Hida Elliptic Curves and Arithmetic Invariants
Hida Elliptic Curves and Arithmetic Invariants
Haruzo Hida
Elliptic Curves
and Arithmetic
Invariants
Springer Monographs in Mathematics
123
Haruzo Hida
Department of Mathematics
University of California
Los Angeles, CA, USA
ISSN 1439-7382
ISBN 978-1-4614-6656-7 ISBN 978-1-4614-6657-4 (eBook)
DOI 10.1007/978-1-4614-6657-4
Springer New York Heidelberg Dordrecht London
Library of Congress Control Number: 2013937713
Mathematics Subject Classification: 11F11, 11F25, 11G18, 11G15, 11F80, 11R23, 11F67, 11G05, 11F80
v
vi Preface
Once the algebraic part of critical L-values is well defined, we might want
to relate it to a new arithmetic object produced from the starting object. For
example, if we are given an imaginary quadratic field, from its reciprocity
law we can produce a Dirichlet L-function whose value at the critical point
0 is essentially the order of the class group of the imaginary quadratic field
(Dirichlet’s celebrated class number formula). Dirichlet’s formula is the iden-
tity between two arithmetic invariants: One is given by the value at 0 of the
L-function associated with the imaginary quadratic field, and the other is the
order of a finite abelian group specifically produced out of the integer ring of
the field.
Going further, we might want to search a p-adic analytic L-function having
the same value (up to a simple p-adic modification factor similar to an Euler
factor) at critical evaluation points. Again, we therefore get a p-adic period
and a p-adic regulator and an arithmetic object (now somehow p-adically
made) whose size is (often conjecturally) given by the value or the derivative
(possibly up to the p-adic regulator).
The p-adic modification factor rarely (but sometimes) vanishes at some in-
teger points (producing an exceptional zero of the p-adic L-function), making
the p-adic critical value close to an Archimedean noncritical value. The p-adic
derivative of the p-adic L-function is equal up to a factor L, similar to the
regulator, to the (algebraic part of) Archimedean L-value, getting another in-
variant, called the L-invariant. Mazur–Tate–Teitelbaum systematically stud-
ied the L although the value of the L was studied and even determined for
some specific L-functions earlier than their work.
One may look into the growth of L-values over points densely populated in
the domain of the L-function. Suppose that a p-adic L-function Lp is defined
on a unit disk of p-adic numbers. If its p-adic absolute values |Lp (s)|p are
bounded by the supremum p−μ for 0 ≤ μ ∈ Q, this μ is the μ-invariant
first studied by Iwasawa in the 1960s and 1970s. Iwasawa then defined his
λ-invariant to be the number of zeros of the p-adic L-function (counted with
multiplicity) in order to measure the growth of a p-adic Dirichlet L-function
and the growth of the p-part of the class numbers of the cyclotomic fields of
p-power roots of unity.
If we are given a compatible system of l-adic Galois representations, its
trace ap of the p-Frobenius element at unramified primes p outside l is al-
gebraic and independent of l. This ap itself is an invariant describing the
structure of the Galois group (and also the local Galois group at p) of the
splitting Galois extension of
the representation, and we may uniquely form
∞
an Archimedean L-function n=1 cn n−s with Euler product converging for
Re(s) 0 having cp = ap for primes p from it. Or we can study the field
generated by ap over the rational numbers Q, which is called the Hecke field
of the system if the system comes from an automorphic/modular form (as we
expect that every system comes from an appropriately chosen automorphic
Preface vii
form). Once a Hecke field is given, its degree over the rational numbers is a
most primitive example of the invariants attached to the system.
Often the definition of the invariant itself does not give us a straight way
of understanding it. So number theorists seek a more direct or theoretical
way to compute the invariant. This is done in many cases through heuristic
understandings of the invariant. For example, the critical/noncritical L-value
is supposed to compute the order of an arithmetically defined abelian group
(an arithmetic cohomology group, called a Selmer group) similarly to the
Dirichlet class number formula. If we could prove this heuristic, it would be a
generalization of Dirichlet’s class number formula. Some mathematicians have
tried to specify what kind of arithmetic cohomology group should appear this
way (i.e., the Bloch–Kato conjecture for general motives; more specifically,
the Birch–Swinnerton-Dyer conjecture if the L-function is associated with a
rational elliptic curve and the evaluation point of interest is s = 1). This is
an identification problem of two different invariants, one analytically defined
and the other arithmetically defined. Similarly, Shimura and Deligne have
studied the equality of the automorphic period and the geometric period.
All of these problems (which are central to present-day number theory) can
be summarized into a set of problems targeting an identity of two or more
different definitions/incarnations of the invariant.
Since miraculous (and possibly profound) formulas are scattered through-
out number theory, mathematicians working in this area tend to seek a new
one directly or indirectly relating different arithmetic invariants. Most of the
celebrated conjectures in this research area fall into this category out of the
desire to convince ourselves of the harmony of our universe. I am not at all
against the idea. However, before setting out to see the God-given harmony of
numbers, we would do better to worry about the nontriviality or triviality of
the invariant. For example, Ferrero–Washington proved the vanishing of the
(original) Iwasawa μ-invariant long ago in an elementary way. The vanishing
is equivalent to the nontriviality of the p-adic L-function modulo p. From their
result, if Iwasawa’s λ-invariant vanishes (i.e., is trivial) very often, we get the
infinity of regular primes (does the name indicate that regular primes must
be “regularly” found?). This type of banal outcome, though possibly deep, is
often more intriguing in its proof than in the proof of a nice formula connect-
ing far-reaching arithmetic objects. Furthermore, as the outcome is banal, one
may not be able to develop a conjectural formula from the problem (which
satisfies mathematicians’ aesthetic desires).
The critical L-values must be generically nonzero (with interesting and
rare exceptions) unless the vanishing is forced by some trivial reason, and its
first derivatives must be nonzero most of the time if the vanishing is forced.
This is basically telling us that normally we are working in an easy environ-
ment (without the need to worry about the triviality of the invariant). We
should be working in “regular” cases (of nonvanishing L-values) to produce
the formulas in most instances, rather than working against hostility, requiring
viii Preface
construction (which I found about 20 years ago) of the p-adic operator, which
connects directly with Archimedean operators and which requires only the
deformation theory of ordinary elliptic curves as a tool (and should be easily
generalized to the ordinary locus of Shimura varieties of PEL type).
In Chap. 2, without becoming too technical, we recall the geometric defi-
nition of elliptic modular forms (and moduli problems of elliptic curves over
fields) just for our use in Chap. 3 to state the main results.
Almost all the main themes of the book appear in Chap. 3, possibly with-
out a detailed proof. We prove the fast growth of the degree of Hecke fields for
non-CM analytic families of slope 0, the nonvanishing (in most cases) of the
L-invariant of the modular adjoint L-function at s = 1 (restricting ourselves
to simplest cases), and a sketch of a proof of the vanishing of the μ-invariant
of Katz p-adic L-functions. At the end, we state a principle of distinguishing
mod-p Shimura subvarieties of a Shimura variety as a subvariety stable under
Hecke correspondences (Chai–Oort’s rigidity principle, an example of prob-
lem A), which is heavily used in the proof of the vanishing of the μ-invariant.
This includes a brief outline of the theory of modular curves as a Shimura
variety (and the Igusa tower over it).
Chapter 4 is an introduction to functorial scheme theory (and tools we
need for fully proving some results illustrated in Chap. 3). Proofs of some the-
orems are given in a different way from other books for algebraic geometers
(which is often unilluminating for number theorists). Some ring-theoretic re-
sults and more geometric results are quoted from different sources. Chapter 5
is a continuation of Chap. 4 but gathers more classical geometric results for
varieties over a field. Readers who are familiar with functorial algebraic geom-
etry or who just want to know the results for problem B without going into
technicalities may skip these two chapters.
Chapter 6 is a detailed description of modular curves of individual levels
over general rings. In particular, it contains a detailed description of deforma-
tion coordinates of the modular curve around an ordinary elliptic curve with
complex multiplication. The study of the move of the deformation coordinate
under the natural action of the stabilizer of the origin of the deformation space
(carrying the CM elliptic curve) is a main tool to prove the rigidity principle.
In Chap. 7, we interpret the theory given in Chap. 6 in Shimura’s way as
the simplest of Shimura varieties. In addition, we complete and legitimize the
sketchy construction of p-adic invariant differential operators first given in
Chap. 1 (in Sect. 7.2.6). We also prove the irreducibility of the Igusa tower in
the way given in [PAF] in more down-to-earth terms (which is the basis of the
q-expansion principle essential in computing μ-invariants).
In Chap. 8, we generalize Sinnott’s method (reproduced in Chap. 1) of
proving the nonvanishing modulo p of Dirichlet L-values to Hecke L-values
via the theory of Shimura varieties.
Chapter 9 is devoted to the proof of the vanishing of the μ-invariant of
the p-adic Hecke L-functions (of each imaginary quadratic field) via a detailed
analysis of the q-expansion of Eisenstein series. The rigidity principle and the
x Preface
module L of finite type, we write L = L⊗Z Z = lim L/N L, L (Σ) = L⊗Z Z (Σ) ,
←−N
and L = L ⊗Z Z
(p) .
(p)
xiii
xiv Contents
References 427
Let p and l be primes. Studying the p-primary part pen of the “minus” part
h−
n of the class number hn of the cyclotomic field Q(μln ) is a classic problem.
When p = l, there is a formula by Iwasawa (cf. [ICF] Theorem 7.14):
Lp (1 − k, χω k ) = (1 − χω k (p)pk−1 )L(1 − k, χω k )
for all positive integers k. Iwasawa showed this function to be analytic of the
form Lp (s, χ) = Φχ (γ s −1) for a power series Φχ (T )
∈ Zp [[T ]], where γ = 1+p
∞
(p > 2) and (1 + p)s is given by the p-adic limit n=0 ns pn . The product
Φ(T ) = χ=1;χ(−1)=1 Φχ (T ) can be decomposed into pm φ(T ) in Zp [[T ]] with
p φ(T ). By Weierstrass’s preparation theorem, φ(T ) has finitely many zeros
αj inQp with multiplicity mj . By the class number formula, we conclude
λ = j mj and μ = m. See [EDM] Sect. 5 and [ICF] for more details.
Once a p-adic L-function s → Lp (γ s − 1) for a power series Lp (T ) in-
terpolating good L-values is given, we may ask what the μ-invariant and
λ-invariant of the power series Lp (T ) ∈ W [[T ]] are. Here we allow a more
general coefficient ring W , which is a complete discrete valuation ring over
Zp . In this book, we study the vanishing of the μ-invariant for some p-adic
L-functions [other than Lp (s, χ)]. The vanishing of Iwasawa’s μ-invariant is
nonvanishing modulo pZp [[T ]] of Φ(T ); thus, this is a nontriviality problem
for Φ(T ) modulo p.
When p = l, in 1978, Washington ([Ws]) proved that en is bounded inde-
pendently of n. The proof in the two cases p = l and p = l are philosophically
close, although technically there is a difference. Washington achieved this by
showing that Dirichlet L-values L(0, χ) do not vanish modulo p for almost
all odd Dirichlet characters χ of the l-power conductor. A new proof of these
facts later given by Sinnott [Sn1] and [Sn2] in the mid-1980s was found more
amenable to generalization. In Sect. 1.2, we describe an “automorphic” version
of the proof of Sinnott in the case of p = l.
In Chaps. 2 and 6, we recall the moduli theory of elliptic curves and the
theory of modular curves. Then in Sect. 1.3 in this chapter and Sect. 3.5 in
Chap. 3, and in Chaps. 8 and 9, we will discuss how to generalize the argument
applied to Dirichlet L-values to Hecke L-values of imaginary quadratic fields
and to more general L-values via the moduli theory. We fix an odd prime p
and denote by l another odd prime different from p.
1.1 Arithmetic Invariants in Iwasawa Theory 5
1.1.2 L-Invariant
d
Hp (X) = det(1 − F robp |gri (ρ)(−i) pi X) = (1 − αj X).
i j=1
and set e = |{j|βj = p}|. Then we define modification factors (which is con-
jectured to appear in front of the complex L-values divided by a period in the
evaluation formula of the conjectural p-adic L-function of ρ) by
d
d
E(ρ) = (1 − βj p−1 ) and E + (ρ) = (1 − βj p−1 ).
j=1 j=1,βj =p
1.1 Arithmetic Invariants in Iwasawa Theory 7
The number e is the number of the vanishing factors (1 − βj p−1 ), and e > 0
if ρ has eigenvalue 1 for the Frobenius element F robp at p. The complex
L-function of ρ is defined by L(s, ρ) = l El (l−s )−1 .
The L-function L(s, ρ) is conjectured to have a meromorphic continuation
to the whole complex plane, and if ρl does not have a subquotient isomorphic
to a power Nlk ((l) = l ∩ Z) for an integer k, the continuation must be analytic
everywhere. Moreover, we have a well-defined Γ -factor Γρ (s) (a product of
specific translates of Γ -functions; e.g., [HMI] Sect. 1.2.1) and a constant ε with
complex absolute value 1; they are supposed to satisfy a functional equation
of the form Γρ (w + 1 − s)L(s, ρ̃) = εN −s Γρ (s)L(s, ρ), where N is a positive
integer called the conductor of ρ and ρ̃ = tρ−1 .
We assume that ρ (that is, each ρl ) does not contain N k as a subquotient.
An integer m is called critical for ρ if Γρ (w + 1 − s) and Γρ (s) are both
finite at s = m. We assume that the value at 1 is critical for L(s, ρ) (this
condition should follow from the condition a1 > 0 ≥ ad conjecturally). Assume
that the compatible system comes in an algebraic way as a factor of the
étale cohomology of an algebraic variety (after twisting by the cyclotomic
character). As conjectured by Deligne [D6] after the earlier discovery of the
rationality/algebraicity of many automorphic L-values by Shimura in [Sh6],
[Sh7], [Sh8] and [Sh9]; see Sect. 8.1 in the text and [HMI] Sect. 1.2.1), we
×
expect to have L(1,ρ⊗ε)
c+ (ρ(1)) ∈ Q for a well-defined period c (ρ(1)) ∈ C
+
for
all finite order characters ε : Z× ×
p → μp∞ (Q). Let γ = 1 + p ∈ Zp so that
Zp
1 + pZp = γ . Then we should have the following (cf. [MzTT] and [Gr]):
Conjecture 1.2 Suppose that s = 1 is critical for ρ. Then there exist a power
series Φan (X) ∈ W [[X]] and a p-adic L-function Lan an 1−s
p (s, ρ) = Φρ (γ − 1)
interpolating L(1, ρ ⊗ ε) for p-power order characters ε such that
L(1, ρ ⊗ ε)
ρ (ε(γ) − 1) ∼ E(ρ ⊗ ε)
Φan
c+ (ρ(1))
Lan
p (s, ρ) L(1, ρ)
lim = Lan (ρ)E + (ρ) + ,
s→1 (s − 1)e c (ρ(1))
Sect. 1.2.1, [D6] and [H94]). To avoid technical details of identifying the ratio,
we just left it ambiguous, inserting “∼” in place of “=.” Thus, once c+ (ρ(1))
is fixed, the power series Φanρ is uniquely determined.
When Katz constructed his p-adic Hecke L-functions in [K2], as we will
see later, he astutely avoided this ambiguity of choosing c+ (ρ(1)) when ρ is an
induced representation of a Hecke character of a CM field, and conjecturally
his trick should also apply to general ρ (see [SGL] for some idea of how to
make the p-adic L-function unique). In any case, Φan ρ is unique up to nonzero
constant in T ; thus, the above conjecture determines Lan (ρ) uniquely.
When e > 0, by our definition of E, we find E(ρ) = 0; so, by the above
interpolation formula, we have Lan p (1, ρ) = 0; hence, in such a case, we say
that Lan
p (s, ρ) has an exceptional zero at s = 1. The exceptional zero appears
if ρ(F robp ) has eigenvalue 1, and the number e is the multiplicity of the
eigenvalue 1. Since e linear factors of E(ρ) vanish, we “naively” guessed that
the order of zero at s = 1 is actually e, though each of these factors cannot be
regarded as a special value at s = 1 of any analytic function dividing Lp (as
the power function n → pn for integers n cannot be interpolated p-adically).
×
Example 1.3 Start with a Dirichlet character
√ χ : (Z/N Z) → Q with
χ(−1) = −1. Then c(ρ
+
(1))
= 2πi for i = −1 (as known from Euler’s time).
If we suppose χ = −D· for a square-free positive
integer
D, the modifying
−D
factor vanishes at s = 1 if the Legendre symbol p equals 1 (⇔ (p) = pp
in OQ[ −D] with p = {x ∈ OQ[ −D] : |x|p < 1}). By a work of Kubota–
√ √
Leopoldt and Iwasawa (e.g., [LFE] Chaps. 3 and 4), we have a p-adic analytic
L-function Lan an 1−s
p (s, χ) = Φ (γ −1) for a power series Φan (X) ∈ Λ = W [[X]]
and γ = 1 + p such that for E(χN m ) = (1 − χ(p)pm−1 )
L(m, χ)
Lan an 1−m
p (m, χ) = Φ (γ − 1) = E(χN m )L(1 − m, χ) ∼ E(χN m )
(2πi)m
for all positive integers m as long as |nm − n|p < 1 for all n prime to p.
If we have an exceptional zero at 1, it appears that we lose the exact con-
nection of the p-adic L-value and the corresponding complex L-value. How-
ever, the conjecture says we can recover the complex L-value via an appro-
priate derivative of the p-adic L-function up to the factor Lan (χ). We may
χ
regard χ as a Galois character Gal(Q[μN ]/Q) = (Z/N Z)× − → {±1}, and we
have the exceptional zero if χ(F robp ) = 1. √
For our later use, for the class number h of Q[ −D], we write the generator
√
of ph as ; hence, ph = () for ∈ Q[ −D]. We then define logp (p) =
1 1
h logp () and logp (p) = h logp (). Here logp is the Iwasawa logarithm, which
is an analytic function defined by the power series t → ∞
n
n+1 (t−1)
n=1 (−1) n
on the open disk of radius 1 centered at 1 extended to the entire C× p by
logp (pα ) = 0 for all α ∈ Q. Thus, we have 0 = logp () = logp () + logp (),
and hence logp (p) = − logp (p). This number shows up as a factor of the L-
invariant for Lp (s, χ).
1.1 Arithmetic Invariants in Iwasawa Theory 9
Assuming p > 2, we put x = p expp (p logp (x)) for x ∈ C×
p (which is
the p-torsion-free projection of x). Here expp is the p-adic exponential map,
√ ∞ 1/p
n=0 n (t − 1) , which is
n
which is convergent at p logp (x), and p t =
again convergent at t = expp (p logp (x)) (see [LFE] Sect. 1.3). We have another
example studied in [MzTT]:
Example 1.4 If ρp is ordinary semistable, we can formulate a conjecture
similarly. Here is such an example. Start with a rational elliptic curve E/Q ,
which yields a compatible system ρE := {Tl E}l given by the l-adic Tate mod-
ule Tl E = limn E[ln ](Q) for E[ln ] = {x ∈ E(Q)|ln x = 0E } (see Sect. 6.2.2).
←−
Suppose that E mod p removed singular points is isomorphic to Gm/Fp (so,
p ∈ S). Then
Hp (X) = (1 − X)(1 − pX), Ep (X) = (1 − X), E(ρE ) = 0 and E + (ρ) = 1.
By the solution of the Shimura–Taniyama conjecture ∞by Wiles (see [GME]
exists an elliptic cusp form f = n=1 a(n, f )q n of weight
Sect. 5.2.4), there
∞
2 with L(s, f ) = n=1 a(n, f )n−s = L(s, E). This L-function has a p-adic
an
analog Lp (s, E) constructed by Mazur (see [LFE] Sect. 6.5). There exists
−1
G(ε )L(1,E,ε)
E (X) ∈ Λ such that ΦE (ε(γ) − 1) = E(ρE ⊗ ε)
Φan an
ΩE for all p-
×
power order characters ε : Zp → μp (Qp ); in other words, Lp (s, E) =
∞
an
Φan
E (γ
1−s
− 1). Here ΩE is a period of the nowhere vanishing differential of
E (see Sect. 2.1.2 for differentials over a field and Corollary 2.33 in Sect. 2.2.1
for nowhere vanishing differentials; see also Sect. 6.1.4). Thus, if F robp has
eigenvalue 1 on Tl E, the exceptional zero appears at s = 1, as in the case
of Dirichlet character. The F robp has eigenvalue 1 if and only if E has split
multiplicative reduction mod p.
The problem of L-invariant is to compute explicitly the L-invariant Lan (ρ)
without recourse to the p-adic L-function. The L-invariant in the above
examples was computed in the 1970s through 1990s, i.e., the cases where
ρ = χ = −D · and ρ = ρE for E with split multiplicative reduction. The
results are as follows:
Theorem 1.5 Let the notation and the assumption be as given above.
logp (q) logp (q)
1. We have Lan (χ) = = − for q ∈ Cp given by q = /
h
ordp (q) h
√
(p = (), p = ()) and the class number h of Q[ −D].
h
explained in [Gr] (10) (and if we evaluate the value at s = 0, the minus sign
disappears, as in [FG]).
Example 1.6 Starting with an ordinary p-adic Galois representation ρ :
Gal(Q/Q) → GLd (W ), there is a systematic way to create many Galois rep-
resentations whose eigenvalues of F robp contain 1. Indeed, let ρ act on the
d × d matrices Md (W ) by conjugation. We have
for the trace 0 space sld (W ), which is stable under the conjugation. We write
Ad(ρ) for this Galois representation sld (W ) x → ρ(σ)xρ(σ)−1 ∈ sld (W ).
Then the action of Ad(ρ)(F robp ) on sld (W ) has eigenvalue 1 with multiplicity
≥ d−1. However, it is easy to check that the system Ad(ρ) is often not critical
if d > 2 even if we start with geometric ρ. Thus, in this book, we study the
case where d = 2. Take a representation ρF : Gal(Q/F ) → GL2 (W ) associated
with a p-ordinary Hilbert–Hecke eigenform of “weight ≥ 2” (i.e., belonging to a
discrete series at ∞) over a totally real field F . We make Ad(ρF ) and consider
the induced representation ρQ = IndQ F Ad(ρF ) : Gal(Q/Q) → GLn (W ) for
n = 2[F : Q]. Then ρQ (F robp ) has eigenvalue 1 with multiplicity e > 0. Here
the number e is the number of prime factors of p in F . The system ρQ is
critical at s = 1. The L-invariant of ρQ is studied in [HMI] Sect. 3.4.
Recall the group scheme Gm = Spec(Z[t, t−1 ]) as a covariant functor from the
category of commutative rings (i.e., Z-algebras) into the category of groups.
We take it as a functor from commutative Z-algebras into the category of
abelian groups sending an algebra A to its multiplicative group. We have
another functor Ga sending A to the additive group A. We may identify
Ga (A) with the set of algebra homomorphisms from a polynomial ring Z[T ]
into A by sending ϕ ∈ Homalg (Z[T ], A) to its value ϕ(T ). In this sense, we
can write Ga = Spec(Z[T ]) and OGa = Z[T ].
Each Laurent polynomial φ(t) ∈ OGm := Z[t, t−1 ] (the affine ring of the
scheme Gm ) can be considered to be a function Gm (A) = A× a → φ(a) ∈
A = Ga (A). Thus, we may regard φ : Gm → Ga as a morphism of functors.
P (t)
Abusing the language, for any fraction φ(t) = Q(t) with P (t), Q(t) ∈ Z[t], we
regard φ : Gm → Ga , although the map is not well defined at a zero z of the
P (t)
denominator Q(t) [i.e., if Q(z) = 0]. If we can write φ(t) = Q(t) with Q(z) = 0
and P (z) = 0, we just simply put φ(z) = ∞ (and call z a pole of φ). Such a
function φ(t) is called a rational function of Gm . Let Q(t) for the field of all
rational functions on Gm , and define the stalk at 1 ∈ Gm (Fp ) by
P (t)
OGm ,1 = φ(t) = ∈ Q(t)P (t), Q(t) ∈ Z[t], Q(1) ≡ 0 mod pZ(p) ,
Q(t)
where Z(p) = ab a, b ∈ Z, p b . The ring Z(p) is a local ring (or more
specifically, a discrete valuation ring) with maximal ideal pZ(p) . Note that
OGm ,1 is the localization of the ring OGm at maximal ideal (p, t − 1); so
OGm ,1 = Z[t, t−1 ](p,t−1) (see [CRT] Sect. 4 for localization of rings).
Set 1 < N ∈ Z prime to p. We fix a Dirichlet character λ of (Z/N Z)×
i∞ ip
with λ(−1) = −1 and two embeddings C ← Q → Qp . We regard λ as having
values in any of the three fields. Consider a rational function:
∞
N
λ(a)ta
Φ(t) = Φλ (t) = n
λ(n)t = a=1
∈ Z[t, t−1 ](p,t−1) = OGm ,1 ,
n=1
1 − tN
12 1 Nontriviality of Arithmetic Invariants
where OGm ,1 is the stalk at the closed point (p, t − 1). Since the numerator
N
a=1 λ(a)t is divisible by (t − 1), the rational function Φ is finite at t = 1.
a
Then the boundedness of the p-primary part of h− n is clear from the theorem,
−
and that of hn = h+ h
n n follows from this; see [ICF] Theorem 5.34 (an analytic
proof) or [ICF] Theorem 10.9 (an algebraic proof via Kummer theory).
Write C(Zl , A) for the space of all continuous functions on the ring of l-
adic integers Zl with values in A. Continuity is with respect to the l-adic
profinite topology on Zl . Often we equip A with the discrete topology; hence,
φ ∈ C(Zl , A) is a function factoring through Z/ln Z for sufficiently large n.
Our automorphic proof of this theorem has the following steps.
(Step1) Hecke operators: We introduce a Hecke operator U (l) acting on ratio-
nal functions on Gm/Fp and functions on μl∞ . For any given algebraic
group, we can define Hecke operators as was first done by Mordell
and Hecke for GL(2)/Q . These operators are very useful in character-
izing algebraic/rational functions on the group and often have strong
integrality properties. Because of the use of such operators, we call
our method of the proof “automorphic.”
(Step2) Measure associated with Hecke eigenforms: Choose a sequence of
d
generators ζN in μN so that ζN = ζN/d for 0 < d|N (to make
things simple, we choose ζN = exp 2πi N ). Fix an isomorphism
Zl ∼= Zl (1) = ←limn μln given by μln ζlan ↔ a ∈ Z/ln Z. For a
− ×
Hecke eigenform f |U (l) = af with unit eigenvalue a ∈ Fp , step 2
is the construction of a measure df on Z× l with a+ln Zl df f (ζln )
a
1.2 Dirichlet L-Values 13
for the image ζlan of ζ a in μln . Here we note that Φ is a Hecke eigen-
form and that we will take f = Φ. Such a construction was given for
elliptic Hecke eigenforms in [Mz1] when l = p (see [LFE] Sect. 4.4 for
a version for p-adic Dirichlet L-functions), but we do this here for
rational functions on Gm and l = p. Here a measure μ on Zl with
values in Fp is an Fp -linear functional defined on the space C(Zl ; Fp )
of continuous functions: Zl → Fp . Thus, μ : C(Zl ; Fp ) → Fp is an
Fp -linear map. We often write μ(φ) = Zl φdμ.
×
(Step3) Evaluation formula: For a character χ : Γ = Z×l /μl−1 → Fp , we
prove
L(0, λχ−1 ) χdΦ = χdΨ for Ψ (ζ) = Φ(ζ ε ).
Z×
l Γ ε∈μl−1
Slightly more generally, we start with Gm over a p-adic integral local domain
B whose quotient field is K. This means that we regard the functor Gm as
defined over the category of B-algebras (changing the base ring appears to be a
trivial maneuver but has a strong impact on the structure of the corresponding
scheme; see Sect. 4.1).
Take an algebraic closure K of K. We suppose that a prime l is invertible
in B and μl∞ (K) ⊂ B (that is, all l-power roots of unity in K are contained
in B). Fix a prime l prime to the characteristic of K. Let K(Gm ) = K(t) be
the rational function field of Gm = Spec(B[t, t−1 ]) [i.e., K(t) is made up of
P (t)
Q(t) for all polynomials 0 = Q(t), P (t) ∈ B[t]]. Define a linear operator U (l)
acting on rational functions φ ∈ K(t) on Gm by
1
φ|U (l)(t) = φ(ζt1/l ). (U)
l
ζ∈μl
Then we have U (l2 ) = U (l)2 . Similarly, we may define U (ln ), which is equal
to U (l)n .
Expand φ into a Laurent series n
−∞ a(n, φ)tn . Then we have
1
φ|U (l)(t) = ζ n tn/l = a(nl, φ)tn ,
n
l n
ζ∈μl
because
0 if l n,
ζn =
ζ∈μl
l if l|n.
1.2 Dirichlet L-Values 15
where y = yGm . The boundedness of μ follows from the fact that μ(x) ∈ B
for all n and j; so we can forget about it. The condition (Dist1) is equivalent
to the following condition for all φ factoring through Gn :
16 1 Nontriviality of Arithmetic Invariants
φ(x) dμ = φdμ = φ(y) dμ for all n > m, (Dist2)
x x G y y
We now give the definition of the measure associated with each eigen-
function φ ∈ OGm ,1/B of U (l). Write D∞ (φ) for the polar divisor of φ (see
Sect. 2.1.5 for divisor and polar divisor); thus, its support |D∞ (φ)| is the finite
set given by {α ∈ K|φ(α) = ∞} for an algebraic closure K of K.
Lemma 1.12 Let 0 = φ ∈ OGm ,1/B with φ|U (l) = aφ for a ∈ B × . If l = p,
then we have |D∞ (φ)| = |D∞ (φ)l |, and φ(ζ) is in B for all ζ ∈ μl∞ (K).
Proof. We first show that φ(ζ) ∈ K. We have |D∞ (φ)| = |D∞ (φ|U (l))| (as
φ|U (l) = aφ). Note τ ∈ |D∞ (φ|U (l))| ⇒ ζτ 1/l ∈ |D∞ (φ)| for some ζ ∈ μl .
Thus, |D∞ (φ|U (l))| = |D∞ (φ)| ⊂ |D∞ (φ)l |. This shows |D∞ (φ)| = |D∞ (φ)l |.
n
In particular, if ζ ∈ μln (K) ∩ |D∞ (φ)|, 1 = ζ l ∈ |D∞ (φ)|, which is absurd.
Therefore, φ is finite at all ζ ∈ μl∞ (K). Applying the above argument to φ
mod mB (and replacing both K and B by B/mB ), we conclude φ(ζ) ∈ B.
This is well defined (by Lemma 1.12) independently of the choice of Z/ln Z
through which f factors, because the condition (Dist2) follows from
1.2 Dirichlet L-Values 17
al · φ(ζ y ) = φ|l · U (l)(ζ y ) = φ(ζ x )
x≡y mod lm
Remark 1.13 We may define φ|U (l) for any function φ : μl∞ (K) → B by
the same formula (U). Then if φ|U (l) = aφ for a function φ : μl∞ (K) → B
with a ∈ B × , we can define the associated measure dφ again by the same
formula (1.2.3).
∞ B = Zpm
Take [μN ] ⊂ Qp . Applying
the computation in Sect. 1.2.3 to φ = Φλ =
m=0 λ(m)t , we find χdΦ(t) = G(χ)(lλ(l))−n Φχ−1 λ (t), and evaluating
the result at t = 1, we get from (1.2.1) (applied to χ−1 λ in place of λ)
χdΦ = (lλ(l))−n G(χ)Φχ−1 λ (1) = (lλ(l))−n G(χ)L(0, χ−1 λ).
there exists 0 < N ∈ Z such that bj mod lN are distinct. Thus, the function
ζ → f (ζ a1 , . . . , ζ ar ) has only finitely many zeros in μl∞ ; so, f = 0 in F . Thus,
the map R0 → F is injective.
To conclude the assertion of (step 4), we need to make the following vari-
able change: Let A be the additive subgroup of Zl generated by μl−1 and take a
Z-basis I = {a1 , . . . , ar } of A. Take a complete set of representatives ε1 , . . . , εn
μl−1 /{±1}
for μl−1 /{±1} and write εj = i cij ai (cij ∈ Z). Put T = Gm =
−1
Spec(R) [resp., Y = Spec(R0 ) = Gm ] for R = Fp [t1 , t1 , . . . , tn , t−1
I
n ] [resp.,
R0 = Fp [y1 , y1−1 , . . . , yr , yr−1 ]]. Here tj is the variable on the factor Gm in-
dexed by εj . Consider the ring homomorphism π ∗ : R → R0 sending tj to
cij
i yi . This algebra homomorphism is the pullback of a group homomor-
phism π : Y → T given as follows: Note Y = Gm ⊗Z Zr (A = Zr via the
basis {ai }) and T := Spec(R) = Gm ⊗Z Zn . Regard Zr and Zn as row vector
modules. For the r × n matrix C = (cij ), C : Zr → Zn given by x → xC
id ⊗C
induces π : Y = Spec(R0 ) ∼ = Gm ⊗Z Zr −−−−→ Gm ⊗Z Zn ∼ = Spec(R) = T ;
hence, π is a morphism of algebraic groups. Since a1 , . . . , ar is a basis of A
and ε1 , . . . , εn generate A, the matrix C has rank r; thus, C : Zr → Zn is
× ×
injective. This shows that Ker(π : Y (Fp ) = (Fp )r → (Fp )n = T (Fp )) is finite.
Proof. Note
c
ζ εj = ζ i ai cij
⇔ tj |tj =ζ εj = yi ij |yi =ζ ai = π(tj )|yi =ζ ai .
i
Hint: This follows from Exercise 1.15 and the minimal polynomial of ζln :
n−1 n−1
X l (l−1) + X l (l−2) + · · · + 1 = 0 over Z.
For simplicity, we assume that λ has values in F×p . Then, applying the
Frobenius automorphism F (x) = xp , we find for χ = χj
pn pn
au
0= χ(x)dϕ(ax) = χ(u)Ψ (ζln )
Γ u
n
pn pn n −1 n
= χ (u)Ψ (ζlaup
n ) = χ (p ) χp dϕ(ax)
u Γ
1.3 CM Periods and L-Values 21
for all n. Thus, taking the trace from the field Fp [χj ] generated by the values
of χj to Fp [μl ], we find that
χj (u)Ψ (ζlau
nj ) = 0
nj
u∈χ−1
j (Fp [μl ])⊂Γ/Γ
l
nj −m nj
for all a ∈ Γ . The above sum only involves u ∈ Γ l /Γ l for a constant m
independent of j. Indeed, writing the order of the l-primary part of (Fp [μl ])×
as lm and taking nj so that nj ≥ 2m, we can identify the multiplicative group
nj −m nj
Γl /Γ l with the additive one Z/lm Z by Z/lm Z v → 1 + lnj −m v = u.
We can then write χj (u) = ζlmj for bj ∈ (Z/lm Z)× and ζlau
b v a v
n = ζln ζlm . Since
v mod lm
∞
= ζlbv
m λ(n)(ζlm z)n
v n=1
(b+n)v
= λ(n)z n ζlm
n≥1 v mod lm
= lm λ(n)z n ,
n>0,n≡−b mod lm
What are modular forms? In the simplest case of elliptic modular forms,
writing w = t (w1 , w2 ) for two linearly independent complex numbers [with
Im(z) > 0 (z = w 1 /w2 )],
aweight- k modular
form is a holomorphic function f
of w satisfying f ac db w = f (w) for ac db ∈ SL2 (Z) and f (aw) = a−k f (w)
for a ∈ C× . In other words, f is a “holomorphic” function on the set Lat
of lattices in C with f (aL) = a−k f (L) (see Sect. 2.3.4). If we regard f as a
function of z ∈ H = {z ∈ C| Im(z) > 0} by setting f (z) = f (2πi ( z1 )), the two
relations can be stated as a single one:
az + b
f = (cz + d)k f (z)
cz + d
for ac db ∈ SL2 (Z). This latter definition might be more familiar for some
of the readers. Here we put 2πi to define f (z) to make the q-expansion of f
rational if f is algebro-geometrically rational [since f (2πiw) = (2πi)−k f (w),
this is just the division by a power of (2πi)]. This division has to be done
because of exp : C/2πi(Z + Zz) ∼ = Gm (C)/q Z for q = exp(2πiz). We want to
prove algebraicity and integrality of the value f (w) when w is a basis of an
imaginary quadratic field (up to a canonical complex number, called the CM
period of the elliptic curve). This rationality/integrality result is due to some
mathematicians, notably, Hurwitz, Shimura, and Katz.
We now describe an algebraic interpretation of modular forms. Waiting
until Chaps. 2 and 6 for a more thorough treatment of the algebraization,
here we give an outline of how to define modular forms algebraically. Pick
two linearly independent numbers w = (w1 , w2 ) ∈ C2 . Writing u for the
variable on C, the quotient C/Lw of C by a lattice Lw = Zw1 + Zw2 gives
rise to a pair (E, ω) of elliptic curves and the differential ω = du of first kind
(nowhere vanishing differential). First of all, we note E(C) ∼ = C/Lw . Then we
embed E into P2 by u → (x(u), y(u), 1) ∈ P2 (C) for Weierstrass ℘-functions
(Sect. 2.3.4):
1 1 1
1 g2 g3
x(u) = ℘(u; Lw ) = 2 + − 2 = 2 + u2 + u4 + · · ·
u (u − l)2 l u 20 28
0=l∈Lw
−4
and y = dx
du , where g2 (w) = 60 0=l∈Lw l and g3 (w) = 140 0=l∈Lw l−6 .
Then the equation satisfied by x and y is y 2 = 4x3 −g2 x−g3 and ω = du = dx
y
24 1 Nontriviality of Arithmetic Invariants
(see Sect.
2.3.4). The pair w (or lattice Lw ) can be recovered by ω so that
wi = γi ω for a basis (γ1 , γ2 ) of the Betti homology group H1 (E(C), Z) =
π1top (E(C)) (see Sect. 2.3.3 for the topological fundamental group π1top (E)).
For a given (E, ω)/C , we write LE for its lattice. Here C/Lw is an additive
group, and this is the canonical group structure of E(C).
Conversely, start with a pair (E, ω)/A defined over a ring A made of an
elliptic curve (a smooth curve of genus 1 with a specific point 0 = 0E ∈
E(A)) and a nowhere vanishing differential ω (see Sect. 2.1.2 for a definition
of algebraic differentials over a field and Sect. 4.1.7 over rings/schemes). Then
take an algebraic parameter u around 0 so that ω = du. Write [0] for the
relative Cartier divisor given by 0 (see Sect. 2.1.5 for divisor and Sect. 6.1.2
for Cartier divisor). Since the line bundle L(m[0]) (made of meromorphic
functions having only pole at 0 of order at most m) is free of rank m if m > 0
(by the existence of ω), we can find x ∈ H 0 (E, L(2[0])) having a pole of order
2 whose Laurent expansion has its leading term u−2 . If 6−1 ∈ A, there is a
unique way of normalizing x and u so that y 2 = 4x3 − g2 x − g3 with y = dx du
for a unique pair (g2 = g2 (E, ω), g3 = g3 (E, ω)) ∈ A2 (see Sect. 2.2.2 for
Weierstrass’s computation of the equation). Since E/A is smooth, this pair
(g2 , g3 ) has to satisfy Δ = g23 − 27g32 ∈ A× (Proposition 2.37). This shows (see
Sect. 2.2.3)
{(g2 , g3 ) ∈ A2 |Δ ∈ A× }
1 to 1 and onto
[(E, ω)/A ] ↔
1 1 1 1
= HomALG (Z[ , g2 , g3 , ], A) = Spec(Z[ , g2 , g3 , ])(A).
6 Δ 6 Δ
Here HomALG denotes the set of algebra homomorphisms, the ring Z[ 16 ][g2 , g3 ]
is a polynomial ring with variables g2 , g3 , and the straight brackets [·] indi-
cates the set of isomorphism classes of the objects inside; thus, [(E, ω)/A ] =
{(E, ω)/A }/ ∼
=.
An important fact is that all these functions g2 , g3 and Δ have Fourier ex-
pansions in Z[ 16 ][[q]] for q = exp(2πiz). Indeed, g2 and g3 are rational multiple
of the following Eisenstein series for k = 4, 6:
(k − 1)! 1
2(2πi)k (mz + n)k
(m,n)∈Z2 \{(0,0)}
∞
1
= ζ(1 − k) + qn dk−1
2 n=1 0<d|n
1
= ζ(1 − k) + q mn (1.3.1)
2
(m,n)∈Z2 \{(0,0)}/Z× ,mn>0
for even integers 2 < k ∈ Z. This follows from the partial fraction expansion
of the cotangent function of z = log q
2πi ([EDM] (2.18) or [EEK] Chap. II):
1.3 CM Periods and L-Values 25
∞ ∞
"
1 1 1
+ + = π cot(πz) = πi 1 − 2 qn , (1.3.2)
z n=1 (z + n) (z + n) n=1
k
d k
and its derivatives by (2πi)−1 dz d
= q dq :
∞
∞
1 (−1)k (2πi)k k−1 n
= n q . (1.3.3)
n=−∞
(z + n)k (k − 1)! n=1
In particular, keeping in mind the fact that gk (z) = gk (2πi ( z1 )), we find
⎧ ⎫
∞ ⎨ ⎬ ∞
1 1 n3 q n
g2 (q) = + 20 d3 q n = + 20 ,
12 ⎩ ⎭ 12 1 − qn
n=1 0<d|n n=1
⎧ ⎫
7 ⎨ 5⎬ n
∞ ∞
1 1 7 n5 q n
g3 (q) = − + d q =− + , (1.3.4)
216 3 n=1 ⎩ ⎭ 216 3 n=1 1 − q n
0<d|n
∞
Δ(q) = q (1 − q n )24 .
n=1
To obtain the product expansion of Δ, one needs to work a little more (see
[EEK], IV (36) and [IAT] (4.6.1)).
We can then think of the Tate curve Tate(q) defined over the power series
ring Z[ 16 ][[q]] by the (affine) equation y 2 = 4x3 − g2 (q)x − g3 (q). As shown
by Tate (Sect. 2.4.2), we have Tate(q)(Qp [[q]]) ⊃ (Qp [[q]])× , and therefore we
get a natural inclusion φN,can : μN (Qp ) → Tate(q)[N ], which comes from a
unique morphism of finite flat group schemes (see Sect. 4.3 for finite flat group
schemes). A Tate curve also has a canonical differential ωcan = dx y , which is a
unique (up to scalar multiple) nowhere vanishing differential on Tate(q) (see
Sect. 6.1.4). A Tate curve is not an elliptic curve over Z[ 16 ][[q]] but is an elliptic
curve over Z[ 16 ]((q)) := Z[ 16 ][[q]][q −1 ] because the nonunit q is a factor of the
discriminant Δ(q) of the elliptic curve.
Let B be a Z[ 16 ]-algebra. For general elliptic curve E over a B-algebra
R, a level Γ1 (N ) structure is an embedding φN : μN → E[N ] of finite flat
group schemes (i.e., a group homomorphism of functors inducing an embed-
ding of group schemes we describe later in Sect. 4.3). For the first reading,
if the reader is not familiar with the theory of group schemes, just assume
N = 1 to forget about φN . If the triple (E, φN , ω) is all defined over R,
we call (E, φN , ω)/R a test object over R. The elements g2 , g3 , Δ in Z[g2 , g3 ]
are functions of pairs (E, ω)/R of an elliptic curve E and a differential ω de-
fined over R with values in R and can be considered as a function of test
objects (E, φN , ω) disregarding φN . This motivates the following algebraic
definition/theorem (see Theorem 2.48) of B-integral elliptic modular forms of
level Γ1 (N ) as functions of test objects (E, φN : μN → E[N ], ω) satisfying
26 1 Nontriviality of Arithmetic Invariants
where (E, φN , ω)/R ⊗R,ϕ R means that we regard (E, φN , ω)/R as de-
fined over R via ϕ [so, for example, if N = 1, (E, ω) ⊗R,ϕ R is defined
by the equation y 2 = 4x3 − ϕ(g2 (E, ω))x − ϕ(g3 (E, ω))].
(G3) f ((E, φN , a · ω)/R ) = a−k f (E, φN , ω) for a ∈ R× .
(G4) f (q) = f ((Tate(q), φN , ωcan )/R[[q1/d ]][q−1/d ] ) ∈ R[[q 1/d ]] for any level N -
structures φN on Tate(q) defined over R[[q 1/d ]][q −1/d ] for any factor
d|N . In particular, we have
In 1897, Hurwitz studied in [Hz1] and [Hz2] an analog of the Riemann zeta
function: 1
L(4k) =
(a + bi)4k
a+bi∈Z[i]−{0}
for Gaussian integers Z[i] and showed that for a positive integer k,
1
L(4k) dx dx
∈ Q (Ω∞ = 2 √ = : CM period of the lemniscate).
4k
Ω∞ 0 1 − x4 γ y
Here λ = λ4k is a Hecke ideal character of Q[i], with λ4k ((α)) = α−4k and
L(4k) = 4L(0, λ), and a runs over all nonzero ideals of the Gaussian integer
ring Z[i].
The plane affine curve G defined by y 2 = 1 − x4 was first studied by
Gauss around the same time when he finished writing his first major treatise,
Disquisitiones Arithmeticae, in 1798 (he seems to have claimed that he could
extend his theory of drawing a regular N -gon in a circle to this lemniscate
or more general curves, which means, if true, he knew how to compute N
section/torsion points of G[N ]). This Gauss’s lemniscate is equivalent to Y 2 =
(4+4i)y (1+i)x+1+i
4X 3 − 4X by Y = ((i−1)x+(1−i)) 2 and X = (i−1)x+1−i ; hence, it is an elliptic
which is a function of lattices satisfying E2k (αL) = α−2k E2k (L). The quo-
tient C/L gives rise to an elliptic curve X(L) ⊂ topP2 by Weierstrass theory.
Since we can recover the lattice L as { γ du|γ ∈ π1 (X(L))} out of the pair
(X(L), du) for the variable u on C, we can think of E2k as a function of the
pairs (E, ω) of an elliptic curve E and a nowhere vanishing differential ω sat-
isfying E2k (E, αω) = α−2k E2k (E, ω). We have E4 (E, ω) = 120 1
g2 (E, ω) and
1
E6 (E, ω) = 280 g3 (E, ω). Since E2k is a rational isobaric polynomial of g2 and
g3 , E2k is a modular form f of weight 2k.
Weierstrass’s function has the following Laurent expansion:
xL (u) = u−2 + 6E4 (E, ω)u2 + · · · + 2(2k + 1)E2k+2 (E, ω)u2k + · · · . (1.3.5)
Taking the derivative, we get
yL (u) = −2u−3 +12E4 (E, ω)u+· · ·+4k(2k+1)E2k (E, ω)u2k−1 +· · · . (1.3.6)
2
Plugging these into the relation yL = 4x3L − g2 (E, ω)xL − g3 (E, ω), we get
a relation of E2k (noted by Hurwitz but known to Weierstrass earlier; see
[EDM] Sect.10.2). Solving the relation, we can compute the isobaric polyno-
mial Φ2k (X, Y ) ∈ Q[X, Y ], with E2k = Φ2k (g2 , g3 ). See [EDM] Sect. 10 for
more of these relations.
Returning to Gauss’s curve (E, ωE ) defined by y 2 = 4x3 − 4x, we have
LE = Z[i] · Ω∞ . Thus,
−2k −2k
E2k (E, ΩE ) = E2k (Z[i] · Ω∞ ) = Ω∞ E2k (Zi + Z) = Ω∞ E2k (i),
where at the extreme right, we regard E2k as a function on H by E2k (z) =
E2k (2πi(Zz + Z)). On the other hand,
E2k (E, ωE ) = Φ2k (g2 (E, ωE ), g3 (E, ωE )) = Φ2k (4, 0) ∈ Q.
Note that E2k (i) = L(0, λ2k ). Thus, we get the following result:
1.3 CM Periods and L-Values 29
L(0,λ4k )
Theorem 1.22 We have Ω∞4k ∈ Q for all 0 < k ∈ Z.
See Hurwitz’s paper [Hz1] and [Hz2] for numerical values of L(0,λ Ω∞
4k )
4k and a
generalization of the von Staut theorem (an integrality result) for these num-
bers.
√ curve defined by y = x − 1 that
2 3
We can apply the same technique to the
has complex multiplication by the field Q[ −3], as a cubic root of unity ζ acts
on the curve by (x, y) → (ζx, y). See [H81a] Sect. 8 for numerical examples in
this cubic root case and [EDM] Sect. 13 for more CM elliptic curves.
This is an obvious analog of the rationality (and recurrence relations) of
Riemann zeta values by Euler/Bernoulli (see [EDM] Sect. 11):
∞
ζ(2k) 1 1
2k
= ∈ Q.
(2πi) (2πi) n=1 n2k
2k
*
Indeed, 2πi is the integral |z|=1 dz
z of the invariant differential of the multi-
plicative group Gm/Z sending a ring A to its multiplicative group A× , and
∼
also note Gm (C) = C× ←−− C/2πiZ, while E(C) ∼ = C/Ω∞ · Z[i].
exp
The rationality/integrality results of Hurwitz is now generalized to all
imaginary quadratic field M and L-values of the form L(0, λk,j ) with integers
j < 0 ≤ k for characters of the form λk,j (α) = αk αj by the effort of many
mathematicians, notably, Shimura and Katz, and the value is algebraic up to a
power of the period Ω∞ (E, ω) and a power of 2πi. Here Ω∞ (E, ω) is given by
γ
ω for a pair (E, ω) defined over a number field with complex multiplication
by the integer ring of M . Elementary treatments of this generalization can be
found [EDM] Chap. IV and [EEK].
We can further ask
(Q1) Is there a good expression by an L-value of the value of a non-Eisenstein
series f ∈ Gk (1; Q) at (E, ωE ) (or at more general elliptic curves with
complex multiplication)?
(Q2) What happens if we evaluate a modular form in Gk (1; Q) at a rational
elliptic curve (E, ω)/Q without complex multiplication?
To get a reasonable answer, we need to assume that the modular form in ques-
tion is an eigenvector of all Hecke operators (as Eisenstein series are examples
of such eigenforms). Under this assumption, as for (Q1), there are good an-
swers by Waldspurger [Wa] (see [H10b] for an elementary treatment) and also
by Gross–Zagier [GZ] (which promoted much progress toward solution of the
Birch–Swinnerton-Dyer conjecture). As for (Q2), not much is known yet.
not. Take a p-adic algebra B as a base ring. As we will learn in more detail in
Sect. 7.2.4, the space of B-integral p-adic modular form V (B) = VΓ1 (N )/B is
a collection of rules f assigning a value f ((E, φN , φp : μp∞ → E[p∞ ])/R ) ∈ R
for p-adic B-algebras R satisfying the following conditions (V0–3) (where φp
is a group homomorphism of functors inducing an embedding of Barsotti–Tate
groups we describe later in Sect. 4.3.7):
(V0) f assigns a value f ((E, φN , φp )/R ) ∈ R to each test object
(E, φN , φp : μp∞ → E[p∞ ])/R
defined over a p-adic B-algebra R. Here R is also a variable.
(V1) The value f ((E, φN , φp )/R ) ∈ R depends only on the isomorphism class
of (E, φN , φp )/R .
(V2) If ϕ : R → R is an B-algebra homomorphism continuous under the
p-adic topology, we have
f ((E, φN , φp )R ⊗ R ) = ϕ(f ((E, φN , φp )/R )).
(V3) f (q) = f ((Tate(q), zφp,can , φN )/R[[q1/d ]][q−1/d ] ) ∈ R[[q 1/d ]] for any level
N -structures φN on Tate(q) over R[[q 1/d ]][q −1/d ] for 0 < d|N and
z ∈ Z× p . Again, we note f (q) = f ((Tate(q), φp,can , φN,can )/R[[q]][q−1 ] ) ∈
R[[q]].
Here we have simply written φp,can for
φp∞ ,can := lim φpn ,can : μp∞ → Tate(q)[p∞ ].
−→n
We note that VΓ1 (N )/B = V (B) is a p-adic B-algebra, as we see in Sect. 7.2.4.
Replacing (V3) by the following condition:
(s) f (q) = f ((Tate(q), φp,can , φN )/R[[q1/d ]][q−1/d ] ) ∈ q 1/d R[[q 1/d ]] for any level
structures φN on Tate(q) defined over R[[q 1/d ]][q −1/d ] for 0 < d|N ,
we define the subspace of p-adic cusp forms VΓcusp1 (N )/B
.
As we learn in Sect. 4.4.3, the knowledge of μp∞ /Zp = limn μpn /Zp is
−→
equivalent to the knowledge of the formal group G m/Z = Spf(Z[t, t−1 ])
p
for Z[t, −1 n
t−1 ] := limn Zp [t, t ]/(t − 1)) (see Example 4.79 and Sect. 4.4
p
←−
for the functor “Spf”). Then φp : μp∞ → E induces an identification
φp : G
m ∼=E = lim E[pn ]◦ with the formal completion E along the identity
←−n
of E (in short, the formal group of E). Since G m has a canonical “invari-
ant” differential t , φp induces a nowhere vanishing differential ωp = φp,∗ dt
dt
t .
Thus, f ∈ Gk (Γ1 (N pn ); B) can be regarded as a p-adic modular form by
f ((E, φN , φp : μp∞ → E[p∞ ])/R ) = f (E, φN , φp |μpn , φp,∗ dt
t ) ∈ R, and we
have a canonical B-linear map Gk (Γ1 (N pn ); B) → V (B) = VΓ1 (N )/B . The
following fact is called the q-expansion principle based on these two facts
(as we will see later): the irreducibility of the Igusa tower over Fp and the
existence of the Tate curve (see Sects. 7.2 and 2.4):
1.3 CM Periods and L-Values 31
We suppose further p splits into a product of two primes pp in O
(so, nas
∞
we will see later, Op = O ⊗Z Zp = O ⊗Z Zp = Op ). Let E[p ] = n E[p ].
32 1 Nontriviality of Arithmetic Invariants
Then we may assume that E[p∞ ]/V is isomorphic to μp∞ after extending
scalars to the strict henselization W ⊂ Q of V (see Sect. 4.3.9); hence, we
have φ = φE : μp∞ ∼ = E[p∞ ]. If p = pp in O but not in O (so, O is an
order of conductor divisible by p), we just assume that E has an isogeny
π
E − → E defined over a finite extension W ⊂ Q of W such that E has
complex multiplication by O and Ker(π) ∩ E [p∞ ] = 0. Under this setting,
φ : μp∞ ∼= E [p∞ ] induces φ = φE = π ◦ φE : μp∞ → E[p∞ ]. We study in
deoth later the sheaf ΩE/W of differentials on the scheme E (see Sect. 6.1.4).
An important point is that its global sections H 0 (E, ΩE/W ) are a free W -
module of rank 1. Pick ω ∈ H 0 (E, ΩE/W ) so that H 0 (E, ΩE/W ) = W ω.
As described above, we fix an embedding φp : μp∞ /W → E[p∞ ]/W and
an isomorphism E(C) = C/i∞ (a) for a fractional O -ideal a. In this case, we
write E = E(a). Thus, we may assume that a = Z + Zz0 (z0 ∈ M × ∩ H).
Note W ∼ = lim W/pn W, and set W = limn W /pn W , and identify W with
←−n ←−
the closure W of ip (W ) in Cp . Then we have two numbers Ω∞ ∈ C× and
×
Ωp ∈ W such that
dt
ω = Ω∞ du = Ωp φp,∗ .
t
The following result is plainly a generalization of Theorem 1.22:
Theorem 1.23 Suppose that p = pp with p = p in O. Let f = f∞ ∈
Gk (Γ1 (N pn )) be a W-integral form and let W/W be a finite extension of W,
which is discrete valuation ring. Write fp ∈ VΓ1 (N ) (W ) as the corresponding
p-adic modular form coming from f∞ . If (E(a), ω)/W has complex multipli-
cation by an order O of O, we have
Here the identity (1) is the base of Shimura’s rationality theory of special
values of modular/automorphic forms (cf. [EDM] Sect. 12) and (2) is a key to
the p-adic version by Katz.
Remark 1.24 Replacing W by its quotient field, the assertion of the theorem
is valid, for an obvious reason, also for f = hg if h ∈ Gk+k (Γ1 (N pn ); W) and
g ∈ Gk (Γ1 (N pn ); W) with g(E(a), φN , du)) = 0.
Remark 1.25 To have a well-defined p-adic period Ωp , a key point here is
to have an embedding φp : μp∞ → E[p∞ ] of Barsotti–Tate groups for a CM
elliptic curve E not E having multiplication by the full integer ring O. We
will scrutinize φp for E(C) = C/i∞ (a) for a lattice a in M with ap not equal
to Op in Sect. 6.4.5. Then the above theorem is valid over a finite extension
W /W [over which E(a) is defined]. The minimum ring W is given by W[μpn ]
for n depending on how far ap differs from Op (see Sects. 6.3.4 and 6.4). So
we state the result hereafter without assuming Op = Op .
1.3 CM Periods and L-Values 33
Shimura studied the effect on modular forms of the following invariant differ-
ential operators on the upper half-complex plane H indexed by k ∈ Z:
1 ∂ k
δk = √ + √ for z = x + iy ∈ H and δkr = δk+2r−2 · · · δk ,
2π −1 ∂z 2y −1
(1.3.7)
where r ∈ Z with r ≥ 0. For some other properties of these operators not
stated here, see [LFE] 10.1. Here are easy identities:
Exercise 1.26 Show the following formulas:
1. δk+l (f g) = gδk f + f δl g.
2. For a holomorphic function f : H → C, δkr (f |kα) = (δkr f )|k+2r α if
f |k α(z) = det(α)k/2 f (α(z))(cz + d)−k for α = ac db ∈ GL2 (R) with
det(α) > 0.
By the exercise, if f ∈ Gk (Γ1 (N ); C), δkr (f ) satisfies δkr (f )|k+2r γ = δkr (f ) for
all γ ∈ Γ1 (N ). Although δkr (f ) is not a holomorphic function, defining
α−k−2 (δk f )(z0 ) = (δk f )|k+2 ρ(α)(z0 ) = (δk f )(z0 )h(z0 ) + f (z0 )(δ0 h)(z0 )
= (δk f )(z0 )α−k + f (z0 )(δ0 h)(z0 ),
d
P (X, f ) = (X − f |k γ) = aj (z)X j ,
γ∈Γ1 (N )\SL2 (Z) j=0
d
Qj (g2 (E, ω), g3 (E, ω))X j = 0.
j=0
Thus, this gives an algorithm to compute the value f (E(a), φN , φp , ω). Once
we know the value f (E(a), φN , φp , ω) = f (E(a),φ
Ωk
N ,du)
, we can then compute
∞
r
δk (f (z0 ))
Ω∞ k+2rfollowing the above proof (in particular, the induction process). See
[H81a] Sect. 8 for numerical examples of derived values of Eisenstein series.
We can let a ∈ Z×
p act on f ∈ V (B) by
36 1 Nontriviality of Arithmetic Invariants
a
f |a(E, φN , φp ) = f | 0
0 a−1
(E, φN , φp ) = f (E, φN , a · φp ).
The operator d has the invariance properties described in the following two
lemmas. So it is natural to call it the p-adic invariant differential operator.
Lemma 1.32 If f ∈ Gk (Γ1 (N pm ); W ), then we have
0
(dr f )| a0 a−1 = a−k−2r (dr f )
for a ∈ Z×
p.
0
Proof. We can approximate p-adically g := a0 a−1 by an element γn ∈
Γ1 (N pm ) ∩ Γ0 (p2n ) so that γn ≡ g mod pn M2 (Zp ). By Exercise 1.31, we
have
The above action of a [or g ∈ T1 (Zp ))] is the local one concentrated at p (as
φN is intact), where T1 (A) is the subgroup of SL2 (A) made up of diagonal
matrices for any ring A. We have a global action f → f |γ for γ ∈ GL2 (Z(p) ).
Since γ changes the level structure φN , the definition is more involved. For
that, we actually need to extend the
level structure to the prime-to-p Tate
(p∞) 2 ∼
(p)
module η : (A ) = V (E) = limpm E[m] ⊗Z Q so that we can let
(p)
←−
γ act by η → η ◦ γ. This is tantamount to introducing the structure of
(p) (p)
Proof. By the deformation theory of E(a) (see Remark 7.49), the differential
operator d commutes with the action η (p) → η (p) ◦ g for g ∈ SL2 (A(p∞) ).
Then the effect of d on the operation f → f |ρ(α) can be computed by
Lemma 1.32. Note that (E(a), φp ◦ αp , η (p) ◦ ρ(p) (α)) ∼
= (E(a), φp , η (p) ) by
complex multiplication.
