LONDON MATHEMATICAL SOCIETY LECTURE NOTE SERIES

Managing Editor: Professor M. Reid, Mathematics Institute,

University of Warwick, Coventry CV4 7AL, United Kingdom

The titles below are available from booksellers, or from Cambridge University Press at
http://www.cambridge.org/mathematics

299 Kleinian groups and hyperbolic 3-manifolds, Y. KOMORI, V. MARKOVIC & C. SERIES (eds)
300 Introduction to Möbius differential geometry, U. HERTRICH-JEROMIN
301 Stable modules and the D(2)-problem, F.E.A. JOHNSON
302 Discrete and continuous nonlinear Schrödinger systems, M.J. ABLOWITZ, B. PRINARI & A.D. TRUBATCH
303 Number theory and algebraic geometry, M. REID & A. SKOROBOGATOV (eds)
304 Groups St Andrews 2001 in Oxford I, C.M. CAMPBELL, E.F. ROBERTSON & G.C. SMITH (eds)
305 Groups St Andrews 2001 in Oxford II, C.M. CAMPBELL, E.F. ROBERTSON & G.C. SMITH (eds)
306 Geometric mechanics and symmetry, J. MONTALDI & T. RATIU (eds)
307 Surveys in combinatorics 2003, C.D. WENSLEY (ed.)
308 Topology, geometry and quantum field theory, U.L. TILLMANN (ed)
309 Corings and comodules, T. BRZEZINSKI & R. WISBAUER
310 Topics in dynamics and ergodic theory, S. BEZUGLYI & S. KOLYADA (eds)
311 Groups: topological, combinatorial and arithmetic aspects, T.W. MÜLLER (ed)
312 Foundations of computational mathematics, Minneapolis 2002, F. CUCKER et al (eds)
313 Transcendental aspects of algebraic cycles, S. MÜLLER-STACH & C. PETERS (eds)
314 Spectral generalizations of line graphs, D. CVETKOVIC, P. ROWLINSON & S. SIMIC
315 Structured ring spectra, A. BAKER & B. RICHTER (eds)
316 Linear logic in computer science, T. EHRHARD, P. RUET, J.-Y. GIRARD & P. SCOTT (eds)
317 Advances in elliptic curve cryptography, I.F. BLAKE, G. SEROUSSI & N.P. SMART (eds)
318 Perturbation of the boundary in boundary-value problems of partial differential equations, D. HENRY
319 Double affine Hecke algebras, I. CHEREDNIK
320 L-functions and Galois representations, D. BURNS, K. BUZZARD & J. NEKOVÁR (eds)
321 Surveys in modern mathematics, V. PRASOLOV & Y. ILYASHENKO (eds)
322 Recent perspectives in random matrix theory and number theory, F. MEZZADRI & N.C. SNAITH (eds)
323 Poisson geometry, deformation quantisation and group representations, S. GUTT et al (eds)
324 Singularities and computer algebra, C. LOSSEN & G. PFISTER (eds)
325 Lectures on the Ricci flow, P. TOPPING
326 Modular representations of finite groups of Lie type, J.E. HUMPHREYS
327 Surveys in combinatorics 2005, B.S. WEBB (ed)
328 Fundamentals of hyperbolic manifolds, R. CANARY, D. EPSTEIN & A. MARDEN (eds)
329 Spaces of Kleinian groups, Y. MINSKY, M. SAKUMA & C. SERIES (eds)
330 Noncommutative localization in algebra and topology, A. RANICKI (ed)
331 Foundations of computational mathematics, Santander 2005, L.M PARDO, A. PINKUS, E. SÜLI & M.J. TODD (eds)
332 Handbook of tilting theory, L. ANGELERI HÜGEL, D. HAPPEL & H. KRAUSE (eds)
333 Synthetic differential geometry (2nd Edition), A. KOCK
334 The Navier–Stokes equations, N. RILEY & P. DRAZIN
335 Lectures on the combinatorics of free probability, A. NICA & R. SPEICHER
336 Integral closure of ideals, rings, and modules, I. SWANSON & C. HUNEKE
337 Methods in Banach space theory , J.M.F. CASTILLO & W.B. JOHNSON (eds)
338 Surveys in geometry and number theory, N. YOUNG (ed)
339 Groups St Andrews 2005 I, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
340 Groups St Andrews 2005 II, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
341 Ranks of elliptic curves and random matrix theory, J.B. CONREY, D.W. FARMER, F. MEZZADRI & N.C. SNAITH (eds)
342 Elliptic cohomology, H.R. MILLER & D.C. RAVENEL (eds)
343 Algebraic cycles and motives I, J. NAGEL & C. PETERS (eds)
344 Algebraic cycles and motives II, J. NAGEL & C. PETERS (eds)
345 Algebraic and analytic geometry, A. NEEMAN
346 Surveys in combinatorics 2007, A. HILTON & J. TALBOT (eds)
347 Surveys in contemporary mathematics, N. YOUNG & Y. CHOI (eds)
348 Transcendental dynamics and complex analysis, P.J. RIPPON & G.M. STALLARD (eds)
349 Model theory with applications to algebra and analysis I, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY &
A. WILKIE (eds)
350 Model theory with applications to algebra and analysis II, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY &
A. WILKIE (eds)
351 Finite von Neumann algebras and masas, A.M. SINCLAIR & R.R. SMITH
352 Number theory and polynomials, J. MCKEE & C. SMYTH (eds)
353 Trends in stochastic analysis, J. BLATH, P. MÖRTERS & M. SCHEUTZOW (eds)
354 Groups and analysis, K. TENT (ed)
355 Non-equilibrium statistical mechanics and turbulence, J. CARDY, G. FALKOVICH & K. GAWEDZKI
356 Elliptic curves and big Galois representations, D. DELBOURGO
357 Algebraic theory of differential equations, M.A.H. MACCALLUM & A.V. MIKHAILOV (eds)
358 Geometric and cohomological methods in group theory, M.R. BRIDSON, P.H. KROPHOLLER & I.J. LEARY (eds)
359 Moduli spaces and vector bundles, L. BRAMBILA-PAZ, S.B. BRADLOW, O. GARCÍA-PRADA & S. RAMANAN (eds)
360 Zariski geometries, B. ZILBER
361 Words: Notes on verbal width in groups, D. SEGAL
362 Differential tensor algebras and their module categories, R. BAUTISTA, L. SALMERÓN & R. ZUAZUA
363 Foundations of computational mathematics, Hong Kong 2008, F. CUCKER, A. PINKUS & M.J. TODD (eds)
364 Partial differential equations and fluid mechanics, J.C. ROBINSON & J.L. RODRIGO (eds)
365 Surveys in combinatorics 2009, S. HUCZYNSKA, J.D. MITCHELL & C.M. RONEY-DOUGAL (eds)
366 Highly oscillatory problems, B. ENGQUIST, A. FOKAS, E. HAIRER & A. ISERLES (eds)
367 Random matrices: High dimensional phenomena, G. BLOWER
368 Geometry of Riemann surfaces, F.P. GARDINER, G. GONZÁLEZ-DIEZ & C. KOUROUNIOTIS (eds)
369 Epidemics and rumours in complex networks, M. DRAIEF & L. MASSOULIÉ
370 Theory of p-adic distributions, S. ALBEVERIO, A.YU. KHRENNIKOV & V.M. SHELKOVICH
371 Conformal fractals, F. PRZYTYCKI & M. URBANSKI
372 Moonshine: The first quarter century and beyond, J. LEPOWSKY, J. MCKAY & M.P. TUITE (eds)
373 Smoothness, regularity and complete intersection, J. MAJADAS & A. G. RODICIO
374 Geometric analysis of hyperbolic differential equations: An introduction, S. ALINHAC
375 Triangulated categories, T. HOLM, P. JØRGENSEN & R. ROUQUIER (eds)
376 Permutation patterns, S. LINTON, N. RUŠKUC & V. VATTER (eds)
377 An introduction to Galois cohomology and its applications, G. BERHUY
378 Probability and mathematical genetics, N. H. BINGHAM & C. M. GOLDIE (eds)
379 Finite and algorithmic model theory, J. ESPARZA, C. MICHAUX & C. STEINHORN (eds)
380 Real and complex singularities, M. MANOEL, M.C. ROMERO FUSTER & C.T.C WALL (eds)
381 Symmetries and integrability of difference equations, D. LEVI, P. OLVER, Z. THOMOVA & P. WINTERNITZ (eds)
382 Forcing with random variables and proof complexity, J. KRAJÍCEK
383 Motivic integration and its interactions with model theory and non-Archimedean geometry I, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
384 Motivic integration and its interactions with model theory and non-Archimedean geometry II, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
385 Entropy of hidden Markov processes and connections to dynamical systems, B. MARCUS, K. PETERSEN &
T. WEISSMAN (eds)
386 Independence-friendly logic, A.L. MANN, G. SANDU & M. SEVENSTER
387 Groups St Andrews 2009 in Bath I, C.M. CAMPBELL et al (eds)
388 Groups St Andrews 2009 in Bath II, C.M. CAMPBELL et al (eds)
389 Random fields on the sphere, D. MARINUCCI & G. PECCATI
390 Localization in periodic potentials, D.E. PELINOVSKY
391 Fusion systems in algebra and topology, M. ASCHBACHER, R. KESSAR & B. OLIVER
392 Surveys in combinatorics 2011, R. CHAPMAN (ed)
393 Non-abelian fundamental groups and Iwasawa theory, J. COATES et al (eds)
394 Variational problems in differential geometry, R. BIELAWSKI, K. HOUSTON & M. SPEIGHT (eds)
395 How groups grow, A. MANN
396 Arithmetic differential operators over the p-adic integers, C.C. RALPH & S.R. SIMANCA
397 Hyperbolic geometry and applications in quantum chaos and cosmology, J. BOLTE & F. STEINER (eds)
398 Mathematical models in contact mechanics, M. SOFONEA & A. MATEI
399 Circuit double cover of graphs, C.-Q. ZHANG
400 Dense sphere packings: a blueprint for formal proofs, T. HALES
401 A double Hall algebra approach to affine quantum Schur–Weyl theory, B. DENG, J. DU & Q. FU
402 Mathematical aspects of fluid mechanics, J.C. ROBINSON, J.L. RODRIGO & W. SADOWSKI (eds)
403 Foundations of computational mathematics, Budapest 2011, F. CUCKER, T. KRICK, A. PINKUS & A. SZANTO (eds)
404 Operator methods for boundary value problems, S. HASSI, H.S.V. DE SNOO & F.H. SZAFRANIEC (eds)
405 Torsors, étale homotopy and applications to rational points, A.N. SKOROBOGATOV (ed)
406 Appalachian set theory, J. CUMMINGS & E. SCHIMMERLING (eds)
407 The maximal subgroups of the low-dimensional finite classical groups, J.N. BRAY, D.F. HOLT & C.M.
RONEY-DOUGAL
408 Complexity science: the Warwick master’s course, R. BALL, V. KOLOKOLTSOV & R.S. MACKAY (eds)
409 Surveys in combinatorics 2013, S.R. BLACKBURN, S. GERKE & M. WILDON (eds)
410 Representation theory and harmonic analysis of wreath products of finite groups, T. CECCHERINI-SILBERSTEIN,
F. SCARABOTTI & F. TOLLI
411 Moduli spaces, L. BRAMBILA-PAZ, O. GARCÍA-PRADA, P. NEWSTEAD & R.P. THOMAS (eds)
412 Automorphisms and equivalence relations in topological dynamics, D.B. ELLIS & R. ELLIS
413 Optimal transportation, Y. OLLIVIER, H. PAJOT & C. VILLANI (eds)
414 Automorphic forms and Galois representations I, F. DIAMOND, P.L. KASSAEI & M. KIM (eds)
415 Automorphic forms and Galois representations II, F. DIAMOND, P.L. KASSAEI & M. KIM (eds)
416 Reversibility in dynamics and group theory, A.G. O’FARRELL & I. SHORT
417 Recent advances in algebraic geometry, C.D. HACON, M. MUSTATA & M. POPA (eds)
418 The Bloch–Kato conjecture for the Riemann zeta function, J. COATES, A. RAGHURAM, A. SAIKIA &
R. SUJATHA (eds)
419 The Cauchy problem for non-Lipschitz semi-linear parabolic partial differential equations, J.C. MEYER &
D.J. NEEDHAM
420 Arithmetic and geometry, L. DIEULEFAIT et al (eds)
421 O-minimality and Diophantine geometry, G.O. JONES & A.J. WILKIE (eds)
422 Groups St Andrews 2013, C.M. CAMPBELL et al (eds)
423 Inequalities for graph eigenvalues, Z. STANIĆ
424 Surveys in combinatorics 2015, A. CZUMAJ et al (eds)
London Mathematical Society Lecture Note Series: 420

Arithmetic and Geometry

Edited by

L U I S D I E U L E FA I T
Universitat de Barcelona

G E R D FA LT I N G S
Max-Planck-Institut für Mathematik (Bonn)

D . R. HEATH-BROWN
University of Oxford

Y U . V. M A N I N
Max-Planck-Institut für Mathematik (Bonn)

B . Z . M O RO Z
Max-Planck-Institut für Mathematik (Bonn)

JEAN-PIERRE WINTENBERGER
Université de Strasbourg
 University Printing House, Cambridge CB2 8BS, United Kingdom

Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning and research at the highest international levels of excellence.

www.cambridge.org
Information on this title: www.cambridge.org/9781107462540

c Cambridge University Press 2015
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2015
Printed in the United Kingdom by Clays, St Ives plc
A catalogue record for this publication is available from the British Library
Library of Congress Cataloguing in Publication data
Arithmetic and geometry / edited by Luis Dieulefait, Universitat de Barcelona
[and five others].
pages cm. – (London Mathematical Society lecture note series ; 420)
Papers presented at the trimester on “Arithmetic and Geometry” at the Hausdorff
Research Institute for Mathematics (University of Bonn), January–April 2013.
ISBN 978-1-107-46254-0
1. Number theory – Congresses. 2. Algebraic number theory – Congresses.
3. Geometry of numbers – Congresses. I. Dieulefait, Luis, editor.
QA241.A695 2015
510–dc23
2015001829
ISBN 978-1-107-46254-0 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication,
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.
 Contents

Preface page vii

Introduction ix
List of participants x
Trimester Seminar xiii
Workshop on Serre’s conjecture xxiii
The research conference xxviii

1 Galois groups of local fields, Lie algebras and ramification 1

Victor Abrashkin
2 A characterisation of ordinary modular eigenforms with CM 24
Rajender Adibhatla and Panagiotis Tsaknias
3 Selmer complexes and p-adic Hodge theory 36
Denis Benois
4 A survey of applications of the circle method to rational points 89
T.D. Browning
5 Arithmetic differential equations of Painlevé VI type 114
Alexandru Buium and Yuri I. Manin
6 Differential calculus with integers 139
Alexandru Buium
7 Un calcul de groupe de Brauer et une application arithmétique 188
Jean-Louis Colliot-Thélène
8 Connectedness of Hecke algebras and the Rayuela
conjecture: a path to functoriality and modularity 193
Luis Dieulefait and Ariel Pacetti

v
vi Contents

9 Big image of Galois representations and congruence ideals 217

Haruzo Hida and Jacques Tilouine
10 The skew-symmetric pairing on the Lubin–Tate formal module 255
M. A. Ivanov and S. V. Vostokov
11 Equations in matrix groups and algebras over number fields
and rings: prolegomena to a lowbrow noncommutative
Diophantine geometry 264
Boris Kunyavskiı̆
12 On the -adic regulator as an ingredient of Iwasawa theory 283
L. V. Kuz’min
13 On a counting problem for G-shtukas 318
Ngo Dac Tuan
14 Modular forms and Calabi-Yau varieties 351
Kapil Paranjape and Dinakar Ramakrishnan
15 Derivative of symmetric square p-adic L-functions via
pull-back formula 373
Giovanni Rosso
16 Uniform bounds for rational points on cubic hypersurfaces 401
Per Salberger
17 Descent on toric fibrations 422
Alexei N. Skorobogatov
18 On filtrations of vector bundles over P1Z 436
A. Smirnov
19 On the dihedral Euler characteristics of Selmer groups of
Abelian varieties 458
Jeanine Van Order
20 CM values of higher Green’s functions and regularized
Petersson products 493
Maryna Viazovska
 Preface

The trimester on “Arithmetic and Geometry” at the Hausdorff Research

Institute for Mathematics (University of Bonn) took place in January – April
2013. In the next few pages the reader will find a list of the participants of
the trimester, the descriptions of the session on Serre’s conjecture, conducted
by L.V. Dieulefait and J.-P. Wintenberger, and of the session on counting
rational points on algebraic varieties, conducted by D.R. Heath-Brown, the
programmes of the workshop on Serre’s conjecture and of the final research
conference, and a list of the talks given at the HIM during the trimester. The
participants were invited to submit their papers for publication in this volume.
The papers appearing in the volume have been carefully refereed.
Acknowledgements. We wish to record our gratitude to the Hausdorff
Research Institute, and in particular to its director, Professor Dr W. Lück, for
the hospitality and financial support we received.

L.V. Dieulefait, G. Faltings, D.R. Heath-Brown, Yu.I. Manin,

B.Z. Moroz, and J.-P. Wintenberger (editors).

vii
 Introduction

The main theme of the trimester was the interplay of different methods used
in modern number theory. We wish to emphasize the new results and conjec-
tures in arithmetic geometry, having direct bearing on the classical number
theoretic problems. Two sessions, on the recently proved Serre’s conjecture
from 15 January to 14 February (organizers: L. Dieulefait and J.-P. Winten-
berger) and on counting rational points on algebraic varieties from 15 March
to 14 April (organizer: D.R. Heath-Brown), as well as a couple of shorter
workshops, several seminars, and mini-courses were organized. The trimester
culminated in a research conference from 15 to 19 April.
The aim of the session “Serre’s conjecture” was to report on recent works
linked to that conjecture, in particular about Galois representations and
automorphic representations. During the weeks starting on 14 January and
21 January, Henri Carayol lectured on his work on the algebraic properties
of Griffiths-Schmid varieties. The Griffiths-Schmid varieties are analytic vari-
eties classifying Hodge structures. Studying their algebraic properties might be
a step towards constructing Galois representations associated to automorphic
representations appearing in the cohomology of these varieties. Our second
theme related to the recent work of Michael Harris, Kai-Wen Lan, Richard
Taylor and Jack Thorne, who have constructed Galois representations asso-
ciated to not necessarily self-dual automorphic representations. The proof
heavily relies on p-adic properties of automorphic representations.
The aim of the session “counting rational points on algebraic varieties” was
to report on recent works on the existence, frequency and distribution of ratio-
nal points on algebraic varieties. Thus the main themes were local to global
principles, Manin’s conjecture, developments of the Hardy-Littlewood method
and the determinant method.

ix
 List of participants

Victor Abrashkin (University of Durham)

Rajender Adibhatla (Universität Regensburg)
Shabnam Akhtari (University of Oregon)
Paloma Bengoechea (College de France)
Denis Benois (Université Bordeaux 1)
Tobias Berger (University of Sheffield)
Yuri Bilu (IMB Université Bordeaux I)
Marco Boggi (University of Los Andes)
Pierre Le Boudec (Institute of Advanced Study)
Régis de la Bretéche (Institut de Mathématiques de Jussieu – Paris Rive
Gauche, UMR 7586 Université Paris-Diderot)
Christophe Breuil (Université Paris-Sud)
Tim Browning (University of Bristol)
Jörg Brüdern (Universität Göttingen)
Roman Budylin (Steklov Mathematical Institute)
Alexandru Buium (University of New Mexico)
Alberto Cámara (University of Nottingham)
Henri Carayol (L’Université de Strasbourg et du CNRS)
Magnus Carlson (University of Stockholm)
Tommaso Centeleghe (Universität Heidelberg)
Huan Chen (Ecole normale superieure ENS)
Narasimha Kumar Cheraku (Ruprecht Karls Universität Heidelberg)
Przemyslaw Chojecki (Institut Mathématique de Jussieu)
Laurent Clozel (Université de Paris Sud 11)
Jean-Louis Colliot-Thélène (Université Paris-Sud)
David Mendes da Costa (University of Bristol)
Tuan Ngo Dac (Université de Paris 13)

x
 List of participants xi

Ulrich Derenthal (Universität München)

Fred Diamond (King’s College London)
Rainer Dietmann (Royal Holloway, University of London)
Luis Dieulefait (Universitat de Barcelona)
Gerd Faltings (Max-Planck-Institut für Mathematik)
Ivan Fesenko (University of Nottingham)
Nuno Freitas (Universitat de Barcelona)
Wojciech Jerzy Gajda (The Adam Mickiewicz University)
David Geraghty (Institute for Advanced Study)
Sergey Gorchinskiy (Steklov Mathematical Institute)
Frank Gounelas (Humboldt University)
Xavier Guitart (Universität Duisburg-Essen)
Shuvra Gupta (University of Iowa)
Shai Haran (Technion – Israel Institute of Technology)
Michael Harris (Institut de Mathématiques de Jussieu)
Roger Heath-Brown (University of Oxford)
Florian Herzig (University of Toronto)
Alexander Ivanov (Universität Heidelberg)
Mikhail Ivanov (Saint-Petersburg State University)
Andrew Kresch (Universität Zürich)
Lars Kühne (SNS Pisa)
Boris Kunyavskii (Bar-Ilan University)
Leonid Kuzmin (National Research Center Kurchatov Institute)
Kai-Wen Lan (University of Minnesota)
Dmitry Logachev (Universidad Simon Bolivar)
Oliver Lorscheid (IMPA)
Daniel Loughran (Leibniz Universität Hannover)
Yuri Manin (Max-Planck-Institut für Mathematik)
Oscar Marmon (Georg-August-Universität Göttingen)
Lilian Matthiesen (University of Bristol)
David McKinnon (University of Waterloo)
Boris Moroz (Universität Bonn)
Jeanine Van Order (The Hebrew University of Jerusalem)
Denis Osipov (Steklov Mathematical Institute)
Ambrus Pál (Imperial College London)
Aftab Pande (Cidade Universitária Ilha do Fundao)
Alexej Parshin (Steklov Mathematical Institute)
Florian Pop (University of Pennsylvania)
Dinakar Ramakrishnan (Caltech)
Giovanni Rosso (Université Paris 13)
xii List of participants

Mohamed Saidi (Exeter University)

Per Salberger (Chalmers University of Technology)
Damaris Schindler (University of Bristol)
Mehmet Haluk Sengun (University of Warwick)
Evgeny Shinder (Max-Planck-Institut für Mathematik)
Ceclia Salgado Guimaraes da Silva (UFRJ)
Alexei Skorobogatov (Imperial College London)
Arne Smeets (Université Paris-Sud 11)
Aleksander Smirnov (Steklov Institute of Mathematics in St. Petersburg)
Efthymios Sofos (University of Bristol)
Cesar Alejandro Soto Posada (Universität Tübingen)
Mike Swarbrick Jones (University of Bristol)
Jack Thorne (Harvard University)
Jacques Tilouine (Université Paris 13)
Yuri Tschinkel (New York University)
Pankaj Hemant Vishe (University of York)
Sergei Vostokov (University of St. Petersburg)
Yosuhiro Wakabayashi (Kyoto University)
Gabor Wiese (Université du Luxembourg)
Nicholas Williams (University of Exeter)
Jean-Pierre Wintenberger (Université Strasbourg)
Trevor Wooley (University of Bristol)
Yanhong Yang (Universität Mainz)
Adrin Zenteno (Universidad Nacional Autónoma de México)
 Trimester Seminar

January 8, Jeanine Van Order, Iwasawa main conjectures for G L(2) via
Howard’s criterion (abstract). In this talk, I will present the Iwasawa main
conjectures for Hilbert modular eigenforms of parallel weight two in dihe-
dral or anticyclotomic extensions of CM fields. The first part will include an
overview of known results, as well as some discussion of open problems and
applications (e.g. to bounding Mordell-Weil ranks), and should be accessible
to the non-specialist. The second part will describe the p-adic L-functions
in more detail, as well as the non-vanishing criterion of Howard (and its
implications for the main conjectures).
January 15, Oliver Lorscheid, A blueprinted view on F1 -geometry
(abstract). A blueprint is an algebraic structure that “interpolates” between
multiplicative monoids and semirings. The associated scheme theory applies
to several problems in F1 -geometry: Tits’s idea of Chevalley groups and build-
ings over F1 , Euler characteristics as the number of F1 -rational points, total
positivity, K-theory, Arakelov compactifications of arithmetic curves; and it
has multiple connections to other branches of algebraic geometry: Lambda-
schemes (after Borger), log schemes (after Kato), relative schemes (after Toen
and Vaquie), congruence schemes (after Berkovich and Deitmar), idempotent
analysis, analytic spaces and tropical geometry. After a brief overview and an
introduction to the basic definitions of this theory, we focus on the combinato-
rial aspects of blue schemes. In particular, we explain how to realize Jacques
Tits’s idea of Weyl groups as Chevalley groups over F1 and Coxeter complexes
as buildings over F1 . The central concepts are the rank space of a blue scheme
and the Tits category, which make the idea of “F1 -rational points” rigorous.
January 16, Jean-Pierre Wintenberger, Introduction to Serre’s modu-
larity conjecture (abstract). This lecture is intended for non-specialists. We
state Serre’s modularity conjecture and give some consequences and hints on
its proof.

xiii
xiv Trimester Seminar

January 17, Henri Carayol, Realization of some automorphic forms and

rationality questions (Part I) (abstract). In this first (and mostly introductory)
talk I shall recall some (well-known) facts on the realization of automorphic
forms in the cohomology groups of some geometric objects, and the rela-
tion with the arithmetic properties of such forms. I shall introduce locally
symmetric varieties, Shimura varieties and the more exotic Griffiths-Schmid
varieties. I shall discuss the case of automorphic forms whose archimedean
component is a limit of discrete series. In the case of degenerate limits, the
only known realization uses the coherent cohomology of Griffiths-Schmid
varieties.
January 21, Günter Harder, Modular construction of mixed motives
and congruences (Part I) (abstract). Starting from a Shimura variety S and
its compactification S v we construct certain objects, which can be thought of
as being mixed motives. These mixed motives give rise to certain elements
of E xt1 groups. We can use the theory of Eisenstein cohomology to compute
the Hodge-de-Rham extension classes of these extensions. We also have some
conjectural formulas for these extensions as Galois modules. Assuming the
correctness of these formulas for the Galois extension class we can derive con-
gruences between eigenvalues of Hecke operators acting on the cohomology of
different arithmetic groups, these congruences are congruences modulo primes
l dividing certain special values of L-functions. These congruences have been
verified experimentally in many cases. They imply the reduciblity of certain
Galois-representations mod l.
January 22, Yuri Manin, Non-commutative generalized Dedekind
symbols (abstract). Classical Dedekind symbol was introduced and stud-
ied in connection with functional equation of Dedekind eta-function. Later
it was generalized and had multiple applications, in particular to topologi-
cal invariants. I will define and study generalized Dedekind symbols with
values in non-necessarily commutative groups, extending constructions of
Sh. Fukuhara done in the commutative context. Basic examples of such sym-
bols are obtained by replacing period integrals of modular forms by iterated
period integrals. I will also explain the interpretation of such symbols in terms
of non-commutative 1-cocycles.
January 23, Henri Carayol, Realization of some automorphic forms
and rationality questions (Part II) (abstract). This talk is a continuation of
part I.
January 24, Michael Harris, Eisenstein cohomology and construction
of Galois representations (Part I) (abstract). I will report on some aspects of
the joint work with Lan, Taylor, and Thorne, which attaches compatible fami-
lies of l-adic Galois representations to a cuspidal cohomological automorphic
 Trimester Seminar xv

representation of G L(n) of a CM field. Earlier work by many authors had

treated the case where the automorphic representation is dual to its image
under complex conjugation; under this hypothesis, the Galois representa-
tions in question, or closely related representations, can be obtained directly
in the cohomology with twisted coefficients of Shimura varieties attached
to unitary groups. Without the duality hypothesis, this is no longer possi-
ble; instead, the representations are constructed by p-adic approximation of
Eisenstein cohomology classes by cuspidal classes in an appropriate (infinite-
dimensional) space of p-adic modular forms. The lectures will concentrate on
the construction of Eisenstein classes, the relation to p-adic modular forms and
the definition of Galois representations by p-adic approximation.
January 25, Michael Harris, Eisenstein cohomology and construction
of Galois representations (Part II) (abstract). This talk is a continuation of
part I.
January 28, Günter Harder, Modular construction of mixed motives
and congruences (Part II) (abstract). This talk is a continuation of part I.
January 28, Fred Diamond, The weight part of Serre’s conjecture for
G L(2) over totally real fields (abstract). I will review the statement of the
weight part of Serre’s conjecture for G L(2) over totally real fields. I will
describe what has been proved by Gee and his coauthors, and give a brief
overview of the methods.
January 30, Luis Dieulefait, Non-solvable base change for G L(2)
(abstract). We will show that any classical cuspidal modular form can be lifted
to any totally real number field. The proof uses a recent Modularity Lifting
Theorem proved by Barnet-Lamb, Gee, Geraghty and Taylor (plus a variant of
it proved by Gee and the speaker) and another one by Kisin that is used in the
“killing ramification” step. The core of the proof is the construction of a “safe”
chain of congruences linking to each other any given pair of cuspforms. The
safe chain that we will construct is also a key input in the proof of other cases
of Langlands functoriality, but this will be explained in another talk (see the
abstracts for the conference week).
January 31, Kai-Wen Lan, Galois representations for regular algebraic
cuspidal automorphic representations over CM fields (part I) (abstract). I
will report on my joint work with Michael Harris, Richard Taylor and Jack
Thorne on the construction of p-adic Galois representations for regular alge-
braic cuspidal automorphic representations of G L(n) over CM (or totally real)
fields, without hypotheses on self-duality or ramification. (This should be
considered as part III of a series of four talks, the first two being given by
Michael Harris in the previous week.)
xvi Trimester Seminar

February 1, Kai-Wen Lan, Galois representations for regular algebraic

cuspidal automorphic representations over CM fields (part II) (abstract).
This is a continuation of part I.
February 12, David Geraghty, The Breuil-Mezard conjecture for
quaternion algebras (abstract). The Breuil-Mezard conjecture relates the
complexity of certain deformation rings for mod p representations of the
Galois group of Q p with the representation theory of G L 2 (F p ). Most cases
of the conjecture were proved by Kisin who established a link between the
conjecture and modularity lifting theorems. In this talk I will discuss a gen-
eralization of the conjecture to quaternion algebras (over an arbitrary finite
extension of Q p ) and show how it follows from the original conjecture for
G L(2). This is a joint work with Toby Gee.
February 18, Boris Kunyavskii, Geometry and arithmetic of word maps
in simple matrix groups (abstract). We wil discuss various geometric and
arithmetic properties of matrix equations of the form

P(X 1 , . . . , X d ) = A,

where the left-hand side is an associative non-commutative monomial in X i ’s

and their inverses, and the right-hand side is a fixed matrix. Solutions are
sought in some group G ⊂ G L(n, R). We will focus on the case where
the group G is simple, or close to such. We will give a survey of classical
and recent results and open problems concerning this equation, concentrating
around the following questions (posed for geometrically and/or arithmetically
interesting rings and fields R): is it solvable for any A?, is it solvable for a
“typical” A?, does it have “many” solutions?, does the set of solutions pos-
sess “good” local–global properties?, to what extent does the set of solutions
depend on A?. The last question will be discussed in some detail for the case
G = S L(2, q) and d = 2, where criteria for equidistribution were obtained in
our recent joint work with T. Bandman.
February 19, Shai Haran, Non-additive geometry (abstract). We give a
language for algebraic geometry based on non-additive generalized rings. In
this language, number fields look more like curves over a finite field. The ini-
tial object of generalized rings is the “field with one element”. This language
“sees” the real and complex primes of a number field, and there is a compacti-

ficaton Spec O K of Spec O K , O K being the ring of integers of a number field

K . The arithmetic surface Spec O K  Spec  O K exists and is not reduced to
its diagonal. And yet most of the Grothendieck algebraic geometry works with
generalized rings replacing commutative rings.
 Trimester Seminar xvii

February 20, Aleksander Smirnov, The internal and external prob-

lems of algebraic geometry over F1 (abstract). An introduction to Durov’s
approach will be given. The theory will be illustrated with several explicit
examples. Besides, we plan to discuss some problems caused by both the
development of the theory and the demands of its applications.
February 21, Marina Viazovska, CM values of higher Green’s func-
tions and regularized Petersson products (abstract). Higher Green functions
are real-valued functions of two variables on the upper half-plane, which are
bi-invariant under the action of a congruence subgroup, have a logarithmic sin-
gularity along the diagonal, and satisfy the equation  f = k(1−k) f ; here  is
a hyperbolic Laplace operator and k is a positive integer. The significant arith-
metic properties of these functions were disclosed in the paper of B. Gross and
D. Zagier “Heegner points and derivatives of L-series” (1986). In the particular
√
case when k = 2 and one of the CM points is equal to −1, the conjecture
has been proved by A. Mellit in his Ph.D. thesis. In this lecture we prove that
conjecture for arbitrary k, assuming that all the pairs of CM points lie in the
same quadratic field. The two main parts of the proof are as follows. We first
show that the regularized Petersson scalar product of a binary theta-series and
a weight one weakly holomorphic cusp form is equal to the logarithm of the
absolute value of an algebraic integer and then prove that the special values
of weight k Green’s function, occurring in the conjecture of Gross and Zagier,
can be written as the Petersson product of that type, where the form of weight
one is the k − 1st Rankin-Cohen bracket of an explicitly given holomorphic
modular form of weight 2 − 2k and a binary theta-series. Algebraicity of regu-
larized Petersson products was also proved at about the same time by W. Duke
and Y. Li by a different method; however, our result is stronger since we also
give a formula for the factorization of the algebraic number in question.
February 26, David Mendes da Costa, Integral points on elliptic curves
and the Bombieri-Pila bounds (abstract). In 1989, Bombieri and Pila found
upper bounds for the number of integer points of (naive exponential) height at
most B lying on a degree d affine plane curve C. In particular, these bounds are
both uniform with respect to the curve C and the best possible with this con-
straint. It is conjectured though that if we restrict to curves with positive genus
then the bounds can be broken. In this talk we shall discuss progress towards
this conjecture in the case of elliptic curves and an application to counting
rational points on degree 1 del Pezzo surfaces.
February 27, Lars Kühne, Effective and uniform results of André-Oort
type (abstract). The André-Oort Conjecture (AOC) states that the irreducible
components of the Zariski closure of a set of special points in a Shimura
variety are special subvarieties. Here, a special variety means an irreducible
xviii Trimester Seminar

component of the image of a sub-Shimura variety by Hecke correspondence.

The AOC is an analogue of the classical Manin-Mumford conjecture on the
distribution of torsion points in abelian varieties. In fact, both conjectures are
considered as special instances of the far-reaching Zilber-Pink conjecture(s).
I will present a rarely known approach to the AOC that goes back to Yves
André himself: Before the model-theoretic proofs of the AOC in certain cases
by the Pila-Wilkie-Zannier approach, André presented in 1998 the first proof
of the AOC in a non-trivial case, namely, a product of two modular curves.
In my talk, I discuss several results in the style of André’s method, allowing
to compute all special points in a non-special curve of a product of two mod-
ular curves. These results are effective – as opposed to the results that could
be obtained by the Pila-Wilkie-Zannier approach – and have also the further
advantage of being uniform in the degrees of the curve and its field of defini-
tion. For example, this allows to show that, in fact, there are no two singular
moduli x and y satisfying x + y = 1.
February 28, Nuno Freitas, Fermat-type equations of signature (r, r, p)
(abstract). In this talk I plan to discuss how a modular approach via the Hilbert
cusp-forms can be used to attack equations of the form x r + y r = C z p , where r
is a fixed prime and p varies. We first relate a possible solution of that equation
to solutions of several related Diophantine equations over certain totally real
fields F. Then we attach Frey curves E over F to the solutions of the latter
equations. After proving modularity of E and irreducibility of certain Galois
representations attached to E we can use the modular approach. We apply the
method to solve equations in the particular case of signature (13, 13, p).
March 4, Leonid Kuzmin, l-adic regulator of an algebraic number field
and Iwasawa theory (abstract). We give a new definition of the l-adic regula-
tor, which makes sense for any (not necessarily totally real) algebraic number
field, present a few results and conjectures, relating to that notion, and discuss
the behaviour of the l-adic regulator in a Zl -cyclotomic extension of the field.
March 4, Alexander Ivanov, Arithmetic and anabelian geometry of
stable sets of primes in number fields (abstract). We define a new class of
sets – stable sets – of primes in number fields. For example, Chebotarev sets
are very often stable. Those sets have positive (but arbitrary small) Dirichlet
density and generalize the notion of a set of density 1, in the sense that arith-
metic theorems like certain Hasse principles, the Grünwald-Wang theorem, the
Riemann existence theorem, etc. hold for them. Geometrically, this allows to
give examples of infinite sets with arbitrarily small positive density such that
the corresponding arithmetic curves are algebraic K (π, 1) and, using some fur-
ther ideas, to generalize (a part of) the Neukirch-Uchida birational anabelian
theorem to stable sets.
 Trimester Seminar xix

March 5, Tuan Ngo Dac, On the problem of counting shtukas (abstract).

I will introduce the stacks of shtukas, explain its role in the Langlands program,
and then report on my work on the problem of counting shtukas.
March 7, Rajender Adibhatla, Modularity of certain two-dimensional
mod p n representations of G Q (abstract). For an odd rational prime p and
integer n > 1, we consider certain continuous representations

ρn : G Q → G L 2 (Z/ pn Z)

with fixed determinant, whose local restrictions “look like arising” from
modular Galois representations, and whose mod p reductions are odd and
irreducible. Under suitable hypotheses on the size of their images, we use
deformation theory to lift ρn to ρ in characteristic 0. We then invoke a
modularity lifting theorem of Skinner-Wiles to show that ρ is modular.
March 11, Frank Gounelas, Rationally connected varieties and free
curves (abstract). The first part of this talk will be a general introduction to
rationally connected varieties. I will then discuss various ways in which a vari-
ety can be “connected by curves of a fixed genus, mimicking the notion of
rational connectedness”. At least in characteristic zero, in the specific case
of the existence of a single curve with a large deformation space of mor-
phisms to a variety implies that the variety is in fact rationally connected. Time
permitting I will discuss attempts to show this result in positive characteristic.
March 12, Tommaso Centeleghe, On the decomposition of primes in
torsion fields of an elliptic curve (abstract). Let E be an elliptic curve over a
number field K and N be a positive integer. In this talk we consider the prob-
lem of describing how primes P of K of good reduction for E and away from
N decompose in the extension K (E[N ])|K . As it turns out, the class Frob P in
Gal(K (E[N ]|K )) can be completely described, apart of finitely many primes
P, in terms of the error term a P (E) and the j -invariant of E. The Hilbert
class polynomials, associated to imaginary quadratic orders, play a role in the
description. The main result relies on a theorem on elliptic curves over finite
fields.
March 13, Shuvra Gupta, Noether’s problem and rationality of invari-
ant spaces (abstract). In the early 1900s Emmy Noether asked the following
question: If a group G acts faithfully on a vector space V (over a field k), is
the field of invariants k(V )G rational, i.e. purely transcendental over k? The
answer (for k = Q or a number field) in general is no, and we will discuss
some consequences and variants of Noether’s problem. We will also discuss
the problem when the field k is algebraically closed, and techniques of testing
rationality of the fields of invariants using unramified cohomology groups.
xx Trimester Seminar

March 14, Mohamed Saidi, Some problems/results related to the

Grothendieck anabelian section conjecture (abstract). I will discuss some
(major) problems related to the Grothendieck anabelian section conjecture.
I will discuss two new results related to these problems and the section con-
jecture. First result: there exists a local–global principle for torsors under the
geometric prosolvable fundamental group of a proper hyperbolic curve over a
number field. Second result: the passage in the section conjecture from number
fields to finitely generated fields is possible under the assumption of finiteness
of suitable Shafarevich-Tate groups.
March 18, Sergey Gorchinskiy, Parameterized differential Galois
theory (abstract). Classical Galois theory studies symmetry groups of solu-
tions of algebraic equations. Differential Galois theory studies symmetry
groups of solutions of linear differential equations. We discuss the so-called
parameterized differential Galois theory which studies symmetry groups of
solutions of linear differential equations with parameters. The groups that
arise are linear differential groups given by differential equations (not nec-
essarily linear) on functions in parameters. We also discuss, in this connection,
derivations on Abelian categories and differential Tannakian categories.
March 19, Alexandru Buium, The concept of linearity for an arithmetic
differential equation (abstract). The concept of an ordinary differential equa-
tion has an arithmetic analogue in which the derivation operator is replaced
by a Fermat quotient operator. We would like to understand which arithmetic
differential equations should be considered as being “linear”.
Classical linear differential equations arise from differential cocycles of lin-
ear algebraic groups into their Lie algebras and their differential Galois groups
are algebraic groups with coefficients in the field of constants. On the other
hand one can prove that there are no such cocycles in the arithmetic context.
This leads one to introduce, in the arithmetic context, a new concept of “Lie
algebra”, “cocycles”, “linear” equations, and “differential Galois groups”; the
latter can be viewed as subgroups of the general linear group with coefficients
in the algebraic closure of the “field with one element”.
March 20, Roman Budylin, Adelic Bloch formula (abstract). Chern class
c2 (X ) is involved in the functional equation for two-dimensional schemes. To
get functional equation by the Tate method we need a local decomposition of
the Chern class, satisfying some properties. Bloch proves that the second Chern
class of a vector bundle with trivial determinant can be obtained by the bound-
ary homomorphism for the universal central extension of the sheaf SL(O X ). In
the talk, this construction will be used to get an adelic formula for the second
Chern class in terms of trivializations in scheme points. We will also discuss a
generalization of this formula for cn of vector bundles with ci = 0 for i < n.
 Trimester Seminar xxi

March 21, Alexej Parshin, A generalization of the Langlands corre-

spondence and zeta-functions of a two-dimensional scheme (abstract). We
introduce Abelian Langlands correspondence for algebraic surfaces defined
over a finite field. When the surface is a semi-stable fibration over an alge-
braic curve, we define two operations, automorphic induction and base change,
which connect this correspondence with the classical Langlands correspon-
dence on the curve. Some conjectural properties of these operations imply the
standard theorems for zeta- and L-functions on the surface (analytic continua-
tion and functional equation). In this approach we do not need to use the étale
cohomology theory.
March 22, Denis Osipov, Unramified two-dimensional Langlands
correspondence (abstract). We will describe the local unramified Langlands
correspondence for two-dimensional local fields (following an approach of
M. Kapranov). For this goal, we will construct a categorical analogue of
principal series representations of general linear groups of even degrees over
two-dimensional local fields and describe their properties. The main ingredi-
ent of this construction is a central extension of a general linear group defined
over a two-dimensional local field or over an adelic ring of a two-dimensional
arithmetic scheme. We will prove reciprocity laws for such central extensions,
i.e. splittings of the central extensions over some subgroups defined over rings
constructed by means of points or by integral one-dimensional subschemes of
a two-dimensional arithmetic scheme.
March 25, Rainer Dietmann, On quantitative versions of Hilbert’s
irreducibility theorem (abstract). If f (X, Y ) is an irreducible rational poly-
nomial, then by Hilbert’s irreducibility theorem for infinitely many rational
specialisations of X the resulting polynomial in Y is still irreducible over the
rationals. In this talk we want to discuss quantitative versions of this result,
using recent advances from the determinant method on bounding the number
of points on curves.
March 27, Roger Heath-Brown, Pairs of quadratic forms in 8 variables
(abstract). We show that a smooth intersection of two quadrics in P 7 , defined
over a number field, satisfies the Hasse principle and weak approximation. The
proof is based on the work of Colliot-Thélène, Sansuc and Swinnerton-Dyer
on Chatelet surfaces, which enables one to reduce the problem to a purely local
problem. The first part of the talk will discuss the background and the overall
strategy of the proof, and the second part will look in a little more detail at
some of the methods involved.
April 2, Pankaj Vishe, Cubic hypersurfaces and a version of the circle
method over number fields (abstract). A version of the Hardy-Littlewood
circle method is developed for number fields K and is used to show that any
xxii Trimester Seminar

non-singular projective cubic hypersurface over K of dimension ≥ 8 always

has a K -rational point. This is a joint work with T. Browning.
April 3, Efthymios Sofos, Counting rational points on the Fermat
surface (abstract). In this talk we shall discuss progress towards finding lower
bound for the number of rational points of bounded height on the Fermat
cubic surface. The argument is based on a uniform asymptotic estimate for
the associated counting function on conics.
April 5, Jörg Brüdern and Trevor Wooley, Systems of cubic forms at
the convexity barrier (abstract). We describe recent joint work concerning
the validity of the Hasse principle for systems of diagonal cubic forms. The
number of variables required meets the convexity barrier. Certain features of
our methods are motivated by work of Gowers on Szemeredi’s theorem.
April 8, Mike Swarbrick Jones, Weak approximation on cubic hyper-
surfaces of large dimension (abstract). A natural question in arithmetic
geometry is to investigate weak approximation on varieties. If the dimension
of the variety is large compared to the degree, our most successful tool is the
circle method, however there are cases where using this is not feasible given
our current state of knowledge. In this talk I will sketch a proof that weak
approximation holds for generic cubic hypersurfaces of dimension at least 17,
in particular discussing a fibration method argument that applies to the cases
where the usual application of the circle method is not possible.
April 10, Arne Smeets, Local–global principles for fibrations in torsors
under tori (abstract). This talk is a report on work in progress about local–
global principles for varieties fibred over the projective line. In particular,
we will study the Brauer-Manin obstruction to the Hasse principle and weak
approximation for certain fibrations in torsors under tori, e.g. (multi-)norm
form equations. Our results are conditional on Schinzel’s hypothesis.
April 12, David Mendes da Costa, On uniform bounds for integral
points on elliptic curves (abstract). In 1989, Bombieri and Pila proved that
given a plane algebraic affine curve of degree d there are no more than
O(N 1/d+ ) integral points on the curve within a box of size N × N . More-
over, the implied constant in their bound depended only on the degree of the
curve and not on the equation. Such bounds are, in general, the best possible,
however, it is believed that by restricting to curves which have positive genus
one can do much better. In this talk we consider the problem of improving
these uniform bounds for integral points on elliptic curves. An application of
this work to degree one del Pezzo surfaces will be presented.
 Workshop on Serre’s conjecture

Monday, February 4
10:00 – 11:00 Fred Diamond (London King’s college): Explicit Serre weights
for two-dimensional Galois representations.
11:30 – 12:30 Denis Benois (Université Bordeaux 1): Trivial zeros of p-adic
L-functions and Iwasawa theory.
14:30 – 15:30 Tommaso Centeleghe (Universität Heidelberg): Computing the
number of certain mod p Galois representation.
16:00 – 17:00 Wojciech Gajda (UAM Poznań): Abelian varieties and l-adic
representations.
Tuesday, February 5
10:00 – 11:00 Tobias Berger (University of Sheffield): Eisenstein congruences
and modularity of Galois representations.
11:30 – 12:30 Christophe Breuil (Université de Paris-Sud): Ordinary represen-
tations of GLn (Q p ) and fundamental algebraic representations I.
14:30 – 15:30 Jeanine Van Order (EPFL Lausanne): Critical values of GL(2)
Rankin-Selberg L-functions.
16:00 – 17:00 Luis Dieulefait (Universitat de Barcelona): Some new cases of
Langlands functoriality solved.
Wednesday, February 6
10:00 – 11:00 Aftab Pande (Cidade Universitária Ilha do Fundao): Deforma-
tions of Galois representations and the theorems of Sato-Tate, Lang-Trotter
and others.

xxiii
xxiv Workshop on the Serre’s conjecture

11:30 – 12:30 Florian Herzig (University of Toronto): Ordinary representations

of GLn (Q p ) and fundamental algebraic representations II.
14:30 – 15:30 Mehmet Sengun (University of Warwick): Mod p Cohomology
of Bianchi groups and Mod p Galois representations.
16:00 – 17:00 Jack Thorne (Harvard University): Symmetric power functori-
ality for GL(2).
Thursday, February 7
10:00 – 11:00 Dinakar Ramakrishnan (California Institute of Technology):
Picard modular surfaces, residual Albanese quotients, and rational points.
11:30 – 12:30 Jacques Tilouine (Université Paris 13): Image of Galois and
congruence ideals, a program of a joint work with H. Hida.
Friday, February 8
10:00 – 11:00 David Geraghty (IAS Princeton): Modularity lifting beyond the
“numerical coincidence” of the Taylor-Wiles method.
11:30 – 12:30 Florian Pop (University of Pennsylvania): Faithful representa-
tions of absolute Galois groups.

Abstracts
Denis Benois. Trivial zeros of p-adic L-functions and Iwasawa theory. We
prove that the expected properties of Euler systems imply quite general Mazur-
Tate-Teitelbaum type formulas for derivatives of p-adic L-functions. We also
discuss the Iwasawa theory of p-adic representations in the trivial zero case.
Tobias Berger. Eisenstein congruences and modularity of Galois representa-
tions. I will report on joint work with Kris Klosin (CUNY) on congruences of
Eisenstein series and cuspforms modulo prime powers and its application in
proving the modularity of residually reducible Galois representations.
Tommaso Centeleghe. Computing the number of certain mod p Galois rep-
resentations. We report on computations aimed to obtain, for a given prime
p, the number R( p) of two-dimensional odd mod p Galois representations of
G Q which are irreducible and unramified outside p. Thanks to Serre’s conjec-
ture, this amounts to a computation of the number of non-Eisenstein systems
of Hecke eigenvalues arising from mod p modular forms of level one. Using
well-known dimension formulas for modular forms (and the fact that any mod
p eigensystem can be Tate-twisted to one arising from a weight ≤ p + 1),
 Workshop on Serre’s conjecture xxv

an explicit upper bound U ( p) of R( p) can be drawn. While discussing the

reasons which might make the “error term” U ( p) − R( p) large, we stress how,
in practice, one can control it from above using only one of the first Hecke
operators. This gives a lower bound for R( p), which coincides in many cases
with R( p) itself.
Fred Diamond. Explicit Serre weights for two-dimensional Galois represen-
tations. I will discuss joint work with Savitt making the set of Serre weights
more explicit for indecomposable two-dimensional mod p representations of
Galois groups over ramified extensions of Q p . In particular the results indicate
a structure on the set of such weights.
Luis Dieulefait. Some new cases of Langlands functoriality solved. We com-
bine the method of Propagation of Automorphy with recent Automorphy
Lifting Theorems (A.L.T.) of Barnet-Lamb, Gee, Geraghty, and Taylor to
prove some new cases of Langlands functoriality (tensor products and sym-
metric powers). In particular, we establish automorphy for lots of Galois
representations of G Q of arbitrarily large dimension (and their base changed
counterparts). We also prove some variants of the available A.L.T., which
are needed at some steps of our proof. Remark: Some technical improve-
ments required to extend some A.L.T. to the case of “small primes” were
accomplished with the kind cooperation of R. Guralnick and T. Gee.
Wojciech Gajda. Abelian varieties and l-adic representations. We will discuss
monodromies for abelian varieties, and independence (in the sense of Serre)
for families of some geometric l-adic representations over finitely generated
fields.
David Geraghty. Modularity lifting beyond the “numerical coincidence” of
the Taylor-Wiles method. Modularity lifting theorems have proven very useful
since their invention by Taylor and Wiles. However, as explained in the intro-
duction to Clozel-Harris-Taylor, they only apply in situations where a certain
numerical coincidence holds. In this talk, I will describe a method to overcome
this restriction. The method is conditional on the existence of Galois represen-
tations associated to integral cohomology classes (which can be established in
certain cases). This is joint work with Frank Calegari.
Aftab Pande. Deformations of Galois Representations and the Theorems of
Sato-Tate, Lang-Trotter and others. We construct infinitely ramified Galois rep-
resentations ρ such that the sequences al (ρ) have distributions in contrast to
the statements of Sato-Tate, Lang-Trotter and others. Using similar methods,
we deform a residual Galois representation for number fields and obtain an
xxvi Workshop on the Serre’s conjecture

infinitely ramified representation with very large image, generalizing a result

of Ramakrishna.
Florian Pop. Faithful representations of absolute Galois groups. In his
“Esquisse d’un Programme” Grothendieck suggested to study the absolute
Galois group of the rationals via its representations on the algebraic funda-
mental group of natural categories of varieties, e.g. the Teichmueller modular
tower. This leads to the intensive study of the so-called Grothendieck-
Teichmueller group and its variants, and the I/OM (Ihara/Oda-Matsumoto
conjecture). I plan to explain variants of I/OM, and discuss its state of
the art.
Dinakar Ramakrishnan. Picard modular surfaces, residual Albanese quotients,
and rational points. The Picard modular surfaces X are at the crossroads
of rich interplay between geometry, Galois representations, and automorphic
forms on G = U (2, 1) associated to an imaginary quadratic field K . The
talk will introduce an ongoing project with M. Dimitrov on the quotients
of the albanese variety Alb(X ) coming from residual automorphic forms on
G, give examples with finite Mordell-Weil group, and investigate possible
consequences, inspired by classical arguments of Mazur, for the K -rational
points on X .
Mehmet Sengun. Mod p Cohomology of Bianchi Groups and Mod p Galois
Representations. Given an imaginary quadratic field K with ring of integers
R, consider the Bianchi group GL(2, R). It is suspected since the numerical
investigations of Fritz Grunewald in the late 1970s that there is a connection
between the Hecke eigenclasses in the mod p cohomology of (congruence
subgroups of) Bianchi groups and the two-dimensional continuous mod p rep-
resentations of the absolute Galois group of K . Most of the basic tools used
for establishing this connection (and its surrounding problems) in the classi-
cal setting fail to work in the setting of Bianchi groups. The situation has an
extra layer of complication by the fact that there are “genuinely mod p” Hecke
eigenvalue systems, resulting from the existence of torsion in the integral
cohomology. In this talk I will elaborate on the above, presenting numer-
ical examples for illustration. Towards the end, I will also talk about how
the “even” two-dimensional continuous mod p representations of the absolute
Galois group of Q come into the picture.
Jack Thorne. Symmetric power functoriality for GL(2). We discuss some
new automorphy lifting theorems, and their applications to the existence of
new cases of Langlands’ functoriality for GL(2). This is joint work with
L. Clozel.
 Workshop on Serre’s conjecture xxvii

Jacques Tilouine. Image of Galois and congruence ideals, a programme of a

joint work with H. Hida. In a recent preprint, H. Hida showed that the image
of the Galois representation associated to a non-CM Hida family contains a
congruence subgroup of GL(2) over , whose level is given in terms of p-
adic L-functions. We try to generalize this to Hida families for bigger groups,
replacing p-adic L-functions by congruence ideals.
Jeanine Van Order. Critical values of GL(2) Rankin-Selberg L-functions. The
aim of this talk is to explain the subtle but powerful link between the algebraic-
ity of critical values of automorphic L-functions, the existence of associated
p-adic L-functions, and the generic non-vanishing of these values, particularly
in the setting of Rankin-Selberg L-functions of GL(2) over a totally real num-
ber field. More precisely, the aim is to explain how to extend the conjectures of
Mazur to the non self-dual setting, thereby extending the works of Vatsal, Cor-
nut and Cornut-Vatsal, via a combination of techniques from Iwasawa theory,
analytic number theory and the theory of automorphic forms. If time permits,
then some open problems will also be introduced.
 The research conference

Monday, April 15
09:30 – 10:30 Trevor Wooley (University of Bristol): Applications of efficient
congruencing to rational points.
11:00 – 12:00 Per Salberger (Chalmers University of Technology): Heath-
Brown’s determinant method and Mumford’s geometric invariant theory.
13:30 – 14:30 Jeanine Van Order (EPFL): Stable Galois averages of Rankin-
Selberg L-values and non-triviality of p-adic L-functions.
15:00 – 16:00 Przemyslaw Chojecki (Institut Mathématique de Jussieu): On
mod p non-abelian Lubin-Tate theory for GL(2).
Tuesday, April 16
09:30 – 10:30 Alexei Skorobogatov (Imperial College London): Applications
of additive combinatorics to rational points.
11:00 – 12:00 Yuri Bilu (IMB Université Bordeaux I): Integral points on
modular curves.
13:30 – 14:30 Ulrich Derenthal (Universität München): Counting points over
imaginary quadratic number fields.
15:00 – 16:00 Oscar Marmon (Universität Göttingen): The density of twins of
k-free numbers.
Wednesday, April 17
09:30 – 10:30 Jörg Brüdern (Universität Göttingen): Random Diophantine
equations.

xxviii
 The research conference xxix

11:00 – 12:00 Florian Pop (University of Pennsylvania): Local–global princi-

ples for rational points.
13:30 – 14:30 Mohamed Saidi (Exeter University): On the anabelian section
conjecture over finitely generated fields.
15:00 – 16:00 Victor Abrashkin (University of Durham): p-extensions of local
fields with Galois groups of nilpotence class less than p.
Thursday, April 18
09:30 – 10:30 Ambrus Pál (Imperial College London): New two-dimensional
counter-examples to the local–global principle.
11:00 – 12:00 David McKinnon (University of Waterloo): Approximating
points on varieties.
13:30 – 14:30 Tim Browning (University of Bristol): Norm forms as products
of linear polynomials, I.
15:00 – 16:00 Lilian Matthiesen (University of Bristol): Norm forms as
products of linear polynomials, II.
Friday, April 19
09:30 – 10:30 Gabor Wiese (Université du Luxembourg): Symplectic Galois
representations and applications to the inverse Galois problem.
11:00 – 12:00 Damaris Schindler (University of Bristol): Manin’s conjecture
for certain smooth hypersurfaces in biprojective space.
13:30 – 14:30 Roger Heath-Brown (University of Oxford): Simultaneous
representation of pairs of integers by quadratic forms.
15:00 – 16:00 Jean-Louis Colliot-Thélène (Université Paris-Sud): Strong
approximation in a family.

Abstracts
Victor Abrashkin. p-extensions of local fields with Galois groups of nilpotence
class less than p. Nilpotent analogue of the Artin-Schreier theory was devel-
oped by the author about twenty years ago. It has found already applications in
an explicit description of the ramification filtration modulo p-th commutators
and the proof of an analogue of the Grothendieck Conjecture for local fields.
We remind basic constructions of this theory and indicate further progress in
the study of local fields of mixed characteristic, especially, higher dimensional
local fields.
xxx The research conference

Yuri Bilu. Integral points on modular curves. The problem of determination

of rational points on modular curves reduces grosso mode to three types of
curves of prime level, corresponding to three types of maximal subgroups of
the linear group GL2 (F p ): the curve X 0 ( p), corresponding to the Borel sub-
group; the curve X sp + ( p), corresponding to the normalizer of a split Cartan

subgroup; the curve X ns + ( p), corresponding to the normalizer of a non-split

Cartan subgroup. The rational points on the curves of the first two types are
determined (almost) completely: Mazur (1978), B.-Parent-Rebolledo (2012).
In particular, it is proved that for p > 13 the rational points are either
cusps or the CM-points. Little is known, however, about the rational points
+ ( p). I will speak about recent progress in a simpler problem: classifica-
on X ns
tion of integral points on X ns + ( p) (i.e. rational points P such that j (P) ∈ Z).

My students Bajolet and Sha obtained a rather sharp upper bound for the
size of integral points. Also, in a joint work with Bajolet we proved that for
7 < p < 71 there are no integral points on X ns + ( p) other than the CM-points;

this improves on a recent work of Schoof and Tzanakis, who proved that for
p = 11.
Tim Browning, Lilian Matthiesen. Norm forms as products of linear polyno-
mials. We report on recent progress using additive combinatorics to prove
the Hasse principle and weak approximation for certain varieties defined by
systems of equations involving norm forms. This is used to show that the
Brauer-Manin obstruction controls weak approximation on normic bundles of
the shape N K (x 1 , . . . , xn ) = P(t), where P(t) is a product of linear polyno-
mials all defined over the rationals and K is an arbitrary degree n extension of
the rationals.
Jörg Brüdern. Random Diophantine equations. We address the classical ques-
tions, concerning diagonal forms with integer coefficients. Does the Hasse
principle hold? If there are solutions, how many? If there are solutions, what is
the size of the smallest solutions? In a joint work with Dietmann, nearly opti-
mal answers to such questions were obtained for almost all forms (in the sense
typically attributed to “almost all” in the analytic theory of numbers) provided
that the number of variables exceeds three times the degree of the forms under
consideration.
Przemyslaw Chojecki. On mod p non-abelian Lubin-Tate theory for GL(2).
Non-abelian Lubin-Tate theory for GL(2) describes the l-adic cohomology
of the Lubin-Tate tower for GL(2, Q p ) in terms of the Langlands program.
Until recently, all results were stated under the assumption that p is different
from l. We discuss the case when l = p and give a partial description of the
 The research conference xxxi

mod p étale cohomology of the Lubin-Tate tower in terms of the mod p local
Langlands correspondence.
Jean-Louis Colliot-Thélène. Strong approximation in a family. Let ai (t), i =

1, 2, 3 and p(t) be polynomials in Z[t]. Assume the product p(t) ai (t) is not

constant and has no square factor. Consider the equation ai (t)xi2 = p(t).

Assume that for all t ∈ R the real conic ai (t)xi = 0 has point over R.
2

Then strong approximation holds for the integral solutions of this equation.
The set of solutions with coordinates in Z is dense in the product of integral
local solutions at all finite primes. In particular, there is a local–global principle
for integral points. The case where all the ai (t) are constant was dealt with in
an earlier paper with F. Xu (Beijing). The result above is itself a special case
of a general theorem on families of homogenous spaces of semisimple groups,
obtained jointly with D. Harari (Paris-Sud).
Ulrich Derenthal. Counting points over imaginary quadratic number fields. For
Fano varieties over number fields, the distribution of rational points is predicted
by Manin’s conjecture. One approach uses universal torsors and Cox rings;
over the field Q of rational numbers, this was started by Salberger for toric
varieties and was extensively studied for many examples of del Pezzo surfaces.
In this talk, I present an extension of this approach to imaginary quadratic
number fields, in particular for some singular quartic del Pezzo surfaces (joint
work with C. Frei).
Roger Heath-Brown. Simultaneous representation of pairs of integers by
quadratic forms. This is joint work with Lillian Pierce. We examine the simul-
taneous integer equations Q 1 (x1 , . . . , xn ) = m 1 and Q 2 (x1 , . . . , xn ) = m 2 ,
and show, under a smoothness condition, that “almost all” pairs m 1 , m 2 which
have local representations also have a global solution, as soon as n is at least 5.
One can use this to attack Q 1 (x1 , . . . , xn ) = Q 2 (x1 , . . . , xn ) = 0 when n is at
least 10. The proof uses a two-dimensional circle method with a Kloostermann
refinement.
Oscar Marmon. The density of twins of k-free numbers. For k ≥ 2, we con-
sider the number Ak (Z ) of positive integers n ≤ Z such that both n and
n + 1 are k-free. In joint work with Dietmann, we prove an asymptotic formula
Ak (Z ) = ck Z + O(Z 14/9k+ε ), where the error term improves upon previously
known estimates. The main tool used is the approximative determinant method
of Heath-Brown.
David McKinnon. Approximating points on varieties. Famous theorems of
Roth and Liouville give bounds on how well algebraic numbers can be
xxxii The research conference

approximated by rational numbers. In my talk, I will describe not-yet-famous

generalizations of these theorems to arbitrary algebraic varieties, giving
bounds on how well algebraic points can be approximated by rational points.
This is joint work with Mike Roth of Queen’s University, who is not yet related
to famous-theorem Roth.
Ambrus Pál (Imperial College London). New two-dimensional counterex-
amples to the local–global principle. In my talk I will describe some new
two-dimensional counter-examples to the local–global principle.
Florian Pop. Local–global principles for rational points. One of the most excit-
ing developments originating from capacity theory (Fekete-Szego, Robinson,
Cantor, Rumely) is Rumely’s local–global principle, and its applications to the
solvability of incomplete global Skolem problems (Cantor-Roquette, Roquette,
and Moret-Bailly). My talk will be about complete Skolem problems and
how they relate to some recent “complete capacity theory” results for curves
developed by Rumely.
Mohamed Saidi. On the anabelian section conjecture over finitely generated
fields. After introducing the section conjecture and some basic facts I will
discuss in the first part of the talk a conditional result on the section con-
jecture over number fields. In the second part of the talk I will discuss the
following result: the section conjecture holds over all finitely generated fields
if it holds over all number fields, under the condition of finiteness of suitable
Tate-Shafarevich groups.
Per Salberger. Heath-Brown’s determinant method and Mumford’s geometric
invariant theory. Heath-Brown’s p-adic determinant method is used to count
rational points on hypersurfaces and was extended to subvarieties of codimen-
sion > 1 in projective space by Broberg and the author. The determinants
that occur give rise to embeddings of Hilbert schemes in Grassmannians,
which enables the use of techniques from geometric invariant theory. We
describe some Diophantine applications and an unexpected link to the theory
of Donaldson and Tian on Kähler metrics of constant scalar curvature.
Damaris Schindler. Manin’s conjecture for certain smooth hypersurfaces in
biprojective space. So far, the circle method has been a very useful tool to
prove many cases of Manin’s conjecture. Work of B. Birch back in 1961 estab-
lishes this for smooth complete intersections in projective space as soon as
the number of variables is large enough depending on the degree and num-
ber of equations. In this talk we are interested in subvarieties of biprojective
space. There is not much known so far, unless the underlying polynomials are
 The research conference xxxiii

of bidegree (1, 1). In this talk we present recent work which combines the cir-
cle method with the generalized hyperbola method developed by V. Blomer
and J. Brüdern. This allows us to verify Manin’s conjecture for certain smooth
hypersurfaces in biprojective space of general bidegree.
Alexei Skorobogatov. Applications of additive combinatorics to rational
points. This is a joint work with Y. Harpaz and O. Wittenberg. In 1982 Colliot-
Thélène and Sansuc noticed that Schinzel’s Hypothesis (H), used in a fibration
method going back to Hasse, has strong implications for local–to-global prin-
ciples for rational points. In the case of the ground field Q when the degenerate
fibres are all defined over Q, we show that in their method Hypothesis (H) can
be replaced by the finite complexity case of the generalised Hardy-Littlewood
conjecture, which is a recent theorem of Greeen, Tao and Ziegler. We sketch
some of the applications of this observation.
Jeanine Van Order. Stable Galois averages of Rankin-Selberg L-values and
non-triviality of p-adic L-functions. I will present the notion of a stable
Galois average of central values of GL(2) Rankin-Selberg L-functions, and
then explain how to derive some consequences for the nonvanishing of certain
families of p-adic L-functions (with interesting applications).
Gabor Wiese (Université du Luxembourg). Symplectic Galois representations
and applications to the inverse Galois problem. We give an account of joint
work with Sara Arias-de-Reyna and Luis Dieulefait about compatible systems
of symplectic Galois representations and how they can possibly be applied
to the inverse Galois problem. In the beginning of the talk the overall strategy
will be outlined, starting from a previous joint work with Dieulefait on the two-
dimensional case. We will then explain the existence of a minimal global field
such that almost all the residual representations (of the compatible system) can
be defined projectively over its residue fields. Moreover, we shall report on a
very simple classification of the symplectic representations containing a non-
trivial transvection in their image. Finally, we shall combine the two points in
an application to the inverse Galois problem.
Trevor Wooley. Applications of efficient congruencing to rational points. We
provide an overview of the implications of the new “efficient congruencing”
method, first developed for Vinogradov’s mean value theorem, so far as appli-
cations to problems involving rational points are concerned. Some of these
conclusions are close to the convexity barriers in the circle method.
 1
Galois groups of local fields, Lie algebras and
ramification
Victor Abrashkin
Department of Mathematical Sciences, Durham University, Science Laboratories,
South Rd, Durham DH1 3LE, United Kingdom & Steklov Institute, Gubkina str. 8,
119991, Moscow, Russia
E-mail address: [email protected]

A BSTRACT. Suppose K is a local field with finite residue field of characteris-

tic p = 2 and K < p (M) is its maximal p-extension such that Gal(K < p (M)/K )
has period p M and nilpotent class < p. If char K = 0 we assume that K
contains a primitive p M -th root of unity. The paper contains an overview of
methods and results describing the structure of this Galois group together with
its filtration by ramification subgroups.

Introduction
Everywhere in the paper p is a prime number. For any profinite group and
s ∈ N, Cs ( ) denotes the closure of the subgroup of commutators of order s.
Let K be a complete discrete valuation field with a finite residue field
k F p N0 , N0 ∈ N. Let K sep be a separable closure of K and K =
Gal(K sep /K ). Denote by K ( p) the maximal p-extension of K in K sep . Then
K ( p) = Gal(K ( p)/K ) is a profinite p-group. As a matter of fact, the major
information about K comes from the knowledge of the structure of K ( p).
This structure is very well known and is related to the following three cases
(ζ p is a primitive p-th root of unity) [17]:

— char K = p;
— char K = 0, ζ p ∈
/ K;
— char K = 0, ζ p ∈ K .

Date: Nov 8, 2013

Key words and phrases: local field, Galois group, ramification filtration.

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

1
2 Victor Abrashkin

In all these cases the maximal abelian quotient of period p of K ( p) is

isomorphic to K ∗ /K ∗ p . Therefore, K ( p) has infinitely many generators in
the first case, has [K : Q p ] + 1 generators in the second case and [K : Q p ] + 2
generators in the third case. In the first two cases K ( p) is free and in the last
case it has one relation of a very special form, cf. [17, 23, 24].
The above results can’t be considered as completely satisfactory because
they do not essentially reflect the appearance of K ( p) as a Galois group of
an algebraic extension of a local field. In other words, let LF be the category
of couples (K , K sep ) where the morphisms are compatible continuous mor-
phisms of local fields and let PGr be the category of profinite groups. Then the
functor (K , K sep ) → K ( p) (as well as the functor (K , K sep ) → K ) is not
fully faithful.
The situation can be cardinally improved by taking into account a natural
additional structure on K ( p) and K given by the decreasing filtration of
ramification subgroups. The ramification filtration { K ( p)(v) }v0 of K ( p)
(v)
(as well as the appropriate filtration { K }v0 of K ) has many non-trivial
properties. For example, it is left-continuous at any v0 ∈ Q, v0 > 1, i.e.
 (v) = (v0 ) , but is not right-continuous, i.e. the closure
v<v K ( p) K ( p)
0
of v>v0 K ( p) is not equal to K ( p)(v0 ) . Another example [14, 15], for
(v)

any v1 , v2 < v0 , ( K ( p)(v1 ) , K ( p)(v2 ) ) ⊂ K ( p)(v0 ) and ( K ( p)(v1 ) ) p ⊂

(v0 ) and in some sense the groups (v0 ) have no “simple”
K ( p) K ( p)/ K ( p)
relations [5].
The significance of study of ramification filtration was very well understood
long ago, e.g. cf. Shafarevich’s Introduction to [17]. (The author also had inter-
esting discussions on this subject in the IAS with A.Weil, P.Deligne and F.Pop.)
As a matter of fact, the knowledge of ramification filtration is equivalent to the
knowledge of the original field K due to the following local analogue of the
Grothendieck conjecture.

Theorem 1.1. The functor (K , K sep ) → ( K ( p), { K ( p)(v) }v0 ) from LF to

the category of profinite p-groups with filtration is fully faithful.

This result was first proved in the mixed characteristic case in the context of
the whole Galois group K by Mochizuki [21] as a spectacular application of
p-adic Hodge-Tate theory. The case of arbitrary characteristic was established
by the author by a different method in [7] under the assumption p = 2. Note
that the characteristic p case was obtained via the explicit description of rami-
fication filtration modulo the subgroup of third commutators from [2]. Then the
mixed characteristic case was deduced from it via the Fontaine-Wintenberger
field-of-norms functor. In paper [11] we removed the restriction p = 2 and
reproved the statement in the context of the pro- p-group K ( p).
 Galois groups of local fields, Lie algebras and ramification 3

The study of ramification filtration in full generality seems not to be a realis-

tically stated problem: it is not clear how to specify subgroups of a given profi-
nite p-group. If we replace K ( p) by its maximal abelian quotient K ( p)ab
then the appropriate ramification filtration is very well known but reflects very
weak information about the original filtration of K ( p). This can be seen from
class field theory where we have the reciprocity map K ∗ −→ ab K and the ram-
ification subgroups appear as the images of the subgroups of principal units of
K ∗ . In particular, we can observe only integral breaks of our filtration.
As a matter of fact, the ramification subgroups can be described on the
abelian level without class field theory. The reason is that cyclic extensions
of K can be studied via much more elementary tools: we can use the Witt-
Artin-Schreier theory in the characteristic p case and the Kummer theory in
the mixed characteristic case. Trying to develop this approach to the case of
nilpotent Galois groups we developed in [1, 2] a nilpotent analogue of the
Witt-Artin-Schreier theory. This theory allows us to describe quite efficiently
p-extensions of fields of characteristic p with Galois p-groups of nilpotent
class < p. Such groups arise from Lie algebras due to the classical equiv-
alence of the categories of p-groups and Lie F p -algebras of nilpotent class
< p, [20]. In [1, 2, 4] we applied our theory to local fields K = k((t0 )), where
k F p N0 , and constructed explicitly the sets of generators of the appropriate
ramification subgroups. This result demonstrates the advantage of our tech-
niques: it is stated in terms of extensions of scalars of involved Lie algebras
but this operation does not exist in group theory.
A generalization of our approach to local fields K of mixed characteris-
tic was sketched earlier by the author in [6]. This approach allowed us to
pM
work with the groups K / K C p ( K ) under the assumption that a primitive
p M -th root of unity ζ p M ∈ K . At that time we obtained explicit constructions
of our theory only modulo subgroup of third commutators. Recently, we can
treat the general case. First results are related to the case M = 1 and can be
found in [11] (we discuss them also in Subsection 1.3.6 of this paper). The
case of arbitrary M as well as the case of higher dimensional local fields will
be considered in upcoming papers. In the case of local fields it would be very
interesting to relate our theory to constructions of “nilpotent class field theory”
from [19].
Note that the main constructions of the nilpotent Artin-Schreier theory do
not suggest that the basic field is local. They can be applied also to global fields
but it is not clear what sort of applications we can expect in this direction.
On the other hand, we can’t expect the existence of an easy “nilpotent
Kummer theory” for global fields. According to anabelian philosophy, for
global fields E, the quotient of E ( p) by the subgroup of third commutators
should already reflect all basic properties of the field E.
4 Victor Abrashkin

1.1 Nilpotent Artin-Schreier theory

In this section we discuss basic constructions of nilpotent Artin-Schreier the-
ory. The main reference for this theory is [2]. We shall call this version
contravariant and introduce also its covariant analogue, cf. Subsection 1.1.2
below. Everywhere M is a fixed natural number.

1.1.1 Lifts modulo p M , M ∈ N.

Suppose K is a field of characteristic p and K sep is a separable closure of
K . Let {xi }i ∈I be a p-basis for K . This means that the elements xi mod K ∗ p ,
i ∈ I , form a basis of the F p -module K ∗ /K ∗ p . Note that if E is any subfield
of K sep containing K then {xi }i∈I can be taken also as a p-basis for E.
Let W M be the functor of Witt vectors of length M. For a field K ⊂
E ⊂ K sep , define O M (E) as the subalgebra in W M (E) generated over
W M (σ M−1 E) by the Teichmuller representatives [xi ] ∈ W M (K ) ⊂ W M (E)
of all xi . Then O M (E) is a lift of E modulo p M : it is a flat W M (F p )-algebra
such that O M (E)/ pO M (E) = E. The system of lifts O M (E) essentially
depends on the original choice of a p-basis in K . If σ is the absolute Frobenius
(i.e. the morphism of p-th powers) then W M (σ ) induces a σ -linear morphism
on O M (E) and we usually denote it again by σ . Note that O M (E)|σ =id =
W M (F p ), if E is normal over K then the Galois group Gal(E/K ) acts on
O M (E) and the invariants of this action coincide with O M (K ).
A (continuous) automorphism ψ ∈ Aut(E) generally can’t be extended
to AutO M (E) if ψ changes the original p-basis. But the morphism σ M−1 ψ
admits “almost a lift” σ M−1 O M (E) −→ O M (E) given by the following
composition

W M (σ M−1 ψ)
σ M−1 O M (E) ⊂ W M (σ M−1 E) −−−−−−−−−−→ W M (σ M−1 E) ⊂ O M (E).

The existence of such lift allowed us to extend the modulo p methods from [1]
to the modulo p M situation in [2, 4].

1.1.2 Covariant and contravariant nilpotent Artin-Schreier

theories
Suppose L is a Lie algebra over W M (F p ). For s ∈ N, let Cs (L) be an
ideal of s-th commutators in L, e.g. C2 (L), resp., C 3 (L), is generated by
the comutators [l1 , l2 ], resp. [[l1 , l2 ].l3 ], where all li ∈ L. The algebra L has
nilpotent class < p if C p (L) = 0.
 Galois groups of local fields, Lie algebras and ramification 5

The basic ingredient of our theory is the equivalence of the categories of

p-groups of nilpotent class < p and the category of Lie Z p -algebras of the
same nilpotent class. This equivalence can be described on the level of objects
killed by p M as follows.
Suppose L is a Lie W M (F p )-algebra of nilpotent class < p. If A is envelop-
ing algebra for L and J is the augmentation ideal in A then there is a natural
embedding of L into A/J p (and L can be recovered as a submodule of
the module of primitive elements modulo J p in A, cf. [1] Section 1.1). The
Campbell-Hausdorff formula is the map L × L −→ L,
1
(l1 , l2 ) → l1 ◦ l2 = l1 + l2 + [l1 , l2 ] + . . .
2
such that in A mod J p we have e xp(l1 )
exp(l2 ) = e
xp(l1 ◦ l2 ), where e
xp(x) =
 i /i ! is the truncated exponential. The set L can be provided with
0i< p x
the composition law (l1 , l2 ) → l1 ◦ l2 which gives a group structure on L. We
denote this group by G(L). Clearly, this group has period p M . Then the cor-
respondence L → G(L) is the above mentioned equivalence of the categories
of p-groups of period p M and Lie W M (F p )-algebras.
Here and below we shall use the notation L K := L ⊗W M (F p ) O M (K ) and
L K sep = L ⊗W M (F p ) O M (K sep ). Then K and the absolute Frobenius σ act
through the second factor on L K sep , L K sep |σ =id = L and (L K sep ) K = L K .
The covariant nilpotent Artin-Schreier theory states that for any e ∈ G(L K ),
the set F(e) = { f ∈ G(L K sep ) | σ ( f ) = e ◦ f } is not empty and the map
g → (− f ) ◦ g( f ) is a group homomorphism π f (e) : K −→ G(L). The
correspondence e → π f (e) has the following properties:

a) if f  ∈ F(e) then f  = f ◦ c, where c ∈ G(L), and π f (e) and π f  (e) are

conjugated via c;
b) for any π ∈ Hom( K , G(L)), there are e ∈ G(L K ) and f ∈ F(e) such
that π f (e) = π;
c) for appropriate elements e, e ∈ G(L K ) and f, f  ∈ G(L K sep ), we have
π f (e) = π f  (e ) iff there is an x ∈ G(L K ) such that f  = x ◦ f and
(therefore) e = σ (x) ◦ e ◦ (−x); e and e are called R-equivalent via
x ∈ G(L K ).

According to above properties a)–c), the correspondence e → π f (e) estab-

lishes an identification of the set of all R-equivalent elements in G(L K ) and
the set of all conjugacy classes of Hom( K , G(L)).
The above theory can be proved in a similar way to its contravariant version
established in [2]. In the contravariant theory for any e ∈ G(L K ), the set { f ∈
G(L K sep ) | σ ( f ) = f ◦ e} is not empty, the correspondence g → g( f ) ◦ (− f )
6 Victor Abrashkin

establishes a group homomorphism from 0K to G(L K ), where 0K coincides

with K as a set but has the opposite group law (g1 g2 )0 = g2 g1 . (Equivalently,
if a ∈ K sep then (g1 g2 )a = g2 (g1 a).) We have also the properties similar
to above properties a)–c) but in c) there should be f  = f ◦ x and e =
x ◦ e ◦ (−σ x).
The both (covariant and contravariant) theories admit a pro-finite version
where L becomes a profinite W M (F p )-Lie algebra and the set Hom( K , G(L))
is the set of all continuous group morphisms.

1.1.3 Identification η0
Suppose K = k((t0 )) where t0 is a fixed uniformiser in K and k F p N0 with
N0 ∈ N. Then {t0 } is a p-basis for K, and we have the appropriate system of
lifts O M (E) modulo p M for all subfields K ⊂ E ⊂ Ksep . In addition, fix an
element α0 ∈ W (k) such that Tr(α0 ) = 1, where Tr is the trace map for the
field extension W (k) ⊗Z p Q p ⊃ Q p .
Let Z+ ( p) = {a ∈ N | (a, p) = 1} and Z0 ( p) = Z+ ( p) ∪ {0}.
For M ∈ N, let LM be a profinite free Lie Z/ p M -algebra with the (topolog-
ical) module of generators K∗ /K∗ p and L M = L
M
M /C p (LM ). From time to
time we drop the subscript M off to simplify the notation.
Let L = L M . Then Lk := L ⊗ W M (k) has the generators

{D0 } ∪ {Dan | a ∈ Z+ ( p), n ∈ Z/N0 }

due to the following identifications (where t = [t0 ] is the Teichmuller

representative of t0 ):

K∗ /K∗ p ⊗W M (F p ) W M (k) =
M

HomW M (F p ) (O M (K)/(σ − id)O M (K), W M (k)) =

HomW M (F p ) ((W M (F p )α0 ) ⊕a∈Z+ ( p) (W M (k)t −a ), W M (k)) =

W M (k)D0 × W M (k)Dan
a∈Z+ ( p)
n∈ZmodN0

Note that the first identification uses the Witt pairing, D0 appears from t0 ⊗ 1 ∈
K∗ /K∗ p ⊗ W M (k) and for all a ∈ Z+ ( p) and w ∈ W M (k), Dan (wt −a ) =
M

σ n w.
For any n ∈ Z/N0 , set D0n = t ⊗ (σ n α0 ) = (σ n α0 )D0 .

Let e0 = a∈Z0 ( p) t −a Da0 ∈ L K , choose f 0 ∈ F(e0 ) and set η0 = π f0 (e0 ).
Then η0 is a surjective homomorphism from K to G(L) and it induces a
pM
group isomorphism K/ K C p( K) G(L). Note that the construction of
 Galois groups of local fields, Lie algebras and ramification 7

η0 depends up to conjugacy only on the original choice of the uniformizer

t0 and the element α0 mod p M ∈ W M (k). On the level of maximal abelian
quotients of period p M , η0 induces the isomorphism of local class field theory
K∗ /K∗ p .
M
K ⊗Z p W M (F p )
ab

1.1.4 Why Campbell-Hausdorff?

In this subsection it will be explained that in our theory, we are, essentially,
forced to use the Campbell-Hausdorff composition law.
Assume for simplicity, that M = 1 and K = F p ((t0 )). Let K( p) be the
maximal p-extension of K and K ( p) = Gal(K( p)/K). For s ∈ N and
a1 , a2 , . . . , as , . . . ∈ Z0 ( p), consider the elements Ta1 ...as ∈ K( p) such that:

p −a
Ta1 − Ta1 = t0 1 ,
Ta1 a2 − Ta1 a2 = t0−a1 Ta2
p

.........
Ta1 ...as − Ta1 ...as = t0−a1 Ta2 ...as
p

.........

Then the system {Ta1 ...as | s  0, ai ∈ Z0 ( p)} is linearly independent over

K and if M = F p Ta1 ...as then K( p) = M ⊗F p K and K ( p) acts on
a1 ,...,as
s0
M via a natural embedding K ( p) → GLF p (M). This construction would
have given us an efficient approach to an explicit construction of the maxi-
mal p-extension K( p) if we could describe explicitly the image of K ( p) in
GLF p (M).
Analyze the situation at different levels s  1.

● 1st level. Here all equations are independent and we can introduce a min-
imal system of generators τa , a ∈ Z0 ( p), of K ( p) with their explicit
action via τa : Ta1 → Ta1 + δ(a, a1 ) at this level. (Here and below δ is
the Kronecker symbol.)
● 2nd level. Here the roots Ta1 a2 are not (algebraically) independent. For
example, the following identity
−a1 −a2 −a1 −a2 −(a1 +a2 )
(Ta1 Ta2 ) p = (Ta1 +t0 )(Ta2 +t0 ) = Ta1 Ta2 +t0 Ta2 +t0 Ta1 +t0

implies under the assumption (a1 + a2 , p) = 1 (and after a suitable choice

of involved roots of Artin-Schreier equations) that

Ta1 Ta2 = Ta1 a2 + Ta2 a1 + Ta1 +a2 .

8 Victor Abrashkin

The presence of the term Ta1 +a2 creates a problem: τa1 +a2 should act non-
trivially on either Ta1 a2 or Ta2 a1 but they both do not depend on the index
a1 + a2 . The situation can be resolved by a slight correction of involved
equations. Namely, let Ta1 a2 be such that
−(a1 +a2 )
Ta1 a2 − Ta1 a2 = t0−a1 Ta2 + η(a1 , a2 )t0
p

where the constants η(a1 , a2 ) ∈ k, a1 , a2 ∈ Z0 ( p), satisfy the relations

η(a1 , a2 ) + η(a2 , a1 ) = 1. (1.1)

With the above correction, the elements Ta1 a2 , a1 , a2 ∈ Z0 ( p), can be

chosen in such a way that we have the following:

— relations: Ta1 Ta2 = Ta1 a2 + Ta2 a1 ;

— Galois action: τa (Ta1 a2 ) = Ta1 a2 + Ta1 δ(a2 , a) + η(a1 , a2 )δ(a1 , a2 , a).

Relation (1.1) will look more natural if we introduce the constants on

the first level via η(a) = 1, a ∈ Z0 ( p). Then (1.1) can be rewritten as
η(a1 )η(a2 ) = η(a1 , a2 ) + η(a2 , a1 ). These relations can be satisfied only if
p = 2 and the simplest choice is η(a1 , a2 ) = 1/2 for all a1 , a2 .
The above picture can be generalized to higher levels as follows.

● s-th level, s < p. Here we have:

– the equations: Ta1 ...as = Ta1 ...as + η(a1 )t −a1 Ta2 ...as + . . .
p
−(a +···+as−1 ) −(a +···+as )
+η(a1 , . . . , as−1 )t0 1 Tas + η(a1 , . . . , as )t0 1

– the relations: Ta1 ...ak Tb1 ...bl = Tinsertions of a’s into b’s , where k + l < p;
– the Galois action: τa (Ta1 ...as ) = Ta1 ...as + Ta1 ...as−1 δ(a, as )η(as )+
· · ·+Ta1 δ(a, a2 , . . . , as )η(a2 , . . . , as )+δ(a, a1 , . . . , as )η(a1 , . . . , as )
– the constants: if k + l < p then

η(a1 , . . . , ak )η(b1 , . . . , bl ) = η(insertions of a’s into b’s)
with their simplest choice η(a1 , . . . , as ) = 1/s!

Remark. An insertion of the ordered collection a1 , . . . , ak into the ordered

collection b1 , . . . , bl is the ordered collection c1 , . . . , ck+l such that

— {1, . . . , k + l} = {i 1 , . . . , i k } { j1 , . . . , jl };
— i 1 < . . . < i k and j1 < . . . < jl ;
— a1 = ci1 ,. . . , ak = cik and b1 = c j1 ,. . . , bl = c jl .

The following formalism allows us to present the above information on all

levels 1  s < p in the following compact way.
 Galois groups of local fields, Lie algebras and ramification 9

Let A  be a pro-finite associative F p -algebra with the set of free generators

{Da | a ∈ Z0 ( p)}. Introduce the elements of the appropriate extensions of
scalars of A
−(a1 +···+as )
E =1+ η(a1 , . . . , as )t0 Da 1 . . . Da s
1s< p
ai ∈Z0 ( p)

= e
xp( t0−a Da ) ∈ AK ,
a∈Z0 ( p)

F = 1+ η(a1 , . . . , as )Ta1 ...as Da1 . . . Das ∈ AKsep

1s< p
ai ∈Z0 ( p)

Define the diagonal map as the morphism of F p -algebras

 : A mod deg p −→ A ⊗ A mod deg p

such that for any a ∈ Z0 ( p), Da → Da ⊗ 1 + 1 ⊗ Da . Then we have the

following properties:

— (E) ≡ E ⊗ E mod deg p; (F) ≡ F ⊗ Fmod deg p ;

— σ (F) ≡ EF mod deg p; τa (F ) ≡ F e
xp(Da ) mod deg p .

Now we can verify the existence of f ∈ LKsep such that F = e xp( f )

 −a
modulo deg p, and recover the basic relations σ ( f ) = ( a t0 Da ) ◦ f and
τa ( f ) = f ◦ Da , a ∈ Z0 ( p), of our nilpotent Artin-Schreier theory.

1.2 Ramification filtration in L = L M+1

In this section we describe and illustrate the main trick used in papers [1, 2,
4]. This trick allowed us to find explicit generators of ramification subgroups
under the identification η0 from Subsection 1.1.3. Remind that we work over
K = k((t0 )), where k F p N0 , N0 ∈ N.

1.2.1 Auxiliary field K = K(r ∗ , N ), [1, 2, 4]

The field K is a totally ramified extension of K in Ksep . It depends on two
parameters: r ∗ ∈ Q such that r ∗ > 0 and v p (r ∗ ) = 0, and N ∈ N such that
if q = p N then b∗ := r ∗ (q − 1) ∈ N. Note that for a given r ∗ , there are
infinitely many ways to choose N , in particular, we can always assume that N
is sufficiently large.
10 Victor Abrashkin

By definition, [K : K] = q and the Herbrand function ϕK /K has only

one edge point (r ∗ , r ∗ ). It can be proved that K = k((t0 )), where t0 =
q ∗
t0 E(−1, t0 b ). Here for w ∈ W (k),
n
E(w, X ) = exp(w X + σ (w)X p / p + · · · + σ n (w)X p / p n + . . . ) ∈ Z p [[X ]]

is the Shafarevich version of the Artin-Hasse exponential.

Note that if r ∗ ∈
/ N, K /K is neither Galois nor a p-extension.

1.2.2 The criterion

Consider the following lifts modulo p M+1 with respect to the p-basis {t0 } of K

O M+1 (K) = W M+1 (k)((t)) = W M+1 (σ M K)[t]

O M+1 (Ksep ) = W M+1 (σ M Ksep )[t] ⊂ W M+1 (Ksep ).

Remember that t = [t0 ] ∈ O M+1 (K) is the Teichmüller representative of t0 in

W M+1 (K).
For K = K(r ∗ , N ) and its uniformiser t0 from Subsection 1.2.1 consider

the appropriate lifts O M+1 (K ) and O M+1
  ). If t  = [t  ] then t and t  can
(Ksep 0

be related one-to-another in W M+1 (K ) via
∗ M−1 b∗ M b∗
t p = t p exp(− p M t  b − · · · − pt  p )E(−1, t  p
M Mq
).

This implies the following relations between the lifts for K and K

σ M O M+1 (K) ⊂ W M+1 (σ M K) ⊂ O M+1 (K )
 
σ M O M+1 (Ksep ) ⊂ W M+1 (σ M Ksep ) ⊂ O M+1 (Ksep )
 −a D
As earlier, take e0 = a∈Z0 ( p) t a0 ∈ LK , f 0 ∈ LKsep such that
σ f 0 = e0 ◦ f 0 and consider π f0 (e0 ) : K −→ G(L) . Similarly, let

e0 = a∈Z0 ( p) t −a Da,−N , choose f 0 ∈ LKsep such that σ f 0 = e0 ◦ f 0 and
consider π f  (e0 ) : K −→ G(L).
0
For Y ∈ LKsep and an ideal I in L, define the field of definition of
Y mod IKsep over K as K(Y mod IKsep ) := Ksep H , where H = {g ∈

K | g(Y ) ≡ Y mod IKsep }.

For any finite field extension E/K in Ksep define its biggest ramification
(v)
number v(E/K) = max{v | K acts non-trivially on E}.
(v0 )
For v0 ∈ Q>0 , let the ideal L(v0 ) of L be such that G(L(v0 ) ) = η0 ( K ).
 
Let f M = σ f 0 and f M = σ f 0 . Our method from [1, 2, 4] is based on the
M M

following criterion.
 Galois groups of local fields, Lie algebras and ramification 11

Proposition 1.2. Let X ∈ LKsep be such that f M = X ◦ σ N ( f M  ). Suppose

∗ ∗ ∗ (v )
v0 , r ∈ Q>0 , v p (r ) = 0 and r < v0 . Then L 0 is the minimal ideal in the
family of all ideals I of L such that
v(K (X mod IKsep )/K )  v0 q − b∗ .

The proof is quite formal and is based on the following properties of upper
ramification numbers. If v = v(K( f mod IKsep )/K) then:
a) v(K ( f  mod IKsep )/K ) = v;
b) v(K ( f  mod IKsep )/K) = ϕK /K (v);
c) if v > r ∗ then ϕK /K (v) < v.

1.2.3 Illustration of the criterion

The criterion from Proposition 1.2 was applied in [1, 2, 4] to describe the
(v)
structure of η0 ( K ) by induction by proceeding from the situation modulo
p M to the situation modulo p M+1 and from the situation modulo Cs (L) to the
situation modulo Cs+1 (L), where 2  s < p.
Typically, for an ideal I ⊂ L, we used the knowledge of the structure
of L(v0 ) modI to prove (after choosing r ∗ sufficiently close to v0 ) that X ∈
(v )
LK mod (LK0 + IK ). Then we could apply our Criterion to L(v0 ) modJ for
an appropriate (slightly smaller than I ) ideal J because X mod JKsep satisfied
over LK just an Artin-Schreier equation of degree p. Notice that f 0 mod JKsep
and f 0 mod JKsep satisfy very complicated relations over L K  . We give below
two examples to illustrate how our method works in more explicit but similar
situations.

1.2.3.1 First example

Suppose M  0 and F ∈ O M+1 (Ksep ) is such that F − σ N0 F = t −a , a ∈
Z0 ( p).
If M = 0 then F is a root of the Artin-Schreier equation F − F q0 = t −a
with q0 = p N0 and directly from the definition of ramification subgroups it
follows that v(K(F)/K) = a. The case of arbitrary M corresponds to the Witt
theory. Here the left-hand side F − σ N0 F is already a Witt vector of length
M + 1, and careful calculations with components of Witt vectors give that
v(K(F)/K) = p M a. Our criterion allows us to obtain this result in a much
easier way.
Take the field K = K(r ∗ , N ) where r ∗ and N (recall that q = p N ) are
such that
ap M > r ∗ > ap M−1 q/(q − 1). (1.2)
12 Victor Abrashkin

  ) and let F  ∈ O 
(Ksep 
Consider the appropriate lift O M+1 M+1 (Ksep ) be such
 
that F − σ 0 F = t .
N  −a

Set FM = σ M F and FM  = σ M F .

Clearly, K(FM ) = K(F) and K (FM  ) = K (F  ).

According to restrictions (1.2) we have

∗ M−1 b∗ M b∗
t −ap = t  −aq p exp(ap M t  b + · · · + apt  p )E(a, t  p
M M
)
 −aq p M M  −ap M q+b∗
=t + ap t + f0,

where f 0 ∈ t  W M (k)[[t  ]] ⊂ O M+1

 (K ). Therefore,

FM = σ N FM + pM X + σ i N0 ( f 0 ),
i 0

∗
where X − σ N0 X = at  −ap q+b .
M

In this situation an analogue of our criterion states that

v(K(FM )/K) = ap M ⇔ v(K (X )/K ) = ap M q − b∗

But the right-hand side of this assertion corresponds to the case M = 0 and
was explained in the beginning of this section.

1.2.3.2 Second example

Consider the following modulo p situation, i.e. the situation where M = 0.
(The appropriate case of arbitrary M can be considered similarly.)
Let F, G ∈ Ksep be such that

F − σ N0 F = t0−a , G − σ N0 G = t0−b σ n 0 (F) (1.3)

with a, b ∈ Z0 ( p) and 0  n 0 < N0 . Let

A = max a + bp−n 0 , ap −N0 +n 0 + b

Prove that v(K(F, G)/K) = A if either a = b or 2n 0 = N0 .

Take K = K (r ∗ , N ), where r ∗ and q = p N satisfy the following
restrictions
qA qA q
< r∗ < , r∗ > max{a, ap −N0 + b}
2(q − 1) q −1 q −1
(We can take r ∗ ∈ (A/2, A) such that r ∗ > max{a, b} and then choose suf-
ficiently large N to satisfy these conditions.) Consider F  , G  ∈ Ksep
 such
that
F  − σ N0 F  = t0 −a , G  − σ N0 G  = t0 −b σ n 0 (F  )
 Galois groups of local fields, Lie algebras and ramification 13

Notice that
 −aq  −aq+b∗
t0−a = t0 + at0 + o1
 −aq+2b∗
where o1 ∈ t0 k[[t0 ]]. Therefore,

F = F  q + TF + σ i N 0 o1 , (1.4)
i0

 −aq+b∗
where TF − σ N0 TF = at0 . We can choose F  in such a way that (use
 −aq+b∗  p N0 (−aq+b∗ )
that −aq + b∗ > 0) TF = at0 + o2 , where o2 ∈ t0 k[[t0 ]].
As earlier,
 −bq  −bq+b∗
t0−b = t0 + bt0 + o3 (1.5)
 −bq+2b∗
with o3 ∈ t0 k[[t0 ]]. Then (1.4) and (1.5) imply that
 −bq  −bq+ p n0 (−aq+b∗ )
t0−b σ n 0 F = (t0 σ n 0 F  )q + at0 +
 −bq+b∗
bt0(σ n 0 F  )q + o3 (σ n 0 F  )q + o4

where o4 ∈ t0 k[[t0 ]]. Therefore, G = G  q + TG + i 0 σ i N0 o4 , where
n 0 (−aq+b ∗ )
TG − σ N0 TG = at  −bq+ p
∗
+bt  −bq+b (σ n 0 F  )q + o3 (σ n 0 F  )q

An appropriate analogue of our criterion gives that

v(K(F, G)/K) = A ⇔ v(K (TF , TG )/K ) = q A − b∗

First of all, TF ∈ k[[t0 ]] and, therefore, we should prove that

v(K (TG )/K ) = q A − b∗ . Notice that
n 0 −N0 n 0 −N0  −aq p n 0 −N0
F qp = (σ N0 F  )q p = F qp
n0
− t0 .

Let OK (F  ) and mK (F  ) be the valuation ring and, resp., the maximal ideal for
K (F  ). Clearly, t0 a σ N0 F  ∈ OK (F  ) and, therefore,
 −bq+b∗ n 0 −N0
t0 F qp ∈ mK (F  )

(use that b∗ > q(ap n 0 −2N0 + b)) and o3 (σ n 0 F  )q ∈ mK (F  ) (use that 2b∗ −
(ap n 0 −N0 + b) > 0).
This implies that K (TG ) ⊂ K (TG , F  ) = K (T, F  ), where
 
−n 0 ∗  −(ap n0 −N0 +b)q+b∗
T − σ N0 T = aσ n 0 t  −(a+bp )q+b − bt0 ,

and v(K (TG )/K ) = v(K (T, F  )/K ).

14 Victor Abrashkin

If either a = b or 2n 0 = N0 the right-hand side of this equation is not trivial

and (use that v(K (F  )/K ) = a < A) v(K (T, F  )/K ) = v(K (T )/K ) =
q A − b∗ .

pM
1.2.4 Ramification subgroups modulo K C p( K)
(v)
As earlier, let L = L M and for v  0, let L(v) = η0 ( K ) ⊂ L. The ideal L(v)
was described in [4] as follows.
For γ  0 and N ∈ Z, introduce the elements Fγ0,−N ∈ Lk via
Fγ0,−N = a1 p n 1 η(n 1 , . . . , n s )[. . . [Da1 n̄ 1 , Da2 n̄ 2 ], . . . , Das n̄ s ]
1s< p
ai ,n i
Here:
— all ai ∈ Z0 ( p), n i ∈ Z, n 1  0, n 1  n 2  · · ·  n s  −N , n̄ s =
n s mod N0 ;
— a 1 p n 1 + a2 p n 2 + · · · + as p n s = γ ;
— if n 1 = · · · = n s1 > · · · > n sr−1 +1 = · · · = n sr then η(n 1 , . . . , n s ) =
(s1 ! . . . (sr − sr−1 )!)−1 .
Let L(v) N be the minimal ideal of L such that L(v) N ⊗ W M (k) contains all
Fγ0,−N with γ  v. Then there is an N M∗ (v) ∈ N and an ideal L(v) of L such
∗
that for all N  N M (v), L(v) N = L(v).

Theorem 1.3. For any v  0, L(v) = L(v).

This statement was obtained in the contravariant setting in [4] and uses
the elements Fγ ,−N given by the same formula (as for Fγ0,−N ) but with the
factor (−1)s−1 . Indeed, the contravariant version of Theorem 1.3 appears by
replacing the Lie bracket [l1 , l2 ] in L by the bracket [l1 , l2 ]0 = [l2 , l1 ]. There-
fore, [. . . [D1 , D2 ], . . . , Ds ] should be replaced by [Ds , . . . , [D2 , D1 ] . . . ] =
(−1)s−1 [. . . [D1 , D2 ], . . . , Ds ].

Remark. For the ideal L(v) mod C 2 (L) we have the generators coming from
Fγ0,−N taken modulo the ideal of second commutators. Such generators are
non-zero only if γ is integral. Therefore, L(v) ⊗ W (k) is generated on the
abelian level by the images of p n Dam , where m ∈ Z/N0 and p n a  v.

The proof of Theorem 1.3 is quite technical and it would be nice to put it into
a more substantial context. This could be done on the basis of the following
interpretation.
 Galois groups of local fields, Lie algebras and ramification 15

Assume for simplicity that M = 0. Choose r ∗ , N as in Subsection 1.2.1. We

can assume that N ≡ 0 mod N0 .
q
If we replace t0 by t0 then the identification η0 will be not changed. Indeed,
  −aq
e0 = a∈Z0 ( p) t0−a Da0 is replaced by ι(e0 ) = a∈Z0 ( p) t0 Da0 = (σ c) ◦
e0 ◦ (−c), where c = (σ N −1 e0 ) ◦ · · · ◦ (σ e0 ) ◦ e0 .
When proving above Theorem 1.3 in [4] we actually established that L(v0 )
appears as the minimal ideal I of L such that the replacement (deformation)
q ∗
d(1) : t0 → t0 E(−1, t0b ) does not affect the identification η0 mod I. The
q ∗
same holds also for the one-parameter deformation D(u) : t0 → t0 E(−u, t0b )
with parameter u. In terms of the nilpotent Artin-Schreier theory this means
the existence of c = c(u) ∈ LK[u] such that

D(u)(e0 ) ≡ (σ c) ◦ e0 ◦ (−c) mod IK[u]

(we assume that σ (u) = u). This condition is satisfied on the linear level
(i.e. for the coefficients of u) if and only if the above elements Fγ ,−N , where
γ  v and N  N M ∗ (v), belong to I . As a matter of fact, the main difficulty
k
we resolved in [1, 4] was that on the level of higher powers u i , i > 1, we do
not obtain new conditions. We do expect to obtain this fact in an easier way by
more substantial use of ideas of deformation theory.

1.3 The mixed characteristic case

In this section we sketch main ideas which allowed us to apply the above
characteristic p results to the mixed characteristic case.
Suppose K is a finite etension of Q p with the residue field k. We fix a choice
of uniformising element π0 in K and assume that K contains a primitive p M -th
root of unity ζ M .

1.3.1 The field of norms functor [26]

 be a composite of the field extensions K n = K (πn ), where n  0,
Let K
p
K 0 = K , and πn+1 = πn . The field of norms functor X provides us with:

1) a complete discrete valuation field K = X ( K ) of characteristic p. The

residue field of K can be canonically identified with k, and K has a
fixed uniformizer t0 : by definition, K∗ = lim K n∗ , where the connecting
←−
morphisms are induced by the norm maps, and t0 = lim πn ;
←−
, then X (E) is separable over K, and the
2) if E is an algebraic extension of K
correspondence E → X (E) gives equivalence of the category of algebraic
16 Victor Abrashkin

 and the category of separable extensions of K. In particular,

extensions of K
X gives the identification of K = Gal(Ksep /K) with K ⊂ K ;
3) the above identification K = K is compatible with the ramification fil-
trations in K and K ; this means that if ϕ K/K = limn→∞ ϕ K n /K then for
(x) (y)
any x  0, K = K ∩ 
K with y = ϕ K/K (x).

1.3.2 Three questions

Let
pM pM
Cp( K)
K < p (M) = K̄ K C p( K ) , K< p (M) = Ksep
K
.
) ⊃ K, and there is a subgroup H M of
Then K< p (M) ⊃ X (K < p (M) K
K (M) := Gal(K< p (M)/K), such that
)/K) =
Gal(X (K < p (M) K K (M)/H M .

Under the identification η0 from Subsection 1.1.3 we have H M G(J M ),

where J M is an ideal of the Lie algebra L M .
Question A. What is the ideal J M ?
Remember that the Lie algebra L M ⊗ W (k) has a system of generators
{Dan | a ∈ Z+ ( p), n ∈ Z/N0 } ∪ {D0 }. So, more precisely,
What are explicit generators of the ideal J M ?
It is easy to see that K < p (M) ∩ K  := K M = K (π M ). Therefore, for
K (M) := Gal(K <p (M)/K ), we have the following exact sequence of
p-groups

−→ τ0 Z/ p −→ 1,
M
1 −→ K (M)/H M −→ K (M)

where τ0 ∈ Gal(K M /K ) is defined by the relation τ0 (π M ) = ζ M π M . Using the

equivalence of the category of Lie Z/ p M -algebras of nilpotent class < p and
of the category of p-groups of the same nilpotent class, cf. Subsection 1.1.2,
we can rewrite the above exact sequence of p-groups as the following exact
sequence of Lie Z/ p M -algebras
0 −→ L M /J M −→ L M −→ (Z/ p M )τ0 −→ 0
Here L M is the Lie Z/ p M -algebra such that G(L M ) = K (M). This sequence
of Lie algebras splits in the category of Z/ p M -modules and, therefore, can be
given by a class of differentiations adτ̂0 of L M , where τ̂0 is a lift of τ0 to an
automorphism of L M .
Question B. What are the differentiations adτ̂0 ?
More precisely,
Find the elements ad(τ̂0 )(Da0 ), a ∈ Z0 ( p).
 Galois groups of local fields, Lie algebras and ramification 17

As we have mentioned in Subsection 1.3.1 the ramification filtrations in K

and K are compatible.
The ramification filtration of Gal(K M /K ) = Z/ p M τ0 has a very simple
structure. Let e K be the absolute ramification index of K and for s ∈ Z, s  0,
vs = e K p/( p − 1) + se K . Then

if 0  v  v0 , then (Z/ p M τ0 )(v) is generated by τ0 ;

if vs < v  vs+1 , 0 ≤ s < M, then (Z/ p M τ0 )(v) is generated by p s τ0 ;
if v > v M , then (Z/ p M τ0 )(v) = 0.

Therefore, we shall obtain a description of the ramification filtration

(v) of
K (M) K (M) by answering the following question.
Question C. How to construct ”good“ lifts p s τ ∈ L (vs ) , 0 ≤ s ≤ M?
0 M
Below we announce partial results related to the above questions.

1.3.3 The ideal J M

Consider the decreasing central filtration by the commutator subgroups
{Cs ( K (M))}s≥2 . This filtration corresponds to the following decreasing
central filtration of ideals of L M

J1 := L M ⊃ J2 ⊃ · · · ⊃ J p = J M

We can treat L M as a free pro-finite object in the category of Lie Z/ p M -

algebras of nilpotent class < p. Its module of generators is K∗ /K∗ p . In this
M

subsection we announce an explicit description of the filtration {Js }2s p .

Particularly, in the case s = p we obtain an answer to the above question A.
Let U be the submodule of K∗ /K∗ p generated by the images of prin-
M

cipal units. U can be identified with a submodule in the power series ring
W M (k0 )[[ t ]] via the correspondences

E(w, t0a ) mod K∗ p

M
→ wt a mod p M W (k)[[ t ]],

where w ∈ W (k), a ∈ Z+ ( p) = {a ∈ N | (a, p) = 1} and E(w, X ) is the

Shafarevich function from Subsection 1.2.1.
0 =
Let H0 ∈ W (k)[[ t ]] be a power series such that ζ M = H0 (π0 ). Let H
H0 mod p M ∈ W M (k)[[ t ]]. Then there is a unique S ∈ W M (k)[[ t ]], such
M
 p = E(1, S). Note that the differential d S = 0, in particular, S ∈
that H0
W M (k0 )[[ t p ]], and therefore, SU ⊂ U under the above identification of U
with the submodule ⊕ W M (k) t a in W M (k)[[ t ]].
a∈Z+ ( p)
18 Victor Abrashkin

For s ≥ 1 define a decreasing filtration of K∗ /K∗ p as follows:

M

 
M (1)
 
M (s)
K∗ /K∗ p = K∗ /K∗ p and K∗ /K∗ p
M
= S s−1 U, if s ≥ 2.
This filtration determines a decreasing filtration of ideals {L M (s)}s1 in L M
which can be characterized as the minimal central filtration of L M such that
for all s, (K∗ /K∗ p )(s) ⊂ L M (s).
M

Theorem 1.4. For 1  s  p, Js = L M (s).

M
Remark. a) The element H  p − 1 appears as the denominator in the
0
Brückner-Vostokov explicit formula for the Hilbert symbol and can be replaced
in that formula by S, cf. [3]; this element S can be considered naturally as an
element of Fontaine’s crystalline ring of p-adic periods and coincides with the
p-adic period of the multiplicative p-divisible group.
b) A first non-abelian case of the above theorem corresponds to s = 2 and
is equivalent to the formula

SU = Ker( K (M)
ab
−→ K (M) ).
ab

By Laubie’s theorem [18] the functor X is compatible with the reciprocity

maps of class field theories of K and K cf. also [12]. Therefore,

SU = Ker(N : K∗ /K∗ p −→ K ∗ /K ∗ p ),
M M
(1.6)

where N is induced by the projection K∗ = lim K n∗ −→ K 0∗ = K ∗ . The

← −
right-hand side of (1.6) can be interpreted as the set of all (wa )a∈Z+ ( p) ∈
+ 
W (k)Z ( p) mod p M such that a∈Z+ ( p) E(wa , π0a ) ∈ K ∗ p . So, formula (1.6)
M

can be deduced from the Brückner-Vostokov explicit reciprocity law.

1.3.4 Differentiation adτ̂0 ∈ Diff(L M )

It can be proved that there are only finitely many different ideals L(v)
M mod J M .
Therefore, we can fix sufficiently large natural number N1 (which depends
only on N0 , e K and M), set Fγ0 := Fγ0,−N1 and use these elements to describe
(v)
the ramification filtration {L M modJ M }v≥0 .
The elements Fγ0 , γ > 0, can be given modulo the ideal of third
commutators C3 (L M k ) as follows :
if γ = apl ∈ N, where a ∈ Z+ ( p) and l ∈ Z0 , then

Fγ0 = apl Dal¯ + η(n)a1 p s [Da1 s̄ , Da2 s̄−n̄ ];

s,n,a1 ,a2
 Galois groups of local fields, Lie algebras and ramification 19

if γ ∈
/ N, then
F(γ ) = η(n)a1 ps [Da1 ,s̄ , Da2 ,s̄−n̄ ].
s,n,a1 ,a2
In the above sums 0  s < M, 0  n < N1 , a1 , a2 ∈ Z0 ( p), p s (a1 +
a2 p −n ) = γ , η(n) = 1 if n = 0 and η(0) = 1/2.
Let S ∈ W M (k)[[ t ]] be the element introduced in Subsection 1.3.3.
Remember that S = σ S  , where S  ∈ W M (k0 )[[ t ]]. For l  1, let αl ∈ W M (k)
be such that
S − pS  = αl t l .
l1
Note that
a) if l < e K p/( p − 1) then αl = 0;
b) for any l  1, we have lαl = 0.

Theorem 1.5. There is a lift τ̂0 ∈ L M of τ0 such that

adτ̂0 (D0 ) = σ n (αl Fl0 )modC3 (L M )
l1
0n<N0

and, for a ∈ Z+ ( p),

adτ̂0 (Da0 ) = σ − f (αl Fl+ap
0
f )modC 3 (L Mk ).
f ∈Z
l1

1.3.5 Good lifts p

sτ , 0 ≤ s < M
0

If s = 0, let τ̂0 ∈ L M be a lift from Theorem 1.5. Define the lifts p s τ by

0
induction on s as follows
s τ = p( p
1
p 0
s−1 τ ) +
0 αl a1 [Da1 n , Da2 n ].
2 +
l1 a1 ,a2 ∈Z ( p)
0n<N0 ps e K
a1 +a2 = p−1

Theorem 1.6. A lift τ̂0 from Theorem 1.5 can be chosen in such a way
that all ps τ , 0 ≤ s < M, are “good” modulo C (L
0 3 Mk ), i.e. p
sτ ∈
0
ϕ(vs )
L M mod C3 (L Mk ), where vs = e K p /( p−1) and ϕ is the Herbrand function
s

of the extension K < p (M)/K .

Theorem 1.6 gives a complete description of the ramification filtration of

K (M) modulo C 3 ( K (M)). This result can be compared with the description
of the filtration K (1)(v) ∩ C2 ( K (1)) modulo C3 ( K (1)) in [27].
20 Victor Abrashkin

1.3.6 The modulo p case.

In this subsection we give an overview of the results from [11] related to the
modulo p aspect of problems discussed in this paper.
Let M = 1 and c0 := e K p/( p − 1)(= v1 ). We can simply drop off M from
all above notation instead of substituting M = 1.
The ideals L(s), 1  s  p, can be described now quite explicitly
as follows. Define a weight filtration on L by setting for a ∈ Z0 ( p) and
n ∈ Z/N0 , wt(Dan ) = s ∈ N if (s − 1)c0  a < sc0 . Then for all s,
L(s) = {l ∈ L | wt(l)  s}, cf. Section 17.3 of [11].
The field-of-norms K admits a standard embedding into R0 = Frac R, where
R is Fontaine’s ring [26]. This also identifies R0 with the completion of Ksep
and, therefore, we have a natural embedding of K< p into R0 . In particular, f 0
can be considered as an element of LR0 .
Choose a continuous automorphism h 0 of K such that

h 0 (t) ≡ τ0 (t) mod t c0 ( p−1) m R ,

where m R is the maximal ideal in R. The formalism of nilpotent Artin-Schreier

theory allows us to describe efficiently the lifts ĥ 0 ∈ AutK< p of h 0 , [2]. This
can be done by specifying the image ĥ 0 ( f 0 ) in the form

ĥ 0 ( f 0 ) = c(ĥ 0 ) ◦ (Adĥ 0 ⊗ 1) f 0

where c(ĥ 0 ) ∈ LK and Adĥ 0 is the conjugation of G(L) via ĥ 0 .

It can be proved then that the lifts τ̂0 ∈ AutK < p satisfy

τ̂0 ( f 0 ) ≡ ĥ 0 ( f 0 ) mod t c0 ( p−1) M R0 ,


where M R0 = 1s< p t −c0 s L(s)m R + L( p) R0 , and are uniquely determined
by this conditions. Therefore, the lifts τ̂0 can be uniquely described via the
morphisms Adτ̂0 and the elements c(ĥ 0 ) mod t c0 ( p−1) MK , where MK =
 −c0 s L(s)
1s< p t mK + L( p)K .
The above elements c(ĥ 0 ) satisfy complicated relations but the whole
situation can be linearized as follows.
In [11] we proved that the action of the group ĥ 0 Z/ p on f 0 comes
from the action of the additive group scheme Ga,F p = SpecF p [u] on
MK< p /t c0 ( p−1) MK< p , where MK< p is defined similarly to M R0 . This
implies that if c[u] = c0 + c1 u + · · · + c p−1 u p−1 ∈ LK [u] is the polyno-
mial with coefficients in LK such that for 0  k < p, c[u]|u=k = c(ĥ k0 ) then
its residue modulo t c0 ( p−1) MK is well defined and can be uniquely recovered
from its first coefficient c1 mod t c0 ( p−1) MK .
Now we can state the main results from [11].
 Galois groups of local fields, Lie algebras and ramification 21

1.3.6.1
There is a bijection
τ̂0 → (c1 , {adτ̂0 (Da0 ) | a ∈ Z0 ( p)})
of the set of lifts τ̂0 and solutions (c1 , {Va | a ∈ Z0 ( p)}), where c1 ∈ LK ,
V0 ∈ α0 L, Va ∈ Lk with a ∈ Z+ ( p), of the following equation
σ c1 − c1 + t −a Va =
a∈Z0 ( p)
1 l−(a1 +···+ak )
t αl [. . . [a1 Da1 0 , Da2 0 ], . . . , Dak 0 ]
k!
k,l1
1 −(a1 +···+ak )
− t [. . . [Va1 , Da2 0 ], . . . , Dak 0 ]
k!
k2
1 −(a2 +···+ak )
− t [. . . [σ c1 , Da2 0 ], . . . , Dak 0 ]
k!
k1

(Remember that αl ∈ k were defined in Subsection 1.3.4 and they equal 0 if

l = c0 , c0 + p, c0 + 2 p, . . . .)

1.3.6.2
A lift τ̂0 is “good”, i.e. τ̂0 ∈ L (c0 ) iff all Va ∈ L(c
k
0)
and
(c )
c1 ≡ − σ i (αl Fγ0,−i t −γ +l ) mod (LK0 + t c0 ( p−1) MK )
γ >0 0i<N ∗
l1

1.3.6.3
Let the operators F0 and G 0 on Lk be such that for any l ∈ Lk ,
1 1
F0 (l) = [. . . [l, D00 ], . . . , D00 ], G 0 (l) = [. . . [l, D00 ], . . . , D00 ]
k!    k!   
k1 k−1 times k0 k times

Then the correspondence

(c10 , V0 ) → (c10 − σ i (αl Fl,−i
0
), V0 )
0i<N ∗
l1

establishes a bijection between the set of all good lifts τ̂0 and the set of all
(c )
(x, y) such that x ∈ Lk 0 , y ∈ α0 L(c0 ) , and
∗
 
(G 0 σ − id)(x) + F0 (y) = σ N αl Fl,−N
0
∗ (1.7)
l1
22 Victor Abrashkin

1.3.6.4
For any above solution (x, y) of (1.7) we have

(c0 )
y ≡ α0 Trk/F p (αl Fl,−N
0
∗ )modα0 [D0 , L ]
l1

This implies the existence of a good lift τ̂0 such that the (only) relation in the
Lie algebra L k (recall that G(L) = Gal(K < p /K )) appears in the form

adτ̂0 (D0 ) = Trk/F p (αl Fl,−N

0
∗) .
l1

References
[1] V.A.A BRASHKIN, Ramification filtration of the Galois group of a local field,
Proceedings of the St. Petersburg Mathematical Society III, Amer. Math. Soc.
Transl. Ser. 2, (1995) 166, Amer. Math. Soc., Providence, RI
[2] V.A. A BRASHKIN , Ramification filtration of the Galois group of a local field.
II, Proceedings of Steklov Math. Inst. (1995) 208
[3] V.A BRASHKIN, The field of norms functor and the Brückner-Vostokov formula,
Math. Ann., (1997) 308, no.1, 5-19
[4] V.A BRASHKIN, Ramification filtration of the Galois group of a local field.
III, Izvestiya RAN: Ser. Mat., 62, no.5 (1998), 3-48; English transl. Izvestiya:
Mathematics 62, no.5, 857-900
[5] V.A. A BRASHKIN, A group-theoretical property of the ramification filtration,
Izvestiya RAN: Ser. Mat., (1998) 62, no.6 3-26; English transl. Izvestiya:
Mathematics (1998), 62, no.6 1073-1094
[6] V.A BRASHKIN, Report on the ramification filtration of the Galois group of
a local field, Proceedings of the Research Conference on “Number Theory
and Arithmetical Geometry: Arithmetical applications of Modular Forms” (San
Feliu de Guixols, Spain, 24-29 October, 1997) Inst. fur Exp. Math. Universitat
Essen, 1998, 47-53
[7] V. A BRASHKIN , On a local analogue of the Grothendieck Conjecture, Int. J.
Math. (2000) 11, no.1, 3-43
[8] V. A BRASHKIN, Ramification theory for higher dimensional fields, Contemp.
Math. (2002) 300, 1-16
[9] V. A BRASHKIN, Characteristic p case of the Grothendieck conjecture for
2-dimensional local fields, Proceedings of Steklov Institute (2003) 241, 1-35
[10] V. A BRASHKIN , Characteristic p analogue of modules with finite crystalline
height, Pure Appl. Math. Q., (2009) 5, 469-494
[11] V.A BRASHKIN, Automorphisms of local fields of period p and nilpotent class
< p (in preparation)
[12] V.A BRASHKIN , R.J ENNI, The field-of-norms functor and the Hilbert symbol
for higher local fields, J.Théor. Nombres Bordeaux, (2012) 24, no.1, 1-39
 Galois groups of local fields, Lie algebras and ramification 23

[13] P.D ELIGNE Les corps locaux de caractéristique p, limites de corps locaux de
caractéristique 0, Representations of reductive groups over a local field, Travaux
en cours, Hermann, Paris, 1973, 119-157
[14] N.L.G ORDEEV, Ramification groups of infinite p-extensions of a local field,
Soviet Math. (1982) 20, no.6 2566-2595
[15] N.L.G ORDEEV, Infinity of the number of relations in the Galois group of the
maximal p-extension of a local field with bounded ramification, Izvestia AN
SSSR, Ser. Matem. (1981) 45, 592-607
[16] U.JANNSEN , K.W INGBERG, Die Struktur der absoluten Galoisgruppe
p-adischer Zahlkörper, Invent. math. (1982) 70, 71-98
[17] H.KOCH, Galoissche Theorie der p-Erweiterungen, Mathematische Monogra-
phien, Band 10, VEB Deutsche Verlag der Wissenschaften, Berlin 1970
[18] F. L AUBIE , Extensions de Lie et groupes d’automorphismes de corps locaux,
Comp. Math., (1988) 67, 165-189
[19] F.L AUBIE, Une théorie du corps de classes local non abélien, Compos. Math.,
(2007) 143, no. 2, 339-362.
[20] M. L AZARD, Sur les groupes nilpotentes et les anneaux de Lie, Ann. Ecole
Norm. Sup. (1954) 71, 101-190
[21] S H .M OCHIZUKI, A version of the Grothendieck conjecture for p-adic local
fields, Int. J. Math., (1997) 8, no.4, 499-506
[22] J.-P.S ERRE, Local Fields, Berlin, New York: Springer-Verlag, 1980
[23] J.-P.S ERRE, Cohomologie Galoisienne, Springer Verlag, Berlin-Gottingen-
Heidelberg-New York, 1964
[24] J.-P.S ERRE, Structure de certains pro-p-groupes (d’aprés Demuŝkin) Séminaire
Bourbaki, 8, Exp. No. 252, 145-155, Soc. Math. France, Paris, 1995.
[25] I.R. S HAFAREVICH, On p-extensions (In Russian), Mat. Sbornik (1947) 20,
351-363
[26] J.-P. W INTENBERGER , Le corps des normes de certaines extensions infinies des
corps locaux; application, Ann. Sci. Ec. Norm. Super., IV. Ser, (1983) 16 59-89
[27] W. Z INK, Ramification in local Galois groups; the second central step, Pure
Appl. Math. Q. (2009) 5, no. 1, 295-338.
 2
A characterisation of ordinary modular
eigenforms with CM
Rajender Adibhatla1 and Panagiotis Tsaknias2
1 Mathematics Research Unit, University of Luxembourg, 6 rue Coudenhove-Kalergi,
L-1359 Luxembourg
E-mail address: [email protected]
2 Mathematics Research Unit, University of Luxembourg, 6 rue Coudenhove-Kalergi,
L-1359 Luxembourg
E-mail address: [email protected]

A BSTRACT. For a rational prime p ≥ 3 we show that a p-ordinary modular

eigenform f of weight k ≥ 2, with p-adic Galois representation ρ f , mod
p m reductions ρ f,m , and with complex multiplication (CM), is characterised
by the existence of p-ordinary CM companion forms h m modulo p m for all
integers m ≥ 1 in the sense that ρ f,m ∼ ρh m ,m ⊗ χ k−1 , where χ is the p-adic
cyclotomic character.

2.1 Introduction
For a rational prime p ≥ 3 let f be a primitive, cusp form with q-expansion

an ( f )q n and associated p-adic Galois representation ρ f . In this paper, we
prove some interesting arithmetic properties of such a form which, in addi-
tion, has complex multiplication (CM). One of the reasons that CM forms
have historically been an important subclass of modular forms is the simplicity
with which they can be expressed. They arise from algebraic Hecke characters
of imaginary quadratic fields and this makes them ideal initial candidates on
which one can check deep conjectures in the theory of modular forms. The
specific arithmetic property we establish involves higher congruence compan-
ion forms which were introduced in [1]. The precise definition and properties
of these forms are given in Section 2.2, but for now we will only remark that

Date: June 2014

2010 Mathematics Subject Classification, 11F33, 11F80

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

24
 A characterisation of ordinary modular eigenforms with CM 25

companion forms mod p m are defined as natural analogues of the classical

(mod p) companion forms of Serre and Gross [6].
The main theorem of this work (Theorem 2.6, stated and proved in Sec-
tion 2.4) establishes that given a p-ordinary CM form f , one can always find
a CM companion form mod p m for any integer m ≥ 1. The proof explicitly
finds the desired companion forms in a Hida family of CM forms, thereby
circumventing the deformation theory and modularity lifting approach of the
companion form theorem in [1]. In addition, we show that the converse is true
as well: Any p-ordinary form f which has CM companions mod p m for each
m ≥ 1 must necessarily have CM. We therefore have a complete arithmetical
characterisation of p-ordinary CM forms. In Section 2.5 we present a different
proof of the main theorem. It is elementary in the sense that we work directly
with the Hecke characters and avoid the heavy machinery of Hida families
and overconvergent modular forms. Moreover, the result in this section is, in
fact, a strengthening of the main theorem because it shows the existence of
companion forms for odd composite moduli M and not just modulo pm . On
the other hand, the argument appears to work only under the hypothesis that
the class number of the imaginary quadratic extension from which f arises is
coprime to M.

2.2 Higher companion forms

Let p ≥ 3 be a rational prime and let f be a primitive cusp form of weight

k ≥ 2, level N prime to p, Nebentypus  and q-expansion an ( f )q n . Here,
f is primitive in the sense that it is a normalised newform that is a common
eigenform for all the Hecke operators. For a place p| p in K f , the number field
generated by the an ( f )’s, let ρ f : Gal(Q/Q) −→ GL2 (K f,p ) be a continuous,
odd, irreducible Galois representation that can be attached to f . We may, after
conjugation, assume that ρ f takes values in the ring of integers of some finite
extension of Q p . This allows us to consider reductions of ρ f modulo p m (for
integers m ≥ 1) which, with the exception of the mod p reduction ρ, we
will denote by ρ f,m . For the rest of this paper we assume that ρ is absolutely
irreducible. We can, and will, define congruences and reduction mod p m even
when the elements do not lie in Z p by using the following notion of congruence
due to Ventossa and Wiese [12].
Fix an embedding Q → Q p and let Z p denote the integers of Q p . Let
K be a number field or a local field with valuation υ p , normalised such that
υ p ( p) = 1. Let α, β ∈ Z p and n ∈ N. Then, α ≡ β mod p n , if and only if
υ p (α − β) > n − 1.
26 Rajender Adibhatla and Panagiotis Tsaknias

With f an eigenform as above, p-adic cyclotomic character χ and 2 ≤ k ≤

p m−1 ( p − 1) + 1, we define a companion form of f to be as follows.

Definition 2.1. Let km ≥ 2 be the smallest integer such that km +k−2 ≡ 0 mod
p m−1 ( p − 1). Then, a companion form g of f , modulo p m , is a p-ordinary
normalised eigenform of weight km such that ρ f,m ρg,m ⊗χ k−1 mod p m . An
equivalent formulation of the above criterion in terms of the Fourier expansions
is: an ( f ) ≡ n k−1 an (g) mod p m for (n, N p) = 1.

Remark 2.2. The equivalence between the Galois side and the coefficient side
in the above definition perhaps needs further justification. One direction is
immediate if we take the traces of Frobenii of the Galois representations. The
other direction follows from the absolute irreducibility of ρ, Chebotarev and
Théorème 1 [3], which is essentially a generalisation of the Brauer-Nesbitt
theorem to arbitrary local rings.

Following Wiles [14], we say that f is ordinary at p (or simply

p-ordinary) if a p ( f ) ≡ 0 mod p for each prime p| p (in K f ). Then, by Wiles
[14] and Mazur-Wiles [9], we have
 
χ k−1 λ(( p)/a p ) ∗
ρ f |G p ∼
0 λ(a p )

where G p is a decomposition group at p and, for any α in (O K f )× , λ(α) is the

unramified character taking Frob p to α.
If ρ f mod p is absolutely irreducible, then we note that the reductions
ρ f,m |G p = ρ f |G p mod p m for m ≥ 1 are independent of the choice of lattice
used to define ρ f . A natural question is to ask when the restriction ρ f,m |G p
actually splits. It is an easy check that a sufficient condition for ρ f,m to split at
p is that f has a p-ordinary companion form modulo p m .
That the splitting at p of ρ f,m implies the existence of a companion form
mod p m is considerably more difficult to prove. This was shown in [1] but
only under the hypothesis that ρ f has full image. However, when f has
CM, Im(ρ f ) is necessarily projectively dihedral. This renders that method of
loc. cit. inapplicable even though, a priori, we know that ρ f |G p splits and
therefore expect that f should have a companion mod p m for all integers m
prime to N .
In the sequel we avoid the use of lifting theorems altogether and exhibit the
companion forms by working directly with the Hecke character associated to
the CM form and visualising them as part of a Hida family.
 A characterisation of ordinary modular eigenforms with CM 27

2.3 Hecke characters and CM forms

In this section we briefly describe the well-known connection between forms
with complex multiplication (CM forms) and Hecke characters of imaginary
quadratic fields. The interested reader may consult [11, §3] and [10, Chapter
VII, §6].
Let K be a number field, O = O K its ring of integers and m an ideal of K .
We denote by J m the group of fractional ideals of O K that are coprime to m.

Definition 2.3. A Hecke character ψ of K , of modulus m, is a group homo-

morphism ψ : J m → C∗ such that there exists a character ψ ∞ : (O K /m)∗ →
C∗ and a group homomorphism ψ∞ : K R∗ → C∗ , where K R := K ⊗Q R, such
that:

ψ((α)) = ψ ∞ (α)ψ∞ (α) for all α ∈ O K prime to m. (2.1)

We refer to ψ ∞ and ψ∞ as the finite and the infinity type of ψ respectively.

The conductor of ψ is the conductor of ψ ∞ and we will call ψ primitive
if its conductor and modulus are equal. We now specialise to the imaginary
√
quadratic field case. Let K = Q( −d), where d is a squarefree positive inte-
ger. We will denote by D its discriminant and by (D|.) the Kronecker symbol
associated to it. The only possible group homomorphisms ψ∞ are of the form
σ u , where σ is one of the two conjugate complex embeddings of K and u ≥ 0
is an integer. The following well-known theorem [8, Theorem 4.8.2] due to
Hecke and Shimura then associates a modular form to the Hecke character
ψ of K .

Theorem 2.4. Given a Hecke character ψ of infinity type σ u and finite type
ψ ∞ with modulus m, assume u > 0 and let Nm be the norm of m. Then,
 
f = n ( Na=n ψ(a))q n is a cuspidal eigenform in Su+1 (N, ), where N =
|D|Nm and (m) = (D|m)ψ ∞ (m) for all integers m prime to N .

The eigenform associated with a Hecke character ψ is new if and only if ψ is

primitive. (See Remark 3.5 in [11].) An important feature of forms which arise
in this way is that they coincide with CM forms whose definition we recall.

Definition 2.5. A newform f is said to have complex multiplication, or just

CM, by a quadratic character φ : G Q −→ {±1} if aq ( f ) = φ(q)aq ( f ) for
almost all primes q. We will also refer to CM by the corresponding quadratic
extension.
28 Rajender Adibhatla and Panagiotis Tsaknias

It is clear that the cusp form f in the above theorem has CM by (D|.).
Indeed, the coefficient aq ( f ) in the Fourier expansion of f is 0 if no ideal of
K has norm equal to q. Since (D|q) = −1 exactly when this holds for q,
aq ( f ) = (D|q)aq ( f ), and f has CM. On the other hand, Ribet [11] shows
that if f has CM by an imaginary quadratic field K then it is induced from a
Hecke character on K .

2.4 The main theorem

We first describe the setting in which we prove our main result. Let Mk† (N )
denote the space of overconvergent modular forms of level N and weight k in
the sense of Coleman [4] and let M † (N ) be the graded ring of overconvergent

modular forms of level N . The θ operator, θ := q dq d
, then maps an q n

to nan q n . Note in particular that if f has (finite) slope α then θ ( f ) has
slope α + 1. Furthermore, by [4, Proposition 4.3],
†
θ k−1 : M2−k (N ) → Mk† (N ),
and classical CM forms of slope k − 1 lie in this image ([4, Proposition 7.1]).
We exploit this property of CM forms to prove our main theorem which is the
following.

Theorem 2.6. Let p ≥ 3 be a rational prime and f = an ( f )q n be a
p-ordinary primitive CM cuspidal eigenform of weight k ≥ 2 and level N
prime to p. Then for every integer m ≥ 1, there exists a p-ordinary CM eigen-

form h m = an (h m )q n of weight km that is a companion form for f mod p m .
√
Proof. Let K = Q( −d) be the imaginary quadratic field by which f has
CM, D be the discriminant of K and σ : K → C one of its two complex
embeddings which we fix for the rest of this section. Let g be the ( p-old)
eigenform of level N p and weight k, whose p-th coefficient has p-adic valua-
tion equal to k − 1 and whose q-expansion agrees with that of f at all primes
n coprime to p. Clearly g has CM by K as well. Let ψ be the Hecke character
over K of modulus m and infinity type σ k−1 associated with the CM form g.
The p-ordinariness of f implies that p = pp̄ is split in K . If r is a rational
prime coprime to m that splits in K , say r = rr̄, then we have:
ar (g) = ψ(r) + ψ(r̄)
In the proof of Proposition 7.1 in [4] it is shown how to obtain a p-ordinary
(CM) p-adic eigenform h 2−k of weight 2 − k such that
 A characterisation of ordinary modular eigenforms with CM 29

θ k−1 (h 2−k ) = g.
It can be easily seen that θ k−1 has the following effect on q-expansions:
θ k−1 ( cn q n ) = n k−1 cn q n .
Therefore,
an ( f ) = n k−1 an (h 2−k ) for (n, p) = 1. (2.2)
We will denote by L the extension of Q p generated by the coefficients of ψ
and λ and by P its prime ideal above p. We will also denote by E the extension
of Q p in which λ takes its values. It is clear from (2.2) that it is enough to find
classical forms h km of weight km that are congruent to h 2−k mod Pm , where
P is the ideal above p in L. By Proposition 7.1 in [4],
h 2−k = ψ̄ −1 (a)q Na
a
where the sum runs over all the integral ideals of a of K away from m. Ghate in
[5, pp 234-236], following Hida, shows how to construct a p-adic CM family
admitting a specific CM form as a specialisation. We outline the construc-
tion. Let λ be a Hecke character of conductor p and infinity type σ . We have
that O× ∼ ×
E = μ E × W E , where W E is the pro- p part of O E . Let  denote
the projection from O×E to W E . One then gets (part of) the family mentioned
above by
G(w) := ψ̄ −1 (a)λ(a)w−(2−k) q Na .
a

For any integer w ≥ 2, ψw (a) = ψ̄ −1 (a)λ(a)w−(2−k) defines a Hecke

character of infinity type w − 1, so that by Theorem 3.1, G(w) is a p-adic
CM eigenform of weight w. Moreover all of them are p-ordinary ([5, pp
236]) and therefore classical for weight w ≥ 2 ([7, Theorem I]). Clearly
G(2 − k) = h 2−k . Notice also that all the ψw have coefficients in L. Let
k E , k L be the residue fields of E and L, respectively, and consider the com-
position O× × ×
E → k E → k L , where the first map is the obvious surjection and
the second one is the obvious injection. The image has prime-to- p order so the
kernel contains W E . In particular λ ≡ 1 mod P. This implies:

λ(a)w−w ≡ 1 mod P for all w, w  ∈ Z p .
It then follows easily that if w ≡ w mod p m−1 ( p − 1) then,
 
λ(a)w−w ≡ 1 mod Pm ,
where m  = (m − 1)e + 1, with e being the ramification degree of L over Q p .
Consider the members of the family with weight km ≥ 2 which is the smallest
30 Rajender Adibhatla and Panagiotis Tsaknias

integer such that k + km ≡ 2 mod p m−1 ( p − 1). The previous identity then
gives:

λ(a)km −(2−k) ≡ 1 mod Pm . (2.3)
   
Since G(w) = n −1 w−(2−k) q n , the r -th coefficient of
a=n ψ̄ (a)λ(a)
G(w) (for r a rational prime is):

ar (G(w)) = ψ̄ −1 (r)λ(r)w−(2−k) + ψ̄ −1 (r̄)λ(r̄)w−(2−k) .

The identity rr̄ = r along with (2.3) then shows that


aq (G(km )) ≡ aq (h 2−k ) mod Pm .

For the primes r that are inert in K the above equivalence is trivially true since
in this case ar (G(km )) = 0 = ar (h 2−k ). We thus get that for all primes r away
from N p the following holds:

ar (G(km )) ≡ ar (h 2−k ) mod Pm .

As we mentioned before, all the members of G with weight w ≥ 2 are classical

forms so every h m := G(km ) is classical. Finally the last identity implies that

an (h m ) is congruent to an (h 2−k ) modulo Pm for all (n N , p) + 1, as required.

Note that the Fourier coefficient version of Definition 2.1 was used to show
companionship in the above proof. As noted in Remark 2.2 following the
definition, this formulation can be reconciled with the Galois representation
formulation if one knows that ρ f is absolutely irreducible. For f as in the
theorem above, ρ f is irreducible because it has projectively dihedral image.
Absolute irreducibility then follows because ρ f is odd and p ≥ 3.
We conclude this section with by observing that the converse to Theo-
rem 2.6 is also true; therefore giving a complete arithmetic characterisation of
p-ordinary CM forms. Assume that f has CM companions h m for all m ≥ 1
and assume that h m has CM with respect to a non-trivial quadratic character
m : (Z/Dm Z)× −→ {±1} ⊂ C× . The companionship property between f
and each of the h m s enforces the following compatibility congruences:

h m 1 ≡ h m 2 mod p m 1 for all m 2 ≥ m 1

m 1 ≡ m 2 mod p m 1 for all m 2 ≥ m 1

where ≡ means “away from p”. The second compatibility congruence, com-
bined with the fact that the characters m are valued in ±1 and p ≥ 3, implies
that there exists a non-trivial quadratic character  such that m =  for all
 A characterisation of ordinary modular eigenforms with CM 31

m ≥ 1. In particular  has conductor D = Dm . Since h m is a companion of f

mod p m and has CM by ,
a ( f ) ≡ l k−1 a (h m ) ≡ ()l k−1 a (h m ) ≡ ()a ( f ) mod p m
for all m ≥ 1 and primes l  D N p. Thus al ( f ) = (l)al ( f ) for l  D N p, and
f has CM by definition.

2.5 An elementary approach

In the section we present a proof of Theorem 2.6 which relies purely on care-
fully manipulating the Hecke character from which the CM form f arises and
then appealing to Theorem 2.4. The weakness of this approach, which we have
been unable to overcome, is a condition on the class number of the imaginary
quadratic field K from which f arises. Nevertheless, this method allows us to
prove that f , in fact, has companion forms modulo an odd integer M ≥ 3 –
not necessarily a prime power. Let k  ≥ 2 be the smallest integer such that
k + k  ≡ 2 mod φ(M), with φ being the Euler totient function. We say that
f has a companion h modulo M of weight k  if, keeping the terminology of
Definition 2.1, an ( f ) ≡ n k−1 an (h) mod M for (n, N M) = 1.

Theorem 2.7. Let M ≥ 3 be an odd integer and f = an ( f )q n be a CM
eigenform of weight k ≥ 2, level N coprime to M and p-ordinary for each
p|M. Assume that the class number of the imaginary quadratic field K from
which f arises is coprime to M. Let k  ≥ 2 be the smallest integer such that
k + k  ≡ 2 mod φ(M), with φ being the Euler totient function. Then, there

exists a CM eigenform h = an (h)q n of weight k  that is p-ordinary for
every prime p dividing M and such that it is a companion form for f mod M.
√
Proof. Let K = Q( −D) be the quadratic imaginary field by which f has

C M and M  = ri=1 pi for primes pi dividing M. Let g be the eigenform of
level N M  and weight k, whose pi -th coefficient has pi -adic valuation equal
to k − 1 for each pi |M  and whose q-expansion agrees with that of f at all
primes away from N M  . Such a g exists and one can construct it iteratively as
follows. Let g1 be the p1 -twin oldform of weight k and level N p1 . Now g1 is
still p2 -ordinary so that one can associate to it the p2 -twin eigenform of weight
k and level N p1 p2 . Proceeding iteratively in this manner we have the desired
eigenform g = gr of weight k, and level N M  . Clearly g has CM by K as well.
Let ψg : J m → C∗ be the Hecke character over K associated with g. We will
denote by ψg∞ its finite type and by ψg,∞ its infinity type. Let σ : K → C be
32 Rajender Adibhatla and Panagiotis Tsaknias

the identity embedding of K in C. Then ψg,∞ = σ k−1 . Theorem 2.4 implies

that Nm|N M  .
−1
Consider the Hecke character ϕ = ψ¯g . We have that ϕ ∞ = ψg∞ and
ϕ∞ = σ 1−k = σ (2−k)−1 . The reason we are interested in this character is the
identity

s k−1 (ϕ(s) + ϕ(s̄)) = ψg (s) + ψg (s̄) = as (g), (2.4)

for every rational prime s that splits in K as ss̄. In view of this identity it
is enough to find a Hecke character ψ  congruent to ϕ and with infinity type

σ k −1 .
Assume for simplicity that D = 1, 3. It is then easy to see (for instance
the discussion in Section 2.4 of [13]) that the compatibility of a finite and
an infinity type is guaranteed by the identity ψg∞ (−1) = (−1)k−1 . One can
therefore find a Hecke character with finite type equal to ψg∞ and infinity type

σ k −1 , for any k  ≡ 2 − k mod 2. This Hecke character is not unique. In fact it
turns out (see for example Lemma 4.1 in [13]) that there are h K choices, where
h K is the class number of K . We explain this in more detail. Consider the group
J m /P m where J m is as in Section 2.2 and P m is the subgroup of principal
fractional ideals coprime to m. It is a finite abelian group of order h K . Let
ai denote a fixed choice of representatives that generate J m /P m , ci be their
respective orders and αi be the generator of the ideal aici . Then, given a pair

(ψ ∞ , σ k −1 ) consisting of a finite and an infinity type, any Hecke character ψ 
corresponding to that pair is completely determined by the values ψ  (ai ). Let
di (ψ  ) be a non-negative integer and ζci denote a primitive ci -th root of unity.
The only possible values for ψ  (ai ) are of the form
 
 
ψ ∞ (αi ) i αik −1 ζcdii (ψ )
ci c

√ 

where c
i ψ ∞ (αi ) and ci
αik −1 are fixed choices of ci -th roots of ψ ∞ (αi ) and

αik −1 respectively.
Let k  be the integer that we defined previously. Since k  has the same parity
as 2−k we can define a Hecke character ψ  of infinity type k  −1 and finite type
  σ

ψg∞ . Let L be the extension of K with ci ψg∞ (αi ), ci αik −1 and ζci adjoined

for all i (and for the fixed k  ). Let M = p p t p . Furthermore, for every p|M,
let p be a prime in K such that p = pp̄ and pick a prime P above p in L. Our
goal is to show that the di (ψ  )’s can always be chosen so that

t
ϕ(ai ) ≡ ψ  (ai ) (mod P P )
 A characterisation of ordinary modular eigenforms with CM 33

 = e(P/p)(t − 1) + 1. First, notice that if α ≡ 0 (mod p)

for all i, where tP p i
then the desired congruence is immediate. In what follows we can therefore
t
freely assume that αi is a unit in O K /pt p ⊆ O L /P P . Moreover, O K /pt p is
isomorphic to Z/ p t p Z because of the splitting condition on the primes divid-

ing M. Since the order of (Z/ p t p Z)∗ √ is φ( pt p ), we get that α k −(2−k) ≡ 1

(mod pt p ). This in turn implies that i α k −(2−k) is a ci -th root of unity in
c

t
O L /P P . Since h K is coprime to M this ci -th root of unity will lift to one in
d (ϕ)−di (ψ  )
O L so all one has to do is to choose di (ψ  ) so that ζcii is this root. In
conclusion, the CM form h that one may associate to the Hecke character ψ 
by Theorem 2.4 is the desired companion of f mod M. Equation (5.1) ensures
that h is p-ordinary for each p dividing M.

Example 2.8. Let f be the CM newform of weight 3 and level 8 with the
following Fourier expansion,

q − 2q 2 − 2q 3 + 4q 4 + 4q 6 − 8q 8 − 5q 9 + 14q 11 − 8q 12 + 16q 16 + 2q 17
+10q 18 − 34q 19 − 28q 22 + 16q 24 + 25q 25 + · · ·

It is ordinary at 3, 11 and 17. Using MAGMA [2] we find companions

h 19 , h 31 and h 59 modulo 33, 51 and 99 respectively. The indices denote the
weights of the companions; each has level 8 and CM by the quadratic Dirichlet
character of conductor 8. Their Fourier expansions are:

h 19 = q − 512q 2 − 3266q 3 + 262144q 4 + 1672192q 6 − 134217728q 8 −

376753733q 9 − 354349618q 11 − 856162304q 12 + 68719476736q 16 +
119842447106q 17 + 192897911296q 18 + 335013705758q 19 +
181427004416q 22 + 438355099648q 24 + 3814697265625q 25 + · · ·

h 31 = q − 32768q 2 − 26595314q 3 + 1073741824q 4 + 871475249152q 6 −

35184372088832q 8 + 501419594663947q 9 + 6656187998706302q 11 −
28556500964212736q 12 + 1152921504606846976q 16 −
4422784932886529086q 17 − 16430517277948215296q 18 −
23964789267887608402q 19 − 218109968341608103936q 22 +
935739423595322933248q 24 + 931322574615478515625q 25 − · · ·

h 59 = q − 536870912q 2 + 57281430144478q 3 + 288230376151711744q 4 −

30752733642330195623936q 6 − 154742504910672534362390528q 8 −
1428966457849531926967711205q 9 +
2900908653579886134108518505134q 11 +
16510248157050893929760929349632q 12 +
34 Rajender Adibhatla and Panagiotis Tsaknias

83076749736557242056487941267521536q 16 +
955027058519269179716584293727217282q 17 +
767170525443087764404272509180968960q 18 −
15840463221028561793718151601779174594q 19 −
1557413474476125533674994636820167262208q 22 −
8863871985422232654486014081904498704384q 24 +
34694469519536141888238489627838134765625q 25 − · · ·

Acknowledgments
The first author thanks DFG GRK 1692 at Regensburg, the Hausdorff Research
Institute for Mathematics (HIM) and the AFR Grant Scheme of the Fonds
National de la Recherche Luxembourg (FNR) for postdoctoral support dur-
ing the course of this work. The second author acknowledges the support of
the DFG Priority Program SPP 1489, the FNR and the invitation extended by
the HIM during its Arithmetic and Geometry Trimester Program.

References
[1] R. Adibhatla and J. Manoharmayum. Higher congruence companion forms. Acta
Arith., 156: 159–175, 2012.
[2] W. Bosma, J. Cannon and C. Playoust. The Magma algebra system. I. The user
language J. Symbolic Comput., 24(3-4):235–265, 1997.
[3] H. Carayol Formes modulaires et représentations galoisiennes à valeurs dans un
anneau local complet Contemp. Math., 165:213–237, 1994.
[4] Robert F. Coleman. Classical and overconvergent modular forms. Invent. Math.,
124(1-3):215–241, 1996.
[5] Eknath Ghate. Ordinary forms and their local Galois representations. In Algebra
and number theory, pages 226–242. Hindustan Book Agency, Delhi, 2005.
[6] Benedict H. Gross. A tameness criterion for Galois representations associated to
modular forms (mod p). Duke Math. J., 61(2):445–517, 1990.
[7] H. Hida. Galois representations into GL2 (Z p [[X ]]) attached to ordinary cusp
forms. Invent. Math., 85(3):545–613, 1986.
[8] Toshitsune Miyake. Modular forms. Springer Monographs in Mathematics.
Springer-Verlag, Berlin, english edition, 2006. Translated from the 1976 Japanese
original by Yoshitaka Maeda.
[9] B. Mazur and A. Wiles. On p-adic analytic families of Galois representations.
Compositio Math., 59(2):231–264, 1986.
[10] Jürgen Neukirch. Algebraic number theory, volume 322 of Grundlehren
der Mathematischen Wissenschaften [Fundamental Principles of Mathematical
Sciences]. Springer-Verlag, Berlin, 1999. Translated from the 1992 German
original and with a note by Norbert Schappacher, With a foreword by G. Harder.
 A characterisation of ordinary modular eigenforms with CM 35

[11] Kenneth A. Ribet. Galois representations attached to eigenforms with Nebenty-

pus. In Modular functions of one variable, V (Proc. Second Internat. Conf., Univ.
Bonn, Bonn, 1976), pages 17–51. Lecture Notes in Math., Vol. 601. Springer,
Berlin, 1977.
[12] Xavier Taixési Ventosa and Gabor Wiese. Computing congruences of modular
forms and Galois representations modulo prime powers. In Arithmetic, geometry,
cryptography and coding theory 2009, volume 521 of Contemp. Math., pages
145–166. Amer. Math. Soc., Providence, RI, 2010.
[13] P. Tsaknias. A possible generalization of Maeda’s conjecture. Computational
aspects of modular forms, Conference Proceedings Heidelberg, pages 1–21.
Springer-Verlag 2013.
[14] A. Wiles. On ordinary λ-adic representations associated to modular forms. Invent.
Math., 94(3):529–573, 1988.
 3
Selmer complexes and p-adic Hodge theory
Denis Benois
Institut de Mathématiques, Université de Bordeaux, 351, cours de la Libération 33405
Talence, France
E-mail address: [email protected]

Contents
Introduction 37
3.0.1 Selmer groups 37
3.0.2 p-adic L-functions 38
3.0.3 Selmer complexes 39
3.0.4 The Main Conjecture 42
3.0.5 Organization of the paper 45
Acknowledgement 45
3.1 Overview of the theory of (ϕ, )-modules 45
3.1.1 Galois representations and (ϕ, )-modules 45
3.1.2 Cohomology of (ϕ, )-modules 47
3.1.3 Relation to the p-adic Hodge theory 50
3.1.4 Families of (ϕ, )-modules 51
3.1.5 Derived categories 54
3.1.6 Iwasawa cohomology 56
3.1.7 Crystalline extensions 59
3.1.8 Cohomology of isoclinic modules 61
3.2 The Main Conjecture 63
3.2.1 Regular submodules 63
3.2.2 Iwasawa cohomology 67
3.2.3 The complexes R Iw (V , D) and R (V, D) 69
3.2.4 The Main Conjecture 72
3.3 Local structure of p-adic representations 73
3.3.1 Filtration associated to a regular submodule 73
3.3.2 The weight  −2 case 76

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

36
 Selmer complexes and p-adic Hodge theory 37

3.4 p-adic height pairings 81

3.4.1 Extended Selmer groups 81
3.4.2 p-adic height pairings 83
References 85

Introduction
3.0.1 Selmer groups
In this paper we discuss some applications of p-adic Hodge theory to the al-
gebraic formalism of Iwasawa theory. Fix an odd prime number p. For every
finite set S of primes containing p, we denote by Q(S) /Q the maximal alge-
braic extension of Q unramified outside S∪{∞} and set G Q,S = Gal(Q(S) /Q).
Let G Qv = Gal(Qv /Qv ) denote the decomposition group at v, and let Iv be its
inertia subgroup.
Let M be a pure motive over Q. The complex L-function L(M, s) is a
Dirichlet series which converges for s  0 and is expected to admit a
meromorphic continuation on the whole C with a functional equation of the
form
(M, s)L(M, s) = ε(M, s) (M ∗ (1), −s)L(M ∗ (1), −s).

Here (M, s) is a product of -factors explicitly defined in terms of the Hodge

structure of M, and ε(V, s) is a factor of the form ε(M, s) = abs , where
a ∈ C∗ and b is a positive integer (see [55], [27]).
Let V denote the p-adic realization of M. We consider V as a p-adic repre-
sentation of the Galois group G Q,S , where S contains p, ∞ and the places
where M has bad reduction. We will write HS∗ (Q, V ) and H ∗ (Qv , V ) for
the continuous cohomology of G Q,S and G Qv respectively. The Bloch–Kato
Selmer group H 1f (Q, V ) is defined as
 
 H 1 (Qv , V )
H f (Q, V ) = ker HS (Q, V ) →
1 1
, (3.1)
v∈S
H 1f (Qv , V )

where the “local conditions” H 1f (Qv , V ) are given by


ker(H 1 (Qv , V ) → H 1 (Iv , V )) if v = p,
H 1f (Qv , V ) = (3.2)
ker(H 1 (Q p, V ) → H 1 (Q p, V ⊗ Bcris )) if v = p

(see [13]). Here Bcris is Fontaine’s ring of crystalline periods [25]. Note that,
using the inflation–restriction sequence, it is easy to see that the definition
(3.1) does not depend on the choice of S (see [27], Chapter II). The Beilinson
38 Denis Benois

conjecture (in the formulation of Bloch and Kato [13]) relates the rank of
H 1f (Q, V ∗ (1)) to the order of vanishing of L(M, s), namely one expects that

ords=0 L(M, s) = dimQ p H 1f (Q, V ∗ (1)) − dimQ p H 0 (Q, V ∗ (1)).

3.0.2 p-adic L-functions

Assume that V satisfies the following three conditions (see Subsection 3.2.1
below):

C1) HS0 (Q, V ) = HS0 (Q, V ∗ (1)) = 0.

C2) The restriction of V on the decomposition group at p is semistable in the
sense of Fontaine [26].
We denote by Dst (V ) the semistable module associated to V. Recall
that Dst (V ) is a filtered Q p -vector space equipped with a linear Frobenius
ϕ : Dst (V ) → Dst (V ) and a monodromy operator N : Dst (V ) → Dst (V )
such that N ϕ = pϕ N . Let Dcris (V ) = Dst (V ) N =0 .

C3) Dcris (V )ϕ=1 = 0.

Recall the notion of a regular submodule of Dst (V ), which plays the key role
in this paper. It was first introduced by Perrin-Riou [51] in the crystalline case.
See also [5] and [7]. Replacing, if necessary, M by M ∗ (1), we will assume that
M is of weight wt (M)  −1. We consider two cases:
The weight  −2 case. Let wt (M)  −2. From Jannsen’s conjecture
[34] (see Subsection 3.2.1 for more detail) it follows that, in this case, the
localization map
H 1f (Q, V ) → H 1f (Q p , V )

should be injective. The condition C3) implies that the Bloch–Kato exponen-
tial map
expV : Dst (V )/Fil0 Dst (V ) → H 1f (Q p , V )

is an isomorphism, and we denote by

logV : H 1f (Q p , V ) → Dst (V )/Fil0 Dst (V )

its inverse. A (ϕ, N )-submodule D of Dst (V ) is regular if D∩Fil0 Dst (V ) = {0}

and the composition of the localisation map with logV induces an isomorphism

r V,D : H 1f (Q, V ) → Dst (V )/(Fil0 Dst (V ) + D).

Note that the map r V,D is closely related to the syntomic regulator.
 Selmer complexes and p-adic Hodge theory 39

The weight −1 case. Let wt (M) = −1. In this case, we say that a (ϕ, N )-
submodule D of Dst (V ) is regular if

Dst (V ) = D ⊕ Fil0 Dst (V )

as vector spaces. Each regular D gives a splitting of the Hodge filtration on

Dst (V ) and therefore defines a p-adic height pairing

 , V,D : H 1f (Q, V ) × H 1f (Q, V ∗ (1)) → Q p

(see [45], [46], [49]), which is expected to be non-degenerate1 .

