one-to-one correspondence between the isomorphism
classes of elliptic Α-modules of rank d over Κ having stable reduction, and the iso-
morphism classes of pairs (Χ, Γ), where X is an elliptic Α-module over Κ of rank
d, {dj < d) with good reduction, and Γ is a lattice in X of dimension d — dy It re-
mains only to apply Galois descent. D

§8. Complex multiplication (d ~ l)

Theorem 1. The scheme Μ is the spectrum of the ring of integers of the maximum
abelian extension of k, completely split at <χ>. The action of 21,/& on Μ coincides
with the action in class field theory.

Proof. Let / C A be an ideal satisfying the conditions of Proposition 5-3. It fol-

lows from Proposition 7.1 that the morphism M{ -> Spec A is finite. It follows from
Proposition 5-4 that Mj is a smooth curve, and that the morphism M{ -> Spec A rami-
fies only over V(/). Let ν €. Spec A, v £ V(l), and let 77 be a prime element in A . It
follows from Proposition 5-5 that π acts on the fiber of M, over ν like the Frobenius
of the field Α /{π). Therefore each connected component of Mj is invariant with re-
spect to 21*. On the other hand, Μ KkJ) = M^kJ = U*/k* (cf. §3), and this identifica-
tion is consistent with the action of 21 /k . Therefore Μ is connected, the action of
21 ,/k* on Μ is exact, and 2I*/M = Spec A. It remains only to apply class field
theory. D

Corollary. Over any algebraically closed field over A there exist elliptic A-
modules of any rank.

^9. Compactification of modular surfaces (d = 2)

A) Construction of the boundary. Let X be an elliptic module over S. We denote
by X the bundle of projective lines over S obtained by joining to X an infinitely dis-
tant section. The multiplicative semigroup A acts on X.
The functor that associates to the scheme S over A the set of isomorphism
classes of elliptic Λ-modules X of rank 1 over S, together with a consistent struc-
ture of level / (for all / C A, I ^ 0) and homomorphism k -> Mor(5, X), is represented
by a scheme Ν .
The group B, consisting of matrices of the form (Q p with χ £ 21, and y e 21
acts on Ν . Let / C A be an ideal satisfying the conditions of Proposition 5.3.
586 V. G. DRINFELD

Let Vj be the group of matrices (ΐ •(), where x, y £A, x = 1 mod /, y = 0 mod / and
Nj = Vj\N . It is easy to check that the morphism N, -* M, is smooth. Let Νj be
x
1 . . . . "~"' 1
the completion of Nj along an infinitely distant
Λ Njsection. (We shall consider Ν j as a
scheme and not as a formal scheme.) Let Ν,ι C 'N,ι be the complement at the infinitely
Λ Λ
°"l ~1 1 · 1 ^2 2
distant section, and let Ν = lim N, and /V = lim N,. Let Λ1 and ΛΙ be the schemes
induces by the schemes Ν and Ν with respect to the imbedding Β C GL(2, ^,)/k .
Let / C A be an ideal satisfying the conditions of Proposition 5.3, and let M,2 =
Uj\M and Mj = U,\M . Let 0 be a complete discrete normed ring over A with field
of fractions K.

Proposition 9.1. 1) The morphism Spec Κ -> Μ. extends to a morphism Spec 0 ->
Mj if and only if the corresponding elliptic module over Κ has potentially good reduction.
2) There exists a unique GL(2, W·,)·invariant morphism s: Μ -* Μ2 that induces a
one-to-one mapping from the set of those morphisms Spec 0 -» Λ1. such that the pre~
* 2 7
image of Mj is Spec Κ into the set of those morphisms Spec Κ -* Μ, which do not ex-
tend to morphisms Spec Ο -> Μ,.
Proof. Assertion 1) is clear. Let a morphism Spec Κ -» Mj not extend to a morphism
Spec 0 -» Μ j . To this morphism corresponds an elliptic Α-module over Κ of rank 2, having
potentially bad reduction, with structure of level /. This is the same (cf. Proposition 7.2) as an
elliptic A-module X over Κ of rank 1 together with a one-dimensional lattice Γ in X such that
Gal(Ks/K) acts on Γ1Υ/Υ trivially and an epimorphism il~l/A)2 -» Γ1Υ/Υ. (By
definition, / " Γ is the set of points of X{KS) that fall in Γ under "multiplication"
by any element α £ /.) Every automorphism of the Α-module /~ Γ which is trivial on
/~ Γ/Γ is the identity. Therefore /"" Γ C Χ(Κ). The bijection 2) is thus constructed.
It is easy to check that it induced by a morphism Μ -> Μ . Ο

Corollary. Let ν £ S p e c Α, ν £ VU), and let Υ be the fiber of Mj over ν and Υ a

smooth compactification of Y, The completion of Υ along Υ — Υ is canonically iso-
morphic to the fiber Mj over v.

