CDM Template PDF
CDM Template PDF
Contents
1. Introduction 2
2. Part I 4
2.1. Primes 4
2.3. The Riemann zeta function 5
2.5. Dirichlet L-functions 7
2.6. Artin L-functions 8
2.8. L-functions of p-adic representations 10
2.13. The Sato–Tate conjecture 17
2.15. Langlands L-functions 19
2.20. Reciprocity and the proof of the Sato–Tate conjecture 24
2.23. Evidence for the reciprocity conjecture 27
2.27. Our main theorem 30
3. Part II 30
3.1. First ideas on how to prove the main theorem 30
3.6. The eigencurve 36
3.8. Analytic continuation of functoriality 39
The author’s work received funding from the European Research Council (ERC) under
the European Union’s Horizon 2020 research and innovation programme (grant agreement
No 714405).
c 2022 International Press
1
2 J. A. THORNE
1. Introduction
Consider Ramanujan’s function τ (n) : N → Z, defined by the formula
Y X
∆(q) = q (1 − q n )24 = τ (n)q n .
n≥1 n≥1
−s ,
P
He also considered the Dirichlet series L(∆, s) = n≥1 τ (n)n conjectur-
ing the following properties:
• There is an Euler product expansion
Y
L(∆, s) = (1 − τ (p)p−s + p11−2s )−1 .
p prime
and conjectured that they admit an analytic continuation to the whole com-
plex plane C, satisfying the functional equation Λm (s) = Λm (11m + 1 − s),
where Λm is the completed L-function (a product of Lm (s) and certain
explicitly given Γ-factors). Deligne’s construction [Del71a] of the represen-
tations ρ∆,p showed that these L-functions may be placed within the broader
class of L-functions associated to the p-adic representations of GQ appearing
in the étale cohomology of algebraic varieties over Q (defined in [Ser70]).
Indeed, Deligne’s construction implies that the composite
Symm ρ∆,p : GQ → GL2 (Qp ) → GLm+1 (Qp )
of ρ∆,p with the mth symmetric power of the standard representation of GL2
has such a realisation, and its L-function is Lm (s). The functions Lm (s) are
the symmetric power L-functions associated to Ramanujan’s modular form
∆.
Around the same time, Langlands introduced in [Lan70] a class of
L-functions associated to pairs (π, R) consisting of an automorphic rep-
resentation π of a reductive group G over Q and an L-homomorphism
R : L G → GLn from the Langlands dual group. His fundamental functo-
riality conjecture predicted that these L-functions should be associated to
a ‘functorial lift’ R∗ (π), which would be an automorphic representation of
GLn (AQ ). In the special case where G = GL2 , the possible choices of R are
(up to twist, and restricting to irreducible representations R) precisely the
symmetric powers Symm : GL2 → GLm+1 of the standard representation
of GL2 . When π = π∆ is the automorphic representation of GL2 (AQ ) asso-
ciated to the Ramanujan ∆ function, the L-functions L(π∆ , R, s) are none
other than the L-functions Lm (s) introduced above from the point of view
of arithmetic geometry.
The goal of this article is give an introduction to the proof, by Newton
and the author, of the existence of Langlands’s functorial lift Symm ∗ (π∆ ),
and therefore the analytic continuation of Serre’s L-function Lm (s) [NT21a,
NT21b]. (More generally, we establish the existence of the symmetric power
4 J. A. THORNE
2. Part I
2.1. Primes. Let us begin by playing the following game. We list the
prime numbers
2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, . . .
2, 3, 5, 7, 1, 3, 7, 9, 3, 9, 1, 7, 1, 3, 7, 3, 9, 1, 7, 1, . . .
1 3 7 9
100 < p < 1000 35 35 40 33
1000 < p < 10000 266 268 262 265
10000 < p < 100000 2081 2092 2103 2087
We see straight away that the last digits of primes are remarkably evenly
distributed between the different residue classes. That this should be the case
is a consequence of Dirichlet’s theorem1 on primes in arithmetic progressions:
Theorem 2.2 (Dirichlet, 1837). Let a, N ∈ N be coprime. Then
#{p < X prime | p ≡ a mod N } 1
lim = ,
X→∞ #{p < X prime} φ(N )
where φ(N ) = #(Z/N Z)× .
In other words, for any modulus N , the primes are distributed evenly
between the different possible residue classes modulo N . By the end of the
first part of this article, we will have seen sweeping generalisations of both
the statement of this theorem and the methods used in the proof.
2.3. The Riemann zeta function. Before we get to the proof, let us
take a step back and consider how the primes themselves are distributed.
Define
π(X) = #{p < X prime},
the prime counting function. The asymptotic behaviour of π(X) is the sub-
ject of the Prime Number Theorem:
Theorem 2.4 (Hadamard, de la Vallée Poussin, 1896). As X → ∞,
π(X) ∼ X/ log(X).
The first proofs of the Prime Number Theorem used the connection
between the primes and the Riemann zeta function
X
ζ(s) = n−s .
n≥1
We consider this sum in the first instance as defined for complex numbers
s such that Re(s) > 1. In this case the sum is absolutely convergent, and
ζ(s) then defines a holomorphic function in the right half-plane Re(s) > 1.
The first thing that we need to know is the Euler product expression for the
Riemann zeta function:
X Y
ζ(s) = n−s = (1 − p−s )−1 ,
n≥1 p prime
again valid in the region Re(s) > 1. This can be viewed as an analytic
expression of the fundamental theorem of arithmetic: we may expand the
1In fact, the statement given here, which computes the natural density of primes in
a given residue class, is stronger than Dirichlet’s original statement (often given in terms
of the Dirichlet density of primes in a given residue class).
6 J. A. THORNE
where Λ(n) is the von Mangoldt function, taking the value log p when n = pk
is a prime power and the value 0 otherwise. An elementary argument shows
that the prime number theorem is equivalent to the asymptotic
X
(2.1) Λ(n) ∼ X as X → ∞.
n<X
If ζ(s) has the claimed properties, then its logarithmic derivative is mero-
morphic in C and holomorphic in the region Re(s) ≥ 1 (excepting the point
s = 1, where it has a simple pole). A Tauberian theorem2 can then be applied
which leads directly to the asymptotic (2.1).
How does one obtain these properties of the Riemann zeta function?
The existence of the meromorphic continuation of the zeta function to the
whole complex plane was known already to Riemann, who gave two proofs
of its existence in his famous monograph [Rie60]. The proof that is of the
greatest interest to us is the expression of the zeta function as an integral
transform of the Jacobi theta function θ(τ ). This is our first encounter in
this article with an automorphic form.
The theta function
2
X
θ(τ ) = eπin τ
n∈Z
2For example, Ikehara’s Tauberian theorem [CQ15, Theorem 3.5.2], which states
that if f (s) = n≥1 an n−s is a Dirichlet series with non-negative coefficients, absolutely
P
convergent in Re(s) > 1, and such that f (s) admits a meromorphic continuation to a
neighbourhood of the half-plane Re(s) ≥ 1 which is holomorphic
P on the line Re(s) = 1,
except for a simple pole of residue C at s = 1), then n<X a n ∼ CX as X → ∞. A
version of this result where the an may be complex numbers is given in [Lan94, Ch. XV,
§3].
RECIPROCITY AND FUNCTORIALITY 7
is defined for τ in the complex upper half plane h = {τ ∈ C | Im(τ ) > 0}. It
follows quickly from the definition that θ(τ ) is holomorphic and satisfies the
transformation property θ(τ + 2) = θ(τ ). Its relation with ζ(s) is expressed
by the formula (valid a priori in Re(s) > 1):
1 ∞
Z
−s/2 dy
(2.2) π Γ(s/2)ζ(s) = (θ(iy) − 1)y s/2 .
2 y=0 y
Let ξ(s) denote the left-hand side of (2.2) (which might be called the com-
pleted ζ-function). We now use that the function θ(τ ) satisfies the additional
symmetry
p −1
(2.3) θ(τ ) = τ /i θ(−1/τ )
for all τ ∈ h. Splitting the integral (2.2) into two pieces gives an expression
and applying this gives a new expression
1 ∞
Z
1 dy
(2.4) ξ(s) = + (θ(iy) − 1)(y s/2 + y (1−s)/2 ) ,
s(s − 1) 2 y=1 y
where the integral is now absolutely convergent for every value of s ∈ C. We
see that ξ(s) therefore admits a meromorphic continuation to C (with poles
only at s = 0, 1) and satisfies the functional equation ξ(s) = ξ(1 − s). The
non-vanishing of ζ(s) on the line Re(s) = 1 relies on a more subtle argument
that we won’t discuss here (but see e.g. [New98]).
The equations relating θ(τ + 2) and θ(−1/τ ) with θ(τ ) should be seen
as generating a whole group of symmetries which preserve θ. The group
SL2 (R) acts on the complex upper half plane h by Möbius transformations:
a b aτ + b
·τ = .
c d cτ + d
1 2 0 −1
The matrices and together generate a finite index
0 1 1 0
subgroup Γ of SL2 (Z). In general, automorphic forms may be viewed as
functions on homogeneous spaces of real groups satisfying both differential
equations (here taken to be the Cauchy–Riemann equations, expressing the
fact that θ is holomorphic) and some kind of invariance property under an
arithmetic group such as Γ. They are acted upon by automorphic represen-
tations. We will return to this topic in §2.15.
Using Fourier analysis on the finite group (Z/N Z)× (in other words, the
character theory of finite groups), we can write the indicator function of the
residue class a mod N as
1 X
1a mod N (n) = χ(a−1 )χ(n mod N ).
φ(N ) χ
as χ varies over the characters of (Z/N Z)× . We already understand the case
of the trivial character, which is just the prime number theorem. For non-
trivial characters χ we need to show that the sum (2.6) is o(X/ log X), and
this can be shown to follow if L(χ, s) has a meromorphic continuation to C
which is holomorphic and non-vanishing on the line Re(s) = 1 (including
now at the point s = 1). The non-vanishing at s = 1 is the hardest part of
Dirichlet’s proof.
2.6. Artin L-functions. Let us now consider the problem of general-
ising Dirichlet’s L-functions. There are at least two reasons for doing this:
first, we might hope to be able to prove statements generalising Dirichlet’s
theorem on primes in arithmetic progressions. Second, the functions L(χ, s)
are beautiful objects in their own right; for example, their values at integer
arguments are expected to have deeper arithmetic significance, as exempli-
fied by the Dirichlet class number formula, which describes the value at
s = 1 when χ is a quadratic character.
To see the path to such a generalisation, we first revisit the definition
of the Dirichlet L-function, using results which go back to Gauss [Gau66].
For N ∈ N, we introduce the cyclotomic field KN = Q(e2πi/N ). This is a
Galois extension of Q, and its Galois group is isomorphic to the group of
units modulo N :
∼
(2.7) Gal(KN /Q) → (Z/N Z)×
RECIPROCITY AND FUNCTORIALITY 9
σa 7→ a,
where σa (e2πi/N ) =e2πia/N . In particular, if p is a prime number not dividing
N then there is an associated automorphism σp ∈ Gal(KN /Q) which satis-
fies σp (e2πi/N ) = e2πip/N . Identifying χ now with a character of Gal(KN /Q),
we have an equivalent expression
Y
L(χ, s) = (1 − χ(σp )p−s )−1 ,
p prime,p-N
We are now ready to define the L-factor Lp (ρ, s): writing V for the vector
space on which ρ acts, it is given by the formula
Lp (ρ, T ) = det(1 − σp T : V Ip → V Ip ).
Because the characteristic polynomial of a matrix depends only on its conju-
gacy class, this really does depend only on the prime number p and not the
choice of prime ideal p lying above it. It is a simple exercise to check that
when K = KN , ρ = χ, and χ is primitive, this recovers the local L-factor
associated to the Dirichlet L-function.
