Pure and Applied Mathematics Quarterly

Volume 2, Number 2
(Special Issue: In honor of
John H. Coates, Part 2 of 2 )
519—538, 2006

Iwasawa Theory and Modular Forms

R. Sujatha
Long ago a capital stood here
And its road leads back into the past
The flowers are in bloom
And who would think that springtime
Should pass as in a dream?

Dedicated to John Coates for his sixtieth birthday

The Iwasawa theory of elliptic curves without complex multiplication has

been studied extensively in recent years ([2], [4], [17], [30], [22], [7]) culmi-
nating in the formulation of a Main Conjecture due to Coates, et al [3]. The
main theme of this paper is to study the GL2 Iwasawa theory of another
important class of p-adic Galois representations, namely those arising from
modular forms. We shall content ourselves with studying the algebraic ques-
tions that occur in this context and obtain results parallel to those proved
for the Galois representations arising from elliptic curves.

More precisely, we study the Selmer groups (see §2) associated to the
Galois representations coming from modular forms which are not CM (see
§1). We shall thus be interested mainly in the case when the image G
of the Galois representation is an open subgroup of GL2 (Zp ), and hence
non-abelian. The Selmer group is then a finitely generated discrete (left)
module over the non-commutative Iwasawa algebra Λ(G) of G (see §2 for
details). We remark that the Selmer group that we consider is actually the
strict Selmer group in Greenberg’s terminology [9] but all our main results
(Theorem 2.8 and Theorem 4.1) also hold for the Selmer group as defined
in [9] as the two Selmer groups differ by a cofinitely generated Zp -module.
We prove (Theorem 2.8) that the dual Selmer group tensored with Qp is

Received October 5, 2005.

520 R. Sujatha

infinite dimensional over Qp . Let L be a finite extension of Q such that

Gal(K∞ /L) is pro-p, where K∞ denotes the trivializing extension of the
representation associated to the modular form. We also show that the dual
Selmer group over Lcyc is infinite for the primitive modular forms of level 1
and even weight and that it even has positive λ-invariant under additional
hypotheses (see Theorem 4.1), which is conjectured to hold in general. The
corresponding results for the representations coming from elliptic curves can
be found in [17], [4], [22] and our methods are similar.

The paper consists of four sections. Section 1 is preliminary in nature

where we gather various notation and definitions. In §2, we define the Selmer
group which is the main object of study and prove that its dual tensored
with Qp is infinite dimensional as a Qp -vector space. In §3, we show that the
dual Selmer group has no non-zero pseudo-null submodules and use these
results in §4 to get information on the dual Selmer group over cyclotomic
extensions.

Acknowledgements: It is a great pleasure to thank John Coates for the

numerous mathematical discussions over the years on Iwasawa theory which
were immensely insightful and beneficial. We gratefully acknowledge the
hospitality of the Morningside Centre and thank the organisers of the con-
ference held in Beijing in honour of J. Coates. We would also like to thank
L. Clozel, E. Ghate, D. Prasad and R. Greenberg for useful discussions and
the referee for his detailed comments.

1. Preliminaries

Let f ∈ Sk (N ) be a primitive cuspidal modular form of positive weight

k ≥ 2, level N and trivial character for the group Γ0 (N ). For simplicity,
we assume that the Fourier coefficients of f lie in the field Q of rational
numbers. Thus the Fourier expansion of f is of the form
∞
X
f (z) = an q n , q = e2πiz , an ∈ Q.
n=1

Let Q (resp. Qp ) denote a fixed algebraic closure of Q (resp. Qp ). We fix an

embedding ip of Q in Qp and i∞ of Q into the field C of complex numbers.
By the results of Eichler, Shimura and Deligne, there is an associated Galois
representation, which we denote by

(1) ρf : Gal(Q/Q) −→ GL2 (Qp )

 Iwasawa Theory and Modular Forms 521

having the property that at all primes l - N p, the representation ρf is

unramified and
tr(ρf (Frobl )) = al , det(ρf (Frobl )) = lk−1 ,
where al is the lth Fourier coefficient of f and Frobl denotes the Frobenius
element at l.

Let V be the two dimensional Qp -vector space associated to ρf . We

choose a lattice T left invariant by the Galois action. Then the image of
ρf may be assumed to be a compact p-adic Lie subgroup of GL2 (Qp ). After
conjugation, one may therefore assume that ρf takes values in GL2 (Zp ). Let
(2) ρf,n : Gal(Q/Q) −→ GL2 (Z/pn )
be the reduction of ρf mod pn . Note that the lattice T is not unique and
therefore the representation ρf,n depends on the choice of the lattice T .
We denote by K∞ the trivializing extension for the representation ρf
so that the Galois group G := Gal(K∞ /Q) is isomorphic to the image of
ρf . For n ≥ 1, we define the field Kn as the finite extension of Q in K∞
corresponding to the extension trivializing the representation ρf,n which
implies that Gal(Kn /Q) is isomorphic to the image of ρf,n . The Galois
group Gal(K∞ /K1 ) is a pro-p group, being contained in the kernel of the
reduction map from GL2 (Zp ) to GL2 (Z/p). Let
(3) χ : Gal(Q/Q) → Z×
p

be the cyclotomic character giving the action of the Galois group of Q on all
p-power roots of unity. As det(ρf ) = χk−1 and the fixed field of the kernel
of χk−1 clearly contains the cyclotomic Zp -extension Qcyc of Q, we see that
Qcyc ⊆ K∞ . Note also that if the reduced representation ρf,1 is reducible
and contains µp as a sub-representation, then the field Q(µp ) is contained
in K1 and in this case the Galois group G(K∞ /Q(µp )) is also pro-p.

Hypothesis: We shall assume throughout the paper that f is ordinary at

p.

