on local galois representations attached to automorphic forms - IWR
on local galois representations attached to automorphic forms - IWR
on local galois representations attached to automorphic forms - IWR
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
ON LOCAL GALOIS REPRESENTATIONS<br />
ATTACHED TO AUTOMORPHIC FORMS<br />
A Thesis<br />
Submitted <strong>to</strong> the<br />
Tata Institute of Fundamental Research, Mumbai<br />
for the degree of Doc<strong>to</strong>r of Philosophy<br />
in Mathematics<br />
by<br />
VG NARASIMHA KUMAR CH<br />
School of Mathematics<br />
Tata Institute of Fundamental Research<br />
Mumbai<br />
November, 2010
DECLARATION<br />
This thesis is a presentati<strong>on</strong> of my original research work. Wherever c<strong>on</strong>tributi<strong>on</strong>s<br />
of others are involved, every effort is made <strong>to</strong> indicate this clearly, with due reference<br />
<strong>to</strong> the literature, and acknowledgement of collaborative research and discussi<strong>on</strong>s.<br />
The work was d<strong>on</strong>e under the guidance of Professor Eknath Ghate, at the Tata<br />
Institute of Fundamental Research, Mumbai.<br />
[VG Narasimha Kumar CH]<br />
In my capacity as supervisor of the candidate’s thesis, I certify that the above<br />
statements are true <strong>to</strong> the best of my knowledge.<br />
[Professor Eknath Ghate]<br />
Date:
C<strong>on</strong>tents<br />
1 Synopsis 11<br />
1.1 (p, p)-Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> au<strong>to</strong>morphic <strong>forms</strong> <strong>on</strong> GL n . . . 11<br />
1.2 2-adic Hida theory and applicati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . . . 14<br />
2 Introducti<strong>on</strong> 17<br />
3 Modular <strong>forms</strong> and Hecke algebras 25<br />
3.1 Modular <strong>forms</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25<br />
3.2 Hecke algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27<br />
3.2.1 Hecke opera<strong>to</strong>rs for Γ 1 (N) . . . . . . . . . . . . . . . . . . . . . . . 27<br />
3.2.2 Old<strong>forms</strong> and New<strong>forms</strong> . . . . . . . . . . . . . . . . . . . . . . . . 28<br />
3.2.3 Complex multiplicati<strong>on</strong> . . . . . . . . . . . . . . . . . . . . . . . . 29<br />
4 Galois Representati<strong>on</strong>s and Au<strong>to</strong>morphic Forms <strong>on</strong> GL n 31<br />
4.1 Motivati<strong>on</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31<br />
4.2 p-adic Hodge theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33<br />
4.2.1 F<strong>on</strong>taine’s rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33<br />
4.2.2 Newt<strong>on</strong> and Hodge numbers . . . . . . . . . . . . . . . . . . . . . 34<br />
4.2.3 Potentially semistable representati<strong>on</strong>s . . . . . . . . . . . . . . . . 35<br />
4.2.4 Weil-Deligne representati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . . . 36<br />
4.3 The case of GL 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37<br />
4.3.1 Proof of Wiles’ theorem . . . . . . . . . . . . . . . . . . . . . . . . 40<br />
4.4 The case of GL n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40<br />
4.4.1 Local Langlands corresp<strong>on</strong>dence . . . . . . . . . . . . . . . . . . . 40<br />
4.4.2 Au<strong>to</strong>morphic <strong>forms</strong> <strong>on</strong> GL n . . . . . . . . . . . . . . . . . . . . . . 42<br />
4.4.3 Galois representati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . . . . . . . 42<br />
4.4.4 A variant, following [CHT08] . . . . . . . . . . . . . . . . . . . . . 43<br />
4.4.5 Newt<strong>on</strong> and Hodge filtrati<strong>on</strong> . . . . . . . . . . . . . . . . . . . . . 44<br />
4.4.6 Ordinary and quasi-ordinary representati<strong>on</strong>s . . . . . . . . . . . . 44<br />
5 On Local Galois Representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> Au<strong>to</strong>morphic Forms 45<br />
5.1 Introducti<strong>on</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45<br />
5.2 Principal series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46<br />
5.2.1 Spherical case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46<br />
5.2.2 Ramified principal series case . . . . . . . . . . . . . . . . . . . . . 48<br />
5.3 Steinberg case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49<br />
5
5.3.1 Unramified twist of Steinberg . . . . . . . . . . . . . . . . . . . . . 49<br />
5.3.2 Ramified twist of Steinberg . . . . . . . . . . . . . . . . . . . . . . 51<br />
5.4 Supercuspidal ⊗ Steinberg . . . . . . . . . . . . . . . . . . . . . . . . . . . 52<br />
5.4.1 m = 2 and n = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53<br />
5.4.2 τ unramified supercuspidal of dim m ≥ 2 and n ≥ 2 . . . . . . . . 56<br />
5.4.3 General case: m ≥ 2 and n ≥ 2 . . . . . . . . . . . . . . . . . . . . 58<br />
5.5 General Weil-Deligne representati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . . . 62<br />
5.5.1 Sum of twisted Steinberg . . . . . . . . . . . . . . . . . . . . . . . 62<br />
5.5.2 General ordinary case . . . . . . . . . . . . . . . . . . . . . . . . . 64<br />
5.6 Further remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66<br />
6 2-adic Hida theory and Applicati<strong>on</strong>s 69<br />
6.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69<br />
6.1.1 Group and sheaf cohomology . . . . . . . . . . . . . . . . . . . . . 70<br />
6.1.2 Hecke acti<strong>on</strong> <strong>on</strong> the cohomology groups . . . . . . . . . . . . . . . 71<br />
6.1.3 Geometry of Riemann surfaces . . . . . . . . . . . . . . . . . . . . 72<br />
6.1.4 Hida’s idempotent opera<strong>to</strong>r . . . . . . . . . . . . . . . . . . . . . . 72<br />
6.2 Relati<strong>on</strong>s between cohomology groups with coefficients . . . . . . . . . . . 73<br />
6.3 Main theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76<br />
6.4 C<strong>on</strong>trol theorem for cohomology . . . . . . . . . . . . . . . . . . . . . . . 77<br />
6.5 Freeness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82<br />
6.6 C<strong>on</strong>stant rank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85<br />
6.7 Λ-adic Hecke algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87<br />
6.8 C<strong>on</strong>trol theorem for ordinary Hecke algebras . . . . . . . . . . . . . . . . 89<br />
6.9 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91<br />
6.9.1 2-adic Λ-adic <strong>forms</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . 92<br />
6.9.2 Tame level N = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />
6.9.3 Uniqueness result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />
6.10 An applicati<strong>on</strong> <strong>to</strong> Galois representati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . 94<br />
6.10.1 Buzzard’s result . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95<br />
6.10.2 Λ-adic Galois representati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . . . 96<br />
6.10.3 Local splitting for Λ-adic <strong>forms</strong> . . . . . . . . . . . . . . . . . . . . 97<br />
6.10.4 Descending <strong>to</strong> the classical situati<strong>on</strong> . . . . . . . . . . . . . . . . . 98<br />
A A 2-adic C<strong>on</strong>trol Theorem for Modular Curves 99<br />
A.1 Introducti<strong>on</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99<br />
A.2 Notati<strong>on</strong>s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99<br />
A.3 Hecke opera<strong>to</strong>rs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101<br />
A.4 Ordinary parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104<br />
A.5 Iwasawa modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105<br />
A.6 Limits of cohomology modules . . . . . . . . . . . . . . . . . . . . . . . . 107<br />
6
This thesis is dedicated <strong>to</strong><br />
my parents for their endless Love, Encouragement, and Belief.
ACKNOWLEDGEMENTS<br />
This thesis would not have been possible without the warm encouragement and the guidance<br />
of my supervisor Prof. Eknath Ghate. I owe my deepest gratitude <strong>to</strong> him for his<br />
patience and care. During all these years, I learnt not <strong>on</strong>ly mathematics but also many<br />
other things which made me a better human being.<br />
I take this opportunity <strong>to</strong> thank all the members of the School of Mathematics for<br />
their patience and help.<br />
I wish <strong>to</strong> thank Prof. S. Bhattacharya, Prof. N. Nitsure, and Prof. V. Srinivas for the<br />
excellent first year graduate courses. I would also like <strong>to</strong> thank Prof. S. M. Bhatwadekar,<br />
Prof. R. V. Gurjar, and Prof. D. Prasad for the sec<strong>on</strong>d year graduate courses.<br />
I wish <strong>to</strong> thank Prof. N. Nitsure not <strong>on</strong>ly for his help at the beginning of my Ph. D.,<br />
but also for keeping an eye <strong>on</strong> me throughout. Over the years, I gradually unders<strong>to</strong>od<br />
Prof. V. Srinivas’s thoughtful comments about life and mathematics. I really appreciate<br />
them now.<br />
I wish <strong>to</strong> express my deep-felt regards <strong>to</strong> Prof. D. Prasad for the mathematical<br />
discussi<strong>on</strong>s at TIFR and also at Number theory seminars at IIT, Mumbai. I also thank<br />
him for his encouragement and support at various stages.<br />
I would like <strong>to</strong> thank Prof. R. Sujatha for various discussi<strong>on</strong>s at TIFR and also during<br />
workshops at Chennai and Guwahati. I would also like <strong>to</strong> thank Prof. C. S. Rajan and<br />
Prof. Saradha for discussing mathematics <strong>on</strong> several occasi<strong>on</strong>s.<br />
I was surprised by the amount of energy and enthusiasm of Prof. M. S. Raghunathan<br />
even at this age. I thank him for explaining various mathematical c<strong>on</strong>cepts, their origins,<br />
and many more things. I also thank Prof. Ravi Rao for being very co-operative at many<br />
places, and also for providing helpful suggesti<strong>on</strong>s whenever needed.<br />
I wish <strong>to</strong> thank Prof. H. Hida for answering several of my questi<strong>on</strong>s patiently at several<br />
stages of my Ph. D. Thanks <strong>to</strong> Prof. K. Buzzard and Prof. M. Emert<strong>on</strong> for their helpful<br />
e-mail communicati<strong>on</strong>s at the beginning of my Ph. D. I would like <strong>to</strong> thank Prof. U. K.<br />
Anandavardhanan and Prof. R. Raghunathan for making me a part of various Number<br />
theory activities at IIT, Mumbai. Special thanks <strong>to</strong> Prof. U. K. Anandavardhanan for<br />
being friendly and helpful since the beginning of my Ph. D.<br />
This work was partly completed during a visit <strong>to</strong> Université Paris-Sud-11, Paris supported<br />
by the ARCUS program (Régi<strong>on</strong> Ile-de-France). I am grateful <strong>to</strong> Prof. L. Clozel<br />
and Prof. J. Tilouine for arranging the visit. I would also like <strong>to</strong> thank Hausdorff Research<br />
Institute for Mathematics, B<strong>on</strong>n, where part of this work was carried out during<br />
the Trimester Program “Algebra and Number Theory”.<br />
I would like <strong>to</strong> thank Profs. C. Breuil, D. A. Elwood, A. Mézard, M-F. Vignéras, and<br />
J-P. Wintenberger, the organizers of the “Galois Trimester” at IHP, Paris for providing<br />
me a platform <strong>to</strong> talk <strong>to</strong> hundreds of mathematicians from all over the world. I benefitted<br />
from the various courses and the c<strong>on</strong>ferences held during the “Galois Trimester”.<br />
Words are inadequate in offering my thanks <strong>to</strong> my teacher Prof. M. Perisastri, whose<br />
influence first made mathematics interesting for me during my undergraduate days.<br />
I take this opportunity <strong>to</strong> thank all the members of Department of Mathematics and<br />
Statistics, University of Hyderabad. My special thanks <strong>to</strong> Prof. T. Amaranath, Prof. M.<br />
9
S. Datt, Prof. V. Kannan, Prof. C. Musili (late), Prof. V. Suresh, and Prof. R. Tand<strong>on</strong><br />
for their excellent courses at M. Sc., and also for their encouragement in pursuing further<br />
studies.<br />
I have benefitted from Mathematics Training and Talent Search, Advanced Foundati<strong>on</strong>al<br />
Schools and Advanced Instructi<strong>on</strong>al Schools. I thank all the organizers, especially<br />
Prof. S. Kumaresan, Prof. A. R. Shastri, and Prof. J. K. Verma, for c<strong>on</strong>ducting such<br />
perfect programs.<br />
I take this opportunity <strong>to</strong> thank my M. Sc. batchmates Ayyangar, Damodar, Devakar,<br />
Karthik, Keshav, Pratyusha, Preena, Sanjay, Siva, and Srikanth for their c<strong>on</strong>tinuous<br />
encouragement. I am also thankful <strong>to</strong> Prem, Shyam, Ved and Ranjana.<br />
I would like <strong>to</strong> thank all my friends at TIFR for making my stay enjoyable. I would<br />
specially like <strong>to</strong> thank Alok, Amala, Amit, Arati, Arijit, Arnab, Chandrasheel and Prachi,<br />
Debargha, Illangovan, Niraj, R<strong>on</strong>nie, Sagar, Sandeep, Satadal, Shanta, Shilpa, Somnath,<br />
Souradeep, Sudarshan, Vaibhav, and Vivek from Mathematics, Argha, Aditya, Bhargav,<br />
Manjusha, Nikhil, Partha, Rakesh, Sangeetha, Sashi, and Satej from Physics, Atul, Krishnamohan,<br />
Manoj, and Ram from Chemistry, Aksar, Benny, Harshit, Kalyan, Mallickarjun,<br />
Santanu, and Saswata from Computer Science, Harinath, and Sunilnoothi from Biology.<br />
Each <strong>on</strong>e has some good qualities which are unique <strong>to</strong> them. I tried <strong>to</strong> acquire them but<br />
I succeeded <strong>on</strong>ly a bit.<br />
I am short of words <strong>to</strong> express my sincere thanks <strong>to</strong> Chandrakant, Naren HR, Narendra,<br />
Sarang, and Vijay for our l<strong>on</strong>g discussi<strong>on</strong>s <strong>on</strong> various things, and also for the amount<br />
of help that I <strong>to</strong>ok from them in, and out of, mathematics.<br />
I would like <strong>to</strong> thank all the members, particularly Mr. D. B. Sawant and Mr. K.<br />
G. Jayaraj, of the School of Mathematics office, for making all administrative issues so<br />
simple and smooth. Prof. P. A. Gastesi and Mr. V. Nandagopal deserve special thanks<br />
for their help with software related issues.<br />
Above all, I take immense pleasure in thanking my parents, whose support has been<br />
instrumental in overcoming several hurdles in my life, my siblings and their families for<br />
their extended support, and finally <strong>to</strong> my late grand mother and my late uncle Ch. S. P.<br />
Ranga Rao, who <strong>to</strong>ld me about many aspects of life during my childhood. I owe every<br />
bit of my existence <strong>to</strong> them.<br />
10
Chapter 1<br />
Synopsis<br />
My thesis c<strong>on</strong>sists of two parts, <strong>on</strong>e a general result and the other rather specific. The former<br />
deals with the irreducibility of the (p, p)-Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> au<strong>to</strong>morphic<br />
representati<strong>on</strong>s of GL n (A Q ), for n ≥ 3, and the latter deals with the semisimplicity<br />
of the <strong>local</strong> Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> normalized ordinary cuspidal eigen<strong>forms</strong><br />
(n = 2), for the prime p = 2.<br />
The starting point for both parts of my thesis is the following theorem, which is due<br />
<strong>to</strong> Wiles. Let f = ∑ ∞<br />
n=1 a n(f)q n be a normalized cuspidal eigenform of weight k ≥ 2. Let<br />
K f denote the number field generated by the a n (f)’s. Let G p denote the decompositi<strong>on</strong><br />
group at ℘, where ℘ denotes the prime induced by a fixed embedding of ¯Q in ¯Q p . There<br />
exists a Galois representati<strong>on</strong><br />
ρ f,℘ : Gal( ¯Q/Q) → GL 2 (K f,℘ ),<br />
associated <strong>to</strong> f (and ℘). If f is ordinary at ℘, i.e., a p (f) is a ℘-adic unit, then the<br />
representati<strong>on</strong> ρ f,℘ | Gp , the restricti<strong>on</strong> of ρ f,℘ <strong>to</strong> G p , is reducible [Wil88]. Moreover, the<br />
powers of the p-adic cyclo<strong>to</strong>mic character χ cyc,p occurring in the semisimplificati<strong>on</strong> of<br />
ρ f,℘ | Gp are distinct and related <strong>to</strong> the weight k of f.<br />
The <strong>local</strong> representati<strong>on</strong> above is referred <strong>to</strong> as the (p, p)-Galois representati<strong>on</strong> <strong>attached</strong><br />
<strong>to</strong> f.<br />
1.1 (p, p)-Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> au<strong>to</strong>morphic<br />
<strong>forms</strong> <strong>on</strong> GL n<br />
In the first part our thesis (cf. [GK1]), we study the <strong>local</strong> reducibility at p of the p-adic<br />
Galois representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ),<br />
assuming that the global representati<strong>on</strong> lives in a strictly compatible system of Galois<br />
representati<strong>on</strong>s. In the case that the underlying Weil-Deligne representati<strong>on</strong> is Frobenius<br />
semisimple and indecomposable, we analyze the reducibility completely. We use methods<br />
from p-adic Hodge theory, and work under a transversality assumpti<strong>on</strong> <strong>on</strong> the Hodge and<br />
Newt<strong>on</strong> filtrati<strong>on</strong>s in the corresp<strong>on</strong>ding filtered module.<br />
Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ) with infinitesimal character<br />
H c<strong>on</strong>sisting of n-distinct integers. The source of (p, p)-Galois representati<strong>on</strong>s is<br />
the following c<strong>on</strong>jecture [Tay04, C<strong>on</strong>j. 3.4] which has been proved by Kottwitz, Clozel,<br />
11
and Harris-Taylor, for certain self-dual cuspidal π’s, and is also true when n = 2 and<br />
π = π(f), for f as above.<br />
C<strong>on</strong>jecture: Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ) with infinitesimal<br />
character H c<strong>on</strong>sisting of integers. Then there is an irreducible geometric<br />
str<strong>on</strong>gly compatible system of l-adic representati<strong>on</strong>s with Hodge-Tate weights H, such<br />
that Local-Global compatibility holds for all primes p.<br />
For every prime p, the c<strong>on</strong>jecture provides a semisimple p-adic Galois representati<strong>on</strong><br />
ρ π,p : Gal( ¯Q/Q) → GL n ( ¯Q p ). The (p, p)-Galois representati<strong>on</strong> ρ π,p | Gp , the restricti<strong>on</strong> of<br />
ρ π,p <strong>to</strong> G p , is potentially semistable with Hodge-Tate weights H. By work of Colmez-<br />
F<strong>on</strong>taine, we can associate a crystal D(ρ π,p | Gp ) <strong>to</strong> ρ π,p | Gp . The study of the irreducibility<br />
of the representati<strong>on</strong> ρ π,p | Gp reduces <strong>to</strong> studying D(ρ π,p | Gp ), because the existence of an<br />
admissible submodule of D(ρ π,p | Gp ) is equivalent <strong>to</strong> the reducibility of the representati<strong>on</strong><br />
ρ π,p | Gp .<br />
Let WD(ρ π,p | Gp ) denote the (p, p)-Weil-Deligne representati<strong>on</strong> associated <strong>to</strong> ρ π,p | Gp .<br />
By the definiti<strong>on</strong> of a strictly compatible system, we know WD(ρ π,p | Gp ) from the (p, l)-<br />
Weil-Deligne representati<strong>on</strong> for l ≠ p. Deligne classified all such Frobenius semisimple<br />
(p, l)-Weil-Deligne representati<strong>on</strong>s as the direct sum of the indecomposable representati<strong>on</strong>s<br />
τ m ⊗ Sp(n), where τ m is an irreducible representati<strong>on</strong> of dimensi<strong>on</strong> m of W p , the<br />
Weil group of Q p , and Sp(n) denotes the Steinberg representati<strong>on</strong> of dimensi<strong>on</strong> n.<br />
Principal series<br />
Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL m (A Q ) with infinitesimal character<br />
given by the integers −β 1 > · · · > −β m , i.e., H = {β 1 , . . . , β m }. Suppose π p , the <strong>local</strong><br />
comp<strong>on</strong>ent of π at p, is an unramified principal series representati<strong>on</strong> with Satake parameters<br />
α 1 , . . . , α m . In this case, WD(ρ π,p | Gp ) is a direct sum of m unramified characters.<br />
Theorem 1.1.1. If π is ordinary at p, i.e., β i + v p (α i ) = 0 for all i, then ρ π,p | Gp ∼<br />
⎛<br />
α<br />
λ( 1<br />
) · χ −β ⎞<br />
1<br />
p vp(α 1 ) cyc,p ∗ · · · ∗<br />
α<br />
0 λ( 2<br />
) · χ −β 2<br />
p<br />
⎜<br />
vp(α 2 ) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ ,<br />
α<br />
0 0 0 λ( m<br />
)) · χ −βm<br />
p vp(αm) cyc,p<br />
where λ(x) is the unramified character taking arithmetic Frobenius <strong>to</strong> x. In particular,<br />
the representati<strong>on</strong> ρ π,p | Gp is ordinary.<br />
Indecomposable case<br />
Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL mn (A Q ) with infinitesimal character<br />
given by the integers −β 1 > · · · > −β mn . Suppose that the Weil-Deligne representati<strong>on</strong><br />
<strong>attached</strong> <strong>to</strong> π p is Frobenius semisimple and indecomposable, i.e.,<br />
WD(ρ π,p | Gp ) ∼ τ m ⊗ Sp(n).<br />
12
Twist of Steinberg<br />
When m = 1, we prove:<br />
Theorem 1.1.2. If π is ordinary at p, i.e., β 1 + v p (α) = 0, then the β i are necessarily<br />
c<strong>on</strong>secutive integers, and ρ π,p | Gp ∼<br />
⎛<br />
α<br />
χ 0 · λ( ) · χ −β ⎞<br />
1<br />
p vp(α) cyc,p ∗ · · · ∗<br />
α<br />
0 χ 0 · λ( ) · χ −β 1−1<br />
p<br />
⎜<br />
vp(α) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ ,<br />
α<br />
0 0 0 χ 0 · λ( ) · χ −β 1−(n−1)<br />
p vp(α) cyc,p<br />
where the character τ 1 decomposes as χ 0 · χ ′ , where χ 0 is the ramified part, and χ ′ is an<br />
unramified character mapping arithmetic Frobenius <strong>to</strong> α.<br />
If π is not ordinary at p, then the <strong>local</strong> representati<strong>on</strong> ρ π,p | Gp is irreducible.<br />
The last statement generalizes a well-known fact for n<strong>on</strong>-ordinary (p, p)-Galois representati<strong>on</strong>s<br />
<strong>attached</strong> <strong>to</strong> normalized cuspidal eigen<strong>forms</strong> (n = 2) at Steinberg primes.<br />
Supercuspidal ⊗ Steinberg<br />
Now we assume that τ m is an irreducible representati<strong>on</strong> corresp<strong>on</strong>ding <strong>to</strong> a supercuspidal<br />
representati<strong>on</strong> of GL m for m ≥ 2. Let E be the field of coefficients of the representati<strong>on</strong><br />
ρ π,p | Gp and let F be a finite Galois extensi<strong>on</strong> of Q p such that τ m | IF = 1. There is an<br />
equivalence of categories between (ϕ, N, F, E)-modules and Weil-Deligne representati<strong>on</strong>s.<br />
We write D τm and D Sp(n) for the (ϕ, N, F, E)-modules corresp<strong>on</strong>ding <strong>to</strong> τ m and Sp(n),<br />
respectively. We prove:<br />
Theorem 1.1.3. All the (ϕ, N, F, E)-submodules of D = D τm ⊗ D Sp(n) are of the form<br />
D τm ⊗ D Sp(r) , for some 1 ≤ r ≤ n.<br />
When τ m ∼ Ind Wp<br />
W χ, where χ is a character of W p m pm, we also write down the filtered<br />
(ϕ, N, F, E)-module with underlying Weil-Deligne representati<strong>on</strong> τ m , generalizing results<br />
in [GM09] for m = 2.<br />
Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL mn (A Q ) with infinitesimal character<br />
c<strong>on</strong>sisting of distinct integers {β i,j } i=n,j=m<br />
i=1,j=1<br />
ordered as follows: β i1 ,j 1<br />
> β i2 ,j 2<br />
, if<br />
i 1 > i 2 , or if i 1 = i 2 and j 1 > j 2 . Let t N (D τm ) denote the Newt<strong>on</strong> number of the<br />
(ϕ, N, F, E)-module D τm . Using Theorem 1.1.3, we prove:<br />
Theorem 1.1.4. Suppose that<br />
WD(ρ π,p | Gp ) ∼ τ m ⊗ Sp(n).<br />
If π is ordinary at p, i.e., t N (D τm ) = ∑ m<br />
j=1 β 1,j, then ρ π,p | Gp is reducible, in which case<br />
m = 1, τ 1 is a character, and ρ π,p | Gp is quasi-ordinary as in Theorem 1.1.2.<br />
If π is not ordinary at p, then ρ π,p | Gp is irreducible. In particular, when m ≥ 2, the<br />
representati<strong>on</strong> ρ π,p | Gp is always irreducible.<br />
The theorems above gives complete informati<strong>on</strong> about the reducibility of the (p, p)-<br />
Galois representati<strong>on</strong> in the indecomposable case.<br />
13
Decomposable case<br />
Suppose π is a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL N (A Q ) with infinitesimal character<br />
given by the integers −β 1 > · · · > −β N . Suppose that N = ∑ r<br />
i=1 m in i and<br />
WD(ρ π,p | Gp ) ∼ ⊕ r i=1 τ m i<br />
⊗ Sp(n i ). In this case, we can also define p-ordinariness in terms<br />
of the t N (D τi )’s and β i ’s and show that there exists an admissible flag of submodules in<br />
the corresp<strong>on</strong>ding (ϕ, N, F, E)-module.<br />
Theorem 1.1.5. If π is ordinary at p, then m i = 1 for all i, the β i occur in r blocks of<br />
c<strong>on</strong>secutive integers, of lengths n i , for 1 ≤ i ≤ r, and<br />
⎛<br />
⎞<br />
ρ n1 ∗ · · · ∗<br />
0 ρ n2 · · · ∗<br />
ρ π,p | Gp ∼ ⎜<br />
⎟<br />
⎝ 0 0 · · · ∗ ⎠ ,<br />
0 0 0 ρ nr<br />
where each ρ ni is an n i -dimensi<strong>on</strong>al representati<strong>on</strong> with shape similar <strong>to</strong> that in Theorem<br />
1.1.2. In particular, ρ π,p | Gp is quasi-ordinary.<br />
In Theorem 1.1.5, we show that our ordinariness c<strong>on</strong>diti<strong>on</strong> implies that a particular<br />
complete flag is admissible. We have c<strong>on</strong>structed examples for which not all complete flags<br />
are necessarily admissible, even under our ordinariness assumpti<strong>on</strong>. In the decomposable<br />
case, we give an example of a reducible (p, p)-Galois representati<strong>on</strong> such that there is no<br />
complete flag of reducible submodules. Finally, in the n<strong>on</strong>-ordinary decomposable case,<br />
we have c<strong>on</strong>structed examples where the crystal is irreducible.<br />
1.2 2-adic Hida theory and applicati<strong>on</strong>s<br />
In the sec<strong>on</strong>d part of our thesis (cf. [GK2], [Kum]), we study the semisimplicity of the<br />
<strong>local</strong> Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> normalized ordinary cuspidal eigen<strong>forms</strong> (n = 2),<br />
for the prime p = 2.<br />
2-adic Hida theory<br />
We first prove a c<strong>on</strong>trol theorem for the Hida’s ordinary Λ-adic Hecke algebra for the<br />
prime p = 2 by following the approach of Hida [Hid86a] which deals with primes p ≥ 5.<br />
Let T 2 = Q 2 /Z 2 and Γ r = 1 + 2 r Z 2 of Γ 2 = 1 + 4Z 2 = 〈u〉. For r ≥ 2, we write<br />
V r = H 1 (X r , T 2 ) and W r = H 1 (Y r , T 2 ), where X r denotes the smooth compactificati<strong>on</strong><br />
of the complex manifold Y r = Γ 1 (N2 r )\H, with (2, N) = 1. Let Vr 0 and Wr 0 denote the<br />
ordinary parts of V r and W r , respectively, where the ordinary parts are defined by using<br />
Hida’s idempotent opera<strong>to</strong>r. There is a natural acti<strong>on</strong> of Γ 0 (N2 r )/Γ 1 (N2 r ) <strong>on</strong> Vr<br />
0 and<br />
Wr 0 . Let V 0 and W 0 denote the direct limit of Vr 0 and Wr 0 , respectively. Since (Z/N2 r Z) ×<br />
acts <strong>on</strong> Vr 0 and Wr 0 , Z × 2 acts <strong>on</strong> V0 and W 0 . Let V 0 and W 0 denote the P<strong>on</strong>tryagin duals<br />
of V 0 and W 0 , respectively. These are Λ-modules for Λ = Z 2 [[Γ 2 ]].<br />
A first step <strong>to</strong>wards proving a c<strong>on</strong>trol theorem for the Hida’s ordinary Λ-adic Hecke<br />
algebra is <strong>to</strong> prove a c<strong>on</strong>trol theorem for the cohomology modules above.<br />
14
Theorem 1.2.1. 1. For each positive integer r ≥ 2, the restricti<strong>on</strong> morphism of cohomology<br />
groups induces an isomorphism of Vr 0 <strong>on</strong><strong>to</strong> (V 0 ) Γr . The same result also<br />
holds for W 0 .<br />
2. Let N > 1. The modules V 0 and W 0 are free of finite rank over Λ.<br />
The proof of the freeness of V 0 depends heavily <strong>on</strong> the Z 2 -freeness of the module<br />
e(H 1 (Φ 2 , Z 2 (ω))/H 1 p(Φ 2 , Z 2 (ω))), where Φ 2 = Γ 1 (N)∩Γ 0 (4), and <strong>on</strong> the following theorem.<br />
Theorem 1.2.2. For each positive integer n ≡ a (mod 2),<br />
rank Z2 h 0 n+2(Φ 2 , Z 2 ) = r(a),<br />
where r(a) is the Z 2 -rank of the Hecke algebra h 0 2 (Φ 2, ω a , Z 2 ), where ω is the mod 4<br />
cyclo<strong>to</strong>mic character, i.e., ω(x) = ±1, if x ≡ ±1 (mod 4).<br />
For p ≥ 5, the theorem above follows from [Hid86b, Thm. 3.1 and Cor. 3.2]. But their<br />
proofs depend <strong>on</strong> the theory of Katz modular <strong>forms</strong>, the theory of mod p modular <strong>forms</strong>,<br />
and some results in these theories <strong>on</strong>ly hold for p ≥ 5. For p = 2, we give an alternative<br />
proof of this theorem, which completely avoids both these technicalities.<br />
Let K be a finite extensi<strong>on</strong> of Q 2 and O K be the integral closure of Z 2 in K. Write Λ<br />
again for O K [[Γ 2 ]]. Let ι denote the natural inclusi<strong>on</strong> of Γ 2 in Λ. For each weight k ≥ 2,<br />
let h 0 k (Γ 1(N2 ∞ ), O K ) denote the ordinary Λ-adic Hecke algebra lim h 0 ←− k (Γ 1(N2 r ), O K ). For<br />
r≥2<br />
k 1 ≥ k 2 ≥ 2, there exists a surjecti<strong>on</strong><br />
h 0 k 1<br />
(Γ 1 (N2 ∞ ), O K ) ↠ h 0 k 2<br />
(Γ 1 (N2 ∞ ), O K ),<br />
and this is an isomorphism by [Hid88b, Thm. 3.2]. Thus the ordinary Λ-adic Hecke<br />
algebra h 0 k (Γ 1(N2 ∞ ), O K ) is independent of the weight, and we denote it by h 0 (N, O K ).<br />
From Theorem 1.2.1 and 1.2.2, we deduce the following theorem, which is the 2-adic<br />
analogue of the c<strong>on</strong>trol theorem for Hida’s ordinary Λ-adic Hecke algebra h 0 (N, O K ).<br />
Theorem 1.2.3 (C<strong>on</strong>trol theorem). For r ≥ 2, the map<br />
ρ k,ɛ : h 0 (N, O K ) ⊗ Λ Λ Pk,ɛ /P k,ɛ Λ Pk,ɛ ↠ h 0 k (Φ2 r, ɛ, K),<br />
where Λ Pk,ɛ /P k,ɛ Λ Pk,ɛ is identified with K, via u corresp<strong>on</strong>ds <strong>to</strong> u k ɛ(u), is an isomorphism,<br />
where P k,ɛ = (ι(u) − ɛ(u)u k ) is an ideal of Λ and ɛ is a character of Γ/Γ r with values in<br />
O K .<br />
As a c<strong>on</strong>sequence of the c<strong>on</strong>trol theorem, we are able <strong>to</strong> deduce that in a primitive 2-<br />
ordinary Hida family, either all arithmetic specializati<strong>on</strong>s are CM <strong>forms</strong> or no arithmetic<br />
specializati<strong>on</strong> is a CM form. Hence, we can speak of CM and n<strong>on</strong>-CM primitive 2-ordinary<br />
Hida families.<br />
An applicati<strong>on</strong><br />
It is known that if f is ordinary at ℘ and f has complex multiplicati<strong>on</strong> (CM), then ρ f,℘ | Gp<br />
splits. Greenberg has asked whether the c<strong>on</strong>verse also holds. Serre has proved that the<br />
15
c<strong>on</strong>verse holds in the case of elliptic curves, i.e., for an elliptic curve E, the representati<strong>on</strong><br />
ρ E,℘ splits at p if and <strong>on</strong>ly if E has CM.<br />
In [GV04], Ghate and Vatsal have proved the c<strong>on</strong>verse for odd primes p and for primitive<br />
p-ordinary Λ-adic <strong>forms</strong> F, using a result of Buzzard [Buz03] for weight <strong>on</strong>e modular<br />
<strong>forms</strong> and under some c<strong>on</strong>diti<strong>on</strong>s <strong>on</strong> the residual representati<strong>on</strong> ¯ρ F . As a c<strong>on</strong>sequence,<br />
they are able <strong>to</strong> deduce that all arithmetic specializati<strong>on</strong>s of a primitive p-ordinary n<strong>on</strong>-<br />
CM Hida family have n<strong>on</strong>-split <strong>local</strong> Galois representati<strong>on</strong>, except for a possible finite<br />
set of excepti<strong>on</strong>s. By Hida’s c<strong>on</strong>trol theorem for odd primes p, it is known that this<br />
excepti<strong>on</strong>al set does not c<strong>on</strong>tain any CM <strong>forms</strong>.<br />
We prove a similar result for the case of p = 2, assuming a relevant result of Buzzard<br />
c<strong>on</strong>tinues <strong>to</strong> hold for p = 2 in the residually dihedral setting.<br />
Theorem 1.2.4. Let p = 2. Let F be a primitive p-ordinary Hida family of eigen<strong>forms</strong><br />
with the property that<br />
1. ¯ρ F is p-distinguished,<br />
2. ¯ρ F is absolutely irreducible, when restricted <strong>to</strong> Gal( ¯Q/Q(i)),<br />
3. ¯ρ F (c) ≠ 1 and ¯ρ F is both α-modular and β-modular.<br />
Then,<br />
(a) F is CM if and <strong>on</strong>ly if ρ F | Gp<br />
splits.<br />
(b) If F is n<strong>on</strong>-CM, then for all but except possibly finitely many arithmetic specializati<strong>on</strong>s<br />
f of F, the representati<strong>on</strong> ρ f,℘ | Gp is n<strong>on</strong>-split. Moreover, the possible excepti<strong>on</strong>s<br />
are necessarily n<strong>on</strong>-CM <strong>forms</strong> by Theorem 1.2.3.<br />
16
Chapter 2<br />
Introducti<strong>on</strong><br />
In this thesis, we study the p-adic Galois representati<strong>on</strong>s of G p := Gal( ¯Q p /Q p ) <strong>attached</strong><br />
<strong>to</strong> au<strong>to</strong>morphic representati<strong>on</strong> of GL n , using p-adic Hodge theory and Hida theory.<br />
Let f = ∑ ∞<br />
n=1 a n(f)q n be a normalized eigenform (i.e., a primitive form) of weight<br />
k ≥ 2, level N ≥ 1, and nebentypus χ : (Z/NZ) × → C × . Let ℘ be the prime of ¯Q<br />
determined by a fixed embedding of ¯Q in<strong>to</strong> ¯Q p . Let ℘ also denote the induced prime of<br />
K f = Q(a n (f)), the Hecke field of f, and let K f,℘ denote the completi<strong>on</strong> of K f at ℘.<br />
There is a global Galois representati<strong>on</strong><br />
ρ f,℘ : G Q → GL 2 (K f,℘ )<br />
associated <strong>to</strong> f (and ℘) by Deligne which has the property that for all primes l ∤ Np,<br />
trace(ρ f,℘ (Frob l )) = a l (f) and det(ρ f,℘ (Frob l )) = χ(l)l k−1 ,<br />
where Frob l denotes the arithmetic Frobenius. It is a well-known result of Ribet that<br />
the global representati<strong>on</strong> ρ f,℘ is irreducible. However, if f is ordinary at ℘, i.e., a p (f)<br />
is a ℘-adic unit, then an important theorem of Wiles says that the corresp<strong>on</strong>ding <strong>local</strong><br />
representati<strong>on</strong> is reducible. Let χ cyc,p denote the p-adic cyclo<strong>to</strong>mic character.<br />
Theorem 2.0.5 ([Wil88]). Let f be a ℘-ordinary primitive form as above. Then the<br />
restricti<strong>on</strong> of ρ f,℘ <strong>to</strong> the decompositi<strong>on</strong> subgroup G p is reducible. More precisely, there<br />
exists a basis in which<br />
(<br />
)<br />
χ p · λ(β/p k−1 ) · χ k−1<br />
cyc,p u<br />
ρ f,℘ | Gp ∼<br />
,<br />
0 λ(α)<br />
where χ = χ p χ ′ is the decompositi<strong>on</strong> of χ in<strong>to</strong> its p and prime-<strong>to</strong>-p-parts, λ(x) : G p →<br />
K × f,℘ is the unramified character which takes arithmetic Frobenius <strong>to</strong> x, and u : G p → K f,℘<br />
is a c<strong>on</strong>tinuous functi<strong>on</strong>. Here α is (i) the unit root of X 2 − a p (f)X + p k−1 χ(p) if p ∤ N,<br />
(ii) the unit a p (f) if p||N, p ∤ c<strong>on</strong>d(χ) and k = 2, and (iii) the unit a p (f) if p|N,<br />
v p (N) = v p (c<strong>on</strong>d(χ)). In all cases αβ = χ ′ (p)p k−1 .<br />
Moreover, in case (ii), a p (f) is a unit if and <strong>on</strong>ly if k = 2, and <strong>on</strong>e can easily show that<br />
ρ f,℘ | Gp is irreducible when k > 2.<br />
Urban has generalized Theorem 2.0.5 <strong>to</strong> the case of primitive Siegel modular cusprepresentati<strong>on</strong>s<br />
of genus 2. For a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> π <strong>on</strong> GSp 4 (A Q )<br />
17
whose Archimedean comp<strong>on</strong>ent π ∞ bel<strong>on</strong>gs <strong>to</strong> the discrete series with cohomological<br />
weights (a, b; a + b) with a ≥ b ≥ 0, Laum<strong>on</strong>, Taylor, and Weissauer have defined a<br />
four-dimensi<strong>on</strong>al Galois representati<strong>on</strong><br />
ρ π,p : G Q → GL 4 ( ¯Q p )<br />
with standard properties. For an unramified prime p for π, Tilouine and Urban have<br />
generalized the noti<strong>on</strong> of ordinariness for such primes p in three ways. For example, in<br />
the Borel case, the p-ordinariness of π implies that the valuati<strong>on</strong>s of the roots of the Hecke<br />
polynomial of π p are 0, b + 1, a + 2 and a + b + 3. In this case, the restricti<strong>on</strong> of ρ π,p<br />
<strong>to</strong> the decompositi<strong>on</strong> subgroup G p is upper-triangular. More precisely, there is a basis in<br />
which the <strong>local</strong> representati<strong>on</strong> ρ π,p | Gp<br />
⎛<br />
⎞<br />
λ(δ/p a+b+3 ) · χ a+b+3<br />
cyc,p ∗ ∗ ∗<br />
0 λ(γ/p a+2 ) · χ a+2<br />
cyc,p ∗ ∗<br />
∼ ⎜<br />
⎝ 0 0 λ(β/p b+1 ) · χ b+1 ⎟<br />
cyc,p ∗ ⎠ ,<br />
0 0 0 λ(α)<br />
where α, β, γ and δ are the roots of the Hecke polynomial of π p with increasing valuati<strong>on</strong>s.<br />
In the first part of the thesis (cf. Chapters 4 and 5), motivated by the above two<br />
examples, we define a noti<strong>on</strong> of p-ordinariness for cuspidal au<strong>to</strong>morphic representati<strong>on</strong>s<br />
π <strong>on</strong> GL n (A Q ), for n ≥ 3, and we study the reducibility or the irreducibility of the <strong>local</strong><br />
Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> π [GK1]. We show that p-ordinariness for π implies that<br />
the associated <strong>local</strong> Galois representati<strong>on</strong> at p is quasi-ordinary, in the sense of Greenberg<br />
(cf. §4.4.6). In this case we explicitly write down the characters <strong>on</strong> the diag<strong>on</strong>al. In<br />
certain cases, we also show that the n<strong>on</strong>-ordinariness of π implies the irreducibility of the<br />
p-adic <strong>local</strong> Galois representati<strong>on</strong>. These results can be thought of as a generalizati<strong>on</strong> of<br />
Wiles’ theorem for GL 2 (cf. Theorem 2.0.5), when we interpret the primitive form f as a<br />
cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL 2 (A Q ).<br />
In Chapter 4, after stating the necessary results from p-adic Hodge theory, we reprove<br />
Theorem 2.0.5. The proof serves <strong>to</strong> illustrate the techniques that will be used in Chapter<br />
5. Let f be a primitive form which is ℘-ordinary. A key ingredient in our proof is that the<br />
representati<strong>on</strong> ρ f,℘ lives in a strictly compatible system of Galois representati<strong>on</strong>s (ρ f,λ ),<br />
where λ varies over the primes of K f . By F<strong>on</strong>taine theory, <strong>on</strong>e can associate a crystal<br />
<strong>to</strong> the potentially semistable representati<strong>on</strong> ρ f,℘ | Gp . Using the explicit descripti<strong>on</strong> of the<br />
crystal (cf. [Bre01], [GM09]) in the Unramified, Steinberg, and ramified principal series<br />
cases, we study the reducibility or the irreducibility of the representati<strong>on</strong> ρ f,℘ | Gp .<br />
In Chapter 5, we prove structure theorems for the <strong>local</strong> Galois representati<strong>on</strong>s <strong>attached</strong><br />
<strong>to</strong> cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ).<br />
Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ) with infinitesimal character<br />
H, where H c<strong>on</strong>sists of a set of distinct integers. We assume that the global p-adic<br />
Galois representati<strong>on</strong> ρ π,p <strong>attached</strong> <strong>to</strong> π exists, and that it satisfies several natural properties,<br />
e.g., it lives in a strictly compatible system of Galois representati<strong>on</strong>s, and satisfies<br />
Local-Global compatibility.<br />
By a result of Flath, π is a restricted tensor product π = ⊗ ′ pπ p (cf. [Bum97, Thm.<br />
3.3.3]) of <strong>local</strong> au<strong>to</strong>morphic representati<strong>on</strong>s. Langlands classified that all irreducible admissible<br />
representati<strong>on</strong>s of GL n (Q p ) are isomorphic <strong>to</strong> Q(∆ 1 , . . . , ∆ r ) for some segments<br />
18
{∆ i } r i=1 , such that for i < j, ∆ i does not precede ∆ j (cf. Theorem 4.4.1). There exists<br />
a bijecti<strong>on</strong> between isomorphism classes of irreducible admissible representati<strong>on</strong>s of<br />
GL n (Q p ) and isomorphism classes of admissible n-dimensi<strong>on</strong>al representati<strong>on</strong>s, over ¯Q l<br />
for l ≠ p, of W p, ′ the Weil-Deligne group of Q p (cf. the discussi<strong>on</strong> after Theorem 4.4.2).<br />
Deligne classified all such admissible n-dimensi<strong>on</strong>al representati<strong>on</strong>s of W p, ′ over an<br />
algebraically closed fields of characterstic 0, as the direct sum of the indecomposable<br />
representati<strong>on</strong>s τ m ⊗ Sp(n), where τ m is an irreducible representati<strong>on</strong> of dimensi<strong>on</strong> m of<br />
W p , the Weil group of Q p , and Sp(r) denotes the Steinberg representati<strong>on</strong> of dimensi<strong>on</strong><br />
r.<br />
Let WD(ρ π,p | Gp ) denote the (p, p)-Weil-Deligne representati<strong>on</strong> associated <strong>to</strong> ρ π,p | Gp .<br />
By the definiti<strong>on</strong> of a strictly compatible system, we know WD(ρ π,p | Gp ) from the (p, l)-<br />
Weil-Deligne representati<strong>on</strong> for l ≠ p, and which is determined by the <strong>local</strong> au<strong>to</strong>morphic<br />
representati<strong>on</strong> of π at p.<br />
The noti<strong>on</strong> of p-ordinariness is defined by using p-adic valuati<strong>on</strong> v p of certain parameters<br />
coming from the <strong>local</strong> au<strong>to</strong>morphic representati<strong>on</strong> π at p. The parameters vary in<br />
different cases, but can be made completely precise.<br />
For instance, suppose that the <strong>local</strong> au<strong>to</strong>morphic representati<strong>on</strong> π p is an unramified<br />
representati<strong>on</strong> of GL m (Q p ) (in the classical setting, we are in this case if p ∤ N), i.e.,<br />
π p = Q(χ 1 , . . . , χ m ), for some unramified characters χ i of Q × p . We can parametrize<br />
the isomorphism class of this representati<strong>on</strong> by the Satake parameters α 1 , . . . , α m , for<br />
α i = χ i (ω), where ω is a uniformizer for Q p . In this case, WD(ρ π,p | Gp ) is a direct sum<br />
of unramified characters χ i ’s. The <strong>local</strong> representati<strong>on</strong> ρ π,p | Gp is crystalline with Hodge-<br />
Tate weights H. Let D be the corresp<strong>on</strong>ding filtered ϕ-module. Let the jumps in the<br />
filtrati<strong>on</strong> <strong>on</strong> D be β 1 < · · · < β m (so that H = {−β 1 , . . . , −β m }). In this case, we say<br />
that the cuspidal au<strong>to</strong>morphic representati<strong>on</strong> π is p-ordinary if<br />
for all i = 1, . . . , m.<br />
β i + v p (α i ) = 0,<br />
Theorem 2.0.6 (Spherical case). Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of<br />
GL m (A Q ) with infinitesimal character given by the integers −β 1 > · · · > −β m and such<br />
that π p is in the unramified principal series with Satake parameters α 1 , . . . , α m . If π is<br />
p-ordinary, then ρ π,p | Gp ∼<br />
⎛<br />
α<br />
λ( 1<br />
) · χ −β ⎞<br />
1<br />
p vp(α 1 ) cyc,p ∗ · · · ∗<br />
α<br />
0 λ( 2<br />
) · χ −β 2<br />
p<br />
⎜<br />
vp(α 2 ) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ .<br />