This definition of the L-function L(ρ, s) attached to a representation
ρ : Gal(K/Q) → GLn (C) (commonly called an Artin representation) was
given by Artin in 1924 [Art24]. One application of the Artin L-function is
to prove the Chebotarev density theorem:3
Theorem 2.7. Let K/Q be a Galois extension, and let C ⊂ Gal(K/Q)
be a conjugacy class. Then
#{p < X prime, unramified in K | σp ∈ C} #C
lim = .
X→∞ #{p < X prime} #G
This beautiful statement is a constantly useful tool in algebraic number
theory. When K = KN = Q(e2πi/N ), it reduces (using the isomorphism
(2.7)) to Dirichlet’s theorem. The proof is similar: using the character the-
ory of finite groups, we can write the indicator function 1C : G → C of
the conjugacy class C as a linear combination of characters of irreducible
representations of G:
X
1C (g) = hρ, 1C i · tr ρ(g).
ρ
3Although other approaches exist, particularly if one is willing to vary the base number
field (taken here to be Q). In particular, Chebotarev’s original proof, the history of which
is described in the enjoyable article [SL96], goes by reduction to the case of cyclotomic
extensions of an arbitrary base number field.
RECIPROCITY AND FUNCTORIALITY 11
a profinite group endowed with its Krull topology. Any Artin representation
determines, by inflation, a representation ρ : GQ → GLn (C), with no need
to mention a finite Galois extension through the Galois group of which ρ
factors. In fact, one can show that any continuous representation ρ : GQ →
GLn (C) necessarily factors through a finite quotient Gal(K/Q) (and in fact
is continuous even when C is endowed with the discrete topology).
The next step therefore is to introduce replace the field C of coefficients
by a non-archimedean local field, such as Qp , the completion of Q with
respect to its p-adic absolute value (for some prime number p). The class of
continuous representations
ρ : GQ → GLn (Qp )
is much larger than the class of representations ρ : GQ → GLn (C) (and in
fact too large: it is necessary to put restrictions on the representations under
consideration in order to be able to do things such as write down reasonable
L-functions).
The most basic example of a p-adic representation is the p-adic cyclo-
tomic character . This can be constructed from the homomorphisms
n : GQ → Gal(Kpn /Q) → (Z/pn Z)× .
These are compatible as n varies, in the sense that n+1 mod pn = n . Pass-
ing to the inverse limit, we obtain a continuous homomorphism (where Zp
denotes the ring of p-adic integers in Qp )
: GQ → lim(Z/pn Z)× = Z×
p = GL1 (Zp ) ⊂ GL1 (Qp ).
←−
n
One important property of that generalises is that if we evaluate it at a
Frobenius element σl ∈ GQ at a prime l 6= p, then the number (σl ) is not
just a p-adic number but in fact an integer, namely l.
The next examples of p-adic representations we consider are those as-
sociated to elliptic curves over Q. Let (E, ∞) be an elliptic curve over Q:
thus E is a smooth, projective curve over Q of genus 1 and ∞ ∈ E(Q) is
a marked rational point. There is a unique way to make E into a commu-
tative algebraic group with identity ∞. For any field extension K/Q, the
set E(K) of K-rational points is thus an abelian group, and E(Q) is even
a Z[GQ ]-module. To construct a p-adic representation, we consider just a
small part of E(Q), namely the subgroup of p∞ -torsion points (i.e. those
which are pn -torsion for some n ≥ 1; this is a Z[GQ ]-submodule because the
group law of E is defined over Q).
We can use complex analytic considerations to describe E(Q)[p∞ ] as
an abelian group. There is an isomorphism of topological groups E(C) ∼ =
S 1 × S 1 , hence
E(C)[p∞ ] = (S 1 × S 1 )[p∞ ] ∼
= (Q/Z × Q/Z)[p∞ ] ∼ = Qp /Zp × Q/Zp .
Again using the fact that the group law of E is defined over Q, we see
that the points of E(C)[p∞ ] are in fact defined over Q, and finally that
12 J. A. THORNE
E(Q)[p∞ ] ∼
= (Qp /Zp )2 . If we define
Tp E = HomZ (Qp /Zp , E(Q)[p∞ ]),
then Tp E is a free Zp -module of rank 2 which receives an action of GQ which
is continuous when Tp E is endowed with its p-adic topology. This is the p-
adic Tate module of E. Writing Vp E = Tp E ⊗Zp Qp , we see that Vp E is a
2-dimensional Qp -vector space and that we have constructed a continuous
representation
ρE,p : GQ → GLQp (Vp E).
How can we construct an L-function from the representation ρE,p ? Following
the construction of Artin, we would like to define4
Ll (ρE,p , T ) = det(1 − ρE,p (σl )T : (Vp E)Il → (Vp E)Il )−1 .
However, it is not a priori clear that this makes sense, since we want this
local L-factor to be an element of C(T ), but Vp E is a vector space over Qp ,
which is not in any natural way a subfield of C!
The representations ρE,p have some remarkable properties which save
the day. We first consider the case where l 6= p is a prime where the elliptic
curve has good reduction (the case for all but finitely many primes l 6= p,
namely those that do not divide the minimal discriminant ∆E of E). In
this case the inertia group Il acts trivially and ρE,p (σl ) is an endomorphism
of Vp E, so might be represented by a matrix with entries in Qp – but the
characteristic polynomial of this matrix has coefficients in Z! We in fact have
the identity
det(1 − σl T : Vp E → Vp E) = 1 − al T + lT 2 ,
where al is given by the formula
(2.8) al = l + 1 − #E(Fl )
(The set of Fl -points of E makes sense because of our assumption that E has
good reduction.) In particular, al is an integer. A similar formula holds when
l 6= p is a place of bad reduction. What about when l = p? The story here
is more complicated. The formula det(1 − σp T : (Vp E)Ip → (Vp E)Ip ) would
give the “wrong” answer. For example, if p is a prime of good reduction we
would like to obtain 1 − ap T + pT 2 , where ap is defined as above. However,
I
in this case we never have Vp = Vp p . This reflects the fact that something
special is happening at the prime p.
4Here we are using notation generalising that introduced in the last section for finite
Galois extensions. For each prime number l, we can extend the embedding Q → Ql to
an embedding Q → Ql of algebraic closures. This choice determines a homomorphism
GQl → GQ , where GQl is the absolute Galois group of Ql . If Fl denotes the residue field
of the valued field Ql (which is indeed an algebraic closure of Fl ) and GFl = Gal(Fl /Fl ),
then the reduction map GQl → GFl is surjective, its kernel Il ⊂ GQl is a closed subgroup,
and it make sense to speak of the coset σl Il ⊂ GQl ⊂ GQ which is the pre-image of the
Frobenius automorphism in GFl . Up to conjugacy in GQ , this coset depends only on the
prime l.
RECIPROCITY AND FUNCTORIALITY 13
The subject of p-adic Hodge theory explains that ρE,p contains infor-
mation both about #E(Fp ) and about the Hodge decomposition of the de
Rham cohomology of E(C). This begins with the work of Tate [Tat67],
who showed that for an elliptic curve (or more generally abelian variety)
with good reduction over Qp , there is a canonical isomorphism (of Cp [GQp ]-
modules, Cp being the completion of Qp , and Cp (k) being the twist of Cp
by the k th power of the p-adic cyclotomic character ):
HomQ (Vp E, Cp ) ∼
p = (H 1 (E, O) ⊗Q Cp ) ⊕ (H 0 (E, Ω1 ) ⊗Q Cp (−1)).
p p
If we did have Vp E = (Vp E)Ip then we’d have h1,0 = 2 and h0,1 = 0 contra-
dicting the fact that both of these numbers are equal to 1 (because E is a
curve of genus 1). In general, if V is a finite-dimensional Qp [GQp ]-module
then we have the inequality
X
dimQp HomQp (V, Cp (k))GQp ≤ dimQp V,
k∈Z
which inspires the following definition:
Definition 2.9. Let V be a finite-dimensional Qp [GQp ]-module. We say
that V is Hodge–Tate if k∈Z dimQp HomQp (V, Cp (k))GQp = dimQp V . If
P
V is Hodge–Tate, we define the multiset of Hodge–Tate numbers of V to
be the set of integers k such that dimQp HomQp (V, Cp (−k))GQp 6= 0, each
appearing with multiplicity equal to the dimension of this finite-dimensional
Qp -vector space.
These ideas were greatly extended and refined by Fontaine (see e.g.
[Fon82, Fon94a, Fon94c]) who introduced several more interesting cate-
gories of Qp [GQp ]-modules, each contained inside the next:
(Hodge–Tate) ⊃ (de Rham) ⊃ (potentially semi-stable)
⊃ (semi-stable) ⊃ (crystalline) .
As an indication of the utility of these categories, Fontaine defined a functor
Dcrys from the category of crystalline Qp [GQp ]-modules to the category of
filtered φ-modules: i.e. the category of finite-dimensional Qp -vector spaces D
endowed with a Qp -linear map φ : D → D and a decreasing filtration Fil• D.
Moreover, he showed that if E is an elliptic curve with good reduction over
Qp , then there is a functorial isomorphism with the algebraic de Rham
cohomology
Dcrys (Vp E) ∼
= HdR1
(E/Qp )
and an equality
det(1 − φT : Dcrys (Vp E) → Dcrys (Vp E)) = 1 − ap T + pT 2 .
14 J. A. THORNE
Thus we can refine the Hodge–Tate decomposition and define the correct
local L-factor in this case. More generally, Fontaine gave a recipe to extract
from any potentially semi-stable Qp [GQp ]-module a Weil–Deligne represen-
tation (r, N ) [Fon94b].5 Even without the hypothesis of good reduction at
the prime p, the representation ρE,p |GQp is potentially semi-stable and one
can show that the associated Weil–Deligne representation gives the correct
local L-factor.
Having defined local L-factors at every prime, we can define
Y
L(E, s) = Ll (ρE,p , l−s ).
l prime
5 Weil–Deligne representations are a technical but useful tool, so let us say a few
words about them. By definition, the Weil group WQp is the subgroup of GQp consisting
of automorphisms which induce an integer power of Frobenius on the residue field Fp . It
is endowed with the topology which makes IQp into an open subgroup (not the subspace
topology of GQp ). A Weil–Deligne representation over a field Ω (say of characteristic 0)
is a pair (r, N ) where r : WQp → GLn (Ω) is a homomorphism with open kernel and
N : Ωn → Ωn is a nilpotent linear map satisfying r(σ)N = p−k N r(σ) for any σ ∈ WQp
k
inducing x 7→ xp on the residue field Fp .
The utility of this notion is that it is independent of the topology of the base field (cf.
[Tat79, §4.2]), and that one can associate to any potentially semi-stable representation
(as we will see below, this includes all p-adic representations which arises from geometry)
a Weil–Deligne representation.
RECIPROCITY AND FUNCTORIALITY 15
the p-adic cyclotomic character would fail on both counts). For representa-
tions which do have these properties, we’d like to define local L-factors as
for the p-adic representations attached to elliptic curves. If there is a number
field M/Q such that all the local
Q L-factors Ll (ρ, T ) have coefficients in M ,
then we can view L(ρ, s) = l prime Ll (ρ, l−s ) as a formal Dirichlet series
with coefficients in M . Given a choice of embedding ι : M → C, we would
then obtain a formal Dirichlet series L(ιρ, s) with coefficients in C. This
discussion motivates the following definition:
Definition 2.10. Let K/Qp be a finite extension. We say that a contin-
uous semi-simple representation ρ : GQ → GLn (K) is algebraic if ρ|GQp is
potentially semi-stable and for all but finitely many prime numbers l, ρ|GQl
is unramified.
A very rich source of p-adic representations which are algebraic in this
sense is the étale cohomology groups of algebraic varieties over Q. This
includes the p-adic Tate modules of elliptic curves. A basic computation in
the étale cohomology of algebraic curves shows that there is an isomorphism
of Qp [GQ ]-modules6
H 1 (E, Qp ) ∼
= HomQp (Vp E, Qp ).
This computation explains a slight change in normalisations that we may
as well mention now. We now write Frobl ∈ GQl ⊂ GQ for the inverse of
the element σl introduced previously, and call it the geometric Frobenius
element. From now on the local L-factors we consider will be defined using
geometric Frobenius elements as Ll (ρ, T ) = det(1 − Frobl T : V Il → V Il )−1 .
