(Advanced Courses in Mathematics - CRM Barcelona) Laurent Berger, Gebhard Böckle, Lassina Dembélé, Mladen Dimitrov, Tim Dokchitser, John Voight (Auth.)-Elliptic Curves, Hilbert Modular Forms and Galoi

Coursesamatics
CRM Barcelona
Advanced Courses
in Mathematics
CRM Barcelona
Laurent Berger
Gebhard Bckle
Lassina Dembl
Mladen Dimitrov
Tim Dokchitser
John Voight
Elliptic Curves,
Hilbert Modular Forms
and Galois
Deformations
Advanced Courses in Mathematics

CRM Barcelona
Centre de Recerca Matemtica
Managing Editor:
Carles Casacuberta
For further volumes:

http://www.springer.com/series/5038
Laurent Berger
Gebhard Bckle
Lassina Dembl
Mladen Dimitrov
Tim Dokchitser
John Voight
Elliptic Curves,
Hilbert Modular Forms and
Galois Deformations
Editors for this volume:
Henri Darmon (McGill University)
Fred Diamond (Kings College London)
Luis V. Dieulefait (Universitat de Barcelona)
Bas Edixhoven (Universiteit Leiden)
Vctor Rotger (Universitat Politcnica de Catalunya)
Laurent Berger
UMPA-ENS Lyon
Lyon, France
Lassina Dembl
Warwick Mathematics Institute
University of Warwick
Coventry, UK
Tim Dokchitser
Department of Mathematics
University of Bristol
Bristol, UK
Gebhard Bckle
Interdisciplinary Center for Scientific
Computing (IWR)
Universitt Heidelberg
Heidelberg, Germany
Mladen Dimitrov
Cit Scientifique, UFR Mathmatiques
Universit Lille 1
Villeneuve dAscq, France
John Voight
Department of Mathematics and Statistics
University of Vermont
Burlington, VT, USA
ISBN 978-3-0348-0617-6
ISBN 978-3-0348-0618-3 (eBook)
DOI 10.1007/978-3-0348-0618-3
Springer Basel Heidelberg New York Dordrecht London
Library of Congress Control Number: 2013941399
Mathematical Subject Classification (2010): Primary: 11F80, 11G18, 11G05, 11G07; Secondary: 11R23, 11F41
Springer Basel 2013
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now
known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection with
reviews or scholarly analysis or material supplied specifically for the purpose of being entered and executed
on a computer system, for exclusive use by the purchaser of the work. Duplication of this publication or parts
thereof is permitted only under the provisions of the Copyright Law of the Publishers location, in its current
version, and permission for use must always be obtained from Springer. Permissions for use may be obtained
through RightsLink at the Copyright Clearance Center. Violations are liable to prosecution under the
respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publication,
neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors or
omissions that may be made. The publisher makes no warranty, express or implied, with respect to the
material contained herein.
Printed on acid-free paper
Springer Basel i s part of Springer Science+Business Media (www.birkhauser-science.com)
Foreword
A research programme in Arithmetic Geometry was organized at the Centre de Recerca Matemàtica by F. Bars (Universitat Autònoma de Barcelona), L. Dieulefait
(Universitat de Barcelona) and V. Rotger (Universitat Politècnica de Catalunya)
in the academic year 20092010. It provided an excellent opportunity to bring together specialists and researchers in this eld and germane lines of research, having
expertise in dierent but always overlapping aspects of this beautiful branch of
mathematics. The aim of the research programme was to learn about the details
of major developments in the area, to explore new directions and perspectives,
to promote fruitful collaborations among participants and to serve as a valuable
centre for the advancement of young researchers.
The ultimate goal of the research programme was to explore several key objects from the arithmetic of number elds/function elds which constitute a living
proof of this fruitful cross-fertilization between the two worlds, such as Drinfeld
and classical modular varieties, classical and p-adic L-functions, the Langlands
programme, the Birch and Swinnerton-Dyer conjecture, Iwasawa theory, and the
BlochKato conjecture.
Already in the 19th century, mathematicians had realized the striking similarities between number elds and function elds; later, prominent number theorists
such as Artin and Weil made essential use of function elds of one variable over
a nite eld to reach a deeper understanding of L-functions and the Riemann
hypothesis. Here we can see a rst instance of a recurrent theme in the number
elds/function eld dialectic: the transposition into one setting of questions rst
asked over the other (often a far from trivial task) allows us to gain new insight
and, in many cases, to progress more than in the situation previously considered,
thanks to a richer structure to be exploited. In particular, function elds have
become a standard testing ground to get evidence for hard conjectures proposed
over number elds.
The rst and third terms of the programme focused on the arithmetic of elliptic curves and more general motives, modular forms, automorphic representations,
L-functions and Shimura varieties over number elds.
During the last few years there have been fundamental breakthroughs in
the subject. Several long-standing conjectures have been settled and fundamental progress has been achieved towards others. In addition, precise variants and
vi
Foreword
generalizations of these conjectures have been raised, and these demand renements of the existing methods as well as completely new ideas in order to be
approached.
Some of these major achievements take the form of modularity results, which
aim to establish a bridge between algebraic varieties or abstract geometric Galois
representations on one side and modular or automorphic forms on a Shimura
variety on the other. Prominent instances of this philosophy are Serres Modularity
Conjecture and the FontaineMazur Conjecture, among others.
In parallel, the diophantine study of algebraic varieties has received a dramatic impulse over the last years provided they can be shown to be modular.
Thanks to the celebrated result of Wiles and his collaborators, elliptic curves over
the eld of rational numbers provide the simplest (but highly non-trivial) examples
of this phenomenon, and their arithmetic still oers deep questions which remain
unsolved. The construction of rational points on them, the Birch and SwinnertonDyer Conjecture, its p-adic formulations and generalizations to higher dimensional
abelian varieties over number elds are certainly one of them. So far, most insights
to these problems exploit and are conditional to the fact that they appear as a
factor of the Jacobian of a suitable modular or Shimura curve.
Computation has played a crucial role in the formulation of some of the
most important conjectures involving modular forms and elliptic curves in the last
decades. Nowadays a wide variety of algorithms, packages and software is available
and, for this reason, the research programme also focused on the computational
side and explicit approaches to the subject.
Several workshops and advanced courses on the above topics were organized
during this research programme at the Centre de Recerca Matemàtica. This volume collects the notes of some of the lecture series delivered during the rst and
third terms of the programme. What follows is a summary of them.
The notes by L. Berger provide an introduction to p-adic Galois representations and Fontaine rings. These are useful in particular to describe many local
deformation rings at p considered in Galois deformation theory (see the notes of
B
ockles course).
The notes by G. B
ockle give a comprehensive course on Galois deformation
theory, starting from the foundational results of Mazur and discussing in detail
the theory of pseudo-representations and their deformations, local deformations
at places = p and local deformations at p with geometric conditions (in the sense
of the FontaineMazur conjecture). The nal chapter of these notes presents the
results of Bockle and Kisin on presentations of global deformation rings over local
ones, from which a very useful lower bound for global deformation rings with local
conditions can be derived.
The notes by M. Dimitrov contain the basics of the arithmetic theory of
Hilbert modular forms and varieties, with an emphasis on the study of the images
of the attached Galois representations, on modularity lifting theorems over totally
real number elds and on the cohomology of Hilbert modular varieties with integral
coecients.
Foreword
vii
The notes by L. Dembele and J. Voight describe methods for performing explicit computations in spaces of Hilbert modular forms. These methods depend on
the JacquetLanglands correspondence and on computations in spaces of quaternionic modular forms, both for the case of denite and indenite quaternion algebras. The notes contain several examples and include applications to modularity
of Galois representations.
The notes by T. Dokchitser describe the proof, obtained by the author in joint
work with V. Dokchitser, of the parity conjecture for elliptic curves over number
elds under the assumption of niteness of the TateShafarevich group. The notes
contain the statement of the Birch and Swinnerton-Dyer conjecture and a detailed
study of local and global root numbers of elliptic curves and their classication.
Bellaterra, December 2012
Contents
Foreword
Galois Deformations
On p-adic Galois Representations

Laurent Berger
Introduction . . . . . . . . . . . . . . . . . . . .
1
The Galois group of Q . . . . . . . . . . .
2
Ramication of local elds, I . . . . . . .
3
p-adic representations with = p . . . . .
4
Rings of periods . . . . . . . . . . . . . .
5
Galois cohomology . . . . . . . . . . . . .
6
Ramication of local elds, II . . . . . . .
7
Cyclotomic extensions . . . . . . . . . . .
8
The cohomology of Cp . . . . . . . . . . .
9
Witt vectors . . . . . . . . . . . . . . . . .
+ . . . . . . . . . . .
+ and B
10 The rings E
11 The eld BdR . . . . . . . . . . . . . . . .
12 De Rham representations . . . . . . . . .
13 The rings Bmax and Bst . . . . . . . . . .
14 Crystalline and semistable representations
15 Admissible ltered (, N )-modules . . . .
16 The groups H1 (K, V ) . . . . . . . . . . .
17 A p-adic period pairing . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Deformations of Galois Representations

Gebhard B
ockle
Introduction . . . . . . . . . . . . . .
Notation . . . . . . . . . . . . . . . .
1
Deformations of representations
1.1
Deformation functors . .
1.2
A niteness condition .
3
3
4
4
5
6
7
7
8
9
10
11
11
12
13
14
16
17
19
21
. . . . . . . . . . .
. . . . . . . . . . .
of pronite groups
. . . . . . . . . . .
. . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
21
24
26
26
27
ix
Contents
1.3
Representability . . . . . . . . . . . . . . . . . . . . . . . .
1.4
The tangent space . . . . . . . . . . . . . . . . . . . . . . .
1.5
Presentations of the universal ring RVF . . . . . . . . . . . .
1.6
Groupoids over categories . . . . . . . . . . . . . . . . . . .
1.7
Appendix: Schlessingers axioms . . . . . . . . . . . . . . .
1.8
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Deformations of pseudo-representations . . . . . . . . . . . . . . .
2.1
Quotients by group actions . . . . . . . . . . . . . . . . . .
2.2
Pseudo-representations . . . . . . . . . . . . . . . . . . . . .
2.3
Deformations of pseudo-representations . . . . . . . . . . .
2.4
Deforming a representation and the pseudo-representation
Tr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5
Representable subgroupoids of RepF . . . . . . . . . . . .
2.6
Completions of RepF . . . . . . . . . . . . . . . . . . . . .
2.7
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.8
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Deformations at places not above p and ordinary deformations . .
3.1
The generic ber of a deformation functor . . . . . . . . . .
3.2
WeilDeligne representations . . . . . . . . . . . . . . . . .
3.3
Deformation rings for 2-dimensional residual representations
of GF and their generic ber . . . . . . . . . . . . . . . . .
3.4
Unramied deformations for = p . . . . . . . . . . . . . .
3.5
Deformations of Steinberg type for = p . . . . . . . . . . .
3.6
On the proof of Theorem 3.3.1 . . . . . . . . . . . . . . . . .
3.7
Ordinary deformation at p . . . . . . . . . . . . . . . . . . .
3.8
Complements . . . . . . . . . . . . . . . . . . . . . . . . . .
3.9
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Flat deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1
Flat deformations . . . . . . . . . . . . . . . . . . . . . . .
4.2
Weakly admissible modules and smoothness of the generic
ber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3
The FontaineLaaille functor and smoothness when e = 1
. . . . . . . . . . . . . . . . . . . . .
4.4
The dimension of DVat
F
4.5
Complements . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.7
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Presenting global over local deformation rings . . . . . . . . . . . .
5.1
Tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . .
5.2
Relative presentations . . . . . . . . . . . . . . . . . . . . .
5.3
Numerology . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.4
Geometric deformation rings . . . . . . . . . . . . . . . . .
5.5
Odd deformations at real places . . . . . . . . . . . . . . . .
27
29
30
31
35
35
36
37
38
42
44
47
50
52
56
58
58
61
64
65
66
71
72
76
77
84
85
85
86
89
91
94
95
99
100
102
103
105
105
107
Contents
xi
5.6
Proof of Key Lemma 5.2.2. . . . . . . . . . . . . . . . . . . . 107
5.7
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
II
117
Arithmetic Aspects of Hilbert Modular Forms and Varieties

Mladen Dimitrov
119
Hilbert modular forms . . . . . . . . . . . . . . . . . . . . . . .

1.1
Congruence subgroups . . . . . . . . . . . . . . . . . . .
1.2
Hilbert modular forms as automorphic forms on GL2 (A)
1.3
Hecke operators and newforms . . . . . . . . . . . . . .
2
Galois representations associated to Hilbert modular forms . .
2.1
Galois representations . . . . . . . . . . . . . . . . . . .
2.2
Images of Galois representations . . . . . . . . . . . . .
2.3
Modularity lifting theorems . . . . . . . . . . . . . . . .
3
Hilbert modular varieties . . . . . . . . . . . . . . . . . . . . .
3.1
Denition . . . . . . . . . . . . . . . . . . . . . . . . . .
3.2
Connected components . . . . . . . . . . . . . . . . . .
3.3
Cusps and compactications . . . . . . . . . . . . . . .
3.4
Smoothness . . . . . . . . . . . . . . . . . . . . . . . . .
3.5
Etale
coverings . . . . . . . . . . . . . . . . . . . . . . .
3.6
Integral models . . . . . . . . . . . . . . . . . . . . . . .
3.7
Betti cohomology with p-adic coecients . . . . . . . .
3.8
Hecke correspondences . . . . . . . . . . . . . . . . . . .
3.9
Poincare duality . . . . . . . . . . . . . . . . . . . . . .
4
Cohomology of Hilbert modular varieties . . . . . . . . . . . . .
4.1
Freeness results . . . . . . . . . . . . . . . . . . . . . . .
4.2
Results on morphisms . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Explicit Methods for Hilbert Modular Forms

Lassina Dembele and John Voight
Introduction . . . . . . . . . . . . . . . . . . .
1
Classical (elliptic) modular forms . . . .
2
Classical Hilbert modular forms . . . . .
3
Quaternionic modular forms . . . . . . .
4
Denite method . . . . . . . . . . . . . .
5
Indenite method . . . . . . . . . . . . .
6
Examples . . . . . . . . . . . . . . . . .
7
Adelic quaternionic modular forms . . .
8
Denite method, arbitrary class number
119
120
120
121
122
122
123
124
124
124
125
126
126
127
129
129
130
130
131
132
132
133
135
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
135
136
139
143
147
153
163
173
183
xii
Contents
9
Indenite method, arbitrary class number . . . . . . . . . . . . . . 187
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
III
Elliptic Curves
199
Notes on the Parity Conjecture

Tim Dokchitser
1
BirchSwinnerton-Dyer and parity . . . . . . . . . . . . . .

1.1
Conjectures and the main result . . . . . . . . . . .
1.2
BirchSwinnerton-Dyer II and isogeny invariance . .
1.3
Parity example . . . . . . . . . . . . . . . . . . . . .
2
The p-parity conjecture . . . . . . . . . . . . . . . . . . . .
2.1
Proof of Theorem A . . . . . . . . . . . . . . . . . .
2.2
Local formulae for the Selmer parity . . . . . . . . .
2.3
Parity in S3 -extensions . . . . . . . . . . . . . . . . .
2.4
Brauer relations and regulator constants . . . . . . .
2.5
Parity in dihedral extensions . . . . . . . . . . . . .
2.6
The KramerTunnell theorem . . . . . . . . . . . . .
3
L-functions and root numbers . . . . . . . . . . . . . . . . .
3.1
L-functions . . . . . . . . . . . . . . . . . . . . . . .
3.2
WeilDeligne representations . . . . . . . . . . . . .
3.3
Epsilon-factors . . . . . . . . . . . . . . . . . . . . .
3.4
Root numbers of elliptic curves . . . . . . . . . . . .
3.5
Root numbers of elliptic curves with an l-isogeny . .
4
Parity over totally real elds . . . . . . . . . . . . . . . . .
5
The 2-isogeny theorem . . . . . . . . . . . . . . . . . . . . .
5.1
Complex places . . . . . . . . . . . . . . . . . . . . .
5.2
Real places . . . . . . . . . . . . . . . . . . . . . . .
5.3
The correction term . . . . . . . . . . . . . . . . . .
5.4
Good reduction at v 2 . . . . . . . . . . . . . . . .
5.5
Split multiplicative reduction at v 2 . . . . . . . . .
5.6
Nonsplit multiplicative reduction at v 2 . . . . . . .
5.7
Deforming to totally real elds . . . . . . . . . . . .
6
The p-isogeny conjecture . . . . . . . . . . . . . . . . . . . .
7
Local compatibility in S3 -extensions . . . . . . . . . . . . .
8
Parity predictions . . . . . . . . . . . . . . . . . . . . . . . .
8.1
Semistable curves in cubic extensions . . . . . . . . .
8.2
Number elds K such that w(E/K) = 1 for all E/Q
8.3
Goldfelds conjecture over Q . . . . . . . . . . . . . .
8.4
No Goldfeld over number elds . . . . . . . . . . . .
8.5
No local expression for the rank . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
201
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
201
201
203
204
206
207
208
209
211
213
214
215
216
220
220
223
224
226
228
229
229
229
231
231
233
234
235
237
238
239
240
240
241
242
245
Part I
Galois Deformations
On p-adic Galois
Representations
Laurent Berger
Introduction
These are the notes for my part of the CRM course p-adic Galois Representations
and Global Galois Deformations. My aim was to give a short introduction to the
p-adic Hodge theory necessary for formulating the local conditions imposed on
deformations of p-adic representations. I also included some material on the technical tools used for proving the properties of Fontaines rings of periods, although
I omitted most proofs of results. In these notes there are a few changes from the
actual course; for example, I exchanged and p at various places in order to follow
the notation of the other courses.
The Galois group of Q
Let E be a nite extension of Qp and let GQ = Gal(Q/Q). A p-adic representation

of GQ is a nite dimensional continuous E-linear representation V of GQ . We wish
to study p-adic representations of GQ , either individually or in families.
Let be a prime number (which may or may not be equal to p) and let
be a place of Q above . The group D = {g GQ such that g() = } is
the decomposition group of . If is another place above , then D and D
are conjugate and we write D for the resulting group, which is well-dened up
to conjugation in GQ .
The choice of is equivalent to the choice of an embedding of Q into Q and
this gives rise to a map D GQ , which is easily seen to be an isomorphism.
The groups GQ are easier to understand than GQ , thanks to ramication theory
(recalled in 2).
L. Berger et al., Elliptic Curves, Hilbert Modular Forms and Galois Deformations, Advanced Courses
in Mathematics - CRM Barcelona, DOI 10.1007/978-3-0348-0618-3_1, Springer Basel 2013
Given a p-adic representation V , one then studies its restriction to D for

various primes , and the following result says that we do not lose too much
information when doing so.
Proposition 1.1. If S is a set of prime numbers of density 1 and V is a semisimple
representation of GQ , then V is determined by its restriction to the D with S.
Ramication of local elds, I
Let be a prime number, let K be a nite extension of Q , and let OK , mK , kK

and K denote its ring of integers, maximal ideal, residue eld and a uniformizer,
respectively. Let K unr denote the maximal unramied extension of K and let
K tame denote the maximal tamely ramied extension of K.
and topologically generated by
The group Gal(F /kK ) is isomorphic to Z
m
Frm = x x , where m = Card(kK ). The inertia subgroup IK of GK is the kernel
of the natural map GK Gal(F /kK ), and we then have IK = Gal(Q /K unr ).
Likewise, we have
1/n
K tame = n K unr K ,
so that Gal(K tame /K unr ) = limn n , where the map is given by
1/n 1/n
g g K /K n1 .
In particular, if Gal(K tame /K unr ) and the image of Gal(K tame /K) in
()
Gal(F /kK ) is Frm , then 1 = m . Finally, IK = Gal(Q /K tame ) is the
-Sylow subgroup of IK , called the wild inertia subgroup.
p-adic representations with = p
An easy corollary of the equation 1 = m is Grothendiecks monodromy

theorem:
Theorem 3.1. If V is a p-adic representation of GK , with K as above, and = p,
then there exists a nite extension L of K such that V |IL is unipotent.
We say that a p-adic representation V of GK has good reduction if V |IK
is trivial, and we say that V is semistable if V |IK is unipotent. Grothendiecks
theorem above then says that every p-adic representation of GK is potentially
semistable (recall that = p).
There is a useful way of describing the p-adic representations of GK . Let K
dened by g = Frn(g) . The Weil
be as before, so that there is a map n : GK Z
m
group WK is {g GK such that n(g) Z}. A WeilDeligne representation is the
datum of a representation V of WK (given by a map : WK End(V )) and of a
nilpotent map N End(V ) such that N (g) = mn(g) (g)N .
4. Rings of periods
Choose a compatible sequence {n }n0 of primitive n -th roots of 1, and let
1/n
t(g) 1/n
t : IK Z be the map determined by g(K ) = n K . Choose also GK
such that n() = 1. If V is a p-adic representation of GK , then, by Grothendiecks
theorem, there exists a nite extension L of K such that (g) is unipotent if g IL .
In this case, the map N = log (g)/t(g) End(V ) is well-dened and independent
of g IL . We attach to V a WeilDeligne representation (WD , NWD ) on the same
underlying space V by the formulas WD (w) = (w) exp(t( n(w) w) N ) and
NWD = N . The isomorphism class of the resulting representation does not depend
on the choices made, and we can easily recover V from (WD , NWD ).
Rings of periods
From here to the end of these notes, we assume that = p, so that K is now a nite
extension of Qp . We would like to have a classication of p-adic representations of
GK similar to the one above, but this is harder to obtain. Indeed, let : GK Z
p
be the cyclotomic character, dened by
(g)
g(pn ) = pn .
We can write = with p1 and 1+pZp , and we can then consider
r s with r Z/(p 1)Z and s Zp . All such characters are representations
of GK , but it turns out that they are good only if s Z. It is hard to distinguish
such characters merely by looking at their image or kernel, and in order to classify
p-adic representations, one therefore needs more than mere ramication theory.
The main tool for doing so is Fontaines construction of rings of periods.
A ring of periods is a Qp -algebra B endowed with an action of GK and, possibly,
some supplementary structures compatible with the action of GK (for example,
a ltration, a Frobenius map, a monodromy map, etc.). We require that B is a
domain, that (1) Frac(B)GK = B GK and that (2) if y B is such that Qp y is
stable under GK , then y B . For example, all these conditions are automatically
fullled if B is a eld.
If V is a p-adic representation of GK , we then dene DB (V ) = (B Qp V )GK ,
which is a B GK -vector space. There is a natural map
: B B GK DB (V ) B Qp V
and condition (1) above implies that is injective, so that DB (V ) is of dimension
dimQp (V ). We say that V is B-admissible if DB (V ) is of dimension dimQp (V ).
By condition (2) above, this is the case if and only if is surjective. If V is E-linear,
then we say that it is B-admissible if the underlying Qp -linear representation is
B-admissible.
In this way, we have dened the subcategory of B-admissible p-adic representations of GK inside the category of all p-adic representations of GK . This
subcategory is stable under subquotients, direct sums, tensor products and duals.
If B has some supplementary structures, then these descend to DB (V ) and in this

way we obtain some nontrivial invariants of B-admissible representations, which
can then be used to classify them.
Galois cohomology
In this section, we recall a few facts about Galois cohomology groups. If V is a

p-adic representation of GK , then we write H i (K, V ) for H i (GK , V ). We also write
hi (V ) = dimQp H i (K, V ). Let V be the dual of V and let V (1) = V .
If i = 0, 1 or 2, then the cup product

: H i (K, V ) H 2i K, V (1) H 2 K, V V (1)
gives rise to a pairing H i (K, V ) H 2i (K, V (1)) H 2 (K, Qp (1)). We then have
the following theorem of Tate.
Theorem 5.1. The groups H i (K, V ) are nite dimensional Qp -vector spaces and
they are {0} if i 3. We have H 2 (K, Qp (1)) = Qp , the pairing
H i (K, V ) H 2i (K, V (1)) Qp
is perfect, and h0 (V ) h1 (V ) + h2 (V ) = [K : Qp ] dim(V ).
p
= yn . Let
If y K , let {yn }n0 be a sequence such that y0 = y and yn+1
(y) : GK Zp be the map determined by the equation
(y)(g)
g(yn ) = pn
yn ,
so that (y)(gh) = (y)(g) + (g)(y)(h). We then have (y) H 1 (K, Qp (1)) and
the map y (y) is the Kummer map. It extends to a map

H 1 (K, Q (1)),
: Q p Zp K
p
which is an isomorphism by Kummer theory. In particular, we have h1 (Qp (1)) =
[K : Qp ] + 1, which is compatible with Theorem 5.1.
If B is a ring of periods, then W = B Qp V is a semilinear representation
of GK , i.e., it is a free B-module with a semilinear action of GK . If we choose a
basis of such a W , then g Mat(g) gives a cocyle on GK with values in GLd (B),
and choosing a dierent basis gives a cohomologous cocycle. In this way, we get
[W ] H 1 (GK , GLd (B)), and the original representation V is then B-admissible if
and only if [B Qp V ] is the trivial cohomology class. The following result, known
as Hilberts theorem 90, is then useful.
Theorem 5.2. If L/K is nite Galois, then H 1 (Gal(L/K), GLd (L)) = {0}.
6. Ramication of local elds, II
As a consequence, we see for example that, if Qp B, then potentially

B-admissible representations are already B-admissible.
We can deduce from Theorem 5.2 that H 1 (Gal(Fp /kK ), GLd (Fp )) = {0},
and an argument of successive approximations then shows that

unr ) = {0}.
H 1 GK /IK , GLd (Q
p
unr -admissible.
In this way we see that unramied representations of GK are Q
p
Ramication of local elds, II
In this section, we collect various statements about the ramication of extensions

of Qp , which are useful for proving some of the properties of Fontaines rings of
periods. We recall in particular a few facts about the conductor and the dierent
of a nite extension K/F .
Let valK () be normalized by valK (K ) = Z. Recall that, if u 1, then
one denes the ramication ltration Gal(K/F )u = {g Gal(K/F ) such that
valK (gx x) u + 1 for all x OK }. Herbrand dened a function K/F such
that, if we dene
Gal(K/F )v = Gal(K/F )K/F (v) ,
then Gal(K/F )v is the image of Gal(L/F )v whenever L is an extension of K. One
can then dene GvF for v 1.
u
If K/F is Galois, then we dene K u = K Gal(K/F ) and, if K is not Galois,
then we set K u = Lu K, where L/F is Galois and contains K. For example, we
have
Gal(Qp (pn )/Qp )i = Gal(Qp (pn )/Qp (pi ))
and therefore Qp (pn )u Qp (pu ). The conductor of K (with respect to F ) is
the inmum of the real numbers u with K u = K.
1 , where O
K is the dual of
Recall also that we have the dierent dK/F = O
K
OK with respect to the pairing (x, y) TrK/F (xy). Dierent and conductor are
related by the following formula.
Proposition 6.1. We have

valp (dK/F ) =
1
1
[K : K u ]

du.
Cyclotomic extensions
Let F = Qp , let Fn = Qp (pn ) for n 1, and let F = n1 Fn . We know

that Fn is a totally ramied extension of F of degree pn1 (p 1), and also that
OFn = Zp [pn ]. If n 1 and y F , then y Fn+k for some k 0, and
Rn (y) = pk TrFn+k /Fn (y) does not depend on k. The map Rn : F Fn is
then a GF -equivariant projection. We have Rn (1) = 1, while Rn (pn+k ) = 0 if

1 n pk 1, so that Rn (OFn+k ) OFn . This implies that, if y F , then
valp (Rn (y)) valp (y)1/(pn1 (p1)), and therefore that Rn extends by uniform
continuity to a projection Rn : F Fn . In addition, we have Rn (y) = y if y F
and n 0, so that, if y F , then Rn (y) y as n .
Let K be a nite extension of Qp , let Kn = K(pn ) for n 1, and let
K = n1 Kn . If n 0, then Kn+1 /Kn is totally ramied of degree p, and
Kn /Fn is of degree d = K /F if n n(K). Proposition 6.1 and the fact
that Fnu Fu can be used to show that the sequence {pn valp (dKn /Fn )}n1 is
eventually constant. In particular, if > 0, then there exist n() n(K) such
that, if n n(), then valp (dKn /Fn ) . This implies that, if n n(), then there
exists a basis e1 , . . . , ed of OKn over OFn such that valp (ei ) .
d
If y OKn+k , then we can write y = j=1 yj ej , where yj = TrK /F (yej )
d
belongs to OFn+k , and we set Rn (y) = j=1 Rn (yj )ej . The resulting map
Rn : K Kn
is then a GK -equivariant projection which satises
valp (Rn (y)) valp (y) 1/(pn1 (p 1)) ,
Kn
and therefore Rn extends, by uniform continuity, to a projection Rn : K
such that Rn (y) y as n as above.
The cohomology of Cp
Let Cp be the p-adic completion of Qp , so that Cp is a complete and algebraically

closed eld. If L is a subeld of Qp , then the action of GL on Qp extends by
continuity to Cp and we have the following result of AxSenTate.
L

Theorem 8.1. If L Qp , then CG
= L.
p
If L is as above and Qp , then we set L () = inf gGL valp (g()). The

main ingredient of the proof of Theorem 8.1 is the following result of Le Borgne,
which improves upon a similar result of Ax.
Lemma 8.2. If Qp , then there exists L with valp () L ()1/(p1).
Let : GK Z
p be a character which is trivial on ker() = Gal(Qp /K )
(for example, one could take = h with h Z).
Theorem 8.3. If has innite order, then H 0 (K, Cp ()) = {0}.
If has nite order, then Hilberts Theorem 90 implies that Cp () = Cp ,
and then H 0 (K, Cp ()) = K by Theorem 8.1 above. We now give a sketch of
9. Witt vectors
the proof of Theorem 8.3. If H 0 (K, Cp ()) = {0}, then there exists a nonzero
y Cp such that g(y) = (g)y for g GK . We apply the maps Rn from 7; since
Rn (y) y, we have Rn (y) = 0 for n 0. The formula g(Rn (y)) = (g)Rn (y)
now implies that is trivial on Gal(Qp /Kn ), and therefore has nite order.
By proving more rened results about K , one can also prove that one has
H 1 (K, Cp ()) = {0} if has innite order. Finally, H 1 (K, Cp ) is a 1-dimensional
K-vector space generated by [g logp (g)].
Witt vectors
We say that a ring R is perfect if p = 0 and x xp is a bijection on R. We

say that a ring A is a perfect p-ring if p is not a zero divisor, A is separated and
complete for the p-adic topology, and A/pA is perfect. If x A/pA, we denote by
i
1/p
xpi }i0 then
x
a lift of x to A. Let x0 = x and let xi+1 = xi . The sequence {
converges to an element [x] A, which is independent of all choices
and is called
the Teichm
uller lift of x. Every element of A can be written as i0 pi [xi ] in a
unique way.
1/p 1/p
Let R = Fp X i
,Y i
, and let S be the p-adic completion of the
i0
1/p 1/p
, Yi
, so that S is a perfect p-ring with residue ring R. There
ring Zp Xi
i0
exist elements {Si }i0 and {Pi }i0 of R such that

pi X i +
pi Y i =
pi [Si ],
i0
i0
p Xi
i
i0
i0
p Yi =
i0
pi [Pi ].
i0
If A is a perfect p-ring and {xi }i0 and {yi }i0 are two sequences of elements
of R, then we have a map : S A given by (Xi ) = [xi ] and (Yi ) = [yi ]. By
applying to the two equations above, we see that

pi [xi ] +
pi [yi ] =
pi Si (x, y) ,
i0

i0
i0
pi [xi ]
i0
pi [yi ] =
i0

pi Pi (x, y) ,
i0
so that addition and multiplication of elements of A, written as

given by universal formulas.

i0
pi [xi ], are
Theorem 9.1. If R is a perfect ring, then there exists a unique perfect p-ring W (R)
such that W (R)/pW (R) = R.
The discussion above shows that one can take

pi [xi ] with xi R ,
W (R) =
i0
10
addition and multiplication being given by the universal formulas. The ring W (R)
is called the ring of Witt vectors over R.
Proposition 9.2. If R is a perfect ring and A is complete for the p-adic topology
and f : R A/pA is a homomorphism, then f lifts to a unique homomorphism
W (f ) : W (R) A.
In the notation of the beginning of this section, we must have
(xn ))p ,
W (f )([x]) = lim (f
n
and it remains to check that this does give a ring homomorphism. For example,
the map R R given by x xp gives rise to the Frobenius map on W (R).
Finally, if R is equipped with
a valuation val(), then we can dene semivaluations wk () on W (R) by wk ( i0 pi [xi ]) = minik val(xi ). The weak topology
of W (R) is the one dened by these semivaluations.
Proposition 9.3. If R is complete for val(), then W (R) is complete for the weak
topology.
+ and B
+
10 The rings E
Fix some 0 < < 1/(p 1) and let I = {x OCp with valp (x) 1/(p 1) }.
+ = {(x0 , x1 , . . .) where xi OC /I and xp = xi }, so that E
+ is
We dene E
p
i+1
I
I
a perfect ring (addition and multiplication being termwise). We have a map from
+ , which can be shown
{(x(0) , x(1) , . . .) where x(i) OCp and (x(i+1) )p = x(i) } to E
I
+ does not depend on I and we denote it by E
+ . If
to be a bijection, so that E
I
+
(0)
+
, we set valE (x) = valp (x ), and this denes a valuation on E
for which
xE
it is complete.
n
+ , and this gives an injective map Fp E
+.
If Fp , then ([1/p ])n0 E
+,
The choice of a sequence {pn }n0 gives rise to an element = (1, p , . . .) E

and we dene = 1, so that valE () = p/(p 1). In particular, Fp [[]] E+ .
The theorem below is not needed in the sequel, but it gives an idea of the structure
+.
of E
+ [1/] is the completion of the algebraic closure of
Theorem 10.1. The eld E
Fp (()).
+ which we have given has the adThe more complicated denition of E
+
is equipped with an action of GQ . We then set
vantage of showing that E
p
+
+
+

+ [1/p], and both rings are also equipped with an

A = W (E ) and B = A
+ OC /p
action of GQp , as well as the Frobenius map . The homomorphism E
p
+ OC given explicitly by
extends,
by
Theorem
9.2,
to
a
homomorphism
:
A
p

i
i (0)
i0 p [xi ] =
i0 p xi . For example, ([] 1) = 0.
11. The eld BdR
11
Proposition 10.2. The ideal ker() is generated by any element y ker() such
that valE (y) = 1.
This is the case with y = ([] 1)/([1/p ] 1) (Fontaines element ), or with
+ is such that p(0) = p.
y = [
p] p, where p E
11
The eld BdR
+ / ker()h
+ be the ring constructed in 10 and, for h 1, let Bh = B
Let B
+
(in particular, we have B1 = Cp ). We let BdR = limh1 Bh , so that B+
dR is a
complete local ring with maximal ideal ker() and residue eld Cp , and is also
equipped with an action of GQp . An element y B+
dR is invertible if and only if
,
since
([1/p ] 1) = 0. We dene
(y) = 0. For example, ker() = ([] 1)B+
dR
+
BdR = Frac(BdR ), so that it is a ring of periods equipped with the additional
structure of the ltration given by Fili BdR = ker()i .
The series ([] 1) ([] 1)2 /2 + ([] 1)3 /3 converges to an element
+
(g)
,
t B+
dR which also generates ker(), so that BdR = BdR [1/t]. Since g() =
we have g(t) = (g)t.
Remark 11.1. The ring B+
dR is isomorphic to Cp [[t]], but only as abstract rings,
and there is no such isomorphism which is compatible with the action of GQp (as
we will see in 16).
The ring B+
dR is complete for the ker()-adic topology, but it is also complete
for a ner topology. Each ring Bh is a Banach space (the unit ball being the image
+ ), and this gives B+ the structure of a Frechet space. Note that there is no
of A
dR
such thing as a p-adic topology on B+
dR .
If P (X) Qp [X] is a polynomial with simple roots, then it splits completely in Cp and hence, by Hensels lemma, it also splits completely in B+
dR , since
+
+
/tB
=
C
.
In
this
way
we
see
that
Q
B
.
A
theorem
of
Colmez
shows
B+
p
p
dR
dR
dR
+
that, actually, Qp is dense in BdR for its Frechet topology.
K
Proposition 11.2. We have BG
dR = K.
To prove this, we write the exact sequence

h +
0 th+1 B+
dR t BdR Cp (h) 0
and use the computation of H 0 (K, Cp (h)) carried out in 8.
12
De Rham representations
We now carry out the constructions of 4 with B = BdR . If V is a p-adic representation of GK , then we set DdR (V ) = (BdR Qp V )GK , which is a ltered K-vector
space (if V is E-linear, then DdR (V ) is an E Qp K-module). We say that V is
12
de Rham if it is BdR -admissible. Note that, since Qp BdR , Theorem 5.2 implies
that potentially de Rham representations are de Rham. If V is de Rham, then a
HodgeTate weight of V is an integer h such that Filh DdR (V ) = Filh+1 DdR (V ).
The functor DdR : {de Rham representations} {ltered K-vector spaces}
forgets a lot of information about V . For instance, if V is potentially unramied, then it is de Rham, but then DdR (V ) is the ltered vector space for which
Fil0 DdR (V ) = DdR (V ) and Fil1 DdR (V ) = {0}.
The following theorem of Faltings proves a conjecture of Fontaine and shows
that representations of GK coming from geometry are de Rham.
i
(XQp , Qp ), then
Theorem 12.1. If X is proper and smooth over K, and V = Het
i
(X/K).
V is a de Rham representation of GK and DdR (V ) = HdR
Conversely, we have the following conjecture of Fontaine and Mazur. If F is

a number eld, then we say that a representation of GF comes from geometry if
it is a subquotient of the etale cohomology of some algebraic variety over F .
Conjecture 12.2. If F is a number eld and V is an irreducible p-adic representation of GF which is unramied at almost every place of F and is de Rham at
every place of F above p, then V comes from geometry.
If, in addition, dim(V ) = 2 and F = Q, then we actually expect V to come
from a modular form; this has been proved in most cases by Emerton and Kisin.
13
The rings Bmax and Bst
Recall that in 10 we constructed the ring

+
k
+

=
p [xk ] where xk E
.
B
k
+ by the formula
If r 0, then we dene a valuation V(, r) on B
V(x, r) = inf valE (xk ) + k
k
pr
,
p1
+ for V(, r) (note that, more

[0;r] to be the completion of B
and we dene B
[r;s] , which explains the heavy notation).
generally, one can dene some rings B
[0;r] . The ring B+ is B
[0;r ] ,
[0;s] B
If s r, then we have an injective map B
max
0
+
unr

where r0 = (p 1)/p. It contains B (and hence Qp ), but also the element t
[0;r] for all r > 0), and we set Bmax = B+ [1/t].
dened in 11 (which belongs to B
max
+
+ gives rise to a bijection : B
[0;r ] B
[0;pr ] ,
B
The Frobenius map : B
0
0
+
B
.
We
use
the
ring
B
and hence to an injective map : B+
max instead
max
max
of Fontaines Bcris for technical reasons, but they are almost equal; for example,
(Bmax ) Bcris Bmax .
14. Crystalline and semistable representations
13
+ and there + Bh is continuous for the valuation V(, r0 ) on B

The map B
+
+
fore it extends to a continuous map Bmax BdR . Recall that if K is a nite
unr
be the maximal unramiextension of Qp , then K B+
dR . Let K0 = K Qp
ed extension of Qp contained in K, so that K0 B+
max .
+
Theorem 13.1. The natural map K K0 B+
max BdR is injective.
+ B+ is injective and, in
One can easily prove that the map K K0 B
dR
order to prove the theorem, one needs to show that the map remains injective
after completing the left-hand side, which is rather delicate.
As a corollary, we get that K K0 Bmax BdR is also injective, and using
GK
K
= K0 .
the fact that BG
dR = K, we get that Frac(Bmax )
+
+
Let u be a variable and let Bst = Bmax [u] and Bst = Bmax [u]. We extend
the action of GQp from Bmax to Bst by g(u) = u + a(g)t, where a(g) is dened by
n
n
a(g)
g(p1/p ) = pn p1/p . We also extend by (u) = pu, and we dene a monodromy
map N : Bst Bst by N = d/du, so that N = pN .
The series log([
p]/p) = log(1 + ([
p]/p 1)) converges in B+
dR and, if we choose
log(p) (usually, we choose log(p) = 0), then we can talk about log([
p]) B+
dR . We
+
+
+
p]), which is a
then extend the map Bmax BdR to Bst by sending u to log([
GQp -equivariant map.
+
Theorem 13.2. The natural map K K0 B+
st BdR is injective.
This implies that K K0 Bst BdR is injective and that Frac(Bst )GK = K0 .
Finally, we have the following result (condition (2) of the denition of a ring of
periods in 4).
unr
Theorem 13.3. If y Bst and Qp y is stable by GK , then y = y0 th with y0 Q
p
and h Z.
In particular, such a y actually belongs to Bmax and is invertible in Bmax .
14
Crystalline and semistable representations
We now carry out the constructions of 4 with B = Bmax or Bst . If V is a p-adic

representation of GK , then we set Dcris (V ) = (Bmax Qp V )GK and Dst (V ) =
(Bst Qp V )GK . They are both K0 -vector spaces; Dst (V ) is a (, N )-module and
Dcris (V ) = Dst (V )N =0 is a -module. We say that V is crystalline or semistable if
V is Bmax -admissible or Bst -admissible, respectively. Theorem 13.2 implies that
K K0 Dst (V ) injects into DdR (V ), so that, if V is semistable, then it is also
de Rham. The space Dst (V ) is then a ltered (, N )-module over K, that is, a
K0 -vector space D with an invertible semilinear endomorphism , an endomorphism N such that N = pN , and a ltration on DK = K K0 D. In the
next section, we will see how technical properties of Bmax and Bst translate into
properties of Dcris () and Dst ().
14
The property of being crystalline or semistable is the analogue of having

good reduction or being semistable for = p as in 3. For example, we have
the following result (due to Iovita for crystalline and to Breuil for semistable),
which is a p-adic analogue of the NeronOggShafarevich criterion for = p.
Theorem 14.1. If A is an abelian variety over K, then Vp A is crystalline if and
only if A has good reduction, and Vp A is semistable if and only if A has semistable
reduction.
We say that V is potentially semistable if there exists some nite Galois
extension L of K such that V |GL is semistable. In this case, Dst (V |GL ) is a ltered
(, N, Gal(L/K))-module over L. Potentially semistable representations are de
Rham, and we have the following result, which may be seen as a p-adic analogue
of Theorem 3.1.
Theorem 14.2. Every de Rham representation is potentially semistable.
Just as in 3, we can attach a WeilDeligne representation WD(V ) to a
potentially semistable representation V of GK . If D = Dst (V |GL ), then D is the
space of this representation, and NWD = N and WD (w) = whn(w) if w WK ,
where WK acts on D through Gal(L/K), and q = ph = Card(kK ). The fact that
N = pN implies that NWD WD (w) = q n(w) WD (w)NWD . Contrary to the
case = p, this WeilDeligne representation is not enough to recover V , since it
does not take into account the ltration.
If f is a modular eigenform, then one can attach to it a p-adic representation
Vp f , as well as a smooth admissible representation p f of GL2 (Qp ), and we then
have the following result of Saito.
Theorem 14.3. If f is a modular eigenform, then Vp f is potentially semistable,
and WD(Vp f ) is the WeilDeligne representation attached to p f by the local
Langlands correspondence.
If, in addition, p N , then Vp f is crystalline and the above theorem completely determines Dcris (Vp f ), because there is only one choice for the ltration (in
this case, Theorem 14.3 was previously proved by Scholl). We get Dcris (Vp f ) =
Dk,ap , where k = k(f ) and ap = ap (f ), and Dk,ap = Ee1 Ee2 with
Mat() =
15
pk1
1
ap

and
Fili Dk,ap
Dk,ap
= Ee1
{0}
if i 0,
if 1 i k 1,
if i k.
Admissible ltered (, N )-modules
By the constructions of the previous section, we have a functor Dst () from the category of semistable representations of GK to the category of ltered (, N )-modules
over K. In this section, we explain how technical properties of the ring Bst can be
15. Admissible ltered (, N )-modules
15
used to prove some properties of the functor Dst (). In particular, we will see that
it is fully faithful and will give a characterization of its image.
+
Theorem 15.1. We have B=1
max BdR = Qp .
As a corollary, we see that one can recover Qp from the ltered (, N )-module
=0,=1
structure of Bst , since we have Qp = BN
Fil0 BdR . In this way we get the
st
following full faithfulness result.
Corollary 15.2. The functor V Dst (V ) is fully faithful.
Indeed, if V is semistable, then Bst K0 Dst (V ) = Bst Qp V , so that

N =0, =1

Fil0 BdR K DdR (V ) .
V = Bst K0 Dst (V )
Let us now characterize the image of Dst (). If D is a 1-dimensional ltered
(, N )-module over K, we dene tN (D) to be valp (Mat()) and tH (D) to be the
integer h such that Filh DK = DK and Filh+1 DK = {0}. If D is of arbitrary
dimension, then we let tN (D) = tN (det D) and tH (D) = tH (det D).
If V is a semistable representation, then det Dst (V ) = Dst (det V ), and, in the
notation of Theorem 13.3, we have tN (Dst (V )) = tH (Dst (V )) = h.
If D is a 1-dimensional subobject of Dst (V ) and y D, then (y) = y for
some K0 of valuation tN (D), and y Filh DK for h = tH (D). As a corollary
h +
of Theorem 15.1, we get that if h valp () + 1, then B=
max t BdR = {0}.
This implies that tH (D) tN (D). If D is a subobject of Dst (V ) of dimension r,
then det D is a 1-dimensional subobject of Dst (r V ), and we can apply the above
reasoning to get again tH (D) tN (D).
We say that a ltered (, N )-module D over K is admissible if tH (D) =
tN (D) and tH (D ) tN (D ) for every subobject D of D.
Proposition 15.3. If V is a semistable representation, then Dst (V ) is an admissible
ltered (, N )-module over K.
Fontaine had conjectured that, conversely, every admissible ltered (, N )module over K is the Dst of some semistable representation V of GK , and this is
now a theorem of Colmez and Fontaine:
Theorem 15.4. The functor Dst () gives rise to an equivalence of categories between
the category of semistable representations of GK and the category of admissible
ltered (, N )-modules over K.
Thus, in principle, one can answer any question about a semistable representation by merely looking at the attached ltered (, N )-module. In practice, this
can be quite hard. For example, computing the reduction modulo p of the crystalline representation attached to the ltered -module Dk,ap given at the end of
14 is (as of January 2013) an open problem.
16
16
The groups H1 (K, V )
If V is a p-adic representation of GK , then H 1 (K, V ) classies extensions E of Qp

by V , that is, representations E inside an exact sequence 0 V E Qp 0.
More generally, extensions of Y by X are classied by H 1 (K, X Y ). Given
some property of representations, we are interested in the subset of H 1 (K, V )
corresponding to extensions having that property. In particular, we denote by
1
(K, V ) or Hg1 (K, V ) the classes of extensions which are crystalline,
Hf1 (K, V ), Hst
semistable or de Rham, respectively. If V is crystalline, then
Hf1 (K, V ) = ker(H 1 (K, V ) H 1 (K, Bmax Qp V )),
1
and Hg1 . The following result is an easy
and we have similar statements for Hst
consequence of Theorem 14.2, but had been proved before by Hyodo and was then
seen as evidence for Theorem 14.2.
1
(K, V ) = Hg1 (K, V ).
Theorem 16.1. If V is semistable, then Hst
We also dene He1 (K, V ) = ker(H 1 (K, V ) H 1 (K, B=1

max Qp V )). Recall
that, by Theorem 5.1, there is a perfect pairing H 1 (K, V ) H 1 (K, V (1)) Qp .
The following theorem of Bloch and Kato computes the orthogonals of the H1 .

Theorem 16.2. If V is crystalline, then we have Hf1 (K, V ) = Hf1 K, V (1) and

He1 (K, V ) = Hg1 K, V (1) .
One can compute the dimensions of the H1 (K, V ) by using the so-called
+
fundamental exact sequence 0 Qp B=1
max BdR /BdR 0, which, when
used along with the fact that 1 : Bmax Bmax is surjective, gives rise to
x((1)x, x)
0 Qp Bmax Bmax BdR /B+

dR 0.
By tensoring the above exact sequence with a crystalline representation V and
taking invariants under GK , one nds
0 V GK Dcris (V ) Dcris (V )DdR (V )/Fil0 DdR (V ) Hf1 (K, V ) 0,
GK
= DdR (V )/Fil0 DdR (V ) if V is
where we use the fact that (BdR /B+
dR Qp V )
1
de Rham. This tells us that (if we write h for dimQp H1 )

h1f (K, V ) = [K : Qp ] dimQp V dimK Fil0 DdR (V ) + dimQp V GK .
Likewise, we can prove that

h1e (K, V ) = h1f (K, V ) dimQp Dcris (V )=1 ,

=1
.
h1g (K, V ) = h1f (K, V ) + dimQp Dcris V (1)
For example, let V = Qp (r) and d = [K : Qp ]. By using the above formulas,
we nd the following dimensions for the various H1 (K, Qp (r)):
17. A p-adic period pairing
17
1
2
h (K, Qp (r))
d+1
d+1
=e
=f
=g
d+1
Let us make a few comments about this table.

1. For r 2, we see that every extension of Qp by Qp (r) is crystalline.
2. For r = 1, they are all semistable, and we saw in 5 that the Kummer map

H 1 (K, Q (1)) is an isomorphism. The subset H 1 (K, Q (1))
: Q p Zp K
p
p
f

then corresponds to the image of Q O , which is the characteristic zero

p
Zp
analogue of peu ramiees extensions.

3. For r = 0, the h1 counts the number of independent Zp -extensions of K, and
Hf1 (K, Qp ) corresponds to the unramied one.
4. For r 1, it is easy to show that all extensions are Cp ((t))-admissible. Yet,
since no nontrivial ones are BdR -admissible, BdR is not isomorphic to Cp ((t)).
If V is de Rham and we tensor the exact sequence
+
0 Qp B=1
max BdR /BdR 0
by V and take GK -invariants, then we nd a connecting map

DdR (V )/Fil0 DdR (V ) He1 (K, V ),
which is denoted by expV and called BlochKatos exponential. If A is an abelian
variety (or a formal group), then Vp A is de Rham, DdR (Vp A)/Fil0 DdR (Vp A) is
identied with the Lie Algebra of A, and if A denotes the Kummer map, then
the following diagram commutes, which helps explain the terminology:
Lie(A)
/ Q Z A(OK )
exp
DdR (Vp A)/Fil0 DdR (Vp A)
expV

/ H 1 (K, Vp A).
It also shows that the image of the Kummer map always lies in He1 .
17
A p-adic period pairing
Let K be a nite unramied extension of Qp and let G be a 1-dimensional formal

group of height h over OK , whose addition law is given by X Y OK [[X, Y ]].
18
We denote by [n](X) the multiplication by n power series. The Tate module of

G is
Tp G = {(u0 , u1 , . . .) where ui mCp , u0 = 0 and [p](ui+1 ) = ui }.
The space Vp G = Qp Zp Tp G is a p-adic representation of GK of dimension h,
which we know is crystalline. We will see here a more precise version of this result.
A dierential form on G is (X) = (X) dX, where (X) K[[X]], and
we denote by F (X) the unique power series such that dF (X) = (X) and
F (0) = 0. We say that
1. is invariant if F (X Y ) = F (X) + F (Y );
2. is exact if F (X) K OK OK [[X]];
3. is of the second kind if F (X Y ) F (X) F (Y ) K OK OK [[X, Y ]].
The rst de Rham cohomology group of G is then given by
1
HdR
(G/K) = {second kind}/{exact}.
This is a K-vector space of dimension h, equipped with the following ltration:

1
1
1
1
Fil0 HdR
= HdR
, Fil1 HdR
= {invariant} and Fil2 HdR
= {0}.
+ is such that
Theorem 17.1. If is of the second kind, u Tp G, and u
n A
(
un ) = un for every n 0, then

1. the sequence {pn F (
un )}n0 converges in B+
max to an element u ;
2. this element only depends on u and on the class of ;
1
3. the resulting map HdR
(G/K) Vp G B+
max is a perfect pairing, compatible
with the action of GK and the ltrations.
For example, if G = Gm , (X) = dX/(1 + X)

and u = (0, p 1, . . .), then
n
one can take u
n = [1/p ] 1 for n 0, and then u = t.
As a consequence of Theorem 17.1, we recover the fact that Vp G is crys1
(G/K) , using the p-adic period pairing. This
talline and that Dcris (Vp G) = HdR
construction can be extended to the case of abelian varieties.
Bibliography
[1] L. Berger, An introduction to the theory of p-adic representations, in: Geometric Aspects of Dwork Theory, vols. I, II, Walter de Gruyter GmbH &
Co. KG, Berlin, 2004, 255292.
[2] S. Bloch and K. Kato, L-functions and Tamagawa numbers of motives, in:
The Grothendieck Festschrift, vol. I, Progress in Math., vol. 86, Birkhauser,
Boston, MA, 1990, 333400.
[3] P. Colmez, Periodes p-adiques des varietes abeliennes, Math. Ann. 292 (1992),
no. 4, 629644.
[4] J.-M. Fontaine and B. Mazur, Geometric Galois representations, in: Elliptic
Curves, Modular Forms & Fermats Last Theorem (Hong Kong, 1993), Ser.
Number Theory, I, Int. Press, Cambridge, MA, 1995, 4178.
[5] J.-M. Fontaine, Le corps des periodes p-adiques, Asterisque (1994), no. 223,
59111, with an appendix by Pierre Colmez, Periodes p-adiques (Bures-surYvette, 1988).
[6] J.-M. Fontaine, Representations p-adiques semi-stables, Asterisque (1994),
no. 223, 113184, Periodes p-adiques (Bures-sur-Yvette, 1988).
[7] J.-M. Fontaine, Representations -adiques potentiellement semi-stables,
Asterisque (1994), no. 223, 321347, Periodes p-adiques (Bures-sur-Yvette,
1988).
[8] J.-P. Serre, Corps locaux, Hermann, Paris, 1968, 2ème edition, Publications
de lUniversite de Nancago, no. VIII.
[9] J. Tate, p-divisible groups, in: Proc. Conf. Local Fields (Driebergen, 1966),
Springer, Berlin, 1967, 158183.
[10] L. Washington, Galois cohomology, in: Modular Forms and Fermats Last
Theorem (Boston, MA, 1995), Springer, New York, 1997, 101120.
19
Deformations of Galois
Representations
Gebhard Bockle
Introduction
These lecture notes give an introduction to deformations of Galois representations
with an eye toward the application of this theory in the proof of the Serre conjecture [29, 30] by KhareWintenberger. There exist several other surveys such
as [18, 23, 37, 40]. We nevertheless hope that with the above scope in mind and
by the arrangement and detail of the material presented we can add something
useful to the existing literature. Clearly, we claim no originality in the material
presented and all errors are to be blamed on the present author.
The idea of studying deformations of Galois representations on their own
right goes back to the seminal article [39] of Mazur. Mazurs motivation was to
give a conceptual, if at the time conjectural, framework for some discoveries of
Hida [25] on ordinary families of Galois representations. It was the work of Wiles
on Fermats Last Theorem which made clear the importance of deformation theory
developed by Mazur. The theory was a key technical tool in the proof [64, 59] by
Wiles and TaylorWiles of Fermats Last Theorem.
Mazurs theory yields a universal deformation ring which can be thought of as
a parameter space for all lifts of a given residual representation (up to conjugation).
The ring depends on the residual representation and on supplementary conditions
that one imposes on the lifts. If the residual representation is modular and the
deformation conditions are such that the p-adic lifts satisfy conditions that hold for
modular Galois representations, then one expects in many cases that the natural
homomorphism R T from the universal ring R to a suitably dened Hecke
algebra T is an isomorphism. The proof of such isomorphisms, called R = T
theorems or modularity theorems, is at the heart of the proof of Fermats Last
Theorem. It expresses the fact that all p-adic Galois representations of the type
described by R are modular and, in particular, that they arise from geometry.
21
22
Many renements of Wiles methods have since been achieved and the theory has been vastly generalized to various settings of automorphic forms. R = T
theorems lie at the basis of the proof of the TaniyamaShimura conjecture by
Breuil, Conrad, Diamond and Taylor; the SatoTate conjecture by Clozel, Harris,
Shepherd-Barron and Taylor; and the already mentioned Serre conjecture. The
proof of Fermats Last Theorem was also the rst strong evidence to the conjecture of Fontaine and Mazur [22]. The conjecture asserts that if a p-adic Galois
representation satises certain local conditions that hold for Galois representations
which arise from geometry, then this representation occurs in the p-adic etale cohomology of a variety over a number eld. In fact, it is a major motivation for the
formulation of the standard conditions on deformation functors. These conditions
should (mostly) be local and reect a geometric condition on a representation. Due
to work of Emerton and independently Kisin [36], there has been much progress
on the FontaineMazur conjecture over Q.
The present notes are based on an advanced course given jointly with Laurent
Berger at the CRM Barcelona. The course provided basic material on p-adic Hodge
theory and deformation theory of Galois representations, motivated by the proof of
the Serre conjecture by Khare and Wintenberger. The lectures by Berger focused
on p-adic Hodge theory [6] and our part on deformation theory.
The contents of our lectures are as follows: Lecture 1 recalls the foundations of
Mazurs theory of deformations of Galois representations with some additional material added from the work of Kisin. Lecture 2 introduces pseudo-representations
and studies their deformations. Pseudo-representations are functions that have
the formal properties of traces of representations. They are important because
completely reducible representations can be recovered from their traces. Moreover, p-adic Galois representations are often given in terms of traces of Frobenius
automorphisms, i.e., as a pseudo-representation. The representation itself is not
directly accessible.
Lecture 3 considers universal deformations of a mod p representation of the
absolute Galois group of a nite extension of the eld Q for = p. The corresponding theory of p-adic Galois representations is well understood in terms of
WeilDeligne representations. It will turn out that also the universal deformation
can be given a natural description in terms of such parameters (or rather inertial WeilDeligne types). This leads to conditions for deformation functors of a
residual mod p representation at places not above p. WeilDeligne representations
are naturally linked to p-adic Galois representations arising from geometry: for
instance, one may consider the Galois representation on the p-adic Tate module
of an elliptic curve (or an abelian variety) over a number eld and restrict this
to a decomposition group at a prime v above . If the curve has good reduction
at v, by the criterion of NeronOggShafarevich, the representation is unramied
and vice versa; moreover the associated inertial WeilDeligne type is trivial. If it
has potentially good reduction, the representation is potentially unramied and
the inertial WeilDeligne type is non-trivial but has trivial monodromy operator.
In the remaining case the representation is potentially unipotent and the mon-
Introduction
23
odromy operator is non-trivial. At the end of Lecture 3 (for technical reasons) and
in Lecture 4, we consider the deformation theory of a mod p Galois representation of the absolute Galois group of a nite extension of Qp . More precisely, we
study some subfunctors of Mazurs functor that satisfy conditions which hold for
representations arising from geometry. This is technically the by far most subtle
part and we only work out some of the simplest cases. To formulate and study
the resulting deformation functors, p-adic Hodge theory aka Fontaine theory are
needed; see [6]. It enables one to describe local conditions for deformations of
2-dimensional representations arising from
(a) nite at group schemes, in ordinary and non-ordinary cases;
(b) crystalline Galois representations of low HodgeTate weights (0, k), where
1 k p 1;
(c) weight 2 semistable non-crystalline Galois representations.
On the geometric side, such representations arise from
(a) the p-power torsion of an elliptic curve with good ordinary or supersingular
reduction at p;
(b) p-adic Galois representations associated with a modular form of weight k,
where 2 k p;
(c) the p-adic Tate module of an elliptic curve with multiplicative reduction at p.
Lecture 5 ends the lecture series with the following result: the global universal
deformation ring R for 2-dimensional totally odd residual representations of the
absolute Galois group of a totally real eld with (suitable) geometric conditions
at all primes, xed determinant and ramication at most at a xed nite set of
places of the base eld, has Krull dimension at least 1. Together with results of
Taylor on potential modularity, covered in a lecture series by J.-P. Wintenberger
during the advanced course, the lower bound in fact suces in many cases to
show that the p-power torsion elements form a nite ideal I of R such that R/I
is nite at over Zp . This implies an important lifting result needed in the proof
by Khare and Wintenberger. The result is also in line with the expectation that
typically R should be isomorphic to a Hecke algebra on a nite-dimensional space
of p-adic modular forms, which is clearly nite at over Zp and thus of exact Krull
dimension 1.
During the lecture series we also cover a number of technically important
issues for the theory of deformations of Galois representations: framed deformations, deformation functors via groupoids on a category, pseudo-representations
and their deformations, the completion of a deformation functor at closed points
of its generic ber, and resolutions of deformation functors. Some lectures have
appendices that, for the convenience of the reader, recall technical terms needed
in the main body. To give a sample: there are appendices on Schlessingers axioms,
formal schemes, nite at group schemes, ltered -modules, etc.
24
Much of the current perspective on deformations of Galois representations is

due to work of M. Kisin, as is clear to everyone familiar with the topic. Moreover
we found his lecture notes [37] very helpful in preparing the present lecture series.
Several parts of our exposition follow closely his notes.
Acknowledgments. Let me rst thank Mark Kisin for allowing me to base parts of
the present notes on [37] and for helpful correspondence. I would also like to thank
L. Berger, B. Conrad, K. Fujiwara. G. Hein, R. Schoof and J.-P. Wintenberger
for answering some questions regarding the present material, and R. Butenuth,
K. N. Cheraku and H. Verhoek for many suggestions to improve the present notes.
I thank the CRM Barcelona for the invitation to present this lecture series during
an advanced course on modularity from June 14 to June 25, 2010 and for the
pleasant stay at CRM in the spring of 2010, during which much of these lecture
notes was written. I also thank the Postech Winter School 2011 on Serres modularity conjecture for the invitation to give a lecture series based on the present
notes. This very much helped to improve the original draft. I acknowledge nancial
support by the Deutsche Forschungsgemeinschaft through the SFB/TR 45.
Notation
The following list can be regarded as a reference page for the notation. Throughout
the notes it will be introduced step by step.
p will be a rational prime.
F will denote a nite eld of characteristic p and W (F) its ring of Witt
vectors.
O will denote the ring of integers of some p-adic eld which is nite and
totally ramied over W (F)[1/p], so that O has residue eld F.
ArO will denote the category of pairs (A, A ) where A is a nite local Artinian
O-algebra with a surjective homomorphism A : A F and maximal ideal
mA = Ker A .
O will denote the category of pairs (A, A ) where A is a complete Noethe Ar
rian local O-algebra with a surjective homomorphism A : A F and maximal ideal mA = Ker A .
G will denote a pronite group.
VF will be a (continuous) representation of G over F with d = dimF VF < .
ad = EndF (VF )
= VF F VF is the adjoint representation of VF ; it is again a
G-module.
ad0 ad is the subrepresentation on trace zero matrices.
: G O will denote a xed lift of det VF .
Introduction
25
For an arbitrary eld K, we let K denote a xed algebraic closure and

write GK = Gal(K/K) for the absolute Galois group of K. We denote the
GK -representation lim
n (K) by Zp (1).
n p
For any ring A and any free nitely generated A-module M , we denote by
M = HomA (M, A) its linear dual. If M carries an A-linear action by G then
so does M .
W (F) and a continuous representation M of GK on a free nitely
For A Ar
generated A-module, we dene its Cartier dual M as HomA (M, A(1)), where
Zp Zp (1).
A(1) = A
In Lecture 5 the following notation pertaining to number elds will be relevant:
F will be a number eld.
S will denote a nite set of places of F typically it will contain all places
above p and .
GF,S or simply GS will denote the Galois group of the maximal outside S
unramied extension of F inside F .
For a place v of F , we will denote by GFv or simply Gv the absolute Galois
group of the completion of F at v.
For each place v of F , we x a homomorphism F alg Fvalg extending F
Fv , yielding a homomorphism Gv GF GS .
26
Deformations of representations of pronite groups
Throughout this lecture series, p will be a prime and F a nite eld of characteristic p. The ring of Witt vectors of F will be denoted by W (F). By G we denote a
pronite group and by VF a nite F[G]-module on which G acts continuously. We
set d = dimF VF and x an F-basis F of VF .
In the rst two lectures, G will mostly be arbitrary but subjected to a certain
niteness condition. Later on, G will either be the absolute Galois group of a nite
extension of Qp or Q for some = p, or a quotient of the absolute Galois group
of a number eld.
In this lecture we discuss basic denitions, notions and results. The material
is fairly standard, although framed deformations are not treated in older surveys
such as [40]. We mainly follow Kisins notation, as in [37]. The lecture ends with a
discussion on groupoids over a category, which can be thought of as an alternative
means to describe deformation functors. This is taken from [34, Appendix].
1.1
Deformation functors
W (F) denote the category of complete Noetherian local W (F)-algebras with

Let Ar
residue eld F, and ArW (F) denote the full subcategory of nite local Artinian
W (F) is denoted by mA . Note that,
W (F)-algebras. The maximal ideal of A Ar
W (F) is canonically
via the W (F)-structure, the residue eld A/mA of any A Ar
identied with F.
Let A be in ArW (F) . A deformation of VF to A is a pair (VA , A ) such that
(a) VA is an A[G]-module which is nite free over A and on which G acts continuously, and
(b) A is a G-equivariant isomorphism A : VA A F VF .

A framed deformation of (VF , F ) to A is a triple (VA , A , A ), where (VA , A ) is a
deformation of VF to A and A is an A-basis of VA which reduces to F under A .
One denes functors DVF , DV2F : ArW (F) Sets by setting, for A ArW (F) ,
DVF (A) = {isomorphism classes of deformations of VF to A},
DV2F (A) = {isomorphism classes of framed deformations of (VF , F ) to A},
and with the obvious extension to morphisms.
Remarks 1.1.1. (a) The xed basis F identies the vector space underlying VF
with Fd and thus allows us to view VF as a representation : G GLd (F).
Then DV2F (A) is the set of continuous representations
: G GLd (A)
lifting . In terms of representations, DVF (A) is the set of such representations
modulo the action by conjugation of Ker(GLd (A) GLd (F)).
1. Deformations of representations of pronite groups
27
(b) It is often useful to consider deformation functors on ArO , where O is the

ring of integers of a nite totally ramied extension of W (F)[1/p], so that F
is still the residue eld of O, and where ArO is the category of local Artinian
O-algebras with residue eld F. We shall do this in later lectures without
further mentioning.
(c) In Section 1.6 we reformulate deformation functors in terms of groupoids
over a category. This gives a dierent viewpoint on the theory and will be
important for certain applications.
1.2
A niteness condition
Denition 1.2.1 (Mazur [38]). A pronite group G satises the niteness condition
p if, for all open subgroups G G, the Fp -vector space Homcont (G , Fp ) of
continuous group homomorphisms is nite-dimensional.
By the Burnside basis theorem (see Exercise 1.8.1), the group G satises
dimFp Homcont (G , Fp ) < if and only if the maximal pro-p quotient of G is
topologically nitely generated.
Examples 1.2.2. The group Homcont (G , Fp ) is isomorphic to Homcont (G , Fp ).
Thus class eld theory shows that the following groups satisfy Condition p :
ab
(a) The absolute Galois group of a nite extension of Qp .