The special value L ∗ (M, n) of L(M, s) at n ∈ Z is defined as the first
non-zero coefficient in the Taylor expansion of L(M, s) at s = n. The theory
of Perrin-Riou [51] suggests that to each regular D one can associate a p-
adic L-function L p (M, D, s) interpolating the rational parts of L ∗ (M, n). Set
r (M) = ords=0 L(M, s). The interpolation formula relating special values of
the complex and the p-adic L-functions at s = 0 should have the form
L p (M, D, s) L ∗ (M, 0)
lim = E(V , D)  p (M, D) .
s→0 s r(M) ∞ (M)
Here ∞ (M) is the irrational part of L ∗ (M, 0) predicted by Beilinson–
Deligne’s conjectures, and  p (M, D) is essentially the determinant of r V ,D
(in the weight  −2 case) or the determinant of  , V ,D (in the weight −1
case). Finally, E(V, D) is an Euler-like factor. Its explicit form in the crys-
talline case was conjectured by Perrin-Riou [51]. Namely, let E p (M, X ) =
det(1 − ϕ X |Dcris (V )). Note that E p (M, p −s ) is the Euler factor of L(M, s) at
p. One expects that
 
1 − p −1 ϕ −1
E(V, D) = E p (M, 1) det |D if V is crystalline at p
1−ϕ
(see [51], Chapter 4 and [17], Conjecture 2.7).

3.0.3 Selmer complexes

We introduce basic notation of the Iwasawa theory. Let μ pn denote the group
∞
of p n -th roots of unity and Qcyc = ∪ Q(μ pn ). We fix a system (ζ p n )n0
n=1
p
of primitive p n th roots of unity such that ζ pn+1 = ζ pn for all n. The Galois

1 The non-degeneracy of the p-adic height pairing was first conjectured by Schneider [56] for
abelian varieties having good ordinary reduction at p.
40 Denis Benois

group = Gal(Qcyc /Q) is canonically isomorphic to Z∗p via the cyclotomic

character
χ (g)
χ : → Z∗p , g(ζ pn ) = ζ p n for all g ∈ .
Let  = Gal(Q(μ p )/Q) and 0 = Gal(Qcyc /Q(μ p )). We denote by F∞ =
Q(μ p∞ ) the cyclotomic Z p -extension of Q and set Fn = Q(μ pn+1 ) , n ∈ N.
Fix a generator γ ∈ and set γ0 = γ p−1 ∈ 0 . Let  denote the Iwasawa
algebra Z p [[ 0 ]]. The choice of γ fixes an isomorphism  Z p [[X ]] such
that γ0 → 1 + X. Let H denote the ring of power series, which converge on
the open unit disk. We consider  as a subring of H via the above chosen
isomorphism  Z p [[X ]]. Note that H is the large Iwasawa algebra intro-
duced in [50]. The map τ → τ −1 , τ ∈ 0 defines a canonical involution on
, which we denote by ι :  → .
Let T be a Z p -lattice of V stable under the action of G Q,S . We denote by
T ⊗ ι the tensor product T ⊗Z p  equipped with the following structures:
a) The continuous action of G Q,S defined by
g(a ⊗ λ) = g(a) ⊗ ḡλ, g ∈ G Q,S ,
where ḡ ∈ 0 is the image of g under the canonical projection G Q,S → 0;
b) The structure of a -module defined by
τ (a ⊗ λ) = a ⊗ ι(τ )(λ) = a ⊗ τ −1 λ, τ ∈ 0.

Note that the action of G Q,S on T ⊗ ι is -linear.

Let C • (G Q,S , T ⊗ ι ) (respectively C • (G Qv , T ⊗ ι )) be the com-
plex of continuous cochains of G Q,S (respectively G Qv ) with coefficients in
T ⊗ ι . We denote by R Iw,S (Q, T ) and R Iw,S (Qv , T ) these complexes
viewed as objects of the bounded derived category D b (Λ) of -modules.
i
Set HIw,S (Q, T ) = Ri Iw,S (Q, T ) and HIwi
(Qv , T ) = Ri Iw (Qv , T ). From
Shapiro’s lemma it follows that
i
HIw,S (Q, T ) = lim HSi (Fn , T ), i
HIw (Qv , T ) = lim HSi (Fn,v , T )
← − ← −
n n

as -modules. In [46], Nekovář considers the diagrams of the form

(3.3)
 Selmer complexes and p-adic Hodge theory 41

The cone of this diagram can be viewed as the derived version of Selmer
groups, where the complexes Uv• (T ) play the role of local conditions. For
v = p, the natural choice of Uv• (T ) is to put Uv• (T ) = R Iw, f (Qv , T ), where
 
ι Frv −1 ι
R Iw, f (Qv , T ) = T ⊗  −−−→ T ⊗  .
Iv Iv

Here Frv denotes the geometric Frobenius, and the terms are placed in degrees
0 and 1. Note that the cohomology groups of
 
Iv Frv −1
R Iw, f (Qv , T ) ⊗L Q
 p = V −−− → V Iv

are H 0 (Qv , V ) and H 1f (Qv , V ). The local conditions at p are more delicate to
define. First assume that V satisfies the Panchishkin condition, i.e. the restric-
tion of V on the decomposition group at p has a subrepresentation F + V ⊂ V
such that
DdR (V ) = DdR (F + V ) ⊕ Fil0 DdR (V )
as vector spaces. Set U p• (T ) = R Iw (Q p , F + T ). Then U p• (T ) is the de-
rived version of Greenberg’s local conditions [28], and the cohomology of
Nekovář’s Selmer complex is closely related to the Pontryagin dual of Green-
berg’s Selmer group [46]. If we assume, in addition, that Dcris (V ∗ (1))ϕ=1 = 0,
then
H 1f (Q p , V ) = ker(H 1 (Q p , V ) → H 1 (I p , V /F + V )),
and we can compare Greenberg and Bloch–Kato Selmer groups. Roughly
speaking, in this case different definitions lead to pseudo-isomorphic -
modules, which have therefore the same characteristic series in the case they
are -torsion (see [20], [46], Chapter 9 and [48]). If Dcris (V ∗ (1))ϕ=1 = 0, the
situation is more complicated. The analytic counterpart of this problem is the
phenomenon of extra-zeros of p-adic L-functions studied in [29], [5], [7].
We no longer assume that V satisfies the Panchishkin condition. The theory
of (ϕ, )-modules (see [23], [15], [9]) associates to V a finitely generated free
module D†rig (V ) over the Robba ring R equipped with a semilinear actions of
the group and a Frobenius ϕ, which commute with each other (see Sub-
section 3.1.1 below). The category of (ϕ, )-modules has a nice cohomology
theory, whose formal properties are very close to properties of local Galois
cohomology [32], [33], [42]. In particular, H ∗ (D†rig (V )) is canonically isomor-
phic2 to the continuous Galois cohomology H ∗ (Q p , V ). Moreover, to each
(ϕ, )-module D we can associate the complex
2 Up to the choice of a generator of
0.
42 Denis Benois

 
ψ−1
R Iw (D) = D −−→ D ,

where ψ is the left inverse to ϕ (see [14]), and the first term is placed in degree
∗ (D) for the cohomology of R
1. We will write HIw Iw (D). The action of 0

induces a natural structure of H -module on D . From a general result of
Pottharst ([54], Theorem 2.8) it follows that
∼
i
HIw (Q p , T ) ⊗ H → HIw
i
(D†rig (V ))

as H -modules.
The approach to Iwasawa theory discussed in this paper is based on the
observation that the (ϕ, )-module of a semistable representation looks like
an ordinary Galois representation, in particular, it is a successive extension
of (ϕ, )-modules of rank 1. This was first pointed out by Colmez in [19],
where the structure of trianguline (ϕ, )-modules of rank 2 over Q p was stud-
ied in detail. Therefore in the non-ordinary setting we can adopt Greenberg’s
approach working with (ϕ, )-modules instead of p-adic representations. This
idea was used in [2] and [3] to study families of Selmer groups, and in [5],
[6], [7] to study extra-zeros of p-adic L-functions. Pottharst [52] developed
a general theory of Selmer complexes in this setting and related it to the
Perrin-Riou’s theory from [53].

3.0.4 The Main Conjecture

Fix a regular submodule D of Dst (V ). By [11], we can associate to D a
canonical (ϕ, )-submodule D of D†rig (V ). Consider the diagram

in the derived category of H -modules, where the local conditions Uv• (V, D)
are

R Iw, f (Qv , T ) ⊗L
H if v = p
Uv• (V , D) =
R Iw (D) if v = p.

Consider the Selmer complex associated to these data

 Selmer complexes and p-adic Hodge theory 43

  
  
•
Iw (V , D) = cone Iw, f (Qv , T ) ⊗ H Uv (V, D) →
L
R R
v∈S


L
R Iw (Qv , T ) ⊗ H [−1]. (3.4)
v∈S
Set
i
HIw (V , D) = Ri Iw (V, D).
We propose the following conjecture.

Main Conjecture. Let M/Q be a pure motive of weight  −1 which does not
contain submotives of the form Q(m). Assume that the p-adic realization V
of M satisfies the conditions C1–3) from Subsection 3.0.2. Let D be a regular
submodule of Dst (V ). Then
i
i) HIw (V, D) = 0 for i = 2.
2 (V, D) is a coadmissible3 torsion H -module, and
ii) HIw
 
2
charH HIw (V, D) = ( f D ),

where L p (M, D, s) = f D (χ (γ0 )s − 1).

Remarks. 1) Since H ∗ = Λ[1/ p]∗ , our Main Conjecture determines f D up

to multiplication by a unit in Λ[1/ p].
2) Assume that M is critical and that V is ordinary at p. Then the restriction of
V to the decomposition group at p is equipped with an increasing filtration
(Fi V )i∈Z such that for all i ∈ Z
gri (V ) = Wi (i),
where Wi is an unramified representation of G Q p . From the definition of
a regular submodule it follows immediately that D = Dst (F 1 V ) is regu-
lar. Coates and Perrin-Riou [16] conjectured that in this situation f D ∈ .
In [28], Greenberg defined a cofinitely generated -module (Greenberg’s
Selmer group) S(F∞ , V ∗ (1)/T ∗ (1)) and conjectured that its Pontryagin
dual S(F∞ , V ∗ (1)/T ∗ (1))∧ is related to the p-adic function L p (V, D, s)
as follows
charΛ S(F∞ , V ∗ (1)/T ∗ (1))∧ = f D Λ.
Using the results of [46], one can check that this agrees with our Main
Conjecture, but Greenberg’s conjecture is more precise because it deter-
mines f D up to multiplication by a unit in Λ. See [53] and Section 3.2.4
for more detail.
3 See Section 3.1.6 for the definition of a coadmissible module and its characteristic ideal.
44 Denis Benois

3) Assume that V is crystalline at p. In [51], for any regular D, Perrin-Riou

defined a free -module LIw (D, V ) together with a canonical trivialization
∼
i V ,D : LIw (D, V ) → H .

See also [7] for the interpretation of Perrin-Riou’s theory in terms of Selmer
complexes. Perrin-Riou’s main conjecture says that

i V ,D (LIw (D, V )) =  f D .

In [53], Pottharst proved that

i V,D (LIw (D, V )) ⊗Z p Q p = charH HIw

2
(V , D)

i.e. for crystalline representations our conjecture is compatible with Perrin-

Riou’s theory.

∞
4) Let f = an q n be a newform of weight k on 0 (N ), and let W f be the p-
n=1
adic representation associated to f. Then W f can be seen as the p-adic real-
ization of a pure motive M f of weight k −1. The representation W f is crys-
talline if ( p, N ) = 1 and semistable non-crystalline if p  N . (As usual,
we write p  N if N = pN  with ( p, N  ) = 1.) Fix 1  m  k − 1. Then
the motive M f (m) is critical, and we denote by V = W f (m) its p-adic re-
alization. The subspace Fil0 Dst (V ) is one dimensional. Using the extension
of scalars to a finite field extension E/Q p , we can assume that Dst (V ) =
Eeα + Eeβ , where ϕ(eα ) = αeα and ϕ(eβ ) = βeβ for some α, β ∈ E. If
( p, N ) = 1, the eigenspaces Dα = Eeα and Dβ = Eeα are regular and

L(M f (m), Dα , s) = L p,α ( f, s+m), L(M f (m), Dβ , s) = L p,β ( f, s+m),

where L p,α ( f, s) and L p,β ( f, s) are the usual p-adic L-functions asso-
ciated to α and β [43], [61], [44]. If p  N , we have N eβ = eα and
N eα = 0 and Dα is the unique regular subspace of Dst (V ). The results of
Kato [36] have the following interpretation in terms of our theory. Assume
that ( p, N ) = 1 and set L(V , Dα , s) = f Dα (χ (γ0 )s −1). Then HIw
2 (V, D )
α
is H -torsion and
2
charH HIw (V, Dα ) ⊂ ( f Dα )

(see [53], Theorem 5.4). In the ordinary case, the opposite inclusion was
recently proved under some technical conditions by Skinner and Urban
[58]. It would be interesting to understand whether their methods can be
generalized to the non-ordinary case.
5) It is certainly possible to formulate the Main Conjecture for families of
p-adic representations in the spirit of [30].
 Selmer complexes and p-adic Hodge theory 45

3.0.5 Organization of the paper

The first part of the paper is written as a survey article. In Section 3.1, we re-
view the theory of (ϕ, )-modules. In particular, we discuss recent results of
Kedlaya, Pottharst and Xiao [40] about the cohomology of families of (ϕ, )-
modules and its applications to the Iwasawa cohomology. In Section 3.2, we
apply this technique in the context of global Iwasawa theory. The notion of
regular submodule is discussed in Section 3.2.1. In Subsections 3.2.2 and
3.2.3, we construct the complex R Iw (V, D) and review its basic properties
following Pottharst [53].
The Main Conjecture is formulated in Subsection 3.2.4. In the rest of the
paper, we discuss the relationship between the nullity of HIw 1 (V , D) and the

structure of the semistable module Dst (V ). In Section 3.3, we consider the

weight  −2 case and prove that HIw 1 (V , D) = 0 (and therefore H 2 (V , D)
Iw
is H -torsion) if the -invariant (V, D), constructed in [5], [7], does not van-
ish. In Section 3.4, we consider the weight −1 case and prove that the nullity
of HIw1 (V, D) follows from the non-degeneracy of the p-adic height pairing.

The proof is based on a generalization of Nekovář’s construction of the p-

adic height pairing [46] to the non-ordinary case. A systematic study of p-adic
heights via (ϕ, )-modules will be done in [8].

Acknowledgement
I would like to thank the referee for numerous suggestions in improving the
paper and for a very patient correction of many minor points.

3.1 Overview of the theory of (ϕ, )-modules

3.1.1 Galois representations and (ϕ, )-modules
In this section we summarize the results about (ϕ, )-modules, which will be
used in subsequent sections. The notion of a (ϕ, )-module was introduced
by Fontaine in his fundamental paper [23]. We consider only (ϕ, )-modules
over Q p and their families, because this is sufficient for applications we have
in mind.
Let p be a prime number. Fix an algebraic closure Q p of Q p and set G Q p =
cyc cyc
Gal(Q p /Q p ). Let Q p = Q p (μ p∞ ) and = Gal(Q p /Q p ). The cyclotomic
character χ : G Q p → Z∗p induces an isomorphism of onto Z∗p , which we
denote again by χ : → Z∗p . Let C p be the p-adic completion of Q p . We
denote by | · | p the p-adic absolute value on C p normalized by | p| p = 1/ p.
46 Denis Benois

We fix a coefficient field E, which is a finite extension of Q p , and consider the

following objects:
● The field E E of power series f (X ) = ak X k , ak ∈ E such that ak are
k∈Z
p-adically bounded and ak → 0 when k → −∞. Thus, E E is a complete
discrete valuation field with residue field k E ((X )), where k E is the residue
field of E.
● For each 0  r < 1 the ring R E(r) of p-adic functions

f (X ) = ak X k , ak ∈ E, (3.5)
k∈Z

which are holomorphic on the p-adic annulus

A(r, 1) = {z ∈ C p | p −1/r  |z| p < 1}.

The Robba ring of power series with coefficients in E is defined as

(r)
R E = ∪R E .
r

Note that R E is a Bézout ring (every finitely generated ideal in R E is princi-

pal) [41] but it is not noetherian. Its group of units coincides with the group
of units of O E [[X ]] ⊗ Q p , where O E is the ring of integers of E.
● For each 0  r < 1 the ring E E†,r of p-adic functions (3.5) that are bounded
on A(r, 1). Then E E† = ∪E E†,r is a field, which is contained both in E E and
r
in R E . Its elements are called overconvergent power series.
The rings E E , E E† and R E are equipped with an E-linear continuous action
of defined by

g( f (X )) = f ((1 + X )χ (g) − 1), g∈

and a linear operator ϕ called the Frobenius and given by

ϕ( f (X )) = f ((1 + X ) p − 1).

Note that the actions of and ϕ commute with each other. Set t = log(1 +
∞
Xn
X) = (−1)n−1 . Then t ∈ RQ p and γ (t) = χ (γ ) t, ϕ(t) = p t.
n
n=1

Definition. i) A (ϕ, )-module over R = E E , E E† or R E is a finitely gener-

ated free R-module D equipped with commuting semilinear actions of and
ϕ and such that A ⊗ϕ(A) ϕ(D) = D. In other words, if e1 , . . . , ed is an R-basis
of D, then ϕ(e1 ), . . . , ϕ(ed ) is also a basis of D over R.
 Selmer complexes and p-adic Hodge theory 47

ϕ,
We denote by M R the category of (ϕ, )-modules over R. The category
ϕ, ϕ, ,ét
ME E contains the important subcategory ME E of étale modules. A (ϕ, )-
module D over E E is étale if it is isoclinic of slope 0 in the sense of Dieudonné–
Manin’s theory. More explicitly, D is étale if there exists a (ϕ, )-stable lattice
OE E e1 + · · · + OE E ed of D over the ring of integers OE E of E E such that
the matrix of ϕ in the basis {e1 , . . . , ed } is invertible over OE E . The category
ϕ, ,ét
M † of étale modules over E E† can be defined by the same manner.
EE
Let Rep E (G Q p ) denote the ⊗-category of p-adic representations of the Ga-
lois group G Q p = Gal(Q p /Q p ) on finite dimensional E-vector spaces. Using
the field-of-norms functor [62], Fontaine constructed in [23] an equivalence of
⊗-categories
ϕ, ,ét
D : Rep E (G Q p ) → ME E
and conjectured that each p-adic representation V is overconvergent, i.e. D(V )
has a canonical E E† -lattice D† (V ) stable under the actions of ϕ and . This was
proved later by Cherbonnier and Colmez in [14].
The relationship between p-adic representations and (ϕ, )-modules over
the Robba ring can be summarized in the following diagram

A striking fact is that the vertical arrow is a fully faithful functor. This fol-
lows from Kedlaya’s generalization of Dieudonné–Manin theory [38]. More
precisely, the functor D → D ⊗E † R E establishes an equivalence between
E
ϕ, ,ét
M and the category of (ϕ, )-modules over R E of slope 0 in the sense of
E E†
Kedlaya. See [19], Proposition 1.7 for details.

3.1.2 Cohomology of (ϕ, )-modules

On the categories of (ϕ, )-modules one can define cohomology theories
whose formal properties are very similar to the properties of continuous Galois
cohomology of local fields. The main idea is particulary clear if we work with
ϕ, ,ét
the category of étale modules. Since ME E is equivalent to Rep E (G Q p ) it is
48 Denis Benois

possible to compute the continuous Galois cohomology H ∗ (Q p , V ) in terms of

D(V ). In practice, since the category of p-adic representations has no enough
of injective objects, one should first work in the context of pm -torsion modules
and consider the category of inductive limits of (ϕ, )-modules over OE E / p m .
This leads to the following results [32], [33]. Let  = Gal(Q p (μ p )/Q p ). Then
 × 0 where 0 is a procyclic p-group. Fix a topological generator γ0 of
pn 
0 and set γn = γ0 , n  0. Let K n = Q p (μ p n ) , and K ∞ = ∪ K n . Then
n0
0 = Gal(K ∞ /Q p ). If M is a topological module equipped with a continuous
action of and an operator ϕ which commute with each other, we denote by
• (M) the complex
Cϕ,γ n

• d0 d1
Cϕ,γ n
(M) : 0 → M  −
→ M ⊕ M −
→ M  → 0, n  0, (3.6)
where d0 (x) = ((ϕ − 1)x, (γn − 1)x) and d1 (y, z) = (γn − 1)y − (ϕ − 1)z
and define
H ∗ (K n , M) = H ∗ (Cϕ,γ
•
n
(M)).
Let D be an étale (ϕ, )-module over E E , and let V be a p-adic representa-
tion of G Q p such that D = D(V ). Then there are isomorphisms
H ∗ (K n , D) H ∗ (K n , V ),
which are functorial and canonical up to the choice of the generator γ ∈ (see
[32]). This gives an alternative approach to the Euler-Poincaré characteristic
formula and the Poincaré duality for Galois cohomology of local fields [32],
[33]. If D is an an étale (ϕ, )-module over E E† , the theorem of Cherbonnier-
Colmez implies that again H i (K n , D) H i (K n , V ), where V is a p-adic
†
representation such that D = Drig (V ). The cohomology of (ϕ, )-modules
over R E was studied in detail in [42] using a non-trivial reduction to the slope
0 case. For any (ϕ, )-module D over R E we consider D(χ ) = D ⊗ E E(χ )
equipped with diagonal actions of and ϕ (here ϕ acts trivially on E(χ )). The
main properties of the cohomology groups H i (K n , D) of D are:
1) Long cohomology sequence. A short exact sequence of (ϕ, )-modules
over R E
0 → D → D → D → 0
gives rise to an exact sequence

δ0
0 → H 0 (K n , D ) → H 0 (K n , D) → H 0 (K n , D) −
→ H 1 (K n , D ) →
δ1
→ H 1 (K n , D ) → H 1 (K n , D) → H 1 (K n , D ) −
→ H 2 (K n , D ) →
→ H 2 (K n , D) → H 2 (K n , D ) → 0. (3.7)
 Selmer complexes and p-adic Hodge theory 49

2) Euler–Poincaré characteristic. Let D be a (ϕ, )-module over R E . Then

for all i, H i (D) are finite dimensional E-vector spaces, and
2
χ (K n , D) = (−1)i dim E H i (K n , D) = −[K n : Q p ] rankR E (D). (3.8)
i=0

3) Computation of the Brauer group. The map

pn
cl( f ) → − res( f dt), (3.9)
log χ (γn )

where res( f dt) denotes the residue of the differential form f dt = f 1+X ,
dX
is
∼
well defined and induces an isomorphism inv K n : H 2 (K
n , R E (χ )) → E.
4) Cup-products. Let D and D be two (ϕ, )-modules over R E . For all i
and j such that i + j  2 define a bilinear map

∪ : H i (K n , D ) × H j (K n , D ) → H i+ j (K n , D ⊗ D )

by

cl(x) ∪ cl(y) = cl(x ⊗ y) if i = j = 0,

cl(x) ∪ cl(y1 , y2 ) = cl(x ⊗ y1 , x ⊗ y2 ) if i = 0, j = 1,
cl(x1 , x2 ) ∪ cl(y1 , y2 ) = cl(x2 ⊗ γn (y1 ) − x1 ⊗ ϕ(y2 )) if i = 1, j = 1,
cl(x) ∪ cl(y) = cl(x ⊗ y) if i = 0, j = 2.
(3.10)

These maps commute with connecting homomorphisms in the usual sense.

5) Duality. Let D∗ = HomR E (D, R E ). For i = 0, 1, 2 the cup product
∪
H i (K n , D) × H 2−i (K n , D∗ (χ )) −
→ H 2 (K n , R E (χ )) E (3.11)

is a perfect pairing.
6) Comparison with Galois cohomology. Let D be an etale (ϕ, )- module
over E E† . Then the map
• •
Cϕ,γn
(D) → Cϕ,γn
(D ⊗ R E ),

induced by the natural map D → D ⊗E † R E , is a quasi-isomorphism. In

E
particular, if V is a p-adic representation of G Q p , then
∼
H ∗ (K n , V ) → H ∗ (K n , D†rig (V )).
50 Denis Benois

3.1.3 Relation to the p-adic Hodge theory

In [23], Fontaine proposed to classify the p-adic representations arising in the
p-adic Hodge theory in terms of (ϕ, )-modules (Fontaine’s program). More
precisely, the problem is to recover the classical Fontaine’s functors DdR (V ),
Dst (V ) and Dcris (V ) (see for example [26]) from D†rig (V ). The complete solu-
tion was obtained by Berger in [9], [11]. His theory also allowed him to prove
that each de Rham representation is potentially semistable. See also [18] for
introduction and relation to the theory of p-adic differential equations. In this
section, we review some of results of Berger in the case where the ground field
is Q p . Consider the following categories:
● The category MF E of finite dimensional E-vector spaces M equipped with
an exhaustive decreasing filtration (Fili M)i∈Z .
●
ϕ,N
The category MF E of finite dimensional E-vector spaces M equipped
with an exhaustive decreasing filtration (Fili M)i ∈Z , a linear bijective Frobe-
nius map ϕ : M → M and a nilpotent operator (monodromy) N : M →
M such that ϕ N = p ϕ N .
ϕ ϕ,N
● The subcategory MF E of MF E formed by filtered (ϕ, N )-modules M
such that N = 0 on M.
cyc
Let Q p ((t)) be the ring of formal Laurent power series equipped with the
cyc cyc
filtration Fili Q p ((t)) = t i Q p [[t]] and the action of given by
 
g ak t k = g(ak )χ (g)k t k .
k∈Z k∈Z
cyc
The ring R E can not be naturally embedded in E ⊗ Q p ((t)) but for any r > 0
small enough and n  0 there exists a -equivariant embedding i n : R E(r ) →
cyc n
E ⊗ Q p ((t)) which sends X to ζ pn et/ p − 1. Let D be a (ϕ, )-module over
(r)
R E . One can construct for such r a natural -invariant R E -lattice D(r) . Then
 
DdR (D) = E ⊗ Q p ((t)) ⊗in D(r)
cyc

is a finite dimensional E-vector space equipped with a decreasing filtration

 
Fili DdR (D) = E ⊗ Fili Q p ((t)) ⊗in D(r) ,
cyc

which does not depend on the choice of r and n.

Let R E [ X ] denote the ring of power series in variable  X with coefficients
in R E . Extend the actions of ϕ and to R E [ X ] by setting
   
ϕ(X ) τ (X )
ϕ( X ) = p X + log , τ ( X ) =  X + log , τ∈ .
Xp X
 Selmer complexes and p-adic Hodge theory 51

(Note that log(ϕ(X )/ X p ) and log(g(X )/ X ) converge in R E .) Define a mon-

 
1 −1 d
odromy operator N : R E [ X ] → R E [ X ] by N = − 1 − . For
p d X
any (ϕ, )-module D define

Dst (D) = D ⊗R E R E [ X , 1/t] , t = log(1 + X ), (3.12)
Dcris (D) = Dst (D) N =0 = (D[1/t]) . (3.13)

Then Dst (D) is a finite dimensional E-vector space equipped with natural ac-
tions of ϕ and N such that N ϕ = p ϕ N. Moreover, it is equipped with a
canonical exhaustive decreasing filtration induced by the embeddings i n . We
have therefore three functors
ϕ, ϕ, ϕ,N ϕ, ϕ
DdR : MR → MF E , Dst : MR → MF E , Dcris : MR → MF E .
E E E

Theorem 3.1 (B ERGER). Let V be a p-adic representation of G Q p . Then

D∗ (V ) D∗ (V ), ∗ ∈ {dR, st, cris}.

Proof. See [9].

For any (ϕ, )-module over R E one has

dim E Dcris (D)  dim E Dst (D)  dim E DdR (D)  rkR E (D).

One says that D is de Rham (resp. semistable, resp. crystalline) if

dim E DdR (D) = rkR E (D) (resp. dim E Dst (D) = rkR E (D), resp.
ϕ, ϕ,
dim E Dcris (D) = rkR E (D)). Let MR ,st and MR ,cris denote the categories
E E
of semistable and crystalline (ϕ, )-modules, respectively. Berger proved (see
[11]) that the functors
ϕ, ϕ,N ϕ, ϕ
Dst : MR → MF E , Dcris : MR → MF E (3.14)
E ,st E ,cris

are equivalences of ⊗-categories. If D is de Rham, the jumps of the filtration

Fili DdR (D) will be called the Hodge–Tate weights of D.

3.1.4 Families of (ϕ, )-modules

In this section, we review the theory of families of (ϕ, )-modules and its
relation to the families of p-adic representations following [12], [39], [52] and
[40]. Let A be an affinoid algebra over Q p . For each 0  r < 1 the ring
(r )
RQ p
is equipped with a canonical Fréchet topology (see [9]) and we define
(r ) (r) (r)
! Q p RQ . Set R A =
R A = A⊗ ∪ R A . The actions of ϕ and on RQ p
p 0r<1
52 Denis Benois

(r)
extend by linearity to R A . Note that R A is stable under the action of , and
(r ) (r 1/ p ) (r) (r)
that ϕ(R A ) ⊂ R A (but ϕ(R A ) ⊂ R A ).

(r)
Definition. i) Let 0  r < 1. A (ϕ, )-module over R A is a finitely
(r)
generated projective R A -module D(r) equipped with the following
structures:
a) A ϕ-semilinear injective map
(r 1/ p )
ϕ : D(r) → R A ⊗R (r) D(r)
A

such that
 
R (r
1/ p )
ϕ D(r) = R (r
1/ p )
⊗   ⊗R (r) D(r) .
A ϕ R (r)
A
A A

b) A semilinear continuous action of on D(r) which commutes with ϕ.

ii) D is a (ϕ, )-module over R A if D = D(r) ⊗R (r) R A for some (ϕ, )-
A
(r)
module D(r) over R A .

The following proposition shows that, if A = E, this definition is

compatible with the definition given in Subsection 3.1.1.

Proposition 3.2. Let D be a (ϕ, )-module over R E . Then there is 0 

r (D) < 1 such that for any r (D)  r < 1 there is a unique free
(r )
R E -submodule D(r) of D with the following properties:
(s)
a) D(s) = D(r) ⊗R (r ) R E for r  s < 1.
E
b) D = D(r) ⊗R (r) R E .
E
c) The Frobenius ϕ induces isomorphisms
(r 1/ p ) ∼
ϕ ∗ : D(r) ⊗R (r) ,ϕ R E → D(r
1/ p )
, r (D)  r < 1.
E

Proof. This is Theorem 1.3.3 of [11].

Let Rep A (G Q p ) be the category of projective A-modules of finite rank
equipped with an A-linear continuous action of G Q p . The construction of the
functor D†rig can be directly generalized to this case, cf. [12], [39]. This means
the existence of a fully faithful exact functor
ϕ,
D†rig, A : Rep A (G Q p ) → MR A ,
which commutes with base change. More precisely, let X = Spm(A). For
each x ∈ X we denote by mx the maximal ideal of A associated to x and set
 Selmer complexes and p-adic Hodge theory 53

ϕ,
E x = A/mx . If V (resp. D) is an object of Rep A (G Q p ) (resp. of MR ), we set
A
Vx = V ⊗ A E x (resp. Dx = D ⊗ A E x ). Then the diagram

commutes, i.e. D†rig,A (V )x D†rig (Vx ). Note that in general the essential image
of D†rig,Adoes not coincide with the subcategory of étale modules. See [12]
[40], [31] for further discussion.
Let D be a (ϕ, )-module over R A . As in the case A = E, we attach
to D the Fontaine–Herr complexes Cϕ,γ• (D) and consider the associated co-
n
homology groups H ∗ (K n , D). We summarize the main properties of these
cohomology in the theorem below. The key result here is the finiteness of the
rank of H ∗ (K n , D).

Theorem 3.3. Let A be an affinoid algebra over Q p and let D be a (ϕ, )-

module over R A . Then
i) Finiteness. The cohomology groups H i (K n , D) are finitely generated A-
modules. More precisely, for each n  0 the complex Cϕ,γ • (D) is
n
quasi-isomorphic to the complex of projective A-modules of finite rank
concentrated in degrees 0, 1 and 2.
ii) Base change. If f : A → B is a morphism of affinoid algebras, then
• • ∼
Cϕ,γ (D) ⊗L !
R A R B → C ϕ,γn (D⊗R A R B ).
n

In particular, if x ∈ X , then
• ∼ •
Cϕ,γn
(D) ⊗L
R A E x → C ϕ,γn (Dx ).

iii) Euler-Poincaré formula. One has

2  
χ (K n , D) = (−1)i rk A H i (K n , D) = −[K n : Q p ] rkR A (D)
i=0

where the rank is considered as a locally constant function rk A :

Spm(A) → N.
iv) Duality. Formulas (3.10) define a duality
• ∼ •
Cϕ,γn
(D) → RHom A (Cϕ,γ n
(D∗ (χ )), A)[−2].
54 Denis Benois

v) Comparision with Galois cohomology. Let V be a p-adic representation

with coefficients in A. Then there are functorial isomorphisms
∼
H ∗ (K n , V ) → H ∗ (K n , D†rig, A (V ))

Proof. See [40], Theorems 4.4.1, 4.4.2, 4.4.5 and [52], Theorem 2.8.

3.1.5 Derived categories

There is a derived version of the comparision isomorphisms v) of Theorem 3.3,
which is important for the formalism of Iwasawa theory [7], [52]. Let A be an
affinoid algebra over Q p . For any (ϕ, )-module D over R A define
 
• γn −1
Cγn (D) = D −−−→ D ,

• (D) can be defined as the

where the first term is placed in degree 0. Then Cϕ,γ n
total complex
 
• • • ϕ−1 •
Cϕ,γn
(D) = Tot C γn (D) −−→ C γn (D) .

If D and D are two (ϕ, )-modules over R A , the cup product

∪γ : Cγ•n (D ) ⊗ Cγ•n (D ) → Cγ•n (D ⊗ D )

defined by

∪γ (xi ⊗ y j )

if xi ∈ Cγi n (D ), y j ∈ Cγn (D ) and i + j = 0 or 1,
j
xi ⊗ γ n (y j ),
=
0, if i + j  2
gives rise to a map of complexes
•
∪ : Cϕ,γn
(D ) ⊗ Cϕ,γ
•
n
(D ) → Cϕ,γ
•
n
(D ⊗ D ).

Explicitly

∪ ((xi−1 , xi ) ⊗ (y j −1 , y j )) = (xi ∪γ y j −1 + (−1) j xi−1 ∪γ ϕ(y j ), xi ∪γ y j )

(3.15)
if (xi−1 , xi ) ∈ Cϕ,γ (D  ) = C i−1 (D ) ⊕ C i (D ) and (y , j) ∈
i y
n γn γn j −1
Cϕ,γn (D ) = Cγn (D ) ⊕ Cγn (D ). It is easy to see that this is a bilinear
j j−1 j

map, which induces the cup-product (3.11) on cohomology groups.

cyc
Let HQ p = Gal(Q p /Q p ). In [9], Berger constructed a topological ring

B†,r †,r
rig,A equipped with commuting actions of ϕ and G Q p such that Drig, A (V ) ⊂
 Selmer complexes and p-adic Hodge theory 55

V ⊗A B†,r
rig, A for each Galois representation V. Moreover, one has an exact
sequence
ϕ−1
0→ A→ †,r
1/ p
B†,r
rig,A −−→ Brig, A → 0 (3.16)

(see [10], Lemma 1.7). Set B†rig,A = limB†,r . Passing to limits in (3.16), we
−→ rig, A
0<r
obtain the following exact sequence
ϕ−1
0→ A→
B†rig,A −−→ 
B†rig, A → 0. (3.17)

Set G K n = Gal(Q p /Q p (μ pn )). Tensoring (3.17) with a p-adic representation

V, and taking continuous cochains, one obtains an exact sequence
 
0 → C • (G K n , V ) → C • G K n , V ⊗  B†rig, A
ϕ−1
 
−−→ C • G K n , V ⊗  B†rig,A → 0.

Define
   ϕ−1  
K • (K n , V ) = Tot• C • G K n , V ⊗ 
B†rig, A −−→ C • G K n , V ⊗ 
B†rig,A .
(3.18)
Consider the diagram

(3.19)
where the maps αV and βV are constructed as follows. Let αV,γ : C γ•n (V ) →
C • (G Q p , V ⊗ 
B†rig, A ) be the morphism defined by

αV ,γ (x0 ) = x0 , if x0 ∈ C γ0n (V ),
g−1
αV,γ (x 1 )(g) = (x1 ), if x1 ∈ Cγ1n (V ).
γn − 1
Then αV is the map induced by αV,γ by passing to total complexes. The
map βV is defined by
βV : C • (G K n , V ) → K • (K n , V ),
xn → (0, xn ), xn ∈ C n (G K n , V ).
Let R (K n , V ) and R (K n , D†rig (V )) denote the images of complexes
C • (G K n , V ) and Cϕ,γ
• (D† (V )) in the derived category Db (A) of A-modules.
n rig
56 Denis Benois

Proposition 3.4. i) In the diagram (3.19), the maps αV and βV are quasi-
isomorphisms and therefore
∼
R (K n , V ) → R (K n , D†rig (V )) (3.20)

in Db (A).

Proof. See [7], Proposition A.1.

3.1.6 Iwasawa cohomology

Let O E denote the ring of integers of the field of coefficients E. Recall that
cyc
0 denote the Galois group of Q p /Q p (μ p ). We equip the Iwasawa algebra
 = O E [[ 0 ]] with the involution ι :  →  defined by ι(τ ) = τ −1 ,
τ ∈ 0 . Let V be a p-adic representation of G Q p with coefficients in E. Fix
a O E -lattice T of V stable under the action of G Q p . The induced module
Ind K ∞ /Q p (T ) = ⊗O E T is equipped with the diagonal action of G Q p and the
natural structure of a -module given by λ ∗ m = ι(λ)m (see for example [46],
Chapter 8). To fix these structures we will write Ind K ∞ /Q p (T ) = ( ⊗Z p T )ι .
Let R Iw (Q p , T ) denote the class of C • (G Q p , Ind K ∞ /Q p (T )) in the derived
category D b (). Define

i
HIw (Q p , T ) = Ri Iw (Q p , T ), i ∈ N.

From Shapiro’s lemma it follows that there are canonical and functorial
isomorphisms in Db (Z p [G n ]), where G n = Gal(K n /Q p ),

i
HIw (Q p , T ) lim H i (K n , T ),
← −
cores K n /K n−1

Iw (Q p , T ) ⊗ Z p [G n ] R (K n , T ).
L
R

We review the computation of Iwasawa cohomology in terms of (ϕ, )-

modules. It was found by Fontaine (unpublished, but see [15]). Let

ψ : E E† → E E†

denote the operator

1
ψ( f (X )) = Tr † † ( f (X )).
p E E /ϕ(E E )
 Selmer complexes and p-adic Hodge theory 57

More explicitly, the polynomials 1, (1 + X ), . . . , (1 + X ) p−1 form a basis of

E E† over ϕ(E E† ), and one has
⎛ ⎞
p−1
ψ⎝ ϕ( f i )(1 + X )i ⎠ = f 0 .
i=0

In particular, ψ ◦ ϕ = id. Let e1 , . . . , ed be a base of D† (V ) over E E† . Then

ϕ(e1 ), . . . , ϕ(ed ) is again a base of D† (V ), and we define

ψ : D† (V ) → D† (V )
 

d 
d
by ψ ai ϕ(ei ) = ψ(ai )ei . Note that this definition extends to (ϕ, )-
i=1 i=1
modules over the Robba ring.
Consider the complex
• ψ−1
CIw (D† (T )) : D† (T ) −−→ D† (T ) ,

Iw (D
where the first term is placed in degree 0, and denote by R † (T )) the
corresponding object in Db ().

Theorem 3.5. The complexes R Iw (Q p , T ), R Iw (D

† (T )) and R (Q p , D†
(Ind K ∞ /Q p T )) are isomorphic in Db ().

Proof. See [15] and [7] for a derived version.

An analog of previous results for (ϕ, )-modules over the Robba ring was
obtained by Pottharst [53]. We start with some preliminary results about coad-
missible H -modules and the Grothedieck duality. Let B(0, 1) = {z ∈ C p |
|z| p < 1} denote the open unit disk. Define

H = { f (X ) ∈ E[[X ]] | f (X ) converges on B(0, 1)} .

In the context of Iwasawa theory, the ring H appeared in [50]. A natural

description of H is as follows. Write

B(0, 1) = ∪Wn ,
n

where, for every n, Wn = {z ∈ C p | |z| p  p−1/n } is the closed disk of radius

p −1/n < 1, which we consider as an affinoid space. Then
 ∞
&
(Wn , OWn ) = f (X ) = ak X k | |ak | p p −k/n → 0 when k → +∞
k=0
58 Denis Benois

and H = lim (Wn , OWn ). We consider  and  ⊗O E E as subalgebras of

← −
n
H . Set Hn = (Wn , OWn ). It is easy to see that all Hn are euclidian rings.
A coadmissible H -module M is the inverse limit of a system (Mn )n1 where
each Mn is a finitely generated Hn -module and the natural maps Mn ⊗Hn
Hn−1 → Mn−1 are isomorphisms [57]. The structure of admissible modules
is given by the following proposition ([53], Proposition 1.1).

Proposition 3.6. i) A coadmissible H -module is torsion free if and only if it

is a finitely generated H -module.
ii) Let M be a coadmissible torsion H -module. Then
 n
M H /pi i , (3.21)
i ∈I

where (pi )i∈I ⊂ ∪ Spec(Hn ) is a system of maximal ideals such that for
n1
each n there are only finitely many i with pi ∈ Spec(Hn ).

Proof. The proof follows easily from Lazard’s theory [41].

By Proposition 3.6, any coadmissible H -torsion module M can be decom-


posed into direct product M H /pin i and from [41], Théorème 1, it
i∈I
follows that there exists a unique, up to multiplication by a unit of Λ[1/ p],

element f ∈ H such that div( f ) = n i pi .
i∈I

Definition. The characteristic ideal charH M is defined to be the principal

ideal generated by f.

b (H ) denote the category of bounded complexes of H -modules

Let Dcoad
with coadmissible cohomology. Let K denote the field of fractions of H .
Consider the complex
' (
ω = cone K → K /H [−1],

and for any object C • of Dcoad

b (H ) define

D(C • ) = HomH (C • , ω).

Then D : Dcoadb (H ) → D b (H ) is an anti-involution, which can be seen

coad
as the “limit” of the Grothendieck dualizing functors RHom(−, Hn ). For any
coadmissible module M, let D k (M) denote the k-th cohomology group of
D([M]). Then
 Selmer complexes and p-adic Hodge theory 59

 
 
D (M) = HomH (M/Mtor , H ),
0
D 1
H /pin i = pi−n i /H ,
i∈I i∈I

and D k (M) = 0 for all k  2 ( see [53], Section 3.1).

Let D be a (ϕ, )-module over R E . Consider the complexes
• ψ−1
CIw (D) : D −−→ D
• (D), where D = D⊗ ι
and Cϕ,γ Q p H , and denote by R Iw (D) and R (Q p , D)
b (H ). For each
the corresponding objects of the derived category Dcoad
H -module M we set M 0 = M ⊗H E.

Theorem 3.7. Let D be a (ϕ, )-module over R E . Then

i
i) HIw (D) (i = 1, 2) are coadmissible H -modules. Moreover, HIw 1 (D)
tor
and HIw (D) are finite dimensional E-vector spaces.
2

ii) The complexes CIw• (D) and C • (D) are quasi-isomorphic, and therefore
ϕ,γ

R Iw (D) R (Q p , D).
iii) One has an isomorphism
• • ∼
Cϕ,γ (D) ⊗H E → Cϕ,γ (D),
which induces the Hochschild–Serre exact sequences
i+1
0 → HIw
i
(D) → H i (D) → HIw (D) → 0.
b
iv) One has a canonical duality in Dcoad (H )
∗
DR Iw (D) R Iw (D (χ ))ι [2].
v) Let V be a p-adic representation of G Q p . There are canonical and
functorial isomorphisms

R Iw (Q p , T ) ⊗
L
H R (Q p , T ⊗Z p H ι ) R (Q p , D†rig (V )).

Proof. See [53], Theorem 2.6.

3.1.7 Crystalline extensions

In this subsection we consider (ϕ, )-modules over the Robba ring R E . As
usual, the first cohomology group H 1 (Q p , D) can be interpreted in terms of
extensions. Namely, to any cocycle α = (a, b) ∈ Z 1 (Cϕ,γ (D)) we associate
the extension
0 → D → Dα → R E → 0
60 Denis Benois

such that Dα = D ⊕ R E e with ϕ(e) = e + a and γ (e) = e + b. This defines a

canonical isomorphism
H 1 (Q p , D) Ext1 (R E , D).
We say that cl(α) ∈ H 1 (Q p , D) is crystalline if
dim E Dcris (Dα ) = dim E Dcris (D) + 1,
and define
H 1f (Q p , D) = {cl(α) ∈ H 1 (Q p , D) | cl(α) is crystalline }.

It is easy to see that H 1f (Q p , D) is a subspace of H 1 (Q p , D). If D is semistable

(even potentially semistable), the equivalence (3.14) between the category of
semistable (ϕ, )-modules and filtered (ϕ, N )-modules allows us to compute
H 1f (Q p , D) in terms of Dst (D). This gives a canonical exact sequence

0 → H 0 (Q p , D) → Dcris (D) → Dcris (D) ⊕ tD (Q p ) → H 1f (Q p , D) → 0,

(3.22)
where tD (Q p ) = Dst (D)/Fil Dst (D) is the tangent space of D ([5], Proposition
0

1.4.4 and [45], Sections 1.19–1.21). In particular, one has

H 0 (Q p , D) = Fil0 Dst (D)ϕ=1,N =0 ,
dim E H 1f (Q p , D) = dim E tD (Q p ) + dim L H 0 (Q p , D) (3.23)

(see [5], Proposition 1.4.4 and Corollary 1.4.5). Moreover, H 1f (Q p , D) and

H 1f (Q p , D∗ (χ )) are orthogonal complements to each other under duality
(3.11) ([5], Corollary 1.4.10).
We will call exponential map the connecting map in (3.22)
expD : tD (Q p ) → H 1f (Q p , D). (3.24)
Let V be a potentially semistable representation of G Q p . Let

tV (Q p ) = Dst (V )/Fil0 Dst (V )

denote the tangent space of V , and let H 1f (Q p , V ) be the subgroup of
H 1 (Q p , V ) defined by (3.2). In [13], Bloch and Kato constructed a map
expV : tV (Q p ) → H 1f (Q p , V ) (3.25)
using the fundamental exact sequence relating the rings of p-adic periods Bcris
and BdR . It follows from Theorem 3.1 that tV (Q p ) is canonically isomorphic
to tD† (V ) (Q p ) and that
rig

H 1f (Q p , D†rig (V )) H 1f (Q p , V )
 Selmer complexes and p-adic Hodge theory 61

(see [5], Proposition 1.4.2). Moreover, the diagram

commutes ([7], Section 2). Therefore our definition of the exponential map
(3.24) agrees with the original one of Bloch and Kato in [13].

3.1.8 Cohomology of isoclinic modules

The results of this section are proved in [5] (see Proposition 1.5.9 and Section
1.5.10 of op. cit.). Let D be a semistable (ϕ, )-module over R E of rank d.
Assume that Dst (D)ϕ=1 = Dst (D) and that all Hodge–Tate weights of D are 
0. Since N ϕ = pϕ N , this implies that N = 0 on Dst (D), and D is crystalline.
The canonical map D → Dcris (D) is an isomorphism, and therefore

H 0 (Q p , D) Dcris (D) = D

is a E-vector space of dimension d. The Euler–Poincaré characteristic formula

gives

dim E H 1 (Q p , D) = d + dim E H 0 (Q p , D) + dim E H 0 (Q p , D∗ (χ )) = 2d.

On the other hand, dim E H 1f (Q p , D) = d by (3.23). The group H 1 (Q p , D)

has the following explicit description. The map

i D : Dcris (D) ⊕ Dcris (D) → H 1 (Q p , D),

i D (x, y) = cl(−x, log χ (γ ) y)

is an isomorphism. (Remark that the sign −1 and log χ (γ ) are normaliz-

ing factors.) Let i D, f and i D,c denote the restrictions of i D on the first and
second summand, respectively. Then Im(i D, f ) = H 1f (Q p , D), and we set
Hc1 (Q p , D) = Im(i D,c ). Thus we have a canonical decomposition

H 1 (Q p , D) H 1f (Q p , D) ⊕ Hc1 (Q p , D). (3.26)

−1
Now consider the dual module D∗ (χ ). It is crystalline, Dcris (D∗ (χ ))ϕ= p =
Dcris (D∗ (χ )), and the Hodge–Tate weights of D∗ (χ ) are all  0. Let

[ , ]D : Dcris (D∗ (χ )) × Dcris (D) → E

62 Denis Benois

denote the canonical pairing. Define

i D∗ (χ ) : Dcris (D∗ (χ )) ⊕ Dcris (D∗ (χ )) → H 1 (Q p , D∗ (χ ))

by
i D∗ (χ ) (α, β) ∪ i D (x, y) = [β, x]D − [α, y]D .

As before, let i D∗(χ ), f and i D∗ (χ ), c denote the restrictions of i D∗ (χ ) on the

first and second summand respectively. From H 1f (Q p , D∗ (χ )) = H 1f (D)⊥ it
follows that Im(i D∗ (χ ), f ) = H 1f (Q p , D∗ (χ )), and we set Hc1 (Q p , D∗ (χ )) =
Im(i D∗ (χ ), c ). Again we have a decomposition

H 1 (Q p , D∗ (χ )) H 1f (Q p , D∗ (χ )) ⊕ Hc1 (Q p , D∗ (χ )).

Now we will compute the Iwasawa cohomology of some isoclinic modules.

Recall that the well-known computation of universal norms of Q p (1) gives that
1
HIw (Q p , Q p (1)) pZ (3.27)

under the Kummer map. The following proposition generalizes this result to
(ϕ, )-modules. It will be used in the proof of Proposition 3.13 below.
−1
Proposition 3.8. Let D be an isoclinic (ϕ, )-module such that Dcris (D)ϕ= p
= Dcris (D) and Fil0 Dcris (D) = 0. Then
1
HIw (D) 0 = Hc1 (Q p , D).

Proof. For any continuous character δ : Q∗p → E ∗ , let R E (δ) denote the
(ϕ, )-module of rank one R E eδ such that ϕ(eδ ) = δ( p)eδ and γ (eδ ) =
δ(χ (γ ))eδ . In [19], Proposition 3.1, Colmez proved that any (ϕ, )-module
of rank 1 is isomorphic to a unique module of the form R E (δ) for some δ.
Set Dm = R E (|x|x m ). Let em denote the canonical generator of Dm . It is not
difficult to see that an isoclinic (ϕ, )-module D which satisfies the conditions
d
of Proposition 3.8 is isomorphic to the direct sum ⊕ Dm i for some m i  1 ([5],
i=1
Proposition 1.5.8). Therefore we can assume that D = Dm for some m  1.
The E-vector space H 1 (Dm ) is generated by the cohomology classes cl(αm )
and cl(βm ) defined by
   
1 1 1 1
αm = ∂ m−1 + , a em , (1 − ϕ) a = (1 − χ (γ )γ ) + ,
X 2 X 2
   
1 1
βm = ∂ m−1 b, em , (1 − ϕ) = (1 − χ (γ ) γ ) b,
X X
 Selmer complexes and p-adic Hodge theory 63

where ∂ = (1 + X ) d/d X ([19], Sections 2.3–2.5). Moreover, cl(αm ) ∈

H 1f (Dm ) and cl(αm ) ∈ Hc1 (Dm ) ([5], Theorem 1.5.7). For m = 1, the mod-
ule D1 = R E (χ ) is étale and corresponds to the p-adic representation E(1).
Using the formula
p−1  
1 1
= (1 + X )i ϕ ,
X X
i=0
 
it can be checked directly that ψ(1/ X ) = 1/ X. Thus ψ X1 e1 = X1 e1 , and we
ψ=1
proved that 1
X e1 ∈ D1
. Then easy induction (use the identity ψ∂ = p∂ψ)
 
ψ=1
shows that for each m  1 one has ∂ m−1 X1 em ∈ Dm . Since the image of
 
∂ m−1 X1 em under the map HIw 1 (D ) → H 1 (D )
m Iw m 0 ⊂ H (Dm ) is βm , this
1

1 (D )
m 0 ⊂ Hc (Dm ). On the other hand, Theorem 3.7 iii) gives
proves that HIw 1

the Hochschild–Serre exact sequence

0 → HIw
1
(Dm ) 0 → H 1 (Dm ) → HIw
2
(Dm ) 0 → 0.
By the same theorem, 2 (D )
HIw is finite dimensional over E, and therefore
m
2
dim E HIw (Dm ) 0 = dim E HIw
2
(Dm ) 0 = dim E H 2 (Dm ) = 1.
Thus
1
dim E HIw (Dm ) 0 = dim E H 1 (Dm ) − dim E H 2 (Dm ) = 1,
1 (D )
and we proved that dim E HIw = dim E Hc1 (Dm ). This gives the
m 0
proposition.

Remark. It can be shown that the image of p ∈ Q∗p under the Kummer map
is (1 − 1/ p) log(χ )cl(β1 ) (see [4], Proposition 2.1.5).

3.2 The Main Conjecture

3.2.1 Regular submodules
In this section, we apply the theory of (ϕ, )-modules to Iwasawa theory of
p-adic representations. In particular, we propose a conjecture, which can be
seen as a (weak) generalization of both Greenberg’s [28] and Perrin-Riou’s
[51] main conjectures. Fix a prime number p and a finite set S of primes of Q
containing p. For each number field F, we denote by G F,S the Galois group
of the maximal algebraic extension of F unramified outside S ∪ {∞}. For any
topological module M equipped with a continuous action of G F,S , we write
64 Denis Benois

HS∗ (F, M) for the continuous cohomology of G F,S with coefficients in M. Let
V be a p-adic representation of G Q,S with coefficients in a finite extension
E of Q p . Recall that in Subsection 3.0.1 we defined the Bloch–Kato Selmer
group H 1f (Q, V ) and, for each v ∈ S, the subgroup H 1f (Qv , V ) of H 1 (Qv , V ).
The Poitou–Tate exact sequence gives the following exact sequence
 H 1 (Qv , V )
0 → H 1f (Q, V ) → HS1 (Q, V ) → → H 1f (Q, V ∗ (1))∗ →
v∈S
H 1f (Qv , V )

HS2 (Q, V ) → H (Qv , V ) → HS0 (Q, V ∗ (1))∗ → 0
2
(3.28)
v∈S

(see [27], Proposition 2.2.1). Together with the well-known formula for the
Euler characteristic, this implies that

dim E H 1f (Q, V ) − dim E H 1f (Q, V ∗ (1)) − dim E HS0 (Q, V )+

+ dim E HS0 (Q, V ∗ (1)) = dim E tV (Q p ) − d+ (V ), (3.29)

where d± (V ) = dim E V c=±1 and c denotes the complex conjugation.

In the rest of this section, we assume that V satisfies the following
conditions:
C1) HS0 (Q, V ) = HS0 (Q, V ∗ (1)) = 0.
C2) V is semistable at p.
C3) Dcris (V )ϕ=1 = 0.
C4) V satisfies one of the following conditions
a) H 1f (Q, V ∗ (1)) = 0.
b) V is a self dual representation, i.e. it is equipped with a non-degenerate
bilinear form V × V → E(1).
Note that the conditions C1–3) coincide with the conditions from Subsec-
tion 3.0.2. In support of these assumptions we notice that representations which
we have in mind appear as p-adic realizations of pure motives over Q. Let
X/Q be a smooth proper variety having good reduction outside a finite set S of
places containing p. Consider the motives h i (X )(m), where 0  i  2 dim(X )
and m ∈ Z. Let H pi (X ) denote the p-adic étale cohomology of X. The p-adic
realization of h i (X )(m) is V = H pi (X )(m). The action of Gal(Q/Q) fac-
tors through G Q,S . Moreover, the restriction of H pi (X ) on the decomposition
group at p is semistable if X has semistable reduction at p (and potentially
semistable in general) [21], [60]. The Poincaré duality and the hard Lefschetz
theorem give a canonical isomorphism

H pi (X )∗ H pi (X )(i ), (3.30)
 Selmer complexes and p-adic Hodge theory 65

and therefore
V ∗ (1) V (i + 1 − m).

The motive h i (X )(m) is pure of weight w = i − 2m, and replacing, if

necessary, V by V ∗ (1) we can assume that w  −1. By the comparison
isomorphism of the p-adic Hodge theory, [60] Dst (H pi (X )) is isomorphic
−cris (X/Q p ). The weight monodromy
to the log-crystalline cohomology Hlog i

conjecture of Deligne–Jannsen [34] predicts that the absolute values of the

eigenvalues of ϕ acting on Dcris (H pi (X )) = Dst (H pi (X )) N =0 are  i/2, and
therefore Dcris (V )ϕ=1 should be 0 if w  −1. If X has good reduction at p, we
do not need the weight monodromy conjecture, and the nullity of Dcris (V )ϕ=1
follows unconditionally from the result of Katz and Messing [37].
i +1
Assume that w = −1. Then i is odd, m = and the isomorphism (3.30)
2
shows that V is self dual and therefore satisfies C4b).
Assume that w  −2. One expects that

H 1f (Q, V ∗ (1)) = 0 if w  −2. (3.31)

Namely, this follows from conjectural properties of the category MM of

mixed motives over Q. Consider the Yoneda groups

H n (Q, h i (X )(m)) = ExtnMM (Q(0), h i (X )(m)), n = 0, 1,

and denote by H 1f (Q, h i (X )(m)) the subgroup of extensions having “good

reduction”. In [13], Bloch and Kato conjectured that the p-adic realization
functor induces an isomorphism
∼
R p : H 1f (Q, h i (X )(m)) → H 1f (Q, H pi (X )(m)), ∀m ∈ Z (3.32)

(see also [34] and [24]). In particular, we should have an isomorphism

∼
H 1f (Q, h i (X )(i + 1 − m)) → H 1f (Q, V ∗ (1)).

On the other hand, from the semisimplicity of the category of pure motives
[35] it follows that H 1 (Q, M) = 0 for any pure motive M of weight  0.
In particular, H 1 (Q, h i (X )(i + 1 − m)) should vanish if −i − 2 + 2m  0.
Together with (3.32) this imples C4b). To sum up, from the motivic point of
view the conditions C4a) and C4b) correspond to the weight  −2 and the
weight −1 cases respectively. We consider these two cases separately below.
The weight  −2 case. In this subsection, we assume that V satisfies the
conditions C1–3) and C4a). From C3) it follows that the exponential map
tV (Q p ) → H 1f (Q p , V ) is an isomorphism, and we denote by logV its inverse.
66 Denis Benois

Compositing logV with the localization map H 1f (Q, V ) → H 1f (Q p , V ), we

obtain a map
r V : H 1f (Q, V ) → tV (Q p ).
If V = H pi (X )(m) with m  i/2 + 1, this map is related to the syntomic
regulator Rsyn via the commutaive diagram

( [47], see also [59]). Our next assumption reflects the hope that the syntomic
regulator is an injective map.
C5a) The localization map
loc p : H 1f (Q, V ) → H 1f (Q p , V )
is injective.
In support of this assumption consider the case where V = H pi (X )(m) with
w = i − 2m  −2. The weight monodromy conjecture of Deligne predicts
that
H 0 (Qv , H pi (X )(m)) = 0, if v = p and w = i − 2m  −2. (3.33)
Since dim E H 1f (Qv , V ) = dim E H 0 (Qv , V ), this implies that H 1f (Qv , V )
should vanish for v = p. On the other hand, in [34], Jannsen conjectured
that
HS2 (Q, H pi (X )(m)) = 0, if i + 1 < m or i + 1 > 2m.
In particular, HS2 (Q, V ∗ (1)) should vanish. From the Poitou–Tate exact se-
quence we obtain an exact sequence

HS2 (Q, V ∗ (1)) → H 1f (Q, V ) → H 1f (Qv , V ).
v∈S
Therefore the condition C5a) follows from Jannsen’s conjecture together with
the weight monodromy conjecture. Also note that if w < −2, then C5a)
follows directly from (3.33) (see [34], Lemma 4 and Theorem 3).
We introduce the notion of regular submodule, which first appeared in
Perrin-Riou’s book [51] in the context of crystalline representations (see
also [5]).

Definition (P ERRIN -R IOU). Assume that V is a p-adic representation which

satisfies the conditions C1–3), C4a) and C5a).
 Selmer complexes and p-adic Hodge theory 67

i) A (ϕ, N )-submodule D of Dst (V ) is regular if D ∩ Fil0 Dst (V ) = 0 and the

map
r V ,D : H 1f (Q, V ) → Dst (V )/(Fil0 Dst (V ) + D),
induced by r V , is an isomorphism.
ii) A (ϕ, N )-submodule D of Dst (V ∗ (1)) is regular if

D + Fil0 Dst (V ∗ (1)) = Dst (V ∗ (1))

and the map
D ∩ Fil0 Dst (V ∗ (1)) → H 1f (Q, V )∗ ,
induced by the dual map r V∗ : Fil0 Dst (V ∗ (1)) → H 1f (V )∗ , is an
isomorphism.

Remark. Assume that H 1f (Q, V ) = H 1f (Q, V ∗ (1)) = 0. Then D is regular

if the canonical projection D → tV (Q p ) is an isomorphism of vector spaces,
and our definition agrees with the definition given in [5].

The weight −1 case. In this subsection we assume that V satisfies the

conditions C1–3) and C4b). Then formula (3.29) gives
dim E tV (Q p ) = d+ (V ).

Definition (P ERRIN -R IOU [51]). Assume that V is a p-adic representation

which satisfies the conditions C1–3) and C4b). A (ϕ, N )-submodule D of
Dst (V ) is regular if the canonical map D → tV (Q p ) is an isomorphism.

In the both cases C4a) and C4b) from (3.29) it follows immediately that for
a regular submodule D one has
dim E D = d+ (V ). (3.34)
Remark. Our definition of a regular module in the weight −1 case is slightly
different from Perrin-Riou’s definition. If we assume the non-degeneracy of
p-adic height pairings  , V,D (see Section 3.4), then our regular modules are
regular also in the sense of [51], but the converse in not true.

3.2.2 Iwasawa cohomology

We keep previous notation and conventions. Let Qcyc = ∪ Q(μ pn ) and
n1
= Gal(Qcyc /Q). Then =  × 0 where  = Gal(Q(μ p )/Q) and
= cyc /Q(μ )). Set F = Q(μ 
0 Gal(Q p n pn+1 ) and F∞ = ∪ Fn . The Galois
n1
68 Denis Benois

group Gal(F∞ /F) is canonically isomorphic to 0 . Let  = O E [[ 0 ]] denote

the Iwasawa algebra of 0 .
Let V be a p-adic representation of G Q,S . Fix a O E -lattice T of V sta-
ble under the action of G Q,S and consider the Iwasawa cohomology groups
∗ (Q, T ) and H ∗ (Q , T ) defined in Subsection 3.0.3. The main properties
HIw,S Iw v
of these groups are summarized below (see [49]).
i
i) HIw,S (Q, T ) = 0 and HIw
i
(Qv , T ) = 0 if i = 1, 2;
ii) If v = p, then HIw (Qv , T ) are finitely generated -torsion modules. In
i

particular,
Iv ι
1
HIw (Qv , T ) H 1 (Qur
v /Qv , ( ⊗O E T ) ). (3.35)
iii) If v = p, then 2 (Q , T )
HIw is a finitely generated -torsion module.
p
Moreover
 
1 1
rk HIw (Q p , T ) = d, HIw (Q p , T )tor H 0 (Q p (ζ p ∞ ) , T ).
2 (Q , T )
Note that, by local duality, HIw p H 0 (F∞, p , V ∗ (1)/ T ∗ (1)),
where F∞, p is the completion of F∞ at p.
The -module structure of HIw,S
i (Q, T ) depends heavily on the following
conjecture formulated by Greenberg [28].

Weak Leopoldt Conjecture. Let V be a p-adic representation of G Q,S which

is potentially semistable at p. Then
HS2 (F∞ , V /T ) = 0.

We have the following result, proved in [51], Proposition 1.3.2.

Proposition 3.9. Assume that the weak Leopoldt conjecture holds for V. Then
2 (Q, T ) is -torsion, and
HIw,S
1
rk HIw,S (Q, T ) = d− (V ).

Passing to projective limits in the Poitou–Tate exact sequence for V , we

obtain an exact sequence

0 → H 2 (F∞ , V ∗ (1)/T ∗ (1))∧ → HIw,S
1
(Q, T ) → 1
HIw (Qv , T ) →
v∈S

→ HS1 (F∞ , V ∗ (1)/ T ∗ (1))∧ → 2
HIw,S (Q, T ) → 2
HIw (Qv , T ) →
v∈S
→ HS0 (F∞ , V ∗ (1)/T ∗ (1))∧ → 0, (3.36)
 Selmer complexes and p-adic Hodge theory 69

where (−)∧ denotes the Pontryagin dual. Therefore, if the weak Leopoldt
conjecture holds for V ∗ (1), we have an exact sequence

0 → HIw,S
1
(Q, T ) → 1
HIw (Qv , T ) → HS1 (F∞ , V ∗ (1)/T ∗ (1))∧ →
v∈S

2
HIw,S (Q, T ) → 2
HIw (Qv , T ) → HS0 (F∞ , V ∗ (1)/T ∗ (1))∧ → 0.
v∈S

3.2.3 The complexes R Iw (V, D) and R (V, D)

In this section, we construct the Selmer complexes R Iw (V , D) and
R (V, D), which play a central role in our approach to the Iwasawa theory.
Nekovář’s book [46] provides a detailed study of Selmer complexes associated
to Greenberg’s local conditions. For the purposes of this paper one should work
with local conditions associated to general (ϕ, )-submodules of D†rig (V ). In
this context, the general formalism of Selmer complexes was developed by
Pottharst in [52], and we refer to op. cit. and [53] for further information and
details.
Let V be a p-adic representation of the Galois group G Q,S with coef-
ficients in E. We fix a G Q,S -stable lattice T of V. Recall that we denote
by C • (G Q,S , T ⊗ Λι ) and C • (G Qv , T ⊗ Λι ) the complexes of continuous
cochaines of G Q,S and G Qv respectively with coefficients in T ⊗ Λι and by
R Iw,S (Q, T ) and R Iw (Qv , T ) these complexes viewed as objects of D b (Λ).
Recall that Shapiro’s lemma gives canonical isomorphisms

Ri Iw,S (Q, T )
i
HIw,S (Q, T ), Ri Iw (Q p , T )
i
HIw (Q p , T ).

The derived version of the Poitou–Tate exact sequence for Iwasawa coho-
mology reads

R Iw,S (Q, T ) → R Iw (Qv , T ) → R (F∞ , V ∗ (1)/T ∗ (1))∧ [−2]
v∈S

(see [46], Proposition 8.4.22). Passing to cohomology in this exact triangle, we

obtain (3.36).
Let V = V ⊗Q p H ι . Then

S (Q, V ) Iw,S (Q, T ) ⊗Λ H ,

L
R R
R (Qv , V ) R Iw (Qv , T ) ⊗Λ H .
L

Assume that V is potentially semistable. To each (ϕ, N )-submodule D of

Dst (V ) we associate a complex R Iw (V, D), which can be seen as a direct gen-
eralization of Selmer complexes determined by Greenberg’s local conditions
70 Denis Benois

• (V, D)
and studied in Chapters 7–8 of [46]. We define local conditions UIw,v
(v ∈ S). For v = p we set
•
UIw,v (V, D) = R Iw, f (Qv , T ) ⊗ H,
where
 
Frv −1
R Iw, f (Qv , T ) = T Iv ⊗ ι −−−→ T Iv ⊗ ι .

Here Iv denotes the inertia group at v, Frv is the geometric Frobenius, and the
f (Qv , T ) denote the cohomology of
i
first term is placed in degree 0. Let HIw,
f (Qv , T ) = 0 for i = 1, and
i
this complex. From (3.35) it follows that HIw,
HIw, f (Qv , T ) = HIw (Qv , T ) is Λ-torsion.
1 1

Now we define the local condition at p. Let D be the (ϕ, )-submodule of

D†rig (V ) associated to D by Theorem 3.1. Set
 
•  ψ−1 
UIw, p (V, D) = R Iw (D) = D −−→ D .

From Theorem 3.7 it follows that we have a canonical map UIw, • (V , D) →

p
R Iw (Q p , T ) ⊗L
Λ H . This gives the following diagram in Dcoad (H )
b

and we denote by R Iw (V ,
D) the associated Selmer complex

 
•
R Iw (V, D) = cone R Iw,S (Q, T ) ⊗LH UIw,v (V, D) →
v∈S


Iw (Qv , T ) ⊗ H [1]. (3.37)
L
R
v∈S

Now we define the complex R (V , D). For v = p, let

 
Frv −1
R f (Qv , V ) = V Iv −−−→ V Iv ,

where the first term is placed in degree 0. It is clear that R0 f (Qv , V ) =

H 0 (Qv , V ) and R1 f (Qv , V ) coincides with the group H 1f (Qv , V ) defined
by (3.2). Define
 Selmer complexes and p-adic Hodge theory 71


R f (Qv , V ), if v = p,
Uv• (V , D) = (3.38)
R (Q p , D), if v = p.
From Theorem 3.7 it follows that
Uv• (V , D) = UIw,v
•
(V, D) ⊗L
H E, v ∈ S.
Consider the Selmer complex R (V, D) associated to this data
 
 
R (V, D) = cone R S (Q, T ) Uv• (V, D) → R (Qv , V ) [1].
v∈S v∈S
(3.39)
∗ (V , D) and H ∗ (V, D) the cohomology of R
We denote by HIw (V , D) and
Iw
R (V, D) respectively. The main properties of our Selmer complexes are
summarized below.

Proposition 3.10 (P OTTHARST). i) The complex R Iw (V, D) has cohomol-

ogy concentrated in degrees [1, 3] consisting of coadmissible H -modules.
ii) One has
1
rkH HIw (V, D) = rkH HIw 2
(V, D).
 ∗
Moreover, HIw 3
(V, D) = T ∗ (1) HQ,S ⊗Λ H , where HF,S =
(S)
Gal(Q /F∞ ).
iii) The complex R (V , D) has cohomology concentrated in degrees [0, 3]
consisting of finite dimensional E-vector spaces, and
R Iw (V, D) ⊗L
H E R (V , D).
In particular, we have canonical exact sequences
i +1
0 → HIw
i
(V , D) 0 → H i (V, D) → HIw (V, D) 0 → 0, i ∈ N.
(3.40)
iv) There are canonical dualities
Hom E (R (V , D), E) R (V ∗ (1), D ⊥ )[3],
DR Iw (V , D)ι R Iw (V
∗
(1), D ⊥ )[3].
In particular, we have canonical exact sequences
4−i
0 → D 1 HIw (V, D) → HIw
i
(V ∗ (1), D ⊥ )ι → D 0 HIw
3−i
(V, D) → 0.
(3.41)
v) Assume that D satisfies the following conditions:
Fil0 D = 0 and Fil0 (Dst (V )/D) = Dst (V )/D;
72 Denis Benois

(Dst (V )/D)ϕ=1,N =0 = 0 and D/(N D + ( pϕ − 1)D) = 0.

Then H 1 (Q p , D) = H 1f (Q p , V ) and H 1 (V, D) = H 1f (Q, V ).

Proof. See [53], Theorem 4.1 and [54], Proposition 3.7.

3.2.4 The Main Conjecture

In this subsection, we use the formalism of Selmer complexes to formulate a
version of the Main Conjecture of Iwasawa theory. Let M be a pure motive
over Q of weight w  −1. Since the category of pure motives is semisimple,
we can assume that M is simple and = Q(m), m ∈ Z. Let V denote the p-adic
realization of M. Assume that V is semistable. One expects (see [51]) that for
each regular (ϕ, N )-submodule D of Dst (V ) there exists a p-adic L-function
of the form
L p (M, D, s) = f D (χ (γ0 )s − 1), fD ∈ H
interpolating algebraic parts of special values of the complex L-function
L(M, s) (see Section 0.2). Assume that V satisfies the conditions C1–4) of
Section 2.1. We propose the following conjecture.

Main Conjecture. Let M/Q be a pure simple motive of weight  −1. Assume
that M = Q(m), m ∈ Z. Let D be a regular submodule of Dst (V ). Then
1 (V, D) = 0,
i) HIw
2 (V, D) is H -torsion and
ii) HIw
charH (HIw
2
(V, D)) = ( f D ).

Remarks. 1) By Proposition 3.10, the nullity of HIw 1 (V , D) implies that

HIw (V, D) is H -torsion.

2

2) The condition M = Q(m) implies that (V ∗ (1)) HQ,S = 0, and therefore

HIw3 (V, D) should vanish by Proposition 3.10 ii). To sum up, we expect that

under our assumptions the cohomology of R Iw (V , D) is concentrated in

degree 2.
3) Let D ⊥ denote the dual regular submodule. One can easily formulate the
Main Conjecture for the dual pair (M ∗ (1), D ⊥ ). Assume that HIw
1 (V, D) =

0. From (3.41) it follows that

2
HIw (V ∗ (1), D ⊥ )ι 2
HIw (V , D).
This isomorphism reflects the conjectural functional equation relating
L p (M ∗ (1), D ⊥ , s) and L p (M, D, s).
 Selmer complexes and p-adic Hodge theory 73

4) Assume that V is ordinary at p, i.e. that the restriction of V on G Q p is

equipped with an increasing exhaustive filtration F i V such that for all
i ∈ Z the inertia group I p acts on gri (V ) = F i V /F i +1 V by χ i . In [28],
Greenberg works with the Selmer group defined as follows. For each place
w of F∞ including the unique place above p, fix a decomposition group
Hw in H F,S = Gal(Q(S) /F∞ ) and denote by Iw its inertia subgroup. Set
A = V /T and F 1 A = F 1 V /F 1 T. Define

H 1 (Hw /Iw , A Iw ), if w = p,
1
HGr (F∞,w , A) = 
ker H 1 (F∞, p , A) → H 1 (I p , A/F 1 A) if w = p.
(3.42)
The Greenberg Selmer group is defined as follows
 
 H 1 (F∞,w , A)
S(F∞ , V /T ) = ker HS1 (F∞ , A) → .
H 1 (F
w∈S Gr ∞,w
, A)

It is well-known that each semistable representation is ordinary, and we

set D = Dst (F 1 V ). Then R Iw (D) = R Iw (Q p , F 1 V ) ⊗L [1/ p] H , and
directly from definition (3.37) it follows that the complex R Iw (V , D) is
) f,Iw (F∞ /Q, T ) ⊗L )
isomorphic to R [1/ p] H , where R f,Iw (F∞ /Q, T )
denotes Nekovář’s Selmer complex associated to the local condition given
by F 1 V. In [46], Chapter 9, it is proved that under some technical condi-
tions the characteristic ideal of the Pontryagin dual S(F∞ , V ∗ (1)/T ∗ (1))∧
of S(F∞ , V ∗ (1)/T ∗ (1)) coincides with the characteristic ideal of the sec-
ond cohomology group H 2 (T ) of R) f,Iw (F∞ /Q, T ). This allows us to
f
compare our conjecture to the Main Conjecture from [28].

3.3 Local structure of p-adic representations

3.3.1 Filtration associated to a regular submodule
Let V be a semistable p-adic representation of G Q p . In this subsection, we
assume that V satisfies the following condition:
S) The endomorphism ϕ : Dst (V ) → Dst (V ) is semisimple at 1 and p −1 .

Note that if X/Q p is a smooth proper variety having semistable reduction, then
the semisimplicity of the Frobenius action on the log-crystalline cohomology
−cris (X/Q p ) is a well-known open question.
i
Hlog
74 Denis Benois

Let D be a (ϕ, N )-submodule of Dst (V ) such that D ∩ Fil0 Dst (V ) = {0}.

We associate to D an increasing filtration (Di )−2i2 on Dst (V ) setting
⎧
⎪
⎪ 0 if i = −2,
⎪
⎪
⎪
⎪ −1 −1 ϕ=1
⎨(1 − p ϕ ) D + N (D
⎪ ) if i = −1,
Di = D if i = 0,
⎪
⎪
⎪
⎪ D + Dst (V ) ϕ=1 −1
∩ N (D ϕ= p −1
) if i = 1,
⎪
⎪
⎪
⎩D (V )
st if i = 2.

It can be easily proved (see[5], Lemma 2.1.5) that (Di )2i=−2 is the unique
filtration on Dst (V ) such that
D1) D−2 = 0, D0 = D and D2 = Dst (V );
D2) (Dst (V )/D1 )ϕ=1,N =0 = 0 and D−1 = (1 − p −1 ϕ −1 )D−1 + N (D−1 );
−1
D3) (D0 /D−1 )ϕ= p = D0 /D−1 and (D1 /D0 )ϕ=1 = D1 /D0 .
In addition, for the dual regular submodule D ⊥ , one has
Di⊥ = HomQ p (Dst (V )/D−i , Dst (Q p (1))).
Let (Di )i=−2
2 be the filtration associated to D. By (3.14), it induces a filtration
†
of Drig (V ), which we will denote by (Di )i2=2 . Define

W = D1 /D−1 , W0 = gr0 D†rig (V ), W1 = gr0 D†rig (V ).

Then we have an exact sequence
0 → W0 → W → W1 → 0, (3.43)
where, by D1-3),
−1
Dst (W0 ) = Dst (W0 )ϕ= p , Fil0 Dst (W0 ) = 0, (3.44)
ϕ=1
Dst (W1 ) = Dst (W1 ) . (3.45)

Proposition 3.11. Let D be a regular submodule of Dst (V ). Then

i) There are canonical inclusions
H 1 (D−1 ) ⊂ H 1f (D) ⊂ H 1 (D1 ) ⊂ H 1 (Q p , V );
ii) H 1f (D−1 ) = H 1 (D−1 );
iii) The sequences
0 → H 1 (D−1 ) → H 1 (D1 ) → H 1 (W) → 0,
0 → H 1 (D−1 ) → H 1f (D1 ) → H 1f (W) → 0
are exact.
 Selmer complexes and p-adic Hodge theory 75

Proof. 1) By D2) one has

Hom(D−1 , Dst (Q p (1)))ϕ=1,N =0 = 0,
and by the Poincaré duality
H 2 (D−1 ) H 0 (D∗−1 (χ ))∗ = 0. (3.46)
Now (3.8) and (3.23) imply that H 1f (D−1 ) = H 1 (D−1 ), and ii) is proved.
2) The exact sequence
0 → D1 → D†rig (V ) → gr2 D†rig (V ) → 0
gives an exact sequence
 
0 → H 0 gr2 D†rig (V ) → H 1 (D1 ) → H 1 (Q p , V ).

From D2) it follows that

 
H 0 gr2 D†rig (V ) = Fil0 (Dst (V )/D1 )ϕ=1,N =0 = 0,

and the injectivity of H 1 (D1 ) → H 1 (Q p , V ) is proved. Since

Fil0 Dcris (D) = 0 and Dcris (D)ϕ=1 = Dcris (V )ϕ=1 = 0 by C3), the exact
sequence (3.22) induces a commutative diagram

where the exponential maps and the left vertical map are isomorphisms.
Thus H 1f (D) → H 1f (Q p , V ) is an injection. The same argument, together
with ii), proves that H 1 (D−1 ) ⊂ H 1f (D). This implies i).
3) Since the sequence
0 → H 1 (D−1 ) → H 1 (D1 ) → H 1 (W) → H 2 (D−1 )
is exact, iii) follows from ii) and (3.46).

Assume now that the canonical projection Dst (D) → tV (Q p ) is an

isomorphism, i.e. that
Dst (V ) = D ⊕ Fil0 Dst (V ) (3.47)
as E-vector spaces. In this case, the structure of W can be completely
determined if we make the following additional assumption (see [29], [5]):
76 Denis Benois

U) The (ϕ, )-module D†rig (V ) has no saturated crystalline subquotient U

sitting in a non-split exact sequence of the form
0 → R E (|x|x k ) → U → R E (x m ) → 0. (3.48)
Note that if k  m, then H 1f (R E (|x|x k−m ))
= 0, and there is no non-
trivial crystalline extension (3.48). If k > m, it follows from (3.23) that
dim E H 1f (R E (|x|x k−m )) = 1, and therefore there exists a unique (up to
isomorphism) crystalline (ϕ, )-module U of the form (3.48).

Proposition 3.12. Let D be a (ϕ, N )-submodule of Dst (V ) which satisfies

(3.47). Assume that the condition U) holds. Then

i) There exists a unique decomposition

W A0 ⊕ A1 ⊕ M, (3.49)
where A0 and A1 are direct summands of W0 and W1 of ranks
dim E H 0 (W∗ (χ )) and dim E H 0 (W) respectively. Moreover, M is inserted
in an exact sequence
f g
0 → M0 −
→M−
→ M1 → 0,

where W0 A0 ⊕ M0 , W1 A1 ⊕ M1 , and rk(M0 ) = rk(M1 ).

ii) One has

dim E H 1 (M) = 2e, dim E H 1f (M) = e, where e = rk(M0 ) = rk(M1 ).

iii) Consider the exact sequence

δ0 f1 g1 δ1
0 → H 0 (M1 ) −
→ H 1 (M0 ) −
→ H 1 (M) −
→ H 1 (M1 ) −
→ H 2 (M0 ) → 0.

Then H 1 (M0 ) Im(δ0 ) ⊕ H 1f (M0 ), Im( f 1 ) = H 1f (M), and H 1 (M1 )

Im(g1 ) ⊕ H 1f (M1 ).

Proof. See [5], Proposition 2.1.7 and Lemma 2.1.8.

3.3.2 The weight  −2 case

We return to the study of the cohomology of p-adic representations. Let V be
the p-adic realization of a pure simple motive M = Q(m). In this subsection,
we assume that V satisfies the conditions C1–4a) of Subsection 3.2.1 and the
conditions S) and U) of Subsection 3.3.1. Suppose D is a regular submodule of
Dst (V ) and D is the associated (ϕ, )-submodule of D†rig (V ). In the remainder
of this paper, we write H 1 (D) instead H 1 (Q p , D).
 Selmer complexes and p-adic Hodge theory 77

Define
⎛ ⎞
 H 1 (Qv , V )
H 1f,{ p} (Q, V ) = ker ⎝ HS1 (Q, V ) → ⎠.
H 1 (Qv , V )
v∈S−{ p}

Then (3.28) gives an exact sequence

H 1 (Q p , V )
0 → H 1f (Q, V ) → H 1f,{ p} (Q, V ) → → 0. (3.50)
H 1f (Q p , V )
Using the fact that D is regular, we obtain the following decomposition
H 1f (Q p , V ) = H 1f (Q, V ) ⊕ H 1f (D).
Therefore the restriction map induces an isomorphism
H 1 (Q p , V )
H 1f,{ p} (Q, V ) .
H 1f (D)

Let H D1 (Q, V ) be the inverse image of H 1 (D1 )/H 1f (D) under this isomor-
phism. By Proposition 5 iii), we have an injection
κ D : H D1 (Q, V ) → H 1 (W) (3.51)

Since, by D3), Dcris (W0 ) ϕ= p−1= Dcris (W0 ) and Fil0 Dcris (W0 ) = 0, the
results of Section 3.1.8 give a canonical decomposition
H 1 (W0 ) = H 1f (W0 ) ⊕ Hc1 (W0 ). (3.52)

Proposition 3.13. Assume that the weak Leopoldt conjecture holds for V ∗ (1)
and that
a) H 0 (W) = 0;
b) The map Hc1 (W0 ) → H 1 (W) is injective;
c) One has
Im(κ D ) ∩ Hc1 (W0 ) = {0} in H 1 (W). (3.53)
1 (V, D) = 0, and therefore H 2 (V, D) is H -torsion.
Then HIw Iw

Proof. The reader can compare this proof to the proof of Theorem 5.1.3 of
[7]. Recall that 0 = Gal(Qcyc /Q(μ p )), and that for each H-module M we
set M 0 = M ⊗H E. We will use the following elementary lemma.
Lemma 3.14. Let A and B be two submodules of a finitely-generated free H-
module M. Assume that the natural maps A 0 → M 0 and B 0 → M 0 are
both injective. Then A 0 ∩ B 0 = {0} implies that A ∩ B = {0}.
78 Denis Benois

Proof. This is Lemma 5.1.4.1 of [7].

We prove Proposition 3.13. Because the weak Leopoldt conjecture holds

for V ∗ (1), the group HIw,S
1 (Q, T ) injects into ⊕
v∈S HIw (Qv , T ). Since
1
1 (Q , T ) = H 1 (Q , T ) for v = p, by the definition of the complex
HIw, f v Iw v
R Iw (V, D) one has
 
1
HIw (V, D) = HIw,S
1
(Q, T ) ⊗Λ H ∩ HIw 1
(D) in HIw 1
(Q p , T ) ⊗Λ H .

The Λ-torsion part of HIw,S

1 (Q, T ) is isomorphic to T HQ,S = 0, and therefore

A = HIw,S
1 (Q, T )⊗ H is a free H -module. Moreover, A
Λ 0 ⊂ H f,{ p} (Q, V )
1

Set B = HIw 1 (D)/H 1 (D)

H −tor . The H -torsion part of HIw (D) is contained
1
Iw     0
in T HQ p ⊗Λ H , and dim E V HQ p = dim E V HQ p = 0, and therefore
0
B = HIw
0
1 (D) .
0
Prove that B 0 = HIw
1 (D)
0 injects into H 1 (D1 ) ⊂ H 1 (Q p , V ). The exact
sequence
0 → D → D1 → W1 → 0

gives rise to an exact sequence 0 → H 0 (W1 ) → H 1 (D) → H 1 (D1 ), and

therefore it is sufficient to show that

H 0 (W1 ) ∩ HIw
1
(D) 0 =0 in H 1 (D).

Set Z = H 0 (W1 ) ∩ HIw1 (D) . Let f : H 1 (D) → H 1 (W ) denote the map

0 0
induced by the natural projection D → W0 . Since H 0 (W) = 0, the exact
sequence 0 → W0 → W → W1 → 0 induces an injection H 0 (W1 ) →
H 1 (W0 ). Thus, Z ∩ ker( f ) = {0}. On the other hand, by Proposition 3.8,

f (HIw
1
(D) 0 ) ⊂ HIw
1
(W0 ) = Hc1 (W0 ).

From the exact sequence

0 → H 0 (W1 ) → H 1 (W0 ) → H 1 (W)

and the injectivity of Hc1 (W0 ) → H 1 (W) we obtain that

Hc1 (W0 ) ∩ H 0 (W1 ) = {0},

and therefore f (Z ) = 0. This proves that Z = {0}, and the map HIw
1 (D)
0 →
H (D1 ) is injective.
1

To complete the proof of Proposition it will be enough, by Lemma 3.14, to

show that

H 1f,{ p} (Q, V ) ∩ HIw

1
(D) 0 =0 in H 1 (Q p , V ). (3.54)
 Selmer complexes and p-adic Hodge theory 79

1 (D)
Since HIw ⊂ H 1 (D1 ) and the map κ D is injective, (3.54) is equivalent to
0

H D1 (Q, V ) ∩ Hc1 (W0 ) = 0,

and the proposition is proved.

We want to discuss how the above result is related to the phenomenon of
trivial zeros of p-adic L-functions studied in [29], [5], [7]. We consider two
cases.
A) Assume in addition that Dst (V )ϕ=1 = 0. Then D1 = D and W = W0 .
By Proposition 3.11 iii), one has
H D1 (Q, V ) H 1 (D)/H 1f (D) H 1 (W0 )/H 1f (W0 ).

Consider the commutative diagram

where p D, f and p D,c are projections given by (3.52) and ρ D, f and ρ D,c are
defined as the unique maps making this diagram commute.

Definition ( see [7]). Let V be a p-adic representation which satisfies the con-
ditions C1–3), C4a), C5a) from Subsection 3.2.1 and the conditions S) and
U) from Subsection 3.3.1. Assume that Dst (V )ϕ=1 = 0. Let D be a regular
submodule of Dst (V ). The determinant
 
(V, D) = det ρ D, f ◦ ρ −1
D,c | Dcris (W0 )

will be called the -invariant associated to V and D.

B) Assume that
H 1f (Q, V ) = H 1f (Q, V ∗ (1)) = 0, (3.55)

and that, in addition, V satisfies:

M) The condition U) of Section 3.3.1 holds and in the decomposition (3.49)
A0 = A1 = 0.
80 Denis Benois

Note that the typical example we have in mind is W f (k), where W f is the p-
adic representation associated to a split multiplicative newform f of weight 2k
on 0 (N ) with p | N .
From Proposition 3.12 one has H 1f (M0 ) = H 1f (M) and H 1 (M)/H 1f (M0 )
H 1 (M1 )/H 1f (M1 ). Thus, H D1 (Q, V ) H 1 (M1 )/H 1f (M1 ) is a E-vector space
of dimension e = rank(M0 ) = rank(M1 ), and one has a commutative diagram

(3.56)
where the map κ̄ D is injective. As H 0 (M) = H 2 (M) = 0, the upper row
is exact. This implies that Im(g1 ) is a E-vector space of dimension e. Thus,
Im(κ̄ D ) = Im(g1 ), and again one has a commutative diagram

Definition (see[5]). Let V be a p-adic representation which satisfies the con-

ditions C1–3) and (3.55). Let D be a regular submodule of Dst (V ). Assume
that the condition M) holds. The determinant
 
(V , D) = det ρ D, f ◦ ρ −1
D,c | Dcris (M1 )

will be called the -invariant associated to V and D.

In particular, in this case the -invariant is local, i.e. depends only on the
restriction of the representation V on the decomposition group at p.

Remark. In [5], the -invariant is defined in a slightly more general situation,

where only W0 vanishes. We do not include it here to avoid additional technical
complications in the formulation of our results.
The following conjecture was formulated in [5] and [7].
 Selmer complexes and p-adic Hodge theory 81

Conjecture (E XTRA ZERO CONJECTURE ). Let V be a p-adic representation

which satisfies the conditions C1–4a). Let D be a regular subspace of Dst (V ),
and let e = rank(W0 ). Then, in both cases A) and B), the p-adic L-function
L p (V, D, s) has a zero of order e at s = 0 and

L p (V , D, 0) L(M, 0)
lim = (V , D) E + (V , D)  p (M, D) ,
s→0 se ∞ (M)
where E + (V, D) is obtained from E(V, D) by excluding zero factors.

Recall (see Section 3.0.2) that  p (M, D) denote the determinant of the reg-
ulator map r V,D . Note that  p (M, D) = 1 if H 1f (Q, V ) = H 1f (Q, V ∗ (1)) = 0.
We refer to [5], [6] [7] for precise formulation of this conjecture and a survey
of known cases. Note that the non-vanishing of (V , D) is a difficult open
problem which is solved only for Dirichlet motives Q(η) [22] and for elliptic
curves [1]. The following result shows that it is closely related to the expected
vanishing of HIw1 (V, D).

Proposition 3.15. Let V be a p-adic representation which satisfies the condi-

tions C1-3). Assume that the weak Leopoldt conjecture holds for V ∗ (1). Then,
in the both cases A) and B) above, the non-vanishing of (V, D) implies that
HIw1 (V, D) = 0, and therefore H 2 (V, D) is H -torsion.
Iw

Proof. In the case A), we have W0 = W, and the statement follows directly
from Proposition 3.13 and the definition of the -invariant. In the case B),
the diagram (3.56), together with Proposition 3.12, show that the injectivity
of Hc1 (M0 ) → H 1 (M) and the condition (3.53) are both equivalent to the
following condition:
Im(δ0 ) ∩ Hc1 (M0 ) = {0}.

The statement follows now from [5], Proposition 2.2.4, where it is proved that
(V, D) can be computed as the slope of the map δ0 : H 0 (M1 ) → H 1 (M)
with respect to the decomposition H 1 (M) H 1f (M) ⊕ Hc1 (M).

3.4 p-adic height pairings

3.4.1 Extended Selmer groups
Let V be a p-adic representation of G Q,S which satisfies the conditions C1–3)
of Subsection 3.2.1. Fix a submodule D of Dst (V ) such that Dst (V ) = D ⊕
82 Denis Benois

Fil0 Dst (V ). We will always assume that the condition M of Section 3.3.2 holds
for D. Then, by Proposition 3.12, one has an exact sequence

0 → M0 → M → M1 → 0,

where Dcris (M0 ) = D/D−1 , Dcris (M1 ) = D1 /D and rk(M0 ) = rk(M1 ).

Proposition 3.16. There exists an exact sequence

0 → H 0 (M1 ) → H 1 (V, D) → H 1f (Q, V ) → 0, (3.57)

where H 1f (Q, V ) is the Bloch–Kato Selmer group and

dim E H 0 (M1 ) = rk(M1 ).

Proof. By the definition of R (V, D), the group H 1 (V, D) is the kernel of
the map
⎛ ⎞
 
HS1 (Q, V ) ⊕ ⎝ H 1f (Qv , V )⎠ ⊕ H 1 (D) → H 1 (Qv , V ).
v∈S−{ p} v∈S

The proposition follows directly from this description of H 1 (V, D) together

with the following facts:

a) H 0 (M1 ) = ker(H 1 (D) → H 1 (Q p , V ));

b) Im(H 1 (D) → H 1 (Q p , V )) = H 1f (Q p , V ).

The proof of a) and b) can be extracted from the construction of the

-invariant in [5], Section 2.2.1, but we recall the arguments for reader’s
convenience. Consider the exact sequence

0 → D1 → D†rig (V ) → gr2 D†rig (V ) → 0.

 
In the proof of Proposition 3.11, we saw that H 0 gr2 D†rig (V ) = 0, and there-
fore the map H 1 (D1 ) → H 1 (Q p , V ) is injective. Taking the long cohomology
sequence associated to

0 → D → D1 → M1 → 0,

we obtain
0 → H 0 (M1 ) → H 1 (D) → H 1 (D1 ),

and a) is proved.
 Selmer complexes and p-adic Hodge theory 83

 
Using (3.22), it is not difficult to show that H 1f gr2 D†rig (V ) = 0, and
therefore H 1f (D1 ) H 1f (Q p , V ). Consider the exact sequence
0 → D−1 → D1 → M → 0.
Since H 0 (M) = 0 and H 1f (D−1 ) = H 1 (D−1 ), by [5], Corollary 1.4.6, one has
an exact sequence
0 → H 1 (D−1 ) → H 1f (D1 ) → H 1f (M) → 0.
On the other hand, from Proposition 3.12 it follows that
Im(H 1 (M0 ) → H 1 (M)) = H 1f (M).

Thus, Im(H 1 (D) → H 1 (D1 )) = H 1f (D1 ) and b) is proved.

Definition. We will call H 1 (V , D) the extended Selmer group associ-

ated to D.

Remarks. 1) Extended Selmer groups associated to ordinary local conditions

were studied in [45], [46].
2) If M = 0, the group H 1 (V, D) coincides with the Bloch–Kato Selmer
group. One expects that, if D is regular, the appearance of H 0 (M1 ) in the
short exact sequence (3.57) reflects the presence of extra-zeros of the p-adic
L-function L p (V , D, s). This question will be studied in [8].

3.4.2 p-adic height pairings

In [46], Nekovář found a new construction of the p-adic height pairing on
extended Selmer groups defined by Greenberg’s local conditions. In this sec-
tion, we follow his approach verbatim working with local conditions defined
by (ϕ, N )-submodules. Let V be a p-adic representation of G Q,S which sat-
isfies the conditions C1–3) and the condition C4b) from Subsection 3.2.1. In
particular, we assume that V is equipped with a bilinear form which induces
an isomorphism V V ∗ (1). We keep notation and conventions of Subsec-
tion 3.4.1. Set A = H /(X 2 ). Then A = E[X ]/(X 2 ), and one has an exact
sequence
X
0→E−
→ A → E → 0. (3.58)
Set V A = V ⊗ E Aι , and D A = D ⊗ E Aι , and consider the complex
R (V A , D A ). The sequence (3.58) induces a distinguished triangle
R (V , D) → R (V A , D A ) → R (V, D),
84 Denis Benois

which gives the coboundary map

β
H 1 (V, D) −
→ H 2 (V, D).
Let D⊥ ⊂ Dst (V ∗ (1)) denote the dual submodule.

Definition. Let V be a p-adic representation which satisfies the conditions

C1–4b) from Subsection 3.2.1, the condition S) from Subsection 3.4.1, and let
D be a (ϕ, N )-submodule of Dst (V ) such that the condition M) holds for D.
We define the p-adic height pairing associated to D as the bilinear map
 , V,D : H 1 (V , D) × H 1 (V ∗ (1), D ⊥ ) → E (3.59)
given by x, yV ,D = β(x)∪ y, where ∪ : H 2 (V, D)× H 1 (V ∗ (1), D ⊥ ) → E
denotes the duality defined in Proposition 3.10.

Remarks. 1) In the ordinary setting, it is possible to work over Z p , and

Nekovář’s descent machinery gives very general formulas of the Birch and
Swinnerton-Dyer type. In the general setting, we are forced to work over
Q p if we want to use the theory of (ϕ, )-modules.
2) The pairing (3.59) will be studied in detail in [8]. In particular, we compare
our construction to the p-adic height pairing constructed by Nekovář in
[45] and relate it to universal norms.
The following result can be seen as an analog of Proposition 3.15 in the
weight −1 case.

Proposition 3.17. Let V be a p-adic representation which satisfies the condi-

tions C1–4b) from Subsection 3.2.1 and the condition S) from Subsection 3.3.1.
Let D be a (ϕ, N )-submodule of Dst (V ) such that the condition M) holds for
D. Assume that
a) The weak Leopoldt conjecture holds for V ∗ (1);
b) The pairing  , V ,D is non-degenerate;
c) V does not contain subrepresentations of the form Q p (m).
1 (V, D) = 0, and therefore H 2 (V, D) is H -torsion.
Then HIw Iw

Proof. If the p-adic height pairing is non-degenerate, the map β is injective.

The diagram
 Selmer complexes and p-adic Hodge theory 85

gives rise to a commutative diagram with exact rows

1 (V, D)
Thus HIw 0 = 0. On the other hand, as the weak Leopoldt conjecture
holds for V ∗ (1) and V HQ,S = 0, the Λ[1/ p]-module HIw,S1 (Q, V ) is free and

injects into HIw1 (Q , V ). Therefore, as in the proof of Proposition 3.13, one

p
has
 
HIw1
(V , D) = HIw,S
1
(Q, T ) ⊗ H ∩ HIw 1
(D).
1 (V , D) is H -free, and the vanishing of H 1 (V, D)
This implies that HIw Iw 0
gives HIw (V , D) = 0.
1

References
[1] K. Barré-Sirieix, G. Diaz, F. Gramain, G. Philibert, Une preuve de la conjecture
de Mahler-Manin, Invent. Math. 124 (1996), 1–9.
[2] J. Bellaïche, G. Chenevier, p-adic families of Galois representations and higher
rank Selmer groups, Astérisque 324 (2009), 314 pages.
[3] J. Bellaïche, Ranks of Selmer groups in an analytic family, Trans. Amer. Math.
Soc. 364 (2012), 4735-4761.
[4] D. Benois, On Iwasawa theory of p-adic representations, Duke Math. J. 104
(2000), 211–267.
[5] D. Benois, A generalization of Greenberg’s L -invariant, Amer. J. Math. 133
(2011), 1573-1632.
[6] D. Benois, Trivial zeros of p-adic L-functions at near central points, J. Inst. Math.
Jussieu 13 (2014), 561–598.
[7] D. Benois, On extra zeros of p-adic L-functions: the crystalline case, In: Iwa-
sawa theory 2012. State of the Art and Recent Advances (T. Bouganis and O.
Venjakob, eds.), Contributions in Mathematical and Computational Sciences, vol.
7, Springer, 2014 (to appear).
[8] D. Benois, p-adic height pairings and p-adic Hodge theory, preprint (2014),
available at arXiv.org/pdf/1412.7305v1.pdf.
[9] L. Berger, Représentations p-adiques et équations différentielles, Invent. Math.
148 (2002), 219–284.
[10] L. Berger, Bloch and Kato’s exponential map: three explicit formulas, Doc.
Math., Extra Volume: Kazuya Kato’s Fiftieth Birthday (2003), pp. 99–129.
[11] L. Berger, Equations différentielles p-adiques et (ϕ, N )-modules filtrés, In:
Représentations p-adiques de groupes p-adiques I (L. Berger, P. Colmez, Ch.
Breuil eds.), Astérisque 319 (2008), 13–38.
86 Denis Benois

[12] L. Berger, P. Colmez, Familles de représentations de de Rham et monodromie

p-adique, In: Représentations p-adiques de groupes p-adiques I (L. Berger, P.
Colmez, Ch. Breuil eds.), Astérisque 319 (2008), 303–337.
[13] S. Bloch, K. Kato, L-functions and Tamagawa numbers of motives, In: The
Grothendieck Festschrift, vol. 1 (P. Cartier, L. Illusie, N. M. Katz, G. Laumon,
Yu. Manin, K. Ribet, eds.), Progress in Math. vol. 86, Birkhäuser Boston, Boston,
MA, 1990, pp. 333–400.
[14] F. Cherbonnier, P. Colmez, Représentations p-adiques surconvergentes, Invent.
Math., 133 (1998), 581–611.
[15] F. Cherbonnier, P. Colmez, Théorie d’Iwasawa des représentations p-adiques
d’un corps local, J. Amer. Math. Soc., 12 (1999), 241–268.
[16] J. Coates, B. Perrin-Riou, On p-adic L-functions attached to motives over Q, In:
Algebraic Number Theory – in honor of K. Iwasawa (J. Coates, R. Greenberg, B.
Mazur and I. Satake, eds.), Adv. Studies in Pure Math., 17 (1989), pp. 23–54.
[17] P. Colmez, Fonctions L p-adiques, In: Seminaire Bourbaki. Vol. 1998/99,
Astérisque, 266 (2000), 21–58.
[18] P. Colmez, Les conjectures de monodromie p-adiques, In: Seminaire Bourbaki.
Vol 2001/2002, Asterisque 290 (2003), 53–101.
[19] P. Colmez, Représentations triangulines de dimension 2, In: Représentations
p-adiques de groupes p-adiques I (L. Berger, P. Colmez, Ch. Breuil eds.),
Astérisque 319 (2008), 213-258.
[20] M. Flach, A generalization of the Cassels–Tate pairing, J. Reine Angew. Math.
412 (1990), 113–127.
[21] G. Faltings, Crystalline cohomology and p-adic Galois representations, In: Al-
gebraic Analysis, Geometry and Number Theory (Baltimore, MD, 1988), John
Hopkins University Press, Baltimore MD, 1989, pp. 25–80.
[22] B. Ferrero, R. Greenberg, On the behavior of p-adic L-functions at s = 0, Invent.
Math. 50 (1978/79), 91–102.
[23] J.-M. Fontaine, Représentations p-adiques des corps locaux, In: The
Grothendieck Festschrift, vol. 2 (P. Cartier, L. Illusie, N. M. Katz, G. Laumon,
Yu. Manin, K. Ribet, eds.), Progress in Math. vol. 87, Birkhäuser Boston, Boston,
MA, 1990, pp. 249–309.
[24] J.-M. Fontaine, Valeurs spéciales de fonctions L des motifs, In: Séminaire
Bourbaki, Vol. 1991/1992, exposé 751, Astérisque, 206 (1992), 205–249
[25] J.-M. Fontaine, Le corps des périodes p-adiques, In: Périodes p-adiques (J.-M.
Fontaine, ed.), Astérisque, 223 (1994), 59-102.
[26] J.-M. Fontaine, Représentations p-adiques semi-stables, In: Périodes p-adiques
(J.-M. Fontaine, ed.), Astérisque, 223 (1994), 59-102.
[27] J.-M. Fontaine, B. Perrin-Riou, Autour des conjectures de Bloch et Kato; co-
homologie galoisienne et valeurs de fonctions L, In: Motives (U. Jannsen, S.
Kleiman, J.-P. Serre, eds.), Proc. Symp. in Pure Math., 55, (1994), 599–706.
[28] R. Greenberg, Iwasawa theory for p-adic representations, In: Algebraic Num-
ber Theory – in honor of K. Iwasawa (J. Coates, R. Greenberg, B. Mazur and I.
Satake, eds.), Adv. Studies in Pure Math., 17 (1989), 97–137.
[29] R. Greenberg, Trivial zeros of p-adic L-functions, In: p-Adic Monodromy and
the Birch and Swinnerton-Dyer Conjecture (B. Mazur and G. Stevens, eds.),
Contemp. Math. 165, (1994), 149–174.
 Selmer complexes and p-adic Hodge theory 87

[30] R. Greenberg, Iwasawa Theory and p-adic Deformations of Motives, In: Motives
(U. Jannsen, S. Kleiman, J.-P. Serre, eds.), Proc. of Symp. in Pure Math., 55,
(1994), 193–223.
[31] E. Hellmann, Families of trianguline representations and nite slope spaces,
preprint available on http://arXiv:1202.4408v1.
[32] L. Herr, Sur la cohomologie galoisienne des corps p-adiques, Bull. Soc. Math.
France 126 (1998), 563–600.
[33] L. Herr, Une approche nouvelle de la dualité locale de Tate, Math. Annalen 320
(2001), 307–337.
[34] U. Jannsen, On the -adic cohomology of varieties over number fields and its
Galois cohomology, In: Galois Groups over Q (Y. Ihara, K. Ribet and J.-P. Serre,
eds.) , Math. Sciences Research Inst. Publ., vol. 16, Springer, 1989, pp. 315–360.
[35] U. Jannsen, Mixed motives and algebraic K-theory, Lecture Notes in Math. vol.
1400, Springer, Berlin, 1990.
[36] K. Kato, p-adic Hodge theory and values of zeta-functions of modular forms,
In: Cohomologies p-adiques et applications arithmétiques (III) (P. Berthelot, J.-
M. Fontaine, L. Illusie, K. Kato, M. Rapoport, eds.), Astérisque 295 (2004), pp.
117-290.
[37] N. M. Katz, W. Messing, Some consequences of the Riemann hypothesis for
varieties over finite fields, Invent. Math. 23 (1974), 73–77.
[38] K. Kedlaya, A p-adic monodromy theorem, Ann. of Math. 160 (2004), 94–184.
[39] K. Kedlaya, R. Liu, On families of (ϕ, )-modules, Algebra and Number Theory,
4 (2010), 943967.
[40] K. Kedlaya, J. Pottharst, L. Xiao, Cohomology of arithmetic families of (ϕ, )-
modules, Preprint available on http://arxiv.org/abs/1203.5718.
[41] M. Lazard, Les zéros des fonctions analytiques d’une variable sur un corps valué
complet, Publ. Math. IHES 14 (1962), 47–75.
[42] R. Liu, Cohomology and Duality for (ϕ, )-modules over the Robba ring, Int.
Math. Research Notices (2007), no. 3, 32 pages.
[43] Yu. Manin, Periods of cusp forms and p-adic Hecke series, Math USSR Sbornik
92 (1973), 371–393.
[44] B. Mazur, J.Tate, J. Teitelbaum, On p-adic analogues of the conjectures of Birch
and Swinnerton-Dyer, Invent. Math. 84 (1986), 1–48.
[45] J. Nekovář, On p-adic height pairing, In: Séminaire de Théorie des Nombres,
Paris, 1990/91 (S. David, ed.), Progress in Mathematics, vol. 108, Birkhäuser
Boston, 1993, pp. 127202.
[46] J. Nekovář, Selmer complexes, Astérisque, 310 (2006), 559 pages.
[47] J. Nekovář, W. Nizioł, Syntomic cohomology and p-adic regulators for varieties
over p-adic fields, Preprint available on http://arxiv.org/abs/1309.7620.
[48] T. Ochiai, Control theorem for Bloch–Kato’s Selmer groups of p-adic represen-
tations, J. Number Theory 82 (1990), 69–90.
[49] B. Perrin-Riou, Théorie d’Iwasawa et hauteurs p-adiques, Invent. Math. 109
(1992), 137–185.
[50] B. Perrin-Riou, Théorie d’Iwasawa des représentations p-adiques sur un corps
local, Invent. Math. 115 (1994), 81–149.
[51] B. Perrin-Riou, Fonctions L p-adiques des représentations p-adiques, Astérisque
229 (1995), 198 pages.
88 Denis Benois

[52] J. Pottharst, Analytic families of finite slope Selmer groups, Algebra and Number
Theory 7 (2013), 1571-1611.
[53] J. Pottharst, Cyclotomic Iwasawa theory of motives, Preprint available on
http://math.bu.edu/people/potthars/.
[54] J. Pottharst, Analytic families of finite slope Selmer groups, Preprint available on
math.bu.edu/people/potthars/writings/.
[55] J.-P. Serre, Facteurs locaux des fonctions zêta des variétés algébriques (défini-
tions et conjectures), Séminaire Delange-Pisot-Poitou. Théorie des nombres, 11
no. 2 (1969-1970), Exp. No. 19, 15 pages.
[56] P. Schneider, p-adic height pairings. I, Invent. Math. 69 (1982), 401409.
[57] P. Schneider, J. Teitelbaum, Algebras of p-adic distributions and ad- missible
representations, Invent. Math. 153 (2003), 145–196.
[58] C. Skinner, E. Urban, The Iwasawa Main Conjectures for GL2 , Invent. Math. 195
(2014), 1–277.
[59] G. Tamme, Karoubi’s relative Chern character, the rigid syntomic regulator, and
the Bloch–Kato exponential map, Preprint available on http://arxiv.org/abs/1111.
4109.
[60] T. Tsuji, p-adic etale cohomology and crystalline cohomology in the semistable
reduction case, Invent. Math. 137 (1999), 233–411.
[61] M. Vishik, Non-archemedian measures connected with Dirichlet series, Math.
USSR Sbornik 28 (1976), 216–228.
[62] J.-P. Wintenberger, Le corps des normes de certaines extensions infinies de corps
locaux; applications, Ann. Sc. ENS 16 (1983), 59–89.
 4
A survey of applications of the circle method to
rational points
T.D. Browning
School of Mathematics, University of Bristol, Bristol, BS8 1TW, UK
E-mail address: [email protected]

Introduction
Given a number field k and a projective algebraic variety X defined over k, the
question of whether X contains a k-rational point is both very natural and very
difficult. In the event that the set X (k) of k-rational points is not empty, one
can also ask how the points of X (k) are distributed. Are they dense in X under
the Zariski topology? Are they dense in the set

X (Ak ) = X (kv )
v∈k

of adèlic points under the product topology? Can one count the points in X (k)
of bounded height? In favourable circumstances, the Hardy–Littlewood circle
method can systematically provide answers to all of these questions.
The focus of this survey will be upon the first and most basic of these
questions: when is X (k) = ∅ for a given variety X defined over k? In
considering what the circle method has to say about this we shall usually
restrict our attention to varieties X defined over k that are geometrically inte-
gral, non-singular and projective. Recall that such a family is said to satisfy
the Hasse principle if any variety in the family has a k-point as soon as it
has a point in every completion of k. Sometimes we will drop the assump-
tion on non-singularity, saying that the smooth Hasse principle holds for
a family of such varieties when the Hasse principle holds for the smooth
locus of X .

Date: April 2, 2014.

2010 Mathematics Subject Classification. 11P05 (11D72, 14G05).

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

89
90 T.D. Browning

Quadrics are among the first examples of families satisfying the Hasse prin-
ciple. Beyond this simple setting, it is quite rare to find varieties which satisfy
the Hasse principle. In dimension 2 we have the following counter-example.

Example 4.1. A generalised Châtelet surface over k is a proper smooth

model of an affine surface

Y 2 − a Z 2 = f (T ),

where a ∈ k ∗ \ k ∗ 2 and f ∈ k[T ] is a polynomial. An infinite family of

counter-examples to the Hasse principle was discovered by Colliot-Thélène,
Coray and Sansuc [21, Prop. C]. This corresponds to taking k = Q, a = −1
and f (T ) = (−T 2 + c)(T 2 − c + 1), where c is a positive integer congruent
to 3 modulo 4. Letting X c /Q denote this Châtelet surface, it follows from this
work that X c (R) = ∅ and X c (Q p ) = ∅ for all primes p, but X c (Q) = ∅.

A systematic and unified explanation of various explicit failures of the Hasse

principle was provided by Manin in his 1970 ICM address [51] in terms of
the Brauer–Grothendieck group Br(X ) = Hét2 (X, Gm ), which can be asso-
ciated to any geometrically integral, projective variety X . We shall discuss
the Brauer–Manin obstruction to the Hasse principle in Section 4.3.1. The
counter-examples X c /Q were shown to conform to this explanation in [21].
It was later shown by Colliot-Thélène, Sansuc and Swinnerton-Dyer [24, 25]
that this obstruction actually explains all counter-examples to the Hasse prin-
ciple among generalised Châtelet surfaces, with deg f (T )  4, defined over
any number field k. In particular, it follows from their work that such sur-
faces satisfy the Hasse principle when f is irreducible over k, which is the
generic situation. In Section 4.3.3 we discuss a partial extension of this result
when k = Q.
The topic of rational points on varieties is vast and we shall only touch upon
those aspects which have involved the circle method intrinsically (and even
then we will not be able to give many details). This survey will focus on the
following subjects:

§1 The circle method and complete intersections

§2 The circle method and diagonalisation
§3 The circle method and descent

Acknowledgements. The author is grateful to Jean-Louis Colliot-Thélène,

Boris Moroz, Sean Prendiville, Damaris Schindler, Alexei Skorobogatov and
 A survey of applications of the circle method to rational points 91

Efthymios Sofos for comments on an earlier draft of this paper. While working
on this paper the author was supported by ERC grant 306457.

4.1 The circle method and complete intersections

4.1.1 The circle method over Q
Let us begin with some generalities about the circle method over Q. Sup-
pose that we are given a system of polynomials f = ( f 1 , . . . , f s ), with
f i ∈ Z[x1 , . . . , xn ] for 1  i  s. Let B ⊂ Rn be a fixed box, with side
lengths parallel to the coordinate axes, and write BB for the dilation of this
region by a parameter B  1. Define the counting function

N (B) = #{x ∈ Zn ∩ BB : f i (x) = 0 for 1  i  s},

for B  1. When f 1 , . . . , f s are homogeneous and cut out a non-singular

complete intersection X ⊂ Pn−1 , with dim(X )  3, this amounts to counting
integral points of bounded height on the universal torsor over X . (Note that
the affine cone over X in An \ {0} is the unique universal torsor over X up to
isomorphism.)
In order to show that the variety f 1 = · · · = f s = 0 has an integral point
it suffices to show N (B) > 0 if B is taken to be sufficiently large and B is
chosen suitably. Let e(z) = e2πi z for any z ∈ R. By orthogonality one has
.
N (B) = S(α)dα, (4.1)
(0,1]s

where
S(α) = e (α1 f 1 (x) + · · · + αs f s (x)) .
x∈Zn ∩BB

The classical circle method is concerned with an asymptotic analysis of this

integral, as B → ∞, based on a suitable dissection of the range of integration.
Suppose that deg f i = di for 1  i  s. Let Q satisfy 1  Q  B maxi di .
For given integers ai , q such that 1  ai  q  Q, we define
/ /
/ /
M(q, a; Q) = α ∈ (0, 1]s : /αi − aqi /  Q B −di for 1  i  s .

The set of major arcs of level Q is the union

0 0
M(Q) = M(q, a; Q).
qQ 1ai q
gcd(a,q)=1
92 T.D. Browning

The general philosophy behind the circle method is that the contribution from
integrating over the major arcs should produce a main term in our asymptotic
formula for N (B). Generally, this main term takes the shape
 
B n− 1i s di · σv , (4.2)
v∈Q

where σv are local densities for the variety f 1 = · · · = f s = 0. More precisely,

. .
σ∞ = e(z 1 f 1 (x) + · · · + z s f s (x)) dx dz
Rs B

is the real density and

1 2
σ p = lim p −t (n−s) # x ∈ (Z/ p t Z)n : f i (x) ≡ 0 (mod p t ) for 1  i  s
t→∞

is the p-adic density, for each prime p.

On the other hand, the minor arcs, which are defined to be

m(Q) = (0, 1]s \ M(Q),

should produce a smaller overall contribution, with

. 
S(α)dα = o(B n− 1i s di ).
m(Q)

We refer the reader to Davenport [27] or Vaughan [71] for a detailed account
of the circle method. The fundamental tool here is Weyl’s inequality and its
generalisations. One such generalisation, which was used to spectacular effect
by Heath-Brown [40] to analyse cubic forms in 14 variables, is the method of
van der Corput differencing.
Suppose for the moment that s = 1 and let f ∈ Z[x1 , . . . , xn ] be a poly-
nomial of degree d. Let H ∈ Z ∩ [1, B]. It will be convenient to write,
temporarily,

e(α f (x)), if x ∈ BB,
w(x) =
0, otherwise.

Let H = Zn ∩ (0, H ]n . Then the kernel of the van der Corput method is the
observation that

#H S(α) = w(x + h) = w(x + h).

h∈H x∈Zn x∈Zn h∈H
 A survey of applications of the circle method to rational points 93

An application of Cauchy’s inequality yields

/ /2
/ /
H 2n |S(α)|2 $ B n / w(x + h)/
x∈Zn h∈H

=B n
w(x + h1 )w(x + h2 )
h1 ∈H h2 ∈H x∈Zn

= Bn w(y + h1 − h2 )w(y)
h1 ∈H h2 ∈H y∈Zn

= Bn N (h) w(y + h)w(y),

h∈Zn y∈Zn
|h|H

where N (h) = #{h1 , h2 ∈ H : h = h1 − h2 } $ H n . We therefore conclude

that
B 2n Bn
|S(α)|2 $ H −n B n |Th (α)| $ + |Th (α)|, (4.3)
Hn Hn
h 0<|h|H

where

Th (α) = e α( f (x + h) − f (x)) .
x∈Zn ∩BB
x+h∈BB

The significance of this is that, for each non-zero h ∈ Zn , the exponential sum
Th (α) is an exponential sum involving the polynomial f (x + h) − f (x) of
degree d − 1. Taking H = B recovers the first step in the proof of Weyl’s
inequality
|S(α)|2 $ B n + |Th (α)|. (4.4)
0<|h|B

The presence of a free parameter H in (4.3) brings a lot of extra flexibility into
the analysis, compared with (4.4). This process can be repeated, using (4.3) or
(4.4) at each step, until one arrives at exponential sums of smaller degree that
can be estimated efficiently using other methods.
Suppose that X ⊂ Pn−1 is a complete intersection cut out by s hypersurfaces
of degrees d1 , . . . , ds , such that X is geometrically integral, non-singular and
projective. Assuming that one is fortunate enough to show that the major and

minor arc contributions behave as above, with v σv absolutely convergent,
one can conclude that the Hasse principle holds for X . Indeed, if X (R) = ∅,
one takes B to be a small box centred on a (necessarily non-singular) real
point. This is enough to ensure that σ∞ > 0. Likewise, one can prove that

p σ p > 0 if X (Q p ) = ∅ for every p. Hence N (B) > 0 if B is sufficiently
large, which implies that X (Q) = ∅, as desired.
94 T.D. Browning

As α runs through elements of (0, 1]s one might expect the numbers
e (α1 f 1 (x) + · · · + αs f s (x))
to be randomly scattered around the unit circle as we vary over x ∈ Zn . This
leads via the central limit theorem to the expectation that S(α) should usually
be of order roughly B n/2 . On the other hand, as in (4.2), one heuristically

expects the main term in any asymptotic formula to be of order B n− 1is di .
Thus we can only hope to succeed with the circle method when
n>2 di . (4.5)
1is

We call this the square root barrier.

Example 4.2. In most applications we remain very far from the square root
barrier. One notable exception is given by considering varieties defined by
norm forms. Let k/Q be a number field of degree d  2 and fix a basis
{ω1 , . . . , ωd } for k as a Q-vector space. We denote the norm form by
Nk/Q (x1 , . . . , xd ) = Nk/Q (x1 ω1 + · · · + xd ωd ),
where Nk/Q denotes the field norm. This is a homogeneous polynomial of
degree d defined over Q. Let X ⊂ P2d denote the hypersurface
cx0d + Nk1 /Q (x1 , . . . , xd ) + Nk2 /Q (xd+1 , . . . , x2d ) = 0,
for any c ∈ Q∗ . This variety is geometrically integral and projective, but it is
singular. Birch, Davenport and Lewis [5] established the smooth Hasse prin-
ciple for X . Note that this is at the square-root barrier, since the number of
variables is 2d + 1 and the degree is d.
The Birch–Davenport–Lewis result has been generalised to arbitrary number
fields independently by Schindler and Skorobogatov [62] and by Swarbrick
Jones [68]. Their work covers the hypersurfaces
cx0d + N K 1 /k (x1 , . . . , xd ) + N K 2 /k (x d+1 , . . . , x2d ) = 0, (4.6)
where K 1 , K 2 are degree d extensions of any number field k and c ∈ k ∗ .

4.1.2 The circle method over k

We proceed to discuss one aspect of the circle method over a general number
field k, in the context of hypersurfaces (i.e. s = 1). When k = Q a key inno-
vation, due to Kloosterman [47], involves decomposing (0, 1] in (4.1) using
a Farey dissection to keep track of the precise end points of the intervals.
 A survey of applications of the circle method to rational points 95

This allows one to introduce non-trivial averaging over the numerators of the
approximating fractions a/q, an approach that is usually called the “Klooster-
man refinement”. This method is not immediately available to us when passing
to general k, since aside from work of Cassels, Ledermann and Mahler [17]
√
about the imaginary quadratic fields Q(i) and Q( −3), no generalisation is
known of the Farey dissection to the number field analogue of (0, 1].
An alternative approach is provided via the smooth δ-function technology
of Duke, Friedlander and Iwaniec [32], as revisited by Heath-Brown
[39, Thm. 1] and recently extended to general number fields by Browning and
Vishe [14]. Suppose o is the ring of integers of our number field k, which we
assume to have degree d. The ideal norm will be designated Na = #o/a for any
integral ideal a ⊆ o. In line with our description of the Hardy–Littlewood cir-
cle method over Q, we would like to use Fourier analysis to find an alternative
description of the indicator function δk : o → {0, 1}, given by

1, if α = 0,
δk (α) =
0, otherwise.

The following result is a special case of [14, Thm. 1.2].

Theorem 4.3. Let Q  1 and let α ∈ o. Then there exists a positive constant
c Q and an infinitely differentiable function h(x, y) : (0, ∞) × R → R such
that
 
cQ ∗ Nb Nk/Q (α)
δk (α) = 2d σ (α)h , ,
Q Qd Q 2d
(0)=b⊆o σ (mod b)


where the notation ∗σ (mod b) means that the sum is taken over primitive
additive characters modulo b. The constant c Q satisfies

c Q = 1 + O N (Q −N ),

for any N > 0. Furthermore, we have h(x, y) $ x −1 for all y and h(x, y) = 0
only if x  max{1, 2|y|}.

In [14, Thm 1.1] this result is used to establish that X (k) = ∅, for any
non-singular cubic hypersurface X ⊂ Pn−1 defined over k, with n  10.

Question 4.4. Can one apply Theorem 4.3 to generalise Heath-Brown’s work
on quadratic forms [39] to arbitrary number fields k?
96 T.D. Browning

4.1.3 Complete intersections

Throughout this section let X d,s ⊂ Pn−1 be a complete intersection over a
number field k cut out by s hypersurfaces Fi = 0 of degree d, with forms
Fi ∈ k[x1 , . . . , x n ]. We will always assume that our complete intersection
is geometrically integral and non-degenerate, but it need not be non-singular.
When X d,s is non-singular it has been conjectured by Colliot-Thélène (see [58,
App. A] and [8, App.]) that the Hasse principle holds when n + 1  ds (i.e.
X d,s is Fano) and n − 1 − s  3 (i.e. X d,s has dimension at least 3). We will
summarise conditions on d, s and the structure of X d,s under which the smooth
Hasse principle has been established for X d,s using the circle method.

The local problem

With the Hasse principle in place for X = X d,s one is naturally led to seek
further conditions under which X (kv ) = ∅ for every completion kv of k. One
might expect this purely local problem to be easier, but apart from a few excep-
tions, the bounds required on n needed to ensure local solubility are rather poor.
Let us discuss this briefly in the case s = 1 of hypersurfaces. When d = 2 we
have X (kv ) = ∅ for every non-archimedean place v ∈ k , provided only that
n  5. When d = 3 work of Lewis [49] (and Dem’yanov [29] when the char-
acteristic of the residue field is distinct from 3) secures the same conclusion
provided that n  10.
For every non-archimedean completion kv of k and for every d  2 it is
fairly easy to construct examples of hypersurfaces of degree d in d 2 variables
which do not have kv -points. This shows that the above results for d ∈ {2, 3}
are best possible. The situation for d  4 is much less satisfactory. Specialising
to hypersurfaces X defined over Q, we know that there is a number vd such
that X (Q p ) = ∅ for every prime p, as soon as n > vd . Brauer [7] achieved the
first result in this direction using an elementary argument based on multiple
nested inductions. The resulting value of vd was too large to write down, but
the underlying ideas have since been revisited and improved upon by Wooley
d
[74], with the outcome that we may take vd ≤ d 2 in general. An account of
recent developments concerning the local solubility problem for forms in many
variables can be found in Heath-Brown’s 2010 ICM address [41].

The global problem

We begin by summarising two conditions that any application of the circle
method naturally imposes on the complete intersection X d,s . The first involves
the h-invariant h(F) of a form F ∈ k[x1 , . . . , xn ]. This is defined to be the
least positive integer h such that F can be written identically as
 A survey of applications of the circle method to rational points 97

A1 B1 + · · · + Ah Bh ,
for forms Ai , Bi ∈ k[x1 , . . . , xn ] of positive degree. We define the h-invariant
h(X d,s ) of X d,s to be min{h(F)}, with the minimum taken over all forms F
of the rational pencil c1 F1 + · · · + cs Fs , with [c1 , . . . , cs ] ∈ Ps−1 (k). A rich
seam of results is founded on h(X d,s ) being “sufficiently large”, in terms of d,
n and s. The most important contribution adopting this point of view is due to
Schmidt [63].
A second point of view requires the singular locus of X d,s to be “sufficiently
small”. Define the Birch singular locus for X d,s to be the projective variety
{[x] ∈ Pn−1 : rank(J (x)) < s},
where J (x) is the Jacobian matrix of size s × n formed from the gradient
vectors ∇ F1 (x), . . . , ∇ Fs (x). We define  ∈ {−1, 0 . . . , n − 1} to be the
dimension of the Birch singular locus, with  = −1 if and only if the variety
is empty. There are several places in the literature (see Peyre [55, Cor. 5.4.9],
for example) where it has been erroneously assumed that  = −1 when X d,s
is non-singular. However, this is not true and in general one only has
  s − 1 + dim sing(X d,s ),
where sing(X d,s ) is the singular locus of X d,s . This can be seen by intersecting
the Birch singular locus with X d,s , before applying the projective dimension
theorem. When X d,s is non-singular it is possible to improve this inequal-
ity slightly. Thus it follows from work of Browning and Heath-Brown [10,
Lemma 3.1] that
s−2
when X d,s is non-singular. In particular, when s = 1 this shows that  = −1
when X d,1 is non-singular. It is easy to show, moreover, that  = s − 2 for any
complete intersection cut out by diagonal hypersurfaces.
At this point we should observe that independent works of Dietmann [31]
and Schindler [61] allow one to replace  by an alternative invariant, which we
denote temporarily by  . One can show in complete generality that  ≤ ,
but that  can be strictly less than  in appropriate cases. We shall work with
 in this survey, however.
We are now ready to recall the gold standard, by which all subsequent results
have been measured. Let us put
νd,s =  + 2 + s(s + 1)(d − 1)2d−1 .
The following result is due to Birch [4] when k = Q and to Skinner [65] for
general number fields. (See Aleksandrov and Moroz [1] for more on the role of
98 T.D. Browning

Table 4.1 Improvements (via the circle method) to the Birch bound over Q

(d, s) νd,s n ? Conditions? Who?

(2, 1) 5 3 X non-singular Heath-Brown [39]

(3, 1)  + 18 16 — Davenport [26]
14 — Heath-Brown [40]
13 X splits off a form Browning [8]
11 X splits off two forms Xue and Dai [75]
10 X non-singular Heath-Brown [38]
9 X non-singular Hooley [44]
9 X contains isolated Hooley [45]
double points
(4, 1) 49 41 X non-singular Browning and
Heath-Brown [9]
40 X non-singular Hanselmann [36]
(5, 1) 129 111 X non-singular Browning and Prendiville
[13]
(2, 2)  + 14 9 X contains 2 conjugate Browning and Munshi
double points over Q(i) [12]
14 11 X non-singular Munshi [54]

 in this result.) It is heavily influenced by the pioneering work of Davenport

[26] on cubic forms.

Theorem 4.5. The smooth Hasse principle holds for X d,s if n  νd,s .

We say that a form F ∈ k[x1 , . . . , xn ] splits off a form if F can

be written identically as A1 + A2 , for forms A1 ∈ k[x1 , . . . , xm ] and
A2 ∈ k[xm+1 , . . . , xn ] of positive degree, for some m ∈ {1, . . . , n}. In
Table 4.1 we summarise the improvements that have been made to Theorem 4.5
in the special case k = Q, using the circle method.

Remark. Some of the results in Table 4.1 pertain to non-singular complete

intersections X d,s ⊂ Pn−1 with n  n 0 , for appropriate n 0 depending on d
and s. It is easy to extend these results to establish the smooth Hasse principle
for complete intersections in n  n 0 +1+σ variables, where σ is the dimension
of the singular locus. To do so one argues by induction on σ  −1, the case
σ = −1 being represented in the table. If σ  0 one intersects the variety with
a generic Q-rational linear space of dimension n − 2 − σ . Bertini’s theorem
ensures that the resulting variety is a non-singular complete intersection of
codimension s and degree d in Pn−2−σ , from which the claim easily follows.

The analogue of Table 4.1 for complete intersections over arbitrary number
fields k is much smaller, with Theorem 4.5 still providing the best bounds in
 A survey of applications of the circle method to rational points 99

most cases. One notable exception concerns cubic hypersurfaces X ⊂ Pn−1

defined over k. In this setting a folklore conjecture predicts that X (k) = ∅ as
soon as n  10. When no constraints are placed on the singular locus of X , the
work of Pleasants [57] shows that n  16 variables are needed to achieve this
conclusion. When X is non-singular, building on Heath-Brown’s [38] break-
through result when k = Q, Skinner [64] showed that n  13 variables suffice.
The loss of 3 variables is entirely due to the lack of a suitable Kloosterman
methodology, a situation that is remedied by Theorem 4.3. Using this, Brown-
ing and Vishe [14] have established the conjecture for all non-singular cubic
hypersurfaces over any number field. Over Q, as indicated in Table 4.1, Heath-
Brown has shown that n  14 suffices when no restrictions are placed on X ,
which suggests the following.

Question 4.6. Can Heath-Brown’s [40] work on cubic forms in 14 variables

be generalised to arbitrary number fields?

So far we have only talked about complete intersections over number fields
k which can be handled using the circle method. For certain cases discussed
in Table 4.1 significant improvements are available through other channels.
Using descent theory, for example, Colliot-Thélène and Salberger [19] have
shown that the Hasse principle holds for any cubic hypersurface over k in
Pn−1 containing a set of three conjugate singular points, provided only that
n  4. As suggested by Example 4.2, the circle method can also be brought
to bear on cubics of this type, but the number of variables required is greater.
The case (d, s) = (2, 2) of pairs of quadrics is a further important class of
varieties that can be handled via descent methods. Assuming that X ⊂ Pn−1
is the common zero locus of two quadratic forms q1 , q2 ∈ k[x1 , . . . , xn ],
which we assume to be geometrically integral and not a cone, the work
of Colliot-Thélène, Sansuc and Swinnerton-Dyer [24] demonstrates that the
smooth Hasse principle holds if n  9. This can be reduced to n  8
when X is non-singular (see Heath-Brown [42]) or n  5 when X con-
tains a pair of conjugate singular points and does not belong to a certain
explicit class of varieties for which the Hasse principle is known to fail
(see [24, Thm. A]).

4.1.4 Recent developments

The technology that lies behind Theorem 4.5 is remarkably robust. In this sec-
tion we summarise some of the most interesting ways in which it has recently
been refined and extended.
100 T.D. Browning

Cubics
It turns out that when the singular locus is not too large, cubic hypersurfaces are
most efficiently handled by conjoining Theorem 4.3 with Poisson summation,
rather than using the division into major and minor arcs from Section 4.1.1
(and the squaring and differencing approach based on (4.3) and (4.4)). This
leads to estimates for complete exponential sums over finite fields, which
can be estimated very efficiently using Deligne’s [28] resolution of the Rie-
mann Hypothesis for varieties over finite fields (however, the methods become
less effective as the dimension of the singular locus grows). Deligne’s bounds
handle sums to prime, or square-free, moduli, but sums to prime power mod-
uli remain a considerable problem. Circumventing these difficulties, Hooley
[46] has recently established the Hasse principle for non-singular cubic hyper-
surfaces over Q in only 8 variables, provided that one assumes Hypothesis
HW. This amounts to some technical (as yet unproved) assumptions about the
analytic properties of certain associated Hasse–Weil L-functions.

Improvements for every degree

In Table 4.1 a saving of 9 variables is recorded for non-singular quartic hyper-
surfaces. Given a quartic form f ∈ Z[x1 , . . . , xn ], the idea behind the proof in
[9] is to first carry out van der Corput differencing, as in (4.3) (the saving of an
extra variable comes through an averaged version of this in [36]). The expo-
nential sum Th (α) now involves the cubic polynomial f (x + h) − f (x), which
has cubic part equal to h.∇ f (x). The idea is then to estimate these exponential
sums directly, using Poisson summation, which appears to be the only other
weapon we have in our arsenal.
Let f ∈ Z[x1 , . . . , xn ] be a non-singular form of degree d  5. It is natural
to ask whether a similar method can be adapted to improve Theorem 4.5 in the
setting of hypersurfaces, which we recall establishes the Hasse principle for
n  νd,1 = 1 + (d − 1)2d . This is answered affirmatively in forthcoming work
of Browning and Prendiville [13], where it is shown that one can take
 √ 
d
n d− 2d .
2
The proof of this result involves throwing every available tool at the problem,
depending on the Diophantine approximation properties of the number α in the
exponential sum S(α). For generic α it is most efficient to apply (4.3) d − 3
times, producing a family of cubic exponential sums with associated parame-
ters H1 , . . . , Hd−3 ∈ Z∩[1, B], which one estimates using Poisson summation
as in [9]. For other α it is more advantageous to apply Weyl differencing (4.4)
multiple times, and for others still, one is better off applying a mixture of the
 A survey of applications of the circle method to rational points 101

approaches. It turns out that the method is most efficient when d is small. When
d = 5, as in Table 4.1, one is able to save 18 variables over Birch’s theorem.

Forms of differing degree

Theorem 4.5 only applies to complete intersections cut out by hypersurfaces
of the same degree. It is not entirely clear how this method can be adapted to
handle a system of forms of differing degree, since the process of Weyl differ-
encing (4.4) involved in the proof eradicates the presence of the lower degree
forms. Schmidt encounters the same problem in the work [63] cited above. A
more efficient treatment of this issue is a key ingredient in work of Browning,
Dietmann and Heath-Brown [15], which shows how the circle method can be
used to analyse non-singular varieties cut out by a cubic and quadric hyper-
surface in n  29 variables. In fact p-adic solubility is guaranteed for n in
this range for every p and so one gets Q-points provided only that there are
R-points.
Browning and Heath-Brown [10] have looked at generalising Birch’s result
to complete intersections cut out by hypersurfaces of arbitrary degree. The
final result is too complicated to state here, but it has a number of particularly
succinct corollaries. First, for a non-singular complete intersection X ⊂ Pn−1
cut out by a hypersurface of degree D and a hypersurface of degree E, where
D > E ≥ 2, the Hasse principle holds for X whenever

n > (2 + E)(D − 1)2 D−1 + E2 E−1 .

In the case of one quadratic and one cubic we find that n ≥ 37 suffices,
which is superseded by the result of Browning, Dietmann and Heath-Brown
[15, Thm. 1.3].
Second, suppose that V ⊆ Pm is a geometrically integral, non-singular and
projective variety over Q. Then V satisfies the Hasse principle provided only
that
dim(V )  (deg(V ) − 1)2deg(V ) .

Note that when V is a hypersurface this result reduces to Theorem 4.5 over
Q. It is worth emphasising here that our hypotheses make no reference to the
shape of the defining equations for V . In particular, V is not required to be a
complete intersection!

Complete intersections in biprojective space

Birch’s theorem is concerned with complete intersections embedded in projec-
tive space Pn−1 . As formalised by Peyre [56], one expects that the methods
can be generalised to handle varieties which are embedded inside more
102 T.D. Browning

general toric varieties. Schindler [60] has carried out this plan for complete
intersections in Pm−1 × Pn−1 . Whereas Birch works with forms of total degree
d and applies (4.4) d − 1 times to obtain linear exponential sums, Schindler
used the structure of bihomogeneous forms Fi (x; y) of bidegree (d1 , d2 ) to
obtain results in which fewer variables are needed than would arise through
the most naive adaptation of Birch’s work. Indeed, Schindler profits by observ-
ing that it suffices to apply the Weyl differencing process d1 − 1 times for the
x-variables and d2 − 1 times for the y-variables, leading to a total of only
d1 + d2 − 2 differencing steps.

Question 4.7. Can one generalise Birch’s theorem to complete intersections

embedded in weighted projective space P(a1 , . . . , an ), where a1 , . . . , an are
pairwise coprime positive integers?

Linear spaces on complete intersections

So far we have been concerned with the Hasse principle for complete intersec-
tions X ⊂ Pn−1 over number fields k. The Hasse principle is concerned with
the existence of k-points on X , or equivalently, the k-rational linear spaces of
dimension 0 contained in X . It is natural to ask whether a local–global prin-
ciple is valid for k-rational linear spaces of arbitrary given dimension m  1.
Brandes [6] has shown how Birch’s result can be adapted to handle this kind
of question when n is sufficiently large.

Complete intersections over Fq [t]

One can also ask whether the circle method carries over to other global fields.
Suppose that X d,s ⊂ Pn−1 is defined over the function field K = Fq [t]. It
follows from the Lang–Tsen theorem that X d,s (K ) = ∅ whenever n > sd 2 .
Thus the interest here lies with the quantitative distribution of K -points on X.
Lee [48] has shown that the proof of Theorem 4.5 can be extended to address
this question in the function field setting.

4.2 The circle method and diagonalisation

In this survey we have concentrated on varieties cut out by general polyno-
mials. When the underlying equations have extra additive structure (such as
being diagonal) one can usually obtain much sharper results. In this section
we content ourselves with summarising what is known for diagonal cubic
hypersurfaces X ⊂ Pn−1 defined over a number field k by
c1 x13 + · · · + cn xn3 = 0, (4.7)
 A survey of applications of the circle method to rational points 103

for c1 , . . . , cn ∈ k ∗ . We would like conditions on n and the coefficients under

which the Hasse principle holds for X . When k = Q this problem was solved
for n  7, for any non-zero coefficients, by Baker [2] (using the version of the
circle method developed by Vaughan [70]). Using completely different meth-
ods, when n = 4 and k is general (but doesn’t contain the primitive cube roots
of unity), it follows from work of Swinnerton-Dyer [69] that the Hasse princi-
ple holds, subject to (very mild) conditions on the coefficients and the (rather
strong) assumption that the Tate–Shafarevich group of every relevant ellip-
tic curve is finite. Using the fibration method, moreover, Colliot-Thélène,
Kanevsky and Sansuc [23, Prop. 7] have shown that in order to deduce that
the Hasse principle holds for (4.7) for k = Q and any n  5, it suffices to
show that the Brauer–Manin obstruction is the only obstruction to the Hasse
principle for k = Q and n = 4.
As pioneered by Birch [3], it turns out that for any general system of poly-
nomial equations over a number field k one can carry out a diagonalisation
process to reduce its analysis to that of an appropriate system of diagonal equa-
tions. This idea was refined substantially by Wooley [73], in such a way as to
incorporate ideas of Lewis and Schulze-Pillot [50], enabling him to avoid the
previous inductive approach. Given positive integers d, s, let μd,s (k) denote
the least integer (if any such integer exists) with the property that whenever
n > μd,s (k), and f 1 , . . . , f s ∈ k[x1 , . . . , xn ] are forms of degree d, then the
variety f 1 = · · · = f s = 0 has a k-point. If no such integer exists, we define
μd,s (k) = ∞. Then it follows from [73, Thms. 1 and 2] that
6
μ3,s (Q) < (90s)8 log5 (27s) and μ5,s (Q) = o(es ).
The bounds for μd,s (Q) for odd d > 5 are much weaker.
As we saw in Section 4.1.3, a great deal of work has been invested in under-
standing μ(k) = μ3,1 (k) in the case (d, s) = (3, 1) of cubic hypersurfaces.
The best bounds that we have are μ(k) < 16 for any number field k (Pleasants
[57]) and μ(Q) < 14 (Heath-Brown [40]).

Question 4.8. Can one (conditionally) reduce the bound for μ(k) by com-
bining an efficient diagonalisation argument with the work of Swinnerton-
Dyer [69]?

4.3 The circle method and descent

4.3.1 The Brauer–Manin obstruction
We now recall the Brauer–Manin obstruction to the Hasse principle, as
introduced by Manin in 1970 using class field theory. Let X be a variety over a
104 T.D. Browning

number field k that is geometrically integral, non-singular and projective. Let

Br(X ) = Hét2 (X, Gm ) be the associated Brauer group. For any place v ∈ k let

invv : Br(kv ) → Q/Z

be the invariant map from local class field theory. For any field K ⊇ k, each
A ∈ Br(X ) gives rise to an evaluation map evA : X (K ) → Br(K ). Global
class field theory (the Hasse reciprocity law) gives the exact sequence
 
v invv
0 −→ Br(k) −→ Br(kv ) −−−−→ Q/Z −→ 0. (4.8)
v

We then have the commutative diagram


where X (Ak ) = v X (kv ). Let #A : X (Ak ) → Q/Z denote the composed
map. Then it follows that the image of X (k) in X (Ak ) under the diagonal
embedding is contained in ker #A for all A ∈ Br(X ). We call
3
X (Ak )Br = ker #A
A ∈Br(X)

the Brauer set. This gives an obstruction to the Hasse principle, since if
X (Ak )Br is empty we cannot expect X to have k-rational points. We say that
the Brauer–Manin obstruction to the Hasse principle is the only one if
X (Ak )Br = ∅ implies that X (k) = ∅.
In seeking to determine whether there is a Brauer–Manin obstruction it is
enough to consider elements A belonging to the quotient Br(X )/Br(k). Indeed
it follows from the exact sequence (4.8) that elements of Br(k) are orthogonal
to any adelic point under the pairing X (Ak ) × Br(X ) → Q/Z.

Example 4.9. Taking up the example of generalised Châtelet surfaces X

from Example 4.1, let p(T ) be any monic irreducible factor of f (T ). If a
is a square in k p = k[T ]/( p(T )) then X is k-rational and it follows that
Br(X )/Br(k) = 0. Let n denote the number of monic irreducible factors p(T )
of f (T ) such that a ∈ k 2p . Then Br(X )/Br(k) ∼
= (Z/2Z)n−1 if either all such
p(T ) have even degree and deg f (T ) is even or else if deg f (T ) is odd. Alter-
natively, if deg f (T ) is even, but some p(T ) with a ∈ k 2p has odd degree, then
Br(X )/Br(k) ∼ = (Z/2Z)n−2 .
 A survey of applications of the circle method to rational points 105

In particular, invv (evA (Mv )) takes the values 0 or 1/2 in Q/Z, for any
(Mv ) ∈ X (Ak ). An obstruction to the Hasse principle arises if and only if
X (Ak )Br = ∅, that is to say, if and only if for any (Mv ) ∈ X (Ak ) there exists
A ∈ Br(X ) such that
1
invv (evA (Mv )) = . (4.9)
v
2

In order to obtain counter-examples to the Hasse principle, it will be neces-

sary for invv (evA (Mv )) to be constant for every valuation v. Indeed, if there
is a valuation v for which invv (evA (Mv )) takes both values 0 and 1/2, as a
function of Mv , then (4.9) is clearly impossible. This can be used to explain the
failure of the Hasse principle for the surfaces X c /Q considered in Example 4.1,
where Br(X c )/Br(Q) ∼ = Z/2Z and a generator is given by the quaternion
algebra (−1, T 2 − c + 1) (see [21, §5] for details).

The main conjecture in this field is the following one due to Colliot-Thélène
(see [18]).

Conjecture 4.10. The Brauer–Manin obstruction is the only obstruction to

the Hasse principle for any geometrically integral, non-singular and projective
variety which is geometrically rationally connected.

The class of geometrically rationally connected varieties covers all Fano

varieties and all generalised Châtelet surfaces. It would appear that the circle
method has not yet been made to say anything significant about varieties which
are not geometrically rationally connected.

4.3.2 Descent varieties

A powerful and very general descent machinery has been developed by Colliot-
Thélène and Sansuc [20]. In favourable circumstances, this allows one to
reduce the task of establishing Conjecture 4.10 to proving the Hasse principle
for some associated descent varieties. We refer the reader to Skorobogatov
[67] for a general account of the theory. A fruitful development in the last
decade has been the use of the circle method to analyse these descent varieties.
It is this aspect that we proceed to discuss here.
One particularly rich seam of varieties arises through the Diophantine
equations
P(t) = N K /k (x1 , . . . , xn ), (4.10)
106 T.D. Browning

defined over a number field k, where P(t) ∈ k[t] is a polynomial and N K /k

is a norm form associated to an arbitrary degree n extension K /k (as in
Example 4.2). Let X be a smooth and projective model of the affine variety
in An+1 that (4.10) defines. Then X is geometrically rational and so is covered
by Conjecture 4.10. Note that when K /k is quadratic this is an example of the
generalised Châtelet surface that we met in Example 4.1. In particular we have
already seen that X need not satisfy the Hasse principle in general.
It turns out that understanding the Hasse principle for X becomes harder
as the number of distinct roots of P(t) grows larger. When P(t) has at most
one distinct root (so that either P(t) is constant or else P(t) = ct d for some
c ∈ k × and d ∈ Z>0 ), the affine variety (4.10) is a principal homogeneous
space for an algebraic k-torus, and the work of Sansuc [59] and Voskresenskiı̆
[72] establishes Conjecture 4.10 for X . When P(t) has precisely two distinct
roots, both of which are defined over k, it can be written
P(t) = ct d1 (t − 1)d2 , (4.11)
after a possible change of variables. In this case there are obvious rational
points on the affine variety (4.10), coming from the roots of P(t). However,
these will be singular as soon as min{d1 , d2 } > 1. When P(t) is given by
(4.11), it turns out that the descent varieties involve equations of the shape
(4.6), which can be analysed using the circle method. This programme was
first carried out by Heath-Brown and Skorobogatov [43] when k = Q and
gcd(n, d1 , d2 ) = 1, but both of these conditions have subsequently been
removed (see [22, 62, 68]). For other work on the arithmetic of (4.10), together
with a survey of the literature, we refer the reader to Derenthal, Smeets and Wei
[30] and the references therein.
We proceed to record some descent varieties associated to two particular
families of varieties. Example 4.11 has recently been solved using additive
combinatorics (see Section 4.3.3). Example 4.12, on the other hand, is still
wide open.

Example 4.11. Working over Q, suppose that P takes the form


r
P(t) = c (t − ei )m i ,
i=1

for c ∈ Q∗ , m 1 , . . . , m r ∈ Z>0 and pairwise distinct e1 , . . . , er ∈ Q. This

generalises the example (4.11) that we met above. Let π be the projection of
the affine variety given by (4.10) to A1 , and let U0 ⊂ A1 be the open subset

on which ri=1 (t − ei ) = 0. Let U = π −1 (U0 ) and let T = R 1K /Q be the
torus given by the affine equation N K /Q (x) = 1. It is possible to construct
 A survey of applications of the circle method to rational points 107

a partial compactification Y of U and show that vertical torsors T → Y

exist (see Colliot-Thélène, Harari and Skorobogatov [22] and Schindler and
Skorobogatov [62, §2]). These are torsors T → Y whose type is the injective
map of Gal(Q/Q)-modules T !r → Pic(Y ). It follows from [62, Lemma 2.2]
that the restriction TU of T to U ⊂ Y is E × V , where E is a principal
homogeneous space for T and V ⊂ Anr +1 is defined by
t − ei = λi N K /Q (xi ) = 0, (1  i  r ),
for λ1 , . . . , λr ∈ Q∗ . This relies crucially on the local description of torsors
laid down by Colliot-Thélène and Sansuc [20, Thm. 2.3.1]. Finally, it follows
from [62, Thm. 2.1] that in order to establish Conjecture 4.10 for a smooth
and projective model of (4.10), it suffices to show that V satisfies the Hasse
principle. But V is isomorphic to the variety cut out by the system of equations
e1 − ei = λi N K /Q (xi ) − λ1 N K /Q (x1 ), (2  i  r ).
By an obvious change of variables it suffices to establish the Hasse principle
for the variety in Anr +r defined by the system of equations
0 = (e1 − ei )N K /Q (y) = λi N K /Q (xi ) − λ1 N K /Q (x1 ), (2  i  r ).
But this variety is isomorphic to the non-singular variety W ⊂ An(r+1)+2
given by
v = N K /Q (y) = 0,
(4.12)
u − ei v = λi N K /Q (xi ) = 0, (1  i  r ).
We shall return to this example in Section 4.3.3.

Example 4.12. Suppose that X ⊂ P3 is a non-singular diagonal cubic surface

defined over k, with equation
a1 x13 + · · · + a4 x 43 = 0,
for a1 , . . . , a4 ∈ k ∗ . When k = Q, Colliot-Thélène, Kanevsky and Sansuc [23]
have undertaken an extensive investigation of the Brauer–Manin obstruction
for X . Assuming without loss of generality that a1 , . . . , a4 ∈ Z are cube-
free, it follows from this work that the Brauer–Manin obstruction to the Hasse
principle is empty if there is a prime p which divides precisely one of the
coefficients.
It is still an open problem to establish Conjecture 4.10 unconditionally for
diagonal cubic surfaces X . One approach would involve establishing the Hasse
principle for some associated intermediate torsors (see [23, Prop. 10]). These
possess a Zariski open subset which is Q-isomorphic to an intersection of two
108 T.D. Browning

cubics in A9 involving norm forms (see [23, §10] for details). Over Q these
take the shape
ax1 x2 x3 + by1 y2 y3 + cz 1 z 2 z 3 = 0
d x1 y1 z 1 + ex2 y2 z 2 + f x3 y3 z 3 = 0,

for non-zero a, . . . , f ∈ Q. Alas this system is beyond the square-root bar-

rier (4.5) and so it seems unlikely that the circle method could be used to say
anything useful in its present form.

Let C denote the family of all systems of Diophantine equations over the
ring of integers ok of a number field k, for which the circle method has proved
the Hasse principle for the system. Let T denote the family of varieties which
are either affine space or a principal homogeneous space for a torus.

Question 4.13. What is the most general class of geometrically integral, non-
singular and projective varieties, for which universal torsors exist and are
k-birational to a product of varieties from C and T ?

In Example 4.11, according to [22] and [62], any smooth and projective
model of (4.10) with P(t) = ct d1 (t − 1)d2 belongs to the class of vari-
eties considered in Question 4.13. Indeed, the universal torsor in this case is
k-birational to a product of V and a principal homogeneous space for the torus
N K /k (x) = 1.

4.3.3 The nilpotent circle method

Let h : Z → R be an arithmetic function and let L 1 , . . . , L r ∈ Z[u, v] be
a system of binary linear forms, no two of which are proportional. A pow-
erful new development concerns the ability (at least in principle) to produce
asymptotic formulae for sums of the shape
h(L 1 (u, v)) . . . h(L r (u, v)), (4.13)
(u,v)∈Z2 ∩BB

as B → ∞, for any bounded convex region B ⊂ R2 . The underlying tech-

nology comes from additive combinatorics and is due to Green and Tao [33]
and Green, Tao and Ziegler [35], where it is used to tackle this problem when
h =  is the von Mangoldt function. The term nilpotent circle method orig-
inates from [34] and refers to the methods and concepts that were developed
in the course of this work.
So far, the analysis of (4.13) has been carried out for a fairly restrictive fam-
ily of arithmetic functions h. Matthiesen [52, 53] has done so when h(N ) is the
 A survey of applications of the circle method to rational points 109

function which counts the number of (inequivalent) representations of N ∈ Z

by a given binary quadratic form. It was noticed by Browning, Matthiesen and
Skorobogatov [16] that, once unravelled in terms of counting solutions to a
system of underlying Diophantine equations, this is exactly what is needed to
establish the Hasse principle for the variety W in (4.12), in the special case
n = 2 of quadratic extensions K /Q. This therefore yields a proof of Conjec-
ture 4.10 over Q, for the generalised Châtelet surfaces considered in Examples
4.1 and 4.9, provided that the polynomial f (T ) decomposes completely as a
product of linear polynomials with coefficients in Q. The significance of this is
that there was previously no unconditional proof of Conjecture 4.10 for gener-
alised Châtelet surfaces in which f (T ) has r > 4 distinct roots (the case r  4
is covered by [24, 25]).
Very recently, Browning and Matthiesen [11] have used the nilpotent circle
method to generalise this considerably. Here the Hasse principle is established
for the variety W in (4.12), without any restriction on the degree or type of
the number field K (note that when K /Q is cyclic, Harpaz, Skorobogatov and
Wittenberg [37] have shown how to arrive at the same conclusion by combin-
ing the fibration method with the work of Green and Tao [33] and Green, Tao
and Ziegler [35] on the generalised Hardy–Littlewood primes conjecture). A
consequence of this is that Conjecture 4.10 holds for the varieties X considered
in Example 4.11 for any r  2 (generalising the case r = 2 handled in [43]).
Using descent, Skorobogatov [66] has shown how the main result in [11] can
be used to study the Diophantine equation (4.10) when the norm form on the
right hand side is replaced by a product of norm forms.
Eliminating u, v from (4.12) the variety W is seen to be Q-birational to the
system of equations

N K /Q (xi ) + bi N K /Q (y) = ci N K /Q (z), (1 ≤ i ≤ r − 1),

for suitable bi , ci ∈ Q. This can be viewed as a variety in P(r+1)n−1 cut out by

r − 1 hypersurfaces of degree n. The main result in [11] shows that the smooth
Hasse principle holds for this variety. By contrast, the inequality (4.5) becomes
(r + 1)n > 2n(r − 1), which is true if and only if 3 > r . The significance of
[11], therefore, is that it goes beyond the square-root barrier.