The reason for making this change is that the L-function associated to e.g.
the H 1 of an elliptic curve will agree with the L-function L(E, s) introduced
above. Geometric Frobenius elements are also more natural from the point
of view of étale cohomology (where they can be compared with the action
on étale cohomology induced by the geometric Frobenius endomorphism of
the reduction of E modulo l).
In general, let X be a smooth proper algebraic variety over Q, and let
X be its base change to Q. The étale cohomology groups H ∗ (X, Qp ) have
the following properties:
• For all but finitely many primes l (including those primes l 6= p
where X has good reduction7), the action of GQl on H ∗ (X, Qp ) is
unramified.
• If w is an integer, then H w (X, Qp ) is pure of weight w: for any
prime l 6= p where X has good reduction, the eigenvalues of Frobl on
H w (X, Qp ) are algebraic numbers, all of whose complex conjugates
have absolute value lw/2 .
6Where E denotes the base change to our fixed algebraic closure Q and the cohomol-
ogy is p-adic étale cohomology.
7In the sense that X arises as the generic fibre of a smooth proper scheme over Z .
l
16 J. A. THORNE
al = l + 1 − #E(Fl ) = αl + βl
√
respect to the Sato–Tate measure π2 1 − t2 dt: more precisely, for any con-
tinuous function f : [−1, 1] → R, we have
1/2 )
P
l<X,l-∆E f (al /2l 2 1
Z p
lim P = f (t) 1 − t2 dt.
X→∞ l<X,l-∆E 1 π t=−1
How is this analogous to the Chebotarev density theorem? That theo-
rem states that if ρ : GQ → GLn (C) is a continuous representation, then
the images ρ(σl ) of Frobenius elements are equidistributed within the con-
jugacy classes of ρ(GQ ). This formulation does not quite make sense here
since ρE,p is a p-adic representation, but the Sato–Tate conjecture con-
cerns the distribution of the real numbers al /2l1/2 . However, the polynomial
1 − al /l1/2 T + T 2 can arise as the characteristic polynomial det(1 − gT )
of an element g ∈ SU2 (R), so we can ask if the characteristic polynomials
det(1−ρE,p (σl )T ) of Frobenius elements are equidistributed within the set of
characteristic polynomials of elements of SU2 (R). The Sato–Tate conjecture
asserts that this is indeed the case: the map
SU2 (R) → [−1, 1]
1
g 7→ tr g
2
has fibres precisely the conjugacy classes of the group SU2 (R), and the Weyl
integration formula for SU2 (R) implies that for any continuous function
f : [−1, 1] → R there is an equality
Z Z π
1 1
f ( tr g) dg = |eiθ − e−iθ |2 f (cos θ) dθ.
G∈SU2 (R) 2 2π θ=0
a priori absolutely convergent and therefore holomorphic in the region Re(s) >
1, admits a meromorphic continuation to some neighbourhood of the line
Re(s) = 1 which is holomorphic and non-vanishing on the line Re(s) = 1.
This is our first point of contact with symmetric power L-functions! The
non-trivial irreducible representations of SU2 (R) are easy to describe: we
have the standard (or identity) representation of SU2 (R) ⊂ GL2 (C), and
its symmetric powers Symm C2 for each m ≥ 2 – and each irreducible rep-
resenation of SU2 (R) is isomorphic to exactly one of these. The standard
L-function L(E, C2 , s) (so-called because it corresponds to the standard rep-
resentation) equals the shifted usual Hasse–Weil L-function L(E, s+1/2), up
to finitely many Euler factors. The higher symmetric power L-functions are
genuinely new, but can similarly be described as shifts L(Symm ρE,p , s+m/2)
of the L-functions associated to the symmetric power Galois representations
Symm ρE,p : GQ → GLm+1 (Qp ).
These representations also come from geometry: the Künneth formula im-
plies that they can be realised inside the étale cohomology of the self-product
Em.
Serre has described a generalisation of the Sato–Tate conjecture in which
the elliptic curve E may be replaced by an arbitrary smooth, proper alge-
braic variety X over Q (or more generally, pure motive over Q). The key
point is the definition of the ‘Sato–Tate group’ K, a compact Lie group
which is the analogue of the group SU2 (R), which can be done either from
the point of view of the motivic Galois group [Ser94] or from the point of
view of p-adic representations [Ser12, Ch. 8]. In either case the generali-
sation of the Sato–Tate conjecture may be seen to follow from conjectural
properties of the family of L-functions associated to a sequence of conjugacy
classes of K and indexed by the set of irreducible representation of K.
l11−2s )−1 .
(2) Let l be a prime number, and factorise 1 − τ (l)T + l11 T 2 = (1 −
αl T )(1 − βl T ). Then |αl | = |βl | = l11/2 .
The first part of the conjecture was proved soon afterwards by Mordell
[Mor20], basically as a consequence of the fact that the fact that the mod-
ular form ∆ is an eigenvector for the Hecke operators Tl – of which more in
a moment. The second part of the conjecture remained unsolved for much
longer. A strategy to prove it was suggested by Langlands as part of his
functoriality conjectures [Lan70]. Langlands’s suggestion was to think of
the polynomial 1 − τ (l)l−11/2 T + T 2 as the determinant det(1 − tl T ) of a
uniquely defined conjugacy class of semisimple elements tl ∈ SL2 (C). For
any non-trivial irreducible algebraic representation R of SL2 (C), one can
then write down the L-function
Y
L(∆, R, s) = det(1 − l−s R(tl ))−1 .
l prime
If each of these L-functions, a priori absolutely convergent in some right
half-plane, admits an analytic continuation to the whole complex plane, then
the Ramanujan conjecture |αl | = |βl | = l11/2 holds.8 Since the non-trivial
algebraic representations of SL2 (C) are precisely the symmetric powers of
the standard representation C2 , this is our second point of contact with
symmetric power L-functions.
Langlands’s functoriality conjectures were in fact made in the context of
automorphic forms on an arbitrary reductive group G (say over a number
field M ), and the good properties of these L-functions expected to follow
as a consequence of the existence of functorial lifts, i.e. automorphic repre-
sentations of various other reductive groups associated to homomorphisms
of L-groups. For the sake of simplicity we explain some more of these ideas
here just in the case where G = GLm is a general linear group.
8This is an exercise using Landau’s lemma, namely that if P −s
n≥1 an n is a Dirichlet
series with non-negative real coefficients, absolutely convergent in Re(s) > σ and which
admits an analytic continuation to the half-plane Re(s) > σ0 for some σ0 < σ, then the
series is in fact absolutely convergent in the half-plane Re(s) > σ0 .
RECIPROCITY AND FUNCTORIALITY 21
(This makes sense because any element of GL2 (AQ ) can be written as a
product γg ∞ g∞ ; the fact that the value obtained is independent of the
choice of expression is essentially equivalent to the fact that ∆ is invariant
under the classical weight 12 action of SL2 (Z).) The function φ∆ is then
a cuspidal automorphic form, and it generates an irreducible GL2 (A∞ Q) ×
(Mn (C), On (R))-module which is a cuspidal automorphic representation π∆ .
The Hecke operators Tl which appear in the proof of the first part of Conjec-
ture 2.16 may be seen to be a shadow of the action of the group GL2 (A∞ Q)
on π∆ . Although the point of view of automorphic representations of adele
groups involves introducing a certain amount of new language and notation
(some of which we have elided here), it has value even for studying classical
22 J. A. THORNE
modular forms, since many questions become much more transparent from
the point of view of representation theory.
One example of this is the definition of the standard L-function asso-
ciated to an automorphic representation π of GLm (AQ ). We can describe
this in a way analogous to our definition of the L-function of a p-adic Galois
representation by using the local Langlands correspondence for the groups
GLm (Ql ). This is a bijection recQl between the following two sets of objects:
• The set of isomorphism classes of irreducible smooth representa-
tions of GLm (Ql ).
• The set of isomorphism classes of Frobenius semi-simple Weil–
Deligne representations (r, N ) of WQp of dimension n. (We have
defined Weil–Deligne representations in footnote 5; the condition
‘Frobenius semi-simple’ means that the representation r is semi-
simple.)
This bijection restricts to a bijection between unramified smooth represen-
tations and unramified Weil–Deligne representations. This unramified cor-
respondence is all that was known to exist at the time Langlands’s article
[Lan70], but is enough to describe L-functions up to finitely many Euler
factors. The standard L-function is given by the formula
Y
L(π, s) = Ll (recQl (πl ), l−s )−1 .
l prime
The modularity conjecture was proved for semistable elliptic curves over Q
by Wiles and Taylor [Wil95, TW95], and in general by Breuil–Conrad–
Diamond–Taylor [BCDT01]. The Sato–Tate conjecture was proved for most
elliptic curves over Q by Clozel–Shepherd-Barron–Harris–Taylor in 2007
[CHT08, Tay08, HSBT10] and in general by Barnet-Lamb–Geraghty–
Harris–Taylor in 2011 [BLGHT11]. However, the strategy used by these
authors to prove Sato–Tate was different to the one described above: they
established a weaker ‘potential’ version of Langlands functoriality for holo-
morphic newforms that suffices to establish the necessary properties of the
L-functions needed for the Sato–Tate conjecture.
In order to explain this more carefully, it is helpful to first back up and
describe the conjectural relation between Galois representations and auto-
morphic representations. This combines conjectures of Langlands, Clozel,
and Fontaine–Mazur, and includes the modularity conjecture for elliptic
curves over Q (and more generally, abelian varieties over any number field)
as a very special case. Which objects will participate in this correspon-
dence? We have already singled out irreducible algebraic Galois representa-
tions ρ : GQ → GLm (Qp ) as the right objects on the Galois side to which to
attach L-functions. We have also asserted that automorphic representations
of GLm (AQ ) have associated L-functions. However, it is not the case that
every automorphic representation should be associated to a Galois repre-
sentation. The simplest examples arise in the case m = 1, in which case
automorphic representations may be thought of equivalently as continuous
Hecke characters χ : Q× \A× ×
Q → C . There is no sensible way to associate
a p-adic Galois representation to e.g. the character x ∈ A× α
Q 7→ kxk , for
a transcendental complex number α.10 More interesting examples are given
by the automorphic representations of GL2 (AQ ) generated by non-algebraic
Maass forms; see [Gel75, §7] for explicit examples.
The right automorphic representations for this purpose are the algebraic
ones. Algebraicity is a condition on the infinite component π∞ of an auto-
morphic representation π, which may be phrased using the local Langlands
correspondence for GLm (R):11
10By contrast, when α is an integer, reciprocity for GL (which is just class field
1
theory) associates to this character an integral power of the p-adic cyclotomic character.
11Here we follow the definition given by Clozel [Clo90]. The twist by | · |(1−m)/2 is
included in order to ensure that the class of algebraic representations includes those that
contribute to the cohomology of congruence subgroups of GLn (Z). In the article [BG14],
Buzzard and Gee consider different possible notions of algebraicity, both for GLn and other
reductive groups, all of which however may ultimately be related by character twists.
26 J. A. THORNE
We remark that it can be seen already at this point that there is at most
one bijection with these properties. The strong multiplicity one theorem
[JS81] implies that π ∞ determines π∞ , so there is at most one automorphic
representation corresponding to any given ρ. On the other hand, the Cheb-
otarev density theorem implies that the character of an algebraic p-adic
representation ρ is determined by the characteristic polynomials det(1 −
T ρ(Frobv )) at unramified places v. If ρ is irreducible, then it is itself deter-
mined up to isomorphism by its character. If π and ρ are related, then we
usually write ρ = rι (π).
