TOWARDS A p-ADIC LANGLANDS PROGRAMME

by

Laurent Berger & Christophe Breuil

WARNING: These notes are informal and we apologize for the inaccuracies,
flaws and English mistakes that they surely contain.

Contents
1. Aim and contents of the course (C.B.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.1. Aim of the course . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2. Why Banach spaces? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3. Some notations and contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2. p-adic Hodge theory (L.B.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1. Some rings of periods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2. The rings Bcris and BdR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3. Crystalline representations in dimensions 1 and 2 . . . . . . . . . . . . 10
3. (ϕ, Γ)-modules (L.B.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.1. Some more rings of periods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2. The theory of the field of norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3. (ϕ, Γ)-modules and the operator ψ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4. (ϕ, Γ)-modules and p-adic Hodge theory (L.B.) . . . . . . . . . . . . . . . . . . . . . . 16
4.1. Overconvergent representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2. A large ring of periods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3. Crystalline representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.4. Eigenvectors of ψ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.5. A review of the notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5. Crystalline representations and Wach modules (L.B.) . . . . . . . . . . . . . . . . 22
5.1. Wach modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.2. The weak topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
5.3. The operator ψ and Wach modules . . . . . . . . . . . . . . . . . . . . . . . . . . 24
5.4. The module D0 (T ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.5. Wach modules and Dcris (V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6. Preliminaries of p-adic analysis (C.B.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.1. Functions of class C r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.2. Tempered distributions of order r . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
7. Crystalline representations of GL2 (Qp ) (C.B.) . . . . . . . . . . . . . . . . . . . . . . 33
7.1. Preliminary Banach spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2 LAURENT BERGER & CHRISTOPHE BREUIL

7.2. Definition of Π(V ) and first results . . . . . . . . . . . . . . . . . . . . . . . . . . 36

8. Representations of GL2 (Qp ) and (ϕ, Γ)-modules I (C.B.) . . . . . . . . . . . . 42
8.1. Back to (ϕ, Γ)-modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
8.2. Back to distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
9. Representations of GL2 (Qp ) and (ϕ, Γ)-modules II (C.B.) . . . . . . . . . . . . 48
9.1. The map Π(V )∗ → (limψ D(V ))b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
9.2. The map (limψ D(V ))b → Π(V )∗ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
9.3. Irreducibility and admissibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
10. Reduction modulo p (L.B.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
10.1. Statement of the conjecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
10.2. Lifting Dcris (V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
10.3. Continuity of the Wach module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

1. Aim and contents of the course (C.B.)

1.1. Aim of the course. — In 1998, M. Harris and R. Taylor ([25]), and slightly later
G. Henniart ([26]), proved the celebrated local Langlands correspondence for GLn (see
the course of Bushnell-Henniart for the case n = 2, which was known much earlier).
Let p be a prime number and K, E two finite extensions of Qp with E “big” (in
particular containing K). Recently, Schneider and Teitelbaum introduced a theory of
locally analytic and continuous representations of p-adic groups such as GLn (K) over
E-vector spaces (see their course and also [20]).
The aim of this course is to present some ongoing research on the (quite ambitious)
project of looking for a “Langlands type local correspondence” between, on the one
hand, some continuous representations of WK := Weil(Qp /K) on n-dimensional E-vector
spaces and, on the other hand, some continuous representations of GLn (K) on infinite-
dimensional E-vector spaces (although what we are going to study here is rather micro-
scopic with respect to such a programme!). Here and throughout, E will always denote
the coefficient field of the vector spaces on which the groups act. For some technical
reasons, we will keep in this course E finite over Qp but will always tacitly enlarge it if
necessary (which explains the above “big”).

Example 1.1.1. — The case n = 1 is just local class field theory. Let WK → E ×
be a continuous 1-dimensional representation, then, using the (continuous) reciprocity
ab ∼
isomorphism τK : WK → K × , we deduce a continuous representation K × → E × .

We would like to extend the case n = 1 to n > 1. One problem is that the category of
p-adic ( = continuous) finite dimensional representations of WK or GK := Gal(Qp /K) is
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 3

BIG, even for n = 2. Fontaine has defined important subcategories of p-adic representa-
tions of GK called crystalline, semi-stable and de Rham with crystalline ( semi-stable (
de Rham. They are important because they are related to algebraic geometry. Moreover,
these categories are now well understood. Therefore it seems natural to first look for
their counterpart, if any, on the GLn -side.
In this course, we will do so for n = 2, K = Qp and crystalline representations. As in
the `-adic case (i.e. the usual local Langlands correspondence), we will have an “F -semi-
simplicity” assumption. We will also need to assume that the Hodge-Tate weights of the
2-dimensional crystalline representation are distinct. Finally, mainly for simplicity, we
will also add the small assumption that the representation is generic (which means that
we do not want to be bothered by the trivial representation of GL2 (Qp ); the non-generic
cases behave a bit differently because of this representation). If f is an eigenform of weight
k ≥ 2 on Γ1 (N ) for some N prime to p, then the p-adic representation ρf |GQp of GQp
associated to f is known to be crystalline generic with Hodge-Tate weights (0, k − 1) and
is conjectured to be always “F -semi-simple” (and really is if k = 2), so these assumptions
should not be too restrictive. The case of equal Hodge-Tate weights correspond to weight
1 modular forms, that we thus disregard.
Our basic aim in this course is to define and start studying some p-adic Banach spaces
endowed with a continuous action of GL2 (Qp ), together with their locally analytic vectors.
These Banach will be associated to 2-dimensional “F -semi-simple” generic crystalline
representations of GQp with distinct Hodge-Tate weights. Independently of the quest for
a “p-adic Langlands programme”, we hope that these Banach spaces will also be useful in
the future for a “representation theoretic” study of the Iwasawa theory of modular forms
(see e.g. the end of §9.2).

1.2. Why Banach spaces?— We will focus on Banach spaces rather than on locally
analytic representations, although there are some very interesting locally analytic vectors
inside the Banach (and we will study them). In order to explain why, let us first recall,
without proof, some results in the classical `-adic case, due to M.-F. Vignéras.
Recall that, if ` is any prime number and E is a finite extension of Q` , an E-Banach
space is a topological E-vector space which is complete with respect to the topology given
by a norm (we don’t specify the norm in the data of a Banach space).

Definition 1.2.1. — Let G be a group acting E-linearly on an E-vector space V . We

say a norm || · || on V is invariant (with respect to G) if ||g · v|| = ||v|| for all g ∈ G and
v ∈V.
4 LAURENT BERGER & CHRISTOPHE BREUIL

Definition 1.2.2. — Let G be a topological group. We call a unitary G-Banach space

on E any E-Banach space Π equipped with an E-linear action of G such that the map
G × Π → Π, (g, v) 7→ g · v (v ∈ Π, g ∈ G) is continuous and such that the topology on Π
is given by an invariant norm.

Unitary G-Banach spaces on E form an obvious category where the morphisms are the
continuous E-linear G-equivariant maps.
Now, let us assume ` 6= p so that K is p-adic and E is `-adic.

Definition 1.2.3. — A smooth representation of GLn (K) on an E-vector space π is said

to be integral if there exists an invariant norm on π (with respect to GLn (K)).

It is easy to see that the absolutely irreducible smooth integral representations of

GLn (K) are exactly those absolutely irreducible smooth representations of GLn (K) that
correspond, under the classical local Langlands correspondence, to those F -semi-simple
`-adic representations of WK that extend to GK (recall that an `-adic representation of
WK on a finite dimensional E-vector space is the same thing by Grothendieck’s `-adic
monodromy theorem as a smooth representation of the Weil-Deligne group on the same
vector space).
Let π be a smooth integral representation of GLn (K) on an E-vector space and let || · ||
be an invariant norm on π. Then the completion of π with respect to || · || is obviously
a unitary GLn (K)-Banach space on E. Conversely, let Π be a unitary GLn (K)-Banach
space and denote by π ⊂ Π its subspace of smooth vectors (i.e. vectors fixed by a compact
open subgroup of GLn (K)). Then π is obviously a smooth integral representation of
GLn (K).

Theorem 1.2.4 (M.-F. Vignéras). — The functor Π 7→ π that associates to a unitary

GLn (K)-Banach space its subspace of smooth vectors induces an equivalence of categories
between the category of unitary GLn (K)-Banach spaces on E which are topologically of
finite length and the category of finite length smooth integral representations of GLn (K)
on E. An inverse functor π 7→ Π is provided by the completion with respect to any
invariant norm on π. Moreover, length(Π)= length(π).

In particular, all the invariant norms on an integral representation π of finite length in-
duce the same topology (in fact, Vignéras has proved that the lattices {v ∈ π | ||v|| ≤ 1}
for an invariant norm || · || are always finitely generated over OE [GL2 (Qp )], and thus
are all commensurable). Moreover π is irreducible if and only Π is topologically irre-
ducible. Thus, we see that the classical local Langlands correspondence can be refor-
mulated as a correspondence between (F -semi-simple) `-adic representations of GK and
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 5

unitary GLn (K)-Banach spaces on finite extensions of Q` that are topologically abso-
lutely irreducible. Hence, we could work with Banach spaces in the `-adic case, although
everybody agrees that smooth representations are much more convenient to handle than
Banach spaces. But we know, thanks to Th.1.2.4, that it wouldn’t make any difference.
We look for an analogous picture in the p-adic case, i.e. where E is a finite extension
of Qp and where the smooth vectors inside the unitary Banach spaces are replaced by the
locally analytic vectors. However, there is no straightforward generalization of Th.1.2.4
because:
1) a strongly admissible locally analytic representation of GLn (K) on E (see the course
of Schneider and Teitelbaum) can have infinitely many non-equivalent invariant norms
(ex: one can prove this is the case for the locally algebraic representation Symk−2 E 2 ⊗E
Steinberg of GL2 (Qp ) if k > 2)
2) thanks to a theorem of Schneider and Teitelbaum ([34]), we know that if Π is a
unitary GLn (K)-Banach on E that is admissible (as a representation of some compact
open subgroup) with locally analytic vectors π, then length(Π) ≤ length(π) (in fact,
unitarity is useless here). However, in general, this is only a strict inequality (ex: the
continuous Steinberg is topologically irreducible whereas the locally analytic Steinberg
has length 2).
So it seems natural in the p-adic case to rather focus on Banach spaces since the number
of Jordan-Hölder factors is smaller. Also, in the crystalline GL2 (Qp )-case, we will see that
the unitary GL2 (Qp )-Banach spaces are somewhat closer to the Galois representations
than their locally analytic vectors.

1.3. Some notations and contents. — Here are some notations we will use: p is a
prime number, Qp an algebraic closure of the field Qp of p-adic rationals, Zp the ring of
integers in Qp , Fp its residue field and GQp the Galois group Gal(Qp /Qp ). We denote by
× ×
ε : Gab
Qp → Zp the p-adic cyclotomic character, unr(x) the unramified character of Qp
sending p to x (whatever x is), val the p-adic valuation normalized by val(p) = 1 and
1
| · | = pval(·) the p-adic norm. The inverse of the reciprocity map Q× ab
p ,→ GQp is normalized
so that p is sent to a geometric Frobenius. In particular, for x ∈ Q×
p , we have ε(x) = x|x|
via this map. We let B(Qp ) ⊂ GL2 (Qp ) be the subgroup of upper triangular matrices.
We will loosely use the same notation for a representation and its underlying E-vector
space. We will use without comment basics of p-adic functional analysis ([30]) as well as
Schneider and Teitelbaum’s theory of locally analytic representations and the structure
theorem of locally analytic principal series for GL2 (Qp ) ([31],[32]).
Here is a rough description of the contents of the course.
6 LAURENT BERGER & CHRISTOPHE BREUIL

In lectures 2 to 5, the first author will introduce Fontaine’s rings Bcris and BdR and
will define crystalline and de Rham representations of GQp using the theory of weakly
admissible filtered modules (this restricted setting will be enough for our purposes). He
will give the explicit classification of F -semi-simple crystalline representations of GQp in
dimension 1 and 2 (in higher dimension, it becomes more involved). Then he will give
a second construction of crystalline representations of GQp using the theory of (ϕ, Γ)-
modules and he will describe the non trivial links between p-adic Hodge theory and
(ϕ, Γ)-modules. He will then introduce and study the Wach module of a crystalline
representation of GQp (a “smaller” module than the (ϕ, Γ)-module). These Wach modules
will be useful in the sequel for the applications to GL2 (Qp ).
In lectures 6 to 9, the second author will define and start studying the unitary GL2 (Qp )-
Banach spaces associated to 2-dimensional crystalline representations of GQp satisfying
the above assumptions (genericity, F -semi-simplicity and distinct Hodge-Tate weights).
He will first introduce the necessary material on distributions and functions on Zp (func-
tions of class C r , tempered distributions of order r, etc.) and will then turn to the
definition of the unitary GL2 (Qp )-Banach space Π(V ) associated to the (2-dimensional)
crystalline representation V . When V is reducible, Π(V ) is admissible, has topological
length 2 and is split (as a representation of GL2 (Qp )) if and only if V is split (as a
representation of GQp ). When V is absolutely irreducible, it seems hard to prove any-
thing on Π(V ) directly. For instance, it is not even clear in general on its definition that
Π(V ) 6= 0. To study Π(V ) for V irreducible, one surprisingly needs to use the theory of
(ϕ, Γ)-modules for V . More precisely, in that case we prove that there is a “canonical”
(modulo natural choices) E-linear isomorphism of vector spaces:

b
lim
←− D(V ) ' Π(V )∗
ψ

where D(V ) is the (ϕ, Γ)-module attached to V , ψ : D(V )  D(V ) is a certain canoni-
cal surjection (that will be described), b means “bounded sequences” and Π(V )∗ is the
Banach dual of Π(V ). Such an isomorphism (i.e. the link with (ϕ, Γ)-modules) was first
found by Colmez ([16]) in the case V is irreducible semi-stable non crystalline (and was
inspired by work of the second author [10], [11]) but we won’t speak of these cases here.
The left hand side of the above isomorphism being known to be non zero, we deduce
Π(V )∗ 6= 0 and thus Π(V ) 6= 0. The left hand side being “irreducible” and “admissible”
(what we mean here is explained in the text), we also deduce that Π(V ) is topologically
irreducible and admissible as a GL2 (Qp )-Banach space. It is hoped that this isomorphism
will be also useful in the future for a more advanced study of the representations Π(V ),
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 7

e.g. of their locally analytic vectors and of their reduction modulo the maximal ideal of
E.
Finally, in the last lecture, the first author will state a conjecture (contained in [9])
relating the semi-simplifications of Π(V ) and V modulo the maximal ideal of E and will
prove in some cases this conjecture using his Wach modules and previous computations
of the second author.
We didn’t include all the proofs of all the results stated or used in this text. But we
have tried to include as many proofs, or sketches of proofs, or examples, or references, as
our energy enabled us to. We apologize for the proofs that perhaps should be, and are
not, in this course.

2. p-adic Hodge theory (L.B.)

The aim of this lecture is to explain what crystalline representations are and to classify
them (at least in dimensions 1 and 2). Let us start with a p-adic representation V ,
that is a finite dimensional E-vector space endowed with a continuous linear action of
the group GQp . For the time being, we will consider the underlying Qp -vector space of
V and p-adic representations will be Qp -vector spaces. We want to single out certain
subcategories of the category of all p-adic representations. There is a general strategy for
cutting out subcategories of the category of p-adic representations. Suppose that we are
given a topological Qp -algebra B which is equipped with a continuous linear action of
GQp . We say that V is B-admissible if B ⊗Qp V = B dimQp (V ) as B[GQp ]-modules. The set
of all B-admissible representations is then a subset of the set of all p-adic representations.
Furthermore, if B is equipped with extra structures which commute with the action of
GQp then DB (V ) := (B⊗Qp V )GQp inherits those structures and provides us with invariants
which we can use to classify B-admissible representations. In this lecture, we will define
a ring of periods Bcris which has two extra structures: a filtration and a Frobenius map
ϕ. Hence, a Bcris -admissible representation V gives rise to a filtered ϕ-module Dcris (V )
and we will see that one can actually recover the representation V from Dcris (V ).

2.1. Some rings of periods. — Let Cp be the completion of Qp for the p-adic topology
and let
e = lim Cp = {(x(0) , x(1) , · · · ) | (x(i+1) )p = x(i) },
E ←−p
x7→x
e+
and let E be the set of x ∈ Ee such that x(0) ∈ OCp . If x = (x(i) ) and y = (y (i) ) are two
elements of E,
e we define their sum x + y and their product xy by:
j
(x + y)(i) = lim (x(i+j) + y (i+j) )p and (xy)(i) = x(i) y (i) ,
j→+∞
8 LAURENT BERGER & CHRISTOPHE BREUIL

which makes E e into a field of characteristic p. If x = (x(n) )n≥0 ∈ E,

e let vE (x) = vp (x(0) ).
This is a valuation on Ee for which Ee is complete; the ring of integers of E e +.
e is E
Let  = ((i) ) ∈ E
e + where (0) = 1 and (i) is a primitive pi th root of 1. It is easy to
see that Fp (( − 1)) ⊂ E e + [( − 1)−1 ] and one can show that E
e = E e is a field which is
the completion of the algebraic (non-separable!) closure of Fp (( − 1)), so it is really a
familiar object.
Let Ae + be the ring W (E
e + ) of Witt vectors with coefficients in E
e + and let
X
e+ = A
B e + [1/p] = { pk [xk ], xk ∈ E
e +}
k−∞

where [x] ∈ Ae + is the Teichmüller lift of x ∈ Ee + . The topology of A

e + is defined by taking
the collection of open sets {([π]k , pn )A
e + }k,n≥0 as a family of neighborhoods of 0, so that
Q e+ e + which to (xk )k assigns P pk [xk ] is a homeomorphism when
the natural map E → A
e + is given its valued field topology. This is not the p-adic topology of A
E e + , which makes
P k e + is given the discrete topology.
the map (xk )k 7→ p [xk ] homeomorphism when E
The ring Be + is endowed with a map θ : B e + → Cp defined by the formula
!
X X (0)
θ pk [xk ] = p k xk .
k−∞ k−∞

The absolute Frobenius ϕ : E e+ → E e + lifts by functoriality of Witt vectors to a map

ϕ:Be+ → Be + . It’s easy to see that ϕ(P pk [xk ]) = P pk [xp ] and that ϕ is bijective.
k
Recall that  = ((i) )i≥0 ∈ E
e + where (i) is as above, and define π = []−1, π1 = [1/p ]−1,
ω = π/π1 and q = ϕ(ω) = ϕ(π)/π. One can easily show that ker(θ : A e + → OCp ) is the
principal ideal generated by ω.
Here is a simple proof: obviously, the kernel of the induced map θ : E e + → OCp /p
e + such that vE (x) ≥ 1. Let y be any element of A
is the ideal of x ∈ E e + killed by θ
whose reduction modulo p satisfies vE (y) = 1. The map y Ae + → ker(θ) is then injective,
and surjective modulo p; since both sides are complete for the p-adic topology, it is an
isomorphism. Now, we just need to observe that the element ω is killed by θ and that
vE (ω) = 1.

2.2. The rings Bcris and BdR . — Using this we can define BdR ; let B+
dR be the ring
+
obtained by completing B for the ker(θ)-adic topology, so that
e

B+
dR = ←−
e + /(ker(θ))n .
lim B
n

In particular, since ker(θ) = (ω), every element x ∈ B+

dR can be written (in many ways)
P+∞ n + +
as a sum x = n=0 xn ω with xn ∈ B e . The ring B is then naturally a Qp -algebra. Let
dR
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 9

us construct an interesting element of this ring; since θ(1 − []) = 0, the element 1 − []
is “small” for the topology of B+
dR and the following series
+∞
X (1 − [])n
−
n=1
n

will converge in B+ dR , to an element which we call t. Of course, one should think of t as

t = log([]). For instance, if g ∈ GQp , then

g(t) = g(log([])) = log([g((0) , (1) , · · · )]) = log([ε(g) ]) = ε(g)t

so that t is a “period” for the cyclotomic character ε.

We now set BdR = B+
dR [1/t], which is a field that we endow with the filtration defined
by Fili BdR = ti B+ +
dR . This is the natural filtration on BdR coming from the fact that BdR
is a complete discrete valuation ring. By functoriality, all the rings we have defined are
GQ
equipped with a continuous linear action of GQp . One can show that BdRp = Qp , so that
if V is a p-adic representation, then DdR (V ) = (BdR ⊗Qp V )GQp is naturally a filtered
Qp -vector space. Note that dimQp (DdR (V )) ≤ d = dimQp (V ) in general.

Definition 2.2.1. — We say that V is de Rham if dimQp DdR (V ) = d.