(b) The Galois group GF,S = Gal(FS /F ), where F is a number eld, S is a
nite set of places of F , and FS F denotes the maximal extension of F
unramied outside S.
Both of these examples will be important in later lectures.
1.3
Representability
Proposition 1.3.1 (Mazur [38]). Assume that G satises Condition p . Then:

W (F) .
(a) DV2F is pro-representable by some RV2F Ar
W (F) .
(b) If EndF[G] (VF ) = F then DVF is pro-representable by some RVF Ar
One calls RV2F the universal framed deformation ring and RVF the universal
deformation ring of VF .
Remarks 1.3.2. (a) Recall that (pro-)representability (e.g., for DV2F ) means that
there exists an isomorphism
DV2F (A)
= HomW (F) (RV2F , A)
which is functorial in A ArW (F) . This universal property implies that
RV2F is unique up to unique isomorphism. Moreover the identity map in
Hom(RV2F , RV2F ) gives rise to a universal framed deformation over RV2F .
28
(b) Originally, Mazur only considered the functor DVF . It describes representations lifting VF up to isomorphism. The additional choice of basis is not a very
interesting datum. However, the functor DVF is not always representable. A
good way to remedy this problem is to rigidify the situation by adding a
choice of basis to a given representation and thus to consider the functor
DV2F instead. This is important for residual representations VF of the absolute Galois group of a number eld F , in the sense that VF may have trivial
centralizer as a representation of GF and yet the restriction of VF to a decomposition group may no longer share this property.
(c) Without Condition p , the universal ring RV2F still exists (as an inverse limit
of Artinian rings), but it may no longer be Noetherian.
(d) Due to the canonical homomorphism F EndF[G] (VF ), it is justied to write
= in EndF[G] (VF ) = F.
Proof of Proposition 1.3.1. We prove part (a). Suppose rst that G is nite, say
with a presentation g1 , . . . , gs | r1 (g1 , . . . , gs ), . . . , rt (g1 , . . . , gs ) . Dene
k
R = W (F)[Xi,j
| i, j = 1, . . . , d; k = 1, . . . , s]/I,
where I is the ideal generated by the coecients of the matrices

rl (X 1 , . . . , X s ) id,
l = 1, . . . , t,
k
with X k the matrix (Xi,j
)i,j . Let J be the kernel of the homomorphism R F
k
dened by mapping X to (gk ) for k = 1, . . . , s, with as in Remark 1.1.1(a).
Then RV2F is the J -adic completion of R and 2
VF is the unique representation
G GLd (RV2F ) mapping gk to the image of X k in GLd (RV2F ).
G/Hi
We may write any pronite group G as a ltered inverse limit G = lim
i
over some index set I of open normal subgroups Hi Ker(
). For each i the above
construction yields a universal pair (Ri2 , 2
i ). By the universality of these pairs,
one can form their inverse limit over the index set I. This yields
(RV2F , 2
(Ri2 , 2
VF ) = lim
i ),
which clearly satises the required universal property. By denition, RV2F lies in
W (F) . It remains to show that R = R2 is Noetherian. Since R is complete, it
Ar
VF
suces to show that mR /(m2R , p) is nite-dimensional as a vector space over F. It
is most natural to prove the latter using tangent spaces. We refer to the proof of
Lemma 1.4.3, where we shall see how Condition p is used.
The proof of part (b) in [39] uses Schlessingers representability criterion
(Theorem 1.7.2). Following Kisin, we shall indicate a dierent proof in Section 2.1.
d denote the completion of
The following is a preview of Kisins proof. Let PGL
d acts on the
the group PGLd over W (F) along its identity section. Then PGL
29
functor DV2F by conjugation and hence it acts on the formal scheme Spf RV2F . The
condition EndF[G] (VF ) = F ensures that this action is free, and the idea is to dene
d.

Spf RVF = Spf RV2 /PGL
F
1.4
The tangent space
Let F[] = F[X]/(X 2 ) denote the ring of dual numbers. The set DVF (F[]) is
naturally isomorphic to Ext1F[G] (VF , VF ), as an element of DVF (F[]) gives rise to
an extension1
0 VF VF[] VF 0,
where we have identied VF with VF , and, conversely, any extension of one copy
of VF by another VF can be viewed as an F[]-module, with multiplication by
identifying the two copies of VF . In particular, DVF (F[]) is naturally an F-vector
space.
Denition 1.4.1. The F-vector space DVF (F[]) is called the Zariski tangent space
of DVF . (The same terminology will be used for DV2F and other deformation functors.)
W (F) , its (mod p) Zariski tangent space
Remark 1.4.2. Recall that, for any A Ar
is the F-vector space tA = HomW (F) (A, F[]). Thus, if DVF is pro-representable,
then the tangent spaces of DVF and of the universal ring representing DVF agree.
Lemma 1.4.3. (a) Dening adVF as the G-representation EndF (VF ), there is a
canonical isomorphism
DVF (F[]) H 1 (G, adVF ).
(1.1)
(b) If G satises Condition p , then DVF (F[]) is a nite-dimensional F-vector

space.
(c) One has dimF DV2F (F[]) = dimF DVF (F[]) + d2 h0 (G, adVF ).
Remark 1.4.4. The symbol h? (. . .) always denotes dimF H ? (. . .).
Proof. Part (a) is immediate from the isomorphism Ext1F[G] (VF , VF )
= H 1 (G, adVF )
proved in Exercise 1.8.4.
We now prove part (b), thereby completing the proof of Proposition 1.3.1(a).
Let G = Ker(
), which is an open subgroup of G. The ination-restriction sequence yields the left exact sequence
0 H 1 (G/G , adVF ) H 1 (G, adVF ) (Hom(G , Fp ) Fp adVF )G/G .

The term on the left is nite because G/G and adVF are nite. The term on the
right is nite because of Condition p for G. Hence (b) is proved.
1 By
Exti we denote the continuous extension classes.
30
To prove part (c), x a deformation VF[] of VF to F[]. The set of F[] bases
of VF[] lifting a xed basis of VF is an F-vector space of dimension d2 . Let and
be two such lifted bases. Then there is an isomorphism of framed deformations
(VF[] , )
= (VF[] , )
if and only if there is an automorphism 1 + of VF[] , where adVF , which
takes to , so that adVF G . Thus the bers of
D2 (VF[] ) DVF (VF[] )
are a principal homogeneous space under adVF /(adVF )G .
Denition 1.4.5. Let : D D be a natural transformation of functors from

ArW (F) to Sets. The map will be called formally smooth if, for any surjection
A A ArW (F) , the map
D (A) D (A ) D(A ) D(A)
is surjective.
Essentially the same proof as that of Lemma 1.4.3(c) implies the following:
Corollary 1.4.6. The natural transformation DV2F DVF , (VA , A ) VA is formally smooth. Thus, if RVF is representable, then RV2F is a power series ring over
RVF of relative dimension d2 h0 (G, adVF ).
Remark 1.4.7. The above corollary says that the singularities of the two local
W (F)-algebras RVF and RV2F are in some sense equivalent, provided that DVF is
representable. Even if DVF is not representable, there is a sense in which it has
an intrinsic geometry. However, this is best formulated in terms of groupoids;
cf. Section 1.6.
1.5
Presentations of the universal ring RVF
By Remark 1.4.2 and Lemma 1.4.3 we have shown part (a) of the following result:
Proposition 1.5.1. Suppose that G satises Condition p and RVF is representable.
Then:
(a) dim tRVF = h1 (G, adVF ) =: h and so there is a surjection
: W (F)[[X1 , . . . , Xh ]] RVF .
(b) For any as in (a), the minimal number of generators of the ideal Ker is
bounded above by h2 (G, adVF ). More precisely, given , one has a canonical
monomorphism
(Ker /(p, X1 , . . . , Xh ) Ker ) H 2 (G, adVF ),
where, for a vector space V , we denote its dual by V .
31
For the proof of (b) we refer to [39] or [8, Thm 2.4]. A similar proof is given
in Lemma 5.2.2.
Corollary 1.5.2. Assume that the hypotheses of Proposition 1.5.1 hold. Then, if
h2 (G, adVF ) = 0 in this case, VF is called unobstructed the ring RVF is
smooth over W (F) of relative dimension h1 (G, adVF ).
Remarks 1.5.3. (a) If G = GF,S for a number eld F and a nite set of places S
containing all places above p and , all of the scarce evidence is in favor of the
following conjecture: if EndF[G] (VF ) = F, then RVF is a complete intersection
and at over W (F) and of relative dimension
h1 (G, adVF ) h0 (G, adVF ) h2 (G, adVF ).
For S not containing all places above p, there are counterexamples [7].

(b) Let f =
an q n be a newform of weight k 2, level N and character .
Let S be any nite set of places of Q containing the innite place and all
primes dividing N . Let K be the number eld over Q generated by all the an .
Then, by work of Eichler, Shimura, Deligne and Serre, for any prime of K
one has a semisimple 2-dimensional Galois representation
f, : GQ,S{} GL2 (F )
over the residue eld F of K at associated to f in a natural way. Let V
denote the corresponding continuous representation of G := GQ,S{} . The
representation V is known to be absolutely irreducible for almost all .
By work of Mazur for k = 2 and f associated with an elliptic curve, and
by Weston for general f (cf. [62]), the following is shown. If k 3, then V
(with respect to G ) is unobstructed for almost all primes of K. If k = 2,
then V is unobstructed outside an exceptional set of density zero.
1.6
Groupoids over categories
Universal deformation rings can be very singular at their unique closed point. The
standard way in algebraic geometry to resolve singularities are blow-ups along the
singular locus. If we apply a formal blow-up to (the formal spectrum of) a universal
ring along a subscheme containing its closed point, the resulting (formal) scheme
may have many closed points. Since we deal with universal rings representing a
functor, it is natural to look for other functors whose representing objects realize
this blow-up. This means that we can no longer consider functors on Artin rings
only. An approach, breaking with all traditions in the area, would be to reformulate the whole local theory developed so far in terms of schemes. Functors of
which one hopes that they are representable (by a formal scheme) could then be
described as stacks over the category of schemes. If we want to stay within the
32
realm of rings at least in the description of functors then one has to reformulate the theory of stacks in terms of rings. The spectra of these rings should be
thought of as giving coverings of the schemes that one should have in mind. This
has been done successfully by Kisin. Instead of studying (pre-)stacks, which are
categories (of schemes) bered in groupoids, he considers categories (of rings with
supplementary structures) cobered in groupoids. While this introduces the right
level of generality to describe resolutions of the functors one is interested in, the
theory is still close to the original theory of functors on ArW (F) .
In the present section we shall give an outline of this, hoping that it will
be useful for the interested reader who wishes to consult Kisins work, e.g. [34].
Moreover we shall make use of this in later parts of these lecture notes.
Let us rst recall the denition of a groupoid category: a groupoid category is
a category in which all morphisms are isomorphisms. However, it is not required
that between any two objects there is a morphism. There can be many isomorphism
classes these are also referred to as the connected components of the groupoid,
thinking of a category as a kind of graph. The set of endomorphisms of an object,
which is the same as the set of its automorphisms, is a group under composition.
The neutral element is given by the identity morphism of this object. One can
easily show that the automorphism groups of any two objects which are connected
are (non-canonically) isomorphic.
We shall now, following Kisin [34], reformulate the theory of deformations of
Galois representations in terms of groupoids over categories.
Fix a base category C which in many applications will be ArW (F) . We consider
a second category F and a functor : F C, and we say that
Ob(F) lies above T Ob(C) if () = T , and
( ) MorF lies above (T S) MorC if () = f .

Each object T together with the morphism idT forms a subcategory of C. By
F(T ) F we denote the subcategory over this particular subcategory of C.
Denition 1.6.1. The triple (F, C, ) is a groupoid over C (or, more ocially, a
category cobered in groupoids over C) if the following hold:
(a) For any pair of morphisms and in F over the same morphism
u
T S in C, there exists a unique morphism in F over idS such that

u=.
f
(b) For any Ob(F) and any morphism T S in C with over T there exists
a morphism in F over f .
In particular, for any T in C, the category F(T ) is a groupoid, i.e., a category
in which all morphisms are isomorphisms. It is natural to specify a groupoid by
f
specifying for any T in C the category in F over T , and for any morphism T S
in C the class of morphisms above f , and we shall often do so.
33
Remark 1.6.2. Let : F C be a functor and opp : Fo Co the induced functor

between the opposite categories. Then denes a category cobered in groupoids
over C if and only if opp denes a category bered in groupoids over C. The latter
structure is well known in the theory of stacks. This is no accident: in the theory
of stacks, the base category is typically the category of schemes. Now the opposite
category of ane schemes is the category of rings and we may look at a subclass
of schemes corresponding to a subclass of rings. Since one base category will be the
ring category ArW (F) , it is natural to work with categories cobered over it. Note
also that stacks have to satisfy some gluing conditions. The corresponding opposite
conditions are not imposed in the present (admittedly very simple) setting.
If for each T Ob(C) the isomorphism classes of F(T ) form a set, we associate
to the category F over C a functor |F| : C Sets by sending T to the set of
isomorphism classes of F(T ).
Example 1.6.3. Let C = ArW (F) . To the representation VF of G we associate a
groupoid DVF over C as follows:
(a) For A ArW (F) , the objects of DVF over A are pairs (VA , A ) in DVF (A).
(b) A morphism (VA , A ) (VA , A ) over a morphism A A in ArW (F) is an
isomorphism class
{ : VA A A VA an isomorphism | A = A }/(A ) .

In the terminology now introduced, the functor previously denoted by DVF would
be the functor |DVF |. For simplicity, we shall often omit the absolute value signs,
if no confusion is likely.
When VF has non-trivial automorphisms, then so do the objects in the categories DVF (A). In this situation, the groupoid DVF captures the geometry of the
deformation theory of VF more accurately than its functor of isomorphism classes.
Representability of a groupoid : F C
For Ob(F), dene the category (the category under ) as the category whose
objects are morphisms with source and whose morphisms from an object
to are morphisms in F such that u = . (We do not assume

that and lie over the same object of C and so u may not be an isomorphism.)
Denition 1.6.4. The groupoid F over C is called representable if there is an object
in F such that the canonical functor F is an equivalence of categories.
In the same way as , one denes the category T for any T C. One has a
natural commutative diagram of categories
34

/F

()

/ C.
(1.2)
are groupoids over C and the top horizontal and left vertical
Both and ()
homomorphisms are homomorphisms of groupoids over C. Because of the axioms
of a groupoid over a category, the left vertical homomorphism is an equivalence of
categories.
implies that , as
If F is representable, say by , the equivalence ()
well as (), are well-dened up to canonical isomorphism. One says that ()
represents F over C. Under the same hypothesis, any two objects of F(()) are
canonically isomorphic and there is an isomorphism of functors
HomC (T, ) |F|,

so that T represents |F| in the usual set theoretic sense. Conversely, if |F| is representable and for any T in C any two isomorphic objects of F(T ) are related by
a unique isomorphism, then F is representable.
Remark 1.6.5. The groupoid of Example 1.6.3 is usually not representable. To
W (F) . This
have a representability result, one needs to extend it to the category Ar
can be done canonically and is explained in [34, A.7].
The main reason why, in some circumstances, one needs to introduce the
language of groupoids, is that formation of ber products is not compatible with
the passage from a groupoid F over C to its associated functor |F|. This is a serious
technical issue, since Denition 2.4.4 of relative representability depends on the
formation of ber products. We illustrate this with a simple example taken from
[34, A.6].
Following Example 1.6.3, we dene the groupoid DV2F on C = Arw(F) as follows. An object over A ArW (F) is a triple (VA , A , A ), where (VA , A ) DVF (A)
and A is an A-basis of VA mapping under A to the basis F of VF . A morphism
=
(VA , A , A ) (VA , A , A ) over A A is an isomorphism : VA A A VA
2
taking A to A . There is an obvious morphism of groupoids DVF DVF .
Consider now the situation when the group G is trivial and x = (VA , A )
DVF (A) for some A ArW (F) . Then DVF DV2F can be identied with quadruples

2

(VA A
VA , A ), where (VA A
, : VA A A
, A ) DV (A ) and morF

phisms over idA are isomorphisms of VA reducing to the identity of VF . It follows
that this category is a principal homogeneous space for the formal group obtained
by completing PGLd /W (F) along its identity section. Hence |
DVF DV2F |(A ) is iso
morphic to the kernel Ker(PGLd (A ) PGLd (F)). On the other hand, |DV2F (A )|
is a singleton and hence the same holds for |
| |DVF | |DV2F |(A ).
1.7
35
Appendix: Schlessingers axioms
Denition 1.7.1. Let D : ArW (F) Sets be a functor such that D(F) is a point.
For any A, A , A ArW (F) with morphisms A A and A A, we have a map

D A A A D(A ) D(A) D(A ).
(1.3)
The axioms of Schlessinger in [51] are as follows:
(H1) If A A is small surjective, then (1.3) is surjective.
(H2) If A A is F[] F, then (1.3) is bijective.
(H3) dimF D(F[]) is nite.
(H4) If A A is small surjective and A = A , then (1.3) is bijective.
Note that D(F[]) carries a natural structure of F-vector space. An epimorphism A A in ArW (F) is called small surjective if its kernel is a principal ideal
which is annihilated by mA .
The following is one of the main theorems of [51]:
Theorem 1.7.2 (Schlessinger). If D satises (H1) to (H4), then D is pro-representable.
1.8
Exercises
Exercise 1.8.1. Show that for a pronite group G the following conditions are
equivalent:
(a) For all open subgroups G G the maximal pro-p quotient of G is topologically nitely generated.
(b) For all open subgroups G G the vector space dimFp Homcont (G , Fp ) is
nite.
(c) For all open subgroups G G and nite continuous F[G]-modules M one
has dimF H 1 (G , M ) < .
Exercise 1.8.2. Give a proof of Proposition 1.3.1 by verifying Schlessingers axioms
(see Denition 1.7.1).
Exercise 1.8.3. Show that the natural transformation DV2F DVF is formally
smooth.
=
Exercise 1.8.4. Show that Ext1G (VF , VF ) H 1 (G, adVF ).
Exercise 1.8.5. Show that the groupoid DVF of Example 1.6.3 is representable if
EndF[G] (VF )
= F.
Exercise 1.8.6. Describe the groupoid corresponding to the functor DV2F . What are
d
its morphism sets? For G the trivial group, show that W (F) with its standard
d
d
basis and the canonical homomorphism W (F) F represent DV2F .
36
Exercise 1.8.7. Fill in the details of the remarks following Denition 1.6.4. In
particular, show that if F is representable then Aut() = id for all Ob(F).
Exercise 1.8.8. Let : F F and : F F be morphisms of categories.
Dene F F F as the category whose objects over T are triples ( , , ), where
=
Ob(F (T )), Ob(F (T )) and is an isomorphism ( ) ( )

over idT , and whose morphisms ( , , ) ( , , ) above T S are pairs
( , ) over T S such that the following diagram in F commutes:

( )

( )
/ ( )

/ ( ).
For example, if F F is a morphism of groupoids over C and F, one
can form F := F F .
Let now S be a scheme. Then using the construction in 1.6 we may consider
over S-schemes. Suppose that X Y and
an S-scheme X as a groupoid X

X Y are morphisms of S-schemes. Show that there is an isomorphism
=

X
X
Y X .
X
Y
2 Deformations of pseudo-representations
We start this lecture by giving a proof of the representability of DVF under the
hypothesis that EndF[G] (VF ) = F, following [37, Lecture III]. Then we turn to
pseudo-representations and their deformations. Pseudo-representations as considered here were introduced in [56]. Their deformation theory was rst studied in
[44] and [49]. The treatment of the deformation theory here is taken from [37].
The material is not directly needed in relation to the proof of Serres conjecture, but it is foundational and deserves further attention. Pseudo-representations
are useful when a representation is not absolutely irreducible. They appeared rst
in work of Wiles (in somewhat dierent form; see Appendix 2.7.2). The present definition goes back to Taylor [56]. In both instances, they were used in the construction of p-adic Galois representations by a patching argument which relied on the existence of a sequence fm of mod pm modular forms such that fm fm+1 mod pm
for all m. More relevant in relation to deformation theory is their use in the
construction of p-adic families of Galois representations in the work of Bellache
Chenevier [4], Buzzard [14] or ColemanMazur [17]. Pseudo-representations also
play an important role in Kisins work [36] on the FontaineMazur conjecture.
If the dimension is larger than the characteristic, pseudo-representations do not
behave well. We shall not discuss a recent variant introduced by Chenevier [16],
which works well in all characteristics.
2. Deformations of pseudo-representations
37
In the appendix to this chapter, we provide a short introduction to formal

schemes and recall the denition of pseudo-representations in the sense of Wiles.
2.1
Quotients by group actions
Quotients by nite (formal) group actions are often representable, and indeed
there are general results which guarantee this in certain situations. In this section
we assume that G satises Condition p . Our rst aim is the proof of Proposition 1.3.1(b) from Lecture 1, whose proof had been postponed.
Theorem 2.1.1. Suppose EndF[G] (VF ) = F. Then DVF is representable.
Proof. We saw that DV2F is representable by the formal scheme2 XVF := Spf RV2F ,
W (F) . Let PGL
d denote the formal completion of the W (F)where RV2F is in Ar
group scheme PGLd along its identity section, i.e., the formal neighborhood of
d acts on the
PGLd of the closed point id PGLd (F). The formal group PGL
formal scheme XVF :
d XVF XVF , (g, x) gx.
PGL
The action can most easily be understood if the schemes involved are considered as
d (A) = Ker(PGLd (A)
functors on rings A ArW (F) : to every matrix g in PGL
PGLd (F)) and representation A : G GLd (A) (given by (VA , A , A )), one assigns gg 1 . This action can be converted into the following equivalence relation:

/
d X VF
PGL
/ XVF , g, x x, gx .
A pair (x, y) X X lies in the image of the relation if and only if x and y lie in
d -orbit.
the same PGL
d on XVF is free. This
By the hypothesis EndF[G] (VF ) = F, the action of PGL
implies that the induced map

d XVF XVF XVF , g, x (x, gx)
PGL
(2.1)
is a monomorphism as a functor of points, and thus a closed immersion of formal
schemes; see Exercise 2.8.1.
d as a formal scheme amounts to the same as conConstructing XVF /PGL
structing a formal scheme representing the above equivalence relation; indeed,
d . To see that the latter is
both schemes parameterize orbits of the action of PGL
possible we need to recall a result from [20].
W (F) is the category of complete local Noetherian W (F)-algeRecall that Ar
W (F) )o is equivalent to the category of formal
bras. Thus, the opposite category (Ar
Noetherian spectra of such W (F)-algebras with underlying space consisting of one
point and residue eld F.
2 See
Appendix 2.7.1 for some background on formal schemes.
38
Denition 2.1.2. An equivalence relation R

phisms such that
W (F) )o is a pair of mor/ X in (Ar
(a) R X X is a closed embedding, and

W (F) )o the subset R(T ) (X X)(T ) is an equivalence
(b) for all T (Ar
relation.
W (F) )o and a free action
We have seen above that, for a group object G in (Ar
G X X, the map
/
GX
/ X, (g, x) (x, gx)
denes an equivalence relation.
W (F) )o is said to be a quotient of
Denition 2.1.3. A at morphism X Y in (Ar
X by R, and one also writes Y = X/R, if the embedding R X X induces an
isomorphism R
= X Y X.
/
Theorem 2.1.4 ([20, VIIb, Thm. 1.4]). Let p0 , p1 : R
/ X be an equivalence
W (F) )o such that the rst projection p1 : R X is at. Then the
relation in (Ar
quotient of X by R exists. It represents the functor on points dened by the equivalence relation. If X = Spf B and R = Spf C, then X/R = Spf A, where
A = {b B | p0 (b) = p1 (b) in C}.
d XVF
Theorem 2.1.4 applied to the equivalence relation PGL
completes the proof of Theorem 2.1.1.
2.2
/ XVF
Pseudo-representations
Absolutely irreducible representations of nite groups are determined by their

trace functions. A result of Carayol [15] and Mazur [39] says that the analogous
result holds also for deformations:
Theorem 2.2.1 (Carayol, Mazur). Suppose that VF is absolutely irreducible. If A is
in ArW (F) and VA , VA DVF (A) are deformations such that Tr(|VA ) = Tr(|VA )
for all G, then VA and VA are isomorphic deformations.
Proof. The following proof is due to Carayol. Fix bases for VA and VA and extend
the resulting representations to A-linear maps
A , A : A[G] Md (A).
We have to show that the bases can be chosen so that A = A .
Let mA be the radical of A ArW (F) , and I = (a) A be an ideal such
that mA a = 0. By induction on the length of A, we may assume that A A
modulo I, and write A = A + , where for A[G] the matrix () Md (I)
has trace 0.
39
As A and A are multiplicative, we nd that, for 1 , 2 A[G],

(2 ).
(1 2 ) = (1 v)(2 ) + (1 v)
(2.2)
), we have that (1 2 ) = (1 )(2 ) for all 1 A[G]; therefore

If 2 Ker(
Tr(
(1 )(2 )) = 0 for all 1 A[G]. But by Burnsides theorem (F[G]) = Md (F)
as is absolutely irreducible. Hence, Tr(X(2 )) = 0 for any X Md (F), so
(2 ) = 0.
It follows that : Md (F) Md (I)
= Md (F) F I
= Md (F) is an F-linear
derivation on Md (F). Such a derivation is always inner; see e.g. [61, Lemma 9.2.1,
Thm. 9.2.11]. Hence there exists U Md (I) such that () = ()U U () and
A = (1 U )A (1 + U ).

The above result gives a clue that in many important cases also the representation theory of pronite groups is governed by traces. The idea of pseudorepresentations, introduced by Wiles [63] for odd 2-dimensional representations
and by Taylor [56] for an arbitrary group, is to try to characterize those functions
on G which are traces and to study deformation theory via deformations of the
trace functions.
Denition 2.2.2. Let R be a (topological) ring. A (continuous) R-valued pseudorepresentation of dimension d, for some d N0 , is a continuous function T : G R
with the following properties:
(a) T (id) = d where id G is the identity element and d! is a non-zero-divisor
of R.
(b) For all g1 , g2 G one has T (g1 g2 ) = T (g2 g1 ) (T is central).
(c) d 0 is minimal such that the following condition holds. Let Sd+1 denote
the symmetric group on d + 1 letters and let sign : Sd+1 {1} denote its
sign character. Then, for all g1 , . . . , gd+1 G,

sign() T (g1 , . . . , gd+1 ) = 0,
Sd+1
where T : Gd+1 R is dened as follows. Suppose that Sd+1 has cycle

decomposition

(1)
(s)
= i1 , . . . , i(1)
. . . i1 , . . . , ir(s)
= 1 . . . s .
(2.3)
r1
s

Then T (g1 , . . . , gd+1 ) = T gi(1) . . . gi(1) . . . T gi(s) . . . gi(s) .
1
r1
rs
Remarks 2.2.3. Let T be a pseudo-representation of G of dimension d.

(a) It is shown in [49, 2] that, if condition (c) holds for some d, then it holds
for all d d. It is also shown in [49, Prop. 2.4] that conditions (b) and (c)
for d minimal imply T (id) = d, provided that R is a domain.
40
(b) The condition that d! is a non-zero-divisor of R is suggested by the work of

Bellache and Chenevier [4]. In fact, they require that d! be a unit of R. Being
somewhat restrictive, this condition does avoid a number of pathologies. For
instance, we shall make use of it in Lemma 2.3.4. In [4, Footnote 13] it is also
observed that the condition d! R is needed for Lemma 2.14, Lemma 4.1
and Theorem 5.1 in the article [49].
(c) In Taylors work [56] he is primarily interested in rings R of characteristic
zero. Then d! is automatically a non-zero-divisor in R.
(d) In the recent preprint [16], Chenevier replaces the notion of pseudo-representation by that of a determinant a notion dened in [16]. Its main advantage
is that it requires no condition on d!, and hence it is a good notion over rings
of any characteristic. The preprint [16] also studies deformations of such and
applies this theory to rigid analytic Galois representation arising from p-adic
families of modular forms.
(e) It is often convenient to consider the R-linear extension T : R[G] R of a
pseudo-representation. The relations in Denition 2.2.2(c) are then satised
for all (g1 , . . . , gd+1 ) R[G]d+1 .
Theorem 2.2.4 (Taylor, Rouquier). (a) If : G GLd (R) is a representation,
then Tr is a pseudo-representation of dimension at most d.
(b) Suppose R is an algebraically closed eld 3 of characteristic Char(R) > d or
Char(R) = 0. Then for any pseudo-representation T of dimension d there
exists a unique semisimple representation : G GLd (R) with Tr = T .
(c) If G is (topologically) nitely generated, then for every integer d 1 there
is a nite subset S G, depending on d, such that a pseudo-representation
T : G R of dimension d is determined by its restriction to S. (Recall that
our hypotheses imply that d! is a non-zero-divisor in R.)
Except for the level of generality of part (b), the above theorem is due to
Taylor; cf. [56]. In op. cit., part (b) is only proved for algebraically closed elds of
characteristic zero. Taylors arguments are based on results of Procesi on invariant
theory; see [46]. Part (b) as stated is from Rouquier [49, 4], who also gives a direct
and self-contained proof of part (a) independent of the results in [46]. Below we
follow Rouquier.
Proof. We only give the arguments for part (a). We let T = Tr and dene
: Md (R) R as the map
3 By [49, Thm. 4.2] it is necessary and sucient to assume that R is a eld with trivial Brauer
group. As an example, consider D = K a division algebra over a p-adic eld K with OD a
maximal order and G the group of units of OD . Then the reduced trace is a pseudo-representation
O
2
G = OD
K K of dimension d such that d = [D : K]. However, the asserted representation
in (b) only exists over a splitting eld L K of D. For another hypothesis under which (b)
holds, see Theorem 2.4.1 due to Nyssen and Rouquier.
(g1 , . . . , gd+1 ) =
41
sign()T (g1 , . . . , gd+1 ).
Sd+1
We shall show that 0. It suces to prove this for G = GLd (R) and = id.
Writing R as a quotient of a domain R of characteristic 0, it suces to prove the
result for R ; this case is then easily reduced to that where R is an algebraically
closed eld of characteristic 0. This we assume from now on.
Let V = Rd and dene W = V V = EndR V . Note that is invariant under the action of Sd+1 (since, if one applies T to the permutation of g1 , . . . , gd+1
under some Sd+1 , one obtains T1 ). Hence, if we extend to a multilinear
map : W (d+1) R, it is determined by its values on Symd+1 W W (d+1) . As
we are in characteristic zero, a simple argument based on homogeneous polynomials of degree d + 1 in dim W variables shows that as an R-vector space Symd+1 W
is spanned by the image of the diagonal map
: W W (d+1) , w w w.
Thus it suces to show that ((w)) = 0 for all w W . As the semisimple
elements in AutR (V ) are Zariski dense in W , it is enough to show ((w)) = 0
for all semisimple w W . For small values of d, this can be veried explicitly. For
general d, one has the following argument:
Choose a basis
{e1 , . . . , ed } of V in which w is diagonal and consider the
action of := w Sd+1 sign() on W (d+1) , where Sd+1 acts by permuting
the factors and w acts as (w). We claim that Tr = ((w)). Assuming the
claim, we observe that obviously

d+1

V = 0,
sign() V (d+1)
Sd+1
and so the proposition follows.

We now prove the claim: suppose that w has diagonal entries 1 , . . . , d with
respect to e1 , . . . , ed . The trace of is then given by

w
sign() ei1 eid+1 , ei1 eid+1
i=(i1 ,...,id ){1,...,d}d+1
Sd+1
sign()
Sd+1

!
w e(i1 ) e(id+1 ) , ei1 eid+1 ,
i{1,...,d}d+1
where , is 1 if both entries are the same and zero otherwise. Thus if we write
in its cycle decomposition 1 . . . s as in (2.3) on page 39, the expression

!
e(i1 ) e(id+1 ) , ei1 eid+1
is non-zero (and thus equal to 1) exactly if the tuple i is constant on the support
of each of the cycles k . Moreover on each such support we can choose the value
42
of i freely. Moreover if (j1 , . . . , js ) denotes the tuple of values on the s supports of

these cycles (cycles may have length 1), then w applied to e(i1 ) e(id+1 )
| |
| |
results in multiplication by j1 1 . . . js s , where |k | is the length of the cycle.

Summing over all i (for xed ) yields

Tr w|1 | . . . Tr w|s | .
This expression clearly agrees with Tr (w) and so the claim is shown.
2.3
Deformations of pseudo-representations
Let F : G F be a pseudo-representation. For A in ArW (F) dene

DF (A) = pseudo-representations A : G A lifting F .
Proposition 2.3.1. Suppose G satises Condition p . Then DF is pro-representable
by a complete local Noetherian W (F)-algebra RF .
For an example of a universal pseudo-deformation, see Exercise 2.8.2. We
note that there is no simple expression for the tangent space of the functor DF
similar to that given in Lemma 1.4.3 for DVF . Interesting results on this can be
found in [3].
For the proof of Proposition 2.3.1, we need some preparation:
Denition 2.3.2. For any pseudo-representation T : G R, dene

Ker T = h G | g G : T (gh) = T (g) .
If we view T as an R-linear map T : R[G] R, then we set

Ker T = h R[G] | g R[G] : T (gh) = 0 .
Lemma 2.3.3. (a) Ker T is a closed normal subgroup of G.
(b) Ker T is an ideal of R[G].
(c) If R is nite, then Ker T is open in G.
Proof. We leave parts (a) and (b) as exercises. Let us prove (c). For each r R,
let Ur = {g G | T (g) = r}. Since T is continuous and R is nite, the Ur form a
partition of G by open subsets. Now note that the condition T (gh) = T (g) for all
g G is equivalent to UT (g) h UT (g) for all g G. Thus
Ker T =
"

h G | Ur h Ur .
rR
The latter is clearly open in G and this proves (c).
43
Inspired by [37, (2.2.3)], we show the following, where F is as in Proposition 2.3.1. For a pronite group G and m N, we denote by Gm the closed
subgroup generated by {g m | g G}. It is clearly normal in G.
Lemma 2.3.4. Set G = Ker F . Then for any A ArW (F) there exists a constant
m
mA N such that for all A DF (A) one has (G )p A Ker A . In particular, if

H G denotes the closed normal subgroup such that G /H is the maximal pro-p
quotient of G , then Ker A contains H.
Proof. For any g G , h G and i 1, one has A ((g 1)i h) mA . Taking
gj = (g 1)ij and gd+1 = (g 1)id+1 h with all ij 1 in Denition 2.2.2(c), and
using the centrality of Denition 2.2.2(b), one nds that d! A ((g 1)i h) m2A
for all g G , h G and i (d + 1). Since d! is a non-zero-divisor, we may cancel.
Proceeding by induction yields
j
A ((g 1)i h) m2A for all g G , h G and j 1 and i (d + 1)j .

m
Since A is Artinian, we can nd m N such that A ((g 1)p h) = 0 for all

g G and h G. By enlarging m if necessary we also assume that pm A = 0.
2m
In particular, it follows that p i = 0 in A for all i < pm . But then binomial
expansion yields
2m
A ((g
p2m
1)h) = A ((((g 1) + 1)
p2m
p

p2m
1)h) =
A ((g 1) h)
i
i=pm
= 0,
where in the last step we use A ((g 1)i h) = 0 for all i pm , g G and h G.
The rst part of the lemma follows with mA = 2m, since Ker A is a closed normal
m
subgroup of G. For the second part, observe that H is a subgroup of (G )p A , since
mA
G /(G )p
is a nite p-group. Hence, by the rst part, H Ker A .

Proof of Proposition 2.3.1. Suppose rst that G is nite. Let RG be the quotient
of W (F)[Xg : g G] by the ideal I generated by the relations Xe d, Xgh Xhg
for all g, h G, and the relations

sign() Xg (1) ...g (1) . . . Xg (s) ...g (s) ,
Sd+1
i1
ir
i1
i rs
where each is given in its cycle decomposition as in (2.3) on page 39. By the
denition of I, mapping XgH to F (g) yields a well-dened homomorphism RG F
W (F) . The completion R
G of RG at the kernel of this homomorphism is the
in Ar
wanted universal ring; the corresponding universal deformation of F is the induced
map
G , g gH XgH .
G : G R
G/Hi for a basis of the identity by
Let now G be arbitrary. Write G = lim
iI

open normal subgroups Hi , i I. By the universality of the (R
G/H , R/H ), they
i
44
form an inverse system and their inverse limit is the wanted universal deformation
G , G ). It remains to see that under Condition p the ring R
G is Noetherian.
(R
By the previous lemma, the elements in D (F[]) factor via RG/(G )pm for some
m
xed m N. Condition p implies that the group G/(G )p is nite, and hence
dimF D (F[]) < .

Unlike for the case of deformations of representations, nding an exact formula for dimF D (F[]) seems dicult in general. For some recent partial results,
see [3].
2.4
Deforming a representation and the

pseudo-representation Tr
For absolutely irreducible representations we have the following:

Theorem 2.4.1 (NyssenRouquier). Suppose that G satises Condition p and
that : G GL(VF ) is absolutely irreducible. Set F = Tr . Then there is an
=
isomorphism of functors DVF DF on ArW (F) .
This theorem does not require that d! be invertible in R. However, note that
is given a priori.
The theorem has the following consequences, the rst of which is due to
Carayol:
W (F) and VA be a representation over A with reduction VF
(a) Let A be in Ar
such that VF is absolutely irreducible. Let A0 A be the subring generated
by Tr(VA )(G). Then VA is dened over A0 , i.e., there exists VA0 DVF (A0 )
such that VA
= VA0 A0 A: by the above theorem, it suces to prove the
analogous statement for pseudo-representations. There it is trivial.
W (F) and : G A is a pseudo-representation such
(b) Suppose that A Ar
that mod mA arises from some as in the theorem. Then there exists a
representation : G GLd (A) whose trace is equal to . It is unique up to
isomorphism by Theorem 2.2.1.
(c) Lastly it gives another proof of the representability of the functor DVF in the
case where VF is absolutely irreducible.
The situation becomes more involved if the initial residual representation
is no longer absolutely irreducible which was one of the main reasons for
introducing pseudo-representations. Suppose therefore that : G GL(VF ) is
arbitrary and set F = Tr . One still has the canonical morphism of functors
DV2F DF , VA Tr VA .
Let us look at the following concrete example. Let 1 , 2 : G F be characters and c1 , c2 Ext1 (2 , 1 ) or rather ci Z 1 (G, 2 1
1 ). Then

1 c1 + T c2
0
2
45
is a representation G GL2 (F[T ]), i.e., a family of representations over A1F . More
naturally, one obtains a family of representations of G over P(Ext1 (2 , 1 )), the
projectivization of Ext1 (2 , 1 ), which all have the pseudo-character 1 +2 . Note
that the projectivized representation consists of a vector bundle of rank 2 over
P(Ext1 (2 , 1 )), which carries an action of a nite quotient of G such that, over
any suciently small ane Spec R P(Ext1 (2 , 1 )), the action is isomorphic to
a true representation G GL2 (R).
To fully express the relationship between the deformations of VF and those of
the pseudo-representations = Tr VF , it will be convenient to work with groupoids.
The underlying category will however not be ArW (F) : the point is that, as we have
seen above, the ber over VF of the natural transformation DV2F DF is no longer
a single point!
Following Kisin, we consider the category AugW (F) . Its objects are morphisms
A B where A is in ArW (F) and B is an A-algebra with no niteness condition
assumed. Morphisms (A B) (A B ) are pairs of homomorphisms A A
and B B which yield a commutative diagram
A
/B

A

/ B.
So, even if A = F, the second entry B can be any F-algebra, e.g. the coordinate
ring of an ane subvariety of P(Ext1 (2 , 1 )).
Remarks 2.4.2. There are several variants of the category AugW (F) .
W (F)
W (F) . Its objects are pairs (B, A) where A Ar
(a) We may also consider Aug
and B is an A-algebra which is topologically complete with respect to the
topology dened by mA B.
(b) Another natural category to consider is that of pairs (S, A) where S is an
A-scheme or even the inverse limit category of it, as described in part (a).
The point is that the pro-representing object of a groupoid bered over
AugW (F) may not be an ane scheme. In the example above it is suggested
that this scheme could be projective. Working therefore with schemes instead
of rings, the universal object would still be within the category considered.
(c) For instance, in [34] Kisin works with yet another denition of AugW (F) .
The denition here emphasizes the underlying ring A in A B. But for
other purposes phrasing the deformation problem in such a relative way is
unnecessary.
Denition 2.4.3. Fix a pseudo-representation F : G F. Dene a groupoid RepF
over AugW (F) by
46

RepF A B = VB , A | A DF (A),

VB
=.
= B d a G-representation, Tr VB = A ) /
Similarly we dene Rep2

F (A B) as the groupoid over AugW (F) with

VB , B , A | VB , A RepF A B ,
Rep2
F A B =

B a B-basis of VB /
=.
Finally we extend DF to a groupoid on AugW (F) by setting

DF (A B) = DF (A).
We shall indicate in Remark 2.6.3 why it is desirable and useful to study the
functor Rep2
F .
We introduce the following notion:
Denition 2.4.4. A morphism : F F of groupoids over C is called relatively
representable if for each Ob(F ) the 2-ber product
F = F F
is representable.
Note that if F is representable and : F F is relatively representable,
then F is representable.
Proposition 2.4.5. If G satises Condition p , then Rep2
F is representable by a
morphism of ane formal schemes Spf R2F Spf RF which is formally of nite
type.
By Theorem 2.2.4(b) we know that Rep2
F = only if p > d.
Proof. This is Exercise 2.8.6.
Suppose that F is the trace of a semisimple representation , and so that

2
Rep2
F is non-empty. We give an explicit description of RepF above the point
x = F DF (F).
Proposition 2.4.6. The functor x
DF Rep2
F is representable by the same F-algebra
as the functor on F-algebras which to any such algebra S assigns the set
{ : G GLd (S) | Tr = F }.
Any such representation : G GLd (S) factors via G/(G )p , where G is the
kernel of . Furthermore the semisimplication of is isomorphic to .
If G is nite, then the ring RF2 := R2F RF F representing x
DF Rep2
F is
g
given as follows. Let R = W (F)[Xi,j | i, j = 1, . . . , d; g G]. Then R0 = R/I
d
g
F (g), g G, together
for I R the ideal generated by the elements i=1 Xi,i
g1
gs
with the components of the dd-matrices r(X , . . . , X )id, where r(g1 , . . . , gs )
g
ranges over all relations among the elements of G and X g = (Xi,j
)i,j=1,...,d .
d
2.5
47
Representable subgroupoids of RepF
The groupoid RepF will not be representable in general. This is for instance the
case in the situation of page 44. In this section we shall investigate a resolution of
a particular subgroupoid of RepF which will turn out to be representable. The
material follows [37, 3.2]. Further details will appear in a planned future version
of [37]. In this section, we require for all A B in AugW (F) that the ring B be of
nite type over A.
We consider the following situation: suppose that for i = 1, . . . , s we are given
pairwise
representations i : G GLdi (F). We set
sdistinct absolutely irreducible
2
2
F =
Tr
.
Let
Rep
Rep
i
F
F be the full subgroupoid over AugW (F)
i=1
on objects (VB , B , A ) Rep2
(A
B) such that the following holds: there

F
exists an ane cover of Spec B/mA B by Spec Bi such that
VB B Bi
c1
2
...
..
.
..
.
..
.
cs1
s
(2.4)
i+1 , i ) for all i = 1, . . . , s1.

with nowhere vanishing extension classes ci Ext1 (
Remark 2.5.1. The condition on the ci has the following consequence: since the i
are absolutely irreducible and pairwise non-isomorphic, the centralizer of the matrix on the right of (2.4) is contained in the set of diagonal matrices which are
scalar along the blocks i . The non-triviality of ci implies that the scalar along
i is the same as the scalar along i+1 . Therefore the centralizer of the representation on the left is precisely the set of scalar matrices. One deduces that the
isomorphism in (2.4) is unique up to a scalar.
Remark 2.5.2. In (2.4) the diagonal blocks of the matrix on the right will always
occur in the order indicated. This will be important in the sequel; see for instance
Corollary 2.6.2.
Theorem 2.5.3. There exists a locally closed formal subscheme, formally of nite
W (F) .
type, X2F Spf R which represents RepF2 on Aug
Idea of proof: The special ber of Rep2
F is a scheme of nite type over F. Any
specialization R2F k for k a nite eld containing F admits an isomorphism,
as in (2.4), for at least one permutation of the i and without the non-vanishing
requirement for the ci . The condition that the chosen order occurs along the diagonal denes a closed subscheme of the special ber Spec RF R F. Similarly,
one argues that the additional conditions 0 = [ci ] Ext1 (
i+1 , i ) under any such
specialization dene an open condition. Details will appear in the nal version
of [37, 3.2].

In Remark 2.5.1 we observed that the action of PGLd on the special ber
d on X2
of RepF2 is free. It easily follows that the conjugation action of PGL
F
48
is free. Even though X2F is typically not a formal local scheme, the quotient
d still exists. The proof is formally similar to that of TheoXF := X2F /PGL
rem 2.1.1. However, here a theorem on representability of free group actions on
formal schemes over Artin rings is needed. Such a result ts well the framework
of Mumfords book on geometric invariant theory, but over Artin rings is not to

be found there. An application of Schlessingers criterion is not possible, as X2
F is
typically not local. Nevertheless, the result holds. One proof is due to B. Conrad;
details will appear in the nal version of [37] by Kisin.
As was explained to us by Kisin, it is not so straightforward to dene the
functor which is represented by XF . Over X2F Spf RF there is a universal
object represented by this arrow: on X2F we have a trivial vector bundle with a
basis and a representation

VX2 , X2 , X2 ;
F
on Spf RF we have the universal pseudo-representation u , and the morphism

X2F Spf RF is induced from the pseudo-representation Tr 2F on X2F . While
d has a well-dened and free action on X2 , it does not act on the universal
PGL
F
d acts on this triple. Since its center acts trivially on the
triple. Only the group GL
d is a projective bundle over X (if X is local, it
base X2F , the quotient VX2 /GL
F
F
F
carries up to isomorphism a unique vector bundle of rank d and one can take it as
the quotient). In the global situation the Picard group of the special ber of XF
need not be trivial. Then it is not clear whether a vector bundle quotient should
exist. Therefore one cannot expect that the universal object on XF is given by a
vector bundle with a G-action up to isomorphism.
One natural way to bypass the above problem the nal version of [37]
might follow a dierent approach is to consider projective bundles equipped
with a G-action instead of vector bundles with a G-action, and to ensure that the
projective action does lift locally to a linear one. This allows one to give a natural
interpretation of XF as a formal moduli space as follows.

Denition 2.5.4. For p > d =
di , denote by RepF the groupoid on AugW (F)
which for any (A B) AugW (F) is the category RepF (A B) whose objects

are tuples PB , PB,s1 , . . . , PB,1 , A satisfying the following conditions:
(a) PB is a projective bundle on Spec B with a continuous action G AutB (PB ).
(b) Setting PB,s = PB B B/mA B, the PB,i , i = 1, . . . , s, are projective G-bundles on Spec B/mA B of rank d1 + + di 1 and they form a ag of G-stable
projective subbundles PB,1 PB,2 . . . PB,s of PB,s .
(c) A : G A is a pseudo-representation lifting Tr .
(d) For any ane open Spec C Spec B/mA over which, disregarding the G-action, PB,1 PB,2 . . . PB,s is isomorphic to Proj of C d1 C d1 +d2
. . . C d , the induced action of G on
#
49
C d1 ++di /C d1 ++di1 modulo scalars
i=1,...,s

is conjugate under GLd1 (C) GLds (C) to i i F C modulo scalars.
(e) In the notation of (d), for any i = 1, . . . , s1 there is a well-dened extension
class in
Ext1C[G] (
i+1 F C, i F C)
and we assume that its specialization to any closed point Spec k Spec C
is non-trivial.
(f) Let the notation be as in (d) and let Spec BC Spec B denote the pullback
of Spec C Spec B/mA B it is ane because mA B is nilpotent. Then
under the above hypotheses one can show that
(i) there exists a unique linear representation C : G GLd (C) with
det C = det and attached projective representation equal to G
AutC (PB B C), and
(ii) there exists a unique linear representation C : G GLd (BC ) with C
(mod mA BC ) = C , det C = det A and attached projective representation equal to G AutBC (PB B BC ).
In addition to (a)(e), we also require that Tr C = A under A BC .
The denition of morphisms on the so-dened objects is left to the reader.
Remarks 2.5.5. (i) By a projective bundle we mean Proj of a vector bundle. By
a ag of projective bundles we mean that, Zariski locally on the base, there
exists a ag of vector bundles (with all factors being again vector bundles)
to which Proj associates a ag isomorphic to the given ag of projective
bundles.

(ii) As we assume p > d =
i di , the concept of pseudo-deformation is well
behaved. By our hypotheses, is multiplicity free in the terminology of [4].
Thus, by [16, Remark 1.28] any pseudo-representation A : G A for A
W (F) and with A A F = Tr has a well-dened determinant det A : G
Ar
A .
(iii) It might be necessary to strengthen the condition EndF[G] (
i ) = F, which
follows from the absolute irreducibility of i , to the condition that the centralizer of the projective representation attached to i is F GLdi (F). This
notion of projective absolute irreducibility is in general stronger than absolute
irreducibility; cf. Exercise 2.8.8. If there exists a subgroup Gi of G such that
i |Gi is absolutely irreducible and such that the di -torsion of Hom(Gi , F ) is
trivial, then i is projectively absolutely irreducible.
(iv) We leave the assertions in (f) as an exercise (perhaps a non-trivial one) to
the reader. Observe however that from the existence and uniqueness of the
50
local linear representations C one cannot deduce the existence of a global

linear representation on some vector bundle: the uniqueness of C implies
that the transition maps on the level of vector bundles are unique only up to
units in BC
. Thus, one can only glue the local patches if the Picard group of
the special ber of XF is trivial.
The following result will, in some form, be fully explained in the nal version
of [37].
Theorem 2.5.6. The groupoid RepF DF is representable by the proper formal
scheme XF over Spf RF .

Sketch of proof: Consider the universal object VX2 , X2 , X2 on X2F together
F
F
F
with the universal pseudo-representation u on Spf RF and the universal ltration
by sub-vector-bundles
VX2 RF F = Vs Vs1 . . . V1
F
m GL
d.
on X2F RF F given by (2.4). These objects carry natural actions by G
m takes us from vector bundles to projective bundles and
The central action of G
m acts trivially on X 2 .
a ag of such on the special ber X2F RF F; the center G
F
2

The free action of PGLd on XF yields the object

d
P VX = VX 2 , Vs1 , . . . , V1 , u /GL
F
in RepF (XF Spf RF ). Assuming assertion (ii) in Denition 2.5.4(f), it is not

hard to see that P VX is the universal object for RepF over XF Spf RF , as
F
follows. Let P VB be in RepF (A B). Choose an ane cover Spec Bi of Spec B
such that on each Spec Bi we have a representation Ci as in Denition 2.5.4(f)
for P VB B Bi . The Ci yield a unique morphism Spec Bi X2F . The induced
morphisms Spec Bi XF agree on overlaps since by construction they are unique.
This shows the universality of (P VX , XF ).

F
2.6
Completions of RepF2
Proposition 2.6.1. Let x be in Rep2

F (F F) and let VF be the corresponding
representation of G.
2

(a) The canonical surjection R
RV2F is an isomorphism.
F x
(b) For x RepF2 (F F) the surjection from (a) induces an isomorphism
O
X , x RVF .
F
Proof. For part (a) observe that the completion of Rep2

F at x can be described
as follows: it is given by the functor on ArW (F) which maps any A ArW (F) to
2
{(A , A , A ) Rep2
F (A A) | A DVF (A)}.
51
The datum A is clearly superuous and hence this functor is isomorphic to DV2F .
This proves (a). Part (b) follows from the construction of the rings as quotients
under the same group action.

Corollary 2.6.2. Let E/W (F)[1/p] be a nite extension and x : RepF E a
point 4 such that the corresponding E-valued pseudo-representation x is absolutely
irreducible. Then the map
RepF Spf RF
is an isomorphism over a formal neighborhood of x .
Proof. We rst observe that x is the only point of RepF lying over x . To see
this, suppose that x is another such point. Denote by Vx and Vx the corresponding G-representations. By the properness of RepF over DF , the points x and
x arise from OE -valued points, which in turn correspond to G-stable lattices
Lx Vx and Lx Vx . Since Vx and Vx are absolutely irreducible with the
same trace, they are isomorphic. We choose an isomorphism so that it induces a
G-homomorphism Lx Lx whose reduction modulo mOE is non-zero. The representations Lx /mOE Lx and Lx /mOE Lx are both of the type described in (2.4).
Since the order of the i is xed and the extension of I+1 by i is never trivial
(see Remark 2.5.2), the semisimplication of the image of Lx /mOE Lx is of the
form i s while the semisimplication of the image as a submodule of
Lx /mOE Lx is of the form 1 j . Since the i are pairwise non-isomorphic,
and the image is non-trivial, we deduce that the morphism of the reductions is an
isomorphism. It follows that Lx
= Lx .

Next, let RF x be the completion of RF at x and O
X , x be that of XF at x.
F
Denote by VF the mod mOE reduction of Lx . We need to show that the canonical
homomorphism

R
F x OX , x = OX ,VF
F
F
2.6.1(b)
x

R
VF x
E is an isomorphism. This follows from the theorem of Carayol and Mazur

in Ar
(Theorem 2.2.1), which can easily be adapted to the case of residue eld E instead
of F.

Remark 2.6.3. Suppose is reducible and is the reduction of a p-adic representation attached to a modular form f . If one wants to study mod p congruences
of f , then one is interested in all modular forms f whose attached p-adic Galois representation has a mod p reduction whose semisimplication is equal to
the semisimplication of . From this perspective it is natural to consider the
deformation space of the pseudo-representation F = Tr .
Next, suppose that the p-adic representation V attached to f is absolutely
irreducible; for cusp forms this is a natural hypothesis. Suppose further that V is
4 Talking about an E-valued point of a formal scheme over W (F) is a slight abuse of notation.
What is meant is the point on the generic ber attached to the formal scheme; see Appendix 2.7.1.
On this analytic space one considers the completion of the stalk at E.
52
residually multiplicity free. Then we can always nd a lattice L V whose mod p

reduction VF satises condition (2.4) for a suitable ordering of the irreducible
constituents of ss . The results in this section show that innitesimally near V
the universal pseudo-representation space, the universal deformation of VF and
the completion of RepF at V agree. Moreover, the completion of RepF at VF is
isomorphic to RVF .
Suppose now that XF has large dimension (as a formal scheme). Then we
expect RF to be highly singular at its closed point: each closed point VF of RepF
has a dierent universal ring RVF ; in RF all these rings are glued together at the
special ber, while irreducible representations p-adic deformations lie in only one
of the spaces Spec RVF . Thus, RepF appears to be a partial desingularization of a
part of RF : partial because the rings RVF which occur for VF in the special ber
of RepF by completion of RepF at VF may themselves still be singular. But by
gluing them one expects to create a more dicult singularity. Note that in the particular case in which the semisimplication of consists of two (non-isomorphic)
summands only, one can in fact regard RepF as a partial desingularization of all
of RF . In this case, any p-adic representation V contains a lattice whose reduction
orders the two irreducible constituents in any given order.
To recapitulate the above, in order to study all mod p congruences of a
residually reducible representation, one needs to consider RF . However, it is to be
expected that its geometry is highly singular at the closed point. A formal scheme
with a less singular special ber that sees many congruences is XF . However, in
general it will not contain all representations with congruent mod p Frobenius
eigenvalue systems. This can only be guaranteed if the ss is a direct sum of
exactly two absolutely irreducible non-isomorphic representations. An alternative,
but also highly singular candidate that would in its generic ber see all p-adic
representations which have congruent mod p Frobenius eigenvalue systems is the
versal hull that arises from the deformation functor of ss . Here the reason is
that, at least after enlarging the coecient eld, any p-adic representation V with
residually reducible mod p reduction contains a lattice whose mod p reduction is
semisimple.
A second reason for choosing RepF instead of DF is that the former groupoid
allows it to impose local conditions quite easily in the case where G is the absolute
Galois group of a global eld. This can be done as in the case of DVF , which we
shall discuss in the following three sections. For DF it is perhaps slightly more
dicult to impose and study local conditions.
2.7
2.7.1
Appendix
Formal schemes
In this appendix we recall the denition of a Noetherian formal scheme. In the

following, we x a Noetherian ring R and an ideal a of R. We assume that R is
53
a-adically complete, i.e., that the canonical homomorphism

a = lim R/an
R R
to the a-adic completion of R is an isomorphism.

We dene a topological space Spf R (with respect to a), which will soon also
be given a structure sheaf:
A prime ideal p of R is called open (with respect to a) if p contains a.
The underlying set of the topological space Spf R consists of the open prime
ideals of R, so that it is in bijection with Spec R/a.
The topology on Spf R is the topology induced from the bijection between
Spf R and Spec R/a.
For instance, if R ArW (F) and a = mR is the maximal ideal of R, then Spf R
consists of a single point.
To dene a structure sheaf on Spf R, let us recall the Zariski topology on
:= R/a. Dene D(f) as the set of
Spec R/a: for f R denote by f its image in R
prime ideals p of R such that f is non-zero at p. This set is in bijection with the
set of open prime ideals p of R at which f is non-zero (under reduction). The sets
D(f) dene a basis for the topology on Spf.
Now, for f R dene
Rf 1 = lim
R[f 1 ]/an .
It is not dicult to verify that the assignment D(f) Rf 1 denes a sheaf
on Spf R. (The main task is to verify the sheaf property by coverings of an open
subset D(f) by sets D(fi ).)
Let us see that the sheaf dened above is locally ringed, i.e., that its stalks
are local rings. Suppose for this that x Spf R corresponds to the open ideal
p = px of R. Then the stalk at x is
Ox = lim
Rf 1 .
xD(f)
It is a good exercise to show that Ox is a local ring with maximal ideal pOx . (To
see the latter one needs that p be nitely generated, which is true in our case since
R is Noetherian.)
Denition 2.7.1. The formal scheme Spf R of R (with respect to a) is the locally
ringed space (X, OX ) where X = Spec R/a as a topological space and the structure
sheaf OX is dened by OX (D(f)) = Rf 1 , for all f R.
We ignore all subtleties necessary for the denition of non-Noetherian formal
schemes.
54
Example 2.7.2. Let X = PGLd /W (F). Its ane coordinate ring consists of the set
of homogeneous rational functions of degree zero in the ring

R = W (F) Xi,j , det((Xi,j )1 ) | i, j = 1, . . . , d .
Consider the morphism id : R F, Xi,j i,j corresponding to the identity element of PGLd (F). The completion of PGLd along the kernel of id is a Noetherian
d in the proof of Theorem 2.1.1.
ane formal scheme, denoted by PGL
Denition 2.7.3. A Noetherian formal scheme is a locally topologically ringed space
(X, OX ) such that each point x X admits an open neighborhood U such that
(U, OX |U ) is isomorphic to an ane formal scheme Spf R, as dened above.
Morphisms of formal schemes are morphisms of topologically ringed spaces.
So a morphism from (X, OX ) to (Y, OY ) is a pair (f, f # ) where f : X Y is a
morphism of topological spaces and f # : OY f OX is a continuous homomorphism of sheaves, i.e., locally such a homomorphism is given as a ring homomorphism
A B mapping a power of the dening ideal a of A into the dening ideal b B.
A particular construction of a formal scheme is the following. Let X be a
scheme and I OX be an ideal sheaf. Then the completion of X along I is a
formal scheme whose underlying topological space is the closed subscheme Z of
X dened by I. On an ane cover of Z one applies the construction indicated
in page 53, and then one glues the so obtained formal ane schemes. We write
for it, or X
I if the need arises to indicate the ideal sheaf. A formal scheme
X
constructed in this way is called algebraizable.
W (F) . Consider the closed immersion A1 A1
Example 2.7.4. Let R be in Ar
R
F
R be the formal scheme
obtained by pulling back A1F along Spec F Spec R. Let A
obtained as completion along this closed immersion. Its underlying topological
space is AF . However, its structure sheaf can be quite enormous. For instance, its
ring of global sections is Rx = lim
R[x]/mnR [x]. To describe the latter ring,
dene
i
ht(r) = max{i N | r
/ mR }. Then Rx is the subring of R[[x]] of series
ri x i
such that ht(ri ) for i .
W (F) and consider the projective scheme P1 =

Example 2.7.5. Let R be in Ar
R
1
PZ Z Spec R over Spec R. The maximal ideal of R denes an ideal sheaf on Spec R
and via pullback an ideal sheaf on P1R . The completion of P1R along this pullback is
a formal scheme with a natural homomorphism to Spf R. It is the formal projective
R from the previous
line over Spf R. It can be obtained by gluing two copies of A
m,R . Carry out the construction in detail to make sure that you
example along G
fully understand the corresponding formal scheme and the morphism of formal
schemes.
The generic ber of a formal scheme over W (F)
Given a formal scheme X with a morphism to Spf W (F), one can, following Berthelot, associate a rigid space over W (F)[1/p] to it. The detailed construction can
55
W (F) . Suppose rst

be found in [19, 7]. Let us give the idea for Spf R with R Ar
that R = W (F)[[X1 , . . . , Xk ]]. Then
W (F)[1/p]{X1 , . . . , Xk } W (F)[[X1 , . . . , Xk ]] W (F) W (F)[1/p]
(W (F)[1/p])[[X1 , . . . , Xk ]],
where on the left we have the Tate algebra over W (F)[1/p], i.e., the ring of those
power series which converge on the closed disc OCk p of radius 1. The ring in the
middle consists of power series whose coecients have uniformly bounded norm.
These converge on the open unit disc of dimension k of radius 1, i.e., on
k
k
OCp = {(x1 , . . . , xk ) OCp | |xi | < 1 for i = 1, . . . , k}.
It is a rigid analytic space. The anoid discs of radius 1

admissible cover.
W (F) can be written as
A general R Ar
1
n
around 0 form an
R = W (F)[[X1 , . . . , Xk ]]/(f1 , . . . , fm ).
Then the zero locus of the functions fj , j = 1, . . . , m, denes a rigid analytic
subspace of the open unit disc. This will be the rigid analytic space (Spf R)rig
1 R and
associated to Spf R. As a next example, one could work out the case of P
show that the associated rigid analytic space is the projective line over OkCp .
Functors on formal schemes
Representability of functors is also an important question for formal schemes.
Schlessingers representability criterion (Theorem 1.7.2) or the theorem of Grothendieck behind it can be regarded as a theorem on the representability of
formal schemes: Schlessingers criterion studies the pro-representability of a co W (F) .
variant functor ArW (F) Sets by an object in Ar
Formally passing to opposite categories, we obtain a functor from (ArW (F) )o
to Setso . The category (ArW (F) )o is the category of formal Artin schemes on one
point over Spf W (F) with residue eld Spf F. In this sense, Schlessingers criterion
provides necessary and sucient conditions for a functor on such formal Artin
W (F) )o .
schemes (to Sets) to be representable by a Noetherian formal scheme in (Ar
2.7.2
Pseudo-representations according to Wiles
The rst occurrence of pseudo-representations in the theory of Galois representations was in the work of Wiles [63] for 2-dimensional odd Galois representations of
the absolute Galois group of a number eld. His denition appears to be dierent
from
2.2.2. In the presence of a complex conjugation whose image is
1 0 Denition

,
Wiles
denition
can easily seen to be equivalent to Denition 2.2.2 for
0 1
d = 2.
56
Denition 2.7.6. A pseudo-representation in the sense of Wiles consists of continuous functions a, d : G R and x : G G R such that for all g, g , h, h G
one has
(a) a(gh) = a(g) + a(h) + x(g, h),
d(gh) = d(g) + d(h) + x(h, g),
x(gh, g h ) = a(g)a(h )x(h, g ) + a(h )d(h)x(g, g ) + a(g)d(g )x(h, h )

+ d(h)d(g )x(g, h ).
(b) a(1) = d(1) = 1, x(1, h) = x(g, 1) = 1 for all g, h G.
(c) x(g, h)x(g , h ) = x(g, h )x(g , h).
Proposition 2.7.7. Suppose that : G GL2 (R), g
representation. Then
a(g) b(g)
c(g) d(g)

is a continuous
(a) (a, d, x(g, h) = b(g)c(h)) forms a pseudo-representation

in the sense of Wiles.
1 0
(b) If there exists c G such that (c) = 0 1 , then the pseudo-representation
from (a) depends only on the trace of , because

1
Tr (g) Tr (gc) ,
2

1
d(g) =
Tr (g) + Tr (gc) ,
2
x(h, g) = a(gh) a(g)a(h).
a(g) =
Note that part (a) is in fact the motivation for the axioms of a pseudorepresentation in the sense of Wiles. They simply arise from the rules for multiplying two matrices.
2.8
Exercises
d (A) = Ker(PGLd (A) PGLd (F)) for all

Exercise 2.8.1. (a) Show that PGL
A ArW (F) cf. Example 2.7.2.
(b) Verify that the morphism (2.1) on page 37 is a closed immersion by using
the following criterion (which is actually not hard to prove): a morphism of
ane formal schemes Spf A = X Y = Spf B is a closed immersion if and
only if it is a monomorphism (of functors); cf. [20, VIIA.1.3].
(c) Verify that the quotient constructed in the proof of Theorem 2.1.1 does indeed
represent RVF .
Exercise 2.8.2. Let p > 2, let G = Zp considered as an additive pronite group
and let F : G F, g 2 be the trivial 2-dimensional pseudo-representation.
(a) Show that the tangent space of DF is at most 2-dimensional. (Hint: Deduce
from condition (b) of Denition 2.2.2 for d = 2 that any 2 DF (F[]) is
determined by its values on 1 and 2; use the continuity of 2 .)
57

1+X z
. By studying Tr(),
(b) Consider : Zp GL2 (W (F)[[X, Y ]]), z 1+X
Y 1+Y
show that W (F)[[X, Y ]] is a quotient of the universal pseudo-representation
ring for F .
(c) Prove that Tr() is the universal pseudo-deformation of F .
(d) Prove that is the universal deformation of := mod (X, Y ) despite
the fact that the representability criterion of Proposition 1.3.1 fails
Exercise 2.8.3 ([36, 1.4]). (a) Give an example where VF is not absolutely irreducible and there exist non-isomorphic deformations VA , VA DVF (A)
with the same traces. (Hint: Consider two characters 1 , 2 : G F with
dimF Ext1 (2 , 1 ) > 1.)
(b) Show that if 1 , 2 : G F are distinct characters such that Ext1 (2 , 1 )
is 1-dimensional and VF is a non-trivial extension of 1 by 2 , then the
analogue of Carayols theorem holds for VF : two deformations in DVF (A) are
non-isomorphic if and only if their traces are dierent.
Exercise 2.8.4. Show that if : F F is a relatively representable morphism of
groupoids over C, and F is representable, then so is F.
Exercise 2.8.5. A morphism of groupoids : F F over C is formally smooth
if the induced morphism of functors || : |F| |F | is formally smooth, i.e., if for
any surjective morphism T S in C, the map
$ $
$ $
$ $
$F$(T $F$ S |F |(S) $F $ T
is surjective. Show that : F F is formally smooth if and only if for every
F the morphism |F | || is formally smooth.
Exercise 2.8.6. Prove Proposition 2.4.5. (Strategy: To establish relative representability, x A0 ArW (F) , B0 some A-algebra and 0 : G A0 a pseudo-representation lifting F . This denes an element in DF (A0 B0 ). Describe the groupoid
Rep2
F , . Show that all representations described by it are representations of a xed
nite quotient G of the originally given group G the quotient depends on . It
suces to consider the case B0 = A0 . Then write down the universal object for
G in a way similar to the proof of Proposition 2.3.1 or Proposition 1.3.1(a). The
wanted formal scheme is obtained by an inverse limit of such situations. To see that
the morphism is of nite type, it suces to consider the case A0 = B0 = R/m2R .)
1 and P1

1
Exercise 2.8.7. (a) Show that for A ArW (F) one has A1A
=A
A = PA
A
as locally ringed topological spaces.
(b) Work out all details in Examples 2.7.4 and 2.7.5.
Exercise 2.8.8. Let p be the characteristic of an algebraically closed eld k. Fix
n > 4 prime to p and denote by D2n the dihedral group of order 2n. Consider the
representation
: D2n GL2 (k) sending a generator of the rotations in D2n to

0
for k a primitive n-th root of unity and a reection to 10 10 . Show
0 1
that is absolutely irreducible but not projectively absolutely irreducible.
58
Deformations at places not above p and ordinary

deformations
The material on generic bers is from [34, 2.3]. The results on deformations at
= p can either be found in the lecture notes [45], in [30, 3.3] or in [34, 2.6]. For
the last section on ordinary deformations we refer to [35, 2.4] or [30, 3.2].
In the appendix to this chapter we recall various topics used in the main
body. We recall the canonical subgroups of the absolute Galois group of a local
eld, we present basic results on Galois cohomology, we give a short introduction
to WeilDeligne representations and we provide some basic results on nite at
group schemes.
3.1
The generic ber of a deformation functor
A deformation functor D may have dicult singularities at its closed point. If

the functor is representable, this means that the corresponding universal ring R
is highly singular at its maximal ideal. However, in many concrete and important situations it turns out that the generic ber of the formal scheme Spf Rrig
associated with R (see Appendix 2.7.1) has no singularity or at most very mild
singularities. Moreover, in the cases we have in mind, closed points on the generic
ber are of utmost interest: a closed point of Spf Rrig is a W (F)-algebra homomorphism R E for some nite extension E of W (F)[1/p]. Thus if R = RV2F ,
as in Lecture 1, such points are precisely the p-adic representations G GLd (E)
which possess a conjugate G GLd (OE ) whose mod mOE -reduction is . A similar interpretation holds for the closed points of (Spf RVF )rig . Using functors on
Artin rings (over nite extensions of W[1/p]), one cannot recover Spf Rrig . However, given any closed point on Spf Rrig , starting from D one can construct such
a functor D which describes the innitesimal neighborhood of on the generic
ber. Moreover, in many concrete examples, this functor can be easily written
down explicitly; cf. Theorem 3.1.2. In particular, this often gives a simple way to
compute the tangent space at such a point, e.g. Remark 3.1.3, and to check for
smoothness. Let me nish this introduction by giving one example why it should
be simpler to work on the generic ber. Suppose G is a nite group. Then the
representation theory of G over a eld of characteristic zero, as is any nite extension of Qp , is completely dominated by Wedderburns theorem. It says that the
abelian category of nite-dimensional representations is semisimple. On the other
hand, if p divides the order of G, then the category of nite A[G]-modules for any
nite Zp -algebra is a rather complicated object.
The above observations regarding the generic ber have been exploited crucially by Kisin in several instances, e.g. [34]. In this section, we briey recall Kisins
constructions and some basic results. We shall consistently work with groupoids.
3. Deformations at places not above p
59
Groupoids for closed points on the generic ber

Let E be a nite extension of W (F)[1/p] with ring of integers OE . Dene ArE as
the category of nite, local W (F)[1/p]-algebras B with residue eld E. Since B is
a nite W (F)[1/p]-module, the residue homomorphism : B E is canonically
split5 and thus B is an E-algebra in a canonical way.
For B ArE denote by IntB the category of nite OE -subalgebras A B
such that A[1/p] = B. The morphisms in IntB are the natural inclusions. The category IntB is ordered by inclusion and ltering, i.e., any two objects are contained
in a third one. For the (ltered) direct limit of the A IntB one obtains
lim A = 1 (OE ).
AIntB
The limit is taken in the category of rings.