Recall that this means that ap is a p-adic unit. Let p be the prime of Q
induced by the embedding ip . We shall again denote by p the restriction
of p to any subfield of Q. Let Dp denote the decomposition group at p
determined by the embedding ip . Recall [9] that the representation V is
ordinary at p if it has a one dimensional Qp -subspace F + (V ) which is left
invariant under the action of Dp with the property that the inertial subgroup
Ip acts via a power of the cyclotomic character on F + (V ) and trivially on
the one dimensional quotient. The work of Mazur-Wiles [21] shows that if
f is ordinary at p, then the associated representation is ordinary at p and
522 R. Sujatha

that the image of the decomposition group under ρf is upper triangular. Let
F/Q be any finite extension of Q, and for each place w of F lying above
p, fix a place w of Q lying above w and let Dw denote the decomposition
group for w. It is easily seen that a Galois conjugate of F + V then provides
a one dimensional subspace Fw+ (V ) ⊂ V with the property that the inertia
subgroup acts by a power of the cyclotomic character on this subspace and
trivially on the one dimensional quotient.
We shall need the following results on CM and non-CM modular forms
(cf. [25]). If f is a CM modular form, then the Lie algebra of the image of
ρf is abelian over Qp of dimension 2 while for non-CM modular forms, the
Lie algebra of the image of ρf is the Lie algebra of GL2 (Qp ) and consists of
all 2 × 2 matrices over Qp . Suppose in addition that the modular form f is
ordinary at p. Then the image under ρf of a decomposition group at p, as
mentioned above, is upper triangular. More precisely, there is a basis of V
in which the restriction ρf | Dp consists of matrices of the form
µ ¶
δu
0²
where δ, ² : Dp → Q× ×
p are characters with ² unramified and u : Dp → Qp is
a continuous function. The representation ρf | Dp is said to be split at p if
V = V1 ⊕ V2 where each Vi is a line, stable under Dp . It is conjectured that
ρf | Dp is split if and only if f is a CM modular form. It is known that if
f is a CM modular form, then ρf | Dp is split and there are partial results
providing strong evidence for the converse (see [14]).
From now on, we fix a lattice T which is stable under the Galois action
and consider the discrete module A := V /T , so that A ' (Qp /Zp )2 . Let Cp
be the discrete submodule of A which is the image of F + (V ) in A and put
Wp for the discrete quotient A/Cp .
For a finite extension F of Q, let S(F ) denote the finite set of primes in F
consisting of those that lie above the primes that divide pN . We shall often
omit the reference to the field F and use just S to lighten notation. As is
usual, we denote by FS the maximal extension of F unramified outside of S
and the archimedean primes. Note that the field QS contains the extension
K∞ and the cyclotomic Zp -extension Qcyc of Q.

2. Selmer groups

Let p be an odd prime. For any profinite group G, we shall define the
Iwasawa algebra of G by Λ(G). Recall that this is defined as the inverse
limit
lim Zp [G/H]
←−
 Iwasawa Theory and Modular Forms 523

where H varies over the open normal subgroups of G and the inverse limit
is taken with respect to the natural maps.

We consider a modular form f as in §1, its associated representation ρf

and denote the image of ρf by G. Let S denote the set of primes of Q
containing the primes dividing N p defined above. The main object of study
will be a Selmer group which is a (left) module over the Iwasawa algebra
Λ(G) as explained below. Let F/Q be any finite extension of Q, and for
each place w of F , fix a place w of Q lying above w and let Dw denote
the decomposition group for w. As f is an ordinary modular form, the
representation is ordinary at p. Further, as mentioned before, it follows
from the results of Mazur-Wiles that for each place w of F lying above p,
there exists a 1-dimensional subspace Fw+ V ⊂ V such that
(i) Fw+ (V ) is stable under the action of the decomposition group Dw at w,
(ii) The inertia subgroup of Dw acts on Fw+ V via a power of the cyclotomic
character,
(iii) The quotient V /Fw+ V is unramified.

Let Cw denote the image of Fw+ V in V /T and put Ww for the quotient
A/Cw . For each place v of S = S(F ) and a finite extension L of F , define

 1
 ⊕ H (Dw , A) if v 6= p
w|v
Jv (L) = 1

 ⊕ H (Dw , Ww ) if v = p.
w|v

For an infinite extension L of Q contained in QS , we set

Jv (L) = lim Jv (L)
→
where L varies over finite extensions of F in L and the direct limit is taken
with respect to the restriction maps. Following Greenberg [9], we define the
Strict Selmer group SA (L) for any extension L of Q in K∞ by
(4) SA (L) = Ker (λ(L) : H 1 (QS /L, A)→ ⊕ Jv (L)),
v∈S

where λ(L) is the natural map induced by restriction.

The definition does in fact depend on the choice of the lattice. However
the conclusions of the theorems in this section and the subsequent ones are
valid for any lattice T of V and in that sense, our main results are inde-
pendent of the chosen lattice. In Greenberg’s terminology [9], the Selmer
group, which we denote here by SA 0 (L), is defined by replacing the decom-

position groups in the definition of Jv (L) by the corresponding inertial sub-

groups. With these definitions, the strict Selmer group is clearly contained
in the Selmer group. These Selmer groups are discrete (left) modules over
524 R. Sujatha

Λ(G) and we shall consider their compact duals. Let XA (K∞ ) (respectively
XA 0 (K )) be the Pontryagin dual of S (K ) (resp. of S 0 (K )). Our
∞ A ∞ A ∞
main result in this section is that XA (K∞ ) ⊗ Qp is infinite dimensional as
a Qp -vector space. By the above remark, this will also imply the infinite
dimensionality of the Selmer group as defined by Greenberg. We shall there-
fore consider only the Strict Selmer group and by abuse of terminology refer
to it as the Selmer group. It can be checked that the Strict Selmer group
and the Selmer group differ in general by a cofinitely generated Zp -module.
We also remark that in the special case of the representations associated to
elliptic curves over number fields, the definition of the Selmer group given
here differs slightly from the usual definition. However, we shall mainly be
working with Selmer groups over deeply ramified extensions, and over these
infinite extensions the two groups coincide.
As it is important to work with pro-p groups, we shall most often put
K1 = K and consider the Selmer group and its dual as modules over the
Iwasawa algebra Λ(GK ), where GK is the pro-p group GK = Gal(K∞ /K)
(cf. §1). It is then an easy consequence of Nakayama’s lemma (cf. [2]) that
XA (K∞ ) is finitely generated as a Λ(GK )-module. We want to show that
XA (K∞ ) ⊗ Qp is finite dimensional as a Qp -vector space. The corresponding
result for the case of the ordinary p-adic representation coming from elliptic
curves without complex multiplication is in [4]. The proof there follows ideas
outlined by Greenberg and our proof is somewhat similar.
We have the commutative diagram
L
0 −−−−→ SA (K∞ )Gn −−−−→ H 1 (QS /K∞ , A)Gn −−−−→ Jv (K∞ )Gn
v∈Sn
x x x

fn 

gn 

(5) hn 
L
0 −−−−→ SA (Kn ) −−−−→ H 1 (QS /Kn , A) −−−−→ Jv (Kn )
v∈Sn
where Gn := Gal(K∞ /Kn ), Sn := S(Kn ) and hn = ⊕ hn,v is the sum of
v∈Sn
the local restriction maps.
Lemma 2.1. The kernel and cokernel of gn are finite.

