Anticyclotomic Main Conjectures

J. reine angew. Math.
606 (2007), 41—78 Journal für die reine und

DOI 10.1515/CRELLE.2007.034 angewandte Mathematik
( Walter de Gruyter
Berlin New York 2007
Anticyclotomic main conjectures

for CM modular forms
By Trevor Arnold at Ann Arbor
Abstract. Generalizing work of Agboola-Howard on CM elliptic curves, we prove

a version of the main conjecture of Iwasawa theory over the anticyclotomic Zp -extension
of a quadratic imaginary field K for Selmer groups attached to modular forms with com-
plex multiplication by K. We deduce a statement to the e¤ect that if the L-function of a
CM modular form f of even weight has odd order of vanishing at its central point, then a
Selmer group attached to f over Q is infinite.
Introduction
Let E be an elliptic curve defined over Q with complex multiplication by the ring of
integers in a quadratic imaginary field K. Agboola and Howard [1] have proved a so-called
main conjecture for the Selmer group of E over the anticyclotomic Zp -extension of K for
primes p where E has good, ordinary reduction. The main conjecture computes the rank
of the anticyclotomic Selmer group as a module over the Iwasawa algebra and describes
the characteristic ideal of the torsion submodule of this Selmer group in terms of a p-adic
L-function.
As the results of Agboola-Howard can be phrased entirely in terms of the Grös-
sencharakter c ¼ cE of K associated to E, it is natural to ask whether their methods
can be generalized to ‘‘higher weight’’ Grössencharaktere, i.e., Grössencharaktere of type
ðw 1; 0Þ, w f 2 (all of which are associated to cuspidal newforms of weight w). The pur-
pose of this paper is to give an a‰rmative answer to this question. In particular, we show
that the anticyclotomic Selmer group associated to certain Tate twists of a Grössenchar-
akter c of K of type ðw 1; 0Þ is usually a torsion Iwasawa module with characteristic ideal
generated by the constant term of the associated (2-variable) p-adic L-function. The excep-
tions occur when the twist is chosen to be a specific value (depending on the weight) and the
sign in the function equation of c is equal to 1; in this case, the rank of the Selmer group
is 1, and the characteristic ideal of the torsion submodule is related to the linear term of the
p-adic L-function via a p-adic regulator. We note that a similar anticyclotomic main con-
jecture was proved by Tilouine [13] for finite-order characters.
As a consequence of the above, we are able to prove a statement towards a general-
ization of the Birch and Swinnerton-Dyer conjecture. We show that if the sign in the func-
42 Arnold, Anticyclotomic main conjectures for CM modular forms
tion equation of the L-function associated to a CM modular form f of even weight is 1,
then a Selmer group associated to f over Q has positive rank. See Theorem 3.11 for a pre-
cise statement.
Much of the di‰culty involved in generalizing the methods of Agboola-Howard lies

in the fact that information arising from the geometry of CM elliptic curves is not available
for Grössencharaktere associated to higher-weight CM forms. For example, the representa-
tions we consider are not in general self-dual, and there is no analogue of the fact that the
p-torsion of a CM elliptic curve is a non-trivial Galois module locally at p. In addition, we
have had to take a slightly di¤erent approach to prove non-vanishing results about the
p-adic L-functions we work with.
For precise statements of the main theorems and outlines of their proofs, see 2.1. The
reader familiar with Iwasawa theory is advised to proceed to §2 after reading 1.1, returning
to §1 as needed to review notation.
The author would like to thank Tobias Berger and Chris Skinner for helpful discus-
sions.
1. Notation and definitions
Let Q be the algebraic closure of Q in C. We view algebraic extensions of Q as being

contained in Q. If F is such an extension, we denote by GF the absolute Galois group
GalðQ=F Þ of F . Given a GF -module M and an algebraic extension F 0 =F , we denote by
MðF 0 Þ the module of GF 0 -invariants M GF 0 ¼ H 0 ðGF 0 ; MÞ.
1.1. CM forms and representations. Let f A Sw ðN; wÞ be a normalized newform of

level G1 ðNÞ, weight w f 2, and character w and let Kf be the number field generated by
the Fourier coe‰cients of f . Then, by work of Deligne [3], there is for each prime P of
Kf (which we fix in the sequel) a representation rf ¼ rf ; P : GQ ! GL2 ðKf ; P Þ whose trace
and determinant satisfy
ð1:1Þ trace rf ðFrobl Þ ¼ al ð f Þ; det rf ðFrobl Þ ¼ l w1 wðlÞ;
for any l F Np, where p is the rational prime below P and al ð f Þ is the lth Fourier coe‰-
cient of f . This representation is unramified outside of Np.
Recall that f is said to be P-ordinary for a prime P j p of Kf if ap ð f Þ is a unit at P. If

f is P-ordinary, then Deligne showed that the restriction of rf to the decomposition group
Dp at p takes the form

a
rf jDp G ;
b
with b unramified.
We say that f has complex multiplication, or is a CM form, if there is a quadratic

imaginary field K and Grössencharakter c of K of type ðw 1; 0Þ such that f is the cusp
form associated to c (see, e.g., [9], p. 34). In this case, the representation associated to f
will satisfy rf G rf n wK , where wK is the Dirichlet character associated to K. This implies
that
Arnold, Anticyclotomic main conjectures for CM modular forms 43

cp
ð1:2Þ rf jGK G
cpt
for some character cp , where cpt ¼ cp t is the conjugate character of cp (t denotes the
involution of GK induced by complex conjugation).
For each split prime q of K lying over a rational prime q, choose an embedding
iq : Q ,! Qq in such a way that iq and iq t are conjugate. Given a Grössencharakter f of K
and a split prime q of K, one associates a p-adic Galois character fq : GK ! C p by the rule
fq ðsa Þ ¼ iq fðaÞ , where sa is the Frobenius associated to a fractional ideal a of K prime to
Np. This notation is consistent with (1.2) in the sense that the Galois character cp of (1.2) is
exactly the p-adic Galois character of K induced by the Grössencharakter c with respect to
the prime p (provided that P is the prime of Kf induced by the embedding ip ). From this
and the above formula (1.1) for the determinant of rf , we see that cp cpt ¼ e w1 w, where e is
the p-adic cyclotomic character.
If f is P-ordinary for P j p, then p splits in K: cp is ramified at p, whereas cpt is

unramified at p, so p 3 p t . Set p ¼ p t , so pOK ¼ pp . Denote by O the integers of
F ¼ Kf ; P . Given a character x : GK ! O , we denote by Ox a free O-module of rank 1 on
which GK acts via x. The GK -module we study in the sequel is T ¼ Ocpt w1 ð2 w þ cÞ, where
w1 w1
c3 is any integer (the assumption that c 3 is needed to ensure that our
2 2
representations do not split locally; see Lemma 2.5). Actually, we are more interested in
1
the Tate dual T ¼ Ocp ðcÞ of T, especially in the case c ¼ w 1. For ease of nota-
2
tion, we set h ¼ cpt e 2wþc w1 and h ¼ cp ec . Additionally, define V ¼ T nO F and
W ¼ V =T ¼ T nO ðF=OÞ, and similarly for V and W . Note that when c ¼ c t, we
have cpt ¼ cp (so that w is trivial), f ¼ f where f denotes the conductor of c, and the places
ramified for T are the same as the places ramified for T .
Remark. In order to make our notation consistent with [11], we have switched the
roles of T and T compared to the corresponding notation in [1]. A similar remark will
apply when we define the associated p-adic measures m and m .
In what follows, we assume chosen a prime P F 6N at which f is ordinary and let p be

the rational prime over which P lies. We moreover choose an isomorphism Cp G C which
is compatible with the embedding ip : Q ,! Qp .
1.2. Selmer groups. Let F be a finite extension of K. Following Greenberg [5],

Rubin [11], §1.3, and others, we define local conditions for our representations as follows:
8
> 1
<Hur ðFv ; V Þ if v F p;
Hf1 ðFv ; V Þ ¼ 0 if v j p;
>
: 1
H ðFv ; V Þ if v j p ;
8
> 1
<Hur ðFv ; V Þ if v F p;
Hf1 ðFv ; V Þ ¼ H 1 ðFv ; V Þ if v j p;
>
:
0 if v j p :
The exact sequence
0!T !V !W !0
gives the sequence of cohomology groups (exact in the middle)
H 1 ðFv ; TÞ ! H 1 ðFv ; V Þ ! H 1 ðFv ; W Þ:
We define the local conditions Hf1 ðKv ; W Þ, resp. Hf1 ðKv ; TÞ, as the image, resp. preimage, of
Hf1 ðKv ; V Þ under the maps in this sequence, and similarly for W and T . Let S ¼ SF be
the set of primes of F lying over p. Then for any M for which the notation has been de-
fined, we define Selmer groups

L 1
Sel S ðF ; MÞ ¼ ker H 1 ðF ; MÞ ! H ðFv ; MÞ=Hf1 ðFv ; MÞ ;
vFp

S L 1 1
SelðF ; MÞ ¼ ker Sel ðF ; MÞ ! H ðFv ; MÞ=Hf ðFv ; MÞ ;
vjp

L 1
SelS ðF ; MÞ ¼ ker SelðF ; MÞ ! H ðFv ; MÞ :
vjp
1.3. Iwasawa modules. For any abelian extension F =K (though F will always be a
finite extension of a Zp - or Zp2 -extension of K), define the Iwasawa algebra LðF Þ to be the
completed group ring O7GalðF =KÞ8 ¼ lim
O½GalðF 0 =KÞ, where the limit is taken over
0
fields F L F which are finite over K. This O-algebra has an involution i : LðF Þ ! LðF Þ
which acts as iðgÞ ¼ g1 on g A GalðF =KÞ. Given any LðF Þ-module M, we can define
M i ¼ M nLðF Þ LðF Þ;
where we view LðF Þ as an algebra over itself via i.
If M ¼ T or T and F is as above, we define
SelðF ; MÞ ¼ lim

SelðF 0 ; MÞ;
where the limit is taken with respect to corestriction maps between finite extensions F 0 of K.
Similarly, if M ¼ W or W , we define the LðF Þ-module
SelðF ; MÞ ¼ lim
!
SelðF 0 ; MÞ;
the limit now being taken with respect to restriction maps. We define

X ðF Þ ¼ HomO SelðF ; W Þ; F=O ;

X ðF Þ ¼ HomO SelðF ; W Þ; F=O ;
and make the analogous definitions for X S ðF Þ, XS ðF Þ, X ; S ðF Þ, and XS ðF Þ. We will also
have occasion to use the semi-local cohomology groups
8 L 1 0
>
>lim H ðFv ; MÞ if M is compact;
< vjq
H 1 ðFq ; MÞ ¼ L 1 0
>
> H ðFv ; MÞ if M is discrete;
:lim
!
vjq
where q is a prime of K. Tate local duality gives a perfect pairing
H 1 ðFq ; TÞ H 1 ðFq ; W Þ ! F=O

satisfying ðlx; yÞ ¼ x; iðlÞy for any l A LðF Þ. In order to make the Tate pairing LðF Þ-
equivariant, we agree to view X ðF Þ as a LðF Þ-module via ðl f ÞðxÞ ¼ f iðlÞx and similarly
for X ðF Þ, X S ðF Þ, etc.
Denote by Cy , Dy , and Ky the cyclotomic Zp -extension, the anticyclotomic Zp -

extension, and the unique Zp2 -extension of K, respectively. Set G ¼ GalðDy =KÞ, so
G G Zp . Thus we have a field diagram:
Ky
Cy Dy
G
K:
We will often work with the Iwasawa algebras LðKy Þ and LðDy Þ, so it is convenient to
define the ideal I L LðKy Þ to be the kernel of the restriction map LðKy Þ ! LðDy Þ.
We need one further definition in order to define the p-adic L-functions we use. Let
Qpur be the maximal unramified extension of Qp , and let O~ be the integers in the completion
~ ðF Þ ¼ O~7GalðF =KÞ8 for any abelian extension F =K.
~ of Q ur F. We define L
F p
1.4. p-adic L-functions. This and the following subsection are substantially similar
to their counterparts in [1]. The p-adic L-function we work with is defined as a suitable
‘‘twist’’ of the (2-variable) p-adic L-function of Katz interpolating the L-functions of Grös-
sencharaktere of K.
For any integral ideal f of K, denote by Kðfpy Þ the union of the ray class fields Kðfp n Þ
of conductor fp n . The Katz p-adic L-function is a measure

~ Kðfpy Þ
mAL
which has the property that for any Grössencharakter f of K of type ðk; jÞ and conductor
dividing fpy such that
ð1:3Þ k > 0 and j e 0;
there is an interpolation formula

fðpÞ
fp ðmÞ ¼ dðfÞ 1 Ly; fp ðf1 ; 0Þ;
p
where fp is the p-adic character associated to f and dðfÞ is an explicit nonzero constant.
(Note that to make sense of this equation, we need to make use of our chosen isomorphism
Cp G C.) For more information about these L-functions, see the book of de Shalit [2], Ch.
II, especially II.4.14. Note that II.6.7 of [2] extends the interpolation range to that given in
(1.3).

y

For any fractional ideal f of K and any O -valued character x of Gal Kðfp Þ=K ,
~ ~
define the O-algebra automorphism of L Kðfp Þ y

~ Kðfpy Þ ! L
Twx : L ~ Kðfpy Þ

by the rule Twx ðgÞ ¼ xðgÞg for all g A Gal Kðfpy Þ=K . If L L Kðfpy Þ is any extension
of
~ y
K, we denote by mðL; xÞ the image of Twx ðmÞ under the projection L Kðfp Þ ! LðLÞ. ~
Additionally, we define, for every integral ideal a L OK prime to fp, the element lðL; x; aÞ
to be the image of Twx ðsa NaÞ in LðLÞ (sa is the Frobenius associated to a). Then set
mðL; x; aÞ ¼ mðL; xÞlðL; x; aÞ.
If we assume in addition
that the ideal f satisfies f ¼ f, then complex conjugation de-
fines an involution t of L~ Kðfpy Þ . We set m ¼ tðmÞ and define in the same way as above
the measures m ðL; xÞ and m ðL; x; aÞ ¼ m ðL; xÞlðF ; x; aÞ. As noted above, the roles of m
and m are switched from the corresponding notation in [1].
These twisted L-functions of course also satisfy interpolation properties. Suppose that
x ¼ yp is the p-adic character associated to a Grössencharakter y. Then for any Grössen-
charakter f such that fp , resp. fp , factors through L, we have

fp mðL; x; aÞ ¼ fyðaÞ Na ðfyÞp ðmÞ;
resp.

fp m ðL; x; aÞ ¼ fyðaÞ Na ðfyÞp ðm Þ:
If we take x to be the p-adic character h associated to cNc , then provided that fcNc ,
resp. fcNc , has type ðk; jÞ satisfying (1.3), we get

ð1:4Þ fp mðL; h ; aÞ

fcðpÞ
¼ dðfcNc Þ fcNc ðaÞ Na 1 1þc Ly; fp ðf1 c1 ; cÞ;
p
resp.