W (F),O as the category of W (F)-algebra homomorphisms A OE ,
Dene Ar
E
W (F) , as A[1/p] need not be Artinian. In particular, IntB may
where A lies in Ar
W (F),O for all B ArE . If E is totally ramied
be regarded as a subcategory of Ar
E
over W (F)[1/p], so that OE has residue eld F, then the last assertion is true in the
obvious sense. Otherwise one proceeds as follows. Denote by OE : OE OE /mOE
the homomorphism to the residue eld, which is a nite extension of F and dene
1
O OE as the inverse image O
(F). Then, given (A B) IntB , the pair
E
1

(A (O ) B) lies in ArW (F),O .
E
W (F) (so that

Let F be a groupoid over ArW (F) . Extend it canonically to Ar
F has some continuity property with regard to inverse limits). Fix F(OE ) and
W (F),O by
dene a groupoid on Ar
E
F() (A OE ) = { F(A) | lies over A OE }
W (F),O , i.e., we consider deformations of to objects A OE

for (A OE ) Ar
E
W (F) . This groupoid gives rise to a groupoid over ArE , again
with A still in Ar
denoted F() , by setting
F() (B) =
lim F() (A)
AIntB
for B ArE .
Lemma 3.1.1. Suppose F is pro-represented by a complete local W (F)-algebra R
and is given by : R OE . Then the groupoid F() on ArE is pro-represented by
5 Lacking a reference, here is an indication of proof. The key is that E is algebraic over
W (F)[1/p]; so, for x E, consider its minimal polynomial f over the other eld. Pick an arbitrary
lift to B. Use the Newton method to nd the unique lift which is a root of f . This denes a
canonical lift.
60
obtained by completing R W (F) E along

the complete local W (F)[1/p]-algebra R
I = Ker(R W (F) E
W (F) E
/ E) .
Proof. When discussing representability of groupoids, we observed that a groupoid

is representable if and only if isomorphic objects over the identity are isomorphic
via a unique isomorphism and the functor |F| is representable (see the comments
below Denition 1.6.4). Using this fact, the lemma is a simple exercise left to the
reader.
Application to the generic ber of DVF and DV2F

To see a rst example for the usefulness of F() , we consider the case F = DV2F (or
F = DVF ). We dene two groupoids related to = (VOE , OE ) DV2F (OE ). Set
V = VOE OE E, which is a continuous representation of G, and the basis OE
canonically extends to a basis of V . Dene DV and DV2 as groupoids on ArE
as follows. For any B ArE , let DV (B) denote the set of deformations of V to a
free B-module VB with a continuous GK -action. Similarly, let DV2 (B) denote the
set of deformations of (V , ) to a free B-module VB with a continuous GK -action
together with a choice of basis B lifting .
Theorem 3.1.2. There are natural isomorphisms of groupoids over ArE
DVF ,() DV and DV2F ,() DV2 .
Proof. We sketch the proof in the rst case (in fact, we shall only sketch the proof
for the functors associated to the groupoids). We begin by dening the natural
transformation. Let B be in ArE . Then an element in DVF ,() (B) is an element in
lim D() (A),
AIntB
where in turn an element of D() (A) is a continuous GK -representation on a free

A-module VA together with an isomorphism VA A OE
= VOE under the homomorphism A OE for A IntB . So an element of DVF ,() (B) is a direct system
of (VA )AIntB of such. We have observed earlier that lim
A = 1 (OE ) for
AInt
B

V 1 (OE ) B denes an
: B E the structure map of B. Hence lim
AIntB A
element in DV (B).
To prove surjectivity, suppose VB DV (B). Since V arises from VO via
OE E, the representation VB contains a natural subrepresentation V1 (OE ) on
a free nitely generated 1 (OE )-module. Since the A IntB exhaust 1 (OE )
and since the action of G is continuous, and B is Artinian, we can nd A
IntB and a subrepresentation VA V1 (OE ) which is free as an A-module and
with a canonical homomorphism onto VOE . This completes the proof of essential
surjectivity. The proof of injectivity is left as an exercise; see [34, (2.3.5)].
61
Remark 3.1.3. One has the isomorphism DV (E[])

= H 1 (G, adV ) for the tangent
space of DVF , , i.e., for that of Spec RVF [1/p] at .
3.2
WeilDeligne representations
Let F be a nite extension of Q with uniformizer and residue eld k. Set

q = #k. Consider the diagram
1
/ IF
/ GF
O
/ IF
?
/ WF
/Z
k/k)
= Gal(
O
/1
?
/Z
= Frobk
/1
dening the Weil group WF , where the bottom row is the pullback of the top row
along the inclusion on the right. We regard n as a homomorphism WF Z and
x an inverse image WF of 1 Z (where 1 is identied with Frobk ).
n
roots of
. This %
induces an
Consider a compatible system p of p-power
%
n
isomorphism from Zp to the Galois group of n F (pn , p ) over n F (pn ).
The action of GF on this copy of Zp is via the cyclotomic character and whence
we write Zp (1). The action of IF on this compatible system denes a surjective
GF -equivariant homomorphism
tp : IF Zp (1).
From the above and standard results about tame ramication of local elds one
deduces the isomorphism
p

(G
F ) Zp (1) Zp
for the pro-p completion of GF with the Zp on the right being the pro-p completion
p
= Gal(k/k). More explicitly, one can nd t Zp (1) and s (G

of Z
F ) mapping
p

to Frobenius in Zp , such that (GF ) is the pro-p-completion of
s, t | sts1 = tq .
(3.1)
Let E/Qp be a nite extension and a continuous representation of GF on

a nite-dimensional E-vector space V .
Theorem 3.2.1 (Grothendieck). There exists a unique nilpotent N End(V ), the
logarithm of monodromy, and a nite index subgroup I1 of IF such that, for all
g I1 ,
(g) = exp(tp (g)N ).
One can verify that for g IF and n Z one has ( n g)N = q n N ( n g).
62
Proof. We indicate the construction of N . By continuity of , one can nd a free

OE -submodule V with V = [1/p] which is preserved under GF . Let F F
be the xed eld of the kernel of the representation GF GLd (/2p) induced
from . The kernel of GLd () GLd (/2p) is a pro-p group. Thus the action
of GF on V is via its pro-p completion. Denote by As and At the matrices of the
images of s, t from (3.1) for the eld F . The relation sts1 = tq implies that At
and (At )q have the same eigenvalues. Thus, the nite set of eigenvalues is invariant
under q , and so its elements must be roots of unity. Since Zp Zp /2pZp
is injective on roots of unity, all eigenvalues of At must be 1. Dene N as the
logarithm of the nilpotent endomorphism At idV . Then the assertion of the
theorem holds for I1 = IF .

The above result yields immediately the following bijection.
Corollary 3.2.2. There is a bijection between isomorphism classes of representations : GF GLd (Qp ) and isomorphism classes of pairs ( , N ) (WeilDeligne
representations; cf. Appendix 3.9.3) such that
(a) : WF GLd (Qp ) is a continuous representation with the discrete topology
on Qp ;
(b) N Md (Qp ) is nilpotent and satises ( m g)N = q m N ( m g) for all
g IF and m Z;
(c) () is bounded (see Appendix 3.9.3).
The bijection sets the pair ( , N ) in correspondence with if and only if for all
m Z and g IF one has

m g = m g exp tp (g)N .
Remark 3.2.3. (a) By the continuity of , it is clear that (IF ) is nite.
(b) The representation depends on the choice of . Its restriction to IF does
(obviously) not.
Denition 3.2.4. Given , we call the pair (|IF , N ) its inertial WD-type.
Corollary 3.2.5. If d = 2, then either
(a) N = 0 (this happens if and only if (IF ) is nite; note that in this case is
semisimple), or
(b) N has rank 1. This happens if and only if (IF ) is innite. Then is a nonsplit extension of some character : GF E by (1), the twist of by the
cyclotomic character; cf. Corollary 3.9.6.
Denition 3.2.6. If V has dimension 2 and N = 0, we call V a representation of
(twisted) Steinberg type.
63
O for O the ring of integers of a nite ramied extension

Let now R be in Ar
of W (F)[1/p]. Let VR be a free nitely generated R-module carrying a continuous
R-linear action by GF . Let GF be as above. Denote by VR[1/p] the representation VR R R[1/p].
Theorem 3.2.7. There exists a unique nilpotent N End(VR[1/p] ) and a nite
index subgroup I1 of IF such that, for all g I1 ,

(g) = exp tp (g)N .
The assignment
: WF GLd (VR[1/p] ), n g ( n g) exp(tp (g)N )
for n Z and g IF denes a continuous representation of WF where we regard
VR[1/p] as a topological vector space with the discrete topology; in particular, (IF )
is nite.
Proof. The argument is basically the same as that for the proof of Theorem 3.2.1.
Dene F as the xed eld of the kernel of GF GLd (R[1/p]/2pR[1/p]). One
veries that (t)id modulo the nilradical of R[1/p] is nilpotent for t any generator
of IF . But then (t) id itself is nilpotent. Dene, as before, N = log((t)) and
I 1 = IF .

Corollary 3.2.8. Suppose Spec R[1/p] is geometrically irreducible over W (F)[1/p].
Let x, y X = Spec R[1/p] be closed geometric points. Let x and y denote the
representations on Vx and Vy obtained from VR by base change. Then x and y
are isomorphic as representations of IF .
Another way to paraphrase the corollary is to say that on geometrically
irreducible components of (Spf R)rig the representation |IF is constant. This result
will be applied in Theorem 3.3.1 to a universal deformation ring.
Proof. By the construction of ( , N ) in the previous theorem, the representation
(x ) of the specialization at x (or y) is the specialization of the representation
for R[1/p]. (It may however happen that N for is non-zero while Nx for x
it is zero.) Hence it suces to consider the representation . As the image of
IF under is nite, we may regard as a representation of the nite group
G = IF / Ker( |IF ).
Since the nilradical of R[1/p] is contained in the kernels of the specializations
at x and y, we may assume that R[1/p] is reduced and hence a domain. Let m
be the exponent of G. Let E be the extension of W [1/p] obtained by adjoining
all m-th roots of unity, and let R = R[1/p] W[1/p] E . Then R is an E -algebra
and by geometric irreducibility it is still an integral domain. We need to show
that, for any two homomorphisms x, y : R Qp , the specializations x and y
of : G GLd (R ) are isomorphic. By the choice of E and ordinary character
64
theory for representations of nite groups, it suces to show that x and y have
the same traces.
Now the E -algebra structure of R is inherited by all specializations. But
then it is obvious that under specialization the traces of x and y will be the
same.

Example 3.2.9. The following example shows that geometric irreducibility is necessary in the above corollary. Suppose that 1 (mod p) and let F be Q . By
local class eld theory, the abelianization Gab
F has a tamely ramied quotient isomorphic to F . Since 1 (mod p), it has a quotient of order p. Hence there is
a surjective homomorphism : GF Z/(p), g (g) such that the xed eld of
its kernel is totally and tamely ramied. Let p (X) = (X p 1)/(X 1). Consider
the representation

(g)
GF GL1 W (F)[X]/ p 1 + X , g 1 + X
.
The ring R = W (F)[X]/(p (1+X))[1/p]
= Qp (p ) is not geometrically irreducible
over Qp . In fact, one has p 1 dierent embeddings Q(p ) Qp over Qp . Clearly
each embedding gives rise to a dierent representation of IF on GL1 (Qp ).
3.3
Deformation rings for 2-dimensional residual

representations of GF and their generic ber
We continue to denote by F a nite extension of Q for some = p. Let VF be

a F[GF ]-module on which GF acts continuously. let O be the ring of integers
of a nite extension of W (F)[1/p], and let : GF O be a character whose
reduction modulo mO agrees with det VF .
We dene subgroupoids DVF DVF and DV,2
DV2F over ArW (F) by
F

DVF (A) = {(VA , A ) DF (A) | det VA ) = },
DV,2
(A) = {(VA , A , A ) DF2 (A) | det(VA ) = },
F
for A ArW (F) . By showing that these functors are relatively representable
is pro-representable (by RV,2
) and, if
as subfunctors, one deduces that DV,2
F
F
EndF[G] (VF ) = F, then also DVF is pro-representable (by RV,2

).
F
Let ( , N ) denote the WeilDeligne representation attached to ,2
VF by Theorem 3.2.7 and assume that O contains all pm -th roots of unity if pm is the maximal
p-power divisor of the exponent of the nite group (IF ) (so that Corollary 3.2.8
.
is applicable). The following is the main theorem concerning DV,2
F
Theorem 3.3.1. The following hold (where component always means of the generic
ber):
(a) The generic ber (Spf RV,2
)rig is the union of nitely many smooth compoF
nents of dimension 3.
65
(b) The restriction of IF to any component is constant (in the sense of Corollary 3.2.8).
(c) The components are in bijection with the inertial WD-types which arise from
p-adic representations of GF that possess a conjugate reducing to .
(d) There is at most one component, which we call CN , whose inertial WD-type
has non-trivial monodromy (at some point). This component occurs if and
by (1).
only if VF is an extension of a mod p character
whose inertial WD-type is of the form

(e) There is at most one component Cnr
( id, 0) for some character . This component occurs if and only if a twist
of VF is unramied.
.
(f) The only generic components which can possibly intersect are CN and Cnr
(g) Spf RV,2

is covered by Spf Ri for domains Ri which are in bijection to its
F
generic components.
(h) dimKrull RV,2
= 4.
F
Remarks 3.3.2. (a) If EndFF (VF ) = F, then, using that the tangent space dimension drops by three if we pass from framed deformations to deformations, it
follows that dimKrull RVF [1/p] = 0. Hence, by generic smoothness, the ring
RVF [1/p] is a product of elds.
(b) The theorem tells us that the natural and only possible subfunctors of DV,2
F
are those given by selecting a nite number of components of the generic
ber, i.e., a nite number of inertial WD-types.
(c) The theorem makes no distinction between p = 2 and p > 2. The case p > 2
is somewhat simpler, in the sense that, independently of the global choice of
determinant, one can read o from the residual representation whether the
components CN or Cnr
appear. For p = 2, the added complication is that
not every character GF O with trivial reduction possesses a square root.
We shall in the following three sections indicate parts of the proof of the
above theorem. For convenience we assume p > 2. For a complete proof, see
[45]. We begin in Section 3.4 with a brief discussion of the very simple case of
unramied representations (up to twist). This corresponds to (f) in the theorem
above. In the subsequent section we treat, rather completely, the case of (twisted)
Steinberg type lifts. This concerns part (e). Due to formal similarities, the case
of Steinberg type deformations is also helpful for the investigation of ordinary
deformations above p later in Section 3.7. Section 3.6 indicates many of the steps
toward the proof of Theorem 3.3.1.
3.4
Unramied deformations for = p
Throughout Sections 3.4 to 3.6 we shall keep the hypotheses of the previous section.
66
the subgroupoid of
Proposition 3.4.1. Suppose VF is unramied. Denote by DV,nr,2
F ,O
,2
DV,2
over
Ar
of
unramied
framed
deformations,
i.e.,
of
(V
,
O
A A A ) DVF (A)
F
such that IF acts trivially on VA . Then this subgroupoid is representable by a ring
O which is smooth over O of relative dimension 3.
Ar
RV,nr,2
F ,O
The case where VF is a twist of the trivial representation can easily be reduced
to Proposition 3.4.1. We denote by (RV,nr,,2
, V,nr,,2
) the pair
F
F
, V,nr,,2
)
(RV,nr,2
F
F
for some character : GF O .
is a free group topologically generated by
Proof. Because F := Gal(F nr /F )
=Z
the Frobenius automorphism , the functor DV2F is smooth: representations of F
are determined by the image of and the only requirement for the image is that
it lies in a compact subgroup, which is vacuous for images in GLd (A), A ArO .
Now for any surjection A A and representation to A , one can lift the image of
to A and with determinant equal to (). Alternatively, one can simply appeal
to the fact that H 2 (F , adVF ) = 0.
can be computed from that
By smoothness, the relative dimension of RV,nr,2
F ,O
of the tangent space:
dimF DV,nr,2
(F[]) = dimF DV,nr
(F[]) + (d2 1) h0 (F , adVF )
F ,O
F ,O
= 3 + h1 (F , adVF ) h0 (F , adVF ) = 3.
is free so that h1 (. . .) = h0 (. . .). One can also give a
The last equality uses that Z
(F[]) is a principal homogenous space
short direct argument proving that DV,nr,2
F ,O
3

under Ker(PGL2 (F[]) PGL2 (F)) = F .
3.5
Deformations of Steinberg type for = p
In this section, we analyze following [34, 2.6] the deformation functor for deformations where the monodromy N is typically non-zero. Thus VF has a basis F
such that
(1)(g)
: GF GL2 (F), g
.
(3.2)
0
(g)
is twisted by the mod p cycloRemarks 3.5.1. (a) Writing (1) indicates that
tomic character.
(b) Since = p, the mod p cyclotomic character is unramied and depending on
it may be trivial.
can be ramied,
(c) If the lift is of Steinberg type, then the character lifting
even if is not. This is possible precisely if the ramication subgroup of Gab

F
contains non-trivial p-torsion, i.e., if q 1 (mod p).
67
we shall for the remainder of this section

After twisting by the inverse of ,
assume the following:
(a) dim VF = 2.
(b) det VF is equal to the mod p reduction of the cyclotomic character
: GF Zp .
(c) VF (1)GF = 0, i.e., there is a subrepresentation of dimension at least 1 (and
exactly 1 unless mod p is trivial) on which GF acts via modulo p.
We now dene the groupoid L,2

VF (resp. LVF ) over Aug see page 45 for the
,2
denition of the base category. The groupoid L,2
VF maps naturally to DVF over
Aug and serves, as we shall see, as a smooth resolution of the subgroupoid of DV,2
F
of deformations of Steinberg type. It can only be understood over AugW (F) and
not over ArW (F) ; cf. Exercise 3.10.3. For (A B) AugW (F) , so that A ArW (F)
and B is an A-algebra, the set of objects in L,2
VF over (A B) is the set of tuples
(VA , A , A , LB ), where
(A),
(VA , A , A ) DV,2
F
LB VB := VA A B is a B-line, i.e., a submodule such that VB /LB is a
projective B-module of rank 1,
LB VB is a subrepresentation on which GF acts via .
Lemma 3.5.2. The functor L,2
VF is represented by a projective algebraizable morphism, which we denote by
,2
VF : L,2
VF Spf RVF .
. Consider the projective space P1R . It classies

Proof. We abbreviate R = RV,2
F
2
R-lines L inside VR = (R) . (Since R is local, these lines, as well as the quotients
1 R the completion
by them, are free of rank 1 and not just projective.) Denote by P
of the above space along its specialization under R F. This is a formal scheme.
It classies R-lines of VR over an F-line of VF .
1
Let L,2
VF be the closed subscheme of P R dened by the equations gv (g)v
1 R and all g GF . By formal GAGA it is
for all v in the universal line over P
a projective scheme over the formal scheme Spf R. Algebraizability is automatic

from Grothendiecks existence theorem in formal geometry; cf. [26]. But it can also
be shown directly, as in [30, Proof of Proposition 3.6]. From the construction of

L,2
VF , one deduces its universal property.
,2 rig
Lemma 3.5.3. L,2
is
VF is formally smooth over W (F). Its generic ber (LVF )
connected.
68
Proof. To prove smoothness, consider a surjective homomorphism B B with

nilpotent kernel of algebras over A A in ArW (F) and let (VA , A , LB ) be an
,2

object of |L,2
VF (A B )|. Note that we want to show that LVF is smooth over
in ArW (F) with A
B
W (F) and not over DV,2
. Hence it suces to nd A A
F

and A mapping to A and (VA, A, LB ) lifting the above triple.
Set VB = VA A B . Since P1 is smooth (over any base), the pair (VB , LB )
lifts to the pair (VB , LB ) consisting of a free rank 2 module over B together with
a B-line. The next step is to show that the extension
0 LB VB VB /LB 0
lifts. This amounts to proving that the natural homomorphism
Ext1B[GF ] (VB /LB , LB (1)) Ext1B [GF ] (VB /LB , LB (1))
(3.3)
is surjective. Clearly the map P := LB (VB /LB ) P := LB (VB /LB ) is

a surjective homomorphism of B-modules. Identifying the Ext1 with an H 1 , the
surjectivity of (3.3) follows from the next lemma.
Lemma 3.5.4. For any A ArW (F) and any A-module M (which is not necessarily
nitely generated), the natural homomorphism
H 1 (GF , Zp (1)) Zp M H 1 (GF , M (1))
is an isomorphism.
Proof. Using the commutativity of H i with direct limits, it suces to prove this
for A = M = Z/(pn ). Then the latter homomorphism arises from the long exact
cohomology sequence for 0 Zp Zp Z/(pn ) 0, which yields that the map
is injective with cokernel H 2 (GF , Zp (1))[pn ]. By local Tate duality, the H 2 is the
Pontryagin dual of Qp /Zp , i.e., Zp , which has no p-power torsion.

To complete the proof of smoothness, we apply the following lemma:
Lemma 3.5.5. Suppose that (A B) (A B ) is a homomorphism in AugW (F)

such that B B is surjective with nilpotent kernel. Suppose also that VB is a
continuous representation of G on a free B-module and VA DVF (A ) such that
V A A B
= VB B B . Then there exists a factorization of in AugW (F) ,
(A B) (A0 B) (A B ),
and a VA0 in DVF (A0 ) such that VA0 A0 B
= VB and VA0 A0 A
= V A .
Proof. We rst observe that
= A B B Ker(B B ),
A
lim

A
69
Ar of A B B which contain the

where the limit ranges over all subrings A
image of A under the homomorphism A A B B deduced from the universal
property of A B B. The continuous representation VB restricts to a continuous
representation VAB B and, by linearity, to a continuous representation VA for
We need to nd A
such that V B = VB . Since VA A B = VB ,
all A.
A
A
0 such that
we have VAB B AB B B
= VB , and hence it suces to nd A

m

VA0 A0 (A B B) = VAB B = lim
. Choose m such that mA = 0 and m
VA
A

such that Ker(B B ) has m -th power equal to zero. Then for m = m +m one

has mm
= 0 for all A. By the universal property of R (we equip VA with a basis
A
which induces one on all VA), there exist (unique) compatible homomorphisms

R/mm
, where the A range over the above direct
R A such that VR induce VA
system. Now take A0 as the subring of A B B generated by the images of A
A B B and R/mm

R A B B.
over W (F), the
It remains to prove connectivity. By smoothness of L,2
VF
,2
,2
idempotent sections of L,2
[1/p],
of
L
and
of
L
F
are
in bijection.
W (F)
VF
VF
VF
Next, as a scheme over the local scheme Spf R, the idempotents of L,2
VF are in
bijection with those of Z := L,2
F;
see
[24,
III,
4.1.5].
The
scheme
Z (which
R
VF
is not formal) is isomorphic to a closed subscheme of P1 . Depending on whether
the image of is scalar or not, this subscheme is all of P1 or a single point;
cf. Exercise 3.10.3(a), and thus in either case it has no non-trivial idempotent
sections. Finally one observes (see [34, 2.4.1] based on [19, 7.4.1]) that one has a
,2 rig
bijection between the idempotents of L,2

VF [1/p] and those of (LVF ) .
Lemma 3.5.6. Let E/Qp be a nite extension, let be in L,2
VF (OE ), and denote
,2
by also its image in DVF (OE ).
,2
The morphism of groupoids L,2
() D() on ArE is fully faithful. If the
E-representation V of GF corresponding to is indecomposable, then this morph
ism is an equivalence, D()
is representable and its tangent space is 0-dimensional.
Note that, on the level of functors, full faithfulness means that |L,2
() | is a
,2
|.
subfunctor of |D()
Proof. We rst prove full faithfulness. Since both functors are representable, the
homomorphism sets of objects are singletons or empty. Suppose we have two objects of L,2
() . It is clear that if there is a morphism between them, then there
will be one between the images. What we need to rule out is the possibility that
there is a homomorphism between the images but no homomorphism between the
objects.
,2
To see this, let B be in ArE and (VB , B ) in D()
(B). We have to show that
if VB admits a B-line LB such that LB (1) is GF -invariant, then LB is the unique
such line. Since detB VB = , we have HomB[GF ] (B(1), VB /LB ) = 0 we are in
characteristic zero , so that HomB[GF ] (B(1), VB ) = HomB[GF ] (B(1), LB ), and
70
the uniqueness of LB follows. (The point is simply that while the mod p reduction
of can be zero, itself is never trivial.)
Suppose further that V is indecomposable. Then V satises EndE (V )
= E,
so that D()
is representable. We have to show that any VB contains at least
one B-line LB VB on which GF acts via . For this, it suces to show that
the tangent space of D()

is trivial, since then VB
= V E B, for E B the
canonical splitting, is the trivial deformation which inherits the required B-line
from VE . However, the dimension of this tangent space is computed by
Exer.

dimE H 1 GF , ad0 V = dimE H 0 GF , ad0 V + dimE H 0 GF , ad0 V (1) = 0,
where the zero at the end follows from the indecomposability of .
)
For as above, dene a groupoid over ArE by dening DV (B) (resp. DV,2
(B)
for B ArE as the category whose objects are deformations of V to B with
determinant (and in addition a basis lifting the given one.)
Proposition 3.5.7. Let Spf RV,1,2
denote the scheme theoretic image of the morphF
ism VF of Lemma 3.5.2. Then
(a) RV,1,2
is a domain of dimension 4 and RV,1,2
[1/p] is formally smooth over
F
F
W (F)[1/p].
(b) Let E/Qp be a nite extension. Then a morphism : RV,2
E factors
F
if
and
only
if
the
corresponding
2-dimensional
E-representhrough RV,1,2
F
tation V is an extension of E by E(1).
proceeds
One way to avoid formal schemes in the denition of Spf RV,1,2
F
,2
as follows. Since L,2
Spf
R
is
relatively
representable,
for
any
Artinian
VF
VF
,2
quotient A of RV,2
,
the
pullback
of
L
to
A
is
a
honest
scheme,
say
L,2
VF
VF /A .
F
The morphism VF of Lemma 3.5.2 induces a projective morphism of schemes
L,2
VF /A Spec A. Its scheme theoretic image thus denes a closed subscheme of
Spec A, say Spf RV,1,2

. These are schemes in ArW (F) which form an inverse limit
F /A
system and their inverse limit is precisely Spf RV,1,2
.
F
,1,2
Proof. The scheme L,2
is
VF is smooth over W (F) and connected. The ring RVF
,2
the ring of global sections of this scheme over Spf RVF and hence it must be a
domain.
,2
,1,2
Since L,2
. If we invert
VF is projective over Spf RVF , it surjects onto Spf RVF
,2
p, then by the previous lemma LVF [1/p] is a closed subscheme of Spec RV,2
[1/p]
F
,1,2
and hence it is isomorphic to Spec RVF [1/p], which shows that the latter is
formally smooth over W (F)[1/p].
By construction, RV,1,2
is p-torsion free. To compute its dimension it thereF
fore suces to compute the dimension of its generic ber (and add 1). This we may
71
is representable
do at an indecomposable V . By Lemma 3.5.6, the functor D()
,2
is formally smooth over Spf E of dimension 3.
by Spf E. Hence D()
The last assertion follows essentially from Lemma 3.5.6, as factors through
if and only if it lifts to a (necessarily unique) point of L,2
RV,1,2
VF , i.e., if and only
F
if V admits an E-line LE such that LE (1) is GF -invariant.
Let O be the ring of integers of a nite extension E of W (F)[1/p]. By twisting

the rings in the previous proposition with any global characters : GF O , one
obtains:
Corollary 3.5.8. Dene RV2F ,O = RV2F W (F) O. Then there exists a unique quotient
of RV2F ,O with the following properties:
RV,,2
F ,O
is a domain of dimension 4 and RV,,2
[1/p] is formally smooth over
(a) RV,,2
F ,O
F ,O
O.
(b) Let E/Qp be a nite extension. Then a morphism : RV,2
E factors
F ,O
,,2
through RVF ,O if and only if the corresponding 2-dimensional E-representation V is an extension of by (1).
3.6
On the proof of Theorem 3.3.1
Proof of Theorem 3.3.1. Parts (b), (d) and (e) follow directly from Corollary 3.2.8
and our results in Sections 3.4 and 3.5, respectively. We now prove (f). Let E
be a nite extension eld of O[1/p] and : RV,2
[1/p] E be an O-algebra
F ,O
homomorphism. The smoothness at in the generic ber, i.e., the smoothness of
(RV,2
) , certainly holds if H 2 (GF , ad0 V ) = 0 and in this case the tangent space
F ,O,
dimension is

dimE H 1 GF , ad0 V + dimE ad0 V dimE H 0 GF , ad0 V = 3.
Suppose now that is such that H 2 (GF , ad0 V ) = 0 and recall that, by local
duality,

dimE H 2 GF , ad0 V = dimE H 0 GF , ad0 (1) .
This last group is non-trivial if and only if V is isomorphic to a sum of characters
(1). Moreover, in this case dimE H 2 (GF , ad0 V ) = 1 and the tangent space
has dimension 4.
Our particular point lies on the components
[1/p] and Spec RV,nr,,2
[1/p].
Spec RV,,2
F ,O
F
The containment in the second is obvious. For the containment in the rst, observe that has deformations with N non-trivial: take any non-zero class of
H 1 (GF , E(1)) which is thus ramied and consider the corresponding deformation
to E[].
72
In fact, the two components identied are the only ones through . By twisting the entire situation by 1 we may assume that is an extension of E by
E(1). Let C be an irreducible component through it. If C contains a point where
N is non-zero, then C = CN , else N = 0 on all of C. Now the triviality of |IF at
implies that |IF is zero on all of C and hence is unramied on C. Thus C is
the other component we have already identied.
For the proof of (c) and (g) one denes deformation functors for all inertial
WD-types that can occur. (Their classication is not dicult.) In several cases
one needs to consider a resolution by adding the datum of an additional line such
as in the analysis we gave in the (twisted) Steinberg case. In each case one directly
shows that the functor is representable by a domain which proves the bijection
in (c) between inertial WD-types and components. To see (g) one shows that any
deformation is described by one of the functors so-dened. See [45] for details.
We now prove (a). The niteness of the number of components is clear, since
RV,2
is Noetherian. By the proof of (f), all (closed) points on the generic ber
F
except for those in the intersection of CN and Cnr

are smooth and have tangent
dimension 3. But we also know that CN and Cnr are smooth of dimension 3. Hence
(a) is proved.
The proof of (h) is now simple. By (g) the rings Ri are p-torsion free and
thus their dimension is one more than the dimension of their generic ber. Hence
(h) follows from (a).
3.7
Ordinary deformation at p
In this section, we let K be a nite extension of Qp . We shall investigate the ordinary at deformation ring of a 2-dimensional representation of GK following [35].
Let : GF Zp denote the cyclotomic character. The deformation functor and
its resolution are very similar to those used in the Steinberg case in Section 3.5.
One key dierence occurs in the proof of the smoothness of the resolution. In the
Steinberg case this relied on the surjectivity of the homomorphism (3.3). Here
we need a dierent argument. The rst results of this section recall the necessary
background for this. Then we closely follow the discussion in the Steinberg case.
IK
Set K nr = K
and K = GK /IK = Gal(K nr /K). Let M be a discrete
possibly innite K -module over Zp on which p is nilpotent.
For any nite subrepresentation M M consider the twist
M (1) = M Zp Zp (1).
Since Zp (1) arises from a p-divisible group over OK and since M is unramied, the
representation M (1) arises from a nite at group scheme over OK (we shall give
some background on this in Appendix 3.9.4). A proof is given in Corollary 3.9.14.
We now introduce the group Hf1 (GK , M (1)). It classies, for any n N such
that M is annihilated by pn , extensions of Z/pn by M (1) which arise from the
generic ber of a nite at group scheme over OK . The group Hf1 (GK , M (1)) is
73
Hf1 (GK , M (1)), where M ranges over

then dened as the direct limit lim
M M
the nite K -submodules of M .
To dene Hf1 (GK , M (1)), we rst consider H 1 (GK , M (1)). Restriction in
cohomology yields a homomorphism
Res : H 1 (GK , M (1)) H 1 (IK , M (1))K .
If M (1)IK = 0, i.e., if (mod p) is non-trivial on IK , then the ination restriction
sequence shows that the above map is an isomorphism it is also an isomorphism
if M is a free Zp -module. To obtain an alternative expression for H 1 (IK , M (1))
suppose that pn annihilates M . Then
H 1 (IK , M (1))
= H 1 (IK , pn ) Z M
= lim H 1 (GL , pn ) Z M
L/K nr
Kummer
lim L /L p Z M
= lim L Z M
= (K nr ) Z M .
n
L/K nr
L/K nr
Denote this isomorphism by and dene Hf1 (GK , M (1)) as the kernel of
H 1 (GK , M (1))
Res
/ H 1 (IK , M (1))/ 1 (O nr Z M ).
K
(3.4)
/OK
.
Observe that by Kummer theory one has Hf1 (GK , Z/pZ(1))
= OK
K
If is non-trivial on IK , then Hf1 (GK , M (1))
(O
M
)
.
In
general,
= K nr Z
1

by [35, Lem. 2.4.2] the group Hf (GK , M (1)) is isomorphic to

K
K
K
(OK
/image (OK
(OK
,
(3.5)
nr Zp M0 )
nr Zp M1 )
nr Zp M0 )
where 0 M1 M0 M 0 is a short exact sequence of Zp [[K ]]-modules

such that M0 (and hence also M1 ) is free and nitely generated over Zp .6 To
obtain (3.5), one rst considers the long exact Tor-sequence for the resolution of
M and OK
. Since OK
is divisible and M0 is free over Zp , it yields the short exact
sequence 0 Tor1 (M, OK

) M1 Z p O K
M0 Z p OK
0. Using the Kummer
pn
n
sequence for OK OK , x x with n such that p M = 0, the Tor-term can
be evaluated as M (1). One now obtains
K
. . . (OK
(OK
nr Z M0 )
nr Z M1 )
H 1 (GK , M (1)) H 1 (GK , OK

nr Z M1 ) . . .
from the long exact GK -cohomology sequence. Kisin shows that
H 1 (GK , OK
nr Z M1 ) H (IK , OK nr Z M1 )
is injective and that a class lies in Hf1 (GK , M (1)) if its image in H 1 (IK , OK
nr Z
M1 ) is zero.
6 Note that given any short exact sequence 0 M M M 0 of Z -modules, with
p
1
0
M0 free and nitely generated over Zp , one can always extend the K action from M to a
continuous action on M0 .
74
Lemma 3.7.1. On discrete Zp [K ]-modules M on which p is nilpotent, the functor

M Hf1 (GK , M (1)) is right exact.
Proof. The assertion is immediate from formula (3.5), since it suces to verify
right exactness for sequences of nite Zp [K ]-modules.

Suppose now that VF is 2-dimensional, at and ordinary. As before, let DV,2
F
be the full subgroupoid of DV2F consisting of those framed deformations (VA , A , A )
ord,,2
over AugW (F) as follows: an
such that det VA
= . Dene the groupoid DVF
ord,,2
over (A B) is a quadruple (VA , A , A , LB ), where
object of DVF
(a) (VA , A , A ) DV,2
(A);
F
(b) LB VB := VA A B is a B-line, i.e., a projective B-submodule of rank 1
such that VB /LB is projective;
(c) LB is GK -stable subrepresentation and IK acts on LB via ;
(d) the extension class of 0 LB VB VB /LB 0 in
Ext1B[GK ] (VB /LB , LB )
= H 1 (GK , LB B (VB /LB ) )
lies in Hf1 (GK , LB B (VB /LB ) ).
of groupoids over AugW (F) is representable
Proposition 3.7.2. The functor DVord,,2
F
by a projective algebraizable morphism of formal schemes, which we denote by
ord,,2
: Lord,,2
Spf RV,2
.
VF
VF
F
becomes a closed embedding after inverting p. The scheme
The morphism ord,,2
VF
ord,,2
L
is formally smooth over W (F).
Proof. The proof of the rst part is more or less the same as that of Lemma 3.5.2.
For the second assertion one proceeds as in Lemma 3.5.6. For any closed point on
the generic ber of R, one proves that the completion of ord,,2
at is either zero
VF
or an innitesimal isomorphism (by showing that the respective functor is fully
faithful). Because the morphism ord,,2
is also projective, it must be a closed
VF
immersion.
It remains to verify formal smoothness. The proof proceeds as the proof of
Lemma 3.5.3. The key input is Lemma 3.7.1. It provides the desired lifting of an
extension 0 LB/I VB/I VB/I /LB/I 0 over B/I to an extension over B
for B B/I surjective with nilpotent kernel.

Corollary 3.7.3. Dene Spf RVord,,2
as the scheme theoretic image of ord,,2
.
VF
F
Then:
75
E is an E-valued
(a) If E/W (F)[1/p] is a nite extension and x : RV,2
F
point, then x factors through RVord,,2
if
and
only
if
the 2-dimensional EF
representation
of
G
corresponding
to
x
is
crystalline
and has the form
K

with : GK E an unramied character.

0 1
(b) RVord,,2
[1/p] is formally smooth over W (F)[1/p].
F
is a domain unless VF
(c) RVord,,2
= 1 2 for distinct characters i : GK F
F
such that 1 |IK = 2 |IK = |IK .
Proof. Let OE denote the ring of integers of E. Then x arises from an OE -valued
point and Vx has the properties listed in part (a) if and only if x lifts to an OE . By the previous proposition and the valuative criterion
valued point of Lord,,2
VF
for properness, the OE -valued points of Spec RVord,,2
and of Lord,,2
are in bijecVF
F
tion. Taking into account the relation between at deformations and crystalline
representations with weights in {0, 1}, cf. Proposition 4.2.1, this proves part (a).
Part (b) follows directly from the previous proposition.
The arguments used to prove connectivity in Lemma 3.5.3 can be applied
to part (c). Thus the number of connected components of RVord,,2
is in bijection
F
,2 F. The latter is a
with the number of connected components of Lord,,2
VF
R
VF
subscheme of P1F . If VF is not semisimple, it is a point, and if VF is scalar then it

is all of P1 . If VF is semisimple but the characters after restriction to inertia are
dierent, then again the subscheme is a point. In the remaining case, it consists
of two points.

Proposition 3.7.4. Let E be a nite totally ramied extension of W (F)[1/p] with
ring of integers O. Let : GK O . Consider now all groupoids over ArO (or
of RV2F such that:
AugO ). Then there exists a quotient RVord,,2
F
(a) If E /E is a nite extension and x : R E is an E -valued point, then x
if and only if the 2-dimensional E -representation
factors through RVord,,2
F

with
of GK corresponding to x is crystalline and has the form
0
1
: GK E an unramied character.
[1/p] is formally smooth over W (F)[1/p] of relative dimension equal
(b) RVord,,2
F
to 3 + [K : Qp ].
is a domain unless VF
(c) RVord,,2
= 1 2 for distinct characters i : GK F
F
such that 1 |IK = 2 |IK = |IK .
Proof. We only give the proof for p > 2. Twisting VF with a square root of
reduces us to the case = 1 treated in the previous corollary. The only claim that
remains to be proved is that on the dimension of RVord,,2
[1/p]. The computation
F
will be indicated later; cf. Proposition 4.2.4. For ordinary crystalline deformations,
so that Vx is an extension of two 1-dimensional crystalline representations, the
computation is particularly simple.
76
3.8
Complements
The methods of the previous section on at (hence weight 2) ordinary deformations may be generalized to ordinary deformations of arbitrary weight. Except for the computation of the Ext-group describing extensions of a twist of the
(k 1)-th power of the cyclotomic character by an unramied character, there are
few changes. One again uses an auxiliary functor L?,2
VF . Therefore we only describe
the setting and the result note however that the computation of the Ext1 groups and in particularly the surjectivity analogues to Lemmas 3.5.4 and 3.7.1 is
in certain cases quite involved.

1
We x K/Qp an unramied nite extension. We suppose VF
=
0 2
with 2 for a basis F of VF . We let : GK O be an unramied character
with 1/2 well-dened. Let RVk,,2
be the universal ring for framed deformations
F
k1
with the p-adic cyclotomic character.
of determinant
Theorem 3.8.1 ([30, 3.2]). Suppose 2 k p is an integer or that k = p + 1,
K = Fp and p > 2. Then:
(a) Rk,,2 has a quotient RVord,k,2
for k > 2; RVord,at,2
for k = 2 if VF is at and
F
F
ord,,2
RV F
for k = 2 if VF is non-at, such that, for all E/O[1/p] nite and for
all x : RVk,,2
E, the following equivalences hold:
F

1
(i) If k > 2, then x factors through RVord,k,2
if
and
only
if
V
x
0
2
F
with
2 unramied.
(ii) If k = 2 and VF is at, then x factors through RVord,at,2
if and only if
F

1
Vx 0 2 and Vx is at.
if and only
(iii) If k = 2 and VF is non-at, then x factors through RVord,,2
F

1/2
.
if Vx
0
1/2
Note that, as a quotient of Rk,,2 , one automatically has det Vx = k1 .
ord,?,2
(b) Except for k > 2 and 1 1
just dened are formally
2 = , the rings RVF
smooth over Qp and of relative dimension 3 + [K : Qp ], unless they are 0.
(c) The number of components of RVord,?,2

is given as follows:
F
(i) Case ? = k > 2: the number is 2 if 1 is unramied and = triv, else
it is 1.
(ii) Case ? = at: the number is 2 if 1 is unramied and = triv, else it
is 1.
(iii) Case ? = : the number is 1.
3.9
3.9.1
77
Appendix
The canonical subgroups of the absolute Galois group of a local eld
Let k be the residue eld of the nite extension F Q . The extension F nr F

is Galois and one has

Gal F nr /F
= Gal k/k
= Z,
where the rst isomorphism is the canonical one arising from reduction.
The absolute Galois group GF of F admits two canonical subgroups. First
there is the inertia subgroup IF of F , which is the kernel of the surjective homomorphism GF Gal(F nr /F ), i.e., one has the short exact sequence

1 IF GF Gal k/k 1.
Second, by the structure theory of IF , the pro- Sylow subgroup PF of IF is a closed
t
normal subgroup, the wild ramication subgroup
& of F . The quotient IF := IF /PF
is the tame quotient of IF ; it is isomorphic to q= Zq where the product is over
all rational primes q = . Thus, one has a short exact sequence
'
1 PF IF
Zq 1.
q=
The subgroup PF is also normal in GF : it is the pro- Sylow subgroup of IF .

Suppose now that = p. In order to focus on representations of GF into
GLd over rings in ArW (F) it is convenient to introduce the subgroup PF,p of IF as
follows. From the structure theory of IF it is apparent that there is a surjective
homomorphism tp : IF Zp . Introduce PF,p as the kernel, so that one has the
exact sequence
1 PF,p IF Zp 1.
The quotient Zp carries an action of Gal(F nr /F ) via the cyclotomic character;
this is immediate from the Kummer isomorphism

(

Zp
F nr pn p /F nr .
= Gal
n
3.9.2
Galois cohomology
We recall some results on the cohomology of GF for F Q nite, where

can be equal to or dierent from p; see [43]. For = p we set [F : Qp ] = 0
for = p it is, as usual, the extension degree of F over Qp . Let E be either
a nite extension of Fp or Qp , and let V be a continuous representation of GF
on a nite-dimensional E-vector space. Recall that V = Hom(V, E(1)). We write
hi (F, V ) = dimE H i (GF , V ).
78
Theorem 3.9.1 (Tate). The groups H i (GF , V ) are nite-dimensional E-vector

spaces for i Z and zero for i
/ {0, 1, 2}. Moreover:
h0 (F, V ) h1 (F, V ) + h2 (F, V ) = [F : Qp ] dimE V, and h2 (F, V ) = h0 (F, V ).
Corollary 3.9.2. Suppose dimE V = 1. Then h1 (F, V ) = [F : Qp ] unless V is trivial
or GF acts via the (mod p) cyclotomic character (if E is nite).

Let : GF GL2 (E), g 0 b be a reducible continuous representation
for characters , : GF E . Then
b Ext1GF (, )
= Ext1GF (1l, 1 )
= H 1 (GF , 1 ).
The class b is trivial if and only if the extension splits, i.e., if and only if is
semisimple.
Corollary 3.9.3. Suppose = p. If is non-split, then 1 is trivial or the cyclotomic character (mod p).
Let F nr be the maximal unramied extension of F inside a xed algebraic
closure F of F .
Proposition 3.9.4. Suppose that dimE V = 1 and V is unramied, i.e., IF acts
trivially on V . Then the groups H i (Gal(F nr /F ), V ), i Z, are nite-dimensional
over E. Moreover H 1 (Gal(F nr /F ), V ) = 0 unless V is trivial, in which case
h1 (Gal(F nr /F ), V ) = 1.
Corollary 3.9.5. Any unramied 2-dimensional representation is either split or an
extension of an unramied character by itself.
Combining the previous two corollaries, one deduces:
Corollary 3.9.6. Suppose = p. If as above is non-split and not unramied up
to twist, then 1 is the cyclotomic character (mod p).
3.9.3
Let F be a nite extension of Q as above with = p and residue eld k. Consider

the canonical homomorphism : GF Gk . The arithmetic Frobenius automorphism is a canonical topological generator of Gk . The Weil group of WF is dened
as 1 ( Z ), so that one has a short exact sequence 1 IF WF Z 1. Let
q = |k| and dene
|| || : WF Q, n g q n
for g IF and n Z.
Denition 3.9.7. Let L be a eld of characteristic 0 equipped with the discrete
topology. A WeilDeligne representation 7 over L is a triple (VL , , N ) such that
7 We follow the conventions of [54, 4] except that we express everything in terms of an
arithmetic Frobenius.
79
(a) VL is a nite-dimensional L-vector space,

(b) : WF AutL (VL ) is a continuous representation with respect to the discrete topology on VL , and
(c) N is a nilpotent endomorphism of VL such that (w)N (w)1 = ||w||N for
all w WF .
If L is a complete discretely valued eld, then A AutL (VL ) is called bounded if all
its eigenvalues have valuation zero, or equivalently if the characteristic polynomials
of A and A1 have coecients in the ring of integers of L. The WeilDeligne
representation (VL , , N ) is called bounded if () is bounded.
Observe that condition (b) is equivalent to the assertion that Ker( |IF ) is
nite.
Let E be a p-adic eld. For any continuous representation : GF Aut(VE ),
where VE is a nite-dimensional E-vector space, we may consider its restriction
to WF . One has the following important and elementary result due to Deligne:
Theorem 3.9.8 (Deligne). The following assignment sets up a bijection between
pairs (VE , ), where VE is a nite-dimensional E-vector space with the p-adic topology and : WF AutE (VE ) is a continuous representation, and WeilDeligne
representations (VE , , N ) over E (where, as in (b) above, VE is given the discrete topology). Given (VE , , N ), one denes
( n g) = ( n g) exp(tp (g)N ) for g IF , n Z.
The key step in the proof is Theorem 3.2.1, due to Grothendieck. The assignment (VE , , N ) (VE , ) is less explicit. It can be deduced from the proof
of Theorem 3.2.1.
The advantage of the WeilDeligne representation associated to a p-adic representation is that it can be expressed without any use of the p-adic topology
involved at the expense of introducing N . The concept is enormously important in the denition of a strictly compatible system of Galois representations to
have a good description also at the ramied places! For instance, let E/Q be an
elliptic curve with semistable but bad reduction at the prime . For any prime
p = , consider the representation Vp of GQ on the p-adic Tate module of A. Then
the action of IQ on Vp is unipotent and non-trivial, i.e., it is a non-trivial action
via the unique quotient Zp of IQ . In particular, the representation depends on p.
However, as the reader may verify, the associated
representation is

WeilDeligne
independent of p. One has (Ip ) = 1 and N 00 10 .
It is possible to describe a WeilDeligne representation also as a representation between algebraic groups. For this, one needs to dene the WeilDeligne
group WDF . As a group one has WF = lim
WF /J, where J ranges over the
JIF
open subgroups of IF . For any discrete group H (which may be innite) and any
80
ring R, denote by H R the constant group scheme on R with group H. Then the
group schemes
WF /J Q , J IF an open subgroup,
form an inverse system. One denes
WF Q = lim
WF /J Q .
Suppose E is any eld of characteristic zero. Then a homomorphism of algebraic

groups WF E AutE (VE ) will factor via WF /J for some open J IF and hence
is nothing else but a continuous representation WF Aut(VE ), where VE carries
the discrete topology.
Denition 3.9.9. The WeilDeligne group is the semidirect product
(
(WDF )Q = Ga WF Q = Ga
n IF Q ,
nZ
where multiplication is obtained as the inverse limit of the action of WF /J on Ga

dened by
(a, w) (a , w ) = (a + a ||w||, ww ), where a, a Ga (R), w, w WF /J.
The following result is straightforward:
Proposition 3.9.10. For any (discrete) eld E of characteristic zero, there is a
canonical bijection between d-dimensional WeilDeligne representations and algebraic representations (WDF )E GLd,E .
For further background we refer to [54].
3.9.4
Finite at group schemes
Let R be a commutative ring (or a scheme). By a nite at group scheme over

R one means a group scheme G which is nite at over R. In particular G is
ane. Let A denote its coordinate ring. It is a locally free (sheaf of) algebra(s)
over R. The rank of G is the rank of A over R. The group scheme structure on G
translates into a cocommutative Hopf algebra structure on A. This means that A
is an R-algebra equipped with R-linear maps : A A R A (comultiplication),
: A R (counit), : A A (coinverse) satisfying axioms which are dual to
those satised by a group (scheme).
Example 3.9.11. (a) For an abstract nite group , the ring A = Maps(, R)
is naturally an R-algebra. Moreover, with (f )(s, t) = f (st) as comultiplication, (f )(s) = f (1) as counit and (f )(s) = f (s1 ) as coinverse, it is a
cocommutative Hopf algebra.
81
(b) Let A = R[X]/(X m 1) with (X) = X X, (X) = X 1 and (X) = 1.

This denes the multiplicative group scheme m . It is etale over R if and
only if m is invertible in R.
For a xed nite extension K of Qp we now present some basic properties on
nite at group schemes over OK and at representations of GK .
A at representation of GK is a continuous representation of GK on a nite
abelian group V such that there exists a nite at group scheme G over OK so that
V
= G(K) as a Z[GK ]-module. Such a representation on V can be decomposed
into its primary parts. Flatness for components of order prime to p is characterized
by the following result it will not be needed in the main part of the lecture,
but we include it for completeness:
Proposition 3.9.12 ([52, 4, Corollary 3]). Suppose G is a nite at group scheme
over OK of order prime to p. Then the following three equivalent conditions hold:
(a) G is etale.
(b) The action of GK on G(K) is via 1 (Spec OK ).
(c) The action of GK is unramied.
Conversely (see Exercise 3.10.5), any unramied continuous representation of GK
on a nite abelian group is at.
From now on we assume that V is of p-power order. The following descent
result is presumably well known. Lacking a reference, we give a proof. Its idea was
suggested to us by J.-P. Wintenberger.
Lemma 3.9.13. Suppose that V is a continuous linear representation of GK on a
nite abelian p-group. If V is at over K nr , then it is at over K.
Proof. The Hopf algebra over OK nr giving the atness of V restricted to GK nr is
already dened over a nite unramied extension L/K such that GL acts trivially
on V . Let AOL denote a Hopf algebra over OL whose associated nite at group
scheme GOL satises GOL (K)
= V as Z[GL ]-modules. Dene AL as AOL OL L.
By Exercise 3.10.5(c), the invariants of AL under Gal(L/K) form a nite Hopf
algebra AK over K such that GK (K)
= V as Z[GK ]-modules for the associated
group scheme GK .
Dene AOK = (AOL )Gal(L/K) . We shall prove that
A O K O K OL
= AOL
(3.6)
is an isomorphism under the naturally given homomorphism (which regards AOK

as a subring of AOL and AOL as a OL -algebra). In other words, we shall show
that AOL satises Galois descent for Gal(L/K); see for example [60, 17]. By
Galois descent one also shows that the Hopf algebra structure descends from AOL
to AOK = AK AOL . For instance, to see that the comultiplication descends to
AOK , one may proceed as follows. Since comultiplication on AL arises by base
82
change from AK , the comultiplication : AL L AL AL is Galois equivariant.

Its restriction to AOL OL AOL maps to AOL . Hence it induces an OK -linear
homomorphism (AOL OL AOL )Gal(L/K) (AOL )Gal(L/K) . However (3.6) allows
us to identify the left-hand side with AOK OK AOK . Further details are left to
the reader.
We now prove (3.6). Let denote a uniformizer of OK . Because L/K is
unramied, is also a uniformizer of OL . We shall prove by induction on n that
for all n one has a natural isomorphism
(AOL / n AOL )Gal(L/K) OK /n OK OL / n OL
= AOL / n AOL .
(3.7)
For n = 1 recall that, by Hilbert 90, or rather the normal basis theorem, one has
kL
= kK [Gal(L/K)] as Galois modules, where kL and kK denote the residue elds
of L and K, respectively. Since AOL is a free OL -module, say of rank r, it follows
that

r
r
AOL /AOL
kK kK [Gal(L/K)]
= kL
= kK Gal(L/K)]r
= kK
as kK [G]-modules. One immediately deduces (3.7) for n = 1. For the induction
step, consider the sequence
n
0 AOL /AOL AOL / n+1 AOL AOL / n AOL 0.

Abbreviate G = Gal(L/K). Observe rst that taking G-invariants is exact. This
is so because the group H 1 (G, AOL /AOL ) vanishes again by the normal basis theorem. Tensoring the resulting short exact sequence with OL over OK and
comparing it with the given sequence yields
(AOL /AOL )G OK OL
AOL /AOL
/ (AOL / n+1 AOL )G OK OL
/ (AOL / n AOL )G OK OL

/ AOL / n+1 AOL

/ AOL / n AOL .
By induction hypothesis, the right and left vertical arrows are isomorphisms. By
the Snake Lemma, the same follows for the middle term. This proves (3.7). The
isomorphism (3.6) now follows by taking the inverse limit of (3.7).

Corollary 3.9.14. If M is a nite continuous Z[Gal(K nr /K)]-module and G is a
nite at group scheme over OK , then the representation M G(K) arises from
a nite at group scheme. In particular, M arises from a nite at group scheme.
Proof. Let V be the representation of GK on G(K) M . Because M is discrete
and unramied, there is a nite unramied extension L/K over which M becomes
trivial. Since G, being at over OK , will also be at over OL , we may apply the
previous result to V .
For the second assertion, note that for any n N the trivial GK -module
Z/(pn ) arises from a nite at group scheme G over OK ; cf. Exercise 3.10.5(d).
83
= Gal(K nr /K). Our next aim is to provide some background on

as dened in (3.4) for M a discrete (possibly innite) representation of Zp [Gal(K nr /K)]. Recall that the group schemes pn are at over any ring.
The following result follows from [27, Prop. 8.10.5] and its proof.
Write K
Hf1 (GK , M (1))
Proposition 3.9.15 ([27]). The group Hf1 (GK , Zpn (1))

/OK
= OK
isomorphic to the group of at extensions
is naturally
0 pn ,K V Z/pn Z 0
of GK -modules such that V is pn -torsion within the group
H 1 (GK , Zpn (1))
= K /K p
of all such extensions of GK -modules.

Note that the identication of the group of all extensions 0 pn (K)
V Z/pn Z 0 of GK -modules with H 1 (GK , Zpn (1)) is via Kummer theory.
The result in [27] is based on the construction of a universal at extension
of Z/pn by pn over Z[x1 ]. To have such a at extension dened over a ring R,
such as R = OK , the image of x has to be a unit in R.
Proposition 3.9.16. Let M be any nite discrete K -module which is pn -torsion, so
that in particular M is at. Then the group Hf1 (GK , M (1)) is naturally isomorphic
to the group of at extensions
0 pn ,K V M 0
of GK -modules as a subgroup of the group H 1 (GK , M (1)) of all such extensions
of GK -modules.
Mi for nite abelian p-groups Mi with a linear action of K ,
If M = lim
iI
the lemma gives a corresponding interpretation for Hf1 (GK , M (1)).
Proof. By [53, 4.2], the map that associates to any at extension 0 pn ,K
V M 0 the corresponding extension of GK -modules on the generic
ber is injective. Hence any at extension is described by a unique class c in
H 1 (GK , M (1)). By Lemma 3.9.13, the extension is at if and only if it is at over
some, or any unramied extension of K. Passing to a suitable such extension, we
may assume that K acts trivially on M . So then M is a nite direct sum of trivial
group schemes Z/pi Z. But then by Proposition 3.9.15, atness of c is equivalent
to being a class in OK
Z M K Z M . Again by Lemma 3.9.13 we can pass to
the limit over all unramied extensions of K, and hence c is at if and only if its
nr
image lies OK
Z M . By the denition of Hf1 , the latter condition
nr Z M K
1
is equivalent to c Hf (GK .M (1)).
84
3.10
Exercises
Exercise 3.10.1. Let = p and R = Zp [[x]]. Construct a continuous representation

Q )p GL2 (R) such that there are two closed points x, y on the generic
: (G

ber Spec R[1/p] whose logarithmic monodromy satises Nx = 0 and Ny = 0.
Exercise 3.10.2. Let : GQ GL2 (Fp ) be the trivial representation. Determine,
depending on the (non-zero) residue of mod p, the set of all bounded WeilDeligne
representations (VQp , , N ) whose corresponding p-adic representation admits a
model GQ GL2 (Zp ) whose mod p reduction is .
Exercise 3.10.3. (a) Let AugF be the full subcategory of AugW (F) whose objects
are pairs (A B) in AugW (F) with A = F. Prove that the restriction of
1
the groupoid L,2
VF to AugF is either the scheme Spec F or the scheme PF ,
depending on VF .
(b) Let AugF be the full subcategory of AugF whose objects are pairs (F B) in
AugF such that B is in ArW (F) . Prove that if L,2
VF over AugF is represented

by the scheme P1F , then L,2
over
Aug
is
represented
by the 0-dimensional
F
VF
scheme lim
1 X, where X runs over the zero-dimensional (not necessarily
XP
F
reduced) subscheme of P1F .
Exercise 3.10.4. Compute h1 (K, Qp (n)) and h1cris (K, Qp (n)) for a nite extension
K of Qp , for all n Z. Hints: Use without proof the following results; cf. [42].
(a) The dimension formulae derived from Tates local duality theory; cf. Theorem 3.9.1.
(b) For V a crystalline representation and Dcris (V ) its associated ltered -module8 one has
h1cris (K, V ) = dimQp Ext1cris (1l, V )
= h0 (K, V ) + [K : Qp ](dimQp V dimK Fil0 (Dcris (V ))).
(c) For K0 K the maximal subeld unramied over Qp and the Frobenius
automorphism of K0 one has Dcris (Qp (n)) = (K0 , = pn , Filn = K,
Filn+1 = 0).
Exercise 3.10.5. Let K be a nite extension of Qp and let V be a nite abelian
group carrying a continuous linear action of GK . Let L/K be a nite Galois extension where GL acts trivially on V . Dene a at OL -algebra AOL = Maps(V, OL )
as in Example 3.9.11(a). Show the following:
(a) The algebra AOL is the cocommutative Hopf algebra underlying a nite at
group scheme GOL over OL .
8 See
Section 4.2 and Appendix 4.6.2.
4. Flat deformations
85
(b) If for f : V OL and g Gal(L/K) one denes (gf )(v) = g(f (g 1 v)), then
this denes an action of Gal(L/K) on AOL which is compatible with the
Hopf algebra structure.
(c) The L-algebra AL = AOL OL L inherits a Hopf algebra structure from AOL .
The invariants (AL )Gal(L/K) form a cocommutative Hopf algebra over K
dening a nite at group scheme GK over Spec K such that GK (K)
= V as
a Z[GK ]-module and GK Spec K Spec L
= GOL Spec OL Spec L.
(d) Suppose L/K is unramied. Then AOK = (AOL )Gal(L/K) is the cocommutative Hopf algebra underlying a nite at group scheme GOK over Spec OK
with generic ber GK and base change to OL isomorphic to GOL .
Hint: Part (c) is proved by etale descent: one may use the additive Hilbert 90
theorem for L/K, which implies that L
= K[Gal(L/K)] as a Galois module. For
(d) one needs to show that the canonical homomorphism AOK OK OL AOL
is an isomorphism, i.e., that AOL descends to OK . The argument is similar to the
proof of Proposition 3.9.13.
Exercise 3.10.6. Adapt the proof of Theorem 2.2.1 to show the following. Let
E/Qp be nite, VE a continuous absolutely irreducible G-representation of E, and
B ArE . Suppose that VB , VB are deformations of VE to B such that Tr(|VB ) =
Tr(|VB ) for all G. Then VB and VB are isomorphic deformations.
4 Flat deformations
We follow mostly [37]. Some parts are motivated by the course of L. Berger on
p-adic Galois representation and discussions with K. Fujiwara and J.-P. Wintenberger during the course. As a reference for much of the p-adic Galois representations, the lecture notes [5] by L. Berger are highly recommended.
The appendix for this chapter summarizes very briey some basic results and
denitions on p-divisible groups, on (weakly admissible) ltered -modules and on
FontaineLaaille modules.
4.1
Flat deformations
Let K/Qp be a nite extension eld with residue eld k; write W = W (k) and
K0 = W [1/p]; x an algebraic closure K of K, and let GK = Gal(K/K). Denote
by VF a continuous representation of GK on a nite-dimensional F-vector space.
A representation V of GK on a nite abelian p-group is called at if it arises
from a nite at group scheme G over OK , i.e., if V
= G(K) as Z[GK ]-modules.
The following result is essentially due to Ramakrishna [47].
Proposition 4.1.1. Let A be in ArW (F) and VA in DVF (A). There exists a quotient
Aat of A such that, for any morphism A A in ArW (F) , VA = VA A A is at
if and only if A A factors through Aat .
86
Proof. Let V denote any at representation of GK on a nite abelian p-group,

say V
= G(K), and let V be any subrepresentation. Dene G G as the scheme
theoretic closure of V G(K); cf. [48, 2.1]. Then G is a nite at group scheme
over OK and V = G(K). Moreover the functor G/G is representable by a nite
at group scheme over OK with generic ber V /V . Let us give some details.
Let A be the ane coordinate ring of G. It is a free OK -module of nite rank
and carries the structure of a cocommutative Hopf algebra. The K-points of V
correspond to OK -homomorphisms A K. The intersection of the kernels of
these homomorphisms is an ideal I of A. Since it is the same as the intersection
of the corresponding ideal IK of the generic Hopf algebra AK = A OK K with A,
the ideal I is saturated as an OK -submodule of A. Over Spec K the subgroup
V is represented by the nite at subscheme Spec A/I of GK . This shows that
IK AK is a Hopf ideal. The latter property is inherited by I. Hence A = A/I
is a Hopf algebra which is nite at over OK . One veries that G = Spec A is
the desired subgroup scheme of G.
The above shows that, if : A A is a morphism in ArW (F) , then VA is
at if and only if V(A) is at. (For one direction use that, if (A)r A is a
r
(A)-module epimorphism, then VA is a quotient of V(A)
.) Similarly, if I, J A
are ideals such that VA/I and VA/J are at, then VA/(IJ) VA/I VA/J is at.
The second assertion implies the existence of a largest quotient A0 of A such that
VA0 is at. By the rst assertion, this A0 is the desired Aat .