1.3 CM Periods and L-Values 37
1 ∂h(ρ(α)(z0 ))
d(h ◦ ρ(α))(z0 ) =
2πi ∂z
1 ∂h
= α−2 ((ρ(α)(z0 )) = α−2 (dh) ◦ ρ(α)(z0 ).
2πi ∂z
Since dh = g1 /g2 for g1 ∈ Gk+2 (Γ1 (N ); W ) and g2 ∈ Gk (Γ1 (N ); W ) for
sufficiently large k, we have (Remark 1.24)
(dh)(E(a), φN , φp,∗ dt
t )
2
= (dh)(E(a), φN , ω) = (δ0 h)(E(a), φN , ω) ∈ Q.
Ωp
(1.3.11)
For sufficiently large k, we will see later in Sect. 6.2.6 that the line bundle
generated by weight k forms is very ample; thus, we can write any modular
function of level N as a ratio of two modular forms of weight k, and hence,
any weight-2 meromorphic modular form is also a ratio of forms, one of weight
k + 2 and another of weight k.
38 1 Nontriviality of Arithmetic Invariants
p,∗ dt )
f (E(a),φ
Since Ωpk
t
= f (E(a), φN , ω) ∈ W , we first show
r
r r
d (f )|k+2r ρ(α) = (d f )h + (dr−s f )(ds h).
s
0<s≤r
αk
Evaluating this at (E(a), φN , φp ), we get for C = (α−2r −1)
r
(dr f )(E(a), φN , φp ) = C· (dr−s f )(E(a), φN , φp )(ds h)(E(a), φN , φp ).
s
0<s≤r
and
(ds h)(E(a), φN , φp ) (ds−1 dh)(E(a), φN , φp ) (δ2s−1 dh)(E(a), φN , du)
= = .
Ωp2s Ωp2s 2s
Ω∞
1.3 CM Periods and L-Values 39
inside Q. Since the left-hand side of the above identity is in the completion
W of ip (W ) in Cp , we conclude the identity in W = i−1
p (W ).
Theorem 1.36 Let V = limn V /pn V be any p-adic Banach algebra over W
←−
with norm | · |p ; then the space
of continuous functions C(Zp ; V ) of Zp with
values in V has a basis { m x
}n=0,1,... over V .In other words, Φ ∈ C(Zp ; V )
∞
n=0 an (Φ) n for an (Φ) ∈ V satisfying
x
has a unique expansion Φ(x) =
|an (Φ)|p → 0 as n → ∞.
One can find an easy proof of this theorem in [LFE] Sect. 3.2. Note here
that the space C(Zp ; V ) is again a Banach space under the norm |Φ|p =
Supx∈Zp |Φ(x)|p . Therefore, any bounded V -linear functional φ : C(Zp ; V ) →
V (called a bounded on Zp with values in V ) is determined uniquely
measure
by the value {φ( nx ) = Zp nx dφ}n=0,1,... . In short, we have
m
x
Φ(x)dφ = lim an (Φ) .
Zp m→∞
n=0 Zp n
Introducing
∞ x n a formal variable T , we have a formal identity (1 + T )x =
T . Thus, by associating a formal power series (1 + T )x dφ(x) =
∞
x n
n=0 n
n=0 Zp n
dφ T ∈ V [[T ]] to a bounded measure φ as above, we have an
isomorphism between the power series ring V [[T ]] and the space M(Zp ; V ) of
all bounded measures on Zp with values in V .
Corollary 1.37 We have a canonical isomorphism of p-adic Banach spaces:
M(Zp ; V ) ∼ ]], where the Banach norm on M(Zp ; V ) is given by |φ|p =
= V [[T
SupΦ=0 Zp Φdφ /|Φ|p .
p
Exercise 1.38 Prove |φ|p = Supn Zp nx dφ , and show M(Zp ; V ) ∼
= V [[T ]]
p
as p-adic Banach spaces.
By Mahler’s theorem, a given
continuous x function Φ : Zp → W can be
expanded uniquely as Φ(x) = m≥0 am (Φ) m for am (Φ) ∈ W tending to 0
x d
as m → ∞. Thus, defining Zp m df (x) = m f (E, φN , φp ), we have a unique
p-adic measure df : C(Zp ; W ) → W , where C(Zp ; W ) is the p-adic Banach
space of continuous functions on Zp with values in W .
Theorem 1.39 Let the notation and the assumption
xbe
as in Theorem 1.34,
and suppose f ∈ Gk (pn ; W). If Φ(x) = m≥0 am (Φ) m ∈ C(Zp ; W ), we have
d
Φ(x)df (x) = Φ(x) = am (Φ) f (E(a), φN , φp ).
Zp m
m≥0
In particular, we have
r
Zp x df (x) dr f (E(a), φN , φp ) δkr f (E(a), du)
= = ∈ W.
Ωpk+2r Ωpk+2r Ω∞k+2r
1.3 CM Periods and L-Values 41
x m E
Then, writing m = j=0 cj xj and defining m = j cj Ej+1 as a function
of (E, ω), we find its q-expansion in Z[[q]]:
a
E
= q n
∈ Z[[q]].
m m
n>0,pn 0<a|n
2 L(0, λk )
−k
ek (E, ω) = Ω∞ α−k = ,
(k − 1)! (k − 1)!Ω∞ k
0=α∈Z[i]
where λk ((α)) = α−k . More generally, the differential operator δkr brings λk
into λk+r,−r , with λk+r,−r ((α)) = α−k (α/α)r , and we get a p-adic measure
L(0,λ )
interpolating Ω k+2rk+r,−r
(2πi)−r
over Z2p [up to an explicit constant like (k − 1)!]. We
∞
may regard λk+r,−r as a continuous character of Z× ×
p × Zp with values in Zp
×
where 0 ≤ j < k and W[λ] is the ring generated over W by λ(A) for A running
over all fractional ideals of M prime to p. For the simple constant c(λ), see
Theorem 9.4.
The determination of the μ-invariant and the λ-invariant of the Katz p-
adic L-functions is a basic question. We would like to answer this question
about μ in the present book.
2
Elliptic Curves and Modular Forms
We now describe basics of elliptic curves and modular curves in three steps:
1. as plane curves over a field;
2. as scheme/group functor over a ring;
3. modular forms as functorial rules on modular curves.
Elliptic curves and modular curves are perhaps the most important objects
studied in number theory. The theory gives a base of the proof by Wiles
(through Ribet’s work) of Fermat’s last theorem, it supplies us with the sim-
plest (and perhaps the most beautiful) example of Shimura varieties (cf. [IAT]
Chaps. 6 and 7), a fast prime factorization algorithm (cf. [REC] IV), and so
on. We give a sketch of the theory of modular curves and modular forms and
(a minimal amount of) notation and terminology for us to be able to state
main themes in the next chapter. For graduate students just starting to learn
elliptic curves and modular forms, this chapter hopefully serves as an intro-
ductory illustration of the theory. Experienced readers may skip this chapter,
going directly to the Chap. 3.
Let a be a principal ideal of the polynomial ring k[X, Y ]. Note that polyno-
mial rings over a field are
a unique factorization domain. We thus have the
prime factorization a = p pe(p) with principal primes p. We call a square-
free if 0 ≤ e(p) ≤ 1 for all principal primes p. Fix a square-free a. The set of
A-rational points for any k-algebra A of a plane curve defined over k is given
by the zero set
Va (A) = (x, y) ∈ A2 f (x, y) = 0 for all f (X, Y ) ∈ a .
but this does not depend on the choice f of generators and depends only on the
ideal a; thus, it is more appropriate to write it as Va . As an exceptional case,
we note V(0) (A) = A2 . Geometrically, if a = {0}, we think of Va (Ω) as a curve
in Ω 2 = V(0) (Ω). This is intuitively more geometric if we take k ⊂ Ω = C (a
“curve” is a two-dimensional “plane” as a real manifold). In this sense, for any
algebraically closed field K over k, a point x ∈ Va (K) is called a geometric
point with coefficients in K, and V(f ) (K) ⊂ V(0) (K) is called the geometric
curve in V(0) (K) = K 2 defined by the equation f (X, Y ) = 0.
By Hilbert’s zero theorem (Nullstellensatz; see [CRT] Theorem 5.4 and
[ALG] Theorem I.1.3A), writing a for the principal ideal of k[X, Y ] generated
by a, we have
a = g(X, Y ) ∈ k[X, Y ]g(x, y) = 0 for all (x, y) ∈ Va (k) . (2.1.1)
The plane curve Vp (for each prime p|a) is called an irreducible component of
Va . Since p is a principal prime, we cannot have a finer nontrivial decomposi-
tion Vp = V ∪ W with plane curves V and W defined over k. A prime ideal
p ⊂ k[X, Y ] generates an ideal in p in k[X, Y ], which may decompose into a
product of primes in k[X, Y ]. If the ideal p remains prime in k[X, Y ], we call
Vp geometrically irreducible.
Suppose that we have a map FA = F (φ)A : Va (A) → Vb (A) given by
two polynomials φX (X, Y ), φY (X, Y ) ∈ k[X, Y ] (independent of A) such that
FA (x, y) = (φX (x), φY (y)) for all (x, y) ∈ Va (A) and all k-algebras A. Such a
map is called a regular k-map or a k-morphism from a plane k-curve Va into
Vb . Here Va and Vb are plane curves defined over k. If A1 = Vb is the affine
line, that is, Vb (A) ∼= A for all A (taking, for example, b = (y)), a regular
k-map Va → A1 is called a regular k-function. Regular k-functions are just
functions induced by the polynomials in k[x, y] on Va ; hence, Ra is the ring
of regular k-functions of Va defined over k. Let An = (A1 )n .
We write Homk-curves (Va , Vb ) for the set of regular k-maps from Va into
Vb . Obviously, only φ? mod a (? = X, Y ) can possibly be unique. We have a
commutative diagram for any k-algebra homomorphism σ : A → A :
F
Va (A) −−−A−→ Vb (A)
⏐ ⏐
⏐
σ.
⏐σ
.
Va (A ) −−−−→ Vb (A ).
FA
Indeed,
We write Ra = k[X, Y ]/a and call it the affine ring of Va . Here is a useful (but
tautological) lemma that is a special case of Yoneda’s lemma:
Lemma 2.1 We have a canonical isomorphism:
HomCOF (Va , Vb ) ∼
= Homk-curves (Va , Vb ) ∼
= Homk−alg (Rb , Ra ).
Thus, for functors, we have Va (?) ∼= HomALG/k (Ra , ?). We identify the two
functors A → Va (A) and A → Hom(Ra , A) in this way. Then the main point
of the proof of the lemma is to construct, from a given natural transformation
F ∈ HomCOF (Va , Vb ), a k-algebra homomorphism F : Rb → Ra , giving F by
F
Va (A) = HomALG/k (Ra , A) φ →A
φ ◦ F ∈ HomALG/k (Rb , A) = Vb (A). Then
the following exercise finishes the proof, as, clearly, if we start with F , the
above association leads to F .
Exercise 2.2 Let F = FRa (idRa ) ∈ VRb (Ra ) = HomALG/k (Rb , Ra ), where
idRa ∈ Va (Ra ) = HomALG/k (Ra , Ra ) is the identity map. Then prove that F
does the required job.
Recall that Va is irreducible (resp., geometrically irreducible) if a is a prime
ideal of k[x, y] (resp., a = ak[X, Y ] is a prime ideal in k[X, Y ]).
Exercise 2.3 1. Prove that for any unique factorization domain R, R[X] is
a unique factorization domain.
2. Give an example of two distinct principal prime ideals a, b of Q[X, Y ] with
Va (Q) = Vb (Q).
3. If a and b are two distinct principal prime ideals of Q[X, Y ], prove
Va (Q) = Vb (Q).
4. For a principal ideal a = (f ) ⊂ k[X, Y ], prove a ∩ k[X, Y ] = a.
5. Show that F : k[X, Y ]/b → k[X, Y ]/a is uniquely determined by the mor-
phism F : Va → Vb of functors.
An element in the total quotient ring of Ra is called a rational k-function on
Va . If Va is irreducible, then rational k-functions form a field. This field is
called the rational function field of Va over k and is written as k(Va ).
2.1 Curves over a Field 47
∂f ∂f
(a, b)(X − a) + (a, b)(Y − b) = 0.
∂X ∂Y
We write the corresponding linear space as TP = Vb for the principal ideal b
∂f ∂f
generated by ∂X (a, b)X + ∂Y (a, b)Y . We say that Va is nonsingular or smooth
at P = (a, b) ∈ Va (K) for a subfield K ⊂ Ω if this TP is really a line, in other
∂f
words, if ( ∂X ∂f
(P ), ∂Y (P )) = (0, 0).
Example 2.4 Let a = (f ) for f (X, Y ) = Y 2 − X 3 . Then for (a, b) ∈ Va (K),
we have
∂f ∂f
(a, b)(X − a) + (a, b)(Y − b) = −3a2 (X − a) + 2b(Y − b).
∂X ∂Y
Thus, this curve is singular only at (0, 0).
Example 2.5 Suppose that k has characteristic different from 2. Let a =
(Y 2 − g(X)) for a cubic polynomial g(X) = X 3 + aX + b. Then the tangent
line at (x0 , y0 ) is given by 2y0 (X −x0 )−g (x0 )(Y −y0 ). This equation vanishes
if 0 = y02 = g(x0 ) and g (x0 ) = 0 and hence is singular at only (x0 , 0) for a
multiple root x0 of g(X). Thus, Va is a nonsingular curve if and only if g(X)
is separable if and only if its discriminant −4a3 − 27b2 = 0.
Suppose that K/k is an algebraic field extension. Then K[X, Y ]/aK[X, Y ]
contains Ra as a subring. A maximal ideal (X −a, Y −b) ⊂ K[X, Y ]/aK[X, Y ]
induces a maximal ideal P = (X − a, Y − b) ∩ Ra of Ra . Then the local ring
OV,P at P is the localization
/a 0
OV,P = b ∈ R, b ∈ R \ P ,
b
where ab = ab if there exists s ∈ R \ P such that s(ab − a b) = 0. Write the
maximal ideal of OV,P as mP . Then mP ∩ R = P .
Lemma 2.6 The linear vector space TP (K) is the dual vector space of
P/P 2 = mP /m2P .
∂f ∂f
f mod (X − a, Y − b)2 = (a, b)(X − a) + (a, b)(Y − b)
∂X ∂Y
if a = (f ), since we have
∂φ ∂φ
φ(X, Y ) ≡ (a, b)(X − a) + (a, b)(Y − b) mod (X − a, Y − b)2 .
∂X ∂Y
Thus, the above duality between Ω(a,b) and DA,(a,b) induces the duality
ΩV,P = P/P 2 and TP (K) given by (ω, t) = t(ω), where we regard t as a
derivation OV,P → K.
We call TP the tangent space at P and ΩP = ΩV,P the cotangent space
at P of V . More generally, a k-derivation ∂ : Ra → Ra is a k-linear map
satisfying the Leibniz condition ∂(φϕ) = φ∂(ϕ) + ϕ∂(φ) and ∂(k) = 0. For a
k-derivation as above, f ∂ : ϕ → f ·∂(ϕ) because f ∈ Ra is again a k-derivation.
The totality of the k-derivation DerVa /k is therefore an Ra -module.
To make TP explicit, first take a = (0); thus, Va = A2 . We have
for ∂ ∈ DerA2 /k ; hence, ∂ is determined by its value ∂(X) and ∂(Y ). Note
∂ ∂
that (∂X) ∂X + (∂Y ) ∂Y in DerA2 /k , and the original ∂ has the same value at
X and Y ; thus, we have
∂ ∂
∂ = (∂X) + (∂Y ) .
∂X ∂Y
∂ ∂
Thus, ∂X , ∂Y gives a basis of DerA2 /k .
Assuming Va nonsingular [including A2 = V(0) ], we write the Ra -dual as
ΩVa /k := Hom(DerVa /k , Ra ) (the space of k-differentials) with the duality
pairing
(·, ·) : ΩVa /k × DerVa /k → Ra .
We have a natural map d : Ra → ΩVa /k given by φ → (dφ : ∂ → ∂(φ)) ∈
DerVa /k . Note
for all ∂ ∈ DerVa /k . Thus, we have d(φϕ) = φdϕ + ϕdφ, and d is a k-linear
derivation with values in ΩVa /k .
Again, let us first look into ΩA2 /k . Then, by definition, (dX, ∂) = ∂X
∂ ∂
and (dY, ∂) = ∂Y ; hence, {dX, dY } is the dual basis of ∂X , ∂Y . We have
∂Φ ∂Φ
dΦ = ∂X dX + ∂Y dY , as we can check easily that the left-hand and the
right-hand sides have the same value on any ∂ ∈ DerA2 /k .
If ∂ : Ra = k[X, Y ]/(f ) → Ra is a k-derivation, we can apply it to any
polynomial Φ(X, Y ) ∈ k[X, Y ] and hence regard it as Y ] → Ra . By
∂∂ : k[X,
∂
the above argument, Derk (k[X, Y ], Ra ) has a basis ∂X , ∂Y now over Ra .
Since ∂ factors through the quotient k[X, Y ]/(f ), it satisfies ∂(f (X, Y )) =
(df, ∂) = 0. Thus, we have the following:
Lemma 2.7 We have an inclusion DerVa /k → (Ra ∂X ∂
⊕ Ra ∂Y
∂
) whose image
is given by {∂ ∈ Derk (k[X, Y ], Ra )|∂f = 0}. This implies
Thus, L is free of rank 1 over A and hence has a generator (x0 , . . . , xn ) with
xi ∈ A× . Then (x0 , . . . , xn ) → ( xx0i , . . . , xxni ) ∈ An gives rise to a natural
transformation of Di onto An (which is an isomorphism of functors).
If K is local (in particular, a field), we write (x0 :x1 : · · · :xn ) for the point of
Pn (K) represented by (x0 , . . . , xn ), as only the ratio matters.
Exercise 2.14 Is there any example of a point in X ∈ P1 (A) (and a ring
A) such that the projections to the first and second coordinates are both not
surjective?
We assume that K is a field for a while. When n = 1, we see P1 (K) =
K × {∞} by (x:y) → xy ∈ K {∞}. Thus, P1 (R) [resp., P1 (C)] is isomorphic
(topologically under Euclidean topology) to a circle (resp., a Riemann sphere).
We now assume that n = 2. Set L = {(x:y:0) ∈ P2 (K)}. Then P1 ∼ =L
by (x:y) → (x:y:0); hence, L is isomorphic to the projective line. We have
P2 (K) = D(K) L for fields K, where D = D2 . Thus, geometrically (i.e.,
over algebraically closed fields K), P2 (K) is the union of the affine plane and
a projective line L ∼
= P1 (K). We let L = L∞ (the line at ∞).
and G(X, Y, Z) = g( X Y
Z , Z )Z
deg(g)
. Multiplying h or g by a power of Z does
not change the above identity f (x, y, 1) = h(x,y) g(x,y) , because Z = 1 on C ∩ D2 .
Thus, adjusting in this way, we get G and H.
2.1.5 Divisors
Lemma 2.22 Let C be a smooth projective plane curve. For any f ∈ k(C),
deg(div(f )) = 0, and f is a constant in k if f ∈ k(C) is regular at all P ∈ C.
Lemma 2.23 If f ∈ k(C) satisfies deg(div0 (f )) = deg(div∞ (f )) = 1, f :
C → P1 induces an isomorphism of projective plane curve over k.
Write Div0 (C) = {D ∈ Div(C/k )| deg(D) = 0}. Inside Div0 (C), we have the
subgroup {div(f )|f ∈ k(C)× }. We call two divisors D, D linearly equivalent
if D = div(f ) + D for f ∈ k(C). We say that D and D are algebraically
equivalent if deg(D) = deg(D ). The quotient groups
are called the jacobian and the Picard group of C, respectively. Sometimes,
J (C) is written as Pic0 (C) (the degree-0 Picard group).
Clearly, L(D) is a vector space over k. It is known that #(D) = dimk L(D) <
∞. For φ ∈ k(C)× , L(D) f → f φ ∈ L(D − div(φ)) is an isomorphism.
Thus, #(D) depends only on the class of D in Pic(C).
Example 2.24 Let C = P1 . Take a divisor D = a∈k ma [a] with ma ≥ 0
and ma > 0 for some a, regarding a ∈ k as a point [a] ∈ P1 (k) = k {∞}.
g(x)
On A1 (k) = k, forgetting about ∞, div(f ) + D ≥ 0 if f = (x−a) ma for
a
a polynomial g(x). If deg(D) ≥ deg(g(x)), the function f does not have a
pole at ∞. Thus, L(D) = {g(x)| deg(g(x)) ≤ deg(D)}, and we have #(D) =
2.1 Curves over a Field 57
and from this, we conclude deg(K) = 2g(C) − 2. One can find a proof of this
theorem in any introductory book of algebraic geometry (e.g., [ALG] IV.1 or
[GME] Theorem 2.1.3).
Corollary 2.27 If g(C) = 1 and deg(D) > 0, then we have #(D) = deg(D)
and #(−D) = 0.
E(K) ∼
= J (E)Gal(k/K) = {D ∈ J (E)|σ(D) = D for all σ ∈ G},
60 2 Elliptic Curves and Modular Forms
where σ ∈ Gal(k/k) acts on D = P mP [P ] by σ(D) = P mP [σ(P )].
Basically, by definition, we have
{D ∈ Pic0 (E)|σ(D) = D}
J (E)(K) := J (E)Gal(k/K) = .
{div(f )|f ∈ K(E)× }
Since any subfield K ⊂ k is a union of finite extensions, the identity E(K) ∼ =
J (E)(K) is also true for an infinite extension K/k inside K. Actually, we have
a good definition of Pic(E)(A) for any k-algebra A, and we can generalize the
identity E(K) ∼= J (E)(K) to all k-algebras A in place of fields K inside k
(see Theorem 6.3).
The function Δ(E, ω) is called the discriminant function and also Ramanu-
jan’s Δ-function. An equation of an elliptic curve E as in (2.2.1) is called a
Weierstrass equation of E, which is determined by the pair (E, ω).
Proof. By the dimension formulas, counting the order of poles at 0 of mono-
mials of x and y, we have
L(4[0]) = k + kx + ky + kx2 ,
L(5[0]) = k + kx + ky + kx2 + kxy and
L(6[0]) = k + kx + ky + kx + kxy + kx3
2
= k + kx + ky + kx2 + kxy + ky 2 ,
2.2 Elliptic Curves 61
y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 with aj ∈ k, (2.2.2)
Similarly, [k(E) : k(y)] = 3. Since [k(E) : k(C)] is a common factor of the two
degrees [k(E) : k(x)] = 2 and [k(E) : k(y)] = 3, we get k(E) = k(C). Thus,
if C is smooth, E ∼ = C by φ as a smooth geometrically irreducible curve is
determined by its function field. Therefore, assuming C is smooth, E/k can
be embedded into P2/k via P → (x(P ), y(P )). The image is defined by (2.2.2).
Let T be a local parameter at 0E normalized so that
ω = (1 + higher-degree terms)dT.
(y + ax + b)2 + a1 x(y + ax + b) + a3 (y + ax + b)
= y 2 + (2a + a1 )xy + (2b + a3 )y + polynomial in x.
y 2 = (x + a )3 + b2 (x + a )2 + b4 (x + a ) + b6 = x3 + (3a + b2 )x2 + · · · .
This shows
dx −2T −3(1 + · · · )
= dT = (1 + higher-degree terms)dT = ω.
y −2T −3(1 + · · · )
nowhere vanishing differential forms ω and ω . We call two pairs (E, ω) and
(E , ω ) isomorphic if we have an isomorphism ϕ : E → E with ϕ∗ ω = ω and
ϕ(0E ) = 0E . Here, for ω = f dg, ϕ∗ ω = (f ◦ ϕ)d(g ◦ ϕ); in other words, if
σ : k(E ) → k(E) is the isomorphism of the function fields associated with ϕ,
ϕ∗ ω = σ(f )d(σ(g)). Let T be the parameter at the origin 0E of E adapted
to ω . If ϕ : (E, ω) ∼ = (E , ω ), then the parameter T = ϕ∗ T mod T 2 is
adapted to ω [because ϕ∗ ω = ω and ϕ(0E ) = 0E ]. We choose coordinates
(x, y) for E and (x , y ) for E relative to T and T as above. By the uniqueness
of the choice of (x, y) and (x , y ), we know ϕ∗ x = x and ϕ∗ y = y. Thus, the
Weierstrass equations of (E, ω) and (E , ω ) coincide. We write g2 (E, ω) and
g3 (E, ω) for the g2 and g3 of the coefficients of the Weierstrass equation of
(E, ω). If a field K has characteristic different from 2 and 3, we have
1 2
℘(K) := (E, ω)/K ∼
= (g2 , g3 ) ∈ K 2 Δ(E, ω) = 0
∼ 1 1
= HomALG (Z[ , X, Y, 3 ], K),
6 X − 27Y 2
where [·] indicates the set of isomorphism classes of the objects inside the
brackets and Spec(R)(K) for a ring R is the set of all algebra homomorphisms:
R → K. The last isomorphism sends (g2 , g3 ) to the algebra homomorphism
φ with φ(X) = g2 and φ(Y ) = g3 . We will see later this identity is actually
valid any algebra A in ALG/Z[ 16 ] in place of a field K.
Exercise 2.39 If k has characteristic 2, show that we cannot have any ring
R such that 1 2
℘(K) = (E, ω)/K ∼ = HomALG (R, K)
for all field extensions K/k. Here the isomorphism is a natural transformation
between the functors K → [(E, ω)/K ] and K → HomALG (R, K) from the
category of fields into SET S.
We now classify elliptic curves E, eliminating the contribution of the dif-
ferential from the pair (E, ω). If ϕ : E ∼
= E for (E, ω) and (E , ω ), we have
∗ × ∗
ϕ ω = λω with λ ∈ K , because ϕ ω is another nowhere vanishing dif-
ferential. Therefore, we study K × -orbit: (E, ω) mod K × under the action
of λ ∈ K × given by (E, ω)/K → (E, λω)/K , computing the dependence of
gj (E, λω) (j = 2, 3) on λ for a given pair (E, ω)/K . Let T be the parameter
adapted to ω. Then λT is adapted to λω. We see
g2 (E, λω) = λ−4 g2 (E, ω) and g3 (E, λω) = λ−6 g3 (E, ω). (2.2.3)
Thus, we have the following result:
Theorem 2.40 If two elliptic curves E/K and E/K are isomorphic, then
choosing nowhere vanishing differentials ω/E and ω/E , we have gj (E , ω ) =
for a 12th root λ of Δ(E, ω)/Δ(E , ω ). Note that the 12th root λ may not be
in K if K is not algebraically closed.
Conversely, for a given j ∈ {0, 1}, the elliptic curve defined by
27j
y 2 = 4x3 − gx − g for g =
j−1
has J-invariant 123 j. If j = 0 or 1, we can take the following elliptic curve with
J = 0 or 123 . If J = 0, then y 2 = 4x3 − 1, and if J = 123 , then y 2 = 4x3 − 4x
(Gauss’s lemniscate). Thus, we have the following:
Corollary 2.41 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) ∈ K.
Exercise 2.42 1. Prove that gj (E , ω ) = λ−2j gj (E, ω) for suitable ω and
ω and a suitable 12th root λ of Δ(E, ω)/Δ(E , ω ) if J(E) = J(E ).
2. Explain what happens if J(E) = J(E ) but E ∼ E over a field K not
=
necessarily algebraically closed.
What we have done over fields can also be done over general noetherian rings
A. We sketch the theory (see Chap. 6 for detailed proofs). Here is a defi-
nition of a plain projective curve over a ring A as a subfunctor C ⊂ P2 .
Recall L ∈ P2 (R) for an A-algebra R is a locally free R-submodule of R3 of
rank 1 with locally free quotient R3 /L. For a given homogeneous polynomial
Φ(X, Y, Z) ∈ A[X, Y, Z], we define Φ(L) = 0 if Φ(#) = 0 for all # ∈ L. Assume
that F (X, Y, Z) is not a zero divisor in A[X, Y, Z]. Then a homogeneous poly-
nomial F (X, Y, Z) ∈ A[X, Y, Z] defines a subfunctor (called a plane projective
A-curve) by
R → C(R) = {L ∈ P2 (R)|F (L) = 0}.
A[X,Y,Z]
Clearly, C is a covariant subfunctor of P2 . If the residue ring (F (X,Y,Z)) modulo
its nilradical is an integral domain, we call C irreducible.
Exercise 2.43 If A is a field k, verify that this definition is equivalent to the
definition of irreducibility of the plane k-curve already given earlier.
We define HomA-curves (C, C ) := HomCOF (C, C ), and in this way, we get
the category of plane projective A-curves. Fix such a curve C ⊂ P2/A . Suppose
that A is a local ring with maximal ideal m. Write k for A/m. We then define
and C2 can be identified with the functor sending R to the zero set of
F (X, Y, 1) in R2 . If R is a local ring, we know C(R) = C0 (R)∪C1 (R)∪C2 (R).
For any finite field extension K of k, a point P ∈ Ci (K) gives rise to an A-
algebra homomorphism φ : Ri → K; hence, Ker(φ) is a maximal ideal of Ri .
The cotangent space at P is defined by P/P 2 and the tangent space at P over
K is by definition its dual HomK (P/P 2 , K). As before, the tangent space is
isomorphic to the space of K-derivations ∂ : OC,P → K.
We sketch a general definition of smoothness, but before starting this sub-
tle process of defining smoothness over a ring, we point out that—precise
definitions aside—an important point is that we can again prove that an el-
liptic curve defined by y 2 = 4x3 − g2 x − g3 is smooth over A = Z[ 16 , g2 , g3 ]
if and only if Δ ∈ A× . If the reader is not very familiar with the notion of
smoothness over rings, he or she can just admit this fact for a while to go
through this section and the next (as we learn more about this in Chap. 4).
Here is a formal definition of smoothness. For A-algebras R and R ,
we define the R -module of derivations DerA (R, R ) by the R -module of
derivations trivial over A [hence, (∂ : R → R ) ∈ DerA (R, R ) satisfies
∂(ϕφ) = ϕ∂(φ) + φ∂(ϕ) and ∂(a) = 0 for all a ∈ A].
Consider m-adic completions
= lim A/mn and O
A C,P = lim OC,P /mn .
← −
n
← −
n
P
Then O C,P is naturally an algebra over A. Write m P for the maximal ideal of
OC,P . We call P ∈ C(K) smooth over A if O C,P is free of finite positive rank
]] for a variable T ∈ O
over A[[T C,P and any adically continuous derivation of
A[[T ]] over A with values in any Artinian O C,P -algebra extends uniquely to
OC,P ; i.e., the ring-theoretic tangent spaces of OC,P and A[[T
]] are equal. This
last point means that ΩOC,P /A[[T ]] = 0 (see Sect. 4.1.7 and Lemma 4.32). In
short, O ]]; i.e., Spec(O
C,P is an étale algebra over A[[T C,P ) and Spec(A[[T
]])
are locally isomorphic in the sense of algebraic geometry; see Sect. 4.1.8 for a
more thorough discussion of étale/smooth morphisms).
If C is smooth over A at all maximal points P ∈ C(K), we call C smooth
over A. Assuming that k is algebraically closed, C is smooth over A if and
only if O C,P ∼ ]] as A-algebras
= A[[T for all maximal points P ∈ C.
For general A not necessarily local, we call C smooth over A if C is smooth
over the localization of A at every maximal ideal of A.
Exercise 2.45 Prove that if C is a smooth plane projective curve over an
integral local domain A with algebraically closed residue field, C is smooth
over the quotient field of A.
We recall the Ri -module of derivations DerCi /A = DerA (Ri , Ri ), which
is the Ri -module of ∂ : Ri → Ri such that ∂(ϕφ) = ϕ∂(φ) + φ∂(ϕ) and
∂(a) = 0 for all a ∈ A. The Ri -dual ΩCi /A of DerCi /A is called the Ri -module
of 1-differentials over Ci . Each ∂ ∈ DerCi /A gives rise to an AP -derivation
∂P : OC,P → OC,P given by ∂P ( ab ) = ∂(a)b−a∂(b)
b2 for a maximal point P ∈ Ci ,
where AP is the localization of A at P ∩ A (regarding P as a prime ideal
of Ri ). By duality, ω ∈ ΩCi /A therefore gives rise to the cotangent vector
2.3 Modular Forms 67
In other words, for a given pair (E, ω)/A of an elliptic curve E over A and a
nowhere vanishing differential ω, there exists unique (g2 (E, ω), g3 (E, ω)) ∈ A2
such that E is canonically isomorphic to an elliptic curve defined by
Y 2 Z = 4X 3 − g2 (E, ω)XZ 2 − g3 (E, ω)Z 3
and ω induces the differential dX
Y on E2 = E ∩ D2 under this isomorphism.
Thus, we have the following:
1. If (E, ω) is defined over a Z[ 16 ]-algebra A, we have gj (E, ω) ∈ A, which
depends only on the isomorphism class of (E, ω) over A;
2. gj ((E, ω) ⊗A A ) = σ(gj (E, ω)) for each Z[ 16 ]-algebra homomorphism σ :
A → A ;
3. gj (E, λω) = λ−2j gj (E, ω) for all λ ∈ A× .
68 2 Elliptic Curves and Modular Forms
Thus, φRA (idRA ) ∈ RA ; hence, write φRA (idRA ) = Φ(g2 , g3 ) for a two-
variable rational function Φ(x, y) ∈ A[x, y, x3 −27y
1
2 ]. Let E/RA be the uni-
In the following three sections, we would like to give a sketch of the classical
theory of elliptic curves defined over the complex field C created by Weier-
strass. By means of Weierstrass ℘-functions, we can identify E(C) (for each
elliptic curve E/C ) with a quotient of C by a lattice L. In this way, we can
identify [(E, ω)/C ] with the space of lattices in C. This method is analytic.
We can deduce from the analytic parameterization (combined with geo-
metric technique of Weil–Shimura) many results on the moduli space of elliptic
curves, such as the exact field of definition of the moduli, determination of
the field of moduli (of each member), and so on (e.g., [IAT] Chap. 6). We have
come here in a reverse way: Starting algebraically, mainly by the Riemann–
Roch theorem, we have determined a unique Weierstrass equation over A for
a given pair (E, ω)/A , and therefore, we know the exact shape of the moduli
space before setting out to study the analytic method. After studying analytic
theory over C, combining these techniques, we start studying modular forms.
We start with classical homotopy theory for complex manifolds. A path
γ on a complex manifold M is a piecewise smooth continuous map γ from
the closed interval [0, 1] into a complex manifold M . We write the path as
(γ : x → y) with γ(0) = x and γ(y) = 1. Morally, we start the point x at
time 0 and “walking” to reach y by the unit time 1. Two paths (α : x → y)
and (β : x → y) are homotopy equivalent (for which we write α ≈ β) if we
have a bicontinuous map φ : [0, 1] × [0, 1] → M such that α(t) = φ(0, t) and
β(t) = φ(1, t) (i.e., we can fill the area encircled by α and β in M , intuitively).
Consider the space Z = Z(M ) of homotopy classes of paths starting from a
fixed point x ∈ M . An open neighborhood U of x is called simply connected
if we have a bijection: Z(U ) ∼ = U by projecting (γ : x → y) down to y [i.e.,
no holes in Z(U ) so no path circling the hole]. Shrinking the neighborhood of
x (to avoid holes), we can always find a simply connected open neighborhood
of x ∈ M . For example, if U is diffeomorphic to an open disk with center x, it
is simply connected (that is, every loop is equivalent to x). If γ : x → y and
γ : y → z are two paths, we define their composed path γγ : x → z by
γ(2t) if 0 ≤ t ≤ 1/2
γγ (t) =
γ (2t − 1) if 1/2 ≤ t ≤ 1.
where
1 1
g2 = g2 (L) = 60 and g3 = g3 (L) = 140 .
#4 #6
∈L−{0} ∈L−{0}
Equating poles, ϕ = yL 2
− 4x3L + g2 xL + g3 is holomorphic everywhere. Since
these functions factor through the compact space C/L, ϕ is bounded and hence
must be a constant. Since the constant terms of xL and yL both vanish, we
conclude ϕ = 0. We obtain a holomorphic map (xL , yL ) : C/L − {0} → A2/C .
Counting the order of poles at 0, we check that the map (xL , yL ) has degree
1, that is, an isomorphism onto its image and extends to
Thus, we rediscover the elliptic curve EL = E(L) = Φ(C/L) = E(g2 (L), g3 (L))
and the induced differential
dxL
ωL = = du.
yL
This shows
Theorem 2.51 (Weierstrass) We have ℘(C) = [(E, ω)/C ] ∼
= Lat.
We now relate the space Lat with the upper half-complex plane H. Two
complex numbers w1 , w2 ∈ (C× )2 span a lattice if and only if Im(w1 /w2 ) = 0.
Recall that H = {z ∈ C| Im(z) > 0}. By changing the order of w1 and w2
without affecting their lattice, we may assume that Im(w1 /w2 ) > 0. Thus, we
have a natural isomorphism of complex manifolds:
/ 0
× 2 ∼
B = v = (w 1
w2 ) ∈ (C ) Im(w1 /w2 ) > 0 = C× × H
w2 ) → (w2 , w1 /w2 ).
via ( w1
Two vectors v and v span the same lattice L if and only if v = αv for
α ∈ SL2 (Z); hence,
Lat ∼= SL2 (Z)\B.
a b
This action of α = c d ∈ SL2 (Z) on B can be interpreted on C× × H as
follows:
az + b
α(u, z) = (cu + d, α(z)) for α(z) = .
cz + d
Exercise 2.52 Check that the above action is well defined; that is, show that
(αβ)(u, z) = α(β(u, z)) for α, β ∈ SL2 (Z).
2.3 Modular Forms 73
The last expansion of Ramanujan’s function tells us that Δ does not vanish
on H analytically although this is geometrically clear, as Δ(E, ω) = 0 for any
elliptic curve (see [EEK] for a concise proof of the product expansion). In
particular, we have
(12g2 )3
J= = q −1 + · · · ∈ q −1 (1 + Z[[q]]).
Δ
74 2 Elliptic Curves and Modular Forms
(E∞ , ω∞ ) gives an elliptic curve over the Laurent series ring Z[1/6]((q)) with
nowhere vanishing differential ω∞ . For any f ∈ Gw (A),
f (q) = f ((E∞ , ω∞ ) ⊗Z[ 16 ]((q)) A((q))) ∈ A[[q]]
∞
1 1 1
℘L (u) = 2 + −
u (u + 2πim)2 (2πim)2
m=−∞, m=0
∞
∞
"
1 1
+ −
n=1 m=−∞
(−u + 2πim + n log q)2 (2πim + n log q)2
∞
1 1
+ −
n=1
(u + 2πim + n log q)2 (2πim + n log q)2
∞
∞ ∞
2ζ(2)
= mwm − 2
+ mw−m q mn + mwm q mn − 2q mn .
m=1
(2πi) n=1 m=1
d 1
∞
Applying dw to 1−w = m=0 wm , we get
∞
w
= mwm .
(1 − w)2 m=1
π2
Then, from the fact ζ(2) = 6 (see [LFE] Sect. 2.1), we see
1
xL (u) = ℘L (u) = tL (w) + , (2.4.2)
12
76 2 Elliptic Curves and Modular Forms
where
∞
∞
qm w qm
t(q, w) = tL (w) = − 2 . (2.4.3)
m=−∞
(1 − q w)
m 2
m=1
(1 − q m )2
This new formula confirms t(q, w) ∈ Z[w, w−1 , (1 − w)−1 ][[q]]. Let Aw =
Z[w, w−1 , (1 − w)−1 ] as a localized polynomial ring, which is plainly a finitely
generated Z-algebra. Thus, we have t(q, w) ∈ Aw [[q]].
d d
Again applying w dw = du to xL , we get
where
∞
∞
(q m w)2 qm
s(q, w) = sL (w) = + . (2.4.5)
m=−∞
(1 − q w)
m 3
m=1
(1 − q m )2
2
From the identity: yL = 4x3L − g2 (L)xL − g3 (L), we get
where
∞
1 1 n3 q
b2 (q) = b2 (L) = g2 − =5 ∈ qZ[[q]]
4 12 n=1
1 − qn
∞
1 g2 1 7n5 + 5n3 qn
b3 (q) = b3 (L) = g3 + − = ∈ qZ[[q]].
4 12 432 n=1
12 1 − qn
Note here that the identity is the algebraic identity in the power series ring
Aw [[q]], because the identity is valid over the open set |q| < |w| < |q|−1 in C2 .
We recall the following identity in Aw [[q]]:
3 2
1 b2 1
Δ= g23 − 27g32= 4b2 + − 27 4b3 − −
12 3 216
∞
= b3 + b22 + 72b2 b3 − 432b23 + 64b32 = q (1 − q n )24 . (2.4.7)
n=1
Exercise 2.53 1. Show that the projective plane curve C over a field k de-
fined by X 3 − XY Z − Y 2 Z = 0 is singular at (0:0:1), which is an ordinary
double point.
2. Show the function field of the curve C for k = Fp as above is isomorphic
w2
to Fp (w) with x = X w Y
Z = (1−w)2 and y = Z = (1−w)3 .
By the computation we have done in the previous section, the projective plane
curve E∞ is defined over Z[[q]] by the following new equation in P2/Z[[q]] with
homogeneous coordinate (S:T :U ):
S 2 U + T SU − T 3 + b2 (q)T U 2 + b3 (q)U 3 = 0.
An important point is that the curve originally defined over Z[ 16 ][[q]] is ex-
tended to over Z[[q]]. It has an integral point 0 given by (S, T, U ) = (1, 0, 0).
More generally, we can think of a surjective homomorphism:
taking (S, T, U ) to (S, 0, 0). To compute the tangent space at 0, we use the
affine equation of u = U/S and t = T /S given by u + tu = t3 − b2 u2 t − b3 u3 ,
and we have
ΩOE ,0 /Z[[q]] = Z[[q, t]]dt.
∞
ing q) has one singular point, that is, E∞ mod q is singular only at P with
coordinate (s, t) = (0, 0) (which is not the origin of E ∞ = E∞ mod q), and
the (completed) stalk O E∞ ,P is isomorphic to Z[[q]][[t, s]]/(ts − q), which is a
regular ring (cf. [CRT] Sect. 19). Thus, the local ring at every geometric point
of E∞/Z is a regular ring of relative dimension 1 over Z[[q]] (we call such a
curve a regular curve). The smooth locus of E ∞ is isomorphic to P1 removed
two points, say 0 and ∞, which is Gm (here we may think Gm as a covariant
functor sending a ring A to its multiplicative group A× ). The description of
78 2 Elliptic Curves and Modular Forms
in Z[w, w−1 ][[q]]. Thus, if w ∈ A× , the series ((1 − w)3 s(q, w), (1 − w)3 t(q, w))
converges in A2 under the q-adic topology; hence, we get a point
for q ∈ K × with |q| < 1. Here |·| is the absolute value of K, and convergence is
under the topology of K. We simply put π(w) = 0 ∈ EA (K) if w ∈ q Z . Thus,
π : K × /q Z → EA (K) is well defined by (2.4.8), and by the first assertion,
π −1 (0) = q Z . (2.4.13)
we may assume that π(u) = P , π(v) = Q, and π(w) = R are all different
from 0 and that P = Q. Write P = (s, t), Q = (s , t ), and R = (s , t ).
By (2.4.10), (2.4.11), and (2.4.12), π(u) + π(v) = π(w) is equivalent to the
following simultaneous identities:
Assuming w = uv, we want to show that this identity (2.4.14) holds for
π(u) = P , π(v) = Q, and π(w) = R. Since w = uv, (2.4.14) is the identity in
Exercise 2.55 Let (G, ·) and (H, +) be infinite abelian groups and let π :
G → H be a map with infinite image sending the identity to the identity such
that π(v · w) = π(v) + π(w) as long as π(v) = ±π(w). Show that π is a
homomorphism.
Corollary 2.56 Let W be a complete discrete valuation ring finite flat over
Zp inside Qp and put K for its field of fractions. Let q ∈ W with |q|p < 1
and Eq be the elliptic curve over W with Tate period q. Then, for a prime
l, we have E[ln ] ∼
n
= μln × Z/ln Z by the map sending ζq m/l ∈ Eq [ln ] with
ζ ∈ μln (Qp ) and m ∈ Z to (ζ, (m mod ln )) ∈ μln (Qp ) × Z/ln Z. On Tl Eq =
limn Eq [ln ] ∼
= Zl (1) × Zl , Gal(Qp /K) acts by a nonsemisimple representation
←−
isomorphic to σ → N0l u1 , where Nl is the l-adic cyclotomic character such
that ζ σ = ζ Nl (σ) for ζ ∈ μl∞ (Qp ) and Nl (σ) ∈ Zl × and for a choice of
q 1/l ∈ Qp and the primitive ln th root ζln = ip (i−1
n
∞ (e( ln ))) ∈ μln (Qp ), we have
1
n u(σ) n
(q 1/l )σ = ζln q 1/q (for all n) with u(σ) ∈ Zl . In particular, the ln -torsion
points of Eq for n sufficiently large generate a nonabelian soluble extension
of K.
Proof. By Theorem 2.54 (2), the field of rationality of Eq [ln ] is the Kummer
n
extension K[μln , q 1/l ]. Then the above form of representation is given by the
n
Galois action on the basis {ζln , q 1/l } of Eq [ln ](Qp ).
is called the Neben type, or the “Neben” character (taking the terminology of
Hecke). These spaces are defined in [IAT] Sect. 3.5 along with the Hecke oper-
ators T (n) acting on it, where Sk+1 (Γ0 (N pr+1 ), ψ) = Sk+1 (Γ0 (N pr+1 ), ψ; C)
is defined. We already defined the spaces of modular forms and cusp forms
Gk+1 (Γ1 (N )) = Gk+1 (Γ1 (N ); C) and Sk+1 (Γ1 (N )) = Sk+1 (Γ1 (N ); C) [see
(G1–4) and (s) in Sect. 1.3.1]. We will give a more detailed definition of these
spaces and operators later (see Definitions 6.30 and 6.21). In this chapter
only, we write the weight of the modular form as k + 1 since the l-Frobenius
eigenvalue has absolute value lk/2 for the Galois representation of a cuspi-
dal eigenform f of weight k + 1. Here we say that f is a Hecke eigenform if
f |T (n) = an f for all n. Actually, the motive associated with f has Hodge (and
Hodge–Tate) weight (k, 0), so it might be more natural to call this weight k as
the weight of modular form, though we use the traditional weight k + 1 in this
book (see [MTV] for motives and [HMI] Sect. 4.4.1 for Hodge–Tate weight).
Let Z[ψ] be the subring of Q generated by the values of ψ. The Hecke
algebra over Z[ψ] is
h ⊗Λ,εk W [ε] ∼
= hk+1,εψk /W [ε] for ψk := ψ1 ω 1−k ,
Tr(ρI (F robl )) = a(l) (for the image a(l) in I of T (l)) for all primes l N p.
interpolating
L(1, Ad(ρP ))
Lp (P ) := P (Lp ) = (Lp mod P ) = for all arithmetic P .
period
The assumptions are that for the connected component Spec(T) of Spec(h)
containing Spec(I), T is a Gorenstein ring (see [MFG] Sect. 5.3.4 for Gorenstein
ring) and that Spec(I) is normal (see [H11c] Sect. 3.1 for more explanation).
Actually, so far, all known computed examples of the normalization of I are
regular isomorphic to a one-variable power series ring over a valuation ring
finite over Zp , and as for T, non-Gorenstein examples are rare, but they exist
(see [KiW]).
3.1 Abelian Component of the “Big” Hecke Algebra 89
Writing the image of T (n) in I (resp., X) as a(n) [resp., b(n)] and regarding
them as functions on Spec(I) and Spec(X), we have a(n)(P ) = b(n)(P ) for
P in the intersection Spec(I) ∩ Spec(X); in other words, we have a(n) ≡ b(n)
mod P (congruence between two Hecke eigenvalues). This is a reason for our
calling this criterion a “congruence” criterion. The study relating congruence
primes and adjoint L-values was started in [H81a]. See also [DoHI] for a his-
torical account of the early research on congruence between Hecke eigenforms
in the late 1970s and some open conjectures (in particular, in the mid-1970s,
Doi found computationally highly nontrivial numerical examples of congru-
ence primes between primitive forms of the same conductor, the same Neben
character, and the same weight).
Congruence modulo a prime l of a fixed Hecke eigenform f in a fixed
space Sk+1 (Γ0 (N pr+1 ), ψ) with another Hecke eigenform g in the same space is
controlled by the canonical integer part c(f ) of the adjoint L-value L(1, Ad(f ))
[i.e., basically, l|c(f ) ⇔ the existence of g with g ≡ f mod l for a prime l|l].
Thus, c(f ) is a product of the prime factors of the discriminant/conductor
of the Hecke order Z[f ] = Z[λ(T (n)]n inside Q(f ) and more if we have g
nonconjugate (but congruent) to f . For this reason, the number c(f ) is called
the congruence number of f . This implication from the adjoint L-value to
congruences of a cusp form was first discovered by the author in [H81a] and
[H81b] and developed into the criterion in the above form in [H86a] (0.7). See
[H12b] Sects. 5 and 6 and [MFG] Sect. 5.3.6 for an up-to-date treatment of the
criterion and the construction of Lp . The criterion characterizes Lp , and at
this moment, Lp is well defined up to units in I (see [H11c] Conjecture 3.1 on
how to normalize Lp without ambiguity).
If we interpolate the adjoint L-values including the cyclotomic variable,
that is, adding a variable s that interpolates L(s, Ad(ρP )), moving s, we need
to multiply the L-value by a nontrivial modifying Euler p-factor as described
in Sect. 1.1.2. For this enlarged two-variable adjoint L-function, the modifying
factor vanishes at s = 1 (see Example 1.6); so Lp (s, Ad(ρI )) has an exceptional
zero at s = 1, and for an L-invariant 0 = Lan (Ad(ρI )) ∈ I[ p1 ], we expect to have
?
Lp (1, Ad(ρI )) = Lan (Ad(ρI ))Lp (in the style of Mazur–Tate–Teitelbaum).
Greenberg proposed a definition of a number L(Ad(ρP )) conjectured to
be equal to Lan (Ad(ρP )) for arithmetic P . We interpolate Greenberg’s L-
invariant L(Ad(ρP )) over arithmetic P to get a function L(Ad(ρI )) = 0 in
I[ p1 ] so that L(Ad(ρI ))(P ) = L(Ad(ρP )) for all arithmetic P .
Proof. If f ∈ Gk+1 (Γ0 (N pr+1 ), ψ; Q), by the rationality (see Theorem 2.48
and its generalization
∞ in Sect. 6.2), the Galois conjugate f σ given by the
q-expansion n=0 a(n, f )σ q n for σ ∈ Gal(Q/Q) is again a modular form
in Gk+1 (Γ0 (N pr+1 ), ψ σ ; Q) [where ψ σ (n) = ψ(n)σ for n ∈ Z]. If σ ∈
Gal(Q/K[ψ1 , ω]) fixes aP (p), fPσ is still ordinary (i.e., slope 0) Hecke eigen-
forms of the same level and the same Neben character. The number of such
forms is bounded by rankZp [[T ]] h [by principle (b)]. Thus,
We gather here some results on Weil numbers for the proof of Theorem 3.2.
For a prime l, a Weil l-number α ∈ C of integer weight k ≥ 0 satisfies
∞
Nn −ns
ζ(s, X) = exp( q )
n=1
n
∞
∞
2 ∞
3
Nn −ns 1 Nn −ns 1 Nn −ns
=1+ q + q + q + ···
n=1
n 2! n=1
n 3! n=1
n
(3.1.4)
2d
g(i)
Pi (q −s )(−1)
i+1
ζ(s, X) = for Pi (T ) = (1 − αij T ).
i=0 j=1
The conjecture of Hasse-Weil (e.g., [IAT] Sect. 7.1) tells us that this L-function
has meromorphic continuation to the entire complex plane and satisfies a
functional equation after supplementing with missing Euler factors and Γ -
factors appropriately.
94 3 Invariants, Shimura Variety, and Hecke Algebra
Here is one more example of the natural appearance of Weil numbers. For
any Hecke eigenform f ∈ Sk+1 (Γ0 (N pr+1 ), ψ) with f |T (l) = al f , if l is prime
to N p, then the roots α, β of X 2 − al X + ψ(l)lk = 0 are Weil numbers of
weight k. When f is Ramanujan’s Δ-function, which spans S12 (SL2 (Z)), this
fact is conjectured by Ramanujan; for general k, it was conjectured by Peters-
son [cf. [IAT] Sect. 3.5 (Z) and (Z )]. If f = fP for an arithmetic point P of
Spec(h)(Qp ), the numbers α, β are the eigenvalues of ρP (F robl ) for the Frobe-
nius element F robl at the prime l in Gal(Q/Q), because Tr(ρP (F robl )) = al
k
and det(ρP (F robl )) = ψ(l)l by (3.1.1). We define themodular Hecke L-
∞
function of f by L(s, f ) = l (1 − al l−s + ψ(l)lk−s )−1 = n=1 a(n, f )n−s .
As we described in the earlier chapters (and we go deeper into this subject
in Sect. 6.2.4), each modular form is a function classifying elliptic curves with
additional structures, for example, (E, ω) for the level-1 case and (E, φN , ω)
(with a level-N structure φN ) for level-N cases. Moreover, we remarked that
X1 (N ) is defined over Z[ N1 ] [i.e., X1 (N )(Q) is a zero set in a projective space
Pn (Q) of homogeneous polynomials defined over Z[ N1 ]]. We prove in Theo-
rem 6.25 that X1 (N ) is (at least) defined over Z[ 6N 1
], throwing away primes 2
and 3. Thus, we can think of the reduction X1 (N )/Fp modulo p and degree-1
part of the zeta function L(s, X1 (N )) = L1 (s, X1 (N )).
When k = 1 (that is, for weight-2 cusp forms), the Hasse–Weil conjecture
for X1 (N ) and the Ramanujan–Petersson conjecture were proven by Shimura
by computing L(s, X1 (N )) as a product of L(s, f ) for Hecke eigenforms of
weight 2 on X1 (N ) via identifying the roots α, β of X 2 − al X + ψ(l)l = 0
with l-Frobenius eigenvalues (see [Sh1] and [IAT] Sect. 7.5) and then reducing
√
the proof of |α| = |β| = p to Weil’s work for curves. For higher-weight
modular forms, Deligne went down a similar path to prove the Ramanujan–
Petersson conjecture, reducing it to his proof of Weil’s conjecture taking the
k−1
!
fibered product E ×X E ×X · · · ×X E (see [D1] and also [Sch]). For weight-1
cusp forms (i.e., the case for k = 0), the problem is reduced analytically by
Deligne and Serre to higher-weight cases (see [DS]). Thus, one might expect
to have a Weil number of weight k even for Hecke eigenvalues of U (p) for a
cusp form of weight k + 1 ≥ 2, as “old” Hecke eigenforms of level N p coming
from a form of the original level p N have α or β as their U (p)-eigenvalue.
This is not always true for eigenforms properly of level N pr+1 (r ≥ 0), called
primitive forms (see [MFM] Sect. 4.6 for primitive forms). For example, if
pr+1 = Cp3 (ψ)pe for the p-conductor Cp (ψ) of ψ, then ap is equal to 0 if
e ≥ 2 or ± ψ0 (p)p(k−1)/2 if e = 1 (Weil number of weight k − 1). Here ψ0 is
the primitive character associated with ψ if ψ is imprimitive at p. See [MFM]
Corollary 4.6.18 for these facts. The case where we have Weil number of weight
k − 1 for k ≥ 1 is called the Steinberg (or special) case, as the automorphic
representation generated by the cusp form has p-component isomorphic to the
Steinberg representation (see [AAG] about Steinberg/special representations
and automorphic representations).
3.1 Abelian Component of the “Big” Hecke Algebra 95
field.
Here is another lemma due to Kronecker:
Lemma 3.6 If ζ is a Weil number of weight 0, then there exists a positive
integer N such that ζ N = 1.
Note that (σ −1)(τ −1) = (στ −1)−(σ −1)−(τ −1). Then (σ−1)(τ −1) ∈ Uj+1
because, again by L-adic expansion, (σ − 1) for σ ∈ Ij (L/F ) takes Uj into
Uj+1 for all j. Thus, it is a homomorphism.
The above claim shows that (Ij : Ij+1 ) is a l-power for j > 0 and for
r = dimFF FL with FF := OF /l and FL := OL /L,
98 3 Invariants, Shimura Variety, and Hecke Algebra
(I0 : I1 )|F×
L | = |FF | − 1,
r
I(q) ∼
= Zq (1) as Gal(F ur /F )-module, (3.1.7)
is a finite set. If α/β ∈ μp∞ (K), we have K(α) = K(β) inside L and hence
S1 (α) = S1 (β). By Lemma 3.11 following this proposition, the group of roots
of unity in the composite L of L for all L ∈ KL in Q contains μp∞ (K) as
a subgroup of finite index; hence, S1 (α) = S1 (β) up to a permutation of
valuations in VL if α ∼ β.
3.1 Abelian Component of the “Big” Hecke Algebra 101
We consider for each s ∈ S the set W(s) = {α ∈ W(l, k, L)|S1 (α) = s}.
By definition, we have a decomposition
4
W(l, k, L) = W(s).
s∈S
We want to show that W(s)/μp∞ (K) is a finite set. Write O for the integral
closure of Zl in L (the l-adic integer ring of L). By definition, for α, β ∈
W(s), the principal O-ideal (α) is equal to (β), because S1 (α) = S1 (β) = s
determines the prime factorization of (α) = (β) in L. Write the tensor product
L ⊗Kl L = L1 ⊕ L2 ⊕ · · · ⊕ Ln for fields Li and write Vi for the valuation
of Li normalized so that it extends a member of V. Then the possibility of
ν(α) = (Vi ((α ⊗ 1), Vi (1 ⊗ α))i for α running over W(s) is finitely many,
and if ν(α) = ν(β), α/β is a Weil l-number of weight 0, because the prime
factorization of (α) and (β) in Lα ⊗K Lβ is equal. Thus, α/β is a root of
unity by Kronecker’s theorem (Lemma 3.6), and hence α ∼ β. As already
remarked (see Lemma 3.11), {roots of unity in L} = ζ∈Z ζ · μp∞ (K) for a
finite set Z of roots of unity. Thus, W(s)/μp∞ (K) is finite, which implies
W(l, k)/{p-power roots of unity} is finite.
Exercise 3.10 Explain in details why in the above proof we can replace the
set Kd by KG .
{(σL )L∈K |σL ∈ Gal(L/K), σL |L∩L = σL |L∩L for all pairs (L, L ) ∈ K2 }.
Thus, the order of any σ ∈ Gal(L/K) is a factor of |G| and, hence, bounded.
Pick any infinite subgroup ν ⊂ μ(p) , and decompose it into a product of #-
primary subgroups ν for primes # = p. Let Ξ = {#|ν = 1}. If Ξ is finite,
we have ν ⊃ μ∞ for a prime # ∈ Ξ, and Gal(K(ν)/K) surjects down to
102 3 Invariants, Shimura Variety, and Hecke Algebra
Recall the base ring W , which is a p-adically complete discrete valuation ring
inside Cp . We start with a rigidity lemma:
Lemma 3.12 Let Φ(T ) ∈ W [[T ]]. If there is an infinite subset Ω ⊂ μp∞ (K)
such that Φ(ζ − 1) ∈ μp∞ (Qp ) for all ζ ∈ Ω, then there exist ζ0 ∈ μp∞ (W )
∞
and s ∈ Zp such that ζ0−1 Φ(T ) = ts = n=0 ns T n (t = 1 + T ).
By the assumption, for s ∈ Z× p sufficiently close to 1, ζ → ζ is an automor-
s
phism of W [[μp∞ ]] over W ; hence, Φ(ζ −1) = Φ(ζ−1) ⇔ Φ(t −1) = Φ(t−1)s ,
s s s
and the power series is the desired form as stated in a remark of Chai’s [Ch1]
Remark 6.6.1 (iv). Here is an elementary proof supplied to the author by
Kiran Kedlaya (a proof following Chai can be found in [H11a] Sect. 5; see also
Lemma 3.30 in the text).
Proof. Making the variable change T → ζ1−1 (T + 1)− 1 for a ζ1 ∈ Ω (replacing
W by its finite extension if necessary), we may replace Ω by ζ1−1 Ω 1; hence,
rewriting ζ1−1 Ω as Ω, we may assume that 1 ∈ Ω. Note that t = 1 ⇔ T = 0.