ð1:5Þ fp m ðL; h ; aÞ

fcðp Þ
¼ dðfcN Þ fcN ðaÞ Na 1 1þc Ly; fp ðf1 c1 ; cÞ:
c
c
p
In the case that L ¼ Ky , if we choose a topological generator g of GalðKy =Dy Þ, we

get power series expansions
mðKy ; h ; aÞ ¼ La; 0 þ La; 1 ðg 1Þ þ La; 2 ðg 1Þ 2 þ ;
mðKy ; h Þ ¼ L 0 þ L1 ðg 1Þ þ L2 ðg 1Þ 2 þ ;
~ ðDy Þ.
where La; i and Li are elements of L
1.5. Euler systems. The Katz p-adic L-function is related by a theorem of Yager to
the so-called Euler system of elliptic units. It is therefore not surprising that the twisted
L-functions constructed in 1.4 are related to a twisted version of the elliptic unit Euler sys-
tem. It is not instructive to give the definition of this Euler system, so we content ourselves
by simply stating its properties.
Let a L OK be an ideal prime to fp. The elliptic unit Euler system (with respect to a)
consists of, for each ideal i L OK prime to a, a certain unit Yði; aÞ A KðiÞ. These units satisfy
the following relations:

Yði; aÞ if j j i;
ð1:6Þ NmKij =Ki Yðij; aÞ e ¼
Yði; aÞ 1sj if j F i;
where sj is the Frobenius of j and e ¼ wi =wij (w? being the number of roots of unity of K
congruent to 1 mod ?).
If q ¼ p or p and F is a finite abelian extension of K, we define

L ð1Þ
Uq ðF Þ ¼ OFv ;
vjq
ð1Þ
where OFv is the group of units of Fv congruent to 1 mod v. If F is an infinite abelian exten-
sion of K, we define
Uq ðF Þ ¼ lim

Uq ðF 0 Þ;
where the limit is taken with respect to norm maps between finite extensions F 0 of K con-
tained in F . Since the relations (1.6) show that the Yðfp n ; aÞ are norm-compatible
for n > 0,
y
we may define yðaÞ, resp. y ðaÞ, to be the image of this sequence in Up Kðfp Þ , resp.
y
Up Kðfp Þ . The following theorem of Yager relates this Euler system to the Katz p-adic

L-function:

Theorem 1.1. There is an injection of L ~ Kðfpy Þ -modules

Uq Kðfpy Þ n ~ Kðfpy Þ
^ Z O~ ! L
p

with image the ideal J L L ~ Kðfpy Þ generated by fsb Nb j ðb; fpÞ ¼ 1g. The image of yðaÞ,
resp. y ðaÞ, under this map is ðsa NaÞm, resp. ðsa NaÞm .
Proof. This theorem is proved in [2], Prop. III.1.4. (As remarked there, our assump-
tion that p 3 2; 3 allows us to avoid taking 12th roots.) r
By Kummer yði; aÞ ¼ NmKðifpÞ=KðiÞ Yðifp; aÞ as being an

theory, we may view the unit S
element of H 1 KðiÞ; Zp ð1Þ . If we define Ka ¼ KðiÞ, then the relations (1.6) show that
ði; aÞ¼1

the yði; aÞ form an Euler system for Zp ð1Þ; fp; Ka in the sense of Rubin’s book [11], §2.1.
The following proposition (essentially Prop. 2.3.1 of [1]) shows that we can twist this Euler
system into an Euler system for T which is related to the twisted p-adic L-function defined
in 1.4.
Proposition 1.2. There is an Euler system ca for ðT; fp; Ka Þ and injective ‘‘Coleman’’
maps

Col : H 1 Kðfpy Þp ; T n ~ Kðfpy Þ ;
^ O O~ ! L

Col : H 1 Kðfpy Þp ; T n ~ Kðfpy Þ
^ O O~ ! L
such that if we define z, resp. z , to be the local image of

ca Kðfpy Þ ¼ lim c Kðfp k Þ A lim
a
H 1 Kðfp k Þ; T
k k

in H 1 Kðfpy Þp ; T , resp. H 1 Kðfpy Þp ; T , then Colðz n 1Þ ¼ m Kðfpy Þ; h ; a and
Col ðz n 1Þ ¼ m Kðfpy Þ; h ; a . The image of these maps is the ideal Twh ðJÞ generated
by fsb hðsb Þ j ðb; fpÞ ¼ 1g.
Proof. Recall that h ¼ cpt e 2wþc w1 ¼ eðh Þ1 , so there is a commutative diagram of
O-modules
b
lim
H 1 Kðfpy Þ; Oð1Þ ! lim
H 1 Kðfpy Þ; T
? ?
? ?
?
loc p y ?locp
y
1
1

Up Kðfpy Þ G H Kðfpy Þp ; Oð1Þ ! H Kðfpy Þp ; T
bp

where b and b p are O-isomorphisms satisfying b Tw1 y
h ðlÞ ¼ l b for l A L Kðfp Þ and
similarly for b p .
We define theEuler system ca to be the twist of yði; aÞ in the sense of [11], Thm. 6.3.5,
so that z ¼ b p yðaÞ . The map Col is just the composite
b1
p Twh
H 1 Kðfpy Þp ; T ! Up Kðfpy Þ ! J ! Twh ðJÞ;
where the second arrow is the isomorphism of Theorem 1.1. The construction of Col is
analogous. r
We will need to view this Euler system as an Euler system for ðDy ; TÞ, but this is
potentially problematic: Rubin’s definition of Euler system requires that all primes of K be
finitely decomposed in Dy , which is not the case. In order to apply main results of [11], we
make use of [11], §9.2, which states that it is enough to show that for every prime q of K
splitting completely in Dy =K and every field F L Ka finite over K, the class ca ðF Þ is locally
unramified at every place above q. Thus the following lemma certainly su‰ces:
Lemma 1.3. For every field F L Ka finite over K, the class ca ðF Þ is locally unramified
at every place of F not lying over p.
Proof. Use [11], Cor. B.3.5, and the fact that ca ðF Þ is a universal norm in the cyclo-
tomic direction. r
2. Divisibility from Euler systems

w1
2.1. Statement of results. Recall that we have chosen an integer c 3 and have
2
set T ¼ Ocp ðcÞ, W ¼ T n ðF=OÞ, and h ¼ cp ec . We are mostly interested in the
1
structure of X ðDy Þ ¼ HomO SelðDy ; W Þ; F=O when c ¼ w 1 and c t ¼ c, i.e.,
2
the‘‘sign’’ in the functional equation of c (and hence of the modular form f associated to c)
makes sense.
1
Theorem 2.1. Assume that if c t ¼ c and c ¼ w 1, then the sign in the functional
2

equation of c is þ1. Then SelðDy ; TÞ ¼ 0 and X ðDy Þ is a torsion LðDy Þ-module. More-
~ ðDy Þ n ~ F
over there is an equality of ideals in L ~
O

char X ðDy Þ ¼ mðDy ; h Þ :
In this section, we will prove the first two claims and that the characteristic ideal of
X ðDy Þ divides the p-adic L-function using some Galois cohomology and the theory of
Euler systems as exposed in Rubin’s book [11]. The key point is to prove that the Euler
system employed, a twisted version of the elliptic unit Euler system, is nontrivial. This will
be achieved by using a theorem of Greenberg on the non-vanishing of L-functions of Grös-
sencharaktere of K. The next section, §3, will be devoted to the proof of the remaining di-
visibility. We use the two variable main conjecture together with a descent argument to
achieve this.
The next result shows what happens in the exceptional case:
1
Theorem 2.2. Assume that c t ¼ c, c ¼ w 1, and the sign in the functional equa-
2
tion of c is 1. Then X ðDy Þ is a LðDy Þ-module of rank 1. Moreover, there is an equality
~ ðDy Þ n ~ F
of ideals in L ~
O :

char X ðDy Þtors R ¼ ðL1 Þ:
Here R denotes a certain p-adic regulator, defined in §4. The rank statement of The-
orem 2.2 will be proved simultaneously with Theorem 2.1. In §4, we use the theory of
p-adic heights to prove the statement about the characteristic ideal.
2.2. Non-triviality of p-adic L-functions. In order for the general theorems of Rubin
[11] on Euler systems to provide non-trivial information, we need to prove the non-
vanishing over Dy of the Euler system ca constructed in Proposition 1.2. What we first
prove is the non-vanishing over Dy of the associated p-adic L-function. We would then
like to be able, roughly speaking, to fill in a diagram like
Col
~ Kðfpy Þ
H 1 Kðfpy Þp ; T ! L
? ?
? ?
ð2:1Þ ?
cory ?
y
H 1 ðDy; p ; TÞ ! ~ ðDy Þ

L
b?
in order to conclude the non-vanishing of the restriction of the Euler system to Dy on the
left from the non-vanishing of an appropriate p-adic L-function on the right. Such a map
~ . Our immediate goal, therefore, is to show:
will exist after extending scalars to F
1
Proposition 2.3. If c t ¼ c and c ¼ w 1, then mðDy ; h Þ is nonzero if and only if
2
the sign in the functional equation of c is þ1 and m ðDy ; h Þ is nonzero if and only if the sign
in the functional equation of c is 1. Otherwise both measures are nonzero.
Proof. Let f0 ¼ c=c. This is a Grössencharakter of type ðw 1; 1 wÞ and, as such,

its corresponding p-adic character f0; p factors through a finite extension of Dy . Choose a
positive integer m (which we may enlarge later on) such that f0m has trivial conductor and
ðf0m Þp factors through Dy . Set f ¼ f0m . We will apply the interpolation formulas (1.4) and
(1.5) to the characters fl (which have appropriate type for l g 0).

Since fl cNc , resp. fl cNc , has type ðlm þ 1Þðw 1Þ c; lmð1 wÞ c , resp.
lmðw 1Þ c; ðlm 1Þð1 wÞ c , we can apply equation (1.4), resp. (1.5), to get the
formula

ð2:2Þ flp mðDy ; h Þ ¼ ðÞLy; fp ðf1 c1 ; cÞ

¼ ðÞLy; fp c 2lmþ1 ; c þ 1 þ lmðw 1Þ ;
resp.

ð2:3Þ flp m ðDy ; h Þ ¼ ðÞLy; fp ðf1 c1 ; cÞ

¼ ðÞLy; fp c 2lm1 ; c þ 1 þ ðlm 1Þðw 1Þ ;
where for the second equalities in each formula we have used the functional equation and
the fact that cc ¼ N w1 (here ðÞ represents a nonzero constant consisting of gamma fac-
tors and Euler factors at p). In both cases, as c has type ðw 1; 0Þ, the given L-values are
1
central if and only if c ¼ w 1.
2
Equations (2.2) and (2.3) hold for all l g 0 (and in particular for infinitely many
characters). The proposition will then follow from work of Greenberg [4], Thm. 1, on the
(non)vanishing of classical L-values of Grössencharaktere of type ðn; 0Þ if we can show that
the sign in the functional equation of c 2lmþ1 , resp. c 2lm1 , is the same as, resp. the negative
of, the sign in the functional equation of c whenever the sign makes sense. To show this, we
use the following formula of Weil [14], p. 161, as stated in [1], Prop. 2.1.6:
Proposition 2.4. For two C -valued idèle class characters y1 , y2 of types ðk1 ; j1 Þ and
ðk2 ; j2 Þ, with relatively prime conductors f1 and f2 , and of absolute value 1, we have