DVF denote the subfunctor corresponding to at deCorollary 4.1.2. Let DVat
F
formations. Then DVat
D
VF is relatively representable.
F
Proof. Relative representability for groupoids over categories was dened in Definition 2.4.4. It simply means that for all in DVF the functor (Dat ) is representable. The latter is the functor of at representations arising from DVF (A)
via a homomorphism A A . The corollary follows from Lemma 4.1.1.
4.2
Weakly admissible modules and smoothness of the generic ber
Proposition 4.2.1. Let E/W (F)[1/p] be a nite extension and DVat

(OE ) with
F
corresponding representation V over E. Then there is a natural isomorphism of
groupoids over ArE ,
DVat
,
DVat
F ,()
where DVat
is the subgroupoid of DV of representations which are crystalline with
HodgeTate weights in {0, 1}. In particular, one has

DVat
(E[])
= Ext1cris V , V .
F ,()
W (F) such that A is at over OE and becomes
Moreover, for any (A OE ) in Ar
B E in ArE after inverting p, and for any VA DVF (A) mapping to V under
, one has
87
(A) VA
VA DVat
= Tatep G for G/OK a p-divisible group
F
VA A B is crystalline with weights in {0, 1}.
(4.1)
(4.2)
Proof. The equivalence in (4.1) is a result of Raynaud: VA lies in DVat

(A) if and
F
only if for all n N the representation VA A A/mnOE A is nite at. By [48, 2.3.1],
the latter is equivalent to VA being isomorphic to the Tate module of a p-divisible
group.
The equivalence in (4.2) uses Breuils result that a crystalline representation
with all HodgeTate weights equal to 0 or 1 arises from a p-divisible group [11,
Thm. 5.3.2], [32, 2.2.6].
From (4.1) and (4.2) and the denition of DVat
, the equivalence of functors
F ,()
is immediate, as is then the identication of the tangent space.

be the quotient of
Suppose that DVF is pro-represented by RVF and let RVat
F
at the completion
.
For
as
above,
denote
by
R
RVF which pro-represents DVat
F
(after W (F) E) along the kernel of . The equivalence in the above proposition
at is formally smooth over E
allows one to use Fontaine theory to show that R
and to compute its relative dimension; see Corollary 4.2.4. If K/Qp is ramied,
the diculty of RVat
lies in its special ber. As shown in [34], its analysis may
F
require delicate arguments.
To compute Ext1cris (V , V ), we recall some facts on weakly admissible ltered -modules from the lectures of L. Berger; cf. [6] see also Appendix 4.6.2.
Consider the fully faithful functor

Dcris : crystalline representations of GK =: Repcris
Q p GK

Fil
K := ltered -modules on K .
It is elementary to extend this equivalence to an equivalence with E-coecients
for any nite extension E/Qp

Dcris : Repcris
GK Fil
E
K,E .
Denote by D the image of V under Dcris . From the denitions and properties of
Dcris one deduces that

D , D .
Hf1 GK , adV = Ext1cris V , V
= Ext1Fil
K,E
Hf1 ,
For the denition of

see [6]. Using the period rings Bcris and BdeR in [6], the
following formula is derived:

dimE Hf1 GK , adV = dimE Hf0 GK , adV + d2 dimE Fil0 adD .
(4.3)
Let us rederive the latter dimension formula by an elementary approach given, for
instance, in [37]. For any weakly admissible ltered -module D over K, denote
by C (D) the complex
D
(1, id)
/ D DK /Fil0 DK
(4.4)
88
concentrated in degrees 0, 1.
Lemma 4.2.2. There is a canonical isomorphism

=
Ext1w.a. 1l, D H 1 C (D)
where 1l = K0 denotes the unit object in the category of weakly admissible modules.
Proof. Consider an extension
1l 0
0 D D
(4.5)
be a lift of 1 1l = K0 . Since (4.5)

of weakly admissible ltered modules. Let d D
is short exact, so is the sequence of Fil0 -terms of the induced sequence obtained
by base change from K0 to K (by the denition of exactness for ltered modules).
This shows that
=
0
DK /Fil0 DK D
K /Fil DK
is an isomorphism, and so we may regard d as an element of DK /Fil0 DK . Moreover
D (because 1 = 1l (1)). We thus associate the class
(1 )(d)
d)
H 1 (C (D))
((1 )d,
to the given extension.
Suppose now that (d0 , d1 ) D DK /Fil0 DK . To construct a corresponding
= D K0 on underlying K0 -vector spaces, dene
extension of 1l by D, set D

on D by ((d, 1)) = (D (d) + d0 , 1), and a ltration by
K = Fili DK + K d1 for all i 0
Fili D
K = Fili DK for i > 0. The extensions which arise from elements in the
and Fili D
image of (1, id) in (4.4) are split extensions. It is the content of Exercise 4.7.2 to
show that these two constructions induce the asserted isomorphism and its inverse.
(Note: The proof uses that the category of weakly admissible ltered -modules
is closed under extensions within the category ltered of -modules. Hence any
extension of weakly admissible modules is again weakly admissible.)

Let the notation be as in Proposition 4.2.1 and let D = Dcris (V ) be in
Fil
K,E . For B ArE , following Kisin one denes the category FilK,B of ltered modules on K over B: the objects are free and nitely generated K0 Qp B-modules
DB with a K0 idB -linear automorphism together with a ltration on DB K0 K
such that the associated graded pieces are free over B (but not necessarily over
K B). An object is weakly admissible if and only it is so if considered in Fil
K.
cris
One now denes the groupoid DVcris
over
Ar
by
dening
D
(B)
as
the
category
E
V
w.a.
of crystalline deformations of V to B, and similarly DD
over ArE by dening
w.a.
DD
(B)
as
the
category
of
all
weakly
admissible
deformations
of D to B.
89
w.a.
DD
Lemma 4.2.3. The functor Dcris induces an equivalence of groupoids DVcris
over ArE . Moreover, each of these groupoids is formally smooth.
Proof. For the proof of the rst statement, see Exercise 4.7.5 below. The proof
w.a.
of the second statement for DD
is rather straightforward. Indeed, one lifts the
free K Qp B/I-module DB/I to a free K Qp B-module DB . The isomorphism

B/I : DB/I DB/I lifts (non-uniquely) to a K Qp B-linear isomorphism
B : DB DB . To lift the ltration, one can use a complete set of idempotents
for K Qp E; via the canonical E-module structure of B, these idempotents lift
uniquely to K Qp B.

Corollary 4.2.4. Let the notation and hypotheses be as in Proposition 4.2.1 and
at is formally smooth of dimension
let D = Dcris (V ). Then the E-algebra R
0
1 + dimE adD,K /Fil adD,K .
The corollary assumes that DVF is representable. One could instead work
. The functor DVat,2
is then formally smooth of dimension
with DV2F and DVat,2
F
F ,()
d2 + dimE adD,K /Fil0 adD,K .
Proof. By Proposition 4.2.1 and the previous lemma, formal smoothness is clear.
The complex (4.4) shows that
h1w.a. (GK , adD ) h0w.a. (GK , adD ) = dimE adD,K /Fil0 adD,K .
As we assume the representability of the groupoid DVF , it has no extra automorphisms and so EndF[T ] (VF )
= VF , which implies that h0w.a. (GK , adD ) = 1. Now
use that

dimE Ext1cris V , V = dimE Ext1w.a. D , D

= dimE Ext1w.a. 1l, adD = dimE H 1 C adD
to obtain the assertion on the dimension from Lemma 4.2.2. Alternatively one can
simply use (4.3).
4.3
The FontaineLaaille functor and smoothness when e = 1
So far we have seen that the generic ber of DVat

is smooth. In general its special
F
ber may have a complicated structure. However, in the case where K/Qp is
unramied the groupoid is smooth over W (F). The principal tool to prove this is
FontaineLaaille theory, which we now recall.
The FontaineLaaille category MF1tor is dened as follows. Its objects are
nite, torsion W -modules M together with a submodule M 1 M and Frobenius
semilinear maps
: M M and 1 : M 1 M
such that
90
(a) |M 1 = p1 ;
(b) (M ) + 1 (M 1 ) = M .
The category MF1tor is an abelian subcategory of the category of ltered W -modules of nite length [21, 9.1.10]. In particular, any morphism on MF1tor is strict for
ltrations.
Note that, if p M = 0, then (M 1 ) = 0, and so comparing the lengths of
the two sides of (b) above shows that 1 is injective and
(M ) 1 (M 1 ) M.
(4.6)
Theorem 4.3.1 (FontaineLaaille, Raynaud). Suppose that K = K0 and p > 2.

Then there exist (covariant) equivalences of abelian categories
MF1tor
=
FL
nite at group schemes/W

at reps. of GK .
Raynaud
Proof. The rst equivalence is obtained by composing the anti-equivalence [21,

9.11] with Cartier duality. The second follows from Raynauds result [48, 3.3.6]
that, when e(K/K0 ) < p 1, the functor G G(K) is fully faithful and the
category of nite at group schemes over OK is abelian.

Remark 4.3.2. For A ArW (F) , one denes a category MF1A as follows: its objects
are quadruples (M, M 1 , , 1 ), where M is a nitely generated W Zp A-module,
M 1 M is a W Zp A-submodule, and : M M and 1 : M 1 M are
W idA -linear homomorphisms such that (a) and (b) hold. This is an A-linear
abelian category; see Exercise 4.7.5.
Theorem 4.3.3. Suppose K = K0 and p > 2. Then DVat
is formally smooth.
F
Independently of the condition EndF[GK ] (VF ) = F, the proof below will also
. Without using frames it is considerably
show the formal smoothness of DVat,2
F
at
more dicult to study DVF and its properties if EndF[GK ] (VF ) F. This problem
had been considered by K. Fujiwara.
Proof. Let MF MF1tor denote the object corresponding to VF . Then MF lies
naturally in MF1F by the full faithfulness of Theorem 4.3.1. Its underlying module
is free and nite over W Zp F, and the submodule MF1F is a W Zp F-direct
summand by (4.6). Let DMF denote the groupoid over ArW (F) such that DMF (A)
is the category of tuples (MA , MA1 , A , 1A , A ) such that (MA , MA1 , A , 1A ) lies in
MF1A and MA is a nite free W Zp A-module, MA1 is a W Zp A-direct summand,
and conditions (a) and (b) hold, and moreover A is an isomorphism

A
MF , MF 1 , F , F 1 .
MA , MA1 , A , 1A A F
Note that W Zp F will in general not be a local ring because in fact k Fp F will
not be a eld whenever the elds k and F over Fp are not linearly disjoint inside Fp .
91
However, observe that, since W (F) is a ring of Witt vectors, the complete set of
indecomposable idempotents e1 , . . . , en for k Fp F will lift to the unique such set
over W Zp W (F) and in turn induce the unique such set on W Zp A for any
A ArW (F) .
The following result is immediate from Exercise 4.7.5.
Lemma 4.3.4. The FontaineLaaille functor of Theorem 4.3.1 induces an equivalence of categories
=
FL : DMF DVat
.
F
Having the lemma at our disposal, to prove the formal smoothness of DVat
F
it suces to prove the formal smoothness of DMF . Let A be in ArW (F) , I A an
ideal and MA/I in DMF (A/I). We have to show that MA/I lifts to an object of
DMF (A). Consider rst the given data displayed in the following diagram:
MA/I
O
?
1
MA/I
A/I
/ MA/I
O
p
1A/I
/ MA/I .
1
The module MA/I
is a direct summand of MA/I as a W Zp A-module, by denition
of DMF . The homomorphism 1A/I is injective, because this holds true for 1F . This
1
in turn implies that LA/I := 1A/I (MA/I
) is a projective W Zp A/I-module of
nite rank. Using the idempotents mentioned above and the fact that any free
submodule of a local Artin ring is a direct summand, one can see that LA/I is a
direct summand of MA/I .
We can thus choose a free W Zp A-module MA , and projective W Zp A1
modules MA1 and LA lifting MA/I , MA/I
, LA/I , respectively, and in such a way
1
that MA and LA are direct summands of MA . By the projectivity of MA1 , one can
=
lift the W idA/I -linear homomorphism 1A/I to an isomorphism 1A : MA1 LA .
Using a complement to MA1 inside MA it is also straightforward to show that A/I
can be lifted.
4.4
The dimension of DVat

F
We wish to compute the dimension of the mod p tangent space of DVat

in the case
F
K = K0 . A direct way using FontaineLaaille theory is described in [37, 5.3.3].
We take a dierent route by working over the generic ber. There it amounts to
nding a more explicit form of the formula in Corollary 4.2.4. The computation
here is valid for all K and is taken from [34]. For K = K0 one can relate the
nal result via FontaineLaaille theory to the ltered torsion -module MVF
associated to VF . Moreover, in that case DVat
is smooth and so the result also
F
yields its dimension.
92
, say with values in

Let be a closed point on the generic ber of RVat
F
the nite extension E of Qp and associated E-representation V . Let D be the
corresponding ltered -module and G a p-divisible group over OK whose Tate
module Tate G = G(K) satises V
= VG := Tate G Zp Qp (see the proof of
Proposition 4.2.1).
Denote by tG the tangent space of G and by G its Cartier dual. Clearly,
tG is the tangent space of the connected component G 0 of G. By [53, Prop. 1],
the p-divisible group G 0 arises as the p-power torsion from a unique p-divisible
smooth formal Lie group over W . The dimension dim G of G is dened to be the
dimension of or, equivalently, the dimension of tG .
Let Cp denote the completion of K. The following isomorphism of continuous
GK -modules is taken from [53, p. 180, Corollary 2]:
V G Q p Cp
= tG (Cp )(1) tG (Cp ) .
(4.7)
Here, for any complete eld L Cp containing K, one has tG (L)

= Ldim G as
L[GK ]-modules and similarly for tG . Hence VG has HodgeTate weights 1 and 0
with multiplicities dim G and dim G .
=
To relate this to the functor MF1tor {nite at group schemes/W } from
Theorem 4.3.1, we observe that the inverse of this functor, extended to the isogeny
category of p-divisible groups, takes the form
G DG := Dcris (VG (1)) HomGK (Vp G , Bcris ).

In particular, as K-vector spaces, we have
Fil1 DG,K
= tG (K) .
(4.8)
However this is not quite sucient for the desired dimension calculation! The
point is that, so far, on the side of the p-divisible group we have ignored the action
of E (or its ring of integers). The action of E on V induces an action on G and
hence on tG as well, as its Cartier dual. This makes tG (K) into a K E-module
and the isomorphism (4.8) one of K E-modules.
To unify the arguments,
we assume that E contains the Galois closure of
&
K/Qp . Then K E
= : KE E, where ranges over the embeddings of K
into K these factor via E. Let e be the corresponding idempotents. Write d
for dimE e tG (K). From equation (4.7) one deduces that
d d = dimE e t
G (K) = dimE e tG (K) .
Theorem 4.4.1.
dimE DVat
(E[]) = 1 +
d (d d ).
93
Proof. By Proposition 4.2.4, we need to compute dimE adDG,K /Fil0 adDG,K . Now
adDG,K = DG,K KE DG,K

. The ltration of DG,K satises Fil0 = DG,K
is given by Fili (DG,K

) = (Fil1i DG,K )
Fil1 Fil2 = 0. The ltration on DG,K
where D DG,K
denotes the annihilator of D DG,K under the duality pairing
from linear algebra. Thus
Fil1 DG,K
= DG,K
Fil0 DG,K
= (Fil1 DG,K ) Fil1 DG,K
= 0,
+ DG,K (Fil1 DG,K ) . We

and it follows that Fil0 adDG,K = Fil1 DG,K DG,K
deduce that
adDG,K /Fil0 adDG,K
)/(DG,K /Fil1 DG,K (Fil1 DG,K ) )

= (DG,K /Fil1 DG,K DG,K
= (DG,K /Fil1 DG,K ) (D /(Fil1 DG,K ) )

G,K
.
= (DG,K /Fil1 DG,K ) Fil1 DG,K
Using the idempotents introduced above and the isomorphism in (4.8), the asserted
dimension for dimF DVat
(E[]) can easily be veried.
Suppose now that K = K0 . Then DVat

is smooth. In particular there is
F
a unique nite at group scheme G1 mod p which gives rise to VF . Moreover
we can assume that G has coecients W (F)[1/p]. One has G1 = G[p] and the
dimension of the tangent space tG (as well as its decomposition into -equivariant
parts) only depend on G1 . Moreover, by the theory of FontaineLaaille modules,
MF 1 agrees with tG (F) as an F-module. We introduce integers d as above for
the automorphisms of K0 = K. The following is an immediate corollary of
Theorems 4.3.3 and 4.4.1.
is formally smooth, then
Corollary 4.4.2. If K = K0 and p > 2, so that DVat
F
dimF DVat
(F[) = 1 +
F
d (d d ).
For arbitrary K/Qp , an important result of Kisin [34, Cor. 2.1.13] constructs
a projective F-scheme GRVF ,0 such that, for any nite extension F of F, the nite
at group scheme models of VF F F are in bijection with the F -valued points of
this scheme. The connected components of the scheme GRVF ,0 are in bijection with
the connected components of the generic ber Spec Rat [1/p], by [34, Cor. 2.4.10].
The latter components are smooth and their dimension is given by Theorem 4.4.1.
Since the tangent space of a p-divisible group G depends on G[p] only, the dimension
can be computed from any model from GRVF ,0 in the corresponding component.
Dierent components for the same VF can have dierent dimensions. The tuple (d )
is called a p-adic Hodge type in [34]. If K is unramied over Qp , then GRVF ,0
=
Spec F which follows already from Raynauds results.
94
Suppose again that K/Qp is an arbitrary nite extension. Assume now that
G is isogenous to G , at least after restriction to a nite extension of K, i.e., that G
is potentially (Cartier) self-dual. This happens in the following situations relevant
to deformations of Galois representations associated to weight 2 Hilbert modular
forms:
(a) G is the p-divisible group associated with an abelian variety over OK (i.e.,
with good reduction): a polarization exists over OK for a nite extension K

of K. It induces an isomorphism G
.
= G over OK
(b) G is the p-divisible group associated with a parallel weight 2 Hilbert modular Hecke eigenform f whose level is prime to p. At least if f arises from a
Shimura curve C over a totally real eld F , then the p-adic Galois representation of f arises from a subfactor of the Jacobian JC of C over F , which has
good reduction at p. Essentially by part (a) the associated p-divisible group
is potentially self-dual.
tG (K ) and that this isoFrom the isogeny over K it follows that tG (K ) =
morphism is compatible with extra endomorphisms such as those coming from E.
In particular, d is even and d = d/2 for all . Thus
dimE DVat
(E[]) = 1 + [K : Qp ](d/2)2 .
(E[]) = 1 + [K : Qp ].
For d = 2 one recovers the expected result dimE DVat
Note that the argument basically rests on the fact that the HodgeTate
weight is invariant under nite extensions of the base eld.
4.5
Complements
Suppose that VF is an irreducible 2-dimensional representation of GQp of any Serre

weight 2 k(VF ) p. Then the methods of the present lecture on at (hence
weight 2) deformations can be generalized to study (low weight) crystalline deformations. The reason is simply that in this range of weight (2 k p), the theory
of FontaineLaaille is still applicable. On the FontaineLaaille side, one considers 2-dimensional ltered torsion modules of weight at most k. An A-representation
(A ArW (F) ) is then said to be crystalline of weight k if it arises via the (inverse)
FontaineLaaille functor from an FL-module of weight k. In this perspective, the
analogue of Lemma 4.3.4 is no longer an assertion but a denition. We simply state
the results from [30], in particular [30, 3.2.3]. (Analogous results hold whenever
K/Qp is unramied.)
Theorem 4.5.1 ([30, 3.2]). Suppose that 2 k p, that VF is irreducible of
Serre weight k, and that p > 2. Then the deformation functor for framed weight k
crystalline deformations of VF of determinant k1 is formally smooth over W (F)
of relative dimension 4.
95
There is one further deformation condition considered in [30, 3.2] for VF irreducible and of weight 2: semistable deformations with associated WeilDeligne parameter (in the sense of Fontaine and for p-adic lifts) given by the pair (( mod p)
1, N ) and with N non-trivial. The result in this case is due to Savitt [50, Thm.
6.2.2(3)].
Theorem 4.5.2 ([30, 3.2]). Suppose that VF is irreducible of Serre weight 2 and
p > 2. Let O be the ring of integers of a totally ramied extension of K0 . Then
the deformation functor for framed weight 2 semistable deformations of VF of determinant on ArO is representable. Provided that O is suciently large, it is
isomorphic to O[[X1 , . . . , X5 ]]/(X4 X5 p).
4.6
Appendix
4.6.1 p-divisible groups

We only recall the most basic notions on p-divisible groups. As a reference we
recommend Tates seminal article [53] and his notes [55].
Denition 4.6.1. Let h 0 be an integer and let S be a scheme. A p-divisible
group G of height h over a scheme S is an inductive system

G = Gn , n n0
where, for each n,
(a) Gn is a nite at commutative group scheme over S of order pnh , and
(b) the sequence
pn
n
Gn+1 Gn+1
0 Gn
is exact (i.e., (Gn , n ) can be identied with the kernel of the homomorphism
multiplication by pn on Gn+1 ).
A homomorphism f : G H of p-divisible groups G = (Gn , n ), H = (Hn , n )
is a compatible system f = (fn )n0 of S-group homomorphisms fn : Gn Hn
such that n fn = fn+1 n for all n 0.
If G = (Gn ) is a p-divisible group, we shall often use the perhaps more intuitive
notation G[pn ] for Gn (see Examples 4.6.2 below).
Note that, if p is invertible in S, then the Gn will be etale over S.
Examples 4.6.2. Let A S be an abelian scheme over S. Then multiplication
pn
by pn is a nite at homomorphism A A of group schemes. Thus the kernel,

denoted by A[pn ], is a nite at commutative group scheme over S. Denote by
n : A[pn ] A[pn+1 ] the canonical inclusion. If g denotes the dimension of A, then
A[p ] := (A[pn ], n )n0 is a p-divisible group of height 2g.
Consider the particular case where A = E is an elliptic curve over a nite
extension K of Qp . Then E[p ] is a p-divisible group over K. It is completely
96
determined by the Tate module of E at p. Suppose that E has good reduction and
denote by E a model over the ring of integers O of K. Then E[p ] is a p-divisible
group over Spec O of height 2.
4.6.2 Weakly admissible ltered -modules
Much of the material of this and the following section goes back to Fontaine and
his coauthors. We suggest [5] and [13, Ch. 8, 12.4] as references. They contain
many further references.
Fix an algebraic closure Qp of Qp and let K, E be nite extensions of Qp
inside Qp . The eld K will take the role of the base and the eld E that of a
coecient ring. Suppose k is the residue eld of K, so that K contains K0 :=
W (k)[1/p]. Let v be the valuation on Qp such that v(p) = 1. Let : K0 K0 be
the Frobenius automorphism induced on k, e.g. via the Witt vector construction.
Denition 4.6.3. A ltered -module of rank n on K over E is a tuple

D = D, , Fili DK iZ
consisting of
(a) a free K0 Qp E-module D of rank r,
(b) an isomorphism : ( idE ) D D, i.e., a K0 -semilinear automorphism ,
and
(c) an exhaustive separating decreasing ltration
i

Fil DK iZ
of DK := D K0 K by K Qp E-submodules.
A morphism : D D between ltered -modules D = (D, D , {Fili DK }iZ )

and D = (D , D , {Fili DK
}iZ ) is a K0 Qp E-linear homomorphism : D D
which is compatible with the action of and preserves the ltration.
The category of all ltered -modules on K over E is denoted by MF
K,E .
Note that the ltration in (c) need not satisfy any compatibilities with the
previous data. However, the ltration datum imposes a strong restriction on the
morphisms in the category MF
K,E . In particular it limits the set of subobjects of a
given ltered -module. Note also that the Fili DK need not be free over K Qp E.
Denition 4.6.4. Suppose : D D and : D D are morphisms in MF
K,E .
Then D D D is a short exact sequence, written
0 D D D 0,
97
if 0 D D D 0 is an exact sequence of K0 -vector spaces and for all

i Z the induced sequences 0 Fili DK
Fili DK Fili DK
0 are exact as
sequences of K-vector spaces.
One says that D is an extension of D by D if there exists a short exact
sequence 0 D D D 0.
For D MF
K,E one can dene exterior and symmetric powers as well as
duals, where one takes the induced endomorphisms and ltrations.
If D is in MF
K,E , it is clearly also in MFK,Qp . Under this forgetful functor,
the dimension will increase by a factor of dimQp E. By detK0 D we denote the
element
dimK0 D
D
K0
in MF
K,Qp of rank 1. By the previous remark,
dimK0 D = dimK0 Qp E D dimQp E.
Denition 4.6.5. The Hodge slope of D MF
K,E is dened as
tH (D) = max{i Z : Fili (detK0 D)K = 0},
and its Newton slope is dened as
tN (D) = v(detK0 (x)/x), for any x detK0 D
{0}.
The Newton slope is well-dened, since detK0 D is of rank 1 over K0 and since for
any x K0
{0} one has v((x)) = v(x).
Hodge and Newton slopes are used to dene a semistability condition on
ltered -modules:
Denition 4.6.6. A ltered (, K, E)-module is called (weakly) admissible if
tH (D) = tN (D)

and for all subobjects D D in the category MF
K,Qp one has tH (D ) tN (D ).
This is a priori a rather tricky denition, since the subobjects to be considered

for weak admissibility are subobjects in MF,N
K,Qp . However, one has the following
[12, Prop. 3.1.1.5]:
Proposition 4.6.7. A ltered (, K, E)-module is admissible i tH (D) = tN (D)
and for all -stable sub-K0 Qp E-modules D D one has tH (D ) tN (D ),
where D carries the induced and ltration.
Note that the test objects D need not be free over K0 Qp E, and so they
may not lie in MF,N
F,E .
Categorically, the introduction of the semistability concept has the following
remarkable consequence (for a proof, see [13, Thm. 8.2.11]):
Theorem 4.6.8 (Fontaine). The full subcategory of MF
K,E of weakly admissible
objects is abelian and closed under extensions.
98
4.6.3

FontaineLaaille modules
We assume that K = K0 and so we drop the subscript K at DK . Let

D = (D, , {Fili DF }iZ )
be an admissible ltered -module with coecients in E. Let W = OK0 = W (k).
Suppose that D is eective, i.e., that Fil0 D = D, and moreover Filp D = 0.
Denition 4.6.9. A strongly divisible OE -lattice in D is a free W Zp OE -submodule
D such that
(a) [1/p] = D,
(b) is stable under ,
(c) (Fili ) pi for all i 0, where Fili = Fili D, and

i
i
(d)
i0 p (Fil ) = .
A strongly divisible lattice is called connected if is topologically nilpotent
for the p-adic topology on .
For the following, see [21] or [13, Thm. 12.4.8]:
Theorem 4.6.10 (FontaineLaaille). There are exact quasi-inverse anti-equivalences between the category of strongly divisible lattices with Filp = 0 and the
category of OE [GK ]-lattices in crystalline GK -representations with HodgeTate
weights in the set {0, . . . , p 1}.
Denition 4.6.11. A FontaineLaaille module M = (M, M , (Fili M )iZ ) over W
is a nite length W -module M equipped with a nite and separated decreasing
ltration (Fili M ) and -semilinear endomorphisms iM : Fili M M such that
(a) for all i 0, the following diagram commutes:
FiliO M
?
Fili+1 M
(b)
/M
O
p()
i+1
/ M,
Im(iM ) = M , and
(c) Fil0M = M .
The category of such is denoted by MFtor . If the ltration step 1 is non-zero,
but step 2 is zero, then we write MF1tor . One says that M is connected if 0M is
nilpotent.
99
Example 4.6.12. If is a strongly divisible lattice, then for each n > 0 we obtain
a FontaineLaaille module M by setting M = /pn , taking Fili M to be the
image of Fili under the natural quotient map, and letting iM be the reduction
of pi restricted to FiliM .
This FontaineLaaille module is connected if and only if is connected.
The following result is stated in [13, Thm. 12.4.12] unfortunately without
proof.
Theorem 4.6.13. Consider the contravariant functor
M HomFil, (M, Acris Qp /Zp )
from the category of FontaineLaaille modules M with one-step ltration that
satises Fil0 M = M and Filp M = 0 to the category of p-power torsion discrete
GK -modules. If p > 2, this is an exact and fully faithful functor into the category
Reptor GK , i.e., continuous p-torsion GK -modules. If p = 2, the same statement
holds if one restricts the functor to connected FontaineLaaille modules.
4.7
Exercises
Exercise 4.7.1. Formulate and prove Proposition 4.2.1 for framed deformations
and verify the assertion made after Corollary 4.2.4 .
Exercise 4.7.2. Check that the two constructions in the proof of Proposition 4.2.2
of the isomorphism

=
Ext1w.a. 1l, D H 1 C (D)
are well-dened and inverse.
Exercise 4.7.3. Give an explicit description of the isomorphism

Ext1w.a. D , D
= Ext1w.a. 1l, adD
used in the proof of Corollary 4.2.4.
w.a.
Exercise 4.7.4. Prove that the functor DD
in Corollary 4.2.3 is formally smooth.
E
Exercise 4.7.5. Let C be a ring (commutative with 1). Recall that an additive
category C is C-linear if for all M C one has a homomorphism M : C
EndC (M ) such that for all M, N C and all HomC (M, N ) diagram (4.9)
commutes (this also makes HomC (M, N ) into a C-module). This exercise provides
a categorical approach to equipping suitable subcategories of a C-linear category
with a larger endomorphism ring than C. It will be applied to several of the
categories in this lecture.
Let now C be a C-linear abelian category in which all objects have nite
length over C.
100
(a) For A ArC , dene a category CA as follows. Objects of CA are pairs (M, )
with M C and : A EndC (M ) a C-linear homomorphism. Morphisms
from (M, M ) to (N, N ) in CA are morphisms : M N such that for
all a A the following diagram commutes:
M
M (a)
/M
N (a)

/ N.

N
(4.9)
Show that CA is abelian and A-linear. Show also that for any nitely generated A-module N the tensor product A N is well-dened.
Hint: If N is free over A, this is obvious. Else use a 2-step resolution of N by
free nitely generated A-modules.
(b) For C = Zp and C = MF1tor , describe CA for A ArC . An object of CA
contains in particular an inclusion M 1 M of A-modules. Show that M 1
must be a direct summand.
Hint: Use the notion of pure submodule from [38, Appendix to 7] and the
abelianness of C.
(c) Suppose that D is a second C-linear abelian category in which all objects
are of nite C-length and that F : C D is an exact C-linear functor. Show
that for all A ArC it induces via the construction in (a) an exact C-linear
functor FA : CA DA .
(d) The functor FA from (c) is compatible with the operation A N for any
nitely generated A-module N and it restricts to an exact subfunctor on
objects which are A-at.
(e) For C a nite extension of Qp and C the category of weakly admissible -modules on K over C, describe CB for B ArC . An object of CB is equipped
with a ltration (over K Qp B). Show that the subobjects of this ltration
are direct summands as K Qp B-modules.
Presenting global over local deformation rings
A p-adic Galois representation of the absolute Galois group of a number eld is

called geometric if it is unramied outside nitely many places and at all places
above p it is de Rham in the sense of Fontaine. When the number eld F is
totally real and the representation is into GL2 , one typically also requires the
representation to be totally odd. Conjecturally, the latter should automatically be
satised if not all HodgeTate weights are equal but no proof is known. By the
FontaineMazur conjecture, geometric 2-dimensional odd Galois representations
5. Presenting global over local deformation rings
101
over totally real elds should (up to twisting by powers of the cyclotomic character)
arise from Hilbert modular forms. Then they are called modular. This is proven
in some instances. But even for Q the proof of the FontaineMazur conjecture is
not complete.
In practice, it is important to construct geometric Galois representations,
even in situations when it is not known that they are modular. An important
method is to combine the proof of the potential modularity theorem by Taylor
[57, 58], i.e., an R = T theorem over an enlarged (totally real) base eld, with
a technique from deformation theory. The method has proved useful in many
instances beyond GL2 , such as Galois representations of unitary or symplectic
type, e.g. [1, 2].
In the present lecture, the focus will be on the deformation theoretic part:
we shall construct and analyze universal deformations rings RS for representations of Galois groups of global elds which locally satisfy conditions that ensure that the deformations are geometric in the above sense. Concretely, one requires that the deformations are unramied outside nitely many places and odd.
At those primes = p where ramication is allowed, one xes a nite set of inertial
WD-types and imposes these on the deformations. Finally, at places above p one
chooses deformation conditions that lead to p-adic Galois representations which
are crystalline or ordinary of low weight, semistable of weight 2, or of potential
BarsottiTate type (for instance).
The main result of this lecture see Theorem 5.4.1 for a precise statement
is the following dimension bound: if RS is non-zero, then
dimKrull RS 1.
(5.1)
Let us indicate how to derive from Taylors results on potential modularity,

e.g. [57, 58], the existence of lifts of mod p Galois representations satisfying the
conditions in the denition of RS . If Taylors result is applicable, then it implies
that RS /(p) is nite. Since the length of RS /(pn ) is at most n times the length
of RS /(p), and since RS is p-adically complete, one deduces that RS is a nitely

generated Zp -module. The lower bound (5.1) thus implies that

RS
) 1 *
p
red
= E1 Er
for suitable p-adic elds Ei (i.e., nite extensions of Qp ). The dening properties of
RS yield geometric Galois representations GF GL2 (Ei ) satisfying the conditions

prescribed by the corresponding functor at all places above p and and possibly
at some further places.
Lower bounds as in (5.1) were rst obtained in [8]; cf. also [9] and in
fact the results therein were sucient for the proof of Serres conjecture in the
level 1 case [28]. However, the results in [8] required the local deformation rings
102
to be complete intersections. In recent work [31], Kisin gave a dierent approach

to obtain such bounds. This greatly enlarged the range where a lower bound as
in (5.1) can be proved. Moreover it simplied the arguments considerably. So here
we follow Kisins approach. As a further reference we recommend [30, Ch. 14].
In this lecture we x the following notation pertaining to number elds:
F will be a number eld and S will denote a nite set of places of F containing
all places v | p and v | .
By GF,S , or simply GS , we denote the Galois group of the maximal outside
S unramied extension of F inside a xed algebraic closure F of F .
For any place v of F , we denote by Gv the absolute Galois group of the
completion Fv of F at v. We x for each v a homomorphism F F v
extending F Fv . This yields a group homomorphism Gv GS .
By VF we denote a continuous F[GF,S ]-module of nite dimension d over F.
We write ad0 ad = adVF for the subrepresentation on trace zero matrices.
All deformation functors (or categories of groupoids) considered will be func O , where O is the ring of integers of a
tors on either the category ArO or Ar
totally ramied extension eld of W [1/p] and thus with residue eld F.
We x a lift : GS O of det VF . This denes subfunctors D and D,2
of D and D2 by requiring det VA = for lifts.
5.1
Tangent spaces
We provide some complements to Section 1.4. Here G stands either for GF,S or Gv ,
and VF for VF|G . As in Lecture 1, one proves:
O is formally
Proposition 5.1.1. (a) The functor D,2 D of groupoids over Ar
,2
smooth. The functor D
is always representable and the functor D is
0
representable when h (G, ad) = 1.
(b) The tangent space D (F[]) is isomorphic to

H 1 G, ad0 := Im H 1 G, ad0 H 1 G, ad .
(c) There is a short exact sequence

0 ad0 /H 0 G, ad0 D,2 F[] D F[] 0.
Remark 5.1.2. If p does not divide the degree d of VF , then ad
= ad0 F as a
G-representation, and in this case H 1 (. . .)
= H 1 (. . .). However, for d = 2 and
p = 2 (for instance) one needs H 1 (. . .) .
Applying Proposition 5.1.1(b) and (c) to the rst ve terms in the long exact
cohomology sequence obtained from 0 ad0 ad F 0, one nds
103
Corollary 5.1.3. One has

dimF D,2 (F[]) = d2 1 + h1 (G, ad0 ) h0 (G, ad0 ).
If in the deformation problem one xes m bases of VA instead of just one, then
one has to add (m 1)d2 to the right-hand side of the above formula.
5.2
Relative presentations
We now turn to a situation which is closer to our nal aim. Thus, from now on,
the representation : GS AutF (VF ) is absolutely irreducible, and
we x a subset of S which is assumed to contain all places v of F dividing
p or .
Corresponding to the above set-up, we introduce the following deformation
functors and associated universal deformation rings:
deformation functor universal ring
v :
Dv = D,2
= DVF|G|Gv
v
,2
D
S = DVF |G
D,2
,S
Rv,2
RS
,2
R,S
,

where the functor D,2
,S : ArO Sets is dened by the assignment
$
+
,
$
A (VA , A , (v )v ) $ (VA , A ) DS (A), (v )v are bases /
=.
of VA with A (v ) = F v
The functor D,2
,S provides the crucial link between the global and local situation:
D,2
,S
smooth, rel.
dim 4|| 1
(VA , A , (v )v ) ((VA )|Gv , A , v )v
&
v
Dv
(VA , A , (v )v ) (VA , A )
D
S .
The formal smoothness of D,2

,S over DS follows from Proposition 5.1.1(a). The formula for the relative dimension is proved in the same way as Proposition 5.1.1(c).
,2
Corollary 5.2.1. (a) R,S
= RS [[x1 , . . . , x4||1 ]].
,2
.
(b) There is a natural homomorphism Rloc := v Rv,2 R,S
104
The ring RS was rst studied by Mazur in [39]. It is an interesting object,

since, for F totally real and odd, it can be naturally compared with a big Hecke
algebra of Hilbert modular forms. On the other hand, it can be recovered from
,2
the universal representation S : GS GLd (R,S
) as the ring generated by the
traces of S over O. Ultimately it is a quotient of RS which will be of interest
to us. The local rings at the places in will be useful in order to pass from
,2
R,S
to this quotient. The use of framed deformations is a clean way to deal with
non-representability issues of the functors DVF|G|Gv .

v
Key Lemma 5.2.2. Consider the canonical homomorphisms

D,2
,S (F[])
2,1
#
/
D,2
v (F[]),
v
0
H (GS , ad )
H 2 (Gv , ad0 ).
v
,2
has a
Set r = dimF Ker 2,1 and t = dimF Ker 2 + dimF Coker 2,1 . Then R,S
presentation

,2
Rloc [x1 , . . . , xr ]]/(f1 , . . . , ft )
= R,S .
Remarks 5.2.3. (a) The proof will be given in Section 5.6.

(b) The lemma makes no assumption about the shape of the fi . Some of the fi
could be zero. Therefore t is only an upper bound for the minimal number
of relations.
(c) The value of r is optimal, since 2,1 is the homomorphism of mod mO tangent
,2
Spec Rloc .
spaces induced from Spec R,S
(d) Before we compute r t in the following section, let us determine the dimension of Coker 2 . The diagram dening 2 is part of the terms 79 of the
9-term PoitouTate sequence
H 2 (GS , ad0 )
H 2 (Gv , ad0 ) H 0 (GS , (ad0 ) ) 0.
v(S)
Using local Tate duality H 2 (Gv , ad0 )

= H 0 (Gv , (ad0 ) ) and some elementary linear algebra, we nd (indeed!)
:= dimF Coker 2

#

= dimF Ker H 0 GS , (ad0 )
H 0 Gv , (ad0 ) .
vS
If H 0 (GS , (ad0 ) ) = 0, which is for instance the case whenever the image
of is non-solvable, or if S
= , and thus by our hypothesis on the
dierence contains a nite prime, then = 0.
105
requires actually more bookkeep(e) The analogous computation for Spec R,S
ing due to the innite places. For Spec R,S

, the set is supposed to only
0
Gv
contain places at which ad VF
= 0; however, the innite places do not
satisfy this requirement.
5.3
Numerology
Lemma 5.3.1. If contains all places above p and , then r t + = || 1.

Proof. Tates duality theory for global (and local) elds gives us the following
formulas for the EulerPoincare characteristic of Galois cohomology (which is
dened to be the alternating sum of the dimension of the zeroth, rst and second
term of Galois cohomology):

(GS , ad0 ) = [F : Q] dim(ad0 ) +
h0 (Gv , ad0 ),
(5.2)
v|
dim(ad )[Fv : Qp ]
0
(Gv , ad ) =
h0 (Gv , ad0 )
if v | p,
if v | ,
otherwise.
(5.3)
We deduce that
r t + = dimF Ker 2,1 dimF Coker 2,1 dimF Ker 2 + dimF Coker 2

= dimF D,2
dimF Dv (F[]) h2 (GS , ad0 ) +
h2 (Gv , ad0 )
,S (F[])
v
Cor. 5.1.3
||d 1 + h (GS , ad ) h (GS , ad ) h (GS , ad )

d2 1 + h1 (Gv , ad0 ) h0 (Gv , ad0 ) h2 (Gv , ad0 )

2
= (GS , ad0 ) +
dim(ad0 )[F : Q]

v|p
h0 (Gv , ad0 )
v|
dim(ad )[Fv : Qp ] +
0
(Gv , ad0 ) + || 1
v
(5.2), (5.3)
h0 (Gv , ad0 ) + 0 + || 1 = || 1,
v|

since [F : Q] = v|p [Fv : Qp ]. Note that after the third and fth = the rst line
contains the global and the second the local contribution.
5.4
Geometric deformation rings
In this and in the following section we assume the following:
106
F is totally real.
is odd and of degree 2 over F (and still absolutely irreducible).
contains all places above p and (as before).
,2
For each place v in , choose a relatively representable subfunctor Dv Dv

,2
such that the corresponding universal ring Rv (a quotient of Rv,2 ) satises:
,2
Rv
is O-at,
) *
,2
Rv
1
p
3
3 + [Fv : Qp ]
is regular of dimension
if v | p, ,
if v | p,
if v | .
Suitable deformation conditions for v | p, were described in Lecture 3. The natural choice is to x a set of inertial WD-types for lifts to the generic ber.
Rings of the above type for v | p were constructed at the end of Lecture 3 and
in Lecture 4. Possible deformation conditions are: low weight crystalline at v if
is absolutely irreducible, low weight ordinary at v for ordinary , and potentially
BarsottiTate. In some cases, Fv = Qp is required; in others, that Fv is unramied
over Qp , etc. Here we shall simply assume that we do have (framed) deformation
functors at places above p which satisfy the above requirements.
For v | we shall shortly describe the deformations and the corresponding
rings. They describe odd deformations.
,2
The above hypotheses on Rv have the following consequences:
,2
O-at. Its generic
ber is regular of dimension
(a) The ring Rloc := v Rv is
3|| (this uses the fact that
[F
:
Q
]
=
v
p
v|p
v| 1 = [F : Q]). Hence
dimKrull Rloc 3|| + 1.
,2
(b) The corresponding functors DS and D,S (where the latter again includes
,2
a choice of || bases of VA ) are representable, where (for instance) D,S is

dened as the pullback in
D,2
,S
&
v
Dv
Dv .
&
,2
,2
(c) The global universal ring R,S is isomorphic to R,S
Rloc Rloc , and therefore
Lemma 5.2.2 yields
,2
R,S
= Rloc [[x1 , . . . , xr ]]/(f1 , . . . , ft )
107
with r, t as in that lemma. Since r t = || 1 by Lemma 5.3.1, part (a)

,2
yields dimKrull R,S 4|| .
,2
By Remark 5.1.2(d), the map D,S DS is formally smooth of relative

dimension 4|| 1. We deduce the following from part (c):
Theorem 5.4.1. If = 0, then dimKrull RS 1.
5.5
Odd deformations at real places
At a real place, any 2-dimensional odd residual representation is of the form

: Gal(C/R) GL2 (F),
with det (c) = 1 in F, for c the complex conjugation in Gal(C/R). Up to
conjugation, one of the following three cases occurs:

(i) p > 2, (c) = 10 10 ;
(ii) p = 2, (c) = 10 11 ;

(iii) p = 2, (c) = 10 01 .
Any framed representation of Gal(C/R) is determined by the image M of c
and the latter is subject to the condition that M 2 = id. If we further want to ensure
that M has eigenvalues 1 and 1, we need to x its characteristic polynomial to be
X 2 1. Let M(X 2 1) denote the moduli space of 2 2 matrices of characteristic
polynomial X 2 1. Its completion at the matrix (c) is the wanted universal
ring (as may be checked easily). This is precisely the construction that was used in
the proof of existence of RV2F in Proposition 1.3.1. Let us carry out this procedure
explicitly for case (iii) (the other onesbeing similar but simpler):
b
If we start with M = 1+a
c
1+d , then the conditions Tr = 0 and det = 1
lead to
M(X 2 1) = Spec O[a, b, c, d]/((1 + a) + (1 + d), (1 + a)(1 + d) + 1 bc)
= Spec O[a, b, c]/((1 + a)2 + 1 bc)
= Spec O[a, b, c]/(2a a2 bc)
,2
O[[a, b, c]]/(2a + a2 + bc). The latter is a domain with generic
and hence R,odd =
regular ber of dimension 2.
,2
In cases (i) and (ii), similar calculations lead to R,odd
= O[[x1 , x2 ]].
5.6
Proof of Key Lemma 5.2.2
Note that we now work again with representations of general degree d. To simplify
,2
notation, we set Rgl = R,S
. For r as in Lemma 5.2.2, we choose a surjective ring
homomorphism
:= Rloc [[x1 , . . . , xr ]] Rgl .
: R
108
by mgl , mloc ,
We set J = Ker and denote the maximal ideals of Rgl , Rloc and R
t. The

m, respectively. By Nakayamas lemma, we need to show that dimF J/mJ

module J/mJ appears as the kernel in the sequence
mJ

R/
R/J
0 J/mJ
= Rgl 0.
(5.4)
is similar to the one given by

The argument to bound the dimension of J/mJ
Mazur in [39] to bound the number of relations in presentations of universal deformation rings as quotients of power series rings over O. The idea is to consider the lifting problem associated to the above sequence for the universal lift
gl : GS GLd (Rgl ). The dierence with Mazurs argument is that some lifting
problems do have a solution and one needs to properly interpret this.
F) Ker 2
Formally, we shall construct a homomorphism : Hom(J/mJ,
and show that the kernel of can be interpreted as a subspace of Coker 2,1 . This
will imply the lemma, since then
= dimF Ker() + dimF Im() dimF Coker 2,1 + dimF Ker 2 = t.
dimF J/mJ
F). Then pushout under u of the sequence (5.4) denes
Fix u HomF (J/mJ,
an exact sequence
u
0 Iu Ru Rgl 0,
where Iu
= F. It is not hard to construct a set-theoretic lift u so that the diagram
GS
u
/ GLd (Ru )
gl
'

GLd (Rgl )
GLd (u )
commutes, and so that det u = . (Regarding GLd (Ru ) as the set-theoretic product of diagonal matrices with diagonal entries (ru , 1, 1, . . . , 1) with SLd (Ru ), it
suces to construct a continuous splitting of SLd (Ru ) SLd (Rgl ). This can be
done using the smoothness of SLd .)
The kernel of GLd (u ) is (1 + Md (Iu ), ) and can thus be identied with
ad F Iu
= ad. Via these identications, the set-theoretic lift yields a continuous
2-cocycle
cu Z 2 (GS , ad0 )
given by 1 + cu (g1 , g2 ) = u (g1 , g2 )
u (g2 )1 u (g1 )1 . Its image [cu ] H 2 (GS , ad0 )
is independent of the choice of the set-theoretic lifting. The representation gl can
be lifted to a homomorphism GS GLd (Ru ) precisely if [cu ] = 0. The existence
of homomorphisms Rloc Ru Rgl together with the universality of Rloc imply
that the restrictions [cu|Gv ] H 2 (Gv , ad0 ) are zero for all v . Thus we have
constructed the desired homomorphism
F) Ker 2 , u [cu ].
: Hom(J/mJ,
109
It remains to analyze the kernel of . Let u be in the kernel, so that [cu ] = 0

and gl can be lifted. By the universality of Rgl we obtain a splitting s of Ru Rgl .
Consider the surjective map of mod mO cotangent spaces
ctu : mRu /(m2Ru + mO ) mgl /(m2gl + mO ).
O which induces an isomorphism on
Any surjective homomorphism A B in Ar
mod mO cotangent spaces and which has a splitting is an isomorphism (exercise!).
In our situation, this implies that Iu can be identied with the kernel of ctu .

Rgl by
m)
The map ctu itself is induced from the homomorphism R/(J
pushout and from the analogous map
m
2 + mO ) mgl /(m2gl + mO ).
ct : m/(
m
2 + mO ) mRu /(m2Ru + mO ) is surjective, the induced homomorphBecause m/(
ism u : Ker(ct ) Iu of F-vector spaces is non-zero. Remembering that Iu is
really just another name for F to indicate that it is an ideal in Ru , we have thus
constructed an injective F-linear monomorphism
Ker() HomF (Ker(ct ), F).
(5.5)
By the choice of r and its minimality, it follows that we have Ker(ct ) = Ker(ct )
for the canonical homomorphism
ct : mloc /(m2loc + mO ) mgl /(m2gl + mO ).
Since ct = (2,1 ) , the map (5.5) is the desired homomorphism Ker()
Coker 2,1 . The proof of Lemma 5.2.2 is thus complete.

Remark 5.6.1. For each v S, dene a subspace Lv of H 1 (Gv , ad0 ) by Lv =
H 1 (Gv , ad0 ) for v S
and by Lv = Ker(H 1 (Gv , ad0 ) H 1 (Gv , ad0 ) ) for
v , and denote by HL1 (GS , (ad0 ) ) the corresponding dual Selmer group (cf.
[30, Ch. 4] for a precise denition). It naturally sits in a short exact sequence
0 Coker 2,1 HL1 (GS , (ad0 ) ) Ker 2 0.
In [30, proof of Prop. 4.4], it is proved directly that there is an injective homomorphism
F) HL1 (GS , (ad0 ) ) .
Hom(J/mJ,
This gives an alternative, more conceptual method to derive the desired bound
t.
dim J/mJ
5.7
Exercises
Exercise 5.7.1. Verify all unproven assertions in Section 5.1.
110
Exercise 5.7.2. Check the assertions made about the cocycle cu in the proof of
Lemma 5.2.2: that [cu ] does not depend on the set-theoretic lifting u and that
the class is trivial if and only if u can be chosen to be a homomorphism.
O which
Exercise 5.7.3. Prove that any surjective homomorphism A B in Ar
has a splitting (as O-algebras) and induces an isomorphism ctA ctB on mod mO
cotangent spaces is an isomorphism.
Exercise 5.7.4. Let O be the ring of integers of a nite totally ramied extension
of W (F)[1/p] and let R be an O-algebra which is nite over O. Show that R is
at over O if and only if R is p-torsion free. Hint: Deduce from TorO
1 (R, O/p) = 0
that TorO
(R,
F)
=
0
and
hence
the
assertion.
1
Bibliography
[1] T. Barnet-Lamb, T. Gee, D. Geraghty, R. Taylor, Potential automorphy and
change of weight, preprint, arXiv:1010.2561.
[2] T. Barnet-Lamb, D. Geraghty, M. Harris, R. Taylor, A family of Calabi-Yau
varieties and potential automorphy II, Publ. Res. Inst. Math. Sci. 47 (2011),
no. 1, 2998.
[3] J. Bellache, Pseudodeformations, preprint,
http://www.math.columbia.edu/~jbellaic/pseudo.pdf
[4] J. Bellache, G. Chenevier, Families of Galois representations and higher rank
Selmer groups, Asterisque 324, Soc. Math. France, 2009.
[5] L. Berger, Galois representations and (, )-modules, Course given at the
Galois trimester at IHP, 2010,
http://www.umpa.ens-lyon.fr/~lberger/ihp2010.html
[6] L. Berger, On p-adic Galois representations, Course given at the CRM
Barcelona, June 2010, in this volume.
[7] F. Bleher, T. Chinburg, Universal deformation rings need not be complete
intersections, Math. Ann. 337 (2007), 739767.
[8] G. Bockle, A local-to-global principle for deformations of Galois representations, J. Reine Angew. Math. 509 (1999), 199236.
[9] G. Bockle, Presentations of universal deformation rings, in: Proceedings of
the LMS Durham Symposium, L-functions and Galois Representations, University of Durham, July 1930, 2004.
[10] S. Bosch, Lectures on formal and rigid geometry, Preprintreihe SFB 478,
http://wwwmath.uni-muenster.de/sfb/about/publ/heft378.pdf
[11] C. Breuil, Groupes p-divisibles, groupes nis et modules ltres, Ann. of Math.
152 (2000), 489549.
[12] C. Breuil, A. Mezard, Multiplicites modulaires et representations de GL2 (Zp )
et de Gal(Qp /Qp ) en l = p, Duke Math. J. 115 (2002), no. 2, 205310.
111
112
Bibliography
[13] O. Brinon, B. Conrad, Lecture notes on p-adic Hodge theory, Notes for the
Clay Mathematics Institute 2009 Summer School on Galois Representations,
June 15 July 10, 2009, University of Hawaii at Manoa, Honolulu, Hawaii.
[14] K. Buzzard, Eigenvarieties, in: L-functions and Galois Representations, London Math. Soc. Lecture Note Ser. 320, Cambridge Univ. Press, 2007, 59120.
[15] H. Carayol, Formes modulaires et representations galoisiennes `
a valeurs
dans un anneau local complet, in: p-adic Monodromy and the Birch and
Swinnerton-Dyer Conjecture (Boston, MA, 1991), Contemp. Math. 165,
Amer. Math. Soc., Providence, RI, 1994, 213237.
[16] G. Chenevier, The p-adic analytic space of pseudocharacters of a pronite group and pseudorepresentations over arbitrary rings, preprint,
arXiv:0809.0415.
[17] R. Coleman, B. Mazur, The eigencurve, in: Galois Representations in Arithmetic Algebraic Geometry, London Math. Soc. Lecture Note Ser. 254, Cambridge Univ. Press, 1998.
[18] H. Darmon, F. Diamond, R. Taylor, Fermats last theorem, in: Current Developments in Mathematics, 1995, International Press, Cambridge, MA, 1996,
1154.
[19] A. J. de Jong, Crystalline Dieudonne module theory via formal and rigid ge 82 (1995), 596.
ometry, Publ. Math. IHES
[20] M. Demazure, A. Grothendieck, Schemas en groupes I, II, III, Lecture Notes
in Math. 151153, Springer, Berlin, 1970.
[21] J.-M. Fontaine, G. Laaille, Construction de representations p-adiques, Ann.
Sci. ENS 15 (1982), 547608.
[22] J.-M. Fontaine, B. Mazur, Geometric Galois representations, in: Elliptic
Curves, Modular Forms and Fermats Last Theorem (Hong Kong, 1993),
J. Coates and S.-T. Yau, eds., International Press, Cambridge, MA, 1995,
4178.
[23] F. Gouvea, Deformations of Galois representations, in: Arithmetic Algebraic
Geometry (1999), IAS/Park City Mathematics Institute Lecture Series 7,
Amer. Math. Soc., Providence, RI, 2001, 233406.
ements de geometrie algebrique I, II, III,
[24] A. Grothendieck, J. Dieudonne, El
4, 8, 11, 17, 20, 24, 28, 32 (196167).
IV, Publ. Math. IHES
[25] H. Hida, Galois representations into GL2 (Zp [[X]]) attached to ordinary cusp
forms, Invent. Math. 85 (1986), 545613.
[26] L. Illusie, Grothendiecks existence theorem in formal geometry (with a letter
of J.-P. Serre), in: Fundamental Algebraic Geometry, Math. Surveys Monographs 123, Amer. Math. Soc., Providence, RI, 2005, 179233.
Bibliography
113
[27] N. Katz, B. Mazur, Arithmetic moduli of elliptic curves, Princeton Univ.

Press, Princeton, New Jersey, 1985.
[28] C. Khare, Serres modularity conjecture: the level one case, Duke Math. J.
134 (2006), no. 3, 557589.
[29] C. Khare, J.-P. Wintenberger, Jean-Pierre Serres modularity conjecture I,
Invent. Math. 178 (2009), no. 3, 485504.
[30] C. Khare, J.-P. Wintenberger, Jean-Pierre Serres modularity conjecture II,
Invent. Math. 178 (2009), no. 3, 505586.
[31] M. Kisin, Modularity of 2-dimensional Galois representation, in: Current Developments in Mathematics 2005, 191230, International Press, Somerville,
MA, 2007.
[32] M. Kisin, Crystalline representations and F -crystals, Algebraic Geometry and
Number Theory, Progress in Math. 253, Birkhauser, Boston, 2006, 459496.
[33] M. Kisin, Potentially semi-stable deformation rings, J. Amer. Math. Soc. 21
(2008), 513546.
[34] M. Kisin, Moduli of nite at group schemes and modularity, Ann. of Math.
170 (2009), 10851180.
[35] M. Kisin, Modularity of 2-adic Barsotti-Tate representations, Invent. Math.
178 (2009), no. 3, 587634.
[36] M. Kisin, The Fontaine-Mazur conjecture for GL2 , J. Amer. Math. Soc. 22
(2009), 641690.
[37] M. Kisin, Lecture notes on deformations of Galois representations, Notes for
the Clay Mathematics Institute 2009 Summer School on Galois Representations, June 15 July 10, 2009, University of Hawaii at Manoa, Honolulu,
Hawaii.
[38] H. Matsumura, Commutative Algebra, W. A. Benjamin Inc., New York, 1970.
[39] B. Mazur, Deforming Galois representations, in: Galois Groups over Q,
Y. Ihara, K. Ribet, J.-P. Serre, eds., MSRI Publ. 16 (1987), Springer, New
York, 1989, 385437.
[40] B. Mazur, An introduction to the deformation theory of Galois representations, in: Modular Forms and Fermats Last Theorem (Boston, MA, 1995),
G. Cornell, J. H. Silverman, and G. Stevens, eds., Springer, New York, 1997,
243311.
[41] D. Mumford, Geometric Invariant Theory (3rd ed.), Springer, New York,
1994.
[42] J. Nekov
ar, On p-adic height pairings, Seminaire de Theorie des Nombres,
Paris, 199091, Progress in Math. 108, Birkhauser, Boston, 1993, 127202.
114
Bibliography
[43] J. Neukirch, A. Schmidt, K. Wingberg, Cohomology of Number Fields, Grund.

