original citation - university of...
TRANSCRIPT
http://wrap.warwick.ac.uk/
Original citation: Loeffler, David and Zerbes, Sarah Livia. (2011) Wach modules and critical slope p-adic L-functions. Journal für die reine und angewandte Mathematik (Crelles Journal), 2013 (679). pp. 181-206. Permanent WRAP url: http://wrap.warwick.ac.uk/43664 Copyright and reuse: The Warwick Research Archive Portal (WRAP) makes this work of researchers of the University of Warwick available open access under the following conditions. Copyright © and all moral rights to the version of the paper presented here belong to the individual author(s) and/or other copyright owners. To the extent reasonable and practicable the material made available in WRAP has been checked for eligibility before being made available. Copies of full items can be used for personal research or study, educational, or not-for-profit purposes without prior permission or charge. Provided that the authors, title and full bibliographic details are credited, a hyperlink and/or URL is given for the original metadata page and the content is not changed in any way. Publisher’s statement: http://dx.doi.org/10.1515/crelle.2012.012 A note on versions: The version presented in WRAP is the published version or, version of record, and may be cited as it appears here. For more information, please contact the WRAP Team at: [email protected]
J. reine angew. Math. 679 (2013), 181—206
DOI 10.1515/crelle.2012.012
Journal fur die reine undangewandte Mathematik( Walter de Gruyter
Berlin � Boston 2013
Wach modules and critical slope p-adicL-functions
By David Loe¿er at Coventry and Sarah Livia Zerbes at Exeter
Abstract. We study Kato and Perrin-Riou’s critical slope p-adic L-function at-tached to an ordinary modular form using the methods of A. Lei, D. Loe¿er and S. L.Zerbes, Wach modules and Iwasawa theory for modular forms, Asian J. Math. 14 (2010),475–528. We show that it may be decomposed as a sum of two bounded measures multi-plied by explicit distributions depending only on the local properties of the modular format p. We use this decomposition to prove results on the zeros of the p-adic L-function, andwe show that our results match the behaviour observed in examples calculated by Pollackand Stevens in ‘‘Overconvergent modular symbols and p-adic L-functions’’, Ann. Sci. Ec.Norm. Super. (4) 44 (2011), no. 1, 1–42.
1. Introduction
1.1. Background. Let pf 3 be prime, and let f be a normalised, new modular ei-genform of level N, character � and weight k f 2, with N prime to p. Then a classical con-struction of Amice–Velu and Visik ([1], [15]) gives rise to p-adic L-functions for f , whichare distributions on Z�p interpolating the critical values of the L-functions of f and itstwists by Dirichlet characters of p-power conductor.
The construction depends on a choice of root of the Hecke polynomial of f at p, andrequires that the root be of ‘‘non-critical slope’’, i.e. that its p-adic valuation should bestrictly less than k � 1. The L-function corresponding to the root a is the unique distribu-tion Lp;a on Z�p of order h ¼ ordpðaÞ whose values at ‘‘special’’ characters of Z�p , i.e. thoseof the form z 7! z joðzÞ where 0e j e k � 2 and o is a finite-order character of conductorpn, are given by
ÐZ�p
z joðzÞ dLp;a ¼ð1� p ja�1Þ
�1� �ðpÞpk�2�ja�1
�~LLð f ; 1; j þ 1Þ if n ¼ 0;
a�npnð jþ1Þ~LLð f ;o�1; j þ 1Þ
Gðo�1Þ if nf 1;
8><>:ð1:1Þ
The first author is supported by EPSRC Postdoctoral Fellowship EP/F04304X/1. The second author is
supported by EPSRC Postdoctoral Fellowship EP/F043007/1.
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
where ~LLð f ;o�1; j þ 1Þ is the complex L-value Lð fo�1 ; j þ 1Þ of the twisted form fo�1 di-vided by certain explicit transcendental factors (see equation (6.1) below), and Gðo�1Þ isthe Gauss sum.
If the Hecke eigenvalue apð f Þ is not a p-adic unit, then both roots have non-criticalslope, and one obtains two p-adic L-functions, both of which are uniquely determined bythe corresponding interpolation formula (1.1). If apð f Þ is a p-adic unit (the ordinary case),then one root has non-critical slope (and in fact gives rise to a bounded measure) but theother does not, so one can only construct one p-adic L-function by these methods. Twoconstructions exist that redress the balance by constructing a ‘‘critical slope L-function’’for ordinary eigenforms: a p-adic analytic approach via the theory of overconvergent mod-ular symbols ([13]), and an algebraic approach via p-adic Hodge theory, using Kato’s Eulersystem ([9]). Both approaches give a distribution of order k � 1 on Z�p with the same inter-polation property at special characters, which depends on the restriction of the Galois rep-resentation of f to a decomposition group at p. If the local representation is non-split, thenthe values of both of these critical-slope L-functions at special characters are given by (1.1);if the local representation is split, the L-functions vanish at all such characters. However,these values do not uniquely determine a distribution of order k � 1, and we cannot neces-sarily deduce that the L-functions arising from the two approaches are equal.
Since Z�p GD� Zp, where D is cyclic of order p� 1, the D-isotypical components ofthese L-functions may be interpreted as distributions on Zp and hence (via the Amice trans-form) as power series on the p-adic open unit disc. In [13], §9, Pollack and Stevens calculatethe Newton polygon of the trivial D-component of the analytic critical-slope L-functionLPS
p;b in some explicit examples. They observe that the distribution of the zeros follows inter-
esting patterns which seem to be governed by the Iwasawa m and l-invariants of the unit-root L-function Lp;a: when p ¼ 3 and f is the twist of X0ð11Þ by a quadratic character ofconductor D prime to 3, for example, the numerical values suggest that the number of zeros
inside the open disc of radius rn ¼1
pnðp� 1Þ is pnðp� 1Þ þ lD, where lD is the l-invariant
of Lp;a. The same behaviour occurs when p ¼ 5 and the discriminant of the character isnegative. However, when p ¼ 5 and the discriminant is positive, the number of zeros insidethe open disc of radius rn is pn � 1þ �, for some mysterious non-negative integer � depend-ing on f .
1.2. Statement of the main results. Let p and f be as above, and assume thatk e p� 1 and that f is ordinary. Let a and b be the roots of the Hecke polynomial, andfix a to be the unit root. We choose a prime of the coe‰cient field of f above p. In thisintroduction, let us assume for simplicity that the completion of the coe‰cient field atthis prime is Qp. Let V �f be the dual of the p-adic representation attached to f , so it is a2-dimensional Qp-vector space which is crystalline with Hodge–Tate weights 0 and k � 1.Let G ¼ Gal
�QpðmpyÞ=Qp
�, HðGÞ the algebra of Qp-valued distributions on G, and write
LV �f: H 1
IwðQp;V�
f Þ !HðGÞnQpDcrisðV �f Þ
for the Perrin-Riou regulator map. (Here H 1IwðQp;V
�f Þ is the Iwasawa cohomology group,
whose definition we recall in §1.3.3 below. For a definition of LV �f
, see [11], Definition 3.4.)For any z A H 1
IwðQp;V �f Þ, we write LaðzÞ (resp. LbðzÞ) for the projection of LV �fðzÞ into the
a (resp. b) eigenspace of j�1. If zKato A H 1IwðQp;V
�f Þ is Kato’s zeta element, then (for appro-
182 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
priate normalisations of the Frobenius eigenvectors) we have LaðzKatoÞ ¼ Lp;a, and it is
conjectured that LbðzKatoÞ agrees with the critical slope p-adic L-function LPSp;b constructed
by Pollack and Stevens (cf. [14], Remark 8.5).
To simplify the notation, write Lp;b for LbðzKatoÞ. In this paper, we study Lp;b usingthe description via Wach modules developed in [10] and [11]. This gives rise to a canonicalsubspace �
j�NðV �f Þ�c¼0
LHðGÞnQpDcrisðV �f Þ;
stable under G and of rank 2 as a LðGÞ-module, through which the map LV �f
factors. InSection 4, we explicitly construct a basis n1, n2 of the Wach module NðV �f Þ. By comparing
this basis with the j-eigenvector basis of DcrisðV �f Þ, we obtain the following result:
Theorem A. If V �f is not locally split, then there exist Lp;1;Lp;2 A LQpðGÞ such that
aLp;a ¼ Lp;2;
bLp;b ¼ Lp;1M�1 ð1þ pÞj t
p
� �k�1 !
