Current Developments in Mathematics

Reciprocity and symmetric power functoriality

Jack A. Thorne

Abstract. Symmetric power functoriality is the one of the first inter-

esting cases of the Langlands functoriality conjectures, which predict
the existence of liftings of automorphic forms from one reductive group
to another. These conjectures are closely tied, through the theory of L-
functions, to questions in number theory of independent interest, most
famously the Sato–Tate conjecture.
In this article we first give an introduction to these L-functions and
their connection with the Langlands programme, before giving a guide
to our proof, with James Newton, of symmetric power functoriality for
holomorphic modular forms.

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

3.12. Proof of the analytic continuation principle 40

3.15. Vanishing of the adjoint Bloch–Kato Selmer group 45
3.18. Seed points 48
3.20. The case of level N > 1 53
3.25. The defects of our proof 61
References 62

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

This function appears in the statement of the following theorem, proved by

Ramanujan in 1916 [Ram16]:
Theorem 1.1. If n ∈ N, then let
24
X
r24 (n) = {(x1 , . . . , x24 ) ∈ Z24 | x2i = n}
i=1

be the number of ways of writing n as a sum of 24 squares. Then for any

odd natural number n we have the formula
16 X 11 33152
r24 (n) = d + τ (n).
691 691
d|n

Ramanujan observed several properties of the function τ (n). For exam-

ple, there is the famous congruence
X
τ (n) ≡ d11 mod 691.
d|n

−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

• For any prime number p, factorise

(1 − τ (p)p−s + p11−2s ) = (1 − αp p−s )(1 − βp p−s )
for complex numbers αp , βp . Then we have

|αp | = |βp | = p11/2 .

 RECIPROCITY AND FUNCTORIALITY 3

Ramanujan’s conjectures have turned out to be connected to some of the

most beautiful aspects of algebraic number theory in the last century. In the
1960’s Serre observed that many of these connections could be explained by
the existence of a compatible family of p-adic Galois representations
ρ∆,p : Gal(Q/Q) → GL2 (Qp )
characterised by the requirement that the number τ (l) (for a prime number
l 6= p) appear as the trace, under ρ∆,p , of a Frobenius element of the Galois
group GQ = Gal(Q/Q) at the prime l. In the paper [Ser69], Serre associated
to ∆ the L-functions
m
Y Y
Lm (s) = (1 − αpm−i βpi p−s )−1
p prime i=0

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

liftings of all automorphic representations of GL2 (AQ ) associated to holo-

morphic Hecke eigenforms.) We divide the exposition into two parts.
In Part I we describe more fully some of the ideas leading up to the for-
mulation of the conjectures described in this introduction, starting with the
Riemann zeta function and the first examples (due to Dirichlet and Artin)
of L-functions associated to representations of Galois groups. The endpoint
is a formulation of the general reciprocity conjecture for p-adic Galois repre-
sentations which describes the relation they should have with automorphic
representations of general linear groups, and which helps to motivate our ap-
proach to Langlands functoriality, which applies only for those automorphic
representations which do have associated Galois representations. Our sketch
is far from complete, both from a mathematical perspective and a histori-
cal one: for a more detailed survey covering much of the same ground, see
[Eme21], while for a more careful approach to the reciprocity conjecture,
including in particular a precise definition of automorphic representation,
see [Tay04].
In Part II we get stuck into the details of our proof of symmetric power
functoriality for holomorphic newforms. Our approach to functoriality is
through the reciprocity Conjecture 2.22: we try to prove the automorphy of
the symmetric powers of the 2-dimensional Galois representations associated
to holomorphic newforms. As emphasised by Mazur [Maz89], p-adic Galois
representations often come in p-adic families, which should be reflected in
the existence of p-adic families of (necessarily non-classical) automorphic
forms. In fact, there are many different notions of p-adic automorphic forms,
corresponding to the different conditions one might impose on p-adic Galois
representations, that lead to larger or smaller families. A feature of our proof
is that we use several different notions of p-adic automorphic form, selecting
the most appropriate one for each step in the argument. For applications
to Ramanujan’s modular form ∆, the most important class of p-adic auto-
morphic forms is the class of finite slope overconvergent Hecke eigenforms,
which together form the Coleman–Mazur eigencurve. For another view on
the ideas appearing in our proof, see [New22].

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, . . .

and then throw away everything but the last digit:

2, 3, 5, 7, 1, 3, 7, 9, 3, 9, 1, 7, 1, 3, 7, 3, 9, 1, 7, 1, . . .

We can then take a census of all of these last remaining digits:

 RECIPROCITY AND FUNCTORIALITY 5

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

Euler product as a product of geometric series, and the fundamental theorem

of arithmetic implies that each term n−s (n ∈ N) appears exactly once.
The Euler product also allows us to give an analytic proof that there are
infinitely many primes. Indeed, ζ(s) diverges as s tends
P to 1 through values
in the interval (1, ∞) (because the harmonic series n≥1 n−1 diverges). On
the other hand, if p is a prime number, then (1 − p−s )−1 tends to a finite
limit as s → 1, so if there were only finitely many primes we would obtain
a contradiction.
More refined information about ζ(s) leads to more refined information
about the distribution of the prime numbers. For example, in order to prove
the prime number theorem, it is enough to know that ζ(s) admits a mero-
morphic continuation to C which is holomorphic and non-vanishing on the
line Re(s) = 1 (except for the point s = 1, where there is a simple pole). Let
us first explain how these properties lead to a proof of the prime number
theorem. One deduction proceeds by considering the logarithmic derivative
ζ 0 (s) X
=− Λ(n)n−s ,
ζ(s)
n≥1

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.

2.5. Dirichlet L-functions. Let us now come back to the proof of

Dirichlet’s theorem on primes in arithmetic progressions. Fix a modulus
N ∈ N and a base a ∈ N prime to N . Based on our experience with
the Riemann zeta function, we might hope to prove Dirichlet’s theorem by
analyzing the asymptotics of the function
X
(2.5) Λ(n),
n<X
n≡a mod N
8 J. A. THORNE

perhaps by considering the logarithmic derivative of the partial Euler prod-

uct
Y 1
.
1 − p−s
p prime
p≡a mod N
Analysing this function is difficult when at the start of the proof we do not
even know if the product is finite or infinite! It turns out to be more fruitful to
introduce the Dirichlet L-functions L(χ, s), associated to a homomorphism
χ : (Z/N Z)× → C× :
X Y
L(χ, s) = χ(n mod N )n−s = (1 − χ(p)p−s )−1 .
n≥1,(n,N )=1 p prime,p-N