W ðy1 y2 Þ if ðk1 j1 Þðk2 j2 Þ f 0;
W ðy1 ÞW ðy2 Þy1 ð f2 Þy2 ð f1 Þ ¼
ð1Þ n W ðy1 y2 Þ if ðk1 j1 Þðk2 j2 Þ < 0;
where W ðyi Þ is the root number of yi and n ¼ minfjk1 j1 j; jk2 j2 jg. r
We proceed as in [4], p. 83, making use of the fact that W ðyÞ ¼ W ðy=jyjÞ for any
Grössencharakter y. By the proposition, since c m has trivial conductor, we have that
W ðc mr Þ ¼ W ðc m Þ r for all r f 0. Moreover, the fact that c m is unramified everywhere
shows that W ðc m Þ is a root of unity. Thus by enlarging m we may assume that W ðc m Þ ¼ 1.
To deal with the c 2lmþ1 case, we take y1 ¼ c 2lm =jc 2lm j (which has trivial conductor)
and y2 ¼ c=jcj. In this case ðk1 j1 Þðk2 j2 Þ f 0. By choice of m, we have f1 ¼ 1, so
y2 ð f1 Þ ¼ 1, and Weil’s formula gives that
W ðc 2lmþ1 Þ ¼ W ðcÞal ;
where a ¼ c 2m ð f2 Þ=jc 2m ð f2 Þj. In the case we are concerned about (i.e., when c t ¼ c), we
have that f2 ¼ f 2 , which implies that a ¼ G1. Replacing m with 2m, we may assume that
a ¼ 1, which is what we want.
Finally, in the c 2lm1 case, we take y1 as before and y2 ¼ c1 =jc1 j. Then
ðk1 j1 Þðk2 j2 Þ < 0. Proceeding as above, we apply Weil’s formula to get
W ðc 2lm1 Þ ¼ W ðc1 Þ ¼ W ðcÞ ¼ W ðcÞ: r
2.3. Cohomological preparations. The following lemma will be used throughout the
w1
sequel, and explains our assumption that c 3 , which ensures that the hypotheses of
2
the lemma are satisfied for the characters h and h defined in 1.1.
Lemma 2.5. Suppose w ¼ yp is the p-adic character associated to a Grössencharakter

y of K of type ða; bÞ with a 3 b. Then wjGDy 3 1 and for any prime q of K, wjGK q 3 1. If
q j p, then moreover wjDy; v 3 1 for any place v of Dy lying over q.
Proof. If wjGDy ¼ 1, then w t ¼ w1 , which implies that a ¼ b. Suppose first that
q j p and choose an abelian extension L of K which is finite over Ky and over which w splits.
It follows from [2], Prop. II.1.9, e.g., that q is finitely decomposed in L, so GalðLv =K q Þ is a
finite-index subgroup of GalðL=KÞ (for any place v of L over q). Thus if w were to split over
Dy; v , some power of w would split over Dy , so again a ¼ b.
If q F p, then by raising w to a suitable power, we may assume that w is unramified at

q. Let p A OK be a generator of some power q d of q. Then, as a 3 b,
wðFrobqd Þ ¼ yðpÞ d ¼ ðp a p b Þ d 3 1;
so wjGK q 3 1. r
~,
As mentioned above, we cannot fill in diagram (2.1); however, after tensoring with F
we can prove
~ ðDy Þ n F
Proposition 2.6. For q ¼ p or p , there is an isomorphism of L ~ -modules
^ F
H 1 ðDy; q ; TÞ n ~ ðDy Þ n F
~ !L ~
which maps locq ca ðDy Þ to mðDy ; h ; aÞ if q ¼ p or to m ðDy ; h ; aÞ if q ¼ p .
~ ðDy Þ-module map induced by the Coleman map of Proposition

Proof. We use the L
1.2

H 1 Kðfpy Þq ; T n ~ ðDy Þ ! L
^ L ~ ðDy Þ;

where the tensor product is taken over the ring L ~ Kðfpy Þ . Thus we would like a map fill-
ing in the diagram
ColðÞ ~
H 1 Kðfpy Þq ; T n ~ ðDy Þ
^ L ! LðDy Þ
? !
?
?
y b?

^ O~
H 1 ðDy; q ; TÞ n
which we construct by proving that the left-hand map (the natural corestriction map) is an
~.
isomorphism, at least after tensoring with F
We break up the extension Kðfpy Þ=Dy into two pieces and deal with each separately.
First consider the finite extension Kðfpy Þ=Ky . The inflation-restriction sequence shows that
the restriction map
Q Q
res : H 1 ðKy; v ; W Þ ! H 1 Kðfpy Þv ; W
vjq vjq
has finite kernel and cokernel since there are only finitely many primes of Kðfpy Þ over p,
and is an isomorphism if p F ½KðfÞ : K . (Note p j ½KðfÞ : K if and only if p j ½Kðfpy Þ : Ky .)
Thus, by local duality, the map

H 1 Kðfpy Þq ; T nLðKðfpy ÞÞ LðKy Þ ! H 1 ðKy; q ; TÞ
has the same property.
Now we consider the Zp -extension Ky =Dy . We claim that the corestriction map
ð2:4Þ H 1 ðKy; q ; TÞ nLðKy Þ LðDy Þ ! H 1 ðDy; v ; TÞ
is injective with finite cokernel. If v is any of the finitely many places of Ky above q, we
have the local inflation-restriction sequence

0 ! H 1 Ky; v =Dy; v ; W ðKy; v Þ ! H 1 ðDy; v ; W Þ
! H 1 ðKy; v ; W Þ GalðKy; v =Dy; v Þ ! 0;
where the final map is surjective because GalðKy; v =Dy; v Þ G Zp has cohomological dimen-
sion 1. Set M ¼ W ðKy; v Þ, which is either finite or equal to W . By [11], Lemma B.2.8,
H 1 ðKy; v =Dy; v ; MÞ G M=ðg 1ÞM;
where g is a topological generator of GalðKy; v =Dy; v Þ. M=ðg 1ÞM is finite in the case M
is finite and trivial in the case M ¼ W , since g does not act trivially on W by Lemma 2.5.
Our claim thus follows from local duality.
^ F
Hence, if we apply n ~ to the map

H 1 Kðfpy Þq ; T nLðKðfpy ÞÞ LðDy Þ ! H 1 ðDy; q ; TÞ;
it becomes an isomorphism. Taking the inverse and composing with the appropriate Co-
leman map (i.e., Col or Col ) of Proposition 1.2 yields the desired map, a priori an in-
jection (since the Coleman maps are). Proposition 1.2 also describes the image of the
Coleman map asy Twh ðJÞ. Since KðhÞDy M Ky by Lemma 2.5, we can certainly find
g1 ; g2 A Gal Kðfp Þ=K such that hðg1 Þ 3 hðg2 Þ but g1 jDy ¼ g2 jDy . This shows that the
map is also surjective. r
2.4. Non-triviality of Euler systems. We can now use Proposition 2.6 to demonstrate
the nontriviality of the Euler systems we constructed in Proposition 1.2. After establishing
the relevant notation, we first prove Lemma 2.8, which will be used often in the sequel. Fix
an integral ideal a of K prime to fp and for any extension F L Ky of K, set
ca ðF Þ ¼ lim c ðF 0 Þ A H 1 ðF ; TÞ;
a
the inverse limit being taken over finite extensions F 0 L F of K.
Proposition 2.7. For any F L Ka , we have
ca ðF Þ A Sel S ðF ; TÞ:
Proof. This follows from Lemma 1.3 and [11], Lemma 2.3.5(ii). r
Hence the LðF Þ-submodule Ca ðF Þ of H 1 ðF ; TÞ generated by ca ðF Þ is contained

in Sel S ðF ; TÞ. Define CðF Þ L Sel S ðF ; TÞ to be the LðF Þ-submodule generated by
fca ðF Þ j ða; fpÞ ¼ 1g and set
Za ðF Þ ¼ Sel S ðF ; TÞ=Ca ðF Þ;
ZðF Þ ¼ Sel S ðF ; TÞ=CðF Þ:
Lemma 2.8. For q ¼ p or p , H 1 ðDy; q ; TÞ is torsion-free of rank 1 over LðDy Þ.
Proof. The proofs of Propositions 2.1.3 and 2.1.6 in Perrin-Riou’s article [7] work
without modification if the representation under consideration has coe‰cients in F rather
than Qp . Thus the former implies that H 1 ðDy; q ; TÞ is of rank 1, and the latter implies that,
for any place v of Dy over q, the torsion submodule of H 1 ðDy; v ; TÞ is naturally isomorphic
to TðDy; v Þ, which is trivial by Lemma 2.5. r
1
Proposition 2.9. If c t ¼ c and c ¼ w 1, then for q ¼ p, resp. p , the localization
2
locq CðDy Þ L H 1 ðDy; q ; TÞ is nonzero if and only if the sign in the functional equation of c is
þ1, resp. 1. Otherwise, locq CðDy Þ is always nonzero.
1
Proof. If c ¼ w 1 and the sign is þ1, Proposition 2.3 shows that mðDy ; h Þ 3 0.
2
Choosing a so that mðDy ; h ; aÞ 3 0 and applying Proposition 2.6, we see that
locp ca ðDy Þ 3 0. Thus locp CðDy Þ 3 0. To see that in this case locp CðDy Þ ¼ 0, note
that Proposition 2.3 implies that for any a, the measure m ðDy ; h ; aÞ vanishes. Thus Pro-
position 2.3, together with Lemma 2.8, shows that locp Ca ðDy Þ ¼ 0 for all a. A similar
1
argument applies when the sign is 1 or when c 3 w 1. r
2
2.5. Towards the main conjecture. After proving several preliminary results relating
the various Selmer groups we have defined, we will be in a position to apply the main re-
sults of [11] over the anticyclotomic extension Dy =K to show that the characteristic ideal
of X ðDy Þ is contained in the characteristic ideal of ZðDy Þ.
Proposition 2.10. Let F be a finite extension of K, v a prime of F not dividing p,

and V a F½GFv -module, finite as vector space over F. If V and V (Tate dual) both have
no GFv -invariants, then H 1 ðFv ; V Þ ¼ 0.
Proof. By [12], Lemma B.2.4, it su‰ces to show that H 1 ðFv ; TÞ is torsion provided
that TðFv Þ and T ðFv Þ are trivial. By [12], Lemma B.2.3, it therefore su‰ces to show that
H 1 ðFv ; T=p n TÞ is bounded independently of n. The triviality of the local Euler characteris-
tic [12], Ch. II, Prop. 17, together with local duality, shows that
KH 1 ðFv ; T=p n TÞ ¼ KðT=p n TÞðFv Þ KðT =p n T ÞðFv Þ;
which is bounded independently of n whenever TðFv Þ ¼ T ðFv Þ ¼ 0. r
The next lemma, in conjunction with Lemma 2.8, will be fundamental in extracting
information from the general results on Euler systems in [11], §2.3.
Lemma 2.11. There are canonical 5-term exact sequences
ð2:5Þ 0 ! SelðDy ; TÞ ! Sel S ðDy ; TÞ

loc p
! H 1 ðDy; p ; TÞ ! X ðDy Þ ! XS ðDy Þ ! 0
and
ð2:6Þ 0 ! SelS ðDy ; TÞ ! SelðDy ; TÞ

loc p
! H 1 ðDy; p ; TÞ ! X ; S ðDy Þ ! X ðDy Þ ! 0:
Proof. If we can show that Hf1 ðDy; p ; TÞ ¼ 0, then we will have
Hf1 ðDy; p ; W Þ ¼ H 1 ðDy; p ; W Þ
by local duality. Also, by global duality, the images of the maps

loc p
SelðDy ; TÞ ! H 1 ðDy; p ; TÞ;
loc p
SelðDy ; W Þ ! H 1 ðDy; p ; W Þ
are exact orthogonal complements (see [11], Thm. 1.7.3). Thus the exact sequences (2.5) and
(2.6) will follow from the definition of the Selmer groups.
To show that Hf1 ðDy; p ; TÞ ¼ 0, let v be a place of Dy above p. The fact that Dy; v
does not split V (Lemma 2.5) together with the cohomology sequence associated to
0!T !V !W !0
shows that, for any F L Dy; v finite over Kp ,
Hf1 ðF ; TÞ G H 0 ðF ; W Þ:
We want to show that taking the inverse limit (with respect to corestriction) over all such F
gives lim

H 1 ðF ; TÞ ¼ 0. The extension of Kp generated by W is unramified (recall that by
assumption p is coprime to the conductors of c and w), but Dy; v contains an infinite rami-
fied pro-p extension of Kp . Thus the intersection of Dy; v with Kp ðW Þ is finite over Kp . This
gives what we want, since Kp ðW Þ is not finite over Kp (the primes over p are finitely decom-
posed in Ky ). r
In order to make full use of the previous lemma (especially in the case c ¼ c t), we
will need to establish some relationship between the compact and discrete Selmer groups
(Proposition 2.13).
Lemma 2.12. Let S be the set of places of K which divide py or are ramified for T. If
KS =K is the maximal extension of K unramified outside of S, then for any finite extension
L=K contained in KS ,
Sel S ðL; TÞ ¼ H 1 ðKS =L; TÞ:
Proof. If v B S, then Hf1 ðLv ; TÞ ¼ Hur 1

ðLv ; TÞ by [11], Lemma 1.3.5, which
shows one inclusion. For v A S not dividing py, both V and V are ramified at v,
so that H 1 ðLv ; V Þ ¼ 0 by Proposition 2.10 (since Lv =Kv is unramified). Thus
Hf1 ðLv ; TÞ ¼ H 1 ðLv ; TÞ, which gives equality. r
Proposition 2.13. There is a canonical isomorphism of LðDy Þ-modules

Sel S ðDy ; TÞ ! HomLðDy Þ X S ðDy Þ; LðDy Þ :
Proof. This is [1], Lemma 1.1.9, the proof of which follows [7], Prop. 4.2.3. Let
L L Dy be a finite extension of K and let S be the set of places of K dividing fpy, where
f ¼ cond c. Note that S is also the set of places which are either ramified for T or divide
py. Let KS be the maximal extension of K unramified outside of S. Then Lemma 2.12
gives an equality
Sel S ðL; TÞ ¼ H 1 ðKS =L; TÞ ¼ lim

H 1 ðKS =L; T=p k TÞ:
k
The cohomology sequence associated to

pk
0 ! T=p k T ! W ! W ! 0
gives a surjective map H 1 ðKS =L; T=p k TÞ ! H 1 ðKS =L; W Þ½ p k whose kernel is
W ðLÞ=p k W ðLÞ. Define XS ðLÞ ¼ H 1 ðKS =L; W Þ4. Then the augmentation map LðLÞ ! O
induces an isomorphism