We say that a (say irreducible, algebraic) p-adic Galois representation ρ
is automorphic if it corresponds to an algebraic cuspidal automorphic repre-
sentation π of GLm (AM ). The reciprocity conjecture suggests an alternate
path to proving Langlands functoriality for algebraic automorphic represen-
tations of general linear groups. Many conjectured functorial properties of
automorphic representations become transparent after transporting to the
Galois side. If R : GLm → GLM is an algebraic representation, and π corre-
sponds to ρ, then the existence of the functorial lift R∗ (π) is implied by the
automorphy of the representation R ◦ ρ. Another expected property of au-
tomorphic representations is base change, or in other words a transfer from
automorphic representations of GLm (AM ) to automorphic representations
of GLm (AM 0 ) for any extension M 0 /M of number fields. On the Galois side,
this corresponds simply to replacing ρ by its restriction ρ|GM 0 to a finite
index subgroup.
We can now explain how the Sato–Tate conjecture for elliptic curves
over Q was proved: namely, by establishing that the symmetric power rep-
resentations Symm ρE,p are potentially automorphic, in the sense that there
([Eme], generalised further by Pan [Pan21]) using the p-adic local Lang-
lands correspondence for GL2 (Qp ).
Going beyond the case of GL2 (AQ ) presents myriad technical difficulties.
The main source of Galois representations in general is the étale cohomol-
ogy of Shimura varieties. We recall that, in Deligne’s formulation [Del71b],
a Shimura variety is a variety (or rather, family of varieties), defined over
a number field M and attached to a Shimura datum (G, X) consisting of
a reductive group G over Q and a homogeneous space X for G(R) satis-
fying certain axioms. Modular curves are the Shimura varieties associated
to the pair (GL2 , C − R). The étale cohomology of Shimura varieties (or
more generally p-adic local systems on Shimura varieties) receives an ac-
tion of GM × G(A∞ Q ). The basic idea, laid out in [Lan77], is to analyse
the cohomology by understanding the action of GM on the G(A∞ Q )-isotypic
pieces in terms of automorphic representations of G(AQ ), by comparing the
Grothendieck–Lefschetz trace formula in étale cohomology with the Arthur–
Selberg trace formula in the theory of automorphic forms.
The Shimura varieties with the best understood cohomology (beyond the
case of Shimura curves) are those attached to unitary groups of CM number
fields M (such as imaginary quadratic fields). One prediction of Langlands
functoriality is that the automorphic representations of these unitary groups
should be related to automorphic representations π of GLm (AM ) which are
conjugate self-dual, in the sense that the contragredient of π is isomorphic
to its image under the automorphism of GLm (AM ) induced by complex
conjugation on M . After several decades of work by many mathematicians,
beginning with that of Kottwitz [Kot92] and Clozel [Clo91], the Galois
representations associated to regular algebraic, cuspidal, conjugate self-dual
(RACSDC) automorphic representations of GLm (AM ) are now known to
exist in complete generality and to satisfy the expected local-global com-
patibility at every place.14 The case of essentially self-dual representations
of GLm (AM 0 ), where M 0 is a totally real field, can often be reduced to
this setting using a combination of base change and patching results (such
as the one exposited in [Sor20]). Strong automorphy lifting theorems are
also available for conjugate self-dual Galois representations; see especially
[BLGGT14, PT15] for automorphy lifting theorems and potential auto-
morphy theorems for compatible systems for Galois representations which
go far beyond what is needed to prove the Sato–Tate conjecture for ellip-
tic curves over totally real number fields. Our proofs of symmetric power
functoriality for automorphic representations of GL2 (AQ ) take place in this
powerful and flexible context.
Some work has also been done outside the RACSDC context, both for
Galois representations which are algebraic but not regular algebraic (starting
14We cannot begin to survey this story here, but see [Shi20] for a description of
these developments, which rely on many important advances in arithmetic geometry and
harmonic analysis.
30 J. A. THORNE
with [DS74] and continuing up to the recent work [BCGP21], which es-
tablishes the potential automorphy of the Galois representations associated
to abelian surfaces over totally real fields) and for Galois representations
over CM fields which are regular algebraic but not conjugate self-dual (as in
[ACC+ 18], which proves the Sato–Tate conjecture for elliptic curves over
CM fields). In both cases the needed Galois representations, which in general
do not appear in the étale cohomology of Shimura varieties [JT20], are con-
structed as p-adic limits of Galois representations which do appear in the
étale cohomology of Shimura varieties. Understanding the limits of these
techniques is an interesting topic for future work but leads in a different
direction to the one of interest for these notes. We invite the reader to look
at the survey [Cal21] for more about these exciting recent developments.
2.27. Our main theorem. Let us state again our main theorem.
Theorem 2.28. Let π be a regular algebraic, cuspidal automorphic rep-
resentation of GL2 (AQ ), without CM. Then for each m ≥ 1, there exists
a regular algebraic, cuspidal automorphic representation Π of GLm+1 (AQ )
such that for every prime number l, recQl (Πl ) = Symm ◦recQl (πl ).
Here are two important corollaries:
Corollary 2.29. (1) For each m ≥ 1, the L-function L(∆, Symm , s)
admits an analytic continuation to C.
(2) Let E be an elliptic curve over Q. Then for each m ≥ 1, the L-
function L(E, Symm , s) admits an analytic continuation to C.
We emphasise that the existence of a meromorphic continuation of these
L-functions (and the fact that they satisfy functional equations) is already a
consequence of the potential automorphy of the associated symmetric power
Galois representations. This potential automorphy also implies the Sato–
Tate conjecture. Our results have applications beyond this. For example,
Thorner [Tho21b] has used Theorem 2.28 to give a form of the Sato–Tate
conjecture with an effective rate of convergence.
Corollary 2.29 answers a question asked more than 50 years ago. Our
proof, however, relies on many ideas in the theory of p-adic automorphic
forms that have been developed much more recently. We will give an overview
of this proof in the second part of this article.
3. Part II
3.1. First ideas on how to prove the main theorem. The proof of
Theorem 2.28 occupies the articles [NT21a, NT21b] which in turn rely on
the papers [NT20, ANT20, AT20]. In this introductory section, we give an
overview of some of the themes that inform the most important arguments
in these papers, focusing on the case of symmetric power functoriality for
level 1 modular forms (such as ∆); as we will see, this special case serves as
a starting point for the proof of the general case.
RECIPROCITY AND FUNCTORIALITY 31
and consider auxiliary sets Q of prime numbers such that, allowing ramifi-
cation at the primes of Q, the maps
fl
Rρ,S∪Q → T(H 1 (Γ1 (NS∪Q ), O))m
fl
Rρ,S / T(H 1 (Γ1 (NS ), O))m .
fl
This means in essence that the tangent space of Rρ,S∪Q is the same size as
fl
that Rρ,S ; this is equivalent, thanks to duality theorems in Galois cohomol-
ogy, to the vanishing of a certain dual Selmer group (with coefficients in
the Tate dual ad ρ(1)), and can be arranged under some hypotheses using a
clever application of the Chebotarev density theorem. Roughly speaking, the
idea is that as Q varies, the rings T(H 1 (Γ1 (NS∪Q ), O))m become ‘closer and
closer’ approximations to OJX1 , . . . , Xg+1 K – this requires a kind of lower
bound on the growth of the spaces H 1 (Γ1 (NS∪Q ), O)m of modular forms
fl
as Q varies. Since Rρ,S∪Q is sandwiched between OJX1 , . . . , Xg+1 K and the
Hecke algebra T(H 1 (Γ1 (NS∪Q ), O))m , this should lead to useful information
about the relation between the Galois deformation ring R and Hecke algebra
T.
What are the drawbacks to using this approach to prove the automorphy
of representations of type Symm rι (π)? The most significant is that using the
Taylor–Wiles method requires the residual representation ρ to be quite non-
degenerate, not only so that the ring Rρ,S exists but also so that we can ‘kill
the dual Selmer group’. This will force us to restrict to primes p ≥ m + 1
(as otherwise the residual representation is forced by to reducible, because
Symm is a reducible representation of GL2 in characteristic p < m+1). Then
there is the separate issue of establishing the residual automorphy of these
residual representations GQ → GLm+1 (k), which generally have insoluble
image – which seems hard, if not impossible. We would rather be able to
take p = 2, π = π∆ , in which case the residual representation Symm rι (π∆ )
is trivial, but this takes us well outside of the realm in which one can apply
the Taylor–Wiles method.
In another approach, taken by Emerton in his work on the Fontaine–
Mazur conjecture [Eme], we keep the ring Rρ,S associated to a (supposed
absolutely irreducible) representation ρ : GQ → GL2 (k), and rather enlarge
the Hecke algebra T under consideration, by allowing non-classical (and
therefore purely p-adic) systems of Hecke eigenvalues. The approach taken in
[Eme] to define this enlarged Hecke algebra uses the completed cohomology
34 J. A. THORNE
points a bit more. Given a Hecke eigenform f ∈ Sk (Γ1 (pr )), we can associate
various quantities: first, the slope s(f ) = vp (ap (f )), where vp : Qp → Q ∪
{∞} is the p-adic valuation, normalised to have vp (p) = 1. We say that f
has finite slope if s(f ) < ∞, or equivalently if ap (f ) 6= 0.
We note that f need not be a newform. For example, the modular
form ∆ ∈ S12 (SL2 (Z)) is not an eigenvector for the Hecke operator Up ∈
18Coleman–Mazur also impose the condition p 6= 2. Buzzard extended the definition
of eigencurve to include the case p = 2 and also to the case of general tame level [Buz07].
Note as well that we consider only the cuspidal part of the eigencurve here (i.e. we excise
the contribution of the ordinary Eisenstein series). Many equivalent definitions of the
eigencurve now exist, some of which will be discussed in greater detail below.
RECIPROCITY AND FUNCTORIALITY 37
reason at all a priori why this space should have reasonable properties! The
fundamental result is proved in [CM98]:
Theorem 3.7. Let W = Hom(Z× p , Gm ), and κ : Ep → W be the map
−1 20
κ(ρ, δ) = δ|1×Z× . Then κ is, locally on Ep , finite flat: each point of Ep
p
admits an affinoid neighbourhood U such that κ|U is finite flat onto the image
κ(U) ⊂ W. In particular, Ep is equidimensional of dimension 1.
This theorem gives a sense to the idea that the points xπ,χ lie in p-adic
families indexed by the weight κ(xπ,χ ) ∈ W. In order to prove the theorem, it
is necessary to actually construct these families, and Coleman–Mazur do this
by describing all of the points of Ep (not just the classical points) in terms of
overconvergent modular forms of finite slope. Overconvergent modular forms
of a given weight w ∈ W lie in a Qp -Banach space Sw , which receives an ac-
tion of the Hecke operators Tl and a compact action of the operator Up , which
therefore has a discrete spectrum with finite-dimensional eigenspaces. Since
the work of Coleman–Mazur, many alternative constructions of the eigen-
curve have been given, using varying notions of p-adic modular form (see
e.g. [AS97, Che04, Eme06, Buz07, Urb11, AIP15, Han17]). What is
important for us is not so much the precise construction but the following
properties of the eigencurve, which are very useful in applications:
• Classicality criterion: we can define a continuous map s : Ep (Qp ) →
Q by s(ρ, δ) = vp (δ(p, 1)). If xπ,χ is the classical point associated to
a Hecke eigenform f of finite slope s(f ) = vp (ap (f )), then s(xπ,χ ) =
s(f ). If f has weight k, then s(f ) ∈ [0, k − 1]. Coleman’s classicality
criterion states that if x ∈ Ep (Qp ) and the character κ(x) agrees
with t 7→ tk−2 on an open subgroup of Z× p , and moreover s(x) ∈
[0, k − 1), then x is in fact a classical point.
• Accumulation property of classical points: classical points of the
eigencurve Ep satisfy a density property stronger than Zariski den-
sity in the ambient space X × T . More precisely, if x ∈ Ep (Qp ) is
a classical point, then for any affinoid neighbourhood U of x in Ep ,
the classical points of U are Zariski dense in U.