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

This reduces an analysis of the function (2.5) to an analysis of the functions

X
(2.6) χ(p mod N ) log(p)
pk <X
p-N

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 can now explain the perspective of Artin L-functions: replace KN by

any Galois extension K/Q, and the character χ by any representation ρ :
Gal(K/Q) → GLn (C). We then want to define
Y
L(ρ, s) = Lp (ρ, p−s ),
p prime

where Lp (ρ, T ) is the corresponding Euler factor (a rational function in

T ). In order to define Lp (ρ, T ), we need to say what is the analogue of
the element σp ∈ Gal(KN /Q). This is the so-called Frobenius element: a
misnomer, since it is in fact a conjugacy class (or even a conjugacy class of
cosets) in Gal(K/Q). We first consider the case where the prime number p
is unramified in K (this is the case for all but finitely many prime numbers,
namely the primes which do not divide the discriminant of K). If OK denotes
the ring of integers of K, then there is a unique factorisation of ideals
pOK = p1 . . . pr ,
where the pi are distinct prime ideals of OK . Choose any prime ideal p
among these: then k(p) = OK /p is a finite extension of Z/pZ = Fp , and as
such the Galois group Gal(k(p)/Fp ) has a canonical generator, the Frobenius
automorphism φp : x 7→ xp . The Galois group Gal(K/Q) acts on the set of
prime factors pi (because it acts on OK by ring automorphisms) and there
is a natural map
StabGal(K/Q) (p) → Gal(k(p)/Fp ),
which is in fact an isomorphism – so we can lift the Frobenius φp uniquely
to an element σp ∈ Gal(K/Q). This element depends on the choice of prime
factor p, but Gal(K/Q) acts transitively on the factors pi and so the con-
jugacy class of σp ∈ Gal(K/Q) depends only on p, and we write σp for any
representative of this conjugacy class.
When p is ramified, there is a factorisation
pOK = (p1 . . . pr )e
for some integer e ≥ 2. There is still a surjective homomorphism
StabGal(K/Q) (p) → Gal(k(p)/Fp )
but now its kernel, the inertia group Ip , is non-trivial (in fact of order e).
The pre-image of φp is a coset of Ip whose conjugacy class in Gal(K/Q)
depends only on p.
10 J. A. THORNE

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

We may therefore reduce the proof of the Chebotarev density theorem to

the problem of showing that the Artin L-function L(ρ, s) of a non-trivial
irreducible representation ρ of G has a meromorphic continuation to C which
is holomorphic and non-vanishing on the line Re(s) = 1. This can be reduced
to the abelian case [Wei74, Ch. XIII, §12].
2.8. L-functions of p-adic representations. Artin L-functions are
only the beginning of the story. We can greatly expand the class of arith-
metic L-functions by enlarging the class of representations of Galois groups
under consideration. In order to do this, we first replace the Galois groups of
finite extensions of Q by the absolute Galois group (with respect to a fixed
algebraic closure Q/Q)
GQ = Gal(Q/Q) = lim Gal(K/Q),
←−
K⊂Q Galois

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

We can then recover the Hodge numbers hp,q

= dimQp H p (E, ΩqE ) using the
formulae
H 1 (E, O) ∼= HomQp (Vp E, Cp )GQp ,
H 0 (E, Ω1 ) ∼
= HomQ (Vp E, Cp (1))GQp .
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

How do we interpret this expression? First, we may think of it as a formal

Dirichlet series with coefficients in C, because the L-factors Ll (ρE,p , T ) are
rational functions of T with rational coefficients. Second, we may think of
it as defining a holomorphic function in the right half-plane Re(s) > 3/2.
Indeed, the Hasse–Weil theorem states that if we factorise the local L-factor
Ll (ρE,p , T ) at a prime of good reduction as
1 1
Ll (ρE,p , T ) = = ,
1 − al T + lT 2 (1 − αl T )(1 − βl T )
then the numbers αl , βl (which satisfy αl βl = l) both have absolute value
l1/2 . An elementary argument then shows that the formal Dirichlet series
defining L(E, s) converges absolutely when Re(s) > 3/2.
The L-function we have associated to the elliptic curve E is precisely
the usual Hasse–Weil L-function, which appears in the formulation of the
Birch–Swinnerton-Dyer conjecture [Wil06]. Some general remarks are now
in order. In order to make sense of L(ρE,p , s) as a function of a complex
variable s, we have relied on the fact that the local L-factors Ll (ρE,p , T )
are rational functions of T with integer coefficients (even though the repre-
sentation ρE,p is defined over Qp ). We have also made use of the fact that
ρE,p |GQp is potentially semi-stable, in order to be able to define the local
L-factor at p. These properties cannot be expected to hold for an arbitrary
p-adic representation GQ → GLn (Qp ) (for example a non-integer power of

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

• H ∗ (X, Qp ) is a potentially semi-stable Qp [GQp ]-module, in the

sense of Fontaine.
The last two bullet points rely on some of the most important advances in
arithmetic geometry of the 20th century. The purity of H w (X, Qp ) (which
generalises Hasse’s theorem for elliptic curves) is part of the Weil conjectures,
proved by Deligne [Del74], while the potential semi-stability of H ∗ (X, Qp )
follows from Tsuji’s proof of Fontaine’s conjecture Cst and de Jong’s the-
ory of alterations [Tsu98, dJ96]. The local L-factors associated to these
representations were first defined (in the case l 6= p) by Serre [Ser70], who
conjectured further that they have coefficients in Q. Fontaine made the
same conjecture also in the case l = p (with the further expectation that
Ll (H ∗ (X, Qp ), T ) is independent of p). It is known that Ll (H w (X, Qp ), T )
has rational coefficients whenever X has good reduction at l. This is also
known in some cases when X has bad reduction (including when w = 1, or
X is an abelian variety, or dim X ≤ 2 [Sai03]) but not in general. In order to
make unconditional statements, we can choose an isomorphism ι : Qp → C
and form the L-function L(ιH w (X, Qp ), s); this will converge in a right half-
plane, by purity. The conjectures in [Ser70, Fon94b] would imply that this
L-function is independent of the choice of ι.
More generally, we can notice that the representations H w (X, Qp ) may
be reducible, and decompose them into pieces. This leads to the following
definition.
Definition 2.11. Let K/Qp be a finite extension and let ρ : GQ →
GLn (K) be a continuous semi-simple representation. We say that ρ arises
from geometry if there is a smooth proper variety X over Q and integers
w, j such that ρ is isomorphic to a subquotient of H w (X, Qp ) ⊗Qp K(j).
It is reasonable to expect that if ρ arises from geometry then there is a
number field M such that the local L-factors Ll (ρ, T ) (including for l = p)
all have coefficients in M .
One special case that is worth remarking on is the case w = 0. Then
H 0 (X, Qp ) may be identified with Qp [π0 (X)], the permutation represen-
tation of GQ on the set of connected components of X. In particular,
the action of GQ factors through a finite quotient, and any representation
GQ → GLn (Qp ) with finite image appears as a submodule of H 0 (X, Qp )
for some X (which we may even take to be 0-dimensional). In this way,
Artin representations may be viewed as the ‘dimension 0’ case of p-adic
representations arising from geometry.
A very important perspective on the class of representations arising from
geometry is given by the Fontaine–Mazur conjecture [FM95], which aims
to give an internal classification of such representations. More precisely:
Conjecture 2.12. Let K/Qp be a finite extension, and let ρ : GQ →
GLn (K) be a continuous semi-simple representation. Then ρ is algebraic if
and only if it arises from geometry.
 RECIPROCITY AND FUNCTORIALITY 17