� Lp;2M�1�ð1þ pÞjðaÞ
�;
8>><>>:
where, as an element of QpJtK, we have
a ¼ ðk � 2Þ!a 1
1� p1�kmþ ð�1Þk
Pnfk
n� 1
k � 2
� �Bntn
n!ð1� mpn�kþ1Þ
!:
(Here M denotes the Mellin transform HðGÞ !G ðBþrig;QpÞc¼0, whose definition we re-
call in §1.3 below.)
Since Lp; i is bounded for i ¼ 1; 2, the distribution of the zeros of Lp;b is determined by
the Newton polygons oft
p
� �k�1
and a. In Section 7, we consider the case k ¼ 2 and deter-
mine which of these two terms dominates, depending on the behaviour of the m-invariantof Lp;a.
Theorem B. Let h be a character of D and let lh1 , mh
1 , lh2 , mh
2 be the Iwasawa l- and
m-invariants of the h-isotypical components of Lp;1 and Lp;2. Suppose that V �f is non-split at p.
(a) If mh2 <
1
ðp� 1Þ2þ m
h1 , then for ng 0, L
hp;b has pnðp� 1Þ2 zeros of valuation
1
pnðp� 1Þ2, and the total number of zeros of valuation > rn is pnðp� 1Þ þ l
h2 .
(b) If mh2 >
1
p� 1þ m
h1 , then for ng 0, L
hp;b has pnðp� 1Þ zeros of valuation
1
pnðp� 1Þ2, and the number of zeros of valuation > rn is pn � 1þ l
h1 .
Under the assumption that Kato’s zeta element is integral, which is known in manycases, this explains the numerical phenomena observed by Pollack and Stevens (see the endof Section 7).
183Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
1.3. Notation. As above, fix a prime pf 3, and let G ¼ Gal�QpðmpyÞ=Qp
�. Note
that the cyclotomic character w gives an isomorphism GGZ�p . We write
G ¼ D� G1;
where D is cyclic of order p� 1 (corresponding to the roots of unity in Z�p ) and
G1 ¼ Gal�QpðmpyÞ=QpðmpÞ
�is the preimage of 1þ pZp (non-canonically isomorphic to Zp). We denote the absoluteGalois group of Qp by GQp
.
We write Bþrig;Qpfor the ring of power series f ðpÞ A QpJpK such that f ðXÞ converges
everywhere on the open unit p-adic disc. Equip Bþrig;Qpwith actions of G and a Frobenius
operator j by g:p ¼ ðpþ 1ÞwðgÞ � 1 and jðpÞ ¼ ðpþ 1Þp � 1. We can then define a left in-verse c of j satisfying
j � c�
f ðpÞ�¼ 1
p
Pz p¼1
f�zð1þ pÞ � 1
�:ð1:2Þ
Inside Bþrig;Qp, we have subrings AþQp
¼ ZpJpK and BþQp¼ Qp nZp
AþQp. Moreover, the
actions of j, c and G restrict to these rings. Finally, we write t ¼ logð1þ pÞ A Bþrig;Qpand
q ¼ jðpÞ=p A AþQp. A formal power series calculation shows that gðtÞ ¼ wðgÞt for g A G and
jðtÞ ¼ pt.
If E is a finite extension of Qp, with ring of integers OE , we extend the actions of j, cand G to E nQp
Bþrig;Qpby E-linearity; these obviously preserve the subrings E nQp
BþQpand
OE nZpAþQp
.
We also let AQpbe the p-adic completion of AþQp
½1=p�, and BQp¼ AQp
½p�1�. Theserings appear in Fontaine’s theory of ðj;GÞ-modules (see [8]): if T is any finite-rank freeOE-module with a continuous action of GQp
, its ðj;GÞ-module DðTÞ is a finite-rank freeOE nZp
AQp-module with commuting semilinear actions of j and G, satisfying a certain
etaleness condition; and the functor Dð�Þ is an equivalence of categories. Similarly, for arepresentation of GQp
on an E-vector space V , one obtains a ðj;GÞ-module DðVÞ over thering E nQp
BQp.
1.3.1. The Mellin transform. We write LZpðGÞ for the Iwasawa algebra of G over
Zp,
LZpðGÞ ¼ lim �
ULGU open
Zp½G=U �:
We define LQpðGÞ ¼ Qp nZp
LZpðGÞ, and similarly for any finite extension E=Qp we define
LOEðGÞ and LEðGÞ by extending scalars. We use also the corresponding notations for G1 in
place of G.
Let HðGÞ be the algebra of distributions on G (the continuous dual of the space oflocally analytic functions on G, with multiplication given by convolution). If we choose a
184 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
generator g1 of G1, then the Amice transform allows us to identify HðGÞ with the algebra offormal power series
f f A Qp½D�JXK : f converges everywhere on the open unit p-adic discg;
where X corresponds to g1 � 1. We may identify LQpðGÞ with the subring of HðGÞ consist-
ing of power series with bounded coe‰cients.
The action of G on Bþrig;Qpgives an isomorphism of HðGÞ with ðBþrig;Qp
Þc¼0, theMellin transform
M : HðGÞ ! ðBþrig;QpÞc¼0;
f ðg� 1Þ 7! f ðg� 1Þðpþ 1Þ:ð1:3Þ
In particular, LZpðGÞ corresponds to ðAþQp
Þc¼0 under M. Similarly, we define HðG1Þ as thesubring of HðGÞ defined by power series over Qp, rather than Qp½D�. Then, HðG1Þ (respec-tively LZp
ðG1Þ) corresponds to ð1þ pÞjðBþrig;QpÞ (respectively ð1þ pÞjðAþQp
Þ) under M.
1.3.2. Crystalline representations. Let E again denote a finite extension of Qp, withring of integers OE . Fix a uniformizer $E . For a crystalline E-linear representation V ofGQp
, denote its Dieudonne module by DcrisðVÞ; this is an E-vector space of the same di-mension as V with an E-linear Frobenius j and a filtration, whose jumps are the negativesof the Hodge–Tate weights.
The following result is shown in [3], §II.1 and §III.4: if V is an E-linear representation,then V is crystalline with Hodge–Tate weights in ½a; b� if and only if there exists a (nec-essarily unique) E nQp
BþQp-module NðVÞ contained in the ðj;GÞ-module DðVÞ of V such
that the following conditions are satisfied:
(1) NðVÞ is free of rank d ¼ dimEðVÞ over E nQpBþQp
.
(2) The action of G preserves NðVÞ and is trivial on NðVÞ=pNðVÞ.
(3) j�pbNðVÞ
�H pbNðVÞ and pbNðVÞ=j�
�pbNðVÞ
�is killed by qb�a where
q ¼ jðpÞp
. (If M is an R-module equipped with a Frobenius j where R is any ring, then
j�ðMÞ denotes the R-module generated by jðMÞ.)
If the Hodge–Tate weights of V are e0, we may take b ¼ 0 above, so j preservesNðVÞ. In this case, if we endow NðVÞ with the filtration
Fil i NðVÞ ¼ fx A NðVÞ : jðxÞ A qiNðVÞg;
then NðVÞ=pNðVÞ is a filtered E-linear j-module, and as shown in [3], §III.4, we have anisomorphism of filtered j-modules NðVÞ=pNðVÞGDcrisðVÞ. Also, as shown in [3], II.2.1,DcrisðVÞ is contained in NðVÞnBþ
QpBþrig;Qp
, and in particular we can recover DcrisðVÞ as
DcrisðVÞ ¼�NðVÞnBþ
QpBþrig;Qp
�G:
185Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Moreover, we have a comparison isomorphism
NðVÞnBþQpBþrig;Qp
½t�1�GDcrisðVÞnQpBþrig;Qp
½t�1�;ð1:4Þ
which holds independent of the Hodge–Tate weights of V .
If T is a GQp-stable lattice in V , then NðTÞ ¼ NðVÞXDðTÞ is an OE nZp
AþQp-lattice
in NðVÞ, and by [3], §III.4, the functor T ! NðTÞ gives a bijection between the GQp-stable
OE-lattices T HV and the OE nZpAþQp
-lattices in NðVÞ satisfying the conditions (1)–(3)above.
Finally, it is easy to see from the construction that for all j A Z, NðVÞ and N�Vð jÞ
�are related by
N�Vð jÞ
�¼ p�jNðVÞn ej;
where ej is a basis for Qpð jÞ.
1.3.3. Iwasawa cohomology. If V is a p-adic representation of GQp, and T is a GQp
-stable Zp-lattice in V , we define the Iwasawa cohomology of T to be
H 1IwðQp;TÞ :¼ lim �
n
H 1�QpðmpnÞ;T
�:
As shown in [12], this is a LZpðGÞ-module of finite rank. Define
H 1IwðQp;VÞ :¼ Qp nZp
H 1IwðQp;TÞ
for any GQp-stable lattice T ; this definition is independent of the choice of T .