Proof. This can be proved using Lie algebra techniques as in [28]. As the
Lie algebra of Gn is either abelian of dimension 2 over Qp or equal to the Lie
algebra of GL2 (Qp ) depending on whether f is CM or non-CM, the proof
is particularly simple and we refer the reader to [6, Appendix] for the main
ideas. ¤
Let Kncyc denote the cyclotomic Zp -extension of Kn where the fields Kn
are as in §1 and denote by Hn the Galois group Gal(K∞ /Kncyc ) for n ≥ 1.
We consider the Selmer groups SA (Kncyc ) in the diagram below:
 Iwasawa Theory and Modular Forms 525

L
0 −−−−→ SA (K∞ )Hn −−−−→ H 1 (QS /K∞ , A)Hn −−−−→ Jv (K∞ )Hn
v∈Sn
x x x

αn 

βn 

γn 
(6)
cyc
λ(Kn ) L
0 −−−−→ SA (Kncyc ) −−−−→ H 1 (QS /Kncyc , A) −−−−−→ Jv (Kncyc ).
v∈Sn

Here the vertical maps are the natural restriction maps and γn = ⊕γn,v as v
runs over all the primes of Sn := S(Kn ). Recall that there are only a finite
number of primes in Kncyc lying over any finite prime of Kn .

Lemma 2.2. The kernel and cokernel of βn are finite.

Proof. We have Ker βn = H 1 (Hn , A) and Coker βn ⊆ H 2 (Hn , A). One

sees that the cohomology groups H 1 (Hn , A) and H 2 (Hn , A) are finite by
an argument entirely analogous to that in [5]. The key observation needed
is that AHn = A(Kncyc ) is finite. As the Lie algebra of Hn is either one
dimensional or equal to the Lie algebra of SL2 (depending on whether f
is CM or non-CM), the above finiteness statements follow easily from the
vanishing of H 0 (h, V ) where h is the Lie algebra of Hn . Note that the Lie
algebras of Hn coincide for all n as Hn+1 is an open subgroup of Hn . If
f is a CM modular form, then h is abelian of dimension 1 and consists of
diagonal matrices of trace zero. If f is a non-CM modular form then h has
dimension 3 and consists of all 2 × 2 matrices of trace zero. In both these
cases, it is then easily verified that H 0 (h, V ) is zero. This in turn implies
the vanishing of H i (h, V ) for i = 1, 2 as in [5] and the lemma follows. ¤

Lemma 2.3. Let Γn denote the Galois group Gal(Kncyc /Kn ). If SA (Kncyc ) is co-
torsion as a Λ(Γn )-module, then the homomorphism λ(Kncyc ) in (6) is surjective.

Proof. This is certainly well-known and classical for the case of Galois
representations coming from elliptic curves, see [16, Proposition 2.3] for a
proof. Similar arguments with the Poitou-Tate sequence show that the proof
carries over in this general case as well. ¤
Given a prime l of Q in S, let w denote a prime of K∞ above l, where
by a prime of K∞ , we mean a compatible sequence of primes at each finite
extension of Kn in K∞ . Let Gn,w ⊆ Gn and Hn,w ⊂ Hn denote the in-
tersection respectively of Gn and Hn with the corresponding decomposition
groups of w over l. Recall that there are only finitely many primes in Kncyc
lying above any prime of S.
526 R. Sujatha

Lemma 2.4. For a prime v of S(Kn ), let hn,v (respectively γn,v ) denote the
corresponding local component of the local restriction map hn (resp. γn ). Then
(
H 1 (Gn,w , A) if v - p,
(7) Ker(hn,v ) = 1
H (Gn,w , A/Cw ) if v | p

⊕H 1 (Hn,w , A) if v - p,
(8) Ker(γn,v ) = v0
⊕H 1 (Hn,w , A/Cw ) if v | p,
v0

where in (8), the sum varies over the finite set of primes v 0 of Kncyc which lie
above v, and w denotes some fixed prime of K∞ lying above v 0 .

Proof. We indicate the proof for (7), the other case being entirely parallel.
Given v ∈ S(Kn ), we first note that by Shapiro’s lemma, we have
(
Gn H 0 (Gn,w , H 1 (K∞,w , A)) if v - p
(Jv (K∞ )) '
H 0 (Gn,w , H 1 (K∞,w , A/Cw ) if v | p
where K∞,w denotes the completion of K∞ at any prime w of K∞ lying
above v. Therefore
Ker(hn,v ) = Ker(H 1 (Kn,v , B) → H 1 (K∞,w , B)Gn,w )
where B = A or A/Cw according as v - p or v | p. By the usual inflation-
restriction sequence, we see that (7) holds. ¤
Lemma 2.5. Let l be any prime, distinct from p, which divides N .
(i) If l2 - N, and v | l in S(Kn ), then for n >> 0, the group Ker(γn,v ) has Zp -
corank at least 1.
(ii) If l2 | N, then Ker(γn,v ) is trivial for a prime v | l in S(Kn ) for n >> 0,
provided the image of the inertial subgroup at v is finite, and Ker(γn,v ) is infinite
otherwise.

Proof. (i) Let w be a prime of Q lying above v | l. Our hypotheses that the
prime l2 - N and l 6= p ensures that the image Gw of the decomposition group
at w under ρf is of dimension 2. This follows on using the results of Carayol
([1], cf. [29, §3]). Indeed, in this case, the corresponding inertial subgroup
Iw of Gw is one dimensional and hence not finite. By the classical local
monodromy theorem of Grothendieck (cf. [13]), as l 6= p, the representation
ρf when restricted to an open subgroup of the inertia subgroup at w has
image consisting of matrices which are unipotent. Let Hn,w = Hn ∩ Gw . As
the trace of the elements in the Lie algebra of Hn,w is zero, we see that the
Lie algebra of the image of the inertial subgroup is contained in the image of
the Lie algebra of Hn,w . But Hn,w is a closed subgroup of Gw of dimension
at most one less than the dimension of Gw . In fact, it is true that in this
 Iwasawa Theory and Modular Forms 527