HomLðLÞ XS ðLÞ; LðLÞ G HomO XS ðLÞ; O :
Moreover, there is an isomorphism

HomO XS ðLÞ; O G lim

H 1 ðKS =L; W Þ½ p k :
k
Putting all this together gives a short exact sequence

0 ! lim

W ðLÞ=p k W ðLÞ ! H 1 ðKS =L; TÞ ! HomLðLÞ XS ðLÞ; LðLÞ ! 0:
k
Note that the size of lim

W ðLÞ=p k W ðLÞ ¼ W ðLÞ is finite and bounded independently of
L L Dy because Dy does not split W by Lemma 2.5. We claim that taking the inverse
limit over L L Dy gives an isomorphism

ð2:7Þ Sel S ðDy ; TÞ ! HomLðDy Þ XS ðDy Þ; LðDy Þ :
Indeed, if L 0 is large enough that W ðL 0 Þ ¼ W ðDy Þ, then for L L Dy a finite extension

of L 0 , the corestriction map W ðLÞ ! W ðL 0 Þ is multiplication by ½L : L 0 . It follows that
lim

W ðLÞ ! W ðL 0 Þ is the zero map for all L 0 su‰ciently large, so lim

W ðLÞ ¼ 0.
By [11], Lemma 1.3.5, we have Hf1 ðLv ; W Þ ¼ Hur

1
ðLv ; W Þ whenever v does not lie over
a place in S. Thus we get an exact sequence
L
ð2:8Þ 0 ! Sel S ðL; W Þ ! H 1 ðKS =L; W Þ ! H 1 ðLv ; W Þ:
vjf
By
L local Tate duality, the last term in this sequence is isomorphic to the Pontryagin dual of
H 1 ðLv ; T Þ. The Lv Galois cohomology sequence associated to
vjf
0 ! T ! V ! W ! 0
contains the exact sequence
V ðLv Þ ! W ðLv Þ ! H 1 ðLv ; T Þ ! H 1 ðLv ; V Þ:
The first term in the sequence is 0 since V is ramified at v and Lv =Kv is unramified, and the
last term is likewise 0 by Proposition 2.10, so the middle two terms are isomorphic.
Taking the Pontryagin dual of the sequence (2.8) and taking the direct limit over L
then gives us a sequence
L
W ðDy; v Þ ! XS ðDy Þ ! X S ðDy Þ ! 0:
vjf

The first term in this sequence is a torsion O-module
(since
W is) and a fortiori torsion
over LðDy Þ. Applying the functor HomLðDy Þ ; LðDy Þ then gives an isomorphism
G
HomLðDy Þ X S ðDy Þ; LðDy Þ ! HomLðDy Þ XS ðDy Þ; LðDy Þ :
Comparing this with (2.7) gives the proposition. r
The following theorem gives all of the information which we will obtain from apply-
ing the Euler system machinery directly over Dy .
Theorem 2.14. XS ðDy Þ and ZðDy Þ are torsion LðDy Þ-modules and there is a di-
visibility of characteristic ideals char XS ðDy Þ j char ZðDy Þ. Sel S ðDy ; TÞ is torsion-free of
rank 1 over LðDy Þ.
1
If c ¼ c t, c ¼ w 1, and the sign in the functional equation of c is 1, then
2
X ðDy Þ has rank 1 over LðDy Þ and SelðDy ; TÞ ¼ Sel S ðDy ; TÞ. Otherwise, X ðDy Þ is a
torsion LðDy Þ-module.
Proof. Choose a such that Ca ðDy Þ 3 0 (possible by Proposition 2.9). The fact that
XS ðDy Þ is a torsion LðDy Þ-module follows from [11], Thm. 2.3.2 (cf. the discussion pre-
ceding Lemma 1.3). The divisibility of characteristic ideals follows from [11], Thm. 2.3.3.
That ZðDy Þ is a torsion LðDy Þ-module will follow from Proposition 2.9 once we show that
Sel S ðDy ; TÞ is torsion-free of rank 1 over LðDy Þ. The fact that Sel S ðDy ; TÞ is torsion free
of positive rank follows from Propositions 2.13 and 2.9 (given that Proposition 2.7 holds).
The exact sequence (2.5), together with Lemma 2.8 and the fact that XS ðDy Þ is torsion,
shows that the rank of X ðDy Þ is at most 1.
Note that if X ðDy Þ has rank 1, then locp must be trivial, as H 1 ðDy; p ; TÞ is torsion-
1
free. If this occurs, Proposition 2.9 shows that we must have c t ¼ c, c ¼ w 1, and the
2
sign in the functional equation of c is 1, so locp is non-trivial. Thus, the exact sequence
(2.6) and Lemma 2.8 imply that X ; S ðDy Þ has rank 1. Otherwise, if X ðDy Þ is torsion, we
see (again from (2.6) and Lemma 2.8) that X ; S ðDy Þ has rank at most 1. Briefly inter-
changing the roles of T and T , p and p , we see that X S ðDy Þ has rank at most 1 as well.
Thus Proposition 2.13 implies that Sel S ðDy ; TÞ must have rank exactly 1.
The remainder of the proof will be a study of the exact sequences in Lemma 2.11
using Lemma 2.8 and Propositions 2.9 and 2.13. We first deal with the ‘‘otherwise’’ case,
1
i.e., assume that if c ¼ c t and c ¼ w 1, then the sign in the functional equation of c
2
is þ1. In this case, the map locp in (2.5) is nonzero by Proposition 2.9. Thus Lemma 2.8,
together with the fact that XS ðDy Þ is torsion, shows that X ðDy Þ is likewise torsion over
LðDy Þ.
1
We now turn to the case that c ¼ w 1, c ¼ c t, and the sign in the functional
2
equation of c is 1. We know that for some a, the class ca ðDy Þ (which lies in SelðDy ; TÞ
by Proposition 2.9) is not torsion, since its image under locp is nonzero and H 1 ðDy; p ; TÞ
is torsion-free of rank 1. But then Sel S ðDy ; TÞ=SelðDy ; TÞ is torsion, and so must be zero,
since it injects via locp into the torsion-free module H 1 ðDy; p ; TÞ. Hence locp is trivial and
X ðDy Þ has rank 1. r
3. Divisibility from descent
In this section, we prove the reverse divisibility of the first part of Theorem 2.14 and
determine the characteristic ideal of X ðDy Þ when the sign in the functional equation of c
is not 1. The method involves twisting Rubin’s 2-variable main conjecture [9] to give a
2-variable main conjecture for T over Ky . We descend from Ky to Dy to complete the
description of char XS ðDy Þ. This section closely follows Section 2.4 of [1].
3.1. Twisting the 2-variable main conjecture. In order to apply the main results of
Rubin’s paper [9], we first need to establish some notation.
Let Ly be an abelian extension of K which is finite over Ky and splits both V and
V . There is a decomposition GalðLy =KÞ G GalðKy =KÞ D, where D G GalðLy =Ky Þ is
finite. (This decomposition is not canonical if p j ½Ly : Ky .) For any F L Ly , define
M S ðF Þ, resp. MðF Þ, resp. MS ðF Þ, to be the maximal abelian p-extension of F unramified
outside of p, resp. outside of p, resp. everywhere.
For an extension F L Ly finite over K, let EðF Þ be the completion of the global units
OF n O. If F L Ly is infinite over K, set EðF Þ ¼ lim
EðF 0 Þ (the inverse limit being taken,
0
as always, over subextensions F L F finite over K). If a L K is an integral ideal prime to
fp, we define Ua ðF Þ L EðF Þ to be the submodule generated by the elliptic unit Euler system
fyði; aÞg discussed in 1.5. We further let UðF Þ be the submodule generated by all the Ua ðF Þ.
Lemma 3.1. There are O-module isomorphisms
a : H 1 ðLy ; F=OÞ ! H 1 ðLy ; W Þ;

1

b : lim

H L; Oð1Þ ! lim

H 1 ðL; TÞ
(the inverse limits being taken with respect to fields L L Ly finite over K) satisfying
a Twh ðlÞ ¼ l a and b Tw1
h ðlÞ ¼ l b for all l A LðLy Þ. Moreover, the restriction of
a gives an (O-module) isomorphism

a : Hom Gal M S ðLy Þ=Ly ; F=O ! Sel S ðLy ; W Þ;
and similarly for MðLy Þ and MS ðLy Þ. The restriction of b gives isomorphisms
b : EðLy Þ n O ! Sel S ðLy ; TÞ;

b : UðLy Þ n O ! CðLy Þ:
(See 2.4 for the definition of CðKy Þ.)
Proof. This is [1], Lemma 2.4.5. The existence of a and b is clear from the fact that

W and T are trivial as GLy -modules and isomorphic as groups to F=O and O, respectively
(note that the characters defining W and T are Tate duals of each other). In order to prove
the statements about a, it su‰ces by [11], Lemma 1.3.5, to check that the condition of being
unramified at a prime v of Ly is equivalent to being locally trivial at v. This is [11], Lemma
ur
B.3.3: Ly; v contains the unique unramified Zp -extension of Kv , so that GalðLy; v =Ly; v Þ has
1 ur
trivial pro-p part, whence H ðLy; v =Ly; v ; F=OÞ ¼ 0 (see the proof of [11], Lemma B.3.3, for
more details).
Turning our attention to b, we see that by the above and local duality, the local con-
ditions defining Sel S ðLy ; TÞ are the same as the unramified conditions, whence the first iso-
morphism. That b maps UðLy Þ n O onto CðLy Þ follows from the definition of our Euler
systems ca (cf. the proof of Proposition 1.2). r
This lemma allows us to apply the main results of Rubin’s paper [9] to get a
2-variable main conjecture for our representation W .
Theorem 3.2. X ðKy Þ is a torsion LðKy Þ-module, Sel S ðKy ; TÞ has rank 1 over
LðKy Þ, and there are equalities of ideals in LðKy Þ n F

char X ðKy Þ ¼ char H 1 ðKy; p ; TÞ=locp CðKy Þ
and
char XS ðKy Þ ¼ char ZðKy Þ:
(See 2.4 for the definition of ZðKy Þ.)
Proof. We view h as a character of the group GalðLy =KÞ G GalðKy =KÞ D, and
so write h ¼ k n , where n is a character of the finite group D. Taking D-invariants of
the various LðLy Þ-modules of Lemma 3.1 (and keeping in mind how we have defined the
action of LðLy Þ on the X s in 1.3) yields the following formulas:
S n
Tw1
h char Gal M ðLy Þ=Ly ¼ char X ; S ðKy Þ i ;
similarly for MðLy Þ and MS ðLy Þ, and

n
Tw1
h char EðLy Þ=UðLy Þ ¼ char ZðKy Þ i :
For the descent of Selmer groups from Ly to Ky , see the proof of Proposition 2.6. The
statement of the theorem is then a consequence of the usual 2-variable main conjecture for
K ([9], Thm. 4.1(i); also cf. [11], Thms. 3.3.1 and 3.3.2) via the twisting theorems of [11],
Ch. 6. r
3.2. Preparations for descent. Unfortunately, we will need several technical results
in order to extract information about our situation over Dy from Theorem 3.2, which deals
with the 2-variable situation over Ky . Recall that we have defined the ideal I L LðKy Þ to
be the kernel of the natural surjective map r : LðKy Þ ! LðDy Þ, so that I ¼ ðg 1Þ for any
topological generator g of GalðKy =Dy Þ.
Lemma 3.3. If q ¼ p or p , then the restriction map

H 1 ðDy; q ; W Þ ! H 1 ðKy; q ; W Þ
has finite kernel.
Proof. This map is dual to the map (2.4) in the proof of Proposition 2.6, the coker-
nel of which was proved to be finite. r
Lemma 3.4. The kernel of the restriction map

L L
H 1 ðDy; v ; W Þ ! H 1 ðKy; v ; W Þ
vjf vjf
has finite exponent.
Proof. By the inflation-restriction sequence, the kernel is

L
H 1 Ky; v =Dy; v ; W ðKy; v Þ :
vjf
Note that Ky; v =Dy; v is a Zp -extension, so by [11], Lemma B.2.8, this group is isomorphic
to
L
W ðKy; v Þ=ðg 1ÞW ðKy; v Þ;
vjf
where g is a topological generator of GalðKy; v =Dy; v Þ. Also, Ky; v =Kv is the unique unrami-
fied Zp -extension of Kv . By definition of f, W is ramified at v, so W ðKy; v Þ is finite. The
size of W ðKy; v Þ only depends on the prime of K over which v lies. r
Proposition 3.5. The Pontryagin dual of the restriction map
Sel S ðDy ; W Þ ! Sel S ðKy ; W Þ½I ;
i.e., the LðDy Þ-module homomorphism
X ; S ðKy Þ=IX ; S ðKy Þ ! X ; S ðDy Þ;
has finite cokernel and finite-exponent kernel. The same is true of X and XS .
In particular, XS ðDy Þ is a torsion LðDy Þ-module exactly when the quotient
XS ðKy Þ=IXS ðKy Þis.
Proof. Let S be the set of places of K dividing fpy and let KS be the maximal ex-
tension of K unramified outside S. The definition of Sel S gives a commutative diagram with
exact rows:
L
0 ! Sel S ðDy ; W Þ ! H 1 ðKS =Dy ; W Þ ! H 1 ðDy; v ; W Þ
? ? vjf ?
? ? ?
res?
y
?
y
?
y
L
0 ! Sel S ðKy ; W Þ½I ! H 1 ðKS =Ky ; W Þ½I ! H 1 ðKy; v ; W Þ:
vjf
The kernel of the middle vertical map is

H 1 Ky =Dy ; W ðKy Þ G W ðKy Þ=IW ðKy Þ;
which is finite: if Ky splits W , then (by Lemma 2.5) a topological generator g of