If V is a de Rham representation, the Hodge-Tate weights of V are the integers h such

that Fil−h DdR (V ) 6= Fil−h+1 DdR (V ). The multiplicity of h is dim Fil−h DdR (V )/ Fil−h+1 DdR (V )
so that V has d Hodge-Tate weights.
One unfortunate feature of B+ dR is that it is too coarse a ring: there is no natural
e+ → B
extension of the natural Frobenius ϕ : B e + to a continuous map ϕ : B+ → B+ .
dR dR
+ (0) 1/p
For example, if p̃ ∈ E is an element such that p̃ = p, then θ([p̃ ] − p) 6= 0, so
e
that [p̃1/p ] − p is invertible in B+
dR , and so 1/([p̃
1/p
] − p) ∈ B+
dR . But if ϕ is a natural
+
e →B + 1/p
e , then one should have ϕ(1/([p̃ ] − p)) = 1/([p̃] − p), and since
extension of ϕ : B
/ B+
θ([p̃] − p) = 0, 1/([p̃] − p) ∈ dR . One would still like to have a Frobenius map, and there
is a natural way to complete B e + (where one avoids inverting elements like [p̃1/p ] − p) such
that the completion is still endowed with a Frobenius map.
e + is defined by taking the collection of open sets {([π]k , pn )A
Recall that the topology of B e + }k,n≥0
as a family of neighborhoods of 0. The ring B+ cris is defined as being
X ω n
B+cris = { an e + is sequence converging to 0},
where an ∈ B
n≥0
n!

and Bcris = B+
cris [1/t].
e + . The ring Bcris injects canonically
One could replace ω by any generator of ker(θ) in A
into BdR (B+ +
cris is naturally defined as a subset of BdR ) and, in particular, it is endowed
with the induced Galois action and filtration, as well as with a continuous Frobenius ϕ,
10 LAURENT BERGER & CHRISTOPHE BREUIL

extending the map ϕ : Be+ → Be + . For example, ϕ(t) = pt. Let us point out once again
e + = ∩+∞ ϕn (B+ ).
that ϕ does not extend continuously to BdR . One also sets B rig n=0 cris

Definition 2.2.2. — We say that a representation V of GQp is crystalline if it is Bcris -

admissible.

e + [1/t]-admissible (the periods of crys-

Note that V is crystalline if and only if it is B rig
talline representations live in finite dimensional Qp -vector subspaces of Bcris , stable by
ϕ, and so in fact in ∩+∞ n +
n=0 ϕ (Bcris )[1/t]); this is equivalent to requiring that the Qp -vector
space
e + [1/t] ⊗Qp V )GQp
Dcris (V ) = (Bcris ⊗Qp V )GQp = (B rig

be of dimension d = dimQp (V ). Then Dcris (V ) is endowed with a Frobenius ϕ induced by

that of Bcris and (BdR ⊗Qp V )GQp = DdR (V ) = Dcris (V ) so that a crystalline representation
is also de Rham and Dcris (V ) is a filtered Qp -vector space.
If V is an E-linear representation, then we can do the above constructions (of Dcris (V )
and DdR (V )) by considering V as a Qp -linear representation, and then one can check
that Dcris (V ) and DdR (V ) are themselves E-vector spaces and that: (1) ϕ : Dcris (V ) →
Dcris (V ) is E-linear (2) the filtration on DdR (V ) is given by E-vector spaces. The Hodge-
Tate weights of V are then by convention taken to be with multiplicity dimE Fil−h DdR (V )/ Fil−h+1 DdR (V
and so V has dimE (V ) Hodge-Tate weights.
Remark: one can also define a ring Bst and study semistable representations, but we
will not do so in this course.

2.3. Crystalline representations in dimensions 1 and 2. — It is not true that

every filtered ϕ-module D over E arises as the Dcris of some crystalline representation
V . If D is a filtered ϕ-module of dimension 1, and e is some basis, define tN (D) as the
valuation of the coefficient of ϕ in that basis. Define tH (D) as the largest integer h such
that Filh D 6= 0. If D is a filtered ϕ-module of dimension ≥ 1, define tN (D) := tN (det D)
and tH (D) := tH (det D). Note that if D = Dcris (V ), then tH (D) is minus the sum of the
Hodge-Tate weights. We say that D is an admissible filtered ϕ-module if tN (D) = tH (D)
and if for every subspace D0 of D stable under ϕ, we have tN (D0 ) − tH (D0 ) ≥ 0.
One can prove that if V is a crystalline representation, then the filtered ϕ-module
Dcris (V ) is admissible and conversely, Colmez and Fontaine proved the following (cf [17]):

Theorem 2.3.1. — If D is admissible, then there exists a crystalline representation V

of GQp such that Dcris (V ) = D. The functor V 7→ Dcris (V ) is then an equivalence of
categories, from the category of E-linear crystalline representations to the category of
filtered ϕ-modules over E.
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 11

This allows us to make a list of all crystalline representations: we just have to make
a list of all possible admissible filtered ϕ-modules. We will now do this in dimensions 1
and 2. Let D = Dcris (V ) be an admissible filtered ϕ-module.
If dimE (D) = 1 we can write D = E · e where ϕ(e) = pn λe with λ ∈ OE∗ and if h is
the Hodge-Tate weight of V , then Fil−h (D) = D and Fil−h+1 (D) = 0. The admissibility
condition in this case is nothing more than n = −h. We can then explicitly describe the
representation V . Let µλ be the unramified character of GQp sending σ to λ. One can
then check that V = εn µλ . If λ = 1 and E = Qp , then V = εn is called Qp (n) and it
has the following property: if V is a de Rham representation with Hodge-Tate weights
h1 , · · · , hd , then V (n) := V ⊗Qp Qp (n) has Hodge-Tate weights h1 + n, · · · , hd + n. The
representation V (n) is called a twist of V .
If dimE (D) = 2 we can once again make a list of the possible D’s. However, we won’t
be able to make a list of the corresponding V ’s because in general, it’s very hard to give
“explicitly” a p-adic representation. This is why the theory of filtered ϕ-modules is so
useful.
Given a crystalline representation V , we can always twist it by a suitable power of the
cyclotomic character, so that its Hodge-Tate weights are 0, k − 1 with k ≥ 1, so now we
assume this. Suppose furthermore that the Hodge-Tate weights of V are distinct, so that
k ≥ 2, and that ϕ : D → D is semi-simple (as we said in the introduction, we’re only
interested in those cases).
We can then enlarge E enough so that it contains the eigenvalues of ϕ, which we will
call α−1 and β −1 . We then have D = Eeα ⊕ Eeβ with ϕ(eα ) = α−1 eα and ϕ(eβ ) = β −1 eβ .
If α = β, we choose any basis of D.
We have tH (D) = −(k − 1) and tN (D) = − val(α) − val(β) so if D is admissible, then
val(α) + val(β) = k − 1. But this is not the only condition. The filtration on D is given
as follows: Fili D is D if i ≤ −(k − 1), it’s a line ∆ = Eδ if −(k − 2) ≤ i ≤ 0 and it’s
0 if i ≥ 1 (by definition of the Hodge-Tate weights). Note in particular that tH (eα ) and
tH (eβ ) are both either −(k − 1) or 0.
The admissibility condition says that − val(α) = tN (eα ) ≥ tH (eα ) ≥ −(k − 1) and
similarly for val(β). This implies that (k − 1) ≥ val(α) ≥ 0 and (k − 1) ≥ val(β) ≥ 0.
Suppose that val(β) ≤ val(α) (switch α and β if necessary). There are three cases to
consider.

1. 0 < val(β) ≤ val(α). If δ were eα or eβ , then ∆ would be a sub-ϕ-module of D and

the admissibility condition says that − val(α) or − val(β) should be ≥ tH (δ) = 0
which is a contradiction. Therefore δ cannot be eα nor eβ . In particular, this implies
that α 6= β. We can then rescale eα and eβ so that δ = eα + eβ .
12 LAURENT BERGER & CHRISTOPHE BREUIL

2. val(β) = 0 and val(α) = (k − 1). Then δ can be eβ or (up to rescaling as previously)

eα +eβ . Suppose it is eα +eβ . Then Eeα is a subobject of D which is itself admissible,
so it corresponds to a subrepresentation of V so that V is reducible. It is however
non-split, because D has no other admissible subobjects.
3. val(β) = 0 and val(α) = (k − 1) and δ = eβ . In this case, D is the direct sum of the
two admissible objects Eeα and Eeβ and so V is the direct sum of an unramified
representation and another unramified representation twisted by Qp (k − 1).
As an exercise, one can classify the remaining cases (when k = 1 or ϕ is not semisimple).
Finally, note that in the special case α + β = 0, it is possible to give an “explicit”
description of V .
To conclude this section, we note that if V is a representation which comes from a
modular form, whose level N is not divisible by p, then V is crystalline and we know
exactly what Dcris (V ) is in terms of the modular form. This makes the approach via
filtered ϕ-modules especially interesting for concrete applications.

3. (ϕ, Γ)-modules (L.B.)

In this lecture, we will explain what (ϕ, Γ)-modules are. The idea is similar to the
one from the previous lecture: we construct a ring of periods and we use it to classify
representations. In this case, we can classify all p-adic representations, as well as the GQp -
stable OE -lattices in them. The objects we use for this classification (the (ϕ, Γ)-modules)
are unfortunately much more complicated than filtered ϕ-modules and not very well
understood. In the next lecture, we will explain the links between the theory of filtered
ϕ-modules and the theory of (ϕ, Γ)-modules for crystalline representations.
Once more, we will deal with Qp -linear representations and then at the end say what
happens for E-linear ones (the advantage of doing this is that we can avoid an “extra
layer” of notations).

3.1. Some more rings of periods. — Let us go back to the field E e which was defined
e contains Fp (( − 1)) and is actually
in the previous lecture. As we pointed out then, E
isomorphic to the completion of Fp ((−1))alg . Let E be the separable closure of Fp ((−1))
(warning: not the completion of the separable closure, which by a theorem of Ax is
e and write EQp for Fp (( − 1)) (the reason for this terminology will
actually equal to E)
be clearer later on). Write GE for the Galois group Gal(E/EQp ) (a more correct notation
for GE would be GEQp but this is a bit awkward).
Let U be an Fp -representation of GE , that is a finite dimensional (=d) Fp -vector space
U with a continuous linear action of GE . Since U is a finite set, this means that GE acts
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 13

through an open subgroup. We can therefore apply “Hilbert 90” which tells us that the
E-representation E ⊗Fp U is isomorphic to Ed and therefore that D(U ) := (E ⊗Fp U )GE is
a d-dimensional EQp -vector space and that E ⊗Fp U = E ⊗EQp D(U ). In addition, D(U )
is a ϕ-module over EQp , meaning a vector space over EQp with a semilinear Frobenius ϕ
such that ϕ∗ D(U ) = D(U ).
It is therefore possible to recover U from D(U ) by the formula U = (E ⊗EQp D(U ))ϕ=1
and as a consequence, we get the following standard theorem going back at least to Katz:

Theorem 3.1.1. — The functor U → 7 D(U ) is an equivalence of categories from the

category of Fp -representations of GE to the category of ϕ-modules over EQp .

Now let AQp be the p-adic completion of Zp [[X]][1/X] inside A

e where we have written
X for π = [] − 1. We can identify AQp with the set of power series ∞ i
P
i=−∞ ai X where
ai ∈ Zp and a−i → 0 as i → ∞. The ring AQp is a Cohen ring for EQp . In particular,
BQp = AQp [1/p] is a field and we let B denote the p-adic completion of the maximal
e so that A = B ∩ A
unramified extension of BQp in B e is a Cohen ring for E.
Let T be a Zp -representation of GE , that is a finite type Zp -module with a continuous
linear action of GE . We can lift the equivalence of categories given by theorem 3.1.1 to get
the following one: the functor T 7→ D(T ) := (A ⊗Zp T )GE is an equivalence of categories
from the category of Zp -representations of GE to the category of ϕ-modules over AQp .
Finally, by inverting p we get:

Theorem 3.1.2. — The functor V 7→ D(V ) := (B ⊗Qp V )GE is an equivalence of cate-

gories from the category of Qp -representations of GE to the category of étale ϕ-modules
over BQp .

Notice the appearance of “étale” in the last statement. We say that a ϕ-module D over
BQp is étale if we can write D = BQp ⊗AQp D0 where D0 is a ϕ-module over AQp (which
includes the requirement ϕ∗ D0 = D0 ).
This way, we have a convenient way of studying representations of GE using ϕ-modules
over various rings. In the next paragraph, we will answer the question: what does this
have to do with representations of GQp ? The answer is given by an amazing construction
of Fontaine and Wintenberger.

3.2. The theory of the field of norms. — Recall that we have chosen a primitive
pn th root of unity (n) for every n. Set Fn = Qp ((n) ) and F∞ = ∪n≥0 Fn . The Galois
group ΓQp of F∞ /Qp is isomorphic to Z∗p (by the cyclotomic character). In this paragraph,
we will explain a striking relation between HQp , the Galois group of Qp /F∞ and GE the
14 LAURENT BERGER & CHRISTOPHE BREUIL

Galois group of E/EQp . Indeed, Fontaine and Wintenberger proved that we have a natural
isomorphism HQp ' GE . We will now recall the main points of their construction.
Let K be a finite extension of Qp and let Kn = K((n) ) as above. Let NK be the
lim Kn where the transition maps are given by NKn /Kn−1 , so that NK is the set of
set ←−
n
sequences (x(0) , x(1) , · · · ) with x(n) ∈ Kn and NKn /Kn−1 (x(n) ) = x(n−1) . If we define a ring
structure on NK by

(xy)(n) = x(n) y (n) and (x + y)(n) = lim NKn+m /Kn (x(n+m) + y (n+m) ),
m→+∞

then NK is actually a field, called the field of norms of K∞ /K. It is naturally endowed
with an action of HK . Furthermore, for every finite Galois extension L/K, NL /NK is
a finite Galois extension whose Galois group is Gal(L∞ /K∞ ), and every finite Galois
extension of NK is of this kind so that the absolute Galois group of NK is naturally
isomorphic to HK .
We will now describe NQp . There is a map NQp → E e which sends (x(n) ) to (y (n) ) where
m
y (n) = lim(x(n+m) )p and it is possible to show that this map is a ring homomorphism
and defines an isomorphism from NQp to EQp (the point is that by ramification theory,
the NKn /Kn−1 maps are pretty close to the x 7→ xp map). As a consequence of this
and similar statements for the NK ’s, we see that there is a natural bijection between
separable extensions of NQp and separable extensions of EQp . Finally, we get out of this
that HQp ' GE .
Let us now go back to the main results of the previous paragraph: we constructed an
equivalence of categories from the category of Zp -representations of GE to the category of
ϕ-modules over AQp and by inverting p an equivalence of categories from the category of
Qp -representations of GE to the category of étale ϕ-modules over BQp . Since HQp ' GE ,
we get equivalences of categories from the category of Zp -representations (resp. Qp -
representations) of HQp to the category of ϕ-modules over AQp (resp over BQp ).
Finally, if we start with a Zp -representation or Qp -representation of GQp , then what we
get is a ϕ-module over AQp (resp over BQp ) which has a residual action of GQp /HQp ' ΓQp .
This is an étale (ϕ, Γ)-module: it is a module over AQp (resp over BQp ) with a Frobenius
ϕ (required to be étale) and an action of ΓQp which commutes with ϕ. Note that on AQp ,
we have

ϕ(X) = ϕ([] − 1) = ([p ] − 1) = (1 + X)p − 1

and also

γ(X) = γ([] − 1) = ([ε(γ) ] − 1) = (1 + X)ε(γ) − 1.

 TOWARDS A p-ADIC LANGLANDS PROGRAMME 15

Theorem 3.2.1. — The functors T 7→ D(T ) and V 7→ D(V ) are equivalences of cate-
gories from the category of Zp -representations (resp. Qp -representations) of GQp to the
category of étale (ϕ, Γ)-modules over AQp (resp over BQp ).

This theorem is proved in [23], in pretty much the way which we recalled.

Example 3.2.2. — For example, let V = Qp (r) = Qp · er . The restriction of V to HQp

is trivial, so the underlying ϕ-module of V is trivial: D(V ) = BQp · er and ϕ(er ) = er .
The action of ΓQp is then given by γ(er ) = ε(γ)r er . In the lecture on Wach modules, we
will see non-trivial examples of (ϕ, Γ)-modules.

To conclude this paragraph, note that if E is a finite extension of Qp , and V is an

E-linear representation (or T is an OE -representation) then D(V ) (resp D(T )) is a (ϕ, Γ)-
module over E⊗Qp BQp (resp OE ⊗Zp AQp ) and the resulting functors are again equivalences
of categories.

3.3. (ϕ, Γ)-modules and the operator ψ. — Let V be a p-adic representation of GQp
and let T be a lattice of V (meaning that T is a free Zp -module of rank d = dimQp (V )
such that V = Qp ⊗Zp T and such that T is stable under GQp ).
In this paragraph, we will introduce an operator ψ : D(T ) → D(T ) which will play a
fundamental role in the sequel. Note that ϕ : B → B is not surjective, actually B/ϕ(B)
is an extension of degree p of local fields, whose residual extension is purely inseparable.
This makes it possible to define a left inverse ψ of ϕ by the formula:
1
ϕ(ψ(x)) = TrB/ϕ(B) (x)
p
and then some ramification theory shows that ψ(A) ⊂ A. Obviously, ψ(ϕ(x)) = x and
more generally ψ(λϕ(x)) = ψ(λ)x, and also ψ commutes with the action of GQp on B.
Therefore, we get an induced map ψ : D(T ) → D(T ) which is a left inverse for ϕ. One
can give a slightly more explicit description of ψ. Given y ∈ D(T ), there exist uniquely
Pp−1
determined y0 , · · · , yp−1 such that y = i=0 ϕ(yi )(1 + X)i and then ψ(y) = y0 . Finally,
if y = y(X) ∈ AQp , we can give a slight variant of the definition of ψ:
1X
ϕ(ψ(y)) = y((1 + X)η − 1).
p ηp =1

The Zp -module D(T )ψ=1 will be of particular importance to us. It turns out to be
of independent interest in Iwasawa theory. Indeed, it is possible to give explicit maps
h1n : D(T )ψ=1 → H 1 (Fn , T ) for all n ≥ 0, and these maps are compatible with each
other and corestriction (meaning that corFn+1 /Fn ◦h1n+1 = h1n ) so that we get a map
D(T )ψ=1 → ←−
lim H 1 (Fn , T ) and this map is, by a theorem of Fontaine, an isomorphism.
n
16 LAURENT BERGER & CHRISTOPHE BREUIL

lim H 1 (Fn , T ) is denoted by HIw

The Zp [[ΓQp ]]-module ←− 1
(Qp , T ) and is called the Iwasawa
n
cohomology of T . Explaining this further is beyond the scope of the course, but it is
an important application of the theory of (ϕ, Γ)-modules. One use we will make of the
previous discussion is the following: for any T , D(T )ψ=1 6= {0} (see corollary 9.2.3 for
1
another application). Indeed, in Iwasawa theory one proves that HIw (Qp , T ) is a Zp [[ΓQp ]]-
module whose torsion free part is of rank rkZp (T ).
Note that in a different direction, Cherbonnier and Colmez have proved that D(T )ψ=1
contains a basis of D(T ) on AQp . We see that therefore, the kernel of 1 − ψ on D(T ) is
rather large. On the other hand, Cherbonnier and Colmez have proved that D(T )/(1−ψ)
is very small: indeed, if V has no quotient of dimension 1, then D(T )/(1 − ψ) is finite
(and so D(V )/(1 − ψ) = 0). The reason is that we have an identification D(T )/(1 − ψ) =
2
HIw (Qp , T ). We will use this to prove the following proposition due to Colmez which will
be useful later on:

Proposition 3.3.1. — Let V be an E-linear representation such that Qp ⊗Qp V has no

quotient of dimension 1. If P ∈ E[X], then P (ψ) : D(V ) → D(V ) is surjective.

Proof. — By enlarging E if necessary, we can assume that P (X) splits completely in E

and so we only need to show the proposition when P (X) = X − α with α ∈ E. There
are several cases to consider:
1. α = 0. In this case for any x ∈ D(V ), we can write x = ψ(ϕ(x)).
2. vp (α) < 0. In this case, we use the fact that D(T ) is preserved by ψ so that the
series (1 − α−1 ψ)−1 = 1 + α−1 ψ + (α−1 ψ)2 + · · · converges and since (ψ − α)−1 =
−α−1 (1 − α−1 ψ)−1 we’re done.
3. vp (α) > 0. By the same argument as the previous one, (1 − αϕ)−1 converges on
D(V ). If x ∈ D(V ), we can therefore write ϕ(x) = (1 − αϕ)y and taking ψ, we get
x = (ψ − α)y.
4. vp (α) = 0. Let µα be the unramified character sending the frobenius to α. Then
D(V )/(ψ − α) = D(V ⊗ µα )/(1 − ψ) and as we have recalled above, D(V ⊗ µα )/(1 −
ψ) = 0 if Qp ⊗Qp V has no quotient of dimension 1.

4. (ϕ, Γ)-modules and p-adic Hodge theory (L.B.)

In this chapter, we study the link between p-adic Hodge theory (the theory of Bcris
and BdR ) and the theory of (ϕ, Γ)-modules. The latter theory provides a very flexible
way of studying p-adic representations, since to describe p-adic representations, one only
needs to give two matrices over BQp (one for ϕ and one for a topological generator of ΓQp
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 17

(which is procyclic if p 6= 2)) satisfying some simple conditions (which say that ϕ and
ΓQp commute and that ϕ is étale).

4.1. Overconvergent representations. — In order to link p-adic Hodge theory and

the theory of (ϕ, Γ)-modules, we will apply the usual strategy: construct more rings of
e be given and write x = P k
periods. Let x ∈ B k−∞ p [xk ] with xk ∈ E. Note that if
e
e then [y] ∈ B+ but still, there is no reason for the series x =
y ∈ E,
P
pk [xk ] to
dR k−∞
converge in B+
dR because the xk ’s could “grow too fast” (recall that E is a valued field).
e
We will therefore impose growth conditions on the xk ’s. Choose r ∈ R>0 and define
e †,r = {x ∈ B,
X p−1
B e x= pk [xk ], k + vE (xk ) → +∞}.
k−∞
pr

If rn = pn−1 (p − 1) for some n ≥ 0, then the definition of B e †,rn boils down to requiring
k (n)
P
that k−∞ p xk converge in Cp , which in turn is equivalent to requiring that the
1/pn
series ϕ−n (x) = k−∞ pk [xk ] converge in B+ †,r
P
dR . For example X ∈ B for all r’s and
−n 1/p n (n) t/pn +
ϕ (X) = [ ]−1= e − 1 ∈ BdR .
Let B†,r = B
e †,r ∩B and let B† = ∪r>0 B†,r . This is the subring of elements of B which are
“related to BdR ” in a way. We say that a p-adic representation V of GQp is overconvergent
if D(V ) has a basis which is made up of elements of D† (V ) = (B† ⊗Qp V )HQp . If there
is such a basis, then by finiteness it will actually be in D†,r (V ) = (B†,r ⊗Qp V )HQp for r
large enough.
We therefore need to know which p-adic representations are overconvergent, and this
is given to us by a theorem of Cherbonnier and Colmez (cf [12]):

Theorem 4.1.1. — Any p-adic representation V of GQp is overconvergent.