case the dimension of Hn,w is one (cf. [29, p. 71]). Combining this with
the above inclusion, we see that the Lie algebras of Hn,w and Iw coincide
and is in fact nilpotent. Let n denote this Lie algebra. By Engel’s theorem,
we therefore have H 0 (n, V ) has dimension at least 1. This translates to the
result that H 0 (Hn,w , V ) has dimension at least 1 for n >> 0 (we simply take
n large enough to include the finite extension L of Ql such that the image
of the inertial subgroup of L is unipotent). We thus get a corresponding
divisible subgroup B of A isomorphic to Qp /Zp on which Hn,w acts trivially.
Hence
Ker(γn,v ) = H 1 (Hn,w , A) ⊃ H 1 (Hn,w , B) = Hom(Hn,w , Qp /Zp ).
This latter group is easily seen to be infinite with Zp -corank at least 1.
(ii) If l2 | N , then the representation ρf restricted to the decomposition
group has dimension at most 2. Indeed, if the image of the inertial subgroup
is finite, then Gn,w is isomorphic to Zp and is topologically generated by the
Frobenius for n sufficiently large. In this case, the subgroup Hn,w is trivial
for n >> 0 and the same is true for Ker(γn,v ). If the image of the inertial
subgroup is infinite, then the arguments as in (i) apply and the lemma
follows. ¤
Remark 2.6. Consider a prime w of Q lying over a prime q ∈ S \ {p}, and its
restriction to K∞ . It is clear from the above proof (see also [29, §3]) that the
image Gw of the decomposition group Dw at w is a p-adic Lie group of dimension
at most 2 if f is non-CM and of dimension 1 if f is CM. Therefore its dimension is
strictly less than the dimension of G. Hence the corresponding number of primes
of Kn lying above q is unbounded as n → ∞ in these cases. If w lies over p and f
is not CM, then the decomposition group Dw is of dimension at most three and
hence the number of primes of Kn lying above q is again infinite. We shall use
this later in the proof of the main theorem of this section.
Proposition 2.7. Assume that the modular form f is non-CM. Then the group
Ker(hn ) contains (Z/pkn )tn where kn ≥ 1 and both kn , tn → ∞ as n → ∞.

Proof. As we have assumed the modular form f to be ordinary, the rep-

resentation is ordinary at p. As remarked earlier, for any prime w lying
above p in the finite extensions Kn of Q, there is then a submodule Cw of
A isomorphic to Qp /Zp which is invariant under the action of the decom-
position group Dw and such that on the quotient Ww = A/Cw the action
of Dw is unramified. Consider a prime v | p in S(Kn ). By Lemma 2.4, we
have Ker(hn,v ) = H 1 (Gn,w , Ww ) where Gn,w ⊆ Gn is the image of the de-
composition group at a prime w of K∞ lying over v. Let In,w ⊂ Gn,w denote
the image under ρf of the corresponding inertial subgroup. The subgroup
In,w is of codimension 1 in Gn,w (cf. [29, §3.2]). As f is not CM, the group
Gn,w consists of upper triangular matrices and hence has dimension at most
528 R. Sujatha

3 while In,w has dimension at most 2 and acts trivially on Ww . By the

Hochschild-Serre spectral sequence, we have a surjection
(9) H 1 (Gn,w , Ww ) ³ H 0 (Γn,w , H 1 (In,w , Ww ))
where Γn,w = Gn,w /In,w is isomorphic to Zp generated by the appropriate
Frobenius element. We now use the fact that In,w acts trivially on Ww and
hence
(10) H 1 (In,w , Ww ) = Hom(In,w /[In,w , In,w ], Ww ).
On the other hand, by the above remarks on the inertial subgroups (cf. [29,
Theorem 4], one sees easily that In,w /[In,w , In,w ] is abelian of dimension
1 for n >> 0. Hence the group in (10) is isomorphic to Qp /Zp (²) as a
Γn,w -module for some character ² : Γn,w → Z∗p . If the action is trivial, then
Ker(hn,v ) is clearly infinite. If on the other hand ² is non-trivial, then the
group (10) is finite of order pkn say, with kn 6= 0. In fact, in this case it is
easily seen that
(11) pkn = H0 (Γn,w , Zp (²)) = χ(Γn,w , Zp (²)),
where for a compact module M over Γ isomorphic to Zp ,
χ(G, M ) = # H0 (G, M )/# H1 (G, M )
is the Euler characteristic which is defined when the homology groups are
finite.
For m ≥ 0, the index [Γn,w : Γn+m,w ] = pm and as χ(Γn+m,w , M ) =
pm χ(Γn,w , M ), we see that
(12) pkn+m = pm · pkn ,
and therefore
(13) H 1 (Gn,w , Ww ) ⊇ (Z/pkn Z)
with kn ≥ 1 and kn → ∞ as n → ∞. Clearly
(14) Ker(hn ) ⊇ ⊕ H 1 (Gn,w , Ww )
w|p

where the sum is taken over all the primes w in Kn dividing p. The propo-
sition now follows from (13), (14) and Remark 2.6. ¤

We can now prove

Theorem 2.8. Let f be primitive cuspidal modular form of positive weight k ≥ 2,
level N and ordinary at p. Assume further that f is not CM. With notation as
before, the group XA (K∞ ) ⊗ Qp is infinite dimensional over Qp .
 Iwasawa Theory and Modular Forms 529

Proof. To prove that XA (K∞ ) ⊗ Qp is infinite dimensional over Qp , as in

[4], we first consider the restriction map
(15) αn : SA (Kncyc ) → SA (K∞ )Hn .
By Lemma 2.2, it has finite kernel. Let Γn = Gal(Kncyc /Kn ). If XA (Kncyc )
is not Λ(Γn )-torsion, then XA (K∞ ) is obviously infinite dimensional. We
therefore assume that XA (Kncyc ) is Λ(Γn )-torsion for all n = 1, 2, · · · . By
Lemma 2.3, this implies that the map λ(Kncyc ) in the bottom row of (6)
is surjective. The infinite dimensionality will then follow if we show that
the Zp -corank of Coker(αn ) is unbounded as n → ∞. By the snake lemma
applied to (6), this is equivalent to showing that
(16) Zp − corank of (Ker γn ) is unbounded as n → ∞
where γn is the map as in (6).
If the set S(Kn ) contains primes dividing N with the property that the
image of the corresponding inertia groups under ρf is infinite, then the
theorem is true. Indeed, observing that the number of primes in Kn dividing
such primes of S tends to infinity as n → ∞, we see that (16) is true by
Lemma 2.5 and Remark 2.6. We therefore assume that S(Kn ) contains only
primes dividing p and analyse the diagram (5). We consider the restriction
map
fn : SA (Kn ) → SA (K∞ )Gn .
Then the same argument as in [4, Lemma A.2] shows that
(17) Ker(gn ) ' (Z/pn )6 , for n ≥ 1.
Consider the homomorphism
λ(Kn ) : H 1 (QS /Kn , A) → ⊕ Jv (A)
v∈S