GalðKy =Dy Þ acts nontrivially on W , which is divisible, so W ðKy Þ=IW ðKy Þ ¼ 0.
Thus the restriction map (the leftmost vertical arrow) likewise has finite kernel.
By Lemma 3.4, the kernel of the right-hand vertical map is finite-exponent. Note
that the middle vertical map is surjective by the inflation-restriction sequence, as
GalðKy =Dy Þ G Zp has cohomological dimension 1. Thus (by the snake lemma), the coker-
nel of the restriction map also has finite exponent. Since the Pontryagin dual of a finite-
exponent module is likewise of finite exponent, we have proved the claim for X ; S .
To prove the claim for X , we use the commutative diagram
0 ! SelðDy ; W Þ ! Sel S ðDy ; W Þ ! H 1 ðDy; p ; W Þ

? ? ?
? ? ?
res?
y
?
y
?
y
0 ! SelðKy ; W Þ½I ! Sel S ðKy ; W Þ½I ! H 1 ðKy; p ; W Þ
with exact rows (cf. Lemma 2.11). The kernel of the middle vertical map is finite by the
above, as is the kernel of the right-hand vertical map (Lemma 3.3). Moreover, the cokernel
of the middle map is of finite exponent (again by the above), so that another application of
the snake lemma proves the claim for X . The proof for XS is the same using the diagram
0 ! SelS ðDy ; W Þ ! SelðDy ; W Þ ! H 1 ðDy; p ; W Þ

? ? ?
? ? ?
res?
y
?
y
?
y
0 ! SelS ðKy ; W Þ½I ! SelðKy ; W Þ½I ! H 1 ðKy; p ; W Þ
(again cf. Lemma 2.11). r
Following [1], we define the decent defect
D ¼ charLðDy Þ XS ðKy Þ½I :
Recall that we have denoted by r : LðKy Þ ! LðDy Þ the natural map.
Corollary 3.6. D is nonzero and

charLðDy Þ XS ðDy Þ ¼ D r charLðKy Þ XS ðKy Þ
as ideals in LðDy Þ.
Proof. By Theorem 2.14, XS ðDy Þ is a torsion LðDy Þ-module. The previous propo-
sition (Proposition 3.5) implies that XS ðKy Þ=IXS ðKy Þ is likewise torsion. The fact that D
is nonzero now follows from [9], Lemma 6.2(i), and the equality of characteristic ideals is
[9], Lemma 6.2(ii). r
3.3. Descent to the anticyclotomic tower. Using the results of 3.2 we descend the
twisted 2-variable main conjecture to the anticyclotomic tower. The ultimate goal will be
Proposition 3.8, of which the main theorem of §3, Theorem 3.9, is an easy consequence.
Recall that we have defined ZðF Þ ¼ Sel S ðF ; TÞ=CðF Þ, where CðF Þ is the module generated
by the Euler systems ca ðF Þ as a runs over ideals of K prime to fp (cf. 2.4).
Proposition 3.7. The LðDy Þ-module homomorphisms
Sel S ðKy ; TÞ=I Sel S ðKy ; TÞ ! Sel S ðDy ; TÞ;

ZðKy Þ=IZðKy Þ ! ZðDy Þ
are injective and the cokernels are torsion of characteristic ideal divisible by D.
Proof. This is essentially [1], Prop. 2.4.15. As before, let S be the set of places of K
dividing fpy and denote by KS the maximal extension of K unramified outside of S. For
any L L Ky finite over K, [8], Prop. 4.1.1 shows there is an exact sequence
L
0 ! XS ðLÞ ! H 2 ðKS =L; TÞ ! H 0 ðLv ; W Þ4;
v A SL
where SL is the set of places of L lying over places in S. Taking the inverse limit over L
yields
L 0
0 ! XS ðKy Þ ! lim
H 2 ðKS =L; TÞ ! H ðKy; v ; W Þ4:
v j fp
Shapiro’s Lemma gives

2 2

lim

H ðKS =L; TÞ G H KS =K; T n LðKy Þ ;
so taking I -torsion of the above exact sequence yields

L 0
0 ! XS ðKy Þ½I ! H 2 KS =K; T n LðKy Þ ½I ! H ðKy; v ; W Þ4½I :
v j fp
The final term of this sequence is dual to

L L
W ðKy; v Þ=IW ðKy; v Þ G H 1 Ky; v =Dy; v ; W ðKy; v Þ ;
v j fp v j fp
i.e., the kernel of the restriction map

L L
H 1 ðDy; v ; W Þ ! H 1 ðKy; v ; W Þ:
v j fp v j fp
This
kernel has finite exponent
by Lemmas 3.3 and 3.4, so that Corollary 3.6 shows that
H 2 KS =K; T n LðKy Þ ½I is torsion over LðDy Þ with characteristic ideal divisible by D.
Taking the Galois cohomology of the sequence

g1
0 ! T n LðKy Þ ! T n LðKy Þ ! T n LðDy Þ ! 0
gives the exact sequence

0 ! H 1 KS =K; T n LðKy Þ nLðKy Þ LðDy Þ

! H 1 KS =K; T n LðDy Þ ! H 2 KS =K; T n LðKy Þ ½I ! 0:
Thus, Lemma 2.12 and another application of Shapiro’s Lemma shows that the natural
map
Sel S ðKy ; TÞ=I Sel S ðKy ; TÞ ! Sel S ðDy ; TÞ
is injective with torsion cokernel of characteristic ideal divisible by D. The map
CðKy Þ=ICðKy Þ ! CðDy Þ
is surjective (since the elliptic units are universal norms in the cyclotomic direction), so the
Snake Lemma implies the statement for Z. r
Proposition 3.8. There is an equality of ideals in LðDy Þ
char XS ðDy Þ ¼ char ZðDy Þ:
Proof. The divisibility

char XS ðDy Þ j char ZðDy Þ
is part of Theorem 2.14. Proposition 3.7 implies ZðKy Þ=IZðKy Þ is a torsion LðDy Þ-
module (because ZðDy Þ is by Theorem 2.14). Thus [10], Lemma 6.2(i) shows that
ZðKy Þ½I is likewise a torsion LðDy Þ-module. An application of [10], Lemma 6.2(ii) gives
an equality of characteristic ideals

r charLðKy Þ ZðKy Þ charLðDy Þ ZðKy Þ½I ¼ charLðDy Þ ZðKy Þ=IZðKy Þ:
By Proposition 3.7, we may write the characteristic ideal of the cokernel of
ZðKy Þ=IZa ðKy Þ ! ZðDy Þ
as the product JD for some ideal J L LðDy Þ. Using the above, we then get the equality
charLðDy Þ ZðDy Þ ¼ JD charLðDy Þ ZðKy Þ=IZðKy Þ

¼ JDr charLðKy Þ ZðKy Þ charLðDy Þ ZðKy Þ½I :
The 2-variable main conjecture, Theorem 3.2, gives the equality

r charLðKy Þ ZðKy Þ ¼ r charLðKy Þ XS ðKy Þ ;
so that we may apply Corollary 3.6 to get
charLðDy Þ ZðDy Þ ¼ J charLðDy Þ XS ðDy Þ charLðDy Þ ZðKy Þ½I ;
which proves the remaining divisibility. r

We can now prove what one might refer to as the ‘‘anticyclotomic main conjecture’’
for Grössencharaktere of K, i.e., Theorem 2.1 and the rank statement of Theorem 2.2. Un-
fortunately, information about the associated ‘‘m-invariant conjecture’’ is lost in Proposi-
tion 2.6 (though the proof shows that this holds in some special cases: e.g., when p does
not divide ½KðfÞ : K and K q ðW ½pÞ 3 K q for q ¼ p and p , p a uniformizer of O).
1
Theorem 3.9. If c ¼ w 1, c ¼ c t, and the sign in the function equation of c is
2
1, then SelðDy ; TÞ is torsion free of rank 1 over LðDy Þ, X ðDy Þ has rank 1, and there is
an equality of characteristic ideals

char XS ðDy Þ ¼ char SelðDy ; TÞ=CðDy Þ :
Otherwise, X ðDy Þ is a torsion LðDy Þ-module and there is an equality of ideals in

~ ðDy Þ n ~ F
L ~
O

charLðDy Þ X ðDy Þ ¼ mðDy ; h Þ :
1
Proof. In the case that c ¼ w 1, c ¼ c t, and the sign in the function equation
2
of c is 1, the statements follow from Theorem 2.14 and Proposition 3.8.
In the remaining case, the fact that X ðDy Þ is torsion is again part of Theorem 2.14,
and we can compute its characteristic ideal by considering the exact sequence
0 ! ZðDy Þ ! H 1 ðDy; p ; TÞ=locp CðDy Þ ! X ðDy Þ ! XS ðDy Þ ! 0
coming from the sequence (2.5) of Lemma 2.11. Proposition 3.8 shows that the middle two
terms have the same characteristic ideals, so the theorem now follows from Proposition 2.6.
r
To complete the proof of Theorem 2.2, it remains to identify the characteristic ideal
of X ðDy Þtors in the case that the sign in the functional equation of c is 1. §4 will be de-
voted to showing that this characteristic ideal is generated by the linear term of the associ-
ated 2-variable p-adic L-function.
3.4. Application to Selmer groups over Q. Before proceeding to the proof of Theo-
rem 2.2, we explore a consequence of Theorem 3.9 towards a generalization to CM modular
forms of the Birch and Swinnerton-Dyer conjecture. This will follow from the following
‘‘control theorem’’.
Proposition 3.10. If X ðDy Þ has positive rank as a LðDy Þ-module, then X ðKÞ has
positive rank as an O-module.
Proof. This is a special case of Proposition 4.3. r
Remark. By Proposition 3.10, the natural map
X ðDy Þ=IX ðDy Þ ! X ðKÞ

has finite kernel and cokernel, where I is the augmentation ideal of LðDy Þ. Hence, if n
denotes the number of summands killed by a power of ðg 1Þ in a cyclic decomposition
(up to pseudo-isomorphism) of X ðDy Þ as a LðDy Þ-module, then we have the formula
rkO X ðKÞ ¼ n þ rkLðDy Þ X ðDy Þ:
As in 1.1, we let rf : GQ ! GL2 ðKf ; P Þ be the p-adic representation associated to a

P-ordinary newform f A Sw ðN; wÞ. Choose a stable lattice T in the representation space
V of rf ðcÞ and denote by W the discrete GQ -module V=T. We can then define the
Selmer groups SelðF ; WÞ for finite extensions F of Q as in 1.2, where the finite conditions
Hf1 ðFv ; WÞ are defined as in Greenberg’s article [5], for example. Finally, we denote by W t
the discrete GK -module ðF=OÞ n Ocpt ðcÞ.
Theorem 3.11. Let f be a CM modular form of even weight w with associated Grös-
1
sencharakter c. If c ¼ w 1, c ¼ c t, and the sign in the functional equation of f is 1
2
(so Lð f ; sÞ has odd order of vanishing at its central point), then SelðQ; WÞ4 has positive rank
as an O-module.
Proof. Recall that rf jGK ¼ cp l cpt , so
SelðK; WÞ ¼ SelðK; W Þ l SelðK; W t Þ:
Since rf is a representation of GQ , complex conjugation t A GQ acts on this Selmer group

1
by exchanging the summands. When c ¼ w 1, Theorem 3.9 and Proposition 3.10
2
imply that SelðK; W Þ4 has positive O-rank. Thus the GalðK=QÞ-invariant submodule
fðb; tbÞ j b A SelðK; W Þg of SelðK; WÞ is of positive O-corank. Since p is odd, we have
SelðK; WÞ GalðK=QÞ ¼ SelðQ; WÞ, which shows that SelðQ; WÞ also has positive O-corank,
as desired. r
4. p-adic heights and the linear term
Theorem 3.9 gives, in case the sign in the functional equation of c is not 1, a de-
scription of the characteristic ideal of X ðDy Þ in terms of a suitable p-adic L-function. In
the case that the sign is 1, we would like to similarly determine the characteristic ideal of
the torsion submodule X ðDy Þtors of X ðDy Þ. To do so, we must first determine the extent
to which X ðDy Þtors and XS ðDy Þ di¤er; this is achieved by using various duality results.
Properties of the p-adic height pairing then allow us to express this di¤erence in terms of
the linear term of the twisted Katz p-adic L-function of 1.4. In what follows, we assume
1
that c ¼ w 1, c ¼ c t, and the sign in the function equation of c is 1.
2
4.1. Duality. The main results of this section, Propositions 4.1 and 4.2, can be
thought of as p-adic functional equations for characteristic ideals of Selmer groups. The
proofs of both propositions use ideas of Greenberg (cf. [5], proof of Thm. 2) and follow
[1], Section 1.2. Recall our convention (made in 1.3) that LðDy Þ acts on the modules
X ; S ðDy Þ, X ðDy Þ, and XS ðDy Þ via the involution i.
Proposition 4.1. X ðDy Þ and X ðDy Þ have the same rank over LðDy Þ, and there is an
equality of characteristic ideals
i
char X ðDy Þtors ¼ char X ðDy Þtors ;
viewed as ideals in LðDy Þ n F (i.e., equality as ideals of LðDy Þ up to powers of a uniform-

izer of O).
Proposition 4.2. The LðDy Þ-rank of X S ðDy Þ is one greater than the LðDy Þ-rank of
XS ðDy Þ,
and there is an equality of characteristic ideals
i
char X S ðDy Þtors ¼ char XS ðDy Þ;
viewed as ideals in LðDy Þ n F.
We reduce the problem, via suitable control theorems, to one of comparing the Selmer
groups of finite GK -modules, and then use a formula of Wiles relating these groups. We con-
sider Selmer groups attached to twists of W and W by characters of G ¼ GalðDy =KÞ.
To be precise, suppose w : GalðDy =KÞ ! ðO 0 Þ is a continuous character, where O 0 is a
(fixed) extension of O in which some polynomial generators of the characteristic ideals of
X ðDy Þtors , X ðDy Þtors , X S ðDy Þtors , and XS ðDy Þtors split into linear factors. Given such a
w, let Ow0 be O 0 with the GK -module structure for which g A GK acts by multiplication by
wðgÞ. If M is an O½GK -module, we define MðwÞ ¼ M nO O 0 ðwÞ.
Given any such w, we define the Selmer groups associated to W ðwÞ and W ðw1 Þ ex-
actly as in 1.2 (i.e., using the analogous local conditions). Because the local conditions de-
fining the Selmer groups Sel S ðF ; W Þ and SelS ðF ; T Þ are not orthogonal under the Tate
pairing, we will need to define the auxiliary Selmer group