Math. Wiss. 323, Springer, Berlin, 2000.
[44] L. Nyssen, Pseudo-representations, Math. Ann. 306 (1996), 257283.
[45] V. Pilloni, The study of 2-dimensional p-adic Galois deformations in the = p
case, Lecture notes for the summer school at Luminy, 2007,
http://www.math.columbia.edu/~pilloni/defo.pdf
[46] C. Procesi, The invariant theory of n n matrices, Adv. Math. 19 (1976),
306381.
[47] R. Ramakrishna, On a variation of Mazurs deformation functor, Comp.
Math. 87 (1993), 269286.
[48] M. Raynaud, Schemas en groupes de type (p, . . . , p), Bull. Soc. Math. France
102 (1974), 241280.
[49] R. Rouquier, Caracterisation des caractères et pseudo-caractères, J. Algebra
180 (1996), 571586.
[50] D. Savitt, On a conjecture of Conrad, Diamond, and Taylor, Duke Math. J.
128 (2005), no. 1, 141197.
[51] M. Schlessinger, Functors of Artin rings, Trans. Amer. Math. Soc. 130 (1968),
208222.
[52] S. Shatz, Group schemes, formal groups and p-divisible groups, in: Arithmetic
Geometry (Storrs, Conn., 1984), Springer, New York, 1986, 2978.
[53] J. Tate, p-divisible groups, in: Proceedings of a Conference on Local Fields
(Driebergen, 1966), T. A. Springer, ed., Springer, Berlin, 1967, 158183.
[54] J. Tate, Number theoretic background, Proc. Symp. Pure Math. 33 (1979) no.
3, 326.
[55] J. Tate, Finite at group schemes, in: Modular Forms and Fermats Last
Theorem (Boston, MA, 1995), G. Cornell, J. H. Silverman, and G. Stevens,
eds., Springer, New York, 1997, 121154.
[56] R. Taylor, Galois representations associated to Siegel modular forms of low
weight, Duke Math. J. 63 (1991), 281332.
[57] R. Taylor, Remarks on a conjecture of Fontaine and Mazur, Inst. Math.
Jussieu 1 (2002), no. 1, 125143.
[58] R. Taylor, On the meromorphic continuation of degree two L-functions, Doc.
Math., Extra Volume: John Coates Sixtieth Birthday, 2006, 729779.
[59] R. Taylor, A. Wiles, Ring theoretic properties of certain Hecke algebras, Ann.
of Math. 141 (1995), 553572.
Bibliography
115
[60] W. C. Waterhouse, Introduction to ane group schemes, Graduate Texts in

Math. 66, Springer, New York, 1979.
[61] C. Weibel, An introduction to homological algebra, Cambridge Stud. Adv.
Math. 38, Cambridge Univ. Press, 1994.
[62] T. Weston, Unobstructed modular deformation problems, Amer. J. Math. 126
(2004), 12371252.
[63] A. Wiles, On ordinary -adic representations associated to modular forms,
Invent. Math. 94 (1988), 529573.
[64] A. Wiles, Modular elliptic curves and Fermats last theorem, Ann. of Math.
141 (1995), 443551.
Part II
Arithmetic Aspects of Hilbert

Modular Forms and Varieties
Mladen Dimitrov
Hilbert modular forms and varieties are the natural generalization of elliptic modular forms and curves, when the ground eld of rational numbers is replaced by a
totally real number eld. The aim of these notes is to present the basics of their
arithmetic theory and to describe some of the recent results in the area. A special
emphasis will be put on the following two subjects: images of Galois representations associated to Hilbert modular forms and cohomology of Hilbert modular
varieties with integral coecients.
The exposition follows the lectures given at the CRM Barcelona in June 2010.
We would like to thank the CRM for the excellent conditions for research, as well
as the programme coordinators Luis Dieulefait and Vctor Rotger for their work
and, of course, all the other participants in the lectures.
Hilbert modular forms
Let F be a totally real number eld of degree d > 1 and ring of integers OF , and
denote by JF the set of all embeddings of F into R.
The torus F is quasi-split over Qand its group of characters can be identied with Z[JF ] as follows: for any k = JF k Z[JF ] and for any Q-algebra
A splitting F , we consider the character
x (F Q A) xk =
'
(x)k A .
JF
The norm character NF/ Q : F Q then corresponds to the element

t=
Z[JF ].
JF
119
120
1.1
Arithmetic aspects of Hilbert modular forms and varieties
Congruence subgroups
the pronite completion of Z. The ring A of adeles of F is the product

Denote by Z
with innite adeles F = F Q R. Denote by F + the
of nite adeles Af = F Z
.
open cone of totally positive elements in F
we denote by C U (resp. C + )
For an open compact subgroup U of (OF Z)
U
+
).
the class group A /F U F (resp. the narrow class group A /F U F
For an integral ideal N of OF , we consider the following open compact subgroup of A
f :
$
,
+
,
$$ x 1 N Z
U (N ) = x (OF Z)
and the following open compact
.
a
K0 (N ) =
c
.
a
K1 (N ) =
c
.
a
K11 (N ) =
c
.
a
K(N ) =
c
1.2
subgroups of GL2 (Af ):

/

$
b
,
$$ c N Z
GL2 (OF Z)
d

/
$
b
$
K0 (N ) $ d U (N ) ,
d

/
$
b
$
K1 (N ) $ a U (N ) ,
d

/
$
b
$

K11 (N ) $ b N Z .
d
Hilbert modular forms as automorphic forms on GL2 (A)
The group G+
= F SL2 (F ) acts by linear fractional transformations on the
d-fold product of upper half planes
+

F Q C .
HF = F + 1 1 F
+
the stabilizer of 1
Denote by K
1 in G+
.
Denition 1.1. A weight (k, w0 ) Z[JF ] Z is arithmetic (or cohomological) if,

for all JF , k 2 and k w0 (mod 2).
Denition 1.2. The space Mk,w0 () of classical Hilbert modular forms of arithmetic
+
weight (k, w0 ) and level (a congruence subgroup of GL2 (F ) GL2(Af )G
)
a
b
consists of all holomorphic functions f : HF C such that for every = c d
we have

az + b
f
= det()(w0 tk)/2 j(, z)k f (z),
cz + d
where j(, z) = cz + d (F C) is the usual automorphic cocycle.
The subspace Sk,w0 () of cuspforms consists of those f vanishing at all cusps
of HF .
1. Hilbert modular forms
121
Note that the action of on HF is via the embeddings of F in R, hence it

cannot be decomposed as a product.
The spaces Mk,w0 () and Sk,w0 () are nite dimensional C-vector spaces.
However, as it will become clear, they are not stable under the action of Hecke
operators in general, which motivates the use of the following adelic denition.
Denition 1.3. The space Mk,w0 (K) of (adelic) Hilbert modular forms of weight
(k, w0 ) and level K (an open compact subgroup of GL2 (Af )) consists of all functions f : GL2 (A) C which are left GL2 (F )-invariant, right K-invariant and such
that for all g GL2 (Af ) the function
(w0 tk)/2
j(, 1 1)k f (g)
G+
det()
+
factors through a homolorphic function on G+
/K HF , denoted by fg .
If moreover F \ A f (( 10 x1 ) g) dx = 0 for all g GL2 (A), then f is called a
cuspform, and we denote by Sk,w0 (K) the space of cuspforms.
We say that a form f has central character , a Hecke character of F of
innity type w0 t, if f (y) = (y)f for all y A , and we denote by Sk,w0 (K, )
Sk,w0 (K) the corresponding subspace.
Using the strong approximation theorem for GL2 , one can compare the adelic
and the classical denition as follows. Choose elements gi GL2 (Af ), 1 i h
such that (det(gi ))1ih forms a set of representatives of C +
det(K) . Then the map
f (fgi )1ih induces isomorphisms
#
#
Mk,w0 (K)
Mk,w0 (gi ) and Sk,w0 (K)
Sk,w0 (gi ),
1ih
1ih
where for g GL2 (Af ) we put g = GL2 (F ) gKg
1.3
G+
.
Hecke operators and newforms
The space Mk,w0 (K) admits a left action of the Hecke algebra C c (K\ GL2 (Af )/K)
of bi-K-invariant compactly supported functions on GL2 (Af ). In more concrete
[KgK], corresponding
to the
terms, for every g GL2 (Af ), the Hecke operator
0

characteristic function of KgK, sends f to i f (gi ), where KgK = i gi K. The
subspace Sk,w0 (K) is stable under this action.
For aprime v, let v denote a uniformizer of Fv . The standard Hecke operator
[Kv 10 0v Kv ] is denoted by Tv if Kv is a maximal open compact subgroup of
GL2 (Fv ), and by Uv otherwise.
The Hecke algebra is not commutative in general, and one rather works with
the commutative subalgebra generated by the standard Hecke operators and the
center.
There is a direct sum decomposition
#
Sk,w0 (K, ),
Sk,w0 (K)
122
where runs over all Hecke characters of level K A

f and type w0 t at innity.
The Peterson inner product

0
f1 (g) f2 (g) | det(g)|w
(f1 , f2 ) =
A dg
GL2 (F ) A \GL2 (A)
endows Sk,w0 (K, ) with a structure of a hermitian space with respect to which
the operators Tv are normal. It follows that Sk,w0 (K, ) can be decomposed as
a direct sum of eigenspaces for all the Tv s. Note that while the Uv s form a
commutative family of operators preserving this decomposition, they are not semisimple in general, hence Sk,w0 (K, ) does not always have a basis of eigenforms
for all standard Hecke operators.
The theory of Atkin and Lehner addresses this problem for K = K1 (N ).
More precisely, if one considers the subspace of primitive forms in Sk,w0 (K1 (N ), )
(those orthogonal with respect to the Peterson inner product to all forms coming
from lower level), their theory implies that the standard Hecke operators preserve
this space and are semi-simple. The Strong Multiplicity One Theorem states that
a primitive form f which is an eigenform for Tv , v N , is uniquely determined,
up to a multiple, by its eigenvalues c(f, v) (hence it is necessarily an eigenform for
Uv , v | N , too).
Recall that the Weak Multiplicity One Theorem for GL2 states that an element of Sk,w0 (K1 (N ), ) which is an eigenform for Tv , v N , and for Uv , v | N ,
is uniquely determined, up to a multiple, by its eigenvalues.
A suitably normalized primitive eigenform in Sk,w0 (K1 (N ), ) is called a
newform.
There is a natural bijection between newforms f in Sk,w0 (K1 (N ), ) and
cuspidal automorphic representations of GL2 (A) of conductor N , central character and such that belongs to the holomorphic discrete series of arithmetic
weight (k, w0 ) (see [3]). It is uniquely characterized by the property that, for all
GL (O
)
v N , c(f, v) is the eigenvalue of Tv acting on the new line v 2 F,v .
Galois representations associated to Hilbert

modular forms
The absolute Galois group of a eld L is denoted by GL . Recall that we have an

1 and that the Weil group WF is dened
exact sequence 1 Iv GFv Z
v
as the inverse image of Z. Local class eld theory gives an isomorphism between
Fv and the maximal abelian quotient of WFv , that we normalize so that v is
sent to a geometric Frobenius Frobv .
2.1
Galois representations
Let be a cuspidal automorphic representation of GL2 (A) of conductor N , such

that belongs to the holomorphic discrete series of arithmetic weight (k, w0 ).
2. Galois representations associated to Hilbert modular forms
123
0
The central character of is a Hecke character of weight w0 t, that is, | |w
A
is of nite order. In classical terms, corresponds to a Hilbert modular newform
f over F of level N , weight (k, w0 ) and central character (see 1.3).
For a prime p and an embedding p : Q Qp one can associate to and p
a p-adic representation (cf. [25, 26])
,p : GF GL2 (Qp ),
(2.1)
which is irreducible, totally odd, unramied outside N p and characterized by the

property that for each prime v not dividing N p we have
tr(,p (Frobv )) = p (c(f, v)).
Moreover det ,p = p , where p denotes the p-adic cyclotomic character.
At all places v not dividing p, ,p |WFv is related to v by the local Langlands
correspondence (see [5]).
0
The embedding p denes a partition JF = v JFv , where v runs over the
primes of F dividing p and JFv denotes the set of embeddings of Fv in Qp .
At places v dividing
p, the representation ,p |GFv is known to be de Rham
of HodgeTate weights 12 (w0 k ) + 1, 12 (w0 + k ) J and crystalline for p
Fv
large enough (cf. [1], [2], [16] and [19]).
2.2
Images of Galois representations
The representation ,p is dened over the ring of integers O of a nite extension

E of Qp . Let ,p : GF GL2 (Fp ) be the semi-simplication of the reduction of
,p modulo a uniformizer of O.
The following theorem established in [7, 3] generalizes results of Serre and
Ribet [22] on classical modular forms to the case of Hilbert modular forms.
Theorem 2.1.
(i) For all but nitely many primes p, ,p is irreducible.
(ii) Assume that has no CM. Then, for all but nitely many primes p, the
image of ,p contains (a conjugate of ) SL2 (Fp ).
(iii) Assume that has no CM and that it is not a twist of a base change. Then
for all but nitely many primes p, = ,p fullls the following condition:
(LI) the image of contains SL2 (Fp ) and none of its twists extends to
a representation of GF for any strict subeld F of F .
For the determination of the image of ,p itself, we refer to [20, Appendix B],
where the author adapts the classical theory of inner twists for Hilbert modular
forms.
124
2.3
Modularity lifting theorems
A conjecture of Fontaine and Mazur (see [11] for F = Q) states that any twodimensional, irreducible, totally odd p-adic representation of GF unramied outside a nite set of primes and de Rham at all primes v dividing p, with distinct
HodgeTate weights for each Fv Qp , is automorphic, that is, it can be obtained
as in 2.1.
In the approach initiated by Wiles [30] and TaylorWiles [28], and extended
by Diamond [6] and Fujiwara [12], this conjecture splits naturally in two parts.
The rst is a conjecture of Buzzard, Diamond and Jarvis [4], generalizing Serres
modularity conjecture (now a theorem of Khare and Wintenberger [14, 15]) to
totally real number elds, stating that every two-dimensional, irreducible, totally
odd representation of GF over a nite eld has an automorphic lift. Since there
is no general result in this direction, we consider the following assumption:
(Mod) p is unramied in F and there exists a cuspidal automorphic representation of level prime to p and weight (k, w0 ) such that
w0 = max (k 2), p 1 >
JF
w 0 + k
and ,p .
2
JF
The second part of the conjecture states that, if has an automorphic lift,
then all suitable lifts of are automorphic. Here we quote one such result (see
[12], [17], [23], [24] and [27] for many other modularity lifting theorems):
Theorem 2.2 ([8, Theorem A]). Let : GF GL2 (Fp ) be a continuous representation satisfying (LI) and (Mod). Then all crystalline lifts of of HodgeTate
weights between 0 and p 2 which are unramied outside a nite set of primes are
automorphic.
3
3.1
Hilbert modular varieties

Denition
For an open compact subgroup K of GL2 (Af ), we dene the Hilbert modular
variety of level K as
+
,
YK = GL2 (F )\GL2 (A)/KK
+
where K
= SO2 (F )F
.
We dene the adjoint Hilbert modular variety of level K as
+
YKad = GL2 (F )\GL2 (A)/ A KK
.
By denition there is a natural homomorphism YK YKad and the latter can be

rewritten in terms of the adjoint group PGL2 as follows:
PSO2 (F ),
YKad = PGL2 (F )\PGL2 (A)/K
3. Hilbert modular varieties
125
is the image of K in PGL2 (Af ).

where K
The inclusion of open compact subgroups K K induces natural homomorphisms YK YK and YKad YKad .
3.2
Connected components
By the strong approximation theorem for GL2 , the set 0 (YK ) of connected compoad
nents of YK is isomorphic to the class group C +
det(K) , whereas 0 (YK ) is isomorphic
2
to the quotient of C +
, hence it is a 2-group.
det(K) by the image of A
For any open compact subgroup K K there are exact sequences
1
det(K)

+ 0 (YK ) 0 (YK ) 1,
det(K )(det(K) F F
)
(3.1)
det(K)
1
0 (YKad ) 0 (YKad ) 1.

det(K )(det(K) A2 F )
, then 0 (YK ) is isomorphic to the narrow class group
If det(K) = (OF Z)
2
+
+ 2
ad
of F , while 0 (YK ) is isomorphic to the genus group C +
F / C F C F /(C F )
of F .
We will now express each connected component of YK in more classical terms
+
as a quotient of G+
/K HF (the d-fold product of upper half planes) by a
certain congruence subgroup of the Hilbert modular group.
Choose elements gi GL2 (Af ), 1 i h, such that (det(gi ))1ih forms a
set of representatives of 0 (YK ) C +
det(K) . By the strong approximation theorem
+
for GL2 , the maps i G gi i GL2 (A), 1 i h induce an isomorphism
C+
F
+
gi \G+
/K
1ih
+
GL2 (F )\GL2 (F )gi KG+
/KK = YK ,
(3.2)
1ih
where for g GL2 (Af ) we write g = GL2 (F ) gKg 1 G+

.
Similarly, each connected component of YKad can be dened more classically
using subgroups of the HurwitzMaass extension of the Hilbert modular group
(see [29, Ch. I]). Explicitly,
1
1ih
+
+
ad
gi \G /K
+
ad
GL2 (F )\GL2 (F )gi KG+
/ A KK = YK ,
1ih
1 +
where for g GL2 (Af ) we write ad
G and the gi
g = GL2 (F ) A gKg
GL2 (Af ), 1 i h are chosen so that (det(gi ))1ih is a set of representatives
for 0 (YKad ).
126
3.3
Cusps and compactications
The analytic varieties YK and YKad are quasi-projective, but never projective. The
minimal compactication of YK is dened using (3.2) as
1

YK
gi \ HF P1 (F ) .
1ih
The complement of YK in Y K consists of a nite number of points (the cusps).

Since d > 1, the cusps are always singular points. For a Hilbert modular surface
(d = 2) the resolution of the cusp singularities was found by Hirzebruch (see
[29, Ch. II]). The variety YK has toroidal compactications YK , depending on
some combinatorial data (see [10, 3]). The varieties YK are proper and smooth
at innity (that is to say smooth if YK is smooth). There exists a projection
pr : YK Y K inducing identity on the open YK and such that pr1 ({cusps}) is
a divisor with normal crossings.
3.4
Smoothness
The analytic varieties YK and YKad are smooth if K is suciently small in a sense
that we will now make precise.
Denition 3.1. We say that K is neat if, for all g GL2 (Af ), the quotient of the
gKg 1 F
is torsion free.
group g = GL2 (F ) gKg 1 G+
by its center F
Similarly, we say that A K is neat if, for all g GL2 (Af ), the group (GL2 (F )
is torsion free.
A gKg 1 G+
)/F
Lemma 3.2. Let K K be two open compact subgroups of GL2 (Af ).
(i) The variety YK (resp. YKad ) is an orbifold.
(ii) The variety YK (resp. YKad ) is smooth if and only if K (resp. A K) is neat.
(iii) If K is neat, then K is neat. If A K is neat, then A K and K are neat.
Proof. (i) Recall that YK admits a complex uniformization as in (3.2). For every
+
g GL2 (Af ) the kernel of the action of g on G+
/K is precisely given by
its center F KF . The latter is a subgroup of OF of nite index. It follows

+
that, for every G+
, locally at the point GL2 (F )gKK , YK is a quotient of
+
+
G /K (the d-fold product of upper half planes) by the group
g, =
+ 1
GL2 (F ) gKg 1 K
.
F KF
We will now show that g, is nite. The determinant maps g, to

+
F det(K)F
2 ,
(F KF
)
(3.3)
127
which is nite as a quotient of two nite index subgroups of O

F . Finally, the kernel
+ 1
, which is nite since
of the determinant is generated by SL2 (F ) gKg 1 K
+ 1
SL2 (A) is compact. This
SL2 (F ) SL2 (A) is discrete while gKg 1 K
ad
shows that YK is an orbifold. Since YK is a quotient of YK by the nite group
C KA , it is an orbifold too.
f
(ii) By (i), YK is a manifold if and only if g, is trivial for all g and , which
is equivalent to K being neat (one uses here that a nite-order linear fractional
transformation of HF has a xed point). Similarly, YKad is a manifold if and only if
ad
g, =
+ 1
GL2 (F ) A gKg 1 K
F
is trivial for all g and , which is equivalent to A K being neat.

Note that we have an exact sequence
1 g, ad
g, C KA ,
f
where the last homomorphism is induced from

+ 1
uk A gKg 1 K
u
A
A
=
+ 1
.
A gKg 1 K
A KF
Part (iii) follows from the fact that for all g GL2 (Af ) we have inclusions
GL2 (F ) gK g 1 G+
GL2 (F ) gKg 1 G+
GL2 (F ) A gKg 1 G+
.
+
+
F
F gK g 1 F
F gKg 1 F
The following lemma can be easily deduced from [8, Lemmas 2.1, 2.2] and
shows that K (resp. A K) will be neat if we carefully chose its local component
at one place.
Lemma 3.3. Let u be a prime ideal of F satisfying NF/ Q (u) 1 (mod 4) for
all prime numbers such that [F ( 1) : F ] = 2. Suppose that K = K0 (u) K (u)

(u)
(u)
GL2 (Fu ) GL2 (Af ), where Af denotes the ring of nite adeles outside u.
(i) If the image of the uniformizer u in C U (4) is trivial, then A K is neat.
(ii) If the image of the uniformizer u in the 2-part of C F is trivial, then K is
neat.
3.5 Etale
coverings
Proposition 3.4. Let K K be two open compact subgroups of GL2 (Af ).
(i) If YK is smooth, then YK is smooth and the natural morphism YK YK is
).
etale with group K/K (K F F
128
(ii) If YKad is smooth, then YK is smooth and the natural morphism YK YKad
is etale with group C KA .
f
(iii) If
is smooth, then YKad is smooth and the natural morphism YKad YKad
is etale with group K/K (K A
f ).
YKad
) acts on the bers of the morphism YK

Proof. (i) The group K/K (K F F
YK . Under the assumption that K is neat, we will show that the action is free.
+
on YK , where g GL2 (Af )
Suppose that k K xes the point GL2 (F )gK K
+
1
1
+ 1
. Since K is neat, the
and G , that is to say gkg GL2 (F )gK g K
group g, dened in (3.3) is trivial, hence
+ 1
gKg 1 GL2 (F )K
= K F F
.
It follows immediately that gkg 1 gK g 1 (KF F

) hence k K (KF F
)
as desired.
Alternatively, one can reason componentwise using the fact that for g
). By (3.1), the claim

GL2 (Af ) the fundamental group of g \H is g /(F KF
would follow from the exactness of the following sequence:
K
det(K)
g
f
det

+ 1,

KF )
K (K F F )
det(K )(det(K) F F
)
g (F
where f stands for the projection onto the nite adeles followed by conjugation
by g. Surjectivity is obvious, while injectivity follows from

GL2 (F ) gK g 1 G+
(F KF ) = g (F KF ).
+
Finally, the exactness in the middle is equivalent to det(f (g )) = det(K) F F
.
+
Let k K be such that det(k) F F . Then

1
0 gkg 1 gKg 1 SL (A ) SL (F ) =
SL2 (F ) det(k)
2
f
2
0
1
as an intersection of a dense and an open subset, hence det(k) det(f (g )).

(ii) We already mentioned that the group C KA acts on the bers of the
f
morphism YK YKad and we will now show that this action is free. Suppose that
+
, where g GL2 (Af ) and G+
u A xes the point GL2 (F )gKK
, that is
1
+ 1
to say u GL2 (F )gKg K . Since A K is neat, the proof of Lemma 3.2(ii)
yields
+ 1
= F ,
GL2 (F ) A gKg 1 K
+ 1
+ 1
hence GL2 (F )gKg 1 K
A = F (gKg 1 K
A ) and thus u
F (KF A ) as desired.
ad
ad
(iii) The group K/K (K A
f ) acts on the bers of the morphism YK YK .
The freeness of this action can be established either as in (i), by showing for
example the exactness of
ad
K
det(K)
f
det
g

1,
)(det(K) A2 F )
g ad
K (K A
)
det(K
f
129
or alternatively as follows. Consider the commutative diagram

YK
YK
&
x
&
YKad .
YKad
Since by (i) and (ii) we already know that the other three morphisms are etale, to
show that YKad YKad is etale of group K/K (K A
f ) it is enough to check that
)
* )
* )
*
: C
,
) = K : K (K A
)
C
K : K (K F F

f
K A
KA
f
which is true, since

* )
*
)

K (K A
f ) : K (K F F ) = K Af : (K F F )(K Af )
)
*

= F F
(K A
)
:
F
F
(K
A
)
.
f
f
From now on we will only consider open compact subgroups K which are
neat.
3.6
Integral models
Since YK and YKad are Shimura varieties for the algebraic groups GL2 (F ) and
PGL2 (F ) over Q, they have canonical models over a number eld, which is Q if
for example K = K0 (N ) or K1 (N ).
Since YK and YKad turn out to be (coarse) moduli spaces classifying Hilbert
Blumenthal abelian varieties with some additional structures, Mumfords Geometric Invariant Theory yields integral models which are smooth away from the
discriminant of F and away from primes v where Kv is not maximal.
Finally, the YK s have smooth rational and integral models over the same
base as YK (see [21] for K = K(N ) and [9] for K = K0 (N ), K1 (N ) and K11 (N )).
3.7
Betti cohomology with p-adic coecients
p,
We x a prime p and a p-adic eld E containing the Galois closure of F in Q
and denote by O its ring of integers. We x an embedding of Q in Qp allowing us

to identify GL2 (OF O) with GL2 (O)JF .
For any arithmetic weight (k, w0 ) and any O-algebra A, we consider the
following algebraic representation of GL2 (OF A) GL2 (A)JF :
2
1
Det 2 (w0 k )+1 Symk 2 (A2 ).
(3.4)
LK (k, w0 ; A) =
JF
130

Let LK (k, w0 ; A) be the sheaf of locally constant sections of
+
YK ,
GL2 (F )\(GL2 (A) L(k, w0 ; A)) /KK
&
where the action on L(k, w0 ; A) is via Kp = v|p Kv GL2 (OF Zp ).
We consider Betti cohomology groups H (YK , LK (k, w0 ; O)) and their compactly supported versions Hc (YK , LK (k, w0 ; O)). We will see in 4.1 that under
certain conditions these groups are torsion free.
3.8
Hecke correspondences
Note that, for K K, there is a natural projection pr : YK YK and

pr LK (k, w0 ; A) = LK (k, w0 ; A).
For g GL2 (Af ) we dene the Hecke correspondence [KgK] on YK by the usual
diagram:
g
/ Yg1 KgK
(3.5)
YKgKg1
pr1
YK
pr2
'
YK .
According to [13, 7], if gp M2 (OF Zp ) then [KgK] induces an endomorphism of H (YK , LK (k, w0 ; A)) and of Hc (YK , LK (k, w0 ; A)).
If Kv is maximal, we dene the standard Hecke operators

Tv = [Kv 10 0v Kv ] = [Kv 0v 10 Kv ]

and Sv = [Kv 0v 0v Kv ] = [ 0v 0v Kv ]. For all other v we dene the Hecke
1 0
operator Uv = [Kv 0 v Kv ].
Similarly, we have Betti cohomology groups H (YKad , LK (k, w0 ; A)) and a
Hecke action on them. In particular, if Kv is maximal, there is a Hecke operator
Tv (the operator Sv acts by NF/Q (v)w0 ).
3.9
Poincare duality
In this section we will endow the middle degree cohomology of a Hilbert modular
variety with various pairings coming from Poincare duality.
We consider the dual sheaves LO = L(k, w0 ; O) and L
O = L(k, w0 ; O)
on YK (see (3.4)). The cup product followed by the trace map induces a pairing
2d
[ , ] : Hdc (YK , LO ) Hd (YK , L
O ) Hc (YK , O) O,
(3.6)
which becomes perfect after extending scalars to E. The dual of the Hecke operator [KgK] under this pairing is the Hecke operator [Kg 1 K] (cf. [12, 3.4]).
In particular, the dual of Tv (resp. Sv ) is Tv Sv1 (resp. Sv1 ). We will modify the
pairing (3.6) in a standard way, in order to make it Hecke equivariant.
4. Cohomology of Hilbert modular varieties
131
First, the involution g g = (det g)1 g of GL2 induces a natural isomord

phism Hd (YK , L
O ) H (YK , LO ).
Assume next that K has
levelN , an ideal of OF prime to p, in the sense
that K = K, where = N0 10 . Then LO LO and there is a natural
isomorphism
Hd (YK , LO ) Hd (YK 1 , LO ) = Hd (YK , LO ).
Since for all diagonal g we have g 1 = g 1 , the following diagram commutes:
Hd (YK , L
O)

[Kg 1 K]
Hd (YK , L
O)
/ Hd (YK , LO )

[KK ]
/ Hd (YK 1 , LO )
Hd (YK , LO )
[K (g 1 ) K ]
/ Hd (YK , LO )
[KK ]
/ Hd (YK 1 , LO )
[KgK]
Hd (YK , LO ).
(3.7)
This shows that the modied Poincare pairing

, = [ , ] : Hdc (YK , LO ) Hd (YK , LO ) O
(3.8)
is equivariant for all the standard Hecke operators.

The interior cohomology group Hd! (YK , LO ) is the image of Hdc (YK , LO ) in
d
H (YK , LO ). From the commutativity of the diagram
Hdc (YK , LO ) Hdc (YK , LO )

Hd (YK , LO ) Hdc (YK , LO )
/ Hdc (YK , LO ) Hd (YK , LO )

,
,

/O
and from (3.8) we deduce a Hecke equivariant pairing

, : Hd! (YK , LO ) Hd! (YK , LO ) O .
(3.9)
We will see in 4.1 that under certain conditions this pairing is perfect.
Cohomology of Hilbert modular varieties
&
Let K = v Kv GL2 (Af ) be a neat open compact subgroup such that Kv is
maximal for all primes v dividing p. Fix an arithmetic weight (k, w0 ) and, for every
O-algebra A, put LA = L(k, w0 ; A).
Let : GF GL2 (Fp ) be a continuous representation such that (Mod) and
(LI) hold. Consider the maximal ideal

(Frobv )), Sv det (
(Frobv )) NF/Q (v)1
m = , Tv tr (
of the abstract Hecke algebra T = O[Tv , Sv | Kv maximal, v p].
The Betti cohomology groups H (YK , LO ) dened in 3.7 are T-modules and
we denote by H (YK , LO ) the localization at m.
132
4.1
Freeness results
Under the above assumptions the following theorem is proved in [7, Theorems 4.4,
6.6] and [8, Theorem 2.3] (see [18] for vanishing theorems for the cohomology
without localization).
Theorem 4.1.
rank.
(i) The O-module H (YK , LO ) = Hd (YK , LO ) is free of nite
(ii) Hd (YK , LE/O ) is a divisible O-module of nite corank and the Pontryagin
pairing Hd (YK , LO ) Hd (YK , LE/O ) E/O is a perfect duality.
(iii) The pairing (3.8) yields a perfect duality of free O-modules
, : Hd (YK , LO ) Hd (YK , LO ) O .
Moreover, if K A is neat then (i) and (ii) remain valid if we replace YK by YKad .
4.2
Results on morphisms
Keep the assumptions from the beginning of this section.

Theorem 4.2 ([8, Theorem 2.4]). Suppose given an etale morphism of smooth
Hilbert modular varieties YK YK with group . Assume that is an abelian
p-group and that O is large enough to contain the values of all its characters. Then
Hd (YK , LO ) is a free O[]-module and there is an isomorphism of T-modules
Hd (YK , LO ) O[] O Hd (YK , LO ).
Let v be a prime not dividing p. Assume that Kv is maximal and consider
the degeneracy maps pr1 , pr2 : YKK0 (v) YK used to dene the Hecke correspondence Tv in 3.8. The following theorem generalizes Iharas lemma on the
rst cohomology groups of modular curves to the middle degree cohomology of
Hilbert modular varieties.
Theorem 4.3 ([8, Theorem 3.1]). The T-linear homomorphism
d
pr1 + pr2 : Hd (YK , LO )2
H (YKK0 (v) , LO )
is injective with at cokernel.
Bibliography
[1] D. Blasius and J. Rogawski, Motives for Hilbert modular forms, Invent. Math.
114 (1993), 5587.
[2] C. Breuil, Une remarque sur les representations locales p-adiques et les congruences entre formes modulaires de Hilbert, Bull. Soc. Math. France 127
(1999), 459472.
[3] D. Bump, Automorphic forms and representations, Cambridge Stud. Adv.
Math. 55, Cambridge University Press, Cambridge, 1997.
[4] K. Buzzard, F. Diamond and F. Jarvis, On Serres conjecture for mod Galois
representations over totally real elds, Duke Math. J. 155 (2010), 105161.
[5] H. Carayol, Sur les representations l-adiques associees aux formes modulaires
de Hilbert, Ann. Scient. Ecole

Norm. Sup. 19 (1986), 409468.
[6] F. Diamond, The Taylor-Wiles construction and multiplicity one, Invent.
Math. 128 (1997), 379391.
[7] M. Dimitrov, Galois representations modulo p and cohomology of Hilbert mod
ular varieties, Ann. Sci. Ecole
Norm. Sup. 38, Issue 4 (2005), 505551.
[8]
, On Iharas lemma for Hilbert Modular Varieties, Compos. Math. 145,

Issue 5 (2009), 11141146.
[9]
, Compactications arithmetiques des varietes de Hilbert et formes modulaires de Hilbert pour 1 (c, n), in: Geometric Aspects of Dwork Theory,
A. Adolphson, F. Baldassarri, P. Berthelot, N. Katz, and F. Loeser, eds.,
Walter de Gruyter, 2004, 527554.
[10] M. Dimitrov and J. Tilouine, Varietes et formes modulaires de Hilbert arithmetiques pour 1 (c, n), in: Geometric Aspects of Dwork Theory,
A. Adolphson, F. Baldassarri, P. Berthelot, N. Katz, and F. Loeser, eds.,
Walter de Gruyter, 2004, 555614.
[11] J.-M. Fontaine and B. Mazur, Geometric Galois representations, in: Elliptic Curves, Modular Forms & Fermats Last Theorem (Hong Kong, 1993),
J. Coates and S.-T. Yau, eds., Internat. Press, 1997, 190227.
[12] K. Fujiwara, Deformation rings and Hecke algebras in the totally real case,
arXiv:math/0602606.
133
134
Bibliography
[13] H. Hida, On p-adic Hecke algebras for GL2 over totally real elds, Ann. of
Math. (2) 128 (1988), 295384.
[14] C. Khare and J.-P. Wintenberger, Serres modularity conjecture I, Invent.
Math. 178 (2009), 485504.
[15]
, Serres modularity conjecture II, Invent. Math. 178 (2009), 505586.
[16] M. Kisin, Potentially semi-stable deformation rings, J. Amer. Math. Soc. 21

(2008), 513546.
[17]
, The Fontaine-Mazur conjecture for GL2 , J. Amer. Math. Soc. 22

(2009), 641690.
[18] K-W. Lan and J. Suh, Vanishing theorems for torsion automorphic sheaves
on general PEL-type Shimura varieties, preprint.
[19] T. Liu, Lattices in ltered (, N )-modules, J. Inst. Math. Jussieu 11 (2012),
Issue 3, 659693.
[20] J. Nekov
ar, Level raising and Selmer groups for Hilbert modular forms of
weight two, Canad. J. Math. 64 (2012), Issue 3, 588-668.
[21] M. Rapoport, Compactication de lespace de modules de Hilbert-Blumenthal,
Compos. Math. 36 (1978), 255335.
[22] K. Ribet, On modular representations of Gal(Q/ Q) arising from modular
forms, Invent. Math. 100 (1990), 431476.
[23] C. Skinner and A. Wiles, Residually reducible representations and modular
forms, Inst. Hautes Etudes

Sci. Publ. Math. 89 (1999), 5126.
[24]
, Nearly ordinary deformations of irreducible residual representations,

Ann. Fac. Sci. Toulouse Math. 10 (2001), 185215.
[25] R. Taylor, On Galois representations associated to Hilbert modular forms,

Invent. Math. 98 (1989), 265280.
[26]
, On Galois representations associated to Hilbert modular forms II, in:

Elliptic Curves, Modular Forms & Fermats Last Theorem (Hong Kong,
1993), J. Coates and S.-T. Yau, eds., International Press, 1997, 333340.
[27]
, On the meromorphic continuation of degree two L-functions, Doc.

Math., Extra Volume: John Coates Sixtieth Birthday (2006), 729779.
[28] R. Taylor and A. Wiles, Ring-theoretic properties of certain Hecke algebras,

Ann. of Math. 141 (1995), 553572.
[29] G. van der Geer, Hilbert modular surfaces, Ergeb. Math. Grenzgeb. (3) 16,
Springer-Verlag, 1988.
[30] A. Wiles, Modular elliptic curves and Fermats last theorem, Ann. of
Math. 141 (1995), 443551.
Explicit Methods for Hilbert

Modular Forms
Lassina Dembele and John Voight
Introduction
The study of modular forms remains a dominant theme in modern number theory,
a consequence of their intrinsic appeal as well as their applications to a wide
variety of mathematical problems. This subject has seen dramatic progress during
the past half-century in an environment where both abstract theory and explicit
computation have developed in parallel. Experiments will remain an essential tool
in the years ahead, especially as we turn from classical contexts to less familiar
terrain.
In these notes, we discuss methods for explicitly computing spaces of Hilbert
modular forms, refashioning algorithms over Q to the setting of totally real elds.
Saving denitions for the sections that follow, we state our main result.
Theorem. There exists an algorithm that, given a totally real eld F , a nonzero
ideal N of the ring of integers of F , and a weight k (Z2 )[F :Q] , computes the
space Sk (N) of Hilbert cusp forms of weight k and level N over F as a Hecke
module.
This theorem is the work of the rst author [15] together with Donnelly [18]
combined with work of the second author [69] together with Greenberg [30].
The outline of these notes is as follows. After briey recalling methods for
classical (elliptic) modular forms in 1, we introduce our results for computing
Hilbert modular forms in the simplest case (of parallel weight 2 over a totally real
eld of strict class number 1) in 2. In brief, our methods employ the Jacquet
Langlands correspondence to relate spaces of Hilbert modular forms to spaces of
quaternionic modular forms that are more amenable to computation; we discuss
this matter in 3, and consider two approaches (denite and indenite) in 45.
In 6 we consider several examples in detail. Having thereby established the main
135
136
ideas, we turn to an adelic description of Hilbert modular forms and their quaternionic incarnations in 7, then give a complete and general description of our
algorithms in adelic language in 89.
Although it is our intention to keep these notes as self-contained as possible,
we will assume that the reader has a basic familiarity with classical modular forms
and the methods employed to compute them. The algorithms exhibited below have
been implemented in the computer algebra system Magma [3] and our examples
are computed using this implementation. Donnelly and the second author [21] are
using this implementation to compute Hecke data for thousands of forms over
totally real elds up to degree 6.
These notes arose from lectures at the Centre de Recerca Matem`
atica (CRM)
in Barcelona; it is our great pleasure to thank the CRM for the invitation to
speak and the hospitality of the organizers, Luis Dieulefait and Vctor Rotger. The
authors would also like to thank Matthew Greenberg, Ariel Pacetti, Aurel Page,
Jeroen Sijsling, and the referee for many helpful comments as well as Benedict
Gross for his remarks which we include at the end of Section 6. The rst author is
supported by a Marie Curie Fellowship, and the second author by an NSF Grant
no. DMS-0901971.
Classical (elliptic) modular forms
To motivate our study of Hilbert modular forms, we begin by briey considering

algorithms for classical (elliptic) modular forms. For a more detailed introduction
to modular forms, see the books by Darmon [11] and Diamond and Shurman [19],
and for more information on computational aspects see Cremona [9], Kilford [40],
Stein [64], and the many references contained therein.
Let H = {x + yi C : y > 0} denote the upper half-plane and let H =
H P1 (Q) denote the completed upper half-plane with cusps P1 (Q). The group

GL+
2 (Q) = GL2 (Q) : det > 0
acts on H by linear fractional transformations. For N Z>0 , we consider the
subgroup of those integral matrices of determinant 1 that are upper-triangular
modulo N ,
.

/
a b
+
SL2 (Z) : N | c SL2 (Z) = GL+
0 (N ) = =
2 (Z) GL2 (Q).
c d
The group P0 (N ) = 0 (N )/{1} is a discrete subgroup of PSL2 (R). A modular
form of weight k Z>0 and level N is a holomorphic function f : H C such
that

az + b
(1.1)
f (z) = f
= (cz + d)k f (z)
cz + d
for all 0 (N ) and such that f (z) tends to a nite limit as z tends to any cusp
(i.e., f is holomorphic at the cusps).
1. Classical (elliptic) modular forms
137
One can equivalently write this as follows. For GL2 (R) and z H we
dene j(, z) = cz + d. We then dene a weight k action of GL+
2 (Q) on the space
of complex-valued functions on H by
(f |k )(z) =
(det )k1
f (z).
j(, z)k
(1.2)
Then (1.1) is equivalent to f |k = f for all 0 (N ).

Note that the determinant factor (det )k1 in our denition is dierent from
the usual (det )k/2 , which is an analytic normalization. Consequently, the central
k2
f.
subgroup Q GL+
2 (Q) acts by f |k =
Remark 1.3. For simplicity we treat only the case of 0 (N )-level structure in these
notes. If desired, one could without diculty extend our methods to more general
level structures with characters, and so on.
The C-vector space of modular forms of weight k and level N is nite-dimensional and is denoted by Mk (N ).
If f Mk (N ), then f (z + 1) = f (z), so f has a Fourier expansion
f (z) =
an q n = a0 + a1 q + a2 q 2 + a3 q 3 +
(1.4)
n=0
where an C and q = exp(2iz).

We say that f is a cusp form if f (z) 0 as z tends to any cusp (i.e., f
vanishes at the cusps). The C-vector space of cusp forms of weight k and level N
is denoted by Sk (N ). We have Mk (N ) = Sk (N ) Ek (N ), where Ek (N ) is spanned
by the Eisenstein series of level N . Note that, when k 2, (1.1) is equivalent to

k1
f (z) d(z)
= f (z) (dz)k1 ,
so one may equivalently think of a cusp form f Sk (N ) as a holomorphic dierential (k 1)-form on the modular curve X0 (N ) = 0 (N )\H . (Because of our
normalization, such dierential forms will be global sections of the line bundle corresponding to the algebraic local system Symk2 (C2 ), corresponding to bivariate
homogeneous polynomials of degree k 2. Some authors use an analytic normalization instead.)
The spaces Mk (N ) and Sk (N ) are equipped with an action of pairwise commuting diagonalizable Hecke operators Tn for each integer n Z>0 . The Hecke
operators can be thought of in several dierent ways: they arise from correspondences on the modular curve X0 (N ), as averaging operators over lattices of
index n, or more formally from double coset decompositions for the group 0 (N )
inside SL2 (Z). The action of the Hecke operator Tn is determined by the action
of Tp for p | n, and the latter for p N in weight k are given simply by the formula
1
f
p a=0
p1
(Tp f )(z) = pk1 f (pz) +
z+a
p
138
(For primes p | N one omits the rst term, and there are also operators called
AtkinLehner involutions.) We say therefore that Sk (N ) is a Hecke module, namely,
= Z[Tp ]p =
an abelian group equipped with an action of the Hecke algebra T
Z[T2 , T3 , . . . ], a polynomial ring in countably many variables over Z indexed by
the primes. Our Hecke modules will always be nite-dimensional C-vector spaces.
A form f Sk (N ) is an oldform (at d) if f (z) = g(dz) for some g Sk (M )
with M | N a proper divisor and d | N/M ; we say that f is a newform if f is a
normalized eigenform which is orthogonal to the space of oldforms (with respect
to the Petersson inner product).
The space Sk (N ) consequently has a basis of eigenforms, i.e., functions that
are eigenfunctions for each Hecke operator Tn . If f is an eigenform, normalized
so that a1 = 1 in its q-expansion (1.4), then Tn f = an f . Moreover, the eld
Q({an }) = E C is a number eld and each Hecke eigenvalue an is an algebraic
integer in E.
In this way, the system of Hecke eigenvalues (ap )p for a normalized eigenform
f Sk (N ) determine the form f : H C. These eigenvalues also determine the
L-series
1 '
'

an
1
ap
ap
=
+
L(f, s) =
1
ns
ps
p2s+1k
ps
n=1
pN
p|N
associated to f (dened for Re s > 1), as well as the -adic Galois representations
f, : Gal(Q/Q) GL2 (Z )
associated to f with the property that for any prime p N , we have
Tr(f, (Frobp )) = ap (f ) and det(f, (Frobp )) = pk1 .
Several methods have been proposed for making the Hecke module Sk (N )
into an object amenable to explicit computation. With a view to their generalizations to Hilbert modular forms, we mention two approaches which have seen wide
application. (We neglect the method of graphs [47] and a method which uses the
EichlerSelberg trace formula [35].) For simplicity, we restrict our discussion to
the case of weight k = 2.
The rst method goes by the name modular symbols and has been developed by Birch, Swinnerton-Dyer, Manin, Mazur, Merel, Cremona [9], Stein [64],
and many others. The Hecke operators Tp act naturally on the integral homology H1 (X0 (N ), Z; cusps) linear combinations of paths in the completed upper
half-plane H whose endpoints are cusps and whose images in X0 (N ) are linear
combinations of loops and integration denes a nondegenerate Hecke-equivariant
pairing which gives rise to an isomorphism (the EichlerShimura theorem)
H1 (X0 (N ), C; cusps)
= S2 (N ) S2 (N ),
2. Classical Hilbert modular forms
139
where denotes complex conjugation. The formalism of modular symbols then

presents the space H1 (X0 (N ), Z; cusps) explicitly in terms of paths in H whose
endpoints are cusps (elements of P1 (Q)) and whose images in X0 (N ) are a linear
combination of loops. We have an explicit description of the action of the Hecke
operators on the space of modular symbols, and the Manin trick (the Euclidean algorithm) yields an algorithm for writing an arbitrary modular symbol as a Z-linear
combination of a nite set of generating symbols, thereby recovering S2 (N ) as a
Hecke module.
The second method goes by the name Brandt matrices and goes back to
Brandt, Eichler [23, 24], Kohel [45], Pizer [48], and others. In this approach, a
basis for S2 (N ) is obtained by linear combinations of theta series associated to
(right) ideals in a quaternion order of discriminant N . These theta series are
generating series which encode the number of elements in the ideal with a given
reduced norm, and the Brandt matrices which represent the action of the Hecke
operators are obtained via this combinatorial (counting) data.
Classical Hilbert modular forms
We now consider the situation where the classical modular forms from the previous section are replaced by forms over a totally real eld. References for Hilbert
modular forms include Freitag [25], van der Geer [26] and Goren [29].
Let F be a totally real eld with [F : Q] = n and let ZF be its ring of integers.
The case n = 1 gives F = Q and this was treated in the previous section, so we
assume throughout this section that n > 1. Let v1 , . . . , vn : F R be the real
places of F , and write vi (x) = xi . For M2 (F ) we write i = vi () M2 (R).
For simplicity, in these rst few sections (26) we assume that F has strict
class number 1; the general case, which is more technical, is treated in 7 and
thereafter.
The group
GL+
2 (F ) = { GL2 (F ) : det i > 0 for i = 1, . . . , n}
acts naturally on Hn by coordinatewise linear fractional transformations

a i zi + b i
.
z z = (i zi )i =
ci zi + di i=1,...,n
For a nonzero ideal N ZF , let
.

/
a b
+
GL2 (ZF ) : c N GL+
0 (N) = =
2 (ZF ) GL2 (F ).
c d
+
Let P0 (N) = 0 (N)/Z
F PGL2 (ZF ). Then the image of P0 (N) under
n
the embeddings (i )i is a discrete subgroup of PGL+
2 (R) .
140

Under the assumption that F has strict class number 1, we have
2
Z
F,+ = {x ZF : xi > 0 for all i} = ZF
and hence GL+

2 (ZF ) = ZF SL2 (ZF ), and so alternatively we may identify
.

/
a b
P0 (N) = =
SL2 (ZF ) : c N /{1}
c d
in analogy with the case F = Q.

Denition 2.1. A Hilbert modular form of parallel weight 2 and level N is a holomorphic function f : Hn C such that
4
3'
n
a 1 z1 + b 1
(ci zi + di )2
a n zn + b n
,...,
f (z)
(2.2)
f (z) = f
=
c 1 z1 + d 1
c n zn + d n
det i
i=1
for all 0 (N).
We denote by M2 (N) the space of Hilbert modular forms of parallel weight 2
and level N; it is a nite-dimensional C-vector space. The reader is warned not to
confuse M2 (N) with the ring M2 (R) of 2 2-matrices over a ring R.
Remark 2.3. There is no holomorphy condition at the cusps in Denition 2.1
as there was for classical modular forms. Indeed, under our assumption that
[F : Q] = n > 1, this follows automatically from
principle [26, 1].
Koechers

u 0
GL+
Note also that, if u ZF , then (u) =
2 (ZF ) acts trivially
0 u
on H and at the same time gives a vacuous condition in (2.2), explaining the
appearance of the determinant term which was missing in the classical case.
Analogous to (1.2), we dene
4
3 n
' det i
f (z)
(2.4)
(f | )(z) =
j(i , z)2
i=1
for f : Hn C and GL+
2 (F ); then (3.3) is equivalent to (f | )(z) = f (z) for
all 0 (N).
1
1
n
The group GL+
2 (F ) also acts naturally on the cusps P (F ) P (R) . We
say that f M2 (N) is a cusp form if f (z) 0 whenever z tends to a cusp, and we
denote the space of cusp forms (of parallel weight 2 and level N) by S2 (N). We have
an orthogonal decomposition M2 (N) = S2 (N) E2 (N), where E2 (N) is spanned
by Eisenstein series of level N; for level N = (1), we have dim E2 (1) = # Cl+ ZF ,
where Cl+ ZF denotes the strict class group of ZF .
Hilbert modular forms admit Fourier expansions as follows. For a fractional
ideal b of F , let
b+ = {x b : xi > 0 for i = 1, . . . , n}.
2. Classical Hilbert modular forms
141
Let d be the dierent of F , and let d1 denote the inverse dierent. A Hilbert
modular form f M2 (N) admits a Fourier expansion

a e2i Tr(z)
(2.5)
f (z) = a0 +
(d1 )+
with a0 = 0 if f is a cusp form.

Let f M2 (N) and let n ZF be a nonzero ideal. Then, under our hypothesis that F has strict class number 1, we may write n = d1 for some d+ ;
we then dene an = a . The transformation rule (2.2) implies that an does not
depend on the choice of , and we call an the Fourier coecient of f at n.
The spaces M2 (N) and S2 (N) are also equipped with an action of pairwise
commuting diagonalizable Hecke operators Tn indexed by the nonzero ideals n of
ZF . For example, given a prime p N and a totally positive generator p of p, we
have

z+a
1
f
(Tp f )(z) = N (p)f (pz) +
,
(2.6)
N (p)
p
aFp
where Fp = ZF /p is the residue eld of p; this denition is indeed independent of

the choice of generator p. Using the notation (2.4), we can equivalently write

(Tp f )(z) =
(f | a )(z),
(2.7)
aP1 (Fp )

p 0
1 a
and a =
for a Fp .
0 1
0 p
If f S2 (N) is an eigenform, normalized so that a(1) = 1, then Tn f = an f ,
and each eigenvalue an is an algebraic integer which lies in the number eld E =
Q({an }) C (see Shimura [58, Section 2]) generated by the Fourier coecients
of f . We again have notions of oldforms and newforms, analogously dened (so
that a newform is in particular a normalized eigenform).
Associated to an eigenform f S2 (N) we have an L-function
an
L(f, s) =
N ns
n
where =
and l-adic Galois representations

f,l : Gal(F /F ) GL2 (ZF,l )
for primes l of ZF such that, for any prime p lN, we have
Tr(f,l (Frobp )) = ap (f ) and det(f,l (Frobp )) = N p.
Each of these is determined by the Hecke eigenvalues an of f , so we are again
content to compute S2 (N) as a Hecke module.
We are now ready to state the rst version of our main result.
142
Theorem 2.8 (Dembele [15], GreenbergVoight [30]). There exists an algorithm

which, given a totally real eld F of strict class number 1 and a nonzero ideal
N ZF , computes the space S2 (N) of Hilbert cusp forms of parallel weight 2 and
level N over F as a Hecke module.
In other words, there exists an explicit nite procedure which takes as input
the eld F and the ideal N ZF encoded in bits (in the usual way; see e.g. Cohen
[7]), and outputs a nite set of sequences (ap (f ))p encoding the Hecke eigenvalues
for each cusp form constituent f in S2 (N), where ap (f ) Ef Q. (This algorithm
will produce any nite subsequence in a nite amount of time, but in theory it will
produce the entire sequence if it is left to run forever.) Alternatively, this algorithm
can simply output matrices for the Hecke operators Tp ; one recovers the constituent
forms using linear algebra.
Example 2.9. Let F = Q( 5). Then ZF = Z[w], where w = (1 + 5)/2 satisfying

w2 w 1 = 0. Let N = (3w 14) ZF ; we have N (N) = 229, which is prime.
We compute that dim S2 (N) = 4. There are two Hecke irreducible subspaces of
dimensions 1 and 3, corresponding to newforms f and g (and its Galois conjugates).
We have the following table of eigenvalues, where we write p = (p) for p ZF :
p
(2)
(w + 2)
(3)
(w + 3)
(w 4)
Np
11
11
ap (f )
ap (g)
3
t
t 4t + 1
t + 2t + 2
t 2t 3
3t + 8t + 1
Here, the element t Q satises t3 3t2 t + 1 = 0 and E = Q(t) is an S3 -eld

of discriminant 148.
Recall that, in the method of modular symbols, a cusp form f S2 (N ) corresponds to a holomorphic dierential (1-)form (2i)f (z) dz on X0 (N ) and so, by
the theorem of EichlerShimura, arises naturally in the space H 1 (X0 (N ), C). In a
similar way, a Hilbert cusp form f S2 (N) gives rise to a holomorphic dierential
n-form (2i)n f (z1 , . . . , zn ) dz1 dzn on the Hilbert modular variety X0 (N), the
desingularization of the compact space 0 (N)\(Hn ) , where (Hn ) = Hn P1 (F ).
But now X0 (N) is an algebraic variety of complex dimension n and f arises in
the cohomology group H n (X0 (N), C). Computing with higher-dimensional varieties (and higher degree cohomology groups) is not an easy task! So we seek an
alternative approach.
Langlands functoriality predicts that S2 (N) as a Hecke module occurs in the
cohomology of other modular varieties as well. This functoriality was already
evident by the fact that both modular symbols and their quaternionic variant,
Brandt matrices, can be used to compute the classical space S2 (N ). In our situation, this functoriality is known as the JacquetLanglands correspondence, which
ultimately will allow us to work with varieties of complex dimension 1 or 0 by
considering twisted forms of GL2 over F arising from quaternion algebras. In dimension 1, we will arrive at an algorithm which works in the cohomology of a
3. Quaternionic modular forms
143
Shimura curve, analogous to a modular curve, and thereby give a kind of analogue
of modular symbols; in dimension 0, we generalize Brandt matrices by working
with theta series on (totally denite) quaternion orders.
Quaternionic modular forms
In this section, we dene modular forms on quaternion algebras; our main reference
is Hida [34]. We retain the notation of the previous section; in particular, F is a
totally real eld of degree [F : Q] = n with ring of integers ZF .
A quaternion algebra B over F is a central simple algebra of dimension 4.
Equivalently, a quaternion algebra B over F is an F -algebra generated by elements
i, j satisfying
(3.1)
i2 = a, j 2 = b, and ji = ij

a, b
for some a, b F ; we denote such an algebra by B =
. For more informaF
tion about quaternion algebras, see Vigneras [66].
Let B be a quaternion algebra over F . Then B has a unique involution
: B B, called conjugation, such that xx F for
allx B; we dene the

a, b
as in (3.1) and x =
reduced norm of x to be nrd(x) = xx. For B =
F
u + vi + zj + wij B, we have
x = u (vi + zj + wij) and nrd(x) = u2 av 2 bz 2 + abw2 .
A ZF -lattice of B is a nitely generated ZF -submodule I of B such that
F I = B. An order O of B is a ZF -lattice which is also a subring of B. A maximal
order of B is an order which is not properly contained in any other order. Let
O0 (1) B be a maximal order in B.
A right fractional O-ideal is a ZF -lattice I such that its right order OR (I) =
{x B : xI I} is equal to O; left ideals are dened analogously.
Let K F be a eld containing F . Then BK = B F K is a quaternion
algebra over K, and we say that K splits B if BK
= M2 (K).
Let v be a noncomplex place of F , and let Fv denote the completion of F at v.
Then there is a unique quaternion algebra over Fv which is a division ring, up to
isomorphism. We say that B is unramied (or split) at v if Fv splits B; otherwise
we say that B is ramied at v. The set S of ramied places of B is a nite set of
even cardinality which characterizes B up to isomorphism, and, conversely, given
any such set S there is a quaternion algebra over B ramied exactly at the places
in S. We dene the discriminant D of B to be the ideal of ZF given by the product
of all nite ramied places of B.
Let N ZF be an ideal which is coprime to the discriminant D. Then there
is an isomorphism
N : O0 (1) O0 (1) ZF ZF,N
= M2 (ZF,N ),
144
where ZF,N denotes the completion of ZF at N. Let

O0 (N) = x O0 (1) : N (x) is upper triangular modulo N .
The order O0 (N) is called an Eichler order of level N. We abbreviate O = O0 (N).
We number the real places v1 , . . . , vn of F so that B is split at v1 , . . . , vr and
ramied at vr+1 , . . . , vn , so that
B Q R
= M2 (R)r Hnr ,

1, 1
where H =
is the division ring of Hamiltonians. If B is ramied at
R
all real places (i.e., r = 0) then we say that B is (totally) denite, and otherwise
we say that B is indenite. The arithmetic properties of the algebra B and its
forms are quite dierent according to whether B is denite or indenite, and so
we consider these two cases separately. Using an adelic language, one can treat
them more uniformly (though to some extent this merely repackages the dierence)
we refer to 8 for this approach.
First, suppose that B is indenite, so that r > 0. The case B
= M2 (Q)
corresponds to the classical case of elliptic modular forms; this was treated in 1,
so we assume B
M2 (Q). Let
=
: B M2 (R)r
denote the map corresponding to the split embeddings v1 , . . . , vr . Then the group

= B : det i = (nrd )i > 0 for i = 1, . . . , r

B+
acts on Hr by coordinatewise linear fractional transformations. Let
= O B+
.
O+
Under the assumption that F has strict class number 1, which we maintain, we
have
= Z
O+
F O1 ,
where O1 = { O : nrd() = 1}. Let
+
r
= B
0 (N) = (O+ ) GL2 (R)
+
r
and let P = /Z
F . Then P is a discrete subgroup of PGL2 (R) which can be
identied with
P
= (O1 )/{1} PSL2 (R).
Denition 3.2. Let B be indenite. A quaternionic modular form for B of parallel

weight 2 and level N is a holomorphic function f : Hr C such that
4
3'
r
a 1 z1 + b 1
(ci zi + di )2
a r zr + b r
f (z) = f
,...,
f (z)
(3.3)
=
c 1 z1 + d 1
c r zr + d r
det i
i=1
for all B
0 (N).
3. Quaternionic modular forms
145
Analogously to (2.4), we dene

(f | )(z) = f (z)
r
'
det i
j(i , z)2
i=1
(3.4)
for f : Hr C and B+
; then (3.3) is equivalent to (f | )(z) = f (z) for all
B
0 (N).
We denote by M2B (N) the space of quaternionic modular forms for B of
parallel weight 2 and level N, a nite-dimensional C-vector space.
A quaternionic modular form for B = M2 (F ) is exactly a Hilbert modular
form over F ; our presentation in these three sections has been consciously redundant so as to emphasize this similarity. (We could recover the denition of cusp
forms given in Section 1 if we also impose the condition that the forms vanish at
the cusps.) As we will see later, this similarity is less apparent when the general
and more technical theory is exposited.
The Hecke operators are dened on M2B (N) following their denition in (2.7).
Let p be a prime of ZF with p N, and let p be a totally positive generator of p.
Dene

(p) = O+
\ O+ : nrd()ZF = p = O+
\ O+ : nrd() = p .
The set (p) has cardinality N p + 1. The Hecke operator Tp is then given by

f | (z).
(3.5)
(Tp f )(z) =
(p)
The set (p) admits an explicit description as follows. As above, let Fp =

ZF /p be the residue eld of p, and let p : O M2 (ZF,p ) be a splitting. Then the
set (p) is in bijection with the set of left ideals of O by O. This set of
left ideals is in bijection [41, Lemma 6.2] with the set P1 (Fp ): explicitly, given the
splitting p , the left ideal corresponding to a = (x : y) P1 (Fp ) is

x y
Ja = O1
+ Op.
(3.6)
p
0 0
By strong approximation [66, Theorème III.4.3], each of the ideals Ja is principal,
so Ja = Oa with nrd(a ) = p for all a P1 (Fp ). Therefore, we have
(p) = {a : a P1 (Fp )}.
This denition reduces
to
the one given
in(2.7) for Hilbert modular forms

p 0
1 a
with the choices =
and a =
for a Fp .
0 1
0 p
Having treated Hilbert modular forms in the previous section, now suppose
that B
M2 (F ), or equivalently that B is a division ring. Then a modular form is
=
146
vacuously a cusp form as there are no cusps! We then refer to quaternionic modular
forms equally well as quaternionic cusp forms and let S2B (N) = M2B (N). Here, a
cusp form f gives a holomorphic dierential r-form (2i)r f (z1 , . . . , zr ) dz1 dzr
r
on the associated quaternionic Shimura variety X0B (N) = B
0 (N)\H , a complex
variety of dimension r.
+
The important case for us will be when r = 1. Then B
0 (N) PGL2 (R) acts
on the upper half-plane and the quotient B
0 (N)\H can be given the structure of
a Riemann surface, known as a Shimura curve. In this simple case, a cusp form
2
for B is simply a holomorphic map f : H
C such
that f (z) = (c1 z + d1 ) f (z)
a
b
1
1
for all B
and v1 is the unique split real
0 (N), where 1 = v1 () =
c 1 d1
place of F .
Next, suppose that B is denite, so that r = 0. Recall that O = O0 (N)
O0 (1) is an Eichler order of level N. A right fractional O-ideal is invertible if there
exists a left fractional O-ideal I 1 such that I 1 I = O, or equivalently if I is locally
principal, i.e., for each (nite) prime ideal p of ZF , the ideal Ip is a principal right
Op -ideal. If I is invertible, then necessarily I 1 = {x B : xI O}.
Let I, J be invertible right fractional O-ideals. We say that I and J are in
the same right ideal class (or are isomorphic) if there exists an x B such that
I = xJ, or equivalently if I and J are isomorphic as right O-modules. We write [I]
for the equivalence class of I under this relation and denote the set of invertible
right O-ideal classes by Cl O. The set Cl O is nite and H = # Cl O is independent
of the choice of Eichler order O = O0 (N) of level N.
Denition 3.7. Let B be denite. A quaternionic modular form for B of parallel
weight 2 and level N is a map
f : Cl O0 (N) C.
The space of quaternionic modular forms M2B (N) is obviously a C-vector
space of dimension equal to H.
A modular form for B which is orthogonal to the (1-dimensional) subspace of
constant functions is called a cusp form for B; the space of such forms is denoted
by S2B (N).
The Hecke operators are dened on M2B (N) as follows. Let p be a prime
ideal of ZF with p N. For a right O-ideal I with nrd(I) coprime to p, the Hecke
operator Tp is given by

Tp f [I] =
f [J] ,
(3.8)
JI
nrd(JI 1 )=p
where the sum is taken over all invertible right O-ideals J I such that nrd(J) =
p nrd(I). As in (3.5), this sum is also naturally over P1 (Fp ), indexing the ideals of
norm index p. This denition does not depend on the choice of representative I in
its ideal class and extends by linearity to all of S2B (N).
4. Denite method
147
Consequent to the denitions in the previous paragraphs, we may now consider the Hecke modules of quaternionic cusp forms over F for the dierent quaternion algebras B over F . These spaces are related to each other, and thus to
spaces of Hilbert modular forms, according to their arithmetic invariants by the
JacquetLanglands correspondence as follows.
Theorem 3.9 (EichlerShimizuJacquetLanglands). Let B be a quaternion algebra over F of discriminant D and let N be an ideal coprime to D. Then there is
an injective map of Hecke modules

S2B N S2 DN
whose image consists of those Hilbert cusp forms which are new at all primes p | D.
Proof. See Jacquet and Langlands [36, Ch. XVI], Gelbart and Jacquet [27, 8]
and work of Hida [32]; another useful reference is Hida [33, Proposition 2.12],
who deduces Theorem 3.9 from the representation theoretic results of Jacquet and
Langlands.

Theorem 3.9 yields an isomorphism

S2B N
= S2 N
when the quaternion algebra B has discriminant D = (1). Since a quaternion
algebra must be ramied at an even number of places, when n = [F : Q] is even
we can achieve this for the denite quaternion algebra B which is ramied at
exactly the real places of F (and no nite place). When n is odd, the simplest
choice is to instead take B to be ramied at all but one real place of F (and still
no nite place), and hence B is indenite (and g = 1).
Remark 3.10. Note that in general a space of newforms can be realized as a
Hecke module inside many dierent spaces of quaternionic cusp forms. Indeed, for
any factorization M = DN with D squarefree and N coprime to D, we consider
a quaternion algebra B of discriminant D (ramied at either all or all but one
real place of F ) and thereby realize S2B (N)
= S2 (M)D-new . For example, if p, q
new
of classical newforms can, after
are (rational) primes, then the space S2 (pq)
splitting o old subspaces, be computed using an indenite quaternion algebra
of discriminant 1 or pq (corresponding to a modular curve or a Shimura curve,
respectively) or a denite quaternion algebra of discriminant p or q.
Our main conclusion from this section is that to compute spaces of Hilbert
cusp forms it suces to compute instead spaces of quaternionic cusp forms. The
explicit description of S2B (N) as a Hecke module varies according to whether B is
denite or indenite.
Denite method
In this section, we discuss a method for computing Hilbert modular forms using a
denite quaternion algebra B. We continue with our notation and our assumption
148
that F has strict class number 1. We accordingly call the method in this section the
denite method : it is a generalization of the method of Brandt matrices mentioned
briey in 2 and was rst exhibited by Eichler [22] and Pizer [48] for F = Q, but
the rst explicit algorithm was given by Socrates and Whitehouse [60].
Let I1 , . . . , IH be a set of representative right ideals for Cl O, where H =
# Cl O. As vector spaces, we simply have that
5H
M2B (N) = Map(Cl O, C)
= i=1 C Ii ,
associating to each ideal (class) its characteristic function. Let Oi = OL (Ii ) be the
left order of Ii and let wi = #(Oi /Z
F ).
The action of the Hecke operators is dened by (3.8): we dene the pth-Brandt
matrix T (p) for O to be the matrix whose (i, j)th entry is equal to

(4.1)
b(p)i,j = # J Ij : nrd(JIj1 ) = p and [J] = [Ii ] Z.
The Brandt matrix T (p) is an H H-matrix with integral entries such that the
sum of the entries in each
5 column is equal to N p + 1. The Hecke operator Tp
then acts by T (p) on
i CIi (on the right), identifying an ideal class with its
characteristic function.
The Brandt matrix is just a compact way of writing down the adjacency
matrix of the graph with vertices X = Cl O, where there is a directed edge from Ii
to each ideal class which represents an ideal of index N p in Ii . Indeed, consider the
graph whose vertices are right O-ideals of norm a power of p and draw a directed
edge from I to J if nrd(JI 1 ) = p. Then this graph is a k-regular tree with N p + 1
edges leaving each vertex. The above adjacency matrix is obtained by taking the
quotient of this graph by identifying two ideals if they are in the same ideal class.
Alternatively, we may give an expression for the Brandt matrices in terms
of elements instead of ideals. A containment J Ij of right O-ideals with [Ii ] =
[J] corresponds to an element x JIi1 Ij Ii1 via J = xIi , and we have
nrd(JIj1 ) = p if and only if nrd(x) nrd(Ii )/ nrd(Ij ) = p.
Writing JIi1 = xOi , we see that x is unique up to multiplication on the
right by Oi . We have Oi = (Oi )

1 ZF and (Oi )1 ZF = {1}, so 2ei = #(Oi )1 .
To eliminate the contribution of the factor ZF , we normalize as follows. Let p

be a totally positive generator for p and similarly qi for nrd(Ii ) for i = 1, . . . , H.
Then x Ij Ii1 can be chosen so that nrd(x)(qi /qj ) = p and is unique up to
multiplication by (Oi )
1 . Thus,
.
/
qi
1
# x Ij Ii1 : nrd(x) = p .
(4.2)
b(p)i,j =
2ei
qj
The advantage of the expression (4.2) is that it can be expressed simply in
terms of a quadratic form. Since B is denite, the space B B Q R
= Hn
= R4n
comes equipped with the positive denite quadratic form Tr nrd : B R. If J is
a ZF -lattice, then J
= Z4n embeds as a Euclidean lattice J R4n with respect
4. Denite method
149
to this quadratic form. It follows that one can compute b(p)i,j by computing all
elements x Ii Ij1 such that Tr(qj /qi ) nrd(x) Tr p, a nite set.
Before giving references for the technical details about how the Brandt matrices above are computed explicitly, we pause to give three examples.

1, 23
Example 4.3. Consider the quaternion algebra B =
, so that B is
Q
generated by i, j subject to i2 = 1, j 2 = 23 and ji = ij. We have the maximal
order O = O0 (1) = Z Zi Zk Zik, where k = (j + 1)/2.
We consider

the

0 1
1 0
prime p = 2; we have an embedding O M2 (Z2 ), where i, k
,
1 0
0 0
(mod 2).
We begin by computing the ideal classes in O. We start with C1 = [O]. We
have three ideals of norm 2, namely I(0:1) = 2O + ikO, I(1:1) = 2O + (i + 1)kO,
and I(1:0) = 2O + kO. If one of these ideals is principal, then it is generated by an
element of reduced norm 2. The reduced norm gives a quadratic form
nrd : O Z,
x + yi + zk + wik x2 + xz + y 2 + yw + 6z 2 + 6w2 .
We see immediately that nrd(x + yi + zk + wik) = 2 if and only if z = w = 0 and
x = y = 1. This shows that I(1:1) = (i + 1)O is principal but I(1:0) and I(0:1) are
not. Note also that we nd 2e1 = 4 solutions matching (4.2). We notice, however,
that iI(1:0) = I(0:1) , so we have just a second ideal class C2 = [I(0:1) ].
(4)
Now, of the two ideals contained in I(0:1) of norm 4, we have that I(0:1) =
(4)
4O + i(k + 2)O belongs to C1 whereas I(2:1) = 4O + (2i + 2k + ik)O gives rise to

a new ideal class C2 = [I(2:1) ]. If we continue, each of the two ideals contained in
(4)
I(0:1) of norm 8 belong to C1 , and it follows that H = # Cl O = 3.

From this computation, we have also computed the Brandt matrix
1 1 0
T (2) = 2 1 3 .
0 1 0
Indeed, the rst column encodes the fact that of the three right O-ideals of reduced
norm 2, there is one which is principal and hence belongs to C1 and two that belong
to C2 . We think of this matrix as acting on the right on row vectors.
The characteristic polynomial of T (2) factors as (x3)(x2 +x1). The vector
(1, 1, 1) is an eigenvector with eigenvalue 3 which generates the space of constant
functions and gives rise to the Eisenstein series having eigenvalues ap = p+1 for all
primes p = 23. The space of cusp forms S2B (1) is correspondingly of dimension 2
and is irreducible as a Hecke module. The Hecke module S2B (1) can be explicitly
identied with S2 (23) using theta correspondence. For example, the series

2
2
2
2
q nrd() =
q x +xz+y +yw+6z +6w = 1+4q 2 +4q 4 +4q 8 +8q 10 +
1 (q) =
O
x,y,z,wZ
150
is the q-expansion of a modular form of level 23 and weight 2 and corresponds to

(the characteristic of) C1 . For more details, we refer to Pizer [48, Theorem 2.29],
where the rst computer algorithm for computing Brandt matrices over Q is also
described.
Now we give an example over a quadratic eld.
Example 4.4. Let F = Q( 5). Then ZF = Z[w], where w = (1 + 5)/2 satises

w2 w 1 = 0. The prime 61 splits in F ; let N = (3w + 7)ZF be
one of the
primes
1, 1
above it. We consider the (Hamilton) quaternion algebra B =
over F
F
of discriminant D = (1). We have the maximal order
O0 (1) = ZF iZF kZF ikZF , where k =
(1 + w) + wi + j
,
2
and the Eichler order O O0 (1) of level N given by

O = ZF (3w + 7)iZF (30i + k)ZF (w + 20i + ik)ZF .
The class number of O is H = 3. The following ideals give a set of representatives for Cl O: we take I1 = O,
I2 = 2O + ((w + 2) (2w + 2)i + (1 + 3w)ik)O
= 2ZF (6w + 14)iZF ((w + 1) + (2w + 5)i k)ZF (1 6i + wik)ZF
and I3 = 2O + ((w + 1) + (1 w)i + (2 2w)k)O.
We compute the orders ei = #(Oi /Z
F ) as e1 = 2, e2 = 5 and e3 = 3. For
example, the element u = (2w)i k wik O satises the equation u2 + 1 = 0,
and so yields an element of order 2 in O /Z
F . It follows that none of these orders
are isomorphic (i.e., conjugate) in B.
The rst few Brandt matrices are:
4 0 3
4 5 6
1 5 3
T (2) = 2 0 0 , T ( 5) = 0 1 3 , T (3) = 2 0 3 ,
2 5 0
2 0 2
4 5 1
4 10 6
6 5 6
T (w + 3) = 4 2 0 , T (w 4) = 2 2 3 .
4 0 6
4 5 3
We note that N (w +3) = N (w 4) = 11. For example, the rst column of the
matrix T (2) records the fact that of the 5 = N (2) + 1 right O-ideals of norm (2),
there is exactly one which is principal, two are isomorphic to I2 and the other two
are isomorphic to I3 .
The space S2B (N) of cusp forms is an irreducible 2-dimensional Hecke module, represented by a constituent form f , whose corresponding eigenvector is
(2, 5w, 3w3). In particular, the ring of Hecke operators Tf = Z[ap ] restricted to
4. Denite method
151
f is equal to Tf = Z[w], by coincidence. We have the following table of eigenvalues

for f :
p
(2)
(w + 2)
(3)
(w + 3)
(w 4)
Np
11
11
ap (f )
2w 2
3w + 1
w 2
4w 2
For further discussion of the geometric objects which arise from this computation,
see the discussion in Section 6.
Finally, one interesting example.
Example 4.5. Let F = Q( 15) and let N = (5, 15). Then there exists a cusp
form constituent of an irreducible space of dimension 8 in S2 (N) such that no
single Hecke eigenvalue generates the entire eld E of Hecke eigenvalues.