α<br />
0 0 0 λ( n<br />
) · χ −βm<br />
p vp(αm) cyc,p<br />
In particular, ρ π,p | Gp<br />
is ordinary.<br />
A similar result was also obtained by D. Geraghty in the course of proving modularity<br />
lifting theorems for GL n . However, the main point of our thesis is <strong>to</strong> treat other cases (in<br />
the classical language, the primes of bad reducti<strong>on</strong>).<br />
In the n<strong>on</strong>-principal series case, we no l<strong>on</strong>ger have the Satake parameters of π p at our<br />
disposal. However, we can replace these numbers by the corresp<strong>on</strong>ding eigenvalues of l-<br />
adic Frobenius in the l-adic Weil-Deligne representati<strong>on</strong> corresp<strong>on</strong>ding <strong>to</strong> π p , for l ≠ p, or<br />
19
equivalently, by using the properties of strictly compatible systems, with the eigenvalues<br />
of crystalline Frobenius <strong>on</strong> the filtered module <strong>attached</strong> <strong>to</strong> π p (as in [GM09] for n = 2).<br />
To keep the analysis of the structure of the <strong>local</strong> Galois representati<strong>on</strong> ρ π,p | Gp within<br />
reas<strong>on</strong>able limits in this thesis, we shall assume that the Newt<strong>on</strong> filtrati<strong>on</strong> is in general<br />
positi<strong>on</strong> with respect <strong>to</strong> the Hodge filtrati<strong>on</strong> <strong>on</strong> the associated crystal.<br />
Now suppose π p is a twist of the Steinberg representati<strong>on</strong>, i.e., π p = Q(∆) for a<br />
segment ∆ = [σ, σ(n − 1)], where σ is a supercuspidal representati<strong>on</strong> of GL m (Q p ) (in the<br />
classical setting, we are in this case if v p (N) = 1 and v p (c<strong>on</strong>d(χ)) = 0). We know that<br />
the corresp<strong>on</strong>ding Frobenius semisimplificati<strong>on</strong> of the Weil-Deligne representati<strong>on</strong> is of<br />
the form τ m ⊗ Sp(n). In this case, we not <strong>on</strong>ly show that the p-ordinariness of π implies<br />
quasi-ordinariness of the representati<strong>on</strong> ρ π,p | Gp but also show that the n<strong>on</strong>-ordinariness<br />
of π implies the representati<strong>on</strong> ρ π,p | Gp is irreducible, unlike in the principal series case.<br />
When m = 1, we prove:<br />
Theorem 2.0.7 (Twist of Steinberg). Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of<br />
GL n (A Q ) with infinitesimal character given by the integers −β 1 > · · · > −β n . Suppose<br />
π p is a twist of the Steinberg representati<strong>on</strong>, and WD(ρ π,p | Gp ) is Frobenius semisimple,<br />
hence<br />
WD(ρ π,p | Gp ) ∼ τ 1 ⊗ Sp(n).<br />
If π is p-ordinary, i.e., β 1 + v p (α) = 0, then the β i are necessarily c<strong>on</strong>secutive integers,<br />
and ρ π,p | Gp ∼<br />
⎛<br />
α<br />
χ 0 · λ( ) · χ −β ⎞<br />
1<br />
p vp(α) cyc,p ∗ · · · ∗<br />
α<br />
0 χ 0 · λ( ) · χ −β 1−1<br />
p<br />
⎜<br />
vp(α) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ ,<br />
α<br />
0 0 0 χ 0 · λ( ) · χ −β 1−(n−1)<br />
p vp(α) cyc,p<br />
where the character τ 1 decomposes as χ 0 · χ ′ , where χ 0 is the ramified part, and χ ′ is an<br />
unramified character mapping arithmetic Frobenius <strong>to</strong> α.<br />
If π is not ordinary at p, then the <strong>local</strong> representati<strong>on</strong> ρ π,p | Gp is irreducible.<br />
The last statement generalizes a well-known fact for n<strong>on</strong>-ordinary (p, p)-Galois representati<strong>on</strong>s<br />
<strong>attached</strong> <strong>to</strong> normalized cuspidal eigen<strong>forms</strong> (n = 2) at Steinberg primes.<br />
We now turn <strong>to</strong> the case where m ≥ 2. Recall that π p = Q(∆) for some segment<br />
∆ = [σ, σ(n − 1)], where σ is supercuspidal representati<strong>on</strong> of GL m (Q p ). Suppose the<br />
infinitesimal character of π c<strong>on</strong>sists of distinct integers {β i,j } i=n,j=m<br />
i=1,j=1<br />
ordered as follows:<br />
β i1 ,j 1<br />
> β i2 ,j 2<br />
, if i 1 > i 2 , or if i 1 = i 2 and j 1 > j 2 . Before defining the p-ordinariness in<br />
this case, we slightly deviate and study the structure of the associated crystal.<br />
Let E be the field of coefficients of the representati<strong>on</strong> ρ π,p | Gp and let F be a finite<br />
Galois extensi<strong>on</strong> of Q p such that τ m | IF = 1. By [BS07, Prop. 4.1], there is an equivalence<br />
of categories between (ϕ, N, F, E)-modules and Weil-Deligne representati<strong>on</strong>s. We write<br />
D τm and D Sp(n) for the (ϕ, N, F, E)-modules corresp<strong>on</strong>ding <strong>to</strong> τ m and Sp(n), respectively.<br />
We prove:<br />
Theorem 2.0.8. All the (ϕ, N, F, E)-submodules of D = D τm ⊗ D Sp(n) are of the form<br />
D τm ⊗ D Sp(r) , for some 1 ≤ r ≤ n.<br />
20
When τ m ∼ Ind Wp<br />
W χ, where χ is a character of W p m pm, we also write down the filtered<br />
(ϕ, N, F, E)-module with underlying Weil-Deligne representati<strong>on</strong> τ m , generalizing results<br />
in [GM09] for m = 2.<br />
Let t N (D τm ) denote the Newt<strong>on</strong> number of the (ϕ, N, F, E)-module D τm . Using Theorem<br />
2.0.8, we prove:<br />
Theorem 2.0.9 (Indecomposable case). Suppose that<br />
WD(ρ π,p | Gp ) ∼ τ m ⊗ Sp(n).<br />
If π is ordinary at p, i.e., t N (D τm ) = ∑ m<br />
j=1 β 1,j, then ρ π,p | Gp is reducible, in which case<br />
m = 1, τ 1 is a character, and ρ π,p | Gp is quasi-ordinary as in Theorem 2.0.7.<br />
If π is not ordinary at p, then ρ π,p | Gp is irreducible. In particular, when m ≥ 2, the<br />
representati<strong>on</strong> ρ π,p | Gp is always irreducible.<br />
The theorems above give complete informati<strong>on</strong> about the reducibility of the (p, p)-<br />
Galois representati<strong>on</strong> in the indecomposable case.<br />
Now suppose π is a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL N (A Q ) with infinitesimal<br />
character given by the integers −β 1 > · · · > −β N , and π p = Q(∆ 1 , . . . , ∆ r ) for some<br />
segments ∆ i = [σ i , σ i (n i −1)] (up<strong>to</strong> permutati<strong>on</strong>) such that for i < j, ∆ i does not precede<br />
∆ j , where σ i is a supercuspidal representati<strong>on</strong> of GL mi (Q p ). Clearly, N = ∑ r<br />
i=1 m in i and<br />
WD(ρ π,p | Gp ) ∼ ⊕ r i=1 τ m i<br />
⊗ Sp(n i ). In this case, we can also define p-ordinariness in terms<br />
of the t N (D τi )’s and β i ’s and show that there exists an admissible flag of submodules in<br />
the corresp<strong>on</strong>ding (ϕ, N, F, E)-module.<br />
Theorem 2.0.10. If π is ordinary at p, then m i = 1 for all i, the β i occur in r blocks of<br />
c<strong>on</strong>secutive integers, of lengths n i , for 1 ≤ i ≤ r, and<br />
⎛<br />
⎞<br />
ρ n1 ∗ · · · ∗<br />
0 ρ n2 · · · ∗<br />
ρ π,p | Gp ∼ ⎜<br />
⎟<br />
⎝ 0 0 · · · ∗ ⎠ ,<br />
0 0 0 ρ nr<br />
where each ρ ni is an n i -dimensi<strong>on</strong>al representati<strong>on</strong> with shape similar <strong>to</strong> that in Theorem<br />
2.0.7. In particular, ρ π,p | Gp is quasi-ordinary.<br />
In Theorem 2.0.10 we show that our ordinariness c<strong>on</strong>diti<strong>on</strong> implies that a particular<br />
complete flag is admissible. In §5.6, we have c<strong>on</strong>structed examples for which not all<br />
complete flags are necessarily admissible, even under our ordinariness assumpti<strong>on</strong>. In<br />
the decomposable case, we give an example of a reducible (p, p)-Galois representati<strong>on</strong><br />
such that there is no complete flag of reducible submodules. Finally, in the n<strong>on</strong>-ordinary<br />
decomposable case, we have c<strong>on</strong>structed examples where the crystal is irreducible.<br />
This gives an overview of all the results in the first part of the thesis.<br />
In the sec<strong>on</strong>d part of the thesis (cf. Chapter 6 and Appendix A), we show that<br />
some of the basic results of Hida theory in the literature stated for odd primes p (and<br />
originally for p ≥ 5 in [Hid86b], [Hid86a]) remain valid for the prime p = 2. We prove<br />
a 2-adic analogue of the c<strong>on</strong>trol theorem for the Hida’s Hecke algebra [GK2]. As an<br />
applicati<strong>on</strong>, we study the semisimplicity of the <strong>local</strong> Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong><br />
ordinary primitive <strong>forms</strong> (n = 2), for the prime p = 2.<br />
21
In Chapter 6, we start by proving a c<strong>on</strong>trol theorem for cohomology groups. To explain<br />
briefly, for every positive integer r ≥ 2, we write V r = H 1 (X r , T 2 ), W r = H 1 (Y r , T 2 ), where<br />
T 2 = Q 2 /Z 2 . The Hecke algebra h 2 (Γ 1 (N2 r ), Z 2 ) acts <strong>on</strong> V r and therefore V 0 r , the ordinary<br />
part of V r , is a module over h 0 2 (Γ 1(N2 r ), Z 2 ), the ordinary part of h 2 (Γ 1 (N2 r ), Z 2 ). There<br />
is also an acti<strong>on</strong> of Γ 0 (N2 r )/Γ 1 (N2 r ) <strong>on</strong> V 0 r and W 0 r . Let V 0 denote the direct limit of V 0 r ,<br />
for r ≥ 2, and similarly for W 0 . One sees that V 0 and W 0 become c<strong>on</strong>tinuous modules<br />
(if we equip them with the discrete <strong>to</strong>pology) over the Iwasawa algebra Λ = Z 2 [[Γ 2 ]],<br />
where Γ r = 1 + 2 r Z 2 , and Γ = Γ 2 = 〈u〉. Let V 0 and W 0 denote the P<strong>on</strong>tryagin duals<br />
of V 0 and W 0 , respectively. The modules V 0 and W 0 are modules over Λ. We prove (cf.<br />
Theorem 6.3.1)<br />
Theorem 2.0.11. Let p = 2. We have:<br />
1. For each positive integer r ≥ 2, the restricti<strong>on</strong> morphism of cohomology groups<br />
induces an isomorphism of V 0 r <strong>on</strong><strong>to</strong> (V 0 ) Γr . The same result also holds for W 0 .<br />
2. Let N > 1. The modules V 0 and W 0 are free modules of finite rank over Λ.<br />
A proof of the freeness of W 0 <strong>forms</strong> the c<strong>on</strong>tent of Appendix A. For the other statements,<br />
especially the freeness of V 0 , by closely following the proof for odd primes, <strong>on</strong>e<br />
realizes that <strong>on</strong>e needs certain results from the theory of Katz modular <strong>forms</strong> and the<br />
theory of mod p modular <strong>forms</strong>. Since these results are not known for p = 2, we prove<br />
the theorem above with other ingredients (cf. §6.6).<br />
Let K be a finite extensi<strong>on</strong> of Q 2 and let O K denote the integral closure of Z 2 in<br />
K. By [Hid88b, Thm. 3.2], we know that the ordinary Λ-adic Hecke algebra is defined<br />
independently of the weight and we denote this Hecke algebra by h 0 (N, O K ). In §6.8, we<br />
prove a 2-adic versi<strong>on</strong> of the Hida’s c<strong>on</strong>trol theorem for h 0 (N, O K ).<br />
Theorem 2.0.12 (C<strong>on</strong>trol theorem). Let ɛ be a character of Γ/Γ r with values in O K ,<br />
and let P k,ɛ denote the ideal (ι(u) − ɛ(u)u k ) of Λ, where ι is an inclusi<strong>on</strong> of Γ in Λ. For<br />
r ≥ 2, the map<br />
ρ k,ɛ : h 0 (N, O K ) ⊗ Λ Λ Pk,ɛ /P k,ɛ Λ Pk,ɛ ↠ h 0 k (Φ2 r, ɛ, K),<br />
where Λ Pk,ɛ /P k,ɛ Λ Pk,ɛ is identified with K, via u corresp<strong>on</strong>ds <strong>to</strong> u k ɛ(u), is an isomorphism.<br />
As a c<strong>on</strong>sequence, we deduce a uniqueness theorem (cf. Theorem 6.9.10) for 2-<br />
stabilized ordinary primitive <strong>forms</strong> in Hida families. Namely, we show that such <strong>forms</strong><br />
live in unique primitive 2-ordinary cuspidal families, up <strong>to</strong> Galois c<strong>on</strong>jugacy, a well-known<br />
result when p is odd. As an applicati<strong>on</strong> of this result we show that the noti<strong>on</strong> of CM-ness<br />
is pure with respect <strong>to</strong> 2-adic families, i.e., we show that for a primitive 2-ordinary Λ-adic<br />
cuspidal eigenform F, either all arithmetic specializati<strong>on</strong>s are CM <strong>forms</strong> or no arithmetic<br />
specializati<strong>on</strong> is a CM form (cf. Propositi<strong>on</strong> 6.9.11). In view of this result, we may and<br />
do speak of CM and n<strong>on</strong>-CM 2-adic Hida families.<br />
Recall that f is a primitive ℘-ordinary form of weight at least two. Furthermore, if<br />
f has complex multiplicati<strong>on</strong> (CM), then ρ f,℘ | Gp splits at p. R. Greenberg has asked<br />
whether the c<strong>on</strong>verse also holds.<br />
In the case of mod p cusp <strong>forms</strong>, there is an (almost) complete answer <strong>to</strong> the above<br />
questi<strong>on</strong> due <strong>to</strong> Gross [Gro90] and Coleman-Voloch. Let g = ∑ a n q n be a p-ordinary mod<br />
22
p cuspidal eigenform of Serre weight 2 ≤ k ≤ p, level N with (p, N) = 1, and nebentypus ψ.<br />
Deligne c<strong>on</strong>structed a Galois representati<strong>on</strong> ρ g associated <strong>to</strong> g with standard properties.<br />
It is a well-known result of Gross and Coleman-Voloch, that ρ g is tamely ramified at<br />
p if and <strong>on</strong>ly if g has a compani<strong>on</strong> form, when k ≠ 2 or k ≠ p. If the form g is p-<br />
distinguished, then ρ g is tamely ramified if and <strong>on</strong>ly if ρ g splits at p (cf. [Gha05]). Hence,<br />
if g is p-distinguished, then ρ g splits at p if and <strong>on</strong>ly if g has a compani<strong>on</strong> form.<br />
If p is prime <strong>to</strong> N, then it is a theorem of Serre and Tate that if a modular form f<br />
corresp<strong>on</strong>ds <strong>to</strong> an elliptic curve E defined over Q, then ρ f,℘ = ρ E,p splits at p if and <strong>on</strong>ly<br />
if E, and therefore f, has CM.<br />
In [GV04], Ghate and Vatsal have proved the c<strong>on</strong>verse for odd primes p and for primitive<br />
p-ordinary Λ-adic <strong>forms</strong> F, using a result of Buzzard [Buz03] for weight <strong>on</strong>e modular<br />
<strong>forms</strong> and under some c<strong>on</strong>diti<strong>on</strong>s <strong>on</strong> the residual representati<strong>on</strong> ¯ρ F . As a c<strong>on</strong>sequence,<br />
they are able <strong>to</strong> deduce that all arithmetic specializati<strong>on</strong>s of a primitive p-ordinary n<strong>on</strong>-<br />
CM Hida family have n<strong>on</strong>-split <strong>local</strong> Galois representati<strong>on</strong>, except for a possible finite<br />
set of excepti<strong>on</strong>s. By Hida’s c<strong>on</strong>trol theorem for odd primes p, it is known that this<br />
excepti<strong>on</strong>al set does not c<strong>on</strong>tain any CM <strong>forms</strong>.<br />
Finally, we study the semisimplicity of the <strong>local</strong> Galois representati<strong>on</strong>s and prove a<br />
similar result for the case of p = 2, assuming a result of Buzzard (cf. Theorem 6.10.2)<br />
c<strong>on</strong>tinues <strong>to</strong> hold for p = 2 in the residually dihedral setting. We prove (cf. Theorem 6.10.4<br />
and Theorem 6.10.5):<br />
Theorem 2.0.13. Let p = 2. Let F be a primitive p-ordinary Hida family of eigen<strong>forms</strong><br />
with the property that<br />
1. ¯ρ F is p-distinguished,<br />
2. ¯ρ F is absolutely irreducible, when restricted <strong>to</strong> Gal( ¯Q/Q(i)),<br />
3. ¯ρ F (c) ≠ 1 and ¯ρ F is both α-modular and β-modular.<br />
Then,<br />
(a) F is CM if and <strong>on</strong>ly if ρ F | Gp<br />
splits.<br />
(b) If F is n<strong>on</strong>-CM, then for all but except possibly finitely many arithmetic specializati<strong>on</strong>s<br />
f of F, the representati<strong>on</strong> ρ f,℘ | Gp is n<strong>on</strong>-split. Moreover, the possible excepti<strong>on</strong>s<br />
are necessarily n<strong>on</strong>-CM <strong>forms</strong> by Theorem 2.0.12.<br />
23
Chapter 3<br />
Modular <strong>forms</strong> and Hecke algebras<br />
In this chapter, first we recall the definiti<strong>on</strong>s of modular <strong>forms</strong> and cusp <strong>forms</strong> of weight<br />
k for c<strong>on</strong>gruence subgroups. Then we define the abstract Hecke opera<strong>to</strong>rs for arbitrary<br />
c<strong>on</strong>gruence subgroups. Finally, we recall the c<strong>on</strong>cept of old<strong>forms</strong>, new<strong>forms</strong> and the<br />
noti<strong>on</strong> of complex multiplicati<strong>on</strong> (CM). We closely follow the expositi<strong>on</strong> in [DS05].<br />
3.1 Modular <strong>forms</strong><br />
Each element of the modular group SL 2 (Z) can be viewed as an au<strong>to</strong>morphism of the<br />
Riemann sphere Ĉ = C ∪ {∞} by the fracti<strong>on</strong>al linear transformati<strong>on</strong>:<br />
( )<br />
γ(τ) = aτ + b<br />
cτ + d , for γ = a b<br />
and τ ∈<br />
c d<br />
Ĉ.<br />
For every N ∈ N, we define<br />
{(<br />
a<br />
Γ 0 (N) :=<br />
c<br />
{(<br />
a<br />
Γ 1 (N) :=<br />
c<br />
{(<br />
a<br />
Γ(N) :=<br />
c<br />
)<br />
}<br />
b<br />
∈ SL 2 (Z) | c ≡ 0 (mod N) ,<br />
d<br />
)<br />
}<br />
b<br />
∈ Γ 0 (N) | a ≡ 1 (mod N) ,<br />
d<br />
)<br />
}<br />
b<br />
∈ Γ 1 (N) | b ≡ 0 (mod N) .<br />
d<br />
Definiti<strong>on</strong> 3.1.1. A subgroup Γ of SL 2 (Z) is a c<strong>on</strong>gruence subgroup, if Γ(N) ⊆ Γ for<br />
some N ∈ N, in which case Γ is a c<strong>on</strong>gruence subgroup of level N.<br />
Let H denote the upper half plane, and let GL + 2 (R) denote the set of matrices in<br />
GL 2 (R) with positive determinant.<br />
For γ = ( )<br />
a b<br />
c d ∈ GL<br />
+<br />
2 (R) and for any integer k, define the weight-k opera<strong>to</strong>r [γ] k <strong>on</strong><br />
functi<strong>on</strong>s f : H → C by<br />
(f[γ] k )(τ) = (det(τ)) k−1 (cτ + d) −k f(γ(τ)), for τ ∈ H.<br />
Since the term cτ + d is never zero or infinity, if f is meromorphic then f[γ] k is also<br />
meromorphic and has the same zeros and poles as f.<br />
25
Definiti<strong>on</strong> 3.1.2. Let Γ be a c<strong>on</strong>gruence subgroup of SL 2 (Z) and let k be an integer. A<br />
meromorphic functi<strong>on</strong> f : H → C is a weakly modular of weight k with respect <strong>to</strong> Γ, or<br />
simply weakly modular with respect <strong>to</strong> Γ, if for all γ ∈ Γ,<br />
f[γ] k = f.<br />
To define modular <strong>forms</strong> for c<strong>on</strong>gruence subgroups Γ, we need <strong>to</strong> make sense of the<br />
“holomorphy”c<strong>on</strong>diti<strong>on</strong> at ∞.<br />
Let D = {q ∈ C : |q| < 1} be the open complex unit disk, let D ′ = D − {0}. Each<br />
c<strong>on</strong>gruence subgroup Γ of SL 2 (Z) c<strong>on</strong>tains a translati<strong>on</strong> matrix of the form ( )<br />
1 h<br />
0 1 : τ ↦→<br />
τ + h for some minimal h ∈ N. Every functi<strong>on</strong> f : H → C that is weakly modular<br />
with respect <strong>to</strong> Γ is hZ-periodic and thus has a corresp<strong>on</strong>ding functi<strong>on</strong> g : D ′ → C, i.e.,<br />
f(τ) = g(exp( 2πiτ<br />
h<br />
)). If f is also holomorphic <strong>on</strong> H, then g is holomorphic <strong>on</strong> D′ and so<br />
it has a Laurent series expansi<strong>on</strong> at q = 0. Define such f <strong>to</strong> be holomorphic at ∞, if g<br />
extends holomorphically <strong>to</strong> D. Thus f has a Fourier expansi<strong>on</strong><br />
f(τ) =<br />
∞∑<br />
a n (f) exp(2πiτn/h).<br />
n=0<br />
A modular form with respect <strong>to</strong> a c<strong>on</strong>gruence subgroup Γ should be holomorphic not<br />
<strong>on</strong>ly at ∞ but also at some other points, namely the cusps.<br />
The group SL 2 (Z) acts <strong>on</strong> Q ∪ {∞}. We define the cusps of Γ <strong>to</strong> be the Γ-equivalence<br />
classes of Q ∪ {∞}. Since Γ is of finite index in SL 2 (Z), we see that the number of cusps<br />
is finite. Writing any s ∈ Q ∪ {∞} as s = α(∞), holomorphy at s is defined in terms<br />
of holomorphy at ∞ via the [α] k opera<strong>to</strong>r. Since f[α] k is holomorphic <strong>on</strong> H and weakly<br />
modular with respect <strong>to</strong> α −1 Γα, the noti<strong>on</strong> of its holomorphy at ∞ makes sense.<br />
Definiti<strong>on</strong> 3.1.3. Let Γ be a c<strong>on</strong>gruence subgroup of SL 2 (Z) and let k be an integer. A<br />
functi<strong>on</strong> f : H → C is a modular form of weight k with respect <strong>to</strong> Γ, if<br />
1. f is holomorphic, and weakly modular with respect <strong>to</strong> Γ,<br />
2. f[α] k is holomorphic at ∞ for all α ∈ SL 2 (Z).<br />
If in additi<strong>on</strong>, a 0 (f[α] k ) = 0 in the Fourier expansi<strong>on</strong> of f[α] k for all α ∈ SL 2 (Z), then f<br />
is a cusp form of weight k with respect <strong>to</strong> Γ. The modular <strong>forms</strong> and cusp <strong>forms</strong> of weight<br />
k with respect <strong>to</strong> Γ are denoted by M k (Γ) and S k (Γ), respectively.<br />
Let N be a natural number. For each Dirichlet character χ modulo N, define the<br />
χ-eigenspace of M k (Γ 1 (N)) and S k (Γ 1 (N)) as<br />
M k (N, χ) := {f ∈ M k (Γ 1 (N)) | f[γ] k = χ(d γ )f, for all γ ∈ Γ 0 (N)} , and<br />
S k (N, χ) := M k (N, χ) ∩ S k (Γ 1 (N)).<br />
(Here d γ denotes the lower right entry of γ.) In particular, the eigenspace M k (N, 1) is<br />
M k (Γ 0 (N)). The vec<strong>to</strong>r spaces M k (Γ 1 (N)) and S k (Γ 1 (N)) decomposes as the direct sum<br />
of the eigenspaces, i.e.,<br />
M k (Γ 1 (N)) = ⊕ χ M k (N, χ) and S k (Γ 1 (N)) = ⊕ χ S k (N, χ).<br />
26
3.2 Hecke algebras<br />
We now briefly recall the definiti<strong>on</strong> of the Hecke opera<strong>to</strong>rs, which play an important role<br />
in the study of modular <strong>forms</strong>, and cusp <strong>forms</strong>. For more details, refer <strong>to</strong> [Miy89, §2.8]<br />
For any c<strong>on</strong>gruence subgroup Γ of SL 2 (Z), we put<br />
˜Γ = {g ∈ GL + 2 (Q) | gΓg−1 and Γ are commensurable}.<br />
Let Γ 1 and Γ 2 be c<strong>on</strong>gruence subgroups of SL 2 (Z). It is easy <strong>to</strong> see that ˜Γ 1 = ˜Γ 2 . Let ∆<br />
be subsemigroup of ˜Γ 1 . For any α ∈ ∆, we set<br />
Γ 1 αΓ 2 = {γ 1 αγ 2 | γ 1 ∈ Γ 1 , γ 2 ∈ Γ 2 }<br />
is a double coset in GL + 2 (Q). Using these double cosets, we define Hecke opera<strong>to</strong>rs and<br />
they transform modular <strong>forms</strong> with respect <strong>to</strong> Γ 1 in<strong>to</strong> modular <strong>forms</strong> with respect <strong>to</strong> Γ 2 .<br />
Definiti<strong>on</strong> 3.2.1. For c<strong>on</strong>gruence subgroups Γ 1 and Γ 2 of SL 2 (Z) and α ∈ ∆, the weightk<br />
opera<strong>to</strong>r [Γ 1 αΓ 2 ] k takes functi<strong>on</strong>s f ∈ M k (Γ 1 ) <strong>to</strong><br />
f[Γ 1 αΓ 2 ] k = ∑ j<br />
f[β j ] k ∈ M k (Γ 2 )<br />
where {β j } are orbit representatives, i.e., Γ 1 αΓ 2 = ∪ j Γ 1 β j is a disjoint uni<strong>on</strong>.<br />
By [Miy89, Thm. 2.8.1], the weight-k opera<strong>to</strong>r is well-defined and it maps S k (Γ 1 ) in<strong>to</strong><br />
S k (Γ 2 ). Again by the same theorem, the weight-k opera<strong>to</strong>r maps the space of modular<br />
<strong>forms</strong> M k (Γ 1 , χ) (resp., S k (Γ 1 , χ)) <strong>to</strong> M k (Γ 2 , χ) (resp., S k (Γ 2 , χ)).<br />
3.2.1 Hecke opera<strong>to</strong>rs for Γ 1 (N)<br />
The group Γ 0 (N) acts <strong>on</strong> M k (Γ 1 (N)), and it induces an acti<strong>on</strong> of the quotient (Z/NZ) ∗ .<br />
The acti<strong>on</strong> of α = ( a b<br />
c d<br />
)<br />
, determined by d (mod N) and denoted 〈d〉, is<br />
〈d〉 : M k (Γ 1 (N)) → M k (Γ 1 (N))<br />
given by 〈d〉f = f[α] k for any α ∈ Γ 0 (N) such that d α ≡ d (mod N). This Hecke opera<strong>to</strong>r<br />
is called as diam<strong>on</strong>d opera<strong>to</strong>r. By definiti<strong>on</strong>, we see that<br />
M k (N, χ) = {f ∈ M k (Γ 1 (N)) | 〈d〉f = χ(d)f for all d ∈ (Z/NZ) ∗ }.<br />
Since the weight-k opera<strong>to</strong>r [ ] k preserves the space of cusp <strong>forms</strong>, we see that the diam<strong>on</strong>d<br />
opera<strong>to</strong>r preserves the spaces of cusp<strong>forms</strong> S k (Γ 1 (N)).<br />
The sec<strong>on</strong>d type of Hecke opera<strong>to</strong>r is also a weight-k opera<strong>to</strong>r [Γ 1 (N)αΓ 1 (N)] k , but<br />
now α = ( )<br />
1 0<br />
0 p , for a prime p. This opera<strong>to</strong>r is denoted Tp . For primes p dividing N,<br />
sometimes we write U p for T p . Thus<br />
is given by<br />
T p : M k (Γ 1 (N)) → M k (Γ 1 (N)),<br />
T p (f) = f[Γ 1 (N) ( 1 0<br />
0 p<br />
)<br />
Γ1 (N)] k .<br />
Using a left coset decompositi<strong>on</strong> of Γ 1 (N) ( )<br />
1 0<br />
0 p Γ1 (N), we can explicitly write down the<br />
acti<strong>on</strong> of [Γ 1 (N) ( )<br />
1 0<br />
0 p Γ1 (N)] k <strong>on</strong> modular <strong>forms</strong>.<br />
27
Propositi<strong>on</strong> 3.2.2. Let Γ 1 = Γ 2 = Γ 1 (N), and let α = ( )<br />
1 0<br />
0 p where p is a prime. The<br />
opera<strong>to</strong>r T p = [Γ 1 (N) ( )<br />
1 0<br />
0 p Γ1 (N)] k <strong>on</strong> M k (Γ 1 (N)) is given by<br />
⎧<br />
∑p−1<br />
f[ ( 1 j<br />
)<br />
0 p ]k if p | N,<br />
⎪⎨<br />
j=0<br />
T p (f) =<br />
∑p−1<br />
⎪⎩ f[ ( 1 j<br />
)<br />
0 p ]k + f[ ( m n<br />
)( p 0<br />
)<br />
N p 0 1 ]k if p ∤ N, where mp − nN = 1.<br />
j=0<br />
Letting Γ 1 = Γ 2 = Γ 0 (N) instead and keeping same α as above gives the same orbit<br />
representatives for Γ 1 αΓ 2 , but in this case ( m n<br />
)( p 0<br />
) (<br />
N p 0 1 can be replaced by p 0<br />
)<br />
0 1 .<br />
Using the propositi<strong>on</strong> above, we can describe the effect of T p <strong>on</strong> Fourier coefficients<br />
as follows: If f ∈ M k (N, χ) then also T p (f) ∈ M k (N, χ) and<br />
a n (T p f) = a np (f) + χ(p)p k−1 a n/p (f) for f ∈ M k (N, χ),<br />
where a n/p (f) = 0, if p ∤ n. Moreover, the acti<strong>on</strong> of T p preserves the subspaces S k (Γ 1 (N))<br />
and S k (N, χ).<br />
So far the Hecke opera<strong>to</strong>rs 〈d〉 and T p are defined for d ∈ (Z/NZ) ∗ and p prime. Now<br />
we extend the definiti<strong>on</strong>s <strong>to</strong> n and T n for all n ∈ N. For n ∈ N with (n, N) = 1, 〈n〉 is<br />
determined by n (mod N). For n ∈ N with (n, N) > 1, define 〈n〉 = 0, the zero opera<strong>to</strong>r.<br />
To define T n , set T 1 = 1 (the identity opera<strong>to</strong>r); T p is already defined for primes p.<br />
For prime powers, define inductively<br />
T p r = T p T p r−1 − p k−1 〈p〉T p r−2, for r ≥ 2.<br />
Extend the definiti<strong>on</strong> multiplicatively <strong>to</strong> T n for all n,<br />
T n = ∏ i<br />
T p e i<br />
where n = ∏ i<br />
p e i<br />
.<br />
Similar <strong>to</strong> the above, we can explicitly write the Fourier coefficients of T n (f), for any<br />
f ∈ M k (Γ 1 (N)). If f ∈ M k (Γ 1 (N)), then a m (T n f) = ∑ d|(m,n) dk−1 a mn/d 2(〈d〉f). In<br />
particular, if f ∈ M k (N, χ), then a m (T n f) = ∑ d|(m,n) χ(d)dk−1 a mn/d 2(f).<br />
For N ∈ N, we let h k (Γ 1 (N), C) denote the subring of End C (S k (Γ 1 (N))) generated over<br />
C by the Hecke opera<strong>to</strong>rs T p for all primes p and the opera<strong>to</strong>rs 〈d〉 acting <strong>on</strong> S k (Γ 1 (N)).<br />
Each element f in S k (Γ 1 (N)) has the Fourier expansi<strong>on</strong> f(z) = ∑ n a n(f)q n , for a n (f) ∈<br />
C. Hence, we may embed S k (Γ 1 (N)) in<strong>to</strong> the power series C[[q]]. One may then give<br />
a rati<strong>on</strong>al structure <strong>on</strong> S k (Γ 1 (N)) by defining the A-rati<strong>on</strong>al subspace S k (Γ 1 (N), A) for<br />
each subalgebra A of C or C p , by S k (Γ 1 (N), A) := S k (Γ 1 (N)) ∩ A[[q]]. Similar <strong>to</strong> above,<br />
<strong>on</strong>e can define the Hecke algebra over Z, and hence over A. We denote this Hecke algebra<br />
by h k (Γ 1 (N), A).<br />
3.2.2 Old<strong>forms</strong> and New<strong>forms</strong><br />
There is a way <strong>to</strong> move between levels is <strong>to</strong> observe that, if M|N then S k (Γ 1 (M)) ⊆<br />
S k (Γ 1 (N)). Another way <strong>to</strong> embed S k (Γ 1 (M)) in<strong>to</strong> S k (Γ 1 (N)) is by composing with the<br />
multiply-by-d map, where d is any fac<strong>to</strong>r of N/M. For any such d, let α d = ( d 0<br />
0 1<br />
)<br />
so that<br />
(f[α d ] k )(τ) = d k−1 f(dτ)<br />
28
for f : H → C. The injective linear map [α d ] k takes S k (Γ 1 (M)) <strong>to</strong> S k (Γ 1 (N)), lifting the<br />
level from M <strong>to</strong> N.<br />
Definiti<strong>on</strong> 3.2.3. For each divisor d of N, let i d be the map<br />
i d : (S k (Γ 1 (Nd −1 ))) 2 → S k (Γ 1 (N))<br />
given by (f, g) → f + g[α d ] k . The subspace of old<strong>forms</strong> of level N is given by<br />
S k (Γ 1 (N)) old = ∑ p|N<br />
i p ((S k (Γ 1 (Np −1 ))) 2<br />
and the subspace of new<strong>forms</strong> of level N is the orthog<strong>on</strong>al complement with respect <strong>to</strong><br />
the Peterss<strong>on</strong> inner product,<br />
S k (Γ 1 (N)) new = (S k (Γ 1 (N)) old ) ⊥ .<br />
For all n ∈ N, the Hecke opera<strong>to</strong>rs T n and 〈n〉 preserve the space of old<strong>forms</strong> and<br />
new<strong>forms</strong>. Moreover, the spaces S k (Γ 1 (N)) old and S k (Γ 1 (N)) new have orthog<strong>on</strong>al bases<br />
of eigen<strong>forms</strong> for the Hecke opera<strong>to</strong>rs away from the level, {T n , 〈n〉 : (n, N) = 1}.<br />
Definiti<strong>on</strong> 3.2.4. A n<strong>on</strong>zero modular form f ∈ M k (Γ 1 (N)) that is an eigenform for the<br />
Hecke opera<strong>to</strong>rs T n and 〈n〉 for all n ∈ N is a Hecke eigenform or simply an eigenform.<br />
The eigenform f(τ) = ∑ ∞<br />
n=0 a n(f)q n is normalized, when a 1 (f) = 1. A newform is a<br />
normalized eigenform in S k (Γ 1 (N)) new .<br />
Theorem 3.2.5 ([DS05]). Let f ∈ S k (Γ 1 (N)) new be a n<strong>on</strong>zero eigenform for the Hecke<br />
opera<strong>to</strong>rs T n and 〈n〉 for all n with (n, N) = 1. Then f is an eigenform.<br />
3.2.3 Complex multiplicati<strong>on</strong><br />
We briefly recall the definiti<strong>on</strong> of a newform <strong>to</strong> have complex multiplicati<strong>on</strong>. For more<br />
details, refer <strong>to</strong> [Rib77, §3].<br />
Let f ∈ S k (N, χ) be a newform and let ϕ be a Dirichlet character. We let f ⊗ ϕ be<br />
f ⊗ ϕ := ∑ n<br />
ϕ(n)a n (f)q n .<br />
By [Shi71, Prop. 3.64], f ⊗ ϕ is a modular form of weight k <strong>on</strong> Γ 1 (ND 2 ) of nebentypus<br />
χϕ 2 , if ϕ is defined mod D. We have (f ⊗ ϕ)| Tp = ϕ(p)a p (f)(f ⊗ ϕ) for p ∤ ND.<br />
Definiti<strong>on</strong> 3.2.6. Suppose ϕ ≠ 1. The form f has complex multiplicati<strong>on</strong> (or CM for<br />
short) by ϕ, if ϕ(p)a p (f) = a p (f) for all primes p in a set of primes of density 1.<br />
Remark 3.2.7. If f has CM by ϕ, then we have χϕ 2 = χ, and ϕ(p)a p (f) = a p (f) for<br />
all p ∤ DN. The first equati<strong>on</strong> implies that ϕ is a quadratic character. If the kernel of ϕ<br />
defines a imaginary quadratic field F , we say that f has CM by F .<br />
29
Chapter 4<br />
Galois Representati<strong>on</strong>s and<br />
Au<strong>to</strong>morphic Forms <strong>on</strong> GL n<br />
We start the chapter by stating the results (cf. §4.1) which motivated the first part of<br />
the thesis. Then we recall some definiti<strong>on</strong>s and some facts from p-adic Hodge theory (cf.<br />
§4.2). In §4.3, we reprove Theorem 4.1.1, using the work of Colmez-F<strong>on</strong>taine [CF00], and<br />
it serves <strong>to</strong> illustrate the techniques that will be used. In §4.4, we recall some generalities<br />
about Galois representati<strong>on</strong>s associated <strong>to</strong> au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ).<br />
4.1 Motivati<strong>on</strong><br />
For a global or a <strong>local</strong> field F , we set G F = Gal( ¯F /F ). For simplicity, we write G p for<br />
G Qp . Let ℘ be the prime of ¯Q determined by a fixed embedding of ¯Q in<strong>to</strong> ¯Q p .<br />
Let f = ∑ ∞<br />
n=1 a n(f)q n be a normalized eigenform (i.e., primitive form) of weight<br />
k ≥ 2, level Np r ≥ 1, with (N, p) = 1, and nebentypus χ. Let ℘ also denote the induced<br />
prime of K f = Q(a n (f)), the Hecke field of f, and let K f,℘ denote the completi<strong>on</strong> of K f<br />
at ℘. There is a global Galois representati<strong>on</strong><br />
ρ f,℘ : G Q → GL 2 (K f,℘ ) (4.1.1)<br />
associated <strong>to</strong> f (and ℘) by Deligne which has the property that for all primes l ∤ Np,<br />
trace(ρ f,℘ (Frob l )) = a l (f) and det(ρ f,℘ (Frob l )) = χ(l)l k−1 .<br />
Thus det(ρ f,℘ ) = χχ k−1<br />
cyc,p, where χ cyc,p is the p-adic cyclo<strong>to</strong>mic character.<br />
It is a well-known result of Ribet that the global representati<strong>on</strong> ρ f,℘ is irreducible.<br />
However, if f is ordinary at ℘, i.e., a p (f) is a ℘-adic unit, then an important theorem of<br />
Wiles, valid more generally for Hilbert modular <strong>forms</strong>, says that the corresp<strong>on</strong>ding <strong>local</strong><br />
representati<strong>on</strong> ρ f,℘ restricted <strong>to</strong> G p is reducible.<br />
Theorem 4.1.1 ([Wil88]). Let f be a ℘-ordinary primitive form as above. Then the<br />
restricti<strong>on</strong> of ρ f,℘ <strong>to</strong> the decompositi<strong>on</strong> subgroup G p is reducible. More precisely, there<br />
exists a basis in which<br />
(<br />
)<br />
χ p · λ(β/p k−1 ) · χ k−1<br />
cyc,p u<br />
ρ f,℘ | Gp ∼<br />
, (4.1.2)<br />
0 λ(α)<br />
31
where χ = χ p χ ′ is the decompositi<strong>on</strong> of χ in<strong>to</strong> its p and prime-<strong>to</strong>-p-parts, λ(x) : G p →<br />
K × f,℘ is the unramified character which takes arithmetic Frobenius <strong>to</strong> x, and u : G p → K f,℘<br />
is a c<strong>on</strong>tinuous functi<strong>on</strong>. Here α is (i) the unit root of X 2 − a p (f)X + p k−1 χ(p) if p ∤ N,<br />
(ii) the unit a p (f) if p||N, p ∤ c<strong>on</strong>d(χ) and k = 2, and (iii) the unit a p (f) if p|N,<br />
v p (N) = v p (c<strong>on</strong>d(χ)). In all cases αβ = χ ′ (p)p k−1 .<br />
Moreover, in case (ii), a p (f) is a unit if and <strong>on</strong>ly if k = 2, and <strong>on</strong>e can easily show that<br />
ρ f,℘ | Gp is irreducible when k > 2.<br />
Urban has generalized Theorem 4.1.1 <strong>to</strong> the case of primitive Siegel modular cusprepresentati<strong>on</strong>s<br />
of genus 2. We briefly recall this result here. Let π be a cuspidal au<strong>to</strong>morphic<br />
representati<strong>on</strong> of GSp 4 (A Q ) whose Archimedean comp<strong>on</strong>ent π ∞ bel<strong>on</strong>gs <strong>to</strong> the discrete<br />
series, with cohomological weights (a, b; a + b) with a ≥ b ≥ 0. For each prime p, Laum<strong>on</strong>,<br />
Taylor and Weissauer have defined a four-dimensi<strong>on</strong>al Galois representati<strong>on</strong><br />
ρ π,p : G Q → GL 4 ( ¯Q p )<br />
with standard properties. Let p be an unramified prime for π. Then Tilouine and Urban<br />
have generalized the noti<strong>on</strong> of ordinariness for such primes p in three ways <strong>to</strong> what they<br />
call Borel ordinary, Siegel ordinary, and Klingen ordinary (these terms come from the<br />
underlying parabolic subgroups of GSp 4 (A Q )). In the Borel case, the p-ordinariness of π<br />
implies that the Hecke polynomial of π p , namely<br />
(X − α)(X − β)(X − γ)(X − δ),<br />
has the property that the p-adic valuati<strong>on</strong>s of α, β, γ and δ are 0, b+1, a+2 and a+b+3,<br />
respectively.<br />
Theorem 4.1.2 ([Urb05], [TU99]). Say π is a Borel p-ordinary au<strong>to</strong>morphic cusp representati<strong>on</strong><br />
for GSp 4 (A Q ) which is stable at ∞ with cohomological weights (a, b; a + b).<br />
Then the restricti<strong>on</strong> of ρ π,p <strong>to</strong> the decompositi<strong>on</strong> subgroup G p is upper-triangular. More<br />
precisely, there is a basis in which ρ π,p | Gp<br />
⎛<br />
⎞<br />
λ(δ/p a+b+3 ) · χ a+b+3<br />
cyc,p ∗ ∗ ∗<br />
0 λ(γ/p a+2 ) · χ a+2<br />
cyc,p ∗ ∗<br />
∼ ⎜<br />
⎝ 0 0 λ(β/p b+1 ) · χ b+1 ⎟<br />
cyc,p ∗ ⎠ ,<br />
0 0 0 λ(α)<br />
where λ(x) is the unramified character which takes arithmetic Frobenius <strong>to</strong> x.<br />
We remark that ρ π,p above is the c<strong>on</strong>tragredient of the <strong>on</strong>e used in [Urb05] (we also use<br />
the arithmetic Frobenius in defining our unramified characters), so the theorem matches<br />
exactly with [Urb05, Cor. 1 (iii)]. Similar results in the Siegel and Klingen cases can be<br />
found in the other parts of [Urb05, Cor. 1].<br />
The <strong>local</strong> Galois representati<strong>on</strong>s appearing in Theorems 4.1.1 and 4.1.2 are sometimes<br />
referred <strong>to</strong> as (p, p)-Galois representati<strong>on</strong>s. The main point of our thesis is <strong>to</strong> prove<br />
structure theorems for the <strong>local</strong> (p, p)-Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> au<strong>to</strong>morphic<br />
representati<strong>on</strong> of GL n (A Q ).<br />
In this study, <strong>on</strong>e of the key ingredients is p-adic Hodge theory, especially, the equivalence<br />
of categories between potentially semistable representati<strong>on</strong>s and filtered (ϕ, N)-<br />
modules with coefficients and descent data.<br />
32
4.2 p-adic Hodge theory<br />
We start by recalling some definiti<strong>on</strong>s from p-adic Hodge theory. For more details, see<br />
[F<strong>on</strong>94], [FO], and [GM09].<br />
Let p be a prime. Let F be a finite Galois extensi<strong>on</strong> of Q p and let F 0 be the maximal<br />
unramified extensi<strong>on</strong> of Q p c<strong>on</strong>tained in F . Let E be another finite extensi<strong>on</strong> of Q p .<br />
Definiti<strong>on</strong> 4.2.1. A filtered (ϕ, N, F, E)-module (with descent data, i.e., Gal(F/Q p )-<br />
acti<strong>on</strong>) is a free of finite rank (F 0 ⊗ Qp E)-module D endowed with<br />
1. the Frobenius endomorphism: an F 0 -semi-linear, E-linear, bijective map ϕ : D → D,<br />
2. the m<strong>on</strong>odromy opera<strong>to</strong>r: an F 0 ⊗ Qp E-linear, nilpotent endomorphism N : D → D<br />
which satisfies Nϕ = pϕN,<br />
3. an F 0 -semi-linear, E-linear acti<strong>on</strong> of Gal(F/Q p ) (the acti<strong>on</strong> <strong>on</strong> F 0 is via the projecti<strong>on</strong><br />
<strong>to</strong> Gal(F 0 /Q p )), which commutes with the acti<strong>on</strong> of ϕ and N,<br />
4. a decreasing filtrati<strong>on</strong> (Fil i D F ) i∈Z of F ⊗ Qp E-submodules of D F = F ⊗ F0 D satisfying<br />
Fil i D F = 0 for i ≫ 0 and Fil i D F = D F for i ≪ 0 and which are stable under<br />
the acti<strong>on</strong> of Gal(F/Q p ).<br />
For simplicity, we call a filtered (ϕ, N, F, E)-module as a filtered module or a crystal,<br />
if the parameters are clear from the c<strong>on</strong>text.<br />
4.2.1 F<strong>on</strong>taine’s rings<br />
We wish <strong>to</strong> recall briefly the definiti<strong>on</strong> of some of F<strong>on</strong>taine’s rings of periods.<br />
F<strong>on</strong>taine’s rings are <strong>to</strong>pological Q p -algebras B equipped with an acti<strong>on</strong> of G p such<br />
that B G F<br />
is a field for every finite extensi<strong>on</strong> F of Q p . For us the main example will be<br />
B st which sits (n<strong>on</strong>-can<strong>on</strong>ically) between B cris and B dR . Let<br />
Ẽ + = lim ←−<br />
x→x p O Cp = {x = (x (0) , x (1) , ..., ) | x (i) ∈ O Cp , (x (i+1) ) p = x (i) }.<br />
Additi<strong>on</strong> given by (x + y) (i) = lim<br />
n→∞ (x(i+n) + y (i+n) ) pn and comp<strong>on</strong>entwise multiplicati<strong>on</strong><br />
turn Ẽ+ in<strong>to</strong> a perfect ring of characteristic p. Let Ã+ = W (Ẽ+ ) be the ring of Witt<br />
vec<strong>to</strong>rs, and let ˜B + = Ã+ [1/p]. Elements of Ã+ , respectively ˜B + , may be written as<br />
∑ p k [x k ] where the sum is over k ≥ 0, respectively k ≥ −n, for some n ≥ 0, and [ ]<br />
denotes Teichmüller representative.<br />
There is a ring homomorphism θ : ˜B + → C p which maps ∑ p k [x k ] <strong>to</strong> ∑ p k x (0)<br />
k<br />
, and<br />
ker(θ) = (ω) is a principal ideal. We let B + dR be the completi<strong>on</strong> of ˜B + with respect <strong>to</strong><br />
ker(θ). So elements of B + dR can be written as ∑ b n ω n with b n ∈ ˜B + .<br />
Fix an element ɛ = (ɛ (0) , ɛ(1), . . .) ∈ Ẽ+ where ɛ (n) is a primitive p n -th root of unity.<br />
Then θ(1 − [ɛ]) = 0 and t := log[ ] c<strong>on</strong>verges as a series in B + dR . Define B dR = B + dR [1/t].<br />
Then B dR has a natural filtrati<strong>on</strong> given by Fil i B dR = t i B + dR<br />
. Also, under the natural<br />
Galois acti<strong>on</strong>, B G F<br />
dR = F for any finite extensi<strong>on</strong> F of Q p.<br />
Now set<br />
B + cris = {x = ∑ w n<br />
b n ∈ B +<br />
n!<br />
dR | b n ∈ ˜B + , b n → 0}<br />
33
where the c<strong>on</strong>vergence of the b n is in the p-adic <strong>to</strong>pology. Finally set B cris = B + cris [1/t].<br />
Then B cris has a Frobenius ϕ : B cris → B cris , and B G F<br />
cris = F 0.<br />
Now let B st = B cris [Y ] be the polynomial ring in <strong>on</strong>e variable over B cris . Extend<br />
the Frobenius <strong>to</strong> B st by defining ϕ(Y ) = pY . There is also the m<strong>on</strong>odromy opera<strong>to</strong>r<br />
N = − d<br />
dY<br />
<strong>on</strong> B st which satisfies Nϕ = pϕN. Finally the Galois acti<strong>on</strong> <strong>on</strong> Y may be<br />
specified, and again <strong>on</strong>e has B G F<br />
st = F 0 . Set ˜p = (p (0) , p (1) , ...) ∈ Ẽ+ where p (n) is a<br />
primitive p n -th root of p. We will think of B st as a sub-ring of B dR by mapping Y <strong>to</strong><br />
log[˜p], though the definiti<strong>on</strong> of this last element depends <strong>on</strong> a choice, that of log p (p),<br />
which we will take <strong>to</strong> be 0.<br />
4.2.2 Newt<strong>on</strong> and Hodge numbers<br />
We start by recalling the definiti<strong>on</strong>s of Newt<strong>on</strong> and Hodge numbers, and some results.<br />
Let D be a filtered (ϕ, N, F, E)-module. Then by forgetting the E-module structure,<br />
D is also a filtered (ϕ, N, F, Q p )-module. Let d = dim F0 D. Then Λ d F 0<br />
D is a filtered<br />
(ϕ, N, F, Q p )-module of dimensi<strong>on</strong> 1 over F 0 . Set<br />
t H (D) = max{i ∈ Z | Fil i (F ⊗ F0 Λ d F 0<br />
D) ≠ 0}, t N (D) = v p (λ),<br />
where for a n<strong>on</strong>-zero element x of Λ d F 0<br />
D, ϕ(x) = λx, with λ ∈ F 0 , where v p is a normalized<br />
valuati<strong>on</strong> such that v p (p) = 1. One says that D is admissible (originally weakly<br />
admissible), if t H (D) = t N (D), and<br />
• for any F 0 -submodule D ′ of D stable by ϕ and N, t H (D) ≤ t N (D), where D ′ F ⊂ D F<br />
is equipped with the induced filtrati<strong>on</strong>.<br />
It turns out that the filtered (ϕ, N, F, E)-module D is admissible, if the sec<strong>on</strong>d c<strong>on</strong>diti<strong>on</strong><br />
above is replaced by the following weaker c<strong>on</strong>diti<strong>on</strong>:<br />
• for any (F 0 ⊗ Qp E)-submodule D ′ of D stable by ϕ and N, t H (D) ≤ t N (D), where<br />
again D ′ F ⊂ D F is equipped with the induced filtrati<strong>on</strong>.<br />
From now, we shall assume that all the c<strong>on</strong>jugates of F are c<strong>on</strong>tained in E.<br />
Lemma 4.2.2. Suppose D 2 ⊆ D 1 are two free of finite rank modules over F ⊗ Qp E. Then<br />
D 1 /D 2 is also free of finite rank = rank(D 1 ) − rank(D 2 ).<br />
Proof. We start with a basis of D 2 . For any F ⊗ Qp E-module D, we have that D ≃<br />
∏<br />
σ:F ↩→E D σ, where D σ = D ⊗ F ⊗E,σ E, a E-vec<strong>to</strong>r space. In each projecti<strong>on</strong> D 1σ of D 1 ,<br />
we can extend the basis of D 2σ <strong>to</strong> D 1σ . Now pull back the extended basis vec<strong>to</strong>rs in each<br />