(Both Definition 2.10 and Definition 2.10 extend to representations of

GM , where M is a number field, and the Fontaine–Mazur conjecture makes
sense in this setting too.) Put another way, suppose given a continuous
semi-simple representation ρ : GQ → GLn (K) which is unramified at all but
finitely many prime numbers l. The Fontaine–Mazur conjecture asserts that
the necessary condition for ρ to arise from geometry, namely that ρ|GQp is
potentially semi-stable, is in fact also sufficient. This conjecture, and the
question of how to classify the Galois representations that satisfy weaken-
ings of this condition, have become very useful organizing principles in the
arithmetic part of the Langlands programme. We will return to these ideas
at the beginning of Part II of this article.

2.13. The Sato–Tate conjecture. We have motivated the impor-

tance of L-functions using their applications to results such as Dirichlet’s
theorem on primes in arithmetic progressions and the Chebotarev density
theorem. The L-functions of p-adic representations arising from geometry
can also conjecturally be applied this way. This was first suggested by Tate,
in connection with the p-adic representations of self-products E k of elliptic
curves over Q [Tat65], and by Serre, who gave a beautiful statement in the
generality of a pure motive over Q [Ser16].
Let us begin with the Sato–Tate conjecture. If E is an elliptic curve over
Q and l is a prime of good reduction then purity implies, as we have already
seen, that the integer

al = l + 1 − #E(Fl ) = αl + βl

is of absolute value |al | ≤ 2l1/2 (by the triangle inequality). Re-normalizing,

we obtain a real number al /2l1/2 ∈ [−1, 1], defined for all but finitely many
prime numbers l. The Sato–Tate conjecture concerns the distribution of
these numbers as l varies. To formulate it, we need to split into cases. The
first case is where E has complex multiplication, in the sense that End(E)
is strictly bigger than Z. In this case, End(E) ⊗Z Q = M is an imaginary
quadratic field, and one can show that the p-adic representation ρE,p of E
G
has the form IndGQ M
χ, for some p-adic character χ of GM = Gal(Q/M ).
In particular, there is an additional symmetry: if the prime l is inert in
M (as is the case for half of all prime numbers l, by Dirichlet’s theorem!)
then tr ρE,p (Frobl ) = al = 0. The remaining, generic, case is where E has
no complex multiplication, in which case Serre showed that the image of
ρE,p : GQ → GL2 (Qp ) is Zariski dense in GL2 (and even of finite index in
GL2 (Zp [Ser72]).
The Sato–Tate conjecture concerns the case where E has no complex
multiplication:

Conjecture 2.14. Let E be an elliptic curve over Q without complex

multiplication. Then the quantities al /2l1/2 are equidistributed in [−2, 2] with
18 J. A. THORNE

√
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 change of variable now recovers the formulation in Conjecture 2.14.

As observed by Serre, this representation-theoretic formulation of the
Sato–Tate conjecture also suggests a strategy for its proof. Choose, for each
prime l - ∆E , a representative tl ∈ SU2 (R) of the conjugacy class of elements
with characteristic polynomial 1 − al /l1/2 T + T 2 . In order to show that
Conjecture 2.14 holds for every continuous function f : [−1, 1] → R, it
is enough (by the Peter–Weyl theorem) to show that for each irreducible
representation R : SU2 (R) → GLn (C), we have
P
l<X,l-∆E (tr R)(tl )
Z
lim P = (tr R)(g) dt.
X→∞ l<X,l-∆E 1 g∈SU2 (R)

The right-hand side is computed by character orthogonality: it is 1 if R is

the trivial representation, and 0 otherwise. The left-hand side is also easily
seen to be 1 if R is the trivial representation, so we are led to the problem
of showing the identity
P
l<X,l-∆E (tr R)(tl )
lim P =0
X→∞ l<X,l-∆E 1
 RECIPROCITY AND FUNCTORIALITY 19

for every non-trivial representation R of SU2 (R). Just as in the proofs of

Dirichlet’s theorem and the Chebotarev density theorem, this identity would
follow if we could show that for each non-trivial representation R the L-
function
Y
L(E, R, s) = det(1 − R(tl )l−s )−1 ,
l-∆E

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.

2.15. Langlands L-functions. How do we actually go about prov-

ing the analytic properties of L-functions of p-adic representations needed
to prove the Sato–Tate conjecture and its generalisations? The preceding
discussion covers roughly the years 1924 – 1968 (the years spanning the
publication of Artin’s article [Art24] and Serre’s book [Ser68]). During
these years a parallel development was taking place, involving a different
class of L-functions which do not obviously arise from representations of
Galois groups. This eventually led to the definition, by Langlands, of fami-
lies of L-functions indexed by algebraic group representations, in remarkable
20 J. A. THORNE

likeness to the L-functions appearing in Serre’s criterion for the Sato–Tate

conjecture to hold.
To begin this side of the story, we recall again the definition of Ramanu-
jan’s modular form
∞
Y X
∆=q (1 − q n )24 = τ (n)q n .
n=1 n≥1
As described in the introduction to this article, Ramanujan made the fol-
lowing conjectures concerning the coefficients τ (n) of the q-expansion of ∆:
Conjecture 2.16. Let L(∆, s) = n≥1 τ (n)n−s . Then:
P