Egal = Q( 17, i, 2, u)
E = Q( 17, i, u)
Q(i,
Q( 17)
17)
Q(i)

Q( 17, u)

Q( 17, u)
Q( 17)
Here, u = (5 + 17)/2 and Egal is the Galois closure of E. Each Hecke eigenvalue
ap for this form f generates a proper subeld of E. (There are also examples
of this phenomenon over Q, and they are related to the phenomenon of inner
twists; this was analyzed over Q by Koo, Stein, and Wiese [44].)
With these examples in hand, we now give an overview of how these computations are performed; for more detail, see work of the rst author [15]. It is clear
that we need several algorithms to compute Brandt matrices. First, we need a
basic suite of algorithms for working with quaternion orders and ideals; these are
discussed in work of Kirschmer and the second author [41, Section 1], and build on
basic tools for number rings by Cohen [7]. As part of this suite, we need a method
to compute a maximal order, which is covered by work of the second author [70].
Next, we need to compute a set of representatives for Cl O and to test if two right
O-ideals are isomorphic: this is covered by Kirschmer and the second author [41,
Sections 67], including a runtime analysis. To compute a set of representatives,
we use direct enumeration in the tree as in Example 4.3 and a mass formula due
to Eichler as an additional termination criterion. To test for isomorphism, we use
lattice methods to nd short vectors with respect to the quadratic form Tr nrd.
In this method, to compute with level N one must compute a set of representatives for Cl O = Cl O0 (N) anew. The rst author has given an improvement
152
on this basic algorithm, allowing us to work always with ideal classes Cl O0 (1) belonging to the maximal order at the small price of a more complicated description
of the Hecke module. The proof of correctness for this method is best explained
in the adelic language, so we refer to Section 8 for more detail.
Let I1 , . . . , Ih be representatives for Cl O0 (1) and suppose that N is relatively
prime to nrd(Ii ) for each i this is made possible by strong approximation. As
before, let p be a prime of ZF with p D. Let p be a totally positive generator for
p, and let qi be a totally positive generator for nrd(Ii ). By our notation, we have
OL (Ii ) = O0 (1)i . Let i = O0 (1)

i /ZF . For each i, j, consider the set
/
.
:
qi
(p)i,j = j x Ij Ii1 : nrd(x) = p ,
qj
where j acts by the identication O0 (1)

i /ZF = O0 (1)1 /{1}. Via a splitting
isomorphism
N : O0 (1) O0 (1) ZF,N
= M2 (ZF,N ),
1
O0 (1)i ZF,N for
the group O0 (1) acts on P (ZF /N) and, since O0 (1) ZF,N =
/Z
each i (since nrd(Ii ) is prime to N), the group i = O0 (1)
i
F similarly acts on
1
P (ZF /N).
5h
We then dene a Hecke module structure on i=1 C[i \P1 (ZF /N)] via the
map

i (x)
C j \P1 (ZF /N) C i \P1 (ZF /N) , j x
(p)i,j
on each component. It is a nontrivial but nevertheless routine calculation that this

Hecke module is isomorphic to the Hecke module M2B (N) dened by the Brandt
matrices at the beginning of this section.
Example 4.6. We keep the notations of Example 4.4. The quaternion algebra B
has class number 1, thus the maximal order O0 (1) is unique up to conjugation.
Here = O0 (1) /Z
F has cardinality 60. We consider the splitting map

N : O0 (1) O0 (1) ZF ZF /N
= M2 ZF /N
given by
N (i) =
11
37
0
47
, N (k) =
50
18
58
29
, N (ik) =
33
16

28
.
14
(One directly veries that N (ik) = N (i)N (k).) We let act on P1 (ZF /N) on the
left via N . By the above discussion, we have

M2B (N)
= C \P1 ZF /N .
The action of on P1 (ZF /N) has three orbits, which are represented by x1 =
(1 : 0), x2 = (1 : 1) and x3 = (23 : 1), whose stabilizers have cardinality e1 = 2,
5. Indenite method
153
e2 = 5 and e3 = 3. Thus M2B (N) is a free module generated by the orbits x1 ,

x2 and x3 . Writing down the Hecke action in that basis, we obtain the same
Hecke operators as in Example 4.4.
Remark 4.7. The approach presented above has some advantages over the usual
denition of Brandt matrices above. First of all, it is better suited for working with
more general level structures, such as those that do not come from Eichler orders.
For example, adding a character in this context is quite transparent. Secondly,
when working over the same number eld, a substantial amount of the required
data can be precomputed and reused as the level varies, and consequently one
gains signicantly in the eciency of the computation.
Indenite method
In this section, we discuss a method for computing Hilbert modular forms using
an indenite quaternion algebra B with r = 1. We accordingly call our method the
indenite method. The method is due to Greenberg and the second author [30]. We
seek to generalize the method of modular symbols by working with (co)homology.
We continue to suppose that F has strict class number 1, and we assume that
B
M2 (Q) for uniformity of presentation.
=
+
Recall that in this case we have dened a group = B
0 (N) GL2 (R)
+
B
such that P = /ZF PGL2 (R) is discrete; the quotient X = X0 (N) = \H
is a Shimura curve and quaternionic cusp forms on B correspond to holomorphic
dierential 1-forms on X. Integration gives a Hecke-equivariant isomorphism which
is the analogue of the EichlerShimura theorem, namely

S2B (N) S2B (N) H 1 X0B (N), C .

We recover S2B (N) by taking the +-eigenspace for complex conjugation on both
sides. Putting this together with the JacquetLanglands correspondence, we obtain

Dnew
S2 DN
= H 1 (X0B (N), C)+ .
= S2B (N)
We have the identications

B

H X, C = H 1 X0B (N), C
= H 1 B
0 (N), C = Hom 0 (N), C = Hom , C .

1
To complete this description, we must relate the action of the Hecke operators.
Let p DN be prime and let p be a totally positive generator of p. As in (3.5), let

\ O+ : nrd() = p
(5.1)
(p) = O+
and choose representatives a for these orbits labeled by a P1 (Fp ). Then any
permutes the elements of (p) by right multiplication, and hence there is a
unique permutation of P1 (Fp ) such that for all a P1 (Fp ) we have
a = a a
154
with a . For f H 1 (, C) = Hom(, C), we then dene

Tp f () =
f (a ).
(5.2)
a (p)
In a similar way, we compute the action of complex conjugation T via the relation

T f ) ) = f (),
where = and O \ O1 .
We begin with two examples, to illustrate the objects and methods involved.
Then ZF = Z[w] is the ring of integers of F , where

Example 5.3.
Let F = Q( 29).
w = (1 + 29)/2 satises w2 w 7 = 0. Indeed F has strict class number 1 and
u = w + 2 is a fundamental
unit of F .

1, u
Let B =
, so B is generated over F by i, j subject to i2 = 1,
F
is ramied at the prime ideal 2ZF and
j 2 = u, and ji = ij. The algebra B
the nonidentity real place of B (taking 29 29) and no other place. The
identity real place gives an embedding
: B B F R
= M2 (R),

2.27...
u
0
0 1
=
i, j
,
0
0 u
1 0

0
.
2.27...
Let
O = O0 (1) = ZF ZF i ZF j ZF k,
where k = (1 + i)(w + 1 + j)/2. Then O is a maximal order of B with discriminant
D = 2ZF . Let = (O1 ) SL2 (R) be as above and let X = X(1)C = \H be
the associated Shimura curve.
Although they are not an intrinsic part of our algorithm, we mention that
the area of X (normalized so that an ideal triangle has area 1/2) is given by
4
4 3
3
3/2
d
(2)(D)
=
29 F (2)(4 1) = ,
F
F
(2)2n
(2)4
2
&
where (D) = p|D (Np+1), and the genus of X is given by the RiemannHurwitz
formula as

1
eq 1
A = 2g 2 +
,
q
q
A=
where eq is the number of elliptic cycles of order q Z2 in . An explicit formula

for eq given in terms of class numbers and Legendre symbols yields e2 = 3 and
eq = 0 for q > 2. Thus 2g 2 = 0, so g = 1. For more details on these formulas
and further introduction, see work of the second author [68] and the references
given there.
5. Indenite method
155
Next, we compute a fundamental domain for , yielding a presentation for ;

we consider this as a black box for now. The domain, displayed in the unit disc,
is as follows:
We obtain the presentation
= , , 1 , 2 , 3 : 12 = 22 = 32 = [, ]1 2 3 = 1 ,
where
= w 1 + i 2j + k,
= 2 + 2i + (w 1)j (w 1)k,
1 = (2w + 2) + wi + j + 4k,
2 = i,
3 = (w + 1) + (2w + 3)i j k.
The above method gives the isomorphisms of Hecke modules

S2 (2ZF )new
= S2B (1) = f : H C | f (gz) d(gz) = f (z) dz for all g
= H 1 (, C)+ .
We compute that H 1 (, C)+
= Hom(, C)+ = Cf , where f is the characteristic
function of , i.e., f () = 1, f ( ) = 0 and f (i ) = 0 for i = 1, 2, 3.
We compute the Hecke operator Tp for p odd according to the denition given
in (5.2). Let p = (w + 1)ZF . Then N p = p = 5. We compute the action of Tp on
H 1 (, C)+ given by Tp f = ap (f )f . The Hecke operators act as a sum over p + 1
left ideals of reduced norm p, indexed by P1 (F5 ). Let
156

p : O M2 (ZF,p )
= M2 (Z5 ),
0 1
1
0
0
i, j, ij
,
,
1 0
0 1
1

1
0
(mod 5).
Specifying the images modulo p gives them uniquely, as they lift to M2 (Z5 ) using
Hensels lemma. (Note that j 2 = u 1 (mod w + 1).)
Let J , J0 , . . . , J4 be dened by

x y
Ja = J(x:y) = O1
+ pO
p
0 0
as in (3.6). Then Ja = Oa are principal left O-ideals by strong approximation.
For example, J0 = O(i ij) + (w + 2)O = O0 , where 0 = (w + 3) + wi + j + ij.
We compute the Hecke operators as in (5.2). For f : C and , we
compute elements a indexed by a P1 (Fp ) and a permutation of P1 (Fp )
such that a = a a for all a P1 (Fp ); then

Tp f () =
f (a ).
aP1 (Fp )
The contribution to the sum for f simply counts the number of occurrences of in
the product a a = a . Carrying out this computation for various primes,
we obtain the following table:
Np
ap (f )
N p + 1 a(p)
13
23
29
53
59
67
71
83
5 1 6
10
10
13
10
20
55
50
60
80
70
15
30
Here we list only the norm of the prime, as the eigenvalue does not depend on the
choice of prime p of the given norm. This suggests that f corresponds to a base
change of a form from Q.
So we look through tables of elliptic curves over Q whose conductor is divisible
only by 2 and 29. We nd the curve E labelled 1682c1, where 1682 = 2 292 ,
given by
E : y 2 + xy = x3 + x2 51318x 4555676.
Let EFdenote the base change of E to F . We compute that the twist EF of EF
by u 29, given by
EF : y 2 + (w + 1)xy = x3 + (w + 1)x2 + (11w 20)x + (23w + 52),
has conductor 2ZF . Since the extension F/Q is abelian, by base change theorems
we know that there exists a Hilbert cusp form associated to EF over F which is
new of level 2ZF , which therefore must be equal to f . This shows that the Jacobian
J(1) of X(1) is isogenous to EF . We verify that #EF (Fp ) = N p + 1 a(p) and
(as suggested by the table) that E has a 5-torsion point, (1, 2w 5), and
consequently so too does J(1).
5. Indenite method
157
Example 5.4. To illustrate the JacquetLanglands correspondence

in action(The
5)
and
w
= (1 + 5)/2.
orem 3.9), we return to Example 4.4. Recall
that
F
=
Q(

w, (3w + 7)
, which is ramied at N =
We nd the quaternion algebra B =
F
(3w + 7)ZF , a prime of norm 61, and one innite place. The order O = O0 (1)
with ZF -basis 1, i, k, ik, where k = ((w + 1) + wi + j)/2, is maximal. As above, we
compute that A(X) = 10 and g(X) = 2 = dim S2B (1) = dim S2 (N)new . Now we
have the following fundamental domain:
We obtain the presentation

;
<
(1)
= 1 , 1 , 2 , 2 : [1 , 1 ][2 , 2 ] = 1 .
(In particular, (1) is a torsion-free group.) We have dim H 1 (X, C) = 4, and on
the basis of characteristic functions given by 1 , 1 , 2 , 2 the action of complex
conjugation is given by the matrix
1
1
0
0
0 1
0
0
.
0
1
0 1
0 1 1
0
This is computed in a way we thus obtain a basis for H = H 1 (X, C)+
= S2B (1).
Computing Hecke operators as in the previous example, we nd as in Example 4.4
that S2B (1) is irreducible as a Hecke module, and nd that

2
2
5 3
4 1
, H | T5 =
, H | T3 =
,
H | T2 =
2 4
3
4
1 1

6
4
2 1
H | Tw+3 =
, H | Tw4 =
.
4 6
1
1
158
Happily, the characteristic polynomials of these operators agree with those computed using the denite method.
We now give an overview of how these computations are performed: for more
details, see the reference by Greenberg and the second author [30]. To compute
eectively the systems of Hecke eigenvalues in the cohomology of a Shimura curve,
we need several algorithms. First, we need to compute an explicit nite presentation of with a solution to the word problem in , i.e., given , write as an
explicit word in the generators for . Secondly, we need to compute a generator
(with totally positive reduced norm) of a left ideal I O. The rst of these problems is solved by computing a Dirichlet domain; the second is solved using lattice
methods. We discuss each of these in turn.
Let p H have trivial stabilizer p = {g : gp = p} = {1}. The Dirichlet
domain centered at p is the set

D(p) = z H : d(z, p) d(gz, p) for all g ,
where d denotes the hyperbolic distance. In other words, we pick in every -orbit
the closest points z to p. The set D(p) is a closed, connected, hyperbolically convex fundamental domain for whose boundary consists of nitely many geodesic
segments, called sides, and comes equipped with a side pairing, a partition of the
set of sides into pairs s, s with s = g(s) for some g . (We must respect a
convention when a side s is xed by an element of order 2, considering s to be the
union of two sides meeting at the xed point of g.)
The second author has proven [67] that there exists an algorithm which computes a Dirichlet domain D for , a side pairing for D, and a nite presentation for
with a minimal set of generators together with a solution to the word problem
for . This algorithm computes D inside the unit disc D, and we consider now
acting on D by a conformal transformation which maps p 0. We nd then that
D can be computed as an exterior domain, namely, the intersection of the exteriors
of isometric
circles
I(g) for elements g , where I(g) = {z C : |cz + d| = 1}
a b
and g =
SU (1, 1). We nd generators for the group by enumerating
c d
short vectors with respect to a certain positive denite quadratic form depending
on p.
The solution to the word problem comes from a reduction algorithm that
nds elements which contribute to the fundamental domain and removes redundant elements. Given a nite subset G \ {1}, we say that is G-reduced if for
all g G we have d(0, 0) d(g0, 0). We have a straightforward algorithm to
obtain a G-reduced element, which we denote redG (): if d(0, 0) > d(g0, 0)
for some g G, set := g and repeat. When the exterior domain of G is a
fundamental domain, we have redG () = 1 if and only if . This reduction is
analogous to the generalized division algorithm in a polynomial ring over a eld.
5. Indenite method
159
Example 5.5. Let F be the (totally real) cubic subeld of Q(13 ) with discriminant
dF = 169. We have F = Q(b), where b3 + 4b2 + b 1 = 0, and F has strict class
number 1.

1, b
The quaternion algebra B =
has discriminant D = (1) and is
F
ramied at two of the three real places of F . We take O to be an Eichler order of
level p = (b + 2), a prime ideal of norm 5; explicitly, we have
O = ZF (b + 2)iZF
b2 + (b + 4)i + j
b + (b2 + 4) +
ZF
ZF .
2
2
We compute a fundamental domain for the group = 0 (p) = (O1 )/{1}.

We take p = 9/10i H.
(1)
We rst enumerate elements of O by their absolute reduced norm. Of the

rst 260 elements, we nd 29 elements of reduced norm 1.
Let G be the set of elements which contribute to the boundary. For each
g G, we compute redG (g). Each in fact reduces to 1, so we are left with 8
elements.
160
v7
We next enumerate elements in O moving the center in the direction of the

innite vertex v7 . We nd an enveloper, an element g such that v7 lies in the
interior of I(g), and reduce to obtain the following:
The domain now has nite area. We next attempt to pair each vertex to
construct a side pairing.
5. Indenite method
161
I(g1)
I(g7) = I(g11 )
v1
v8
For example, the rst vertex v1 pairs with v8 , pairing the isometric circle
I(g1 ) of g1 with I(g7 ). We continue, but nd that v9 does not pair with another
vertex.
v4
v6
v5
v3
v2
v7
v8
v9
v1
v11
I(g9) = I(g91 )
Indeed, g9 (v9 ) does not lie in the exterior domain. We compute the reduction
redG (g9 ; v9 ), analogously dened. We then obtain a domain with the right area,
so we are done!
162
v7
v9
v8
v6
v5
v4
v3
v2
v10
v11
v14
v12
v17 v1
v16
(In order to get an honest side pairing, we agree to the convention that a
side I(g) which is xed by an element g, necessarily of order 2, is in fact the union
of two sides which meet at the unique xed point of g; the corresponding vertex
then appears along such a side.)
As to the second problem, that of principalizing ideals, we refer to work of
Kirschmer and the second author [41, 4], which relies again upon enumeration in
a lattice. This method can be improved using the computation of a fundamental
domain, which allows one to further eectively reduce lattice elements.
As in the previous section, when the level N varies, by changing the coecient
module one can work always with the group (1) associated to the maximal order
O0 (1) rather than recomputing a fundamental domain for = 0 (N) each time.
This is a simple application of Shapiros lemma: the isomorphism

(1)
H 1 (, C)
= H 1 (1), Coind C
(1)
is an isomorphism of Hecke modules. We give Coind C = Hom((1)/, C) the

(1)
natural structure as a (1)-module, and for a cochain f H 1 (, Coind C), we
6. Examples
dene
163

Tp f () =
f (a )
(5.6)
a (p)
as in (5.2). We still require that (p) O+ ; after enumerating a set of representatives for \(1), we can simply multiply elements a O0 (1) by the representative
for its coset.
To conclude, we compare the above method to the method of modular symbols introduced when F = Q. The Shimura curves X = X0B (N) do not have cusps,
and so the method of modular symbols does not generalize directly. However, the
side pairing of a Dirichlet domain for gives an explicit characterization of the
gluing relations which describe X as a Riemann surface, hence one obtains a complete description for the homology group H1 (X, Z). Paths are now written {v, v}
for v a vertex on a side paired by G. The analogue of the Manin trick in the
context of Shimura curves is played by the solution to the word problem in , and
so in some sense this can be seen as a partial extension of the Euclidean algorithm
to totally real elds. Our point of view is to work dually with cohomology, but
computationally these are equivalent [30, 6].
Examples
In this section we provide some detailed examples to illustrate applications of the

algorithms introduced in the previous sections.
Our rst motivation comes from the association between Hilbert modular
forms and abelian varieties. The EichlerShimura construction (see Knapp
[38,

Ch. XII] or Shimura [59, Ch. 7]) attaches to a classical newform f (z) = n an q n
S2 (N )new an abelian variety Af with several properties. First, Af has dimension
equal to [Ef : Q], where Ef = Q({an }) is the eld of Fourier coecients of f and
EndQ (Af ) Q
= Ef . Second, Af has good reduction for all primes p N . Last,
we have an equality of L-functions

L Af , s =
'

L f , s ,
Hom(Ef ,C)
where f is obtained by letting act on the Fourier coecients of f , so that

L(f , s) =

an
.
ns
n=1
The abelian variety Af arises as the quotient of J0 (N ) by the ideal of End J0 (N )

generated by Tn an .
The simplest example of this construction is the case of a newform f with
rational Fourier coecients Q({an }) = Q. Then Af is an elliptic curve of conductor
164
N obtained analytically via the map

X0 (N ) Af ,
z = 2i
f (z) dz =
i

an 2i
e
.
n
n=1
The equality of L-functions is equivalent to the statement #Af (Fp ) = p+1ap for
all primes p, and this follows in the work of Eichler and Shimura by a comparison
of correspondences in characteristic p with the Frobenius morphism.
The EichlerShimura construction extends to the setting of totally real elds
in the form of the following conjecture (see e.g. Darmon [11, Section 7.4]).
Conjecture 6.1 (EichlerShimura). Let f S2 (N)new be a Hilbert newform of
parallel weight 2 and level N over F . Let Ef = Q({an }) be the eld of Fourier
coecients of f . Then there exists an abelian variety Af of dimension g = [Ef : Q]
dened over F of conductor Ng such that EndF (Af ) Q
= Ef and

L Af , s =
'

L f , s .
Hom(Ef ,C)
In particular, if f has rational Fourier coecients, Conjecture 6.1 predicts

that one can associate to f an elliptic curve Af dened over F with conductor N
such that L(Af , s) = L(f, s).
Conjecture 6.1 is known to be true in many cases. It is known if f appears
in the cohomology of a Shimura curve (as in Sections 3 and 5 above; see also
Zhang [74]): this holds whenever n = [F : Q] is odd or if N is exactly divisible by a
prime p. If f appears in the cohomology of a Shimura curve X and f has rational
Fourier coecients, then when the conjecture holds there is a morphism X Af
and we say that Af is uniformized by X.
However, Conjecture 6.1 is not known in complete generality: by the JacquetLanglands correspondence, one has left exactly those forms f over a eld
F of even degree n and squarefull level (i.e., if p | N then p2 | N), for example
N = (1). One expects still that the conjecture is true in this case: Blasius [1],
for example, shows that the conjecture is true under the hypothesis of the Hodge
conjecture.
The converse to the EichlerShimura construction (Conjecture 6.1) over Q is
known as the ShimuraTaniyama conjecture. An abelian variety A over a number
eld F is of GL2 -type if EndF A Q is a number eld of degree dim A. The
ShimuraTaniyama conjecture states that given an abelian variety A of GL2 -type
over Q, there exists an integer N 1 and a surjective morphism J0 (N ) A.
This conjecture is a theorem, a consequence of the proof of Serres conjecture by
Khare and Wintenberger [39]. In the setting of totally real elds, the analogous
conjecture is as follows.
Conjecture 6.2 (ShimuraTaniyama). Let A be an abelian variety of GL2 -type over
a totally real eld F . Then there exists a Hilbert newform f of parallel weight 2
6. Examples
165
such that Ef
= EndF (A) Q and
L(A, s) =
'
L(f , s).
Hom(Ef ,C)
If both of these conjectures were true, then, by Faltings isogeny theorem,

the abelian variety A in Conjecture 6.2 would be isogenous to the abelian variety
Af constructed in Conjecture 6.1. An abelian variety of GL2 -type over a totally
real eld F which satises the conclusion of the ShimuraTaniyama conjecture
(Conjecture 6.2) is called modular.
There has been tremendous progress in adapting the techniques initiated by
Wiles [71] which led to the proof of the ShimuraTaniyama conjecture in the
case F = Q to totally real elds. For example, there are modularity results
which imply that wide classes of abelian varieties of GL2 -type are modular: see
for example work of Skinner and Wiles [61, 62], Kisin [42, 43], Snowden [63] and
Geraghty [28], and the results below. However, a complete proof of Conjecture 6.2
remains elusive.
With these conjectures
in mind, we turnto some computational examples.
We begin with F = Q( 5) and let w = (1 + 5)/2. For some further discussion

on this case, see work of the rst author [14].
Example 6.3. We consider the spaces S2 (N) with N (N) 30. We compute by
means of the denite method of Section 4, using the quaternion algebra over F
which is ramied at no nite place. We nd that dim S2 (N) = 0 for all ideals N
of ZF with N (N) 30 and dim S2 (N) = 1 for N with norm 31.
Now let N be either of the primes above 31. By the JacquetLanglands
correspondence (Theorem 3.9) and the accompanying discussion, the space S2 (N)
can also be computed using the indenite quaternion algebra B ramied at N and
one real place. We nd agreeably that the Shimura curve X0B (1) has genus 1 and
thus its Jacobian is an elliptic curve. By a search, we nd the elliptic curve
A : y 2 + xy + wy = x3 (1 + w)x2
of conductor N = (5 + 2w) of norm 31.
We now show that A is modular. In brief, we verify that the Galois representation A,3 : Gal(F /F ) GL2 (F3 ) GL2 (F9 ) is irreducible, solvable, and
ordinary since #A(F9 ) = 8 (so 3 a3 (A) = 9 + 1 8 = 2), so A is modular by
Skinner and Wiles [61]. Since A has conductor p and the space S2 (p) has dimension 1, it follows by Faltings isogeny theorem that A is isogenous to the Jacobian
of the Shimura curve X0B (1).
In particular, the elliptic curves over F = Q( 5) with smallest conductor

have prime conductor dividing 31 and are isogenous to A or its Galois conjugate.
Example 6.4. Let N ZF be such that N (N) 100. If N 61, then
dim S2 (N)new 2,
166
since each newform f of conductor N has rational Fourier coecients and is either
the base change of a classical modular form over Q or associated to an elliptic
curve which is uniformized by a Shimura curve.
We compute each space in turn using the denite method. Those forms f
which have the property that ap (f ) = ap (f ), where denotes the Galois involution,
are candidates to arise from base change over Q. For each of these forms f , we
nd a candidate classical modular form g using the tables of Cremona [9]: such
a form will have conductor supported at 5 and the primes dividing N (N). Each
curve comes with a Weierstrass equation, and we nd a quadratic twist of the base
change of this curve to F which has conductor N. Since every elliptic curve over
Q is modular, the base change and its quadratic twist are also modular, and since
we exhaust the space this way we are done.
For example, when N = (8) we have dim S2 (8)new = 1 and nd the form f
with Hecke eigenvalues a(w+2) = 2, a(3) = 2, a(w+3) = a(w+7) = 4, a(w5) =
a(w+4) = 4, and so on. In the tables of Cremona, we locate the form
g(q) = q 2q 3 2q 7 + q 9 4q 11 4q 13 4q 19 + S2 (200)
associated to an elliptic curve Ag with label 200b and equation y 2 = x3 +x2 3x2.
We nd that the quadratic twist of (the base change of) Ag by (w + 2) = 5

given by the equation y 2 = x3 + (w 1)x2 wx has conductor (8), so we conclude
that f is the base change of g. Note that this example is not covered by the known
cases of the EichlerShimura construction, since the level (8) = (2)3 is squarefull.
Example 6.5. Now consider the prime level N = (3w + 7) | 61. Then S2 (N)new is a
2-dimensional irreducible Hecke module arising from the Jacobian of the Shimura
curve X0B (1), where B is the quaternion algebra ramied at one of the innite
places and N. This follows from Examples 4.4 and 5.4.
One is naturally led to ask: can one identify the genus 2 Shimura curve
X = X0B (1) with B of discriminant N?
Consider the hyperelliptic curve C : y 2 + Q(x)y = P (x) over F with
P (x) = wx4 + (w 1)x3 + (5w + 4)x2 + (6w + 4)x + 2w + 1,
Q(x) = x3 + (w 1)x2 + wx + 1.
This curve was obtained via specialization of the BrumerHashimoto family [4,31]
of curves (see also Wilson [73]) whose Jacobian has real multiplication by Q( 5),
so in particular so does the Jacobian J = Jac(C) of C. The discriminant of C is
disc(C) = N2 . One can show that J is modular using the theorem of Skinner and
Wiles [62, Theorem A] and the fact that J has a torsion point of order 31. Since C
is hyperelliptic of genus 2, we compute that the conductor of (the abelian surface)
J is N2 and that its level is N. Since J is modular of level N, it corresponds to
the unique Hecke constituent of that level. It follows then by Faltings isogeny
theorem that J and the Jacobian of the Shimura curve X are isogenous.
6. Examples
167
These examples exhaust all modular abelian varieties of GL2 -type over Q( 5)
with level N of norm N (N) 100 (up to isogeny).
We conclude with some (further) cases where Conjectures 6.1 and 6.2 are not
known.
Example 6.6. The rst case of squarefull level with a form that is not a base change
is level N = (w + 36) = (w + 3)2 so N (N) = 121. We have dim S2 (w + 36) = 1 and
the corresponding newform f has the following eigenvalues (where t2 = 3):
p
(2) (w + 2) (3) (w + 3) (w 4) (w 5) (w + 4) (w + 5) (w 6)
Np
11
11
19
19
29
29
ap (f )
2t
4t
5t
In this case, the EichlerShimura

conjecture predicts the existence of an abelian
with
good reduction away from p = (w + 3) and
surface dened over F = Q( 5)
with real multiplication by Q( 3). The form f is not a base change from Q since
ap (f ) = ap (f ) in general. Can one nd an explicit genus 2 curve over F , analogous
to the previous example, with the L-function of its Jacobian given by the above
Frobenius data?
Example 6.7. We next consider a situation when we are successful in establishing
the correspondence in a case which is not covered by known results. Consider
N = (17w 8) = (w + 4)2 , so N (N) = 361. There exist forms f, g S2 (N) with
Hecke eigenvalues as follows:
p
Np
(2) (w + 2) (3) (w + 3) (w 4) (w 5) (w + 4) (w + 5) (w 6)
4
11
11
19
19
29
29
ap (f )
ap (g)
The form g is a quadratic twist of f by w(w + 4), and the forms f, g are not base
changes from Q.
We use the method of Cremona and Lingham [10] without an attempt to
be exhaustive to nd an elliptic curve E over F with good reduction away from
p = (w + 4). We nd the curve
E : y 2 + (w + 1)y = x3 + wx2 x w
(which could also be found by a naive search) with conductor N. We verify that
#E(Fp ) = N p + 1 ap (f ) for all primes p up to the limit of the computation. We
prove that E ?
is modular using the fact that E has a 3-isogeny, with kernel dened
over F = F ( w(w + 4)) generated by the points with x = w 1, so we can apply
the theorem of Skinner and Wiles. Therefore E corresponds indeed to f .
We now turn to the existence of elliptic curves with everywhere good reduction over totally real elds. The ShimuraTaniyama conjecture predicts that such
curves arise from cusp forms of level (1).
168
Example 6.8. Let F = Q() = Q(13 )+ be the totally real subeld of the cyclo1
. The space S2 (1) over F has dimension 1,
tomic eld Q(13 ) with = 13 + 13
represented by the cusp form f . Conjecture 6.1 predicts the existence of an elliptic curve Af over F with everywhere good reduction. We note that Af must be
a Q-curve since the twist f of f by an element Gal(F/Q) is again a form
of level 1 and so f = f (since dim S2 (1) = 1), and consequently Af must be
isogenous to its Galois conjugates.
Indeed, let
A : y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6
be the elliptic curve with the following coecients:
a1 = 2 + + 1,
a2 = 3 + 2 + ,
a3 = 5 + 3 + ,
a4 = 5 4 113 2 + 3,
a6 = 135 314 483 + 522 + 33 10.
We verify that A has everywhere good reduction and that #A(F )tors = 19 with
A(F )tors generated by the point
5

4 43 + 32 + 2 1, 5 + 4 + 2 .
Therefore, by Skinner and Wiles [61, Theorem 5] applied to the 3-adic representation attached to A or [62, Theorem A] to the 19-adic one, we conclude that A
is modular, and thus A is our abelian variety Af .
The elliptic curve A was obtained from a curve over Q( 13) F as follows.
There is a 2-dimensional Hecke constituent g in the space of classical modular
forms S2 (169)new over Q given by
g(q) = q + 3q 2 + 2q 3 + q 4 3q 5 + .
dened over Q. Roberts and WashingLet Ag be the associated abelian surface,
ton [49] showed that Ag twisted by 13 is isomorphic over F to E E , where

3483 13 12285
E : y 2 = x3 +
+ 74844 13 270270 ,
2
a Q-curve of conductor (13) and is the nontrivial automorphism of Q( 13).

The abelian variety Ag is related to the elliptic curve A as follows. Roberts
and Washington show that J1 (13) and Ag are isogenous over Q. The abelian
surface J1 (13) is isogenous over F to A A, since J1 (13) obtains good reduction
over F (see Mazur and Wiles [46, Chapter 3.2, Proposition 2] and Schoof [52]).
6. Examples
169
point of order
Indeed, Serre showed that the abelian surface J1 (13) has a rational
19 and Mazur and Tate showed that J1 (13) twisted by Q( 13) is a product of
two elliptic curves. We thereby determine an equation for A by showing that it
is a quadratic twist of E by some unit in F . We would like to thank Elkies and
Watkins for pointing us to the rich history of this curve.
The next example gives the list of all modular elliptic curves (up to isogeny)
with everywhere good reduction over a real quadratic eld F of discriminant
1000 and having strict class number 1.
Example 6.9. Let 0 D 1000 be a fundamental discriminant such that
the
eld
F = Q( D) has strict class number 1. We have ZF = Z[w] with w = D or
(1 + D)/2, according to whether D 0, 1 (mod 4). Let A be a modular elliptic
curve over F with everywhere good reduction. Then A corresponds to a cusp form
f of level 1 with rational Fourier coecients. Computing the spaces S2 (1), we
nd that either A is a Q-curve (so A is isogenous to its Galois conjugate), or
D {509, 853, 929, 997} and the isogeny class of A is represented by the curve
with coecients as below, up to Galois conjugacy:
D
a1
a2
a3
509
2 + 2w
853
1 w
929
1w
997
1 + 20w
1
a4
a6
162 + 3w
71 + 34w
5921 + 429w
145066 9437w
1738 82w
11808 201w
289 24w
2334 144w
The Q-curves over F are listed in Cremona [8]. For example, the curve
y 2 + xy + uy = x3
with u = (5 + 29)/2 is the smallest such and was discovered by Tate [54].
The curve for discriminant D = 509 was discovered by Pinch [52, p. 415]. It
was the rst example of an elliptic curve having everywhere good reduction which
is not a Q-curve, and it was proven to be modular by Socrates and Whitehouse [60]
using the method of Faltings and Serre. The curve for D = 853 was discovered
by Cremona and Watkins independently; the one for D = 929 by Elkies; and the
one for D = 997 by Cremona. Their modularity can be established by applying the theorem of Skinner and Wiles [61, Theorem 5] to the associated 3-adic
representation.
As a third and nal application, we consider the existence of number elds
with small ramication these are linked to Hilbert modular forms via their
associated Galois representations.
In the late 1990s, Gross proposed the following conjecture.
Conjecture 6.10. For every prime p, there exists a nonsolvable Galois number eld
K ramied only at p.
170
Conjecture 6.10 is known to be true for p 11 by an argument due to Serre

and Swinnerton-Dyer [53]. For k = 12, 16, 18, 20, 22 or 26, let k Sk (1) be the
unique newform of level 1 and weight k. For each such k and every prime p, there
is an associated Galois representation k,p : Gal(Q/Q) GL2 (Fp ) ramied only
at p. In particular, the eld K xed by the kernel of k,p is a number eld ramied
only at p with Galois group Gal(K/Q) = img k,p .
We say that the prime p is exceptional for k if k is not surjective. The
open image theorem of Serre [55] implies that the set of primes where k is
exceptional is nite. Serre and Swinnerton-Dyer produced the following table of
exceptional primes:
k
Exceptional p
12
2, 3, 5, 7, 23, 691
16
2, 3, 5, 7, 11, 31, 59, 3617
18
2, 3, 5, 7, 11, 13, 43867
20
2, 3, 5, 7, 11, 13, 283, 617
22
2, 3, 5, 7, 13, 17, 131, 593
26
2, 3, 5, 7, 11, 17, 19, 657931
(The large primes that occur are divisors of the numerators of Bernoulli numbers.)
It follows that Conjecture 6.10 is true for every p 11.
Conversely, it is a consequence of the proof of Serres conjecture by Khare
and Wintenberger [39] together with standard level lowering arguments that if
p 7, then any odd representation of the absolute Galois group Gal(Q/Q) to
GL2 (Fp ) which is ramied only at p is necessarily reducible (and thus solvable).
To address Gross conjecture, we instead look at the residual representations
associated to Hilbert cusp forms of parallel weight 2 and level 1 over elds F
ramied only at a prime p 7. This idea was rst pursued by the rst author [16]
following a suggestion of Gross, who found such a form with F = Q(32 )+ , the
totally real subeld of the cyclotomic eld Q(32 ), settling the case p = 2. This
line of inquiry was followed further by the authors with Matthew Greenberg [17],
settling the cases p = 3, 5. (The case p = 7 was recently settled by Dieulefait [20]
by considering the mod 7 Galois representation attached to a genus 2 Siegel cusp
form of level 1 and weight 28. Therefore Gross conjecture is now a theorem.)
We sketch below the resolution of the case p = 5. We take the base eld
F = Q(b) to be the subeld of Q(25 ) of degree 5, where b5 +5b4 25b2 25b5 = 0.
The eld F has strict class number 1. Let E be the elliptic curve over F with
j-invariant given by
(5 711 )j =
16863524372777476 b4 88540369937983588 b3 + 11247914660553215 b2
+ 464399360515483572 b + 353505866738383680
6. Examples
171
and minimal conductor N. Then N = p5 p7 , where p5 = ((2b3 12b2 +31b+25)/7)

is the unique prime above 5 and p7 = ((2b4 9b3 + 8b2 + 53b + 6)/7) is one of the
ve primes above 7. Roberts [50] showed that the mod 5 Galois representation

E,5 : Gal F /F End E[5]
= GL2 Fp5 = GL2 F5
has projective image PGL2 (F5 ) and is ramied only at the prime p5 (and not p7 ).
(We will verify below that E has indeed the right conductor.) The representation
E,5 and its Galois conjugates give an extension K of Q with Galois group
5

Gal K/Q
= PSL2 F5 Z/10Z.
By work of Skinner and Wiles (as in Example 6.3), we prove that the 3-adic
representation E,3 associated to E is modular, and hence E itself is modular. Since
E is modular and F has odd degree, E is uniformized by a Shimura curve. Namely,
let B be the quaternion algebra over F ramied at four of the ve real places.
Let X0B (p5 p7 ) be the Shimura curve associated to an Eichler order of level p5 p7
contained in a maximal order O, and let J0B (p5 p7 ) be the Jacobian of X0B (p5 p7 ).
We compute that J0B (p5 p7 )new has dimension 203. We then nd the unique Hilbert
newform fE of parallel weight 2 and level p5 p7 with integer Fourier coecients
which corresponds to E.
The elliptic curve E found by Roberts [50] was obtained from our computations of Hilbert modular forms at level p5 over F . The space S2 (p5 )new has
two Hecke constituents of dimension 10 and 20, respectively. Let f be a newform
in the 20-dimensional constituent. Let Tf be the restriction of the Hecke algebra
T = Z[Tp ]p to this constituent: this is the constituent which yields the Galois representation obtained in the 5-torsion of Roberts curve. Let Ef = Tf Q = Q(ap (f ))
be the eld of Fourier coecients of f and let = Aut(Ef ). By direct calculations, we see that Kf = Ef is the (totally real) quartic eld generated by a root
of the polynomial x4 + 2x3 75x2 112x + 816.
The Galois group Gal(F/Q) acts on T via its action on ideals of ZF , namely
(Tp ) = T(p) . This action preserves the decomposition of T into Hecke-irreducible
components. By work of Shimura, the action of on Fourier expansions preserves
the Hecke constituents. In particular, both these actions preserve Tf and hence
Kf and must be compatible. Therefore, for each Gal(F/Q), there is a unique
= () such that, for all prime ideals p ZF , we have

a(p) (f ) = ap (f ) .
The map thus yields a homomorphism

Gal F/Q ,
().
(6.11)
By direct calculation, we show that this map is an isomorphism. Since Gal(F/Q)

is abelian, the eld Ef must be a ray class eld over Kf ; in fact, we compute that
172
it has conductor equal to a prime above 71 which splits completely in Ef . The

ideal 5ZKf factors as 5ZKf = P2 P . The prime P splits completely in Ef , and
the primes above it in Ef are permuted by Gal(Ef /Kf ) = Aut(Ef ); the prime P
is inert.
We have a Galois representation

f : Gal F /F GL2 Tf F5
with Tr(f (Frobp )) ap (f ) (mod 5) and det(f (Frobp )) N p (mod 5) for all
(i)
p 5. Let mf , i = 1, . . . , 5 be the ve maximal ideals above (the image of) P.

They give rise to the ve residual representations

(i)
f : Gal F /F GL2 F5 .
(i)
Let Li be the xed eld of ker(f ), and L the compositum of the Li . Then L is a
Galois extension of F and is ramied only at 5. Since F is Galois and ramied only
at 5, and by the above Gal(F/Q) permutes the elds Li so preserves L, we see that
L is a Galois extension of Q ramied only at 5. By a result of Shepherd-Barron
and Taylor [56], each extension Li can be realized in the 5-torsion of an elliptic
curve Ei /F .
Recall that the projective representation P E,5 from Roberts elliptic curve E
is surjective, ramied at p5 but unramied at p7 . Therefore, the level p5 p7 is a
(i)
nonoptimal level for E,5 ; thus, by Mazurs Principle [37], we have E,5
= f
for some 1 i 5. In other words, the extension constructed by Roberts is
isomorphic to our eld L.
Roberts [50] has given an explicit equation for the number eld L (obtained
from the 5-division polynomial of the elliptic curve E): the eld L is the splitting
eld of the polynomial
x25 25x22 + 25x21 + 110x20 625x19 + 1250x18 3625x17
+ 21750x16 57200x15 + 112500x14 240625x13 + 448125x12
1126250x11 + 1744825x10 1006875x9 705000x8 + 4269125x7
3551000x6 + 949625x5 792500x4 + 1303750x3 899750x2
+ 291625x 36535.
To conclude, we consider a question that touches on each of the above three
subjects. We reconsider Gross conjecture (Conjecture 6.10) in the case p = 2 [16].
The nonsolvable eld which is ramied only at 2 arises from the Galois representation associated to a constituent eigenform f S2 (1) in a 16-dimensional
subspace of the space of Hilbert cusp forms of parallel weight 2 and level 1 over
F = Q(32 )+ . Let Ef = Q(ap ) be the eld of Fourier coecients of f and let
= Aut(Ef ). Let Kf be the xed eld of , so that Gal(Ef /Kf ) = . The
map (6.11) in this context is again an isomorphism, and so Ef is abelian over
7. Adelic quaternionic modular forms
173
Kf and 8= [F : Q] divides [Ef : Q] = 16. By direct calculations, we show that

Kf = Q( 5).
The EichlerShimura construction (Conjecture 6.2) predicts the existence
of a 16-dimensional abelian variety Af dened over F with everywhere good reduction and real multiplication by Ef associated to f . More should be true, as
communicated to us by Gross (private communication). In fact, Af should descend
to an abelian variety of dimension 16 over Q, and we should have
L(Af , s) = L(f, s) L(f , s),
where Gal(Ef /Q) is any element that restricts to the nontrivial element of
Kf = Q( 5). The endomorphisms of Af over Q should be the ring of integers
of Kf , and over Q, the variety Af would have bad reduction only at the prime 2;
the nonsolvable extension would then arise as its 2-division eld. The conductor
N = 2124 = d4 of Af over Q can be computed from the functional equation of
L(f, s), where we note that d = 231 is the discriminant of F . We note that Af
is not of GL2 -type over Q itself as it would be modular by the proof of Serres
conjecture.
Although one knows that the associated Galois representation exists by work
of Taylor [65], as Gross says: With such nice properties, its a shame that we
cant even prove that the abelian variety Af exists! Thats an advantage you have
when F has odd degree.
Adelic quaternionic modular forms
In this section, we begin again and revisit the denition of Hilbert and quaternionic
modular forms, allowing F to have arbitrary class number; we quote Hida [34] as
a reference for this section.
We renew our notation. Let F be a totally real eld of degree n = [F : Q]
with ring of integers ZF . Let B be a quaternion algebra over F of discriminant
D. Let v1 , . . . , vn be the real places of F (abbreviating as before xi = vi (x) for
x F ), and suppose that B is split at v1 , . . . , vr and ramied at vr+1 , . . . , vn , i.e.,
B B = B Q R M2 (R)r Hnr .
(7.1)
Let i denote the ith projection B M2 (R) and = (1 , . . . , r ). Let

F+ = x F : xi > 0 for all i
be the group of totally positive elements of F and let Z

F,+ = ZF F+ .
Let O0 (1) B be a maximal order. Let N be an ideal of ZF coprime to D
and let O = O0 (N) O0 (1) be an Eichler order of level N.
With a view towards generalizations, rather than viewing modular forms as
functions on (a Cartesian power of) the upper half-plane which transform in a
174
. Let
certain way, we now view them instead more canonically as functions on B
H = C \ R be the union of the upper and lower half-planes. Via the embeddings
(7.1), the group

B
acts on (H )r
v1 , . . . , vr , corresponding to the rst r factors in
on the right transitively with the stabilizer of ( 1, . . . , 1 ) Hr being

r nr
K = R SO2 (R) H
.
Therefore we can identify
r
B
/K H ,
g z = g
1, . . . ,

1 .
(7.2)
From this perspective, it is natural to consider the other (nonarchimedean)

places of F at the same time. Let
= lim Z/nZ =
Z
'
Zp

(where denotes the restricted direct product) and let denote tensor with Z

B
over Z. We will dene modular forms on B as analytic functions on B
in a specied
which are invariant on the left by B and transform by K O
way.
We must rst dene the codomain of these functions to obtain forms of
arbitrary weight. Let k = (k1 , . . . , kn ) (Z2 )n and suppose that the ki have the
same parity; we call k a weight. Let

k0 = max ki , mi = k0 ki /2, and wi = ki 2.
i
For an integer w 0, let Pw = Pw (C) be the subspace of C[x, y] consisting

of homogeneous polynomials of degree w. For GL2 (C), let be the adjoint
of , so that = = det . Dene a right action of GL2 (C) on Pw (C) by

(q )(x, y) = q (x y)
= q dx cy, bx + ay

a b
for =
GL2 (C) and q Pw (C). For m Z, GL2 (C) also acts on
c d
Pw (C) via the character (det )m . By twisting the above action by this
character, we get a right GL2 (C)-module denoted by Pw (m)(C). Dene the right
GL2 (C)nr -module

Wk (C) = W (C) = Pwr+1 mr+1 (C) Pwn (mn )(C).
(7.3)
(By convention, if r = n then we set Wk (C) = C.) For the ramied real places
vr+1 , . . . , vn of F , we choose splittings
i : B B F C
= M2 (C).
175
We abbreviate as above i = i () for B. Then Wk (C) becomes a right

B -module via (r+1 , . . . , n ) GL2 (C)nr . From now on, Wk (C) will be
endowed with this action, which we denote by x x for x Wk (C) and B .
One may identify Wk (C) with the subspace of the algebra C[xr+1 , yr+1 , . . . , xn , yn ]
consisting of those polynomials q which are homogeneous in (xi , yi ) of degree wi
but with a twisted action.
Wk (C), with a right action

B
We consider the space of functions : B

of K B dened by
3 r
4

' j i , 1 ki

1

|k (, ) (g,
) =
1 ,
(7.4)

mi +ki 1 g ,
i=1 det i
a i bi
where we recall that j(i , 1 ) = ci 1+di C if i =
. (The presence
c i di
of the inverses is forced as we want a right action on functions via multiplication of
,
B
the argument on the right. This almost extends to a right action of B
except that j(gh, z) = j(g, hz)j(h, z) = j(g, z)j(h, z) unless h xes z.)
Denition 7.5. A (quaternionic) modular form of weight k and level N for B is an
analytic function
Wk (C)
: B
B
we have:
such that for all (g,
) B
B

;
(i) |k (, u
) (g,
) = (g,
) for all K and u
O
(ii) (g,
) = (g,
) for all B .
In other words, Denition 7.5 says that a quaternionic modular form of
weight k and level N is an analytic function which is B -invariant on the left
)-equivariant on the right under the action (7.4). In particular,
and (K O

) g,
u
= g,

(7.6)
g,
u
= |k (1, u
and
for all u
O
(g,
) = |k (, 1) (g,
) =
ki 4
r
'
j i , 1
) .

mi +ki 1 (g,
det
i
i=1
(7.7)
We denote by MkB (N) the space of quaternionic modular forms of weight k

and level N for B.
/O
,
) (H )r B
Let be a quaternionic
O
modular
form. For (z,
such that g( 1, . . . , 1 ) = z and dene

choose g B
/O
Wk (C),
f : (H )r B
3 r
4

' det gi mi +ki 1

1

(g,
)g .
f z,
O =
ki
i=1 j gi , 1
(7.8)
176
The map f in (7.8) is well-dened by (7.6) and (7.7): we have

3 r
4
4
3 r

' det gi i mi +ki 1
' det gi mi +ki 1

(g)1

g1
=
g,
u

g,

ki
ki
i=1 j gi i , 1
i=1 j gi , 1
using the fact that j(gi i , 1 ) = j(gi , 1)j(i , 1 ).

in Denition 7.5(ii) is translated
as
follows. Let B . Then
The identity
g( 1, . . . , 1 ) = z is equivalent to (g)( 1, . . . , 1 ) = z, so
3 r
4

' j i gi , 1 ki

(g)1

f z,
O =

mi +ki 1 g,
i=1 det i gi
3 r
43 r
4
ki

'
' j gi , 1 ki

g1 1
j i , zi
=

mi +ki 1

mi +ki 1 g,
i=1 det i
i=1 det gi
3 r
4
ki

'
1

j i , zi
,
=
O

mi +ki 1 f z,
i=1 det i
(7.9)

a i bi
where now j(i , zi ) = ci zi + di C if i =

for all B and we have
c i di
the relation j(, z) = j(, z)j(, z) for all z H and g, h B . Accordingly, we
dene a right action of B on the space of functions in (7.8) by
4
3 r

' det i mi +ki 1

=
.
f |k z,
O
(7.10)
f z,
O
ki

j i , zi
i=1
Then

.
= f z,
O
O
f |k z,
Note that the central subgroup F B acts by

= NF/Q (a)k0 2 f z, a
, for a F .
f |k a z,
O
O
The C-vector space of modular forms of weight k and level N on B is nitedimensional and is denoted by MkB (N).
Lemma 7.11. There is a bijection between MkB (N) and the space of functions
/O
Wk (C) that are holomorphic in the rst variable and locally
f : (H )r B
constant in the second one and such that f |k = f for all B .
From now on, we will only work with modular forms f as presented in
Lemma 7.11. We dene the quaternionic Shimura variety of level N associated
to B as the double coset

/O
= B \ (H )r B
/O
.
/K B
X0B (N)(C) = B \ B
177
The set X0B (N)(C) can be equipped with the structure of a complex (possibly
disconnected) Riemannian manifold of dimension r.
Example 7.12. We recover rst the denition of classical modular forms when
F = Q and B = M2 (Q). For simplicity, we take N = 1. In this case, r = 1, so
Wk (C) = C and m = 0.
The action (7.10) is simply
k1

= (det )
.
f |k z,
O
f z,
O
k
j(, z)
We take the denition (7.8) as our starting point. The element

1 0
B = GL2 (Q)
0 1
identies the upper and lower half-planes, so a modular form
GL2 (Z)
C
f : H GL2 (Q)/

0 (N ) C, and the subgroup of
is determined by its restriction f : H GL2 (Q)/
+
GL2 (Q) which preserves H is exactly GL2 (Q).
We wish to recover the classical action using this new action of GL+
2 (Q), so
GL2 (Z).
An element of
we are led to consider the double coset GL+
(Q)\GL
(
Q)/
2
2
up to right-multiplication
this double coset is specied by an element
GL2 (Q)
2
2

.
by GL2 (Z), i.e., a Z-lattice in Q , i.e., Lat(Q ), specied by the rows of
2 ) given by
is a bijection, with inverse
But the map Lat(Q2 ) Lat(Q

Q2 . And since GL+ (Q) acts transitively on the left on the set of lattices
2

Lat(Q2 ), we conclude that # GL+
2 (Q)\GL2 (Q)/ GL2 (Z) = 1.
) for z H,
It follows that f is uniquely specied by the function f (z, O
which by abuse of notation we write simply f : H C. The stabilizer of GL+
2 (Q)
by multiplication on the left is GL+ (Z), so we recover the
acting on GL2 (Z)
2
condition

(det )k1
f (z)
f |k (z) =
j(, z)k
for all GL+
2 (Z), which is exactly the denition given in Section 1.
The interested reader can modify this argument for N > 1; alternatively, we
give a general derivation below.
M2 (F ), then we are in the situation of
We now dene cusp forms. If B =
Hilbert modular forms (but over a eld with arbitrary class number), so we dene
a cusp form to be a form f such that f (z) 0 whenever z tends to a cusp
P1 (F ) P1 (R)n . Otherwise, the Shimura variety X0B (N)(C) is compact. If B
is indenite, so 0 < r n, then there are no cusps, and we dene the space of
cusp forms to be SkB (N) = MkB (N). Finally, suppose B is denite; then r = 0. If
178
k = (2, . . . , 2), then we dene EkB (N) to be the space of those f MkB (N) such
F ; otherwise, we set E B (N) = 0. Then there
that f factors through nrd : B
k
is an orthogonal decomposition MkB (N) = SkB (N) EkB (N) and we call SkB (N)
the space of cusp forms for B.
The spaces MkB (N) and SkB (N) come equipped with an action of pairwise
commuting diagonalizable Hecke operators Tn indexed by the nonzero ideals n
, we dene a Hecke
B
of ZF , dened as follows. Given f SkB (N) and
operator associated to
by writing
@

=
O
O
O
i
(7.13)
i
and letting
T f

.
=
z,
O
f z,

i1 O
(7.14)
(Again, although it may seem unnatural, the choice of inverse here is to make the
denitions agree with the classical case.)
For a prime p of ZF with p DN, we denote by T
operator T ,
p the Hecke

p
0
is such that
v = 1 for v = p and
p =
O ZF ZF,p
where
B
=
0 1
M2 (ZF,p ), where p ZF,p is a uniformizer at p.
F such that pZ
F ZF = p, we dene
Equivalently, for a prime p and p Z

: nrd(
\
O
) = p ,
(p) = O
a set of cardinality N p + 1, and dene

,
=
Tp f z,
O
f z,

1 O
(7.15)
(p)
for the orbits in (p). For

where we have implicitly chosen representatives
O
an ideal n of ZF , the operator Tn is dened analogously.
We say that a cusp form f is a newform if it is an eigenvector of the Hecke
operators which does not belong to Mk (M) for M | N.
To unpack this denition further, and to relate this denition with the denitions given previously, we investigate the structure of the Shimura variety

/O
= B \ (H )r B
/O
.
/K B
X0B (N)(C) = B \ B
By Eichlers theorem of norms [66, Theorème III.4.1], we have

nrd(B ) = F(+)
= a F : vi (a) > 0 for i = r + 1, . . . , n ,
i.e., the norms from B consist of the subgroup of elements of F positive at all

real places which are ramied in B. In particular, B /B+
= (Z/2Z)r , where

= B : nrd() F+
B+
179
is the subgroup of B whose elements have totally positive reduced norm.
acts on Hr , therefore we may identify

The group B+

/O

X0B (N)(C) = B+
\ Hr B
/O
can be uniquely recovered from its reand a modular form on (H )r B
/O
. Now we have a natural (continuous) projection map
striction to Hr B

X0B (N)(C) B+
\B /O .
The reduced norm gives a surjective map

+

\B /O F+ \F /Z
nrd : B+
F = Cl ZF ,
(7.16)
where Cl+ ZF denotes the strict class group of ZF , i.e., the ray class group of ZF
with modulus equal to the product of all real (innite) places of F . The theorem of
strong approximation [66, Theorème III.4.3] implies that (7.16) is a bijection if B
is indenite. So our description will accordingly depend on whether B is indenite
or denite.
First, suppose that B is indenite. Then the space X0B (N)(C) is the disjoint
union of connected Riemannian manifolds indexed by Cl+ ZF , which we identify
explicitly as follows. Let the ideals a ZF form a set of representatives for Cl+ ZF ,
F ZF = a. (For the trivial class a = ZF , we
F be such that
aZ
and let
a Z
such that nrd(

) =
a. We let
choose
a = 1 .) By (7.16), there exists
B
1
B, so that O(1) = O, and we put a = O = O
a B . Then we
O
Oa =
a,+
+
have
@
@
r

H
B+
O
a \Hr ,
(7.17)
X0B (N)(C) =
[a]Cl+ (ZF )
[a]Cl+ (ZF )
where the last identication is obtained via the bijection

z.

z,
O
\ Hr
O
a \Hr , B+
B+
(7.18)
/O
Wk (C) satises f |k = f
Now let f MkB (N), so that f : (H )r B
B
for all B . Let Mk (N, a) be the space of functions fa : Hr Wk (C) such
that fa |k = fa for all a , where we dene
3 r
4
' (det i )mi +ki 1
(fa | )(z) =
fa (z)
ki
j(
,
z)
i
i=1
for B . Then, by (7.18), the map
#
MkB (N)
MkB (N, a),
[a]Cl+ (Z
F)
f (fa ),
(7.19)
180
where
),
fa (z) = f (z,
O
fa : Hr Wk (C),
is an isomorphism.
We now explain how the Hecke module structure on the left-hand side of
(7.19), dened in (7.14)(7.15), is carried over to the right-hand side. We follow
Shimura [57, Section 2]. We consider the action on the summand corresponding to
[a] Cl+ (ZF ). Extending the notation above, among the representatives chosen,
F ZF = b, and let B
be such that
let b be such that [b] = [ap1 ], let bZ
= b. By denition,
nrd()

.
=
Tp f a (z) = Tp f z,
O
f z,

1 O
(p)
Let
(p). Then, by strong approximation, we have
1 B = 1 Ob

1 O
(7.20)
with B , since this lattice has reduced norm [ap1 (ap1 )1 ] = [(1)]. There such that
1 = 1 , whence
fore, there exists u
O

1 u

=
.
(7.21)
Tp f a (z)
f z,

1 O
f z, 1 O
(p)
The second sum runs over a choice of as in equation (7.20) corresponding to

each
(p). This latter sum can be identied with a sum over values of fb as
follows. We have

= f |k z, 1 O

f z, 1 O
3 r
4
' (det i )mi +ki 1

= fb |k (z).
f z, O
=
k
i
j(i , zi )
i=1
The rst equality follows from the B -invariance of f and the others by denition
of the slash operators. Putting these together, we have

fb |k (z).
(7.22)
Tp f a (z) =

(The naturality of this denition explains the choice of inverses above.)

B and
O
This adelic calculation can be made global as follows. Let Ia =

(p), we have
Ib = O B. For
1 B = 1 Ob ,

1 O
hence

1
1

O

1 B = Ib Ia1 J.
1 = O

Ob = O
(7.23)
181
The elements thus obtained are characterized by their norms (in the right
lattice), as with the Hecke operators dened previously in (5.1). We analogously
dene

: nrd Ib Ia1 p = (nrd())

(p)a,b = b \ Ib Ia1 B+
(7.24)

= b \ Ib Ia1 B+
: nrd()b = ap .
Then, for fb MkB (N, b), we have Tp fb MkB (N, a) and

fb |k ,
Tp f b =
(p)a,b
where [b] = [ap1 ].

Example 7.25. If F has strict class number 1, then Ia = Ib = O = Oa , so we have
Ob a = Oa as in Section 5.
We note that the isomorphism (7.19) preserves the subspace of cusp forms
in a way that is compatible with the Hecke action, so we obtain a decomposition
#
SkB (N)
SkB (N, a).
[a]Cl+ (ZF )
B= M2 (F ), and let O = O0 (N) O0 (1) = M2 (ZF ). Then we

0
GL2 (F ) and so we nd simply that
1

ZF
a
1
M2 (ZF )
B =
= O0 (N, a).
Oa =
Na1 ZF
Example 7.26.
Let

a
may take
=
0
Let
0 (N, a) =
O0 (N, a)
+
.
=
Then
X0B (N)(C) =
a
c
b
d

O0 (N, a) : det
Z
F,+
/
.

0 N, a \Hn
[a]Cl+ (ZF )
is a disjoint union.
A Hilbert modular form of weight k and level N is a tuple (fa ) of holomorphic
functions fa : Hn C, indexed by Cl+ ZF , such that for all a we have

fa |k (z) = fa (z) for all 0 N, a
(with the extra assumption that f is holomorphic at the cusps if F = Q). Or, put
another way, let Mk (N, a) be the set of holomorphic functions Hn C such that
(f |k )(z) = f (z) for all 0 (N, a); then
#

Mk N, a .
Mk (N) =
[a]
182
In particular, we recover the denitions in Section 2 when F has strict class number 1.
A modular form f Mk (N, a) admits a Fourier expansion

f (z) = a0 +
a e2iTr(z)
(ad1 )+
analogous to (2.5). We say that f Mk (N, a) is a cusp form if f (z) 0 as z

tends to any cusp. Letting Sk (N, a) be the space of such cusp forms, we have
#

Sk (N) =
Sk N, a .
[a]
Let f = (fa ) Sk (N) be a Hilbert cusp form and let n ZF be an ideal.

Suppose that [n] = [ad1 ] amongst the representatives chosen for Cl+ (ZF ), and
let ZF be such that n = ad1 . We dene an = m a (fa ); the transformation
rule implies that an only depends on n and we call an the Fourier coecient of f
at n.
Now suppose that B is denite. Then the Shimura variety is simply
/O
= Cl O
X0B (N)(C) = B \B
and so is canonically identied with the set of right ideal classes of O. Note that
the reduced norm map (7.16) here is the map nrd : Cl O Cl+ ZF , which is
surjective but not a bijection in general. A modular form f MkB (N) is then
Wk (C) such that f |k = f for all B . Such a
/O
just a map f : B
function is completely determined by its values on a set of representatives of Cl O;
by
B, the stabilizer of B acting on
O
moreover, given any right ideal I =
O
1

O
B . Therefore, there is an isomorphism
left multiplication is OL (I) =
of complex vector spaces given by
#

MkB (N)
Wk (C)(I) , f f (
) ,
(7.27)
[I]Cl(O)

I=
OB

where (I) =
O
1 B = OL (I) and Wk (C)(I) is the (I)-invariant
subspace of Wk (C).
Having now discussed both the denite and indenite cases in turn, we return
to a general quaternion algebra B. Let f SkB (N)new be a newform. A theorem
of Shimura states that the coecients an are algebraic integers and Ef = Q({an })
is a number eld. The Hecke eigenvalues an determine the L-series
L(f, s) =

nZF
1 '
1
'
1
ap
ap
an
=
+
1
1
N ns
N ps
N p2s+1k0
N ps
pN
p|N
8. Denite method, arbitrary class number
183
associated to f (dened for Re s > 1). Moreover, associated to f is a Galois

representation: for l a prime of ZEf and Ef,l the completion of Ef at l, there is
an absolutely irreducible, totally odd Galois representation

f,l : Gal F /F GL2 Ef,l
such that, for any prime p lN, we have

Tr f,l (Frobp ) = ap (f ) and det(f,l Frobp ) = N pk0 1 .
The existence of this representation is due to work of BlasiusRogawski [2], Carayol [6], Deligne [13], Saito [51], Taylor [65], and Wiles [72].
The statement of the JacquetLanglands correspondence (3.9) reads the same
in this more general context.
Theorem 7.28 (JacquetLanglands). There is an injective map of Hecke modules
SkB (N) Sk (DN)
whose image consists of those forms which are new at all primes dividing D.
We are now ready to state the main general result of these notes, generalizing
the result of Theorem 2.8 to arbitrary class number and arbitrary weight.
Theorem 7.29 (DembeleDonnelly [18], Voight [69]). There is an algorithm which,
given a totally real eld F , a nonzero ideal N ZF , and a weight k (Z2 )[F :Q] ,
computes the space Sk (N) of Hilbert cusp forms of level N over F as a Hecke
module.
The proof of this theorem is discussed in the next two sections. It falls again
naturally into two methods, denite and indenite, which overlap just as in Remark 3.10.
Denite method, arbitrary class number
In this section, we return to the totally denite case but allow arbitrary class
number. As explained above, the space MkB (N) of modular forms of level N and
Wk (C) such that f |k = f
/O
weight k on B is the space of functions f : B
for all B .
We can use the identication (7.27) to compute the space SkB (N) as in the
direct approach of Section 4, with the appropriate modications. Let I1 , . . . , IH
be a set of representatives for Cl O such that nrd(Ii ) is coprime to DN for all i.
O = Ii , and let i = OL (Ii ) . Then, dualizing the
be such that
i O
Let
i O
isomorphism (7.27), we have
MkB (N)
=
H
#
i=1
Wk (C)i .
184
The Hecke module structure on this space is dened similarly as in Section 4, as

the following example illustrates.
Example 8.1. Consider the totally real quartic eld F = Q(w), where w4 5w2
2w + 1 = 0. Then F has discriminant 5744 = 24 359 and Galois group S4 . We
have Cl+ ZF = 2 (but Cl ZF = 1).