D 1σ , we get a basis of D 1 , which extends the basis of D 2 .<br />
Lemma 4.2.3 (Newt<strong>on</strong> number). Suppose D is a filtered (ϕ, N, F, E)-module of rank n,<br />
such that the acti<strong>on</strong> of ϕ is E-semisimple, i.e., there exists a basis {e 1 , · · · , e n } of D such<br />
that ϕ(e i ) = α i e i , for some α i ∈ E × . Then<br />
t N (D) = [E : Q p ] ·<br />
n∑<br />
v p (α i ).<br />
Proof. The proof is standard from the definiti<strong>on</strong> of the Newt<strong>on</strong> number.<br />
34<br />
i=1
Lemma 4.2.4 (Hodge number). Suppose D is a filtered (ϕ, N, F, E)-module of rank n.<br />
Then<br />
t H (D) = [E : Q p ] · ∑<br />
i · rank F ⊗Qp E gr i D F .<br />
i∈Z<br />
Proof. Since D is a filtered module, there exists a filtrati<strong>on</strong> (Fil i D F ) i∈Z of (F ⊗ Qp E)-<br />
submodules of D F . Then, by forgetting the E-module structure, D is also a filtered<br />
(ϕ, N, F, Q p )-module and each term of the filtrati<strong>on</strong> {Fil i D F } can also thought of as an<br />
F -module. By a standard formula (see, e.g., [FO, Prop. 6.45]), we have:<br />
t H (D) = ∑ i∈Z<br />
i · dim F gr i D F ,<br />
where gr i D F = Fil i D F /Fil i+1 D F with gr i D F thought of as an F -module. By [Sav05] and<br />
by Lemma 4.2.2, we see that each Fil i D F and gr i D F are free of finite rank over F ⊗ Qp E,<br />
respectively. Since dim F D = [E : Q p ] · rank F ⊗Qp ED, we obtain the lemma.<br />
Remark 4.2.5. By the last two lemmas, <strong>on</strong>e can drop the comm<strong>on</strong> fac<strong>to</strong>r of [E : Q p ]<br />
when checking the admissibility of a filtered (ϕ, N, F, E)-module.<br />
Corollary 4.2.6. Suppose D is a filtered (ϕ, N, F, E)-module of rank 1. Then<br />
t H (D) = [E : Q p ] · β,<br />
where β is the unique integer such that 0 = Fil β+1 D F Fil β D F = D F .<br />
Lemma 4.2.7. Let D 1 , D 2 be two filtered (ϕ, N, F, E)-modules, of rank r 1 , r 2 , respectively.<br />
Assume that the acti<strong>on</strong> of ϕ <strong>on</strong> D 1 , D 2 is semisimple. Then<br />
t N (D 1 ⊗ D 2 ) = rank(D 1 ) t N (D 2 ) + rank(D 2 ) t N (D 1 ),<br />
t H (D 1 ⊗ D 2 ) = rank(D 1 ) t H (D 2 ) + rank(D 2 ) t H (D 1 ).<br />
(4.2.1a)<br />
(4.2.1b)<br />
Proof. These formulas are well-known if E = Q p . The proof of (4.2.1a) is an easy check.<br />
It is enough <strong>to</strong> prove (4.2.1b) when D 1 , D 2 are of rank 1, since<br />
∧ r 1r 2<br />
(D 1 ⊗ D 2 ) = ∧ r 2<br />
(D 2 ) ⊗ · · · ⊗ ∧ r 2<br />
(D 2 ) ⊗ ∧ r 1<br />
(D<br />
} {{ }<br />
1 ) ⊗ · · · ⊗ ∧ r 1<br />
(D 1 ),<br />
} {{ }<br />
r 1 -times<br />
r 2 -times<br />
where the tensor products are taken over F ⊗ Qp E. In this case (4.2.1b) follows from<br />
Corollary 4.2.6.<br />
4.2.3 Potentially semistable representati<strong>on</strong>s<br />
Let V be a finite-dimensi<strong>on</strong>al vec<strong>to</strong>r space over E.<br />
Definiti<strong>on</strong> 4.2.8. A representati<strong>on</strong> ρ : G p → GL(V ) is said <strong>to</strong> be semistable over F , or<br />
F -semistable, if dim F0 D st,F (V ) = dim F0 (B st ⊗ Qp V ) G F<br />
= dim Qp V , where F 0 = B G F<br />
st . If<br />
such an F exists, ρ is said <strong>to</strong> be a potentially semistable representati<strong>on</strong>. If F = Q p , we<br />
say that ρ is semistable.<br />
Remark 4.2.9. If ρ is F -semistable, then ρ is F ′ -semistable for any finite extensi<strong>on</strong> of<br />
F ′ /F . Hence we may and do assume that F is Galois over Q p .<br />
35
The following fundamental theorem plays a key role in subsequent arguments.<br />
Theorem 4.2.10 ([CF00]). There is an equivalence of categories between F -semistable<br />
representati<strong>on</strong>s ρ : G p → GL n (E) with Hodge-Tate weights −β n ≤ · · · ≤ −β 1 and admissible<br />
filtered (ϕ, N, F, E)-module D of rank n over F 0 ⊗ Qp E such that the jumps in the<br />
Hodge filtrati<strong>on</strong> Fil i D F <strong>on</strong> D F := F ⊗ F0 D are at β 1 ≤ · · · ≤ β n .<br />
The jumps in the filtrati<strong>on</strong> <strong>on</strong> D F = F ⊗ F0 D st,F (ρ) are the negatives of the Hodge-<br />
Tate weights of ρ, i.e., if h is a Hodge-Tate weight, then Fil −h+1 (D F ) Fil −h (D F ). The<br />
equivalence of categories in the theorem is induced by F<strong>on</strong>taine’s func<strong>to</strong>r D st,F . The<br />
Frobenius ϕ, m<strong>on</strong>odromy N, and filtrati<strong>on</strong> <strong>on</strong> B st induce the corresp<strong>on</strong>ding structures <strong>on</strong><br />
D st,F (V ). There is also an induced acti<strong>on</strong> of Gal(F/Q p ) <strong>on</strong> D st,F (V ). As an illustrati<strong>on</strong><br />
of the power of the theorem we recall the following useful (and well-known) fact:<br />
Corollary 4.2.11. Every potentially semistable character χ : G p → E × is of the form<br />
χ = χ 0 · λ(a 0 ) · χ i cyc,p, where χ 0 is a finite order character of Gal(F/Q p ), for a cyclo<strong>to</strong>mic<br />
extensi<strong>on</strong> F of Q p , −i ∈ Z is the Newt<strong>on</strong> number of D st,F (χ), and λ(a 0 ) is the unramified<br />
character that takes arithmetic Frobenius <strong>to</strong> the unit a 0 = p −i /a ∈ O × E<br />
, where a = ϕ(v)/v<br />
for any vec<strong>to</strong>r v in D st,F (χ).<br />
Proof. Every potentially semistable χ : G Q → E × is F -semistable for a sufficiently large<br />
cyclo<strong>to</strong>mic extensi<strong>on</strong> of Q p . Let D st,F (χ) be the corresp<strong>on</strong>ding filtered (ϕ, N)-module with<br />
coefficients and descent data. Suppose that the induced Gal(F/Q p )-acti<strong>on</strong> <strong>on</strong> D st,F (χ)<br />
is given the character χ 0 . Now c<strong>on</strong>sider the F -semistable E-valued character χ ′ := χ 0 ·<br />
λ(a 0 ) · χ i cyc,p. One easily checks that D st,F (χ ′ ) = D st,F (χ). By Theorem 4.2.10, we have<br />
χ = χ ′ = χ 0 · λ(a 0 ) · χ i cyc,p.<br />
4.2.4 Weil-Deligne representati<strong>on</strong>s<br />
We now recall the definiti<strong>on</strong> of the Weil-Deligne representati<strong>on</strong> associated <strong>to</strong> an F -<br />
semistable representati<strong>on</strong> ρ : G p → GL n (E), due <strong>to</strong> F<strong>on</strong>taine. Let W F denote the Weil<br />
group of F , and when F = Q p , we denote W Qp by W p . For any (ϕ, N, F, E)-module D,<br />
we have the decompositi<strong>on</strong><br />
D ≃<br />
[F 0 :Q p]<br />
∏<br />
i=1<br />
D i , (4.2.2)<br />
where D i = D ⊗ (F0 ⊗ Qp E,σ i ) E, and σ is the arithmetic Frobenius of F 0 /Q p .<br />
Definiti<strong>on</strong> 4.2.12 (Weil-Deligne representati<strong>on</strong>). Let ρ : G p → GL n (E) be an F -<br />
semistable representati<strong>on</strong>. Let D be the corresp<strong>on</strong>ding filtered module. Noting W p /W F =<br />
Gal(F/Q p ), we let<br />
g ∈ W p act <strong>on</strong> D by (g mod W F ) ◦ ϕ −α(g) , (4.2.3)<br />
where the image of g in Gal(¯F p /F p ) is the α(g)-th power of the arithmetic Frobenius at p.<br />
We also have the an acti<strong>on</strong> of N via the m<strong>on</strong>odromy opera<strong>to</strong>r <strong>on</strong> D. These acti<strong>on</strong>s induce<br />
a Weil-Deligne acti<strong>on</strong> <strong>on</strong> each D i in (4.2.2) and the resulting Weil-Deligne representati<strong>on</strong>s<br />
are all isomorphic. This isomorphism class is defined <strong>to</strong> be the Weil-Deligne representati<strong>on</strong><br />
WD(ρ) associated <strong>to</strong> ρ.<br />
36
Remark 4.2.13. If F/Q p is <strong>to</strong>tally ramified and Frob p ∈ W p is the arithmetic Frobenius,<br />
then WD(ρ)(Frob p ) acts by ϕ −1 .<br />
Lemma 4.2.14. Let ρ : G p → GL n (E) be a potentially semistable representati<strong>on</strong>.<br />
WD(ρ) is irreducible, then so is ρ.<br />
If<br />
Proof. Suppose the space V which affords ρ is reducible. By Theorem 4.2.10, the reducibility<br />
of V is equivalent <strong>to</strong> the existence of a n<strong>on</strong>-trivial admissible filtered (ϕ, N, F, E)-<br />
submodule of D st,F (V ). From the definiti<strong>on</strong> of the Weil-Deligne acti<strong>on</strong> given above, this<br />
submodule is both W p and N stable. Thus WD(ρ) is reducible, a c<strong>on</strong>tradicti<strong>on</strong>.<br />
4.3 The case of GL 2<br />
Let f be a primitive ℘-ordinary form. Let ρ f,℘ be the associated Galois representati<strong>on</strong>.<br />
In this secti<strong>on</strong>, we shall reprove Theorem 4.1.1 (reducibility of ρ f,℘ | Gp ).<br />
Wiles’ original proof in [Wil88] involves some amount of p-adic Hodge theory. More<br />
precisely it uses some Dieud<strong>on</strong>né theory for the abelian varieties associated by Shimura <strong>to</strong><br />
cusp <strong>forms</strong> of weight 2. The result for <strong>forms</strong> of weight greater than 2 is then deduced by<br />
a clever use of certain auxiliary Λ-adic Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> Hida families<br />
of ordinary <strong>forms</strong> [Hid86a] (see [BGK10, §6] for a detailed expositi<strong>on</strong> of the argument).<br />
The proof we give below avoids Hida theory completely, and (as the expert will note) is<br />
a simple extensi<strong>on</strong> of Wiles’ weight 2 argument. We remark that this proof could not<br />
have been given in [Wil88], since the equivalence of categories of Colmez and F<strong>on</strong>taine<br />
(Theorem 4.2.10) was of course unavailable at the time.<br />
A key ingredient in our proof is the fact that Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> primitive<br />
<strong>forms</strong> live in strictly compatible systems of Galois representati<strong>on</strong>s [Sai97]. The c<strong>on</strong>sequent<br />
ability <strong>to</strong> transfer informati<strong>on</strong> about the Weil-Deligne parameter between various<br />
members of the family has been used <strong>to</strong> great effect in recent times (e.g., in the Khare-<br />
Wintenberger proof of Serre’s c<strong>on</strong>jecture) and is important for us as well. We start by<br />
recalling the definiti<strong>on</strong> of such a system of Galois representati<strong>on</strong>s following [KW09, §5].<br />
Let K be a number field, l be a prime, and let ρ : G K → GL n ( ¯Q l ) be a c<strong>on</strong>tinuous<br />
Galois representati<strong>on</strong>.<br />
Definiti<strong>on</strong> 4.3.1. We say that ρ is geometric, if it is unramified outside a finite set<br />
of primes of K and its restricti<strong>on</strong>s <strong>to</strong> the decompositi<strong>on</strong> group at primes above l are<br />
potentially semistable.<br />
For every prime q of K, a geometric representati<strong>on</strong> defines a representati<strong>on</strong> of W ′ q,<br />
the Weil-Deligne group of Q q , with values in GL n ( ¯Q l ), well-defined up <strong>to</strong> c<strong>on</strong>jugacy.<br />
For primes q of characteristic not l, the definiti<strong>on</strong> comes from the theory of Deligne-<br />
Grothendieck, and for primes q of characteristic l, the definiti<strong>on</strong> comes from F<strong>on</strong>taine<br />
theory (cf. Definiti<strong>on</strong> 4.2.12).<br />
Definiti<strong>on</strong> 4.3.2. For a number field L, we call an L-rati<strong>on</strong>al, n-dimensi<strong>on</strong>al strictly<br />
compatible system of geometric representati<strong>on</strong>s (ρ l ) of G K the data of:<br />
1. For each prime l and each embedding i : L ↩→ ¯Q l , a c<strong>on</strong>tinuous, semisimple representati<strong>on</strong><br />
ρ l : G K → GL n ( ¯Q l ) that is geometric.<br />
37
2. For each prime q of K, an F -semisimple (Frobenius semisimple) representati<strong>on</strong><br />
r q : W ′ q → GL n (L) such that:<br />
• r q is unramified for all q outside a finite set.<br />
• for each l and each i : L ↩→ ¯Q l , the Frobenius semisimple Weil-Deligne representati<strong>on</strong><br />
W ′ q → GL n ( ¯Q l ) associated <strong>to</strong> ρ l | Gq is c<strong>on</strong>jugate <strong>to</strong> r q (via the<br />
embedding i : L ↩→ ¯Q l ).<br />
• There are n-distinct integers β 1 < · · · < β n , such that ρ l has Hodge-Tate<br />
weights {−β 1 , . . . , −β n } (the minus signs arise since the weights are the negatives<br />
of the jumps in the Hodge filtrati<strong>on</strong> <strong>on</strong> the associated filtered module).<br />
By work of Faltings, it is known that ρ f,℘ | Gp is a potentially semistable representati<strong>on</strong><br />
with Hodge-Tate weights {0, k − 1}. Let D be the admissible filtered (ϕ, N, F, E)-module<br />
associated <strong>to</strong> this representati<strong>on</strong> for a suitable choices of F , and E = K f,℘ . By Theorem<br />
4.2.10, the study of the structure of the (p, p)-Galois representati<strong>on</strong> ρ f,℘ | Gp reduces<br />
<strong>to</strong> that of the study of the filtered module D. In particular, ρ f,℘ | Gp is reducible if and<br />
<strong>on</strong>ly if D has a n<strong>on</strong>-trivial admissible submodule.<br />
As menti<strong>on</strong>ed above, a key ingredient in our proof of Theorem 4.1.1 is that the representati<strong>on</strong><br />
ρ f,℘ lives in a strictly compatible system of Galois representati<strong>on</strong>s (ρ f,λ ), where<br />
λ varies over the primes of K f . This result is the culminati<strong>on</strong> of the work of several<br />
people, including Langlands, Deligne, Carayol, Katz-Messing, and most recently Sai<strong>to</strong><br />
[Sai97]. In particular, <strong>on</strong>e may read off the Weil-Deligne representati<strong>on</strong> WD(ρ f,℘ | Gp ) <strong>on</strong><br />
D (cf. Definiti<strong>on</strong> 4.2.12) by looking at the Weil-Deligne representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> ρ f,λ | Gp<br />
for a place λ of K f with λ ∤ p. As a c<strong>on</strong>sequence, <strong>on</strong>e may read off, e.g., the characteristic<br />
polynomial of crystalline Frobenius ϕ purely in terms of a λ-adic member of the strictly<br />
compatible family for λ ∤ p.<br />
Let π be the cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL 2 (A Q ) corresp<strong>on</strong>ding <strong>to</strong> f. It is<br />
well-known that f is ordinary at ℘ <strong>on</strong>ly if the underlying <strong>local</strong> au<strong>to</strong>morphic representati<strong>on</strong><br />
π p is in the principal series or (an unramified twist of) the Steinberg representati<strong>on</strong>. We<br />
obtain:<br />
Theorem 4.3.3. The characteristic polynomial P (X) of the (inverse of) crystalline<br />
Frobenius ϕ <strong>on</strong> D coincides with that of ρ f,λ (Frob p ) for a place λ of K f with λ ∤ p.<br />
More precisely, in the cases we need, P (X) is given by:<br />
• (Unramified principal series) If p ∤ N, then ρ f,℘ | Gp is crystalline and P (X) =<br />
X 2 − a p (f)X + χ(p)p k−1 .<br />
• (Steinberg) If p||N and v p (c<strong>on</strong>d(χ)) = 0, then ρ f,℘ | Gp is Q p -semistable and P (X) =<br />
(X − a p (f)) (X − pa p (f)).<br />
• (Ramified principal series) If p|N with v p (N) = v p (c<strong>on</strong>d(χ)), then ρ f,℘ | Gp is potentially<br />
crystalline and P (X) = (X − a p (f))(X − χ ′ (p)ā p (f)), where χ ′ denotes the<br />
prime-<strong>to</strong>-p part of the character χ.<br />
More generally, the complete rank 2 filtered (ϕ, N, F, E)-module D can be written down<br />
quite explicitly in all the above cases (cf. [Bre01], [GM09]). We now give a proof of<br />
Theorem 4.1.1 using this structure of D, which we first recall below in various cases.<br />
38
Good reducti<strong>on</strong>: p ∤ N<br />
In this case F = Q p and ρ f,℘ | Gp is crystalline (N = 0). Let D be the corresp<strong>on</strong>ding filtered<br />
(ϕ, N)-module. Let α and β be the two roots of P (X) = X 2 − a p (f)X + χ(p)p k−1 , with<br />
v p (α) = 0 and v p (β) = k − 1, since v p (a p (f)) = 0. The structure of the filtered module<br />
D is well-known (cf. [Bre01, p. 30-32], where the normalizati<strong>on</strong>s are a bit different).<br />
There are essentially two possibilities for D depending <strong>on</strong> whether D is decomposable or<br />
indecomposable. If D is decomposable, then D = Ee 1 ⊕ Ee 2 , and<br />
D =<br />
{<br />
ϕ(e1 ) = α −1 e 1 , ϕ(e 2 ) = β −1 e 2 ,<br />
Fil i (D F ) = 0 if i ≥ 1, Ee 1 if 2 − k ≤ i ≤ 0, D if i ≤ 1 − k.<br />
If D is indecomposable, then D = Ee 1 ⊕ Ee 2 , and<br />
D =<br />
{<br />
ϕ(e1 ) = α −1 e 1 + p 1−k e 2 , ϕ(e 2 ) = β −1 e 2 ,<br />
Fil i (D F ) = 0 if i ≥ 1, Ee 1 if 2 − k ≤ i ≤ 0, D if i ≤ 1 − k.<br />
Steinberg case: p ‖ N and p ∤ c<strong>on</strong>d(χ)<br />
In this case F = Q p , ρ f,℘ | Gp is semistable over Q p but n<strong>on</strong>-crystalline, and v p (a p (f)) =<br />
k−2<br />
2 . Note v p(a p (f)) = 0 if and <strong>on</strong>ly if k = 2. Let D be the corresp<strong>on</strong>ding filtered<br />
(ϕ, N)-module (cf. [GM09, §3.1]). Set α = a p (f) and β = pa p (f). Then D = Ee 1 ⊕ Ee 2 ,<br />
and<br />
⎧<br />
⎪⎨<br />
ϕ(e 1 ) = pβ −1 e 1 , ϕ(e 2 ) = β −1 e 2 ,<br />
D = N(e 1 ) = e 2 , N(e 2 ) = 0,<br />
⎪⎩<br />
Fil i (D F ) = 0 if i ≥ 1, E(e 1 − Le 2 ) if 2 − k ≤ i ≤ 0, D if i ≤ 1 − k,<br />
for some unique n<strong>on</strong>-zero L ∈ E.<br />
Ramified principal series: m = v p (c<strong>on</strong>d(χ)) = v p (N) ≥ 1<br />
In this case, ρ f,℘ | Gp becomes crystalline over the <strong>to</strong>tally ramified abelian extensi<strong>on</strong> F =<br />
Q p (µ p m) of Q p . Decompose χ = χ p χ ′ in<strong>to</strong> its p-part and prime-<strong>to</strong>-p part. Let D be the<br />
associated admissible filtered (ϕ, N, F, E)-module. An explicit descripti<strong>on</strong> of this module<br />
was given in [GM09, §3.2]. Set α = a p (f) and β = χ ′ (p)ā p (f). Then D = Ee 1 ⊕ Ee 2 , and<br />
D =<br />
{<br />
ϕ(e1 ) = α −1 e 1 , ϕ(e 2 ) = β −1 e 2 ,<br />
g(e 1 ) = e 1 , g(e 2 ) = χ p (g)e 2 ,<br />
for g ∈ Gal(F/Q p ). Since v p (a p (f)) = 0, the module D is either D ord-split or D ord-n<strong>on</strong>-split<br />
(cf. [GM09, §3.2]). The corresp<strong>on</strong>ding filtrati<strong>on</strong>s in these cases are given by<br />
Fil i (D F )<br />
= 0 if i ≥ 1, (F ⊗ E)e 1 if 2 − k ≤ i ≤ 0, D F if i ≤ 1 − k, and,<br />
Fil i (D F ) = 0 if i ≥ 1, (F ⊗ E)(xe 1 + ye 2 ) if 2 − k ≤ i ≤ 0, D F if i ≤ 1 − k,<br />
respectively, where x and y are explicit n<strong>on</strong>-zero quantities in F ⊗ E (cf. [GM09, §3.2]).<br />
39
4.3.1 Proof of Wiles’ theorem<br />
Let D n be the submodule of D generated by e n , for n = 1, 2. Since, in all cases, the ‘line’<br />
which determines the interesting step in the filtrati<strong>on</strong> <strong>on</strong> D F , is transverse <strong>to</strong> D 2,F , the<br />
induced filtrati<strong>on</strong> <strong>on</strong> D 2,F is given by<br />
· · · = Fil 2−k (D 2,F ) = 0 Fil 1−k (D 2,F ) = D 2,F (4.3.1)<br />
Thus t H (D 2 ) = 1 − k in all cases.<br />
In the principal series cases (i.e., the first and third cases above), we see that t N (D 2 ) =<br />
v p (β −1 ) = 1−k, so that D 2 is an admissible ϕ-submodule of D. In the Steinberg case, D 2 is<br />
the <strong>on</strong>ly (ϕ, N)-submodule of rank 1, since N(e 2 ) = 0. Moreover, t N (D 2 ) = v p (β −1 ) = − k 2<br />
which equals 1 − k = t H (D 2 ) if and <strong>on</strong>ly if k = 2. Thus D 2 is an admissible submodule<br />
if and <strong>on</strong>ly if k = 2, and D is irreducible if k > 2. Thus, in all (ordinary) cases, we have<br />
shown the existence of an admissible submodule D 2 of D, associated <strong>to</strong> ρ f,℘ | Gp , such that<br />
<strong>on</strong> D/D 2 , crystalline Frobenius ϕ acts by an explicit element α −1 of valuati<strong>on</strong> zero.<br />
Assume that we are in the first two cases, so that F = Q p . By Theorem 4.2.10, the<br />
representati<strong>on</strong> ρ f,℘ | Gp is clearly reducible, with a <strong>on</strong>e-dimensi<strong>on</strong>al submodule given by the<br />
character λ(β/p k−1 )χ k−1<br />
cyc,p, and quotient given by the unramified character λ(α) (see Corollary<br />
4.2.11), proving the theorem in these cases. In the last case (when F ≠ Q p ), again<br />
by Theorem 4.2.10, ρ f,℘ | Gp is reducible. Indeed, the module D 2 is a filtered (ϕ, N, F, E)-<br />
module with descent data given by the character χ p of Gal(F/Q p ). By Corollary 4.2.11,<br />
D 2 corresp<strong>on</strong>ds <strong>to</strong> the character ψ = χ p λ(β/p k−1 )χ k−1<br />
cyc,p of G p . Since D 2 ⊂ D as filtered<br />
modules with descent data, we see V (ψ) is a <strong>on</strong>e dimensi<strong>on</strong>al submodule of ρ f,℘ | Gp with<br />
unramified quotient given by λ(α), proving the theorem in this case as well.<br />
4.4 The case of GL n<br />
In the next chapter, we prove various generalizati<strong>on</strong>s of Theorem 4.1.1 for the <strong>local</strong> (p, p)-<br />
Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ).<br />
In this secti<strong>on</strong> we collect <strong>to</strong>gether some facts about such au<strong>to</strong>morphic representati<strong>on</strong>s<br />
and their Galois representati<strong>on</strong>s needed for the proof. The main results we need are the<br />
Local Langlands corresp<strong>on</strong>dence (now a theorem of Henniart [Hen00] and Harris-Taylor<br />
[HT01]), and the existence of a strictly compatible systems of Galois representati<strong>on</strong>s<br />
<strong>attached</strong> <strong>to</strong> cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (much progress has been made<br />
<strong>on</strong> this by Clozel, Harris, Kottwitz and Taylor [CHT08]).<br />
4.4.1 Local Langlands corresp<strong>on</strong>dence<br />
We state a few results c<strong>on</strong>cerning the Local Langlands corresp<strong>on</strong>dence. We follow Kudla’s<br />
article [Kud94], noting this article follows Rodier [Rod82], which in turn is based <strong>on</strong> the<br />
original work of Bernstein and Zelevinsky.<br />
Let F be a complete n<strong>on</strong>-Archimedean <strong>local</strong> field of residue characteristic p, let n ≥ 1,<br />
and let G = GL n (F ). For a partiti<strong>on</strong> n = n 1 +n 2 +· · ·+n r of n, let P be the corresp<strong>on</strong>ding<br />
parabolic subgroup of G, M the Levi subgroup of P , and N the unipotent radical of P . Let<br />
δ P denote the modulus character of the adjoint acti<strong>on</strong> of M <strong>on</strong> N. If σ = σ 1 ⊗σ 2 ⊗· · ·⊗σ r<br />
40
is a smooth representati<strong>on</strong> of M <strong>on</strong> V , we let<br />
I G P (σ) = {f : G → V | f smooth <strong>on</strong> G and f(nmg) = δ 1 2<br />
P<br />
(m)(σ(m)(f(g))},<br />
for n ∈ N, m ∈ M, g ∈ G. G acts <strong>on</strong> functi<strong>on</strong>s in IP G(σ) by right translati<strong>on</strong> and IG P<br />
(σ) is<br />
the usual induced representati<strong>on</strong> of σ. It is an admissible representati<strong>on</strong> of finite length.<br />
A result of Bernstein-Zelevinsky says that if all the σ i supercuspidal, and σ is irreducible,<br />
smooth and admissible, then IP G(σ) is reducible if and <strong>on</strong>ly if n i = n j and<br />
σ i = σ j (1) for some i ≠ j. For the partiti<strong>on</strong> n = m + m + · · · + m (r times), and for a<br />
supercuspidal representati<strong>on</strong> of σ of GL m (F ), call the data<br />
(σ, σ(1), . . . , σ(r − 1)) = [σ, σ(r − 1)] = ∆<br />
a segment. Clearly IP G (∆) is reducible. By [Kud94, Thm. 1.2.2], the induced representati<strong>on</strong><br />
IP G (∆) has a unique irreducible quotient Q(∆), which is essentially square-integrable.<br />
Two segments<br />
∆ 1 = [σ 1 , σ 1 (r 1 − 1)] and ∆ 2 = [σ 2 , σ 2 (r 2 − 1)]<br />
are said <strong>to</strong> be linked, if ∆ 1 ∆ 2 , ∆ 2 ∆ 1 , and ∆ 1 ∪ ∆ 2 is a segment. We say that ∆ 1<br />
precedes ∆ 2 if ∆ 1 and ∆ 2 are linked and if σ 2 = σ 1 (k), for some positive integer k.<br />
Theorem 4.4.1 (Langlands classificati<strong>on</strong>). Given segments ∆ 1 , . . . , ∆ r , assume that for<br />
i < j, ∆ i does not precede ∆ j . Then<br />
1. The induced representati<strong>on</strong> IP G(Q(∆ 1) ⊗ · · · ⊗ Q(∆ r )) admits a unique irreducible<br />
quotient Q(∆ 1 , . . . , ∆ r ), called the Langlands quotient. Moreover, r and the segments<br />
∆ i up <strong>to</strong> permutati<strong>on</strong> are uniquely determined by the Langlands quotient.<br />
2. Every irreducible admissible representati<strong>on</strong> of GL n (F ) is isomorphic <strong>to</strong> the Langlands<br />
quotient Q(∆ 1 , . . . , ∆ r ), for some ∆ i ’s.<br />
3. The induced representati<strong>on</strong> I G P (Q(∆ 1) ⊗ · · · ⊗ Q(∆ r )) is irreducible if and <strong>on</strong>ly if no<br />
two of the segments ∆ i and ∆ j are linked.<br />
We now turn <strong>to</strong> the Galois side.<br />
Representati<strong>on</strong>s of the Weil-Deligne group<br />
A representati<strong>on</strong> of W F is said <strong>to</strong> be Frobenius semisimple, if the arithmetic Frobenius<br />
acts semisimply. An admissible representati<strong>on</strong> of W<br />
F ′ , the Weil-Deligne group of F , is <strong>on</strong>e<br />
for which the acti<strong>on</strong> of W F is Frobenius semisimple. Let Sp(r) denote the Weil-Deligne<br />
representati<strong>on</strong> of order r with the usual definiti<strong>on</strong>. When F = Q p , there is a basis {f i }<br />
of Sp(r) for which ϕf i = p i−1 f i , and Nf i = f i−1 for i > 1 and Nf 1 = 0.<br />
Deligne classified all such admissible representati<strong>on</strong>s of W<br />
F ′ as the direct sum of the<br />
indecomposable representati<strong>on</strong>s τ m ⊗ Sp(r), where τ m is an irreducible admissible representati<strong>on</strong><br />
of dimensi<strong>on</strong> m of W F and r ≥ 1 (cf. [Roh94, §5, Cor. 2]).<br />
Theorem 4.4.2 (Local Langlands corresp<strong>on</strong>dence). ([HT01, VII.2.20], [Hen00], [Kut80]).<br />
There exists a bijecti<strong>on</strong> between isomorphism classes of irreducible admissible representati<strong>on</strong>s<br />
of GL n (F ) and isomorphism classes of admissible n-dimensi<strong>on</strong>al representati<strong>on</strong>s of<br />
W ′ F , the Weil-Deligne group of F . 41
The corresp<strong>on</strong>dence is given as follows. The key point is <strong>to</strong> c<strong>on</strong>struct a bijecti<strong>on</strong> Φ F<br />
between the set of isomorphism classes of supercuspidal representati<strong>on</strong>s of GL n (F ) and<br />
the set of isomorphism classes of irreducible admissible representati<strong>on</strong>s of W F . This is<br />
due <strong>to</strong> Henniart [Hen00] and Harris-Taylor [HT01]. Then, <strong>to</strong> Q(∆), for the segment ∆ =<br />
[σ, σ(r−1)], <strong>on</strong>e associates the indecomposable admissible representati<strong>on</strong> Φ F (σ)⊗Sp(r) of<br />
W<br />
F ′ . More generally, <strong>to</strong> the Langlands quotient Q(∆ 1, . . . , ∆ r ), where ∆ i = [σ i , σ i (r i −1)],<br />
for i = 1, . . . , k, <strong>on</strong>e associates the admissible representati<strong>on</strong> ⊕ k i=1 Φ F (σ i ) ⊗ Sp(r i ) of W<br />
F ′ .<br />
4.4.2 Au<strong>to</strong>morphic <strong>forms</strong> <strong>on</strong> GL n<br />
We now briefly recall what are (cuspidal) au<strong>to</strong>morphic <strong>forms</strong> <strong>on</strong> GL n (A Q ). More details<br />
can be found in [Tay04].<br />
The Harish-Chandra isomorphism identifies the center z n of the universal enveloping<br />
algebra of the complexified Lie algebra gl n of GL n , with the commutative C-algebra<br />
C[X 1 , X 2 , . . . , X n ] Sn , where the symmetric group S n acts by permuting the X i . Given a<br />
multiset H = {x 1 , x 2 , . . . , x n } of n complex numbers <strong>on</strong>e obtains an infinitesimal character<br />
of z n give by χ H : X i ↦→ x i .<br />
Cuspidal au<strong>to</strong>morphic <strong>forms</strong> with infinitesimal character χ H (or more simply just H)<br />
are smooth functi<strong>on</strong>s f : GL n (Q)\GL n (A Q ) → C satisfying the usual finiteness c<strong>on</strong>diti<strong>on</strong><br />
under a maximal compact subgroup, a cuspidality c<strong>on</strong>diti<strong>on</strong>, and a growth c<strong>on</strong>diti<strong>on</strong> for<br />
which we refer the reader <strong>to</strong> [Tay04]. In additi<strong>on</strong>, if z ∈ z n then z ·f = χ H (z)f. The space<br />
of such functi<strong>on</strong>s is denoted by A ◦ H (GL n(Q)\GL n (A Q )). This space is a direct sum of irreducible<br />
admissible GL n (A (∞)<br />
Q<br />
)×(gl n, O(n))-modules each occurring with multiplicity <strong>on</strong>e,<br />
and these irreducible c<strong>on</strong>stituents are referred <strong>to</strong> as cuspidal au<strong>to</strong>morphic representati<strong>on</strong><br />
of GL n (A Q ) with infinitesimal character χ H . Let π be such an au<strong>to</strong>morphic representati<strong>on</strong>.<br />
By a result of Flath, π is a restricted tensor product π = ⊗ ′ pπ p (cf. [Bum97, Thm.<br />
3.3.3]) of <strong>local</strong> au<strong>to</strong>morphic representati<strong>on</strong>s.<br />
4.4.3 Galois representati<strong>on</strong>s<br />
Let π be an au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ) with infinitesimal character χ H ,<br />
where H is a multiset of integers. The following very str<strong>on</strong>g, but natural, c<strong>on</strong>jecture<br />
seems <strong>to</strong> be part of the folklore.<br />
C<strong>on</strong>jecture 4.4.3. Let H c<strong>on</strong>sist of n distinct integers. There is a strictly compatible<br />
system of Galois representati<strong>on</strong>s (ρ π,l ) associated <strong>to</strong> π, with Hodge-Tate weights H, such<br />
that Local-Global compatibility holds.<br />
Here Local-Global compatibility means that the underlying semisimplified Weil -<br />
Deligne representati<strong>on</strong> at p in the compatible system (which is independent of the residue<br />
characteristic l of the coefficients by hypothesis) corresp<strong>on</strong>ds <strong>to</strong> π p via the Local Langlands<br />
corresp<strong>on</strong>dence. C<strong>on</strong>siderable evidence <strong>to</strong>wards this c<strong>on</strong>jecture is available for self-dual<br />
representati<strong>on</strong>s thanks <strong>to</strong> the work of Clozel, Kottwitz, Harris and Taylor. We quote the<br />
following theorem from Taylor’s paper [Tay04], referring <strong>to</strong> that paper for the original<br />
references (e.g., [Clo91]).<br />
Theorem 4.4.4 (cf. [Tay04], Thm. 3.6). Let H c<strong>on</strong>sist of n distinct integers. Suppose<br />
that the c<strong>on</strong>tragredient representati<strong>on</strong> π ∨ = π ⊗ ψ for some character ψ : Q × \A × Q → C× ,<br />
42
and suppose that for some prime q, the representati<strong>on</strong> π q is square-integrable. Then there<br />
is a c<strong>on</strong>tinuous representati<strong>on</strong><br />
ρ π,l : G Q → GL n ( ¯Q l )<br />
such that ρ π,l | Gl is potentially semistable with Hodge-Tate weights given by H, and such<br />
that for any prime p ≠ l, the semisimplificati<strong>on</strong> of the Weil-Deligne representati<strong>on</strong> <strong>attached</strong><br />
<strong>to</strong> ρ π,l | Gp is the same as the Weil-Deligne representati<strong>on</strong> associated by the Local<br />
Langlands corresp<strong>on</strong>dence <strong>to</strong> π p , except possibly for the m<strong>on</strong>odromy opera<strong>to</strong>r.<br />
Taylor and Yoshida [TY07] show that the two Weil-Deligne representati<strong>on</strong>s in the<br />
theorem above are in fact the same (i.e., the m<strong>on</strong>odromy opera<strong>to</strong>rs also match).<br />
In any case, from now <strong>on</strong>, we shall assume that C<strong>on</strong>jecture 4.4.3 holds. In particular,<br />
we assume that the Weil-Deligne representati<strong>on</strong> at p associated <strong>to</strong> a p-adic member of the<br />
compatible system of Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> π using F<strong>on</strong>taine theory is the<br />
same as the Weil-Deligne representati<strong>on</strong> at p <strong>attached</strong> <strong>to</strong> an l-adic member of the family,<br />
for l ≠ p.<br />
4.4.4 A variant, following [CHT08]<br />
A variant of the above result can be found in [CHT08]. We state this now using the<br />
notati<strong>on</strong> and terminology from [CHT08, §4.3].<br />
Say π is an RAESDC (regular, algebraic, essentially self dual, cuspidal) au<strong>to</strong>morphic<br />
representati<strong>on</strong> if π is a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> such that<br />
• π ∨ = π ⊗ χ for some character χ : Q × \A × Q → C× .<br />
• π ∞ has the same infinitesimal character as some irreducible algebraic representati<strong>on</strong><br />
of GL n .<br />
Let a ∈ Z n satisfy<br />
a 1 ≥ · · · ≥ a n . (4.4.1)<br />
Let Ξ a denote the irreducible algebraic representati<strong>on</strong> of GL n with highest weight a. We<br />
say that an RAESDC au<strong>to</strong>morphic representati<strong>on</strong> π has weight a, if π ∞ has the same<br />
infinitesimal character as Ξ ∨ a . In this case there is an integer w a such that a i +a n+1−i = w a<br />
for all i.<br />
Let S be a finite set of primes of Q. For v ∈ S, let ρ v be an irreducible square<br />
integrable representati<strong>on</strong> of GL n (Q v ). Say that an RAESDC representati<strong>on</strong> π has type<br />
{ρ v } v∈S , if for each v ∈ S, π v is an unramified twist of ρ ∨ v .<br />
With this setup, Clozel, Harris, and Taylor <strong>attached</strong> a Galois representati<strong>on</strong> <strong>to</strong> an<br />
RAESDC π.<br />
Theorem 4.4.5 ([CHT08], Prop. 4.3.1). Let ι : ¯Q l ≃ C. Let π be an RAESDC au<strong>to</strong>morphic<br />
representati<strong>on</strong> as above of weight a and type {ρ v } v∈S . Then there is a c<strong>on</strong>tinuous<br />
semisimple Galois representati<strong>on</strong><br />
with the following properties:<br />
r l,ι (π) : Gal( ¯Q/Q) → GL n ( ¯Q l )<br />
43
1. For every prime p ∤ l, we have<br />
r l,ι (π)| ss<br />
G p<br />
= (r l (ι −1 π p ) ∨ )(1 − n) ss ,<br />
where r l is the reciprocity map defined in [HT01].<br />
2. If l = p, then the restricti<strong>on</strong> r l,ι (π)| Gp is potentially semistable and if π p is unramified<br />
then it is crystalline, with Hodge-Tate weights −(a j + n − j) for j = 1, . . . , n.<br />
4.4.5 Newt<strong>on</strong> and Hodge filtrati<strong>on</strong><br />
Let ρ π,p | Gp be the (p, p)-Galois representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> an au<strong>to</strong>morphic representati<strong>on</strong><br />
π and D be the corresp<strong>on</strong>ding filtered (ϕ, N, F, E)-module (for suitable choices of F and<br />
E).<br />
Note that there are are two natural filtrati<strong>on</strong>s <strong>on</strong> D F , the Hodge filtrati<strong>on</strong> Fil i D F<br />
and the Newt<strong>on</strong> filtrati<strong>on</strong> defined by ordering the slopes of the crystalline Frobenius (the<br />
valuati<strong>on</strong>s of the roots of ϕ.) To keep the analysis of the structure of the (p, p)-Galois<br />
representati<strong>on</strong> ρ π,p | Gp within reas<strong>on</strong>able limits in this thesis, we shall make the following<br />
assumpti<strong>on</strong>.<br />
Assumpti<strong>on</strong> 4.4.6. The Newt<strong>on</strong> filtrati<strong>on</strong> <strong>on</strong> D F is in general positi<strong>on</strong> (or transverse)<br />
with respect <strong>to</strong> the Hodge filtrati<strong>on</strong> Fil i D F .<br />
Here, if V is a space and Fil i 1V and Fil j 2V are two filtrati<strong>on</strong>s <strong>on</strong> V then we say they<br />
are in general positi<strong>on</strong> if the each Fil i 1V is as transverse as possible <strong>to</strong> each Fil j 2 V . (cf.<br />
(4.3.1)). We remark that the c<strong>on</strong>diti<strong>on</strong> above is in some sense generic, since two random<br />
filtrati<strong>on</strong>s <strong>on</strong> a space tend <strong>to</strong> be in general positi<strong>on</strong>.<br />
We end this chapter by recalling the noti<strong>on</strong> of ordinary and quasi-ordinary for p-adic<br />
representati<strong>on</strong>s.<br />
4.4.6 Ordinary and quasi-ordinary representati<strong>on</strong>s<br />
As menti<strong>on</strong>ed earlier, our goal is <strong>to</strong> prove that the (p, p)-Galois representati<strong>on</strong> <strong>attached</strong> <strong>to</strong><br />
π is ‘upper-triangular’ in several cases. To this end it is c<strong>on</strong>venient <strong>to</strong> recall the following<br />
terminology (see, e.g., Greenberg [Gre94, p. 152] or Ochiai [Och01, Def. 3.1]).<br />
Definiti<strong>on</strong> 4.4.7. Let F be a number field. A p-adic representati<strong>on</strong> V of G F is called<br />
ordinary (respectively quasi-ordinary), if the following c<strong>on</strong>diti<strong>on</strong>s are satisfied:<br />
1. For each places v of F over p, there is a decreasing filtrati<strong>on</strong> of G Fv -modules<br />
· · · Fil i vV ⊇ Fil i+1<br />
v V ⊇ · · ·<br />
such that Fil i vV = V for i ≪ 0 and Fil i vV = 0 for i ≫ 0.<br />
2. For each v and each i, I v acts <strong>on</strong> Fil i vV/Fil i+1<br />
v V via the character χ i cyc,p, where χ cyc,p<br />
is the p-adic cyclo<strong>to</strong>mic character (respectively, there exists an open subgroup of I v<br />
which acts <strong>on</strong> Fil i vV/Fil i+1<br />
v V via χ i cyc,p).<br />
44
Chapter 5<br />
On Local Galois Representati<strong>on</strong>s<br />
<strong>attached</strong> <strong>to</strong> Au<strong>to</strong>morphic Forms<br />
This chapter c<strong>on</strong>tains the main results of the first part of the thesis. We prove structure<br />
theorems for the <strong>local</strong> Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> cuspidal au<strong>to</strong>morphic representati<strong>on</strong><br />
of GL n (A Q ).<br />
5.1 Introducti<strong>on</strong><br />
Recall that we have assumed that the global p-adic Galois representati<strong>on</strong> ρ π,p <strong>attached</strong><br />
<strong>to</strong> π exists, and it satisfies several natural properties, e.g., it lives in a strictly compatible<br />
system of Galois representati<strong>on</strong>s, and satisfies Local-Global compatibility. Recently, much<br />
progress has been made <strong>on</strong> this fr<strong>on</strong>t: such Galois representati<strong>on</strong>s have been <strong>attached</strong> <strong>to</strong><br />
what are referred <strong>to</strong> as RAESDC au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ) by Clozel,<br />
Kottwitz, Harris and Taylor, and for c<strong>on</strong>jugate self-dual au<strong>to</strong>morphic representati<strong>on</strong>s<br />
over CM fields these representati<strong>on</strong>s were shown <strong>to</strong> satisfy Local-Global compatibility by<br />
Taylor-Yoshida, away from p.<br />
Under some standard hypotheses (e.g., Assumpti<strong>on</strong> 4.4.6), we show that in several<br />
cases the corresp<strong>on</strong>ding <strong>local</strong> representati<strong>on</strong> ρ π,p | Gp has an ‘upper-triangular’ form, and<br />
completely determine the ‘diag<strong>on</strong>al’ characters. In other cases, and perhaps more interestingly,<br />
we give c<strong>on</strong>diti<strong>on</strong>s under which this <strong>local</strong> representati<strong>on</strong> is irreducible. For instance,<br />
we directly generalize the comment about irreducibility made just after Theorem 4.1.1.<br />
One of the main theorems is proved in §5.3 and §5.4, using methods from p-adic Hodge<br />
theory. It is well-known that the category of Weil-Deligne representati<strong>on</strong>s is equivalent<br />
<strong>to</strong> the category of (ϕ, N)-modules [BS07, Prop. 4.1]. In §5.4, we classify the (ϕ, N)-<br />
submodules of the (ϕ, N)-module associated <strong>to</strong> the indecomposable Weil-Deligne representati<strong>on</strong><br />
is the theorem. This classificati<strong>on</strong> plays a key role in analyzing the (p, p)-Galois<br />
representati<strong>on</strong> <strong>on</strong>ce the Hodge filtrati<strong>on</strong> is introduced. Al<strong>on</strong>g the way, we take a slight<br />
de<strong>to</strong>ur <strong>to</strong> write down explicitly the filtered (ϕ, N)-module <strong>attached</strong> <strong>to</strong> an m-dimensi<strong>on</strong>al<br />
‘unramified supercuspidal’ representati<strong>on</strong>, since this might be a useful additi<strong>on</strong> <strong>to</strong> the<br />
literature (cf. [GM09] for the two-dimensi<strong>on</strong>al case).<br />
Some results in the decomposable case (where the Weil-Deligne representati<strong>on</strong> is a<br />
direct sum of indecomposables) are given in §5.5, though the principal series case is<br />
45
treated completely a bit earlier, in §5.2 (in the spherical case our results overlap with<br />
those in D. Geraghty’s recent thesis).<br />
5.2 Principal series<br />
Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q ) with infinitesimal character<br />
H, for a set of distinct integers H. In this secti<strong>on</strong>, we study the behaviour of the (p, p)-<br />
Galois representati<strong>on</strong> assuming that π p , the <strong>local</strong> au<strong>to</strong>morphic representati<strong>on</strong>, is in the<br />
principal series.<br />
5.2.1 Spherical case<br />
Assume that π p is an unramified principal series representati<strong>on</strong>. Then there exist unramified<br />
characters χ 1 , . . . , χ n of Q × p such that π p is the Langland’s quotient Q(χ 1 , . . . , χ n ).<br />
We can parametrize the isomorphism class of this representati<strong>on</strong> by the Satake parameters<br />
α 1 , . . . , α n , for α i = χ i (ω), where ω is a uniformizer for Q p .<br />
Note that ρ π,p | Gp is crystalline with Hodge-Tate weights H. Let D be the corresp<strong>on</strong>ding<br />
filtered ϕ-module. Let the jumps in the filtrati<strong>on</strong> <strong>on</strong> D be β 1 < β 2 < · · · < β n (so<br />
that the Hodge-Tate weights H are −β 1 > · · · > −β n .<br />
Definiti<strong>on</strong> 5.2.1. Say that the au<strong>to</strong>morphic representati<strong>on</strong> π is p-ordinary, if β i +<br />
v p (α i ) = 0, for all i = 1, . . . , n.<br />
Theorem 5.2.2 (Spherical case). Let π be an au<strong>to</strong>morphic representati<strong>on</strong> of GL n (A Q )<br />
with infinitesimal character given by the integers −β 1 > · · · > −β n and such that π p is<br />
in the unramified principal series with Satake parameters α 1 , . . . , α n . If π is p-ordinary,<br />
then ρ π,p | Gp ∼<br />
⎛<br />
α<br />
λ( 1<br />
) · χ −β ⎞<br />
1<br />
p vp(α 1 ) cyc,p ∗ · · · ∗<br />
α 0 λ( 2<br />
) · χ −β 2<br />
p vp(α 2 ) cyc,p · · · ∗<br />
0 0 · · · ∗<br />
.<br />
⎜<br />
α<br />
⎝<br />
0 0 λ( n−1<br />
) · χ −β n−1<br />
p vp(α n−1 ) cyc,p<br />
∗ ⎟<br />
⎠<br />
α<br />
0 0 0 λ( n<br />
)) · χ −βn<br />
p vp(αn) cyc,p<br />
In particular, ρ π,p | Gp<br />
is ordinary.<br />
Proof. Since π p is p-ordinary, we have that v p (α n ) < v p (α n−1 ) < · · · < v p (α 1 ). By strict<br />
compatibility, the characteristic polynomial of the inverse of crystalline Frobenius of D n<br />
is equal <strong>to</strong> ∏ i (X − α i).<br />
Since the v p (α i ) are distinct, there exists a basis of eigenvec<strong>to</strong>rs of D n for the opera<strong>to</strong>r<br />
ϕ, say {e i }, with corresp<strong>on</strong>ding eigenvalues {αi<br />
−1 }. For any integer 1 ≤ i ≤ n, let D i be<br />
the ϕ-submodule generated by {e 1 , . . . , e i }. The filtrati<strong>on</strong> <strong>on</strong> D n is<br />
· · · ⊆ 0 Fil βn (D n ) ⊆ · · · Fil β 1<br />
(D n ) = D n ⊆ · · · (5.2.1)<br />
Since D n is admissible, we have that<br />
n∑<br />
β i = −<br />
i=1<br />
46<br />
n∑<br />
v p (α i ). (5.2.2)<br />
i=1
Clearly, we have that the jumps in the induced filtrati<strong>on</strong> <strong>on</strong> D n−1 are β 1 , . . . , β n−1 . By<br />