whose kernel is SA (Kn ). Using the Poitou-Tate sequence, the arguments in

[4, p. 231-232] along with Proposition 2.7 apply in this situation to show
that
(Z/pkn )tn ,→ Ker(hn ) ∩ Im(λ(Kn ))
and both tn , kn → ∞ as n → ∞. Combining this with (17), the snake
lemma applied to (5) proves that
0 0
(18) (Z/pkn )tn ,→ Coker(fn ).
0 0
Thus SA (K∞ )Gn maps onto Coker(fn ) which contains a subgroup (Z/pkn )tn
with the property that both kn0 , t0n → ∞ as n → ∞. As the p-primary
torsion in SA (K∞ ) is of finite exponent, one concludes as in [4, Appendix].
¤
530 R. Sujatha

3. Non-existence of pseudo-null submodules

Recall that G = Gal(K∞ /Q) and let Λ(G) be the corresponding Iwasawa
algebra. We consider XA (K∞ ) as a compact finitely generated Λ(G)-module.
The aim of this section is to prove that XA (K∞ ) has no non-trivial pseudo-
null Λ(G)-submodules. Recall ([30], [7]) that a finitely generated left module
M over a noetherian Auslander regular ring R is said to be pseudo-null if the
group E i (M ) := ExtiR (M, R) = 0 for i = 0, 1. We follow the arguments of
Ochi-Venjakob in [22] where they consider the case of p-adic representations
coming from abelian varieties. The one point that we want to stress here is
that for the case of representations coming from non-CM ordinary modular
forms, it is still a conjecture (albeit with considerable evidence, cf. [14])
that the local decomposition group Gp has dimension at least 3, which is
part of the hypotheses of [22, Theorem 5.2]. However, an extension of their
methods allows us to prove the result in general for non-CM modular forms.
For any extension L of Q in K∞ , let λ(L) denote the map

(19) λ(L) : H 1 (QS /L, A) → ⊕ Jv (L).

v∈S(L)

We assume that the map λ(K∞ ) is surjective. By standard arguments (see

[2, Theorem 4.5] for the case of elliptic curves), this is equivalent to the
module XA (K∞ ) being Λ(G)-torsion. Let XS = XS (A/K∞ ) denote the
compact Pontryagin dual of H 1 (QS /K∞ , A), considered as a Λ(G)-module.
We note also that the Galois group H 2 (QS /K∞ , A) is trivial (cf. [2, §2.5]).
Using this along with the Fox-Lyndon resolution technique (see [20], [22,
§4.1]), one sees that XS (A/K∞ ) is contained in a Λ(G)-module Y which
has projective dimension at most 1 and hence has no non-trivial pseudo-null
Λ(G)-submodule (cf. [22, Theorem 4.6]).

Theorem 3.1. Let XA (K∞ ) be the dual of the Selmer group where A = V /T
is the divisible module attached to the representation ρf for a non-CM ordinary
cuspidal primitive modular form f of weight ≥ 2 and level N which is square
free. Assume that XA (K∞ ) is a torsion Λ(G)-module. Then XA (K∞ ) has no
non-trivial pseudo-null submodule.

Proof. As mentioned above, the result follows from [22, Theorem 5.2]
when the local image Gp has dimension at least 3, noting that our hypothesis
ensures that for l ∈ S \ {p}, the image Gl has dimension 2 (see the proof
of Lemma 2.5). We assume therefore that Gp has dimension 2. Now Gp
consists of upper triangular matrices, and as the representation is assumed
to be ordinary, the Frobenius acts non-trivially on the unramified quotient
Wp . In particular, this implies that the local image Gp is necessarily split.
 Iwasawa Theory and Modular Forms 531

For a prime l ∈ S, we have the compact Λ(Gl )-modules

(
H 1 (Qp,∞ , Wp )∨ if l = p
Xl :=
H 1 (Ql,∞ , A)∨ if l ∈ S \ p
where, as before, Wp denotes the quotient A/Cp and for a discrete Λ(G)-
module M , M ∨ denotes the Pontryagin dual. As in [22, §5], let U denote
the Λ(G)-module
U = IndG G
Gp Xp ⊕ ⊕ IndGl Xl .
l∈S\p
By our assumption that Gp has dimension 2 and the remarks above on Gp ,
it follows from [20, 5.2 (c)] that
∼
(20) Xp ⊕ Λ(Gp )=Yp
where Yp is the module defined by the exact sequence
0 → Xp → Yp → Ip → 0
and Ip = Ker(Λ(Gp ) → Zp ) (cf. [20, §4]) is the kernel under the augmen-
tation map. Recall further that there is a Λ(G)-module Y [22, §4], and a
similar exact sequence
(21) 0 → XS → Y → I → 0
where I = Ker(Λ(G) → Zp ) is the augmentation kernel. The Λ(Gp )-module
Yp has projective dimension at most 1 [20], [22, §4], and Y has projec-
tive dimension at most 1 as a Λ(G)-module. By (20), it is clear that the
projective dimension of Xp is also at most 1. As in [22, Proposition 2.4
and §5], the proof of the theorem will be complete if we can show that
E i E i (XA (K∞ )) = 0 for all i ≥ 2 where E i (M ) := ExtiΛ(G) (M, Λ(G)). Since
we have assumed that XA (K∞ ) is Λ(G)-torsion, which is equivalent to the
map λ(K∞ ) being surjective (cf. [2]), we have the following exact sequence
of Λ(G)-modules:
0 → U → XS → XA (K∞ ) → 0.
As the dimension of Gl is 2 for l ∈ S \ p (see the proof of Lemma 2.5), the
proof of [22, Lemma 5.4 (ii)] applied to this case shows that Xl is zero. The
same arguments as in [22, Lemma 5.6] imply that the projective dimensions
of XS and XA (K∞ ) over Λ(G) are at most 2 and hence E i E i (XA (K∞ )) = 0
for i ≥ 3. We therefore need only to prove that E 2 E 2 (XA (K∞ )) = 0. The
above exact sequence gives an exact sequence
s
(22) E 1 (XA (K∞ )) → E 1 (XS )→E 1 (U ) → E 2 (XA (K∞ )) → E 2 (XS ) → E 2 (U ).
Let B denote the image of s in the above sequence. Then, as in the proof
of [22, Theorem 5.2], it suffices to prove that E 2 (B) = 0 as this will imply
that E 2 E 2 (XA (K∞ )) = 0.
532 R. Sujatha