This theorem is quite hard, and we will not comment its proof. For de Rham repre-
sentations, one can give a simpler proof, based on the results of Kedlaya.
If V is overconvergent and if there is a basis of D(V ) in D†,r (V ), then D†,r (V ) is a
B†,r †,r HQp
Qp := (B ) -module. Warning: ϕ(B†,r ) ⊂ B†,pr so one has to be careful about the
fact that D†,r (V ) is not a ϕ-module. In any case, we need to know what B†,r
Qp looks like.
The answer turns out to be very nice. A power series f (X) ∈ BQp belongs to B†,r
Qp if and
only if it is convergent on the annulus 0 < vp (z) ≤ 1/r (the fact that f (X) ∈ BQp then
implies that it is bounded on that annulus). In particular, one can think “geometrically”
about the elements of B†,r
Qp and this turns out to be very convenient.
Given r > 0, let n(r) be the smallest n such that pn−1 (p − 1) ≥ r so that ϕ−n (x) will
converge in B+ for all x ∈ B
dR
e †,r and n ≥ n(r). We then have a map ϕ−n : D†,r (V ) →
(B+
dR ⊗ V )
HQp
and the image of this map is in (B+
dR ⊗ V )
HQp
if n ≥ n(r). This is used by
18 LAURENT BERGER & CHRISTOPHE BREUIL

Cherbonnier and Colmez to prove a number of reciprocity laws, but it is not enough for
our purposes, which is to reconstruct Dcris (V ) from D† (V ) if V is crystalline.
We do point out, however, that

Proposition 4.1.2. — If V is de Rham, then the image of the map ϕ−n : D†,r (V ) →
(B+
dR ⊗ V )
HQp
lies in Fn ((t)) ⊗Qp DdR (V ) and so if V is an E-linear de Rham representa-
tion, then the image of the map ϕ−n : D†,r (V ) → (B+
dR ⊗ V )
HQp
lies in (E ⊗Qp Fn ((t))) ⊗E
DdR (V ).

This will be used later on.

4.2. A large ring of periods. — In this paragraph, we take up the task of recon-
structing Dcris (V ) from D† (V ) if V is crystalline. For this purpose, we will need a few
P∞
more rings of periods. Let B†,r
rig,Qp be the ring of power series f (X) =
i
i=−∞ ai X where
ai ∈ Qp and f (X) converges on the annulus 0 < vp (z) ≤ 1/r (but is not assumed to be
bounded anymore). For example, t = log(1 + X) belongs to B†,r †,r
rig,Qp but not to BQp . Let
B†rig,Qp = ∪r>0 B†,r
rig,Qp . Note that this ring is often given a different name, namely RQp .
It is the “Robba ring”. As Colmez says, RQp is its “first name” and B†rig,Qp is its “last
name”. Similarly, let B†Qp = ∪r>0 B†,r † HQp
Qp = (B ) . The first name of that ring is EQ† p .
The main result which we have in sight is the following:

Theorem 4.2.1. — If V is a p-adic representation of GQp , then

 ΓQp
† †
Brig,Qp [1/t] ⊗B† D (V )
Qp

is a ϕ-module and as a ϕ-module, it is isomorphic to Dcris (V ). If V is crystalline, then

in addition we have
 ΓQp
B†rig,Qp [1/t] ⊗Qp B†rig,Qp [1/t] †
⊗B† D (V ) = B†rig,Qp [1/t] ⊗B† D† (V ).
Qp Qp

The proof of this result is quite technical, and as one suspects, it involves introducing
more rings of periods, so we will only sketch the proof. Let us mention in passing that
there are analogous results for semistable representations.
e + [1/t] ⊗Qp V )GQp as we
Recall that D† (V ) = (B† ⊗Qp V )HQp and that Dcris (V ) = (B rig
have explained in paragraph 2.2. The main point is then to construct a big ring B e†
rig
† † +
which contains both B (and hence B ) and B . This way, we have inclusions
e e
rig
 GQp
(1) e † [1/t] ⊗Qp V
Dcris (V ) ⊂ B rig

and
 ΓQp  GQp
(2) B†rig,Qp [1/t] ⊗B† D† (V ) ⊂ Be † [1/t] ⊗Qp V
rig .
Qp
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 19

We then need to prove that these two inclusions are equalities. There are essentially
two steps. the first is to use the fact that all the above are finite dimensional Qp -vector
spaces stable under ϕ and that ϕ−1 tends to “regularize” functions (think of the fact that
e †,r ) = B
ϕ(B e † are those which are in B
e †,rp ). The “most regular elements” of B e + which
rig rig
explains why inclusion (1) is an equality.
For inclusion (2), we first remark that
e † [1/t] ⊗Qp V )HQp = B
e† †
(B rig rig,Qp [1/t] ⊗B† D (V ),
Qp

e† e † HQp and in a way, the ring B

e† −m
(B†rig,Qp )
where B rig,Qp = (Brig ) rig,Qp is the completion of ∪m≥0 ϕ
e † . Inclusion (2) then becomes the statement that:
in B rig
 ΓQp  ΓQp
B†rig,Qp [1/t] ⊗B† D† (V ) = Be†
rig,Qp [1/t] ⊗ B† D †
(V ) ,
Qp Qp

and the proof of this is a “ϕ-decompletion” process, similar to the one used by Cherbon-
nier and Colmez to prove that all p-adic representations are overconvergent, and about
which we shall say no more.

4.3. Crystalline representations. — In the previous paragraph, we explained the

ideas of the (admittedly technical) proof of the facts that if V is a crystalline representa-
tion, then
 ΓQp
Dcris (V ) = B†rig,Qp [1/t] ⊗B† D† (V )
Qp

and
B†rig,Qp [1/t] ⊗Qp Dcris (V ) = B†rig,Qp [1/t] ⊗B† D† (V ).
Qp

In this paragraph, we will explain one important consequence of this theorem, namely
that p-adic representations are of finite height. First, let us explain what this means. Let
A+ = Ae + ∩ A and B+ = B e + ∩ B. since B+ ⊂ B e + , we have a natural map B+ → B+ and
dR
+ †,r
the rings B are “even better behaved” than the B ’s. In addition, they are stable under
ϕ. We say that a p-adic representation V is of finite height if D(V ) has a basis made of
elements of D+ (V ) := (B+ ⊗Qp V )HQp . This last space is a module over B+
Qp and one can
prove that B+
Qp = Qp ⊗Zp Zp [[X]] which is indeed a very nice ring. We will explain the
proof of Colmez’ theorem (cf [13]):

Theorem 4.3.1. — Every crystalline representation of GQp is of finite height.

This means that we can study crystalline representations effectively using Qp ⊗Zp
Zp [[X]]-modules which we will do in the next lecture.
Recall that Cherbonnier and Colmez have proved that D(V )ψ=1 contains a basis of
D(V ). The strategy for our proof that if V is crystalline, then V is of finite height is to
20 LAURENT BERGER & CHRISTOPHE BREUIL

prove that if V is crystalline, and S is the set

S = X, ϕ(X), ϕ2 (X), · · · ,
then D(V )ψ=1 ⊂ (B+ [S −1 ] ⊗Qp V )HQp . This will show that D(V ) has a basis of elements
which live in (B+ ⊗Qp V )HQp . This proof is due to Colmez; there is a different proof,
due to the first author, which relies on a result of Kedlaya. We will use the fact (due to
Cherbonnier) that D(V )ψ=1 ⊂ (B† ⊗Qp V )HQp .
For simplicity, we will assume that ϕ is semi-simple on Dcris (V ), and we will consider
V as an E-linear crystalline representation, where E contains the eigenvalues of ϕ on
Dcris (V ). Suppose then that Dcris (V ) = ⊕E · ei where ϕ(ei ) = αi−1 ei . If x ∈ D† (V ), then
xi ⊗ ei with xi ∈ E ⊗Qp B†rig,Qp [1/t] and if ψ(x) = x then, since ψ
P
we can write x =
acts by ϕ−1 on Dcris (V ), we have xi ∈ (E ⊗Qp B†rig,Qp [1/t])ψ=1/αi .
Let B+ i
P
rig,Qp be the ring of power series f (X) = i≥0 ai X which converge on the open
unit disk. The first name of that ring is R+
Qp . By an argument of p-adic analysis,
which we will give in paragraph 4.4, we can prove that for any α ∈ E, we have (E ⊗Qp
B†rig,Qp [1/t])ψ=α = (E ⊗Qp B+
rig,Qp [1/t])
ψ=α
. We conclude that D(V )ψ=1 ⊂ B+
rig,Qp [1/t] ⊗Qp
+ + ψ=1 +
e [1/t] ⊗Qp V )HQp . Let us
Dcris (V ). Finally, B
rig,Qp⊂Be so that we have D(V )
rig ⊂ (B rig
ψ=1 † HQp
recall that D(V ) ⊂ (B ⊗Qp V ) . This means that the periods of the elements of
ψ=1 +
e [1/t] ∩ B† and one can easily prove (using the definition of all these
D(V ) live in B rig
+ e + as “holomorphic
rings) that Brig [1/t] ∩ B† ⊂ B+ [S −1 ]. Indeed, one should think of B
e
rig
(algebraic) functions on the open unit disk” and of B† as “bounded (algebraic) functions
on some annulus”. The statement that B e + [1/t] ∩ B† ⊂ B+ [S −1 ] then says that a function
rig
which is both meromorphic with poles at the (n) − 1 and bounded toward the external
boundary is the quotient of a bounded holomorphic function on the disk (an element of
B+ ) by ϕn (X k ) for k, n large enough. Therefore, V is of finite height.
In other words, the (ϕ, Γ)-module of a crystalline representation has a very special
form: one can choose a basis such that the matrices of ϕ and γ ∈ ΓQp have coefficients
in B+ +
Qp (in E ⊗Qp BQp if V is E-linear).

4.4. Eigenvectors of ψ. — In this paragraph, we investigate further the action of

the operator ψ on certain rings of power series. The goal is to explain the proof of the
following proposition:

Proposition 4.4.1. — If α ∈ E ∗ , then:

(E ⊗Qp B†rig,Qp [1/t])ψ=α = (E ⊗Qp B+
rig,Qp [1/t])
ψ=α
.

Proof. — Since ψ(t−h f ) = ph t−h ψ(f ), it is enough to prove that

(E ⊗Qp B†rig,Qp )ψ=α = (E ⊗Qp B+
rig,Qp )
ψ=α
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 21

whenever vp (α) < 0 (if we are given f such that ψ(f ) = αf , then ψ(th f ) = p−h α · th f so
to make vp (α) < 0 we just need to take h  0) To avoid cumbersome formulas, we will
do the case E = Qp where the whole argument already appears.
† †
Recall that we have the following rings of power series: B+rig,Qp and Brig,Qp and BQp and
B+ +
Qp = Qp ⊗Zp Zp [[X]] and BQp as well as AQp and AQp = Zp [[X]]. Note that each of these
rings is stable under ψ. Let A− ai X i ∈ AQp such that ai = 0 if
P
Qp be the set of f (X) =
i ≥ 0 (this is not a ring!) so that AQp = A− + − −
Qp ⊕ AQp and let BQp = AQp [1/p].
Let us first prove that ψ(A− −
Qp ) ⊂ AQp . To see this, recall that q = ϕ(X)/X = 1 + [] +
· · · + []p−1 so that if ` ≥ 1, then ψ(q ` ) is a polynomial in [] of degree ≤ ` − 1 (since
ψ([i ]) = 0 if p - i and [i/p ] otherwise) and so ψ(q ` ) is a polynomial in X = [] − 1 of
degree ≤ ` − 1. We’ll leave it as an exercise to show that its constant term is p`−1 (this
will be useful later). Now observe that

ψ(X −` ) = ψ(q ` /ϕ(X)` ) = X −` ψ(q ` ),

so that ψ(X −` ) ∈ A− − −
Qp and consequently ψ(AQp ) ⊂ AQp .
If f ∈ B†rig,Qp then we can write f = f − + f + with f + ∈ B+
rig,Qp and f
−
∈ B−
Qp . In
order to prove our main statement, and since ψ commutes with f 7→ f ± , we therefore
only need to study the action of ψ on B− − −
Qp . But since ψ(AQp ) ⊂ AQp , ψ cannot have any
eigenvalues α with vp (α) < 0 on B−
Qp , so we are done.

†
Exercise: using this, determine all the eigenvalues of ψ on B−
Qp ∩ BQp .

4.5. A review of the notation. — The point of this paragraph is to review the
notation for the various rings that have been introduced so far. There are some general
guidelines for understanding “who is who”. The rings BdR and Bcris do not follow these
patterns though.
Most rings are E∗∗ or A∗∗ or B∗∗ . An “E” indicates a ring of characteristic p, an “A” a
ring of characteristic 0 in which p is not invertible, and a “B” a ring of characteristic 0
in which p is invertible.
The superscripts are generally + or † (or nothing). A “+” indicates objects defined on
the whole unit disk (in a sense) while a “†” indicates objects defined on an annulus (of
specified radius if we have a “†, r”) and nothing generally indicates objects defined “on
the boundary”. A “+” also means that X = π is not invertible. So a “†” indicates that
X is invertible but not too much (as on an annulus).
The subscript “rig” denotes holomorphic-like growth conditions toward the boundary,
while no subscript means that we ask for the much stronger condition “bounded”. The
subscript “Qp ” means we take invariants under HQp .
22 LAURENT BERGER & CHRISTOPHE BREUIL

A tilde “e” means that ϕ is invertible (eg in the algebraic closure E

e of EQp ) while no
tilde means that ϕ is not invertible (eg the separable closure E of EQp ).
Finally, as we said, BdR and Bcris do not follow these patterns, in part because they are
not defined “near the boundary of the open disk” like the other ones. It is better to use
e + [1/t] than the ring Bcris which is linked to crystalline cohomology but has no
the ring B rig
other advantage. The notation B e + is consistent with our above explanations. Note that
rig
this ring is also called B+
cont by some authors (with inconsistent notation).

5. Crystalline representations and Wach modules (L.B.)

This lecture will be a little different from the previous three: indeed we have (finally!)
defined all of the period rings which we needed and constructed most of the objects which
describe, in various ways, crystalline representations. It remains to construct the Wach
module associated to a crystalline representation V , and then to prove some technical
statements about the action of ψ on this module, which will be crucial for the applications
to representations of GL2 (Qp ).

5.1. Wach modules. — Recall from the previous lecture that if V is a crystalline
representation, then it is of finite height, meaning that we can write D(V ) = BQp ⊗B+
Qp

D+ (V ) where D+ (V ) = (B+ ⊗Qp V )HQp is a free B+

Qp = Qp ⊗Zp Zp [[X]]-module of rank d.
If V is an E-linear representation, then we have that D+ (V ) is a free E ⊗Qp B+
Qp = E ⊗OE
OE [[X]]-module of rank dimE (V ). Once again, we will deal with Qp -linear representations
in this chapter and at the end explain how everything remains true for E-linear objects
in an obvious way.
The starting point for this paragraph is that all crystalline representations are of finite
height, but there are finite height representations which are not crystalline (for example:
a non-integer power of the cyclotomic character). Wach has characterized which repre-
sentations V of finite height are crystalline: they are the ones for which there exists a B+
Qp -
submodule N of D+ (V ) which is stable under ΓQp and such that: (1) D(V ) = BQp ⊗B+ N
Qp

(2) there exists r ∈ Z such that ΓQp acts trivially on (N/X · N )(r).
Suppose that the Hodge-Tate weights of V are ≤ 0. One can then take r = 0 above and
one can further ask that N [1/X] = D+ (V )[1/X]. In this case, N is uniquely determined
by the above requirements and we call it the Wach module N (V ) associated to V . It is
then stable by ϕ. So to summarize, the Wach module N (V ) associated to V satisfies the
following properties:

Proposition 5.1.1. — The Wach module N (V ) ⊂ D(V ) has the following properties if
the weights of V are ≤ 0:
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 23

1. it is a B+
Qp -module free of rank d
2. it is stable under the action of ΓQp and ΓQp acts trivially on N (V )/X · N (V )
3. N (V ) is stable under ϕ.

Suppose that the Hodge-Tate weights of V are in the interval [−h; 0] and let ϕ∗ (N (V ))
be the B+ ∗
Qp -module generated by ϕ(N (V )). One can show that N (V )/ϕ (N (V )) is killed
by (ϕ(X)/X)h . Recall that we write q for the important element ϕ(X)/X.
Note that the Wach module of V (−h) is simply X h N (V )e−h (here e1 is a basis of
Qp (1) and ek = e⊗k1 ). If the Hodge-Tate weights of V are no longer ≤ 0 it is still
possible to define the Wach module of V by analogy with the above formula: we set
N (V ) = X −r N (V (−r))er where r is large enough so that V (−r) has negative Hodge-Tate
weights. This obviously does not depend on the choice of r. The module N (V ) ⊂ D(V )
is then stable under ΓQp and ΓQp acts trivially on N (V )/XN (V ) but N (V ) is no longer
stable under ϕ. What is true is the following:

Proposition 5.1.2. — If in prop 5.1.1, the Hodge-Tate weights of V are in the interval
[a; b] with b not necessarily ≤ 0, then ϕ(X b N (V )) ⊂ X b N (V ) and X b N (V )/ϕ(X b N (V ))
is killed by q b−a .

We finish this paragraph with a few examples. If V = Qp (r) then N (V ) = X −r B+

Qp er
which means that N (V ) is a B+ −r
Qp -module of rank 1 with a basis nr such that ϕ(nr ) = q nr
and γ(nr ) = (γ(X)/X)−r ε(γ)r nr if γ ∈ ΓQp .
Suppose now that V is the representation attached to a supersingular elliptic curve
(with ap = 0), twisted by Qp (−1) to make its weights ≤ 0. This representation is
crystalline and its Hodge-Tate weights are −1 and 0 (and the eigenvalues of ϕ on Dcris (V )
are ±p1/2 ). The Wach module N (V ) is of rank 2 over B+
Qp generated by e1 and e2 with
ϕ(e1 ) = qe2 and ϕ(e2 ) = −e1 . The action of γ ∈ ΓQp is given by:
log+ (1 + X) log− (1 + X)
γ(e1 ) = e1 and γ(e2 ) = e2
γ(log+ (1 + X)) γ(log− (1 + X))
where
Y ϕ2n+1 (q) Y ϕ2n (q)
log+ (1 + X) = and log− (1 + X) = ,
n≥0
p n≥0
p
so that t = log(1 + X) = X log (1 + X) log− (1 + X).
+

Finally, we point out that if V is an E-linear representation then N (V ) is a free

E ⊗ OE [[X]]-module satisfying all the properties given previously.

5.2. The weak topology. — Until now, we have largely ignored issues of topology,
but for the sequel it will be important to distinguish between two topologies on D(V ).
There is the strong topology which is the p-adic topology defined by choosing a lattice T
24 LAURENT BERGER & CHRISTOPHE BREUIL

of V and decreeing the p` D(T ) to be neighborhoods of 0. This does not depend on the
choice of T . Then, there is the weak topology which Colmez has defined for all p-adic
representations V , but which we will only define for crystalline ones, using the Wach
module. Choose a lattice T in V and define the Wach module associated to T to be
N (T ) = N (V ) ∩ D(T ). One can prove that N (T ) is a free Zp [[X]]-module of rank d such
that N (V ) = Qp ⊗Zp N (T ).
For each k ≥ 0, we define a semi-valuation νk on D(T ) as follows: if x ∈ D(T ) then
νk (x) is the largest integer j ∈ Z ∪ {+∞} such that x ∈ X j N (T ) + pk D(T ). The weak
topology on D(T ) is then the topology defined by the set {νk }k≥0 of all those semi-
valuations. Remark that the weak topology induces on N (T ) the (p, X)-adic topology.
The weak topology on D(V ) is the inductive limit topology on D(V ) = ∪`≥0 D(p−` T ).
This does not depend on the choice of T . Concretely, if we have a sequence (vn )n of
elements of D(V ), and that sequence is bounded for the weak topology, then there is a
GQp -stable lattice T of V such that vn ∈ D(T ) for every n ≥ 0 and furthermore, for every
k ≥ 0, there exists f (k) ∈ Z such that vn ∈ X −f (k) N (T ) + pk D(T ).
In lecture 8, we will use the weak topology on D(T ).

5.3. The operator ψ and Wach modules. — In this lecture, we assume that V is
a crystalline representation with ≥ 0 Hodge-Tate weights. The goal of this paragraph is
to show that

Theorem 5.3.1. — If V is a crystalline representation with ≥ 0 Hodge-Tate weights

and if T is a GQp -stable lattice in V , then
ψ(N (T )) ⊂ N (T )
and more generally, if ` ≥ 1, then
ψ(X −` N (T )) ⊂ p`−1 X −` N (T ) + X −`+1 N (T ).

Proof. — Say that the weights of V are in [0, h]. We know that ϕ(X h N (T )) ⊂ X h N (T )
and that X h N (T )/ϕ∗ (X h N (T )) is killed by q h . If y ∈ N (T ), then q h X h y ∈ ϕ∗ (X h N (T )) =
ϕ(X h )ϕ∗ (N (T )) and since by definition q h X h = ϕ(X h ) this shows that y ∈ ϕ∗ (N (T ))
and therefore N (T ) ⊂ ϕ∗ (N (T )). If y ∈ ϕ∗ (N (T )) then by definition we can write
P P
y= yi ϕ(ni ) with yi ∈ OE [[X]] and ni ∈ N (T ) and ψ(y) = ψ(yi )ni ∈ N (T ). This
proves that ψ(N (T )) ⊂ N (T ).
Now choose y ∈ N (T ) and ` ≥ 1. Once again, since N (T ) ⊂ ϕ∗ (N (T )), we can write
yi ϕ(ni ) with yi ∈ OE [[X]] and ni ∈ N (T ) so that ψ(X −` y) = ψ(X −` yi )ni .
P P
y =
Therefore, in order to prove that
ψ(X −` N (T )) ⊂ p`−1 X −` N (T ) + X −`+1 N (T ),
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 25

it is enough to prove that

ψ(X −` OE [[X]]) ⊂ p`−1 X −` OE [[X]] + X −`+1 OE [[X]],

which we did in §4.4.