B) Compactification of M{.

Proposition 9 . 2 . Let X, and X2 be normal surfaces, π- Χ • -» S p e c A proper

morphisms, D . C X. closed subschemes finite over S p e c A, and φ: Χ.—D^^X^-
D 2 a finite morphism over S p e c A. Then φ extends.to a finite morphism X ^ -> X 2 ·

Proof. Let GCX, x X , be the graph of φ, and let G be its closure. Since G is
finite over Χ , - D , and X _ — D 2 , it follows that G is closed in (Xj - D j ) x X 2 and
Χ ι χ (X 2 - D 2 ) . Therefore G - G is finite over Spec A. It follows from this that the
projection G-. -> X, is a finite morphism and birational isomorphism, i.e. G * Χ γ •

Proposition 9.3. 1) Let I C A be an ideal satisfying the conditions of Proposition

? 2
exists a unique (up to isom
5.3. There exists a unique (up to isomorphism) smooth surface M{ containing M/ as
verywhere dense
an open everywhere dense set
set such
such that
th the morphism Mj -> Spec A extends to a proper
morphism Μ. -» Spec A, and M{ — Mj is finite over Spec A.
 ELLIPTIC MODULES 587

2) // / C /, then the projection Μ , -» M* extends to a finite morphism M, -> M,.

Set Μ = limM,. 77?e action of GL{2, 21,) o« Μ extends to an action on Μ .
* —2 ~? 2 ^2
3) 77? e completion of Mj along M. - M^ z's canonically isomorphic to M,. In
particular, the morphism M, -> S p e c Λ z s smooth over S p e c Λ — \/(/).

Proof. Let β e A , |a| > 1 and k = log j<at |. Let the image of a have the form
Ζ
Σ#.Γ in the universal elliptic module over M,.
α e
We set t = α,Ι^Ι / 2
£· ^ obtain a morphism φ: bA, -» Spec A[i]. It follows from
Proposition 9.1 that φ is a finite morphism. It is easy to check that φ is flat. On the
other hand, it is easy to construct a finite flat morphism π: Μ, -•*• Spec A[[l/V]] such
that φ ο s coincides with the composition MJ -» Spec A((l/t)) -» Spec A[t]. (Here
A((l/ t)) is the localization of A[[l/7]] at 1/t, and π' is obtained from π by change
of basis.) There arises a morphism

σ : Μ) -> Μ) X SpecA[t] Spec Α Ιί~-\) .

Lemma. Let X = {1, 2, 3}, and take the sets φ, X, \ 1, 2|, |2, 3i aW {2| ίο ^e open.
Let Ο χ be a sheaf of rings on X such that

W{{\, 2}, Ox)=A[[y]],

i
H°({3, 2},Ox)=A[y,y- ], H°({2}, 0x) =A((y))
The category of quasicoherent 0y-modules is equivalent to the category of A\y\-
modules.

Proof. 1) Let Μ be a module over A[y]. We set H°(U, M) = H°(U, 0y) (gt.r ι Μ.
Μ is a quasicoherent sheaf, and Μ —
f • Μ is a functor. It remains to prove that the func-
torial homomorphisms Μ -> Η (Χ, Μ) and Η (Χ, F) -> F (where F is a quasicoherent
sheaf) are isomorphisms.
2) 0 -> A[y] -* A[y, y~ ! ] χ A[[y]] -» /A((y)) -> 0 is an exact sequence of flat A[y| -mod-
ules. Therefore H°(X, Μ) = Μ and Hl{X, M) = 0.
3) The restriction of the homomorphism Η (X, F) -* F on {2, 31 is an isomoφhism.
4) If Supp F C{1\, then WU(X, F ) ~ F .
5) If H°(X,'F) = 0 , then F = 0. (This follows from 3) and 4).)
6) Let 0 -» Gl -* //°(X, F) -» F -* G ? -• 0 be an exact sequence. Then H Q (X, G ^ =
0, and so Gj = 0. Since f/^X, //"(X, F)) = 0, we have tf°(X, G 2 ) = 0. Therefore
G2 = 0 .
It remains to check that σ is an isomorphism. It is easy to check that deg φ •=
deg π. (It is necessary to use the Corollary to Proposition 9.1.) It follows from Propo-
sition 9· 1 that the fibers of σ consist of one point (and not necessarily reduced). On
the other hand, one can check that the restriction of σ to each connected component
of Mj is a closed imbedding. D