(1) There is an Euler product expansion L(∆, s) = l prime (1−τ (l)l−s +

Q

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

We first need to introduce the notion of automorphic representation of

the group GLm (AQ ). We first describe this group a little bit more. First,
the ring A∞ = 0l Ql of finite adeles is a restricted direct product, which
Q
Q
contains l Zl as an open subring. The ring of adeles is AQ = A∞
Q
Q0 Q × R.
The group GLm (A∞ Q Q) = l GL m (Ql ) may also be realised as a restricted
direct product, with l GLm (Zl ) as an open subgroup, and GLm (AQ ) =
GLm (A∞ Q ) × GLm (R).
An automorphic representation π of GLm (AQ ) is a tensor product π ∞ ⊗
π∞ , where:
• π ∞ is an irreducible C[GLm (A∞
Q )]-module which is smooth, in the
∞
sense that each vector v ∈ π has open stabilizer.
• π∞ is an irreducible admissible (Mn (C), On (R))-module (see e.g.
[Wal88, Ch. 3]).
• The tensor product π ∞ ⊗ π∞ is a subquotient of the space of auto-
morphic forms on GLm (AQ ) (a subspace of the space of all func-
tions on the quotient GLm (Q)\GLm (AQ ), see [BJ79]).
Within the space of automorphic forms lies the subspace of cuspidal auto-
morphic forms; an automorphic representation is said to be cuspidal if it is
isomorphic to a subquotient of the space of cuspidal automorphic forms.
If π = π ∞ ⊗π∞ is an automorphic representation of GLm (AQ ), then π ∞
itself admits a restricted tensor product decomposition π ∞ = ⊗0l πl , where
each πl is an irreducible smooth representation of GLm (Ql ) and all but
finitely many of these representations are unramified, in the sense that the
GL (Z )
space πl m l is non-zero. As an example of an automorphic representa-
tion, we consider the function φ∆ : GL2 (Q)\GL2 (AQ ) → C defined by the
formula
 
∞ ai + b
φ∆ (γg g∞ ) = ∆ (ci + d)−12 ,
ci + d
where
 
∞ a b
γ ∈ GL2 (Q), g ∈ GL2 (Z),
b and g∞ = ∈ GL2 (R)det>0 .
c d

(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

An important point is that these standard L-functions are known to ad-

mit a meromorphic continuation to C and to satisfy a functional equation,
which may be described explicitly. If m > 1 and π is cuspidal, then the
continuation is even holomorphic everywhere. For example, the automor-
phic L-function
P L(π∆ , s), which turns out to equal Ramanujan’s L-function
L(∆, s) = n≥1 τ (n)n−s , admits a holomorphic continuation to C and sat-
isfies the functional equation Λ(∆, s) = Λ(∆, 12 − s), where

Λ(∆, s) = (2π)−s Γ(s)L(∆, s).

This may be proved by using the expression

Z ∞
dy
Λ(∆, s) = ∆(iy)y s
y=0 y

analogous to Riemann’s expression for ζ(s) as an integral transform of the

Jacobi theta function. Godement–Jacquet used analogous ideas to prove the
continuation and functional equation of L(π, s) in general [GJ72].
To go beyond the standard L-function, suppose given an algebraic rep-
resentation R : GLm → GLn and define
Y
L(π, R, s) = Ll (R ◦ recQl (πl ), l−s )−1 .
l prime
 RECIPROCITY AND FUNCTORIALITY 23

Langlands’s functoriality conjecture implies that these L-functions are the

standard L-functions of automorphic representations of higher rank general
linear groups. More precisely:
Conjecture 2.17. There exists an automorphic representation Π =
R∗ (π) of GLn (AQ ) such that for each prime number l, recQl (Πl ) = R ◦
recQl (πl ). 9
We see that, essentially by definition, we would have the equality
L(π, R, s) = L(R∗ (π), s),
showing why functoriality would lead to the analytic continuation of the
non-standard L-functions L(π, R, s). In particular, taking π = π∆ to be the
automorphic representation of GL2 (AQ ) associated to the Ramanujan ∆
function, we see that the analytic continuation of the L-functions L(∆, R, s)
is implied by the following theorem, namely the existence of the Langlands
functorial lift of π∆ along the symmetric powers of the standard represen-
tation of GL2 :
Theorem 2.18. For each m ≥ 1, there exists a cuspidal automor-
phic representation Π of GLm+1 (AQ ) such that for every prime number
l, recQl (Πl ) = Symm ◦recQl (π∆,l ).
This is a special case of the main theorem of [NT21a]. The most general
statement we prove may be stated somewhat informally as follows:
Theorem 2.19. Let f be a holomorphic newform of weight k ≥ 2 and
let πf be the associated cuspidal automorphic representation of GL2 (AQ ).
Suppose that f does not have CM (equivalently, there is no non-trivial Hecke
character χ : Q× \A× × ∼
Q → C such that π = π ⊗ (χ ◦ det)). Then for each
m ≥ 1, there exists a cuspidal automorphic representation Π of GLm+1 (AQ )
such that for every prime number l, recQl (Πl ) = Symm ◦recQl (πl ).
(The theorem is true, and much easier, also in the case where πf does
have CM, except the symmetric power lifts will no longer be cuspidal in this
case.)
We end this section with some brief remarks concerning the general case
of Langlands’s functoriality conjectures, which concern liftings of automor-
phic representations associated to L-homomorphisms
L
H → LG
of L-groups of reductive groups H, G over a fixed number field M . As Lang-
lands has emphasised, even the case where H is trivial and G = GLn is
interesting; in this case an L-homomorphism is essentially the data of a
9One can also specify what Π should be after introducing the local Langlands corre-
∞
spondence recR for GLm (R) [Lan89], which classifies the irreducible admissible represen-
tations of GLm (R) in terms of continuous semisimple representations of the Weil group
WR = C× t C× , where 2 = −1 and z−1 = z for z ∈ C× .
24 J. A. THORNE

continuous representation ρ : GQ → GLn (C), and functoriality is equiva-

lent to the strong Artin conjecture, which states that the Artin L-function
L(ρ, s) is in fact the standard L-function of an automorphic representation
of GLn (AQ ). This basic case of Langlands’s functoriality conjecture remains
open (although some cases are known). The arguments in the second part
of this article make constant use of known cases of functoriality for this
broader class of L-homomorphisms (going beyond algebraic representations
GLm → GLn ).

2.20. Reciprocity and the proof of the Sato–Tate conjecture.

We have now seen two situations where families of symmetric power L-
functions arise:
• Starting with an elliptic curve E over Q (say without CM), with
associated p-adic representation ρE,p : GQ → GL2 (Qp ), the sym-
metric power L-functions
L(Symm ◦ρE,p , s).
The holomorphic continuation of these L-functions, non-vanishing
on the line Re(s) = 1+m/2, would imply the Sato–Tate conjecture.
• Starting with a holomorphic newform f with associated cuspidal
automorphic representation π = πf of GL2 (AQ ) (say without CM),
the symmetric power L-functions
L(π, Symm , s).
Langlands functoriality would imply that they admit a holomorphic
continuation to the entire complex plane.
In fact, one of these classes of L-functions contains the other! If f is a
newform of weight 2 with rational coefficients, then there is an associated
elliptic curve Ef such that
L(Symm ◦ρE,p , s) = L(π, Symm , s)
for each m ≥ 1. This is the construction of Eichler and Shimura. More
generally, Deligne [Del71a] showed that if f is a newform of weight k ≥ 2
then for any isomorphism ι : Qp → C there exists an associated p-adic
representation rι (f ) : GQ → GL2 (Qp ), and in fact it is the case that
L(ι Symm ◦rι (f ), s) = L(πf , Symm , s)
for each m ≥ 1. We see that one path to the Sato–Tate conjecture is to
follow the two steps:
• Show that if E is an elliptic curve over Q, then there is a newform
f such that L(πf , s) = L(E, s). This is the modularity conjecture
for elliptic curves over Q.
• Establish the existence of the symmetric power lifts Symm ∗ (πf ), as
predicted by Langlands.
 RECIPROCITY AND FUNCTORIALITY 25

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

Definition 2.21. Let π be an automorphic representation of GLm (AQ ).