Although the eigencurve is a curve, in the sense of being a 1-dimensional rigid
space, it is very far from being an algebraic curve, and its geometry remains
rather mysterious. Some results on the local geometry are available (see
e.g. [DL16]) but it is not known in any case whether, for example, the set
of irreducible components is finite or infinite. One very intriguing question
concerns the geometry ‘close to the boundary of weight space’ – we will come
back to this very soon. Since the eigencurve is constructed as a finite cover
of the spectral variety of the Up operator, such questions are closely tied
up with questions about slopes of p-adic modular forms. Many mysterious
phenomena have been observed here which are yet to be explained – see for
example the fascinating conjectures in the articles [Buz05, BG16, BP19].
Theorem 3.9. Fix m ≥ 1. Let xπ,χ , xπ0 ,χ0 be classical points of the
eigencurve Ep which lie on a common irreducible component. Suppose that
the following conditions hold:
(1) Neither of π, π 0 has CM, and neither of π, π 0 is ι-ordinary.
(2) The refinement χ0 = χ01 ⊗χ02 is (m+1)-regular ([NT21a, Definition
2.23]). This means that for each i = 1, . . . , m we have (χ01 )i 6= (χ02 )i .
(3) Symm rι (π) is automorphic.
Then Symm rι (π 0 ) is automorphic.
We will sketch the proof of this theorem in the next section. First we
describe how it may be applied to obtain the following corollary:
One ideal scenario would be that the eigencurve Ep was irreducible (for
some prime p). In general this cannot be true since there are at least as
many components of Ep as there are odd, semisimple residual representations
ρ : GQ,{p} → GL2 (Fp ). As we have already remarked, there is a unique such
representation if we stick to the case p = 2. However, even in this case we
don’t know how to say anything about the set of irreducible components of
E2 .
Remarkably, however, Buzzard–Kilford [BK05] were able to compute
explicitly a large part of E2 , namely the part ‘at the boundary of weight
space’. In order to say what this means, let us make our description of
weight space W = Hom(Z× p , Gm ) slightly more explicit. When p = 2, we
can decompose Z× p = {±1}×(1+4Z 2 ). The eigencurve E2 is supported above
the connected component W+ of W corresponding to characters which are
trivial on −1onZ× p (essentially because there are no level 1 modular forms of
odd weight). The group 1+4Z2 is procyclic, so we may identify W+ with the
2-adic open unit disc {|w| < 1} by sending a character δ to w(δ) = δ(5) − 1.
Here is the result of Buzzard–Kilford:
40 J. A. THORNE
Theorem 3.11. Let A = {w | |8| < |w| < 1}. Then there is a decom-
position κ−1 (A) = t∞i=1 Xi as a countable disjoint union of admissible open
subspaces. The components Xi have the following properties:
(1) For each i ≥ 1, the restriction κ|Xi : Xi → A is an isomorphism.
(2) If x ∈ Xi (Qp ), then s(x) = ivp (w(κ(x))).
More informally, the pre-image of the boundary annulus A of weight
space W+ in E2 decomposes as an infinite disjoint union of copies of A, each
of which maps isomorphically to A. Moreover, on each copy the slope is a
simple function of the weight.
In order to prove Corollary 3.10, we want to be able to reach any point
from any other point by analytic continuation. This is too much to ask.
However, another move is available to us. This can be seen most trans-
parently for the classical points associated to everywhere unramified regular
algebraic cuspidal automorphic representations π. Such a representation will
admit two accessible refinements χ, χ0 (cf. the two p-stabilizations of ∆ writ-
ten down in (3.2) which determine two distinct classical points xπ,χ , xπ,χ0
on the eigencurve Ep . Although these points have the same image under the
map κ, they (usually) have distinct slopes! If π has weight k then the slopes
are related by the formula
s(xπ,χ ) + s(xπ,χ0 ) = k − 1.
Since the existence (or otherwise) of Symm
∗ (π) depends only on π, and not
on the choice of refinement, swapping refinements allows us to move be-
tween different components Xi of κ−1 (A) while preserving the property of
the existence of Symm 21
∗ (π) on that component . Moreover, every irreducible
component of E2 meets κ−1 (A) because the image of any irreducible compo-
nent of E2 under κ has finite complement in W+ , by [CM98, Theorem B].
Therefore every irreducible component of E2 contains some Xi . These moves
taken together are enough to establish Corollary 3.10.
3.12. Proof of the analytic continuation principle. We now sketch
the proof of Theorem 3.9, which is based on ideas going back to Kisin’s work
[Kis03] on the eigencurve and the Fontaine–Mazur conjecture. Recall that
we have defined the (tame level 1) eigencurve Ep as the Zariski closure inside
X × T of the set of classical points xπ,χ . One can ask how close the closed
immersion Ep ⊂ X × T is to being an isomorphism. The answer is: rather
far, since dim X = 3 and dim T = 2, while dim Ep = 1. Even ignoring the
factor of T , it is still the case that X contains points corresponding to many
21There is a slight wrinkle here: the classical points corresponding to level 1 forms
have even weights, and so do not lie above A (cf. the discussion after [NT21a, Lemma
3.3]). One can also define a ‘swapping refinements’ map for classical points which are
ramified at p by introducing a character twist, in order to ensure that the condition ‘χ1
unramified’ is preserved. Ultimately we find that if xπ,χ is a classical point lying in some
Xi , and xπ0 ,χ0 is its image under this swapping map, then xπ0 ,χ0 lies in Xi0 , where i, i0 are
related by the formula i + i0 = (k − 1)/vp (w(xπ,χ )).
RECIPROCITY AND FUNCTORIALITY 41
more Galois representations ρ : GQ,{p} → GL2 (Qp ) than the ones associated
to finite slope overconvergent Hecke eigenforms. We need to impose a more
stringent condition on Galois representations appearing in X .
In [Kis03], Kisin proposed such a condition. Roughly speaking, he de-
fined a closed subvariety Xf s ⊂ X × T (where f s stands for ‘finite slope’)
with the following properties:
• For ‘generic’ points (t, δ) ∈ X × T with t = tr ρ, ρ : GQ → GL(V ),
we have (t, δ) ∈ Xf s if and only if there is a GQp -equivariant ho-
momorphism
+
(3.3) V ,→ (Bcris ⊗Qp K)ϕ=δ(p,1) .
• There is an inclusion Ep ⊂ Xf s .
• At ‘generic’ classical points xπ,χ ∈ Ep (K), the inclusion Ep ⊂ Xf s
is a local isomorphism provided that the Bloch–Kato Selmer group
Hf1 (Q, ad rι (π)) is zero.
(We will recall the definition of the Bloch–Kato Selmer group in the next
section. Here we use the word generic in the loose sense that there are
certain technical conditions which are satisfied in many examples.) The ring
+
Bcris is the positive part of Fontaine’s ring of crystalline periods. Condition
(3.3) arises since it is classified for level 1 classical points (for which the
space of GQp -equivariant homomorphisms V → Bcris ⊗Qp K is in fact two-
dimensional, because rι (π)|GQp is crystalline) and behaves well in families.
Optimistically, one might ask if the inclusion Ep ⊂ Xf s is in fact an equality,
and this has been proved in many cases [Eme, Theorem 1.2.4].
Kisin’s condition was reformulated by Colmez [Col08], using a different
set of ideas from p-adic HodgeP theory. Let R be the Robba ring, i.e. the
n
ring of formal series f (X) = n∈Z an X (an ∈ Qp ) which converge in
the annulus (ρ, 1) for some 0 < ρ(f ) < 1. It is equipped with a Frobenius
endomorphism ϕ : R → R and a commuting action of the group Γ = Z× p . By
definition, a (ϕ, Γ)-module of rank n is a finite free R-module D equipped
with commuting semilinear actions of ϕ and Γ such that ϕ(D) generates
D. The category of (ϕ, Γ)-modules is not abelian, but Kedlaya proved that
its objects have a canonical slope decomposition [Ked04]. The fundamental
theorem relating (ϕ, Γ)-modules to p-adic representations is as follows (cf.
[CC98, Ked04]):
Theorem 3.13. There are mutually inverse equivalences of categories
Finite dimensional
continuous Qp [GQp ]-modules ↔ ((ϕ, Γ)-modules over R of slope 0) .
Symm
(X × T )tri / (Xm+1 × Tm+1 )tri ,
EG2
? / EG
m+1
i2 im+1
Symm
(X2,M × T2,M )tri / (Xm+1,M × Tm+1,M )tri ,
where:
24Recall this stands for regular algebraic, conjugate self-dual, cuspidal – see §2.23
of this article. The base change automorphic representation is the one whose associated
Galois representation is rι (π)|GM .
25At least, when we stick to regular algebraic automorphic representations π such
that rι (π)|GM is irreducible – see [BLGHT11, Lemma 1.4] for a positive statement.
26The work [KMSW14], the second half of which has yet to appear, will give a
complete description of the discrete automorphic representations of any unitary group.
44 J. A. THORNE
27The classicality theorem we need is somewhat stronger than the ‘numerically non-
critical’ implies classical’ theorem which is the naive generalisation of Coleman’s criterion
vp (ap ) < k − 1. This is because the symmetric powers of a 2-dimensional Galois repre-
sentation of small slope are no longer necessarily of small slope. However, they will (as
observed by Chenevier [Che11, Example 3.26] – this requires the (m + 1)-regular condi-
tion in the statement of Theorem 3.9) satisfy the weaker condition that each triangulation
is non-critical, cf. [NT21a, Lemma 2.7]. The criticality (or otherwises) of classical points
is closely tied to the local geometry of the eigenvariety around such points and has been
studied in great detail by Breuil–Hellmann–Schraen [BHS17, BHS19].
RECIPROCITY AND FUNCTORIALITY 45
We now come back to the result proved by Kisin [Kis03] for the eigen-
curve Ep : namely that the closed immersion Ep → Xf s (his analogue of our
map i2 ) is a local isomorphism in an affinoid neighbourhood of a classical
point xπ,χ such that the adjoint Bloch–Kato Selmer group Hf1 (Q, ad rι (π))
is 0 (and satisfying some other mild technical conditions). This holds in
our context, as can be seen using a tangent space calculation first done in
ths unitary group case by Bellaı̈che–Chenevier [BC09, Corollary 2.6.1]. The
essential point is that the local surjectivity of the eigenvariety over weight
space gives a lower bound for the dimension of the eigenvariety. If the adjoint
Bloch–Kato Selmer group vanishes and the character δ is sufficiently generic,
one can show that an upper bound for the tangent space of (X2,M × T2,M )tri
that is equal to this lower bound, showing that the eigenvariety is smooth
and that the map i2 is an isomorphism in a suitable affinoid neighbourhood.
Exactly the same argument is valid for EGm+1 . Thus the proof is complete,
provided we can show that
Hf1 (M + , ad Symm rι (π)|GM ) = 0.
This is non-trivial and requires a different set of ideas, which we address in
the next section.
where the product runs over the set of all finite places v of M + , and we
define
Hf1 (Mv+ , ad ρ) = Hur
1
(Mv+ , ad ρ)
to be the unramified subgroup if v - p and
Hf1 (Mv+ , ad ρ) = ker(H 1 (Mv+ , ad ρ) → H 1 (Mv+ , ad ρ ⊗Qp Bcrys ))
46 J. A. THORNE
3.18. Seed points. Let m ≥ 1. The work described in the last few
sections shows that, if we want to establish the existence of Symm ∗ (π) for
every everywhere unramified regular algebraic cuspidal automorphic repre-
sentation of GL2 (AQ ), then it is enough to show this existence for a single
RECIPROCITY AND FUNCTORIALITY 49
×
(1) There are continuous characters χ1 , . . . , χn : GM → Fp such that
there is an isomorphism
ρ∼= ⊕n χ , i=1 i
and moreover for each i = 1, . . . , n, χi χci = 1−n , and for each
1 ≤ i < j ≤ n, the character χi /χj |GM (ζp ) has order strictly greater
than 2n.
(2) n ≥ 3 and p > n.
(3) ρ is unramified at all but finitely many places of M and there is
λ = (λτ,i )Hom(M,Qp ) such that for each τ ∈ Hom(M, Qp ), λτ is
dominant in the sense that λτ,1 ≥ λτ,2 ≥ · · · ≥ λτ,n , and ρ is
ordinary of weight λ, in the sense that for each place v|p of M
there is an isomorphism
ψv,1 ∗ ∗ ∗
0 ψv,n ∗ ∗
ρ|GFv ∼
= .. ,
.. ..