§10. Connection with automorphic forms (d = 2)

A) Etale cohomology of rigid analytic spaces. Let Κ be a field complete with
respect to a nonarchimedean absolute value, let Κ be the separable closure of K,
588 V. G. DRINFELD

and let η be a natural number such that l/n € K. Let X be a rigid analytic space
over K.
Definition. 1) The set of pairs (L , φ), where L i s an invertible sheaf over X and
φ: 0x~Ln, is denoted W^(X, μη)- (There is a group structure on H^iX, μ ).)
2) Ηϊμ, Ζ/(η))Μ Η° 8 . ά (χΓζ/(«)) and H^X, μη)Μ H°igJX, J£).
(On the right side of these equalities Z/(n) is the constant sheaf and μ is the sheaf
of nth roots of unity; this is a sheaf in the rigid topology.)

3) Hit (X (x) /C, μη) = lim Hlt (X <g) L, μ π ),

~t
s
where Κ C L C K and [L: K] < eo. In the same way we define #^ t (X <S> Ks, μη) and
H°^(X ® K*, Z/(«)). (Gal(/CVK) acts on all of these groups.)
"' 4) //It (X <g> Ks, Z/(n)) ^= /& (X (8) KS, μ«) ® μ; 1 ·
(Here μ~ i s a Gal(/CV/0-module homomoφhism from the group of nth roots of unity
of the field Κ into Z/(n).
Properties of ''Stale cohomologies".
1) If X = Υ , where Κ i s a protective scheme over K, then the "cohomology"
of X coincides with the cohomology of Y. (This follows from theorems of type GAGA;
cf. [11] and [12].)
2) There exists an exact sequence

0-+H0 (Χ, μ π ) — H° (X, 0 ^ H° (Χ, Οχ) -+ Η1 (Χ, μΒ) — Η1 (Χ, 0*χ) ^Η\Χ, Οχ).

(Here Ητ{Χ, 0 ^ ) is the rigid cohomology.)

3) Let the sets X. form an admissible open covering of X whose nerve has
dimension not greater than 1. There exists an exact sequence

0 -> H° (Χ, μη) - Π //° (Χό μ«) -> Π Η° (Χ, Π Χ/, μ«)

-> Ζ/1 (Χ, μπ) -> Π Ζ/1 (^νμ«) -> ΓΙ //ι (Χι Π Xy, μΒ).
t ΙΦΪ
4) If Χ , -• Χ 2 i s a finite e t a l e morphism of rigid analytic s p a c e s with Galois
group G whose order i s prime to n, then Η ( Χ 2 ? μ ) = Η (Χ^, μ^) .

Proposition 10.1. Let ζ. ε Κ, c. e K s ( l < i < k) and ce Ks. Let \z \ < \c\,
0 < \c.\ < \c\, and let \z.-z.\ > \c.\ for i £ j . Set X = \z\\z\ < \c\, \z - ζ .\ > \c.\\.
Then
H°{X®KS, Z/{n))=Z/(n), Hl(X®Ks, μη) ^ (Z/(n) ) \
(Gal{Ks/K) acts trivially.)

Proof, a) The circle of radius \c\ is the union of circles of radius | c j and sets
X. Since the circle i s connected, and the circles which intersect X and circles of
radius | c . | are connected, it follows that X is connected. Analogously, X ® L is
connected if L is a finite extension of K.
b) Hl{X, 0^·) = 0 since every divisor on X is principal. The set of rational func-
tions without poles on X is everywhere dense in Η (Χ, 0χ). Clearly, if / 6 Η (Χ, Οχ)
 ELLIPTIC MODULES 589

and s u p z e X |/(z) - l| < 1, then / = g ,

n
°
where g ε Η°(Χ, 0 χ ) . Therefore every element
of H°(X, 0 χ ) ® Z/{n) can be represented as a rational function. It is easy to check
that if / is a rational function whose divisor on P , is contained in one of the k small
circles or is outside of the large circle, then there exists c€ Κ such that