We say that π is algebraic if recR (π∞ | det |(1−m)/2) )|C× is a sum of charac-
ters of the form z 7→ z a z b where a, b are integers.

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

A similar definition applies to automorphic representations of GLn (AM )

for any number field M , making it possible to formulate the following con-
jecture:
Conjecture 2.22 (Reciprocity conjecture). Let M be a number field,
let p be a prime number, and let ι : Qp → C be an isomorphism. Then there
is a unique bijection between the following two sets of objects:
(1) The set of irreducible algebraic representations ρ : GM → GLm (Qp ),
up to isomorphism.
(2) The set of algebraic cuspidal automorphic representations π of GLm (AM ).
with the following property: if ρ and π correspond, then for each finite place
v of M , ι−1 recMv (πv | det |(1−m)/2 ) ∼
= WD(ρ|GMv ).12 In particular, if ρ and
π correspond then L(ιρ, s) = L(π, s − m−1 2 ).

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

12Here WD denotes the Weil–Deligne representation (see footnote 5) associated to

ρ|GMv using the recipe described in [Tat79]. Since ι is not continuous, Weil–Deligne
representations are useful here as a way to describe Galois representations in a way in-
dependent of the topology of the base field. If ρ|GMv is unramified, then we simply have
WD(ρ|GMv ) = ρ|WMv .
 RECIPROCITY AND FUNCTORIALITY 27

exists a finite Galois extension M/Q (perhaps depending on m) such that

the Galois representation Symm ρE,p |GM is automorphic. Using Brauer’s in-
duction theorem, one can express the L-function L(ι Symm ρE,p , s) as a ra-
tio of automorphic L-functions associated to automorphic representations
of GLm+1 (AMi ), for some intermediate fields M/Mi /Q. Using the known
analytic continuation of these automorphic L-functions, one obtains the
meromorphic continuation of L(ι Symm ρE,p , s), and crucially the required
properties (holomorphy and non-vanishing) on the line Re(s) = 1 + m/2.
However, one does not in this way have any control over the poles in the
region Re(s) < 1 + m/2.
2.23. Evidence for the reciprocity conjecture. Historically the
reciprocity conjecture has been attacked in two stages: first, show that any
automorphic representation π in a given class admits an associated p-adic
representation rι (π). Then, try to show that the induced map π 7→ rι (π)
is surjective. The greatest progress so far has been for the class of regular
algebraic objects.
Definition 2.24. Let K/Qp be a finite extension. An algebraic repre-
sentation ρ : GQ → GLn (K) is said to be regular algebraic if the multiset of
Hodge–Tate numbers of ρ|GQp is in fact a set, i.e. is multiplicity-free.
This condition holds for the p-adic representations associated to elliptic
curves, but not for the p-adic representations associated to H 1 (A, Qp ) for
abelian varieties A of dimension g > 1 (in which case the multiset of Hodge–
Tate numbers has elements 0, 1, each appearing with multiplicity g).
Definition 2.25. An algebraic automorphic representation π of GLn (AQ )
is said to be regular algebraic if the restriction recR (π∞ | det |(1−n)/2 )|C× is
multiplicity-free.
A refined version of the reciprocity conjecture would describe the ex-
pected relation between the Hodge–Tate numbers of ρ and the composition
factors of recR (π∞ | det |(1−n)/2 )|C× , explaining why these two definitions
should match up. One can generalise these definitions to a general base
number field M . If ρ : GM → GLn (Qp ) is an algebraic representation, then
there is one associated multiset Hτ (ρ) of Hodge–Tate numbers for every em-
bedding τ : M → Qp . If ι : Qp → C is an isomorphism, then this set would
in turn be related to the Langlands parameter of πv , where v is the infinite
place of M corresponding to the embedding ιτ : M → C.
Let us make this more explicit in the case n = 2. The regular algebraic
cuspidal automorphic representations of GL2 (AQ ) are precisely the ones as-
sociated to holomorphic newforms of weight k ≥ 2 (see [Clo90, §1.2.3]).
As mentioned above, Deligne showed [Del71a] how to construct the p-adic
representations associated to such newforms inside the étale cohomology of
a p-adic local system on a modular curve, by decomposing this cohomology
under the action of the adele group GL2 (A∞ Q ). The first major success in
28 J. A. THORNE

showing a broad class of regular algebraic p-adic representations are auto-

morphic came in the works [Wil95, TW95] on the modularity conjecture
for elliptic curves: they proved the first automorphy lifting theorems. Since
the idea of automorphy lifting theorems will be very important in the second
part of this article, we expand on this here.
If M is a number field and ρ : GM → GLm (Qp ) is a continuous rep-
resentation, then we can find (using the compactness of GM ) a finite ex-
tension K/Qp such that, after replacing ρ by a conjugate, ρ takes values in
GLm (OK ) (here OK is the ring of integers of K). Pushing forward along
the homomorphism OK → Fp , we obtain a representation GM → GLm (Fp ).
This representation might depend on the choice of conjugate of ρ valued
in GLm (OK ), but its semisimplification is independent, up to isomorphism,
of any choices, and we write ρ : GM → GLm (Fp ) for this semisimplified
residual representation. The archetypal automorphy lifting theorem takes
the following form:
Theorem 2.26 (Ideal). Let M be a number field, and let ρ, ρ0 : GM →
GLm (Qp ) be irreducible regular algebraic Galois representations. Suppose
that the following conditions are satisfied:
(1) ρ0 is automorphic.
(2) There is an isomorphism ρ ∼ = ρ0 .
Then ρ is automorphic.
Unconditional theorems of this type, proved under varying additional
technical conditions on ρ and ρ0 , are the main tool we have in proving the
automorphy of Galois representations. The first automorphy lifting theorems
in the case M = Q, m = 2 included the requirement that ρ be irreducible
and ρ, ρ0 have Hodge–Tate numbers 0, 1 (a necessary condition to arise from
a newform of weight 2). The automorphy of the residual representation ρ was
verified, in the case that ρ arises from an elliptic curve, using the fact that
the image is constrained to lie in the soluble group GL2 (F3 ) – in fact this was
deduced from a known case of the Artin conjecture [Tun81]. The difficulty
in verifying this residual automorphy hypothesis in general is one of the
main challenges in successfully applying automorphy lifting theorems and is
the reason that the proof of the Sato–Tate conjecture given in [BLGHT11]
goes by the route of establishing only potential automorphy of the symmetric
power Galois representations.
Two advances have led to very strong results towards the surjectiv-
ity part of the reciprocity conjecture for regular algebraic representations
of GL2 (AQ ). First, the proof of Serre’s modularity conjecture by Khare–
Wintenberger [KW09a, KW09b] implies that any irreducible residual rep-
resentation ρ : GQ → GL2 (Fp )13 is automorphic. Second, powerful auto-
morphy lifting theorems have been proved by Kisin [Kis09a] and Emerton
13Satisfying the necessary condition to arise from a regular algebraic automorphic
representation, namely oddness.
 RECIPROCITY AND FUNCTORIALITY 29