. . . ∗
0 ··· 0 ψv,n
×
where for each i = 1, . . . , n, ψv,i : GFv → Qp is a continuous
character such that for some open subgroup Uv,i ⊂ OF×v and for all
u ∈ Uv,i , we have ψv,i ◦ ArtFv (u) = τ :Mv →Q τ (u)−λτ,n−i+1 +i−1 .
Q
p
(4) There is a RACSDC automorphic representation π of GLn (AM )
such that rι (π) ∼ = ρ and π is ι-ordinary.
(5) [F (ζp ) : F ] = p − 1.
(6) There exists a place v0 - p of M such that πv0 is an unramified twist
of the Steinberg representation and there is an unramified character
× ∼
ψ : GMv0 → Qp such that ρ|ss GM = ⊕ni=1 ψ1−i .
v0
developed Hida theory for definite unitary groups and used it to prove au-
tomorphy lifting theorems for ordinary, conjugate self-dual Galois represen-
tations in arbitrary rank (although with a residual irreducibility condition).
Skinner–Wiles used Hida theory for classical modular forms (and more gen-
erally, p-adic Hilbert modular forms) to prove an automorphy lifting the-
orem for residually reducible ordinary Galois representations of dimension
2. (Ordinary modular forms can be viewed as a precursor of overconvergent
modular forms of finite slope, of which they are a special case. They remain
useful since they generally have easy-to-understand integral structures.)
The starting point in [ANT20] for the proof of the above theorem is the
idea of applying the Skinner–Wiles strategy in Geraghty’s context of Hida
theory for definite unitary groups. We start with a diagram
Rord ← P ord → Tord ,
where Tord is the ‘big ordinary Hecke algebra’ (a finite Λ-algebra), P ord is a
pseudodeformation ring of tr ρ with an ‘ordinary of variable weight’ condi-
tion imposed at the p-adic places of M , and Rord is the corresponding Galois
deformation ring. (A pleasant feature of this situation is that although ρ is
reducible, Rord exists, because ρ is irreducible as a representation valued in
the Langlands dual group of the unitary group Gn – in other words, is Schur
in the sense of [CHT08]. This is in contrast to the situation considered in
[SW99], where one needs to consider one Galois deformation ring for every
possible indecomposable residual representation with the same pseudochar-
acter as ρ.) The gain in using modular forms of variable weight is that Tord
now has Krull dimension dim Λ = 1 + n[M + : Q], and one can find plentiful
‘generic’ prime ideals p ⊂ Tord such that the specialisation of the universal
pseudocharacter to Frac Tord /p is absolutely irreducible, and suitable for
application of a modified version of the Taylor–Wiles method. This makes
it possible to prove that Ppord → Tord p is an isomorphism.
Thinking geometrically, this implies that any irreducible component
of Spec P ord which contains p is in fact contained in the closed subspace
Spec Tord . The wonderful idea introduced in [SW99] is to use the fact that
for any two irreducible components Z0 , Z1 of Spec P ord , one can find a chain
W1 , W2 , . . . , Wd of irreducible components of Spec P ord such that all of the
intersections
Z0 ∩ W1 , W1 ∩ W2 , W2 ∩ W3 , . . . , Wd−1 ∩ Wd , Wd ∩ Z1
have a dimension which is ‘not too small’.33 Using that the dimension is not
too small, we can find generic prime ideals in each of these intersections, and
33The lower bound dimension of these intersections is the so-called ‘connectedness
dimension’ of the ring P ord and is more easily computed for the ring Rord . In the set-up
of this section, the relationship between Rord and P ord is rather simple (in fact, [ANT20,
Proposition 3.2] identifies P ord as the subring of invariants of Rord under a finite abelian
group). If we have a presentation Rord ∼ = OJX1 , . . . , Xg K/(f1 , . . . , fr ), then the connected-
ness dimension of Rord is at least g − r. In practice this can be made as large as desired
by first replacing the field M by a soluble CM extension.
52 J. A. THORNE
m ≥ 1 [NT21a, Lemma 3.4]. In general, this condition can fail (for example,
take the newform corresponding to an elliptic curve E and any refinement
at a prime p ≥ 5 at which E has good supersingular reduction).
The result that can be proved most naturally by induction using Theo-
rem 3.21 is the following one:
Proposition 3.22. Fix m ≥ 1. If π is a regular algebraic cuspidal auto-
morphic representation of GL2 (AQ ) without CM such that for every prime
number p such that πp is ramified, πp admits an accessible refinement χp
which is (m + 1)-regular, then Symm
∗ (π) exists.
35The needed conditions are: p > max(2(m + 2), mk), r (π) contains a conjugate
ι
of SL2 (Fp ), and πp is unramified. This implies in particular that for any isomorphism
ι : Qp → C, Symm rι (π) is potentially diagonalisable and we are in a position to apply the
very general automorphy lifting theorems for (m + 1)-dimensional Galois representations
established in the paper [BLGGT14].
56 J. A. THORNE
otherwise. In particular, the above theorem is valid for any prime p (in-
cluding the case p = 2!) and when p is small relative to m (in which case
the residual representation Symm rι (π) could have arbitrarily many Jordan–
Hölder factors). This is also reflected in the proof, where instead of perform-
ing Taylor–Wiles patching on a single (pseudo-)deformation ring, we patch
two (pseudo-)deformation rings (corresponding to deformations of rι (π) and
tr Symm rι (π) respectively) and the map between them induced by the rep-
resentation Symm : GL2 → GLm+1 .
3.24.1. Getting to the end. Before getting on to the proof of Theorem
3.24, let’s sketch how it leads to the completion of the proof of symmet-
ric power functoriality for regular algebraic cuspidal automorphic represen-
tations of GL2 (AQ ). Let π be a cuspidal automorphic representation of
GL2 (AQ ) of weight k and without CM. We induct on the number sc(π) of
primes l such that πl is supercuspidal, the base case sc(π) = 0 having been
already established.
The first reduction is that we can assume k = 2. This is familiar from
the proof of Serre’s conjecture: we can choose a prime p, large enough with
respect to π so that for any isomorphism ι : Qp → C, rι (π)(GQ ) contains a
conjugate of SL2 (Fp ), πp is unramified, and another cuspidal automorphic
representation π 0 of GL2 (AQ ) of weight 2 such that at primes l 6= p, πl0
and πl are unramified twists of each other, but such that πp0 is now ramified
(of principal series type, say). Then rι (π 0 ) will be potentially Barsotti–Tate
and so potentially diagonalisable, in the sense of [BLGGT14]. Here we are
using a result from [GK14], itself making use of Kisin’s description of the
irreducible components of the local lifting rings of 2-dimensional Barsotti–
Tate representations [Kis09c]. The automorphy lifting theorems established
m 0
in [BLGGT14] then show that Symm ∗ (π) exists if and only if Sym∗ (π ) does.
The next reduction is that, selecting a prime p such that πp is super-
cuspidal, we can assume that there is an isomorphism ι : Qp → C such
that the residual representation rι (π) satisfies the ‘large image’ hypothesis
(4) of Theorem 3.24. The idea that one can force large image (and in par-
ticular, non-dihedral image) of the mod p Galois representation by adding
in additional primes of ramification is important in the proof of [KW09a],
where ‘good dihedral primes’ are introduced that force the residual image
rι (π) to be large for all primes p up to a specified bound. Using a similar
idea, we can modify π so that it is ‘seasoned’36, in the sense of [NT21b,
Definition 3.6], and therefore has large residual image with respect to at
least one isomorphism ι : Qp → C for each prime number p such that πp is
supercuspidal.
M/F and let t = tr Symm ρ|GM , and write P ∈ CO for the object classify-
ing conjugate self-dual pseudodeformations of t which are semistable with
Hodge–Tate weights in the interval [0, m]; the same ring whose existence
we have exploited already in §3.15. We introduce the definite unitary group
Gm+1 over F and write HG for a space of algebraic modular forms on Gm+1
with integral structure, on which the ring P acts. We observe that there is
a tautological ring homomorphism
P → Rρ,Sp ,−1 ,
which classifies the natural transformation which associates to a deformation
ρ0 of ρ the pseudocharacter tr Symm ρ0 |GM . After introducing Taylor–Wiles
primes and carrying out a patching argument, we can construct, in addition
to the objects R∞ , HD,∞ considered above, an S∞ -algebra P∞ and a P∞ -
module HG,∞ with the following properties:
• There is an isomorphism P∞ ⊗S∞ O ∼ = P and a compatible isomor-
phism HG,∞ ⊗S∞ O ∼ H
= G .
• HG,∞ is finite free as S∞ -module.
• There is a ring homomorphism P∞ → R∞ making the diagram
P∞ / R∞
P /R
commute.
What we do not have is any control on the Krull dimension of the set of
irreducible components of P∞ : getting such information for the ring R∞
relies on being able to compute the tangent space in Galois cohomological
terms and to choose Taylor–Wiles primes which kill the dual Selmer group,
which is not possible for P∞ because of the potentially degenerate form of
the residual representation Symm ρ|GM .
As a substitute for this, we will use the vanishing of the adjoint Bloch–
Kato Selmer group Hf1 (F, ad Symm ρ). Let p, p0 ⊂ P denote the prime ideals
which are the kernels of the maps P → O associated to the two representa-
tions Symm rι (π)|GM and Symm rι (π 0 )|GM , and let p∞ , p0∞ ⊂ P∞ denote the
pullbacks of these prime ideals to P∞ . The vanishing of the adjoint Bloch–
Kato Selmer group implies that the Zariski tangent space of the ring Pp is 0,
and therefore that the Zariski tangent space of the ring P∞,p∞ has dimension
at most dim S∞ [1/p]. On the other hand, if we assume that Symm ∗ (π) exists
then p∞ is in the support of HG,∞ , which is a free S∞ -module, implying that
the Krull dimension of P∞,p∞ is at least dim S∞ [1/p]. It follows that P∞,(p∞ )
is a regular local ring and that there is a unique irreducible component Z
of Spec P∞ containing p∞ , which is necessarily in the support of HG,∞ as
P∞ -module.
We now consider the map Spec R∞ → Spec P∞ . The map P∞ → R∞
is finite, so this map has closed irreducible image of dimension dim S∞ and
RECIPROCITY AND FUNCTORIALITY 61
containing both the points p∞ , p0∞ . We conclude that this image equals Z
and that p0∞ is in the support of HG,∞ as P∞ -module. Passing back to P ,
we find that p0 is in the support of HG as P -module, or in other words that
Symm rι (π 0 )|GM is automorphic and Symm 0
∗ (π ) exists, as required.
3.25. The defects of our proof. We conclude this article with a re-
flection on the reach of our methods. There is no doubt that reciprocity is
an effective tool for studying Langlands functoriality for those automorphic
representations which have associated Galois representations. After the the-
orems discussed in this article, the next goal would be to establish symmetric
power functoriality for all automorphic representations associated to Hilbert
modular forms (in other words, regular algebraic cuspidal automorphic rep-
resentations of GL2 (AF ), without CM, where F is an arbitrary totally real
field), beyond the case F = Q. Many of our techniques apply equally well in
this case: in particular, the ‘analytic continuation of functoriality’ principle
on the eigenvariety, the vanishing of the adjoint Bloch–Kato Selmer group
for Galois representations associated to RACSDC automorphic representa-
tions, and the ‘functoriality lifting theorem’ discussed in the previous section
(which is in fact stated in [NT21b] over an arbitrary totally real base field).
However, there is one critical ingredient which is missing in general,
namely the beautifully simple structure of the 2-adic tame level 1 eigencurve.