Write the valuation of W as v (and use the same symbol v for an extension
of v to W [μp∞ ]). Normalize v so that v(p) = 1. We are trying to show that
Φ(T ) = (1+T )s ζ for some s ∈ Zp and some p-power root of unity ζ . Anyway,
−1
we write Φ(0) = ζ ∈ μp∞ (Qp ). Replacing Φ by ζ Φ (and extending the
scalars to a finite extension of W if necessary), we may assume thatΦ(0) = 1.
∞
Suppose that Φ(T ) ∈ W (nonconstant). Write Φ(T ) − 1 = i
i=1 ai T .
Since W is a discrete valuation ring, there is a least index j > 0 for which
v(aj ) is minimized. For ( sufficiently small, if v(τ ) = (, then v(Φ(τ ) − 1) =
v(aj ) + j(. In particular, for ζ a p-power root of unity, taking τ = ζ − 1,
we have v(ζ − 1) = p−m /(p − 1) for some nonnegative integer m, so we have
infinitely many relations of the form jp−m /(p−1)+v(aj ) = p−n /(p−1). Then
we may assume m → ∞ ⇒ n → ∞, replacing Ω by its infinite subset. Indeed,
by the nonconstancy of τ → Φ(τ ), the image of Ω under Φ is an infinite set,
and hence over Ω, the order of Φ(ζ − 1) is unbounded. Thus, taking limits
under m → ∞ yields v(aj ) = 0. Also, j must be a power of p, say j = ph , and
for m large, we have n = m − h.
Since v(aj ) = 0, aj mod mW is in F× . For the moment, assume F = Fp .
That is, aj reduces to an integer b0 coprime to p in the residue field of W .
We can thus replace Φ(T ) by Φ1 (T ) defined by Φ(T ) = Φ1 (T ) × (1 + T )s for
some s (namely, s = b0 j = b0 ph0 for h0 := h) so as to increase the least index
3.1 Abelian Component of the “Big” Hecke Algebra 103
j
j for which v(aj ) = 0. Indeed, writing Φ(T ) = n=0 an T n + T j+1 f (T ) with
f (T ) ∈ W [[T ]], we have
j
h0 h0
an T n ≡ 1 + b 0 T p ≡ (1 + T p )b0 ≡ (1 + T )s mod (mW + (T j+1 )).
n=0
We also have Φ1 (T ) ≡ 1+T j+1 f (T )(1+T )−s ≡ 1 mod (mW +(T j+1 )). Thus,
if we write j1 for the j for this new Φ1 , j1 > j, and j1 = ph1 with h1 > h0
∞
and aj1 ≡ b1 mod mW for b 1 ∈ Z. Repeating this, for s = k=0 bk p
hk
∈ Zp ,
Φ(T )/(1 + T )s − 1 = n=1 an T n no longer has a least j with minimal v(aj );
hence, Φ(T )/(1 + T )s = 1, and we get Φ(T ) = (1 + T )s .
Suppose now that F = Fp . We have the Frobeniusautomorphism φ fixing
Zp [μp∞ ] ⊂ W [μp∞ ]. Letting φ act on power series by ( n an T n )φ = n aφn T n ,
we find Φφ (tφ ) = Φ(t)φ . Since Φ(ζ − 1) is a p-power root of unity for ζ in an
infinite set Ω ⊂ μp∞ , we have Φφ (ζ − 1) = Φφ (ζ φ − 1) = Φ(ζ − 1)φ = Φ(ζ − 1).
Since Ω ⊂ G m is Zariski dense, we find that Φφ = Φ, which shows Φ ∈ W φ [[T ]]
for the subring W φ ⊂ W fixed by φ. Note that the residue field of W φ is Fp ,
and the earlier argument applies to Φ ∈ W φ [[T ]].
3.1.5 CM Components
where γ s = (1 + p)s ∈ Z×
p for s ∈ Zp and [x, Qp ] is the local Artin symbol.
for a root of unity ζ0 , where logp (p) = h1 logp (α), taking h such that ph = (α)
with α ∈ M . Thus, in this case, the derivative da(p)dT
|T =0
is essentially logp (p),
M/Q
which is closely related to the L-invariant of χ = mentioned in The-
orem 1.5 (2) and also in the statement “Conjecture/Question” in Sect. 3.1.1.
This is not a coincidence, and we relate the adjoint L-invariant to the deriva-
tive da(p)
dT for the non-CM family in Theorem 3.31.
By a similar computation, we confirm, for any prime l of M outside N p,
logp (α0 )
A(X) = c0 (1 + X)s with c0 = α1 α0 −1 and s = .
logp (γ)
Under the assumption of Theorem 3.2, the constant c0 turns out to be a root
of unity, as we prove later (see Theorem 3.15). In any case, it is close to the
form of the eigenvalues in the CM case discussed in Sect. 3.1.5.
Proof. In this book, we give a proof assuming I = Λ = W [[T ]], referring to
[H13c] Proposition 5.2 for a proof dealing with the general case. By Proposi-
tion 3.9 (and a remark after Lemma 3.11), we have only a finite number of
Weil l-numbers of weight k in P ∈A LP up to multiplication by roots of unity
in μp∞ (K), and hence AP for P ∈ A hits one such Weil l-number α of weight
k infinitely many times, up to p-power order roots of unity, except for the case
where αl,P is not a Weil l-number of weight k. If αl,P is not a Weil l-number
of weight k, then fP is Steinberg at l and l is a factor of N ; thus, this case is
excluded by our assumption l N .
The automorphic representation generated by fP for arithmetic P has l-
component in the principal series if l N or k(P ) > 1. Indeed, in the Steinberg
case at p, we have aP (p) = p(k(P )−1)/2 up to roots of unity, which has to have
slope 0. Hence, it is limited to k(P ) = 1. This follows from the fact that in
the Steinberg case, the central character (i.e., the Neben character) is trivial
at p, and we have aP (p) = p(k(P )−1)/2 up to roots of unity [e.g., [MFM]
Theorem 4.6.17 (2)]. If εP = 1, the p-conductor and the p-level match, and as
already explained, the U (p)-eigenvalue is a Weil number of weight k = k(P ).
Note that Spf(W [[T ]]) = G m (the formal multiplicative group; see Exam-
ple 4.79) as G m = Spf(W [t, t−1 ]) for W[t, t−1 ] = limn W [t, t−1 ]/(tn − 1) ∼
=
←−
W [[T ]] by t = 1 + T . As a functor, Gm sends each local p-profinite W -algebra
R to the following multiplicative group:
{x ∈ R× |x ≡ 1 mod mR }
Recall the root A(l) ∈ Q (we fixed in Sect. 3.1.6) of det(X − ρI (F robl )) = 0 for
a prime l N p. Extend I so that A(l) ∈ I. Extending I further to its integral
closure in its quotient field, we assume that I is normal. We have written
αl,P = P (A(l)) for P ∈ Spec(I)(Qp ). For l|N p, we simply put αl,P = P (a(l))
for the image a(l) in I of U (l).
The horizontal theorem follows from the following theorem:
Theorem 3.15 Let the notation be as above, and write K := Q[μp∞ ] and
EP = K(αp,P ) for each arithmetic point P with k(P ) = k. Suppose that there
exists an infinite set A of arithmetic points with k(P ) = k ≥ 1 satisfying
(bd) EP /K is a finite extension of bounded degree independent of P ∈ A.
Then we have an imaginary quadratic field M in which p splits such that the
component I has complex multiplication by M . In particular, for α0 , α1 given
in Proposition 3.13, we conclude that α1 = α0 up to roots of unity.
108 3 Invariants, Shimura Variety, and Hecke Algebra
In this book, as before, we give a proof under the assumption that I = W [[T ]]
(see [H13c] for the treatment when I = W [[T ]]).
Proof. By (bd) and Lemma 3.3 (2) combined, [K(A(l)P ) : K] is bounded by
a constant B independent of P ∈ A and l N p, and hence the condition (BT)
of Proposition 3.13 is satisfied if l > B!. Assume that l > B! and write A =
Al = A(l). We use the symbols introduced in the proof of Proposition 3.13.
As shown in Proposition 3.13, slightly modifying the two Weil l-numbers of
weight 1 there (by multiplying roots of unity if necessary) and writing them
again as α0 and α1 , we have Al (ζγ κ ) = ζ κ α1 ακ−1 0 (for all ζ ∈ μp∞ ). We now
vary l > B!.
Pick a p-power root of unity ζ = 1 of order m = pe . Write f = fP for
P = (t − γ k ) and g for the cusp form fP for P = (t − ζγ k ). Put αf,l =
αl = α1 = Al (γ k ) and αg,l = βl = Al (ζγ k ). Thus, αf,l = ζl αg,l with ζlm = 1.
Consider the compatible system of Galois representations associated with f
and g. Write ρf (resp., ρg ) for the p-adic member ρf,p (resp., ρg,p ) of the
system associated with f (resp., g).
For a continuous representation ρ : Gal(Q/F ) → GL2 (Qp ), let ρsym⊗j de-
note the jth symmetric tensor representation into GLj+1 (Qp ) introduced in
the previous section. Suppose that f [and hence g by the equivalence of (cm1–
4)] does not have complex multiplication. Then by [Ri3] (see [GME] Theo-
rem 4.3.18), Im(ρf,p ) and Im(ρg,p ) contain an open subgroup of SL2 (Zp ). Then
ρsym⊗j
?,p (? = f, g) is absolutely irreducible for all j ≥ 0. Since βl = ζl αl for an
mth root of unity ζl , we conclude that βlm = αm e
l for the integer m = p . Thus,
Tr(ρf,p (F robl )) = Tr(ρg,p (F robl )) for all prime l N p, where Tr(ρ?,p )(g) is
m m m
just the trace of the mth matrix power ρm ?,p (g). Since the continuous func-
tions Tr(ρm m
f,p ) and Tr(ρg,p ) match on the dense subset of Frobenius conjugacy
classes of all primes outside N p, we find that Tr(ρm m
f,p ) = Tr(ρg,p ) on Gal(Q/Q).
From Lemma 3.14, we get, by the absolute irreducibility of the representations
involved,
ρsym⊗m ∼ sym⊗m
= ρg,p .
f,p
? ∗
Write, by (3.1.2), ρ?,p |Gal(Qp /Qp ) ∼
= 0 δ? with unramified δ? (? = f, g).
Then we conclude {(jf δfm−j |j = 0, . . . , m} = {(jg δgm−j |j = 0, . . . , m} from
ρsym⊗m ∼
= ρg,p
sym⊗m
. By (Gal), we have δ? ([γ, Qp ]) = 1 and (? ([γ, Qp ]) =
f,p
ζ? γ for γ ∈ Z×
k
p and a root of unity ζ? . Thus, we conclude that (jf δfm−j =
(jg δgm−j ⇔ ((f /(g )j = (δg /δf )m−j for all j. From (3.1.2), we conclude the
triviality (f /(g = 1 = (P /(P on the inertia group at p as δ? is unramified.
This is a contradiction, as we have chosen 1 = ζ = (P /(P ([γ, Qp ]). Thus, f
and hence g have complex multiplication by the same imaginary quadratic
field, and hence α1 /α0 is a root of unity by the explicit form of A(l) described
in Sect. 3.1.5.
3.2 Finiteness of Abelian Varieties 109
“modular” rank-2 motives proven by Brown and Ghate [BrG] Corollary 2.4.2,
once two rank-2 Grothendieck motives are (both) realized in the cohomology
of modular curves (not just as compatible systems), Galois endomorphisms
come from motivic endomorphisms (as Tate’s conjecture predicts).
An abelian scheme A of dimension g over a field κ of characteristic p is
called ordinary if we can embed the finite flat group scheme μgp into A over
an algebraic closure of κ. An abelian scheme over a number field F has a good
reduction at a prime p if A extends to an abelian scheme over the valuation
ring Op of p. It has potentially good reduction if there exists a valuation ring
V finite flat over Op over which A has good reduction. We say that A has
ordinary potential good reduction if the reduction modulo mV of the extended
abelian scheme over V is ordinary (for the maximal ideal mV of V ). In the
following conjecture, the term number field means a finite extension of Q.
Conjecture 3.16 For a given base number field F , fixing a prime ideal p of
F , there are only finitely many twist-equivalence classes of non-CM F -simple
abelian varieties of GL(2)-type with good reduction everywhere outside p and
potentially ordinary good reduction modulo p.
3.2.1 CM Type
Ord2 (I) = {P |AP has ordinary good reduction over Zp [μpr(P )+1 ]}.
By the corollary above, this set is a finite set. What are the numbers
Thus, we have slopes 0, 1/2, and 1, each of multiplicity 2, for the above
polynomial. These abelian varieties are associated with slope-0 forms but do
not have ordinary good reduction over Z13 . See [BrG] Table I for some other
information on this example. Numerically, we can find AP with nonordinary
reduction modulo p, but they are rather rare for small-level primes p; so
Ord2 (I) could be large although it is a finite set.
Proof. Though the result is valid even for p = 2, we assume p > 2 here for
simplicity. We use the notation introduced above the proposition. In partic-
ular, take K = Qp , and M/K is the quadratic extension specified above the
proposition. Let E be the subfield of Q generated over Q by the Hecke eigen-
values of f ; hence, we identify E with the E-linear endomorphism algebra
End0E (Af /Q ).
As we noted, Af has potentially good reduction modulo p. Take a finite
extension L ⊂ Qp of Qp such that Af extends to an abelian scheme A over
the integer ring of L. By extending L, we may assume that L contains the
field M and is a Galois extension of K. Let à be the special fiber of A. Define
×
another character ψ : Gal(Qp /M ) → Ql by ψ(g) = ϕ(τ gτ −1 ) for an element
τ ∈ Gal(Qp /K) inducing a nontrivial automorphism on M . Write ξL for the
restriction of ξ to Gal(Qp /L) ⊂ Gal(Qp /M ) for any character ξ of Gal(Qp /M ).
Then the action of Gal(Qp /L) on the λ-adic Tate module Tλ Ã is isomorphic
to ϕL ⊕ψL . The two characters ϕL and ψL are unramified. Thus, for the prime
element of L and its Artin symbol φ := [, L], ϕL (φ) + ψL (φ) ∈ E, and
ϕL (φ)ψL (φ) is equal to the value at φ of the l-adic cyclotomic character (after
extending L further if necessary); hence, we have ϕ(φ)ψ(φ) = |NL/Qp ()|−1 p .
Thus, P (X) = (X − ϕ(φ))(X − ψ(φ)) is in E[X]. Then the characteristic
polynomial over Q of the Frobenius endomorphism Φ of à over the residue
field F of L is given by σ:E→Q P σ (X) for conjugates P σ (X) of P (X) under
field embeddings σ : E → Q. If ()e = (p) in the integer ring of L, we get
ψ σ (φe ) = ψ σ ([p, L]) = ϕσ (τ [p, L]τ −1 ) = ϕσ ([pτ , L]) = ϕσ ([p, L]) = ϕσ (φe ).
This implies |ϕσ (φ)|p = |ψ σ (φ)|p , and by ϕ(φ)ψ(φ) = |NL/Qp ()|−1 p already
remarked, we have |ϕσ (φe )ψ σ (φe )|p = |ϕ(φ)|2e
p = |ψ(φ)|p = |NL/Qp (p)|p . So
2e
we get |ϕσL (φ)|p = |ψ σ (φ)|p = (NL/Q (p))−1/2e . Thus, 0 < |ψ σ (φ)|p < 1, and
hence, all eigenvalues of Φ have a positive slope between 0 and 1. Thus, Ã
cannot be ordinary, and, moreover, Ã[p](Fp ) = {0}.
We prepare
Proposition 3.26 Let F = {fP }P ∈Spec(I)(Qp ) be a p-adic analytic family of
classical p-ordinary Hecke eigenforms and A ⊂ Spec(I)(Qp ) be an infinite
set of arithmetic points P with r(P ) ≤ r for a fixed r ≥ 0 VA (I) is a finite
extension of Q. Assume for P0 ∈ A that
1. α0 = aP0 (p) is a Weil number,
2. Σα0 = σ : Q(α0 ) → Q : |ip (ασ0 )|p = 1 is a CM type of Q(α0 ).
Then there exists a Weil p-number α of weight 1 with |ip (α)|p = 1 such that
a(p, fP ) = ζ(εP (γ))logp (α)/ logp (γ) αk(P ) for a root of unity ζ for all arithmetic
P with k(P ) ≥ 1, where α = expp (logp (ip (α))).
Proof. In this book, to simplify the argument, we only deal with the case where
M := VA (I) is an imaginary quadratic field and r = 0; hence, M = Q(α0 )
(see [H11a] Proposition 7.2 for the general case). Since we assumed r = 0, we
can forget about εP (γ) = 1. Take P ∈ A with k(P ) > 1. Then αp,P is a Weil
number of weight k(P ) > 1 with |αp,P |p = 1. Thus, (p) has to split in M , and
so (p) = pp in M . Thus, Σαp,P consists of the single element ι = ip |M , and
for each k, there exists at most one Weil number αk ∈ M of weight k (up to
roots of unity in M ) such that |αk |p = 1. In M , (αk ) = pk for the prime
√ ideal
p of M corresponding to ip |M . Fix such a k. Taking a kth root α = k αk , we
have αl = αl up to roots of unity for all l as (αl ) = pl .
Since A is an infinite set, we find an infinite sequence, P1 , P2 , . . . , Pn , . . .
in A, with increasing weight k(P1 ) < k(P2 ) < · · · such that (aPj (p)) = pk(Pj )
for all j > 0. Put
1
α = exp( logp (a(p, fP0 )) = exp(logp (α)).
k(P0 )
Since (aPj (p)) = p(k(Pj )) , aPj (p)/αk(Pj ) is a Weil number of weight 0; that is,
it is an algebraic integer with all its conjugates having absolute value 1. Then,
by Kronecker’s theorem, we find aPj (p) = ζPj αk(Pj ) for a root of unity ζPj .
Note that α is contained in a finite extension M /M . Since there are finitely
3.4 Nonconstancy of Adjoint L-Invariant 119
Suppose that VA (I) is a finite extension and assume the existence of an arith-
metic point P0 as in the theorem. Therefore, the assumptions of Proposi-
tion 3.26 are met. By Proposition 3.26, we find a Weil number α of weight 1
and a power series Φα (T ) ∈ W [[T ]] such that a(p, fP ) = Φα (εP (γ)γ k(P ) −1) =
ζ(εP (γ))logp (α)/ logp (γ) αk(P ) for all arithmetic P , where ζ is a root of unity
independent of P ; in short, a(p) = Φα ∈ W [[T ]] ⊂ I. Then, for the entire set
B of arithmetic points P with k(P ) = 1, we find HB (I) ⊂ Q(μp∞ (p−1) )(ζ, α),
which is a finite extension of Q(μp∞ ). By the horizontal theorem, I has com-
plex multiplication. The converse is easier (in the same manner as in the proof
of Theorem 3.2; see [H11a] Sect. 4 for more details).
We could make the following conjecture stronger than Conjecture 3.24:
Conjecture 3.27 Let A ⊂ Spec(I)(Qp ) be an infinite set of arithmetic points
P with bounded level r(P ) ≤ r. Suppose that I does not have complex multi-
plication. Then we have
lim sup[Q(a(p, fP )) : Q] = ∞.
P ∈A
FI = F = {fP |P ∈ Spec(I)(Qp )}
We prove this theorem assuming a formula relating da(p)dT with the L-invariant
(Theorem 3.31); after the proof, we revisit Greenberg’s definition of his L-
invariant in Sects. 3.4.2–3.4.5 and prove the derivative formula in Sect. 3.4.6.
By this theorem, if F is a non-CM family, P → L(Ad(ρP )) is a nonconstant
function; hence, except for finitely many Galois representations in the family,
Greenberg’s conjecture (see [Gr]) predicting the nonvanishing of L(Ad(V )) is
true. We have a Hilbert modular slope-0 family as described in [HMI]. We can
make the following conjecture for Hilbert modular families.
Conjecture 3.29 For a slope-0 parallel weight family of Hilbert modular
Galois representations {ρP }P ∈Spec(I) of Gal(Q/F ) for a totally real field F ,
P → L(Ad(ρP )) is constant if and only if the family has CM.
If the conjecture holds, the L-invariant L(IndQ
F Ad(ρP )) vanishes only over a
thin Zariski closed subset of Spec(I).
Before starting the proof of the theorem, we prepare to state another form
of the rigidity lemma. Identify Λ with W
n
[t, t−1 ] = limn W [t]/(tp − 1) by
←−
t = 1 + T . Then t → t for z ∈ Zp extends linearly to an endomorphism ιzn
z
n n n
of the ring W [t]/(tp − 1). Indeed, in W [t]/(tp − 1), tp = 1; hence, cutting
∞
the p-adic expansion z = m=0 am p (with integers 0 ≤ am ≤ p − 1) to
m
n−1 n
z0 = m−0 am pm , we legitimately define tz by tz = tz0 in W [t]/(tp − 1),
p −1
n p −1
n
and we then define ιzn ( j=0 cj tj ) = j=0 cj tzj . Clearly, ιzn is a W -algebra
n
endomorphism of W [t]/(tp −1). Then passing to the limit, z ∈ Zp gives rise to
an algebra endomorphism ιz = limn ιzn of Λ. In this way, Z× acts on Λ through
←− ∞ p
z → ιz ∈ Aut(λ/W ). Note that tz = (1 + T )z = n=0 nz T n ∈ W [[T ]].
W Λ as W [t, t −1 , t , t −1 ] = lim W [t, t−1 , t , t −1 ]/(tp −1, t p −1)
n n
Write Λ⊗
←− n
for the variable t of the left factor and t of the right factor, and let z ∈ Z× p
act on W [t, t
−1 , t , t −1 ] by (t, t ) → (tz , t z ) for z ∈ Z× . Here is a version of
p
the rigidity lemma (Lemma 3.12), which we prove later as Lemma 4.85 after
some preparation:
3.4 Nonconstancy of Adjoint L-Invariant 121
We assume p > 2 in the rest of this section. We may assume that the integral
closure of Zp in I is equal to W (by extending scalars if necessary). As we will
prove in the following sections as Theorem 3.31 (for which we need to assume
Conjecture 3.32), L(Ad(ρP )) is a constant multiple of
da(p)
a(p)−1 X=0
,
dX
da(p) da(p)
a(p)−1 (1 + T ) = a(p)−1 t .
dT dt
db da db da d logp (a) 1
= = =s·a· = s
dx dx da dx da a
from the chain rule. Then b must be a linear function of x with slope s:
Z ∼= Spf(J) in G m × G m (see Sect. 4.4 for formal scheme and the functor
∞
“Spf”). Since Φ(t) is a power series ts = n=0 ns T n ∈ F [[T ]] convergent
over a closed disk D = {x ∈ F : |x|p ≤ (} for some ( ∈ |F × |p , it is in an
affinoid ring F {T }, which is the convergent power series ring over D (see
[Cd] for rigid geometry and general affinoid rings). Since the action t → tz for
z ∈ Z×p extends to F {T }, we have Φ(t ) = Φ(t) in F {T } as Φ(t) = t . Thus,
z z s
L∞ (1, E)
Lp (1, E) = (1 − a−1
p ) ,
period
L∞ (1,E)
to the Archimedean value period times a theoretically explicit error factor
Lan (E), the so-called L-invariant:
L∞ (1, E)
Lp (1, E) = Lan (E) .
period
The problem regarding L-invariants of motives is to find an explicit formula in
terms of their p-adic realization without recourse to p-adic L-functions. Writ-
×
ing E(Qp ) = Qp /q Z for the Tate period q ∈ pZp , the solution [Theorem 1.5
(2)] conjectured by [MzTT] and proved by Greenberg–Stevens [GS] is
logp (q)
Lan (E) = .
ordp (q)
da(p)
Lan (E) = −2 logp (γ) ,
dX X=0
where γ is the generator of Γ = 1 + pZp corresponding to 1 + X under the
identification Zp [[Γ ]] = Zp [[X]]. These formulas are actually proven in [GS]
for abelian varieties A split multiplicative at p associated with weight-2 cusp
forms in place of elliptic curve E.
Here is an analogous formula for Greenberg’s L-invariant of Ad(ρP ) in
[H04a] whose proof we give in Sect. 3.4.6 after recalling Galois deformation
theory and Greenberg’s definition of his L-invariant:
Theorem 3.31 Let p be an odd prime. Then we have
da(p)
L(Ad(ρP )) = −2 logp (γ)aP (p)−1 .
dX X=0
In particular, if fP = fE as above, we have
logp (q)
L(Ad(ρP )) = Lan (E) = .
ordp (q)
The last formula follows from the first and the Greenberg–Stevens’ formula
we described before stating the theorem. We prove the first assertion in the
rest of this section.
124 3 Invariants, Shimura Variety, and Hecke Algebra
ι ◦ det(ρ) = det(ρA );
∗
(D3) we have an exact sequence ρA |Dp ∼
= 0A δA with δA ≡ δP mod mA .
Condition (D3) is the near-ordinarity, and we call the character δA of Dp the
nearly ordinary character of ρ. By the work started by Wiles/Taylor (and
practically ended by Kisin), we know the following conjecture for most cases:
Note here that IP ∼= κ(P )[[X]]. Though we have taken the field κ as a
base ring, we could have taken W = W (F) in place of κ and the deformation
functor from the category of Artinian W -algebras similarly defined. This type
of deformation over W was studied by Taylor–Wiles, Kisin, and others, and
from their results, we can conclude the above conjecture in the following cases:
1. ρP is p-distinguished
√ [as defined above (D1)] and ρ := ρmI is absolutely
irreducible over Q[ p∗ ] for p∗ = (−1)(p−1)/2 p (e.g., essentially the original
R = T theorem of Wiles/Taylor; see [DiFG], [MFG] Theorem 5.29, and
[HMI] Proposition 3.78).
2. This case is via the theory of eigenvariety (cf. [Em]). Supposing that ρ
is absolutely irreducible, the conjecture holds if the slope is less than
k(P )/2, since the eigenvariety is étale around P over the arithmetic point
below P on the weight space; see also [Ki1] Theorem 11.10). The result
includes the identification of the rigid-analytic full (dimension-3) Galois
deformation space with the maximal spectrum of the full Hecke algebra
(p-inverted) by Gouvêa/Mazur, Coleman/Mazur, Kisin, and Chenevier
(see [Ki2] Theorem 8.2, [Cn1], and [Cn2]). Here the word “full” means we
do not impose reducibility on the restriction of the deformation to Dp .
3. ρ ∼
= χ1 ⊕ χ2 for two Galois characters with χ1 |Gal(Q/K) = χ2 |Gal(Q/K) for
K = Q[μp∞ ] ([Ki1] Theorem 11.10 and [Ki2] Theorem 8.2).
The only remaining case is when ρ is reducible and χ1 |Gal(Q/Q[μp∞ ]) =
χ2 |Gal(Q/Q[μp∞ ]) (which means p-distinguishedness as the representation is
odd). In this exceptional case, the Skinner–Wiles result [SW] applies (so at
least IP is equal to the reduced part of the universal ring).
Case (1) is the combination of the result in [DiFG] and the method in
[HMI]. Indeed, if we have an R = T theorem for one arithmetic point P , it
extends by the method of [HMI] Chap. 3 to the entire connected component
Spec(T) of Spec(h) containing Spec(I). The specific one point P with R = T is
essentially covered by either [DiFG] or [HMI] Chap. 3 (especially Sect. 3.2.4).
In cases (2) and (3), the full Galois deformation ring R [without imposing
(D3)] around an arithmetic point is smooth over the weight space by the work
quoted there. The full deformation ring often does not have information on
a(p) [or U (p)], because we often have a deformation whose restriction on Dp
does not satisfy (D3). Then the natural projection Spec(R) → Spec(R) (for
the ordinary deformation ring R) has an image on one of the two leaves (of
the infinite fern) crossing at P if the slope ordp (aP (p)) is 0 [or more gen-
erally, ordp (aP (p)) is less than k(P )/2]; hence, it is étale around the point
P̃ ∈ Spec(R) coming from the arithmetic point P ∈ Spec(I). Thus, we have
étale smooth
Spec(RP̃ ) −−−→ Spec(RP ) −−−−→ Spec(W [[T ]]P ), which shows that the com-
= IP = κ(P )[[X]].
posite is étale, and hence RP̃
In the following sections, we start with a brief review of Greenberg’s defi-
nition of the Selmer group and his L-invariant.
126 3 Invariants, Shimura Variety, and Hecke Algebra
We describe Greenberg’s definition of his Selmer group associated with the ad-
joint square Galois representation. For simplicity, we assume that S = {p, ∞}
(so N = 1). One can find the definition in the general case in [Gr] and in
[HMI] Sect. 1.2.3. We may assume that κ has a p-adic integer ring W . Let
QS be the maximal extension over Q unramified outside S. All Galois coho-
mology groups are continuous cohomology groups described in [MFG] 4.3.3.
Write GS = Gal(QS /Q) and Ip for the inertia subgroup of the decomposition
subgroup Dp ⊂ GS . The group Dp is expected to be isomorphic to Gal(Qp /Q)
(and this fact is close to be true; see [CnC]).
Write V for the space of ρP . Let GS act on Endκ (V ) by conjugation and
put Ad(V ) ⊂ Endκ (V ) (the trace 0 subspace of dimension 3). We have a
filtration:
V F + V {0} (ord)
stable under the decomposition group Dp such that Dp acts on the quotient
V /F + V by δP . Then Ad(V ) has the following three-step filtration stable un-
der Dp :
Ad(V ) ⊃ F − ⊃ F + ⊃ {0}, (F)
where
F − = F − Ad(V ) = {φ ∈ Ad(V )|φ(F + V ) ⊂ F + V } (upper triangular),
F + = F + Ad(V ) = {φ ∈ Ad(V )|φ(F + V ) = 0} (upper nilpotent).
Ad(V )
Up (Ad(V )) = Ker(Res : H 1 (Dp , Ad(V )) → H 1 (Ip , )).
F + (Ad(V ))
Then we define
H 1 (Dp , V )
Sel(Ad(V )) = Ker(H 1 (GS , Ad(V )) → ). (3.4.1)
Up (V )
Ad(V )
Up− (Ad(V )) = Ker(Res : H 1 (Dp , Ad(V )) → H 1 (Ip , ))
F − (Ad(V ))
for p|p, we define a bigger “−” Selmer group Sel− (Ad(V )) ⊃ Sel(Ad(V )).
In the above definition, replacing GS by the stabilizer GS∞ := Gal(QS /Q∞ )
of the cyclotomic Zp -extension Q∞/Q and Ad(V ) by A = Ad(V )/L for a Ga-
lois stable lattice L of Ad(V ), one can define the Selmer group SelQ∞ (A).
The characteristic power series Φ(T ) of the Pontryagin dual of SelQ∞ (A) is
conjectured to be the adjoint p-adic L-function Lp (s, Ad(ρP )) (the adjoint
3.4 Nonconstancy of Adjoint L-Invariant 127
We have the Poitou–Tate exact sequence (e.g., [MFG] Theorem 4.50 (5)):
H 1 (Dp , Ad(V ))
0 → Sel(Ad(V )) → H 1 (GS , Ad(V )) → → Sel(Ad(V )∗ (1))∗ .
Up (Ad(V ))
H 1 (Dp , Ad(V ))
H 1 (GS , Ad(V )) ∼
= . (I)
Up (Ad(V ))
Greenberg defined in [Gr] his invariant L(Ad(V )) in the following way. Write
F − H 1 (Dp , Ad(V )) for the image of H 1 (Dp , F − Ad(V )) in H 1 (Dp , Ad(V )). By
F − H 1 (Dp ,Ad(V ))
the definition of Up (Ad(V )), the subspace inside the right-
Up (Ad(V ))
−∼
hand side of (I) is isomorphic to Sel (Ad(V )) = κ. Namely, we have
Note that
Hom(Dpab , κ) ∼
=κ×κ
L : κ → κ,
Hom(Dpab , κ) is δP−1 ∂δ|X=0 . We know that δP−1 δ([p, Qp ]) = aP (p)−1 a(p) and
δP−1 δ([u, Qp ]) = (ζγ k )− logp (u)/2 logp (γ) tlogp (u)/2 logp (γ) = (1 + X)logp (u)/2 logp (γ)
3.4 Nonconstancy of Adjoint L-Invariant 129
da(p)
c([p, Qp ]) = aP (p)−1 ,
dX X=0
(3.4.2)
d(1 + X)logp (u)/2 logp (γ)
c([u, Qp ]) = .
dX X=0
Taking u = γ s , we have
dδ([γ s , Qp ]) s dδ([γ s , Qp ]) 1
= − and logp (γ s )−1 = − .
dX X=0 2 dX X=0 2 logp (γ)
Thus, we get the desired result from the definition of L(Ad(V )).
κ,k )
Lp (λ Lp (γσκσ − 1, γ k − 1) π κ L(0, ψλκ,k )
:= = ∗E(ψλκ,k ) for γσ = 1 + p.
ΩpkΣ+2κ ΩpkΣ+2κ Ω∞kΣ+2κ
1 → O×
p /{±1} → C → Cl → 1
Gk,ζ (a)
and La−1 (0, ε̃ζ λ0,k ),
ε̃ζ λ0,k (a)
where “” indicates that we need to multiply the Euler-like factor E(?).
134 3 Invariants, Shimura Variety, and Hecke Algebra
a1−c ∈ C×
δkκ Gk (a1−c τ )|τ =0 a1−c κ δkκ Gk (z0 (a)) δ κ Gk (z0 (a))
= = k La−1 (0, λκ,k ).
λ0,k (a) λ0,k (a) λκ,k (a)
There is a canonical p-adic Serre–Tate parameter t around z0 (as a point
of a modular curve). Heuristically, logp (t) behaves like τ : τ = 0 ⇔ t = 1 and
1−c
t ◦ ρ(α) = tα (see Proposition 7.44 for a proof). For θ := t ddt , by Katz, with
a specific p-adic period Ωp ∈ W × of E(O) (we recall later),
κ,k ) θκ (Gk (τ a ))|τ =1
Lp (λ
1−c
δ κ (Gk (Ω∞ a)) π κ L(0, λκ,k )
= = k
Ωpk+2κ a
λ0,k (a) a
λκ,k (a) Ω∞k+2κ
for a running through ideal classes. Thus, we can compute the Taylor expan-
Gk (ta1−c )
sion of E = a λ0,k (a) with respect to x = logp (t) by computing the
derivative with respect to θ.
Since E is defined over W , out of this identification of the Taylor expansion,
we conclude that Lp (x, γ k − 1) is almost the expansion of E with respect to
x T = t − 1 (which we call t-expansion by an abuse of symbol instead of
T -expansion, as this T has nothing to do with the weight variable “T ” of the
Hecke algebra h). Strictly speaking, first, the t-expansion is the expansion of
the measure given by E as a measure on Zp , not on 1 + pZp . Second, we want
to know the nonvanishing of the t-expansion modulo p of the restriction of the
measure on Cl(p∞ ) to 1 + pOp ∼ = Γ 2 . Thus, we need to replace G := {Gk,ζ }
by a family {Gk,ζ,b } of Eisenstein series of level p2 (somehow restricting the
measure to the torsion-free part 1 + pOp increases the level; see Sect. 9.2,
as this is subtle). Since La−1 (s, λ) (resp. G) is a sum of partial L-functions
for a class modulo
p (resp., a sum of Eisenstein series of level p2 ), we have
La−1 (s, λ) = b≡a mod p,[b]∈ClM (p) Lb−1 (s, λ), and
(tb )
1−c
Gk,ζ,b
Eζ =
ε̃ζ λ0,k (b)
b∈ClM (p)
3.5 Vanishing of the μ-Invariant of Katz L-Functions 135
F[[(t − 1)]] over F, which enables us to compute the μ(Lp (x, ζγ k − 1)) [via
the almost identity Lp (x, ζγ k − 1) the t-expansion of Eζ under x t − 1]
c−1
by q-expansion of fζ,b = c∼b Gk,ζ,b (τ α(c) ). This is because Eζ is a linear
combination of fζ,b ((tb
1−c
). Our conclusion is
lowing functor defined over Q-algebras (see Theorem 7.11 in Sect. 7.1.3 for a
proof):
E Q (A) = {(E, η : (A(∞) )2 ∼
= V (E))/A }/isogenies.
Here A(∞) is the finite adele ring. Then g ∈ G(A) sends a point (E, η)/A ∈
E Q (A) to (E, η ◦ g (∞) )/A for the projection g (∞) of g to A(∞) .
Take the quotient Y (p) = limpN Y (N ) = Y /G(Zp ). Put V (p) (E) =
←−
T (E) ⊗Z A(p∞) , and consider the prime-to-p-level structure η (p) : (A(p∞) )2 ∼
=
V (p) (E). By Theorem 7.13, Y (p) over Z(p) represents the following functor
defined over Z(p) -algebras:
write α ◦ η (p) = η (p) ◦ ρ(p) (α) for ρ(p) (α) ∈ G(A(p∞) ). Since α is “non-
central” (i.e., α ∈ Z), ρ(p) (α) is not in the center of G(A(p∞) ). Thus, if
z0 = (E, η (p) ) ∈ Y (p) (A) (A = W and F), we find that ρ(p) (α)(z0 ) = z0 ,
and
× ∼
O×
(p) /Z(p) −
→ {g ∈ Aut(Y (p) )|g(z0 ) = z0 }.
ρ
◦ dt
ω(O) = Ωp · (
ηp,∗ ) for Ωp ∈ W × . (3.5.1)
t
This is the p-adic period we discussed earlier.
Write ηp = (η ◦ , ηpet ) : μp∞ × Qp /Zp ∼ ∞
= E[p∞ ] × E[p ], and define a homo-
×
morphism ρp of O(p) into the diagonal torus of G(Zp ) by α ◦ ηp = ηp ◦ ρp (α)
for α ∈ O(p) . Thus, ηp◦ ◦ ρp (α) = αηp◦ , identifying Op with Zp and ηpet ◦ ρp (α) =
αc ηpet .
Fix a base w1 , w2 of O (p) ∼ (p) , and identify M (p∞) with
= T (p) (E) over Z A
(A (p∞) 2
) . The choice induces a prime-to-p-level structure η (p) : (A(p∞) )2 ∼ =
O⊗Z A(p∞) = V (p) (E). We put η = ηp ×η (p) . Define ρ : O× (p) → G(Zp ×A
(p∞)
)
×
by η ◦ ρ(α) = α ◦ η. Since α ∈ O(p) induces an isogeny α : E → E sending
αη (p) = η (p) ρ(p) (α), the point z0 (O) = (E(O), η (p) ) ∈ Y (p) = Y /G(Zp ) is
fixed by ρ(α). Pick a fractional ideal a ⊂ M prime to p; hence, a = (aO) ∩M
for an idele a ∈ MA× with ap = a∞ = 1. Then we have z0 (a) = (E(a), η(a)) =
ρ(a)−1 (z0 (O)).
Consider the formal completion Y = Yz/W of Y/W along z = z0 (a) ∈
(p)
Y (p) (F). See Sect. 4.4 for completing the process of a scheme. Then, by the
universality of Y (p) , Y satisfies
Y (A) ∼
= E(A) := X/A : elliptic curves over AX ⊗A F = E(a)/F / ∼ =,
For this identification, we used ηp◦ : μp∞ ∼ = E(a)[p∞ ] and its dual inverse ηpet :
Qp /Zp ∼= E(a)[p∞
]. The above identification only depends on ηp = (ηp◦ , ηpet ),
and multiplication by an idele a with ap = 1 does not affect ηp . Thus the
above identification is independent of a and a.
Since ρ(α) fix z0 (a), it acts on Y . Since α ∈ O acts on X[p∞ ]et = Qp /Zp
by α → αc and on X[p∞ ]◦ = μp∞ by α → α, only reasonable action of M × on
the deformation space t ∈ Hom(Qp /Zp /A , μp∞ /A ) = G m is by t → α◦t◦α−c =
1−c
tα . This turns out to be true, as we verify as Proposition 7.44 in Sect. 7.2.5
(see also, [H10a] Proposition 3.4):
This can be proven via Chai’s theory of Hecke invariant subvariety of the
Shimura variety. We sketch the proof in the following section (and give full
details in Chap. 11).
3.5.4 Conclusion
Theorem 3.38
μ(Lp (t − 1, γ k ζ − 1)) = min(ordp (a(n, fζ,aj )),
n,j
0 ≤ μ(Lp ) ≤ inf μ(Lp (t − 1, ζγ k − 1)) ≤ inf (ordp (1 + ζ logp ()/ logp (γ) #k−1 )) = 0,
ζ ζ
as the p-power order of the root of unity ζ logp ()/ logp (γ) grows indefinitely.
This concludes the sketch of the proof of Theorem 3.37 in this chapter.
More generally, as Theorem 9.15, we will prove the following:
Corollary 3.41 Suppose F = Q. Then the μ-invariant of the anticyclotomic
Katz p-adic L-function L−
p (x) = Lp (x, 0) also vanishes.
When F = Q, writing L− −
p (xσ ) = Lp (xσ , 0), μ(Lp ) could be positive though
μ(Lp (xσ , y)) = 0 always (see [H11b]). This rare positivity μ(L− p ) > 0 only
occurs if [F : F (μp )] = 2 and M/F is unramified everywhere at finite places
[see (M1–M3) in [H10a] for a precise set of conditions for μ(L− p ) > 0].
Hilbert modular Shimura varieties). Any essential ingredients for the proof
of the general case show up in the simplest case of modular curves. Write
G = GL(2)/Z with center Z ∼ = Gm/Z . In this section, the word “variety”
means a finite-type reduced scheme over F = Fp .
We recall the following lemma we mentioned already in Sect. 3.5.1:
Lemma 3.42 Identifying Y with G m = Spf(lim W [t, t−1 ]/(t − 1)pn ), if α ∈
←−n
M × , we have ρ(α)(t) = tα
1−c
for complex conjugation c.
Note that if α ∈ M × is not prime to p, the action of ρ(α) is an endomorphism
of Y (p) , not an automorphism. We prove this as Proposition 7.44 (more general
cases are treated in [H10a] as Proposition 3.4). The action of ρ(α) on the Serre–
1−c
Tate coordinate is given by t → tα factoring through G(A(p∞) )/Z(Q), since
Z(Q) acts trivially on the Shimura variety Y (p) .
(α ∈ T ). Then the formal completion H along z0n is also stable under T , since
n
z0 is fixed by T . By the Serre–Tate theory, H ⊂ In ∼ =Gn .
m/F
As we have seen, if a power series Φ(T ) = Φ(t) (T = t − 1) satisfies
Φ(tz ) = Φ(t)z for all z in an open subgroup of Z×p , then Φ(t) = t for s ∈ Zp
s
m = Spf(W
G [t, t−1 ]) = Spf(W [[T ]]).
n ) of
where L runs over the (finitely many) Zp -direct summand of X∗ (G m
rank r.
The proof is given in [Ch3] but is technical and long, so we postpone the proof
of this lemma until Chap. 10.
We apply this to the formal completion H along z0n = (z0 , z0 , . . . , z0 ) ∈ I n
inside I = G
n . Since H is stable under t → tα1−c for α ∈ O× /Z× 1−c
n
m (p) (p) → Z× ,
p
by continuity, H is stable under the closure Z× of {α1−c |α ∈ O× }. Since H is
p (p)
an excellent irreducible scheme, H is reduced equidimensionally of dimension
n − 1. Thus, by the above lemma,
+
H= m ⊗Z L ⊂ G
G n = In , (3.6.1)
p m
L
where L runs over the (finitely many) Zp -direct summand of X∗ (In ) of rank
n − 1. Let H → H be the normalization of H. Since H is irreducible, H is
irreducible. By (3.6.1), each point over z0n of H is indexed by {L}, and for the
point yL ∈ H over z0n corresponding to L, H yL is étale over G
m ⊗ L. Write
I = I × I for I = I
n n−1
and I = I for the last component.
Lemma 3.46 The scheme H is finite flat over I around z0n−1 . In particular,
each L is of rank n − 1 and projects down to an open subgroup of X∗ (G n−1 ) ∼
=
m
n−1
Zp . If one of L surjects down onto X∗ (I ) (I = Gm ), all of L surjects
n−1
down onto X∗ (I ), and the projection H → I is étale finite around z0n−1 .
142 3 Invariants, Shimura Variety, and Hecke Algebra
datum (cf., [GME] Sect. 1.11) with respect to O/O, obtaining the desired U .
Proof of Theorem 3.43. Since the two projections πj : H → I (j = L, R)
are dominant, we have End(A) ⊗ Q = End(B) ⊗ Q = Q. Let Y/H = A ×H B.
Thus, there are only two possibilities for EndQ (Y) = End(Y/H ) ⊗ Q: Either
EndQ (Y) = Q × Q or EndQ (Y) = M2 (Q).
Suppose that EndQ (Y) = M2 (Q). By the semisimplicity of the category
of abelian schemes, we have two commuting idempotent e? ∈ EndQ (Y) such
that eA (Y) = A and eB (Y) = B. Since EndQ (Y) = M2 (Q), we can find
an invertible element β̃ in G(Z(p) ) ⊂ M2 (Q) such that β̃ ◦ eA = eB ; hence,
β̃ : A → B is an isogeny with β̃ ◦ ηA = ηB , whose specialization to the fiber
of A and B at y gives rise to an endomorphism β ∈ End(E(O)) ⊗ Q. Thus,
the isogeny β̃ is induced by ρ(β), and we conclude that Δ1,β = H.
We suppose that EndQ (Y) = Q×Q and try to get a contradiction [in order
to prove that EndQ (Y) = M2 (Q)]. We pick a sufficiently small open compact
subgroup K ⊂ G(A(p∞) ) maximal at p so that the normalization HK of
HK ⊂ YK × YK is smooth at the image of y. The variety YK is naturally
defined over a finite extension Fq /Fp as the solution of the moduli problem
E (p) /K. The universal elliptic curve EK is therefore defined over IK/Fq , and
HK is a finite-type variety over Fq . Let η be the generic point of HK/Fq , and
write η for the geometric point over η and Fq (η)sep for the separable algebraic
closure Fq (η)sep of Fq (η) in Fq (η).
Take an odd prime # = p, and consider the #-adic Tate module T (Yη )
for the generic fiber Yη of Y. We consider the image of the Galois action
G := Im(Gal(Fq (η)sep /Fq (η))) in GLO ×O (T (Yη )). Then, by Zarhin’s result
([DAV] Theorem V.4.7), the Zariski closure over Q of G is a reductive subgroup
G of GLQ ×Q (T (Yη )⊗Q), and G is an open subgroup of G(Q ). Moreover, by
Zarhin’s theorem, the centralizer of G in EndQ ×Q (T (Yη )⊗Q) is End(Y)⊗Q.
Since the reductive subgroups of GL(2) either are tori or contain SL(2) (cf.
[GME] Corollary 4.3.14 in Sect. 4.3.5), the derived group G1 (Q ) of G(Q ) has
to be SL2 (Q × Q ).
By a geometric version of Chebotarev’s density theorem for function fields
(cf. [Se1] Theorem 7; see also [L]), we can find a set of closed points u ∈
HK (F) with positive density such that the Zariski closure in G of the subgroup
generated by the Frobenius element F robu ∈ Im(Gal(Fq (η)sep /Fq (η))) at u
with πj (u) = uj [uj ∈ IK (F)] is a torus containing a maximal torus Tu =
(Tu1 × Tu2 ) ∩ G1 of the derived group G1 of G. In particular, the centralizer
of Tu in G1 is itself. Thus, Yu is isogenous to a product of two nonisogenous
elliptic curves Y1 = Eu1 and Y2 = Eu2 defined over a finite field.
The endomorphism algebra Mj = EndQ (Yj ) is an imaginary quadratic field
of Q generated over Q by the relative Frobenius map φj induced by F robu ,
and M1 = M2 . The relative Frobenius map F robu acting on X∗ (Iu1 ) ∼= Zp has
(1−c)σ
one eigenvalue: φ1 for the CM type Σ1 = {σ} of Y1 , which differs from
144 3 Invariants, Shimura Variety, and Hecke Algebra
{t, t β 1−c
)|t ∈ Gm } for nonzero β ∈ O(p) . Suppose that y corresponds to L,
and so H m ⊗L. On the other hand, we have the skew-
y ⊂ I× I coincides with G
diagonal Δβ = Δ1,β = {(z, ρ(β)(z))|z ∈ I} ⊂ I ×I. The formal completion Δ β
along (z0 , z0 ) therefore coincides with H y and Δβ = Gm ⊗ L ⊂ H (z ,z ) inside
0 0
I . Thus, Δβ ⊂ H. By the irreducibility of H, we conclude that H = Δβ .
2
There are two ways of proving Theorem 3.44. One is an induction reducing
things to Theorem 3.43, whose details we give in Chap. 11. Here is a brief
sketch of the second proof, which is based on the argument in [Ch1] Sect. 8. By
using Chai’s globalization of the Serre–Tate coordinate in [Ch1] Sect. 2, we find
a dense open subscheme Y lin ⊂ Y [y ∈ Y lin ] such that Y is locally linear at
every closed point of Y lin (see Sect. 11.1.1 for local linearity). The existence of
Y lin ⊂ Y follows basically from Proposition 5.3 in [Ch1] and its proof (applied
to Y not X). Consider the abelian scheme Y = Em ×V m Y . Then EndQ (Y/Y ) is
isomorphic to either Qm or Qm−2 ×M2 (Q), because Y is dominant over V m−1
and V by the two projections. Then in a manner similar to the above sketch
of the proof of Theorem 3.43, we can prove the impossibility of EndQ (Y/Y ) ∼ =
m Q
Q . Thus, we have End (Y/Y ) = Q ∼ m−2
× M2 (Q); in other words, for an
index i < m, the ith factor Yi obtained by pulling back the ith factor E
of Em to Y is isogenous to the last factor Ym . This isogeny is induced by
a nonzero α/β ∈ M ∼ Q
= End (Ex ) with α, β ∈ O. Then we conclude that
X ⊃ S m−1
× Δ(α,β) for Δ(α,β) plugged in the product of the ith and mth
copies of V in V m , and hence X has the desired form.
4
Review of Scheme Theory
Thus, we write this closed set as V = V (I) for the ideal I. Since k[X] is
noetherian, an ideal I is generated by finitely many polynomials; hence, any
Zariski closed subset has the form V (I) with an ideal I.
5 V (k[X]) = ∅, V ((0)) = k , V (I ∩ J) = V (I) ∪ V (J),
n
It iseasy to verify
and V ( i∈I Ii ) = i∈I V (Ii ). Thus, the family of subsets of the form V (I)
for ideals I satisfies the axiom of closed subsets of k n , giving the Zariski
√
topology on k n . √
By Hilbert’s zero theorem (cf. [CRT] Sect. 5), V (I) = V ( I)
for the radical I = {f ∈ k[X]|f n ∈ I for some 0 < n ∈ Z}. Thus, the
association {ideals of k[X]} → {Zariski closed subsets of k n } given by I →
V (I) is surjective but not injective.
For any k-algebra R, define
4.1.2 Categories
We give a brief outline of the definition of categories. A category C consists
of two data: objects of C and morphisms of C. For two objects X and Y of C,
we have a set HomC (X, Y ) of morphisms satisfying the following three rules:
(Ct1) For three objects X, Y , Z, there is a composition map:
HomC (Y, Z) × HomC (X, Y ) → HomC (X, Z) : (g, f ) → g ◦ f.
f g h
(Ct2) (Associativity). For three morphisms X − → Y −→ Z − → W , we have
h ◦ (g ◦ f ) = (h ◦ g) ◦ f .
(Ct3) For each object X, there is a specific element 1X ∈ HomC (X, X) such
that 1X ◦ f = f and g ◦ 1X = g for all f : Y → X and g : X → Z.
E −−−−→ E
⏐ ⏐
⏐ ⏐
. .
S −−−−→ S
with the top arrow sending the origin of E to that of E (see Sect. 6.1.1 for a
more precise description of this).
A category C is a subcategory of C if the following two conditions are
satisfied:
(i) Each object of C is an object of C and HomC (X, Y ) ⊂ HomC (X, Y ).
(ii) The composition of morphisms is the same in C and C .
A subcategory C is called a full subcategory of C if the sets of homomorphisms
are equal: HomC (X, Y ) = HomC (X, Y ) for any two objects X and Y in C .
4.1.3 Functors
Example 4.2 Let X be a topological space. Then the category O(X) of open
sets consists of open subsets and inclusions:
Inc : U → V if U ⊂ V
HomO(X) (U, V ) =
∅ otherwise.
Spec(A) for SpecB (A). Although Spec(A) has another aspect of a topological
local ringed space as described in, for example, [GME] Sect. 1.2, here we con-
centrate on its functorial property. Since the category SCH/B of B-schemes
is a full subcategory of COF (ALG/B , SET S), this is legitimate (see [GME]
Sect. 1.4.3 and Lemma 4.17 in the text). Hereafter we simply write CB for
COF (ALG/B , SET S), which is the biggest “category” containing our geo-
metric objects. As we remarked already, COF (ALG/B , SET S) is not really
a category in the true sense [i.e., HomCOF (F, G) may not be a set], but our
category of schemes SCH/B is defined later, imposing a sheaf condition to
ensure that it is a category by Yoneda’s lemma (cf. Lemma 4.17).
The set SA (R) is called the set of R-rational points (or R-points) of SA .
A B-morphism φ : SA → SC is given by φ(P ) = P ◦ φ for an underlying
B-algebra homomorphism φ : C → A satisfying the following commutative
diagram:
P
A −−−−→ R
7 7
⏐
φ⏐
⏐P ◦φ
⏐
C C.
In other words, HomALG/B (C, A) → HomAF F (SA , SC ) (by φ → φ). By defi-
nition, we have the following properties of the functor SA :
f g
(F1) If R −→ R − → R are B-algebra homomorphisms, then we have maps
f∗ : SA (R) → SA (R ) and g∗ : SA (R ) → SA (R ) given by f∗ (P ) = f ◦P
and g∗ (Q) = g ◦ Q, and we have (g ◦ f )∗ = g∗ ◦ f∗ .
(F2) If R = R and g as above is the identity map idR : R → R , we
have idR ,∗ ◦f∗ = f∗ . If R = R and f as above is the identity map
idR : R → R, we have g∗ ◦ idR,∗ = g∗ .
(F3) For the identity map idR : R → R, idR,∗ : SA (R) → SA (R) is the
identity map of the set SA (R).
In short, R → SA (R) is a covariant functor of ALG/B into SET S, hence an
object in CB . For two affine schemes S and T over B, a morphism φ : S → T
is a family of maps φR : S(R) → T (R) indexed by B-algebras R such that the
following diagram commutes for any B-algebra homomorphism α : R → R :
φR
S(R) −−−−→ T (R)
⏐ ⏐
⏐
α∗ .
⏐α
. ∗
S(R ) −−−−→ T (R ).
φR
We often write HomB (S, T ) = HomCB (S, T ) for the set of all morphisms from
S into T . If S = SA and T = SC , we have HomB (S, T ) = HomALG/B (C, A)
by Yoneda’s lemma (see (4.1.5), Exercise 2.2, and Lemma 4.17).
By definition, we also have the following properties of affine schemes:
152 4 Review of Scheme Theory
φ ψ
(cf1) If A −
→C−
→ D are B-algebra homomorphisms, then we have morphisms
ψ φ
of schemes SD − → SC − → SA such that φ ◦ ψ is associated with ψ ◦ φ.
(cf2) If A = C and φ in (cf1) is the identity map idA of A, we have idA ◦ψ = ψ.
If C = D and ψ in (cf1) is the identity map idC of C, we have φ◦idC = φ.
(cf3) For the identity map idA : A → A, idA : SA (R) → SA (R) is the identity
map for all B-algebras R.
Thus, the functor A → SA is a contravariant functor from ALG/B into
CB . One of the most basic facts in functorial algebraic geometry is the full-
faithfulness of this functor (e.g., [GME] Sect. 1.4.3):
HomALG (A , A) ∼
/B = HomB (SA , SA ) via α ↔ α. (4.1.5)
A main point of the proof of this fact is to construct from a given morphism
φ ∈ HomB (SA , SA ) a B-algebra homomorphism ϕ : A → A such that ϕ = φ.
Exercise 4.3 Let ϕ = φA (idA ) ∈ SA (A) = HomALG/B (A , A), where idA ∈
SA (A) = HomALG/B (A, A) is the identity map. Then prove that ϕ = φ.
Here are some examples of affine schemes:
Example 4.4 Take f (X, Y, Z) = X p + Y p − Z p for a prime p, and let B = Z.
Then consider A = Z[X, Y, Z]/(f (X, Y, Z)). For each algebra R, we claim
SA (R) ∼
= {(x, y, z) ∈ R3 |xp + y p = z p }.
Indeed, for each solution P = (x, y, z) of Fermat’s equation in R, we define
an algebra homomorphism φ : B[X, Y, Z] → R by sending polynomials
Φ(X, Y, Z) to its value Φ(x, y, z) =: φ(Φ) ∈ R. Since Φ ∈ (f (X, Y, Z)) ⇔
Φ = Ψ f , we find that φ(Ψ f ) = Ψ (x, y, z)f (x, y, z) = 0; so φ factors through
the quotient A, getting φ ∈ SA (R). In this way, we get an injection from the
right-hand side to SA (R). If we start from φ : A → R in SA (R), we find
SA (Z) ∼
= {(a, 0, a), (0, b, b), (c, −c, 0)|a, b, c ∈ Z} if p is a prime ≥ 3.
HomALG/A (A /m2A , A/m2A ) = {ϕ2 } and EndALG/A (A /m2A ) = {idA /m2 }
A
Obviously, for the ideal generated (I) by I, U(I) = UI , and we may assume
that I is an ideal. For I = (0A ), U(0) (R) = ∅ for any B-algebra R, and so ∅
is an open subscheme. Similarly, UA (R) = SA (R) for all R, so SA itself
is an
open subscheme. If {Ii }i∈I is a family of ideals, we have Ui∈I Ii ⊃ i∈I UIi
and Ui∈I Ii (k) = i∈I UIi (k) if k is a field.
ι : HomB (SA , X) ∼
= X(A)
in other words, writing Resi : HomCB (Y, X) → HomCB (Ui , X) for the
restriction to Ui and Resi,j : HomCB (Ui , X) → HomCB (Ui ∩ Uj , X) for
the restriction to Ui ∩ Uj , if Resi,j (αi ) = Resj,i (αj ) for all (i, j) with
4.1 Functorial Algebraic Geometry 159
(αi ) ∈ i∈I HomCB (Ui , X), then αi = Resi (α) for all i for a unique α ∈
HomCB (Y, X). Thus, any morphism from Y to X is determined by the local
data αi . This is the least requirement for X being a geometric object, as a
morphism from a geometric object must be determined by local data.
A B-scheme is a local functor in CB that has an open covering by affine
B-schemes. For a B-scheme S, an S-scheme is a pair consisting of a morphism
ιX : X → S and a scheme X. We write SCH/S for the category of S-schemes.
A morphism of S-schemes φ : X → Y by definition satisfies ιY ◦ φ = ιX . If
X/S and Y/S are S-schemes, the fibered product functor X ×S Y is clearly an
S-scheme (so SCH/B has a fibered product; see [GME] Sect. 1.5.2).
A morphism X → Y of a B-scheme is separated if the diagonal subfunctor
ΔX/Y (R) = {(x, x) ∈ X(R) ×Y (R) X(R)|x ∈ X(R)} is a closed subscheme
of X ×Y X. We define a morphism of scheme f : X → S to be proper if the
following two conditions are satisfied by f :
1. f is separated and of finite type.
2. For any S-scheme T , the projection fT : XT = X ×S T → T is a closed
map (of the associated topological spaces).
f
A morphism X − → Y of a B-scheme is quasi-compact if, for any open
subscheme U ⊂ Y , the pullback functor f −1 (U ) is quasi-compact [i.e., any
open covering of f −1 (U ) has a finite subcovering].
As expected, we have (cf. [RAG] 1.8)
Lemma 4.19 An affine B-scheme X = SR is a B-scheme.
Here is a sketch of a proof.
Proof. We have X(A) = HomCB (SA , SR ) = HomALG/B (R, A) by Lemma 4.17
and (2.2). Thus, for an open affine covering {SAi }i∈I (for fraction rings Ai )
of SA , we need to prove exactness for X = SR of
HomCB (SA , X) → HomCB (SAi , X) ⇒ HomCB (SAi ∩ SAj , X),
i∈I i,j∈I
Since O(SA ) has a base of open subsets of the form SAf , we may assume
Ai = Afi for {fi }i∈I . Then Ai ⊗A Aj = Afi fj by a⊗ b → ab, and the exactness
of the above sequence follows from standard commutative ring theory (cf.
[ALG] II.2.2 or [GME] Sect. 1.2.1).