([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

Our strategy, as suggested by the reciprocity conjecture, will be to prove

the automorphy of the Galois representations Symm rι (π) (for regular alge-
braic cuspidal automorphic representations π of GL2 (AQ ), and for some
choice of prime number p and isomorphism ι : Qp → C). The main tools
will be deformation theory and the p-adic variation of p-adic Galois repre-
sentations. We now introduce some of the basic objects in this theory.
We fix a finite extension K/Qp with ring of integers O, uniformizer
$, and residue field O/($) = k. Suppose given a number field M and a
continuous representation ρ : GM → GLm (k). Let CO denote the category of
complete Noetherian local O-algebras A with residue field k (i.e. equipped
with a local homomorphism O → A inducing an isomorphism O/$O = k ∼ =
A/mA ).
Definition 3.2. A lifting of ρ to an object A ∈ CO is a homomorphism
ρA : GM → GLm (A) such that the composite map
GM → GLm (A) → GLm (k)
induced by reduction modulo mA equals ρ. A deformation of ρ is a ker(GLm (A) →
GLm (k))-conjugacy class of liftings.
Mazur [Maz89] initiated Galois deformation theory by defining defor-
mation functors of Galois representations and studying them using Galois
cohomology. For example, suppose we fix a finite set S of places of M , in-
cluding the p-adic places of M , and let
Def ρ,S : CO → Sets
denote the functor which associates to any A ∈ CO the set of deformations of
ρ to A which are unramified outside S. Mazur showed that if ρ is absolutely
irreducible then the functor Def ρ,S is representable. The spectrum Spec Rρ,S
of the representing object (or perhaps, its rigid generic fibre in the sense of
Berthelot [dJ95, §7.1]) then deserves to be called the universal deformation
space of ρ. Mazur also introduced a tangent-obstruction theory for Def ρ,S :
there is a canonical isomorphism15
H 1 (MS /M, ad ρ) ∼
= Homk (mR /($, m2 ), k),
ρ,S Rρ,S
and using Tate’s Euler characteristic formula in Galois cohomology one can
show that there is a presentation
(3.1) Rρ,S ∼
= OJX1 , . . . , Xg K/(f1 , . . . , fr ),
where g − r = [M : Q]n2 + 1 − v|∞ dimk H 0 (Mv , ad ρ). Here we use the
P

traditional notation ad ρ for the composition of ρ : GM → GLn (k) with the

adjoint representation GLn (k) → Endk (Mn (k)). This gives in particular a
lower bound on the Krull dimension of the ring Rρ,S , which in many cases
15Here the H 1 is the continuous group cohomology of Gal(M /M ), the profinite
S
Galois group of the maximal subextension MS of M /M which is unramified outside S. We
use similar notation where MS /M is replaced by Mv for a place v of M (and Gal(MS /M )
by the absolute Galois group GMv .
32 J. A. THORNE

one might expect to be an equality. For example when n = 2, M = Q, and

ρ is odd, we obtain the formula (cf. [Maz89, Corollary 3])
dim Rρ,S ≥ 4.
With refinements, Mazur’s deformation theory has become a very effective
tool for proving the automorphy of p-adic Galois representations. Let us
describe two such approaches that have been applied in the case m = 2,
M = Q, and under the assumption that ρ is irreducible (so that Rρ,S exists).
In the first, which is the approach of [Wil95], one first cuts down Rρ,S by
imposing conditions from p-adic Hodge theory. For example, if ρ = ρE,p for
an elliptic curve E with good reduction at an odd prime p, then one can
fl
define a quotient Rρ,S of Rρ,S , which represents the functor of deformations
of ρ such that ρ|GQp comes from a finite flat group scheme over Zp . This
property is in particular satisfied for the lifts GQ → GL2 (Zp ) of ρ which
come from elliptic curves with good reduction at p, or more generally those
lifts GQ → GL2 (Zp ) of ρ arising from newforms of weight 2 and level prime
to p. We are led to consider a homomorphism
Rρ,S → T(H 1 (Γ(NS ), O))m ,
where H 1 (Γ(NS ), O) is the group cohomology of the congruence subgroup
Γ1 (NS ) ≤ SL2 (Z) of level NS depending on S with O-coefficients, T(H 1 (Γ(NS ), O))
is the subalgebra of EndO (H 1 (Γ(NS ), O)) generated by the Hecke operators
Tl for prime numbers l 6∈ S, and m ⊂ T(H 1 (Γ(NS ), O)) is the maximal ideal
which is the kernel of the homomorphism T(H 1 (Γ(NS ), O)) → k which sends
Tl to tr ρ(Frobl ). The existence of this maximal ideal is equivalent to the as-
sertion that ρ is the residual representation attached to an automorphic
representation π which contributes to H 1 (Γ(NS ), O). The homomorphism
from Rρ,S to the integral Hecke algebra can be constructed, using Carayol’s
lemma [Car94], by gluing together the Qp -representations rι (π) for those
automorphic representations π which contribute to H 1 (Γ(NS ), O)m . Local-
global compatibility at the prime p for the representations rι (π) implies that
this homomorphism factors through a surjective homomorphism
fl
Rρ,S → T(H 1 (Γ1 (NS ), O))m ,
and to prove an automorphy lifting theorem here one wants to prove that this
map is an isomorphism (or at least induces a bijection on Qp -points). The
ring T(H 1 (Γ(NS ), O))m is a finite flat O-algebra; a computation analogous
fl
to the one leading to the presentation (3.1) implies that the ring Rρ,S has
‘expected dimension’ 1, but we know rather little about it a priori. Several
tools were introduced in the papers [Wil95, TW95] in order to prove that
such maps are isomorphisms, but the one that has had the most lasting
impact is the Taylor–Wiles method. The basic idea of the Taylor–Wiles
method is to start with a presentation
fl
OJX1 , . . . , Xg+1 K/(f1 , . . . , fg )  Rρ,S
 RECIPROCITY AND FUNCTORIALITY 33