By [2], Theorem A.3, if V is crystalline with Hodge–Tate weightsf 0, and V hasno quotient isomorphic to the trivial representation, there is a canonical isomorphism ofLQpðGÞ-modules
H 1IwðQp;VÞGNðVÞc¼1;ð1:5Þ
which also identifies H 1IwðQp;TÞ with NðTÞc¼1 for each lattice T . These constructions
clearly commute with the additional OE-linear and E-linear structures when V is anE-linear representation and T is a GQp
-stable OE-lattice.
2. Setup
Let p be prime and let f be a normalised new modular eigenform of level N, charac-ter � and weight k f 2, with N prime to p and k e p� 1. Let F be the coe‰cient field of f ;we fix a choice of prime of F above p and let E be the completion of F at that prime, whichwe regard as a subfield of Qp.
We assume that f is ordinary, i.e. apð f Þ is a p-adic unit. Thus the roots of theHecke polynomial X 2 � apð f ÞX þ pk�1e, where e ¼ �ðpÞ, are elements of E with valua-
186 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
tions 0 and k � 1. Let a be the unit root and b ¼ pk�1ea�1 the non-unit root. We definem :¼ pk�1a=b ¼ e�1a2, which is a p-adic unit. By the Deligne–Ramanujan–Peterssonbound on japj, all embeddings of m into C have complex absolute value pk�1; in particular,m3 1.
We define ðrf ;Vf Þ to be the ‘‘cohomological’’ p-adic representation attached to f , sothe characteristic polynomial of geometric Frobenius at primes lFNp is
X 2 � alX þ lk�1�ðlÞ.
Then it is well known that
rf jGQpG
lðaÞ �0 w1�klðea�1Þ
� �
where lðxÞ denotes the unramified character of GQpmapping geometric Frobenius to x. In
particular, rf jGQphas an unramified subrepresentation.
Since f has level prime to p, rf jGQpis crystalline, and the characteristic polynomial
of j on Dcrisðrf Þ is X 2 � apX þ pk�1e. The unramified subrepresentation of rf jGQpcorres-
ponds to the j ¼ a eigenspace.
We shall mostly work with the ‘‘homological’’ representation ðr�f ;V �f Þ, the linear dualof Vf . This is crystalline at p and the arithmetic Frobenius at primes lFNp has the Heckepolynomial as its characteristic polynomial. On the decomposition group it is given by
r�f jGQpG
wk�1lðe�1aÞ �0 lða�1Þ
� �;
so the j-eigenvalues on DcrisðV �f Þ are b�1 ¼ p1�ke�1a (corresponding to the subrepresenta-tion with Hodge–Tate weight k � 1) and a�1.
We let Tf and T �f denote the canonical GQ-stable OE-lattices in Vf and V �f definedusing the cohomology of the modular curve X1ðNÞQ with coe‰cients in Zp, as in [9], §8.3.
3. A sequence of polynomials
In this section we define a certain sequence of polynomials which are related to theEulerian polynomials.
Definition 3.1. Let the polynomials hjðXÞ, for j f 0, be defined by h0ðXÞ ¼ 1 and
hjðXÞ ¼ ð1þ XÞ �Xd
dXþ j
� �hj�1ðX Þ:ð3:1Þ
It is immediate that hj is a monic polynomial of degree j with integral coe‰cients andconstant term hjð0Þ ¼ j!.
187Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Proposition 3.2. For any j f 0, the following identity of formal power series holds in
QJtK:
t
et � 1
� �jþ1
hjðet � 1Þ ¼ j!
1þ ð�1Þ j P
nfjþ1
n� 1
j
� �Bntn
n!
!;
where Bn is the n-th Bernoulli number.
Proof. The case j ¼ 0 of the proposition is the definition of the Bernoulli numbers,
t
et � 1¼ 1þ
Pnf1
Bntn
n!:
If Dj is the operator �td
dtþ j
� �, then we compute that
Dj �t
et � 1
� �j
hj�1ðet � 1Þ ¼ t
et � 1
� �jþ1
hjðet � 1Þ
and
Dj � ð j � 1Þ!
1þ ð�1Þ j�1Pnfj
n� 1
j � 1
� �Bntn
n!
!¼ j!
1þ ð�1Þ j P
nfjþ1
n� 1
j
� �Bntn
n!
!:
So the proposition holds for all j by induction. r
4. Calculating the (j,G)-module
Let V1 ¼ E�wk�1lðe�1aÞ
�and V2 ¼ E
�lða�1Þ
�. The extensions
0! V1 ! V ! V2 ! 0
in the category of E-linear continuous representations of GQpare parametrised by
H 1ðQp;V1 nE V �2 Þ ¼ H 1�Qp;E
�wk�1lðmÞ
��:
By [6], Theorem 0.2 (ii), this cohomology group is 1-dimensional over E. Hence up to iso-morphism there are exactly two such extensions: one split and one non-split. In this sectionwe give an explicit description of the ðj;GÞ-module corresponding to the non-split exten-sion V .
Since the ðj;GÞ-module functor is exact, DðVÞ is a free module of rank 2 overE nBQp
with a basis v1, v2, where v1 is a basis for the ðj;GÞ-module of V1 and the imageof v2 is a basis for the ðj;GÞ-module of the quotient V2. Thus in this basis ðv1; v2Þ, j acts viathe matrix
P ¼ e�1a x
0 a�1
� �
188 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
and an element g A G acts via
G ¼ wðgÞk�1 ay
0 1
!
for some x; y A E nBQp. In the split case, we may clearly take x ¼ y ¼ 0; so let us as-
sume we are in the non-split case. Since the actions of j and G commute, we must havePjðGÞ ¼ GgðPÞ, so x and y satisfy
ðmj� 1ÞðyÞ ¼�wðgÞk�1g� 1
�ðxÞ:
This, of course, is exactly the requirement that ðx; yÞ forms a 1-cocycle in the Herr complex(see e.g. [4], §I.4) calculating H 1
�Qp;E
�wk�1lðmÞ
��. We shall construct an explicit choice of
ðx; yÞ realising the non-split extension.
Proposition 4.1. If xk ¼ p1�khk�2ðpÞ, where hj are the polynomials of Definition 3.1,then we have
�wðgÞk�1g� 1
�xk A AþQp
for all g A G; and hk�2 is the unique monic polynomial of degree k � 2 such that this holds.
Proof. Let f be any element of QpJtK with f ð0Þ ¼ 1, and let F ¼ p1�kf . Then wefind that
�wðgÞk�1g� 1
�F ¼ t1�kðg� 1Þðtk�1FÞ ¼ t1�kðg� 1Þ
t
p
� �k�1
f
!:
Hence we have�wðgÞk�1g� 1
�F A QpJtK if and only if ðg� 1Þ
t
p
� �k�1
f
!A tk�1QpJtK, or
equivalentlyt
p
� �k�1
f A 1þ tk�1QpJtK. By construction, f ¼ hk�2ðpÞ ¼ hk�2ðet � 1Þ satis-
fies this (and it is obvious that there is a unique polynomial with this property up to scal-ing). Since F lies in AQp
, we have�wðgÞk�1g� 1
�F A AQp
XQpJtK ¼ AþQp. r
Recall that m :¼ e�1a2 3 1.
Lemma 4.2. There is no z A E nQpBQp
such that ðmj� 1ÞðzÞ ¼ p1�khk�2ðpÞ.
Proof. We assume (for simplicity of notation) that E ¼ Qp. Observe that we havea decomposition AQp
¼ pAþQplAe0
Qp, where Ae0
Qpconsists of those series with only non-
positive powers of p; thus Ae0Qp
is a subring isomorphic to ZphXi, where X ¼ p�1. Also, jpreserves Ae0
Qp, since the series expansion of jðp�1Þ lies in Ae0
Qp; indeed this gives an action
of j on ZphXi lifting the canonical Frobenius on Fp½X �.
If z is such that ðmj� 1ÞðzÞ ¼ p1�khk�2ðpÞ, then we must have z A Be0Qp¼ Ae0
Qp½p�1�,
and the constant term of z is zero. If z3Ae0Qp
, then there is some j > 0 such that z 0 ¼ p jz isin Ae0
Qpand its mod p reduction z 0 is a non-zero element of XFp½X �. But then z 0 must satisfy
189Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
ðmj� 1Þðz 0Þ ¼ 0, which is clearly impossible as j increases the degree of any non-constantpolynomial.