160 4 Review of Scheme Theory
and {ιAi : SAi → SB }i is in the kernel of the second double arrows, and so
we have a unique structure morphism ιX : X → SB .
Let X be a B-scheme. A presheaf F on X is a contravariant functor from
O(X) to AB. A morphism of presheaves F → G is a morphism of contravari-
ant group functors. A sheaf F is a presheaf that satisfies the following exact
sequence (of abelian groups) for any open covering {Ui }i∈I of any object U
in O(X):
Resi i,j (Resi −Resj )
0 → F (U ) −−−
i
−−→ F (Ui ) −−−−−−−−−−−→ F (Ui ∩ Uj ),
i∈I i,j∈I
where Resi indicates the restriction map F (U ) → F (Ui ), and Resi − Resj is
taken in F (Ui ∩ Uj ) after restricting to Ui ∩ Uj . The category of sheaves S(X)
is a full subcategory of the category P S(X) of presheaves. For a presheaf F
and an open subscheme U , often we write H 0 (U, F ) for F (U ) following the
tradition of cohomology theory. For any open subscheme U ⊂ X of a scheme
and a sheaf F on X, F induces a sheaf on U as any open subscheme of U is
open in X. We write F |U for the induced sheaf.
The stalk at P ∈ X(k) (for a field k) of a sheaf F is an abelian group given
by FP = limU H 0 (U, F ), where U runs over affine open subschemes such that
−→
P ∈ U (k) (such an open subscheme is called an open neighborhood of P ).
Since the stalk at P is determined by open affine neighborhoods of P , any
property of affine schemes determined by stalks can be extended to general
schemes; so, for example, flatness, smoothness, étaleness, and so on are well
defined for the general scheme morphism f : X → Y .
We call P a maximal point if it has an affine open neighborhood SA with
P ∈ SA (k) = HomALG/B (A, k) surjecting down to k. A point P ∈ X(k)
is called a geometric point if k is an algebraically closed field. A geometric
point may not be maximal as P : A → k may not be surjective for any open
neighborhood SA of P . For a geometric point P ∈ X(k), we sometimes write
k(P ) for the field of fraction of the image of P in k. For any integral domain
R, P ∈ X(R) induces a geometric point P ∈ X(k) for an algebraic closure k
of the field of fraction of R as we have a natural map X(R) → X(k).
Since a sheaf has values in abelian groups, it has a natural structure of
an additive category. We call a sequence of sheaves F → G → H exact if
FP → GP → HP is exact for stalks at all maximal points P . Then the
category of sheaves on X becomes an abelian category (see [GME] Sect. 1.4.4
4.1 Functorial Algebraic Geometry 161
S red = SpecS (O/ n) called the reduced part of S. Clearly, the topological spaces
of S and S red are identical. A scheme S is integral if H 0 (U, OS ) is an integral
domain for all open subschemes U ⊂ S. The reduced part S red is a union
of maximal integral closed subschemes. Each maximal integral closed sub-
scheme of S is called an irreducible component of S. If S = Spec(A), for each
minimal prime ideal p of A, Spec(A/p) gives an irreducible component; thus,
minimal prime ideals are in bijection with irreducible components. Similarly,
irreducible components of a general scheme S are in bijection to subsheaves of
minimal prime ideals of OS . We call S an irreducible scheme if S red is integral.
See [ALG] Chap. 1 for the topological definition of irreducibility.
An OX -module F on a scheme X is called invertible or a line bundle if
there exists an open covering {Ui }i∈I of X such that F |Ui ∼ = OUi for each i.
An OX -module F is called quasi-coherent if we have a sheaf exact sequence
OXI
→ OX J
→ F → 0, where OX I
is the product of copies of OX indexed by
any set I. If we can take I and J as finite, the OX -module F is called coherent.
Remark 4.20 If X = SA is affine, for a quasi-coherent OX -module F ,
putting M = F (X), we have F (SAf ) = Mf = Af ⊗A M ; thus, F is
determined by M and sometimes written as M6. For any given A-module, we
6 0 6
have a unique sheaf M such that H (SAf , M ) = Mf (cf. [GME] Sect. 1.2.3).
Indeed, if we write Ga ⊗A M for the A-module functor R → R ⊗A M in CA ,
we have M6(U ) = HomCA (U, Ga ⊗A M ).
is equivalent to having
i
K ←−−
X
−− A
7 7
⏐
∪⏐
⏐∪
⏐
V ←−−−− A.
iY
ΩR/A ⊗R (A ⊗A R) ∼
= I/I 2 ⊗R R ∼
= I /I ∼
2
= ΩR /A
canonically. Thus, we get the following:
Corollary 4.29 Let R ← A → A be morphisms of algebras. Then we have
ΩR/A ⊗R (A ⊗A R) ∼
= ΩA ⊗A R/A .
166 4 Review of Scheme Theory
By the first proof of Proposition 4.27, any element of ΩR/A can be written
as i fi dri for some ri and fi in R. By Corollary 4.29, taking A = S −1 A for
a multiplicative subset S in A, we get
descent data for getting A → E (see [CRT] Sect. 28 or [EGA] IV.22, and
[GME] Sect. 1.11 for descent).
{φi ∈ Hom(SR ×S Si , Si )}
Thus, S is a local functor covered with “finitely” many affine open sub-
schemes (this finiteness is “quasi-compactness,” as all affine schemes are quasi-
compact; see [GME] Sect. 1.2.2).
A B-morphism φ : S → T of schemes is a morphism in CB ; hence, the
category of B-schemes SCH/B is a full subcategory of CB . The category
SCH/B contains the category of affine B-schemes as a full subcategory (by
Lemma 4.17).
Let φ : S → T be a morphism of B-schemes.
Covering T by affine schemes
{SAi }i for B-algebras Ai , that is, T = i SAi , then φ−1 (SAi ) can be covered
again by suitable affine schemes SAij for Ai -algebras Aij . If SAij is smooth
(resp., étale) over SAi for all i and j, we call S is smooth (resp., étale) over T .
(OV,x /mx )[[X1 , . . . , Xn ]] (hence, also if and only if OV,x is regular for all max-
imal points x). A normal variety is a variety V whose stalk OV,x is integrally
closed in its total quotient ring at any closed point (hence at any point) of
V . In particular, a smooth variety is normal (as power series rings over a field
are normal).
A variety having dimension 1 is called a curve. Since any normal noetherian
domain of dimension 1 is a discrete valuation ring (cf. [CRT] Theorem 11.2),
a curve is smooth over a perfect field k if and only if it is normal (cf. Corol-
lary 2.11). A maximal point of a smooth curve C over a perfect field k gives
rise to a discrete valuation of the function field k(C) as seen in Sect. 2.1.5.
Conversely, each discrete valuation ring associated with a valuation trivial
over k gives rise to a unique maximal point of C (see [PAF] Sect. 2.4). In
short, a smooth curve C over a perfect field is determined by its function field
k(C). This type of exact relation between discrete valuations and points of a
variety is only valid for curves.
Pn (A) ∼
= {(x0 , . . . , xn )|xj ∈ A× for at least one j}/A× .
Let R be a general noetherian graded B-algebra. Suppose we have a
finite set of homogeneous generators x0 , . . . , xn of degree 1. Define a surjective
B-algebra homomorphism B[T0 , . . . , Tn ] sending Tj to xj . Then the algebra
homomorphism induces a surjection B[T0 , . . . , Tn ][Ti−1 ]0 → R[x−1 i ]0 , which
−1
in turn gives rise to a closed
subscheme V i = Spec(R[x i ] 0 ) ⊂ D i . Then we
define Proj(R) = i Vi ⊂ i Di = Pn .
We can generalize this definition to any graded B-algebra R generated by
finitely many homogeneous elements (not necessarily) of degree 1; see [GME]
Sect. 1.3), because Proj(R) = Proj(R(n) ) for R(n) = d Rnd . If R× contains
a homogeneous element of nonzero degree, we have Proj(R) ∼ = Spec(R0 ) by
definition (compare with [GME] Lemma 1.3.1); however, if R has no negative-
degree elements, Proj(R) is not affine.
If we can realize a B-scheme X as a closed subscheme of Pn (i.e., if we
have a closed immersion X → Pn/B ), X is called B-projective. More generally,
if we can realize a morphism f : X → Y so that it factors through a closed
immersion X → PnY = Pn/B ×SB Y , f is called a projective morphism. This
is equivalent to having an invertible
) sheaf L with generators x0 , . . . , xn on
X whose graded algebra R = n≥0 H 0 (X, L⊗n ) has Proj(R) isomorphic to
X. Here L⊗n = L ⊗OX L ⊗OX · · · ⊗OX L (n times). The module R is a ring
under the tensor product; that is, for x ∈ H 0 (X, L⊗n ) and y ∈ H 0 (X, L⊗m ),
the product x · y in R is given by x ⊗ y ∈ H 0 (X, L⊗m+n ). Since Lx ∼ = OX,x ,
the tensor product commutes: a ⊗ b = b ⊗ a; hence, R is a commutative ring.
Then x0 , . . . , xn are regarded as homogeneous generators of R. The morphism
X → Proj(R) is the canonical one described in [GME] Sect. 1.5.5.
If (X, f : X → Y ) is isomorphic in SCH/Y to an open subscheme (U, g :
open
U → Y ) of a closed subscheme (C, h : C → Y ) of Pn/Y (i.e., U −−−→ C as
⊂
Y -schemes), f is called quasi-projective.
Exercise 4.33 Prove that any projective morphism X → Y of B-schemes is
a proper morphism (hint: use the valuative criterion: Proposition 4.25).
Exercise 4.34 Suppose that the base ring B is an algebraically closed field
k. Prove that for any k-morphism f from Pn/k into an affine scheme S, fK :
Pn (K) → S(K) has one point image for any field K/k. Using this fact, show
further that Pn is not affine if n ≥ 1.
Similar to the relative spectrum SpecS (A) for a quasi-coherent OS -algebra,
) construct relative “Proj” for a quasi-coherent graded OS -algebras
we can
A = n≥0 An . We assume that A is finitely generated over A0 and A0 is
a coherent OS -algebras (so An is also coherent). Then for any affine open
Spec(R) ⊂ S, A(Spec(R)) is a graded algebra of finite type over R, and we
have Proj(A(Spec(R)). Clearly, if we cover S by an affine open subscheme
Spec(Ri ), {Proj(A(Spec(Ri )))}i glue naturally over S, giving rise to an
S-scheme P rojS (A) → S (which is locally projective).
4.1 Functorial Algebraic Geometry 171
is exact (a usual sheaf condition). Thus, we can form the category of τ -sheaves
S(Sτ ) on Sτ .
Recall that we have a sheafication functor P → P # from the category
of presheaves into S(Sτ ) that is universal with respect to morphisms of
presheaves P into abelian τ -sheaves F ; more precisely, we have a canoni-
cal morphism ι : P → P # , and if we have a morphism a : P → F for a
sheaf F , there is a unique morphism a# : P # → F such that a# ◦ ι = a
(see [ECH] Theorem II.2.11). Hence, in the category of abelian τ -sheaves, we
have a quotient sheaf. So S(Sτ ) is an abelian category (see [GME] Sect. 1.4.4
for abelian category); that is, for a monomorphism i : H → G of abelian
τ -sheaves, first making the naive presheaf (G/H)P : U → G/H(U ), its sheafi-
cation G/H := ((G/H)P )# gives the sheaf quotient. This is why the category
of abelian τ -sheaves is abelian. See Sect. 4.2.3 for monomorphisms.
This category S(Sτ ) has enough injectives (that is, any sheaf G can be
embedded into an injective sheaf I on Sτ ; see [ECH] Proposition III.1.1).
Taking an injective resolution G → I • , for any left exact functor F from
S(Sτ ) into an abelian category C, we can define the right-derived (cohomology)
i
→I i+1 )
functor Ri F : S(Sτ ) → C given by Ri F (G) = Ker(IIm(I i−1 →I i ) for i > 0 [and we
put R0 F (G) = F (G)]. The object Ri F (G) is independent of the choice of I •
up to canonical isomorphisms. If F : G → G(S) ∈ AB, we write H i (Sτ , G) =
Hτi (S, G) for Ri F (G). By a standard argument, an exact sequence 0 → H →
G → K → 0 of sheaves over Sτ produces a long exact sequence with connection
maps δq (see [GME] Sect. 1.10.1):
δ
0 → R0 F (H) → R0 F (G) → R0 F (K) −→
0
R1 F (H) → R1 F (G)
δq
→ R1 F (K) → · · · → Rq F (K) −→ Rq+1 F (H) → · · · . (4.1.9)
for π1 (S, s)], and from this, it is easy to see Tl E is the dual of H 1 (E, Zl )
q
(cf. (7.1.8)). If G is a coherent sheaf, HZar (S, G) := Rq F (G) is more classical
coherent cohomology (Serre studied) with respect to Zariski topology. This
does not give a Weil cohomology, as its cohomological dimension is known to
be dim S (see [ALG] III.7), while the cohomology Weil predicted must have
cohomological dimension 2 dim S.
We recall some properties of excellent rings from [EGA] IV.7 without much
proof. All rings in this section are commutative with multiplicative identity.
Though the definition itself is not very practical, we recall the definition
of excellent rings for completeness. A ring is called catenary if, for any two
prime ideals p ⊂ p , we can find a sequence p = p0 p1 · · · pn−1 pn =
p of prime ideals without any more subdivision (i.e., saturated) and any
such sequence has constant (finite) length (that is, n). For example, a field is
obviously catenary, and any polynomial ring k[X1 , . . . , Xn ] (0 ≤ n < ∞) over
a field is catenary (cf. [CRT] Exercise 5.1 and Theorem 31.4). By definition,
if A is catenary, A/a for any ideal a of A is catenary, as we can pull back
a prime ideal saturated sequence of A/a to a prime ideal saturated sequence
of A. Similarly, any localization of a catenary ring is catenary. The Krull
dimension of a ring is the maximal length of saturated prime ideal chain. Any
noetherian ring has finite dimension (cf. [CRT] Sect. 15).
A noetherian ring A is called universally catenary if every finitely gener-
ated A-algebra is catenary (so k[X1 , . . . , Xn ] is universally catenary). Since
any residue ring of a catenary ring is catenary (and any finitely generated
ring over A is a residue ring of A[X1 , . . . , Xn ]), A is universally catenary if
A[X1 , . . . , Xn ] is catenary for all n > 0. Any localization of universally cate-
nary ring is universally catenary.
A noetherian ring is called regular if, for the localization Am at any max-
imal ideal m of A, dim Am = dimA/m m/m2 , where dim Am is the Krull
dimension of Am . If A is regular, any localization is regular; in particular,
Ap is regular for any prime ideal p ([CRT] Theorem 19.3). A regular ring A
over a field k is called a geometrically regular ring over k if, for any finite
extension field k /k, A ⊗k k is regular. A regular “local” ring is an integral
174 4 Review of Scheme Theory
domain ([CRT] Theorem 14.3). Since Fp (x)⊗Fp (xp ) Fp (x) for the rational func-
tion field Fp (x) has nilpotent element x ⊗ 1 − 1 ⊗ x, Fp (x) is not geometrically
regular over Fp (xp ), though it is geometrically regular over itself.
For a prime ideal p of a ring A, we write k(p) for the quotient field of
the integral domain A/p. A morphism of rings A → A is called a regular
morphism if it is A-flat and A ⊗A k(p) is geometrically regular over k(p) for
any prime ideal p of A.
A ring A is called excellent if the following three conditions are satisfied:
(e1) A is universally catenary.
(e2) For any prime ideal p of A, Ap → A p is regular, where A
p is the p-adic
completion of Ap .
(e3) For any integral domain A of finite type over A, there exists 0 = a ∈ A
such that A [ a1 ] is regular.
Remark 4.35 Clearly, the field K is regular and excellent. Since properties
(e1–3) are stable under localization, any localization of excellent ring is excel-
lent. It is known that a Dedekind domain with characteristic 0 quotient field
is excellent. Any algebra of finite type over an excellent ring is excellent.
See [EGA] IV.7.8 for (the location of) a proof of nontrivial facts in the above
remark.
A ring is called semilocal if it has finitely many maximal ideals. If A is
semilocal, A = lim A/J(R)n denotes the J(R)-adic completion of A [here
←−n
J(R) is the Jacobson radical, which is the intersection of all maximal ideals
of
5 A]. We have A = i Ami , writing its maximal ideals m1 , . . . , mn as J(R) =
i mi (cf. [CRT] Theorem 8.15). We need the following fact later (whose proof
can be found in [EGA] IV.7.6.4).
Theorem 4.36 (Zariski–Nagata) Let A be a semilocal noetherian ring.
Then the following three conditions are equivalent:
= A ⊗A A
1. For any reduced finite A-algebra A , A is reduced.
9 = A/p ⊗A A
2. For any prime ideal p of A, A/p is reduced and is a separable
algebra over k(p).
9 ⊗A k(p) is a separable algebra over k(p).
3. For any prime ideal p of A, A/p
For a field K, a K-algebra A is separable if A ⊗K L is reduced for any finite
extension L/K.
Corollary 4.37 If A is an excellent reduced noetherian local ring with max-
= lim A/mn is reduced.
imal ideal m, then A
←−n
9 ⊗A k(p) for all prime ideals p of A is geometrically
By condition (e2), A/p
regular over k(p); in particular, it is reduced and separable over k(p). Thus,
by the theorem, A is reduced.
4.2 Fundamental Groups 175
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 4.38 (E. Galois) Suppose that L/K is a finite Galois extension.
For an intermediate field L/M/K, L/M is a Galois extension, and we have
(1) There is one-to-one onto correspondence
Proof. If M/K is a finite extension, for any finite Galois extension L with
L ⊃ L ⊃ M , we can extend σ ∈ Gal(M/K) to τ ∈ Gal(L /K) [assertion
(3) of Theorem 4.38]. Since L is a union of finite Galois extensions of L ,
σ ∈ Gal(L /K) extends to τ ∈ 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 σ ∈ Gal(M/K) can be extended to
τ ∈ Gal(L/K). Thus, σ → σ|M is a surjective homomorphism of Gal(L/K)
onto Gal(M/K). Its kernel is given by Gal(L/M ), which is normal.
Basically, the same proof as above yields the following result, which appears
stronger than the lemma.
Corollary 4.40 Let M be an intermediate extension of L/K. Then σ → σ|M
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 M into L.
Proving this corollary is left to the reader as an exercise (Exercise 4.47 (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 M gal is still a
finite Galois extension of K; so we find a normal open subgroup Gal(L/M gal )
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 σ ∈ Gal(L/K), we define the system of neighborhoods of
σ to be σU = Uσ = {σN = N σ|N ∈ U}. By this, if U 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) → ab and inverse a → a−1 are continuous. The axiom is verified by the
next result:
Lemma 4.41 We have
(1) For σ = 1, there exists N ∈ U with σ ∈ N (so, G is a Hausdorff group).
(2) N, H ∈ U ⇒ N ∩ H ∈ U.
Proof.
5 By (4.2.1), L = M M for finite Galois extensions M/K. If σ ∈
M Gal(L/M ), σ is the identity over all M , and so σ = 1 in Gal(L/K).
Thus, there exists a finite Galois extension M/K inside L such that σ ∈ N =
Gal(L/M ) if σ = 1. By definition, we have N ∈ U, and so assertion (1) follows.
4.2 Fundamental Groups 177
The topology we have defined was first introduced by Krull and is therefore
called the Krull topology on Gal(L/K).
Proposition 4.42 For an infinite Galois extension L/K, Gal(L/K) is a
compact group under the Krull topology.
Proof. Since Gal(L/K) is Hausdorff, we need to show that for any given infi-
nite subset Σ in Gal(L/K), there exists an element σ ∈ Gal(L/K) such that
Σ ∩ Hσ is infinite for all H ∈ U; that is, Σ has a limit point σ ∈ Gal(L/K).
Since H = Gal(L/M ) for a finite Galois extension M/K, there are finitely
many automorphisms of M/K. Thus, there exists an infinite subset ΣH of Σ
that induces a single automorphism σM on M . For any extension σ ∈ ΣH of
σM to L, the coset Hσ is the collection of τ ∈ Gal(L/K) (by Lemma 4.39)
with τ |M = σM . We make the construction of σM compatible with an in-
creasing sequence of Galois extensions Mj of K inside L. In other words,
we find a sequence of finite Galois extensions Mj /K (j = 1, 2,. . . ) with
K ⊂ Mj ⊂ Mj+1 ⊂ L and σj = σMj ∈ Gal(Mj /K) such that L = j Mj and
σj+1 |Mj = σj for all j. We have found an element σ ∈ Gal(L/K) such that
σ|Mj = σMj for all j. This σ is a limit point of Σ.
(2) Let {Ni }i∈I for an index set I be a collection of closed subgroups of the
Galois group Gal(L/K).5Writing Ni = Gal(L/Ki ), for the composite M
of all {Ki }i∈I , we have i∈I Ni = Gal(L/M ).
(3) If {Ki }i∈I is a collection of intermediate fields of L/K, then the group
Gal(L/ ∩i∈I Ki ) is the closure of the subgroup in Gal(L/K) generated by
Gal(L/Ki ) for all i.
(4) For σ ∈ Gal(L/K), we have Gal(L/σ(M )) = σ · Gal(L/M )σ −1 .
(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 σ → σ|M is
an isomorphism of topological groups.
Proof. Let M be the set of all intermediate fields of L/K and let H be the set
of all closed subgroups of Gal(L/K). For any M ∈ M, Gal(L/M ) is a closed
subgroup of Gal(L/K) by Corollary 4.43; hence, M → Gal(L/M ) defines a
map M → H. We define a reverse map H → M by H → LH .
We first prove that Gal(L/LH ) = H. Since Gal(L/LH ) ⊂ Gal(L/K)
is the collection of all automorphisms of L fixing LH , we confirm that
H ⊂ Gal(L/LH ). Conversely, for each finite Galois extension M/LH inside
L, H acts on M nontrivially if M = LH , and the image under σ → σ|M
in Gal(M/LH ) is a subgroup H of Gal(M/LH ) and LH = M H . By Theo-
rem 4.38 (1), H = Gal(M/LH ). In other words, for each σ ∈ Gal(L/LH ),
we find hM ∈ H such that hM |M = σ|M . We consider the infinite set
Σ = {hM }M ⊂ 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 Σ has a unique accumulation point h. Then for each finite Galois
extension M/LH , σ|M = h|M . Since L = M M for finite Galois extensions
of M/LH , we find that h = σ and H = Gal(L/LH ).
We now prove LGal(L/M) = M . By Corollary 4.43, H = Gal(L/M ) ∈ H.
By definition, M ⊂ LH . Supposing that LH = M , we try to get a con-
tradiction. Pick ξ ∈ LH − M . Then M [ξ]/M is a finite extension. Thus,
M [ξ]gal ⊂ L because L/K is normal. Since ξ ∈ M , Gal(M [ξ]gal /M ) = {1}.
Pick σ ∈ Gal(M [ξ]gal /M ) with σ(ξ) = ξ. Then by Lemma 4.39, we find
τ ∈ Gal(L/M ) = H such that τ (ξ) = σ(ξ) = ξ. This is wrong since σ(ξ)
has to be ξ because ξ ∈ LH . Thus, we find that LGal(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; thus, it
is finite [see Exercise 4.47 (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 4.39 and Corollary 4.40 that Gal(L/K)/H = HomK (LH , L), and
hence LH /K is a finite extension by Corollary 4.40. This shows (1).
Assertion (2) follows from the fact that σ ∈ Gal(L/K) is the identity on
every Ki if and only if σ is the identity on the composite of all Ki [Exer-
cise 4.47 (2)].
Let Gal(L/ ∩i∈I Ki ) = H . Then, by definition, H is a closed subgroup
containing all Hi = Gal(L/Ki ), and it contains the closure H of the subgroup
4.2 Fundamental Groups 179
h (ξ) ⇔ hH = h H , the roots of f (X) are all distinct, and so LH (ξ)/LH is a
separable extension. Since h(ξ) ∈ L for all h ∈ H, all the roots of f (X) are in
L. The Galoisclosure M of LH (ξ) over LH is a Galois extension of LH inside
L. Since L = ξ∈L LH (ξ), we find that L/LH is a Galois extension. Then by
Theorem 4.44, we find that H = Gal(L/LH ). Starting from M ∈ M, we find
that Gal(L/M ) ∈ H and M = LGal(L/M) by Theorem 4.44.
Following closely the exposition in [PAF] Sect. 4.4, we give a short description
of how one can axiomatize Galois theory (the idea is due to Grothendieck,
and the main reference is [SGA] V.4). This subsection interprets the result in
Sect. 4.2.1 in terms of category theory.
We recall some terminology from category theory. Let C be a category.
f
A morphism X − → Y in this category is called an epimorphism if, for any
g
third object Z and morphisms Y − → Z, g ◦ f = h ◦ f ⇔ g = h (so f is a
h
right unit). In other words, for any other object Z in C, the pullback map
f∗
HomC (Y, Z) −→ HomC (X, Z) is injective if f is an epimorphism. Similarly, a
f
morphism X − → Y in this category is called a monomorphism if, for any third
g
object Z and morphisms Z −→ X, f ◦ g = f ◦ h ⇔ g = h (so f is a left unit). In
h
many categories, for example, SET S and AB, monomorphisms are injections,
and epimorphisms are surjections, but, for example, in the abelian category
of sheaves, say, S(X) and S(Xτ ), an epimorphism may not be a surjection of
sections of sheaves.
Recall that a covariant (resp., contravariant) functor F : C → SET S is
called representable if there is an object F in C such that we have a system
of functorial bijections φX : F (X) → HomC (F , X) [resp., φX : F (X) →
HomC (X, F )]. Here the word “functorial” means that f∗ ◦ φX = φY ◦ F (f )
in the covariant case and f ∗ ◦ φY = φX ◦ F (f ) in the contravariant case for
f
any morphism X − → Y in C, where f∗ (g) = f ◦ g for g ∈ HomC (F , X) and
f ∗ (h) = h ◦ f for h ∈ HomC (Y, F ). The object in C representing a functor is
unique up to isomorphisms (a characterization by a universal property; see
Lemma 4.17).
If a covariant functor F0 sending every object in C to a singleton {0}
is representable in C, the object representing the functor is called an initial
4.2 Fundamental Groups 181
In the above example of the two categories (of semisimple separable finite-
dimensional commutative algebras over k and of étale coverings over S),
this assertion is obvious. However, this is just a consequence of axioms
(C3) and (f2–3). An axiomatic proof is in [SGA] V.4.e.
6. The following conditions are equivalent.
ιPi
a) The natural injection HomC (Pi , Pi ) → Hom(P, Pi ) −−→ F (Pi ) is sur-
jective.
b) Aut(Pi ) acts on F (Pi ) transitively.
c) Aut(Pi ) acts on F (Pi ) simply transitively.
We say that Pi is of Galois type if one of the above equivalent properties
is satisfied by Pi .
This is again obvious for our two examples in item 4 of C, because End(Pi )
is the monoid of field endomorphisms of a finite extension over a given
base field k. In general, we apply the functor F and prove the fact in the
image of F by using the previous assertions (1), (4), and (5).
7. Every object X in C is covered by an object Pi of Galois type so that the
unique map P → X in 3.a)factors through Pi .
8. Let I gal be the subset of I made of all indices i with Pi of Galois type.
Then we have Hom(P, P ) = Aut(P ) = limi∈I gal AutC (Pi ), which is set-
←−
theoretically in bijection with F (P ) = limi F (Pi ). We set Π = Aut(P ).
←−
9. Define a functor G : C(Π) → C so that
HomC (G(Z), ?) ∼
= HomΠ (Z, Hom(P, ?))
In the first example of item 4 (that is the example of finite extensions K/k
inside Ω sep ), we write π1 (Spec(k), s) for π1 (C, F ). Then the following lemma
is clear from the above steps:
Lemma 4.49 Let k be a field with separable closure Ω sep whose algebraic
closure we denote by Ω; thus, s = Spec(Ω) → Spec(k) is a geometric point of
Spec(k). Then we have π1 (Spec(k), s) = Gal(Ω sep /k).
We now quote two more propositions from [SGA] V Propositions 5.5–6.
Proposition 4.50 Let {Qi }i∈I be a pro-object in C = C(Π) (i.e., a projective
system of objects in C). Suppose that Qi is not an initial object for any i and
that Qi is connected for all i. Define a covariant functor G : C → SET S by
G(X) = HomC (Q, X). Then Q is isomorphic to Π/H for a closed subgroup
H of Π, and the functor G is isomorphic to the functor E → E H for the
fixed-point subset E H of E under H.
The subgroup H is given by the stabilizer of one point q ∈ Q. Since Qi is
connected, the action of Π is transitive, and Q ∼
= Π/H via πq ↔ πH.
Proposition 4.51 Let (C, F, P ) be as above (a Galois category, a fundamen-
tal functor, and the fundamental pro-object with respect to F ). Let P =
limj∈J Pj be a pro-object in the procategory of C, and define F : C → SET S
←−
by F (X) = Hom(P , X) = limj∈J HomC (Pj , X). Then the following three
−→
conditions are equivalent:
1. P ∼
= P and also F ∼
= F;
2. P is fundamental and also F is fundamental;
where Γ runs over all normal subgroups of finite index in π1top (X, x); in other
words, the algebraic fundamental group π1 (X, x) is the profinite completion
of the topological fundamental group π1top (X, x). If we pick another point
x ∈ X(C) and choose a path σ from x to x , we find π1top (X, x) ∼
= π1 (X, x )
top
f∗ i∗
which is induced by the sequence of functors ET/Y −→ ET/X −→ ET/X y .
Theorem 4.54 Suppose that X and Y are connected locally Noetherian and
that f is proper, surjective, and separable with f∗ (OX ) = OY . Then the
sequence (4.2.4) is an exact sequence of profinite groups.
Here is a sketch of the proof (see [SGA] X.1.3 for more details).
g
Proof. If Y −
→ Y is a connected étale finite covering, then f ∗ Y /X is a finite
étale covering and has to be connected because f∗ (OX ) = OY . This shows
that Fx (f ∗ Y ) = Fy (Y ), and hence f ∗ induces a surjection of the universal
pro-object of ET/X onto that of ET/Y . Since the universal pro-object is in
f∗
bijection with the fundamental group, we find that π1 (X, x) −→ π1 (Y, y) is a
surjection.
Since the fiber g −1 (y) of Y over y is a finite set of closed points, i∗ ◦ f ∗ (Y )
is a disjoint union of X y indexed by points of g −1 (y). Since X y is the final
object of ET/X y , we find that Im(i∗ ) ⊂ Ker(f∗ ).
To prove the reverse inclusion, Im(i∗ ) ⊃ Ker(f∗ ), we need to show that
h
for any finite étale Galois covering X −→ X, we have an étale Galois covering
Y /Y so that X ∼ = Y ×Y X if and only if X y /X y has a section (i.e., a
disjoint union of copies of X y ). The only nontrivial point is the “if” part. By
the properness of f and the finiteness of h, A = (f ◦h)∗ (OX ) is a sheaf of OY -
algebras finite over OY (see Lemma 5.5 and also [EGA] III.3.2.1). Let Y be the
relative spectrum SpecY (A) defined in Sect. 4.1.6. Since f∗ (OX ) = OY , f is a
surjection onto the topological space of Y . By the separability of f , Yy /y is flat
for all y ∈ Y , and hence Y /Y is flat (by the fiber-by-fiber criterion of flatness;
4.3 Group Schemes 189
Example 4.58
1. Let A = B[X1 , . . . , Xn ]. Then Gna (R) := SA (R) = Rn (as already
remarked), which is an additive group. Since
1
5. Consider the ring D = B[Xij , det(X) ] for n2 variables Xij and the vari-
able matrix X = (Xij ). Then SD (R) = GLn (R) and SD is a group
scheme under matrix multiplication, which is a subscheme of SC because
GLn (R) ⊂ Mn (R) for all R. This group scheme SD is written as GL(n).
In particular, SB[t,t−1 ] = GL(1) is equal to Gm .
6. For a given B-module X free of rank n, we define XR = X ⊗B R (which
is R-free of the same rank n) and
GLX (R) = α ∈ EndR (XR )there exists α−1 ∈ EndR (XR ) .
(idG , μ ⊗ 1) ∈ ⏐ × M (A)
G(A) −→ M
⏐ (A)
¯
↓ .G(g) × (idM ⊗g) .idM ⊗g
(g, μ ⊗ 1) ∈ G(R) × M (R) −→ M (R).
We see that g(μ ⊗ 1) = (idM ⊗g) ◦ idG (μ ⊗ 1) = (idM ⊗g) ◦ Δ(μ), and Δ
actually determines the destination of μ ⊗ 1 under g. By the associativity of
the action, for the multiplication m : G×S G → G and the identity e : SB → G
[sending SB (R) = {ιR } to the identity eR of the group G(R)], we get
−1
Example 4.62 Suppose G = Gm = Spec(B[t, t ]) for a variable t. Then we
have Δ(μ) = n∈Z pn (μ)tn , where pn ∈ EndB (M ). Then we see from e(t) = 1
and m(t) = t ⊗ t that
idM = (idM ⊗e) ◦ Δ = pn and
n
pk ◦ pj t ⊗ t = (Δ ⊗ idA ) ◦ Δ = (idM ⊗m) ◦ Δ =
k j
p n tn ⊗ tn .
k,j∈Z n
and we have
:
If G = Gm and M is a G-module, then M = Mk . (4.3.2)
k∈Z
m∗ : A∗ ⊗B A∗ → A∗ by (φ, η) → φ · η = (φ ⊗B η) ◦ m.
A ⊗B A −−−−−→ A ⊗B A ⊗B A
idA ⊗m
(φ · η) · ξ = (φ ⊗ η ⊗ ξ) ◦ (m ⊗ idA ) ◦ m
= (φ ⊗ η ⊗ ξ) ◦ (idA ⊗m) ◦ m = φ · (η · ξ)
for φ, η, ξ ∈ A∗ .
2. The commutativity defining co-inverse e:
m
A −−−−→ A ⊗B A
⏐ ⏐
⏐
.
⏐e⊗id
. A
A A
implies the existence of the identity:
e · φ = (e ⊗ φ) ◦ m = φ ◦ (e ⊗ idA ) ◦ m = φ ◦ idA = φ.
φ · η = (φ ⊗ η) ◦ m = (η ⊗ φ) ◦ m = η · φ,
where idA : A → A is the identity map of the set A and sw : A⊗B A → A⊗B A
is switching the coordinates: sw(a ⊗ b) = b ⊗ a. We can check the distributive
law for A∗ as follows:
(φ + η) · ξ = ((φ + η) ⊗ ξ) ◦ m = (φ ⊗ ξ + η ⊗ ξ) ◦ m = φ · ξ + η · ξ.
A ⊗ A ⊗ A −−−−→ A ⊗ A μ∗
μ⊗idA A∗ −−−−→ A∗ ⊗ A∗
⏐ ⏐ ⏐ ⏐
⏐
idA ⊗μ.
⏐μ ⏐ ⏐μ∗ ⊗id ∗
. ⇒ μ∗ . . A
∗
Writing G = SA∗ , the above commutative diagram implies the associativity
of multiplication m∗ induced by μ∗ . Recall that ιA : B → A denotes the
structure B-algebra homomorphism. We see easily again that ι induces the
identity e∗ ∈ G∗ . and that i∗ = (i)∗ gives the inverse of G∗ . Thus, the affine
scheme G∗ is a locally free commutative group scheme of constant rank N . By
definition, (G∗ )∗ ∼
= G canonically. The group scheme G∗ is called the Cartier
dual of G. The association A → A∗ is a contravariant involutive functor from
locally free B-bialgebras into itself, and at the same time, G → G∗ is a
contravariant involutive functor from locally free group schemes over B into
itself.
X → HomGSCH/X (G ×S X, Gm ×S X)
where φ and η run over all elements in (A ⊗B R)∗ . We write Ξ for the last
set of ξ in the above formula. Here the second identity from the last follows
from the fact
(A ⊗B R)∗ ⊗R (A ⊗B R)∗ ∼
= (A ⊗B A ⊗B R)∗ .
If ξ ∈ Ξ, then
This shows
SA∗ (R) = ξ ∈ (A ⊗B R)× e(ξ) = 1 and m(ξ) = ξ ⊗ ξ .
We now compute HomGSCH/R (G/R , Gm/R ) and show that this is isomorphic
to the above set: Note that the comultiplication m0 , the co-inverse i0 , and the
co-identity e0 of Gm/R = Spec(R[t, t−1 ]) are given by m0 (t) = t ⊗ t, e0 (t) = 1,
and i0 (t) = t−1 . Then we see
Exercise 4.64 Let the notation and assumption be as in the theorem. Prove
(Z/N Z)∗ = μN and μ∗N = Z/N Z.
)
Here (Z/N Z) = Spec( n∈Z/N Z B); so, for a B-scheme X, writing π0 (X)
for the set of connected components of the topological space of X, we have
(Z/N Z)(X) = (Z/N Z)π0 (X) , which is a locally free group.
4.3 Group Schemes 199
Up to now, we have studied affine group schemes (except for elliptic curves). A
group scheme A/S is called an abelian scheme over a base B-scheme S if A → S
is smooth, proper, and geometrically connected. Here the phrase geometrically
connected means that the fiber Ak = A ×S,φ Sk over φ : Spec(k) → S for any
algebraically closed field k over B is connected (i.e., irreducible). Since A is
proper and geometrically connected, it cannot be affine unless A = S.
Lemma 4.65 If A/S is an abelian scheme, then A as a group functor has
values in the category of abelian groups.
We give a sketch of a proof (see [ABV] Sect. 4 for details).
Proof. For simplicity, we assume that S = Spec(k) for a field k. Then we con-
sider the stalk OA,e at the group identity e. Consider OA,e and Mn = OA,e /mne
a schematic representation of A on which x ∈ A(R) acts by conjugation (as
y → xyx−1 preserves the identity e, hence induces an action on the stalk).
Thus, we have a morphism of functors A(R) → EndR (Mn ⊗B R). Since
R → EndR (Mn ⊗B R) is an affine ring scheme Spec(Cn ), we get a mor-
phism of schemes A → SCn . Since A is proper and geometrically connected,
the image of this morphism is a point (as there is no proper geometrically
connected subscheme inside an affine scheme except for a maximal point).
Since e induces the identity of SCn , we find that the entire A fixes Mn for
all n, and hence fixes the stalk OA,e . Since A is smooth and geometrically
connected, A is irreducible; hence, OA (U ) → OA,e if U is an open subscheme
containing e, and the conjugation has to be trivial on A. Thus, A has values
in abelian groups. The above proof works over any local ring B and hence
over any scheme, as schemes are local ringed spaces.
Thus, G is an abelian sheaf on SCH/B under the f ppf -topology (so we call
it an abelian f ppf -sheaf). As before, we denote by S(Bf ppf ) the category
of abelian fppf sheaves over B. More generally, for a general scheme S, we
call an object in S(Sf ppf ) an abelian fppf sheaf over S. As we remarked in
Sect. 4.1.10, the category S(Sf ppf ) is an abelian category.
Since a finite flat group scheme G over B is affine represented by a B-
bialgebra A finite flat over B, A is finitely presented faithfully flat; thus, G
gives rise to an object in S(Bf ppf ). Thus, the category F F G/B of finite flat
group schemes is a full subcategory of S(Bf ppf ). If H is a finite flat subgroup
scheme of a flat group scheme (giving rise to an fppf abelian sheaf), the
quotient G/H originally in S(Bf ppf ) is actually represented by a flat group
scheme (see, for example, [GME] Proposition 1.12.2). The multiplication of
H induces a functorial representation of H on OG and, therefore, we have
a sheaf of rings OG H
made up of H-invariant sections. If G is finite flat, we
indeed have SpecB (OG H
) = G/H. Though any morphism f : G → G in
F F G/B may not have cokernel in F F G/B , if f is a closed immersion, we have
the quotient Coker(f ) in F F G/B . If f is an epimorphism in S(Bf ppf ), we
have Ker(f ) ∈ F F G/B (see [GME] Lemma 1.12.1). In short, F F G/B is an
exact category (but may not be an abelian category).
4.3 Group Schemes 201
then R →
By Lemmas 4.67 and 4.66 combined, if A/S is an abelian scheme,
A(R) = HomCB (SR , A) is an fppf abelian sheaf, and A[p∞ ] = n A[pn ] for
: A → A) is a Barsotti–Tate p-divisible group. In the same
A[pn ] = Ker(p n
For a given locally free group scheme G/S over a connected scheme S, intu-
itively, we expect that the connected component G◦ would be a locally free
group scheme with étale quotient G/G◦ . This is not generally true, but we
can prove the following fact:
202 4 Review of Scheme Theory
Proposition 4.68 For a locally free group scheme G/S of finite rank over
S = Spec(A) for a Henselian local ring A, we have a canonical exact sequence
of locally free group schemes:
0 → G◦ → G → Get → 0,
For the moment, assume that S = Spec(k) for a field k. Then Aet /k is the
maximal separable extension of k inside A. Thus, Aet ⊗k Aet is the maximal
separable extension of k inside A ⊗k A (cf. [TCF] II.2.5.3), and
Taking the unique closed point x ∈ S and looking into stalks at x, we thus
have
0 → Gmult → G → Get → 0
such that Get is the maximal étale quotient and Gmult is the maximal multi-
plicative subgroup.
Lemma 4.71 Suppose that S is a reduced Z(p) -scheme and either that p = 0
all over S or that μp∞ (k(η)) is a finite group of bounded order for all generic
points η of S. If the exact sequence Gmult → G Get of ordinary Barsotti–
Tate groups splits over S, the splitting is unique.
204 4 Review of Scheme Theory
HomZ/pn Z,Sfppf (E[pn ], T [pn ]) (resp., Ext1Z/pn Z,Sfppf (E[pn ], T [pn ]))
Proof. The uniqueness follows from the previous lemma. To prove the exis-
tence, first suppose that E[pn ] is constant over R. Since X → E[pn ] is the
epimorphism as abelian fppf R-sheaves, we can find an fppf extension R /R
such that the set-theoretic group homomorphism X(R ) → E[pn ](R ) is sur-
jective. We may assume that Spec(R ) is connected, replacing Spec(R ) by its
connected component. Choosing a basis x1 , . . . , xr of E[pn ](R ) over Z/pn Z,
we can lift xj to a point xj in X(R ). Then we have pn xj = 0 because X(R )
is a Z/pn Z-module. Then the group subscheme generated by x1 , . . . , xr is a
constant scheme over R , splitting X E[pn ] over R . By extending scalars
to a finite étale extension, we may always make E[pn ] constant, and thus we
can find an fppf extension of R over which a splitting of X E[pn ] takes
place.
Am −−−−→ An
7 7
⏐ ⏐
⏐ ⏐
Bm −−−−−−−→ Bn .
mod m
n
The morphism set HomALG /B (A, A ) is made up of continuous B-algebra
homomorphisms. Since any Bn -algebra is mB -adically complete and any
(∞)
Bn -algebra homomorphism is mB -adically continuous, ALG/B is a full sub-
/B . Then we put
category of ALG
time, we may identify CB with COF (ALG , SET S). If a is a closed ideal of
(∞)
/B
A, we define A = A/a, and the projection A A induces an inclusion of func-
tors SA → SA . Such a formal subscheme is called a closed formal subscheme,
and the morphism SA → SA is called a closed immersion. For a subset {fi }i∈I
of A, define a subfunctor U I (R) = {P ∈ SA (R)| P (fi )R = R} of SA .
i∈I
Then, by Nakayama’s lemma (Lemma 10.8 in the text, [CRT] Theorem 2.2
and Theorem 8.4, or [BCM] II.3.2),
P (fi )R = R ⇔ P (fi )R/mB R = R/mB R.
i∈I i∈I
Thus, UI only depends on {fi mod mB A}i∈I , and for this reason, the topolog-
ical space associated with Spf B (A) is given by the space of SpecB1 (A/mB A).
Example 4.73 Formal additive group. Let A = B[X] = B[X1 , . . . , Xn ] for
lim B/mn [X].
X = (X1 , . . . , Xn ) (the n-variable polynomial ring), and set A=
←−n α
B
Each element Φ(X) ∈ A is a power series Φ(X) = α aα X (with
αj
coefficients in B) for α = (α1 , . . . , αn ) with X α = j X j such that Φ
mod mnB ∈ B/mnB [X] for all n > 0; so, writing |α| = j αj , we have
Almost all definitions of formal schemes are verbatim translations of the cor-
responding ones of schemes, replacing CB by CB . An open formal subfunctor
(resp., closed formal subfunctor) U ⊂ X for X ∈ CB is defined to be a func-
tor such that f −1 (U ) is an open (resp., closed) affine formal scheme for any
morphism f : SA → X in CB . Since morphisms of ALG/B are just B-algebra
(∞)
(∞)
homomorphisms [i.e., ALG/B is a full subcategory of ALG/B ], we can still
apply Lemma 4.18 to the functor X ∈ CB and obtain
Lemma 4.76 If U, U ⊂ X are open subfunctors of X ∈ CB , then U = U if
and only if U (k) = U (k) for all fields k in ALGB/mB .
(∞)
Note here any field in ALG/B actually is an object of ALGB/mB ; so, in the
above statement, we could have replaced ALGB/mB by the (possibly) bigger
(∞)
category ALG/B . This also supports the idea of associating the topological
space Spec(A/mB A) with SA .
Again, an open (formal) covering {Ui }i∈I of a functor X in CB is made
of open subfunctors Ui in CB such that X(k) = i∈I Ui (k) for all fields k
in ALGB/m . A functor X ∈ CB is called local if it satisfies the following
B
208 4 Review of Scheme Theory
set-theoretic exact sequence for any open covering {Yi }i∈I of any object Y in
CB :
HomCB (Y, X) → HomCB (Yi , X) ⇒ HomCB (Yi ∩ Yj , X).
i∈I i,j∈I
A formal B-scheme is a local formal functor in CB that has an open covering
made up of affine formal schemes. The structure sheaf OX is sending each
open formal B-subscheme U ⊂ X to HomCB (U, G a/B ). For each maximal
point, we have completed stalk OX,x ∈ ALG
/B given by lim OX,x /mnx for
←−n
OX,x = limUx OX (U ) for U running over an open neighborhood under the
−→
order given by inclusion.
Example 4.77 Formal completion along the special fiber. Let X in CB be a
B-scheme with structure morphism ιX : X → SB . Take an integer n > 0, and
consider Xn ∈ CB/mnB given by Xn (R) = X(R) for any B/mnB -algebra R (i.e.,
Xn is the restriction of X to the full subcategory CB/mB n of CB ). Note that,
for B/mn -algebra R,
(X ×SB SB/mn )(R) = {(x, y) ∈ X(R) × SB/mn (R)|ιX (x) = ιB/mn (y)}
= Xn (R/mn R) = Xn (R).
x to
residue field k, without losing generality, we can restrict the functor X
the category CL/B of local B-algebras with residue field k complete under the
adic topology of the maximal ideal. The morphism in CL/B is a B-algebra
homomorphism whose pullback of the maximal ideal is maximal in the source
(i.e., local B-algebra homomorphisms).
Example 4.79 Formal multiplicative group. We have already defined the ad-
ditive formal groups G a/B and G a . If we start with the multiplicative group
Gm = Spec(B[t, t−1 ]) for a variable t, we can make a formal completion along
the identity element 1 ∈ Gm (B/mB ) = (B/mB )× . We write G m for the result-
×
ing formal scheme. For the point 1A ∈ Gm (A) = A , its kernel as a B-algebra
homomorphism contains T := t − 1. Thus, the kernel of 1 is m1 = (T, mB ),
and OGm ,1 = B[t, t−1 ]m1 (the localization at 1). This shows that
OGm ,1 = lim B[t, t−1 ]m1 /mn1 = lim B[T ]/(T n) ∼
= B[[T ]].
← −
n
← −
n
Assume that B is a local ring with residue field k. Recall the category CL/B of
local complete B-algebras with residue field k (introduced in Definition 4.78).
A covariant functor D ∈ COF (CL/B , SET S) is called a deformation functor
if D(k) is a singleton (i.e., |D(k)| = 1). A deformation functor is representable
if D ∼
= SA over CL/B for an object A in CL/B . Let CL/B be the full sub-
(∞)
∼ ∼
0 (R) = {(G/R , ι)|G/R ∈ LIE/R , ι : G ×R k = G0/k }/ = .
LIE
DG
By Lemma 4.81, μpn is rigid (i.e., we have only constant deformations over
Artinian B-algebras); so μp∞ is also rigid, and by the equivalence between
LIE/B and CBT/B , the rigidity of μp∞ (Lemma 4.81) tells us that
LIE
Corollary 4.83 The deformation functor DG
is represented by B.
m
Lemma 4.84 Let G be a commutative formal Lie group over a complete local
ring B with residue field inside F = Fp . Write the group operation additively
as (x, y) → x + y. Then NG : G → G given by x → N x induces multiplication
by N on the tangent space TG/B of the identity 0.
Proof. Assume first that S = Spec(k) for a field k; so G = Spf B (R) for
R := B[[X]] with X = (X1 , . . . , Xr ). Let T = TG/k := Hom((X)/(X)2 , k) =
HomR (ΩG/k (0), k) [the tangent space at the identity 0; so ΩG/k (0) =
ΩG/k ⊗R R/(X)]. Then the addition “+”: G×G → G induces a k-linear binary
operation T ⊕T → T , which we write for the moment as a“+”b. This operation
is associative, commutative (by the associativity/commutativity of the group
structure on G), and k-linear. We claim that on T , “+” and the natural addi-
tion on T coincide up to k-linear automorphism of T . Since G[p] = Ker(pG )
and G share the same tangent space (as G[p] is a closed connected subgroup
of G finite flat over k), we may replace G by G[p]. If r = 1, then T ∼ = k,
and there is a unique k-vector space structure on k up to isomorphism; so the
claim holds. If r > 1, by extending scalars to k = F, we may assume to have an
exact sequence 0 → G1 → G → G2 → 0 with 0 < dim Gi < r as the category
of height-1 commutative group schemes over an algebraically closed field k of
4.4 Completing a Scheme 213
/ r 0
−1 p u pr u
τ : (t, t ) → (tt−1
0 , t t 0 ), τ (Z) contains Ξ0 = (t0 , t 0 ) u ∈ Zp .
Since Z is finite flat over Spf(Λ) with dominant projection to the right
d−1
5 Gm , by flatness, J is a free module over Λ (with basis 1, t , . . . , t
factor ),
and P ∈Ξ0 (P ∩ Λ)J = 0 [regarding P ∈ Ξ0 as a prime ideal P ∈ Spec(J)]. So
2 . Then τ (Z) must
τ (Z) is the Zariski closure of the infinite subgroup Ξ0 of G m
be a formal subgroup of Gm of codimension 1, and Z is a coset (t0 , t0 )τ (Z) in
2
m ×G
G m . Since Z contains the identity 1 of G
m ×G m as Φ(1, 1) = 0, we must
have τ (Z) = Z. Since the left projection π : Z → Spf(Λ) = G m is finite flat
of degree d, the kernel Ker(π) is a finite flat group scheme over W , which is
214 4 Review of Scheme Theory
therefore a finite flat subgroup of G 2 [pn ] = μpn × μpn for sufficiently large n.
m
Since the Cartier dual of μpn × μpn is the constant group (Z/pn Z) × (Z/pn Z)
(see Exercise 4.64), Ker(π) is isomorphic to μpm × μpm (for some m ≤ n and
m ≤ n). Thus, Z is an extension of G m by μpm × μ m .
p
By the geometric irreducibility of Z = Spec(J), multiplication by p induces
a dominant morphism on Z → Z, and hence Z[p∞ ] is p-divisible. Since Ker(π)
is finite flat, Z[p∞ ] is a Barsotti–Tate group fitting into an exact sequence
0 → Ker(π) → Z[p∞ ] → μp∞ → 0; so we may assume that m = 0. By
Cartier duality (see Sect. 4.3.5) applied to this sequence, the dual of Z[p∞ ] is
étale, and hence Z[p∞ ] must be isomorphic to μp∞ . Therefore, Z ∼ = lim μ n
−→n p
is a formal multiplicative group, and Z is a formal subtorus of Gm . Writing
2
the proof.
Remark 4.86 In the above proof, the two conditions Φ(1, 1) = 0 and geo-
metric irreducibility are both indispensable. If we do not suppose Φ(1, 1) = 0,
for a one-dimensional formal subtorus G ⊂ G m × G m , a coset (ζ, ζ )G for
(1, 1) = (ζ, ζ ) ∈ μpm (W ) × μpm (W ) is stable under U = 1 + pm Zp but is not a
formal subgroup. As a special instance of this example, we have the case when
pn
Φ(t, t ) is an irreducible factor Ψ (t ) of t − 1 [and in this case, Z = G m × ζ
for a root ζ of Ψ (t ) = 0 in W ; so, actually, Ψ (t, t ) = t − ζ by geometric
irreducibility]. Similarly, if we do not suppose the geometric irreducibility of
Spec(J), Z ⊂ G m ×G m could be of the form (μpm × μpr ) · G (with m + r > 0),
which is not a formal subtorus but a geometrically reducible formal subgroup.
In the above example of Φ(t, t ) = Ψ (t ), the geometric irreducibility of Spec(J)
forces Φ(t ) = t − 1 (and hence d = 1 and b = 0 under the notation of the
above lemma).
Lemma 4.87 Let A be an excellent reduced local ring with maximal ideal m.
n
!
Suppose that pA = 0 or A ⊂ A ⊗Z Q (characteristic 0). Then A ⊗A · · · ⊗A A
is reduced.
is regular. Indeed, for any prime ideal p ⊂ A and its residue field k(p),
n n
! !
(A ⊗A k ∼
⊗A · · · ⊗A A) = (A ⊗ A k
) ⊗ k(p) · · · ⊗ k(p) (A ⊗ A k
)
is regular for any finite extension k /k(p) if this is true for n = 1. In particular,
A ⊗A · · · ⊗A A
is reduced.
splits into Y ×S S = (Y ×S S)
mult × Y et over the formal completion S of S at
/S
s0 . Then there exists an open neighborhood U of s0 in S such that we have a
decomposition Y = G1 × G2 over U of Barsotti–Tate O-modules that induces
a splitting of 0 → Yu/
→ Y ×U Uu → Yu/U → 0 for all closed points u ∈ U .
mult et
U u u
Let U be the image of the middle arrow in V × V and let Z be the Zariski
closure of U . The properness of V tells us that V × V → V is a closed map
(by the definition in Sect. 4.1.6); hence, the image of Z in V is closed, and it
contains U , and so Z maps onto V . Since Z is a curve and V is normal, the
morphism Z → V is an isomorphisms (as a smooth curve is determined by its
function field, as remarked at the end of Sect. 4.1.8). Therefore, f is defined
everywhere.
Now we deal with the general case. Take a representative of f : U → V .
The schematic complement V − U is a closed subscheme X. Since codim X is
at least 1, we may assume that X has codimension 1 (otherwise, the theorem
follows). Then X contains an irreducible component V1 , which gives a prime
divisor of V . Since V has a dense affine open subvariety, SA with SA ∩ V1
is still an irreducible prime divisor in SA , that is, SA ∩ V1 is defined by a
prime ideal P ⊂ A. Since V is normal, A is a normal domain, and hence
AP is a discrete valuation ring. Since SA ∩ U is a dense open subvariety of
V , f induces the morphism of the spectrum of the field of fraction of A into
V . By a valuative criterion of properness (Proposition 4.25), we then have
f
→ V well defined. This implies that at the generic point of V1 ,
Spec(AP ) −
f is defined, a contradiction. Thus, f extends to an open subvariety U with
codim (V − U ) ≥ 2.
If we have rational maps f : V V and g : V V with f ◦ g equivalent to
idV and g ◦ f equivalent to idV , we call f and g birational, and V is birational
to V .
Proposition 5.2 Let f : V → V be a morphism of irreducible varieties
making V birational to V . Let V nfl be the Zariski closure in V of all maximal
points x ∈ V such that there exists a maximal point y ∈ V with OV ,y not
flat over OV,x . If V is proper, V nfl has at least codimension 2.
The closed scheme V nfl is called the nonflat locus in V .
5.1 Variety over a Field 219
Spec(k(X)) −−−−→ X
⏐ ⏐
⏐ ⏐
. .
Spec(V ) −−−−→ Y.
Proof. Since φ is proper and dominant, φ is surjective [so φ−1 (y0 ) is nonempty].
Also, φ is flat over an open dense subscheme O ⊂ Y . We may assume that
O is maximal under this property, since a union of open subschemes is still
open. Thus, we may assume that Y nfl = Y − O. Take the Stein factorization
X → Y st → Y . Replacing Y by Y st , we may assume that φ has connected
fibers. Since at y0 , it has one point fiber, we have [k(X) : k(Y )] < ∞. Thus,
over O, fibers of φ have constant dimension 0. Since φ : f −1 (O) → O is proper
and quasi-finite, it is finite flat by the above result. In other words, we get
Y nfi ⊂ Y − O = Y nfl , as desired.
5.1.4 Provariety
Thus, as a functor in CS := COF (ALG/OS , SET S), limi Si is well defined and
←−
given by SA∞ . Clearly, the functor SA∞ is local and coincides with the functor
associated with the scheme S∞ = SpecS (A∞ ). Thus, the projective limit S∞
exists in the category of schemes. Write πi : S∞ → Si for the projection to
Si . Define the projective limit topology on S∞ to be the weakest topology
such that all πi is continuous. If Uj ⊂ Sj is open, πij (Uj ) ⊂ Si is open for
i ≺ j, as flat πij is an open map. Thus, U ⊂ S∞ is open under the projective
limit topology if U = πj−1 (Uj ) for an open subscheme Uj for sufficiently large
j. Since πij : Sj → Si is finite flat, by the going-up/-down theorems ([CRT]
Sect. 9), it induces a surjective open map on their topological space.
Lemma 5.13 Let {Si = SpecS (Ai ), πij }i∈I be a projective system of affine
finite flat morphisms of S-varieties. Then S∞ = SpecS (A∞ ) for limi Ai gives
−→
a projective limit of {Si , πij }i∈I as a scheme, and the Zariski topology on S∞
is the projective limit topology of the Zariski topology on Si .
224 5 Geometry of Variety
Let {Si , πij }i≺j∈I be a projective system as above of finite flat affine mor-
phisms of S-varieties. Let Zi ⊂ Si be a closed subvariety forming a projective
system {Zi , πijZ
:= πij |Zj } of proper and dominant morphisms of S-varieties.
Z
Since πij is proper and dominant, its image is closed Zariski-dense; thus, it
induces a surjection of associated topological spaces. Let Z∞ = limi Zi , which
←−
is a subfunctor of S∞ and is a closed subscheme of S∞ by the above corollary.
Since a proper morphism is a closed map, V ⊂ Z∞ is a closed subscheme if
V = πj−1 (Vj ) for a closed subscheme Vj of Zj for a sufficiently large j.
6
Elliptic and Modular Curves over Rings
E −−−−→ E E −−−−→ E
⏐ ⏐ 7 7
⏐ ⏐ and ⏐ ⏐0
. . 0E ⏐ ⏐ E
S −−−−→ S S −−−−→ S .
The covariant functor ELL E/S → S ∈ SCH is called the fiber functor of
ELL/SCH .
We have seen in Sect. 2.2.2, by the Riemann–Roch theorem, that an el-
liptic curve E over an algebraically closed field k can be embedded into the
projective space P2/k . In this chapter, more generally, we will see that E/S can
be embedded into the projective space P2/S over S of dimension 2 as a curve
of degree 3. If S = Spec(k) for an algebraically closed field k, the intersection
of E ⊂ P2/k with a linear hyperplane has three points {P, Q, R} (counting
with multiplicity). As we have seen in Sect. 2.2.1, defining P + Q + R = 0E ,
we can put a group structure on E(k). Indeed, even E/S over general scheme
S has a unique structure of a group scheme, making 0E the identity.
Lemma 6.1 An abelian scheme f : A → S of relative dimension 1 is an
elliptic curve.
6.1 Basics of Elliptic Curves over a Scheme 227
In Sect. 6.1.3, we will see not only that an elliptic curve E/S has the natural
structure of a group scheme with identity 0E , but also that any morphism
f : E → E over S between elliptic curves sending 0E to 0E is a group ho-
momorphism. Thus, the category of elliptic curves is identical to the category
of abelian schemes of dimension 1. We prove here that an abelian variety of
dimension 1 over S is an elliptic curve.
Proof. By definition, an abelian scheme of relative dimension 1 is a proper
curve with a specific point given by the identity 0A . Thus, we need to prove
f∗ ΩA/S is locally S-free of rank 1. Since A/S is smooth proper, f∗ ΩA/S
is a locally free OS -module of finite type (by Lemma 5.5). To compute
rankOS f∗ ΩA/S , we may assume that S = Spec(k) for an algebraically closed
field [by pulling back via Spec(k) → S].