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

can be made to fit into a diagram

OJX1 , . . . , Xg+1 K / / Rf l / T(H 1 (Γ1 (NS∪Q ), O))m

ρ,S∪Q

 
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

of modular curves. We define

He 1 (Γ1 (NS ), O) = lim lim H 1 (Γ1 (NS pr ), O/$s O).
←− −→
s r
Then one can show that there is a surjective homomorphism
Rρ,S → T(H e 1 (Γ1 (NS ), O))m .
The idea now is to use softer techniques to show that this map is an iso-
morphism (for example, if we know in advance that Rρ,S ∼ = OJX1 , X2 , X3 K
1
and that T(H (Γ1 (NS ), O))m has Krull dimension at least 4, then the map
e
is forced to be an isomorphism for dimension reasons). If we are given a ho-
momorphism Rρ,S → O associated to an algebraic deformation ρ of ρ, then
it comes from a homomorphism T(H e 1 (Γ1 (NS ), O))m → O, and one might
hope to be able to prove a classicality result asserting that the algebraicity16
of ρ implies that this homomorphism factors through the homomorphism
T(He 1 (Γ1 (NS ), O))m → T(H 1 (Γ1 (NS ), O))m
to the classical Hecke algebra, and therefore corresponds to the Hecke eigen-
values of an algebraic automorphic representation. This is exactly what
Emerton does, using the p-adic local Langlands correspondence for GL2 (Qp ).
One could object that this approach also relies on the irreducibility of
ρ, in order to have the existence of Rρ,S . However, we can circumvent this
difficulty using the theory of pseudocharacters or determinants, which are
a substitute for representations with many good properties. They were first
applied in number theory by Wiles in order to construct the p-adic Galois
representations associated to p-ordinary Hilbert modular forms [Wil88]. A
theory that works well even in small characteristics was given by Chenevier
using results in geometric invariant theory [Che14]. A further generalisa-
tion of the notion of pseudocharacter, in which GLn may be replaced by
an arbitrary reductive group G, was given by V. Lafforgue in his work on
the Langlands correspondence over global fields of positive characteristic
[Laf18]. Emerson showed that when G = GLn , this notion is equivalent
to Chenevier’s notion [Eme18]. Let us recall Lafforgue’s definition of pseu-
docharacter for GLn here:
Definition 3.3. Let Γ be a group and let A be a ring. A pseudocharacter
of Γ of dimension n over a ring A is a collection Θ = (Θk )k≥1 of algebra
homomorphisms Θk : Z[GLkn ]GLn → Map(Γk , A) satisfying the following
conditions:
(1) For all k, l ≥ 1 and for each map ζ : {1, . . . , k} → {1, . . . , l}, each
f ∈ Z[GLkn ]GLn , and for each γ1 , . . . , γl ∈ Γ, we have
Θl (f ζ )(γ1 , . . . , γl ) = Θk (f )(γζ(1) , . . . , γζ(k) ),
where f ζ ∈ Z[GLln ]GLn is defined by f ζ (g1 , . . . , gl ) = f (gζ(1) , . . . , gζ(k) ).
16In other words, the potential semi-stability of ρ|
GQp
 RECIPROCITY AND FUNCTORIALITY 35

(2) For each k ≥ 1, for each γ1 , . . . , γk+1 ∈ Γ, and for each f ∈

Z[GLkn ]GLn , we have
Θk+1 (fˆ)(γ1 , . . . , γk+1 ) = Θk (f )(γ1 , . . . , γk−1 , γk γk+1 ,
where fˆ ∈ Z[GLk+1
n ]
GLn is defined by fˆ(g , . . . , g
1 k+1 ) = f (g1 , . . . , gk−1 , gk gk+1 ).
If Γ is a profinite group and A is a topological ring, we say that Θ is con-
tinuous if for each k ≥ 1, Θk takes values in Mapcts (Γk , A).
It follows from the definition that if ρ : Γ → GLn (A) is a (continu-
ous) representation, then tr ρ = (Θk )k≥1 ,17 defined by Θk (f )(γ1 , . . . , γk ) =
f (ρ(γ1 ), . . . , ρ(γk ))), is a pseudocharacter. Pseudocharacters have good func-
torial properties, and the following theorem gives some evidence as to why
they are a good proxy for representations:
Theorem 3.4. Let Γ be a profinite group and let F = Fp or Qp . Then
the map ρ 7→ tr ρ sets up a bijection between the set of conjugacy classes
of semisimple continuous representations ρ : Γ → GLn (F ) and the set of
continuous pseudocharacters Θ of Γ over F of rank n.
Coming back to the situation of a number field M equipped with a finite
set of finite places S, we can consider the problem of deforming a continuous
pseudocharacter t of GM,S over k. More precisely, we can define a functor
PDef t,S : CO → Sets
by assigning to A ∈ CO the set of pseudocharacters tA of GM,S over A
such that tA mod mA equals t. (‘PDef’ stands for ‘pseudodeformation’, i.e.
a deformation of a pseudocharacter.)
Theorem 3.5. (1) The functor PDef t,S is representable. We write
Pt,S ∈ CO for the representing object.
(2) Suppose that t = tr ρ for a continuous representation ρ : GM,S →
GLn (k). Then the map ρA 7→ tr ρA determines a natural transfor-
mation Def ρ,S → PDef t,S . If ρ is absolutely irreducible, then this is
a natural isomorphism.
When ρ is not absolutely irreducible, the ring Pt,S can thus stand in for
the universal deformation ring, although it is harder to get a handle on that
Rρ,S . For example, the paper [Che14] gives only a weak upper bound for the
dimension of the Zariski tangent space of Pt,S . Wang-Erickson [WE20] has
given a refined tangent-obstruction theory for residually multiplicity-free for
Pt,S , but it remains to be seen whether this theory will have applications to
e.g. proving automorphy lifting theorems.
Nevertheless, we can describe Pt,S in some situations. For example, sup-
pose again that M = Q, p = 2, and S = {2}, and that t is the trivial
17This is an abuse of notation, since tr ρ already denotes the usual character of the
representation ρ (which is a function Γ → A). Note however that Θ1 is equivalent to the
data, for each γ ∈ Γ, of the characteristic polynomial of ρ(γ), and that Θ1 determines Θk
for each k ≥ 1.
36 J. A. THORNE