5.4. The module D0 (T ). — The goal of this paragraph is to explain the proof of the
following proposition:

Proposition 5.4.1. — If V is crystalline irreducible and Qp ⊗Qp V has no quotient of

dim 1 (e.g. V is absolutely irreducible), then there exists a unique non-zero Qp -vector sub-
space D0 (V ) of D(V ) possessing an Zp -lattice D0 (T ) which is a compact (for the induced
weak topology) Zp [[X]]-submodule of D(V ) preserved by ψ and ΓQp with ψ surjective.

If V has a quotient of dim 1, then such an Zp [[X]]-module exists but is not unique. For
example, if V = Qp , the we can take either Qp ⊗Zp Zp [[X]] or X −1 Qp ⊗Zp Zp [[X]]. If V
is not crystalline, it still exists (this is a result of Herr) but in our case, the situation is
greatly simplified if we assume V to be crystalline. Indeed, in this case D0 (V ) is free of
rank d over Qp ⊗Zp Zp [[X]].

Proof. — First, we’ll prove the existence of such a module. Suppose that the Hodge-
Tate weights of V are in [−h; 0]. We can always twist V to assume this, because the
underlying ϕ-module of D(V ) is unchanged by twists. Choose a GQp -stable lattice T in
V . Our assumption implies that N (T ) ⊂ ϕ∗ (N (T )) and that ϕ∗ (X h N (T )) ⊂ X h N (T ) as
we’ve seen previously. As a consequence, ψ(N (T )) ⊂ N (T ) and X h N (T ) ⊂ ψ(X h N (T )).
Consider the sequence

X h N (T ), ψ(X h N (T )), ψ 2 (X h N (T )), · · ·

It is increasing and all its terms are contained in N (T ). Since Zp [[X]] is noetherian, this
implies that the sequence is eventually constant, and so there exists m0 such that if m ≥
m0 , then ψ m+1 (X h N (T )) = ψ m (X h N (T )). One can then set D10 (T ) = ψ m0 (X h N (T ))
and D10 (V ) = Qp ⊗Zp D10 (T ). This proves the existence of one module D10 (T ) (and
the associated D10 (V )) satisfying the conditions of the proposition. Before we prove
the uniqueness statement, we want to set D0 (V ) to be the largest such module and
first we need to prove some properties of those modules satisfying the conditions of the
proposition.
Suppose therefore that some module M (V ) satisfies the condition of the proposition,
that is we can write M (V ) = Qp ⊗Zp M (T ) with ψ : M (T ) → M (T ) surjective and
M (T ) compact in D(T ) for the induced weak topology. This last property means that
26 LAURENT BERGER & CHRISTOPHE BREUIL

for any k ≥ 0, there exists f (k) such that M (T ) ⊂ X −f (k) N (T ) + pk D(T ). Recall that
in proposition 5.3.1, we’ve proved that if f (k) ≥ 1, then:

ψ(X −f (k) N (T )) ⊂ pf (k)−1 X −f (k) N (T ) + X −f (k)+1 N (T ).

Since ψ(M (T )) = M (T ), we have

M (T ) = ψ m (M (T )) ⊂ pm(f (k)−1) X −f (k) N (T ) + X −f (k)+1 N (T ) + pk D(T )

and if m is large enough, we finally get that M (T ) ⊂ X −f (k)+1 N (T ) + pk D(T ). What this
shows is that we can take f (k) = 1 for all k ≥ 0 and so M (T ) ⊂ X −1 N (T ) + pk D(T ) for
all k ≥ 0 which implies that M (T ) ⊂ X −1 N (T ). Therefore, any such M (T ) is contained
in ∩m≥0 ψ m (X −1 N (T )).
On the other hand, if we set D0 (T ) = ∩m≥0 ψ m (X −1 N (T )), then obviously D0 (T )
satisfies the conditions of the proposition, and it contains D10 (T ) constructed previously
so it is certainly non-zero, and D0 (V ) is therefore a free B+
Qp -module of rank d contained
−1
in X N (V ).
Any other M (T ) satisfying the conditions of the proposition is included in D0 (T ). If
M (V ) was a free B+
Qp -module of rank < d, then we would have a submodule of N (V )
stable under ψ and Γ. One can prove that it would then have to be stable under ϕ (this
is easy to see in dimension d = 2, less so in higher dimensions) and this would correspond
to a subrepresentation of V , which we assumed to be irreducible. Therefore, any such
M (V ) is a B+
Qp -module of rank d contained in X
−1
N (V ).
Recall that by proposition 3.3.1, for any polynomial P , the map P (ψ) : D(V ) → D(V )
is surjective. By continuity, we can find a set Ω ⊂ D(T ) bounded for the weak topology
and ` ≥ 0 such that P (ψ)(Ω) ⊃ p` D0 (T ). This means that for any k ≥ 0, there exists
f (k) such that

p` D0 (T ) = ψ m (p` D0 (T )) ⊂ ψ m ◦ P (ψ)(X −f (k) N (T ) + pk D(T ))

for all m ≥ 0, and just as above this proves that p` D0 (T ) ⊂ P (ψ)X −1 N (T ) and then
that

p` D0 (T ) ⊂ P (ψ) ∩m≥0 ψ m (X −1 N (T )) = P (ψ)(D0 (T )),

so that ψ : D0 (V ) → D0 (V ) is surjective.
We now prove the uniqueness of an M (V ) satisfying the conditions of the proposi-
tion. By the previous discussion, D0 (V )/M (V ) is a finite dimensional Qp -vector space
stable under ψ and such that the operator P (ψ) is surjective for all P ∈ Qp [X] which is
impossible unless D0 (V ) = M (V ).
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 27

5.5. Wach modules and Dcris (V ). — We end this lecture with two ways of recovering
Dcris (V ) from the Wach module N (V ) of a crystalline representation V .
Let V be a crystalline representation. Recall that N (V ) ⊂ D(V ) is not necessarily
stable under ϕ, but that in any case ϕ(N (V )) ⊂ N (V )[1/q]. We can define a filtration
on N (V ) in the following way: Fili N (V ) = {x ∈ N (V ), ϕ(x) ∈ q i N (V )}. The results of
5.1 show that if the weights of V are in [a; b], then Fil−b N (V ) = N (V ).
Note also that since X and q = ϕ(X)/X are coprime, we have a natural identifica-
tion N (V )/X · N (V ) = N (V )[1/q]/X · N (V )[1/q] and we endow N (V )/X · N (V ) with
the induced filtration from N (V ), and the Frobenius induced from N (V )/X · N (V ) =
N (V )[1/q]/X · N (V )[1/q]. This makes N (V )/X · N (V ) into a filtered ϕ-module. The
main result is then (see [5, §III.4]):

Theorem 5.5.1. — The two filtered ϕ-modules N (V )/X · N (V ) and Dcris (V ) are iso-
morphic.

We shall now prove a technical result which will be used later on. Suppose now that
the weights of V are ≥ 0. We’ve seen that both N (V ) and X −1 N (V ) are stable under ψ,
so we get a map ψ : X −1 N (V )/N (V ) → X −1 N (V )/N (V ).

Proposition 5.5.2. — If we identify X −1 N (V )/N (V ) with Dcris (V ) by the map X −1 y 7→

y, then the map ψ : X −1 N (V )/N (V ) → X −1 N (V )/N (V ) coincides with the map ϕ−1 :
Dcris (V ) → Dcris (V ).

Proof. — Indeed, given y ∈ N (V ), choose z ∈ N (V )[1/q] such that y − ϕ(z) ∈ X ·

N (V )[1/q]. We then have ψ(y/X) = ψ(ϕ(z)/X + w) with w ∈ N (V ) and ψ(ϕ(z)/X) =
z/X so that ψ(y/X) ≡ z mod N (V )[1/q].

Now, we will give a different relation between Dcris (V ) and N (V ). Recall that we have
an isomorphism
B†rig,Qp [1/t] ⊗Qp Dcris (V ) ' B†rig,Qp [1/t] ⊗B† D† (V )
Qp

and in particular an inclusion

N (V ) ⊂ D† (V ) ⊂ B†rig,Qp [1/t] ⊗Qp Dcris (V ).
One can then prove that:

Proposition 5.5.3. — If V is crystalline, then

N (V ) ⊂ B+
rig,Qp [1/t] ⊗Qp Dcris (V ),

and
B+ +
rig,Qp [1/t] ⊗B+ N (V ) = Brig,Qp [1/t] ⊗Qp Dcris (V ).
Qp
28 LAURENT BERGER & CHRISTOPHE BREUIL

Furthermore, if the weights of V are ≥ 0, then

N (V ) ⊂ B+
rig,Qp ⊗Qp Dcris (V ).

Exercise: work out explicitly all the above identifications for the Wach modules given
at the end of 5.1.

6. Preliminaries of p-adic analysis (C.B.)

We give here some classical material of p-adic analysis which will crucially be used in
the coming lectures. The results can be found in [28], [14], [29] or any treatise on p-adic
analysis. Thus, we do not provide most of the proofs.

6.1. Functions of class C r . — Fix E a finite extension of Qp . We have extensively

seen in Schneider and Teitelbaum’s course what a locally analytic function f : Qp → E
is. Let r be an element in R+ . Here, we study the larger class of functions f : Zp → E
which are of class C r . For f : Zp → E any function and n ∈ Z≥0 , we first set:
n  
i n
X
an (f ) := (−1) f (n − i).
i=0
i

Definition 6.1.1. — A function f : Zp → E is of class C r if nr |an (f )| → 0 in R+ when

n → +∞.

Of course, if f is of class C r , then it is also of class C s for any s ≤ r. Functions of class

C 0 are functions f such that an (f ) → 0 in E, which can easily be seen to be equivalent to
f continuous (see e.g. [15],§V.2.1). Hence, a function of class C r for r ∈ R+ is necessarily
at least continuous. Moreover, one can write ([15],§V.2.1):
+∞  
X z
(3) f (z) = an (f )
n=0
n

where z ∈ Zp and nz := z(z−1)···(z−n+1)

n!
. The E-vector space of functions of class C r


is a Banach for the norm ||f ||r := supn (n + 1)r |an (f )| which contains locally analytic
functions on Zp , hence also locally polynomial functions. We denote this Banach by
C r (Zp , E).

Theorem 6.1.2. — Let d be an integer such that r − 1 < d, then the E-vector space of
locally polynomial functions f : Zp → E of degree at most d is dense in C r (Zp , E).

This theorem is a consequence of the Amice-Vélu condition as will be seen in §6.2

(Cor.6.2.7).
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 29

We now state results on the spaces C r (Zp , E) when r = n is in Z≥0 (although this is
actually probably a useless assumption for some of the results below).
For n ∈ Z≥0 and f : Zp → E any function, define f [n] (z, h1 , · · · , hn ) : Zp ×(Zp −{0})n →
E by induction as follows:
f [0] (z) := f (z)
f [n−1] (z + hn , h1 , · · ·, hn−1 ) − f [n−1] (z, h1 , · · ·, hn−1 )
f [n] (z, h1 , · · ·, hn ) := .
hn
Theorem 6.1.3. — A function f : Zp → E is of class C n if and only if the functions f [i]
for 0 ≤ i ≤ n extend continuously on Zi+1
p . Moreover, the norm sup0≤i≤n sup(z,h1 ,···,hi )∈Zi+1
p
|f [i] (z, h1 , · · ·, hi
is equivalent to the norm || · ||n .

See [15],§V.3.2 or [28],§54. Note that, in particular, any function f of class C n admits
a continuous derivative f 0 := f [1] (z, 0).

Theorem 6.1.4. — Assume n ≥ 1. The derivative map f 7→ f 0 induces a continuous

topological surjection of Banach spaces C n (Zp , E)  C n−1 (Zp , E) which admits a contin-
uous section.

Note that, with our definition, it is not transparent that f 0 ∈ C n−1 (Zp , E) (and indeed,
this requires a proof). We refer to [28],§§78 to 81. The topological surjection obviously
follows from the existence of a continuous section. Let us at least explicitly give such a
section Pn : C n−1 (Zp , E) ,→ C n (Zp , E):
n−1 (i)
+∞ X
X f (zj )
Pn f (z) := (zj+1 − zj )i+1
j=0 i=0
(i + 1)!
Pj−1 P+∞
where zj := l=0 al pl if z = l=0 al pl with al ∈ {0, · · · , p−1}. The fact P n f ∈ C n (Zp , E)
if f ∈ C n−1 (Zp , E) is proved in [28],§81. The fact P n is continuous is proved in [28],§79
for n = 1 and in [29],§11 for arbitrary n.
By an obvious induction, we obtain:

Corollary 6.1.5. — The n-th derivative map f 7→ f (n) induces a continuous topological
surjection of Banach spaces C n (Zp , E)  C 0 (Zp , E) which admits a continuous section.

Let d be an integer such that d ≥ n. We have already seen that the closure in C n (Zp , E)
of the vector subspace of locally polynomial functions of degree at most d is C n (Zp , E)
itself (Th.6.1.2). We we will need an analogous result when d < n:

Theorem 6.1.6. — Assume n ≥ 1 and d < n. The closure in C n (Zp , E) of the vector
subspace of locally polynomial functions of degree at most d is the closed subspace of
C n (Zp , E) of functions f such that f (d+1) = 0.
30 LAURENT BERGER & CHRISTOPHE BREUIL

This is proved in [28],§68 for n = 1 (and d = 0) and in [29],§8 for n and d arbitrary.

6.2. Tempered distributions of order r. — We let RE be the Robba ring of power

series with coefficients in E converging on an open annulus and R+
E ⊂ RE be the subring
P+∞
of power series converging on the open (unit) disk. We thus have R+ E = {
n
n=0 an X |
an ∈ E, limn |an |rn = 0 ∀ r ∈ [0, 1[} that we equip with the natural Fréchet topology
given by the collection of norms supn (|an |rn ) for 0 < r < 1. We start by recalling Amice’s
famous result:

Theorem 6.2.1. — The map:

+∞  
X z
(4) µ 7→ µ Xn
n=0
n
induces a topological isomorphism between the dual of the E-vector space of locally analytic
functions f : Zp → E and R+
E.

We now fix r ∈ R+ .

Definition 6.2.2. — We define the space of tempered distributions of order r on Zp as

the Banach dual of the Banach space C r (Zp , E).

People sometimes also say “tempered distributions of order ≤ r”. A tempered dis-
tribution of order r being also locally analytic (as C r (Zp , E) contains locally analytic
functions), the space of tempered distributions of order r corresponds by Th.6.2.1 to a
subspace of R+E.
P+∞
Definition 6.2.3. — An element w = n=0 an X
n
∈ R+ −r
E is of order r if n |an | is
bounded (in R+ ) when n varies.

The following corollary immediately follows from Def.6.1.1 and (3):

Corollary 6.2.4. — The map (4) induces a topological isomorphism between tempered
distributions of order r and the subspace of R+
E of elements w of order r endowed with
the norm ||w||r := supn ((n + 1)−r |an |).

Let µ be a locally analytic distribution, that is, an element of the dual of the E-
vector space of locally analytic functions f : Zp → E. If f : Zp → E is a locally
analytic function which is 0 outside of a + pn Zp (a ∈ Zp , n an integer), we use the
R
convenient notation a+pn Zp f (z)µ(z) := µ(f ). For any r ∈ R+ , we denote ||µ||r :=
supn (n + 1)−r | Zp nz µ(z)| ∈ R+ ∪ ∞. We define another norm ||µ||0r as follows:
R

Z
0 n(j−r)
||µ||r := supa∈Zp supj,n∈Z≥0 p (z − a)j µ(z) ∈ R+ ∪ ∞.
a+pn Zp
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 31

Lemma 6.2.5. — The two norms || · ||r and || · ||0r are equivalent (when defined).

P+∞
Proof. — We are going to use an intermediate norm. For w = m=0 am X m ∈ R +
E,
define:
00
||w||r := supn p−nr supm (|am [ pmn ]!|) ∈ R+ ∪ ∞

where [ pmn ] is the largest integer smaller than m

pn
. Then one can prove that there exists
00
c1 , c2 ∈ R+ such that c1 ||w||r ≤ ||w||r ≤ c2 ||w||r (this is purely an exercise in R+
E : see
m
[15],Lem.V.3.19 and note that |[ pmn ]!| is p− (p−1)pn up to a bounded scalar). If ||µ||r := ||w||r
00 00

00
when w is the Amice transform of µ, it is thus enough to prove that || · ||r is equivalent to
0
|| · ||r . Fix n ∈ Z≥0 and let:
Z  z − a j
||µ||0r,n := p −nr
supa∈Zp supj∈Z≥0 µ(z)
pn a+pn Zp
      
00 −nr z m
||µ||r,n := p supm∈Z≥0 µ !
m pn
00 00
(so that ||µ||0r = supn ||µ||0r,n and ||µ||r = supn ||µ||r,n ). Note that one can replace a ∈ Zp by
a ∈ {1 · · · , pn } in the definition of ||µ||0r,n . One can prove that each function mz [ pmn ]! can



j
be written as a (finite) linear combination of the functions 1a+pn Zp z−a p n (for j ∈ Z≥0
and a ∈ {1 · · · , pn }) with coefficients in Zp . Conversely, one can also write the function

j z
1a+pn Zp z−a

m
p n as an (infinite) linear combination of the functions [ ]! with coefficients
m pn
0
in Zp converging toward 0 (see [15],§V.3.1). This implies that ||µ||r,n is in R+ if and only
00 00 00 0
if ||µ||r,n is in R+ and, if so, ||µ||0r,n = ||µ||r,n . Likewise with ||µ||r and ||µ||r . This finishes
the proof.

Lemma 6.2.6. — Let µ be a linear form on the space of locally polynomial functions
f : Zp → E of degree at most d ∈ Z≥0 . Assume there exists a constant Cµ ∈ E such that
∀a ∈ Zp , ∀j ∈ {0, · · · , d} and ∀n ∈ Z≥0 :
Z
(5) (z − a)j µ(z) ∈ Cµ pn(j−r) OE .
a+pn Zp

If d > r − 1, there is a unique way to extend µ as a locally analytic distribution on Zp

such that ||µ||0r ∈ R+ (i.e. is bounded).

Proof. — We give the idea of the proof in the case d = 0 and r < 1, leaving the technical
details and the general case (which is analogous) to the reader. We will first define
32 LAURENT BERGER & CHRISTOPHE BREUIL

R
a+pn Zp
(z − a)j µ(z) by induction on j. For j = 1 (it is known for j = 0), we have:
Z X Z X Z
(z − a)µ(z) = (z − a)µ(z) = (z − b)µ(z) +
a+pn Zp b≡a(pn ) b+pn+1 Zp b≡a(pn ) b+pn+1 Zp

X Z X X Z
(b − a) µ(z) = (z − c)µ(z) +
b≡a(pn ) b+pn+1 Zp b≡a(pn ) c≡b(pn+1 ) c+pn+2 Zp

X X Z X Z
(b − c) µ(z) + (b − a) µ(z) = · · ·
b≡a(pn ) c≡b(pn+1 ) c+pn+2 Zp b≡a(pn ) b+pn+1 Zp

We have
Z
(b − c) µ(z) ∈ pn+1 Cµ p−r(n+2) OE = p(n+1)(1−r) p−r Cµ OE
c+pn+2 Zp

µ(z) ∈ pn(1−r) p−r Cµ OE . Decomposing again c + pn+2 Zp , we see

R
and (a − b) b+pn+1 Zp
R
that we get a converging sum of terms indexed by N of the type pN α+pN +1 Zp µ(z) ∈
pN (1−r) p−r Cµ OE with N growing together with a remainder which is a sum of terms of
R R
the type α+pN +1 Zp (z − α)µ(z). Since α+pN +1 Zp (z − α)µ(z) must be p-adically very small
if µ extends thanks to the condition ||µ||0r ∈ R+ (which in particular implies α+pN +1 Zp (z −
R

α)µ(z) ∈ ||µ||0r p(N +1)(1−r) OE ), we immediately see that we can canonically extend µ on
R
locally polynomial functions of degree at most 1 by setting a+pn Z(zp
− a)µ(z) :=the above
2
converging sum. Replacing (z − a) by (z − a) , etc., an easy induction shows that µ
extends in a similar way to all locally polynomial functions and by continuity to all
locally analytic functions and moreover that ||µ||0r ∈ R+ . Unicity also follows obviously
from the above decomposition and the condition ||µ||0r ∈ R+ .

The condition (5) in Lem.6.2.6 is sometimes called the “Amice-Vélu condition”.

Corollary 6.2.7. — 1) When d > r − 1, the subspace of locally polynomial functions

f : Zp → E of degree at most d is dense in C r (Zp , E).
2) Let µ be as in Lem.6.2.6 with d > r − 1, then µ uniquely extends as a tempered
distribution on Zp of order r.

Proof. — 1) If it is not dense, there exists a non zero µ which is tempered of order r and
vanishes on locally polynomial functions of degree at most d. Because of Lem.6.2.6, µ
vanishes on all locally polynomial functions, hence in particular on nz for any n, hence

µ has 0 as Amice transform which is impossible by Th.6.2.1 since µ 6= 0. 2) The unicity

follows from 1) and the existence follows from Lem.6.2.6, Lem.6.2.5 and Cor.6.2.4.

Assuming Th.6.1.6 and using Th.6.1.3, one can also prove that, when r is a positive
integer and d < r, any linear form µ as in Lem.6.2.6 extends uniquely to a continuous
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 33

linear form on the closed subspace of C r (Zp , E) of functions f such that f (d+1) = 0. We
leave this as an exercise to the reader.