References
[1] A.G. Aleksandrov and B.Z. Moroz. Complete intersections in relation to a paper
of B. J. Birch. Bull. London Math. Soc. 34 (2002), 149–154.
[2] R.C. Baker, Diagonal cubic equations, II. Acta Arith. 53 (1989) 217–250.
[3] B.J. Birch, Homogeneous forms of odd degree in a large number of variables.
Mathematika 4 (1957), 102–105.
110 T.D. Browning

[4] B.J. Birch, Forms in many variables. Proc. Roy. Soc. Ser. A 265 (1961), 245–263.
[5] B.J. Birch, H. Davenport and D.J. Lewis, The addition of norm forms. Mathe-
matika 9 (1962), 75–82.
[6] J. Brandes, Forms representing forms and linear spaces on hypersurfaces. Proc.
London Math. Soc. 108 (2014), 809–835.
[7] R. Brauer, A note on systems of homogeneous algebraic equations. Bull. Amer.
Math. Soc. 51 (1945), 749–755.
[8] T.D. Browning, Rational points on cubic hypersurfaces that split off a form.
Compositio Math. 146 (2010), 853–885.
[9] T.D. Browning and D.R. Heath-Brown, Rational points on quartic hypersur-
faces. J. reine angew. Math. 629 (2009), 37–88.
[10] T.D. Browning and D.R. Heath-Brown, Forms in many variables and differing
degrees. J. Eur. Math. Soc., to appear.
[11] T.D. Browning and L. Matthiesen, Norm forms for arbitrary number fields as
products of linear polynomials. Submitted, 2013. (arXiv:1307.7641)
[12] T.D. Browning and R. Munshi, Rational points on singular intersections of
quadrics. Compositio Math. 149 (2013), 1457–1494.
[13] T.D. Browning and S. Prendiville, Improvements in Birch’s theorem on forms
in many variables. J. reine angew. Math., to appear.
[14] T.D. Browning and P. Vishe, Cubic hypersurfaces and a version of the circle
method for number fields. Duke Math. J. 183 (2014), 1825–1883.
[15] T.D. Browning, R. Dietmann, D.R. Heath-Brown, Rational points on intersec-
tions of cubic and quadric hypersurfaces. J. Inst. Math. Jussieu, to appear.
[16] T.D. Browning, L. Matthiesen and A.N. Skorobogatov, Rational points on pen-
cils of conics and quadrics with many degenerate fibres. Annals of Math. 180
(2014), 381–402.
[17] J.W.S. Cassels, W. Ledermann and K. Mahler, Farey section in k(i) and k($).
Philos. Trans. Roy. Soc. London. Ser. A. 243 (1951), 585–626.
[18] J.-L. Colliot-Thélène. Points rationnels sur les fibrations. Higher dimensional
varieties and rational points (Budapest, 2001), 171–221, Springer-Verlag, 2003.
[19] J.-L. Colliot-Thélène and P. Salberger, Arithmetic on some singular cubic
hypersurfaces. Proc. London Math. Soc. 58 (1989), 519–549.
[20] J.-L. Colliot-Thélène and J.-J. Sansuc, La descente sur les variétés rationnelles,
II. Duke Math. J. 54 (1987), 375–492.
[21] J.-L. Colliot-Thélène, D. Coray and J.-J. Sansuc, Descente et principe de Hasse
pour certaines variétés rationnelles. J. reine angew. Math. 320 (1980), 150–191.
[22] J.-L. Colliot-Thélène, D. Harari and A.N. Skorobogatov, Valeurs d’un polynôme
à une variable représentés par une norme. Number theory and algebraic
geometry, 69–89, London Math. Soc. Lecture Note Ser. 303 Cambridge Uni-
versity Press, 2003.
[23] J.-L. Colliot-Thélène, D. Kanevsky and J.-J. Sansuc, Arithmétique des surfaces
cubiques diagonales. Diophantine approximation and transcendence theory,
1–108, Springer Lecture Notes Math. 1290 Springer-Verlag, 1987.
[24] J.-L. Colliot-Thélène, J.-J. Sansuc and P. Swinnerton-Dyer, Intersections of
two quadrics and Châtelet surfaces, I. J. reine angew. Math. 373 (1987),
37–107.
 A survey of applications of the circle method to rational points 111

[25] J.-L. Colliot-Thélène, J.-J. Sansuc and P. Swinnerton-Dyer, Intersections of two

quadrics and Châtelet surfaces, II. J. reine angew. Math. 374 (1987), 72–168.
[26] H. Davenport, Cubic forms in 16 variables. Proc. Roy. Soc. A 272 (1963),
285–303.
[27] H. Davenport, Analytic methods for Diophantine equations and Diophantine
inequalities. 2nd ed., edited by T.D. Browning, Cambridge University Press,
2005.
[28] P. Deligne, La conjecture de Weil, I. Inst. Hautes Études Sci. Publ. Math. 43
(1974), 273–307.
[29] V.B. Dem’yanov, On cubic forms in discretely normed fields. (Russian) Doklady
Akad. Nauk SSSR 74 (1950), 888–891.
[30] U. Derenthal, A. Smeets and D. Wei, Universal torsors and values of quadratic
polynomials represented by norms. Math. Annalen, to appear.
[31] R. Dietmann, Weyl’s inequality and systems of forms, Quart. J. Math. 66 (2015),
97–110.
[32] W. Duke, J.B. Friedlander and H. Iwaniec, Bounds for automorphic L-functions.
Invent. Math. 112 (1993), 1–8.
[33] B. Green and T. Tao, Linear equations in primes. Annals of Math. 171 (2010),
1753–1850.
[34] B. Green and T. Tao, The Möbius function is strongly orthogonal to nilse-
quences. Annals of Math. 175 (2012), 541–566.
[35] B. Green, T. Tao and T. Ziegler, An inverse theorem for the Gowers Us+1 [N ]-
norm. Annals of Math. 176 (2012), 1231–1372.
[36] M. Hanselmann, Rational points on quartic hypersurfaces. Ph.D. thesis, Ludwig
Maximilians Universität Munchen, 2012.
[37] J. Harpaz, A.N. Skorobogatov and O. Wittenberg, The Hardy–Littlewood
conjecture and rational points. Compositio Math. 150 (2014), 2095–2111.
[38] D.R. Heath-Brown, Cubic forms in ten variables. Proc. London Math. Soc. 47
(1983), 225–257.
[39] D.R. Heath-Brown, A new form of the circle method, and its application to
quadratic forms. J. reine angew. Math. 481 (1996), 149–206.
[40] D.R. Heath-Brown, Cubic forms in 14 variables. Invent. Math. 170 (2007),
199–230.
[41] D.R. Heath-Brown, Artin’s conjecture on zeros of p-adic forms. Proceedings
of the International Congress of Mathematicians. Vol. II, 249–257, Hindustan
Book Agency, New Delhi, 2010.
[42] D.R. Heath-Brown, Zeros of pairs of quadratic forms. Submitted, 2013.
(arXiv:1304.3894)
[43] D.R. Heath-Brown and A.N. Skorobogatov, Rational solutions of certain equa-
tions involving norms. Acta Math. 189 (2002), 161–177.
[44] C. Hooley, On nonary cubic forms. J. reine angew. Math. 386 (1988), 32–98.
[45] C. Hooley, On nonary cubic forms: IV. J. reine angew. Math. 680 (2013), 23–39.
[46] C. Hooley, On octonary cubic forms. Proc. London Math. Soc. 109 (2014),
241–281.
[47] H.D. Kloosterman, On the representation of numbers in the form ax 2 + by 2 +
cz 2 + dt 2 . Acta. Math. 49 (1926), 407–464.
112 T.D. Browning

[48] S.-L. A. Lee, Birch’s theorem in function fields. Submitted, 2012.

(arXiv:1109.4953)
[49] D.J. Lewis, Cubic homogeneous polynomials over p-adic fields. Annals of Math.
56 (1952), 473–478.
[50] D.J. Lewis and R. Schulze-Pillot, Linear spaces on the intersection of cubic
hypersurfaces. Monatsh. Math. 97 (1984), 277–285.
[51] Y.I. Manin, Le groupe de Brauer–Grothendieck en géométrie diophantienne.
Actes du Congrès International des Mathématiciens (Nice, 1970). Tome 1,
401–411, Gauthier-Villars, Paris, 1971.
[52] L. Matthiesen, Linear correlations amongst numbers represented by positive
definite binary quadratic forms. Acta Arith. 154 (2012), 235–306.
[53] L. Matthiesen, Correlations of representation functions of binary quadratic
forms. Acta Arith. 158 (2013), 245–252.
[54] R. Munshi, Pairs of quadrics in 11 variables. Compositio Math., to appear.
[55] E. Peyre, Hauteurs et mesures de Tamagawa sur les variétiés de Fano. Duke
Math. J. 79 (1995), 101–218.
[56] E. Peyre, Torseurs universels et méthode du cercle. Rational points on algebraic
varieties, 221–274, Progr. Math. 199, Birhäuser, 2001.
[57] P.A.B. Pleasants, Cubic polynomials over algebraic number fields. J. Number
Theory 7 (1975), 310–344.
[58] B. Poonen and J.F. Voloch, Random Diophantine equations. Arithmetic of
higher-dimensional algebraic varieties (Palo Alto, CA, 2002), 175–184, Progr.
Math. 226, Birkhäuser, 2004.
[59] J.-J. Sansuc, Groupe de Brauer et arithmétique des groupes algébriques linéaires
sur un corps de nombres. J. reine angew. Math. 327 (1981), 12–80.
[60] D. Schindler, Bihomogeneous forms in many variables. J. Théorie Nombres
Bordeaux 26 (2014), 483–506.
[61] D. Schindler, A variant of Weyl’s inequality for systems of forms and applica-
tions. To be submitted, arXiv:1403.7156.
[62] D. Schindler and A.N. Skorobogatov, Norms as products of linear polynomials.
J. London Math. Soc. 89 (2014), 559–580.
[63] W. Schmidt, The density of integer points on homogeneous varieties. Acta Math.
154 (1985), 243–296.
[64] C.M. Skinner, Rational points on nonsingular cubic hypersurfaces. Duke Math.
J. 75 (1994), 409–466.
[65] C.M. Skinner, Forms over number fields and weak approximation. Compositio
Math. 106 (1997), no. 1, 11–29.
[66] A.N. Skorobogatov, Descent on toric fibrations. This volume, pp. 422–435.
[67] A.N. Skorobogatov, Torsors and rational points. Cambridge University Press,
2001.
[68] M. Swarbrick Jones, A note on a theorem of Heath-Brown and Skorobogatov.
Quart. J. Math. 64 (2013), 1239–1251.
[69] P. Swinnerton-Dyer, The solubility of diagonal cubic surfaces. Annales Sci.
École Normale Sup. 34 (2001), 891–912.
[70] R.C. Vaughan, On Waring’s problem for cubes. J. reine angew. Math. 365 (1986)
122–170.
 A survey of applications of the circle method to rational points 113

[71] R.C. Vaughan, The Hardy-Littlewood method. 2nd ed., Cambridge Tracts in
Mathematics 125, Cambridge University Press, 1997.
[72] V.E. Voskresenskiı̆, Algebraic groups and their birational invariants. Transla-
tions of Math. Monographs 179, American Math. Soc., 1998.
[73] T.D. Wooley, Forms in many variables. Analytic number theory (Kyoto, 1996),
361–376, London Math. Soc. Lecture Note Ser. 247, Cambridge University
Press, 1997.
[74] T.D. Wooley, On the local solubility of Diophantine systems. Compositio Math.
111 (1998), 149–165.
[75] B. Xue and H. Dai, Rational points on cubic hypersurfaces that split off two
forms. Bull. London Math. Soc. 46 (2014), 169–184.
 5
Arithmetic differential equations of
Painlevé VI type
Alexandru Buium1 and Yuri I. Manin2
1 University of New Mexico, Albuquerque
2 Max–Planck–Institut für Mathematik, Bonn, Germany

A BSTRACT. Using the description of Painlevé VI family of differential equa-

tions in terms of a universal elliptic curve, going back to R. Fuchs (cf. [Ma96]),
we translate it into the realm of Arithmetic Differential Equations (cf. [Bu05]),
where the role of derivative “in the p-adic direction” is played by a version of
Fermat quotient.

Introduction and brief summary

This article is dedicated to the study of differential equations of the Painlevé
VI type “with p-adic time”, in which the role of time derivative is played
by a version of p-Fermat quotient. Richness of p-adic differential geometry
was already demonstrated in many papers of the first-named author: see in
particular the monograph [Bu05].
Applicability of this technique to the Painlevé equations is ensured by the
combination of two approaches: R. Fuchs’s treatment of the classical case
modernized in [Ma96], and constructions of p-adic differential characters in
[Bu95].
In sec. 5.1 below we present a short introduction to the p-adic differential
geometry. In sec. 5.2 and 5.3 we introduce several versions of p-adic PVI.
Sec. 5.4 is dedicated to the problem of transposing into p-adic domain of main
features of Hamiltonian formalism. We have not found a definitive answer to
this problem, and Painlevé equations serve here as a very stimulating testing
ground. Finally, sec. 5.5 treats another problem, suggested by comparison of
several versions of p-adic PVI, but presenting an independent interest.

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

114
 Arithmetic differential equations of Painlevé VI type 115

Acknowledgement. This article was conceived during the Spring 2013

Trimester program “Arithmetic and Geometry” of the Hausdorff Institute for
Mathematics (HIM), Bonn. The authors are grateful to HIM for a stimulating
atmosphere and working conditions.

5.1 Arithmetic differential equations: background

We start with a brief summary of relevant material from [Bu05], Chapters 2, 3.
Fix a prime p; in our applications it will be assumed that p ≥ 5.

5.1.1 p-derivations
Recall that, in the conventional commutative algebra, given a ring A and an
A-module N , a derivation of A with values in N is any map ∂ : A → N
such that the map A → A × N : a → (a, ∂a) is a ring homomorphism, where
A × N is endowed with the structure of commutative ring with componentwise
addition, and multiplication (a, m) · (b, n) := (ab, an + bm). Notice that {0} ×
N is the ideal with square zero in A × N .
Similarly, in arithmetic geometry a p-derivation of A with values in an A-
algebra B, f : A → B, is a map δ p : A → B such that the map A → B × B :
a → ( f (a), δ p (a)) is a ring homomorphism A → W2 (B) where W2 (B) is the
ring of p-typical Witt vectors of length 2. Again, if p B = {0}, Witt vectors of
the form (0, b) form the ideal of square zero.
Explicitly, this means that δ p (1) = 0, and
δ p (x + y) = δ p (x) + δ p (y) + C p (x, y), (5.1)
δ p (x y) = f (x) · δ p (y) + f (y) · δ p (x) + p · δ p (x) · δ p (y),
p p
(5.2)
where
X p + Y p − (X + Y ) p
C p (X, Y ) := ∈ Z[X, Y ]. (5.3)
p
In particular, this implies that for any p-derivation δ p : A → B the respective
map φ p : A → B defined by φ p (a) := f (a) p + pδ p (a) is a ring homomor-
phism satisfying φ p (x) ≡ f (x) p mod p, that is “a lift of the Frobenius map
applied to f ”.
Conversely, having such a lift of Frobenius, we can uniquely reconstruct the
respective derivation δ p under the condition that B has no p-torsion.
We will often work with p-derivations A → A with respect to the identity
map A → A and keep p fixed; then it might be kept off the notation. Such
a pair (A, δ) is called a δ-ring. Morphisms of δ-rings are algebra morphisms
compatible with their p-derivations.
116 Alexandru Buium and Yuri I. Manin

Moreover, our rings (and more generally, schemes) will be R-algebras

where R = W (k) (ring of infinite p-typical Witt vectors) is the comple-
tion of the maximal unramified extension of Z p , with residue field k := an
algebraic closure of Z/ pZ. By φ : R → R we denote the automorphism
acting as Frobenius x → x p on k, and by δ the respective p-derivation:
δ(x) = (φ(x) − x p )/ p. The R-algebra structure on a δ-ring is always assumed
to be compatible with this p-derivation.

5.1.2 Prolongation sequences and p-jet spaces

In the classical situation invoked in 1.1, there exists an universal derivation
d : A → 1 ( A) (5.4)
with values in the A-module of differentials.
For p-derivations, (5.4) might be replaced by the following construction
(however, see the Remark 5.1 below).
Let A be an R-algebra. A prolongation sequence for A consists of a family
of p-adically complete R-algebras Ai , i ≥ 0, where A0 = A! is the p-adic
completion of A, and of maps ϕi , δi : Ai → Ai+1 satisfying the following
conditions:
a) ϕi are ring homomorphisms, each δi is a p-derivation with respect to ϕi ,
compatible with δ on R.
b) δi ◦ ϕi−1 = ϕi ◦ δi −1 for all i ≥ 1.
Prolongation sequences form a category with evident morphisms, ring
homomorphisms f i : Ai → B i commuting with ϕi and δi , and in its
subcategory with fixed A0 there exists an initial element, defined up to
unique isomorphism (cf. [Bu05], Chapter 3). It can be called the universal
prolongation sequence.
In the geometric language, if X = Spec A, the formal spectrum of the i-th
ring Ai in the universal prolongation sequence is denoted J i (X ) and called
the i-th p-jet space of X . Conversely, Ai = O(J i (X )), the ring of global
functions.
The geometric morphisms (of formal schemes over Z) corresponding to φi
are denoted φ i : J i (X ) → J 0 (X ) =: X!(formal p-adic completion of X ).
This construction is compatible with localisation so that it can be applied to
the non-necessarily affine schemes: cf. [Bu05], Chapter 3.

Remark 5.1. Classically, (5.4) extends to the universal map of A to the dif-
ferential graded algebra ∗ (A), and there is a superficial similarity of this
 Arithmetic differential equations of Painlevé VI type 117

map with the one, say of A to the inductive limit of its universal prolongation
sequence in the p-adic arithmetics context.

However, the classical differential acts in Z2 -graded supercommutative

algebras and is an odd operator with square zero, whereas δ p are even.
The differential geometry of smooth schemes in characteristic p > 0 sug-
gests a perspective worth exploring. Namely, the sheaf of differential forms on
such a scheme is endowed with the so-called Cartier operator C, which is dual
to the Frobenius operator F : ∂ → ∂ p acting upon vector fields. This oper-
ator C is F −1 -linear. One could consider studying p-adic lifts of the Cartier
operator from the closed fibre of the relevant scheme to its p-adic comple-
tion, following the lead of [Bu05]. For a recent survey of F −1 -linear maps
cf. [BlSch12] and references therein.

5.1.3 Flows
Let now X be a smooth affine scheme over R = W (k). Each element of
O(J r (X )) induces a function f : X (R) → R. Such functions are called δ-
functions of order r on X , and we may and will identify them with respective
elements of O(J r (X )). For r = 0, O(J 0 (X )) = O(X )!, the p-adic completion
of O(X ).

Definition 5.2. a) A system of arithmetic differential equations of order r on

X is a subset E of O(J r (X )).
b) A solution of E is an R-point P ∈ X (R) such that f (P) = 0 for all f ∈ E.
The set of solutions of E is denoted Sol(E) ⊂ X (R).
c) A prime integral of E is a function H ∈ O(X )! such that δ(H(P)) = 0 for
all P ∈ Sol(E).

We will also denote by Z r (E) the closed formal subscheme of J r (X )

generated by E.
Now, let δ X be a p-derivation of O(X )!. From the universality of the jet
sequence explained in 1.2, it follows that such derivations are in a bijection
with the sections of the canonical morphism J 1 (X ) → J 0 (X ).

Definition 5.3. The δ-flow associated to δ X is the system of arithmetic dif-

ferential equations of order 1 which is the ideal in O(J 1 (X )) generated by
elements of the form δ f i −δ X f i where f i ∈ O(X ) generate O(X ) as R-algebra.

We use the word “flow” in this context in order to suggest that in our main
applications we consider the p-adic axis as an arithmetic version of the time
axis.
118 Alexandru Buium and Yuri I. Manin

The derivation δ X is completely determined by its δ-flow. If δ X corresponds

to the section s : J 0 (X ) → J 1 (X ) of J 1 (X ) → J 0 (X ) then Z (E(δ X )) ⊂
J 1 (X ) coincides with the image of ths section. One easily checks that if H ∈
O(X )! is such that δ X H = 0 then H is a prime integral for E(δ X ). All of
the above can be transposed to the case when X a p-formal scheme, locally a
p-adic completion of a smooth scheme over R.
In what follows we choose a smooth affine scheme Y and apply the con-
structions discussed above to X := J 1 (Y ). In this case one can define a special
class of δ-flows on J 1 (Y ) which will be called canonical δ-flows.

Definition 5.4. A canonical δ-flow is a δ-flow E(δ J 1 (Y ) ) with the property that
the composition of δ J 1 (Y ) : O(J 1 (Y )) → O(J 1 (Y )) with the pull back map
O(Y ) → O(J 1 (Y )) equals the universal p-derivation δ : O(Y ) → O(J 1 (Y )).

Notice that in view of the universality property of p-jet spaces, one gets a
natural closed embedding ι : J 2 (Y ) → J 1 (J 1 (Y )). This induces an injective
map (which we view as an identification) from the set of sections of J 2 (Y ) →
J 1 (Y ) to the set of sections of J 1 (J 1 (Y )) → J 1 (Y ). The sections of J 2 (Y ) →
J 1 (Y ) are in a natural bijection with canonical δ-flows on J 1 (Y ), whereas
the sections of J 1 (J 1 (Y )) → J 1 (Y ) are in a bijection with (not necessarily
canonical) δ-flows on J 1 (Y ).
Finally, consider a system of arithmetic differential equations of order 2,
F ⊂ O(J 2 (Y )).

Definition 5.5. F defines a δ-flow on J 1 (Y ) if the map Z 2 (F) → J 1 (Y ) is an

isomorphism.

In this case then Z 2 (F ) → J 1 (Y ) defines a section of J 2 (Y ) → J 1 (Y )

and hence a canonical δ-flow E(δ J 1 (Y ) ) on J 1 (Y ) such that ι(Z 2 (F)) =
Z 1 (E(δ J 1 (Y ) )).
The differential algebra counterpart of the above definition yields the natural
concept of flow on the (co)tangent space defining a second-order differential
equation.

5.2 Painlevé VI and differential characters of elliptic curves

5.2.1 Classical case
The family of sixth Painlevé equations depends on four arbitrary constants
(α, β, γ , δ) and is classically written as
 Arithmetic differential equations of Painlevé VI type 119

    
d2 X 1 1 1 1 dX 2 1 1 1 dX
= + + − + + +
dt 2 2 X X −1 X −t dt t t −1 X − t dt
 
X (X − 1)(X − t) t t −1 t (t − 1)
+ α+β 2 +γ +δ .
t (t − 1)
2 2 X (X − 1)2 (X − t)2
(5.5)

As R. Fuchs remarked in 1907, (5.5) can be rewritten as the differential equa-

tion for a (local) section P := (X (t), Y (t)) of the generic elliptic curve
E = E(t) : Y 2 = X (X − 1)(X − t):
  . (X,Y )
d2 d 1 dx
t (1 − t) t (1 − t) 2 + (1 − 2t) − √ =
dt dt 4 ∞ x(x − 1)(x − t)
tY (t − 1)Y 1 t (t − 1)Y
= αY + β 2 + γ + (δ − ) (5.6)
X (X − 1)2 2 (X − t)2
The l.h.s. of (5.6) can be called the additive differential character μ of order
two of E: it is a non-linear differential expression in coordinates of P such
that μ(P + Q) = μ(P) + μ(Q) where P + Q means addition of points of
the generic elliptic curve E, with infinity as zero. In particular, μ(Q) = 0 for
points of finite order. The point is that the integral in the l.h.s. of (5.6) already
has such additivity property, but it is defined only modulo periods, and the
latter are annihilated by the Gauss differential operator.
Thus μ(P) is defined up to multiplication by an invertible function of t.
If we choose a differential of the first kind ω on the generic curve and the
symbol of the Picard–Fuchs operator of the second-order annihilating periods
of ω, the character will be defined uniquely. In particular, if we pass to the
analytic picture replacing the algebraic family of curves E(t) by the analytic
one E τ := C/(Z + Zτ ) → τ ∈ H , and denote by z a fixed coordinate on C,
then (5.5) and (5.6) can be equivalently written in the form
3
d2z 1 Tj
= α j ℘z (z + , τ) (5.7)
dτ 2 (2πi )2 2
j =0

where (α0 , . . . , α3 ) := (α, −β, γ , 12 − δ) and

 
1 1 1
℘ (z, τ ) := + − . (5.8)
z2 (z + mτ + n)2 (mτ + n)2
(m,n) =(0.0)

Moreover, we have

℘z (z, τ )2 = 4(℘ (z, τ ) − e1 (τ ))(℘ (z, τ ) − e2 (τ ))(℘ (z, τ ) − e3 (τ )) (5.9)

120 Alexandru Buium and Yuri I. Manin

where
Ti
ei (τ ) = ℘ ( , τ ), (T0 , . . . , T3 ) = (0, 1, τ, 1 + τ ) (5.10)
2
so that e1 + e2 + e3 = 0.
For a more geometric description, cf. sec. 5.4.3 below.

5.2.2 p-adic differential additive characters

The form (5.7) is more suggestive than (5.5) for devising a p-adic version
of PVI. In purely algebraic terms, z can be described as the logarithm of the
formal group law, and the r.h.s. of (5.7) is simply a linear combination of shifts
of ℘z - (or Y -)coordinate by sections of the second order.
More precisely, using basic conventions of [Bu05], let p ≥ 5 be a prime, k
an algebraic closure of Z/ pZ, R = W (k) the ring of p-typical Witt vectors.,
as in 1.1.
Consider a smooth projective curve of genus one E over R, with four marked
and numbered R-sections Pi , i = 0, . . . , 3, such that all divisors 2(Pi − P j )
are principal ones. Choosing, say, P0 as zero section, we may and will iden-
tify E with its Jacobian and represent E as the closure of the affine curve
y 2 = 4x 3 + ax + b = 4(x − e1 )(x − e2 )(x − e3 ), with ei ∈ R, correspond-
ing to Pi − P0 . Assume that E is not a canonical lift of its (good) reduction,
that is, does not admit a lift of the Frobenius morphism of E ⊗ k.
Choose the differential ω := d x/y. Whenever we work with several elliptic
curves simultaneously, we will write x E,ω , y E,ω in place of former x, y etc.
The curve E has a canonical p-differential character ψ E ,ω of order 2
([Bu05], pp. 201 and 197), which corresponds to (2πi)2 d 2 z/dτ 2 in (5.7).

5.2.3 Painlevé p-adic equation and the problem of constants

Now we can directly write a p-adic version of (5.7) as
3
ψ E,ω (Q) = α j s ∗j (y(Q)) (5.11)
j =0

where Q is a variable section of E/R, and s j : E j → E j is the shift by P j .

At this point we have to mention two problems.
(A) In (5.7), α j must be absolute constants rather than, say, functions of t.
Directly imitating this condition, we have to postulate that in (5.11), α j must
be roots of unity or zero, i.e. δ p -constants. It is desirable to find a justification
of such a requirement (or a version of it) in a more extended p-adic theory of
 Arithmetic differential equations of Painlevé VI type 121

Painlevé VI, e.g. tracing its source to the arithmetic analog of isomonodromy
deformations.
(B) The relevance and non-triviality of the problem of “constants” in the
p-adic differential equations context is also implicit in our exclusion of those
E that are canonical lifts of their reductions.
A formal reason for this exclusion was the fact that for such E the basic
differential character ψ has order 1 rather than 2, thus being outside the
framework of Painlevé VI. But the analogy with functional case suggests
that canonical lifts should be morally considered as analogs of families with
constant j -invariant in the functional case.
This agrees also with J. Borger’s philosophy that Frobenius lift(s) should be
treated as descent data to an “algebraic geometry below Spec Z” (cf. [Bo09]).
In our particular case the latter might be called “geometry over p-typical field
with one element” (Borger’s suggestion in e-mail to Yu. M. of April 22, 2013).
Indeed, canonical lifts X in such a geometry are endowed with an isomor-
phism X φ → X ( p) that can be seen as a categorification of the identity cφ = c p
defining roots of unity.
In the p-adic case, however, if we decide to declare j-invariants of canon-
ical lifts “constants” in some sense, this will require a revision of the latter
notion. These invariants generally are not roots of unity: cf. Finotti’s papers in
http://www.math.utk.edu/finotti/.

5.3 Symmetries and variants

Lemma 5.6. (i) (Landin’s transform). In the notations of 2.3, denote for each
i = 1, 2, 3 by πi : E → E i := E/Pi  the respective isogeny. Let ωi be
the 1-form on E i such that π ∗ (ωi ) = ω. Then
πi∗ (y Ei ,ωi ) = y E,ω + si∗ (y E,ω ). (5.12)
(ii) We have
πi∗ (ψ Ei ,ωi ) = ψ E ,ω . (5.13)

Proof. (i) If we choose an embedding of R in C identifying E(C) with

C/(Z+Zτ ) in such a way that Pi becomes the point τ/2 (modulo periods),
then (5.12) turns into the classical Landin identity (cf. [Ma96], sec. 1.6):
℘ (z, τ/2) = ℘ (z, τ ) + ℘ (z + τ/2, τ )
(ii) The identity (5.13) can be stated and proved in wider generality. Namely,
let E, E  be two smooth elliptic curves over R, not admitting lifts of
122 Alexandru Buium and Yuri I. Manin

Frobenius. Let π : E → E  be an isogeny of degree prime to p and

and ω, ω such bases of 1-forms on E, E  that π ∗ (ω ) = ω. Then

π ∗ (ψ E  ,ω ) = ψ E,ω . (5.14)

In fact from [Bu05], Theorem 7.34, it follows that π ∗ (ψ E  ,ω ) = cψ E,ω

for some c ∈ R.
Now consider the second jet space J 2 (E). Then ψ = ψ E,ω is an element
of O(J 2 (E)) (cf. [Bu05], p. 201). Denote by φ i : J i (E) → J 0 (E) the map
defined at the end of sec. 5.1.2. Put
(φ i )∗ (ω)
ω(i ) := , i = 0, 1, 2. (5.15)
pi

Then in view of [Bu05], p. 203 (where our ω(i) were denoted ωi ), we have

dψ = pω(2) + λ1 ω(1) + λ0 ω(0) (5.16)

with λ1 ∈ R, λ1 ∈ R × depending on (E, ω). Notice that here d means

the usual differential taken in the “geometric”, or “vertical” direction, that is
dc = 0 for any c ∈ R. Moreover, in [Bu97] it was proved that if E is defined
over Z p ⊂ R, then λ0 = 1, λ1 = −a p where a p is the trace of Frobenius on
the reduction E mod p.
Returning to the proof of (5.14), we can now compare the ω(2) -contributions
to dψ E  ,ω in two different ways. First, we have

π ∗ (dψ E  ,ω ) = π ∗ ( p(ω )(2) + . . .) = p (π ∗ (ω )(2) ) + . . .

Second,

π ∗ (dψ E  ,ω ) = dπ ∗ (ψ E  ,ω ) = d(c ψ E,ω ) = c ( pω(2) + . . .).

This shows that c = 1.

5.3.1 Two versions of PVI


Put Y := E \ ∪3i=0 Pi . Denote by r := 3j=0 α j s ∗j (y) ∈ O(Y ) (cf. (5.11)).
Below we will denote by ρ either r , or φ(r ) ∈ O(J 1 (Y )). Although the
introduction of the version with φ(r ) is not motivated at this point, we will
see in sec. 5.4 below that exactly this version admits a “ p-adic Hamiltonian”
description.
The character ψ induces the map of sets, that we will also denote ψ :
E(R) → R; similarly, ρ induces the map of sets ρ : Y (R) → R.
 Arithmetic differential equations of Painlevé VI type 123

Proposition 5.7. Denote by the symbol P V I (E, ω, P0 , P1 , P2 , P3 , α0 , α1 ,

α2 , α3 ) any of the two equations

ψ(P) = ρ(P). (5.17)

Let π := π2 : E → E  := E/P2  as in Lemma 5.6. Put P0 = π(P0 ) (zero

point), P1 = π(P1 ), and choose remaining two points of order two P2 , P3 .
Assume that π ∗ (ω ) = ω.
In this case for any solution Q to (5.17), the point Q  := π(Q) will be a
solution to

P V I  = P V I (E  , ω , P0 , P1 , P2 , P3 , α0 , α1 , 0, 0), (5.18)

and conversely, if Q  is a solution to P V I  , then Q is a solution to P V I .

Proof. Let ρ = φ j (r ), j = 0 or 1. Our statement results from the following

calculation, using (5.12), (5.13):

ψ E  ,ω (π(Q)) = ψ E,ω (Q) =

φ (α0 (y E ,ω (Q) + y E,ω (Q + P2 )) + α1 (y E,ω (Q + P1 )+
j

y E,ω (Q + P1 + P2 ))) =
φ j (α0 y E  ,ω (π(Q)) + α1 y E  ,ω (π(Q) + π(P1 ))).

Of course, similar statement will hold if π2 is replaced by π1 or π3 .

5.3.2 Two more versions of PVI

The character ψ is of course not algebraic, but the equation (5.16) shows that
it has an algebraic vertical differential. The same is obviously true for the r.h.s.
of (5.11): one easily sees that if E is given in the Weierstrass form y 2 = f (x),
then
3
1
dr = − α j s ∗j ( f  (x))) ω. (5.19)
2
j=0

Hence we may consider two more versions of the arithmetic P V I : the

condition of vanishing of one of the following 1-forms on J 2 (E):
 
dr
dψ − dr = pω(2) + λ1 ω(1) + λ0 − ω(0) , (5.20a)
ω
  
dr
dψ − dφ(r ) = pω(2) + λ1 − pφ ω(0) + λ0 ω(0) . (5.20b)
ω
124 Alexandru Buium and Yuri I. Manin

But in the p-adic situation solutions to (5.17) and (5.20a,b) respectively are
not related to each other in the way we would expect by analogy with usual
calculus.
Indeed, let us consider a smooth function f (x, y, y  , . . . , y (r) ) in r + 2 vari-
ables defined on Rr+2 , where the latter is viewed as the r -th jet space of the
first projection R2 → R, (x, y) → x. Let u = u(x) be an unknown smooth
function u : R → R. Then the equation

f (x, u(x), u  (x), . . . , u (r) (x)) = 0 (5.21)

is related to the 1-form d f on Rr+2 as follows. If u solves (5.21), then taking

the derivative of (5.21) with respect to x we get

∇ r (u)∗ (d f ) = 0 (5.22)

where ∇ r (u) : R → Rr+2 is given by ∇ r (u)(x) = (x, u(x), . . . , u (r) (x)).

However in the case of arithmetic jet spaces the situation is different.
Indeed, let f ∈ O(J r (A1 )) = R[y, y  , . . . , y (r) ]! where J r (A1 ) is the
arithmetic jet space of the affine line over R. Furthermore, let u ∈ R be a
solution of
f (u, δu, . . . , δr u) = 0 (5.23)

where δ = δ p : R → R is the standard p-derivation on R. Applying δ to (5.23)

we get:
1 (φ) p
( f (u + pδu, (δu) p + pδ 2 u, . . . , (δr u) p + pδr+1 u)
p
− f (u, δu, . . . , δr u) p ) = 0 (5.24)

where f (φ) is f with coefficients twisted by the Frobenius φ. To compare this

latter equation with (5.22), one can apply the Taylor formula and get
r
f (φ) (∇ r (u) p ) − f (∇ r (u)) p ∂ f (φ) r
+ (∇ (u) p )(δi +1 u) + M = 0, (5.25)
p ∂ y (i)
i=0

where ∇ r (u) = (u, δu, . . . , δr u).

The first term here is an arithmetic version of the pullback of the term ∂∂xf d x
in d f . The second term clearly involves d f . However, M involves higher par-
tial derivatives of f ; it is highly non-linear, it is divisible by p, and “the more
non-linear its terms are the more they are divisible by p”. In some sense, since
p is small, one can view (5.25) as a non-linear deformation of (5.22).
We will continue discussion of this formalism in sec. 5.5 below. The reader
may wish to skip the next section, or to postpone reading it.
 Arithmetic differential equations of Painlevé VI type 125

5.4 Hamiltonian formalism

5.4.1 Classical case
The classical PVI equation written in the form (5.7) can be represented as a
Hamiltonian flow on the variable two-dimensional phase space (twisted cotan-
gent spaces to a versal family of elliptic curves, cf. [Ma96] and the end of this
section), with time-dependent Hamiltonian:
dz ∂H dy ∂H
= , =− , (5.26)
dτ ∂ y dτ ∂z
where
3
y2 1 Tj
H := H(α0 , . . . , α3 ) := − α j ℘ (z + , τ ). (5.27)
2 (2πi)2 2
j=0

In more geometric terms, this means that solutions to the PVI become leaves
of the null-foliation of the following closed two-form:
ω = ω(α0 , . . . , α3 ) := 2πi(dy ∧ dz − dH ∧ dτ ) =
3
1 Tj
= 2πi(dy ∧ dz − ydy ∧ dτ ) + α j ℘z (z + , τ )dz ∧ dτ. (5.28)
2πi 2
j =0

The extra factor 2πi makes ω defined over Q[αi ] on a natural algebraic model
of (twisted) relative cotangent bundle to the respective versal family of elliptic
curves.
In the expression (5.28), the summand 2πi dy ∧ dz is the canonical fibre-
wise symplectic form on the relative cotangent bundle. The terms involving dτ
uniquely determine the differential of the (time-dependent) Hamiltonian.
Moreover, ω is not just closed, but is a global differential: ω = dν where
the form
1
ν = ν(α0 , . . . , α3 ) := 2πi (ydz − y 2 dτ ) + dlog θ (z, τ ) + 2πi G 2 (τ )dτ +
2
3
1 Tj
+ α j ℘ (z + , τ )dτ (5.29)
2πi 2
j=0

also descends to an appropriate algebraic model, and the Hamiltonian H is

again encoded in the dτ -part of ν. Here is a convenient way to represent this
encoding:
 2 
y 1
H(α0 , . . . α3 ) = i ∂τ dτ + (ν(0, . . . , 0) − ν(α0 , . . . , α3 )) , (5.30)
2 2πi
∂
where ∂τ := ∂τ .
126 Alexandru Buium and Yuri I. Manin

Finally, the last summand in (5.28) is simply the additive differential

d 2z
character 2πi 2 that is generally denoted ψ in the p-adic case.
dτ
In the following subsections, we try to imitate this description of Hamil-
tonian structure for p-adic PVI equations. The reader should be aware that
our treatment is somewhat ad hoc, and must be considered as a tentative step
towards a more coherent vision of Hamiltonian flows with p-adic time. In fact,
we do not yet have an appropriate version of dp replacing dτ and generally do
not know what are differential forms involving “differentials in the arithmetical
direction”.

5.4.2 Arithmetical case: preparation

Let Y be a formal affine scheme over R = W (k). Modules of vertical
differential forms on Y are defined as

Y = lim inv Yn /Rn

where Rn = R/ p n+1 R, Yn = Y ⊗ R Rn .
Let now Z ⊂ J n (Y ) be a closed formal subscheme defined by the ideal
I Z ⊂ O(J n (Y )). Put
 J n (Y )
Z := (5.31)
I Z  J n (Y ) , d I Z 
Given a system of arithmetic differential equations F ⊂ O(J r (Y )), denote
by Z r := Z r (F ) the ideal generated by F . For each s ≤ r , there is a natural
map πr,s : Z r → J s (Y ).
Generally, the natural maps φ ∗ respect degrees of differential forms, one can
define natural maps φ ∗ / pi : iJ r−1 (Y ) → iJ r (Y ) and, for f ∈ O(J 2 (Y )), they
induce maps which we will denote
φ ∗Z
: iJ 1 (Y ) → iZ 2 ( f ) (5.32)
pi

Definition 5.8. We say that F ⊂ O(J r (Y )) defines a generalized canonical

δ-flow on J s (Y ), if the induced map
∗
πr,s  J s (Y ) → Z r

is injective, and its cokernel is annihilated by a power of p.

The cokernel here intuitively measures “how singular” F is on the closed

fibre of Y .
 Arithmetic differential equations of Painlevé VI type 127

Definition 5.9. a) Let X be a smooth surface over R (or the p-adic comple-
tion of such a surface).
A symplectic form on X is an invertible 2-form on X .
A contact form on X is an 1-form on X such that dν is symplectic.
b) Let Y be a smooth curve over R. An 1-form on X := J 1 (Y ) is called
canonical, if ν = fβ, where f ∈ O(X ) and β is an 1-form lifted from Y .

Notice that any closed canonical 1-form on X = J 1 (Y ) is lifted from Y .

We now come to the main definition.

Definition 5.10. Let Y be a smooth affine curve over R and let f ∈ O(J 2 (Y ))
be a function defining a generalized canonical δ-flow on J 1 (Y ).

a) The respective generalized δ-flow is called Hamiltonian with respect to the

symplectic form η on J 1 (Y ), if φ ∗Z η = μ · η in 2 Z 2( f )
for some μ ∈ p R
called the eigenvalue.
b) Assume that moreover η = dν for some canonical 1-form ν on J 1 (Y ).
Then we call
φ Z ν − μν
 := ∈ Z 2 ( f ) (5.33)
p
an Euler–Lagrange form.

We consider (5.33) as an (admittedly, half-baked) arithmetical analog of the

expression i ∂τ (ydz − Hdτ ) (cf. (5.30)) in the same sense as the p-derivation
φ(x) − x p
δ p (x) =
p
is an analog of ∂τ .
Now we pass to the arithmetical PVI. Let again E be an elliptic curve over R
that does not admit a lift of Frobenius and let ψ ∈ O(J 2 (E)) be the canonical
δ-character of order 2 attached to an invertible 1-form ω on E. Consider the
symplectic form η = ω(0) ∧ ω(1) on J 1 (E): cf. (5.15). Let Y ⊂ E be an affine
open set and let r ∈ O(Y ). Assume in addition that Y has an étale coordinate.
(The basic example is E with sections of the second order deleted.)
Denoting such an étale coordinate by T , put A2 = K [[T, T  ]], A3 =
K [[T, T  , T  ]], where K is the quotient ring of R.

Proposition 5.11. The following assertions hold:

1) The function f = ψ − φ(r ) defines a generalized canonical δ-flow on

J 1 (Y ) which is Hamiltonian with respect to η.
128 Alexandru Buium and Yuri I. Manin

2) There exists a canonical 1-form ν on X such that dν = η; in particular the

symplectic form η is exact and if  := 1p (φ ∗Z ν − μν) is the Euler–Lagrange
form then p is closed.
3) Let  be Euler–Lagrange form and f = ψ − φ(r). Then we have the
following equality in A3 :
1 ∗
 = f ω(1) − (φ − μ)ν.
p
4) Let r1 , r2 ∈ O(Y ) be such that r2 − r1 = ∂s, for some s ∈ O(Y ), where
∂ is the derivation on E dual to ω. (This holds for two right-hand sides of
any two PVI equations.) Consider the equations ψ − φ(r1 ) and ψ − φ(r2 )
respectively. Then there exists a canonical 1-form ν on J 1 (Y ) such that
dν = η and such that, if 1 , 2 are the corresponding Euler–Lagrange
forms, then:
1
1 − 2 = dφ(s) ∈ A2 .
p

Remark. We will deduce from this Proposition below (cf. Corollary 5.16)
that in fact p-adic PVI in the form (5.17) defines a generalized canonical
δ-flow. However, it does not define a δ-flow in the sense of Definition 5.5.
This motivated our Definition 5.8.

Proof. From (5.15), we get the following equality in 2J 2 (Y ) :

φ∗η φ ∗ ω(0) φ ∗ ω(1)

= ∧ = ω(1) ∧ ω(2) .
p2 p p
Recall the formula (5.20b):

d f = pω(2) + (λ1 − pφ(∂r ))ω(1) + λ0 ω(0)

in  J 2 (E) where λ1 ∈ R, λ0 ∈ R × . Hence, if we keep notation ω(0) , ω(1) also

for the images of the respective forms in Z 2 ( f ) , in view of ω(1) ∧ d f = 0 in
Z 2 ( f ) , we have the following equality in 2
Z 2( f )
:

φ ∗Z η = − p · ω(1) ∧ ((λ1 − pφ(dr ))ω(1) + λ0 ω(0) ) = pλ0 η,

This completes the proof of 1).

Write now ω = d L = dT dL
dT where L = L(T ) ∈ T K [[T ]]. (For instance,
we can take L to be the formal logarithm of E.) Then

A3 = K [[T, φ(T ), φ 2 (T )]] = K [[L , φ(L), φ 2 (L)]].

 Arithmetic differential equations of Painlevé VI type 129

So the image of ψ in A3 is
1 2
ψ= (φ (L) + λ1 φ(L) + pλ0 L) + λ−1
p
for some λ−1 ∈ R. Therefore the maps

 A3 2A3
A2 → , 2A2 →
 f A3 , d f   f 2A3 , d f ∧ A3 

are isomorphisms and so we have induced Frobenii maps φ ∗f : A2 → A2

and φ ∗f : 2A2 → 2A2 . Since T is étale the lift of Frobenius T → T p on
!→ Y
A1 extends to a lift of Frobenius φ0 : Y ! of the p-adic completion of Y .
Also the derivation dT on R[T ] extends to a derivation still denoted by dT
d d
on
!
O(Y ). We claim that
φ(L) − φ0 (L)
,
p
which a priori is an element of A2 , actually belongs to O(J 1 (Y )). Indeed we
have the following expansion in A2 :
φ(L) − φ0 (L) L (φ) (T p + pT  ) − L (φ) (T p )
=
p p
∞ (φ) ∞  i 
p i−1 i
d L  i pi −1 d L
= p
(T )(T ) = φ0 (T  )i
i! dT i i! dT i
i=1 i=1
 i −1  
ω 
∞ i −1
p d
= φ0 (T  )i ∈ O(J 1 (Y )),
i! dT dT
i =1

where the superscript (φ) means twisting coefficients by φ. The latter inclusion
ω
follows because T  = δT ∈ O(J 1 (Y )), dT ∈ O(Y ) ⊂ O(Y !), and the latter is
stable under dT and φ0 . Now set
d

φ(L) − φ0 (L)
ν := − ω ∈ O(J 1 (Y )) ω ∈  J 1 (Y ) .
p
Then
 
φ(L) − φ0 (L)
dν = −d ∧ω
p
   
φ(L) φ0 (L)
= −d ∧ω+d ∧ω
p p
= −ω(1) ∧ ω(0) = η.
This completes the proof of 2).
130 Alexandru Buium and Yuri I. Manin

Next for f = ψ − φ(r ) we have the following computation in A3 :

p = φ ∗f ν − μν
 
∗ φ(L) − φ0 (L) φ(L) − φ0 (L)
= −φ f ω +μ ω
p p
φ(L) − φ0 (L)
= −φ f φ(L)ω(1) + φφ0 (L)ω(1) + μ ω
p
φ(L) − φ0 (L) (0)
= (λ1 φ(L) + pλ0 L + φφ0 (L) − pφ(r ) + pλ−1 )ω(1) + μ ω
p
= (λ1 φ(L) + pλ0 L + φφ0 (L) − pφ(r ) + pλ−1 )ω(1)
 
∗ φ(L) − φ0 (L) (0)
+φ ω − (φ ∗ − μ)ν
p
 
= φ 2 (L) + λ1 φ(L) + pλ0 L − pφ(r ) + pλ−1 ω(1) − (φ ∗ − μ)ν
= p f ω(1) − (φ ∗ − μ)ν.

This ends the proof of assertion 3). Assertion 4) follows from the fact that
1
1 − 2 = φ(∂s)ω(1) = d(φ(s)).
p
Remarks. a) Some of our arguments above break down if ψ −φ(r ) is replaced
by ψ − r . This is our main motivation for studying ψ − φ(r ).
A 3
b) Assertion 3) implies that if []2 , [ f ω(1) ]2 ∈ (φ ∗ −μ) A
are the images of
2
, f ω(1) then
[]2 = [ f ω(1) ]2 .

This justifies our suggestion that f is the “Euler–Lagrange equation

attached to our Hamiltonian data”. Assertion 4) says that the Euler–
Lagrange forms of various PVI equations differ by exact forms.
c) Finally, we could treat in this way also the multicomponent version of PVI
and the degeneration PV as they are described in [Ta01].

5.4.3 Geometry of PVI in various categories

Below we essentially reproduce from [Ma96] a geometric description of the
natural habitats of various forms of PVI including its arithmetic version.
Our series of constructions starts with a non–constant family (“pencil”) of
elliptic curves over one-dimensional base. in one of several natural categories:
schemes over a field of characteristic zero or a ring of algebraic integers, or a
p-adic completen of the latter, analytic spaces etc.
 Arithmetic differential equations of Painlevé VI type 131

a. Let (π : E → B; D0 , . . . , D3 ) be a pencil of compact smooth curves of

genus one, with variable absolute invariant, endowed with four labelled sec-
tions Di such that if any one of them is taken as zero, the others will be of
order two.
We will call E a configuration space of PVI (common for all values of
parameters.) Solutions to all equations will be represented by some multisec-
tions of π.
b. Let F be the subsheaf of the sheaf of vertical 1-forms 1E/B (D3 ) on E
with pole at D3 and residue 1 at this pole. It is an affine twisted version of
1E/B which is the sheaf of sections of the relative cotangent bundle TE/B ∗ .

Similarly, F itself “is” the sheaf of sections of an affine line bundle F = FE/B
on E. More precisely, we can construct a bundle λ : F → E and a form
ν F ∈ (F, 1F/B (λ−1 (D3 )) such that the map

{local section s of F} → s ∗ (ν F )

identifies the sheaf of sections of F/E with F .

We will call F a phase space for PVI (again, common for all parameter
values). It is this space that carries a canonical symplectic form (relative over
the base) rather than the usual cotangent bundle.
c. E carries a distinguished family of algebraic/arithmetic curves transversal
to the fibers of E: considered as multisections of E/B they are of finite order
(if any of Di is chosen as zero). It is important that each curve of this family
has a canonical lifting to F (for its description, see [Ma96], especially (2.12)
and (2.29)).
d. F carries a closed two-form ω which can be characterised by the
following two properties:

i). The vertical part of ω, i.e. its restriction to the fibers of π ◦ λ : F → B,

coincides with d F/B (ν F ).
ii). Any canonical lift to F of a connected multisection of finite order of
E → B, referred to above, is a leaf of the null-foliation of ω.

e. E also carries four distinguished closed two-forms ω0 , . . . , ω3 . They are

determined, up to multiplication by a constant, by the following properties.
D j Dk Dl
i). The divisor of ωi is where {i, j, k, l} = {0, 1, 2, 3}.
Di3
ii). Identify the sheaves 2E and π ∗ (1E/B )⊗3 on E using the Kodaira–
Spencer isomorphism π ∗ (1B ) ∼
= (1E/B )⊗2 and the exact sequence
0 → π ( B ) →  E →  E/B → 0. Then the image of ωi in π ∗ (1E/B )⊗3
∗ 1 1 1
132 Alexandru Buium and Yuri I. Manin

considered in the formal neighborhood of Di is the cube of a vertical

1-form with a constant residue along Di .
Notice that up to introduction of ω, all constructions were valid both in
geometric and arithmetic cases. It is precisely the absence of the differential
in arithmetic direction was the object of our concerns in 4.2–4.4 above.

The affine space P0 := ω + i3=0 Cλ∗ (ωi ) of closed two-forms on F is
our version of the moduli space of the PVI equations replacing the classical
(α, β, γ , δ)-space.
We can now summarize our geometric definition of PVI equations and their
solutions.

Definition 5.12. a) A Painlevé two-form on F is a point  ∈ P0 .

b) The Painlevé foliation corresponding to  is the null-foliation of .
c) The solutions to the respective Painlevé equation are the leaves of this
foliation (in the Hamiltonian description).
The form ω corresponds to (α0 , α1 , α2 , α3 ) = (0, 0, 0, 0).

5.5 Pfaffian congruences

5.5.1 Notation
We will continue here the discussion started in 3.3.
For u ∈ R, define the following infinite vectors in R × R × . . .:
∇(u) := (u, δu, δ 2 u, . . .),
δ∇(u) := (δu, δ 2 u, δ 3 u, . . .),
∇(u) p := (u p , (δu) p , (δ 2 u) p , . . .).
For g ∈ R[y, y  , . . . , y (r) ]!, denote by g (φ) be the series obtained from g by
applying φ to the coefficients of g. Moreover, put
r
∂g
dg := dy (i) ,
∂ y (i)
i=0
∂g 1
:= (g (φ) (y p , . . . (y (r) ) p ) − g(y, . . . , y (r) ) p ),
∂p p
 
∂ ∂ ∂ ∂
:= , , ... ,
∂∇(y) ∂ y ∂ y  ∂ y 
Denoting by  
,  the basic pairing between 1-forms and vector fields, and
t
∂ ∂
writing ∂∇(y) for the column transpose of ∂∇(y) , we get for any f ∈
R[y, y  , . . . , y (r) ]! and i ≥ 0:
 Arithmetic differential equations of Painlevé VI type 133

δ i +1 f (u) = δδ i f (u)
1 i (φ) p
= ((δ f ) (u + pδu, (δu) p + pδ 2 u, . . .) − (δi f )(u, δu, . . .) p )
p
r
∂δ i f ∂(δ i f )(φ)
≡ (∇(u) p ) + (∇(u) p )δ j +1 u mod p
∂p ∂ y( j )
j =0
 t
∂δ f
i
(φ) ∂
≡ (∇(u) ) +  d[(δ f ) ](∇(u) ), δ∇(u) ·
p i p
 mod p.
∂p ∂∇(y)
It is known ([Bu05], Lemma 3.20), that if b ∈ R then b = 0 if and only if
δ i b ≡ 0 mod p for all i ≥ 0.
Combining the above facts we get:

Proposition 5.13. An element u ∈ R is a solution to f (u, δu, . . . , δ r u) = 0 if

and only if the following hold:
a) f (∇(u)) ≡ 0 mod p,  t
∂(δ i f ) (φ) ∂
b) (∇(u) ) ≡ − d[(δ f ) ](∇(u) ), δ∇(u) ·
p i p  mod p,
∂p ∂∇(y)
i ≥ 0.
Moreover, u is a solution to the equation δ f (u, δu, . . . , δr u) = 0 if and
only if b) above holds.

The above discussion can be obviously generalized to the case when y is

a tuple of variables and f is a tuple of equations. It shows that the equation
f (u, δu, . . . , δr u) = 0 is controlled by a system of “Pfaffian” congruences
involving the 1-forms
d((δi f )(φ) ) = d(δi ( f (φ) )), i ≥ 0.
If f has Z p -coefficients then the above forms are, of course,
d(δ i f ), i ≥ 0.
In what follows we analyse such forms relevant for P V I equations.

5.5.2 The forms δ i (ψ − ρ)

We start by defining inductively certain universal δ-polynomials.
Consider a family of commuting free variables z = (z 0 , z 1 , . . . , zr ) and let
w be another variable. As in the jet theory in [Bu05], denote by z  , z  , . . ., resp.
w , w , . . . new (families of) independent variables indexed additionally by the
134 Alexandru Buium and Yuri I. Manin

formal order of derivative. Consider the ring Z[z, z  , z  , . . . , w, w  , w  , . . .],

(k) (k+1)
equipped with the tautological p-derivation δz i = z i etc., and hence
with the lift of Frobenius φ(F) = F + pδ F. Define the elements Am,i =
p

Am,i (z, z  , z  , . . . , w, w , w . . .) of this ring (m ≥ 0, i = 0, . . . , m + r ) by

induction:

A0,i := z (i) , i = 0, . . . , r
Am+1,0 := −(w(m) ) p−1 Am,0 , m≥0 (5.34)
(m) p−1
Am+1,i := φ(Am,i−1 ) − (w ) Am,i , i = 1, . . . , m + r, m ≥ 0
Am+1,m+r+1 := φ(Am,m+r ), m ≥ 0.

It is easy then to check that

Am,0 = (−1)m−1 (ww  . . . w(m−1) ) p−1 z (0) , m ≥ 1,

Am,m+r = φ m (z (r ) ), m ≥ 0,
Am,m+r−1 ≡ φ m (z (r−1) ) mod (z (r) , φ(z (r) ), . . . , φ m−1 (z (r) )), m ≥ 0
(5.35)
Am,m+i ≡ φ m (z (i ) ) mod (w, w , . . . , w(m−1) ), m ≥ 1, i = 0, . . . , r,
Am,i ≡ 0 mod (w, w , . . . , w (m−1) ), i = 0, . . . , m − 1.

In the following statement Y is an affine smooth curve over R and

f ∈ O(J r (Y )). Then by [Bu05], there exist a0 , . . . , ar ∈ O(J r (Y )) such that
r
df = ai ω(i ) .
i=0

Put also a = (a1 , . . . , ar ). Then we have:

Proposition 5.14. Put am,i = Am,i (a, δa, δ 2 a, . . . , f, δ f, δ 2 f, . . .). Then

m+r
d(δ m f ) = am,i ω(i ) .
i=0

Proof. We proceed by induction on m ≥ 0. The case m = 0 is trivial. The

passage from m to m + 1 runs as follows:
 
φ(δ m f ) − (δ m f ) p
d(δ m+1 f ) = d(δδ m f ) = d
p
φ∗
= (d(δ m f )) − (δ m f ) p−1 d(δ m f )
p
 Arithmetic differential equations of Painlevé VI type 135

m+r m+r
φ∗
= ( am,i ω(i) ) − (δ m f ) p−1 ( am,i ω(i ) )
p
i=0 i =0
m+r m+r m+1+r
=( φ(am,i )ω(i+1) ) − (δ m f ) p−1 ( am,i ω(i) ) = am+1,i ω(i) .
i=0 i=0 i =0

This ends the proof.

Now we apply this to the case when Y is an affine open subset of the elliptic
curve E over Z p without lift of Frobenius. Denote by a p ∈ Z the trace of
Frobenius on the reduction E mod p.

Corollary 5.15. Let f = ψ − ρ, ρ = φ(r ), where r ∈ O(Y ), ψ the canonical

 (i )
δ-character of order 2 attached to ω. Let d(δ m f ) = m+2
i=0 am,i ω , m ≥ 0,
am,i ∈ O(J m+2 (Y )). So
dr
a0,2 = p, a0,1 = −(a p + pφ( )), a0,0 = 1.
ω
Then, for m ≥ 1:

1) am,m+2 = p;
2) am,m+1 ≡ −a p mod p;
3) am,m+1 ≡ −(a p + pφ m+1 ( dr ω )) mod ( f, δ f, . . . , δ
m−1 f );

4) am,m ≡ 1 mod ( f, δ f, . . . , δ m−1 f );

5) am,i ≡ 0 mod ( f, δ f, . . . , δ m−1 f ) for i = 0, . . . , m − 1;
6) am,0 = (−1)m−1 ( f · δ f · · · δ m−1 f ) p−1 , for m ≥ 1.
In particular if E has ordinary reduction then am,m+1 is invertible in
O(J m+2 (Y )).

One can prove a similar statement for ψ − r in place of ψ − φ(r ).

Corollary 5.16. In the same situation, let f = ψ − ρ, where ρ = r or ρ =

φ(r ). Let ψ be the canonical δ-character of order 2 attached to ω.
Denote by Z m+2 = Z m+2 ( f, δ f, . . . , δ m f ) the closed formal subscheme of
J m+2 (Y ) defined by the ideal generated by f , δ f , . . . , δ m f . Let πm : Z m+2 →
J (Y ) be the canonical projection.
1

Then the map

πm∗  J 1 (Y ) → Z m+2

is injective with cokernel annihilated by p m+1 . In particular, for any m ≥ 0,

the system { f, δ f, . . . , δ m f } ⊂ O(J m+2 (Y )) defines a generalised δ-flow on
136 Alexandru Buium and Yuri I. Manin

J 1 (Y ). If moreover E has ordinary reduction then the above cokernel is a

cyclic O(Z m+2 )-module generated by the class of ω(m+2) .

Proof. Consider the case ρ = φ(r ); a similar argument holds for ρ = r. Recall
that  J n (Y ) is a free O(J n (Y ))-module generated by ω(0) , . . . , ω(n) . Also, by
definition,
O(J m+2 (Y ))
O(Z m+2 ) =
( f, δ f, . . . , δ m f )
and
O(J m+2 (Y ))ω(0) ⊕ . . . ⊕ O(J m+2 (Y ))ω(m+2)
Z m+2 = ,
(δ i f )ω( j) , d(δi f )
where   means O(J m+2 (Y ))-linear span and j = 0, . . . , m+2, i = 0, . . . , m.
By Corollary 5.15 we have:
O(Z m+2 )ω(0) ⊕ . . . ⊕ O(Z m+2 )ω(m+2)
Z m+2 =
 pω(2+i) − (a p + pφ i +1 ( dr
ω ))ω
(1+i) + ω(i) 

where i = 0, . . . , m and   means here O(Z m+2 )-linear span. Note that
πm∗  J 1 (Y ) is a free O(J m+2 (Y ))-module with basis ω(0) , ω(1) . So in order to
prove that the map
πm∗  J 1 (Y ) → Z m+2 (5.36)

is injective we need to check that no O(Z m+2 )-linear combination of ω(0) ,

ω(1) can be a O(Z m+2 )-linear combination of elements pω(2+i ) − (a p +
pφ i+1 ( dr (1+i ) + ω(i) which is clear. For j = 2, . . . , m + 2 let
ω ))ω

Z m+2
ω( j ) ∈
ω(0) , ω(1) 

be the image of ω( j ) . Clearly pω(2) = 0 hence p 2 ω(3) = 0, etc. We conclude

that p m+1 annihilates the cokernel of (5.36).
Finally assume E has ordinary reduction. We want to show that the
above cokernel is generated by ω(m+2) . It is enough to show that for all
j = 1, . . . , m + 2 we have

ω( j−1) ∈ p O(Z m+2 ) ω( j ) .

We proceed by induction on j. For j = 1 this is clear. Now assume the above

is true for some 1 ≤ j < m + 2. We have
dr
pω( j+1) − (a p + pφ j ( ))ω( j ) + ω( j−1) = 0.
ω
 Arithmetic differential equations of Painlevé VI type 137

By induction ω( j−1) = p c ω( j ) for some c ∈ O(Z m+2 ). Since a p is invertible

in R it follows that a p − pc(1 + φ j−1 (dr/ω)) is invertible hence

dr
ω( j) = p(a p + pφ j ( ) − pc)−1 ω( j +1) ,
ω
which ends the proof.

Remark 5.17. Let Y be a smooth affine scheme over R and let F ⊂ O(J r (Y ))
be a system of arithmetic differential equations. Consider the set

{F , δF , . . . , F} ⊂ O(J r+m (Y ))

which can be referred to as the m-th prolongation of F . Let

Z r+m := Z r +m (F, δF, . . . , δ m F ) ⊂ J r+m (Y )

be the closed subscheme defined by this prolongation and let

Z ∞ = Z ∞ (δ ∞ F)

be defined by
Z ∞ = lim Z r+m ,
←

the projective limit (defined as the Sp f of the p-adic completion of the induc-
tive limit of O(Z r+m ) as m varies; this is a generally non-Noetherian formal
scheme). Then Z ∞ is a closed horizontal formal subscheme of

J ∞ (Y ) := lim J n (Y );
←

by horizontal we mean here that its ideal is sent into itself by δ. So there is an
induced p-derivation δ Z ∞ on O(Z ∞ ).
If Z ∞ happens to be a smooth formal scheme then δ Z ∞ defines a (genuine)
δ-flow E(δ Z ∞ ) ⊂ O(J 1 (Z ∞ )) on Z ∞ (in the sense of our previous definition).
In case Z ∞ is not necessarily a smooth formal scheme one can still define the
affine formal scheme J 1 (Z ∞ ) and the δ-flow E(δ Z ∞ ) on Z ∞ by copying the
definitions from the smooth case.
The following problem needs to be investigated. Assume that F ⊂
O(J 2 (Y )) defines a δ-flow E(δ J 1 (Y ) ) ⊂ O(J 1 (J 1 (Y ))) on J 1 (Y ). Consider
the formal scheme Z ∞ = Z ∞ (δ ∞ F ) as above. Prove that Z ∞ is naturally
isomorphic to J 1 (Y ) and E(δ J 1 (Y ) ) is naturally identified with E(δ Z ∞ ). This
would show the naturality of the above definitions.
138 Alexandru Buium and Yuri I. Manin

References
[BlSch12] M. Blickle, K. Schwede. p−1 -linear maps in algebra and geometry.
arXiv:1205.4577
[Bo09] J. Borger. Lambda-rings and the field with one element. arXiv:0906.3146
[Bu95] A. Buium. Differential characters of abelian varieties over p-adic fields.
Inv. Math., vol. 122 (1995), 309–340.
[Bu97] A. Buium. Differential characters and characteristic polynomial of Frobe-
nius. J. reine u. angew. Math. 485 (1997), 209–219.
[Bu05] A. Buium. Arithmetic Differential Equations. Math. Surveys and Mono-
graphs, 118, MS, Providence RI, 2005. xxxii+310 pp.
[Ma96] Yu. Manin. Sixth Painlevé equation, universal elliptic curve, and mirror
of P2 . In: Geometry of Differential Equations, ed. by A. Khovanskii,
A. Varchenko, V. Vassiliev. Amer. Math. Soc. Transl. (2), vol. 186 (1998),
131–151. Preprint alg–geom/9605010.
[Ma08] Yu. Manin. Cyclotomy and analytic geometry over F1 . In: Quanta of Maths.
Conference in honour of Alain Connes. Clay Math. Proceedings, vol. 11
(2010), 385–408. Preprint math.AG/0809.2716.
[Ta01] K. Takasaki. Painlevé–Calogero correspondence revisited.
Journ. Math. Phys., vol. 42, Nr 3 (2001), 1443–1473.
 6
Differential calculus with integers
Alexandru Buium
Department of Mathematics and Statistics,
University of New Mexico,
Albuquerque, NM 87131, USA
E-mail address: [email protected]

A BSTRACT. Ordinary differential equations have an arithmetic analogue in

which functions are replaced by numbers and the derivation operator is
replaced by a Fermat quotient operator. In this survey we explain the main
motivations, constructions, results, applications, and open problems of the
theory.

The main purpose of these notes is to show how one can develop an arith-
metic analogue of differential calculus in which differentiable functions x(t)
are replaced by integer numbers n and the derivation operator x → ddtx is
p
replaced by the Fermat quotient operator n → n−n p , where p is a prime
integer. The Lie–Cartan geometric theory of differential equations (in which
solutions are smooth maps) is then replaced by a theory of “arithmetic differ-
ential equations” (in which solutions are integral points of algebraic varieties).
In particular the differential invariants of groups in the Lie–Cartan theory are
replaced by “arithmetic differential invariants” of correspondences between
algebraic varieties. A number of applications to diophantine geometry over
number fields and to classical modular forms will be explained.
This program was initiated in [11] and pursued, in particular, in [12]–[35].
For an exposition of some of these ideas we refer to the monograph [16]; cf.
also the survey paper [58]. We shall restrict ourselves here to the ordinary
differential case. For the partial differential case we refer to [20, 21, 22, 7].
Throughout these notes we assume familiarity with the basic concepts of alge-
braic geometry and differential geometry; some of the standard material is
being reviewed, however, for the sake of introducing notation, and “setting
the stage”. The notes are organized as follows. The first section presents some

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

139
140 Alexandru Buium

classical background, the main concepts of the theory, a discussion of the main
motivations, and a comparison with other theories. The second section presents
a sample of the main results. The third section presents a list of open problems.
Acknowledgements. The author is indebted to HIM for support during part
of the semester on Algebra and Geometry in Spring 2013. These notes are
partially based on lectures given at the IHES in Fall 2011 and MPI in Summer
2012 when the author was partially supported by IHES and MPI, respectively.
Partial support was also received from the NSF through grant DMS 0852591.

6.1 Main concepts

6.1.1 Classical analogies
The analogies between functions and numbers have played a key role in the
development of modern number theory. Here are some classical analogies. All
facts in this subsection are well known and entirely classical; we review them
only in order to introduce notation and put things in perspective.

6.1.1.1 Polynomial functions

The ring C[t] of polynomial functions with complex coefficients is analogous
to the ring Z of integers. The field of rational functions C(t) is then analogous
to the field of rational numbers Q. In C[t] any non-constant polynomial is a
product of linear factors. In Z any integer different from 0, ±1 is up to a sign
a product of prime numbers. To summarize
C ⊂ C[t] ⊂ C(t)
are analogous to
{0, ±1} ⊂ Z ⊂ Q

6.1.1.2 Regular functions

More generally rings O(T ) of regular functions on complex algebraic affine
non-singular curves T are analogous to rings of integers O F in number
fields F. Hence curves T themselves are analogous to schemes Spec O F .
Compactifications
T ⊂ T = T ∪ {∞1 , . . . , ∞n } (compact Riemann surface of genus g)
are analogous to “compactifications”
H om(F, C)
Spec O F ⊂ Spec O F = (Spec O F ) ∪
conjugation
 Differential calculus with integers 141

6.1.1.3 Formal functions

The inclusions
C ⊂ C[[t]] ⊂ C((t))
(where C[[t]] is the ring of power series and C((t)) is the ring of Laurent
series) are analogous to the inclusion
{0} ∪ μ p−1 = {c ∈ Z p ; c p = c} ⊂ Z p ⊂ Q p
(where Z p is ring of p-adic integers and Q p = Z p [1/ p]). Recall that
∞
Z p = lim Z/ p n Z = { ci pi ; ci ∈ {0} ∪ μ p−1 }
←
n=0

So {0} ∪ μ p−1 plays the role of “constants” in Z p . Sometimes we need more

“constants” and we are led to consider, instead, the inclusions:
0
{0} ∪ 
μ pν −1 ⊂ Z 
ur ⊂ Z ur [1/ p]
p p
ν

where
∞ 0
ν

Z ur = Z [ζ ; ζ p −1 = 1, ν ≥ 1]ˆ = {
p ci pi ; ci ∈ {0} ∪ μ pν −1 }.
p
i =0 ν

Here the upper hat on a ring A means its p-adic completion:

! := lim A/ p n A.
A
←

So in the latter case the monoid {0} ∪ ν μ pν −1 should be viewed as the set
of “constants” of Z ur ; this is consistent with the “philosophy of the field with
p
one element” to which we are going to allude later. Let us say that a ring is a
local p-ring if it is a discrete valuation ring with maximal ideal generated by
a prime p ∈ Z. Then Z p and Z  ur are local p-rings. Also for any local p-ring
p
R we denote by k = R/ p R the residue field and by K = R[1/ p] the fraction
field of R. Sometimes we will view local p-rings as analogues of rings C{x}
of germs of analytic functions on Riemann surfaces and even as analogues of
rings of global analytic (respectively C ∞ functions) on a Riemann surface T
(respectively on a 1-dimensional real manifold T , i.e. on a circle S 1 or R).

6.1.1.4 Topology
Fundamental groups of complex curves (more precisely Deck transformation
groups of normal covers T  → T of Riemann surfaces) have, as analogues,
Galois groups G(F  /F) of normal extensions number fields F ⊂ F  . The
genus of a Riemann surface has an analogue for number fields defined in terms
142 Alexandru Buium

of ramification. All of this is very classical. There are other, less classical, topo-
logical analogies like the one between primes in Z and nodes in 3-dimensional
real manifolds [61].

6.1.1.5 Divisors
The group of divisors

Div(T ) = { n P P; n P ∈ Z}
P∈T

on a non-singular complex algebraic curve T is analogous to the group of

divisors

Div(Spec O F ) = { ν P P; ν P ∈ Z if P is finite, ν P ∈ R if P is infinite}

One can attach divisors to rational functions f on T (Div( f ) is the sum of

poles minus the sum of zeroes); similarly one can attach divisors to elements
f ∈ F. In both cases one is lead to a “control” of the spaces of f s that have
a “controlled” divisor (the Riemann–Roch theorem). One also defines in both
settings divisor class groups. In the geometric setting the divisor class group
of T is an extension of Z by the Jacobian

J ac(T ) = Cg /(period lattice of T )

where g is the genus of T . In the number theoretic setting divisor class groups
can be interpreted as “Arakelov class groups”; one recaptures, in particular, the
usual class groups Cl(F). Exploring usual class groups “in the limit”, when
one adjoins roots of unity, leads to Iwasawa theory. We will encounter Jaco-
bians later in relation, for instance, to the Manin–Mumford conjecture. This
conjecture (proved by Raynaud) says that if one views T as embedded into
J ac(T ) (via the “Abel–Jacobi map”) the the intersection of T with the torsion
group of J ac(T ) is a finite set. This particular conjecture does not seem to
have an analogue for numbers.

6.1.1.6 Families
Maps
M→T

of complex algebraic varieties, complex analytic, or real smooth manifolds,

where dim T = 1, are analogous to arithmetic schemes, i.e. schemes of finite
type
X → Spec R
 Differential calculus with integers 143

where R is either the ring of integers O F in a number field F or a complete

local p-ring respectively. Note however that in this analogy one “goes arith-
metic only half way”: indeed for X → Spec R the basis is arithmetic yet
the fibers are still geometric. One can attempt to “go arithmetic all the way”
and find an analogue of M → T for which both the base and the fiber are
“arithmetic”; in particular one would like to have an analogue of T × T which
is “arithmetic” in two directions. This is one of the main motivations in the
search for F1 , the “field with one element”.

6.1.1.7 Sections
The set of sections
(M/T ) = {s : T → M; π ◦ σ = 1}
of a map π : M → T is analogous to the set
X (R) = {s : Spec R → X ; π ◦ s = 1}
of R-points of X where π : X → Spec R is the structure morphism. This
analogy suggests that finiteness conjectures for sets of the form X (R), which
one makes in Diophantine geometry, should have as analogues finiteness con-
jectures for sets of sections (M/T ). A typical example of this phenomenon
is the Mordell conjecture (Faltings’ theorem) saying that if X is an algebraic
curve of “genus” ≥ 2, defined by polynomials with coefficients in a num-
ber field F, then the set X (F) is finite. Before the proof of this conjecture
Manin [55] proved a parallel finiteness result for (M/T ) where M → T is a
“non-isotrivial” morphism from an algebraic surface to a curve, whose fibers
have genus ≥ 2. Manin’s proof involved the consideration of differential equa-
tions with respect to vector fields on T . Faltings’ proof went along completely
different lines. This raised the question whether one can develop a theory of
differential equations in which one can differentiate numbers.
All these examples of analogies are classical; cf. work of Dedekind, Hilbert,
Hensel, Artin, Weil, Lang, Tate, Iwasawa, Grothendieck, and many others.

6.1.2 Analogies proposed in [11]–[35]

One thing that seems to be missing from the classical picture is a counterpart,
in number theory, of the differential calculus (in particular of differential equa-
tions) for functions. Morally the question is whether one can meaningfully
consider (and successfully use) “arithmetic differential equations” satisfied by
numbers. In our research on the subject [11]–[35] we proposed such a theory
based on the following sequence of analogies:
144 Alexandru Buium

6.1.2.1 Derivatives
The derivative operator δt = dtd
: C[t] → C[t] is analogous to the Fermat
p
quotient operator δ = δ p : Z → Z, δ p a = a−ap . More generally the deriva-
tive operator δt = dt d
: C ∞ (R) → C ∞ (R) (with t the coordinate on R) is
analogous to the operator δ = δ p : R → R on a complete local p-ring R,
δ p α = φ(α)−α
p
p (where φ : R → R is a fixed homomorphism lifting the
p-power Frobenius map on R/ p R). The map δ = δ p above is, of course, not a
derivation but, rather, it satisfies the following conditions:

δ(1) = 0
δ(a + b) = δ(a) + δ(b) + C p (a, b)
δ(ab) = a p δ(b) + b p δ(a) + pδ(a)δ(b),

where C p (x, y) ∈ Z[x, y] is the polynomial C p (x, y) = p −1 (x p + y p

−(x + y) p ). Any set theoretic map δ : A → A from a ring A to itself sat-
isfying the above axioms will be referred to a p-derivation; such operators
were introduced independently in [49, 11] and they implicitly arise in the the-
ory of Witt rings. For any such δ, the map φ : A → A, φ(a) = a p + pδa is a
ring homomorphism lifting the p-power Frobenius on A/ p A; and vice versa,
given a p-torsion free ring A and a ring homomorphism φ : A → A lifting
Frobenius the map δ : A → A, δa = φ(a)−a
p
p is a p-derivation.

6.1.2.2 Differential equations

Differential equations F(t, x, δt x, . . . , δtn x) = 0 (with F smooth) satisfied by
smooth functions x ∈ C ∞ (R) are replaced by arithmetic differential equations
F(α, δ p α, . . . , δ np α) = 0 with F ∈ R[x0 , x1 , . . . , xn ]ˆ satisfied by numbers
α ∈ R. As we shall see it is crucial to allow F to be in the p-adic completion
of the polynomial ring rather than in the polynomial ring itself; indeed if one
restricts to polynomial F’s the main interesting examples of the theory are
left out.

6.1.2.3 Jet spaces

More generally, the Lie–Cartan geometric theory of differential equations has
an arithmetic analogue which we explain now. Let M → T be a submersion
of smooth manifolds with dim T = 1, and let

J n (M/T ) = {Jtn (s); s ∈ (M/T ), t ∈ T }

be the space of n-jets Jtn (s) of smooth sections s of M → T at points

t ∈ T . Cf. [1, 48, 62] for references to differential geometry. If M = Rd × R,
T = R, M → T is the second projection, and x = (x1 , . . . , xd ), t are global
 Differential calculus with integers 145

coordinates on Rd and R respectively then J n (M/T ) = R(n+1)d × R with jet

coordinates x, x  , . . . , x (n) , t. So for general M → T the map J n (M/T ) →
M is a fiber bundle with fiber Rnd where d + 1 = dim(M). One has the total
derivative operator
δt : C ∞ (J n (M/T )) → C ∞ (J n+1 (M/T ))
which in coordinates is given by
n d
∂ ( j +1) ∂
δt = + xi ( j)
.
∂t ∂ xi
j=0 i=1

In the arithmetic theory the analogues of the manifolds J n (M/T ) are cer-
tain formal schemes (called arithmetic jet spaces or p-jet spaces) J n (X ) =
J n (X/R) defined as follows. Assume X is affine, X = Spec R[x] ( f ) with x, f
tuples; the construction that follows is easily globalized to the non-affine
case. Let x  , . . . , x (n) , . . . be new tuples of variables, consider the polyno-
mial ring R{x} := R[x, x  , x  , . . .], let φ : R{x} → R{x} be the unique
ring homomorphism extending φ on R and sending φ(x) = x p + px  ,
φ(x  ) = (x  ) p + px  ,. . . , and let
δ = δ p : R{x} → R{x}
be the p-derivation
φ(F) − F p
δF = .
p
Then one defines
R[x, x  , . . . , x (n) ]ˆ
J n (X ) = Sp f .
( f, δ f, . . . , δ n f )

6.1.2.4 Differential equations on manifolds

Usual differential equations are defined geometrically as elements of the ring
C ∞ (J n (M/T )); alternatively such elements are referred to as (time depen-
dent) Lagrangians on M. Their analogue in the arithmetic theory, which
we call arithmetic differential equations [16], are the elements of the ring
On (X ) := O(J n (X )). For group schemes the following concept [11] plays
an important role: if G is a group scheme of finite type over R then we may
consider the R-module Xn (G) = H om gr (J n (G), G !a ) ⊂ On (G).
Going back to arbitrary schemes of finite type X/R note that δ p : R{x} →
R{x} induces maps δ = δ p : On (X ) → On+1 (X ) which can be viewed as
arithmetic analogues of the total derivative operator δt : C ∞ (J n (M/T )) →
C ∞ (J n+1 (M/T )). The latter is a “generator” of the Cartan distribution defined
146 Alexandru Buium

by d xi − xi dt, d xi − xi dt, etc. Note however that the forms in differential
geometry defining the Cartan distribution do not have a direct arithmetic ana-
logue; for one thing there is no form “dp” analogous to dt. On the other hand
in the arithmetic case we have induced ring homomorphisms φ : On (X ) →
On+1 (X ) which have no analogue in differential geometry.

6.1.2.5 Differential functions

Any differential equation F ∈ C ∞ (J n (M/T )) defines a natural differential
function F∗ : (M/T ) → C ∞ (T ); in coordinates sections s ∈ (M/T )
correspond to functions x = x(t) and then F∗ sends x(t) into F∗ (x(t)) =
F(x(t), δt x(t), . . . , δtn x(t)). Analogously any arithmetic differential equation
f ∈ O(J n (X )) defines a map of sets f ∗ : X (R) → R, referred to as a
δ-function, which in affine coordinates sends α ∈ X (R) ⊂ R N into f ∗ (α) :=
F(α, δ p α, . . . , δ np α) ∈ R if F ∈ R[x, x  , . . . , x (n) ]ˆ represents f . If X = G is
in addition a group scheme and ψ ∈ Xn (G) then ψ∗ : G(R) → Ga (R) = R
is a group homomorphism called a δ-character of G. For X/R smooth and
R/ p R algebraically closed f is uniquely determined by f ∗ so one can identify
f with f ∗ and ψ with ψ∗ .

6.1.2.6 Prolongations of vector fields

For any vertical vector field
d
∂
ξ := ai (t, x)
∂ xi
i =1

on M/T one can consider the canonical prolongations

n d
∂
ξ (n) :=
j
(δt ai (t, x)) ( j)
j =0 i=1 ∂ xi

on J n (M/T ). The map

ξ (n) : C ∞ (J n (M/T )) → C ∞ (J n (M/T ))

is the unique R-derivation whose restriction to C ∞ (M) is ξ and which com-

mutes with the total derivative operator δt . The above construction has an
arithmetic analogue that plays a key technical role in the development of the
theory. Indeed for any affine smooth X/R and any R-derivation ξ : O(X ) →
O(X ) the canonical prolongation

ξ (n) : On (X )[1/ p] → On (X )[1/ p]

 Differential calculus with integers 147

is defined as the unique K = R[1/ p]-derivation whose restriction to O(X ) is

ξ and which commutes with φ. This construction then obviously globalizes.

6.1.2.7 Infinitesimal symmetries of differential equations

Some Galois theoretic concepts based on prolongations of vector fields have
arithmetic analogues. Indeed recall that if L ⊂ C ∞ (J n (M/T )) is a linear sub-
space of differential equations then a vertical vector field ξ on M/T is called
an infinitesimal symmetry of L if ξ (n) L ⊂ L; it is called a variational infinites-
imal symmetry if ξ (n) L = 0. Similarly given an R-submodule L ⊂ On (X ) an
infinitesimal symmetry of L is an R-derivation ξ : O(X ) → O(X ) such that
ξ (n) L ⊂ L[1/ p]. One says ξ is a variational infinitesimal symmetry of L if
ξ (n) L = 0.

6.1.2.8 Total differential forms on manifolds

Recall that a total differential form ([62], p. 351) on M/T is an expression that
in coordinates looks like a sum of expressions
F(t, x, x  , . . . , x (n) )d x j1 ∧ . . . ∧ d x ji ,
It is important to introduce an arithmetic analogue of this which we now
explain. We consider the case of top forms (i = d) which leads to what we
will call δ-line bundles. Denote by On the sheaf U → On (U ) on X for the
Zariski topology. Define a δ-line bundle of order n on X to be a locally free
On -module of rank 1. Integral powers of bundles need to be generalized as fol-
lows. Set W = Z[φ] (ring of polynomials with Z-coefficients in the symbol
 
φ). For w = as φ s write deg(w) = as . Also let W+ be the set of all

w= as φ s ∈ W with as ≥ 0 for all s. If L is a line bundle on X given by a

cocycle (gi j ) and w = ns=0 as φ s ∈ W , w = 0, an = 0, then define a δ-line

bundle L w of order n by the cocycle (giwj ), giwj = s φ s (gi j )as . With all these
definitions in place we may define the following rings which, by the way, are
the main objects of the theory:

Rδ (X, L) = H 0 (X, L w ).
0=w∈W+

Note that the above is a graded ring without unity. The homogeneous elements
of Rδ (X, L) can be viewed as arithmetic analogues of Lagrangian densities
and can also be referred to as arithmetic differential equations [16].

6.1.2.9 Differential forms on jet spaces and calculus of variations

The spaces
i↑ (J n (M/T ))
148 Alexandru Buium

of vertical smooth i -forms on J n (M/T ) (generated by i -wedge products of

forms d x, d x  , . . . , d x (n) ) play an important role in the calculus of variations
[62]. These spaces fit into a deRham complex where the differential is the
vertical exterior differential

d↑ : i↑ (J n (M/T )) → i↑+1 (J n (M/T ))

with respect to the variables x, x  , . . . , x (n) . On the other hand we have unique
operators
δt : i↑ (J n (M/T )) → i↑ (J n+1 (M/T ))

that commute with d↑ , induce a derivation on the exterior algebra, and for
i = 0 coincide with the total derivative operators. Then one can define spaces
of functional forms ([62], p. 357)

i↑ (J n (M/T ))
i∗ (J n (M/T )) =
I m(δt )
and the (vertical part of the) variational complex ([62], p. 361) with differen-
tials
d : i∗ (J n (M/T )) → i+1
∗ (J (M/T )).
n

4
The class of ω ∈ i↑ (J n (M/T )) in i∗ (J n (M/T )) is denoted by ωdt. For
4 4
i = 0, ω = F, have the formula d( Fdt) = E L(F)dt in i∗ (J 2n (M/T ))
d
where E L(F) = i =1 Fi d xi is the Euler–Lagrange total differential form,
n
 
j j ∂F
Fi = (−1) δt ( j)
.
j =0 ∂ x i

Noether’s theorem then says that for any vertical vector field ξ on M/T we
have the formula
ξ, E L(F) − ξ (n) (F) = δt G

for some G ∈ C ∞ (J 2n−1 (M/T )); G is unique up to a constant. If ξ is a vari-

ational infinitesimal symmetry of F then G is referred to as the conservation
law attached to this symmetry; in this case if x(t) is a solution of all Fi = 0
then G evaluated at x(t) will be a constant.
Analogously, for X/R smooth and affine, one can consider the modules
On (X )/R defined as the exterior powers of the inverse limit of the Käher
i

differentials
On (X )⊗(R/ p n R)/(R/ pn R) ,
 Differential calculus with integers 149

and the exterior differential

d : iOn (X )/R → i+1
On (X )/R .

Also one may consider the operators

φ ∗ : iOn (X )/R → iOn+1 (X )/R ;
again φ ∗ and d commute. Also note that any element in the image of φ ∗ is
uniquely divisible by pi ; for any ω ∈ iOn (X )/R and r ≥ 0 we then set ωr =
p −ir φ ∗r ω. The operation p −ir φ ∗r is a characteristic zero version of the inverse
Cartier operator. For any element μ ∈ R (which we refer to as eigenvalue) one
can define groups
iOn (X )/R
i∗ (J n (X )) =
I m(φ ∗ − μ)
that fit into a variational complex with differentials
d : i∗ (J n (X )) → i+1
∗ (J (X )).
n
4
The class of ω ∈ iOn (X )/R in i∗ (J n (X )) is denoted by ωdp. For i = 0,
4 4 
μ = 1, ω = F ∈ On (X ) we have d( Fdp) = { di=1 Fi ωni }dp in
1∗ (J 2n (X)) where Fi ∈ O2n (X ), ωi is a basis of O(X)/R , and ωni =
p −n φ ∗n ωi . Also an analogue of the Noether theorem holds in this context with

(F) := id=1 Fi ωni playing the role of the Euler–Lagrange form; indeed for
any vector field ξ on X there exists G ∈ O2n−1 (X ) such that
ξ (n) , (F) − ξ (n) (F) = G φ − G.
If ξ is a variational infinitesimal symmetry of F then G can be referred to as a
conservation law; in this case if P ∈ X (R) is a point which is a solution to all
Fi∗ (P) = 0 then G ∗ (P)φ = G ∗ (P).

6.1.2.10 Flows and Hamiltonian formalism [31]

We place ourselves in either the smooth or the complex analytic setting.
Assume one is given a section σ : M → J 1 (M/T ) of the projection
J 1 (M/T ) → M. To give such a σ is equivalent to giving a vector field δ M
on M lifting δt ; such a vector field can be referred to as a δt -flow on M/T . In
coordinates, if σ is defined by (t, x) → (t, x, s(t, x)), then
∂
δM = si (t, x) .
∂ xi
The composition
δt σ∗
L δ M : i↑ (M/T ) −→ i↑ (J 1 (M/T )) −→ i↑ (M/T )
150 Alexandru Buium

induces a derivation on the exterior algebra of vertical forms on M/T , com-

mutes with exterior vertical differentiation d↑ , and coincides with δ M for i = 0;
one can refer to L δ M as the “Lie derivative” attached to the vector field δ M (it
is a relative version of the usual Lie derivative).
We consider now the “time-dependent” Hamiltonian formalism. The classi-
cal treatment of this formalism (e.g. [56]) usually involves the differential of
time, dt. Since “dp” does not exist in the arithmetic setting what we will do
will be to first find a presentation of the classical time dependent Hamiltonian
formalism without reference to dt and then we shall transpose the latter to the
arithmetic case. Assume for simplicity that M has dimension 3 so the fibers
of M → T are 2-dimensional. A vertical symplectic form on M is a nowhere
vanishing vertical 2-form on M. A vertical contact form on M is a vertical
1-form ν on M such that dν is symplectic. We shall drop the word “vertical”
in what follows. Recall that a δt -flow on M is a vector field δ M on M that
lifts the vector field δt on T ; and recall that to give a δ-flow on M is the same
as to give a section of the projection J 1 (M/T ) → M. Given a δ-flow δ M as
above we may consider the relative Lie derivative L δ M on the vertical forms
i (M/T ). A δt -flow δ M will be called Hamiltonian with respect to a sym-
plectic form η on M if L δ M η = μ · η for some function μ on T . Note that if
δt γ = −μ · γ for some function γ on T and if η1 = γ η then L δ M η1 = 0.
If ν is a contact form, η = d↑ ν, and δ M is a δt -flow which is Hamiltonian
with respect to η (hence L δ M η = μη) then  := L δ M ν − μν is closed; the
latter can be referred to as an Euler–Lagrange form attached to the symplectic
form η and the δt -flow δ M . If  happens to be exact, i.e. L δ M ν − μν = dL
for some function L on M, then L can be referred to as a Lagrangian attached
to η and δ M . (This Euler–Lagrange formalism is related to, but does not coin-
cide with, the Euler–Lagrange formalism in 6.1.2.9.) In a series of interesting
applications (especially in the complex analytic setting of the above, e.g. to
Painlevé equations [56]) M is taken to be J 1 (N /T ) where N → T is a family
of curves; in this case a 1-form on M is called canonical if it is a function on
M times the pull-back of a 1-form on N . Also one is interested, in this situ-
ation, in δt -flows defined by sections of J 1 (M/T ) → M = J 1 (N/T ) that
arise from sections of J 2 (N/T ) → J 1 (N/T ) (where J 2 (N /T ) is naturally
embedded into J 1 (M/T )); such δt -flows are referred to as canonical and are
defined by order 2 differential equations (functions on J 2 (N )).
A justification for the above terminology comes, for instance, from the
following local algebraic considerations. Let F be a field equipped with a
derivation δ : F → F (e.g. F = C((t)), δ = δt ) and let A be a ring isomor-
phic to a power series ring in 2 variables with coefficients in F. Choose formal
coordinates x, y ∈ A (i.e. A = F[[x, y]]), and set  A = A · d x + A · dy,
 Differential calculus with integers 151

2A = A · d x ∧ dy. A form η = ud x ∧ dy ∈ 2A is called invertible if

u(0, 0) = 0. (These definitions are independent of the choice of the formal
coordinates x, y.) The following is a translation into the language of differen-
tial algebra of the classical Hamiltonian/Lagrangian formalism in 2 analytic
coordinates. Assume one is given a derivation δ : A → A such that δ F ⊂ F
and sending the maximal ideal of A into itself. Then the following are easy to
check:
1) Assume we are given an invertible η ∈ 2A . Then there exist formal coor-
dinates x, y ∈ A such that η = dy ∧ d x. Furthermore if δη = 0 then and
there exists an element H ∈ A such that
∂H ∂H
δy = − , δx = .
∂x ∂y
2) Assume there exist formal coordinates x, y ∈ A, and a series H ∈ A such
that δy = − ∂H ∂H
∂ x and δx = ∂y . Let η = dy ∧ d x, ν = yd x,  = δν, and
L = y ∂H
∂y − H = yδx − H. Then

dν = η, δη = 0, dL = .
∂2H
3) Assume the notation in 2) and assume the series ∂ y2
is invertible in
∂
F[[x, y]]. Then there exists a unique F-derivation denoted by ∂δx on
F[[x, y]] sending x → 0 and δx → 1. Moreover we have
 
∂L ∂L ∂H ∂L ∂L
= y, =− , δ = .
∂δx ∂x ∂x ∂δx ∂x
We want to introduce an arithmetic analogue of the above formalism;
cf. [31]. To simplify our discussion let X be a smooth surface over R (or the
p-adic completion of such a surface). Recall that by an i-form on X we under-
stand a global section of iX (projective limit of usual Kähler differentials of
X mod p n ). A symplectic form on X is an invertible 2-form on X . A contact
form on X is a 1-form on X such that dν is symplectic. By a δ-flow on X we
will understand a section X → J 1 (X ) of the projection J 1 (X ) → X . Let us
further specialize the situation to the case when X = J 1 (Y ) where Y is a curve
(affine, to simplify). A 1-form ν on X = J 1 (Y ) is called canonical if ν = gβ
with g ∈ O(X ) and β a 1-form on Y . If ν is a canonical 1-form on X = J 1 (Y )
and is closed, i.e. dν = 0, then ν is (the pull-back of) a 1-form on Y . Note
that J 2 (Y ) naturally embeds into J 1 (J 1 (Y )). A δ-flow on X = J 1 (Y ) will be
called canonical if the section J 1 (Y ) = X → J 1 (X ) = J 1 (J 1 (Y )) factors
through a section J 1 (Y ) → J 2 (Y ). This concept is, however, too restrictive
for applications. We will need to relax the concept of canonical δ-flow as fol-
lows. To simplify the discussion we restrict to the case of order 2 differential
152 Alexandru Buium

equations. Let f ∈ O(J 2 (Y )) and let Z = Z 2 ( f ) ⊂ J 2 (Y ) be the closed

formal subscheme defined by f . For i ≥ 1 we define
iJ n (Y )
iZ = .
 f iJ n (Y ) , d f ∧ iJ−1
n (Y ) 

So 1 
Z =  Z . Also we have natural wedge products
j i+ j
∧ : iZ ×  Z →  Z ,
and exterior differentiation maps
d : iZ → i+1
Z .

We say that f defines a generalized canonical δ-flow on J 1 (Y ) if the natural

map
π ∗  J 1 (Y ) → Z
is injective and has a cokernel annihilated by a power of p. Note now that the
natural maps φ ∗ : iJ 1 (Y ) → iJ 2 (Y ) induce natural maps denoted by

φ ∗Z : iJ 1 (Y ) → iZ .
We say that the generalized canonical δ-flow is Hamiltonian with respect
to a sympletic form η on J 1 (Y ) if φ ∗Z η = μ · η in 2
Z 2( f )
for some
μ ∈ p R. If in addition η = dν for some canonical 1-form ν on J 1 (Y ) we
call  := 1p (φ ∗Z ν − μν) ∈ Z 2 ( f ) an Euler–Lagrange form. (This arithmetic
Euler–Lagrange formalism is related to, but does not coincide with, the arith-
metic Euler–Lagrange formalism in 6.1.2.9.) In the notation above p is closed
in the sense that d( p) = 0 in 2 Z 2( f )
. The above is the framework for the main
results in [31] about the arithmetic analogue of Painlevé VI.

6.1.2.11 “Category of differential equations”

A categorical framework for differential equations on manifolds can be intro-
duced (cf. [1], for instance). In one variant of this the objects are locally
isomorphic to projective systems of submanifolds of J n (M/T ) compatible
with the total derivative operator and morphisms are smooth maps between
these, again compatible with the total derivative operator. This categorical
framework has an arithmetic analogue as follows. Note first that φ acts nat-
urally on Rδ (X, L) but δ does not. Nevertheless for any homogeneous s ∈
Rδ (X, L) of degree v the ring Rδ (X, L)s of all fractions f /s w with f homo-
geneous of degree wv has a natural p-derivation δ on it. This inductive system
X δ (L) = (Rδ (X, L)s , δ)s of rings equipped with p-derivations δ is an object
 Differential calculus with integers 153

of a natural category underlying a geometry more general than algebraic geom-

etry which we refer to as δ-geometry [16]. This geometry is an arithmetic
analogue of the categorical setting in [1] and also an arithmetic analogue of
the Ritt–Kolchin δ-algebraic geometry [52, 9]. If one restricts to étale maps
of smooth schemes we have a functor X → X δ = X δ (K ν ) from “algebraic
geometry” to our “δ-geometry” by taking L = K ν to be a fixed power of the
canonical bundle; the natural choice later in the theory turns out to be the anti-
canonical bundle L = K −1 . In δ-geometry X δ should be viewed as an infinite
dimensional object.

6.1.2.12 Differential invariants

Recall the concept of differential invariant which plays a key role in the
“Galois theoretic” work of Lie and Cartan. Assume that a group G acts
on M and T such that M → T is G-equivariant. (In this discussion
assume d := dim T is arbitrary.) Then G acts on J n (M/T ) and the ring
C ∞ (J n (M/T ))G of G-invariant elements in the ring C ∞ (J n (M/T )) is called
the ring of differential invariants. There are two extreme cases of special inter-
est: the case when G acts trivially on T and the case when G acts transitively
on T . A special case of the first extreme case mentioned above is that in which
M = F × T , with G a Lie group acting trivially on T and transitively on
F; this leads to the context of Cartan’s moving frame method and of Cartan
connections. For dim T = 1, T is the “time” manifold, F is the “physical
space”, and sections in (M/T ) correspond to particle trajectories. A special
case of the second extreme case mentioned above is the situation encountered
in the study of “geometric structures” and in the formulation of field theories,
in which G is the group Diff(T ) of diffeomorphisms of T , T is viewed as
the “physical space” or “physical space-time”, and M is a natural bundle over
T , i.e. a quotient M = \Repn (T ) of the bundle Repn (T ) → T of n-jets
of frames (Rd , 0) → T by a Lie subgroup of the group Autn (Rd , 0) of
n-jets of diffeomorphisms of (Rd , 0); cf. [1], pp. 150-153 and 183. (E.g. the
Riemannian metrics on T identify with the sections in (M1,O(d) /T ) where
O(d) < G L(d) = Aut1 (Rd , 0) is the orthogonal group.) In the case when the
action of G on T is trivial the ring of differential invariants above turns out
to have an interesting arithmetic analogue, namely the ring of δ-invariants of
a correspondence; cf. the discussion below. The special case of Cartan con-
nections has also an arithmetic analogue: the arithmetic logarithmic derivative
attached to a δ-flow. The latter has a flavor different from that of δ-invariants
of correspondences and will be discussed later. There are also interesting can-
didates for arithmetic analogues of the situation when G = Diff(T ); cf. our
discussion of Lie groupoids below.
154 Alexandru Buium

We next explain the ring of δ-invariants of a correspondence. Pursuing an

analogy with usual geometric invariant theory assume we are given (X, L), a
correspondence on X (i.e. a morphism σ = (σ1 , σ2 ) : X̃ → X × X ), and a
linearization of L (i.e. an isomorphism β : σ1∗ L σ2∗ L). Then we may define
the ring of δ-invariants of σ by
Rδ (X, L)σ = { f ∈ Rδ (X, L); βσ1∗ f = σ2∗ f }.
(This ring, again, has no unity!) The homogeneous elements s of Rδ (X, L)σ
can be viewed as arithmetic analogues of total differential forms invariant
under appropriate symmetries. The inductive system of rings
(Rδ (X, L)σs , δ)
equipped with p-derivations δ can be viewed as an incarnation of the quotient
space “X δ /σδ ” in δ-geometry (and, under quite general hypotheses, is indeed
the categorical quotient of X δ by σδ ); note that, in most interesting examples
(like the ones in Theorem 6.6 below), the quotient X/σ does not exist in usual
algebraic geometry (or rather the categorical quotient in algebraic geometry
reduces to a point).
It is worth revisiting the sequence of ideas around differential invariants.
Classical Galois theory deals with algebraic equations (satisfied by numbers).
Lie and Cartan extended Galois’ ideas, especially through the concept of differ-
ential invariant, to the study of differential equations (satisfied by functions);
roughly speaking they replaced numbers by functions. In the theory presented
here functions are replaced back by numbers. But we did not come back
to where things started because we have added new structure, the operator
δ = δ p . In particular the δ-invariants mentioned above, although arithmetic in
nature, and although attached to algebraic equations (defining X , X̃ , σ ), are
nevertheless not “Galois theoretic” in any classical sense.

6.1.2.13 Lie groupoids

Let us go back to the (interrelated) problems of finding arithmetic analogues
of T × T and of Diff(T ). The arithmetic analogue of S := T × T is usually
referred to as the hypothetical tensor product “Z ⊗F1 Z” over the “field with
one element”; cf. [57] for some history of this. The arithmetic analogue of
Diff(T ) could be thought of as the Galois group “Gal(Q/F1 )”. According to a
suggestion of Borger [6] one should take “Z⊗F1 Z” to be, by definition, the big
Witt ring of Z. Since we are here in a local arithmetic situation it is convenient,
for our purposes, to take, as an arithmetic analogue of S = T × T , the schemes
'm = Spec Wm (R), where Wm is the p-typical functor of Witt vectors of
length m + 1 (we use Borger’s indexing) and R is a complete local p-ring. On
 Differential calculus with integers 155

the other hand an infinitesimal analogue of Diff(T ) is the Lie groupoid of T

defined as the projective system of groupoids Gn (T ) := J n (S/T )∗ (where the
upper ∗ means taking “invertible” elements and S = T × T → T is the second
projection.) So an arithmetic analogue of the system J n (S/T )∗ (which at the
same time would be an infinitesimal analogue of “Gal(Q/F1 )”) would be the
system J n ('m ) equipped with the natural maps induced by the comonad map.
This system is a rather non-trivial object [29]. It is worth noting that a good
arithmetic analogue of J n (S/T )∗ could also be the “usual” jet spaces (in the
sense of [8]) of Wm (R)/R (which in this case can be constructed directly from
the module of Käher differentials Wm (R)/R ). By the way Wm (R)/R is also the
starting point for the construction of the deRham–Witt complex [44]. However
 involves usual derivations (rather than Fermat quotients) so taking  as a
path to an arithmetic analogue of J n (S/T )∗ seems, again, like “going arith-
metic halfway”. On the contrary taking the system J n ('m ) as the analogue of
J n (S/T )∗ seems to achieve, in some sense, the task of “going arithmetic all
the way”.
We end by remarking that if we denote π1 and π2 the source and target pro-
jections from Gn (T ) into T then there are natural “actions” ρ of the groupoids
Gn (T ) on all natural bundles Mn, := \Repn (T ) → T fitting into diagrams:
ρ
Mn, ×T ,π1 Gn (T ) −→ Mn,
p2 ↓ ↓
π2
Gn (T ) −→ T
where p2 is the second projection. The above induce “actions”
ρ
J n (Mn, /T ) ×T ,π1 Gn (T ) −→ J n (Mn, /T )
p2 ↓ ↓
π2
Gn (T ) −→ T
where p2 is again the second projection. One can consider rings of differential
invariants
C ∞ (J n (Mn, /T ))ρ := {F ∈ C ∞ (J n (Mn, /T )); F ◦ ρ = F ◦ p1 }
where p1 : J n (Mn, /T ) ×T ,π1 Gn (T ) → J n (Mn, /T ) is the fist projec-
tion. There should be arithmetic analogues of the above “actions” and rings
of differential invariants. One can argue that the analogue of Repn (T ) is,
again, the system J n ('m ). Then for = 1 the analogue of J n (Mn, /T )
might be J (J ('m )); the analogue, for
n n = 1, of the “action” ρ above
could then be the map J n (J n ('m  ,m  )) → J n (J n ('m  +m  )) where 'm  ,m  =
Spec Wm  (Wm  (R)). A challenge would then be to find arithmetic analogues
of non-trivial s.
156 Alexandru Buium

6.1.2.14 Differential Galois theory of linear equations

This subject is best explained in a complex (rather than real) situation. Classi-
cally (following Picard-Vessiot and Kolchin [51]) one starts with a differential
field F of meromorphic functions on a domain D in the complex plane C and
one fixes an n × n matrix A ∈ gln (F) in the Lie algebra of the general linear
group G L n (F). The problem then is to develop a “differential Galois theory”
for equations of the form
δz U = A · U

where U ∈ G L n (G), G a field of meromorphic functions on a subdomain of

D, and z a coordinate on D. The start of the theory is as follows. One fixes a
solution U and introduces the differential Galois group G U/F of U/F as the
group of all F−automorphisms of the field F(U ) that commute with d/dz.
One can ask for an arithmetic analogue of this theory. There is a well devel-
oped difference algebra analogue of linear differential equations [65]; but the
arithmetic differential theory is still in its infancy [33, 34, 35]. What is being
proposed in loc. cit. in the arithmetic theory is to fix a matrix α ∈ gln (R) and
define δ-linear equations as equations of the form

δu = α · u ( p)

where u = (u i j ) ∈ G L n (R), δu := (δu i j ), and u ( p) := (u i j ). Note that the

p

above equation is equivalent to φ(u) =  · u ( p) where  = 1 + pα which is not

a difference equation for φ in the sense of [65]; indeed difference equations
for φ have the form φ(u) =  · u. To define the δ-Galois group of such an
equation start with a δ-subring O ⊂ R and let u ∈ G L n (R) be a solution of
our equation. Let O[u] ⊂ R the ring generated by the entries of u; clearly
O[u] is a δ-subring of R. We define the δ-Galois group G u/O of u/O as the
subgroup of all c ∈ G L n (O) for which there exists an O-algebra automorphism
σ of O[u] such that σ ◦ δ = δ ◦ σ on O[u] and such that σ (u) = uc. The theory
starts from here.

6.1.2.15 Maurer–Cartan connections

The Maurer–Cartan connection attached to a Lie group G is a canonical map
T (G) → L(G) from the tangent bundle T (G) to the Lie algebra L(G); for
G = G L n it is given by the form dg · g −1 and its algebraic incarnation is
Kolchin’s logarithmic derivative map [51] G L n (G) → gln (G), u → δz u ·
u −1 . In our discussion of linear equations above the arithmetic analogue of the
Kolchin logarithmic derivative map is the map G L n (R) → gln (R),

u → δu · (u ( p) )−1 .
 Differential calculus with integers 157

This map is naturally attached to the lift of Frobenius φG L n ,0 : G L n → G Ln

whose effect on the ring O(G L n ) = R[x, det(x)−1 ]ˆ is φG L n ,0 (x) = x ( p) .
The latter lift of Frobenius behaves well (in a precise sense to be explained
later) with respect to the maximal torus T ⊂ G L n of diagonal matrices
and with respect to the Weyl group W ⊂ G L n of permutation matrices but
it behaves “badly” with respect to other subgroups like the classical groups
S L n , S On , Spn . (This phenomenon does not occur in the geometry of Lie
groups where the Maurer–Cartan form behaves well with respect to any Lie
subgroup of G L n , in particular with respect to the classical groups.) In order
to remedy the situation one is lead to generalize the above constructions by
replacing φG L n ,0 with other lifts of Frobenius φG L n : GL n → G L n that are
adapted to each of these classical groups. Here is a description of the resulting
framework.
First we define an arithmetic analogue of the Lie algebra gln of G L n as the
set gln of n ×n matrices over R equipped with the non-commutative group law
+δ : gln × gln → gln given by

a +δ b := a + b + pab,

where the addition and multiplication in the right hand side are those of gln ,
viewed as an associative algebra. There is a natural “δ-adjoint” action *δ of
G L n on gln given by

a *δ b := φ(a) · b · φ(a)−1 .

Assume now one is given a ring endomorphism φG L n of O(G L n ) lifting

Frobenius, i.e. a ring endomorphism whose reduction mod p is the p-power
Frobenius on O(G L n )ˆ/( p) = k[x, det(x)−1 ]; we still denote by φG L n :
G L n → G L n the induced morphism of p-formal schemes and we refer to
it as a lift of Frobenius on G L n or simply as a δ-flow on G L n . Consider the
matrices +(x) = (φG L n (xi j )) and x ( p) = (xi j ) with entries in O(G L n )ˆ; then
p

+(x) = x ( p) + p(x) where (x) is some matrix with entries in O(G L n )ˆ.
Furthermore given a lift of Frobenius φG L n as above one defines, as usual, a
φ ( f )− f p
p-derivation δG L n on O(G L n )ˆ by setting δG L n ( f ) = G L n p .
Assume furthermore that we are given a smooth closed subgroup scheme
G ⊂ G L n . We say that G is φG L n -horizontal if φG L n sends the ideal of G
into itself; in this case we have a lift of Frobenius endomorphism φG on G, !
equivalently on O(G)ˆ.
Assume the ideal of G in O(G L n ) is generated by polynomials f i (x). Then
recall that the Lie algebra L(G) of G identifies, as an additive group, to the
158 Alexandru Buium

subgroup of the usual additive group (gln , +) consisting of all matrices a

satisfying
“ −1 ” f i (1 + a) = 0,
(φ)
where  2 = 0. Let f i be the polynomials obtained from f i by applying φ
to the coefficients. Then we define the δ-Lie algebra L δ (G) as the subgroup of
(gln , +δ ) consisting of all the matrices a ∈ gln satisfying
(φ)
p −1 f i (1 + pa) = 0.

The analogue of Kolchin’s logarithmic derivative (or of the Maurer–Cartan

connection) will then be the map, referred to as the arithmetic logarithmic
derivative, lδ : G L n → gln , defined by
1 
lδa := φ(a)+(a)−1 − 1 = (δa − (a))(a ( p) + p(a))−1 .
p
For G a φG L n -horizontal subgroup lδ above induces a map lδ : G → L δ (G).
Now any α ∈ L δ (G) defines an equation of the form lδu = α, with unknown
u ∈ G; such an equation will be referred to as a δ-linear equation (with respect
to our δ-flow). Later in the paper we will explain our results about existence
of δ-flows on G L n compatible with the classical groups. These δ-flows will
produce, as explained above, corresponding δ-linear equations.

6.1.2.16 Symmetric spaces

In the classical Cartan theory of symmetric spaces one starts with a (real) Lie
group G equipped with a Lie group automorphism x → x τ of order 1 or
2 (which we refer to as an involution). One defines the fixed group of the
involution
G + := {a ∈ G; a τ = a} (6.1)

and homogeneous spaces G/S where (G + )◦ ⊂ S ⊂ G + are Lie groups inter-

mediate between the identity component (G + )◦ and G + . Then τ acts on the Lie
algebra L(G) and one denotes by L(G)+ and L(G)− the + and − eigenspaces
of this action. We have a decomposition L(G) = L(G)+ ⊕ L(G)− which can
be referred to as a Cartan decomposition corresponding to τ . One can also
consider the closed subset of G defined by

G − := {a ∈ G; a τ = a −1 }. (6.2)

This is not a subgroup of G. Set a −τ = (a τ )−1 . The map H : G → G − ,

H(a) := a −τ a identifies G/G + with a subset of G − and there is an obvious
compatibility between the tangent map to H and the Cartan decomposition.
 Differential calculus with integers 159

All of the above has an arithmetic analogue as follows. We start with a linear
smooth group scheme G over R and an involution τ on G. We define G + and
G − as in 6.1 and 6.2 respectively. One also has an action of τ on the δ-Lie alge-
bra L δ (G). One considers the subgroup L δ (G)+ ⊂ L δ (G) of all b ∈ L δ (G)
such that bτ = b. One also considers the closed subscheme L δ (G)− ⊂ L δ (G)
whose points b satisfy bτ +δ b = 0; L δ (G)− is not a subgroup. For G = G L n
one easily proves that the map +δ : L δ (G)+ × L δ (G)− → L δ (G) is a bijec-
tion on points; this can be viewed as an analogue of the Cartan decomposition.
Later we will state results having the above as background.
The classical theory of symmetric spaces also considers bilinear (positive
definite) forms B on L(G) such that G + acts on L(G) via the adjoint action
by orthogonal transformations with respect to B; it would be interesting to find
an arithmetic analogue of this condition.

6.1.3 Main task of the theory

At this point we may formulate the main technical tasks of the theory. Let, from
now on, R = Z  ur . First given a specific scheme X (or group scheme G) the
p
task is to compute the rings On (X ) (respectively the modules Xn (G)). More
generally given a specific pair (X, L) we want to compute the ring Rδ (X, L).
Finally given (X, L), a correspondence σ on X , and a linearization of L, we
want to compute the ring Rδ (X, L)σ . The main applications of the theory
(cf. the subsection below on motivations) arise as a result of the presence of
interesting/unexpected elements and relations in the rings On (X ), Rδ (X, L),
Rδ (X, L)σ .

6.1.4 Motivations of the theory

There are a number of motivations for developing such a theory.

6.1.4.1 Diophantine geometry

Usual differential equations satisfied by functions can be used to prove dio-
phantine results over function fields (e.g. the function field analogues of the
Mordell conjecture [55] and of the Lang conjecture [8]). In the same vein
one can hope to use “arithmetic differential equations” satisfied by numbers
to prove diophantine results over number fields. This idea actually works in
certain situations, as we will explain below. Cf. [12, 19]. The general strat-
egy is as follows. Assume one wants to prove that a set S of points on an
algebraic variety is finite. What one tries to do is find a system of arithmetic
differential equations Fi (α, δ p α, . . . , δ np α) = 0 satisfied by all α ∈ S; then
160 Alexandru Buium

one tries, using algebraic operations and the “differentiation” δ p , to eliminate

δ p α, . . . , δ np α from this system to obtain another system G j (α) = 0 satisfied
by all α ∈ S, where G j do not involve the “derivatives” anymore. Finally one
proves, using usual algebraic geometry that the latter system has only finitely
many solutions.

6.1.4.2 “Impossible” quotient spaces

If X is an algebraic variety and σ : X̃ → X × X is a correspondence on X
then the categorical quotient X/σ usually reduces to a point in the category
of algebraic varieties. In some sense this is an illustration of the limitations
of classical algebraic geometry and suggests the challenge of creating more
general geometries in which X/σ becomes interesting. (The non-commutative
geometry of A. Connes [38] serves in particular this purpose.) As explained
above we proposed, in our work, to replace the algebraic equations of usual
algebraic geometry by “arithmetic differential equations”; the resulting new
geometry is called δ p -geometry (or simply δ-geometry). Then it turns out that
important class of quotients X/σ that reduce to a point in usual algebraic
geometry become interesting in δ-geometry (due to the existence of interest-
ing δ-invariants). A general principle seems to emerge according to which this
class coincides with the class of “analytically uniformizable” correspondences.
Cf. [16].

6.1.4.3 “Impossible” liftings to characteristic zero

A series of phenomena belonging to algebraic geometry in characteristic p,
which do not lift to characteristic 0 in algebraic geometry, can be lifted nev-
ertheless to characteristic 0 in δ-geometry. This seems to be a quite general
principle with various incarnations throughout the theory (cf. [15, 16, 25])
and illustrates, again, how a limitation of classical algebraic geometry can be
overcome by passing to δ-geometry.

6.1.5 Comparison with other theories

It is worth noting that the paradigm of our work is quite different from the
following other paradigms:

1) Dwork’s theory of p-adic differential equations [41] (which is a the-

ory about δt acting on functions in Q p [[t]] and not about δ p acting on
numbers; also Dwork’s theory is a theory of linear differential equa-
tions whereas the theory here is about analogues of non-linear differential
equations),
 Differential calculus with integers 161

2) Vojta’s jet spaces [66] (which, again, although designed for arithmetic
purposes, are nevertheless constructed from Hasse–Schmidt derivations
1 n
“ n! δt ” acting on functions and not from operators acting on numbers),
3) Ihara’s differentiation of integers [47] (which, although based on Fermat
quotients, goes from characteristic zero to characteristic p and hence,
unlike our δ p , cannot be iterated),
4) the point of view of Kurokawa et al. [53] (which uses an operator on num-
bers very different from δ p namely ∂α ∂ p := np
n−1 β for α = p n β ∈ Z,

p |β),
5) the theory of Drinfeld modules [39](which is entirely in characteristic p),
6) the difference geometry in the work of Cohn, Hrushovski-Chatzidakis
[37], and others (in which the jet spaces are n-fold products of the original
varieties as opposed to the jet spaces in our work which are, as we shall
see, bundles over the original varieties with fibers affine spaces),
7) Raynaud’s deformation to Witt vectors W2 (k) over a field k of charac-
teristic p [64] (which again leads to operators from characteristic zero
to characteristic p which cannot be iterated; loosely speaking W2 (k) in
Raynaud’s approach is replaced in our theory by W2 (W (k))).
8) the work of Soulé, Deitmar, Connes, Berkovich, and many others on the
“geometry over the field F1 with one element”. In their work passing from
the geometry over Z to the geometry over F1 amounts to removing part
of the structure defining commutative rings, e.g. removing addition and
hence considering multiplicative monoids instead of rings. On the contrary
our theory can be seen as a tentative approach to F1 (cf. the Introduction
to [16]) that passes from Z to F1 by adding structure to the commuta-
tive rings, specifically adding the operator(s) δ p . This point of view was
independently proposed (in a much more systematic form) by Borger [6].
Borger’s philosophy is global in the sense that it involves all the primes
(instead of just one prime as in our work) and it also proposes to see
“positivity” as the corresponding story at the “infinite” prime; making our
theory fit into Borger’s larger picture is an intriguing challenge.
9) the work of Joyal [49] and Borger [4, 5] on the Witt functor; the Witt
functor is a right adjoint to the forgetful functor from “δ-rings” to rings
as opposed to the arithmetic jet functor which is a left adjoint to the same
forgetful functor. As it is usually the case the left and right stories turn out
to be rather different.
10) the theory of the Greenberg transform, cf. Lang’s thesis and [42] (which
attaches to a scheme X/R varieties G n (X ) over k; one can show [12] that
G n (X ) J n (X ) ⊗ R k so the arithmetic jet spaces are certain remarkable
liftings to characteristic zero of the Greenberg transforms. The operators
162 Alexandru Buium

δ on On (X ) do not survive after reduction mod p as operators on the

Greenberg transforms.)
11) the work on the deRham–Witt complex, cf., e.g. [44] (which has as its
starting point the study of Kähler differential of Witt vectors; on the con-
trary, what our research suggests, cf. [29], is to push arithmetization one
step further by analyzing instead the arithmetic jet spaces of Witt vectors.)
12) the theory φ-modules (which is a theory about linear equations as opposed
to the theory here which is non-linear).

6.2 Main results

We present in what follows a sample of our results. We always set A = A ⊗Z
F p = A/ p A, X = X ⊗Z F p for any ring A and any scheme X respectively.
! the p-adic completion of A; for X Noetherian we
Recall that we denote by A
denote by !X the p-adic completion of X . Also, in what follows, R := Zur ,
p
k := R/ p R. We begin with completely general facts:

6.2.1 Affine fibration structure of p-jet spaces

Theorem 6.1. [11]

1) If X/R is a smooth scheme of relative dimension d then X has an affine

covering X i such that J n (X i ) 5 
Xi × A nd in the category of p-adic formal

schemes.
2) If G/R is a smooth group scheme of relative dimension d, with formal
group law F ∈ R[[T1 , T2 ]]d (T1 , T2 d-tuples), then the kernel of J n (G) →
! is isomorphic to A
G  nd with composition law obtained from the formal

series
(n) (n)
δF, . . . , δ n F ∈ R[[T1 , T2 , . . . , T1 , T2 ]]d

by setting T1 = T2 = 0.

Note that after setting T1 = T2 = 0 the series δ n F become restricted,

(n) (n)
i.e. elements of (R[T1 , T2 , . . . , T1 , T2 ]ˆ)d so they define morphisms in the
category of p-adic formal schemes. Assertion 1) in the theorem makes p-jet
spaces resemble the usual jet spaces of the Lie–Cartan theory. Note however
that, even if G is commutative, the kernel of J n (G) → G ! is not, in general
the additive group raised to some power. Here is the idea of the proof of 1)
for n = 1. We may assume X = Spec A, A = R[x]/( f ), and there is an
 Differential calculus with integers 163

étale map R[T ] ⊂ A with T a d-tuple of indeterminates. Consider the unique

ring homomorphism R[T ] → W1 (A[T  ]) sending T → (T, T  ) where W1 is
the functor of Witt vectors of length 2 and T  is a d-tuple of indeterminates.
Using the fact that R[T ] ⊂ A is formally étale and the fact that the first projec-
tion W1 (A[T  ]/( p n )) → A[T  ]/( p n ) has a nilpotent kernel, one constructs
a compatible sequence of homomorphisms A → W1 (A[T  ]/( p n )), T →
(T, T  ). Hence one gets a homomorphism A → W1 (A[T  ]ˆ), a → (a, δa).
Then one defines a homomorphism R[x, x  ]ˆ/( f, δ f ) → A[T  ]ˆ by sending
x → a := class(x) ∈ A, x  → δa. Conversely one defines a homomorphism
A[T  ]ˆ → R[x, x  ]ˆ/( f, δ f ) by sending T  → δT . The two homomorphisms
are inverse to each other which ends the proof of 1) for n = 1.

6.2.2 δ-functions and δ-characters on curves

The behavior of the rings On (X ) for smooth projective curves X depends on
the genus of X as follows:

Theorem 6.2. [11, 12, 32] Let X be a smooth projective curve over R of
genus g.

1) If g = 0 then On (X ) = R for all n ≥ 0.

2) Let g = 1. If X is not a canonical lift then O1 (X ) = R (hence X1 (X ) = 0)
and X2 (X ) is a free module of rank 1; in particular O2 (X ) = R. If, on the
other hand, X is a canonical lift then O1 (X ) = O(A 51 ) and X1 (X ) is free of
rank one.
3) If g ≥ 2 then J n (X ) is affine for n ≥ 1; in particular O1 (X ) separates the
points of X (R).

The proof of 1) is a direct computation. The idea of proof of the statements

about Xn in 2) is as follows. Let N n = K er (J n (X ) → ! X ). Then one first
proves (using Theorem 6.1) that H om(N n , G!a ) has rank at least n over R and
one computes the ranks of Xn (X ) by looking at the exact squence
!a ) → H om(N 2 , G
H om(J 2 (X ), G !a ) → H 1 (X, O).

Here H om = H om gr . The proof of 3) is based on representing J 1 (X ) as

P(E)\D where E is a rank 2 vector bundle on X , and D is an ample divisor.
If g = 1 and X is not a canonical lift then a basis ψ for X2 (X ) can be viewed
as an analogue of the “Manin map” of an elliptic fibration [55]. Also note
that assertion 3) in Theorem 6.2 implies the effective version of the Manin–
Mumford conjecture [12]. Indeed Manin and Mumford conjectured that if X
164 Alexandru Buium

is a complex curve of genus ≥ 2 embedded into its Jacobian A then X ∩ Ator s

is a finite set. This was proved by Raynaud [64]. Mazur later asked [59] if
-(X ∩ Ator s ) ≤ C(g) where C(g) is a constant that depends only on the genus
g of X . Using 3) in Theorem 6.2 one can prove:

Theorem 6.3. [12] For a smooth projective complex curve X in its Jacobian
A we have -(X ∩ Ator s ) ≤ C(g, p) where C(g, p) is a constant that depends
only on the genus g and (in case X is defined over Q) on the smallest prime p
of good reduction of X .

Roughly speaking the idea of proof is as follows. First one can replace
the complex numbers by R and Ator s by its prime to p torsion subgroup
< A(R). Then one embeds X (R) ∩ (via the “jet map”) into the the set
of k-points of J 1 (X ) ∩ p J 1 (A). But the latter is a finite set because J 1 (X )
is affine, p J 1 (A) is projective, and both are closed in J 1 (A). Moreover the
cardinality of this finite set can be bounded using Bézout’s theorem.
One can ask for global vector fields on a smooth projective curve X/R
that are infinitesimal symmetries for given δ-functions on X . The only non-
trivial care is that of genus 1 (elliptic curves); indeed for genus 0 there are no
non-constant δ-functions (cf. Theorem 6.2) while for genus ≥ 2 there are no
non-zero vector fields. Here is the result:

Theorem 6.4. [11, 16] Let X be an elliptic curve over R with ordinary
reduction.

1) If X has Serre–Tate parameter q(X ) ≡ 1 mod p2 then there exists a non-

zero global vector field on X which is a variational infinitesimal symmetry
for all the modules Xn (X ), n ≥ 1.
2) If X is a canonical lift (equivalently has Serre–Tate parameter q(X ) = 1)
then there is no non-zero global vector field on X which is a variational
infinitesimal symmetry of X1 (X ).

Finally here is a computation of differentials of δ-characters.

Theorem 6.5. [13, 16] Assume X is an elliptic curve over R which is not a
canonical lift and comes from an elliptic curve X Z p over Z p . Let a p ∈ Z be
the trace of Frobenius of the reduction mod p of X Z p and let ω be a basis for
the global 1-forms on X Z p . Then there exists an R-basis ψ of X2 (X ) such that

p · dψ = (φ ∗2 − a p φ ∗ + p)ω.
 Differential calculus with integers 165

In particular for the eigenvalue μ = 1 we have

.  .
p·d ψdp = (1 − a p + p) · ωdp.

A similar result holds in case X is a canonical lift.

6.2.3 δ-invariants of correspondences

Here is now a (rather roughly stated) result about δ-invariants of correspon-
dences on curves; for precise statements we refer to [16].

Theorem 6.6. [16] The ring Rδ (X, K −1 )σ is “δ-birationally equivalent” to

the ring Rδ (P1 , O(1)) if the correspondence σ on X “comes from” one of the
following cases:

1) (spherical case) The standard action of S L 2 (Z p ) on P1 .

2) (flat case) A dynamical system P1 → P1 which is post-critically finite with
(orbifold) Euler characteristic zero.
3) (hyperbolic case) The action of a Hecke correspondence on a modular (or
Shimura) curve.

Here by saying that σ “comes from” a group action on X (respectively from

an endomorphism of X ) we mean that (“up to some specific finite subschemes”
for which we refer to [16]) X̃ is the disjoint union of the graphs of finitely
many automorphisms generating the action (respectively X̃ is the graph of
the endomorphism). Also δ-birational equivalence means isomorphism (com-
patible with the actions of δ) between the p-adic completions of the rings of
homogeneous fractions of degree zero with denominators not divisible by p.
The proofs behind the spherical case involve direct computations. The proofs
behind the flat case use the arithmetic Manin map, i.e. the space X2 (X ) in
Theorem 6.2, plus an induction in which canonical prolongations of vector
fields play a crucial role. The proofs behind the hyperbolic case of Theorem 6.6
are based on a theory of δ-modular forms which we quickly survey next.

6.2.4 δ-modular forms

Cf. [15, 2, 16, 26, 17, 25]. Let X 1 (N ) be the complete modular curve over R of
level N > 4 and let L 1 (N ) → X 1 (N ) be the line bundle with the property that
the sections of its various powers are the classical modular forms on 1 (N ) of
various weights. Let X be X 1 (N ) minus the cusps and the supersingular locus
166 Alexandru Buium

(zero locus of the Eisenstein form E p−1 of weight p − 1). Let L → X be

the restriction of the above line bundle and let V be L with the zero section
removed. So L 2 K . The elements of M n = On (V ) are called δ-modular
∞
functions. Set M = ∪M n . The elements of On (X ) are called δ-modular
forms of weight 0. For w ∈ W , w = 0, the elements of H 0 (X, L w ) ⊂ M ∞
are called δ-modular forms of weight w. We let σ = (σ1 , σ2 ) : X̃ → X × X
be the union of all the (prime to p) Hecke correspondences. Any δ-modular
function f ∈ M n has a “δ-Fourier expansion” in R((q))[q  , . . . , q (n) ]ˆ; setting
q  = q  = . . . = 0 in this δ-Fourier expansion one gets a series in R((q))ˆ
called the Fourier expansion of f . Finally let a4 and a6 be variables; then
consideration of the elliptic curve y 2 = x 3 + a4 x + a6 yields an R-algebra map
R[a4 , a6 , −1 ] → M 0
where  = (a4 , a6 ) is the discriminant polynomial. By universality we have
induced homomorphisms
(n) (n)
R[a4 , a6 , . . . , a4 , a6 , −1 ]ˆ → M n
that are compatible with δ.

Example 6.7. [17, 19] For f = an q n a classical newform over Z of weight
2, we get a δ-modular form of weight 0 and order 2
J (+)
2 ψ
5a
f - : J 2 (X ) ⊂ J 2 (X 1 (N )) → J 2 (E f ) → G
where + : X 1 (N ) → E f is the Eichler–Shimura map to the corresponding
elliptic curve E f (assumed for simplicity to be non-CM), and ψ the “unique”
δ-character of order 2. Then the δ-Fourier expansion of f - is congruent
mod p to
an n   q    q p
mp2
q − ap am q mp + a m q ;
n qp qp
(n, p)=1

hence the Fourier expansion of f - is congruent mod p to an n
(n, p)=1 n q .

Example 6.8. [15] There exists a unique δ-modular form f 1 ∈ H 0 (X ,

L −φ−1 ) with δ-Fourier expansion
 
p n−1 q  n q
(−1)n−1 = + p (. . .) .
n qp qp
n≥1

Hence the Fourier expansion of f 1 is 0. By the way the above δ-Fourier expan-
sion has the same shape as a certain function playing a role in “explicit” local
 Differential calculus with integers 167

class field theory. Here is the rough idea for the construction of f 1 . One con-

siders the universal elliptic curve E = Ui → Spec M ∞ , one considers
!i → J 1 (Ui ) of the projection, one considers the differences
sections si : U
!i ∩ U
si − s j : U !j → N 1 !a ,
G
! O) = H 1 (E, O).
and one considers the cohomology class [si − s j ] ∈ H 1 ( E,
Then f 1 ∈ M ∞ is defined as the cup product of this class with the canonical
generator of the 1-forms; in fact f 1 is the image of some element
f 1 (a4 , a6 , a4 , a6 ) ∈ R[a4 , a6 , a4 , a6 , −1 ]ˆ.
By the way it is a result of Hurlburt [45] that
2a a  − 3a a 
p p
f 1
(a4 , a6 , a4 , a6 ) ≡ E p−1 4 6 p 6 4 + f 0 (a4 , a6 ) mod p

where f 0 ∈ R[a4 , a6 , −1 ] and we recall that E p−1 is the Eisenstein form of
weight p − 1. (The polynomial f 0 is related to the Kronecker modular polyno-
mial mod p 2 .) This plus the δ-Fourier expansion of f 1 should be viewed as an
arithmetic analogue of the fact (due to Ramanujan) that
2a4 da6 − 3a6 da4 dq
→
 q
via the map between the Kähler differentials induced by the Fourier expansion
map C[a4 , a6 , −1 ] → C((q)).

Example 6.9. [2] There exists unique δ-modular forms f ∂ ∈ H 0 (X, L φ−1 )
and f ∂ ∈ H 0 (X, L 1−φ ) with δ-Fourier expansions 1 (and hence Fourier expan-
sions 1). They satisfy f ∂ · f ∂ = 1. The form f ∂ is constructed by applying the
“top part” of the canonical prolongation of the “Serre operator” to the form f 1 .
More precisely f ∂ is a constant times the image of
∂ ∂
(72φ(a6 )  − 16φ(a4 )2  − p · φ(P))( f 1 (a4 , a6 , a4 , a6 ))
∂a4 ∂a6

where P ∈ R[a4 , a6 , −1 , E −1

p−1 ]ˆ is the Ramanujan form.

Theorem 6.10. [2, 15, 16] The ring Rδ (X, K −1 )σ is “δ-generated” by f 1

and f ∂ .

Note that f 1 and f ∂ do not actually belong to Rδ (X, K −1 ) so the above

theorem needs some further explanation which we skip here; essentially, what
happens is that f 1 and f ∂ belong to a ring slightly bigger than Rδ (X, K −1 )
and they “δ-generate” that ring. We also note the following structure theorem
168 Alexandru Buium

for the kernel and image of the δ-Fourier expansion map, in which the forms
f 1 and f ∂ play a key role:

Theorem 6.11. [25]

1) The kernel of the Fourier expansion map M ∞ → R((q))ˆ is the p-adic
closure of the smallest δ-stable ideal containing f 1 and f ∂ − 1.
2) The p-adic closure of the image of the Fourier expansion map M ∞ →
R((q))ˆ equals Katz’ ring W of generalized p-adic modular forms.

The proof of the Theorem above is rather indirect and heavily Galois-
theoretic. Statement 2) in Theorem 6.11 says that all Katz’ divided congru-
ences between classical modular forms can be obtained by taking combinations
of “higher p-derivatives” of classical modular forms. Statement 1) above is a
lift to characteristic zero of the Serre and Swinnerton–Dyer theorem about
the kernel of the classical Fourier expansion map for classical modular forms
mod p. Theorem 6.10 can also be viewed as a lift to characteristic zero of
results of Ihara [46] about the Hasse invariant in characteristic p. These mod p
results do not lift to characteristic zero in usual algebraic geometry but do lift,
as we see, to characteristic zero in δ-geometry.
We mention the following remarkable infinitesimal symmetry property;
recall the classical Serre operator ∂ : O(V ) → O(V ). Also consider the Euler
derivation operator D : O(V ) → O(V ) given by multiplication by the degree
on each graded component of O(V ). Finally let P be the Ramanujan form (in
the degree 2 component of O(V )) and let θ : O(V ) → O(V ) be the derivation
θ = ∂ + PD.

Theorem 6.12. [16] The operator θ is an infinitesimal symmetry of the

R-module generated by f 1 , f ∂ , and f ∂ in M 1 = O1 (V ). Also θ is a vari-
ational infinitesimal symmetry of the R-module generated by f ∂ and f ∂ in
M 1 = O1 (V ).

Here is a calculation of differentials of the forms f 1 , f ∂ , f ∂ .

Theorem 6.13. [16] Let ω, α be the basis of O(V )/R dual to θ, D. Then
d( f ∂ ) = f ∂ · (φ ∗ α − α)
d( f ∂ ) = − f ∂ · (φ ∗ α − α)
d( f 1 ) = − f 1 · (φ ∗ α + α) − f ∂ · ω + f ∂ · p −1 φ ∗ ω.
In particular, for the eigenvalue μ = 1 we have
. .
d( f ∂ ) d( f ∂ )
{ ∂ }dp = { }dp = 0.
f f∂
 Differential calculus with integers 169

By the way, the forms f - and f 1 introduced above can be used to

prove some interesting purely diophantine results. For instance we have the
following:

Theorem 6.14. [19] Assume that + : X = X 1 (N ) → A is a modular

parametrization of an elliptic curve. Let p be a sufficiently large “good” prime
and let Q ∈ X (R) be an ordinary point. Let S be the set of all rational primes
that are inert in the imaginary quadratic field attached to Q. Let C be the
S-isogeny class of Q in X (R) (consisting of points corresponding to isogenies
of degrees only divisible by primes in S). Then there exists a constant c such
that for any subgroup ≤ A(R) with r := rank( ) < ∞ the set +(C) ∩ is
finite of cardinality at most cpr .

Other results of the same type (e.g. involving Heegner points) were proved
in [19]. In particular an analogue of the above Theorem is true with C replaced
by the locus C L of canonical lifts. To have a rough idea about the arguments
involved assume we want to prove that +(C L) ∩ is finite (and to bound the
cardinality of this set) in case is the torsion group of A(R). One considers
the order 2 δ-modular form f - : X 1 (N )(R) → R and one constructs, using
f 1 , a δ-function of order 1, f 0 : X (R) → R, on an open set X ⊂ X 1 (N )
which vanishes exactly on C L ∩ X (R). Then any point P in the intersection
X (R) ∩ +(C L) ∩ satisfies the system of “differential equations of order ≤ 2
in 1 unknown”

f - (P) = 0
f 0 (P) = 0

One can show that this system is “sufficiently non-degenerate” to allow the
elimination of the “derivatives” of the unknown; one is left with a differential
equation f 0 (P) = 0 “of order 0” which has, then, only finitely many solutions
(by Krasner’s theorem). By that theorem one can also bound the number of
solutions.
We end the discussion here by noting that the main players in the theory
above enjoy a certain remarkable property which we call δ-overconvergence.
Morally this is a an overconvergence property (in the classical sense of Dwork,
Monsky, Washnitzer) “in the direction of the variables x  , x  , . . . , x (n) ” (but
not necessarily in the direction of x). We prove:

Theorem 6.15. [27] The δ-functions f - , f 1 , f ∂ are δ-overconvergent.

170 Alexandru Buium

6.2.5 δ-Hecke operators

Next we discuss the Hecke action on δ-modular forms. For (n, p) =
1 the Hecke operators Tm (n) act naturally on δ-series (i.e. series in
R((q))[q  , . . . , q (r) ]ˆ) by the usual formula that inserts roots of unity of order
prime to p which are all in R. However no naive definition of Tm ( p) seems
to work. Instead we consider the situation mod p and make the following
definition. Let x be a p-tuple x1 , . . . , x p of indeterminates and let s be the
p-tuple s1 , . . . , s p of fundamental symmetric polynomials in x. An element
f ∈ k[[q]][q  , . . . , q (r ) ] is called δ- p-symmetric mod p if the sum
(r)
f (x1 , . . . , x1 ) + . . . + f (x p , . . . , x (r )  (r)
p ) ∈ k[[x]][x , . . . , x ]

is the image of an element

(r)
f ( p) = f ( p) (s1 , . . . , s p , . . . , s1 , . . . , s (r)  (r)
p ) ∈ k[[s]][s , . . . , s ].

For f that is δ- p-symmetric mod p define

“ pTm ( p) f = f ( p) (0, . . . , 0, q, . . . , 0, . . . , 0, q (r) )

+ p m f (q p , . . . , δ r (q p )) ∈ k[[q]][q  , . . . , q (r ) ].

Eigenvectors of “ pTm ( p) will be automatically understood to be

δ- p-symmetric. Also let us say that a series in k[[q]][q  , . . . , q (r) ] is primi-
tive if the series in k[[q]] obtained by setting q  = . . . = q (r) = 0 is killed by
the classical U -operator. Then one can give a complete description (in terms of
classical Hecke eigenforms mod p) of δ-“eigenseries” mod p of order 1 which
are δ-Fourier expansions of δ-modular forms of arbitrary order and weight:

Theorem 6.16. [26] There is a 1 − 1 correspondence between:

1) Series in k[[q]] which are eigenvectors of all Tm+2 (n), Tm+2 ( p), (n, p) =
1, and which are Fourier expansions of classical modular forms over k of
weight ≡ m + 2 mod p − 1.
2) Primitive series in k[[q]][q  ] which are eigenvectors of all nTm (n),
“ pTm ( p)”, (n, p) = 1, and which are δ-Fourier expansions of δ-modular
forms of some order ≥ 0 with weight w, deg(w) = m.

Note that the δ-Fourier expansion of the form f - discussed in Example 6.7 is
an example of series in 2) of Theorem 6.16. (Note that f - has order 2 although
its δ-Fourier expansion reduced mod p has order 1!) More generally the series
in 1) and 2) of Theorem 6.16 are related in an explicit way, similar to the way
f and f - of Example 6.7 are related. The proof of Theorem 6.16 involves a
 Differential calculus with integers 171

careful study of the action of δ-Hecke operators on δ-series plus the use of the
canonical prolongations of the Serre operator acting on δ-modular forms.

6.2.6 δ-functions on finite flat schemes

The p-jet spaces of finite flat schemes over R seem to play a key role in many
aspects of the theory. These p-jet spaces are neither finite nor flat in general
and overall they seem quite pathological. There are two remarkable classes of
examples, however, where some order seems to be restored in the limit; these
classes are finite flat p-group schemes that fit into p-divisible groups and finite
length p-typical Witt rings. Recall that for any ring A we write A = A/ p A.
Then for connected p-divisible groups we have:

Theorem 6.17. [28] Let F be a formal group law over R in one variable x,
assume F has finite height, and let F[ p ν ] be the kernel of the multiplication by
p ν viewed as a finite flat group scheme over R. Then
k[x, x  , x  , . . .]
lim On (F[ p ν ]) ν ν ν
→
n (x p , (x  ) p , (x  ) p , . . .)

sending x, δx, δ 2 x, . . . into the classes of x, x  , x  , . . . .

A similar result is obtained in [28] for the divisible group E[ p ν ] of an ordi-

nary elliptic curve; some of the components of J n (E[ p ν ]) will be empty and
exactly which ones are so is dictated by the valuation of q − 1 where q is the
Serre–Tate parameter. The components that are non-empty (in particular the
identity component) behave in the same way as the formal groups examined in
Theorem 6.17 above.
In the same spirit one can compute p-jet spaces of Witt rings. Let us consider
the ring Wm (R) of p-typical Witt vectors of length m+1, m ≥ 1, and denote by
'm = Spec Wm (R) its spectrum. Set vi = (0, . . . , 0, 1, 0, . . . , 0) ∈ Wm (R),
(1 preceded by i zeroes, i = 1, . . . , m), set π = 1 − δv1 ∈ O1 ('m ), and let
m = {1, . . . , m}. The following is a description of the identity component of
the limit of p-jet spaces mod p:

Theorem 6.18. [29] For n ≥ 2 the image of π p in On ('m ) is idempotent and

we have an isomorphism
k[xi(r) ; i ∈ m ; r ≥ 0]
lim On ('m )π (r)
→
n (xi x j , (xi ) p ; i, j ∈ m , r ≥ 1)

sending each δr vi into the class of the variable xi(r) .

172 Alexandru Buium

A similar description is obtained in [29] for the p-jet maps induced by the
Witt comonad maps. We recall that the data consisting of On ('m ) and the maps
induced by the comonad maps should be viewed as an arithmetic analogue of
the Lie groupoid of the line.

6.2.7 δ-Galois groups of δ-linear equations

Recall that for any solution u ∈ G L n (R) of a δ-linear equation

δu = α · u ( p)

(where α ∈ gl n (R)) and for any δ-subring O ⊂ R we defined the δ-Galois

group G u/O ⊂ G L n (O). We want to explain a result proved in [34]. Con-
sider the maximal torus T ⊂ G L n (R) of diagonal matrices, the Weyl group
W ⊂ G L n (R) of permutation matrices, the normalizer N = W T = T W of T
in G L n (R), and the subgroup N δ of N consisting of all elements of N whose
entries are in the monoid of constants R δ . We also use below the notation K a
for the algebraic closure of the fraction field K of R; the Zariski closed sets Z
of G L n (K a ) are then viewed as varieties over K a . The next result illustrates
some “generic” features of our δ-Galois groups; assertion 1) of the next the-
orem shows that the δ-Galois group is generically “not too large”. Assertions
2) and 3) show that the δ-Galois group are generically “as large as possible”.
As we shall see presently, the meaning of the word generic is different in each
of the 3 situations: in situation 1) generic means outside a Zariski closed set;
in situation 2) generic means outside a thin set (in the sense of diophantine
geometry); in situation 3) generic means outside a set of the first category (in
the sense of Baire category).

Theorem 6.19. 1) There exists a Zariski closed subset  ⊂ G L n (K a ) not

containing 1 such that for any u ∈ G L n (R)\ the following holds. Let
α = δu · (u ( p) )−1 and let O be a δ-subring of R containing α. Then G u/O
contains a normal subgroup of finite index which is diagonalizable over K a .
2 2
2) Let O = Z( p) . There exists a thin set  ⊂ Qn such that for any α ∈ Zn \
there exists a solution u of the equation δu = αu ( p) with the property that
G u/O is a finite group containing the Weyl group W .
3) There exists a subset  of the first category in the metric space

X = {u ∈ G L n (R); u ≡ 1 mod p}

such that for any u ∈ X \ the following holds. Let α = δu · (u ( p) )−1 . Then
there exists a δ-subring O of R containing R δ such that α ∈ gl n (O) and
such that G u/O = N δ .
 Differential calculus with integers 173

The groups W and N δ should be morally viewed as “incarnations” of the

groups “G L n (F1 )” and “G L n (Fa1 )” where “F1 ” and “Fa1 ” are the “field with
element” and “its algebraic closure” respectively [6]. This suggests that the
δ-Galois theory we are proposing here should be viewed as a Galois theory
over “F1 ”.

6.2.8 δ-flows for the classical groups

The main results in [34] concern the existence of certain δ-flows on G Ln
that are adapted, in a certain precise sense, to the various classical subgroups
G L n , S L n , S On , Spn .
Let G = G L n and let H be a smooth closed subgroup scheme of G. We
say that a δ-flow φG is left (respectively right) compatible with H if H is
φG -horizontal and the left (respectively right) diagram below is commutative:
!× G
H ! → !
G !× H
G ! → !
G
φ H × φG ↓ ↓ φG φG × φ H ↓ ↓ φG
!× G
H ! → !
G !× H
G ! → !
G

where φ H : H !→ H ! is induced by φG and the horizontal maps are given by

multiplication.
Next recall that by an involution on G we understand an automorphism τ :
G → G over R, x → x τ , of order 1 or 2. For each such τ one may consider the
subgroup G + = {a ∈ G; a τ = a}. The identity component (G + )◦ is referred
to as the subgroup defined by τ . Also set H : G → G, H(x) = x −τ x and
B : G × G → G, B(x, y) = x −τ y.
! A lift of Frobenius φG on G
Let us fix now a lift of Frobenius φG,0 on G. ! is
said to be H-horizontal (respectively B-symmetric) with respect to φG,0 if the
left (respectively right) diagram below is commutative:
φG φG,0 ×φG
! −→
G !
G G! −→ !×G
G !
H↓ ↓H φG × φG,0 ↓ ↓B
φ B
! −→
G
G,0
!
G !×G
G ! −→ !
G

If this is the case and φG,0 (1) = 1 then the group S := (G + )◦ defined by τ is
φG -horizontal; in particular there is an induced lift of Frobenius φ S on !
S. Also
note that if we set φG,0 (x) = x ( p) , viewing H as a matrix H(x) with entries
in R[x, det(x)−1 ]ˆ, we have that δG H = 0, which can be interpreted as saying
that H is a prime integral for our δ-flow φG .
The basic split classical groups G L n , S On , Spn are defined by involutions
on G = G L n as follows. We start with G L n itself which is defined by x τ = x.
174 Alexandru Buium

We call τ the canonical involution defining G L n . We also recall that if T ⊂ G

is the maximal torus of diagonal matrices, N is its normalizer in G, and W ⊂ G
is the group of permutation matrices then N = T W = W T and W is isomor-
phic to the Weyl group N /T . Throughout our discussion we let φG,0 (x) be
the lift of Frobenius on G L n defined by φG,0 (x) := x ( p) ; one can prove that
this φG,0 (x) is the unique lift of Frobenius on G ! that is left and right com-
patible with N and extends to a lift of Frobenius on gl5n (where we view G Ln
5
as an open set of gln ). On the other hand the groups Sp2r , S O2r , S O2r +1 are
defined by the involutions on G = G L n given by x τ = q −1 (x t )−1 q where q is
equal to
⎛ ⎞
    1 0 0
0 1r 0 1r
, , ⎝ 0 0 1r ⎠ ,
−1r 0 1r 0
0 1r 0

n = 2r, 2r, 2r + 1 respectively, x t is the transpose, and 1r is the r × r identity

matrix. We call this τ the canonical involution defining Sp2r , S O2r , S O2r+1 ,
respectively. All these groups are smooth over R. If S is any of these groups
and we set TS = T ∩ S and N S = N ∩ S then these groups are smooth, TS is
a maximal torus in S and N S is the normalizer of TS in S. Call a root of one of
these groups abnormal if it is a shortest root of a group S On with n odd.

Theorem 6.20. Let S be any of the groups G L n , S On , Spn and let τ be the
canonical involution on G = G L n defining S. Then the following hold.

1) (Compatibility with involutions) There exists a unique lift of Frobenius

φG on G ! that is H-horizontal and B-symmetric with respect to φG,0 . In
particular if lδ, lδ0 : G L n → gln are the arithmetic logarithmic deriva-
tive associated to φG , φG,0 then for all a ∈ S we have that the Cartan
decomposition
lδ0 (a) = (lδ0 (a))+ +δ (lδ0 (a))−

of lδ0 (a) satisfies (lδ0 (a))+ = lδ(a).

2) (Compatibility with the normalizer of maximal torus.) φG is right compat-
ible with N and left and right compatible with N S . In particular for all
a ∈ N S and b ∈ G L n (alternatively for all a ∈ G L n and b ∈ N ) we have

lδ(ab) = a *δ lδ(b) +δ lδ(a).

3) (Compatibility with root groups.) If χ is a root of S which is not abnormal

then the corresponding root group Uχ Ga is φG L n -horizontal.
 Differential calculus with integers 175

Note that a similar result can be proved for S L n ; the involution τ lives, in this
case, on a cover of G L n rather than on G L n itself. Note also that the exception
in assertion 3 of the Theorem (occurring in case χ is a shortest root of S On
with n odd) is a curious phenomenon which deserves further understanding.

6.2.9 Arithmetic Painlevé VI

The classical Painlevé VI equation has an elegant geometric description due to
Manin [56]. In [31] an arithmetic analogue of this equation has been introduced
and studied. We present below the main result in [31]. Recall our framework
in 6.1.2.10 above.

Theorem 6.21. [31] Let E be an elliptic curve over R which does not have a
lift of Frobenius and let ψ ∈ O(J 2 (E)) be δ-character of order 2 attached to
∗
an invertible 1-form ω on E. Consider the symplectic form η = ω ∧ φ pω on
J 1 (E). Let Y = E\E[2] and let r ∈ O(Y ). Then the following hold:

1) f = ψ − φ(r ) defines a generalized canonical δ-flow on J 1 (Y ) which is

Hamiltonian with respect to η.
2) There exists a canonical 1-form ν on X such that dν = η; in particular η
is exact and if  is the associated Euler–Lagrange form then p is closed.

Note that f in the Theorem does not define a δ-flow on J 1 (E) (equivalently,
J 1 (E) has no lift of Frobenius, equivalently the projection J 1 (J 1 (E)) →
J 1 (E) does not have a section):

Theorem 6.22. [32]. Let E be an elliptic curve over R which does not have a
lift of Frobenius. Then J 1 (E) does not have a lift of Frobenius.

6.3 Problems
Problem 6.1. Study the arithmetic jet spaces J n (X ) of curves X (and more
general varieties) with bad reduction.

This could be applied, in particular, to tackle Mazur’s question [59] about

bounding the torsion points on curves unifromly in terms of the genus; in other
words replacing C(g, p) by C(g) in Theorem 6.3. Our proof in [12] is based on
the study of the arithmetic jet space of J 1 (X ) at a prime p of good reduction.
A study of the arithmetic jet space of curves at primes of bad reduction might
lead to dropping the dependence of C(g, p) on p. Evidence that arithmetic jet
176 Alexandru Buium

spaces can be handled in the case of bad reduction comes in particular from
the recent paper [27].

Problem 6.2. Study the δ-modular forms that vanish on arithmetically inter-
esting Zariski dense subsets of Shimura varieties (such as CM loci or indi-
vidual non-CM isogeny classes). Compute δ-invariants of higher dimensional
correspondences.

This could be applied to extend results in [19], e.g. Theorem 6.14 above.
A deeper study of differential modular forms may allow one, for instance, to
replace S-isogeny class with the full isogeny class. The arguments might then
be extended to higher dimensional contexts and to the global field rather than
the local field situation. That such a deeper study is possible is shown by papers
like [25], for instance. For the higher dimensional case the theory in [3] might
have to be further developed to match the one dimensional theory in [15, 16].
In a related direction one might attempt to use the methods in [19] to tackle
Pink’s conjectures in [63]. In [19] it was shown that behind finiteness theo-
rems in diophantine geometry one can have reciprocity maps that are somehow
inherited from δ-geometry (and that provide effective bounds); a similar pic-
ture might hold for (cases of) Pink’s conjecture. The first case to look at for
such reciprocity maps would be in the case of the intersection between a mul-
tisection X of an abelian (or semiabelian) scheme G → S over a curve S
with the set of torsion points lying in special (CM or otherwise) fibers; more
general situations, in which torsion points are replaced by division points of
a group generated by finitely many sections, can be considered. Results of
André, Ribet, and Bertrand are pertinent to this question.

Problem 6.3. Compare the δ-geometric approach to quotient spaces with the
approach via non-commutative geometry.

The quotients X/σ for the correspondences appearing in Theorem 6.6 do

not exist, of course, in usual algebraic geometry. As Theorem 6.6 shows
these quotients exist, however, and are interesting in δ-geometry. Remarkably
such quotients also exist and are interesting in non-commutative geometry
[60]. More precisely the 3 cases (spherical, flat, hyperbolic) of Theorem
6.6 are closely related to the following 3 classes of examples studied in
non-commutative geometry:
P1 (R)
1) (spherical) S L 2 (Z) , non-commutative boundary of the classical modular
curve;
 Differential calculus with integers 177

1
τ  (θ ∈ R\Q): non-commutative elliptic curves;
S
2) (flat) e2πi
3) (hyperbolic) Non-commutative space Sh nc containing the classical
Shimura variety Sh (2-dimensional analogue of Bost–Connes systems).

It would be interesting to understand why these 3 classes appear in both con-

texts (δ-geometry and non-commutative geometry); also one would like to see
whether there is a connection, in the case of these 3 classes, between the 2
contexts.
Note that non-commutative geometry can also tackle the dynamics of
rational functions that are not necessarily post-critically finite of Euler char-
acteristic zero. It is conceivable that some post-critically finite polynomials of
non-zero Euler characteristic possess δ-invariants for some particular primes
(with respect to the anticanonical bundle or other bundles). A good start would
be to investigate the δ-invariants of σ (x) = x 2 − 1. Another good start would
be to investigate δ-invariants of post-critically finite polynomials with Euler
characteristic zero that are congruent modulo special primes to post-critically
finite polynomials with Euler characteristic zero.

Problem 6.4. Study the de Rham cohomology of arithmetic jet spaces. Find
arithmetic analogues of Kähler differentials  and D-modules. Find an object
that is to D what O1 is to Sym().

The study of de Rham cohomology of arithmetic jet spaces was started in

[7] where it is shown that the de Rham cohomology of J n (X ) carries informa-
tion about the arithmetic of X . The de Rham computations in [7] are probably
shadows of more general phenomena which deserve being understood. Also
the de Rham setting could be replaced by an overconvergent one; overconver-
gence is known to give an improved picture of the de Rham story and, on the
other hand, as already mentioned, it was proved in [27] that most of the remark-
able δ-functions occurring in the theory possess a remarkable overconvergence
property in the “δ-variables” called δ-overconvergence. Finally one is tempted
to try to relate the de Rham cohomology of arithmetic jet spaces J n (X ) to
the de Rham–Witt complex of X in characteristic p and in mixed character-
istic. Note further that since the arithmetic jet space J 1 (X ) is an analogue of
the (physical) tangent bundle T (X ) of X it follows that the sheaf O1 is an
arithmetic analogue of the sheaf Sym( X/R ), symmetric algebra on the Käh-
ler differentials. But there is no obvious arithmetic analogue of the sheaf  X/R
itself. Also there is no obvious arithmetic analogue of the sheaf D X of differen-
tial operators on X and of D X -modules. The absence of immediate analogues
of  and D is of course related to the intrinsic non-linearity of p-derivations. It
178 Alexandru Buium

would be interesting to search for such analogues. It is on the other hand con-
ceivable that there is a sheaf in the arithmetic theory that is to D what O1 is to
Sym(). Recall that the associated graded algebra of D is canonically isomor-
phic to the algebra of functions on the (physical) cotangent bundle O(T ∗ (X ))
(and not on the tangent bundle); this looks like a discrepancy but actually the
arithmetic jet space J 1 (X ) has a sort of intrinsic self-duality (cf. [16]) that
is missing in the classical algebro-geometric case where the tangent bundle
T (X ) and the cotangent bundle T ∗ (X ) and not naturally dual (unless, say, a
symplectic structure is given).
The δ-overconvergence property mentioned in Problem 6.4 may hold the
key to:

Problem 6.5. Define and study the/a maximal space of δ-modular forms on
which Atkin’s U operator can be defined.

Indeed the Hecke operators T (n) with p |n are defined on δ-modular forms
and have a rich theory in this context [15]. In contrast to this T ( p) and hence U
are still mysterious in the theory of δ-modular forms. A step in understanding
U was taken in [26] where the theory mod p for series of order 1 was given
a rather definitive treatment. However the theory in characteristic zero seems
elusive at this point. There are two paths towards such a theory so far. One path
is via δ-symmetry [24, 26]; this is a characteristic 0 analogue of the concept of
δ- p-symmetry mod p discussed above. Another path is via δ-overconvergence
[27]. The two paths seem to lead into different directions and this discrepancy
needs to be better understood. Assuming that a good theory of U is achieved,
this might lead to a Hida-like theory of families of differential modular forms,
including Galois representations attached to such forms. It is conceivable that
families in this context are not power series but Witt vectors. Part of the quest
for a U theory of δ-modular forms is to seek a δ-analogue of Eisentein series.
It is conceivable that the rings On (X 1 (N )) contain functions that do not vanish
at the cusps and are eigenvectors of the Hecke operators; such functions could
be viewed as “δ-Eisenstein” forms of weight zero.

Problem 6.6. Interpret information contained in the arithmetic jet spaces

J n (X ) as an arithmetic Kodaira-Spencer “class” of X .

Indeed some of these arithmetic Kodaira–Spencer classes (e.g in the case of

elliptic curves or, more generally, abelian schemes) were studied in [15, 16]
and lead to interesting δ-modular forms. For general schemes (e.g for curves
of higher genus) these classes were explored in Dupuy’s thesis [40]. They are
 Differential calculus with integers 179

non-abelian cohomology classes with values in the sheaf of automorphisms

of p-adic affine spaces ! Ad (in the case of curves d = 1). These classes arise
from comparing the local trivializations of arithmetic jet spaces. In this more
general case these classes may hold the key to a “deformation theory over the
field with one element”. On the other hand Dupuy proved in [40] that if X is a
smooth projective curve of genus ≥ 2 over R then J 1 (X ) is a torsor for some
line bundle over X ; this is rather surprising in view of the high non-linearity of
the theory. One should say that the line bundle in question is still mysterious
and deserves further investigation.

Problem 6.7. Further develop the partial differential theory in [20, 21, 22].

Indeed in spite of the extensive work done in [20, 21, 22] the arithmetic
partial differential theory is still in its infancy. The elliptic case of that theory
[20] (which, we recall, involves operators δ p1 and δ p2 corresponding to two
primes p1 and p2 ) is directly related to the study of the de Rham cohomology
of arithmetic jet spaces [7]; indeed one of the main results in [7] shows that
the arithmetic Laplacians in [20] are formal primitives (both p1 -adically and
p2 -adically) of global 1-forms on the arithmetic jet spaces (these forms being
not exact, although formally exact, and hence closed). By the way analogues
of these results in [7] probably exist in the case of modular curves; in the one
prime case a beginning of such a study was undertaken in [16], where some of
the main δ-modular forms of the theory were shown to satisfy some remark-
able systems of Pfaff equations. The hyperbolic/parabolic case of the theory
[21, 22] (which, we recall, involves a p-derivation δ p with respect to a prime p
and a usual derivation operator δq ) could be further developed as follows. One
could start by “specializing” the variable q in δq to elements π in arbitrarily
ramified extensions of Z p . This might push the theory in the “arbitrarily ram-
ified direction” which would be extremely desirable for arithmetic–geometric
applications. Indeed our ordinary arithmetic differential theory is, at present,
a non-ramified (or at most “boundedly ramified”) theory. A further idea along
these lines would be to use the solutions of the arithmetic partial differential
equations in [21, 22] to let points “flow” on varieties defined over number
fields. Some of the solutions in [21, 22] have interesting arithmetic features
(some look like hybrids between quantum exponentials and Artin–Hasse expo-
nentials, for instance) so the “flows” defined by them might have arithmetic
consequences. The challenge is to find (if at all possible) “special values”
of these solutions that are algebraic. One should also mention that the arith-
metic hyperbolic and parabolic equations in [21, 22] have, in special cases,
well defined “indices” that seem to carry arithmetic information; the challenge
180 Alexandru Buium

would be to make the index machinery work in general situations and to study
the variation of indices in families.