Sel 0 K; W ðwÞ L Sel S K; W ðwÞ

to be the set of classes t A Sel S K; W ðwÞ such that locq t lies in H 1 K q ; W ðwÞ div for q ¼ p
or p . Because w is a character of GalðDy =KÞ, we have an identification of LðDy Þ-
modules

H 1 Dy ; W ðwÞ ¼ H 1 ðDy ; W ÞðwÞ;
and similarly for W and for the local cohomology groups. This restricts to an identifica-
tion of Selmer groups

Sel Dy ; W ðwÞ ¼ SelðDy ; W ÞðwÞ;
and similarly for the other Selmer groups defined over Dy .
Proposition 4.3. The natural restriction homomorphism

H 1 K; W ðwÞ ! H 1 ðDy ; W ÞðwÞ G
induces maps

Sel 0 K; W ðwÞ ! Sel S ðDy ; W ÞðwÞ G ;

Sel K; W ðwÞ ! SelðDy ; W ÞðwÞ G ;

SelS K; W ðwÞ ! SelS ðDy ; W ÞðwÞ G ;
whose kernels and cokernels are finite, of cardinality bounded independently of w.
Remark. Of course, the same statements hold when W ðwÞ is replaced by W ðw1 Þ.
Proof. By the inflation-restriction sequence, the homomorphism

H 1 K; W ðwÞ ! H 1 ðDy ; W ÞðwÞ G
is surjective with kernel bounded by the size of W ðwÞðDy Þ. Since Dy does not split W (by
Lemma 2.5) and w is a character of GalðDy =KÞ, W ðwÞðDy Þ has finite order bounded in-
dependently of w.
We first show that the restriction map

Sel K; W ðwÞ ! SelðDy ; W ÞðwÞ G
has finite cokernel, and then explain how to modify the argument to prove the same is true
G
1
proposition. Any element c A SelðDy1;W ÞðwÞ is the
for the other maps of the restriction of
some d A H K; W ðwÞ , and any such d satisfies res locv d A Hf Dy; v ; W ðwÞ for any prime
v of Dy . As SelðDy ; W ÞðwÞ G is a cofinitely generated
Zp -module,
it therefore
1 su‰ces to
1 1
find a bound (uniform in w and v) for the index of Hf Kv ; W ðwÞ in res Hf ðDy; v ; W ÞðwÞ .
If v Fp , then Hf1 ðDy; v ; W ÞðwÞ ¼ 0, so we need to bound the size of

ker res ¼ H 1 Dy; v =Kv ; W ðwÞðDy; v Þ . First note that we may assume GalðDy; v =Kv Þ G Zp ,
as otherwise this Galois group is trivial. We first consider those v F p at which W ðwÞ is un-
ramified. For such v, either W ðwÞðDy; v Þ or W ðwÞðDy; v Þ ¼ W ðwÞ (depending on whether
W ðwÞ½pLj W ðwÞðK
v Þ or W ðwÞ½p L W ðwÞðKv Þ, respectively, for a uniformizer p of O),
since Kv W ðwÞ must contain Dy; v , the maximal unramified abelian p-extension of Kv . In
case W ðwÞðDy; v Þ ¼ W ðwÞ, we have

H 1 Dy; v =Kv ; W ðwÞðDy; v Þ ¼ W ðwÞ=ðg 1ÞW ðwÞ
for a topological generator g of GalðDy; v =Kv Þ. This group is trivial, since g acts by a non-
trivial scalar and W ðwÞ is divisible. Thus it remains to uniformly bound ker res for each
of the finitely many primes v of Dy such that either v j p or all of the following hold:
GalðDy; v =Kv Þ G Zp , W ðwÞ is ramified, and W ðwÞðDy; v Þ is finite. In either of these cases,

H 1 Dy; v =Kv ; W ðwÞðDy; v Þ ¼ W ðwÞðDy; v Þ=ðg 1ÞW ðwÞðDy; v Þ;
which is finite, bounded independently of w.

If v j p , then we consider the larger group Selfp g K; W ðwÞ M Sel K; W ðwÞ (defined
in the same way as the usual Selmer group, save that we pose no condition locally at p ).

We claim that Sel K; W ðwÞ has finite index in Selfp g K; W ðwÞ . From the definitions, we
see that the quotient of these two groups injects into

H 1 Kp ; W ðwÞ =H 1 Kp ; W ðwÞ div ;

which is finite, as H 1 Kp ; W ðwÞ is a cofinitely generated O-module. Thus we see that some
finite multiple of locp d lies in Hf1 Kp ; W ðwÞ , as desired.
The restriction

Sel 0 K; W ðwÞ ! Sel S ðDy ; W ÞðwÞ G
likewise has finite cokernel: if q ¼ p or p , then we argue as above that

H 1 K q ; W ðwÞ =H 1 K q ; W ðwÞ div
is finite, which gives what we want, since

Hf1 K q ; W ðwÞ ¼ H 1 K q ; W ðwÞ div :
The rest of the argument (i.e., for primes v F p) is the same.
Similarly, to show that the map

SelS K; W ðwÞ ! SelS ðDy ; W ÞðwÞ G
has finite cokernel, it su‰ces (given the above) to note that the kernel
W ðwÞðDy; p Þ=ðg 1ÞW ðwÞðDy; p Þ
of the restriction map over p is finite. r
The essential content of the proofs of Propositions 4.1 and 4.2 is a formula of Wiles
for the Selmer groups of finite Galois modules. The following lemma will allow us to relate
our Selmer groups to the Selmer groups of the associated finite Galois modules.
Lemma 4.4. The natural maps

Sel K; W ðwÞ½ p k ! Sel K; W ðwÞ ½ p k ;

Sel 0 K; W ðwÞ½ p k ! Sel 0 K; W ðwÞ ½ p k ;

SelS K; W ðwÞ½ p k ! SelS K; W ðwÞ ½ p k
are surjective with kernels of order bounded independently of w and k. The same is true for
W ðw1 Þ.
Remark. We define the various Selmer groups associated to W ðwÞ½ p k by the local
conditions induced from those defining the Selmer groups associated to W ðwÞ. Cf. Mazur-
Rubin [6], §1.1.
Proof. The short exact sequence

p k
0 ! W ðwÞ½ p k ! W ðwÞ ! W ðwÞ ! 0
gives rise to the exact sequence of cohomology groups

W ðwÞðKÞ ! H 1 K; W ðwÞ½ p k ! H 1 K; W ðwÞ ½ p k ! 0:
The fact that Dy does not split W (Lemma 2.5) implies that W ðwÞðDy Þ is finite, of order
bounded independently of w, so certainly the same holds for W ðwÞðKÞ. This proves the
statement about the kernels. The surjectivity follows from the definition of the local condi-
tions cutting out the Selmer groups associated to W ðwÞ½ p k ; cf. [6], Lemma 3.5.3. r
4 4.5. The di¤erence

Proposition 4 between the orders of the torsion submodules of
Sel K; W ðwÞ and Sel K; W ðw1 Þ is bounded as w varies. The same is true of the torsion
4 4
submodules of Sel 0 K; W ðwÞ and SelS K; W ðw1 Þ . Moreover, for any w,
4 4
rkO 0 Sel K; W ðwÞ ¼ rkO 0 Sel K; W ðw1 Þ ;
4 4
rkO 0 Sel 0 K; W ðwÞ ¼ 1 þ rkO 0 SelS K; W ðw1 Þ :
Proof. In view of Lemma 4.4, it su‰ces to show that

lengthO 0 Sel K; W ðwÞ½ p k lengthO 0 Sel K; W ðw1 Þ½ p k ¼ cðkÞ
and

lengthO 0 Sel 0 K; W ðwÞ½ p k lengthO 0 SelS K; W ðw1 Þ½ p k ¼ v 0 ðpÞk þ dðkÞ;
where v 0 ðpÞ is the O 0 -valuation of p and jcðkÞj and jdðkÞj are bounded independently of k
and w. From Wiles’s formula [6], Prop. 2.3.5, we get the formula

lengthO 0 Sel K; W ðwÞ½ p k lengthO 0 Sel K; W ðw1 Þ½ p k
¼ lengthO 0 W ðwÞ½ p k ðKÞ lengthO 0 W ðw1 Þ½ p k ðKÞ

P
lengthO 0 W ðwÞ½ p k ðKv Þ lengthO 0 Hf1 Kv ; W ðwÞ½ p k
v j py
and a similar formula relating Sel 0 and SelS . In the former case, this formula gives

lengthO 0 Sel K; W ðwÞ½ p k lengthO 0 Sel K; W ðw1 Þ½ p k

¼ cðkÞ þ lengthO 0 Hf1 Kp ; W ðwÞ½ p k lengthO 0 W ðwÞ½ p k ðCÞ;
whereas the latter case gives

lengthO 0 Sel 0 K; W ðwÞ½ p k lengthO 0 SelS K; W ðw1 Þ½ p k

¼ dðkÞ þ lengthO 0 Hf1 Kp ; W ðwÞ½ p k

þ lengthO 0 Hf1 Kp ; W ðwÞ½ p k lengthO 0 W ðwÞ½ p k ðCÞ
for functions cðkÞ and dðkÞ with the desired properties. The proposition follows. r
We now prove Proposition 4.1; the proof of Proposition 4.2 is similar. For ease
of notation, set XO 0 ¼ X ðDy Þ nO O 0 and XO;0 i ¼ X ðDy Þ i nO O 0 . Let m be the maxi-
mal ideal of O 0 . By our choice of O 0 , we have pseudo-isomorphisms of modules over
LO 0 ¼ LðDy Þ nO O 0 :
L mðzÞ
L eði; zÞ
XO 0 @ LO 0 = ðg 1Þ z l LOr 0 l ðO 0 -torsionÞ;
z A m i¼1
L nðzÞ
L f ð j; zÞ
XO;0 i @ LO 0 = ðg 1Þ z l LOs 0 l ðO 0 -torsionÞ;
z A m j¼1
where g is a topological generator of GalðDy =KÞ; only finitely many summands are non-
zero. Proposition 4.1 is the assertion that r ¼ s and that
mðzÞ
P nðzÞ
P
ð4:1Þ eði; zÞ ¼ f ð j; zÞ
i¼1 j¼1
for all z A m. Let A L m be the subset of all z such that eði; zÞ ¼ 0 for all 1 e i e mðzÞ and
f ði; zÞ ¼ 0 for all 1 e i e nðzÞ. For each z A A define a character wz : GalðDy =KÞ ! O 0 by
the rule wz ðgÞ ¼ ð1 þ zÞ1 .
To see that r ¼ s, note that

4
XO 0 = ðg 1Þ z XO 0 ¼ SelðDy ; W Þðwz Þ G
for any z A A. In particular,

4
r ¼ rkO 0 SelðDy ; W Þðwz Þ G :
Similarly,
G 4
s ¼ rkO 0 SelðDy ; W Þðw1
z Þ :
Propositions 4.3 and 4.5 then imply that r ¼ s.
We now prove (4.1). Let z A m be such that one of eði; zÞ or f ð j; zÞ is nonzero (for
some i or j). Choose any sequence fzk g L A converging to z, and for each k define vðkÞ to
be the valuation of z zk . Then
mðzÞ
P
lengthO 0 XO 0 = ðg 1Þ zk XO 0 tors ¼ vðkÞ eði; zÞ þ cðkÞ
i¼1
and
nðzÞ
P
lengthO 0 XO;0 i = ðg 1Þ zk XO;0 i tors ¼ vðkÞ f ð j; zÞ þ dðkÞ;
j¼1
where jcðkÞj and jdðkÞj are bounded independently of k. Applying Propositions 4.3 and 4.5,
we see that
mðzÞ
P c 0 ðkÞ nðzÞ
P d 0 ðkÞ
eði; zÞ þ ¼ f ð j; zÞ þ ;
i¼1 vðkÞ j¼1 vðkÞ
where jc 0 ðkÞj and jd 0 ðkÞj are bounded independently of k. Proposition 4.1 follows by letting
k ! y.
4.2. The p-adic height pairing. There exists, quite generally, a p-adic height pairing
on the Selmer groups associated to any suitably nice p-adic Galois representation of a num-
ber field. Following [1] and an article of Rubin [10], we prove a formula for how this height
pairing behaves when evaluated on classes arising from the Euler system of twisted elliptic
units constructed in 1.5. First we establish some notation.
Set G ¼ GalðDy =KÞ and G ¼ GalðCy =KÞ. We choose an isomorphism l : G G Zp

sending a chosen topological generator g of G to 1 A Zp . Recall from 1.4 the power series
expansions (for a prime to fp)
mðKy ; h ; aÞ ¼ La; 0 þ La; 1 ðg 1Þ þ La; 2 ðg 1Þ 2 þ ;
mðKy ; h Þ ¼ L 0 þ L1 ðg 1Þ þ L2 ðg 1Þ 2 þ ;
~ ðDy Þ and L 0 ¼ mðDy ; h Þ. Because we are assuming

where La; i and Li are elements of L
1
that c ¼ w 1, c ¼ c t and the sign in the function equation of c is 1, we see that
2
L 0 ¼ La; 0 ¼ 0
S
by Proposition 2.9. We write Dy ¼ Dn , where ½Dn : K ¼ p n and set Fn ¼ Dn Cy .
~ satisfying
^ F
Lemma 4.6. For each n there is a unique element bn A H 1 ðFn; p ; TÞ n
ðg 1Þb n ¼ locp ca ðFn Þ.
If an is the image of bn in H 1 ðDn; p ; TÞ n ~ , then the an are norm-compatible. The