First, suppose that k has characteristic p. Since A[p∞ ] = n A[pn ] for
A[pn ] = Ker(x → pn x) is a Barsotti–Tate group, its connected component
A◦ [p∞ ] is prorepresented by the smooth, connected, formal group A = A|CL
k
(that is, A is the formal completion of A along the origin 0A ; see Lemma 4.82).
Thus, A = Spec(k[[T ]]) for a parameter T at the identity 0A , and it has a
unique invariant differential ω on A (induced by the cotangent vector at 0A ) up
to constant multiple. Since A(k) acts freely and transitively on A(k), we have
ΩA/k (x) ∼ = ΩA/k (0A ) ∼
= k at each point x ∈ A(k) by translation a → a + x.
Then the invariant differential ω extends uniquely to an invariant differential
over A, which generates H 0 (A, ΩA/k ); thus, H 0 (A, ΩA/k ) ∼
= k (see [LAG] 9.2,
[RAG] 7.7 or [ABV] Sect. 11). Therefore, f∗ ΩA/S is OS -locally free of rank 1,
and hence A/S is an elliptic curve.
Suppose now that k has characteristic 0 under the circumstance of the
lemma. Since A is a curve, by the Riemann–Roch theorem, A has an ample
line bundle; hence, A is projective defined by finitely many equations (and
its group structure is also defined by finitely many equations). Thus, the
group scheme A is actually defined over a subfield of k over Q of finite type.
Thus, we may assume that k → C, and by base change, we may assume that
k = C. Then, by the Archimedean uniformization described in Sect. 2.3.3,
A(C) ∼= C/L for a lattice L ∼
top
= π1 (A(C), 0A ) and A is defined by a Weierstrass
equation.
Here is another basic lemma:
Lemma 6.2 Any nonconstant S-morphism f : E → E of elliptic curves over
S is finite flat. In particular, Ker(f ) is a locally free group scheme of finite
rank if f is nonconstant.
For any scheme X, we define Pic(X) as the set of all isomorphism classes
of invertible sheaves on X. The association X → Pic(X) is a contravariant
functor by the pullback of invertible sheaves, and Pic(X) is actually a group
by the tensor product. In short, Pic : SCH → AB is a contravariant group
functor.
f
Let E −→ S be an elliptic curve. We define, for each S-scheme φ : T → S,
Since invertible sheaves are determined by its restriction to open covering, the
functor Pic is local [in the sense of (4.1.7)], and hence its subfunctor PicE/S
is local, giving a local group functor from SCH/S into AB.
We can extend the degree map to deg : PicE/S (T ) → Zπ0 (T ) for the set of
connected components π0 (T ). Indeed, for any algebraically closed field k and
a geometric point s : Spec(k) → T , the fiber E(s) = E ×S,φ◦s s = ET ×T s
is an elliptic curve over the field k and deg(L) = deg(L(s)) for the pullback
L(s) at s, which is well defined independently of the choice of s in a connected
component. We embed Z in Zπ0 (T ) diagonally, and we define
PicrE/S (T ) = L ∈ PicE/S (T ) deg(L) = r .
PicrE/S (T ) ∼
= E(T ) = HomS (T, E) (6.1.1)
by L([P ]) ⊗ L([0])r−1 ↔ P .
We repeat here a proof given in [AME] Sect. 2.1 and [GME] Sect. 2.2.2.
Proof. We write ET for E ×S T . Since the degree is kept by pullback, Picr is
a contravariant functor. Note that Picr ∼
= Pic0 via L → L ⊗ I(r[0]).
We want to show
230 6 Elliptic and Modular Curves over Rings
E∼
= Pic1 ∼
= Pic0 via P → I(P )−1 → I(P )−1 ⊗ I(0). (6.1.2)
Since Pic0 is a group functor with the identity OE under the multiplication
L·L = L⊗L , if (6.1.2) is true, the theorem is proven except for the uniqueness
of the group structure. Before stating the theorem we showed that
The proof of this fact is the same as in the case of elliptic curves (Lemma 6.2;
see also [GME] Sect. 4.1.3). Since a flat module of finite type over connected
affine scheme Spec(A) is locally free of constant rank (e.g., [CRT] Theo-
rem 7.12), by the connectedness of C (and covering C by connected affine
open subschemes), rank φ∗ OC is constant everywhere. We write this number
6.1 Basics of Elliptic Curves over a Scheme 231
;deg(φ)
as deg(φ). Thus, φ∗ OC is an invertible sheaf on C . This construction
gives a covariant functoriality to the originally contravariant functor PicC/S
(called the Albanese functoriality). If φ : E → E is an S-morphism of elliptic
curves, by our convention, φ takes 0E to 0E and, hence, at the side of the
;deg(φ)
Picard scheme, it is just L → φ∗ L. Thus, obviously, φ is a homomor-
phism of group functors. In particular, a morphism of elliptic curves (taking
zero to zero) is a homomorphism of group schemes (so the group structure on
E is unique and only dependent on 0E ).
Exercise 6.4 Let the notation and the assumption be as above. Give a de-
tailed proof of the commutativity of the following diagram:
φ
E −−−−→ E
⏐ ⏐
⏐
.Abel
⏐
.Abel
Pic0 (E) −−−−
−−−−−−→ Pic (E ).
0
L
→ deg(φ)
φ∗ L
0 → I(f (P )) → OE → k(f (P )) → 0.
I(f −1 (f (P )) = T−P
∗
I(Ker(f ))
<
N
I(P )−N ⊗ I(0)N ∼
= I(P + f −1 (0))−1 ⊗ I(f −1 (0)) = (I(P + Qi )−1 ⊗ I(Qi )).
i=1
Proof. Since EndS (E) is a ring under the composition product (f, g) → f ◦ g,
we hereafter simply write f g for f ◦ g. Then
6.1 Basics of Elliptic Curves over a Scheme 233
By (E3), for a dense affine open subset Spec(A) of S, we have H 0 (E, ΩE/A ) =
Aω for a 1-differential ω. For each point P ∈ E(S), TP : x → x + P gives
an automorphism of the curve E (sending 0E to P ). Thus, we can bring any
given cotangent vector at 0 isomorphically to a cotangent vector at P , and
as seen in the proof of Lemma 6.1, each cotangent vector at 0 extends to a
global section of ΩE/S . Thus, TP∗ ω = ω. Or differently, if we write λ(P ) for
the constant given by TP∗ ω = λ(P )ω, λ : E(S) → Ga (S) clearly gives rise to
a morphism of functors E → Ga defined over S; thus, it is a morphism of
schemes by Lemma 4.17. Since Ga is affine and E is proper and geometrically
connected over S, the image of λ is a single point. Thus, λ = 1 as λ(0E ) = 1.
The proof of Lemma 4.84 tells us the following fact, for which we give a
shorter proof.
Lemma 6.7 Let E be an elliptic curve over a general base S. Write the group
operation additively as (x, y) → x + y. Then NE : E → E given by x → N x
induces multiplication by N on the tangent space of the identity and on the
invariant differentials ΩE/S .
Proof. We only need to prove this for ΩE/S as the other case follows from
duality. First, suppose that E has an invariant differential ω generating ΩE/S .
N
!
Consider the diagonal embedding Δ : E → E := E ×S E ×S · · · ×S E given
N
HomS (X, Y ) ∼
= HomCS (X, Y )
φ f
via (X − → Y ) → (φ(T ) : X(T ) → Y (T )) given by φ(T )(T −→ X) = φ ◦ f . 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 the same as the proof of Lemma 4.17 and is left to the reader as
an exercise [the inverse is given by HomCT F (X, Y ) F → F (X)(1X ), where
F (X) : X(X) → Y (X) = HomS (X, Y )]. We often write X for X hereafter.
As we will see later in Corollary 6.16, deg NE = N 2 , and E[N ](k) ∼ =
(Z/N Z)2 for all geometric points Spec(k) → S as long as N is invertible in k
(see Lemma 4.70). We consider the following functor:
1 2
℘Γ1 (N ) (R) = (E, P, ω)/R (6.1.4)
from the category ALG of Z-algebras into SET S, where ω is a nowhere van-
ishing invariant differential, P is a point of order exactly N , that is, m → mP
induces a closed immersion (Z/N Z)/R → E of group schemes defined over
R, and [ ] = { }/ ∼ = is the set of all isomorphism classes of the objects
inside the brackets. Here Z/N Z as a group functor associates ) with T the
group (Z/N Z)π0 (T ) as in Sect. 4.3.5. Therefore, O(Z/N Z)/Z = Z/N Z Z [and
)
O(Z/N Z)/R = Z/N Z R], and the structure sheaf of Z/N Z is R-free of finite
rank N .
Recall another prototypical example of locally free group schemes: Start-
ing with the multiplicative group Gm [as a group functor Gm (R) = R×
and as an affine scheme Gm = Spec(Z[t, t−1 ])], we consider the kernel μN
of N [as a group functor μN (R) = {ζ ∈ R|ζ N = 1} and as a scheme
μN = Spec(Z[t]/(tN − 1)) = Spec(Z[(Z/N Z)])]. Then μN is a locally free
group scheme of rank N . If N > 1, it is nonisomorphic to Z/N Z, since for
6.1 Basics of Elliptic Curves over a Scheme 235
The two functors ℘Γ1 (N ) and ℘Γ1 (N ) are isomorphic by the duality of locally
free group schemes of finite rank (see Sect. 4.3.5): Here is a brief recall of the
theory. If G is a locally free group scheme of rank N over S, write G/T =
G ×S T for an S-scheme T . Then there exists a group scheme G∗/S such that
G∗ (T ) = HomT (G/T , Gm/T ) = HomT (G/T , μN/T ), where Gm/S = Gm × S
and μN/S = μN × S over Spec(Z). We have (G∗ )∗ ∼ = G in an obvious manner,
and (Z/N Z)∗ = μN by ζ(m) = ζ m for ζ ∈ μN (R) and m ∈ (Z/N Z)(R).
f
Let E − → S be an elliptic curve and fT : E/T → T be the base change to
an S-scheme T → S with zero section 0T : T → E/T . Recall the splitting for
the identity section 0 : S → E:
π∗ : E ∼
= Pic0E /S → Pic0E/S ∼
=E
is an S-homomorphism.
Lemma 6.9 If π is nonconstant, Ker(π) is locally free of finite rank, and we
have Ker(π ∗ ) = Ker(π)∗ for the dual Ker(π)∗ of Ker(π). In particular, E[N ]
is a Cartier self-dual.
Proof. Since π is nonconstant, it is finite flat (Lemma 6.2), and so 0 <
deg(π) ∈ Z. Since π ∗ ◦ π = deg(π) (Theorem 6.6), π ∗ is nonconstant, and
π ∗ is finite flat. Take L ∈ Ker(π ∗ ). If U = Spec(A) ⊂ E is an affine open
subset, L|U = L̃ for a locally free A-module L. Shrinking U if necessary, we
may assume that L ∼ = φ−1
U A ⊂ Q(A) for the total quotient ring Q(A) of A.
Thus, we may write L|U= φ−1
U OU for a small open subscheme U ⊂ E . Take
an open covering E = i Ui such that L|Ui = φi OUi for a generator φ−1
−1
i .
Since 0∗ L = OS , we may normalize φi so that φi ◦ 0E = φj ◦ 0E for all i = j
on 0−1 ∗ −1
E (Ui ∩ Uj ). Let ϕi = φi ◦ π; we have π L|Vi = ϕi OVi for Vi = π
−1
(Ui ).
Let P ∈ Ker(π); then
ϕi ◦ P = φi ◦ π ◦ P = φi ◦ 0E = φj ◦ 0E = ϕj ◦ P.
236 6 Elliptic and Modular Curves over Rings
This implies that the ϕi ◦ P s glue to give a global section ϕ ◦ P ∈ Γ (S, OS× ) =
Gm (S), getting a homomorphism Ker(π ∗ ) → Ker(π)∗ . Thus, we get a pairing
eπ : Ker(π) × Ker(π ∗ ) → Gm .
= E, we have ℘ ∼
Since (E/C)/φN (μN ) = E/E[N ] ∼ =℘∼
= ℘ .
Corollary 6.10 For each N ≥ 1, we have a canonical isomorphism of func-
tors: ℘Γ1 (N ) ∼
= ℘Γ1 (N ) defined over ALG/Z[ N1 ] .
Let p be a prime and E/S be an elliptic curve over a connected base S. Since
S is connected, OS cannot be decomposed into a direct sum of two nontrivial
sheaves of rings (cf. [GME] Sect. 1.2.4). Since E[pn ]/S is a locally free group
scheme of rank p2n , we have E[pn ] = SpecS (A) for a locally free sheaf of
OS -bialgebras A. By Proposition 4.68, over the henselization S̃x of S at each
closed point x, we have an exact sequence of abelian fppf sheaves:
where E[pn ]◦/S is defined in the same way as above, replacing S̃x by S. Here
E[pn ]et
/S is for the moment just the quotient abelian sheaf of E[p ]/S by
n
n ◦ n ◦
E[p ]/S . The abelian fppf sheaf E[p ] may not be flat but affine over the
base S; so we write E[pn ]◦ = SpecS (A◦ ) and E[pn ] = SpecS (A). Thus, the
nonflat locus E[pn ]◦,nf l is over a proper closed subscheme S nf l ⊂ S. If S nf l is
empty, E[pn ]◦ is finite flat over S and the quotient E[pn ]et is étale finite over
S. Suppose that E[pn ]◦ is flat over S. By the group structure, over an étale ex-
tension S /S so that P ∈ E et (S ) with P = 0E[pn ]et , regarding P ∈ E[pn ](S ),
the translation by P brings A◦ isomorphic to the maximal indecomposable
OS -algebra direct summand of A through which P : A → OS factors. Thus,
over a finite étale extension S , E[pn ]et/S is a disjoint union of schemes isomor-
phic to S . Thus, E[pn ]et/S is étale, and by descent, E[pn ]et
/S is étale over S.
The étale group scheme E[p ] is called the maximal étale quotient of E[pn ].
n et
If p is invertible over S, E[pn ] = E[pn ]et as pnE is étale. Suppose that p is not
invertible over S. Since μpn is connected, by self-duality E[pn ] × E[pn ] → μpn ,
E[pn ]◦ is nontrivial. Thus, rank E[p]et ≤ p. Writing rank E[p]et = pr(E) , the
number r(E) is called the p-rank of E. We call E supersingular at p if r(E) = 0,
and otherwise, we call E ordinary at p (or p-ordinary). By the self-duality, we
have (E[pn ]et )∗ → E[pn ]◦ . Thus, rank E[p] = p2r(E)+s(E) for s(E) given by
ps(E) = rank E[p]◦ / rank(E[p]et )∗ . Summarizing this, we have the following:
Lemma 6.11 Suppose that p is locally nilpotent over S. Then over an open
dense subscheme of S, E[pn ]◦ is finite flat. If E[pn ]◦ is flat over S, locally un-
der étale topology, the étale quotient E[p]et is isomorphic to (Z/pZ)r(E) for a
constant r(E) called the p-rank of E, and E[p]◦ has the maximal multiplicative
part (E[p]et )∗ isomorphic to μp . In particular, we have 0 ≤ r(E) ≤ 1.
r(E)
Exercise 6.12 Where did we use the assumption of the local nilpotency of p
in the proof of the above lemma?
isomorphism
(E, ω)/S ∼
= ϕ∗ (E, ω) = (E, ω) ×M1 S.
In short, the classification functor ℘Γ1 (1) is represented by the fine moduli
scheme M1 = Spec(R).
If we change ω by λω for λ ∈ H 0 (S, OS× ) = Gm (OS ), the parameter T
will be changed to λT and hence (x, y) is replaced by (λ−2 x, λ−3 y). Thus,
(E, λω)/S will be defined by
as long as 6 is invertible in S.
Since the moduli schemeM1 is a fine modulus, it is determined by the
restriction of ℘ = ℘Γ1 (1) to ALG/Z[ 16 ] .
Corollary 6.16 The multiplication NE by N on an elliptic curve E over a
scheme S is a finite flat endomorphism of E with deg(NE ) = N 2 . If S =
Spec(k) for an algebraically closed field k, E[N ](k) ∼
= (Z/N Z)2 as long as N
is prime to the characteristic of k.
Pp (X) = X 2 − (φ + φ∗ )X + q = 0
in End(Ep ) (by Theorem 6.6). Here Tr(φ) = φ + φ∗ ∈ Z. The Frobenius
element F robp relative to P in DP acts via the relative Frobenius map φ :
E → E on Tl E, and F robP satisfies Pp (X) = 0, which is independent of l.
Thus, {Tl E}l gives rise to a compatible system of Galois representations.
6.2 Moduli of Elliptic Curves 243
Proposition 6.18 Let E and E be elliptic curves defined over a field k. The
module Homk (E, E ) is torsion-free of finite rank ≤ 4 over Z.
Q(x1 , . . . , xr ) := (x1 , . . . , xr ) →
deg(f ) = f ∗ f = (x1 f1∗ + · · · + xr fr∗ )(x1 f1 + · · · + xr fr )
Let E/M1 be the universal elliptic curve. Since E[l] → M1 for a prime l is a
finite morphism (Lemma 6.2), it is affine. Write E[l] = Spec(A). Then I(0)2 is
principal generated by φ (the image of x−1 in A), and we remove 0 by adding
φ to A, we get an affine scheme E[l] − {0} = Spec(A[ φ ]).
1 1
ϕE : (E, ω)/R ∼
= ϕ∗ (E, ω) = (E, ω) ×M1 Spec(R).
E7 −
→ M
71
ϕE ⏐ ⏐ϕ
P
Spec(R) −
→ E → Spec(R) .
−
℘Γ1 (N ) (R) ∼
= ℘Γ1 (N ) (R) ∼
= HomZ[ 6N
1
]−alg (RΓ1 (N ) , R) = MΓ1 (N ) (R)
for all Z[ 6N
1
]-algebras R. The scheme MΓ1 (N ) /M1 is an étale finite covering
×
over M1/Z[ 6N 1
] of degree ϕ(N ) = |(Z/N Z) | for the Euler function ϕ.
The fact that the covering is étale finite follows from the same fact for E[N ]
since E[N ](k) ∼
= (Z/N Z)2 for all algebraically closed fields with characteristic
not dividing N . Since M1 is affine, any finite covering of M1 is affine. Fix a
6.2 Moduli of Elliptic Curves 245
base Z[ 6N
1
]-algebra B. Then the statement of the theorem is plainly valid for
the functors ℘Γ1 (N ) and ℘Γ1 (N ) restricted to ALG/B (by base change to B
from Z[ 6N
1
]).
If we find an S-scheme M for a given contravariant functor F : SCH/S →
SET S such that we have an isomorphism of functors M ∼ = F for M(S ) =
HomS (S , M), we say that F is representable (or represented) by an S-scheme
M. The scheme M is called the moduli scheme of the functor F . Then the
statement of the theorem is equivalent to the representability of the functors
℘Γ1 (N ) and ℘Γ1 (N ) by an affine Z[ 6N
1
]-scheme MΓ1 (N ) .
Even if F is not representable by a scheme, we often find a rough approx-
imation of the moduli called a coarse moduli scheme so that F (Spec(k)) ∼ =
M(k) for the spectrum of any algebraically closed field k that is an S-scheme.
Here we do not give the scheme-theoretic characterization of coarse moduli
schemes, only referring to [GME] Sect. 2.3.2, though we have some coarse
moduli schemes in this book.
where on f ∈ Rκ (Γ1 (N ); B), Gm acts by the character −κ. Thus, RΓ1 (N )/B
is a graded algebra.
Since f ∈ Rκ (Γ1 (N ); B) is a morphism of functors:
f
MΓ1 (N ) (R) = ℘Γ1 (N ) (R) −
→ Ga (R) = R,
we may regard f as a function of (E, φN , ω)/R (for any B-algebra R) with
f ((E, φN , ω)/R ) ∈ R satisfying the following:
(G0) f ((E, φN , λω)/R ) = λ−κ f ((E, φN , ω)/R ) for λ ∈ R× = Gm (R).
(G1) If (E, φN , ω)/R ∼
= (E , φN , ω )/R , then we have
f ((E, φN , ω)/R ) = f ((E , φN , ω )/R ).
(G2) If ρ : R → R is a morphism of B-algebras, then we have
f ((E, φN , ω)/R ×R R ) = ρ(f ((E, φN , ω)/R ))).
246 6 Elliptic and Modular Curves over Rings
E −−−−→ E/C,
π
1
f |T (l)(E, φN , ω) = f (E/C, πC,∗ φN , πC,∗ ω),
l
C∈Cl
which is a B-linear operator called the Hecke operator acting on Rκ (Γ1 (N ); B).
Often, for a prime factor l of N , the operator T (l) is written as U (l) be-
cause the definition of Cl excludes theimage of μl . For any positive inte-
ger n, taking a decomposition n = n l|N le(l) for n prime to N , we put
T (n) = T (n ) l|N U (l)e(l) . We write l = (T (l)2 − T (l2 )) for primes l out-
side N , and put n = lN le(l) if n = lN le(l) . We call n the diamond
operator.
6.2 Moduli of Elliptic Curves 247
Exercise 6.22 Prove the following formulas (cf. [IAT] Theorem 3.24):
m
(1) T (lm )T (ln ) = i=0 li T (lm+n−2i ) if m ≤ n and l N ;
(2) T (m)T (n) = 0<d|(m,n),(d,N )=1dT ( mn d2 ).
−1
We compute T (l)2 for l N . For each C ∈ Cl (E) and C ∈ Cl (E/C), πC (C )
2 ⊥
is cyclic of order l except for C = E[l]/C. Then we have a commutative
diagram:
π
E −−−C−→ E/C
⏐ ⏐
⏐
lE .
⏐π ⊥
. C
∼
E −−−−→ (E/C)/C ⊥ .
ιl
1
f |T (l)2 = f ((E/C)/C , πC ,∗ πC, ∗φN , πC ,∗ πC, ∗ω)
l2
C∈Cl (E) C ∈Cl (E/C)
⎛ ⎞
1 ⎝
= 2 f (E/C , πC ,∗ φN , πC ,∗ ω) + l · f (E, lE ◦ φN , l−1 · ω)⎠
l C ∈Cl2
Since SL2 (Z) acts on the upper half-complex plane H discretely by z → az+b
cz+d ,
we can make a quotient Riemann surface Γ1 (N )\H (see [IAT] Chap. 1).
6.2 Moduli of Elliptic Curves 249
Thus, ω is an ample line bundle over X? (N ) that gives rise to the projective
embedding of X? (N ) via homogeneous coordinates given by modular forms;
hence, we could have written ω k as OX? (N ) (k).
Here is a version of Gk (Γ0 (N ); B) with “Neben” character:
Definition 6.30 For each character ψ : (Z/N Z)× → B × , we define
Gk (Γ0 (N ), ψ; B) = f ∈ Gk (Γ1 (N ); B)f |d = dk−1 ψ(d)f ∀d ∈ (Z/N Z)× .
6.2 Moduli of Elliptic Curves 251
6.2.8 q-Expansion
We look into locally free subgroup of rank l in Tate(q) for a prime l. Let
us fix a primitive lth root of unity ζ ∈ μl ⊂ G m /q Z . We may assume that
−1
ζ = ip (i∞ (exp( l ))). Write x for the subgroup generated by x in Tate(q).
2πi
For a prime l, the set Cl rank-l subgroups in Definition 6.21 for Tate(q) is then
given by
{ζ u q 1/l |u ∈ Z/lZ} if l|N ,
Cl =
{ζ u q 1/l |u ∈ Z/lZ} {ζ} if l N .
and
πζ u ,∗ ωcan = ωcan and π∗ ωcan = l−1 · ωcan .
Proof. If we write G
m = Spf(Z[w, w−1 ]), the l isogeny w → wl on Gm induces
w
→w l
an exact sequence 1 → C → Tate(q) −−−−→ Tate(q l ) → 1. This shows the
second identity. Clearly,
N = φN .
and πζ u ,∗ φcan
can
By definition, we get
Then, by a direct computation via the definition of T (l), we get the following:
254 6 Elliptic and Modular Curves over Rings
Fix a base Z[ 6N 1
, ζN ]-algebra B, where ζN = exp(2πi N1 ). We consider the
following classification problem of level Γ (N ):
1 2
℘Γ (N ),ζN /B (R) = (E, ω, φN : (Z/N Z)2 ∼= E[N ])φN (1, 0), φN (0, 1) = ζN ,
for all B-algebras R. Writing f for the map ·, · : E[N ] ×M1 E[N ] → μN
as a morphism of group schemes over B, we clearly see that ℘Γ (N ),ζN /B is
represented over B by the closed affine subscheme MΓ (N ),ζN /B = f −1 (ζN )
of E[N ] ×M1 E[N ]. Writing
) MΓ (N ),ζN /B = Spec(RΓ (N ),ζN /B ) for a graded
algebra RΓ (N ),ζN /B = k Rk (Γ (N ); B), we define an affine curve
for each Z[ 6N
1
, ζN ]-algebra B, and we have a well-defined q-expansion at ∞
of each section in H 0 (X(N )ζN /B , ω k/B ). Here Gk (Γ (N ); B) is the degree-k
component of GΓ (N ),ζN /B = GΓ (N ),ζN ⊗Z[ N1 ,ζN ] B.
We can let a constant group α ∈ SL2 (Z/N Z) act on Y (N )ζN [and hence
det(α)
on X(N )ζN ] by (E, φ) → (E, φ ◦ α). Since φ ◦ α(1, 0), φ ◦ α(0, 1) = ζN ,
6.2 Moduli of Elliptic Curves 255
N ◦ α, ωcan )/Z[ N
fα (q) = f ((Tate(q), φcan 1
,ζN ]((q1/N )) ) ∈ B((q)).
N ◦ α, ωcan )/Z[ N
fα (q) = f ((Tate(q), φcan 1
,ζN ]((q1/N )) ).
If f ∈ Gk (Γ (N ); B) for a Z[ 6N 1
, ζN ]-algebra B, fα (q) = 0 ⇔ we have
f ((E, φN , ω)/R ) = 0 for all elliptic curves with level-Γ (N ) structure φN de-
fined over any B-algebra R.
In particular, for a prime p outside N , supposing W is a Z[ζN ]-algebra, this
implies that f ≡ 0 mod pW[[qN ]] (qN = q 1/N ) for f ∈ Gk (Γ (N ); W) =
H 0 (X(N )ζN /W , ω k ) if and only if the image of f in Gk (Γ (N ); F) vanishes.
) Since X(N )ζN /W = Proj(GΓ (N ) (W)) for the graded algebra GΓ (N ) (W) =
k≥0 Gk (Γ (N ); W), the function field K(X(N )ζN ) (K = F rac(W)) is made
up of fg for some f, g ∈ Gk (Γ (N ); W) and integer k ≥ 0. This implies
K(X(N )ζN ) is embedded into the Laurent series ring K((qN )); hence, fg =
n
−∞ c(n, g )qN with c(n, g ) ∈ K. Writing ordp for the standard addi-
f n f
Thus, we get
H(E, λω) = λ1−p H(E, ω).
Then H is a modular form of weight p − 1 defined over Fp :
Proof. We only need to prove the last assertion. Suppose that (E, ω)/k is an
ordinary elliptic curve over a field k of characteristic p. Replacing k by its finite
extension, we have E[p] ∼ = μp × Z/pZ. The differential ω induces a differential
×
λ dw
w (λ ∈ k ) on
μp = Spec(k[w]/(wp − 1)).
Since μp ⊂ Gm , and they share the tangent space at 1, we see from the above
argument H(E, ω) = λ1−p = 0.
Suppose that (E, ω)/k is supersingular. Let C be the connected component
of E[p]. If C = E[p], then we have locally under the étale topology a constant
group scheme (Z/pZ)r as a quotient of the locally free group scheme E[p] for
0 < r = r(E) ≤ 2. Since E[p] is isomorphic to its Cartier dual (see Lemma 6.9)
and the dual of Z/pZ is the connected group scheme μp , r has to be 1 and E has
to be ordinary, which is a contradiction against supersingularity. Thus, C =
E[p] and therefore, E[p] = Spec(R) for a local ring R of characteristic p. Since
E is one-dimensional, the local ring OE,0 at the origin 0 of E is a valuation
ring with residue field k. If C = Spec(R ) ⊂ E is a connected (locally free)
subgroup scheme of E of rank m, R = OE,0 /mm for the maximal ideal m of
OE,0 because C is connected. Thus, there is at most one connected subgroup
of E for a given rank. This shows that Ker(F 2 ) = C for the Frobenius map
F : E → E (p) , because deg(F 2 ) = p2 = rank(C). Thus, F 2 = p up to an
∗
automorphism of E, and hence Fabs η = 0. That is, H(E, ω) = 0, as desired.
Remark 6.38 The zero locus of the nonzero section H of ω p−1 is an effective
divisor; hence, for any algebraically closed field k of characteristic p for p N ,
the points in X1 (N )(k) carrying supersingular elliptic curves are finitely many
and algebraic over Fp .
Corollary 6.39 Let (E, ω) be an elliptic curve and an invariant differential
defined over a finite field Fpn of characteristic p. Let F : E → E be the
258 6 Elliptic and Modular Curves over Rings
F + V ∈ Z. Then
Proof. If a(p) ≡ 0 mod p, then Ker(F )[p] = Ker(V )[p] and hence, Ker(pn ) =
Ker(F V ) = Ker(F 2 ). Thus, Ker(p) is connected; hence, E is supersingular. If
a(p) ≡ 0 mod p, then Ker(F )[p] = Ker(V )[p], and thus E has two subgroup
schemes of rank p. Since E can have only one connected subgroup of rank p
(dim E = 1), E has to be p-ordinary. In this case, Ker(V ) is étale.
Corollary 6.40 Let the notation be as in the above corollary. The elliptic
curve E is ordinary defined over a finite field Fpn of characteristic p if and
only if the endomorphism algebra EndF (E)⊗Z Z(p) is isomorphic to the integral
closure O(p) of Z(p) in an imaginary quadratic field M in which the prime p
splits into pp. We can normalize the splitting of (p) so that E[pn ] = Ker(F )
and E[pn ] = Ker(V ). The ring O(p) is generated over Z(p) by the Frobenius
map F , and O(p) = EndF (E) ⊗Z Z(p) for an algebraic closure F of Fpn and
E := E ⊗Fpn F.
Proof. Suppose that E is ordinary. By the above corollary, F satisfies
X 2 − a(p)X + pn = 0.
for a unit α ∈ (Z/pZ)× , by Hensel’s lemma (cf. [CRT] Theorem 8.3), we have
op ∼= Zp ⊕ Zp ; thus, p splits in o, and hence, o(p) is equal to O(p) (see the
exercise below).
Since Op = Zp ⊕ Zp acts in the Barsotti–Tate group E[p∞ ], we have
E[p∞ ] = C ⊕ E for C = (1, 0)E[p∞ ] and E = (0, 1)E[p∞ ]. Since the first factor
Zp is identified with Op for a prime factor p of p in O(p) , we have C = E[p∞ ]
and E = E[p∞ ]. Since E is ordinary, E[p∞ ] is an ordinary Barsotti–Tate group
(see Sect. 4.3.9). We have the connected étale exact sequence
6.2 Moduli of Elliptic Curves 259
which is stable under the action of Dp and E[p∞ ]◦ ∼ = μp∞ /F . Thus, one of
E[p∞ ] and E[p∞ ] is the connected component; so we normalize the factor-
ization (p)= pp so that E[p∞ ] is the connected component. Then E[p∞ ] =
E[F ∞ ] = n Ker(F n ) as F is an inseparable isogeny. Thus, Ker(F ) = E[p],
and this determines the factor p of p.
Conversely, if EndF (E) ⊗Z Z(p) is the integral closure O(p) of Z(p) in an
imaginary quadratic field where p splits, we have E[p∞ ] = E[p∞ ] ⊕ E[p∞ ].
If E[p∞ ] is connected, the one-dimensional formal group E has to split into
the direct product of Ep and Ep , corresponding to the Barsotti–Tate groups
E[p∞ ] and E[p∞ ] by Lemma 4.82. This is impossible as each of E p and E p
has dimension at least 1; so one of E[p∞ ] and E[p∞ ] must be étale. Thus,
E[p∞ ] is ordinary, and hence E is ordinary. This finishes the proof.
Mm = X1 (N )/Wm = X1 (N ) ×W Wm .
nilpotent; so E and E are invertible at the same time. Thus, the scheme Sm
does not depend on the choice of the lift E as long as p is nilpotent in the base
ring. We write S∞ for the formal completion limm Sm of S0 along S1 .
←−
Since we have defined X1 (N ) by Proj(GΓ1 (N ) ), the invertible sheaf ω k
(k > 0) associated with the kth graded piece is ample. To see for which k,
ωk becomes very ample, we recall that an invertible sheaf of degree ≥ 2g + 1
over a curve of genus g is very ample by the Riemann–Roch theorem (see
[GME] Proposition 2.1.4). Computing the genus of X1 (N ) (e.g., [GME] The-
orem 3.1.2), the invertible sheaf ωk/R corresponding to Gk (Γ1 (N ); R) is very
ample if k ≥ 2 and N ≥ 4 (or k > 2). Thus, Sm is affine, and Sm = Spec(Vm,0 )
for a Wm -flat algebra Vm,0 (the degree-0 component of GΓ1 (N ) (Wm )[ E1 ]).
Let pALG/Wm be the category of p-adically complete Wm -algebras R with
p-adically continuous morphisms. Here the words “p-adically complete” mean
that R = limn R/pn R. When we write W∞ , it means W = limm Wm . We
←− ←−
define the following two functors, Eαord , E α : pALG/Wm → SET S, by
ord
1 2 1 2
E α (R) = (E, P, φN )/R
ord
and Eαord (R) = (E, μpα → E[pα ], φN )/R ,
if m is finite. Here E[pα ]et is the maximal étale quotient of E[pα ]; thus, writing
E[pα ] as a relative affine spectrum SpecSm (R) (cf. [GME] Sect. 1.5.4) of a
sheaf of OSm -algebras R, we have E[pα ]et = SpecSm (Ret ), where Ret is the
maximal subalgebra of R étale over OSm [see Sect. 4.3.8 in the text and [T1]
(1.4)]. We write Im,α/Sm = E[pα ]et − E[pα−1 ]et /Sm , which is an étale Galois
covering with the Galois group (Z/pα Z)× . Then the scheme Im,α represents
the functors E α and Eαord over Wm . The curve Im,α/Wm is called the Igusa
ord
We prove this in the following chapter in Sect. 7.2 after reinterpreting the tower
of modular curves {Y (N )}N as an example of a Shimura variety limN Y (N ).
←−
Since we have a universal elliptic curve E, we can identify the functor
T → I1,α (T ) = Eαord (T ) with
GI (R) = Ker(G(R) → G(R/IR)) and G(R) = Ker(G(R) → G(R/mR )),
where mR is the maximal ideal of R. When G(R) = HomCL/B (R, R)(= G(R))
for R = B[[T1 , . . . , Tn ]] [that is, G/B = Spf(R)/B ] and the identity element 0
is corresponding to the ideal (T1 , . . . , Tn ), we call G a formal (Lie) group. If G
is formal, then the map HomCL/B (R, R) φ → (φ(T1 ), . . . , φ(Tn )) identifies
GI (R) with the set IR × IR × · · ·× IR (n times) endowed with a formal group
law.
Suppose that G/B is a formal Lie group. For any integer m, the endomor-
phism mG sending x to mx of G induces a continuous algebra endomorphism
262 6 Elliptic and Modular Curves over Rings
Then
(1) The modules HomS(Bf ppf ) (G, H) and HomS((B/I)f ppf ) (G0 , H0 ) are
p- torsion-free, where the symbol “HomS(Xf ppf ) ” stands for the homo-
morphisms of abelian f ppf sheaves over CL/X .
(2) The natural map
is injective.
(3) For any f0 ∈ HomS((B/I)f ppf ) (G0 , H0 ), there exists a unique homomor-
phism Φ ∈ HomS(Bf ppf ) (G, H) such that Φ mod I = N ν f0 . We write as
in [K1] “N ν f ” for Φ even if f exists only in HomS(Bf ppf ) (G, H) ⊗Z Q.
(4) In order for f ∈ HomS(Bf ppf ) (G, H), it is necessary and sufficient that
“N ν f ” kills G[N ν ].
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 = 0.
We have an exact sequence: 0 → HI → H → H0 → 0; so we have another
exact sequence:
mod I
0 → Hom(G, HI ) → Hom(G, H) −−−−−→ Hom(G, H0 ) = Hom(G0 , H0 ),
φ : (A0 , D, () → (A0 , D , ( )
variety is a p-divisible abelian fppf sheaf. This proof of injectivity only uses
f0 : A0 → A0 induced by f .
We now show surjectivity using f |A[p∞ ] . Applying Theorem 6.44 (3) again
to G = A and H = A , for any given f0 ∈ HomA/B/I (A0 , A0 ), we have a lift
g:=“N ν f ” in HomA (A, A ) of N ν f0 . By Theorem 6.44 (4), we need to show
that g kills A[N ν ]. By the unicity of the lift, N ν f = g over A[p∞ ]. Since N
is a p-power, A[N ν ] ⊂ A[p∞ ] is killed by N ν f ; hence, A[N ν ] ⊂ Ker(g), as
desired. Thus, f = g/N ν , and the map is surjective.
We now show that (A0 , D, () ∈ DEF (B, B/I) comes from an abelian
scheme over B. When E0 is an elliptic curve, by Corollary 6.17, there ex-
ists an elliptic curve E/B such that E ×S S0 = E0 for S = Spec(B) and
S0 = Spec(B/I). In general, it is known that we can lift A0 to an abelian
scheme over B. This follows from the deformation theory of Grothendieck
([GIT] Sect. 6.3 and [CBT] Chap. V). This lifting is particularly easier if A0
is ordinary and B/I is F. Indeed, by a theorem of Tate’s, A0 has complex
multiplication. By the theory of abelian varieties with complex multiplica-
tion, A0 can be lifted to a unique abelian scheme over B with (the same)
complex multiplication (the canonical lift), because isomorphism classes of
such abelian varieties of the CM type correspond bijectively to lattices (up to
scalar multiplication) in a CM field. In any case, we do not need to specifically
lift A0 to a CM abelian variety (just the existence of a lift is sufficient).
Write X/B for a lift of A0 ; hence, X ×S S0 ∼
= A0 . Then we have an isomor-
phism α0 : X0 [p∞ ] → A0 [p∞ ]. Then we have a unique lifting g : X[p∞ ] → D
of N ν α0 [by Theorem 6.44 (4) applied to G = X0 [p∞ ] and H = D]. Clearly,
g is an isogeny, whose (quasi-)inverse is the lift of N ν (α0 )−1 . Thus, Ker(g) is
a finite flat group subscheme of X. The geometric quotient of X by a finite
flat group subscheme exists as remarked in Sect. 4.3.7 (see also [ABV] Sect. 12
and [GME] Sect. 1.8). The quotient has to be a proper geometrically connected
group scheme and hence is an abelian scheme over B. Thus, dividing X by
the finite flat group scheme Ker(g), we get the desired A/B ∈ A/B .
Recall that F is the fixed algebraic closure of Fp . Pick a complete local ring B
with residue field F. Recall the category CL/B of complete local B-algebras
with residue field F. Then ART/B is the full subcategory of CL/B made up
of Artinian local B-algebras with residue field F.
Let (E0 , ω0 )/F be a pair of an elliptic curve E over F and a generator
ω0 of H 0 (E0 , ΩE0 /F ) over F. Write π0 : E0 → Spec(F) for the structure
morphism with the zero section 0. Let E 0 be the formal completion of E0
along the origin 0. Suppose that (E0 , ω0 )/F is an ordinary elliptic curve. By
Lemma 4.72, we have E0 [pn ] ∼ = μpn × (Z/pn Z). This shows E 0 = μp∞ as
functors from ART/F into AB (Sect. 4.4.3); so η : E 0 ∼ m if we fix an
= G
isomorphism ηp◦ : μp∞ ∼ = E[p∞ ]◦ .
6.3 Deformation of Ordinary Elliptic Curves 265
Write T E[p∞ ]et = limn E[pn ](F) for the Tate module of the maximal étale
←−
quotient of E[p∞ ], which is a free Zp -module of rank 1 (if we ignore the Galois
action). We consider the following deformation functor E : CL/B → SET S:
1 2
EE0 (R) = (E/R , ιE ) E/R is an abelian scheme and ιE : E ×SR SF ∼ = E0 .
(6.3.1)
As before “[ ]” indicates the set “{ }/ ∼=” of isomorphism classes of the objects
inside the straight brackets, and f : (E, ιE )/S ∼= (E , ιE )/S if f : E → E is
an isomorphism of abelian schemes with the following commutative diagram:
f0
E ⊗B F −−−−→ E ⊗B F
⏐ ⏐
⏐
ιE . ιE ⏐
.
E0 E0 .
Since the category of abelian schemes over S of relative dimension 1 is equiv-
alent to the category of elliptic curves over S (Lemma 6.1), we have the
following definition equivalent to (6.3.1):
1 2
EE0 (R) = (E/R , ιE ) E/R is an elliptic curve and ιE : E ×SR SF ∼
= E0 .
(6.3.2)
Lemma 6.46 Assume that E0 is ordinary. Take (E, ιE )/B in EE0 (R) for an
Artinian local B-algebra R with residue field F. Let E be the formal com-
pletion of E along the origin of E. Then there exists a canonical Zp -linear
homomorphism ϕE/R : Tp E[p∞ ](F) → E(R).
The construction of ϕE/R is given in the following proof.
Proof. Since R is an Artinian local ring with maximal ideal mR , we find a
positive integer ν so that mν+1 = 0, and hence pν+1 mR = 0. Since G = E
R
ν
is a formal Lie group over R, by Lemma 6.43, p kills GmR = E. Since E is
ordinary, the connected component E[p∞ ]◦ is a deformation of E0 [p∞ ]◦ ∼
= μp∞
over F, and by Corollary 4.83, E[p∞ ]◦ ∼
= μp∞ over R. Thus, the Barsotti–Tate
group E[p∞ ] is ordinary, and we have the multiplicative/étale extension of
Barsotti–Tate groups over ART/B
Taking the prime-to-p part into our account, we have the following exact
sequence of abelian groups:
266 6 Elliptic and Modular Curves over Rings
0 → E(R) → E(R) → E(F) → 0.
The above argument only uses the fact that mν+1 = 0; thus,
E[pν ]et (F) → E. R
it works for any integer n ≥ ν in place of ν, getting a commutative diagram
for any n ≥ ν:
“pn+1 ”
E[pn+1 ]et (F) −−−−−
→ E(R)
⏐ ⏐
pE .
⏐ ⏐
.
n
E[pn ]et (F)
“p ”
−−−−→ E(R).
Passing to the limit, we get ϕE/R : Tp E[pn ]et (F) → E(R), sending x =
limn xn ∈ Tp E[p∞ ]et (F) to limn→∞ “pn ”(xn ). Here xn ∈ E[pn ](F) with
←−
pE (xn+1 ) = xn .
Theorem 6.47 (Serre–Tate) Let Tp E0 = limn E0 [pn ](F) for a given ordi-
←−
nary elliptic curve E0/F , and consider it a constant group scheme over R (for
any given Artinian local B-algebra R with residue field F). The functor EE0
is representable by the formal torus
m (R)) ∼
R → HomZp (Tp E0 ⊗Zp Tp E0 , G =G m (R),
and each deformation (E/R , iE ) ∈ EE0 (R) gives rise to the Serre–Tate coor-
dinate tE/R = tE/R,E0 : Tp E0 × Tp E0 → G m (R). The last isomorphism is
∼
induced by a chosen isomorphism η : μp∞ = E0 [p∞ ]mult and its dual inverse
◦
η et = (η ◦ )∗ : Qp /Zp ∼
= E[p∞ ](F) [so Tp E0 ∼
= Zp by η et ].
Here we simply wrote HomZp for the abelian group of Zp -linear maps; thus,
strictly speaking, it is HomMOD/Zp .
Proof. This follows from Theorem 6.45 and the above lemma. An elliptic
curve E/R ∈ EE0 (R) deforming E0/F is determined by the Serre–Tate the-
orem by the extension (6.3.3), which gives rise to an extension class in
By the Cartier self-duality of E[N ] (see Lemma 6.9),
ExtS(Rf ppf ) (E[p∞ ]et , E).
∼
= HomGSCH/R (Tp E0 ⊗Z p−n Z/Z, μpn ),
The right-hand side is independent of the choice of E/R ∈ EE0 (R). Thus, the
extension group ExtBT (E[p∞ ]et , E[p∞ ]mult ) is isomorphic to
m) .
ExtS(Rf ppf ) Tp E0 ⊗ Qp /Zp , HomZp (Tp E0 , G
φi ◦ x φ ◦ f ◦ x φi ◦ 0E
= i = = 1.
φj ◦ x φj ◦ f ◦ x φj ◦ 0E
ef : Ker(f ) × Ker(f ∗ ) → Gm
∼ Ker(f ∗ ).
inducing Ker(f )∗ =
We apply the above argument to f = pnE : E → E, write the pairing as en ,
and record the following points already shown from the construction in the
above proof:
Lemma 6.48 We have
(P1) en (α(x), y) = en (x, α∗ (y)) for α ∈ End(E/R ).
(P2) Fix the identification E0 [pn ]mult = μpn ⊂ E0 [pn ]. Then en induces an
isomorphism of group schemes: E[pn ]mult ∼ n et
/R = HomGSCH/R (E0 [p ] , μpn /R ),
n
where we regard E0 [p ] as a constant group scheme over R with fiber
E0 [pn ].
(P3) Taking the limit of the above isomorphisms with respect to n, we find
∼
E m) ∼
= Hom(Tp E[p∞ ]et , G m)
= Hom(Tp E0 , G
The value tn (xn ) becomes stationary if n ≥ n0 , and taking the limit of tn (xn )
as n → ∞, we get t(x) ∈ E(R) ∼
= HomZp (Tp E0 , G m (R)). Then we define
tE/R (x, y) = t(x)(y). Taking the dual (η0 ) : E0 [p ](F) ∼
◦ ∗ ∞
= Qp /Zp of the level
structure η0◦ : μp∞ ∼
= E0 [p∞ ]mult , we identify Tp E0 with Zp . Then we redefine
tE/R ∈ G m (R) by tE/R (1)(1) for 1 ∈ Zp .
∗⏐ ⏐ ⏐
◦f0 . .? .f0 ⊗1
m ) −−−−→ E [p∞ ] −−−−→ Tp E ⊗Zp (Qp /Zp ).
→
HomZp (Tp E0 , G 0
There are two ways of constructing ?: We may take the vertical arrow ?L :
E[p∞ ] → E [p∞ ] made by the push-out of the left square, and we may take ?R :
E[p∞ ] → E [p∞ ] made by the pullback (a fiber product) of the right square.
The morphism f |E[p∞ ] (making the above diagram commutative) exists if and
only if ?L =?R (= f |E[p∞ ] ). By the construction of the Serre–Tate coordinate,
the push-out ?L is determined by an element tE/R (·, f ∗ (·)) of
Ext(Tp E(F) ⊗
Qp Qp
Zp , Hom(Tp E (F), Gm )) = Ext(Tp E0 ⊗ Zp , Hom(Tp E0 , Gm ))
%↓ %↓
m)
Hom(Tp E(F) ⊗ Tp E (F), G = m ).
Hom(Tp E0 ⊗ Tp E0 , G
Here Ext = ExtS(Rfppf ) and Hom = HomZp . In the same way, the pullback ?R
is determined by an element tE /R (f (·), ·) of the same extension group. Thus,
having fE[p∞ ] is equivalent to tE/R (·, f ∗ (·)) = tE /R (f (·), ·).
Assertion (2) follows from the self-duality E ∼ = E ∗ = Pic0E/S (Albanese
covariant functoriality) and the canonical identity E ∼ = (E ∗ )∗ .
Let S = Gm/Z and S = G m/W . Thus, S = Spf(W [t, t−1 ]) for W[t, t−1 ] =
−1
limn W [t, t ]/(t − 1), which is isomorphic to the formal power series ring
n
←−
W [[T ]] for T = t − 1. Fix a level structure η0◦ : μp∞ ∼ = E0 [p∞ ]mult . By
Cartier self-duality, we get the identification η0 : Qp /Zp ∼
et
= E0 [p∞ ]et . We
consider the universal deformation E/S of E0/F . Since E[p ]/S is étale and
n et
S = Spf(W [t, t−1 ]), E[pn ]et = Spec(An ) for an étale algebra
An over W [t, t−1 ].
n et ∼ −n n
Since E[p ] = p Z/Z by η0et , An ⊗W [[T ]] F = Fp = p−n Z/Z F as F-algebras.
Then, by Hensel’s lemma (see [BCM] III.4.6), An ∼ = p−n Z/Z W [[T ]], so that
we have a commutative diagram:
∼
An −−−−→ p−n Z/Z W [[T ]]
⏐ ⏐
mod p.
⏐ ⏐ mod (p,T )
.
n ∼
Fp −−−et−→ p−n Z/Z F.
η0
6.3 Deformation of Ordinary Elliptic Curves 271
μpn /S → E[pn ]/S → p−n Z/Z → 0 is nonsplit, C is not defined over S but
is defined over an fppf extension of S. The following lemma, determining the
minimal fppf extension carrying C, is taken from [Br1] Sect. 7, though our
proof is slightly different from the one given there.
Proposition 6.51 The above group scheme C cyclic of order pn is étale over
OS[μpn , t1/p ] = W
n n
[t, t−1 ][μpn , t1/p ]. Once we embed W in Qp , there are
pn different choices of étale group scheme Cζ cyclic of order pn indexed by
ζ ∈ μpn (Qp ), which is finite flat over W
n
[t, t−1 ][ζ · t1/p ] but cannot descend to
a proper W [t, t−1 ]-subalgebra of W
n
[t, t−1 ][ζ · t1/p ].
In the following proof, we make explicit a canonical way of associating μpn
ζ → Cζ .
Proof. Since C is étale cyclic of order pn , its image covers E[pn ]et ; hence, we
can lift η et ( p1n ) to a generator γ of C. Note that taking a lift xm of η et ( p1m )
with xn = γ, we have t := limm→∞ “pm ”(xm ). In other words, γ gives rise to
γ = (“pn ”)−1 (limn→∞ “pm ”(xm )), which is a pn th root of t. Thus, γ = t1/p
n
at the lower exact sequence of the above diagram. We have another com-
mutative diagram, making a quotient of E[p∞ ]/E[pn ]◦ by the image of C in
E[p∞ ]/E[pn ]◦ (which is isomorphic to the original E):
μp∞ −−−−→ E[p∞ ]/E[pn ]◦ −−−−→ Qp /Zp
⏐ →
⏐ ⏐
⏐
.
⏐ ⏐x
→pn x
V n. . (6.3.9)
μp∞ −−−−→ E[p∞ ] −−−−→ Qp /Zp .
→
E0 [p∞ ], F f lifts to φ◦ ∈ EndW (E0 [p∞ ]◦ ); so we have the lift φdiv = (φ◦ ⊕φet ) ∈
EndW (E0 [p∞ ]) = EndDEF ((E0 , E0 [p∞ ], idE0 )) of F f . Then by the equivalence
of category in Theorem 6.45,
The elliptic curve E0/W as in the above corollary is called the canonical lift
of E0/F .
In the previous section, we started with an elliptic curve defined over a finite
field Fq and lifted it to W (F). By the existence of Frobenius endomorphism,
any ordinary elliptic curve over a finite field has complex multiplication by
an imaginary quadratic field M . In this section, as was traditionally done by
Deuring and more generally by Shimura–Taniyama–Weil [ACM], we start with
a CM elliptic curve over C and describe its canonical association to orders of
M . We fix a prime p and an identification Cp ∼ = C as fields compatible with
ip i∞
our embeddings Cp ← Q → C.
Fix an imaginary quadratic field M ⊂ Q in which the fixed prime p splits into
(p) = pp, where ip : Q → Cp indices
the
prime p; in other words, writing O
for the integer ring of M , p = x ∈ O|x|p < 1 .
For each lattice a ⊂ M whose p-adic completion ap is identical to Op =
O ⊗Z Zp , we consider the complex torus E(a)(C) = C/a, where we wrote a
for the lattice {i∞ (a) ∈ C|a ∈ a} in C. By a theorem in [ACM] 12.4, this
complex torus is algebraizable to an elliptic curve E(a) having CM by M over
a number field.
Let F be an algebraic closure of the finite field Fp of p-elements. We write
W for the p-adically closed discrete valuation ring inside Cp unramified over
Zp with residue field F. Thus, W is isomorphic to the ring W (F) of Witt
274 6 Elliptic and Modular Curves over Rings
O(a) = {α ∈ M |αa ⊂ a}
Let O(a) = {α ∈ O|αa ⊂ a}. Then O(a) is an order of M , that is, a subring
of the integer ring O that spans M . The order O(a) is determined by its
conductor ideal c ⊂ Z, which is the annihilator of the Z-module O/O(a). In
other words, O(a) = Oc := Z + cO.
Exercise 6.54 If c is the annihilator of O/O(a) in Z, prove that O(a) =
Z + cO and that any order is of the form Z + cO for an ideal c of Z.
Fix a prime ideal l = (l) ⊂ Z and set Oe = Z + le O (the order of conductor
le ). The following three conditions for a fractional Oe -ideal a are equivalent
(cf. [IAT] Proposition 4.11 and (5.4.2) and [CRT] Theorem 11.3):
(I1) a is Oe -projective;
(I2) a is locally principal;
(I3) a is a proper Oe -ideal [that is, Oe = O(a)].
Here a is locally principal means that for each prime ideal q of Oe , the local-
ization a(q) is a principal ideal of the local ring Oe,q . An R-module M for a
noetherian ring R is R-projective if R is isomorphic to a direct summand of an
R-free module of finite rank. An Oe -ideal is Oe -projective if it is Oe -projective
as an Oe -module. Thus, Cle := Pic(Oe ) is the group of Oe -projective frac-
tional ideals modulo principal ideals. The group Cle is finite and is called the
ring class group modulo le .
Since f∗ ΩE(a)/W = 0∗E(a) ΩE(a)/W is locally free of rank 1 for the stricture
morphism f : E(a) → Spec(W), it is free of rank 1 as W is a discrete valuation
ring. We choose and fix a differential ω = ω(O) on E(O)/W so that
As long as E(O ∩ a) is an étale covering of both E(a) and E(O), ω(O) induces
a differential ω(a) first by pullback to E(O ∩ a) and then by pullback inverse
from E(O∩a) to E(a). Writing ap for the p-adic closure of a in Mp , if ap = Op ,
π πR
this étaleness of the covering E(O) ←−L− E(O ∩ a) −−→ E(a) holds, as Ker(πL )
and Ker(πR ) have order prime to p and hence étale. So ω(a) is well defined
with H 0 (E(a), ΩE(a)/W ) = Wω(a).
Exercise 6.55 Prove the following facts:
π
→ E of elliptic curves over W is étale if deg(π) is prime
1. An isogeny E −
to p;
2. E(a ∩ O)/W → E(a)/W is étale if ap = Op .
276 6 Elliptic and Modular Curves over Rings
There is one more case where ω(a) is well defined. Take the étale subgroup Cζ
for a primitive pn th root ζ ∈ μpn (Cp ) of unity as in Proposition 6.51 of the
universal deformation E/W of E0 = E(O) ⊗W F. Note that the canonical lift
E0 of E0 is exactly E(O) ×W W by Corollary 6.52. Let C ζ = Cζ ∩ E(O) =
Cζ ×E E(O), regarding E(O) as the fiber of E at t = 1 for the canonical
coordinate t of the Serre–Tate deformation space. Then, for a suitable proper
ideal aζ of On for l = p which we describe later, we have E(aζ ) = E(O)/C ζ
and π : E(O) → E(aζ ) is an étale covering defined over W [ζ] by Proposi-
tion 6.51, where pn is the rank of the locally free group scheme C ζ over W [ζ].
Thus, the pullback inverse (π ∗ )−1 : ΩE(O)/W[ζ] → ΩE(aζ )/W[ζ] is a surjective
isomorphism, and for ω(aζ ) = (π ∗ )−1 ω(O), we thus have
This level structure induces if (a) : E(a)[f] f−1 /Z by the Cartier self-duality
of E(a)[f]. In this way, we get many quintuples:
(E(a), if (a), if(a), C(a)[l], ω(a)) ∈ M(Γ0 (f2 ) ∩ Γ1 (f) ∩ Γ0 (l))(W) (6.4.2)
Here the ordering of writing—that is, Mp /Op comes at the right side of
exp
Mp /Op —is important. In complex uniformization E(a)(C) = C/(Zz + Z) = ∼
C× /q Z for a basis (z, 1)
of a andq = exp(2πiz) by the exponential map
1
u → e(u) := exp(2πu), n pn Z + Zz/Zz is sent to the canonical level
structure μp∞ → E(a)(C). This corresponds to Mp /Op → E(a)[p∞ ], and 1
in the basis (z, 1) is put on the right end traditionally.
u
→e(u)
Identify Mp /Op with Qp /Zp and then Qp /Zp with n p−n Z/Z −−−−−→
μp∞ (C). We get η∞ ◦
(a) : μp∞ ∼= E(a)[p∞ ] canonically given. By the Cartier
dual inverse, we get η∞ (a) : Qp /Zp ∼
et ∞
= E(a)[p ]. Since W is strictly Henselian,
η∞ (a) descends to the étale level structure η et (a) = ηpet (a) over W, again
et
by the Cartier dual inverse, we get the descent η ◦ (a) = ηp◦ (a) to W of the
◦
connected level structure η∞ (a). Since E(a)/W = E(a) ×W W is the canonical
lift of E(a) ×W F, we recover a canonically chosen two-level structure
a(p) ∼
η (p) = η (p) (a) : = T (p) E(a) = ←
lim E(a)[N ]. (6.4.4)
−
N,pN
By Proposition 6.53, the elliptic curve E(a) is defined over a number field
K and extends to an elliptic curve over the integer ring OK . Write O(a) for
the order of a. Since ap = Op , we have O(a)p = Op . Thus, we may regard
p as a proper O(a)-ideal [i.e., we just write p for p ∩ O(a)]. By the proof of
Corollary 6.52, for the Frobenius automorphism φ : W → W , Eφ0 = E0 /E0 [p]◦ .
Thus, we get E(a)φ = E(p−1 a) as E(a)[p] = E(a)[p]◦ . Consider Tl (E(a)) =
limn E(a)[ln ] ∼
= al = Tl (E(p−1 a)) (for a prime l = p) as al = (p−1 a)l . Thus,
←−
the action of φ induces a Zl -linear automorphism
(p) )× .
Consider (φl )l an element in (O
×
Define ρ : (MA(p∞) ) → G(A(p∞) ) by αη (p) (a) = η (p) (a) ◦ ρ(p) (α). Since
(p)
Here, again, we regard NK/M (P) as a proper prime ideal of the order O(a) =
{α ∈ M |αa ⊂ a} [in other words, for any prime ideal L of O with O(a)L = OL ,
we consider L to be a proper prime ideal of O(a) by taking the intersection
L∩O(a)]. In particular, NK/M (P) is a principal O(a)-ideal generated by ψ(P).
Exercise 6.58 Give a detailed proof of the fact that NK/M (P) is a principal
O(a)-ideal generated by ψ(P).
7
Modular Curves as Shimura Variety
of prime-to-p level and the Igusa tower over it, as we seek applications to
integrality of L-values and p-integral modular forms defined over the tower.
a b
Γ0 (pm )p = ∈ G(Zp )c ∈ pm Zp ,
c d
/ 0
Γ1 (pm )p = a b ∈Γ
c d
0 (pm )p d − 1 ∈ pm Zp .
any algebra over smaller Z[ N1 , μN ]). This type of level structure has already
been studied in earlier sections (cf. Sect. 2.2), and over Z(p) [μpn ] (or over a
complete discrete valuation ring W ⊃ Z(p) [μpn ]), we know the representability
(sometimes as a coarse modulus) of the functor
1 2
R → EΓ (N ),Γ (R) := (E, φN , φp )/R
by a modular curve Y (N, Γ ). In Sect. 6.2.12, this functor was denoted by Eαord
if Γ = Γ1 (pα )p (to emphasize the exclusion of supersingular elliptic curves
of characteristic p). We consider the p-integral Shimura curve (of prime-to-p
level)
(p)
Sh/Z(p) = lim Y (N )/Z(p)
←−
pN
(regarding these schemes as Q-schemes). We will later see that ShΓ/Q = Sh/Γ
as a geometric quotient as in [GME] Sects. 1.8.3 and 3.3. Write μ× N ⊂ μN for
the set of all primitive N th roots of unity, and let Z(1) = limN μ×
×
←− N . For each
×
ζ = (ζN )N ∈ Z(1) , let
Y (N, Γ )ζ/B = Y (N, Γ )ζN ×Z(p) [μN ] B, Y (N )ζ/B = Y (N )ζN ×Z(p) [μN ] B.