Problem 6.8. Find an arithmetic analogue of Sato hyperfunction solutions of

both “ordinary” and “partial” arithmetic differential equations.

Indeed Sato’s hyperfunctions, in their simplest incarnation, are pairs

( f (x), g(x)) of functions on the unit disk (corresponding to the distribution
f (x) − g(x −1 )) modulo (c, c), c a constant. The derivative of a pair is then
( dd xf (x), −x 2 ddgx (x)).) One could then try to consider, in the arithmetic case,
pairs (P, Q) of points of algebraic groups with values in δ-rings modulo
an appropriate equivalence relation and with an appropriate analogue of dif-
ferentiation with respect to p; this framework could be the correct one for
“non-analytic” solutions of the equations in [20, 21, 22].

Problem 6.9. Construct p-adic measures from δ-modular forms.

Indeed one of the main ideas in Katz’s approach to p-adic interpolation [50]
was to lift some of the remarkable Z p -valued ( p-adic) measures of the theory
to W-valued measures where W is the ring of (generalized) p-adic modular
forms. One can hope that some of these W-valued measures of Katz can be
further lifted to measures with values in the p-adic completion of the ring
of δ-modular functions M ∞ . Indeed recall from Theorem 6.11 that there is a
canonical homomorphism M ∞ → W whose image is p-adically dense, hence
the “lifting” problem makes sense. These lifted M  ∞ -valued measures could

then be evaluated at various elliptic curves defined over δ-rings to obtain new
Z p -valued measures (and hence new p-adic interpolation results) in the same
way in which Katz evaluated his measures at special elliptic curves. Another
related idea would be to interpret the solutions in Z p [[q]] of the arithmetic
partial differential equations in Problem 6.7 above as measures (via Iwasawa’s
representation of measures as power series). There is a discrepancy in this
approach in that the derivation of interest in Iwasawa’s theory is (1 + q) dq d
d
whereas the derivation of interest in Problem 6.7 is q dq ; nevertheless one
should pursue this idea and understand the discrepancy.

Problem 6.10. Find arithmetic analogues of classical theorems in the theory

of differential algebraic groups and further develop the δ-Galois theory in [35].
Also further develop the arithmetic analogue of the theory of symmetric spaces
in [34], especially on the Riemannian side.
 Differential calculus with integers 181

Indeed the theory of groups defined by (usual) differential equations (“dif-

ferential algebraic groups”) is by now a classical subject: it goes back to
Lie and Cartan and underwent a new development, from a rather new angle,
through the work of Cassidy and Kolchin [36, 52]. It is tempting to seek an
arithmetic analogue of this theory: one would like to understand, for instance,
the structure of all subgroups of G L n (R) that are defined by arithmetic dif-
ferential equations. The paper [14] proves an arithmetic analogue of Cassidy’s
theorem about Zariski dense subgroups of simple algebraic groups over dif-
ferential fields; in [14] the case of Zariski dense mod p groups is considered.
But Zariski dense groups such as G L n (Z p ) are definitely extremely interesting
(and lead to interesting Galois theoretic results such as in [16], Chapter 5). So
a generalization of [14] to the case of Zariski dense (rather than Zariski dense
mod p) groups, together with a generalization of the Galois theoretic results in
[16], would be very desirable. For instance it would be interesting to classify
all δ-subgroups of the multiplicative group Gm (R) = R × (or more generally
of G L 2 (R)) and find the invariants of such groups acting on R{x}  ( p) (where
 
R{x} = R[x, x , x , . . .]). Also remark that, as in the case of usual derivations,
there are interesting (“unexpected”) homomorphisms in δ-geometry between
G L 1 = Gm and G L 2 , for instance
 
a aψ∗ (a)
Gm (R) → G L 2 (R), a→
0 a

where ψ∗ is a δ-character, i.e. ψ ∈ Xn (Gm ). Cf. [43] for interesting devel-

opments into this subject. The main open problem in the δ-Galois theory of
G L n [35] seems at this point to decide if the δ-Galois groups always con-
tain a subgroup of the diagonal matrices as a subgroup of finite index. Other
problems are: to establish a Galois correspondence; to understand the relation
(already hinted at in [35]) between δ-Galois groups and Galois problems aris-
ing from the dynamics on Pn ; to generalize the theory by replacing G L n with
an arbitrary reductive group. With regards to the analogue in [34] of the the-
ory of symmetric spaces the main open problems are to find analogues of the
Riemannian metrics and curvature. An analogue of Lie brackets can, by the
way, be developed; it would be interesting to use it to develop an “arithmetic
Riemannian theory”.

Problem 6.11. Compose some of the basic δ-functions of the theory in [16]
(e.g. the δ-modular forms on Shimura curves) with p-adic uniformization maps
(e.g. with Drinfeld’s uniformization map of Shimura curves).
182 Alexandru Buium

Indeed this might shed a new light (coming from the analytic world) on
δ-geometry. These composed maps would belong to a “δ-rigid geometry”
which has yet to be developed in case these maps are interesting enough to
require it. By the way it is not at all clear that the functor that attaches to a
formal p-adic scheme its arithmetic jet space can be prolonged to a functor
in the rigid category. The problem (which seems to boil down to some quite
non-trivial combinatorially flavored calculation) is to show that the arithmetic
jet space functor sends blow-ups with centers in the closed fiber into (some
version of) blow-ups.
In the spirit of the last comments on Problem 6.11 one can ask:

Problem 6.12. Study the morphisms J n (X ) → J n (Y ) induced by non-étale

finite flat covers of smooth schemes X → Y .

Indeed note that if X → Y is finite and étale then it is well known that
J n (X ) → J n (Y ) are finite and étale; indeed J n (X ) J n (Y ) ×Y X . But note
that if X → Y is only finite and flat then J n (X ) → J n (Y ) is neither finite
nor flat in general. One of the simplest examples which need to be investigated
(some partial results are available [28], cf. Theorem 6.17) is that of the covers
[ p ν ] : G → G of smooth group schemes (or formal groups) given by multipli-
cation by pν . The geometry of the induced endomorphisms of the arithmetic jet
spaces is highly complex and mysterious. Understanding it might be, in partic-
ular, another path towards introducing/understanding the Atkin operator U on
δ-modular forms referred to in Problem 6.5. An obviously closely related prob-
lem is to understand the p-jet spaces of p-divisible groups; cf. Theorem 6.17.
Yet another example of interest is the study of J n (X ) → J n (X/ ) for a
finite group acting on a smooth X ; even the case X = Y n with Y a curve
and ' = Sn the symmetric group acting naturally is still completely myste-
rious. This latter case is related to the concept of δ-symmetry mentioned in
Problem 6.5 and appeared in an essential manner in the paper [24]: the failure
of invariants to commute with formation of jet spaces (in the ramified case) is
directly responsible for the existence (and indeed abundance) of δ-functions on
smooth projective curves which do not arise from δ-characters of the Jacobian.

Problem 6.13. Understand which analytic functions X (Z p ) → Z p for smooth

schemes X/Z p are induced by δ-functions.

Indeed it was proved in [23] that a function f : Z p → Z p is analytic

if, and only if, there exists m such that f can be represented as f (x) =
F(x, δx, . . . , δ m x), where F ∈ Z p [x0 , x1 , . . . , xm ]ˆ is a restricted power series
 Differential calculus with integers 183

with Z p -coefficients in m + 1 variables. This can be viewed as a “differential

interpolation result”: indeed f (x) is given by a finite family of (unrelated)
power series Fi (x) convergent on disjoint balls Bi that cover Z p and the result
says that one can find a single power series F(x, δx, . . . , δ m x) that equals
Fi (x) on Bi for each i. One can ask for a generalization of this by asking
which analytic functions f : X (Z p ) → Z p defined on the Z p -points of a
smooth scheme X/Z p come from a δ-function f˜ ∈ On (X ) (i.e. f = f˜∗ ); of
course such a f˜ cannot be, in general, unique. If X = A1 is the affine line then
the result in [23] says that any f comes from some f˜. This is probably still the
case if X is any affine smooth scheme. On the other hand this fails if X = P1
is the projective line simply because there are no non-constant δ-functions in
On (P1 ) [12] but, of course, there are plenty of non-constant analytic functions
P1 (Z p ) → Z p . There should be a collection of cohomological obstructions to
lifting f to some f˜ that should reflect the global geometry of X . This seems to
us a rather fundamental question in understanding the relation between p-adic
analytic geometry and δ-geometry.
In the light of Theorem 6.17 one can ask:

Problem 6.14. Let α0 , . . . , αν be elements in the algebraic closure of the frac-

tion field of R which are integral over R. Let X i = Spec R[αi ] be viewed as
a closed subscheme of the line A1 over R. Compute/understand the arithmetic
jet spaces J n (∪νi=0 X i ).

This is a problem “about the interaction” of algebraic numbers. Indeed the

case ν = 0 is clear; for instance if R[α] is totally ramified over R (and = R)
then the arithmetic jet spaces are empty: J n (Spec R[α]) = ∅ for n ≥ 1.
However, for ν ≥ 1, an interesting new phenomenon occurs. Indeed if αi = ζ pi
(primitive pi -th root of unity) then although J n (Spec R[ζ pi ]) = ∅ for n ≥ 1
and i ≥ 1 we have that J n (∪i=0ν Spec R[ζ ]) = J n (μ ν ) is non-empty and
pi p
indeed extremely interesting; cf. Theorem 6.17.

Problem 6.15. Find arithmetic analogues of Hamiltonian systems and of

algebraically completely integrable systems. Find arithmetic analogues of the
formal pseudo-differential calculus.

This problem is motivated by the link (due to Fuchs and Manin [56])
between the Painlevé VI equation (which has a Hamiltonian structure) and the
Manin map of an elliptic fibration. Painlevé VI possesses an arithmetic ana-
logue whose study was begun in [31]; this study is in its infancy and deserves
further attention. More generally there is an intriguing possibility that other
184 Alexandru Buium

physically relevant differential equations (especially arising in Hamiltonian

contexts, especially in the completely integrable situation, both finite and infi-
nite dimensional) have arithmetic analogues carrying arithmetic significance.
Finally, it is conceivable that a meaningful arithmetic analogue of the formal
pseudo-differential calculus in one variable [62]. p. 318 can be developed; in
other words one should be able to bring into the picture negative powers of
the p-derivation δ : On (X ) → On+1 (X ) in the same way in which formal
pseudo-differential calculus brings into the picture the negative powers of the
total derivative operator acting on functions on jet spaces.

Problem 6.16. Find an arithmetic analogue of the differential groupoids/Lie

pseudogroups in the work of Lie, Cartan, Malgrange [39].

Indeed recall that one can view the p-jet spaces J n ('m ) of 'm =
Spec Wm (R) and the jet maps induced by the comonad maps as an arith-
metic analogue of the Lie groupoid J n (R × R/R)∗ of the line; cf. [29] and
Theorem 6.18 above for results on this. It would be interesting to investigate
subobjects of the system J n ('m ) that play the role of analogues of differential
sub-groupoids and to find arithmetic analogues of the differential invariants
of diffeomorphism groups acting on natural bundles arising from frame bun-
dles. A theory of J n ('m ), suitably extended to other rings and to arithmetic
analogues of poly-vector fields, could also be interpreted as an arithmetic ana-
logue of the theory of the deRham–Witt complex [44] because it would replace
the usual Kähler differentials  by constructions involving the operators δ p .

Problem 6.17. Compute On (T(N, κ)) where T(N , κ) is the Hecke Z-algebra
attached to cusp forms on 1 (N ) of level κ.

This could lead to a way of associating differential modular forms f - to

classical newforms f of weight = 2. (Cf. Theorem 6.16.) This might also lead
to a link between δ-geometry and Galois representations. The problem may be
related to the study of double coset sets of G L 2 (or more generally G L n ) and
a link of this to the study in [33, 34, 35] is plausible.
We end by stating two of the most puzzling concrete open problems of the
theory.

Problem 6.18. Compute On (X ) for an elliptic curve X .

Note that O1 (X ) and Xn (X ) have been computed (Theorem 6.2) and one
expects Xn (X ) to “generate” On (X ). In particular assume X is not a canonical
 Differential calculus with integers 185

lift and ψ is a basis of X2 (X ); is it true that On (X ) = R[ψ, δψ, . . . , δ n−1 ψ]ˆ?

The analogue of this in differential algebra is true; cf. [9].

Problem 6.19. Compute Xn (G) for G an extension of an elliptic curve by Gm .

One expects that this module depends in an arithmetically interesting way

on the class of the extension. The problem is directly related to that of
understanding the cohomology of arithmetic jet spaces.

References
[1] D. V. Alexeevski, V. V. Lychagin, A. M. Vinogradov, Basic ideas and concepts
of differential geometry, in: Geometry I, R. V. Gamkrelidze, Ed., Encyclopedia
of Mathematical Sciences, Volume 28, Springer 1991.
[2] M.Barcau, Isogeny covariant differential modular forms and the space of elliptic
curves up to isogeny, Compositio Math., 137 (2003), 237-273.
[3] M. Barcau, A. Buium, Siegel differential modular forms, International Math.
Res. Notices, 28 (2002), 1459-1503.
[4] J. Borger, The basic geometry of Witt vectors, I: the affine case, Algebra and
Number Theory, 5 (2011), No. 2, 231-285.
[5] Borger, J., The basic geometry of Witt vectors, II: Spaces, Mathematische
Annalen, 351 (2011), No. 4, 877-933.
[6] J. Borger, -rings and the field with one element, arXiv:0906.3146v1
[7] J. Borger, A. Buium, Differential forms on arithmetic jet spaces, Selecta Math.,
17 (2011), no. 2, 301-335.
[8] A. Buium, Intersections in jet spaces and a conjecture of S.Lang, Annals of
Math. 136 (1992), 557-567.
[9] A. Buium, Differential algebra and diophantine geometry, Hermann, 1994.
[10] A. Buium, On a question of B.Mazur, Duke Math. J., 75 (1994), no. 3, 639-644.
[11] A. Buium, Differential characters of Abelian varieties over p-adic fields, Invent.
Math., 122 (1995), 309-340.
[12] A. Buium, Geometry of p-jets, Duke J. Math., 82, (1996), no. 2, 349-367.
[13] A. Buium, Differential characters and characteristic polynomial of Frobenius,
Crelle J. 485 (1997), 209-219.
[14] A. Buium, Differential subgroups of simple algebraic groups over p-adic fields,
Amer. J. Math. 120 (1998), 1277-1287.
[15] A. Buium, Differential modular forms, J. reine angew. Math., 520 (2000),
95-167.
[16] A. Buium, Arithmetic differential equations, Math. Surveys and Monographs,
118, American Mathematical Society, Providence, RI, 2005. xxxii+310 pp.
[17] A. Buium, Differential eigenforms, J. Number Theory, 128 (2008), 979-1010.
[18] A. Buium, B. Poonen, Independence of points on elliptic curves arising from
special points on modular and Shimura curves, I: global results, Duke Math. J.,
147 (2009), No. 1, 181-191.
186 Alexandru Buium

[19] A. Buium, B. Poonen, Independence of points on elliptic curves arising from

special points on modular and Shimura curves, II: local results, Compositio
Math., 145 (2009), 566-602.
[20] A. Buium, S.R. Simanca, Arithmetic Laplacians, Adv. Math., 220 (2009),
246-277.
[21] A. Buium, S.R. Simanca, Arithmetic partial differential equations, I, Advances
in Math. 225 (2010), 689-793.
[22] A. Buium, S.R. Simanca, Arithmetic partial differential equations II, Advances
in Math., 225 (2010), 1308-1340.
[23] A. Buium, C. Ralph, S.R. Simanca, Arithmetic differential operators on Z p , J.
Number Theory, 131 (2011), 96-105.
[24] A. Buium, Differential characters on curves, in: Number Theory, Analysis and
Geometry: In Memory of Serge Lang, D. Goldfeld et al. Editors, Springer, 2011,
pp. 111-123.
[25] A. Buium, A. Saha, The ring of differential Fourier expansions, J. Number
Theory, 132 (2012), 896-937.
[26] A. Buium, A. Saha, Hecke operators on differential modular forms mod p, J.
Number Theory, 132 (2012), 966-997
[27] A. Buium, A. Saha, Differential overconvergence, in: Algebraic methods in
dynamical systems; volume dedicated to Michael Singer’s 60th birthday, Banach
Center Publications, Vol 94 (2011), 99-129.
[28] A. Buium, p-jets of finite algebras, I: p-divisible groups, Documenta Math.,
18 (2013), 943-969.
[29] A. Buium, p-jets of finite algebras, II: p-typical Witt rings, Documenta Math.,
18 (2013), 971-996.
[30] A. Buium, Galois groups arising from arithmetic differential equations, to
appear in: Proceedings of a Conference in Luminy, Seminaires et Congrès, Soc.
Math. France.
[31] A. Buium, Yu. I. Manin, Arithmetic differential equations of Painlevé VI type,
arXiv:1307.3841
[32] A. Buium, A. Saha, The first p-jet space of an elliptic curve: global functions
and lifts of Frobenius, arXiv:1308.0578, to appear in Math. Res. Letters.
[33] A. Buium, T. Dupuy, Arithmetic differential equations on G L n , I: differential
cocycles, arXiv:1308.0748v1.
[34] A. Buium, T. Dupuy, Arithmetic differential equations on G L n , II: arithmetic
Lie theory, arXiv:1308.0744.
[35] A. Buium, T. Dupuy, Arithmetic differential equations on G L n , III: Galois
groups, arXiv:1308.0747
[36] P. Cassidy, Differential algebraic groups, Amer. J. Math, 94 (1972), 891-954.
[37] Z. Chatzidakis, E. Hrushovski, Model theory of difference fields, Trans. AMS,
351 (1999), 2997-3071.
[38] A. Connes, Non-commutative Geometry, Academic Press, 1994.
[39] V. G. Drinfeld, Elliptic modules, Math. Sbornik 94 (1974), 594-627.
[40] T. Dupuy, Arithmetic deformation theory of algebraic curves, PhD Thesis
(UNM).
[41] B. Dwork, G. Gerotto, F. J. Sullivan, An introduction to G-functions. Annals
of Mathematics Studies, 133. Princeton University Press, Princeton, NJ, 1994.
xxii+323 pp.
 Differential calculus with integers 187

[42] M.Greenberg, Schemata over local rings, Annals of Math., 73 (1961), 624-648.
[43] A. Herras-Llanos, PhD thesis (UNM), in preparation.
[44] L. Hesselholt, The big deRham–Witt complex, preprint, 2011.
[45] C. Hurlburt, Isogeny covariant differential modular forms modulo p, Composi-
tio Math., 128, (2001), No. 1, 17-34.
[46] Y. Ihara, An invariant multiple differential attached to the field of elliptic
modular functions of characteristic p, Amer. J. Math., XCIII (1971), No. 1,
139-147.
[47] Y. Ihara, On Fermat quotient and differentiation of numbers, RIMS Kokyuroku
810 (1992), 324-341, (In Japanese). English translation by S. Hahn, Univ. of
Georgia preprint.
[48] T. A. Ivey, J. M. Landsberg, Cartan for beginners: differential geometry via
moving frames and exterior differential systems, GTM 61, AMS 2003.
[49] A. Joyal, δ−anneaux et vecteurs de Witt, C.R. Acad. Sci. Canada, VII, (1985),
No. 3, 177-182.
[50] N. Katz, p-adic interpolation of real Eisenstein series, Ann. of Math.
104 (1976), 459-571.
[51] E.R. Kolchin, Differential algebra and algebraic groups. Pure and Applied
Mathematics, Vol. 54. Academic Press, New York-London, 1973. xviii+446 pp.
[52] E.Kolchin, Differential Algebraic Groups, Academic Press, 1985.
[53] N. Kurokawa, H. Ochiai, M. Wakayama, Absolute derivations and zeta func-
tions, Documenta Math.,Extra Volume Kato (2003), 565-584.
[54] B. Malgrange, Le groupoïde de Galois d’un feuilletage, Monographie 28 Vol 2,
L’enseignement Mathématique (2001).
[55] Yu. I. Manin, Rational points on algebraic curves over function fields, Izv. Acad.
Nauk USSR, 27 (1963), 1395-1440.
[56] Yu. I. Manin, Sixth Painlevé equation, universal elliptic curve, and mirror of P2 ,
arXiv:alg-geom/9605010.
[57] Yu. I. Manin, Cyclotomy and analytic geometry over F1 , arXiv:0809.1564.
[58] Yu. I. Manin, Numbers as functions, p-Adic Numbers, Ultrametric Analysis, and
Applications, 5 (2013), No. 4, 313-325.
[59] B.Mazur, Arithmetic on curves, Bull. Amer. Math. Soc., 14, 2 (1986), 207-259.
[60] M.Marcolli, Lectures on Arithmetic Non-commutative Geometry, Univ. Lecture
Series 36, AMS, 2005.
[61] M. Morishita, Knots and primes, Springer, 2012.
[62] P. J. Olver, Appications of Lie Groups to Differential Equations, GTM 107,
Springer, 2000.
[63] R. Pink, A combination of the conjectures of Mordell-Lang and André-Oort, in:
Progress in Math 235, Birkhauser 2005, pp. 251-282.
[64] M.Raynaud, Courbes sur une variété abélienne et points de torsion, Invent.
Math. 71 (1983), 207-235.
[65] M. Singer, M. van der Put, Galois theory of difference equaltions, LNM,
Springer 1997.
[66] P. Vojta, Jets via Hasse-Schmidt derivations, in: Diophantine geometry,
pp. 335-361, CRM Series, 4, U. Zannier Ed., Sc. Norm., Pisa, 2007.
 7
Un calcul de groupe de Brauer et une
application arithmétique
Jean-Louis Colliot-Thélène
C.N.R.S., Université Paris Sud, Mathématiques, Bâtiment 425,
91405 Orsay Cedex, France
E-mail address : [email protected]

R ÉSUMÉ . Browning et Matthiesen ont récemment démontré le principe de

Hasse et l’approximation faible pour une certaine classe de variétés algé-
briques. Skorobogatov a utilisé ce résultat pour montrer que sur une deuxième
classe, plus large, de variétés, les points rationnels sont denses dans l’ensemble
de Brauer–Manin. On considère ici une troisième classe de variétés, inter-
médiaire entre la première et la seconde, pour laquelle on établit, de façon
purement algébrique, que le groupe de Brauer non ramifié est trivial. Le résul-
tat de Skorobogatov donne alors le principe de Hasse et l’approximation faible
pour ces variétés.

7.1 Un calcul algébrique

Soient k un corps, K = k une clôture séparable de k et G = Gal(K /k) le
groupe de Galois absolu. Pour toute k-variété X , on note X K = X ×k K .
On a la proposition bien connue (cf. [CTS, (1.5.0)]) :

Proposition 7.1 Pour toute k-variété X , on a une suite exacte naturelle

0 → H 1 (G, K [X ]× ) → Pic X → (Pic X K )G → H 2 (G, K [X ]× ) →

Ker[Br X → Br X K ] → H 1 (G, Pic X K )

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

188
 Un calcul de groupe de Brauer et une application arithmétique 189

C’est la suite exacte des termes de bas degré de la suite spectrale de

Hochschild-Serre pour le faisceau Gm pour la topologie étale, et le revêtement
XK → X.
Soient L i (x1 , . . . , xn ), i = 1, . . . , m des formes linéaires à coefficients
dans k, linéairement indépendantes deux à deux, et soient K i /k, i =
1, . . . , m, des k-algèbres finies étales.
Soit Y la k-variété affine définie par le système d’équations affines
L i (x1 , . . . , xn ) = Norm K i /k (1i ), i = 1, . . . , m.
Ici 1i varie dans la k-variété R K i /k A1 définie par la restriction à la Weil de K i
à k de la droite affine A1K i .
Soit p : Y → Ank la projection sur l’espace affine de dimension n définie par
les coordonnées (x1 , . . . , xn ). Soit F ⊂ An le complémentaire de la réunion
des fermés de codimension 2 définis par L i = L j = 0, i < j. Soit U ⊂ Ank
l’ouvert complémentaire de F et soit X = YU ⊂ Y l’ouvert image inverse de
U dans Y .

Proposition 7.2 (i) La k-variété X est lisse et géométriquement intègre.

(ii) Toute fonction inversible sur X est constante : k × = k[X ]× .
(iii) Le groupe de Picard Pic X est nul.
(iv) Pour toute compactification lisse Z de X , le groupe de Picard Pic Z K est
un G-réseau de permutation.
(v) Supposons k de caractéristique zéro. Pour toute k-variété Z projec-
tive, lisse, géométriquement intègre k-birationnelle à Y , l’appplication
naturelle de groupes de Brauer Br k → Br Z est un isomorphisme.

Démonstration Pour les fonctions inversibles, on a k[X ]× = (K [X ]× )G . Le

théorème 90 de Hilbert donne H 1 (G, K × ) = 0. En utilisant la proposition 7.1,
on voit que pour établir les énoncés (i), (ii) et (iii) on peut se limiter au cas
k = K séparablement clos, ce que nous supposons jusqu’à nouvel ordre.
Chaque norme Norm K i /k (1i ) est alors un produit de variables indépen-
dantes :

Norm K i /k (1i ) = 1i,r .
r∈Ii

Au-dessus de l’ouvert V ⊂ U ⊂ Ank défini par i L i = 0, la fibration
p : X V → V est clairement lisse, les fibres sont des tores. Plus précisément,
pour i = 1, . . . , m , on a

1i,1 = L i (x1 , . . . , xn ). 1−1
i,r
r∈Ii , r>1
190 Jean-Louis Colliot-Thélène

si bien que X V est le produit de V ⊂ Ank et d’un k-tore de groupe des caractères
ayant pour base les 1i,r pour r > 1. Il est alors clair que l’on a Pic X V = 0. Le
groupe des fonctions inversibles sur X V est engendré par les L i et les 1i,r , où,
pour i donné, r varie dans Ii . Il est donc engendré par les 1i,r . Ces éléments
sont indépendants multiplicativement.
Comme L 1 n’est pas identiquement nul, la k-variété définie par
L 1 (x1 , . . . , xn ) = Norm K 1 /k (11 )
est isomorphe à un espace affine, donc intègre et lisse sur k. La projection de
la variété définie par
0 = L i (x1 , . . . , xn ) = Norm K i /k (1i ), i = 2, . . . , m,
sur Ank est un morphisme lisse. La k-variété X 1 ⊂ X définie par les équations
L 1 (x1 , . . . , xn ) = Norm K 1 /k (11 )

0 = L i (x1 , . . . , xn ) = Norm K i /k (1i ), i = 2, . . . , m,

On voit aisément qu’elle est intègre. Définissant X i , pour chaque i = 2, . . . , m
de façon analogue à X 1 , on voit que la k-variété X , qui est la réunion des
ouverts X i , est une k-variété lisse (cf. [Sk1, Lemma 4.4.5]). Notons i,r le
diviseur de X défini par 1i,r = 0, qui est inclus dans L i = 0. Un élément
n
f ∈ k[X ]× produit d’éléments 1i,ri,r avec l’un des n i,r ∈ Z non nul n’est pas
inversible. Ceci établit k = k[X ]× .
×

Le complémentaire de X V dans X est la réunion des diviseurs i,r , et

chacun d’eux est principal sur X . Comme on a Pic X V = 0, ceci implique
Pic X = 0.
Sur tout corps k, désormais quelconque et de clôture séparable K , on a donc
établi les points (i) à (iii), et la suite exacte de la proposition 7.1 donne de plus
un isomorphisme :
Br k → Ker[Br X → Br X K ].
Pour Z comme en (iv), le groupe Div∞ Z K des diviseurs à support dans
le complémentaire de X K dans Z K est un G-réseau de permutation (pos-
sédant une base globalement respectée par l’action de G), et ce groupe est
G-isomorphe à Pic Z K , comme le montre la suite exacte
K [Z ]× → K [X ]× → Div∞ Z K → Pic Z K → Pic X K .
Ceci établit (iv).
On a vu plus haut que l’ouvert X V ,K de X K est le produit d’un ouvert de
AnK et d’un K -tore. La K -variété X K est donc K -birationnelle à un espace
projectif.
 Un calcul de groupe de Brauer et une application arithmétique 191

Supposons désormais k de caractéristique zéro. Soit Z une k-variété propre,

lisse, géométriquement intègre k-birationnelle à X . Il existe donc un ouvert
W ⊂ X contenant tous les points de codimension 1 de X et un k-morphisme
birationnel f : W → Z, d’où l’on déduit par les propriétés standards du
groupe de Brauer des injections Br Z → Br W (lissité des k-variétés), puis
Br X → Br W (pureté pour le groupe de Brauer en caractéristique zéro, [Gr,
§6]), donc Br Z → Br X . L’invariance birationnelle du groupe de Brauer
en caractéristique zéro [Gr, §7], sa nullité sur l’espace affine AnK en carac-
téristique zéro et la K -rationalité de la K -variété propre et lisse Z K donnent
Br Z K = 0. On a donc

Br Z = Ker[Br Z → Br Z K ] → Ker[Br X → Br X K ],

et comme la flèche Br k → Ker[Br X → Br X K ] est un isomorphisme, il en

résulte que la flèche Br k → Br Z est un isomorphisme, ce qui établit (v). 

Remarque 7.3 Le calcul ci-dessus est une généralisation du calcul fait au

Lemme 2.6.1 de [CTS].

7.2 Application arithmétique

Le calcul du paragraphe précédent a été motivé par le théorème 1.3 de l’ar-
ticle [BM] de T. Browning et L. Matthiesen. Pour toute Q-variété définie par
un système

0 = L i (x1 , . . . , xn ) = Norm K i /Q (1i ), i = 1, . . . , m

comme à la proposition 7.2, les algèbres K i étant de plus supposées être des
corps, les auteurs montrent que le principe de Hasse et l’approximation faible
valent pour les points rationnels. Il ne saurait donc y avoir d’obstruction de
Brauer–Manin associée au groupe de Brauer d’une compactification lisse.
On eût pu difficilement imaginer un tel théorème si le groupe de Brauer
d’une telle compactification n’avait pas été réduit au groupe de Brauer du corps
de base, ce qu’établit la proposition 7.2.
A. Skorobogatov [Sk2] a récemment combiné la théorie de la descente et
le théorème de Browning et Matthiesen. Il a ainsi établi que l’obstruction de
Brauer–Manin est la seule obstruction au principe de Hasse pour certaines
variétés plus générales que celles considérées par Browning et Matthiesen.
La conjonction de [Sk2, Prop. 3.5] et de la proposition 7.2 ci-dessus donne
l’énoncé suivant, qui étend directement celui de Browning et Matthiesen.
192 Jean-Louis Colliot-Thélène

Théorème 7.4 Soient L i (x1 , . . . , xn ), i = 1, . . . , m des formes linéaires

à coefficients dans Q, linéairement indépendantes deux à deux, et soient
K i /Q, i = 1, . . . , m, des Q-algèbres finies étales, c’est-à-dire des produits
d’extensions finies de Q.
Soit Y la Q-variété affine définie par le système d’équations affines
0 = L i (x1 , . . . , xn ) = Norm K i /k (1i ), i = 1, . . . , m.
Ici 1i varie dans la Q-variété R K i /Q A1 définie par la restriction à la Weil de
K i à Q de la droite affine A1K i .
Le principe de Hasse et l’approximation faible valent pour les points
rationnels de Y . 

Ce travail, stimulé par un séjour à l’Institut Hausdorff à Bonn en avril

2013, a bénéficié d’une aide de l’Agence Nationale de la Recherche portant
la référence ANR-12-BL01-0005.

Références
[BM] T. Browning et L. Matthiesen, Norm forms for arbitrary number fields as
products of linear polynomials, http://arxiv.org/abs/1307.7641
[CTS] J.-L. Colliot-Thélène et J.-J. Sansuc, La descente sur les variétés rationnelles, II,
Duke Math. J. 54 (1987) 375–492.
[Gr] A. Grothendieck, Le groupe de Brauer III, in Dix exposés sur la cohomologie
des schémas, Masson & North-Holland, 1968.
[Sk1] A. N. Skorobogatov, Torsors and rational points, Cambridge tracts in mathe-
matics 144, Cambridge University Press, 2001.
[Sk2] A. N. Skorobogatov, Descent on toric fibrations. This volume, pp. 422–435.
 8
Connectedness of Hecke algebras and the
Rayuela conjecture: a path to functoriality and
modularity
Luis Dieulefait1 and Ariel Pacetti2
1 Departament d’Àlgebra i Geometria, Facultat de Matemàtiques, Universitat de
Barcelona, Gran Via de les Corts Catalanes, 585. 08007 Barcelona
E-mail address: [email protected]
2 Departamento de Matemática, Facultad de Ciencias Exactas y Naturales,
Universidad de Buenos Aires and IMAS, CONICET, Argentina
E-mail address: [email protected]

A la memoria de Julio Cortazar, autor de “Rayuela”, al cumplirse hoy

cien años de su nacimiento.

A BSTRACT. Let ρ1 and ρ2 be a pair of residual, odd, absolutely irreducible

two-dimensional Galois representations of a totally real number field F. In
this article we propose a conjecture asserting existence of “safe” chains of
compatible systems of Galois representations linking ρ1 to ρ2 . Such conjecture
implies the generalized Serre’s conjecture and is equivalent to Serre’s conjec-
ture under a modular version of it. We prove a weak version of the modular
variant using the connectedness of certain Hecke algebras, and we comment
on possible applications of these results to establish some cases of Langlands
functoriality.

8.1 Introduction
Let F be a totally real number field. In [KK03] Khare and Kiming conjectured
that given ρ1 and ρ2 two odd, absolutely irreducible, two-dimensional residual
representations of the absolute Galois group of F with values on finite fields of
Date: August 26, 2014
2010 Mathematics Subject Classification. 11F33
Keywords and phrases. Base Change, Rayuela Conjecture
L.D. partially supported by MICINN grants MTM2012-33830 and by an ICREA Academia
Research Prize.
AP was partially supported by CONICET PIP 2010-2012 GI and FonCyT BID-PICT 2010-0681.

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

193
194 Luis Dieulefait and Ariel Pacetti

different prime characteristics  and  , satisfying certain local compatibilities

(for all primes q different from  and  , there exists a Weil-Deligne parameter
and a choice of integral model such that their reductions modulo  and  are
isomorphic to the restrictions of ρ1 and ρ2 to a decomposition group at q),
there exists a modular form which residually coincides with ρ1 at  and ρ2 at
 . This is a very strong statement and no evidence for its truth is known. In
particular it implies that given two Hilbert modular forms f and g over F, and
two prime numbers  and  , if we assume suitable local compatibilities, then
there exists a Hilbert modular form h which is congruent to f modulo  and to
g modulo  .
The purpose of this article is to state a weaker conjecture called The
Rayuela Conjecture involving “chains of congruent systems” connecting two
given Galois representations. The idea is not to link any given pair of residual
Galois representations via a single compatible system of Galois representations
but via a chain of such systems, meaning that each of them has to be congru-
ent to the next one in the chain, modulo a suitable prime, and it will also be
required that some reasonable properties are being satisfied at each of these
congruences (this will be done in Section 8.2). Such conjecture has important
consequences, like Serre’s conjectures for totally real number fields or base
change for totally real number fields (the second one follows from a weaker
version that we will explain later). The conjecture for the field of rational num-
bers follows from the results proved by Dieulefait in [Die12a] (the reader can
easily check that the method used to prove base change in loc. cit. implies, in
fact is based on, the fact that this conjecture is true over Q at least for modular
Galois representations, and this combined with the truth of Serre’s conjecture
over Q implies the conjecture).
Although we are not able to prove the conjecture itself, in Sections 8.4
and 8.5 we prove that some (weak) variant of it holds in the modular world.
This is a generalization of a Theorem of Mazur, a result that he proved for the
case of weight 2 modular forms of prime level N .
In this article, for general weights and levels, and for any totally real number
field, we extend Mazur’s result, and the control we get in the level and weights
of the modular forms involved might be used to prove the modular version of
the conjecture (although there are still some ingredients missing to get the full
statement).
Also we will show that our Conjecture is equivalent to the generalized
Serre’s conjecture proposed in [BDJ10], at least if some strengthening of the
available Modularity Lifting Theorems is assumed. What we show is how one
can manipulate a pair of modular Galois representations to end up in a con-
trolled situation where both representations have the same Hodge-Tate weights
 Connectedness of Hecke algebras and the Rayuela conjecture 195

and ramification in a small controlled set of primes (and then this is combined
with Mazur’s result).
Combining the ideas of the present article with some nowadays standard
tricks (like the use of Micro Good Dihedral primes), one can prove for some
small real quadratic fields a Base Change theorem, all that is left is a finite
computational check (to ensure that some Hecke Algebra of known level and
weight is connected in a good way). This is part of a work in progress of the
authors but more will be said at the end of the present article.
For most of the chains constructed in this paper, the construction carries
on for the case of abstract Galois representations (the reader should keep in
mind that the possibility of constructing congruences between abstract Galois
representations where some local information changes and the existence of
compatible systems containing most geometric p-adic Galois representations
are the two main technical ingredients in the proofs of Serre’s conjecture over
Q), and then again the use of Micro Good Dihedral primes combined with the
results in this paper is enough to reduce the proof of Serre’s conjecture over a
given small real quadratic field to some special cases of Galois representations
with small invariants (Serre’s weights and conductor). Thus it is perfectly con-
ceivable that one can give a complete proof of Serre’s conjecture over a small
real quadratic field F, by just completing this process with a few extra steps
designed to remove the Micro Good Dihedral prime from the level (eventually
relying on results of Skinner and Wiles [SW99] if the residually reducible case
has to be considered) and end up in some “base case for modularity” over F,
such as the modularity/non-existence
√ of semistable abelian varieties of conduc-
tor 3 over Q( 5) proved by Schoof ([Sch12]). We plan to check over which
real quadratic fields, such proof of Serre’s conjecture can be completed in a
future work.
Section 8.3 contains the Modularity Lifting Theorems (that will be denoted
MLT) used and needed in the present article. We have included a mixed case
that has only been proved in weight 2 situations, yet we assume its truth in
more general situations. We believe that such a result can be deduced from the
techniques in [BLGGT], but we haven’t formally checked this.

Conventions and notations: If F is a number field, we denote by O F its ring

of integers. By G F we denote the Galois group Gal( F̄/F). All Galois repre-
sentations (residual or -adic) are assumed to be continuous. Let p be a prime
ideal in O F . We denote by Frobp a Frobenius element over p.

To ease notation, instead of specifying a prime π in the field of coefficients

of a Galois representation and reducing mod π , sometimes we will simply say
196 Luis Dieulefait and Ariel Pacetti

that we reduce “mod p”, where p is the rational prime below π . We expect
that this will be no cause of confusion.
Concerning local types of strictly compatible systems of Galois rep-
resentations, we will use the words “Steinberg”, “principal series” and
“supercuspidal”, to denote the local representations they correspond to under
local Langlands.

Acknowledgment: we want to thank Fred Diamond for useful comments and

remarks and for the proof of Lemma 8.32.

8.2 The Rayuela Conjecture

There are nowadays many MLT that can be applied to propagate modular-
ity through a congruence between two -adic Galois representations. For us,
the main interest in congruences between Galois representations is in the case
where a MLT holds in both directions (i.e. modularity of either of the two
representations implies modularity of the other one), so we will call a congru-
ence an MLT congruence if the hypothesis of some MLT theorem are fulfilled
in both directions. Since MLT hypothesis are becoming less restrictive with
time, some of the proofs we give make use of tricks that are required to reduce
to situations in which some of the available MLT applies, but some of these
tricks are very likely to become obsolete in the near future (when new MLT
are proved). There are also steps in the chains that we are going to build in our
attempt to connect two given modular or abstract Galois representations that
involve congruences that are not known to be MLT. This is the unique reason
why we are not able to prove any strong result in this paper, such as relative
base change (see the discussion at the end of the article).
Let us now state our conjecture for abstract Galois representations.

Conjecture 8.1. The Rayuela Conjecture: Let F be a totally real number

field, 0 , ∞ prime numbers and

ρi : Gal( F̄/F) → GL2 (Fi ), i = 0, ∞,

two absolutely irreducible odd Galois representations. Then there exists a fam-
ily of odd, absolutely irreducible, two-dimensional strictly compatible systems
of Galois representations {ρi,λ }ni=1 such that:
● ρ0 ≡ ρ1,λ0 (mod λ0 ), with λ0 | 0 .
● ρ∞ ≡ ρn,λ∞ (mod λ∞ ), with λ∞ | ∞ .
 Connectedness of Hecke algebras and the Rayuela conjecture 197

● for i = 1, . . . , n − 1 there exist λi prime such that ρi,λi ≡ ρi+1,λi (mod λi ),

● all the congruences involved are MLT.

Remark 8.2. The primes involved always are taken to be in the field of coef-
ficient of the corresponding system of representations (for the first and second
condition) or in the compositum of the two relevant such fields (for the third
conditions). From now on, this remark applies to all congruence between
compatible systems appearing in this paper.
The last hypothesis of the conjecture implies that modularity of one of the
given representations can be propagated through the chain of congruences pro-
duced by the conjecture allowing to prove modularity of the other one. It is thus
clear that Conjecture 8.1 implies Serre’s generalized conjecture over F by just
taking ρ∞ to be an irreducible residual representation attached to any cuspidal
Hilbert modular form over F.
At this time we think Conjecture 8.1 is out of reach, but since it does
not involve directly modular forms, it might lead to a different attack to
Serre’s conjecture. Note that in the hypothesis of the Conjecture, one can put
ρ0 inside a strictly compatible system of Galois representations as done by
[Die04] for representations over Q and by [Sno09] for totally real fields (using
Theorem 7.6.1 and Corollary 1.1.2). The main difficulty for abstract represen-
tations is to connect them. During this work we will show how to manipulate
abstract representations such that if we start with any pair of them we can
connect both, through suitable chains of MLT congruences, to representations
having common values for their Serre’s level and weights, but we cannot go
any further so far.
For different purposes, such as applications to Langlands functoriality, it is
interesting to study the previous conjecture in the modular setting. In this case
we are able to prove part of it, namely, we are able to build the chain of con-
gruences but we can not ensure that congruences are MLT at all steps. Still,
we find this interesting because advances in MLT theorems may eventually
lead to a proof of this modular variant of the conjecture, and this will be evi-
dence for the truth of the Rayuela conjecture. In fact it will give a proof that
this conjecture is equivalent to the generalized Serre’s conjecture over F (the
other implication being trivial as already remarked). The modular variant is the
following:

Conjecture 8.3. Let F be a totally real number field, 0 , ∞ prime numbers

and
ρi : Gal( F̄/F) → GL2 (Fi ), i = 0, ∞,
198 Luis Dieulefait and Ariel Pacetti

two absolutely irreducible odd Galois representations attached to Hilbert mod-

ular newforms f 0 and f ∞ , respectively. Then there exists a family of odd,
absolutely irreducible, two-dimensional strictly compatible systems of Galois
representations {ρi,λ }ni=1 such that:
● ρ0 ≡ ρ1,λ0 (mod λ0 ), with λ0 | 0 .
● ρ∞ ≡ ρn,λ∞ (mod λ∞ ), with λ∞ | ∞ .
● for i = 1, . . . , n − 1 there exist λi prime such that ρi,λi ≡ ρi+1,λi (mod λi ),
● all the congruences involved are MLT.

The aim of this article is to prove part of Conjecture 8.3 and show some
implications of it related to modularity and functoriality. It is clear that all the
representations appearing in Conjecture 8.3 are modular. Clearly we have:

Theorem 8.4. Serre’s generalized conjecture + Conjecture 8.3 imply Conjec-

ture 8.1.

Proof. This is clear, since if ρ0 and ρ∞ are any two Galois representation and
if Serre’s conjecture holds, they are modular, and then they are connected in
the right way by Conjecture 8.3.

Conjecture 8.3 can be proved using standard arguments mainly due to

Mazur, if we remove the condition that the congruences are MLT.

Theorem (Mazur). Let F be a totally real number field and f 0 , f ∞ two

Hilbert modular forms, whose weights are congruent modulo 2. Then there
exists Hilbert modular forms {h i }ni=1 such that h 1 = f 0 , h n = f ∞ and for
i = 1, . . . , n − 1 there exist λi prime such that ρh i ,λi ≡ ρh i +1 ,λi (mod λi ).

Remark 8.5. Although connectedness of the Hecke algebra was proved by

Mazur for classical modular forms of weight 2 and prime level N using the
curve X 0 (N ), the proof we will give is strongly based in his argument.
Remark 8.6. If modularity would propagate via any congruence, the previous
Theorem would be equivalent to Conjecture 8.3. Unfortunately this is not clear
even for classical modular forms.
The main result of the present article is the following:

Theorem 8.7. Let F be a totally real number field,  f , g prime numbers and

ρ f : Gal( F̄/F) → GL2 (F f ), ρg : Gal( F̄/F) → GL2 (Fg ),

 Connectedness of Hecke algebras and the Rayuela conjecture 199

two absolutely irreducible odd Galois representations attached to Hilbert mod-

ular newforms f and g, respectively. Then there exists prime ideals q ⊂ O F
and p ∈ Q which splits completely in F, Hilbert modular forms f ∞ and g∞ of
parallel weight 2 and level 0 ( pq2 ) and two families of odd, absolutely irre-
ducible, 2-dimensional strictly compatible systems of Galois representations
f nf g ng
{ρi,λ }i=1 , {ρi,λ }i =1 , such that:
f
● ρ f ≡ ρ1,λ f (mod λ f ), with λ f |  f .
f
● ρ f∞ ,λ∞ ≡ ρn f ,λ∞ (mod λ∞ ), for some prime ideal λ∞ .
f f
● for i = 1, . . . , n − 1 there exist λi prime such that ρi,λi ≡ ρi+1,λi (mod λi ),
g
● ρg ≡ ρ1,λg (mod λg ), with λg | g .
g
● ρg∞ ,λ∞ ≡ ρn g ,λ∞ (mod λ∞ ), for some prime ideal λ∞ .
g g
● for i = 1, . . . , n − 1 there exist λi prime such that ρi,λi ≡ ρi+1,λi (mod λi ),
● all the congruences involved are MLT.

Remark 8.8. The importance of our main result is that we can translate Con-
jecture 8.3 to a situation where the level and the weight are known. Some of
the primes involved are not explicit, since they come from some application of
Tchebotarev density theorem. See Section 8.6.

8.3 MLT theorems used

In this section we just enumerate the MLT theorems that will be used during
this work. By E we denote a finite extension of Q .

Theorem 8.9 (MLT1). Let F be a totally real number field, and  ≥ 5 a prime
number which splits completely in F. Let ρ : G F → GL2 (E) be a continuous
irreducible representation such that:
● ρ ramifies only at finitely many primes,
● ρ is odd,
● ρ|G Fv is potentially semi-stable for any v |  with different Hodge-Tate
weights.
● The restriction ρ|G F(ξ ) is absolutely irreducible.
● ρ̄ ∼ ρ¯f for a Hilbert modular form f .

Then ρ is automorphic.

Proof. This is Theorem 6.4 of [HT13].

200 Luis Dieulefait and Ariel Pacetti

Theorem 8.10 (MLT2 - ordinary case). Let F be a totally real field,  and odd
prime and ρ : G F → GL2 (E) be a continuous irreducible representation such
that:
● ρ is unramified at all but finitely many primes.
● ρ is de Rham at all primes above .
● The reduction ρ̄ is irreducible and ρ̄(G F(ξ ) ) ⊂ GL2 (F ) is adequate.
● ρ is ordinary at all primes above .
● ρ̄ is ordinarily automorphic.

Then ρ is ordinarily automorphic. If ρ is also crystalline (resp. potentially

crystalline), then ρ is ordinarily automorphic of level prime to  (resp.
potentially level prime to ).

Proof. This is just Theorem 2.4.1 of [BLGGT].

Theorem 8.11 (MLT3 - pot. diagonalizable case). Let F be a totally real field,
 ≥ 7 be and odd prime and ρ : G F → GL2 (E) be a continuous irreducible
representation such that:
● ρ is unramified at all but finitely many primes.
● ρ is de Rham at all primes above , with different Hodge-Tate numbers.
● ρ|G Fλ is potentially diagonalizable for all λ | .
● The restriction ρ̄|G F(ξ ) is irreducible.
● ρ̄ is either ordinarily automorphic or potentially diagonalizable automor-
phic.

Then ρ is potentially diagonalizable automorphic (of level potentially prime

to ).

Proof. This is Theorem 4.2.1 of [BLGGT]. Note that although it is stated only
for CM fields, one can chose a suitable CM extension and get the same result
for totally real fields using solvable base change.

Remark 8.12. The hardest condition to check in the previous Theorem is that
of ρ being potentially diagonalizable, but this is satisfied if, in particular, 
is unramified in F, ρ is crystalline at all primes above  and the Hodge-Tate
weights are in the Fontaine-Laffaille interval.
Remark 8.13. The last result is stated for  ≥ 7, BUT if  is odd and ρ|G F(ξ )
is adequate, the same holds (see [DG12]).
We also need a mixed variant. Recall the following property.
 Connectedness of Hecke algebras and the Rayuela conjecture 201

Lemma 8.14. Let F be a totally real field, and {ρλ } be a strictly compatible
system of continuous, odd, irreducible, parallel weight 2 representations, then
ρλ is potentially Barsotti-Tate or ordinary at λ.

Proof. If the restriction to λ is not potentially crystalline, then it is potentially

semistable non-crystalline, in which case ordinariness is known. In fact, using
potential modularity and the semistability at λ, the system is known to cor-
respond to an abelian variety with potentially semistable reduction at λ, and
potentially (over a suitable extension) to a parallel weight 2 Hilbert modular
form which is Steinberg at λ.

The following variant of the current MLT is not yet known:

Assumption 8.15 (MLT4 - mixed case). Let F be a totally real field,  be and
odd prime and ρ : G F → GL2 (E) be a continuous irreducible representation
such that:

● ρ is unramified at all but finitely many primes.

● ρ is de Rham at all primes above , with different Hodge-Tate numbers.
● ρ|G Fλ is potentially diagonalizable for some λ |  and is ordinary at the
others.
● The restriction ρ̄|G F(ξ ) is irreducible and adequate.
● ρ̄ is either ordinarily automorphic or potentially diagonalizably automor-
phic at the same places as ρ̄.

Then ρ is potentially automorphic (of level potentially prime to ).

Remark 8.16. We believe that using the tools developed in [BLGGT] this
result is accessible. Not only we consider the above result accessible, but also
if we restrict to the case where both Galois representations involved are of
parallel weight 2 (thus, because of the previous Lemma, locally potentially
Barsotti-Tate or ordinary at all primes above p), it is a Theorem as proved in
[BD], Theorem 3.2.2.

Remark 8.17. We will need also need Assumption 8.15 to hold for p = 2 and
parallel weight 2.

Remark 8.18. To apply the previous MLTs we need ρ(G F(ξ ) ) to be irreducible
and to be adequate. When  ≥ 7, adequacy is equivalent to irreducibility,
so this imposes no extra hypothesis. Nevertheless, when  = 3 and  = 5
this is not the case. Theorem 1.5 of [Gur12] implies that if  ≥ 3 and
202 Luis Dieulefait and Ariel Pacetti

SL2 (F pr ) ⊂ Im(ρ) for r > 1 then the residual image of the representa-
tion (and the same for the image of its restriction to any abelian extension)
is adequate as well. In what follows, we will only apply any of the previous
MLTs in characteristics 3 or 5 AFTER having added a good dihedral prime
(which is added modulo a prime greater than 5) and from this it follows that
this result of Guralnick can be applied ensuring adequacy (see [Die12a] for
more details).

8.4 Abstract representations

As mentioned in the introduction, most of the level/weight manipulations that
we are going to perform, work not only for modular representation but for
abstract ones as well. In this section we will work in the greatest generality
possible, and in the next section we will restrict to modular representations
putting emphasis on the results that are nowadays only known for modular
representations. The results in this section are enough to prove Mazur connect-
edness Theorem. We begin by recalling the following well-known definition
(see [Ser98] for example). If λi is a prime in a number field K , we denote by
i the rational prime below λi and by L i the set of primes in K dividing i
(which clearly contains λi ).

Definition 8.19. A compatible system of Galois representations over F is a

family of continuous Galois representations

ρλ : Gal(F/F) → GL2 (K λ ),

where K is a finite extension of Q and λ runs through the prime ideals of O K ,

which satisfy:

(1) There exists a finite set of primes S (independent of λ) such that ρλ is

unramified outside S ∪ L  .
(2) For each pair of prime ideals (λ1 , λ2 ) in O K and for each prime ideal
p ∈ S ∪ L 1 ∪ L 2 , the characteristic polynomials Q p (x) of ρλi (Frobp ) lie
in K [x] and are equal.

The most important examples of such families are the ones arising from the
étale cohomology of a variety defined over F. In this case we also have some
control on the roots of the characteristic polynomials, and some control on the
λ-adic representation at primes in L. This motivate the following definition
(see [BLGGT], Section 5).
 Connectedness of Hecke algebras and the Rayuela conjecture 203

Definition 8.20. A rank 2 strictly compatible system of Galois representations

R of G F defined over K is a 5-tuple

R = (K , S, {Q p (x)}, {ρλ }, {Hτ }),

where

(1) K is a number field.

(2) S is a finite set of primes of F.
(3) for each prime p ∈ S, Q p (x) is a degree 2 polynomial in K [x].
(4) For each prime λ of K , the representation

ρλ : G F → GL2 (K λ ),

is a continuous semi-simple representation such that:

● If p ∈ S and p  , then ρ is unramified at p and ρ (Frob ) has
λ λ p
characteristic polynomial Q p (x).
● If p | , then ρ|
G Fp is de Rham and in the case p ∈
/ S, crystalline.
(5) for τ : F → K , Hτ contains 2 different integers such that for any
K → K λ over K , we have that HTτ (ρλ ) = Hτ .
(6) For each finite place p of F there exists a Weil-Deligne representation
WDp (R) of W Fp over K such that for each place λ of K not dividing the
residue characteristic of p and every K -linear embedding ι : K → K λ ,
the push forward ι WDp (R) WD(ρλ |G Fp ) K -ss .

Remark 8.21. If one starts with a two-dimensional continuous, odd, Galois

representations over a totally real number field F, under some minor hypoth-
esis (which are exactly the hypothesis for an MLT theorem to hold), one
can prove that such representations is potentially modular. In particular, this
implies that the representations is part of a strictly compatible system. Since
all the congruences we will work with are where an MLT theorem works
in both directions, without loss of generality, we will assume that all the
representations come in strictly compatible systems.

Definition 8.22. Let {ρλ } be a strictly compatible system of Galois representa-

tions. We say that the system is dihedral if the images are compatible dihedral
groups, i.e. if there exists a quadratic extension L/F (independent of λ) such
that ρλ is induced from a λ-adic character of L.

Lemma 8.23. The family {ρλ } is dihedral if and only if one representation is
dihedral.
204 Luis Dieulefait and Ariel Pacetti

Proof. It is clear that if the whole family is dihedral, in particular any of them
is dihedral. For the converse, let λ be a prime ideal of O F , and suppose that ρλ0
is dihedral. Then there exists a quadratic extension L/F and a λ-adic character
χλ0 : Gal(L/L) → K λ0 such that ρλ0 = IndG G L χλ0 . We know that χλ0 is
F

part of a strictly compatible system of one-dimensional representations {χλ },

so we are led to prove that ρλ IndG G L χλ for all primes λ ⊂ O F . This comes
F

from a straightforward computation, since the values of the traces (an even the
whole characteristic polynomial) of ρλ0 (Frobp ) are given in terms of the values
of χλ0 . For split primes in the extension L/F, the trace of ρλ0 (Frobp ) equals
−χλ0 (p1 ) − χλ0 (p2 ), where pO L = p1 p2 and for inert primes, the trace is zero.
In particular, the same happens to ρλ (Frobp ), for all λ so ρλ and IndGG L χλ have
F

the same trace and are thus isomorphic.

To apply most MLTs theorems we will need to have some control of the
image of our residual Galois representations. In particular we will need the
image of its restriction to a cyclotomic extension to be adequate. To avoid
checking this particular condition at each step of our chain of congruences, we
will move to families with “big image”, in the sense that for all but finitely
many primes p (and in most steps, for all primes of bounded size), the residual
representation has an image containing SL2 (F p ).

Proposition 8.24. Let ρλ be a strictly compatible system of odd dihedral rep-

resentations. Then there exists a strictly compatible system of odd non-dihedral
representations {ρ2,λ } and a prime p such that ρp ≡ ρ2,p (mod p) and MLT
holds in both directions.

Proof. It is well known that dihedral compatible families do not have Stein-
berg primes in the level, so we just need to add a Steinberg prime to our
representation by some raising the level argument. Let λ be a prime over a
prime p > 5 and such that:

● p splits completely in F.
● λ ∈ S, i.e. ρq is unramified at λ if q = λ.
● L and F(ξ p ) are disjoint, i.e. F(ξ p ) ∩ L = F.

We want to add a Steinberg prime q modulo λ, and by the previous choice, we

are in the hypothesis of Theorem 7.2.1 of [Sno09] which says that it is enough
to raise the level locally. The local problem is standard, and can be achieve
by Tchebotarev’s Theorem as follows: take q inert in the extension L/F, so
that aq = 0. Then the local raising the level condition becomes Nq ≡ −1
(mod p), so we chose any such prime and get a global representation with the
 Connectedness of Hecke algebras and the Rayuela conjecture 205

desired properties. The existence of a strictly compatible system attached to

such global representation follows from Taylor’s potential modularity result
(see [BLGGT]) plus the argument from [Die04], as generalized in [BLGGT].
That MLT holds in both directions comes from Theorem 8.9.

Proposition 8.25. Let {ρλ } be a strictly compatible system of odd, non-

dihedral representations. Then there exists an integer B, such that if N(λ) > B
then SL2 (F p ) ⊂ Im(ρ λ ).

Proof. According to Dickson’s classifications of subgroups of PGL2 (Fλ ),

when we consider the residual representations, they might:

(1) contain PSL2 (Fλ ),

(2) be reducible,
(3) be dihedral,
(4) be isomorphic to A4 , S4 or A5 .

We want to prove under the running hypothesis, there are only finitely many
primes where the first case does not hold. But the second case is exactly
Lemma 5.4 of [CG11], the third case Corollary 5.2 of [CG11] (here we use
the assumption that the system is not dihedral), and the last case is Lemma 5.3
of [CG11].

Remark 8.26. Another way to prove the last Proposition (following the classi-
cal approach of Ribet) it to first apply some potential automorphy result (like
in [BLGGT]) to deduce that the system is potentially modular. Then its restric-
tion is isomorphic to that of a Hilbert modular form. Furthermore, since our
abstract representations are not dihedral, the respective Hilbert modular form
has no CM. But for Hilbert modular forms, such result is proven in [Dim05]
Proposition 0.1.
By the above considerations, from now on we will only consider non-
dihedral families, so we will skip writing this hypothesis in the next results.

Proposition 8.27. Let {ρ1,λ } be a strictly compatible system of odd irreducible

Galois representations. Then there exists a compatible system {ρ2,λ } of parallel
weight 2 representations and a prime p such that ρ1, p ≡ ρ2, p (mod p) and
MLT holds in both directions.

Proof. Let p > 5 be a prime which splits completely in F, does not divide the
level of ρ, is larger than all weights of the system and such that the image of
ρ 1,p is large. Let det(ρ) = ψχ p , where χ p is the reduction of the cyclotomic
206 Luis Dieulefait and Ariel Pacetti

character, and let ψ be any lift of ψ. Then Theorem 7.6.1 of [Sno09] implies
that ρ admits a weight two lift to Q p which ramifies at the same primes as ρ
and ψ, with determinant ψχ p . MLT hold in both directions by the same proof
as the previous Proposition 8.24.

8.4.1 Adding a good dihedral prime

As already mentioned, while working with MLT one needs to ensure that
residual images are big enough to be in the hypothesis of such a theorem.
Sometimes, one needs the restriction of the residual representation to the cyclo-
tomic extension of p-th roots of unity to have adequate image, but for most
MLT requiring irreducibility of this restriction is enough (and for p > 5 it is
known that both properties are equivalent). A way to get this property guaran-
teed at most steps is by introducing to the level an extra prime which forces the
image modulo all primes up to a certain bound to be “non-exceptional”, i.e. it is
irreducible, and its projectivization is not dihedral, nor the exceptional groups
A4 , S4 , A5 . A way to get this, is by adding a “good dihedral prime” (with
respect to the given bound) as was introduced by Khare and Wintenberger in
their work on Serre’s conjecture.
The difference with the classical setting is that since we work with two
strictly compatible systems at the same time, we need to add the same good
dihedral prime to both of them.

Proposition 8.28. Let {ρ1,λ } and {ρ2,λ } be two strictly compatible systems
of continuous, odd, irreducible representations of parallel weight 2. Fix B a
positive integer, larger than 5 and than all primes in the conductors of both
systems. Let p ≡ 1 (mod 4) be a rational prime such that:

● p is bigger than B.
● p splits completely in the compositum of the coefficient fields of ρ1 and ρ2 .
● p is relatively prime to the conductors of both systems.
● Im(ρ i,p ) = GL2 (F p ), i = 1, 2, for some prime p over p.

Then there exists a prime q not dividing the conductor of the systems such
that:

● q ≡ −1 (mod p).
● q splits completely in the extension F  given by the compositum of all
quadratic extensions of F ramified only at primes above rational primes
 < B.
 Connectedness of Hecke algebras and the Rayuela conjecture 207

● q ≡ 1 (mod 8).
● There exists a prime ideal q in F over q such that the image of ρ 1,p (Frobq )
and ρ 2,p (Frobq ) both have eigenvalues 1 and −1.
With this choice of primes, there exists two strictly compatible systems of
continuous representations {$1,λ } and {$2,λ } of parallel weight 2 such that:
(1) ρ 1,p $1,p .
(2) ρ 2,p $2,p .
(3) {$i,λ }, for i = 1, 2, is locally good dihedral at q (w.r.t. the bound B).
(4) $i,p , is Barsotti-Tate at all primes dividing p and has the same type as ρi,p
locally at any prime other that q for i = 1, 2.
(5) The congruences are MLT.

Remark 8.29. We will not reproduce here the precise definition of good-
dihedral prime (it can be found in [KW09] and [Die12b]) or what it means for
a compatible system to be locally good dihedral at a prime w.r.t. certain bound
B. Let us just recall that a good-dihedral prime is a supercuspidal prime in a
compatible system such that the ramification at this prime forces the system to
have residual images containing SL2 (F p ) every time the residual characteristic
p is bounded by B (and a bit more for p = 2 or 3: also in these characteristics
the image is forced to be non-solvable).
Remark 8.30. We do not want to make precise what condition (4) of the last
statement means (the local type), since it will not be important for our pur-
poses, but what we prove is the following: for abstract representations, we will
use Theorem 7.2.1 of [Sno09], where a “type” is understood as a Weil type (no
information on the monodromy), while for modular representations (actually
residually modular ones), we will use Theorem 3.2.2 of [BD], which uses the
complete notion of type.
Proof. The existence of the primes p and q follows with almost the same
arguments as [Die12b] (Lemma 3.3). The only difference is that we need to
consider the compositum of Q(i ), F and the coefficient field of ρ1 and ρ2 .
Take p big enough (for the images to be large) and split in such extension.
Then q is chosen using Tchebotarev density Theorem, with the condition that
it hits complex conjugation in the same suitable field.
We are lead to prove the existence of a lift of ρ̄1,p with the desired prop-
erties. By Theorem 7.2.1 of [Sno09], or by Theorem 3.2.2 of [BD] (see the
last remark), we know that a global representation with the desired properties
exists if and only if locally the corresponding lifts do exist, so we only need to
show which are the local deformation conditions:
208 Luis Dieulefait and Ariel Pacetti

● At the primes l =
√
q, that of ρi |G l .
Q [ ] √
● $i | Dq = IndQqq (χ ), where Qq [ ] is the unique quadratic unramified
extension of Qq , and χ is a character with order p (this is called type C in
[Sno09]).

This proves the existence of $i,p . The congruence is MLT in both directions
because of Theorem 8.9 (in this case both forms are of parallel Hodge-Tate
weight 2 which is smaller than the prime p). Since we are in the hypothesis
of an MLT theorem, we can put such representation into a strictly compatible
system and get the result.

Remark 8.31. Although we stated the result for representations of paral-

lel weight 2, in general we can first move to parallel weight 2 (using
Proposition 8.27) and then add the good dihedral prime.

With this result, we can prove Mazur’s Theorem.

Theorem (Mazur). Let F be a totally real number field, f 0 and f ∞ two

Hilbert modular forms, whose weights are congruent modulo 2, then there
exists Hilbert modular forms {h i }ni=1 such that h 1 = f 0 , h n = f ∞ and such
that for i = 1, . . . , n − 1 there exist λi prime such that ρh i ,λi ≡ ρh i+1 ,λi
(mod λi ).

Proof. By the previous results, we can assume that both forms are of parallel
weight 2, and do not have complex multiplication (or we move to such a sit-
uation using Proposition 8.24 and Proposition 8.27), for some common level
1 (n) (of course they might not be new at the same level). The idea now is to
find both modular forms in the cohomology of a Shimura curve, and for this
purpose we need to add an auxiliary prime to the level where both forms are
not principal series (in case [F : Q] is odd). So what we do is to raise the level
of both forms at an auxiliary prime p as in Proposition 8.28 (we could also add
a Steinberg prime). Now we consider the Shimura curve X p (n) ramified at all
infinite places of F but one, and at the auxiliary prime p if needed (depending
whether [F : Q] is odd or even) and with level n.
We apply Mazur’s argument just as in [Maz77] (proposition 10.6, page 98)
with some minor adjustments. The main idea of his proof is the following: if
the Jacobian J is a product of two abelian varieties A × B, since J decom-
poses (up to isogeny) as a product of simple factors with multiplicity one,
there are no nontrivial homomorphisms from A to B̂ nor from B to Â. Then
the principal polarization of J induces principal polarizations in A and B, but a
 Connectedness of Hecke algebras and the Rayuela conjecture 209

Jacobian cannot decompose as a nontrivial direct product of principally polar-

ized abelian varieties (which follows from the irreducibility of its θ -divisor, see
[ACGH85]). From this it follows that Spec T is connected, where T denotes the
Hecke algebra acting on J .
In his original article Mazur was dealing with the curve X 0 (N ), with N
a prime number, so there are no old forms appearing in J0 (N ). To use the
same argument in our context, we have to deal with old forms as well, and
the problem is that the abelian varieties A f corresponding to old forms do
not appear with multiplicity one in the decomposition (up to isogeny) of the
Jacobian of a modular or a Shimura curve. But this is not a problem if we
observe that for what we want it is not necessary to prove the connectedness of
Spec T, it is enough to show that the anemic Hecke algebra T0 generated only
by the Hecke operators with index prime to the level is connected. Therefore,
what we need is to discard the cases where the Jacobian of X p (n) decomposes
as a product of abelian varieties A × B with every simple factor in A and every
simple factor in B being orthogonal (recall that now these simple factors need
not appear with multiplicity one). In such case, the same proof as in Mazur’s
article applies, and gives the connectedness we are looking for.

8.5 Killing the level

Now that we have a good dihedral prime, which controls the image of the
residual representations in the families of the two representations for small
primes, we want to connect them at a chosen level. From know on, we will
only consider modular representations, pointing out in each case the needed
result for abstract ones. To lower the level we will need the following result,
whose proof is due to Fred Diamond.

Lemma 8.32. Let ρ f : Gal(F/F) → GL2 (K p ) be a modular representation

of level dividing n pr (where p | p and p  n) which satisfies the following
hypothesis:

● The form f has a supercuspidal prime q  p.

● The residual image ρ f is absolutely irreducible and not bad dihedral.
● The determinant det(ρ f | Iv ) = χ pn for all v| p and some n independent of v.

Then ρ has a modular lift of level dividing n. Moreover we can assume the
weight of the form is parallel and the character has order prime to p.
210 Luis Dieulefait and Ariel Pacetti

Proof. We do not assume that p is unramified in F because the results we need

from [BDJ10] can be generalized to the ramified case in a straightforward way.
We will follow the notation of the aforementioned article.
By Proposition 2.10 and Corollary 2.11 of [BDJ10], ρ is modular of some
weight σ = ⊗v| p σv and central character ⊗v| p (Nkv /F p )n−1 . The results
of [Roz12] show that σ is a Jordan-Holder constituent of the reduction of
⊗τ Symmk−1 O2E for some sufficiently large k, where the tensor product is
over embeddings τ : F → E for a sufficiently large number field E, viewed
as contained both in C and in Q p . Another application of Proposition 2.10
of [BDJ10] shows that ρ is modular of parallel weight k and level prime to p.
Moreover the presence of a supercuspidal prime q allows us to use an indefinite
quaternion algebra of discriminant dividing q and hence assume the open com-
pact subgroup has level dividing n in the first application of Proposition 2.10
of [BDJ10]. Moreover ρ is not badly dihedral in the sense of Lemma 4.11 of
[BDJ10], so the arguments there ensure that we can assume the divisibility of n
by the level is preserved in the second application of Proposition 2.10 as well.
We can similarly ensure the conclusion on the central character.
For further details and more general results along these lines, see forthcom-
ing work of Diamond and Reduzzi.

8.5.1 Modifying the non-Steinberg primes

From now on we work under the assumption that MLFMT (Assertion 8.15) is
true. We call an abstract strictly compatible system of continuous, odd, irre-
ducible Galois representations modular if there exists a Hilbert modular form,
whose attached Galois representations matches the abstract one.

Theorem 8.33. Let {ρ1,λ } be a strictly compatible system of continuous, odd,

irreducible, Galois representations attached to a Hilbert newform f over a
totally real number field F of parallel weight 2, containing in its conductor a
locally good dihedral prime q (w.r.t. some sufficiently large bound B). Then:
● There exists a strictly compatible system of continuous, odd, irreducible,
parallel weight 2 representations {ρ2,λ }, which is semistable at all primes
except the same good dihedral prime q and such that the Steinberg ramified
primes are bounded in norm by B.
● There exists a chain of congruences of compatible systems linking {ρ1,λ } and
{ρ2,λ } such that all congruences involved occur in residual characteristics
bounded by B and are MLT.
In particular the system {ρ2,λ } is also modular.
 Connectedness of Hecke algebras and the Rayuela conjecture 211

Proof. Let λ be a prime which is supercuspidal or principal series. Let p be a

prime number dividing the order of the character corresponding to ramification
at λ. We consider two different cases: if p is relative prime to λ, we call it the
tamely ramified case, while the other case we call it the wildly ramified case.
The tamely ramified case. Let p be a prime ideal in K (the coefficient field)
dividing p, and consider the residual mod p representation. Then the p-part of
the ramification is lost, so we take a minimal lift with the same parallel weight
2 (it exists by Theorem 3.2.2 in [BD]). Observe that at this step we are not only
modifying the ramification type at λ, but also at any other prime supercuspi-
dal or principal series in the prime-to- p part of the conductor with ramification
given by a character of order divisible by p. Then by Lemma 8.14, we might be
in a mixed situation of potentially Barsotti-Tate and ordinary representations.
Using Theorem 3.2.2 of [BD] (recall that this is a special case of our Assump-
tion 8.15), we get rid of this p-part of the inertia with an MLT congruence.
Iterating this process, we end up killing all tamely ramified ramification given
by characters, i.e. the prime-to- p part of the ramification at primes dividing p
is killed. So we are reduced to the case where all primes in the conductor are
either Steinberg or with ramification given by a prime order character whose
order is divisible by the ramified prime.
The wildly ramified case. In this case, we will move the wildly ramified
primes (up to twist) to tamely ramified primes, so the previous argument ends
our proof. For a prime t in the conductor of the wildly ramified case, let us call
t the rational prime below t and consider a mod t congruence, up to twist by
some finite order character ψ, with the Galois representation corresponding to
a Hilbert newform H of parallel weight 2 with at most t to the first power in the
level (i.e. 1 at t), and the same for the other primes dividing t. The existence
of such a form comes from Lemma 8.32.
By level-lowering, we can assume that the only extra primes in the level of
H are those primes that have been introduced to the residual conductor while
twisting by a character ψ. It is easy to see that such character can be chosen
such that at primes other than those dividing t it has square-free conductor. To
this congruence, Theorem 3.2.2 of [BD] applies (the conditions for this the-
orem are preserved by twisting, and modularity too) so we are reduced to a
case where we have a system that is either Steinberg or tamely ramified princi-
pal series at primes dividing t, and tamely ramified principal series at all extra
primes introduced by ψ. If we iterate this process at all wildly ramified primes,
we end up with a system with no wildly ramified primes. We repeat the previ-
ous case procedure of killing all ramification given by tamely ramified charac-
ters, but now in the absence of wild ramification we finish with a compatible
212 Luis Dieulefait and Ariel Pacetti

system such that all its ramified primes other than the good-dihedral prime are
Steinberg. It is not hard to see that in all this process, for a suitably chosen
bound B, all auxiliary primes can be taken to be smaller than B.

Remark 8.34. Except for the application of Corollary 2.12 of [BDJ10] at a

key point, and a better control of the local types (which seems reasonable
for abstract representations), all congruences in the above proof are known
to exist for abstract Galois representation, so an analogue of the above result
for abstract compatible systems can be proved assuming that this result from
[BDJ10] generalizes to the abstract setting (thus relating the modularity of any
geometric compatible system to that of a system with only Steinberg primes).

8.5.2 Killing the Steinberg primes

This part of the process is a little more delicate, and the MLT theorems are
more restrictive, so what we do first is move the Steinberg primes to primes
which split completely in F .

Theorem 8.35. Suppose that Assumption 8.15 is true. Let {ρ1,λ } be a system of
modular continuous, odd, irreducible, parallel weight 2 representations with
a big locally good dihedral prime q. Let L |  ( ∈ Z) be a Steinberg prime
of the system which does not split completely in F. We also assume that the
system is either unramified or Steinberg at all primes dividing . Then:
● there exists a strictly compatible system {ρ2,λ } of continuous, odd, irre-
ducible, parallel weight 2 representations, which has the same ramification
behavior at all primes except those dividing , where it is unramified, and
has at most a set of extra Steinberg primes, all of them dividing the same
rational prime which splits completely in F.
● there exists a chain of MLT congruences linking the two systems.

In particular, the system {ρ2,λ } is also modular.

Proof. We look at ρ1,λ for a prime λ dividing  and we reduce it modulo

. We can construct then a modular lift which is unramified at L and at all
primes dividing , and with weights among those predicted by the Serre’s
weights of the residual representation. This follows from iterated applications
of Lemma 8.32 since the determinant locally at primes above  is a fixed power
of the cyclotomic character, and the form (without  in the level) has parallel
weight as well. Note that this congruence is MLT, at least under Assump-
tion 8.15, which applies because at Steinberg primes, weight 2 representations
 Connectedness of Hecke algebras and the Rayuela conjecture 213

are ordinary, and the crystalline lift of higher weight can also be taken to be
ordinary (observe that to deduce modularity of the weight 2 family from the
other one we can apply Theorem 3.2.2 in [BD]).
Let p be a big prime (bigger than the weights of this second family) which
splits completely in F, then by [Sno09] we can construct a third family with
parallel weight 2 by looking modulo p, and this congruence is MLT because of
the results in [BLGGT] (we are comparing a Fontaine-Laffaille with a poten-
tially Barsotti-Tate representation, so they are both potentially diagonalizable).
We apply the previous section method to this new family to end up with a
representation which is at most Steinberg at all primes dividing p as desired.
Observe that in this last step no extra ramified primes are introduced because
we do not have wild ramification at p (the parallel weight 2 lift implies tamely
ramified principal series at all primes above p).
Now that we have only totally split Steinberg primes in our family, we can
get rid of the Steinberg primes.

Theorem 8.36. Let {ρ1,λ } be a system of modular, irreducible, parallel weight

2 representations, whose ramification consists of a big locally good dihedral
prime q and Steinberg primes which split completely in F. Then:
● There exists a strictly compatible system {ρ2,λ } of continuous, odd, irre-
ducible representations which are only ramified at the locally good dihedral
prime q.
● There exists a chain of MLT congruences linking the two systems.
In particular, the system {ρ2,λ } is also modular.

Proof. The procedure is quite similar to the one in the previous Theorem, but
now we can use Theorem 8.9. For each Steinberg prime of residual characteris-
tic  we look at ρ 1,λ with λ | . There exists a minimal lift which is crystalline
at all primes of residual characteristic  by Lemma 8.32. Now, since the prime
is split, the congruence is MLT by the cited Theorem. The only delicate point
here is that to apply Theorem 8.9 we need the residual characteristic to be dif-
ferent from 2 and 3, so if we have such a small Steinberg prime, we first apply
Theorem 8.35 to transfer the ramification to Steinberg ramification at some
larger split prime.
Remark: In the previous two Theorems, modularity of the given system
was only used to ensure existence of the lift corresponding to a system with
no  in its conductor, and this was deduced from results of [BDJ10]. In the
case of abstract Galois representations, after computing the Serre’s weights of
214 Luis Dieulefait and Ariel Pacetti

the residual representation one should be able to propose locally at all primes
above  a crystalline lift of the residual local representation, but then the prob-
lem is that in order to apply results such as those in [BLGGT] that guarantee
existence of a global lift with such local conditions, one should be in a poten-
tially diagonalizable case. In particular, under the conjecture that all potentially
crystalline representations are potentially diagonalizable, the previous two the-
orems shall generalize to the abstract setting (of course, the conclusion that the
last system is modular shall also be removed).
Proof of Theorem 8.7. With the machinery developed in the previous chap-
ters, starting with two modular representations, we can take each of them to
representations which are only ramified at the good dihedral prime q. The only
problem is that while killing the Steinberg primes, we lost control over the
weights, so to take the families to parallel weight 2, we chose an auxiliary
prime p which splits completely in F and consider a parallel weight 2 lift of
each family modulo p, with the same local type at q. Note that we cannot assure
that the forms we get at the end of the process will be newforms for 0 ( pq2 ),
because at some of the prime ideals of F dividing p our representations could
be unramified.

8.6 Further developments

As mentioned before, although we have some good control on the level of the
forms we started with, the primes in the level are not explicit. One can go
further and change the primes in the level for smaller and concrete ones, so as
to check whether the connectedness of the Hecke algebras in Mazur’s Theorem
corresponds to a chain where all congruences are MLT or not. For this purpose,
one can use the notion of micro good dihedral primes. Adding a micro good
dihedral prime (which is chosen asking some splitting behavior in the base
field) one can get rid of the good dihedral prime, and also bound the Steinberg
primes in the level. These ideas, although standard (see for example [Die12b])
are more delicate, and involve some technicalities that we prefer to avoid in the
present article. With this control, one can give an algorithm that given a totally
real number field, checks whether our approach implies Base Change over that
field via a finite computation involving Hilbert modular forms. See [DP] for
more details.
Concerning Serre’s conjecture over a specified small real quadratic field, one
should carefully check that the chain of congruences constructed carries on to
the case of abstract representations, and then once having reduced the prob-
lem to the case of representations of concrete small invariants (those where
 Connectedness of Hecke algebras and the Rayuela conjecture 215

the above process concludes, after the introduction of the micro good dihedral
prime) one should connect such representations to some “base case” where
modularity or residual reducibility is known (applying for example the result
of Schoof recalled in the introduction), checking on the way that all congru-
ences are MLT (with the advantage that over quadratic fields there are also
MLT of Skinner and Wiles that deal with the residually reducible case, under
suitable assumptions). We plan to check in a future work if this strategy suc-
ceeds in giving a proof of Serre’s conjecture over some small real quadratic
field.

References
[ACGH85] E. Arbarello, M. Cornalba, P. A. Griffiths, and J. Harris. Geometry of
algebraic curves. Vol. I, volume 267 of Grundlehren der Mathematis-
chen Wissenschaften [Fundamental Principles of Mathematical Sciences].
Springer-Verlag, New York, 1985.
[BD] Christophe Breuil and Fred Diamond. Formes modulaires de Hilbert mod-
ulo p et valeurs d’extensions entre caractères galoisiens. Ann. Scient. de
l’E.N.S., to appear.
[BDJ10] Kevin Buzzard, Fred Diamond, and Frazer Jarvis. On Serre’s conjecture
for mod  Galois representations over totally real fields. Duke Math. J.,
155(1):105–161, 2010.
[BLGGT] Thomas Barnet-Lamb, Toby Gee, David Geraghty, and Richard Taylor.
Potential automorphy and change of weight. Ann. of Math., to appear.
[CG11] Frank Calegari and Toby Gee. Irreducibility of automorphic galois repre-
sentations of GL(n), n at most 5. arXiv:1104.4827 [math.NT], 2011.
[DG12] Luis Dieulefait and Toby Gee. Automorphy lifting for small  - appendix
b to “Automorphy of Symm 5 (G L(2)) and base change”. Journal de Math.
Pures et Appl., to appear
[Die04] Luis V. Dieulefait. Existence of families of Galois representations and new
cases of the Fontaine-Mazur conjecture. J. Reine Angew. Math., 577:147–
151, 2004.
[Die12a] Luis Dieulefait. Automorphy of Symm 5 (GL(2)) and base change. Journal
de Math. Pures et Appl., to appear.
[Die12b] Luis Dieulefait. Langlands base change for GL(2). Ann. of Math. (2),
176(2):1015–1038, 2012.
[Dim05] Mladen Dimitrov. Galois representations modulo p and cohomology of
Hilbert modular varieties. Ann. Sci. École Norm. Sup. (4), 38(4):505–551,
2005.
[DP] Luis Dieulefait and Ariel Pacetti. Examples of base change for real
quadratic fields. In preparation.
[Gur12] Robert Guralnick. Adequacy of representations of finite groups of lie type
– appendix a to “automorphy of symm 5 (gl(2)) and base change”. Journal
de Math. Pures et Appl., to appear
216 Luis Dieulefait and Ariel Pacetti

[HT13] Yongquan Hu and Fucheng Tan. The Breuil-Mezard conjecture for non-
scalar split residual representations. Ann. Sci. ENS, to appear.
[KK03] Chandrashekhar Khare and Ian Kiming. Mod pq Galois representations
and Serre’s conjecture. J. Number Theory, 98(2):329–347, 2003.
[KW09] Chandrashekhar Khare and Jean-Pierre Wintenberger. Serre’s modularity
conjecture. I. Invent. Math., 178(3):485–504, 2009.
[Maz77] B. Mazur. Modular curves and the Eisenstein ideal. Inst. Hautes Études
Sci. Publ. Math., (47):33–186 (1978), 1977.
[Roz12] Sandra Rozensztajn. Asymptotic values of modular multiplicities for GL2 .
arXiv:1209.5666, J. Théorie des Nombres Bordeaux, 2012, to appear.
[Sch12] René Schoof. Semistable abelian varieties with good reduction outside 15.
Manuscripta Math., 139(1-2):49–70, 2012.
[Ser98] Jean-Pierre Serre. Abelian l-adic representations and elliptic curves, vol-
ume 7 of Research Notes in Mathematics. A K Peters Ltd., Wellesley, MA,
1998. With the collaboration of Willem Kuyk and John Labute, Revised
reprint of the 1968 original.
[Sno09] Andrew Snowden. On two dimensional weight two odd representations of
totally real fields. arXiv:0905.4266v1 [math.NT], 2009.
[SW99] C. M. Skinner and A. J. Wiles. Residually reducible representations and
modular forms. Inst. Hautes Études Sci. Publ. Math., (89):5–126 (2000),
1999.
 9
Big image of Galois representations and
congruence ideals
Haruzo Hida and Jacques Tilouine
Department of Mathematics, UCLA, Los Angeles, CA 90095-1555, U.S.A.,
Université Paris 13, Sorbonne Paris-Cité, LAGA, CNRS (UMR 7539), 99 av. J.-B.
Clément, F-93430, Villetaneuse
E-mail address: [email protected], [email protected]

Contents
9.1 Introduction 218
9.2 Galois representations associated to Siegel modular forms 219
9.3 Fullness of the image for Galois representations in GSp(4) 220
9.3.1 Irreducibility and open image 220
9.3.2 Connected components of the image of Galois 223
9.3.3 Connected semi-simple subgroups of GSp(4) 224
9.3.4 Z p -Fullness for GSp(4) and its subgroups 225
9.3.5 Pink’s Lie algebra theory 227
9.3.6 -fullness for the endoscopic cases 229
9.3.7 Z p [[S1 ]] ⊕ Z p [[S2 ]]-fullness in Case (Y) 233
9.4 Fullness of the Galois image in general Spin groups over
Iwasawa algebras 234
9.4.1 Spin groups 234
9.4.2 Genus n Hida families 235
9.4.3 n -fullness 240
9.5 Galois level and congruence ideals 247
9.5.1 Congruence ideals 247
9.5.2 Comparison between Galois level and congruence ideals 248
References 252
Date: August 16, 2014
The first author is partially supported by the NSF grant: DMS 0753991. The second author is
partially supported by the ANR grant: ArShiFo ANR-10-BLAN-0114

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

217
218 Haruzo Hida and Jacques Tilouine

9.1 Introduction
Let n ≥ 1; we consider the n × n antidiagonal unit matrix  s =
0 s
(δi,n+1− j )1≤i, j≤n and the 2n × 2n antisymmetric matrix J = ;
−s 0
we denote by G = GSp2n the Chevalley group of symplectic similitude matri-
ces for J and by B = T N its standard Borel consisting of upper triangular
matrices in G; let ρ be the half-sum of positive roots for (G, B, T ) and
λ ∈ X ∗ (T ) a dominant weight. Let π be a cuspidal automorphic represen-
tation on G(AQ ) of level M ≥ 1, whose infinity type is in the discrete series,
with infinitesimal character λ + ρ. It occurs in the cohomology of the genus
n Siegel variety of level M, with coefficients in the local system of highest
weight λ. It implies that there exists a number field K 0 whose ring of integers
O0 contains all Hecke eigenvalues of π (for prime-to-M Hecke operators).
Let G Q = Gal(Q/Q) and G ! be the dual Chevalley group for (G, B, T ); one
!
has G = GSpin2n+1 (see below); we assume that the compatible system of
Galois representations ρπ,ι : G Q → G(Q !  ) associated to π and to embed-
dings ι : Q → Q is constructed (for the moment, it is known for n = 1, 2).
Let p |M be a rational prime. Assume that π is ordinary for ι p ; there exists an
n-variable Hida family passing by π ; more precisely, let n be the n-variable
Iwasawa algebra and T M be the ordinary prime-to-M Hida-Hecke algebra
[MT02], [H02], [Pi12]; it is finite torsion-free over n . A Hida family is a
n -algebra homomorphism θ : T M → I where I is an integrally closed finite
torsion-free extension of n .
By interpolating the pseudo-representations associated to the arithmetic spe-
cializations of θ (see Sect. 4.2), one constructs a continuous homomorphism
ρθ : G Q → G(I)! associated to θ (see Lemma 9.12 in the text). We assume
throughout that the residual representation is irreducible. Following the ideas
of [H13b], we prove first, under some assumptions (mostly I = n and Z p -
regularity), that if the Hida family θ is “generic”, the image of ρθ is “n -full”
(see 9.17); more precisely, there is a non-zero ideal l of n such that the image
of Galois contains the principal congruence subgroup G! (l) in G ! (n ), where
!  !
G denotes the derived group of G. An important point of the proof is to create
a n -structure on the submodule of a Lie algebra associated to Im ρθ ; in the
case n = 1, Hida [H13b] used conjugation by certain elements of the inertia;
a suitable generalization using repeated use of Poisson brackets provides the
desired structure of n -module. Moreover, we ask the relation between the
maximal such ideal lθ , called the Galois level of θ , and the schematic inter-
section of the irreducible component Spec (I) with other components, more
precisely, at least if I = n , we ask whether the set of primes containing lθ
coincides with the union of the zero loci in n of the greatest common divisor
 Big image of Galois representations and congruence ideals 219

of the congruence ideal of θ and those of other non-generic families congruent

to θ. The equality of these has been established for n = 1 by Hida [H13b].
We prove this equality for n = 2 in the case of twisted Yoshida lift congru-
ence. In a subsequent paper, we plan to formulate more precisely this relation
(including exponents of height one prime ideals) and generalize it to more
general groups. The question of replacing n -fullness by I-fullness, or rather
I -fullness for a suitable subring of I is currently investigated, by a student of
one of the authors1 .

9.2 Galois representations associated to Siegel

modular forms
Let G = GSp2n be the Chevalley group of symplectic similitude matrices for
J . Let ν : G → Gm be the similitude factor character. Its kernel Sp2n coincides
with the derived group G  of G. We denote by B = T U its standard Borel
(consisting of upper triangular matrices in G).
(n)
Let Sk (M) be the space of genus n holomorphic Siegel cusp forms of
weight k = (k1 , . . . , kn ) with k1 ≥ k2 ≥ . . . ≥ kn ≥ n + 1, and level M (say,
for the principal congruence subgroup of level M). Let TkM be the M-spherical
Z[ M1 ]-algebra generated by the Hecke operators T,i for all rational primes 
prime to M and all i = 1, . . . , n, acting on Sk(n) (M). Recall that T,i denotes
the action of the double class (M) · d,i · (M) where d,1 = diag(1n ,  · 1n )
(n)
and d,i = diag(1n+1−i ,  · 12i , 2 · 1n+1−i ) for i = 2, . . . n. Let f ∈ Sk (M)
be an eigenform for TkM ; let K 0 be a number field containing its eigenvalues;
let O0 be its ring of integers. It gives rise to a character θ f : TkM → O0 . For
any prime  not dividing M, we denote by P f, (X ) the Hecke polynomial (see
[FC90] Chapt.VII, Sect. 1); it is monic of degree 2n .
Fix an odd prime p relatively prime to M, an embedding ι p : Q → Q p and
assume that f is ordinary for ι p . It means that the multiset of p-adic valuations
of the (images by ι p of the) roots of the Hecke polynomial P f, p (X ) coincides

with the multiset { i∈I (ki − i); I ⊂ [1, n]} of Hodge weights.
To the data (G, B, T ), one can associate a dual reductive group over Z[ 12 ]
with a standard Borel and a standard torus (G, ! ! !), with an identification
B, T
X ∗ (T ) = X ∗ (T!). By comparing the root data (with épinglage, see [MT02]
Sect. 3.2.2), one has a canonical identification G ! = GSpin2n+1 over Z[ 1 ].
2
!
Therefore G is endowed with a linear representation spin : G ! → GL2n and an
orthogonal representation π : G ! → SO2n+1 whose kernel is the center Gm of
! (the central inclusion Gm → G
G ! is dual to the similitude factor G → Gm ). It
!
also carries a similitude factor G → Gm , dual to the central inclusion Gm →
G; its kernel coincides with the derived group G ! = Spin2n+1 .
1 H. Hida
220 Haruzo Hida and Jacques Tilouine

For any number field F, we put G F = Gal(F/F) and for any finite place v
of F, we denote by Dv a decomposition subgroup at v in G F ; let : G F → Z× p
be the p-adic cyclotomic character. One still denotes by  its restriction to
Dv . We assume that there exists a Galois representation ρ f : G Q → G(Q ! p)
unramified outside M p and such that for any rational prime  not divid-
ing M p, the characteristic polynomial Char(spin ◦ρ f (Frob )) coincides with
ι p (P f, (X )). It is called the Galois representation associated to f . The integer
w = k1 + · · · + kn − n(n+1) 2 is called the motivic weight of ρ f .
The existence of ρ f is known for n = 1 (Deligne) and for n = 2 (due to R.
Taylor, Laumon and Weissauer, in [Ta93], [We05] and [Lau05]). By compact-
ness, there exists a p-adic field K ⊂ Q p such that ρ f is defined over K . We
may assume that ι p (K 0 ) ⊂ K , hence if O denotes its valuation ring, we have
ι p (O0 ) ⊂ O. One can choose an O-lattice stable in the orthogonal represen-
tation π ◦ ρ f . We assume it is unimodular. It implies that ρ f takes values in
!
G(O). Let 2 be a uniformizing parameter of O and F = O/2 O its residue
field. Let ρ f : G Q → G(F) ! be the residual representation.
Under the assumption of “automorphic ordinarity”, it is conjectured that the
n
following “Galois ordinarity” holds. Let ρ  = i=1 2i be the character of T
sum of all the fundamental weights of G; it is given by ρ on T  = T ∩G  and by
! via X ∗ (T ) =
n(n+1)
z → z 2 on the center. Let us view λ+ ρ as a cocharacter of T
X ∗ (T!). For ! g, !h ∈ G, ! let us put ! !
h = !
g g !
−1 h!g . Write D p for a (chosen)
decomposition group at p inside Gal(Q/Q) with the inertia subgroup I p .
(GO) There exists ! !
g ∈ G(O) such that ρ f (D p ) ⊂ ! !
g · B(O) g −1 and for any
·!
!
g !(O)) is given by (λ + ρ
σ ∈ I p , ρ f (σ ) (mod N ) ◦ (σ ).
“Automorphic ordinarity” implies “Galois ordinarity” is known for n = 1
(due to Wiles) and for n = 2 (due to Urban [Ur05]). We assume in the sequel it
holds for any n. It would follow from Katz-Messing theorem if one knew that
spin ◦ρ f is motivic.
In order to motivate the assumptions we shall make in genus n, let us prove
some results in genus 2, where the results have less conditions.

9.3 Fullness of the image for Galois representations

in GSp(4)
9.3.1 Irreducibility and open image
Let n = 2 and f be a genus 2 cusp eigenform of weight k = (k1 , k2 ),
k1 ≥ k2 ≥ 3, level M, as above. Fix an odd prime p prime to M at which
f is ordinary; let P f, p (X ) = (X − α)(X − β)(X − γ )(X − δ) be the
Hecke polynomial of f at p, with roots ordered with increasing p-adic val-
uation (so that ord p (α) = 0, ord p (β) = k2 − 2, ord p (γ ) = k1 − 1 and
 Big image of Galois representations and congruence ideals 221

ord p (α) = k1 + k2 − 3); we assume that all the Hecke eigenvalues, together
with α, β, γ and δ, are contained in O and that the Galois representation
associated to f ([Ta93], [We05] and [Lau05]) is defined over O: ρ f : G Q →
GSp4 (O) where the group of symplectic similitudes is relative to the matrix
 0 0 0 1
J = 00 −1 0 10
0 0 . Recall that the spin representation provides an isomorphism
−1 0 0 0
GSpin5 ∼ = GSp4 ⊂ GL4 . Let (L , ψ) be the O-module of the representation
endowed with a unimodular symplectic pairing ψ; we fix a symplectic O-basis
(e1 , e2 , e3 , e4 ) of L with ψ(e1 , e4 ) = ψ(e2 , e3 ) = 1 so that we have an identi-
fication GSp(L , ψ) = GSp4 (O); let V = L[ 1p ] be the K -vector space spanned
by L. We have
⎛ k +k −3 δ
⎞
 1 2 ur ( k1 +k 2 −3
) ∗ ∗ ∗
⎜ p
⎟
⎜ 0  k1 −1 ur ( pkγ1 −1 ) ∗ ∗ ⎟
ρ f |Dp ∼ ⎜ ⎜ ⎟.
⎝ 0 0  k2 −2 ur ( kβ2 −2 ) ∗ ⎟ ⎠
p
0 0 0 ur (α)
Here ur (x) : D p → O× is the unramified character sending Frob p to x ∈ O× .
We shall assume in the sequel that the conjugation needed for this description
is the identity. We also put ρ : G Q → GSp4 (F) its reduction modulo 2 .
We first show
Proposition 9.1. If f is neither CAP nor endoscopic, and if either
(i) Langlands transfer holds from GSp(4) to GL(4) (no need of the ordinarity
assumption then),
(ii) p > 7, the semisimplification (ρ|G Q(ζ p ) )ss contains a copy of SL2 (F p ),
and the p-adic units k1 +k δ
2 −3
, kγ1 −1 , β
and α are distinct modulo 2 ,
p p pk2 −2
then ρ f is absolutely irreducible.
In the first case, one can use [CaGe13]. The transfer has been established by
Arthur for Sp(4), but not yet for GSp(4), unless f admits a generic form with
the same eigenvalues.
In the second case, we proceed by case inspection. If there is a stable line
D = e2  in L, let P1 = e2 , e3  be a hyperbolic plane of L containing D and
P2 = e1 , e4  its orthogonal, so that P1 ⊥ P2 = L be a decomposition of L
into two orthogonal hyperbolic planes (it is possible by unimodularity of the
symplectic pairing on L). By taking (e1 , e2 , e3 , e4 ) as symplectic basis of L
adapted to this decomposition, we see in this basis that
⎛ ⎞
∗ 0 • ∗
⎜• χ • •⎟
ρf = ⎜ ⎝ 0 0 χ 0 ⎠ .
⎟

∗ 0 • ∗
222 Haruzo Hida and Jacques Tilouine

hence,
● ρ f admits a degree 1 subrepresentation χ which is locally algebraic at p
and finitely ramified, hence it is given by an (A0 )-type Hecke character of
Q, which cannot be of finite order because χ ( p) cannot be of archimedean
absolute value one by Deligne purity for α, β, γ and δ, so that χ cannot have
Hodge-Tate weight 0: it must be k2 − 2 or k1 − 1
● ρ f has a degree 2 subquotient σ = ( ∗∗ ∗∗ ), which is odd because its
determinant is the similitude factor of ρ, which is odd.
Moreover, σ is p-ordinary with two distinct Hodge-Tate weights (0 and k1 +
k2 −3) and its residual representation is irreducible over G Q(ζ p ) by assumption.
Therefore, the representation σ is modular: σ = σg for an ordinary form g of
weight  ≥ 2 and level prime to p by Emerton’s theorem on Fontaine-Mazur
Conjecture ([Em14, Cor.1.2.2]); this implies by the classification of Vogan-
Zuckerman (see [Ta93] Section 1 or [Ti09] Section 6) that, k1 = k2 = k,
 = 2k − 2, χ =  k−1 · ( f inite), and that the L-function of f is of the form
L(χ , s)L(g, s)L(χ , s − 1), hence f is CAP of Siegel type (i.e. is a Saito-
Kurokawa lift), contrary to the assumption.
Similarly, if there is a stable isotropic plane, ρ f contains a two-dimensional
subrepresentation; it is p-ordinary with distinct Hodge-Tate weights, p > 7,
and the image of its reduction modulo 2 restricted to G Q(ζ p ) contains SL2 (F),
it is odd by Cor.1.3 of Calegari [Ca13]; so again by [Em14, Cor.1.2.2],
Fontaine-Mazur Conjecture holds and shows that f is CAP of Klingen type,
contradiction. Finally, if there is a stable hyperbolic plane, we see by a simi-
lar argument (but without using Calegari’s argument, hence without assuming
p > 7 and the SL2 (F p ) condition, because the oddness is obvious in this case),
that f is a Yoshida lift, so it is endoscopic. QED.

Proposition 9.2. Assume that the adjoint representation of ρ f on the Lie alge-
bra sp4 is irreducible. Then (without assuming ordinarity), the image of Galois
is full, i.e. contains a conjugate in G(Q p ) of a congruence subgroup of G(Z p ).

We first show that the image of Galois is open in its Zariski closure. Say that
Im ρ f ⊂ GSp4 (E) for a finite extension E/Q p , and let G = Res E/Q p GSp4 .
Let g, resp. G, be the Lie algebra over Q p of image of Galois (with its natural
structure of p-adic Lie group), resp. of its Q p -Zariski closure in G. For any
embedding σ : E → Q p , let Gσ resp. gσ be the σ -projection of G ⊗ Q p resp.
g ⊗ Q p ; By a theorem of Chevalley [Che51, p.177, Th.15], we see that, for any
σ , the derived Lie algebras gσ and Gσ coincide. By absolute irreducibility of
ρ f , the center of gσ has rank at most one. Moreover, since the similitude factor
 Big image of Galois representations and congruence ideals 223

ν ◦ ρ is a non-trivial power of the p-adic cyclotomic character, the rank of the

center is exactly one for any σ , so that g ⊗ Q p = G ⊗ Q p .
The possibilities for G ⊗ Q p are listed below

(F) G ⊗ Q p ∼ sp4 .
(Y) G ⊗ Q p ∼ sl(2) × sl(2)
(K) G ⊗ Q p ∼ sl(2) with sl(2) embedded into a Klingen parabolic
subalgebra.
(C) G ⊗ Q p ∼ sl(2) via the symmetric cube representation of SL(2),
(S) G ⊗ Q p ∼ sl(2) with sl(2) embedded into a Siegel parabolic subalgebra.
(T) G ⊗ Q p ∼ {1}

Let us show that in all cases except (F), the adjoint representation of ρ f on
gsp4 (denoted by adding the superscript “ad”) would be reducible. Indeed, in
 ad
all cases except (F), we see that Gad is reducible (note that Symm3 sl(2)
is also reducible). This shows that the Q p -Lie algebra g of Im ρ f is a Q p -form
of gsp4 (Q p ) contained in gsp4 (Q p ). Since Aut(gsp4 ) = Int(gsp4 ), it implies
that there exists α ∈ G(Q p ) such that

g = α · gsp4 (Q p ) · α −1 .

In order to get a more complete result without assuming adjoint irre-

ducibility, one needs to analyze the connected components of the Galois
image.

9.3.2 Connected components of the image of Galois

Let ρ = {ρl }l be an n-dimensional compatible system of l-adic representations
of Gal(Q/k) satisfying the assertions of [P98a, Theorems 3.2–3]. Write Vl ∼ =
Ql for the space of ρl . Replacing ρl by its semi-simplification, we may assume
n

that ρl is semi-simple for all l. Define an algebraic group G p/Q p over Q p by

the Q p -Zariski closure in GL(V p/Q p ) of Im(ρ p ) for a prime p, and let G ◦p
be its connected component. Since ρ p is semi-simple, G p is reductive. Then
by a result of Serre, G p /G ◦p and rank G ◦p are independent of p (see [P98a,
Theorem 3.6]). Let k conn be the fixed field of ρ −1 ◦
p (G p (Q p )). We fix a prime p
and write l for a prime ideal of k conn prime to p. Let ρ p (Frobl )ss be the semi-
simplification of ρ p (Frobl ) inside G p (Q p ). Write Tl for the smallest algebraic
subgroup over Q p containing ρ(Frobl )ss inside G ◦p (Q p ). Then the connected
component Tl◦ is a torus, and it is isomorphic to a base-change to Q p of a
Q-torus in GL(n). If Tl is a maximal torus of G ◦p , then ρ p (Frobl ) is already
semi-simple.
224 Haruzo Hida and Jacques Tilouine

Lemma 9.3 (Serre). Let p be a prime. The algebraic group G ◦p has a dense
Zariski open subset U stable under conjugation such that Tl is connected and
maximal if ρ p (Frobl )ss is in U (Q p ). In particular, for any prime l in a density
one subset of primes of k conn , Tl is maximal.

This follows from the argument in [Ch92, Theorem 3.7]. Though Chi assumes
that ρ comes from an abelian variety, his argument works in general.

9.3.3 Connected semi-simple subgroups of GSp(4)

Let ρ = {ρl }l be a strict compatible system of symplectic semi-simple rep-
resentations of Gal(k/k) into GSp4 (Tl ) for a number field T (associated to a
motive). Assume that T is the smallest coefficient field (i.e. it is generated by
trace of Frobenii). Here l runs over primes of T . Let l be the residual char-

acteristic of l. Let Res ρ = {ρl }l for ρl = l|l ρl regarded to have values in
GSp4 (T ⊗Q Ql ). Then Res ρ is a compatible system with coefficients in Q act-
ing on the Ql -vector spaces Vl = (T ⊗Q Ql )4 . Consider the Ql -Zariski closure
G l of Im(ρl ) in GL(Vl ).
As mentioned earlier, G l is a reductive subgroup of GL(Vl ) defined over Ql ,
rank G l◦ is independent of l and G l /G l◦ is independent of l and is isomorphic
to Gal(k conn /k) for the fixed field k conn of ρl−1 (G l◦ )
Since G l commutes with the action of T on Vl , for each prime factor l, we
have its projection G l in GSp4 (Tl ). Let G l◦ be the connected component of G l
(and G ◦l for the l-component of G l◦ ; so, G ◦l is the connected component of G l ).
Let G l be the derived group of G ◦l . We may regard G l ⊂ ResTl /Ql Sp(4).
Thus the projections of G l (Ql ) to each simple component Sp4 (Ql ) of
(ResTl /Ql Sp(4))(Ql ) are mutually isomorphic. We write G l ∼ G  for a semi-
simple connected subgroup G  of Sp(4)/Ql if its projections to any simple
factor Sp4 (Ql ) of ResTl /Ql Sp(4)/Ql are isomorphic to G  over Ql . Assume
that ρ has weight w, i.e. det ρ is, up to finite character, given by {lw }l for the
l-adic cyclotomic character l .
Let G be a connected semi-simple subgroup of Sp(4)/k for a field k, and
write g for its Lie algebra. Write Vn for the symplectic space of dimension 2n
on which Sp(2n) acts. Then by [LS98, Theorem 1], the largest proper semi-
simple connected subgroup G of Sp(4) is isomorphic to SL(2) × SL(2) over
an algebraic closure k. Making an identification V2 ∼ = V1 ⊕ V1 as symplectic
spaces, this group G acts diagonally on V1 ⊕ V1 . As for smaller semi-simple
connected subgroups, we have two possibilities G = 1 and G ∼ = SL(2). The
second possibility includes the case where the isomorphism SL(2) ∼ = G is
given by the symmetric cube representation of SL(2). Thus irreducibility of
 Big image of Galois representations and congruence ideals 225

G-module g (under the adjoint action) is satisfied if the derived group G is

either SL(2) (the image of symmetric cube) or Sp(4).
For a prime P over p of k, we suppose to have
⎛ w1 ⎞
 ∗ ∗ ∗
⎜ 0  w2 ∗ ∗ ⎟
ρ| IP ∼ ⎜
⎝ 0
⎟ (w1 > w2 > w3 > w4 ) (Reg)
0  w 3 ∗ ⎠
0 0 0  w4
up to finite error for the p-adic cyclotomic character  =  p of the inertia
group IP of P. Thus we have six possibilities of the derived group G p (the
group G ◦p is split over Tp by (Reg)):

(F) G p ∼ Sp(4).
(Y) G p ∼ SL(2) × SL(2) with ρpconn = ρ1,p ⊕ ρ2,p , where ρ1,p is not twist
equivalent to ρ2,p .
(K) G p ∼ SL(2) with ρpconn = χ ⊕χ  ⊕ρ1,p with irreducible two-dimensional
ρ1,p .
(C) G p ∼ SL(2) via the symmetric cube representation of SL(2).
(S) G p ∼ SL(2) with ρpconn = ρ1,p ⊕ (ρ1,p ⊗ χ ) with absolutely irreducible
two-dimensional ρ1,p .
(T) G p ∼ {1} with ρp = IndQF χ for a degree 4 CM field F/Q.

Here we write ρpconn for the restriction of ρp to Gal(Q/k conn ).

9.3.4 Z p -Fullness for GSp(4) and its subgroups

Note that Sp4 = ∼ Spin over Z p (for p > 2). Suppose that ρ is the (semi-
5
simple) system associated to a Siegel cusp form (so k = Q) with regular weight
and that ρp (for p| p) is ordinary as in Proposition 9.1. Recall that G p is the
Zariski closure of Im(ρp ), that G ◦p is the identity connected component of G p
and that G p is the derived group of G ◦p for a prime p| p of T . For a semi-
simple Q-split subgroup G of Sp(4) and a semi-simple Tp subgroup H , we
write H ∼ G if H ∼ = G ×Q Tp . Then as seen in Sect. 9.3.3, we have the
following six possibilities of G p :

(F) G p ∼ G for G = Sp(4).

(Y) G p ∼ G for G = SL(2) × SL(2) with ρpconn = ρ1,p ⊕ ρ2,p , where ρ1,p is
not twist equivalent to ρ2,p .
(K) G p ∼ G for G = SL(2) with ρpconn = χ ⊕ χ  ⊕ ρ1,p with irreducible
two-dimensional ρ1,p .
(C) G p ∼ G for G = SL(2) via the symmetric cube representation of SL(2).
226 Haruzo Hida and Jacques Tilouine

(S) G p ∼ G for G = SL(2) with ρpconn = ρ1,p ⊕ (ρ1,p ⊗ χ ) with absolutely

irreducible two-dimensional ρ1,p .
(T) G p ∼ {1} with ρp = IndQ
F χ for a degree 4 CM field F/Q.

Here Im(ρ j,p ) ( j = 1, 2) contains an open subgroup of SL2 (Z p ) by the def-

inition of G p , and χ , χ  are characters. In the automorphic side, Case (Y)
corresponds to the Yoshida lift from an automorphic form GL(2) × GL(2)/Q
or Res F/Q GL(2) for a real quadratic field F, under the hypothesis that the
starting automorphic form is not endo-scopic (i.e. non-CM). Cases (K) and (S)
are either CAP or Eisenstein associated to the Siegel parabolic (S) and Klin-
gen parabolic (K), assuming the corresponding GL(2) automorphic form is not
endoscopic. Case (C) is the symmetric cube lift from GL(2)/Q .
If G is absolutely simple, let g be the Lie algebra of G and put gp = g⊗Q Tp
and regard it as the Lie algebra of G p . Since Im(ρp ) is Zariski dense in G p ,
the adjoint action of ρp is absolutely irreducible. Write Ad(ρp ) for the adjoint
representation acting on gp , and let E p be the subfield of Tp generated over Q p
by Tr(Ad(ρp )) on the Galois group. Let W be the p-adic integer ring of E p .
If G = SL(2) × SL(2) (i.e. we are in Case (Y)), we consider Ad(ρ j,p ) acting
on sl2 (Tp ). We then define E j by the subfield of Tp generated over Q p by the
values of Tr( Ad(ρ j,p )) over Gal(Q/k conn ). Then we write W j for the p-adic
integer ring of E j .

Proposition 9.4. Let the notation be as above. Suppose p > 3. If G is

absolutely simple, Im(ρp ) contains an open subgroup of G(W ). If G =
SL(2) × SL(2) (i.e. in Case (Y)), then Im(ρp ) contains an open subgroup
of SL2 (W1 ) × SL2 (W2 ).

Proof. Suppose first that G is absolutely simple (i.e. we are not in Case
(Y)); so, G is either SL(2) or Sp(4). Thus the adjoint representation of G
on its Lie algebra g is absolutely irreducible. We regard Ad(ρp ) restricted
to Gal(Q/k conn ) as having values in G ad (the adjoint group of G). Since
Im(ρpconn ) is Zariski dense in G, Ad(ρpconn ) is absolutely irreducible. Then
by a theorem of Weisfeiler (see [P98b, Theorem 0.7]), there exists a linear
algebraic group H defined over E p in Tp such that H × Ep Tp = G ad ×Q Tp
with Im(Ad(ρpconn ) being an open subgroup of H (W ). Since the Siegel cusp
form has regular weight, Im(ρpconn ) contains p-inertia regular element g such
that any of its power g n (0 < n ∈ Z) has a split torus of maximal rank in G as
its centralizer. Thus H is split, and the result follows as G → G ad is a central
isogeny.
Now we assume that we are in Case (Y). Applying the above argument to
ρ j,p , Im(ρ j,p ) contains an open subgroup of SL2 (W j ). Thus we need to show
 Big image of Galois representations and congruence ideals 227

that Ad(ρ1 ) × Ad(ρ2 ) contains an open subgroup of PGL2 (W1 ) × PGL2 (W2 ).
Let P G be PGL(2) defined over F; so, P G(F) = PGL2 (Tp ) × PGL2 (Tp ).
By [P98b, Main Theorem 0.2 (a)], there exist a semi-simple Q p -subalgebra
E of F := Tp ⊕ Tp and a connected adjoint group H/E and an isogeny ϕ :
H × E F → P G such that Im(ρpconn )) is an open subgroup of ϕ(H (E)). If E
is a field, we have E = E 1 = E 2 , and ρ1,P ∼= ρ2,p , and hence we are in Case
(S), a contradiction. Thus E = E 1 ⊕ E 2 , and H (E) = PGL2 (E 1 ) × PGL2 (E 2 )
as desired.

9.3.5 Pink’s Lie algebra theory

Let G be a split (smooth) connected reductive Z p -group with maximal split
torus T . Let G  be the derived group of G, and put G A (c) to be the kernel
of the reduction map G  (A)  G  (A/c) for a non-zero ideal c of A for a
Z p -algebra A. For a Z p -algebra B ⊂ A, a subgroup H of G(A) is B-full
if H ⊃ G B (c) for a non-zero B-ideal c. Let ρ : Gal(Q/K ) → G(A) be a
continuous representation for a finite extension K /Q. Suppose A is an integral
domain with quotient field Q(A). We call ρ B-full with respect to G if after
replacing ρ by a conjugate of ρ under an element of G(Q(A)), Im(ρ) is B-full.
Suppose that B is a local ring with maximal ideal m B . We call ρ B-regular if
Im(ρ) contains a regular element b in the torus T (B) such that α(b) ≡ β(b)
mod m B for all distinct roots α and β.
Via Pink’s theory of tight correspondence between p-profinite Lie subalge-
bras of sl(2) and p-profinite subgroups, we want to show that if a p-profinite
subgroup of Spin2n+1 contains sufficiently large unipotent subgroups, it is full.
Let us recall Pink’s theory briefly. To study a general p-profinite subgroup G
of SL2 (A) for a general p-profinite ring A, we want to have an explicit rela-
tion between p-profinite subgroups G of the form SL2 (A) ∩ (1 + X ) for a Lie
Z p -subalgebra X ⊂ gl2 (A). Assuming p > 2, Pink [P93] found a functorial
explicit relation between closed subgroups in SL2 (A) and Lie subalgebras of
gl2 (A) (valid even for A of characteristic p). We call subgroups of the form
SL2 (A) ∩ (1 + X ) (for a p-profinite Lie Z p -subalgebra X of gl2 (A)) basic
subgroups following Pink.
We prepare some notation to quote here the results in [P93]. Let A be a semi-
local p-profinite ring (not necessarily of characteristic p). Since Pink’s result
allows semi-local p-profinite algebra, we do not assume A to be local in the
exposition of his result. We assume p > 2. Define maps # : SL2 (A) → sl2 (A)
and ζ : SL2 (A) → Z (A) for the center Z (A) of the algebra M2 (A) by
   
1 1 0 1 1 0
#(x) = x − Tr(x) and ζ (x) = (Tr(x) − 2) .
2 0 1 2 0 1
228 Haruzo Hida and Jacques Tilouine

For each p-profinite subgroup G of SL2 (A), define L by the closed additive
subgroup of sl2 (A) topologically generated by #(x) for all x ∈ G. Then we
put C = Tr(L · L). Here L · L is the closed additive subgroup of M2 (A)
generated by {x y|x, y ∈ L} for the matrix product x y, similarly L j is the
closed additive subgroup generated by iterated products ( j times) of elements
in L. We then define L 1 = L and inductively L j+1 = [L , L j ]; so, L 2 =
[L , L], where [L , L j ] is the closed additive subgroup generated by Lie bracket
[x, y] = x y − yx for x ∈ L and y ∈ L n . Then by [P93, Proposition 3.1],
we have
[L , L] ⊂ L, C · L ⊂ L , L
3 3
= L 1 ⊃ · · · ⊃ L j ⊃ L j+1 ⊃ · · · and Ln = L j = 0. (9.1)
j≥1 j ≥1

In particular, L is a Lie Z p -subalgebra of sl2 (A). Put M j (G) = C 10 01 ⊕
L j ⊂ M2 (A) = gl2 (A), which is a closed Lie Z p -subalgebra by (9.1). Define
H j = {x ∈ SL2 ( A)|#(x) ∈ L j , Tr(x) − 2 ∈ C} for j ≥ 1.

If x ∈ Hn , then x = #(x) + ζ (x) + 10 01 ; thus, H j ⊂ SL2 (A) ∩ (1 + M j (G)).
If we pick x ∈ SL2 (A) ∩ (1 + M j (G)), then x = 1 +  1c0· 1 + y with y ∈ Ln
and c ∈ C. Thus Tr(x) − 2 = 2c ∈ C and #(x) = 0 1 + c · 0 1 + y − 1 0
1 0
2 (2 + 2c) · 0 1 = y. This shows
1

H j = SL2 (A) ∩ (1 + M j (G)).

Here is a result of Pink (Theorem 3.3 combined with Theorem 2.7 both in
[P93]):

Theorem 9.5 (Pink). Let the notation be as above. Suppose p > 2, and let A
be a semi-local p-profinite commutative ring with identity. Take a p-profinite
subgroup G ⊂ SL2 (A). Then we have
(1) G is a normal closed subgroup of H1 (defined as above for G),
(2) Hn is a p-profinite subgroup of SL2 (A) inductively given by H j+1 =
(H1 , H j ) which is the closed subgroup topologically generated by com-
mutators (x, y) with x ∈ H1 and y ∈ H j ,
(3) {H j } j≥2 coincides with the descending central series of {G j } j≥2 of G,
where G j+1 = (G, G j ) starting with G1 = G.
In particular, we have
(P) The topological commutator subgroup G  of G is the subgroup given by
SL2 (A) ∩ (1 + M2 (G)) for the closed Lie subalgebra M2 (G) ⊂ M2 (A)
as above.
 Big image of Galois representations and congruence ideals 229

Recall that G is a split (smooth) connected reductive Z p -group with max-

imal split torus T . Let G  be the derived group of G. Now let G be a
p-profinite subgroup of G  (A). Then on G1 = G ∩ G A ( p) the logarithm
∞ m+1 (x−1)m converges to an element Log(x) ∈ Lie(G)(A). Note
m=1 (−1) m
that in general Log(G) = {Log(x)|x ∈ G} is neither an abelian group nor a Lie
algebra. Fix a Borel subgroup B/Z p of G containing T with B = T U + for the
unipotent radical U + . Write U − for the unipotent subgroup opposite to U + ;
so, T U − is the opposite Borel subgroup of B.

Lemma 9.6. Suppose that A is a p-profinite complete local integral domain

+ −
over Z p for p > 2. If G2 contains UA (a) and UA (b) for non-zero ideals a
and b of A, then G2 (and hence G) contains GA (a b ).
2 2

Proof. We write u+ (resp. u− , t) for the Lie subalgebra of Lie(G) correspond-

ing to U + (resp. U − , T ). We start with the case where G = SL(2). Note that
GSpin3 ∼ = SL(2).
In our setting, we assume that A is local. Then by the definition as
above, under the notation in Theorem 9.5, we have Log(G1 ∩ U + (A)) =
M1 (G1 ) ∩ U + (A) and Log(G −
1 10 a∩ /U (A)) 2 = M
−
1 (G1 ) ∩1U ( A).
/ Thus2
+
M1 (G1 ) contains u (a) = /a ∈ a and u (a ) =
−  0 0 /a ∈ a .
0 01 / 2 a 0
0 /a ∈ aa is contained in M (G  ).
Thus [u+ (a), u− (a )] ⊃ t(aa ) := a0 −a 1
In other words, M1 (G1 ) ⊃ M1 ( G (aa  )). This implies G  contains the
A 1
 G 2 2
derived group of G A (aa ) which contains A (a a ). This finishes the proof
for n = 1.
For general G, we pick a positive root α of T with root subgroup Uα ⊂ U + .
Then we have an embedding i α : SL(2) → G sending the upper (resp. lower)
triangular unipotent subgroup of SL(2) to Uα (resp. U−α ⊂ U − ). Applying the
argument in the case of SL(2) to Gα := i α−1 (G), we have Gα ⊃ A (a2 b2 ).
SL(2)
SL(2)
It is easy to see that {i α ( A (a2 b2 ))}α generates G A (a b ), and hence the
2 2

desired assertion holds.

9.3.6 -fullness for the endoscopic cases

We return to the setting of Sect. 9.3.4. We now study -fullness in the one
variable case of  = 1 = Z p [[T ]] in the endoscopic cases (Y), (K), (C) and
(S) below. In each endoscopic case, we have a factor GL(2) of G. Even if the
component of the Hecke algebra has two variables, the Galois image of the
projected factor falls in GL2 (I) for I finite over  (for a suitable quotient 
of 2 ). Thus we do not lose any generality assuming that I is finite over one
230 Haruzo Hida and Jacques Tilouine

variable . To describe the quotient map 2 →  explicitly, we normalize

first 2 = Z p [[T1 , T2 ]] so that the Galois image to contain
T = {diag[(t1 t2 )s , t1s , t2s , 1] ∈ T (2 )|s ∈ Z p , t j = 1 + T j }
as the wild p-inertia image. Then we normalize the Galois representation
attached an I-adic ordinary Hecke eigenform form as follows depending on
cases for ρ = ρI .
 
● In Case (Y), we write ρ ∼
a(σ ) b(σ )
= φ × ϕ with φ(σ ) = c(σ ) d(σ ) , ϕ(σ ) =
 
α(σ ) β(σ )
γ (σ ) δ(σ ) and
⎛ ⎞
a(σ ) 0 b(σ ) 0
⎜ 0 α(σ ) 0 ⎟
β(σ )
ρ(σ ) = ⎜ ⎟ and  = 2 /(t1 t2 − t1 )
⎝ 0 γ (σ ) 0 ⎠ with t1 → t = 1 + T .
δ(σ )
c(σ ) 0 d(σ ) 0
 
● In Case (K), ρ ∼ a(σ ) b(σ )
= χ ⊕ χ  ⊕ φ with φ(σ ) = c(σ ) d(σ ) and
⎛ ⎞
a(σ ) 0 0 b(σ )
⎜ 0 χ (σ ) 0 ⎟
ρ(σ ) = ⎜
0 ⎟ and  = 2 /(t1 − t2 )
⎝ 0 0 χ  (σ ) 0 ⎠ with t1 t2 → t = 1 + T .
c(σ ) 0 0 d(σ )
● In Case (S), ρ ∼ = (φ ι ⊗ χ ) ⊕ φ; i.e.,
 
χ (σ )s t φ(σ )ι s −1 0
ρ(σ ) = and  = 2 /(t1 − 1) with t2 → t = 1 + T.
0 φ(σ )
● In Case (C), and  = 2 /(t1 − t22 ) with t2 → t = 1 + T .
Suppose
(y) In Case (Y), n = 2, det ϕ = det ρ = νρ and for σ ∈ I p ,
 
(t1 t2 )log p ((σ )/ log p (1+ p) ∗
φ(σ ) = ,
0 1
 log ((σ )/ log (1+ p) 
t1 p p ∗
ϕ(σ ) = ,
0 t2 log p ((σ )/ log p (1+ p)
and φ has values in a finite extension Iφ ⊂ I of φ = Z p [[Tφ ]] for Tφ =
t1 t2 − 1 and ϕ ⊗ κ −1 has values in a finite extension Iϕ ⊂ I of ϕ =
Z p [[Tϕ ]] for Tϕ = t1 t2−1 − 1, where κ : G Q → × ϕ is a character such
log p ((σ )/ log p (1+ p)
that κ(σ ) = t2 .
 Big image of Galois representations and congruence ideals 231

log p ((σ )/ log p (1+ p)

(k) In Case (K), n = 2 and for σ ∈ I p , χ (σ ) = t1 , χ  (σ ) =
log ((σ )/ log p (1+ p)
t2 p and
 
(t1 t2 )log p ((σ )/ log p (1+ p) ∗
φ(σ ) = ,
0 1

and φ has values in a finite extension Iφ ⊂ I of φ = Z p [[Tφ ]] for Tφ =

t1 t2 − 1.
log ((σ )/ log p (1+ p)
(s) In Case (S), n = 2 and χ (σ ) = t1 p and φ(σ ) =
 log p ((σ )/ log p (1+ p) 
t2 ∗ for σ ∈ I , and φ has values in a finite extension
p
0 1
Iφ ⊂ I of φ = Z p [[Tφ ]] for Tφ = t2 − 1.
(c) In Case (C), n = 1 and for σ ∈ I p ,
⎛ ⎞
t1 3 log p ((σ )/ log p (1+ p) ∗ ∗ ∗
⎜ 0 t1 2 log p ((σ )/ log p (1+ p) ∗ ∗⎟
ρ(σ ) = ⎜
⎝
⎟
log p ((σ )/ log p (1+ p) ⎠ .
0 0 t1 ∗
0 0 0 1

Note here GSpin3 = ∼ SL(2) and GSpin = GSp(4) over Z p (for p odd).
5
Let I be a finite flat normal extension of  as before, and ρI : Gal(Q/Q) →
GSpin2n+1 (I) be a Galois representation. Suppose

(G0) G ⊃ T ; so, rank G = rank GSpin2n+1 ,

(G1) there exists a point P ∈ Spec ()(Q p ) such that ρ P = (ρI mod P) is
Z p -full with respect to G, regarding I/PI as a Z p -algebra.
(G2) the semi-simplification of ρI restricted to the wild p-inertia group I pw is
isomorphic to the diagonal representation diag[κ 2n−1 , κ 2n−2 , · · · , 1] for
the character κ : I pw → × such that κ([1 + p, Q p ]) = t = 1 + T .
(G3) (ν ◦ ρI )/κ n(2n−1) has finite order.
(G4) We have diag[κ 2n−1 , κ 2n−2 , · · · , 1] in ρI (I pw ).

Conjecture 9.7. Assume p > 2n − 1 > 2 in addition to (G0–4). Then ρI is

-full with respect to G.

Theorem 9.8. Suppose that n = 1, 2 and that we are not in Case (F) nor in
Case (T) (so we are in the endoscopic cases). Let ρI be associated to an I-adic
ordinary Hecke eigen cusp form. Assume Z p -regularity if I =  or in Case
(Y). Then the above conjecture holds.

We prove below the -fullness including p = 3.

232 Haruzo Hida and Jacques Tilouine

Proof. When n = 1, this follows from [H13b, Theorem I]. If I = , by [Z14],

we have a non-trivial upper unipotent element in ρI (I p ) and also a lower one;
so, (G4) is actually not necessary if I = . If I = , Z p -regularity is used
in [H13b] to show non-trivial unipotents can be found in SL2 () (not just in
SL2 (I) shown in [Z14]). Since we are in the cases different from (F) and (T),
G  is either split SL(2) or SL(2) × SL(2). This implies that, by GL(2)-theory,
we can find an arithmetic prime P ∈ Spec () (sufficiently regular) such that
ρ := ρP for every prime P ∈ Spec (I) above P belongs to the same case
except for Case (F). Then by Proposition 9.4 applied to ρ, ρ j,p is Z p -full, and
hence the argument in [H13b] works.
Since we have one simple component SL(2) except for Case (Y), if we
are not in Cases (Y) and (F), there is no difference in the argument prov-
ing the theorem from the one given in [H13b]. Thus we give more details
in Case (Y). In this case, ρI has values in GL2 (I) × GL2 (I). We write
ρ L , ρ R : Gal(Q/Q) → GL2 (I) for the projection of ρI to the left and right
factor GL2 (I). Each projection ρI, j ( j = 1, 2) to each simple component
GL(2) is -full by the argument in [H13b]. Pink’s theory can be also applied
SL(2)
to semi-local p-profinite rings, like A = I ⊕ I. Put G = Im(ρI ) ∩ I⊕I (mI )
SL(2)
for the maximal ideal mI . and G P = Im(ρ P ) ∩ I/P⊕I/P (mI/P ). Since the
reduction map modulo P is onto for Pink’s Lie algebra, we have a surjective
morphism M j (G)  M j (G P ). Adding the superscript “+” (resp. “−”, “0”)
to Pink’s Lie algebra, we indicates the upper nilpotent subalgebra (resp. the
lower nilpotent subalgebra, the diagonal subalgebra). Then by Z p -regularity,
M j (G) = M+j (G) ⊕ M0j (G) ⊕ M−j (G) and the same for M j (G P ). Thus
the reduction map modulo P induces a sujection M∗j (G) → M∗j (G P ) for
∗ = ±, 0.
Let 1 L (resp. 1 R ) be the idempotent of the left (resp. the right) factor I of A.
Since G P is Z p -full, for M j (G P ), 1 L M∗j (G P ) = 0 and 1 R M∗j (G P ) = 0
for ∗ = ±, 0. Take g ∈ Im(ρI ) giving the Z p -regularity, Ad(g) acts on
1 L M±j (G P ) ∩ M±j (G P ) and 1 R M±j (G P ) ∩ M±j (G P ) by the different set
of eigenvalues. Thus 1? M±j (G P ) ∩ M±j (G P ) = 0 for ? = L , R. We
have eigenspace decompositions under the action of Ad(g): M±j (G P ) =
(1 L M±j (G P ) ∩ M±j (G P )) ⊕ (1 L M±j (G P ) ∩ M±j (G P )) and M±j (G) =
(1 L M±j (G) ∩ M±j (G)) ⊕ (1 L M±j (G) ∩ M±j (G)). This implies 1 L M±j (G) ∩
M±j (G) = 0 and 1 R M±j (G) ∩ M±j (G) = 0 as the modulo P reduction map of
1? M±j (G) ∩ M±j (G) to 1? M±j (G P ) ∩ M±j (G P ) is surjective for ? = L , R.
This implies that (1? M±j (G) ∩ M±j (G)) = 1? M±j (G) for ? = L , R and
for all j > 0. After replacing ρI by a conjugate under G(Q(I)), by [H13b,
SL(2) SL(2)
Lemma 2.9], that Im(ρ L ) ⊃  (c L ) and Im(ρ R ) ⊃  (c R ) for non-zero
 Big image of Galois representations and congruence ideals 233

SL(2)
-ideals c L and c R . In particular, by Lemma 9.6, we get Im(ρI ) ⊃ A (c)
for c = c2L ∩ c2R as desired.

9.3.7 Z p [[S1 ]] ⊕ Z p [[S2 ]]-fullness in Case (Y)

We can think of fullness property in a two-variable setting. Case (F) will be
treated in the following section in the general setting of GSpin representations
as GSpin5 ∼ = GSp(4). In the cases other than (F) and (Y), the abelian part
consumes one variable out of the two, we do not have a properly two-variable
Galois representation ρI . Thus we assume that there exists a finite extension
k/Q such that ρ := ρI |Gal(Q/k) has values in
G 2 = {(x, y) ∈ GL(2) × GL(2)| det(x) = det(y)}.
We suppose that G 2 is embedded into GSp(4) in the following way:
⎛ ⎞
a 0 0 b
   
a b α β ⎜0 α β 0 ⎟
G2 , ( , )→⎜ ⎟
⎝0 γ δ 0 ⎠ ∈ GSp(4).
c d γ δ
c 0 0 d
As before, we suppose that ρ(I pw ) for the wild p-inertia group I pw contains the
following split torus
1 2
T = diag[t1s t2s , t1s , t2s , 1] ∈ T |s ∈ Z p
for t j = 1 + T j and 2 = Z p [[T1 , Ts ]]. We make the following variable
change s1 = t1 t2 and s2 = t1 /t2 ; so, Z p [[S1 , S2 ]] = 2 for S j = s j − 1. We
! Z p A2 .
put A j = Z p [[S j ]]; so, 2 = Z p [[S1 , S2 ]] = A1 ⊗

Proposition 9.9. Let ρI is associated to I-adic ordinary Hecke eigenform on

GSp(4) for an algebra I finite torsion-free over 2 . Suppose Z p -regularity for
ρI and that for an arithmetic point P ∈ Spec (2 )(Q p ) and for all primes
P ∈ Spec (I)(Q p ) above P, ρP = ρ mod P belongs to Case (Y) in Sec-
tion 9.3.4. Then, if p ≥ 3, replacing ρ by a conjugate of ρ by an element in
GL2 (Q(A1 )) × GL2 (Q( A2 )) if necessary, there exist non-zero A j -ideals c j
such that Im(ρ) contains
SL(2) SL(2)
A1 (c1 ) × A2 (c2 ) ⊂ SL2 (A1 ) × SL2 ( A2 ) ⊂ G 2 (2 ).

Proof. Let {i, j} = {1, 2}. We can write P = Pi ⊗ P j for P? ∈ Spec (A? )(Q)
for ? = i, j. Consider the projection Pi × id : 2 := Ai ⊗ !Z p A j →

Wi ⊗Z p A j = Wi [[S j ]] for Wi = Ai /Pi . Take a prime Pi ∈ Spec (I) above
234 Haruzo Hida and Jacques Tilouine

Ker(Pi × id). Let ρ j = ρ mod P  i . Then by Theorem 9.8 (or more precisely,
by the same argument proving the theorem), replacing ρ j by its conjugate
SL(2)
under an element of GL2 (Q(A j )), Im(ρ j ) contains A j (a j ) for a non-zero
A j -ideal a j .
Sp(4)
Put G = Im(ρ) ∩ 2 (2 ), and consider Pink’s Lie algebra M2 (G) with
respect to A = A1 ⊕ A2 regarding SL2 (A1 )×SL2 (A2 ) ⊂ G 2 (2 ) as SL2 (A1 ⊕
A2 ). Write u+j for the upper nilpotent Lie algebra of the j -th component of
SL(2) × SL(2), and let u−j be the opposite algebra of u+j . By Z p -regularity of
ρ, we have σ ∈ Gal(Q/K ) normalizing G whose adjoint action Ad(ρ(σ ) has
distinct Z p -eigenvalues on u±j for j = 1, 2.
±
:= M2 (G) ∩ u±j is non-trivial. By the
SL(2)
Since Im(ρ j ) contains A j (a j ), n j
adjoint action of T , n±j is a A j -module. Then by [H13b, Lemma 2.9] applied to
each projection of ρ to GL(2), we conclude a suitable conjugate of ρ has image
containing the product of congruence subgroups as in the proposition.

9.4 Fullness of the Galois image in general Spin groups over

Iwasawa algebras
9.4.1 Spin groups
Let us first recall the definition and the basic properties of the Z[ 12 ]-group
scheme GSpin2n+1 of spinorial similitudes. Consider the quadratic form
2n+1
Q(x) = 2x1 x2n+1 +2x2 x2n +. . .+2xn−1 xn+1 +xn2 on L = i=1 Z[ 12 ]·ei . Let
C = C(L , Q) be the Clifford algebra associated to (L , Q) and C = C + ⊕ C −
be its decomposition as Z/2Z-graded algebra into even and odd Clifford
elements; C + is a subalgebra. The main (anti-)involution ∗ sends a pure ele-
ment v1 · . . . · vr to vr · . . . · v1 . It leaves C + stable; the module C, resp.
C + , is locally free of rank 22n+1 , resp. 22n ; L embeds in C − by v → v.
Then GSpin2n+1 = {x ∈ (C + )× ; x · L · x ∗ = L} and Spin2n+1 = {x ∈
(C + )× ; x · L · x ∗ = L , x ∗ x = 1}. We have two exact sequences of group
schemes
μ
1 → Spin2n+1 → GSpin2n+1 → Gm → 1
by μ : x → x ∗ x, and
1 → Gm → GSpin2n+1 → SO(L , Q) → 1
by x → (v → xvx −1 ). Since 2 is invertible, we can write L =
W ⊕ W ∨ ⊕ U where W , resp. W ∨ , is the standard lagrangian generated
by e1 , . . . , en , resp. its dual, identified to the direct factor generated by
 Big image of Galois representations and congruence ideals 235

en+2 , . . . , e2n+1 , and U = Zen+1 . There is a natural isomorphism C + ∼ =

8
End( • W ) (Fulton-Harris, Lemma 20.16, the proof works with C replaced
by Z[ 12 ]); by restriction to GSpin, it induces the spin representation
8
spin : GSpin → GL( • W ) = GL2n . Let Tspin be the diagonal maximal
torus of GSpin2n+1 , and put Tspin  = Tspin ∩ Spin2n+1 . We denote by
(t1 , . . . , tn ; μ) → [t1 , . . . , tn ; μ] the standard isomorphism Gnm × Gm ∼ = Tspin ,
with μ([t1 , . . . , tn ; μ]) = μ. In particular (t1 , . . . , tn ) → [t1 , . . . , tn ; 1] is an
isomorphism Gnm ∼  . We define the upper, resp. lower, triangular Borel
= Tspin

subgroup Bspin = Tspin · Uspin resp. t Bspin = Tspin  · t Uspin , of Spin2n+1 as
spin−1 (B2n ),resp. spin−1 (t B2n ), where B2n = T2n · U2n denotes the upper
triangular Borel subgroup of GL2n .
The triple (GSpin2n+1 , Bspin , Tspin ) identifies to the dual (G, ! ! B, T!) of
the triple (G, B, T ) (see [MT02, Sect. 3.2]). In particular, the parametriza-
tion of (the semi-simple part of) the standard torus Tspin  = T ! of

! = GSpin2n+1 by [t1 , . . . , tn ; 1] can be written as
i [ei − 2 f ](ti ) via
1
G
T = Gm ⊗ X (T ), where ei : (x1 , . . . , xn , νxn , . . . , νx −1 ) → xi and
! ∗ −1
1
f : (x1 , . . . , xn , νxn−1 , . . . , νx1−1 ) → ν.

9.4.2 Genus n Hida families

n(n+1)
Let n ≥ 1 and d = 2 . Let H be the genus n Siegel half space; it is a
d-dimensional hermitian symmetric domain. Let G = GSp2n with its stan-
dard Borel subgroup B and its diagonal torus T as before. For any torsion-free
compact open subgroup K ⊂ G(A∞ ), we form the d-dimensional Shimura
Siegel variety X K = G(Q)\(H × G(A∞ )/K ). These are non-connected d-
dimensional complex varieties; the inclusions K  ⊂ K induce a projective
system of finite coverings π K  ,K : X K  → X K . These varieties and morphisms
are algebraic and admit quasi-projective canonical models over Q which we
still denote by X K and π K  ,K . Let λ ∈ X ∗ (T ) be a dominant weight for
(G, B, T ) and Vλ the irreducible representation of G over Q associated to λ.
It defines a projective system of étale local systems Vλ,K (Q p ) over the X K ’s:
π K∗  ,K Vλ,K (Q p ) = Vλ,K  (Q p ). Let M ≥ 3 and K (M) be the level M principal
congruence subgroup of G(! Z). Let us fix an odd prime p not dividing M and
+
consider the subgroups Im ⊂ Im of K p = G(Z p ) where Im = {g ∈ K p ; g
(mod p m ) ∈ B(Z/ p m Z)} and Im+ = {g ∈ K p ; g (mod p m ) ∈ N (Z/ p m Z)}
denotes its (normal) pro- p-Sylow. Let K m = K (M) p × Im+ . Consider the
middle degree cohomology

H = lim H d (X K m × Q, Vλ,K m (Q p ))
m
236 Haruzo Hida and Jacques Tilouine

Note that any Z p -model Vλ of the representation of G defines a compatible

system of integral étale sheaves Vλ,K m (Z p )), hence an integral structure HZ p
of H: H = HZ p ⊗Q p ; note that HZ p may have torsion. Let  = (1, . . . , 1; 0) ∈
X ∗ (T ); we put w = λ + ρ, ; it is the motivic weight of these cohomology
groups.
For any prime  not dividing M, let d,1 = diag(1n ,  · 1n ) and for any
i = 2 . . . , n, let d,i = diag(1n+1−i ,  · 12n−i , 2 · 1n+1−i ). Consider the
−1 −1
groups K m = d,i K m d,i and K m = d,i K m d,i , we have an isomorphism
π(d,i ) : X K m → X K m . We define an action of the double class K m d,i K m
on the pair (X K m , Vλ,K m (Q p )) by using the two finite morphisms π K m ,K m ,
π K m ,K m and the morphism of sheaves
∗
(π(d,i ), d,i ) : Vλ,K m (Q p ) → Vλ,K m (Q p )
where for t ∈ T (Q p ), we write t ∗ = ν(t) · t −1 ∈ T (Q p ); and we consider its
action on Vλ (Q p ). Note that for any  = p, d,i ∗ = d ∗ = 1 on V (Q ), in
,i λ p
∗
particular, it preserves any integral structure of Vλ . For  = p, we see that d,i
defines an isomorphism of Vλ (Q p ), that it preserves Vλ (Z p ) = Ind B − λ, but
max G

has non-trivial cokernel; we won’t make use of this integral structure in the
sequel. We define for any prime  not dividing M, and any i = 1, . . . , n:
∗
[K m d,i K m ] = π K m ,K m ,∗ ◦ (π d,i , d,i ))∗ ◦ π K∗  ,K m .
m

These actions are compatible when m varies and define Q p -linear endomor-
phisms on H, denoted by T,i for  prime to M p, resp. by U p,i for  = p. These
operators are the Hecke correspondences acting on H. The space H carries a
Galois action which commutes to the T,i s and U p,i s. These actions preserve
the integral structures.
It carries also a normal action of the torus T (Z p ) viewed as the quotient
B(Z p )/N (Z p ) = lim Im /Im+ . This action is continuous. One can decompose
←−
T (Z p ) as T f × T1 where T f is the torsion subgroup and T1 is the pro- p-Sylow,
kernel of the reduction modulo p. We fix a topological basis (u 1 , . . . , u n )
of T1 by fixing u = 1 + p as topological basis of 1 + pZ p and defin-
ing u i as the image of u by the cocharacter t → diag(E i (t), E n+1−i (t −1 ))
where E i (t) = diag(1, . . . , t, . . . , 1) (t is the i -th component). By this choice,
we identify the completed group algebra Z p [[T1 ]] to the n-variable Iwa-
sawa algebra n = Z p [[X 1 , . . . , X n ]] via u i → X i + 1. H is therefore a
n [T f ]-module. This action commutes to the T,i and the U p,i s; moreover
it preserves any integral structure given by a Z p -model Vλ of the represen-
tation of G. We consider the largest subspace H0d (X K m × Q, Vλ,K m (Q p )),
resp. H0 , of H d (X K m × Q, Vλ,K m (Q p )), resp. of H, on which the eigenval-
ues of the operators U p,i are p-adic units; it is the sum of all the generalized
 Big image of Galois representations and congruence ideals 237

eigenspaces for such eigenvalues. We denote by TλM (K m ) the Z p -algebra gen-

erated by the T,i s, the U p,i s and the group action of T (Z/ p m Z) on H0d (X K m ×
Q, Vλ,K m (Q p )); similarly, let T M be the n -subalgebra of Endn H0 gener-
ated by the T,i s, the U p,i s and the action of T f . Recall ([M04] Prop. 6.4.1)
that H0 is a finite n -module. Therefore, T M is a finite n -algebra. Note how-
ever that it is not cuspidal, in general (this would mean that all specializations
at cohomological weights would be cuspidal). It is conjectured that cuspidal-
ity holds for the localization of T M at a “weakly non-Eisenstein” maximal
!
ideal (assuming the existence of the G-valued p-adic Galois representation and
its ordinarity at p, “weak non-Eisenstein-ness” means that the corresponding
residual Galois representation is absolutely irreducible).
However, another obstacle in this definition (due to the presence of tor-
sion in the cohomology groups) is that the exact control theorem has not
been established for all cohomological weights, unless one makes some strong
extra assumptions. This seems to prevent to show in full generality that T M is
torsion-free as a n -algebra. Let λ ∈ X ∗ (T ) be a dominant weight such that
λ |T f = λ|T f , its restriction to T1 defines by linearity and continuity an algebra
homomorphism λ : n → Z p ; we denote by Pλ the prime ideal of n kernel
of λ .
Let C > 0 be a real number. The set of C-regular weights is the set of λ ’s
such that λ , α > C for all simple roots α of G.

Definition 9.10. We say that control theorem, resp. m-control theorem, holds
for the weight λ (congruent to λ mod. p − 1), if the surjective morphism
T M ⊗n ,λ Z p → TλM (K 1 ), resp. Tm
M ⊗
n ,λ Z p → Tλ (K 1 )m , has finite
M

kernel.

If one assumes that p −1 > w (for the motivic weight w) and if the maximal
ideal m is “strongly non-Eisenstein” (in the sense of [MT02] Sect. 9: it means
that the image of the residual representation contains the normalizer of the
F p -points of the standard maximal torus T !) then cuspidality and n -torsion-
freeness hold for the m-localized Hecke algebra and m-control theorem do
hold for all cohomological weights λ congruent to λ mod. p − 1 (see Th.10 of
[MT02]), otherwise, without assuming the existence of the Galois representa-
tions and without localization, there exists a constant C depending on the data
(λ, p) such that control theorem holds for all C-regular weights (but it does
not imply the torsion-freeness of the Hecke algebra, except for Siegel-Hilbert
varieties of genus 2, see [TU99], or for unitary groups in three variables, see
[M04]).
238 Haruzo Hida and Jacques Tilouine

In order to obtain a control theorem for all weights without localization, one
needs to follow another approach via coherent cohomology, started by Hida
[H02]. We identify the character group X ∗ (T ) to the index two sublattice of
Zn+1 by sending the character
λ : diag(t1 , . . . , tn , νtn−1 , . . . , νt1−1 ) → t1k1 . . . tnkn ν (c−k1 −...−kn )/2
to (k1 , . . . , kn ; c). Let t = (1, . . . , 1; 1); we fix the cohomological weight
k = λ + (n + 1)t = (k1 , . . . , kn ; k1 + · · · + kn ) and the sheaf ωcusp k

over X K 1 (see [H02, Sect. 2.3] and [Pi12, Sect. 4]). These objects are now
viewed on the rigid analytic site over Q p . One starts from a holomorphic
cusp form f ∈ H 0 (X K 1 , ωcusp k ); instead of the whole tower of finite cover-

ings X K m → X K 1 , one forms the Igusa tower Tm = Isom X 0 (μnpm , A[ p m ]◦ )

K1
over the ordinary locus X 0K 1 of X K 1 (here, A[ p m ]◦ denotes the connected
component of the pm -torsion of the universal abelian variety over X 0K 1 ); Tm
can be viewed as the rigid tube of the “multiplicative type” connected com-
ponent of the ordinary locus in the reduction modulo p of X K m . Consider
the Q p -vector space HC = limm H 0 (Tm , ωcusp k ); one defines similarly action

of the Hecke operators T,i s and U p,i s (i = 1, . . . , n); these are compati-
ble with the transition morphisms of the Igusa tower; the Galois group of the
Igusa tower is T (Z p ) which therefore acts on H 0 (Tm , ωcusp k ) and HC. One

defines the Z p -Hecke algebras T M (Tm ), resp. T M (T∞ ) as generated by T,i s,

U p,i s and the image of T (Z p ) acting on H 0 (Tm , ωk ), resp. HC. Similarly, one
holcusp
could define T M,holcusp as the Z p -algebra acting on the ordinary part H0
of Hholcusp = limm H 0 (X K m , ωcusp
k ); by density of classical Siegel modular

cusp forms in p-adic modular cusp forms, the natural morphism T M (T∞ ) →
T M,holcusp is an isomorphism of n -algebras (note though that one has only
surjections T M (Tm ) → T M,holcusp (K m )). By Hodge–Tate decomposition (see
Faltings–Chai Chapter VI, using BGG theory), T M,holcusp is a quotient of T M
(it cannot be injective since the left-hand side contains non-cuspidal families,
but after localization at any weakly non-Eisenstein maximal ideal should be an
isomorphism).
By [Pi12, Th.6.7 and Th.1.2], and [PiSt, Th.6.3.1], T M,holcusp is finite
torsion-free over n and (cuspidal) Control Theorem holds without localiza-
tion (and without assuming the existence of the Galois representations); in
other words, for any dominant weight λ such that λ |T f = λ|T f , the surjective
M,holcusp
morphism T M,holcusp ⊗n ,λ Z p → Tλ (K 1 ) has finite kernel.
Let I be the normal closure of an irreducible component of Spec (T M,holcusp ).
Let θ : T M,holcusp → I be the corresponding Hida family. Assuming the
existence of the Galois representations ρ f : G Q → G(Z ! p ) associated to the
 Big image of Galois representations and congruence ideals 239

holomorphic cusp forms of level K 1 occurring in the family, let us construct

the Galois representation G Q → G(I) ! associated to θ . We say that a cusp
eigenform f of level K 1 , weight k  = λ + (n + 1)t and eigensystem θ f 


occurs in the family if there exists a prime P  of I above Pλ such that, via the
control theorem, one has θ f  = θ (mod P  ).

Definition 9.11. We say that the irreducible component I is weakly non-

Eisenstein if for some (hence all) cusp eigenform f  occurring in the family,
the residual representation composed with spin is absolutely irreducible.

Lemma 9.12. For any weakly non-Eisenstein irreducible component I, there

exists a continuous homomorphism ρθ : G Q → G(I) ! such that for any domi-
nant weight λ such that λ |T f = λ|T f , and for any cusp eigenform f  of weight
 

! p)
k  = λ + (n + 1)t in the family, the representation ρ f  is conjugate in G(Q
to ρθ (mod P  ).

Proof. We give the proof only for I = n . Let us consider the Z p -morphism
Ad : G ! → SO2n+1 ; note that Ker(Ad) is the center ! Z ∼ ! We first
= Gm of G.
construct a representation R : G Q → SO2n+1 (I) and then lift it to G(I). ! For
any arithmetic prime P of I above a dominant weight λ such that λ |T f =
 

λ|T f , let f P  be a cusp eigenform occurring in the family with this weight;
by residual irreducibility of spin ◦ρ f P  and Carayol’s theorem [Car94], we can
assume that this representation is defined over I/P  ; by injectivity of spin, it
!
implies that ρ P  = ρ f P  takes values in G(I/P P Q !
 ): ρ  : G → G(I/P  ).

On the other hand, by residual irreducibility, it follows from Carayol’s the-

orem [Car94] that Ad ◦ ρ lifts to a continuous representation R : G Q →
GL2n+1 (I). Since Ad ◦ ρ is orthogonal, and each specialization modulo an
arithmetic prime is orthogonal, we see that R is orthogonal. Since I is profi-
nite and by formal smoothness of the group G, ! there exists a continuous lift
: G Q → G(I),
R ! which gives rise to a 2-cocyle cσ,τ ∈ Z 2 (G Q , I× ).
For any prime P  , the existence of the representation ρ P  implies that the
image of cσ,τ is a coboundary: cσ,τ = bσ τ bσ−1 bτ−1 over (I/P  )× . Take I = n
n n
and consider the Iwasawa ideal n,k generated by the (1 + Ti ) p − u ki p ,
= 1, . . . , n. The quotient ring n,k = /n,k embeds in the product of
cyclotomic rings (ζ1 , . . . , ζn ) = n /I (ζ1 , . . . , ζn ) where

I (ζ1 , . . . , ζn ) = (1 + Ti − u ki ζi , i = 1 . . . , n)

where ζi is a pn -th root of unity. The condition for a family

(b(ζ1 , . . . , ζn ))ζ1 ,...,ζn to be in the image of the embedding is that for any pair
240 Haruzo Hida and Jacques Tilouine

(ζ1 , . . . , ζn ) and (ζ1 , . . . , ζn ) the components b(ζ1 , . . . , ζn ) and b(ζ1 , . . . , ζn )
are congruent modulo I( ζ1 , . . . , ζn ) + I (ζ1 , . . . , ζn ). Let us check it is the
case for the family of 1-cochains (b(ζ1 , . . . , ζn )σ ) associated to the specializa-
tion cσ,τ of cσ,τ over (n,k )× . Indeed, let us compare the reductions modulo
n,k

 )b(ζ1 , . . . , ζn )σ
I (ζ1 , . . . , ζn ) + I (ζ1 , . . . , ζn ) of the representations spin ◦ R(σ
and spin ◦ R(σ  )b(ζ  , . . . , ζn )σ ; they have the same characteristic polynomial
1
and they both lift the irreducible representation spin ◦ρ. By Carayol’s theo-
rem [Car94], they are conjugate, so that b(ζ1 , . . . , ζn )σ = b(ζ1 , . . . , ζn )σ in
n /I (ζ1 , . . . , ζn ) + I (ζ1 , . . . , ζn ).
n,k
From this, we conclude that there exists a 1-cochain bσ with values in
(n,k )× such that cσ,τ = bσ τ bσ −1 bτ −1 . This construction is compatible
n,k n,k n,k n,k

when n grows and gives rise to a 1-cochain bσ with values in × n as desired.

The resulting map σ → R(σ  )bσ is the desired representation G Q → G( ! n ).

Moreover ρθ is ordinary at p. More precisely, recall that  denotes the p-adic

cyclotomic character; we define for any σ ∈ Gal(Q/Q) the cyclotomic loga-
log ((σ ))
rithm (σ ) = log p (1+ p) (for the p-adic cyclotomic character ). Let ti = 1+ X i
p
(i = 1, . . . , n), then

Lemma 9.13. Assume p − 1 > w. The representation ρθ is ordinary: there

exists ! ! such that ρθ (D p ) ⊂ !
g ∈ G(I) g! g −1 and for any σ ∈ I p ,
B(I)!
(σ ) (σ )
g −1ρθ (σ )!
! g ≡ [(σ )−1t1 , (σ )−2t2 . . . , (σ )−n tn(σ ) ; (σ )−d ] (mod U (I))
n(n+1)
where d = 2 .

Proof. Since all Borel subgroups are conjugated in G, ! it is enough to apply

n
the spin representation and to find a D p -stable flag in I2 with action of I p on
 (σ )
the graded pieces given by the mutually distinct characters i∈I (σ )−i ti

for all subset I s of {1, . . . , n}. Over J = P  O P  as in the previous proof, we
n
have such a flag (in J2 ). If we assume p − 1 > w so that all the characters
above take values in I and are mutually distinct modulo p, we see that the
n
intersection of this flag with I2 induces a flag as desired.

9.4.3 n -fullness
In the sequel, we consider an irreducible component I of T M such that
cyc
I = n = Z p [[X 1 , . . . X n ]] (with p > 2). Let u = 1 + p. Let Q p be
the p-power-cyclotomic extension of Q p . For any weight k = (k1 , . . . , kn )
 Big image of Galois representations and congruence ideals 241

dominant for (G, B, T ), and cohomological, we denote by Pk the point of

Spec n given by Pk = (1 + X 1 − u k1 , . . . , 1 + X n − u kn ). the point
We define for any σ ∈ Gal(Q/Q) the cyclotomic logarithm (σ ) =
log p ((σ ))
log p (1+ p) (for the p-adic cyclotomic character ). We put, as above, ti = 1+ X i
(i = 1, . . . , n), and we assume that the semi-simplification of ρ restricted to
the p-inertia group I p is
(σ ) (σ )
[(σ )−1 t1 , (σ )−2 t2 . . . , (σ )−n tn(σ ) ; (σ )−d (t1 · . . . · tn )(σ ) ],

where d = n(n+1) 2 . This is conjecturally the case if ρ is associated to an

n-variable p-adic Hida family of Siegel forms of genus n, passing by a form f
of level prime to p and weight k 0 = (k10 , . . . , kn0 ), with k10 ≥ . . . ≥ kn0 ≥ n + 1.
It is actually proven if n = 2.
We consider the Galois representation ρ = ρθ : Gal(Q/Q) →
GSpin2n+1 (n ) associated to the family θ : T M,holcusp → I = n ; the rep-
resentation is ρ ordinary at p; After conjugation, we can and will assume in
the sequel that ρ(D p ) ⊂ ! B(n ). For any root α of (G, ! T
!), we denote by
Uα the one-dimensional Lie subgroup of Spin2n+1 , by u α : Ga → Uα the
corresponding one-parameter subgroup and by uα = Lie(Uα ) ⊂ spin2n+1
the corresponding Lie subalgebra (say over Z p or any base change thereof).
Because of these notations, it will be convenient to write U instead of N ! for
the unipotent radical of the standard Borel subgroup ! B = T !N !. We have the
following key lemma

Lemma 9.14. Let ρ : Gal(Q/Q) → GSpin2n+1 (n ) be the Galois represen-

tation associated to a family θ : T M,holcusp → I = n . Suppose

(1) there is a cohomological weight k 0 for which the specialization ρ Pk 0 = ρ f

has Z p -full image; in other words, Im(ρ Pk 0 ) contains an open subgroup of
Spin2n+1 (Z p ),
(2) strong regularity, (compare to 9.3.5) Im(ρ) contains an element δ ∈ T !(n )

such that for any two roots α = α of SO2n+1 , α(Ad(δ)) ≡ α (Ad(δ)) 

(mod mn ),
! T
Then for any root α of (G, !), Im ρ ∩ Uα (n ) = 0

Remarks. 1) Note that the roots of (G ! = GSpin2n+1 , T !) are the same as

those of (SO2n+1 , TSO2n+1 ).
2) If instead of Assumption 2 we only assume that there exists an element
δ ∈ ! !(n ) is strongly regular mod.
B(n ) such that its projection δ1 ∈ T
mn , by which we mean that for any two roots α = α  of SO2n+1 ,
242 Haruzo Hida and Jacques Tilouine

α(Ad(δ1 )) ≡ α  (Ad(δ1 )) (mod mn ), then we can modify it by conjuga-

tions by elements u α (x) for positive roots α roots and suitable elements
x so that it belongs to T !(n ) and it satisfies the exact statement of
Assumption 2.
3) If we assume furthermore that p − 1 > 2w and k 0 = (k10 , . . . , kn0 ) is such
that for any i = j , ki0 − i ≡ ±(k 0j − j) (mod p − 1), we see by Lemma
4.4 above that if we choose σ ∈ I p such that κ(σ ) = ω(σ ) is of exact
order p − 1, we can take δ = ρ(σ ).