7. Crystalline representations of GL2 (Qp ) (C.B.)

We define and start studying unitary GL2 (Qp )-Banach spaces Π(V ) associated to the
2-dimensional crystalline representations V of GQp as in the introduction. When V is
reducible, we prove that Π(V ) is admissible of topological length 2.

7.1. Preliminary Banach spaces. — Let V be a crystalline representation of GQp

on a two-dimensional E-vector space with distinct Hodge-Tate weights. Replacing V
by V ⊗E εn for some n ∈ Z, we can assume that the Hodge-Tate weights of V are
(0, k − 1) with k ∈ Z, k ≥ 2. Assuming moreover that V is F -semi-simple, i.e. that
the Frobenius ϕ on Dcris (V ) is semi-simple, we have seen in Lecture 2 that there are two
elements α, β ∈ OE such that α 6= β, val(β) ≤ val(α), val(α) + val(β) = k − 1 and
Dcris (V ) = D(α, β) = Eeα ⊕ Eeβ with ϕ(eα ) = α−1 eα , ϕ(eβ ) = β −1 eβ and:

1) V absolutely irreducible: val(α) > 0, val(β) > 0, Fili D(α, β) = D(α, β) if i ≤

−(k − 1), Fili D(α, β) = E(eα + eβ ) if −(k − 2) ≤ i ≤ 0 and Fili D(α, β) = 0 if i > 0,

2) V reducible and non-split: val(α) = k − 1, val(β) = 0, Fili D(α, β) = D(α, β) if

i ≤ −(k − 1), Fili D(α, β) = E(eα + eβ ) if −(k − 2) ≤ i ≤ 0 and Fili D(α, β) = 0 if i > 0,

3) V reducible and split: val(α) = k − 1, val(β) = 0, Fili D(α, β) = D(α, β) if

i ≤ −(k − 1), Fili D(α, β) = Eeβ if −(k − 2) ≤ i ≤ 0 and Fili D(α, β) = 0 if i > 0.

The genericity hypothesis alluded to in the introduction just means that we moreover
want α 6= pβ. Note that D(α, β) ' D(β, α) when val(α) = val(β).
We define now preliminary GL2 (Qp )-Banach spaces B(α), B(β), L(α), L(β) and, when
val(α) = k − 1, N (α).
Let B(α) be the following Banach space. Its underlying E-vector space is the vector
space of function f : Qp → E such that f |Zp is of class C val(α) and (αpβ −1 )val(z) z k−2 f (1/z) |Zp
can be extended as a function of class C val(α) on Zp . As a vector space we thus have:

(6) B(α) ' C val(α) (Zp , E) ⊕ C val(α) (Zp , E), f 7→ f1 ⊕ f2

34 LAURENT BERGER & CHRISTOPHE BREUIL

where, for z ∈ Zp , f1 (z) := f (pz) and f2 (z) := (αpβ −1 )val(z) z k−2 f (1/z). Hence B(α) is a
Banach space for the norm:


||f || := Max ||f1 ||C val(α) , ||f2 ||C val(α) .

We endow B(α) with an E-linear action of GL2 (Qp ) as follows:

   dz − b 
a b
(7) (f )(z) = α−val(ad−bc) (αpβ−1 )val(−cz+a) (−cz + a)k−2f
c d −cz + a
and to check that this action induces a continuous map GL2 (Qp ) × B(α) → B(α) is a
simple exercise that we leave to the reader (recall that by the Banach-Steinhaus The-
orem (see [30]) it is enough to check that the map GL2 (Qp ) → HomE (B(α), B(α)) is
continuous, where the right hand side is endowed with the weak topology of pointwise
convergence). Likewise, we define B(β) with a continuous action of GL2 (Qp ). Note that
we have a continuous GL2 (Qp )-equivariant injection:
 an
GL2 (Qp ) −1 k−2 −1
(8) LA(α) := IndB(Qp ) unr(α ) ⊗ x unr(pβ ) ,→ B(α)

given by:
 

0 1

(9) h : GL2 (Qp ) → E →
7 z ∈ Qp 7→ h( )
−1 z
where the left hand side is a locally analytic principal series in the sense of Schneider and
Teitelbaum ([31]) with left action of GL2 (Qp ) by right translation on functions. We thus
see that the Banach B(α) is nothing else than:
GL (Q )
C val(α) 00
“ IndB(Q2p ) p unr(α−1 ) ⊗ xk−2 unr(pβ −1 ) .

We have analogous properties with B(β) and LA(β) by interchanging α and β.

When val(α) = k − 1, let N (α) ( B(α) be the closed E-vector subspace of functions
(k−1) (k−1)
f such that f1 = f2 = 0 (k − 1-derivative) so that we have an exact sequence of
Banach spaces:

(10) 0 → N (α) → B(α) → C 0 (Zp , E)2 → 0

(k−1) (k−1)
where the third map sends (f1 , f2 ) to (f1 , f2 ). Note that this is well defined since
k−1
f1 and f2 are C on Zp (the topological surjection on the right follows from Cor.6.1.5).
Now, let us look at the action of GL2 (Qp ). When χ1 , χ2 are two characters on Q×p with
values in OE× (i.e. two integral characters), we denote by:
GL (Q )
C 0
IndB(Q2p ) p χ1 ⊗ χ2

the Banach of continuous functions h : GL2 (Qp ) → E such that:

 
a b
h( g) = χ1 (a)χ2 (d)h(g)
0 d
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 35

equipped with the topology of the norm ||h|| := Supg∈GL2 (Zp ) |h(g)| and with an action of
GL2 (Qp ) by right translation. It is easily seen to be a unitary GL2 (Qp )-Banach. We can
also describe this Banach as before using (9) in terms of continuous functions f : Qp → E
satisfying a certain continuity assumption at infinity.

Lemma 7.1.1. — Assume val(α) = k − 1 (hence val(β) = 0), then the map f 7→ f (k−1)
induces an exact sequence of GL2 (Qp )-Banach spaces on E:
GL (Q )
C 0
0 → N (α) → B(α) → IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 ) → 0.

Proof. — This is a standard and formal computation which is the same as in the locally
analytic context and that we leave to the reader. Note that the characters in the right
hand side parabolic induction are OE× -valued.

Let L(α) (resp. L(β)) be the following closed subspace of B(α) (resp. B(β)): it is the
closure of the E-vector subspace generated by the functions z j and (αpβ −1 )val(z−a) (z −
a)k−2−j for a ∈ Qp and j ∈ Z, 0 ≤ j < val(α) (resp. 0 ≤ j < val(β) with the convention
L(β) = 0 if val(β) = 0). The reader can check using (7) that L(α) (resp. L(β)) is
preserved by GL2 (Qp ). In order for L(α) (resp. L(β)) to be a subspace of B(α) (resp.
B(β)), it is enough to check (using (6) and up to a translation on z):

Lemma 7.1.2. — For 0 ≤ j < val(α), the function z 7→ (αpβ −1 )val(z) z k−2−j is of class
C val(α) on Zp . Likewise interchanging α and β.

Proof. — Fix j as in the statement and let f (z) := (αpβ −1 )val(z) z k−2−j which is easily
seen to extend to a continuous function on Zp by setting f (z) = 0 (look at the valuation
of f (z) for z 6= 0). Let f0 be the zero function on Zp and, for n ∈ Z, n > 0, consider
on Zp the functions fn (z) := (αpβ −1 )val(z) z k−2−j if val(z) < n and fn (z) := 0 otherwise.
It is clear that fn is of class C val(α) on Zp since it is locally polynomial. If we can prove
that fn+1 − fn → 0 in C val(α) (Zp , E) when n → +∞, we deduce that ∞
P
n=0 (fn+1 − fn ) ∈
C val(α) (Zp , E) since C val(α) (Zp , E) is complete. But this function is clearly f (as can be
checked for any z ∈ Zp ). If B is any Banach space, we have a closed embedding B ,→ (B ∗ )∗
(see [30],Lem.9.9), hence, to check that the sequence (fn+1 − fn )n converges toward 0 in
B, we can do it inside (B ∗ )∗ . It is thus enough to prove that for any tempered distribution
µ on Zp of order val(α), we have:
R
Zp
(fn+1 (z) − fn (z))µ(z)
sup −→ 0 when n → +∞.
µ ||µ||val(α)
36 LAURENT BERGER & CHRISTOPHE BREUIL

But:
Z Z Z 
−1 n k−2−j
(fn+1 (z) − fn (z))µ(z) = (αpβ ) z µ(z) − z k−2−j µ(z)
Zp pn Zp pn+1 Zp
−1
∈ ||µ||0val(α) pn(2val(α)−k+2) pn(k−2−j−val(α)) OE

using that val(α) + val(β) = k − 1 and that µ is of order val(α) (see §6.2). This implies
(f (z) − fn (z))µ(z) ∈ Cµ ||µ||val(α) −1 pn(val(α)−j) OE hence the result since j < val(α).
R
Zp n+1

If val(α) = k − 1, note that the functions f in L(α) give rise to pairs (f1 , f2 ) with zero
(k −1)-derivative, hence L(α) ⊆ N (α) ( B(α) in that case. If val(α) < k −1, the authors
do not know how to prove directly that L(α) is distinct from B(α).

7.2. Definition of Π(V ) and first results. — 1) Assume V absolutely irreducible,

or equivalently 0 < val(α) < k − 1. Then we define:

Π(V ) := B(α)/L(α)

with the induced action of GL2 (Qp ). We will prove in the next lectures that Π(V ) is
always non zero, but this will use (ϕ, Γ)-modules.

2) Assume V is reducible and non-split. This implies val(α) = k − 1 and val(β) = 0.

We define again:
Π(V ) := B(α)/L(α) 6= 0
with the induced action of GL2 (Qp ). By Lem.7.1.1, we have an exact sequence:
N (α) GL (Q )
C 0
(11) 0→ → Π(V ) → IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 ) → 0.
L(α)
We will identify below the Banach N (α)/L(α).

3) Assume V is reducible and split. This implies val(α) = k − 1 and val(β) = 0. Then
we define:
GL (Q )
C 0
Π(V ) := B(β) ⊕ IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 )
GL (Q )
C 0
= IndB(Q2p ) p unr(β −1 ) ⊗ xk−2 unr(pα−1 ) ⊕
GL (Q )
C 0
IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 ) .

In that case, Π(V ) can also be written:

GL (Q )
C 0
IndB(Q2p ) p unr(β −1 ) ⊗ εk−2 unr(pk−1 α−1 ) ⊕
GL (Q )
C 0
IndB(Q2p ) p ε(εk−2 unr(pk−1 α−1 )) ⊗ ε−1 unr(β −1 )
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 37

GL (Q ) 0 GL (Q ) 0
where one can see some symmetry (IndB(Q2p ) p χ1 ⊗ χ2 )C ⊕ (IndB(Q2p ) p εχ2 ⊗ ε−1 χ1 )C .
We will now try to “justify” these definitions, or at least explain where they are coming
from. We also start studying Π(V ).
GL (Q )
Let π(α) := Symk−2 E 2 ⊗E IndB(Q2p ) p unr(α−1 ) ⊗ unr(pβ −1 ) (the parabolic induction
is here the usual smooth one): π(α) is an irreducible locally algebraic representation of
GL2 (Qp ) in the sense of D. Prasad (see appendix to [32]). Likewise, we define π(β) :=
GL (Q )
Symk−2 E 2 ⊗E IndB(Q2p ) p unr(β −1 ) ⊗ unr(pα−1 ). There are closed GL2 (Qp )-equivariant
embeddings π(α) ,→ LA(α) (resp. π(β) ,→ LA(β)) identifying the source with the
subspace of functions f : Qp → E such that f |Zp and (αpβ −1 )val(z) z k−2 f (1/z) |Zp (resp.
(βpα−1 )val(z) z k−2 f (1/z) |Zp ) are locally polynomial functions on Zp of degree at most k −2
(using (9)). Moreover, the usual intertwining operator between smooth generic principal
series induces a GL2 (Qp )-equivariant isomorphism:
(12) I : π(α) ' π(β).
See (24) and (25) later in the text for an explicit description of I.
GL (Q )
Remark 7.2.1. — The smooth principal series IndB(Q2p ) p unr(α−1 ) ⊗ unr(pβ−1) is natural
to introduce here since it is exactly the smooth representation of GL2 (Qp ) that corre-
sponds under the local Langlands correspondence (slightlytwisted)
 to the unramified
α 0
representation of WQp sending an arithmetic Frobenius to = matrix of ϕ−1 on
0 β
D(α, β).

Theorem 7.2.2. — 1) Assume val(α) < k −1 (hence val(β) > 0) then B(α)/L(α) (resp.
B(β)/L(β)) is isomorphic to the completion of π(α) (resp. π(β)) with respect to any OE -
lattice of π(α) (resp. π(β)) which is finitely generated over GL2 (Qp ).
2) Assume val(α) = k−1 (hence val(β) = 0) , then N (α)/L(α) (resp. B(β)) is isomorphic
to the completion of π(α) (resp. π(β)) with respect to any OE -lattice of π(α) (resp. π(β))
which is finitely generated over GL2 (Qp ).

Note that we don’t know a priori in 1) that such a lattice exists in π(α) or π(β). In
case not, the statement means B(α)/L(α) = 0 = B(β)/L(β). Before giving the proof of
Theorem 7.2.2, we mention the following four corollaries:

Corollary 7.2.3. — Assume val(α) < k−1, then B(α)/L(α) and B(β)/L(β) are unitary
GL2 (Qp )-Banach spaces and we have a commutative GL2 (Qp )-equivariant diagram:
B(α)/L(α) ' B(β)/L(β)
↑ ↑
I
π(α) ' π(β)
where I is the intertwining operator of (12).
38 LAURENT BERGER & CHRISTOPHE BREUIL

Proof. — The statement follows from Th.7.2.2 together with the fact that the isomor-
phism I obviously sends an OE -lattice of π(α) of finite type over GL2 (Qp ) to an OE -lattice
of π(β) of finite type over GL2 (Qp ).

We will see later on that B(α)/L(α) and B(β)/L(β) are topologically irreducible and
admissible.

Corollary 7.2.4. — Assume val(α) = k − 1, then we have a commutative GL2 (Qp )-

equivariant diagram:
N (α)/L(α) ' B(β)
↑ ↑
I
π(α) ' π(β)
where I is the intertwining operator of (12). Moreover, N (α)/L(α) is a topologically
irreducible admissible unitary GL2 (Qp )-Banach and B(α)/L(α) is an admissible unitary
GL2 (Qp )-Banach of topological length 2 which is a non-trivial extension of the Banach
GL (Q )
C 0
IndB(Q2p ) pxk−1unr(α−1) ⊗ x−1unr(pβ−1) by the Banach N (α)/L(α) ' B(β).

Proof. — The first part follows from Th.7.2.2 as previously. The unitary GL2 (Qp )-
Banach B(β) is admissible and topologically irreducible. The admissibility is true for
GL (Q )
C 0
any Banach IndB(Q2p ) p χ1 ⊗ χ2 with integral χi and follows from the fact that re-
GL (Z )
C 0
striction of functions to GL2 (Zp ) induces an isomorphism with IndB(Z2p ) p χ1 ⊗ χ2 ,
the   which is E ⊗ (OE [[GL2 (Zp )]] ⊗OE [[B(Zp )]] OE ) (the map B(Zp ) → OE be-
dual of
a b
ing 7→ χ1 (a)χ2 (d)) which is obviously of finite type over E ⊗OE OE [[GL2 (Zp )]]
0 d
(this proof is extracted from [33]). The irreducibility can be proved as follows: if B
is a closed non zero GL2 (Qp )-equivariant E-vector subspace in B(β), then B is also
admissible and B an 6= 0 by the density of locally analytic vectors (see Schneider and
Teitelbaum’s course). But the locally analytic vectors in B(β) are just LA(β) hence
π(β) ⊂ B since π(β) is the unique irreducible subobject of LA(β). But π(β) is dense
by Prop.7.2.2, hence B = B(β). This proves the statement on N (α)/L(α). The proof
that B(α)/L(α) is admissible of topological length 2 is analogous. The exact sequence
is non-split because the analogous exact sequence with locally analytic vectors is non-
split thanks to Cor.7.2.6 below (and the structure of locally analytic principal series for
GL2 (Qp )). Finally, let us prove that B(α)/L(α) is unitary. We will use the fact that the
GL (Q )
C 0
surjection B(α)/L(α)  IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 ) admits a continuous
section (not GL2 (Qp )-equivariant of course). This follows from (6), (10) and the fact
that, for any positive integer n, the nth -derivation map C n (Zp , E)  C 0 (Zp , E) admits a
continuous section (Cor.6.1.5). Thus, as a Banach space, we can write B(α)/L(α) as:
GL (Q )
C 0
(13) B(β) ⊕ IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 ) .
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 39

 
1 0
Since the matrix acts continuously on B(α)/L(α), replacing the unit ball B(β)0
0 p
of B(β) by p−n B(β)0 for a convenient integer n, we can assume, using (13) and the
fact that the two Jordan-Hölder
 factors
 of B(α)/L(α) are unitary, that the unit ball of
1 0
B(α)/L(α) is preserved by and GL2 (Zp ), hence by GL2 (Qp ) using the Cartan
0 p
decomposition and the integrality of the central character.

Corollary 7.2.5. — Assume val(α) < k − 1 and B(α)/L(α) 6= 0 (this fact will be proved
later on), then we have a continuous GL2 (Qp )-equivariant injection:

LA(α) ⊕π(α) LA(β) ,→ (B(α)/L(α))an

where π(α) embeds into LA(β) via the intertwining (12).

Proof. — We have a continuous equivariant injection LA(α) ,→ (B(α)/L(α))an by (8) (it

is injective since, otherwise, π(α) would necessarily map to zero and we know this is not
true because of Th.7.2.2). Likewise with β. The corollary then follows from Cor.7.2.3.

Because of Cor.7.2.6 below, I am tempted to conjecture that the injection in Cor.7.2.5 is

actually a topological isomorphism. The statement of Cor.7.2.5 (without the description
of the completion of π(α) as B(α)/L(α)) was also noted in [19],Prop.2.5.

Corollary 7.2.6. — Assume val(α) = k − 1, then we have a topological GL2 (Qp )-

equivariant isomorphism:
∼
LA(α) ⊕π(α) LA(β) → (B(α)/L(α))an

where π(α) embeds into LA(β) via the intertwining (12).

Proof. — Since all the Banach in (11) are admissible by Cor.7.2.4, a theorem of Schneider
and Teitelbaum tells us that we have an exact sequence of locally analytic representations
from the sequence (11):
 B(α) an
an
GL (Q )
(14) 0 → LA(β) → → IndB(Q2p ) p xk−1unr(α−1 ) ⊗ x−1unr(pβ−1 ) →0
L(α)
where we have used (N (α)/L(α)an ' B(β)an ' LA(β). But we also have a continuous
GL2 (Qp )-equivariant injection LA(α) ,→ (B(α)/L(α))an as in Cor.7.2.5, hence a contin-
uous GL2 (Qp )-equivariant map LA(α) ⊕π(α) LA(β) → (B(α)/L(α))an . To prove it is a
topological bijection, note that from the exact sequence of locally analytic representa-
tions:
GL (Q )
an
0 → π(α) → LA(α) → IndB(Q2p ) p xk−1 unr(α−1 ) ⊗ x−1 unr(pβ −1 ) → 0,
40 LAURENT BERGER & CHRISTOPHE BREUIL

we deduce an exact sequence similar to (14) with (B(α)/L(α))an replaced by LA(α) ⊕π(α)
LA(β)), together with an obvious commutative diagram between the two exact sequences.

Proof of Th.7.2.2: Note first that the completion of π(α) doesn’t depend on which
lattice of finite type is chosen (if any) since all these lattices are commensurable and
thus give rise to equivalent invariant norms. Moreover, this completion is also the
same as the completion with respect to any OE -lattice finitely generated over B(Qp ).
Indeed, any OE -lattice of finite type over GL2 (Qp ) is of finite type over B(Qp ) since
GL2 (Qp ) = B(Qp )GL2 (Zp ) and the GL2 (Zp )-span of any vector in π(α) is finite dimen-
sional. Conversely, one can check using the compacity of GL2 (Zp ) that any OE -lattice of
finite type over B(Qp ) is contained in an OE -lattice of finite type over GL2 (Qp ) (and is
thus commensurable to it). If π(α) admits an OE -lattice preserved by B(Qp ), then one
can check that the following OE -submodule:
k−2
X k−2
X
OE [B(Qp )]z j 1Zp + OE [B(Qp )](αpβ −1 )val(z) z j 1Qp −pZp
j=0 j=0

(viewing π(α) as embedded into LA(α)) is necessarily an OE -lattice of π(α) which is

finitely generated over B(Qp ). If π(α) doesn’t admit any OE -lattice preserved by B(Qp ),
then one can check that the above OE -submodule is the E-vector space π(α). The dual
of the sought after completion of π(α) is thus isomorphic to:

(15) {µ ∈ π(α)∗ | ∀g ∈ B(Qp ), ∀j ∈ {0, · · · , k − 2}, |µ(g(z j 1Zp ))| ≤ 1

and |µ(g((αpβ −1 )val(z) z j 1Qp −pZp ))| ≤ 1} ⊗OE E.
 