The action of τ (z, g) on FΓ is on the left and has the following property
(shown by Shimura;
0 e.g.,
[IAT] Theorem 6.23 or [PAF] Theorem 4.14). Simply
write a := (a, a0 a−1 ) ∈ GL1 (Z(p) ) × SL2 (Z(p) ) ⊂ GI [the matrix a0 a−1
0
in
T (Z(p) ) being diagonally embedded in SL2 (A )]. Then, for f ∈ FΓ , we have
(∞)
Proof. The first assertion follows directly from (7.1.3). By the existence
(p)
of a smooth compactification X/W of Y (p) over W, Zariski’s connected-
(p)
ness theorem (Theorem 5.7) tells us that Y/F = Y (p) ×W F is irreducible.
(p)
Thus, the vertical Weil prime divisor X/F on the smooth arithmetic surface
(p) (p)
X/W gives rise to a unique valuation. To see this, writing Y/W = Spec(R)
(because it is affine), we have R = pN RN for the noetherian normal domain
RN := R0 (Γ (N ), W) of Krull dimension 2 with Y (N ) = Spec(R0 (Γ (N ), W)).
(p)
So the principal ideal (p) = pR is the prime ideal corresponding to Y/F =
Spec(R/(p)). Since Y (N ) is finite (integral) over P1 (J)−{∞} over the discrete
(p)
valuation ring W, the ring RN has Krull dimension 2. The tower Y/W is an
étale cover over Y (N )ζN /W for N ≥ 3 prime to p, as Y (N )ζN is a fine modulus
of EΓ (N ),ζN . Thus, the localization RN,(p) of RN at (p) is a normal noetherian
local domain of dimension 1, and it is a discrete valuation ring(e.g., [CRT]
Theorem 11.2). Again by the étaleness of Y (p) /Y (N ), R(p) = N RN,(p) is
(p)
a discrete valuation ring. By the irreducibility of X/F , a W-integral modu-
(p)
lar form of level away from p vanishes on the divisor X/F if and only if its
q-expansion vanishes modulo p by Lemma 6.36 and the remark following the
lemma. Thus, the valuation v0 is the one corresponding to the valuation ring
(p) (p)
R(p) and to the vertical prime divisor X/F ⊂ X/W .
1
Write G/Z for the affine algebraic group GL(2) = Spec(Z[a, b, c, d, ad−bc ])
defined over Spec(Z) and Z/Z for its center [see Example 4.58 (5)].
Write
a b the homogeneous coordinate of P1 (C) as column vectors [ xy ]. Then
g = c d ∈ G(C) acts on P1 (C) by the usual matrix multiplication:
? @
[ xy ] → g [ xy ] = ax+by
cx+dy .
Since the center Z(C) acts trivially, this action factors through P GL2 (C) =
G(C)/Z(C). Since G(R) leaves P1 (R) stable, G(R) acts on P1 (C) \ P1 (R) ∼ =
C − R, where we identify P1 (C) \ P1 (R) with C − R, sending the homogeneous
coordinate [ xy ] to the inhomogeneous one z = xy ∈ C.
Let H = {z ∈ C|i(z − z) > 0} (the upper half-complex plane); hence,
C − R = H H for the lower half-plane H (the complex conjugate of H). Thus,
∼
P (C) \ P (R) = H H. Tracking down the
1 1
G(R) acts
a on above isomorphisms,
for g = c d ∈ G(R) and z ∈ C − R, we have g(z) = az+b
b
cz+d . Put
N to its image in Z× ×
N ⊂ Z (see Exercise 7.9). By the strong approximation
theorem, SL2 (Q) is dense in SL2 (A(∞) ) = SL2 (A)/SL2 (R) (see Exercise 7.9
again), det : G(A) → A× induces G(Q)\G(A)/Γ(N )G(R)+ ∼ = Cl(N ). Indeed,
from density, we see
SL2 (Q)U (N ) = SL2 (A(∞) ) = Ker(det : G(A(∞) → Gm (A(∞) ))).
Thus, we have 4
G(A) = G(Q)tc Γ(N )G(R)+
c∈Cl(N )
a b √
a√ −1+b
for tc = ( 0c 01 ). The projection G(R)+ c d → c −1+d
∈ H induces
Since Y (N )ζNc /Q[μN ] (C) = Γ (N )\H and Y (N )/Q (C) = c Y (N )ζNc /Q[μN ] (C),
we have Y (N )/Q (C) = G(Q)\G(A)/Γ(N )Z(Q). Passing to the limit, we get
(7.1.6)
as J1 t gJ1−1 = g ι = det(g)g −1 , where h is supposed to be induced from a mor-
phism of group schemes from ResC/R Gm into GL(2)/R . In this way, we identify
H H with X. The isomorphism H H ∼ = X sending z to hz is equivariant
under the action of G(R), making the following diagram commutative:
v
→gv
V (R) −−−−→ V (R)
⏐ ⏐
⏐
ριg−1 (z) (u)=t hg−1 (z) (u).
⏐ ι
.ρz (u)=t hz (u) (7.1.7)
V (R) −−−−→ V (R).
v
→gv
7.1 Shimura Curve 289
set H 0 (π1 (S , s ), Isom(V (A(Σ∞) ), Vs (E/S ))) [of π1 (S , s )-fixed points in
(Σ)
Isom(V (A(Σ∞) ), Vs (E/S ))], where s is a unique geometric point of S such
(Σ)
that s = Fs (φi )(ϕi ). Here Isom(V (A(Σ∞) ), Vs (E/S )) is the set of A(Σ∞) -
(Σ)
(Σ)
linear isomorphisms of the V (A(Σ∞) ) onto Vs (E/S ). This is the sheaf of
level structures. Pick η (Σ) ∈ Isom(V (A(Σ∞) ), Vs (E/S )). Then γ ∈ π1 (S , s )
(Σ)
(Σ)
[resp., g ∈ G(A(Σ∞) )] acts on η (Σ) through its action on Vs (E/S )) [resp.,
the natural left action of G(A(Σ∞) ) on V (A(Σ∞) )].
Lemma 7.10 The right action of G(A(Σ∞) ) and the action of π1 (S , s ) com-
mute, and the definition of the sheaf is independent of the choice of s .
for the right action η (Σ) → η (Σ) ◦ u [u ∈ Γ (Σ) ]. Here η (Σ) runs over
Isom(V (A(Σ∞) ), Vs (E/S )), and if S is not connected, the right-hand side is
(Σ)
the disjoint union of the sets (each defined for each connected component). We
define a level-Γ structure η (Σ) over S to be a section η (Σ) ∈ (LV (Σ) /Γ )(S).
When Γ is sufficiently small (i.e., a so-called neat subgroup, for example,
7.1 Shimura Curve 291
Proof. Since E1Q (S) = limΓ EΓQ (S) = limN E Q (S) for Γ running over open
←− ←− Γ (N )
compact subgroups with the obvious projection (E, η Γ ) → (E, η Γ ) if Γ ⊂ Γ ,
we study EΓQ for (sufficiently small) open compact subgroups Γ .
Let G1 = SL(2) ⊂ G and Gad = G/Z = P GL(2). First, note that the
images of gΓ g −1 ∩ G1 (Q) and gΓ g −1 ∩ G(Q)+ in Gad (Q) are equal, where
G(R)+ = {x ∈ G(R)| det(x) > 0} and G(Q)+ = G(Q) ∩ G(R)+ . Indeed,
if γ = gΓ g −1 ∩ G(Q)+ , det(γ) ∈ Q× × × = {1}; hence,
+ ∩ det(Γ ) ⊂ Q+ ∩ Z
γ ∈ SL2 (Z). So, by Lemma 7.8, ShΓ (C) can be embedded into ShΓ1 (C) for
Γ1 = G1 (A(∞) ) ∩ Γ . This also follows from the fact that the moduli problem
with respect to Γ1 is neat without having any nontrivial automorphisms. Let
292 7 Modular Curves as Shimura Variety
Thus, for any Z-lattice M of V (E) = T (Es ) ⊗Z Q, we have a unique elliptic
curve EM/S and an isogeny f : EM → E such that π(f ) ⊗ Q : V (EM ) →
7.1 Shimura Curve 293
Here ·, ·E : T (E ) × T (E ) → Z(1)( μΓ ) is the projective limit of the
self-duality pairing E [N ] × E [N ] → μN , and ∼
= is the isomorphism (not
an isogeny). As we now see, following the argument we presented earlier,
this functor EΓ,c is represented by a scheme Y (c, Γ ) over a specific abelian
extension kΓ of Q dependent on Γ (see below for a description of kΓ for some
specific Γ ’s). See [PAF] Sect. 4.2 for this process for more general families of
abelian varieties (of higher dimension).
Recall the principal congruence subgroup of level N
Γ(N ) = Ker(GL(L)
GL(L/N L)) = Ker(GL2 (Z)
GL2 (Z/N Z))
Thus, EΓ(N ),c is the standard modulifunctor classifying the level structure for
the principal congruence subgroup Γ(N ):
φ : (Z/N Z)2 ∼
= E[N ]
EΓ(N ),c (S) = EΓ (N ),ζNc (S) = (E, φ)/S /∼
=.
and φ(1, 0), φ(0, 1) = ζN
c
which implies
4
ShΓ(N )/Q[μN ] = Y (N )ζNc over Q[μN ]. (7.1.10)
c∈Cl(N )
Since the Galois group permutes Y (N )ζ , the right-hand side of (7.1.10) de-
scends to the base Q, and we obtain the model ShΓ(N ) over Q, because Y (N )
is a quasi-projective normal curve (if N ≥ 3). Since a normal curve over a
field is smooth (cf. Corollary 2.11), Y (N ) (with N ≥ 3) is smooth over Q.
Since E1Q (S) = limN EΓQ(N ) (S), the procurve Sh/Q = limN Y (N )/Q represents
←− ←−
E1Q , and also Sh/Q is smooth. This finishes the proof of the theorem. We will
give a more direct proof of smoothness below in the proof of Theorem 7.13.
Remark 7.12 The set ℘(C) of isomorphism classes of elliptic curves over C
with differential is in one-to-one onto correspondence to the set of lattices Lat
in C as seen in Sect. 2.3.3. The set E1Q (C) is a union of G(A(∞) )-orbits each
isomorphic to
9 = {L
Lat = Zw
1 + Zw
2 |(w1 , w2 ) is a basis of (A(∞) )2 over A(∞) }.
To describe this fact in a more down-to-earth manner, pick (E, η)/C ∈ E1Q (C)
and write E(C) = C/Lw for Lw = Zw1,∞ + Zw2,∞ with a basis w =
= lim L/N L, which has a basis w = (w1 , w2 )
(w1,∞ , w2,∞ ) of C over R. Let L
←−N
induced by (w1,∞ , w2,∞ ). Since E[N ] ∼
over Z = L/N L canonically, we have
T (E) = L = lim L/N L; thus, we get ηw : VA(∞) (a, b) → aw1 + bw2 ∈
←−N
T (E). Then we have a bijection
7.1 Shimura Curve 295
9
{(E, ηw ◦ g)|g ∈ G(A(∞) )} ↔ Lat
This explains well why the action
given by ηw ◦ g → (ηw ◦ g)−1 (T (E)) = g −1 L.
9
η → η ◦ g is interpreted as the action of g −1 on Lat.
η (p) ((1, 0)), η (p) ((0, 1)) = c(η (p) )Λ0 ((1, 0), (0, 1)) = c(η (p) ).
Here ·, · : T (p) E × T (p) E → Z (p) (1) is the limit bilinear form lim ·, ·N
←−N,pN
of the self-duality pairing ·, ·N on E[N ]. Identifying Z with Z(1), taking
1 ∈ Z to lim ζN for ζN = exp( 2πi ), we consider ·, · to have values in
←−N N
(p) ⊂ A(p∞) . Then we consider the following functor from Z(p) -schemes into
Z
SET S.
? @
(p)
EΓ (S) = (E, η (p) )/S . (7.1.11)
Here [·] indicates isomorphism classes in ELL(p) (so, in the classical language,
“prime-to-p isogeny classes”).
296 7 Modular Curves as Shimura Variety
(p)
Theorem 7.13 The p-integral Shimura curve Sh/Z(p) represents the functor
(p)
EG(Zp ) over Z(p) for the closed subgroup G(Zp ) of G(A(∞) ). The procurve
(p)
Sh/Z(p) is smooth over Z(p) and satisfies the following canonical identities:
Proof. Let Z× × ×
(p)+ = Zp ∩ Q+ . As long as Γ is maximal at p, we can iden-
× ×
tify Cl(Γ ) = A(∞) /Q× + det(Γ ) with A
(p∞)
/Z×
(p)+ det(Γ
(p)
). Thus, we may
(p) ×
choose the representative set {c} in (Z ) (and hence we may assume the
self-duality of L at p). By the same process of bringing EΓQ isomorphically to
EΓ/Q [for an open, compact subgroup Γ ⊂ G(Z) maximal at p], the functor
is equivalent to EΓ/Z(p) defined over SCH/Z(p) ; thus, it is representable over
(p)
Z(p) , giving a canonical model ShΓ/Z(p) over Z(p) . The functor EΓ/Z (p)
is a
disjoint union of functors EΓ,c indexed by c ∈ Cl(Γ ), where
? @
EΓ,c (S) = (E, η (p) )/S η (p) (L)
= T (p) (Es ) and c(η (p) ) ≡ c mod det(Γ ) .
(7.1.12)
Since η (p) ∈ (LV (p) /Γ )(S), at the geometric point s ∈ S, we have a small
étale neighborhood U s such that η (p) is the image of a full prime-to-p-level
c ) := η (p) (L
structure η (p) over U , and we define η (p) (L c ) [which is independent
of the choice of η because Γ · L = L].
(p)
(p)
A subtle point is to relate Sh/Z(p) to Sh/Q . Since the above process of iden-
tifying functors EΓQ(N ) and EΓ (N ) ends up with the functor EΓ (N ) independent
(p)
are smooth over Zp and hence over Z(p) (as smoothness only depends on the
completion).
Example 7.14 Let us return to the setting of Sect. 6.4; thus, we start with
an imaginary quadratic field M in which the prime p splits into (p) = pp,
with p corresponding to ip . For each lattice a ⊂ M , we made a tuple in
(6.4.4) x(a)/W = (E(a), η (p) : a(p) ∼
= T (p) E(a)) of a CM elliptic curve E(a)
defined over W and level structures. Fixing a basis of a(p) over Z (p) and
hence identifying a ∼
(p)
= (Z ) , we get a W-point (E(a), η )/W of Sh(p) (W)
(p) 2 (p)
Slightly more generally, we take a discrete valuation ring W flat over Z(p)
W is not necessarily equal to the dis-
as a base ring B. Hence, hereafter
crete valuation ring i−1
p (W ∩ ip ( pN Q[μN ])). We consider the base change
(p) (p)
ShΓ/W = Sh/Z(p) ×Spec(Z(p) ) Spec(W). Consider the following open compact
subgroups for an integer N ≥ 1:
/ 0
Γ0 (N ) = c ∈ N Z
a b ∈ G(Z)
c d
/ 0
Γ1 (N ) =
0 (N )d − 1 ∈ N Z
a b ∈ Γ
c d
/ 0 (7.1.13)
Γ11 (N ) =
1 (N )a − 1 ∈ N Z
a b ∈ Γ
c d
/ 0
Γ(N ) = 1 (N )b ∈ N Z
a b ∈ Γ
c d 1
.
for any x, x ∈ E[N ] projecting down to 1 ∈ Z/N Z for a given exact sequence
φN : μN → E[N ] Z/N (of finite flat group schemes).
Then we have the following expression of EΓ1 (N ),c :
1
1 2
EΓ1 (N ),c (R) = (E, φN : μN → E[N ] Z/N Z)/R φN (ζN ), φN (1) = ζN
c
.
1
(7.1.14)
Exercise 7.16 Verify the expression (7.1.14) if N ≥ 4.
=
If N ≥ 4, Aut((E, φN )/R ) = {1} for (E, φN ) ∈ EΓ1 (N ),c (R) [so if η (p) (L)
1
T (p) (E), η (p) is the really the orbit η (p) ◦ Γ11 (N ) before sheafication of
LV /Γ11 (N )]. This point is essential to answer the exercise above.
By this triviality of the automorphism of level-Γ11 (N ) test objects, EΓ1 (N ),c
1
is representable by Y (Γ11 (N ))ζNc /W (if p N ≥ 4). In any case, if p N , we
have a coarse moduli scheme Y (Γ )/W for EΓ,c with respect to the above three
choices of Γ , since we have Y (Γ ) = Y (N )/Γ , where γ ∈ Γ acts on level-Γ(N )
test objects (E, φN : (Z/N Z)2 ∼ = E[N ]) by φN → φN ◦ γ for the image γ of γ
in G(Z/N Z) (see [GME] Sect. 2.6).
Let TG ⊂ G be the diagonal torus given by
TG (R) = ( a0 d0 ) ∈ G(R)a, d ∈ Gm (R) .
which gives a right action of G(A(∞) ) on the functor. This action obvi-
ously extends the action of u ∈ Γ ⊂ G(Z) (defined earlier) to G(A(∞) ). By
Lemma 4.17, this induces a right action of G(A(∞) ) on Sh/Q . If g ∈ G(A(p∞) ),
(p)
the same action as above, (E, η (p) ) → (E, η (p) ◦ g), is well defined on E/W and
(p)
therefore extends to a p-integral action on Sh/W .
For a general open compact subgroup Γ ⊂ G(A(∞) ) maximal at p [not
we define Y (Γ ) = Sh(p)
necessarily in G(Z)], Γ/Z(p) by the (geometric) quotient
Sh(p) /Γ , as Γ acts on Sh(p) through a constant finite group (see [GME] 1.8.3
for geometric quotients). Over Q, for any open compact subgroup Γ , we have
ShΓ/Q = Sh/Γ . Let Σ = ∅ or Σ = {p}.
(Σ)
Lemma 7.17 If Γ is sufficiently small with trivial Aut((E, η Γ )) for all test
(Σ) (Σ)
objects (E, η Γ ), Sh(Σ) is étale over ShΓ , and each geometric fiber of Sh(Σ)
(Σ)
over ShΓ is isomorphic to the group Γ .
We give a proof assuming 16 ∈ W [general cases can be treated using Y (lq)
for two primes l and q as we have done many times now].
Proof. Since the proof is the same for any choice of Σ, we assume Σ = ∅.
Since we have a commutative diagram
∼
Sh −−−−→ Sh
g
⏐ ⏐
⏐ ⏐
. .
∼
ShΓ −−−−→ Shg−1 Γ g
g
Γ(24) acts freely on Sh [because Aut((E, φ24 )/R ) = {1}], we find that any
geometric fiber is one-to-one onto Γ(24), as desired.
(Σ) (Σ)
ShZ(A(Σ∞) )Γ = Sh (Σ) )Γ .
Z(Z
(p) )Γ = Z(Z)Γ
If Σ = {p}, Γ is maximal at p; hence, Z(Z . Thus, assuming
Σ ⊂ {p}, we have
(Σ) (Σ)
ShZ(A(p∞) )Γ = Sh .
Z(Z)Γ
on
The action of (a, d) ∈ TG (Z) sends EΓ
to EΓ,adc
EΓ,c
, as we have observed.
Thus, z ∈ Z(Z) acts by sending the index c to z c [and hence, Y (Γ 1 (N ))ζ c ∼
2
= 1 N
Y (Γ11 (N ))ζ cz2 ]. This shows
N
4
Sh
(p) ∼
= Y (Γ11 (N ))ζNc /W and
Γ1 (N )/W
Z(Z) 1
c∈Cl(N )/(Cl(N ))2
Recall the right action of g ∈ G(A(∞) ) on Sh/Q given by (E, η) → (E, η ◦ g).
Consider G := G(A(∞ )/Z(Q) (see [IAT] Theorem 6.23). The following fact
has been shown in [IAT] Theorem 6.23 (see also [PAF] Theorem 4.14):
Theorem 7.18 The action of the group G(A(∞) ) on Sh leaves any geometri-
cally irreducible component Y ⊂ Sh stable, and we have G = G(A(∞ )/Z(Q) =
AutSCH (Y/Q ). Each geometrically irreducible component Y is geometrically
defined over the maximal cyclotomic extension Qab /Q. The right action of
g ∈ G on [x, g ] is given by
rise to x̃ ∈ Sh(p) (W) (assuming that W has residue field F) and by definition
τ (g)(x̃) = τ
(g)(x) for g ∈ G(A(p∞) ). Note that {x̃ ∈ Sh(p) (W)|x ∈ Shord (F)}
(p)
is Zariski dense in Sh/W . Thus, {g ∈ G(A(p∞) )| det(g) ∈ Z(p) }/Z(Z(p) ) →
(p) (p)
AutSCH (Sh/F ), and by smoothness, the set π0 (Sh/C ) of connected
7.1 Shimura Curve 303
(p)
components is in bijection to the corresponding mod p set π0 (Sh/F ) (Zariski’s
connectedness theorem: Theorem 5.7). Therefore, we get
(p)
{g ∈ G(A(p∞) )| det(g) ∈ Z(p) }/Z(Z(p) ) → AutSCH (Y/F )
(p)
for each irreducible component Y (p) of Sh/F . We have proved the injectivity of
Exercise 7.21 Why does the Zariski density of {x̃ ∈ Sh(p) (W)|x ∈ Shord (F)}
(p)
in Sh/W imply the injectivity of the above theorem?
sits on [z, gζ ], where “≈” indicates that there is an isogeny αζ : E(aζ ) → E(a)
such that αζ ◦ η z−u n
= ηz ◦ γu . Thus, we get the following result:
p
In [IAT] Theorem 6.31, for a modular function f ∈ F finite at [z, 1], it is proven
that f ([z, 1])[b,Mz ] = f τ (rz (b)) ([z, 1]) = f ([z, 1] · rz (b)), which is equivalent
to the above proposition as functions determine the point. Also, note that
7.1 Shimura Curve 305
Let l = (l) for a prime l > 0 be a prime ideal in Z. Then, for a level-N prime
to l with Γ ⊃ Γ(N ), if lA = A (for a Z[ 6N 1
]-algebra A), we define a morphism
of functors
1 2 1 2
[l] : ℘Γ ∩Γ0 (l) (A) = (E, φN , C, ω)/A → ℘Γ (A) = (E, φN , ω)/A
containing G1 = SL2 (A(p∞) ), and the second key point is that Aut(I/S1 )
contains GL1 (Zp ) and the two actions on I/Fp commute. There are at least
two ways of showing irreducibility via the symmetry action on the Shimura
variety: (i) the use of the automorphism group of the function field of the
Shimura variety of characteristic 0 (cf. [PAF] Sects. 6.4.3 and 8.4.4), which
uses characteristic-0 results to prove the characteristic-p assertion; and (ii)
a purely characteristic-p proof following a line close to (i) (see [H09a]). In
this book, we follow the first method. There are some other arguments purely
in characteristic p (for example, Igusa’s original argument in [I] relying on
ramification at a supersingular point) to prove the same result (covering dif-
ferent families of reductive groups giving the Shimura variety) as sketched
in [Ch4] for the Siegel modular variety. In this section, we write F = Fp (an
algebraic closure of Fp ) and take W = W (F) (the ring of Witt vectors with
coefficients in F in [BCM] IX.1, or equivalently, the completed p-adic integer
ring of the maximal unramified extension of Qp in Cp ). Hereafter we assume
W = i−1p (W ), which is a strict henselization of Z(p) (see [NMD] Sect. 2.2 for
Henselian rings).
elliptic modular Shimura curve classifies elliptic curves E/S (abelian schemes
of dimension 1; Lemma 6.1, Theorems 2.32 and 6.3) and level structure η :
A2 ∼ = V (E). In this simplest case, the fixed endomorphism algebra is Q ⊂
End(E/S ) ⊗Z Q, and the polarization is well hidden as each elliptic curve has
a canonical polarization whose polar divisor is [0E ]. Indeed, the Néron–Severi
∼
group N S(E/S ) := Pic(E/S )/ Pic0 (E/S ) −−→ Z has a canonical generator
deg
[0E ]. For a higher-dimensional abelian scheme A/S , N S(A/S ) has a higher
rank, and a single A could have different polarizations.
In the setting of the Shimura variety Sh of PEL type (of level away from
a given finite set Σ of places), assuming we have a smooth integral compact-
ification of Sh over a p-adic discrete valuation ring W (see [ACS] 6.4.1), we
can verify the axioms (A1–2) for the following reasons: Compatibility with
the action in (A1) and the transitivity in (A2) follow from the definition. In
the case of a Shimura variety of PEL type, S is usually the ordinary locus
(the open locus on which a generalized Hasse invariant is invertible) of the
modulo p Shimura variety Sh/F of level away from a given finite set Σ of
places containing p and ∞. Then for the adele ring A(Σ) away from Σ, G1 is
often the adele group G1 (A(Σ) ) for a semisimple group G1/Q (G1 is the de-
rived group of the starting reductive group in Shimura’s data), and T is the
Zp -points T (Zp ) of the reductive part T of a parabolic subgroup P of G1 [as
this notation suggests, we apply our axiomatic approach to (G1 , P, T ) in Defi-
nition 7.1 to prove irreducibility of the elliptic Igusa tower]. Here P modulo its
unipotent radical is reductive by definition, and the quotient can be realized
as a subgroup of P , called the Levi subgroup (or the reductive part; see [LAG]
Sect. 30). If we choose Σ so that G1 (Ql ) is generated by unipotent elements
for all l ∈ Σ, G1 has no nontrivial finite quotient group (as unipotent groups
over characteristic-0 field are divisible). For any finite subcovering I /S of I,
G1 acts on the finite set π0 (I ) through a finite quotient of G1 ; so the action
is trivial, getting (A1).
In the above discussion of how to verify (A1), a key ingredient is that G1 is
large enough not to have finite (nontrivial) quotient. For the general Shimura
variety, this is deduced from the existence of smooth toroidal compactification
(if the Shimura variety is not projective; see [CSM], [DAV], and [ACS]) and
a characteristic-0 determination of the automorphism group of the Shimura
variety (principally due to Shimura; cf. [Sh4] II, [IAT] Theorem 6.23, and
[PAF] Theorems 4.14 and 7.7). The compactification in our Shimura curve
case is as described in Sect. 6.2.7. Alternatively, one can prove that G1 is large
by showing that the l-adic monodromy representation for primes l = p has
a large open image in G1 (A(Σ) ) (here an l-adic monodromy representation
is given by the action of the fundamental group of the Shimura variety on
the l-adic Tate module of the fiber at the base point). Indeed, C.-L. Chai
[Ch2] (in the symplectic case) has deduced the open image result via group
theory from the semisimplicity theorem of Grothendieck–Deligne of the l-adic
7.2 Igusa Tower 309
We define a generic version of the Igusa tower and prepare necessary lemmas
to count the number of generic points of the Igusa tower (so, our goal in the
next subsection is to show that this number is equal to 1, i.e., irreducibility).
Recall the neutral component Y (p) of Sh(p) and its function field F(p) over
the fraction field K of W. Let S be the ordinary locus Y (p) [ H1 ]/F for the Hasse
invariant H. Then S is irreducible over F, because H is a global section of the
ample line bundle ω ⊗(p−1) over the compactification X/F of Y/F .
(p) (p)
We look into the Igusa tower in a local neighborhood of the generic point
(p) (p)
of the vertical irreducible divisor X/F ⊂ X/W . Consider the valuation ring V
of F(p) of the valuation v0 in Definition 7.4. Thus, the residue field V/mV is
the function field F(S) of S. Let E/Y (p) be the universal elliptic curve. Then
we consider the Cartesian diagram for EV = E ×Y (p) Spec(V):
→
EV −−−−→ E
⏐ ⏐
⏐ ⏐
. .
Spec(V) −−−−→ Y (p) .
→
(∗)
m ◦ ∼
((Z/p Z)/V
m m et
IV,m
= IsomV
(μpm /V
, EV
[p ] ) = IsomV , EV
[p ] )
(∗∗)
∼ [p ] − EV
= EV m et
[p
m−1 et
] ,
where the identity (∗) is given by taking the inverse of the Cartier dual map
and (∗∗) is given by φ → φ(1) (1 ∈ Z/pm Z) for φ ∈ IsomV ((Z/p Z), E[p ] ).
m m et
In particular, IV,m
is an étale finite covering over V,
/Spf(V) and GL1 (Zp )
naturally acts on IV,m
by its action on (Z/p Z)/V
m
.
m ◦
a closed immersion sending identity to identity, it factors through EV
[p ] .
Comparing the rank, we find that it is an isomorphism. Thus, we get
m ∼ m ◦
IV,m
= ImmV
(μpm /V
, EV
[p ]) = IsomV
(μpm /V
, EV
[p ] ).
m ◦
In any case, EV [p ] is isomorphic to μpm /V
étale locally. Since the Cartier
dual of μpm is isomorphic to Z/pm Z (Exercise 4.64), by the self-duality of
E[pm ] (see Lemma 6.9), we get the dual-inverse isomorphism
m ◦ ∼
(Z/p Z, EV
[p ] ) ⊂ Z/p Z ×V
m m et m m et
(μpm , EV
IsomV [p ] ) = IsomV EV
[p ] .
Thus, IV,m
is an étale finite covering over V.
/Spf(V)
Since IV,m
it is affine (Lemma 4.24), and we may
is étale finite over Spf(V),
write I = Spf(V m ). Then V
m is a semilocal normal V-algebra
étale finite
V,m
flat over V; so it is a product of complete discrete valuation rings whose
maximal ideal is, by étaleness, generated by the rational prime p.
(p)
Recall W = limn W/pn W, and take a modular form E on X/W lift-
←−
ing the Hasse invariant H (as in Sect. 6.2.12). Let X (p) [resp., Y (p) ] be a
formal completion of X [ E ]/W (resp., Y [ E ]/W ) along X (p) [ H1 ]/F (resp.,
(p) 1 (p) 1
S := Y (p) [ H1 ]/F ).
(p) [resp., Y (p) ] does not depend
Lemma 7.28 The p-adic formal scheme X
on the choice of the lift E.
Proof. Since the proof is the same, we deal with the case of X (p) . Take an-
other lift E of H . By replacing E and E by its power, we may assume
a
Define, for Γ = Γ1 (pm )p , a W-scheme YΓ = ImmY (p) (μpm , E) and a p-adic
formal scheme YΓ/W = ImmY (p) (μpm , E) ∼ = E[pm ]et − E[pm−1 ]et over Y (p) ,
which is étale finite over Y (p) . We obtain YΓ/K as in (7.1.1) by the base
change YΓ/K ∼ = YΓ/W ×W K. We may regard YΓ/W as the formal completion
of YΓ/W along YΓ/F := IsomS (μpm , E[pm ]◦ ). By definition, we have an open
immersion
I → Y m
V,m Γ1 (p )p /W Y/W
× (p) Spf(V),
312 7 Modular Curves as Shimura Variety
1
By definition, the moduli scheme Y (N, Γ )/F is a scheme over Y (N )[ H ]
(forgetting the level-p structure). We put IV,m = Spec(Vm ) and YΓ/F =
limpN Y (N, Γ )/F . Then
←−
YΓ1 (pm )p /F = IsomS (μpm , E[pm ]◦ ) =: Im
the group GI := GL1 (Zp ) × SL2 (A(p∞) ) acts on IV,m and hence on IV,m
(m = 1, 2, . . . , ∞), Spec(V) [by Lemma 7.5 (2)], FΓ , Im , YΓ/F , and YΓ/K .
Thus, we can make the étale quotient IΓ0 (pm ) := IV,m /GL1 (Z/pm Z).
Lemma 7.29 We have IΓ0 (pm ) = Spec(VΓ0 (pm ) ), and VΓ0 (pm ) is a valuation
ring finite flat over V sharing the same residue field with V.
Proof. As seen in Sect. 4.3.8, the connected component of a finite flat group
scheme over a Henselian local ring R with residue field F is uniquely deter-
mined. Thus, there is a unique connected subgroup of E (isomorphic to μpm
étale locally) if (E, φN )/R gives rise to an R-point of Y (N, Γ0 (pm ))/W . Thus,
for any m > 0, we have S/F = limpN Y (N, Γ0 (pm ))/F . This shows that the
←−
residue field of VΓ0 (pm ) is the function field of S and that the quotient field
of VΓ0 (pm ) is the field FΓ0 (pm ) .
and Vm is étale finite over VΓ0 (pm ) . This shows the following:
7.2 Igusa Tower 313
where v is a p-adic valuation of FΓ1 (pm )p unramified (of degree 1) over v0 and
Max(Vm ) is the set of maximal ideals of Vm .
The correspondence is given by
◦ ◦
v ↔ mv = {x ∈ Vm |v(x) > 0} ↔ Im with F(Im ) = Vm /mv .
Lemma 7.31 The action of G1 := SL2 (A(p∞) ) fixes vm = vΓ1 (pm )p and each
element of π0 (Im ).
Proof. Since FΓ1 (pm )p /FΓ0 (pm )p is a finite Galois extension, the set of exten-
sions of vΓ0 (pm ) to FΓ1 (pm )p is a finite set, and by the above lemma, it is in
bijection with π0 (Im ). Thus, the action of SL2 (A(p∞) ) on π0 (Im ) gives a fi-
nite permutation representation of SL2 (A(p∞) ). Since SL2 (k) of any field k
of characteristic 0 does not have any nontrivial finite quotient group (because
it is generated by divisible unipotent subgroups by the following lemma), the
action of SL2 (A(p∞) ) fixes every irreducible component in π0 (Im ).
Exercise 7.32 Why does unipotent generation imply that SL2 (k) of any field
k of characteristic 0 does not have any nontrivial finite quotient group?
Here is the lemma on the unipotent generation of SL(2):
Lemma 7.33 Let K be a field. Then SL2 (K) and P SL2 (K) are generated
by unipotent elements.
subgroup of P SL2 (K). Since the upper triangular unipotent group acts by
translation on the affine space A1 (K) = K, it acts transitively on A1 (K). The
opposite unipotent subgroup t U acts transitively on P1 (K) − {0} ∼ = A1 (K),
and H acts transitively on P (K). Thus, for any g ∈ P SL2 (K), we find
1
Since T (Z(p) ) and SL2 (A(p∞) ) fix v∞ (Lemmas 7.5 and 7.31), by the density
in Exercise 7.34, we conclude (G) for x given by the generic point associated
with v∞ in Sect. 7.2.1:
Theorem 7.35 We have D = GI /{±1}, and hence I is irreducible.
Note that Gal(I ◦ /S) is the stabilizer of I ◦ in Aut(I/S). Then, the irre-
ducibility of I follows from D = GI /{±1}, because this implies Aut(I/S) =
Gal(I ◦ /S), as argued in Sect. 7.2.1.
Let K (p) be an open compact subgroup of SL2 (A(p∞) ) and K = K (p) ×
GL2 (Zp ). Put YK = Y (p) /K (p) (which is the level-K modular curve). Let
IK = I/K (p) , which is the Igusa tower over YK . Since I is irreducible by
the above theorem, the quotient IK = I/K (p) is irreducible (as I IK is a
dominant morphism). Thus, we have reproved
Corollary 7.36 (Igusa) The Igusa tower IK over YK/F is irreducible for
K = GL2 (Zp ) × K (p) for any open compact subgroup K (p) of SL2 (A(p∞) ).
The following corollary is called the q-expansion principle and is a con-
sequence of the irreducibility of Iα combined with the fact that the local
q-parameter gives the uniformizing parameter at a cusp of Y1 (N ) over Z[ N1 ].
Corollary 7.37 (q-expansion principle) Let L/Fp be a line bundle realized
in the function field of Iα/Fp (α = 1, 2, . . . , ∞). Let s be a cusp and U be any
Zariski open neighborhood of s ∈ Iα/Fp . If the q-expansion at s of a section
f ∈ H 0 (U, L) vanishes, then f vanishes on U .
Ican = φp : μpα ∼
= Ẽ[pα ]◦ and t −1
Ican : Z/pα Z ∼
= Pα = Ẽ[pα ]et ,
−1
where t Ican is the Cartier dual inverse of Ican . We then define Vm,∞ = α Vm,α
and
V = VΓ = lim Vm,∞ and V = VΓ = V (Γ ; W ) = lim Vm,∞ .
−m
→ ←−
m
p , φN ) ∈ W [[q
f (Tate(q), zφcan 1/N
]] for all z ∈ Z×
p.
p , φN ) ∈ q
f (Tate(q), zφcan 1/N
W [[q 1/N ]]
an invariant differential over E, we have H 0 (E, ΩE/R ) = H 0 (E, Ω ); so
E/R
p on E such that
we have a unique invariant differential ωp induced by ω
H 0 (E, ΩE/R ) = Rωp .
We start with a fixed CM point x = [z, g] and the associated elliptic curve
(Ex , ηx ) with complex multiplication by Mx . See Sect. 7.1.8 for the notation.
We choose z ∈ H ∩ idx (M ) so that O = Z + Z · z. Unless confusion seems
likely, we write M for Mx . We suppose that O → End(Ex ) for the integer
ring O of M . Take W = W (Fp ) and consider the reduction E0 modulo (p)
of Ex for the maximal ideal (p) ⊂ W . Suppose that E0 is ordinary; thus, p
splits (p) = pp in O. We assume E0 [p∞ ] ∼ = μp∞ (see Corollary 6.40). By this
choice, idx = idz defined in Sect. 7.1.8 coincides with ip |M for the embedding
ip : Q → Cp we fixed at the beginning. We write v for the p-adic place of M
associated with idx . Thus, the ordinarity of E0 is equivalent to
(p)
the forgetful morphism: (E, ιE , ηE )/R → (E, ιE )/R of ℘ into the original
deformation functor EE0 in (6.3.2) induces an isomorphism of functors; so
they have canonically isomorphic deformation spaces (though it would be
better to view that they are not equal).
We consider the Serre–Tate deformation space Sx representing ℘.
We take
(p)
the Shimura–Igusa–Kottwitz model Sh/W over W and consider x = [z, g] as a
(p) (p)
W -point of a geometrically connected component Y/W of Sh/W . Abusing the
notation, we let S = Y (p) [ E1 ] [instead of X (p) ]; that is, we invert over Y (p) a
lift E of a power of the Hasse invariant H. Recall that the formal completion
Y (p) of Y (p) along Sm = S ⊗W Wm is uniquely determined independently of
318 7 Modular Curves as Shimura Variety
the choice of E and gives the ordinary locus of the universal elliptic curve E
(see Lemma 7.28). By our choice, x gives rise to a closed point of Y (p) .
Since the Serre–Tate deformation space G m carries the universal deforma-
(p) m ) = E (p) (G
tion E = (E, η )/Sx , which is an element of EG(Zp ) (G
(p) m ), by the
universality of the Shimura variety, we have an inclusion
choice of gp , which brings the base of Lp to the base given by the idempotents
−1p := (0, 1) of Op and 1p := (1, 0) of Op in Op × Op = Zp × Zp (1). We write
η ord = ηpord ×η (p) [this η ord incorporates the Galois action on Zp ×Zp (1), and if
we want to make explicit about ignoring the action, we write η = ηpord × η (p) ].
This ordering of Zp and Zp (1) is important as the canonical inclusion μpn →
C× /q Z for q = exp(2πiz) is induced by
1 1 1
→ n ∈ (2πi)(Z n + Zz)/(2πi)(Z + Zz),
pn p p
and as a convention, the basis of (2πi)(Z + Zz) is written as 2πit (z, 1) [i.e.,
the right-hand side 1 corresponding to the level-Γ1 (pn ) structure]. In other
−1 −1
words, as we have remarked, the action η → η ◦ g corresponds
? @to J?1 g @ J1
0 −1 1 1
on the lattice side for J1 = 1 0 , and by this choice J1 −1pp = 1pp , we
do not need to worry about conjugating by J1 (so we forget about J1 ).
We can think of the deformation of (E0 , ι0 , η0ord )/Fp for η0ord = η ord mod p.
The p-part of the level-p structure η0ord provides the canonical identification
of the deformation space S with G m . For any complete local W -algebra C
and any deformation E/C of E0 , E[p∞ ]et is étale over Spec(C); so, again,
the deformation is insensitive to the ordinary level structure. Thus, we get a
canonical immersion:
Here Eord (resp., E) denotes the universal ordinary triple over I (resp., the
m ).
universal triple over G
7.2 Igusa Tower 319
Note that the identification of Hom(T E0 [p∞ ]et , G m ) with the formal group E
of E is given by the Cartier duality composed with the polarization; so if α
= limn α(“pn ”α−c (vn )) (if we allow fraction
1−c
is p-integral, α sends t to tα
←−
power of t, this is valid without assuming the integrality of α). Thus, the effect
of α on t is given by t → tα . Once the identification of S with G
1−c
m is given
∞ ∼ ∞ et
(that is, a level-p structure η0 : Qp /Zp = E0 [p ] is chosen), nonzero
ord
group Gm , and all Zp -linear automorphisms are obtained in this way. On a
point u ∈ G m (W ) = 1 + mW , the action induces u → us .
The inclusion Z → O induces an identification of p-adic rings Op with
Zp , which we fix in this paper and use always in the sequel. Note that Op =
Op × Op . This same inclusion, Z → O, induces an inclusion of Z(p) -tori
320 7 Modular Curves as Shimura Variety
T (Z(p) ) → Z×
p = T0 (Zp ). (7.2.5)
Thereby, the action of T (Zp ) on S and that of T (Z(p) ) are compatible. The
torus T (Z(p) ) is isomorphic to the image of Tx (Z(p) ) in Shimura’s auto-
morphism group Aut(X (p) ) = G(A(p∞) )/Z(Q), and its action on Sx factors
through the action of GI on I via (7.2.2). The Zp -module structure of S given
by t → ts therefore commutes with the isogeny action of Tx on S.
By the level structure ηpord (and its dual), we identify E x with G m and
Ex [p∞ ]et with Qp /Zp . In this way, we may identify the torus T with the
diagonal torus T of G1 = SL(2). The action of t ∈ T0 (Zp ) = Z× p on the
quotient Ex [p∞ ]et is given by the multiplication by t ∈ Z× , and hence the two
−1 p
tori are identified by T0 (Zp ) t → 0 t ∈ T (Zp ) ⊂ SL2 (Zp ), taking the
t 0
the coordinate given by η ord (and its dual) under, which we identified as T0
and T . By the universality of Sh/Q , there exists an isogeny α : Ex → Ex such
that η (p) ◦ h̃ = α ◦ η (p) and η ord ◦ h̃ = α ◦ η ord . Since α ∈ End(Ex ) = O, we
have h̃ = rx (α) modulo Z(Q), and therefore, h is the image of rx (α) in GI .
Then the assertion follows from the above discussion.
We have defined the two-level structure ηpet and ηp◦ of the canonical lift Ex
of E0 := Ex ⊗W F, which automatically (and uniquely) extends to the level
structure η et ∼ ∞ et
and η ◦p : μp∞ ∼
= E[p∞ ]◦ over G m/W . Let
p : Qp /Zp = E[p ]
ip
ζn = e( p1n ) [the canonical generator of μpn (Cp ) we chose in Proposition 6.51].
By Proposition 6.51, after extending scalars to Spf(W [t1/p n
, t−1/pn ][ζn ]), the
finite flat group scheme E[pn ] splits as E[pn ]◦ × C1 , and η p mod pn induces
this splitting (though the splitting is up to the truncated Barsotti–Tate group
of order pn ). We consider an open compact subgroup Γ ⊂ G(Z (p) ) containing
/ 0
Γ(N )(p) = x ∈ G(Z (p) )x ≡ 1 mod N · M2 (Z (p) )
(p)
for a positive integer N prime to p. Let IΓ be the Igusa tower over ShΓ =
Sh(p) /Γ . Then we have the following result in [Br1] Sect. 7:
E ×G m Spf(W [t1/pn
, t−1/pn ][ζ])
π
−−−−→ E/Cζ /Spf(W [t1/p
n −1/pn
⏐ ⏐ ,t ][ζ])
⏐ ⏐
. .
Spf(W [t1/pn
, t−1/pn ][ζ]) −−−−−−→
Spf(W [t1/pn
, t−1/pn ][ζ])
n
t
→t1/p ζ
pn u pn 0
brings η ord to η ord ◦ g, where gp = 0 1
and gl = 0 1
for all primes l = p.
322 7 Modular Curves as Shimura Variety
Proof. Our proof is slightly different from the proof given in [Br1]. By Propo-
sition 6.51, the induced map in the bottom line of the commutative diagram
n
in the proposition is given by t → ζt1/p for the canonical coordinate t with
respect to η. Thus, we need to determine g. Since the connected-étale exact
sequence of E[pn ] splits over an fppf extension of G m , it is canonical over
W [t1/pn
, t−1/p ], and the splitting is given by C1 (i.e., u = 0). Thus, up
n
Exercise 7.46 Let the assumption be as in the above lemma. Prove that the
quotient map E → E/η(μ n ◦
1 p0n ) = E/E[p ] is defined over Gm induced by η →
η ◦ h for h with hl = 0 p n for all primes l and that the following diagram
is commutative:
π
E −−−−→ E/E[pn ]◦
⏐ ⏐
⏐ ⏐
. .
Spf(W
[t, t−1 ]) −−−−→
n
Spf(W
[t, t−1 ]).
t
→tp
This follows from [K1] Theorem 4.3.1, but we give a different proof without
using the Gauss–Manin connection on the deformation space.
Proof. We use the notation introduced between (1.3.9) and (1.3.10). Since a p-
adic modular form is a p-adic limit of classical modular forms, we only need to
verify the result for
classical forms. Then for a classical modular
form f with q-
expansion f (q) = m∈Z a(m, f )q m and t-expansion f (t) = m∈Z c(m, f )tm
we defined in (1.3.10) the operator d as
(over S),
df (q) = lim f |φn (q) = na(n, f )q n ,
n→∞
n
as desired.
1−c
Remark 7.49 The action of ρ(α) for α ∈ Tx (Z(p) ) induces t → tα by
Proposition 7.44. Then the above lemma supplies us with the final missing
item in the proof of Lemma 1.33; so Lemma 1.33 has now been proven.
V ord [k] ∼
= eH 0 (S0 , ωk ⊗ Tp ) = eH 0 (X, ω k ⊗ Tp ) (Tp = Qp /Zp ),
where Tp = Qp /Zp . The last identity follows from the affineness of S0 . When
we generalize this point to classical groups bigger than G = GL(2)/Q , the
affineness is no longer true; so we need to work out this point (the base-
change property) carefully to bring Tp inside the cohomology group outside
326 7 Modular Curves as Shimura Variety
the cohomology group. In the elliptic modular case, simply by the affineness
of S0 , (7.3.2) shows that V[k] is p-divisible, and its direct summand eV[k] is
also p-divisible. We show that e contracts V[k] into H 0 (S0/W , ω k ⊗ Tp ).
We consider the following condition:
(F) corankW eV[k] = rankW Hom(eV[k], Tp ) is finite for an integer k.
In practice, this condition is often proven by showing
where
The left-hand side (LHS) of (7.3.3) is p-divisible, since S0 is affine. The right-
hand side (RHS) is of finite corank, since X1 (N ) is projective. Thus, eV[k] is
p-divisible of finite corank.
Decompose Z× p = ΓT × Δ for a p-profinite group ΓT and a prime-to-p
finite group Δ. For simplicity, suppose that p > 2. Then Δ has order p − 1
prime to p, and ΓT is isomorphic to Zp . For the generator γ = 1 + p of
ΓT , we have W [[ΓT ]] ∼
= W [[x]] = Λ via γ → 1 + x [i.e., γ s → (1 + x)s =
∞ s j × ord
j=0 j x ], and W [[Zp ]] = Λ[Δ]. Let V be the Pontryagin dual module of
eV. See [FAN] Chaps. 1 and 3 for generalities on topological groups (including
Pontryagin duality and profinite groups), and [LFE] Chap. 8 also contains a
short exposition on this subject. The Pontryagin dual for p-torsion modules M
(equipped with discrete topology) is given by M ∗ = Homcont (M, Tp ) for Tp =
Qp /Zp with discrete topology. Writing M = limi∈I Mi for finite submodules
−→
Mi of order pn(i) , we have
∗
and the topology on M is the profinite topology with respect to the finite
discrete groups Hom(Mi , p−n Z/Z). Then M ∗ is a compact profinite group.
The duality is perfect; so if M → Y Z is an exact sequence of p-torsion
modules, Z ∗ → Y ∗ M ∗ is an exact sequence of profinite groups. In par-
ticular, if M is p-divisible and p-torsion, the multiplication by p is surjective;
so, by duality, multiplication by p on M ∗ is injective, which shows that M ∗
is torsion-free if M is p-divisible and p-torsion.
Assume that a profinite (noetherian) W -algebra R acts on M continuously.
For any closed ideal a ⊂ R, Pontryagin’s dual of M [a] = {x ∈ M |ax = 0} is
given by M ∗ /aM ∗ , where R acts on M ∗ continuously by the pullback (dual)
action. To see this, we choose a finite set of generators (a1 , . . . , ar ) of a and look
x
→(a1 x,...,ar x)
into the exact sequence: 0 → M
[a] → M −−−−−−−−−−→ M r . Taking the dual,
(x1 ,...,xr )
→ ai x i
we find that M ∗ r −−−−−−−−−−−
i
−−→ M ∗ → M ∗ /aM ∗ → 0 is exact, which
7.3 Elliptic Modular Forms of Slope 0 327
In the middle equality, we have assumed (7.3.3). Here the subscript or super-
script ord indicates the image of e.
Write Δ = Hom(Δ, W × ). Note here that each χ ∈ Δ is induced by in-
∗
finitely many 0 < k ∈ X (T ). Decompose V ord
into the sum of χ-eigenspaces,
:
Vord = Vord [χ].
χ∈Δ
where s = s(χ) = corankW eV[k]. If (F) holds for one k, it holds for all
κ ∈ X ∗ (T ) inducing χ, and π is an isomorphism for the following reason: s
is the minimum number of generators of Vord [χ] ⊗Λ,κ W over W . We know
that this module is W -free, because its dual V[κ] is p-divisible; so it is free of
5 an isomorphism moduloκ (1 + x) − γ for all
κ
rank s. The morphism π induces
κ inducing χ. Then Ker(π) ⊂ κ Ker(π mod (1 + x − γ )) = 0, and we get
)
Theorem 7.50 Suppose that (F) holds for one k. Then Vord = χ Vord [χ]
is Λ-free of finite rank, and if (7.3.3) holds for κ ∈ X ∗ (T ) with κ|Δ = k|Δ ,
then
328 7 Modular Curves as Shimura Variety
Vord ⊗W [[Z×
p ]],κ
W ∼
= HomW (Gord
κ (Γ1 (N ); W ), W ).
Vord,cusp ⊗W [[Z×
p ]],κ
W ∼
= HomW (Sκord (Γ1 (N ); W ), W ).
for the projector e [H86b] Section 4. Here L(k; K) = kj=0 KX j Y k−j is the
K-vector space of homogeneous polynomials φ(X, Y ) in (X, Y ) of degree k on
which γ ∈ Γ1 (N ) acts by γ · φ(X, Y ) = φ((X, Y )γ). We can ease the condition
k ≥ 3 to k ≥ 2 if we replace Gordk (Γ1 (N ); K) by Gk (Γ1 (N ) ∩ Γ0 (p); K) in
ord
Let (E, φp , φN )/R be a p-adic test object (i.e., φp : μp∞ → E[p∞ ] and R is a
p-adic ring). For the moment, we assume that R is of characteristic 0. Each
subgroup C of order p outside the image of φp is étale; so we can think of the
quotient (E/C, φp , φN ) defined over a (possibly “radiciel”) finite extension R
of R. To make this R explicit, we may assume that R is a complete local
p-adic algebra (by taking a completion). Then, writing t for the Serre–Tate
coordinate of (E, φp , t φ−1
p ) for the Cartier dual inverse φp
t −1
of φ, we have
1/p
R = R[t ] by (the version specialized to E of) Proposition 6.51. Thus,
R is R-free of finite rank; so, identifying R ∼ = Rp (column vector space)
as an R-module, we can write multiplication by r ∈ R on R as a matrix
form ρ(r) ∈ Mp (R); hence, t (a1 r, . . . , ap r) = ρ(r)t (a1 , . . . , ap ). Then we define
Tr = TrR /R : R → R by Tr(r) = Tr(ρ(r)), where right-hand side is the
matrix trace of ρ(r).
Definition 7.52 We define the U (p) operator acting on a (p-adic or classical)
geometric modular form f by
1 (∗) 1
f |U (p)(E, φp , φN ) = Tr(f (E/C, φp , φN )) = f (E/C, φp , φN ).
p p
C
We have a natural inclusion map: GΓ1 (N ) (R) → VΓ1 (N )/R , since we may regard
f ∈ Gk (Γ1 (N pα ); R) as a p-adic modular form by putting f (E, φp , φN ) =
f (E, φp |μpα , φN , ωp ) as in Proposition 7.41. One can prove the p-adic density
of D = GΓ1 (N ) (W )[ 1p ] ∩ V in V (see [GME] Corollary 3.2.4). Using this fact,
we can show that we can write V as a union of finite modules stable under
U (p); so e = limn→∞ U (p)n! exists. The final result is as follows.
Theorem 7.53 (Vertical control theorem) We have
∼ HomW (Gord
k (Γ1 (N ); W ), W ) if k ≥ 3,
V ⊗W [[Z×
ord
W =
2 (Γ1 (N ) ∩ Γ0 (p); W ), W )
p ]],k
HomW (Gord if k = 2.
G(χ; Λ) ⊗Λ,k W ∼
= HomΛ (Vord [χ], Λ) ⊗Λ,k W
∼
= HomW (Vord [χ] ⊗Λ,k W, W ) ∼
= HomW (HomW (Gord (Γ1 (N ); W ), W ))
k
∼
= Gord
k (Γ1 (N ); W ) k (Γ1 (N ) ∩ Γ0 (p); W ).
= Gord
7.4 Hecke Algebras 331
Here is the explanation of how to obtain the last isomorphism in the above
sequence of isomorphisms. As is well known (see Corollary 6.34 and (7.4.1)),
the Hecke operator T (p) acts on f ∈ Gk (Γ1 (N ); W ) by
n
a(n, f |T (p)) = a(pn, f ) + a( , f |p),
p
By the theory of primitive forms ([MFM] Theorem 4.6.17; see also [H85] Sec-
tion 3), this inclusion is surjective if k ≥ 3 (not surjective if k = 2, for example,
if N = 1 and p = 11 or p ≥ 17). We leave the verification of assertion (3) to
the reader.
For a given element Φ ∈ G(χ; Λ), we get a family of p-ordinary elliptic modular
forms {Φ(γ k − 1)}k≥2 whose q-expansion coefficients depend (p-adic) analyt-
ically on the weight k ∈ Homcont (T (Zp ), Z×p ). Each element Φ ∈ G(χ; Λ)
is called a p-ordinary Λ-adic form, and {Φ(γ k − 1)}k≥2 is called the p-adic
analytic family of modular forms associated with a Λ-adic form Φ.
Remark 7.55 As we mentioned in Remark 7.51, there is a version of this for
Γ0 (N ) and cusp forms. For k ≥ 2, we have
1. F (Γ0 (N ), χ; Λ) is Λ-free of finite rank;
2. F (Γ0 (N ), χ; Λ) ⊗Λ,k W ∼ −k
= Fkord (Γ0 (N p), χω ; W );
3. the above identification is induced by Φ → n a(n, Φ)(γ k − 1)q n ∈ W [[q]],
where χ is a character modulo N p and F = G and S (see [GME] Chap. 3).
We put F (Γ0 (N ), χ; I) = F (Γ0 (N ), χ; Λ) ⊗Λ I for a Λ-algebra I.
1
f |T (n)(E, φN , φp ) = f (E/C, πC,∗ φN , πC,∗ φp ).
n
C∈Cn (E)
mn
a(m, Φ|T (n)) = a( , Φ|d). (7.4.2)
d2
0<d|(m,n),(d,N p)=1
for all n, and therefore T (n) gives rise to the Λ-linear operator T (n) on
F (Γ0 (N ), χ; Λ). As seen in [GME] Theorem 3.2.16, we may regard the space
F (Γ0 (N ), χ; Λ) as a subspace of VΓ1 (N )/Λ , and we can verify the operator
T (n) introduced in the previous section preserves F0 (Γ0 (N ), χ; Λ), inducing
the T (n) formally defined by (7.4.2).
Proof. Since the space over I is the scalar extension of the space over Λ, we
may assume that I = Λ. For simplicity, we write S = S(Γ0 (N ), χ; I) and
h = h(Γ0 (N ), χ; I). Let m be the maximal ideal of Λ with F = Λ/m. By
definition, h/mh surjects down to h(S/mS) as the two algebras are generated
by T (n). This shows the morphism h → HomΛ (S, Λ) induced by the pairing
gives rise to
again by the result over now the field Q. Thus, i is an isomorphism. Since h is
the Λ-dual of the Λ-free module S, h is Λ-free. Then, by applying HomΛ (?, Λ)
to h ∼
= HomΛ (S, Λ), we recover
as desired.
By Remark 7.55 and the above proposition, we have now proven the result
we used in Sect. 3.1:
Corollary 7.61 We have a canonical isomorphism
h/((1 + x) − ε(γ)γ k )h ∼
= hord
k (Γ0 (N p
r+1
), εχω −k ; W )
√
We return to the setting of Sect. 6.4; thus, M = Q[ −D] ⊂ Q with dis-
criminant −D is an imaginary quadratic field in which the fixed prime (p)
splits into a product of two primes pp with p = p. As before, assume that
p = α ∈ O|ip (α)|p < 1 and denote by c the nontrivial field automorphism
of M (complex conjugation). For an ideal C of the integer ring O of M , we
consider the ideal group I(C) of fractional ideals prime to C. We put
P(C) = (α)α ∈ M × and α ≡ 1 mod C in OC .
As is well known, Cl(C) = I(C)/P(C) is a finite group (see [LFE] Sect. 1.2).
An ideal Hecke character of M modulo C is a character (i.e., a group homo-
×
morphism) λ : I(C) → Q such that there exist integers k, j with λ(α) =
α−k α−j = α−k α−jc for all principal ideals (α) in P(C). Here M α → α ∈ C
is induced by i∞ and α = αc is the complex conjugation of α ∈ M induced
by i∞ .
The linear combination ∞(λ) := k · i∞ + j · c ∈ Z[Gal(M/Q)] is called
the infinity type of λ [note here the sign change from the exponent of λ(α) =
α−k−jc ; the reason will be explained later]. Since λ has values in M in the
subgroup P(C) ⊂ I(C) of finite index, it has values in a finite extension of M
inside Q all over I(C). Also, a Hecke character has finite order if and only if
k = j = 0. A character λ of I(C) is called anticyclotomic if λ(Ac ) = λ(A)−1 for
all A ∈ I(C), while a character satisfying λ(Ac ) = λ(A) is called cyclotomic.
If C|C , we have a natural inclusion I(C ) → I(C); hence, we can pull back
a Hecke character modulo C to a Hecke character modulo C . If C C , the
character obtained by pullback is called imprimitive. A character is called
primitive if it is not imprimitive. If λ is primitive modulo C, the ideal C is
called the conductor of λ. The norm N (A) = [O:O∩A]
[A:O∩A] is an example of a Hecke
character with ∞(λ) = −i∞ − c. If a primitive λ has conductor C, we extend
×
λ : I(C) → Q to I(O) → Q by putting λ(A) = 0 if A ∈ I(C). Then the
(primitive) L-function of λ is defined by an absolutely and locally uniformly
convergent Dirichlet series/Euler product on a right half-complex plane:
λ(L)
−1
L(s, λ) = λ(A)N (A)−s = 1− ,
N (L)s
A L
where A runs over all O-ideals and L runs over all prime ideals of O. It is a re-
sult of Hecke that this L-function can be continued to a meromorphic function
on the whole complex plane with holomorphy everywhere except for the case
of a power of the norm character λ = N m (see [LFE]
/ Sect.
2.7 and0 Chap. 8).
λ(Q)
For a prime ideal 0 = q of Z, we put L (s, λ) =
(q)
Q|q 1 − N (Q)s L(s, λ)
(the imprimitive L-function whose Euler q-factor is removed).