Proof. Since the argument is the same for positive and negative roots, we
restrict ourselves to the “upper triangular” unipotent radical U (correspond-
ing to positive roots α) of Spin(2n + 1). Let K = Im ρ ⊂ G( ! n ). We
write P = Pk 0 for the arithmetic point such that the specialization ρ P of the
representation ρ is Z p -full. Fix a positive root α and define
j
Uα (P ) = {x ∈ Spin2n+1 (n )|x mod P j ∈ Uα (n /P j )}
Bα (P
j
) = {x ∈ Spin2n+1 (n )|x !(n /P j )Uα (n /P j )}
mod P j ∈ T
(9.2)
and
K Uα (P j ) = K ∩ Uα (P
j
)
(9.3)
K Bα (P ) = K ∩
j
Bα (P
j
)
Since U (n ) and n (P) are p-profinite, the groups Uα (P
j) and K Uα (P j )
for all j ≥ 1 are also p-profinite. Recall

n (P ) = {x ∈ Spin2n+1 (n )|(x mod P i ) = 1}

i

is the kernel of the reduction morphism πj : Spin2n+1(n ) → Spin2n+1(n/P j ).

Recall the embedding ια : SL(2) → Spin2n+1 associated to the root α (i.e. the
+ −
upper (resp. lower) unipotent subgroup USL(2) (resp. USL(2) ) of SL(2) is sent
to Uα (resp. U−α ) by ια ). Thus, we see

Uα (P )= n (P )ια (USL(2) (n )).

j j

Note that
     
a b e f bg − c f 2(a f − be)
, = .
c −a g −e 2(ce − ag) c f − bg
From this, we have
 
Lemma 9.15. If X, Y ∈ sl2 (n ) ∩ P j Pk with i ≥ j ≥ k, [X, Y ] ∈
  Pi P j
P i+k P j+k .
P i+ jP i+k
 Big image of Galois representations and congruence ideals 243

Uα (P
This tells us, for the topological commutator subgroup D j) :=
( Uα (P j ), Uα (P j )) that
j 2j j
D Uα (P )⊂ Bα (P )∩ Uα (P ). (9.4)

Using now the Z p -regularity assumption (in Sect. 9.3.5), we consider an ele-
ment δ ∈ T !(n ) ∈ K , with distinct root values modulo mn . Replacing δ by
m
limm→∞ δ p , we may assume that δ ∈ T !(Z p ) has finite order a. The order a
is prime to p (indeed it is a factor of p − 1). Note that δ normalizes K Uα (P j )
and Bα (P j ). Since H is p-profinite, x ∈ H has unique a-th root inside H .
We define

−1 −2
α (x) = x · (δxδ −1 )α(δ) · (δ 2 xδ −2 )α(δ) · · · (δa−1 xδ 1−a )α(δ) ∈ H.
a 1−a

Lemma 9.16. If u ∈ Uα (P
j) ( j ≥ 1), then 2α (u) ∈ Uα (P
2j) and
π j (α (u)) = π j (u).

Proof. If u ∈ Uα (P j ), we have π j (α (u)) = π j (u) as α is the iden-

tity map on Uα (n /P j ). Let D Uα (P j ) be the topological commutator
subgroup of Uα (P j ). Since α induces the projection of the Z p -module
Uα (P )/D Uα (P ) onto its α(δ)-eigenspace for Ad(δ), it is a projec-
j j

tion onto Uα (n )D Uα (P j )/D Uα (P j ). The fact that this is exactly the
α(δ)-eigenspace comes from the Iwahori decomposition of Uα (P j ) which
shows it is generated by the Uβ (P j n ) for all roots β = α and by
Uα (n ), hence a similar direct sum decomposition holds in the abelianization
Uα (P )/D Uα (P ).
j j

By (9.4) D Uα (P j ) ⊂ Bα (P 2 j ) ∩ Uα (P j ). Since the α(δ)-eigenspace of

Uα (P )/D Uα (P ) is inside Bα (P ), α projects u Uα (P ) to
j j 2j j

α (u) ∈ ( Bα (P
2j
)∩ Uα (P
j
))/D Uα (P
j
).

In particular, α (u) ∈ Bα (P ) ∩ Uα (P ). Again apply α . Since

2j j

Bα (P )/ n (P ) is sent to to Uα (P )/ n (P ) by α , we get
2j 2j 2j 2j

α (u) ∈ Uα (P ) as desired.
2 2 j

We can now prove the key Lemma 9.14. Pick 0 = u α ∈ Uα (A/P)∩Im(ρ P ).

By Z p -fullness of ρ P , u α exists. Since the reduction map Im(ρ) → Im(ρ1 )
is surjective, we have u α ∈ Im(ρ) such that u α mod P = u α . Take v1 ∈
Uα (n ) such that v1 mod P = u α . Such a v1 exists as the reduction map
Uα (n ) → Uα (n /P) is onto. Then u α v1−1 ∈ n (P), we find that u α ∈
K Uα (P).
By the compactness of K Uα (P), by Lemma 9.16, starting with u α as above,
limm→∞ m ∞
α (u α ) converges P-adically to α (u α ) ∈ Uα (n ) ∩ Im(ρ) with
244 Haruzo Hida and Jacques Tilouine

∞ (P) (i.e., u α = (∞

Spin(2n+1) Spin(2n+1)
α (u α ) n (P) = u α n α (u α ) mod P)).
∞
Thus α (u α ) is a non-trivial unipotent element in Uα (n ) ∩ Im(ρ).

From this lemma we can deduce the following main theorem (we still
assume that ρ(D p ) ⊂ !
B(n )).
Theorem 9.17. Let ρ : Gal(Q/Q) → GSpin2n+1 (n ) be a continuous
representation. Suppose

(1) there is a cohomological weight k 0 for which the specialization ρ Pk 0 = ρ f

has Z p -full image; in other words, Im(ρ Pk 0 ) contains an open subgroup of
Spin2n+1 (Z p ),
(2) Im(ρ) contains an element δ ∈ ! B(n ), whose projection δ1 ∈ T !(n ) is
spin2n+1 -regular mod. mn , that means that for any two roots α = α  of
SO2n+1 , α(Ad(δ1 )) ≡ α  (Ad(δ1 )) (mod mn ),
Then Im ρ contains a principal congruence subgroup of Spin2n+1 (n )
Definition 9.18. The largest ideal of n such that n (lθ ) ⊂ Im ρθ is called
the Galois level of θ , and is denoted by lθ .
There are two proofs. It can either be deduced from Pink’s theory using
Lemma 9.6, as the only missing point was the non-triviality of Uα (n )∩Im(ρ)
for all αs (which follows from Lemma 9.14), or it can be proven using Lie
algebras as follows.
Let H = Im ρ∩ n ( p). We have seen that for any root α, Im ρ∩Uα (n ) =
{1}; therefore, by taking possibly a p-th power of the non-trivial element in this
intersection, we can assume that for any root α, H ∩ Uα (n ) = {1}. The series
defining Log still provides a bijection between H and the set Log H which is
contained in p · spin2n+1 (n ), but may not be stable by addition; however,
the Log map induces group isomorphisms from H ∩ Uα (n ) to the (possibly
infinite dimensional) Z p -Lie algebras Log(H ) ∩ uα (), which are therefore
non-trivial Z p -modules for all roots α. Recall that the Q p · Log(H ) is a Q p -Lie
algebra, denoted LogQ p H below.
Using the adjoint action of δ, we have a decomposition into root spaces

(∗) LogQ p H = (LogQ p H )0 ⊕ (LogQ p H )α
α∈+

compatible with the root decomposition


!) ⊕
spin2n+1 = Lie(T uα
α

which is valid over n , hence also over n [ 1p ]

 Big image of Galois representations and congruence ideals 245


Let [t1 , . . . , tn ] = i [ei − 2 f ](ti ) denote the parametrization of (the
1

semi-simple part of) the standard torus T  = T ! of G ! = GSpin2n+1 via

!
T = Gm ⊗ X (T ), where ei : (x1 , . . . , xn , νxn , . . . , νx1−1 ) → xi and
∗ −1

f : (x1 , . . . , xn , νxn−1 , . . . , νx1−1 ) → ν. For any σ ∈ I p , we put

t (σ ) = [(1 + T1 )(σ ) , . . . , (1 + Tn )(σ ) ]

We know that for any σ ∈ I p ,

ρ  (σ ) ≡ t (σ ) (mod U (n ))

The simple roots of G ! (for the standard “upper triangular” Borel) are
αi : [t1 , . . . , tn ] → ti /ti +1 (i = 1, . . . , n − 1) and αn : [t1 , . . . , tn ] → tn .
The positive roots are then ti /t j for 1 ≤ i < j ≤ n, ti , i = 1, . . . , n, and ,
ti t j for 1 ≤ i < j ≤ n; they can be written, respectively, as αi + . . . + α j−1 ,
αi +. . .+αn , and αi +. . .+α j −1 + 2(α j +. . .+αn ). Conjugation by ρ(σ ) acts
on LogQ p Hα = LogQ p H ∩ uα (n ) by α(t (σ )); one can choose σ ∈ I p such
1+X i
that (σ ) = 1. We find that we can let 1+X j
, resp. 1+ X i , resp. (1+ X i )(1+ X j )
act on (LogQ p H )ti /t j resp. LogQ p Hti , resp. (LogQ p H )ti t j . We have an action
1+X
of 1 + X n on the non-zero subgroup (LogQ p H )αn and of 1+Xn−1
n
on the non-
zero subgroup (LogQ p H )αn−1 ; by forming the Poisson bracket of these two
submodules, we find a non-zero modules of (LogQ p H )αn−1 +αn on which both
1+X
(1+ X n ) and 1+Xn−1 n
act; therefore (1+ X n ) and 1+ X n−1 both act on this non-
zero subgroup of (LogQ p H )αn−1 +αn . By repeating this procedure, we find, for
α = α1 + . . . + αn , a non-zero submodule of (LogQ p H )α on which all 1 + X i s
act. This is a non-zero n -submodule of Uα (n ), contained in (Log H )α . We
then use successive Poisson brackets with (LogQ p H )−α1 ,. . . ,(LogQ p H )−α j
to construct inside all (LogQ p H )αi ’s a non-trivial n -submodule of uαi (n ).
Once we have non-zero n -submodules in (LogQ p H )αi for all the simple
roots, one can construct non-trivial n -submodules in (LogQ p H )α for all
the roots α. After multiplying by a power of p, we obtain non-trivial n -
submodules in (LogQ p H )α for all roots. Taking the exponential, we see that
there exists a non-zero ideal l of n such that Uα (l ) ⊂ Im ρ. Since these
subgroups generate a group containing a principal congruence subgroup of
Spin2n+1 (n ), we see that there exists a non-zero ideal l of n such that
n (l) ⊂ Im ρ as desired.

We give a lemma which follows from the proof above. For any simple root
αi of spin2n+1 , let uαi = uαi (n [ 1p ]) ∩ LieQ p Im ρθ . For any characteristic
zero quotient A of n , let ρ A be the specialization of ρθ to A and let uαAi be the
246 Haruzo Hida and Jacques Tilouine

image of uαi in spin2n+1 (A[ 1p ]). Let uαA,• j the sequence defined by uαA,1
•
= uαA1 ,
uαA,2
•
= [uαA1 , uαA2 ], and for j ≤ n − 1:

uαA,• j+1 = [uαA,• j , uαAj +1 ]

Note that by calculations in the proof of Theorem 9.17, uαA,n

•
is an A[ 1p ]-
module.

Lemma 9.19. For any n ≥ 1, assume ρθ is full over n . For any characteristic
zero quotient A of n , there exists a unique Lie subalgebra Lie A of the Q p -Lie
algebra LieQ p Im ρ A which is an A⊗Q p -Lie algebra, such that Lie A ∩u A
αi =i
uαA,n
•
and which is stable by the adjoint action of Im ρ A . If n → A → B are
two characteristic zero quotients of n , then Lie A → Lie B is surjective.

Note however that for some quotients A, this algebra may be zero because
uαA,n
•
may be zero, although it is not the case over n . This is why we’ll need a
modified construction later (see Lemma 9.24).
Proof. To simplify notations, we prove it for sp4 . Let α1 = t1 t2−1 and α2 = t22
be the two simple roots. We form uαA,2 •
= [uαA1 , uαA2 ] which is an A-module
by calculations at the end of the proof of Theorem 9.17. If it is non-zero, by
Poisson bracket of uαA,2
•
with all the uγA s for all the root (positive or nega-
tive) of sp4 , we construct A[ 1p ]-submodules uαA,
1 +α2 +γ
of uαA1 +α2 +γ . Note that
A,
in particular, uαA,1 +α2
= uαA1 +α2 . We also define a Cartan subalgebra t by
A, A,
[uαA,
1
, u−α 1
] ⊕ [uαA,
2
, u−α 2
], and we then define an A ⊗ Q p -module by
A,

Lie A = t ⊕ uγA,
γ

Using the Jacobi identity for the Poisson bracket, one sees easily that it is an
A ⊗ Q p -Lie algebra which has all the desired properties.
It follows from the construction above that if ρθ is full over n ; then
for all roots γ , uγ which is the γ (δ)-eigenspace for the action of δ as in
Theorem 9.17) on Lien , is non-zero. More precisely,

Corollary 9.20. If ρθ is full over n , there exists a non-zero largest n [ 1p ]-

ideal lθ such that
lθ · sp4 (n ⊗ Q p ) ⊂ Lien

We have lθ ⊂ lθ ⊗ Q p and these two ideals have the same radical: lθ =

lθ ⊗ Q p in n [ 1p ].
 Big image of Galois representations and congruence ideals 247

9.5 Galois level and congruence ideals

9.5.1 Congruence ideals
Let θ : T M,holcusp → I0 be an irreducible component of Spec T M,holcusp ; we
denote by I the normalization of I0 and by K its field of fractions; let TI =
T M,holcusp ⊗n I and θI = m I ◦ θ ⊗ IdI the composition of the extension of
scalars with the multiplication on I; by the diagonalisability of T M,holcusp over
K, the character θI : TI → I splits over K, and there is an isomorphism of
K-algebras:
TI ⊗I K ∼
= K × TK

with the first projection pr1 given by θK = θI ⊗ K.

(Mult 1) We assume that θI does not factor through pr2 . In other words, the
family θI occurs with multiplicity 1 in TI .
If n = 1, this assumption amounts to saying that the family θI is M-new (i.e.
all the forms occurring in the family are M-new); if n = 2, using the Roberts-
Schmidt theory of newforms, and assuming a folklore conjecture which says
that automorphic forms on GSp4 which are neither CAP nor endoscopic satisfy
multiplicity one, one can prove that it holds for M-new families of squarefree
level M. For n ≥ 3, we simply assume (Mult 1).
Let T be the image of TI by the second projection pr2 . We define then
c1 = TI ∩(I×{0}), which is an ideal of TI and of I×{0}, and c2 = TI ∩({0}×T )
which is an ideal of TI and of ×T × {0}. Note that c1 = Ker pr2 and c2 =
Ker pr1 , and that for i = 1, 2, one can identify ci with its image pri (ci ). We
also define c = c1 + c2 . The projections pri induce isomorphisms of I-modules

pr1 : TI /c → I/c1 and pr2 : TI /c → T /c2

We thus obtain the congruence isomorphism pr2 ◦ pr−1 ∼ 

1 : I/c1 = T /c2 .

Definition 9.21. The ideal cθ = c1 is called the congruence ideal associated to

the family θ .

For any prime P of I containing cθ , there exists another family θ  : TI → J

(for some finite normal extension J of I) and a prime Q of J above P such that
θI ∼
= θ  (mod Q).
Let θ2 : T M,holcusp → J be a second family, different from θ , hence it factors
through pr2 . Let cθ2 ⊂ J be its congruence ideal; if Q is a prime of J contain-
ing both the ideals cθ and cθ2 , and if P = I ∩ Q, the congruence isomorphism
induces a congruence injection I/P ∼ = J/Q which expresses a congruence
248 Haruzo Hida and Jacques Tilouine

between the two specific families θ and θ2 modulo Q. Therefore the intersec-
tion locus V (cθ + cθ2 ) in Spec T M,holcusp of the (normalizations of the) two
irreducible components Spec I and Spec J is also the congruence locus of the
two families θ and θ2 . Note that this locus is empty unless both families fac-
tor through the same local factor of the semilocal algebra T M,holcusp ; in other
words a necessary condition for this locus to be non empty is that the residual
representations ρ θ and ρ θ2 coincide.

9.5.2 Comparison between Galois level and congruence ideals

In this section we assume again that the family θ : T M,holcusp → I is n -
valued, i.e. I = n . We also assume that its residual Galois representation
ρ θ is irreducible and that ρ = ρθ satisfies the assumptions of Theorem 9.17:
there is an arithmetic specialization for which Im ρ Pk 0 is full and that there is a
Z p -regular element δ in Im ρ. By Theorem 9.17, the representation ρθ is full;
we denote by lθ the Galois level of θ , that is, the largest ideal of n such that
n (lθ ) ⊂ Im ρθ .
Let us assume that there exists another family θ2 in the same local com-
ponent of T M,holcusp as θ (so that its residual representation coincides with
that of θ, in particular it is irreducible) but such that ρθ2 is not full. Let
c(θ, θ2 ) = cθ ·J +cθ2 be the ideal of J defining the congruence locus between θ

and θ2 and let c(θ ) = θ2 c(θ, θ2 ) ∩ n , where the intersection of the c(θ, θ2 )s
is taken over all the families θ2 congruent to θ but not full. Let Q be a prime
ideal of J containing c(θ, θ2 ), and let P = Q∩n . Then, we have a congruence
ρθ ≡ ρθ2 (mod Q); then ρθ (mod P) is not full; hence we must have lθ ⊂ Q.
Since for any prime P of n containing c(θ ) there exists a prime Q above P

in a finite extension which contains θ2 c(θ, θ2 ), we see that any prime ideal
P containing c(θ ) contains lθ . For any ideal a of n , let V (a) be the set of
prime ideals containing a. Recall that in the case n = 1, Hida has proved in
[H13b] that if the family θ is not CM but there is another family θ2 which
is CM for a given imaginary quadratic field, then the sets V (lθ ) and V (c(θ ))
coincide; actually under slightly stronger assumptions (see [H13b, Th.7.2]), he
could use Pink’s Lie algebra theory for SL2 to compare the exponents of the
height one primes occurring in (the reflexive envelopes of) the two ideals lθ
and c(θ ), see [H13b, Th.8.5.].
For general n, a natural question is to ask whether the converse is true,
namely that if a prime P divides lθ , there exists a non full family θ2 con-
gruent to θ at a prime Q above P. Assume again n = 2, recall that given
a real quadratic field F of discriminant D, and a Hilbert modular cusp form
for F of level n and weight (m 1 , m σ ) with m 1 = m σ + 2r with r ≥ 1 and
 Big image of Galois representations and congruence ideals 249

m σ ≥ 2, eigen for the Hecke operators and p-ordinary, one can define its theta
lift to GSp4 (see [Y79] and [R01, Th.8.6]), which we call its twisted Yoshida
lift. It gives rise to an eigensystem of T M,holcusp with M = N (n)D 2 of weight
(k1 , k2 ) with k1 = m σ + r and k2 = r + 2, (so, k1 ≥ k2 ≥ 3), see for instance
[MT02, Sect. 7.3]. Assuming moreover that n is prime to p and that g is p-
ordinary, there exists a unique two-variable p-adic family of p-ordinary Hilbert
modular forms g of auxiliary level n passing through g (after p-stabilization).
It defines a homomorphism of 2 -algebras θg : TnF → J of the ordinary Hida
Hecke algebra TnF for Hilbert cusp forms over F, of auxiliary level n. For
any place w of F prime to n p, this homomorphism sends the Hecke operator
Tw to the eigenvalue θg (Tw ) on g. The map (m 1 , m σ ) → (k1 , k2 ) defines an
automorphism of 2 , denoted by τ .

Lemma 9.22. Given (F, g) of a real quadratic field and a two-variable family
of p-ordinary cusp forms θg : TnF → J, then, for M = N (n)D 2 , there exists a
two-variable p-adic family θ2 : T M,holcusp → Jτ , which is a homomorphism of
2 -algebras via τ : θ2 (α · T ) = τ (α) · θ2 (T ), such that for every prime q not
dividing M p,
θ2 (Tq,1 ) = λq (g) and θ2 (Tq,2 ) = μq (g)

where the Hecke eigenvalues λq (g) and μq (g) are defined as follows
● If q splits in F, let w1 and w2 the places above q, then λq (g) = θg (Tw1 ) +
θg (Tw2 ) and q · μq (g) = q 2 + qθg (Tw1 )θg (Tw2 ) − 1
● If q is inert, λq (g) = 0 and q · μq (g) = −(q 2 + qθg (Tw ) + 1)

We call θ2 = θ2 (F, g) the twisted Yoshida lift of (F, g)

The proof is by comparing for each prime q not dividing M p the character-
G
istic polynomial of IndG QF ρh for all classical Hilbert forms h in the family g
and the GSp4 -Hecke polynomial at q of its Yoshida lift.
From now on, we fix a real quadratic field F of discriminant D, an auxiliary
level n in F prime to p and we pose M = N (n)D 2 ; we give ourselves a 2 -
valued family θ : T M,holcusp → 2 which satisfies the assumptions of Sect.
9.4.3, so that its Galois representation ρθ : G Q → GSp4 (2 ) is full.
In the following theorem, we assume there exists a Hilbert modular form g
of weight m 1 = m σ + 2r with r ≥ 1 (hence the form g does not descend to Q)
and m σ ≥ 2, such that
● ρ g is “full”: that is, SL2 (F p ) ⊂ Im ρ g ,
×
● there exists no character ξ : G Q → F p such that ρ gσ ∼
= ρg ⊗ ξ .
250 Haruzo Hida and Jacques Tilouine

G
We finally assume that ρ θ = IndG QF ρ g . By the previous lemma, there exists
a “twisted Yoshida lift” family θ2 = θ2 (F, g) in the same local component of
T M,holcusp as θ . Note that if there is another quadratic field F  and a Hilbert
G
form g  on F  such that ρ θ = IndG Q ρ g  then F  = F. Indeed, if this equality
F
holds and F  = F, we have an isomorphism of G Q -modules ρ θ ⊗ ξ ∼ = ρθ
where ξ is the (non-trivial) quadratic character associated to F  ; hence, by
irreducibility of the G F -modules ρ g and ρ g σ , we have an isomorphism of G F -
modules ρ g ⊗ ξ ∼ = ρ g (which implies that ρ g ∼ = Ind FF F  ϕ for a character ϕ
∼
and is impossible by fullness of ρ g ) or ρ g ⊗ ξ = ρ g σ which is impossible by
assumption.

Theorem 9.23. For n = 2, assume that ρθ satisfies the assumptions of Theo-

rem 9.17, hence is full, and that there is a Hilbert cusp form g for F as above
G
such that ρ θ = IndG QF ρ g , then we have V (c(θ )[ 1p ]) = V (lθ [ 1p ]).

Note that we are not yet able to study the prime ( p); more precisely, we
can’t prove that if ( p) ∈ V (lθ ), then ( p) ∈ V (c(θ )). We need first a variant
of 9.19 and 9.20, because in our case 9.19 applied to the quotient A of interest
G
for us yields a trivial A ⊗ Q p -Lie algebra. Since ρθ is congruent to IndG QF ρg
and since its image contains the diagonal element δ with distinct eigenvalues,
we see that the diedral group generated by δ and by the element w of the Weyl
group which exchanges t1 and t2 in diag(t1 , t2 , t2−1 , t1−1 ) lifts to characteristic
zero as a subgroup of Im ρθ (by a theorem of P. Hall, as the kernel of Im ρθ →
G
Im IndG QF ρg is pro- p); hence Im ρθ contains the element w of the Weyl group
which exchanges t1 and t2 . Let α1 = t1 t2−1 and α2 = t22 be the two simple
roots. We recall the notations uγA = uγ (A) ∩ LieQ p Im ρ A for all roots γ s.

Lemma 9.24. For any characteristic zero quotient A of 2 , there exists a Lie
subalgebra Lie A of the Q p -Lie algebra LieQ p Im ρ A which is an A ⊗ Q p -
Lie algebra, which contains uαA2 and which is stable by the adjoint action of
Im ρ A . If 2 → A → B are two characteristic zero quotients of 2 , then
Lie A → Lie B is surjective.

Proof. By action of the inertia as in the proof of Theorem 9.17, we see that uα2 ,
resp. u2α1 +α2 , carries an action of (1 + T1 )(1 + T2 )−1 , resp. of (1 + T1 )(1 + T2 );
moreover uα2 = w(u2α1 +α2 ) hence uα2 carries a structure of A-module by
transport of structure. We construct Lie A as the Lie algebra generated by uαA2 ,
u2α1 +α2 and their Poisson brackets by the uαA2 s for all the root (positive or
negative) of sp4 . This defines the desired A ⊗ Q p -Lie algebra.
 Big image of Galois representations and congruence ideals 251

Again, it follows from the construction above that if ρθ is full over 2 ; then
for all roots γ , the γ (δ)-eigenspace for the action of δ as in Theorem 9.17) on
Lie2 , is non-zero. More precisely,

Corollary 9.25. If ρθ is full over 2 , there exists a non-zero largest n [ 1p ]-

ideal lθ such that
1
lθ · sp4 (n [ ]) ⊂ Lie2
p

We have lθ ⊂ lθ ⊗ Q p and these two ideals have the same radical: lθ =

lθ ⊗ Q p .

Proof. Simply review the construction of Lie2 starting from uγ (lθ ) ⊂ uγ for
all γ s, and take Poisson brackets.

Proof of the theorem. Let us prove that if P is a characteristic zero prime con-
taining lθ , then it contains c(θ ). We can assume that P is an isolated component
in the primary decomposition of the ideal lθ defined in the previous corollary.
Let q be the primary component of lθ whose radical is P. Let  P be the local-
ization of 2 at P, κ P = Frac(2 /P) its residue field. Let   = 2 /q and  P
its localization at P; it is an artinian  P -algebra with residue field κ P .
Let ρ be the reduction of ρθ modulo q; it takes values in G().  We consider
an ideal a of 2 containing q and  a = a/q, such that  a P is one-dimensional
over κ p . Consider the   ⊗ Q p -Lie algebra Lie  of Lemma 9.24 in gsp4 () 
and the Lie subalgebra s =    inside the Lie algebra 
a · sp4 () ∩ Lie 
a · sp4 (),
after localization at P, the latter is isomorphic to sp4 (κ P ). So we can identify
s P to a κ P -Lie subalgebra of sp4 (κ P ). Let us prove that s P = sp4 (κ P ). Indeed
if equality holds, we have   ⊂ Lie
a · sp4 ()  . hence a · sp4 (2 ) ⊂ Lie2 + q ·
sp4 (2 ) so after localizing at P, noting that lθ,P = q P and that lθ · sp4 (2 ) ⊂
Lie2 , we have a P · sp4 ( P ) ⊂ Lie P . Consider the ideal b = lθ + a, it is
strictly larger than lθ and we have b · sp4 (2 ) ⊂ Lie2 ; this contradicts the
definition of lθ .
In conclusion, s P is a strict κ P -Lie subalgebra of sp4 (κ P ). It is semi-simple
because ρ  is irreducible. Hence it is either 0, or an sl2 in the Levi of a maxi-
mal parabolic, or Sym3 sl2 , or the endoscopic sl2 × sl2 . Since it is normalized
by Im ρ P (recall ρ P is the representation over 2 /P), Im ρ P is contained in
the corresponding normalizer. However the second and third cases are impos-
sible. Indeed, for any root γ occurring in sl2 × sl2 , one can show using the
projection γ as Lemma 9.14 and Lemma 9.16, that Uγ ( ) ∩ Im ρ  maps sur-
jectively to Uγ (F p ) ∩ Im ρ. Hence by the assumption of fullness of ρ g and
252 Haruzo Hida and Jacques Tilouine

ρ gσ , these groups are non-trivial. For instance for γ = α2 , by taking a power

of p and taking the Log of a non-trivial element, we find a non-trivial element
 
of uα2 ⊂ Lie  , since both sides are -modules, one can assume that this non-

zero element is in uα2 (
a) ⊂ Lie , hence s P = 0. The only remaining case
is the endoscopic sl2 × sl2 and Im ρ P ⊂ N (GL2 × GL2 )◦ ). Comparing with
the reduction modulo the maximal ideal, we find a two-dimensional Galois
representation of G F over 2 /P congruent to g modulo the maximal ideal of
2 /P.
From this, one can deduce, using an R F = T F of Skinner-Wiles [SW01],
that there exists a Hilbert family g such that the family θ2 : T M,holcusp → J
associated to (F, g) which is congruent to θ modulo a prime Q of J above P.

Questions: 1) By this theorem, we can ask more precisely about the primary
ideals (for minimal primes) occurring in c(θ )[ 1p ] and lθ [ 1p ]. Are they equal?
2) Note that for a Hilbert form g on a real quadratic field F, the L function
G
of Ad IndG QF ρg can be decomposed as

G G
L(Ad IndG QF ρg , s) = L(IndG QF Adρg , s)L(Rg , s)

where Rg is the extension to G Q of the representation ρg ⊗ ρg∨σ of G F

G
appearing as a factor of Ad IndG QF ρg .
By a theorem of [Di05], the normalized special value at 1 of the first factor
control congruences between the Hilbert form g and other Hilbert forms, hence
a similar normalized special value at 1 of the second factor should control the
congruences between the Yoshida lift of g and other Siegel forms which are
not theta lifts of Hilbert forms. The p-adic L function interpolating the values
of the first factor for g varying in a Hida family g has been defined by Hida.
It might be possible to define the p-adic L function interpolating the second
factor for g varying in a Hida family. Then one could ask how it is related to
the determination of lθ for a family θ congruent to #(g).
3) More generally, we hope to treat other cases of congruences between
a “full” family and other families of automorphic forms with “small but
irreducible Galois image”, for GSp(4) and other groups.

References
[Ca13] F. Calegari, Even Galois representations and the Fontaine-Mazur conjec-
ture, Invent math. (DOI 10.1007/s00222-010-0297-0)
 Big image of Galois representations and congruence ideals 253

[CaGe13] F. Calegari, T. Gee, Irreducibility of automorphic Galois representations of

GL(n), n at most 5, to appear in Ann. Inst. Fourier
[Car94] H. Carayol, Formes modulaires et représentations galoisiennes à valeurs
dans un anneau local complet, in “p-adic monodromy and the Birch and
Swinnerton-Dyer conjecture”, Contemp. Math. vol.165, AMS 1994, pp.
213-237
[Che51] C. Chevalley, Théorie des groupes de LIE, II et III, Publ. Inst. Math. Univ.
Nancago, Paris, Hermann 1951 et 1955.
[Ch92] W.-C. Chi, l-adic and λ-adic representations associated to abelian varieties
defined over number fields. Amer. J. Math. 114 (1992), 315–353.
[Di05] M. Dimitrov, On Ihara’s Lemma for Hilbert modular varieties, Compositio
Mathematica 145, Issue 05 (2009), 1114-1146.
[Em14] M. Emerton, Local-global compatibility in the p-adic Langlands pro-
gramme for GL2/Q , preprint
[FC90] G. Faltings, C.-L. Chai, Degeneration of Abelian Varieties, Erg. Math.
Series 3-22, Springer Verlag 1990
[H02] H. Hida, Control theorems of coherent sheaves on Shimura varieties of PEL
type, J. Inst. Math. Jussieu, 1, 2002, pp.1-76
[H13a] H. Hida, Image of -adic Galois representations modulo p, Invent. Math.
194 (2013), 1–40.
[H13b] H. Hida, Big Galois representations and p-adic L-functions, preprint, 2012,
51 pages, to appear in Compositio Math.
[Lau05] G. Laumon, Fonctions zêta des variétés de Siegel de dimension trois,in
Formes Automorphes (II), le cas du groupe GSp(4), Astérisque 302, SMF,
2005
[LS98] Martin W. Liebeck and Gary M. Seitz, On the subgroup structure of
classical groups, Invent. math. 134 (1998), 427–453
[M04] D. Mauger, Algèbres de Hecke quasi-ordinaires universelles, Ann. Sci.
École Norm. Sup. (4) 37 (2004), 171–222.
[MT02] A. Mokrane, J. Tilouine, Cohomology of Siegel varieties with p-adic
integral coefficients and applications, in Asterisque 280, 2002
[Pi12] V. Pilloni, Sur la théorie de Hida pour le groupe GSp2g , Bulletin de la SMF
140 (2012), 335-400
[PiSt] V. Pilloni, B. Stroh, Surconvergence et classicité: le cas déployé, submitted
[P93] R. Pink, Classification of pro- p subgroups of SL2 over a p-adic ring, where
p is an odd prime, Compositio Math. 88 (1993), 251–264
[P98a] R. Pink, l-adic algebraic monodromy groups, cocharacters, and the
Mumford-Tate conjecture, J. reine angew. Math. 495 (1998), 187–237
[P98b] R. Pink, Compact subgroups of linear algebraic groups. J. Algebra 206
(1998), 438–504.
[R01] B. Roberts, Global L-packets for GSp2 and Theta lifts, Doc. Math. 6 (2001),
pp.247-314
[SW01] C. Skinner, A. Wiles, Nearly ordinary deformations of irreducible residual
representations, Ann. Fac. Sci. Toulouse Math. (6), 10 (1):185-215, 2001
[Ta93] R. Taylor, On the cohomology of Siegel threefolds, Invent. Math. 114
(1993), pp.289-310
254 Haruzo Hida and Jacques Tilouine

[TU99] J. Tilouine, E. Urban, Several variable p-adic families of Siegel-Hilbert

cusp eigensystems and their Galois representations, Ann. Sci. E.N.S., 4
série, t. 32, p. 499-574, 1999
[Ti09] J. Tilouine, Cohomologie des variétés de Siegel et représentations galoisi-
ennes associées aux représentations cuspidales cohomologiques de GSp(4),
Publ. Math. Besançon, Algèbre et Théorie des Nombres, 2009
[Ur05] E. Urban, Sur les représentations p-adiques associées aux représentations
cuspidales de GSp4 (Q), in Formes Automorphes (II) , le cas du groupe
GSp(4), pp. 151-176, Astérisque 302, SMF, 2005
[Ur11] E. Urban, Eigenvarieties for Reductive Groups, Annals of Math. 174 (2011),
1685-1784
[We05] R. Weissauer, Four-dimensional Galois representations, in Formes Auto-
morphes (II) , le cas du groupe GSp(4), pp., Astérisque 302, SMF,
2005
[Y79] H. Yoshida, Weil’s representations and Siegel’s modular forms. Lectures on
harmonic analysis on Lie groups and related topics (Strasbourg, 1979), pp.
319-341, Lectures in Math., 14, Kinokuniya Book Store, Tokyo, 1982.
[Z14] B. Zhao, Local indecomposability of Hilbert modular Galois represen-
tations, to appear in Ann. Inst. Fourier (Grenoble) (posted in web:
arXiv:1204.4007v1 [math.NT])
 10
The skew-symmetric pairing on the
Lubin–Tate formal module
M. A. Ivanov1 and S. V. Vostokov2
1 Saint Petersburg state university, 198504, Universitetsky prospekt,
28, Peterhof, St. Petersburg, Russia
E-mail address: [email protected]
2 Saint Petersburg state university, 198504, Universitetsky prospekt,
28, Peterhof, St. Petersburg, Russia
E-mail address: [email protected]

§10.1 Introduction
The classical Hilbert symbol determines a skew-symmetric pairing on the mul-
tiplicative group of a local field. Explicit formulae for the Hilbert symbol were
obtained independently by Brückner ([3]) and Vostokov ([2]) in 1978. Subse-
quently, these formulae were generalized to multi-dimensional local fields (see
[7]). In this note, we shall generalize these results to the formal modules.
Let F be a commutative formal group over the ring of integers ok of a local
field k; let k0 ⊂ k be the subfield such that ok0 is the endomorphism ring of F.
Let k  |k be a finite extension and let M := π ok  , where π is a prime element of
k  . Our goal is to construct a skew-symmetric bilinear pairing (α, β) : F(M)×
F(M) → ker[π n ] F .
Due to the short exact sequence

0 → FE (M) → E 0 (k) → E ns (k) → 0,

where E is an elliptic curve, E 0 is the set of points with non-singular reduction,

E ns is the set of non-singular points, and FE is a formal group, associated to
E (cf. [9, Ch. 7]), our pairing defines a pairing on the group E 0 (k) of the
elliptic curve E with non-singular reduction. In the multiplicative case this

The authors are grateful to the Hausdorff Research Institute for Mathematics and Max Planck
Institute for Mathematics in Bonn (Germany) for the excellent working conditions. This work has
been partially supported by the RFFI (grant 11-01-00588-a)

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

255
256 M. A. Ivanov and S. V. Vostokov

construction determines the wild Hilbert symbol (a pairing on E ns (k) gives

the tame symbol).
We shall begin with the simplest case of formal groups, that is with the
Lubin–Tate formal group, corresponding to an elliptic curve with complex
multiplication. The endomorphism ring of such a group coincides with ok and
the kernel of the isogeny is generated by the unique element denoted by ξ .

§10.2 The construction

We introduce the basic notation of this article:

● k0 is a local field (a finite extension of Q p ), p > 2,

● o0 is the ring of integers of the field k0 ,
● π0 is a prime element of o0 ,
● F(X, Y ) ∈ o0 [[X, Y ]] is the formal series, defining a Lubin–Tate formal
group over o0 ,
● λ(X ) is the logarithm of F (cf. [8, Ch. 8]),
● ξ is a fixed primitive root of the isogeny [π0n ](X ),
● k|k0 is a finite extension with ξ ∈ k,
● π is a prime element of k,
● e is the ramification index of the extension k|k0 ,
● the Lubin–Tate formal group F defines a formal module F(M) on the
maximal ideal M = (π ) of k,
● T is the inertia subfield in k/k0 ,
● oT is the ring of integers of T ,
● . is the Frobenius endomorphism of T |k0 ,
● R is the Teichmüller system in T ,
● for α ∈ M, let α(X ) ∈ X oT [[X ]] be such that α(π ) = α.

Our goal is to define a bilinear skew-symmetric pairing

(·, ·) : F(M) × F(M) → κn := ker[π0n ] F . (10.1)

At first, we shall introduce a pairing on the rings of formal series

·, · : X oT [[X ]] × X oT [[X ]] → oT /π0n oT . (10.2)

The pairing (2) allows us to define a pairing on the formal module

(α, β) = [α(X ), β(X )](ξ ), (10.3)

 The skew-symmetric pairing on the Lubin–Tate formal module 257

As usual, we define the action of the Frobenius automorphism . on the


series φ = di X i (lying either in oT ((X )) or in oT {{X }}) by
.
.φ = φ . = di X qi . (10.4)

Let us also define the Artin–Hasse logarithm of the Lubin–Tate formal

group:

(φ) = 1 − π.0 λ(φ), for φ ∈ X oT [[X ]],

and the Artin–Hasse exponent (inverse to (φ)):

. .2
E(ψ) = λ−1 (1 + π0 + + . . . )(ψ).
π02

For ·, · we take the pairing constructed similarly to the multiplicative case:
 
1
α(X ), β(X ) = Tr res +(α(X ), β(X )) (10.5)
s(X )
with s(X ) = [π0n ](ξ (X )), Tr = TrT /k0 , and

+(α(X ), β(X ))
= (α(X )) d(β(X )) − (α(X )) dλ(β(X )) + (β(X )) dλ(α(X )).

Simple properties
The pairing (5) is well defined because +(α, β) ∈ oT [[X ]] ((X ) ∈ oT [[X ]]
and dλ(X ) ∈ oT [[X ]] (see [4, §2.3])).
The bilinearity of the pairing (5) (and of the pairing (3) as consequence of
it) follows from the properties of λ(X ) and (X ) (see [4, §0.3, §2.3]):

α1 + F α2 , β = α1 , β + α2 , β,

α, β1 + F β2  = α, β1  + α, β2 
(10.6)
[a] F α, β = aα, β,
α, [c] F β = cα, β,

(α1 + F α2 , β) = (α1 , β) + F (α2 , β),

(α, β1 + F β2 ) = (α, β1 ) + F (α, β2 )
(10.7)
([a] F α, β) = [a] F (α, β),
(α, [c] F β) = [c] F (α, β)

with a, c ∈ o0 .
258 M. A. Ivanov and S. V. Vostokov

By Lemma 10.2 below,

1
res d((α) (β)) ≡0 mod π0n ;
s
on the other hand,
1 1
α, β + β, α = res[+(α, β) + +(β, α)] = res d((α) (β)) ,
s s
therefore the pairing ·, · is skew-symmetry.

The Shafarevich basis

A suitable basis of the formal module F(M) can viewed as a generalization
the Shafarevich canonical basic of the group of principal units of a local field
(see [1, 2]). It can be described as follows. Let α ∈ F(M), then

α= (F) E F (ai X
i
)| X =π + F ω(a), ai , a ∈ o T ,
i

where 1  i < qe/(q − 1), q  i, and ω(a) = E F (as(X ))| X =π ∈ oT is

a π0n -primary element (see [4]). Besides the ok0 -module  of the π0n -primary
elements is generated by ω(a0 ), so that a0 ∈ oT and Tr a0 ∈ / π ok 0 .
An element α belongs to [π0n ] F F(M) if and only if Tr a ∈ π n ok and ai ∈
π n oT . There are also other generalizations of the Shafarevich basis [4, 6].
To compare pairing (3) with the known pairings on the multiplicative group
[2] and on the product of the multiplicative group and the formal Lubin–Tate
module [4], let us remark that in those two cases the proof of the explicit for-
mulae depends on the study of the pairs (π, β), where π is the uniformizing
element and β is a principal unit. Such a method does not work for pairing
(3). Loosely speaking, we define a pairing on the product of two groups of
“principal units”.

§10.3 Invariance of the pairing

Let us prove that our pairing (5) is invariant under a change of variable and
therefore the pairing (3) does not depend on the choice of the prime element
π. We refer to [2, §2.4] for analogous considerations in the multiplicative case.
It suffices to establish invariance of E(a X u ), E(bX v ) with a, b ∈ R
because any element in oT is a linear combination of some elements in R,
the pairing (5) is bilinear, and ε = E((ε)) for ε ∈ X oT [[X ]], where E is an
additive bilinear function. Let

X = g(Y ) = cψ(Y ) with c ∈ R and ψ(0) = 1.

 The skew-symmetric pairing on the Lubin–Tate formal module 259

Since . and E depend on the choice of X , let us write . X and .Y instead

of . when appropriate, and similarly for E and res.
In these notations,

E X (a X u ) = E Y (1 − .π0Y )S , (10.8)
where
∞ r
(ag u )q
a ∈ R, and S = ,
π0r
r=0
for

E Y (1 − .π0Y )S = λ−1 (1 + .π0Y + . . . )(1 − .π0Y )S = λ−1 S =
 (ag m )q   (a X m )q 
= λ−1 ag m + +. . . = λ−1 a X m + +. . . = E X (a X m ).
π0 π0
.Y
Let us note that the series (1 − π0 )S has integer coefficients.

Lemma 10.1. We have

k k−1 .
hq − hq
∈ oT [[X ]],
π0k
for h ∈ oT [[X ]].

Proof. This Lemma can be easily proved by induction on k.

Lemma 10.2. We have

1
res dU ≡0 mod π0n ,
s
for U ∈ oT [[X ]].

Proof. See [4, (18)].

Proposition 10.3. For p = 2 the following congruence holds true:

 
E X (a X u ), E X (bX v ) X ≡ E Y (1 − .π0Y )S1 , E Y (1 − .π0Y )S2 Y mod π0n .

Proof. It follows from (5) that

1
E X (a X u ), E X (bX v ) X = Tr res X +(X ) ,
s(X )
where
 (bX v )q 
+(X ) = a X u d(bX v ) − a X u d bX v + + ...
π0
 (a X u )q 
+ bX v d a X u + + ... .
π0
260 M. A. Ivanov and S. V. Vostokov

Moreover,
 .Y  .Y 1
E Y (1 − π0 )S1 , E Y (1 − π0 )S2 Y = Tr resY (Y ) ,
s(g(Y ))
where
.Y
(Y ) = (1 − )S1 d(1 − .π0Y )S2 − (1 − .π0Y )S1 dS2 + (1 − .π0Y )S2 dS1 =
π0
 (bg v )q   (ag u )q 
= ag u d(bg v ) − ag u d bg v + + . . . + bg v d ag u + +... +
π0 π0
+ σr + τr,s = σr + τr,s + +(g(Y )) dg(Y )
r1 k,s1 r1 k,s1

and
 (bg v )q k (bg v )q
k−1 .
Y 
σk = ag d u
− + (10.9)
π0k π0k
 (bg v )q k (bg v )q . y 
k−1  (bg v )q − (bg v )q .Y 
k k−1

+ d(ag u ) · − = d ag u · .
π0k π0k π0k

 (ag u )q s  (bg v )q k − (bg v )q k−1 .Y

τk,s =d · −
π0s π0k
(ag u )q − (ag u )q .Y
s s−1  (bg v )q k−1 .Y 
− · d .
π0s π0k
The series
(bg v )q − (bg v )q
k k−1 .
Y
ag · u
π0k
has integer coefficients (see Lemma 10.1), and therefore
∞
σr = dϕ1 (Y ), ϕ1 (Y ) ∈ oT [[X ]].
r=1

It follows also that

 
uq s 1  u qs v qk u q s−1 . y v q r−1 .Y

τk,s = d (ag ) (bq ) − (ag ) (bg ) −
uq s + vq k π0s+k
 
(bq v )q .Y  u q s 
k−1
u q s−1 . y
−d (ag ) − (ag )
π0s+k
 The skew-symmetric pairing on the Lubin–Tate formal module 261

Let us verify that the power series under the differential have integer
coefficients. In view of
.Y
s −uq s−1 . s −uq s−1 . uq s (1− )
g uq Y = ψ uq Y = exp log(ψ q ) = exp(uq s m (ψ)),

one obtains
uq s 1  1 
exp((uq s +vq k ) m ψ)−1 − s+k exp(uq s m ψ)−1 =
uq + vq π0
s k s+k
π0
uq s (uq s + vq k )n−1 − (uq s )n
= (m ψ)n .
n2
π0s+k n!

It is not hard to see that for q = 2 all the coefficients are integers, hence
∞
τk,s = dϕ2 (Y ), ϕ2 (Y ) ∈ oT [[X ]].
k,s

The above calculations, when combined with Lemma 10.2, show that

  1 
1
resY (Y ) ≡ resY +(g(Y )) dg(Y ) + dϕ1 + dϕ2 ≡
s(g(Y )) s(g(Y ))
 
1
≡ res X +(X ) mod π0n .
s(X )
The proposition is proved.

From now on, let us fix the prime π and define the pairing (·, ·) by (3). We
have just proved that the pairing (·, ·) is bilinear, skew-symmetric, and invariant
under the choice of the prime element π .
Moreover, it can be checked in a standard way (see, for instance, [5]) that
the pairing (·, ·) does not depend on the expansions of α and β of α and β.

§10.4 The kernel of the pairing

Let us restrict ourselves, for simplicity, to the totally ramified extensions.

Proposition 10.4. 1) The module of the π0n -primary elements is contained in

the kernel of the pairing.
2) The pairing
F(M)/ × F(M)/ → ker[π0n ] (10.10)
262 M. A. Ivanov and S. V. Vostokov

is nondegenerate modulo [π0n ], that is, if (α, β) = 0 for all β ∈ F(M)/ ,

then α ∈ [π0n ](F(M)).

Proof. To prove 1), let us calculate the pairing ω(a), ε. We have
 
1
ω(a), ε = Tr res −as(X ) d( π.0 λ(s(X ))) +
s(X )
 
1
+ Tr res (ε(X )) d(1 + π.0 + .2 + . . . )(as(X ))
2

π0 s(X )
Obviously, the first term is equal to zero; as for the second term, it suffices to
note that
s.
r
qr
d r = X q −1 r (ds(X )). ≡ 0 mod π0n ,
r r

π0 π0
see, for example, [4, §1.4]. This proves 1).
For α ∈ F(M)/, let us constuct an element β such that α, β = 0. We
have:
α = α(X )| X =π = E(w(X ))| X =π ,
pe
with w(X ) = w1 X m 1 + . . . + wt X m t ∈ oT [[X ]], where p  m i and m i < p−1
for 1  i  t.
Let us suppose first that α is not divisible by [π0 ], that is, there exists j such
that a j ∈ o∗T . Let j0 = j0 (α) be the minimal such index. Then

pe pe
− j0 − j0
+(α(X ), E(X p−1 )) ≡ X p−1 dw(X ) ≡
pe pe
≡ w j0 j0 X p−1 −1 mod (π0 , X p−1 ),
and one can prove [6, §10.25] that
1 c0 c1
≡ pe + pe + ... mod π0 , c0 ∈ o∗T ,
s(X ) X p−1 X p−1
−p

It follows from those two congruences that

pe
− j0
α, E(X p−1 )|π0  ≡ Tr c0 w j0 j0 ≡ 0 mod π0 .
In the general case, if
α ∈ [π0k ]F(M) and α ∈ [π0k+1 ]F(M),
pe
− j0 (α  )
then there is α  such that α = [π0k ]α  . Let β = E(X p−1 )|π0 , then it
follows from our calculation that
α, β = [π0k ]α  , β ≡ 0 mod π0k .
This proves 2).
 The skew-symmetric pairing on the Lubin–Tate formal module 263

References
[1] I. R. Shafarevich, A general reciprocity law, Mat. Sbornik N.S. 26(68) (1950),
113–146 (Russian). MR0031944 (11,230f)
[2] S. V. Vostokov, An explicit form of the reciprocity law, Izv. Akad. Nauk SSSR Ser.
Mat. 42 (1978), no. 6, 1288–1321, 1439 (Russian); English transl., Math. USSR-
Izv. 13 (1979), no. 3, 557–588. MR522940 (80f:12014)
[3] H. Brückner, Explizites Reziprozitätsgesetz und Anwendungen, Vorlesungen aus
dem Fachbereich Mathematik der Universität Essen [Lecture Notes in Mathemat-
ics at the University of Essen], vol. 2, Universität Essen Fachbereich Mathematik,
Essen, 1979 (German). MR533354 (80m:12015)
[4] S. V. Vostokov, A norm pairing in formal modules, Izv. Akad. Nauk SSSR Ser.
Mat. 43 (1979), no. 4, 765–794, 966 (Russian); English transl., Math. USSR-Izv.
15 (1980), no. 1, 25–51. MR548504 (81b:12017)
[5] ______, The Hilbert symbol in a discretely valued field, Zap. Nauchn. Sem.
Leningrad. Otdel. Mat. Inst. Steklov. (LOMI) 94 (1979), 50–69, 150 (Russian);
English transl., Journal of Soviet Mathematics 19 (1982), no. 1, 1006–1019. Rings
and modules, 2. MR571515 (81g:12019)
[6] ______, Symbols on formal groups, Izv. Akad. Nauk SSSR Ser. Mat. 45 (1981),
no. 5, 985–1014, 1198 (Russian); English transl., Math. USSR-Izv. 19 (1982), no.
2, 261–284. MR637613 (83c:12016)
[7] ______, Explicit construction of the class field theory for a multidimensional local
field, Izv. Akad. Nauk SSSR Ser. Mat. 49 (1985), no. 2, 283–308, 461 (Rus-
sian); English transl., Math. USSR-Izv. 22 (1986), no. 2, 263–288. MR791304
(86m:11096)
[8] I. B. Fesenko and S. V. Vostokov, Local fields and their extensions, 2nd ed.,
Translations of Mathematical Monographs, vol. 121, American Mathematical So-
ciety, Providence, RI, 2002. With a foreword by I. R. Shafarevich. MR1915966
(2003c:11150)
[9] Joseph H. Silverman, The arithmetic of elliptic curves, 2nd ed., Graduate Texts in
Mathematics, vol. 106, Springer, Dordrecht, 2009. MR2514094 (2010i:11005)
 11
Equations in matrix groups and algebras over
number fields and rings: prolegomena to a
lowbrow noncommutative Diophantine
geometry
Boris Kunyavskiı̆
Department of Mathematics, Bar-Ilan University, 5290002 Ramat Gan, ISRAEL
E-mail address: [email protected]

In memory of Valentin Evgenyevich Voskresenskiı̆

A BSTRACT. We give a brief survey of some problems related to the solvability

of word equations in matrix groups and polynomial equations in matrix alge-
bras, focusing on the cases where the matrix entries belong to a global field or
its ring of integers.

What is needed is a corpus of explicit concrete cases and a

middlebrow arithmetic theory which would provide both a
practicable means to obtain them and a framework to
understand any unexpected regularities.

J. W. S. Cassels, E. V. Flynn, Prolegomena to a Middlebrow

Arithmetic of Curves of Genus 2, Cambridge, 1996

11.1 Introduction
In the present survey, by a noncommutative Diophantine problem we mean the
question on the existence of a solution to an equation of the form

F(A1 , . . . , Am , X 1 , . . . , X d ) = 0, (11.1)

where the Ai are fixed elements of an associative or Lie algebra A, or of a

group G, F is an associative or Lie polynomial P, or a group word w, and

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

264
 Equations in matrix groups and algebras 265

solutions are d-uples (X 1 , . . . , X d ) sought in Ad or G d , respectively (in the

group case we change 0 to 1 in the right-hand side of (11.1)). We restrict our
attention to a particular class of equations:
F(X 1 , . . . , X d ) = A, (11.2)
where A is fixed and only scalar constants are permitted in the left-hand side
of (11.2).
If the algebra A or the group G admits a faithful matrix representation over
a field k or a ring R, then we can regard the matrix entries as variables and
thus reduce our noncommutative Diophantine problem to a commutative one
(in the group case, this reduction may require either cancelling denominators
arising from negative powers of X i ’s in group words, or using representations
in SL rather than in GL, which can easily be achieved by embedding GLn into
SLn+1 ).
This “abelianization” has several drawbacks. First, we replace a single
(though noncommutative) polynomial equation with a system of polynomial
equations which may be really huge (though commutative), require computer
investigation, and go far beyond software limitations. Second, this abelianiza-
tion process is somehow forgetful: essential symmetries built in the original
algebra or group are hidden in the arising polynomial system, and it is not an
easy task to reveal them.
However, sometimes one should be ready to pay this price, having in mind
a potential reward: full access to a mill driven by arithmetic-geometric power,
allowing one to grind hard grain. Such a reward is particularly achievable in
the finite group case, when one can use representations over finite fields and
apply to the resulting problem of the existence of rational points on algebraic
varieties over finite fields the whole strength of high-tech homological algebra
developed over the twentieth century (Weil–Grothendieck–Deligne). Perhaps
the first, quite impressive instance of such tour de force is due to Bombieri
[Bom1] (see also a follow-up in [Bom2]), who ingeniously applied elimination
techniques to polynomial systems over finite fields which arise in classifi-
cation of finite simple groups of Ree type (note that computer calculations
played a critically important role). Over the past decade, some new appli-
cations of arithmetic-geometric methods appeared, which resulted in solving
long-standing problems in the theory of finite [BGGKPP1]–[BGGKPP2] and
infinite [BS1]–[BS2] groups; see surveys [Sh1]–[Sh3], [GKP], [BGaK] for
more details (including the discussion of the famous Ore problem [Mall],
where the impressive final stop was recently put in [LOST]).
In the present paper, our focus is, however, somewhat different: we are go-
ing to discuss the situation when matrix representations described above are
266 Boris Kunyavskiı̆

defined over some infinite field (or ring) of arithmetic nature. Our primary
interest is addressed to the field Q of rational numbers and the ring Z of
rational integers. Here, comparing to the finite field case, achievements are
much more modest, to put it mildly. We will give a brief survey of some
of those, trying to put the relevant problems in a general context of local–
global considerations, traditional for classical “commutative” Diophantine
geometry.
Note that perhaps the only existing general result, Borel’s theorem [Bor],
guaranteeing, in the case where G is a connected semisimple linear algebraic
group over an algebraically closed field and w is any non-identity group word,
the solvability of equation (11.2) with a “general” (in the sense of the Zariski
topology) right-hand side, gives very little for smaller fields: A. Thom showed
that even over reals, if G is compact, the solutions of the relevant equation may,
for an appropriately chosen w, be included in an arbitrarily small (in the sense
of real topology) neighbourhood of the identity [Th]. Once again, it is worth
quoting the book of Cassels and Flynn [CF], the title of which inspired (and
was shamelessly plagiarised by) the author of this survey: ‘Modern geometers
take account of the ground field, but regard a field extension as a cheap ma-
noeuvre. For practical computations it is desperately expensive: the theory we
seek must avoid them at almost all costs.’
The author has to apologise that the corpus of presented explicit examples,
as well as the theory brought for explaining some regularities, seem, at the cur-
rent stage, too poor. What is badly needed is some replacement of the missing
algebraic chain in Manin’s triad appearing in the title of [Man]: by now, we
are not aware of any reasonable analogues of Galois-cohomological machin-
ery and descent methods, which proved their power in clarifying ties between
geometric and arithmetic properties of varieties considered in [Man] and [CF]
(and in many other sources). This explains why the term “middlebrow” has
been downgraded in the title of this paper.

11.2 Local obstructions

First, it should be emphasised once again that in the present text we restrict our
attention to considering word equations in simple algebraic groups, as well
as to polynomial equations in simple matrix algebras (and close to such).
Apart from the reason mentioned in the introduction (that we want to use
algebraic–geometric machinery), there are other arguments in favour of such
limitations: roughly, in a “random” group a “random” word equation has little
chance to be solvable; even if we drop one of the assumptions on the group
 Equations in matrix groups and algebras 267

(either simplicity, or algebraicity), even the simplest word equations, such as

the commutator equation [x, y] = g, may turn out to be nonsolvable; see
[KBKP] for relevant discussions and counter-examples.
So throughout below all word (resp. polynomial) equations will be consid-
ered in G(k), where G is a connected simple linear algebraic k-group (resp. in
the matrix algebra Mn (k)).
As our focus is on arithmetic properties of matrix equations, k will typically
stand for a number field (usually k = Q) or its ring of integers.
We are going to discuss the traditional local–global approach to solving ma-
trix equations over a number field or its ring of integers, so the first question
is the existence of local obstructions. Here we notice the first unusual phe-
nomenon: even an innocent looking matrix equation may have no solutions
over C. Below are the simplest instances.

11.2.1 Power maps

Let w : G → G be defined by w(x) = x m where m ≥ 2 is an integer. Looking
at the equation
 
−1 1
x2 = (11.3)
0 −1

in G = SL2 (C), we immediately notice that it has no solutions (apply Jor-

dan’s theorem). This obstruction can easily be removed by factoring out the
centre: indeed, one can show that in PSL2 (C) any power equation x m = a
is solvable. However, this observation is somewhat misleading because for
simple algebraic groups of types other than A the obstruction over C does
exist:

Theorem 11.1 ([Ch1], [Ch2], [Ste]). The map x → x m is surjective on G(k)

(k is an algebraically closed field of characteristic exponent p, G is a con-
nected semisimple algebraic k-group) if and only if m is prime to pr z, where
z is the order of the centre of G and r is the product of “bad” primes.

So even in characteristic zero (when p = 1), even for groups of adjoint type
(when z = 1), the power map x → x m is not surjective whenever m is divisible
by a bad prime r , which can be 2,3 or 5, depending on the type of the Dynkin
diagram of G.
Other local obstructions ( p-adic and real) to the surjectivity of power maps
are also completely described, see [Ch3], [Ch4].
Let us now go over to more complicated maps.
268 Boris Kunyavskiı̆

11.2.2 Group case

As above, here G stands for a connected semisimple algebraic k-group,
G = G(k), w is a word in d letters. We denote by the same letter the map
w : G d → G, (g1 , . . . , gd ) → w(g1 , . . . , gd ).
Assume that G is of adjoint type, so G has no centre. Here is a brief account
of what is known by now.

(i) For k = C it is a challenging open question whether or not there exist

a nonpower word w (i.e. w cannot be represented as a proper power of
another word v) and a group G such that the map w is not surjective; see
[KBKP] for a survey of known surjectivity results.
(ii) For k = R, A. Thom [Th] constructed words w in two variables such
that the image of the corresponding word map on the unitary group lies
within a neighbourhood of the identity matrix, and this neighourhood can
be made as small as we wish by an appropriate choice of w.
(iii) This construction can easily be extended to an arbitrary compact simple
real group and also (using a theorem of Bruhat–Tits–Rousseau, see [Pr]
for a short proof) to an arbitrary anisotropic p-adic simple group (which
is necessarily of type A).

If examples as in (i) above do exist, one can ask whether the surjectivity can
be tested on real points:

Question 11.2. Do there exist a word w and a simple algebraic R-group

G such that the map wR : (G(R))d → G(R) is surjective but the map
wC : (G(C))d → G(C) is not?

Similar questions can be asked for justification of “global-to-local” surjec-

tivity principles:

Question 11.3. Do there exist a word w and a simple algebraic Q-group

G such that the map wQ : (G(Q))d → G(Q) is surjective but the map
wR : (G(R))d → G(R) is not?

Question 11.4. Do there exist a word w, a simple algebraic Q-group G and a

prime number p such that the map wQ : (G(Q))d → G(Q) is surjective but the
map w p : (G(Q p ))d → G(Q p ) is not?

The author is not aware of any results in this direction.

 Equations in matrix groups and algebras 269

11.2.3 Algebra case

For the solvability of polynomial equations (11.2) in matrix algebras there are
several obstructions: obvious, such as the case where F is an identical relation
of the algebra under consideration, or the case where F is a Lie polynomial
and hence its image consists of trace zero matrices; less obvious but known
for decades, such as the case where all values of F lie in the centre of the
algebra; more recent, described in [KBMR1]–[KBMR3], [Sp] (for matrices
over an algebraically closed or, more generally, a quadratically closed field)
and [Male] (for matrices over a real closed field); see these papers, as well as
the survey [KBKP] for more details and some open problems.
The case of Lie polynomials on simple Chevalley Lie algebras was treated
in some detail in [BGKP], where among other things, one can find counter-
examples to the surjectivity as well. However, the cases of real and p-adic
algebras remain almost unexplored, for both associative and Lie algebras. In
particular, the author does not know anything regarding analogues of Questions
11.2–11.4.
The next questions refer to an eventual analogue of Thom’s construction
described in 12.4(ii):

Questions 11.5. Do there exist a simple real Lie algebra g and a Lie polyno-
mial P such that the image of the map P : gd → g lies within an open (in the
real topology) neighbourhood of zero U (different from g)? Can one choose U
as small as we wish by an appropriate choice of P? Can one arrange this for
g = su(n), the algebra of skew-hermitian trace zero matrices? Can one do the
same for some simple p-adic Lie algebra?

11.3 Local–global principles

In light of Questions 11.2–11.4, it seems reasonable to formulate local–global
principles for the surjectivity of word (or polynomial) maps as follows. Let G
denote a simple linear algebraic Q-group. For p ≤ ∞ we denote by Q p the
completion of Q at p where Q∞ := R.

Question 11.6. Let w(x1 , . . . , xd ) = g (resp. P(X 1 , . . . , X d ) = a) be a word

(resp. polynomial) equation in the group G(Q) (resp. in the associative algebra
Mn (Q)). Suppose that for every g ∈ G(Q) (resp. for every a ∈ Mn (Q)) the
equation is solvable in G(Q p ) (resp. in Mn (Q p )) for all p ≤ ∞. Is it true that
for every g ∈ G(Q) (resp. for every a ∈ Mn (Q)) it is solvable in G(Q) (resp.
in Mn (Q))?
270 Boris Kunyavskiı̆

If the answer to this question is “yes”, we say that the pair (w, G) (resp.
(P, Mn )) satisfies weak Grunwald–Wang principle (see the next section for
justification of this terminology).
One can also consider a variant of this principle, where the local require-
ments are weakened so that the global solvability is implied by the local
solvability at all but finitely many p. In this case we say that the pair under
consideration satisfies strong Grunwald–Wang principle.
Naturally, both versions can be discussed when Q is replaced with any
number field (or, more generally, any global field).
As in the case of obstructions over C, let us start with power maps.

11.3.1 Power maps

Variants of Question 11.6 can be posed for G other than simple linear alge-
braic groups, say, for commutative algebraic groups. In this context, the case
of power words w = x m was thoroughly investigated. For the multiplica-
tive group Gm , the answer follows from the Grunwald–Wang theorem [AT,
Chapter X], [NSW, Chapter IX], [Co, Appendix A] (see [Roq] for histori-
cal details). Given a global field k and a natural number m, it says whether
there is an element of k which is an m th power everywhere (or almost ev-
erywhere) locally but not globally. In particular, if m is divisible by 8, there
are counter-examples to strong Grunwald–Wang principle for (x m , Gm,Q ). For
(x m , Gm,Q(√7) ) there are well-known counter-examples to weak Grunwald–
√
Wang principle: 16 is not an 8th power in Q( 7) though it is an 8th power
everywhere locally.
For power maps on more general commutative algebraic groups, perhaps
the first counter-example to weak Grunwald–Wang principle was constructed
by Cassels and Flynn [CF, 6.9]. In their example, m = 2 and G is an abelian
Q-surface. Over the past decade, both versions of Grunwald–Wang principles
(weak and strong) for power maps have been studied in detail, which led to
almost conclusive results for algebraic tori [DZ1], [Il], elliptic curves [Wo],
[DZ1]–[DZ3], [Pa1]–[Pa3], [PRV], [Cr2], and higher-dimensional abelian
varieties [Wo], [Cr1]. The interested reader is referred to [Cr1], [Cr2] for
a detailed account of these results as well as for various ramifications of
local–global principles for higher cohomology.
These considerations lead to counter-examples to weak Grunwald–Wang
principle for power maps
√ on simple linear algebraic groups defined over num-
ber fields such as Q( 7): indeed, the diagonal 2 × 2 matrix with diagonal
entries 16 and 1/16 cannot be an 8th power of a matrix from SL2 (Q) though it
is an 8th power everywhere locally.
 Equations in matrix groups and algebras 271

As to more general words, the author is not aware of any counter-examples

to Grunwald–Wang principles.

11.3.2 Digression on relations with rationality

and approximation problems
It is well known that counter-examples to Grunwald–Wang principles are
closely related to counter-examples to E. Noether’s problem on the rationality
of the field of invariants Q(x1 , . . . , xn )G with respect to the permutation action
of a finite group G on the xi . Perhaps this relation was first noticed by Saltman
[Sal1], [Sal2] (see also [Sw] for a survey of these results), who provided an
argument based on the existence (or, more precisely, non-existence) of generic
G-extensions and described relations with lifting and approximation problems.
There is another approach, where a bridge between the Grunwald–Wang and
Noether problems goes through birational invariants of certain algebraic tori.
Namely, according to Sansuc [San, §§1–2], the cohomological obstruction
to the classical strong Grunwald–Wang principle is of purely algebraic nature:
the deviation from this principle is measured by the finite abelian group
⎡ ⎤

X1ω (k, μm ) = ker ⎣ H 1 ( , μm ) → H 1 (γ  , μm )⎦ , (11.4)
γ ⊂

where μm is the group of m th roots of unity, = Gal(K /k) is the Galois

group of the splitting field K of μm , and the sum in the right-hand side is
taken over all cyclic subgroups γ  of . Note that the deviation from the
weak Grunwald–Wang principle is measured by the subgroup X (k, μm ) of
1

Xω (k, μm ), which is of arithmetic nature, as well as their quotient, measuring

1

the gap between the weak and strong principles (see [San, Lemme 1.4 and
Sec. 2a]).
On the other hand, one can show (see [CTS]) that the same finite abelian
group Xω (k, μm ) provides a Galois-cohomological obstruction to the ra-
1

tionality of the field of invariants in Noether’s problem, and give another

interpretation of this obstruction. Namely, following Voskresenskiı̆ (see [Vo1],
[Vo2, Chapter VII]), consider, for any finite abelian group C (say, for C =
Z/m), the k-torus Q C = S/C where S is a quasitrivial torus (say, for
C = Z/m we have S = R K /k Gm where K is the splitting field of the character
group Ĉ = μm , i.e. the fixed field with respect to the action of Gal(k) on Ĉ).
Then the field of invariants in Noether’s problem for C is none other than the
field of rational functions k(Q C ). From the exact sequence of -modules
272 Boris Kunyavskiı̆

0 → Q̂ C → Ŝ → Ĉ → 0,

using the fact that Ŝ is a permutation -module and hence the group
⎡ ⎤

X2ω (k, Ŝ) := ker ⎣ H 2 ( , Ŝ) → H 2 (γ  , Ŝ)⎦
γ ⊂

is trivial [San, Lemme 1.9], we obtain that Xω (k, Ĉ) ∼ = Xω (k, Q̂ C ). This
1 2

latter group is a birational invariant of the k-torus Q C . It is isomorphic to

Br(X )/ Br(k), where X is a smooth projective k-variety containing Q C as
an open subset [CTS, Proposition 9.5(ii)] (the unramified Brauer group of
k(Q C )/k, using an alternative terminology), so whenever this invariant does
not vanish, one can conclude that the field in Noether’s problem for C is not
rational. Thus any counter-example to strong Grunwald–Wang principle yields
a counter-example to Noether’s problem.
It is worth noting that the abelian group Xω (k, μm )/X (k, μm ), men-
1 1

tioned above (or, more precisely, its Pontryagin dual), carries somewhat
different arithmetical information, which is in a sense complementary to weak
Grunwald–Wang principle. Namely, whenever this group vanishes, the classi-
cal version of the Grunwald–Wang theorem for the field k and the cyclic group
C = Z/m holds: for any finite set S of places of k and any collection of local
field extensions K (v) /kv with Galois group v (v ∈ S) embeddable into C,
there exists an extension K /k with Galois group C such that for every w|v one
has K w ∼ = K (v) (see [NSW, IX.2], [Co], [Ne] for other versions). This property
of the triple (k, C, S) is sometimes called weak approximation with respect to
S. The pair (k, C) is said to satisfy weak (resp. very weak) approximation if
the triple (k, C, S) satisfies weak approximation for every finite S (resp. for
every finite S disjoint from some finite set S0 depending from k and C). The
classical Grunwald–Wang theory establishes very weak approximation for all
abelian groups C and all global fields k and describes all cases when weak
approximation holds. These properties also make sense for nonabelian finite
groups C and are equivalent to weak (resp. very weak) approximation for quo-
tients SLn /C (see, e.g., [Ha2]). According to a well-known observation due
to Colliot-Thélène, very weak approximation for (k, C) implies that the in-
verse Galois problem for C over k is solvable; moreover, it is solvable under a
weaker assumption of so-called hyper-weak approximation [Ha2]; a recent pa-
per [DG] describes other interrelations between Grunwald–Wang and inverse
Galois problems.
For nonabelian groups C, conjectural relations of weak approximation prop-
erties to the unramified Brauer group are much more tricky comparing with the
 Equations in matrix groups and algebras 273

abelian case. We leave them aside referring the interested reader to [Ha2] and
more recent papers [De], [BDH], [LA], [Ne]. In the latter paper there has been
defined another cohomological invariant, detecting the failure of weak approx-
imation for some semi-direct products, and there has been introduced a finer
subdivision of Grunwald problems into tame (where all primes dividing the
order of C are outside S) and wild ones. It was shown in [Ne] that for every
C there are wild problems with negative solution. For tame problems no such
examples are known.
To finish with this digression, let us mention a recent generalization of
the Grunwald–Wang theorem, viewed as a statement on characters of Gm , to
automorphic representations of semisimple groups [So, §3].

11.3.3 Back to general words

It is not clear to what extent power words can serve as a prototypical toy-model
for investigating local–global properties in the general case. It is a common
understanding that, say, commutator words are in a certain sense opposite to
powers (cf., e.g., [HSVZ, Section 6.5]). Moreover, looking at equidistribution
properties of fibres of word maps on simple algebraic groups over finite fields,
one can observe that the behaviour of all nonpower words is in striking contrast
with the case of power words [BK1]. However, the author believes that the lack
of experimental evidence does not prevent from raising various questions of
local–global nature, such as Question 11.6, for nonpower words as well. Note
that collecting experimental data is not an easy task: it is not obvious how to
produce a nonpower word w such that for some simple algebraic Q-group G
the map w : G(Q)d → G(Q) is not surjective. Here is an example of such a
word.

Proposition 11.7. For w(x, y) = x 2 [x −2 , y −1 ]2 , the map w : SL2 (Q)2 →

SL2 (Q) is not surjective.

The assertion is an immediate consequence of results obtained by Jambor,

Liebeck and O’Brien [JLO]. We revisit their proof, adjusting it to our setting,
to use this opportunity for demonstrating the power of the trace method.
This method dates back to classical works of Vogt, Fricke and Klein.
The needed results are quoted below from [Ho] (see [BGrK], [BGG], [BG],
[BGaK] for more details and applications).
Let Fd denote the free group on d generators. For G = SL2 (k) (k is any
commutative ring with 1) and for any u ∈ Fd denote by tr(u) : G d → G the
trace character, (g1 , . . . , gd ) → tr(u(g1 , . . . , gd )).
274 Boris Kunyavskiı̆

Theorem 11.8. [Ho] If w is an arbitrary element of Fd , then the character of

w can be expressed as a polynomial

tr(w) = Pw (t1 , . . . , td , t12 , . . . , t12...d )

with integer coefficients in the 2d − 1 characters ti1 i2 ...iν = tr(xi1 xi2 . . . xiν ),
1 ≤ ν ≤ d, 1 ≤ i 1 < i 2 < · · · < i ν ≤ d.
d −1
Let G = SL2 (k), and let π : G d → A2 be defined by

π(g1 , . . . , gd ) = (t1 , . . . , td , t12 , . . . , t12...d )

in the notation of Theorem 11.8.

Let Z d := π(G d ) ⊂ A2 −1 . Let w : G d → G be a word map. It fol-
d

lows from Theorem 11.8 that for every d there exists a polynomial map
ψ : A2 −1 → A1 such that the following diagram commutes:
d

(11.5)

Moreover, for small d we have a more precise information: one can take
Z 2 = A3 and Z 3 ⊂ A7 an explicitly given hypersurface. This diagram al-
lows one to reduce the study of the image and fibres of w to the corresponding
problems for ψ, which may be much simpler. To ease the notation, we de-
note the coordinates of Z 2 = A3 by (s, t, u). For any prime p we denote by
Pw, p (s, t, u) the reduction of the integer polynomial Pw (s, t, u) modulo p.

Proof of the proposition. We will use diagram (11.5) for k = Q and k = F p .

Corollary 3 of [JLO] implies that for infinitely many primes p the word map
w : SL2 (F p )2 → SL2 (F p ) induced by the word w from the statement of the
proposition is not surjective. More precisely, looking at the proof of this corol-
lary, one can see that for these primes the image of ψ does not contain 0. In
other words, for any such p all values of the polynomial Pw, p are nonzero. Let
P denote the set of all these primes.
Let us show that the image of the map w : SL2 (Q)2 → SL2 (Q) does not
contain matrices with trace zero. Assume to the contrary that there exist A, B ∈
SL2 (Q) such that the trace of w(A, B) equals zero. Let s = tr(A), t = tr(B),
u = tr(AB). From diagram (11.5) we have Pw (s, t, u) = 0. Since the set of
prime divisors of denominators of rational numbers s, t, u is finite and P is
 Equations in matrix groups and algebras 275

infinite, there exists p ∈ P such that the reductions s̄, t¯, ū ∈ F p are well-
defined and Pw, p (s̄, t¯, ū) = 0, contradiction.

11.3.4 Approximation problems

By analogy with the commutative case, along with local–global problems
such as Question 11.6, one can consider various approximation problems. We
restrict ourselves to stating them in the group case.
For a word w = w(x1 , . . . , xd ) and g ∈ G(Q), where G, as above, stands for
a simple linear algebraic Q-group embedded into an affine Q-space A N , denote
by X w,g the affine algebraic variety defined over Q by the system of polyno-
mial equations in A N arising from the matrix equation w(x1 , . . . , xd ) = g.
Assume that for every g ∈ G(Q) the variety X w,g has a Q-point.

Question 11.9. (i) Is it true that for any finite set S of primes p ≤ ∞ and for

every g ∈ G(Q) the set X w,g (Q) is dense in p∈S X w,g (Q p )?
(ii) Does there exist a finite set S0 such that for any finite set S of primes
p ≤ ∞ disjoint from S0 and for every g ∈ G(Q) the set X w,g (Q) is dense

in p∈S X w,g (Q p )?

If Question 11.9(i) (resp. (ii)) is answered in the affirmative, we say that the
word w satisfies weak (resp. very weak) approximation in G.
Although Question 11.9 does not seem to be easy for an arbitrary G,
there is a hope that the trace method may be helpful in the case G = SL2 .
Say, for d = 2 diagram (11.5) then induces a fibration π : X w,g → Yw,g
where the surface Yw,g ⊂ A3Q is given by the equation ψ(s, u, t) =
tr(g). For the commutator word w = [x, y] this approach will hopefully
lead to (at least very) weak approximation in SL2 ; using [Ha1], one can
show that in this case for X w,g the Brauer–Manin obstruction to weak ap-
proximation is the only obstruction. Details will follow in a forthcoming
paper [BK2].

11.3.5 Strong approximation and local–global

principles over integers
Diophantine questions similar to those discussed above can be posed in the
case where G is a group scheme over Z (or, more generally, over the ring of
integers O K of a global field K ) such that its generic fibre is a simple linear
algebraic group over Q (or K ).
276 Boris Kunyavskiı̆

Little is known in such a set-up. Even local obstructions are not completely
understood though some recent results give certain hope for conclusive an-
swers, at least for G = SLn . Say, for the commutator word w = [x, y] the
situation is roughly as follows: obstructions for representing an element of Z p
as a commutator vanish (after excluding necessary congruence obstructions)
as soon as p is sufficiently large with respect to the Lie rank of G (see [GS,
Theorem 5], [AGKS, Theorem 3.5]). Here is an explicit instance of such a phe-
nomenon [AGKS, Theorem 3.8]: if n is a proper divisor of p−1, then every ele-
ment of PSLn (Z p ) is a commutator. The reader is referred to [AGKS] for more
results in this direction, some of those answer the questions raised in [Sh2].
As to the global case, practically nothing is known, even for simplest words
and groups. Here is a tempting question raised by Shalev [Sh2], [Sh3]: is every
element of SLn (Z) (n ≥ 3) a commutator? For n = 2 the answer to a similar
question is obviously negative. However, the author is not aware of any answer
to the following question.

Question 11.10. Let A ∈ SL2 (Z) be representable as a commutator in all

groups SL2 (Z p ). Is A a commutator in SL2 (Z)?

(Apparently, only p = 2 and p = 3 provide nontrivial local obstructions.)

Recent developments in understanding strong approximation for varieties
admitting nice fibrations [CTX], [CTH] give some hope for successful appli-
cation of the trace method, as in approximation problems mentioned above.
One can consider analogues of Question 11.10 for other arithmetically in-
teresting cases, such as Bianchi groups PSL2 (Od ), where Od is the ring
√
of integers of the imaginary quadratic field Q( −d). Higher-dimensional
analogues, such as Picard modular groups SU(n, 1; Od ), look even more
challenging.

11.3.6 Case of simple algebras

In the case where P(X 1 , . . . , X d ) is an associative (or Lie) polynomial, the
author is not aware of any example of failure of analogues of Grunwald–Wang
principles for the map of simple associative (or Lie) algebras Ad → A induced
by P (apart from the power maps discussed above), neither over global fields,
nor over their rings of integers. Some natural cases to be considered are, say,
the following ones:

(i) P = X 12 + · · · + X d2 ;
(ii) additive commutator P = [X 1 , X 2 ] = X 1 X 2 − X 2 X 1 .
 Equations in matrix groups and algebras 277

Some interesting results for case (i) were obtained in [Va1], [KG], [Pu],
however, local–global questions as above seem to remain open. For the addi-
tive commutator in matrix algebras, the induced map A2 → A0 , where A0
denotes the collection of all trace zero elements of A, is surjective in each of
the following cases (see [KBKP], [Sta] for more details):

(i) A = Mn (D), where D is a central division algebra over a field k and

n ≥ 2 [AR];
(ii) for n = 1 this is only known in the case where k is a local field and the
degree of D is prime [Ros];
(iii) A = M2 (R), where R is a principal ideal domain [Li], [Va2], [RR];
(iv) A = Mn (Z) [LR] or, more generally, A = Mn (R), where R is a principal
ideal ring [Sta].

Remark 11.11. Regarding case (iv), for A = Mn (R) where n ≥ 3 and R is

an arbitrary Dedekind ring, the question on the surjectivity of the commutator
map A2 → A0 remains open, see a discussion in [Sta].

Remark 11.12. If the commutator map as above is regarded as a map of Lie

algebras sl(n, R) × sl(n, R) → sl(n, R), the situation becomes more subtle:
it is not clear whether in the Laffey–Reams–Stasinski theorem on representing
any trace zero matrix in the form AB − B A one can choose A and B to be
trace zero matrices. Say, this is not always possible in the case n = 2, R = Z:
if it were the case, it would also be possible in g = sl(2, F2 ), but [g, g] = g.
(I thank N. Gordeev for this remark.)
It is tempting to use the local–global approach to the surjectivity problem
for this map as well. Naturally, similar problems make sense for any Lie poly-
nomial P(X 1 , . . . , X d ), when one can ask about the surjectivity of the induced
map of Lie rings g(R)d → g(R), where g is a Chevalley Lie algebra over Z
and R is the ring of integers of a global field.

Remark 11.13. Perhaps, the first class of polynomials to which one can try to
extend results known for additive commutator are additive Engel polynomials
E n (X 1 , X 2 ) = [[[X 1 , X 2 ], . . . , X 2 ]]. Nothing is known about the surjectivity
of induced maps. Both the associative and Lie case are completely open, even
over fields, even for the simplest polynomial E 2 (X 1 , X 2 ) = [[[X 1 , X 2 ], X 2 ]].
Note, however, that for a somewhat similar polynomial, generalised commuta-
tor P(X 1 , X 2 , X 3 ) = X 1 X 2 X 3 − X 3 X 2 X 1 , the surjectivity of the induced map
A3 → A is known for any division algebra A [GKKL] and for any matrix
algebra A = Mn (R) over a principal ideal domain R [KL].
278 Boris Kunyavskiı̆

Remark 11.14. To conclude, it is worth emphasizing the importance of pur-

suing parallels between matrix equations in groups and algebras, especially in
the Lie algebra case that can be viewed as a bridge between groups and asso-
ciative algebras. Certain considerations in [AGKS] and [BRL] give some hope
for making this into a working tool.

Acknowledgements. The author was supported in part by the Israel Science

Foundation, grant 1207/12, and by the Minerva Foundation through the Emmy
Noether Research Institute for Mathematics; a part of this work was done when
he participated in the trimester program “Arithmetic and Geometry” in the
Hausdorff Research Institute for Mathematics (Bonn).
The author thanks T. Bandman, N. Gordeev, D. Neftin and A. Shalev for use-
ful discussions and correspondence and T. Bauer for providing reference [Sta].

References
[AR] S. A. Amitsur, L. H. Rowen, Elements of reduced trace 0, Israel J. Math.
87 (1994), 161–179.
[AT] E. Artin, J. Tate, Class Field Theory, AMS Chelsea Publishing, Provi-
dence, RI, 2009.
[AGKS] N. Avni, T. Gelander, M. Kassabov, A. Shalev, Word values in p-adic
and adelic groups, Bull. London Math. Soc. 45 (2013), 1323–1330.
[BG] T. Bandman, S. Garion, Surjectivity and equidistribution of the
word x a y b on PSL(2, q) and SL(2, q), Internat. J. Algebra Comput.
22 (2012), no. 2, 1250017, 33 pp.
[BGG] T. Bandman, S. Garion, F. Grunewald, On the surjectivity of Engel words
on PSL(2, q), Groups Geom. Dyn. 6 (2012), 409–439.
[BGaK] T. Bandman, S. Garion, B. Kunyavskiı̆, Equations in simple matrix
groups: algebra, geometry, arithmetic, dynamics, Central European J.
Math. 12 (2014), 175–211.
[BGKP] T. Bandman, N. Gordeev, B. Kunyavskiı̆, E. Plotkin, Equations in simple
Lie algebras, J. Algebra 355 (2012), 67–79.
[BGGKPP1] T. Bandman, G.-M. Greuel, F. Grunewald, B. Kunyavskiı̆, G. Pfister, E.
Plotkin, Two-variable identities for finite solvable groups, C. R. Acad.
Sci. Paris, Sér. A 337 (2003), 581–586.
[BGGKPP2] T. Bandman, G.-M. Greuel, F. Grunewald, B. Kunyavskiı̆, G. Pfister, E.
Plotkin, Identities for finite solvable groups and equations in finite simple
groups, Compos. Math. 142 (2006), 734–764.
[BGrK] T. Bandman, F. Grunewald, B. Kunyavskiı̆ (with an appendix by
N. Jones), Geometry and arithmetic of verbal dynamical systems on
simple groups, Groups Geom. Dyn. 4 (2010), 607–655.
[BK1] T. Bandman, B. Kunyavskiı̆, Criteria for equidistribution of solutions of
word equations on SL(2), J. Algebra 382 (2013), 282–302.
 Equations in matrix groups and algebras 279

[BK2] T. Bandman, B. Kunyavskiı̆, Local-global properties of word varieties,

in preparation.
[Bom1] E. Bombieri (with appendices by A. Odlyzko and D. Hunt), Thompson’s
problem (σ 2 = 3), Invent. Math. 58 (1980), 77–100.
[Bom2] E. Bombieri, Sastry automorphisms, J. Algebra 257 (2002), 222–243.
[Bor] A. Borel, On free subgroups of semisimple groups, Enseign. Math.
29 (1983) 151–164; reproduced in Œuvres - Collected Papers, vol. IV,
Springer-Verlag, Berlin–Heidelberg, 2001, pp. 41–54.
[BS1] A. Borisov, M. Sapir, Polynomial maps over finite fields and residual
finiteness of mapping tori of group endomorphisms, Invent. Math. 160
(2005), 341–356.
[BS2] A. Borisov, M. Sapir, Polynomial maps over p-adics and redisual prop-
erties of mapping tori of group endomorphisms, Intern. Math. Res.
Notices 2009, 3002–3015.
[BDH] M. Borovoi, C. Demarche, D. Harari, Complexes de groupes de type mul-
tiplicatif et groupe de Brauer non ramifié des espaces homogènes, Ann.
Sci. Éc. Norm. Supér. 46 (2013), 651–692.
[BRL] K. Bou-Rabee, M. Larsen, Linear groups with Borel’s property, J. Europ.
Math. Soc., to appear.
[CF] J. W. S. Cassels, E. V. Flynn, Prolegomena to a Middlebrow Arithmetic
of Curves of Genus 2, London Math. Soc. Lecture Note Ser., vol. 230,
Cambridge Univ. Press, Cambridge, 1996.
[Ch1] P. Chatterjee, On the surjectivity of the power maps of algebraic groups
in characteristic zero, Math. Res. Lett. 9 (2002), 741–756.
[Ch2] P. Chatterjee, On the surjectivity of the power maps of semisimple
algebraic groups, Math. Res. Lett. 10 (2003), 625–633.
[Ch3] P. Chatterjee, On the power maps, orders and exponentiality of p-adic
algebraic groups, J. reine angew. Math. 629 (2009), 201–220.
[Ch4] P. Chatterjee, Surjectivity of power maps of real algebraic groups, Adv.
Math. 226 (2011), 4639–4666.
[CTH] J.-L. Colliot-Thélène, D. Harari, Approximation forte en famille, J. reine
angew. Math., to appear, doi:10.1515/crelle-2013-0092.
[CTS] J.-L. Colliot-Thélène, J.-J. Sansuc, Principal homogeneous spaces under
flasque tori: applications, J. Algebra 106 (1987), 148–205.
[CTX] J.-L. Colliot-Thélène, F. Xu, Strong approximation for total space of
certain quadric fibrations, Acta Arith. 157 (2013), 169–199.
[Co] B. Conrad, Lifting global representations with local properties, preprint,
2011, available at http://math.stanford.edu/~conrad/
papers/locchar.pdf.
[Cr1] B. Creutz, A Grunwald–Wang type theorem for abelian varieties, Acta
Arith. 154 (2012), 353–370.
[Cr2] B. Creutz, On the local–global principle for divisibility in the cohomol-
ogy of elliptic curves, arXiv:1305.5881.
[DG] P. Dèbes, N. Ghazi, Galois covers and the Hilbert–Grunwald property,
Ann. Inst. Fourier 62 (2012), 989–1013.
[De] C. Demarche, Groupe de Brauer non ramifié d’espaces homogènes à
stabilisateurs finis, Math. Ann. 346 (2010), 949–968.
280 Boris Kunyavskiı̆

[DZ1] R. Dvornicich, U. Zannier, Local-global divisibility of rational points

in some commutative algebraic groups, Bull. Soc. Math. France
129 (2001), 317–338.
[DZ2] R. Dvornicich, U. Zannier, An analogue for elliptic curves of the
Grunwald–Wang example, C. R. Acad. Sci. Paris, Ser. I 338 (2004),
47–50.
[DZ3] R. Dvornicich, U. Zannier, On a local–global principle for the divisi-
bility of a rational point by a positive integer, Bull. Lond. Math. Soc.
39 (2007), 27–34.
[GS] N. Gordeev, J. Saxl, Products of conjugacy classes in Chevalley groups
over local rings, Algebra i Analiz 17 (2005), no. 2, 96–107; English
transl. in St. Petersburg Math. J. 17 (2006), 285–293.
[GKP] F. Grunewald, B. Kunyavskiı̆, E. Plotkin, Characterization of solvable
groups and solvable radical, Intern. J. Algebra Computation 23 (2013),
1011–1062.
[GKKL] R. N. Gupta, A. Khurana, D. Khurana, T. Y. Lam, Rings over which the
transpose of every invertible matrix is invertible, J. Algebra 322 (2009),
1627–1636.
[Ha1] D. Harari, Flèches de spécialisation en cohomologie étale et applications
arithmétiques, Bull. Soc. Math. France 125 (1997), 143–166.
[Ha2] D. Harari, Quelques propriétés d’approximation reliées à la cohomolo-
gie galoisienne d’un groupe algébrique fini, Bull. Soc. Math. France.
135 (2007), 549–564.
[HSVZ] R. Hazrat, A. Stepanov, N. Vavilov, Z. Zhang, Commutator width in
Chevalley groups, Note Mat. 33 (2013), 139–170.
[Ho] R. D. Horowitz, Characters of free groups represented in the two-
dimensional special linear group, Comm. Pure Appl. Math. 25 (1972),
635–649.
[Il] M. Illengo, Cohomology of integer matrices and local–global divisibility
on the torus, J. Théor. Nombres Bordeaux 20 (2008), 327–334.
[JLO] S. Jambor, M. W. Liebeck, E. A. O Brien, Some word maps that are non-
surjective on infinitely many finite simple groups, Bull. London Math.
Soc. 45 (2013), 907–910.
[KBKP] A. Kanel-Belov, B. Kunyavskiı̆, E. Plotkin, Word equations in sim-
ple groups and polynomial equations in simple algebras, Vestnik St.
Petersburg Univ. Math. 46 (2013), no. 1, 3–13.
[KBMR1] A. Kanel-Belov, S. Malev, L. Rowen, The images of non-commutative
polynomials evaluated on 2 × 2 matrices, Proc. Amer. Math. Soc.
140 (2012), 465–478.
[KBMR2] A. Kanel-Belov, S. Malev, L. Rowen, The images of non-commutative
polynomials evaluated on 3 × 3 matrices, arXiv:1306.4389, to
appear in Proc. Amer. Math. Soc.
[KBMR3] A. Kanel-Belov, S. Malev, L. Rowen, Power-central polynomials on
matrices, arXiv:1310.1598.
[KG] S. A. Katre, A. S. Garge, Matrices over commutative rings as sums of
k-th powers, Proc. Amer. Math. Soc. 141 (2013), 103–113.
 Equations in matrix groups and algebras 281

[KL] D. Khurana, T. Y. Lam, Generalized commutators in matrix rings, Linear

Multilin. Algebra 60 (2012), 797–827.
[LR] T. J. Laffey, R. Reams, Integral similarity and commutators of integral
matrices, Linear Algebra Appl. 197/198 (1994), 671–689.
[LOST] M. W. Liebeck, E. A. O Brien, A. Shalev, P. H. Tiep, The Ore conjecture,
J. Europ. Math. Soc. 12 (2010), 939–1008.
[Li] D. Lissner, Matrices over polynomial rings, Trans. Amer. Math. Soc.
98 (1961), 285–305.
[LA] G. Lucchini Arteche, Groupe de Brauer non ramifié des espaces ho-
mogènes à stabilisateur fini, J. of Algebra 411 (2014), 129–181.
[Male] S. Malev, The images of non-commutative polynomials evaluated on 2×2
matrices over an arbitrary field, J. Alg. Appl. 13 (2014), no. 6, 1450004,
12pp.
[Mall] G. Malle, The proof of Ore’s conjecture (after Ellers–Gordeev and
Liebeck–O’Brien–Shalev–Tiep). Astérisque 361 (2014), exp. 1069,
325–348.
[Man] Yu. I. Manin, Cubic Forms: Algebra, Geometry, Arithmetic, Nauka,
Moscow, 1972; English transl., 2nd ed., North-Holland, Amsterdam,
1986.
[Ne] D. Neftin, The tame and wild Grunwald problems, preprint available at
http://www-personal.umich.edu/~neftin/manu.htm.
[NSW] J. Neukirch, A. Schmidt, K. Wingberg, Cohomology of Number Fields,
Grundlehren Math. Wiss., vol. 323, Springer-Verlag, Berlin, 2000.
[Pa1] L. Paladino, Local–global divisibility by 4 in elliptic curves defined over
Q, Ann. Mat. Pura Appl. (4) 189 (2010), 17–23.
[Pa2] L. Paladino, Elliptic curves with Q(E[3]) = Q(ζ3 ) and counterexamples
to local–global divisibility by 9, J. Théor. Nombres Bordeaux 22 (2010),
139–160.
[Pa3] L. Paladino, On counterexamples to local–global divisibility in commu-
tative algebraic groups, Acta Arith. 148 (2011), 21–29.
[PRV] L. Paladino, G. Ranieri, E. Viada, On local–global divisibility by p n in
elliptic curves, Bull. Lond. Math. Soc. 44 (2012), 789–802.
[Pr] G. Prasad, Elementary proof of a theorem of Bruhat–Tits–Rousseau and
of a theorem of Tits, Bull. Soc. Math. France 110 (1982), 197–202.
[Pu] S. Pumplün, Sums of d-th powers in non-commutative rings, Beiträge
Algebra Geom. 48 (2007), 291–301.
[Roq] P. Roquette, The Brauer–Hasse–Noether theorem in historical perspec-
tive, Schriften der Mathematisch-Naturwissenschaftlichen Klasse der
Heidelberger Akademie der Wissenschaften, vol. 15, Springer-Verlag,
Berlin–New York, 2005.
[Ros] M. Rosset, Elements of trace zero and commutators, Ph.D. Thesis, Bar-
Ilan Univ., 1997.
[RR] M. Rosset, S. Rosset, Elements of trace zero that are not commutators,
Comm. Algebra 28 (2000), 3059–3072.
[Sal1] D. J. Saltman, Generic Galois extensions and problems in field theory,
Adv. Math. 43 (1982), 250–283.
282 Boris Kunyavskiı̆

[Sal2] D. J. Saltman, Retract rational fields and cyclic Galois extensions, Israel
J. Math. 47 (1984), 165–215.
[San] J.-J. Sansuc, Groupe de Brauer et arithmétique des groupes algébriques
linéaires sur un corps de nombres, J. reine angew. Math. 327 (1981),
12–80.
[Sh1] A. Shalev, Commutators, words, conjugacy classes and character meth-
ods, Turkish J. Math. 31 (2007), 131–148.
[Sh2] A. Shalev, Applications of some zeta functions in group theory, in:
“Zeta Functions in Algebra and Geometry” (A. Campillo et al., Eds.),
Contemp. Math., vol. 566, Amer. Math. Soc., Providence, RI, 2012,
pp. 331–344.
[Sh3] A. Shalev, Some results and problems in the theory of word maps, in:
“Erdös Centennial” (L. Lovász, I. Ruzsa, V. T. Sós, Eds.), Bolyai Soc.
Math. Studies, vol. 25, Springer, 2013, pp. 611–649.
[So] C. M. Sorensen, A proof of the Breuil–Schneider conjecture in the
indecomposable case, Ann. Math. 177 (2013), 367–382.
[Sta] A. Stasinski, Similarity and commutators of matrices over principal
ideal rings, arXiv:1211.6872, to appear in Trans. Amer. Math. Soc.
[Ste] R. Steinberg, On power maps in algebraic groups, Math. Res. Lett.
10 (2003), 621–624.
[Sw] R. G. Swan, Noether’s problem in Galois theory, in: “Emmy Noether in
Bryn Mawr” (B. Srinivasan, J. Sally, Eds.), Springer, New York–Berlin,
1983, pp. 21–40.
[Sp] Š. Špenko, On the image of a noncommutative polynomial, J. Algebra
377 (2013), 298–311.
[Th] A. Thom, Convergent sequences in discrete groups, Canad. Math. Bull.
56 (2013), 424–433.
[Va1] L. N. Vaserstein, On the sum of powers of matrices, Linear Multilin.
Algebra 21 (1987), 261–270.
[Va2] L. N. Vaserstein, Noncommutative number theory, in: “Algebraic
K-theory and Algebraic Number Theory (Honolulu, HI, 1987)” (M. R.
Stein, R. K. Dennis, Eds.), Contemp. Math., vol. 83, Amer. Math. Soc.,
Providence, RI, 1989, pp. 445–449.
[Vo1] V. E. Voskresenskiı̆, On the question of the structure of the subfield of
invariants of a cyclic group of automorphisms of the field Q(x 1 , . . . x n ),
Izv. Akad. Nauk SSSR Ser. Mat. 34 (1970), 366–375; English transl. in
Math. USSR Izv. 4 (1970), 371–380.
[Vo2] V. E. Voskresenskiı̆, Algebraic Tori, Nauka, Moscow, 1977. (Russian.)
[Wo] S. Wong, Power residues on abelian varieties, Manuscripta Math.
102 (2000), 129–138.
 12
On the -adic regulator as an ingredient of
Iwasawa theory
L. V. Kuz’min
NRC “Kurchatov Institute”
E-mail address: [email protected]

Contents
12.1 Introduction 283
12.2 The main definitions and conjectures 287
12.3 Local and semi-local duality 291
12.4 Duality results for non-Abelian local fields 295
12.5 Explicit formulae for the norm residue symbol 303
12.6 Applications to the feeble conjecture on the -adic regulator 307
12.7 An Analog of the Riemann-Hurwitz formula for some
-extensions of algebraic number fields 310
12.8 On a new type of the -adic regulator 312
References 316

12.1 Introduction
A starting-point of the Iwasawa theory is an analogy between algebraic number
fields and fields of algebraic functions in one indeterminate over a finite field
of constants [3]. Let X be a curve over a finite field Fq of q = p n elements,
K = k(X ) the function field of X . For a given prime  = p let T (K ) be
the Galois group of L/(F̄q K ), where L is the maximal Abelian unramified
-extension of F̄q K . The Frobenius automorphism ϕ ∈ G(F̄q /Fq ) acts on
T (K ) ⊗Z Q hence we have a characteristic polynomial P(X ) ∈ Z[X ] of
this action and the system of its eigenvalues {αi }, 1  i  2g, where g is
the genus of X . This polynomial is equal to the numerator of the zeta-function
√
of X . The eigenvalues αi satisfy the Riemann conjecture, that is, |αi | = q.
Moreover, there is a skew-symmetric Weil scalar product on T (K ).

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

283
284 L. V. Kuz’min

The Iwasawa theory tries to generalize all these results to algebraic number
fields. For such a field K one defines K ∞ to be the cyclotomic Z -extension
of K . Then, as in the functional case, one defines L to be the maximal Abelian
unramified -extension of K ∞ and T̄ (K ) = T̄ (K ∞ ) = G(L/K ∞ ). So T̄ (K )
is a -module, where  = Z [[ ]] is a completed group ring of and
= G(K ∞ /K ). This module is known as the Tate module of K or the Iwa-
sawa module. Consider the simplest case when K is a cyclotomic field. In this
case one has T̄ (K ) = T̄ + (K ) ⊕ T̄− (K ) corresponding to the action of the
automorphism of complex conjugation j. Then there is the so called “the main
conjecture of Iwasawa theory” [5, 20] (which is stated for cyclotomic fields
and in some other cases) that expresses an action of on T̄− (K ) in terms of
some functions (Iwasawa series), which may be defined in terms of the -adic
L-functions.
By the analogy with the functional case one can expect that, under assump-
tion that K contains an -th primitive root of unity, there exists an analog of
Weil scalar product on T̄ (K ). But one can see easily that it is impossible.
Indeed, this product must be compatible with the action of the Galois group on
T̄ (K ) and hence it must induce a non-degenerate product T̄ − (K )× T̄ + (K ) →
V , where V is the projective limit of -primary roots of unity in K ∞ . But the
group T̄− (K ) is infinite (if it is non-zero), whereas T̄+ (K ) is supposed to be
finite by the Greenberg conjecture [2].
We can resume the situation as follows. The group T̄− (K ) is a right object
but it is only one half of the right Iwasawa module for an algebraic number
field. In [8] the author undertook an attempt to define a good Iwasawa module
for a C M-field K . (It was denoted by A (K ) in that paper). In brief, the def-
inition is as follows. Let K n be an n-layer of K ∞ /K , that is, [K n : K ] = n ,
and Mn the maximal Abelian -extension of K n unramified outside . By class
field theory the Galois group X n = G(Mm /K n ) enters the exact sequence
α  ×
U S (K n )[] −→ K n,v [] −→ X n −→ Cl ˜  (K n ) −→ 1, (12.1)
v|

where K n,v is the completion of K n relative to the place v, U S (K n ) the group

of S-units in K n (S is the set of all v|), symbol [] means pro--completion
and Cl ˜  (K n ) is a factor of -class group of K n by the subgroup generated by
all prime divisors in S.
For a local field K n,v we put H̃ (K v ) = (K n,v × /μ(K
n,v )[]) and H̃ (K n ) =

v| H̃ (K n,v ). Let Ũ S (K n ) be the image of the natural map U S (K n )[] →
H̃ (K n ) induced by α in (12.1). Then (12.1) induces an exact sequence (in
additive notations)
˜  (K n ) −→ 0.
0 −→ Ũ S (K n ) + H̃ (K n )− −→ H̃ (K n ) −→ Yn −→ Cl
 On the -adic regulator as an ingredient of Iwasawa theory 285

Thus Yn is a factor module of X n , and by Galois theory there is the unique

subfield L n ⊆ Mn such that the Galois group G(L n /K n ) equals Yn . We put
L = ∪n L n and A (K ) = G(L/K ∞ ) = lim Yn .
← −
In [8] we had studied the -module A (K ) for a C M-field K . The main
result of that paper was a construction of an inner skew-symmetric product

A (K ) × A (K ) −→ V , V = lim μ (K n ),
←−
that is completely analogous to Weil scalar product.
The module A (K ) is -torsion and, assuming that the Iwasawa μ-invariant
μ (K ) vanishes, we see that A (K ) is a free Z -module of finite rank, which
we denote by λ(A (K )). In this case we have the following theorem, which is
an analog of Riemann-Hurwitz formula.

Theorem 12.1. [8, Prop. 5] Let K be a field of C M-type containing a primi-

tive -th root of unity and let K  be a field of C M-type, which is an -extension
of K . Suppose that μ (K ) = 0. Then one has

λ(A (K  )) − 2 = [K ∞

: K ∞ ](λ( A (K )) − 2) + (ep − 1),
p∈S1

where S1 consists of all places of the field K ∞  which split completely in the

extension K /K and either are divisors of  or ramify in the extension K  /K .

+

If p is a place, ep denotes the local degree of the extension K ∞  /K at p, and

∞
 /K .
for p  , ep is also equal to the ramification index of p in K ∞ ∞

So there is a problem to generalize the construction of A (K ) to the case of

an arbitrary algebraic number field K . This problem yields a number of inter-
esting conjectures and results, which will be the theme of this paper. Because
of lack of the space we will omit full proofs of presented results. Instead of this
we will explain our goals and the problems we meet with. Note that another
approach to this problem was suggested in [22]. We shall not discuss it here,
since it is based on absolutely different ideas.
Our paper is organized as follows.
In Section 12.2 we define the -adic regulator R (K ) of an algebraic num-
ber field K and the relative -adic regulator R (K /k) of a finite extension of
algebraic number fields and formulate our main conjectures: the conjecture on
the -adic regulator (Conjectures 12.3) and the feeble conjecture on the -adic
regulator (Conjecture 12.6). Then we state the first results on this conjectures.
These conjectures follow from the -adic Schanuel conjecture (see Conjec-
ture 12.2 at the end of this section) and hence are very probable. Nevertheless,
it is interesting to give an unconditional proof of them in some particular cases.
286 L. V. Kuz’min

In brief, an analog of H + (K n ) is well defined in any case, and we are going to

define H̃ − (K n ) for an arbitrary K as an orthogonal complement to H̃ + (K n )
relative to the scalar product induced by local scalar products
U (K n,v ) × U (K n,v ) → Q , (x, y) → Sp K n /Q (log x · log y), (12.2)
where U (K n,v ) is the group of units of K n,v and log means the -adic
logarithm.
In Section 12.3 we introduce some tools that will be used in sequel. Namely,
we define a Galois module V (K n,v ) that is close to U (K n,v ) and, assuming that
K n,v is Abelian, define an inner product of the form V (K ∞,v ) × V (K ∞,v ) →
 p , where  p = Z [[G(K ∞ /K p )]], V (K ∞,v ) = lim V (K n,v ) and the limit
← −
is taken with respect to the norm maps (12.23). Let K be an algebraic number
field Abelian over . This means that the completions K v are Abelian for all
 
v|. Then we put V (K ) = v| V (K v ) and V (K ∞ ) = v| V (K ∞,v ). If p
is sufficiently large then the products (12.23) for all v| yields a semi-local
product T∞ : V (K ∞ ) × V (K ∞ ) →  p (see (12.25)). Thus we see immedi-
ately that there exists an orthogonal complement V − (K ∞ ) of the  p -module
V + (K ∞ ) = V (K ∞ ) ∩ H + (K ∞ ). We show (Theor. 12.7) that Conjecture 12.6
is equivalent to the assertion that V + (K ∞ ) ∩ V − (K ∞ ) = 0.
In Section 12.4 we generalize the results of preceding section to the case of
non-Abelian local fields. Using some complicated techniques (Laurent series
with infinite principal part), we state the results (Theor. 12.10) analogous but
more feeble than those of Section 12.3. Nevertheless these results are sufficient
for application to Conjecture 12.6 that we give in Section 12.6.
In Section 12.5 we consider a relation between the product (12.25) and the
norm residue symbol. Using this last symbol man can define an inner product
 ,  : V (K ∞ ) × V (K ∞ ) →  p , which looks as (12.25). So, if K is Abelian
over , then there is an isomorphism ϕ : V (K ∞ ) → V (K ∞ ) such that x, y =
T∞ (x, ϕ(y)) for any x, y ∈ V (K ∞ ). Since V + (K ∞ ) is a maximal isotropic
submodule of V (K ∞ ), we obtain that V + (K ∞ ) ∼ = V − (K ∞ ) as H -modules
for any finite group of automorphisms H (Prop. 5.1).
In Section 12.6 we use the obtained results to prove Conjecture 12.3 in
some cases. We consider a finite -extension of algebraic number fields K /k.
Assuming that Conjecture 12.6 holds for k and k satisfies some other natural
conditions, we prove that Conjecture 12.6 holds for K as well.For example, if
k is Abelian then Conjecture 12.6 holds for any -extension of k. Moreover, if
K is Abelian over  then we obtain a bit more (see Theorems 12.15, 12.16).
In Section 12.7 we study some other aspects of the situation. Firstly, we
prove that Theor. 12.1 can be generalized to the case when K /k is a finite
-extension of algebraic number fields, k contains a primitive -th root of unity
 On the -adic regulator as an ingredient of Iwasawa theory 287

if K is Abelian over  and k satisfies the conditions of Theorems 12.15, 12.16

(see Theor. 12.21, 12.22). Note that there appears some new module D(K ∞ )
(D  (K ∞ ) if  = 2) , which is zero in C M-case.
In Section 12.8 we introduce a new notion of the -adic regulator R (K ),
assuming that  splits completely in K . In distinction with R (K ), this last
regulator is defined not in terms of the -adic logarithms but in terms of divi-
sors. Conjecture 12.2 yields that R (K ) = 0 for any K . Moreover, we give
an unconditional proof of this last inequality for some Abelian extensions of
imaginary quadratic field.
In conclusion we will state a particular case of the -adic Schanuel conjec-
ture that we shall use in this paper.

Conjecture 12.2. Let x1 , . . . , xn ∈ Q̄×

 be algebraic over Q and , x 1 , . . . , x n
are multiplicatively independent. Then the -adic logarithms log x1 , . . . , log xn
are algebraically independent over Q.

A part of results of this last section was obtained while the author was a guest
of the Trimester Program “Arithmetic and Geometry”, Hausdorff research
institute for mathematics, Bonn. The author is pleased to thank the collabo-
rators of the institute for hospitality and the participants of the Trimester for
stimulating communion.

12.2 The main definitions and conjectures

Now let K be any algebraic number field. In this case there is the cyclotomic
Z -extension K ∞ /K and a Galois module H (K ∞ ). This last module may be
defined in the same manner as in the case of CM-fields. Namely, if K ∞ =
∪n K n , where [K n : K ] = n then we put H (K ∞ ) = lim H̃ (K n ), where
 ← −
× []/μ(K
H̃ (K n ) = v| K n,v n,v ) and the limit is taken with respect to the
norm maps. Since there are only finitely many places v| in K ∞ , we can also

write H (K ∞ ) = v| H (K ∞,v ), where v runs through all places over  in
K ∞ and H (K ∞,v ) = lim K n,v× []/μ(K
n,v ). If K is a CM-field then there is
← −
the automorphism of complex conjugation j and for  = 2 and any Z -module
A one has a decomposition A = A+ ⊕ A− , where A+ = e+ A and A− = e− A,
and e+ = (1 + j)/2, e− = (1 − j )/2. If  = 2 then we put A+ = (1 + j)A,
and A− = (1 − j)A. Thus, if  = 2 and A has no 2-torsion then A/( A+ ⊕ A− )
has period 2.
Our goal here is to generalize this decomposition to the case of any field,
that is, to define Galois submodules H + (K ∞ ), H − (K ∞ ) ⊆ H (K ∞ ), whose
properties are analogous to those of H + (K ∞ ) and H − (K ∞ ) for CM-fields.
288 L. V. Kuz’min

There is quite natural way to define H + (K ∞ ). Indeed, put Ū (K ∞ ) =

lim(U (K n )[]/μ(K n )). Then Ū (K ∞ ) ⊆ H (K ∞ ) and Ū (K ∞ ) is a -module
← −
of rank r1 + r2 (see [7, Theorem 7.2]), where r1 and r2 is the number of
real and complex places of K , respectively. Then we define H + (K ∞ ) as
the minimal -submodule of H (K ∞ ) such that Ū (K ∞ ) ⊆ H + (K ∞ ) and
H (K ∞ )/H + (K ∞ ) has no torsion.
The problem of definition of H − (K ∞ ) is much more complicated. It seems
convenient to introduce some product on H (K ∞ ), which is non-degenerate,
and such that its restriction to H + (K ∞ ) is non-degenerate as well. Then man
can define H − (K ∞ ) as an orthogonal complement of H + (K ∞ ) with respect to
this product. This is a general idea, and to realize it we have to overcome many
difficulties. Moreover, we can realize this idea completely only under some
additional assumptions. Nevertheless, this idea leads us to many interesting
conjectures, phenomena and results.
To start with, for a -extension K ∞ /K of algebraic number fields we put

Ã(K n ) = v| U (K n,v )/μ(K n,v ) and Ā(K ∞ ) = lim Ã(K n ), where the limit
← −
is taken with respect to the norm maps. We wish to define our product on
Ā(K ∞ ), but at first we have to settle the situation for a finite level. Note that
we have a product on the Galois module Ã(K ). Indeed, for a local field K v we
have a product

( , )v : U (K v )/μ(K v ) × U (K v )/μ(K v ) −→ Q , (12.3)

where (x, y)v = Sp K v /Q (log x log y) and log : U (K v ) → K v is the -adic
logarithm in the sense of Iwasawa. At first glance the product (12.3) is far
from being good, though it is non-degenerate. We shall study its properties
more carefully in the next section.
 
If x, y ∈ Ã(K ) then x = v| xv , y = v| y , where we have xv , yv ∈
U (K v )/μ(K v ), and we can put

S(x, y) = (xv , yv )v ∈ Q . (12.4)

v|

Note that the sum of local traces is a global one, thus, one has S(x, y) =

Sp K /Q (log x log y), where log means a homomorphism Ã(K ) → v| K v ,
which coincides with the -adic logarithm log on each v-component, and

Sp K /Q denotes the map v| K v ∼ = K ⊗Q Q → Q induced by the
trace map from K to Q. So we obtain a non-degenerate product of the form
S : Ã(K ) × Ã(K ) → Q .
There is a natural map α : U (K )[] → Ã(K ), and it is natural to assume
that S rests non-degenerate after restriction to its image. Thus we come to the
following conjecture.
 On the -adic regulator as an ingredient of Iwasawa theory 289

Conjecture 12.3. (Conjecture on the -adic regulator.)[9] Let (K , ) be

an algebraic number field and a rational prime number respectively. Let
ε1 , . . . , εr be a system of fundamental units of K and ε0 = 1 +  (ε0 = 5
if  = 2). Define the -adic regulator of K by

R (K ) = det(S(εi , ε j )) ∈ Q , 0  i, j  r. (12.5)

Then for any (K , ) one has R (K ) = 0.

Note that R (K ) does not depend on the choice of the basis in U (K ), since
it is defined up to the square of the determinant of transition matrix, which is
unimodular. We note also that for a totally real K the regulator R (K ) in (12.5)
is almost a square of the Leopoldt -adic regulator R (K ). More precisely, man
has a relation [17, Proposition 2.1]

R (K ) = (log ε0 )2 [K : Q]2 R (K )2 .

Thus non-vanishing of R (K ) for a totally real field is equivalent to the valid-

ity of Leopoldt conjecture. In general, Conjecture 12.3 is essentially stronger.
It asserts that ε0 , . . . , εr generate a submodule in Ã(K ) of Z -rank r + 1
and, moreover, the product (12.4) is non-degenerate being restricted to this
submodule.

Theorem 12.4. Conjecture 12.3 is a consequence of Conjecture 12.2.

We can assume that K is Galois over Q and contains an imaginary quadratic

field k. Then there exists an Artin unit ε in U (K ). This means that τ (ε) = ε,
where τ is an automorphism of complex conjugation in K , and the elements
h(ε), h ∈ H = G(K /k), generate a subgroup of finite index in U (K ).
Then, putting ε = ε0 ε and Th = log(h(ε )) for h ∈ H , we see that R (K )
is a non-zero polynomial in Th for h ∈ H of degree 2r with coefficients in Z.
Since the elements h(ε ), h ∈ H , are multiplicatively independent and hence
their logarithms are algebraically independent over Q by Conjecture 12.2, we
get R (K ) = 0. This proves the theorem.
There is one particular case, when Conjecture 12.3 can be proved uncondi-
tionally.

Theorem 12.5. [9, Theorem 3.1] Let K be an Abelian extension of an imagi-

nary quadratic field k. Suppose that K is Galois over Q and the only non-trivial
automorphism τ ∈ G(k/Q) acts on the Galois group H = G(K /k) by the rule
τ (h) = τ hτ −1 = h −1 for any h ∈ H . Then R (K ) = 0 for any prime .
290 L. V. Kuz’min

Proof. For any h i , h j ∈ H one has

τ (Sp K /k (log h i (ε  ) log h j (ε  )) = Sp K /k (log τ h i (ε ) log τ h j (ε ))

= h i hj Sp K/k (h i hj (log h i−1(ε ) log h −1   
j (ε ))) = SpK/k (log h j(ε ) log h i(ε )).

Thus, one has Sp K /Q (log u 1 log u 2 ) = 2Sp K /k (log u 1 log u 2 ) for any
u 1 , u 2 ∈ U (K ), and we can finish the proof by the same arguments, which
used Brumer in his proof of Leopoldt conjecture for Abelian fields [1].
Our definition of the -adic regulator allows us to introduce an impor-
tant notion of the relative -adic regulator. Let K /k be a finite extension of
algebraic number fields. Let U (K /k) be the relative group of units of K /k,
that is,

U (K /k) = { x ∈ U (K ) | N K /k (x) ∈ μ(k) }, (12.6)

where N K /k is the norm map from K to k.

Definition. Let u 1 , . . . , u t be some basis of U (K /k). Then the -adic regula-

tor R (K /k) is defined by

R (K /k) = det(Sp K /Q (log u i log u j )) ∈ Q , 1  i, j  t.

For a finite extension of algebraic number fields K /k and prime  the con-
dition R (K /k) = 0 means that α induces an injection (U (K /k)/μ )[] →
Ã(K ) an the product (12.4) is non-degenerate on the image of this injection.
Let K /k be an extension of algebraic number fields. Then R (K ) = 0
if and only if R (k) = 0 and R (K /k) = 0. Moreover, there is an
explicitly computable non-zero constant c(K /k) ∈ Q such that R (K ) =
c(K /k)R (k)R (K /k) [17, Prop. 2.3](see also Sect. 7).

Conjecture 12.6. Feeble conjecture on the -adic regulator. Let K be an

algebraic number field and K ∞ the cyclotomic Z -extension of K . Then there
is an index n 0 , which depends only on K and , such that for any m > n  n 0
man has R (K m /K n ) = 0.

This conjecture is in the same position to Conjecture 12.3 as the feeble

Leopoldt conjecture to Leopoldt conjecture. The feeble Leopoldt conjecture
is, in fact, a theorem, whereas Conjecture 12.6 is stated only for some partic-
ular cases (see Sections 12.6 and 12.8). Conjecture 12.6 is more tractable than
Conjecture 12.3, and in the next section we discuss some approach to it.
 On the -adic regulator as an ingredient of Iwasawa theory 291

12.3 Local and semi-local duality

Let k be a finite cyclotomic extension of Q that contains a primitive root
of unity of degree  (of degree 4 if  = 2). By ζn we denote a primitive
root of unity of degree n+1 (of degree 2n+2 for  = 2). By ξr we denote a
primitive root of unity of degree r for some r prime to . Thus, k is of the
form k = Q (ζn 0 , ξr ) for n 0  0, and H := Q (ξr ) is the maximal unramified
subfield of k. Put kn = k(ζn ). Then kn = k for n  n 0 , [kn : k] = n−n 0 for
n > n 0 , and k∞ := ∪n kn is the cyclotomic Z -extension of k. If A is a full
lattice in kn then we put
A⊥ = { y ∈ kn | Spkn /Q (x y) ∈ Z for all x ∈ A }.
Our first goal here is to determine the dual lattice Bn := (log U (kn ))⊥ , where
U (kn ) is the group of units of k.
Note that this group was determined by Iwasawa in [4], and this remarkable
result was a starting point for our paper [10].
Let O H be the ring of integers of H , [H : Q ] = f , and ϕ the Frobenius
automorphism of H . Let α, β ∈ O H form integer normal dual bases of H . It
means that

1 if i = 0,
Sp H/Q (ϕ i (α) · β) = (12.7)
0 if i = 1, . . . , f − 1.
Put
m−1
#n (β, i) = qn−1 ζn−i ϕ s+1−m (β),
s
(12.8)
s=0

where m = n + 1 if  = 2 (m = n + 2 if  = 2) and qn = m .
Generalizing the results of Iwasawa, we had proved in [10, Prop. 1.2] that
for any indices i, i 1 , i 2 prime to  we have
n (β, i 1 , i 2 ) : = #n (β, i 1 ) − #n (β, i 2 ) ∈ Bn , (12.9)
n (β, i) : = #n (ϕ(β), i ) − #n (β, i ) − qn−1 ϕ(β) ∈ Bn . (12.10)
By M(kn ) (resp. by X(kn )) we denote an additive Galois Z -submodule of kn
generated by #n (β, 1) (resp. by all n (β, i 1 , i 2 ) and n (β, i)).
Thus, M(kn ) is a cyclic free Z [G(kn /Q )]-module, which is generated by
any of two elements #n (α, 1) and #n (β, −1), where α and β are as in (12.7).
We wish to introduce a construction that unites the products (17.10) for all n.
To do this, we at first define this construction for M(kn ), then for M(kn )⊥ ,
and at last for X(kn )⊥ . Note that X(kn )⊥ is close to the group log U (kn ).
√
Indeed, let u n ∈ U (kn ) be an element such that kn ( qn u n ) is the unramified
292 L. V. Kuz’min

extension of kn of degree qn . We denote by V (kn ) a multiplicative group gen-

√
erated by U (kn )/μ(kn ) and qn u n . Then V (kn ) is a G(kn /Q )-module and
V (kn )/(U (kn )/μ(kn )) ∼
= μ (kn ).
To construct our product, we put for any x, y ∈ kn

Tn (x, y) = Spkn /Q (σ −1 (x) · y)σ ∈ Q [G(kn /Q )]. (12.11)

σ ∈G(kn /Q )

One can easily check that (12.11) is compatible with the trace maps, that is, for
any m > n there is a commutative diagram

(12.12)

where Spm,n is the trace from km to kn and πm,n is the natural projection of the
group rings.
Now we assume that the roots of unity ζn ∈ kn are coherent for all n, that is,

ζn+1 = ζn . Then, passing to projective limits in (12.12), we obtain a product

˜ := Q [[G(k∞ /Q )]],

T∞ : k̃∞ × k̃∞ −→  (12.13)

where k̃∞ = lim kn and  ˜ = lim Q [G(kn /Q )]. This product is non-degene-
← − ←−
rate and sesquilinear in the sense that
˜
T∞ (λx, μy) = λT∞ (x, y)μ∗ for any λ, μ ∈ , (12.14)

where ∗ means an anti-automorphism of  ˜ induced by the map σ → σ −1 for

any σ ∈ G(k∞ /Q ).
Now, restricting (12.13) to the Galois module M(k∞ ) := lim M(kn ), we
←−
obtain a product of the form
˜
M(k∞ ) × M(k∞ ) −→ . (12.15)

Note that, under our choice of the roots ζn , the generators #n (α, 1) and
#n (β, −1) behave well with respect to the trace maps. To be precise,
the sequences {#n (ϕ −n (α), 1)} and {#n (ϕ −n (β), −1)} are compatible with
respect to the trace maps, and we can put #∞ = lim #n (ϕ −n (α), 1), #∞ =
←−
lim #n (ϕ −n (β), −1).
← −
By next step we can compute explicitly the element +∞ := T∞ (#∞ , #∞ ).
Indeed, we have +∞ = lim +n , where +n = Tn (#n (ϕ −n (α), 1),
←−
#n (β, −1)) = (#n (α, 1), #n (β, −1)). The calculation of this last element
 On the -adic regulator as an ingredient of Iwasawa theory 293

reduces to calculation of the traces of different powers of ζn . These calcu-

lations were carried out in [10, Prop. 3.1]. It was stated there that +n =
−−(n+2) SpG n + n if  = 2 and +n = −1 n if  = 2, where G n =
G(kn /H ) and
= n−1
>
−(n+2)

n =   Spn,0 +  Spn,r+1 − Spn,r . (12.16)
r=0

In this formula Spn,i means Spkm /ki = σ ∈G(kn /ki ) σ , and we put ki = H (ζi )
for any i  0.
Our next goal is to pass from M(kn ) to M(kn )⊥ . To do this, we note that
the element +n is invertible in Q [G(kn /Q )] for any n, whereas the elements
+−1
n can be given explicitly. Namely, we have [10, Prop. 3.2]
n
n−1 = n+1 − ( − 1)n+1−2r Spn,n−r (12.17)
r=1

and for  = 2
+−1 −1
n = n + 
1−n
SpG n .
Then, applying an analog of (12.14) to the product Tn , we obtain immediately
Tn (+−1  −1
n (#(kn )), # (kn )) = +n +n = 1. In other words, this means that
M(kn ) = +n M(kn ) and M(kn )⊥ is a free cyclic Galois module over the
⊥ −1

group ring Z [G(kn /Q )], which may be generated by any of two elements
ξ(kn ) = +−1  −1 
n (#(k n )) and ξ (kn ) = +n (# (k n )) [10, Prop.3.3]. Using explicit
representations (12.8) and (12.17), we obtain
m−1
−s ζn− ϕ s−m+1 (α),
s
ξ(kn ) = δα + (12.18)
s=0
m−1
ξ  (kn ) = δβ + −s ζn ϕ s−m+1 (β),
s
(12.19)
s=0

where δ = − 1−m if  = 2 and δ = 0 if  = 2. The elements ξ(kn ) and ξ  (kn )

are compatible for different n with respect to the trace maps and, passing to
inverse limits, we obtain two elements ξ(k∞ ) := lim ξ(kn ) and ξ  (k∞ ) :=
← −
lim ξ  (kn ), each of which generates a free Z [[(G(k∞ /Q )]]-module
← −
M(k∞ )⊥ = lim M(kn )⊥ . Moreover, we have T∞ (ξ(k∞ , ξ  (k∞ )) = +−1∞.
←−
⊥
If x, y ∈ M(k∞ ) then there are the unique presentations x = λξ(k∞ ), y =
μξ  (k∞ ) for some λ, μ ∈  := Z [[G(k∞ /Q )]]. Thus, T∞ (x, y) =
λ+−1 μ∗ = λμ ∗ +−1 , since the Galois group G(k∞ /Q ) is Abelian.
So, by definition the product T∞ takes its values in the very large ring
294 L. V. Kuz’min

Q [[G(k∞ /Q )]]. But in fact these values belong to a free -module of rank
˜ −1 .
1, namely to +
Fix some index p > 0. For any n  p there is a natural
restriction map resn : Q [G(kn /Q )] → Q [G(kn /k p )], where we put
 
resn ( σ ∈G(kn /Q ) aσ σ ) = σ ∈G(kn /k p ) aσ σ . Passing to inverse limit, we get a
map res∞ : Q [[G(k∞ )]] → Q [[G(k∞ /k p )]]. Then, taking into account that
(γ p − 1)+−1 −1
n ∈ Q [G(kn /k p )], hence (γ p − 1)+∞ ∈ Q [[G(k∞ /k p )]], and
putting T̄∞ (x, y) = res∞ (T∞ (x, y)), we get an inner product [10, Prop. 3.5]

T̄∞ : (γ p − 1)M(k∞ )⊥ × M(k∞ )⊥ −→ (γ p − 1)B1 , (12.20)

where (γ p − 1)B1 = (γ p − 1) p +−1∞ and  p := Z [[G(k∞ /k p )]] is the ring

of formal power series in T p = γ p − 1.
Since  p is an integral domain, the multiplication by (γ p − 1) induces an
isomorphism (γ p − 1) :  p → (γ p − 1) p . So, we can modify (12.20) once
more, putting T̃∞ (x, y) = h ∈  p if T̄(x, y) = (γ p − 1)h+−1∞ . In such a way
we obtain an inner product

T̃∞ : M(k∞ )⊥ × M(k∞ )⊥ −→  p . (12.21)

Next step is passing from M(kn )⊥ to X(kn )⊥ = log V (kn ). We prove [10,
Theor. 2.1] that (σ − 1)X(kn )⊥ = (σ − 1)M(kn )⊥ for any σ ∈ G n =
G(kn /H ). So, passing to the limit, we get an isomorphism (γ p − 1)X(k∞ )⊥ =
(γ p − 1)M(k∞ )⊥ . Since (12.21), in fact, depends only on the elements
(γ p − 1)x, (γ p − 1)y, which are the elements of (γ p − 1)X(k∞ )⊥ , this product
defines an inner product of the form

T̂∞ : X(k∞ )⊥ × X(k∞ )⊥ −→  p . (12.22)

Since the -adic logarithm induces an isomorphism log : V (k∞ ) → X(k∞ )⊥ ,

we obtain an inner product

T∞ : V (k∞ ) × V (k∞ ) −→  p , (12.23)

where for x, y ∈ V (k∞ ) we put T∞ (x, y) = T̂(log x, log y).

It is easy to generalize our results to the case when k is an arbitrary finite
Abelian extension of Q . Indeed, put F = k(ζ0 ). Then  := G(F∞ /k∞ ) is a
cyclic group and |||(−1) (|| = 2 if  = 2). If we put V (k∞ ) = N V (F∞ ),
then we have a non-degenerate inner product Tk∞ : V (k∞ ) × V (k∞ ) →  p ,
which is defined as follows. Let x, y ∈ V (k∞ ). We have two natural maps: the
inclusion i : V (k∞ ) → V (F∞ ) and the norm map N : V (F∞ ) → V (k∞ ).
Let z ∈ V (F∞ ) be such an element that N (z) = y. Then we put Tk∞ (x, y) =
TF∞ (i(x), z), where TF∞ is the product (12.23) defined for F∞ .
 On the -adic regulator as an ingredient of Iwasawa theory 295

Now we consider the semi-local situation. Let K be an algebraic number

field such that the completions K v are Abelian for all v|. Let p be such an
index that any place v| is purely ramified in K ∞ /K p . Thus, for any v| we
have a canonical isomorphism G(K ∞ /K p ) ∼ = G(K n,v /K p,v ). So, we have for
any v| a product of the form (12.23)
Tv,∞ : V (K ∞,v ) × V (K ∞,v ) −→  p,v , (12.24)

where we can identify  p,v with  p . Then, putting V (K ∞ ) = v| V (K ∞,v ),
we obtain an inner product
T∞ : V (K ∞ ) × V (K ∞ ) −→  p , (12.25)
 
where for x, y ∈ V (K ∞ ), x = v| xv , y = v| yv we put T∞ (x, y) =

v| T∞,v (x v , yv ). Note that (12.25) is defined via the sum of local traces that
coincides with global one.
Let be given a G(K ∞ /K p )-module A ⊂ V (K ∞ ) and B = A∧ is an orthog-
onal complement of A with respect to (12.25). Since (12.25) is sesquilinear,
B is also a G(K ∞ /K p )-module. One has also rk A + rk B = rk V (K ∞ ) =
[K p : Q], where rk means the rank over  p . Let a ∈ A, b ∈ B, where

a = lim an , b = lim bn , an , bn ∈ V (K n ) = v| V (K n,v ). Then it follows
← − ← −
from the definition of T∞ that Sp K n /Q ((γ p − 1) log (an ) log (bn )) = 0 for all
n > p. Applying this to the case A = V + (K ∞ ) and denoting (V + (K ∞ ))∧ by
V − (K ∞ ), we get the following statement [10, Theor. 5.1].

Theorem 12.7. Let K be Abelian over , that is the completions K v are

Abelian for all v|. Put C(K ∞ ) = V + (K ∞ ) ∩ V − (K ∞ ). Then C(K ∞ ) = 0 if
and only if Conjecture 12.6 holds for (K , ).

Problem 12.1. Let a, b ∈ V (K ∞ , a ∈ V + (K ∞ ), b ∈ V − (K ∞ ) and, for a

given n, let an and bn be the images of a and b relative to the natural projection
V (K ∞ ) → V (K n ). Is it true that S(an , bn ) = 0, where S is the product (12.4)?

12.4 Duality results for non-Abelian local fields

Now we consider the case, when k is any (non-Abelian) finite extension of Q .
We wish to generalize the results of the preceding section to the case of the
cyclotomic Z -extension k∞ /k. To do this, we have firstly to formulate the
results of the preceding section in the form, which is maximally independent
of the special properties of cyclotomic fields. For example, such elements as
ξ(kn ) and #(kn ) cannot be defined in the general setting. So our first goal
296 L. V. Kuz’min

is to formulate the results of Section 12.3 in the form, which is suitable for
generalization. This was done in [12].
Thus for a while we suppose that k is a cyclotomic field of the form k =
H (ζ0 ), kn = H (ζn ) and k∞ = ∪n kn , where H is an unramified extension of
Q . Put = G(k∞ /k) and n = G(k∞ /kn ). Let γ be a fixed topological
generator of and γn = γ  . We define the groups V (kn ) and V (kn )⊥ as in
n

Section 12.3. We put for n > 0, as in Section 12.3,

= n−1
>
+n = qn−1 −1 δSpG n + Sp / n
+ Fn(i) ∈ Q [G n ],
i=0

where δ = −1 if  = 2, δ = 0 if  = 2, G n = G(kn /H ) and

Fn(i ) = Sp n−i / n

− Sp n−i −1 / n
∈ Q [G n ].

Theorem 12.8. Let kn be as above. Then for a fixed index p  0 and any
n > p we have

Ker Spkn /k p (log V (kn )⊥ ) = (γ p − 1)+n (log V (kn )),

where Ker Spkn /k p is the kernel of the corresponding trace map.

This theorem was stated in [12, Theor. A] for p = 0 and then generalized in
[15] for any p.
We shall say a few words on the proof given in [12], since this proof can be
generalize to the non-Abelian case. The proof breaks in two parts. At first, we
prove that
(γ p − 1)+n (log U (kn )) ⊆ (log U (kn ))⊥ , (12.26)

where U (kn ) is the group of units of kn . In the second part we prove that
(12.26) implies an inclusion

(γ p − 1)+n (log V (kn )) ⊆ Ker Spkn /k p (log V (kn ))⊥ , (12.27)

which in fact is an equality.

While the second part of the proof is rather routine, the first part is more
interesting, and we will to discuss it in some details. Obviously, we have to
prove that
A := Spkn /Q ((γ p − 1)+n x · y) ∈ Z (12.28)

for any x, y ∈ log U (kn ). The group log U (kn ) is generated by the Artin-Hasse
∞ −r
ϕ (α)−r πna , where α belongs to the ring of
r
logarithms ηn (α, a) := r=0
integers O H of H , ϕ is the Frobenius automorphism of H , a = 1, 2, . . . and
 On the -adic regulator as an ingredient of Iwasawa theory 297

πn = 1 − ζn is a local parameter of kn . So it is enough to prove (12.28) for

x = ηn (β, b) and y = ηn (α, a).
(i)
So we put Cni (α, a, r ) = Fn (ϕr (α)−r πna ). For r < 0 we
r

put Cni (α, a, r ) = 0 by definition. Taking into account that 1 =

n−1 −(i+1) (i) (i ) ( j)
i=0  Fn + −n Sp / n and the elements Fn , Fn are orthogonal
for i = j, we may rewrite (12.28) for the above-mentioned x and y in the form
n−1 ?
A= qn−1 −(i +2) Spkn /Q Cni (α, a, r + i)
−∞<r,s<∞ i =0
 @
× (γ − 1) Cni (β, b, s + i) .

To prove the theorem, it is enough to show that for any integers r, s we have
n−1 ?
B(r, s) := qn−1 −(i+2) Spkn /Q Cni (α, a, r + i )
i=0
 @
× (γ − 1) Cni (β, b, s + i) ≡ 0 (mod Z ). (12.29)
r+1 r
Since one can easily prove by induction on r a congruence πn+1
a
≡ πna
+1
(mod  ), it follows that (see [12, Lemma 3])
r

∞ ?
B(r, s) ≡ qn−1 −(n+1) Spk j /Q C n−1
j (ϕ n− j (α), a, n + r − 1)
j=n
 @
× (γ − 1) C n−1
j (ϕ n− j (β), b, n + s − 1) (mod Z ). (12.30)

Namely, we prove that the i-th summand in (12.29) is congruent to the

(n + i)-th summand in the last sum modulo Z and all summands that
correspond j  2n − 1 are congruent 0 modulo Z .
Thus instead of sum (12.29), which include summands of different form but
all belonging to kn , we obtain another sum, all terms of which look uniformly,
but belong to different fields. Note that we can also add to the sum (12.30) all
terms with j < n, since all these terms are zero because of the factor (γ − 1).
The next step is to introduce some universal power series that presents
the terms of (12.30). This is done in [12, Sect. 3]. Namely, we prove that
there is a formal power series F(T ) ∈ O H [[T ]] such that the j -th term
of (12.30) equals Spk j /Q (F (− j ) (π j )) times some power of  independent
of j. Here F − j (T ) means the result of action of ϕ − j on the coefficients of
F(T ). Obviously, this twisting by a power of Frobenius does not change the
trace. The final almost obvious fact we need to state the theorem is the fol-
lowing [12, Lemma 6]. For any formal power series F(T ) ∈ O H [[T ]] we
298 L. V. Kuz’min


have ∞ j=−1 Spk j /Q F
(− j) (π ) = 0, where π
j −1 = 0. (A slight modification
is necessary if  = 2.) This proves the theorem.
Now we have to repeat this proof for arbitrary finite (non-Abelian) exten-
sion of Q . This was done in [14]. The first problem is to define a coherent
system of local parameters. So let K be any finite extension of Q and K ∞ the
cyclotomic Z -extension of K . Assume that K contains a primitive -th root of
√
unity ( −1 ∈ K if  = 2). Let H be the maximal unramified subfield of K ∞ ,
kn = H (ζn ) and K n = K · kn . Then for all sufficiently large n the extension
K n /kn is purely ramified of some degree e, which does not depend on n.

Proposition 12.9. [14, prop. 1.2] Put R = O H [[T ]]. Then there is an irre-
ducible (over R) monic polynomial F(X, T ) ∈ R[X ] of degree e such that
F(X, πn ) is irreducible in O(kn )[X ] for any n  0. For all sufficiently large n
there is a local parameter λn in K n that satisfies F (−n) (λn , πn ) = 0. There is a
positive integer δ (independent of n) such that the different D(K n /kn ) satisfies
the condition D(K n /kn ) = (λδn ) for all sufficiently large n. Moreover, one can
choose the elements λn so that for all sufficiently large n

λn+1 ≡ λn (mod λ−δ ). (12.31)

Remark. The equality D(K n /kn ) = (λδn ) was originally stated in [23].

Sketch of the proof. Let q be such an index that ν (D(K q /kq ))  1/2, where
ν is the -adic exponent. Let λq be any local parameter of K q and f (X )
the minimal polynomial of λq over kq . Thus one can write f (X ) in the form
f (X ) = X e + ae−1 x e−1 + · · · + a0 , where ai = ai (πq ) are formal power series
in πq . We put f (X, T ) = X e + ae−1 (T )X e−1 + · · · + a0 (T ) and F(X, T ) =
f (q) (X, T ). So we have F (−q) (λq , πq ) = 0. Now let λq+1 be a root of
F −(q+1) (X, πq+1 ). Since (a + b) ≡ a  + b (mod ) for any integer ele-
ments a, b, we obtain that 0 = (F −(q+1) (λq+1 , πq+1 )) ≡ F (−q) (λ q+1 , πq )
(mod ). Therefore λ q+1 is a good approximation to a root of F (−q) (X, πq ).
Moreover, in this way we obtain one-to-one correspondence between all roots
of F (−(q+1)) (X, πq+1 ) and F (−q) (X, πq ). We take for λq+1 such a root that

λq+1 is close to λq .
To show that kq+1 (λq+1 ) = K q+1 it is enough to prove that λq ∈
F := kq+1 (λq+1 ). But it follows from the Hensel lemma, since λq+1 is an
approximation in F of the root λq . Then we define λq+2 in the same way and
so on. This proves the proposition.
∞ −r
ϕ (α)−r λa
r
Now we put ηn (α, a) = r=0 n . We would like to show that
(12.28) holds for y = ηn (β, b) and x = ηn (α, a). But (12.28) fails to be true
 On the -adic regulator as an ingredient of Iwasawa theory 299

in this case. So we can state only some weak version of this result. Thus we
shall proceed as follows. Reasoning as in the proof of Theorem 12.8, where we
have proved that the right sides of (12.29) and (12.30) are congruent modulo
Z , we prove, using (12.31), (see [14, Theor. 2.1] that for some p that depends
only on K and any n > p one has a congruence

A = A(α, a, β, b) := Sp K n /Q ((γ p − 1)+n (x) · y) (12.32)

∞ ?  (n− p−1)
≡ −m−n+ p−1 Sp K j /Q (γ p − 1) F j (η j (ϕ n− j (α), a))
j=n
@
(n− p−1)
×F j (η j (ϕ n− j (β), b)) (mod Z ).

Again, as in the proof of Theorem 12.8, we wish to define some universal series

G(T ) such that G − j (λ j ) = F j
n− p−1
(η j ϕ n− j (α, a)) . So we consider the ring
of formal power series R := O H [[T ]](ζσ ) and an automorphism σ of this ring
defined by σ (T ) = T −(1−T )(1−ζσ ), where ζσ is some root of unity (maybe
non-primitive) of degree n− p . All such σ form a cyclic group of order n− p .
Note that after specialization Y = π j we obtain an automorphism σ [ j] of O K j
of the form σ [ j](ζ j ) = ζ j ζσ . Now let λ be a root of the polynomial F(X, T )
defined in Prop. 12.9. Then one can present T as a formal power series in λ,
hence R[λ] ∼ = O H [[λ]]. Our goal is to extend the action of the automorphisms
σ on R(λ, μ), where μ is the group of all roots of unity of degree n− p , but to
do this we have to consider instead of O H (μ)[[λ]] the ring of formal Laurent
power series with infinite principal part. So let R̃ = O H (μ){{λ}} be the ring

of Laurent power series of the form ∞ i=−∞ ai λ , where ai ∈ O H (μ) and
i

ai tends to zero if i tends to −∞. Let Fσ (X, T ) ∈ R(μ)[X ] be a result of

the action of σ on the coefficients ai (T ) of F (X, T ). Then Fσ (λ, T ) ≡ 0
(mod 1 − ζσ ). So λ is an approximation of the root of Fσ , and we can, using
the Hensel lemma, extend λ uniquely to the root σ (λ) ∈ R̃ of Fσ . (We have
σ (λ)0 = λ, σ (λ)1 = λ − Fσ (λ)/Fσ (λ) and so on.) Using this expressions for
σ (λ) for any ζσ ∈ μ, one can define the above-mentioned series G(T ). Note
that in our case G(T ) is not a formal power series, but a formal Laurent series
with coefficients in O H and infinite principal part. Moreover, we obtain an
explicit estimate for the -adic exponents of its coefficients at negative powers
of T .
n− p−1
The next step is as follows. Though F j ( f (λ j )) is a function in λ j , in
(0)
fact it is a function in λ j−s because F sj = F j −s Sp K j /K j −s , where s = n− p−1.
We can translate all this into the language of formal Laurent series. That is, we

can define a new Laurent series D(X ) = i∞=−∞ di X i ∈ O H {{X }} such that
n− p−1
D n− j (λ j−n+ p+1 ) = F j (η j (ϕ n− j (α), a)) for all sufficiently large j . The
300 L. V. Kuz’min

advantage of using D(X ) instead of G(T ) is that its coefficients at the negative
powers of X tends to zero very fastly. Namely, we prove [14, Theor. 5.2] that
one has
i
ν (di ) > − (12.33)
2δ
for any i < 0, where δ is as in Prop. 12.9.
Another sort of automorphisms we have to deal with is the automorphism
×
γ p defined by γ p (ζ j ) = ζ κ
j , where κ = κ(γ p ) ∈ Z . Again, we can consider
γ p as an automorphism of R defined by γ p (T ) = 1 − (1 − T )κ . Using the
Hensel lemma one can uniquely extend the action of γ p to R̃. We prove [14,
Theor. 6.3] that there is a formal Laurent series E(X ) such that for any j  n
and  = 2 one has

(γ p − 1)D (− j) (λ j −s ) = π j −n E − j (λ j−s ), (12.34)

where s = n − p − 1. (For  = 2 (12.34) needs a slight modification.)

Note that the coefficients of E(X ) at the negative powers of X satisfies the
condition (12.33). Now we can state and prove the main theorem of [14].

Theorem 12.10. Let K be an arbitrary finite extension of Q , K n = K · kn

and K ∞ = ∪K n . Then there are an index p and a non-negative constant
C, depending only on K ∞ /K , such that for any n > p there exists a Z -
submodule Mn of the Z -module log(U (K n )) with the property that the group
(log(U (K n ))/Mn can be generated by at most Cn generators and

(γ p − 1)+n (Mn ) ⊆ (log U (K n ))⊥ .

Moreover, one can effectively determine p and C.

Sketch of the proof. Let the elements ηn,1 = ηn (α1 , a1 ), . . . , ηn,t =

ηn (αt , at ) generate the Z -module log(U (K n )) for a given n. Then any element
ηn ∈ log(U (K n )) can be written as
t
ηn = ci ηn,i
i=1

with some ci ∈ Z .
For each i = 1, . . . , t there is a Laurent series Ei (X ) such that
(− j )
(γ p − 1)F (s)
j (η j,i ) = π j−n Ei (λ j−s ), s = n − p − 1,
 On the -adic regulator as an ingredient of Iwasawa theory 301

where η j,i = η j (ϕ n− j (αi ), ai ). Putting

t t
ηj = ci η j,i , E(X ) = ci Ei (X ),
i=1 i=1

for any j  n, we see that

(s)
(γ p − 1)F j (η j ) = π j −n E (− j ) (λ j −s ), (12.35)

for every j  n.

The coefficient eν of X ν in E(X ) has the form eν = t ci ei,ν , where ei,ν
is the coefficient of X ν in Ei (X ). Thus eν is a linear function in ci with coef-
ficients ei,ν , and we can consider a subgroup Mn of log(U (K n )) consisting of
those ηn that the corresponding series E(X ) has zero coefficients eν whenever
ν satisfies |ν|  (n − p + 2)δ. Thus there are rn := 8(n − p + 2)δ + 1 coeffi-
cients eν in E(X ) that must vanish, that is, Mn is defined in log(U (K n )) by rn
linear equations with coefficients in O H or by rn f equations with coefficients
in Z , where f = [H : Q ].
We claim that for any ηn ∈ Mn and any ηn,i as above one has

A := Sp K n /Q (+n ηn · ηn,i ) ∈ Z .

Let D(X ) be the series corresponding to ηn,i , that is, for any j  n one has
(s)
F j (η j,i ) = D (− j) (λ j −s ). (12.36)

Then by (12.32), (12.35) and (12.36) we have

∞
A ≡ −m−n+ p−1 Sp K j /Q (π j−n E (− j) (λ j−s )D (− j) (λ j−s )) (mod Z ),
j=n

where π j−n must be replaced by π j−n−1 if  = 2.


If E(X ) = ∞ ν
ν=−∞ eν X , then we put

E(X ) = E1 (X ) + E2 (X ),

where
−4(n− p+2)−1 ∞
E1 (X ) = eν X ν , E2 (X ) = eν X ν .
ν=−∞ ν=4(n− p+2)+1

We define D1 (X ) and D2 (X ) analogously. Then

A ≡ Aα,β (mod Z ),
1α,β2
302 L. V. Kuz’min

where
∞
(− j)
Aα,β = −m−n+ p−1
Sp K j /Q (π j−n Eα(− j) (λ j−s )Dβ (λ j−s )).
j=n

We can prove that A1,1 ≡ A1,2 ≡ A2,1 ≡ 0 (mod Z ), using the estimates
of coefficients of these series of the form (12.33). The series A2,2 is a formal
power series, and we can treat it as in the proof of Theorem 12.8. This finishes
the proof of the theorem.
This theorem means that the situation for non-Abelian local fields is in
some sense close to that for Abelian fields. We shall see in Section 12.6
that this fact yields interesting consequences concerning Conjecture 12.6 (see
Theorem 12.19).
To analyze the non-Abelian situation, we introduced in [15] two sequences
( p) ( p)
of finite Abelian -groups An and Bn . Namely, if K is a non-Abelian finite
extension of Q and K n are defined as above, we put Vn = V (K n ) and
( p) ( p)
An = (γ p − 1)+n (log Vn )/Mn ,
( p) ( p)
Bn = Ker Sp K n /K p ((log Vn )⊥ )/Mn ,

( p)
where Mn = (γ p − 1)+n (log Vn ) ∩ Ker Sp K n /K p ((log Vn )⊥ ).

Proposition 12.11. [15, Theor. 2.1] For all sufficiently large n we have
( p) ( p) ( p) ( p) ( p) ( p)
d(An ) = d(Bn ), An = An+1 , Bn = Bn+1 ,

where d(A) means the minimal number of generators of an Abelian -group A.

In [16] it was stated by direct calculation that for any non-Abelian K and
( p) ( p)
 > 3 one has A p+1 = 0 and B p+1 = 0 for all sufficiently large p. This result
holds also for  = 3 with one exceptional case. It was also proved there that
( p) ( p)
An ∼ = Bn for all sufficiently large n and p [16, Theor. 3.3].
Moreover, recently the author have proved that for any non-Abelian K the
( p) ( p)
groups An and Bn are non-zero for any fixed p and all sufficiently large n
( p) ( p)
and d(An ) → ∞, d(Bn ) → ∞ if n → ∞.
To say the truth, our understanding of pairing (17.10) for non-Abelian local
fields is far from being good. We shall return to this question at the end of
Section 12.6. Here we should like to note the following problem.
 On the -adic regulator as an ingredient of Iwasawa theory 303

Problem 12.2. Let K be a finite non-Abelian extension of Q and the groups

( p) ( p)
An , Bn be as above. Is there any connection between these groups as Galois
modules?

12.5 Explicit formulae for the norm residue symbol

In this section we consider the norm residue symbol (the Hilbert symbol) for
local number fields. It appears that in Abelian case this symbol is very close to
the product T∞ from the preceding section. It explains why these formulae are
applicable to the -adic regulator. Though our main results are valid only for
Abelian local fields, we start with considering a general case.
Let k be a finite extension of Q .√Suppose that k contains a primitive -th
root of unity if  = 2 (k contains −1 if  = 2). Let k∞ = ∪n kn be the
cyclotomic Z -extension of k and kn = k(ζn ). Let μ (kn ) be the -component
of μ(kn ). Hence μ (kn ) = ζn  for all sufficiently large n.
Then for any n one has the non-degenerate Hilbert symbol
( , )n : kn× × kn× −→ μ (kn ). (12.37)
√ √
If a, b ∈ kn× and qn = |μ (kn )| then (a, b) = ϑ(a)( qn b)/ qn b, where ϑ(a) is
the element of the Galois group G ab of the maximal Abelian extension of kn
that corresponds to a in the sense of local class field theory. Let kn× [] be the
pro--completion of kn× and put H (kn ) = Ĥ (kn )/μ (kn ), where Ĥ (kn ) is the
group of universal norms in kn× [] from k∞ . It follows from local class field
theory that (12.37) induces a non-degenerate product
H (kn ) × H (kn ) −→ μ (kn ). (12.38)

We choose the roots ζn ∈ kn in such a way that ζn+1 = ζn for all n and identify
μ (kn ) with Z(m) = Z/ Z by putting 1 ↔ ζn , where m = qn . We put
m

H (kn )(m) = H (kn )/(H (kn )) and define H (kn )(m)[−1] as H (kn )(m) with
m

twisted Galois action, that is, σ [−1](y) = κ(σ )σ (y), where y ∈ H (kn )(m)
κ(σ )
and κ(σ ) is the cyclotomic character defined by σ (ζn ) = ζn . Then (12.38)
turns into the product
H (kn )(m) × H (kn )(m)[−1] −→ Z(m),
satisfying the relation (x, y)n = (σ (x), σ [−1](y))n for any x, y ∈ H (kn )(m).
Now, as in Section 12.3, we choose some index p and for all n > p put
x, yn = (σ −1 (x), y)n σ ∈ Z(m)[G(kn /k p )]. (12.39)
σ ∈G(kn /k p )
304 L. V. Kuz’min

The products (12.39) are compatible relative to the norm maps and, after
passing to inverse limit, we obtain an inner sesquilinear product [11, Prop. 2.2]

 , ∞ : H (k∞ ) × H (k∞ )[−1] −→  p , (12.40)

where  p was defined in Section 12.3.

Since (γ p −1)H (k∞ ) ⊆ V (k∞ ), where γ p and V (k∞ ) are as in Section 12.3,
we can extend (12.40) to the product on V (k∞ ) × V (k∞ ). This last product
has its values, generally speaking, in the quotient field L p of  p , but in fact all
these values lie in  p , and (12.40) induces an inner sesquilinear product [11,
Theor. 3.1]
 , ∞ : V (k∞ ) × V (k∞ )[−1] −→  p . (12.41)

In what follows, we assume that k is an Abelian field. Since k contains

also a primitive -th root of unity, we may assume k to be a cyclo-
tomic field. So we have also the product T∞ defined in Section 12.3.
Thus we have two isomorphisms f 1 : V (k∞ ) ∼ = Hom p (V (k∞ ),  p ) and
f 2 : V (k∞ )[−1] ∼ = Hom p (V (k p ),  p ), where f 1 and f 2 are induced by
T∞ and by (12.41) respectively. In other words, there is an isomorphism
ψ = f 1−1 ◦ f 2 : V (k∞ )[−1] → V (k∞ ) such that x, y∞ = T∞ (x, ψ(y))
and the problem is to describe ψ explicitly.
At first we prove that ψ does not depend on p [11, Prop. 4.4] and then prove
the main theorem of [11].

Theorem 12.12. [11, Theor. 4.1] Let k be a finite Abelian extension of Q

and k∞ the cyclotomic Z -extension of k. Then the twisted automorphism ψ
satisfies the following two equivalent conditions:

ψ(η(k∞ )) = η(k∞ ), ψ(η (k∞ )) = −η (k∞ ) (12.42)

(we use an additive notation for operations in V (k∞ )). Each of these condi-
tions defines uniquely the automorphism ψ.

The proof of the theorem given in [11] is rather complicated, but its idea is
simple. We use the explicit formulae for the norm residue symbol of Sen [21],
which express this symbol in terms of the logarithmic derivative. To be precise,
one has
−qn Sn (ζn ( f  (πn )/ f (πn )) log α)
(β, α)n = ζn , (12.43)

where α, β ∈ kn× , πn = 1 − ζn is a local parameter of kn , β = f (πn ), f (X ) ∈

O H [X ] and O H is the ring of integers of the maximal unramified subfield H
 On the -adic regulator as an ingredient of Iwasawa theory 305

of kn . Note that the element α in (12.43) satisfies rather restrictive condition

α ≡ 1 (mod λ2 ), where λ = 1 − ζ0 (λ = 2 if  = 2).
We prove that the setting β → −qn ζn f  (πn )/ f (πn ), where f (X ) is as
in (12.43), defines a map

δn : U (1) (kn ) −→ −qn O(kn )/−qn D(kn ), (12.44)

where O(kn ) and D(kn ) are the ring of integers and the different of kn
respectively.
Passing in (12.44) to inverse limit (relative to the norm maps on the left
and to the trace maps on the right side), we get a map δ∞  : U (k ) → k̃
∞ ∞ :=
lim kn . Using the behaviour of the symbol (12.37) under variation of the ground
← −
field, we can get rid of restrictions on α. As a result we obtain a formula
u, v∞ = T∞ (log u, δ∞  (v)) that is valid for any u, v ∈ U (k ) [11, Cor.
∞
to Lemma 5.2]. So, to prove the theorem we have to calculate δ∞  (v) for a

certain special element v ∈ U (k∞ ).