1 Qp
Granting the fact that the central character is integral, we can even take g ∈ ×
R 0 Qp
in (15). For f ∈ π(α), seeing f as a function on Qp via (9), we write Qp f (z)µ(z)
instead of µ(f ) in the sequel. A short computation then gives that the conditions in
(15) are equivalent to the existence for each µ of a constant Cµ ∈ E such that ∀a ∈ Qp ,
∀j ∈ {0, · · · , k − 2} and ∀n ∈ Z, n > val(a) if a 6= 0:
Z
(16) (z − a)j µ(z) ∈ Cµ pn(j−val(α)) OE
a+pn Zp
Z
αp
val(z−a)
(17) (z − a)k−2−j µ(z) ∈ Cµ pn(val(α)−j) OE
n
Qp −(a+p Zp ) β
R R
where U f (z)µ(z) := Qp f (z)1U (z)µ(z) for any open U ⊂ Qp . Let us now first assume
val(α) < k − 1. From (16), we deduce (modifying Cµ if necessary):
Z
αp
val(z−a)
(18) (z − a)k−2−j µ(z) ∈ Cµ pn(val(α)−j) OE .
(a+p n−1 n
Zp )−(a+p Zp ) β
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 41

Writing Qp − (a + pn Zp ) = Qp − (a + pn−1 Zp ) q (a + pn−1 Zp ) − (a + pn Zp ), using (18) for

the right hand side and decomposing again Qp − (a + pn−1 Zp ) = Qp − (a + pn−2 Zp ) q (a +

val(z−a)
pn−2 Zp ) − (a + pn−1 Zp ), we see by induction that Qp −(a+pn Zp ) αp
R
β
(z − a)k−2−j µ(z)
for j > val(α) is uniquely determined and that (17) for j > val(α) is a consequence of
(16). Furthermore, we see by the same argument that (17) for j = val(α) (if val(α) is a
positive integer) and all a follows from (17) for j = val(α), a = 0 and from (16) (develop
(z − a)k−2−j when n  0). Moreover, we know by Lem.6.2.6 that (16) for j > val(α) is
a consequence of (16) for j ≤ val(α). Hence, (16) and (17) are equivalent to: (16) for
j ≤ val(α) and a ∈ Qp + (17) for j < val(α) and a ∈ Qp + (17) for j = val(α) and
a = 0. Any µ ∈ π(α)∗ is equivalent to the data of a pair (µ1 , µ2 ) where, if f ∈ π(α),
f (z)µ(z) = Zp f (pz)µ1 (z) + Zp (αpβ −1 )val(z) f (1/z)µ2 (z). Assuming j ≤ val(α), we
R R R
Qp
then see by an easy computation (left to the reader) that [(16) for j = val(α) + (16) for
j < val(α) and a ∈ Q× p + (16) for j < val(α), a = 0 and n ≥ 0 + (17) for j = val(α)
and a = 0 + (17) for j < val(α), a = 0 and n ≤ 0] is equivalent to µ1 and µ2 satisfying
condition (5) of Lem.6.2.6 with r = val(α) and d = k − 2. Hence µ1 and µ2 are in fact
tempered distributions on Zp of order val(α) by Cor.6.2.7. Thus we can integrate any
function in B(α) against µ, in particular the functions of L(α). By making n tends to
−∞, we see that (16) for j < val(α) and a = 0 implies µ cancels the functions z j ∈ L(α).
By making n tends to +∞, we see that (17) for j < val(α) implies µ cancels the functions
αp val(z−a)


β
(z −a)k−2−j ∈ L(α). Now, the reader can check that µ1 and µ2 being tempered
of order val(α) together with these cancellations is in fact equivalent to (16) for j ≤ val(α)
and a ∈ Qp + (17) for j < val(α) and a ∈ Qp + (17) for j = val(α) and a = 0. This gives
a GL2 (Qp )-equivariant isomorphism between the dual of the completion of π(α) in (15)
and (B(α)/L(α))∗ . A closer examination shows that this is a topological isomorphism
of Banach spaces and thus (B(α)/L(α))∗ is a unitary GL2 (Qp )-Banach space. Using
the closed embedding B(α)/L(α) ,→ ((B(α)/L(α))∗ )∗ , we see that the GL2 (Qp )-Banach
B(α)/L(α) is unitary since it is closed in a unitary GL2 (Qp )-Banach. This means that the
canonical map π(α) → B(α)/L(α) induces a continuous GL2 (Qp )-equivariant morphism
from the completion of π(α) (with respect to any finite type lattice) to B(α)/L(α). This
morphism is, as we have seen, an isomorphism on the duals for the strong topologies,
hence also for the weak topologies (using the theory of [33]). By biduality, we finally
get 1) for B(α)/L(α) and also for B(β)/L(β) by interchanging α and β. The proof
of 2) is similar (and actually simpler since we only have to deal with j < val(α)), the
only difference being that here, the conditions (16) and (17) only allow you to integrate
functions in N (α) against µ (see the end of §6.2).
42 LAURENT BERGER & CHRISTOPHE BREUIL

8. Representations of GL2 (Qp ) and (ϕ, Γ)-modules I (C.B.)

We now focus on the study of Π(V ) when V is irreducible. Quite surprisingly, this will
crucially use the (ϕ, Γ)-module D(V ) associated to V . We define here a topological space

b
lim
←−ψ D(V ) and give a first explicit description of it using the Wach module N (T )
associated to a Galois lattice in V . We then give preliminary results on distributions on

b
Qp naturally arising from ←− lim D(V ) via Amice transform.
ψ

8.1. Back to (ϕ, Γ)-modules. — Let V be as in case 1) of §7.1, Π(V ) as in §7.2 and
D(V ) be the (ϕ, Γ)-module associated to V in §3.1. Recall that we have defined in §3.3 a
map ψ : D(V )  D(V ) which is a left-inverse of ϕ (and hence a surjection). Define the
following E-vector space:

b
lim
←− D(V ) := {(vn )n∈Z≥0 | vn ∈ D(V ), ψ(vn+1 ) = vn , (vn )n a bounded sequence in D(V )},
ψ

(recall that bounded means bounded for the weak topology which was defined in §5.2).
We equip the above space with the following structures:
1) a left OE [[X]]-module structure, by s((vn )n ) := (ϕn (s)vn )n , if s ∈ OE [[X]]
2) a bijection ψ given by ψ((vn )n ) := (ψ(vn ))n
3) an action of Γ given γ((vn )n ) := (γvn )n .

The aim of this lecture and the next one is to prove a “canonical” topological isomor-
phism:

Theorem 8.1.1. — Assume that V is absolutely irreducible. There is an isomorphism

of topological E-vector spaces (for the weak topologies on both sides):

b
lim D(V ) ' Π(V )∗
←−
ψ
 
1 0
such that the action of on Π(V )∗ corresponds to (vn )n 7→ (ψ(vn ))n , the action of
0 p
   
1 0 × 1 Zp
to that of Γ ' Zp and the action of to the multiplication by (1+X)Zp .
0 Z×p 0 1

Such an isomorphism was inspired by an analogous isomorphism due to Colmez in the

case V is semi-stable non crystalline ([16]).
Before proving this theorem, we will need to establish a number of preliminary results.
Let T be a GQp -stable lattice in V . First of all, because V is assumed to be crystalline,
we can write D(T ) = AQp ⊗Zp [[X]] N (T ) where N (T ) is the Wach module which we have
constructed in §5. This is a free OE [[X]]-module of finite rank and thus it is equipped with
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 43

the (p, X)-adic topology. The first result which we will need in order to prove theorem
8.1.1 is the following one:

Proposition 8.1.2. — Let T be a GQp -stable lattice in a crystalline, irreducible repre-

sentation V with Hodge-Tate weights 0 and r > 0 and let ←− lim N (T ) denote the set of
ψ
ψ-compatible sequences of elements of N (T ). Then the natural map:


b
lim
←− N (T ) ⊗ O E
E → lim
←− D(V )
ψ ψ

is a topological isomorphism.

Proof. — The main content of this proposition is the explicit description of the topology
of D(V ) in terms of N (T ) which we have given in §5 and its interaction with ψ. Recall that
for each k ≥ 0, we define a semi-valuation νk on D(T ) as follows: if x ∈ D(T ) then νk (x)
is the largest integer j ∈ Z ∪ {+∞} such that x ∈ X j N (T ) + pk D(T ). The weak topology
on D(T ) is then the topology defined by the set {νk }k≥0 of all those semi-valuations.
The weak topology on D(V ) is the inductive limit topology on D(V ) = ∪`≥0 D(p−` T ).
Concretely, if we have a sequence (vn )n of elements of D(V ), and that sequence is bounded
for the weak topology, then there is a GQp -stable lattice T of V such that vn ∈ D(T )
for every n ≥ 0 and furthermore, for every k ≥ 0, there exists f (k) ∈ Z such that
vn ∈ X −f (k) N (T ) + pk D(T ). Recall that we have proved (cf. Th.5.3.1) that if V is
crystalline with positive Hodge-Tate weights, and ` ≥ 1, then:

ψ(X −` N (T )) ⊂ p`−1 X −` N (T ) + X −(`−1) N (T )

by iterating this m times we get:

ψ m (X −` N (T )) ⊂ pm(`−1) X −` N (T ) + X −(`−1) N (T ).

Choose k ≥ 0 and ` ≥ 2 such that vn ∈ X −` N (T ) + pk D(T ). Since vn = ψ m (vn+m ) for

every m ≥ 1, we see that vn ∈ pm(`−1) X −` N (T ) + X −(`−1) N (T ) + pk D(T ) for all m and
by taking m large enough we get that vn ∈ X −(`−1) N (T ) + pk D(T ) so that if ` ≥ 2 then
we may replace it by ` − 1. Hence, for every k ≥ 0, we have vn ∈ X −1 N (T ) + pk D(T ).
This being true for every k, we get vn ∈ X −1 N (T ). We will now prove that vn ∈ N (T )
if V is irreducible (it is in fact enough to assume that the ϕ-module Dcris (V ) has no
part of slope 0). Indeed, both X −1 N (V ) and N (V ) are stable by ψ and we have seen
in proposition 5.5.2 that the map ψ : π −1 N (V )/N (V ) → π −1 N (V )/N (V ) coincides,
via the identification π −1 N (V )/N (V ) = Dcris (V ) given by π −1 x 7→ x, with the map
ϕ−1 : Dcris (V ) → Dcris (V ). If Dcris (V ) has no part of slope 0, then ϕ−m (x) → 0 as
m → ∞ and so for every ` there exists m such that ψ m (X −1 N (T )) ⊂ p` X −1 N (T )+N (T ).
44 LAURENT BERGER & CHRISTOPHE BREUIL

Consequently, we have vn ∈ N (T ) for all n ≥ 0. To finish the proof of the proposition,

note that ←−
lim N (T ) is compact and that the map:
ψ

b
lim
←− N (T ) → lim
←− D(T )
ψ ψ

is a continuous isomorphism, so that it is a topological isomorphism. The proposition

follows by inverting p.

Next, recall that in Prop.5.5.3 we have seen that if V is a crystalline irreducible rep-
resentation whose Hodge-Tate weights are ≥ 0, then we have an injective map N (T ) ,→
R+Qp ⊗Qp Dcris (V ). Now let V = Vcris (D(α, β)), V irreducible. Given a sequence (vn )n ∈
lim N (T ) we can write vn = wα,n ⊗eα +wβ,n ⊗eβ . Recall that we write R+
←−
+
E for E ⊗Qp RQp .
ψ

Proposition 8.1.3. — Given (vn )n ∈ (lim ←−ψ N (T )) ⊗OE E, the sequences (wα,n )n and
+
(wβ,n )n of elements of RE defined above satisfy the following properties:
1) ∀ n ≥ 0, wα,n (resp. wβ,n ) is of order val(α) (resp. val(β)) and kwα,n kval(α) (resp.
kwβ,n kval(β) ) is bounded independently of n
2) ∀ n ≥ 0 and ∀ m ≥ 1, we have
ϕ−m (wα,n ⊗ eα + wβ,n ⊗ eβ ) ∈ Fil0 ((Qp (µpm ) ⊗Qp E)[[t]] ⊗E D(α, β))
3) ∀ n ≥ 1, ψ(wα,n ) = α−1 wα,n−1 and ψ(wβ,n ) = β −1 wβ,n−1 .

Proof. — We can always change T in order to have (vn )n ∈ ←−

lim N (T ) which we now
ψ
+
assume. Let cα : N (T ) → RE be the map which to x = xα ⊗ eα + xβ ⊗ eβ assigns xα .
To prove that cα (N (T )) is of order ≤ val(α), we use the following characterization of
elements of order r in R+ . Recall that R+ is included in a larger ring R
E E
e + := ∩n≥0 ϕn (B+ )
E cris
on which ϕ is a bijection and which has a topology defined by the family of semi-norms
k·kD where D runs over all closed disks of radii < 1. Fix one such closed disk D = D(0, ρ)
of radius ρ < 1. We say that f ∈ R e + is of order r if the sequence {p−mr kϕ−m f kD } is
E
bounded independently of m. We can then define kf kD,r := supm (p−mr kϕ−m f kD ). The
norms k · kD,r (which depend on the choice of D) are equivalent to the norms k · kr defined
in §6.1. Indeed, if Dr has radius r, then kϕ−m f kDr = kf kDr1/pm and one recovers the
definition of order as “order of growth”.
Now if x ∈ N (T ) and x = xα ⊗ eα + xβ ⊗ eβ then:
ϕ−m (x) = ϕ−m (xα )α−m ⊗ eα + ϕ−m (xβ )β −m ⊗ eβ
and the set of ϕ−m (x) is bounded for any k · kD norm (because x ∈ A+ [1/X] ⊗Zp T )
so that the sets {ϕ−m (xα )α−m } and {ϕ−m (xβ )β −m } are both bounded. It thus follows
that xα is of order val(α) and xβ of order val(β). We have proved that the image of cα
is made up of elements of order val(α) and it is bounded in the k · kval(α) norm because
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 45

N (T ) is an OE [[X]]-module of finite type. The same holds with β in place of α of course.

For the second point, we use the fact that vn ∈ N (T ) so that for all m ≥ 1 we have
+
ϕ−m (vn ) ∈ BdR ⊗Qp V . On the other hand by the results of §4.1 (see Prop.4.1.2) we have
ϕ−m (vn ) ∈ (Qp (µpm ) ⊗Qp E)((t)) ⊗E D(α, β) and therefore:

ϕ−m (vn ) ∈ (BdR

+


⊗Qp V ) ∩ (Qp (µpm ) ⊗Qp E)((t)) ⊗E D(α, β)
= Fil0 (Qp (µpm ) ⊗Qp E)((t)) ⊗E D(α, β) .

Since the weights of the filtration on D(α, β) are negative, we can replace (Qp (µpm ) ⊗Qp
E)((t)) by (Qp (µpm ) ⊗Qp E)[[t]] in the Fil0 . Finally, the third point is obvious since
vn−1 = ψ(vn ) and ψ acts as ϕ−1 on Dcris (V ).

Proposition 8.1.4. — Conversely, if we are given two sequences (wα,n )n and (wβ,n )n of
elements of R+ E satisfying properties 1)-3) of Prop.8.1.3 and if we set vn := wα,n ⊗ eα +
wβ,n ⊗ eβ , then (vn )n ∈ (lim
←− N (T )) ⊗OE E.
ψ

Proof. — It is enough to show that there exists a lattice T such that vn ∈ N (T ) because
condition 3) of Prop.8.1.3 ensures that the sequence (vn )n is ψ-compatible. Choose any
lattice U . By the results of §4 and §5 (more specifically proposition 5.5.3), we know that
vn ∈ R+ E [1/t] ⊗OE [[X]] N (U ).
Condition 2) of proposition 8.1.3 then implies that vn ∈ R+
E [1/X]⊗OE [[X]] N (U ), because
the filtration condition for m ≥ 1 is equivalent to an order-of-vanishing condition at
(m) − 1, and implies that vn ∈ R+ E [ϕ
m−1
(q)/t] ⊗OE [[X]] N (U ). By using this for each
m ≥ 1, we get that vn ∈ R+E [1/X] ⊗OE [[X]] N (U ).
The ψ-compatibility allows us to get rid of the denominators in X. Finally, condition
1) tells us that there exists ` independent of n such that vn ∈ p−` N (U ) so we can take
T = p−` U .

It is important to notice that the map ψ : N (T ) → N (T ) is not surjective. Recall that

we have in particular proved in Prop.5.4.1 the following result (we actually proved more,
and we proved it for Qp -linear representations, but as usual the E-linear result follows
immediately):

Proposition 8.1.5. — If V is crystalline absolutely irreducible, then there exists a unique

non-zero E-vector subspace D0 (V ) of D(V ) possessing an OE -lattice D0 (T ) which is a
compact OE [[X]]-submodule of D(V ) preserved by ψ and Γ with ψ surjective.

In fact, as we saw, the lattice D0 (T ) is of finite type over OE [[X]]. The following
corollary will be used in §9.3 to prove the irreducibility of Π(V ) (for V irreducible):
46 LAURENT BERGER & CHRISTOPHE BREUIL

Corollary 8.1.6. — Let M ⊆ ←−

lim N (T ) be a closed non zero OE [[X]]-submodule which
ψ
is stable by ψ, ψ −1 and Γ, then:

b
M ⊗OE E = (lim
←− N (T )) ⊗O E
E = lim
←− D(V ) .
ψ ψ

lim N (T ) is compact, so is M (being closed in a compact

Proof. — Note first that since ←−
ψ
set). For m ∈ Z≥0 , denote by prm : ←−
lim N (T ) → N (T ), (vn )n →
7 vm and define
ψ
Mm := prm (M). As M 6= 0, there exists m such that Mm 6= 0. Using that M is
stable by ψ and ψ −1 , it is straightforward to check that Mm doesn’t depend on m and
we denote it by M . Via M = M0 , it is an OE [[X]]-submodule of N (T ) (necessarily of
finite type since N (T ) is) stable by Γ and ψ and such that ψ is surjective. Moreover,
the canonical map M → ←− lim M is easily checked to be an isomorphism since M is
ψ
lim M and compact (the density follows from M = prm (M) ∀ m). Doing
both dense in ←−
ψ
lim N (T ), we find another (finite type) OE [[X]]-submodule M 0 of N (T )
the same with ←−
ψ
containing M stable by Γ and ψ with ψ surjective and such that ←− lim M 0 .
lim N (T ) = ←−
ψ ψ
By Prop.8.1.5, we have thus M ⊗OE E = M 0 ⊗OE E and there exists a ∈ E × such that
M ⊂ M 0 ⊂ aM . Hence, we have (lim
←−ψ M ) ⊗OE E ' (lim
0
←−ψ M ) ⊗OE E which finishes the
proof.

8.2. Back to distributions. — Recall from §6.2 that µ 7→ +∞ z

P R
n
n=0 Zp n µ(z)X gives
a bijection between R+
E and “E-valued” locally analytic distributions on Zp (some people
write “µ 7→ Zp (1 + X)z µ(z)”) which induces a bijection between elements of R+
R
E of order
r and tempered distributions of order r. We will need the following easy lemma (we leave
its proof to the reader):

Lemma 8.2.1. — If w ∈ R+ +
E corresponds to µ, then ψ(w) ∈ RE corresponds to the
unique locally analytic distribution ψ(µ) on Zp such that:
Z Z
f (z)ψ(µ)(z) := f (z/p)µ(z)
Zp pZp

for any locally analytic function f on Zp .

We have an analogous lemma for tempered distributions of order r and functions of

class C r (as one easily checks that ψ preserves elements in R+
E of order r). In particular, if
(µn )n∈Z≥0 is a sequence of tempered distributions on Zp of order r such that ψ(µn+1 ) = µn ,
we can define a linear form µ on the E-vector space LPolc,k−2 of locally polynomial
functions f : Qp → E with compact support of degree less than k − 2 as follows:
Z Z Z
f (z)µ(z) = f (z)µ(z) := f (z/pN )µN (z)
Qp p−N Zp Zp
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 47

where N ∈ Z≥0 is such that supp(f ) ⊂ p−N Zp . One readily checks using Lem.8.2.1 that
it doesn’t depend on such an N .
Let (vn )n = (wα,n ⊗eα +wβ,n ⊗eβ )n ∈ (lim
←−ψ N (T ))⊗OE E. Recall this means (Prop.8.1.3
and Prop.8.1.4):
1) ∀ n ≥ 0, wα,n (resp. wβ,n ) is of order val(α) (resp. val(β)) and ||wα,n ||val(α) (resp.
||wβ,n ||val(β) ) is bounded independently of n
2) ∀ n ≥ 0, ∀ m ≥ 1, ϕ−m (wα,n ⊗ eα + wβ,n ⊗ eβ ) ∈ Fil0 ((Qp (µpm ) ⊗Qp E)[[t]] ⊗E D(α, β))
3) ∀ n ≥ 1, ψ(wα,n ) = α−1 wα,n−1 , ψ(wβ,n ) = β −1 wβ,n−1 .

Lemma 8.2.2. — Let wα , wβ ∈ R+

E , m ∈ N and µα , µβ the locally analytic distributions
on Zp corresponding to wα , wβ . The condition:

ϕ−m (wα ⊗ eα + wβ ⊗ eβ ) ∈ Fil0 ((Qp (µpm ) ⊗Qp E)[[t]] ⊗E D(α, β))

is equivalent to the equalities:

Z Z
m
α z j ζpzm µα (z) =β m
z j ζpzm µβ (z)
Zp Zp

for all j ∈ {0, · · · , k − 2} and all primitive pm -roots ζpm of 1.