Hence z A Ae0Qp
. Then z A Fp½X � and ðmj� 1ÞðzÞ is a non-constant polynomial in X ofdegree k � 1 < p. It is clear that no such polynomial can lie in the image of mj� 1. r
We now define xk ¼ p1�khk�2ðpÞ � dk, where dk ¼ ð�1Þk Bk�1
ðk � 1Þ . Since
k � 1 < p� 1; dk A Zp;
and hence xk A AQp. Moreover, the choice of d implies that tk�1xk ¼ ðk � 2Þ!þOðtkÞ in
QpJtK; hence�wðgÞk�1g� 1
�ðxkÞ A pAþQp
.
Proposition 4.3. There exists yk A OE nAþQpsolving
ðmj� 1ÞðykÞ ¼�wðgÞk�1g� 1
�ðxkÞ:
Proof. It is clear that mj� 1 is surjective on pðOE nAþQpÞ for any m A OE : if
z A pAþQp, then we have jnðzÞ A jnðpÞAþQp
, and jnðpÞ tends to zero in the ðp; pÞ-adic topo-logy of AþQp
. Thus the series expansion �P
nf0
mnjnðzÞ converges, and its limit is clearly a
preimage of z. r
Note that yk depends on k, m, and g, but xk depends only on k. By construction,ðxk; ykÞ determines a class in H 1 of the Herr complex, and Lemma 4.2 shows that this classis not zero. Since H 1
�Qp;E
�wk�1lðmÞ
��is 1-dimensional over E, as noted above, we de-
duce that the ðj;GÞ-module we have constructed corresponds to the unique non-split exten-sion of the two factors.
5. Calculations in Wach modules
We deduce that the ðj;GÞ-module of the unique non-split extension V has anðE nBQp
Þ-basis ðv1; v2Þ for which the matrices of j and the generator g, in the basisðv1; v2Þ, are given by
P ¼ e�1a xk
0 a�1
� �
and
G ¼ wðgÞk�1 ay
0 1
!:
We let n1 ¼ p1�kv1 and n2 ¼ v2. Then the matrices of j and g in the basis ðn1; n2Þ are givenby
P 0 ¼pk�1
jðpÞk�1e�1a pk�1xk
0 a�1
0B@
1CA
190 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
and
G 0 ¼pk�1
gðpk�1Þ wðgÞk�1 pk�1ay
0 1
0B@
1CA:
It follows that if N is the E nBþQp-span of n1 and n2, the module N satisfies the con-
ditions (1)–(3) of §1.3.2, so N ¼ NðVÞ, the Wach module of V . Moreover, the OE nAþQp-
span of ðn1; n2Þ is a Wach module over OE nAþQp, and hence is NðTÞ for a GQp
-stableOE-lattice T in V .
Lemma 5.1. For each c; d A Z with ce d, there is a lattice Tc;d A V whose Wach
module is the OE nAþQp-span of $c
En1, $dEn2, and every GQp
-stable OE-lattice in V is one of
these. The residual representation Tc;d=pETc;d is non-split if c ¼ d and split otherwise.
Proof. It is clear that $cEn1, $d
En2 span a Wach module for every ce d, and hencecorrespond to a lattice in V . By construction the mod$E reduction of the cocycle ðxk; ykÞis nontrivial, so T0;0 ¼ T is residually non-split, and hence the same holds for Tc; c forany c.
Let T 0 be any lattice in V ; by scaling we may assume that it is contained in T , andnot contained in $ET . Then the image of T 0 in T=$ET is a nontrivial Galois-stable sub-space; so it is the subspace corresponding to the reduction of n1, as T=$ET is non-split and
thus has a unique 1-dimensional subspace. By devissage, we deduce that T 0 ¼ $jET þ T1
for some j, where T1 ¼ T XV1 is the lattice corresponding to n1. This corresponds to thecocycle ð$ j
Exk; $jEykÞ, and hence is residually split if j f 1. r
Proposition 5.2. The space�j�NðVÞ
�c¼0is free of rank 2 as a LEðGÞ-module, and a
basis is given by ð1þ pÞjðn1Þ, ð1þ pÞjðn2Þ. More specifically, if Tc;d denotes the lattice in V
defined in Lemma 5.1, then $cEð1þ pÞjðn1Þ, $d
Eð1þ pÞjðn1Þ are a basis of�j�NðTc;dÞ
�c¼0
as a free LOEðGÞ-module.
Proof. It is easy to check that the functor�j�ð�Þ
�c¼0is exact, so we have a short
exact sequence of LOEðGÞ-modules
0!�j�NðT1Þ
�c¼0 !�j�NðTÞ
�c¼0 !�j�NðT2Þ
�c¼0 ! 0:
Now as shown in [10], Theorem 3.5,�j�NðT1Þ
�c¼0is a free LOE
ðGÞ-module of rank 1,and there exists a basis n 01 of NðT1Þ which is congruent to n1 mod p such that ð1þ pÞjðn 01Þis a LOE
ðGÞ-basis of�j�NðT1Þ
�c¼0. Observe that an element in OE nAþQp
is invertible ifand only if its constant term is a unit in OE . Write n1 ¼ an 01 with a A ðAþQp
Þ�. By Prop-osition 3.10 in op.cit., jðpÞ ið1þ pÞjðn 01Þ A ð1� gÞ
�j�NðT1Þ
�c¼0. It follows that if we
write ð1þ pÞjðn1Þ ¼ a:ð1þ pÞjðn 01Þ, then a A LOEðGÞ�, so ð1þ pÞjðn1Þ is also a basis of�
j�NðT1Þ�c¼0
.
Similarly, we can show that if n2 denotes the image of n2 in NðT2Þ, then ð1þ pÞjðn2Þis a LOE
ðGÞ-basis of�j�NðT2Þ
�c¼0. As ð1þ pÞjðn2Þ is a lift of ð1þ pÞjðn2Þ, this implies the
result. r
191Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
We now calculate DcrisðVÞ using Berger’s comparison isomorphism (1.4). Since thehighest Hodge–Tate weight is k � 1, we have
DcrisðVÞ ¼
t
p
� �1�k
Bþrig;QpnBþ
QpNðVÞ
!G:
It is clear thatt
p
� �1�k
n1 is G-stable, and that it is a j-eigenvector with the following eigen-
value: b�1 ¼ p1�ke�1a.
To find an a�1 eigenvector lifting n2 (which is clearly an eigenvector in DcrisðV2Þ) we
must find a; b A E nBþrig;Qpsuch that
t
p
� �1�k
an1 þ bn2 is G-stable and killed by aj� 1,
with b ¼ 1 modulo p. Comparing coe‰cients of n2, we find that b ¼ 1. Writing out the
equations gðvÞ ¼ v and jðvÞ ¼ a�1v, where v ¼ t
p
� �1�k
an1 þ n2, we need to have
ð1� gÞðaÞ ¼ tk�1ay
and
ð1� p1�kmjÞðaÞ ¼ tk�1axk:
The existence and uniqueness of a solution to these equations is a consequence of the factthat V is known to be crystalline. If a is the solution, then we have the following formulaefor the eigenvectors in DcrisðVÞ:
Proposition 5.3. A basis of j-eigenvectors in DcrisðVÞ is given by
vb�1 ¼ �að0Þ t
p
� �1�k
n1;
va�1 ¼ t
p
� �1�k
an1 þ n2;
where a is the unique solution in Bþrig;Qpto ð1� p1�kmjÞðaÞ ¼ tk�1axk.
The motivation for the factor �að0Þ is the following lemma:
Lemma 5.4. The subspace Fil2�k DcrisðVÞ ¼ Fil0 DcrisðVÞ of DcrisðVÞ is spanned by
the vector va�1 þ vb�1 ¼ n2 þ�a� að0Þ
�n1.