Claim E 2 (B) = 0.
To prove the claim, we first note that as the dimension of G is 4 and
E i (Zp ) = 0 for all i 6= 4 [20, Corollary 2.6], it follows from (21) that
(23) E 1 (XS ) = E 1 (Y ).
Further by (20), we have
(24) EG1p (Xp ) = EG1p (Yp )

where EGi p (M ) = ExtiΛ(Gp ) (M, Λ(Gp )). Let ZA and Zp (Wp ) respectively de-
note the global and local dualising modules of A over Λ(G) and Wp over
Λ(Gp ), as in [22, Lemma 4.9(i), Proposition 4.10(i)]. As Y and Yp have pro-
jective dimensions 1 respectively over the Iwasawa algebras Λ(G) and Λ(Gp ),
we have
(25) E 1 (Y ) = ZA , EG1p (Yp ) = Zp (Wp ).
Here we have used the fact that ZA is torsion as a Λ(G)-module (cf. [22,
Proposition 4.10]). Further by [22, Lemma 4.9(i)], it follows that Zp (Wp ) is
a free Zp -module of rank 1. By [22, Lemma 5.4(iii)], we have Xl = 0. We
therefore have
E 1 (U ) = IndG 1
Gp EGp (Xp )
We drop the subscript and superscript in the induced module henceforth to
lighten notation. By (24) and (25), we see that
(26) E 1 (U ) = Ind Zp (Wp ).
Combining this with (23) we get a commutative diagram
u
ZA −−−−→ Ind Zp (Wp )
 
 
y y
E 1 (Y ) = E 1 (XS ) −−−−→ E 1 (U )
 
 
y y
0 0
where the vertical arrows are isomorphisms by (25). Consequently, by (22),
the module B is isomorphic to the cokernel of the natural map ZA →
Ind Zp (Wp ). On the other hand, it is clear that the group Coker(u) is
a quotient of Coker(u0 ), where u0 is the natural map u0 : ZA → Ind Zp (A).
Here Zp (A) is the dualising module of A over Λ(Gp ) and there is clearly a
natural surjective map f : Zp (A) ³ Zp (Wp ). On applying projective lim-
its to the classical Poitou-Tate sequence (see [23, 1.3.1]), one sees that the
cokernel of u0 is a free Zp -module of finite rank. In particular, the module
 Iwasawa Theory and Modular Forms 533

B is a finitely generated Zp -module. As the dimension of G is 4, we have

E 2 (B) = 0 which proves the claim.
We now use this in the exact sequence (22). As Xp has projective dimen-
sion atmost 1 over Λ(Gp ), we see that E 2 (U ) = Ind E 2 (Xp ) = 0. Therefore
(22) gives an exact sequence
0 → B → E 2 (XA (K∞ )) → E 2 (XS ) → 0.
As XS is contained in Y which has projective dimension at most 1, we know
that XS has no non-trivial pseudo-null Λ(G)-submodule [22, Corollary 4.7
and Proposition 2.4]. Hence E i E i (XS ) = 0 for i ≥ 2. Combining this with
the vanishing of E 2 (B), it follows that E 2 E 2 (XA (K∞ )) = 0 and the theorem
is proved. ¤
Remark 3.2. (i) When f is a CM modular form, it is unclear whether the above
methods apply as in this case both G and Gp have dimension 2. However, the
technique used by Perrin-Riou in [24] should imply a similar result.

4. Selmer groups over the cyclotomic extension

In this section, we apply the results of the previous sections to study the
Selmer group XA (Lcyc ) over Lcyc where L is a finite extension of Q contained
in K∞ such that Gal(K∞ /L) is pro-p and Lcyc denotes the cyclotomic Zp -
extension of L. Let GL denote the image of Gal(K∞ /L) under ρf . The
main reason for working over such ground fields L is that the Iwasawa al-
gebra Λ(GL ) is then a local noetherian ring and we can apply Nakayama’s
lemma. Recall that K = K1 ⊂ K∞ is the trivializing extension for the rep-
resentation ρf,1 and that Gal(K∞ /K) is pro-p. We shall further make the
hypothesis that p ≥ 5 so that the Iwasawa algebras have no zero divisors. If
the representation ρf,1 is reducible and contains µp as a subrepresentation,
then the image of Gal(Q/Q(µp )) under ρf is also pro-p. We may then take
the field L to be F := Q(µp ) which is an abelian extension of Q, instead
of K. The advantage in working with F is that one can apply the deep
result of Kato [19] proving that the Selmer group XA (F cyc ) is torsion as
a Λ(ΓF )-module where ΓF = Gal(F cyc /F ), which in turn implies that the
homomorphism λ(F cyc ) in (19) is surjective. For any field L as above, let
GL := Gal(K∞ /L) and HL := Gal(K∞ /Lcyc ), and let G and H respec-
tively denote the Galois groups GQ and HQ . For any Λ(G)-module M ,
let M (p) denote the submodule consisting of all p-power torsion elements.
We shall denote by MH (G) the abelian category of finitely generated Λ(G)-
modules M such that M/M (p) is finitely generated when considered as a
Λ(H)-module by restricting scalars. It seems reasonable to conjecture (see
[3]) that the Selmer groups XA (K∞ ) are always in MH (G). With notation
as before, our main result is the following:
534 R. Sujatha

Theorem 4.1. Let ρf be the representation associated to an ordinary non-CM

modular form f and let L be a finite extension of Q contained in K∞ such that
the Galois group G := Gal(K∞ /L) is pro-p. Assume that p ≥ 5 and that f has
level 1. Then XA (Lcyc ) is infinite. Moreover, if XA (K∞ ) is in MH (G), then
XA (Lcyc ) ⊗ Qp has positive dimension over Qp .