(5.2.2), we have<br />
n−1<br />
∑<br />
n−1<br />
∑<br />
t H (D n−1 ) = β i = − v p (α i ) = t N (D n−1 ), (5.2.3)<br />
i=1<br />
since β n = −v p (α n ) and hence the module D n−1 is admissible. Moreover, D n /D n−1 is<br />
also admissible, because t H (D n /D n−1 ) = β n = −v p (α n ) = t N (D n /D n−1 ), since ϕ acts <strong>on</strong><br />
D n /D n−1 by αn −1 . Therefore, the Galois representati<strong>on</strong> ρ π,p | Gp looks like<br />
( )<br />
ρn−1 ∗<br />
ρ ∼<br />
α<br />
0 λ( n<br />
,<br />
) · χ −βn<br />
p vp(αn) cyc,p<br />
where ρ n−1 is the (n − 1)-dimensi<strong>on</strong>al representati<strong>on</strong> of G p which corresp<strong>on</strong>ds <strong>to</strong> D n−1 .<br />
Repeating this argument successively for D n−1 , D n−2 , . . . D 1 , we obtain the theorem.<br />
Corollary 5.2.3. When n = 2, and π corresp<strong>on</strong>ds <strong>to</strong> f, we recover part (i) of Theorem<br />
4.1.1 (at least when k ≥ 3 is odd).<br />
Proof. Say 0 ≤ v p (α) ≤ v p (β) ≤ k − 1. It is well-known (cf. [Bum97]) that the Satake<br />
parameters at p satisfy the formulas α 1 = β · p 1−k<br />
2 and α 2 = α · p k−1<br />
2 . In fact L(s, π) =<br />
L(s + k−1<br />
2 , f) and ρ π,p = ρ f,℘ ⊗ χ 1−k<br />
2<br />
cyc,p, where (k − 1)/2 ∈ Z if k is odd. In particular, the<br />
Hodge-Tate weights of ρ π,p | Gp are −β 2 = 1−k<br />
2<br />
and −β 1 = k−1<br />
2<br />
, distinct integers if k ≥ 3 is<br />
odd. By the p-ordinariness c<strong>on</strong>diti<strong>on</strong> for π, we have that v p (α 2 ) = 1−k<br />
2<br />
and v p (α 1 ) = k−1<br />
2 .<br />
By the theorem above, we obtain<br />
⎛<br />
⎞<br />
ρ π,p | Gp ∼ ⎝ λ(α 1/p (k−1)/2 ) · χ k−1<br />
2<br />
cyc,p<br />
∗<br />
⎠ .<br />
0 λ(α 2 /p (1−k)/2 ) · χ 1−k<br />
2<br />
cyc,p<br />
Twisting both sides by χ k−1<br />
2<br />
cyc,p, we recover part (i) of Theorem 4.1.1.<br />
Variant, following [CHT08]<br />
Let π now be an RAESDC representati<strong>on</strong> of weight a as in §4.4.4 and let π p denote<br />
the <strong>local</strong> p-adic au<strong>to</strong>morphic representati<strong>on</strong> associated <strong>to</strong> π. For any i = 1, . . . , n, set<br />
β n+1−i ′ := a i + n − i, where a i ’s are as in (4.4.1). We have that β n ′ > β n−1 ′ > · · · > β′ 1 ,<br />
and the Hodge-Tate weights are −β n ′ < −β n−1 ′ < · · · < −β′ 1 .<br />
Assume that π p is in the unramified principal series, so π p = Q(χ 1 , χ 2 , . . . , χ n ), where<br />
χ i ’s are unramified characters of Q × p . Set α i ′ = χ i(ω)p n−1<br />
2 . Let t (j)<br />
p denote the eigenvalue<br />
of T p<br />
(j) <strong>on</strong> πp<br />
GLn(Zp) , where T p<br />
(j) is j-th Hecke opera<strong>to</strong>r as in [CHT08], and πp<br />
GLn(Zp) is<br />
spanned by GL n (Z p )-fixed vec<strong>to</strong>r, unique up <strong>to</strong> a c<strong>on</strong>stant. We would like <strong>to</strong> compute<br />
the right hand side in the display in part (1) of Theorem 4.4.5. By [CHT08, Cor. 3.1.2],<br />
in the spherical case, <strong>on</strong>e has<br />
(r l (ι −1 π p ) ∨ )(1 − n)(Frob −1<br />
p ) = ∏ i<br />
i=1<br />
(X − α ′ i)<br />
= X n − t (1)<br />
p X n−1 + · · · + (−1) j p j(j−1)<br />
2 t (j)<br />
p<br />
X n−j + · · · + (−1) n p n(n−1)<br />
2 t (n)<br />
p ,<br />
47
where Frob −1<br />
p is geometric Frobenius. Let s j denote the j-th elementary symmetric polynomial.<br />
Then from the equati<strong>on</strong> above, for any j = 1, · · · , n, we have p j(j−1)<br />
2 t (j)<br />
p =<br />
s j (α i ′) = p j(n−1)<br />
2 s j (χ i (p)) and hence t (j)<br />
p = s j (χ i (p))p j(n−j)<br />
2 . In this setting we make:<br />
Definiti<strong>on</strong> 5.2.4. Say that the au<strong>to</strong>morphic representati<strong>on</strong> π p is p-ordinary if β i ′ +<br />
v p (α i ′ ) = 0 for all i = 1, · · · , n.<br />
Note that by strict compatibility, crystalline Frobenius has characteristic polynomial<br />
exactly that above. The following theorem follows in a manner identical <strong>to</strong> that used <strong>to</strong><br />
prove Theorem 5.2.2.<br />
Theorem 5.2.5 (Spherical case, variant). Let π be a cuspidal au<strong>to</strong>morphic representati<strong>on</strong><br />
of GL n (A Q ) of weight a, as in §4.4.4. Let r p,ι (π) be the corresp<strong>on</strong>ding p-adic Galois<br />
representati<strong>on</strong>, with Hodge-Tate weights −β n+1−i ′ := a i + n − i, for i = 1, . . . , n. Suppose<br />
π p is in the principal series with Satake parameters α 1 , . . . , α n , and set α i ′ = α ip n−1<br />
2 . If<br />
π is p-ordinary, then r p,ι (π)| Gp ∼<br />
⎛<br />
α<br />
λ(<br />
′ 1<br />
) · ⎞<br />
p vp(α′ 1 ) χ−β′ 1<br />
cyc,p ∗ · · · ∗<br />
α 0 λ(<br />
′ 2<br />
) · p vp(α′ 2 ) χ−β′ 2<br />
cyc,p · · · ∗<br />
.<br />
⎜<br />
⎝ 0 0 · · · ∗ ⎟<br />
⎠<br />
α<br />
0 0 0 λ(<br />
′ n<br />
)) · p vp(α′ n ) χ−β′ n<br />
cyc,p<br />
In particular, r p,ι (π)| Gp<br />
is ordinary.<br />
The result above was also obtained by D. Geraghty in the course of proving modularity<br />
lifting theorems for GL n (see Lem. 2.7.7 and Cor. 2.7.8 of [Ger10]).<br />
5.2.2 Ramified principal series case<br />
Returning <strong>to</strong> the case where π is an au<strong>to</strong>morphic representati<strong>on</strong> with infinitesimal character<br />
H, we assume now that π p = Q(χ 1 , . . . , χ n ), where χ i are possibly ramified characters<br />
of Q × p .<br />
By the Local Langlands corresp<strong>on</strong>dence, we can think of the χ i as characters of W p .<br />
In particular, χ i | Ip have finite image. By strict compatibility, WD(ρ)| Ip ≃ ⊕ i χ i| Ip . The<br />
characters χ i | Ip fac<strong>to</strong>r through Gal(Q nr<br />
p (ζ p m)/Q nr<br />
p ) ≃ Gal(Q p (ζ p m)/Q p ) for some m ≥ 1.<br />
Set F := Q p (ζ p m). Observe that F is a finite abelian <strong>to</strong>tally ramified extensi<strong>on</strong> of Q p .<br />
Note that ρ π,p | GF is crystalline. Let D be the corresp<strong>on</strong>ding filtered module. Then<br />
D = Ee 1 + · · · + Ee n , where g ∈ Gal(F/Q p ) acts by χ i <strong>on</strong> e i . A small computati<strong>on</strong> shows<br />
that ϕ(e i ) = αi −1 e i , where α i = χ i (ω F ), for ω F a uniformizer of F . Using Corollary 4.2.11,<br />
and following the proof of Theorem 5.2.2, we obtain:<br />
Theorem 5.2.6 (Ramified principal series). Say π p = Q(χ 1 , . . . , χ n ) is in the ramified<br />
principal series. If π is p-ordinary, then ρ π,p | Gp ∼<br />
⎛<br />
α<br />
χ 1 · λ( 1<br />
) · χ −β ⎞<br />
1<br />
p vp(α 1 ) cyc,p ∗ · · · ∗<br />
α<br />
0 χ 2 · λ( 2<br />
) · χ −β 2<br />
p<br />
⎜<br />
vp(α 2 ) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ .<br />
α<br />
0 0 0 χ n · λ( n<br />
)) · χ −βn<br />
p vp(αn) cyc,p<br />
48
In particular, ρ π,p | Gp is quasi-ordinary.<br />
5.3 Steinberg case<br />
In this secti<strong>on</strong> we treat the case where the Weil-Deligne representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> π p is<br />
a twist of the special representati<strong>on</strong> Sp(n).<br />
5.3.1 Unramified twist of Steinberg<br />
We start with the case where the Weil-Deligne representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> π p is of the<br />
form χ ⊗ Sp(n), where χ is an unramified character.<br />
Let D be the filtered (ϕ, N, Q p , E)-module <strong>attached</strong> <strong>to</strong> ρ π,p | Gp . Thus D is a vec<strong>to</strong>r<br />
space over E. Note N n = 0 and N n−1 ≠ 0 so that there is a basis {f i } of D with<br />
f i−1 := Nf i , for 1 < i ≤ n, and Nf 1 = 0. We set α := χ(Frob p ). Since Nϕ = pϕN,<br />
we may assume that ϕ(f i ) = αi<br />
−1 f i , ∀i, where αi<br />
−1 = pi−1<br />
α<br />
. When α = 1, D = Sp(n) as<br />
menti<strong>on</strong>ed in §4.4.1.<br />
For each 1 ≤ i ≤ n, let D i denote the subspace 〈f i , . . . , f 1 〉. Clearly, dim(D i ) = i and<br />
D 1 D 2 · · · D n = D. We have:<br />
Lemma 5.3.1. For every integer 1 ≤ r ≤ n, there is a unique N-submodule of D of rank<br />
r, namely D r .<br />
Proof. Let D ′ be a N-submodule of D of rank r. Say the order of nilpotency of N <strong>on</strong><br />
D ′ is i. Thus, D ′ ⊆ Ker(N i ). Since Ker(N i ) = 〈f i , . . . , f 1 〉, dim(Ker(N i )) = i and hence<br />
r ≤ i. Clearly, the order of nilpotency of N <strong>on</strong> D ′ is less than or equal <strong>to</strong> r. Hence i = r<br />
and D ′ = Ker(N r ) = D r .<br />
Let β n > · · · > β 1 be the jumps in the Hodge filtrati<strong>on</strong> <strong>on</strong> D. We assume that the<br />
Hodge filtrati<strong>on</strong> is in general positi<strong>on</strong> with respect <strong>to</strong> the Newt<strong>on</strong> filtrati<strong>on</strong> given by the<br />
D i (cf. Assumpti<strong>on</strong> 4.4.6). An example of such a filtrati<strong>on</strong> is<br />
〈f n 〉 〈f n , f n−1 〉 · · · 〈f n , . . . , f 2 〉 〈f n , . . . , f 1 〉.<br />
The following elementary lemma plays an important role in later proofs.<br />
Lemma 5.3.2. Let m be a natural number. Let {a i } n i=1 be an increasing (resp., decreasing)<br />
sequence of integers such that |a i+1 −a i | = m. Let {b i } n i=1 be another increasing (resp.,<br />
decreasing) sequence of integers, such that |b i+1 − b i | ≥ m. Assume that ∑ i a i = ∑ i b i.<br />
If a n = b n or a 1 = b 1 , then a i = b i , ∀ i.<br />
Proof. Let us prove the lemma when a n = b n and a i are increasing. The proof in the<br />
other cases is similar. We have:<br />
m(n − 1 + n − 2 + · · · + 1) ≤<br />
n∑<br />
(b n − b i ) =<br />
i=1<br />
n∑<br />
(a n − a i ) = m(n − 1 + n − 2 + · · · + 1)<br />
The first equality in the above expressi<strong>on</strong> follows from a n = b n . From the above equati<strong>on</strong>,<br />
we see that b n − b i = a n − a i , for every 1 ≤ i ≤ n. Since a n = b n , we have that a i = b i ,<br />
for every 1 ≤ i ≤ n.<br />
i=1<br />
49
By Lemma 5.3.1, the D i are the <strong>on</strong>ly (ϕ, N)-submodules of D. The following propositi<strong>on</strong><br />
shows that if two ‘c<strong>on</strong>secutive’ submodules D i and D i+1 are admissible then all the<br />
D i are admissible.<br />
Propositi<strong>on</strong> 5.3.3. Suppose there exists an integer 1 ≤ i ≤ n such that both D i and<br />
D i+1 are admissible. Then each D r , for 1 ≤ r ≤ n, is admissible. Moreover, the β i are<br />
c<strong>on</strong>secutive integers.<br />
Proof. Since D i and D i+1 are admissible, we have the following equalities:<br />
i∑<br />
β 1 + β 2 + · · · + β i = − v p (α r ),<br />
r=1<br />
i+1<br />
∑<br />
β 1 + β 2 + · · · + β i+1 = − v p (α r ).<br />
r=1<br />
(5.3.1)<br />
Define a r = −v p (α r ) and b r = β r , for 1 ≤ r ≤ n. Hence, we have<br />
a n > · · · > a i+2 > a i+1 > a i > · · · > a 1 ,<br />
b n > · · · > b i+2 > b i+1 > b i > · · · > b 1 .<br />
(5.3.2)<br />
By (5.3.1), we have a i+1 = −v p (α i+1 ) = β i+1 = b i+1 . By Lemma 5.3.2 and by (5.3.1), we<br />
have that a r = b r , for all 1 ≤ r ≤ i + 1. Since D n is admissible, we have<br />
t H (D n ) =<br />
n∑<br />
β r = −<br />
r=1<br />
From (5.3.1) and (5.3.3), we have that<br />
n∑<br />
r=i+1<br />
β r = −<br />
n∑<br />
v p (α r ) = t N (D n ). (5.3.3)<br />
r=1<br />
n∑<br />
r=i+1<br />
v p (α r ).<br />
Again, by (5.3.2) and Lemma 5.3.2, we have that a r = b r , for all i + 1 ≤ r ≤ n. Hence<br />
β r = −v p (α r ), for all 1 ≤ r ≤ n. This shows that all other D i ’s are admissible. Also, the<br />
β i are c<strong>on</strong>secutive integers since the v p (α i ) are c<strong>on</strong>secutive integers.<br />
Corollary 5.3.4. Keeping the notati<strong>on</strong> as above, the admissibility of D 1 or D n−1 implies<br />
the admissibility of all other D i .<br />
Theorem 5.3.5. Assume that the Hodge filtrati<strong>on</strong> <strong>on</strong> D is in general positi<strong>on</strong> with respect<br />
<strong>to</strong> the D i (cf. Assumpti<strong>on</strong> 4.4.6). Then the crystal D is either irreducible or reducible,<br />
in which case each D i , for 1 ≤ i ≤ n is admissible.<br />
Proof. If D is irreducible, then we are d<strong>on</strong>e. If not, there exists an i, such that D i<br />
is admissible. If D i−1 or D i+1 is admissible, then by Propositi<strong>on</strong> 5.3.3, all the D r are<br />
admissible. So, it is enough <strong>to</strong> c<strong>on</strong>sider the case where neither D i−1 nor D i+1 is admissible<br />
50
(and D i is admissible). We have:<br />
r=i−1<br />
∑<br />
β 1 + β 2 + · · · + β i−1 < − v p (α r ),<br />
r=1<br />
r=i<br />
∑<br />
β 1 + β 2 + · · · + β i = − v p (α r ),<br />
r=1<br />
r=i+1<br />
∑<br />
β 1 + β 2 + · · · + β i+1 < − v p (α r ).<br />
r=1<br />
(5.3.4a)<br />
(5.3.4b)<br />
(5.3.4c)<br />
Subtracting (5.3.4b) from (5.3.4a), we get −β i < v p (α i ). Subtracting (5.3.4b) from<br />
(5.3.4c), we get β i+1 < −v p (α i+1 ) = −v p (α i ) + 1. Adding these inequalities, we obtain<br />
β i+1 − β i < 1. But this is a c<strong>on</strong>tradicti<strong>on</strong>, since β i+1 > β i . This proves the theorem.<br />
Definiti<strong>on</strong> 5.3.6. Say π is p-ordinary, if β 1 + v p (α) = 0.<br />
Note that if π is p-ordinary, then D 1 is admissible, so the flag D 1 ⊂ D 2 ⊂ · · · ⊂ D n<br />
is an admissible flag by Theorem 5.3.5 (an easy check shows that if π is p-ordinary then<br />
Assumpti<strong>on</strong> 4.4.6 holds au<strong>to</strong>matically). Applying the above discussi<strong>on</strong> <strong>to</strong> the <strong>local</strong> Galois<br />
representati<strong>on</strong> ρ π,p | Gp , we obtain:<br />
Theorem 5.3.7 (Unramified twist of Steinberg). Say π is a cuspidal au<strong>to</strong>morphic representati<strong>on</strong><br />
with infinitesimal character given by the integers −β 1 > · · · > −β n . Suppose<br />
that π p is an unramified twist of the Steinberg representati<strong>on</strong>, i.e., WD(ρ π,p | Gp ) ∼<br />
χ ⊗ Sp(n), where χ is the unramified character mapping arithmetic Frobenius <strong>to</strong> α. If π<br />
is ordinary at p (i.e., v p (α) = −β 1 ), then the β i are necessarily c<strong>on</strong>secutive integers and<br />
ρ π,p | Gp ∼<br />
⎛<br />
α<br />
λ( ) · χ −β ⎞<br />
1<br />
p vp(α) cyc,p ∗ · · · ∗<br />
α<br />
0 λ( ) · χ −β 1−1<br />
p<br />
⎜<br />
vp(α) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ ,<br />
α<br />
0 0 0 λ( ) · χ −β 1−(n−1)<br />
p vp(α) cyc,p<br />
α<br />
where λ( ) is an unramified character taking arithmetic Frobenius <strong>to</strong> , and in<br />
p vp(α) p vp(α)<br />
particular, ρ π,p | Gp is ordinary. If π is not p-ordinary, and Assumpti<strong>on</strong> 4.4.6 holds, then<br />
ρ π,p | Gp is irreducible.<br />
Proof. By strict compatibility, D is the filtered (ϕ, N, Q p , E)-module <strong>attached</strong> <strong>to</strong> ρ π,p | Gp .<br />
If π is p-ordinary, then we are d<strong>on</strong>e and the characters <strong>on</strong> the diag<strong>on</strong>al are determined<br />
by Corollary 4.2.11.<br />
If π is not p-ordinary, then D is irreducible. Indeed, if D is reducible, then by Theorem<br />
5.3.5, all D i , and in particular D 1 , are admissible, so π is p-ordinary.<br />
5.3.2 Ramified twist of Steinberg<br />
Theorem 5.3.8 (Ramified twist of Steinberg). Let the notati<strong>on</strong> and hypotheses be as in<br />
Theorem 5.3.7, except that this time assume that WD(ρ π,p | Gp ) ∼ χ⊗Sp(n), where χ is an<br />
α<br />
51
arbitrary, possibly ramified, character. Write χ = χ 0 · χ ′ where χ 0 is the ramified part of<br />
χ, and χ ′ is an unramified character taking arithmetic Frobenius <strong>to</strong> α. If π is p-ordinary<br />
(β 1 = −v p (α)), then the β i are c<strong>on</strong>secutive integers and ρ π,p | Gp ∼<br />
⎛<br />
α<br />
χ 0 · λ( ) · χ −β ⎞<br />
1<br />
p vp(α) cyc,p ∗ · · · ∗<br />
α<br />
0 χ 0 · λ( ) · χ −β 1−1<br />
p<br />
⎜<br />
vp(α) cyc,p · · · ∗<br />
⎝ 0 0 · · · ∗<br />
⎟<br />
⎠ .<br />
α<br />
0 0 0 χ 0 · λ( ) · χ −β 1−(n−1)<br />
p vp(α) cyc,p<br />
If π is not p-ordinary, and Assumpti<strong>on</strong> 4.4.6 holds, then ρ π,p | Gp<br />
is irreducible.<br />
Proof. Let F be a <strong>to</strong>tally ramified abelian (cyclo<strong>to</strong>mic) extensi<strong>on</strong> of Q p such that χ 0 | IF =<br />
1. Then the reducibility of ρ π,p | GF over F can be shown exactly as in Theorem 5.3.7,<br />
and the theorem over Q p follows using the descent data of the underlying filtered module.<br />
If π is not p-ordinary, then by arguments similar those used in proving Theorem 5.3.7,<br />
ρ π,p | GF is irreducible, so that ρ π,p | Gp is also irreducible.<br />
5.4 Supercuspidal ⊗ Steinberg<br />
We now turn <strong>to</strong> the case where the Weil-Deligne representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> π p is indecomposable.<br />
We assume that WD(ρ π,p | Gp ) is Frobenius semisimple. Thus, it is of the<br />
form τ ⊗ Sp(n), where τ is an irreducible m-dimensi<strong>on</strong>al representati<strong>on</strong> corresp<strong>on</strong>ding <strong>to</strong><br />
a supercuspidal representati<strong>on</strong> of GL m , for m ≥ 1.<br />
We classify the (ϕ, N, F, E)-submodules of D, the crystal <strong>attached</strong> <strong>to</strong> the <strong>local</strong> representati<strong>on</strong><br />
ρ π,p | Gp , where WD(ρ π,p | Gp ) = τ⊗Sp(n), for m ≥ 1 and n ≥ 1. This classificati<strong>on</strong><br />
will be used later <strong>to</strong> study the structure of ρ π,p | Gp , taking the filtrati<strong>on</strong> <strong>on</strong> D in<strong>to</strong> account.<br />
Recall that there is a equivalence of categories between (ϕ, N)-modules with coefficients<br />
and descent data, and Weil-Deligne representati<strong>on</strong>s [BS07, Prop. 4.1]. Write D τ ,<br />
respectively D Sp(n) , for the (ϕ, N)-modules corresp<strong>on</strong>ding <strong>to</strong> τ, respectively Sp(n), etc.<br />
The main result of the first few subsecti<strong>on</strong>s is:<br />
Theorem 5.4.1. All the (ϕ, N, F, E)-submodules of D = D τ ⊗ D Sp(n) are of the form<br />
D τ ⊗ D Sp(r) , for some 1 ≤ r ≤ n.<br />
We prove the theorem in stages. However, the reader may turn straight <strong>to</strong> the §5.4.3,<br />
where the general case is treated. Note that the case m = 1 was treated in the previous<br />
secti<strong>on</strong> (twist of Steinberg), and the case n = 1 is vacuously true. The next simplest case<br />
is when m = 2 and n = 2, and we start with this case in the next subsecti<strong>on</strong>.<br />
The following lemma will be useful in our analysis throughout.<br />
Lemma 5.4.2. The theory of Jordan can<strong>on</strong>ical <strong>forms</strong> can be extended <strong>to</strong> nilpotent opera<strong>to</strong>rs<br />
<strong>on</strong> free of finite rank (F 0 ⊗ E)-modules, and we call the number of blocks in the<br />
Jordan decompositi<strong>on</strong> of the m<strong>on</strong>odromy opera<strong>to</strong>r N as the ‘index’ of N.<br />
Proof. One simply extends the usual theory of Jordan can<strong>on</strong>ical <strong>forms</strong> <strong>on</strong> each projecti<strong>on</strong><br />
under (4.2.2) <strong>to</strong> modules over F 0 ⊗ E-modules.<br />
52
5.4.1 m = 2 and n = 2<br />
We start with the case when τ corresp<strong>on</strong>ds <strong>to</strong> a supercuspidal representati<strong>on</strong> of GL 2 (Q p )<br />
whose Weil-Deligne representati<strong>on</strong> is the 2-dimensi<strong>on</strong>al representati<strong>on</strong> obtained by inducing<br />
a character from a quadratic extensi<strong>on</strong> K of Q p , and n = 2. The methods used in<br />
this case will be used <strong>to</strong> deal with more general cases in subsequent subsecti<strong>on</strong>s.<br />
We also assume, for simplicity, that the quadratic extensi<strong>on</strong> of Q p menti<strong>on</strong>ed above is<br />
K = Q p 2, the unramified quadratic extensi<strong>on</strong> of Q p . By abusing language a bit we shall<br />
say that τ corresp<strong>on</strong>ds <strong>to</strong> an unramified supercuspidal representati<strong>on</strong>. The argument in<br />
the ramified supercuspidal case (i.e., K/Q p ramified quadratic) was also worked out in<br />
detail, but is excluded here for the sake of brevity.<br />
Thus we assume that there is a character χ of W p 2, the Weil group of Q p 2, which does<br />
not extend <strong>to</strong> W p (i.e., χ ≠ χ σ <strong>on</strong> W p 2, where 1 ≠ σ ∈ Gal(Q p 2/Q p )) such that<br />
τ| Ip ≃ Ind Wp<br />
W p 2 χ| I p<br />
≃ χ| Ip ⊕ χ σ | Ip .<br />
The corresp<strong>on</strong>ding (ϕ, N)-module in this case can be written down quite explicitly (cf.<br />
[GM09, §3.3]). Briefly there is an abelian field F (depending <strong>on</strong> χ), generated by σ (a<br />
lift <strong>to</strong> F of the au<strong>to</strong>morphism σ above) and elements g of the inertia group I(F/K) of<br />
Gal(F/K), and an element t ∈ E such that D τ = 〈e 1 , e 2 〉, with N = 0, and satisfies:<br />
D τ = D unr−sc [a : b] =<br />
⎧<br />
⎪⎨<br />
⎪⎩<br />
ϕ(e 1 ) = 1 √<br />
t<br />
e 1 , ϕ(e 2 ) = 1 √<br />
t<br />
e 2 ,<br />
σ(e 1 ) = e 2 , σ(e 2 ) = e 1 ,<br />
g(e 1 ) = (1 ⊗ χ(g)) e 1 ,<br />
g(e 2 ) = (1 ⊗ χ σ (g)) e 2 .<br />
We do not write down the Hodge filtrati<strong>on</strong> since we do not need it here.<br />
For the sec<strong>on</strong>d fac<strong>to</strong>r, we recall that the module D Sp(2) has a basis f 1 , f 2 with properties<br />
described at the start of §5.3.1. Then D = D τ ⊗ D Sp(2) has a basis of 4 vec<strong>to</strong>rs<br />
{v i } 4 i=1 defined as follows v 1 = e 1 ⊗ f 2 v 2 = e 1 ⊗ f 1<br />
v 3 = e 2 ⊗ f 2 v 4 = e 2 ⊗ f 1<br />
Observe that <strong>on</strong> D, the m<strong>on</strong>odromy opera<strong>to</strong>r N = 1 ⊗ N + N ⊗ 1 = 1 ⊗ N has kernel of<br />
rank 2, and this space is generated by the vec<strong>to</strong>rs v 2 and v 4 .<br />
For notati<strong>on</strong>al simplicity, write z for 1 ⊗ z, if z ∈ E and note that for z = ∑ i z i ⊗ e i ∈<br />
F 0 ⊗ E, σ(z) = ∑ i σ(z i) ⊗ e i . Then the induced acti<strong>on</strong> <strong>on</strong> D, the tensor product of the<br />
two (ϕ, N)-modules D τ and D Sp(2) , is summarized in the following table:<br />
ϕ<br />
v 1 = e 1 ⊗ f 2 v 2 = e 1 ⊗ f 1 v 3 = e 2 ⊗ f 2 v 4 = e 2 ⊗ f 1<br />
√<br />
p<br />
t<br />
v 1 √t 1<br />
v 2 √t<br />
p<br />
v 3 √t 1<br />
v 4<br />
N v 2 0 v 4 0<br />
σ v 3 v 4 v 1 v 2<br />
g χ(g)v 1 χ(g)v 2 χ σ (g)v 3 χ σ (g)v 4<br />
Lemma 5.4.3. There are no rank 1 (ϕ, N, F, E)-submodules of D.<br />
53
Proof. Let 〈v〉 be a free module of rank-1 admissible (ϕ, N, F, E)-submodule of D. Write<br />
v = av 1 + bv 2 + cv 3 + dv 4 where a, b, c, d ∈ (F 0 ⊗ Qp E). Since the rank of 〈v〉 is 1, we have<br />
Nv is zero. From the above table, it is easy <strong>to</strong> see that v has <strong>to</strong> be equal <strong>to</strong> bv 2 + dv 4 , for<br />
some b, d. Assume that ϕ(v) = α ϕ v. Since ϕ is bijective, α ϕ is a unit. We compute:<br />
α ϕ (bv 2 + dv 4 ) = α ϕ v (1)<br />
= ϕv (2)<br />
= σ(b) 1 √<br />
t<br />
v 2 + σ(d) 1 √<br />
t<br />
v 4<br />
(equality (1) follows from the fact that 〈v〉 is ϕ-stable and equality (2) from the table).<br />
By comparing coefficients, we have<br />
α ϕ<br />
√<br />
tb = σ(b), αϕ<br />
√<br />
td = σ(d). (5.4.1)<br />
Lemma 5.4.4. Suppose x is a n<strong>on</strong>-zero and a n<strong>on</strong>-unit element of F 0 ⊗ E. Then σ(x) ≠<br />
cx, for every c ∈ F 0 ⊗ E.<br />
From the above relati<strong>on</strong>s for b (resp., d) and by lemma above, we c<strong>on</strong>clude that b<br />
(resp., d) is either zero or a unit.<br />
Suppose that σ(v) = c σ v, for some c σ ∈ (F 0 ⊗ E), with c 2m<br />
σ = 1 (this m is the m of<br />
[GM09, §3.3], and not the m of this paper which is presently 2). Then<br />
c σ (bv 2 + dv 4 ) = c σ v (1)<br />
= σv (2)<br />
= σ(b)v 4 + σ(d)v 2<br />
(equality (1) follows from the fact that 〈v〉 is σ-stable and equality (2) from the table).<br />
By comparing coefficients, we have<br />
c σ b = σ(d), c σ d = σ(b). (5.4.2)<br />
From these relati<strong>on</strong>s, we can c<strong>on</strong>clude that b is zero if and <strong>on</strong>ly if d is zero. Since v is<br />
n<strong>on</strong>-zero, we have that both b and d are both units.<br />
For every g ∈ I(F/K), we have<br />
c g (bv 2 + dv 4 ) = c g v (1)<br />
= g · v (2)<br />
= χ(g)bv 2 + χ σ (g)dv 4<br />
(equality (1) follows from the fact that 〈v〉 is I(F/K)-stable and equality (2) from the<br />
table). Now, by comparing coefficients, and using the fact that b and d are units, we have<br />
c g = χ(g),<br />
c g = χ σ (g).<br />
Hence χ(g) = χ σ (g), for every g ∈ I(F/K). This is a c<strong>on</strong>tradicti<strong>on</strong>.<br />
Remark 5.4.5. The above argument shows that there are no (ϕ, Gal(F/Q p ))-stable submodules<br />
of rank 1 of either 〈v 2 , v 4 〉 or, as <strong>on</strong>e can show similarly, 〈v 1 , v 3 〉. There is also a<br />
simpler proof of this Lemma which avoids working with the above explicit manipulati<strong>on</strong>s<br />
which works for general m and n (cf. the proof of Lemma 5.4.11).<br />
Lemma 5.4.6. The <strong>on</strong>ly rank 2 (ϕ, N, F, E)-submodule of D is 〈v 2 , v 4 〉 = D τ ⊗ D Sp(1) .<br />
54
Proof. Let D ′ be a filtered module of rank 2. Suppose the index of N = 1, i.e., D ′ =<br />
〈w 1 , N(w 1 )〉, such that N 2 (w 1 ) = 0. It is easy <strong>to</strong> see that 〈N(w 1 )〉 is a rank 1 (ϕ, N, F, E)-<br />
submodule of D, a c<strong>on</strong>tradicti<strong>on</strong> by Lemma 5.4.3. Therefore, the index of N ≠ 1.<br />
Suppose the index of N = 2, i.e., D ′ = 〈w 1 , w 2 〉 such that N(w 1 ) = N(w 2 ) = 0. We<br />
know that the kernel of N is 〈v 2 , v 4 〉. Therefore, we have that 〈w 1 , w 2 〉 ⊆ 〈v 2 , v 4 〉. But<br />
both are free modules of the same rank, hence the equality holds.<br />
Lemma 5.4.7. There are no rank 3 (ϕ, N, F, E)-submodules of D.<br />
Proof. Let D ′ be a filtered module of D of rank 3. The index of nilpotency of N cannot<br />
be 3, since the kernel of N has rank 2.<br />
Suppose the index of nilpotency of N is 2, that is, there exists a basis, say w 1 , w 2 ,<br />
and N(w 2 ), of D ′ such that N(w 1 ), N 2 (w 2 ) are both zero. Clearly 〈w 1 , N(w 2 )〉 = 〈v 2 , v 4 〉.<br />
Thus there is a vec<strong>to</strong>r w 2 ′ = a 1v 1 + a 3 v 3 such that D ′ = 〈w 2 ′ , v 2, v 4 〉. But 〈w 2 ′ 〉 is stable<br />
by ϕ. Indeed ϕ acts by a scalar <strong>on</strong> w 2 ′ since, a priori, ϕw′ 2 is a linear combinati<strong>on</strong> of<br />
w 2 ′ , v 2 and v 4 , but the v 2 and v 4 comp<strong>on</strong>ents do not appear since ϕ preserves the space<br />
〈v 1 , v 3 〉. Similarly 〈w 2 ′ 〉 is also Gal(F/Q p)-stable. But then 〈w 2 ′ 〉 ⊂ 〈v 1, v 3 〉 is a rank 1<br />
module stable by ϕ and Gal(F/Q p ), which is not possible (cf. Remark 5.4.5).<br />
Finally, suppose the index of nilpotency of N is 1. But this case does not arise since<br />
the index of nilpotency of N <strong>on</strong> D is 2.<br />
Proof of Theorem 5.4.1 when m = 2 and n = 2<br />
This follows immediately from Lemmas 5.4.3, 5.4.6, and 5.4.7, when τ is an unramified<br />
supercuspidal representati<strong>on</strong> of dimensi<strong>on</strong> m = 2, and n = 2. As we remarked earlier, the<br />
case when τ is a ramified supercuspidal representati<strong>on</strong> is proved in a very similar manner<br />
using the notati<strong>on</strong> in [GM09, §3.4] (the <strong>on</strong>ly difference in the computati<strong>on</strong>s are the role<br />
of σ is now played by the au<strong>to</strong>morphism ι there).<br />
We menti<strong>on</strong> some immediate corollaries of Theorem 5.4.1 in the present case. Let<br />
π be an au<strong>to</strong>morphic representati<strong>on</strong> of GL 4 (A Q ) with infinitesimal character c<strong>on</strong>sisting<br />
of distinct integers −β 4 < · · · < −β 1 . Let ρ = ρ π,p | Gp be the corresp<strong>on</strong>ding (p, p)-<br />
Galois representati<strong>on</strong>. Suppose that WD(ρ) ∼ τ ⊗ Sp(2), where τ is a supercuspidal<br />
representati<strong>on</strong> of dimensi<strong>on</strong> 2, as above. Let D = D(ρ) be the corresp<strong>on</strong>ding admissible<br />
filtered (ϕ, N, F, E)-module. Note that β 4 > β 3 > β 2 > β 1 are also the drops in the Hodge<br />
filtrati<strong>on</strong> <strong>on</strong> D F .<br />
Corollary 5.4.8. With notati<strong>on</strong> as above, the crystal D is irreducible if and <strong>on</strong>ly if<br />
〈v 2 , v 4 〉 is not an admissible (ϕ, N, F, E)-submodule of D.<br />
Proof. By Lemmas 5.4.3 and 5.4.7, there are no rank 1 and rank 3 filtered submodules<br />
of D. By Lemma 5.4.6, there exists a unique rank 2 (ϕ, N, F, E)-submodule of D, which<br />
is 〈v 2 , v 4 〉. Hence the admissibility of 〈v 2 , v 4 〉 is equivalent <strong>to</strong> the reducibility of D.<br />
Corollary 5.4.9. The crystal D is irreducible if no two of the four β i add up <strong>to</strong> −v p (t).<br />
In particular, ρ is irreducible in this case.<br />
Proof. By Corollary 5.4.8, D is irreducible if and <strong>on</strong>ly if D ′ = 〈v 2 , v 4 〉 is not an admissible<br />
submodule. The submodule D ′ is not admissible if no two of the four β i add up <strong>to</strong> t N (D ′ )<br />
which is −v p (t), from the table above.<br />
55
5.4.2 τ unramified supercuspidal of dim m ≥ 2 and n ≥ 2<br />
We now prove Theorem 5.4.1 for general m and n, assuming τ is an ‘unramified supercuspidal<br />
representati<strong>on</strong>’. Let us explain this terminology. We assume that τ is an induced<br />
representati<strong>on</strong> of dimensi<strong>on</strong> m, i.e., τ ≃ Ind Wp<br />
W K<br />
χ, where K is a p-adic field such that<br />
[K : Q p ] = m, and χ is a character of W K . This is known <strong>to</strong> always hold if (p, m) = 1 or<br />
p > m. For simplicity, we shall assume that K is the unique unramified extensi<strong>on</strong> of Q p ,<br />
namely Q p m. We refer <strong>to</strong> τ in this case as an unramified supercuspidal representati<strong>on</strong>.<br />
Following the methods of [GM09, §3.3], we explicitly write down the crystal D = D τ<br />
whose underlying Weil-Deligne representati<strong>on</strong> is an unramified supercuspidal representati<strong>on</strong><br />
τ of dimensi<strong>on</strong> m. This is d<strong>on</strong>e in the next few subsecti<strong>on</strong>s. The arguments are<br />
similar <strong>to</strong> those given in [GM09, §3.3], with some minor modificati<strong>on</strong>s. We outline the<br />
steps now.<br />
Let σ be the genera<strong>to</strong>r of Gal(Q p m/Q p ) and let I p m denote the inertia subgroup of<br />
Q p m. Then<br />
τ| Ip=Ip m ≃ (Ind Wp<br />
W χ)| p m I p m ≃ ⊕ m i=1χ σi | Ip m .<br />
Since τ is irreducible, by Mackey’s criteri<strong>on</strong>, we have that χ ≠ χ σi , for all i, <strong>on</strong> W p m<br />
also <strong>on</strong> I p m. Moreover, we have that χ σj ≠ χ σi , for any i ≠ j.<br />
and<br />
Descripti<strong>on</strong> of Gal(F/Q p )<br />
First, we need <strong>to</strong> c<strong>on</strong>struct a finite extensi<strong>on</strong> F over K such that τ| IF is trivial and which<br />
simultaneously has the property that F/Q p is Galois. The c<strong>on</strong>structi<strong>on</strong> of an explicit such<br />
field F is given in [GM09, §3.3.1], in the case K = Q p 2 using Lubin-Tate theory. The<br />
structure of Gal(F/Q p ) is described in the same place. The case K = Q p m may be treated<br />
in a very similar manner. Write d for what was called in m in [GM09, §3.3.1], since m<br />
already has meaning here. Then F may be chosen such that Gal(F/Q p ) is the semi-direct<br />
product of a cyclic group 〈σ〉 with σ md = 1, with the product of the cyclic groups ∆ = 〈δ〉,<br />
with δ pm−1 = 1, and Γ = ∏ m<br />
i=1 〈γ i〉, with each γ pn<br />
i<br />
= 1, for some n. Moreover, the maximal<br />
unramified extensi<strong>on</strong> F 0 of F is F 0 = Q md<br />
p and Gal(F 0 /Q p ) = 〈˜σ〉 = Z/md, such that ˜σ| K<br />
is the genera<strong>to</strong>r of Gal(K/Q p ). By abuse of notati<strong>on</strong>, we denote ˜σ by σ itself.<br />
Descripti<strong>on</strong> of the Galois acti<strong>on</strong><br />
Recall that D is a free module of rank m over F 0 ⊗ Qp E . Let D i = D ⊗ F0 ⊗E,σi E, for<br />
i = 0, 1, . . . , md−1, be the comp<strong>on</strong>ent of D corresp<strong>on</strong>ding <strong>to</strong> σ i . Each D i is a Weil-Deligne<br />
representati<strong>on</strong> with an acti<strong>on</strong> of W p .<br />
By the definiti<strong>on</strong> of the Weil-Deligne representati<strong>on</strong>, the acti<strong>on</strong> of I p matches with the<br />
acti<strong>on</strong> of the inertia subgroup of Gal(F/Q p ), namely ∆×Γ. The restricti<strong>on</strong> of χ <strong>to</strong> I p can<br />
be written as χ| Ip = ωm r ∏ m<br />
i=1 χ i, where ω m is the fundamental character of level m, r ≥ 1,<br />
and χ i is the character of Γ which takes γ i <strong>to</strong> a p n -th root of unity ζ i , for i = 1, . . . , m.<br />
We see that each D i has a basis v i,1 , v i,2 , . . . , v i,m such that if i ≡ k (mod m), for some<br />
0 ≤ k ≤ m − 1, then<br />
g · v i,j = χ σj−k−1 (g)v i,j , (5.4.3)<br />
56
for all j = 1, . . . , m. Since σ takes D i <strong>to</strong> D i+1 , using (5.4.3), we may assume that<br />
Descripti<strong>on</strong> of acti<strong>on</strong> of ϕ<br />
σ(v i,j ) = v i+1,j ∀i, j.<br />
The opera<strong>to</strong>r ϕ acts in a cyclic manner as well, taking D i <strong>to</strong> D i+1 . Since ϕ commutes<br />
with the acti<strong>on</strong> of inertia, we see that<br />
ϕ(v i,j ) = c j v i+1,j+1 ,<br />
for some c j , for all 1 ≤ j ≤ m. Observe that c j ’s does not depend <strong>on</strong> i, since ϕ commutes<br />
with σ.<br />
1<br />
For all 1 ≤ k ≤ m, define,<br />
t k<br />
:= ∏ k<br />
j=1 c j and 1 t 0<br />
= 1. Replace the extensi<strong>on</strong> E with a<br />
finite extensi<strong>on</strong>, again denoted by E, so that it c<strong>on</strong>tains all m-th roots of all c j . Let m√ c j<br />
denote a particular m-th root, for each j.<br />
We now write down a basis of D, say {e i } m i=1 , such that ϕ(e i) = 1 m √ e<br />
t i . First, we<br />
m<br />
shall define e 1 and the other e i ’s are defined by e i = σ i−1 e 1 . The vec<strong>to</strong>r e 1 is given by<br />
e 1 =<br />
md−1<br />
∑<br />
j=0<br />
j≡j 0 (mod m), 0≤j 0 ≤m−1<br />
(t m ) j 0<br />
m<br />
t j0<br />
v j,j+1 .<br />
Here we use the obvious c<strong>on</strong>venti<strong>on</strong> that if j is such that j ≡ j 0 (mod m), with 1 ≤ j 0 ≤ m,<br />
then v i,j := v i,j0 . A small computati<strong>on</strong> shows that ϕ(e 1 ) = 1 m √ t m<br />
e 1 . Since ϕ commutes<br />
with σ, we have ϕ(e i ) = 1 m √ e<br />
t i , for all 1 ≤ i ≤ m. We obtain the D τ is a free rank m<br />
m<br />
module over F 0 ⊗ E with basis e i , i = 1, . . . , m such that<br />
⎧<br />
1<br />
ϕ(e i ) = m √ e<br />
t i ,<br />
m<br />
⎪⎨<br />
N(e i ) = 0,<br />
D τ =<br />
(5.4.4)<br />
σ(e i ) = e i+1 ,<br />
⎪⎩<br />
g(e i ) = (1 ⊗ χ σi−1 (g))(e i ), g ∈ I(F/K),<br />
for all 1 ≤ i ≤ m.<br />
When m = 2, this (ϕ, N)-module is exactly the <strong>on</strong>e given in [GM09, §3.3] (though the<br />
e i used here differ by a scalar from the e i used there).<br />
Descripti<strong>on</strong> of the filtrati<strong>on</strong><br />
For the sake of completeness, let us make some brief comments about the filtrati<strong>on</strong> <strong>on</strong><br />
D τ , even though we shall not need <strong>to</strong> use the filtrati<strong>on</strong> later.<br />
Let D be a filtered (ϕ, N, F, E)-module and write D F = F ⊗ F0 D. It is known that<br />
every Galois stable line in D F is generated by a Galois stable vec<strong>to</strong>r v (cf. [GM09, Lem.<br />
3.1]). The proof uses the fact H 1 (Gal(F/Q p ), (F ⊗ Qp E) × ) = 0. In fact, for any n ≥ 1:<br />
∏<br />
H 1 (Gal(F/Q p ), GL n (F ⊗ Qp E)) = H 1 (Gal(F/Q p ), GL n (E)),<br />
F ↩→E<br />
=<br />
(1) H1 (Gal(F/Q p ), Ind Gal(F/Qp)<br />
{e}<br />
GL n (E)),<br />
=<br />
(2) H1 ({e}, GL n (E)) = {0},<br />
57
(where (1) follows from the permutati<strong>on</strong> acti<strong>on</strong> of Gal(F/Q p ) <strong>on</strong><br />
follows from Shapiro’s lemma). Now, we can prove:<br />
∏<br />
F ↩→E<br />
GL n (E) and (2)<br />
Lemma 5.4.10. Every Gal(F/Q p )-stable submodule of D F has a basis c<strong>on</strong>sisting of Galois<br />
invariant vec<strong>to</strong>rs.<br />
Proof. Let D ′ be a Galois stable submodule of D F . By [Sav05, Lem. 2.1], we have that<br />
any F ⊗ Qp E-submodule of a filtered module with descent data is free. Hence D ′ is a<br />
free module of finite rank, say r. If {v 1 , v 2 , . . . , v r } is a basis of D ′ , then for every g ∈<br />
Gal(F/Q p ), we have g ·(v 1 , v 2 , · · · , v n ) t = c g (v 1 , v 2 , · · · , v r ) t , for some c g ∈ GL n (F ⊗ Qp E).<br />
Moreover, c g is 1-cocycle, i.e., c g ∈ Z 1 (Gal(F/Q p ), GL n (F ⊗ Qp E)). By the vanishing result<br />
above, c g is coboundary, hence c g = cg(c) −1 , for some c ∈ GL n (F ⊗ Qp E). Replacing the<br />
basis (v 1 , v 2 , · · · , v r ) t with c · (v 1 , v 2 , · · · , v r ) t , we may assume that c g = 1 and that each<br />
vec<strong>to</strong>r in {v 1 , v 2 , . . . , v r } is invariant under Gal(F/Q p ).<br />
In particular each step Fil i (D F ) in the filtrati<strong>on</strong> <strong>on</strong> D F is spanned by Gal(F/Q p )-<br />
invariant vec<strong>to</strong>rs. In [GM09, §3.3.4] the Hodge filtrati<strong>on</strong> <strong>on</strong> D F was written down explicitly<br />
when D = D τ and τ is a 2-dimensi<strong>on</strong>al unramified supercuspidal representati<strong>on</strong>.<br />
Presumably this can be d<strong>on</strong>e also when τ has dimensi<strong>on</strong> m ≥ 2, but we refrain from<br />
pursuing this.<br />
Using the explicit descripti<strong>on</strong> of the crystal above, we can prove Theorem 5.4.1 in the<br />
case above, and for brevity, we skip the proof and prove it in the general case directly.<br />
5.4.3 General case: m ≥ 2 and n ≥ 2<br />
Now, we shall prove Theorem 5.4.1 in general. Thus, we show that the <strong>on</strong>ly (ϕ, N, F, E)-<br />
submodules of D = D τ ⊗ D Sp(n) , for any irreducible representati<strong>on</strong> τ of W p dimensi<strong>on</strong><br />
m ≥ 2, and n ≥ 2, are of the form D τ ⊗ D Sp(r) , for some 1 ≤ r ≤ n. The proof uses ideas<br />
introduced for the special cases proved so far.<br />
Recall that the module D Sp(n) has a basis {f n , f n−1 , . . . , f 1 }, with properties as in<br />
§5.3.1, and say that D τ has a basis {e 1 , e 2 , . . . , e m } over F 0 ⊗ E. Let {v i } mn<br />
i=1 denote the<br />
basis of D = D τ ⊗ D Sp(r) defined by the following table:<br />
v 1 = e 1 ⊗ f n v 2 = e 1 ⊗ f n−1 · · · v n = e 1 ⊗ f 1<br />
v n+1 = e 2 ⊗ f n v n+2 = e 2 ⊗ f n−1 · · · v 2n = e 2 ⊗ f 1<br />
.<br />
. · · ·<br />
v in+1 = e i+1 ⊗ f n v in+2 = e i+1 ⊗ f n−1 · · · v in = e i+1 ⊗ f 1<br />
.<br />
. · · ·<br />
v (m−1)n+1 = e m ⊗ f n v (m−1)n+2 = e m ⊗ f n−1 · · · v mn = e m ⊗ f 1<br />
.<br />
.<br />
Lemma 5.4.11. There are no rank r (ϕ, N, F, E)-submodules of D <strong>on</strong> which N acts<br />
trivially, for 1 ≤ r ≤ m − 1.<br />
Proof. Suppose there exists such a module, say ˜D, of rank r < m. Since N acts trivially<br />
<strong>on</strong> ˜D, we have ˜D ⊆ 〈v n , v 2n , . . . , v mn 〉 = D τ ⊗ D Sp(1) ≃ D τ . But τ is irreducible, so D τ is<br />
irreducible by Lemma 4.2.14, a c<strong>on</strong>tradicti<strong>on</strong>.<br />
58
Corollary 5.4.12. The index of N <strong>on</strong> a (ϕ, N, F, E)-submodules of D is m.<br />
Proof of Theorem 5.4.1. Let D ′ be a (ϕ, N, F, E)-submodule of D = D τ ⊗ D Sp(n) . From<br />
the corollary above, there are m blocks in the Jordan can<strong>on</strong>ical form of N <strong>on</strong> D ′ . Without<br />
loss of generality assume that the blocks have sizes r 1 ≤ r 2 ≤ · · · ≤ r m with ∑ m<br />
i=1 r i =<br />
rank D ′ . Suppose w 1 , . . . , w m are the corresp<strong>on</strong>ding basis vec<strong>to</strong>rs in D ′ such that the<br />
order of nilpotency of N <strong>on</strong> w i is r i , so that the {N j (w i )} i,j form a basis of D ′ . If all the<br />
r i are equal <strong>to</strong> say r, then the usual argument shows D ′ = D τ ⊗ D Sp(r) . We show that<br />
this is indeed the case.<br />
Suppose <strong>to</strong>wards a c<strong>on</strong>tradicti<strong>on</strong> that r i ≠ r i+1 for some 1 ≤ i < m. For 1 ≤ i ≤ n,<br />
let D i be span of the vec<strong>to</strong>rs in the last i columns in the table at the start of this secti<strong>on</strong>.<br />
Observe that D i = D τ ⊗ Ker(N i ) = D τ ⊗ D Sp(i) .<br />
Now, arrange the basis vec<strong>to</strong>rs of D ′ , i.e., the N j w k , in such a way that the last<br />
column in the table c<strong>on</strong>tains the basis vec<strong>to</strong>rs <strong>on</strong> which N acts trivially and the last but<br />
<strong>on</strong>e column c<strong>on</strong>sists of basis vec<strong>to</strong>r which precisely spans Ker(N 2 ) <strong>on</strong> D ′ , and so <strong>on</strong> and<br />
so forth. With respect <strong>to</strong> this arrangement, denote the span of the vec<strong>to</strong>rs in the last i<br />
columns as A i . Since r i ≠ r i+1 , the rank of the space A ri +1/A ri is less than m. Moreover,<br />
A ri +1/A ri is a subspace of D ri +1/D ri , i.e., there is an inclusi<strong>on</strong> of (ϕ, N, F, E)-modules<br />
A ri +1/A ri ↩→ D ri +1/D ri .<br />
Now<br />
D ri +1/D ri = (D τ ⊗ D Sp(ri +1))/(D τ ⊗ D Sp(ri ))<br />
≃ D τ ⊗ (D Sp(ri +1)/D Sp(ri ))<br />
≃ D τ ⊗ D Sp(1)<br />
≃ D τ .<br />
All the isomorphisms above are isomorphism of (ϕ, N, F, E)-modules over F 0 ⊗ E. By<br />
Lemma 5.4.11, the above inclusi<strong>on</strong> is not possible! Hence all the r i are indeed equal. This<br />
finishes the proof of Theorem 5.4.1.<br />
Filtrati<strong>on</strong> <strong>on</strong> D = D τ ⊗ D Sp(n)<br />
We finally can apply the discussi<strong>on</strong> above <strong>to</strong> write down the structure of the (p, p)-<br />
representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> an au<strong>to</strong>morphic representati<strong>on</strong> of GL mn (A Q ).<br />
We start with some remarks. Suppose D 1 and D 2 are two admissible filtered modules.<br />
It is well-known (cf. [Tot96]) that the tensor product D 1 ⊗ D 2 is also admissible. The<br />
difficulty in proving this lies in the fact that <strong>on</strong>e does not have much informati<strong>on</strong> about<br />
the structure of the (ϕ, N)-submodules of the tensor product. If they are of the form<br />
D ′ ⊗ D ′′ , where D ′ and D ′′ are admissible (ϕ, N)-submodules of D 1 and D 2 respectively,<br />
then <strong>on</strong>e could use (4.2.1) <strong>to</strong> prove that D ′ ⊗ D ′′ is also admissible. But not all the<br />
submodules of D 1 ⊗ D 2 are of this form.<br />
However in the previous secti<strong>on</strong> we have just shown (cf. Theorem 5.4.1), that for<br />
D = D τ ⊗ D Sp(n) , all the (ϕ, N, F, E)-submodules of D are of the form D τ ⊗ D Sp(r) , for<br />
some 1 ≤ r ≤ n. This fact allows us <strong>to</strong> study the crystal D and its submodules, <strong>on</strong>ce we<br />
introduce the Hodge filtrati<strong>on</strong>.<br />
59
Filtrati<strong>on</strong> in general positi<strong>on</strong><br />
Assume that the Hodge filtrati<strong>on</strong> <strong>on</strong> D is in general positi<strong>on</strong> with respect <strong>to</strong> the Newt<strong>on</strong><br />
filtrati<strong>on</strong> (cf. Assumpti<strong>on</strong> 4.4.6). Let m be the rank of D τ . Let {β i,j } i=n,j=m<br />
i=1,j=1<br />
be the<br />
jumps in the Hodge filtrati<strong>on</strong> with β i1 ,j 1<br />
> β i2 ,j 2<br />
, if i 1 > i 2 , or if i 1 = i 2 and j 1 > j 2 . Thus<br />
β n,m > β n,m−1 > · · · > β n,1 > β n−1,m > · · · > β 1,m > · · · > β 1,1 .<br />
Define, for every 1 ≤ k ≤ n,<br />
and<br />
b k =<br />
j=m<br />
∑<br />
j=1<br />
β k,j ,<br />