Proof. We know that
Fil j DcrisðVÞ ¼ fv A DcrisðVÞ : jðvÞ A q jNrigðVÞg:
192 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Since vb�1 ¼ �að0Þ t
p
� �1�k
n1, we clearly have vb�1 3Fil2�k DcrisðVÞ. Hence we can
find some scalar l such that Fil2�k DcrisðVÞ is spanned by v ¼ va�1 þ lvb�1 . We have
jðvÞ ¼ a�1va�1 þ lb�1vb�1
¼ t
p
� �1�k�aa�1 � lað0Þb�1
�n1 þ a�1n2:
This is in q2�kNrigðVÞ if and only if�aa�1 � lað0Þb�1
�A E n qBþrig;Qp
; in other words,if and only if aðzp � 1Þ ¼ p1�klmað0Þ where zp is a primitive p-th root of unity. Since
a ¼ tk�1axk þ p1�kmjðaÞ, and t vanishes at zp � 1, we have
aðzp � 1Þ ¼ p1�kmjðaÞðzp � 1Þ ¼ p1�kmað0Þ:
Thus the unique solution is l ¼ 1, as claimed. r
Since the Dieudonne module of a lattice T LV is the image of NðTÞ in
DcrisðVÞ ¼ NðVÞ=pNðVÞ;
we see that:
Corollary 5.5. If Tc;d is the lattice in V defined above, then Fil0 DcrisðTc;dÞ is the
OE-span of $dEðva�1 þ vb�1Þ.
Note that there is a well-defined map Bþrig;Qp! QpJtK, which is determined by send-
ing p to et � 1 and whose image is contained in the ring of power series converging forjtj < p�1=ðp�1Þ.
Proposition 5.6. As elements of EJtK, we have
a
ðk � 2Þ!a ¼1
1� p1�kmþ ð�1Þk
Pnfk
n� 1
k � 2
� �Bntn
n!ð1� mpn�kþ1Þ :
Proof. By definition, we have ð1� p1�kmjÞðaÞ ¼ atk�1xk. Since
xk ¼ p1�khk�2ðpÞ � ð�1ÞkBk�1=ðk � 1Þ;
we have
tk�1xk ¼t
et � 1
� �k�1
hk�2ðet � 1Þ � ð�1Þk Bk�1tk�1
ðk � 1Þ
as elements of QpJtK. Substituting in the formula of Proposition 3.2, the tk�1 terms cancel,and we obtain
tk�1xk ¼ ðk � 2Þ!
1þ ð�1ÞkP
nfk
n� 1
k � 2
� �Bn
n!tn
!:
193Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Since jðtnÞ ¼ pntn, we deduce that
a
ðk � 2Þ!a ¼1
1� p1�kmþ ð�1Þk
Pnfk
n� 1
k � 2
� �Bntn
n!ð1� mpn�kþ1Þ ;
as claimed. r
We now consider the relation between this abstract representation V and the repre-sentation V �f attached to f . We have V �f GV
fðk � 1Þ, where f is the complex conjugate
of f . We may identify DcrisðVfÞ ¼ DdRðVf
Þ with the f -isotypical component of the de
Rham cohomology of X1ðNÞ with coe‰cients in the standard line bundle ok; in particular,f gives a canonical basis for the nontrivial filtration step.
Proposition 5.7. If the representation V �f is non-split as a GQp-representation, there is
a unique E-linear isomorphism of GQp-representations
V �f GV
inducing the map DcrisðV �f Þ ! DcrisðVÞ that sends t1�kf to va�1 þ vb�1 .
Proof. Clear, since V is the unique non-split extension up to scaling by E�, and wemay choose our scale factor so that t1�kf corresponds to va�1 þ vb�1 . r
Corollary 5.8. Suppose that V �f is non-split. Let T be the lattice of Lemma 5.1. Then
T �f MT , with equality if and only if T �f =$ET �f is non-split as a mod p representation of GQp.
Proof. As in [9], §14.22, the element f is a basis for Fil1 DcrisðTfÞ. Twisting by t1�k,
we see that T �f corresponds to a strongly divisible lattice in DcrisðVÞ whose intersection withFil0 DcrisðVÞ is the OE-span of va�1 þ vb�1 . Applying Corollary 5.5, T �f must be Tc;0 forsome ce 0; this is residually non-split if and only if c ¼ 0. r
Note that the local splitness of T �f =$ET �f at p can be explicitly checked in certaincases, using [14], Proposition 6.9.
If V �f is split, then we have an isomorphism V �f GV1 lV2, where
V1 ¼ E�wk�1lðe�1aÞ
�and V2 ¼ E
�lða�1Þ
�as above. If n1 and n2 are the natural basis vectors of the Wach modules of V1 and V2, then
it is clear that vb�1 ¼ t
p
� �1�k
n1 and va�1 ¼ n2 are a basis of eigenvectors of DcrisðV1 lV2Þ,
and the nontrivial filtration step is DcrisðV2Þ ¼ Eva�1 . Hence we may choose an isomor-phism V �f ! V1 lV2 such that the resulting isomorphism DcrisðV �f Þ ! DcrisðV1 lV2Þsends t1�kf to va�1 . Since V1 and V2 have distinct mod p reductions, the only possible lat-tices are direct sums of lattices in the factors, and hence we may assume that the above iso-morphism maps T �f to the lattice corresponding to the span of n1 and n2. However, thisisomorphism is still not canonical, since it is only determined up to multiplying n1 by anarbitrary element of O�E .
194 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
6. Consequences for the L-functions
We now use the results we have collected on the local representation V �f jGQpto de-
scribe the p-adic L-functions of f . We recall the setup from [10], §3.6. Since V �f has non-negative Hodge–Tate weights 0 and k � 1, and the unique 1-dimensional quotient of V �f isnot the trivial representation, the theorem quoted in §1.3.3 shows that
H 1IwðQp;V �f ÞGNðV �f Þ
c¼1:
We let Col denote the composition of this isomorphism with the map
1� j : NðV �f Þc¼1 !
�j�NðV �f Þ
�c¼0:
Note that since the Hodge–Tate weights of V �f aref0, we have
�j�NðV �f Þ
�c¼0L ðBþrig;Qp
Þc¼0 nQpDcrisðV �f Þ:
Let z A H 1IwðQp;V
�f Þ. We thus have an element ColðzÞ A
�j�NðV �f Þ
�c¼0. We consider
it as an element of ðBþrig;QpÞc¼0 nQp
DcrisðVÞ via the above inclusion, and define LaðzÞ andLbðzÞ to be its projections to the eigenspaces, so
ColðzÞ ¼ LaðzÞva�1 þ LbðzÞvb�1 :
We can consider LaðzÞ and LbðzÞ as power series in p lying in ðE nBþrig;QpÞc¼0. Alter-
natively, we may regard them as E-valued distributions on G, via the Mellin transform(1.3). As shown in [11], §3.1, these correspond to the coordinates of LV �
fðzÞ in the basis
ðva�1 ; vb�1Þ, where
LV �f: H 1
IwðQp;V�
f Þ !HðGÞnQpDcrisðVÞ
is Perrin-Riou’s regulator map. Perrin-Riou’s theory implies that LaðzÞ is a distribution oforder 0, while LbðzÞ has order k � 1.
If o is a Dirichlet character (to any modulus), let Lð f ;o; sÞ denote the complexL-function of f twisted by o,
Panð f ÞoðnÞn�s. Recall (e.g. from [5], §3.1.2) that there are
nonzero complex numbers Wþ and W� such that
~LLð f ;o; j þ 1Þ :¼ Gð j þ 1Þð2piÞ jþ1WG
Lð f ;o; j þ 1Þ A Qð f ;oÞð6:1Þ
where WG denotes Wþ if ð�1Þ jþ1wð�1Þ ¼ 1 and W� if ð�1Þ jþ1wð�1Þ ¼ �1. Here Qð f ;oÞdenotes the finite extension of Q generated by the values of o and the coe‰cients anð f Þ.
Theorem 6.1 ([9], Theorem 16.6). There exists an element zKato A H 1IwðQp;V
�f Þ such
that for any finite-order character o of G of conductor pn, and any 0e j e k � 2,
195Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
LaðzKatoÞðw joÞ ¼ð1� p ja�1Þð1� epk�2�ja�1Þ~LLð f ; 1; j þ 1Þ if n ¼ 0;
a�npnð jþ1Þ ~LLð f ;o�1; j þ 1ÞGðo�1Þ if nf 1;
8><>:
where Gðo�1Þ is the Gauss sum.
Hence LaðzKatoÞ is a distribution of order 0 on G whose values at special charactersare given by (1.1); so it is equal to the p-adic L-function Lp;a. We define Lp;b ¼ LbðzKatoÞ; ifV �f is non-split, this satisfies (1.1) for the root b, while if V �f is split, Lp;b vanishes at allspecial characters. We now use the fact that Col factors through
�j�NðV �f Þ
�c¼0, and the
basis of the latter space given by Proposition 5.2, to give a decomposition of these two dis-tributions.