k l n
We obtain an epimorphism φ: (Z/(n)) ^H (X, μη)/[Κ*/{K*) ]. Let X. be a circle of
radius | c . | . We consider the homomorphism Η (Χ, μη) -> Η (Χρ μη). In order to prove that
1 η
φ is an isomorphism, it is sufficient to check that Η (Χ{, μη)/[Κ*/(Κ*) ] ^ Z/(n) for 1 <
m
i < k. This follows from the fact that z is not an ηύι power in K\z, z~ \ for η \ m, (In-
deed, if / = Γ ^ ^ / e K\z, z~l\ and fn = zm, then max y | « J = 1. We reduce the
equation fn = zm modulo the maximal ideal and obtain a contradiction.) Thus we ob-
tain an exact sequence

0 -> Κ*ΜΎ -> Η1 (Χ, μη) -> (Z/(n))k - * 0,

s
and so HHX® K, μ^) « Z/(n)*. D
B) Cohomology of Ω . Let C be an abelian group. A 1-cochain on a graph with
coefficients in C is an antisymmetric function on the set of oriented edges of the
graph with values in C. A 1-cochain is harmonic if the sum of its values on all edges
emanating from a single vertex of the graph is equal to 0.

Proposition 10.2. The group

H° (Ω2 ® k'm Z/(/i)) = Z/(n) · tf1 (Ω 2 (x) us1», μη)
is isomorphic to the group of harmonic l-cochains on S with coefficients in Z/(n).
(The isomorphism is consistent with the action of GL(2, k^); the action of
Galik^/k^) is trivial.)

Proof (cf. §6A). Let 0 < c .< 1, and let Aj C 5" 2 and Δ 2 C 5 2 be simplices. Let
F — ί/f Πί/^ , where F is isomorphic to the space X studied in Proposition 10.1.
Since F is absolutely irreducible and S (R) is connected, we see that Ω is con-
nected. We shall describe Η (F ® k^, μ ) in invariant terms. Let ν be an (additive)
valuation in k^ such that vik^) = Z. Let ^ C L C k^ with [L: A;^] < °°, and let /
be a holomorphic invertible function on F <8> L. There exists a function / : λ(Ρ) -> Q
such that / ο λ = ν ο f. The function / has the following properties:

1) / is linear on every edge intersecting A(p), and its "derivative" is an integer.

2) If χ 6 \(F) \\S , then the sum of its "derivatives" on directions emanating
from χ is equal to 0.
Every function on X(F) satisfying 1) and 2) is equal to / for some L and some
function / on F <8> L. A function / on F ® L corresponds to the zero class in
Η (F ® k^, μ ) if and only if the "derivative" of / on every edge is divisible by n. D
C) Cohomology of M2 <8> K^.

Proposition 10.3. Let Π be a representation of GL(2, k ) in the space of locally

constant functions Ρ (k^) -» Q, factored out by the constants, and let A _ be the space
590 V. G. DRINFELD

of parabolic automorphic forms on GL{2, k)\GL(2, 21) (cf. [10]). Let

K = H o m G L ( a , * o o ) (π, Ao).
Let I 4-p, and let W°° be a 2-dimensional Uadic representation of ^
00
such that there exists a nonsplit exact sequence 0 -> Qj -> W -» QX- l) -> 0. Then
l 2
H (M ® k^, Qj) j ~ V ® W°° (isomorphism of GL(2, U ) χ GalU^/^Vroew/a/es)
^ ^ f and
f
2
H°(M <8> k^, Qp is isomorphic to the space of locally constant functions &*\2I* -> Q.
a c
(GalU^A^) ^ trivially, and g e GL(2, 21 ) izcis according to the formula (gf)(x) -
f(x det g)).

Proof, a) Let / C Λ be an ideal, / ^ 0, \ Φ A. The images of the sets t/f and their
GL(2, 21 ,)-translates form an admissible open covering of Ω . Clearly, the stabilizer
in Uι of any vertex or edge of S2 is a p-group. Using properties 3) and 4) of the
"etale cohomologies", we obtain an isomorphism

H° {Jtl (8)feSo,Z/(/i)) - H° (Uj\S2 (R), Z/(/i))

and an exact sequence

0-+H1 {Uj\S2 (R), Zf(n))^Hl (Jl)® kl, Ζ/(/ι)) - * [Ζ/1 (Ω ® ^ , μη)\! ® μ^ 1 - > 0 ,

Η (Ω ® k^, μ ) coincides (cf. Proposition 10.2) with the group of harmonic 1-cochains
on S with coefficients in Z,/{n).
b) Let /„ C G a K ^ / ^ ) be the inertial group. The mapping

/oX// 1 ^!®^, Z/^-v/i^iZ/V^R), Z/(«))

defined by the formula (χ, σ) (-> σχ — χ is bilinear, and GalCfe^/^) is invariant.
/.//" and ju are isomorphic as Gal(^ s //fe )-modules. Therefore there is a homomorphism

[Η1 (Ω (χ) ^ , μ,,)]" 7 — Ζ/1 ( ί 7 / \ 5 2 (R), Z/(n)).