Proof. We first assume that XA (K∞ ) is in MH (G) and prove the final
assertion of Theorem 4.1. From the hypotheses on f , we see that the set
S := S(L) consists precisely of the primes that lie above p. We may clearly
assume that XA (Lcyc ) is Λ(Γ)-torsion and hence that the map λ(Lcyc ) is
surjective. We thus have the following commutative diagram:
(27) L
0 −−−−→ SA (K∞ )HL −−−−→ H 1 (QS /K∞ , A)HL −−−−→ Jv (K∞ )HL
v∈S
x x x

α

β γ


λ(Lcyc ) L
0 −−−−→ SA (Lcyc ) −−−−→ H 1 (QS /Lcyc , A) −−−−−→ Jv (Lcyc ) −−−−→ 0.
v∈S

We analyse the kernel of γ = ⊕ γv . As in Lemma 2.4,we have Ker(γv ) =

v∈S
⊕H 1 (Hw , Ww ) where Hw = HL ∩ Gw , and the sum varies over the finite set
v0
of primes v 0 of Lcyc lying above v and w denotes a fixed prime of K∞ lying
above v 0 . As p - N , the Galois representation is (potentially) crystalline and
hence the group H 1 (Hw , Ww ) is finite by [8, Theorem 1.5]. Further, the
group Ker(β) is also finite by Lemma 2.2. Applying the snake lemma, we
therefore see that Ker(α) and Coker(α) are finite. Let YA (K∞ ) denote the
quotient module XA (K∞ )/XA (K∞ )(p). Similarly, let YA (Lcyc ) denote the
quotient module XA (Lcyc )/XA (Lcyc )(p), which is a free Zp -module, thanks
to our assumption that XA (Lcyc ) is Λ(ΓL )-torsion. As we have assumed
that XA (K∞ ) ∈ MH (G), it is clear that by restricting scalars, we also have
XA (K∞ ) ∈ MHL (GL ). Arguing as in [3, Lemma 5.3], one sees that there
is an equality of µ-invariants, namely µGL (XA (K∞ )) = µΓL (XA (Lcyc )). On
dualising the map α, the above considerations show that

(28) rankZp YA (Lcyc ) = rankZp (YA (K∞ ))HL

where, as usual for a Λ(HL )-module M , MHL denotes the coinvariants. Fur-
ther, we also have (loc.cit.) Hi (HL , YA (K∞ )) is finite for all i ≥ 1. In
particular, as HL is pro-p and Λ(HL ) is local, we get (see [18])
(29)
X3
r := rankΛ(HL ) YA (K∞ ) = (−1)i rankZp Hi (HL , YA (K∞ )) = rankZp (YA (K∞ ))HL
i=0
 Iwasawa Theory and Modular Forms 535

which by (28) and the finiteness of the higher homology groups, is the same
as rankZp YA (Lcyc ). We claim that this latter rank cannot be zero. Indeed,
if it were zero, then YA (K∞ ) would be Λ(HL )-torsion and therefore pseudo-
null by a result of Venjakob [31]. We need the following general lemma,
whose simple proof below was pointed out to us by the referee:

Lemma 4.2. Let G be a pro-p compact p-adic Lie group with no element of
order p and having a closed normal subgroup H such that G/H is isomorphic
to Zp . Let M be any module in MH (G) such that M has no non-zero pseudo-
null Λ(G)-submodule. Then N := M/M (p) has no non-zero Λ(G)-pseudo-null
submodule.

Proof. As M is noetherian, there is an integer k such that pk annihilates

M (p). Let κ : M → M be the map given by multiplication by pk , so that
Ker(κ) = M (p). Clearly the image of κ is isomorphic to N = M/M (p).
As N is isomorphic to a Λ(G)-submodule of M , it plainly has no non-zero
pseudo-null submodules. ¤
The above lemma therefore implies that the rank r in (29) is not zero
unless YA (K∞ ) = 0. But if the latter were true, then XA (K∞ ) would be
p-primary torsion, thereby contradicting Theorem 2.8. Thus we see that
YA (K∞ ) has positive Λ(HL )-rank which by (28) is equal to the λ-invariant
of XA (Lcyc ) and the final assertion of the theorem is proved.
To prove the first assertion of the theorem, suppose the contrary and
assume that XA (Lcyc ) is finite. Then by Nakayama’s lemma and (27), we
see that XA (K∞ ) is finitely generated over Λ(HL ). By [31] and Theorem 3.1,
we clearly have XA (K∞ )(p) = 0 and hence XA (K∞ ) is in MH (G) and the
above results apply, showing that XA (Lcyc ) has positive λ-invariant. This
is a contradition and therefore XA (Lcyc ) is always infinite. The proof of the
theorem is now complete.
¤

Remark 4.3. (i) We remark that if the level N is equal to pr with r > 1, then
ap = 0 and the form is not ordinary. If N = p, then ρf is ordinary precisely when
the weight k = 2 (see [9, p. 4]). In this case, it is a classical result that the λ-
invariant is positive when the corresponding elliptic curve has split multiplicative
reduction at p, thanks to the existence of trivial zeros of p-adic L-functions [11].

We now consider some numerical examples.

Examples 4.4. Take f = Σn τ (n)q n to be the unique normalized cusp form of

weight 12 and level 1, which has been studied by Ramanujan and many others.
Take p = 691, which is an ordinary prime for this form. It is well-known (see [27,
536 R. Sujatha

§5] , [15]) that the form f is not of CM type. Ramanujan’s congruence

τ (l) ≡ 1 + l11 mod 691
implies easily that K∞ must be a pro-691 extension of the field K = Q(µ691 ).
As remarked by Greenberg [9], it can be shown that, up to homothety, there
are just two GQ -invariant lattices in V , and hence two choices for A. What
we now explain holds for both choices of lattices. Let us consider XA (K cyc ), as
defined in this paper (note that our normalizations are not the same as that used
by Greenberg in [9], [10], and, for simplicity, we do not not spell out in detail
here the precise relationship between Greenberg’s Selmer groups, which involve
the Tate twists of A, and ours). Since the Galois group of K cyc /Q is the direct
product of a cyclic group ∆ of order 690 and a group isomorphic to Z691 , we can
decompose
689
M
XA (K cyc ) = XA (K cyc )(i) ,
i=0
where XA (K cyc )(i) denotes the eigenspace in which the group ∆ acts via the i-th
power of the character giving its action on µ691 . Now Theorem 4.1 shows that if
XA (K∞ ) belongs to the category MH (G), then at least one of the XA (K cyc )(i) ⊗
Qp must have positive Qp -dimension for some i = 0, · · · , 689. On the other hand,
after taking account of Greenberg’s different normalization, one can easily deduce
from Greenberg’s arguments along with the calculations of Manin and the work
of Kato [19] on the Main Conjecture, that
XA (K cyc )(i) ⊗ Qp = 0 for i = 0, 689, 688, · · · , 680 mod 690.
I am very grateful to Ralph Greenberg for pointing out to me that the meth-
ods of his paper with Vatsal [12], for the prime 691, enable one to prove that
XA (K cyc )(i) ⊗ Qp has positive dimension precisely for odd integers i = 1, 189,
491, 679 modulo 690, and in each case the λ-invariant is 1. It does not seem
at all easy to prove this more precise result by our methods. Of course, there
could also conceivably be even integers i such that XA (K cyc )(i) ⊗ Qp has positive
λ-invariant. Finally, we remark that similar phenomena occur for other modular
forms of level 1 and higher weight and primes p (see ([10, p. 230]) for which there
are suitable Ramanujan style congruences.
 Iwasawa Theory and Modular Forms 537