a k = t N (D τ ⊗ D Sp(k) ) − t N (D τ ⊗ D Sp(k−1) ) = t N (D τ ) + m(k − 1),<br />
where the last equality follows from (4.2.1a). Clearly, we have that<br />
b n > · · · > b i+2 > b i+1 > b i > · · · > b 1 ,<br />
a n > · · · > a i+2 > a i+1 > a i > · · · > a 1 .<br />
(5.4.5)<br />
Observe that b i+1 −b i ≥ m 2 and a i+1 −a i = m, for every 1 ≤ i ≤ n. Since D is admissible,<br />
the submodule D τ ⊗ D Sp(i) of D is admissible if and <strong>on</strong>ly if ∑ i<br />
k=1 b k = ∑ i<br />
k=1 a k.<br />
The arguments below are similar <strong>to</strong> the <strong>on</strong>es used when analyzing the Steinberg case.<br />
We start with an analog of Lemma 5.3.2.<br />
Lemma 5.4.13. Let {a i } n i=1 be an increasing sequence of integers, such that a i+1−a i = m,<br />
for every i and for some fixed natural number m. Let {b i } n i=1 be an increasing sequence<br />
of integers, such that b i+1 − b i ≥ m 2 , for every i. Suppose that ∑ i a i = ∑ i b i. If a n = b n<br />
or a 1 = b 1 , then m = 1 and hence a i = b i , for all i.<br />
Theorem 5.4.14. If D τ ⊗D Sp(i) and D τ ⊗D Sp(i+1) are admissible submodules of D, then<br />
m = 1, in which case all the D τ ⊗ D Sp(i) , for 1 ≤ i ≤ n, are admissible.<br />
Proof. Since D τ ⊗ D Sp(i) and D τ ⊗ D Sp(i+1) are admissible, we have:<br />
b 1 + b 2 + · · · + b i =<br />
i∑<br />
a r ,<br />
r=1<br />
i+1<br />
∑<br />
b 1 + b 2 + · · · + b i+1 = a r .<br />
r=1<br />
(5.4.6)<br />
Clearly, we have b i+1 = a i+1 . By (5.4.5), (5.4.6), and by Lemma 5.4.13 we have m = 1<br />
and a i = b i , for all 1 ≤ i ≤ n. This shows that all the D τ ⊗ D Sp(i) are admissible.<br />
Theorem 5.4.15. Let D = D τ ⊗ D Sp(n) and assume that the Hodge filtrati<strong>on</strong> <strong>on</strong> D is in<br />
general positi<strong>on</strong> (cf. Assumpti<strong>on</strong> 4.4.6). Then either D is irreducible or D is reducible,<br />
in which case m = 1 and the (ϕ, N, F, E)-submodules D τ ⊗ D Sp(i) , for 1 ≤ i ≤ n are all<br />
admissible.<br />
60
Proof. Let D i = D τ ⊗ D Sp(i) , for 1 ≤ i ≤ n. If D is irreducible then we are d<strong>on</strong>e. If<br />
not, by Theorem 5.4.1, there exists an 1 ≤ i ≤ n such that D i is admissible. If D i−1 or<br />
D i+1 is also admissible, then by the above theorem m = 1 and hence all the (ϕ, N, F, E)-<br />
submodules of D are admissible. So, assume D i−1 and D i+1 are not admissible, but D i<br />
is admissible. We shall show that this is not possible. Indeed, we have:<br />
b 1 + b 2 + · · · + b i−1 <<br />
r=i−1<br />
∑<br />
r=1<br />
r=i<br />
a r ,<br />
∑<br />
b 1 + b 2 + · · · + b i = a r ,<br />
b 1 + b 2 + · · · + b i+1 <<br />
r=1<br />
r=i+1<br />
∑<br />
r=1<br />
a r .<br />
(5.4.7a)<br />
(5.4.7b)<br />
(5.4.7c)<br />
Subtracting (5.4.7b) from (5.4.7a), we get −b i < −a i . Subtracting (5.4.7b) from (5.4.7c),<br />
we get b i+1 < a i+1 . Adding these two inequalities, we get b i+1 − b i < a i+1 − a i = m. But<br />
this is a c<strong>on</strong>tradicti<strong>on</strong>, since b i+1 − b i ≥ m.<br />
For emphasis we state separately the following corollary:<br />
Corollary 5.4.16. With assumpti<strong>on</strong>s as above, for any m ≥ 2, the crystal D = D τ ⊗<br />
D Sp(n) is irreducible.<br />
Definiti<strong>on</strong> 5.4.17. Say π is ordinary at p if a 1 = b 1 , i.e., t N (D τ ) = ∑ m<br />
j=1 β 1,j.<br />
This c<strong>on</strong>diti<strong>on</strong> implies m = 1, and this definiti<strong>on</strong> then coincides with Definiti<strong>on</strong> 5.3.6.<br />
Applying the above discussi<strong>on</strong> <strong>to</strong> the <strong>local</strong> (p, p)-representati<strong>on</strong> in a strictly compatible<br />
system, we obtain:<br />
Theorem 5.4.18 (Indecomposable case). Say π is a cuspidal au<strong>to</strong>morphic representati<strong>on</strong><br />
with infinitesimal character c<strong>on</strong>sisting of distinct integers. Suppose that<br />
WD(ρ π,p | Gp ) ∼ τ m ⊗ Sp(n),<br />
where τ m is an irreducible representati<strong>on</strong> of W p of dimensi<strong>on</strong> m ≥ 1, and n ≥ 1. Assume<br />
that Assumpti<strong>on</strong> 4.4.6 holds.<br />
• If π is ordinary at p, then ρ π,p | Gp is reducible, in which case m = 1 and τ 1 is a<br />
character and ρ π,p | Gp is (quasi)-ordinary as in Theorems 5.3.7 and 5.3.8.<br />
• If π is not ordinary at p, then ρ π,p | Gp<br />
is irreducible.<br />
Tensor product filtrati<strong>on</strong><br />
One might w<strong>on</strong>der what happens if the filtrati<strong>on</strong> <strong>on</strong> D is not necessarily in general positi<strong>on</strong>.<br />
As an example, here we c<strong>on</strong>sider just <strong>on</strong>e case arising from the so called tensor<br />
product filtrati<strong>on</strong>.<br />
Assume that D τ and D Sp(n) are the usual filtered (ϕ, N, F, E)-modules and equip<br />
D τ ⊗ D Sp(n) with the tensor product filtrati<strong>on</strong>. By the formulas in Lemma 4.2.7 <strong>on</strong>e can<br />
prove:<br />
61
Lemma 5.4.19. Suppose that D = D τ ⊗ D Sp(n) has the tensor product filtrati<strong>on</strong>. Fix<br />
1 ≤ r ≤ n. Then D τ ⊗ D Sp(r) is an admissible submodule of D if and <strong>on</strong>ly if D Sp(r) is an<br />
admissible submodule of D Sp(n) .<br />
We recall that if the filtrati<strong>on</strong> <strong>on</strong> D Sp(n) is in general positi<strong>on</strong> (as in Assumpti<strong>on</strong> 4.4.6),<br />
then we have shown that furthermore D Sp(r) is an admissible submodule of D Sp(n) if and<br />
<strong>on</strong>ly if D Sp(1) is an admissible submodule.<br />
Remark 5.4.20. The lemma can be used <strong>to</strong> give an example where the tensor product<br />
filtrati<strong>on</strong> <strong>on</strong> D is not in general positi<strong>on</strong> (i.e., does not satisfy Assumpti<strong>on</strong> 4.4.6). Suppose<br />
τ is an irreducible representati<strong>on</strong> of dimensi<strong>on</strong> m = 2 and D Sp(2) has weight 2 (cf. [GM09,<br />
§3.1]). Note 〈f 1 〉 is an admissible submodule of D Sp(2) . Hence, by the lemma, D τ ⊗ 〈f 1 〉<br />
is an admissible submodule of D τ ⊗ D Sp(2) . If the tensor product filtrati<strong>on</strong> satisfies<br />
Assumpti<strong>on</strong> 4.4.6, then the admissibility of D τ ⊗ 〈f 1 〉 would c<strong>on</strong>tradict Theorem 5.4.15,<br />
since m = 2.<br />
In any case, we have the following applicati<strong>on</strong> <strong>to</strong> <strong>local</strong> Galois representati<strong>on</strong>s.<br />
Propositi<strong>on</strong> 5.4.21. Suppose that ρ π,p | Gp ∼ ρ τ ⊗ ρ Sp(n) is a tensor product of two (p, p)-<br />
representati<strong>on</strong>s, with underlying Weil-Deligne representati<strong>on</strong>s τ and Sp(n) respectively.<br />
If ρ Sp(n) is irreducible, then so is ρ π,p | Gp .<br />
5.5 General Weil-Deligne representati<strong>on</strong>s<br />
So far, we have studied the (p, p)-Galois representati<strong>on</strong> <strong>attached</strong> <strong>to</strong> π p when the underlying<br />
Weil-Deligne representati<strong>on</strong> is indecomposable. We now make some remarks in the<br />
general setting where the Weil-Deligne representati<strong>on</strong> is a direct sum of indecomposable<br />
representati<strong>on</strong>s.<br />
5.5.1 Sum of twisted Steinberg<br />
For simplicity we start with the case where the indecomposable pieces are twists of the<br />
Steinberg representati<strong>on</strong> by an unramified character. Thus we assume the underlying<br />
Weil-Deligne representati<strong>on</strong> is<br />
D χ1 ⊗ D Sp(n1 ) ⊕ D χ2 ⊗ D Sp(n2 ) ⊕ · · · ⊕ D χr ⊗ D Sp(nr),<br />
where n i ≥ 1 and χ i are unramified characters taking arithmetic Frobenius <strong>to</strong> α i . Let<br />
χ i (ω) = α i where ω is a uniformizer of Q × p . Without loss of generality we may assume<br />
that v p (α 1 ) ≥ v p (α 2 ) ≥ · · · ≥ v p (α r ).<br />
Let n = ∑ r<br />
i n i. Let {β i,j } r,n i<br />
i=1,j=1<br />
be the jumps in the Hodge filtrati<strong>on</strong> such that<br />
β i1 ,j 1<br />
> β i2 ,j 2<br />
, if i 1 > i 2 or i 1 = i 2 and j 1 > j 2 . Thus<br />
β r,nr > · · · > β r,1 > β r−1,nr−1 > · · · > β r−1,1 > · · · ><br />
β 2,n2 > · · · > β 2,1 > β 1,n1 > · · · > β 1,1 .<br />
(5.5.1)<br />
62
Let D be a filtered (ϕ, N, Q p , E)-module with associated Weil-Deligne representati<strong>on</strong><br />
as above. We now define a flag inside D as follows.<br />
⎧<br />
D χ1 ⊗ D Sp(i) if 1 ≤ i ≤ n 1 ,<br />
⎪⎨ D χ1 ⊗ D Sp(n1 ) ⊕ D χ2 ⊗ D Sp(i−n1 ) if n 1 + 1 ≤ i ≤ n 1 + n 2 ,<br />
D i =<br />
.<br />
⎪⎩<br />
⊕ r−1<br />
k=1 D χ i<br />
⊗ D Sp(ni ) ⊕ D χr ⊗ D Sp(i−(n−nr)) if n − n r + 1 ≤ i ≤ n.<br />
Clearly, D n is the full (ϕ, N)-module D. We now show that the above flag is admissible<br />
if and <strong>on</strong>ly if π is ordinary at p in the following sense:<br />
Definiti<strong>on</strong> 5.5.1. Say π is ordinary at p if β i,1 = −v p (α i ) for all 1 ≤ i ≤ r.<br />
We remark that the noti<strong>on</strong> of ordinariness extends the previous definiti<strong>on</strong>s given in<br />
Definiti<strong>on</strong> 5.2.1, when all the n i = 1 and Definiti<strong>on</strong> 5.3.6, when r = 1 and m = 1. We<br />
have:<br />
Theorem 5.5.2. Assume that Assumpti<strong>on</strong> 4.4.6 holds. Then the flag {D i } is an admissible<br />
flag in D (i.e., each D i is admissible) if and <strong>on</strong>ly if π is ordinary at p.<br />
Proof. The ‘<strong>on</strong>ly if’ part is clear. Indeed if the Hodge filtrati<strong>on</strong> is in general positi<strong>on</strong><br />
then the jump in the filtrati<strong>on</strong> <strong>on</strong> D 1 will be the last number in (5.5.1), i.e., β 1,1 , and the<br />
admissibility of D 1 shows that β 1,1 = −v p (α 1 ). Similarly, the jumps in the filtrati<strong>on</strong> <strong>on</strong><br />
D n1 (respectively D n1 +1) are the last n 1 numbers (respectively the last n 1 numbers al<strong>on</strong>g<br />
with β 2,1 ) in (5.5.1) above, and clearly t N (D n1 +1) = t N (D n1 )−v p (α 2 ), so the admissibility<br />
of D n1 and D n1 +1 <strong>to</strong>gether shows that β 2,1 = −v p (α 2 ), etc.<br />
Let us prove the ‘if’ part. Since β 1,1 = −v p (α 1 ), D 1 is admissible. Since D 2 is a<br />
(ϕ, N, Q p , E)-submodule of D we have that<br />
β 1,1 + β 1,2 ≤ (1 − v p (α 1 )) + (−v p (α 1 ))<br />
and hence β 1,2 ≤ 1 − v p (α 1 ) = 1 + β 1,1 . Thus β 1,2 − β 1,1 ≤ 1. But β 1,2 − β 1,1 ≥ 1, by<br />
(5.5.1), hence equality holds, i.e., β 1,2 = (1 − v p (α 1 )). Therefore,<br />
By a similar argument, we see that<br />
β 1,1 + β 1,2 = (1 − v p (α 1 )) + (−v p (α 1 )).<br />
∑n 1<br />
j=1<br />
∑n 1<br />
β 1,j = (j − 1 − v p (α 1 )). (5.5.2)<br />
j=1<br />
This shows that D n1 is admissible. Since D n1 +1 is an (ϕ, N, F, E)-submodule of D, we<br />
have that<br />
∑n 1<br />
∑n 1<br />
β 1,j + β 2,1 ≤ (j − 1 − v p (α 1 )) + (−v p (α 2 )),<br />
j=1<br />
j=1<br />
but this inequality is actually an equality, by (5.5.2), and since β 2,1 = −v p (α 2 ) by assumpti<strong>on</strong>.<br />
This shows that D n1 +1 is also admissible. The admissibility of the other D i is<br />
proved in a similar manner.<br />
63
5.5.2 General ordinary case<br />
We now assume that as a (ϕ, N, F, E)-module,<br />
D = D τ1 ⊗ D Sp(n1 ) ⊕ D τ2 ⊗ D Sp(n2 ) ⊕ · · · ⊕ D τr ⊗ D Sp(nr),<br />
where n i ∈ N and τ i ’s are irreducible representati<strong>on</strong>s of W p of degree m i . Without loss of<br />
generality we may assume t N (D τ1 ) ≤ t N (D τ2 ) ≤ · · · ≤ t N (D τr ).<br />
We now define a flag inside D τ1 ⊗ D Sp(n1 ) ⊕ · · · ⊕ D τr ⊗ D Sp(nr), and show that this<br />
flag is admissible if and <strong>on</strong>ly if there is a relati<strong>on</strong> between some numbers a i (depending<br />
<strong>on</strong> Newt<strong>on</strong> numbers) and b i (depending <strong>on</strong> Hodge numbers). More precisely, define the<br />
flag {D i } in D by<br />
⎧<br />
D τ1 ⊗ D Sp(i) if 1 ≤ i ≤ n 1 ,<br />
⎪⎨ D τ1 ⊗ D Sp(n1 ) ⊕ D τ2 ⊗ D Sp(i−n1 ) if n 1 + 1 ≤ i ≤ n 1 + n 2 ,<br />
D i =<br />
.<br />
⎪⎩<br />
⊕ r−1<br />
k=1 D τ k<br />
⊗ D Sp(nk ) ⊕ D τr ⊗ D Sp(i−(n−nr)) if n − n r + 1 ≤ i ≤ n.<br />
We now define the numbers a i and b i . Let {β i,j } be the jumps in the Hodge filtrati<strong>on</strong><br />
associated <strong>to</strong> D such that β i1 ,j 1<br />
> β i2 ,j 2<br />
, if i 1 > i 2 or if i 1 = i 2 but j 1 > j 2 . Thus, in the<br />
case r = 2, the jumps in the filtrati<strong>on</strong> are:<br />
β m1 +m 2 ,n 2<br />
> β m1 +m 2 −1,n 2<br />
> · · · > β m1 +1,n 2<br />
><br />
β m1 +m 2 ,n 2 −1 > β m1 +m 2 −1,n 2 −1 > · · · > β m1 +1,n 2 −1 > · · · ><br />
β m1 +m 2 ,1 > β m1 +m 2 −1,1 > · · · > β m1 +1,1 ><br />
β m1 ,n 1<br />
> β m1 −1,n 1<br />
> · · · > β 1,n1 ><br />
β m1 ,n 1 −1 > β m1 −1,n 1 −1 > · · · > β 1,n1 −1 > · · · ><br />
Define, for every 1 ≤ k ≤ n 1 ,<br />
b k =<br />
i=m<br />
∑ 1<br />
i=1<br />
β m1 ,1 > β m1 −1,1 > · · · > β 1,1 .<br />
β i,k ,<br />
a k = t N (D τ1 ⊗ D Sp(k) ) − t N (D τ1 ⊗ D Sp(k−1) ) = t N (D τ1 ) + m 1 (k − 1).<br />
Clearly, we have that<br />
b n1 > b n1 −1 > · · · > b 2 > b 1 ,<br />
a n1 > a n1 −1 > · · · > a 2 > a 1 .<br />
Observe that b i+1 − b i ≥ m 2 1 and a i+1 − a i = m 1 , for 1 ≤ i < n 1 . Under Assumpti<strong>on</strong> 4.4.6,<br />
the jumps in the induced Hodge filtrati<strong>on</strong> <strong>on</strong> D τ1 ⊗ D Sp(j) are β m1 ,j > · · · > β 1,1 , so that<br />
D τ1 ⊗ D Sp(j) is an admissible submodule of D if and <strong>on</strong>ly if ∑ j<br />
k=1 b k = ∑ j<br />
k=1 a k.<br />
Similarly define b k and a k for all 1 ≤ k ≤ n = ∑ r<br />
i n i. For example, if n 1 +1 ≤ n 1 +k ≤<br />
n 1 + n 2 , define<br />
b n1 +k =<br />
i=m<br />
∑ 2<br />
i=1<br />
β m1 +i,k,<br />
a n1 +k = t N (D τ2 ⊗ D Sp(k) ) − t N (D τ2 ⊗ D Sp(k−1) ) = t N (D τ2 ) + m 2 (k − 1).<br />
64
Again, we have<br />
b n1 +n 2<br />
> b n1 +n 2 −1 > · · · > b n1 +2 > b n1 +1,<br />
a n1 +n 2<br />
> a n1 +n 2 −1 > · · · > a n1 +2 > a n1 +1,<br />
and b i+1 − b i ≥ m 2 2 and a i+1 − a i = m 2 , for every n 1 + 1 ≤ i < n 1 + n 2 , etc.<br />
Definiti<strong>on</strong> 5.5.3. Say π is ordinary at p if a P i−1<br />
j=1 n j+1 = bP i−1<br />
j=1 n , for 1 ≤ i ≤ r.<br />
j+1<br />
Note that this definiti<strong>on</strong> of ordinariness reduces <strong>to</strong> Definiti<strong>on</strong> 5.4.17 when r = 1, but<br />
also <strong>to</strong> Definiti<strong>on</strong> 5.5.1 since it implies m i = 1, for 1 ≤ i ≤ r.<br />
Theorem 5.5.4. Assume that Assumpti<strong>on</strong> 4.4.6 holds. Then, the flag {D i } is admissible<br />
(i.e., each D i is an admissible submodule of D) if and <strong>on</strong>ly if π is ordinary at p.<br />
Proof. We prove the ‘<strong>on</strong>ly if’ directi<strong>on</strong> for r = 2, since the general case is similar, and <strong>on</strong>ly<br />
notati<strong>on</strong>ally more cumbersome. Thus we have <strong>to</strong> show that if the flag {D i } is admissible,<br />
then a 1 = b 1 and a n1 +1 = b n1 +1. The proof is an easy applicati<strong>on</strong> of Lemma 5.4.13.<br />
Indeed<br />
• The admissibility of D 1 shows that a 1 = b 1 .<br />
• The admissibility of D n1<br />
Lemma 5.4.13).<br />
and a 1 = b 1 shows m 1 = 1 and a i = b i for 1 ≤ i ≤ n 1 (by<br />
• The admissibility of D n1 +1 and D n1 <strong>to</strong>gether shows a n1 +1 = b n1 +1.<br />
• The admissibility of D n1 +n 2<br />
and D n1 and a n1 +1 = b n1 +1 shows m 2 = 1 and a n1 +i =<br />
b n1 +i for 1 ≤ i ≤ n 2 (by Lemma 5.4.13).<br />
Since all m i = 1, the proof of the ‘if’ part of the theorem is exactly the same as the ‘if’<br />
part of the proof of Theorem 5.5.2, noting b P i−1<br />
j=1 n j+1 = β i,1 and and a P i−1<br />
j=1 n j+1 = −v p(α i ),<br />
for 1 ≤ i ≤ r.<br />
Remark 5.5.5. The theorem does not tell us when D is irreducible, since there are a large<br />
number of (ϕ, N, F, E)-submodules of D which are not part of the flag c<strong>on</strong>sidered above.<br />
For instance, it seems hard <strong>to</strong> determine the admissibility of the submodule D τi ⊗D Sp(ni ),<br />
except when i = 1.<br />
Translating the above theorem in terms of the (p, p)-representati<strong>on</strong>, we obtain:<br />
Theorem 5.5.6 (Decomposable case). Suppose π is a cuspidal au<strong>to</strong>morphic representati<strong>on</strong><br />
of GL N (A Q ) with infinitesimal character given by the integers −β 1 > · · · > −β N .<br />
Suppose that N = ∑ r<br />
i=1 m in i and<br />
WD(ρ π,p | Gp ) ∼ ⊕ r i=1τ i ⊗ Sp(n i ),<br />
where τ i is an irreducible representati<strong>on</strong> of dimensi<strong>on</strong>s m i ≥ 1, and n i ≥ 1. If π is<br />
ordinary at p, then m i = 1 for all i, the β i occur in r blocks of c<strong>on</strong>secutive integers, of<br />
lengths n i , for 1 ≤ i ≤ r, and<br />
⎛<br />
⎞<br />
ρ n1 ∗ · · · ∗<br />
0 ρ n2 · · · ∗<br />
ρ π,p | Gp ∼ ⎜<br />
⎟<br />
⎝ 0 0 · · · ∗ ⎠ ,<br />
0 0 0 ρ nr<br />
65
where each ρ ni is an n i -dimensi<strong>on</strong>al representati<strong>on</strong> with shape similar <strong>to</strong> that in Theorem<br />
5.3.8. In particular, ρ π,p | Gp is quasi-ordinary.<br />
5.6 Further remarks<br />
Theorem 5.5.6 treats the general ordinary case. In the general n<strong>on</strong>-ordinary case, the<br />
behaviour of the (p, p)-Galois representati<strong>on</strong> is more complex and we do not have complete<br />
informati<strong>on</strong> about irreducibility (compare with Theorem 5.4.18). In this secti<strong>on</strong> we<br />
c<strong>on</strong>tent ourselves with a few c<strong>on</strong>cluding remarks about new issues that arise.<br />
To fix ideas we assume that the Weil-Deligne representati<strong>on</strong> associated <strong>to</strong> π p is of the<br />
form<br />
χ 1 ⊗ Sp(2) ⊕ χ 2 ⊗ Sp(2),<br />
where χ 1 and χ 2 are unramified characters of W p taking arithmetic Frobenius <strong>to</strong> α 1 and<br />
α 2 , respectively. Let D be the associated (ϕ, N, Q p , E)-module. Let β 4 > · · · > β 1 be the<br />
jumps in the Hodge filtrati<strong>on</strong> <strong>on</strong> D. We c<strong>on</strong>tinue <strong>to</strong> assume that Assumpti<strong>on</strong> 4.4.6 holds.<br />
Just as in previous secti<strong>on</strong>s, we ignore the filtrati<strong>on</strong>, and first classify the (ϕ, N, Q p , E)-<br />
submodules of D.<br />
Classificati<strong>on</strong> of (ϕ, N)-submodules of D<br />
Let e 1 , e 2 = N(e 1 ) be a basis of χ 1 ⊗ Sp(2) and f 1 , f 2 = N(f 1 ) be a basis of χ 2 ⊗ Sp(2).<br />
Sometimes we write 〈e 2 〉 for χ 1 ⊗ Sp(1), etc.<br />
If α 1 = α 2 , then any 1-dimensi<strong>on</strong>al subspace of 〈e 2 , f 2 〉 is a (ϕ, N)-submodule of<br />
D. In particular, there exists infinitely many 1-dimensi<strong>on</strong>al submodules of D. This is<br />
already in striking c<strong>on</strong>trast with the statement of Theorem 5.4.1, which says that in<br />
the indecomposable case there are <strong>on</strong>ly finitely many (ϕ, N)-submodules. So already we<br />
can expect that the analysis in the decomposable case might be much more complex.<br />
We remark, however, that if α 1 ≠ α 2 , then 〈e 2 〉 and 〈f 2 〉 are the <strong>on</strong>ly 1-dimensi<strong>on</strong>al<br />
submodules of D.<br />
Again, if α 1 = α 2 , then the 2-dimensi<strong>on</strong>al (ϕ, N)-submodule of D are of the form<br />
〈ae 1 + cf 1 , ae 2 + cf 2 〉, for some a, c ∈ E. If α 1 ≠ α 2 , but p/α 1 = 1/α 2 , then again<br />
there are again infinitely many 2-dimensi<strong>on</strong>al (ϕ, N)-submodules and they are given by<br />
〈e 1 + bf 2 , e 2 〉 or 〈f 1 + be 2 , f 2 〉, for some b ∈ E. If α 1 ≠ α 2 and p/α 1 ≠ 1/α 2 , then there<br />
are <strong>on</strong>ly finitely many 2-dimensi<strong>on</strong>al (ϕ, N)-submodules: they are χ 1 ⊗ Sp(2), χ 2 ⊗ Sp(2)<br />
and the diag<strong>on</strong>al <strong>on</strong>e χ 1 ⊗ Sp(1) ⊕ χ 2 ⊗ Sp(1).<br />
Finally, like the 1-dimensi<strong>on</strong>al case, if α 1 = α 2 , then all the 3-dimensi<strong>on</strong>al (ϕ, N)-<br />
submodules of D are of form 〈ae 1 + bf 1 , e 2 , f 2 〉, for any a, b ∈ E. If α 1 ≠ α 2 , there are<br />
exactly two 3-dimensi<strong>on</strong>al submodules, namely χ 1 ⊗ Sp(2) ⊕ χ 2 ⊗ Sp(1) and χ 1 ⊗ Sp(1) ⊕<br />
χ 2 ⊗ Sp(2).<br />
Hence, if we choose α 1 and α 2 generically (i.e., α 1 ≠ α 2 and α 1 ≠ pα 2 ), then there are<br />
<strong>on</strong>ly finitely many (ϕ, N)-submodules of D, otherwise there are infinitely many (ϕ, N)-<br />
submodules of D. The following table c<strong>on</strong>tains the possible Newt<strong>on</strong> numbers of the<br />
(ϕ, N)-submodules D ′ of D.<br />
66
dim E D ′ t N (D ′ ) when α 1 ≠ α 2 t N (D ′ ) when α 1 = α 2<br />
1 −v p (α 1 ), −v p (α 2 ) −v p (α 1 )<br />
2 −v p (α 1 ) − v p (α 2 ), 1 − 2v p (α 1 ), 1 − 2v p (α 2 ) 1 − 2v p (α 1 )<br />
3 1 − 2v p (α 1 ) − v p (α 2 ), 1 − v p (α 1 ) − 2v p (α 2 ) 1 − 3v p (α 1 )<br />
4 1 − 2v p (α 1 ) + 1 − 2v p (α 2 ) 2 − 4v p (α 1 )<br />
An irreducible example<br />
We can now easily c<strong>on</strong>struct examples such that the crystal D is irreducible. For instance,<br />
choose any α 1 ∈ E with v p (α 1 ) = 0 and take α 2 = α 1 . Take (β 4 , β 3 , β 2 , β 1 ) = (2, 1, 0, −1).<br />
Using the table above, <strong>on</strong>e can easily check that there are no admissible (ϕ, N)-submodules<br />
of D, except for D itself. We note that since α 1 = α 2 , there are infinitely many (ϕ, N)-<br />
submodules of D, but <strong>on</strong>ly finitely many c<strong>on</strong>diti<strong>on</strong>s <strong>to</strong> check for n<strong>on</strong>-admissibility.<br />
All complete flags cannot be reducible<br />
In proving Theorem 5.5.6 we showed that ordinariness implies that a particular complete<br />
flag is admissible. We now wish <strong>to</strong> point out that not all complete flags in D are necessarily<br />
admissible, even under the ordinariness assumpti<strong>on</strong>. Indeed, in the setting of the example<br />
of this secti<strong>on</strong>, if we choose α 1 and α 2 such that v p (α 1 ) ≠ v p (α 2 ), then any two complete<br />
flags whose 1-dimensi<strong>on</strong>al subspaces are 〈e 2 〉 and 〈f 2 〉, respectively, cannot be admissible<br />
simultaneously, since (under Assumpti<strong>on</strong> 4.4.6) we have both β 1 = −v p (α 1 ) and β 1 =<br />
−v p (α 2 ).<br />
Intermediate cases<br />
Finally, in the general decomposable case, the regularity (distinct Hodge-Tate weights) of<br />
the (p, p)-Galois representati<strong>on</strong> ρ π,p | Gp does not imply that it is either (quasi)-ordinary<br />
in the sense of Definiti<strong>on</strong> 4.4.7 or irreducible (compare with Theorem 5.4.18). We now<br />
give an example of such an ‘intermediate case’, i.e., an example for which the (p, p)-<br />
Galois representati<strong>on</strong> is reducible, but such that there is no complete flag of reducible<br />
submodules.<br />
Let D be as above. Choose α 1 and α 2 such that v p (α 1 ) = 1 and v p (α 2 ) = −10.<br />
Take (β 4 , β 3 , β 2 , β 1 ) = (17, 4, 0, −1). From the table above, we see that χ 1 ⊗ Sp(1),<br />
χ 1 ⊗Sp(2), and D, are admissible and all the other (ϕ, N)-submodules satisfy the c<strong>on</strong>diti<strong>on</strong><br />
that their Hodge numbers are less than or equal <strong>to</strong> their Newt<strong>on</strong> numbers. So D is<br />
reducible. However, since α 1 ≠ α 2 and α 1 ≠ pα 2 , there are <strong>on</strong>ly two 3-dimensi<strong>on</strong>al<br />
(ϕ, N)-submodules of D, and again from the table we see that neither is admissible.<br />
Hence, there is no admissible complete flag of (ϕ, N)-submodules of D.<br />
67
Chapter 6<br />
2-adic Hida theory and<br />
Applicati<strong>on</strong>s<br />
Hida theory for the prime p is the theory [Hid86b], [Hid86a] that deals with p-ordinary<br />
cuspidal families. While generalizati<strong>on</strong>s are now available, for au<strong>to</strong>morphic <strong>forms</strong> <strong>on</strong><br />
groups other than GL 2 and base fields other than Q, many authors tend <strong>to</strong> shy away from<br />
the prime p = 2. We show that the basic results in the literature stated for odd primes p<br />
remain valid for the prime p = 2. We apply these results <strong>to</strong> study the <strong>local</strong> semisimplicity<br />
of ordinary modular Galois representati<strong>on</strong>s.<br />
In §6.1, we briefly recall some useful results about group and sheaf cohomology. In<br />
§6.2, we study the relati<strong>on</strong>s between various cohomology groups with coefficients. In<br />
§6.3, we state a c<strong>on</strong>trol theorem for cohomology groups of modular curves and §6.4, §6.5<br />
c<strong>on</strong>tain its proof. Since some results that we need from the theory of mod 2 modular<br />
<strong>forms</strong> are not known, we replace them with other ingredients (cf. Theorem 6.5.3 and<br />
§6.6) <strong>to</strong> prove the c<strong>on</strong>trol theorem above for p = 2. Finally, in §6.8, we show that Hida’s<br />
c<strong>on</strong>trol theorem for the ordinary Λ-adic Hecke algebra holds in this setting.<br />
In §6.9, we deduce a uniqueness theorem for 2-stabilized ordinary cuspidal new<strong>forms</strong><br />
in Hida families. Namely, we show that such <strong>forms</strong> live in unique primitive 2-ordinary<br />
cuspidal families, up <strong>to</strong> Galois c<strong>on</strong>jugacy, a well-known result when p is odd.<br />
Towards the end (cf. §6.10), we show that a recent pretty applicati<strong>on</strong> of Hida theory<br />
for odd primes p, namely <strong>to</strong> understanding the <strong>local</strong> semisimplicity of p-ordinary modular<br />
Galois representati<strong>on</strong>s, c<strong>on</strong>tinues <strong>to</strong> hold for the prime p = 2. Following the approach<br />
of [GV04] for odd primes p, and assuming that a modularity result of Buzzard [Buz03] for<br />
Artin-like representati<strong>on</strong>s holds in sufficient generality for the prime p = 2, we show that<br />
almost all arithmetic members of a primitive n<strong>on</strong>-CM 2-ordinary cuspidal family have<br />
<strong>local</strong>ly n<strong>on</strong>-split Galois representati<strong>on</strong>s. The uniqueness result menti<strong>on</strong>ed above implies<br />
that n<strong>on</strong>e of these possibly finitely many excepti<strong>on</strong>s are CM <strong>forms</strong>.<br />
6.1 Background<br />
In this secti<strong>on</strong>, we closely follow the expositi<strong>on</strong> in [Hid86a]. Let GL 2 (R) act <strong>on</strong> the upper<br />
half plane H as follows. If α ∈ GL + 2 (R), then we let α act <strong>on</strong> H through linear fracti<strong>on</strong>al<br />
transformati<strong>on</strong>s. Let ɛ = ( )<br />
1 0<br />
0 −1 act <strong>on</strong> H by ɛ(z) = −¯z. If A ∈ GL2 (R) with det(A) < 0,<br />
69
then we let A act H using the decompositi<strong>on</strong> A = ɛ(ɛA). Let ι : M 2 (R) → M 2 (R) be<br />
the involuti<strong>on</strong> defined by α + α ι = Tr(α), i.e., α ι = Adj(α). Let ∆ be a semi-group in<br />
GL 2 (Q) and ∆ ι denote ι(∆). Let M be Z[∆ ι ]-module and Φ be a c<strong>on</strong>gruence subgroup<br />
of SL 2 (Z) c<strong>on</strong>tained in ∆ ∩ ∆ ι . Thus M becomes a left Φ-module. The abstract Hecke<br />
ring is denoted by R(Φ, ∆) (cf. [Shi71, Chap. 4], for more details).<br />
6.1.1 Group and sheaf cohomology<br />
A map u of Φ <strong>to</strong> M is called a 1-cocycle, if u satisfies u(αβ) = u(α) + α · u(β), for all<br />
α, β ∈ Φ. Let Z(Φ, M) denote the group of all 1-cocycles of Φ with values in M, and<br />
B(Φ, M) the subgroup of Z(Φ, M) c<strong>on</strong>sisting of maps of the form u(α) = (α − 1)x, for<br />
x ∈ M. Then the usual first cohomology group for M is defined by<br />
H 1 (Φ, M) := Z(Φ, M)/B(Φ, M).<br />
Now we shall define the parabolic cohomology group H 1 p(Φ, M) as follows: Let U be<br />
any unipotent subgroup of SL 2 (Q) and put Φ U = {±1}U ∩Φ. Then we have the restricti<strong>on</strong><br />
map res U : H 1 (Φ, M) → H 1 (Φ U , M). We shall define<br />
H 1 p(Φ, M) := {c ∈ H 1 (Φ, M) | res U (c) = 0 for all unipotent subgroups U}.<br />
Let U ∞ be the standard unipotent subgroup { ( 1 u<br />
0 1<br />
)<br />
| u ∈ Q}. Every unipotent subgroup<br />
U of SL 2 (Q) can be written as αU ∞ α −1 with α ∈ SL 2 (Z). Then, the corresp<strong>on</strong>dence<br />
U ↦→ α(∞) ∈ P 1 (Q) gives a bijecti<strong>on</strong> between unipotent subgroups and cusps for SL 2 (Z).<br />
For each cusp s ∈ P 1 (Q), we denote the corresp<strong>on</strong>ding unipotent subgroup by U s . We<br />
write Φ s for Φ Us . Another useful descripti<strong>on</strong> of Φ s , which we use more frequently, is given<br />
by Φ s = {α ∈ Φ | α(s) = s}.<br />
Let C(Φ) be a representative (finite) set for the Φ-equivalence classes of cusps. For<br />
each cusp s ∈ P 1 (Q), there exists γ ∈ Φ such that s 0 = γ(s) ∈ C(Φ), hence γΦ s γ −1 = Φ s0 .<br />
This observati<strong>on</strong> simplifies the definiti<strong>on</strong> of the parabolic cohomology group, i.e., there<br />
exists a short exact sequence<br />
0 → H 1 p(Φ, M) → H 1 (Φ, M) → G 1 (Φ, M), (6.1.1)<br />
where G i (Φ, M) = ⊕ s∈C(Φ) H i (Φ s , M), and the last arrow sends each cohomology class <strong>to</strong><br />
the sum of its restricti<strong>on</strong>s <strong>to</strong> Φ s .<br />
To define the sheaf cohomology, we assume that Φ is <strong>to</strong>rsi<strong>on</strong>-free, e.g., Φ = Γ 1 (M)<br />
with M ≥ 4. Write Y for the open complex manifold Φ\H. We give the Φ-module M the<br />
discrete <strong>to</strong>pology and define F (M) = Φ\(H × M). Then F (M) is an étale covering of Y ,<br />
and we can c<strong>on</strong>sider the sheaf of c<strong>on</strong>tinuous secti<strong>on</strong>s of F (M) over Y , which we denote by<br />
the same symbol F (M). Then, we c<strong>on</strong>sider the usual cohomology group H 1 (Y, F (M)) and<br />
that of compact support H 1 c(Y, F (M)). We shall define the parabolic sheaf cohomology<br />
group H 1 p(Y, F (M)) by the image of H 1 c(Y, F (M)) in H 1 (Y, F (M)).<br />
There are well-known isomorphisms (cf. [Hid81, Prop. 1.1]), which make the following<br />
diagram commutative:<br />
H 1 (Y, F (M))<br />
∼ H 1 (Φ, M)<br />
(6.1.2)<br />
H 1 p(Y, F (M))<br />
∼<br />
H 1 p(Φ, M).<br />
70
6.1.2 Hecke acti<strong>on</strong> <strong>on</strong> the cohomology groups<br />
We shall define the acti<strong>on</strong> of the Hecke ring R(Φ, ∆) <strong>on</strong> the cohomology groups H 1 (Φ, M),<br />
H 1 p(Φ, M), and G 1 (Φ, M). Let Φ ′ be another c<strong>on</strong>gruence subgroup c<strong>on</strong>tained in ∆ ∩ ∆ ι .<br />
We shall define an opera<strong>to</strong>r [ΦαΦ ′ ] : H 1 (Φ, M) → H 1 (Φ ′ , M) for each double coset ΦαΦ ′<br />
in ∆. Decompose the double coset ΦαΦ ′ = ∪ i Φα i . The number of left cosets of Φ in<br />
ΦαΦ ′ is finite and for each γ ∈ Φ ′ , by definiti<strong>on</strong>, we can find γ i ∈ Φ such that γ i α j = α i γ<br />
for some α j . Then we can define a map v : Φ ′ → M by v(γ) = ∑ i αι i · u(γ i). Thus the<br />
corresp<strong>on</strong>dence [u] ↦→ [v] defines a morphism:<br />
[ΦαΦ ′ ] : H 1 (Φ, M) → H 1 (Φ ′ , M),<br />
such that it preserves parabolic cocyles, and hence [ΦαΦ ′ ] acts <strong>on</strong> H 1 p(Φ, M). In particular<br />
when Φ ′ = Φ, the Hecke ring R(Φ, ∆) acts <strong>on</strong> H 1 (Φ, M) and H 1 p(Φ, M). Similarly, <strong>on</strong>e<br />
can define an opera<strong>to</strong>r [ΦαΦ ′ ] : H 0 (Φ, M) → H 0 (Φ ′ , M) by x|[ΦαΦ ′ ] = ∑ i αι i · x, for each<br />
x ∈ H 0 (Φ, M). Hence, the Hecke ring R(Φ, ∆) acts <strong>on</strong> H i (Φ, M) for each i = 0, 1.<br />
Now we introduce a morphism [ΦαΦ ′ ] : G i (Φ, M) → G i (Φ ′ , M) for i = 0, 1. If the<br />
number of left Φ t -cosets in Φ t αΦ ′ s for t ∈ C(Φ) and s ∈ C(Φ ′ ) is finite, we can define a<br />
morphism [Φ t αΦ ′ s] : H i (Φ t , M) → H i (Φ ′ s, M) for i = 0, 1 in the same manner as above. Fix<br />
s ∈ C(Φ ′ ) and write ΦαΦ ′ = ∪ disj Φβ i Φ ′ s, for some β i ’s and write each Φβ i Φ ′ s = ∪ disj Φβ i π j ,<br />
for some π j ∈ Φ ′ s.<br />
Lemma 6.1.1 ([Hid86a], Lem. 4.1). Let t = β i (s). Then the uni<strong>on</strong> ∪ j Φ t β i π j coincides<br />
with Φ t β i Φ ′ s and is disjoint. Especially, the number of left cosets in Φ t β i Φ ′ s is finite.<br />
For a given s ∈ C(Φ ′ ), decompose ΦαΦ ′ as above. By definiti<strong>on</strong>, we can find γ ∈ Φ<br />
so that γβ i (s) ∈ C(Φ). Thus, we may assume that β i (s) ∈ C(Φ) by substituting γβ i for<br />
β i , if necessary. Then, by the last Lemma, we can define a morphism<br />
[Φ t βΦ ′ s] : H i (Φ t , M) → H i (Φ ′ s, M),<br />
for t = β i (s). For c ∈ G i (Φ, M) (resp., G i (Φ ′ , M)), let us write c t (resp., c s ) for the<br />
comp<strong>on</strong>ent of c in H i (Φ t , M) (resp., H i (Φ ′ s, M)). Then, we shall define<br />
(c|[ΦαΦ ′ ]) s = ∑ i<br />
c βi (s)|[Φ βi (s)β i Φ ′ s].<br />
The opera<strong>to</strong>r defined above depends <strong>on</strong>ly <strong>on</strong> the double coset ΦαΦ ′ and via this acti<strong>on</strong><br />
the module G i (Φ, M) becomes an R(Φ, ∆)-module.<br />
We now briefly recall the acti<strong>on</strong> of double cosets <strong>on</strong> sheaf cohomology groups (cf.<br />
[Hid81, §3], for more details). Assume that Φ ′ is <strong>to</strong>rsi<strong>on</strong>-free. For each α ∈ ∆, we put<br />
Φ α = Φ ′ ∩ α −1 Φα, Φ α = αΦ α α −1 ,<br />
and Y ′ = Φ ′ \H, Y α = Φ α \H, Y α = Φ α \H. Then the map α : H × M → H × M defined by<br />
α(z, v) = (α −1 (z), α ι v) induces a morphism [α] : F (M)| Y α → F (M)| Yα , which gives rise<br />
<strong>to</strong> a morphism [α] : H i (Y α , F (M)) → H i (Y α , F (M)). Since Y α /Y ′ is an étale covering,<br />
we have the trace map Tr Yα/Y ′ : Hi (Y α , F (M)) → H i (Y ′ , F (M)) and the restricti<strong>on</strong> map<br />
Res Y α /Y : H i (Y, F (M)) → H i (Y α , F (M)). We shall define the acti<strong>on</strong> of double coset<br />
[ΦαΦ ′ ] : H i (Y, F (M)) → H i (Y ′ , F (M))<br />
by Tr Yα/Y ′ ◦ [α] ◦ Res Y α /Y . In exactly the same manner, we define the acti<strong>on</strong> of [ΦαΦ ′ ] <strong>on</strong><br />
H i c(Y, F (M)) and H i (Y, F (M)). These acti<strong>on</strong>s are compatible with the isomorphism (6.1.2).<br />
71
6.1.3 Geometry of Riemann surfaces<br />
As before, assume that Φ is <strong>to</strong>rsi<strong>on</strong>-free. Take a point y ∈ Y . Then Φ can be identified<br />
with the fundamental group π 1 (Y ) of Y with the base point y. Let X be the smooth<br />
compactificati<strong>on</strong> of Y at cusps. Let g denote the genus of X. We choose 2g-curves<br />
{α 1 , β 1 , . . . , α g , β g } passing through y, but not crossing any cusps of X, which form a<br />
system of can<strong>on</strong>ical genera<strong>to</strong>rs of the group π 1 (X). Namely, π 1 (X) is isomorphic <strong>to</strong> the<br />
quotient of the free group generated by {α 1 , β 1 , . . . , α g , β g } by the unique relati<strong>on</strong><br />
[α g , β g ] . . . [α 1 , β 1 ] = 1,<br />
where [α g , β g ] denote the commuta<strong>to</strong>r of α g and β g . By cutting X al<strong>on</strong>g these 2g-curves,<br />
we have a simply c<strong>on</strong>nected polyg<strong>on</strong> of 4g-sides, and inside the polyg<strong>on</strong>, there are cusps<br />
of X. Write X − Y = {x 1 , . . . , x d }, and draw curves π i <strong>on</strong> the polyg<strong>on</strong> from y encircling<br />
each cusp x i and assume that they intersect <strong>on</strong>ly at y. Then, Φ = π 1 (Y ) is generated by<br />
{π 1 , . . . , π d } and {α 1 , β 1 , . . . , α g , β g }, with the <strong>on</strong>ly relati<strong>on</strong> π d · · · π 1 · [α g , β g ] · · · [α 1 , β 1 ] =<br />
1. Let Φ ab be the free Z-module generated by {π 1 , . . . , π d , α 1 , β 1 , . . . , α g , β g } and Φ ∞ ab be<br />
the free submodule of Φ ab generated by {π 1 , . . . , π d }. For each Z-module A, let Φ act <strong>on</strong><br />
A trivially. Then we have a natural commutative diagram<br />
H 1 (Y, A)<br />
∼ H 1 (Φ, A)<br />
∼<br />
{ϕ ∈ Hom(Φ ab , A) | ∑ d<br />
i=1 ϕ(π i) = 0}<br />
(6.1.3)<br />
H 1 (X, A)<br />
∼<br />
H 1 p(Φ, A)<br />
∼<br />
{ϕ ∈ Hom(Φ ab , A) | ϕ(π i ) = 0 ∀ i}<br />
Put P (Φ) = {π 1 , . . . , π d }. By definiti<strong>on</strong>, we can identify P (Φ) with the set of genera<strong>to</strong>rs<br />
of (the free part of) Φ s for s ∈ C(Φ). Then, we have<br />
H 1 (Φ, A) ≃ H 1 (Φ, Z) ⊗ Z A, H 1 p(Φ, A) ≃ H 1 p(Φ, Z) ⊗ Z A, and,<br />
G(Φ, A) = H 1 (Φ, A)/H 1 p(Φ, A) ≃ {ϕ ∈ Hom(Φ ∞ ab , A) |<br />
6.1.4 Hida’s idempotent opera<strong>to</strong>r<br />
∑<br />
π∈P (Φ)<br />
ϕ(π) = 0}.<br />
Let us now introduce the idempotent opera<strong>to</strong>r <strong>attached</strong> <strong>to</strong> T p in the Hecke algebra. Suppose<br />
R is a commutative sub-algebra of an endomorphism algebra End(M), where M is a<br />
finite free Z p -module. Let T be an element of R. The algebra R/pR is finite-dimensi<strong>on</strong>al<br />
over F p . Hence, the image ˜T of T in R/pR can be decomposed in<strong>to</strong> the unique sum s + n<br />
of a semisimple element s and a nilpotent element n. Thus, for a sufficiently large r, the<br />
element ˜T pr coincides with s pr and becomes semisimple. Then, we can choose a positive<br />
integer u so that ˜T pru gives an idempotent of R/pR. This idempotent can be lifted <strong>to</strong> a<br />
unique idempotent e of R. In fact, this idempotent can be given as a p-adic limit in R by<br />
e = lim<br />
r→∞<br />
T pru .<br />
This idempotent is independent of the choice of the integer u.<br />
Let K be a finite extensi<strong>on</strong> of Q p and O K be the integral closure of Z p in K. For r ≥ 1<br />
and k ≥ 2, let h k (Γ 1 (Np r ), O K ) denote the Hecke algebra of level Np r and of weight k.<br />
72
Applying the above discussi<strong>on</strong> with R = h k (Γ 1 (Np r ), O K ), we get an idempotent opera<strong>to</strong>r<br />
e r which is compatible with the opera<strong>to</strong>r e r−1 . Thus we can define an idempotent e of<br />
lim<br />
←−r≥1 h k(Γ 1 (Np r ), O K ). For modules M over the Hecke algebras, we define the ordinary<br />
part of M <strong>to</strong> be eM. Sometimes, we denote the ordinary part of M by M 0 .<br />
6.2 Relati<strong>on</strong>s between cohomology groups with coefficients<br />
When we study the ordinary parts of the cohomology groups of Γ 1 (Np r ) with (p, N) =1,<br />
for different r’s, we also need <strong>to</strong> work with the ordinary parts of cohomology groups of<br />
Φ s r, for r ≥ s ≥ 0, where<br />
{( )<br />
}<br />
Φ s r := Γ 1 (Np s ) ∩ Γ 0 (p r a b<br />
) = ∈ SL 2 (Z) | c ≡ 0 (mod Np r ), a ≡ 1 (mod Np s ) ,<br />
c d<br />
{( )<br />
}<br />
∆ s a b<br />
r := ∈ M 2 (Z) | ad − bc > 0, c ≡ 0 (mod Np r ), a ≡ 1 (mod Np s ) .<br />
c d<br />
When studying the acti<strong>on</strong> of the Hecke opera<strong>to</strong>rs <strong>on</strong> (usual or parabolic) cohomology<br />
groups, often <strong>on</strong>e needs <strong>to</strong> decompose them in<strong>to</strong> disjoint uni<strong>on</strong> of left cosets with a clever<br />
choice of coset representatives. Such useful decompositi<strong>on</strong>s can be found in [Hid86a, Lem.<br />
4.3]. We recall with proof <strong>on</strong>ly part (ii) of that lemma, since the hypotheses of the original<br />
statement are mildly misstated and the remaining parts are stated as in that lemma.<br />
Lemma 6.2.1. Let r, m ≥ 1, r ≥ s. For every integer u ∈ Z, let α u ∈ M 2 (Z) be such<br />
that<br />
α u ≡ ( 1 u<br />
(mod Np max(m,r) ) and det(α u ) = p m .<br />
0 p m )<br />
Then we have a disjoint decompositi<strong>on</strong><br />
Φ s ( 1 u<br />
)<br />
r 0 p m Φ<br />
s<br />
r = ⋃<br />
Φ s rα u .<br />
u mod p m<br />
Proof. Suppose that m ≥ r. The proof in the other case is similar. Take Γ ′ in (3.3.2) of<br />
[Shi71, p. 67], as Φ s r, h <strong>to</strong> be the kernel of (Z/Np r ) × → (Z/Np s ) × and N <strong>to</strong> be Np r .<br />
By [Shi71, Prop. 3.33], we have that Φ s ( 1 0<br />
)<br />
r 0 p m Φ<br />
s<br />
r = {β ∈ ∆ ′ | det(β) = p m }, where ∆ ′<br />
is as in [Shi71, p. 68]. Since the number of left cosets of Φ s r in Φ s ( 1 u<br />
)<br />
r 0 p m Φ<br />
s<br />
r is p m , and<br />
u ≡ u ′ (mod p m ) if and <strong>on</strong>ly if α u ≡ α u ′ (mod p m ), we see that the lemma follows.<br />
Lemma 6.2.2 ( [Hid86a], Lem. 4.3). 1. For each r, s and m satisfying r − s ≥ m > 0<br />
and s ≥ 1, we have: Φ s ( 1 0<br />
)<br />
r 0 p m Φ<br />
s<br />
r = Φ s ( 1 0<br />
)<br />
r 0 p m Φ<br />
s<br />
r−m . In particular, we have<br />
Φ s ( 1 0<br />
)<br />
r 0 p r−s Φ<br />
s<br />
r = Φ s ( 1 0<br />
)<br />
r 0 p r−s Φ<br />
s<br />
s .<br />
2. For each prime l, we have a disjoint decompositi<strong>on</strong><br />
⎧ (<br />
Φ s 1 0<br />
) ⎨ ∪<br />
r(<br />
0 l Φ<br />
s u mod l Φs 1 u<br />
) ( )<br />
r 0 l ∪ Φ<br />
s<br />
r σ l 0 l 0 1 if l ∤ Np,<br />
r =<br />
)<br />
⎩ ∪<br />
if l | Np,<br />
and<br />
u mod l Φs r<br />
( 1 u<br />
0 l<br />
Φ s rlσ l Φ s r = Φ s rlσ l if l ∤ Np,<br />
where σ l is an element of SL 2 (Z) satisfying σ l ≡ ( l −1 ∗<br />
0 l<br />
)<br />
(mod Np r ).<br />
73
3. Take δ ∈ SL 2 (Z) such that δ ≡ ( )<br />
0 1<br />
−1 0 (mod p 2r ) and δ ≡ 1 (mod N 2 ). We have a<br />
disjoint decompositi<strong>on</strong>:<br />
Φ 0 rδΦ 0 r = ∪ rδ ( )<br />
1 u<br />
u mod p rΦ0 0 1 .<br />
We now introduce certain important modules and we later study the relati<strong>on</strong>s between<br />
cohomology groups with these as coefficients. Firstly, we c<strong>on</strong>sider the column vec<strong>to</strong>r space<br />
L n (Z) = Z n+1 for each n<strong>on</strong>-negative integer n. Let (x, y) t be a variable vec<strong>to</strong>r in L 1 (Z)<br />
and define<br />
( ) x n<br />
= (x n , x n−1 y, . . . , y n ) ∈ L n (Z).<br />
y<br />
We let M 2 (Z) act <strong>on</strong> L n (Z) through the symmetric n-th tensor representati<strong>on</strong> explicitly<br />
given by ( ) ( )<br />
a b x n ( ) ax + by n<br />
· =<br />
.<br />
c d y cx + dy<br />
For any Z-module A, put L n (A) = L n (Z) ⊗ Z A with the natural acti<strong>on</strong> of M 2 (Z) <strong>on</strong> the<br />
left fac<strong>to</strong>r. For Φ = Γ 1 (Np r ), we take ∆ = M 2 (Z) ∩ GL 2 (Q) (cf. §6.1). Let K be a finite<br />
extensi<strong>on</strong> of Q p and let O denote the integral closure of Z p in K. For a natural number<br />
t, we define the map χ t from (Z/p r Z) × → (Z/p r Z) × ⊂ (O/p r O) × by χ t (a) = a t . Finally,<br />
let χ be a character from (Z p /p r Z p ) × <strong>to</strong> either O × or (O/p t O) × .<br />
We shall define a twisted acti<strong>on</strong> of α = ( )<br />
a b<br />
c d ∈ (∆<br />
s<br />
r ) ι <strong>on</strong> L n (O/p t O) by α · x =<br />
χ(d)(αx) for x ∈ L n (O/p t O), where αx is the usual acti<strong>on</strong>. We denote this twisted<br />
module by L n (χ, O/p t O). In the theorem below, we relate the cohomology groups for the<br />
module L n (O/p r O) with those of the module L 0 (χ n , O/p r O). For every natural number<br />
r, we define<br />
i r : L n (O/p r O) → L 0 (χ n , O/p r O)<br />
and<br />
(x 0 , . . . , x n ) → x n<br />
j r : L 0 (χ −n , O/p r O) → L n (O/p r O)<br />
x 0 → (x 0 , 0, . . . , 0)<br />
These are Γ 0 (p r )-module homomorphisms and they covariantly induces the maps:<br />
ι r := (i r ) ∗ : H 1 (Φ s r, L n (O/p r O)) → H 1 (Φ s r, L 0 (χ n , O/p r O))<br />
and<br />
(j r ) ∗ : H 1 (Φ s r, L 0 (χ −n , O/p r O)) → H 1 (Φ s r, L n (O/p r O)).<br />
Let τ ∈ M 2 (Z) such that det(τ) = p r and,<br />
τ ≡ ( )<br />
0 −1<br />
p r 0<br />
(mod p 2r ) and τ ≡ ( 1 0<br />
0 p r )<br />
(mod N 2 ).<br />
One checks that [τ] induces an isomorphism between the groups H 1 (Φ s r, L 0 (χ, O/p r O))<br />
and H 1 (Φ s r, L 0 (χ −1 , O/p r O)) defined as follows: For each cocycle u, let (u|[τ])(γ) =<br />
θ(u(τγτ −1 )), where θ is an isomorphism between L 0 (χ, O/p r O) ≃ L 0 (χ −1 , O/p r O). Let<br />
δ ∈ SL 2 (Z) be such that<br />
δ ≡ ( )<br />
0 1<br />
−1 0<br />
(mod p 2r ) and δ ≡ ( )<br />
1 0<br />
0 1<br />
74<br />
(mod N 2 ).