It is easy to check that this homomorphism associates to a (/,-invariant harmonic

2 · \ ^2
1-cochain on 5 the cohomology class of the corresponding cochain on Uj\S .
c) Let Κ be a complete archimedean field, Υ a smooth projective analytic curve
over K, and D C V a finite subspace. It is easy to show that
H° (Y, Z/(n» - H° (Y — D, Z/(/i)), Η1 (Υ ® ^ , μ η ) C Z/1 ((V — D) (g) /sL, μ«).

More precisely, if Υ - £> = UzffAf ^z ^ s a n a dmissible affine covering such that Υ{ for
each i €. Ν intersects only a finite number of sets of the covering, then

= U Ker [H1 ((Y - D) ® Λ^, μ η ) — /ί 1 (( U V/) ® ^STO, μ»)]

ΦΟ.Ν ΚΖΝ—Φ
(where Φ runs through the finite subsets of N). Therefore

tP (Mf (x) A:5*,, Z/(«)) =7f> ( i / 7 / 5 2 (R), Z/(/i)).

and // HM,2 <S) ks , μ ) c o n s i s t s of those elements of H 1 ^ , <S> k*\, μ ) for which the
1 oo ι γι I OG * γι

corresponding harmonic 1-form is finite modulo Uj (or, equivalently, parabolic).

 ELLIPTIC MODULES 591

d) Going to the limit, we obtain an isomorphism

H° (ΜΪ <g) kl, Q/) - H° (i/,\S~(R), Q/)

and an exact sequence
1 2 2
0 -> H (Uj\S 0 (R), Qt) -> //' (My (§l kl, Qt) -y Vj (x) Q, ( - 1) — 0

where Vj is the space of parabolic t/j-invariant harmonic 1-cochains on S with co-

efficients in Q. It is easy to check that the homomorphism U{ -* Η (Uj\S (R), Q) con-
structed in b) is an isomorphism. It follows from this that Η (Μ ® k^, Q^) ~ V <S> W°°,
where V = lirnj Vj.
e) It remains to show that V - V. For this it suffices to verify that the space of
harmonic 1-cochains on S with coefficients in Q is canonically isomorphic to
Hom(ll, Q). This isomorphism is constructed in the following way. Let χ and y be
adjacent points in S , and let Ρ be the set of points ζ € Ρ (k^) such that y is
between χ and z. If μ is a distribution on Ρ (k^) with values in Q such that
(οι,*, xM ~ 0> then the 1-cochain / which has the value Γρ μ on xy is harmonic. D
F
"ftoo'
J
V- r x y'
§11. Fundamental theorem
The group GalUVO χ GL(2, ^ acts on //*(AI 2 ®/^ s , Q z ). (The group GL(2, ^ ^
acts on Μ on the left, and on the cohomology on the right; we shall consider the
corresponding left action on the cohomology.) We set V <S> Q = © · ~V\, where V • is
an irreducible representation of GL(2, 51,). (It is proved in [10] that V • φ V . for i 4 /').
It follows from Proposition 10.3 that

where Wi is a representation of Gal(&"V&) in the space Q{. Let Vz· =

and let WY be the restriction of Wi on G a l U ^ A v ) . Let VY be the representation
contragredient to Vv..

Theorem 2. 1)
H°{M2®k\ Ql)=H°{Mi®ks, Qt).
(The group GL{2, 21,) acts on Μ by the homomorphism det: GL(2, 21,) -» 21* )
2) For any ζ £ Τ flnii 7^6 Spec A such that Vv is a representation of class \f
Wv. ''coincides" with Vv. {i.e. Ψν{ is unramified and L{s, WV) = L(s - l/2, Vp).