Comparing the explicit formulae for # (kn ) and ξ  (kn ) and reasoning quite
naively, we see that # (kn ) = ζn d(ξ  (kn )), where d means differentiation rel-
ative to ζn . To make this approach correct, we define a special power series
f (X ) ∈ O H [X ] and prove that the elements exp( f n (πn )) ∈ U (kn ) form a
coherent sequence relative to the norm maps. Here f n (X ) means a result of
the action of the power ϕ −n of Frobenius on the coefficients of f (X ). Let γ
be a topological generator of G(k∞ /k p ) and κ = κ(γ ) ∈ Z is defined by
the condition γ (ζn ) = ζnκ for any n. Then we prove [11, Lemma 5.5] that
f n (πn ) = (γ − κ)(γ − 1)ξ  (kn ) and ζn f n (πn ) = κ(κγ − 1)(γ − 1)# (kn ).
This conclude the proof of the theorem.
Now let K√be an algebraic number field containing a primitive -th root
of unity ζ0 ( −1 ∈ K if  = 2). Let K ∞ = ∪n K n be the cyclotomic Z -
extension of K . Then K n = K (ζn ) for all sufficiently large n. There are only
finitely many places v| in K ∞ , and we put
 
H (K ∞ ) = H (K ∞,v ), V (K ∞ ) = V (K ∞,v ),
v| v|

where the products are taken over all places over  in K ∞ . We choose a coher-

ent family of roots of unity ζn ∈ K n such that ζn+1 = ζn for any n and, since
K n ⊂ K n,v , we may consider {ζn }n0 as a coherent sequence of roots of unity
in the sequence of local fields {K n,v }n0 for any v|. Thus, putting 1 ↔ ζn ,
we get isomorphisms μ (K n ) ∼ = Z(m) for any n. So, passing to the limit, we
obtain an isomorphism Z ∼ = T [−1], where T = lim μ (K n ) and [−1] denotes
←−
twisted Galois action as above.
306 L. V. Kuz’min

We choose some index p such that for any n > p we have [K n : K p ] =

[K n,v : K n, p ] = n− p for any v|. Then, denoting the local product (12.39) by
 , n,v , we define a product

x, yn = xv , yv n,v ∈ Z(m)[G(K n /K p )], (12.45)

v|
 
where x = v| xv , y = v| yv ∈ H (K ∞ ).
Just as in the local case, passing to inverse limit in (12.45), we get an inner
sesquilinear product

 , ∞ : H (K ∞ ) × H (K ∞ )[−1] −→  p ,

which, in turn, induces an inner sesquilinear product

 , ∞ : V (K ∞ ) × V (K ∞ ) −→  p .

The latter product is connected with (12.41) by a formula analogous to (12.45).

Since we have also the product T∞ of Section 12.3, we again obtain a
twisted isomorphism ψ : V (K ∞ ) ∼ = V (K ∞ ). Obviously, ψ coincides with
local ψv of Theorem 12.8 on each local component V (K ∞,v ). So we can write

ψ = v| ψv . For any x, y ∈ V (K ∞ ) one has x, y∞ = T∞ (x, ψ(y)). The
group V + (K ∞ ) is the maximal isotropic subgroup of V (K ∞ ). This implies
the following characterization of the complement V − (K ∞ ) defined at the end
of Section 12.3.

Theorem 12.13. [11, Prop. 7.2] Let V − (K ∞ ) be the orthogonal complement

to V + (K ∞ ) relative to the product T∞ . Then V − (K ∞ ) = ψ(V + (K ∞ )).
More generally, for any integer i the module ψ i (V + (K ∞ )) coincides with the
orthogonal complement to ψ 1−i (V + (K ∞ )) relative to the product T∞ and
ψ i (V + (K ∞ )) coincides with the orthogonal complement to ψ −i (V + (K ∞ ))
relative to the product  , ∞ .

As an immediate consequence of Theorem 12.10 we get the following

statement.

Proposition 12.14. Let H be a finite -group of automorphisms of K ∞ Then

V + (K ∞ ) and V − (K ∞ ) are isomorphic as H -modules.

Proof. Since H is finite, the cyclotomic character κ is trivial on H . Hence ψ

is an H -isomorphism.
 On the -adic regulator as an ingredient of Iwasawa theory 307

12.6 Applications to the feeble conjecture on the

-adic regulator
The results of preceding sections enable us to prove Conjecture 12.6 for some
particular cases.

Theorem 12.15. [10, Theor. 6.1] Let  = 2, K be a finite -extension

of an algebraic number field k, k∞ the cyclotomic Z -extension of k and
K ∞ = K · k∞ the cyclotomic Z -extension of K . Let K be Abelian over ,
that is, the completions K v are Abelian for any v|. Suppose that the Iwasawa
μ-invariant μ (k) vanishes. Suppose also that k contains a primitive -th root
of unity, Conjecture 12.6 holds for k∞ /k, (that is, V + (k∞ ) ∩ V − (k∞ ) = 0 by
Theorem 12.4) and D(k∞ ) := V (k∞ )/(V + (k∞ ) ⊕ V − (k∞ )) is finitely gener-
ated over Z . Then V + (K ∞ ) ∩ V − (K ∞ ) = 0, in other words, Conjecture 12.6
holds for K ∞ /K and, moreover, D(K ∞ ) := V (K ∞ )/(V + (K ∞ ) ⊕ V − (K ∞ ))
is finitely generated over Z .

Sketch of the proof. We may assume that K ∞ /k∞ is Galois with a Galois
group G. By Iwasawa theorem on the μ-invariant we have μ (K ) = 0. This
means that the -ranks of the -class groups Cl (K n ) are restricted by some
λ that does not depend on n. We can deduce from this fact that the Tate
cohomology groups H i (G, U (K ∞ )) are finite.
The equality μ = 0 combined with the fact that ζ0 ∈ k yield that
V + (K ∞ )/U (K ∞ ) is a finitely generated Z -module. Hence the cohomolo-
gies H i (G, V + (K ∞ )) are also finite. By Proposition 12.14 we obtain that the
groups H i (G, V − (K ∞ )) are finite as well.
Put A = A/A for an Abelian group A. Then V (K ∞ ) , V + (K ∞ ) and
−
V (K ∞ ) are Z/Z[G]-modules with finite cohomologies, and the natural
injections V + (K ∞ ) → V (K ∞ ) and V − (K ∞ ) → V (K ∞ ) induce injec-
tions V + (K ∞ ) → V (K ∞ ) and V − (K ∞ ) → V (K ∞ ) . If Conjecture 12.6
fails to be true for K ∞ /K then by Theorem 12.4 C(K ∞ ) = 0. Hence
C(K ∞ ) is infinite, but C(K ∞ ) ⊂ V + (K ∞ ) ∩ V − (K ∞ ) . Therefore this
last group is infinite and so is the group (V + (K ∞ ) ∩ V − (K ∞ ) )G . On
the other hand, the natural inclusion V (k∞ ) → V (K ∞ ) implies inclusions
V + (k∞ ) → (V + (K ∞ ) )G and V − (k∞ ) → (V − (K ∞ ) )G , both with finite
co-kernels. If D(k∞ ) is finitely generated over Z then V + (k∞ ) ∩ V − (k∞ )
is also finite. (In fact it is a zero group, since V + (k∞ ) and V − (k∞ ) contain
no non-trivial finite submodules.) This contradicts to infiniteness of the group
(V + (K ∞ ) ∩ V − (K ∞ ) )G . The theorem is proved.
308 L. V. Kuz’min

Note that among other things we have proved in Theorem 12.12 that D(K ∞ )
is finitely generated over Z . We may treat this result as a complement to
Iwasawa theorem on the μ-invariant.
If  = 2 then the situation is a bit more complicated. In this case D(k∞ )
always has non-zero μ-invariant. Indeed, Theorem 12.8 shows that the map
ψ : V (k∞ ) → V (k∞ ) induces the identity map on V (k∞ )2 , hence the groups
V + (k∞ )2 and V − (k∞ )2 coincide as subgroups of V (k∞ )2 . Denote this last
subgroup by P̃, and by P we denote the maximal subgroup of V (k∞ )2 that
contains P̃ as a subgroup of finite index. Let V  (k∞ ) be the pre-image of P
under the natural projection V (k∞ ) → V (k∞ )2 . At last, we put D  (k∞ ) =
V  (k∞ )/(V + (k∞ ) ⊕ V − (k∞ )). We define the groups V  (K ∞ ) and D  (K ∞ )
analogously. Then for  = 2 we have the following version of Theorem 12.15.

Theorem 12.16. [11, Theor. 7.1] Let  = 2, let k be any extension of Q con-
√
taining −1, and let K be a finite 2-extension of the field k, Abelian over 2.
Let k∞ and K ∞ be the cyclotomic Z -extensions of k and K respectively. Sup-
pose that V + (k∞ ) ∩ V − (k∞ ) = 0, the Iwasawa μ-invariant μ2 (k) vanishes
and D  (k∞ ) is finitely generated over Z . Then V + (K ∞ ) ∩ V − (K ∞ ) = 0, that
is Conjecture 12.6 holds for K ∞ /K , the group D  (K ∞ ) is finitely generated
over Z and μ(D(K ∞ )) = [K : Q]/2.

Suppose that k is a C M-field. Then V + (k∞ ) ∩ V − (k∞ ) = 0, since the

complex conjugation j acts on V + (k∞ ) and V − (k∞ ) by multiplication by +1
and -1 respectively. So we have D(k∞ ) = 0 if  = 2, and D(k∞ ) has exponent
two and Iwasawa μ-invariant [k : Q]/2 if  = 2. Hence Theorems 12.15
and 12.16 yield the following result.

Theorem 12.17. In the notation of Theorem 12.15 let k be a C M-field contain-

√
ing ζ0 (containing −1 if  = 2) and such that μ (k) = 0. Then C(K ∞ ) = 0,
that is, Conjecture 12.6 holds for K ∞ /K . If  = 2 then D(K ∞ ) is a finitely
generated Z -module. If  = 2 then D  (K ∞ ) is a finitely generated Z2 -module
and D(K ∞ )/D  (K ∞ ) is of exponent two and has μ-invariant [K : Q]/2.

Note that the condition μ (k) = 0 holds if k is Abelian by the Ferrero-

Washington theorem. Therefore we have the following unconditional result.

Theorem 12.18. Let k be an Abelian algebraic number field and K an

-extension of k Abelian over . Then Conjecture 12.6 holds for the cyclo-
tomic Z -extension K ∞ /K . The module D(K ∞ ) defined above (the module
 On the -adic regulator as an ingredient of Iwasawa theory 309

√
D  (K ∞ ) if  = 2) is finitely generated over Z . If  = 2 and −1 ∈ K then
D(K ∞ )/D  (K ∞ ) is of exponent two and its μ-invariant is equal to [K : Q]/2.

The results of Section 12.4 enable us to generalize all these results to the
case, when K has an arbitrary ramification over . Note that in non-Abelian
case we don’t know if there exists the complement V − (K ∞ ). So we define
V + (K n ) as the image of the natural projection V + (K ∞ ) → V (K n ) and put
for any intermediate field K n in the cyclotomic Z -extension K ∞ /K
V − (K n ) = { x ∈ V (K n ) | Sp K n /Q (log x · log y) = 0 for any y ∈ V + (K n )}.

Theorem 12.19. [15, Theor. 6.1] Let  be an odd prime number and let K be
an algebraic number field that satisfies the following conditions:
(i) K contains a primitive -th root of unity ζ0 ;
(ii) μ (K ) = 0;
(iii) the field K and its cyclotomic Z -extension satisfy the feeble conjecture
on the -adic regulator (Conjecture 12.6);
(iv) There is a constant c (independent of n) such that for all n > 0

d V (K n )/(V + (K n ) + V − (K n )) < cn.
Then the following conditions are satisfied by any field L that is a finite
-extension of K :
(ii’) μ (L) = 0;
(iii’) the field L and its cyclotomic Z -extension L ∞ /L satisfy the feeble
conjecture on the -adic regulator;
(iv’) there is a constant c (independent of n) such that for all n > 0

d V (L n )/(V + (L n ) + V − (L n )) < c n.

The proof is similar to that of Theorem 12.15. We may suppose that L is

a Galois extension of K with Galois group G. Let { An } be a sequence of
G-modules. We say that this sequence has bounded cohomologies in dimen-
sion i if there is c = c(i ) (independent of n) such that d( Ĥ i (G, An )) < cn.
Obviously, two sequences {V (L n )} and {V + (K n )} has bounded cohomolo-
gies. Using Theorem 12.10, one can prove that {V − (K n )} also has bounded
cohomologies. Then, reasoning as in the proof of Theorem 12.15 and putting
C(L n ) = V + (L n ) ∩ V − (L n ), we obtain that d(C(L n )) < c1 n for some
c1 (independent of n). But C(L n ) are G(L n /L)-modules without Z -torsion.
Hence if their ranks grow, this growth cannot be so slow. This contradiction
proves the theorem.
310 L. V. Kuz’min

Again, one has analogs of Theorem 12.19 for the cases of C M-field K or
of Abelian K , which are similar to Theorems 12.17 and 12.18 respectively.
Note that in [18] we have generalized Theorem 12.19 to the case  = 2. This
generalization yields also generalization of Theorems 12.17 and 12.18. In par-
ticular, Conjecture 12.6 holds for  = 2 and for any L that is a 2-extension of
an Abelian field K .

Problem 12.3. Let V − (K n ) be as in Theor. 12.19 and V − (K ∞ ) =

lim V − (K n ). Is it true that rk(V (K ∞ )) = rk(V + (K ∞ )) + rk(V − (K ∞ ))?
← −
In [8] in construction of an analog of the Weil product we used the fact
that H − (K ∞ ) is isotropic relative to the norm residue symbol. The module
V − (K ∞ ), that we have constructed, as a rule has not this property. We may
improve V − (K ∞ ), using the following simple statement.

Proposition 12.20. Let A be a finite free Z/n Z-module and A × A → Z/n Z

a skew-symmetric non-degenerate product on A. Let B ⊂ A be a maximal
isotropic submodule of A such that A ∼ = B ⊕ C for some C. Then there is a
canonically defined C  ⊂ A (C  depends on C) such that A ∼
= B ⊕ C  and C 
is isotropic.

At last, note that in Section 12.8 we shall discuss another approach to

Conjecture 12.6, which is based on completely different ideas.

12.7 An Analog of the Riemann-Hurwitz formula for some

-extensions of algebraic number fields
The theorem 12.1 of introduction can be generalized to some -extensions
K /k, where K are not of C M-type. Let k∞ /k be the cyclotomic Z -extension
of k and N the maximal unramified Abelian  extension of k∞ such that
in N /k∞ all places v| completely split. Let M be the maximal Abelian
-extension of the field k∞ not ramified outside  , and M  the maximal sub-
field of M such that the Galois group G(M  /k∞ ) is a torsion free -module,
where  = Z [[ ]] and = G(k∞ /k). Let N̄ be the maximal Abelian unram-
ified -extension of k∞ . Let N  = N ∩ M  and N̄  = N̄ ∩ M  . We denote the
Galois groups G(N /N  ), G(N  /k∞ ) and G( N̄  /N  ) by T (k∞ ), T (k∞ ) and
R (k∞ ) respectively. These groups are torsion -modules, and we will denote
their λ-invariants by λ (k∞ ), λ (k∞ ) and r  (k∞ ).
 On the -adic regulator as an ingredient of Iwasawa theory 311

Theorem 12.21. [13, Theor. 1] (For the case  = 2 see √ the remark at the end
of [11].) Let k contains a primitive -th root of unity ( −1 ∈ k if  = 2), K a
finite -extension of k, Abelian over . (This means that K v are Abelian for all
v|.) Suppose that D(k∞ ) defined in Section 12.6 is a torsion -module, and
that the torsion -modules T (k∞ ) and D(k∞ ) (D  (k∞ ) if  = 2) have zero
Iwasawa μ-invariants.
Then D(K ∞ ) is a torsion -module, and the following relation holds:

d(K ∞ ) + 2λ (K ∞ ) + 4λ (K ∞ ) + 4r  (K ∞ ) − 4

= [K ∞ : k∞ ][d(k∞ ) + 2λ (k∞ ) + 4λ (k∞ ) + 4r  (k∞ ) − 4] + 2 (ev − 1),
v
(12.46)

where v runs through all places v   of the field K ∞ , ev is the ramification

index of the place v in the extension K ∞ /k∞ , and d(k∞ ) = λ(D(k∞ )).

The idea of the proof is as follows. It is enough to prove the theorem in the
case when K ∞ /k∞ is a cyclic extension of degree . Let G denotes its Galois
group. For a G-module A such that the Tate cohomology groups Ĥ i (G, A)
are finite for all i , we define the Euler characteristic by the formula χ ( A) =
dimF H 2 (G, A) − dimF H 1 (G, A). If the G-module A is finitely generated
over Z then rkA = rkA G −(−1)χ (A). So we compute χ ( A) for all modules
entering the formula of the theorem and it yields the proof.
√ √
As an application of this theorem we take k = Q( −3) and K = k( 3 a),
where a ∈ Z and K ∞ /k∞ ramifies exactly in two places. Then all invari-
ants in (12.46) vanish, and we get a decomposition V (K ∞ ) = V + (K ∞ ) ⊕
V − (K ∞ ). Suppose that K ∞ /k∞ ramifies in three places. Then Theorem 12.21
yields that either λ (K ∞ ) = 2 and λ (K ∞ ) = d(K ∞ ) = r  (K ∞ ) = 0 or
d(K ∞ ) = 4 and all other invariants are zero.√ √
The second case takes place for K = k( 3 550) and K = k( 3 460) (see [13,
examples 6.1 and 6.2]. In these two examples T (K ∞ ) is finite but non-zero.

Problem 12.4. Are there algebraic number fields K such that λ (K ∞ ) > 0?

If K is of C M-type then the condition λ (K ∞ ) = 0 is equivalent to the

Greenberg conjecture [2]. So we can consider the claim λ (K ∞ ) = 0 as a
generalization of Greenberg conjecture that has sense for any algebraic number
field. The problem is if it holds true or not.
312 L. V. Kuz’min

12.8 On a new type of the -adic regulator

The regulator R (K ) that we had discussed in the preceding sections isn’t quite
perfect from the viewpoint of Iwasawa theory.
For example, suppose that prime  splits completely in an algebraic num-
ber field K . In this case for any intermediate subfield K n of the cyclotomic
Z -extension K ∞ /K one has a regulator R (K n ) that characterize the -adic
behaviour of the group of units U (K n ) of K n . As it was shown in [17],
the behaviour of the -adic exponents ν (R (K n )) for increasing n shows
some interesting features. This is connected with the fact that the properties
of the sequence of the group of units {U (K n )}n0 reflect in a single object
Ū (K ∞ ) := lim(U (K n )[]/μ(K n )), which is a Galois module (relative to the
←−
action of = G(K ∞ /K )). We have a natural projection π : Ū (K ∞ ) →
U (K )[]/μ (K ), but in our case this projection is a zero map. (At any rate,
if the Leopoldt conjecture holds for (K , ).) So it is natural to consider instead
of U (K ) another object that behaves better from the viewpoint of Iwasawa
theory.
Namely, let S denotes the set of all places over  in a given field and U S (K n )
the group of S-units in K n . Let U S,1 (K n ) be the subgroup of universal norms
in U S (K n )[]/μ (K n ), that is

U S,1 (K n ) = ∩m>n N K m /K n (U S (K m )[]/μ (K m )). (12.47)

The new -adic regulator that we wish to define should characterize the
-adic behaviour of U S,1 (K n ). But now we shall treat only a particular case
n = 0, that is,  splits completely in K . To give the definition of our new
regulator we have to state some properties of the group U S,1 (K ).
For any S-unit u there is the principal divisor (u) of u. We can extend this
map to the pro--completion of the group of S-units and obtain a well-defined
map Div : U S (K )[] → D(K ), where D(K ) is a free Z -module (written addi-
tively) generated by the set S of all places v| in K . Let l1 , · · · , ln be all prime
n n
divisors of  in K . If x, y ∈ D(K ) and x = i=1 ai li , y = i=1 bi li ,
we put
n
x, y = ai bi ∈ Z . (12.48)
i =1

Thus (12.48) defines a non-degenerate bilinear product of the form D(K ) ×

D(K ) → Z . If K is Galois over some subfield with a Galois group G then the
natural action G on S extends to the action on D(K ), and the product (12.48)
satisfies the condition σ (x), σ (y) = x, y for any x, y ∈ D(K ), σ ∈ G.
Under assumption that the Leopoldt conjecture holds for (K , ), one can state
 On the -adic regulator as an ingredient of Iwasawa theory 313

easily that Div defines an injection Div : U S,1 (K ) → D(K ). The next the-
orem [19, Theor. 1] characterize the group U S,1 (K ) as Galois module (as
Z -module if K isn’t Galois). Note that in this theorem we put no restrictions
on decomposition type of prime l in K .

Theorem 12.22. Let K be a finite Galois extension of Q with a Galois group

G and R = Q [G]. Suppose that the Leopoldt conjecture holds for (K , ).
Then there an isomorphism of R-modules

ϕ : U S,1 (K ) ⊗Z Q ∼
= (U (K ) ⊗Z Q ) ⊕ Q

(here we write the group operation in U S,1 (K ) and U (K ) additively). Without

assumption that K is Galois over Q, the module U S,1 (K ) has Z -rank r1 + r2 ,
where r1 and r2 is the number of real and complex places of K respectively.

The idea of proof is as follows. Let M be the maximal Abelian -extension

of K ∞ unramified outside S and X = G(M/K ∞ ). Let M0 be the maximal
subfield of M Abelian over K . Then we can characterize the Galois group
G(M0 /K ) in two ways. Firstly, M0 is an Abelian extension of K hence its
Galois group over K can be characterized in terms of class field theory. This
characterization depends on U (K ). On the other hand, denoting by Mn the
maximal subfield of M Abelian over K n and putting X n = G(Mn /K n ) we
can characterize any X n via class field theory, as above, and put X = lim X n ,
← −
where the limit is taken relative to the norm maps. Thus, factoring X by the
action of γ , where γ is a topological generator of = G(K ∞ /K ), we obtain
a characterization of X 0 = X/(γ − 1)X ∼ = G(M0 /K ∞ ) in terms of U S,1 (K ).
Comparing these two characterizations, we obtain the theorem.
Using this theorem, we can define our new regulator as follows.

Definition. Let e1 , · · · , et be a Z -basis of U S,1 (K ), where t = r1 + r2 . Then

we put
R (K ) = det(Div(ei ), Div(e j )), 1  i, j  t,

where  ,  is the product (12.48).

Conjecture 12.23. Let K be an algebraic number field and  a prime that

splits completely in K . Then R (K ) = 0.

Theorem 12.24. [19, Theor. 2] Let K be an algebraic number field and prime
 splits completely in K . Then Conjecture 12.2 yields that R (K ) = 0.
314 L. V. Kuz’min

Sketch of the proof. We may assume that K is Galois over Q and K contains
an imaginary quadratic field k. Put G = G(K /Q) and H = G(K /k). Let
α ∈ U S (K ) be such an element that (α) = lh1 , where l1 is a prime divisor
of  in K and h any number such that lh1 is principal. It follows from Con-
jecture 12.2 that the elements log (σ (α)), σ ∈ G generate in K ⊗Q Q a

subspace of co dimension one. (The only relation is σ log (σ (α)) = 0.) The
proof of this fact is similar to the proof that the Leopoldt conjecture follows
from Conjecture 12.2.
Let ε ∈ U (K ) be an Artin unit, that is, τ (ε) = ε for an automorphism
of complex conjugation τ ∈ G and the elements h(ε), h ∈ H , generate a
subgroup of finite index in U (K ). Then there exist coefficients ch , h ∈ H ,
defined up to a constant summand and such that
log ε = (1 + τ ) ch h(α).
h∈H

Using Conjecture 12.2, one can show that the set of n elements ch , h ∈ H ,
where |H | = n, has transcendence degree n − 1 over Q. The element

z = ε−1 (h(α ch )1+τ
h∈H

has the property log (z) = 0. One can deduce from this fact that y = z  ∈
m

U S,1 (K ) for some sufficiently large m. Then the elements h(y), h ∈ H gen-
erate a subgroup of finite index in U S,1 (K ), and it is enough to check that this
system of elements has non-zero regulator R. We show that this last regulator
is a non-zero polynomial with rational integer coefficients in indeterminates
c1 , . . . , cn−1 . Hence R = 0 and R (K ) = 0. This proves the theorem.
It is desirable to give a non-conditional proof of conjecture 12.23 in some
cases. In this direction we have proved the following result.

Theorem 12.25. [19, Theor. 3] Let K be an Abelian extension of an imaginary

quadratic field k with Galois group H = G(K /k). Suppose that K is Galois
over Q and prime  splits completely in K . Let the complex conjugation τ ∈
G = G(K /Q) acts on H by the rule τ (h) = τ hτ −1 = h −1 for any h ∈ H .
Let the exponent m of H be such that the primitive m-th root of unity ζm ∈ Q̄
is conjugated with ζm−1 over Q . Then one has R (K ) = 0.

Sketch of the proof. Let p1 , p2 be two divisors of  in k. Then S = S1 ∪ S2 ,

where Si is the set of all places in K over pi for i = 1, 2. We put Di (K ) =
Z [Si ] and define two products as in (12.48)
 , 1 : D1 (K ) × D1 (K ) −→ Z ,  , 2 : D2 (K ) × D2 (K ) −→ Z .
 On the -adic regulator as an ingredient of Iwasawa theory 315

If x, y ∈ D(K ) then x = x1 +x2 and y = y1 + y2 , where xi , yi ∈ Di , i = 1, 2,

and we can define a new product x, yk = x1 , y1 p1 + x2 , y2 2 p2 .
The first step is as in the proof of Theor. 12.5. We prove, using Theor. 12.22,
that one has τ (x), τ (y)k = x, yk . It means that x, y = 2x1 , y1 1 =
2x2 , y2 2 .
By Theor. 12.22 we have U S,1 (K ) ⊗Z Q ∼ = Q [H ]. Thus, if ϕ is a
Q -irreducible character of H then at least one of two maps πi : U S,1 (K )ϕ →
Di (K ), i = 1, 2 must be an injection. Suppose that for a given ϕ the map
π1 is an injection hence Im(π1 (U S,1 (K )ϕ )) is a subgroup of finite index in

D(K )1,ϕ . Any ϕ is of the form ϕ = χ|ϕ χ , where χ runs through all
absolutely irreducible characters of H that are conjugated over Q . By ϕ̄

we denote the Q -irreducible character χ|ϕ χ −1 . Since the product  , 1
is non-degenerate, we obtain that  , 1 induces a non-degenerate product
D(K )1,ϕ × D(K )1,ϕ̄ → Z . But in our case one has ϕ = ϕ̄. Hence the product
U S,1 (K )ϕ × U S,1 (K )ϕ → Z induced by  ,  is non-degenerate. This proves
the theorem, since ϕ is an arbitrary Q -irreducible character of H .
The next proposition characterize those  that satisfy the conditions of
Theorem 12.25 for a given K .

Proposition 12.26. Let K be Galois over Q and Abelian over an imaginary

quadratic field k, as in Theor. 12.25. Suppose that τ ∈ G(k/Q), τ = 1, acts
on H = G(K /k) by inversion. Let m be an exponent of H , ξm a primitive m-th
root of unity and T the set of all primes that do not divide m.
If K ∩ Q(ξm ) is totally real then there are infinitely many  ∈ T such that
(K , ) satisfy all conditions of Theor. 12.25. Therefore R (K ) = 0 for these .
If K ∩ Q(ξm ) is an imaginary field then there is no  ∈ T that satisfies the
conditions of Theor. 12.25.

In conclusion we return to Conjecture 12.6.

Theorem 12.27. Let K be an algebraic number field and prime  splits com-
pletely in K . Suppose that R (K ) = 0. then the feeble conjecture on the -adic
regulator (Conjecture 12.6) is valid for (K , ).

Sketch of the proof. Let H (K ∞ ) be as in Section 12.5. One has (γ −

1)H (K ∞ ) ⊂ V (K ∞ ), where γ is a topological generator of G(K ∞ /K ).
Therefore one can extend the product (12.25) by -sesquilinearity to the
product

H (K ∞ ) × H (K ∞ ) −→ (γ − 1)−2 .
316 L. V. Kuz’min

In turn, this product leads to the product

H (K ∞ ) × H (K ∞ ) −→ (γ − 1)−2 /(γ − 1)−1  ∼
= Z . (12.49)
Since H (K ∞ )/(γ − 1)H (K ∞ ) ∼= D(K ) and the product (12.49) vanishes on
(γ − 1)H (K ∞ ), the product (12.49) induces a non-degenerate product
D(K ) × D(K ) −→ Z , (12.50)
which up to a constant factor coincides with (12.48).
Suppose that Conjecture 12.6 fails to hold for K ∞ /K . Then by Theor. 12.7
one has C(K ∞ ) = V + (K ∞ ) ∩ V − (K ∞ ) = 0. We define H + (K ∞ ) ⊆ H (K ∞ )
as the minimal  submodule of H (K ∞ ) such that V + (K ∞ ) ∩ H (K ∞ ) ⊆
H + (K ∞ ) and H (K ∞ )/H + (K ∞ ) has no -torsion. We define H − (K ∞ ) anal-
ogously. Let D + (K ) and D − (K ) be the image of the natural projection
π : H + (K ∞ ) → D(K ) and H − (K ∞ ) → D(K ) respectively. If C(K ∞ ) = 0
then F(K ∞ ) := H + (K ∞ ) ∩ H − (K ∞ ) = 0 and H (K ∞ )/F(K ∞ ) has no
-torsion. Let F1 be the image in D(K ) of F(K ∞ ) relative to π .
Since the product (12.49) is zero on F(K ∞ )× H + (K ∞ ), the product (12.50)
is zero on F1 × D + (K ), hence the product (12.50) is degenerate on D + (K ) ×
D + (K ). But D + (K ) contains DivU S,1 (K ) as a subgroup of finite index.
Therefore the product (12.48) is degenerate on Div(U S,1 ) and R (K ) = 0.
This finishes the proof.

Problem 12.5. We started from an analogy between functional and number

cases. One can ask if there are any analogs of the considered results in the
functional case?

References
[1] Brumer A. “On the units of algebraic number fields”. Math. 14 (1967), 121-124.
[2] Greenberg R. “On the Iwasawa invariants of totally real number fields”. Amer. J.
Math. 98 (1976), 263-284.
[3] Hartshorne R. Algebraic geometry, Springer-Verlag, New York, Heidelberg,
Berlin, 1977.
[4] Iwasawa K. “On some modules in the theory of cyclotomic fields”. J. Math. Soc.
Japan. 20 (1964), 42-82.
[5] “Lectures on p-adic L-functions”, Princeton Univ. Press and Univ. of Tokyo
Press, Princeton 1972.
[6] Iwasawa K. “On the μ-invariants of Z -extensions”. Number theory, algebraic
geometry and commutative algebra in honor of Akizuki, Tokyo 1979, 1-11.
[7] Kuz’min L. V. “The Tate module for algebraic number fields”. Math. USSR-Izv.,
6:2 (1972), 263-321.
 On the -adic regulator as an ingredient of Iwasawa theory 317

[8] Kuz’min L. V. “Some duality theorems for cyclotomic -extensions of algebraic

number fields of C M-type”. Math. USSR-Izv., 14:3 (1980), 441-498.
[9] Kuz’min L. V. “Some remarks on the -adic Dirichlet theorem and the -adic
regulator”. Math. USSR-Izv., 19:3 (1982), 445-478.
[10] Kuz’min L. V. “Some remarks on the -adic regulator. II”. Math. USSR-Izv., 35
(1980), 113-144.
[11] Kuz’min L. V. “New explicit formulas for the norm residue symbol, and their
applications”. Math. USSR-Izv., 37:3 (1991), 555-586.
[12] Kuz’min L. V. “A new proof of a duality theorem on the -adic logarithms of
local units”. J.Math Sci. 95:1 (1999), 1996-2005.
[13] Kuz’min L. V. “An analog of the Riemann-Hurwitz formula for one type of
-extensions of algebraic number fields”. Math. USSR-Izv. 36 (1991):2, 325-347.
[14] Kuz’min L. V. “Some remarks on the -adic regulator. III”. Izv. Math. 63:6
(1999), 1089-1138.
[15] Kuz’min L. V. “Some remarks on the -adic regulator. IV”. Izv. Math. 64:2
(2000), 265-310.
[16] Kuz’min L. V. “On some property of the -adic logarithms of units of non-
Abelian local fields”. Izv. Math. 70:5, (2006), 949-974.
[17] Kuz’min L. V. “Some remarks on the -adic regulator.V. Growth of the -adic
regulator in the cyclotomic Z -extension of an algebraic number field”. Izv. Math.
73:5, (2009), 959-1021.
[18] Kuz’min L. V. “The feeble conjecture on the 2-adic regulator for some 2-
extensions”. Izv. Math. 76:2 (2012), 346-355.
[19] Kuz’min L. V. “On a new type of the -adic regulator for algebraic number fields
(The -adic regulator without logarithms)”. Izv. Math., 79:1 (2015), 109–144.
[20] Lang S. “Cyclotomic fields” I, II Grad. Texts in Math., 121 Springer-Verlag, New
York, 1990.
[21] Sen Sh. “On explicit reciprocity laws”. J. Reine Angew. Math. 313 91980), 1-26.
[22] Winberg K. “Duality theorems for -extensions of algebraic number fields”.
Comp. Math. 55 (1985), 333-381.
[23] Wintenberger J.-P. “Le corps des norms de certaines extensions infinies de corps
locaux; applications”. Ann. Sci. Ecole Norm. Sup. (4) 16:1 (1983), 59-89.
 13
On a counting problem for G-shtukas
Ngo Dac Tuan
CNRS - Université de Paris Nord (Paris 13), LAGA - Département de Mathématiques,
99 avenue Jean-Baptiste Clément, 93430 Villetaneuse, France
E-mail address: [email protected]

A BSTRACT. In this paper, we give a fixed point formula of stacks of G-

shtukas with arbitrary modifications, which generalizes the works of Drinfeld
and Lafforgue on stacks of Drinfeld’s shtukas.

Moduli spaces (or stacks) of shtukas were introduced by Drinfeld [4] and they
are associated to the data consisting of a reductive group G over a function field
F and a collection of dominant coweights of this group which are also called
modifications. These algebraic stacks are considered to be an analog version
of Shimura varieties over function field and it is expected that they would play
an important role in the Langlands program over function fields.
In the seventies, Drinfeld studied the so-called moduli space of Drinfeld’s
shtukas for GL2 and proved that the Langlands’ correspondence for GL2 could
be realized in the cohomology of this moduli space [4, 5]. Later, following
the same strategy, Laurent Lafforgue proved the Langlands’ correspondence
for GLn by realizing it in the cohomology of the stack of Drinfeld’s shtukas
for GLn [13, 14, 15]. When the reductive group G is an interior form of GLn ,
different stacks associated to this group have been studied in the literature, see
Laumon [19, 20], Laumon-Rapoport-Stuhler [21], Lau [17, 18] and Ngô Bao-
Châu [24]. In the direction of a general reductive group G, the first result is due
to Varshavsky and Kazhdan [33, 7]. Note that, by a completely new and dif-
ferent method, Vincent Lafforgue [16] has proved a canonical decomposition
of the space of cuspidal automorphic forms for G by studying all the stacks of
G-shtukas.
2010 Mathematics Subject Classification: 22E55, 11G09, 14D20.
Keywords: Drinfeld shtukas, moduli spaces, function fields, fixed point formula, Arthur-Selberg
trace formula.

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

318
 On a counting problem for G-shtukas 319

This paper is a continuation of our previous articles [26, 28] in which we

are interested in counting the number of fixed points for stacks of G-shtukas
under a power of the Frobenius and a Hecke correspondence. We hope that it
would constitute a step towards the study of the cohomology of these stacks
in the same spirit of the works of Drinfeld and Lafforgue. In all the works
cited above, the classification of ϕ-spaces (for GLn ) due to Drinfeld plays a
primordial role in the counting problem. Roughly speaking, it could be con-
sidered as an analog version of the Honda-Tate theory over function fields.
To our knowledge, we do not know any generalization of this classification
for a general reductive group G. In the articles [26, 28], we have suggested
a way to by-pass this obstacle and have given a formula for the number of
fixed points for the elliptic part. Essentially, the number of fixed points for
the elliptic part for G-shtukas is a sum indexed by elliptic semisimple ele-
ments of G(F) of product of twisted orbital integrals and orbital integrals for
some test function. The main goal of this paper is to extend this formula by
dealing with all (not necessarily semisimple) elements of G(F). Since we
are working in positive characteristic, we do not have the canonical Jordan
decomposition for elements of G(F). What saves the day is the observation
that, for any element γ in G(F), there always exists a power of γ which is
semisimple.
The article is organized as follows. In Section 13.1 and Section 13.2, we
recall the definition of stacks classifying G-shtukas, its basic properties, the
morphisms of partial Frobenius and the Hecke correspondences. In Sections
13.3-13.6, we present a formula for the number of fixed points of a power of
the Frobenius and a Hecke correspondence on a fiber of these stacks, see Theo-
rem 13.21. The formula is given in terms of twisted orbital integrals. In Section
13.7, we give a slight generalization of this formula by counting the number of
fixed points of a power of the partial Frobenius and a Hecke correspondence
on a fiber of stacks of G-shtukas, see Theorem 13.25. We also illustrate our
calculation in the case of the general linear group G = GLn , see 13.28.

Acknowledgements. The author thanks Ngô Bao-Châu for his interest and
helpful conversations on this work. Different parts of the work were done while
the author visited the Hausdorff Institute for Mathematics (HIM) and the Viet-
nam Institute for Advanced Study in Mathematics (VIASM), which I thank for
stimulating atmosphere and financial support. My research is partly supported
by the grant ANR-10-BLAN 0114 “Arithmétique des Variétés de Shimura et
des formes automorphes et Applications”, the grant LIA Formath Vietnam and
the grant ARCUS via the Université de Paris-Nord (Paris 13).
320 Ngo Dac Tuan

Notations
In this article, let X be a smooth projective and geometrically connected curve
over a finite field Fq of positive characteristic p. Denote by F the function field
of X . Let x be a place of F, denote by Fx the x-adic completion of F and Ox
its ring of integers. We fix an algebraic closure k of Fq and denote by X̄ , F̄, F̄x
and Ōx the corresponding objects obtained by base change from Fq to k, for
example X̄ = X ×Fq k, F̄ = F ⊗Fq k, etc.
Let G be a (connected) split reductive group over Fq , hence over F, whose
derived group G der is always supposed to be simply connected. This hypothesis
is not necessary. However, it allows us to use several results of Steinberg [32]
and Kottwitz [8], hence simplifies lots of arguments. Denote by B and T a
Borel subgroup and a maximal torus of G defined over Fq , with T ⊂ B.
Denote by B der et T der (resp. B ad and T ad ) the induced Borel subgroup and
the induced maximal torus of G der (resp. G ad ). With these notations, we define
the set of simple roots, dominant (rational) weights and coweights of G with
respect to (T, B).
In this article, every scheme or stack will be defined over Fq . Let Y and Z
be two such stacks, then by Y ×Z, we always mean the fiber product Y ×Fq Z.

13.1 Stack of G-shtukas

We begin by recalling the definition of G-shtukas, the stack classify-
ing such objects and its properties. The principal references of this sec-
tion are Lafforgue’s book [13], Varshavsky’s article [33] and that of Ngô
Bao-Châu [24].

13.1.1 Stack of G-torsors. Modifications

We consider the stack BunG classifying G-torsors over the curve X . Let T be
a finite subscheme of X , and let V, V  be two G-torsors over X . By definition,
a T -modification of V in V  consists of an isomorphism between V and V 
outside T , i.e.
/ ∼ /
φ : V / X −T −→ V  / X −T .

Let x be a geometric point in T . Denote by Ox the completion of O X at the

place x, and by Fx its fraction field. Given a T -modification
/ ∼ /
φ:V / X −T
−→ V  / X −T
 On a counting problem for G-shtukas 321

∼
between two G-torsors V et V  , we get an isomorphism V −→ V  between
∼
the generic fibers of V and V  which induces an isomorphism φx : Vx −→ Vx
between the completion of V et V  at x. If we choose a trivialization of Vx , then
the completion of V at x defines an element of the quotient G(Fx )/G(Ox ).
Similarly, if we choose a trivialization of Vx , then the completion of V  at x
defines an element of the quotient G(Fx )/G(Ox ). Using the above isomor-
phism φx , Vx and Vx can be identified, hence we get a double class in the
double quotient G(Ox )\G(Fx )/G(Ox ). We define the relative position of φ to
be this double class.
Since G is a split reductive group over Fq , the Cartan decomposition
gives us:
A
G(Fx ) = G(Ox )2x̄λ G(Ox )
λ

where λ runs through the set of dominant coweights of G, and 2x̄ denotes a
uniformizing element of X at x. Each double class contains a unique element
of the form 2x̄λ where λ is a dominant coweight. We define this dominant
coweight as the invariant of φ at x.

13.1.2 Hecke stacks

Recall that the set of dominant coweights of G is equipped with an order rela-
tion. Let λ and λ be two dominant coweights of G, we say that λ ≤ λ if and
only if λ − λ is a positive integral linear combination of simple coroots of G.
Following Beilinson and Drinfeld, for each dominant coweight λ of G, we
consider the stack Hecke≤λ classifying simple modifications bounded by λ as
follows. For each scheme S (over Fq ), Hecke≤λ (S) is the groupoid consisting
of the following data:

i) two G-torsors V and V  over X × S,

ii) a morphism x : S −→ X , and we denote by (x) the graph of x,
iii) an isomorphism
/ ∼ /
φ : V / X ×S− (x)
−→ V  / X×S− (x)

such that, for each geometric point s of S, the s-modification φs admits an

invariant (at x(s)) bounded by λ.

Denote by Heckeλ the substack of Hecke≤λ by replacing the condition iii)

by the following condition:
322 Ngo Dac Tuan

iii’) an isomorphism
/ ∼ /
φ : V / X ×S− (x)
−→ V  /
X ×S− (x)

such that, for each geometric point s of S, the s-modification φs admits λ

as the invariant (at x(s)).

It is proved that Heckeλ is an open substack of Hecke≤λ . Furthermore, we

get the following proposition and we send the reader to [2] for the proof:

Proposition 13.1. The morphism

Hecke≤λ −→ X × BunG
(V, V  , x) → (x, V)

is representable and projective. And the restriction of this morphism to the

open substack Heckeλ is smooth.

More generally, to each collection of dominant coweights λ = (λ1 , . . . , λn )

of G, we introduce the Hecke stack Hecke≤λ (resp. the stack Heckeλ ) classify-
ing successive modifications, see [33, 24]. For each scheme S, it classifies the
data consisting of

i) (n + 1) G-torsors V0 , . . . , Vn over X × S,
ii) n morphisms xi : S −→ X (1 ≤ i ≤ n) and we denote by (xi ) the graph
of xi ,
iii) a collection of isomorphisms
/ ∼ /
φi : Vi−1 / X ×S− (x ) −→ Vi / X ×S− (x ) , 1 ≤ i ≤ n,
i i

such that, for each geometric point s of S, the s-modification φi,s admits
an invariant (at xi (s)) bounded by λi (resp. the s-modification φi,s admits
λi as the invariant (en xi (s))).

We have immediately the following proposition:

Proposition 13.2. The morphism

Hecke≤λ −→ X n × BunG
(V0 , V1 , . . . , Vn , x1 , x2 , . . . , xn ) → (x 1 , x2 , . . . , xn , V0 )

is representable and projective. And the restriction of this morphism to the

open substack Heckeλ is smooth.
 On a counting problem for G-shtukas 323

Over the complementary X n \  of the union of the diagonals of X n , we get

a factorization property which will be useful later.

Proposition 13.3. Over the complementary X n \  of the union of the diago-

nals of X n , the Hecke stack of successive modifications Hecke≤λ is canonically
isomorphic to the fiber product of the morphisms Hecke≤λ j −→ FibG for
j = 1, 2, . . . , n.

Proof. See [24].

13.1.3 Stack of G-shtukas

We keep the notations of the previous section. We are ready to introduce the
stack of G-shtukas Cht≤λ associated to the collection of dominant coweights
λ of G. For each scheme S, Cht≤λ (S) classifies the data consisting of
i) (n + 1) G-torsors V0 , . . . , Vn over X × S,
ii) n morphisms xi : S −→ X (1 ≤ i ≤ n) and we denote by (xi ) the graph
of xi ,
iii) a collection of isomorphisms
/ ∼ /
φi : Vi −1 / X ×S− (x ) −→ Vi / X ×S− (x ) , 1 ≤ i ≤ n,
i i

such that, for each geometric point s of S, the s-modification φi,s admits
an invariant (at xi (s)) bounded by λi ,
∼
iv) an isomorphism V0σ := (Id X × Frob S )∗ V0 −→ Vn .
By composition, we get an isomorphism
/ ∼ /
t : V0σ / X×S−n (x ) −→ V0 / X ×S−n (xi )
.
i=1 i i=1

If the graphs (xi ) are disjoint, the isomorphisms φi can be determined by

t so that we have a simpler description of the stack of G-shtukas over the
complementary X n \  of the union of the diagonals of X n .
The stack of G-shtukas fits into the following cartesian diagram:

The horizontal maps are given as follows:

 ∼
V0 , . . . , Vn ; x1 , . . . , xn ; V0σ −→ Vn → V0 ,

V0 , . . . , Vn ; x1 , . . . , x n ) → (V0 , Vn ).
324 Ngo Dac Tuan

The vertical maps are given as follows:

 ∼ 
V0 , . . . , Vn ; x 1 , . . . , xn ; V0σ −→ Vn → V0 , . . . , Vn ; x1 , . . . , xn
V0 → (V0 , V0σ ).
As before, we define the open substack Chtλ of Cht≤λ by replacing the
inequalities ≤ λi in the condition (iii) by the equalities.

13.1.4 Level structures

By a level of X , we always mean a finite subscheme of X .

Definition 13.4. Let I be a level of X and S be a scheme. We consider a

 ∼
G-shtuka V = V0 , . . . , Vn ; x1 , . . . , xn ; V σ −→ Vn over S such that the asso-
0
ciated points x1 , x2 , . . . , xn do not meet I . By definition, an I -level structure
 consists of an isomorphism
of V
/ ∼
u : V0 / I ×S −→ G(O I ) × S
such that the following diagram is commutative:

Denote by Cht≤λ,I the stack classifying G-shtukas equipped with an I -level

structure. We call the characteristic morphism the natural morphism
Cht≤λ,I −→ (X − I )n .
We are now presenting an equivalent definition of level structures [26]. We
consider G as the trivial G-torsor over X with the action of G on itself on the
left and G I the restriction of G on the finite subscheme I of X . We consider the
functor G  I which, for each X -scheme S, classifies the subgroup of automor-
phisms of G × X S which commute with the action of G on the left and induce
the identity action on G I × X S. This functor is representable by a group scheme
G  I on X . It is equipped with a map to G which is an isomorphism outside I .
In general, it is not flat, hence not smooth, we will consider the smooth group
scheme associated to it as explained in Raynaud [3, Théorème 3, page 61]. We
denote it by G I equipped with a canonical morphism G I → G  I .
It is proved [26] that a shtuka V  equipped with an I -level structure is
equivalent to the data consisting of
 On a counting problem for G-shtukas 325

i) (n + 1) G I -torsors V0 , . . . , Vn over X × S,
ii) n morphisms xi : S −→ X (1 ≤ i ≤ n) and we denote by (xi ) the graph
of xi ,
iii) a collection of isomorphisms
/ ∼ /
φi : Vi −1 / X ×S− (x ) −→ Vi / X ×S− (x ) , 1 ≤ i ≤ n,
i i

such that, for each geometric point s of S, the s-modification φi,s admits
an invariant (at xi (s)) bounded by λi (resp. the s-modification φi,s admits
λi as the invariant (en xi (s))).
∼
iv) an isomorphism V0σ := (Id X × Frob S )∗ V0 −→ Vn .
In other words, we have defined a shtuka associated to the group scheme G I
instead of G.

13.1.5 Affine grassmannian and local model

We keep the notations of the previous section. We recall now the notion of the
local and global affine grassmannians. Let λ be a dominant coweight of G.
We define the global affine grassmannian Gr≤λ as follows. For each scheme S,
Gr≤λ (S) classifies the data consisting of
i) two G-torsors V and V  over X × S,
ii) a morphism x : S −→ X , and denote by (x) the graph of x,
iii) an isomorphism
/ ∼ /
φ : V / X ×S− (x) −→ V  / X×S− (x)
such that, for each geometric point s of S, the s-modification φs admits an
invariant (at x(s)) bounded by λ,
iv) and a trivialization of V  , i.e. an isomorphism between V  and the trivial
G-torsor on X × S.
Recall now the definition of the local affine grassmannian Gr. It is an ind-
scheme which classifies G-torsors over the formal local disc equipped with a
trivialization on the punctured disc. The closed sub-ind-scheme Gr≤λ (resp.
the locally closed sub-ind-scheme Grλ ) of Gr are defined by requiring that the
Hodge invariant is bounded by λ (resp. equal to λ).

Proposition 13.5. Locally for the étale topology, the natural morphism
Gr≤λ −→ X

is a fibration with fiber Gr≤λ .

326 Ngo Dac Tuan

Proof. See [33, Lemma A.8].

More generally, let λ = (λ1 , . . . , λn ) be a collection of dominant coweights

of G. Then, the global affine grassmannian Gr≤λ is defined as the stack which,
for each scheme S, classifies the data consisting of

i) G-torsors V0 , . . . , Vn sur X × S,
ii) morphisms xi : S −→ X (1 ≤ i ≤ n), and denote by (xi ) the graph of
xi ,
iii) isomorphisms
/ ∼ /
φi : Vi−1 / −→ Vi /
X ×S− (xi )
, 1 ≤ i ≤ n,
X ×S− (x i )

such that, for each geometric point s of S, the s-modification φi,s admits an
invariant (at xi (s)) ≤ λi (resp. the s-modification φi,s admits an invariant
(at xi (s)) λi ),
iv) and a trivialization of Vn , i.e. an isomorphism between Vn and the trivial
G-torsor over X × S.

It is shown that:

Proposition 13.6. i) Over the complementary X n \  of the union of the


diagonals of X n , Gr≤λ is cannonically isomorphic to the product i Gr≤λi .
ii) Gr≤λ (resp. Grλ ) is a local model of Cht≤λ (resp. Chtλ ). In other words, for
every point u of Cht≤λ , there exists an étale neighborhood Uu −→ Cht≤λ
of u and an étale morphism Uu −→ Gr≤λ .

Proof. See [33, Lemma A.8] and [33, Theorem 2.20].

13.1.6 Morphisms of partial Frobenius

Let λ = (λ1 , λ2 , . . . , λn ) be a collection of dominant coweights of G and let
I be a finite subscheme of X . We have introduced the stack Cht I,≤λ classi-
fying G-shtukas with I -level structures. It is equipped with the characteristic
morphism
Cht I,≤λ −→ (X − I )n .

For each integer i such that 1 ≤ i ≤ n, we consider the morphism of partial

Frobenius of (X − I )n given by

σi (x1 , . . . , xn ) = (x1 , . . . , xi −1 , Frob(xi ), xi +1 , . . . , xn ).

 On a counting problem for G-shtukas 327

Denote by  the set of points (x 1 , . . . , xn ) ∈ (X − I )n such that Frobm (xi ) =

Frobn (x j ) for all couples of positive integers (m, n) and all couples of indexes
i = j; then  is open in (X − I )n .

Proposition 13.7. Over the open subscheme , we have an isomorphism of

partial Frobenius:

σi : Cht I,≤λ ×(X −I )n  −→ Cht I,≤λ ×(X −I )n 

which suits into the commutative diagram:

Moreover, the open substack Cht I,λ ×(X −I )n  is stable under the isomor-
phisms of partial Frobenius.

Proof. To simplify the exposition, we can suppose that i = 1. Let S be a

scheme and let
∼

V = (V0 , V1 , . . . , Vn , x1 , . . . , xn , V0σ −→ Vn , ι)

be an object of the groupoid Cht I,≤λ ×(X−I )n (S). We will delete the G-torsor
V0 from the beginning of the chain and add V0σ at the end to obtain a new object
∼
(V0 , V1 , . . . , Vn , x2 , . . . , xn , σ1 (x1 ), V0σ −→ Vn , ι )

of Cht I,≤(λ2 ,...,λn ,λ1 ) . The associated points in (X −I )n is (x2 , . . . , , xn , σ1 (x1 )).
Since we are working over the open subscheme , the factorization property
13.3 implies that this object induces an object σ1 ( V) of Cht I,≤λ over
(σ1 (x1 ), x2 , . . . , xn ). It is obvious that σ1 verifies all the required properties.
The proof is finished.

The morphisms of partial Frobenius commute with each other and their
composition gives the usual Frobenius.

13.1.7 Truncating parameters

Definition 13.8. Let p be a rational dominant coweight in X ∗ T der ⊗ Q. We
will say that p is sufficiently convex if, for each simple root α of G, α, p is
sufficiently big.
328 Ngo Dac Tuan

Let p be a rational dominant coweight in X ∗ T der ⊗ Q. Let S be a scheme

and V be a G-torsor over the fiber product X × S. We will say that p(V) ≤ p if,
for each geometric point s ∈ S, each B-reduction VsB of Vs and each dominant
weight λ ∈ X ∗ T ad , we have the inequality:

B
deg(Vs,λ ) ≤ λ, p,

where Vs,λB is the line bundle induced by V B via the character λ.

s
≤p
Then, we define the substack BunG of BunG which, for each scheme S,
classifies G-torsors V over the product X × S such that p(V) ≤ p. It is
≤p
proved that the substack BunG is an open substack of the stack BunG , see
[33, Lemma A.3].
Then, let λ = (λ1 , λ2 , . . . , λn ) be a collection of dominant coweights of G
and let I be a finite subscheme of X . Recall that we have defined the stack
Cht I,≤λ classifying G-shtukas with I -level structures. We define the open sub-
≤p
stack Cht I,≤λ of Cht I,≤λ by requiring that the associated G-torsor V0 satisfies
the condition that p(V0 ) ≤ p.

Remark 13.9. If the reductive group G is the general linear group GLn , we
rediscover the notion of truncation by the polygon of Harder-Narasimhan for
vector bundles of rank n over the curve X . It gives a definition of truncating
parameters for GLn -shtukas (or shtukas of rank n) over X . Lafforgue intro-
duced a slightly different definition of truncating parameters for shtukas of
rank n over X , see [13, Chapter II, section 2]. However, he observed that if the
parameter p is sufficiently convex, then the two truncating parameters give the
same open substack. We refer the reader to [13] for more details.

13.2 Hecke operators

13.2.1 Notations

We denote by A the ring of adèles of F, OA = x∈|X | Ox the ring of integers.

Let K = x∈|X | K x be the maximal compact subgroup of G(A) defined by
K x = G(Ox ) for each place x of F. For each finite set of places T of X ,
denote by
 
AT = Fx , OT = Ox ,
x∈T x∈T
AT = A/AT , O T = OA /OT ,
 On a counting problem for G-shtukas 329

and
 
KT = Kx , KT = Kx .
x∈T x ∈T
/

We consider a level I of X which does not meet T . Denote by K I (resp.

K IT ) the kernel of the surjective map K −→ G(O I ) (resp. K T −→ G(O I )).

Then we get K I = x∈|X | K I,x where K I,x is the open compact subgroup of
K x defined as the kernel of the map K x → G(O I ).
The group G(A) (resp. G(AT )) is equipped with a Haar measure dg (resp.
dg T ). It is normalized such that dg(K ) = 1 (resp. dg T (K T ) = 1). We denote
by dgx the normalized Haar measure of G(Fx ) such that dgx (K x ) = 1. Hence,

we obtain dg = dg T × x∈T dgx .
For each place x, we denote by H I,x the vector space of locally constant
functions with compact support f : G(Fx ) → Q, invariant on the left and on
the right by the action of the open compact subgroup K I,x . The characteristic
functions φβx of double-classe βx ∈ K I,x \G(Fx )/K I,x form a basis of H I,x .
The convolution product with respect to the Haar measure dgx on the vector
space H I,x equips it with an algebra structure. For each finite set of places T 
B
of X , we denote by H I,T  = x∈T  H I,x and define H I (resp. H I ) as the
T

inductive limit of H I,T  where T runs through the set of all the finite sets of
places of X (resp. of X − T ): if T  ⊂ T  , we have an injective map
C C
H I,x → H I,x
x∈T  x∈T 

defined by
C
φ →φ⊗ 1 K I,x
x∈T  −T 

where 1 K I,x is the characteristic function of K I,x . The algebra H I has a basis
B  
x∈T  φβx indexed by the set of all the couples (T , (βx ) x∈T  ) where T is a
finite set of places of X and βx ∈ K I,x \G(Fx )/K I,x is a double class modulo
by the obvious equivalent relation.

13.2.2 Hecke operators

Let I be a level of X and T  be a finite set of places of X which can eventually
meet the level I . For each place x ∈ T  , we are given a double class βx ∈
B
K I,x \G(Fx )/K I,x . The Hecke operator x∈T  βx acts on Chtλ,I ⊗(X −I )n (X −
I − T  )n and Cht≤λ,I ⊗(X −I )n (X − I − T  )n by correspondence which will be
introduced in this section.
330 Ngo Dac Tuan

Let x be a closed point of T  . Let Y be a flattening neighborhood of x in

X . A shtuka  V ∈ Chtλ,I ⊗(X −I )n (X − I − T  )n (S) induces by restriction to
Y a G I -torsor VY over the fiber product Y × S equipped with an isomorphism
∼
VYσ −→ VY . Since G I has connected fibers, it is equivalent to give a S-point
4
of the classifying stack of the finite group G I (OY ), i.e. a G I (OY )-torsor VY
over S. If Y runs through the set of all flattening neighborhoods of x in X , we
4
obtain a K I,x -torsor Vx over S where K I,x = G I (Ox ).
A correspondence of type (T  , (βx )x∈T  ) between two shtukas V,  V ∈

Chtλ,I ⊗(X −I )n (X − I − T ) (S) with the same associated points x1 , . . . , x n
n

consists of an isomorphism
∼
 (X −T  )×S −→
t  : V|  |(X −T  )×S
V
between the restrictions of V  and V  over the open complementary of T  , veri-
4
fying the following condition. For each closed point x ∈ T  , let Vx and V  4x
be the K I,x -torsors over S associated to these shtukas  V and V  . The map
4
t gives an isomorphism between the G (Fx )-torsors induced by Vx et V  4x .
 I

This isomorphism gives in turn a function locally constant on S in the set of

double-classe K I,x \G I (Fx )/K I,x . We require that this function is constant and
its value is βx . Moreover, we have the following result:

Proposition 13.10. For each (T  , (βx )x∈T  ) as above, let consider the functor
+(T  , (βx )x∈T  ) over
Chtλ,I ×(X−I )n (X − I − T  )n
which associates to any G-shtuka V  with an I -level structure such that its
associated points do not meet T  the category of couples consisting of a G-
 with an I -level structure and the same associated points as those of
shtuka V

V and a correspondence between V  and V of type (T  , (βx )x∈T  ). Then, the

functor +(T , (βx )x∈T  ) is representable by a finite and étale morphism over
Chtλ,I ×(X−I )n (X − I − T  )n .

Remark 13.11. A similar argument to that given by Lafforgue [13, Chap-

ter I, section 4] in the case G = GLn implies that we could extend the above
construction and define a map from the Hecke algebra H I to the algebra of
étale correspondences in the generic fiber of Chtλ,I over (X − I )n . Moreover,
this action commutes with the morphisms of partial Frobenius, hence with the
Frobenius.

Remark 13.12. Let p be a truncating parameter sufficiently convex with

respect to λ. We consider the correspondence +(T  , (βx )x∈T  ) and we suppose
 On a counting problem for G-shtukas 331

that, for each x ∈ T  , βx lies in K I,x \G(Ox )/K I,x . Then, the open substack
≤p
Chtλ,I ×(X−I )n (X − I − T  )n is stable by this correspondence.

13.3 Counting the number of fixed points by a power

of the Frobenius
13.3.1 Notations
Let s ≥ 1 be a positive integer, x1 , . . . , xn ∈ X (Fq s ) be distinct points
of (X − I ) with value in Fq s . Let T = {x 1 , . . . , x n } where x i is a geo-
metric point over xi . We will fix a finite set of places T  of X which does
not meet the set {x1 , . . . , xn } and for each place x ∈ T  , we fix a double-
class βx ∈ K I,x \G(Fx )/K I,x . Let +(T  , (βx )x∈T  ) be the associated Hecke
correspondence defined in the previous section.
The main goal of this article is to give an expression of the number of fixed
points σ s ◦ +(T  , (βx )x∈T  ) in the fiber Chtλ,I (x 1 , . . . , x n ) de Chtλ,I over
(x 1 , . . . , x n ) by a formula which could be compared to the geometric side
of the Arthur-Selberg trace formula of G for a suitable test function.

13.3.2 Fixed points by a power of the Frobenius

We will give the first description of the category of fixed points σ s ◦
+(T  , (βx )x∈T  ) in Chtλ,I (x 1 , . . . , x n ). Let C I (T, T  , s) be the groupoid of
triples (V, t, t  ) consisting of:

i) V is a G I -torsor over the/geometric curve

/ X,
∼
ii) a T -modification t : V σ / X −T −→ V / X −T ,
 s / ∼ /
iii) a T -modification t  : V σ /  −→ V /
X −T
,
X −T

such that the maps t et t  commute, i.e. the following diagram is commutative:

The definition of an isomorphism between such objects is obvious.

We introduce the sub-category CλI ,β  (T, T  , s) whose objects are triples
T T
(V, t, t  ) ∈ C I (T, T  , s) which verify the following conditions:
332 Ngo Dac Tuan

iv) the invariant invx i (t) is equal to λi ,

v) for each place x ∈ T  , the correspondence t  between the shtuka (V, t) and
the shtuka σ s (V, t) is of type βx , see Section 13.2.2.

In the triple (V, t, t  ) ∈ CλI ,β  (T, T  , s), the couple 

V = (V, t) defines a
T T
G-shtuka with an I -level structure and t  defines a correspondence between
σ s and V
V  of type (T  , (βx )x∈T  ). Hence, we obtain:

Proposition 13.13. The groupoid CλI ,β  (T, T  , s) is the groupoid of fixed

T T
points of σ s ◦ +(T  , (βx )x∈T  ) in Chtλ,I (x 1 , . . . , x n ).

Forgetting the level structures, we will denote by

C(T, T  , s) et CλT ,βT  (T, T  , s)

the resulting groupoids. There are the forgetting functors

C I (T, T  , s) → C(T, T  , s)

et
CλI (T, T  , s) → CλT ,βT  (T, T  , s).
T ,β T 

13.3.3
We consider the counting number LefλI (T, T  , s) defined by the formula
T ,βT 

LefλI (T, T  , s) := dg(K I ) · #CλI (T, T  , s)

T ,β T  T ,βT 
1
= dg(K I ) ·
# Isom(V, t, t  )
where the sum runs through the set of representatives of isomorphism classes
of objects (V, t, t  ) of CλI ,β  (T, T  , s). This number is in general infinite
T T
because the connected components of stack of G-shtukas are not always of
finite type.
Therefore we suggest to modify the counting problem as follows. Let Z be
the center of G. We fix a discrete sub-group J ⊂ Z (AT ∪I ) ⊂ Z (A) which is
cocompact in Z (A)/Z (F). The group J acts on Chtλ,I (x 1 , . . . , x n ) by Hecke
correspondences. We will consider the quotient of the stack Chtλ,I (x 1 , . . . , x n )
by the action of J . It is equivalent to add formally isomorphisms between an
object V  ∈ Chtλ,I (x 1 , . . . , x n ) and its image under the action of an element
j ∈ J . By applying the same procedure to the groupoids C I (T, T  , s) and
CλI ,β  (T, T  , s) we obtain the groupoids C I (T, T  , s) J and CλI ,β  (T, T  , s) J
T T T T
 On a counting problem for G-shtukas 333

We define the number

(T, T  , s) J (T, T  , s) J
I,≤ p I,≤ p
Lefλ := dg(K I ) · #Cλ
T ,βT  T ,β T 
1
= dg(K I ) ·
# Isom(V, t, t  )
where the sum runs through the set of representatives of isomorphism classes
of objects (V, t, t  ) of CλI ,β  (T, T  , s) J with p(V) ≤ p.
T T
If we restrict the formula to a so-called elliptic part, the truncating function
will be always 1 and we will get the equality:

LefλI,ell,β  (T, T  , s) J = dg(K I ) · #CλI,ell 

,β  (T, T , s) J
T T T T
1
= dg(K I ) ·
# Isom(V, t, t  )
where the sum runs through the set of representatives of isomorphism classes
of so-called elliptic objects (V, t, t  ) of Cλ,β
I (T, T  , s) J . In our previous work
T
[26, 28], we have found an expression of this number which is a sum indexed
by the elliptic elements of G(F) of products of twisted orbital integrals and
orbital integral. In this article, we extend this result to get an integral expression
of Lefλ ,β  (T, T  , s) J .
I,≤ p
T T

13.4 Description of isogeny classes

13.4.1 Functor of generic fiber
We consider the groupoid C(T, s) of triples (V , τ, τ  ) consisting of

i) V is a G-torsor over F̄ = F ⊗Fq k,

ii) τ : V σ −→ V is a σ -linear application,
iii) τ  : V σ −→ V is a σ s -linear application,
s

which verify the following conditions:

a) τ et τ  commute with each other,

b) for each place x ∈
/ T , (Vx , τx ) is isomorphic to

! Fq k), id ⊗
(G(Fx ⊗ ! Fq σ ),

c) for each place x ∈ T , (Vx , τx ) is isomorphic to

! Fq Fq s ⊗
(G(Fx ⊗ ! Fq s k), id ⊗
! Fq s σ s ).
334 Ngo Dac Tuan

We will define a functor called the functor of generic fiber from the cate-
gory C(T, T  , s) to the category C(T, s). To each triple (V, t, t  ) of C(T, T  , s),
we associate the object (V , τ, τ  ) in C(T, s) which is called the generic fiber.
This construction can be extended without difficulties to triples (V, t, t  ) de
C I (T, T  , s) using the forgetting functor.
Denote by V the generic fiber of V. It is a G-torsor over F ⊗Fq k. The modi-
fications t and t  induce the isomorphisms τ : V σ −→ V and τ  : V σ −→ V .
s

It is proved [26] that this triple (V , τ, τ  ) is an object in the groupoid C(T, s).
Since the generic fiber is stable by the Hecke action of elements in J , we obtain
a well-defined functor C(T, T  , s) J → C(T, s).

13.4.2 Stable conjugacy classes and Kottwitz’s triples

Following [26], to each triple (V, τ, τ  ) ∈ C(T, s), we will associate a
Kottwitz’s triple (γ0 ; (γx )x ∈T
/ , (δ x ) x∈T ) where

a) γ0 is a stable conjugacy class of G(F),

b) for each place x ∈/ T , γx is a conjugacy class of G(Fx ) which is stably con-
jugate to γ0 . Moreover, for almost every place x ∈/ T , γx is conjugate to γ0 ,
c) for each place x ∈ T , δx is a σ -conjugacy class of G(Fx ⊗ ˆ Fq Fq s ) whose
norm is stably conjugate to γ0 .

We will say that (V, t, t  ) or (V, τ, τ  ) is semisimple (resp. elliptic) if the

associated conjugacy class γ0 of G(F) is semisimple (resp. elliptic). Note that
the notion of Kottwitz’s triples is borrowed from the work of Kottwitz on a
similar counting problem on Shimura varieties [11, 12].

13.4.3 Basic elements after Kottwitz

In this section, we recall important results of Kottwitz on the set B H of σ -
conjugacy classes associated to a connected reductive group H over a non-
archimedian local field. We refer the reader to the excellent articles [9, 30] for
proofs and details.
Let x be a place of F and H be a connected reductive group over Fx . Denote
by x the Galois group of Fx and by H ! the dual group of H . We consider
! Fq k) equipped with the Frobenius action σ which acts trivially
the set H (Fx ⊗
on Fx and acts by the usual Frobenius on k. We define B H (Fx ) the set of
!Fq k).
σ -conjugacy classes of H (Fx ⊗
 On a counting problem for G-shtukas 335

Let T be a maximal torus of H defined over Fx , we denote by W the

corresponding Weyl group. Kottwitz has introduced the Newton map:
ν : B H (Fx ) −→ (X ∗ T / W ) x .
Then, Kottwitz has defined the set B Hbasic (Fx ) of so-called basic elements
characterized by the following property: an element b in B H (Fx ) is basic if
and only if its image ν(b) by the Newton map factorizes through the center
Z (H ) of H . Kottwitz proved:

Theorem 13.14 (Kottwitz). i) An element b in B H (Fx ) is basic if and only

if the group of twisted centralizers
! Fq k) : g −1 bσ (b) = b}
Jb (F) = {g ∈ H (Fx ⊗
is the group of Fx -points of an interior form of H .
ii) There exists an isomorphism of functors from the category of connected
reductive groups over Fx to the category of the set
!) x )∗
B Hbasic (Fx ) −→ (Z ( H
of the set B Hbasic (Fx ) to the group of algebraic characters of Z ( H!) x .
This isomorphism of functors gives an isomorphism of the subgroup
!) x which are trivial on
H 1 (Fx , H ) to the subgroup of characters of Z( H
the neutral component.
In particular, with the above structure, the set B Hbasic (Fx ) is an abelian
group.
iii) The isomorphism of functors B Hbasic (Fx ) −→ (Z ( H !) x )∗ can be
extended to a morphism of functors
!) x )∗ .
B H (Fx ) −→ (Z ( H

13.4.4 Existence of a global element.

Let γ be a semisimple element of G(F). Denote by H the centralizer G γ of γ
in G; it is a connected reductive group over F since G der is simply connected.
We get a natural map

B H (F) −→ B H (Fx )
x

that, for each place x, sends a global σ -conjugacy class b ∈ B H (F) to its
local σ -conjugacy class bx ∈ B H (Fx ). Following [26], we define the global
invariant
inv(b) : Z ( Ĥ ) −→ C∗
336 Ngo Dac Tuan

to be the sum of the local invariants

inv(bx )|Z ( Ĥ) : Z ( Ĥ ) −→ Z ( Ĥ ) x

−→ C∗ .

and it is proved that inv(b) = 0, i.e. the global invariant associated to b is the
trivial character of Z( Ĥ ) . Under certain hypothesis, this condition is suffi-
cient to prove the existence of a global element, see [28]. Remember that the
case of tori is already proved in [26] and also known by Varshavsky [34].

Proposition 13.15. Let H be a connected reductive group over F. Given the

elements bx ∈ B H (Fx ) which are supposed to be basic and trivial for almost

every place x and which verify the condition x inv(bx ) = 0. Then, there
exists an element b of the form b = hz with h ∈ H 1 (F, H ) and z ∈ Z (H )( F̄)
such that b = bx in B H (Fx ).
Moreover, the number of such elements is equal to

ker1 (F, H ) = ker(H 1 (F, H ) −→ H 1 (Fx , H )).
x

13.4.5 A necessary and sufficient condition for stable

conjugacy classes
Definition 13.16. Let γ0 be a stable conjugacy class of G(F). We say that γ0
is a (T, s)-norm if, for each place x ∈ T , there exists a σ -conjugacy class δx
of G(Fx ⊗ ˆ Fq Fq s ) whose norm is stably conjugate to γ0 .

Let γ0 be a stable conjugacy class of G(F) which is also a (T, s)-norm.

pk
There exists a positive integer k big enough such that the element γ0 is
pk
semisimple. We choose a such element and denote it by γ , i.e. γ = γ0 .
For each place x ∈ T , we fix a σ -conjugacy class δx de G(Fx ⊗ ˆ Fq Fq s ) whose
norm is stably conjugate to γ0 . There is an element gx ∈ G( F̄x ) such that
γ0 = gx−1 N δx gx . Then the element τx := gx−1 δx σ (gx ) commutes with γ0 ,
hence, it belongs to G γ0 ( F̄x ). Since G γ0 is a subgroup of G γ , τx defines an
element, says invx (γ0 , s), in BG γ (Fx ).

Lemma 13.17. The class invx (γ0 , s) is basic in BG γ (Fx ). Moreover, we have:

ν(γ0 ) ν(γ )
ν(invx (γ0 , s)) = = k
s p s
where ν is the Newton map constructed by Kottwitz.
 On a counting problem for G-shtukas 337


Proof. Put τx = gx−1 σ s (gx ), then γ0 = τxs τx−1 . Since γ0 and τx commute with
each other, τx and τx commute with each other as well. Since τx is of torsion, it
implies that γ0m = τxms for a certain positive integer m. So we get an equality:
mp k mpk s
γ m = γ0 = τx .
It is obvious that γ is always basic in BG γ (Fx ), we deduce that invx (γ0 , s) is
basic in BG γ (Fx ) and that
ν(γ0 ) ν(γ )
ν(invx (γ0 , s)) = = k .
s p s
The proof is finished.
The above lemma implies that the class invx (γ0 , s) ∈ BG γ ,basic (Fx ) whose
definition depends on the choices of δx and gx is well defined in BG γ ,basic (Fx )
modulo the action of the torsion group H 1 (Fx , G γ ).
Given the local invariants (invx (γ0 , s))x∈T , we define the global invariant
invT (γ0 , s) : Z (Ĝ) −→ C∗
as follows. For each place x ∈ T , the local invariant invx (γ0 , s) is a basic ele-
ment in BG γ ,basic (Fx ). Following Kottwitz, it induces a character invx (γ0 , s) :
Z (Ĝ) x −→ C∗ , then by restriction, we get a character invx (γ0 , s)|Z (Ĝ) :
Z (Ĝ) −→ C∗ . The global invariant invT (γ0 , s) is defined as the sum of local
characters:
invT (γ0 , s) = invx (γ0 , s).
x∈T

Definition 13.18. With the above notations, we say that a stable conjugacy
class γ0 of G(F) is (T, s)-admissible if it is a (T, s)-norm and that its global
invariant invT (γ0 , s) lies in Im( x∈|X | H 1 (Fx , G γ ) −→ H 1 (F, G γ )).

The rest of this section is devoted to the proof of the following proposition:

Proposition 13.19. i) Let (V, τ, τ  ) be an element of C(T, s); denote by

γ0 its associated stable conjugacy class in G(F). Then, γ0 is (T, s)-
admissible.
ii) Conversely, let γ0 be a stable conjugacy class which is (T, s)-admissible.
Then, there exists an element (V , τ, τ  ) in C(T, s) whose associated sta-
ble conjugacy class is γ0 and that τ ∈ Z (G γ0 )( F̄ ). Moreover, any triple
(V, τ0 , τ0 ) in C(T, s) whose associated stable conjugacy class is γ0 can be
written in the form: τ0 = hτ with h ∈ H 1 (F, G γ0 ).
In particular, the set V (γ0 ) of such triples is in bijection with
H (F, G γ0 ).
1
338 Ngo Dac Tuan

Proof. i) Let (V, τ, τ  ) be an element in C(T, s) and let (γ0 ; (γx )x ∈T / ,

(δx )x∈T ) be its associated Kottwitz’s triple. By definition of a Kottwitz’s
triple, it is clear that γ0 is always a (T, s)-norm. Moreover, for each place
x ∈ T , the local invariants invx (γ0 , s) and invx (γ0 , γx , δx ) are the same. So
we get the equality:

invT (γ0 , s) = invx (γ0 , s) = invx (γ0 , γx , δx ).

x∈T x∈T

We know that
a) For each place x ∈ / T , the local invariant invx (γ0 , γx , δx ) lies in the
subgroup H 1 (Fx , G γ ) of BG γ ,basic (Fx ).

b) The global invariant inv(γ0 , γx , δx ) = x∈|X | invx (γ0 , γ x , δ x ) van-
ishes.
It implies that invT (γ0 , s) lies in Im( x∈|X | H 1 (Fx , G γ ) −→
H 1 (F, G γ )), i.e. γ0 is (T, s)-admissible.
ii) Let γ0 a stable conjugacy class of G(F) which is (T, s)-admissible. Since
γ0 is (T, s)-admissible, we can add to the collection (invx (γ0 , s))x∈T
the elements (invx (γ0 , s))x ∈T/ ∈ / H (Fx , G γ ) such that the global
x ∈T
1

invariant

invx (γ0 , s)
x∈|X |

vanishes. Following [28], there exists an element τ0 ∈ G γ ( F̄) such that,

for every place x, we have the equality invx (τ0 ) = invx (γ0 , s). Further, this
element is of the form hτ with h ∈ H 1 (F, G γ ) and τ ∈ Z (G γ )( F̄). In
particular, τ lies in Z(G γ0 )( F̄).
Put τ  = γ0−1 τ s , then we claim that (V , τ, τ  ) is an element of C(T, s)
whose associated stable conjugacy class is γ0 . In fact, we have to show that
this triple satisfies the three properties required in the definition of C(T, s),
see Section 13.4.1. Since τ et γ0 commute with each other, by construc-
tion, τ and τ  commute too. So the first property is shown. Next, for each
place x ∈ / T , since invx (τ0 ) lies in H 1 (Fx , G γ ), so does invx (τ ) . Since
the derived group G der of G is simply connected, the group H 1 (Fx , G)
vanishes, hence we get the second property. Finally, for each place x ∈ T ,
since γ0 and τ s have the same image by the Newton map, it implies that τ 
is of torsion. The same argument implies the third property.
To conclude, we remark that by construction, the stable conjugacy class
associated to the triple (V, τ, τ  ) is γ0 . The proof is finished
 On a counting problem for G-shtukas 339

13.4.6 The automorphism group of (V , τ, τ  )

Let (V, τ, τ  ) be an element of C(T, s) and let (γ0 , (γx )x∈T , (δx )x ∈T
/ ) be the
associated Kottwitz’s triple. In this section, we will show that the automor-
phism group Aut(V, τ, τ  ) of (V, τ, τ  ) is an interior form Jγ0 of G γ0 defined
over F.
Choose a representative of the conjugacy class γ0 in G(F), says γ0 . Suppose
that γ0 = τ s τ −1 . Since γ0 and τ commute, τ acts on the group of centralizers
G γ0 (F ⊗Fq k). In the one hand, the automorphism group Aut(V, τ, τ  ) is the
subgroup of fixed points by τ of the semi-direct product G γ0 (F ⊗Fq k)  τ .
On the other hand, G γ0 is the subgroup of fixed points by σ of the semi-direct
product G γ0 (F ⊗Fq k)  σ . It suffices to verify that for some positive integer
n, we have an isomorphism of semi-direct products:
∼
G γ0 (F ⊗Fq k)  τ n  −→ G γ0 (F ⊗Fq k)  σ n .

In fact, there exists a positive integer m such that τ ms is central in G γ (F).

We write n = ms and obtain the desired isomorphism
∼
G γ0 (F ⊗Fq k)  τ n  −→ G γ0 (F ⊗Fq k)  σ n .

By the Hasse’s principle for adjoint groups, the automorphism group Jγ0 =
Aut(V, τ, τ  ) which is an interior form of G γ0 is completely determined by its
local components Jγ0 ,x :

● For each place x ∈ / T , the local component Jγ0 ,x is the centralizer of γx in

G(Fx ).
● For each place x ∈ T , the local component Jγ0 ,x is the twisted centralizer of
δx in G(Fx ⊗ˆ Fq Fq s ).

13.5 Description of an isogeny class

13.5.1 Adelic description of fixed points
Let V be a G I -torsor over k. The generic fiber V of V is a G-torsor over
F̄ = F ⊗Fq k, hence it is trivial. We will fix an isomorphism between V and
the trivial G-torsor over F̄. Then, to give a G I -torsor V over k equipped with a
trivialization of the generic fiber is equivalent to give, for each place x ∈ |X |,
an element gx in G(Fx ⊗ ! Fq k)/K I,x which is trivial almost everywhere. If we
forget the choice of the trivialization, we will say that two such data (gx )x∈|X|
and (gx )x∈|X | are equivalent if there exists an element g ∈ G( F̄) such that
gx = ggx .
340 Ngo Dac Tuan

Let (V, τ, τ  ) be an element of C I (T, T  , s). Then, to give a such element is

equivalent to give the data consisting of

i) two elements τ, τ  ∈ G( F̄),

ˆ Fq k)/K I,x which is trivial
ii) for each place x ∈ |X |, an element gx of G(Fx ⊗
almost everywhere,

which verify the following conditions:

1) τ and τ  commute, i.e. τ τ  = τ  τ .

2) For each place x ∈ T , gx−1 τ σ (gx ) ∈ K x 2xλx K x .
3) / T , gx−1 τ σ (gx ) ∈ K x .
For each place x ∈
4) For each place x ∈ T , gx−1 τ  σ s (gx ) ∈ K I,x .
5) / T , gx−1 τ  σ s (gx ) ∈ K I,x βx K I,x .
For each place x ∈

Two such data (τ1 , τ1 , (g1,x )x∈|X | ) et (τ2 , τ2 , (g2,x )x∈|X | ) are equivalent if
there exists an element g ∈ G( F̄) such that τ2 = g −1 τ1 σ (g), τ2 = g −1 τ1 σ s (g)
and for each place x, g2,x = gg1,x .

13.5.2 Adelic description of the functor of generic fiber

Let (V, τ, τ  ) be an element of C I (T, T  , s) and let (τ, τ  , (gx )x∈|X | ) be the
associated adelic data as above. The functor of generic fiber will associate to it
the data consisting of two elements τ, τ  ∈ G( F̄) which verify the following
conditions:

1) τ and τ  commute, i.e. τ τ  = τ  τ .

2b) For each place x ∈ ˆ Fq k) such
/ T , there exists an element g0,x ∈ G(Fx ⊗
−1
that g0,x τ σ (g0,x ) = 1.
3b) For each place x ∈ T , there exists an element g0,x ∈ G(Fx ⊗ ˆ Fq k) such
−1  s
that g0,x τ σ (g0,x ) = 1.

In fact, the functor of generic fiber consists of forgetting the local data
((gx )x∈|X | ). By a theorem of Lang, every element of K x can be written in
−1 ˆ Fq k). It implies that the last two
the form g0,x σ (g0,x ) for some g0,x ∈ G(Fx ⊗
conditions 2b) and 3b) are satisfied.
Let (τ, τ  ) be an element of C(T, s). We fix a choice of elements (g0,x )x∈|X |
as above. We will give an adelic description of the Kottwitz’s triple associated
to (τ, τ  ). Put γ0 = τ s τ −1 ; it is an element of G( F̄). We have already seen
that the stable conjugacy class of γ0 is fixed by the Frobenius, hence it contains
an element of G(F). If necessary replacing (τ, τ  ) by an equivalent datum, we
−1
can suppose that γ0 is in G(F). For each place x ∈ / T , since g0,x τ σ (g0,x ) = 1,
 On a counting problem for G-shtukas 341

−1
the element γx := g0,x γ0 g0,x is fixed by the Frobenius, hence it is an ele-
−1  s
ment in G(Fx ). For each place x ∈ T , since g0,x τ σ (g0,x ) = 1, the element
−1 ˆ Fq Fq s ). To con-
δx := g0,x τ σ (g0,x ) is fixed by σ , it is an element in G(Fx ⊗
s

clude, note that the triple (γ0 , (γx )x ∈T

/ , (δx )x∈T ) is exactly the Kottwitz’s triple
associated to (τ, τ  ).

13.5.3 Adelic description of an isogeny class. Choice of base points

Let (τ, τ  ) be an element of C(T, s). We will give an adelic description of fixed
points (V, τ, τ  ) over (τ, τ  ). As before, we always suppose that the element
γ0 = τ s τ −1 is an element in G(F). We fix a choice of base points (g0,x )x∈|X | .
To give a fixed point (V, τ, τ  ) over (τ, τ  ) is equivalent to give a collection of
elements (gx ∈ G(Fx ⊗ ˆ Fq k)/K I,x )x∈|X | which verify the following conditions:

1) They are trivial almost everywhere.

2) For each place x ∈ T , gx−1 τ σ (gx ) ∈ K x 2xλx K x .
3) For each place x ∈ / T , gx−1 τ σ (gx ) ∈ K x .
4) For each place x ∈ T , gx−1 τ  σ s (gx ) ∈ K I,x .
5) For each place x ∈ / T , gx−1 τ  σ s (gx ) ∈ K I,x βx K I,x .
−1
For each place x ∈ |X |, put h x = g0,x gx . The condition 3) requires
that, for each place x ∈ −1
/ |T |, h x σ (h x ) ∈ K x , it is equivalent to ask
that h x ∈ G(Fx )/K I,x . Similarly, the condition 4) requires that, for each
place x ∈ T , h −1 x σ (h x ) ∈ K x , which is equivalent to say that h x ∈
s

G(Fx ⊗ ˆ Fq Fq s )/G(Ox ⊗
ˆ Fq Fq s ).
−1
Recall that we have already defined γx := g0,x γ0 g0,x ∈ G(Fx ) and δx :=
−1 ˆ Fq Fq s ). With these notations, the conditions 2) and 5)
g0,x τ σ (g0,x ) ∈ G(Fx ⊗
are equivalent to the following:
λx
2b) For each place x ∈ T , h −1
x δx σ (h x ) ∈ K x 2x K x .
/ T , h −1
5b) For each place x ∈ x γ x σ (h x ) ∈ K I,x βx K I,x .
s

To summarize, we have proved:

Proposition 13.20. We keep all the above notations. Then, to give a fixed
point (V, τ, τ  ) over (τ, τ  ) is equivalent to give a collection of elements
(gx = g0,x h x )x∈|X | where the elements
 
(h x )x∈|X | ∈ G(Fx ⊗ ˆ Fq Fq s )/G(Ox ⊗
ˆ Fq Fq s ) × G(Fx )/K I,x
x∈T x ∈T
/

verify the following conditions:

342 Ngo Dac Tuan

λx
2b) For each place x ∈ T , h −1
x δx σ (h x ) ∈ K x 2x K x .
5b) For each place x ∈ −1
/ T , h x γx h x ∈ K I,x βx K I,x .

13.5.4 Adelic description of the truncating function

We keep the notations of the previous section. Following Section 13.1.7, for
each truncating parameter p, we have defined a characteristic function of the
set of G-torsors over k bounded by p. In adelic terms, it is just a characteristic
function over G(Ā) which is invariant on the left by G( F̄) and invariant on

the right by the sub-group x∈|X | K x . Let (V, τ, τ  ) be a fixed point and let
(gx = g0,x h x )x∈|X | be the corresponding adèle. Then, we have the equality:
1( p(V) ≤ p)) = 1( p((gx )x∈|X | ) ≤ p)).

13.6 Integral expression of the number of fixed points

for G-shtukas
13.6.1 Integral expression
In this section, we will give a formula for the number of fixed points
Lefλ ,β  (T, T  , s) J . Let (V, τ, τ  ) be an object of C(T, s), denote by
I,≤ p
T T
(γ0 ; (γx )x ∈T
/ , (δx )x∈T ) its associated Kottwitz’s triple. Then, the number of
isomorphism classes
1
dg(K I ) ·
# Isom(V, t, t  )
(V ,t,t  )

where the sum runs through the set of representatives of isomorphism classes
of triples (V, t, t  ) of CλI ,β  (T, T  , s) J satisfying the condition that the
T̄ T
generic fiber is isomorphic to (V, τ, τ  ) and that the shtuka V is bounded by p
is finite and is equal to
. 
 φλx (h −1
x δx σ (h x ))
J Jγ0 (F )\ ˆ Fq Fq s )×G(A T )
G(Fx ⊗
x∈T x∈T

× φβx (h −1
x γx h x ) × 1( p((g0,x h x )x∈|X | ) ≤ p)) · dh
x ∈T
/
where
● for each place x ∈ T , φλx is the spherical function
C C
φλ x = φλ y ∈ Hx ⊗Fq Fq s := Hy ,
y∈T ⊗Fq Fq s y∈T ⊗Fq Fq s
y|x y|x
 On a counting problem for G-shtukas 343

● for each place x ∈ T  , φβx is the characteristic function of the double class
βx ∈ K I,x \G(Fx )/K I,x ,
● for each place x ∈/ T ∪ T  , φβx is the characteristic function of the unit
double class of G(Ox )\G(Fx )/G(Ox ).

We have proved that there exists a triple (V , τ, τ  ) whose stable conjugacy

class is γ0 if and only if γ0 is (T, s)-admissible. Moreover, if γ0 is (T, s)-
admissible, the set V (γ0 ) of triples (V , τ, τ  ) whose stable conjugacy class is
γ0 is in bijection with H 1 (F, G γ0 ).
We obtain then an integral expression of the number of fixed points in the
stack of G-shtukas:

Theorem 13.21. With the notations as before, we have:

.
Lefλ ,β  (T, T  , s) J
I,≤ p
= 
T T
γ0 V (γ0 ) J Jγ0 (F)\ x∈T
ˆ Fq Fq s )×G(AT )
G(Fx ⊗
 
φλx (h −1
x δx σ (h x )) φβx (h −1
x γx h x ) · 1( p((g0,x h x )x∈|X | ) ≤ p)) · dh
x∈T x ∈T
/

where the first sum runs through all (T, s)-admissible stable conjugacy classes
γ0 of G(F), and the second sum runs through the set V (γ0 ) of triples (V , τ, τ  )
whose stable conjugacy class is γ0 .

13.6.2 The elliptic terms

Suppose that γ0 is elliptic. Since γ0 is semisimple, (V , τ, τ  ) is determined by
its associated Kottwitz’s triple (γ0 ; (γx )x ∈T
/ , (δx )x∈T ). Moreover, the charac-
teristic truncating function is always 1 so that the choice of base points does
not appear in this formula. The formula was obtained in the article [28]:

 
LefλI,ell,β  (T, T  , s) J = ker H 1 (F, G γ0 ) −→ H 1 (Fx , G γ0 ) ·
T T
(γ0 ;γx ,δ x ) x∈|X|
 
· vol(J Jγ0 (F)\Jγ0 (A)) Oγx (φβx ) TOδx (φλx )
x ∈T
/ x∈T

where the sum runs through all Kottwitz’s triples (γ0 ; (γx )x ∈T / , (δ x ) x∈T ) such
that γ0 is elliptic and that the global invariant inv(γ0 ; γx , δx ) vanishes.
344 Ngo Dac Tuan

13.7 The number of fixed points under the actions

of partial Frobenius
In this section, we give a slight generalization of the previous results and
obtain a formula for the number of fixed points under the actions of par-
tial Frobenius. This result generalizes a unpublished result of de Ngô Bao
Châu [23]. The proofs are almost straightforward. So we will only state the
results and we let the reader to fill the details of the proofs. We will recall
the definition of the partial Frobenius, then introduce the groupoid of fixed
points by the composition of a power of partial Frobenius with a Hecke
operator. Finally, we present an integral expression for the number of fixed
points.

13.7.1 Fixed points by the actions of partial Frobenius

Let s ≥ 1 be a positive integer and x1 , . . . , xn be distinct places of X − I . Put
T = {x 1 , . . . , x n } where x i is a geometric point over xi . Let s = (s1 , . . . , sn )
be a collection of positive integers verifying that, for every 1 ≤ i ≤ n, si is a
multiple of deg(xi ): we put si = deg(xi )si with some positive integer si . We fix
a finite set of places T  disjoint from {x1 , . . . , xn } and for each place x ∈ T  ,
we fix a double class βx ∈ K I,x \G(Fx )/K I,x . Let ∂ s = (σ1s1 , . . . , σnsn ) be a
composition of partial Frobenius and +(T  , (βx )x∈T  ) be the associated Hecke
correspondence over Chtλ,I (x 1 , . . . , x n ).
As before, we fix a discrete subgroup J ⊂ Z(AT ∪I ) ⊂ Z (A) which
is cocompact in Z (A)/Z (F). The group J acts on Chtλ,I (x 1 , . . . , x n )
by Hecke correspondences. We will consider the quotient of the stack
Chtλ,I (x 1 , . . . , x n ) by the action of J and the groupoid of the fixed points
Fixe(∂ s ◦ +(T  , (βx )x∈T  ), Chtλ,I (x 1 , . . . , x n ) J ) by the correspondence ∂ s ◦
+(T  , (βx )x∈T  ) in Chtλ,I (x 1 , . . . , x n ) J . We refer the reader to the article [23]
for a detailed description of this groupoid for D-shtukas. The generalization to
G-shtukas is straightforward.
We define the number

(T, T  , s) J (T, T  , s) J
I,≤ p I,≤ p
Lefλ = dg(K I ) · #Cλ
T ,βT  T ,βT 
1
= dg(K I ) ·
# Isom(V, t, t  )

where the sum runs through the set of representatives of isomorphism classes
of objects (V, t, t  ) of Fixe(∂ s ◦ +(T  , (βx )x∈T  ), Chtλ,I (x 1 , . . . , x n ) J ) with
p(V) ≤ p.
 On a counting problem for G-shtukas 345

13.7.2 Integral expression of the number of fixed points

We introduce a generalization of the category C(T, s), denoted by C(T, s)
classifying triples (V , τ, γ0 ) consisting of

i) V is a G-torsor over F̄ = F ⊗Fq k,

ii) τ : V σ −→ V is a σ -linear map,
iii) γ0 : V −→ V is a linear map,

which verify the following properties:

1) τ and γ0 commute with each other, i.e. γ0−1 τ σ (γ0 ) = τ ,

2) for each place x ∈
/ T , (Vx , τx ) is isomorphic to
! Fq k), id ⊗
(G(Fx ⊗ ! Fq σ ),

3) for each place xi ∈ T , (Vxi , γ0−1 τ si ) is isomorphic to

! Fq Fq si ⊗
(G(Fxi ⊗ ! F si k), id ⊗
!F si σ si ).
q q

The definition of isomorphisms between two such triples is immediate.

Let (V, τ, γ0 ) be an element in C(T, s). We will choose a trivialization of V ,
so τ and γ0 can be considered as elements of G( F̄). The condition 1) implies
that the stable conjugacy class γ0 is stable under the action of the Frobenius,
hence contains an element of G(F). From now on, we always suppose that γ0
is in G(F).

Definition 13.22. Let γ0 be a stable conjugacy class of G(F). We will say that
γ0 is a (T, s)-norm if, for each place xi ∈ T , there exists a σ -conjugacy class
δxi of G(Fxi ⊗ˆ Fq Fq si ) whose norm is stably conjugate to γ0 .

Let γ0 be a stable conjugacy class of G(F) which is semisimple and is a

(T, s)-norm. By a similar manner as before, we define the global invariant
invT (γ0 , s) which lies in (Z (Ĝ) )∗ modulo x∈|T | H 1 (Fx , G γ ).

Definition 13.23. With the previous notations, we will say that a sta-
ble semisimple conjugacy class γ0 of G(F) is (T, s)-admissible if
it is a (T, s)-norm and that its global invariant invT (γ0 , s) lies in
Im( x∈|X | H 1 (Fx , G γ0 ) −→ H 1 (F, G γ0 )).

We can extend this definition to all the stable conjugacy classes by following
the same procedure as described in Section 13.4.5.
346 Ngo Dac Tuan

Proposition 13.24. i) Let (V , τ, γ0 ) be an element in C(T, s) with γ0 ∈

G(F). Then, γ0 is (T, s)-admissible.
ii) Conversely, let γ0 be a stable conjugacy class which is (T, s)-admissible.
Then, there exists an element (V, τ0 , γ0 ) in C(T, s) with τ0 ∈ Z(G γ0 )( F̄).
Moreover, a triple (V , τ, γ0 ) in C(T, s) can be written in the form: τ = hτ0
with h ∈ H 1 (F, G γ0 ).
In particular, the set V (γ0 ) of such triples (V , τ, γ0 ) ∈ C(T, s) is in
bijection with H 1 (F, G γ0 ).

Let γ0 be a stable conjugacy class of G(F) which is (T, s)-admissible and let
(V, τ, γ0 ) be an element in C(T, s). There exist elements (g0,x ∈ G( F̄x ))x∈|X |
such that

−1
● For each place x ∈/ T , we have: τ = g0,x σ (g0,x ).
−1 si −1 si
● For each place xi ∈ T , we have: γ0 τ = g0,x i
σ (g0,xi ).

We fix a choice of these base points. We can see that:

−1
● For each place x ∈ / T , the element γx := g0,x γ0 g0,x is in G(Fx ).
−1
● For each place xi ∈ T , the element δxi := g0,x i
τ σ (g0,xi ) is in
! Fq Fq si ).
G(Fxi ⊗

We show that the automorphism group Jγ0 of (V , τ, γ0 ) is an interior form of

G γ0 defined over F whose local components are completely determined by the
elements (γx )x ∈T
/ and (δx i )x i ∈T defined above.
Finally, we obtain the integral expression of the number of fixed points:

Theorem 13.25. With the above notation, we get:

.
Lefλ ,β  (T, T  , s) J
I,≤ p
= 
T T
γ0 V (γ0 ) J Jγ0 (F)\ xi ∈T
ˆ Fq Fq si )×G(AT )
G(Fxi ⊗
 
φλxi (h −1
xi δxi σ (h x i )) φβx (h −1
x γx h x ) · 1( p((g0,x h x )x∈|X | ) ≤ p)) · dh
xi ∈T x ∈T
/

where

● the first sum runs through all the stable conjugacy classes γ0 of G(F) which
are (T, s)-admissible,
● the second sum runs through the set V (γ0 ) of the elements (V, τ, γ0 ) in
C(T, s),
 On a counting problem for G-shtukas 347

● for each place xi ∈ T , φλxi is the spherical function

C C
φλxi = φλ y ∈ Hxi ⊗Fq Fq si := Hy ,
y∈T ⊗Fq Fq si y∈T ⊗Fq Fq si
y|xi y|x i

● for each place x ∈ T  , φβx is the characteristic function of the double class
βx ∈ K I,x \G(Fx )/K I,x ,
● for each place x ∈/ T ∪ T  , φβx is the characteristic function of the unit
double class G(Ox )\G(Fx )/G(Ox ).

13.7.3 Example: the case of the general linear group G = GLn

In this section, we consider the case of the general linear group G = GLn . It
is well known that, for any field F, if two elements in GLn (F) are stably con-
jugate, then they are in fact conjugate. In other words, every stable conjugacy
class contains only one conjugacy class. This assertion allows us to obtain a
simpler expression of the number of fixed points for GLn -chtoucas.
As stated in the introduction, different counting problems for GLn -chtoucas
have been treated in the literature. When the modifications are elementary, it
was done by Drinfeld [4, 5] and Lafforgue [13]. For elliptic D-modules, it
was done by Laumon, Rapoport and Stuhler [21]. Contrary to our result, their
formulas contain only orbital integrals. The first formula containing twisted
orbital integrals was appeared in the work of Laumon [19, 20] using a strategy
similar to that of Kottwitz on certain Shimura varieties. Then, Ngô Bao Châu
[24] and Lau [17] have given similar expressions for the number of fixed points
for D-chtoucas. It is important to note that all these works are based on the
classification of φ-spaces of Drinfeld, which is different from our approach.
Let γ0 be an element of GLn (F) and let γ be a semisimple element of
GLn (F) which has the same characteristic polynomial as γ0 . The F-algebra
F[γ ] can be written in the form

F[γ ] = E 1 × E 2 × . . . × E k

where E i are the finite field extensions of F, with kj=1 [E j : F] = n. For each
index 1 ≤ j ≤ k, denote by γ j the projection of γ ∈ F[γ ] in E j . Let X E j
be the normalized curve over X whose function field is E j . Then, γ j induces
a divisor div(γ j ) over X E j and for each point xi ∈ T , denote by divxi (γ j ) the
part of div(γi ) supported by xi .

Proposition 13.26. The element γ0 is (T, s)-admissible if and only if it

satisfies the following conditions:
348 Ngo Dac Tuan

i) For each place xi ∈ T , γ0 is a si -norm in GLn (Fxi ).


ii) For each index 1 ≤ j ≤ k, we have the equality: xi ∈T deg(divxi (γ j )
/si ) = 0.

Proof. The proposition follows from the previous observation that, for each
place x, we have H 1 (Fx , GLn,γ ) = 1.

Proposition 13.27. i) Let (V, τ, γ0 ) be an element of C(T, s). Then, γ0 is

always (T, s)-admissible.
ii) Conversely, if γ0 is a (T, s)-admissible element of GLn (F), then there exists
a unique element (V , τ, γ0 ) in C(T, s). Moreover, τ lies in Z(GLn,γ0 )( F̄)
and the automorphism group of (V , τ, γ0 ) is exactly G γ0 .

Proof. It follows from the fact that H 1 (F, GLn,γ0 ) = 1.

Now, we choose the base points (g0,x )x∈|X| such that g0,x ∈ GLn,γ0 ( F̄x ). In
particular, for each place x ∈
/ T , we can take γx = γ0 .
The expression for the number of fixed points for GLn -chtoucas is given as
follows:

Theorem 13.28. We keep the previous notations. Then, we have:

.
Lefλ ,β  (T, T  , s) J =
I,≤ p

T T
γ0 J G γ0 (F )\ xi ∈T
ˆ Fq Fq si )×G(A T )
G(Fxi ⊗
 
φλxi (h −1
x i δxi σ (h xi )) × φβx (h −1
x γ h x ) × 1( p((g0,x h x ) x∈|X | ) ≤ p)) · dh x
xi ∈T x ∈T
/

where the sum runs through all the conjugacy class γ0 of GLn (F) which are
(T, s)-admissible.

References
[1] A. B EAUVILLE , Y. L ASZLO, Un lemme de descente, C. R. Acad. Sci. Paris
Sér. I Math., 320 (1995), 335-340.
[2] A. B EILINSON , V. D RINFELD, Quantization of Hitchin’s Integrable System
and Hecke Eigensheaves, preprint.
[3] S. B OSCH , W. L UTKEBOHMERT, M. R AYNAUD, Neron models, Ergeb. der
Math. 21. Springer Verlag 1990.
[4] V. D RINFELD, Varieties of modules of F-sheaves, Functional Analysis and its
Applications, 21 (1987), 107-122.
[5] V. D RINFELD, Cohomology of compactified moduli varieties of F-sheaves of
rank 2, J. Soviet Math., 46(2) (1989), 1789-1821.
 On a counting problem for G-shtukas 349

[6] M. G REENBERG, Schemata over local rings: II, Annals of Mathematics, 78(2)
(1963), 256-266.
[7] D. K AZHDAN , Y. VARSHAVSKY, article in preparation.
[8] R. KOTTWITZ, Rational conjugacy classes in reductive groups, Duke Math. J.,
49 (1982), no. 4, 785-806.
[9] R. KOTTWITZ, Isocrystals with additional structure, Compositio Math., 56(2)
(1985), 201-220.
[10] R. KOTTWITZ, Stable trace formula : elliptic singular terms, Math. Ann., 275
(1986) 365-399.
[11] R. KOTTWITZ, Shimura varieties and λ-adic representations, Automorphic
forms, Shimura varieties, and L-functions, Vol. I (Ann Arbor, MI, 1988), 161-
209, Perspect. Math., 10, Academic Press, Boston, MA, 1990.
[12] R. KOTTWITZ, Points on some Shimura varieties over finite fields, J. Amer.
Math. Soc., 5 (1992), 373-444.
[13] L. L AFFORGUE, Chtoucas de Drinfeld et conjecture de Ramanujan-Petersson,
Astérisque, 243 (1997), +329pp.
[14] L. L AFFORGUE, Une compactification des champs classifiant les chtoucas de
Drinfeld, J. Amer. Math. Soc., 11(4) (1998), 1001-1036.
[15] L. L AFFORGUE, Chtoucas de Drinfeld et correspondance de Langlands,
Invent. Math., 147 (2002), 1-241.
[16] V. L AFFORGUE , Chtoucas pour les groupes réductifs et paramétrisation de
Langlands globale, arXiv:1209.5352v4.
[17] E. L AU, On generalised D-shtukas, Dissertation, Bonner Mathematische
Schriften [Bonn Mathematical Publications], Universitat Bonn, Mathematis-
ches Institut, Bonn, 369 (2004), +110 pp.
[18] E. L AU, On degenerations of D-shtukas, Duke Math. J. 140(2) (2007), 351-
389.
[19] G. L AUMON, Cohomology of Drinfeld modular varieties. Part I. Geome-
try, counting of points and local harmonic analysis, Cambridge Studies in
Advanced Mathematics, Cambridge University Press, Cambridge, 41 (1996),
+344 pp.
[20] G. L AUMON, Cohomology of Drinfeld modular varieties. Part II. Automor-
phic forms, trace formulas and Langlands correspondence, with an appendix
by Jean-Loup Waldspurger, Cambridge Studies in Advanced Mathematics,
Cambridge University Press, Cambridge, 56 (1997), +366 pp.
[21] G. L AUMON , M. R APOPORT, U. S TUHLER, D-elliptic sheaves and the
Langlands correspondence, Invent. Math. 113(2) (1993), 217-338.
[22] B.-C. N GÔ, private communication.
[23] B.-C. N GÔ , D-chtoucas de Drinfeld à modifications symétriques et identité de
changement de base, long version, arXiv 2003.
[24] B.-C. N GÔ , D-chtoucas de Drinfeld à modifications symétriques et identité de
changement de base, Annales Scientifiques de l’ENS, 39 (2006), 197-243.
[25] B.-C. N GÔ, Fibration de Hitchin et endoscopie, Invent. Math., 164 (2006),
399-453.
[26] B.-C. N GÔ , T. N GÔ DAC , Comptage de G-chtoucas: la partie régulière
ellitique, Journal de l’Institut Mathématique de Jussieu, 7(1) (2008), 181-203.
350 Ngo Dac Tuan

[27] T. N GÔ DAC, Compactification des champs de chtoucas et théorie

géométrique des invariants, Astérisque, 313 (2007), +124pp.
[28] T. N GÔ DAC, Comptage de G-chtoucas: la partie ellitique, to appear, Compo-
sitio Math.
[29] M. R APOPORT, A guide to the reduction of Shimura varieties, Astérisque, 298
(2005), 271-318.
[30] M. R APOPORT, M. R ICHARTZ, On the classification and specialization of F-
isocrystals with additional structure, Compositio Math., 103 (1996), 153-181.
[31] J-P. S ERRE, Cohomologie galoisienne, 5e édition, Lecture Notes in Mathe-
matics, 5 (1994), +181pp.
[32] R. S TEINBERG, Regular elements of semisimple algebraic groups, Publica-
tions de l’IHES, 25 (1965), 49-80.
[33] Y. VARSHAVSKY, Moduli spaces of principal F-bundles, Selecta Math., 10
(2004), 131-166.
[34] Y. VARSHAVSKY, private communication.
 14
Modular forms and Calabi-Yau varieties
Kapil Paranjape†,1 and Dinakar Ramakrishnan‡,2
† Department of Mathematical Sciences, IISER, Mohali, India
E-mail address: [email protected]
‡ Department of Mathematicss, California Institute of Technology
Pasadena, CA 91125
E-mail address: [email protected]

Introduction

∞
Let f (z) = an q n be a holomorphic newform of weight k ≥ 2 relative
n=1
to 1 (N ) acting on the upper half plane H. Suppose the coefficients an are
all rational. When k = 2, a celebrated theorem of Shimura asserts that there
corresponds an elliptic curve E over Q such that for all primes p  N , a p = p+
1 − |E(F p )|. Equivalently, there is, for every prime , an -adic representation
ρ of the absolute Galois group GQ of Q, given by its action on the -adic Tate
module of E, such that a p is, for any p  N , the trace of the Frobenius Fr p at
p on ρ .
The primary aim of this article is to provide some positive evidence for the
expectation of Mazur and van Straten that for every k ≥ 2, any (normalized)
newform f of weight k and level N should, if it has rational coefficients, have
an associated Calabi-Yau variety X/Q of dimension k − 1 such that

(Ai) The {(k − 1, 0), (0, k − 1)}-piece of H k−1 (X ) splits off as a submotive
M f over Q,
(Aii) a p = tr(Fr p | M f, ), for almost all p, and
(Aiii) det(M f, ) = χk−1 ,

where χ is the -adic cyclotomic character.

In fact we expect (Aiii) to hold for every p not dividing N .

1 Partly supported by the the JCBose Fellowship of the DST (SR/S2/JCB-16/2010) and IMSc,
Chennai
2 Partly supported by the NSF grants DMS-0701089 and DMS-1001916

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

351
352 Kapil Paranjape and Dinakar Ramakrishnan

Furthermore, we even hope that in addition the following holds:

(Aiv) X admits an involution τ which acts by −1 on H 0 (X, k−1 ).
We anticipate that the involution τ can be chosen in such a way to make the
quotient X/τ a rational variety.
This typically holds in our examples below. This extra structure is natural to
want and is needed for understanding a variety of operations like products and
twists. Such a τ obviously exists for k = 2, in which case X is (by Shimura)
an elliptic curve over Q, given by an equation y 2 = f (x), and the involution
τ sends (x, y) to (−x, y), thus acting by −1 on the holomorphic differential
ω = d x/y which spans H 0 (X, 1 ). (Of course X/τ is in this case P1 .)
To fix ideas, one could think of a motive as a semisimple motive M relative
to absolute Hodge cycles ([DMOS]), with avatars (M B , MdR , M ) of Betti, de
Rham and -adic realizations. Since every Calabi-Yau manifold of dimension
1 is an elliptic curve, (Ai) through (Aiv) provide a natural extension of what
one has for k = 2. Of course for k > 2, one knows by Deligne ([Del]) that
there is an irreducible, two-dimensional -adic representation ρ of Gal(Q/Q)
so that (Aii ), (Aiii) hold with M f, replaced by ρ .
In the first part of this article we will focus on the forms f of even weight
for small levels, before moving on in the second part to formulate an analogue
for regular self-dual cusp forms on GL(n)/Q and see how the framework is
compatible with the principle of functoriality.
When the weight is odd, the Q-rationality forces f to be of CM type, and
for weight 3, we refer to the paper of Elkies and Schuett ([ES]) for a beautiful
result.
For non-CM newforms f of weight k > 2 with Q-coefficients, we in fact
hope for more, namely that the cohomology ring of X will be spanned by M
and the Hodge/Tate classes of various degrees; in particular, X should be rigid
in this case.
It is a difficult problem in dimensions > 3 to find smooth models of varieties
with trivial canonical bundles, and for this reason we formulate our ques-
tions for such varieties with mild singularities. By a Calabi-Yau variety over
a field k, we will mean an n-dimensional projective variety X/k on which the
canonical bundle K X is defined such that
(CY1) K X is trivial; and
(CY2) H m (X, O X ) = 0 for all (strictly) positive m < n.
More precisely, we will want such an X to be normal and Cohen-Macaulay, so
that the dualizing sheaf K X is defined, with the singular locus in codimension
at least 2, so that K X defines a Weil divisor; finally, X should be Gorenstein,
 Modular forms and Calabi-Yau varieties 353

so that K X will represent a Cartier

? divisor.
@ Ideally we would like to singular
locus X sing to be of dimension ≤ 2 .
n−1

In addition to these properties, we would ideally also like X to be realized

as a double cover π : X → Y , where Y is a projective smooth rational vari-
ety with negative ample canonical divisor. (Such an X will be automatically
Gorenstein so that K X is defined, and π∗ (O X ) = OY ⊕ L, for a line bundle L
on Y with L 2 giving the branch locus; moreover, π∗ (K X ) = KY ⊕ KY ⊗ L −1 ,
forcing L = KY .)
Here is our first result:

Theorem 14.1. Fix = 1 (N ), N ≤ 5. Let k be the first even weight s.t.

dim(Sk ( )) = 1. Then ∃ a Calabi-Yau variety V ( f )/Q with an involution τ
associated to the new generator f of Sk ( ) satisfying (Ai ) through (Ai v). In
fact, when N ≤ 5, V = V ( f ) is birational over Q to the Kuga-Sato variety
(k−2)
Ẽ N .
∞
In particular, this applies to the Delta function (z) = n=1 τ (n)q =
n

m≥1 (1 − q ) , for N = 1, k = 12. By the Kuga-Sato variety, we mean a
m 24

smooth compactification (over Q) of the fibre product E k−2 N of the universal

elliptic curve E N over the model over Q of the modular curve 1 (N )\H.
It is well known that Sk ( 1 (N )) is one-dimensional when (N, k) equals
(1, 12), (1, 16), (2, 8), (2, 10), (3, 6), (3, 8), (4, 6), (5, 4), (5, 6), (6, 4), (7, 4).
(The values in bold are the cases to which the Theorem applies.) For example,
for the case (2, 8), the generator is
f (z) = q − 8q 2 + 12q 3 − 210q 4 + 1016q 5 + . . .

Recall that a newform f (z) = ∞ n
n=1 an q is of CM-type iff there is an odd,
quadratic Dirichlet character δ such that a p = a p δ( p) for almost all primes p.
Equivalently, if K is the imaginary quadratic field cut out by δ, a p = 0 for all
p which are inert in K .
As a first step, for general N , we may ask for a potential statement, i.e. the
association, over a finite extension F of Q, of a Calabi-Yau variety V /F to a
newform f with Q-coefficients. Here we state a modest result in this direction,
already known in different ways, just to show that it fits into our framework:

Proposition 14.2. Let f be a newform of weight k ≥ 3 of CM type with

rational coefficients. Then ∃ a Calabi-Yau (k −1)-fold X defined over a number
field F such that (Ai), (Aii) and (Aiii) hold over F. This X arises as a Kummer
variety associated to an elliptic curve E with complex multiplication. When
k ≤ 4, X can be taken to be a smooth model.
354 Kapil Paranjape and Dinakar Ramakrishnan

For k = 4, all but one form f have been treated in Cynk-Schuett ([CS]).
Again, when k = 3, there is a much more precise and satisfactory result over
Q in the work of Elkies and Schuett [ES].
In the converse direction, if M is a simple motive over Q of rank 2, with
coefficients in Q, of Hodge type {(w, 0), (0, w)} with w > 0, then the gen-
eral philosophy of Langlands, and also a conjecture of Serre, predicts that M
should be modular and be associated to a newform f of weight w + 1 with
rational coefficients. This is part of a very general phenomenon, and applies
to motives occurring in the cohomology of smooth projective varieties over
Q. In any case, it applies in particular to Calabi-Yau threefolds over Q whose
{(3, 0), (0, 3)}-part splits off as a submotive. In this context, there have been a
number of beautiful results, some of which have been described in the mono-
graphs [YL] and [YYL]. See also [Mey], [GKY] and [GY]. They are entirely
consistent with what we are trying to do in the opposite direction, and also
provide supporting examples. It should perhaps be noted that these results
(relating to the modularity of rigid Calabi-Yau threefolds) can now be deduced
en masse from the proof of Serre’s conjecture due to Khare and Wintenberger
([KW]), with a key input from Kisin ([Kis]). See also the article of Dieulefait
([Die]).
Let us now move to a more general situation. Fix any positive integer n and
suppose that f is a (new) Hecke eigen-cuspform on the symmetric space
Dn := SL(n, R)/SO(n),
relative to a congruence subgroup of SL(n, Z), which is algebraic and reg-
ular. For n = 2, f is algebraic and regular iff it is holomorphic of weight
≥ 2. In general, one considers the cuspidal automorphic representation π of
GL(n, A) which is generated by f , and by Langlands the archimedean compo-
nent π∞ corresponds to an n-dimensional representation σ∞ of the real Weil
group WR , which contains C∗ as a subgroup of index 2. One says ([Clo1]) that
π is algebraic if the restriction of σ∞ to C∗ is a sum of characters χ j of the
form z → z p j z q j , with p j , q j ∈ Z, and it is regular iff χi = χ j when i = j.
Such an f contributes to the cuspidal cohomology Hcusp ∗ ( \D , V ) relative
n
to a local coefficient system V in a specific degree w = w( f ). Moreover, f
is rational over a number field Q( f ), defined by the Hecke action on coho-
mology, which preserves the cuspidal part (loc. cit.). There is conjecturally
a motive M( f ) over Q of rank n, with coefficients in Q( f ), and weight w.
By Clozel [Clo2], [CHL], M ( f ) exists for suitable f which are in addition
essentially self-dual.
Now let f be an algebraic, regular, essentially self-dual newform of weight

w relative to ⊂ SL(n, Z), with L-function L(s, f ) = p L p (s, f ), such
 Modular forms and Calabi-Yau varieties 355

that Q( f ) = Q. Then our question is if there exists a Calabi-Yau variety X/Q

of dimension w with an involution τ such that

(Ci) There is a submotive M( f ) of H w (X ) of rank n such that M( f )(w,0) =

H w,0 (X ),
(Cii) L p (s, f ) = L p (s, M ( f )) for almost all p, and
(Ciii) the quotient of X bt τ is a rational variety.

where L p (s, M ( f )) equals, at any prime p =  where the -adic realization

M ( f ) is unramified, det(I − Fr p p −s | M ( f ))−1 . Again we would like to be
able to find an X having good reduction outside the primes dividing t N , where
n is the level of f and t the order of torsion in , such that (Cii) holds for any
such p = .
Thanks to the principle of functionality, one should be able to obtain a cer-
tain class of Q-rational, regular, algebraic, essentially self-dual newforms f
by transferring forms on (the symmetric domains of) smaller reductive Q-
subgroups G of GL(n). The simplest instance of this phenomenon is given
by the symmetric powers symm (g) of classical Q-rational, non-CM newforms
g of weight k. One knows by Kim and Shahidi ([KS], [Kim]) that for m ≤ 4,
f = symm (g) is a cusp form on GL(m + 1). (Recently, this has been extended
to m = 5 (and further) in the works of Clozel and Thorne, and of Dieulefait.)
Here is our third result:

Theorem 14.3. Let g be a non-CM, elliptic modular newform of weight 2,

level N and trivial character, whose coefficients an lie in Q. Then for for any
m > 0, there is a Calabi-Yau variety X m with an involution τ over Q of dimen-
sion m associated to (g, symm ) such that (Ci), (Cii), (Ciii) hold relative to
M = symm (ρ (g)). Moreover, for m ≤ 3, X m can be taken to be non-singular,
with good reduction outside N .

Here X 2 is just the familiar Kummer surface attached to E × E, where

E = X 1 is the elliptic curve/Q defined by g. But the case m = 3 is
interesting, especially since it is not rigid, thanks to the Hodge type being
{(3, 0), (2, 1), (1, 2), (0, 3)}, with each Hodge piece being one-dimensional.
In fact, in that case, sym3 (g) corresponds (by [RS]) to a holomorphic Siegel
modular cusp form F of genus 2 and (Siegel) weight 3. Such an F contributes
to the cohomology in degree 3 of the Siegel modular threefold V of level N 3 .
Since the geometric genus of V is typically > 1, it cannot be Calabi-Yau.
However, there should be, as predicted by the Hodge and Tate conjectures, an
algebraic correspondence between X 3 and V (for any N ).
356 Kapil Paranjape and Dinakar Ramakrishnan

One also knows (cf. [Ram]) that given two non-CM newforms g, h of
weights k, r ≥ 2 respectively, then there is an algebraic automorphic form
f = g  h on GL(4)/Q, which will be cuspidal and regular if k = r . If g, h
are Q-rational, then so is f . Moreover, f is essentially self-dual because g and
h are.

Theorem 14.4. Let g, h be Q-rational, non-CM newforms as above of respec-

tive weights k, r > 1, with k = r . Suppose we have Calabi-Yau varieties with
involutions (X (g), τg ), (X (h), τh ) over Q attached to g, h respectively, satis-
fying (Ai) through A(iv). Put f = g  h, so that w( f ) = (k − 1)(r − 1). Then
there is a Calabi-Yau variety with involution (X ( f ), τ f ) over Q of dimension
w( f ) such that (Ci) through (Ciii) hold.

The point is that the product Z := X (g) × X (h) has the desired submotive
in degree w, but it also has global holomorphic m-forms for m = k − 1 and
m = r −1. We exhibit an involution τ on Z such that when we take the quotient
by τ , these forms get killed and we get a Calabi-Yau variety with reasonable
singularities. To get unconditional examples of this Theorem, take k = 2 and
choose h to be one of the examples of Theorem A of weight r > 2. To be
specific, we may take g to be the newform of weight 2 and level 11, and h to
be the newform of weight 4 and level 5, in which case f = g  h has level
552 , and X ( f ) is a Calabi-Yau fourfold. A similar inductive construction of
Calabi-Yau varieties is also found in a paper of Cynk and Hulek ([CH]).
In sum, it is a natural question, given Shimura’s work on forms of weight
2, if there are Calabi-Yau varieties with an involution associated to forms of
higher weight with rational coefficients, and a preliminary version of this circle
of questions was raised by the second author in a talk at the Borel memo-
rial conference at Zhejiang University in Hangzhou, China, in 2004, and quite
appropriately, Dick Gross, who was in the audience, cautioned against hoping
for too much without sufficient evidence. Over the past years, there has been
some positive evidence, though small, and even if there is no V in general,
especially for non-CM forms f of even weight, the examples where one has
nice Calabi-Yau varieties V enriches them considerably, and one of our aims
is to understand the Hecke eigenvalues a p in such cases a bit better in terms of
counting points of V mod p. Since then we have learnt from [ES] (and Noriko
Yui) that the question of existence of a Calabi-Yau variety V associated to f
were earlier raised by Mazur and van Stratten. What we truly hope for is that in
addition, V will be equipped with an involution τ acting by −1 on the unique
global holomorphic form of maximal degree, so one can form products, etc.,
and also deal with quadratic twists. We have some interesting examples in the
 Modular forms and Calabi-Yau varieties 357

three-dimensional case (where k = 4), and a sequel to this paper will also
contain a discussion of these matters, as well as a way to get nicer models in
certain higher dimensional examples.
The Ramanujan coefficients τ ( p) of the Delta function have been a source of
much research. Our own work was originally motivated by the desire to express
them in terms of the zeta function of a Calabi-Yau variety, though our path has
diverged somewhat. In a different direction, a very interesting monograph of
Edixhoven, et al. ([Edi]) yields a deterministic algorithm for computing τ ( p)
with expected running time which is polynomial in log p. They do this by
relating the associated Galois representation mod  to the geometry of certain
effective divisors on the modular curve X 1 (5).
We thank all the people who have shown interest in this work, and a
special thanks must go to Matthias Schuett who made a number of use-
ful comments on the earlier version (2008) as well as a recent version (of
two months ago). One of us (K.P.) would like to thank the JCBose Fellow-
ship of the DST (SR/S2/JCB-16/2010) as well as Caltech and the IMSc,
Chennai, where some of the work was done, and the second author (D.R.)
would like to thank the NSF for continued support through the grants DMS-
0701089 and DMS-1001916, and also IMSc, Chennai, for having him visit
at various times. In addition, the second author (D.R.) would like to thank
the Hausdorff Institute for their hospitality during his visit there for a week
in February to attend and speak at the Session organized by Dieulefait and
Wintenberger.

14.1 An intuitive picture of the geometric construction

In this section we will give an idea behind our construction of varieties with
trivial canonical bundles arising as birational models of elliptic modular vari-
eties V with non-positive canonical bundles. (S.T. Yau has informed us that he
earlier had a similar construction, albeit in a different context.) We will use the
intuitive language of divisors and linear systems. The content here will not be
used in the succeeding sections, where we will use sheaves and do everything
precisely in the different cases at hand.
Recall that V arise as fibre products of the universal family of elliptic curves
E with additional structures over the modular curve associated to a congru-
ence subgroup of SL(2, Z). When is 1 (N ) (resp. 0 (N )), the additional
structure is a point (resp. subgroup) of order N . We have a slight preference
for 1 (N ) over 0 (N ), because in the latter case, one gets, due to the existence
of −I , only a coarse moduli space.
358 Kapil Paranjape and Dinakar Ramakrishnan

Suppose we want to parametrize triples of the form (E, S; R) where E

is a curve of genus 1, S a finite set of points on E, and R a finite set of
linear equivalence relations on the points S. We will think of E as a curve
of genus 1 without a specified origin (by making an identification with line
bundles of degree 1 on it), hence with no group structure, and the relations
in R are taken to hold in the divisor class group. (Once we pick a point o of
E, we can of course identify this class group with E × Z by sending (e, n)
to ((e − o) + n.o).) We will take at least a portion of S to consist of general
points, and we assume that R contains all the relations between the chosen
points. Additional relations may hold for special triples (E,S,R) but generically,
only those in R will hold.
Suppose also that there is a surface X , and a linear system P of divisors
linear equivalent to −K X , where K X is the canonical divisor of X . Assume
that for a “general” datum (E, S; R) we have a uniquely determined element of
P, so that the projective space P parametrizes triples as above up to birational
isomorphism. Here by “general datum” we mean a point in an open subset of
the parameter space (or moduli stack to be precise).
Let W ⊂ (X, O(−K X )) be the subspace so that P is the associated
projective space and n = rankW . We have a natural homomorphism
φ : V ⊗ O X n → ⊕in=1 pri∗ O(−K X ),
where pri denotes the projection onto the i -th factor. The divisor D where
the determinant det(φ) vanishes parametrizes, birationally, tuples of the form
(E, S; R; p1 , . . . , pn ) where the pi are n additional points which are not
subject to any additional relation. Moreover, D has trivial canonical bundle.
If n is at most the dimension of the linear system, then the parameter space
is rational. But new things happen when n is larger than that, when one gets
a divisor on a product of rational surfaces, in fact given by the vanishing of
det(φ). This moduli space Vn , say, fibers over the rational variety Vn−1 , with
the general fibre being an elliptic curve. The involution x → −x on the general
fibre gives rise to an involution τ on Vn . So Vn /τ fibers over a rational variety
with P1 fibres. Hence Vn /τ is unirational.

14.2 The modular varieties of interest

In the context of the congruence subgroup 1 (N ) of SL(2, Z), the elliptic mod-
ular varieties V with non-positive canonical bundles are associated with pairs
(N , k) such that there is at most one modular form of level N and weight k.
The complete list of such pairs is given in the following table:
 Modular forms and Calabi-Yau varieties 359

N k
1 ≤ 23, 25, 26, odd
2 ≤ 11, odd
3 ≤8
4 ≤6
5, 6 ≤4
7, 8 ≤3
9, 10, 11, 12, 14, 15 2

One can similarly make the corresponding tables for the groups 0 (N ) and
0 (N
2) ∩
1 (N ).

14.3 The Calabi-Yau 11-fold associated to 

 is a generator of S12 (SL(2, Z)). The object is to show that the Kuga-Sato
variety E (10) is birational to an eleven-dimensional Calabi-Yau variety V . We
will use ≡ to denote birational equivalence. It is easy to see that for any r ≥ 0,
E (r ) ≡ M1 (r + 1),
where Mg (k) is the moduli space of genus g curves with k marked points, with
compactification Mg (k).
So we need to find a birational model V of M1 (11) such that V is
Calabi-Yau. Let S be the surface obtained by blowing up 4 general points
P1 , P2 , P3 , P4 in P2 . Let E be an elliptic curve with n + 5 general points
Q 0 , Q 1 , . . . , Q n+4 , and use |3Q 0 | to define a morphism E → P2 .
Using an automorphism of P2 we may assume: Q i = Pi for 1 ≤ i ≤ 4.
The embedding E → P2 lifts to a morphism ϕ : E → S, and the adjunction
formula gives ϕ(E) ∈ |K−1 S |.
We get a rational map M1 (n + 5) → S n ,
(E, {Q 0 , . . . , Q n+4 }) → (P5 , . . . , Pn+4 ).
W := (S, K−1
S ) has dimension 6, and there exists a hom (of sheaves on S n ):
f n : W ⊗ O S n → K−1 −1
S  · · ·  KS .

Ker( f n ) is the vector space of sections in W vanishing at = P5 , . . . , Pn+4 .

The associated projective space then identifies with the collection of all
(general) points in M1 (11) giving rise to this point on S n .
Put
Vn := Proj S n (coker(t f n ))
360 Kapil Paranjape and Dinakar Ramakrishnan

where
t
fn : K S  · · ·  K S → W ∨ ⊗ O∨
S.

Then Vn is birational to M1 (n +5). We have corank(t f n ) = 6 −n at a general

point of S n . And there is a natural map
π : Vn → S n
n ≤ 5 π is surjective with fibres P5−n . Hence Vn , which is ≡ M1 (n + 5), is a
rational variety in this case. Note that V5 is just S 5 .
n = 6 V = V6 is a (reduced) divisor in S 6 , hence Gorenstein. It is defined by
the vanishing of det( f n ), which is a section of K−1S n . So KV is trivial.
We already know that h (11,0) = 1 and h ( p,0) = 0 for 0 < p < 11
for Ẽ 10 . These also hold for V . So V is Calabi-Yau. Also, the whole
construction is rationally defined.
V = V6 fibers over V5 with fibres of dimension 1, corresponding to
cubics passing through 10 points. The natural involution on the general
fibre, which is an elliptic curve, gives rise to one, call it τ , on V . The
quotient V /τ is unirational because it is a family of rational curves
on a smooth rational surface. Clearly, τ must act by −1 on the one
dimensional space H 0 (V , 11 ).

14.4 A C-Y 7-fold occurring in level 2

Let E be an elliptic curve with origin o ∈ E, x ∈ E a point of order 2
and y, z ∈ E some other (general) points. Under the morphism a : E →
|2[o] + [y]|, the divisors 2[o] + [y] and 2[x] + [y] are linear sections. Of the
four points o, x, y and z we may assume (under the hypothesis of generality on
y and z) that no three become collinear under the morphism a. Thus, we can
identify these with the points (0 : 0 : 1), (1 : 0 : 0), (0 : 1 : 0) and (1 : 1 : 1)
respectively in order to identify |2[o] + [y]| with P2 .
Conversely, let E be a cubic curve in P2 which has the following properties:
(1) E passes through the points (0 : 0 : 1), (0 : 1 : 0), (1 : 0 : 0) and
(1 : 1 : 1).
(2) The line Y = 0 is tangent to E at the point (0 : 0 : 1).
(3) The line Z = 0 is tangent to E at the point (0 : 1 : 0).
Then E is a curve of genus 1 for which we take o = (0 : 0 : 1) as the origin of
a group law. Let x = (0 : 1 : 0), y = (1 : 0 : 0) and z = (1 : 1 : 1). Then we
obtain the relation
2[o] + [y] 2[x] + [y]
 Modular forms and Calabi-Yau varieties 361

It follows that 2x = o. Thus we have obtained (E, o, x, y, z) of the type we

started with.
Direct calculation shows that the linear system of cubics in P2 that satisfy the
conditions above is the linear span of X 2 Y − X Y Z, X 2 Z − X Y Z , Y 2 Z − X Y Z
and Y Z 2 − X Y Z.
Here is an alternate construction in level 2:
Let E be an elliptic curve with origin o ∈ E, x ∈ E a point of order 2 and
y, z ∈ E some other points. The morphism a : E → |2[o]| has fibres 2[o],
[y] + [−y] and [z] + [−z] which we map to 0, 1 and ∞ respectively in order to
identify |2[o]| with P1 . The morphism b : E → |[o] + [x]| has fibres [o] + [x],
[y] + [x − y] and [z] + [x − z] which we map to 0, 1 and ∞ in order to identify
|[o] + [x]| with P1 . Thus we obtain a morphism a × b : E → P1 × P1 which
is constructed canonically from the data (E, o, x, y, z).
Conversely let E be a curve of type (2, 2) in P1 ×P1 which has the following
properties:

(1) E passes through the points (0, 0), (1, 1) and (∞, ∞).
(2) The line {0} × P1 is tangent to E at the point (0, 0).
(3) If (u, 0) is the residual point of intersection of E with P1 × {0}, then {u} ×
P1 is tangent to E at this point.

Then E is a curve of genus 1 for which we take o = (0, 0) as the origin in a

group law. Let x = (u, 0), y = (1, 1) and z = (∞, ∞). We obtain the identity
2[o] 2[x], from which it follows that 2x = o. Thus we have recovered the
data (E, o, x, y, z).

14.5 A remark on elliptic curves with level 3 structure

Let E be an elliptic curve with origin o ∈ E, x ∈ E a point of order 3 and
y ∈ E some other point. The morphism a : E → |2[o]| has fibres 2[o],
[x] + [2x] and [y] + [−y] which we map to 0, 1 and ∞ respectively in order to
identify |2[o]| with P1 . The morphism b : E → |[o] + [x]| has fibres [o] + [x],
2[2x] and [y] + [x − y] which we map to 0, 1 and ∞ respectively in order to
identify |[o] + [x]| with P1 . Thus we obtain a morphism a × b : E → P1 × P1
which is constructed canonically from the data (E, o, x, y).
Conversely, let E be a curve of type (2, 2) in P1 ×P1 which has the following
properties:

(1) E is tangent to the line {0} × P1 at the point (0, 0).

(2) E is tangent to the line P1 × {1} at the point (1, 1).
(3) E passes through the points (1, 0) and ∞, ∞).
362 Kapil Paranjape and Dinakar Ramakrishnan

Then E is a curve of genus 1 and we use o = (0, 0) as the origin of a group law
on E. Let x = (1, 0), y = (∞, ∞) and p = (1, 1). We obtain the identities,
2[o] [x] + [ p] [o] + [x] 2[ p]
It follows that p = 2x and 3x = o. We have thus recovered the data
(E, o, x, y) that we started with.

14.6 Level 3 and a CY 5-fold

We will construct a 5-fold with trivial canonical bundle and singularities only
in dimension 2 or less such that its middle cohomology represents the motive
of the (unique) modular form of level 3 and weight 6.
Consider the linear system P of cubics in P2 that is spanned by the curves
X Y − X Y Z , Y 2 Z − X Y Z and Z 2 X − X Y Z; this system is stable under the
2

cyclic automorphism X → Y → Z → X of P2 . Each curve in the linear

system P is tangent to the line Z = 0 at the point pY = (0 : 1 : 0); similarly,
the curve is tangent to X = 0 at the point p Z = (0 : 0 : 1) and to Y = 0
at the point p X = (1 : 0 : 0). Moreover, each curve passes through the point
p0 = (1 : 1 : 1). We note that the linear system P is precisely the collection of
cubic curves in P2 that satisfy these conditions.
In the divisor class group of a smooth curve in this linear system we obtain
the identities
2 pY + p X = 2 p Z + pY = 2 p X + p Z = p X + p0 + r
where r denotes the remaining point of intersection of the curve with the line
Y = Z that joins p X and p0 . In particular, we note that pY − p X is of order 3
in this class group and p Z − p X = 2( pY − p X ).
Conversely, suppose we are given a smooth curve E of genus 1 and a line
bundle ξ of order 3 on E; moreover, suppose that three distinct points p, q and
r are marked on E. We then obtain two additional points a and b on E such
that a − p = ξ and b − p = 2ξ in the divisor class group of E. Consider the
morphism E → P2 that is given by the linear system of the divisor p + q + r .
Moreover, we choose co-ordinates on P2 so that the point p goes to p X , q goes
to p0 , a goes to pY and b goes to p Z . This gives an embedding of E as a curve
in P2 that belongs to the linear system P.
Let S denote the surface obtained by blowing up P2 at the four points
p X , pY , p Z and p0 , and then further blowing up the resulting surface at the
“infinitely near points” that correspond to Z = 0 at pY , to X = 0 at p Z and
 Modular forms and Calabi-Yau varieties 363

to Y = 0 at p X . Let H denote the inverse image in S of a general line in P2 ;

let E X , E Y , E Z and E 0 denote the strict transforms of the exceptional loci
of the first blow-up over the points p X , pY , p Z and p0 respectively; let FX ,
FY and FZ denote the exceptional divisors of the second blow-up. The anti-
canonical divisor −K S = 3H − E X − E Y − E Z − E 0 − 2(FX − FY − FZ )
has a base-point free complete linear system | − K S | which can be iden-
tified with P. Let T denote the natural incidence locus in S × P. The
variety
X = T ×P T ×P T

is a singular 5-fold which is Gorenstein and has trivial canonical bun-

dle. Moreover, an open subset of X 0 parametrizes tuples of the form
(E, ξ, p, q, r, s, t, u) where E is a curve of genus 1, ξ is a line bundle of order
3 on E and p, q, r , s, t and u are six distinct points on E.
Let L X , L Y , L Z denote the strict transforms in S of the lines in P2 defined
by X = 0, Y = 0, Z = 0 respectively. Further, let R be the strict transform in
S of the curve in P2 defined by

X 2 Z + Y 2 X + Z 2 Y − 3X Y Z = 0

This is the unique cubic in P2 that has a node at p0 and is tangent to X = 0 at

pY , to Y = 0 at p Z and to Z = 0 at p X . It follows that R is a smooth rational
curve that meets E 0 in a pair of distinct points and the triple (R, L X , E Y )
(respectively (R, L Y , E Z ) and (R, L Z , E X )) consists of smooth curves that
meet pairwise transversally.
The morphism S → P ∗ induced by the linear system P can be factorized
via a double cover S → W which is ramified along R. Each of the curves L X
and E Y (respectively L Y and E Z ; L Z and E X ) is mapped isomorphically onto
the same smooth irreducible curve G Z (respectively G X ; G Y ) in W ; the curve
R is mapped isomorphically onto the branch locus Q in W . The morphism
W → P ∗ collapses the curves G X (respectively G Y and G Z ) to a point q X
(respectively qY and q Z ) in P ∗ ; in fact W → P ∗ is identified with the blow-up
of P ∗ at these points. Moreover, Q is mapped to a plane quartic Q which has
cusps at these three points.
Let T denote the incidence locus in P × S as above. It is the pull-back via
S → P ∗ of the natural incidence locus I ⊂ P × P ∗ . The latter can be identified
(via the projection I → P ∗ ) with the projective bundle of one-dimensional
linear subspaces of the tangent bundle of P ∗ . Hence, the exceptional curve
G X (respectively G Y and G Z ) of the blow-up W → P ∗ can be identified with
the fibre I X of I → P ∗ over the point q X (respectively qY and q Z ). Thus we
364 Kapil Paranjape and Dinakar Ramakrishnan

obtain natural maps E α → T and L α → T that are sections of the P1 -bundle

T → S over the curves E α and L α respectively; let E˜α and L˜α denote the
images. Let TX (respectively TY ; TZ ) denote the fibre of T over the point of
intersection of L Y and E Z (respectively L Z and E X ; L X and E Y ).
The tangent direction along Q gives a rational morphism (defined outside
the cusps) from Q to I . It follows that this extends to a section R → T of
T → S over R and gives a curve R̃ in T . The quadruple of curves R̃, TX , L˜Y ,
E˜Z (respectively, R̃, TY , L˜Z , E˜X ; R̃, TZ , L˜X , E˜Y ) meet pairwise transversally
in a single point r X (respectively rY ; r Z ) in T . The curve R̃ in T is mapped to
a nodal cubic R in P for which I X , IY and I Z are inflectional tangents. The
curves TX , L˜Y , E˜Z (respectively TY , L˜Z , E˜X ; TZ , L˜X , E˜Y ) in T lie over I X
(respectively IY ; I Z ) in P.
The singular locus of the morphism T → P consists of the curves R̃, Tα ,
L˜α and E˜α for α = X, Y, Z as described above.
The singular fibres of T → P then have the following description:

(1) If a is a smooth point of R which is not a point of inflection then the fibre
Ca is a rational curve in S with a single ordinary node.
(2) If b which is on an inflectional tangent (i. e. one one of the lines I X , IY ,
I Z ) of R but is not a point of inflection of R then the fibre Cb is a curve
with three components and three nodes (i. e. a “triangle” of P1 ’s).
(3) If c is a point of inflection of the curve R, then Cc consists of three P1 ’s that
pass through a point and (since Cc lies on a smooth surface S) is locally a
complete intersection.
(4) If d is the node of R then the fibre Cd consists of a pair of smooth P1 ’s
in S that meet in a pair of points. In fact the curves are E 0 and the strict
transform in S of the curve in P2 defined by the equation

X 2 Y + Y 2 Z + Z 2 X − 3X Y Z = 0

In particular, the elliptic fibration T → P is semi-stable but for the three fibres
over the points of inflection of R.
Now consider the variety X = T × P T × P T . The singular points of X 0
consist of triples (x, y, z) of points of T , where at least two of these points are
critical points for the morphism T → P. In particular, these points lie over
the union of R and I X , IY and I Z . Since the singular points of each of the
fibres described above are isolated, it follows that the singular locus of X has
components of dimension at most 2.
 Modular forms and Calabi-Yau varieties 365

14.7 Level 4 and a C-Y 5-fold

Let E be an elliptic curve with origin o ∈ E, x ∈ E a point of order 4. Under
the morphism a : E → |3[o]|, the divisors 3[o], 2[2x] + [o], [2x] + 2[x] and
[o] + [x] + [3x] are linear sections of the image curve. Let p denote the point
of intersection of the lines corresponding to 3[o] and [2x] + 2[x]. No three of
the points o, 2x, 3x and p are collinear. Thus we can identify |3[o]| with P2 in
such a way that o is identified with (0 : 1 : 0), 3x is identified with (0 : 0 : 1),
p is identified with (1 : 0 : 0) and 3x is identified with (1 : 1 : 1). Thus we
obtain a morphism a : E → P2 which is constructed canonically from the data
(E, o, x).
Conversely, let E be a cubic curve in P2 which has the following properties:
(1) E passes through the points (0 : 1 : 0), (0 : 0 : 1), (1 : 1 : 1) and
(1 : 0 : 1).
(2) The line Z = 0 is an inflectional tangent to E (at the point (0 : 1 : 0)).
(3) The line X = 0 is tangential to the curve E at the point (0 : 0 : 1).
(4) the line Y = 0 is tangential to the curve E at the point (1 : 0 : 1).
Then E is a curve of genus 1 and we use o = (0 : 1 : 0) as the origin of a
group law on E. Let x = (1 : 0 : 1), p = (0 : 0 : 1) and q = (1 : 1 : 1). We
obtain the identities
3[0] [o] + 2[ p] [ p] + 2[x] [o] + [q] + [x]
It follows that 2 p = o, 2x = p and q = −x = 3x. Thus we have recovered
the data (E, o, x) that we started with.
Direct calculation shows us that the linear system of cubic curves in P2 that
satisfy the above conditions is the linear span of Y Z (Y − Z) and X (X − Z )2 .
Here is an alternate construction in level 4:
Let E be an elliptic curve with origin o ∈ E, x ∈ E a point of order 4
and y ∈ E some other point. The morphism a : E → |2[o]| has fibres 2[o],
2[2x] and [y] + [−y] which we map to 0, 1 and ∞ respectively in order to
identify |2[o]| with P1 . The morphism b : E → |[o] + [x]| has fibres [o] + [x],
[2x]+[3x] and [y]+[x−y] which we map to 0, 1 and ∞ respectively in order to
identify |[o] + [x]| with P1 . Thus we obtain a morphism a × b : E → P1 × P1
which is constructed canonically from the data (E, o, x, y).
Conversely, let E be a curve of type (2, 2) in P1 ×P1 which has the following
properties:
(1) E is tangent to the line {0} × P1 at the point (0, 0).
(2) E is tangent to the line {1} × P1 at the point (1, 1).
366 Kapil Paranjape and Dinakar Ramakrishnan

(3) If E meets P1 × {0} at (0, 0) and (u, 0) and E meets P1 × {1} at (1, 1) and
(v, 1); then u = v.
(4) E passes through the point (∞, ∞).
Then E is a curve of genus 0 and we use o = (0, 0) as the origin of a group
law on E. Let x = (u, 0), p = (v, 1) and q = (1, 1). We obtain the identities,
2[o] 2[q] [o] + [x] [ p] + [q]
2[0] [x] + [ p] (from condition 3 above)
It follows that q = 2x, p = 3x and 4x = o. Let y = (∞, in f t y). We
have thus recovered the data (E, o, x, y) that we started with. Note that in this
construction, E is in S = P1 × P1 , and the Calabi-Yau variety V is a divisor
 8
on S 4 = P1 .

14.8 Forms of weight 4 and Calabi-Yau 3-folds

The construction of this section is adapted from that of Schoen’s article [Sch1],
where he has associated a modular form f of weight 4 in this case to a special
quintic threefold X . However, we need to modify his construction to obtain an
involution τ such that the quotient X/τ is a rational threefold whose homology
is the same as its Chow group. (Hence, the rational variety has no “interest-
ing” motives other than powers of the Tate motive.) Moreover, the “additional”
motive on the C-Y double cover is exactly the one associated (in [Sch2]) to the
modular form f .
Let E be an elliptic curve with o ∈ E as its origin and x ∈ E a point of order
5. Under the morphism a : E → |3[o]|, the divisors 3[o], [o] + [x] + [4x],
2[x]+[3x] and 2[3x]+[4x] are linear sections. There is a unique identification
of |3[o]| with P2 under which these sections are identified with Z = 0, X = 0,
X + Y + Z = 0 and Y = 0 respectively.
Conversely, let E be a cubic curve in P2 which has the following properties:
(1) E passes through the points (0 : 1 : 0), (0 : 1 : −1), (1 : 0 : −1) and
(0 : 0 : 1).
(2) The line Z = 0 is an inflectional tangent to E (at the point (0 : 1 : 0)).
(3) The line X + Y + Z = 0 is tangent to E at the point (1 : 0 : −1).
(4) The line Y = 0 is tangent to E at the point (0 : 1 : −1).
The E is a curve of genus 1. Let o = (0 : 1 : 0), x = (0 : 1 : −1), p = (0 :
0 : 1) and q = (1 : 0 : −1). We use o as the origin of the group law on E. We
obtain the identities,
3[o] [o] + [x] + [ p] 2[x] + [q] 2[q] + [ p].
 Modular forms and Calabi-Yau varieties 367

It follows that q = 2x, p = 4x and 5x = o. Thus we have obtained the data

(E, o, x) that we started with.
Direct calculation shows us that the linear system of cubics in P2 satisfying
the above conditions is the linear span of Y Z (X + Y + Z ) and Y Z(Y + Z ) −
X (X + Z )2 .
Let E be an elliptic curve with o ∈ E as its origin and x ∈ E a point
of order 5. The morphism a : E → |2[o]| has fibres 2[o], [x] + [4x] and
[2x] + [3x], which we map to 0, 1 and ∞ to identify |2[o]| with P1 . Similarly,
the morphism b : E → |2[x]| has fibres 2[x], [o]+[2x] and [3x]+[4x], which
we map to 0, 1 and ∞ to identify |2[x]| with P1 . Thus we obtain a morphism
a ×b : E → P1 ×P1 , which is constructed canonically from the data (E, o, x).
Conversely, let E be a curve of type (2, 2) in P1 × P1 which is

(1) tangent to P1 × {0} at the point (1, 0),

(2) tangent to {0} × P1 at the point (0, 1),
(3) passes through the points (∞, 1), (∞, ∞), and (1, ∞).

Then E is a curve of genus 1. Let o = (0, 1) and x = (1, 0), which are points
on E. We use o as the origin of the group law on E. Let p, q, r denote the
points (∞, 1), (∞, ∞) and (1, ∞) respectively. We obtain the identities,

2[o] [ p] + [q] 2[x] [q] + [r ]

2[o] [x] + [r ] 2[x] [o] + [ p]

We solve these to show that a = 2x, b = 3x, c = 4x, and 5x = o. We have

thus recovered the data (E, o, x) that we started with.
Now, appealing to the fibre product paper of Schoen ([Sch1]), we can deduce
that our canonical object is Calabi-Yau.

14.9 CM forms and C-Y varieties over suitable extensions

Let E be an elliptic curve and n ≥ 1. Put
n+1
B := {x ∈ E n+1
| x j = 0},
j =1

which admits an action by the alternating group An+1 . Consider the quotient

X := B/ An+1 .

The following result is proved in [PR], where the smoothness of the model for
n = 3 appeals to ideas of Cynk and Hulek.
368 Kapil Paranjape and Dinakar Ramakrishnan

p
Theorem. X has trivial canonical bundle, with H 0 (X,  X ) = 0 if 0 < p < n.
If n ≤ 3, there is a smooth model X̃ which is Calabi-Yau.

A submotive M of rank 4 splits off of H 3 ( X̃ ) corresponding to

sym3 (H 1 (E)), of Hodge type
{(3, 0), (2, 1), (1, 2), (0, 3)}. It is simple iff E is not of CM type, and in this
case X̃ is not rigid.
Here is a sketch of proof of Proposition 2. Let  be the Hecke character
(of weight k − 1) of an imaginary quadratic field K attached to f , so that
L(s, f ) = L(s, ). Pick an algebraic Hecke character λ of K of weight 1,
with an associated elliptic curve E over a finite extension F0 . Then /λk−1 is
a finite order character ν. One attaches (by the Theorem above) a Calabi-Yau
variety V , smooth for k ≤ 4, to E k−1 . Then it has the requisite properties
relative to f over the number field F = F0 (ν), having the correct traces
of Frobenius elements at the primes outside a finite set S. There are a lot of
choices for λ, which can be used to make F and S more precise. For example,
when the discriminant −D of K is either odd or divisible by 8, with D > 4,
we may choose λ to be a canonical Hecke character of weight one ([Roh]),
which is K -valued, equivariant relative to the complex conjugation of K , and
is ramified only at the primes dividing D. There is an associated CM elliptic
curve E over an explicit number field, which is a quadratic subextension of
the Hilbert class field of K when D is odd), studied deeply by Gross in [Gro],
which is isogenous to all of its Galois conjugates. We start with the C-Y vari-
ety V associated to E k−1 . In this case S involves only the primes dividing the
level of f and D.
We will discuss Theorem 3 elsewhere in detail, where for the key case m = 3
is partly understood via the descent to GSp(4)/Q ([RS]).

14.10 Behavior under taking products

In this section we exploit an idea of Claire Voisin. M. Schuett has remarked
that Cynk and Hulek have also used such an argument.
Suppose X i is a double cover of smooth variety Yi branched along a smooth
divisor Di for i = 1, 2; let ιi denote the associated involutions on X i .
The product variety X 1 × X 2 carries an action of the involution ι = ι1 × ι2
which has D1 × D2 as its fixed locus.This is a codimension two transverse
intersection of the divisors D1 × X 2 and X 1 × D2 .
Let X 12 be the blow-up of X 1 ×X 2 along D1 ×D2 and E 12 be the exceptional
divisor. Then E 12 is isomorphic to the projective bundle over D1 × D2
 Modular forms and Calabi-Yau varieties 369

associated with the rank two-vector bundle L 1 ⊕ L 2 , where L i = pi∗ N Di / X i .

The strict transform E 1 of D1 × X 2 (respectively E 2 of X 1 × D2 ) in X 12
is a divisor that meets E 12 in a section of this projective bundle; the two
intersections E 1 ∩ E 12 and E 2 ∩ E 12 are disjoint.
Since D1 × D2 is the fixed locus of ι, this involution lifts to X 12 ; we denote
this lift also by ι by abuse of notation. The (scheme-theoretic) fixed locus of
ι on X 12 is the smooth divisor E 12 and hence the quotient Z 12 of X 12 by this
involution is a smooth variety. In other words, X 12 is a double cover of Z 12
branched along the smooth divisor D12 which is the image of E 12 .
The involution ι1 × id X 2 also lifts to X 12 since the base of the blow-up is
contained in its fixed locus. Moreover, it commutes with ι and hence descends
to an involution τ on Z 12 . The (scheme-theoretic) fixed locus of this involution
on Z 12 is the disjoint union of the images of E 1 and E 2 and is thus again a
smooth divisor. Thus Z 12 is the double cover of a variety Y12 branched along
a smooth divisor. Moreover, one checks that Y12 is just the blow-up of Y1 × Y2
along D1 × D2 (where by abuse of notation we are using the same notation for
the divisors Di in X i and for their images in Yi ).
Finally, let us calculate the canonical bundle of Z 12 and check if it is trivial.
First note that the double cover X of a smooth variety Y branched along a
smooth divisor D is obtained by choosing an isomorphism of OY (D) with the
square of some line bundle L on Y . In this case, the canonical bundle of X is
the pull back of K Y ⊗ L.
Next let L i denote the square root of OYi (Di ). The canonical bundle of K X i
is the pull-back of K Yi ⊗ L i ; so if X i is a Calabi-Yau variety, then K Yi is the
dual of L i . If E denotes the exceptional locus of the morphism e : Y12 →
Y1 × Y2 , then we see that if D̂1 denotes the strict transform of D1 × Y2 in Y12
we have
OY12 ( D̂1 + E) = e∗ OY1 ×Y2 (D1 × Y2 )
Similarly, we have
OY12 ( D̂2 + E) = e∗ OY1 ×Y2 (Y1 × D2 )
It follows that
OY12 ( D̂1 + D̂2 ) = e∗ (L 1 ⊗ L 2 ⊗ OY12 (−E))2
Since the canonical bundle of Y12 is
K Y12 = e∗ K Y1 ×Y2 ⊗ OY12 (E) = e∗ (L 1  L 2 )−1 ⊗ OY12 (E)
it follows that the canonical bundle of the double cover Z 12 is trivial.
Theorem 4 now follows by applying this construction.
370 Kapil Paranjape and Dinakar Ramakrishnan

Appendix: A consequence of the Hirzebruch

Riemann-Roch theorem
The object of this appendix to the paper is to say that there is no obstruction
from the Hirzebruch Riemann-Roch theorem for the cohomology of an odd
dimensional Calabi-Yau manifold to be of the structure that we prescribe in
the main conjecture we pursue in this paper. Specifically, there can be C-Y
manifolds X whose Hodge structure looks like a sum of H p, p ’s (for various p)
with H n,0 ⊕ H 0,n , where n is the dimension of X .

Theorem. For any odd dimensional smooth projective variety X of dimension

m ≥ 3, the Chern class cm (TX ) does not enter into the expression for the
Euler characteristic χ (O X ) when it is written as a sum of monomials in Chern
classes.

Proof. Let X be any smooth projective variety. Recall that the Todd classes of
a variety are the multiplicative classes defined by the generating function
t t t2 t4 t6
td(t) = =1+ + − + + O(t 8 )
1 − exp(−t) 2 12 720 30240
Fix an integer m and for i = 1, . . . , m, let βi be algebraic numbers (depending
on m) such that
 m
td(t) ≡ (1 + βi t) mod t m+1
i=1
For X of dimension m, let ci for i = 1, . . . , m be the Chern classes of its
tangent bundle. Let γi for i = 1, . . . , m be the Chern roots so that

m
1 + c1 t + c2 t 2 + · · · + cm t m = (1 + γi t)
i =1
The Todd polynomial of the variety is then given by

m
Todd(t) = td(γi t)
i=1
The coefficient of t m in Todd(t) is the m-th Todd class Toddm of the variety.
Making use of the above expression for td(t) mod t m+1 ,

m m 
 
Todd(t) ≡ (1+β j γi t) ≡ 1 + c1 (β j t) + c2 (β j t)2 + · · · + cm (β j t)m ,
i, j =1 j =1

where ≡ denotes congruence modulo t m+1 . Consequently, the coefficient of


cm in the m-th Todd class Toddm is mj=1 β mj .
 Modular forms and Calabi-Yau varieties 371

We can compute this as follows. Consider the function f (t) = log(td(t)). It

has an expression modulo t m+1 as
m m ∞
(−1)k−1
f (t) = log(1 + β j t) = (β j t)k mod t m+1
k
j =1 j =1 k=1

On the other hand we have f (−t) = f (t) − t so that in the expression of f (t)
as a power series in t, all the odd degree terms except t have coefficient 0. In

particular, it follows that mj=1 β mj is 0 whenever m is odd and m > 1.
To summarize, we have proved that the coefficient of the top Chern class
in the Todd class of an m-dimensional (smooth projective) variety X is 0 if
m is odd and > 1. Since this top Chern class can be identified with the Euler
characteristic of X , we have the result that we claimed.