We define a new morphism H 1 (Φ s r, L 0 (χ n , O/p r O)) → πr<br />
H 1 (Φ s r, L n (O/p r O)), which is the<br />
compositi<strong>on</strong> of the following maps<br />
[τ]<br />
H 1 (Φ s r, L 0 (χ n , O/p r O)) H 1 (Φ s r, L 0 (χ −n , O/p r O))<br />
<br />
(j r) ∗<br />
H 1 (Φ s r, L n (O/p r O)) H 1 (Φ s r, L n (O/p r O)).<br />
[Φ s rδΦ s r]<br />
The abstract Hecke ring R(Φ s r, ∆ s r) acts <strong>on</strong> the cohomology groups for L n (A) as follows.<br />
For each prime l, the Hecke opera<strong>to</strong>rs T (l) and T (l, l) act <strong>on</strong> the cohomology groups for<br />
L n (A) by the acti<strong>on</strong> of double cosets (cf. §6.1.2), i.e., T (l) := [ Φ s 1 0<br />
) ]<br />
r(<br />
0 l Φ<br />
s<br />
r and<br />
{<br />
[Φ<br />
s<br />
r lσ l Φ s r] if l ∤ Np r ,<br />
T (l, l) :=<br />
0 if l | Np r ,<br />
where σ l is an element of SL 2 (Z) such that σ l ≡ ( l −1 ∗<br />
0 l<br />
)<br />
(mod Np r ).<br />
Theorem 6.2.3. For each positive integer r, with r > s ≥ 1, we have:<br />
π r ◦ ι r = T (1, p r ) <strong>on</strong> H 1 (Φ s r, L n (χ, O/p r O)),<br />
and<br />
ι r ◦ π r = T (1, p r ) <strong>on</strong> H 1 (Φ s r, L 0 (χ n , O/p r O)).<br />
Moreover, the maps ι r , π r preserve parabolic classes and the map ι r is T p -equivariant.<br />
Proof. We show that ι r is equivariant under the acti<strong>on</strong> of T = T (l) or T (l, l), for all l.<br />
By Lemma 6.2.2 (2), we can decompose T as disjoint uni<strong>on</strong> of left cosets:<br />
T = ∪ i Φ s rα i with α i ≡ ( )<br />
1 ∗<br />
0 ∗<br />
(mod p r ).<br />
If l | Np, then T (l, l) is zero. If l ∤ Np, then T (l, l) = [Φ s rlσ l Φ s r] = [Φ s rlσ l ] = Φ s rlσ l .<br />
Now, by the choice of σ l , we see the above decompositi<strong>on</strong>. A similar argument holds for<br />
T (l) also. Then, <strong>on</strong> L n (O/p r O), αi ι acts by the matrix of the form:<br />
⎛<br />
⎞<br />
∗ ∗ · · · ∗<br />
⎜<br />
⎟<br />
⎝· · · · · · · · · ∗⎠ ∈ M n+1 (O/p r O).<br />
0 0 · · · 1<br />
To check the map ι r is T p -equivariant, we need <strong>to</strong> show that ι r ◦ (u|T ) = (ι r ◦ u)|T . For<br />
each 1-cocycle u : Φ s r → L n (O/p r O), we have<br />
ι r ((u|T )(γ)) = ι r ( ∑ i<br />
α ι i.u(γ i )) = ∑ i<br />
ι r (u(γ i )) = ∑ i<br />
(ι r ◦ u)(γ i ) = ((ι r ◦ u)|T )(γ),<br />
where γ i ∈ Φ s r is defined by the relati<strong>on</strong> α i γ = γ i α j for some j.<br />
We now prove the identity π r ◦ ι r = T (1, p r ). By definiti<strong>on</strong>, we have:<br />
(π r ◦ ι r (u))(γ) = ([Φ s rδΦ s r] ◦ (j r ) ∗ ◦ [τ] ◦ ι r (u))(γ) =<br />
p∑<br />
r −1<br />
a=0<br />
δ ι a · j r (ι r (τγ a τ −1 )),<br />
(∗)<br />
75
where δ a = δ ( )<br />
1 a<br />
0 1 for δ ∈ SL2 (Z) such that δ ≡ ( )<br />
0 1<br />
−1 0 (mod p 2r ) and δ ≡ 1 (mod N 2 ).<br />
The element γ a is defined by the equati<strong>on</strong> δ a γ = γ a δ b for some b with 0 ≤ b < p r . Since<br />
τ ≡ ( )<br />
0 −1<br />
0 0 (mod p r ), we have τ ι · x = (x n , . . . , 0) t , for x = (x 0 , . . . , x n ) ∈ L n (O/p r O),<br />
and therefore τ ι · x = j r (i r (x)). The expressi<strong>on</strong> in (∗) reduces <strong>to</strong><br />
(π r ◦ ι r (u))(γ) =<br />
p r −1<br />
∑<br />
(τδ a ) ι · j r (ι r (τγ a τ −1 )).<br />
a=0<br />
We have τδ a ≡ ( )<br />
1 a<br />
0 p r (mod Np r ), det(α a ) = p r , and (τδ a )γ = (τγ a τ −1 )(τδ b ). By<br />
Lemma 6.2.1, we see that the required identity holds.<br />
We now check the identity ι r ◦ π r = T (1, p r ) <strong>on</strong> H 1 (Φ s r, L 0 (χ n , O/p r O)). For each<br />
1-cocycle u : Φ s r → L 0 (χ n , O/p r O), we have<br />
p∑<br />
r −1<br />
((ι r ◦ π r )(u))(γ) = ι r ( δa ι · j r (τγ a τ −1 )) =<br />
a=0<br />
p∑<br />
r −1<br />
a=0<br />
ι r (δ ι a · j r (u(τγ a τ −1 ))).<br />
Assume that ι r (δ ι a · j r (x)) = x. Then the expressi<strong>on</strong> in (∗∗) implies the identity, because<br />
((ι r ◦ π r )(u))(γ) =<br />
p∑<br />
r −1<br />
a=0<br />
ι r (δ ι a · j r (u(τγ a τ −1 ))) =<br />
p∑<br />
r −1<br />
a=0<br />
u(τγ a τ −1 ) = (u|T (1, p r ))(γ),<br />
by the decompositi<strong>on</strong> of T (1, p r ). The identity ι r (δa ι · j r (x)) = x follows from the fact that<br />
the matrix δa ι ≡ ( )<br />
−a −1<br />
1 0 (mod p r ) acts <strong>on</strong> L n (O/p r O) by a matrix of the form:<br />
⎛<br />
⎞<br />
∗ ∗ · · · ∗<br />
⎜<br />
⎟<br />
⎝· · · · · · · · · ∗⎠ ∈ M n+1 (O/p r O).<br />
1 0 · · · 0<br />
(∗∗)<br />
6.3 Main theorems<br />
We start this secti<strong>on</strong> by introducing some notati<strong>on</strong>s, and then we state <strong>on</strong>e of the main<br />
theorems of this chapter. From now <strong>on</strong> p = 2, q = 4 and (2, N) = 1, unless explicitly<br />
stated. Let Γ 0 = Γ 1 = Z × p and for r ≥ 2, let Γ r denote the subgroup 1 + p r Z p of Γ, where<br />
Γ = Γ 2 = 1 + qZ p = 〈u〉. For r ≥ s ≥ 0, there is a short exact sequence of groups<br />
0 → Γ 1 (Np r ) → Φ s r → Γ s /Γ r → 0, (6.3.1)<br />
induced by Φ s r ∋ ( a b<br />
c d<br />
)<br />
↦→ ¯d ∈ Γs /Γ r . We write Φ r for Φ 0 r, for r ≥ 0. In Hida theory for<br />
odd primes p, the c<strong>on</strong>gruence group Φ 1 plays an important role, but when p = 2, the role<br />
of this group is played by the group Φ 2 = Γ 0 (4) ∩ Γ 1 (N), which is <strong>to</strong>rsi<strong>on</strong>-free, if N > 1.<br />
Let K be a finite extensi<strong>on</strong> of Q p and O K be the integral closure of Z p in K. By<br />
definiti<strong>on</strong>, there is a tau<strong>to</strong>logical character ι : Γ ↩→ Λ K = O K [[Γ]], which takes u <strong>to</strong> itself<br />
in Λ K . For each character χ : Γ → O × K , the element P χ = ι(u) − χ(u) is a prime element,<br />
and the quotient Λ K /P χ Λ K is isomorphic <strong>to</strong> O K . If χ(u) = ɛ(u)u k , where ɛ is a finite<br />
order character of Γ, we write P k,ɛ for P χ , and simply write P k , if ɛ is trivial. We may<br />
identify Λ K with O K [[X]] sending u <strong>to</strong> 1 + X. When K = Q p , we denote Λ Qp by Λ.<br />
76
Finally, let ω denote the mod 4 cyclo<strong>to</strong>mic character, that is, ω is the mod 4 character<br />
defined by ω(x) = ±1, for x ≡ ±1 (mod 4).<br />
Recall that p = 2, and N is odd. We <strong>on</strong>ly use the c<strong>on</strong>gruence subgroup Γ 1 (Np r ) with<br />
r ≥ 2, which is a <strong>to</strong>rsi<strong>on</strong>-free group. We denote the corresp<strong>on</strong>ding complex Riemann<br />
surface by Y r and its compactificati<strong>on</strong> by X r . Let H i (Y r , M) and H i (X r , M) denote the<br />
corresp<strong>on</strong>ding sheaf cohomology groups for each c<strong>on</strong>stant sheaf M of Z-modules. It is<br />
well-known that<br />
S 2 (Γ 1 (Np r )) ≃ H 1 (X r , R),<br />
where the right hand side of the isomorphism, the sheaf cohomology with R coefficients,<br />
can be identified with de Rham cohomology. The above isomorphism is invariant under<br />
the Hecke acti<strong>on</strong>. The Hecke algebra h 2 (Γ 1 (Np r ), Z) acts <strong>on</strong> H 1 (X r , Z) and therefore<br />
h 2 (Γ 1 (Np r ), Z p ) acts <strong>on</strong> H 1 (X r , Z p ), H 1 (X r , Q p ), H 1 (X r , T p ), where H 1 (X r , M) =<br />
H 1 (X r , Z) ⊗ Z M and T p := Q p /Z p .<br />
For every positive integer r ≥ 2, we simply write<br />
V r = H 1 (X r , T p ), W r = H 1 (Y r , T p ).<br />
Since H 1 (X r , T p ) ≃ H 1 (X r , Q p )/H 1 (X r , Z p ), we see that V r and W r are p-divisible modules<br />
of finite Z p -corank. Therefore End(V r ) and End(W r ) are free of finite rank. By §6.1.4, we<br />
can define Hida’s idempotent opera<strong>to</strong>r e r <strong>attached</strong> <strong>to</strong> the Hecke opera<strong>to</strong>r T p in End(V r )<br />
and End(W r ). Define the ordinary parts of V r <strong>to</strong> be V 0 r = e r V r and similarly for W r .<br />
Therefore, V 0 r is a module for h 0 2 (Γ 1(Np r ), Z p ), the ordinary part of h 2 (Γ 1 (Np r ), Z p ).<br />
There is also an acti<strong>on</strong> of Γ 0 (Np r )/Γ 1 (Np r ) <strong>on</strong> V 0 r and W 0 r . Let V denote the direct<br />
limit of V r and define similarly W, V 0 and W 0 . Since (Z/Np r Z) × acts <strong>on</strong> V 0 r and W 0 r ,<br />
hence the inverse limit, over r ≥ 2, of (Z/Np r Z) × acts <strong>on</strong> V 0 and W 0 . In particular V 0<br />
and W 0 become c<strong>on</strong>tinuous modules over the Iwasawa algebra Λ = Z p [[Γ]], if we equip<br />
them with the discrete <strong>to</strong>pology. Let V 0 and W 0 denote the P<strong>on</strong>tryagin dual modules of<br />
V 0 and W 0 , respectively. Then V 0 and W 0 are compact Λ-modules. We can now state<br />
<strong>on</strong>e of the main theorems of this chapter.<br />
Theorem 6.3.1. Let p = 2. We have:<br />
1. For each positive integer r ≥ 2, the restricti<strong>on</strong> morphism of cohomology groups<br />
induces an isomorphism of V 0 r <strong>on</strong><strong>to</strong> (V 0 ) Γr . The same result also holds for W 0 .<br />
2. Let N > 1. The modules V 0 and W 0 are free modules of finite rank over Λ.<br />
The first part of Theorem 6.3.1 gives c<strong>on</strong>trol of the ordinary parts of the cohomology<br />
modules associated with the decreasing sequence of c<strong>on</strong>gruence subgroups Γ 1 (N2 r ), for<br />
r ≥ 2, and we refer <strong>to</strong> such a result as a c<strong>on</strong>trol theorem (for cohomology).<br />
6.4 C<strong>on</strong>trol theorem for cohomology<br />
Before we start the proof of part (1) of Theorem 6.3.1, we shall state a lemma.<br />
Lemma 6.4.1. Let {M r } r≥2 be an inductive system of compatible modules over Z p [Γ/Γ r ],<br />
respectively. Assume that for all r ≥ t ≥ 2, Mr<br />
Γt = M t . Then (lim M −→r r ) Γt = M t .<br />
77
Proof. Clearly, lim M −→r r is a Z p [[Γ]]-module. Since H q (Γ t , lim M −→r r ) = lim H q (Γ −→r t , M r ), for<br />
q ≥ 0, we have (lim M −→r r ) Γt = M t , where the last equality follows from the assumpti<strong>on</strong>.<br />
Now, we start the proof of the c<strong>on</strong>trol theorem for cohomology.<br />
Lemma 6.4.2. If Φ s r/Γ 1 (Np r ) acts <strong>on</strong> T p trivially, then H 2 (Φ s r/Γ 1 (Np r ), T p ) = 0.<br />
Proof. Since Φ s r/Γ 1 (Np r ) is a finite cyclic group, H 2 (Φ s r/Γ 1 (Np r ), T p ) = T p /N T p , where<br />
N denote the norm map from T p <strong>to</strong> itself. Since Φ s r/Γ 1 (Np r ) acts <strong>on</strong> T p trivially, N is<br />
multiplicati<strong>on</strong> by the index of Γ 1 (Np r ) in Φ s r, hence is surjective.<br />
Lemma 6.4.3. For each r ≥ s ≥ 2, eH 1 (Γ 1 (Np s ), T p ) ≃ eH 1 (Φ s r, T p ).<br />
Proof. Since Φ s r ⊆ Γ 1 (Np s ), there is a restricti<strong>on</strong> map H 1 (Γ 1 (Np s ), T p ) → H 1 (Φ s r, T p ).<br />
We have the following commutative diagram<br />
H 1 (Γ 1 (Np s ), T p )<br />
H 1 (Φ s r, T p )<br />
Tp<br />
r−s<br />
Tp<br />
<br />
r−s<br />
[Φr( s 1 0<br />
0 p<br />
)Φ r−s s s]<br />
H 1 (Γ 1 (Np s ), T p ) H 1 (Φ s r, T p ).<br />
By applying the idempotent opera<strong>to</strong>r, we get that the vertical morphisms are isomorphisms<br />
and hence the diag<strong>on</strong>al map is an isomorphism.<br />
res <br />
For each r ≥ s ≥ 2, we have the inflati<strong>on</strong>-restricti<strong>on</strong> sequence<br />
0 → H 1 (Φ s r/Γ 1 (Np r ), T p )<br />
res<br />
ι<br />
→ H 1 (Φ s r, T p )<br />
→ H 1 (Γ 1 (Np r ), T p ) Γs → H 2 (Φ s r/Γ 1 (Np r ), T p ) = 0<br />
where the last term vanishes by Lemma 6.4.2. The image of the group H 1 (Φ s r/Γ 1 (Np r ), T p )<br />
inside H 1 (Φ s r, T p ) is annihilated by the idempotent e <strong>attached</strong> <strong>to</strong> T p by [Hid86a, Lem. 6.1].<br />
Therefore,<br />
eH 1 (Φ s r, T p ) ≃ eH 1 (Γ 1 (Np r ), T p ) Γs = (W 0 r ) Γs .<br />
By Lemma 6.4.3, we have that Ws<br />
0<br />
these isomorphisms, we get<br />
= eH 1 (Γ 1 (Np s ), T p ) ≃ eH 1 (Φ s r, T p ). By combining<br />
W 0 s ≃ (W 0 r ) Γs .<br />
By Lemma 6.4.1, for any r ≥ 2, we have that (W 0 ) Γr ≃ Wr 0 . This finishes the proof of<br />
the c<strong>on</strong>trol theorem for W 0 . Note that so far the proof works for all primes p ≥ 2.<br />
Now, we prove the c<strong>on</strong>trol theorem for V 0 , c<strong>on</strong>centrating <strong>on</strong> what changes need <strong>to</strong> be<br />
made in Hida’s original proof when p = 2. By (6.1.3), for r ≥ 2, the module W r is given<br />
by {ϕ ∈ Hom(Γ 1 (Np r ), T p ) | ∑ π∈P (Γ 1 (Np r )) ϕ(π) = 0}, and V r is the submodule of W r<br />
given by<br />
{<br />
}<br />
V r = ϕ ∈ Hom(Γ 1 (Np r ), T p ) | ϕ(π) = 0 for π ∈ P (Γ 1 (Np r )) ,<br />
since the group Γ 1 (Np r ) acts trivially <strong>on</strong> T p .<br />
By Lemma 6.4.1, it is enough <strong>to</strong> prove that Vs 0 = (Vr 0 ) Γs , for r ≥ s ≥ 2. Since<br />
Ws 0 ≃ (Wr 0 ) Γs , we have Vs 0 ↩→ (Vr 0 ) Γs . Therefore, it is enough <strong>to</strong> prove the surjectivity of<br />
78
this map. Since (W 0 r ) Γs = W 0 s , given a homomorphism ϕ : Γ 1 (Np r ) → T p invariant under<br />
Γ s and satisfying ϕ| e = ϕ, there exists a homomorphism ψ : Γ 1 (Np s ) → T p satisfying<br />
ψ| e = ψ such that ψ = ϕ <strong>on</strong> Γ 1 (Np r ). Thus, we need <strong>to</strong> show that ψ(π) = 0, for all π ∈<br />
P (Γ 1 (Np s )), assuming the same holds for ϕ with r instead of s. Let [ψ] denote the equivalence<br />
class of ψ in the module eG(Γ 1 (Np s ), T p ) := eH 1 (Γ 1 (Np s ), T p )/eH 1 p(Γ 1 (Np s ), T p ).<br />
We need <strong>to</strong> show that [ψ] = 0. We know that [ψ]| e = [ψ]. If [ψ]| 1−e = [ψ], then [ψ] = 0,<br />
since e is an idempotent. Hence, it is enough <strong>to</strong> show that<br />
[ψ]| 1−e = [ψ]<br />
holds. By following the strategy in [Hid86a], this reduces <strong>to</strong> proving [Hid86a, Thm. 5.8],<br />
which characterizes the elements of (1 − e)G(Γ 1 (Np s ), T p ) as elements of the set<br />
V (T p ) := {ψ ∈ Hom(Γ 1 (Np s ) ∞ ab , T p) | ψ(π) = 0, for all π ∈ P (Γ 1 (Np s ))<br />
corresp<strong>on</strong>ding <strong>to</strong> the unramified cusps},<br />
where the module Γ 1 (Np s ) ∞ ab as in §6.1.3. Under the above equality, if ψ ∈ V (T p), then<br />
[ψ] = (1 − e)[ψ ′ ] and hence [ψ]| 1−e = [ψ] holds. Thus it suffices <strong>to</strong> show that ψ ∈ V (T p ).<br />
Since every unramified cusp of X s over X 0 is under an unramified cusp of X r over X 0 ,<br />
the elements of P (Γ 1 (Np s )) corresp<strong>on</strong>ding <strong>to</strong> unramified cusps in X s can be taken <strong>to</strong><br />
be am<strong>on</strong>g the elements of P (Γ 1 (Np r )) corresp<strong>on</strong>ding <strong>to</strong> unramified cusps in X r . Then<br />
ψ(π) = ϕ(π) = 0, for all π ∈ P (Γ 1 (Np s )), which corresp<strong>on</strong>ds <strong>to</strong> unramified cusps of<br />
Γ 1 (Np s ), as desired.<br />
The proof of [Hid86a, Thm. 5.8] depends, firstly, <strong>on</strong> various relati<strong>on</strong>s between the<br />
dimensi<strong>on</strong>s of the space of Eisenstein series for Γ 1 (Np r ) with coefficients in A and the<br />
boundary cohomology G 1 (Γ 1 (Np r ), L n (A)) (cf. §6.1.1), for any subalgebra A of C or C p ,<br />
and sec<strong>on</strong>dly, <strong>on</strong> the validity of [Hid86a, Prop. 5.7]. The results <strong>on</strong> the dimensi<strong>on</strong>s of<br />
the space of Eisenstein series and the space G 1 (Γ 1 (Np r ), L n (A)) also hold for the prime<br />
p = 2. But, in the proof of [Hid86a, Prop. 5.7], <strong>on</strong>e crucially uses the fact that p ≠ 2.<br />
Thus <strong>to</strong> finish the proof of the c<strong>on</strong>trol theorem for V 0 when p = 2, it suffices <strong>to</strong> check<br />
that the propositi<strong>on</strong> holds.<br />
The difference between [Hid86a, Prop. 5.7] and the analogous result for p = 2 (Propositi<strong>on</strong><br />
6.4.5 below) is that in the former case, i.e., when Φ = Γ 1 (Np r ) for p ≥ 5 and<br />
r ≥ 1, the group Φ is <strong>to</strong>rsi<strong>on</strong>-free with regular cusps, whereas in the latter case, i.e., when<br />
Φ = Γ 1 (Np r ), for p = 2 and r ≥ 2, the group Φ is <strong>to</strong>rsi<strong>on</strong>-free, but its cusps are not<br />
necessarily regular, for example, when N = 1 and r = 2. When p = 2, these irregular<br />
cusps create problems in the proof given in [Hid86a, Prop. 5.7].<br />
Definiti<strong>on</strong> 6.4.4. Let Φ be a c<strong>on</strong>gruence subgroup of Γ 1 (Np r ) for r ≥ 2. Let s ∈ C(Φ)<br />
be a cusp and s 0 be the image of s in C(Γ 1 (N)). We say that s is unramified, if s is<br />
unramified over s 0 , in the sense of finite covers of Riemann surfaces.<br />
Propositi<strong>on</strong> 6.4.5. Let Φ = Γ 1 (Np r ) with r ≥ 2. Let A be either Z p , Z p /p i Z p or any field<br />
of characteristic 0. Let s be any unramified cusp of Φ and ρ s : G 1 (Φ, A) → H 1 (Φ s , A)<br />
be the natural projecti<strong>on</strong> map. Then for any c ∈ G 1 (Φ, A), we have that ρ s (c|e) = ρ s (c).<br />
Proof. For any positive integer M ≥ 3 with M ≠ 4, all the cusps of Γ 1 (M) are regular.<br />
Hence, when p = 2, all cusps of Γ 1 (Np r ), for N ≥ 3 or r ≥ 3 are regular (so in particular<br />
are the unramified <strong>on</strong>es). So, it is enough <strong>to</strong> c<strong>on</strong>sider the case when Φ = Γ 1 (4).<br />
79
By [Hid86a, Lem. 5.1], Γ 1 (4) has an unique unramified cusp, namely ∞. We see that this<br />
cusp is also regular since otherwise we would have<br />
π s = π ∞ = − ( 1 u<br />
0 1<br />
)<br />
,<br />
for some u > 0, which is not an element of Γ 1 (4). Hence, when N = 1 and r = 2, the<br />
unique unramified cusp is also regular.<br />
For any cusp s ∈ C(Φ), the group Φ s is an infinite cyclic group and generated by π s .<br />
The evaluati<strong>on</strong> of 1-cocyles at πs<br />
N gives an isomorphism between H 1 (Φ s , A) and A. Thus<br />
H 1 (Φ s , A) ≃ H 1 (Φ s , Z p ) ⊗ Zp A.<br />
Thus, we may assume that A = Z p . Take an integer m ≥ r such that p m ≡ 1 (mod N).<br />
By replacing s by another cusp in the Φ-equivalence of s if necessary, we may suppose<br />
that α(∞) = s with α ∈ Γ 0 (p m ), by [Hid86a, Lem. 5.1]. Then α −1 π s α = ( )<br />
1 u<br />
0 1 with<br />
positive u prime <strong>to</strong> p, since all the unramified cusps of Φ are regular. We shall choose a<br />
disjoint decompositi<strong>on</strong>:<br />
Φ ( 1 0<br />
0 p m )<br />
Φ = ∪<br />
p m −1<br />
j=0 Φβ j and Φ s β 0 Φ s = ∪ pm −1<br />
j=0 Φ sβ j<br />
with β j ∈ Φ s , for all j, β j π N = β j+1 , if 0 ≤ j < p m − 1, and β p m −1π N = π N β 0 .<br />
Take β j ′ = ( 1 jNu<br />
)<br />
0 p , for 0 ≤ j < p m m . Then, β j ′ ≡ ( )<br />
1 0<br />
0 1 (mod N). If we put βj =<br />
αβ j ′ α−1 , then β j ≡ 1 (mod N). Since α = ( )<br />
a b<br />
c d ∈ Γ0 (p m ), we have β j ≡ ( )<br />
1 −ab+a 2 Nuj<br />
0 0<br />
(mod p m ) and det(β j ) = p m . Since a, N and u are relatively prime <strong>to</strong> p, we see that<br />
−ab + a 2 Nuj runs over all residues modulo p m , as j varies over all residues modulo p m .<br />
By Lemma 6.2.1 and using the fact that m ≥ r, we have the equality<br />
Φ ( 1 0<br />
0 p m )<br />
Φ = ∪<br />
p m −1<br />
j=0 Φβ j.<br />
Since β ′ j (∞) = ∞, we see that β j(s) = s, for all j. If 0 ≤ j < p m − 1, then<br />
(α −1 β j α)(α −1 πα) N = β j( ′ 1 u<br />
) N (<br />
0 1 = 1 (j+1)Nu<br />
)<br />
0 p = β<br />
′ m j+1 ,<br />
and therefore β j π N = β j+1 . When j = p m − 1, a similar calculati<strong>on</strong> shows that β j π N =<br />
π N β 0 holds. Using the properties of β j ’s, we see that Φβ 0 Φ s = Φ ( )<br />
1 0<br />
0 p m Φ = ∪<br />
p m −1<br />
j=0 Φβ j.<br />
Now, by Lemma 6.1.1, we have the disjoint decompositi<strong>on</strong>:<br />
Φ s β 0 Φ s = ∪ pm −1<br />
j=0 Φ sβ j . (6.4.1)<br />
From the above decompositi<strong>on</strong>, we see that, for each 1-cocycle ϕ : Φ s → A, we have<br />
ϕ|[Φ s β 0 Φ s ](π N ) = ϕ(π N ), which shows that ϕ|[Φ s β 0 Φ s ] = ϕ. Since the Hecke opera<strong>to</strong>r<br />
T (p m ) acts <strong>on</strong> G 1 (Φ, A) by ρ s (c|T (p m )) = ρ s (c)|[Φ s β 0 Φ s ] (cf. §6.1.2), we see that<br />
ρ s (c|T (p m )) = ρ s (c), and hence the propositi<strong>on</strong>.<br />
This finishes the proof of the c<strong>on</strong>trol theorem for cohomology. We remark that subsequently<br />
we will <strong>on</strong>ly need <strong>to</strong> c<strong>on</strong>sider N ≥ 3, for applicati<strong>on</strong>s <strong>to</strong> p-adic families.<br />
Before proving the part (2) of Theorem 6.3.1, we prove an important result. The<br />
following propositi<strong>on</strong> was proved in [Hid88a, Prop. 2.3] for Γ 1 (Np r ) when p ≥ 5. We<br />
shall prove a similar result for Φ s r for the prime p = 2 and with N ≥ 3.<br />
80
Propositi<strong>on</strong> 6.4.6. For r ≥ 2, the quotient module eH 1 (Φ s r, L n (Z p ))/eH 1 p(Φ s r, L n (Z p )) is<br />
Z p -free.<br />
Proof. We write ∆ for Φ s r. If eG 1 (∆, L n (Z p )) is Z p -free, then the propositi<strong>on</strong> follows,<br />
since the maps in (6.1.1) are invariant under the acti<strong>on</strong> of the Hecke opera<strong>to</strong>rs. C<strong>on</strong>sider<br />
the l<strong>on</strong>g exact sequence of cohomology groups<br />
G 0 (∆, L n (Q p )) β → G 0 (∆, L n (T p )) → G 1 (∆, L n (Z p )) → G 1 (∆, L n (Q p )),<br />
induced by the short exact sequence 0 → L n (Z p ) → L n (Q p ) → L n (T p ) → 0.<br />
Since (2, N) = 1 and N ≥ 3, all the cusps of ∆ are regular, because irregularity for<br />
∆ implies the irregularity for Γ 1 (N), but there are no irregular cusps for Γ 1 (N). Let<br />
s ∈ C(∆) be a cusp of ∆. Let α s = ( )<br />
a b<br />
c d ∈ SL2 (Z) such that α s (∞) = s. If π s denotes a<br />
genera<strong>to</strong>r for ∆ s , we may write<br />
(<br />
π s = α 1 u<br />
)<br />
s 0 1 α<br />
−1<br />
s = ( )<br />
1−cau a 2 u<br />
−c 2 u 1+cau ∈ ∆, with u ≠ 0. (6.4.2)<br />
We shall show that the idempotent e annihilates G 0 (∆, L n (T p ))/β(G 0 (∆, L n (Q p ))). Since<br />
the image of β is p-divisible, it is sufficient <strong>to</strong> know that if x ∈ G 0 (∆, L n (T p ))[p], then<br />
x|T (p) m ∈ Im(β) for sufficiently large m. We shall divide our argument in<strong>to</strong> the following<br />
three cases: case 1: p ∤ u; case 2: p | u and p | c, and case 3: p | u but p ∤ c. Since<br />
c 2 u ≡ 0 mod Np r , c is au<strong>to</strong>matically divisible by p in case 1. We have the following<br />
decompositi<strong>on</strong>:<br />
∆ ( {<br />
)<br />
1 0 ∐<br />
p−1<br />
i=0 ∆( 1 0<br />
0 p<br />
0 p ∆ =<br />
)<br />
π<br />
i<br />
s = ∆ ( 1 0<br />
0 p<br />
)<br />
∆s , in case 1,<br />
∐ p−1<br />
i=0 ∆( 1 i<br />
0 p<br />
)<br />
∆s , in case 2 and 3.<br />
(6.4.3)<br />
The decompositi<strong>on</strong>s follow from the proof of the Propositi<strong>on</strong> 6.4.5, since p | c and hence<br />
α s ∈ Γ 0 (p). For an element x ∈ G 0 (∆, L n (T p )), we let x s denote the comp<strong>on</strong>ent of x in<br />
H i (∆ s , L n (T p )).<br />
For an element x ∈ G 0 (∆, L n (T p ))[p], we would like <strong>to</strong> write down the comp<strong>on</strong>ent<br />
(x|T (p)) s . In order <strong>to</strong> write down this, first we choose γ ∈ ∆ in case 1 and γ i ∈ ∆ in<br />
)<br />
(s) ∈ C(∆) and γi<br />
( 1 i<br />
0 p<br />
)<br />
(s) ∈ C(∆) and we denote them by<br />
cases 2 and 3, so that γ ( 1 0<br />
0 p<br />
ν i (s) and t respectively. From the definiti<strong>on</strong> of acti<strong>on</strong> of T (p) <strong>on</strong> x, we see that<br />
⎧<br />
∑p−1<br />
(γ ( )<br />
1 0<br />
⎪⎨<br />
0 p π<br />
i<br />
s ) ι · x t = (γ ( )<br />
1 0<br />
0 p )ι · x t , in case 1,<br />
i=0<br />
(x|T (p)) s =<br />
∑p−1<br />
(<br />
⎪⎩ (γ 1 i<br />
)<br />
i 0 p )ι · x γi (s), in case 2 and 3.<br />
i=0<br />
(6.4.4)<br />
In case 1, (x|T (p)) s acts <strong>on</strong> p-<strong>to</strong>rsi<strong>on</strong> points by the nilpotent matrix<br />
⎛<br />
⎞<br />
0 0 · · · ∗<br />
0 0 · · · ∗<br />
⎜<br />
⎟<br />
⎝· · · · · · · · · ∗⎠ , (6.4.5)<br />
0 0 · · · 0<br />
and hence (x|T (p) 2 ) s is zero.<br />
81
In case 2, since a is prime <strong>to</strong> p, for any i 0 , i 1 , . . . , i m ∈ Z, we see that<br />
( 1 im<br />
0 p<br />
)( 1 im−1<br />
0 p<br />
) (<br />
· · ·<br />
1 i0<br />
) a + c ∑ m<br />
j=0 pj i j<br />
0 p (s) =<br />
p m c<br />
∈ Q.<br />
Define ν ij := ( 1 i j<br />
)<br />
0 p . The numera<strong>to</strong>r of the above number is prime <strong>to</strong> p, and thus if m ≥ r,<br />
the cusp ν = ν im ◦ ν im−1 ◦ · · · ◦ ν i0 (s) as a cusp of Γ 1 (Np r ) is unramified over X 1 (N) and<br />
hence as a cusp of ∆. Then we have<br />
1. L n (p −1 Z/Z) ∆ν ⊆ {(y, 0) t | y ∈ (p −1 Z/Z) n }. Let α ν = ( )<br />
a b<br />
c d ∈ SL2 (Z) such that<br />
α ν (∞) = ν. Since ν is an unramified regular cusp, we can choose α ν ∈ Γ 0 (p m ) such<br />
that αν<br />
−1 π ν α ν = ( )<br />
1 u<br />
0 1 with (u, p) = 1. If x = (x0 , . . . , x n ) ∈ L n (p −1 Z/Z) ∆ν , then<br />
αν<br />
−1 x is fixed by ( )<br />
1 u<br />
0 1 , which implies that (n + 1)-th co-ordinate of the tuple α<br />
−1<br />
ν x<br />
is zero. Therefore, x n = 0 (here use the fact that α ν ∈ Γ 0 (p m )).<br />
(<br />
2. (γ 1 i<br />
)<br />
i 0 p ) ι x ν = 0, if x ν ∈ L n (p −1 Z/Z) ∆ν = H 0 (∆ ν , L n (T p ))[p]. If γ i = ( )<br />
a b<br />
c d , then<br />
( ) ι<br />
(<br />
(γ 1 i<br />
)<br />
i 0 p ) ι a ia + pb<br />
=<br />
acts <strong>on</strong> a point as above by the matrix<br />
c ci + pd<br />
⎛<br />
⎞<br />
0 0 · · · (ia) n<br />
0 0 · · · ∗<br />
⎜<br />
⎟<br />
⎝· · · · · · · · · ∗ ⎠ . (6.4.6)<br />
0 0 · · · ∗<br />
Therefore, the element (x|T (p) r+1 ) s = 0. In the last case, the proof is similar <strong>to</strong> the proof<br />
of [Hid88a, Prop. 2.3].<br />
6.5 Freeness<br />
In this secti<strong>on</strong>, we prove that the modules V 0 and W 0 are free of finite rank over Z p [[X]],<br />
completing the proof of Theorem 6.3.1. We restate this formally as:<br />
Theorem 6.5.1. Let p = 2, N > 1 be odd, and Λ = Z p [[X]]. The modules V 0 and W 0<br />
are free modules of finite rank over Λ.<br />
A proof of the freeness of W 0 is proved in Appendix A (cf. Theorem A.5.3). The claim<br />
for V 0 is more subtle and requires more machinery and results. We start by recalling<br />
without proof a lemma [Hid86a, Lem. 6.3] which is useful in proving the freeness of V 0 .<br />
Lemma 6.5.2. A compact c<strong>on</strong>tinuous Λ-module M is free of finite rank r over Λ if and<br />
<strong>on</strong>ly if there is a subset I of positive integers and infinitely many elements {P n } n∈I in<br />
Λ such that M[P n ] ≃ T r p, for all n ∈ I, where M is the P<strong>on</strong>tryagin dual of M and<br />
M[P n ] = {m ∈ M|P n .m = 0}.<br />
We know that the group Z × p (recall p = 2) acts <strong>on</strong> V 0 . In particular, µ 2 = (Z/qZ) ×<br />
acts <strong>on</strong> V 0 (recall q = 4). Write<br />
V 0 = V 0 (0) ⊕ V 0 (1),<br />
82
where V 0 (a) = {v ∈ V 0 | v|ζ = ζ a v, for ζ ∈ µ 2 }. Since the acti<strong>on</strong> of Γ commutes with the<br />
acti<strong>on</strong> of µ 2 , V 0 (a) is also a Λ-module, for a = 0, 1. Let V 0 (a) denote the P<strong>on</strong>tryagin<br />
dual of V 0 (a). We shall show V 0 (a) is a free module of rank 2r(a) over Λ, where r(a) is<br />
the rank of the Hecke algebra h 0 2 (Φ 2, ω a , Z p ).<br />
By part (1) of Theorem 6.3.1, we have that V 0 (a)/a 2 V 0 (a) ≃ V2 0(a),<br />
where a 2 is the<br />
augmentati<strong>on</strong> ideal of Z p [[Γ]]. Since V2 0(a)<br />
is a free module of rank 2r(a) over Z p, by<br />
Nakayama’s lemma, we see that V 0 (a) is a finitely generated Λ-module with minimal<br />
number of genera<strong>to</strong>rs 2r(a). Hence there is a surjecti<strong>on</strong> from Λ 2r(a) ↠ V 0 (a). Hence, by<br />
duality, we have<br />
V 0 (a)[P n ] ↩→ T 2r(a)<br />
p ,<br />
where P n is the prime ideal of Λ defined in §6.3. Now define<br />
H 1 (Γ 1 (Np r ), n; Z p /p r Z p ) := {v ∈ H 1 (Γ 1 (Np r ), Z p /p r Z p ) | v|z = z n v for z ∈ Z × p },<br />
H 1 p(Γ 1 (Np r ), n; Z p /p r Z p ) := H 1 (Γ 1 (Np r ), n; Z p /p r Z p ) ∩ H 1 p(Γ 1 (Np r ), Z p /p r Z p ).<br />
Suppose that the following inclusi<strong>on</strong>s and isomorphisms are true for r ≥ 2:<br />
eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /p r Z p ↩→<br />
(1)<br />
eH 1 p(Φ 2 , L n (Z p /p r Z p ))<br />
≃<br />
(2) eH1 p(Φ r , L n (Z p /p r Z p )) ≃ eH 1<br />
(3)<br />
p(Φ r , Z p /p r Z p (n))<br />
↩→<br />
(4)<br />
eH 1 p(Γ 1 (Np r ), n; Z p /p r Z p ) ↩→<br />
(5)<br />
V 0 (a)[P n ],<br />
(6.5.1)<br />
where the last inclusi<strong>on</strong> holds <strong>on</strong>ly if n ≡ a (mod 2). Then we have<br />
eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /p r Z p ↩→ V 0 (a)[P n ] ↩→ T 2r(a)<br />
p .<br />
Taking direct limits with respect <strong>to</strong> r, we have that<br />
eH 1 p(Φ 2 , L n (Z p )) ⊗ T p ↩→ V 0 (a)[P n ] ↩→ T 2r(a)<br />
p . (6.5.2)<br />
In the next secti<strong>on</strong>, we prove that eH 1 p(Φ 2 , L n (Z p )) is Z p -free (cf. Lemma 6.6.2). More<br />
precisely, we prove in Theorem 6.6.1 that:<br />
Theorem 6.5.3. The Z p -rank of the module eH 1 p(Φ 2 , L n (Z p )) is 2r(a), for n ≡ a (mod 2).<br />
Proof. For p ≥ 5, the theorem follows from [Hid86b, Thm. 3.1 and Cor. 3.2], but<br />
their proofs use results from the theory of Katz modular <strong>forms</strong> and the theory of mod p<br />
modular <strong>forms</strong>. For the prime p = 2, we need different arguments <strong>to</strong> prove the theorem<br />
(cf. Theorem 6.6.1) and we postp<strong>on</strong>e the proof <strong>to</strong> the next secti<strong>on</strong>.<br />
We complete the proof of Theorem 6.5.1, assuming Theorem 6.5.3.<br />
Proof. By Theorem 6.5.3, we have that T 2r(a)<br />
p ≃ eH 1 p(Φ 2 , L n (Z p ))⊗T p . Hence, for all n ≡<br />
a (mod 2), we have V 0 (a)[P n ] ≃ Tp 2r(a) . The theorem now follows from Lemma 6.5.2.<br />
Now, we shall show that the inclusi<strong>on</strong>s and isomorphisms in (6.5.1) hold. This is the<br />
c<strong>on</strong>tent of the next few lemmas and propositi<strong>on</strong>s. The following propositi<strong>on</strong> proves the<br />
inclusi<strong>on</strong> (1) in (6.5.1).<br />
83
Propositi<strong>on</strong> 6.5.4. For all r ≥ 1 and n ≥ 0, we have<br />
eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /p r Z p ↩→ eH 1 p(Φ 2 , L n (Z p /p r Z p )).<br />
Proof. For any Z p -module A, the short exact sequence of modules<br />
0 → eH 1 p(Φ 2 , L n (Z p )) → eH 1 (Φ 2 , L n (Z p )) → eH 1 (Φ 2 , L n (Z p ))/eH 1 p(Φ 2 , L n (Z p )) → 0<br />
induces the l<strong>on</strong>g exact sequence<br />
Tor(M, A) → eH 1 p(Φ 2 , L n (Z p )) ⊗ A α → eH 1 (Φ 2 , L n (Z p )) ⊗ A,<br />
where M = eH 1 (Φ 2 , L n (Z p ))/eH 1 p(Φ 2 , L n (Z p )). If M is Z p -free, then the map α is injective.<br />
Now, for any c<strong>on</strong>gruence subgroup Φ, we have:<br />
eH 1 (Φ, L n (Z p )) ⊗ A ∼ → eH 1 (Φ, L n (A)).<br />
Under this identificati<strong>on</strong>, the map α preserves parabolic classes, proving the theorem.<br />
The Z p -freeness of eH 1 (Φ 2 , L n (Z p ))/eH 1 p(Φ 2 , L n (Z p )) follows from Propositi<strong>on</strong> 6.4.6.<br />
The isomorphisms (2) and (3) in (6.5.1) follows from [Hid86a, Cor. 4.5], noting that<br />
the argument given there works for p = 2 and for Φ 2 , instead of p odd and the Φ 1 there.<br />
The following lemma proves the inclusi<strong>on</strong> (4).<br />
Lemma 6.5.5. For r ≥ 2, we have an inclusi<strong>on</strong><br />
eH 1 p(Φ r , Z p /p r Z p (n)) ↩→ eH 1 p(Γ 1 (Np r ), n; Z p /p r Z p ).<br />
Proof. We have the following inflati<strong>on</strong>-restricti<strong>on</strong> sequence for the groups Γ 1 (Np r ) ⊆ Φ r :<br />
0 → H 1 (Φ r /Γ 1 (Np r ), Z p /p r Z p (n)) → H 1 (Φ r , Z p /p r Z p (n))<br />
→ H 1 (Γ 1 (Np r ), Z p /p r Z p (n)) Φr/Γ 1(Np r) .<br />
Since H 1 (Γ 1 (Np r ), Z p /p r Z p (n)) Φr/Γ 1(Np r) ↩→ H 1 (Γ 1 (Np r ), n; Z p /p r Z p ), we have the following<br />
exact sequence<br />
0 → H 1 (Φ r /Γ 1 (Np r ), Z p /p r Z p (n)) → H 1 (Φ r , Z p /p r Z p (n)) → H 1 (Γ 1 (Np r ), n; Z p /p r Z p ).<br />
By [Hid86a, Lem. 6.1], we have the required claim.<br />
The following lemma proves inclusi<strong>on</strong> (5) in (6.5.1).<br />
Lemma 6.5.6. eH 1 p(Γ 1 (Np r ), n; Z p /p r Z p ) ↩→ V 0 (a)[P n ], if n ≡ a (mod 2).<br />
Proof. Observe that, eH 1 p(Γ 1 (Np r ), n; Z p /p r Z p ) is the subspace of eH 1 p(Γ 1 (Np r ), Z p /p r Z p )<br />
<strong>on</strong> which Z × p act by v|z = z n v. The group V 0 (a)[P n ] is also the subspace of V 0 such that<br />
Z × p acts by v|z = z n v, since µ 2 acts by ζ2 n = ζa 2 and γ ∈ Γ acts by v|γ = γn v.<br />
By [Hid86a, p. 584, (5.4)], we have that eH 1 p(Γ 1 (Np r ), Z p /p t Z p ) ≃ eH 1 p(Γ 1 (Np r ), Z p )⊗<br />
Z p /p t Z p . Since tensor product commutes with direct limits, we have that<br />
lim eH<br />
−→ 1 p(Γ 1 (Np r ), Z p /p t Z p ) ≃ eH 1 p(Γ 1 (Np r ), Z p ) ⊗ Zp T p ≃ eH 1 p(Γ 1 (Np r ), T p ).<br />
t<br />
By part (1) of Theorem 6.3.1 with V 0 , we have eH 1 p(Γ 1 (Np r ), Z p /p r Z p ) ↩→ V 0 . Since this<br />
map respects the acti<strong>on</strong> of Z × p , we have the required claim.<br />
84
6.6 C<strong>on</strong>stant rank<br />
In this secti<strong>on</strong>, we prove that the ranks of certain cuspidal ordinary 2-adic Hecke algebras<br />
of different weights are all equal <strong>to</strong> the rank of a weight 2 cuspidal ordinary Hecke algebra.<br />
As in the previous subsecti<strong>on</strong>, p = 2 and N > 1 is odd.<br />
For a = 0 or 1, recall that r(a) is the rank of the Hecke algebra h 0 2 (Φ 2, ω a , Z p ), where<br />
ω denotes the mod 4 cyclo<strong>to</strong>mic character. For simplicity, we write A(ω n ) for the sheaf<br />
with twisted acti<strong>on</strong> L 0 (ω n , A), for any Z p -module A.<br />
Theorem 6.6.1. For each positive integer n ≡ a (mod 2),<br />
rank Zp<br />
h 0 n+2(Φ 2 , Z p ) = r(a).<br />
Before proving this theorem, we need <strong>to</strong> gather some results, which we do now.<br />
Lemma 6.6.2. For r > s ≥ 0, the module eH 1 (Φ s r, L n (Z p )) is Z p -free, for n ≥ 0.<br />
Proof. The short exact sequence<br />
0 → L n (Z p ) → L n (Q p ) → L n (T p ) → 0<br />
induces a l<strong>on</strong>g exact sequence of cohomology groups for the group Φ s r<br />
H 0 (Φ s r, L n (Q p )) α → H 0 (Φ s r, L n (T p )) β → H 1 (Φ s r, L n (Z p )) γ → H 1 (Φ s r, L n (Q p )).<br />
If n = 0, then the map α is surjective, and hence the map β is zero. Therefore, the<br />
map γ is injective, and hence H 1 (Φ s r, Z p ) is Z p -free. Assume that n > 0. If we can show<br />
that eH 0 (Φ s r, L n (T p )) = 0, then the lemma follows. The opera<strong>to</strong>r T p acts <strong>on</strong> L n (T p )<br />
by x|T p = ∑ p−1( 1 −i<br />
) ι<br />
i=0 0 p · x, where A ι = Adj(A). We see that T p acts <strong>on</strong> any element<br />
of H 0 (Φ s r, L n (T p ))[p] by the matrix ( )<br />
0 ∗<br />
0 0 , and hence T<br />
2<br />
p acts trivially <strong>on</strong> such elements,<br />
hence the idempotent e annihilates H 0 (Φ s r, L n (T p )).<br />
Corollary 6.6.3. For any integer n ≥ 0, the module eH 1 (Φ 2 , Z p (ω n )) is Z p -free.<br />
Proof. If n is even, then ω n = 1, hence this follows from the lemma and when n is odd,<br />
the proof is similar <strong>to</strong> the proof of the lemma.<br />
Lemma 6.6.4. eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /qZ p ≃ eH 1 p(Φ 2 , L n (Z p /qZ p )).<br />
Proof. By Propositi<strong>on</strong> 6.5.4 with r = 2, the map<br />
eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /qZ p → eH 1 p(Φ 2 , L n (Z p /qZ p ))<br />
is injective. For the surjectivity, we work with sheaf cohomology instead of group cohomology.<br />
Let Y be the complex open manifold associated with Φ 2 . Observe that we have<br />
the following commutative diagram:<br />
eH 1 c(Y, F (L n (Z p )))/q<br />
eH 1 c(Y, F (L n (Z p /qZ p )))<br />
≀<br />
eH 1 p(Y, F (L n (Z p )))/q <br />
<br />
eH 1 p(Y, F (L n (Z p /qZ p ))) <br />
eH 1 (Y, F (L n (Z p )))/q<br />
eH 1 (Y, F (L n (Z p /qZ p ))).<br />
By [Hid88b, Cor. 2.2], the first vertical map is an isomorphism. As a result, we get that<br />
the middle vertical map is surjective and the lemma follows.<br />
≀<br />
85
Lemma 6.6.5. For any n ≥ 0, eH 1 (Y, F (Z p (ω n )))/q ≃ eH 1 (Y, F (Z p /qZ p (ω n ))).<br />
Proof. The short exact sequence<br />
induces another short exact sequence<br />
0 → Z p (ω n ) q → Z p (ω n ) → Z p /qZ p (ω n ) → 0<br />
0 → H 1 (Y, F (Z p (ω n ))) ⊗ Z p /qZ p → H 1 (Y, F (Z p /qZ p (ω n ))) → H 2 (Y, F (Z p (ω n )))[q] → 0,<br />
where H 2 (Y, F (Z p (ω n )))[q] = {x ∈ H 2 (Y, F (Z p (ω n ))) | q·x = 0}. This last group vanishes,<br />
since the cohomological dimensi<strong>on</strong> of Φ s r is 1.<br />
Propositi<strong>on</strong> 6.6.6. The module e(H 1 (Φ 2 , Z p (ω))/H 1 p(Φ 2 , Z p (ω))) is Z p -free.<br />
Proof. If the module eG 1 (Φ 2 , Z p (ω)) is Z p -free, then the propositi<strong>on</strong> follows, since the<br />
maps in (6.1.1) are invariant under the acti<strong>on</strong> of the Hecke opera<strong>to</strong>rs. C<strong>on</strong>sider the l<strong>on</strong>g<br />
exact sequence of cohomology groups<br />
G 0 (Φ 2 , Q p (ω)) β → G 0 (Φ 2 , T p (ω)) → G 1 (Φ 2 , Z p (ω)) → G 1 (Φ 2 , Q p (ω)),<br />
induced by the short exact sequence 0 → Z p (ω) → Q p (ω) → T p (ω) → 0.<br />
Since the image of β is p-divisible, it is enough <strong>to</strong> show that, ∀ x ∈ G 0 (Φ 2 , T p (ω))[p],<br />
x|T p bel<strong>on</strong>gs <strong>to</strong> β(G 0 (Φ 2 , Q p (ω))). Then a small computati<strong>on</strong> shows that<br />
e(G 0 (Φ 2 , T p (ω))/β(G 0 (Φ 2 , Q p (ω)))) = 0,<br />
and hence eG 1 (Φ 2 , L n (Z p (ω))) is Z p -free. We now prove, for all x ∈ G 0 (Φ 2 , T p (ω))[p], the<br />
element x|T p bel<strong>on</strong>gs <strong>to</strong> β(G 0 (Φ 2 , Q p (ω))).<br />
Since (2, N) = 1 and N ≥ 3, all the cusps of Φ 2 are regular, because irregularity for<br />
Φ 2 implies the irregularity for Γ 1 (N), but there are no irregular cusps for Γ 1 (N). Let<br />
s ∈ C(Φ 2 ) be a cusp of Φ 2 . Let α s = ( )<br />
a b<br />
c d ∈ SL2 (Z) such that α s (∞) = s. If π s denotes<br />
a genera<strong>to</strong>r for (Φ 2 ) s , we may write<br />
(<br />
π s = α 1 u<br />
)<br />
s 0 1 α<br />
−1<br />
s = ( )<br />
1−cau a 2 u<br />
−c 2 u 1+cau ∈ Φ2 , with u ≠ 0. (6.6.1)<br />
The structure of G 0 (Φ 2 , M(ω)) depends <strong>on</strong> the acti<strong>on</strong> π s <strong>on</strong> M. In order <strong>to</strong> study<br />
this, let us divide the cusps in<strong>to</strong> two types. If p | u, then we refer <strong>to</strong> this cusp as being of<br />
type 1, otherwise of type 2. We assume that Φ 2 acts trivially <strong>on</strong> M, because we are <strong>on</strong>ly<br />
interested in the cases when M = Z p , Q p or T p . Let x be an element of G 0 (Φ 2 , T p (ω))[p]<br />
and let x s denote the comp<strong>on</strong>ent of x in H i ((Φ 2 ) s , T p (ω)).<br />
If s is a cusp of type 1, then we see that H 0 ((Φ 2 ) s , M(ω)) = M by (6.6.1) and<br />
moreover the map β s is surjective, where β s : H 0 ((Φ 2 ) s , Q p (ω)) → H 0 ((Φ 2 ) s , T p (ω)).<br />
Hence (x|T p ) s ∈ β(H 0 ((Φ 2 ) s , Q p (ω)) = Q p ).<br />
Suppose s is a cusp of type 2. If π s acts trivially <strong>on</strong> M, then H 0 ((Φ 2 ) s , M(ω)) = M<br />
and if π s does not act trivially <strong>on</strong> M, then H 0 ((Φ 2 ) s , M(ω)) = M[2]. In the former case,<br />
again (x|T p ) s ∈ β(Q p ). In the latter case,<br />
p−1<br />
∑( (<br />
(x|T p ) s = γ 1 0<br />
) )<br />
0 p π<br />
i ι<br />
s · xt ,<br />
i=0<br />
86