Proof, 1) In §9A) we defined an imbedding Μ -•> Μ consistent with the action

of the group β CGL{2, 21,). It follows from Proposition 10.3 that the corresponding
homomorphism
H° (M 2 <g> !&, Q,) -> ^° ( ^ Ο Λί., Qi)
is an isomorphism.
2) Let

L (s, K,°) = (1 - μ^-Τ 1 (1 - μ^-Γ1·

Just as in [14], it is proved that WV is unramified and each eigenvalue of the Frobenius
592 V. G. DRINFEL'D

Vl 1 l/2
(arithmetical) on WV coincides either with [C^q~ or with μ~ q~ . We shall show
l//l
that if one of the eigenvalues of the Frobenius coincides with μ~^ q~ , then another
is equal to μ~^ q~ . It is known that V. ~ V. for some j £ Τ. If the restriction of V.
to the center of the group GL(2, 21,) is the multiple character ω € det, then V. ~V{
<S> co~ (cf. [10]). It follows from properties of the cup product that one of the eigen-
1/ 0 2 s
values of the Frobenius on WY coincides with μ^~ \ Let se H (M ® k , QX s φ 0,
and gs = &)(det g)s for g € GL(2, 51.). Multiplication by s induces an automorphism of
Η (Μ ® k^, Q z ). It follows from assertion 1) and Theorem 1 that one of the eigenvalues
of the Frobenius on W? is μ^~/2ω(π), where π is a prime element of A . But ω{π) =

Remark. The action of GL(2, 21 ) in [14] is different from the action described in
§5D) on the outer automorphisms of GL(2, 21,). This explains the different formulations
of Theorem 2 and the corresponding theorem in [14].

Received 17/DEC/73

BIBLIOGRAPHY
1. I. M. Gel fand, M. I. Graev and I. I. Pjateckil-Sapiro, Generalized functions. Vol. 6:
Representation theory and automorphic functions,"Nauka", Moscow, 1966; English transl., Saunders,
Philadelphia, P a . , 1969. MR 36 #3725; 38 #2093.
2. D. Mumford, Abelian varieties, Tata Inst. Fund. Res. Studies in Math., no. 5, Oxford
Univ. Press, London, 1970. MR 44 #219.
3· , An analytic construction of degenerating curves over complete local rings,
Compositio Math. 24 (1972), 129-174.
4. I. I. Pjateckil-Sapiro, Induced rings and the reduction of fields of Abelian modular func-
tions, Izv. Akad. Nauk SSSR Ser. Mat. 34 (1970), 532-546 = Math. USSR Izv. 4 (1970), 536-550.
MR 44 #155.
5. F. Bruhat and J. Tits, Groupes reductifs sur un corps local, Inst. Hautes Etudes Sci.
Publ. Math. No. 41 (1972).
6. P. Deligne, Formes modulaires et representations l-adiques, Seminaire Bourbaki 1968/69,
Expose 355, Lecture Notes in Math., no. 179, Springer-Verlag, Berlin, and New York, 1971, pp.
139-185. MR 42 #7460.
7. Albrecht Frolich, Formal groups. Lecture Notes in Math., no. 74, Springer-Verlag, Berlin
and New York, 1968. MR 39 #4164.
8. Hans Grauert and Lothar Gerritzen, Die Azyklizitat der affinoiden Dberdeckungen, Global
Analysis (Papers in Honor of K. Kodaira), Univ. of Tokyo Press, Tokyo, 1969, pp. 159—184.
MR 41 #514.
9. O. Goldman and N. Iwahori, The space of p-adic norms, Acta Math. 109 (1963), 137-177.
MR 26 #2430.
10. Herve Jacquet and R. P. Langlands, Automorphic forms on GL(2), Lecture Notes in
Math., no. 114, Springer-Verlag, Berlin and New York, 1970.
11. Rienhardt Kiehl, Der Endlichkeitssatz fur eigentliche Abbildungen in der nichtarchi-
medischen Funktionentheorie, Invent. Math. 2 (1967), 191-214. MR 35 #1833.
12. , Theorem A und Theorem Β in der nichtarchimedischen Funktionentheorie,
Invent. Math. 2 (1967), 256-273. MR 35 #1834.
13. Jonathan Lubin and John Tate, Formal moduli for one-parameter formal Lie groups,
Bull. Soc. Math. France 94 (1966), 49-59. MR 39 #214.
14. I. I. Pjateckil-Sapiro, Ζ eta-functions of modular curves, preprint, Inst. Appl. Math.,
Moscow, 1972.
Translated by M. B. NATHANSON