References
[Clo1] L. Clozel. Motifs et formes automorphes: applications du principe de foncto-
rialité. In Automorphic forms, Shimura varieties, and L-functions, Vol. I (Ann
Arbor, MI, 1988), volume 10 of Perspect. Math., pages 77–159. Academic
Press, Boston, MA, 1990.
[Clo2] L. Clozel. Représentations galoisiennes associées aux représentations auto-
morphes autoduales de GL(n). Inst. Hautes Études Sci. Publ. Math. (1991),
97–145.
[CHL] L. Clozel, M. Harris, and J.-P. Labesse, Construction of automorphic Galois
representations, I. On the stabilization of the trace formula. Stab. Trace For-
mula Shimura Var. Arithm. Appl. 1, Int. Press, Somerville, MA, 2011, pages
497–527.
[CH] S. Cynk and K. Hulek. Higher-dimensional modular Calabi-Yau
manifolds. Canad. Math. Bull. 50(2007), 486–503. http://dx.
doi.org/10.4153/CMB-2007-049-9
[CS] S. Cynk and M. Schütt. Generalised Kummer constructions and Weil
restrictions. J. Number Theory 129(2009), 1965–1975. http://dx.
doi.org/10.1016/j.jnt.2008.09.010
[Del] P. Deligne. Formes modulaires et reprsentations -adiques. Séminaire Bour-
baki 1968/69 355(1971), 139–172.
[DMOS] P. Deligne, J. S. Milne, Arthur Ogus, and Kuang-yen Shih. Hodge cycles,
motives, and Shimura varieties, volume 900 of Lecture Notes in Mathemat-
ics. Springer-Verlag, Berlin, 1982.
[Die] L. Dieulefait. On the modularity of rigid Calabi-Yau threefolds: Epilogue.
Proceedings of the trimester on Diophantine Equations at the Hausdorff
Institute (2010), 1–7.
[Edi] B. Edixhoven. Introduction, main results, context. In Computational aspects
of modular forms and Galois representations, volume 176 of Ann. of Math.
Stud., pages 1–27. Princeton Univ. Press, Princeton, NJ, 2011.
[ES] N. D. Elkies and M. Schütt. Modular forms and K3 surfaces. Adv.
Math. 240(2013), 106–131. http://dx.doi.org/10.1016/j.aim.
2013.03.008
372 Kapil Paranjape and Dinakar Ramakrishnan

[GKY] F. Q. Gouvêa, I. Kiming, and N. Yui. Quadratic twists of rigid

Calabi-Yau threefolds over Q. In Arithmetic and geometry of K3 sur-
faces and Calabi-Yau threefolds, volume 67 of Fields Inst. Commun.,
pages 517–533. Springer, New York, 2013. http://dx.doi.org/10.
1007/978-1-4614-6403-7_20
[GY] F. Q. Gouvêa and N. Yui. Rigid Calabi-Yau threefolds over Q are
modular. Expo. Math. 29(2011), 142–149. http://dx.doi.org/
10.1016/j.exmath.2010.09.001
[Gro] B. H. Gross. Arithmetic on elliptic curves with complex multiplication, vol-
ume 776 of Lecture Notes in Mathematics. Springer, Berlin, 1980. With an
appendix by B. Mazur.
[KW] C. Khare and J.-P. Wintenberger. Serre’s modularity conjecture. I.
Invent. Math. 178(2009), 485–504. http://dx.doi.org/10.1007/
s00222-009-0205-7
[Kim] H. H. Kim. Functoriality for the exterior square of GL4 and the symmetric
fourth of GL2 . J. Amer. Math. Soc. 16(2003), 139–183 (electronic). With
Appendix 1 by Dinakar Ramakrishnan and Appendix 2 by Kim and Peter
Sarnak.
[KS] H. H. Kim and F. Shahidi. Functorial products for GL2 × GL3 and the
symmetric cube for GL2 . Ann. of Math. (2) 155(2002), 837–893. With an
appendix by Colin J. Bushnell and Guy Henniart.
[Kis] M. Kisin. Modularity of 2-adic Barsotti-Tate representations. Invent. Math.
178(2009), 587–634. http://dx.doi.org/10.1007/s00222-
009-0207-5
[Mey] C. Meyer. Modular Calabi–Yau threefolds. Fields Institute Monographs, 22.
American Mathematical Society, Providence, RI, 2005.
[PR] K. Paranjape and D. Ramakrishnan. Quotients of E n by An+1 and Calabi-
Yau manifolds. In Algebra and number theory, pages 90–98. Hindustan
Book Agency, Delhi, 2005.
[Ram] D. Ramakrishnan. Modularity of the Rankin-Selberg L-series, and multiplic-
ity one for SL(2). Ann. of Math. (2) 152(2000), 45–111.
[RS] D. Ramakrishnan and F. Shahidi. Siegel modular forms of genus 2 attached
to elliptic curves. Math. Res. Lett. 14(2007), 315–332.
[Roh] D. E. Rohrlich. On the L-functions of canonical Hecke charac-
ters of imaginary quadratic fields. Duke Math. J. 47(1980), 547–557.
http://projecteuclid.org/euclid.dmj/1077314180
[Sch1] C. Schoen. On fiber products of rational elliptic surfaces with
section. Math. Z. 197(1988), 177–199. http://dx.doi.org/10.
1007/BF01215188
[Sch2] A. J. Scholl. Motives for modular forms. Invent. Math. 100(1990), 419–430.
http://dx.doi.org/10.1007/BF01231194
[YL] N. Yui and J. D. Lewis, editors. Calabi-Yau varieties and mirror symme-
try, volume 38 of Fields Institute Communications, Providence, RI, 2003.
American Mathematical Society.
[YYL] N. Yui, S.-T. Yau, and J. D. Lewis, editors. Mirror symmetry. V, volume 38 of
AMS/IP Studies in Advanced Mathematics, Providence, RI, 2006. American
Mathematical Society.
 15
Derivative of symmetric square p-adic
L-functions via pull-back formula
Giovanni Rosso∗

In this paper we recall the method of Greenberg and Stevens to calculate

derivatives of p-adic L-functions using deformations of Galois representation
and we apply it to the symmetric square of a modular form Steinberg at p.
Under certain hypotheses on the conductor and the Nebentypus, this proves a
conjecture of Greenberg and Benois on trivial zeros.

Contents
15.1 Introduction 373
15.2 The method of Greenberg and Stevens 376
15.3 Eisenstein measures 381
15.3.1 Siegel modular forms 381
15.3.2 Eisenstein series 384
15.3.3 Families for GL2 × GL2 388
15.4 p-adic L-functions 392
References 398

15.1 Introduction
Let M be a motive over Q and suppose that it is pure of weight zero and
irreducible. We suppose also that s = 0 is a critical integer à la Deligne.
Fix a prime number p and let V be the p-adic representation associated to
M. We fix once and for all an isomorphism C ∼ = C p . If V is semistable, it
∗ PhD Fellowship of Fund for Scientific Research - Flanders, partially supported by a JUMO
grant from KU Leuven (Jumo/12/032), a ANR grant (ANR-10-BLANC 0114 ArShiFo) and a
NSF grant (FRG DMS 0854964).

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

373
374 Giovanni Rosso

is conjectured that for each regular submodule D [Ben11, §0.2] there exists a
p-adic L-function L p (V, D, s). It is supposed to interpolate the special values
of the L-function of M twisted by finite-order characters of 1 + pZ p [PR95],
multiplied by a corrective factor (to be thought of as a part of the local ep-
silon factor at p) which depends on D. In particular, we expect the following
interpolation formula at s = 0;
L(V , 0)
L p (V, D, s) = E(V, D) ,
(V )
for (V ) a complex period and E(V , D) some Euler type factors which con-
jecturally have to be removed in order to permit p-adic interpolation (see
[Ben11, §2.3.2] for the case when V is crystalline). It may happen that cer-
tain of these Euler factors vanish. In this case the connection with what we
are interested in, the special values of the L-function, is lost. Motivated by the
seminal work of Mazur-Tate-Teitelbaum [MTT86], Greenberg in the ordinary
case [Gre94] and Benois [Ben11] have conjectured the following;

Conjecture 15.1. [Trivial zeros conjecture] Let e be the number of Euler-

type factors of E(V, D) which vanish. Then the order of zeros at s = 0 of
L p (V, D, s) is e and
L p (V , D, s) L(V, 0)
lim = L(V, D)E ∗ (V, D) (15.1)
s→0 se (V )
for E ∗ (V, D) the non-vanishing factors of E(V, D) and L(V, D) a non-zero
number called the L-invariant.

There are many different ways in which the L-invariant can be defined.
A first attempt at such a definition could that of an analytic L-invariant
L p (V,D,s)
lims→0 se
Lan (V, D) = L(V ,0)
.
E ∗ (V, D) (V )
Clearly, with this definition, the above conjecture reduces to the statement on
the order of L p (V, D, s) at s = 0 and the non-vanishing of the L-invariant.
In [MTT86] the authors give a more arithmetic definition of the L-invariant
for an elliptic curve, in terms of an extended regulator on the extended
Mordell-Weil group. The search for an intrinsic, Galois theoretic interpre-
tation of this error factor led to the definition of the arithmetic L-invariant
Lar (V, D) given by Greenberg [Gre94] (resp. Benois [Ben11]) in the ordinary
case (resp. semistable case) using Galois cohomology (resp. cohomology of
(ϕ, )-module).
For two-dimensional Galois representations many more definitions have
been proposed and we refer to [Col05] for a detailed exposition.
 Derivative of symmetric square p-adic L-functions 375

When the p-adic L-function can be constructed using an Iwasawa cohomol-

ogy class and the big exponential [PR95], one can use the machinery developed
in [Ben12, §2.2] to prove formula (15.1) with L = Lal . Unluckily, it is a very
hard problem to construct classes in cohomology which are related to special
values. Kato’s Euler system has been used in this way in [Ben12] to prove
many instances of Conjecture 15.1 for modular forms. It might be possible
that the construction of Lei-Loeffler-Zerbes [LLZ12] of an Euler system for
the Rankin product of two modular forms could produce such Iwasawa classes
for other Galois representations; in particular, for V = Sym2 (V f )(1), where
V f is the Galois representation associated to a weight two modular form (see
also [PR98] and the upcoming work of Dasgupta on Greenberg’s conjecture
for the symmetric square p-adic L-functions of ordinary forms). We also refer
the reader to [BDR14].
We present in this paper a different method which has already been used
extensively in many cases and which we think to be more easy to apply at
the current state: the method of Greenberg and Stevens [Gre94]. Under certain
hypotheses which we shall state in the next section, it allows us to calculate
the derivative of L p (V , D, s).
The main ingredient of their method is the fact that V can be p-adically
deformed in a one-dimensional family. For example, modular forms can
be deformed in a Hida-Coleman family. We have decided to present this
method because of the recent developments on families of automorphic forms
[AIP12, Bra13] have opened the door to the construction of families of
p-adic L-function in many different settings. For example, we refer to the
ongoing PhD thesis of Z. Liu on p-adic L-functions for Siegel modular
forms.
Consequently, we expect that one could prove many new instances of
Conjecture 15.1.
In Section 15.4 we shall apply this method to the case of the symmetric
square of a modular form which is Steinberg at p. The theorem which will be
proved is the following:

Theorem 15.2. Let f be a modular form of trivial Nebentypus, weight k0 and

conductor N p, N squarefree and prime to p. If p = 2, suppose then k = 2.
Then Conjecture 15.1 (up to the non-vanishing of the L-invariant) is true for
L p (s, Sym2 ( f )).

In this case there is only one choice for the regular submodule D and the
trivial zero appears at s = k0 − 1. This theorem generalizes [Ros13b, Theorem
1.3] in two ways: we allow p = 2 in the ordinary case and when p = 2 we
376 Giovanni Rosso

do not require N to be even. In particular, we cover the case of the symmetric

square 11-adic L-function for the elliptic curve X 0 (11) and of the 2-adic L-
function for the symmetric square of X 0 (14).
In the ordinary case (i.e. for k = 2), the proof is completly independent of
[Ros13b] as we can construct directly a two-variable p-adic L-function. In the
finite slope case we can not construct a two-variable function with the method
described below and consequently we need the two-variable p-adic L-function
of [Ros13b, Theorem 4.14], which has been constructed only for p = 2.
In the ordinary setting, the same theorem (but with the hypothesis at 2) has
been proved by Greenberg and Tilouine (unpublished). The importance of this
formula for the proof of the Greenberg-Iwasawa Main Conjecture [Urb06] has
been put in evidence in [HTU97].
We improve on the result of [Ros13b] making use of a different construc-
tion of the p-adic L-function, namely that of [BS00]. We express the complex
L-function using Eisenstein series for GSp4 , a pullback formula to the Igusa
divisor and a double Petersson product. We are grateful to É. Urban for having
suggested this approach to us. In Section 15.3 we briefly recall the theory of
Siegel modular forms and develop a theory of p-adic modular forms for GL2 ×
GL2 necessary for the construction of p-adic families of Eisenstein series.

Acknowledgements. This paper is part of the author’s PhD thesis and the au-
thor is very grateful to his director J. Tilouine for the constant guidance and the
attention given. Thanks go to É. Urban for inviting the author to Columbia Uni-
versity (where this work has seen the day) and for the generosity with which
he shared with his ideas and insights. R. Casalis is thanked for interesting
discussions on Kähler differentials and J. Welliaveetil for useful remarks.

15.2 The method of Greenberg and Stevens

The aim of this section is to recall the method of Greenberg and Stevens
[GS93] to calculate analytic L-invariant. This method has been used success-
fully many other times [Mok09, Ros13a, Ros13b]. It is very robust and easily
adaptable to many situations in which the expected order of the trivial zero is
one. We also describe certain obstacles which occur while trying to apply this
method to higher order zeros.
We let K be a p-adic local field, O its valuation ring and  the Iwasawa
algebra O[[T ]]. Let V be a p-adic Galois representation as before.
We denote by W the rigid analytic space whose C p -points are
Homcont (Z∗p , C∗p ). We have a map Z → W defined by k → [k] : z → z k .
 Derivative of symmetric square p-adic L-functions 377

Let us fix once and for all, if p = 2 (resp. p = 2) a generator u of 1 + pZ p

(resp. 1+4Z2 ) and a decomposition Z∗p = μ×1+ pZ p (resp Z∗2 = μ×1+4Z2 ).
Here μ = Gtors
m (Z p ). We have the following isomorphism of rigid spaces:
∧
W∼
= Gtors −
m (Z p ) × B(1, 1 )
κ → (κ|μ , κ(u)),

where the first set has the discrete topology and the second is the rigid open
unit ball around 1. Let 0 < r < ∞, r ∈ R, we define
1 −r
2
W(r ) = (ζ, z)|ζ ∈ Gtorsm (Z p ), |z − 1| ≤ p .

We fix an integer h and we denote by Hh the algebra of h-admissible distribu-

tions over 1 + pZ p (or 1 + 4Z2 if p = 2) with values in K . Here we take the
definition of admissibility as in [Pan03, §3], so measures are one-admissible,
i.e.  ⊗ Q p ∼
= H1 . The Mellin transform gives us a map

Hh → An(0),

where An(0) stands for the algebra of Q p -analytic, locally convergent func-
tions around 0. If we see Hh as a subalgebra of the ring of formal series, this
amounts to T → u s − 1.
We suppose that we can construct a p-adic L-function for L p (s, V, D) and
that it presents a single trivial zero.
We suppose also that V can be deformed in a p-adic family V (κ). Pre-
cisely, we suppose that we are given an affinoid U , finite over W(r). Let us
write π : U → W(r ). Let us denote by I the Tate algebra corresponding to
U. We suppose that I is integrally closed and that there exists a big Galois
representation V (κ) with values in I and a point κ0 ∈ U such that V = V (κ0 ).
We define U cl to be the set of κ ∈ U satisfying the following conditions:
● π(κ) = [k], with k ∈ Z,
● V (κ) is motivic,
● V (κ) is semistable as G Q p -representation,
● s = 0 is a critical integer for V (κ).

We make the following assumption on U cl .

(CI) For every n > 0, there are infinitely many κ in U cl and s ∈ Z such
that:
● |κ − κ0 |U < p −n
● s critical for V (κ),
● s ≡ 0 (mod p)n .
378 Giovanni Rosso

This amounts to asking that the couples (κ, [s]) in U × W with s critical for
V (κ) accumulate at (κ0 , 0).
We suppose that there is a global triangulation D(κ) of the (ϕ, )-module
associated to V (κ) [Liu13] and that this induces the regular submodule used
to construct L p (s, V, D).
Under these hypotheses, it is natural to conjecture the existence of a
two-variable p-adic L-function (depending on D(κ)) L p (κ, s) ∈ I ⊗H ˆ h in-
terpolating the p-adic L-functions of V (κ), for all κ in U . Conjecturally
cl

[Pan94, Pot13], h should be defined solely in terms of the p-adic Hodge theory
of V (κ) and D(κ).
We make two hypotheses on this p-adic L-function.
i) There exists a subspace of dimension one (κ, s(κ)) containing (κ0 , 0) over
which L p (κ, s(κ)) vanishes identically.
ii) There exists an improved p-adic L-function L ∗p (κ) in I such that
L p (κ, 0) = E(κ)L ∗p (κ), for E(κ) a non-zero element which vanishes at κ0 .
The idea is that i) allows us to express the derivative we are interested in in
terms of the “derivative with respect to κ”. The latter can be calculated using
ii). In general, we expect that s(κ) is a simple function of π(κ).

Let log p (z) = − ∞ n=1
(1−z)n
n , for |z − 1| p < 1 and
log p (κ(u r ))
Log p (κ) = ,
log p (u r )
for r any integer big enough.
For example, in [Gre94, Mok09] we have s(κ) = 12 Log p (π(κ)) and in
[Ros13a, Ros13b] we have s(κ) = Log p (π(κ)) − 1. In the first case the line
corresponds to the vanishing on the central critical line which is a consequence
of the fact that the ε-factor is constant in the family. In the second case, the van-
ishing is due to a line of trivial zeros, as all the motivic specializations present
a trivial zero.
The idea behind ii) is that the Euler factor which brings the trivial zero for
V varies analytically along V (κ) once one fixes the cyclotomic variable. This
is often the case with one-dimensional deformations. If we allow deformations
of V in more than one variable, it is unlikely that the removed Euler factors
define p-analytic functions, due to the fact that eigenvalues of the crystalline
Frobenius do not vary p-adically or equivalently, that the Hodge-Tate weights
are not constant.
We now give the example of families of Hilbert modular forms. For simplic-
ity of notation, we consider a totally real field F of degree d where p is split.
Let f be a Hilbert modular form of weight (k1 , . . . , kd ) such that the parity of
 Derivative of symmetric square p-adic L-functions 379

ki does not depend on i . We define m = max(ki − 1) and vi = m+1−k 2

i
. We
suppose that f is nearly-ordinary [Hid89] and let F be the only Hida family to
which f belongs. For each p-adic place pi of F, the corresponding Hodge-Tate
weights are (vi , m − vi ). This implies that the Fourier coefficient api (F) is a
p-adic analytic function only if it is divided by p vi . Unluckily, api (F) is the
number which appears in the Euler type factor of the evaluation formula for
the p-adic L-function of F or Sym2 (F). This is why in [Mok09, Ros13a] the
authors deal only with forms of parallel weight. It seems very hard to gener-
alize the method of Greenberg and Stevens to higher order derivatives without
new ideas.
It may happen that the Euler factor which brings the trivial zero for V is
(locally) identically zero on the whole family; this is the case for the symmetric
square of a modular form of prime-to- p conductor and more generally for the
standard L-function of parallel weight Siegel modular forms of prime-to- p
level. That’s why in [Ros13a, Ros13b] and in this article we can deal only with
forms which are Steinberg at p.
We have seen in the examples above that s(κ) is a linear function of
the weight. Consequently, one needs to evaluate the p-adic L-function
L p (V (κ), s) at s which are big for the archimedean norm. When s is not a crit-
ical integer it is quite a hard problem to evaluate the p-adic L-function. This
is why we have supposed (CI). It is not a hypothesis to be taken for granted.
One example is the spinor L-function for genus two Siegel modular forms of
any weight, which has only one critical integer.
The improved p-adic L-function is said so because L ∗p (κ0 ) is supposed to
be exactly the special value we are interested in.
The rest of the section is devoted to make precise the expression “derive
with respect to κ”.
We recall some facts about differentials. We fix a -algebra I1 . We suppose
that I1 is a DVR and a K -algebra. Let I2 be an integral domain and a local ring
which is finite, flat and integrally closed over I1 . We have the first fundamental
sequence of Kähler differentials
I1 /K ⊗I1 I2 → I2 /K → I2 /I1 → 0.
1 [X ]
Under the hypotheses above, we can write I2 = IP(X) . Then, every K -linear
derivation of I1 can be extended to a derivation of I2 and [Mat80, Theorem 57
ii)] ensures us that the first arrow is injective.
Let P0 be the prime ideal of I corresponding to the point κ0 and P the cor-
responding ideal in O(W(r )). We take I1 = O(W(r )) P and I2 = I P0 . The
assumption that I2 is integrally closed is equivalent to ask that U is smooth
at κ0 . In many cases, we expect that I1 → I2 is étale; under this hypothesis,
380 Giovanni Rosso

we can appeal to the fact that locally convergent series are Henselian [Nag62,
Theorem 45.5] to define a morphism I2 → An(k0 ) to the ring of meromorphic
functions around k0 which extends the natural inclusion of I1 . Once this mor-
phism is defined, we can derive elements of I2 as if they were locally analytic
functions.
There are some cases in which this morphism is known not to be étale; for
example, for certain weight one forms [Dim13, DG12] and some critical CM
forms [Bel12, Proposition 1]. (Note that in these cases no regular submodule
D exists.)
We would like to explain what we can do in the case when I1 → I2 is not
étale. We also hope that what we say will clarify the situation in the case where
the morphism is étale.
We have that /K is a free rank 1 -module. Using the universal property
of differentials

Hom (/K , ) = Der K (, ),

we shall say, by slight abuse of notation, that /K is generated as -module

by the derivation dT d
. Similarly, we identify I1 /K with the free I1 -module
d
generated by dT . As the first arrow in the first fundamental sequence is injec-
tive, there exists an element d ∈ I2 /K (which we see as K -linear derivation
from I2 to I2 ) which extends dT d
. If I1 → I2 is étale, then I2 /I1 = 0 and
the choice of d is unique.
Under the above hypotheses, we can then define a new analytic L-invariant
(which a priori depends on the deformation V (κ) and d) by
/
−1 /
Lan
d (V ) := −log p (u) d(s(κ)) d(E(κ))/
κ=κ0
/
/
d(L p (0, κ)) /
= − d(s(κ))−1 / .
L ∗p (κ) /
κ=κ0

We remark that with the notation above we have log p (u) dT

d
= d
ds . We apply d
to L p (κ, s(κ)) = 0 to obtain

0 = d(L p (κ, s(κ)))

/
d / /
= log p (u)−1
L p (κ, s)// d(s(κ)) + d(L p (κ, s))/s=s(κ) .
ds s=s(κ)

Evaluating at κ = κ0 we deduce
/
d /
L p (V, s)// =Lan ∗
d (V )L p (κ0 )
ds s=0
 Derivative of symmetric square p-adic L-functions 381

and consequently

Lan (V ) = Lan
d (V ).

In the cases in which Lal has been calculated, namely symmetric powers
of Hilbert modular forms, it is expressed in terms of the logarithmic deriva-
tive of Hecke eigenvalues at p of certain finite slope families [Ben10, HJ13,
Hid06, Mok12]. Consequently, the above formula should allow us to prove the
equality Lan = Lal .
Moreover, the fact that the L-invariant is a derivative of a non-constant func-
tion shows that Lan = 0 outside a codimension 1 subspace of the weight space.
In this direction, positive results for a given V have been obtained only in
the cases of a Hecke character of a quadratic imaginary field and of an el-
liptic curve with p-adic uniformization, using a deep theorem in transcendent
number theory [BSDGP96].

15.3 Eisenstein measures

In this section we first fix the notation concerning genus two Siegel forms.
We then recall a normalization of certain Eisenstein series for GL2 × GL2 and
develop a theory of p-adic families of modular forms (of parallel weight) on
GL2 × GL2 . Finally we construct two Eisenstein measures which will be used
in the next section to construct two p-adic L-functions.

15.3.1 Siegel modular forms

We now recall the basic theory of Siegel modular forms. We follow closely the
notation of [BS00] and we refer to the first section of loc. cit. for more details.
Let us denote by H1 the complex upper half-plane and by H2 the Siegel space
for GSp4 . We have explicitly
D  / E
z1 z2 / t
H2 = Z= / Z = Z and Im(Z ) > 0 .
z3 z4 /

It has a natural action of GSp4 (R) via fractional linear transformation; for any
 
A B
M= in GSp+4 (R) and Z in H2 we define
C D

M(Z ) = (AZ + B)(C Z + D)−1 .

382 Giovanni Rosso

(2)
Let = 0 (N ) be the congruence subgroup of Sp4 (Z) of matrices whose
(2)
lower block C is congruent to 0 modulo N . We consider the space Mk (N , φ)
of scalar Siegel forms of weight k and Nebentypus φ:
/
/
F : H2 → C /F(M(Z ))(C Z + D)−k

= φ(M)F(Z ) ∀M ∈ , f holomorphic .

Each F in Mk ( , φ) admits a Fourier expansion

F(Z) = a(T )e2πi tr(T Z ) ,

T
 
T1 T2
where T = ranges over all matrices T positive and semi-defined,
T2 T4
with T1 , T4 integer and T2 half-integer.
We have two embeddings (of algebraic groups) of SL2 in Sp4 :
⎧⎛ ⎞ ⎫
⎪
⎪ a 0 b 0 /
  ⎪
⎪
⎨⎜ ⎟ / ⎬
↑ ⎜ 0 1 0 0 ⎟ / a b
SL2 (R) = ⎝ ∈ SL (R) ,
c 0 d 0 ⎠/ c d
2
⎪
⎪ ⎪
⎪
⎩ ⎭
0 0 0 1
⎧⎛ ⎞ ⎫
⎪
⎪ 1 0 0 0 /   ⎪
⎪
⎨⎜ ⎟ / ⎬
↓ ⎜ 0 a 0 b ⎟ / a b
SL2 (R) = ⎝ ∈ SL (R) .
0 0 1 0 ⎠/ c d
2
⎪
⎪ ⎪
⎪
⎩ ⎭
0 c 0 d
We can embed H1 × H1 in H2 in the following way
 
z1 0
(z 1 , z 4 ) → .
0 z4
If γ belongs to SL2 (R), we have
   
z1 0 γ (z 1 ) 0
γ↑ = ,
0 z4 0 z4
   
z1 0 z1 0
γ↓ = .
0 z4 0 γ (z 4 )
Consequently, evaluation at z 2 = 0 gives us a map

Mk(2) (N , φ) → Mk (N , φ) ⊗C Mk (N , φ),

where Mk (N, φ) denotes the space of elliptic modular forms of weight k, level
N and Nebentypus φ.
 Derivative of symmetric square p-adic L-functions 383

This also induces a closed embedding of two copies of the modular curve
in the Siegel threefold. We shall call its image the Igusa divisor. On points, it
corresponds to abelian surfaces which decompose as the product of two elliptic
curves.
We consider the following differential operators on H2 :
∂ 1 ∂ ∂
∂1 = , ∂2 = , ∂4 = .
∂z 1 2 ∂ z2 ∂z 4
We define
   1

Dl =z 2 ∂1 ∂4 − ∂22 − l − ∂2 ,
2
Dls =Dl+s−1 ◦ . . . ◦ Dl ,
D̊ls =Dls |z 2 =0 .

The importance of D̊ls is that it preserves holomorphicity.

 
T1 T2
Let I = . We define bls (I ) to be the only homogeneous
T2 T4
polynomial in the indeterminates T1 , T2 , T4 of degree s such that

D̊ls e T1 z 1 +2T2 z2 +T4 z 4 = bls (I )eT1 z1 +T4 z 4 .

We need to know a little bit more about the polynomial bls (I ). Let us write

Dls eT1 z1 +2T2 z 2 +T4 z4 = Pls (z 2 , I )e T1 z1 +2T2 z 2 +T4 z4 ,

we have
=
Dls+1 etr(Z I ) = Pl+s
1
(z 2 , I )Pls (z 2 , I )
    >
1 ∂2 1 1 ∂
− z2 + l +s− Pl (z 2 , I ) etr(Z I ) .
s
(15.2)
4 ∂z 22 2 2 ∂ z2

We obtain easily
   
1
Pl1 (z 2 , I ) = z 2 det(I ) − l − T2 ,
2
bls (I ) =Pls (0, I ). (15.3)

We have

bls+1 (I ) =bls (I )bl+s

1
(I ) + (l + s − 1/2)∂2 Pls (z 2 , I )
=(l + s − 1/2)(−T2 bls (I ) + ∂2 Pls (z 2 , I )|z 2 =0 ).
384 Giovanni Rosso

j j j j j j
Let J = { j1 , j2 , j4 }. We shall write ∂ J resp. z J for ∂11 ∂22 ∂44 resp. z 11 z 22 z 44 .
We can write [BS00, page 1381]

D̊ls = clJ ∂ J , (15.4)

j1 + j2 + j4 =s

D̊ls (z J )
where clJ = ∂ J (z J )
. We have easily:

∂ J (z J ) = j1 ! j2 ! j4 !2− j2 ,
∂ J (etr(Z I ) ) =T1 1 T2 2 T4 4 (etr(Z I ) ),
j j j

We pose

cls :=cl0,s,0 ,
s 
 
i (2l − 2 + 2s − 1)!
(−1)s cls = l −1+s− = 2−s .
2 (2l − 2 + s − 1)!
i =1

Consequently, for L | T1 , T4 and for any positive integer d, we obtain

j s− j1 − j4 j
4s bst+1 (I ) ≡ (−1)s 4s J
ct+1 T1 1 T2 T4 4 (mod L)d .
j1 + j4 <d

15.3.2 Eisenstein series

The aim of this section is to recall certain Eisenstein series which can be used to
construct the p-adic L-functions, as in [BS00]. In loc. cit. the authors consider
certain Eisenstein series for GSp4g whose pullback to the Igusa divisor is a
holomorphic Siegel modular form.
We now fix a (parallel weight) Siegel modular form f for GSp2g . We write
the standard L-function of f as a double Petersson product between f and
these Eisenstein series (see Proposition 15.4). When g = 1, the standard L-
function of f coincides, up to a twist, with the symmetric square L-function
of f we are interested in.
In general, for an algebraic group bigger than GL2 it is quite hard to find the
normalization of the Eisenstein series which maximizes, in a suitable sense, the
p-adic behavior of its Fourier coefficients. In [BS00, §2] the authors develop
a twisting method which allow them to define Eisenstein series whose Fourier
coefficients satisfy Kummer’s congruences when the character associated with
the Eisenstein series varies p-adically. This is the key for their construction of
the one variable (cyclotomic) p-adic L-function and of our two-variable p-adic
L-function.
 Derivative of symmetric square p-adic L-functions 385

When the character is trivial modulo p there exists a simple relation be-
tween the twisted and the not-twisted Eisenstein series [BS00, §6 Appendix].
To construct the improved p-adic L-function, we shall simply interpolate the
not-twisted Eisenstein series.
Let us now recall these Fourier developments.
We fix a weight k, an integer N prime to p and a Nebentypus φ. Let f be an
eigenform in Mk (N p, φ), of finite slope for the Hecke operator U p . We write
N = Nss Nnss , where Nss (resp. Nnss ) is divisible by all primes q | N such that
Uq f = 0 (resp. Uq f = 0). Let R be an integer coprime with N and p and N1 a
positive integer such that Nss | N1 | N . We fix a Dirichlet character χ modulo
N1 Rp which we write as χ1 χ  ε1 , with χ1 defined modulo N1 , χ  primitive
modulo R and ε1 defined modulo p. We shall explain after Proposition 15.4
why we introduce χ1 .
 (χ )
Let t ≥ 1 be an integer and Ft+1 w, z, R 2 N 2 p 2n , φ, u be the twisted
Eisenstein series of [BS00, (5.3)]. We define
  (χ ) 

H L ,χ (z, w) := L(t + 1 + 2s, φχ )D̊t+1 F
s t+1
w, z, R N p , φ, u
2 2 2n

/z /w
/U 2/ U 2
L L

for s a non-negative integer and pn | L, with L a p-power. It is a form for

0 (N R p) × 0 (N R p) of weight t + 1 + s.
2 2 2 2

We shall sometimes choose L = 1 and in this case the level is N 2 R 2 p 2n .

For any prime number q and matrices I as in the previous section, let
Bq (X, I ) be the polynomial of degree at most 1 of [BS00, Proposition 5.1].
We pose
21+2t 5
B(t) = (−1)t+1 π2.
(3/2)
We deduce easily from [BS00, Theorem 7.1] the following theorem

Theorem 15.3. The Eisenstein series defined above has the following Fourier
development;

HL ,χ (z, w)|u= 1 −t = B(t)(2πi)s G (χ ) bst+1 (I )(χ)−1 (2T2 )
2
T1 ≥0 T4 ≥0 I

(φχ ) (det(G))|det(G)|
2 2t−1
L(1 − t, σ−det(2I ) φχ )
G∈GL2 (Z)\D(I )
⎞
  
Bq χ φ(q)q t−2 , G −t I G −1 ⎠ e2πi (T1 z+T2 w) ,
q|det(2G −t I G −1 )
386 Giovanni Rosso

 
L 2 T1 T2
where the sum over I runs along the matrices positive definite
T2 L 2 T4
and with 2T2 ∈ Z, and

D(I ) = G ∈ M2 (Z)|G −t I G −1 is a half-integral symmetric matrix .

/ 
/
Proof. The only difference from loc. cit. is that we do not apply / 10 0
N2S
.

In fact, contrary to [BS00], we prefer to work with 0 (N 2 S) and not with

the opposite congruence subgroup.
In particular each sum over I is finite because I must have positive deter-
minant. Moreover, we can rewrite it as a sum over T2 , with (2T2 , p) = 1 and
T1 T4 − T22 > 0.
It is proved in [BS00, Theorem 8.5] that (small modifications
of) these functions HL ,χ (z, w) satisfy Kummer’s congruences. The
key fact is what they call the twisting method [BS00, (2.18)]; the

t+1 w, z, R 2 N 2 p 2n , φ, u (χ ) are obtained weighting
Eisenstein
 series F
Ft+1 w, z, R 2 N 2 p 2n , φ, u with respect to χ over integral matrices modulo
N Rp n . To ensure these Kummer’s congruences, even when p does not divide
the conductor of χ, the authors are forced to consider χ of level divisible by p.
Using nothing more than Tamagawa’s rationality theorem for GL2 , they find
the relation [BS00, (7.13’)]:
 (χ )
Ft+1 w, z, R 2 N 2 p, φ, u
 (χ1 χ  ) // 
1 p

=F w, z, R N p, φ, u /
/ id − p 0 1 .
t+1 2 2

So the Eisenstein series we want to interpolate to construct the improved

p-adic L-function is
  (χ1 χ  ) 
∗
H L ,χ  (z, w) :=L(t + 1 + 2s, φχ )D̊st+1 Ft+1 w, z, R 2 N 2 p, φ, u .

In what follows, we shall specialize t = k − k0 + 1 (for k0 the weight of the

form in the theorem of the introduction) to construct the improved one variable
p-adic L-function.
For each prime q, let us denote by αq and βq the roots of the Hecke
polynomial at q associated to f . We define

Dq (X ) := (1 − αq2 X )(1 − αq βq X )(1 − βq2 X ).

 Derivative of symmetric square p-adic L-functions 387

For each Dirichlet character χ we define

 −1
L(s, Sym2 ( f ), χ ) := Dq (χ (q)q −s ) .
q

This L-function differs from the motivic L-function L(s, Sym2 (ρ f ) ⊗ χ ) by

a finite number of Euler factors at prime dividing N . We conclude with the
integral formulation of L(s, Sym2 ( f ), χ ) [BS00, Theorem 3.1, Proposition 7.1
(7.13)].

Proposition 15.4. Let f be a form of weight k, Nebentypus φ. We put t + s =

k − 1, s1 = 12 − t and H = H1,χ  (z, w)|
u=s1 ; we have
I /  J
/ 0 −1
/
f (w) / ,H 
N 2 p 2n R 2 0 N 2 p2n R 2
 2−k
k,s (s1 ) ps1 (t + 1) n s+3−k N
= (R N1 p ) ×
χ (−1)ds1 (t + 1) N1
× L(s + 1, Sym2 ( f ), χ −1 ) f (z)|U N 2 /N 2
1

for
s s 
ps1 (t + 1) ct+1 s + t − 2i
= s = i=1  ,
ds1 (t + 1) c3 s
i=1 s − 2
i−1
2
 
  (k − t) k − t − 1
1 k 2
k,s − t =22t (−1) 2 π   .
2 3
2

Proof. With the notation of [BS00, Theorem 3.1] we have M = R 2 N 2 p 2n

ds (t+1)
and N = N1 Rpn . We have that ps1 (t+1) H is the holomorphic projection of
1
the Eisenstein series of [BS00, Theorem 3.1] (see [BS00, (1.30),(2.1),(2.25)]).
The final remark to make is the following relation between the standard (or
adjoint) L-function of f and the symmetric square one:

L(1 − t, Ad( f ) ⊗ φ, χ ) = L(k − t, Sym2 ( f ), χ ).

The authors of [BS00] prefer to work with an auxiliary character modulo

N to remove all the Euler factors at bad primes of f , but we do not want to
do this. Still, we have to make some assumptions on the level χ . Suppose that
there is a prime q dividing N such that q  N1 and Uq f = 0, then the above
formula would give us zero. That is why we introduce the character χ1 defined
modulo a multiple of Nss .
388 Giovanni Rosso

At the level of L-function this does not change anything as for q | Nss we
have Dq (X ) = 1.
For f as in Theorem 15.2 we can take N1 = 1.

15.3.3 Families for GL2 × GL2

The aim of this section is the construction of families of modular forms on two
copies of the modular curves. Let us fix a tame level N , and let us denote by X
the compactified modular curve of level 1 (N ) ∩ 0 ( p). For n ≥ 2, we shall
denote by X ( p n ) the modular curve of level 1 (N ) ∩ 0 ( p n ).
We denote by X (v) the tube of ray p −v of the ordinary locus. We fix a p-adic
field K . We recall from [AIS12, Pil13] that, for v and r suitable, there exists
an invertible sheaf ωκ on X (v) × K W(r ). This allows us to define families of
overconvergent forms as

Mκ (N ) := lim H 0 (X (v) × K W(r ), ωκ ).

−→
v

We denote by ωκ,(2) the sheaf on X (v) × K X (v) × K W(r) obtained by base

change over W(r ).
We define

(2)
Mκ,v (N ) := H 0 (X (v) × K X (v) × K W(r ), ωκ,(2) )
= H 0 (X (v) × W(r ), ωκ )⊗
ˆ O(W (r)) H 0 (X (v) × W(r ), ωκ );
Mκ(2) (N ) := lim Mκ,v
(2)
(N ).
−→
v

We believe that this space should correspond to families of Siegel modu-

lar forms [AIP12] of parallel weight restricted on the Igusa divisor, but for
shortness of exposition we do not examine this now.
We have a correspondence C p above X (v) defined as in [Pil13, §4.2]. We
define by fiber product the correspondence C 2p on X (v) × K X (v) which we
extend to X (v) × K X (v) × K W(r ). This correspondence induces a Hecke
operator U p⊗ on H 0 (X (v) × K X (v) × K W(r ), ωκ,(2) ) which corresponds to
2

U p ⊗ U p . These are potentially orthonormalizable O(W(r ))-modules [Pil13,

§5.2] and U p⊗ acts on these spaces as a completely continuous operator (or
2

compact, in the terminology of [Buz07, §1]) and this allows us to write

≤α 
Mκ(2) (N ) = Mκ (N )≤α1 ⊗
ˆ O(W (r)) Mκ (N )≤α2 . (15.5)
α1 +α2 ≤α
 Derivative of symmetric square p-adic L-functions 389

Here and in what follows, for A a Banach ring, M a Banach A-module and U
a completely continuous operator on M, we write M ≤α for the finite dimen-
sional submodule of generalized eigenspaces associated to the eigenvalues of
U of valuation smaller or equal than α. We write Pr≤α for the corresponding
projection.
We remark [Urb11, Lemma 2.3.13] that there exists v > 0 such that
≤α (2) ≤α
Mκ(2) (N ) = Mκ,v  (N )

(2)
for all 0 < v  < v. We define similarly Mκ (N pn ).
We now use the above Eisenstein series to give examples of families. More
precisely, we shall construct a two-variable measure (which will be used for the
two-variable p-adic L-function in the ordinary case) and a one-variable mea-
sure (which will be used to construct the improved one variable L-function)
without the ordinary assumption.
Let us fix χ = χ1 χ  ε1 as before. We suppose χ even. We recall the Kubota-
Leopoldt p-adic L-function;

Theorem 15.5. Let η be a even Dirichlet character. There exists a p-adic L-

function L p (κ, η) satisfying for any integer t ≥ 1 and finite-order character ε
of 1 + pZ p
L p (ε(u)[t], η) = (1 − (εω−t η)0 ( p))L(1 − t, εω−t η),
where η0 stands for the primitive character associated to η. If η is not trivial
then L p ([t], η) is holomorphic. Otherwise, it has a simple pole at [0].

We can consequently define a p-adic analytic function interpolating the

Fourier coefficients of the Eisenstein series defined in the previous section;
log (z)
for any z in Z∗p , we define l z = log p (u) . We define also
p

aT1 ,T4 ,L (κ, κ  )

⎛

=⎝ κ(u 2T2 )χ −1 (2T2 ) (φχ )2 (det(G))|det(G)|−1 κ  (u l|det(G)| )

2

I G∈GL2 (Z)\D(I )
⎞
  
L p (κ  , σ−det(2I ) φχ ) Bq φ(q)κ  (u lq )q −2 , G −t I G −1 ⎠ .
q|det(2G −t I G −1 )

If p = 2, then χ −1 (2T2 ) vanishes when T2 is an integer, so the above above

sum is only on half-integral T2 and κ(u 2T2 ) is a well-defined two-analytic
function. Moreover, it is two-integral.
390 Giovanni Rosso

We recall that if p j | T1 , T4 we have [BS00, (1.21, 1.34)]

4s bst+1 (I ) ≡ (−1)s 4s ct+1

s
T2s (mod p)d ,
for s = k − t − 1. Consequently, if we define
−1
H L (κ, κ  ) = aT1 ,T4 ,L (κ[−1]κ  , κ  )q1T1 q2T4
T1 ≥0 T4 ≥0

we have,

(−1)s 2s ct+1
s
AH L ([k], ε[t]) ≡ 2s HL ,χ εω−s (z, w) (mod L)r ,
with

A = A(t, k, ε) = B(t)(2πi )s G χ εω−s .

We have exactly as in [Pan03, Definition 1.7] the following lemma;

Lemma 15.6. There exists a projector

0 ≤α
Pr≤α
∞ : Mκ(2) (N p n ) → Mκ(2) (N )
n

(2) −i i
which on Mκ (N pn ) is (U p⊗ ) Pr≤α (U p⊗ ) , independent of i ≥ n.
2 2

When α = 0, we shall write Prord ∞.

We shall now construct the improved Eisenstein family. Fix k0 ≥ 2 and
s0 = k0 − 2. It is easy to see from (15.3) and (15.2) that when k varies p-
s s
adically the value bk0 (I ) varies p-adically analytic too. We define bκ0 (I ) to
be the only polynomial in O(W(r ))[Ti ], homogeneous of degree s0 such that
s s0
b[t]0 (I ) = bt+1 (I ). Its coefficients are products of Log p (κ[i]). We let χ =
 −2
χ1 χ ω 0 and we define
k

aT∗1 ,T4 (κ)

⎛
−1
=⎝ bsκ0 (I )(χ  χ1 ) (2T2 ) (φχ  χ1 )2 (det(G))|det(G)|−1 κ(u l2|det(G)| )×
I G∈GL2 (Z)\D(I )
⎞
  
× L p (κ, σ−det(2I ) φχ  χ1 ) Bq φχ  χ1 (q)κ(u lq )q −2 , G −t I G −1 ⎠ .
q|det(2G −t I G −1 )

We now construct another p-adic family of Eisenstein series:

H∗ (κ) = Pr≤α (−1)k0 aT∗1 ,T4 (κ[1 − k0 ])q1T1 q2T2 .

T1 ≥0 T4 ≥0
 Derivative of symmetric square p-adic L-functions 391

Proposition 15.7. Suppose α = 0. We have a p-adic family H(κ, κ  ) ∈

(2) ord
Mκ (N 2 R 2 ) such that
1 2 −2i
H([k], ε[t]) = (U p⊗ ) Prord Hpi ,χ εω−s (z, w).
(−1) ct+1 A
s s

(2) ≤α
For any α, we have H∗ (κ) ∈ Mκ (N 2 R 2 ) such that
∗
H∗ ([k]) = A∗ −1 (−1)k0 Pr≤α H 1,χ1 χ  (z, w)|u= 1 −k+k0 −1 ,
2

for
A∗ = B(k − k0 + 1)(2πi)k0 −2 G(χ  χ1 ).

2j
 (z, w) = H
Proof. From its own definition we have (U p⊗ ) H1,χ
2
(z, w).
p j ,χ
We define
−2 j
H(κ, κ  ) = lim (U p⊗ ) Prord H p j ,χ (κ, κ  ).
2

−→
j

With Lemma 15.6 and the previous remark we obtain

−2i −2 j
4s (U p⊗ ) Prord Hpi ,χ εω−s (z, w) = 4s (U p⊗ ) Prord Hp j ,χ εω−s (z, w)
2 2

−2 j
≡ A(−1)s 4s cts (U p⊗ )
2

× Prord H p j ([k], ε[t]) (mod p)2 j ,

as U p−1 acts on the ordinary part with norm 1.
The punctual limit is then a well-defined classical finite slope form. By the
same method of proof of [Urb13, Corollary 3.4.7] or [Ros13b, Proposition
2.17] (in particular, recall that when the slope is bounded, the ray of overcon-
vergence v can be fixed), we see that this q-expansion defines a family of finite
slope forms.
For the second family, we remark that Log p (κ) is bounded on W(r ), for any
r and we reason as above.
We want to explain briefly why the construction above works in the ordinary
setting and not in the finite slope one.
It is slightly complicated to explicitly calculate the polynomial bst+1 (I ) and
in particulat to show that they vary p-adically when varying s. But we know
that Dls is an homogeneous polynomial in ∂i of degree s. Suppose for now
α = 0, i.e. we are in the ordinary case. We have a single monomial of Dls
which does not involve ∂1 and ∂4 , namely ct+1 s ∂ s . Consequently, in bs (I )
2 t+1
there is a single monomial without T1 and T4 .
392 Giovanni Rosso

When the entries on the diagonal of I are divisible by pi , 4s bst+1 (I ) reduces

modulo pi . Applying U p⊗ many times ensures us that T1 and
2
to (−1)s 4s ct+1
s

T4 are very divisible by p. Speaking p-adically, we approximate Dls by ∂2

(multiplied by a constant). The more times we apply U p⊗ , the better we can
2

approximate p-adically Dls by ∂2s . At the limit, we obtain equality.

2 −1
For α > 0, when we apply U p⊗ we introduce denominators of the order of
α
p . In order to construct a two-variable family, we should approximate Dls with
higher precision. For example, it would be enough to consider the monomials
∂ J for j1 + j4 ≤ α. In fact, ∂1 and ∂4 increase the slope of U p ; we have the
relation
 2   
U p⊗ ∂1 ∂4 = p 2 ∂1 ∂4 U p⊗
2
.
|z 2 =0 |z 2 =0
J
Unluckily, it seems quite hard to determine explicitly the coefficients ct+1 of
(15.4) or even show that they satisfy some p-adic congruences as done in
[CP04, Gor06, Ros13b]. We guess that it could be easier to interpolate p-
adically the projection to the ordinary locus of the Eisenstein series of [BS00,
Theorem 3.1] rather than the holomorphic projection as we are doing here (see
also [Urb13, §3.4.5] for the case of nearly holomorphic forms for GL2 ).
This should remind the reader of the fact that on p-adic forms the Maaß-
Shimura operator and Dwork #-operator coincide [Urb13, §3.2].

15.4 p-adic L-functions

We now construct two p-adic L-functions using the above Eisenstein measure:
the two-variable one in the ordinary case and the improved one for any finite
slope. Necessary for the construction is a p-adic Petersson product [Pan03, §6]
which we now recall. We fix a family F = F(κ) of finite slope modular forms
which we suppose primitive, i.e. all its classical specializations are primitive
forms, of prime-to- p conductor N . We consider characters χ , χ  and χ1 as
in Section 15.3.2. We keep the same decomposition for N (because the local
behavior at q is constant along the family). We shall write N0 for the conductor
of χ1 .

(notCM) We suppose that F has not complex multiplication by χ .

Let C F be the corresponding irreducible component of the Coleman-Mazur

eigencurve. It is finite flat over W(r ), for a certain r. Let I be the coefficients
ring of F and K its field of fraction. For a classical form f , let us denote by f c
 Derivative of symmetric square p-adic L-functions 393

the complex conjugated form. We denote by τ N the Atkin-Lehner involution of

level N normalized as in [Hid90, h4]. When the level is clear from the context,
we shall simply write τ .
Standard linear algebra allows us to define a K-linear form l F on
Mκ (N )≤α ⊗I K with the following property [Pan03, Proposition 6.7];

Proposition 15.8. For all G(κ) in Mκ (N )≤α ⊗I K and any κ0 classical point,
we have
K L
F(κ0 )c |τ N p , G(κ0 ) N p
l F (G(κ))|κ=κ0 = K L .
F (κ0 )c |τ N p , F(κ0 ) N p

We can find H F (κ) in I such that H F (κ)l F is defined over I. We shall refer
sometimes to H F (κ) as the denominator of l F .
(2) ≤α
We define consequently a K-linear form on Mκ (N ) ⊗I K by

l F×F := l F ⊗ l F

under the decomposition in (15.5).

Before defining the p-adic L-functions, we need an operator to lower the
level of the Eisenstein series constructed before. We follow [Hid88, §1 VI].
Fix a prime-to- p integer L, with N |L. We define for classical weights k:

TL/N ,k : Mk (L p, A) → Mk (N p, A)
 
 1 0 .
f → (L/N )k/2 [γ ]∈ (N )/ (N,L/N ) f |k |k γ
0 L/N

As L is prime to p, it is clear that TL/N,k commutes with U p . It extends

uniquely to a linear map

TL/N : Mκ (L) → Mκ (N )

which in weight k specializes to TL/N ,k .

≤α
We have a map Mκ (N )≤α → Mκ (N 2 R 2 ) . We define 1 N 2 R 2 /N to be one
left inverse. We define
 
 N0 −1 
L p (κ, κ ) = 2 2 l F ⊗ l F (U 2 2 ◦ 1 N 2 R 2 /N ) ⊗ TN 2 R 2 /N (H(κ, κ ))
N R N1 N /N1

(for α = 0),
 
N0
L ∗p (κ) = l F ⊗ l F (U −12 ∗
◦ 1 N 2 R 2 /N ) ⊗ TN 2 R 2 /N (H (κ)
N R2 N
2
1 N /N12

(for α < +∞).

394 Giovanni Rosso

We will see in the proof of the following theorem that it is independent of the
left inverse 1 N 2 R 2 /N which we have chosen.
We fix some notations. For a Dirichlet character η, we denote by η0 the
associated primitive character. Let λ p (κ) ∈ I be the U p -eigenvalue of F. We
say that (κ, κ  ) ∈ C F × W is of type (k; t, ε) if:

● κ|W(r) = [k] with k ≥ 2,

● κ  = ε[t] with 1 ≤ t ≤ k − 1 and ε finite order character defined modulo
p n , n ≥ 1.

Let as before s = k − t − 1. We define

E 1 (κ, κ  ) =λ p (κ)−2n 0 (1 − (χ εω−s )0 ( p)λ p (κ)−2 p s ))

where n 0 = 0 (resp. n 0 = n) if χ εω−s is (resp. is not) trivial at p.

If F(κ) is primitive at p we define E 2 (κ, κ  ) = 1, otherwise

E 2 (κ, κ  ) =(1 − (χ −1 ε−1 ωs φ)0 ( p) p k−2−s )×

(1 − (χ −1 ε−1 ωs φ 2 )0 ( p)λ p (κ)−2 p 2k−3−s ).

We denote by F ◦ (κ) the primitive form associated to F(κ). We shall write

W  (F(κ)) for the prime-to- p part of the root number of F ◦ (κ). If F(κ) is not
p-primitive we pose
  
φ0 ( p) p k−1 φ0 ( p) p k−2
S(F(κ)) = (−1)k 1 − 1 − ,
λ p (κ)2 λ p (κ)2

and S(F(κ)) = (−1)k otherwise. We pose

 
−1
Cκ,κ  = i 1−k s!G (χ εω−s ) (χ εω−s )0 ( p n−n 0 )(N1 Rpn 0 )s N −k/2 2−s ,
C κ = Cκ,[k−k0 +1] ,
K L
(F(κ), s) = W  (F(κ))(2πi )s+1 F ◦ (κ), F ◦ (κ) .

We have the following theorem, which will be proven at the end of the section.

Theorem 15.9. i) The function L p (κ, κ  ) is defined on C F × W, it is mero-

morphic in the first variable and bounded in the second variable. For all
classical points (κ, κ  ) of type (k; t, ε) with k ≥ 2, 1 ≤ t ≤ k − 1, we have
the following interpolation formula

L(s + 1, Sym2 (F(κ)), χ −1 ε−1 ωs )

L p (κ, κ  ) = Cκ,κ  E 1 (κ, κ  )E 2 (κ, κ  ) .
S(F(κ))(F(κ), s)
 Derivative of symmetric square p-adic L-functions 395

ii) The function L ∗p (κ) is meromorphic on C F . For κ of type k with k ≥ k0 , we

have the following interpolation formula

L(k0 − 1, Sym2 (F(κ)), χ  −1 χ1−1 )

L ∗p (κ) = Cκ E 2 (κ, [k − k0 + 1]) .
S(F(κ))(F(κ), k0 − 2)

Let us denote by L̃ p (κ, κ  ) the two-variable p-adic L-function of [Ros13b,

Theorem 4.14], which is constructed for any slope but NOT for p = 2. We
can deduce the fundamental corollary which allows us to apply the method of
Greenberg and Stevens;

Corollary 15.10. For α = 0 (resp. α > 0 and p = 2) we have the following

factorization of locally analytic functions around κ0 in C F :

L p (κ, [k − k0 + 1]) = (1 − χ  χ1 ( p)λ p (κ)−2 p k0 −2 )L ∗p (κ)

κ −1 (2) L̃ p (κ, [k − k0 + 1])
(resp. C = (1 − χ  χ1 ( p)λ p (κ)−2 p k0 −2 )L ∗p (κ)),
1 − φ −2 χ 2 κ(4)2−2k0
where C is a constant independent of κ, explicitly determined by the compari-
son of Cκ here and in [Ros13b, Theorem 4.14].

We can now prove the main theorem of the paper;

Proof of Theorem 15.2. Let f be as in the statement of the theorem; we take

χ  = χ1 = 1. For α = 0 resp. α > 0 we define

L p (Sym2 ( f ), s) = Cκ−1
0 ,[1]
L p (κ0 , [k0 − s]),
κ(2) L̃ p (κ, [k − k0 + 1])
L p (Sym2 ( f ), s) = C .
Cκ0 ,[1] (1 − φ −2 χ 2 κ(4)2−2k0 )

The two variables p-adic L-function vanishes on κ  = [1]. As f is Steinberg

at p, we have λ p (κ0 )2 = p k0 −2 .
Consequently, the following formula is a straightforward consequence of
Section 15.2 and Corollary 15.10;

L p (Sym2 ( f ), s) d log λ p (κ) L(Sym2 ( f ), k0 − 1)

lim = −2 |κ=κ0 .
s→0 s dκ S(F(κ))( f, k0 − 2)
From [Ben10, Mok12] we obtain
d log λ p (κ)
Lal (Sym2 ( f )) = − 2 |κ=κ0 .
dκ
396 Giovanni Rosso

Under the hypotheses of the theorem, f is Steinberg at all primes of bad

reduction and we see from [Ros13b, §3.3] that

L(Sym2 ( f ), k0 − 1) = L(Sym2 ( f ), k0 − 1)

and we are done.

Proof of Theorem 15.9. We point out that most of the calculations we need in
this proof and have not already been quoted can be found in [Hid90, Pan03].
If ε is not trivial at p, we shall write p n for the conductor of ε. If ε is trivial,
then we let n = 1.
We recall that s = k − t − 1; we have
I J
K L
N0 F(κ)c |τ N p , U −12 2 F(κ)c |τ N p , TN 2 R 2 /N ,k H([k], ε[t])
N /N1
L p (κ, κ  ) = .
N 2 R 2 N1 F(κ)c |τ, F(κ)2

We have as in [Pan03, (7.11)]

M N
F(k)c |τ N p , U p−2n+1 Prord U p2n−1 g (15.6)
M N
= λ p (κ)1−2n p (2n−1)(k−1) F(κ)c |τ N p |[ p 2n−1 ], g ,

where f |[ p 2n−1 ](z) = f ( p 2n−1 z). We recall the well-known formulae

[Hid88, page 79]:
M N k
M N
f |[ p 2n−1 ], TN 2 R 2 /N ,k g = (N R 2 ) f |[ p 2n N R 2 ], g ,
 −k/2
τ N p |[ p 2n−1 N 2 R 2 ] = N R 2 p2n−1 τ N 2 R 2 p2n ,
K L
F(κ)c |τ N p , F(κ)
= (−1)k W  (F(κ)) p (2−k)/2 λ p (κ)×
F(κ)◦ , F(κ)◦ 
  
φ0 ( p) p k−1 φ0 ( p) p k−2
1− 1 − .
λ p (κ)2 λ p (κ)2

So we are left to calculate

 
k/2 (2n−1) k2 −1
N0 (N R 2 ) λ p (κ)1−2n p
A(−1)s ct+1
s N N 2 R2
1
I M NJ
× F(κ)c |τ N p , U −12 2 F(κ)c |τ N 2 R 2 pn , H1,χ

εω −t (z, w)|s= −t .
1
N /N1 2
 Derivative of symmetric square p-adic L-functions 397

We use Proposition 15.4 and [BS00, (3.29)] (with the notation of loc. cit. β1 =
λ2p (κ)
p k−1
) to obtain that the interior Petersson product is a scalar multiple of

L(Sym2 ( f ), s + 1) Ẽ(κ, κ  )E 2 (κ, κ  )F(κ)|U N 2 /N 2

1
. (15.7)
F(κ)◦ , F(κ)◦  W  (F(κ)) p (2−k)/2 λ p (κ)S(F(κ))
Here

Ẽ(κ, κ  ) = − p −1 (1 − χ −1 ε−1 ωs ( p)λ p (κ)2 p t+1−k )

if χ εω−s is trivial modulo p, and 1 otherwise. The factor E 2 (κ, κ  ) appears

because F(κ) could not be primitive. Clearly it is independent of 1 N 2 R 2 /N and
we have l F (F(κ)) = 1. We explicit the constant which multiplies (15.7):
 
k/2 (2n−1) k −1
N0 (N R 2 ) λ p (κ)1−2n p 2
N1s+1 (Rp n )s+3−k N 2−k ωk−t−1 (−1)k,s (s1 ) ps1 (t + 1)
(−1) s 
−s
B(t)(2πi) ct+1 G (χ εω ) N 2 R 2 N1
s s ds1 (t + 1)
k
N − 2 N0 N1s R s+1 λ p (κ)1−2n p(2−k)/2 p n(s+1) (2s)!2−2s 22t π 3/2
= (−1)− 2
k
 .
21+2t π 2 (2πi)s G (χ εω−s ) 2−s (2s)!
5
s!

If ε1 εω−s is not trivial we obtain

 
k −1
N − 2 N1s R s λ p (κ)1−2n p (2−k)/2 p ns s!G (χ εω−s )
=i 1−k ,
(2πi)s+1 2s
otherwise
 
N − 2 N1s R s λ p (κ)−1 p (2−k)/2 ps+1 s!G (χ1 χ  )−1
k

= i 1−k .
(2πi)s+1 2s χ1 χ  ( p)
The calculations for L ∗p (κ) are similar. We have to calculate
k/2
N0 (N R 2 )
A∗ N 2 R 2 N1
I M NJ
∗
× F(κ)c |τ N p , U −12 F(κ) |τ N 2 R 2 p , (−1)
c k0
H 1,χ1 χ  (z, w)|u=s1 ,
N /N12

where s1 = 12 − k + k0 − 1. The interior Petersson product equals (see [BS00,

Theorem 3.1] with M, N of loc. cit. as follows: M = R 2 N 2 p, N = N1 R)
2−k
R k0 +1−k (N 2 p) 2 (−1)k0 k,k0 −2 (s1 ) ps1 (k − k0 + 2)
F(κ)|U p |U N 2 /N 2 ,
ds1 (k − k0 + 2) F(κ)◦ , F(κ)◦  W  (F(κ)) p (2−k)/2 λ p (κ)S(F(κ)) 1
398 Giovanni Rosso

so we have
k −2
N1 0 R k0 −2 N k/2 (k0 − 2)!
L ∗p (κ) =i 1−k
(2πi )k0 −1 G(χ1 χ  )2k0 −2
E 2 (κ, [k − k0 + 1])L(k0 − 1, Sym2 (F(κ)), (χ1 χ  )−1 )
× .
S(F(κ))W  (F(κ))F(κ)◦ , F(κ)◦ 
We give here some concluding remarks. As F(κ) has not complex mul-
tiplication by χ, we can see exactly as in [Hid90, Proposition 5.2] that
H F (κ)L p (κ, κ  ) is holomorphic along κ  = [0] (which is the pole of the
Kubota Leopoldt p-adic L-function).
We point out that the analytic L-invariant for C M forms has already been
studied in literature [DD97, Har12, HL13].
Note also that our choice of periods is not optimal [Ros13b, §6].

References
[AIP12] Fabrizio Andreatta, Adrian Iovita, and Vincent Pilloni. p-adic families of
Siegel modular cuspforms. to appear Ann. of Math., 2012.
[AIS12] Fabrizio Andreatta, Adrian Iovita, and Glenn Stevens. Overconvergent
modular sheaves and modular forms for GL2/F . to appear Israel J.
Math., 2012.
[BDR14] Massimo Bertolini, Henri Darmon, and Victor Rotger. Beilinson-Flach
elements and Euler systems II: the Birch and Swinnerton-Dyer conjecture
for Hasse-Weil-Artin L-functions. in preparation, 2014.
[Bel12] Joël Bellaïche. p-adic L-functions of critical CM forms. preprint avail-
able at http://people.brandeis.edu/ jbellaic/preprint/CML-functions4.pdf,
2012.
[Ben10] Denis Benois. Infinitesimal deformations and the -invariant. Doc. Math.,
(Extra volume: Andrei A. Suslin sixtieth birthday):5–31, 2010.
[Ben11] Denis Benois. A generalization of Greenberg’s L-invariant. Amer. J.
Math., 133(6):1573–1632, 2011.
[Ben12] Denis Benois. Trivial zeros of p-adic L-functions at near central
points. to appear J. Inst. Math. Jussieu http://dx.doi.org/10.1017/
S1474748013000261, 2012.
[Bra13] Riccardo Brasca. Eigenvarieties for cuspforms over PEL type Shimura
varieties with dense ordinary locus. in preparation, 2013.
[BS00] S. Böcherer and C.-G. Schmidt. p-adic measures attached to Siegel
modular forms. Ann. Inst. Fourier (Grenoble), 50(5):1375–1443, 2000.
[BSDGP96] Katia Barré-Sirieix, Guy Diaz, François Gramain, and Georges Philibert.
Une preuve de la conjecture de Mahler-Manin. Invent. Math., 124(1-3):1–
9, 1996.
[Buz07] Kevin Buzzard. Eigenvarieties. In L-functions and Galois representa-
tions, volume 320 of London Math. Soc. Lecture Note Ser., pages 59–120.
Cambridge University Press, Cambridge, 2007.
 Derivative of symmetric square p-adic L-functions 399

[Col05] Pierre Colmez. Zéros supplémentaires de fonctions L p-adiques de

formes modulaires. In Algebra and number theory, pages 193–210.
Hindustan Book Agency, Delhi, 2005.
[CP04] Michel Courtieu and Alexei Panchishkin. Non-Archimedean L-functions
and arithmetical Siegel modular forms, volume 1471 of Lecture Notes in
Mathematics. Springer-Verlag, Berlin, second edition, 2004.
[DD97] Andrzej Dabrowski and Daniel Delbourgo. S-adic L-functions attached
to the symmetric square of a newform. Proc. London Math. Soc. (3),
74(3):559–611, 1997.
[DG12] Mladen Dimitrov and Eknath Ghate. On classical weight one forms in
Hida families. J. Théor. Nombres Bordeaux, 24(3):669–690, 2012.
[Dim13] Mladen Dimitrov. On the local structure of ordinary hecke algebras at
classical weight one points. to appear in Automorphic Forms and Galois
Representations, proceedings of the LMS Symposium, 2011, to appear,
2013.
[Gor06] B. Gorsse. Mesures p-adiques associées aux carrés symétriques. 2006.
Thesis (Ph.D.)–Université Grenoble 1.
[Gre94] Ralph Greenberg. Trivial zeros of p-adic L-functions. In p-adic mon-
odromy and the Birch and Swinnerton-Dyer conjecture (Boston, MA,
1991), volume 165 of Contemp. Math., pages 149–174. Amer. Math.
Soc., Providence, RI, 1994.
[GS93] Ralph Greenberg and Glenn Stevens. p-adic L-functions and p-adic
periods of modular forms. Invent. Math., 111(2):407–447, 1993.
[Har12] Robert Harron. The exceptional zero conjecture for symmetric powers of
CM modular forms: the ordinary case. to appear Int. Math. Res. Notices
doi: 10.1093/imrn/rns161, 2012.
[Hid88] Haruzo Hida. A p-adic measure attached to the zeta functions associ-
ated with two elliptic modular forms. II. Ann. Inst. Fourier (Grenoble),
38(3):1–83, 1988.
[Hid89] Haruzo Hida. On nearly ordinary Hecke algebras for GL(2) over totally
real fields. In Algebraic number theory, volume 17 of Adv. Stud. Pure
Math., pages 139–169. Academic Press, Boston, MA, 1989.
[Hid90] Haruzo Hida. p-adic L-functions for base change lifts of GL2 to GL3 . In
Automorphic forms, Shimura varieties, and L-functions, Vol. II (Ann Ar-
bor, MI, 1988), volume 11 of Perspect. Math., pages 93–142. Academic
Press, Boston, MA, 1990.
[Hid06] Haruzo Hida. Hilbert modular forms and Iwasawa theory. Oxford Math-
ematical Monographs. The Clarendon Press Oxford University Press,
Oxford, 2006.
[HJ13] Robert Harron and Andrei Jorza. On symmetric power L-invariants
of Iwahori level Hilbert modular forms. preprint available at
www.its.caltech.edu/ ajorza/papers/harron-jorza-non-cm.pdf, 2013.
[HL13] Robert Harron and Antonio Lei. Iwasawa theory for symmetric powers
of CM modular forms at non-ordinary primes. preprint available at
http://arxiv.org/abs/1208.1278, 2013.
[HTU97] Haruzo Hida, Jacques Tilouine, and Eric Urban. Adjoint modular Ga-
lois representations and their Selmer groups. Proc. Nat. Acad. Sci.
400 Giovanni Rosso

U.S.A., 94(21):11121–11124, 1997. Elliptic curves and modular forms

(Washington, DC, 1996).
[Liu13] Ruochuan Liu. Triangulation of refined families. preprint available at
http://arxiv.org/abs/1202.2188, 2013.
[LLZ12] Antonio Lei, David Loeffler, and Sarah Livia Zerber. Euler systems for
Rankin-Selberg convolutions of modular forms. to appear Ann. of Math.,
2012.
[Mat80] Hideyuki Matsumura. Commutative algebra, volume 56 of Mathematics
Lecture Note Series. Benjamin/Cummings Publishing Co., Inc., Reading,
Mass., second edition, 1980.
[Mok09] Chung Pang Mok. The exceptional zero conjecture for Hilbert modular
forms. Compos. Math., 145(1):1–55, 2009.
[Mok12] Chung Pang Mok. L-invariant of the adjoint Galois representation of
modular forms of finite slope. J. Lond. Math. Soc. (2), 86(2):626–640,
2012.
[MTT86] B. Mazur, J. Tate, and J. Teitelbaum. On p-adic analogues of the
conjectures of Birch and Swinnerton-Dyer. Invent. Math., 84(1):1–48,
1986.
[Nag62] Masayoshi Nagata. Local rings. Interscience Tracts in Pure and Applied
Mathematics, No. 13. Interscience Publishers a division of John Wiley &
Sons New York-London, 1962.
[Pan94] Alexei A. Panchishkin. Motives over totally real fields and p-adic L-
functions. Ann. Inst. Fourier (Grenoble), 44(4):989–1023, 1994.
[Pan03] A. A. Panchishkin. Two variable p-adic L functions attached to eigen-
families of positive slope. Invent. Math., 154(3):551–615, 2003.
[Pil13] Vincent Pilloni. Overconvergent modular forms. Ann. Inst. Fourier
(Grenoble), 63(1):219–239, 2013.
[Pot13] Jonathan Pottharst. Cyclotomic Iwasawa theory of motives. preprint
available at http://vbrt.org/writings/cyc.pdf, 2013.
[PR95] Bernadette Perrin-Riou. Fonctions L p-adiques des représentations p-
adiques. Astérisque, (229):198, 1995.
[PR98] Bernadette Perrin-Riou. Zéros triviaux des fonctions L p-adiques, un cas
particulier. Compositio Math., 114(1):37–76, 1998.
[Ros13a] Giovanni Rosso. Derivative at s = 1 of the p-adic L-function of
the symmetric square of a Hilbert modular form. preprint available at
http://arxiv.org/abs/1306.4935, 2013.
[Ros13b] Giovanni Rosso. A formula for the derivative of the p-adic L-function
of the symmetric square of a finite slope modular form. preprint at
http://arxiv.org/abs/1310.6583, 2013.
[Urb06] Éric Urban. Groupes de Selmer et fonctions L p-adiques pour
les représentations modulaires adjointes. preprint available at
http://www.math.jussieu.fr/∼urban/eurp/ADJMC.pdf, 2006.
[Urb11] Eric Urban. Eigenvarieties for reductive groups. Ann. of Math. (2),
174(3):1685–1784, 2011.
[Urb13] Éric Urban. Nearly overconvergent modular forms. to appear in the
Proceedings of conference IWASAWA 2012, 2013.
 16
Uniform bounds for rational points on cubic
hypersurfaces
Per Salberger
Mathematics, Chalmers University of Technology,
SE-412 96 Göteborg, Sweden
E-mail address: [email protected]

A BSTRACT. We use a global version of Heath-Brown’s p-adic determinant

method to show that there are O N,ε (B dim X+1/7+ε ) rational points of height at
most B on a geometrically integral variety X ⊂ P N of degree three defined
over Q. By the same method we also show that there are Oε (B 12/7+ε ) rational
points of height at most B outside the lines on any cubic surface in P3 .

16.0 Introduction
We shall in this paper study the number N (W ; B) of rational points of height
at most B ≥ 1 on quasi-projective subvarieties W of P N defined over Q. The
height H (x) of a rational point x on W will be given by max(|x0 |, . . . , |x N |)
for a primitive integral N + 1-tuple (x0 , . . . , x N ) representing x. We will use
the O notation for functions defined for B ≥ 1. If g is a function from [1, ∞)
to [0, ∞), then we write N (W ; B) = O(g(B)) when there is a constant C > 0
such that N (W ; B) ≤ Cg(B) for all B ≥ 1. If the constant C depends on
parameters, then we include these as indices.
One of the goals of the paper is to establish the following uniform bound, in
which the implicit constant is independent of X .

Theorem 16.1. Let X ⊂ P3 be an integral cubic surface defined over Q and

U be the complement of the union of all lines on X . Then,
N (U ; B) = Oε (B 12/7+ε ).

The first general bound where the growth order on U is less than two is
due to Heath-Brown. He proved in [HB02] that N (U ; B) = Oε (B 52/27+ε ).
Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

401
402 Per Salberger

This was then√improved in [Sal08], where it was shown that N (U√; B) =

Oε (B 2/3+73/36 3+ε ). Finally, it was shown that N (U ; B) = Oε (B 3+ε ) in
[Sal]. The bound in 16.1 is superior to these bounds.
It has been conjectured by Manin that N (U ; B) = O X,ε (B 1+ε ), but this has
not been proved for any non-singular cubic surface with a rational point.
To prove theorem 16.1, we first establish the following lemma (cf. 16.12).

Lemma 16.2. Let X ⊂ P3 be an integral cubic surface over Q and B ≥ 1. Let

ε > 0 and p1 , . . . , pt be distinct primes with q = p1 . . . pt ≥ B 6/7+ε . For i =
1, . . . , t, let Pi be a non-singular F pi -point on X pi and X (Q; B; P1 , . . . , Pt )
be the set of rational points of height at most B on X , which specialize to Pi
on X pi for i = 1, . . . , t. Then there exists a surface Y ⊂ P3 not containing X
of degree bounded solely in terms of ε with X (Q; B; P1 , . . . , Pt ) ⊂ Y .

Here X p is the reduction (mod p) of Proj Z[x0 , . . . , x3 ]/(F(x0 , . . . , x3 ))

for a primitive cubic form F(x0 , . . . , x3 ) defining X ⊂ P3 . This lemma is
inspired by theorem√ 14 in [HB02], from which one gets the same conclusion
for primes q ≥ B 3/2+ε . To establish 16.2, we use a refined version of the p-
adic determinant method in [HB02], where the equations for the tangent planes
play a special role.
The basic idea to deduce 16.1 from 16.2 is to reduce to a counting problem
for curves for each residue class (mod q). But to obtain the estimate in 16.1,
it is important to apply 16.2 to many smooth integers q even if the number of
such q will be bounded in terms of ε. We use thereby a dichotomy (see (16.3.2)
and (16.3.3)) similar to the one used for the global determinant method in [Sal].
It seems at first sight that the method in this paper is cruder than the method
in [Sal], where one makes use of conguences modulo almost all primes. But
we have not been able to deduce 16.1 from 16.2 with the global method in
[Sal]. The present method which might be called “semiglobal” is easier to use
as the degrees of the auxiliary hypersurfaces do not depend on B. But there
are varieties where the method in [Sal] gives better results. It is thus likely that
both methods will be used in the future.
Another goal of this paper is to prove the following result, which improves
upon the previous estimates in [BHBS06] and [Sal].

Theorem 16.3. Let N ≥ 3 and f (y1 , . . . , y N ) ∈ Q[y1 , . . . , y N ] be a cubic

polynomial such that its leading cubic form is absolutely irreducible. Let
n( f ; B) be the number of N -tuples (y1 , . . . , y N ) of integers in [−B, B] such
that f (y1 , . . . , y N ) = 0. Then
n( f ; B) = O N ,ε (B N −2+1/7+ε ).
 Uniform bounds for rational points on cubic hypersurfaces 403

To prove this, we proceed as in [BHBS06] and reduce to the case N = 3 by

means of hyperplane sections. The proof is then similar to the proof of theorem
16.1. On applying theorem 1 in [BHBS06], we obtain the following corollary
of theorem 16.3.

Corollary 16.4. Let X ⊂ P N be a geometrically integral closed subvariety

over Q of degree three. Then N (X ; B) = O N ,ε (B dim X +1/7+ε ).

It is known (see [Sal]) that N (X ; B) = O X,ε (B dim X +ε ) for such varieties.

But the bound in 16.4 is better than previous uniform bounds for varieties of
degree three (cf. [BHBS06] and [Sal]).

16.1 Some consequences of Siegel’s lemma

We present in this section some consequences of Siegel’s lemma on bounds for
solutions to linear equations. These will be needed for the p-adic determinant
N
method to work well. In the sequel a hypersurface X in PQ will always mean
an equidimensional closed subscheme of codimension one. We shall in this
paper use the following notation.

Notation 16.5. Let X ⊂ P N be a hypersurface over Q and F(x0 , . . . , x N ) ∈

N +1
Q[x0 , . . . , x N ] be a form which defines X . Let B = (B0 , . . . , B N ) ∈ R≥0 .
Then,
(i) X (Q; B) is the set of rational points on X which may be represented by
an integral (N + 1)-tuple (x0 , . . . , x N ) with |x j | ≤ B j for 0 ≤ j ≤ N . If
B0 = · · · = B N = B, then we also denote this set by X (Q; B).
(ii) N (X ; B) = #X (Q; B) and N (X ; B) = #X (Q; B).
(iii) V = B0 . . . B N .
1 m m 2
(iv) T = max B0 0 . . . B N N with the maximum taken over all (N +1)-tuples
m
(m 0 , . . . , m N ) for which the corresponding monomial x0 0 . . . x Nm N occurs
in F(x0 , . . . , x N ) with non-zero coefficient.
(v) If m = (m 0 , . . . , m N ) is an (N + 1)-tuple of non-negative integers, then
m m
x m = x0 0 . . . x N N .
(vi) If X ⊂ PQ N is a hypersurface over Q defined by a primitive form

F(x0 , . . . , x N ) = am x m in Z[x0 , . . . , x N ], then X cl =
Proj Z[x0 , . . . , x N ]/(F(x0 , . . . , x N )) and H (X ) = max |am |.
X cl is thus the scheme-theoretic closure of X in PZN . We shall also write
X p = X cl × F p for primes p and make extensive use of the specialization map
from X cl (Z) = X (Q) to X p (F p ).
404 Per Salberger

Lemma 16.6. Let X ⊂ PQ

N
be an integral hypersurface of degree D and B =
N +1
(B0 , . . . , B N ) ∈ R≥1 . Then one of the following two statements holds.

(a) There exists a projective Q-hypersurface Y ⊂ P N of degree D disinct from

X which contains X (Q; B).
(b) H (X ) = O D,N (V (D+N )!/(D−1)!N ! ).

Proof. This is proved in [HB02, th.4] for N = 2 and the same argument can
be used to prove 16.6 for N > 2.

We shall use lemma 16.6 together with the following elementary result.

N +1
Lemma 16.7. Let B = (B0 , . . . , B N ) ∈ R≥1 and X ⊂ PQ N be a hypersurface

of degree D and of height H (X ) = O D,N (V (D+N )!/(D−1)!N ! ). Let x be a non-

singular point in X (Q; B) and πx be the product of all primes p where x

specializes to a singular F p -point on X p . Then,

log πx = O D,N (1 + log V ).


Proof. Let F(x0 , . . . , x N ) = am x m be a primitive form defining X ⊂ P N
and ξ be an integral primitive (N + 1)-tuple representing x ∈ X (Q; B).
Suppose that (δ F/δx j )(ξ ) = 0 for some j. Then, πx | (δ F/δx j )(ξ ) and
log πx ≤ log |(δ F/δx j )(ξ )| = O D,N (1 + log V ).

We shall also need a generalization of 16.6 and 16.7 to arbitrary integral

N +1
closed subschemes Z ⊂ PQ N
. If B = (B0 , . . . , B N ) ∈ R≥0 , then Z(Q; B)
will denote the set of rational points on X , which may be represented by an
integral (N + 1)-tuple (x0 , . . . , x N ) with |x j | ≤ B j for 1 ≤ j ≤ N and Z cl
will denote the scheme-theoretic closure of Z in PZN .

Lemma 16.8. Let Z ⊂ PQ

N be an integral subscheme of degree ≤ D and
N +1
B = (B0 , . . . , B N ) ∈ R≥1 . Let V = B0 . . . B N . Then one of the following
two statements holds.

(a) There exists a hypersurface Y ⊂ PQ N

of degree O D,N (1) which contains
Z (Q; B) but not Z .
(b) If z is a non-singular point in Z(Q; B), then the product π z of all primes
p where z specializes to a singular F p -point on Z p = Z cl × F p satisfies
log π z = O D,N (1 + log V ).

Proof. It follows from lemma 16.7 and lemma 16.8 in [Sal07] that Z ⊂ PQ
N is

defined as a subscheme by homogeneous polynomials of degree δ = O D,N (1).

 Uniform bounds for rational points on cubic hypersurfaces 405

We may therefore apply lemma 5 in [Bro04] and conclude that either (a)
holds or that we may find a set of t = O D,N (1) homogeneous polynomials
F1 , . . . , Ft of degree at most δ with integer coefficients, which define Z ⊂ PQ
N
t +δ
as a subscheme and with i =1 Fi  = O D,N (V ) for g = δ ( δ ). It is now
g 2 N

easy to see from the Jacobian criterion that (b) holds for schemes defined by
such polynomials F1 , . . . , Ft .

16.2 The p-adic determinant method for cubic surfaces

We shall in this section describe an improvement of Heath-Brown’s p-adic
determinant method for the special case of cubic surfaces. We will keep the
notation of 16.5, but only consider the case where X is an integral cubic surface
over Q. If I is a prime ideal in a commutative ring R, then we let R I be the
localization of R at I . In particular, if R = Z and I = ( p), then Z( p) =
{m/n ∈ Q : m, n ∈ Z such that n/ p ∈ / Z}.
The following lemma should be viewed as a refinement of lemma 16.8 in
[Sal07].

Lemma 16.9. Let R be a commutative noetherian regular local ring contain-

ing Z( p) as subring. Let φ1 , . . . , φs be ring homomorphisms from R to Z( p) ,
I be the kernel of φ1 and m = I R I be the maximal ideal of A = R I . Let
W ⊂ A be the vector subspace over Q generated by elements r1 , . . . , rs of

R and f = k≥1 dimQ (W ∩ mk ). Then the determinant of the s × s-matrix
(φi (r j )) is divisible by p f .

Proof. Let I be the kernel of φ1 and I 0 = R. Then I is generated by dim R −1

regular parameters (see [AK70, p.56]). Since these form a regular sequence on
R (see [Eis95, p.243]), we conclude that Symk (I /I 2 ) = I k /I k+1 is a free
R/I -module for each k ≥ 1 (see [FL85, p.75]). So if ur ∈ I k+1 for u ∈ S and
r ∈ I k , then r ∈ I k+1 as (u + I )(r + I k+1 ) = I k+1 . Therefore, mk+1 ∩ I k =
I k+1 R I ∩ I k = I k+1 for k ≥ 0.
Now suppose there are more than dimQ (W ∩ mk )/(W ∩ mk+1 ) ele-
ments among r1 , . . . , rs in I k \ I k+1 for some k ≥ 0. If we denote these
by r1 , . . . , rq , then we may find β1 , . . . , βq ∈ Q, not all 0, such that
β1r1 + · · · + βq−1 rq−1 + βq rq ∈ mk+1 . After scaling and reordering we
may assume that βq = 1 and β1 , . . . , βq−1 ∈ Z( p) . Then ρq = β1 r1 +
. . . βq−1 rq−1 + rq ∈ mk+1 ∩ I k = I k+1 . We may therefore by perform-
ing elementary tranformations of this kind replace r1 , . . . , rs by new elements
ρ1 , . . . , ρs in R such that det(φi (ρ j )) = det(φi (r j )) and such that we have
at most dimQ (W ∩ mk ) − dimQ (W ∩ mk+1 ) elements ρ j in I k \ I k+1 for
406 Per Salberger

each k ≥ 0. But then we have at least dimQ (W ∩ mk ) elements ρ j in I k for

each k ≥ 1. As p k | φi (ρ j ) for i = 1, . . . , s, for each such ρ j , we find that
p f | det(φi (ρ j )), as was to be proved.

We now apply this result to a cubic surface X ⊂ P3Q and let Q(n) be its
   
n+3 n
Hilbert polynomial. Then, Q(n) = − = (3n 2 + 3n + 2)/2. We
3 3
shall also consider the integer e(n) given by

e(n) = Q(n) − (k 2 + k)/2 + Q(2n − k). (16.1)
1≤k≤3n/2 3n/2<k≤2n

Lemma 16.10. Let X ⊂ P3 be an integral cubic surface over Q and

F1 , . . . , FQ(n) ∈ Z[x0 , . . . , x3 ] be homogeneous polynomials of degree n ≥ 0.
Let ξi = (ξi,0 , . . . , ξi,3 ), 1 ≤ i ≤ Q(n) be primitive quadruples of integers
such that all ξi (mod p) represent the same non-singular F p -point P on X p .
Then the determinant of the Q(n) × Q(n)-matrix (F j (ξi )) is divisible by p e(n) .

Proof. We may suppose that F1 , . . . , FQ(n) are linearly independent on X ,

since otherwise det(F j (ξi )) = 0. Let R be the stalk of the structure sheaf of
X cl at P and assume without loss of generality that x0 (P) = 0. Then r j =
F j /x0n ∈ R for j = 1, . . . , Q(n). Let φi : R → Z( p) , 1 ≤ i ≤ Q(n) be the
local ring homomorphisms at P corresponding to the Z( p) -points on X cl given
by ξi for 1 ≤ i ≤ Q(n). Then, φi (r j ) = F j (ξi )/ξi,0
n and


Q(n)
det(F j (ξi )) = det(φi (r j )) ξi,0
n
. (16.2)
i=1

Let I be the kernel of φ1 . Then A = R I is the stalk of the structure sheaf of X

at the non-singular rational point x defined by ξ1 . It is thus a two-dimensional
regular local ring with m = I R I as maximal ideal and dimQ (A/mk ) = 1 +
· · · + k = (k 2 + k)/2 for k ≥ 1.
Let W be the Q-subspace of A generated by r1 , . . . , r Q(n) . Then
dimQ W = Q(n) as F1 , . . . , FQ(n) are linearly independent on X . Further,
dimQ (W/(W ∩ mk )) ≤ dimQ (A/mk ). Hence,

dimQ (W ∩ mk ) = dimQ (W ) − dimQ (W/(W ∩ mk )) ≥ Q(n) − (k 2 + k)/2.

(16.3)
We shall use (16.3) for 1 ≤ k ≤ 3n/2. For 3n/2 < k ≤ 2n, we will
use another lower estimate for dimQ (W ∩ mk ). Let G 1 , . . . , G Q(4n−2k) be
homogeneous polynomials in Q[x0 , . . . , x3 ] of degree 4n − 2k ≥ 0, which
 Uniform bounds for rational points on cubic hypersurfaces 407

are linearly independent on X and W0 be the Q-subspace of A generated by

G j /x04n−2k , 1 ≤ j ≤ Q(4n − 2k). Then, dimQ (W0 ∩ m6n−3k ) ≥ dimQ (W0 ) −
dimQ (A/m6n−3k ) and

dimQ (W0 ∩m6n−3k ) ≥ Q(4n −2k)−((6n −3k)2 +(6n −3k))/2 = Q(2n −k).
(16.4)

There are thus linearly independent Q-linear combinations H1 , . . . , H Q(2n−k)

of G 1 , . . . , G Q(4n−2k) such that H j /x04n−2k ∈ W0 ∩ m6n−3k for 1 ≤ j ≤
Q(2n − k). If now L(x0 , . . . , x3 ) = 0 is a linear equation of the tangent plane
of X of ξ1 , we deduce that L 2k−3n H j /x0n ∈ mk for 1 ≤ j ≤ Q(2n − k) as
L/x0 ∈ m2 .
Let +1 , . . . , + Q(2n−k) be linear combinations of F1 , . . . , FQ(n) such
that + j − L 2k−3n H j vanish on X for 1 ≤ j ≤ Q(2n − k). Then,
+1 /x0n , . . . , + Q(2n−k) /x0n are linearly independent elements of W ∩ mk and

dimQ (W ∩ mk ) ≥ Q(2n − k) for k ∈ (3n/2, 2n]. (16.5)

The assertion now follows from lemma 16.9, (16.2), (16.3) and (16.5).

Lemma 16.11. e(n) = 7n 3 /4 + O(n 2 ).

4k
Proof. Q(n) − (k 2 + k)/2 = 2 (3n
1 2 − x 2 )d x + O(n) for integers k ∈
k−1
4
3n/2 (3n+1)/2
4
[1, 3n/2], while Q(2n−k) = 2 (3n −x )d x +
1 2 2
2 (2n−x) d x +
3 2
(3n−1)/2 3n/2
4k
O(n) for k = (3n +1)/2 if n is odd and Q(2n −k) = 2 (2n −x) d x + O(n)
3 2
k−1
for integers k ∈ [3n/2 + 1, 2n].
Hence, e(n)=I (n)+O(n 2 ) for

.
3n/2 .2n
I (n)= 2 (3n −x )d x
1 2 2
+ 2 (2n
3
− x)2 d x = 7n 3 /4.
0 3n/2

Remark. The arguments in (16.5) and 16.11 can be generalized to inte-

N −1 . Let Q(n) =
X ⊂ P of degree 2 ≤ D < 2
gral N
 hypersurfaces
 
n+N n+N−D
N − N be the Hilbert polynomial of X ⊂ P N , F1 , . . . , FQ(n)
be homogeneous polynomials in Z[x0 , . . . , x N ] of degree n ≥ 0 and ξi =
(ξi,0 , . . . , ξi,N ), 1 ≤ i ≤ Q(n) be primitive (N + 1)-tuples of integers such
that all ξi (mod p) represent the same non-singular F p -point P on X p . Then
408 Per Salberger

the determinant of the Q(n) × Q(n)-matrix (F j (ξi )) is divisible by p e(n) for

an integer e(n) = I (n) + O D,N (n N −1 ) where g(D, N ) = (D/2)1/(N −2) and

. )n
g(D,N
Dn N −1 − x N −1
I (n) = dx
(N − 1)!
0
.2n    N −1
Dn N −1 − (g(D, N )n) N −1 2n − x
+ dx
(N − 1)! 2n − g(D, N )n
g(D,N )n
 
2D D(D/2)1/(N −2)  2 N
= + 1− n
N! (N − 1)! N
> D N /(N −1) n N /(N − 2)!N for 2 ≤ D < 2 N −1 .

We have here used the inequality of arithmetic and geometric means in the
last line. The assertion that p e(n) | det(F j (ξi )) is stronger than (3.11) in
[HB02], which says p f (n) | det(F j (ξi )) for f (n) = D N /(N −1) n N /(N −2)!N +
O D,N (n N −1 ). If X is a quadric, then e(n) = 2n N /(N − 1)! + O N (n N −1 ).
We now introduce some new notation, which will be needed in the sequel.

Notation. Let X ⊂ P N be a closed subscheme over Q and B =

N +1
(B0 , . . . , B N ) ∈ R≥1 . Let p1 , . . . , pt be distinct primes and Pi be an F pi -
point on X pi for 1 ≤ i ≤ t. Then X (Q; B; P1 , . . . , Pt ) is the set of points in
X (Q; B), which specialize to Pi on X pi for 1 ≤ i ≤ t. If B0 = · · · = B N = B,
then we also write X (Q; B; P1 , . . . , Pt ) for this set.

Theorem 16.12. Let X ⊂ P3 be an integral cubic surface over Q and

B = (B0 , . . . , B3 ) ∈ R4≥1 . Let ε > 0 and p1 , . . . , pt be distinct primes
with q := p1 . . . pt ≥ (V /T 1/3 )2/7 V ε (cf. 16.5). For i = 1, . . . , t,
let Pi be a non-singular F pi -point on X pi . Then there exists a surface
Y (B; P1 , . . . , Pt ) ⊂ P3 not containing X of degree bounded solely in terms
of ε with X (Q; B; P1 , . . . , Pt ) ⊂ Y (B; P1 , . . . , Pt ).

Proof. Suppose first that q > 2V 1/4 . We will then show that X (Q; B;
P1 , . . . , Pt ) is contained in a plane. To see this, let ξk = (ξk,0 , . . . , ξk,3 ),
1 ≤ k ≤ 4 be primitive quadruples of integers representing points in
X (Q; B; P1 , . . . , Pt ). Then, by [HB02, lemma 6] we have that pi4 | det(ξk, j )
for i = 1, . . . , t. As | det(ξk, j ))| ≤ 16V by Hadamard’s inequality, we
conclude that det(ξk, j ) = 0 for q > 2V 1/4 . Hence, any four points in
X (Q; B; P1 , . . . , Pt ) are coplanar and X (Q; B; P1 , . . . , Pt ) contained in a
 Uniform bounds for rational points on cubic hypersurfaces 409

plane when q > 2V 1/4 . We may and shall thus in the sequel suppose that
q ≤ 2V 1/4 . We will also assume that V 1/4 ≥ 2 since the assertion is trivial
when V 1/4 < 2.
Suppose that X ⊂ P3 is defined by F(x0 , . . . , x3 ). There is then by the
remark after (5) in [Bro04] a graded monomial ordering of the monomials in
(x0 , . . . , x3 ) such that +(B) = T for the leading monomial +(x0 , . . . , x3 ) =
m m
x0 0 . . . x3 3 occurring in F(x0 , . . . , x3 ) with non-zero coefficient. We also note
that the Q(n) monomials F1 , . . . , FQ(n) ∈ Z[x0 , . . . , x3 ] of degree n not
divisible by + form a basis for the summand of degree n of the homogeneous
coordinate ring of X (see [Bro04, p.163]).  
n+3
Now let In (x0 , . . . , x 3 ) be the product of all monomials of degree
3
 
n
n in (x0 , . . . , x3 ) and Jn (x0 , . . . , x3 ) be the product of all monomials of
3
degreen in (x0 , . . . , x3 ) divisible 
by +. Then F1 . . . FQ(n) = In / Jn , Jn =
n k +3
In−3 + 3
and Ik4 = (x0 x1 x 2 x3 ) 3
for k = n − 3 and n. Hence

log(F1 (B) . . . FQ(n) (B)) (16.6)

     
n+3  n  n
= log V n/4 − log V (n−3)/4 − log(T ),
3 3 3

where the right-hand side is equal to 12 n 3 log(V /T 1/3 ) + 12 n log V + 16 (3n 2 −

2n) log T . Hence as 1 ≤ T ≤ V 3 for B ∈ R4≥1 , we deduce from (16.6) that

log(F1 (B) . . . FQ(n) (B)) ≤ (n 3 /2) log(V /T 1/3 ) + 32 n 2 log V + 12 n log V .

(16.7)
Now fix ε > 0 and assume that ξi = (ξi,0 , . . . , ξi,3 ), 1 ≤ i ≤ Q(n) are
primitive quadruples of integers representing points in X (Q; B; P1 , . . . , Pt ).
Then,
| det(F j (ξi ))| ≤ Q(n) Q(n)/2 F1 (B) . . . FQ(n) (B)

by the inequality of Hadamard. Hence, if det(F j (ξi )) = 0, we obtain from

(16.7) that:

log | det(F j (ξi ))| ≤ (n 3 /2) log(V /T 1/3 ) + (Q(n)/2) log Q(n) + 2n 2 log V .
(16.8)

We now give a lower estimate for log | det(F j (ξi ))|. By the previous two
lemmas det(F j (ξi )) is divisible by q e(n) for a non-negative integer e(n) =
410 Per Salberger

7n 3 /4 + O(n 2 ). We have thus if det(F j (ξi )) = 0 and (V /T 1/3 )2/7 V ε ≤ q ≤

2V 1/4 that

log | det(F j (ξi ))| ≥ e(n) log q ≥ (n 3 /2) log(V /T 1/3 ) (16.9)
ε
+ (7n /4) log V − C1 (n log V ).
3 2

for some absolute constant C1 > 0.

Therefore, by (16.8) and (16.9) we will have that det(F j (ξi )) = 0 if

(7n 3 /4) log V ε > (Q(n)/2) log Q(n) + (C1 + 2)n 2 log V . (16.10)

But if V 1/4 ≥ 2, then it is easy to see that there exists an integer n 0 depend-
ing only on ε > 0 such that (16.10) holds for all n ≥ n 0 . There is thus for such
n a non-trivial linear combination λ1 F1 + · · · + λ Q(n) FQ(n) which vanishes at
X (Q; B; P1 , . . . , Pt ). This completes the proof.

16.3 The semiglobal determinant method for cubic surfaces

We now come to the main lemma of this paper in which we apply theo-
rem 16.12 to several smooth square-free integers q simultaneously.

Main lemma 16.3.1. Let X ⊂ P3 be an integral cubic surface over Q. Let

B = (B0 , . . . , B3 ) ∈ R4≥1 be a quadruple with V 1/4 ≥ 2 for which X is the
only cubic surface containing X (Q, B). Then there exist for each ε > 0 a set 
of primes depending only on B and ε, and a set of surfaces Y (B; P1 , . . . , Pt ) ⊂
P3 indexed by sequences (P1 , . . . , Pt ) of non-singular F pi -points Pi on X pi for
sequences of primes p1 < · · · < pt in  with p1 . . . pt ≥ (V /T 1/3 )2/7 V ε such
that the following properties hold.

(a) If p ∈ , then V ε ≤ p ≤ C0 V ε for some constant C0 depending only on

ε.
(b) # is bounded solely in terms of ε.
(c) X (Q; B; P1 , . . . , Pt ) ⊂ Y (B; P1 , . . . , Pt ) for any sequence (P1 , . . . , Pt )
as above and all surfaces Y (B; P1 , . . . , Pt ) have the same degree d
bounded solely in terms of ε.
(d) Let x be a non-singular point in X (Q; B). Then one of the following
alternatives holds.

(16.3.2) Case (i): There exist primes p0 < · · · < pt in  with p0 . . . pt =

Oε ((V /T 1/3 )2/7 V 3ε ) and p0 . . . pt−1 ≥ (V /T 1/3 )2/7 V ε such that x
specializes to a non-singular F pi -point Pi on X for 0 ≤ i ≤ t and
 Uniform bounds for rational points on cubic hypersurfaces 411

where x lies on a component of X ∩Y (B; P0 , . . . , Pt−1 ) not contained

in Y (B; P1 , . . . , Pt ).
(16.3.2) Case (ii): There exist primes p0 < · · · < pt in  with p0 . . . pt =
Oε ((V /T 1/3 )2/7 V 2ε ) such that x specializes to a non-singular F pi -
point Pi on X for 0 ≤ i ≤ t and where for some i ∈ {0, . . . , t}
with p0 . . . p̂i . . . pt ≥ (V /T 1/3 )2/7 V ε x lies on a component of X ∩
Y (B; P0 , . . . , Pt ) not contained in Y (B; P0 , . . . , P̂i , . . . Pt ). (Here p̂i
means that pi is omitted and P̂i that Pi is omitted.)
(16.3.3) There are primes p1 < · · · < pt in  with p1 . . . pt =
Oε ((V /T 1/3 )2/7 V 2ε ) and p1 . . . pt ≥ (V /T 1/3 )2/7 V ε such that x
specializes to a non-singular F pi -point Pi on X for 1 ≤ i ≤ t
and a component Z of X ∩ Y (B; P1 , . . . , Pt ) containing x where
#Z (Q; B) = Oε (1), x is singular on Z or x specializes to a
non-singular F pi -point Pi on Z pi for each i ∈ {1, . . . , t}.

Proof. As X is the only cubic surface containing X (Q; B) and V 1/4 ≥ 2, we

have by 16.6 and 16.7 that log πx = O(log V ) for any non-singular point x in
X (Q; B). There is thus in this case an absolute constant k such that πx ≤ V k
for any non-singular point x in X (Q; B).
Now consider sequences p1 < · · · < pt of primes with p1 . . . pt ≥
(V /T 1/3 )2/7 V ε and choose a surface Y (B; P1 , . . . , Pt ) ⊂ P3 not containing
X as in theorem 16.12 for each sequence (P1 , . . . , Pt ) of non-singular F pi -
points on X pi . Then all Y (B; P1 , . . . , Pt ) are of the same degree d = Oε (1)
and the components Z of X ∩ Y (B; P1 , . . . , Pt ) of degree at most D = 3d
by the theorem of Bezout in [Ful84, 8.4.6]. By lemma 16.8 there are two
alternatives for these curves. If 16.8(a) holds, then #Z (Q; B) = Od (1) by
[Ful84, 8.4.6] and hence #Z(Q; B) bounded in terms of ε. If 16.8(b) holds,
then log π z = Od (log V ) for any non-singular point z in Z (Q; B). There is
thus for such curves Z a constant l depending only on d such that π z ≤ V l for
any non-singular point z in Z (Q; B).
Let m be the smallest positive integer such that p1 . . . pm ≥ V k+l
(V /T 1/3 )2/7 V ε for the m smallest primes with V ε ≤ p1 < · · · < pm . Let
 = { p1 , . . . , pm }. Then,

p ≥ V ε for all p ∈ . (16.11)


k+l 1/3 2/7 ε
p≥V (V /T ) V . (16.12)
p∈

We next note that p1 p2 . . . pm−1 < V k+l (V /T 1/3 )2/7 V ε ≤ V k+l+2/7 V ε by

the minimality of m. This is also true for m = 1 if we put p1 p2 . . . pm−1 = 1.
412 Per Salberger

Hence by (16.11) we get that V (m−2)ε < V k+l+2/7 and

# < 2 + (k + l + 2/7)/ε. (16.13)
It also follows from (16.11) and Bertrand’s postulate that
p ≤ 2# V ε for all p ∈ . (16.14)
Hence from (16.11), (16.13) and (16.14) we obtain that (a) and (b)
hold, while (c) follows from the definition of Y (B; P1 , . . . , Pt ) for sequences
p1 < · · · < pt in  with p1 . . . pt ≥ (V /T 1/3 )2/7 V ε .
To prove (d), let x be a non-singular point in X (Q; B) and (x) ⊂  be
the subset of all primes p where x specializes to a non-singular F p -point on
X p . We shall also for square-free products q = p1 . . . pt ≥ (V /T 1/3 )2/7 V ε
of primes in (x), write Y (x, q) = Y (B; P1 , . . . , Pt ) where Pi is the
specialization of x in X pi .
From (16.12) and the bound πx ≤ V k , we obtain :

p ≥ V l (V /T 1/3 )2/7 V ε . (16.15)
p∈(x)

Now list the primes in (x) in increasing order and stop as soon as
p1 . . . pt ≥ (V /T 1/3 )2/7 V ε . Then p1 . . . pt−1 < (V /T 1/3 )2/7 V ε , with
p1 . . . pt−1 = 1 for t = 1. Let q0 be the product p1 . . . pt of these primes
and Z be a component of X ∩ Y (x, q0 ) containing x. Further, let (x; Z ) be
the subset of primes p in (x) where x specializes to a non-singular F p -point
on Z p . On applying (16.15) and the bound π x ≤ V l , we obtain the following
inequality:

p ≥ (V /T 1/3 )2/7 V ε . (16.16)
p∈(x;Z )

We now define a finite sequence (q0 , . . . , qm ) of square-free integers recur-

sively as follows. If Z is not a component of X ∩ Y (x, qi ), then we stop. If
all prime factors of qi are in (x; Z) or if all primes in (x; Z ) divide qi ,
then we also stop. Otherwise, we replace the largest prime factor p of qi not
in (x; Z ) by the smallest prime π ∈ (x; Z), which is not a prime factor
of qi and let qi+1 = qi π/ p. Then qi < qi +1 for all i as the prime factors
of q0 are smaller than the other primes in (x). Further, qi+1 ≤ C0 qi by
(a). Hence, as q0 < (V /T 1/3 )2/7 V ε pt ≤ C0 (V /T 1/3 )2/7 V 2ε , we get that
lcm(qi−1 , qi ) < C0i+1 (V /T 1/3 )2/7 V 3ε and qi < C0i+1 (V /T 1/3 )2/7 V 2ε for all
i ≥ 1.
This process will finish after at most t ≤ # steps and produce an integer
qm with (V /T 1/3 )2/7 V 2ε ≤ qm < C0t+1 (V /T 1/3 )2/7 V 2ε satisfying one of the
following alternatives.
 Uniform bounds for rational points on cubic hypersurfaces 413

(16.3.4) m ≥ 1 and Z is a component of X ∩ Y (x, qm−1 ), but not of

X ∩ Y (x, qm ).
(16.3.5) Z is a component of X ∩Y (x, qm ) and all factors of qm are in (x; Z ).
(16.3.6) Z is a component of X ∩ Y (x, qm ) and qm is divisible by all primes
in (x; Z ).

Then, as lcm(qm−1 , qm ) < C0t+1 (V /T 1/3 )2/7 V 3ε all the conditions in the
first case of (16.3.2) will hold if (16.3.4) holds. We also obtain that the
conditions in (16.3.3) hold if (16.3.5) holds.
If (16.3.6) holds, then we define a new sequence (qm , . . . , qm+n ) of square-
free integers starting with qm . If Z is not a component of X ∩ Y (x, qm+ j )
or if all factors of qm+ j are in (x; Z ), then we stop. Otherwise, we let
qm+ j+1 = qm+ j / p for the largest prime factor p of qm+ j , which is not
in (x; Z ). This process will finish after less than t steps and all integers
in the sequence will be divisible by all primes in (x; Z). By (16.16) we
have thus that (V /T 1/3 )2/7 V ε ≤ qm+n . The last integer qm+n will also
satisfy qm+n ≤ qm < C0t+1 (V /T 1/3 )2/7 V 2ε and one of the following
conditions:

(16.3.7) n ≥ 1 and Z is a component of X ∩ Y (x, qm+n−1 ), but not of

X ∩ Y (x, qm+n ).
(16.3.8) Z is a component of X ∩ Y (x, qm+n ) and all factors of qm+n are in
(x; Z ).

If (16.3.7) holds, then all the conditions in the second case of (16.3.2)
hold. If instead (16.3.8) holds, then the conditions in (16.3.3) will hold. This
completes the proof.

Remark. Let (P0 , . . . , Pt ) be a sequence where Pi is a non-singular F pi -

point on X pi for 0 ≤ i ≤ t. It will be important to count the number of
non-singular points x ∈ X (Q; B), which specialize to (P0 , P1 , . . . , Pt ) and
satisfy (16.3.2). In case (i), any such point x belongs to Y (B; P1 , . . . , Pt ) (see
16.3.1(c)) and to a component of X ∩ Y (B; P0 , . . . , Pt−1 ) not contained in
Y (B; P1 , . . . , Pt ). There are thus by the theorem of Bezout [Ful84, 8.4.6] at
most 3d 2 non-singular points in X (Q; B) which specialize to (P0 , P1 , . . . , Pt )
and satisfy the condition in case (i). Similarily, we get that there are at most
3d 2 non-singular points in X (Q; B), which specialize to (P0 , P1 , . . . , Pt )
and satisfy the condition in case (ii) with x ∈ Y (B; P0 , . . . , P̂i , . . . , Pt ) for
a given i∈{0, 1, . . . , t}. There are thus altogether at most 3(t + 2)d 2 points
x∈X (Q; B), which specialize to (P0 , P1 , . . . , Pt ) and satisify (16.3.2).
414 Per Salberger

Lemma 16.13. Let Z ⊂ P3 be an integral curve over Q of degree δ. Let ε > 0

and p1 , . . . , pt be distinct primes with p1 . . . pt ≥ B 2/δ V ε . Let (P1 , . . . , Pt )
be a sequence where Pi is a non-singular F pi -point on X pi for 1 ≤ i ≤ t.
Then there exists a surface W ⊂ P3 of degree Oδ,ε (1) not containing Z with
Z (Q; B; P1 , . . . , Pt ) ⊂ W . In particular, N (Z ; B; P1 , . . . , Pt ) = Oδ,ε (1).

Proof. This is a special case of theorem 3.2 in [Sal07]. To obtain the second
assertion from the first, use the theorem of Bezout in [Ful84, 8.4.6].

Lemma 16.14. Let Z ⊂ P3 = Proj Q[x0 , . . . , x3 ] be an integral curve over

Q of degree δ, which is not contained in the hyperplane H0 ⊂ P3 defined
by x0 = 0. Let ε > 0, B = (1, B, B, B) and p1 , . . . , pt be distinct primes
with p1 . . . pt ≥ B 1/δ V ε . Let (P1 , . . . , Pt ) be a sequence where Pi is a non-
singular F pi -point on X pi for 1 ≤ i ≤ t. Then there exists a surface W ⊂ P3 of
degree Oδ,ε (1) not containing Z with Z (Q; B; P1 , . . . , Pt ) ⊂ W . In particular,
N (Z ; B; P1 , . . . , Pt ) = Oδ,ε (1).

Proof. This is a special case of corollary 3.7 in [Sal07]. To obtain the second
assertion from the first, use the theorem of Bezout in [Ful84, 8.4.6].

16.4 Integral points on affine cubic hypersurfaces

We shall in this section prove theorem 16.3 for affine hypersurfaces.

Lemma 16.15. Let X ⊂ P3 be a cubic surface over some field. Then the sum
of the degrees of the components of the singular locus of X is at most 16.

Proof. This follows from the theorem of Bezout [Ful84, 8.4.6] applied to the
system of quadratic equations given by the gradient of a cubic homogeneous
polynomial defining X ⊂ P3 .
We may now derive the following corollary of 16.3.1 and 16.14. We shall
thereby use the trivial uniform estimate #X p (F p ) = O( p 2 ) for cubic surfaces
X over Q, when we sum over tuples (P1 , . . . , Pt ) of non-singular F pi -points
Pi on X pi .

Lemma 16.16. Let X ⊂ P3 = Proj Q[x0 , . . . , x3 ] be an integral cubic surface

over Q such that the plane defined by x0 = 0 intersects X properly. Let B ≥ 1
and B = (1, B, B, B). Then there exists a set of Oε (B 8/7+ε ) rational lines
 Uniform bounds for rational points on cubic hypersurfaces 415

l ⊂ X such that all but Oε (B 8/7+ε ) points in X (Q; B) lie on one of these
lines.

Proof. Suppose first that there is more than one cubic surface containing
X (Q; B). Then X (Q; B) is contained in the union of at most nine inte-
gral curves C of degree at most nine by [Ful84, 8.4.6]. As #C(Q; B) =
Oδ,ε (B 2/δ+ε ) for any integral curve C ⊂ P3 of degree δ defined over Q (see
[HB02, th.5]), there is thus a set of at most nine rational lines l ⊂ X such that
all but Oε (B 1+ε ) points in X (Q; B) lie on one of these lines in this case. We
may hence assume that X is the only cubic surface containing X (Q; B) and
we may also suppose that B ≥ 2 since the assertion is trivial when B < 2. We
have then by lemma 16.3.1 that any non-singular point in X (Q; B) will satisfy
(16.3.2) or (16.3.3).
It is clear from lemma 16.15 and [HB02, th.5] that there are O(1) lines in the
singular locus and that there are at most Oε (B 1+ε ) singular points in X (Q; B)
outside these lines. It is thus enough to consider the points in X (Q; B) in the
non-singular locus X ns . We also observe that V = B 3 and that V /T 1/3 = B 2
by lemma 1.12 in [Sal07].
Let us first estimate the number of points in X ns (Q; B) satisfying (16.3.2).
We may then fix the t + 1 primes p0 , p1 , . . . , pt in (16.3.2) as the number
of such sets of primes is bounded in terms of ε (cf. 16.3.1(b)). There are
Oε (( p0 p1 . . . pt )2 ) tuples (P0 , P1 , . . . , Pt ) of non-singular F pi -points Pi on
X pi as t + 1 ≤ # is bounded in terms of ε (see 16.3.1(b)). There are, thus,
by (16.3.2) Oε (B 8/7 B 18ε ) such (t + 1)-tuples. There are further by the remark
after (16.3.8) at most 3(t +2)d 2 = Oε (1) points in X ns (Q; B) which specialize
to a given (t + 1)-tuple (P0 , P1 , . . . , Pt ) and satisfy (16.3.2). There are thus
altogether Oε (B 8/7+18ε ) points in X ns (Q; B) satisfying (16.3.2).
We now estimate the number of points in X ns (Q; B) satisfying (16.3.3).
We may then fix the primes p1 , . . . , pt in (16.3.3) as the number of such sets
of primes is bounded in terms of ε. There are Oε (( p1 . . . pt )2 ) and hence
Oε (B 8/7 B 12ε ) t-tuples (P1 , . . . , Pt ) of non-singular F pi -points Pi on X pi
for 1 ≤ i ≤ t. By [Ful84, 8.4.6], there are at most 3d components Z in
X ∩ Y (P1 , . . . , Pt ) and each component is of degree at most 3d. The points
in X ns (Q; B) satisfying (16.3.3) will thus lie on Oε (B 8/7+12ε ) curves Z of
degree ≤ 3d = Oε (1) (see 16.3.1(c)). As there are Od (1) singular points
on each curve of degree ≤ 3d, it only remains to estimate the total contribu-
tion from all Z(Q; B; P1 , . . . , Pt ) such that Pi is a non-singular F pi -point on
Z pi for 1 ≤ i ≤ t. If deg Z ≥ 2, then p1 . . . pt ≥ B 4/7 V ε ≥ B 1/ deg Z V ε
such that #Z (Q; B; P1 , . . . , Pt ) = Oε (1) by 16.14. There are thus altogether
Oε (B 8/7+12ε ) points in X ns (Q; B) satisfying (16.3.3) for some curve Z of
416 Per Salberger

degree at least two. The other points satisfying (16.3.3) will lie on one of
Oε (B 8/7+12ε ) rational lines Z on X , thereby completing the proof.

We shall need the following notation for counting problems for affine
varieties.

Notation 16.17. Let X ⊂ P N be a quasi-projective variety over Q.

(a) S1 (X ; B) is the set of rational points on X which may be represented by an

integral (N + 1)-tuple (1, x1 , . . . , x N ) with |xi | ≤ B for i ∈ {1, . . . , N }.
(b) N1 (X ; B) = #S1 (X ; B).

Lemma 16.18. Let X ⊂ P3 be an integral surface of degree D > 1 over some

field k and H ⊂ P3 be a k-rational plane such that X is not a cone with vertex
in H . Then, the number of lines on X passing through a given point y on H ∩ X
is bounded in terms of D.

Proof. The assertion is proved in [BHBS06, lemma 9] under the stronger

hypothesis that H ∩ X is integral. This hypothesis is only used to ensure that
the schemes Z X,y parametrizing lines l ⊂ X through points y on H ∩ X are
empty or zero-dimensional. Hence as dim(Z X,y ) ≥ 1 would imply that X is a
cone with vertex in y, we obtain the desired result from the proof in (op. cit.).

Theorem 16.19. Let X ⊂ P3 = Proj Q[x0 , . . . , x3 ] be an integral cubic sur-

face such that X is not a cone with vertex in the plane H defined by x0 = 0.
Then N1 (X; B) = Oε (B 8/7+ε ).

Proof. Let X 0 ⊂ X be the open subset where x0 = 0. Then 2N1 (X ; B) =

N (X 0 ; 1, B, B, B). By 16.16 it thus suffices to show that there are
Oε (B 8/7+2ε ) points on a set of Oε (B 8/7+ε ) lines not contained in H . This
follows from the previous lemma and the proof of proposition 1 in [BHBS06].

Theorem 16.20. Let f (y1 , y2 , y3 ) ∈ Q[y1 , y2 , y3 ] be a cubic irreducible

polynomial which is not a polynomial of two linear forms in Q[y1 , y2 , y3 ].
Let n( f ; B) be the number of triples (y1 , y2 , y3 ) of integers in [−B, B] such
that f (y1 , y2 , y3 ) = 0. Then

n( f ; B) = Oε (B 8/7+ε ).
 Uniform bounds for rational points on cubic hypersurfaces 417

Proof. Let F(x0 , x1 , x2 , x3 ) = x03 f (x1 /x0 , x2 /x0 , x3 /x0 ). Then the cubic sur-
face X ⊂ P3 over Q defined by F will satisfy the hypothesis in 16.19. Hence
n( f ; B) = N1 (X ; B) = Oε (B 8/7+ε ).

Theorem 16.21. Let N ≥ 3 and f (y1 , . . . , y N ) ∈ Q[y1 , . . . , y N ] be a cubic

polynomial such that its leading cubic form is absolutely irreducible. Let
n( f ; B) be the number of n-tuples (y1 , . . . , y N ) of integers in [−B, B] such
that f (y1 , . . . , y N ) = 0. Then
n( f ; B) = Oε (B n−2+1/7+ε ).

Proof. This is a special case of theorem 16.20 when N = 3. The general case
is deduced from this case by means of an argument with repeated hyperplane
sections (see [BHBS06, lemma 8]).

Remark. One can use a finer hyperplane section argument than in (op.cit.)
to deduce a more general result than in 16.21 from 16.20. It suffices that
f (y1 , . . . , y N ) ∈ Q[y1 , . . . , y N ] is absolutely irreducible and not a polynomial
of two linear forms in (y1 , . . . , y N ) to have the estimate in 16.21.

16.5 Counting points on families of conics in P3

We shall in this section use the word conic for a closed subscheme of P3 with
Hilbert polynomial 2t +1. A conic may thus be a union of two intersecting lines
or a double line. It is well known (cf. [Har82, 1.b]) that the Hilbert scheme H
of conics in P3 is a P5 -bundle π : H → P3∨ over the dual projective space P3∨ .
This morphism π sends the point on H parametrizing a conic C ⊂ P3 to the
point in the dual plane P3∨ parametrizing the plane  = C ⊂ P3 spanned
by C. We shall also let H X ⊂ H be the subscheme of all conics C ⊂ X on a
cubic surface X ⊂ P3 .

Lemma 16.22. Let X ⊂ P3 be a cubic surface over some field, which is not
covered by its lines. Then the number of lines on X is uniformly bounded.

Proof. This is well known and easy to prove using elimination theory.

Lemma 16.23. Let X ⊂ P3 be an integral cubic surface over some field, which
is not covered by its lines. There is then for each component Y of H X a line
l ⊂ X such that Y parametrizes the residual conics of l in  ∩ X for the planes
 ⊂ P3 passing through l.
418 Per Salberger

Proof. The residual lines l in C ∩ X of the conics C parametrized by Y form

a family of lines over Y . As X and Y are irreducible and X is not covered by
its lines, theses lines will coincide. Hence Y is the family of residual conics of
l in  ∩ X for the planes  passing through l.

Lemma 16.24. Let C ⊂ P3 be a non-singular conic over Q, q = p1 . . . pt

be a square-free integer and (P1 , . . . , Pt ) be a t-tuple where each Pi is a non-
singular F pi -point Pi on C pi . Let H () be the height of the point in the dual
plane P3∨ parametrizing  = C ⊂ P3 . Then the following holds.

(a) #C(Q; B; P1 , . . . , Pt ) ≤ 2 in case H () > 6B 3 /q 3 .

(b) #C(Q; B) = Oε (B/H ()1/3 + B ε ).

Proof. To prove (a), suppose that we had #C(Q; B; P1 , . . . , Pt ) > 2. We may

then find primitive quadruples ξk = (ξk,0 , . . . , ξk,3 ), k = 1, 2, 3 of integers in
[−B, B] representing different points in #C(B; P1 , . . . , Pt ). These cannot be
collinear. So they will span the plane  = C. Its height H () is given by
max0≤ j≤3 | j |/ gcd0≤ j ≤3 ( j ) for the determinants  j , 0 ≤ j ≤ 3 of the four
3 × 3-submatrices of the 3 × 4-submatrix with rows ξk , k = 1, 2, 3. We now
apply 16.9 to the stalk R of the structure sheaf of C cl at Pi . If say x0 (Pi ) = 0,
then r j = x j /x0 ∈ R for 0 ≤ j ≤ 3 and we conclude as in (16.2) that
pi3 |  j for 0 ≤ j ≤ 3. Further, max0≤ j≤3 | j | ≤ 6B 3 as all ξk, j ∈ [−B, B].
This proves that H () ≤ 6B 3 /q 3 when #C(Q; B; P1 , . . . , Pt ) > 2, as
desired.
To prove (b), we first note that (a) also holds when q = 1. Hence
#C(Q; B) ≤ 2 when H () > 6B 3 . It remains to treat the case where
H () ≤ 6B 3 . We may then assume that #C(Q; B; P1 , . . . , Pt ) ≥ 5 and
B ≥ 2. Let ξk = (ξk,0 , . . . , ξk,3 ), k = 1, 2, 3, 4, 5 be primitive quadruples
representing five such points and π ≤ 60B 3 be the product of all primes
p such that three of these quadruples become collinear in C p after reduc-
tion (mod p). There is then by Bertrand’s postulate a prime p not dividing
π such that p > (6B 3 /H ())1/3 and p = Oε (B/H ()1/3 + B ε ). As C p
is non-singular and H () > 6B 3 / p 3 , we may now apply (a) for q = p.
We then get that #C(Q; B; P) ≤ 2 for all F p -points P on C p and that
#C(Q; B) ≤ 2( p + 1) = Oε (B/H ()1/3 + B ε ), as was to be shown.

Lemma 16.25. Let X ⊂ P3 be an integral cubic surface over Q, which is not

covered by its lines. Let q = p1 . . . pt ≥ B 6/7 be a square-free integer with
all prime factors pi ≥ B ε and  be a set of Oε (B 12/7+ε ) tuples (P1 , . . . , Pt )
where Pi is a non-singular F pi -point on X pi for each i. Suppose that for each
 Uniform bounds for rational points on cubic hypersurfaces 419

such λ = (P1 , . . . , Pt ) ∈  we are given a non-singular conic Cλ ⊂ X where

each Pi is a non-singular F pi -point on (Cλ ) pi . Then,

#Cλ (Q; B; P1 , . . . , Pt ) = Oε (B 12/7+ε ).

λ∈

Proof. If H (Cλ ) > 6B 3/7 for λ = (P1 , . . . , Pt ) ∈ , then H (Cλ ) >

6B 3 /q 3 and #Cλ (B; P1 , . . . , Pt ) ≤ 2 by 16.24(a). The total contribution to
the sum from all t-tuples λ with H (Cλ ) > 6B 3/7 is thus Oε (B 12/7+ε ).
For the remaining Cλ , we first consider the contribution from conics C with
H (C) ∈ [R, 2R) for a fixed R ≥ 1. There are O(R 2 ) such conics C on X
by 16.22 and 16.23, and Oε (B/R 1/3 + B ε ) rational points of height ≤ B on
each such conic by 16.24. The sum of #C(Q; B) for all such C will thus be
of order Oε (B R 5/3 + R 2 B ε ) and of order Oε (B 12/7 ) for R ≤ 6B 3/7 . After
summing over O(log B) dyadic intervals [R, 2R), we thus obtain the bound
Oε (B 12/7+ε ) for the total contribution of all C λ with H (Cλ ) ≤ 6B 3/7 . This
completes the proof.

16.6 Rational points on projective cubic surfaces

The following result improves upon theorem 0.1 in [Sal] and theorem 7 in
[HB02].

Theorem 16.26. Let X ⊂ P3 be an integral cubic surface over Q and U be

the complement of the union of the lines on X . Then,

N (U ; B) = Oε (B 12/7+ε ).

Proof. We may just as in the proof of lemma 16.16 assume that X is the only
cubic surface containing X (Q; B) and that B ≥ 2. Then, by 16.3.1 we get
that any non-singular point in X (Q; B) will satisfy (16.3.2) or (16.3.3). We
have also seen in the proof of lemma 16.16 that there are Oε (B 1+ε ) singular
points of height ≤ B on U . It is thus enough to consider the contribution to
N (U ; B) from the non-singular locus X ns of X . Finally, we note that V = B 4
and V /T 1/3 = B 3 .
Let us first estimate the number of points in X ns (Q; B) satisfying
(16.3.2).We may then fix the t + 1 primes p0 , p1 , . . . , pt in (16.3.2) as the
number of such sets of primes is bounded in terms of ε (cf. 16.3.1(b)). There
are Oε (( p0 p1 . . . pt )2 ) (t + 1)-tuples (P0 , P1 , . . . , Pt ) of non-singular F pi -
points Pi on X pi as t + 1 ≤ # is bounded in terms of ε (see 16.3.1(b)). There
420 Per Salberger

are thus by (16.3.2) Oε (B 12/7 B 24ε ) such (t + 1)-tuples. There are also by the
remark after (16.3.8) at most 3(t + 2)d 2 = Oε (1) points in X ns (Q; B) which
specialize to a given (t + 1)-tuple (P0 , P1 , . . . , Pt ) and satisify (16.3.2). There
are hence altogether Oε (B 12/7+24ε ) points in X ns (Q; B) satisfying (16.3.2).
We now estimate the contribution to N (U ; B) from the points in X ns (Q; B)
satisfying (16.3.3). We may then fix the primes p1 , . . . , pt in (16.3.3) as
the number of such sets of primes is bounded in terms of ε. There are
Oε (( p0 p1 . . . pt )2 ) and hence Oε (B 12/7 B 16ε ) tuples (P1 , . . . , Pt ) of non-
singular F pi -points Pi for 1 ≤ i ≤ t. By [Ful84, 8.4.6], there are at most
3d components Z in X ∩ Y (P1 , . . . , Pt ) and each such component is of
degree at most 3d. The points in X ns (Q; B) satisfying (16.3.3) will thus lie
on Oε (B 12/7+16ε ) curves Z of degree ≤ 3d = Oε (1) (see 16.3.1(c)). As there
are Od (1) singular points on each curve of degree ≤ 3d, it only remains to
estimate the total contribution to N (U ; B) from all Z (Q; B; P1 , . . . , Pt ) such
that Pi is a non-singular F pi -point on Z pi for 1 ≤ i ≤ t. If deg Z ≥ 3, then
p1 . . . pt ≥ B 6/7 V ε ≥ B 2/ deg Z V ε such that #Z (Q; B; P1 , . . . , Pt ) = Oε (1)
by 16.13. There are thus altogether Oε (B 12/7+16ε ) points in X ns (Q; B) satisfy-
ing (16.3.3) for some curve Z of degree at least three. By 16.25 we get the same
total bound for the number of points in X ns (Q; B) satisfying (16.3.3) for some
non-singular conic Z . This finishes the proof as the components Z of degree
one and the singular components Z of degree two are disjoint from U .

Remark. The method of this paper applies to other del Pezzo surfaces. If
X ⊂ P3 is an integral quadric, one gets thereby another proof of the bound
N (X ; B) = Oε (B 2+ε ) in [HB02, theorem 2]. If U is the complement of
the lines on X on an anticanonically embedded non-singular del Pezzo sur-
face over Q of degree 3 ≤ d ≤ 8, then one obtains that N (U ; B) =
Oε (B 12/(d+4)+ε ).
One can combine the method of this paper with conic bundle techniques
to get non-uniform upper bounds for classes of cubic surface with rational
lines. If X contains one rational line then N (U ; B) = O X,ε (B 8/5+ε ) while
N (U ; B) = O X,ε (B 3/2+ε ) for a cubic surface with two rational lines.

References
[AK70] Allen Altman and Steven Kleiman. Introduction to Grothendieck duality
theory. Lecture Notes in Mathematics, Vol. 146. Springer-Verlag, Berlin-
New York, 1970.
[BHBS06] T. D. Browning, D. R. Heath-Brown, and P. Salberger. Counting rational
points on algebraic varieties. Duke Math. J., 132(3):545–578, 2006.
 Uniform bounds for rational points on cubic hypersurfaces 421

[Bro04] Niklas Broberg. A note on a paper by R. Heath-Brown: “The density of

rational points on curves and surfaces” [Ann. of Math. (2) 155 (2002), no.
2, 553–595; mr1906595]. J. Reine Angew. Math., 571:159–178, 2004.
[Eis95] David Eisenbud. Commutative Algebra with a View toward Algebraic
Geometry, volume 150 of Graduate Texts in Mathematics. Springer-Verlag,
New York, 1995.
[FL85] William Fulton and Serge Lang. Riemann-Roch algebra, volume 277 of
Grundlehren der Mathematischen Wissenschaften [Fundamental Princi-
ples of Mathematical Sciences]. Springer-Verlag, New York, 1985.
[Ful84] William Fulton. Intersection theory, volume 2 of Ergebnisse der Math-
ematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related
Areas (3)]. Springer-Verlag, Berlin, 1984.
[Har82] Joe Harris. Curves in projective space, volume 85 of Séminaire de Math-
ématiques Supérieures [Seminar on Higher Mathematics]. Presses de
l’Université de Montréal, Montreal, Que., 1982. With the collaboration of
David Eisenbud.
[HB02] D. R. Heath-Brown. The density of rational points on curves and surfaces.
Ann. of Math. (2), 155(2):553–595, 2002.
[Sal] Per Salberger. Counting rational points on projective varieties. submitted.
[Sal07] Per Salberger. On the density of rational and integral points on algebraic
varieties. J. Reine Angew. Math., 606:123–147, 2007.
[Sal08] Per Salberger. Rational points of bounded height on projective surfaces.
Math. Z., 258(4):805–826, 2008.
 17
Descent on toric fibrations
Alexei N. Skorobogatov
Department of Mathematics, South Kensington Campus, Imperial College London,
SW7 2BZ England, U.K. – and – Institute for the Information Transmission Problems,
Russian Academy of Sciences, 19 Bolshoi Karetnyi, Moscow, 127994 Russia
E-mail address: [email protected]

A BSTRACT. We describe descent on families of torsors for a constant torus. A

recent result of Browning and Matthiesen then implies that the Brauer–Manin
obstruction controls the Hasse principle and weak approximation when the
ground field is Q and the singular fibres are all defined over Q.

17.1 Introduction
Let T be a torus over a number field k. Let X be a smooth, proper, geo-
metrically integral variety with a surjective morphism f : X → P1k whose
generic fibre X k(t) is geometrically integral and is birationally equivalent to a
k(t)-torsor for T . The main result of this note says that the set X (k) is dense
in X (Ak )Br if certain auxiliary varieties satisfy the Hasse principle and weak
approximation.
These varieties are given by explicit equations. Choose A1k ⊂ P1k so that the
fibre of f at ∞ = P1k \A1k is smooth. Let P1 , . . . , Pr be closed points of P1k such
that each fibre of the restriction of f to P1k \ (P1 ∪ . . . ∪ Pr ) is split, i.e. contains
an irreducible component of multiplicity 1 that is geometrically irreducible.
By a well-known result (Lemma 17.8) the fibre X Pi , for i = 1, . . . , r , has
an irreducible component of multiplicity 1. We fix such a component in each
X Pi and define ki as the algebraic closure of the residue field k(Pi ) in the
function field of this component. For i = 1, . . . , r let pi (t) ∈ k[t] be the monic
irreducible polynomial such that Pi is the zero set of pi (t) in A1k , and let ai be
the image of t in k(Pi ) = k[t]/( pi (t)). Let u, v be independent variables, and
let z i be a ki -variable, for i = 1, . . . , r . For α = {αi }, where αi ∈ k(Pi )∗ , we

define the quasi-affine variety Wα ⊂ A2k × ri=1 Rki /k (A1ki ) by

αi (u − ai v) = Nki /k(Pi ) (z i ), (u, v) = (0, 0), i = 1, . . . , r. (17.1)

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

422
 Descent on toric fibrations 423

The varieties Wα are smooth and geometrically irreducible. Over an algebraic


closure of k such a variety is given by ri=1 [k(Pi ) : k] equations in 2 +
r
i=1 [ki : k] variables. We can now state our main result.
r ∗
Theorem 17.1. Suppose that for each α ∈ i =1 k(Pi ) the variety Wα
satisfies the Hasse principle and weak approximation. Then X (k) is dense in
X (Ak )Br .

The results of this kind are obtained by the descent method of Colliot-
Thélène and Sansuc, and have a long history. When the relative dimension and
the number of singular geometric fibres of f are small, geometric proofs of
the Hasse principle and weak approximation for Wα were obtained by Colliot-
Thélène, Sansuc, Swinnerton-Dyer and others, see [8, 5, 22, 27, 10] and [24,
Ch. 7]. The analytic tool in these proofs is Dirichet’s theorem on primes in
an arithmetic progression. Over k = Q one can do more if one uses more
advanced analytic methods: the circle method [18, 9], sieve methods [1] (see
also [11] where the results of [1] are used) and recent powerful results from
additive combinatorics [12, 13, 14, 3, 17, 25]. Note that the circle method can
sometimes be applied over arbitrary number fields, see [26, 21].
When k = k(P1 ) = . . . = k(Pr ) = Q, a recent result of Browning and
Matthiesen obtained by methods of additive combinatorics [2, Thm 1.3] estab-
lishes the Hasse principle and weak approximation for Wα . Hence we deduce
the following corollary of Theorem 17.1.

Corollary 17.2. Let X be a smooth, proper, geometrically integral variety over

Q, and let f : X → P1Q be a surjective morphism satisfying the following
properties.
(a) There is a torus T over Q such that the generic fibre X Q(t) of f is
birationally equivalent to a Q(t)-torsor for T ×Q Q(t).
(b) There exists a finite subset E ⊂ P1Q (Q) such that X \ f −1 (E) → P1Q \ E
has split fibres.
Then X (Q) is dense in X (AQ )Br .

This generalizes a recent result due to A. Smeets, namely the unconditional

counterpart of [25, Thm. 1.1, Rem. 1.3]. For a version of this statement over
PnQ see Proposition 17.11.
For a number field k of degree n = [k : Q] we write x = (x1 , . . . , xn )
and denote by Nk/Q (x) the norm form Normk/Q (x1 ω1 + . . . + xn ωn ), where
ω1 , . . . , ωn is a basis of k as a vector space over Q. The following corollary
extends [2, Thm. 1.1] to the case of a product of norm forms. It generalizes
424 Alexei N. Skorobogatov

the unconditional version of [25, Cor. 1.6], a number of statements from [17,
Section 4] and the main result of [21] in the case of the ground field Q.

Corollary 17.3. Let k1 , . . . , kn be number fields and let m 1 , . . . , m n be

positive integers with gcd(m 1 , . . . , m n ) = 1. Let L i ∈ Q[t1 , . . . , ts ] be poly-
nomials of degree 1, for i = 1, . . . , r . Let X be a smooth and proper variety
over Q that is birationally equivalent to the affine hypersurface

r 
n
L i (t1 , . . . , ts ) = Nk j /Q (x j )m j . (17.2)
i =1 j=1

Then X (Q) is dense in X (AQ )Br .

Note that repetitions among L 1 , . . . , L r are allowed here. Corollary 17.3 is

a particular case of Proposition 17.12 which deals with several equations like
(17.2) and extends [2, Thm. 1.3].
This note consists of two sections. In §2 we make preliminary remarks, some
of which, like Corollary 17.7, are not needed in the proof of our main results
but could possibly be of independent interest. In §3 we prove Theorem 17.1,
Corollary 17.3 and their generalizations.
The density of X (Q) in X (AQ )Br can sometimes be proved by the method
of fibration, when rational points are found in specific subvarieties of X , or
by the method of descent, when rational points are found in certain X -torsors
which are varieties with a surjective morphism to X . Our proof of Theorem
17.1 uses descent like [3], [11] or [21], and not the fibration method like [25]
or [17]. It was inspired by the approach of Colliot-Thélène and Sansuc [6] to
degeneration of torsors for tori, and by their computation of universal torsors
on conic bundles [7, Section 2.6]. We apply open descent based on Harari’s
formal lemma [15, Cor. 2.6.1] as in [10], with an improvement found in [9]
(see the proof of Thm. 3.1, pages 84–85).
The author is grateful to Jean-Louis Colliot-Thélène for useful discussions
over the past many years. I would like to thank Daniel Loughran for his
question that led me to Proposition 17.6, and Olivier Wittenberg for useful
comments. This work was carried out at the Institute for the Information Trans-
mission Problems of the Russian Academy of Sciences at the expense of the
Russian Foundation for Sciences (project number 14-50-00150).

17.2 Reduction of varieties over discretely valued fields

Let R be a discrete valuation ring with the field of fractions K , the maximal
ideal m and the residue field κ = R/m. The following lemma is well known,
cf. [6, p. 163] or [24, p. 70].
 Descent on toric fibrations 425

Lemma 17.4. Let T be a torus over R, and let Tκ = T × R κ. Let Tκ◦ be the
!κ◦ is the
torus dual to Tκ , that is, the κ-torus whose module of characters T
Galois module Hom(T !κ , Z).

(i) The restriction map from R to K fits into an exact sequence

!κ◦ ).
H1 (R, T ) → H1 (K , T ) → H1 (κ, T (17.3)

(ii) If R is a subring of a discrete valuation ring R  with the residue field κ  ,

the fraction field K  and the maximal ideal m such that m = mR  , then
the restriction maps give rise to a commutative diagram

H1 (R  , T ) → H1 (K  , T ) !κ◦ )
→ H1 (κ  , T
↑ ↑ ↑ (17.4)
H1 (R, T ) → H1 (K , T ) → H1 (κ, T !κ◦ )

j
Proof. (i) Let Spec(K ) / Spec(R) o i Spec(κ) be the embeddings
of the generic and the special points, respectively. By [19, Example II.3.9]
we have an exact sequence of étale sheaves on Spec(R)

0 → Gm,R → j∗ Gm,K → i ∗ Zκ → 0. (17.5)

For any scheme X and any X -torus T the local-to-global spectral sequence
for Ext gives a canonical isomorphism H1 (X, T ) = Ext1X (T!, Gm ), see [24,
Lemma 2.3.7] (although in loc. cit. X is a variety over a field, the same
!, ·) to (17.5) gives
proof works in general). Thus an application of Ext1R (T
an exact sequence

!, j∗ Gm,K ) → Ext1R (T
H1 (R, T ) → Ext1R (T !, i ∗ Zκ ).

p ! p+q !
The spectral sequence Ext R (T , R q i ∗ Zκ ) ⇒ Extκ (T κ , Z), in view of the
vanishing of R 1 i ∗ Zκ = 0 [19, Example III.2.22], gives rise to the first
canonical isomorphism in

!, i ∗ Zκ ) = Ext1κ (T
Ext1R (T !κ , Z) = H1 (κ, T
!κ◦ ).

The second canonical isomorphism in this line is provided by the spec-

q ! p+q !
tral sequence H p (κ, ExtZ (Tκ , Z)) ⇒ Extκ (Tκ , Z). Hilbert’s theorem
90 gives R 1 j∗ Gm,K = 0, which produces canonical isomorphisms

!, j∗ Gm,K ) = Ext1K (T
Ext1R (T !, Z) = H1 (K , T ).

Putting all these together gives the exact sequence (17.3).

426 Alexei N. Skorobogatov

(ii) The left-hand square in (17.4) is obviously commutative. The commuta-

tivity of the right-hand square follows from the fact that the restriction of
the discrete valuation of R  to R is the given discrete valuation on R.

For an R-scheme X we denote by X K and X κ the generic and special fibres

of X , respectively.

Lemma 17.5. Let Y be a K -torsor for TK . Suppose that there is an integral

normal scheme X and a surjective morphism X → Spec(R) of finite type with
integral fibres such that the generic fibre X K is K -birationally equivalent to Y .

(i) Let κ  be the separable closure of κ in the function field of the special
!◦ ) is in the kernel of the
fibre κ(X κ ). Then the image of [Y ] in H1 (κ, T
restriction map H (κ, T
1 ! ) → H (κ , T
◦ 1  ! ).
◦

(ii) If X κ is geometrically integral, there is an R-torsor Z for T such that

Y ∼= ZK .

Proof. (i) The morphism X → Spec(R) is surjective, so the (integral) spe-

cial fibre X κ has codimension 1. Let R  be the local ring of X κ . Since
X is normal, R  is an integrally closed domain of dimension 1. Since
X → Spec(R) is of finite type, R  is Noetherian, and hence a discrete
valuation ring. As X is integral, the field of fractions of R  is K (X K ). A
local parameter of R is also a local parameter of R  , because X κ is integral.
Thus m = mR  is the maximal ideal of R  , and the residue field R  /m is
the field of functions on the special fibre X κ . In our case the commutative
diagram (17.4) takes the form

H1 (R  , T ) → H1 (K (X K ), T ) !κ◦ )
→ H1 (κ(X κ ), T
↑ ↑ ↑
H1 (R, T ) → H1 (K , T ) → !κ◦ )
H1 (κ, T

The restriction of the diagonal Y → Y × K Y to the generic point of

Y is a K (Y )-point of Y . Hence the torsor Y is split by the field exten-
sion K (X K ) = K (Y ), so that the class [Y ] ∈ H1 (K , T ) goes to zero in
H1 (K (X K ), T ).
The right-hand vertical map in the diagram factorises as follows:
!κ◦ ) −→ H1 (κ  , T
H1 (κ, T !κ◦ )−→H !κ◦ ).
˜ 1 (κ(X κ ), T

The second map here is an isomorphism because T !κ◦ is a finitely generated


free abelian group and κ is separably closed in κ(X κ ). This implies (i).
 Descent on toric fibrations 427

(ii) If the κ-scheme X κ is geometrically integral, the field κ is separably closed

in κ(X κ ), that is, κ  = κ. The exact sequence (17.3) then shows that [Y ] is
in the image of the map H1 (R, T ) → H1 (K , T ), and this proves (ii).

The statement of Lemma 17.5 (i) leaves open the question to what extent
κ  is determined by the field K (X K ). We treat this as a question about inte-
gral, regular, proper schemes over a discretely valued field, see Corollary 17.7
below.
For an integral variety V over a field k, we write k V for the separable closure
of k in the field of functions k(V ).

Proposition 17.6. Let Y and Y  be integral regular schemes that are proper
and flat over R. If there is a dominant rational map from Y K to Y K , then for
any irreducible component C ⊂ Yκ of multiplicity 1 there exists an irreducible
component C  ⊂ Yκ of multiplicity 1 such that κC  ⊂ κC .

Proof. Write F = K (Y K ) and F  = K (Y K ). We are given an inclusion F  ⊂

F, or, equivalently, a morphism Spec(F) → Spec(F  ). Let OC be the local
ring of the generic point of C. Since Y is regular and C has codimension 1
in Y , OC is a discrete valuation ring. The multiplicity of C is 1, that is, the
maximal ideal of OC is mOC = m ⊗R OC . Since Y is integral, the field of
fractions of OC is F. The residue field of OC is κ(C).
After the base change from R to OC we obtain a morphism Y  × R OC → OC .
Its generic fibre Y F = Y K × K F has an F-point Spec(F) → Spec(F  ) → Y F
coming from the F  -point defined by the diagonal morphism Y K → Y K × K Y K .
The morphism Y  × R OC → OC is proper, and by the valuative criterion
of properness any F-point of its generic fibre extends to a section of the
morphism.
Since C has multiplicity 1, the special fibre of Y  × R OC → OC is Yκ ×κ
κ(C). A section of Y  × R OC → OC thus gives rise to a κ(C)-point of Yκ ×κ
κ(C), which can be viewed also as a κ(C)-point of Yκ . Since Y  × R OC is
regular, any section of Y  × R OC → OC meets the special fibre at a smooth
point. In particular, this point belongs to exactly one geometric irreducible
component, moreover, this component has multiplicity 1 (see the calculation
in the proof of [23, Lemma 1.1 (b)], or [28, Lemme 3.8]).
Let U ⊂ Yκ be the smooth locus of Yκ . Since U contains a κ(C)-point, it
is non-empty. By Stein factorization, the structure morphism U → Spec(κ)
factors through the surjective morphism U → Spec(L) with geometrically
connected fibres, where L is an étale κ-algebra. Explicitly, L is the direct sum
428 Alexei N. Skorobogatov

of finite field extensions of κ, each of which is the separable closure of κ in

the function field of an irreducible component of Yκ of multiplicity 1. The
image of the composed map Spec(κ(C)) → Spec(L) is connected, hence this
is Spec(κC  ) for some irreducible component C  ⊂ Yκ of multiplicity 1. Thus
κC  ⊂ κ(C) and hence κC  ⊂ κC .

For a discussion of some related results see Section 17.3 of [4].

Corollary 17.7. Let X be an integral regular scheme that is proper and flat
over R. Let ' X be the partially ordered set of irreducible components of mul-
tiplicity 1 of X κ , where C dominates D if κ D ⊂ κC . The set of finite field
extensions κ ⊂ κC , where C is a minimal element of ' X , is a birational
invariant of the generic fibre X K .

Proof. Suppose that proper R-schemes X and Y are integral and regular with
birationally equivalent generic fibres, that is, K (X K ) ∼
= K (Y K ). Let C be a
minimal element of ' X . By Proposition 19.19 there exists C  ∈ 'Y such that
κC  ⊂ κC . By the same proposition, there is C  ∈ ' X such that κC  ⊂ κC  . By
minimality of C we have κC  = κC , hence κC = κC  . If C  is not minimal in
'Y , then, by Proposition 19.19, C is not minimal in ' X .

This set of finite extensions of the residue field can be explicitly determined
when the generic fibre is a conic, and, more generally, a Severi–Brauer variety,
or a quadric of dimension 2.

Remark. Olivier Wittenberg suggested a somewhat different approach to

Proposition 17.6 and Corollary 17.7. Consider a discrete valuation v :
K (X K )∗ → Z such that the restriction of v to K ∗ is the given discrete val-
uation of K . Let R  be the valuation ring of v, and let κ  be the separable
closure of κ in the residue field of R  . Let us call # X the resulting set of
finite field extensions of κ partially ordered by inclusion. It is clear that # X
is a birational invariant of X K . The discrete valuation associated to an irre-
ducible component of X κ of multiplicity 1 is an example of such a valuation,
so we have an inclusion of partially ordered sets ' X ⊂ # X . Since X is
proper over R and regular, it can be shown that any extension of the given
discrete valuation of K to K (X K ) gives rise to a morphism of R-schemes
Spec(R  ) → X that factors through the smooth locus of X/R. Hence ' X and
# X have the same set of minimal elements, which is thus a birational invariant
of X K .
 Descent on toric fibrations 429

17.3 Torsors over toric fibrations

Let k be a field of characteristic zero. Let k̄ be an algebraic closure of k, and
let k = Gal(k̄/k). For a variety X over k we write X = X ×k k̄. We use the
standard notation Br0 (X ) for the image of the natural map Br(k) → Br(X ),
and Br1 (X ) for the kernel of the natural map Br(X ) → Br(X ).
Let T be a k-torus. We write T ! for the k -module of characters of T .
Let X be a smooth, proper, geometrically integral variety with a surjective
morphism f : X → P1k and the geometrically integral generic fibre X k(t)
which is birationally equivalent to a k(t)-torsor for T .

Lemma 17.8. Each fibre of X → P1k has an irreducible component of

multiplicity 1.

Proof. The generic fibre of X → P1k̄ is birationally equivalent to a tor-

sor for T Gdm,k̄ , where d = dim(T ). By Hilbert’s theorem 90 we have
H1 (k̄(t), T ) = H1 (k̄(t), Gm )d = 0, so this torsor has a k̄(t)-point. By the
lemma of Lang and Nishimura, X k(t) has a k̄(t)-point too. By the valuative
criterion of properness, this point extends to a section of the proper mor-
phism X → P1k̄ . Since X is smooth, by a standard argument (see the proof
of [23, Lemma 1.1 (b)], or [28, Lemme 3.8]) any section intersects each fibre
of X → P1k̄ in an irreducible component of multiplicity 1.

The same conclusion holds in far greater generality: by a theorem of Graber,

Harris and Starr it is enough to assume the generic fibre to be rationally con-
nected. We refer the reader to Section 7 of [4] for a survey of known results, in
particular, see [4, Thm. 7.7].

Proof of Theorem 17.1. We keep the notation of §1. By Lemma 17.8 there
is an irreducible component of multiplicity 1 in each X Pi , for i = 1, . . . , r .
We mark these components. We also mark a geometrically integral irreducible
component of multiplicity 1 in each of the remaining fibres of f . Define Y ⊂ X
as the complement to the union of all the unmarked irreducible components of
the fibres of f : X → P1k . It is clear that Y is a dense open subset of X . The
restriction of f to Y is a surjective morphism f : Y → P1k with integral fibres,
and with proper and geometrically integral generic fibre Yk(t) = X k(t) . An
standard easy argument then shows that k̄[Y ]∗ = k̄ ∗ and Pic(Y ) is torsion-free,
cf. the proof of Thm. A in [10, p. 391].

Let Pic(Y ) → Pic(Yk̄(t) ) be the homomorphism of k -modules induced by

the inclusion of the generic fibre Yk̄(t) into Y . This homomorphism is surjective
430 Alexei N. Skorobogatov

since Y is smooth. Let S be the k-torus defined by the exact sequence of k-

modules
0→!
S → Pic(Y ) → Pic(Yk̄(t) ) → 0. (17.6)

Thus the abelian group ! S is generated by the geometric irreducible components

of the fibres Y P1 , . . . , Y Pr (this holds when r ≥ 1; for r = 0 the group ! S∼ =Z
is generated by the class of a geometric fibre). Recall that a vertical torsor
T → Y is a torsor of S whose type is the injective map ! S → Pic(Y ) from
(17.6). According to [24, Prop. 4.4.1] such torsors exist. For any vertical torsor
T the injectivity of type ! S → Pic(Y ) and k̄[Y ]∗ = k̄ ∗ imply k̄[T ]∗ = k̄ ∗ , see
[7, Prop. 2.1.1] or formula (2.6) in [24]. By (17.6) and [7, Prop. 2.1.1] we have
isomorphisms of Galois modules

Pic(T ) ∼
= Pic(Yk̄(t) ) ∼
= Pic(X k̄(t) ).

The Picard group of the proper rational variety X k̄(t) is torsion-free. It is

well known that the absence of torsion in Pic(T ) implies the finiteness of
H1 (k, Pic(T )), and then k̄[T ]∗ = k̄ ∗ implies the finiteness of Br1 (T )/Br0 (T ).
Recall that for i = 1, . . . , r we write k(Pi ) for the residue field of Pi , and ki
for the algebraic closure of k(Pi ) in the function field k(Y Pi ). The variety Wα
is defined by (17.1), which is also equation (4.34) of [24].

Lemma 17.9. The variety Wα is smooth and geometrically irreducible, and

the morphism π : Wα → P1k given by (u : v) is faithfully flat. We have
k̄[Wα ]∗ = k̄ ∗ and Pic(W α ) = 0.

Proof. The first statement is [24, Lemma 4.4.5]. The second statement is a
straightforward adaptation of [7, Lemme 2.6.1].

The reader may want to compare the following statement with

[7, Thm. 2.6.4].

Proposition 17.10. Let f : Y → P1k be as above. Then any vertical torsor over

Y is birationally equivalent to the product Wα ×k Z , where α ∈ ri=1 k(Pi )∗
and Z is a k-torsor for T .

Proof. The local description of torsors due to Colliot-Thélène and Sansuc, see
[24, Thm. 4.3.1, Cor. 4.4.6], can be stated as follows. (Recall that the fibre of
f : Y → P1k at infinity is assumed to be smooth.) Let

U  = A1k \ {P1 , . . . , Pr }, U = Y ∩ f −1 (U  ), Vα = π −1 (U  ).
 Descent on toric fibrations 431


Then for any vertical torsor T /Y there exists an α ∈ ri=1 k(Pi )∗ such that
the restriction TU = T ×Y U is isomorphic to the fibred product U ×U  Vα .
Let us write W = Wα , V = Vα .
Let j  : Spec(k(t)) → U  be the embedding of the generic point of A1k
into the open set U  . Let i P : P → U  be the embedding of a closed point
P ⊂ U  . Since A1k is smooth, there is the following exact sequence of étale
sheaves on U  :

0 → Gm,U  → j∗ Gm,k(t) → ⊕ P∈U  i P∗ Zk(P) → 0,

see [19, Examples II.3.9, III.2.22]. Repeating the proof of Lemma 17.5, we see
1 (T
that an application of ExtU 
!, ·) to this sequence produces an exact sequence

!k(P)
H1 (U  , T ) → H1 (k(t), T ) → ⊕ P∈U  H1 (k(P), T ◦
). (17.7)

Let ξ ∈ H1 (k(t), T ) be the class of a k(t)-torsor for T birationally equivalent to

X k(t) . The map H1 (k(t), T ) → H1 (k(P), T !◦ ) in (17.7) can be computed in
k(P)
the local ring R = O P of P in A1k . The fibres of f : U → U  are geometrically
integral, hence Lemma 17.5 implies that this map is trivial. Thus we see from
(17.7) that ξ comes from some ξ  ∈ H1 (U  , T ). By Lemma 17.5 (i) applied to
f : Y → P1k the image of ξ  in H1 (k(Pi ), T!◦ ) goes to zero in H1 (ki , T !◦ ).
k(P) k(P)
The equations (17.1) show that the fibre Wi = π −1 (Pi ) is the product of a
torsor for a k(Pi )-torus and the affine k(Pi )-variety defined by Nki /k(Pi ) (z i ) =
0. Thus Wi is integral over k(Pi ) and the field ki is the algebraic closure of
k(Pi ) in the function field of Wi . If we write P0 = ∞, then by setting v = 0
in (17.1) we see that W0 = π −1 (P0 ) is geometrically integral, and thus the
algebraic closure of k(P0 ) = k in k(W0 ) is k0 = k. We have a commutative
diagram with exact rows similar to (17.4):

H1 (W, T ) → H1 (V, T ) !◦ )
→ ⊕ri=0 H1 (k(Wi ), T k(Pi )
↑ ↑ ↑
H1 (k, T ) → H1 (U  , T ) !◦ )
→ ⊕ri=0 H1 (k(Pi ), Tk(Pi )

By the structure of degenerate fibres of π : W → P1k described above, the

!◦ ) is zero, for i = 0, . . . , r . Now the upper
image of ξ  in H1 (k(Wi ), Tk(Pi )
sequence in the diagram shows that π ∗ (ξ  ) ∈ H1 (V, T ) comes from H1 (W, T ).
By Lemma 17.9 we have k̄[W ]∗ = k̄ ∗ and Pic(W ) = 0, and thus the fun-
damental exact sequence of Colliot-Thélène and Sansuc (see [7, Thm. 1.5.1]
or [24, Cor. 2.3.9]) shows that the natural map H1 (k, T ) → H1 (W, T ) is an
isomorphism. It follows that any W -torsor for T is the product of W and a
k-torsor for T .
432 Alexei N. Skorobogatov

End of proof of Theorem 17.1. Let (Mv ) ∈ X (Ak )Br . By a theorem of

Grothendieck, Br1 (X ) is naturally a subgroup of Br1 (Y ). We have k̄[Y ]∗ = k̄ ∗ ,
and this implies that Br1 (Y )/Br0 (Y ) is a subgroup of H1 (k, Pic(Y )), which is
finite because Pic(Y ) is torsion-free. Thus we can use [10, Prop. 1.1] (a conse-
quence of Harari’s formal lemma) which says that the natural injective map of
topological spaces Y (Ak )Br1 (Y ) → X (Ak )Br1 (X ) has a dense image. Thus we
can assume without loss of generality that (Mv ) ∈ Y (Ak )Br1 (Y ) .
The main theorem of the descent theory of Colliot-Thélène and Sansuc states
that for any adelic point in Y (Ak )Br1 (Y ) there exists a universal torsor T0 → Y
such that our adelic point is in the image of the map T0 (Ak ) → Y (Ak ), see
[7, Section 3] or [24, Thm. 6.1.2(a)]. Thus we can find a point (Nv ) ∈ T0 (Ak )
such that the image of Nv in Y is Mv for all v.
The structure group of T0 → Y is the Néron–Severi torus T0 defined by the
property T !0 = Pic(Y ). The dual of the injective map in (17.6) is a surjective
morphism of tori T0 → S. Let S0 be its kernel. Then T = T0 /S0 is a Y -torsor
with the structure group S whose type is the natural map ! S → Pic(Y ), so T
is a vertical torsor. Since T0 is a universal torsor, [7, Prop. 2.1.1] tells us that
k̄[T0 ]∗ = k̄ ∗ and Pic(T 0 ) = 0, hence Br1 (T0 ) = Br0 (T0 ). Let (Pv ) ∈ T (Ak )
be the image of (Nv ). By the functoriality of the Brauer–Manin pairing we see
that (Pv ) ∈ T (Ak )Br1 (T ) .
By Proposition 17.10 the variety T is birationally equivalent to E × Wα for

some α ∈ ri=1 k(Pi )∗ . Let E c be a smooth compactification of E. Then we
have a rational map g from the smooth variety T to the proper variety E c , and
a rational map h from T to Wα . By the valuative criterion of properness there
is an open subset  ⊂ T with complement of codimension at least 2 in T such
that g is a morphism  → E c . Let  ⊂  be a dense open subset such that
(g, h) defines an open embedding  ⊂ E c × Wα .
By Grothendieck’s purity theorem the natural restriction map gives an
isomorphism Br(T )−→Br().
˜ Thus g ∗ Br1 (E c ) ⊂ Br1 () is naturally a sub-
group of Br1 () = Br1 (T ). (The argument given in the last three paragraphs
is inspired by the proof of Thm. 3.1 in [9].)
Evaluating Brauer classes at local points is continuous and hence locally
constant. Since Br1 (T )/Br0 (T ) is finite, by a small deformation we can
assume Pv ∈  (kv ) for each v. Thus (g(Pv )) is a well-defined element
of E c (Ak )Br1 (E ) . By Sansuc’s theorem [20], E(k) is then non-empty and,
c

moreover, E(k) is dense in E c (Ak )Br1 (E ) .

c

Let ' be a finite set of places of k containing all the places where we need to
approximate. By the Hasse principle and weak approximation for Wα we can
find a k-point in  that is arbitrarily close to (g(Pv ), h(Pv )) for v ∈ '. We
conclude that there is a k-point in T that is arbitrarily close to Pv for v ∈ '.
 Descent on toric fibrations 433

The image of this point in Y approximates (Mv ). This finishes the proof of
Theorem 17.1.

Proposition 17.11. Let X be a smooth, projective, geometrically integral

variety over Q, and let f : X → PnQ be a surjective morphism satisfying
the following properties.

(a) There is a torus T over Q such that the generic fibre X Q(η) of f is
birationally equivalent to a Q(η)-torsor for T ×Q Q(η).
(b) There exist hyperplanes H1 , . . . , Hr ⊂ PnQ such that f has split fibres at
all points of codimension 1 of PnQ other than H1 , . . . , Hr .
Then X (Q) is dense in X (AQ )Br .

Proof. We deduce this from a fibration theorem due to D. Harari [16, Thm.
3.2.1]. There is an open subset U ⊂ PnQ such that each fibre of f −1 (U ) → U
is geometrically integral and birationally equivalent to a torsor for T . Choose
a point M ∈ U (Q) and a hyperplane L ⊂ PnQ through M, and let V =
f −1 (PnQ \ L). Obviously, V is a quasi-projective dense open subset of X . The
projective lines through M that are not contained in L are in a natural bijection
with the points of An−1
Q , so we have a surjective morphism p : V → AQ .
n−1

Each fibre of p is split, because the fibre of f at the generic point of the corre-
sponding line through M is geometrically integral and hence its Zariski closure
is a geometrically integral irreducible component of multiplicity 1. Any point
of X M (Q) defines a section of p over Q. The generic fibre of p is birationally
equivalent to a family of torsors for T parameterized by an open subset of the
projective line, hence it is geometrically integral and geometrically rational.
In particular, a smooth and proper model of the generic fibre has trivial geo-
metric Brauer group and torsion-free geometric Picard group. The fibres of p
over Q-points of a non-empty open subset of An−1 Q satisfy the assumptions of
Corollary 17.2, hence all the conditions of [16, Thm. 3.2.1] are satisfied. An
application of this result proves the proposition.

Proposition 17.12. Let k1 , . . . , kn be number fields, and let L 1 , . . . , L r ∈

Q[t1 , . . . , ts ] be polynomials of degree 1. Let m i j ≥ 0, for i = 1, . . . , 
and j = 1, . . . , n, be integers such that the sublattice of Zn generated by
the rows of the matrix (m i j ) is primitive. Finally, let di p ≥ 0 be integers, where
i = 1, . . . ,  and p = 1, . . . , r . Let X be a smooth and proper variety over
Q that is birationally equivalent to the affine variety given by the system of
equations
434 Alexei N. Skorobogatov


r 
n
ci L p (t1 , . . . , ts )di p = Nk j /Q (x j )m i j , i = 1, . . . , , (17.8)
p=1 j=1

where ci ∈ Q∗ for i = 1, . . . , . Then X (Q) is dense in X (AQ )Br .