where γ ∈ Φ 2 such that t = γ ( )<br />
1 0<br />
0 p (s) ∈ C(Φ2 ). Since x is 2-<strong>to</strong>rsi<strong>on</strong>, we see that<br />
(x|T p ) s = ∑ p−1<br />
i=0 (±1) · x t = ∑ p−1<br />
i=0 x t = 2x t = 0 ∈ β(H 0 ((Φ 2 ) s , Q p (ω))) = 0. Hence we have<br />
that for any x ∈ G 0 (Φ 2 , T p (ω))[p], the element x|T p bel<strong>on</strong>gs <strong>to</strong> β(G 0 (Φ 2 , Q p (ω))).<br />
Remark 6.6.7. In the above proof, we have used the fact that p is 2.<br />
Corollary 6.6.8. eH 1 p(Φ 2 , Z p (ω n ))/q ↩→ eH 1 p(Φ 2 , Z p /qZ p (ω n )).<br />
Proof. When n is even, this follows from Propositi<strong>on</strong> 6.5.4 with r = 2. When n is odd,<br />
the injectivity of the first vertical map follows from the following diagram<br />
eH 1 p(Y, F (Z p (ω)))/q α <br />
<br />
eH 1 (Y, F (Z p (ω)))/q<br />
eH 1 p(Y, F (Z p /qZ p (ω))) <br />
eH 1 (Y, F (Z p /qZ p (ω))),<br />
≀<br />
since α is injective by Propositi<strong>on</strong> 6.6.6, and the sec<strong>on</strong>d vertical map is an isomorphism<br />
by Lemma 6.6.5.<br />
Now we shall give a proof Theorem 6.6.1.<br />
Proof. It is enough <strong>to</strong> prove that the Z p -rank of eH 1 p(Φ 2 , L n (Z p )) is the same as the Z p -<br />
rank of eH 1 p(Φ 2 , Z p (ω a )) (the modules are Z p -free by Lemma 6.6.2 and by its corollary).<br />
By (6.5.2), we see that the rank of eH 1 p(Φ 2 , L n (Z p )) is less than or equal <strong>to</strong> the rank of<br />
eH 1 p(Φ 2 , Z p (ω a )).<br />
Again by Lemma 6.6.2 and by its corollary, it is enough <strong>to</strong> show the Z p /qZ p -rank of the<br />
module eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /qZ p is greater than or equal <strong>to</strong> that of eH 1 p(Φ 2 , Z p (ω a )) ⊗<br />
Z p /qZ p . We have the following<br />
eH 1 p(Φ 2 , L n (Z p )) ⊗ Z p /qZ p ≃<br />
(1)<br />
eH 1 p(Φ 2 , L n (Z p /qZ p ))<br />
≃<br />
(2) eH1 p(Φ 2 , Z p /qZ p (ω a ))<br />
←↪<br />
(3) eH1 p(Φ 2 , Z p (ω a )) ⊗ Z p /qZ p ,<br />
(6.6.2)<br />
where the isomorphisms (1), (2) and the inclusi<strong>on</strong> (3) follow from Lemma 6.6.4, the<br />
isomorphism (3) in (6.5.1) with r = 2, and Corollary 6.6.8, respectively. Hence the<br />
theorem is proved.<br />
6.7 Λ-adic Hecke algebras<br />
Recall that our aim is <strong>to</strong> prove a c<strong>on</strong>trol theorem for Hida’s ordinary Hecke algebra, which<br />
we now introduce.<br />
For each subalgebra A of C or C p , and for any r ≥ s ≥ 1, we have a commutative<br />
diagram for all n:<br />
S k (Γ 1 (Np s ), A) S k (Γ 1 (Np r ), A)<br />
T n<br />
S k (Γ 1 (Np s ), A)<br />
T n<br />
S k (Γ 1 (Np r ), A).<br />
87
Thus, we have a A-algebra homomorphism, h k (Γ 1 (Np r ), A) ↠ h k (Γ 1 (Np s ), A) and since<br />
T p ↦→ T p , we have that h 0 k (Γ 1(Np r ), A) → h 0 k (Γ 1(Np s ), A), for each r ≥ s ≥ 1. Now, set:<br />
h k (Γ 1 (Np ∞ ), A) := lim h ←−r k (Γ 1 (Np r ), A), h 0 k (Γ 1(Np ∞ ), A) := lim h 0 ←−r k (Γ 1(Np r ), A),<br />
S k (Np ∞ , A) := ∪ ∞ r=1S k (Γ 1 (Np r ), A).<br />
In Lemma 6.7.2 below, we show that, for weights k 1 ≥ k 2 ≥ 2, there is a surjecti<strong>on</strong><br />
h k1 (Γ 1 (Np ∞ ), A) ↠ h k2 (Γ 1 (Np ∞ ), A), and hence <strong>on</strong> the ordinary parts. Before we state<br />
it, we need <strong>to</strong> define a pairing between certain Hecke algebras and certain spaces of<br />
modular <strong>forms</strong>. Recall that K/Q p is finite, and O K is the integral closure of Z p in K.<br />
Put<br />
S k (Np r , K/O K ) = S k (Γ 1 (Np r ), K)/S k (Γ 1 (Np r ), O K ).<br />
By definiti<strong>on</strong>, <strong>on</strong>e can embed this space via q-expansi<strong>on</strong> in<strong>to</strong> the module of formal series<br />
K/O K [[q]]. We take the injective limit:<br />
S k (Np ∞ , K/O K ) = lim S −→r k (Np r , K/O K ) → K/O K [[q]].<br />
Then S k (Np ∞ , K/O K ) ≃ S k (Np ∞ , K)/S k (Np ∞ , O K ). The algebra h k (Γ 1 (Np ∞ ), O K )<br />
acts <strong>on</strong> S k (Np ∞ , K/O K ). Define the pairing<br />
(, ) : h k (Γ 1 (Np ∞ ), O K ) × S k (Np ∞ , K/O K ) → K/O K ,<br />
by (h, f) = a(1, f|h). Then (h, f|g) = (hg, f), for all h, g ∈ h k (Γ 1 (Np ∞ ), O K ). Equip<br />
S k (Np ∞ , K/O K ) with the discrete <strong>to</strong>pology. We have (cf. [Hid86a, Lem. 7.1]):<br />
Lemma 6.7.1. The pairing above shows that h k (Γ 1 (Np r ), Z p ) and S k (Γ 1 (Np r ), T p ) (respectively,<br />
h 0 k (Γ 1(Np r ), Z p ) and S 0 k (Γ 1(Np r ), T p )), for r = 1, 2, . . . , ∞, are P<strong>on</strong>tryagin<br />
duals.<br />
Lemma 6.7.2. For k 1 ≥ k 2 ≥ 2, there exists a surjecti<strong>on</strong><br />
h k1 (Γ 1 (Np ∞ ), O K ) ↠ h k2 (Γ 1 (Np ∞ ), O K ).<br />
Proof. The proof is similar <strong>to</strong> the proof of [Hid86a, Lem. 7.2]. For p = 2, we need <strong>to</strong> work<br />
with a different Eisenstein series than the <strong>on</strong>e given in that lemma. For r ≥ 2, define a<br />
formal q-expansi<strong>on</strong> for each t ∈ (Z/p r Z) × by<br />
⎛<br />
⎞<br />
G(r, t) = −t 0 p −r + 1 ∞ 2 + ∑ ∑<br />
⎝<br />
sgn(d) ⎠ q n ,<br />
n=1<br />
d|n, d≡t (mod p r )<br />
where t 0 is an integer satisfying 0 ≤ t 0 < p r and t 0 ≡ t mod p r . Then, as shown by Hecke,<br />
G(r, t) gives the q-expansi<strong>on</strong> of an element of M 1 (Γ 1 (Np r ), Q) and satisfies<br />
G(r, t)| 1 = G(r, at) for ( a b<br />
c d<br />
)<br />
∈ Γ0 (Np r ).<br />
Put E(r, t) = −p r G(r, t). For odd primes p, the c<strong>on</strong>gruence E(r, t) ≡ t (mod p r ) holds.<br />
For the prime p = 2, the c<strong>on</strong>gruence that holds is E(r, t) ≡ t (mod p r−1 ). Multiplicati<strong>on</strong><br />
by the Eisenstein series E(r, 1) gives an injective morphism<br />
ι r : S k−1 (Np ∞ , T p )[p r−1 ] → S k (Np ∞ , T p )[p r−1 ].<br />
Using the injective limit of the maps ι r and Lemma 6.7.1, we can finish the proof of the<br />
lemma al<strong>on</strong>g the lines of the proof of [Hid86a, Lem. 7.2].<br />
88
It is known, by [Hid88b, Thm. 3.2], that the map in the lemma above is an isomorphism.<br />
This theorem is stated adelically, but includes the case of p = 2. Thus, the Hecke<br />
algebra h 0 k (Γ 1(Np ∞ ), O K ) is independent of the weight, for all k ≥ 2, and we denote this<br />
by h 0 (N, O K ).<br />
6.8 C<strong>on</strong>trol theorem for ordinary Hecke algebras<br />
In this secti<strong>on</strong>, we prove a c<strong>on</strong>trol theorem for Hida’s ordinary Hecke algebras for the<br />
prime p = 2.<br />
Recall that K is a finite extensi<strong>on</strong> of Q p and O K is integral closure of Z p in K. Let ɛ<br />
be a character of Γ/Γ r with values in O K , with r ≥ 2. In this secti<strong>on</strong>, we write Λ for Λ K<br />
and Q(Λ) for the field of fracti<strong>on</strong>s of Λ K .<br />
We know that h 0 (N, O K ) acts <strong>on</strong> the finite free Λ-module V 0 . Hence the Λ-module<br />
h 0 (N, O K ) is finitely generated and <strong>to</strong>rsi<strong>on</strong>-free, since the acti<strong>on</strong> is faithful <strong>on</strong> V 0 . By<br />
abuse of notati<strong>on</strong>, let P k,ɛ also denote the prime ideal generated by the prime element<br />
P k,ɛ = ι(u) − ɛ(u)u k . By the independence of weight of h 0 (N, O K ), there is a surjective<br />
homomorphisms of O K -algebras, respectively, of Λ-algebras:<br />
ρ : h 0 (N, O K ) ↠ h 0 k (Φ2 r, ɛ, O K ) and Λ Pk,ɛ ↠ Λ Pk,ɛ /P k,ɛ Λ Pk,ɛ = K, (6.8.1)<br />
where Λ Pk,ɛ /P k,ɛ Λ Pk,ɛ is identified with K with ι(u) corresp<strong>on</strong>ding <strong>to</strong> u k ɛ(u), inducing the<br />
map<br />
˜ρ k,ɛ : h 0 (N, O K ) ⊗ Λ Λ Pk,ɛ ↠ h 0 k (Φ2 r, ɛ, O K ) ⊗ OK K,<br />
which in turn fac<strong>to</strong>rs via P k,ɛ <strong>to</strong> give the map:<br />
Theorem 6.8.1. The natural map<br />
is an isomorphism.<br />
ρ k,ɛ : h 0 (N, O K ) ⊗ Λ Λ Pk,ɛ /P k,ɛ ↠ h 0 k (Φ2 r, ɛ, K).<br />
ρ k,ɛ : h 0 (N, O K ) ⊗ Λ Λ Pk,ɛ /P k,ɛ ↠ h 0 k (Φ2 r, ɛ, K)<br />
Proof. Since the module h 0 (N, O K ) is finitely generated <strong>to</strong>rsi<strong>on</strong>-free over Λ, then so is<br />
h 0 (N, O K ) Pk,ɛ over Λ Pk,ɛ . Therefore, the module h 0 (N, O K ) Pk,ɛ is free, since any finitely<br />
generated <strong>to</strong>rsi<strong>on</strong>-free module over a discrete valuati<strong>on</strong> ring is free. Let S(k, ɛ) (resp.,<br />
R(k, ɛ)) denote the rank of h 0 (N, O K ) Pk,ɛ (resp., of h 0 k (Φ2 r, ɛ, K)). A priori, the number<br />
S(k, ɛ) depends <strong>on</strong> k and ɛ. Since<br />
h 0 (N, O K ) Pk,ɛ ⊗ ΛPk,ɛ Q(Λ) ≃ h 0 (N, O K ) ⊗ Λ Q(Λ),<br />
we see that S(k, ɛ) is independent of k and ɛ and we denote this comm<strong>on</strong> value by R.<br />
We first prove the theorem for weights k > 2, by assuming that it holds for k = 2.<br />
The Eisenstein series E(2, 1) (cf. §6.7) has the property that E(2, 1) ≡ 1 (mod 2).<br />
Multiplicati<strong>on</strong> by E(2, 1) k−2 induces an injecti<strong>on</strong><br />
S 0 2(Γ 1 (Np r ), T p )[p] → S 0 k (Γ 1(Np r ), T p )[p].<br />
89
By duality, we have a surjecti<strong>on</strong><br />
h 0 k (Γ 1(Np r ), Z p ) ⊗ Z p /pZ p ↠ h 0 2(Γ 1 (Np r ), Z p ) ⊗ Z p /pZ p .<br />
Then<br />
R[Γ : Γ r ] ≥ ∑ ɛ<br />
R(k, ɛ) = rank OK (h 0 k (Γ 1(Np r ), O K ))<br />
≥ rank OK (h 0 2(Γ 1 (Np r ), O K )) = R[Γ : Γ r ],<br />
where the last equality follows by assumpti<strong>on</strong>. This can happen <strong>on</strong>ly if R(k, ɛ) = R for<br />
all k, ɛ, showing that the map ρ k,ɛ is an isomorphism.<br />
Now, we shall prove the result for k = 2. By Theorem 6.3.1, we have that the Z p -rank<br />
of eH 1 p(Γ 1 (Np r ), Z p ) is equal <strong>to</strong> 2[Γ : Γ r ] rank Zp h 0 2 (Γ 1(Nq), Z p ). Hence,<br />
rank Zp h 0 2(Γ 1 (Np r ), Z p ) = [Γ : Γ r ] rank Zp h 0 2(Γ 1 (Nq), Z p ).<br />
Since h 0 2 (Γ 1(Np r ), K) = ⊕ ɛ h 0 2 (Φ2 r, ɛ, K), the left hand side of the equality above is also<br />
∑<br />
ɛ R(2, ɛ). If rank Z p<br />
h 0 2 (Γ 1(Nq), Z p ) = R, then [Γ : Γ r ]R = ∑ ɛ<br />
R(2, ɛ). Since R ≥<br />
R(2, ɛ), we get R = R(2, ɛ), for each ɛ, as desired. Thus, we need <strong>to</strong> show that R =<br />
rank Zp h 0 2 (Γ 1(Nq), Z p ). This is proved in Theorem 6.8.3 below.<br />
The following lemma is well-known; for the proof refer <strong>to</strong> [Hid86a, Lem. 6.4].<br />
Lemma 6.8.2. For any subfield K of C or C p , H 1 p(Γ 1 (M), L n (K)) is free of rank 2 over<br />
the Hecke algebra h n+2 (Γ 1 (M), K) for each positive integer M.<br />
Set ɛ := ( )<br />
1 0<br />
0 −1 . Let M be a module over the Hecke algebra h<br />
0<br />
2 (Γ 1 (Np r ), Z). Let<br />
M ± denote the subspaces of M defined by {m ± [ɛ]m | m ∈ M}. Since ɛ normalizes<br />
Γ 1 (Np r ), the acti<strong>on</strong> of [ɛ] = [Γ 1 (Np r )ɛΓ 1 (Np r )] commutes with that of h 0 2 (Γ 1(Np r ), Z)<br />
<strong>on</strong> M. Therefore, the modules M ± are stable under the acti<strong>on</strong> of h 0 2 (Γ 1(Np r ), Z). For<br />
simplicity, we write h 0 2 (N, Z p) for the weight-2 Λ-adic Hecke algebra h 0 2 (Γ 1(Np ∞ ), Z p ).<br />
Theorem 6.8.3. The surjective map<br />
is an isomorphism.<br />
ρ 2,triv : h 0 2(N, Z p ) ⊗ Λ Λ P2 /P 2 ↠ h 0 2(Γ 1 (Nq), Q p )<br />
Proof. By Theorem 6.3.1, we have (V 0 ) Γ 2<br />
= V 0 2 , i.e., V0 [P 2 ] = V 0 [ω 2,2 ] = eH 1 (X 2 , T p ) =<br />
eH 1 p(Γ 1 (Nq), T p ), where the last equality follows from [Hid86a, p. 583 (5.3)]. Again by<br />
the same theorem, we have<br />
V 0 /P 2 V 0 ≃ Hom Zp (eH 1 p(Γ 1 (Nq), Z p ), Z p ). (6.8.2)<br />
Since V 0 is direct limit over V 0 r , we see that [ɛ] acts <strong>on</strong> V 0 and the acti<strong>on</strong> commutes with<br />
that of h 0 2 (N, Z p). There is a map V 0+ ⊕ V 0− → V 0 , which is an isomorphism if p is odd.<br />
Since p = 2, we tensor this with Λ P2 <strong>to</strong> get an isomorphism (V 0+ ) P2 ⊕ (V 0− ) P2 ≃ V 0 P 2<br />
.<br />
Let V 0± denote the P<strong>on</strong>tryagin dual of V 0± . Then (V 0+ ) P2 ⊕ (V 0− ) P2 ≃ V 0 P 2<br />
. We can<br />
think of h 0 2 (N, Z p) P2 as a subalgebra of the endomorphism algebra of (V 0+ ) P2 and hence<br />
we shall restrict ourselves <strong>to</strong> the module (V 0+ ) P2 . We now prove that<br />
(V 0+ ) P2 /P 2 (V 0+ ) P2 = h 0 2(Γ 1 (Nq), Q p ).<br />
90
We remark that since p = 2 and we work with (V 0+ ) P2 , the above isomorphism is with<br />
Q p -coefficients, otherwise we would have worked with V 0+ and the above isomorphism<br />
would have been with Z p -coefficients. Since the func<strong>to</strong>r Hom Zp (−, T p ) commutes with<br />
the ±-acti<strong>on</strong> after tensoring with Λ P2 , we see that V 0± ⊗ Λ Λ P2 ≃ (VP 0 2<br />
) ± holds. Hence<br />
(V 0± ) P2 /P 2 (V 0± ) P2 ≃ (VP 0 2<br />
) ± /P 2 (VP 0 2<br />
) ± ≃ (VP 0 2<br />
/P 2 VP 0 2<br />
) ± , where the last isomorphism is<br />
an easy check. We have that<br />
(V 0+ ) P2 /P 2 (V 0+ ) P2 = (V 0 P 2<br />
/P 2 V 0 P 2<br />
) +<br />
= (Hom Z p<br />
(eH 1 p(Γ 1 (Nq), Z p ), Z p ) ⊗ Zp Q p ) +<br />
(6.8.2)<br />
= Hom Qp (eH 1 p(Γ 1 (Nq), Q p ) + , Q p ) = h 0 2(Γ 1 (Nq), Q p ),<br />
(6.8.3)<br />
where the last equality follows from Lemma 6.8.2 and the third equality follows from the<br />
fact that for any Q p -module M, Hom Qp (M, Q p ) ± ≃ Hom Qp (M ± , Q p ).<br />
Let v denote the vec<strong>to</strong>r in (V 0+ ) P2 corresp<strong>on</strong>ding <strong>to</strong> 1 in h 0 2 (Γ 1(Nq), Q p ) in (6.8.3).<br />
Therefore, we have a map h 0 2 (N, Z p) P2 → (V 0+ ) P2 defined by mapping h → hv. This<br />
map is a surjective map by Nakayama’s lemma and by (6.8.3). The map is injective<br />
since the Hecke acti<strong>on</strong> is faithful <strong>on</strong> (V 0+ ) P2 . Therefore, we have h 0 2 (N, Z p) P2 ≃ (V 0+ ) P2 .<br />
Tensoring this isomorphism with Λ P2 /P 2 and using (6.8.3), we obtain the theorem.<br />
6.9 Uniqueness<br />
In this secti<strong>on</strong>, we prove a uniqueness result for Hida families. Suppose f is a p-stabilized<br />
ordinary newform (or an eigenform). Let P f denote the unique height <strong>on</strong>e prime ideal,<br />
induced by f, via the isomorphism in Theorem 6.8.1. Suppose Q = P f lies over the prime<br />
ideal P = P k,ɛ , where the integer k and the character ɛ depend <strong>on</strong> f.<br />
First we show that, for the prime P of Λ, the <strong>local</strong>ized Hecke algebra h 0 (N, O K ) Q is<br />
étale over Λ Pk,ɛ . We deduce the uniqueness result as a c<strong>on</strong>sequence. For simplicity, let us<br />
denote h 0 (N, O K ) by h 0 (N).<br />
Propositi<strong>on</strong> 6.9.1. The <strong>local</strong>ized Hecke algebra h 0 (N) Q is étale over Λ P and Qh 0 (N) Q =<br />
P h 0 (N) Q , i.e., h 0 (N) Q is a regular <strong>local</strong> ring.<br />
Proof. We apply [Nek06, Lem. 12.7.6], with A = Λ, B = h 0 (N) and J = 0 (and also by<br />
switching the roles of P and Q). The first c<strong>on</strong>diti<strong>on</strong> of that lemma, namely the Hecke<br />
algebra h 0 (N) is finitely generated and <strong>to</strong>rsi<strong>on</strong>-free over Λ follows, as menti<strong>on</strong>ed earlier,<br />
from Theorem 6.3.1. By Theorem 6.8.1, the short exact sequence in the sec<strong>on</strong>d part of<br />
that lemma reduces <strong>to</strong><br />
0 → P → h 0 k (Φ2 r, ɛ, K) α → Q p (a n (f)) ∞ n=1 → 0,<br />
where the map α is given by T n → a n (f). By analyzing the proof of that lemma, we see<br />
that if P P = 0, where P P denotes the <strong>local</strong>isati<strong>on</strong>, then the propositi<strong>on</strong> follows. From the<br />
theory of new<strong>forms</strong>, <strong>on</strong>e knows that h 0 k (Φ2 r, ɛ, K) P<br />
∼ → Qp (a n (f)) ∞ n=1 , hence P P = 0.<br />
Remark 6.9.2. The étaleness of h 0 (N) new<br />
P k,ɛ<br />
over Λ Pk,ɛ for the N-new part of Hida’s Hecke<br />
algebra would follow from a c<strong>on</strong>trol theorem for h 0 (N) new instead of h 0 (N) if <strong>on</strong>e knew<br />
such a result. To prove the étaleness, it is enough <strong>to</strong> show Ω h 0 (N) new<br />
P /Λ P = 0. Let K P be the<br />
91
esidue field of Λ P . By Nakayama’s lemma it suffices <strong>to</strong> show that Ω h 0 (N) new<br />
P /Λ ⊗ P Λ P<br />
K P =<br />
Ω h 0 (N) new<br />
P ⊗K P /K P<br />
= 0. The vanishing of the Kähler differentials follows from the fact that<br />
P is an arithmetic point and by the c<strong>on</strong>trol theorem with “h 0 (N) new ”, we have that<br />
h 0 (N) new<br />
P<br />
⊗ K P is a classical new Hecke algebra acting <strong>on</strong> the space of classical new<strong>forms</strong><br />
of fixed level, weight and character. Such Hecke algebras are well-known <strong>to</strong> be semisimple,<br />
since they are generated by Hecke opera<strong>to</strong>rs away from Np.<br />
6.9.1 2-adic Λ-adic <strong>forms</strong><br />
Now we recall the definiti<strong>on</strong> of a 2-adic Λ = Z 2 [[X]]-adic form. Let I denote the integral<br />
closure of Λ in a finite extensi<strong>on</strong> L of Q(Λ). Let ζ denote a 2 r−2 -th root of unity in ¯Q 2 , with<br />
r ≥ 2, and let k ≥ 1 be a positive integer. The assignment X ↦→ ζ(1 + q) k − 1 yields a Z 2 -<br />
algebra homomorphism ϕ k,ζ : Λ → ¯Q 2 . We say that a height <strong>on</strong>e prime P ∈ Spec(I)( ¯Q 2 )<br />
has weight k, if the corresp<strong>on</strong>ding Λ-algebra homomorphism P : I → ¯Q 2 extends ϕ k,ζ <strong>on</strong><br />
Λ, for some k ≥ 1 and for some ζ. In additi<strong>on</strong>, we say that P is arithmetic, if P has<br />
weight k ≥ 2.<br />
From now <strong>on</strong>, p = 2, q = 4 and (2, N) = 1. We will need the following Dirichlet<br />
characters. Let<br />
• ψ be a Dirichlet character of level Nq,<br />
• ω be the mod 4 cyclo<strong>to</strong>mic character,<br />
• ɛ be the character χ ζ mod 2 r for each root of unity ζ of order 2 r−2 with r ≥ 2<br />
defined by first decomposing<br />
(Z p /2 r Z p ) × = (Z p /qZ p ) × × Z/2 r−2 ,<br />
where the sec<strong>on</strong>d fac<strong>to</strong>r is generated by 1 + q, and then by setting<br />
χ ζ = 1 <strong>on</strong> (Z p /qZ p ) × and χ ζ (1 + q) = ζ.<br />
Definiti<strong>on</strong> 6.9.3. Let F = ∑ ∞<br />
n=1 a(n, F)qn ∈ I[[q]] be a formal q-expansi<strong>on</strong>. We say F is<br />
a Λ-adic form of tame level N and character ψ if for each arithmetic point P ∈ Spec(I)( ¯Q p )<br />
lying over ϕ k,ζ , with k ≥ 2 and ζ of order p r−2 , r ≥ 2, the specializati<strong>on</strong><br />
P (F) ∈ ¯Q p [[q]]<br />
of F at P is the q-expansi<strong>on</strong> of a classical cusp form f ∈ S k (N2 r , χ), where χ = ψω −k χ ζ .<br />
Now, we briefly recall various noti<strong>on</strong>s, e.g., primitive, ordinary, for Λ-adic <strong>forms</strong>.<br />
Definiti<strong>on</strong> 6.9.4. We say that F is a Λ-adic eigenform of level N, if every arithmetic<br />
specializati<strong>on</strong> f = P (F) is an eigenvec<strong>to</strong>r for the classical Hecke opera<strong>to</strong>rs T l for all<br />
primes l with l ∤ Np, and for U l if l | Np.<br />
Let f ∈ S k (Np ∞ , χ) be any eigenvec<strong>to</strong>r for all the opera<strong>to</strong>rs T l with l ∤ Np. Atkin-<br />
Lehner theory implies that <strong>on</strong>e can associate <strong>to</strong> f a unique primitive form f ∗ of minimal<br />
level dividing Np ∞ which has the same eigenvalues as f for almost all primes l.<br />
92
Definiti<strong>on</strong> 6.9.5. We say that a Λ-adic eigenform F of level N is a newform of level N,<br />
if for every arithmetic specializati<strong>on</strong> f = P (F) ∈ S k (Np ∞ , χ), the associated primitive<br />
form f ∗ has level divisible by N.<br />
We say that F is normalized, if a(1, F) = 1. Finally, let us say that F is primitive<br />
of level N, if it is normalized and is a newform of level N.<br />
There is a well-known theory of Λ-adic <strong>forms</strong> due <strong>to</strong> Hida under the assumpti<strong>on</strong> of<br />
ordinariness, which we describe now. Let f be a normalized eigenform for all the Hecke<br />
opera<strong>to</strong>rs of level Np r , with (N, p) = 1 and r ≥ 2, and weight k ≥ 2. Assume that f ∗<br />
has level divisible by N. We say that f is ℘-stabilized (or p-stabilized), if either f is<br />
p-new and is ℘-ordinary, or f is p-old, and is obtained from a primitive ℘-ordinary form<br />
f 0 of level N by the formula f(z) = f 0 (z) − βf 0 (pz), where β is the n<strong>on</strong> ℘-adic-unit root<br />
of E p (x) = x 2 − a(p, f 0 )x + χ(p)p k−1 . The cusp form f is an eigenform of all the Hecke<br />
opera<strong>to</strong>rs of level Np, with U p eigenvalue equal <strong>to</strong> α, the unique ℘-adic unit root of E p (x).<br />
Definiti<strong>on</strong> 6.9.6. We say that a primitive Λ-adic form of level N is ℘-ordinary, if each<br />
arithmetic specializati<strong>on</strong> P (F ) is a ℘-stabilized form in S k (N, χ), where k, N, χ are the<br />
weight, level, and character of P respectively.<br />
6.9.2 Tame level N = 1<br />
We remark that there is no 2-adic Hida theory, when N = 1, showing that our assumpti<strong>on</strong><br />
that N > 1 in several previous secti<strong>on</strong>s loses no generality. Indeed, we have that:<br />
Propositi<strong>on</strong> 6.9.7. There are no ordinary Λ-adic eigen<strong>forms</strong> of tame level 1.<br />
Proof. If such a Λ-adic eigenform were <strong>to</strong> exist, then for every integer k ≥ 2 and r ≥ 2,<br />
its’ specializati<strong>on</strong> at P k,ζ , where ζ is a p r−2 -th root of unity, would be an element of<br />
S k (p r , ω a−k χ ζ ), for some a ∈ N. For parity reas<strong>on</strong>s, (−1) a−k = (−1) k , hence a is even.<br />
But, if r = 2 and k is even, then are no 2-ordinary, 2-stabilized eigen<strong>forms</strong> in S k (4, triv).<br />
For new<strong>forms</strong> this follows from [Miy89, Thm. 4.6.17] and for old<strong>forms</strong> from loc. cit. and<br />
the fact that X 0 (2) has genus 0, and from Hatada [Hat79]. Since there are no ordinary<br />
specializati<strong>on</strong>s in even weight, there are no ordinary Λ-adic eigen<strong>forms</strong> of tame level 1.<br />
Corollary 6.9.8. For odd integers k, the space eSk<br />
2-new (4, ω) is zero.<br />
Proof. This follows immediately from the propositi<strong>on</strong> noting that every 2-ordinary eigenform<br />
in the above space must live in a 2-ordinary Hida family of tame level 1.<br />
Remark 6.9.9. It can be checked independently that the dimensi<strong>on</strong>s of eSk<br />
2-new (4, ω) for<br />
k = 3, 5, 7, 9, 11, 13, 15, 17, are indeed all zero, whereas the dimensi<strong>on</strong>s of Sk 2-new (4, ω)<br />
for k = 3, 5, 7, 9, 11, 13, 15, 17 are 0, 1, 2, . . . , 7, respectively.<br />
6.9.3 Uniqueness result<br />
We now turn <strong>to</strong> the uniqueness result for 2-adic families.<br />
Theorem 6.9.10. Any p-stabilized eigenform is an arithmetic specializati<strong>on</strong> of a unique<br />
Hida family, up <strong>to</strong> Galois c<strong>on</strong>jugacy.<br />
93
Proof. By (6.8.1), we know that any p-stabilized eigenform lives in a Hida family. We<br />
want <strong>to</strong> show that such a family is unique, up <strong>to</strong> Galois c<strong>on</strong>jugacy (this last caveat is<br />
necessary since if a form lies in F by specializati<strong>on</strong> under P : I → ¯Q p , then it also lies in<br />
the c<strong>on</strong>jugate family F σ , by specializing under σ −1 ◦ P : I σ → ¯Q p . Note that F and F σ<br />
corresp<strong>on</strong>d <strong>to</strong> the same minimal prime ideal of h 0 (N)).<br />
Assume the c<strong>on</strong>trary. Let λ 1 and λ 2 denote the algebra homomorphisms from h 0 (N)<br />
<strong>to</strong> I and I ′ respectively, where I, I ′ are finite integral extensi<strong>on</strong>s of Λ. Let P 1 and<br />
P 2 denote the minimal prime ideals of h 0 (N) which are the respective kernels of these<br />
homomorphisms. Since λ 1 and λ 2 have <strong>on</strong>e arithmetic specializati<strong>on</strong> in comm<strong>on</strong>, there<br />
are two algebra homomorphisms P : I → ¯Q p and P ′ : I ′ → ¯Q p such that P ◦ λ 1 =<br />
P ′ ◦ λ 2 = λ P,P ′, say. Then the kernel of λ P,P ′ is a height <strong>on</strong>e prime of h 0 (N), denote by<br />
Q, c<strong>on</strong>taining both P 1 and P 2 and lying over P = P k,ζ for some k ≥ 2, ζ.<br />
By Propositi<strong>on</strong> 6.9.1, h 0 (N) Q is a regular <strong>local</strong> ring. But, a regular a <strong>local</strong> ring is a<br />
domain, hence the prime ideals P 1 and P 2 have <strong>to</strong> be equal.<br />
As an applicati<strong>on</strong> of the last result we now show that the noti<strong>on</strong> of CM-ness is pure<br />
with respect <strong>to</strong> families.<br />
Propositi<strong>on</strong> 6.9.11. Let F be a primitive 2-adic Hida family. Then either all arithmetic<br />
specializati<strong>on</strong>s are CM <strong>forms</strong> or no arithmetic specializati<strong>on</strong> is a CM form.<br />
Proof. The proof is the same as for odd prime p, <strong>on</strong>ce <strong>on</strong>e has the uniqueness result for<br />
p = 2. Indeed a CM family is defined <strong>to</strong> be <strong>on</strong>e which is obtained as the theta series<br />
of a Λ-adic Hecke character of an imaginary quadratic field. Clearly all its arithmetic<br />
specializati<strong>on</strong>s are CM <strong>forms</strong>. Now start with an arbitrary CM form. Assume it lives in<br />
a n<strong>on</strong>-CM family (<strong>on</strong>e which is not a theta series). Then explicit interpolati<strong>on</strong> allows us<br />
<strong>to</strong> also c<strong>on</strong>struct a CM family passing through this CM form. Clearly the n<strong>on</strong>-CM family<br />
and the CM family are not Galois c<strong>on</strong>jugate, and this c<strong>on</strong>tradicts uniqueness.<br />
In view of this result from now <strong>on</strong> we may and do speak of CM and n<strong>on</strong>-CM 2-adic<br />
Hida families.<br />
6.10 An applicati<strong>on</strong> <strong>to</strong> Galois representati<strong>on</strong>s<br />
In [GV04], the splitting of the <strong>local</strong> Galois representati<strong>on</strong>s associated <strong>to</strong> ordinary eigen<strong>forms</strong><br />
was studied for odd primes p. We carry out the same analysis for the case of p = 2,<br />
assuming that the relevant result of Buzzard c<strong>on</strong>tinues <strong>to</strong> hold for p = 2 in the residually<br />
dihedral setting. That is, under this assumpti<strong>on</strong>, we prove that in a n<strong>on</strong>-CM 2-adic Hida<br />
family, all arithmetic specializati<strong>on</strong>s have n<strong>on</strong>-split <strong>local</strong> Galois representati<strong>on</strong>, except<br />
for a possible finite set of excepti<strong>on</strong>s. As a c<strong>on</strong>sequence of the uniqueness result proved<br />
earlier, we are also able <strong>to</strong> exclude CM <strong>forms</strong> from this finite excepti<strong>on</strong>al set.<br />
Recall p denotes the prime 2 and q = 4. We recall some preliminaries <strong>on</strong> ordinary<br />
eigen<strong>forms</strong> and their associated Galois representati<strong>on</strong>s.<br />
Let f = ∑ ∞<br />
n=1 a n(f)q n be a normalized eigenform (i.e., a primitive form) of weight<br />
k ≥ 2, level Np r ≥ 1, and nebentypus χ. Let ℘ be the prime of ¯Q determined by a fixed<br />
embedding of ¯Q in<strong>to</strong> ¯Q p . Let ℘ also denote the induced prime of K f = Q(a n (f)), the<br />
94
Hecke field of f, and let K f,℘ denote the completi<strong>on</strong> of K f at ℘. There is a global Galois<br />
representati<strong>on</strong><br />
ρ f = ρ f,℘ : Gal( ¯Q/Q) → GL 2 (K f,℘ ),<br />
associated <strong>to</strong> f (and ℘) which has the property that for all primes l ∤ Np,<br />
trace(ρ f (Frob l )) = a l (f) and det(ρ f (Frob l )) = χ(l)l k−1 .<br />
Recall f is ordinary at ℘ (or ℘-ordinary), if a p (f) is ℘-adic unit. If f is ordinary at ℘,<br />
then the result of Wiles [Wil88] shows that the restricti<strong>on</strong> of ρ f <strong>to</strong> the decompositi<strong>on</strong><br />
group G p is upper triangular, i.e.,<br />
( )<br />
δ ψ<br />
ρ f | Gp ∼ ,<br />
0 ɛ<br />
where δ, ɛ : G p → K × f,℘ are characters with ɛ unramified and ψ : G p → K f,℘ is a c<strong>on</strong>tinuous<br />
functi<strong>on</strong>.<br />
Definiti<strong>on</strong> 6.10.1. We say that the ordinary representati<strong>on</strong> ρ f | Gp splits, if the representati<strong>on</strong><br />
space of ρ f can be written as direct sum of two G p -invariant lines.<br />
6.10.1 Buzzard’s result<br />
We shall assume that a slight strengthening of a result of Buzzard holds. Let O denote<br />
the ring of integers in a finite extensi<strong>on</strong> K of Q p , and λ denote the maximal ideal of O.<br />
Let ρ : G Q → GL 2 (O) be a c<strong>on</strong>tinuous representati<strong>on</strong> and let ¯ρ denote the reducti<strong>on</strong> mod<br />
λ. The following result is proved in [Buz03], and we refer <strong>to</strong> that paper for a detailed<br />
explanati<strong>on</strong> of all the hypotheses.<br />
Theorem 6.10.2 (Buzzard). Assume that<br />
1. ρ is ramified at finitely many primes and ¯ρ is modular,<br />
2. ¯ρ is absolutely irreducible when restricted <strong>to</strong> Gal( ¯Q/Q(i)),<br />
3. ρ| Gp is the direct sum of two 1-dimensi<strong>on</strong>al characters α and β : G p → O x , such<br />
that α(I p ) and β(I p ) are finite, and (α/β) mod λ is n<strong>on</strong>-trivial,<br />
4. ¯ρ(c) ≠ 1,<br />
5. ¯ρ is both α-modular and β-modular, in the sense that there are eigen<strong>forms</strong> f α with<br />
T p -eigenvalue ᾱ(Frob p ) and f β with T p -eigenvalue ¯β(Frob p ) giving rise <strong>to</strong> ¯ρ,<br />
6. The projective image of ¯ρ is not dihedral.<br />
Then ρ is modular, in the sense that there exists an embedding i : K ↩→ C and a<br />
classical weight 1 cuspidal eigenform f such that the composite i ◦ ρ is isomorphic <strong>to</strong> the<br />
representati<strong>on</strong> associated <strong>to</strong> f by Deligne and Serre.<br />
For odd primes p, there is no restricti<strong>on</strong> (6) <strong>on</strong> the projective image of ¯ρ. For p = 2,<br />
this assumpti<strong>on</strong> is due <strong>to</strong> the unavailability of R = T theorems in the residually dihedral<br />
setting. From now <strong>on</strong>, we assume that Theorem 6.10.2 holds, without c<strong>on</strong>diti<strong>on</strong> (6).<br />
95
6.10.2 Λ-adic Galois representati<strong>on</strong>s<br />
Now, we recall a few facts about Λ-adic Galois representati<strong>on</strong>s. Let F ∈ I[[q]] be a<br />
primitive Λ-adic form of level N and with character ψ. There is a Galois representati<strong>on</strong><br />
<strong>attached</strong> <strong>to</strong> F, c<strong>on</strong>structed by Hida, and Wiles in the case of p = 2,<br />
ρ F : G Q → GL 2 (L),<br />
such that for each arithmetic specializati<strong>on</strong> P of I, P (ρ F ), the specializati<strong>on</strong> of ρ F at P ,<br />
is isomorphic <strong>to</strong> the representati<strong>on</strong> ρ f <strong>attached</strong> <strong>to</strong> f = P (F) by Deligne. Note that, if l<br />
is a prime number such that l ∤ Np, then<br />
trace(ρ F (Frob l )) = a(l, F) ∈ I, det(ρ F (Frob l )) = ψ(l)κ(Frob l )l −1 ,<br />
where κ : Gal( ¯Q/Q) → Λ × is the ‘Λ-adic cyclo<strong>to</strong>mic character’.<br />
If F is ordinary, then the restricti<strong>on</strong> of ρ F <strong>to</strong> G p turns out <strong>to</strong> be ‘upper-triangular’.<br />
More precisely, the representati<strong>on</strong> ρ F | Gp has the following shape<br />
( )<br />
δ F u F<br />
ρ F | Gp ∼<br />
,<br />
0 ɛ F<br />
where δ F , ɛ F : G p → L × are characters with ɛ F unramified, and u F : G p → L is a<br />
c<strong>on</strong>tinuous map. Let<br />
c F = ɛ −1<br />
F · u F ∈ Z 1 (G p , L(δ F ɛ −1<br />
F<br />
)) (6.10.1)<br />
be the associated cocycle. Then the representati<strong>on</strong><br />
ρ F | Gp<br />
splits if and <strong>on</strong>ly if [c F ] = 0 in H 1 (G p , L(δ F ɛ −1<br />
F )).<br />
Definiti<strong>on</strong> 6.10.3. We say that F is p-distinguished, if the characters δ F and ɛ F appearing<br />
above, have distinct reducti<strong>on</strong>s modulo the maximal ideal of I.<br />
It follows that F is p-distinguished if and <strong>on</strong>ly if f is p-distinguished for <strong>on</strong>e (therefore<br />
every) arithmetic specializati<strong>on</strong> f of F.<br />
We shall show that for a primitive 2-adic family F whose residual representati<strong>on</strong> satisfies<br />
some technical c<strong>on</strong>diti<strong>on</strong>s (cf. c<strong>on</strong>diti<strong>on</strong>s (1), (2), and (3) below), the corresp<strong>on</strong>ding<br />
representati<strong>on</strong> ρ F splits at p if and if F is CM. As a c<strong>on</strong>sequence, standard descent arguments<br />
(cf. §6.10.4) allow us <strong>to</strong> c<strong>on</strong>clude the following partial result <strong>to</strong>wards Greenberg’s<br />
questi<strong>on</strong> <strong>on</strong> the <strong>local</strong> splitting of ordinary 2-adic modular Galois representati<strong>on</strong>s.<br />
Theorem 6.10.4. Let F be a primitive n<strong>on</strong>-CM 2-ordinary Hida family of eigen<strong>forms</strong><br />
with the property that<br />
1. ¯ρ F is p-distinguished,<br />
2. ¯ρ F is absolutely irreducible, when restricted <strong>to</strong> Gal( ¯Q/Q(i)),<br />
3. ¯ρ F (c) ≠ 1 and ¯ρ F is both α-modular and β-modular.<br />
Then for all but except possibly finitely many arithmetic members f ∈ F, the representati<strong>on</strong><br />
ρ f | Gp is n<strong>on</strong>-split. Moreover the possible excepti<strong>on</strong>s are necessarily n<strong>on</strong>-CM <strong>forms</strong>.<br />
96
6.10.3 Local splitting for Λ-adic <strong>forms</strong><br />
Propositi<strong>on</strong> 6.10.5. Let F be a primitive 2-adic Λ-adic form of fixed tame level N<br />
satisfying c<strong>on</strong>diti<strong>on</strong>s (1)-(3) above. Then ρ F | Gp splits if and <strong>on</strong>ly if F is of CM type.<br />
Proof. The proof is very similar <strong>to</strong> that for odd primes given in [GV04, Prop. 14]. One<br />
shows that the following statements are equivalent.<br />
1. ρ F | Gp splits.<br />
2. F has infinitely many weight <strong>on</strong>e classical specializati<strong>on</strong>s.<br />
3. F has infinitely many weight <strong>on</strong>e classical CM specializati<strong>on</strong>s.<br />
4. F is of CM type.<br />
We prove the implicati<strong>on</strong>s (1) =⇒ (2), <strong>to</strong> show how the strengthened versi<strong>on</strong> of Buzzard’s<br />
result is used. For the remaining implicati<strong>on</strong>s, we refer <strong>to</strong> [GV04, Prop. 14]. We remark<br />
that a shortening of the implicati<strong>on</strong> (3) =⇒ (4) can be found in [DG10].<br />
(1) =⇒ (2): Recall that we have the following characters:<br />
ψ : Gal( ¯Q/Q) → ¯Q × p<br />
κ : Gal( ¯Q/Q) → Λ × ,<br />
ν : Gal( ¯Q/Q) → Z × p<br />
the character of F of c<strong>on</strong>duc<strong>to</strong>r Nq,<br />
the Λ-adic cyclo<strong>to</strong>mic character,<br />
the 2-adic cyclo<strong>to</strong>mic character.<br />
We know that det(ρ F )= ψκν −1 . The specializati<strong>on</strong> of det(ρ F ) at ϕ k,ζ is χν k−1 , where<br />
χ = ψω −k χ ζ . By assumpti<strong>on</strong> ρ F | Gp splits, i.e.,<br />
( )<br />
ψκν −1 0<br />
ρ F | Ip ∼<br />
.<br />
0 1<br />
Let P be a weight <strong>on</strong>e point of I extending ϕ 1,ζ : Λ → ¯Q p . It follows that P (ρ F ) = ρ P (F)<br />
has the following shape <strong>on</strong> I p :<br />
( )<br />
ψω −1 χ<br />
ρ P (F) | Ip ∼<br />
ζ 0<br />
,<br />
0 1<br />
noting that the characters <strong>on</strong> the diag<strong>on</strong>al have finite order. Now by Theorem 6.10.2, we<br />
have that<br />
ρ P (F) ∼ ρ f ,<br />
where f is a primitive weight 1 form of level N2 r , with character ψω −1 χ ζ where ζ is<br />
exactly of order 2 r−2 , r ≥ 2. As we vary the point P , and therefore r ≥ 2, we obtain<br />
infinitely many classical weight 1 specializati<strong>on</strong>s of F as required.<br />
We remark that elementary arguments (cf. [GV04, (2) =⇒ (3) of Prop. 14]), allow<br />
us <strong>to</strong> c<strong>on</strong>clude that infinitely many of these must be of CM type, and in particular the<br />
residual representati<strong>on</strong> must necessarily be dihedral! This explains why it is crucial <strong>to</strong><br />
assume that Buzzard’s result holds in this case as well.<br />
97
6.10.4 Descending <strong>to</strong> the classical situati<strong>on</strong><br />
We now give a proof of Theorem 6.10.4.<br />
Proof. If F does not have CM, then by Propositi<strong>on</strong> 6.10.5 the representati<strong>on</strong> ρ F | Gp is<br />
not split. We now show that this forces ρ f | Gp <strong>to</strong> be n<strong>on</strong>-split for all but except possibly<br />
finitely many arithmetic members f ∈ F. To see this suppose F ∈ I[[q]]. Then<br />
I × = µ d−1 × (1 + m I ),<br />
where d denotes the order of the residue field of I. Since both δ F and ɛ F are I × valued,<br />
we may decompose these characters as δ F = δ t δ w according <strong>to</strong> the decompositi<strong>on</strong> of I ×<br />
above. Let E t denote the uni<strong>on</strong> of the finitely many tamely ramified abelian extensi<strong>on</strong>s<br />
of Q p of order dividing d − 1. Let E w denote the maximal abelian pro-p extensi<strong>on</strong> of Q p ,<br />
and let E = E t · E w . Set H = Gal( ¯Q p /E). Then δ t and ɛ t (resp., δ w and ɛ w ) are trivial <strong>on</strong><br />
Gal( ¯Q p /E t ) (resp., Gal( ¯Q p /E w )). It follows that δ F and ɛ F are trivial <strong>on</strong> H and hence:<br />
( )<br />
1 λ<br />
ρ F | H ∼ ,<br />
0 1<br />
for some homomorphism λ : H → I.<br />
We show that the homomorphism λ is n<strong>on</strong>-zero. Let c F denote the cocycle corresp<strong>on</strong>ding<br />
<strong>to</strong> λ, as we defined in (6.10.1). Since ρ F | Gp is n<strong>on</strong>-split, the corresp<strong>on</strong>ding<br />
cohomology class [c F ] ≠ 0. By [GV04, Lem. 19], the restricti<strong>on</strong> <strong>to</strong> H of the cohomology<br />
class [c F ] is still n<strong>on</strong>-zero. It follows that the homomorphism λ is n<strong>on</strong>-zero.<br />
Let J denote the n<strong>on</strong>-zero ideal of I generated by the image of λ. Since the intersecti<strong>on</strong><br />
of infinitely many height <strong>on</strong>e primes of I is the zero ideal, we see that J has <strong>to</strong> be c<strong>on</strong>tained<br />
in <strong>on</strong>ly finitely many height <strong>on</strong>e primes of I. Let f = P (F) be an arithmetic specializati<strong>on</strong><br />
of F. Then ρ f | H splits if and <strong>on</strong>ly if I ⊂ P . It follows that ρ f | H does not split for all but<br />
finitely many specializati<strong>on</strong>s f of F. In particular, ρ f | Gp is n<strong>on</strong>-split for all but finitely<br />
many f of F. This excepti<strong>on</strong>al set does not c<strong>on</strong>tain any CM <strong>forms</strong> by Propositi<strong>on</strong> 6.9.11.<br />
This proves the theorem.<br />
98
Appendix A<br />
A 2-adic C<strong>on</strong>trol Theorem for<br />
Modular Curves<br />
In this appendix, we study the behaviour of the ordinary parts of the homology groups of<br />
modular curves, associated with the decreasing sequence of c<strong>on</strong>gruence subgroups Γ 1 (N2 r )<br />
for r ≥ 2, and we prove a c<strong>on</strong>trol theorem for these homology groups. As a c<strong>on</strong>sequence,<br />