Definition 6.2. Let n1, n2 be the basis of NðT �f Þ defined above. We let Lp;1 and Lp;2
be the unique elements of LEðGÞ such that
ColðzKatoÞ ¼ Lp;1 � ð1þ pÞjðn1Þ þ Lp;2 � ð1þ pÞjðn2Þ:
Proposition 6.3. If the image of GalðQ=QyÞ in GLðT �f Þ contains a conjugate of
SL2ðZpÞ, then Lp;2 lies in LOEðGÞ. If in addition V �f is split at p, or it is non-split and the re-
sidual representation T �f =$ET �f is also non-split, then the same holds for Lp;1.
Proof. If the hypothesis on the image of the global representation is satisfied, thenzKato A H 1
IwðT �f Þ, by [9], Theorem 12.5 (4) and Theorem 12.6. Since the image of NðT �f Þ inthe quotient is the OE nAþQp
-span of n2, this implies that Lp;2 is integral. The additionalassumptions imply the stronger statement that NðT �f Þ is the span of n1 and n2 (Corollary5.8), so we also obtain integrality for Lp;1. r
Using the formulae of the previous section relating va�1 and vb�1 to n1 and n2, we canwrite the functions Lp;a and Lp;b in terms of the Lp; i.
Theorem 6.4. The following relations hold in E nHðGÞ:
(a) If V �f is locally split, then
aLp;a ¼ Lp;2;
bLp;b ¼ Lp;1M�1
ð1þ pÞj t
p
� �k�1!:
8>><>>:
(b) If V �f is not locally split, then
aLp;a ¼ Lp;2;
�að0ÞbLp;b ¼ Lp;1M�1
ð1þ pÞj t
p
� �k�1!� Lp;2M
�1�ð1þ pÞjðaÞ
�:
8>><>>:
196 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Proof. In the non-split case, we use Proposition 5.3 to write
n1 ¼ �1
að0Þt
p
� �k�1
vb�1 ;
n2 ¼ va�1 þ a
að0Þ vb�1 :
Substituting these into the identity
Lp;a � ð1þ pÞva�1 þ Lp;b � ð1þ pÞvb�1 ¼ Lp;1 � ð1þ pÞjðn1Þ þ Lp;2 � ð1þ pÞjðn2Þ;
where the ‘‘�’’ denotes the action of E nHðGÞ on ðBþrig;QpÞc¼0 nQp
DcrisðVÞ, we obtain
Lp;a � ð1þ pÞva�1 þ Lp;b � ð1þ pÞvb�1
¼ Lp;1 ��1
að0Þb ð1þ pÞj t
p
� �k�1
vb�1 þ Lp;2 � ð1þ pÞ 1
ava�1 þ jðaÞ
að0Þb vb�1
� �:
Since ð1þ pÞva�1 and ð1þ pÞvb�1 are clearly a basis for ðBþrig;QpÞc¼0 nQp
DcrisðVÞ as aE nHðGÞ-module, we can project onto each of these to obtain the proposition.
In the split case, one argues identically using the formulae n1 ¼t
p
� �k�1
vb�1 andn2 ¼ va�1 . r
Note that jt
p
� �k�1
has a zero of order k � 1 at z� 1, for any root of unity z of order
pn, nf 2. It is straightforward to see (using Theorem 5.4 and Lemma 5.9 of [10]) that this
is equivalent to M�1
ð1þ pÞj t
p
� �k�1!
vanishing at every character of G of the form w jo,
where w is the cyclotomic character, 0e j e k � 2 and o is a finite-order character not fac-toring through D. Hence the factor multiplying Lp;1 in Proposition 6.4 vanishes at all butfinitely many of the points corresponding to critical values of the complex L-function.
In fact Lp;1 vanishes at most of the remaining points:
Proposition 6.5. If V �f is locally split, then the distribution Lp;1 vanishes at
z 7! zihðzÞ, for any 0e ie k � 2 and any character h of Z�p factoring through D. If V �f is
not locally split, this is true at all characters of this form with h nontrivial.
Proof. If the representation is split, Lb is known to vanish at all special characters.
Since we have bLp;b ¼ Lp;1M�1
ð1þ pÞj t
p
� �k�1!
, and the second factor on the right-
hand side does not vanish at the characters z 7! zihðzÞ, Lp;1 must do so.
In the non-split case, we suppose that h is nontrivial. Then we must have
bLp;bðw ihÞ ¼ aLp;aðw ihÞ;
197Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
since both sides are equal to p jþ1~LLð f ; h�1; 1þ jÞ=Gðh�1Þ. Substituting the formulae ofProposition 6.4, we obtain
Lp;2ðw jhÞM�1�ð1þ pÞj
�a� að0Þ
��ðw jhÞ ¼ Lp;1ðw jhÞM�1
ð1þ pÞj t
p
� �k�1!:
By construction b :¼ a� að0Þ vanishes to order k at 0, so ð1þ pÞjðbÞ vanishes to order k atzp � 1 for any nontrivial p-th root of unity z. Hence the distributions q jð1þ pÞjðbÞ vanish
at zp � 1, for i ¼ 0; . . . ; k � 2, where q ¼ ð1þ pÞ d
dp; equivalently, ð1þ pÞjðbÞ pairs to zero
with any function on Zp whose restriction to each coset of pZp is a polynomial ofdegreee k � 2. In particular, it pairs to zero with the characters z 7! z jhðzÞ (extended to
functions on Zp zero on pZp). Since M�1
ð1þ pÞj t
p
� �k�1!
does not vanish at these char-acters, Lp;1 must vanish. r
This proposition, together with the preceding discussion, imply that the distribution
Lp;1M�1
ð1þ pÞj t
p
� �k�1!
vanishes at every locally algebraic character of degreee k � 2
that is not algebraic, and at every locally algebraic character in the split case. In the splitcase, this distribution is simply Lp;b, and one deduces the well-known fact that Lp;b ‘‘pre-tends rather convincingly to be 0’’ (see [5], Remarque 4.12).
Remark 6.6. From the formula a ¼ tk�1axk þ p1�kmjðaÞ, we deduce that
aðzp j � 1Þ ¼ p1�kmaðzp
p j � 1Þ
for any j f 1, where zp j is a p j-th root of unity. Since p1�km ¼ a=b, this gives
aðzp j � 1Þ ¼ a
b
� �j
að0Þ:
Since a� að0Þ vanishes to orderf k � 1 at 0, a� a
b
� �j
að0Þ vanishes to order k � 1 at
zp j � 1. This gives a purely analytic proof that for any F A HðGÞ satisfying the interpola-tion property of (1.1) for the critical-slope root b, the distribution
G ¼ aLp;aM�1�ð1þ pÞjðaÞ
�� að0ÞbF
vanishes at all special characters of G of conductor > 1. Hence G factorises as
M�1
ð1þ pÞj t
p
� �k�1!
H
for some distribution H, and if F has order k � 1, then H must be in LEðGÞ.
198 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
In particular, taking F to be the analytic critical-slope L-function LPSp;b, we obtain a
decomposition of LPSp;b analogous to Proposition 6.4. However, without the above interpre-
tation of H via Wach modules, it is not clear how one could determine whether or not H
was integral.
In the non-split case, we also obtain a formula for Lp;1ðw iÞ, for i ¼ 0; . . . ; k � 2, whichallows us to show that it is non-vanishing in some cases:
Proposition 6.7. If V �f is non-split and at least one of the L-values Lð f ; jÞj¼1;...;k�1 is
non-zero, then Lp;1 3 0.
Proof. For 0e j e k � 2, we have
M�1�ð1þ pÞjðaÞ
�ðw jÞ ¼ að0Þ;
since a� að0Þ vanishes to degreef k � 1 at the origin. Thus
Lp;1ðw jÞM�1
ð1þ pÞj t
p
� �k�1!ðw jÞ ¼ að0ÞðaLp;a � bLp;bÞ:
On the right-hand side,
að0Þ ¼ ðk � 2Þ!a1� p1�km
¼ ðk � 2Þ!a1� a=b
¼ ðk � 2Þ!abb � a
;
and substituting the values of Lp;a and Lp;b at w j from equation (1.1) and simplifiying, weobtain (eventually)
Lp;1ðw jÞ �M�1
ð1þ pÞj t
p
� �k�1!ðw jÞ ¼ �ðk � 2Þ!ðp� 1Þpk�2e~LLð f ; 1; 1þ jÞ: r
If k f 3, this is su‰cient to show that Lp;1 3 0, since the complex L-function Lð f ; jÞ
does not vanish for j >k
2. If k ¼ 2, the only character at which we can relate the value of
Lp;1 to the complex L-function is the trivial character, so when Lð f ; 1Þ ¼ 0 (which can ofcourse happen) we cannot show that Lp;1 3 0.