In this chapter, we fix a prime l = (l) (p) of Z. Generalizing the result
(l)
of Sect. 1.2, we prove nonvanishing modulo p of Hecke L-values L (0,λχ)
Ω for
the canonical period Ω for “almost all” anticyclotomic characters χ of finite
order with l-power conductor (under some mild assumptions; Theorem 8.31).
This topic is a generalization of Washington’s result [Ws] (and Sinnott [Sn2])
from Dirichlet L-values to Hecke L-values, and a more general case of Hecke
characters of a CM field has been dealt with in [H04b], [H07a], and [Hs1] by
the same technique, resorting to a Hilbert modular Shimura variety in place
of a Shimura curve here. There will be much room for generalization of the
techniques (we list here only one of them: [Br2]). The phrase “almost all”
means “except for finitely many,” that is, nonvanishing modulo p outside a
(unspecified) finite set.
Here the first identity follows from the fact that ω(a) is induced by ω(O) on
E(O). We suppose that a is prime to p. Recall Shimura’s rationality result for
f ∈ Gk (N ; W) we reproduced in Theorem 1.27:
(δkr f )(x(a), ω∞ )
k+2r
= (δkr f )(x(a), ω(a)) ∈ Q. (Sh)
Ω∞
Katz interpreted the differential operator in terms of the Gauss–Manin
connection of the universal elliptic curve over Sh(p) and gave a purely algebro-
geometric definition of the p-adic invariant differential operator (see [K2]
Chap. II and [HT1] Sect. 1). We have given in Sect. 1.3.6 a more elementary
construction of his operator. As before, we write his operator correspond-
ing to δ∗k as dk : VΓ1 (N )/B → VΓ1 (N )/B for a p-adic base Zp -algebra B.
The level-p structure ηp◦ (a) : μp∞ ∼ = Mp /ap → E(a)[p∞ ] (p corresponding
m ∼
to ip |M : M → Cp ) induces an isomorphism ηp◦ (a) : G
= E(a) for the p-adic
/W obtained by formal completion of E along the origin
formal group E(a)
0E ∈ E(F). Then ω(O) = ω(a) = Ωp ωp (Ωp ∈ W × ) for ωp = ηp◦ (a)∗ dt t . By
Theorem 1.34, we have, for f ∈ Gk (N ; W),
Consider the torus TM = ResM/Q Gm [see Example 4.58(11) for the definition
of ResM/Q G]. Each integral linear combination k · i∞ + j · c ∈ Z[Gal(M/Q)]
is regarded as a character of TM by x → i∞ (x)k c(x)j . We fix an arithmetic
Hecke character λ of infinity type k · i∞ + κ(i∞ − c) for the integers k and κ.
×
Originally, λ is defined as a character λ : I(C) → Q with λ(α) = α−k−κ(1−c)
for (α) ∈ P(C). We can extend λ to idele characters of MA× in two ways.
For each prime L of M , we choose its local generator L ∈ M so that
LO(L) = (L ), where
b
O(L) = a, b ∈ O, aO + L = O (the localization at L).
a
Exercise 8.5 Let Cl(Cp∞ ) = limn Cl(Cpn ) as a profinite compact group and
5 ←−
U (Cp∞ ) = n U (Cpn ). Then prove that (i) TM (Q)\TM (A(∞) )/U (Cpn ) ∼ =
Cl(Cpn ) as a group if n < ∞ and as a compact group if n = ∞ and (ii)
factors through Cl(Cp∞ ).
λ
8.1 Rationality of Hecke L-Values 339
Let λ be an idele character of MA× of infinity type k + κ(1 − c). Fixing two
primes p (we work in the mod p and the p-adic setting) and l (we consider
l-power conductor characters), we assume the following three conditions:
(ct) k > 1 and κ ≥ 0.
(pl) The conductor C of λ is prime to p and (l) = l.
(sp) The ideal C is a product of primes split over Q.
The first condition implies criticality at s = 0 for L(s, λ). The two other
conditions are for simplicity and can be removed [see [Hs1] how to proceed
without assuming (sp)].
Since ℘Γ1 (1) (C) is identified with the complex manifold B made of lattices
L ⊂ C in Sect. 2.3.4, a function ψ as above can be interpreted as a function
of lattices L with level-Γ structure φf = (if , if ) = (i, i ) defined in Sect. 6.4.4.
Here i : X ∗ → f−1 L/L is the level-f structure. We define an Zf -submodule
P V (L) ⊂ L ⊗Z Qf specified by the following conditions:
340 8 Nonvanishing Modulo p of Hecke L-Values
i
0 → f−2 /Z −
ĩ
→ P V (L)/fLf −
→ Z/f → 0.
ĩ i
f−2 /Z −−−−→ P V (L)/fLf −−−−→ Z/f
⏐ ⏐ ⏐
⏐ ⏐ ⏐ (8.1.7)
. f .∩ f .∩
For any positive integer k > 0, we can now define the Eisenstein series Ek .
Writing L = (L, φ = φf ) for simplicity, we define the value Ek (L; ψ) by
8.1 Rationality of Hecke L-Values 341
P ψ(w)
Ek (L; ψ) = (−1)k Γ (k + s) . (8.1.10)
wk |w|2s s=0
w∈f−1 L/Z×
Here “ ” indicates that we are excluding the term corresponding to w = 0
from the summation. As shown by Hecke, this type of series is convergent when
the real part of s is sufficiently large and can be continued to a meromorphic
function well defined at s = 0 [as long as either k ≥ 3 or ψ(a, 0) = 0 for all a;
see Sect. 9.1.1 in the text and [LFE] Sect. 2.5 for analytic continuation].
Lemma 8.8 If either k ≥ 3 or ψ(a, 0) = 0 for all a, the function Ek (ψ) gives
an element in Gk (Γ1 (f2 ); C), whose q-expansion at the cusp ∞ is given by
a k−1 n
Ek (ψ)(q) = 2−1 L(1 − k, ψ) + ψ(a, b) a q , (8.1.11)
|a|
0<n∈Z (a,b)∈(Z×Z)/Z×
ab=n
k
N (n)
ψ(n, 0) |N (n)|−s .
|N (n)|
n∈Z/{±1}×
ψ ◦ (a, b) = G(λ−1 −1
F )λF (a)λFc (b) (8.1.13)
for the Gauss sum G(λ−1 F ) =
−1
u∈Z/f λF (u)eQ (−u
−e(f)
). Here we agree to
−1 × ×
put λF (a)λFc (b) = 0 if (a, b) ∈ (Z/f) × (Z/f ) .
By the Fourier inversion formula (see [LFE] Sect. 4.1(5a)), we have
Then, from the effect of T (l) and U (l) on q-expansion (as in Exercise 6.22),
we verify the following lemma easily.
Lemma 8.10 Let l be a prime ideal of Z outside f generated by 0 < l ∈ Z.
Then we have
(as i is the dual of i under the Cartier self-duality of E[ff ] in Sect. 6.1.6). Note
that f |[l](q) = lk λF (l)f for f ∈ Gk (Γ0 (ff ), (λ ) because as seen in Sect. 7.1.9,
the action
of [l]
corresponds
to the action on the level structure: η → η ◦
1 0 1 0 (l)
0 l−1
l
with 0 l−1 l
≡ 0 01 mod Z(Q). If one wants to compute the
l
action of U (l) and T (l) via q-expansion, we need to compute the corresponding
q-expansion for this two-sided Neben character. Or we can directly compute
344 8 Nonvanishing Modulo p of Hecke L-Values
the effect using the geometric definition of the Hecke operators as in Sect. 6.2.5.
This way of using the geometric definition (Definition 6.21) is adopted in
[H07a] Sect. 1.3.
Combining the effect of the operator (l) explained before the lemma with
the computation of [l] in Sect. 7.1.9, for f = Ek (ψ ◦ ), we get
Remark 8.12 By (7.4.1) and a(n, df ) = n · a(n, f ), the Hecke operators T (l)
and U (l) satisfy T (l) ◦ d = l · d ◦ T (l) and U (l) ◦ d = l · d ◦ U (l) for the Katz
differential operator d. Set E(λ) = dκ Ek (ψ) (if neither k = 2 nor ψ = 1) and
E (λ) = dκ Ek (ψ). Put E(λ) = −dκ (Δ−1 dΔ − Δ−1 dΔ|(l)|[l])/24 if k = 2 and
ψ = 1. Then we have, under the notation of Lemma 8.11,
We now look into the degeneracy operator [q] for a prime q = (q) (with
0 < q ∈ Z) outside the level f. This operator brings a level Γ0 (q)-test object
(E, C, φ) with level f structure i outside q to (E/C, φ), where the level-f struc-
ture i is intact under the quotient map: E → E/C. On the lattice side, taking
the lattice LC with LC /L = C, it is defined as follows:
In short, we have
[q] ◦ U (q) = (q)−1 . (8.1.20)
Exercise 8.14 Make explicit the relation between the two operators l and
(l) for a prime l > 0 outside the level.
Pick a prime l in Z and write l = lO. Recall from Sect. 6.4.2 the order On =
Z + ln O of conductor ln . We determine the type of Hecke L-function obtained
by values of the Eisenstein series at CM points. Results equivalent to the
ones presented here are more detailed in [LAP] V.3.2 (which were obtained
independently by H. Yoshida). Since lOn ⊂ On+1 , a → la gives a map from
the set of On -ideals into the set of nonproper On+1 -ideals.
Proposition 8.15 Let In be the group of all proper fractional On -ideals. As-
sociating with each On+1 -ideal a the On -ideal On a, we get the following ho-
momorphism of groups πn : In+1 → In . The homomorphism πn is surjective,
and the kernel of πn is isomorphic to O× ×
n,l /On+1,l . After taking the quotient
by principal ideals, we still have the following exact sequence:
1 → O× × ×
n,l /On+1,l On → Cln+1 → Cln → 1,
× ×
we get the last assertion, noting that M × ∩ M∞ On = O×
n.
Writing a prime element of L and l as and l, respectively, we find that
On+1,l /O× ×
n+1,l = {1} lOn,l /On+1,l . (8.1.21)
j
When l splits into a product of two primes LL in M , Li L ⊂ On if and only
if i ≥ n and j ≥ n, and hence, we have
4
On,l /O× ∼ j × ×
∞ n+i n+j ×
Ol /O×
n,l = j=0 () On−j,l /On,l
n−1
i,j=0 n,l .
(8.1.22)
where a runs over all proper On -ideals (or equivalently, a runs over O n -integral
× × ×
ideles in In = MA /M∞ On ). Even if χ is a primitive character of Clf with
f > 0, we could have χ(a) = 0 even if a + lOn On (for n ≥ f ) in this
definition, as the proper On -ideal a is equivalent to another proper ideal b
prime to lOn . Thus, the l-Euler factor Lfl (s, χ) of Lf (s, χ) can be nontrivial.
Exercise 8.16 Prove that the L-function Ln (s, χ) has an Euler product
whose Euler factor is given by (1 − χ(Q)N (Q)−s )−1 for each prime Q l.
8.1 Rationality of Hecke L-Values 347
We write L(s, χ) for L0 (s, χ), which is the classical Hecke L-function. The L-
function Ln (s, χ) depends on n, because the set of proper On -ideals depends
on n. We are going to compute the dependence on n. Since each proper ideal
b ⊂ On equivalent to a−1 can be written as b = αa−1 with α ∈ a, we have
χ((α))N (α)−s = χ(a)N (a)−s χ(b)N (b)−s .
α∈a b∼a−1
We write Lna−1 (s, χ) = b∼a−1 χ(b)N (b)−s . Then we get
Ln (s, χ) = Lna−1 (s, χ) = χ(a)−1 N (a)s χ((α))N (α)−s .
a∈Cln a∈Cln α∈a
where If +k,l is the set of all proper Of +k -ideals a with N (a) = lm for some
m ≥ 0. Note that If +k,l ∼ = (Of +k,l ∩ Ml× )/O× ×
f +k,l by a → αOf +k,l for α ∈
Of +k,l with al = (α).
We first deal with the case where l splits in M/Q. As N (L) = l, by (8.1.22),
writing χl = χ|M × , we have χl () = χ(lOf +k ) and
l
f +k−1
Lfl +k (s, χ) = χl ()j l−2sj χl (u)
j=0 u∈Uf +k−j
+ χl ()f l−2(f +k)s χl (i j )l−(i+j)s χl (u),
i,j u∈Uf +k
where Uj = O× ×
f +k−j,l /Of +k,l . We
have u∈Uf +k−j χl (u) = 0 unless j =
k
Lfl +k (s, χ) = χl ()j lj−2sj with χl () = χ(l). (8.1.26)
j=0
k
M/Q
Lfl +k (s, χ) = 2j j−2sj
χl () l 2
with χl () = χ(l) if = 0,
j=0
l
k
M/Q
Lfl +k (s, χ) = j j−2sj
χl () l with χl () = χ(l) if = −1.
j=0
l
(8.1.27)
348 8 Nonvanishing Modulo p of Hecke L-Values
In summary, we have
Lk+1
l (s, χ) − χ(l)l1−2s Lkl (s, χ) = 1 for k > 0 and k ≥ f > 0, (8.1.29)
M/Q
Ll (s, χ) − l −
1
χ(l)l−2s L0l (s, χ) = 1 for f = 0. (8.1.30)
l
Pick the Néron differential ω(O) on E(O) in (Ω1) of Sect. 6.4.3, and recall the
Néron period Ω∞ ∈ C× given by ω(O) = Ω∞ ω∞ in (8.1.2). Since Cl∞ is an
almost pro-l group, all finite order characters of Cl∞ have values in W[μpm ] if
every element of the finite p-Sylow subgroup of Cl∞ is killed by pm . Replacing
W by W [μpm ] = W (Fp )[μpm ], we hereafter write W for i−1 p (W (Fp )[μpm ]). Put
L(l) (s, χ) = Lf (s, χ)Lfl (s, χ)−1 (removing the Euler factor at l). We will prove,
assuming C = 1, the following theorem at the end of this chapter after a long
preparation (a proof in the general case where C = 1 can be found in [H07a]):
aOn /a ⊂ E(a)(C) gives rise to a canonical level-Γ0 (l) structure to x(a) defined
over W (as p = l). We hereafter in this section write
x(a)/W := (E(a), if (a), if(a), C(a))/W , (8.1.31)
1
which is a test object of level Γ1,0 (f2 )∩Γ0 (l). For a p-adic modular form f of the
form d g with classical g ∈ Gk (Γ1,0
κ 1
(f2 ) ∩ Γ0 (l); W), we have by Theorem 1.34
dκ f (x(a), ωp ) κ δkκ f (x(a), ω∞ )
= δ k f (x(a), ω(a)) = .
Ωpk+2κ Ω∞k+2κ
π κ Γ (k+κ) √
We write c0 = (−1)k Im(δ) √
κ DΩ k+2κ
with 2δ = −D. Here Γ (s) is the
∞
Euler’s Gamma function [i.e., Γ (k + κ) = (k + κ − 1)!]. By definition, we
find, for e = [O× ×
n+1 : Z ] (which is equal to 1 if n > 0 since a root of unity
congruent to ±1 modulo l2 is ±1),
(c0 e)−1 δkκ Ek (x(a), ω(a))
λ−1 e(F)
C ( w)
= k+κ(1−c) N
s s=0
−1 ×
w M/Q (w)
w∈F a/On+1
λ−1
C (w)λ
−1
(w(∞) )
= λ−1
C (
e(F)
)
NM/Q (w)s s=0
w∈F−1 a/O×
n+1
(8.1.32)
λ(wF)
= λ−1
C (
e(F)
)
NM/Q (w)s s=0
w∈F−1 a/O×
n+1
λ(wFa−1 )
= λ−1
C (
e(F)
)λ(a)NM/Q (Fa−1 )s
NM/Q (wFa−1 )s s=0
wFa−1 ⊂On+1
= λ−1
C (
e(F)
)λ(a)Ln+1
[Fa−1 ] (0, λ),
where for an ideal class [c] ∈ Cln+1 represented by a proper On+1 -ideal c,
Ln+1
[c] (s, λ) = λ(b)NM/Q (b)−s
b∈[c]
is the partial L-function of the class [c] whose summation is over b running
over all On+1 -proper integral ideals in the class [c] prime to C. In the second
and third lines of (8.1.32), we regard λ as an idele character and in the sum
of the other lines as an ideal character. For an idele a with aO = aO and
aC = 1, we have λ(a(∞) ) = λ(a) and λ(a (p) ) = λ(a).
We put E(λ) = dκ Ek (ψ) and E (λ) = dκ Ek (ψ) as in Remark 8.12. We want
to evaluate E(λ) and E (λ) at x = (x(a), ω(a)) = (E(a), if (a), if (a), C(a), ω(a)).
Then, by definition, we have
E (λ)(x) = δkκ Ek (ψ ◦ )(x) − l−1 δkκ Ek (ψ ◦ )(x(aOn ), ω(aOn ))
(8.1.33)
E(λ)(x) = δkκ Ek (ψ ◦ )(x) − δkκ Ek (ψ ◦ )(x(laOn ), ω(laOn ))
350 8 Nonvanishing Modulo p of Hecke L-Values
because C(a) = aOn /a, and hence [l](x(a)) = x(aOn ) and (l)(x(a)) = x(la).
To simplify the notation, write φ([a]) = λ(a) −1 φ(x(a), ω(a)). By (8.1.15),
for φ = E(λ) and E (λ), the value φ([a]) only depends on the ideal class [a] but
not on the individual a. The formula (8.1.33) combined with (8.1.32) shows
e−1 λC (e(F) )E (λ)([a]) = c0 Ln+1−1
[Fa ] (0, λ) − l −1 n
L −1
[Fa On ] (0, λ)
e−1 λC (e(F) )E(λ)([a]) = c0 L[Fa−1 ] (0, λ) − λ(l)Ln[Fl−1 a−1 On ] (0, λ) ,
n+1
(8.1.34)
κ √
where e = [O× × π Γ (k+κ)
n+1 : Z ] and c0 = (−1) Im(δ)κ DΩ k+2κ , with 2δ =
k √ −D.
∞
Recall Ln (s, λ) as in (8.1.25) for χ = λ. For each primitive character
×
χ : Clf → Q , taking n = max(f, 1), we have
e−1 λC (e(F) ) χ(a)E (λ)([a])
[a]∈Cln+1
= c0 · χ(F) Ln+1 (0, λχ−1 ) − Ln (0, λχ−1 )
e−1 λC (e(F) ) χ(a)E(λ)([a])
[a]∈Cln+1
= c0 · χ(F) Ln+1 (0, λχ−1 ) − λχ−1 (ll )l · Ln (0, λχ−1 ) .
(8.1.35)
As computed in Sect. 8.1.5, the Euler l-factor of Lk (s, χ−1 λ) for k ≥ f is given
by
k−f
(χ−1 λ(l)l)j
if f > 0,
j=0
l2sj
(8.1.36)
−1
(χ−1 λ(l)l)j
k−1 l − M/Q
l (χ λ(l)l)k
2sj
+ 1+2ks
L0l (s, χ−1 λ) if f = 0,
j=0
l l
M/Q
where l is 1, −1, or 0 depending on whether l splits, remains prime,
or ramifies in M/Q, respectively, and L0l (s, χ−1 λ) is the l-Euler factor of the
primitive L-function L0 (s, χ−1 λ). We define a possibly imprimitive L-function
χ−1 λ(L)
L(l) (s, χ−1 λ) = L0l (s, χ−1 λ)−1 L0 (s, χ−1 λ) = L(s, χ−1 λ) (1 − )
N (L)
L|l
λχ−1 (e(F) ll )
χ(a)E (λ)([a]) = c0 · L(0, χ−1 λ) if f > 0, (8.1.37)
el
[a]∈Cln+1
λχ−1 (e(F) )
χ(a)E(λ)([a]) = c0 · L(l) (0, χ−1 λ) if f ≥ 0. (8.1.38)
e
[a]∈Cln+1
Note that L(0, χ−1 λ) = L(l) (0, χ−1 λ) if f > 0. The right-hand side of (8.1.37)
when f = 0 (hrnvr, n = 1) is explicitly computed in [H04b] (4.23) and the
l-Euler factor of L(s, χλ−1 ) at s = 1 shows up in front of L(0, χ−1 λ) with a
p-adic unit factor. Following [Sh6], we have proven that all these values are
algebraic in Q and actually p-integral over W (most of the cases by Theo-
rem 1.34):
κ
π Γ (k+κ)
Theorem 8.18 Let c0 = (−1)k Im(δ) √
κ DΩ k+2κ
for integers k > 1 and κ ≥ 0.
∞
Then the L-value of (8.1.38) is an algebraic p-integer in W if either p ≥ 5 or
f > 0 or k = 2.
Recall that we have modified W, adding some p-power roots of unity to
the strict henselization of Z(p) given by i−1 p (W (Fp )) (so that any finite or-
der character of Cl∞ has values in W × ). As we have seen in Sect. 6.3.4, x(a)
is defined over W[μln ], but W ⊃ μln (Q), as it contains a strict henseliza-
tion of Z(p) ; so the result is clear from the formula (8.1.38) and Theo-
rems 1.27 and 1.34 if dκ E(λ) has q-expansion in W[[q]]. Thus, we need to
show that dκ E(λ) is W-integral. If k = 2 or f > 0, by the explicit q-
expansion, dκ E(λ) has W-integral coefficients except for the constant term.
If κ > 0 and k = 2, the constant term vanishes; hence, the integrality fol-
lows. Even if κ = 0, if f > 0, E (λ) does not have constant term and W-
integral; so by (8.1.37), the integrality follows. Thus, we may assume that
k = 2 and f = 0. Then we have E2 (1) = −Δ−1 δ12 Δ/24 by Lemma 8.8.
Recall E(λ) = −dκ (Δ−1 dΔ − Δ−1 dΔ|(l)|[l])/24. If p ≥ 5, Δ−1 and Δ/24
are in W [[q]], and hence E(λ) is W -integral if p ≥ 5. By Theorems 1.27
and 1.34 applied to dκ (Δ−1 dΔ)/24 = − m+n=κ m κ
dm Δ−1 dn+1 Δ/24 and
κ −1
κ
m −1 n+1
δ2 (Δ δ12 Δ/24) = m+n=κ m δ−12 Δ δ12 Δ/24 [cf. Exercise 1.26(1)] and
their translates by [l] ◦ (l), we have E(λ)([a]) ∈ W ∩ Q = W.
for a proper On+m -ideal a ⊃ On with (aac ) = l−2m , and for some u ∈ Z×
l , we
have
u
1 lm
x(a) = x(On )/C = x(Om+n )| l . (8.2.1)
0 1
u
1 lm
See Sect. 7.1.6 for the action of g = l on the point x(Om+n ).
0 1
Proof. We only need to prove (8.2.1). The basis of On,l is given by αn t (1, w2 )
for αn = 10 l0nl with a prime element ll of Zl . By Remark 7.12, the action on
level structure η → η ◦ g induces the action L for Z-lattices.
→ g −1 L Thus, we
−1 −1
find that αn (x(O)) = x(On ) and α1 (x(On−1 )) = x(On ). Since the general
case of m > 1 follows by iteration of the formula in the case m = 1, we suppose
m = 1. Then the formula becomes, for a suitable u ∈ Z× l ,
u
ll−1 (x(a)) = x(la) = x(On+1 )| 0 l1l
1
(8.2.2)
if x(a) = x(On )/C for C as above. To see this, note that the basis of ll al is
given by
354 8 Nonvanishing Modulo p of Hecke L-Values
1−ln uw
l 2 1 −u 1
ln+1 w2
= ll αn+1 w2 .
l 0 1
1−uln
l w2
Thus, al/On is generated by llmod On,l, which gives the subgroup C
for a suitable choice of u. Since l ∈ Z × , the action of ll is equivalent to the
(l)
1
−1 f |U (l)(x(a)) =
lκ a · f ([a]) = λ(a) f ([au ]). (8.2.4)
λ(l)l u mod l
8.2 Nonvanishing Modulo p of L-Values 355
We need to have the term φ(a−1 )f ([a]) instead of φ(a)f ([a]) because in our
application, f ([a]) for a specific f actually gives the partial L-value for the
ideal class [a−1 ], not [a].
Exercise 8.22 Prove that Cl∞ φdϕf depends only on φ and not on the choice
of n, so that φ factors through Cln (hint: use (8.2.4)).
We choose a complete representative set {R−1 |r ∈ R} for Δalg such that the
set R is a subset of the set of all square-free products of primes in Q ramifying
in M/Q, and R is a unique ideal in M with R2 = r. The set {R|r ∈ R} is a
complete representative set for 2-torsion elements in the class group Cl0 of O
(i.e., the ambiguous classes). We fix a character ν : Δ → F× , and define
8.2 Nonvanishing Modulo p of L-Values 357
fν = −1 (R)f |[r].
λν (8.2.6)
r∈R
Proof. Write Γf,n for the image of Γ in Cln . For proper On -ideal a, by the
above definition of these operators,
−1 f (x(Q−1 R−1 a), ω(Q−1 R−1 a)).
f |[r]|[q]([a]) = λ(a)
For sufficiently large n, χ factors through Cln . Since χ = ν on Δ, we have
χdϕνf = −1 (QRa)f |[r]|[q]([a])
λχ
Γ Q∈Q r∈R a∈Γf,n
= χ(QRa)f ([Q−1 R−1 a]) = χdϕf ,
a,Q,r Cl∞
−1
because Cl∞ = Q,R [Q R−1 ]Γ .
In the next couple of sections, we prove the following result (given in [H04b]
as Theorems 3.2 and 3.3):
Theorem 8.25 Suppose that p = pp with two primes p = p in M/Q (the
factor p corresponding to ip : M → Qp ). Let f = 0 be an eigenform of
U (l) of level Γ0 (l) defined over F with nonzero eigenvalue. Fix a character
ν : Δ → F× , and define fν as in (8.2.6). If f satisfies the following condition:
(H) For any given integer r > 0 and any congruence class u ∈ (Z/lr Z)× , there
exists 0 ≤ ξ ∈ u such that a(ξ, fν ) = 0,
358 8 Nonvanishing Modulo p of Hecke L-Values
then nonvanishing Cl∞ νχdϕf = 0 holds except for finitely many characters
χ of Γ . Here νχ is the character of Cl∞ = Γ ×Δ given by νχ(γ, δ) = ν(δ)χ(γ)
for γ ∈ Γ and δ ∈ Δ.
where [a] runs over all classes in Cln that project down to [A] = [lm−n A] ∈
Clm . We suppose that f |U (l) = (a/λ(l)l)f with a unit a ∈ A. For each
function φ : Cl∞ → A factoring through Clm , we define
φdϕf = a−m φ(a−1 )f ([a]). (8.2.8)
Cl∞ a∈Clm
Writing Fpr for V/P ∩ V, any modular form defined over Fpr is a reduction
modulo P of a classical modular form defined over V of sufficiently high weight.
Thus, the above identity is valid for σ = Φs (s ∈ Z) for the Frobenius element
Φ ∈ Gal(F/Fpr ). In this case, N (b) is a power of a prime ideal p|p in M .
We now take the base ring B to be F = W/mW = Ṽ/P and consider the
measure ϕf as having values in F. Thus, we simply write g ∈ Gk (Γ0 (l); F) for
g mod P for the original g ∈ Gk (Γ0 (l); V) [and f = dκ g ∈ VΓ1 (l)/F ]. Then
(8.2.10) shows that if φ is a character χ of Cl∞ , for σ ∈ Gal(F/Fpr ),
χ(x)dϕf (x) = 0 ⇐⇒ σ ◦ χ(x)dϕf (x) = 0. (8.2.11)
Cl∞ Cl∞
d
TrFq (χ)/Fq ◦ χ(y −1 x)dϕf (x) = n χ−1 (y −1 a)f ([a]),
Cl∞ a −1 −1
a∈Cln :χ (ay )∈Fq
(8.2.12)
because, by Exercise 1.15, for an lth-power root of unity ζ ∈ μln − μl ,
s ln−m ζ s if ζ s ∈ Fq and Fq ∩ μl∞ (F) = μlm (F)
TrFq (μln )/Fq (ζ ) =
0 otherwise.
Suppose Cl∞ χ(x)dϕf (x) = 0 for an infinite set X of characters χ. For
m
sufficiently large m, we always find a character χ ∈ X such that Ker(χ) ⊂ Γ l .
n
Then, writing Ker(χ) = Γ l for n ≥ m, we have the vanishing from (8.2.11)
σ ◦ χdϕf = 0 for all σ ∈ Gal(F/Fq ).
Cl∞
We write Fp [f ] for the minimal field of definition of f ∈ V (F) [i.e., the field
generated by a(ξ, f ) ∈ F for all 0 ≤ ξ ∈ Z]. Similarly, Fp [λ] (resp., Fp [ν])
is the subfield of F generated by the values λ([a]) mod P (resp., ν([a])) for
all [a] ∈ C . Define Fp [f, λ, ν] by the composite of these fields and Fp [μl ].
alg
α(t) = ( 10 1t ) . (8.2.14)
Actually, a → u mod lr gives a bijection of yχ−1 (μlr ) onto O/lr . We write the
element a corresponding to u as α( lur )a0 . This shows, choosing a primitive lr th
l
root of unity ζ = exp(2πi/lr ) and ay ∈ yχ−1 (μlr ) so that χ−1 (α( lur )ay ) = ζ uv
l
for an integer 0 < v < lr prime to l (independent of y), the inner sum of
(8.2.13) is equal to, up to a nonzero factor,
u
ζ uv (fν |α( r ))([ay Q−1 ][Q]Γ ).
l l
u mod l
r
Since SL2 (A(p∞) ) keeps any geometrically irreducible component (see Theo-
rem 7.18), x(a) as above always resides in one component V .
Let F = FΞ denote the F-algebra of functions φ : Ξ → P1 (F) = F {∞}
with |φ−1 (0)| < ∞ and |φ−1 (∞)| < ∞. The profinite class group C = Cn1 :=
Ker(Cl∞ → Cln1 ) acts on F by translation: f (x) → f (xy) (y ∈ Cl∞ ). In
particular, α ∈ O(p) with trivial [(α)] ∈ Cln1 acts on Ξ and such α is p-
adically dense in TM (Zp ). For f ∈ F(V )× , for each x = x(a) ∈ Ξ, expanding
f into a Laurent series f (t) = n an tn ∈ F[[t]][t−1 ] with leading nonzero term
am tm (m ∈ Z), we may define
⎧
⎪
⎨∞ if m < 0,
f (x(a)) = a0 if m = 0,
⎪
⎩
0 if m > 0.
(p)
By the Zariski density of Ξ in V = V/F , we can embed into F the function
field F(V ) of V .
Exercise 8.27 Why is Ξ Zariski dense in V ? Why does density imply the
injectivity of F(V ) into F ?
Proposition 8.28 Take a finite set Δ = {γ1 , . . . , γm } ⊂ Cn1 injecting
into Cl∞ /Clalg . Then the subset Ξ̃ := {(x(δ(a))δ∈Δ |x(a) ∈ Ξ} is Zariski
Δ
dense in the product V/F of Δ copies of V/F . This implies that the fields
γ1 (F(V )), . . . , γm (F(V )) are linearly disjoint over F in FΞ , where γ(F(V ))
is the image of F(V ) ⊂ F under the action of γ ∈ Cn1 . In other words, we
m
!
have injectivity of the map γ1 ⊗ · · · ⊗ γm : OV ⊗F OV ⊗F· · · ⊗F OV → F
sending f1 ⊗ · · · ⊗ fm to an element in F given by x(a) → j fj (x(γj a)).
Proof. We write xm (a) for (x(γj a))j ∈ V m . Let X be the Zariski closure of
Ξ̃ := {xm (a) ∈ V/Fm
|x(a) ∈ Ξ} in V m . The last two assertions follow from
the density of the above set: X = V m . Thus, we only prove X = V m . We
have a homomorphism π : O× (pl) → Cl∞ by sending α to its class [α]. Let T =
π −1 (Cn1 ). We have α ∈ T ⇔ (α) ∩ On1 ,l is a principal On1 -ideal (necessarily
generated by one of ±α), where (α) = αO. Note that [α] = limn [(α) ∩ On,l]n
←−
for the class [(α) ∩ On,l]n in Cln . We only need to deal with Y1 (l) as we
assumed that the level is just l, but we take care of a slightly more general
8.2 Nonvanishing Modulo p of L-Values 363
for proper On -ideals a. We have ((α) ∩ On )−1 in the right end, because of
Remark 7.12 [i.e., on the lattice, the action of g ∈ G(A(∞) ) is given by
→ g −1 L].
L A tricky point is that for a principal ideal (α) of On1 , (α) ∩ On,l
is no longer the principal On -ideal if n n1 . By (8.2.15), the diagonal auto-
morphism τ (ρ(p) (α))Δ ∈ Aut(V/F Δ
) leaves Ξ̃ stable as multiplication by δ ∈ Δ
commutes with multiplication by [α] ∈ Cn1 . Thus, the closure X is stable
under T . Since X is a union of finitely many irreducible components, the
group T permutes them; so, replacing T by a subgroup of finite index, we
may assume that T fixes all of the irreducible components. Replacing X
by an irreducible component, we may assume that X is irreducible of pos-
itive dimension. Since dim X > 0, we have that Ξ̃X = X ∩ Ξ̃ has infinitely
many densely populated distinct points (and stable under T ). Thus, we could
restart with ΞX = {x ∈ Ξ|(δ(x))δ∈Δ ∈ Ξ̃X }. Again, moving by the action
of Cl∞ , we may assume that X contain x(On ). Note that On,p = Op as
l = p. Thus, we have the canonical coordinate t around x(O) (by choos-
ing a level-p∞ structure ηp◦ × ηpet ). Then the action of τ (ρ(p) (α)) is given
1−c
by t → tα (see Proposition 7.44). In particular, the action is p-adically
continuous on the formal completion V Δ along (δ(x(On ))δ∈Δ ∈ V Δ . Thus,
the p-adic closure T of {τ (ρ(p) (α))Δ |α ∈ T } in Aut(V Δ ) leaves stable X.
Identifying V with G m by t, the action of T contains the diagonal action of
an open subgroup U of AutGSCH (G m ) = Z× . Then if dim X < dim V Δ ,
p
by Theorem 3.44 and Corollary 11.11 (or by the theorem in Sect. 3.5.3),
X ⊂ V × V × · · · × V × Δα,β after permutation. Thus, we have two dis-
tinct elements δ, γ ∈ Δ such that δγ −1 = τ (ρ(p) (αβ −1 )) in Aut(V/F ). By
Theorem 7.20, this means δγ −1 = αβ −1 modulo Q× . Since (n) for n ∈ Q× is
trivial in Cl∞ , we have δγ −1 = [(αβ −1 )] ∈ Clalg , which is a contradiction.
We apply Theorem 8.25 to the Eisenstein series E(λ) in (8.1.17) for the Hecke
character λ fixed in Sect. 8.1.2. Choosing a generator π of mW , the exact
l
sequence ω k/W − → ω k/W ωk/F induces a reduction map: H 0 (M, ω k/W ) →
H 0 (M, ω k/F ). We write Ek (ψ ◦ ) mod mW for the image of the Eisenstein series
Ek (ψ ◦ ). Then we put
(−1)k (O× : Z× )
√ .
Im(δ)κ D
The nonvanishing of C modulo mW follows from the unramifiedness of p in
M/Q. We now prove the following a slightly more specific version of Theo-
rem 8.17, assuming C = 1:
Theorem 8.31 Let p be an odd prime splitting in M/Q. Let λ be a Hecke
character of M of conductor C and of infinity type k + κ(1 − c). Suppose
×
(ct), (sp), and (pl) in Sect. 8.1.2. Fix a character ν : Δ → Q . Then for
π κ ΓΣ (k+κ)L(l) (0,ν −1 χ−1 λ)
the Néron period Ω∞ in (8.1.2), we have Ω∞k+2κ ∈ W for
all nontrivial characters χ : Cl∞ → μl∞ (Q) factoring through Γ . Moreover,
except for finitely many characters χ in HomGP (Γ, μl∞ ), we have
holds for
f = (Ek (ψ ◦ ) − N (l)Ek (ψ ◦ )|l|[l]),
because l p. Up to nonzero constant, ψ ◦ (a, b) in (8.1.12) is equal to φ(a, b) =
λF (a)λ−1 ×
Fc (b) for (a, b) ∈ (Z/f) , so by our simplifying assumption that φ is a
constant function 1 with value 1. Thus, we are going to prove
Exercise 8.32 Give a detailed proof showing 1+ξ k−1 ≡ 0 mod p for infinity
of primes ξ in a given class u ∈ (Z/lr )× .
Remark 8.33 If we start with a CM field M in place of an imaginary
quadratic field, a similar theorem holds, but there exist well-specified excep-
tional cases where the canonical integral part of L(0, χ−1 λ) vanishes modulo
mW for all χ. See details in [H04b] and [H07a] (M1–3).
9
p-Adic Hecke L-Functions and Their
μ-Invariants
Recall N (f) = |Z/f| = [Z : f], which is the positive generator of the ideal f. We
put Xα = (Z/pα Z) × (Z/f) and define the partial Fourier transform
"
+
P Φ : (Qp /Zp ) × (f−1 /Z) × Y = p−α f−1 /Z × Y → C
α
where eQ is the standard additive character of A/Q (as in Exercise 8.6) re-
stricted to the local component Qpf at pf.
Exercise 9.1 Show that the definition of P Φ as above does not depend on the
choice of α.
By Pontryagin duality under x, yL as in (8.1.6), the dual map i of i gives
rise to
i : P V (L) Zp × (Z/f).
We regard P Φ as a function on p−∞ f−1 L∩P V (L)= α p−α f−1 L ∩P V (L) by
P Φ(i−1 (w), i (w)) if (w mod L) ∈ Im(i),
P Φ(w) = (9.1.3)
0 otherwise.
Writing L = (L, φpf ) for simplicity, we define, similarly to (8.1.10), the value
Ek (L; Φ) by
9.1 Eisenstein and Katz Measure 369
P Φ(w)
Ek (L; Φ) = (−1)k Γ (k + s) .
wk |w|2s s=0
w∈p−∞ f−2 L/Z×
k
N (n)
Φ(n, 0) |N (n)|−s .
|N (n)|
n∈Z/{±1}×
Proof. For simplicity, assume f = 1 (see [HT2] for the general cases). Let
Lz := (Lz , iz ) for Lz = (2πi)(Z+Zz) with z ∈ H and iz : p−α Z/Z → Q·Lz /Lz
given by iz (a mod Z) = (2πi)a mod Lz . Then we have
P Φ(a, b)
Ek (z; Φ) := Ek (Lz ; Φ) = c(k, s) ,
(a + bz)k |a + bz|2s s=0
(a,b)∈(p−α Z×Z)/Z×
k
where c(k, s) := (−1) Γ (k+s)
(2πi)k (2π)s . Put c = c(k, 0). We can separate the sum for
(a, 0) and (a, b) with b = 0:
P Φ(a, 0)
c−1 Ek (z; Φ) =
ak |a|2s s=0
0<a∈p−α Z
1
+ P Φ(a0 , b) .
(a + a0 + bz)k |a + a0 + bz|2s s=0
0<b∈Z a0 ∈(p−α Z/Z) a∈Z
Put 1
Sk (z; s) = .
(a + z)k |a + z|2s
a∈Z
By Poisson’s summation formula (cf. [EDM] Theorem 2.3 or [LFE] Sect. 8.4),
we get
Sk (z; s) = Ck (a, z; s)
a∈Z
370 9 p-Adic Hecke L-Functions and Their μ-Invariants
∞ eQ (−xt)
for Ck (x, z; s) = −∞ (t+z)k |t+z|2s
dt. We verify from the definition that
P Φ(a, 0)
c−1 Ek (z; Φ) =
ak |a|2s s=0
0<a∈p−α Z
+ { P Φ(a0 , b)eQ (−aa0 )}Ck (a, bz; 0).
0<b∈Z a∈Z a0 ∈p−α Z/Z
We consider the space O 8 = (Zp × (Z/f)) × (Zp × (Z/s)) and write the
variable on O8 as (x, a; y, b) for x, y ∈ Zp and a ∈ Z/f and b ∈ Z/s. We regard
8
O as a ring; then O8 × is the group of invertible elements in O.8 We embed Z×
8 8 × ×
into O diagonally. We also write T = (Zp × (Z/f)) × (Zp × (Z/s)). We let
( ∈ Z× act on O 8 by ((x, a; y, b) = ((x, (a; (y, (b) and define T = T8/Z× and
T× = O 8 × /Z× . These are the profinite compact spaces carrying the Eisenstein
measure.
9.1 Eisenstein and Katz Measure 371
This is the property required to define an Eisenstein series (for even weight k)
in Sect. 8.1.3. Then there exists a unique measure E : C(T, W ) → VΓ11 (fs)/W
with the following two properties:
(1) If Φ has values in Q equipped with discrete topology, then for each positive
integer k > 0,
E(N −k Φ) = Ek (Φ◦ ),
where N : T8 → Z× p is given by N (x, a; y, b) = x. Note here that N
−k ◦
Φ
◦
(for any positive integer k) factors through T ⇐⇒ Φ andsatisfies in-
variance (9.1.1) under Z× required for the definition of the Eisenstein
series.
(2) The q-expansion of E(Φ) at the infinity cusp is given by
qn Φ◦ (a; b)|a|−1 .
0<n∈Z (a,b)∈(Z×Z)/Z× ,ab=n
The verification of the existence and uniqueness of the measure satisfying (1–
2) is a consequence of the q-expansion principle (see Corollary 7.38) and the
q-expansion of the classical Eisenstein series given in Lemma 9.2. We write
E(Φ) for E0 (Φ).
of elliptic curves (see Sect. 6.3), E(A)/W sits at the origin 1 ∈ G m of the
deformation space G m of E(A)/F . Thus, we can uniquely lift η ◦ to a level-p∞
0
structure η ◦ (A); so we may assume that φpf (A) contains a level-p∞ structure
defined over W .
We recall the construction of the Katz measure interpolating the L-values
of arithmetic Hecke characters of conductor dividing Cp∞ , where C is an
integral ideal of M prime to p. We write O for the integer ring of M . We
decompose C = FFc I. Here I consists of inert or ramified primes over Q, FFc
consists of split primes over Q, and
F + Fc = F + Fc = Fc + Fcc = O and Fc ⊃ Fc .
We put f = FI ∩ Q, s = Fc ∩ Q and i = I ∩ Q. We have Op = O ⊗Z Zp =
Op × Op [so that E(O)[p∞ ]◦ = E(O)[p∞ ] over W ]. We recall the following
four conditions:
1. The lattice A is a fractional ideal of M prime to Cp; so we write E = E(A).
2. Choose δ ∈ M so that δ c = −δ, and the imaginary part Im(i∞ (δ)) is
positive and the alternating
√ form u, v = (uc v − uv c )/2δ induces O ∧ O ∼=
Z (in short, 2δ = −D for the discriminant D > 0 of M ).
3. The inclusion Q → M induces a canonical isomorphism Zp ∼ = Op , which in
(∗)
∼ Qp /Zp ∼
turn induces η ◦ (A) : μp∞ = = Mp /AOp ⊂ C/i∞ (A) = E(A)[p∞ ].
◦
We put ip (A)(x) = η (A)(2δx). This ip (A) is the p-part of the level struc-
ture x(A). Here (∗) is given by e( pmn ) ↔ pmn .
4. The prime-to-p part φ(p) = φf sends (x mod Z) to (x mod A), which
induces (f2 )−1 /Z → (Fi)−2 A/A.
Let ClM (i) be the ideal class group of M modulo i and let Cl− (i) be the
quotient of ClM (i) by the image of (Z/i)× . Identifying Op (resp., Op ) with
the first (resp., last) component of Zp of O 8 and embedding Z/f into O/FI
resp., identifying Z/s with O/Fc)] through the inclusion Z → O, we bring T ×
into Z = ClM (Cp∞ ) = limn ClM (Cpn ). Then we have the exact sequence
←−
T× −
→ Z → Cl− (i) → 1,
ι
9.1 Eisenstein and Katz Measure 373
and the kernel of ι is a finite group. We write [A] for the image of the class
of an ideal A prime to Cp in Z. For α ∈ O, we have [(α)] = α−1 , where the
right-hand side is the image of the inclusion O×
Cp → Z. Choosing a complete
−
representative set {A} for Cl (i), we have a decomposition
4
Z= Im(ι)[A]−1 .
A
λ(x∞ ) = xk+κ(1−c)
∞ ,
where k and κ are integers. Then λ has values in Q on the finite part
: M ×(∞) /M × → Q× defined by λ(x)
MA×(∞) of MA× . Moreover, the map λ =
A p
k+κ(1−c)
λ(x)xp is a well-defined continuous character, which is called the p-adic
avatar of λ. We can associate λ with its dual λ∗ given by λ∗ (x) = λ(xc )−1 |x|A .
Then the p-adic avatar of λ∗ is given by λ ∗ (x) = λ(x
c )−1 N (x)−1 for the cy-
clotomic character N . We define a local Gauss sum of λ at prime ideals Q
dividing the conductor of λ by
−e −e −1
G(d, λQ ) = λ(Q ) λQ (u)eM (Q dQ u), (9.1.8)
u∈(OQ /Qe )×
for all Hecke characters λ modulo Cp∞ such that (i) its conductor is divisible
by all prime factors of F, (ii) the infinity type of λ is k + κ(1 − c) for integers
k and κ satisfying either k > 0 and κ ≥ 0 or k ≤ 1 and κ ≥ 1 − k.
An element a ∈ M is considered to be an element of C via i∞ . The L-function
in (9.1.10) is always the primitive one associated with the primitive Hecke
character. We also tacitly agree to put λ(Q) = 0 if Q divides the conductor
of λ.
Remark 9.5 Since Hecke L-functions satisfy functional equation as in [LFE]
Chap. 8, we can deduce a p-adic version from the Archimedean functional
equation. Denoting the measure μ for Z(Cc ) by μc , we have the following
functional equation:
λdμ = W (λ) ∗ dμc
λ
Z(C) Z(Cc )
−1
ΦA (x, y) := λ(A)−1 λ−1
p (x)λp (y).
By the above exercise, noting that we have twisted η ◦ (A) slightly to define
ip (A)(x) := η ◦ (A)(2δx) (to be in line with the existing literature on p-adic
L-functions), we have, for x ∈ p− max(e(p),1) A,
Wp (λ)λ−1 −1
p (x)λp (y) if e(p) > 0 and xpe(p) Op = Op ,
P Φ◦A (x, y) = −1
(1p (x) − p 1p (p x))λp (y) if e(p) = 0,
1
λ−1 e −1 e
p (p α)λp (p α)
= λ(A)−1 λp (pe )
e α∈(B∩O× )/O×
αk+κ(1−c) NM/Q (α)s s=0
$p p
λ−1 −1
p (α)λp (α)
α
→α
e
$p
−1
= λ(A) λp (pe )
(p−e α)k+κ(1−c) NM/Q (p−e α)s s=0
α∈(B∩O×
p )/O
×
λ−1 −1
p (α)λp (α)
= λ(A)−1 λp (pe )p
e(k+κ(1−c))
αk+κ(1−c) NM/Q (p−e α)s s=0
α∈(B∩O×
p )/O
×
λ((α))
λ(B)−1 = LB−1 (0, λ ),
Exercise 9.7
=
NM/Q (α)s s=0
α∈(B∩O×
p /O
×
(9.1.12)
where B = Ap−e (pe ), which is prime to p, and λ is equal to λ if e(p) > 0 but
is an imprimitive character modulo p induced by λ if e(p) = 0 [so λ (p) = 0
in the imprimitive case]. Summing up over the ideal classes B, we get
Z(C) λdμ
= c0 (O× : Z× )Wp (λ)(1 − λ(pc ))L(0, λ). (9.1.13)
Ωpk+2κ
We now assume that e(p) = 0. Then in the same manner as above, we get
k+κ(1−c) λ−1 −1
p (p α)λp (p α)
λp (p )p
= λp (p α)
pλ(A)λp (p ) (p α)k+κ(1−c) NM/Q (α)s s=0
α∈(p−1 A∩O×
p
))/O×
k+κ(1−c) λ−1 (α)
λp (p )p p
=
pλ(A)λp (p ) αk+κ(1−c) NM/Q (p−1 α)s s=0
α∈(B∩O×
p
)/O×
λ((α))
λ(B)−1 λ∗ (pc ) = λ∗ (pc )LB−1 (0, λ ),
Exercise 9.7
=
NM/Q (α)s s=0
α∈(B∩O×
p
)/O×
where B = Ap−1 (p ). Summing up over ideal classes the difference between
the first and second sums, we get
Z(C) λdμ
= c0 (O× : Z× )Wp (λ)(1 − λ(pc ))(1 − λ∗ (pc ))L(0, λ). (9.1.15)
Ωpk+2κ
This finishes the proof.
Exercise 9.7 In the above computation (9.1.12), prove that
Lemma 9.8 We have μ(ϕ) = 0 if lim inf μ(π∗ (( ∗ ϕ)) = 0, where ( runs over
all finite order characters of H.
Actually, we can prove the converse of the lemma, but we do not need it.
Exercise 9.9 Give a detailed proof of the above lemma.
We apply the splitting techniques in Sect. 8.2.2 to the Katz measure to
compute the q-expansion
of the corresponding Eisenstein series. We consider
the measure EA : φ → T φdE(x(A)) for x(A) = E(A), i(A) [on the image of
T × in ClM (p∞ )].
Lemma 9.10 For α ∈ M prime to p, we have
φ(t)dE(ρ(α)−1 (x(A))) = φ(αt)dEA = φ(t)dEαA , (9.2.1)
T× T× T×
Thus, we have
φA (t)dEA (t) = 1H[B−1 A] φA (t)dEA (t)
T× B T×
= 1H (t[BA−1 ])φB (t[BA−1 ])dEA (t) = 1H (t)φB (t)dEB (t),
B T× B T
1 → O× ×
p /O → Z → ClM → 1,
where O× ×
p [−1] = {a ∈ Op |c(a) = a
−1
}. In particular, identifying Op with Zp ,
for a principal ideal (α) prime to p, [(α)]− = (αc−1 )p ∈ T (Zp ) = O× ×
p = Zp .
c−1 1−c
Here as an element of T (Zp ), having (α )p [instead of (α )p ] is correct
because in Z = MA× /M × (O (p) )× M × , the diagonal class of β = α1−c is trivial,
∞
and so the principal ideal (β) corresponds to the ideal class of β (p∞) with
βp β (p∞) = β = 1 in Z [i.e., βp−1 = β (p∞) ]. Therefore, we have, with T (Z(p) ) ⊂
T (Zp ) = Z× p by (7.2.5),
+
[A]− ∈ T (Z(p) ) ⇐⇒ [A] ∈ {[(α)]|α ∈ O× (p) } · Γ × Δ
+
⊂ Z, (9.2.3)
for the intersection Γ of ΓM and the image of O×p . We would like to choose D
so that our computation of q-expansion (of Eisenstein series) becomes easier.
Let I(p) be the group of fractional ideals of M prime to p, and define
I(p)+ = A ∈ I(p)A1−c = α1−c O for α ∈ M × .
+ +
Exercise 9.12 Prove that ClM = H 0 (Gal(M/F ), ClM ) and that ClM is
amb
equal to the ambiguous class group ClM .
Thus, if, furthermore, the class number of M is odd, ClM +
= {1}. We take
− + + +
a complete representative set D (resp., D ) for ClM /ClM [resp., ClM in
I(Cp) ].
+
We make use of degeneracy operator. As seen in Sect. 8.2.2, for each split
prime q = QQ or a square-free product Q of ramified primes with Q2 = q,
we have a unique level-Γ0 (q) structure on E(A) given by C = E(A)[Q]. Thus,
for the degeneration operator [q],
g(E, η) = (E, ηg = η ◦ g)
for the image of a test object (A, η) under the action of g. Here, writing
T (E) = limN E[N ] for the Tate module, recall the level structure on E given
←−
by η : (A(∞) )2 ∼ = T (E) ⊗Q A(∞) , where (A(∞) )2 is made up of column vectors
(∞)
on which G(A ) acts from the left.
On the lattice side, η → η ◦ g corresponds to the action L = η −1 (T (E)) →
= (η ◦ g)−1 (T (E)). Suppose g = z ∈ Z(Zp ). The p-part of the level
g −1 L
structure ηp is given by i : Qp /Zp ∼ = P V (L)/Lp and its dual i : P V (L) → Zp
in (pv1–2). This (i, i ) is sent to (i ◦ z, z −1 i ) as i is the dual of i, or, equiva-
lently, (i−1 , i ) is sent to (z −1 i−1 , z −1 i ). The latter expression indicates that
this is the action of the central element z. Then we sum up the terms in-
volving (P Φ)(z −1 i−1 (w), z −1 i (w)) over P V (L) ∩ Q · L to define the Eisen-
stein series. We have (P Φ)(z −1 i−1 (w), z −1 i (w)) = P (Φ|z)(i−1 (w), i (w)) for
Φ|z(x, y) = Φ(zx, z −1 y). Thus, we get the next result.
Lemma 9.13 If z ∈ Z(Zp ), then E(Φ◦ )(E, η ◦ z) = E(Φ◦ |z −1 )(E, η). In
particular,
Φ(za, z −1 b) ab
E(Φ◦ )|z(q) = q .
|a|
(a,b)∈Z×Z,ab>0
Proof. The action of the degeneracy operator [l] is equal to the action of g
is shown in Sect. 7.1.9. It corresponds to the action of h := J1 g −1 J1−1 on the
lattice side. Thus, we see that P Φ|hL = P (Φ ) for Φ (x, y) = Φ(ll−1 x, y). From
⇔ x ∈ lZ
this, the result follows, as ll−1 x ∈ Z and the multiplication by ll
does not affect Φ as l is prime to the level.
since s = βp ∈ S ⊂ O× −
p (for β ∈ D ). We further sum up over D :
+
ψ(α)E(Φ◦+ )|[ααc ]|ρ(α)−1 = E(Φ◦ ), (9.2.10)
α∈D −
∼ Z/((α) ∩ Q)
where we have chosen (α) to be an integral ideal, with O/(α) =
and (α) ∩ Q = (ααc ), and Φ◦ is given from (9.2.8) by
(9.2.8)
Φ◦ (x; y) = −1 Φ◦ (rx; r−c y)) (∗)
ψ(r) = −1 Φ◦ (r(x; y)), (9.2.11)
ψ(r)
+ +
r∈R r∈R
of (9.2.12) with
(∗∗) follows from the fact that 1Γ ◦ (αβ −1 ) is the characteristic function of
(α−1 β)Γ .
◦
Let F = Inf ψ ψ| Γ · (x−1 y)κ 1Γ |Z = (xy)κ Φ◦ . By the compu-
0
tation given in the proof of Theorem 9.4 and by Theorem 1.34, the partial
L-value for the character ψ (κ) and for the ideal class of A is given by
and
dκ E(E(A)) = E(F )(E(A)) (9.2.14)
for all κ ≥ 0.
−
Now we apply the operator [BBc ] and make a variable change: t → t[B]
in (9.2.14). We may assume that [B]− ∈ Γ , since B without this property
does not contribute to the restriction ϕ−
ψ |Γ . By the computation given in the
proof of Theorem 9.4 [or [HT1] (4.9)], the partial L-value for the character ψκ
and for the ideal class of B−1 is given by, for E(F ),
−
ψ (κ) (B)E(F )(E(B−1 )) = ψ (κ) (B)dκ (E(Φ◦ )|[BBc ](t[B] ))(E(O))
and
dκ E(E(O)) = ψ (κ) (B)E(F )(E(B−1 )) (9.2.15)
B∈D− ,[B]− ∈Γ
Combined with the argument proving Theorem 9.4 [or [HT1] (4.9)], this shows
that the function in the theorem, after applying dκ and evaluating at E(O),
gives rise to the value (9.1.10) for the character λ = ψ(κ) for all κ; so, by
Mahler’s theorem (Theorem 1.36), the measure corresponding the t-expansion
of E is equal to the anticyclotomic Katz measure ϕ− ψ . Since t-expansion is the
Taylor expansion with respect to d, the first assertion follows.
The Igusa tower IgΓ1 (N )/F (p N ) is given by IsomY1 (N )ord /F (μp∞ , E[p∞ ]◦ )
for the universal elliptic curve E over the ordinary locus Y1 (N )ord . Though
Theorem 3.38 and the theorems in Sect. 3.6.1 are stated for global sections
386 9 p-Adic Hecke L-Functions and Their μ-Invariants
OY (p) , they are valid for global sections of OIg , because Ig/Y (p) is étale (i.e.,
the completion of corresponding stalks is the same by Lemma 4.7); so the
statement is intact. Thus, by Theorem 3.38, in view of the independence of
[B]− modulo T (Z(p) ) = O× ×
(p) /Z(p) , the sum in (9.2.16) does not vanish modulo
the maximal ideal mW of W if EB |[BBc ](t) for one of B ∈ D− does not vanish
modulo mW . Take the identity class; we may assume that EO (t) = Ek (Φ◦ )
with Φ◦ (x; y) = x−k χ(x)χ (y) for finite order characters χ and χ of p-power
conductor (indeed, our choice was χ = (−1 and χ = 1). The fact that Φ◦ is
a product of characters as above follows from (9.2.9) and (9.2.11). Thus, Φ is
also a product of characters (x; y) → xk χ−1 (x)χ (y). Then by the computation
already done for such Eisenstein series in Lemma 9.2, for primes l prime to p,
a(l, Ek (Φ◦ )) = χ (l) + χ−1 (l)lk−1 . We leave the reader the verification of the
existence of a prime l such that a(l, Ek (Φ◦ )) ≡ 0 mod mW . Thus, μ(ϕ− ψ ) = 0,
as desired, for ψ = ψk, .
A general character ψ can be decomposed as ψ0 (ψx ◦ πx )ψ − for characters
× ×
ψx : Z× p /μp−1 (Zp ) → Qp and ψ
−
: Γ − → Qp as ΓM = πx (ΓM ) × Γ − .
Since μ− − −
ψ = ψ ∗ μψ for ψ = ψ0 ψx , by Lemma 9.8, μ(μψ ) = 0 follows
−
from μ(ψψ ) = 0. It is an easy exercise to find an infinite sequence {(kj , (j )}j
for integers 0 < kj ∈ Z and characters (j : Z× p /Z
×
→ μp∞ (Qp ) such that
ψ = limj→∞ ψkj ,j under the supremum norm · on C(Z, Qp ). Then we
conclude that μ(ϕ− − −
ψ ) = μ(ϕψ ) = limj μ(ϕψk , ) = 0.
j j
Exercise 9.16 Prove the claim about the sequence {(kj , (j )}j .
10
Toric Subschemes in a Split Formal Torus
W
[t, t−1 ] = lim W [t, t−1 ]/(tn − 1) = lim W [T ]/((1 + T )n − 1) = W [[T ]].
←− ←−
n n
n ) of rank r.
where L runs over (finitely many) Zp -direct summands of X∗ (G m
Since Z has finitely many irreducible components, which are permuted by the
action of the stabilizer of Z under the diagonal action of Z×p , each component
is fixed by an open subgroup of the stabilizer. Therefore, we may assume that
Z is integral (i.e., reduced irreducible). In this chapter, we give a full proof
of the result as Theorem 10.6, assuming irreducibility. This theorem follows
from Corollary 10.7.
One might want to expect a similar result over W in place of F. However,
there is an easy counterexample. Take a diagonal torus Δ ⊂ G m × G m over
W ; so Δ = Gm ⊗Zp D for the diagonal D ⊂ X∗ (Gm/W ). Suppose that μp (W )
2
and hence U ⊂ X 8reg . Since the open subschemes U cover Dreg , we have
Dreg ⊂ X 8reg . Pick x ∈ Dreg (k) ⊂ X 8reg ; then R → O ∼ = k[[x1 , . . . , xd ]] (as
X,x
above), which is a desired embedding.
We need the following lemma in [Ch1] (6.2):
Lemma 10.3 Let k be a field of characteristic p > 0 and q = pr be a p-power
(r > 0). Let P (x1 , . . . , xm ) ∈ k[x1 , . . . , xm ] be a polynomial with coefficients in
n n
k. If there exists n0 ≥ 0 such that P (cq1 , . . . , cqm ) = 0 for all integers n ≥ n0
n n
for a vector c = (c1 , . . . , cm ) ∈ k m , we have P (cq1 , . . . , cqm ) = 0 for all n ∈ Z;
in particular, P (c1 , . . . , cm ) = 0.
We give a version of Chai’s proof in [Ch1].
Proof. Replacing k by its algebraic closure if necessary, we may assume that k
−1
is perfect. Thus,
σ : x → xq is an automorphism of k. Write x = (x1 , . . . , xm )
m α
and xα = j=1 xj j for multiindices α = (α1 , . . . , αm ) ∈ Nm . We define the
total degree deg(f ) = max{α1 + α2 + · · · + αm |aα = 0}, and consider the
filtration V0 ⊂ V1 ⊂ · · · ⊂ Va of k[x] given by Va = {f | deg(f ) ≤ a} (a ∈ N).