1, 1
The quaternion algebra B =
is unramied at all nite places
F
(and ramied at all real places). We compute a maximal order O and nd that
# Cl O = 4. We compute the action of the Hecke operators as in (3.8): we identify
the isomorphism classes of the N p + 1 right ideals of norm p inside each right ideal
I in a set of representatives for Cl O. We compute, for example, that
0 0 1 1
0 0 4 4
T(w3 4w1) =
2 2 0 0 ,
3 3 0 0
where N (w3 4w 1) = 4; note this matrix has a block form, corresponding to the
fact that (w3 4w 1) represents the nontrivial class in Cl+ ZF . Correspondingly,
6 2 0 0
8 12 0 0
T(w2 w4) =
0 0 8 4
0 0 6 10
with N (w2 w 4) = 13 is a block scalar matrix, as (w2 w 4) is trivial in
Cl+ ZF . In this case, the space E2 (1) of functions that factor through the reduced
norm has dimension dim E2 (1) = 2, so dim S2 (1) = 2, and we nd that this space
is irreducible as a Hecke module and so has a unique constituent f .
We obtain the following table of Hecke eigenvalues:
p
(w3 4w 1)
(w 1)
(w2 w 2)
Np
ap (f )
2t
(w2 3)
13
t
(w2 w 4)
(w2 2)
13
17
3t
Here t satises the polynomial t2 6 = 0. As in Section 6, we

predict the existence
of an abelian variety over F with real multiplication by Q( 6) and everywhere
good reduction.
As in Section 4, the disadvantage of the approach used in Example 8.1 is
that for each level N one must compute the set of ideal classes Cl O = Cl O0 (N).
By working with a more complicated coecient module, we can work with ideal
classes only with the maximal order O0 (1), as follows.
Changing notation, let now I1 , . . . , Ih be representatives for Cl O0 (1), with
i O0 (1) B. By strong approximation, we may
h = # Cl O0 (1), and let Ii =
8. Denite method, arbitrary class number
185
assume that each nrd(Ii ) is coprime to DN. Indeed, we may assume each nrd(Ii )
is supported in any set S of primes that generate Cl+ ZF . Let O0 (1)i = OL (Ii ) =
0 (1)
i1 B be the left order of Ii . Then O0 (1)i ZF ZF,N
i O
= O0 (1) ZF ZF,N .
Let a for a P1 (ZF /N) represent the O0 (1), O-connecting ideals of norm
0 (1) and if Ja = O
0 (1)a B then OR (Ja ) = O. Then the set
N; that is, a O

i a OB, covers all isomorphism classes of right O-ideals,
{Ii Ja }i,a , where Ii Ja =
but not necessarily uniquely: two such ideals Ii Ja and Ij Jb are isomorphic if and
only if i = j and there exists O0 (1)
i such that Ja = Jb , comparing the
0 (1). The action of O0 (1) can be equivalently given on
elements
i a ,
j b O
i
the set of indices a P1 (ZF /N): via the (reduction of a) splitting map

(8.2)
N : O0 (1) O0 (1) ZF ZF,N
= M2 ZF,N ,
1
1
0 (1) /O

each O0 (1)
we have O
i acts on the left on P (ZF /N), and5
i
i P (ZF /N).
h
B
We conclude that MkB (N)
= MB
k (N) =
i=1 Mk (N)i , where

1
MB
k (N)i = f : P ZF /N Wk (C) : f |k = f for all O0 (1)i .
In this presentation, the Hecke operators act as follows. For a prime p, let

1
(p)i,j = O0 (1)
: nrd xIi Ij1 = p .
i \ x Ii Ij
B
We then dene the linear map Tp : MB
k (N) Mk (N) on each component by the
rule

B
f
f |k .
(8.3)
(Tp )i,j : MB
k (N)i Mk (N)j ,
(p)i,j
This is indeed an isomorphism of Hecke modules. For further details, see work of
the rst author [15, Theorem 2] which traces these maps under the assumption
that F has narrow class number 1, but this assumption can be easily removed.
Put another way, by the decomposition
=
B
h
@
0 (1) ,
B
i O
i=1
we decompose the set
X0B (N)
/O
=
X0B (N) = B \B
h
@
as
h
@

0 (1) /O

B\ B
i O
O0 (1)
i \O0 (1)i /Oi ,
i=1
i=1
where the last identication is obtained by sending

i u to
i u
i1 . Thus, analogously to (7.17), we get a decomposition
X0B (N) =
h
@
i=1
X0B (N)i =
h
@
i=1

i \P1 ZF /N .
(8.4)
186
In particular, this gives X0B (1) = Cl O0 (1). From this, we get an identication
MkB (N)
h
#
MB
k (N)i ,
f (fi )i ,
(8.5)
i=1
0 (1) such that x =

where we set fi (x) = f (
i ) after choosing
O
i .
i
Again, the decomposition (8.5) is analogous to (7.19), and one shows that it is a
Hecke module isomorphism by arguing similarly.
/O
Wk (C) such that
Now f MkB (N) is by denition a map f : B
f |k = f for all B . Associated to such a map, via the identication (8.4),

0 (1) /O
Wk (C) such
f is uniquely dened by a tuple of maps (fi )i with fi : O
i
i
that f |k = f for all i = O0 (1)i . In other words,
MkB (N)
h
#

0 (1) /O
, Wk (C)
H 0 i , Hom O
i
i
i=1
h
#
i=1

(1)
O
H 0 i , CoindO 0 i Wk (C) .
i
But then, as in (8.2), we have

(1)
O
H 0 i , CoindO 0 i Wk (C)
= MB
k (N)i .
i
Example 8.6. The real quadratic eld F = Q( 106) has strict class number 2 and
class number 2. We compute that the space S2 (1) of Hilbert cusp forms of level 1
and parallel weight 2 has dimension 50. It decomposes into four Hecke constituents
of dimension 1, six of dimension 2, two of dimension 4 and one of dimension 26.
The table below contains the rst few Hecke eigenvalues of the one-dimensional
constituents.
(2, w) (3, w + 1) (3, w 1) (5, w + 1) (5, w 1) (3w + 31) (3w 31)
p
Np
ap (f1 )
ap (f2 )
ap (f3 )
ap (f4 )
The forms f1 and f2 (resp. f3 and f4 ) are interchanged by the action of

Gal(F/Q) (on the ideals p). The forms f1 and f3 (resp. f2 and f4 ) are interchanged
by the action of Cl+ ZF , so these forms are twists via the strict class character of
Gal(F + /F ), where F + denotes the strict class eld of F .
Elkies has found a curve E which gives rise to the above data:
E : y 2 wxy 2wy = x3 + (2 2w)x2 + (10809936 + 1049944w)x
+ (19477819120 + 1891853024w).
9. Indenite method, arbitrary class number
187
The curve E has j-invariant j(E) = 264235 + 25777w and has everywhere good
reduction. We conclude that E is modular using Kisin [42, Theorem 2.2.18] (see
also Kisin [43, Theorem 3.5.5]): we need to verify that 3 is split in F , that E
has no CM, and that the representation 3 : Gal(F /F ) GL2 (Z3 ) has surjective
reduction 3 : Gal(F /F ) GL2 (F3 ) which is solvable hence modular.
We nd that E matches the form f1 ; so its conjugate by Gal(F/Q) corresponds to f2 and the quadratic twist of E by the fundamental unit 4005 389w
(of norm 1) corresponds to f3 (and its conjugate to f4 ).
The input of our algorithm is a totally real number eld F of degree n,
a totally denite quaternion algebra B with discriminant D, an integral ideal
N ZF which is coprime with D, a weight k Zn such that ki 2 and ki kj
(mod 2), and a prime p D. The output is then a matrix giving the action of Tp
5h
in a basis of MkB (N) = i=1 MB
k (N)i which is independent of p. By computing
enough Tp and simultaneously diagonalising, one obtains all Hecke constituents
corresponding to Hilbert newforms of level N and weight k.
The algorithm starts by nding a maximal order O0 (1), then computes a set
of representatives Cl O0 (1) of the right ideal classes of O0 (1) whose norms generate
Cl+ (ZF ) and are supported outside DN. This part of the algorithm uses work of
the second author and Kirschmer [41]; it is the most time-consuming part but can
be seen as a precomputation. Next, the algorithm nds a fundamental domain for
the action of each i on P1 (ZF /N) and computes MkB (N) as the direct sum of
#
Wk (C)x ,
MB
k (N)i =
[x]X0B (N)i
where x is the stabilizer of x in i . From this, one obtains a basis of MkB (N).
Finally, the algorithm computes the sets (p)i,j and then the block matrices which
give the action of Tp in this basis. We refer to [15] and [18] for further details on
the implementation.
Indenite method, arbitrary class number
In this section, we generalize the indenite method to arbitrary class number. We

carry over the notation from Section 7, and now take the quaternion algebra B to
be ramied at all but one real place.
In this case, from (7.17)(7.18) the space
/O
)
\(H B
X(C) = X0B (N)(C) = B+
is the disjoint union of Riemann surfaces indexed by Cl+ ZF . Let {a} be a set of
F be such that
F ZF = a for each a.
aZ
aZ
representatives for Cl+ ZF and let
Then
@
@
a \H =
Xa (C),
(9.1)
X(C) =
[a]Cl+ (ZF )
[a]Cl+ (ZF )
188
1 B and a = O .
O
where Oa =
a,+
Therefore, a modular form of weight k and level N is a tuple (fa ) of functions
fa : H Wk (C), indexed by [a] Cl+ ZF , such that, for all a, we have

fa |k (z) = fa (z)
for all a and all z H. In particular, if k = (2, . . . , 2) is parallel weight 2,
then (fa ) corresponds to a tuple of holomorphic 1-forms ((2i)fa (z) dz)a , one for
each curve Xa (C).
We compute with this space of functions by relating them to cohomology,
and for that we must modify the coecient module. Dene the right GL2 (C)n =
GL2 (C) GL2 (C)n1 -module
Vk (C) =
n
2
Pwi (mi )(C) = Pw1 (m1 )(C) Wk (C).
i=1
The group B acts on Vk (C) via the composite splitting B GL2 (C)n given by
(i )i . The EichlerShimura theorem, combined with the isomorphism (7.19),
applied to each component Xa (C) of X(C) in (9.1), gives the isomorphism of Hecke
modules
#

+
H 1 a , Vk (C) ,
(9.2)
SkB (N)
[a]
where + denotes the +1-eigenspace for complex conjugation.

5 1
In the description (9.2), the Hecke operators act on
H (b , Vk (C)) in the
following way; we follow their denition in (7.24). Let p be a prime ideal of ZF
with p DN. We consider the [b]-summand, and given f H 1 (b , Vk (C)) we will
B and Ib = O
B
dene Tp f H 1 (a , Vk (C)), where [b] = [p1 a]. Let Ia =
O
so that nrd(Ib ) = b and nrd(Ia ) = a, and let

Ib Ia1 : nrd(Ib Ia1 )p = nrd()

(p)a,b = b \ B+

Ib Ia1 : nrd()a = pb ,
= b \ B+
where b = Ob,+
acts by multiplication on the left. Let a , so that Ia = Ia .
Then the map on B induces a bijection (of the equivalence classes) of
(p)a,b . Therefore, for every (p)a,b , there exist a and (p)a,b
such that = . From (7.22) and the EichlerShimura theorem, we have

f ( ) .
(9.3)
(Tp f )() =
(p)a,b
One can similarly dene the AtkinLehner involutions.

Admittedly, this description is complicated, but it can be summarized simply:
a Hecke operator Tp permutes the summands (9.2) in accordance with translation
by [p] in Cl+ ZF , and adjusting for this factor one can principalize as before (when
189
the strict class number was 1). The resulting Hecke matrices are consequently
block matrices.
We illustrate this with an example; we then give a few more details on the
algorithm.
Example 9.4. Let F = Q(w) be the (totally real) cubic eld of prime discriminant 257, with w3 w2 4w + 3 = 0. Then F has Galois group S3 and ZF = Z[w].
The eld F has class number 1 but strict class number 2: the unit (w 1)(w 2)
2
generates the group Z
F,+ /ZF of totally positive units modulo squares.

1, w 1
Let B =
be the quaternion algebra with i2 = 1, j 2 = w 1,
F
and ji = ij. Then B has discriminant D = (1) and is ramied at two of the
three real places and unramied at the place with w 2.19869 . . . , corresponding
to : B M2 (R). The order

O = ZF w2 + w 3 iZF (w2 + w) 8i + j /2ZF (w2 + w 2)i + ij /2ZF
is an Eichler order of level N = (w)2 , where N (w) = 3. A fundamental domain for
) on H is as follows:
the action of = (O+
The ideals (1) and a = (w+1)Z

F represent the classes in the strict class group

Cl+ ZF . The ideal Ja = 2O + (w2 + w + 2)/2 4i + (1/2)j O has nrd(Ja ) = a.
The left order of Ja is OL (Ja ) = Oa , where

Oa = ZF w2 2w3 iZF (w2 +w)/24i+(1/2)j ZF

(1/10) (174w2 343w348)i+(w2 2w2)j +(w2 +2w+2)ij ZF .
) on H is as follows:
A fundamental domain for the action of a = (Oa,+
190
The orders O and Oa are not isomorphic since the connecting ideal Ia (with
left order Oa and right order O) is not principal. This implies that the groups
and a are not conjugate as subgroups of PSL2 (R) but nevertheless are isomorphic
as abstract groups: they both have signature (1; 2, 2, 2, 2), so that
<
;
= a
= , , 1 , . . . , 4 : 12 = = 42 = [, ]1 4 = 1 .
In particular, both X(1) (C) and Xa (C) have genus 1, so

dim H 1 X(C), C = dim H 1 X(1) (C), C + dim H 1 Xa (C), C = 4 = 2 dim S2 (N).
We choose a basis of characteristic functions on , as a basis for H 1 (X(1) (C), C)
and similarly for H 1 (Xa (C), C).
We now compute Hecke operators following the above. Let H = H 1 (X(C), C).
We compute that complex conjugation acts on H by the matrix
1 1
0 0
0
1
0 0
.
H | W =
0
0 1 0
0
0
1 1
Note that, in this case, W preserves each factor. Now consider the Hecke operator
Tp , where p = (2w 1) and N (p) = 7. Then p represents the nontrivial class in
Cl+ ZF . We compute that
0
0 3
2
0
0 2 4
H | Tp =
4 2
0
0
2 3
0
0
and restricting we get
H | Tp =
+

0 2
.
8
0
191
Therefore there are two eigenspaces for Tp with eigenvalues 4, 4. By contrast, the
Hecke operator T(2) acts by the scalar matrix 3 on H, preserving each component.
Continuing in this way, we nd the following table of eigenvalues:
Np
19
25
37
41
43
47
49
53
61
61
61
ap (f )
12
ap (g)
12
Clearly, the form g is the quadratic twist of the form f by the nontrivial character
of the strict class group Gal(F + /F ), where F + is the strict class eld of F . Note
also that these forms do not arise from base change from Q, since ap has three
dierent values for the primes p of norm 61.
We are then led to search for elliptic curves of conductor N = (w)2 , and we
nd two:
Ef : y 2 + (w2 + 1)xy
= x3 x2 + (36w2 + 51w 18)x + (158w2 + 557w 317),
Eg : y 2 + (w2 + w + 1)xy + y
= x3 + (w2 w 1)x2 + (4w2 + 11w 11)x + (4w2 + w 3).
Each of these curves has nontrivial Z/2Z-torsion over F , so as above they are
modular and we match Hecke eigenvalues to nd that Ef corresponds to f and
Eg corresponds to g.
In this situation, although by the theory of canonical models due to Deligne
we know that the variety X(C) = X(1) (C) Xa (C) has a model XF over F , the
curves themselves are not dened over F they are interchanged by the action of
Gal(F + /F ). Nevertheless, the Jacobian of XF is an abelian variety of dimension 2
dened over F which is isogenous to Ef Eg we characterize in this way isogeny
classes, not isomorphism classes.
As in the case of class number 1, the application of Shapiros lemma allows
us always to work with the group associated to a maximal order, as follows. Let
O0 (1) O be a maximal order containing O, and, for each ideal a, let O0 (1)a =
0 (1)
1 B be the maximal order containing Oa , and let (1)a = (O+
).
O
Further, dene
(1)
Vk (C)a = Coinda a Vk (C)
for each a. Then Shapiros lemma implies that

H 1 a , Vk (C)
= H 1 (1)a , Vk (C)a
and so
SkB (N)
#
[a]

+
H 1 (1)a , Vk (C)a .
192
Our algorithm takes as input a totally real eld F of degree [F : Q] = n,

a quaternion algebra B over F split at a unique real place, an ideal N ZF
coprime to the discriminant D of B, a vector k (2Z>0 )n , and a prime p DN,
and outputs the matrix of the Hecke operator Tp acting on the space
#

+
H 1 (1)b , Vk (C)b
H=
b
with respect to some xed basis which does not depend on p. From these matrices,
we decompose the space H into Hecke-irreducible subspaces by linear algebra. We
give a short overview of this algorithm.
First, some precomputation. We precompute a set of representatives [a] for
the strict class group Cl+ ZF with each a coprime to pDN. For each representative ideal a, precompute a right O0 (1)-ideal Ia such that nrd(Ia ) = a and let
O0 (1)a = OL (Ia ) be the left order of Ia . Next, we compute for each a a nite presentation for (1)a consisting of a (minimal) set of generators Ga and relations Ra
together with a solution to the word problem for the computed presentation [67].
Then,
using standard linear algebra techniques, we compute a basis for the space
5
1
H
((1)a , Vk (C)a ).
[a]
The main issue then is to make the description (9.3) amenable to explicit
computation. First, compute a splitting p : O0 (1) M2 (ZF,p ). Then, for each
ideal a, perform the following steps.
First, compute the ideal b with ideal class [b] = [p1 a]. Compute the left
ideals

x y
Ja = Oa 1
+ Oa p
p
0 0
indexed by the elements a = (x : y) P1 (Fp ). Then compute the left Ob -ideals
corresponding to
Ib Ia1 Ja and compute totally positive generators a Oa B+
Ob a = Ib Ia1 Ja ; see [41].
Now, for each in a set of generators Ga for a , compute the permutation
of P1 (Fp ) as in [30, Algorithm 5.8] and then the elements a = a 1
a for
a P1 (Fp ); write each such element a as a word in Gb and then apply the formula

fb (a )a .
Tp fa () =
aP1 (Fp )
The algorithm in its full detail is rather complicated to describe; we refer the
reader to work of the second author [69] for the details.
Bibliography
[1] Don Blasius, Elliptic curves, Hilbert modular forms, and the Hodge conjecture, Contributions to automorphic forms, geometry, and number theory,
Johns Hopkins Univ. Press, Baltimore, 2004, 83103.
[2] Don Blasius and Jonathan Rogawski, Galois representations for Hilbert modular forms, Bull. Amer. Math. Soc. (N.S.) 21 (1989), no. 1, 6569.
[3] Wieb Bosma, John Cannon, and Catherine Playoust, The Magma algebra
system, I. The user language, J. Symbolic Comput. 24 (1997), no. 3-4, 235
265.
[4] Armand Brumer, The rank of J0 (N ), Asterisque 228 (1995), 4168.
[5] Kevin Buzzard, Fred Diamond, and Frazer Jarvis, On Serres conjecture for mod Galois representations over totally real elds, preprint,
arXiv:0810.2106v2.
[6] Henri Carayol, Sur les representations l-adiques associees aux formes mod
ulaires de Hilbert, Ann. Sci. Ecole
Norm. Sup. (4) 19 (1986), no. 3, 409468.
[7] Henri Cohen, A course in computational algebraic number theory, Graduate
Texts in Math., vol. 138, Springer-Verlag, Berlin, 1993.
[8] John Cremona, Modular symbols for 1 (N ) and elliptic curves with everywhere good reduction, Math. Proc. Cambridge Philos. Soc. 111 (1992), no. 2,
199218.
[9] John Cremona, Algorithms for modular elliptic curves, 2nd ed., Cambridge University Press, Cambridge, 1997, available at the authors webpage
http://www.maths.nott.ac.uk/personal/jec/book/.
[10] John Cremona and Mark Lingham, Finding all elliptic curves with good
reduction outside a given set of primes, Expo. Math. 16 (2007), no. 3, 303
312.
[11] Henri Darmon, Rational points on modular elliptic curves, CBMS Series in
Math., vol. 101, Amer. Math. Soc., Providence, RI, 2004.
193
194
Bibliography
[12] Pierre Deligne, Travaux de Shimura, Seminaire Bourbaki, 23ème annee

(1970/71), Exp. No. 389, 123165, Lecture Notes in Math. vol. 244, Springer,
Berlin, 1971.
[13] Pierre Deligne, Formes modulaires et representations l-adiques, Seminaire
Bourbaki (1968/69), Exp. No. 355, 139172, Lecture Notes in Math. vol. 179,
Springer, Berlin, 1971.
[14] Lassina Dembele, Explicit computations of Hilbert modular forms on Q( 5),

Experiment. Math. 14 (2005), no. 4, 457466.
[15] Lassina Dembele, Quaternionic Manin symbols, Brandt matrices and Hilbert
modular forms, Math. Comp. 76 (2007), no. 258, 10391057.
[16] Lassina Dembele, A non-solvable Galois extension of Q ramied at 2 only,
C. R. Acad. Sci. Paris, Ser. I, 347 (2009), 111116.
[17] Lassina Dembele, Matthew Greenberg, and John Voight, Nonsolvable number elds ramied only at 3 and 5, Compos. Math. 147 (2011), no. 3, 716734.
[18] Lassina Dembele and Steve Donnelly, Computing Hilbert modular forms over
elds with nontrivial class group, Algorithmic number theory (Ban, 2008),
Lecture Notes in Comput. Sci., vol. 5011, Springer, Berlin, 2008, 371386.
[19] Fred Diamond and Jerry Shurman, A rst course in modular forms, Graduate Texts in Math., vol. 228, Springer-Verlag, New York, 2005.
[20] Luis V. Dieulefait, A non-solvable extension of Q unramied outside 7,
preprint, arXiv:1005.4209.
[21] Steve Donnelly and John Voight, Tables of Hilbert modular forms and elliptic
curves over totally real elds, in preparation.
[22] Martin Eichler, On theta functions of real algebraic number elds, Acta
Arith. 33 (1977), no. 3, 269292.
[23] Martin Eichler, The basis problem for modular forms and the traces of
the Hecke operators, Modular functions of one variable, I (Proc. Internat.
Summer School, Univ. Antwerp, Antwerp, 1972), Lecture Notes in Math.,
vol. 320, Springer, Berlin, 1973, 75151.
[24] Martin Eichler, Correction to: The basis problem for modular forms and the
traces of the Hecke operators, Modular functions of one variable, IV (Proc.
Internat. Summer School, Univ. Antwerp, Antwerp, 1972), Lecture Notes in
Math., vol. 476, Springer, Berlin, 1975, 145147.
[25] Eberhard Freitag, Hilbert modular forms, Springer-Verlag, Berlin, 1990.
[26] Gerard van der Geer, Hilbert modular surfaces, Springer-Verlag, Berlin, 1988.
Bibliography
195
[27] Stephen Gelbart and Herve Jacquet, Forms of GL(2) from the analytic point
of view, Automorphic forms, representations and L-functions (Corvallis, OR,
1977), vol. 1, eds. A. Borel and W. Casselman, Proc. Sympos. Pure Math.
33, Amer. Math. Soc., Providence, RI, 1979, 213251.
[28] David Geraghty, Modularity lifting theorems for ordinary Galois representations, preprint, http://www.math.harvard.edu/~geraghty/oml.pdf.
[29] Eyal Z. Goren, Lectures on Hilbert modular varieties and modular forms,
CRM Monograph Series, vol. 14, Amer. Math. Soc. Providence, RI, 2002.
[30] Matt Greenberg and John Voight, Computing systems of Hecke eigenvalues
associated to Hilbert modular forms, Math. Comp. 80 (2011), no. 274, 1071
1092.
[31] Ki-ichiro Hashimoto, On Brumers family of RM-curves of genus two, Tohoku Math. J. (2) 52 (2000), no. 4, 475488.
[32] Haruzo Hida, On p-adic Hecke algebras for GL2 over totally real elds, Ann.
of Math. (2) 128 (1988), no. 2, 295384.
[33] Haruzo Hida, On abelian varieties with complex multiplication as factors
of the Jacobians of Shimura curves, Amer. J. of Math. 103 (1981), no. 4,
727776.
[34] Haruzo Hida, Hilbert modular forms and Iwasawa theory, Oxford Sci. Publ.,
Oxford, 2006.
[35] Hiroaki Hijikata, Explicit formula of the traces of Hecke operators for 0 (N ),
J. Math. Soc. Japan 26 (1974), no. 1, 5682.
[36] Herve Jacquet and Robert P. Langlands, Automorphic forms on GL(2), Lectures Notes in Math., vol. 114, Springer-Verlag, Berlin, 1970.
[37] Frazer Jarvis, Mazurs principle for totally real elds of odd degree, Compos.
Math. 116 (1999), no. 1, 3979.
[38] Anthony W. Knapp, Elliptic curves, Math. Notes, vol. 40, Princeton Univ.
Press, Princeton, 1992.
[39] Chandrashekhar Khare and Jean-Pierre Wintenberger, On Serres conjecture
for 2-dimensional mod p representations of the absolute Galois group of the
rationals, to appear in Ann. of Math.
[40] Lloyd J. P. Kilford, Modular forms: A classical and computational introduction, Imperial College Press, London, 2008.
[41] Markus Kirschmer and John Voight, Algorithmic enumeration of ideal
classes for quaternion orders, SIAM J. Comput. (SICOMP) 39 (2010), no. 5,
17141747.
196
Bibliography
[42] Mark Kisin, The Fontaine-Mazur conjecture for GL2 , J. Amer. Math. Soc.
22 (2009), no. 3, 641690.
[43] Mark Kisin, Moduli of nite at group schemes, and modularity, Ann. of
Math. (2) 170 (2009), no. 3, 10851180.
[44] Koopa Tak-Lun Koo, William Stein, and Gabor Wiese, On the generation
of the coecient eld of a newform by a single Hecke eigenvalue, J. Theorie
Nombres Bordeaux 20 (2008), no. 2, 373384.
[45] David Kohel, Hecke module structure of quaternions, Class eld theory: its
centenary and prospect (Tokyo, 1998), ed. K. Miyake, Adv. Stud. Pure
Math., vol. 30, Math. Soc. Japan, Tokyo, 2001, 177195.
[46] Barry Mazur and Andrew Wiles, Class elds of abelian extensions of Q,
Invent. Math. 76 (1984), 179330.
[47] Jean-Francois Mestre, La methode des graphes: exemples et applications,
Proceedings of the international conference on class numbers and fundamental units of algebraic number elds (Katata), Nagoya Univ., Nagoya,
1986, 217242.
[48] Arnold Pizer, An algorithm for computing modular forms on 0 (N ), J. Algebra 64 (1980), no. 2, 340390.
[49] Boyd B. Roberts and Lawrence C. Washington, The modularity of some
Q-curves, Compos. Math. 111 (1998), no. 1, 3549.
[50] David P. Roberts, Nonsolvable polynomials with eld discriminant 5A ,
preprint, http://cda.morris.umn.edu/~roberts/research/five.pdf.
[51] Takeshi Saito, Hilbert modular forms and p-adic Hodge theory, Compos.
Math. 145 (2009), no. 5, 10811113.
[52] Rene Schoof, Abelian varieties over cyclotomic elds with good reduction
everywhere, Math. Ann. 325 (2003), no. 3, 413448.
[53] Jean-Pierre Serre, Congruences et formes modulaires [daprès H. P. F. Swinnerton-Dyer], Seminaire Bourbaki, 24ème annee (1971/72), Exp. No. 416,
319338, Lecture Notes in Math., vol. 317, Springer, Berlin, 1973.
[54] Jean-Pierre Serre, Proprietes galoisiennes des points dordre ni des courbes
elliptiques, Invent. Math. 15 (1972), no. 4, 259331.
[55] Jean-Pierre Serre, Abelian -adic representations and elliptic curves, Research Notes in Mathematics, vol. 7, Wellesley, MA, A. K. Peters, 1997.
[56] N. I. Shepherd-Barron and R. Taylor, Mod 2 and mod 5 icosahedral representations, J. Amer. Math. Soc. 10 (1997), no. 2, 283298.
[57] Goro Shimura, Construction of class elds and zeta functions of algebraic
curves, Ann. of Math. (2) 85 (1967), 58159.
Bibliography
197
[58] Goro Shimura, The special values of the zeta functions associated with Hilbert
modular forms, Duke Math. J. 45 (1978), no. 3, 637679.
[59] Goro Shimura, Introduction to the arithmetic theory of automorphic functions, Publ. Math. Soc. Japan, vol. 11, Princeton Univ. Press, Princeton,
1994.
[60] Jude Socrates and David Whitehouse, Unramied Hilbert modular forms,
with examples relating to elliptic curves, Pacic J. Math. 219 (2005), no. 2,
333364.
[61] C. M. Skinner and Andrew J. Wiles, Nearly ordinary deformations of irreducible residual representations, Ann. Fac. Sci. Toulouse Math. (6) 10 (2001),
no. 1, 185215.
[62] C. M. Skinner and Andrew J. Wiles, Residually reducible representations and
modular forms, Inst. Hautes Etudes

Sci. Publ. Math. 89 (1999), 5126.
[63] Andrew Snowden, On two dimensional weight two odd representations of
totally real elds, preprint, arXiv:0905.4266.
[64] William Stein, Modular forms, a computational approach, with an appendix
by Paul E. Gunnells, Graduate Studies in Math., vol. 79, Amer. Math. Soc.,
Providence, RI, 2007.
[65] Richard Taylor, On Galois representations associated to Hilbert modular
forms, Invent. Math. 98 (1989), no. 2, 265280.
[66] Marie-France Vigneras, Arithmetique des algèbres de quaternions, Lecture
Notes in Math., vol. 800, Springer, Berlin, 1980.
[67] John Voight, Computing fundamental domains for Fuchsian groups, J. Theorie Nombres Bordeaux 21 (2009), no. 2, 467489.
[68] John Voight, Shimura curve computations, Arithmetic Geometry, Clay
Math. Proc., vol. 8, Amer. Math. Soc., Providence, 2009, 103113.
[69] John Voight, Computing automorphic forms on Shimura curves over elds
with arbitrary class number, Algorithmic number theory (ANTS IX, Nancy,
France, 2010), eds. Guillaume Hanrot, Francois Morain, and Emmanuel
Thome, Lecture Notes in Comp. Sci., vol. 6197, Springer, Berlin, 2010, 357
371.
[70] John Voight, Identifying the matrix ring: algorithms for quaternion algebras
and quadratic forms, preprint, arXiv:1004.0994.
[71] Andrew Wiles, Modular elliptic curves and Fermats Last Theorem, Ann. of
Math. (2) 141 (1995), no. 3, 443551.
[72] Andrew Wiles, On ordinary -adic representations associated to modular
forms, Invent. Math. 94 (1988), no. 3, 529573.
198
Bibliography
[73] John
Wilson, Explicit moduli for curves of genus 2 with real multiplication
by Q( 5), Acta Arith. 93 (2000), no. 2, 121138.
[74] Shou-Wu Zhang, Heights of Heegner points on Shimura curves, Ann. of
Math. (2) 153 (2001), no. 1, 27147.
Part III
Elliptic Curves
Notes on the Parity Conjecture

Tim Dokchitser1
The main purpose of these notes is to prove, in a reasonably self-contained way,
that niteness of the TateShafarevich group implies the parity conjecture for
elliptic curves over number elds. Along the way, we review local and global root
numbers of elliptic curves and their classication, and we end by discussing some
peculiar consequences of the parity conjecture.
Essentially nothing here is new, and the notes follow closely the articles
[11][17], all joint with Vladimir Dokchitser. There are only some additional shortcuts replacing a few technical computations of root numbers and Tamagawa numbers by a deforming to totally real elds argument. Also, this is not meant to
be a complete survey of results on the parity conjecture. Many important results,
especially those concerning Selmer groups, are not mentioned.
The exposition is based on lectures given at CRM Barcelona in December
2009. It is a great pleasure to thank CRM for its warm hospitality and Vctor
Rotger and Luis Dieulefait for organising the lecture series. I would also like to
thank its participants, especially Anton Mellit, Ravi Ramakrishna and Thomas de
La Rochefoucauld for many useful comments, Alex Bartel for his helpful lecture
series preceding this one, and Vladimir Dokchitser for proofreading the manuscript.
BirchSwinnerton-Dyer and parity
1.1
Conjectures and the main result
Throughout the notes K denotes a number eld. Suppose E/K is an elliptic curve,
say in Weierstrass form,
y 2 = x3 + ax + b, with a, b K.
The set E(K) of K-rational solutions (x, y) to this equation together with a point
at innity O forms an abelian group, which is nitely generated by famous theorems of Mordell and Weil. The primary arithmetic invariant of E/K is its rank:
1 The
author is supported by a Royal Society University Research Fellowship.
201
202
Notes on the parity conjecture
Denition. The MordellWeil rank rk E/K is the Z-rank of E(K)/torsion.

The celebrated conjecture of Birch and Swinnerton-Dyer relates it to another
fundamental invariant, the L-function L(E/K, s). We dene it in 3, together with
its conductor N Z and the global root number w(E/K) = 1. For now it suces
to say that L(E/K, s) is given by a Dirichlet series

an ns
L E/K, s =
n=1
with an Z, which converges for Re s > 3/2.

HasseWeil Conjecture. The function L(E/K, s) has an analytic continuation to
the whole of C. The completed L-function

d
d ds s/2

L E/K, s = 2s s+1
N L E/K, s ,
2
where d = [K : Q], satises a functional equation

L E/K, 2 s = w(E/K) L E/K, s .
Granting the analytic continuation, we may state
BirchSwinnerton-Dyer Conjecture I. ords=1 L(E/K, s) = rk E/K.
This remarkable conjecture relates arithmetic properties of E to analytic
properties of its L-function. It was originally stated over the rationals, extended
to all abelian varieties over global elds by Tate, and vastly generalised by Deligne,
Gross, Beilinson, Bloch and Kato. One immediate consequence of the two conjectures above is
Parity Conjecture. (1)rk E/K = w(E/K).
In particular, an elliptic curve whose root number is 1 must have innitely
many rational points. This is a purely arithmetic statement which does not involve
L-functions and it might appear to be simpler than the two conjectures above.
However, it is remarkably hard, and we have no approach to resolve it in any
kind of generality. It has several important consequences: for instance, it settles
Hilberts 10th problem over rings of integers of arbitrary number elds [36] and
implies most remaining cases of the congruent number problem.
The diculty is that we know virtually nothing about the rank of a general
elliptic curve, as it is very hard to distinguish rational points from the elements
of X:
Denition. The TateShafarevich group of E/K is dened by
3
4
'

1
1
XE/K = ker H K, E(K)

H Kv , E(Kv ) .
v
1. BirchSwinnerton-Dyer and parity
203
The famous ShafarevichTate conjecture asserts that XE/K is nite. The

main result of these notes is that its niteness does imply the parity conjecture.
(Over function elds, Artin and Tate [58] showed that the full BirchSwinnertonDyer conjecture follows from niteness of X.)
and K(E[n]) for the eld obtained
We write E[n] for the n-torsion in E(K),
by adjoining to K the coordinates of points in E[n].
Theorem A. Let E be an elliptic curve over a number eld K, and suppose that
XE/K(E[2]) has nite 2- and 3-primary parts. Then
(1)rk E/K = w(E/K).
The problem whether niteness of X implies parity has a reasonably long
history. It was solved for elliptic curves over Q with j-invariant 0 or 1728 by Birch
and Stephens [3]; for CM elliptic curves over Q by Greenberg [24] and Guo [25]; for
all (modular) elliptic curves over Q by Monsky [39], and for most modular elliptic
curves over totally real elds by Nekovar [41]. Over arbitrary number elds, the
theorem above is proved in [12] for elliptic curves with decent reduction types at
2 and 3, and for all elliptic curves in [16].
1.2
BirchSwinnerton-Dyer II and isogeny invariance
To explain our approach to the parity conjecture, we need to state the second part
of the BirchSwinnerton-Dyer conjecture.
Notation 1.1. Denote the discriminant of K by K . For an elliptic curve E/K,
write E(K)tors for the torsion subgroup of E(K) and

RegE/K = det Pi , Pj
for the regulator of E/K; here {Pi } is any basis of E(K)/E(K)tors and ,
denotes the NeronTate height pairing over K.
Finally, we dene the product CE/K of Tamagawa numbers and periods. Fix
an invariant dierential on E, and let vo denote the Neron dierential at a nite
place v of K. Set

' $$ $$
'
'
CE/K =
cv $$ o $$
||
2
,
v v v|
v|
v
real
E(Kv )
complex
E(Kv )
with cv the local Tamagawa number at v and | |v the nomalised absolute value
denote the individual terms of CE/K by C(E/Kv , ), so
on Kv . We sometimes
&
of but their
that CE/K = v C(E/Kv , ). The terms depend on the choice &
product does not: replacing by with K changes it by v ||v , which
is 1 by the product formula.
204
BirchSwinnerton-Dyer Conjecture II. The TateShafarevich group XE/K is nite, and the leading coecient of L(E/K, s) at s = 1 is
|XE/K | RegE/K CE/K
?
=: BSDE/K .
|E(K)tors |2 |K |
Over general number elds, the only thing known about BSDE/K is that it is
an isogeny invariant, which is a theorem of Cassels [4]. If : E E is an isogeny
over K, then L(E/K, s) = L(E /K, s) ( induces an isomorphism Vl (E) Vl (E ),
and the L-function is dened in terms of Vl ), so in eect this means that BSD II is
compatible with isogenies. Actually, because isogenous curves have the same rank
( induces E(K) Q
= E (K) Q), so is BSD I as well. We will need a slight
generalisation of Cassels theorem, which relies on isogeny invariance of BSD II
for general abelian varieties:
Lemma 1.2. Suppose
that Ki are number elds, Ei /Ki are elliptic curves and ni
&
are integers. If i L(Ei /Ki , s)ni = 1 and all XEi /Ki are nite, then
'
(BSDEi /Ki )ni = 1.
Proof. Taking the terms with ni < 0 to the other side (and renaming the curves
and elds if necessary), rewrite the assumed identity of L-functions as
'
'

L Ei /Ki , s =
L Ej /Kj , s .
i
&
&
Let A = i ResKi /Q Ei and A = j ResKj /Q Ej be the products of Weil restrictions of the curves to Q. These are abelian varieties with L(A/Q, s) = L(A /Q, s),
so A and A have isomorphic l-adic representations (Serre [52, 2.5, Rmk. (3)]),
and are therefore isogenous (Faltings [19]). Their Xs are assumed to be nite and
BSD-quotients are invariant under Weil restriction (Milne [37, 1]) and isogeny
(TateMilne [38, Thm. 7.3]). This proves the claim.
1.3
Parity example
Why is this relevant to the parity conjecture, which concerns only the rst part of
BSD? Assume for the moment niteness of X, and consider the following example.
The two elliptic curves over Q
E:
E :
y 2 + y = x3 + x2 7x + 5,
2
y + y = x + x + 13x + 42,
E = 7 13
E
= 7 13
3
(91b1)
3
(91b2)
are isogenous via a 3-isogeny : E E dened over Q. Choose and =

dx/(2y + 1) to be the global minimal dierentials, so Cp = cp for all p. The curves
1. BirchSwinnerton-Dyer and parity
205
E, E have split multiplicative reduction at 7 and 13, and their local Tamagawa
numbers and the innite periods c = C(E/R, ), c = C(E /R, ) are
c7 = v7 (E ) = 1,
c13 = v13 (E ) = 1;
c = 6.039 . . .
c7
c13
c = 2.013 . . ..
= v7 (E ) = 3,
= v13 (E ) = 3;
In fact, c = 3c (see the computation below for general isogenous curves). Now
BSDE/Q = BSDE /Q by Cassels theorem, but
CE/Q = 1 1 c = c
CE /Q = 3 3 13 c = 3c
A
are not equal,
so some other terms in the BirchSwinnerton-Dyer constant for E and E must

be unequal as well. Because the two are o by a rational non-square factor, and
the conjectural orders of X (as well as of (torsion)2 ) are squares, the regulators
must be unequal! In other words,
RegE/Q
CE /Q |XE /Q | |E(Q)tors |2
= 3 = 1.
RegE /Q
CE/Q |XE/Q | |E (Q)tors |2
If now rk E/Q were 0, then so would be rk E /Q and we would have
RegE/Q = RegE /Q = 1,
contradicting the above. So, assuming niteness of X, we proved that E/Q has
positive rank. In fact, E/Q has odd rank:
Lemma 1.3. Let : E/K E /K be a K-rational isogeny of degree d. Then
RegE/K
= drk E/K (a rational square).
RegE /K
Proof. Write n = rk E/K = rk E /K, and pick a basis
<
;
:= E(K)/torsion = P1 , . . . , Pn , := E (K)/torsion.
Write t : E E for the dual isogeny, so that t = [d]. Then
<
;
dn RegE/K = det dPi , Pj = dett Pi , Pj
<
;
= det Pi , Pj = RegE /K [ : ()]2 .
Returning to our example, we proved that
3=
RegE/Q
= 3rk E/Q ,
RegE /Q
206
so the rank is odd. Because w(E/Q) = (1)3 = 1 (2 split places + 1 innite;

see 3), we showed that
Finiteness of X
Parity Conjecture
for the curve E = 91b1/Q.

This is the main idea behind the proof of Theorem A in general. We will use
equalities of L-functions to deduce relations between rank parities and the Cs,
and then verify by means of local computations that the latter agree with the
required root numbers.
The p-parity conjecture
To prove Theorem A, along the way we will also need slightly ner statements
that give unconditional results for the Selmer ranks of E/K.
Recall that X is an abelian torsion group, and for every prime p its p-primary
part can be written as
XE/K [p ]
= (Qp /Zp )p (a nite p-group of square order).
Denition 2.1. With p as above, the p-innity Selmer rank of E/K is
rkp E/K = rk E/K + p .
If X is nite as expected, then p = 0 and rkp = rk for all p. But the point
is that we can often say something about rkp without assuming niteness of X.
In particular, the following version of the parity conjecture is often accessible:
p-Parity Conjecture. (1)rkp E/K = w(E/K).
This conjecture is known over Q ([12, 28, 39, 41]) and for most elliptic curves
over totally real elds [16, 42, 43]; see Theorem 4.4 below.
Remark 2.2. It is more conventional to dene rkp as follows. Let

Xp (E/K) = HomZp lim Selpn (E/K), Qp /Zp
n
be the Pontryagin dual of the p -Selmer group of E/K. This is a nitely generated
Zp -module, and
Xp (E/K) = Xp (E/K) Zp Qp
is a Qp -vector space, whose dimension is precisely rkp . Both Xp and Xp are (contravariantly) functorial in E and behave well under eld extensions. Specically,
if F/K is a nite Galois extension, its Galois group acts on Xp (E/F ) and there is
a canonical isomorphism of the Galois invariants with Xp (E/K),
Xp (E/K) = Xp (E/F )Gal(F/K) .
(2.3)
2. The p-parity conjecture
207
(The proof is a simple ination-restriction argument; see e.g. [12, Lemma 4.14].)
For p [F : K] this is even true on the level of Xp . Because the same Galois
invariance holds for E(K)Z Qp Xp (E/K), we get the following corollary (which
can also be proved directly).
Corollary 2.4. Suppose F/K is a nite extension of number elds, and E/K is an
elliptic curve. If XE/F is nite, then so is XE/K .
Remark 2.5. The denitions above and Corollary 2.4 also apply to abelian varieties
in place of elliptic curves.
2.1
Proof of Theorem A
To establish Theorem A, we will need to prove the 2-parity conjecture for elliptic curves with a K-rational 2-torsion point and a special case of the 3-parity
conjecture:
Theorem 2.6 (= Theorem 5.1). Let K be a number eld, and E/K an elliptic curve
with a K-rational 2-torsion point O = P E(K)[2]. Then
(1)rk2 E/K = (1)
ord2
CE/K
C
E /K
= w(E/K),
where E = E/{O, P } is the 2-isogenous curve.

Theorem 2.7. Let F/K be an S3 -extension of number elds, and let M and L be
intermediate elds of degree 2 and 3 over K, respectively. For every elliptic curve
E over K,
C
C2
(1)rk3 E/K+rk3 E/M +rk3 E/L = ( 1)
ord3
E/F E/K
CE/M C 2
E/L
= w(E/K) w(E/M ) w(E/L).
These two parity statements are sucient to deduce Theorem A:

Proof of Theorem A. Take F = K(E[2]), so Gal(F/K) GL2 (F2 )
= S3 . By
Corollary 2.4, the 2- and 3-primary parts of XE/k are nite for K k F .
If E has a K-rational 2-torsion point, the result follows from Theorem 2.6. If
F/K is cubic, then rk E/K and rk E/F have the same parity, because E(F ) Q
is a rational Gal(F/K)
= C3 -representation, so its dimension has the same parity
as that of its C3 -invariants. Also, w(E/K) = w(E/F ) (see [33, p. 167], or prove
this directly using the results of 3), so the result again follows.
We are left with the case when Gal(F/K)
= S3 . Let M be the quadratic
extension of K in F and L one of the cubic ones. By the above argument,
rk E/M even
rk E/L even
w(E/M ) = 1,
w(E/L) = 1.
On the other hand, by Theorem 2.7,

rk E/K + rk E/M + rk E/L is even w(E/K) w(E/M ) w(E/L) = 1,
and Theorem A is proved.
208
2.2
Local formulae for the Selmer parity
There are two totally dierent steps in proving Theorems 2.6 and 2.7, one global
and one local. The rst one is to relate the Selmer parity to the products of periods
and Tamagawa numbers (the rst equality in the two theorems), and the second
one is a local computation comparing the latter to root numbers (the second
equality).
The rst one is in the spirit of the example in 1.3. Recall that there we had
two elliptic curves over Q and an equality of L-functions,

L E/Q, s = L E /Q, s ,
that originated from an isogeny E E , say of degree d. Assuming niteness
of X, it implies an equality of the BirchSwinnerton-Dyer quotients BSDE/Q =
BSDE /Q , which reads, modulo squares,
RegE /Q
CE/Q
RegE/Q
CE /Q
(mod Q2 ).
The lattice index argument (Lemma 1.3) shows that the left-hand side is the same
as drk E/Q modulo squares. So we have an expression for the parity of rk E/Q, which
is a dicult global invariant, in terms of easy local data. This is an absolutely crucial
step in all parity-related proofs. It can be made slightly ner, without assuming
niteness of X, but at the expense of working with Selmer groups. The precise
statement is as follows:
Notation 2.8. For an isogeny : A A of abelian varieties over K, we write
t : (A )t At for the dual isogeny. For a prime p, write
p : Xp (A /K) Xp (A/K)
for the induced map on the dual Selmer groups, and
p,v : A(Kv ) A (Kv )
for the map on local points. We let () = | ker()|/| coker()|.
Theorem 2.9. Let : A A be a non-zero isogeny of abelian varieties dened
over a number eld K. For every prime p,
'
(tp )
CA/K
= p-part of
(p,v ) = p-part of
.
(p )
C
A /K
v
The proof is an application of PoutouTate duality (see e.g. [12, 4.1]). For
example, if : E E is an isogeny of elliptic curves of prime degree p, then
(tp )
CE/K
(tp )p (p ) = ([p]p ) = prkp E/K
=
CE /K
(p )
(mod Q2 ),
which is a formula of Cassels (see Birch [2] or Fisher [20]). This extends the
argument of 1.3 and, in particular, Lemma 1.3 to an unconditional statement for
Selmer groups, and proves the rst equality of Theorem 2.6 (with p = 2).
2.3
209
Parity in S3 -extensions
We proved the rst global step of Theorem 2.6. Now we do the same for Theorem 2.7; in other words, we claim the following2 :
Theorem 2.10. Let F/K be an S3 -extension of number elds, M and L intermediate elds of degree 2 and 3 over K, and E/K an elliptic curve. Then
rk3 E/K + rk3 E/M + rk3 E/L ord3
2
CE/F CE/K
2
CE/M CE/L
(mod 2).
As done in 1.3 for the isogeny case, we start with a slightly simpler version
rst, assuming niteness of X. Again, we will use a relation between L-functions,
but this time it is one of a dierent nature.
Thus, suppose that G = Gal(F/K)
= S3 and M and L are as in the theorem.
Let E/K be an elliptic curve for which XE/F is nite. By Corollary 2.4, XE/K ,
XE/M and XE/L are nite as well.
The group S3 has three irreducible representations, namely 1 (trivial), (sign)
and a 2-dimensional representation , all dened over Q. The list of subgroups of
S3 up to conjugacy is

H = 1, C2 , C3 , S3 ,
and they correspond by Galois theory to F , L, M and K, respectively. Each H H
gives rise to a representation C[G/H] of G, associated to the G-action on the left
cosets of H in G. Because there are four subgroups and only three irreducible
representations, there is a relation between these. Writing out
C[G]
= 1 and C[G/C2 ]
= 1 ,
= 1 , C[G/C3 ]
we nd that the (unique such) relation is
C[S3 ] C[S3 /S3 ]2
= C[S3 /C3 ] C[S3 /C2 ]2 .
(2.11)
Now tensor this relation with the l-adic representation Vl (E/K)C = Vl (E/K)Ql C
(embedding Ql C in some way). By the Artin formalism for L-functions (see 3),
for every H H,

L Vl (E/K)C C[G/H], s = L E/F H , s ,
so we have a relation between L-functions3 ,

2
L E/F, s L(E/K, s)2 = L E/M, s L(E/L, s .
2 This
is [12, Thm. 4.11] with p = 3. Note that the contributions from v | to
2
CE/F CE/K
2
CE/M CE/L
cancel when using the same K-rational over each eld. The denition of C in [12] excludes
innite places, so the formula there does not need the
3 The
2
CE/K
2
CE/L
term, as it is then a rational square.
elliptic curve has nothing to do with this: this relation already exists on the level of
Dedekind zeta-functions of the number elds. This forces relations between the regulators and
class groups of number elds, and they have been studied by Brauer, Kuroda, de Smit, Bartel
and others (see e.g. [1] for references).
210
Applying Lemma 1.2, we get a relation between the regulators and the products
of periods and Tamagawa numbers,
RegE/F Reg2E/K
RegE/M Reg2E/L
2
CE/M CE/L
2
CE/F CE/K
(mod Q2 ) .
(2.12)
How do we interpret the left-hand side, and why is it even a rational number? Tensor the MordellWeil group E(F ) with Q and decompose it as a G-representation,
V = E(F ) Q
= 1a b c .
Next, compute the ranks of E over the intermediate elds of F/K in terms of a,
b and c: using Frobenius reciprocity, for a subgroup H of S3 we have
a,
H = S3 ,
<
;
a
+
b,
H = C3 ,
rk E/F H = dim V H = 1, V |H H = C[G/H], V G =
a
+
c,
H
= C2 ,
a + b + 2c, H = {1}.
So, let P1 , . . . , Pa be a basis of E(K) Q (the trivial part), and complement it to
a basis of E(M ) Q with Q1 , . . . , Qb (the sign part) and to a basis of E(L) Q
with R1 , . . . , Rc . By clearing the denominators, we may assume that the P s, Qs
and Rs are actual points in E(F ). If g Gal(F/K) is an element of order 3, then
<
;
E(F ) Q = P1 , . . . , Pa , Q1 , . . . , Qb , R1 , . . . , Rc , R1g , . . . , Rcg ;
in other words, these points form a basis of a subgroup of E(F ) of nite index.
Now we can compute all the regulators. Consider the three determinants

P = det Pi , Pj F , Q = det Qi , Qj F , and R = det Ri , Rj F ,
where , F is the NeronTate height pairing for E/F . Recall that the pairing is
normalised in such a way that it changes if computed over a dierent eld by the
degree of the eld extension. For instance,
;
Pi , Pj
<
K
<
1;
Pi , Pj F ,
6
so RegE/K is ( 16 )a P , where is the inverse square of the index of the lattice

spanned by the Pi in E(K). As we are only interested in regulators modulo squares,
we may ignore this. Similarly,
a+b
a+c
1
1
RegE/M =
PQ and RegE/L =
PR ,
3
2
because all Pi , Qj F and Pi , Rj F are 0, so both regulators are really products
of two determinants. (The height pairing is Galois-invariant, so dierent isotypical
211
components are always orthogonal to each other with respect to it.) Finally, using
2
Galois invariance again, together with the fact that Ri + Rig + Rig is S3 -invariant
and so orthogonal to Rj , we nd that Ri , Rjg F = 12 Ri , Rj F , so
3
RegE/F = P Q det
A
12 A
4
12 A
A
= P Q 3c R2 ,
where A is the matrix (Ri , Rj )i,j . Combining the four regulators yields
RegE/F Reg2E/K
RegE/M Reg2E/L
3c PQR2 (6a P)2

3a+b+c
3a+b PQ (2a+c PR)2
3a+(a+b)+(a+c)
3rk E/K+rk E/M +rk E/L
(mod Q2 ).
Together with (2.12), this proves Theorem 2.10, assuming niteness of X.
2.4
Brauer relations and regulator constants
The results of 2.3 generalise to arbitrary Galois groups as follows (see [12, 14] for
details).
Denition 2.13. Let G be a nite group, and denote by H a set of representatives
of subgroups of G up to conjugacy. A formal linear combination

ni Hi , with ni Z, Hi H,
=
i
ni
is a Brauer
= 0 as a virtual representation, i.e., if the
relationGif i C[G/Hi ]
character i ni IndHi 1Hi is zero.
Example 2.14. The dihedral group G = D2p for an odd prime p has a relation
= {1} 2C2 Cp + 2 G,
the only one in G up to multiples. For p = 3 this is the relation (2.11).
If G = Gal(F/K)
is a Galois group and E/K is an elliptic curve, every
Brauer relation =
ni Hi in G gives an identity of L-functions
'
L(E/Mi , s)ni = 1, with Mi = F Hi ,
i
and so a relation like (2.12) between regulators and Tamagawa numbers,

assuming
&
that XE/F is nite. As in the case G = S3 , the left-hand side i (RegE/Mi )ni
depends only on E(F ) Q as a G-representation:
212

Theorem 2.15. Let G be a nite group and =
ni Hi a Brauer relation in G.
Suppose that V is a QG-representation and , : V V R is a non-degenerate
bilinear G-invariant pairing 4 . Then

'
$ H ni
1
i
$
, V
C (V ) :=
det
|Hi |
i
is a well-dened number in Q /Q2 , and it is independent of the choice of the
pairing , .

$
The notation det |H1i | , $ V Hi means the following: pick a basis {Pj } of
the invariant subspace V Hi and compute the determinant of the matrix whose
entries are |H1i | Pj , Pk . Up to rational squares, this is independent of the basis
choice and the total product is well-dened in R /Q2 . The theorem asserts that
it is in fact in Q /Q2 , and independent of , .
Remark 2.16. The theorem also holds with Q replaced by any other eld k where
|G| is invertible, V by a self-dual kG-representation, and R by any eld containing k.
The proof of the theorem is reasonably straightforward (see [12, 2] or [14,
2.ii]), in the spirit of what we did in 2.3 explicitly for G = S3 . An immediate
consequence is that if 1 , . . . , k are all irreducible QG-representations, then the
regulator constants, determine everything. In
numbers C (1 ), . . . , C (k ), called5
other words, if we decompose V =
j aj , then
'
C (j )aj ,
C (V ) =
j
as we can obviously pick a diagonal pairing on V which respects the decomposition.

Example 2.17. For G = S3 and = {1} 2C2 C3 + 2 S3 , we have
C (1) = C () = C () = 3 Q /Q2 .
(For example, if we choose the obvious trivial pairing on 1, then
2 1 1 1 2
C (1) = 1 12
3
6 3 (mod Q2 ),
and similarly for and .)
Corollary 2.18. Suppose that E/K
is an elliptic curve, F/K a Galois extension
with Galois group
G,
and
=
ni Hi a Brauer relation in G. If we decompose
5 ak
E(F ) Q =
k into irreducible G-representations, then
'
'
(RegE/F Hi )ni =
C (k )ak (mod Q2 ).
i
4 It
exists for every such V , as every rational representation is self-dual.

Proof. Take V = E(F ) Q and , the NeronTate height pairing on V .
213
For G = S3 , together with (2.12), this corollary once again proves Theorem
2.10, still assuming that XE/F is nite.
Finally, we discuss the modication necessary to turn this into an unconditional statement about Selmer ranks. Suppose that E/K is an elliptic curve,
Gal(F/K) = G as before and p is a prime. Then the dual Selmer Xp (E/F ) (cf.
Remark 2.2) is a Qp G-representation which now plays a role analogous to the
QG-representation E(F ) Q.
Theorem 2.19. (1) X = Xp (E/F ) is a self-dual Qp G-representation. In other
words, it possesses a G-invariant Qp -valued non-degenerate bilinear pairing.

(2) For any such pairing , and a Brauer relation =
ni Hi in G, we have
3
4
n i
'
'
$
1
, $ X Hi
ordp
det
ordp (CE/F Hi )ni (mod 2).
|Hi |
i
i
For the proof, see [13]. In fact, for an explicit group such as G = S3 , it is possible to give a direct proof of this, by constructing an isogeny between the products
of the relevant Weil restrictions and applying Theorem 2.9; this is how it is done
in [12, Thm. 4.11] (with p = 3). Note that, when G = S3 , every G-representation
is rational, so (1) is automatic.
As a corollary, we get Theorem 2.10. (Decompose X3 (E/F ) into Q3 S3 -irreducibles and apply the theorem above.) So we have proved the rst equalities in
Theorems 2.6 and 2.7, and it remains to prove the second ones. For that we need
to look carefully at the root numbers of elliptic curves and carry out the local
computations relating them to Tamagawa numbers.
2.5
Parity in dihedral extensions
The S3 -example generalises to dihedral groups. Suppose G = Gal(F/K) = D2p ,

with p an odd prime, and recall from Example 2.14 that G has a unique Brauer
relation
= {1} 2C2 Cp + 2 G.
Let M and L be the unique quadratic and one of the degree p extensions of K in F ,
respectively. Apply Theorem 2.19 and interpret the left-hand side using regulator
constants (using Theorem 2.15 and Remark 2.16 with k = Qp ). Take any elliptic
curve E/K, and decompose
Xp (E/F ) = 1n1 n n ,
where 1 (trivial), (sign) and (p1-dimensional) are the distinct Qp G-irreducible
p as a sum of 2-dimensionals,
representations; the latter decomposes over Q
p
Q
= 1 12 (p1) .
214
The regulator constants of 1, and are easily seen to be p (as in the case G = S3 ),
and Theorem 2.19 gives the following parity statement:
n1 + n + n ordp
2
CE/F CE/K
2
CE/M CE/L
(mod 2).
This can also be written as

1 + + i , Xp (E/F ) ordp
2
CE/F CE/K
2
CE/M CE/L
(mod 2) for all i,
where , stands for the usual inner product of characters of representations.

In other words, the relation gives a p-parity expression for the twist of E by
1 + + i in terms of Tamagawa numbers and periods.
Remark 2.20. For an alternative expression for exactly the same parity in terms of
local constants, see Mazur and Rubins papers [34, 35]. The parity conjecture for
these twists is now known for all elliptic curves over number elds and all odd p;
see [12, 16] and de La Rochefoucauld [47].
2.6
The KramerTunnell theorem
We mentioned two types of relations between L-functions of elliptic curves: one

comes from a rational isogeny, and one from Brauer relations in Galois groups.
There is a third example, which
is classical: the relation for quadratic twists.
Suppose that M = K( ) is a quadratic extension of number elds, E/K is
an elliptic curve, and E /K is the quadratic twist of E by :
E : y 2 = x3 + ax + b,
E : y 2 = x3 + ax + b (
= y 2 = x3 + 2 ax + 3 b).
The l-adic Tate modules of E and E are related by

Tl (E ) = Tl (E) ,
with : Gal(M/K) {1} the non-trivial character, and the Artin formalism of
Gal(K/K)
L-functions (see 3) applied to IndGal(K/M
1 = 1 yields a relation
)
L(E/M, s) = L(E/K, s) L(E /K, s).
If we assume that XE/M is nite, then Lemma 1.2 gives
RegE/M
CE/K CE /K
RegE/K RegE /K
CE/M
(mod Q2 ),
and it is easy to see that the left-hand side is 2rk E/M (i.e., 2rk E/K+rk E /K ) up to
squares. In fact, the Weil restriction A = ResM/K E admits an isogeny
: A E E
3. L-functions and root numbers
215
such that t = [2], and Theorem 2.9 produces an unconditional version:

rk2 E/M ord2
CE/K CE /K
CE/M
(mod 2).
(2.21)
This was used by Kramer [32] and KramerTunnell [33] to prove the 2-parity conjecture for E/M , by comparing the right-hand side with the root number w(E/M )
by a local computation:
Theorem 2.22 (KramerTunnell). If the primes of additive reduction for E above 2
are unramied in M/K, then the 2-parity conjecture holds for E/M :
(1)rk2 E/M = w(E/M ).
The restrictions on the reduction type can in fact be removed using the
methods of 5.7. We will not need this, so we refer the reader to [16] (proof of the
KramerTunnell conjecture).
L-functions and root numbers
To set up the notation, let K be a number eld and p a prime of K with completion Kp , residue eld Fq of characteristic p, and uniformiser . We write GK for
the absolute Galois group Gal(K/K)

of K, and use a similar notation for other
elds as well. For most of this section we work in the local setting, and we begin
by recalling the structure of the local Galois group at p. The reduction map on
automorphisms puts GKp into an exact sequence
mod p
1 Ip GKp GFq 1,
which denes the inertia group Ip at p. An (arithmetic) Frobenius is any element
Frobp GKp whose reduction mod p is the map x xq . If we choose an embedding
K p , then GK GK via the restriction map, and we can consider Ip as
K
p
a subgroup of GK . Choosing a dierent embedding conjugates Ip , so it is really
only well-dened up to conjugation. Similarly, we can view Frobp as an element
of GK , but it is only dened up to conjugation, and only modulo inertia.
We will
standard characters that come from identications
& often use two
wild &
GKp /Ip
= Zl and Ip / inertia
= l=p Zl . Fix a prime l p.
Denition 3.1. The (l-adic) cyclotomic character : GKp Z
l is dened by
(Ip ) = 1 and (Frobp ) = q. Alternatively, it is the action of GKp on the l-power
roots of unity,
lim (Z/ln Z)
: GKp lim Aut ln =
= Z
l .
n
n
(In this way it can be dened as a character of GK .)
216
Denition 3.2. The (l-adic) tame character : Ip Zl is dened by

n
n
() = ( 1/l )/ 1/l lim ln
= Zl .
n
Recall that a GKp -module M is unramied if Ip acts trivially on M , and,
similarly, a GK -module is unramied at p if Ip < GK acts trivially on it.
Example 3.3. The trivial character 1 and the cyclotomic character both give Zl
a structure of an unramied GKp -module.
3.1
L-functions
Let E/K be an elliptic curve. For every prime l, the Galois group GK = Gal(K/K)
n
n
n
acts on the sets E[l ] = E(K)[l ] of l -torsion points of E for all n 1. The
fundamental arithmetic invariant of E/K is its l-adic Tate module,
Tl E = lim E[ln ],
n
the limit taken with respect to multiplication by l, mapping E[ln+1 ] E[ln ]. This
is a free Zl -module of rank 2, and
Vl E = Tl E Zl Q l
is a 2-dimensional Ql -vector space. By the non-degeneracy and the Galois equivariance of the Weil pairing on E[ln ], we have det Vl E = . The representation Vl E
(or sometimes its dual Vl E ) is called the l-adic representation associated to E/K.
For varying l these form a compatible system of l-adic representations, meaning
that they satisfy two conditions:
(1) All Vl E are unramied at p for almost all primes p of K.
(2) For such p, the characteristic polynomial of the Frobenius Frobp on Vl E is
independent of l for p l.
To explain this, take a prime p of K and any l q. The criterion of Neron
OggShafarevich ([54, Ch. VII]) asserts that
Vl E is unramied at p
E has good reduction at p.
In particular, this happens for almost all primes p of K and it is independent of l.