Proof. The condition on the matrix (m i j ) implies that the affine variety
given by

n
Nk j /Q (x j )m i j = 1, i = 1, . . . , ,
j=1

is a torus. Let us call it T . The affine variety Y given by (17.8) has a morphism
g : Y → AsQ given by the coordinates t1 , . . . , ts . Let Hi be the hyperplane
L i = 0. The restriction of g to AsQ \ (H1 ∪ . . . ∪ Hr ) is a torsor for T . We can
choose a smooth compactification X of the smooth locus Ysm in such a way
that there is a surjective morphism f : X → PsQ extending g : Ysm → AsQ .
Now we apply Proposition 17.11.

For  = 1 this statement is Corollary 17.3.

References
[1] T. Browning and D.R. Heath-Brown. Quadratic polynomials represented by norm
forms. GAFA 22 (2012) 1124–1190.
[2] T. Browning and L. Matthiesen. Norm forms for arbitrary number fields as
products of linear polynomials. arXiv:1307.7641
[3] T. Browning, L. Matthiesen and A.N. Skorobogatov. Rational points on pencils
of conics and quadrics with many degenerate fibres. Ann. of Math. 180 (2014)
381–402.
[4] J.-L. Colliot-Thélène. Variétés presque rationnelles, leurs points rationnels et
leurs dégénérescences. Arithmetic Geometry. P. Corvaja, C. Gasbarri, eds. Lec-
ture Notes Math. 2009, Springer-Verlag, 2011, pp. 1–44.
[5] J.-L. Colliot-Thélène and P. Salberger. Arithmetic on some singular cubic
hypersurfaces. Proc. London Math. Soc. 58 (1989) 519–549.
[6] J.-L. Colliot-Thélène and J.-J. Sansuc. Principal homogeneous spaces under
flasque tori: applications. J. Algebra 106 (1987) 148–205.
[7] J.-L. Colliot-Thélène and J.-J. Sansuc. La descente sur les variétés rationnelles,
II. Duke Math. J. 54 (1987) 375–492.
[8] J.-L. Colliot-Thélène, J.-J. Sansuc and Sir Peter Swinnerton-Dyer. Intersections
of two quadrics and Châtelet surfaces, I, II. J. reine angew. Math. 373 (1987)
37–168.
[9] J.-L. Colliot-Thélène, D. Harari and A.N. Skorobogatov. Valeurs d’un polynôme à
une variable représentées par une norme, Number theory and algebraic geometry,
 Descent on toric fibrations 435

London Math. Soc. Lecture Note Ser. 303. Cambridge University Press, 2003,
pp. 69–89.
[10] J.-L. Colliot-Thélène and A.N. Skorobogatov. Descent on fibrations over P1k
revisited. Math. Proc. Camb. Phil. Soc. 128 (2000) 383–393.
[11] U. Derenthal, A. Smeets and D. Wei. Universal torsors and values of quadratic
polynomials represented by norms. Math. Annalen 361 (2015), 1021–1042.
[12] B. Green and T. Tao. Linear equations in primes. Ann. of Math. 171 (2010)
1753–1850.
[13] B. Green and T. Tao. The Möbius function is strongly orthogonal to nilsequences.
Ann. of Math. 175 (2012) 541–566.
[14] B. Green, T. Tao and T. Ziegler. An inverse theorem for the Gowers U s+1 [N ]-
norm. Ann. of Math. 176 (2012) 1231–1372.
[15] D. Harari, Méthode des fibrations et obstruction de Manin. Duke Math. J. 75
(1994), 221–260.
[16] D. Harari. Flèches de spécialisation en cohomologie étale et applications arith-
métiques. Bull. Soc. Math. France 125 (1997) 143–166.
[17] Y. Harpaz, A.N. Skorobogatov and O. Wittenberg. The Hardy–Littlewood
conjecture and rational points. Compositio Math. 150 (2014), 2095–2111.
[18] R. Heath-Brown and A.N. Skorobogatov. Rational solutions of certain equations
involving norms. Acta Math. 189 (2002) 161–177.
[19] J.S. Milne. Étale cohomology. Princeton University Press, 1980.
[20] J.-J. Sansuc. Groupe de Brauer et arithmétique des groupes algébriques linéaires
sur un corps de nombres. J. reine angew. Math. 327 (1981) 12–80.
[21] D. Schindler and A.N. Skorobogatov. Norms as products of linear polynomials.
J. London Math. Soc. 89 (2014) 559–580.
[22] A.N. Skorobogatov. Arithmetic on certain quadric bundles of relative dimension
2, I. J. reine angew. Math. 407 (1990) 57–74.
[23] A.N. Skorobogatov. Descent on fibrations over the projective line. Amer. J. Math.
118 (1996) 905–923.
[24] A. Skorobogatov. Torsors and rational points. Cambridge University Press, 2001.
[25] A. Smeets. Principes locaux-globaux pour certaines fibrations en torseurs sous un
tore. Math. Proc. Camb. Phil. Soc. 158 (2015), 131–145.
[26] M. Swarbrick Jones. A note on a theorem of Heath-Brown and Skorobogatov.
Quat. J. Math. 64 (2013) 1239–1251.
[27] Sir Peter Swinnerton-Dyer. Rational points on some pencils of conics with 6
singular fibres. Ann. Fac. Sci. Toulouse 8 (1999) 331–341.
[28] O. Wittenberg. Intersections de deux quadriques et pinceaux de courbes de genre
1. Lecture Notes Math. 1901, Springer-Verlag, 2007.
 18
On filtrations of vector bundles over P1Z
A. Smirnov∗
St. Petersburg Department of Steklov Math. Institute
E-mail address: [email protected]

Introduction
A positive definite lattice may be regarded as a vector bundle on a compact-
ification of Spec Z, that is on an arithmetic curve. From this point of view, the
number [−1/2 log disc ] can be thought of as the first Chern class, and the
short vectors of the lattice correspond to the global sections.
On the other hand, there is a well-developed theory of vector bundles on
algebraic surfaces. One could study, for example, the discrete invariants and
the moduli spaces of such bundles. Some very interesting questions relate to
the exceptional collections of bundles on algebraic surfaces.
The study of vector bundles on arithmetic surfaces, for instance on P1Z ,
seems to be a rather promising enterprise and may be regarded as a synthesis
of two directions above. The “right” problem is, of course, a study of bundles
on some compactifications of arithmetic surfaces, for instance, on Arakelov’s
compactifications. As an approach to this problem, one may study vector
bundles on the classical scheme P1Z . We shall make a few steps in this direction
here.
Theorems 18.16, 18.18, and 18.24 are the main results of this paper. The first
theorems give a complete classification of the rank 2 bundles with the trivial
generic fiber and simple jumps. The third of these theorems gives a criterion,
allowing to determine whether the global restrictions for a minimal filtration of
a bundle with linear quotients can be reduced to local ones. We give examples,
showing that it is insufficient to know the local restrictions.
Parts of this paper were written during the program “Arithmetic and
Geometry 2013” at the Hausdorff Institute for Mathematics in Bonn. I thank

∗ Partially supported by RFFI (grant 13-01-00429 ).

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

436
 On filtrations of vector bundles over P1Z 437

the organizers of this program for their invitation and hospitality. It is a pleasure
for me to thank Boris Moroz for useful discussions.

18.1 General considerations

We shall study vector bundles over P1A for a Dedekind domain A, in particular
for A = Z. It is instructive and more natural, however, to state a few results in
greater generality. Thus let A be a Noetherian commutative ring.
We write usually O and O(d) instead of O X and O X (d) for a suitable
scheme X , especially if X can be easily specified. As usual, let

P1A = Proj A[t0 , t1 ], (deg t0 = deg t1 = 1).

In addition, O(U0 ) = Z[x], O(U1 ) = Z[x −1 ], and O(U01 ) = Z[x, x −1 ],

where x = t1 /t0 , Ui denotes the complement to the zero locus of ti , and U01 =
U0 ∩ U1 .
To construct vector bundles on X = P1A , let us define the following gluing
operation.

18.1.1 Gluing
Let σ ∈ GLn (A[x, x −1 ]). To σ one associates a vector bundle E of rank n over
P1A , as follows: E|U0 = Oe1 + · · · + Oen , E|U1 = O f 1 + · · · + O f n , and

[e1 , . . . , en ]σ = [ f 1 , . . . , f n ]
n
over U01 , so that f j = i=1 σi, j ei .
If every projective A-module is free, by the Quillen–Suslin theorem [6], [8],
any vector bundle of rank n over P1A can be obtained in that way. In this case,
the isomorphism class of such a bundle is an element of the double quotient

Vectn (P1A ) = GLn A[x] GLn A[x, x −1 ] GLn A[x −1 ]. (18.1)

Example 18.1. The line bundle O(d) can be given by the matrix σ = [x d ] ∈
GL1 (Z[x, x −1 ]).

Example 18.2. Let A = Z p and E be given by the matrix

 −1 
x p
σ = .
0 x
This bundle is not isomorphic to a sum of line bundles (see 18.1.3).
438 A. Smirnov

18.1.2 Jumps
When studying vector bundles on P1A it is important to understand the phe-
nomenon of jumps. Let us start by stating a few general results of independent
interest.

Theorem 18.3 (Grothendieck, [5]). Let F be a field. Any vector bundle on P1F
is isomorphic to a sum of line bundles with uniquely defined summands.

Theorem 18.4 (Serre, [7]). Any line bundle on PnA is isomorphic to a bundle
of the form p∗ L ⊗ O(d), where L is a line bundle on Spec A and p : PnA →
Spec A is the structure projection.

In particular, any line bundle on P1F is isomorphic to O(d) for some d ∈ Z.

18.1.2.1 Splitting types

Let E be a vector bundle over P1A , r = rk E, and y ∈ Spec A. Let E y be the
pullback of E to P1F , where F is the residue field at y. By the theorems 18.3
and 18.4, the pullback E y is isomorphic to the sum
E y = O(d1 ) + · · · + O(dr ),
where the collection d1 ≤ · · · ≤ dr is uniquely defined. This collection is
called the splitting type of E at y.

Example 18.5. To illustrate the notion of a splitting type, let us use the bundle
E of (18.2). Let y be the closed point and η the generic point Spec Z p . In this
example, E y is obtained by gluing with the matrix
 −1   −1 
x p x 0
σy = = ∈ GL2 (F p [x, x −1 ]).
0 x 0 x
Since σ y is diagonal it is clear that E y ∼
= O(−1) + O(1). On the other hand,
the residue field at the point η coincides with Q p and p is invertible in this
field. Therefore we have
     
1 0 x −1 p 0 −p 1 0
= .
−x/ p 1 0 x 1/ p 1/x 0 1

Thus (18.1) shows that E η ∼

= O2 .

18.1.2.2 Variations of splitting types

To describe the character of these variations we need a function and an order
on its value domain. Fix a bundle E on P1A and consider the following function
 On filtrations of vector bundles over P1Z 439

s(E, y) with values in Zr , where r = rk E. It maps y ∈ Spec A to the splitting

type of the bundle E y . Let

s(E, y) = (d1 (E, y), . . . , dr (E, y)).

Consider the lexicographic order on the set of the nondecreasing collections

(d1 , . . . , dr ), so that (d1 , . . . , dr ) < (e1 , . . . , er ) if dr < er , or dr = er and
dr −1 < er−1 , etc. For (d1 , . . . , dr ) ∈ Zr , the set

{y ∈ Spec A|s(E, y) ≥ (d1 , . . . , dr )}

is Zariski closed, i.e. the splitting type can not decrease under specializations.
Moreover, the sum d1 (E, y) + · · · + dr (E, y) doesn’t depend on y.
These assertions follow easily from the base change theorem [3, Ch. III,
§12].

18.1.2.3 Consequences of the base change theorem

Let us state the necessary results. Let f : X → Y be a proper morphism
of Noetherian schemes and F be a coherent O-module, flat over Y . As usu-
ally, X y is the fiber of f over y ∈ Y and F y = F | X y . Then y → χ (F y ) is
a locally constant function on Y and y → dimk(y) H p (X y , F y ) is an upper
semicontinuous function. Moreover, if Y is reduced and connected, then

(1) the function y → dimk(y) H p (X y , F y ) is constant on Y iff R p f ∗ F is

locally free and the natural arrow R p f ∗ F ⊗OY k(y) → H p (X y , F y ) is an
isomorphism for all y ∈ Y ;
(2) if the equivalent conditions of the previous item hold, then the natural mor-
phism R p−1 f ∗ F ⊗OY k(y) → H p−1 (X y , F y ) is an isomorphism for all
y ∈ Y.

Example 18.6. In the example 18.5 the underlying topological space of

Spec A is the set {y, η} with

s(E, y) = (−1, 1), and s(E, η) = (0, 0).

Thus the splitting type of E at y “jumps” relatively to the splitting type at η.

18.1.3 Linear filtrability of bundles

Any vector bundle on the projective line over a field is the sum of linear bun-
dles. Moreover, the splitting type of the sum of linear bundles on the projective
line over a Dedekind ring is obviously constant. However, the splitting type of
440 A. Smirnov

the bundles in the examples 18.2, 18.5, and 18.6 is not constant and therefore
those vector bundles are not isomorphic to sums of linear bundles.
It is an open question whether any vector bundle on P1A for a Dedekind ring
A admits a filtration with linear bundles as the quotients. Let us cite a few
known results in this direction.

Theorem 18.7 (Horrocks, [4]). If A is a local Dedekind ring, then any vector
bundle on P1A admits a filtration with linear bundles as the quotients.

The following result is valid for arbitrary Dedekind ring A.

Theorem 18.8 (Hanna, [2]). Any vector bundle on P1A admits a filtration with
linear and 2-dimensional bundles as the quotients.

The question is answered completely for any Euclidian ring.

Theorem 18.9 (Hanna, [2]). Any vector bundle over P1A , where A is an
Euclidian ring, admits a filtration with linear bundles as the quotients.

In particular, any bundle on P1Z admits a filtration with linear bundles as the
quotients.

18.1.4 Beilinson’s spectral sequence

Some of the methods of constructing bundles on PnC (see [5]) can be applied to
P1A . For example, it concerns two Beilinson’s spectral sequences.

Theorem 18.10 (Beilinson, [5]). Let F be a vector bundle on P1A and let p :
P1A → Spec A be the structure projection. There exists a spectral sequence
with E 1 = Rp∗ (F( p)) ⊗ − p ( p), which converges to
pq q


F, if i = 0;
F =
i
0, if i = 0.

In particular, the E 1 -term of this sequence is concentrated in the second

quadrant. Moreover, its nontrivial part is concentrated in the first two rows:

d1 / H1 (F) ⊗ O .
H1 (F(−1)) ⊗ O(−1)

d1 / H0 (F) ⊗ O
H0 (F(−1)) ⊗ O(−1)
 On filtrations of vector bundles over P1Z 441

18.1.4.1 Normalizations
The theorem (18.10) is especially useful when H1 (F(−1)) = H1 (F) = 0 and
both H0 (F(−1)) and H0 (F) are not too large.
A bundle F is called normalized if H 1 (F(−1)) = 0 and H 1 (F(−2)) = 0.
The base change theorem (see 18.1.2.3) implies that for a Dedekind ring A
there exists one and only one normalized bundle of the form F(d). Let δ(F) =
d in that case; for example, δ(O) = 0.

18.2 Bundles of rank 2 with simple jumps

Let A be a Dedekind ring. We shall study bundles E with rk E = 2, with the
trivial generic fiber, and with simple jumps. In such a case E η ∼
= O2 for the
∼ ∼
generic point η ∈ Spec A and E y = O or E y = O(−1) + O(1) for each
2

y ∈ Spec A. It follows from these conditions that the jumping points, i.e. the
points y with E y ∼= O(−1) + O(1), form a finite set (see 18.1.2.2).

18.2.1 Classification of the bundles with simple jumps

First of all we construct some bundles of rank 2 with simple jumps. Such a
bundle will be denoted by

F ν,ε1 ,ε2

with (ν, ε1 , ε2 ) ∈ A3 . The triples (ν, ε1 , ε2 ) are subjected to the only condition
that ν and ε1 are coprime. In other words, the row [ν, ε1 ] is supposed to be
unimodular.
We shall see that any 2-bundle with simple jumps can be obtained in this
way. If the ring A is factorial, we can restrict ourselves to the triples with ε2 =
0 (see 18.16). In that case, theorem 18.18 solves the classification problem.

18.2.1.1
We construct the bundle F ν,ε1 ,ε2 as the cokernel of an arrow of the form

φ : O2 (−2) −→ O4 (−1). (18.2)

Let e1 , e2 and f 1 , . . . , f 4 be the standard bases of O 2 and O4 .

The choice of bases fixes an identification Hom (O2 (−2), O4 (−1))
M4,2 (Hom (O, O(1))). Since Hom (O, O(1)) = At0 + At1 , we have φ =
t0 φ0 + t1 φ1 with φ0 , φ1 ∈ M4,2 (A).
442 A. Smirnov

18.2.1.2
The following particular case of φ (see (18.2)) is of special interest.
     
M 1 ν 0
φ0 = , φ1 = 2 , N (ν) = , (18.3)
N (ν) 02 0 1

where M ∈ M2,2 (A), ν ∈ A. In this case Coker φ does not depend on the 2nd
column of M. This can be easily proved by means of the elementary transfor-
mations. They adjust the first two rows of φ by means of the 4th row. Thus
without loss of generality we may assume that
 
ε 0
M = M(ε1 , ε2 ) = 2 . (18.4)
ε1 0

18.2.1.3
The arrow φ in (18.2) is called nondegenerate if F = Coker φ is a bundle
of rank 2. This is equivalent to the assertion that φ can be split locally. For a
nondegenerate φ we have an exact sequence
φ
0 / O2 (−2) / O4 (−1) /F /0. (18.5)

In particular, for the nondegenerate φ there is a canonical isomorphism

H 0 (X, F) → H 1 (X, O(−2))2 A2 .

It arises from the cohomology sequence for (18.5) and the standard identifica-
tion of O(−2) with the dualizing sheaf.

Proposition 18.11. The map φ of (18.2.1.2) is nondegenerate if and only if the

row [ν, ε1 ] is unimodular.

Proof. The map φ is nondegenerate iff the pullbacks of φ to U0 and U1 are

nondegenerate. It is easy to see that φ|U0 is nondegenerate iff the map A[x] →
A[x]3 , 1 → (ν, ε1 , ε2 + x) is an injection with the projective cokernel. This
happens if and only if the inclusion A[x] → A[x]3 splits, that is if and only if
the row [ν, ε1 , ε2 + x] is unimodular. On taking x = −ε2 we see that this is
equivalent to the unimodularity of [ν, ε1 ].
Conversely, suppose that αν + βε1 = 1. It is easy to see that the projectivity
of F|U1 is equivalent to the unimodularity of the row [νy, ε1 y 2 , 1 + ε2 y] in
A[y], where y = t0 /t1 . We show step by step, that y 2 = (αy)·(γ y)+β ·(ε1 y 2 ),
y = y ·(1+ε2 y 2 )−ε2 · y 2 , and 1 = 1·(1+ε2 y)−ε2 · y are linear combinations
of νy, ε1 y 2 and 1 + ε2 y.
 On filtrations of vector bundles over P1Z 443

Thus we have associated to any unimodular row [ν, ε1 ] and any ε2 ∈ A a

vector 2-bundle
F ν,ε1 ,ε2 = Coker φ, (18.6)

where the arrow φ in (18.5) is given by (18.2.1.2).

Proposition 18.12. If A is a field, then


ν,ε1 ,ε2 O2 , if ν = 0;
F
O(−1) + O(1), if ν = 0.

Proof. The exactness of (18.5) shows that Det F O and F O(−d)+O(d)

for some d ≥ 0. The cohomology sequence for (18.5) shows that h 0 (X, F) = 2
and d ≤ 1. To distinguish between the cases d = 0 and d = 1 let us consider
the cohomology sequence for (18.5) twisted by O(−1), that is the sequence

0 → H 0 (F(−1)) → H 1 (O(−3))2 → H 1 (O(−2))4 → H 1 (F(−1)) → 0,

where the scheme X is omitted in the notation for cohomology.

We have to calculate the middle arrow H 1 (φ(−1)). It is more conve-
nient to calculate the adjoint arrow H 0 ([φ(−1)]∨ ⊗ K X ) : H 0 (X, O)4 →
H 0 (X, O X (1))2 , where K X O(−2) is the canonical class. It suffices to note
that H 1 (ψ)∨ = H 0 (ψ ∨ ) and, consequently, the arrow H 0 ([φ(−1)]∨ ⊗ K X )
is given by multiplication with the adjoint matrix φ ∗ ∈ M2,4 (Hom (O, O(1))).
A direct calculation shows that the matrix of φ ∗ with respect to the bases
f 1∗ , . . . , f 4∗ and t1 e1∗ , t1 e2∗ , t0 e1∗ , t0 e2∗ has the form:
⎡ ⎤
1 0 0 0
⎢0 1 0 0⎥
⎢ ⎥.
⎣ε2 ε1 ν 0⎦
0 0 0 1

In particular, H 0 (X, F(−1)) = H 1 (X, F(−1)) = A/(ν).

Corollary 18.13. If A is a domain and ν = 0, then the generic fiber of F ν,ε1 ,ε2
is isomorphic to O2 , all the jumps have the form O(−1) + O(1) and lie exactly
over the divisors of ν.

Corollary 18.14. Let F = F ν,ε1 ,ε2 . Then H 0 (X, F(−1)) H 1 (X, F(−1))
A/(ν). In particular, the ideal, generated by ν, depends on the isomorphism
class of F only.
444 A. Smirnov

Proof. This statement is proved in the course of the proof of Proposition 18.12.

18.2.1.4
Further, let us assume that A is a factorial Dedekind ring. This assumption
allows us to describe explicitly the orbits of the action of GLn (A) on An and
the orbits of the action of certain congruence subgroups on P1 (A). Let K be
the fraction field of A. For a, b ∈ A we write c = (a, b) for any c with
Aa + Ab = (c).
To ν ∈ A one associates the following congruence subgroup
 
˜ 0 (ν) = { α β ∈ GL2 ( A)| γ = 0 (mod ν)}.
γ δ

For v ∈ K 2 \ {0}, let v̄ = K v, v̄ ∈ P1 (K ). Since A is a Dedekind ring, we

have P1 (K ) = P1 (A), so that v̄ ∈ P1 (A). We denote the line 0 × K by ∞.
 
v0
Lemma 18.15. Let v = ∈ A2 be a nontrivial vector. Then v̄ and ∞
v1
belong to the same orbit for the standard (left) action of ˜ 0 (ν) on P1 (A) if and
only if v1 /(v0 , v1 ) and ν are coprime.

Proof. Replacing v by v/(v0 , v1 ) we can assume that (v0 , v1 ) = 1 (here we

use the assumption
  18.2.1.4).
 The orbit
 of ∞ consists of the projectivizations
β α β
of the vectors , where ∈ ˜ 0 (ν) and, in particular, (γ , δ) = 1.
δ γ δ
Since γ = 0 (mod ν), it follows that (ν, δ) = 1.
Conversely, let (v1 , ν) = 1. The condition (v0 , v1 ) = 1 implies the condition
(νv0 , v1 ) = 1. Let αv1 − γ0 νv0 = 1. Then
     
v0 0 α v0
=σ with σ = .
v1 1 νγ0 v1

It remains to note that σ ∈ ˜ 0 (ν).

Theorem 18.16. Let F be a vector 2-bundle on P1A . Assume that its generic
fiber FK on P1K is isomorphic to O2 and its jumps have the form O(−1)+O(1).
Then F is isomorphic to a bundle of the form F ν,ε,0 with coprime ν and ε.

Proof. Taking into account the base change theorem (see 18.1.2.3), the form of
the generic fiber and the form of the jumps it is easy to see that H 1 (P1 , F)
H 1 (P1 , F(−1)) 0, H 0 (P1 , F) A2 , H 0 (P1 , F(1)) A4 . Applying to
 On filtrations of vector bundles over P1Z 445

F(1) the Beilinson theorem (see 18.10) and a twist, we get F as a quotient in
the exact sequence
φ
0 → O2 (−2) −
→ O4 (−1) → F → 0, (18.7)

where φ = t0 φ0 + t1 φ1 and φ0 , φ1 ∈ M4,2 (A). We can identify the bundles,

constructed by means of equivalent matrices, where φ ∼ φ  , iff φ  = ρφσ −1 ,
ρ ∈ GL4 (A), σ ∈ GL2 (A).
Let us show first that any nondegenerate φ (see 18.2.1.3) is given by a matrix
of the form (18.2.1.2). Indeed, the pullback of (18.7) to t0 = 0 shows that
rk φ1 = 2. The theory of elementary divisors (here we use the assumption
18.2.1.4) allows to suppose that φ ∼ φ (1) with
 
(1) 1
φ1 = 2 ∈ M4,2 (A).
02
(1)
The stabilizer of φ1 in GL4 (A) × GL2 (A) consists of the pairs (ρ, α −1 )
 
α β
with ρ = . The theory of elementary divisors implies that φ (1) ∼ φ (2) ,
0 δ
where
   
(2) (1) (2) M ν1 ν 0
φ1 = φ1 , φ0 = , and N = .
N 0 ν1

Since φ is nondegenerate, it follows that ν1 ∈ A∗ . Indeed, otherwise we

(2)
could find a prime π dividing ν1 . But φ0 has only one 2-minor, which
can be nontrivial mod π. This minor, being a homogeneous polynomial of
degree 2, has zeros on P1 ⊗A/(π ). This contradicts to the assumption that φ
is nondegenerate.
Without loss of generality we can assume that
     
1 M ν 0
φ (1) ∼ φ (2) = t1 + t0 with N (ν) = ,
0 N (ν) 0 1
that ν = 0 and that M = M(ε1 , ε2 ); this follows from the invertibility of ν1 ,
(18.2.1.2) and the conditions on the generic fiber of F.
The stabilizer of φ (2) in GL4 (A) × GL2 (A) consists of the pairs (ρ, α −1 )
with δ N (ν)α −1 = N (ν). The integrality of δ is equivalent to the condition
that α ∈ ˜ 0 (ν). It remains to show that for any matrix M = M(ε1 , ε2 ) with
(ν, ε1 ) = 1, we can solve the equation α Mα −1 + β N (ν)α −1 = M(ε, 0) or, in
other words, the equation
     
ε2 0 ν 0 0 0
α +β = α. (18.8)
ε1 0 0 1 ε 0
446 A. Smirnov

Here α = (αi, j ) ∈ ˜ 0 (ν), β = (βi, j ) ∈ M2,2 (A) and ε ∈ A are the unknowns.
 
˜ ε
First of all, we use Lemma 18.15 and choose α ∈ 0 (ν) such that α 2 =
ε1
 
0
λ for some λ ∈ A. This is possible because ν and ε1 are coprime (see
1
18.11). The equation (18.8) takes the form:
     
0 0 νβ1,1 β1,2 0 0
+ = .
λ 0 νβ2,1 β2,2 εα1,1 εα1,2
It remains to find β and ε. Let β1,1 = β1,2 = 0 and choose β2,1 and ε to satisfy
the equation λ = εα1,1 − ξ2,1 ν (it is possible since (α1,1 , ν) = 1). Then take
β2,2 = εα1,2 .

We shall describe morphisms between the bundles in (18.16). In the

following two results it sufficies to assume that A is a domain.

Proposition 18.17. Let F = F ν,ε,0 , G = F μ,ζ,0 with (ν, ε) = 1, (μ, ζ ) =

1, ν = 0, μ = 0. The functor H 0 and the canonical isomorphisms
H 0 (X, F) A2 and H 0 (X, G) A2 (see 18.2.1.3) identify Hom O (F, G)
with the set of those θ in M2,2 ( A), which satisfy the following conditions:
N (μ)θ N (ν)−1 ∈ M2,2 (A), (M(ζ, 0)θ − θ M(ε, 0))N (ν)−1 ∈ M2,2 (A). Here
N (ν)−1 ∈ M2,2 (K ). That identification turns the composition into the product.

Proof. The functoriality of the Beilinson spectral sequence (see 18.10)

reduces the computation of Hom O (F, G) to a description of the commutative
diagrams of the form
φ
O2 (−2) / O4 (−1) /F /0,

θ λ
 ψ

O2 (−2) / O4 (−1) /G /0

where φ and ψ are the arrows in the definition of F and G. We have the
following commutativity equation
⎡ ⎤ ⎡ ⎤
t1 0 t 0
  1
⎢ ζ t0 t1 ⎥ ⎢ ⎥
⎢ ⎥ θ = λ1,1 λ1,2 ⎢εt0 t1 ⎥ .
⎣μt0 0 ⎦ λ2,1 λ2,2 ⎣νt0 0 ⎦
0 t0 0 t0
For t0 = 0 it is reduced to the equalities θ = λ1,1 , λ2,1 = 0. Taking them
into account, we see that for t1 = 0 the commutativity is equivalent to the
 On filtrations of vector bundles over P1Z 447

relations λ2,2 = N (μ)θ N (ν)−1 , λ1,2 = (M(ζ, 0)θ − θ M(ε, 0))N (ν)−1 . Thus
λ is uniquely determined by θ ; moreover, λ1,2 ∈ A and λ2,2 ∈ A.

Theorem 18.18. Let F = F ν,ε,0 , G = F μ,ζ,0 with (ν, ε) = 1, (μ, ζ ) = 1,

ν = 0, μ = 0. The vector bundles F and G are isomorphic iff (ν) = (μ) and
there are η ∈ A∗ and λ ∈ A such that
ζ λ2 ≡ εη (mod ν).

Proof. From the relation F G it follows that (μ) = (ν) (see 18.14). If the
isomorphism F → G is given by θ ∈ GL2 (A), as defined in Prop. 18.17,
then θ and θ −1 satisfy the integrality relations in that Proposition. It can be
proved by a calculation that θ = (θi, j ) defines an isomorphism iff θ2,1 ≡ 0,
εθ2,2 ≡ ζ θ1,1 , εθ1,2 ≡ 0, ζ θ1,2 ≡ 0 mod ν. Since ε and ζ are invertible
modulo (ν), it follows that
θ2,1 ≡ θ1,2 ≡ 0 (mod ν) and εθ2,2 ≡ ζ θ1,1 (mod ν). (18.9)
Let η = det θ ∈ A∗ . Then η ≡ θ1,1 θ2,2 ≡ ζ ε −1 θ1,1
2 (mod ν). This proves that

if F G, then (μ) = (ν) and there are η ∈ A∗ and λ ∈ A with ζ λ2 ≡ εη

(mod ν). Conversely, suppose μ, ν, λ, and ε satisfy those conditions and let
χ ∈ A such that
χ ε ≡ 1 (mod ν). (18.10)
Then it follows from the relation ζ λ2 ≡ εη (mod ν) that η = ζ χ λ2 + κν with
κ ∈ A. Let ω, τ ∈ A so that ωλ = κ + τ ν. Let θ1,1 = λ, θ2,2 = ζ χ λ + ων,
θ2,1 = τ ν, θ1,2 = ν. Then θ induces an isomorphism F → G. The condition
θ ∈ GL2 (A) follows from the definitions of θ, τ , ω, κ and the relation det θ ∈
A∗ . It remains to prove (18.9). The relations θ2,1 ≡ θ1,2 ≡ 0 (mod ν) and
εθ2,2 ≡ ζ θ1,1 follow from the definition of θ and (18.10).

Example 18.19. Theorems 18.16 and 18.18 classify completely the 2-bundles
with the trivial generic fiber and simple jumps in the case of a factorial
Dedekind ring A.
For A = C[t] the isomorphism classes of vector bundles in question can be
identified with the effective divisors on A1 .
Let A = R[t]. If ν is an irreducible polynomial then there is only one iso-
morphism class of bundles related to ν, independently on deg ν. This is related
to the fact that for degree 1 the negative residues are represented by global
units. However, if ν = p1n 1 · · · pen d q1m 1 · · · qdm d with irreducible pi of degree
1 and irreducible qi of degree 2, then for e ≥ 1 there are 2e−1 isomorphism
classes of bundles with this ν.
448 A. Smirnov

We have a similar, but slightly more interesting picture for A = Fq [t] and
A = Z. For example, in the case A = Z there are two bundles with ν = 5,
because the only nontrivial global unit (−1) is a square mod ν (the same is
true for any prime ν ≡ 1 (mod 4)), and only one bundle with ν = 7 (the same
is true for any prime ν ≡ 3 (mod 4)).

18.3 Minimal linear filtrations

Let E be a vector bundle on P1A such that there exists at least one filtration of E
with linear quotients. We are interested in filtrations with simplest quotients.
We shall study this problem in the special case when rk E = 2 and A is a
factorial Dedekind ring. By Serre’s theorem (see 18.4) each linear quotient is
isomorphic to O(d) for some d in this case.

18.3.1 Amplitudes of bundles

Assume additionally that the generic fiber of E is trivial, i.e. that E η ∼
= O2 .
Then for each filtration with linear quotients
0 = E0 ⊂ E1 ⊂ E2 = E (18.11)
the line bundles E 1 and E/E 1 are dual each to other.
Indeed, the linear bundle Det E ∼ = O(d) for some integer d, by the factori-
ality of A and Serre’s theorem (see 18.4). The pullback of this isomorphism to
the generic fiber and the triviality of the generic fiber E η show that d = 0 and
therefore Det E ∼= O. But then E 1 ⊗ E/E 1 ∼ = Det E ∼ = O. This isomorphism
gives the claimed duality.
Thus, for any filtration of E of the form (18.11), there is an integer d such
that
E/E 1 ∼= O(d) and E 1 ∼ = O(−d).
Moreover, d ≥ 0. Indeed, the filtration (18.11) induces an element in
Ext1O (E/E 1 , E 1 ) = Ext1O (O(d), O(−d));
for d < 0 this group is trivial and consequently E ∼ = O(−d) + O(d). This
contradicts the triviality of the generic fiber, however.

Definition 18.20. Let E be a vector bundle on P1A with rk E = 2 and E η ∼ =

O2 . The minimal non-negative integer d, for which there exists a filtration on
E of the form (18.11) with E/E 1 ∼= O(d), is called the amplitude of E. The
amplitude of E is denoted by a(E).
 On filtrations of vector bundles over P1Z 449

18.3.2 Local amplitudes

Let A be a local ring. In this case, the amplitude can be easily computed and
is determined by the splitting type at the generic point. This follows from the
following complement to the Horrocks result (see 18.7).

Proposition 18.21. Let A be a local Dedekind ring, E be a vector bundle on

P1A , y be the closed point of Spec A, and E y = O(d1 ) + · · · + O(dn ) with
d1 ≤ · · · ≤ dn . Then there exists a filtration 0 = E 0 ⊂ E 1 ⊂ · · · ⊂ Er = E,
where E i is a vector bundle and E i /E i−1 ∼ = O(di ) for i = 1, . . . , r .

Proof. By induction, it suffices to find an inclusion O(d1 ) → E such that

E/O(d1 ) is a bundle. We can twist E and, without loss of generality, assume
that d1 = 0. Let H n (E) = R n p∗ E, where p : P1A → Spec A is the structure
projection. Let F be the residue field at y. Making use of the base change
theorem (see 18.1.2.3), one can prove that H 1 (E) = 0 and the pullback
H 0 (E) ⊗ A F → H 0 (E y ) is an isomorphism. Thus we can choose s ∈ H 0 (E),
identifying O with a summand of E y . Then the quotient of s : O → E is
locally free.

18.3.3 Local restrictions for amplitudes

Let
aloc (E) = max a(E y ),
y

where y runs over the closed points of Spec A. It follows from the definition of
the amplitude that
a(E) ≥ aloc (E). (18.12)

18.3.4 Bundles without jumps

Proposition 18.22. Let A be a Dedekind ring and let E be a vector bundle
on P1A . If the function y → d1 (E, y) (see 18.1.2.1) is constant on the set of
the closed points of Spec A, then there is an inclusion O(d) → E such that
E/O(d) is a vector bundle, where d = d1 (E, y) for each closed point y.

Proof. Essentially this is proved in the course of the proof of Proposi-

tion 18.21.

It follows from Proposition 18.22 that E is a trivial bundle if aloc (E) = 0.

450 A. Smirnov

18.3.5 Bundles with simple jumps

Let E be a vector bundle with aloc (E) = 1. As we shall see, the local data, i.e.
the local amplitudes, do not determine the global amplitude. Let

E = F β,α,0 = Coker [O(−2)2 −

→ O(−1)4 ],

where the arrow is given by the matrix

⎡ ⎤
t1 0
⎢ α t1 ⎥
φ=⎢ ⎥ / A∗ .
⎣βt0 0 ⎦ , (β, α) = 1, β = 0, β ∈ (18.13)
0 t0
Since (α, β) = 1, the sheaf E is a bundle of rank 2 (see 18.11). It follows
from the condition β = 0 that the generic fiber is trivial and the jumps are
simple. The condition β ∈/ A∗ ensures the existence of jumps (see 18.13).

18.3.5.1 Gluing matrices

To compute Im φ and the corresponding quotient let us introduce the following
notation. Denote by e1 , e2 the standard O-basis for O2 and by f 1 , f 2 , f 3 , f 4
the standard O-basis for O4 . Let γ , δ ∈ A and

αδ − βγ = 1. (18.14)

In the standard bases on U0 , the arrow φ is given by

⎡ ⎤
x 0
⎢α x⎥
φ|U0 = t0 ⎢ ⎣β
⎥.
0⎦
0 1

Let the basis of Im φ on U0 consist of φ(t0−2 e1 ) and φ(t0−2 e2 ), that is of the

columns of φ|U0 . Extend φ|U0 to an invertible matrix by letting
⎡ ⎤
x 0 0 1
⎢ α x γ 0⎥

φ| ⎢
U0 = t0 ⎣
⎥ , det φ| U0 = t0 .
4
β 0 δ 0⎦
0 1 0 0

The OU0 -module O(−1)4 |U0 is thereby equipped with the basis

t0−1 [g1 , . . . , g4 ] = t0−2 [ f 1 , . . . , f 4 ]φ|U0 , (18.15)

where t0−1 [ḡ3 , ḡ4 ] is a basis of the quotient, i.e. a basis of E|U0 .
 On filtrations of vector bundles over P1Z 451

In the standard bases on U1 the arrow φ is given by

⎡ ⎤
1 0
⎢αy 1 ⎥
φ|U1 = t1 ⎢ ⎥ −1
⎣βy 0 ⎦ , where y = t0 /t1 = x .
0 y
Extend φ|U1 to an invertible matrix, by letting
⎡ ⎤
1 0 0 0
⎢αy 1 0 0⎥

φ| ⎢
U1 = t1 ⎣
⎥ , det φ|
 U1 = t1 .
4
βy 0 1 0⎦
0 y 0 1
The OU1 -module O(−1)4 |U1 is thereby equipped with the basis

t1−1 [h 1 , . . . , h 4 ] = t1−2 [ f 1 , . . . , f 4 ]φ|U1 , (18.16)
whereby t1−1 [h̄ 3 , h̄ 4 ] is a basis of the quotient, i.e. a basis of E|U1 .
Let us compute the gluing matrix on U01 . In order to do it, let us express
t1−1 h 3 and t1−1 h 4 in terms of t0−1 g1 , t0−1 g2 , t0−1 g3 , t0−1 g4 . Let us start by
expressing t0−1 gi , 1 ≤ i ≤ 4, in terms of t0−2 f j , 1 ≤ j ≤ 4, (see (18.15)).
It follows then that
⎡ ⎤
0 δ −γ −δx
−1 ⎢0 0 0 1 ⎥

φ| = t0−1 ⎢
⎣0 −β
⎥.
βx ⎦
U0
α
1 −δx γx δx 2
Relations (18.15) and (18.16) give:
−1 
t1−1 [h 1 , h 2 , h 3 , h 4 ] = x −2 t0−1 [g1 , g2 , g3 , g4 ]φ|
 U0 φ|U1 .

It follows that
⎡ ⎤
y −δ −γ −δx
−1 ⎢0 y 0 1 ⎥

φ|U0 φ| ⎢
U1 = x ⎣
⎥.
0 0 α βx ⎦
0 0 γx δx 2
Thus
 
' ( α βx −1 ' (
t1−1 h 3 h4 = x −2
x t g3 g4 mod (g1 , g2 ).
γx δx 2 0
Therefore the bundle E can be given by the gluing matrix
 
αy β
σ = , det σ = 1. (18.17)
γ δx
452 A. Smirnov

In other words, there are a basis [e1 , e2 ] of the pullback of E to U0 and a

basis [ f 1 , f 2 ] of the pullback of E to U1 with
 −1   
αx β δx −β
[e1 , e2 ] = [ f 1 , f 2 ] or [e1 , e2 ] = [ f 1 , f 2 ] on U01 .
γ δx −γ αx −1
Although E has the trivial generic fiber and simple jumps at the divisors of
β (see 18.12), we need an explicit identification of the corresponding fibers
with O2 and O(−1) + O(1).

18.3.5.2 Generic fibers

Let us fix an isomorphism E ⊗Z[1/β] O2 . Let [g1 , g2 ] be the standard basis
for O2 on U0 and [h 1 , h 2 ] be the standard basis for O2 on U1 , so that
   
1 0 1 0
[g1 , g2 ] = [h 1 , h 2 ] or [g1 , g2 ] = [h 1 , h 2 ] on U01 .
0 1 0 1

The isomorphism E ⊗ Z[1/β] O2 is given as follows:

   
δx −β 1 0
[g1 , g2 ] = [e1 , e2 ] and [ f 1 , f 2 ] = [h 1 , h 2 ].
1 0 −αβ −1 x −1 β −1

18.3.5.3 Special fibers

Let us fix an isomorphism E ⊗ A/β O(−1) + O(1). Choose [t0−1 g1 , t0 g2 ]
and [t1−1 h 1 , t1 h 2 ] as the bases of O(−1) + O(1) on U0 and U1 . Here [g1 , g2 ]
and [h 1 , h 2 ] are the standard bases for O2 on U0 and U1 . Then
 
x −1
0
[t0−1 g1 , t0 g2 ] = [t1−1 h 1 , t1 h 2 ] or [t0−1 g1 , t0 g2 ]
x0
 
−1 x 0
= [t0 h 1 , t0 h 2 ] on U01 .
0 x −1
The isomorphism E ⊗ A/(β) O(−1) + O(1) is given as follows:
   
−1 δ 0 1 0
[t0 g1 , t0 g2 ] = [e1 , e2 ] and [ f 1 , f 2 ] = [t1−1 h 1 , t1 h 2 ].
−γ x α 0 1
Let us check whether a(E) = 1, that is whether there is a filtration of the
form E 1 ⊂ E with E 1 ∼ = O(−1) and E/E 1 ∼ = O(1). The existence of such a
filtration of E is equivalent to the existence of a global nonvanishing section
of V = E(1).
In further calculations, take t0 [e1 , e2 ] and t1 [ f 1 , f 2 ] as bases of V on U0
and U1 , where [e1 , e2 ] is the basis of E on U0 and [ f 1 , f 2 ] is the basis of E
on U1 . Then
 On filtrations of vector bundles over P1Z 453

 
α βx
t0 [e1 , e2 ] = t1 [ f 1 , f 2 ]
γx δx 2
 
δ −βx −1
or t0 [e1 , e2 ] = t1 [ f 1 , f 2 ] on U01 .
−γ x −1 αx −2

18.3.5.4 Calculating H0 (V)

The generic U0 -section s = (s1 e1 + s2 e2 )t0 , where si ∈ A[x], can be written in
the U1 -basis as s = (δs1 −βx −1 s2 )t1 f 1 +(−γ x −1 s1 +αx −2 s2 )t1 f 2 . Therefore
the conditions for s to be a global section are as follows:

c1 : δs1 − βx −1 s2 ∈ A[x −1 ];
c2 : −γ x −1 s1 + αx −2 s2 ∈ A[x −1 ].

It can be easily seen that {x −1 s1 , x −2 s2 } ⊂ A[x −1 ]: it suffices to note that

γ x −1 c1 + δc2 = x −2 s2 . Therefore s1 = u 0 + u 1 x, s2 = v0 + v1 x + v2 x 2 . We
obtain an equation for u i and v j : δu 1 = βv2 . Since α and γ are coprime, then
the generic global section of V has the form:
s = (u 0 + βwx)t0 e1 + (v0 + v1 x + δwx 2 )t0 e2 , where u 0 , v0 , v1 , w ∈ A.
(18.18)

18.3.5.5 Nonvanishing over A[1/β]

Let us determine when the section s, defined by (18.18), vanishes nowhere on
P1 ⊗ A[1/β]. To this end, we shall use the isomorphism in 18.3.5.2 and regard
s as a section of O(1)2 . Choose t0 [g1 , g2 ] as a basis of O2 (1) on U0 , where
[g1 , g2 ] is the standard basis for O2 on U0 . The isomorphism V ⊗ Z[1/β]
O2 (1) (see 18.3.5.2) on U0 is given as follows:
 
δx −β
t0 [g1 , g2 ] = t0 [e1 , e2 ].
1 0
The generic section of O(1)2 is of the form: (a10 t0 + a01 t1 )g1 + (b10 t0 +
b01 t1 )g2 . Let us express the generic section of V in these terms: s = (u 0 +
βwx)t0 e1 + (v0 + v1 x + δwx 2 )t0 e2 = (−βv0 + [δu 0 − βv1 ]x)t0 g1 + (u 0 +
βwx)t0 g2 . Therefore
a10 = −βv0 , a01 = δu 0 − βv1 , b10 = u 0 , b01 = βw. (18.19)
Thus we have constructed a section of O(1)2 over A[β −1 ], i.e. a pair of
polynomials of degree 1, by means of s ∈ H 0 (V ). The existence of their
common zero can be determined by means of the resultant
 
a10 a01
R11 (a10 , a01 ; b10 , b01 ) = det = a10 b01 − a01 b10 .
b10 b01
454 A. Smirnov

Using (18.19) we obtain the following condition

− δu 20 + βu 0 v1 − β 2 v0 w ∈ A[β −1 ]∗ . (18.20)
The rows (u 0 , v0 , v1 , w) ∈ A4 , satisfying this condition, correspond to the
global sections of V without zeros outside of the divisors of β.

18.3.5.6 Nonvanishing over divisors of β

Let us determine when the section s, defined by (18.18), vanishes nowhere
on P1 ⊗ A/(π ), where π is a prime divisor of β. To this end, we shall use
the isomorphism of 18.3.5.3 and regard s as a section of O + O(2). Choose
[g1 , t02 g2 ] as a basis for O + O(2) on U0 , where [g1 , g2 ] is the standard basis
for O2 on U0 . The isomorphism V ⊗ A/π O + O(2) (see 18.3.5.3) on U0
is given as follows:
 
2 δ 0
[g1 , t0 g2 ] = t0 [e1 , e2 ].
−γ x α
The generic section of O + O(2) is of the form: a00 g1 + (b20 t02 + b11 t0 t1 +
b02 t12 )g2 . Let us express the generic section of V in these terms: s (mod π ) =
(u 0 + βwx)t0 e1 + (v0 + v1 x + δwx 2 )t0 e2 = (δu 0 + βδwx)g1 + (αv0 + [αv1 −
γ u 0 ]x + wx 2 )t02 g2 . In other words,
a00 = δu 0 , b20 = αv0 , b11 = αv1 − γ u 0 , b02 = w. (18.21)
Thus, by s ∈ H 0 (V ) we have constructed a section of O + O(2) over A/π ,
i.e. a pair of polynomials of degree 0 and 2. The existence of their common
zero can be determined by means of the resultant
 
a 0
R02 (a00 ; b20 , b11 , b02 ) = det 00 = a00
2
.
0 a00
Using (18.21) we obtain the condition δ 2 u 20 = 0 (mod π ) or
u0 = 0 (mod π ). (18.22)
The rows (u 0 , v0 , v1 , w) ∈ A4 , satisfying this condition, correspond to global
sections of V without zeros over A/(π ).

18.3.5.7 Conclusion
The existence of E 1 ⊂ E with E 1 ∼ = O(−1) and E/E 1 ∼ = O(1) is equiva-
lent to the existence of a row (u 0 , v0 , v1 , w) ∈ A4 , satisfying the following
conditions:

δu 20 − βu 0 v1 + β 2 v0 w ∈ A[β −1 ]∗
(18.23)
u 0 (mod β) ∈ (A/(β))∗ .
 On filtrations of vector bundles over P1Z 455

Of course, the solvability of these conditions can be verified by giving such a

row. However, if we fail to find a solution it is desirable to prove the insolvabil-
ity. We have already seen that there are no local obstructions to the solvability.
However, we can use some global information, namely the knowledge of units
in A[β −1 ], to modify slightly the system (18.23). After such a modification
local obstructions can arise.
Namely, let π1 , . . . , πr be all the prime divisors of β. The solvability of
(18.23) is equivalent to the existence of nonnegative integers n 1 , . . . , nr and of
θ ∈ A∗ such that the system
 n
δu 20 − βu 0 v1 + β 2 v0 w = θ π1 1 · · · πrnr .
(18.24)
v0 (mod β) ∈ ( A/(β))∗
is solvable. Moreover, it is clear that ordπ (δ) = 0 for any p, dividing β. There-
fore the second condition implies that n i = 0 for every i , so that the solvability
of the system (18.24) is equivalent to the existence of a global unit θ ∈ A∗
such that the equation
δu 20 − βu 0 v1 + β 2 v0 w = θ. (18.25)
is solvable.
Thus let the bundle E be given by the gluing matrix
 −1   
αx β α β
σ = , where ∈ SL2 (A).
γ δx γ δ
The existence of E 1 ⊂ E with E 1 ∼ = O(−1) and E/E 1 ∼ = O(1) is equiva-
∗
lent to the existence of θ ∈ A and of a row (u 0 , v0 , v1 , w) ∈ A4 , satisfying
(18.25). In other words, it is equivalent to the existence of θ ∈ A∗ such that
the equation (18.25) can be solved in A.

Example 18.23. Let A = Z and the bundle E be given by the gluing matrix
 −1 
2x 5
σ= .
1 2x
The equality a(E) = 1 is equivalent to the existence of a row (u 0 , v0 , v1 , w) ∈
Z4 with 2u 20 − 5u 0 v1 + 25v0 w = ±1. These equations are inconsistent because
±2 is not a square in Z/(5). It can be shown, however, that there exists a
filtration E 1 ⊂ E with E 1 ∼ = O(−2) and E/E 1 ∼ = O(2). In other words,
a(E) = 2.

Theorem 18.24. Let A = Z. Then a(E) = 1, iff either α (mod β) is a square

and the equation δu 20 − βu 0 v1 + β 2 v0 w = 1 can be solved in Z2 , or (−α
456 A. Smirnov

(mod β)) is a square and the equation δu 20 − βu 0 v1 + β 2 v0 w = −1 can be

solved in Z2 .

Proof. It follows from the equation αδ − βγ = 1 (see (18.17)) that εα

(mod β) is a square iff εδ (mod β) is a square, where ε ∈ {±1}. Therefore
it is enough to prove the theorem with δ instead of α. Further, a(E) = 1 iff
(see 18.3.5.7) at least one of the equations δu 20 − βu 0 v1 + β 2 v0 w = ±1 has a
solution in Z. If the equation δu 20 − βu 0 v1 + β 2 v0 w = 1 has a solution in Z,
then δ (mod β) is a square. And the same is true for the second equation. This
proves the “only if” part.
Conversely, suppose that δ (mod β) is a square and the equation δu 20 −
βu 0 v1 + β 2 v0 w = 1 has a solution in Z2 . We have to show the solvability of
the following equation

δu 20 − βu 0 v1 + β 2 v0 w = 1. (18.26)

We shall apply results of [1]. The definition of a quadratic form in [1, I,

§2] requires the evenness of the coefficients for the products. So consider the
equation

q(u 0 , v0 , v1 , w) = 2, where q(u 0 , v0 , v1 , w) = 2δu 20 − 2βu 0 v1 + 2β 2 v0 w.

(18.27)
It is obtained by multiplication (18.26) with 2. This does not change the solv-
ability. The form q is indefinite and Z-regular in the sense of [1], i.e. it is
nondegenerate over Q. There is an important result (the strong approximation
theorem is applied) concerning such forms [1, Ch.9, Th.1.5]. It says that if
rk q ≥ 4 (in our case rk q = 4), then an integer is represented by q over Z iff
it is represented by q over Z p for every p .
It remains to study the local solvability of (18.27). Since disc(q) = 16β 6 , we
have the local solvability outside the jumps and 2. It follows from the smooth-
ness, the existence of rational points on quadrics over finite fields for dim > 1
and Hensel’s lemma.
Let us study the local solvability at an odd prime p dividing β. If δ (mod p)
is a square, we can take v0 = v1 = w = 0 and use Hensel’s lemma. The
solvability in Z2 is assumed in the statement. The case when (−δ) (mod β) is
a square is similar.

Corollary 18.25. Let A = Z and E has no jump at 2. Then a(E) = 1 iff either
α (mod β) is a square, or (−α) (mod β) is a square.
 On filtrations of vector bundles over P1Z 457

Proof. By Theorem 18.24, it is enough to show that the equations δu 20 −

βu 0 v1 + β 2 v0 w = ±1 are solvable in Z2 if there is no jump at 2. The last
condition is equivalent (see 18.13) to the condition of β being odd. Let u 0 = 1
and v0 = w = 0. The equations in the question take the form δ − βv1 = ±1.
Their solution v1 = −(±1 − δ)/β lies in Z2 since β is invertible in Z2 .
Theorem 18.24 and Corollary 18.25 show when a(E) = 1 for A = Z and
E = F β,α,0 . Theorem 18.16 asserts that any bundle with the trivial generic
fiber and simple jumps is isomorphic to a bundle of the form F β,α,0 . Therefore
Theorem 18.24 and Corollary 18.25 allow to check whether a(E) = 1 for any
bundle with the trivial generic fiber and simple jumps.

References
[1] J.W.S. Cassels. Rational Quadratic Forms. Academic Press, 1978.
[2] Ch.C. Hanna. Subbundles of vector bundles on the projective line. J. Algebra, 52,
no. 2, 322–327, 1978.
[3] R. Hartshorne. Algebraic Geometry. Graduate Texts in Math. 52, Springer Verlag,
1977.
[4] G. Horrocks. Projective modules over an extension of a local ring, Proc. London
Math. Soc. (3) 14 (1964), 714–718.
[5] Ch. Okonek, M. Shneider, H. Spindler. Vector Bundles On Complex Projective
Spaces. Progress in Math 3, Birkhäuser, 1980.
[6] D. Quillen. Projective modules over polynomial rings, Invent. Math., 1976, vol.
36, p. 167–171.
[7] J.-P. Serre. Faisceaux algébriques cohérents. Ann. of Math., 61 (1955), 197–278.
[8] A. Suslin. Projective modules over polynomial rings are free, Doklady Akademii
Nauk SSSR (in Russian), 1976, 229 (5), p. 1063–1066. Translated in Soviet
Mathematics, 1976, 17 (4), p. 1160–1164.
 19
On the dihedral Euler characteristics of Selmer
groups of Abelian varieties
Jeanine Van Order
Einstein Institute of Mathematics, The Hebrew University of Jerusalem
E-mail address: [email protected]

A BSTRACT. This note shows how to use the framework of Euler characteris-
tic formulae to study Selmer groups of abelian varieties in certain dihedral or
anticyclotomic extensions of CM fields via Iwasawa main conjectures, and in
particular how to verify the p-part of the refined Birch and Swinnerton-Dyer
conjecture in this setting. When the Selmer group is cotorsion with respect to
the associated Iwasawa algebra, we obtain the p-part of formula predicted by
the refined Birch and Swinnerton-Dyer conjecture. When the Selmer group
is not cotorsion with respect to the associated Iwasawa algebra, we give a
conjectural description of the Euler characteristic of the cotorsion submodule,
and explain how to deduce inequalities from the associated main conjecture
divisibilities of Perrin-Riou and Howard.

Contents
19.1 Introduction 459
19.2 Ring class towers 465
19.3 Galois cohomology 467
19.4 The torsion subgroup A(K p∞ )tors 471
19.5 The definite case 472
19.6 The indefinite case 481
References 490
1991 Mathematics Subject Classification. Primary 11, Secondary 11G05, 11G10, 11G40, 11R23
Keywords and phrases: Iwasawa theory, abelian varieties, elliptic curves
The author acknowledges partial support from the Swiss National Science Foundation (FNS)
grant 200021-125291.

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

458
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 459

19.1 Introduction
Fix a prime number p. Let G be a compact p-adic Lie group without p-torsion.
Writing O to denote the ring of integers of some fixed finite extension of
Q p , let

(G) = lim O[G/U ]

← −
U

denote the O-Iwasawa algebra of G. Here, the projective limit runs over
all open normal subgroups U of G, and O[G/U ] denotes the usual group
ring of G/U with coefficients in O. Let M be any discrete, cofinitely gen-
erated (G)-module. If the homology groups Hi (G, M) are finite for all
integers i ≥ 0, then we say that the G-Euler characteristic χ (G, M) of M is
well-defined, and given by the (finite) value

|H i (G, M)|(−1) .
i
χ (G, M) =
i≥0

If G is topologically isomorphic to Zδp for some integer δ ≥ 1, then a clearer

picture of this Euler characteristic χ (G, M) emerges. In particular, a clas-
sical result shows that χ (G, M) is well defined if and only if H0 (G, M)
is finite. Moreover, if M in this setting is a pseudonull (G)-module, then
χ (G, M) = 1. Now, the structure theory of finitely generated torsion (G)-
modules shown in Bourbaki [4] also gives the following Iwasawa theoretic
picture of the G-Euler characteristic χ (G, M). That is, suppose M is a discrete,
cofinitely generated (G)-cotorsion module. Let M ∨ = Hom(M, Q p /Z p )
denote the Pontragin dual of M, which has the structure of a compact
(G)-module. Then, there exists a finite collection of non-zero elements
f 1 , . . . , fr ∈ (G), along with a pseudonull (G)-module D ∨ , such that
the following sequence is exact:

r
0 −→ (G)/ f i (G) −→ M ∨ −→ D ∨ −→ 0.
i=1

The product

r
∨ ∨
char(M ) = char(G) (M ) = fi ,
i =1

is unique up to unit in (G), and defines the (G)-characteristic power

series of M ∨ . If the homology group H0 (G, M) is finite, then the G-Euler
characteristics χ (G, M) and χ (G, D) are both well-defined, and given by the
formulae
460 Jeanine Van Order

χ (G, M) = | char(M ∨ )(0)|−1

p
χ (G, D) = 1.
Here, char(M ∨ )(0) denotes the image of the characteristic power series
char(M ∨ ) under the natural map (G) −→ Z p , and D denotes the discrete
dual of D ∨ .
In this note, we show how to use the framework of Euler characteristics
to study the Selmer groups of abelian varieties in certain abelian extensions
of CM fields, and in particular to verify the p-part of the refined Birch and
Swinnerton-Dyer conjecture via the associated Iwasawa main conjectures. To
be more precise, let A be a principally polarized abelian variety defined over a
totally real number field F. We shall assume for simplicity that A is principally
polarized, though all of the arguments given below can be carried through to
the more general case with some extra care. Fix a prime p above p in F. Let
K be a totally imaginary quadratic extension of F. Assume that A has good
ordinary reduction at each prime above p in K . Let K p∞ denote the dihedral
or anticyclotomic Zδp -extension of K , where δ = [Fp : Q p ] is the residue
degree. This p-adic Lie extension, whose construction we recall below, factors
through the tower of ring class fields of p-power conductor over K . Let G
denote the Galois group Gal(K p∞ /K ), which is topologically isomorphic to
Zδp . Let Sel(A/K p∞ ) denote the p ∞ -Selmer group of A over K p∞ . Hence by
definition,
Sel(A/K p∞ ) = lim Sel(A/L),
−→
L

where the inductive limit ranges over all finite extensions L of K contained in
K p∞ of the classically defined Selmer groups
 

Sel(A/L) = ker H (G S (L), A p∞ ) −→
1
Jv (L) .
v∈S

Here, S is any (fixed) finite set of primes of K that includes both the primes
above p and the primes where A has bad reduction. We write K S to denote the
maximal Galois extension of K unramified outside of S and the archimedean
places of K , and G S (L) to denote the Galois group Gal(K S /L). We write A p∞
to denote the collection of all p-power torsion of A over the relevant extension

of K in K S , i.e. A p∞ = n≥0 A pn with A pn = ker([ p n ] : A −→ A), which
is standard notation. In general, however, given an abelian group B, we shall
write B( p) to denote its p-primary component. We define

Jv (L) = H 1 (L w , A(L w ))( p),
w|v
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 461

where the sum runs over primes w above v in L. Note that each Selmer group
Sel(A/L) fits into the short exact descent sequence
0 −→ A(L) ⊗ Q p /Z p −→ Sel(A/L) −→ X(A/L)( p) −→ 0,
where A(L) denotes the Mordell-Weil group of A over L, and X(A/L)( p)
the p-primary part of the Tate-Shafarevich group X(A/L) of A over L. Now,
the p-primary of p ∞ -Selmer group Sel(A/K p∞ ) has the structure of a dis-
crete, cofinitely generated (G)-module. It is often (but not always) the case
that Sel(A/K p∞ ) is (G)-cotorsion. For instance, this is known by a stan-
dard argument to be the case when the p-primary Selmer group Sel(A/K ) is
finite. Let us first consider this case where Sel(A/K p∞ ) is (G)-cotorsion.
This setting is generally known as the definite case in the literature, for rea-
sons that we do not dwell on here.1 In any case, the conjecture of Birch and
Swinnerton-Dyer predicts that the Hasse-Weil L-function L(A/K , s) has an
analytic continuation to s = 1, with order of vanishing at s = 1 given by
the rank of the Mordell-Weil group A(K ). The refined conjecture predicts
that Tate-Shafarevich group X(A/K ) is finite, and moreover that the leading
coefficient of L(A/K , s) at s = 1 is given by the formula
.  .
|X(A/K )| · Reg(A/K ) · +(A/K ) 
dim(A)
· |ω| · 2 ω ∧ ω.
| A(K )tors | · |At (K )tors | · |D| 2 v|∞ A(K v ) v|∞ A(K v )
v:K →R v:K →C
(19.1)
Here, Reg(A/K ) denotes the regulator of A over K , i.e. the determinant of the
canonical height pairing on a basis of A(K )/ A(K )tors , +(A/K ) denotes the
product
/ /
 /ω/
+(A/K ) = cv // ∗ // ,
ω v v
v∞

where cv = cv ( A) = [ A(K v ) : A0 (K v )] is the local Tamagawa factor of A at a

finite prime v of K , ω = ω A a fixed nonzero global exterior form (an invariant
differential if A is an elliptic curve), and ωv∗ = ω∗A,v the Néron differential at v.
Moreover, At denotes the dual abelian variety associated to A, and D = D K
the absolute discriminant of K . We refer the reader to the article of Tate [36]
for details.2 We consider a p-adic analogue of this formula (19.1). Namely,
1 Roughly, when A/F is a modular abelian variety, the associated Hasse-Weil L-function
L(A/K , s) has an analytic continuation a product of Rankin-Selberg L-functions of cuspidal
eigenforms on some totally definite quaternion algebra over F, and this in particular can be
used to deduce that the root number of L(A/K , s) is equal to 1 (as opposed to −1). We refer
the reader to the e.g. works [3], [29], [24], or [38] for some of the many known Iwasawa
theoretic results in this direction.
2 The reader will note that we have included extraneous real places in the formula (19.1) above,
i.e. as K here is totally imaginary.
462 Jeanine Van Order

 v ) denote the reduction of A at a finite prime v of K . Write A(K )( p)

let A(κ
to denote the p-primary component of A(K ), and A(κ  v )( p) that of A(κ
 v ). We
first establish the following result, using standard techniques.

Theorem 19.1 (Theorem 19.30). Assume that A is a principally polarized

abelian variety having good ordinary reduction at each prime above p in
K . Assume additionally that the p ∞ -Selmer group Sel(A/K ) is finite. If
the p ∞ -Selmer group Sel(A/K p∞ ) is (G)-cotorsion, then the G-Euler
characteristic χ (G, Sel(A/K p∞ )) is well defined, and given by the formula

χ (G, Sel(A/K p∞ ))
 
|X(A/K )( p)|    cv (A)
= · | A(κv )( p)| ·
2
ord p .
|A(K )( p)|2 L v (A/K , 1)
v| p v p
∈Sns

(19.2)
Here, Sns denotes the subset of S of primes which do not split completely in
K p∞ , and L v (A/K , 1) denotes the Euler factor at v of the global L-value
L(A/K , 1).

This result has various antecedents in the literature (see e.g. [6], [8], or [42]),
though the setting of dihedral extensions considered here differs greatly from
the cyclotomic or single-variable settings considered in these works. Anyhow,
let us now examine the complementary setting where the p ∞ -Selmer group
Sel(A/K p∞ ) is not expected to be (G)-cotorsion. This setting is generally
known as the indefinite case in the literature, for reasons that we likewise do
not dwell on here.3 To fix ideas, let us assume the so-called weak Heegner
hypothesis that χ (N ) = (−1)d−1 , where χ denotes the quadratic charac-
ter associated to K /F, N the arithmetic conductor of A, and d = [F : Q]
the degree of F. This condition ensures that the abelian variety A comes
equipped with families of Heegner or CM points defined over ring class exten-
sions of K . We refer the reader to [19] or the discussion below for details.
Let us first consider the issue of describing the G-Euler characteristic of
the (G)-cotorsion submodule Sel(A/K p∞ )cotors of Sel(A/K p∞ ), following
the conjecture of Perrin-Riou [26] and Howard [19]. To state this version of
3 Roughly (as before), in the setting where A is a modular abelian variety defined over F, the
Hasse-Weil L-function L(A/K , s) has an analytic continuation given by the product of
Rankin-Selberg L-functions of cuspidal eigenforms on some totally indefinite quaternion
algebra over F (where indefinite here means ramified at all but one real place of F), and this
in particular can be used to deduce that the root number of L(A/K , s) is equal to −1 (as
opposed to 1). We refer the reader to the excellent paper of Howard [19] for an account of the
Iwasawa theory in this setting.
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 463

the Iwasawa main conjecture, let us write S∞ = S(A/K p∞ ) denote the com-
pactified p ∞ -Selmer group of A in K p∞ , which has the structure of a compact
(G)-module. Let H∞ denote the (G)-submodule of S∞ generated by the
images of Heegner points of p-power conductor under the appropriate norm
homomorphisms. We refer the reader to [18] or to the passage below for more
details of this construction. It is often known and generally expected to be the
case that the (G)-modules S∞ and H∞ are torsionfree of (G)-rank one, in
which case the quotient S∞ /H∞ has (G)-rank zero. Hence in this setting,
the quotient S∞ /H∞ is a torsion (G)-module, so has a well defined (G)-
characteristic power series char(S∞ /H∞ ). Let char(S∞ /H∞ )∗ denote the
image of char(S∞ /H∞ ) under the involution of (G) induced by inversion
in G. Given an integer n ≥ 0, let K [pn ] denote the ring class field of con-

ductor pn over K , with Galois group G[pn ]. Write K [p∞ ] = n≥0 K [pn ] to
denote the union of all ring class extensions of p-power conductor, with Galois
group G[p∞ ] = Gal(K [p∞ ]/K ) = limn G[pn ]. Recall that we let K p∞ denote
←−
the dihedral or anticyclotomic Zδp -extension of K . We write G = lim G pn
← −n
to denote the associated profinite Galois
√ group G pn = Gal(K pn /K ). Let Kn
denote the Artin symbol of dn = ( DO K ) ∩ O pn . Here, D = D K denotes
the absolute discriminant of K , and O pn the O F -order of conductor pn in K ,
i.e. O pn = O F + pn O K . Let K = lim Kn . Let  ,  A,K [ pn ] denote the p-adic
←−n
height pairing

 ,  A,K [ pn ] : At (K [ p n ]) × A(K [ p n ]) −→ Q p ,

defined in Howard [18, (9), §3.3] and Perrin-Riou [28]. Let us again assume
for simplicity that A is principally polarized. There exists by a construction of
Perrin-Riou [26] (see also [28]) a p-adic height pairing

hn : S(A/K pn ) × S(A/K pn ) −→ c−1 Z p

whose restriction to the image of the Kummer map A(K [ p n ]) ⊗ Z p →

S(E/K [ p n ]) coincides with the pairing  ,  A,K [ pn ] after identifying At ∼
=A
in the canonical way (cf. [18, Proposition 0.0.4]). Here, c ∈ Z p is some inte-
ger that does not depend on the choice of n. Following Perrin-Riou [26] and
Howard [18], these pairings can be used to construct a pairing

h∞ : S∞ × S∞ −→ c−1 Z p [[G]]
(lim an , lim bn ) −→ lim hn (an , bnσ ) · σ.
← − ← − ← −
n n σ ∈G pn
464 Jeanine Van Order

We then define the p-adic regulator R = R(A/K p∞ ) to be the image of S∞

in c−1 Z p [[G]] under this pairing h∞ . Let e denote the natural projection
e : Z p [[G[ p ∞ ]]] −→ Z p [[G]].
We then make the following conjecture, reformulating those of [26], [18],
and [19].

Conjecture 19.2 (Conjecture 19.32). Let A be a principally polarized abelian

variety defined over F having good ordinary reduction at each prime above
p in K . Assume that A satisfies the weak-Heegner hypothesis with respect to
N and K . Then, the G-Euler characteristic χ (G, Sel(A/K p∞ )cotors ) is well-
defined, and given by the formula
χ (G, Sel(A/K p∞)cotors ) = |char(S∞ /H∞)(0) · char(S∞ /H∞ )∗ (0) · R(0)|−1
p .

Here, R denotes the ideal (e(K))−1 R, which lies in the Z p -Iwasawa algebra
Z p [[G]].

We refer the reader to the divisibility proved in Howard [19, Theorem B]

for results towards this conjecture. We can also deduce the following result
toward this conjecture in the setting where A is an elliptic curve defined over
the totally real field Q. Hence, A is modular by the fundamental work of
Wiles [41], Taylor-Wiles [37], and Breuil-Conrad-Diamond-Taylor [5]. Let
f ∈ S2 ( 0 (N )) denote the cuspidal newform of level N associated to A by
modularity. Let K be an imaginary quadratic field in which the fixed prime
p os not ramified. Let K (μ p∞ ) denote the extension obtained from K by
adjoining all primitive p-power roots of unity, with = Gal(K (μ p ∞ )/K )
its Galois group. Thus, is topologically isomorphic to Z× p . Let L f denote
the two-variable p-adic L-function constructed by Hida [16] and Perrin-Riou
[26], which lies in the Z p -Iwasawa algebra Z p [[G[ p ∞ ] × ]], as explained
in the proof of [39, Theorem 2.9]. Fixing a topological generator γ of , we
write the expansion of L f in the cyclotomic variable γ − 1 as
L f = L f,0 + L f,1 (γ − 1) + L f,2 (γ − 1)2 + . . . , (19.3)
[[G[ p ∞ ]]]
with L f,i ∈ Z p for each i ≥ 0. If we assume the so-called Heegner
hypothesis that each prime dividing N splits in K , then root number of the
Hasse-Weil L-function L(A/K , s) equals −1, which forces the leading term
L f,0 in (19.3) to vanish. Let (G) denote the Z p -Iwasawa algebra Z p [[G]].
We deduce the following result from the (G)-adic Gross-Zagier theorem
of Howard [18, Theorem B] with the two-variable main conjecture shown in
Skinner-Urban [35, Theorem 3] and the integrality of the two-variable p-adic
L-function L f shown in [39].
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 465

Corollary 19.3 (Corollary 19.37). Let A be an elliptic curve defined over Q,

associated by modularity to a newform f ∈ S2 ( 0 (N )). Let K be an imaginary
quadratic extension of Q of discriminant D in which the fixed prime p is not
ramified. Assume that the elliptic curve A has good ordinary reduction at p,
that each prime dividing N splits in K , and that D is odd and not equal to
−3. Assume in addition that the absolute Galois group G K = Gal(K /K )
surjects onto the Z p -automorphisms of T p (A), that χ ( p) = 1, that p does not
divide the class number of K , and that the compactified Selmer group S∞ has
(G)-rank one. Then the G-Euler characteristic of χ (G, Sel(A/K p∞ )cotors )
is well-defined, and given by the formulae
χ (G, Sel p∞ (A/K p∞ )cotors ) ≥ |e(L f,1 (0))|−1
p
= |e(K · log p (γ ))−1 h∞ ( x∞ )(0)|−1
x∞ ,  p
/ /−1
= char(S∞ /H∞ )(0) · char(S∞ /H∞ )∗ (0) · R(0)/ p .
/

Here, log p is the p-adic logarithm (composed with a fixed isomorphism

∼
= Z× p ), 
x∞ is a generator of H∞ constructed from a compatible sequence
of regularized Heegner points (described below), and R is the ideal (e(K ·
log p (γ ))−1 R ∈ (G). Moreover, these formulae for χ (G, Sel(A/K p∞ )cotors )
do not depend on the choice of topological generator γ ∈ .

Though relatively simple to deduce, this result is revealing in that it links

the nonvanishing of the coefficient L f,1 to the nondegeneracy of the height
pairing h∞ and the p-adic regulator R. Anyhow, we have shown here we can
verify the p-part of the refined Birch and Swinnerton-Dyer formula (and varia-
tions) by relatively straightforward computations of Euler characteristics after
knowing partial results towards the associated Iwasawa main conjectures. It
would be interesting to relate these formula more precisely to the modular set-
ting outlined in Pollack-Weston [29], as well as perhaps to the conjectures of
Shimura/Prasanna [30] via Ribet-Takahashi [31]. It would also be interesting
to relate the conjectural formula described above in the indefinite setting to
the conjecture of Kolyvagin [22] on indivisibility of Heegner points (and nat-
ural extensions to CM points on quaternionic Shimura curves). However, such
problems lie beyond the scope of the present note.

19.2 Ring class towers

The dihedral or anticyclotomic Zδp -extension K p∞ of K . Suppose that F is
any totally real number field, and K any totally imaginary quadratic extension
of F. Recall that we fix throughout a rational prime p, and let p denote a fixed
466 Jeanine Van Order

prime above p in F. Given an integral ideal c ⊆ O F , we write Oc = O F +cO K

to denote the O F -order of conductor c in K . The ring class field of conductor
c of K is the Galois extension K [c] of K characterized via class field theory by
the identification

!× / F
K !c× K × −−rec
!× O K
−−→ Gal(K [c]/K ).
Here, rec K denotes the Artin reciprocity map, normalized to send uniformiz-
ers to geometric Frobenius automorphisms. Let us write G[c] to denote the
Galois group Gal(K [c]/K ). We consider the union of all ring class extensions
of p-power conductor
0
K [p∞ ] = K [pn ],
n≥0

along with its Galois group G[p∞ ] = Gal(K [p∞ ]/K ), whose profinite
structure we write as

G[p∞ ] = lim G[pn ].

← −
n

A standard argument in the theory of profinite groups shows that the tor-
sion subgroup G[p∞ ]tors ⊆ G[p∞ ] is finite, and moreover that the quotient
G[p∞ ]/G[p∞ ]tors is topologically isomorphic to Zδp , where δ = [Fp : Q p ]
is the residue degree of p. We refer the reader e.g. to [12, Corollary 2.2] for
details on how to deduce this fact via a basic computation with adelic quotient
groups. Let G p∞ denote the Galois group G[p∞ ]/G[p∞ ]tors ∼ = Zδp , which has
the structure of a profinite group

G p∞ = lim G pn .
← −
n

Let us then write K pn to denote the abelian extension of K contained in K [pn ]

such that G pn = Gal(K pn /K ). When there is no risk of confusion, we shall
simply write G to denote the Galois group G p∞ .

Lemma 19.4. The extension K p∞ over K is unramified outside of p.

Proof. The result is a standard deduction in local class field theory, as shown
for instance in Iwasawa [21, Theorem 1 §2.2]. In general, given any p-adic
field L, the abelianization of G L = Gal(L/L) is a completion of L × . The
subgroup corresponding to ramified extensions is O× L , which is a profinite
abelian group whose open subgroups are pro- p. The result is then easy to
deduce.
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 467

Finally, let us recall that the Galois extension K [p∞ ] is of generalized dihe-
dral type. Equivalently, the complex conjugation automorphism of Gal(K /F)
acts by inversion on G[p∞ ]. It follows that K [p∞ ] is linearly disjoint over K to
the cyclotomic extension K (μ p∞ ), where μ p∞ denotes the set of all p-power
roots of unity, because the complex conjugation automorphism acts trivially
on Gal(K (μ p∞ )/K ). For this reason, the extension K p∞ /K is often called the
anticyclotomic Zδp -extension of K .

Decomposition of primes in K p∞ . Let us now write G to denote the Galois

group G p∞ = Gal(K p∞ /K ). Given a finite prime v of K , let Dv ⊆ G
denote the decomposition subgroup at a fixed prime above v in K p∞ . We
shall often make the standard identification Dv with the Galois group G w =
Gal(K p∞ ,w /K v ), where w is a fixed prime above v in K p∞ .

Proposition 19.5. Let v be a finite prime of K .

(i) If v does not divide p and splits completely in K p∞ , then Dv is trivial.

(ii) If v does not divide p and does not split completely in K p∞ , then Dv can
be identified with a finite-index subgroup of G ≈ Zδp .
(iii) If v divides p, then Dv is topologically isomorphic to Zδp .

Proof. The result is easy to deduce from standard properties of local fields,
see e.g. [13, Proposition 5.10].

19.3 Galois cohomology

We now record for later use some basic facts from Galois cohomology.

p-cohomological dimension. Recall that the p-cohomological dimension

cd p (G) of a profinite group G is the smallest integer n satisfying the condition
that for any discrete torsion G-module M and any integer q > n, the p-primary
component of H q (G, M) is zero ([33, §3.1]). We have the following crucial
characterization of cd p .

Theorem 19.6 (Serre-Lazard). Let G be any p-adic Lie group. Let M be any
discrete G-module. If G does not contain an element of order p, then cd p (G)
is equal to the dimension of G as a p-adic Lie group.

Proof. See [34] and [25].

468 Jeanine Van Order

Since G p∞ ∼
= Zδp has no point of order p, we obtain the following
consequence.

Corollary 19.7. Let G denote the Galois group G p∞ ∼ = Zδp defined above.
Then, cd p (G) = δ. Moreover, given a finite prime v of K ,

(i) cd p (Dv ) = 0 if v does not divide p and splits completely in K p∞ ;

(ii) cd p (Dv ) = δ if v does divide p.

Proof. The claim follows from Theorem 19.6, using Proposition 19.5 for (i)
and (ii).

The Hochschild-Serre spectral sequence. Fix S any finite set of primes of

K containing the primes above p in K , and the primes where A has bad reduc-
tion. Let K S denote the maximal extension of K that is unramified outside of S
and the archimedean places of K . Let G S (K ) = Gal(K S /K ) denote the corre-
sponding Galois group. Observe that there is an inclusion of fields K p∞ ⊂ K S .
Given any intermediate field L with K ⊂ L ⊂ K p∞ , let

G S (L) = Gal(K S /L).

Recall again that we write G to denote the Galois group G p∞ = Gal(K p∞ /K ).

Lemma 19.8. For all integers i ≥ 1, we have bijections

H i (G, H i (G S (K p∞ ), A p∞ ) ∼
= H i+2 (G, A p∞ ). (19.4)

Proof. See [33, § 2.2.6] or [34]. We claim that for all i ≥ 0, a direct application
of the Hochschild-Serre spectral sequence gives the exact sequence

H i+1 (G S (K ), A p∞ ) −→ H i (G, H 1 (G S (K p∞ ), A p∞ )) −→ H i +2 (G, A p∞ ).

On the other hand, it is well known that H i (G S (K ), A p∞ ) = 0 for all i ≥ 2.

Hence, the claim holds for all i ≥ 1.

Some identifications in local cohomology. Let L be a finite extension of K .

Recall that for each finite prime v of K , we define

Jv (L) = H 1 (L w , A(L w ))( p), (19.5)
w|v

where the sum runs over all primes w above v in L. If L ∞ is an infinite exten-
sion of K , then we define the associated group Jv (L ∞ ) by taking the inductive
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 469

limit over finite extensions L of K contained in L ∞ with respect to restriction

maps,
Jv (L ∞ ) = lim Jv (L). (19.6)
−→
L
Given a finite prime v of K , let us fix a prime w above v in K p∞ . Recall that we
then write G w to denote the Galois group Gal(K p∞ ,w /K v ), which we identify
with the decomposition subgroup Dv ⊆ G.

Lemma 19.9. Let v be a finite prime of K , and w any prime above v in K p∞ .

Then, for each integer i ≥ 0, there is a canonical bijection
H i (G, Jv (K p∞ )) ∼
= H i (G w , H 1 (K p∞ ,w , A)( p)). (19.7)

Proof. Recall that we let G pn denote the Galois group Gal(K pn /K ), which is
isomorphic to (Z/ p n Z)δ . Let G pn ,w ⊆ G denote the decomposition subgroup
of the restriction of w to K pn . Shapiro’s lemma implies that for all integers
n ≥ 1 and i ≥ 0, we have canonical bijections
H i (G pn , Jv (K pn )) ∼
= H i (G pn ,w , H 1 (K pn ,w , A)( p)).
Passing to the inductive limit with n then proves the claim.

Corollary 19.10. Let v be any finite prime of K not dividing p which splits
completely in K p∞ , and let w be any prime above v in K p∞ . Then, for each
integer i ≥ 1, we have that H i (G w , Jv (K p∞ )) = 0.

Proof. The result follows from (19.7), since G w is trivial by Proposition 19.5.

Lemma 19.11. Let v be any finite prime of K that divides p, and write w to
denote the prime above v in K p∞ . Then, for each i ≥ δ − 1, we have that
H i (G w , H 1 (K p∞ ,w , A)( p)) = 0.
Moreover, for each 1 ≤ i ≤ δ − 2, we have bijections
H i (G w , H 1 (K p∞ ,w , A)( p)) ∼ v, p∞ ).
= H i +2 (G w , A (19.8)

Proof. See [6, Lemma 3.5]. Let G K v denote the Galois group Gal(K v /K v ),
where K v is a fixed algebraic closure of K v . Let I K v ⊆ G K v denote the inertia
subgroup at a fixed prime above v in K v . Consider the short exact sequence of
G K v -modules
0 −→ C −→ A p∞ −→ D −→ 0. (19.9)
470 Jeanine Van Order

Here, C denotes the canonical subgroup associated to the formal group of the
Néron model of A over O K v (see [7, § 4]). Moreover, (19.9) is characterized
by the fact that C is divisible, with D being the maximal quotient of A p∞ by a
divisible subgroup such that the inertia subgroup I K v acts via a finite quotient.
Since K p∞ ,w is deeply ramified in the sense of Coates-Greenberg [7], we have
by [7, Propositions 4.3 and 4.8] a canonical G w -isomorphism

H 1 (K p∞ ,w , A)( p) ∼
= H 1 (K p∞ ,w , D). (19.10)

Moreover, the G K v -module D vanishes if and only if A has potentially

supersingular reduction at v, as explained in [7, § 4].
Observe now that Gal(K v /K p∞ ,w ) has p-cohomological dimension equal
to 1, as the profinite degree of K p∞ ,w over K v is divisible by p ∞ (see [33]).
Hence, for all i ≥ 2, we have identifications

H i (K p∞ ,w , D) = 0. (19.11)

Taking (19.11) along with the Hochschild-Serre spectral sequence then gives
us for each i ≥ 1 exact sequences

H i+1 (K v , D) −→ H i (G w , H 1 (K p∞ ,w , D)) −→ H i+2 (G w , D). (19.12)

While we would like to follow the argument of [6, Lemma 3.5] in showing that
the terms on either side of (19.12) must vanish, we have not been able to find
an argument to deal with the fact that δ could be greater than 4. Anyhow, we
can deduce that

H i (K v , D) = 0 (19.13)

for all i ≥ 2. That is, we know that cd p G K v = 2, whence (19.13) holds
trivially for all i ≥ 3. To deduce the vanishing for i = 2, we take cohomology
of the exact sequence (19.9) to obtain a surjection from H 2 (K v , A p∞ ) onto
H 2 (K v , D). We know by local Tate duality that H 2 (K v , A p∞ ) vanishes (see
e.g. [8, 1.12 Lemma]). Hence, we deduce that (19.13) holds for all i ≥ 2. It
then follows from the exact sequence (19.12) that we have bijections

H i (G w , H 1 (K p∞ ,w , D)) −→ H i +2 (G w , D)

for all i ≥ 1. Now, observe that both of these groups vanish for all i ≥ δ − 1,
as cd p (G w ) = δ. The result now follows from the natural identification of D
with the G w -module defined by A p∞ ,v = A(K p∞ ,w ) p∞ .
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 471

19.4 The torsion subgroup A(K p∞ )tors

We start with the following basic result, whose proof relies on the fact that
there exist primes v ⊂ O K which split completely in the anticyclotomic
extension K p∞ .

Lemma 19.12. The torsion subgroup A(K p∞ )tors of A(K p∞ ) is finite.

Proof. Fix a finite prime v of F that remains inert in K . Let us also write
v to denote the prime above v in K . By class field theory, v ⊆ O K splits
completely in K p∞ . Hence, fixing a prime w above v in K p∞ , we can identify
the union of all completions K p∞ ,w with K v itself. In particular, we find that
A(K p∞ ,w ) = A(K v ), which gives us an injection A(K p∞ )tors −→ A(K v )tors .
On the other hand, consider the short exact sequence
 v ) −→ 0.
0 −→ A1 (K v ) −→ A0 (K v ) −→ A(κ (19.14)

Here, A1 (K v ) denotes the kernel of reduction modulo v, which can be iden-

! v ), and κv denotes the residue field of K at
tified with the formal group A(m
v. Let n ≥ 1 be any integer prime to the residue characteristic qv . It is well-
! v )[n] = 0. Hence, taking n-torsion in (19.14), we find for any
known that A(m
such prime v ⊂ O K an injection
 v ).
A0 (K v )[n] −→ A(κ (19.15)
 v ) is of course finite. Since we have this injection (19.15) for
The group A(κ
any such prime v ⊂ O K , it follows that A(K p∞ )tors must be finite.

We now consider the G-Euler characteristic of the p-primary subgroup A p∞

of A(K p∞ ), denoted here by χ (G, A p∞ ). Recall that this is defined by the
product

|H i (G, A p∞ )|(−1) ,
i
χ (G, A p∞ ) =
i ≥0

granted that it is well-defined.

Corollary 19.13. We have that χ (G, A p ∞ ) is well-defined and equal to 1.

Proof. This is a standard result, since G is a pro- p group with A p∞ (K p∞ ) a

finite group of order equal to some power of p. See [33, Exercise (a) §4.1].

Recall that given a finite prime v if K , we let G w denote the Galois group
v, p∞ denote
Gal(K p∞ ,w /K v ), where w is a fixed prime above v in K p∞ . Let A
472 Jeanine Van Order

the G w -module defined by A p∞ (K p∞ ,w ). Consider the G w -Euler characteris-

p∞ ,v , which we denote by χ (G w , A
tic of A v, p∞ ). Recall that this is defined by
the product

p∞ ,v ) =
χ (G w , A v, p∞ )|(−1)i ,
|H i (G w , A
i≥0

granted that it is well-defined.

Corollary 19.14. Let v be any finite prime of K , with w a fixed prime above v
in K p∞ . The G w -Euler characteristic χ (G w , Ap∞ ,v ) is well-defined and equal
to 1.

Proof. In any case on the decomposition of v in K p∞ , G w is pro- p, with

p∞ ,v a finite group of order equal to some power of p, whence the result is
A
standard. Note as well that since G w ∼
= Dv is in any case a closed subgroup of
G, the result can also be deduced from Corollary 19.13 above, since Shapiro’s
lemma gives canonical isomorphisms H i (G, A p∞ ) ∼ v, p∞ ) for
= H i (G w , A
each i ≥ 1.

Putting this all together, we have shown the following result.

Proposition 19.15. Let v be any finite prime of K , with w a fixed prime above
v, p∞ ) = 1.
K in K p∞ . Then, χ (G, A p∞ ) = χ (G w , A

19.5 The definite case

Let us keep all of the notations of the section above. Here, we shall assume that
the p ∞ -Selmer group Sel(A/K p∞ ) is (G)-cotorsion, where (G) denotes
the O-Iwasawa algebra (G p∞ ) = O[[G p∞ ]] for O the ring of integers of
some fixed finite extension of Q p . We shall also use here the standard notations
for abelian varieties and local fields, following Coates-Greenberg [7].

Strategy. We consider the snake lemma in the following commutative diagram

ψ S (K p∞ )
Sel(A/K p∞ )G −−−−→ H 1 (G S (K p∞ ), A p∞ )G −−−−−→ v∈S Jv (K p∞ )G
Q Q Q
⏐α ⏐β ⏐γ =
⏐ ⏐ ⏐ v∈S γv

λ S (K )
Sel(A/K ) −−−−→ H 1 (G S (K ), A p∞ ) −−−−→ v∈S Jv (K ).
(19.16)
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 473

Here, recall that S denotes any finite set of places of K containing the primes
above p, as well as all places where A has bad reduction. The map ψ S (K p∞ )
is induced from the localization map

λ S (K p∞ ) : H 1 (G S (K p∞ ), A p∞ ) −→ Jv (K p∞ )
s∈S
defining the p∞ -Selmer group Sel(A/K p∞ ). The vertical arrows are induced
by restriction on cohomology. We also consider the associated diagram
 
Im ψ S (K p∞ ) −−−−→ v∈S Jv (K p∞ ) − −−−→ coker ψ S (K p∞ )
Q Q Q
⏐ ⏐γ = ⏐
⏐1 ⏐ v∈S γv ⏐2
Im (λ S (K )) −−−−→ v∈S Jv (K ) −−−−→ coker (λ S (K )) .
(19.17)
Local restrictions maps. We now compute the cardinalities of the kernel and
cokernel of the restriction map γ = v∈S γv in (19.16), in a series of lemmas
leading to Theorem 19.23 below. Throughout this section, we shall assume for
simplicity that the abelian variety is principally polarized.

Lemma 19.16. Let v be a finite prime of K which does not divide p. Assume
that A is principally polarized. Then, the group Jv (K ) = H 1 (K v , A)( p)
is finite, and its order is given by the exact power of p dividing the quo-
tient cv (A)/L v (A, 1). Here, cv (A) = [A(K v ) : A0 (K v )] denotes the local
Tamagawa factor at v, and L v (A, 1) the Euler factor at v of the Hasse-Weil
L-function L(A, s) at s = 1.

Proof. The result is standard, see e.g. [6, Lemma 3.6].

Lemma 19.17. Let v be a finite prime of K which does not divide p. Let L be
an arbitrary Galois extension of K v with Galois group  = Gal(L/K v ). Let
τv : H 1 (K v , A)( p) −→ H 1 (L , A)( p)
denote the natural map induced by restriction. Then, we have bijections
ker(τv ) ∼
= H 1 (, A p∞ (L))
coker(τv ) ∼
= H 2 (, A p∞ (L)).

Proof. See [6, Lemma 3.7]. That is, consider the commutative diagram
sv
H 1 (K v , A p∞ ) −−−−→ H 1 (L , A p∞ )
⏐ ⏐
⏐ ⏐
S S (19.18)
τv
H 1 (K v , A)( p) −−−−→ H 1 (L , A)( p) .
474 Jeanine Van Order

Here, sv denotes the restriction map on cohomology. The vertical arrows

denote the surjective maps derived from Kummer theory on A. Now, observe
that since v does not divide p, we have identifications

A(K v ) ⊗ Q p /Z p = A(L) ⊗ Q p /Z p = 0.

Hence, the vertical arrows in (19.18) are isomorphisms, and we have bijections

ker(τv ) ∼
= ker(sv )
∼ coker(sv ).
coker(τv ) =

The result now follows from the Hochschild-Serre spectral sequence associ-
ated to sv , using the fact that H 2 (K v , A p∞ ) = 0 by local Tate duality.

Corollary 19.18. Let v be a prime of S not dividing p. If v splits completely

K p∞ , then the local restriction map γv is an isomorphism, i.e. ker(γv ) =
coker(γv ) = 0. If v does not split completely in K p∞ , then γv is the zero
map, and the order of its kernel Jv (K ) = H 1 (K v , A)( p) is the exact power of
p dividing cv (A)/L v (A, 1).

Proof. The first assertion follows from the result of Corollary 19.7, which
implies that cd p (G w ) = 0 for any such prime v of K not dividing p. To see
the second assertion, it can be argued in the same manner as [6, Lemma 3.3]
that the Jv (K p∞ ) vanishes in this setting (i.e. since the Galois group of the
extension K p∞ ,w over K v is isomorphic to a finite number of copies of the
Galois group of the maximal unramified pro- p extension of K v ). The result
then follows from that of Lemma 19.16. Note as well that if we take L =
K p∞ ,w with  = G w in Lemma 19.17, we obtain identifications

ker(γv ) = H 1 (G w , A p∞ )
coker(γv ) = H 2 (G w , A p∞ ),

whence we can also deduce that |H 1 (G w , A p∞ )| = ord p (cv (A)/L v (A, 1))
and that H 2 (G w , A p∞ ) = 0.

To study the localization maps γv with v dividing p in the manner of [6], we

start with the following basic result.

Lemma 19.19. Let v be a prime of K dividing p. Then, we have identifications

ker(γv ) ∼
= H 1 (G w , A(K p∞ ,w ))( p)
coker(γv ) ∼
= H 2 (G w , A(K p∞ ,w ))( p).
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 475

Proof. Taking the Hochschild-Serre spectral sequence associated to K p∞ ,w

over K v with the fact that H 2 (K v , A p∞ ) = 0, the result follows from the
definition of γv .

We now consider the primes v | p of K where A has good ordinary

reduction.

Lemma 19.20. Let v be a prime of K dividing p where A has good

ordinary reduction. Assume that A is principally polarized. Then, we have
identifications

0 if i ≥ 2
!v (m))| =
|H i (K v , A

| A(κv )( p)| if i = 1.

Proof. See [6, Lemma 3.13], the proof carries over without changes to the
setting of principally polarized abelian varieties.

Lemma 19.21. Let v be a prime of K above p where A has good ordinary

reduction. Assume additionally that A is principally polarized. Then, for all
i ≥ 2, we have isomorphisms

H i (G w , A(K p∞ ,w ))( p) ∼ v, p∞ ).

= H i (G w , A

We also have for i = 1 the short exact sequence

!v (m)) −→ H 1 (G w , A(K p∞ ,w ))( p) −→ H 1 (G w , A

H 1 (K v , A v, p∞ ).
(19.19)

Proof. The results are deduced from the lemmas above, following [6, Lemmas
3.12 and 3.14], which work in the same way for principally polarized abelian
varieties using the results of [7].

Putting these identifications together, we obtain the following result. Let us

v, p∞ ) for any integer i ≥ 0.
write h v,i to denote the cardinality of H i (G w , A

Proposition 19.22. Let v be a prime of S above p where A has good ordinary

reduction. Assume additionally that A is principally polarized. Then, ker(γv )
and coker(γv ) are finite, and we have the identification
| ker(γv )| h v,0 · h v,1
= .
|coker(γv )| h v,2
476 Jeanine Van Order

Proof. Lemmas 19.19 and 19.21 imply that |coker(γv )| = h v,2 . Using the
exact sequence (19.19) along with Lemma 19.20 for i = 0, we then deduce
that |coker(γv )| = h v,0 · h v,1 , as required.

To summarize what we have shown in this section, let us for a prime v ∈ S

not dividing p write mv (A) to denote the exponent ord p (cv (A)/L v (A, 1)).

Theorem 19.23. Let Sns denote the subset of primes of the fixed set S which
(i) do not dividing p and (ii) do not split completely in K p∞ . We have the
following description of the global restriction map γ in (19.16):

| ker(γ )|  h v,0 · h v,1 

= × mv (A).
|coker(γ )| v| p
h v,2 v p
v∈S v∈Sns

Global calculation. We start with the easy part of the calculation, which
in some sense is just diagram chasing in (19.16). Let us now assume for
simplicity that A has good ordinary reduction at all primes above p in K .

Lemma 19.24. We have the following bijections for the restriction map β:

ker(β) ∼
= H 1 (G, A p∞ )
coker(β) ∼
= H 2 (G, A p∞ ).

Proof. This follows immediately from the inflation–restriction exact

sequence, using the fact that H 2 (G S (K ), A p∞ ) = 0.

Recall that given a prime v ∈ S and an integer i ≥ 0, we write h v,i

v, p∞ ). Let us also write h i to denote
to denote the cardinality of H i (G w , A
the cardinality of H (G, A p∞ ). We now give the first part of the global
i

calculation:

Lemma 19.25. Assume that Sel(A/K ) and coker(λ S (K )) are finite. Assume
that A has good ordinary reduction at each prime above p in K . Then, the
cardinality |H 0 (G, Sel(A/K p∞ ))| is given by the formula

h2  h v,0 · h v,1 
|Sel(A/K )| · |coker(ψ S (K p∞ ))| · · · mv (A).
h1 · h0 h v,2
v| p v p
v∈Sns

(19.20)
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 477

Proof. Theorem 19.23 implies that

| ker(γ )|  h v,0 · h v,1 

= · mv (A). (19.21)
|coker(γ )| h v,2
v| p v p
v∈Sns

Observe that the numerator and denominator of (19.21) are finite by our
assumption that A has good ordinary reduction at each prime above p in K .
Using this result in the snake lemma associated to (19.17), we deduce that
ker(1 ) and coker(1 ) must be finite. Moreover, we deduce that

| ker(1 )| | ker(γ )| |coker ψ S (K p∞ ) |
= · . (19.22)
|coker(1 )| |coker(γ )| |coker (λ S (K )) |

Now, it is well-known that if Sel(A/K ) and coker(λ S (K )) are finite, then

|coker (λ S (K )) | = A(K ) p ∞ = h 0 . (19.23)

On the other hand, taking the snake lemma in (19.16), we find that

|Sel(A/K p∞ )G | |coker(β)| | ker(1 )|

= · . (19.24)
|Sel( A/K )| | ker(β)| |coker(1 )|

The result now follows from (19.24) by (19.21), (19.22) and (19.23).

Corollary 19.26. If the localization map λ S (K p∞ ) is surjective, then

|Sel(A/K )|  2 
χ (G, Sel(A/K p∞ )) = 2
· h 0,v · mv (A). (19.25)
h0 v| p v p
v∈Sns

Proof. If the localization map λ S (K p∞ ) is surjective, then we have the short

exact sequence

0 −→ Sel(A/K p∞ ) −→ B∞ −→ C∞ −→ 0, (19.26)

where

B∞ = H 1 (G S (K p∞ ), A p∞ )

C∞ = Jv (K p∞ ).
v∈S
478 Jeanine Van Order

Taking G-cohomology of (19.26) then gives the long exact sequence

ψ S (K p∞ )
0 −−−−−→ H 0 (G, Sel(A/K p∞ )) −−−−−→ H 0 (G, B∞ ) −−−−−−→ H 0 (G, C∞ )

−−−−−→ H 1 (G, Sel(A/K p∞ )) −−−−−→ H 1 (G, B∞ ) −−−−−→ H 1 (G, C∞ )

−−−−−→ ...

−−−−−→ H δ (G, Sel(A/K p∞ )) −−−−−→ H δ (G, B∞ ) −−−−−→ H δ (G, C∞ )

−−−−−→ 0.
(19.27)
Let us now write

Wi,v = H i (G w , H 1 (K p∞ ,v , A p∞ )( p)).
We can then extract from (19.27) the long exact sequence
ψ S (K p∞ )
0 −−−−−→ H 0 (G, Sel(A/K p∞ )) −−−−−→ H 0 (G, B∞ ) −−−−−−→ v∈S W0,v

−−−−−→ H 1 (G, Sel(A/K p∞ )) −−−−−→ H 3 (G, A p∞ ) −−−−−→ v, p∞ )

H 3 (G w , A
v∈S

−−−−−→ ...

−−−−−→ H δ (G, Sel(A/K p∞ )) −−−−−→ 0.

(19.28)
Here, the identifications for the central terms come from the Hochschild-Serre
spectral sequence, (19.4). The identifications of the terms on the right come
from Corollary 19.10 for the first row, then from the version of Shapiro’s
lemma given in (19.7) along with the bijections (19.8) for the subsequent rows.
Note that we have ignored the vanishing of higher cohomology groups here.
The reason for this will me made clear below. That is, using the definition
of the map ψS (K p∞ ) in (19.16), we can then extract from (19.28) the exact
sequence

0 −−−−−→ coker ψ S (K p∞ ) −−−−−→ H 1 (G, Sel( A/K p∞ )) −−−−−→ H 3 (G, A p ∞ )

−−−−−→ ··· −−−−−→ H δ (G, Sel( A/K p∞ )) −−−−−→ 0.

(19.29)
Using (19.29), we deduce that
|H 1 (G, Sel(A/K p∞ ))|
χ (G, Sel(A/K p∞ )) =  · χ (3) (G, A p∞ )
|coker ψ S (K p∞ ) |

· χ (3) (G w , Av, p∞ )−1 . (19.30)
v∈S
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 479

Here, we have written


χ (3) (G, A p∞ ) = |H i (G, A p ∞ )|(−1) ,
i

i≥3

and

χ (3) (G w , A p ∞ ) = v, p∞ )|(−1)i .
|H i (G w , A
i ≥3

Substituting (19.20) into (19.30), we then find that χ (G, Sel( A/K p∞ )) is
equal to
h 2  h v,0 h v,1  
v, p∞ )−1
|Sel(A/K )| · · · mv ( A) · χ (3) (G, A p∞ ) · χ (3) (G w , A
h0h1 h v,2
v| p v p v| p
v∈Sns

χ (G, A p∞ )  
= |Sel(A/K )| · · mv ( A) · v, p∞ )−1 .
h 20,v · χ (G w , A
h02
v p v| p
v∈Sns

The result then follows from Proposition 19.15, which gives for each prime
v, p ∞ ) = 1.
v | p the identifications χ (G, A p∞ ) = χ (G w , A

Surjectivity of the localization map. Recall that we write  to denote the

O-Iwasawa algebra (G) = O[[G p∞ ]]. We now show that the localization
map λ S (K p∞ ) is surjective if Sel p∞ (A/K p∞ ) is -cotorsion, at least when A
is known to be prinicipally polarized. Given L a Galois extension of K , let
S(A/L) denote the compactified Selmer group of A over L,
 

S(A/L) = lim ker H (G S (L), A pn ) −→
1
Jv (L) .
← −
n v∈S

Proposition 19.27. Let  = Gal(L/K ) be an infinite, pro- p group. If A(L) p∞

is finite, then there is a ()-module injection

S(A/L) −→ Hom() (X (A/L), ()) .

Proof. The result is well-known, see for instance [15, Theorem 7.1], the proof
of which carries over without change.

Theorem 19.28. (Cassels-Poitou-Tate exact sequence for abelian varieties).

Let At denote the dual abelian variety associated to A. Let S(A/K p∞ )∨
denote the Pontryagin dual of the compactified Selmer group S(A/K p∞ ).
There is a canonical exact sequence
480 Jeanine Van Order

0 −−−−−→ Sel(A/K p∞ ) −−−−−→ H 1 (G S (K p∞ ), A p∞ ) −−−−−→ v∈S H 1 (K p∞ ,w , A)( p)

−−−−−→ S∨ ( At /K p∞ ) −−−−−→ H 2 (G S (K p∞ ), A p∞ ) −−−−−→ 0.

(19.31)

Proof. See Coates-Sujatha [8, § 1.7]. The same deduction using the general-
ized Cassels-Poitou-Tate exact sequence [8, Theorem 1.5] works here, keeping
track of the dual abelian variety At .

Corollary 19.29. Assume that A is principally polarized. If Sel(A/K p∞ ) is

-cotorsion, then the localization map λ S (K p∞ ) is surjective, i.e. there is a
short exact sequence

0 −→ Sel(A/K p∞ ) −→ H 1 (G S (K p∞ ), A p ∞ ) −→ Jv (K p∞ ) −→ 0.
v∈S

Proof. If Sel(A/K p∞ ) is -cotorsion, then X (A/K p∞ ) is -torsion by dual-

ity. It follows that Hom (X (A/K p∞ ), ) = 0. Hence by Proposition 19.27,
we find that S(A/K p∞ ) = 0. It follows that S∨ (A/K p∞ ) = 0. Since we
assume that A is principally polarized, we can assume that there is an isomor-
phism A ∼ = At , in which case it also follows that S∨ (At /K p∞ ) = 0. The claim
then follows from (19.31). Observe that we also obtain from this the vanishing
of H 2 (G S (K p∞ ), A p∞ ).

Hence, we obtain from Corollary 19.26 the following main result of this
section.

Theorem 19.30. Assume that A is a principally polarized, and that A has

good ordinary reduction at each prime above p in K . Assume additionally
that Sel(A/K ) is finite. If Sel p∞ (A/K p∞ ) is (G)-cotorsion, then
 
|X(A/K )( p)|    cv (A)
χ (G, Sel(A/K p∞ )) = · | A(κv )( p)| · ord p
2
.
| A(K )( p)|2 L v (A, 1)
v| p v p
v∈Sns

Proof. The computations above show that the Euler characteristic is well-
defined, and given by the formula
|Sel(A/K )|  2 
χ (G, Sel(A/K p∞ )) = · h 0,v · mv (A).
h 20 v| p v p
v∈Sns

The result then follows from the definitions, as well as the standard identifica-
tion in this setting of Sel(A/K ) with X(A/K )[ p ∞ ].
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 481

19.6 The indefinite case

Let us keep all of the setup above, but with the following crucial condition. Let
N ⊂ O F denote the arithmetic conductor of A. Recall that we write η to denote
the quadratic character associated to K /F. We assume here that the so-called
weak Heegner hypothesis holds, which is that η(N ) = (−1)d−1 (cf. [19, 0.]).
In this setting, the Selmer group Sel(A/K p∞ ) is generally not -cotorsion, but
rather free of rank one over . For results in this direction, see Howard [19,
Theorem B], which generalizes earlier results of Bertolini [1] and Nekovar (cf.
[?] or [?]).

The conjectures of Perrin-Riou and Howard. We refer the reader to [19] or

[26] for background. The version of the Iwasawa main conjectures posed in
these works has the following interpretation in terms of Euler characteristic
formulae. Recall that we fix a prime p above p in F. For each finite exten-
sion L of K , we consider the two p-primary Selmer groups Selp∞ (A/L) and
S(A/L), which can be defined implicitly via their inclusion in the descent
exact sequences

0 → A(L) ⊗O F O Fp → S(A/L) → lim X(A/L)[pn ]

← −
n

0 → A(L) ⊗O F Fp /O Fp → Selp∞ (A/L) → X(A/L)[p∞ ] → 0.

The abelian variety A comes equipped with a family of Heegner (or CM) points
defined over the ring class extensions K [pn ], as described e.g. in [19, §1.2],
which gives rise to the following submodule. Given an integer n ≥ 0, let h n
denote the image under the norm from K [pn+1 ] to K pn of the Heegner/CM
point of conductor pn+1 . Let Hn denote the -module generated by all the
images h m with m ≤ n. Let H∞ = limn Hn denote the associated Heegner
←−
module. Taking the usual limits, let us also define the finitely generated -
modules

S∞ = S(A/K p∞ ) = lim S(A/K pn )

← −
n
Selp∞ (A/K p∞ ) = lim Selp∞ (A/K pn ).
−→
n

Let X (A/K p∞ ) denote the Pontryagin dual of Selp∞ (A/K p∞ ), with

X (A/K p∞ ) its -cotorsion submodule. Given an element λ ∈ , let λ∗ denote
the image of λ under the involution of  induced by inversion in G = G p∞ .
The main conjecture posed by Howard [19, Theorem B], following Perrin-Riou
[26], can then be summarized as in the following way.
482 Jeanine Van Order

Conjecture 19.31 (Iwasawa main conjecture). If A has ordinary reduction at

p, then

(i) The -modules H∞ and S∞ are torsionfree of rank one.

(ii) The dual Selmer group X (A/K p∞ ) has rank one as a -module.
(iii) There is a pseudoisomorphism of -modules

X (A/K p∞ )tors −→ M ⊕ M ⊕ Mp.

Here, the -characteristic power series char(M) is prime to p with the

property that char(M) = char(M)∗ , and char(Mp) is a power of p.
(iv) The following equality of ideals holds in , at least up to powers of p:

(char(M)) = (char (S∞ /H∞ )) .

Remark. Note that many cases of this conjecture (with one divisibility in (iv))
are established by Howard [19, Theorem B], using the relevant nonvanishing
theorem of Cornut-Vatsal [12] to ensure nontriviality.

Let us now re-state this conjecture in terms of Euler characteristic

formulae. Let Selp∞ (A/K p∞ )cotors denote the -cotorsion submodule of
Selp∞ (A/K p∞ ). Recall that given an element λ ∈ , we write λ(0) to denote
the image of λ under the natural map  √ −→ Z p . Recall as well that we let
Kn denote the Artin symbol of dn = ( DO K ) ∩ O pn . Here, D denotes the
absolute discriminant of K , and O p n the O F -order of conductor pn in K , i.e.
O pn = O F + pn O K . Let K = limn Kn . We let  ,  A,K [ pn ] denote the p-adic
←−
height pairing

 ,  A,K [ p n ] : At (K [ p n ]) × A(K [ p n ]) −→ Q p ,

defined e.g. in [18, (9), §3.3] or [28]. Assume for simplicity that A is princi-
pally polarized. There exists by a construction of Perrin-Riou [26] (see also
[28]) a p-adic height pairing

hn : S(A/K pn ) × S(A/K pn ) −→ c−1 Z p

whose restriction to the image of the Kummer map A(K [ p n ]) ⊗ Z p →

S(E/K [ p n ]) coincides with the pairing  ,  A,K [ pn ] after identifying At ∼
=A
in the canonical way (cf. [18, Proposition 0.0.4]). Here, c ∈ Z p is some integer
that does not depend on the choice of n. Following [26] and [18], these pairings
can be used to construct a pairing
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 483

h∞ : S∞ × S∞ −→ c−1 Z p [[G]]
(lim an , lim bn ) −→ lim hn (an , bnσ ) · σ.
← − ← − ← −
n n σ ∈G pn

We then define the p-adic regulator R = R(A/K p∞ ) to be the image of S∞

in c−1 Z p [[G]] under this pairing h∞ . Let e denote the natural projection

e : Z p [[G[ p∞ ]]] −→ Z p [[G]].

We then make the following conjecture in this situation.

Conjecture 19.32. Let A be a principally polarized abelian variety of arith-

metic conductor N defined over F having good ordinary reduction at each
prime above p in K . Assume as well that A satisfies the weak-Heegner
hypothesis with respect to N and K , i.e. that η(N ) = (−1)d−1 . Then, the
G-Euler characteristic χ (G, Sel(A/K p∞ )cotors ) is well-defined, and given by
the expression

χ (G, Sel(A/K p∞ )cotors ) = |char(S∞ /H∞ )(0) · char(S∞ /H∞ )∗ (0) · R(0)|−1
p .

Here, R denotes the ideal (e(K))−1 R, which lies in the Z p -Iwasawa algebra
Z p [[G]]. Thus, we also conjecture that the p-adic regulator R = R(A/K p∞ )
is not identically zero, and hence that the p-adic height pairing h∞ is
nondegenerate.

Modular elliptic curves over imaginary quadratic fields. Let us now give a
more precise description of the conjectural formula for elliptic curves defined
over the totally real field F = Q. We first recall the -adic Gross-Zagier
theorem of Howard [18, Theorem B]. Let p be an odd prime. Fix embeddings
Q → Q p and Q → C. Let E be an elliptic curve of conductor N defined
over Q. Hence, E is modular by fundamental work of Wiles, Taylor-Wiles,
and Breuil-Conrad-Diamond-Taylor. Let f ∈ S2 ( 0 (N )) denote the cuspidal
newform of weight 2 and level N associated to E by modularity. Let K be an
imaginary quadratic field of discriminant D, and associated quadratic character
η. We impose the following

Hypothesis 19.33. Assume (i) that f is p-ordinary in the sense that the image
of its T p -eigenvalue under the fixed embedding Q → Q p is a p-adic unit, (ii)
that the Heegner hypothesis holds with respect to N and K , i.e. that each prime
divisor of N split in K , and (iii) that ( p, N D) = 1.
484 Jeanine Van Order

Let us for clarity also fix the following notations here. We write K [ p ∞ ]
to denote the p ∞ -ring class tower over K , with Galois group Ω =
Gal(K [ p ∞ ]/K ). We then write K p∞ to denote the dihedral or anticyclotomic
Z p -extension of K , with Galois group  = Gal(K p∞ /K ) ≈ Z p . We also
write K (μ p∞ ) to denote the extension of K obtained by adjoining all prim-
itive, p-th power roots of unity, with Galois group Γ = Gal(K (μ p∞ )/K ).
We then write K cyc to denote the cyclotomic Z p -extension of K , with Galois
group = Gal(K cyc /K ) ≈ Z p . We shall consider the compositum extension
R∞ = K [ p ∞ ] · K (μ p∞ ) with Galois group G = Gal(R∞ /K ) = Ω × Γ .
We shall also also consider the Z2p -extension K ∞ of K contained in R∞ , with
Galois group G = Gal(K ∞ /K ) =  × ≈ Z2p . The construction of Hida
[16] and Perrin-Riou [27] (whose integrality is shown in [39, Theorem 2.9])
gives a two-variable p-adic L-function
L f ∈ Z p [[G]] = Z p [[Ω × Γ ]].
This element interpolates the algebraic values
L(E/K , W, 1)/8π 2  f, f  N = L( f × gW , 1)/8π 2  f, f ,
which we view as elements of Q p under our fixed embedding Q −→ Q p . Here,
W is any finite order character of G, and  f, f  N the Petersson inner product of
f with itself. Let us now commit a minor abuse of notation in also writing L f
to denote the image of this p-adic L-function in the Iwasawa algebra Z p [[Ω ×
]]. Fixing a topological generator γ of , we can then write this two-variable
p-adic L-function L f as a power series in the cyclotomic variable (γ − 1),
L f = L f,0 + L f,1 (γ − 1) + L f,2 (γ − 1)2 + . . . (19.32)
Here, the coefficients L f,i are elements of the completed group ring Z p [[Ω]].
Now, since we assume the Heegner hypothesis in Hypothesis 19.33 (ii), it
is a well known consequence that the functional equation for L(E/K , s) =
L( f × g K , s) forces the central value L(E/K , 1) = L( f × g K , 1) to vanish.
In fact, it is also well known that L(E/K , W, 1) = L( f × gW , 1) vanishes for
W any finite order character of Ω in this situation (see e.g. [12, §1]). Hence,
in this situation, it follows from the interpolation property of L f that we must
have L f,0 = 0. The main result of Howard [18] shows that the linear term L f,1
in this setting is related to height pairings of Heegner points in the Jacobian
J0 (N ) of the modular curve X 0 (N ). To be more precise, the Heegner hypoth-
esis implies for each integer n ≥ 0 the existence of a family of Heegner points
h n ∈ X 0 (N )(C) of conductor p n . These Heegner points h n are defined in the
usual way: each h n is a cyclic N -isogeny h n : En → En of elliptic curves En
and En having exact CM by the Z-order O pn = Z + p n O K . As explained in
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 485

[18], the family {h n }n can be chosen in such a way that the following diagram
commutes, where the vertical arrows are p-isogenies:


This in particular makes En−1 a quotient of En by its pO pn−1 -torsion, and En−1

a quotient of En by its pO pn−1 -torsion. By the theory of complex multipli-
cation, the CM elliptic curves En and En , as well as the isogeny h n between
them, can all be defined over the ring class field K [ p n ] of conductor p n
over K . Hence, we have for each integer n ≥ 0 a family of Heegner points
h n ∈ X 0 (N )(K [ p n ]). Let us commit an abuse of notation in also writ-
ing h n to denote the image of any such point under the usual embedding
X 0 (N ) → J0 (N ) that sends the cusp at infinity to the origin.
Let T denote the Q-algebra generated by the Hecke operators Tl at primes
l not dividing N acting on J0 (N ). The semisimplicity of T induces T-module
isomorphism

J0 (N )(K [ p n ]) −→ J (K [ p n ])β .
β

Here, the direct sum runs over all Gal(Q p /Q p )-orbits of algebra homomor-
phisms β : T → Q p , and each summand is fixed by the natural action of
Gal(K [ p n ]/Q). If β(T) is contained in Q p , then the Hecke algebra T acts
on the summand J (K [ p n ])β via the homomorphism β. Our fixed newform
f ∈ S2 ( 0 (N )) determines such a homomorphism, which we shall also denote
by f in an abuse of notation. We shall then write h n, f to denote the image of
a Heegner point h n ∈ J0 (N )(K [ p n ]) in the associated summand J (K [ p n ]) f .
Let α p denote the p-adic unit root of the Hecke polynomial X 2 − a p ( f )X + p,
where a p ( f ) denotes the T p -eigenvalue of f . Using the language of [2], we
define for each integer n ≥ 0 a sequence of regularized Heegner points:
1 1
zn = h n, f − n−1 h n−1, f
α np αp
if n ≥ 1, otherwise
⎧  
σp σ p∗
⎪ −
⎨ 1 σp  1 −
∗ σp h 0, f if η( p) = 1
z0 = · 
1
u ⎪
⎩ 1 − α12 h 0, f if η( p) = −1.
p
486 Jeanine Van Order

Here, u = 12 |O× ∗
K |, and σ p and σ p denote the Frobenius automorphisms in the
Galois group Gal(K [1]/K ) ≈ Pic(O K ) of the Hilbert class field K [1] over
K at the two primes above p in the case where η( p) = 1. It can be deduced
from the Euler system relations shown in Howard [18] that these regularized
Heegner points are compatible under norm and trace maps on the summand
J (K [ p n ]) f . We refer the reader to [18, § 1.2] for details.
The following main result of Howard [18] was first proved by Perrin-Riou
[26] for the√ case of n = 0. To state this result, let Kn denote the Artin symbol
of dn = ( DO K ) ∩ O pn . Let  ,  E,K [ pn ] denote the canonical p-adic height
pairing

 ,  E ,K [ pn ] : E t (K [ p n ]) × E(K [ p n ]) −→ Q p ,

as defined in [18, (9), § 3.3]. Note that this pairing is canonical as a conse-
quence of the fact that E is ordinary at p, along with the uniqueness claims
of [18, Proposition 3.2.1]. The reader will also note that we do not extend this
pairing Q p -linearly as done throughout [18], to obtain the results stated here
after taking tensor products ⊗Q p . Indeed, such extensions are not necessary as
L f is integral by [39, Theorem 2.9], and hence belongs to the Iwasawa alge-
bra Z p [[G]] (as opposed to just Z p [[G]] ⊗Z p Q p ). Let log p denote the p-adic
logarithm, composed with a fixed isomorphism ∼
= Z× p . Let us also write
t
z n denote the image of the regularized Heegner point z n under the canonical
principal polarization J0 (N ) ∼
= J0 (N )t . We deduce from [18, Theorem A] the
following result.

Theorem 19.34 (Howard). Assume that D is odd and not equal to −3.
Assume as well that η( p) = 1. Then, for any dihedral character ρ :
×
Gal(K [ p n ]/K ) −→ Q p , we have the identity

ρ(Kn ) · log p (γ ) · ρ(L f,1 ) = ρ(σ )z nt , z nσ  E,K [ pn ] . (19.33)

σ ∈Gal(K [ pn ]/K )

This identity (19.33) is independent of the choice of topological generator

γ ∈ .

This result has the following implications for our modular elliptic curve
E, which belongs the isogeny class of ordinary elliptic curves associated
to the newform f . Fix a modular parametrization ϕ : X 0 (N ) → E. Let
ϕ∗ : J0 (N ) → E denote the induced Albanese map, and ϕ ∗ : E → J0 (N ) the
induced Picard map. Let

xn = ϕ∗ (z n ) ∈ E(K [ p n ])
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 487

denote the image of the regularized Heegner point z n under the Albanese map
ϕ∗ , and let xnt denote the unique point of E(K [ p n ]) for which ϕ ∗ (xnt ) = z nt .
Hence, we can identify xn with deg(ϕ)xnt . Note that the points xn and xnt are
norm compatible as n varies, thanks to the norm compatibility of the regular-
ized Heegner points z n and z nt . Granted this property, we define the Heegner
p-adic L-function LHeeg in Z p [[Ω]] by

LHeeg = lim xnt , xnσ  E ,K [ pn ] · σ.

← −
n σ ∈Gal(K [ pn ]/K )

We deduce from Theorem 19.34 the following integral version of [18, Theo-
rem B].

Theorem 19.35. Keep the setup and hypotheses of Theorem 19.34 above. Let
us write K = limn Kn ∈ Ω. Then, as ideals in the Iwasawa algebra Z p [[Ω]],
← −
we have

K · log p (γ ) · L f,1 = LHeeg .

This result in turn has applications to the -adic Gross-Zagier theorem

of Mazur-Rubin [23, Conjecture 9], where  denotes the Iwasawa algebra
Z p [[]]. Following [18, § 0], we define a pairing

 ,  E,K [ pn ] : E(K [ p n ]) × E(K [ p n ]) −→

by the implicit relation  ,  E ,K [ pn ] = log p ◦ ,  E,K [ pn ] . Here, we should

mention that the construction of  ,  E,K [ pn ] given in [18, § 3.3] requires as
input an auxiliary (idele class) character
log p
ρ K [ p n ] : A× n × −−−−
K [ pn ] /K [ p ] → −−−−→ Z p .

Anyhow, we then define from this a completed group ring element

LHeeg = lim xn , xnσ  E ,K [ pn ] · σ ∈ Z p [[]] ⊗ .

← −
n σ ∈Gal(K [ pn ]/K )

Let I denote the kernel of the natural projection

Z p [[ × ]] −→ Z p [[]].

Let ϑ denote the isomorphism

ϑ : Z p [[]] ⊗ −→ I /I 2
(λ ⊗ γ ) −→ λ(γ − 1).
488 Jeanine Van Order

As explained in [18], we can derive from this the formula

  deg(ϕ)
ϑ LHeeg = · LHeeg (γ − 1).
log p (γ )

Now, observe that since L f,0 = 0 with our hypothesis (i.e. Hypothesis 19.33
(ii)), the two-variable p-adic L-function L f is contained in the kernel I .
Hence, Theorem 19.35 gives us the following identifications in the quotient
I /I 2 :
 
LHeeg ϑ L Heeg
L f = L f,1 (γ − 1) = · (γ − 1) = . (19.34)
K · log p (γ ) K · deg(ϕ)

Let us now keep the hypotheses of Theorem 19.34 in force. We assume addi-
tionally that the absolute Galois group G K = Gal(K /K ) of K surjects onto the
Z p -automorphisms of the p-adic Tate module T p (E) of E, and that p does not
divide the class number of K . Let us now write x∞ ∈ S(E/K p∞ ) to denote
the projective limit of the sequence of norm compatible elements


xn = Norm K [ p n+1 ]/K pn (xn+1 ) ∈ S(E/K pn ).

We again write H∞ = H∞ (E/K p∞ ) to denote the Heegner or CM sub-

module of the compactified Selmer group S∞ = S(E/K p∞ ) generated by
this x∞ . By the main theorem of Cornut [11] (see also [12]), H∞ has rank
one as a -module. By work of Howard [17] (see also [18]) and Bertolini
[1], we also know (i) that X (E/K p∞ ) is a rank one -module, (ii) that
S∞ = S(E/K p∞ ) is free of rank one over  (whence the quotient S∞ /H∞
is a torsion -module), and (iii) that the following divisibility of ideals holds
in :

char X (E/K p∞ )tors | char(S∞ /H∞ ) · char(S∞ /H∞ )∗ . (19.35)

Here again, we have written X (E/K p ∞ )tors to denote the -torsion submodule
of X (E/K p∞ ). Let us now consider the following result of Perrin-Riou [26],
as described in [18, Proposition 0.0.4].

Theorem 19.36 (Perrin-Riou). There exists a p-adic height pairing

hn : S(E/K pn ) × S(E/K pn ) −→ c−1 Z p (19.36)

whose restriction to the image of the Kummer map E(K [ p n ]) ⊗ Z p →

S(E/K [ p n ]) coincides with the pairing  ,  E,K [ pn ] . Here, c ∈ Z p is some
p-adic integer that does not depend on choice of n.
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 489

Using this pairing (19.36), we can then define a pairing

h∞ : S∞ × S∞ −→ c−1 Z p [[]]
(lim an , lim bn ) −→ lim hn (an , bnσ ) · σ.
← − ← − ← −
n n n σ ∈Gal(K pn /K )

We again define the p-adic regulator R = R(E/K p ∞ ) of E over K p∞ to be the

image in c−1 Z p [[]]] of this pairing h∞ . Let e denote the natural projection
e : Z p [[Ω]] −→ Z p [[]].
Following Howard [18] (using [18, Remark 3.2.2] with the uniqueness claims
of [18, Proposition 3.2.1] and the fact that E is ordinary at p), we deduce that
the height pairing h∞ is norm compatible. This allows us to deduce that

e LHeeg = h∞ ( x ∞ ) = char (S∞ /H∞ ) · char (S∞ /H∞ )∗ R (19.37)
x∞ , 
as ideals in c−1 Z p [[]]. On the other hand, we obtain from (19.34) the
identifications
h∞ (
x∞ , x∞ ) char (S∞ /H∞ ) · char (S∞ /H∞ )∗ R
e(L f,1 ) = =
e(K · log p (γ )) e(K · log p (γ ))
as ideals in  = Z p [[]]. Putting this all together, we obtain the following
result.

Corollary 19.37. Assume that D is odd and not equal to −3. Assume addi-
tionally that the absolute Galois group G K = Gal(K /K ) surjects onto the
Z p -automorphisms of T p (E), that η( p) = 1, and that p does not divide the
class number of K . Then, we have the following formal relations:
χ (G, Sel p∞ (A/K p∞ )cotors )
≥ | char(S∞ /H∞ )(0) · char(S∞ /H∞ )∗ (0)|−1
p
≥ | char(S∞ /H∞ )(0) · char(S∞ /H∞ )∗ (0)R(0)|−1
p
= |e(L f,1 (0))|−1
p
= |h∞ ( x∞ )(0)|−1
x∞ ,  p

Proof. Note that e(K · log p (γ )) is a unit, and so we can ignore contributions
from the obvious e(K · log p (γ ))−1 terms. The first inequality follows from the
main conjecture divisibility shown in [19]. The second inequality is trivial, at
least granted the conjectural nontriviality of R. The third and fourth equalities
follow from the Iwasawa theoretic Gross-Zagier theorem shown in [18], as
described above.
490 Jeanine Van Order

Note as well that this result is formal, as we have not assumed the G-Euler
characteristic to be well-defined, i.e. as we have not assumed that the p-adic
regulator R is nondegenerate. We therefore end this section with the following
speculation.

Conjecture 19.38. Assume the conditions of Hypothesis 19.33 are true. Recall
that we let  denote the Iwasawa algebra (G) = Z p [[G]]. Then, (i) the
G-Euler characteristic χ (G, Sel p∞ (A/K p∞ )) is well-defined, and given by
the formula
χ (G, Sel p∞ (A/K p ∞)cotors ) = |char(S∞ /H∞ )(0) · char(S∞ /H∞ )∗ (0)R(0)|−1
p
= |e(L f,1 (0))|−1
p
= |h∞ ( x∞ )(0)|−1
x∞ ,  p

Moreover, (ii) the image in the Iwasawa algebra  of the two-variable p-adic
L-function L f = L f,1 (γ − 1) is not identically zero, (iii) the -adic height
pairing h∞ of Perrin-Riou is nondegenerate, and (iv) the p-adic regulator R =
R(E/K p∞ ) of E over K p∞ is nondegenerate, and satisfies the relation R =
e(K · log p (γ ))−1 .

Remark. Note that the nonvanishing condition for L f presented in part (i)
of this conjecture does not follow a priori from the nonvanishing theorems of
Cornut [11] or Cornut-Vatsal [12]. Those results show here that the complex
values L  (E/K , ρ, 1) do not vanish, where ρ a ring class character of G[ p∞ ]
of sufficiently large conductor. In particular, what is required is a strengthening
of their result to include twists by characters of the cyclotomic Galois group
= Gal(K (μ p∞ )/K ).

Acknowledgements. The author is grateful to John Coates, Christophe Cor-

nut, Olivier Fouquet, Ben Howard, and Dimitar Jetchev for useful discussions,
as well as to an anonymous referee for pointing out a bad error in a earlier
version of this work (which has since been corrected).

References
[1] M. Bertolini, Selmer groups and Heegner points in anticyclotomic Z p -extensions,
Compositio Math. 99 (1995), 153-182.
[2] M. Bertolini and H. Darmon, Heegner points on Mumford-Tate curves, Invent.
math. 126 (1996), 413-456.
[3] M. Bertolini and H. Darmon, Iwasawa’s Main Conjecture for elliptic curves over
anticyclotomic Z p -extensions, Annals of Math. 162 (2005), 1-64.
 Dihedral Euler characteristics of Selmer groups of Abelian varieties 491

[4] N. Bourbaki, Élements de mathématique, Fasc. XXXI, Algèbre commutative,

Chapitre 7: Diviseurs, Actualiés Scientifiques et Industrielles, 1314, Paris,
Hermann, (1965).
[5] C. Breuil, B. Conrad, F. Diamond, and R. Taylor, On the modularity of elliptic
curves over Q: wild 3-adic exercises, J. Amer. Math. Soc. 14 (2001), 843-939.
[6] J. Coates, Fragments of GL2 Iwasawa Theory of Elliptic Curves without Com-
plex Multiplication, Arithmetic Theory of Elliptic Curves (C.I.M.E.) Ed. Viola,
Lecture Notes in Mathematics, Springer (1999), 1-50.
[7] J. Coates and R. Greenberg, Kummer theory for abelian varieties over local fields,
Invent. math. 124 (1996), 129-174.
[8] J. Coates and R. Sujatha, Galois Cohomology of Elliptic Curves, T.I.F.R. Lectures
in Mathematics 88, Narosa Publishing House (1996).
[9] J. Coates, P. Schneider, and R. Sujatha, Links between cyclotomic and G L 2 Iwa-
sawa theory, Doc. Math. Extra Volume: Kazuya Kato’s 50th birthday (2003),
187-215.
[10] J. Coates and A. Wiles, On the Conjecture of Birch and Swinerton-Dyer,
Inventiones math. 39 (1977), 223-251.
[11] C. Cornut, Mazur’s conjecture on higher Heegner points, Inventiones math. 148
(2002), 495-523.
[12] C. Cornut and V. Vatsal, Nontriviality of Rankin-Selberg L-functions and
CM points, L-functions and Galois Representations, Ed. Burns, Buzzard and
Nekovář, Cambridge University Press (2007), 121-186
[13] D.A. Cox, Primes of the form x 2 +ny 2 : Fermat, Class Field Theory, and Complex
Multiplication, John Wiley & Sons (1989).
[14] R. Greenberg, Iwasawa theory for Elliptic Curves, Arithmetic Theory of Elliptic
Curves (C.I.M.E.) Ed. Viola, Lecture Notes in Mathematics, Springer (1999),
50-142.
[15] Y. Hachimori and O. Venjakob, Completely faithful Selmer groups over Kummer
extensions, Doc. Math. Extra Volume: Kazuya Kato’s 50th birthday (2003), 2-36.
[16] H. Hida, A p-adic measure attached to the zeta functions associated to two elliptic
modular forms I, Inventiones math. 79, 159-195 (1985).
[17] B. Howard, The Heegner point Kolyvagin system, Compositio Math. 140 No. 6
(2004), 1439-1472.
[18] B. Howard, The Iwasawa theoretic Gross-Zagier theorem, Compositio Math. 141
No. 4 (2005), 811-846.
[19] B. Howard, Iwasawa theory of Heegner points on abelian varieties of G L 2 -type,
Duke Math. Jour. 124 No. 1 (2004), 1-45.
[20] H. Imai, A remark on the rational points of abelian varieties with values in
cyclotomic Zl -extensions, Proc. Japan. Acad. Math. Sci. 51 (1975), 12-16.
[21] K. Iwasawa, On the Zl -extensions of algebraic number fields, Ann. of Math., 98
No. 2 (1973), 246-326.
[22] V.A. Kolyvagin On the structure of Selmer groups, Math. Ann 291 (1991), no. 2,
253-259.
[23] B. Mazur and K. Rubin, Elliptic curves and class field theory, Proceedings of the
ICM, Beijing 2002, 2 (2002), 185-196.
[24] J. Nekovar, Level raising and anticyclotomic Selmer groups for Hilbert modular
forms of weight 2, to appear in Canadian J. Math.
492 Jeanine Van Order

[25] M. Lazard, Groupes analytiques p-adiques, I.H.E.S. Publ. Math. (1965),

389-603.
[26] B. Perrin-Riou, Fonctions L p-adiques, théorie d’Iwasawa et points de Heegner,
Bull. Soc. Math. France 115 (1987), 399-456.
[27] B. Perrin-Riou, Fonctions L p-adiques associées à une forme modulaire et à un
corps quadratique imaginaire, Jour. London Math. Soc. (2) 38 (1988), 1-32.
[28] B. Perrin-Riou, Théorie d’Iwasawa et hauteurs p-adiques (cas des variétés
abéliennes), Séminaire de Théorie des Nombres, Paris, 1990-1991, Birkh auser,
203-220.
[29] R. Pollack and T. Weston, On anticyclotomic μ-invariants of modular forms, to
appear in Compositio Math.
[30] K. Prasanna, Arithmetic properties of the theta correspondence and periods
of modular forms, Eisenstein series and Applications, Progress in Math. 258,
Birkhauser (2008), 251-269.
[31] K. Ribet and S. Takahashi, Parametrizations of elliptic curves by Shimura curves
and by classical modular curves, Elliptic curves and modular forms (Washington,
DC, 1996), Proc. Nat. Acad. Sci. U.S.A. 94 (1997), 11110-11114.
[32] D. Rohrlich, On L-functions of elliptic curves and cyclotomic towers, Inventiones
math. 75 (1984), 409-423.
[33] J.-P. Serre, Cohomologie Galoisienne, Lecture Notes in Mathematics, 5th Ed.,
Springer, Berlin (1994).
[34] J.-P. Serre, Sur la dimension des groupes profinis, Topology 3 (1965), 413-420.
[35] C. Skinner and E. Urban, The Main Conjecture for G L(2), Inventiones math.
(2013).
[36] J. Tate, On the conjecture of Birch and Swinnerton-Dyer and a geometric
analogue, Seminaire Bourbaki 1965/66, exposé 306.
[37] R. Taylor and A. Wiles, Ring theoretic properties of certain Hecke algebras, Ann.
of Math., 141 (1995), 553-572.
[38] J. Van Order, On the dihedral main conjectures of Iwasawa theory for
Hilbert modular eigenforms, Can. J. Math. (2013), 403-466, available at
http://arxiv.org/abs/1112.3821
[39] J. Van Order, Some remarks on the two-variable main conjecture of Iwasawa
theory for elliptic curves without complex multiplication, J. of Algebra 350, Issue
1 (2012), 273-299.
[40] V. Vatsal, Uniform distribution of Heegner points, Inventiones math. 148 1-46
(2002).
[41] A. Wiles, Modular elliptic curves and Fermat’s last theorem, Ann. of Math. 141
(1995), 443-551.
[42] S.L. Zerbes, Euler characteristics of Selmer groups I, J. London Math. Soc. 70
no. 3 (2004), 586-608.
 20
CM values of higher Green’s functions and
regularized Petersson products
Maryna Viazovska
Department of Mathematical Analysis, Taras Shevchenko National University of Kyiv,
Volodymyrska 64, 01033 Kyiv, Ukraine
E-mail address: [email protected]

A BSTRACT. This paper is a written version of a talk given in the Trimester

Seminar “Arithmetic and Geometry” on 21.02.2012.
Higher Green functions are real-valued functions of two variables on the
upper half-plane, which are bi-invariant under the action of a congruence
subgroup, have a logarithmic singularity along the diagonal, and satisfy the
equation  f = k(1 − k) f ; here  is a hyperbolic Laplace operator and k is
a positive integer. The significant arithmetic properties of these functions were
disclosed in the paper of B. Gross and D. Zagier “Heegner points and deriva-
tives of L-series”(1986). In particular, it was conjectured that higher Green’s
functions have “algebraic” values at the CM points. In the case when k = 2
√
and one of the CM points is equal to −1, the conjecture was proved by A.
Mellit in his doctoral dissertation. In this lecture we prove this conjecture for
arbitrary k, assuming that both CM points under consideration lie in the same
quadratic imaginary field.
The two main parts of the proof are as follows. We first show that the regular-
ized Petersson scalar product of a binary theta series and a weight one weakly
holomorphic cusp form is equal to the logarithm of the absolute value of an
algebraic number and then prove that the special values of weight k Green’s
function, occurring at the conjecture of Gross and Zagier, can be written as the
Petersson product of that type, where the form of weight one is the (k − 1)-
st Rankin-Cohen bracket of an explicitly given weakly holomorphic modular
form of weight 2 − 2k and a binary theta series. Algebraicity of regularized
Petersson products was also proved at about the same time by W. Duke and Y.
Li by a different method; however, our result is stronger than theirs since we
also give a formula for the factorization of the algebraic number in question.

Arithmetic and Geometry, ed. Luis Dieulefait et al. Published by Cambridge University Press.

c Cambridge University Press 2015.

493
494 Maryna Viazovska

Notations
We shall use the following conventions and notations:
Let N := {n| n ∈ Z, n ≥ 1} be the set of natural numbers;
We denote the real and the imaginary part of a complex number z by 3(z)
and 4(z), respectively;
We shall regard Q as a subfield of C.

20.1 CM values of elliptic j-invariant

Let H := {τ | τ ∈ C, 4(τ ) > 0} be the upper half-plane; the group
1 / 2
SL2 (Z) = ac db / a, b, c, d ∈ Z, ad − bc = 1
+b
acts on H by linear fractional transformations τ → aτ cτ +d . The elliptic
j-invariant is the unique holomorphic function on H such that
  a b
+b
1. j aτcτ +d = j (τ ) for any τ ∈ H and c d ∈ SL2 (Z);

2. j (τ ) = q −1 + 744 + ∞ n=1 an q , where q = e
n 2πiτ and a ∈ Z for n ∈ N.
n

A point z ∈ H that satisfies a quadratic equation

Az2 + Bz + C = 0,

where
A, B, C ∈ Z, (A, B, C) = 1, B 2 − 4 AC = −D < 0,

is called a point of complex multiplication (a CM point) of discriminant −D.

Let Z D be the set of the CM-points of discriminant −D on SL2 (Z)\H.
√
Let K := Q( −D) and let D = f 2 D̃ with f ∈ N and − D̃ a fundamental
discriminant.
The values of j-invariant at the CM points are known as singular mod-
uli. The following classical result is often called “The first main theorem of
complex multiplication”.

Theorem 20.1. For each z ∈ Z D the value j (z) is an algebraic integer. More-
over, the field H D = K ( j (z)) is abelian over K and dihedral over Q; it is the
ring class field of conductor f over K .

Another remarkable property of the singular moduli is that their differ-

ences are highly divisible numbers. An explicit factorization formula for these
 CM values of higher Green’s functions 495

numbers was discovered by B. Gross and D. Zagier [4]. In particular, they

computed the prime factorization of the integer numbers
 
A(D1 , D2 ) := ( j (z1 ) − j (z2 )) ∈ Z, (20.1)
z1 ∈Z D1 z2 ∈Z D2

where −D1 and −D2 are coprime fundamental discriminants, and for the
algebraic numbers
j (z1 ) − j (z2 ), (20.2)

where z1 , z2 ∈ Z D and −D is an odd prime discriminant. Now we will state

the factorization formula for the prime ideal generated by j (z1 ) − j (z2 ).
√
We denote by K the imaginary quadratic field Q( −D), by h the class
number,
p and by H the Hilbert class field of K . For a rational prime p with
D = −1 let P p = {Pi }i=1
h
be the set of prime ideals of H lying above p.
Recall that we consider H as a subset of C. The complex conjugation acts on
H and also on the set P p . There exists the unique prime ideal in P p , say P1 ,
such that P1 = P1 (see [4]). For each prime ideal P ∈ P p there exists a
unique element σ ∈ Gal(H/K ) such that

Pσ = P1 . (20.3)

Let A = A(P) be the ideal class of K that corresponds to σ under the Artin
isomorphism.

Theorem 20.2. ([4, Proposition 3.8]) Let z1 , z2 be two CM points of prime

odd discriminant −D. Denote by B and C, respectively, the ideal classes of
√
Q( −D) containing the fractional ideals Zz1 + Z and Zz2 + Z, respectively.
Then for a rational prime p and a prime P lying above p in the Hilbert class
field H we have
D n   p
ordP ( j (z1 ) − j (z2 )) = rBC (D − n) rBCA2 ord p (n) if = −1,
p D
n=0
(20.4)
 p
ordP ( j (z1 ) − j (z2 )) = 0 if = 1. (20.5)
D
Here rC (n) stands for the number of integral ideals of norm n in the ideal
class C.

The key technique used in the proof of Theorem 20.2 is the computation of
local height pairings between images of CM points on the modular curve X 1 .
These technique was further developed in the well-known paper [5].
496 Maryna Viazovska

20.2 CM values of higher Green’s functions

For any integer k > 1 and subgroup ⊂ PSL2 (Z) of finite index there is
\H
a unique function G k on the product of two upper half planes H × H that
satisfies the following conditions:
\H
1. G k is a smooth function on H × H \ {(τ, γ τ ), τ ∈ H, γ ∈ } with values
in R.
\H \H
2. G k (τ1 , τ2 ) = G k (γ1 τ1 , γ2 τ2 ) for all γ1 , γ2 ∈ .
\H \H
3. i G k = k(1 − k)G k , where i is the hyperbolic Laplacian with
respect to the i-th variable, i = 1, 2.
\H
4. G k (τ1 , τ2 ) = m log |τ1 − τ2 | + O(1) when τ1 tends to τ2 (m is the order
of the stabilizer of τ2 , which is almost always 1).
\H
5. G k (τ1 , τ2 ) tends to 0 when τ1 tends to a cusp.

These functions have been introduces in [5] and are called higher Green’s
functions. Existence of a function satisfying 1–5 is shown in [5] by an explicit
construction and the uniqueness follows from the maximum principle for
subharmonic functions.
\H
In the case k = 1, there is a unique function G 1 (τ1 , τ2 ) satisfying
conditions 1-4 and condition 5 should be slightly modified in that case.
Let S2k ( ) be the space of cusp forms of weight 2k for a group ⊂
PSL2 (Z). A sequence of integer numbers λ = {λm }∞ , which belongs to
∞
∞ m=1
⊕m=1 Z, is called a relation for S2k ( ) if m=1 λm am = 0 for any cusp form

f = ∞ m=1 am q ∈ S2k ( ).
m

We shall study the CM values of the functions

∞
PSL (Z)\H PSL2 (Z)\H
G k,λ 2 := λm m k−1 G k (τ1 , τ2 )|Tm
m=1

where Tm is a Hecke operator and λ is a relation for S2k (SL2 (Z)). We shall
PSL (Z)\H
write for simplicity G k,λ := G k,λ 2 .

Conjecture 20.3. Suppose λ is a relation for S2k (SL2 (Z)). Then for any pair
of CM points z1 ∈ Z D1 and z2 ∈ Z D2 there is an algebraic number α such that
1−k
G k,λ (z1 , z2 ) = (D1 D2 ) 2 log |α|.

This conjecture was formulated in [5] for D1 = D2 and in [6] for any pair
of CM points. Moreover, D. Zagier has made a more precise conjecture about
the field of definition and prime factorization of this number α [12].
 CM values of higher Green’s functions 497

A. Mellit has proved Conjecture 20.1 in the case k = 2, D1 = −4, and

arbitrary D2 in his doctoral dissertation [8]. In the thesis [11] this conjec-
ture is proved for arbitrary k and any pair of CM points under the additional
assumption that both CM points under consideration lie in the same imaginary
quadratic field. Furthermore, in this case we have found the prime factorization
of the algebraic number α.

20.3 Regularized theta lifts

The main technique used in [11] is the theory of regularized theta lifts, which
was developed by Borcherds [1], Bruinier [2], Kudla [7], and others. Let us
give a brief review of this theory.
Let V be a finite dimensional rational vector space and let Q : V → Q be a
quadratic form. The corresponding bilinear form on V × V is defined as
1
(x, y) := (Q(x + y) − Q(x − y)).
2
Let L ⊂ V be a lattice. The dual lattice of L is defined as
L  := {x ∈ V |(x, L) ⊂ Z}.
The lattice L is said to be even if (l, l) ∈ 2Z for all l ∈ L. In that case L  ⊂ L
and L  /L is a finite abelian group. We denote by Aut(L) the group of those
isometries of L ⊗ R that fix each coset of L in L  .
Suppose now that L is an even lattice of signature (b+ , b− ). Denote
by Gr+ (L) the positive Grassmanian, i.e. the set of positive-definite b+ -
dimensional subspaces of L ⊗ R. For lattices L of signature (2, b)
the Grassmanian Gr+ (L) has the structure of a Hermitian symmetric
domain.

Example 20.4. Consider the lattice S2 (Z) of 2×2 symmetric integral matrices
equipped with the quadratic form Q(x) := − det(x). This is an even lattice
of signature (2, 1). The Grassmanian Gr+ (S2 (Z)) is isomorphic to the upper
half-plane H. There is an isomorphism H → Gr+ (S2 (Z)) given by
   
+ + τ2 τ τ2 τ
τ → v (τ ), v (τ ) := 3 R+4 R,
τ 1 τ 1
v + (τ ) ⊂ S2 (Z) ⊗ R. (20.6)
The group SL2 (Z) acts on the lattice S2 (Z) by γ (x) = γ xγ t , x ∈ S2 (Z),
γ ∈ SL2 (Z). This action preserves the quadratic form Q.
498 Maryna Viazovska

Example 20.5. The lattice M2 (Z) of 2 × 2 integral matrices equipped with the
quadratic form Q(x) := − det(x) is an even unimodular lattice of signature
(2, 2). The Grassmanian Gr+ (M2 (Z)) is isomorphic to H×H. An isomorphism
H × H → Gr+ (M2 (Z)) is given by
   
+ + τ1 τ2 τ1 τ1 τ2 τ1
(τ1 , τ2 ) → v (τ1 , τ2 ), v (τ1 , τ2 ) := 3 R+4 R,
τ2 1 τ2 1
(20.7)
v + (τ1 , τ2 ) ⊂ M2 (Z) ⊗ R.
The group SL2 (Z)×SL2 (Z) acts on the lattice M2 (Z) by (γ1 , γ2 )(x) = γ1 xγ2t ,
x ∈ M2 (Z), (γ1 , γ2 ) ∈ SL2 (Z) × SL2 (Z).

Recall that the group SL2 (Z) has a double cover Mp2 (Z), called the
metaplectic group. Elements of the metaplectic group can be written in the
form
  
a b √
, cτ + d
c d
 
a b √
where ∈ SL2 (Z) and cτ + d is one of the two holomorphic func-
c d
tions of τ in the upper half-plane whose square is cτ + d. The multiplication
is defined so that the usual formulas for the transformation of modular forms
of half integral weight work, i.e.
(A, f (τ ))(B, g(τ )) = ( AB, f (B(τ ))g(τ ))
for A, B ∈ SL2 (Z) and f, g are square roots of c A τ + d A and c B τ + d B ,
respectively. The Weil representation ρ L is a unitary representation of Mp2 (Z)

on the vector space C L /L (see [1] for a definition). A vector-valued modular
form of half-integral weight k and representation ρ L is a function f : H →

C L /L that satisfies the following transformation law
    
aτ + b √ 2k a b √
f = cτ + d ρ L , cτ + d f (τ ). (20.8)
cτ + d c d
For an even lattice L of signature (2, b) the vector-valued Siegel theta

function # L : H × Gr+ (L) → C L /L is defined as follows

# L (τ, v + ) := y b/2 eλ e Q(v + )τ + Q(v − )τ̄ . (20.9)
λ∈L  /L ∈L+λ

Here {eλ }λ∈L  /L is the canonical basis of C L /L , e(x) := e2πi x , and v + and
v− denote the orthogonal projections of a vector  ∈ V to the subspaces v +
and (v + )⊥ , respectively.
 CM values of higher Green’s functions 499

Theorem 4.1 in [1] asserts that # L (τ, v + ) satisfies the following properties

1. it is real-analytic in both variables;

2. transforms as a vector-valued modular form of weight 1 − b/2 and
representation ρ L with respect to the variable τ ;
3. is invariant in the variable v + under the group Aut(L) ⊂ O(L ⊗ R).

Let Mk (ρ L ) be the space of real-analytic (vector-valued) modular forms of

weight k and representation ρ L . For f ∈ M1−b/2 (ρ L ) the theta lift + L is
defined as follows:
.  
+ L (v + , f ) := f (τ ), # L (τ, v + )  y −1−b/2 d x d y (20.10)
L /L
SL2 (Z)\H

(here the product (·, ·) L  /L is defined by eμ , eν L  /L = δμ,ν ).
The integral (20.10) is often divergent and has to be regularized. For f, g ∈
Mk (ρ L ) we define the regularized Petersson product as follows
.  
( f, g)reg := lim f (τ ), # L (τ, v + )  y −1−b/2 d x d y,
T →∞ FT L /L

where
1 / 2
FT := τ ∈ H/ − 1/2 < 3(τ ) < 1/2, |τ | > 1, and 4(τ ) < T

is a truncated fundamental domain of SL2 (Z).

The regularized theta lift of f ∈ Mk (ρ L ) is defined as

+ L (v + , f ) := ( f (τ ), # L (v + , τ ))reg .

It is clear that the regularized integral + L (v + , f ) is invariant under the action

of the group Aut(L) on the Grassmannian Gr+ (L).
The regularized theta lift is well-defined for a much larger class of modular
forms than the classical theta lift. For instance, regularized theta lift can be
computed for weakly holomorphic cusp forms. A weakly holomorphic (vector-
valued) modular form is called a weakly holomorphic cusp form if its constant
term in the Fourier expansion is zero. Let Sk! (ρ) be the space of weakly
holomorphic cusp forms of weight k and representation ρ.

!
Theorem 20.6. ([1, Theorem 13.3]) Let f ∈ S1−b/2 (ρ L ), let

f (τ ) = eλ cλ (n) e(nτ )
λ∈L  /L n−∞
500 Maryna Viazovska

and suppose that the Fourier coefficients cλ (n) ∈ Z for n ≤ 0. Then there
is a meromorphic function  L (Z , f ) on Gr+ (L), satisfying the following
conditions:

1.  L is an automorphic function for the group Aut(L) with respect to a

unitary character of Aut(L);
2. The zeros and the poles of  L lie on the rational quadratic divisors ⊥ for
 ∈ L with Q() < 0. The order of  L at l ⊥ is equal to

−4 cxl Q(xl) ;
x∈R+ :
xl∈L 

3.
+ L (v + , f ) = log | L (v + , f )|;

4. One can write an explicit infinite product expansion converging in a

neighborhood of each cusp of Gr+ (L).

Let us consider several examples of regularized theta lifts associated to

the lattice M2 (Z). Recall that the Grassmanian Gr+ (M2 (Z)) is isomorphic
to H × H. The theta function #(τ, τ1 , τ2 ) := #M2 (Z) (τ, v + (τ1 , τ2 )) is a real
analytic modular form of weight 0 in each variable.
∞
Example 20.7. Set J1 := j (τ ) − 744 = q −1 + n=1 an q
n . By Theorem 20.6

+M2 (Z) (v + (τ1 , τ2 ), J1 ) = log | j (τ1 ) − j (τ2 )|.

Example 20.8. Let λ be a relation for S2k (SL2 (Z)). Then by Serre dual-
ity there exists the unique gλ ∈ M2−2k ! with the Fourier expansion gλ =
 −m

m λ m q + n≥0 b n q n . It has been proved in [2] that

+M2 (Z) (v + (τ1 , τ2 ), R k−1 (gλ )) = G k,λ (τ1 , τ2 ).

This result provides the connection between regularized theta lifts and higher
∂
Green’s functions. Here Rk = 2πi 1
( ∂τ + τ −k τ̄ ) is the raising Maass operator
and R k−1 = R−2 ◦ · · · ◦ R2−2k .

20.4 CM-values of the theta lifts

In this section we shall study the CM values of the theta lifts.
 CM values of higher Green’s functions 501

Let L be an even lattice of signature (2, b1 ) and let N ⊂ L be an even lattice

of signature (2, b2 ). Then there is an embedding i : Gr+ (N )/Aut(N ) →
Gr+ (L)/Aut(L).

Example 20.9. The pairs of the CM points z1 , z2 such that Q(z1 ) = Q(z2 )
are in one-to-one correspondence with signature (2, 0) rational subspaces of
M2 (Q).

For simplicity, assume that L = M ⊕ N . Then for f ∈ M1−b1 /2 (ρ L )

+ L (i (v + ), f ) = + N (v + , ( f, # M ) M  /M ).

Therefore, the CM-values of theta lifts can be expressed as regularized Peters-

son products between (real-analytic) weight one modular forms and binary
theta series.
In [11], we have obtained the following result for regularized Petersson
products of such type.

Theorem 20.10. ([11, Theorem 5.6]) Let N be an even lattice of signature

(2, 0) and let f ∈ S1! (ρ N )Z . Then

( f, # N )reg = log |α| (20.11)

for some α ∈ Q.

After this result had been announced [10], the author learned that a similar
result was obtained independently in a slightly different context by W. Duke
and Y. Li [3].
Furthermore, we obtained the prime factorization of the number α.

Theorem 20.11. ([11, Theorem 5.8]) Suppose that −D is an odd prime dis-
√
criminant and K := Q( −D). Given a fractional ideal B ⊂ K , consider the
following even lattice
 
N K /Q (·)
(N , Q) = B, .
N K /Q (B)

Let f ∈ S1! (ρ N ) be a weakly holomorphic cusp form with the Fourier

expansion
t 
f = eν cν (t) e τ .
D
ν∈Z/DZ t∈Z
t−∞
502 Maryna Viazovska

Suppose that the Fourier coefficients cν (n) ∈ Z for n ≤ 0. Then there exists a
number α in the Hilbert class field H of K such that
( f, #B )reg = log |α|.
Moreover, for a rational prime p and a prime P lying above p in the Hilbert
class field H we have
 −t  p
ordP (α) = cν (t) rBA2 ord p (t) in the case = −1,
p D
t<0 ν∈Z/DZ
(20.12)
p
ordP (α) = 0 in the case = 1. (20.13)
D
Here rC (n) denotes the number of integral ideals on norm n in the ideal class C.

Finally, the following result relates CM values of higher Green’s functions

and regularized Petersson products (20.11).

Theorem 20.12. ([11, Theorem 5.4]) Let z1 , z2 be two CM points of nega-

tive fundamental discriminant −D. Denote by B and C, respectively, the ideal
√
classes of Q( −D) containing the fractional ideals Zz1 + Z and Zz2 + Z,
respectively. Let λ be a relation for S2k . Then
 reg
G k,λ (z1 , z2 ) = [gλ , #BC ]k−1 , #BC , (20.14)
where gλ is defined in Example 20.8 and [·, ·]k−1 denotes the (k −1)-st Rankin-
Cohen brackets.

This result combined with Theorem 20.10 proves Conjecture 20.1 for each
pair of CM points lying in the same imaginary quadratic field. Finally, Theo-
rems 20.11 and 20.12 imply the factorization formula for exp(G k,λ (z1 , z2 )).

References
[1] R. E. Borcherds, Automorphic forms with singularities on Grassmannians,
Invent. Math. 132, 491–562 (1998).
[2] J. H. Bruinier, Borcherds products on O(2, l) and Chern classes of Heegner
divisors, Springer Lecture Notes in Mathematics 1780, Springer-Verlag (2002).
[3] W. Duke, Y. Li, Mock-modular forms of weight one, preprint, http://www.
math.ucla.edu/~wdduke/preprints/mock%20weight%20one.
pdf, (2012).
[4] B. Gross and D. Zagier, On singular moduli, J. Reine Angew. Math. 355, 191–220
(1985).
 CM values of higher Green’s functions 503

[5] B. Gross, D. Zagier, Heegner points and derivatives of L-series, Invent. Math. 84,
225–320 (1986).
[6] B. Gross, W. Kohnen, D. Zagier, Heegner points and derivatives of L-series. II
Math. Ann. 278, 497–562 (1987).
[7] S. Kudla, Integrals of Borcherds forms, Compositio Math. 137 (2003), 293–349.
[8] A. Mellit, Higher Greens functions for modular forms, PhD dissertation, Univer-
sity of Bonn (2008).
[9] A. Néron, Quasi-fonctions et hauteurs sur les variétés abéliennes, Ann. of Math.
82, 249–331 (1965).
[10] M. Viazovska, CM Values of Higher Green’s Functions, arXiv:1110.4654
[math.NT] (2011).
[11] M. Viazovska, Modular functions and special cycles, PhD dissertation, University
of Bonn (2013).
[12] D. Zagier, Private communication.