^ F
1 ^ ~
induced element ay A H ðDy; p ; TÞ n F maps under the isomorphism of Proposition 2.6 to
La; 1 .
Proof. This is [1], Lemma 3.1.1. Given that La; 0 ¼ 0, it follows directly from Pro-
position 2.6 and the definition of La; 1 . r
Recall that local Tate duality provides, for each finite extension F of K, a pairing
h ; iFp : H 1 ðFp ; TÞ H 1 ðFp ; T Þ ! O
which becomes perfect after taking the quotient by the O-torsion submodule on each side.
For notational convenience, we set h ; in ¼ h ; iDn; p .
Theorem 4.7. For all n, there is a p-adic height pairing
hn : SelðDn ; TÞ SelðDn ; T Þ ! F;
canonically determined by l up to sign, satisfying:

(1) (bounded image) There is an integer k, independent of n, such that the image of hn
lies in pk O.
(2) (Galois-equivariance) For any a A SelðDn ; TÞ, b A SelðDn ; T Þ, and s A GalðDn =KÞ,
we have
hn ða s ; b s Þ ¼ hn ða; bÞ:
(3) (compatibility) For an A SelðDn ; TÞ and bnþ1 A SelðDnþ1 ; T Þ, we have

hn an ; corðbnþ1 Þ ¼ hnþ1 resðan Þ; bnþ1 ;
where res and cor are the restriction, resp. corestriction, maps with respect to Dnþ1 =Dn .
~)
(4) (height formula) If b A SelðDn ; T Þ, then (after extending scalars to F

hn ca ðDn Þ; b ¼ han ; locp bin :
This and the next subsection will be devoted to the proof of part (4) of this theorem
(parts (1)–(3) are proved in [7], §1.2). We first recall Perrin-Riou’s construction of the
height pairing and give Rubin’s proof of the height formula in 4.3. Those readers not con-
cerned with the proof are advised to proceed to 4.4. In what follows, we fix n and let
Lk ¼ Ck Dn .
Lemma 4.8. Let v be a place of Ly . Then the module of universal norms

T
Hf1 ðDn; v ; TÞ u ¼ cor Hf1 ðLk; v ; TÞ
has finite index in Hf1 ðDn; v ; TÞ which is bounded independently of v and n.
Proof. If v j p, then the definitions show that Hf1 ðDn; v ; TÞ is the image under the
connecting homomorphism of W ðDn; v Þ. The size of W ðDn; v Þ is bounded independently
of n since W generates an infinite unramified extension of Kp and Dy; v is a ramified Zp -
extension of Kp .
If v j p , then we have Hf1 ðLk; v ; TÞ ¼ H 1 ðLk; v ; TÞ, so it su‰ces by local duality to

bound the kernel of the restriction map
H 1 ðDn; v ; W Þ ! H 1 ðLy; v ; W Þ:

By the inflation-restriction sequence, this kernel is H 1 Ly; v =Dy; v ; W ðLy; v Þ , which
is isomorphic to W ðLy; v Þ=ðg 1ÞW ðLy; v Þ, where g is a topological generator of
GalðLy; v =Dn; v Þ. Using the exact sequence
g1
0 ! W ðDn; v Þ ! W ðLy; v Þ ! W ðLy; v Þ
! W ðLy; v Þ=ðg 1ÞW ðLy; v Þ ! 0;
we see that W ðLy; v Þ=ðg 1ÞW ðLy; v Þ has the same size as W ðDn; v Þ, which is bounded
independent of n, since, as above, Dy; v is a ramified Zp -extension of Kp .
Finally, suppose v F p. The fact that V and V are not split locally at v (Lemma
2.5) together with Proposition 2.10 shows that H 1 ðLk; v ; V Þ ¼ 0, so we again have
Hf1 ðLk; v ; TÞ ¼ H 1 ðLk; v ; TÞ. As above, we therefore must bound the order of
W ðLy; v Þ=ðg 1ÞW ðLy; v Þ independently of v and n, where g is a topological generator
of GalðLy; v =Dn; v Þ. Because v F p, Ly; v is the unique unramified Zp -extension of Kv
and in particular does not change as n varies, so that any bound for the order of
W ðLy; v Þ=ðg 1ÞW ðLy; v Þ will be independent of n. If v is ramified for W , then
W ðLy; v Þ is finite, and its order only depends on the prime of K over which v lies, of which
there are finitely many. If Kv ðW ½pÞ 3 Kv where p is a uniformizer of O, then clearly
W ðLy; v Þ ¼ 0, since ½Kv ðW ½pÞ : Kv is prime to p. Thus we may assume that v is unrami-
fied for W and that KðW ½pÞ ¼ K. But in this case, we have W ðLy; v Þ ¼ W , since W
generates an unramified abelian p-extension of Kv . Thus, W =ðg 1ÞW ¼ 0 since g acts
nontrivially on W , which is divisible. r
The importance of the previous lemma is that it su‰ces to define hn ða; bÞ for
a A SelðDn ; TÞ and b A SelðDn ; T Þ which are everywhere locally contained in Hf1 ðDn; v ; TÞ u ,
resp. Hf1 ðDn; v ; T Þ u . In what follows, we fix such a and b.
We may view b as an element of H 1 ðDn ; T Þ, which corresponds to an extension of

GDn -modules
0 ! T ! Mb ! O ! 0:
Taking the Tate dual gives the sequence
ð4:2Þ 0 ! Oð1Þ ! Mb ! T ! 0:
For any finite extension L of Dn and any place v of L, we have a commutative diagram with
exact rows
pL dL
H 1 L; Oð1Þ ! H 1 ðL; Mb Þ ! H 1 ðL; TÞ ! H 2 L; Oð1Þ
? ? ? ?
? ? ? ?
(4.3) ? ? ? ?
y y y y
1
1 pL v 1 dL v 2

H Lv ; Oð1Þ ! H ðLv ; Mb Þ ! H ðLv ; TÞ ! H Lv ; Oð1Þ :
Lemma 4.9. If L is a finite Galois extension of Dn and a 0 A H 1 ðL; TÞ satisfies

corða Þ ¼ a, then a 0 is in the image of pL .
0
Proof. The connecting homomorphism dL is given up to sign by applying W resðbÞ,

where res is restriction from Dn to L. For any place v of L, we have

locv dL ða 0 Þ ¼ locv a 0 W resðbÞ ¼ locv ða W bÞ ¼ 0;
since a and b are orthogonal everywhere under the local Tate pairing. This shows that
dL ða 0 Þ is trivial locally everywhere, hence trivial. r
Lemma 4.10. If v is any place of L, then Hf1 ðLv ; TÞ is contained in the image of
pL v .
Proof. As in Lemma 4.9, dLv is given by W locv resðbÞ, but locv resðbÞ lies in
Hf1 ðLv ; T Þ, which is orthogonal to Hf1 ðLv ; TÞ. r
l
By class field theory, the map GalðLy =Dn Þ ! Zp ,! O gives rise to a map
rv : D
n; v n O ! O for every place v of Dn . Kummer theory then allows us to view rv as a
homomorphism

rv : H 1 Dn; v ; Oð1Þ ! O:
This map can also be described up to sign as

ð4:4Þ W locv l : H 1 Dn; v ; Oð1Þ ! H 2 Dn; v ; Oð1Þ G O;
where we view l as an element of H 1 ðDn ; OÞ.
Setting L ¼ Dn and a 0 ¼ a in Lemma 4.9 allows us to choose a sequence of elements

xk A H 1 ðDn ; Mb Þ such that pDn ðxk Þ ¼ a. Since we are assuming that a is a universal norm
locally everywhere, we may choose, for all places v of Ky and positive integers k, elements
yk; v A Hf1 ðLk; v ; TÞ satisfying cor yk; v ¼ locv a. Lemma 4.10 then gives the existence of
xk; v A H 1 ðLk; v ; Mb Þ such that pLk; v ðxk; v Þ ¼ yk; v . From (4.3), the di¤erence locv xk cor xk; v
is the image of some wk; v A H 1 Dn; v ; Oð1Þ , and we define the p-adic height pairing as
P
ð4:5Þ hn ða; bÞ ¼ lim rv ðwk; v Þ:
k v
It can be checked that this definition makes sense and is independent of the choices made
(see 1.2.4 of Perrin-Riou’s article [7]).
4.3. The height formula. What follows is Rubin’s proof of the height formula, The-
orem 4.7(4), as given in [10]. We have included a fairly complete proof, as it might not be
immediately clear why the proof in [10] works in our more general setting. In the notation
of the previous subsection, we will be considering a ¼ ca ðDn Þ (where ca is the Euler system
of 1.2) and computing its pairing against an arbitrary b A SelðDn ; T Þ which is a universal
norm locally everywhere. The formula for hn ða; bÞ can be given a particularly simple form:
Proposition 4.11. For suitable choices of xk , xk; v , and yk; v , the height pairing on
a ¼ ca ðDn Þ can be computed as

hn ða; bÞ ¼ lim rp ðwk; p Þ þ rp ðwk; p Þ :
k
Proof. Let a ¼ ca ðLk Þ, so that cor aðkÞ ¼ a. As above, Lemma 4.9 provides us with
ðkÞ
xðkÞ A H 1 ðLk ; Mb Þ mapping to aðkÞ under pLk . We define the sequence xk by xk ¼ cor xðkÞ .
If v j p, then choose xk; v A H 1 ðLk; v ; Mb Þ satisfying pLk; v ðxk; v Þ ¼ locv ak yk; v , where
yk; v A Hf1 ðLk; v ; TÞ as before satisfies cor yk; v ¼ locv a. We then define xk; v ¼ locv xðkÞ xk; v ,
so pLk; v ðxk; v Þ ¼ yk; v A Hf1 ðLk; v ; TÞ by the commutativity of (4.3). With this definition of
xk; v , the image of wk; q in H 1 ðDn; q ; Mb Þ is equal to cor xk; q (q ¼ p or p ).
If v F p, we take
P
xk; v ¼ locv ðxðkÞ Þ s ;
sAB
where B is a set of coset representatives for GalðLk =Dn Þ=Dv and Dv is a decomposi-
tion group at v. Then pLk; v ðxk; v Þ A Hf1 ðLk; v ; TÞ because Hf1 ðLk; v ; TÞ ¼ H 1 ðLk; v ; TÞ and
cor pLk; v ðxk; v Þ ¼ pLk ðlocv xk Þ ¼ locv a, so this choice of xk; v satisfies the requisite properties.
The stated formula for the height pairing now follows from the definition and the fact
that for v F p, we have cor xk; v ¼ locv xk , so these terms do not contribute to the sum (4.5)
defining hn ða; bÞ. r
For q ¼ p or p , define the module Hk; q as the kernel of the corestriction map
cor
H 1 ðLk; q ; TÞ ! H 1 ðDn; q ; TÞ:
Note in particular that, by definition, pLk; q ðxk; q Þ A Hk; q . In §4 of Rubin’s article [10], one
finds the definition of a ‘‘derivative’’ mapping
Derk; q : Hk; q ! H 1 ðDn; q ; T=p k TÞ:
(See
P also the introductory remarks of [10], §5.) In k Rubin’s notation, this is the map
DerLk; v =Dn; v ; ck; v , where ck; v : GalðLk; v =Dn; v Þ ! O=p O is induced by l. We now describe
vjq
several properties that the derivative map enjoys. Let

invq : H 2 Dn; q ; ðO=p k OÞð1Þ ! O=p k O
be the sum of the local invariant maps for v j q.
Proposition 4.12. With the choices made above, there is an equivalence up to sign

rq ðwk; q Þ 1 invq Derk; q pLk; q ðxk; q Þ W locq b ðmod p k Þ:
Proof. [10], Prop. 4.3 gives the formula

ðlocq lÞ W wk; q ¼ dDn; q Derk; q pLk; q ðxk; q Þ ;
where in the notation of that proposition we are taking wk; q ¼ corK=k ðcÞ and pLk; q ðxk; q Þ ¼ d.
Applying invq to both sides of this formula gives the result in light of our description (4.4)
for rq . r
Corollary 4.13. The height pairing on a ¼ ca ðDn Þ can be computed as
hn ða; bÞ ¼ lim rp ðwk; p Þ:

k
Proof. By definition of the Selmer group SelðDn ; T Þ, locp b lies in H 1 ðDn ; T Þtors ,
say p m locp b ¼ 0. The proposition implies that rp ðwk; p Þ 1 0 ðmod p km Þ for all k, so
limk rp ðwk; p Þ ¼ 0. r
Lemma 4.14. If dk A Hk; p is a sequence satisfying dk W z ¼ ðlocp ak Þ W z for

all k and all z A H 1 ðLk; p ; T Þ and ek A Hf1 ðLk; p ; T Þ is a sequence satisfying
cor ek ¼ e0 A Hf1 ðDn; p ; T Þ for all k, then
P
lim invp Derk; p ðdk Þ W e0 ¼ lim lðsÞhlocp ak ; sek iLk; p :
k k s A G=p k G
Proof. It su‰ces to show that