The dimension of Va over k is finite.
For each polynomial f (x) = α aα xα ∈ k[x], we write
σ n (f ) = σ n (aα )xα ∈ k[x].
α
as desired.
We keep the notation introduced in Lemma 10.3. In particular, q = pr for
r > 0 and n0 is a nonnegative integer. Here is a sort of Chai’s implicit func-
tion theorem for power series in characteristic p-coefficients [as it gives a uni-
formization of the zero set of a power series f (u, v); see also [BCM] III.4.5]:
Again by Lemma 10.3, if either |β| < M2 or |β| = M2 and |α| < M1 , Aα,β (c) =
n
Aα,β (cq ) = 0 for all n ∈ Z.
We have
n n n
f (g(x), h(x)q ) = f (g(x), H(cq ; yq ))|y=x
n n n n
= Aα,β (cq )xα yq β |x=y = Aα,β (cq )xα+q β . (10.1.2)
α,β α,β
n
Because of limn→∞ dq n = 0 in real numbers R, we can find a positive integer
n1 such that q n1 > 2M1 and q n M2 < dn for all n ≥ n1 . Equations (10.1.1)
and (10.1.2) combined, we get, for all n ≥ n1 ,
mod (x)dn n
0 = f (g(x), Φn (h(x))) ≡ f (g(x), h(x)q )
mod (x)M1 +q
n M +1
2 n n
≡ Aα,β (cq )xα+q β . (10.1.3)
|α|≥M1 ,|β|≥M2
Let G and G be a connected linear algebraic group over Qp with Lie algebra
g/Qp and g/Qp , respectively (see Sect. 10.2.4). As a vector space over Qp , g/Qp is
the tangent space of G at the identity element 1G of G. Thus, any morphism
f : G → G of Qp -schemes sending 1G to 1G induces its differential df :
g → g , which is a Qp -linear map (as it is the dual of the pullback map
f ∗ : ΩG /Qp → ΩG/Qp ). In particular, if G = GLV as in Example 4.58 (6)
for a finite-dimensional Qp -vector space V [i.e., G associates the R-linear
automorphism group AutR (V ⊗Qp R) to any commutative Qp -algebra R], we
have g = EndQp (V ) with [x, y] = x ◦ y − y ◦ x. The commutator map (·, ·) :
G × G → G given by (x, y) → (x, y) := xyx−1 y −1 sends the identity of G × G
to the identity to G, inducing the bracket product [·, ·] = d(·, ·) : g × g → g,
which makes g a Lie algebra (see Sect. 10.2 for more details).
Exercise 10.5 Check that the two bracket products defined above coincide if
G = GLV .
Let ρ : G → AutQp (V ) = GLV be a morphism of group schemes over Qp ,
which is called a rational representation of G. The space V is often written as
V (ρ), and V (ρ) could also be called a rational representation of G [identifying
the G-module V (ρ) and the homomorphism ρ]. Then the differential dρ :
g → EndQp (V ) induces a homomorphism of Lie algebras (see Sect. 10.2.4 for
details about the differential of a representation). As remarked already, the
endomorphism algebra glV := EndQp (V ) is naturally a Lie algebra under the
bracket product [x, y] = x ◦ y − y ◦ x. We pick a Zp -lattice L of V and consider
the open subgroup U of G(Qp ) stabilizing L (that is, U · L = L). We write
G(Zp ) for the maximal subgroup of G(Qp ) with this property: G(Zp ) = {u ∈
G(Qp )|u · L = L}.
Let k be a field of characteristic p > 0 and consider the formal group
m/k = lim μpn /k . Put T = L ⊗Z G
G m . Thus, for any pro-Artinian k-algebra
←−n p
Here the word “integral” means reduced and irreducible. We faithfully follow
the proof given in [Ch1] (6.6).
[note that h(y) = g(x)|x=y ]. If this is the case, any power series f1 ∈ P kills
m ◦ (id ×δX )(Z × Z), and hence m ◦ (id ×δX )(Z × Z) ⊂ Z.
For each X ∈ g/Zp and n ≥ n0 , we have exp(pn X) ∈ U and
pin
ρ(exp(pn X)) = idL +dρ(X) dρ(X)i−1 .
i!
i≥1
in
Let Epn X := i≥1 pi! dρ(X)i−1 as an endomorphism of V . By a well-known
estimate (cf. [LFE] Sect. 1.3), we have
i
ordp (pin /i!) ≥ in − ≥0 (10.1.5)
p−1
in
(by n ≥ n0 ≥ 2) and limi→∞ pi! = 0, and hence Epn X takes L into L; so
p
we regard Epn X ∈ EndZp (L). Then εpn X := Epn X ⊗ idG m is an endomorphism
m = Spf(k[[u]]). Thus, εpn X induces a corresponding k-algebra
of T = L ⊗Zp G
endomorphism ε∗pn X : k[[u]] → k[[u]]. Since
pin
εpn X − pn idL = ≡0 mod pdn
i!
i≥2
pn
for dn = (2(n − 1
p−1 )) ∼ 2n by (10.1.5), then, in particular, limn→∞ pdn =0
n
in R. Thus, if we write φi,n,X (u) := ε∗pn X (ui ),
we have φi,n ≡ upi
mod (u)dn .
Put Φn (u) = Φn,X (u) = (φ1,n,X (u), . . . , φd,n,X (u)) ∈ k[[u]].
Since ρ(exp(pn X)) ⊗ idG m ∈ AutZp (L ⊗ G m ) = AutZ (T ), we have
p
f (u, v) = (m ◦ (idT ×δX ))∗ (f1 (u)) = (idT ×δX )∗ ◦ m∗ (f1 (u)).
Recall ι : k[[u]]/P → k[[x]] and its copy ι : k[[v]]/P → k[[y]] made out of
Proposition 10.2 applied to R = k[[u]]/P . Via ι⊗ι : k[[u]]/P ⊗ k k[[v]]/P →
k[[x, y]], we regard f (u, v) = f (g(x), h(y)) ∈ k[[x, y]]. Then, by the above
computation, we find
where the last inclusion comes from the induction hypothesis. Suppose r = 1.
Then we have
because δi,2 ◦ · · · ◦ δi,ni (Z) ⊂ Z. Thus, we have shown the result when r = 1.
r−1
Now we proceed by induction on r. Since z0 + i=1 δi,1 ◦ δi,2 ◦ · · · ◦ δi,ni (zi )
with z0 , z1 , . . . , zr−1 ∈ Z is in Z, by step 1 applied to δr,1 and Z × δr,2 ◦
· · · ◦ δr,nr (Z) ⊂ Z × Z, we get z0 + ri=1 δi,1 ◦ δi,2 ◦ · · · ◦ δi,ni (zi ) for any
z0 , . . . , zr−1 , zr ∈ Z. This finishes the proof of the first statement of step 2.
Note that σ is just the restriction of s to Z × Δ(Z) for the diagonal map
Δ : Z → Z r−1 . This proves step 2.
Step 3: We have m(Z × Z) ⊂ Z; so Z is stable under multiplication.
Proof of step 3. By Proposition 10.14, we can find a family of elements
{Xi,j ∈ g/Zp |i = 1, . . . , r, j = 1, . . . , ni }
10.2.1 Algebras
V is irreducible ⇒ R is simple.
To study modules over simple algebras, we start slightly more generally. Here
we only assume R to be an Artinian algebra. An R-module V is called com-
pletely reducible if it is a direct sum of irreducible modules. The algebra R is
called semisimple if its radical J = J(R) vanishes. Since maximal left ideals
of R/J(R) correspond bijectively to maximal left ideals of R by the homo-
morphism theorem, we see that J(R/J(R)) = 0. This shows that the quotient
R/J(R) is semisimple. Now, the following three statements are equivalent:
(SS1) R is semisimple.
(SS2) The left R-module R is completely reducible.
(SS3) Every R-module V of finite length is completely reducible.
Proof. We first prove the implication: (SS1)⇒ (SS2): Let Ω be the set of all
maximal left ideals of R. Then m + n = R for 5 two distinct elements m, n ∈ Ω.
Then,)by the Chinese remainder theorem, m∈Ω m = J(R) = 0 implies that
R∼ = m∈Ω R/m. Since R is Artinian, Ω is a finite set. By the homomorphism
theorem, R-submodules of R/m correspond bijectively to left ideals between
m and R. This shows that R/m is irreducible, since m is a maximal left ideal.
)
(SS2)⇒(SS3): Since R ∼ = )m∈Ω R/m, we have minimal left ideals Im indexed
by m ∈ Ω such that R = m∈Ω Im with Im ∼ = R/m as left R-modules. Then
1 = ⊕m∈Ω em for em ∈ Im . Multiplying the left-hand and right-hand sides by
a ∈ Im , we get Im a = a1 = ⊕m∈Ω aem , and therefore, aen = δm,n a, where
δm,n = 1 or 0 depending on whether or not m = n. Thus, Im = Rem . Replacing
a by em , we get e2m = em and em en = 0 if m = n. Consider an R-linear map
ϕ : Im → Im v given by ϕ(i) = iv for v ∈ V . Since Ker(ϕ) is an R-submodule
of Im , the irreducibility of Im tells us that either Ker(ϕ) = 0 or Ker(ϕ) = Im .
400 10 Toric Subschemes in a Split Formal Torus
∼
Thus, Im v = 0, then
either Im v = Im or Im v = 0. In particular, if m∈Ω em v =
v ∈ m∈Ω Im v. From this, we conclude that V = v∈V −{0} m∈Ω Im v. Then
) )
we can find mj and vj such that Imj vj = 0 and V = j Imj vj ∼ = j Imj as
R-modules, since V is of finite length.
(SS3)⇒(SS1): By (SS3), R is the direct sum of irreducible R-submodules Im .
By Lemma 10.8, I) m = J(R)Im . Since Im is irreducible, J(R)Im = 0. Then
J(R) = J(R)R = m J(R)Im = 0.
∼
We now decompose Ω = Ω1 · · · Ω )λ so that m, n ∈ Ωj ⇐⇒ Im = In
as left R-modules. Then we write Rj = m∈Ωj Im . If m ∈ Ωj , Im a for a ∈ R
is isomorphic to eitherIm or 0, and therefore, it has to)be inside Rj . Thus,
Rj R ⊂ Rj ; hence, Rj is a two-sided ideal, and R = j Rj is an algebraic
direct sum. Returning to the original setting and applying the above argument
to R/J(R), we have
(S) An artinian algebra R has only one maximal two-sided ideal if and only
if there is a unique isomorphism class of irreducible R-modules
since the set of maximal two-sided ideals of R naturally corresponds to that
of R/J(R) bijectively. We now claim that the following three assertions are
equivalent (Wedderburn’s theorem):
Exercise 10.11 Write down explicitly the Lie bracket [x, y] for two given
tangent vectors x, y at 1 ∈ G without extending x, y to vector fields over G.
Let G be another connected linear algebraic group defined over E with a
morphism ρ : G → G of group schemes over E. Write g for the Lie algebra of
G . Then its differential dρ = ρ∗ : g → g is a homomorphism of Lie algebras.
Exercise 10.12 Check that dρ is a homomorphism of Lie algebras.
An E-rational representation ρ : G/E → GL(n)/E is a homomorphism of E-
group schemes. Write gln (E) for the Lie algebra of GL(n)/E . Then gln (E)
can be identified with the Lie algebra of n × n matrices Mn (E) with entries
in E whose Lie bracket is given by [x, y] = xy − yx for the matrix product
xy with x, y ∈ Mn (E). Then dρ : g → gln (E) is an E-linear representation of
Lie algebras. Since ρ is complex analytic after extending scalars to C/E (after
embedding E into C), ρ is absolutely irreducible if and only if dρ is absolutely
irreducible. Over C, the exponential map expG : g/C = g ⊗E C → G(C)
∞ n
given by the usual formula expG (X) = n=0 Xn! converges absolutely, giving
an analytic surjection of g/C onto G(C) whose inverse is given by logG (g) =
∞ n+1 (g−1)
n
n=1 (−1) n . Then we see easily that expGL(n) (dρ(X)) = ρ(expG (X))
and logGL(n) (ρ(g)) = dρ(logG (g)) for all X ∈ g/C and g ∈ G(C) (see [GME]
Sect. 4.3.3 for a p-adic version of this). From this, the above equivalence on
absolute irreducibility of ρ and dρ is clear.
Lemma 10.13 Let K be an infinite field and {V1 , . . . , Vb } be a finite set of
finite-dimensional
) K-vector spaces. Let A1 , · · · , Ar be K-linear endomorphism
of V = bj=1 Vj preserving Vj (j = 1, . . . , b). If, for each j = 1, . . . , b, there
exists i = 1, . . . , r with det(Ai |Vj ) = 0,
rthen there exists (λ1 , . . . , λr ) ∈ K
r
b r
f (t) = det( ti Ai |Vj ).
j=1 i=1
r
dρ(wi,1 ) ◦ dρ(wi,2 ) ◦ · · · ◦ dρ(wi,ni )
i=1
is invertible as an endomorphism of V .
r
dρ(w1 ) ◦ dρ(w2 ) ◦ · · · ◦ dρ(wni )
i=1
In this last chapter, we prove that any Hecke invariant subvariety of a prod-
uct of copies of mod-p Shimura curve is a Shimura subvariety, as stated in
Theorems 3.43 and 3.44. From this result, we pull out a linear independence
result (Corollary 11.13) more general than Theorem 3.38.
In [Ch1], local linearity is called Tate linearity, and the definition of locally
linear subvarieties is given for a closed subvariety of the ordinary locus of
the Siegel modular variety. Since (V ord )n has a canonical closed immersion
into a Siegel modular variety, this definition is equivalent to Chai’s definition.
Chai conjectured that a weakly locally linear subvariety is actually locally
linear (see [Ch1] 5.3.1), which has been shown to hold for weakly locally
linear subvarieties of the ordinary locus of Hilbert modular varieties (see [Ch1]
Theorem 8.6 and [H10a] Sect. 3.5).
If Z is a variety with a morphism π : Z → (V ord )m and if, for a closed point
z ∈ Z, π induces an embedding of Zz into the formal completion of (V ord )m
at π(z), we can still speak of locally linearity (and weak locally linearity) at
z of Z (we make this abuse often later).
Take a1 , . . . , am ∈ Z×
p . By the action of aj on S (and hence on OS ) given by
t → t , we have an algebra homomorphism
aj
!
m
S sending f1 ⊗ · · · ⊗ fm to
φ : OV,x ⊗F · · · ⊗F OV,x → O S .
aj (fj ) ∈ O
j=1
(11.1.1)
If aj ’s (j = 1, . . . , m) are pairwise distinct modulo T (Z(p) ), we would like to
prove that φ is injective. Thus, for a nonconstant modular function f ∈ OV,x ,
{a1 (f ), . . . , am (f )} are linearly independent over F. Thus, we study Ker(φ)
for a1 , . . . , am ∈ Z× p.
Since ρ(T (Z(p) )) fixes x (Lemma 7.43), T (Z(p) ) acts on OS by ring au-
tomorphisms, and by Proposition 7.44, this action is compatible (and com-
mutes) with the action of Z× ×
p via the embedding T (Z(p) ) → Zp . Thus, we
have φ(α(f1 ) ⊗ · · · ⊗ α(fm )) = α (φ(f1 ⊗ · · · ⊗ fm )) for all α ∈ T (Z(p) ). In
1−c
We take a more general setting than the example given above, specified as
follows (we use the following notation throughout).
(N0) Let S = Spec(OV,x )/F and SK = Spec(OVK ,x )/F with their formal com-
pletion S and SK along x isomorphic to G m.
⊗m
(N1) For a prime ideal b ⊂ OV,x := (OV,x ⊗ · · · ⊗ OV,x ) (the m-fold tensor
product) stable under a p-adically open subgroup T of T (Z(p) ), we write
⊗m
X/F = Spec(OV,x /b) ⊂ S m and let Y X be the normalization.
(N2) X ⊂ Sm is a formal completion of X along its closed point xm =
(x, . . . , x).
(N3) X/F is the Zariski closure of X in V m (so X is stable under T and X is
the formal completion of X along xm ). Let Y X be the normalization.
Write X = limK XK ⊂ V m with irreducible closed subschemes XK ⊂
←−
VKm (the image of X in VKm ) and YK XK for the normalization of
XK (so Y = limK YK ), where K runs over open compact subgroups of
←−
G(A(∞) ) with K = K (p) × G(Zp ).
Suppose m = 1. Since dimF V = dimF Ig = 1, the irreducible reduced
T-invariant closed subscheme X is either a single point {x} or V itself, and
hence, we conclude b = 0; in particular, Ker(φ) = 0, taking b to be Ker(φ).
We restate Proposition 4.88 for elliptic curves:
Lemma 11.1 Let E → Z be an ordinary elliptic curve over a reduced excel-
lent affine base scheme Z over Fp . For a closed point s ∈ Z(Fp ) and a formal
completion Z splits into a product
along s, if the p-divisible group E[p∞ ] ×Z Z
of its connected component E[p∞ ]◦/Z and étale quotient E[p∞ ]et , then on an
/Z
open subscheme U ⊂ Z containing s, the p-divisible group E[p∞ ]/U canoni-
cally splits into the product E[p∞ ]◦ ⊕ E[p∞ ]et /U for the connected component
E[p∞ ]◦/U and the étale quotient E[p∞ ]et /U .
Here the words “reduced excellent Z” means that Z = Spec(R) with excellent
reduced ring R (see Sect. 4.1.11 about excellent rings).
Exercise 11.2 Give a counterexample against the assertion of the above
lemma if Z = Spec(Z(p) ).
Our goal is to prove the injectivity of φ for general m > 1 in (11.1.1) under
the assumption that the aj ’s are pairwise distinct modulo T (Z(p) ) (as the
result is trivial if m = 1). The injectivity is equivalent to Spec(Im(φ)) = S m ;
hence, we study the local property of Spec(Im(φ)) to show that dim(Im(φ)) <
dim S m implies the equality of two of the aj modulo T (Z(p) ). The following
result dealing with the local structure of Spec(Im(φ)) when m > 1 is a key to
prove the linear independence.
11.1 Stability under Hecke Correspondence 409
the action of T (Z(p) ) (not just T) diagonally embedded into AutGSCH (S) m.
From this, we claim that X and X are stable under T (Z(p) ). Indeed, for the
mxm -adic completion b of b, by [CRT] Theorem 8.1 (ii), X = Spf(O ⊗m
V,x /b) and
m ⊗ L. Then b =
b is stable under T (Z(p) ) as X = L G ⊗m
b ∩ OV,x by faithful
flatness of O over O
⊗m ⊗m
(cf. [BCM] I.2.6); so b is stable under T (Z ), and
V,x V,x (p)
⊗m
hence, X = Spec(OV,x /b) is stable under T (Z(p) ). Then the Zariski closure
X of X is stable under T (Z(p) ).
The normalization Y → X is given by limK YK for the normalization YK
←−
of XK . Naturally, the semigroup EndSCH (X ) of endomorphisms of the scheme
X acts on Y; in particular, T (Z(p) ) acts on Y. The formal completion Yy along
410 11 Hecke Stable Subvariety Is a Shimura Subvariety
not flat over π(G m ⊗ L) S . Define the nonflat locus Y nf l ⊂ Y as the
Zariski closure of the set of closed points y ∈ Y such that OY,y is not flat
over OV ,v for the image v ∈ V of y. By Proposition 5.3, the nonflat locus
yL ∈ Y nf l ⊂ Y is a nonempty closed subscheme strictly smaller than Y ,
because G m ⊗ L0 ⊂ Y = YyL is flat over S and flatness is an open property.
0
the origin in S = S. If δ = 1, we have rankZp L0 = rank X∗ (Sm ), and hence
X = S ; so we are done. We hereafter assume that δ = 0. Since π : G
m m ⊗L →
S is dominant and dim Gm ⊗ L = dim S = rankZp L, π : Gm ⊗ L → S is
finite flat for all L ∈ I. Thus, X → S is finite, and Y is finite flat over S by
the faithful flatness of Yy over S . This proves assertion (3).
To prove assertion (4), consider the differential sheaf ΩY /S . We may as-
sume that (πX ◦ ΠY )∗ : ΘY → ΘS is surjective at y0 = yL0 . Since Yy0 /S is
finite flat, Yy0 is étale over S , and hence Yy0 /S is étale. Thus, ΩY /S |Yy0 = 0
(Lemma 4.32), and hence ΩY /S vanishes on a nonempty open subscheme of
Y . Thus, the support Y ram ⊂ Y of ΩY /S is a closed subscheme of Y strictly
smaller than Y . If YyL /S is not étale, YyL ⊂ Y ram . Thus, Y ram is a closed
subscheme of dimension equal to dim Y ; so it is an irreducible component of
Y , and hence Y is reducible, a contradiction. Thus, Y → S is étale finite.
Remark 11.4 Let the notation and assumption be as in Proposition 11.3.
We suppose that m = 2 and that Y → X S 2 has two dominant projections
onto the left and right factors S of S 2 . We write the Serre–Tate coordinate
(induced by the ordinary level structure on Ex ) of the left factor (resp., the
right factor) of S2 as t (resp., t ). Then by Theorem 10.1, the formal completion
Yy of Y along a point y above x2 = (x, x) is canonically isomorphic to a
formal subtorus of S2 given by G m ⊗ L for a Zp -free direct summand L of
2
Zp . Thus, if dim X = dim V , Yy is defined by the equation tu = t for
v
L = {(x, y) ∈ Z2p |ux = vy}. If two projections are étale, (u, v) can be chosen
to be (1, a) for a unit a = v/u ∈ Z× p (otherwise, one of the projections involves
the pth power map, which cannot be étale).
Corollary 11.5 Let b = Ker(φ) for φ as in (11.1.1). Let y ∈ Y be a point
above xm and let Yy be the formal completion of Y along y. Then Yy for at
least one y ∈ Y contains Δ and for i = 1, 2, . . . , m,
= {(ta1 , . . . , tam )|t ∈ S},
writing Si for the ith copy of S in S , the projection X∗ (Yy ) → X∗ (Si ) is
m
V,x = O
S , Φ(f1 ⊗ · · ·⊗ fm) = m
where O j=1 aj (fj ), and the map ι is the tensor
product of the natural inclusion OV,x ⊂ O V,x . Thus, Ker(Φ) ⊃ ι(b). Note that
Spf(OV,x ⊗F · · · ⊗F OV,x / Ker(Φ)) is the skew diagonal image Δ in Sm . If we
412 11 Hecke Stable Subvariety Is a Shimura Subvariety
take the formal completion along xm , the map ι brings Δ into X because
Ker(Φ) ⊃ ι(b). Thus, an irreducible component G m ⊗ L of X contains Δ.
In
particular, if m = 2, πX ,∗ (L) contains πX ,∗ (X∗ (Δ)) = X∗ (S ). Then the rest
follows from assertion (4) of the above proposition.
X/F ⊂ V × V
414 11 Hecke Stable Subvariety Is a Shimura Subvariety
be an irreducible subscheme with (x, x ) ∈ X ord (F) [X ord = X ∩(V ord ×V ord )]
stable under the diagonal action of a p-adically open subgroup T in T (Z(p) ).
As we mentioned already (in the sketch of step 1), for the use of Zariski’s
main theorem, we need smoothness and properness of our varieties; hence,
we write V8 for a smooth compactification of V , X 8 for the Zariski closure of
X in V8 × V8 , and Π 8 : Y8 → X 8 for the normalization Y8 of X. 8 The action of
T on X extends to Y . For open compact subgroups K1 , K2 ⊂ G(A(p∞) ), we
write V12 for VK1 × VK2 , and we define X12 for the image of X in V12 and
X812 for the image of X8 in V812 := V8K1 × V8K2 . We write Y812 (resp., Y12 ) for the
normalization of X12 (resp., X12 ). Thus, Y8 = limK ×K Y812 . We suppose the
8
←− 1 2
following [see Proposition 11.3 (3)]:
(DE) The two projections Π1 , Π2 : Y → V are finite at a point y ∈ Y above
a point (x, x ) in V (F) × V (F) fixed by the diagonal action of Tx (Z(p) ).
Since x and x are fixed by ρ(Tx (Z(p) )), Ex and Ex have complex multiplica-
tion by the same imaginary quadratic field M . Since elliptic curves over C (and
hence over any characteristic-0 fields) with complex multiplication by a given
imaginary quadratic field are isogenous, applying this to canonical lift (with
Serre–Tate coordinate = 1) of Ex and Ex , they are isogenous and have com-
plex multiplication by M . This isogeny property also follows from Honda–Tate
theory [Ho] and [T2] because M is generated over Q by the same Frobenius
endomorphism φ [with (φ) = pm if Ex and Ex are defined over Fpm ]. We
also have dimF Y = dimF X = dimF V and dimF Y812 = dim X 812 = dim V8K . If
Πj is not étale at (x, x ), Πj,∗ (X∗ (Yy )) ⊂ X∗ (S) = Zp is an Zp -submodule of
finite index, and we find α ∈ Tx (Z(p) ) such that α1−c X∗ (S) = Πj,∗ (X∗ (Yy )) in
−1
X∗ (S) = Zp . Then ρ(α )◦ Πj is étale [by Proposition 11.3 (4)]. The action of
ρ(α) for a p-adic nonunit α1−cis not an automorphism of V but a “radiciel”
endomorphism of V . Indeed, 10 p0 acts on Sh(p) as the relative Frobenius
map of degree p, and hence if rx (α)p K = 10 p0 K, the action of ρ(α) co-
incides with the Frobenius map composed with rx (α(p) ) on ShK . Since any
statement concerning the underlying topological space of our schemes is not
affected by “radiciel” endomorphisms, we may assume (for such statements)
(A) Π1 and Π2 are étale finite at any point y ∈ Y above (x, x ),
by modifying Πj by ρ(α) for α ∈ M × . This condition follows from Proposi-
tion 11.3 (4) and Corollary 11.5 in the case of our interest: b = Ker(φ).
As already remarked, we will show that X is the graph of the action by
an element g ∈ G(A(p∞) ). In this process, we may also assume the following
condition without losing generality:
(B) x = x .
By moving X under the automorphism 1 × ρ(b(p) ) of V × V for a suitable
b ∈ MA× , we can bring x to x, and hence we may assume (B).
11.1 Stability under Hecke Correspondence 415
Let Vj = limK VKj (resp., V8j = limK V8Kj ) (j = 1, 2) be the jth factor in
←− j ←− j
V × V (resp., in V8 × V8 ). We now study the finite locus fi Vj = limK fi VKj in Vj
←− j
of the projection Πj : Y → Vj for j = 1, 2. By definition, the nonfinite locus
nfi 8
VKj is the closure of all closed points v ∈ V8Kj such that OYK ,y is not finite
j
over OVK ,v for at least one point y ∈ Πj−1 (v). We put fi V8Kj = V8Kj − nfi V8Kj .
j
Similarly, we define the nonflat locus nfl V8Kj ⊃ nfi V8Kj by the closure of all
closed points v ∈ V8Kj such that OYK ,y is not flat over OVK ,v for at least one
j j
point y ∈ Πj−1 (v). Since a flat morphism is an open map, by Corollary 5.11,
nfl 8
VKj is a closed subvariety of V8Kj of (at least) codimension 2; so nfl V8Kj = ∅
as dim V8Kj = 1. Thus, Π 8 j is everywhere finite flat.
Similarly, we define et VKj by the maximal open subscheme of VKj over
which Πj is étale (the cuspidal divisors ramify in V8j over V8Kj ; so the étale
locus is in Vj ). We write et VKj = VKj ∩ et V8Kj , and V et = et V1 ∩ et V2 (in other
words, over V et , Π1 and Π2 are both étale).
Proposition 11.6 Suppose (DE). Then we have
1. The morphisms Π8 j : Y8 → V8 (j = 1, 2) in V8 are finite flat.
2. If Π1 : Y → S and Π2 : Y → S are étale, V et = et V1 ∩ et V2 is an open
dense subscheme of V containing (x, x) stable under T.
Proof. The first assertion is already proven, so we prove (2). As explained
before, we may assume (A) and (B). The scheme Y12 is normal dominant
finite over V8Kj (and is generically étale finite). Since V8Kj is smooth if Kj
is sufficiently small, the ramified (non-étale) locus ram V8Kj = V8Kj − et V8Kj of
Y12 over V8Kj is a divisor of VKj ; hence, it is of at least codimension 1 for a
given Kj . We take the Stein factorization Y st → VKj of Y12 → VKj defined in
Sect. 5.1.3. Since V8Kj − VKj is the cuspidal divisor, the ramified (non-étale)
locus ram VKj ⊂ VKj of Y st over VKj is a divisor of VKj ; so it is of at least
codimension 1 for a given Kj . To extend this result to the provariety Vj ,
we need to show that ram VKj is sent into ram VKj under the projection map
VKj VKj . To show this, we look into the following commutative diagram
(removing cuspidal divisors):
finite →
Y12 −−−−→ X12 −−−−→ VK1 × VK2
⏐ ⏐ ⏐
⏐
πY .finite
⏐
πX .finite
⏐
π .étale
→
Y12 −−−−→ X12 −−−−→ VK1 × VK2 .
finite
X12 ×(VK1 ×VK2 ) (VK1 × VK2 ) of X12 to VK1 × VK2 is a closed subscheme of
VK1 × VK2 étale over X12 containing (x, x). By definition, X12 is the Zariski
closure of X (i.e., the image of X in VK1 × VK2 ), which is therefore given by
the above closed subscheme of VK1 × VK2 over X12 containing (x, x). Then
X12 is a closed irreducible subscheme of X12 ×(VK1 ×VK2 ) (VK1 × VK2 ) of equal
dimension, and X12 is the irreducible component of
X12 ×(VK1 ×VK2 ) (VK1 × VK2 )
containing (x, x) and covering X12 , and πX is étale finite. Thus, we have a
commutative diagram:
étale
Y12 ×X12 X12 −−−−→ Y12
⏐ ⏐
⏐ ⏐
. .normalization
X12 −−−−→ X12 .
étale
Since étale morphisms are isomorphisms at the level of completed local rings
(by Lemma 4.7), they commute with the formation of normalization. Thus,
Y12 is an irreducible component of Y12 ×X12 X12 , and therefore πY is étale
finite. Hence, the projection VKj VKj sends ram VKj into ram VKj , and
ram
Vj = limK ram VKj is a closed pro-subscheme of at least codimension 1
←− j
(by Corollary 5.14) of the provariety Vj (whose image in VKj is contained in
ram
VKj ). Hence, V et = et V1 ∩ et V2 is open dense in V . By our assumption,
(x, x) ∈ V et , and V et is stable under T, since étaleness is preserved by the
action of T.
Let v be a closed point of Y/F ord
= Y ×V 2 (V ord )2 above (v1 , v2 ) ∈ V 2 .
We consider the formal completion Yv [resp., VΠj (v) ] along v [resp., Πj (v)]. If
Π1 × Π2 embeds Yv into Vv1 × Vv2 and the equation defining Yv in the product
VΠ (v) × VΠ (v) is given by ta = tb for the Serre–Tate coordinate tj of Vvj ,
1 2 1 2
we call Y locally linear at v. Write Ylin ⊂ Y ord (F) for the subset of all closed
locally linear points.
Proposition 11.7 Suppose (DE), and let the notation be as in (N0–3) in
Sect. 11.1.3 for m = 2. In particular, let X be a Zariski closure of X in V × V
and Y → X be the normalization of X. The subset Ylin ⊂ Y of locally linear
points as defined above contains the set of all closed points of an open dense
subscheme Y lin in Y ord . In other words, at each closed point y ∈ Y lin (F), the
formal completion of Y lin along y is defined by a linear equation.
In the following corollary, we will find that the subvariety X is a graph of the
action of an element in G(A(p∞) ); hence, we conclude that Y lin = X lin = X ord .
Since X ord is locally linear at densely populated v in the image of Y lin , by
[Ch1] Proposition 5.3, X ord is weakly locally linear, and so Y ord is locally
linear. We shall give here an argument (again suggested by Chai) sufficient to
prove the weaker version as stated above.
11.1 Stability under Hecke Correspondence 417
for the Serre–Tate coordinate (t, t ) of S × S for S = Vx , where t and t are
associated with the ordinary level structure ηpord . Since the two projections
Π1 , Π2 : Y → V are étale at y [by (A)], a = v/u is a unit in Zp , and Yy
is defined by t = ta . We write ax ∈ End(Ex ) ⊗Z Zp = End(Ex [p∞ ]) for the
endomorphism inducing a ∈ End(S) as normalized above. Then ax is a unit
in EndGSCH (Ex [p∞ ]).
We consider the universal elliptic curve E/V . We pull it back to Π : Y →
X ⊂ V × V : Y1 = Π1∗ E and Y2 = Π2∗ E. Identifying Ex with the fibers Yj,y of
Yj (j = 1, 2) at y, we regard the unit ax ∈ End(Ex [p∞ ]) as a homomorphism
ax : Y1,x [p∞ ] = Ex [p∞ ] → Ex [p∞ ] = Y2,x [p∞ ].
We now reduce the existence of the desired nonempty open subscheme
Y lin ⊂ Y ord to the existence of an étale irreducible covering U 8 over an open
dense subscheme U ⊂ Y ord containing the given point y ∈ Y ord such that ax
extends to an isomorphism 8 a : Y1 [p∞ ]/U → Y2 [p∞ ]/U of Barsotti–Tate groups
over U 8 . Thus, supposing the existence of the open subscheme U and such an
extension 8 a over U8 , we specify the open subscheme Y lin . Shrinking U ⊂ Y ord
(keeping y inside U ), we may assume that the projections Πj : U → V
(j = 1, 2) are both étale (cf. Proposition 11.6), because the projections are
étale at the specific point y.
By the étaleness of Πj , the formal completion Yu along u ∈ U with Π(u) =
(u1 , u2 ) ∈ V × V is isomorphic by Πj (j = 1, 2) to the universal deformation
space of the Barsotti–Tate group Euj [p∞ ] carrying the universal deformation
Yj [p∞ ]/Yu ∼= Ej [p∞ ]/Vu via Πj : Yu ∼ = Vuj .
j
Choose an ordinary level structures ηj,p et
: Qp /Zp ∼ = Euj [p∞ ]. Then
∞ ∼ ∞
the canonical coordinate tj of E[p ]/Vu = Y[p ]/Yu is given by tj =
j
linear formal subscheme Yu defined by t1 = ta2 u . Since Yu → Vu1 × Vu2 is a
smooth formal subscheme with two isomorphisms Yu ∼ = Vuj induced by Πj ,
we find Yu = Yu , and hence, Yu is defined by t1 = ta2 u . Thus, we may let
Y lin = U .
Next we shall show that ax extends to a morphism a : Y1 [p∞ ] → Y2 [p∞ ]
of Barsotti–Tate groups over Yy . We make a remark first. In the above com-
putation, taking u1 = u2 = x and t = t1 = t2 , the action of ρ(α) (whose
matrix expression is given by α0 α0c inside GL2 (Qp ) by our choice of the
level-p
structure)
acts on the
coordinate t as if we were making conjugate
α1−c
ρ(α) 01 log1p (t) ρ(α)−1 = 1 logp (t ) by Proposition 7.44. By this re-
0 1
mark, identifying Ex with the fibers Yj,x of Yj (j = 1, 2) at y, we regard
the unit ax ∈ End(Ex [p∞ ]) as a homomorphism ax : Y1,y [p∞ ] = Ex [p∞ ] →
Ex [p∞ ] = Y2,y [p∞ ]. The formal completion Y = Yy of Y along y is the max-
imal subscheme of S × S over which this ax extends to a homomorphism
a : Y1 [p∞ ]/Y → Y2 [p∞ ]/Y of Y -group schemes.
To find a, as above, we identify a with an element of Zp by projecting
ax down to EndBT (Ex [p∞ ]et ) = Zp for the maximal étale quotient Ex [p∞ ]et
of the p-divisible group. The isogeny action ρ(α) : E/V → E/V for α ∈
End(Ex ) = O induces
∼ α ∼
ρ(α) : Y1 [pn ]Y −−→ E[pn ]/Vx −
→ E[pn ]/Vx −−−→ Y2 [pn ]/Y .
−1
Π1 Π2
Choose α so that α ≡ a mod pn Op for p = p∈Σp p and α ≡ 1 mod pn Opc ,
and write it as αn . Therefore, the isogeny action of ρ(αn ) on E gives rise to
αn : Y1 [pn ]/Yn → Y2 [pn ]/Yn well defined over an infinitesimal neighborhood
Yn of y (isomorphic to the connected component of Ex [pn ]). Indeed, the em-
∼
bedding Y → S × S is given by (t, ta ), ρ(αn ) sends Y1,v [pn ] −−→ E[pn ]
/Yn Π Vn
1
∼
to Y2,ρ(αn )p (v) [p ]/Yn −−→ E[p ]/Vn , and v = ρ(αn )p (v) as long as v ∈ Yn .
n n a
Π2
Here Vn is the infinitesimal neighborhood of x isomorphic to the connected
component of Ex [pn ]), ρ(αn )p ∈ G(Qp ) is the p-component of ρ(αn ), and
αn gives rise to a|Y1 [pn ] after a base change to Yn . By taking the limit
a = limn→∞ αn |Y2 [pn ]/Y with respect to n, we find out that ax gives rise to
n
a unique homomorphism a : Y1 [p∞ ]/Y → Y2 [p∞ ]/Y of Barsotti–Tate groups
over Y .
To show the existence of U , we follow Chai’s argument in the proof of
[Ch1] Proposition 8.4. We choose a sufficiently small open compact subgroup
K1 ×K2 ⊂ G(Z) 2 maximal at p so that Y12,y = Yy . Since Y ord is irreducible, we
12
only need to find a nonempty open subscheme U ⊂ Y12 ord
with the required
property (then U is given by the pullback of U to Y ord ). We have a well
defined over Y = Y12,y . For any reduced local OY12 ,y -algebra R, a locally
linear isomorphism α : Y1 [p∞ ]/R → Y2 [p∞ ]/R inducing ax on Ex [p∞ ] (if
11.1 Stability under Hecke Correspondence 419
for Yj = Πj∗ E = E ×Vj ,Πj Y . Let Y/Y = Y1 ×Y Y2 . Thus, there are only
two possibilities for EndQ (Y) = End(Y/Y ) ⊗ Q: Either EndQ (Y) = Q × Q
or EndQ (Y) = M2 (Q). Suppose that EndQ (Y) = M2 (Q). By the semisim-
plicity of the category of abelian schemes (e.g., [GME] Theorem 4.1.19), we
have two commuting idempotent ej ∈ EndQ (Y) such that ej (Y) = Yj . Since
EndQ (Y) = M2 (Q), we can find an invertible element β8 in GL2 (O(p) ) ⊂ M2 (Q)
such that β8◦e1 = e2 ; hence, β8 : Y1 → Y2 is an isogeny, whose specialization to
the fiber of Yj (j = 1, 2) at y gives rise to an endomorphism β ∈ End(Ex )⊗ Q.
Thus, the isogeny β8 is induced by ρ(β) [this point is explained more carefully
after proving EndQ (Y) = M2 (Q)].
We suppose EndQ (Y) = Q × Q and try to get a contradiction [in order
to prove that EndQ (Y) = M2 (Q)]. We pick a sufficiently small K1 = K2 =
K ⊂ G(A(p∞) ) maximal at p so that VK is smooth. For the moment, we
assume that K is open compact. The variety VK is naturally defined over a
finite extension Fq /Fp as the solution of the moduli problem EK,c in (7.1.12)
for the polarization ideal c of Ex (the minimal choice of Fq is the residue
field of W ∩ kK for Shimura’s field kK of the definition of VK ⊂ ShK/Q ). The
universal elliptic curve EK is therefore defined over VK/Fq , and EK is a variety
of finite type over Fq . Replacing q by its finite power, we may assume that
X12 ⊂ VK × VK/Fq is stable under the Galois action Gal(F/Fq ), and hence it
has a unique geometrically irreducible model X12/Fq ⊂ VK × VK/Fq defined
over Fq .
Let Y12/Fq be the normalization of X12/Fq . Then Y12 = Y12/Fq ×Fq F.
We write Y12/Y12/Fq for the abelian scheme E2K ×(VK ×VK ) Y12 . Then Y12
is an abelian scheme over the variety Y12/Fq of finite type over Fq . Let η
be the generic point of Y12/Fq , and write η for the geometric point over η
and Fq (η)sep for the separable algebraic closure Fq (η)sep of Fq (η) in Fq (η).
Take an odd prime # = p, and consider the #-adic Tate module T (Yη )
for the generic fiber Yη of Y. We consider the image of the Galois action
G := Im(Gal(Fq (η)sep /Fq (η))) in GLZ ×Z (T (Yη )).
By Zarhin’s result ([Z] and [DAV] Theorem V.4.7), the Zariski closure
over Q of G is a reductive subgroup of GLQ ×Q (T (Yη ) ⊗ Q). Write G for
the connected component of the reductive group. Then G ∩ G(Q ) is an open
subgroup of G(Q ). Moreover, by Zarhin’s theorem, the centralizer of G in
GLQ ×Q (T (Yη )⊗ Q) is End(Y)⊗ Q . Since the reductive subgroups of GL(2)
either are tori or contain SL(2), the derived group G1 (Q ) of G(Q ) has to be
SL2 (Q ×Q). Any noncentral semisimple element in SL2 (Q ) of infinite order
generates an infinite group whose Zariski closure has a torus as its identity
connected component, since SL(2) has only three types of proper algebraic
subgroups (i.e., tori, unipotent subgroups, and Borel subgroups).
Pick a pair (a1 , a2 ) of distinct nonconjugate semisimple elements in SL2
(Q × Q ) inside the Galois image [each belonging to SL2 (Q )] generating
an infinite group whose Zariski closure is a torus. Since Frobenius elements
are dense in G by (geometric) Chebotarev’s density in [Se1] Theorem 7,
11.1 Stability under Hecke Correspondence 421
Tu = (Tu1 × Tu2 ) ∩ G1
m ⊗ L ⊂ X
G (x,x) inside S2 . Thus, Δβ ⊂ X.
Again writing OΔ ,x2 = O⊗2 /bβ
β V,x
and OX,x2 = O⊗2 /b, this implies
V,x bβ ⊃
b for mx2 -adic completions
bβ and
b.
Therefore, bβ ⊃ b by the faithful flatness of O ⊗2 ⊗2
V,x over OV,x ; in other words,
Δβ ⊂ X. By the irreducibility of X, we conclude that X = Δβ . Since Δβ is
smooth, Δβ = Y , and hence X is smooth everywhere.
If condition (A) fails, as explained after statement (DE), the morphisms
ρ(α)−1 ◦ Π1 and ρ(β )−1 ◦ Π2 for suitable nonzero α, β ∈ O(p) are étale, and
so (ρ(α) × ρ(β ))−1 (X) = Δ1,β by the above argument; hence, X = Δα,β β .
This finishes the proof.
Here are two technical lemmas, before going into the case where m > 2.
Lemma 11.9 Let Ni = A for a commutative ring A (i = 1, 2, . . . , m). Let
N ⊂ N1 × N2 × · · · × Nm = Am be an A-free submodule of Am with m ≥ 2.
If A is a product of finitely many local rings, the projection of N to Ni × Nm
is surjective for all i = 1, 2, . . . , m − 1, and the projection π of N to N :=
N1 × N2 × · · · × Nm−1 is surjective, we have N = Am .
Proof. We may assume that A is a local ring. Tensoring its residue field,
by Nakayama’s lemma (Lemma 10.8), we may assume that A is a field k.
Suppose that dim N ≤ m − 1. Since π is surjective, π is an isomorphism, and
either N ∩ (N × zm ) is empty or z ∈ N ∩ (N × zm ) is a unique point with
π (z) = zm . Since N → N1 × Nm is surjective, there exist at least two points
in N ∩ (N × zm ), a contradiction. Thus, dim N = m and N = k m .
Proof. From the definition of the image π(X ), B0 is given by the image of the
π∗ K =
composite OS 2 −→ OS m OX . Let XK ⊂ VKm be as in (N3) such that X
X. Then π(XK ) is an excellent scheme since it is of finite type over a field F [see
Sect. 4.1.11 and [EGA] IV.7.8.3 (ii)]. For the projection XK of X to SK m
, write
π(XK ) = Spec(B0,K ). Then B0,K is a localization of π(XK ) at (x, x) ∈ SK 2
with rankZp # = dim B0 , by Proposition 11.3 (1), for finitely many Zp -direct
summands # ⊂ X∗ (S2 ). This proves assertion (1).
We now give a proof of assertion (2). Let Spec(B) be the normalization
of π(X ). Let πV : VKm → VK2 be the projection that induces π. Let U =
Spec(B0 ) ⊂ π(XK ) be a sufficiently small affine open neighborhood of (x, x)
such that its normalization B/B0 is finite. We can find such B0 because B/B0 is
finite by (1). We take an affine open neighborhood U = Spec(A0 ) ⊂ πV−1 (U )
of xm such that the normalization A/A0 is finite and A0 is of finite type
over B0 ; hence, A is a noetherian domain of finite type over B. As already
explained, A and B are excellent. Let U 8 = Spec(B) and U8 = Spec(A). Since
B ⊂ A and A and B are integral domains, the morphism Spec(A) → Spec(B)
is generically flat. The nonflat locus U 8 [which is the Zariski closure of {P ∈
nf l
U8 |AP is not flat over Bπ (P ) }] is a closed subscheme of U 8 strictly smaller
than U m ⊗ L of A along
8 . If rankZp π∗ (L) < rankZp #, the formal completion G
8
the point yL ∈ U corresponding to L is not flat over the formal completion
m ⊗ # of B along the image π(yL ). Thus, U
G 8 contains a closed subscheme
nf l
of maximal dimension, a contradiction against the irreducibility of U 8 . Thus,
rankZp π∗ (L) = rankZp #, as desired.
Here is a restatement of Theorem 3.44:
Corollary 11.11 Let the notation be as in (N0–3) and the assumption be as
in Proposition 11.3. Then X is smooth everywhere, and X ord is locally linear.
If X is finite over S , X either is given by V m−1 × {x} or is identical to
V m−2 × Δ(α,β) for some nonzero α, β ∈ O(p) (after permuting first m − 1
factors).
We now give a detailed proof.
Proof. We use the symbols introduced in Proposition 11.3 and its proof; in
particular, S = Spec(OV,x ) and S m = S × S ⊂ V m with S = S. By
Corollary 11.8 (m = 2), we may assume that m > 2. Assume that X = S m .
Since X is irreducible, its projection to S is irreducible. If the projection
of X to S is a proper closed subscheme in S (i.e., strictly smaller than
S ), by dim S = 1 (and by the fact that OV,x is a valuation ring), a closed
irreducible subscheme strictly smaller than S is a closed point. Thus, we find
that X = S × {x} as X → S is dominant; hence, we are done. Thus, we may
assume that the projection of X to S is dominant and that X is finite over
S [Proposition 11.3 (3)].
We describe a way of reducing the assertion to Corollary 11.8. Let Π :
Y → X and ΠY : Y → X be the normalization. Then by Proposition 11.3
(3), Y is finite flat over S , and Y is locally linear at every point y ∈ Y above
xm ∈ X (abusing the terminology).
Pick a point y ∈ Y above xm , and write
+
Yy = G
m ⊗ L and X = G m ⊗ L.
L∈I
424 11 Hecke Stable Subvariety Is a Shimura Subvariety
Recall that f (Z) denotes the Zariski closure in Z of the image of the
underlying topological space of Z for a morphism f : Z → Z of schemes. If
Z = Spec(A) and Z = Spec(B), then the topological space of f (Z) is that of
Spec(f ∗ (B)). Write Πi,m := πi,m ◦ Π : Y → Si × S . By Lemma 11.10 (2),
dim πi,m (X ) = rankZp πi,m,∗ (L) < rankZp (X∗ (Si × S )) = dim(Si × S ).
(11.1.3)
By (11.1.3), the reduced image πi,m (X ) ⊂ Si × S is an irreducible
closed subscheme (strictly smaller than Si × S ) invariant under T. Apply-
ing Corollary 11.8 to πi,m (X ), we find that πi,m (X ) = Spec(OΔα,β ,x2 ) and
πi,m (X) ⊂ Δα,β for some nonzero α, β ∈ O(p) . Permuting indices to bring i
to m − 1, we conclude
−1
X ⊂ πm−1,m (Δα,β ) = V m−2 × Δα,β .
Books
[AAF] G. Shimura, Arithmeticity in the Theory of Automorphic Forms. Math-
ematical Surveys and Monographs, vol. 82 (American Mathematical So-
ciety, Providence, RI, 2000)
[AAG] S.S. Gelbart, Automorphic Forms on Adele Groups. Annals of Mathemat-
ics Studies, vol. 83 (Princeton University Press, Princeton, NJ, 1975)
[ABV] D. Mumford, Abelian Varieties. TIFR Studies in Mathematics (Oxford
University Press, New York, 1994)
[ACM] G. Shimura, Abelian Varieties with Complex Multiplication and Modular
Functions (Princeton University Press, Princeton, NJ, 1998)
[ACS] K.-W. Lan, Arithmetic Compactifications of PEL-Type Shimura Vari-
eties. London Mathematical Society Monographs, vol. 36 (Princeton Uni-
versity Press, Princeton, NJ, 2013)
[ALG] R. Hartshorne, Algebraic Geometry. Graduate Texts in Mathematics, vol.
52 (Springer-Verlag, New York, 1977)
[ALR] J.-P. Serre, Abelian l-Adic Representations and Elliptic Curves (A K
Peters, Wellesley, MA, 1998)
[AME] N.M. Katz, B. Mazur, Arithmetic Moduli of Elliptic Curves. Annals of
Mathematics Studies, vol. 108 (Princeton University Press, Princeton,
NJ, 1985)
[AQF] G. Shimura, Arithmetic of Quadratic Forms. Springer Monograph in
Mathematics (Springer, New York, 2010)
[ARG] G. Cornell, J.H. Silverman (eds.), Arithmetic Geometry (Springer-Verlag,
New York, 1986)
[BCM] N. Bourbaki, Algèbre Commutative (Hermann, Paris, 1961–1998)
[BLI] N. Bourbaki, Groupes et Algèbres de Lie (Hermann, Paris, 1972–1985)
[BNT] A. Weil, Basic Number Theory (Springer-Verlag, New York, 1974)
[CBT] W. Messing, The Crystals Associated to Barsotti–Tate Groups; with Ap-
plications to Abelian Schemes. Lecture Notes in Mathematics, vol. 264
(Springer-Verlag, New York, 1972)
[CFN] J. Neukirch, Class Field Theory (Springer-Verlag, Berlin, 1986)
[CMA] H. Matsumura, Commutative Algebra (Benjamin, New York, 1970)
[CPS] G. Shimura, Collected Papers, vols. I, II, III, IV (Springer, New York,
2002)
[DoHI] K. Doi, H. Hida, H. Ishii, Discriminant of Hecke fields and twisted adjoint
L-values for GL(2). Invent. Math. 134, 547–577 (1998)
[Em] M. Emerton, p-Adic families of modular forms (after Hida, Coleman, and
Mazur). Astérisque 339, 31–61 (2011)
[FG] B. Ferrero, R. Greenberg, On the behavior of p-adic L-functions at s = 0.
Invent. Math. 50, 91–102 (1978)
[Fi1] T. Finis, Divisibility of anticyclotomic L-functions and theta functions
with complex multiplication. Ann. Math. 163, 767–807 (2006)
[Fi2] T. Finis, The μ-invariant of anticyclotomic L-functions of imaginary
quadratic fields. J. Reine Angew. Math. 596, 131–152 (2006)
[Fo] J.-M. Fontaine, Il n’y a pas de variété abélienne sur Z. Invent. Math. 81,
515–538 (1985)
[FW] B. Ferrero, L. Washington, The Iwasawa invariant μp vanishes for abelian
number fields. Ann. Math. 109, 377–395 (1979)
[GeJ] S. Gelbart, H. Jacquet, A relation between automorphic representations
of GL(2) and GL(3). Ann. Sci. Éc. Norm. Sup. (4) 11, 471–542 (1978)
[GhV] E. Ghate, V. Vatsal, On the local behaviour of ordinary I-adic represen-
tations. Ann. Inst. Fourier (Grenoble) 54, 2143–2162 (2004)
[Gi1] R. Gillard, Fonctions L p-adiques des corps quadratiques imaginaires et
de leurs extensions abéliennes. J. Reine Angew. Math. 358, 76–91 (1985)
[Gi2] R. Gillard, Remarques sur l’invariant mu d’Iwasawa dans le cas CM,
Sém. Théorie Nombres, Bordeaux 3, 13–26 (1991)
[Gr] R. Greenberg, Trivial zeros of p-adic L-functions. Contemp. Math. 165,
149–174 (1994)
[GS] R. Greenberg, G. Stevens, p-Adic L-functions and p-adic periods of mod-
ular forms. Invent. Math. 111, 407–447 (1993)
[Gs] B.H. Gross, On the factorization of p-adic L-series. Invent. Math. 57,
83–95 (1980)
[GsK] B.H. Gross, N. Koblitz, Gauss sums and the p-adic Γ -function. Ann.
Math. 109, 569–581 (1979)
[GZ] B. Gross, D. Zagier, Heegner points and derivatives of L-series. Invent.
Math. 84, 225–320 (1986)
[H81a] H. Hida, Congruences of cusp forms and special values of their zeta func-
tions. Invent. Math. 63, 225–261 (1981)
[H81b] H. Hida, On congruence divisors of cusp forms as factors of the special
values of their zeta functions. Invent. Math. 64, 221–262 (1981)
[H82] H. Hida, Kummer’s criterion for the special values of Hecke L-functions
of imaginary quadratic fields and congruences among cusp forms. Invent.
Math. 66, 415–459 (1982)
[H85] H. Hida, A p-adic measure attached to the zeta functions associated with
two elliptic modular forms. I. Invent. Math. 79, 159–195 (1985)
[H86a] H. Hida, Iwasawa modules attached to congruences of cusp forms. Ann.
Sci. Éc. Norm. Sup. 4th series 19, 231–273 (1986)
[H86b] H. Hida, Galois representations into GL2 (Zp [[X]]) attached to ordinary
cusp forms. Invent. Math. 85, 545–613 (1986)
[H88a] H. Hida, Modules of congruence of Hecke algebras and L-functions asso-
ciated with cusp forms. Am. J. Math. 110, 323–382 (1988)
[H88b] H. Hida, On p-adic Hecke algebras for GL2 over totally real fields. Ann.
Math. 128, 295–384 (1988)
432 References
[HM] H. Hida, Y. Maeda, Non-abelian base-change for totally real fields. Spe-
cial Issue of Pac. J. Math. in memory of Olga Taussky Todd, 189–217
(1997)
[HT1] H. Hida, J. Tilouine, Anticyclotomic Katz p-adic L-functions and con-
gruence modules. Ann. Sci. Éc. Norm. Sup. 4th series 26, 189–259 (1993)
[HT2] H. Hida, J. Tilouine, On the anticyclotomic main conjecture for CM
fields. Invent. Math. 117, 89–147 (1994)
[Ho] T. Honda, Isogeny classes of abelian varieties over finite fields. J. Math.
Soc. Jpn. 20, 83–95 (1968)
[Hs1] M.-L. Hsieh, On the non-vanishing of Hecke L-values modulo p. American
Journal of Mathematics, 134, 1503–1539 (2012)
[Hs2] M.-L. Hsieh, On the μ-invariant of anticyclotomic p-adic L-functions for
CM fields. J. Reine Angew. Math. (in press). doi:10.1515/crelle-2012-
0056, http://www.math.ntu.edu.tw/~ mlhsieh/research.htm)
[Hs3] M.-L. Hsieh, Eisenstein congruence on unitary groups and Iwasawa main
conjecture for CM fields (preprint posted at http://www.math.ntu.edu.
tw/~mlhsieh/research.htm)
[Hz1] A. Hurwitz, Ueber die Entwicklungscoefficienten der lemniscatischen
Functionen. Göttingen Nachr., 273–276 (1897)
[Hz2] A. Hurwitz, Ueber die Entwickelungscoefficienten der lemniscatischen
Functionen. Math. Ann. 51, 196–226 (1899)
[I] J. Igusa, Kroneckerian model of fields of elliptic modular functions. Am.
J. Math. 81, 561–577 (1959)
[K1] N.M. Katz, Serre–Tate local moduli, in Surfaces Algébriques. Lec-
ture Notes in Mathematics, vol. 868 (Springer-Verlag, Berlin, 1978),
pp. 138–202
[K2] N.M. Katz, p-Adic L-functions for CM fields. Invent. Math. 49, 199–297
(1978)
[Kh1] C. Khare, Serre’s modularity conjecture: The level one case. Duke Math.
J. 134, 557–589 (2006)
[Kh2] C. Khare, Serre’s conjecture and its consequences. Jpn. J. Math. 5, 103–
125 (2010)
[KhW] C. Khare, J.-P. Wintenberger, Serre’s modularity conjecture. I, II. I: In-
vent. Math. 178, 485–504 (2009); II: Invent. Math. 178, 505–586 (2009)
[Ki1] M. Kisin, Overconvergent modular forms and the Fontaine–Mazur con-
jecture. Invent. Math. 153, 373–454 (2003)
[Ki2] M. Kisin, Geometric deformations of modular Galois representations.
Invent. Math. 157, 275–328 (2004)
[KiW] L.J.P. Kilford, G. Wiese, On the failure of the Gorenstein property for
Hecke algebras of prime weight. Exp. Math. 17, 37–52 (2008)
[Ko] R. Kottwitz, Points on Shimura varieties over finite fields. J. Am. Math.
Soc. 5, 373–444 (1992)
[L] S. Lang, Sur les séries L d’une variété algébrique. Bull. Soc. Math. France
84, 385–407 (1956)
[Lx] J.H. Loxton, On two problems of R. M. Robinson about sum of roots of
unity. Acta Arith. 26, 159–174 (1974)
[Mh] K. Mahler, An interpolation series for continuous functions of a p-adic
variable. J. Reine Angew. Math. 199, 23–34 (1958) (Correction: J. Reine
Angew. Math. 208, 70–72 (1961))
434 References
[We2] A. Weil, Exercices dyadiques. Invent. Math. 27, 1–22 (1974) (Œuvres
III, [1974e])
[Y] R. Yager, p-Adic measures on Galois groups. Invent. Math. 76, 331–343
(1984)
[Z] J.G. Zarhin, Endomorphisms of abelian varieties over fields of finite char-
acteristics. Math. USSR Izvestija 9(2), 255–260 (1975)
[Zh] B. Zhao, Local indecomposability of Hilbert modular Galois representa-
tions, preprint, 2012 (posted on web: arXiv:1204.4007v1 [math.NT])
Symbol Index
[·], 24 φf , 370
x, 9 ϕψ , ϕ− ψ , 377
ψP , 87
αl,P , 92 Ω∞ , Ωp , 334, 335
α(t), 358 ΩVa /k , ΩC/k,η , ΩC/A , 48, 55, 66
Γ ± , 353, 376 ΩR/A , ΩX/S , 164, 166
Γ1,0
1
(f2 ), 338 ω(O), ω(a), 275
Γ (N ), Γ? (N ), Γ11 (N ), 285, 296
ω k , 250, 254
Γ0 (pm )p , Γ1 (pm )p , 280 ζN , 12
(γ : x → y), 69
Δalg , 353 [A], 370
Δ(E, ω), 60 A1 , An , 45
δP , P , 87 A/B , 262
∗ ϕ, 375 /B , ALG/B , ART/B , 205, 210, 225
ALG
εP 87 A(l), 92
η (p) , 277, 287 AP , 111
◦
ηpet (a), η∞ (a), ηpord (a), 277 a(l), 86
×
ι : T → Z, 370 B, 72
κ(P ), 86 BIALG/B , 193
Λ, 85 C, 337, 370
λ∗ , λ∗ , 371 C(a), 276
λκ,k , 130 C(Zl , A), C(G, B), C(Z, W ), 12, 15, 377
μ(ϕ), μ(Φ), 130 CL/B , 208, 210, 264
μN , μ× N , 3, 234, 281 CB , 151
π0 (T ), 229 CB , 205
π1 (X, x), 187 CBT/B , 210
π1top , 69 COF , 149
π∗ (ϕ), 375 CT F , 149
π − , 377 C(Π), 181
ρI , ρP , 86, 87 Cζ , 271
ρsym⊗n , 106 (CF ), 183
ρz , 31 Cl(Γ ), 290
Φ◦ , ψ ◦ , 339, 368 Clalg , 353