Although the rough structure of the eigencurve at the boundary of weight
space may be expected to generalise (and this is even proved in many cases,
as in [LWX17]), it is not reasonable to expect such a clean statement to
hold over base fields other than Q. One can draw an analogy here with the
proof of Serre’s conjecture. As we have alluded to above, this may be proved
by reducing to the case of S-type representations ρ : GQ → GL2 (Fp ) which
are unramified outside p. The proof in this case in turn is by induction on the
prime p, beginning with Tate’s theorem [Tat94] that the conjecture holds
vacuously when p = 2 (because the set of S-type representations in this case
is empty).
Serre’s modularity conjecture is expected to hold equally over any to-
tally real number field (or even over any number field at all, in the correct
formulation) [BDJ10]. However, we cannot expect the above strategy to
work without modification because when the base number field F is large
enough, there are plentiful low weight S-type representations which are ram-
ified only at the p-adic places. (For example, Dembélé [Dem09] constructs
a surjective representation GF → SL2 (F28 ) with F = Q(ζ32 ) ∩ R).) There
is no base case for the induction!
In the case of symmetric power functoriality, we can point to two possible
ways out. The first is to try to follow the conjectural programme outlined
in [CT14], which aims to establish the existence of the symmetric power
lifting Symm ∗ (π) by induction on m. Holding us back is the hypothesis that
one already has access to sufficiently many cases of GL2 ×GLr → GL2r tensor
62 J. A. THORNE
References
[AC89] James Arthur and Laurent Clozel. Simple algebras, base change, and the
advanced theory of the trace formula, volume 120 of Annals of Mathematics
Studies. Princeton University Press, Princeton, NJ, 1989.
[ACC+ 18] Patrick B. Allen, Frank Calegari, Ana Caraiani, Toby Gee, David Helm, Bao
V. Le Hung, James Newton, Peter Scholze, Richard Taylor, and Jack A.
Thorne. Potential automorphy over CM fields, 2018.
[AIP15] Fabrizio Andreatta, Adrian Iovita, and Vincent Pilloni. p-adic families of
Siegel modular cuspforms. Ann. of Math. (2), 181(2):623–697, 2015.
[ANT20] Patrick B. Allen, James Newton, and Jack A. Thorne. Automorphy lift-
ing for residually reducible l-adic Galois representations, II. Compos. Math.,
156(11):2399–2422, 2020.
[Art24] E. Artin. Über eine neue art von L-Reihen. Abh. Math. Sem. Univ. Hamburg,
3(1):89–108, 1924.
[AS97] Avner Ash and Glenn Stevens. p-adic deformations of cohomology classes of
subgroups of GL(n, Z). volume 48, pages 1–30. 1997. Journées Arithmétiques
(Barcelona, 1995).
[AT20] Christos Anastassiades and Jack A. Thorne. Raising the level of automorphic
representations of GL2n of unitary type, 2020.
[BC09] Joël Bellaı̈che and Gaëtan Chenevier. Families of Galois representations and
Selmer groups. Astérisque, (324):xii+314, 2009.
[BCDT01] Christophe Breuil, Brian Conrad, Fred Diamond, and Richard Taylor. On the
modularity of elliptic curves over Q: wild 3-adic exercises. J. Amer. Math.
Soc., 14(4):843–939, 2001.
[BCGP21] George Boxer, Frank Calegari, Toby Gee, and Vincent Pilloni. Abelian sur-
faces over totally real fields are potentially modular. Publ. Math. Inst. Hautes
Études Sci., 134:153–501, 2021.
[BDJ10] Kevin Buzzard, Fred Diamond, and Frazer Jarvis. On Serre’s conjecture
for mod l Galois representations over totally real fields. Duke Math. J.,
155(1):105–161, 2010.
[BG14] Kevin Buzzard and Toby Gee. The conjectural connections between auto-
morphic representations and Galois representations. In Automorphic forms
and Galois representations. Vol. 1, volume 414 of London Math. Soc. Lecture
Note Ser., pages 135–187. Cambridge Univ. Press, Cambridge, 2014.
[BG16] Kevin Buzzard and Toby Gee. Slopes of modular forms. In Families of auto-
morphic forms and the trace formula, Simons Symp., pages 93–109. Springer,
[Cham], 2016.
[BHS17] Christophe Breuil, Eugen Hellmann, and Benjamin Schraen. Une in-
terprétation modulaire de la variété trianguline. Math. Ann., 367(3-4):1587–
1645, 2017.
[BHS19] Christophe Breuil, Eugen Hellmann, and Benjamin Schraen. A local model
for the trianguline variety and applications. Publ. Math. Inst. Hautes Études
Sci., 130:299–412, 2019.
[BJ79] A. Borel and H. Jacquet. Automorphic forms and automorphic representa-
tions. In Automorphic forms, representations and L-functions (Proc. Sympos.
RECIPROCITY AND FUNCTORIALITY 63
Pure Math., Oregon State Univ., Corvallis, Ore., 1977), Part 1, Proc. Sym-
pos. Pure Math., XXXIII, pages 189–207. Amer. Math. Soc., Providence,
R.I., 1979. With a supplement “On the notion of an automorphic represen-
tation” by R. P. Langlands.
[BK05] Kevin Buzzard and L. J. P. Kilford. The 2-adic eigencurve at the boundary
of weight space. Compos. Math., 141(3):605–619, 2005.
[BLGGT14] Thomas Barnet-Lamb, Toby Gee, David Geraghty, and Richard Taylor. Po-
tential automorphy and change of weight. Ann. of Math. (2), 179(2):501–609,
2014.
[BLGHT11] Tom Barnet-Lamb, David Geraghty, Michael Harris, and Richard Taylor. A
family of Calabi-Yau varieties and potential automorphy II. Publ. Res. Inst.
Math. Sci., 47(1):29–98, 2011.
[BP19] John Bergdall and Robert Pollack. Slopes of modular forms and the ghost
conjecture. Int. Math. Res. Not. IMRN, (4):1125–1144, 2019.
[Buz05] Kevin Buzzard. Questions about slopes of modular forms. Number 298, pages
1–15. 2005. Automorphic forms. I.
[Buz07] Kevin Buzzard. Eigenvarieties. In L-functions and Galois representations,
volume 320 of London Math. Soc. Lecture Note Ser., pages 59–120. Cambridge
Univ. Press, Cambridge, 2007.
[Cal21] Frank Calegari. Reciprocity in the Langlands program since Fermat’s Last
Theorem, 2021.
[Car94] Henri Carayol. Formes modulaires et représentations galoisiennes à valeurs
dans un anneau local complet. In p-adic monodromy and the Birch and
Swinnerton-Dyer conjecture (Boston, MA, 1991), volume 165 of Contemp.
Math., pages 213–237. Amer. Math. Soc., Providence, RI, 1994.
[CC98] F. Cherbonnier and P. Colmez. Représentations p-adiques surconvergentes.
Invent. Math., 133(3):581–611, 1998.
[Che04] Gaëtan Chenevier. Familles p-adiques de formes automorphes pour GLn . J.
Reine Angew. Math., 570:143–217, 2004.
[Che11] Gaëtan Chenevier. On the infinite fern of Galois representations of unitary
type. Ann. Sci. Éc. Norm. Supér. (4), 44(6):963–1019, 2011.
[Che14] Gaëtan Chenevier. The p-adic analytic space of pseudocharacters of a profi-
nite group and pseudorepresentations over arbitrary rings. In Automorphic
forms and Galois representations. Vol. 1, volume 414 of London Math. Soc.
Lecture Note Ser., pages 221–285. Cambridge Univ. Press, Cambridge, 2014.
[CHT08] Laurent Clozel, Michael Harris, and Richard Taylor. Automorphy for some
l-adic lifts of automorphic mod l Galois representations. Publ. Math. Inst.
Hautes Études Sci., (108):1–181, 2008. With Appendix A, summarizing un-
published work of Russ Mann, and Appendix B by Marie-France Vignéras.
[Clo90] Laurent Clozel. Motifs et formes automorphes: applications du principe de
fonctorialité. In Automorphic forms, Shimura varieties, and L-functions, Vol.
I (Ann Arbor, MI, 1988), volume 10 of Perspect. Math., pages 77–159. Aca-
demic Press, Boston, MA, 1990.
[Clo91] Laurent Clozel. Représentations galoisiennes associées aux représentations
automorphes autoduales de GL(n). Inst. Hautes Études Sci. Publ. Math.,
(73):97–145, 1991.
[CM98] R. Coleman and B. Mazur. The eigencurve. In Galois representations in arith-
metic algebraic geometry (Durham, 1996), volume 254 of London Math. Soc.
Lecture Note Ser., pages 1–113. Cambridge Univ. Press, Cambridge, 1998.
[Col08] Pierre Colmez. Représentations triangulines de dimension 2. Number 319,
pages 213–258. 2008. Représentations p-adiques de groupes p-adiques. I.
Représentations galoisiennes et (φ, Γ)-modules.
64 J. A. THORNE
[Con99] Brian Conrad. Irreducible components of rigid spaces. Ann. Inst. Fourier
(Grenoble), 49(2):473–541, 1999.
[CQ15] D. Choimet and H. Queffélec. Twelve landmarks of twentieth-century anal-
ysis. Cambridge University Press, New York, 2015. Illustrated by Michaël
Monerau, Translated from the 2009 French original by Danièle Gibbons and
Greg Gibbons, With a foreword by Gilles Godefroy.
[CT14] Laurent Clozel and Jack A. Thorne. Level-raising and symmetric power func-
toriality, I. Compos. Math., 150(5):729–748, 2014.
[CT15] Laurent Clozel and Jack A. Thorne. Level raising and symmetric power func-
toriality, II. Ann. of Math. (2), 181(1):303–359, 2015.
[CT17] Laurent Clozel and Jack A. Thorne. Level-raising and symmetric power func-
toriality, III. Duke Math. J., 166(2):325–402, 2017.
[Del71a] Pierre Deligne. Formes modulaires et représentations l-adiques. In Séminaire
Bourbaki. Vol. 1968/69: Exposés 347–363, volume 175 of Lecture Notes in
Math., pages Exp. No. 355, 139–172. Springer, Berlin, 1971.
[Del71b] Pierre Deligne. Travaux de Shimura. In Séminaire Bourbaki, 23ème année
(1970/71), Exp. No. 389, pages 123–165. Lecture Notes in Math., Vol. 244.
1971.
[Del74] Pierre Deligne. La conjecture de Weil. I. Inst. Hautes Études Sci. Publ. Math.,
(43):273–307, 1974.
[Dem09] Lassina Dembélé. A non-solvable Galois extension of Q ramified at 2 only.
C. R. Math. Acad. Sci. Paris, 347(3-4):111–116, 2009.
[dJ95] A. J. de Jong. Crystalline Dieudonné module theory via formal and rigid
geometry. Inst. Hautes Études Sci. Publ. Math., (82):5–96 (1996), 1995.
[dJ96] A. J. de Jong. Smoothness, semi-stability and alterations. Inst. Hautes Études
Sci. Publ. Math., (83):51–93, 1996.
[DL16] Hansheng Diao and Ruochuan Liu. The eigencurve is proper. Duke Math. J.,
165(7):1381–1395, 2016.
[DS74] Pierre Deligne and Jean-Pierre Serre. Formes modulaires de poids 1. Ann.
Sci. École Norm. Sup. (4), 7:507–530 (1975), 1974.
[Eme] Local-global compatibility in the p-adic Langlands programme for GL2 /Q.
[Eme06] Matthew Emerton. On the interpolation of systems of eigenvalues attached
to automorphic Hecke eigenforms. Invent. Math., 164(1):1–84, 2006.
[Eme18] Kathleen Emerson. Comparison of different definitions of pseudocharacter.
PhD thesis, Princeton University, November 2018.
[Eme21] Matthew Emerton. Langlands reciprocity: L-functions, automorphic forms,
and Diophantine equations. In The genesis of the Langlands Program, volume
467 of London Math. Soc. Lecture Note Ser., pages 301–386. Cambridge Univ.
Press, Cambridge, 2021.
[FM95] Jean-Marc Fontaine and Barry Mazur. Geometric Galois representations. In
Elliptic curves, modular forms, & Fermat’s last theorem (Hong Kong, 1993),
Ser. Number Theory, I, pages 41–78. Int. Press, Cambridge, MA, 1995.