P
invp Derk; p ðdk Þ W e0 1 lim lðsÞhlocp ak ; sek iLk; p ðmod p k Þ:
k s A G=p k G
[10], Lemma 4.2 asserts the commutativity of the diagram
Hk; p K! H 1 ðLk; p ; TÞ

? ?
? ?
Derk; p ?
y
? T lðsÞs1
ys A G=p k G
H 1 ðDn; p ; T=p k TÞ ! H 1 ðLk; p ; T=p k TÞ
res
P
so in particular res Derk; p dk ¼ lðsÞs1 dk . Applying first W ek and then invp to this equa-
tion gives the lemma. r
We can now prove the height formula, Theorem 4.7(4). In the above lemma, take
dk ¼ pLk; p ðxk; p Þ and let ek A Hf1 ðLk; p ; T Þ be such that cor ek ¼ locp b. Recall the definition
of b n A H 1 ðLy; p ; TÞ from Lemma 4.6. We denote by b nðkÞ the image of bn in H 1 ðLk; p ; TÞ. In
particular, b ð0Þ 1
n ¼ an A H ðDn; p ; TÞ. We have
hn ða; bÞ ¼ lim rp ðwk; p Þ by Corollary 4:13

k
P
¼ lim lðsÞhlocp ak ; sek iLk; p by Proposition 4:12 and Lemma 4:14
k s A G=p k G
k
pP 1
¼ lim hðg 1Þb nðkÞ ; ig i ek iLk; p by Lemma 4:6
k i¼0
k
¼ lim hb nðkÞ ; res locp biLk; p since lim hb nðkÞ ; p k g p 1
ek i ¼ 0
k k
¼ han ; locp bin ;
which is exactly Theorem 4.7(4).
4.4. The characteristic ideal. We now have at our disposal the p-adic height pairing
of Theorem 4.7. The properties of the pairing together with the duality results of 4.1 render
the proof of Theorem 2.2 for the most part formal. First, however, we must define an
Iwasawa-theoretic version of the pairing.
Following [1], we define an Iwasawa-theoretic Tate pairing

1 i ^ ~

ð4:6Þ h ; iy : H 1 ðDy; p ; TÞ n ~ n~
^ F ~
LðDy Þ H ðDy; p ; T Þ n F ! LðDy Þ
by setting
P
hay ; by iy ¼ lim

hans ; bn in s1 :
s A GalðDn =KÞ
This pairing is in fact an isomorphism. We similarly define the Iwasawa-theoretic p-adic

height pairing
i ^ ~

~ n~
^ F ~
hy : SelðDy ; TÞ n LðDy Þ SelðDy ; T Þ n F ! LðDy Þ
by setting
P
hy ðay ; by Þ ¼ lim

hy ðans ; bn Þs1 :
s A GalðDn =KÞ
~ of Lemma 4.6 satisfies the equa-

^ F
Lemma 4.15. The element ay A H 1 ðDy; p ; TÞ n
tion

hy ca ðDy Þ; by ¼ hay ; locp by iy
for every element by A SelðDy ; T Þ.
Proof. This follows from the definitions of the Iwasawa-theoretic pairings and the
height formula, Theorem 4.7(4). r
Let R be the characteristic ideal of the cokernel of hy . Recall that, under our assump-
tions, the image of ca ðDy Þ in H 1 ðDy; p ; TÞ is trivial (Proposition 2.9), so the exact sequence
(2.5) of Lemma 2.11 shows that ca ðDy Þ A SelðDy ; TÞ; thus it makes sense to define Ra as
the characteristic ideal of the cokernel of
hy jCa ðDy Þ : Ca ðDy Þ nL~ ðDy Þ SelðDy ; T Þ i ! L

~ ðDy Þ:
We additionally define the ideal H L LðDy Þ by

H ¼ char H 1 ðDy; p ; T Þ=locp SelðDy ; T Þ :
Note that H 3 0, since locp is nontrivial (Proposition 2.9 applied to T ).
~ ðDy Þ n ~ F
Lemma 4.16. As ideals of L ~
O ,
Ra ¼ R char SelðDy ; TÞ=Ca ðDy Þ ¼ ðLa; 1 ÞH i :
Proof. The first equality is clear from the definitions. We show that the first term is
the same as the last. The height formula, Theorem 4.7(4), shows that

Imðhy jCa ðDy Þ Þ ¼ ay ; locp SelðDy ; T Þ n ~ i :
^ F
y
The result then follows from the fact that the Tate pairing (4.6) is an isomorphism and the
property of ay described in Lemma 4.6. r
The following, together with Theorem 3.9, completes the proof of Theorem 2.2.
~ ðDy Þ n ~ F
Theorem 4.17. There is an equality of ideals in L ~
O :

char X ðDy tors Þ R ¼ ðL1 Þ:
Proof. Interchanging T and T , p and p , in the exact sequence (2.6) of Lemma 2.11
yields the short exact sequence
0 ! H 1 ðDy; p ; T Þ=locp SelðDy ; T Þ ! X S ðDy Þ ! X ðDy Þ ! 0:
The leftmost term in this sequence is torsion of characteristic ideal H, so that we get the
equality
H char X ðDy Þtors ¼ char X S ðDy Þtors :
Thus, Propositions 4.1 and 4.2 give
H i char X ðDy Þtors ¼ char XS ðDy Þ;

but char XS ðDy Þ ¼ char SelðDy ; TÞ=CðDy Þ (this is part of Theorem 3.9), so the theorem
now follows from Lemma 4.16 by letting a vary. r
References
[1] A. Agboola and B. Howard, Anticyclotomic Iwasawa theory of CM elliptic curves, Ann. Inst. Fourier, to
appear.
[2] E. de Shalit, Iwasawa theory of elliptic curves with complex multiplication, Academic Press, 1987.
[3] P. Deligne, Formes modulaires et représentations l-adiques, Séminaire Bourbaki 1968/1969, exp. 355, Lect.
Notes Math. 179, Springer-Verlag, Berlin (1971), 139–172.
[4] R. Greenberg, On the critical values of Hecke L-functions for imaginary quadratic fields, Invent. Math. 79
(1985), 79–94.
[5] R. Greenberg, Iwasawa theory for p-adic representations, Algebraic number theory, Adv. Stud. Pure Math.
17, Academic Press, Boston, MA (1989), 97–137.
[6] B. Mazur and K. Rubin, Kolyvagin systems, Mem. Amer. Math. Soc. 168 (2004), no. 799.
[7] B. Perrin-Riou, Théorie d’Iwasawa et hauteurs p-adiques, Invent. Math. 109 (1992), 137–185.
[8] K. Ribet, Galois representations attached to eigenforms with Nebentypus, Modular functions of one variable,
V (Proc. Second Internat. Conf., Univ. Bonn, Bonn 1976), Lect. Notes Math. 601, Springer-Verlag, Berlin
(1977), 17–51.
[9] K. Rubin, The ‘‘main conjectures’’ of Iwasawa theory for imaginary quadratic fields, Invent. Math. 103
(1991), 25–68.
[10] K. Rubin, Abelian varieties, p-adic heights and derivatives, Algebra and number theory, Walter de Gruyter
and Co. (1994), 247–266.
[11] K. Rubin, Euler systems, Princeton University Press, 2000.
[12] J. P. Serte, Cohomologie galoisienne, fifth ed., Lect. Notes Math. 5, Springer-Verlag, Berlin 1994.
[13] J. Tilouine, Sur la conjecture principale anticyclotomique, Duke Math. J. 59 (1989), no. 3, 629–673.
[14] A. Weil, Dirichlet series and automorphic forms, Lect. Notes Math. 189, Springer-Verlag, 1971.
Department of Mathematics, University of Michigan, 530 Church Street, Ann Arbor, MI 48109-1043, USA
e-mail: tsarnold@umich.edu
Eingegangen 11. Oktober 2005, in revidierter Fassung 14. Februar 2006

Anticyclotomic Main Conjectures

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Anticyclotomic Main Conjectures

Uploaded by

Copyright:

Available Formats

J. reine angew. Math.

606 (2007), 41—78 Journal für die reine und

Anticyclotomic main conjectures

Abstract. Generalizing work of Agboola-Howard on CM elliptic curves, we prove

Much of the di‰culty involved in generalizing the methods of Agboola-Howard lies

1. Notation and deﬁnitions

Let Q be the algebraic closure of Q in C. We view algebraic extensions of Q as being

1.1. CM forms and representations. Let f A Sw ðN; wÞ be a normalized newform of

ð1:1Þ trace rf ðFrobl Þ ¼ al ð f Þ; det rf ðFrobl Þ ¼ l w1 wðlÞ;

Recall that f is said to be P-ordinary for a prime P j p of Kf if ap ð f Þ is a unit at P. If

We say that f has complex multiplication, or is a CM form, if there is a quadratic

If f is P-ordinary for P j p, then p splits in K: cp is ramiﬁed at p, whereas cpt is

In what follows, we assume chosen a prime P F 6N at which f is ordinary and let p be

1.2. Selmer groups. Let F be a ﬁnite extension of K. Following Greenberg [5],

The exact sequence

gives the sequence of cohomology groups (exact in the middle)

H 1 ðFv ; TÞ ! H 1 ðFv ; V Þ ! H 1 ðFv ; W Þ:

where we view LðF Þ as an algebra over itself via i.

If M ¼ T or T and F is as above, we deﬁne

where q is a prime of K. Tate local duality gives a perfect pairing

H 1 ðFq ; TÞ  H 1 ðFq ; W Þ ! F=O

Denote by Cy , Dy , and Ky the cyclotomic Zp -extension, the anticyclotomic Zp -

there is an interpolation formula

In the case that L ¼ Ky , if we choose a topological generator g of GalðKy =Dy Þ, we

mðKy ; h ; aÞ ¼ La; 0 þ La; 1 ðg 1Þ þ La; 2 ðg 1Þ 2 þ ;

If q ¼ p or p and F is a ﬁnite abelian extension of K, we deﬁne

By Kummer yði; aÞ ¼ NmKðifpÞ=KðiÞ Yðifp; aÞ as being an

such that if we deﬁne z, resp. z , to be the local image of

2. Divisibility from Euler systems

The next result shows what happens in the exceptional case:

H 1 ðDy; p ; TÞ ! ~ ðDy Þ

Proof. Let f0 ¼ c=c. This is a Grössencharakter of type ðw 1; 1 wÞ and, as such,

where W ðyi Þ is the root number of yi and n ¼ minfjk1 j1 j; jk2 j2 jg. r

W ðc 2lm1 Þ ¼ W ðc1 Þ ¼ W ðcÞ ¼ W ðcÞ: r

Lemma 2.5. Suppose w ¼ yp is the p-adic character associated to a Grössencharakter

If q F p, then by raising w to a suitable power, we may assume that w is unramiﬁed at

which maps locq ca ðDy Þ to mðDy ; h ; aÞ if q ¼ p or to m ðDy ; h ; aÞ if q ¼ p .

~ ðDy Þ-module map induced by the Coleman map of Proposition

has the same property.

ð2:4Þ H 1 ðKy; q ; TÞ nLðKy Þ LðDy Þ ! H 1 ðDy; v ; TÞ

! H 1 ðKy; v ; W Þ GalðKy; v =Dy; v Þ ! 0;

H 1 ðKy; v =Dy; v ; MÞ G M=ðg 1ÞM;

the inverse limit being taken over ﬁnite extensions F 0 L F of K.

Proposition 2.7. For any F L Ka , we have

Hence the LðF Þ-submodule Ca ðF Þ of H 1 ðF ; TÞ generated by ca ðF Þ is contained

ZðF Þ ¼ Sel S ðF ; TÞ=CðF Þ:

Lemma 2.8. For q ¼ p or p , H 1 ðDy; q ; TÞ is torsion-free of rank 1 over LðDy Þ.

Proposition 2.10. Let F be a ﬁnite extension of K, v a prime of F not dividing p,

KH 1 ðFv ; T=p n TÞ ¼ KðT=p n TÞðFv Þ KðT =p n T ÞðFv Þ;

which is bounded independently of n whenever TðFv Þ ¼ T ðFv Þ ¼ 0. r

Lemma 2.11. There are canonical 5-term exact sequences

ð2:5Þ 0 ! SelðDy ; TÞ ! Sel S ðDy ; TÞ

ð2:6Þ 0 ! SelS ðDy ; TÞ ! SelðDy ; TÞ

Proof. If we can show that Hf1 ðDy; p ; TÞ ¼ 0, then we will have

Hf1 ðDy; p ; W Þ ¼ H 1 ðDy; p ; W Þ

by local duality. Also, by global duality, the images of the maps

shows that, for any F L Dy; v ﬁnite over Kp ,

Proof. If v B S, then Hf1 ðLv ; TÞ ¼ Hur 1

Proposition 2.13. There is a canonical isomorphism of LðDy Þ-modules

Sel S ðL; TÞ ¼ H 1 ðKS =L; TÞ ¼ lim

The cohomology sequence associated to

Moreover, there is an isomorphism

Putting all this together gives a short exact sequence

Note that the size of lim

H 1 ðFq ; TÞ H 1 ðFq ; W Þ ! F=O

ð2:5Þ 0 ! SelðDy ; TÞ ! Sel S ðDy ; TÞ

ð2:6Þ 0 ! SelS ðDy ; TÞ ! SelðDy ; TÞ

Sel S ðDy ; W Þ ! Sel S ðKy ; W Þ½I ;

0 ! SelðDy ; W Þ ! Sel S ðDy ; W Þ ! H 1 ðDy; p ; W Þ

0 ! SelS ðDy ; W Þ ! SelðDy ; W Þ ! H 1 ðDy; p ; W Þ

D ¼ charLðDy Þ XS ðKy Þ½I :

charLðDy Þ ZðDy Þ ¼ J charLðDy Þ XS ðDy Þ charLðDy Þ ZðKy Þ½I ;