7. Newton polygons and Mellin transforms
In this section, we take k ¼ 2, and present some explicit consequences of the aboveanalysis for the algebraic critical-slope L-function Lp;b. For s > 0 let Cs denote theclosed a‰noid disc fX : jX je p�sg. (For our purposes it will su‰ce to take s rational;if s is irrational, this space is not defined as an a‰noid space, but it can be interpretedas a Berkovich space.) For a rigid-analytic function f on the open unit disc, we writevsð f Þ ¼ inf
x ACs
ordp
�f ðxÞ
�; note that vsð f Þ is clearly an increasing function of s.
199Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Proposition 7.1. The function s 7! vsð f Þ is continuous, piecewise-linear, and concave.
For any sf 0, the left-hand derivative of vsð f Þ at s is the number of zeros of f on Cs (counted
with multiplicity), and the right-hand derivative is the number of zeros of f on the open disc
fX : jX j < p�sg.
Proof. This is simply a restatement of the standard theory of the Newton polygon.r
As above, let g1 be a generator of G1, and let x ¼ g1 � 1, so HðG1Þ is the ring ofpower series in x converging on the unit disc.
Proposition 7.2. Let f A Bþrig;Qpand let g ¼M�1
�ð1þ pÞjð f Þ
�A HðG1Þ. Then for
any s with 0 < s < 1, we have vsð f Þ ¼ vsðgÞ.
Proof. Let us suppose f ¼P
anpn. Then
g ¼Pnf0
anfnðxÞ;
where fnðxÞ ¼M�1�ð1þ pÞjðpÞn
�.
We know that fnðxÞ ¼ tnðxÞxðx� l1Þ � � � ðx� ln�1Þ, where li ¼ wðg1Þi � 1 A pZp and
tn A LQpðG1Þ�. Since ð1þ pÞjðpÞn A AþQp
npAþQp, we must have tn A LZp
ðG1Þ�.
Let us write dn ¼ tnð0Þ. Then we have
vs
�fnðxÞ � dnxn
�¼ vs
�xn�tnðxÞ � dn
�þ tnðxÞ
�xðx� l1Þ � � � ðx� ln�1Þ � xn
��:
We evidently have xn�tnðxÞ � dn
�A xnþ1ZpJxK, so
vs
�xn�tnðxÞ � dn
��f ðnþ 1Þs:
For the second group of terms, the coe‰cient of xn�j in xðx� l1Þ � � � ðx� ln�1Þ � xn isclearly divisible by p j; since tnðxÞ A Z�p , we have
vs
�tnðxÞ
�xðx� l1Þ � � � ðx� ln�1Þ � xn
��f inf
1ejenðn� jÞsþ j
¼ ðn� 1Þsþ 1:
Since 0 < s < 1, both ðnþ 1Þs and ðn� 1Þsþ 1 are strictly bigger than vsðdnxnÞ ¼ ns. Thus,in particular, vs
�fnðxÞ
�¼ ns.
We now write
g ¼P
nf0
anfnðxÞ ¼�P
nf0
andnxn
�þ�P
nf0
an
�fnðxÞ � dnxn
��:
200 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
Clearly we have
vs
�Pnf0
andnxn
�¼ inf
nf0ðnsþ ordp anÞ ¼ vsð f Þ:
On the other hand,
vs
�Pnf0
an
�fnðxÞ � dnxn
��f inf
nf0
�inf�ðn� 1Þsþ 1; ðnþ 1Þs
�þ ordp an
�¼ vsð f Þ þ infðs; 1� sÞ:
Hence we must have vsðgÞ ¼ vsð f Þ. r
Combining the two preceding propositions, we see that the zeros of the power series f
and g lying ‘‘near the boundary’’ must have the same valuations; the zeros inside the closeddisc jX je p�1 are equal in number, but can be in very di¤erent places within this disc, asthe examples f ¼ pn show.
Corollary 7.3. Let m A O�E , m3 1, and let f be the unique element of E nBþrig;Qpsuch
that
1� m
pj
� �ð f Þ ¼ t
pþ t
2:
Then f has ðp� 1Þ zeros of valuationf1
p� 1, piðp� 1Þ2 zeros of valuation
1
piðp� 1Þ2for
each integer i f 0, and no other zeros. Moreover, vsð f Þ < vs
t
p
� �for s <
1
ðp� 1Þ2; and we
have
lim infs!0
vs
t
p
� �� vsð f Þ
!¼ 1
ðp� 1Þ2;
lim sup
vs
t
p
� �� vsð f Þ
!¼ 1
ðp� 1Þ :s!0
(We have f ¼ a=a in the notation of the previous sections, in the case k ¼ 2.)
Proof. Let us calculate vsð f Þ. We suppose first that s >1
p� 1. Then the formal
series expansion
f ¼P
nf0n31
Bntn
n!ð1� pn�1mÞ
201Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
of Proposition 5.6 is convergent, and the disc jpje p�s corresponds to jtje p�s; hence
vf ðsÞ ¼ infnf0n31
ordp
Bn
n!ð1� pn�1mÞ
� �þ ns:
For nf 2, ð1� pn�1mÞ A O�E , and hence ordp
Bn
n!ð1� pn�1mÞ
� �¼ ordp
Bn
n!
� �. We
have
infnf2
ordp
Bn
n!
� �þ ns ¼ inf
jtjep�sordp
t
et � 1� 1þ t
2
� �
¼ infjpjep�s
ordp
t
p� 1þ t
2
� �
¼ infnf2
ordp
1
nþ 1� 1
2n
� �þ ns:
Clearly, if nf p� 1 and n 0 is the largest integere n of the form p j � 1, or n 0 ¼ 2 ifn < p� 1, then we have
ordp
1
n 0 þ 1� 1
2n 0
� �þ n 0s < ordp
1
nþ 1� 1
2n
� �þ ns:
Thus the infimum is attained either at n ¼ 2 or at n ¼ p j � 1 for some j f 1. We calculatethat the term for n ¼ p j � 1 is ðp j � 1Þs� j, which is a strictly increasing function of j for
any s >1
p� 1. Hence the infimum is
inf�ðp� 1Þs� 1; 2s
�:
(If p ¼ 3, then the 2s term does not appear, but the other term is 2s� 1 which is smalleranyway, so the formula is true as stated.)
If we include also the term in the original sum for n ¼ 0, we deduce that for any s inthis range
vsð f Þ ¼ inf�1; 2s; ðp� 1Þs� 1
�:
One checks that if pf 5, this gives
vsð f Þ ¼
ðp� 1Þs� 1 if1
p� 1e se
1
p� 3;
2s if1
p� 3e se
1
2;
1 if sf1
2;
8>>>>>>><>>>>>>>:
202 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
whereas if p ¼ 3, we obtain
vsð f Þ ¼2s� 1 if
1
2e se 1;
1 if sf 1:
8<:
Hence the zeros of f with valuation >1
p� 1are: two zeros of valuation 1 if p ¼ 3; four
zeros of valuation1
2if p ¼ 5; and two of valuation
1
2and p� 3 of valuation
1
p� 3if pf 7.
In each case, the total number of zeros with valuations in this range is p� 1.
We now use this intensive study of vf ðsÞ for relatively large s to describe vf ðsÞ forall smaller s. From the equation f ¼ t=pþ t=2þ p�1mjð f Þ, we deduce that for s in the
interval1
p� 1;
p
p� 1
� �we have
vs=pð f Þf inf�vs=pðt=pÞ;�1þ vsð f Þ
�¼ inf
�vs=pðtÞ � s=p;�1þ vsð f Þ
�
¼ �1þ inf
1� 1
p
� �s; vsð f Þ
!:
We find that if1
p� 1< s <
p
ðp� 1Þ2, the term �1þ vsð f Þ is strictly smaller, whereas for
larger s, the term �1þ 1� 1
p
� �s is strictly smaller; hence this is the exact value of the
left-hand side (by the ultrametric property), and by continuity this is the case at the cross-
over point s ¼ p
ðp� 1Þ2. This determines vsð f Þ in the interval
1
pðp� 1Þ ;1
p� 1
� �: we
have
v 1pð p�1Þð f Þ ¼ �1; v 1
ð p�1Þ2ð f Þ ¼ �1þ 1
p� 1; and v 1
p�1ð f Þ ¼ 0;
and vsð f Þ is linear between these points.