[Fon82] Jean-Marc Fontaine. Sur certains types de représentations p-adiques du
groupe de Galois d’un corps local; construction d’un anneau de Barsotti-
Tate. Ann. of Math. (2), 115(3):529–577, 1982.
[Fon92] Jean-Marc Fontaine. Valeurs spéciales des fonctions L des motifs. Number
206, pages Exp. No. 751, 4, 205–249. 1992. Séminaire Bourbaki, Vol. 1991/92.
[Fon94a] Jean-Marc Fontaine. Le corps des périodes p-adiques. Number 223, pages 59–
111. 1994. With an appendix by Pierre Colmez, Périodes p-adiques (Bures-
sur-Yvette, 1988).
RECIPROCITY AND FUNCTORIALITY 65
[KPX14] Kiran S. Kedlaya, Jonathan Pottharst, and Liang Xiao. Cohomology of arith-
metic families of (ϕ, Γ)-modules. J. Amer. Math. Soc., 27(4):1043–1115, 2014.
[KW09a] Chandrashekhar Khare and Jean-Pierre Wintenberger. Serre’s modularity
conjecture. I. Invent. Math., 178(3):485–504, 2009.
[KW09b] Chandrashekhar Khare and Jean-Pierre Wintenberger. Serre’s modularity
conjecture. II. Invent. Math., 178(3):505–586, 2009.
[Lab11] J.-P. Labesse. Changement de base CM et séries discrètes. In On the stabi-
lization of the trace formula, volume 1 of Stab. Trace Formula Shimura Var.
Arith. Appl., pages 429–470. Int. Press, Somerville, MA, 2011.
[Laf18] Vincent Lafforgue. Chtoucas pour les groupes réductifs et paramétrisation de
Langlands globale. J. Amer. Math. Soc., 31(3):719–891, 2018.
[Lan70] R. P. Langlands. Problems in the theory of automorphic forms. In Lectures in
modern analysis and applications, III, pages 18–61. Lecture Notes in Math.,
Vol. 170. 1970.
[Lan77] R. P. Langlands. Shimura varieties and the Selberg trace formula. Canadian
J. Math., 29(6):1292–1299, 1977.
[Lan89] R. P. Langlands. On the classification of irreducible representations of real al-
gebraic groups. In Representation theory and harmonic analysis on semisim-
ple Lie groups, volume 31 of Math. Surveys Monogr., pages 101–170. Amer.
Math. Soc., Providence, RI, 1989.
[Lan94] Serge Lang. Algebraic number theory, volume 110 of Graduate Texts in Math-
ematics. Springer-Verlag, New York, second edition, 1994.
[Liu07] Tong Liu. Torsion p-adic Galois representations and a conjecture of Fontaine.
Ann. Sci. École Norm. Sup. (4), 40(4):633–674, 2007.
[Liu15] R. Liu. Triangulation of refined families. Comment. Math. Helv., 90(4):831–
904, 2015.
[LWX17] Ruochuan Liu, Daqing Wan, and Liang Xiao. The eigencurve over the bound-
ary of weight space. Duke Math. J., 166(9):1739–1787, 2017.
[Maz89] B. Mazur. Deforming Galois representations. In Galois groups over Q (Berke-
ley, CA, 1987), volume 16 of Math. Sci. Res. Inst. Publ., pages 385–437.
Springer, New York, 1989.
[Mor20] L. J. Mordell. On mr ramanujan’s empirical expansions of modular functions.
Proceedings of the Cambridge Philosophical Society, 19:117–124, 1920.
[New98] Donald J. Newman. Analytic number theory, volume 177 of Graduate Texts
in Mathematics. Springer-Verlag, New York, 1998.
[New22] James Newton. Modularity of Galois representations and Langlands functo-
riality, 2022.
[NT20] James Newton and Jack A. Thorne. Adjoint selmer groups of automorphic
Galois representations of unitary type, 2020.
[NT21a] James Newton and Jack A. Thorne. Symmetric power functoriality for holo-
morphic modular forms. Publ. Math. Inst. Hautes Études Sci., 134:1–116,
2021.
[NT21b] James Newton and Jack A. Thorne. Symmetric power functoriality for holo-
morphic modular forms, II. Publ. Math. Inst. Hautes Études Sci., 134:117–
152, 2021.
[Pan21] Lue Pan. The Fontaine-Mazur conjecture in the residually reducible case,
2021.
[PT15] Stefan Patrikis and Richard Taylor. Automorphy and irreducibility of some
l-adic representations. Compos. Math., 151(2):207–229, 2015.
[Ram16] S. Ramanujan. On certain arithmetical functions. Transactions of the Cam-
bridge Philosophical Society, 22(9):159–184, 1916.
RECIPROCITY AND FUNCTORIALITY 67
[Rie60] B. Riemann. Über die Anzahl der Primzahlen unter einer gegebenen größe.
Monatsberichte der Berliner Akademie 1859, pages 671–680, 1860.
[Sai03] Takeshi Saito. Weight spectral sequences and independence of l. J. Inst.
Math. Jussieu, 2(4):583–634, 2003.
[Ser68] Jean-Pierre Serre. Abelian l-adic representations and elliptic curves. W. A.
Benjamin, Inc., New York-Amsterdam, 1968. McGill University lecture notes
written with the collaboration of Willem Kuyk and John Labute.
[Ser69] Jean-Pierre Serre. Une interprétation des congruences relatives à la fonction
τ de Ramanujan. In Séminaire Delange-Pisot-Poitou: 1967/68, Théorie des
Nombres, Fasc. 1, Exp. 14, page 17. Secrétariat mathématique, Paris, 1969.
[Ser70] Jean-Pierre Serre. Facteurs locaux des fonctions zêta des varietés algébriques
(définitions et conjectures). In Séminaire Delange-Pisot-Poitou. 11e année:
1969/70. Théorie des nombres. Fasc. 1: Exposés 1 à 15; Fasc. 2: Exposés 16
à 24, page 15. Secrétariat Math., Paris, 1970.
[Ser72] Jean-Pierre Serre. Propriétés galoisiennes des points d’ordre fini des courbes
elliptiques. Invent. Math., 15(4):259–331, 1972.
[Ser87] Jean-Pierre Serre. Sur les représentations modulaires de degré 2 de
Gal(Q/Q). Duke Math. J., 54(1):179–230, 1987.
[Ser94] Jean-Pierre Serre. Propriétés conjecturales des groupes de Galois motiviques
et des représentations l-adiques. In Motives (Seattle, WA, 1991), volume 55
of Proc. Sympos. Pure Math., pages 377–400. Amer. Math. Soc., Providence,
RI, 1994.
[Ser97] Jean-Pierre Serre. Répartition asymptotique des valeurs propres de
l’opérateur de Hecke Tp . J. Amer. Math. Soc., 10(1):75–102, 1997.
[Ser12] Jean-Pierre Serre. Lectures on NX (p), volume 11 of Chapman & Hall/CRC
Research Notes in Mathematics. CRC Press, Boca Raton, FL, 2012.
[Ser16] Jean-Pierre Serre. Lettre á Armand Borel. In Frobenius distributions: Lang-
Trotter and Sato-Tate conjectures, volume 663 of Contemp. Math., pages 1–9.
Amer. Math. Soc., Providence, RI, 2016.
[Shi12] Sug Woo Shin. Automorphic Plancherel density theorem. Israel J. Math.,
192(1):83–120, 2012.
[Shi20] Sug Woo Shin. Construction of automorphic Galois representations: the self-
dual case, page 209–250. London Mathematical Society Lecture Note Series.
Cambridge University Press, 2020.
[SL96] P. Stevenhagen and H. W. Lenstra, Jr. Chebotarëv and his density theorem.
Math. Intelligencer, 18(2):26–37, 1996.
[Sor20] Claus M. Sorensen. A Patching Lemma, page 297–305. London Mathematical
Society Lecture Note Series. Cambridge University Press, 2020.
[SW99] C. M. Skinner and A. J. Wiles. Residually reducible representations and
modular forms. Inst. Hautes Études Sci. Publ. Math., (89):5–126 (2000), 1999.
[Tat65] John T. Tate. Algebraic cycles and poles of zeta functions. In Arithmetical
Algebraic Geometry (Proc. Conf. Purdue Univ., 1963), pages 93–110. Harper
& Row, New York, 1965.
[Tat67] J. T. Tate. p-divisible groups. In Proc. Conf. Local Fields (Driebergen, 1966),
pages 158–183. Springer, Berlin, 1967.
[Tat79] J. Tate. Number theoretic background. In Automorphic forms, representa-
tions and L-functions (Proc. Sympos. Pure Math., Oregon State Univ., Cor-
vallis, Ore., 1977), Part 2, Proc. Sympos. Pure Math., XXXIII, pages 3–26.
Amer. Math. Soc., Providence, R.I., 1979.
[Tat94] John Tate. The non-existence of certain Galois extensions of Q unramified
outside 2. In Arithmetic geometry (Tempe, AZ, 1993), volume 174 of Con-
temp. Math., pages 153–156. Amer. Math. Soc., Providence, RI, 1994.
68 J. A. THORNE
[Tay04] Richard Taylor. Galois representations. Ann. Fac. Sci. Toulouse Math. (6),
13(1):73–119, 2004.
[Tay08] Richard Taylor. Automorphy for some l-adic lifts of automorphic mod l Ga-
lois representations. II. Publ. Math. Inst. Hautes Études Sci., (108):183–239,
2008.
[Tho15] Jack A. Thorne. Automorphy lifting for residually reducible l-adic Galois
representations. J. Amer. Math. Soc., 28(3):785–870, 2015.
[Tho21a] Jack A. Thorne. On the vanishing of adjoint Bloch–Kato Selmer groups of
irreducible automorphic Galois representations, 2021.
[Tho21b] Jesse Thorner. Effective forms of the sato–tate conjecture. Research in the
Mathematical Sciences, 8(1), Jan 2021.
[Tho22] Jack A. Thorne. A p-adic approach to level-raising for GLn , 2022.
[Tsu98] Takeshi Tsuji. p-adic Hodge theory in the semi-stable reduction case. In Pro-
ceedings of the International Congress of Mathematicians, Vol. II (Berlin,
1998), number Extra Vol. II, pages 207–216, 1998.
[Tun81] Jerrold Tunnell. Artin’s conjecture for representations of octahedral type.
Bull. Amer. Math. Soc. (N.S.), 5(2):173–175, 1981.
[TW95] Richard Taylor and Andrew Wiles. Ring-theoretic properties of certain Hecke
algebras. Ann. of Math. (2), 141(3):553–572, 1995.
[Urb11] Eric Urban. Eigenvarieties for reductive groups. Ann. of Math. (2),
174(3):1685–1784, 2011.
[Wal88] Nolan R. Wallach. Real reductive groups. I, volume 132 of Pure and Applied
Mathematics. Academic Press, Inc., Boston, MA, 1988.
[WE20] Carl Wang-Erickson. Presentations of non-commutative deformation rings
via A∞ -algebras and applications to deformations of Galois representations
and pseudorepresentations, 2020.
[Wei74] André Weil. Basic number theory. Die Grundlehren der mathematischen Wis-
senschaften, Band 144. Springer-Verlag, New York-Berlin, third edition, 1974.
[Wil88] A. Wiles. On ordinary λ-adic representations associated to modular forms.
Invent. Math., 94(3):529–573, 1988.
[Wil95] Andrew Wiles. Modular elliptic curves and Fermat’s last theorem. Ann. of
Math. (2), 141(3):443–551, 1995.
[Wil06] Andrew Wiles. The Birch and Swinnerton-Dyer conjecture. In The millen-
nium prize problems, pages 31–41. Clay Math. Inst., Cambridge, MA, 2006.
[WWE19] Preston Wake and Carl Wang-Erickson. Deformation conditions for pseu-
dorepresentations. Forum Math. Sigma, 7:Paper No. e20, 44, 2019.