References

[1] H. Carayol, Sur les représentations l-adiques associées aux formes modulaires de Hilbert,
Ann. Sci. ENS 19 (1986), 409–468.
[2] J. Coates, Fragments of the GL2 Iwasawa theory of elliptic curves without complex multi-
plication,Arithmetic theory of elliptic curves (Cetraro, 1997), SLN 1716, Springer, (1999),
1–50.
[3] J. Coates, T. Fukaya, K. Kato, R. Sujatha. O. Venjakob, The GL2 -Main Conjecture for
elliptic curves without complex multiplication, Publ. Math. IHES 101 (2005), 163–208.
[4] J. Coates, S. Howson, Euler characteristics and elliptic curves II, Journal of Math. Society
of Japan, 53 (2001), 175–235.
[5] J. Coates, R. Sujatha, Euler-Poincaré characteristics of abelian varieties, C. R. Acad. Sci.
Paris Sér. I Math. 329 no. 4, (1999), 309–313.
[6] J. Coates, R. Sujatha, Galois cohomology of elliptic curves, TIFR Lecture Notes Series,
Narosa Publishing house (2000).
[7] J. Coates, R. Sujatha, P. Schneider, Links between cyclotomic and GL2 Iwasawa theory,
Documenta Math. Extra Volume in honour of K. Kato, (2003), 187–215.
[8] J. Coates, R. Sujatha. J.-P. Wintenberger, On the Euler-Poincaré characteristics of finite
dimensional p-adic Galois representations, Publ. Math. IHES 93 (2001), 107–143.
[9] R. Greenberg, Iwasawa theory of p-adic representations, Adv. Stud. in Pure Math. 17 (1989),
97–137.
[10] R. Greenberg, Iwasawa theory for motives, L-functions and Arithmetic ed. J. Coates and
M. Taylor, LMS Lecture Notes Series 153 (1991), 211–233.
[11] R. Greenberg, Trivial zeros of p-adic L-functions, p-adic monodromy and the Birch and
Swinnerton-Dyer conjecture, Amer. Math. Soc., Providence, RI, Contemp. Math. 165
(1994), 149–174.
[12] R. Greenberg, V. Vatsal, On the Iwasawa invariants of elliptic curves, Invent. Math. 142
(2000), 17–63.
[13] A. Grothendieck, Letters of 24/9/64 p.183 and 3-5/10/64 p.204 in Correspondance
Grothendieck-Serre, Documents Mathématiques, vol. 2, Soc. Math. France, (2001).
[14] E. Ghate, V. Vatsal, On the local behaviour of ordinary Λ-adic representations, Ann. Inst.
Fourier, Grenoble 54 (2004), 2143-2162.
[15] E. Ghate, On the behaviour of ordinary modular Galois representations, Modular curves
and abelian varieties, Progress in Mathematics 224 Birkhauser-Verlag (2004), 105-124.
[16] Y. Hachimori, K. Matsuno, An analogue of Kida’s formula for the Selmer group of elliptic
curves, J. Alg. Geom 8 (1999), 581–601.
[17] S. Howson, Iwasawa Theory of Elliptic Curves for p-adic Lie Extensions, Cambridge PhD
thesis, (1998).
[18] S. Howson, Euler characteristics as invariants of Iwasawa modules, Proc. London Math.
Soc. 85 (2002) 634–658.
[19] K. Kato, p-adic Hodge theory and values of zeta functions of modular forms, Cohomologies
p-adiques et applications arithmétique, III,Astérisque 295 (2004) 1117–290.
[20] U. Jannsen, Iwasawa modules up to isomorphism, Adv. Stud. in Pure Math. 17 (1989),
171–207.
[21] B. Mazur, A. Wiles, On p-adic analytic families of Galois representations, Comp. Math.
59 (1986), 231–264.
[22] Y. Ochi, O. Venjakob, On the structure of Selmer groups over p-adic Lie extensions J. Alg.
Geom. 11 (2002), 547–576.
[23] B. Perrin-Riou, Fonctions L p-adiques des représentations p-adiques Astérisque 229 (1995).
538 R. Sujatha

[24] B. Perrin-Riou, Groupe de Selmer d’une courbe elliptique à multiplication complexe, Com-
positio Math. 43 (1981), 387–417.
[25] K. Ribet, Galois representations attached to eigenforms with Nebentypus, Modular functions
of one variable V, LNM 601 (1977), 17–52.
[26] A. Scholl, Motives for modular forms, Invent. Math. 100 (1990), 419–430.
[27] J.-P. Serre, Une interprétation des congruences relative à la fonction τ de Ramanujan, Sém.
Delange-Pisot Exp. 14 (1967/68).
[28] J.-P. Serre, Sur les groupes de congruence des variétés abeliénnes. II. Izv. Akad. Nauk SSSR
Ser. Mat. 35 (1971), 731–737.
[29] J. Tilouine, Modular forms and Galois representations, Bull. Greek Math. Soc. 46 (2002),
63–78.
[30] O. Venjakob, On the Iwasawa theory of p-adic Lie extensions, Comp. Math. 138 (2003),
1–54.
[31] O. Venjakob, A non-commutative Weierstrass preparation theorem and applications to Iwa-
sawa theory, Crelle J. (2003), 153–191.

R. Sujatha
School of Mathematics
Tata Institute of Fundamental Research
Homi Bhabha Road
Mumbai 400 005, India
E-mail: [email protected]