We now consider the interval from1
pnþ1ðp� 1Þ to1
pnðp� 1Þ , for nf 1. By iteratingthe functional equation, we find that
vs=pið f Þf�i þ inf
vf ðsÞ; 1� 1
p
� �s; 1� 1
p2
� �s; . . . ; 1� 1
pi
� �s
!:
The terms 1� 1
p2
� �s; . . . ; 1� 1
pi
� �s are clearly strictly larger than 1� 1
p
� �s, so we de-
duce that for sf1
p� 1, we have vf
s
pi
� �¼ �i þ inf
vf ðsÞ; 1� 1
p
� �s
!¼ 1� i þ vf
s
p
� �.
203Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
This gives the locations of the zeros in the statement of the proposition, and shows that
vsð f Þ < vs
t
p
� �for all s <
1
ðp� 1Þ2.
Finally, we establish the formulae for the limits inferior and superior. For1
p� 1e se
p
p� 1and i f 1 we have
vs=pi
t
p
� �� vs=pið f Þ ¼ vs=piðtÞ � s
pi
� �� �i þ inf
1� 1
p
� �s; vsð f Þ
!
¼ 1� 1
pi
� �s� inf
1� 1
p
� �s; ðp� 1Þs� 1
!:
One checks that the minimum value of this expression is attained at s ¼ p
ðp� 1Þ2, where it
is equal topi�1 � 1
pi�1ðp� 1Þ2. Hence for any se
1
piðp� 1Þ , we have
vs
t
p
� �� vsð f Þf
pi�1 � 1
pi�1ðp� 1Þ2
and equality occurs for s ¼ 1
piðp� 1Þ2; so the limit inferior as s! 0 is
1
ðp� 1Þ2, as
claimed. On the other hand, the maximum value ispi�1 � 1
pi�1ðp� 1Þ , attained at both of the
endpoints, so the limit superior is1
p� 1. r
We now use this to describe the L-functions. We note that HðGÞ ¼Lh
ehHðGÞ,
where the sum is over the characters of D and eh is the corresponding idempotent. Forg A HðGÞ, we let gh be the unique element of HðG1Þ such that ehg
h ¼ ehg.
For a character h of D, let lh1 , mh
1 , lh2 , mh
2 be the Iwasawa l- and m-invariants of Lhp;1
and Lhp;2. Note that lh
2 , mh2 are equal to the corresponding invariants of the unit root p-adic
L-function Lp;a, which can be calculated in many cases; we know of no easy way to evalu-ate l
h1 , mh
1 , but Proposition 6.3 gives conditions under which mh2 is forced to be non-negative.
Theorem 7.4. Let h be a character of D and let lh1 , m
h1 , l
h2 , m
h2 be as above. Suppose
that V �f is non-split at p.
(a) If mh2 <
1
ðp� 1Þ2þ m
h1 , then for all su‰ciently small s we have
vsðLhp;bÞ ¼ l
h2 sþ m2 þ vsð f Þ;
where f is as above. In particular, for ng 0, Lhp;b has pnðp� 1Þ2 zeros of valuation
1
pnðp� 1Þ2, and the total number of zeros of valuation > rn is pnðp� 1Þ þ l
h2 .
204 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
(b) If mh2 >
1
p� 1þ m
h1 , then for su‰ciently small s the formula becomes
vsðLhp;bÞ ¼ l
h1 sþ m1 þ vs
t
p
� �;
so for ng 0 there are pnðp� 1Þ zeros of valuation1
pnðp� 1Þ2and the number of zeros of
valuation > rn is pn � 1þ lh1 .
Proof. We have a decomposition
�að0ÞbLp;b ¼ Lp;1M�1
ð1þ pÞj t
p
� �!� Lp;2M
�1�ð1þ pÞjðaÞ
�:
For all su‰ciently small s, we have vsðLhp; iÞ ¼ m
hi þ sl
hi . Hence, using the bounds of Corol-
lary 7.3, the hypotheses of case (a) force the second term to dominate, giving the stated for-mula for vsðLh
p;bÞ. Similarly in case (b), the first term must dominate for all su‰ciently small
s, giving the formula stated. r
This gives a (conditional) explanation of the phenomena observed in the examples of[13], §9, for the critical slope 3-adic and 5-adic L-functions of quadratic twists of the ellipticcurves X0ð11Þ and X0ð14Þ, with h the trivial character of D. In all of these cases the residualrepresentations are locally non-split at p.
The 3-adic representation of X0ð11Þ (and hence of any of its twists) is surjective1).Hence the Kato zeta element for this curve is integral, and any twist such that m
h2 ¼ 0
must satisfy the hypotheses of (a) above.
In the case of 3-adic L-functions of twists of X0ð14Þ, the residual representation isglobally reducible (but still locally non-split). Thus we cannot show that Lp;1 is integral;but if we assume that this is the case, then again any twist of X0ð14Þ with m
h2 ¼ 0 satisfies
the hypotheses of (a), which is consistent with the numerical results of op.cit.
In the case of the 5-adic L-functions of twists of X0ð11Þ, let us also assume that Lp;1 isintegral. From this assumption, it follows that for twists by even quadratic characters,where m
h2 ¼ 0, we obtain the pattern of zeros of part (a) of the theorem; but for twists by
odd quadratic characters, where mh2 > 0, we are in the situation of (b), unless mh
1 is also pos-itive.
Acknowledgement. We would like to thank Chris Wuthrich for useful conversationsrelating to the integrality of Kato’s zeta-elements. This paper was written while the secondauthor was visiting the University of Warwick; she thanks the number theory group fortheir hospitality.
1) This follows from the fact that its mod 3 representation is surjective, as can be explicitly seen by calcu-
lating the Galois group of the 3-torsion field; and its j-invariant does not lie in the image of the rational func-
tion f ðxÞ of [7]. So its mod 9 representation is surjective, which implies that its 3-adic representation is also
surjective.
205Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM
References
[1] Y. Amice and J. Velu, Distributions p-adiques associees aux series de Hecke, Journees Arithmetiques de
Bordeaux (Univ. Bordeaux, 1974), Asterisque 24–25 (1975), 119–131.
[2] L. Berger, Bloch and Kato’s exponential map: Three explicit formulas, Doc. Math. Extra Vol. 3 (2003),
99–129, Kazuya Kato’s fiftieth birthday.
[3] L. Berger, Limites de representations cristallines, Compos. Math. 140 (2004), 1473–1498.
[4] F. Cherbonnier and P. Colmez, Theorie d’Iwasawa des representations p-adiques d’un corps local, J. Amer.
Math. Soc. 12 (1999), 241–268.
[5] P. Colmez, La conjecture de Birch et Swinnerton-Dyer p-adique, Seminaire Bourbaki, Vol. 2002/03,
Asterisque 294 (2004), Exp. No. 919, 251–319.
[6] P. Colmez, Representations triangulines de dimension 2, Asterisque 319 (2008), 213–258.
[7] N. Elkies, Elliptic curves with 3-adic Galois representation surjective mod 3 but not mod 9, preprint 2006.
[8] J.-M. Fontaine, Representations p-adiques des corps locaux. I, The Grothendieck Festschrift, Volume II,
Progr. Math. 87 (1990), 249–309.
[9] K. Kato, p-adic Hodge theory and values of zeta functions of modular forms, in: Cohomologies p-adiques et
applications arithmetiques (III), Asterisque 295 (2004), 117–290.
[10] A. Lei, D. Loe¿er and S. L. Zerbes, Wach modules and Iwasawa theory for modular forms, Asian J. Math.
14 (2010), 475–528.
[11] A. Lei, D. Loe¿er and S. L. Zerbes, Coleman maps and the p-adic regulator, Alg. Number Th., to appear.
[12] B. Perrin-Riou, Theorie d’Iwasawa des representations p-adiques sur un corps local, Invent. Math. 115
(1994), 81–161.
[13] R. Pollack and G. Stevens, Overconvergent modular symbols and p-adic L-functions, Ann. Sci. Ec. Norm.
Super. (4) 44 (2011), no. 1, 1–42.
[14] R. Pollack and G. Stevens, Critical slope p-adic L-functions, preprint 2009.
[15] M. Visik, Nonarchimedean measures associated with Dirichlet series, Mat. Sb. (N.S.) 99(141) (1976),
248–260, 296.
Warwick Mathematics Institute, University of Warwick, Coventry CV4 7AL, UK
e-mail: D.A.Loe¿[email protected]
Department of Mathematics, University of Exeter, Exeter EX4 4QF, UK
e-mail: [email protected]
Eingegangen 6. Dezember 2010, in revidierter Fassung 28. Juli 2011
206 Loeffler and Zerbes, Wach modules and critical slope p-adic L-functions
Brought to you by | University of WarwickAuthenticated
Download Date | 1/25/16 10:51 AM