This is condition (1).
Now, as Ip GKp and Vl E is a GKp -representation, the quotient GKp /Ip acts
on the inertia invariants (Vl E)Ip , so we can talk of the action of the Frobenius
Frobp on (Vl E)Ip , and this is independent of the choice of Frobp . Its characteristic
polynomial does not change under conjugation, therefore it is completely choiceindependent. So we may dene the local polynomial at p,

$
Ip
$
Fp (T ) = det 1 Frob1
.
p T (Vl E )
217
(There are two technical points: we take the geometric Frobenius Frob1
p , the
inverse of the arithmetic one, and we also compute the characteristic polynomial
on the dual Vl E ; both are just standard conventions, and are not too important.)
As explained in [54, Ch. V],
E/Kp has good reduction
Fp (T ) = 1 ap T + qT 2 ,
with ap = q + 1 |E(Fq )|. This polynomial has degree 2, since Vl E = (Vl E)Ip in
the good reduction case. Also, Fp (T ) is in Z[T ] rather than just Zl [T ] and it is
independent of l. This is condition (2).
Denition 3.4. The L-function of E/K is a function of a complex variable s given
by the Euler product
L(E/K, s) =
'
p
Fp (qps )1 =
'

1
$
$ (Vl E )Ip
det 1 qps Frob1
.
p
Here the two products run over all primes p of K, and qp denotes the size of the
residue eld at p. (It follows from the HasseWeil bound that L(E/K, s) converges
for Re s > 3/2.)
In the same way one associates an L-function to any compatible system
= (l )l of l-adic representations. There are obvious notions of direct sums and induction for l-adic representations and it is not hard to verify that their L-functions
satisfy the following:
(i) If , are compatible systems of l-adic representations of GK , then
L( , s) = L(, s) L( , s).
(ii) For as above and a subeld F K,

F
L IndG
GK , s = L(, s).
These properties are known as the Artin formalism for L-functions.
As for elliptic curves, we expect all L-functions of systems of l-adic representations to have meromorphic (and usually analytic) continuation to the whole
of C, and to satisfy a functional equation relating the value at s to that at k s,
where k Z is the weight of .
Remark 3.5. Most L-functions that we know of are supposed to arise from l-adic
representations. For instance, if V /K is any non-singular projective variety and 0
i 2 dim V , the etale cohomology groups Heit (V, Ql ) form a compatible system.
Thus we can dene the corresponding L-function L(H i (V ), s) and it has weight
k = i + 1. In this terminology, L(E/K, s) = L(H 1 (E/K), s). (This the reason for
taking the dual Vl E instead of Vl E in the denition of L(E/K, s).)
218
There is one subtle point, though: for varieties other than curves and abelian
varieties, it is not known that the polynomials Fp (T ) are independent of l when p
is a prime of bad reduction. This is conjectured to be true, and this conjecture is
implicitly assumed when one speaks of L-functions of general varieties.
Remark 3.6 (Conductors). Another arithmetic invariant that enters the functional
equation of any L-function is its conductor N . For an elliptic curve E/K, the
conductor of L(E/K, s) is N = 2K NormK/Q (NE/K ). Here K is the discriminant
of K and NE/K is the conductor of E/K. It is an ideal in OK (see [55]), and we
interpret its norm to Q as a positive integer. It is this N that enters the functional
equation in 1.1.
For elliptic curves, let us determine Fp (T ) in all cases and verify its independence of l explicitly. There are several cases:
E/Kp has good reduction
As mentioned above, Fp (T ) has degree 2, is independent of l and is of the form
Fp (T ) = 1 ap T + qp T 2 .
E/Kp has split multiplicative reduction
In this case, Tates uniformization (theory of the Tate curve) asserts that there
is a (p-adic analytic) isomorphism of GKp -modules
E(K p )
= K p /aZ
for some a Kp with vp (a) = vp (j(E)) > 0. Writing ln K p for a primitive

ln th root of unity, we get that, in particular,
;
n<
E[ln ]
= ln , a1/l K p /aZ
as a Galois module. In this basis, the action of Galois on E[ln ] is of the form
3
4
() vp (a)
GKp !
0
1
with the l-adic cyclotomic character and restricting to the tame character
on Ip . Passing to the inverse limit Tl E = lim E[ln ] and tensoring with Ql we nd
that the actions on Vl E and Vl E are

3
4
4
3
()
1 () 0

,
and
1
0
1
219

1 0
on Vl E , so it has a 1-dimensional inrespectively. The inertia Ip acts as
1
variant subspace, spanned by the second basis vector. The Frobenius acts trivially
on it, so

$
Ip
$
Fp (T ) = det 1 Frob1
= 1 T.
p T (Vl E )
E/Kp has non-split multiplicative reduction
Let Kp ( )/Kp be the unramied quadratic extension, and

?
: GKp Gal(Kp ( )/Kp ) {1}

the associated character of order 2. The quadratic twist E of E has split-multiplicative reduction and Vl E = Vl (E ) as a Galois module. In other words, GKp
acts on Vl E as
1

()()
0

.
()
Because Ip acts in the same way as in the split multiplicative reduction case and
(Frobp ) = 1, we nd that the inertia invariants are again 1-dimensional and
Fp (T ) = 1 + T.
E/Kp has additive potentially multiplicative reduction
This is similar to the non-split multiplicative case, except that here K()/K is
replaced by a ramied quadratic extension (namely, the smallest
extension
where
1 0
E acquires split multiplicative reduction). Now Ip acts through
on Vl E
1
with non-trivial , and has therefore trivial invariants. So
Fp (T ) = 1.
E/Kp has additive potentially good reduction
As in the last case, we claim that (Vl E )Ip = 0 and consequently
Fp (T ) = 1.
Recall that E acquires good reduction after a nite extension of Kp . By the Neron
OggShafarevich criterion, Ip acts on Vl E non-trivially and through a nite quotient. If
Vl E had a 1-dimensional inertia invariant subspace, then Ip would act as

1
in some basis. The bottom right corner is 1 as det Vl E = is unramied,
0
and the top-right corner is then 0 since an action of a nite group is diagonalizable.
So Ip would act trivially, a contradiction. Therefore (Vl E )Ip = 0, as asserted.
220
3.2
Whatever the reduction type of E/Kp is, note that Ip has an open (i.e., nite
index) subgroup which
acts
unipotently on Vl E, namely trivially in the potentially

1
good case and as
in the potentially multiplicative case.
0 1
Grothendiecks monodromy theorem asserts that for any non-singular projective variety V and any i, the action of some open subgroup of Ip on = Heit (V, Ql )
is unipotent. It follows that it has the form 1 + N for some xed nilpotent endomorphism N of . We are going to call such representations Weil representations
(N = 0) and WeilDeligne representations (any N ). There are two technical points:
one is that we x some embedding Ql C and make our representations complex
instead of l-adic from this point onwards; another one is that, to get rid of the
dependence of the l-adic tame character of l, we just remember the nilpotent
endomorphism N and how it commutes with the Galois group.
Denition 3.7. The Weyl group WF is the subgroup of GF generated by the inertia group IF /F and (any) Frobenius element. A Weil representation over F of
dimension n is a homomorphism WF GLn (C) whose kernel contains a nite
index open subgroup of the inertia group IF /F .
Denition 3.8. A WeilDeligne representation over F is a Weil representation
: WF GLn (V ) together with a nilpotent endomorphism N End(V ) such
that (g)N (g)1 = (g)N for all g WF .
Example 3.9. If E/Kp is an elliptic curve, then Vl E C and (Vl E ) C have a
natural structure of WeilDeligne representations. They are Weil representations
if and only if E has integral j-invariant (equivalently, E has potentially good
reduction).
3.3
Epsilon-factors
The current state of aairs is that we are very far from proving the HasseWeil
conjecture for compatible systems of l-adic representations or even for elliptic
curves over number elds. Even for elliptic curves over Q, the proof (via modularity) is rather roundabout. In some sense, the only well-understood situation
is the 1-dimensional case, that of Hecke characters (a.k.a. Grossencharaktere).
Also well-understood are the signs in the conjectural functional equations for all
L-functions. This is the theory of -factors, which we now sketch.
We will not need Hecke characters themselves, but only their local components. Let p be a prime of K, and denote by F some nite extension of Kp .
Denition 3.10. A quasi-character of WF is a one-dimensional WeilDeligne (equivalently, Weil) representation. Alternatively, it is a homomorphism
: WF C
221
which is continuous with respect to the pronite topology on WF and discrete

topology on C .
In his thesis, Tate associated to a quasi-character its epsilon-factor
() C ,
which enters the functional equation for the local L-function of :
Notation 3.11. Composing with the local reciprocity map F Gab
F , consider
also as a character F C . Dene
n() = the conductor exponent of ,
b(F ) = vp (F/Qp ),
n()b(F )
h = any element of F of valuation n() b(F ), e.g. F

(x1 )e2i TrF /Qp (x) dx

for ramied,
hOF

() =
(h1 )
(h1 ) dx =
dx for unramied.
|h|F
hOF
OF
The integrals are in eect nite sums (with the number of terms growing with
the conductor of and the ramication of F over Qp ), so they can be explicitly
computed for a given quasi-character.
Tates theory of signs in the functional equations extends uniquely from quasicharacters to arbitrary WeilDeligne representations
: WF GLn (C),
for all nite F/Kp , and all n:
Theorem 3.12 (LanglandsDeligne [8]). There is a unique way to associate to each
its epsilon-factor () C with the following properties:
(1) (Multiplicativity.) (1 2 ) = (1 )(2 ).
(2) (Inductivity in degree 0.) If 1 , 2 : WF GLn (C) have the same degree,
then5
W

K
IndWF p 1
(1 )
.
= W
K
(2 )
IndWF p 2
(3) (Quasi-characters.) For quasi-characters : WF C , the () are as given
in 3.11.
5 Equivalently,
(W ) = (Ind W ) for any virtual representation W of degree 0.
222
(4) (Semi-simplication.) Writing ss for the semi-simplication of ,

det Frobp |(ss )Ip
.
() = ( )
det ( Frobp |Ip )
ss
Remark 3.13. Uniqueness follows from the Brauer induction argument: every
semisimple WeilDeligne representation is a Z-linear combination of inductions of
quasi-characters. Existence is harder: one has to understand monomial relations
between inductions and use Stickelbergers theorem to prove that the -factors
satisfy those relations.
Denition 3.14. Write sgn z = z/|z| for the sign of z C on the complex unit
circle. The local root number of is dened as
w() = sgn () =
()
.
|()|
Example 3.15. For = a 1-dimensional unramied quasi-character,

b(F )
w() = sgn h1 = sgn Frobp
.
In particular, the trivial representation has w(1) = 1.
Example 3.16. Writing q for the size of the residue eld of F , we nd that the
cyclotomic character also has
w() = (sgn q)... = 1.
Because the -factors are multiplicative in direct sums and inductive in degree 0, clearly so are the root numbers; similarly,

ss Ip
ss sgn det Frobp |( )
w() = w( )
sgn det ( Frobp |()Ip )
as well. Here are some additional properties that are not hard to deduce:
Proposition 3.17 (Tate [61], Deligne [8]).
(1) w( ) = (det )(1), i.e., the image of 1 under
loc. recip.
det
F Gab
C .
F

n( )+b(F ) dim 1
if 2 is unramied.
(2) w(1 2 ) = w(1 )dim 2 sgn(det 2 ) F 1
3.4
223
Root numbers of elliptic curves
Denition 3.18. Let E/Kp be an elliptic curve. Dene its local root number as

w E/Kp = w(), where = Vl E Ql C.
For elliptic curves (and, generally, for abelian varieties) this is known to be
independent of l (with p l) and of the embedding Ql C, and it equals 1.
Denition 3.19. The global root number of an elliptic curve E dened over a
number eld K is the product of the local root numbers over all places of K,

'
w E/Kv .
w E/K =
v
The product is nite since w(E/Kv ) = 1 for primes of good reduction (see below).
Also, we let w(E/Kv ) = 1 for all Archimedean v.
Example 3.20 (Good reduction). If E/Kp has good reduction, then is unramied
by the NeronOggShafarevich criterion. Since det is the cyclotomic character,

w E/Kp = w 1 = w(1)2 sgn det F = sgn q = +1.
Example 3.21 (Split multiplicative reduction). If E/Kp has split multiplicative
reduction, then
0
0
ss
, =
, =
,
Vl E =
0 1
1
0
1
and det = as before. Applying the semi-simplication formula, we nd
0
sgn det Frobp |
1
0
1
0
w() = w
0
1
sgn det(Frobp |1)

q 1 0
sgn det
0
1
1
=
= 1.
=
sgn det (1)
1
Example 3.22 (Non-split multiplicative reduction). If E/Kp has non-split multiplicative reduction, then the same computation shows that w(E/Kp ) = +1.
To compute the root numbers of elliptic curves in the additive reduction case
one has to understand well enough. One case when this works well is when E
admits an isogeny, forcing the Galois action on Vl E to be less complicated than
in general.
224
3.5
Root numbers of elliptic curves with an l-isogeny
As before, let E/Kp be an elliptic curve, and suppose that p l. Assume that there
is a degree l isogeny dened over Kp ,
: E E .
Recall that is a non-constant morphism of curves mapping O to O, and over
K p it is an l-to-1 map; such a map automatically preserves addition on E, and
therefore maps E[ln ] to E [ln ], and hence Tl E to Tl E as well.
Notation 3.23. Write ker for the abelian group of points in the kernel of in
E(K p ); thus ker
= Z/lZ as an abelian group. Write also

F := Kp coordinates of points in ker Kp E[l] .
The points in ker are permuted by GKp , so F/Kp is a Galois extension.
This
also
means that the action of GKp on E[l] is reducible, that is of the form

. The Galois group Gal(F/Kp ) is the image of
0
.
/
action on E[l]
top left corner

F
GKp
l ,
0
so it is cyclic of order dividing l 1.
Notation 3.24. Write (1, F/Kp ) for the Artin symbol
.
+1 if 1 is a norm from F to Kp ,
(1, F/Kp ) =
1 otherwise.
The Artin symbol is the main class-eld-theoretic invariant for cyclic extensions, and the local root number of an elliptic curve with an isogeny turns out to
be closely related to it:
Theorem 3.25. If E/Kp has additive reduction, p l and l 5, then

w E/Kp = 1, F/Kp .
Proof. We leave the potentially multiplicative case to the reader, and do the
(slightly more involved) case of potentially good reduction.
Claim 1: E[l] is unramied over F . Proof: Because det = and GF acts
trivially

1
on ker , the image of GF in Aut(E[l]) = GL2 (Fl ) is contained in
. As
0

1
, so |I| divides l. But
is unramied, the image I of inertia IF is inside
0 1
|I| divides 24 in the potentially good case, so I is trivial and E[l] is unramied
over F .
225
Claim 2: E[ln ] is unramied over F for all n 1. Proof: The image In of IF

in GL2 (Z/ln Z) is of order dividing 24, as before. But the kernel of the map
GL2 (Z/ln Z) GL2 (Z/lZ) is of l-power order, so this map is injective on In .
But by Claim 1 the image is trivial, so In is trivial as well.
Claim 3: E/F has good reduction. Proof: NeronOggShafarevich.
Claim 4: The action of GKp on Vl E is abelian. Proof: We want to show that the
commutator subgroup GKp acts trivially on Vl E. Now,
is abelian, and
GKp IKp as GKp /IKp
= Gal(k/k)
=Z
the image of GKp in Gal(F/Kp ) is trivial as Gal(F/Kp ) is abelian,
so GKp IF . But IF acts trivially on Vl E by Claim 2, as asserted.
Step 5: The local
A semisimple abelian action can be diagonalised
root number.

0
over C, so =
for some quasi-character . Now we can compute
0 1
w() = w() w(1 ) = w() w( )

= w = (1)
= (1)

= 1, F/Kp .
(unramied twist formula)

( formula)
for any primitive character
of Gal(F/Kp ) that agrees
with on inertia
(local class eld theory)
Remark 3.26. The theorem illustrates the fact that, when GKp acts on Vl E through
an abelian quotient, we have enough formulae to determine the local root number.
When the action is not abelian, the following result is very useful:
Theorem 3.27 (Fr

ohlichQueyrut). Suppose that F ( )/F is a quadratic extension
and : GF () C is a quasi-character such that |F = 1. Then

w() = () {1}.
Via the induction formula, the theorem computes the local root number of
the Galois representation w(Ind ). This is a 2-dimensional representation which
is not necessarily abelian, e.g. the Galois image may be dihedral or quaternion
(it has a cyclic subgroup of index 2). This is enough to determine the local root
numbers of all elliptic curves over elds of residue characteristic at least 5 (i.e.,
p 2, 3); see [48].
In residue characteristics 2 and 3 the situation is a bit more complicated;
see [26] for Q2 and Q3 , [30] for p | 3 and [15, 65] for p | 2, and also [16] for a
general root number formula. See also [48, 49] for the root numbers of twists of
elliptic curves.
226
Parity over totally real elds
We refer to Wintenberger [66] for the denition of modularity for elliptic curves
over totally real elds. The two propositions below are essentially due to Taylor
(see [62, Proof of Thm. 2.4] and [63, Proof of Cor. 2.2]).
Proposition 4.1. Let F/K be a cyclic extension of totally real elds, and E/K an
elliptic curve. If E/F is modular, then so is E/K.
Proof. Let be the cuspidal automorphic representation associated to E/F . Pick
a generator of Gal(F/K). Then = and therefore, by Langlands base
change theorem, descends to a cuspidal automorphic representation over K.
Associated to there is a compatible system of -adic representations , , with
varying over the primes of some number eld k (see [66]). Fix a prime of k, let l be
the rational prime below it and write Vl (E/F ) for the Tate module of E/F tensored
with Ql . The restriction of = , to Gal(F /F ) agrees with V = Vl (E/F )Ql k .
Because E cannot have complex multiplication over F (it is totally real), Vl (E/F )
is absolutely irreducible, and therefore the only representations that restrict to V
are W = Vl (E/K) Ql k and its twists by characters of Gal(F/K). Hence and
W dier by a 1-dimensional twist, whence W is also automorphic and E/K is
modular.

Proposition 4.2. Let E be an elliptic curve over a totally real eld K, and p a prime
number. If the p-parity conjecture for E is true over every totally real extension
of K where E is modular, then it is true for E/K.
Proof. Because E is potentially modular (see [66, Thm. 1]), there is a Galois totally real extension F/K over which E becomes modular. By Solomons induction
theorem, there are soluble subgroups Hi < G = Gal(F/K) and integers ni such
that the trivial character 1G can be written as

ni IndG
1G =
H i 1H i .
i
Hi
Write Ki for F . Since Gal(F/Ki ) = Hi is soluble, a repeated application of

Proposition 4.1 shows that E/Ki is modular. By Artins formalism for L-functions,
ni

'
L E/Ki , s .
L E/K, s =
i
On the other hand, writing X = Xp (E/F ), for every H < G we have

<
<
;
;
rkp E/K H = dim X H = X , 1H H = X , IndG
H 1H G .
The rst equality is (2.3) and the last one is Frobenius reciprocity.
By assumption, the p-parity conjecture holds for E/Ki . Therefore

<
;
ni rkp E/Ki
rkp E/K = X , 1G G =

i
ni ords=1 L(E/Ki , s) = ords=1 L(E/K, s)
(mod 2).
4. Parity over totally real elds
227
Remark 4.3. The proof also shows that L(E/K, s) has a meromorphic continuation
to C, with the expected functional equation. This is the same argument as in [63,
Proof of Cor. 2.2].
Theorem 4.4 ([12, 16, 42]). Let K be a totally real eld, and E/K an elliptic
curve with non-integral j-invariant. Then the p-parity conjecture holds for E/K
for every prime p.
Proof. Let p be a prime of K with ordp j(E) < 0. If E has additive reduction
at p,
it becomes multiplicative over some totally real quadratic extension K( ), and
the quadratic twist E /K has multiplicative reduction at p as well. Because

w E/K( ) = w E/K w(E /K) and
(4.5)

rkp E/K = rkp E/K + rkp E /K,
it suces to prove the theorem for elliptic curves with a prime of multiplicative
reduction. Since multiplicative reduction remains multiplicative in all extensions,
by Proposition 4.2 we may also assume that E is modular.
By
FriedbergHosteins theorem [21, Thm. B], there is a quadratic extension
M = K( ) of K with a prescribed behaviour at a given nite set of primes of K,
such that the quadratic twist E has analytic rank 1. If we require that the
multiplicative primes are unramied, then E also has a prime of multiplicative
reduction. By Zhangs theorem ([67, Thm. A]) that generalises the GrossZagier
formula over the rationals, XE /K is nite and the MordellWeil rank of E /K
agrees with its analytic rank; in particular, the p-parity conjecture holds for E /K.
To prove it for E/K we just need to show it for E/M ; cf. (4.5).
Suppose that p = 2. Fix an invariant dierential for E/K, and choose M
so that all innite places, all bad places for E, and the ones where is not minimal
are split in M/K. By (2.21) in 2.6,
rk2 E/K(
?
CE/K() CE /K
) ord2
0
CE/K
(mod 2),
since the other nite primes do not contribute to C. For the same reason, we have
w(E/M ) = (1)2 = 1; see 8, where this is spelled out in more detail, or apply
Theorem 2.22. So the 2-parity holds for E/M , as required.
Finally, suppose that p > 2. Then choose M so that (a) all bad places for
E except one multiplicative place p are split; (b) p is inert, and (c) all real places
are inert, i.e., M is totally complex. Let F /M/K be the p-adic anticyclotomic
tower; thus F is the largest Galois extension of K containing M such that G =
Gal(F /K) is of the form Zrp C2 with C2 acting by 1. The Artin representations
of G are 1 and , which factor through Gal(M/K), plus two-dimensional ones of
r
the form = IndK

M for non-trivial : Zp C . Such a factors through a
dihedral extension F/K in F , and by 2.5 we have
;
<
CE/F
1
1 + + , Xp (E/F ) ordp
CE/M
(mod 2),
228
where the last equality uses again the fact that all bad primes for E and all innite
primes split into pairs in M (and therefore in F as well), except for p, which gives
the 1. For the same reason, we have
w(E/K) w(E/K, ) = w(E/M, ) = 1.
We already know the p-parity conjecture for E ,
w(E/K, ) = w(E /K) = (1)ord2 rk2 E /K = (1), Xp (E/F ) ,
so to show the p-parity conjecture for the trivial twist we just need to verify it
for a single non-trivial . Precisely such a character is provided by the anticyclotomic theory: a combination of CornutVatsals [7, Thm. 4.2] with Nekovars [40,
Thm. 3.2] gives a such that the -component of Xp (E/F ) has multiplicity 1,
which is an odd number, as required.

Remark 4.6. The same idea lies behind the proof of the p-parity conjecture over Q;
see [39] (p = 2) and [12] (odd p) for details.
The 2-isogeny theorem
Theorem 5.1. Let K be a number eld and let E/K be an elliptic curve with
E(K)[2] = 0. Then

(1)rk2 E/K = w E/K .
In particular, niteness of X implies the parity conjecture for E.
The proof will occupy the whole of 5.
Notation 5.2. Fix a 2-torsion point 0 = P E(K)[2], and let : E E be a
2-isogeny whose kernel is {O, P }. Furthermore, translate P to (0, 0), so that E
and E become
(Ea,b =)
E : y 2 = x3 + ax2 + bx,
a, b OK ,
E : y 2 = x3 2ax2 + x,
= a2 4b,
with : E E given by
: (x, y) (x + a + bx1 , y bx2 y).
Notation 5.3. Denote
(E/Kv ) = (1)ord2
| coker v |
| ker v |
= (1)1+ord2 | coker v | ,
where v is the induced map on local points v : E(Kv ) E (Kv ). (Note that
ker v = {O, (0, 0)} is always of size 2.)
5. The 2-isogeny theorem
229
Strategy. Recall from Theorem 2.9 and the denition of the global root number
3.19 that
'
'
(1)rk2 E/K =
(E/Kv ) and
w(E/Kv ) = w(E/K),
(5.4)
v
so a natural strategy is to make a term-by-term comparison at all places.

Write F = Kv for some completion of K.
5.1
Complex places
Suppose F = C. Because it is algebraically closed, v : E(C) E (C) is surjective,

that is | coker v | = 1. Thus,
w(E/F ) = 1 = (E/F ).
5.2
Real places
Suppose F = R. It turns out, somewhat surprisingly, that w(E/R) and (E/R) may not
be equal: w(E/F ) = 1 as in the complex case
but v : E(R) E (R) is not always surjective.
For varying a, b R the picture is given on the
right.
= 0 (a2 = 4b)
w =
-
w =
To see this, consider the structure of the group of real components for E
and E ; recall that the group of real points is connected if and only if the discriminant of the equation is negative:
If (2a)2 4 = 16b < 0, then E (R)
= S 1 , so v is surjective.
If b > 0 and < 0 then E(R)
= S 1 Z/2Z, so v is not
= S 1 and E (R)
surjective.
If b, > 0, then E(R)
= S 1 Z/2Z
= E (R). Here v is surjective if and
only if the points O, (0, 0) of ker lie on the same connected component. (If
they are on dierent components, the image of is connected; otherwise, the
identity component of E(R) maps 2-to-1 to the identity component of E (R),
so the other component maps to the other component since deg = 2.) So
v is surjective if and only if 0 is the rightmost root of x3 + ax2 + bx, and
this is equivalent to a being positive.
5.3
The correction term
To save our strategy, which seems to be somewhat in ruins, we have to introduce

a correction term that measures the dierence between w(E/F ) and (E/F ). It
230
should be trivial at complex places and depend on the signs of a, b and at real
places, so it is natural to consider something like
(a, b)F (a, )F
where (x, y)F is the Hilbert symbol:
Denition 5.5. Let F be a local eld of characteristic zero. For x, y F , the
Hilbert symbol (x, y)F = 1 is

+1 if x is a norm from F ( y) to F ,
(x, y) = (x, F ( y)/F ) =

1 otherwise.
Recall that (x, y) is symmetric, bilinear as a map F F {1}, and
satises (1 x, x) = 1 for x = 0, 1. In a eld of residue characteristic = 2 it is
explicitly determined by
F =C:
(x, y) = 1 always;
F =R:
(x, y) = 1 x, y < 0;
F/Qp nite, p = 2:
(unit, unit) = 1,
(uniformiser, non-square unit) = 1.

The suggested correction term (a, b)F (a, )F is trivial at complex places
and it is precisely set up to give the right signs at the real places. A few experiments
suggest that, up to a missing factor of 2 in front of a (invisible over the reals),
this is the right correction at all completions F = Kv , not just at innite places.
One technical problem is that the Hilbert symbols do not make sense when a = 0,
but this is easy to x: if |a|F is small, then
(a, b)F (2a, a2 4b)F = (a, b)F (2a, 1
a2
4b )F
(2a, 4b)F
= (a, b)F (2a, )F (a, 4b)F (2, 4b)F

= (2, b)F ,
because elements of F close to 1 are squares. So we may state
Conjecture 5.6. Let F be a local eld of characteristic zero, and E/F an elliptic
curve with a 2-isogeny : E E over F . In the notation of 5.2,

(a, b)F (2a, a2 4b)F if a = 0,
w(E/F ) = (E/F )
if a = 0.
(2, b)F
This conjecture, together with (5.4), implies Theorem 5.1. Indeed, if E as
in 5.2 is now dened over a number eld K, the product formula for the Hilbert
symbol
'
(x, y)Kv = 1, for all x, y K
v
231
implies that the correction term disappears globally.

Next, we prove the conjecture in a few cases. As we dealt with innite places
already, assume from now on that F = Kv is non-Archimedean, i.e., a nite
extension of Qp . Also we may deal with the annoying exceptional case a = 0:
if we prove the conjecture in all cases when |a|F is small, then it holds when
a = 0 as well, because both the left- and the right-hand side in the conjecture are
continuous functions of the coecients a and b of E; we already proved this for
the Hilbert symbols, and for C(E/F, ) and w(E/F ) this is a general fact:
Proposition 5.7. Suppose that E/F is an elliptic curve in Weierstrass form,
E : y 2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , with ai F .
There is an > 0 such that changing the ai to any ai with |ai ai |F < does
not change the conductor, minimal discriminant, Tamagawa number, the value of
C(E, dx/(2y + a1 x + a3 )), the root number of E and the Tate module Tl E as a
Gal(F /F )-module for any given l = p.
Proof. The assertion for the conductor, minimal discriminant, Tamagawa number
and C follows from Tates algorithm ([60] or [55, IV.9]). The claim for Tl E is a
result of Kisin ([29, p. 569]), and the root number is a function of Vl E = Tl E
Ql . Alternatively, that the root number is locally constant can be proved in a
more elementary way: see Helfgotts [27, Prop. 4.2] when E has potentially good
reduction; in the potentially multiplicative case it follows from Rohrlichs formula
([48, Thm. 2(ii)]).

So assume from now on that a = 0. Note also that our chosen model y 2 =
x + ax2 + bx is unique up to transformations a u2 a, b u4 b for u K .
As these do not change the Hilbert symbols (a, b), (2a, a2 4b) and (2, b),
we may and will assume that the model is integral and minimal when v 2. Now
we prove the conjecture in all cases when E is semistable with v 2; recall that
ord2 CE/F = ord2 c(E/F ) in this case, since the quotient C/c is a power of the
residue characteristic.
3
5.4
Good reduction at v 2
Here w(E/F ) = 1, (E/F ) = 1 and b, OF . If a OF , then both the

Hilbert symbols are (unit, unit), hence trivial. For a 0 (mod mF ), the expression
b 4b2 (mod mF ) is a non-zero square mod mF , so the product of the Hilbert
symbols is again trivial.
5.5
Split multiplicative reduction at v 2
Write E/F as a Tate curve ([55, V.3])

Eq : y 2 + xy = x3 + a4 (q)x + a6 (q),
E(F )
= F /q Z ,
232
with q mF of valuation v(q) = v(). The coecients have expansions

a4 (q) = 5s3 (q), a6 (q) =
nk q n
5s3 (q) + 7s5 (q)
, sk (q) =
,
12
1 qn
n1
and they start as

a4 (q) = 5q 45q 2 140q 3 365q 4 630q 5 + O(q 6 ),
a6 (q) = q 23q 2 154q 3 647q 4 1876q 5 + O(q 6 ).
The 2-torsion, as a Galois set, is {1, 1, q, q}. For u = 1 in this set, the
corresponding point on E has coordinates

qn u
q n u1
qn
u
+
+
2
X(u, q) =
,
(1 u)2
(1 q n u)2
(1 q n u1 )2
(1 q n )2
n1
Y (u, q) =

u2
+
3
(1 u)
n1
q 2n u2
q n u1
qn
+
+
n
3
n
1
2
(1 q u)
(1 q u )
(1 q n )2
We now have two cases to consider: the 2-torsion point (X(1, q), Y (1, q)) Eq
and (renaming q by q) the 2-torsion point (X(q, q 2 ), Y (q, q 2 )) Eq2 . In both

cases, we have c(E)/c(E ) = 21 and w(E/F ) = 1, so we need that
(a, b) (2a, ) = 1,
(5.8)
where a, b, are the invariants of the curve transformed back to the form 5.2 with
the 2-torsion point at (0, 0). First of all, Eq has a model
y 2 = x3 + x2 /4 + a4 (q)x + a6 (q).
Let r = X(u, q), and write a4 = a4 (q), a6 = a6 (q). Then, after translation, the
curve becomes
E : y 2 = x3 + ax2 + bx,
a = 1/4 3r,
b = 2a4 r/2 + 3r2 .
Suppose we are in Case 1, so r = X(1, q). Then the substitution

x 4x 2r + 1/2,
y 8y + 4x
(5.9)
transforms E into the form

E : y 2 + xy = x3 + 5q 2 + O(q 4 ) x + q 2 + O(q 4 ) .
We use the notation O(q n ) to indicate a power series in q with coecients in OF
that begins with an q n + . In fact, E = Eq2 but we will not need this; it is
233
only important that it is again a Tate curve (in particular, this model is minimal).
From the expansions
r = 1/4 + 4O(q),
a = 1/2 + 4O(q),
b = 1/16 + O(q),
we have
(a, b) = (unit, unit) = 1,

(2a, ) = 1 8O(q), = , = 1.
Case 2 is similar: here

a = 1/4 + 2 O(q),
b = q + O(q 2 ),
= 1/16 + O(q),
so a and are squares in F , and both Hilbert symbols are therefore trivial.
5.6
Nonsplit multiplicative reduction at v 2
Let OF be a non-square unit, so that F ()/F is the unramied quadratic

extension of F . Consider the twist of E by ,
E:
E :
y 2 = x3 + ax2 + bx,
y 2 = x3 + ax2 + 2 bx.
It has split multiplicative reduction, so, by (5.8),

(a, 2 b) (2a, 2 (a2 4b)) = 1.
Now

a, 2 b = a, b = , b a, b ,

2a, 2 (a2 4b) = 2a, a2 4b = , a2 4b 2a, a2 4b ,
so, comparing with the Hilbert symbols (a, b) and (2a, ), we have an extra
term

, b(a2 4b) = , b2 (E )/16(E) = , (E )/(E) .
(5.10)
Because x is a norm from F () to F if and only if v(x) is even, this Hilbert symbol is trivial precisely when v((E )) v((E)) (mod 2). From Tates algorithm
([55, IV.9.4, Step 2]),

1 if v((E )) is odd,
1 if v((E)) is odd,
c(E) =
c(E ) =
2 if v((E )) is even,
2 if v((E)) is even,
so the correction term (5.10) is trivial if and only if c(E)/c(E ) has even 2-valuation. This proves Conjecture 5.6 in the non-split multiplicative case, when v 2.
234
5.7
Deforming to totally real elds
There are several other reduction types, including all cases when v 2 and when E
has good ordinary or multiplicative reduction at v | 2, when Conjecture 5.6 can
be shown to hold directly [11]. It would be satisfying to do this in the remaining
cases as well:
Problem 5.11. If F/Q2 is nite, and E/F has good supersingular or additive
reduction, prove Conjecture 5.6 directly.
Having tried and failed to do this, we will complete the proof of Conjecture 5.6
and of Theorem 5.1 by a global argument. The idea is that if E/K is an elliptic
curve over a number eld and the conjecture holds for it at all primes but one,
the conjecture at the remaining prime is equivalent to the 2-parity conjecture
for E. But there is a large supply of elliptic curves E, namely those dened over
totally real elds with non-integral j-invariants, for which we know the 2-parity
conjecture. There are more than enough of these to approximate any given elliptic
curve over F .
Assumption 5.12. From now on, F/Q2 is nite, a, b F satisfy b = 0, a2 4b = 0,
and Ea,b is an elliptic curve as in 5.2,
Ea,b : y 2 = x3 + ax2 + bx.
Lemma 5.13. There exists a totally real eld K with a unique place v0 | 2, and
a
, b K such that:
(1) Kv0
= F.
(2) Under this identication, |a a
|v0 and |b b|v0 are so small that all terms
in Conjecture 5.6 are the same for Ea,b /F and for Ea,b /Kv0 .
(3) Ea,b is semistable at all primes v = v0 of K and has non-integral j-invariant.
Proof. (1) Say F = Q2 [x]/(f ) with monic f Q2 [x]. If f Q[x] is monic and
2-adically close enough to f , it denes the same extension of Q2 by Krasners
lemma. Now pick such an f which is also R-close to any polynomial in R[x] whose
roots are real (weak approximation), and set K = Q[x]/(f) and v0 to be the prime
above 2 in K.
(2) Next, provided a
, b K are close enough to a, b at v0 , the continuity of
the Hilbert symbol (and the computation preceding Conjecture 5.6) and Proposition 5.7 imply that (2) holds.
(3) As v0 is the unique prime above 2, it suces to guarantee that Ea,b is
semistable at primes v 2 of K and has j(E) OK . The curve has standard
invariants
a2 3b), c6 = 32
a(2
a2 9b), = 16b2 (
a2 4b).
c4 = 16(
6. The p-isogeny conjecture
235
Pick any non-zero a

OK which is close to a at v0 . Next, choose any b OK
which is close to b at v0 and close to 1 at all primes v = v0 that divide a
. Then
either or c4 is a unit at every prime v = v0 , and this ensures that E is semistable
outside v0 ([54, Prop. VII.5.1]). If, in addition, we force b to be divisible by at least
one prime p 2 (weak approximation all the time), this guarantees that j(E) is
non-integral at p.

Claim. Conjecture 5.6 is true.
Proof. First we nish o the case when F has residue characteristic 2. Let F/Q2
and Ea,b be as above, and K and E = Ea,b as in the lemma. We want to show
that Conjecture 5.6 holds for Ea,b /F , equivalently for E/Kv0 .
Because K has a unique prime v0 above 2, and E is semistable at all other
primes, Conjecture 5.6 holds for E/Kv for all v = v0 . In view of (5.4), the conjecture at v0 is equivalent to the 2-parity conjecture for E/K. But E has non-integral
j-invariant and K is totally real, so the latter holds.
Finally, suppose F has odd residue characteristic p, and E/F has additive
reduction. Exactly as in Lemma 5.13, nd a totally real eld K with a place v0 | p
and E/K
which is v0 -adically close to E, and semistable at all v = v0 . Again,
the 2-parity conjecture for E/K

together with Conjecture 5.6 at all v = v0 (now

including v | 2) proves Conjecture 5.6 at v0 as well.
As explained above, reversing the argument yet again proves Theorem 5.1.
The p-isogeny conjecture
What is the analogue of the results of the previous section for an isogeny whose
degree is an odd prime p? Let : E E be such an isogeny, with E, E and
all dened over K. Again, write
|coker v |

E/Kv = (1)ordp |ker v | ,
so that, by Theorem 2.9,

(1)rkp E/K =
'
v
'

E/Kv and
w E/Kv = w E/K .
(6.1)
Again, let us make a term-by-term comparison at all places. Fix a place v of K.

The dierence between w(E/Kv ) and (E/Kv ) is suggested by the root number
computation of Theorem 3.25, and we formulate the following
Conjecture 6.2. Let K be a local eld of characteristic zero, and E/K an elliptic
curve with a p-isogeny : E E over K. Write F = K(ker ). Then

w E/K = E/K 1, F/K .
236
If this conjecture is true, applying it for E/K at completions and taking the
product over all places, we get the p-parity conjecture for E/K. Let us verify that
the conjecture holds in most situations:
If K = C then | ker | = p, | coker | = 1, w(E/K) = 1 and F = K, so the
formula holds. The same is true if K = R and the points in ker are real.
If K = R and the points in ker are not dened over R, then | ker | = 1,
| coker | = p and F = C, so the correction term (1, C/R) = 1 compensates for
the kernel, and the formula holds again.
Suppose K is a nite extension of Ql for l = p. The standard exact sequences
for E0 (K) E(K) and E1 (K) E0 (K) ([54, Ch. VII]) show that

ord
E/K = (1) p
c(E/K)
c(E /K)
where c is the local Tamagawa number. (Use that E and E have the same reduction type and the same number of points over the residue eld, and that [p] is an
isomorphism on the formal groups when p = l.) Now we have a few cases:
If E/K has good reduction, then K(E[p])/K is unramied by the NeronOgg
Shafarevich criterion, and in particular F/K is unramied as well. So all units are
norms in this extension by class eld theory; in particular 1 is. The Tamagawa
numbers are trivial as well, so all terms in the conjecture are 1.
If E/K has multiplicative reduction, then F/K is still unramied, since the
inertia in K(E[p])/K is pro-p in the multiplicative case and [F : K] | (p 1) is
coprime to p; so the Artin symbol is still 1. Now w(E/K) = 1 as opposed to the
c(E/K)
good case, but also the quotient c(E
/K) is either p or 1/p (same argument as in
5.5 for p = 2), and the conjecture holds.
If E/K has additive reduction and p, l are at least 5, then the Tamagawa
numbers have order 4, coprime to p. So the conjecture claims that w(E/K) =
(1, F/K), and this is precisely what we proved in Theorem 3.25; in fact, it is this
computation that suggests the correction term in the conjecture.
When p = l, the problem becomes trickier. It is still manageable when E/K
is semistable, by a careful analysis of which also involves the action of on
the formal groups (see [11]). For p = 3 the whole conjecture can be settled by
a deformation argument, exactly as we did for p = 2 (see [16]; presumably the
argument there also works when p = 5 and 7, when X0 (p) still has genus 0). The
remaining problem is
Problem 6.3. Prove Conjecture 6.2 when p > 7, the eld K has residue characteristic p and E/K has additive reduction.
Coates, Fukaya, Kato and Sujatha [6] have proved this conjecture when E/K
acquires semistable reduction after an abelian extension of K, with a tour de force
computation with crystalline cohomology. Moreover, they proved it for arbitrary
principally polarised abelian varieties; p-isogeny has to be replaced by a pg -isogeny
with a totally isotropic kernel. It would be very interesting to nd at least a
conjectural generalisation for p = 2:
7. Local compatibility in S3 -extensions
237
Problem 6.4. Suppose that K is a local eld of characteristic 0 and : A A

is an isogeny of principally polarised abelian varieties over K such that t = [2]
(up to isomorphisms given by the polarisations). Find a relation between (A/K)
and w(A/K).
This would have important parity implications for abelian varieties. We also
mention here that Trihan and Wuthrich [64] have proved an analogue of Conjecture 6.2 over function elds of characteristic p, when is the dual of the Frobenius
isogeny. Because this isogeny exists for elliptic curves over a function eld, they
thus proved the p-parity conjecture in characteristic p.
Finally, returning to Conjecture 6.2 itself, it would also be very interesting to
see whether it can be settled by a deformation argument, like we did it for p = 2.
The problem is that because X0 (p) has positive genus for large p, it is unclear how
to deform elliptic curves with a p-isogeny while keeping control over other places
of the resulting totally real eld. It is known how to construct points of essentially
arbitrary varieties over totally real elds (see e.g. [46]), but it is not clear whether
these results are sucient in this case.
Local compatibility in S3 -extensions
To complete the proof of Theorem A, it remains to settle the second equality in

Theorem 2.7. Thus, suppose E/K is an elliptic curve, F/K is a Galois S3 -extension
and M/K and L/K are the quadratic and cubic intermediate extensions. We want
to show that
ord3
w(E/K) w(E/M ) w(E/L) = (1)
CE/F C 2
E/K
CE/M C 2
E/L
(glo )
Both sides are products of local terms and, as in the isogeny case, we want to
compare the contributions above each place of K.
Fix such a place v of K and denote by wK , wM , wL and wF the product of
the local root numbers of E over the places of K, M , L and F above v. Fix also
an invariant dierential for E/K and write CK , CM , etc. for the product of the
modied local Tamagawa numbers of E over the places of K, M , etc., computed
using (see end of Notation 1.1). We need to show the following local statement:
wK wM wL = (1)
ord3
2
CF CK
CM C 2
L
(loc )
Lemma 7.1. If there is more than one place above v in F or E has good reduction
at v, then (loc ) holds.
Proof. It is easy to see that for v | both the left-hand side and the right-hand
side are trivial6 , so suppose v is a nite place. Then the right-hand side is the
same as ordp (CF /CM ) mod 2. Now consider the following cases.
6 This
is true in any Brauer relation; see e.g. [14, Proof of Cor. 3.4].
238
Case 1: v splits in M/K. (In particular, this must happen if F/M is inert
above v.) As the number of primes above v in M is even, CF and CM are both
squares, and wM = 1. Since in this case Lv /Kv is Galois of odd degree, wL = wK
by KramerTunnell [33, Proof of Prop. 3.4].
Case 2: the prime above v in M splits in F/M . Then CF = (CM )3 , so CF /CM
is a square. Write G = Gal(F/K)
the decomposition
= S3 . Under the action of
group Dv at v, the G-sets G/ Gal(F/L) and (G/ Gal(F/M )) (G/G) are isomorphic. So the number of primes above v with a given ramication and inertial degree
is the same in L as in M plus in K. It follows that the local root numbers cancel,
wK wM wL = 1.
Case 3: F/M is ramied above v and E is semistable at v. The contributions from cancel modulo squares, and wK = wL . If E has split multiplicative
reduction over a prime of M above v, this prime contributes p to CF /CM and 1
to the root number. If the reduction is either good or non-split, it contributes to
neither.

Proposition 7.2. The formula (loc ) holds in all cases.
Proof. The only case not covered by the lemma above is when E/K has additive
reduction at v and F/K has a unique prime v above v. We will use a continuity
argument to settle this case.
Pick an S3 -extension F/K of totally real number elds with completions
| w in F/K (same argument as in
Kw = Kv and Fw = Fv for some prime w
Proposition 5.7). Choose an elliptic curve E/K which is close enough w-adically
to E/Kv , with semistable reduction at all places = w where F/K is ramied and
at least one prime of multiplicative reduction. By close enough we mean that the
left-hand side and the right-hand side of (loc ) are the same for E/Kv and E/Kw
(Proposition 5.7).
By the 3-parity conjecture for E over the intermediate elds of F/K (Theorem 4.4) and Theorem 2.10, we nd that (glo ) holds. Since the terms in it agree
at all places except possibly w by Lemma 7.1, they must agree at w as well. This

proves (loc ) for E/Kw and hence for E/Kv as well.
This completes the proof of Theorem A.
Parity predictions
The purpose of this nal section is to collect some peculiar predictions of the parity
conjecture concerning ranks of elliptic curves over number elds.

Denition 8.1. Write H pq = max(|p|, |q|) for the usual naive height of a rational
number. We say that a subset S Q has density d if

# a S | H(a) < X

= d.
lim
X # a Q | H(a) < X
8. Parity predictions
239
There is a folklore minimalistic conjecture:

Conjecture 8.2. Let E/Q(t) be an elliptic curve of MordellWeil rank r. For a Q
write Ea for its specialisation t a. Then

r
if w(Ea /Q) = (1)r ,
rk Ea /Q =
r + 1 otherwise,
for a set of rational numbers a of density 1.
Note that Ea is indeed an elliptic curve for all but nitely many vaues of a,
so the conjecture makes sense. What it says is that, generically, the rank of a ber
is a sum of the geometric contribution from the points in the family plus an
arithmetic contribution from the root number.
8.1
Semistable curves in cubic extensions
The simplest elliptic curves are semistable ones, and their root numbers are particularly nice:
Denition 8.3. An elliptic curve over E/K is semistable if it has good or multiplicative reduction at all primes of K.
Because places of good and non-split multiplicative reduction do not contribute to the global root number, and innite and split multiplicative places
contribute 1, for semistable E/K,
w(E/K) = (1)#{v|} (1)#{v split multiplicative for E} .
Thus, the parity conjecture implies
Conjecture 8.4. If E/Q is semistable, then rk E/Q 1 + #{primes p where E has
split multiplicative reduction} (mod 2).
Example 8.5 ([18]). Take E = 19A3 over Q,
E : y 2 + y = x3 + x2 + x,
= 19,
split multiplicative at 19.
Two-descent shows that its rank over Q is 0 (in accordance with the prediction
that it is even).
Now take Km = Q( 3 m) for some cube-free integer m > 1. Then the conjecture asserts that

3 if 19 splits in Km
rk E/Km 2 +
1 (mod 2).
1 otherwise
#{v|}
#{v|19}
So rk E/Km should always be odd, in particular positive. We get the following

surprising statement:
240
Conjecture 8.6. For every m > 1, not

a cube, the equation y 2 + y = x3 + x2 + x
3
has innitely many solutions in Q( m).
It is known that
an elliptic curve over Q acquires rank over innitely many
elds of the form Q( 3 m) ([9, Thm. 1]), but the full conjecture appears to be
completely unapproachable at the moment.
8.2
Number elds K such that w(E/K) = 1 for all E/Q
Let K = Q( 1, 17). This eld has a peculiar property that every place of Q
splits into an even number of places in it (2 or 4):
K has two (complex) places v | .
2 splits in Q( 17), and thus in K as well.
17 splits in Q( 1), and thus in K as well.

Primes p = 2, 17 are unramied in K/Q, so their decomposition groups are
cyclic, Dp = Gal(K/Q) = C2 C2 ; so such p split as well.
Thus, for any elliptic curve E/Q,
w(E/K) =
'
w(E/Kp ) =
'
(1)even = +1,
and we get
Conjecture 8.7. Every elliptic curve E/Q has even rank over Q( 1, 17).
The existence of such number elds was pointed out to us by Rubin. Note
also that the same conjecture may be stated for abelian varieties. As an exercise,
we leave it to the reader to use the same ideas to show that the parity conjecture
implies the following:
Conjecture 8.8. Every elliptic curve over Q with split multiplicative reduction at 2
has innitely many rational points over Q(8 ).
8.3
Goldfelds conjecture over Q
Denition 8.9. For an elliptic curve

E/Q : y 2 = f (x)
and a (usually square-free) integer d, the quadratic twist of E by d is
Ed /Q : dy 2 = f (x).
241
Note that E
= Ed over Q(
d), but not over Q. Now, if d0 < 0 is such that
all primes p | 2E split in Q( d0 ), then it is easy to see that
w(Edd0 /Q) = w(Ed /Q) for all square-free d.
In other words, the involution d dd0 on Q /Q2 changes the sign of w(Ed ). So
w(Ed /Q) = +1 for 50% square-free ds,
w(Ed /Q) = 1 for 50% square-free ds,
meaning that
#{|d| X square-free | w(Ed /Q) = 1}
1
as X .
#{|d| X square-free}
2
(8.10)
The minimalistic conjecture above becomes a famous conjecture of Goldfeld:

Conjecture 8.11 (Goldfeld). Let E/Q be an elliptic curve. Then
rk Ed /Q = 0 for 50% square-free ds,
rk Ed /Q = 1 for 50% square-free ds,
rk Ed /Q 2 for 0% square-free ds.
Note that 0% does not exclude the possibility of E having innitely many
quadratic twists of rank 2. It only says that

r2 (X) := # |d| X square-free | rk Ed /Q 2
is o(X) (the denominator in (8.10) is X/(2) for large X). In fact, it is known
that
r=0 (X) X/ log X
r=1 (X) X 1
r2 (X) CE X 1/7 / log2 X
(OnoSkinner [44]),
(PomykalaPerelli [45]),
(StewartTop [56]).
For some specic elliptic curves it is known that r=0 (X) CX and r=1 (X)
C X, but to get C or C to be 12 seems to be extremely hard.
8.4
No Goldfeld over number elds
Over number elds, Goldfelds Conjecture has to be formulated dierently, because the w = +1 in 50% cases formula (8.10) may not hold. The simplest
counterexample is CM curves:
Example 8.12. Let K = Q(i) and E/K : y 2 = x3 + x. This is a curve with complex
multiplication,
EndK E
= Z[i], [i](x, y) = (x, iy).
242
The set of rational points E(K) is naturally a Z[i]-module, and so is E(F ) for any
extension F/K. Because E(F ) Z Q is a Q(i)-vector space, it is even-dimensional
over Q, so rk E/F is even for every F K. Hence

rk Ed /K = rk E/K d rk E/K 0 (mod 2), for all d K /K 2 ;
in other words, every quadratic twist of E/K has even rank. It is also not hard to
prove that w(Ed /K) = 1 for all d, as expected.
The same applies to any CM curve E/K with endomorphisms dened over K.
Such a K is automatically totally complex (it contains End E as a subring), the
representation GK Aut Tl E has abelian image, and E acquires everywhere
good reduction after an abelian extension of K. Interestingly, these are precisely
the local conditions on an elliptic curve to guarantee that all of its quadratic twists
have the same root number:
Theorem 8.13 ([10]). Let E/K be an elliptic curve. Then w(Ed /K) = w(E/K)
for all d K if and only if
(a) K is totally complex, and
(b) for all primes p of K the curve E/Kp acquires good reduction after an abelian
extension of Kp .
For semistable curves, the second condition simply says that E must have
everywhere good reduction. It is not hard to construct explicit examples:
Example 8.14. The elliptic curve E/Q : y 2 = x3 + 54 x2 2x7 (121C1) has minimal
discriminant 114 and acquireseverywhere good reduction over, e.g., Q( 3 11). If we

take for instance K = Q(3 , 3 11), it is totally complex and E/K has everywhere
good reduction, so

w Ed /K = w E/K = (1)#{v|} = 1, for all d K .
The parity conjecture implies that every quadratic twist of E/K has positive rank,
or, equivalently, that the rank of E must grow in every quadratic extension of K.
Conjecture 8.15. The curve 121C1 over K = Q(3 , 3 11) and all of its quadratic
twists over K have positive rank.
Here is a very elementary way to phrase this:
Conjecture 8.16. Over K = Q(3 , 3 11) the polynomial x3 + 54 x2 2x 7 K[x]

takes every value in K /K 2 .
8.5
No local expression for the rank
The reader may have noticed that the above examples rely not so much on the
precise formulae for the root numbers but mostly just on their existence. In other
words, they explore the fact that (conjecturally) the parity of the MordellWeil
rank is a sum of local invariants:
243
Denition 8.17. Say that the MordellWeil rank (resp. MordellWeil rank modulo n) is a sum of local invariants if there is a Z-valued function (k, E) (E/k)
of elliptic curves over local elds7 , such that, for any elliptic curve E over any
number eld K,

(E/Kv ) (resp. rk E/K
(E/Kv ) (mod n)),
rk E/K =
v
the sum taken over all places of K (and implicitly nite).

The parity conjecture implies that the MordellWeil rank modulo 2 is a sum
of local invariants, namely those dened by the local root numbers,
(1)(E/Kv ) := w(E/Kv ).
One might ask whether there is a local expression like this for the rank modulo 3
or modulo 4, or even for the rank itself. The answer is no:
Theorem 8.18 ([17]). The MordellWeil rank is not a sum of local invariants. In
fact, a stronger statement holds: for n {3, 4, 5} the MordellWeil rank modulo n
is not a sum of local invariants.
Proof. Take E/Q : y 2 = x(x + 2)(x 3), which is 480a1 in Cremonas notation.
Writing p for a primitive pth root of unity, let
if n = 3,
the degree 9 subeld of Q(13 , 103 )

if n = 5,
the degree 25 subeld of Q(11 , 241 )
Fn =
Q( 1, 41, 73)
if n = 4.
Because 13 and 103 are cubes modulo one another, and all other primes are unramied in F3 , every place of Q splits into 3 or 9 in F3 . Similarly in F4 (resp. F5 )
every place of Q splits into a multiple of 4 (resp. 5) places. Hence, if the Mordell
Weil rank modulo n were a sum of local invariants, it would be 0 Z/nZ for
E/Fn . However, 2-descent shows that rk E/F3 = rk E/F5 = 1 and rk E/F4 = 6

(e.g. using Magma over all minimal non-trivial subelds of Fn ).
Remark 8.19. It is interesting to note
L-series of the curve E = 480a1
thatthe
used in the proof over F = F4 = Q( 1, 41, 73) is formally a 4th power, in
the sense that each Euler factor is:
4
1
1
1
L(E/F, s) = 1
1 32s
1 52s
1 + 14 72s + 724s
4
1
....
1 + 6 112s + 1124s
7 Meaning that if k k and E/k and E /k are isomorphic elliptic curves (identifying k
=
with k ), then (E/k) = (E /k ).
244
However, it is not a 4th power of an entire function, as it vanishes to order 6 at

s = 1. (Actually, it is not even a square of an entire function: a computation shows
that it has a simple zero at 1 + 2.1565479 . . . i.)
In fact, by construction of F , for any E/Q the L-series L(E/F, s) is formally
a 4th power and vanishes to even order at s = 1 by the functional equation. Its
square root has analytic continuation to a domain including Re s > 32 , Re s < 12
and the real axis, and satises a functional equation s 2 s, but it is not clear
whether it has any arithmetic meaning.
Added in print
Problem 6.3 (p-parity problem for elliptic curves with a p-isogeny) has now been
solved by K. Cesnavi
cius, The p-parity conjecture for elliptic curves with a pisogeny, preprint 2012, arXiv:1207.0431.
Bibliography
[1] A. Bartel, On Brauer-Kuroda type relations of S-class numbers in dihedral
extensions, J. reine angew. Math. 668 (2012), 211-244.
[2] B. J. Birch, Conjectures concerning elliptic curves, Proc. Sympos. Pure Math.,
vol. VIII, Amer. Math. Soc., Providence, RI, 1965, 106112.
[3] B. J. Birch, N. M. Stephens, The parity of the rank of the Mordell-Weil group,
Topology 5 (1966), 295299.
[4] J. W. S. Cassels, Arithmetic on curves of genus 1, IV, Proof of the Hauptvermutung, J. reine angew. Math. 211 (1962), 95112.
[5] J. W. S. Cassels, Arithmetic on curves of genus 1, VIII: On conjectures of
Birch and Swinnerton-Dyer, J. reine angew. Math. 217 (1965), 180199.
[6] J. Coates, T. Fukaya, K. Kato, R. Sujatha, Root numbers, Selmer groups and
non-commutative Iwasawa theory, J. Algebraic Geom. 19 (2010), 1997.
[7] C. Cornut, V. Vatsal, Nontriviality of Rankin-Selberg L-functions and CM
points, in: L-functions and Galois representations (Durham, July 2004), LMS
Lecture Note Series 320, Cambridge Univ. Press, 2007, 121186.
[8] P. Deligne, Les constantes des equations fonctionnelles, Seminaire DelangePisot-Poitou (Theorie des Nombres), 11ème annee, 1969/70, no. 19 bis,
Secretariat Mathematique, Paris, 1970.
[9] T. Dokchitser, Ranks of elliptic curves in cubic extensions, Acta Arith. 126
(2007), 357360.
[10] T. Dokchitser, V. Dokchitser, Elliptic curves with all quadratic twists of positive rank, Acta Arith. 137 (2009), 193197.
[11] T. Dokchitser, V. Dokchitser, Parity of ranks for elliptic curves with a cyclic
isogeny, J. Number Theory 128 (2008), 662679.
[12] T. Dokchitser, V. Dokchitser, On the BirchSwinnerton-Dyer quotients modulo squares, Ann. of Math. 172 (2010), no. 1, 567596.
[13] T. Dokchitser, V. Dokchitser, Self-duality of Selmer groups, Math. Proc. Cambridge Philos. Soc. 146 (2009), 257267.
245
246
Bibliography
[14] T. Dokchitser, V. Dokchitser, Regulator constants and the parity conjecture,

Invent. Math. 178 (2009), no. 1, 2371.
[15] T. Dokchitser, V. Dokchitser, Root numbers of elliptic curves in residue characteristic 2, Bull. London Math. Soc. 40 (2008), 516524.
[16] T. Dokchitser, V. Dokchitser, Root numbers and parity of ranks of elliptic
curves, J. reine angew. Math. 658 (2011), 3964.
[17] T. Dokchitser, V. Dokchitser, A note on the Mordell-Weil rank modulo n,
J. Number Theory 131 (2011), no. 10, 18331839.
[18] V. Dokchitser, Root numbers of non-abelian twists of elliptic curves, with an
appendix by T. Fisher, Proc. London Math. Soc. (3), 91 (2005), 300324.
[19] G. Faltings, Endlichkeitss
atze f
ur abelsche Variet
aten u
ber Zahlk
orpern, Invent. Math. 73 (1983), no. 3, 349366.
[20] T. Fisher, Appendix to V. Dokchitser, Root numbers of non-abelian twists of
elliptic curves, Proc. London Math. Soc. (3), 91 (2005), 300324.
[21] S. Friedberg, and J. Hostein, Nonvanishing theorems for automorphic
L-functions on GL(2), Ann. of Math. 142 (1995), no. 2, 385423.
[22] A. Frohlich, J. Queyrut, On the functional equation of the Artin L-function
for characters of real representations, Invent. Math. 20 (1973), 125138.
[23] D. Goldfeld, Conjectures on elliptic curves over quadratic elds, Lecture Notes
in Math. 751, Springer, Berlin, 1979, 108118.
[24] R. Greenberg, On the Birch and Swinnerton-Dyer conjecture, Invent. Math.
72 (1983), no. 2, 241265.
[25] L. Guo, General Selmer groups and critical values of Hecke L-functions, Math.
Ann. 297 (1993), no. 2, 221233.
[26] E. Halberstadt, Signes locaux des courbes elliptiques en 2 et 3, C. R. Acad.
Sci. Paris Serie I Math. 326 (1998), no. 9, 10471052.
[27] H. A. Helfgott, On the behaviour of root numbers in families of elliptic curves,
arXiv:math/0408141v3.
[28] B. D. Kim, The Parity Theorem of elliptic curves at primes with supersingular
reduction, Compos. Math. 143 (2007), 4772.
[29] M. Kisin, Local constancy in p-adic families of Galois representations,
Math. Z. 230 (1999), 569593.
[30] S. Kobayashi, The local root number of elliptic curves with wild ramication,
Math. Ann. 323 (2002), 609623.
[31] V. A. Kolyvagin, Euler systems, The Grothendieck Festschrift, Progress in
Math., Birkhauser, Boston, 1990.
Bibliography
247
[32] K. Kramer, Arithmetic of elliptic curves upon quadratic extension, Trans.

Amer. Math. Soc. 264 (1981), no. 1, 121135.
[33] K. Kramer, J. Tunnell, Elliptic curves and local -factors, Compos. Math. 46
(1982), 307352.
[34] B. Mazur, K. Rubin, Finding large Selmer ranks via an arithmetic theory of
local constants, Ann. of Math. 166 (2007), no. 2, 579612.
[35] B. Mazur, K. Rubin, Growth of Selmer rank in nonabelian extensions of number elds, Duke Math. J. 143 (2008), 437461.
[36] B. Mazur, K. Rubin, Ranks of twists of elliptic curves and Hilberts tenth
problem, Invent. Math. 181 (2010), no. 3, 541575.
[37] J. S. Milne, On the arithmetic of abelian varieties, Invent. Math. 17 (1972),
177190.
[38] J. S. Milne, Arithmetic duality theorems, Perspectives in Mathematics, no. 1,
Academic Press, 1986.
[39] P. Monsky, Generalizing the BirchStephens theorem, I: Modular curves,
Math. Z. 221 (1996), 415420.
[40] J. Nekov
ar, The Euler system method for CM points on Shimura curves, in:
L-functions and Galois representations (Durham, July 2004), LMS Lecture
Note Series 320, Cambridge Univ. Press, 2007, 471547.
[41] J. Nekov
ar, Selmer complexes, Asterisque 310 (2006).
[42] J. Nekov
ar, On the parity of ranks of Selmer groups IV, with an appendix by
J.-P. Wintenberger, Compos. Math. 145 (2009), 13511359.
[43] J. Nekov
ar, Some consequences of a formula of Mazur and Rubin for arithmetic local constants, preprint, 2010.
[44] K. Ono, C. Skinner, Non-vanishing of quadratic twists of modular L-functions,
Invent. Math. 134 (1998), no. 3, 651660.
[45] A. Perelli, J. Pomykala, Averages over twisted elliptic L-functions, Acta Arith.
80 (1997), 149-163.
[46] F. Pop, Embedding problems over large elds, Ann. of Math. (2) 144 (1996),
no. 1, 134.
[47] T. de La Rochefoucauld, Invariance of the parity conjecture for p-Selmer
groups of elliptic curves in a D2pn -extension, preprint, arXiv:1002.0554.
[48] D. Rohrlich, Galois Theory, elliptic curves, and root numbers, Compos. Math.
100 (1996), 311349.
[49] D. Rohrlich, Galois invariance of local root numbers, Math. Ann. 351 (2011),
no. 4, 9791003.
248
Bibliography
[50] J.-P. Serre, Zeta and L-functions, in: Arithmetical Algebraic Geometry, Proc.
Conf. Purdue 1963 (1965), 8292.
[51] J.-P. Serre, Proprietes galoisiennes des points dordre ni des courbes elliptiques, Invent. Math. 15 (1972), no. 4, 259331.
[52] J.-P. Serre, Abelian l-adic Representations and Elliptic Curves, AddisonWesley, 1989.
[53] J.-P. Serre, J. Tate, Good reduction of abelian varieties, Ann. of Math. 68
(1968), 492517.
[54] J. H. Silverman, The Arithmetic of Elliptic Curves, GTM 106, SpringerVerlag, New York, 1986.
[55] J. H. Silverman, Advanced Topics in the Arithmetic of Elliptic Curves, GTM
151, Springer-Verlag, New York, 1994.
[56] C. L. Stewart, J. Top, On ranks of elliptic curves and power-free values of
binary forms, J. Amer. Math. Soc. 8 (1995), 943973.
[57] J. Tate, Duality theorems in Galois cohomology over number elds, Proc. ICM
Stockholm, 1962, 234241.
[58] J. Tate, On the conjectures of Birch and Swinnerton-Dyer and a geometric
analog, Seminaire Bourbaki, 18ème annee, 1965/66, no. 306.
[59] J. Tate, Fourier analysis in number elds and Heckes zeta functions, Algebraic Number Theory (J. W. S. Cassels and A. Fr
ohlich, eds.), Academic
Press, London, 1967, 305347.
[60] J. Tate, Algorithm for determining the type of a singular ber in an elliptic
pencil, in: Modular Functions of One Variable IV, Lecture Notes in Math.
476, B. J. Birch and W. Kuyk, eds., Springer-Verlag, Berlin, 1975, 3352.
[61] J. Tate, Number theoretic background, in: Automorphic forms, representations
and L-functions, Part 2 (ed. A. Borel and W. Casselman), Proc. Symp. in Pure
Math. 33, Amer. Math. Soc., Providence, RI, 1979, 326.
[62] R. Taylor, On icosahedral Artin representations II, Amer. J. Math. 125 (2003),
549566.
[63] R. Taylor, Remarks on the conjecture of Fontaine and Mazur, J. Inst. Math.
Jussieu 1 (2002), 119.
[64] F. Trihan, C. Wuthrich, Parity conjectures for elliptic curves over global elds
of positive characteristic, Compos. Math. 147 (2011), no. 4, 11051128.
[65] D. Whitehouse, Root numbers of elliptic curves over 2-adic elds, preprint,
2006, http://www-math.mit.edu/~dw/maths/elliptic2.pdf.
[66] J.-P. Wintenberger, appendix to [42].
Bibliography
249
[67] S. Zhang, Heights of Heegner points on Shimura curves, Ann. of Math. 153
(2001), 27147.

(Advanced Courses in Mathematics - CRM Barcelona) Laurent Berger, Gebhard Böckle, Lassina Dembélé, Mladen Dimitrov, Tim Dokchitser, John Voight (Auth.)-Elliptic Curves, Hilbert Modular Forms and Galoi

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

(Advanced Courses in Mathematics - CRM Barcelona) Laurent Berger, Gebhard Böckle, Lassina Dembélé, Mladen Dimitrov, Tim Dokchitser, John Voight (Auth.)-Elliptic Curves, Hilbert Modular Forms and Galoi

Transféré par

Droits d'auteur :

Formats disponibles

Coursesamatics

Advanced Courses in Mathematics

For further volumes:

On p-adic Galois Representations

Deformations of Galois Representations

Hilbert Modular Forms

Arithmetic Aspects of Hilbert Modular Forms and Varieties

Hilbert modular forms . . . . . . . . . . . . . . . . . . . . . . .

Explicit Methods for Hilbert Modular Forms

Notes on the Parity Conjecture

BirchSwinnerton-Dyer and parity . . . . . . . . . . . . . .

The Galois group of Q

Let E be a nite extension of Qp and let GQ = Gal(Q/Q). A p-adic representation

On p-adic Galois Representations

Given a p-adic representation V , one then studies its restriction to D for

Ramication of local elds, I

Let be a prime number, let K be a nite extension of Q , and let OK , mK , kK

p-adic representations with = p

An easy corollary of the equation 1 = m is Grothendiecks monodromy

On p-adic Galois Representations

If B has some supplementary structures, then these descend to DB (V ) and in this

In this section, we recall a few facts about Galois cohomology groups. If V is a

6. Ramication of local elds, II

As a consequence, we see for example that, if Qp B, then potentially

Ramication of local elds, II

In this section, we collect various statements about the ramication of extensions

Let F = Qp , let Fn = Qp (pn ) for n 1, and let F = n1 Fn . We know

On p-adic Galois Representations

then a GF -equivariant projection. We have Rn (1) = 1, while Rn (pn+k ) = 0 if

Let Cp be the p-adic completion of Qp , so that Cp is a complete and algebraically

If L is as above and Qp , then we set L () = inf gGL valp (g()). The

We say that a ring R is perfect if p = 0 and x  xp is a bijection on R. We

so that addition and multiplication of elements of A, written as

On p-adic Galois Representations

11. The eld BdR

The eld BdR

To prove this, we write the exact sequence

and use the computation of H 0 (K, Cp (h)) carried out in 8.

On p-adic Galois Representations

Conversely, we have the following conjecture of Fontaine and Mazur. If F is

The rings Bmax and Bst

Recall that in 10 we constructed the ring

+ for V(, r) (note that, more

14. Crystalline and semistable representations

+ and there + Bh is continuous for the valuation V(, r0 ) on B

Crystalline and semistable representations

We now carry out the constructions of 4 with B = Bmax or Bst . If V is a p-adic

On p-adic Galois Representations

The property of being crystalline or semistable is the analogue of having

Admissible ltered (, N )-modules

15. Admissible ltered (, N )-modules

On p-adic Galois Representations

The groups H1 (K, V )

If V is a p-adic representation of GK , then H 1 (K, V ) classies extensions E of Qp

We also dene He1 (K, V ) = ker(H 1 (K, V ) H 1 (K, B=1

0 Qp Bmax Bmax BdR /B+

Likewise, we can prove that

17. A p-adic period pairing

Let us make a few comments about this table.

then corresponds to the image of Q O , which is the characteristic zero

analogue of peu ramiees extensions.

by V and take GK -invariants, then we nd a connecting map

DdR (Vp A)/Fil0 DdR (Vp A)

A p-adic period pairing

We say that a ring R is perfect if p = 0 and x xp is a bijection on R. We

W (F) denote the category of complete Noetherian local W (F)-algebras with

W (F) )o is a pair of mor/ X in (Ar