we prove the freeness of W 0 (cf. Theorem 6.5.1 and Theorem A.5.3).<br />
A.1 Introducti<strong>on</strong><br />
Hida theory studies the modular curves associated <strong>to</strong> the following c<strong>on</strong>gruence subgroups,<br />
for primes p ≥ 5, and (p, N) = 1,<br />
· · · ⊂ Γ 1 (Np r ) ⊂ · · · ⊂ Γ 1 (Np). (∗)<br />
Write Y r for the open complex manifold Γ 1 (Np r )\H. One of the important results in<br />
Hida theory [Hid86a] is that W 0 := lim H ←−r 1 (Y r , Z p ) 0 is a free Λ-module of finite rank and<br />
W 0 /a r W 0 = H 1 (Y r , Z p ) 0 ,<br />
for all r ≥ 1, where a r denotes the augmentati<strong>on</strong> ideal of Z p [[1 + p r Z p ]] and Λ = Z p [[1 +<br />
pZ p ]]. Emert<strong>on</strong> gives a proof of the results above for p ≥ 5, using algebraic <strong>to</strong>pology<br />
of the Riemann surfaces Y r (cf. [Eme99]). In this appendix, by following the approach<br />
in [Eme99], we prove that the behaviour of the modular curves as in (∗), for p = 2 and 3,<br />
is quite similar <strong>to</strong> (∗∗) (cf. Theorem A.5.2 and Theorem A.5.3).<br />
The classical versi<strong>on</strong>s of the c<strong>on</strong>trol theorem for p = 2 and 3, do not seem <strong>to</strong> be<br />
explicitly available in the literature, though an adèlic versi<strong>on</strong> of it can be found in [Hid88b].<br />
Emert<strong>on</strong>s proof for primes p ≥ 5 holds for p = 3, with N > 1, verbatim. For the prime<br />
p = 2, a bit more work is needed and <strong>forms</strong> the c<strong>on</strong>tent of this appendix. We follow the<br />
notati<strong>on</strong>s in the previous chapter (cf. §6.2, §6.3), unless explicitly stated.<br />
(∗∗)<br />
A.2 Notati<strong>on</strong>s<br />
Throughout this appendix, let p denote the prime 2 and let q denote 4. Let N be a<br />
natural number coprime <strong>to</strong> p. We look at the modular curves associated <strong>to</strong> the following<br />
99
c<strong>on</strong>gruence subgroups:<br />
· · · ⊂ Γ 1 (Np r ) ⊂ · · · ⊂ Γ 1 (Nq).<br />
If we take the homology of the <strong>to</strong>wer of modular curves, we get a <strong>to</strong>wer of finitely generated<br />
free abelian groups, which is the abelianizati<strong>on</strong> of the above chain of subgroups:<br />
· · · → Γ 1 (Np r ) ab → · · · → Γ 1 (Nq) ab , (A.2.1)<br />
because, for r ≥ 2, H 1 (Y r , Z) = Γ 1 (Np r ) ab . We have a short exact sequence of groups:<br />
1 → Γ 1 (Np r ) → Φ 2 r<br />
η r<br />
→ Γ/Γr → 1,<br />
where η r ( ( a b<br />
c d<br />
)<br />
) = d mod Γr . The acti<strong>on</strong> of Φ 2 r <strong>on</strong> Γ 1 (Np r ) by c<strong>on</strong>jugati<strong>on</strong> induces an<br />
acti<strong>on</strong> of Φ 2 r/Γ 1 (Np r ) = Γ/Γ r <strong>on</strong> Γ 1 (Np r ) ab . Thus Γ acts naturally <strong>on</strong> Γ 1 (Np r ) ab through<br />
its quotient Γ/Γ r . The morphisms in the chain (A.2.1) are Γ-equivariant.<br />
Remark A.2.1. The restricti<strong>on</strong> of η r <strong>to</strong> Φ 2 r ∩Γ 0 (p), denoted by Res(η r ), is also surjective<br />
<strong>on</strong><strong>to</strong> Γ/Γ r . Moreover, we have the following commutative diagram:<br />
Φ 2 r+1<br />
η r+1<br />
Γ/Γ r+1<br />
≀<br />
t −1 −t<br />
Φ 2 r ∩ Γ 0 (p) Res(ηr) Γ/Γ r ,<br />
where the group Γ 0 (p) = { ( )<br />
a b<br />
c d ∈ SL2 (Z) | b ≡ 0 (mod 2)}, and t = ( )<br />
1 0<br />
0 p .<br />
For r ≥ s > 1, if we abelianize the short exact sequence in (6.3.1), we obtain:<br />
Γ 1 (Np r ) ab → Φ s ab<br />
r → Γ s /Γ r → 1.<br />
The group Γ s is procyclic, and let γ s denote a genera<strong>to</strong>r of it. Then, the augmentati<strong>on</strong><br />
ideal a s is generated by γ s − 1. For i ≥ 1, the group Γ s+i is generated by γs pi<br />
, and hence<br />
a s+i is generated by γs<br />
pi<br />
− 1. For any r ≥ s, by definiti<strong>on</strong>, the augmentati<strong>on</strong> ideal of<br />
Z p [Γ s /Γ r ] is a s . Then for any r ≥ s > 1, by definiti<strong>on</strong><br />
a s Γ 1 (Np r ) ab = [Φ s r, Γ 1 (Np r )]/[Γ 1 (Np r ), Γ 1 (Np r )] ⊂ Γ 1 (Np r ) ab ,<br />
and the last inclusi<strong>on</strong> follows, since Γ 1 (Np r ) is a normal subgroup of Φ s r. The extensi<strong>on</strong><br />
1 → Γ 1 (Np r )/[Φ s r, Γ 1 (Np r )] → Φ s r/[Φ s r, Γ 1 (Np r )] → Γ s /Γ r → 1<br />
is a central extensi<strong>on</strong> of a cyclic group, thus the middle group is abelian, implying that<br />
[Φ s r, Φ s r] = [Φ s r, Γ 1 (Np r )]. One way inclusi<strong>on</strong> is clear, because Φ s r ⊇ Γ 1 (Np r ). The other<br />
inclusi<strong>on</strong> follows from the fact that [Φ s r/[Φ s r, Γ 1 (Np r )], Φ s r/[Φ s r, Γ 1 (Np r )]] is trivial.<br />
∼<br />
−→ Γs<br />
Γ r<br />
, by Remark A.2.1. More-<br />
Remark A.2.2. We have an isomorphism<br />
over, we have the following commutative diagram:<br />
Φs r∩Γ 0 (p)<br />
Γ 1 (Np r )∩Γ 0 (p)<br />
Φ s r∩Γ 0 (p) i Φ s<br />
Γ 1 (Np r )∩Γ 0 r<br />
(p) Γ<br />
1 (Np r )<br />
<br />
∼<br />
∼<br />
<br />
Γ s<br />
Γ r<br />
,<br />
where the map i is induced by the inclusi<strong>on</strong> of the groups Φ s r ∩ Γ 0 (p) ⊆ Φ s r. In fact, the<br />
map i is an isomorphism. This observati<strong>on</strong> is useful in Lemma A.3.5.<br />
100
Summarising this discussi<strong>on</strong>, we see that the map Γ 1 (Np r ) ab → Γ 1 (Np s ) ab in the<br />
chain of homology groups as in (A.2.1) can be fac<strong>to</strong>red as the compositi<strong>on</strong> of<br />
Γ 1 (Np r ) ab ↠ Γ 1 (Np r ) ab /a s ↩→ Φ s ab<br />
r → Γ 1 (Np s ) ab . (A.2.2)<br />
It may not be true that sec<strong>on</strong>d and third morphisms of this fac<strong>to</strong>risati<strong>on</strong>s are isomorphisms.<br />
But Hida observed that if <strong>on</strong>e applies a certain projecti<strong>on</strong> opera<strong>to</strong>r arising from<br />
the Atkin U-opera<strong>to</strong>r <strong>to</strong> all these modules then the sec<strong>on</strong>d and third morphisms become<br />
isomorphisms. So we now define the Atkin U-opera<strong>to</strong>r and study their properties.<br />
A.3 Hecke opera<strong>to</strong>rs<br />
Suppose T is a group which c<strong>on</strong>tains subgroups G and H and that t is an element of T<br />
such that t −1 Ht ∩ G has finite index in G. Then <strong>on</strong>e has a transfer map<br />
V : G ab → (t −1 Ht ∩ G) ab .<br />
C<strong>on</strong>jugati<strong>on</strong> by t induces an isomorphism (t −1 Ht∩G) ab ≃ (H ∩tGt −1 ) ab . Inclusi<strong>on</strong> of the<br />
group H ∩tGt −1 in H induces a morphism (H ∩tGt −1 ) ab → H ab . Taking the compositi<strong>on</strong><br />
of all these we obtain a morphism, we call the ‘Hecke opera<strong>to</strong>r’ corresp<strong>on</strong>ding <strong>to</strong> t,<br />
[t] : G ab → H ab .<br />
In the case T = GL 2 (Q), G = H = a c<strong>on</strong>gruence subgroup of SL 2 (Z) of level divisible<br />
by p and t = ( ) (<br />
1 0<br />
0 p , we set U := [t]. For A = a b<br />
)<br />
c d ∈ Φ<br />
s<br />
r , we have:<br />
t −1 At = ( a<br />
bp<br />
c/p d<br />
)<br />
and<br />
tAt −1 = ( a b/p<br />
cp d<br />
and hence t −1 Φ s rt ∩ Φ s r = Φ s r ∩ Γ 0 (p), Φ s r ∩ tΦ s rt −1 = Φ s r+1 .<br />
Remark A.3.1. Observe that (1, 1), (2, 2)-entries of A and of t ±1 At ∓1 are the same.<br />
Thus the Atkin U-opera<strong>to</strong>r is by definiti<strong>on</strong> the compositi<strong>on</strong><br />
)<br />
,<br />
Φ s ab<br />
r<br />
V (Φ s r ∩ Γ 0 (p)) ab t−t −1 <br />
∽ Φ s r+1<br />
ab<br />
Φ s ab<br />
r . (A.3.1)<br />
(The final morphism is just that induced by the inclusi<strong>on</strong> of groups Φ s r+1 ⊂ Φs r.) Define<br />
U ′ <strong>to</strong> be the compositi<strong>on</strong> of the first two of above morphisms.<br />
Lemma A.3.2. Suppose that r ≥ s > 1, r ′ ≥ s ′ > 1, r ≥ r ′ , s ≥ s ′ , so that Φ s r ⊂ Φ s′<br />
r ′ .<br />
Then the following diagram commutes:<br />
Φ s ab<br />
r<br />
U ′<br />
Φ s ab<br />
r+1<br />
Φ s′ ab<br />
r ′ U ′<br />
Φ s′ ab<br />
r ′ +1 .<br />
C<strong>on</strong>sequently, the map U commutes with the map Φ s r<br />
ab<br />
is a Z[U]-module via the acti<strong>on</strong> of the U-opera<strong>to</strong>r.<br />
→ Φ s′ ab<br />
r<br />
. In particular, each Φ s ab<br />
′<br />
r<br />
101
Proof. We can fac<strong>to</strong>rize the above diagram as a compositi<strong>on</strong> of two diagrams<br />
Φ s ab<br />
r<br />
Φ s′ ab<br />
r ′ V<br />
V<br />
(Φ s ab<br />
r ∩ Γ 0 (p)) ab <br />
(Φ s′ ab<br />
r ′<br />
∩ Γ 0 (p)) ab<br />
t−t −1 <br />
Φ s r+1<br />
ab<br />
t−t −1 <br />
Φ s′ ab<br />
r ′ +1<br />
The lower porti<strong>on</strong> of this diagram clearly commutes. Since Φ s r ∩ Γ 0 (p) has index p in<br />
Φ s r, we can take { ( )<br />
1 i<br />
0 1 }<br />
p−1<br />
i=0<br />
as the coset representatives. The coset representatives are<br />
independent of the particular values of r and s, and hence the transfer map V is given by<br />
a formula independent of r and s, so the upper porti<strong>on</strong> of the diagram commutes.<br />
A particular case of Lemma A.3.2 is the case r ′ = r − 1, s = s ′ ≤ r ′ . If we write π for<br />
the morphism π : Φ s r<br />
ab −→ Φ s r−1 ab and π′ for the morphism π ′ : Φ s r+1 ab −→ Φs<br />
ab<br />
r , then the<br />
lemma above yields the following formula:<br />
The same definiti<strong>on</strong> yields the formula<br />
U ′ ◦ π = π ′ ◦ U ′ = U ∈ End Z (Φ s ab<br />
r ). (A.3.2)<br />
π ◦ U ′ = U ∈ End Z (Φ s ab<br />
r−1).<br />
(A.3.3)<br />
Again by the same lemma, the cokernel of the morphism Γ 1 (Np r ) ab → Φ s ab<br />
r , for r ≥ s > 1,<br />
is a Z[U]-module. Hence Γ s /Γ r is a Z[U]-module.<br />
Lemma A.3.3. For r ≥ s > 1, the opera<strong>to</strong>r U acts <strong>on</strong> Γ s /Γ r as multiplicati<strong>on</strong> by p.<br />
Proof. The opera<strong>to</strong>r U acts <strong>on</strong> Γ s /Γ r as a multiplicati<strong>on</strong> by p if and <strong>on</strong>ly if it acts <strong>on</strong><br />
Φ s ab<br />
r<br />
Γ 1 (Np r ) ab<br />
Φ s ab<br />
r<br />
Γ 1 (Np r ) ab<br />
by Ā ↦→ Āp . The opera<strong>to</strong>r U is the compositi<strong>on</strong> of the following maps:<br />
V<br />
−→<br />
(Φ s r ∩Γ0 (p)) ab t−t<br />
(Γ 1<br />
−→<br />
−1<br />
(Np r )∩Γ 0 (p)) ab<br />
Φ s ab<br />
r+1<br />
Φ r ab<br />
r+1<br />
−→<br />
Φ s ab<br />
r<br />
Γ 1 (Np r ) ab<br />
Ā ↦−→ Ā p ↦−→ tĀp t −1 ↦−→ tĀp t −1 . (A.3.4)<br />
Let {α i = ( )<br />
1 i<br />
0 1 }<br />
p−1<br />
i=0 be a set of coset representatives of the group Φs r ∩ Γ 0 (p) in Φ s r. With<br />
this choice, we see that the transfer map in (A.3.4) looks like Ā ↦→ Āp . By Remark A.3.1,<br />
tA p t −1 and A p represent the same coset mod Γ 1 (Np r ) ab , and hence we are d<strong>on</strong>e.<br />
We would like <strong>to</strong> define an acti<strong>on</strong> of Γ <strong>on</strong> Φ s r<br />
ab and call it the nebentypus acti<strong>on</strong>.<br />
For r ≥ 2, if ¯d ∈ Γ/Γ r , then choose an element α ∈ Φ 2 r+1 ∩ Γ0 (p), such that η r (α) = ¯d<br />
(cf. Remark A.2.1). The nebentypus acti<strong>on</strong> of d <strong>on</strong> Φ s r<br />
ab is given by c<strong>on</strong>jugati<strong>on</strong> by<br />
α. This acti<strong>on</strong> is well-defined because, if α 1 and α 2 denote two lifts of ¯d, then α1 −1 α 2 ∈<br />
Γ 1 (Np r+1 ) ∩ Γ 0 (p) ⊆ Φ s r, and hence for any element x ∈ Φ s r, α1 −1 α 2xα2 −1 α 1 = x in Φ s r ab .<br />
Now, we shall show that the acti<strong>on</strong>s of Γ and U commutes.<br />
Lemma A.3.4. If r ≥ s > 1, the acti<strong>on</strong> of U commutes with the acti<strong>on</strong> of Γ <strong>on</strong> Φ s ab<br />
r .<br />
102
Proof. It is easy <strong>to</strong> see that α(Φ s r ∩Γ 0 (p))α −1 = Φ s r ∩Γ 0 (p) for any α ∈ Φ 2 r+1 ∩Γ0 (p), since<br />
αΦ s rα −1 ⊆ Φ s r. We need <strong>to</strong> prove that the diagram below is commutative, because in the<br />
diagram below compositi<strong>on</strong> of the vertical morphisms <strong>on</strong> either side are the opera<strong>to</strong>r U<br />
and it commutes with the au<strong>to</strong>morphism of Φ s r<br />
ab induced by c<strong>on</strong>jugati<strong>on</strong> by α, but we<br />
know that Γ acts <strong>on</strong> Φ s r ab by c<strong>on</strong>jugati<strong>on</strong> by such elements α.<br />
Φ s ab<br />
r<br />
α−α −1<br />
Φ s ab<br />
r<br />
V<br />
(Φ s r ∩ Γ 0 (p))<br />
ab<br />
α−α−1<br />
(α(Φ s r ∩ Γ 0 (p))α −1 ) ab = (Φ s r ∩ Γ 0 (p)) ab<br />
V<br />
Φ s ab<br />
r+1<br />
t−t −1<br />
α−α −1<br />
αtα −1 (−)αt −1 α −1<br />
(αΦ s r+1 α−1 ) ab = Φ s ab<br />
r+1<br />
Φ s ab<br />
r<br />
α−α −1<br />
(αΦ s rα −1 ) ab = Φ s ab<br />
The <strong>to</strong>p square in the diagram above commutes because, if {γ 1 , . . . , γ p } form a set of coset<br />
representatives for the group Φ s r ∩ Γ 0 (p) in Φ s r, then so is {αγ 1 α −1 , . . . , αγ p α −1 }. This<br />
diagram commutes even if α ∈ Φ 2 r ∩ Γ 0 (p). The last square commutes by the func<strong>to</strong>riality<br />
of the transfer map. We now prove the commutativity of the middle square, i.e., the map<br />
αtα −1 (−)αt −1 α −1 : (Φ s r∩Γ 0 (p)) ab → Φ s r+1 ab is t−t−1 . For g ∈ Φ s r∩Γ 0 (p), αtα −1 gαt −1 α −1 =<br />
(αtα −1 t −1 )tgt −1 (αtα −1 t −1 ) −1 . For α ∈ Φ 2 r+1 ∩ Γ0 (p), we have αtα −1 t −1 ∈ Γ 1 (Np r+1 ) ⊆<br />
Φ s r+1 . Thus c<strong>on</strong>jugati<strong>on</strong> by αtα−1 t −1 induces the identity <strong>on</strong> Φ s r+1 ab.<br />
Lemma A.3.5. The transfer morphism V : Φ s r<br />
ab<br />
of U <strong>on</strong> its source and target.<br />
Φ s ab<br />
r<br />
r .<br />
→ Γ 1 (Np r ) ab commutes with the acti<strong>on</strong><br />
Proof. The inclusi<strong>on</strong> of groups Γ 1 (Np r ) ⊆ Φ s r gives rise <strong>to</strong> the another transfer map<br />
V<br />
→ Γ 1 (Np r ) ab . It suffices <strong>to</strong> prove that the following diagram commutes:<br />
Φ s ab<br />
r<br />
V<br />
Φ r ab<br />
r<br />
V<br />
(Φ s r ∩ Γ 0 (p)) ab V (Γ 1 (Np r ) ∩ Γ 0 (p)) ab<br />
Φ s ab<br />
r<br />
t−t −1<br />
V<br />
t−t −1<br />
V Γ 1 (Np r ) ab .<br />
Here V denotes the transfer maps between various abelianizati<strong>on</strong>s and <strong>to</strong> be unders<strong>to</strong>od<br />
from the c<strong>on</strong>text. The commutativity of the <strong>to</strong>p square and the bot<strong>to</strong>m square follows<br />
from the func<strong>to</strong>riality of the transfer map, and from Remark A.2.2, Remark A.3.1, respectively.<br />
If {σ d } denote a set of coset representatives of Γ 1 (Np r ) ∩ Γ 0 (p) in Φ s r ∩ Γ 0 (p),<br />
then the c<strong>on</strong>jugates {tσ d t −1 } form a set of coset representatives of Γ 1 (Np r ) in Φ s r.<br />
To summarise, we have defined the U-opera<strong>to</strong>rs for the c<strong>on</strong>gruence subgroups {Φ s r}<br />
and we have checked that morphisms between these c<strong>on</strong>gruence subgroups respects the<br />
acti<strong>on</strong> of U and this acti<strong>on</strong> commutes with that of Γ. We have also checked that the<br />
transfer morphism V : Φ s ab<br />
r → Γ 1 (Np r ) ab commutes with the acti<strong>on</strong> of U.<br />
103
A.4 Ordinary parts<br />
Let M be a Z p [X]-module, which is finitely generated as a Z p -module. Then, we have a<br />
morphism of Z p -modules<br />
Z p [X] → End Zp (M).<br />
The endomorphism ring End Zp (M) of M is a finite Z p -algebra. Thus the image of Z p [X]<br />
in End Zp (M), will be denoted by A, is also a commutative finite Z p -algebra, and hence it<br />
fac<strong>to</strong>rs as a product of <strong>local</strong> rings. Let A 0 denote the product of all those <strong>local</strong> fac<strong>to</strong>rs of A<br />
in which the image of X is a unit, and A n its complementary fac<strong>to</strong>r, so that A = A 0 × A n .<br />
Each of these is a flat A-algebra, and a subalgebra of End Zp (M). We define<br />
M 0 := M ⊗ A A 0<br />
and call this the ordinary part of M. Clearly, taking ordinary parts is an exact func<strong>to</strong>r.<br />
If we c<strong>on</strong>sider X <strong>to</strong> be the U-opera<strong>to</strong>r, we may c<strong>on</strong>sider the ordinary part of the<br />
Z p -homology of the curve Y r , i.e., the module (Γ 1 (Np r ) ab ⊗ Z p ) 0 , which is a Γ-module by<br />
Lemma A.3.4. We have the following fundamental theorem for the prime p = 2 and it is<br />
similar <strong>to</strong> the theorem, which was proved in [Hid86a] for primes p ≥ 5.<br />
Theorem A.4.1. If r ≥ s > 1, then the morphism of abelian groups<br />
is an isomorphism.<br />
(Γ 1 (Np r ) ⊗ Z p ) 0 /a s ↠ (Γ 1 (Np s ) ⊗ Z p ) 0<br />
Proof. We shall prove the following isomorphisms:<br />
(Γ 1 (Np r ) ab ⊗ Z p ) 0 /a s<br />
∼ −→ (Φ<br />
s ab<br />
r ⊗ Z p ) 0 ∼ −→ (Γ1 (Np s ) ab ⊗ Z p ) 0 .<br />
We have c<strong>on</strong>structed an opera<strong>to</strong>r U ′ : Φ s ab<br />
r−1<br />
and (A.3.3). If π : Φ s ab<br />
r<br />
Φ s ab<br />
r<br />
→ Φ s ab<br />
r−1<br />
, satisfying the equati<strong>on</strong>s (A.3.2)<br />
is the morphism induced by the inclusi<strong>on</strong> of groups<br />
→ Φs<br />
ab<br />
r<br />
⊂ Φ s r−1 ab,<br />
then U ′ ◦ π = U ∈ End(Φ s r<br />
ab ), π ◦ U ′ = U ∈ End(Φ s r−1).<br />
ab<br />
We have the following diagram:<br />
Φ s r−1<br />
ab π<br />
Φ s r−2<br />
ab<br />
<br />
U ′ U<br />
U ′<br />
Φ s r<br />
ab π<br />
Φ s r−1<br />
ab<br />
<br />
U ′ U<br />
U ′<br />
Φ s ab<br />
r+1<br />
π<br />
Φ s r ab .<br />
The existence of U ′ implies that up<strong>on</strong> tensoring over Z p and taking the ordinary parts π<br />
induces an isomorphism (Φ s r<br />
ab ) 0 = (Φ s r−1 ab)0<br />
. By descending inducti<strong>on</strong> <strong>on</strong> r, we obtain the<br />
required isomorphism<br />
(Φ s r<br />
ab ) 0 = (Φ s s ab ) 0 = (Γ 1 (Np s ) ab ⊗ Z p ) 0 .<br />
104
To prove the first isomorphism, we c<strong>on</strong>sider the short exact sequence:<br />
1 → Γ 1 (Np r ) ab /a s → Φ s ab<br />
r → (Γ s /Γ r ) → 1.<br />
Tensor this short exact sequence with Z p and then take the ordinary parts <strong>to</strong> obtain:<br />
1 → (Γ 1 (Np r ) ab ⊗ Z p ) 0 /a s → (Φ s ab<br />
r ⊗ Z p ) 0 → (Γ s /Γ r ) 0 → 1,<br />
because Z p is flat over Z and taking ordinary parts preserves exactness. By Lemma A.3.3,<br />
the opera<strong>to</strong>r U is nilpotent, because Γ s /Γ r is a p-<strong>to</strong>rsi<strong>on</strong> group. Thus Γ s /Γ r has trivial<br />
ordinary part and hence the Theorem follows.<br />
A.5 Iwasawa modules<br />
We have the following inverse system indexed by natural numbers r ≥ 2,<br />
Define the Iwasawa module by<br />
· · · → Γ 1 (Np r ) ab ⊗ Z p → · · · → Γ 1 (Nq) ab ⊗ Z p .<br />
W := lim ←− r≥2<br />
Γ 1 (Np r ) ab ⊗ Z p .<br />
The profinite group Γ acts <strong>on</strong> the Z p -module Γ 1 (Np r ) ab ⊗ Z p through its finite quotient<br />
Γ/Γ r . Thus the Iwasawa module W becomes a module over the completed group algebra<br />
Λ := Z p [[Γ]] = lim Z ←−r≥2 p [Γ/Γ r ]. The Iwasawa module W is difficult <strong>to</strong> understand, because<br />
we do not have a good characterisati<strong>on</strong> of the image of the morphism<br />
Γ 1 (Np r ) ab → Γ 1 (Np s ) ab<br />
(r ≥ s > 1) in general, and so we cannot get a good descripti<strong>on</strong> of the projective limit.<br />
However, Theorem A.4.1 allows us <strong>to</strong> understand the ordinary part of W very well. To<br />
make the statement clear, let us slightly abstract the situati<strong>on</strong>.<br />
Let {M r } r≥2 be a system of Λ-modules, such that each M r is pointwise fixed by Γ r ,<br />
i.e., M r is a module over Λ/a r Λ = Z p [Γ/Γ r ]. For each r ≥ s > 1, we have a map M r → M s<br />
such that it fac<strong>to</strong>rs via<br />
M r /a s M r → M s .<br />
Define W := lim ←−r≥2<br />
M r . For each r ≥ 2, we have a map W → M r , and they fac<strong>to</strong>r as<br />
W/a r W → M r .<br />
Lemma A.5.1. Assume that each M r is p-adically complete and for each r ≥ s > 1,<br />
M r /a s M r → M s is an isomorphism. Then W/a s W ↠ M s is an isomorphism.<br />
Proof. Clearly the maps M r → M s are surjective for all r ≥ s, and hence the can<strong>on</strong>ical<br />
map from W → M s is also surjective. It is enough <strong>to</strong> show that the kernel is a s W . Since<br />
each M r is p-adically complete and we have that M r = lim M ←−i r /p i M r . Moreover, we have<br />
that each M r is pointwise fixed by Γ r and hence M r = lim M ←−i r /n i M r , where γ r is a cyclic<br />
genera<strong>to</strong>r of Γ r and n = (γ r − 1, p).<br />
105
By inducti<strong>on</strong> <strong>on</strong> i, we see that γs<br />
pi<br />
− 1/γ s − 1 is an element of the ideal (γ s − 1, p) i . In<br />
particular, when s = 2, the element γ pr−2<br />
2 − 1/γ 2 − 1 is an element of the maximal ideal<br />
m = (γ 2 − 1, p) and hence n ⊆ m = (a 2 , p). Since m pr−2 ⊆ ((γ 2 − 1) pr−2 , p) ⊆ n ⊆ m,<br />
we see that each M r is m-adically complete, since they are n-adically complete. Once we<br />
have that each M r is m-adically complete, then proving the injectivity of the map above<br />
is quite similar <strong>to</strong> the proof of [Eme99, Prop. 5.1].<br />
The following Theorem is an immediate c<strong>on</strong>sequence of the above Lemma.<br />
Theorem A.5.2. For any r ≥ 2, we have<br />
is the Γ r -coinvariants of W 0 .<br />
W 0 /a r W 0 ∼ = (Γ1 (Np r ) ab ⊗ Z p ) 0<br />
Proof. This follows from Lemma A.5.1 <strong>to</strong>gether with Theorem A.4.1<br />
The theorem above is <strong>on</strong>e of the key step in the proof of the main theorem A.5.3.<br />
Each module Γ 1 (Np r ) ab ⊗ Z p is free of finite rank over Z p , and so is compact in its p-adic<br />
<strong>to</strong>pology. Thus if we give the limiting module W the <strong>to</strong>pology which is the projective<br />
limit of the p-adic <strong>to</strong>pologies <strong>on</strong> each module Γ 1 (Np r ) ab ⊗ Z p it becomes a compact Λ-<br />
module. Furthermore, Λ acts c<strong>on</strong>tinuously <strong>on</strong> W , since Γ acts <strong>on</strong> each of the modules<br />
Γ 1 (Np r ) ab ⊗ Z p through a finite quotient. Since W 0 is a direct fac<strong>to</strong>r of W the same<br />
remarks hold true for W 0 . Furthermore, Theorem A.5.2 implies that the projective limit<br />
<strong>to</strong>pology <strong>on</strong> W 0 coincides with its m-adic <strong>to</strong>pology (where m = (a 2 , p) ⊂ Λ denotes the<br />
maximal ideal of Λ), because the kernels of the projecti<strong>on</strong> Λ → Z p /p r Z p [Γ/Γ r ] are cofinal<br />
with the sequence of ideals m r in Λ. Thus W 0 is a Λ-module, compact in its m-adic<br />
<strong>to</strong>pology such that<br />
W 0 /m = W 0 /(a 2 , p) = (Γ 1 (Nq) ab ⊗ Z p /p) 0<br />
is a finite dimensi<strong>on</strong>al Z p /pZ p -module, of dimensi<strong>on</strong> d (say). By Nakayama’s lemma, we<br />
have that W 0 is a finitely generated Λ-module with a minimal generating set of order<br />
equal <strong>to</strong> d. By Theorem A.5.3 below, d will be equal <strong>to</strong> the Z p -rank of the free Z p -module<br />
(Γ 1 (Np r ) ab ⊗ Z p ) 0 , for r ≥ 2. We have the following theorem for the prime p = 2 and it<br />
is similar <strong>to</strong> the theorem, which was proved in [Hid86a] for primes p ≥ 5.<br />
Theorem A.5.3. The Λ-module W 0 is free of finite rank equal <strong>to</strong> d.<br />
By [NSW08, Prop. 5.1.9], it is enough <strong>to</strong> show that, the module W 0 is a reflexive Λ-<br />
module. We will prove this by c<strong>on</strong>sidering the duality theory of the modules (Γ 1 (Np r ) ab ⊗<br />
Z p ) 0 and showing that they are reflexive as Z p [Γ/Γ r ]-modules.<br />
We briefly recall some results from Duality theory. Suppose R is a commutative ring,<br />
G is a finite group, and M is a left R[G]-module. Write M ∗ = Hom R (M, R), the R-dual<br />
of M. The module M ∗∗ can be made a left module over R[G]. By definiti<strong>on</strong> of M ∗ ,<br />
there is a natural morphism of left R[G]-modules M → M ∗∗ . The module M is called<br />
reflexive, if this map is an isomorphism. The following lemma will be useful later.<br />
Lemma A.5.4 ([Eme99], Lem. 6.2). If M is reflexive as an R-module, then M is reflexive<br />
as an R[G]-module.<br />
Now, we need <strong>to</strong> understand how <strong>to</strong> use the reflexivity results for modules over<br />
Z p [Γ/Γ r ] <strong>to</strong> show the reflexivity of W 0 as a Λ-module.<br />
106
A.6 Limits of cohomology modules<br />
Cohomology is the dual of homology:<br />
H 1 (Y r , Z p ) := Hom Z (Γ 1 (Np r ) ab , Z p ) = Hom Zp (Γ 1 (Np r ) ab ⊗ Z p , Z p ).<br />
The ring Λ acts <strong>on</strong> Γ 1 (Np r ) ab ⊗ Z p through its quotient Λ r := Λ/a r = Z p [Γ/Γ r ]. If<br />
r ≥ s > 1 then the ring Λ s is the quotient of the ring Λ r by the augmentati<strong>on</strong> ideal a s ,<br />
i.e., Λ r ↠ Λ s = Λ r /a s . Thus we get the following sequence of morphisms of Λ r -modules:<br />
Hom Λr (Γ 1 (Np r ) ab ⊗ Z p , Λ r ) → Hom Λr (Γ 1 (Np r ) ab ⊗ Z p , Λ r )/a s<br />
→ Hom Λr (Γ 1 (Np r ) ab ⊗ Z p , Λ s ) = Hom Λs (Γ 1 (Np r ) ab ⊗ Z p /a s , Λ s ).<br />
If M is any Z p [U]-module, which is finitely generated as a Z p -module, then so is the<br />
Z p -dual M ∗ := Hom Zp (M, Z p ) (via the dual acti<strong>on</strong> of U). Clearly (M ∗ ) 0 = (M 0 ) ∗ , i.e.,<br />
taking ordinary parts commutes with taking duals. By Theorem A.4.1 and by taking the<br />
ordinary parts of the homomorphisms above, we obtain a morphism:<br />
Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r )/a s<br />
Lemma A.6.1. The morphism η is an isomorphism.<br />
η<br />
−→ Hom Λs ((Γ 1 (Np s ) ab ⊗ Z p ) 0 , Λ s ).<br />
Proof. By virtue of Lemma A.3.5, we may restrict V <strong>to</strong> the ordinary parts of its source<br />
and target <strong>to</strong> obtain a morphism which we c<strong>on</strong>tinue <strong>to</strong> denote by V<br />
(Φ s ab<br />
r ⊗ Z p ) 0 V<br />
−→ (Γ 1 (Np r ) ab ⊗ Z p ) 0 .<br />
There is a dual morphism Hom Zp ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Z p ) −→ V ∗<br />
Hom Zp ((Φ s r ab ⊗ Z p ) 0 , Z p ),<br />
which sits in the first column of the following commutative diagram:<br />
Hom Zp ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Z p )<br />
V ∗<br />
Hom Zp ((Φ s r ab ⊗ Z p ) 0 , Z p )<br />
∼<br />
Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r )<br />
Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r )/a s<br />
≀<br />
<br />
Hom Λs ((Γ 1 (Np r ) ab ⊗ Z p ) 0 /a s , Λ s )<br />
Hom Zp ((Γ 1 (Np s ) ab ⊗ Z p ) 0 , Z p )<br />
∼ Hom Λs ((Γ 1 (Np s ) ab ⊗ Z p ) 0 , Λ s )<br />
≀<br />
in which the two horiz<strong>on</strong>tal isomorphisms are those provided by Lemma A.5.4, and the<br />
two vertical isomorphisms follow from Theorem A.4.1 and its proof. Thus <strong>to</strong> prove the<br />
lemma it suffices <strong>to</strong> prove that<br />
Hom Zp ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Z p ) V ∗<br />
−→ Hom Zp ((Φ s ab<br />
r ⊗ Z p ) 0 , Z p ) (A.6.1)<br />
is surjective with kernel equal <strong>to</strong> a s (Hom Zp ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Z p )). Since V commutes<br />
with U and taking ordinary parts commutes with taking Z p -duals, we see that the morphism<br />
in (A.6.1) is the ordinary part of the morphism<br />
Hom Zp ((Γ 1 (Np r ) ab ⊗ Z p ), Z p ) V ∗<br />
−→ Hom Zp ((Φ s ab<br />
r ⊗ Z p ), Z p ).<br />
107
Taking ordinary parts is also exact and commutes with the acti<strong>on</strong> of Γ. Thus it suffices <strong>to</strong><br />
show that the map above is surjective with kernel equal <strong>to</strong> a s Hom Zp ((Γ 1 (Np r ) ab ⊗Z p ), Z p ).<br />
This statement was proved in [Eme99, §8] for <strong>to</strong>rsi<strong>on</strong>-free groups H and G such that<br />
H ⊆ G, instead of Γ 1 (Np r ) ⊆ Φ s r. It is known that the group Γ 1 (M) is <strong>to</strong>rsi<strong>on</strong> free for<br />
all M ≥ 4. In particular, the group Γ 1 (Nq) is <strong>to</strong>rsi<strong>on</strong>-free, and so is the group Φ s r.<br />
We now give a proof of Theorem A.5.3.<br />
Lemma A.6.2. There is a can<strong>on</strong>ical isomorphism<br />
Hom Λ (W 0 , Λ) = lim ←− r≥2<br />
Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r ).<br />
Proof. We have the following series of can<strong>on</strong>ical isomorphisms:<br />
Hom Λ (W 0 , Λ) = lim Hom ←−r Λ (W 0 , Λ r ) = lim Hom ←−r Λr (W 0 /a r , Λ r )<br />
= lim ←−r<br />
Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r ),<br />
where the last isomorphism follows from the Theorem A.5.2.<br />
Propositi<strong>on</strong> A.6.3. For r > 1, there is a can<strong>on</strong>ical isomorphism<br />
Hom Λ (W 0 , Λ)/a r = Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r ).<br />
Proof. This follows from Lemma A.6.1, Lemma A.6.2 and Lemma A.5.1.<br />
Theorem A.6.4. The module W 0 is Λ-free.<br />
Proof. Since any finitely generated reflexive Λ-module is free, it suffices <strong>to</strong> show that W 0<br />
is a reflexive Λ-module. This follows from the following series of can<strong>on</strong>ical isomorphisms:<br />
Hom Λ (Hom Λ (W 0 , Λ), Λ) = lim ←−r<br />
Hom Λ (Hom Λ (W 0 , Λ), Λ r )<br />
= lim ←−r<br />
Hom Λ (Hom Λr (W 0 , Λ)/a r , Λ r )<br />
(1)<br />
= lim ←−r<br />
Hom Λr (Hom Λr ((Γ 1 (Np r ) ab ⊗ Z p ) 0 , Λ r ), Λ r )<br />
(2)<br />
= lim ←−r<br />
(Γ 1 (Np r ) ab ⊗ Z p ) 0 = W 0 .<br />
(Equality (1) follows from Propositi<strong>on</strong> A.6.3. Equality (2) follows from Lemma A.5.4,<br />
since (Γ 1 (Np r ) ab ⊗Z p ) 0 is a free Z p -module and so is certainly a reflexive Z p -module).<br />
108
Bibliography<br />
[BGK10] D. Banerjee, E. Ghate, and N. Kumar, Λ-adic Forms and the Iwasawa Main<br />
C<strong>on</strong>jecture, Guwahati Workshop <strong>on</strong> Iwasawa Theory of Totally Real Fields,<br />
Ramanujan Math. Soc. Lect. Notes Ser., vol. 12, 2010, pp. 15–47.<br />
[Bre01]<br />
[BS07]<br />
C. Breuil, Lectures <strong>on</strong> p-adic Hodge theory, deformati<strong>on</strong>s and <strong>local</strong> Langlands,<br />
vol. Advanced course lecture notes, 20, Centre de Recerca Matemàtica, 2001.<br />
C. Breuil and P. Schneider, First steps <strong>to</strong>wards p-adic Langlands func<strong>to</strong>riality,<br />
J. Reine Angew. Math. 610 (2007), 149–180.<br />
[Bum97] D. Bump, Au<strong>to</strong>morphic <strong>forms</strong> and representati<strong>on</strong>s, Cambridge Studies in Advanced<br />
Mathematics, vol. 55, Cambridge University Press, Cambridge, 1997.<br />
[Buz03]<br />
[CF00]<br />
K. Buzzard, Analytic c<strong>on</strong>tinuati<strong>on</strong> of overc<strong>on</strong>vergent eigen<strong>forms</strong>, J. Amer.<br />
Math. Soc. 16 (2003), no. 1, 29–55.<br />
P. Colmez and J.-M. F<strong>on</strong>taine, C<strong>on</strong>structi<strong>on</strong> des représentati<strong>on</strong>s p-adiques semistables,<br />
Invent. Math. 140 (2000), no. 1, 1–43.<br />
[CHT08] L. Clozel, M. Harris, and R. Taylor, Au<strong>to</strong>morphy for some l-adic lifts of au<strong>to</strong>morphic<br />
mod l Galois representati<strong>on</strong>s, Publ. Math. Inst. Hautes Études Sci.<br />
(2008), no. 108, 1–181, With Appendix A, summarizing unpublished work of<br />
Russ Mann, and Appendix B by Marie-France Vignéras.<br />
[Clo91]<br />
[DG10]<br />
[DS05]<br />
[Eme99]<br />
[FO]<br />
L. Clozel, Représentati<strong>on</strong>s <strong>galois</strong>iennes associées aux représentati<strong>on</strong>s au<strong>to</strong>morphes<br />
au<strong>to</strong>duales de GL(n), Inst. Hautes Études Sci. Publ. Math. (1991), no. 73,<br />
97–145.<br />
M. Dimitrov and E. Ghate, Weight <strong>on</strong>e <strong>forms</strong> in Hida families, in preparati<strong>on</strong><br />
(2010).<br />
F. Diam<strong>on</strong>d and J. Shurman, A first course in modular <strong>forms</strong>, Graduate Texts<br />
in Mathematics, vol. 228, Springer-Verlag, New York, 2005.<br />
M. Emert<strong>on</strong>, A new proof of a theorem of Hida, Internat. Math. Res. Notices<br />
(1999), no. 9, 453–472.<br />
J.-M. F<strong>on</strong>taine and Y. Ouyang, Theory of p-adic Galois representati<strong>on</strong>s,<br />
Springer, forthcoming book.<br />
109
[F<strong>on</strong>94] J.-M. F<strong>on</strong>taine, Le corps des périodes p-adiques, Astérisque (1994), no. 223, 59–<br />
111, With an appendix by Pierre Colmez, Périodes p-adiques (Bures-sur-Yvette,<br />
1988).<br />
[Ger10]<br />
[Gha05]<br />
[GK1]<br />
[GK2]<br />
[GM09]<br />
[GV04]<br />
[Gre94]<br />
[Gro90]<br />
D. Geraghty, Modularity lifting theorems for ordinary Galois representati<strong>on</strong>s,<br />
preprint (2010).<br />
E. Ghate, Ordinary <strong>forms</strong> and their <strong>local</strong> Galois representati<strong>on</strong>s, Algebra and<br />
number theory, Hindustan Book Agency, Delhi, 2005, pp. 226–242.<br />
E. Ghate and N. Kumar, (p, p)-Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> au<strong>to</strong>morphic<br />
<strong>forms</strong> <strong>on</strong> GL n , Submitted.<br />
E. Ghate and N. Kumar, 2-adic Hida theory and applicati<strong>on</strong>s, Preprint.<br />
E. Ghate and A. Mézard, Filtered modules with coefficients, Trans. Amer. Math.<br />
Soc. 361 (2009), no. 5, 2243–2261.<br />
E. Ghate and V. Vatsal, On the <strong>local</strong> behaviour of ordinary Λ-adic representati<strong>on</strong>s,<br />
Ann. Inst. Fourier (Grenoble) 54 (2004), no. 7, 2143–2162.<br />
R. Greenberg, Trivial zeros of p-adic L-functi<strong>on</strong>s, p-adic m<strong>on</strong>odromy and the<br />
Birch and Swinnert<strong>on</strong>-Dyer c<strong>on</strong>jecture (Bost<strong>on</strong>, MA, 1991), C<strong>on</strong>temp. Math.,<br />
vol. 165, Amer. Math. Soc., Providence, RI, 1994, pp. 149–174.<br />
B. H. Gross, A tameness criteri<strong>on</strong> for Galois representati<strong>on</strong>s associated <strong>to</strong> modular<br />
<strong>forms</strong> (mod p), Duke Math. J. 61 (1990), no. 2, 445–517.<br />
[Hat79] K. Hatada, Eigenvalues of Hecke opera<strong>to</strong>rs <strong>on</strong> SL(2, Z), Math. Ann. 239 (1979),<br />
no. 1, 75–96.<br />
[Hen00]<br />
[Hid81]<br />
G. Henniart, Une preuve simple des c<strong>on</strong>jectures de Langlands pour GL(n) sur<br />
un corps p-adique, Invent. Math. 139 (2000), no. 2, 439–455.<br />
H. Hida, C<strong>on</strong>gruence of cusp <strong>forms</strong> and special values of their Zeta functi<strong>on</strong>s,<br />
Invent. Math. 63 (1981), no. 2, 225–261.<br />
[Hid86a] H. Hida, Galois representati<strong>on</strong>s in<strong>to</strong> GL 2 (Z p [[X]]) <strong>attached</strong> <strong>to</strong> ordinary cusp<br />
<strong>forms</strong>, Invent. Math. 85 (1986), no. 3, 545–613.<br />
[Hid86b] H. Hida, Iwasawa modules <strong>attached</strong> <strong>to</strong> c<strong>on</strong>gruences of cusp <strong>forms</strong>, Ann. Sci.<br />
École Norm. Sup. (4) 19 (1986), no. 2, 231–273.<br />
[Hid88a] H. Hida, Modules of c<strong>on</strong>gruence of Hecke algebras and L-functi<strong>on</strong>s associated<br />
with cusp <strong>forms</strong>, Amer. J. Math. 110 (1988), no. 2, 323–382.<br />
[Hid88b] H. Hida, On p-adic Hecke algebras for GL 2 over <strong>to</strong>tally real fields, Ann. of Math.<br />
(2) 128 (1988), no. 2, 295–384.<br />
[HT01]<br />
M. Harris and R. Taylor, The geometry and cohomology of some simple Shimura<br />
varieties, Annals of Mathematics Studies, vol. 151, Princet<strong>on</strong> University Press,<br />
Princet<strong>on</strong>, NJ, 2001, With an appendix by Vladimir G. Berkovich.<br />
110
[Kud94]<br />
[Kum]<br />
S. S. Kudla, The <strong>local</strong> Langlands corresp<strong>on</strong>dence: the n<strong>on</strong>-Archimedean case,<br />
Motives (Seattle, WA, 1991), Proc. Sympos. Pure Math., vol. 55, Amer. Math.<br />
Soc., Providence, RI, 1994, pp. 365–391.<br />
N. Kumar, A 2-adic c<strong>on</strong>trol theorem for modular curves, available at http:<br />
//www.math.tifr.res.in/~ganesh.<br />
[Kut80] P. Kutzko, The Langlands c<strong>on</strong>jecture for GL 2 of a <strong>local</strong> field, Ann. of Math. (2)<br />
112 (1980), no. 2, 381–412.<br />
[KW09]<br />
[Miy89]<br />
C. Khare and J.-P. Wintenberger, Serre’s modularity c<strong>on</strong>jecture. I, Invent.<br />
Math. 178 (2009), no. 3, 485–504.<br />
T. Miyake, Modular <strong>forms</strong>, Springer-Verlag, Berlin, 1989, Translated from the<br />
Japanese by Yoshitaka Maeda.<br />
[Nek06] J. Nekovář, Selmer complexes, Astérisque (2006), no. 310.<br />
[NSW08] J. Neukirch, A. Schmidt, and K. Wingberg, Cohomology of number fields, sec<strong>on</strong>d<br />
ed., Grundlehren der Mathematischen Wissenschaften [Fundamental Principles<br />
of Mathematical Sciences], vol. 323, Springer-Verlag, Berlin, 2008.<br />
[Och01]<br />
[Rib77]<br />
[Rod82]<br />
[Roh94]<br />
T. Ochiai, C<strong>on</strong>trol theorem for Greenberg’s Selmer groups of Galois deformati<strong>on</strong>s,<br />
J. Number Theory 88 (2001), no. 1, 59–85.<br />
K. A. Ribet, Galois representati<strong>on</strong>s <strong>attached</strong> <strong>to</strong> eigen<strong>forms</strong> with Nebentypus,<br />
Modular functi<strong>on</strong>s of <strong>on</strong>e variable, V (Proc. Sec<strong>on</strong>d Internat. C<strong>on</strong>f., Univ. B<strong>on</strong>n,<br />
B<strong>on</strong>n, 1976), Springer, Berlin, 1977, pp. 17–51. Lecture Notes in Math., Vol.<br />
601.<br />
F. Rodier, Représentati<strong>on</strong>s de GL(n, k) où k est un corps p-adique, Bourbaki<br />
Seminar, Vol. 1981/1982, Astérisque, vol. 92, Soc. Math. France, Paris, 1982,<br />
pp. 201–218.<br />
D. E. Rohrlich, Elliptic curves and the Weil-Deligne group, Elliptic curves and<br />
related <strong>to</strong>pics, CRM Proc. Lecture Notes, vol. 4, Amer. Math. Soc., Providence,<br />
RI, 1994, pp. 125–157.<br />
[Sai97] T. Sai<strong>to</strong>, Modular <strong>forms</strong> and p-adic Hodge theory, Invent. Math. 129 (1997),<br />
no. 3, 607–620.<br />
[Sav05] D. Savitt, On a c<strong>on</strong>jecture of C<strong>on</strong>rad, Diam<strong>on</strong>d, and Taylor, Duke Math. J.<br />
128 (2005), no. 1, 141–197.<br />
[Shi71]<br />
G. Shimura, Introducti<strong>on</strong> <strong>to</strong> the arithmetic theory of au<strong>to</strong>morphic functi<strong>on</strong>s,<br />
Publicati<strong>on</strong>s of the Mathematical Society of Japan, No. 11. Iwanami Shoten,<br />
Publishers, Tokyo, 1971, Kanô Memorial Lectures, No. 1.<br />
[Tay04] R. Taylor, Galois representati<strong>on</strong>s, Ann. Fac. Sci. Toulouse Math. (6) 13 (2004),<br />
no. 1, 73–119.<br />
111
[TY07]<br />
R. Taylor and T. Yoshida, Compatibility of <strong>local</strong> and global Langlands corresp<strong>on</strong>dences,<br />
J. Amer. Math. Soc. 20 (2007), no. 2, 467–493.<br />
[Tot96] B. Totaro, Tensor products in p-adic Hodge theory, Duke Math. J. 83 (1996),<br />
no. 1, 79–104.<br />
[TU99]<br />
[Urb05]<br />
[Wil88]<br />
J. Tilouine and E. Urban, Several-variable p-adic families of Siegel-Hilbert cusp<br />
eigensystems and their Galois representati<strong>on</strong>s, Ann. Sci. École Norm. Sup. (4)<br />
32 (1999), no. 4, 499–574.<br />
E. Urban, Sur les représentati<strong>on</strong>s p-adiques associées aux représentati<strong>on</strong>s cuspidales<br />
de GSp 4/Q , Astérisque (2005), no. 302, 151–176, Formes au<strong>to</strong>morphes.<br />
II. Le cas du groupe GSp(4).<br />
A. Wiles, On ordinary λ-adic representati<strong>on</strong>s associated <strong>to</strong> modular <strong>forms</strong>, Invent.<br />
Math. 94 (1988), no. 3, 529–573.<br />
112