Proof. — Here ζpzm is considered as a locally constant function Zp → E (enlarging E is

necessary). Fixing ζpm a primitive root of 1, we have:
m m
ϕ−m (X) = (ζpm ⊗ 1)et/p − 1 = (ζpm ⊗ 1)(et/p − 1) + ζpm ⊗ 1 − 1
P+∞ i
in (Qp (µpm ) ⊗Qp E)[[t]] (see previous lectures). Writing wα = i=0 αi X and wβ =
P+∞ i 0
i=0 βi X , the condition on Fil is equivalent to:
+∞
X +∞
X
−m
α m
αi ϕ i
(X) eα + β m
βi ϕ−m (X)i eβ ∈
i=0 i=0
t/pm k−1
(Qp (µpm ) ⊗Qp E)[[t]](eα + eβ ) ⊕ (e − 1) ((Qp (µpm ) ⊗Qp E)[[t]]eα )
m
noting that et/p − 1 generates gr1 (Qp [[t]]) = Qp t. Assuming E contains Qp (µpm ), we
have:
Y
Qp (µpm ) ⊗Qp E = E
Qp (µpm ),→E
m
and writing the binomial expansion of ϕ−m (X)i = (ζpm (et/p − 1) + ζpm − 1)i , a simple
calculation gives that the above condition is equivalent to the equalities:
+∞   +∞  
m
X i i−j m
X i
(19) α αi (ζpm − 1) = β βi (ζpm − 1)i−j
i=j
j i=j
j
48 LAURENT BERGER & CHRISTOPHE BREUIL

z
R

for j ∈ {0, · · · , k − 2} and all primitive pm -roots ζpm of 1. Using αi = Zp i
µα (z) and
the elementary Mahler expansion:
  +∞    
z z−j X i i−j z
ζm = (ζpm − 1)
j p i=j
j i

we obtain:
+∞   Z X +∞   
X i i−j z i i−j

αi (ζp − 1)
m = (ζp − 1)
m µα (z)
i=j
j Zp i=j
i j
Z  
−j z z
= ζpm ζ m µα (z)
Zp j p
and likewise with βi and µβ . Using (19), we easily get the result.

For n ∈ Z≥0 , let µα,n be the tempered distribution on Zp of order val(α) associated
to αn wα,n . We have ψ(µα,n ) = µα,n−1 for n ∈ N and can thus define µα ∈ LPol∗c,k−2
(algebraic dual of LPolc,k−2 ) as before:
Z Z
(20) f (z)µα (z) := f (z/pN )µα,N (z).
p−N Zp Zp

Likewise with β instead of α. By Lem.8.2.2, we have that condition 2) above on (wα,n , wβ,n )n
is equivalent to:
Z Z
m−n
(21) α z j ζpzm µα,n (z) =β m−n
z j ζpzm µβ,n (z)
Zp Zp

for j ∈ {0, · · · , k − 2}, n ≥ 0 and m ≥ 1.

Fixing embeddings Q ,→ C, Q ,→ Qp and using the identification Qp /Zp = Z[1/p]/Z,
we can “see” e2iπz in Qp for any z ∈ Qp .

Corollary 8.2.3. — Condition 2) above is equivalent to the equalities in Qp :

Z  α val(y) Z
j 2iπzy
z e µα (z) = z j e2iπzy µβ (z)
p −N Zp β p −N Zp

for j ∈ {0, · · · , k − 2}, y ∈ Q×

p and N > val(y).

N
Proof. — This follows from (20) and (21) noting that the algebraic number e2iπy/p is a
primitive pN −val(y) -root of 1.

9. Representations of GL2 (Qp ) and (ϕ, Γ)-modules II (C.B.)

We still assume V absolutely irreducible as in §7.1. The aim of this section is to

b ∗
←− D(V )) ' Π(V ) , more precisely a topological
construct a topological isomorphism (lim
ψ
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 49

isomorphism:
∗
←− N (T )) ⊗OE E ' (B(β)/L(β)) .
(lim
ψ

We then deduce from this isomorphism that Π(V ) is non zero, topologically irreducible
and admissible. These three statements were conjectured in [9] and [10] (and already
known in some cases by a “reduction modulo p” method, see [9] and Lecture 10). We
1
also deduce a link with the Iwasawa cohomology HIw (Qp , V ) of V .

9.1. The map Π(V )∗ → (limψ D(V ))b . — We first construct a map from (B(β)/L(β))∗
←−ψ N (T )) ⊗OE E. Note that any f in π(α) or π(β) which has compact
to the space (lim
support as a function on Qp via (9) is an element in LPolc,k−2 . Recall that I : π(α) ' π(β)
is the intertwining operator of (12).

Lemma 9.1.1. — Let µα , µβ ∈ LPol∗c,k−2 . The following statements are equivalent:

1) For j ∈ {0, · · · , k − 2}, y ∈ Q× p and N > val(y), we have:
Z  α val(y) Z
j 2iπzy
(22) z e µα (z) = z j e2iπzy µβ (z).
p−N Zp β p−N Zp

2) For any f ∈ π(α) with compact support (as a function on Qp via (9)) such that
I(f ) ∈ π(β) has compact support (as a function on Qp via (9)), we have:
1 − αβ
Z Z
(23) f (z)µα (z) = α I(f )(z)µβ (z).
Qp 1 − pβ Qp

Proof. — Note that α 6= β and α 6= pβ (see §7.1) hence the constant in (23) is well
GL (Q )
defined and non zero. Recall the intertwining I sm : IndB(Q2p ) p unr(α−1 ) ⊗ unr(pβ −1 ) '
GL (Q )
IndB(Q2p ) p unr(β −1 ) ⊗ unr(pα−1 ) is given by:
 Z    
sm 0 −1
(24) I (h) = g 7→ h g dx
Qp
1 x
where dx is Haar measure on Qp (with values in Q ⊂ E). In terms of locally constant
functions on Qp via (9), we thus get:
 Z
(25) sm
I : f 7→ z 7→ (αpβ −1 )val(x) f (z + x−1 )dx =
Qp
Z 
−1 val(x)
(βpα ) f (z + x)dx .
Qp

This is in fact an algebraic intertwining. This means that we shouldn’t bother too much
about convergence problems in Qp (βpα−1 )val(x) f (z + x)dx since, in fine, we can always
R

replace the infinite sums around 0 or −∞ by well defined algebraic expressions in βpα−1 .
Let j ∈ {0, · · · , k − 2} and fj : Qp → E be a locally constant function with compact
support such that I sm (fj ) also has compact support. Let fˆj be the usual Fourier transform
50 LAURENT BERGER & CHRISTOPHE BREUIL

of fj which is again a locally constant function on Qp with compact support. Recall

fˆj (x) = Qp fj (z)e−2iπzx dz and fj (z) = Qp fˆj (x)e2iπzx dx (see the end of §8.2 for e−2iπzx )
R R

where dx, dz is Haar measure on Qp . The fact that I sm (fj ) has compact support is
easily checked with (25) to be equivalent to fˆj (0) = 0 (for |z|  0, we have I sm (fj )(z) =
(βpα−1 )val(z) Qp fj (x)dx = (βpα−1 )val(z) fˆj (0)). Let N ∈ N be such that fj and I sm (fj )
R

have support in p−N Zp and such that fˆj |pN Zp = 0. For z ∈ p−N Zp , we have:
Z
j
I(z fj )(z) = z j
(βpα−1 )val(x) fj (z + x)dx
p −N Zp
Z Z 
= z j
(βpα )−1 val(x)
fˆj (y)e2iπy(z+x)
dy dx
p−N Zp Qp −pN Zp
Z Z 
= z j
fˆj (y)e 2iπzy
(βpα ) −1 val(x)
e2iπxy
dx dy.
Qp −pN Zp p−N Zp

Decomposing p−N Zp = p−N Z× p qp

−N +1 ×
Zp q · · · , a straightforward computation yields
for N > val(y):
+∞ α  
Z
−1 val(x) 2iπxy
X
−1 i
Z
2iπxy
1 − pβ α val(y)
(βpα ) e dx = (βpα ) e dx =
p−N Zp i=−N p i Z×
p 1 − αβ β
hence:
α
1− pβ
Z  α val(y)
(26) j
I(z fj )(z) = fˆj (y)z j e2iπzy dy
1 − αβ Qp −pN Zp β
for z ∈ p−N Zp and I(z j fj )(z) = 0 otherwise. Likewise, we have:
Z
(27) j
z fj (z) = fˆj (y)z j e2iπzy dy
Qp −pN Zp

for z ∈ p−N Zp and z j fj (z) = 0 otherwise. Note that (26) and (27) are in fact finite sums
over the same finite set of values of y (as fˆj is locally constant with compact support).
Replacing I(z j fj )(z) and z j fj (z) in (28) below by the finite sums (26) and (27), we then
easily deduce that the following equalities for various fj as above with support in p−N Zp :
α Z
Z
j
1 − pβ
(28) I(z fj (z))µβ (z) = β
z j fj (z)µα (z)
p−N Zp 1 − α p−N Zp
are equivalent to the equalities in 1) (final details here are left to the reader). This finishes
the proof.

Let µβ ∈ (B(β)/L(β))∗ ⊂ π(β)∗ . Define a sequence of locally analytic distributions on

Zp , (µβ,N )N ∈Z≥0 , as follows:
Z Z
f (z)µβ,N (z) := f (pN z)µβ (z).
Zp p−N Zp

One obviously has ψ(µβ,N ) = µβ,N −1 by Lem.8.2.1.

 TOWARDS A p-ADIC LANGLANDS PROGRAMME 51

Lemma 9.1.2. — The distributions µβ,N are tempered of order val(β) and ||β −N µβ,N ||0val(β)
is bounded independently of N (see §6.2 for the definition of this norm).

Proof. — The fact that µβ,N is tempered of order val(β) is easy and left to the reader.
Going back to the proof of Th.7.2.2, we know that µβ satisfies in particular (16) and (17).
Hence, for a ∈ Zp , 0 ≤ j ≤ k − 2 and n ∈ Z≥0 :
Z Z
−N −N N j
β j
(z − a) µβ,N (z) = β p (z − p−N a)j µβ (z)
a+pn Zp p−N a+pn−N Zp

∈ Cµβ p−N val(β) pN j p(n−N )(j−val(β)) OE

∈ Cµβ pn(j−val(β)) OE .

For any locally analytic distribution µ on Zp and any r ∈ R+ such that r < k − 1, define:
Z
||µ||0r,k := supa∈Zp supj∈{0,··· ,k−2} supn∈Z≥0 p n(j−r)
(z − a)j µ(z) .
a+pn Zp

An examination of the proof of Lem.6.2.6 shows that the norms || · ||0r and || · ||0r,k are in
fact equivalent (see also [15],§V.3.6). This implies:

||β −N µβ,N ||0val(β) ≤ c||β −N µβ,N ||0val(β),k ≤ c|Cµβ |

for a convenient constant c ∈ R+ .

By Cor.7.2.3, the intertwining I : π(α) ' π(β) extends “by continuity” to an inter-
twining Iˆ : B(α)/L(α) ' B(β)/L(β) and we define:

1 − αβ ˆ ∗
µα := α µβ ◦ I ∈ (B(α)/L(α)) .
1 − pβ

By restriction to LPolc,k−2 , µα and µβ define elements of LPol∗c,k−2 satisfying (23) by

definition, hence (22) by Lem.9.1.1. As in Lem.9.1.2, µα gives rise to a bounded sequence
of tempered distributions µα,N on Zp . From Cor.8.2.3 and Lem.9.1.2, we get that Amice
transform produces from (µα,N )N and (µβ,N )N a sequence (vN )N = (wα,N ⊗eα +wβ,N ⊗eβ )N
satisfying conditions 1), 2), 3) before Lem.8.2.2, hence (vN )N ∈ (lim
←− N (T )) ⊗OE E. This ψ
defines a linear map (B(β)/L(β))∗ → (lim
←−ψ N (T )) ⊗OE E. The continuity follows from
the fact that the function “Amice transform” from the space of tempered distributions
of order r on Zp with norm smaller than 1 endowed with the weak topology of pointwise
convergence to the subspace of R+
E of elements w of order r such that ||w||r ≤ 1 endowed
with its compact topology induced by that of R+
E is a topological isomorphism. Final
details here are left to the reader.
52 LAURENT BERGER & CHRISTOPHE BREUIL

9.2. The map (limψ D(V ))b → Π(V )∗ . — We now construct a map from (lim
←− N (T ))⊗OE ψ
E to (B(β)/L(β))∗ . Let (vn )n = (wα,n ⊗ eα + wβ,n ⊗ eβ )n ∈ (lim
←−ψ N (T )) ⊗OE E and define
∗
µα and µβ as in §8.2. They are elements of Polc,k−2 satisfying (22) by Cor.8.2.3, hence
(23) by Lem.9.1.1.

Lemma 9.2.1. — There is a unique way to extend µβ and µα as elements of respectively

B(β)∗ and B(α)∗ such that for any f ∈ π(α) (seen as a function on Qp via (9)):
1 − αβ
Z Z
(29) f (z)µα (z) = α I(f )(z)µβ (z).
Qp 1 − pβ Qp

Proof. — An easy computation using (25) gives:

1
j
1− p
pβval(z)
(30) I(z 1Zp ) = β
z j 1Z p + z j (1 − 1Zp )
1− α
α
pαval(z) 1 − p1 j pβval(z)
I zj (1 − 1pZp ) = z j 1pZp


(31) + z (1 − 1pZp ).
β 1 − αβ α
R R
Since Qp z j 1Zp µα (z) and Qp z j 1Zp µβ (z) are well defined, we see using (29) and (30) that
Z
pβ
val(z)
zj (1 − 1Zp )µβ (z)
Qp α

val(z)
is uniquely determined. Then, using (31) (and (29)), we see that Qp z j pα
R
β
(1 −
1pZp )µα (z) is also uniquely determined. One readily checks that this defines unique
extensions of µβ and µα as linear forms on respectively π(β) and π(α). To check that
µβ is in fact in B(β)∗ , it is enough by Cor.6.2.7 to check that the relevant Amice-Vélu
condition (as in Lem.6.2.6) is satisfied by µβ on each copy of Zp via (6). We already know
it for the first one since wβ,0 is of order val(β). For the second one, a straightforward
computation via the definition of f2 in (6) (exchanging α and β) shows that we have to
find a constant Cµβ ∈ E such that:
Z pβval(z)
(32) z k−2−j (1 − az)j µβ (z) ∈ Cµβ pn(j−val(β)) OE
a−1 +pn−2val(a) Zp α
for a ∈ Zp − {0}, j ∈ {0, · · · , k − 2} and n > val(a), and:
Z pβval(z)
(33) z k−2−j µβ (z) ∈ Cµβ pn(j−val(β)) OE
−n
Qp −p Zp α
for j ∈ {0, · · · , k − 2} and n ∈ Z≥0 . From condition 1) before Lem.8.2.2, we know there
exists Cµβ ∈ E such that β −N a+pn Zp (z − a)j µβ,N (z) ∈ Cµβ pn(j−val(β)) OE for a ∈ Zp and
R

N, n ∈ Z≥0 . For a ∈ Q× p , let ua := p

val(a) −1
a ∈ Z× p . By (20), the left hand side of (32) is
equal to (up to a unit in OE ):
pβ−val(a) Z
(j−k+2)val(a)
p z k−2−j (z − ua )j µβ,val(a) (z)
α ua +p n−val(a) Zp
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 53

and we see, writing z k−2−j = (z − ua + ua )k−2−j = · · · and using the above bound with
N = val(a), that it belongs to:

Cµβ p(val(α)−val(β)−1)val(a) p(j−k+2)val(a) pval(β)val(a) p(n−val(a))(j−val(β)) OE

which is exactly Cµβ pn(j−val(β)) OE using val(α) + val(β) = k − 1. This proves (32). To
prove (33), one uses the equality (similar to (30)):
 pβval(z)
z k−2−j (1 − 1p−n Zp ) = p−n I(z k−2−j 1p−n Zp ) −
α
1 − p1  α n k−2−j
z 1p−n Zp
1 − αβ pβ
together with (29), the equalities:
Z Z
−n k−2−j (j−k+1)n
p z µα (z) = p z k−2−j µα,n (z)
p−n Zp Zp
 α n Z  α n Z
k−2−j (j−k+2)n
z µβ (z) = p z k−2−j µβ,n (z)
pβ −n
p Zp pβ Zp

and the fact α−n Zp z k−2−j µα,n (z) and β −n Zp z k−2−j µβ,n (z) are bounded independently
R R

of n. The proof that µα extends as a distribution on B(α) is similar.

∗
←−ψ N (T ))⊗OE E ,→ B(β) , (wα,n ⊗
By Lem.9.2.1, we have a continuous injective map (lim
eα + wβ,n ⊗ eβ )n 7→ µβ (the continuity follows as in §9.1 from the continuity of Amice
transform and the injectivity is straightforward using Lem.9.2.1). With the action of
GL2 (Qp ) deduced from (7) on the dual B(β)∗ , it is an easy and pleasant exercise on
R
Amice transform that we leave to the reader (using e.g. the formula Zp (1 + X)z µ(z)) to
 
1 Zp
check that the induced action of on (lim
←−ψ N (T ))⊗OE E is indeed as in Th.8.1.1.
0 Q× p
In particular, it preserves ←−
lim N (T ).
ψ

Lemma 9.2.2. — The distribution µβ ∈ B(β)∗ defined in Lem.9.2.1 sits in (B(β)/L(β))∗ .

Proof. — The non zero compact OE -submodule of B(β)∗ which is the image of ←− lim N (T )
    ψ
1 Zp 1 Qp
is preserved by the action of B(Qp ) since generates and the central
0 Q× p 0 Q× p
character is integral. Let:
( Z )
π(β)0 := f ∈ π(β), f (z)µ(z) ∈ OE ∀ µ ∈ ←−lim N (T ) .
Qp ψ

We have that π(β)0 ⊂ π(β) is thus preserved by B(Qp ). By [30],Lem.13.1(iii), we also

have that π(β)0 ⊗OE E = π(β) (using that B(β)∗ ,→ π(β)∗ and thus that ←− lim N (T ) is
ψ
∗
bounded inside π(β) ). By the same proof as for Th.7.2.2 (using that any µ ∈ ←−
lim N (T )
ψ
54 LAURENT BERGER & CHRISTOPHE BREUIL

is such that ∀ g ∈ B(Qp ), ∀ f ∈ π(β)0 , |µ(g(f ))| ≤ 1), we deduce that any µ ∈ B(β)∗
which is in the image of ←−
lim N (T ) cancels L(β). Whence the result.
ψ

∗
←−ψ N (T )) ⊗OE E ,→ B(β) takes val-
By Lem.9.2.2, the above continuous injection (lim
ues in the subspace (B(β)/L(β))∗ . It is easily checked that this map is an inverse to
the previous one (§9.1), thus finishing the proof of Th.8.1.1. In particular, we have an
isomorphism:
0 1
@
1 0A
D(V )ψ=1 = (lim b

ψ=1
' Π(V )∗ 0 pZ .
←− D(V ))
ψ

Using Fontaine’s canonical isomorphism (see e.g. [15],§4.3):

1
lim H 1 (Gal(Qp /Qp (µpn )), T ) ⊗OE E ' D(V )ψ=1


(34) HIw (Qp , V ) := ←−
n

for any Galois lattice T in V , we deduce the quite surprising corollary:

Corollary 9.2.3. — We have an isomorphism of OE [[Z×

p ]]-modules:
0 1
@
1 0A
1
HIw (Qp , V ) ' Π(V )∗ 0 pZ

where the OE [[Z×

p ]]-module structure on the left hand side is coming from the action of Γ
 
1 0
and on the right hand side from the action of .
0 Z×p

When V is reducible (as in 2) and 3) of §7.1), it is not true that we have an isomorphism
b ∗ k−1
←−ψ D(V )) ' Π(V ) (although it is “almost true”). But if α 6= p
(lim and β 6= 1 (which
always holds if V is coming from a modular form by the Weil conjectures), one0
can 1prove
@
1 0A
that there is still an isomorphism of O [[Z× ]]-modules H 1 (Q , V ) ' Π(V )∗ 0 p .
Z
E p Iw p

9.3. Irreducibility and admissibility. — We deduce from Th.8.1.1 that, for V irre-
ducible as before, Π(V ) is non zero, topologically irreducible and admissible.

Corollary 9.3.1. — The Banach B(α)/L(α) and B(β)/L(β) are non zero.
b ψ=1
←−ψ D(V )) 6= 0 which follows from D(V )
Proof. — This follows from (lim 6= 0 which
follows e.g. from (34).

Corollary 9.3.2. — The Banach B(α)/L(α) and B(β)/L(β) are topologically irreducible.

Proof. — Assume there exists a (necessarily unitary) GL2 (Qp )-Banach V together with
a GL2 (Qp )-equivariant topological surjection B(α)/L(α)  V . We can find norms on
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 55

the two Banach such that we have a surjection of unit balls (B(α)/L(α))0  V 0 . By
duality, we get a GL2 (Qp )-equivariant continuous injection of compact OE -modules:
HomOE (V 0 , OE ) ,→ HomOE ((B(α)/L(α))0 , OE ).
lim N (T ) is a compact OE -lattice of (B(α)/L(α))∗ which is preserved
We have seen that ←−
ψ
by B(Qp ). Hence, multiplying by a scalar in E × if necessary, we get a continuous B(Qp )-
equivariant injection of compact OE -modules HomOE (V 0 , OE ) ,→ ←− lim N (T ). If V 6= 0,
ψ
Cor.8.1.6 together with the translation of the action of B(Qp ) in terms of ψ and Γ imply
∼ ∼
HomOE (V 0 , OE )⊗E → (B(α)/L(α))∗ , which implies B(α)/L(α) → V . Hence B(α)/L(α)
(and B(β)/L(β) by Cor.7.2.3) is topologically irreducible.

We see from the previous proof that B(α)/L(α) and B(β)/L(β) are even topologically
irreducible as B(Qp )-representations.

Corollary 9.3.3. — The Banach B(α)/L(α) and B(β)/L(β) are admissible.

Proof. — We don’t know if the compact OE -module ←−

lim N (T ) is preserved by GL2 (Zp )
ψ
inside
b ∗
←− D(V )) ' (B(α)/L(α)) ,
(lim
ψ

but we can replace it by M := ∩g∈GL2 (Zp ) g(lim

←−ψ N (T )) ⊂ ←−
lim N (T ) which is a non zero
ψ
compact OE [[X]]-submodule as in Cor.8.1.6 (since it is easily checked to be preserved
by B(Qp ) using the Iwasawa decomposition of GL2 (Qp )). Note that there are now two
natural structures of OE [[X]]-module on M : the first is the one already defined and the
second is:    
0 1 0 1
(λ, v) ∈ OE [[X]] × M 7→ λ v.
1 0 1 0
Equivalently,
  the first is such that multiplication by (1 + X) corresponds to the action
Zp

1 Zp
of and the second is such that multiplication by (1 + X)Zp corresponds to the
0 1
 
1 0
action of . We have seen that M = ←− lim M (see proof of Cor.8.1.6) with M of
Zp 1 ψ
finite type over OE [[X]], M ⊂ N (T ) and ψ surjective on M . Let pr0 : M  M be the
projectionon the  first component and define N := Ker(pr0 ) ( M. The map N 7→ M ,
0 1

v 7→ pr0 v is injective: if v maps to 0, its associated distribution µ ∈ B(α)∗ '
1 0
C val(α) (Zp , E)∗ ⊕ C val(α) (Zp , E)∗ (see (6)) is zero on both copies of C val(α) (Zp , E) hence is
zero in (B(α)/L(α))∗ . Thinking in terms of distributions again, we also see that N is
an OE [[X]]-module for the first structure but only a ϕ(OE [[X]])-module for the second
structure (recall from (7) and Th.8.1.1 that multiplication by ϕ(1 + X) = (1 + X)p on
distributions correspond to the change of variables z 7→ z +p on functions). Moreover, for
56 LAURENT BERGER & CHRISTOPHE BREUIL

this second action, the above injection N ,→ M is ϕ(OE [[X]])-linear. Since M is of finite
type over OE [[X]], hence over ϕ(OE [[X]]), we thus get that the ϕ(OE [[X]])-module N for
the second action of ϕ(OE [[X]]) is of finite type. Now fix elements (e1 , · · · , em ) ∈ M (resp.
(f1 , · · · , fn ) ∈ N ) such that pr0 (ei ) (resp. fi ) generate M over OE [[X]] (resp. N over
P
ϕ(OE [[X]])). Let v ∈ M. There exist λ1 , · · · , λm in OE [[X]] such that  v − λi ei ∈N .
P P 0 1 0 1
Then there exist µ1 , · · · , µn in ϕ(OE [[X]]) such that v− λi ei = µ f.
1 0 i 1 0 i
 
1 Zp
Since the λi correspond to the action of elements in the group algebra of and the
0 1
     
0 1 0 1 1 0
µ correspond to the action of elements in the group algebra of ,
1 0 i 1 0 pZp 1
we see in particular that M is of finite type over the group algebra of GL2 (Zp ), whence
the admissibility.

10. Reduction modulo p (L.B.)

In the `-adic case (` 6= p), Vignéras has proved that one can reduce the classical local
Langlands correspondence over Q` modulo the maximal ideal of Z` and thus obtain a
new correspondence between semi-simple representations over F` ([36]).
It is tempting to do the same here with V and Π(V ). That is, for any V as in §7.1, we
have associated a non zero unitary admissible GL2 (Qp )-Banach Π(V ). Let Π(V )0 ⊂ Π(V )
(resp. T ⊂ V ) be any unit ball (resp. any OE -lattice) which is preserved by GL2 (Qp )
(resp. GQp ). In this last lecture, we want to state and prove some cases of a conjecture
relating the semi-simplification of Π(V )0 ⊗OE FE and the semi-simplification of T ⊗OE FE .

10.1. Statement of the conjecture. — In order to state the conjecture, we need to

make lists of some 2-dimensional Fp -representations of GQp and of some Fp -representations
of GL2 (Qp ).
Let us start with those of GQp . Let ω be the mod p cyclotomic character seen as a
character of Q×
p via class field theory, and let µ−1 be the unramified quadratic character
×
of Q×
p . We write IQp for the inertia subgroup of GQp . Let ω2 : IQp → Fp be Serre’s funda-
mental character of level 2 and for s ∈ {0, · · · , p}, let ind(ω2s ) be the unique (irreducible)
representation of GQp whose determinant is ω s and whose restriction to the inertia sub-
group of Gal(Qp /Qp2 ) is ω2s ⊕ ω2ps . We know that as r runs over 0, . . . , p − 1 and as η runs
×
over all characters η : Q× r
p → Fp , the representations ρ(r, η) := (ind(ω2 )) ⊗ η run over all
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 57

irreducible 2-dimensional Fp -representations of GQp and that the only isomorphisms are:

ρ(r, η) ' ρ(r, ηµ−1 )

ρ(r, η) ' ρ(p − 1 − r, ηω r )
ρ(r, η) ' ρ(p − 1 − r, ηω r µ−1 )
2
Now, we turn to representations of GL2 (Qp ). For r ∈ {0, · · · , p − 1}, let Symr Fp
denote the natural representation of GL2 (Zp ) acting through GL2 (Fp ) which we extend
to GL2 (Zp )Q×
p by sending p to 1. Let
GL (Q ) 2
indGL2 (Z p)Q× Symr Fp
2 p p

2
be the Fp -vector space of functions f : GL2 (Qp ) → Symr Fp which are compactly sup-
r
ported modulo Q× p and such that f (kg) = Sym (k)(f (g)) with the left action of GL2 (Qp ).
For r ∈ {0, · · · , p − 1}, and λ ∈ Fp , let
h 2
 i
GL (Q )
π(r, λ, η) := indGL2 (Z p)Q× Symr Fp /(T − λ) ⊗ (η ◦ det),
2 p p

where T is a “Hecke operator” which corresponds to the double coset

 
× 1 0
GL2 (Zp )Qp GL2 (Zp ).
0 p
Those representation are irreducible, and all smooth irreducible Fp -representations of
GL2 (Qp ) having a central character are given by (see [1],[2],[8]):
×
1. η ◦ det where η : Q×
p → Fp is a smooth character.
2. the Sp ⊗ (η ◦ det) where Sp is the “Special” representation.
×
3. the π(r, λ, η) where η : Q×
p → Fp is a smooth character and r ∈ {0, · · · , p − 1} and
λ ∈ Fp \ {−1, 1}.
When λ 6= 0, the π(r, λ, η) have no non trivial intertwinings (unless λ = ±1, in which
case there are some “easy” ones, see [2]) and when λ = 0, the only isomorphisms are:

π(r, 0, η) ' π(r, 0, ηµ−1 )

π(r, 0, η) ' π(p − 1 − r, 0, ηω r )
π(r, 0, η) ' π(p − 1 − r, 0, ηω r µ−1 ).

This certainly suggests that there is a correspondence between the representations ρ

of GQp and the representations π of GL2 (Qp ). More precisely, we define the following
“correspondence” ([8]):

Definition 10.1.1. — If λ = 0, then

π(r, 0, η) ↔ ρ(r, η)
58 LAURENT BERGER & CHRISTOPHE BREUIL

and if λ 6= 0, then
unr(λ−1 )ω r+1
 
ss −1 r+1 ss 0
π(r, λ, η) ⊕ π([p − 3 − r], λ , ω η) ↔ ⊗η
0 unr(λ)
where unr(λ) is the unramified character sending Frobenius to λ, ss denotes the semisim-
plification and [x] is the integer in {0, · · · , p − 2} which is congruent to x mod p − 1.

On the other hand, we have associated in the previous chapters a unitary representation
Π(V ) of GL2 (Qp ) to a crystalline representation V of GQp . Let V mod p denotes the
semisimplification of T /pT where T is some Galois invariant lattice of V and let Π(V )
be the semisimplification of Π0 (V )/p where Π0 (V ) is some GL2 (Qp )-invariant unit ball
of Π(V ).

Conjecture 10.1.2. — The representation Π0 (V )/p is of finite length (so that Π(V ) is
well-defined) and Π(V ) corresponds to V mod p under the correspondence of definition
10.1.1.

One application of this conjecture is that it is in principle easier to compute Π(V ) than
to compute V mod p given Dcris (V ). Note that using [2] and the definition of Π(V ),
this conjecture is easily checked to be true when V is reducible. We thus only consider
V irreducible in the sequel.
For ap ∈ E with positive valuation, let Vk,ap be the crystalline representation with
∗
Hodge-Tate weights (0, k − 1) such that Dcris (Vk,ap
) = Dk,ap = Ee1 ⊕ Ee2 where:

Dk,ap if i ≤ 0,
(
ϕ(e1 ) = pk−1 e2

i
and Fil Dk,ap = Ee1 if 1 ≤ i ≤ k − 1,
ϕ(e2 ) = −e1 + ap e2 
0 if i ≥ k.
In the notation of the previous lectures, we have Vk,ap = V (α, β) where α and β are
the roots of the polynomial X 2 − ap X + pk−1 = 0. The description with k and ap is a bit
more convenient for computations associated to Wach modules.
It is possible to compute Π(Vk,ap ) “by hand” for small values of k and one deduces from
10.1.2 some predictions for what Vk,ap mod p should be. Indeed, we have the following
theorem (see [9, théorème 1.4]):

Theorem 10.1.3. — Suppose that val(ap ) > 0.

1. If 2 ≤ k ≤ p + 1, then Π(Vk,ap ) = π(k − 2, 0, 1).
2. If k = p + 2 and p 6= 2, then
(a) if val(ap ) < 1, then Π(Vk,ap ) = π(p − 2, 0, ω) ' π(1, 0, 1).
(b) if val(ap ) ≥ 1, then Π(Vk,ap ) = π(p − 2, λ, ω)ss ⊕ π(p − 2, λ−1 , ω)ss where λ2 −
(ap /p)λ + 1 = 0.
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 59

If p + 3 ≤ k ≤ 2p and p 6= 2, then
(a) if val(ap ) < 1, then Π(Vk,ap ) = π(2p − k, 0, ω k−1−p ) ' π(k − 1 − p, 0, 1).
(b) if val(ap ) = 1, then Π(Vk,ap ) = π(k − 3 − p, λ, ω)ss ⊕ π(2p − k, λ−1 , ω k−1−p )ss
where λ = (k − 1)ap /p.
(c) if val(ap ) > 1, then Π(Vk,ap ) = π(k − 3 − p, 0, ω).

In the remainder of this chapter, we will prove some of the formulas for Vk,ap mod p
which are predicted by the above theorem (via Conj.10.1.2). In particular, we will explain
the proof of the following theorem:

Theorem 10.1.4. — 1. If k ≤ p + 1, then Vk,ap mod p = ind(ω2k−1 ).

2. If k = p + 2 and val(ap ) > 1 then
√ √
Vk,ap mod p = unr( −1)ω ⊕ unr(− −1)ω.

3. If p + 3 ≤ k ≤ 2p − 1 and val(ap ) > 1 then Vk,ap mod p = ind(ω2k−1 ).

When ap = 0, one can explicitly describe Vk,0 in terms of Lubin-Tate characters and
so we can also describe Vk,0 mod p for all k. When k ≤ p, then the theory of Fontaine-
Laffaille gives us an explicit description of Vk,ap mod p regardless of the valuation of ap
and the theorem is then straightforward. Our method of proof for theorem 10.1.4 is to
show that once we fix k, then if we vary ap a little bit, Vk,ap mod p does not change.
In order to do this, we compute the Wach module associated to Vk,ap as explicitly as
possible.

10.2. Lifting Dcris (V ). — In this paragraph, we’ll explain the general strategy for
constructing the Wach module N (V ) attached to a crystalline representation V if we are
given the filtered ϕ-module Dcris (V ). Recall from theorem 5.5.1 in §5.5 that N (V )/X ·
N (V ) is a filtered ϕ-module isomorphic to Dcris (V ). Furthermore, the functor V 7→
Dcris (V ) is fully faithful, so if we can construct some Wach module N such that N/X ·N =
Dcris (V ), then we necessarily have N = N (V ).
Note that both the ϕ-module structure and the filtration on N/XN depend solely
on the map ϕ : N → N and not at all on the action of ΓQp . Furthermore, given an
admissible filtered ϕ-module D, it is easy to construct a ϕ-module N over OE [[X]] such
that N/XN = D. The problem then is that it is very hard in general to find an action
of ΓQp which commutes with ϕ.
Therefore, in order to construct the Wach module N (V ) starting from Dcris (V ), one
has to find a “lift” of ϕ which is such that we can then define an action of ΓQp which will
commute with that lift. I do not know of any systematic way of doing this in general.
60 LAURENT BERGER & CHRISTOPHE BREUIL

If ap = 0, then we can carry out this program easily enough and the resulting Wach
module was given in §5.1 when k = 2. Let us generalize it to arbitrary k ≥ 2. The Wach
∗
module N (Vk,0 ) is of rank 2 over B+
Qp generated by e1 and e2 with ϕ(e1 ) = q
k−1
e2 and
ϕ(e2 ) = −e1 . The action of γ ∈ ΓQp is given by:
k−1
log+ (1 + X)

γ(e1 ) = e1
γ(log+ (1 + X))
k−1
log− (1 + X)

γ(e2 ) = e2
γ(log− (1 + X))
where
Y ϕ2n+1 (q) Y ϕ2n (q)
log+ (1 + X) = and log− (1 + X) = ,
n≥0
p n≥0
p
as previously.

10.3. Continuity of the Wach module. — One can also think about the above
constructions in terms of matrices. Given a Wach module N , we can fix a basis and let
P ∈ M(d, B+
Qp ) be the matrix of Frobenius in that basis. The problem is then to construct
an action of ΓQp , that is for every γ ∈ ΓQp , we need to construct a matrix G ∈ GL(d, B+
Qp )
such that P ϕ(G) = Gγ(P ). In practice we need do this only for a topological generator
γ of ΓQp (which is procyclic for p 6= 2).
In this paragraph, we use this approach to show that if we know a Wach module N
and we “change” P slightly, then this defines another Wach module. More to the point,
we have the following proposition:

Proposition 10.3.1. — Let N be a Wach module and let P and G be the matrices
of ϕ and γ ∈ ΓQp . Define α(r) = vp (1 − ε(γ)) · (1 − ε(γ)2 ) · · · (1 − ε(γ)r ). If H0 ∈
M(d, pα(k−1)+C A+ C +
Qp ) with C ≥ 0, then there exists H ∈ M(d, p AQp ) such that if we set
Q = (Id + H)P , then the ϕ-module defined by the matrix Q admits a commuting action
of ΓQp .

The proof of this rests on the following technical lemma:

Lemma 10.3.2. — If N is a Wach module and if G ∈ GL(d, A+

Qp ) is the matrix of
γ ∈ ΓQp , and if H0 ∈ M(d, pα(k−1)+C A+ C +
Qp ) with C ≥ 0, then there exists H ∈ M(d, p AQp )
such that:
1. H ≡ H0 mod X.
2. (Id + H)−1 Gγ(Id + H) ≡ G mod X k−1 .

The idea of the proof of proposition 10.3.1 is then that if P, G are the matrices of
ϕ and γ on N , then the lemma above and the fact that P ϕ(G) = Gγ(P ) imply that
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 61

G − (Id + H)P ϕ(G)γ(P (Id + H))−1 ≡ 0 mod X k−1 and one can then construct a matrix
for γ by successive approximations starting from G (see [7, §3.1] for more details).
The proof of lemma 10.3.2 is also by successive approximations and we sketch it below.

Proof. — If we write explicitly the condition (2) of the proposition, then we get:

(Id +XG1 +X 2 G2 +· · · )γ(H0 +XH1 +· · · ) = (H0 +XH1 +· · · )(Id +XG1 +X 2 G2 +· · · )

and bearing in mind that γ(X r ) ≡ ε(γ)r X r mod X r+1 , we can solve the above equation
mod X r in r successive steps. The first is to find H1 such that (1−ε(γ))H1 = G1 H0 −H0 G1
so that
H1 ∈ M(d, pα(k−1)+C−vp (1−ε(γ)) A+
Qp ).

The second is to find H2 such that (1 − ε(γ)2 )H2 = (some Zp -linear combination of
products of G0 , G1 , G2 , H0 , H1 ) so that
2
H2 ∈ M(d, pα(k−1)+C−vp (1−ε(γ))−vp (1−ε(γ) ) A+
Qp ).

Eventually, we get H0 , · · · , Hk−1 ∈ M(d, pC A+

Qp ) and we can then set H = H0 + XH1 +
· · · + X k−1 Hk−1 .

We can then apply proposition 10.3.1 to the matrices

   
0 −1 0 0
P = k−1 and H0 =
q 0 −ap 0
to get a Wach module Nk,ap as soon as the valuation of ap is large enough, and a straight-
∗
forward computation shows that Nk,ap /XNk,ap is then isomorphic to Dcris (Vk,a p
).
An explicit computation of α(k − 1) and some refinements of the computations above
specific to this case then give us the proof of theorem 10.1.4.

Remark 10.3.3. — To conclude, let us point out that using the description of Π(V )∗
with (ϕ, Γ)-modules given in the previous lectures, we can prove that Π0 (V )/p is of finite
length in general and we can also prove conjecture 10.1.2 in many cases, including all the
cases when T /pT is irreducible (forthcoming work).

References
[1] Barthel L., Livné R., Modular representations of GL2 of a local field: the ordinary,
unramified case, J. of Number Theory 55, 1995, 1-27.
[2] Barthel L., Livné R., Irreducible modular representations of GL2 of a local field, Duke
Math. J. 75, 1994, 261-292.
[3] Berger L., Représentations p-adiques et équations différentielles, Inv. Math. 148,
2002, 219-284.
[4] Berger L., Bloch and Kato’s exponential map: three explicit formulas, Doc. Math.
Extra Volume: Kazuya Kato’s Fiftieth Birthday, 2003, 99-129.
62 LAURENT BERGER & CHRISTOPHE BREUIL

[5] Berger L., Limites de représentations cristallines, to appear in Comp. Math.

[6] Berger L., Équations différentielles p-adiques et (ϕ, N )-modules filtrés, preprint.
[7] Berger L., Li H., Zhu H. J., Construction of some families of 2-dimensional crys-
talline representations, Math. Ann. 329, 2004, 365-377.
[8] Breuil C., Sur quelques représentations modulaires et p-adiques de GL2 (Qp ) I, Comp.
Math. 138, 2003, 165-188.
[9] Breuil C., Sur quelques représentations modulaires et p-adiques de GL2 (Qp ) II, J.
Institut Math. Jussieu 2, 2003, 23-58.
[10] Breuil C., Invariant L et série spéciale p-adique, to appear in Ann. Scient. E.N.S.
[11] Breuil C., Série spéciale p-adique et cohomologie étale complétée, available online at:
http://www.ihes.fr/∼ breuil/publications.html.
[12] Cherbonnier F., Colmez P., Représentations p-adiques surconvergentes, Inv. Math.
133, 581-611, 1998.
[13] Colmez P., Représentations cristallines et représentations de hauteur finie, J. Reine
Angew. Math. 514, 1999, 119-143.
[14] Colmez P., Arithmétique de la fonction zêta, available online at:
http://math.polytechnique.fr/xups/vol02.html
[15] Colmez P., La conjecture de Birch et Swinnerton-Dyer p-adique, Séminaire Bourbaki
919, juin 2003.
[16] Colmez P., in preparation.
[17] Colmez P., Fontaine J.-M., Construction des représentations p-adiques semi-stables,
Inv. Math. 140, 2000, 1-43.
[18] Deligne P., Formes modulaires et représentations de GL2 , Modular functions of one
variable II, Lecture Notes 349, 1973, 55-105.
[19] Emerton M., p-adic L-functions and completions of representations of p-adic reduc-
tive groups, preprint 2004.
[20] Emerton M., Locally analytic representation theory of p-adic reductive groups: A
summary of some recent developments, preprint 2003.
[21] Fontaine J-M., Le corps des périodes p-adiques, Astérisque 223, 1994, 59-111.
[22] Fontaine J-M., Représentations p-adiques semi-stables, Astérisque 223, 1994, 113-
184.
[23] Fontaine J-M., Représentations p-adiques des corps locaux I, The Grothendieck
Festschrift II, Progr. Math. 87, Birkhäuser, 1990, 249-309.
[24] Fontaine J-M., Wintenberger J-P., Le “corps des normes” de certaines extensions
algébriques de corps locaux, C. R. Acad. Sci. Paris 288, 1979, 367-370.
[25] Harris M., Taylor R., On the geometry and cohomology of some simple Shimura
varieties, Ann. Math. Studies 151, Princeton Univ. Press, 2001.
[26] Henniart G., Une preuve simple des conjectures de Langlands locales pour GLn sur
un corps p-adique, Inv. Math. 139, 2000, 439-455.
[27] Mazur B., Tate J., Teitelbaum J., On p-adic analogues of the conjectures of Birch
and Swinnerton-Dyer, Inv. Math. 84, 1986, 1-48.
[28] Schikhof W., An introduction to p-adic analysis, Cambridge Studies in Advanced
Math. 4, Cambridge University Press, 1984.
[29] Schikhof W., Non-archimedean calculus, preprint, university of Nijmegen, 1978.
 TOWARDS A p-ADIC LANGLANDS PROGRAMME 63

[30] Schneider P., Nonarchimedean Functional Analysis, Springer-Verlag, 2001.

[31] Schneider P., Teitelbaum J., Locally analytic distributions and p-adic representation
theory, with applications to GL2 , J. Amer. Math. Soc. 15, 2002, 443-468.
[32] Schneider P., Teitelbaum J. (with an appendix by D. Prasad), U (g)-finite locally
analytic representations, Representation Theory 5, 2001, 111-128.
[33] Schneider P., Teitelbaum J., Banach space representations and Iwasawa theory, Israel
J. Math. 127, 2002, 359-380.
[34] Schneider P., Teitelbaum J., Algebras of p-adic distributions and admissible repre-
sentations, Inv. Math. 153, 2003, 145-196.
[35] Serre J.-P., Sur les représentations modulaires de Gal(Q/Q), Duke Math. J. 54,
1987, 179-230.
[36] Vignéras M.-F., Correspondance de Langlands semi-simple pour GLn (F ) modulo ` 6=
p, Inv. Math. 144, 2001, 177-223.
[37] Wach N., Représentations cristallines de torsion, Comp. Math. 108, 1997, 185-240.
[38] Wintenberger J-P., Le corps des normes de certaines extensions infinies des corps
locaux; applications, Ann. Scient. E.N.S. 16, 1983, 59-89.

Hangzhou, 16-27 August 2004

Laurent Berger
Christophe Breuil