pseudocharacter of rank 2 (thus associated to the reduction modulo 2 of

the 2-adic Galois representation attached to Ramanujan’s modular form
∆). In this case, Chenevier [Che14, Theorem 5.1] has shown that the quo-
tient of Pt,S corresponding to odd pseudodeformations of t is isomorphic
to OJX1 , X2 , X3 K – a beautifully simple form for the universal pseudode-
formation ring that might lead us to hope we could apply an argument in
the style of [Eme] to the problem of symmetric power functoriality for the
modular form ∆. Here however there is a critical missing ingredient – cur-
rently the p-adic Langlands correspondence is known to exist only for the
group GL2 (Qp ), and not for higher rank groups, and in particular there are
no known classicality theorems for the completed cohomology of GLn when
n > 2.
We are thus led, Goldilocks-like, to look for a theory of p-adic automor-
phic forms that admits a non-trivial geometry (unlike the case of classical
automorphic forms, for which the spectrum of the Hecke algebra is a fi-
nite set of points) and which admits useful classicality results (so we can
show that p-adic automorphic forms with associated Galois representations
which are algebraic are in fact classical, which will be a necessary step if we
are to establish functoriality). The theory of overconvergent p-adic modular
forms of finite slope has these properties, and the geometry is that of the
Coleman–Mazur eigencurve Ep , to which we now turn.
3.6. The eigencurve. The eigencurve Ep is a p-adic rigid analytic
space (in the sense of Tate) which can be thought of as a moduli space for
families of newforms (equivalently, regular algebraic automorphic automor-
phic representations) of finite slope. We review one of the two descriptions of
the eigencurve given by Coleman–Mazur [CM98] in the case of tame level
1 modular forms.18
The space Sk (Γ1 (pr )) of cuspidal modular forms of weight k and level
Γ1 (pr ) (for some r ≥ 1, and say with Qp coefficients) admits an action of
the Hecke operators Tl (for prime numbers l 6= p) and Up . The eigencurve Ep
admits a Zariski dense set of classical points associated to normalised Hecke
eigenforms (i.e. simultaneous eigenvectors of all these Hecke operators) f =
1 + n≥2 ap (f )q n ∈ Sk (Γ1 (pr )) of finite slope. Let us describe these classical
P

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

Sk (Γ1 (pr )) when p ≥ 1, but generates a 2-dimensional subspace with a basis

of Up -eigenvectors
(3.2) ∆(q) − αp ∆(q p ), ∆(q) − βp ∆(q p ),
corresponding to the factors of the Hecke polynomial 1 − τ (p)T + p11 T =
(1 − αp T )(1 − βp T ). This phenomenon can be explained a bit more trans-
parently by passing to the viewpoint of automorphic representations. If
f ∈ Sk (Γ1 (pr )) is a Hecke eigenform, then the lift of ιf to an automor-
phic form on GL2 (AQ ) generates a cuspidal automorphic representation
π of GL2 (AQ ) such that π∞ is ‘of weight k’, in the sense that its infini-
tesimal character coincides with that of the representation (Symk−2 C2 )∨ ,
and such that π ∞ is unramified away from the prime p. When f has finite
slope, πp is isomorphic to a subrepresentation of the normalized induction
GL2 (Qp )
iB2 (Q p)
χ1 ⊗ χ2 , where19
χ1 ⊗ χ2 : T2 (Qp ) = Q× ×
p × Qp → C
×

is a smooth character such that χ1 is unramified. We can further choose χ1

so that ιap (f ) = p1/2 χ1 (p). A choice of character χ = χ1 ⊗ χ2 of T2 (Qp )
GL2 (Qp )
such that πp occurs as a subrepresentation of iB2 (Q p)
χ1 ⊗ χ2 is called an
accessible refinement of πp . The association f ↔ (π, χ) sets up a bijection
between Hecke eigenforms in Sk (Γ1 (pr )) of finite slope (for varying r ≥ 1)
and cuspidal automorphic representations of weight k, unramified outside p,
and which are equipped with an accessible refinement as above. The paper
[CM98] uses the language of Hecke eigenforms, while later works (particu-
larly those setting up the theory of eigenvarieties for higher rank groups) use
the language of accessible refinements; both are useful. The set of classical
points embeds into the set of Qp -points of the eigencurve Ep .
We can now give one description of the eigencurve Ep . It is a disjoint
union of pieces corresponding to the possible residual representations rι (π)
associated to pairs (π, χ) as above. We fix a choice of residual represen-
tation ρ which arises in this fashion. Let Z(ρ) denote the set of pairs
(π, χ) as above such that rι (π) ∼= ρ. Let Ptr ρ,{p} denote the universal pseu-
dodeformation ring of tr ρ, and let X denote its rigid generic fibre. Let
T = Hom(T2 (Qp ), Gm ) denote the character variety of Q× ×
p × Qp . Thus
× × ×
T (K) is the set of continuous characters Qp × Qp → K . If (π, χ) ∈ Z(ρ)
then we can write down a point xπ,χ ∈ (X × T )(Qp ) as follows:
• The X -component corresponds to the pseudocharacter tr rι (π).
−1/2
• The T -component is the character δ(t1 , t2 ) = t2−k −1
2 ι χ1 (t1 )χ2 (t2 )|t1 /t2 |p .
We can define the eigencurve Ep as the Zariski closure of the set {xπ,χ |
(π, χ) ∈ Z(ρ)} of classical points (i.e. the smallest rigid analytic closed
subvariety of X × T containing all of these points). Of course, there is no
19Here we introduce the notation B for the upper-triangular Borel subgroup in GL ,
n n
and Tn for the diagonal maximal torus of GLn .
38 J. A. THORNE

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

20The sign here is to assist comparison with classical normalisations.

 RECIPROCITY AND FUNCTORIALITY 39

phenomena have been observed here which are yet to be explained – see for
example the fascinating conjectures in the articles [Buz05, BG16, BP19].

3.8. Analytic continuation of functoriality. We now come to the

first key principle that animates our proof: the idea that the existence of
the symmetric power lifting can be analytically continued along irreducible
components of the eigencurve Ep . (Following [Con99], we define irreducible
components as corresponding to connected components of the normaliza-
tion of Ep .) The following theorem, a special case of the results proved in
[NT21a], makes this precise:

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:

Corollary 3.10. Fix m ≥ 1. Suppose there exists an everywhere un-

ramified, regular algebraic cuspidal automorphic representation π of GL2 (AQ )
such that Symm m
∗ (π) exists. Then Sym∗ (π) exists for every such automorphic
representation of GL2 (AQ ).

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) .

Colmez observed that for a 2-dimensional representation V of GQp ,

Kisin’s condition on the existence of a crystalline period can be replaced
by the condition that its associated (ϕ, Γ)-module is reducible in the cat-
egory of all (ϕ, Γ)-modules over R, or in other words an extension of 1-
dimensional objects in this category. The (ϕ, Γ)-modules of rank 1 over R
42 J. A. THORNE

may be described explicitly: they correspond to the continuous characters

δ : Q×
p →K .
× 22

Definition 3.14. Let ρ : GQp → GLd (K) be a continuous representation

and let δ = δ1 ⊗ · · · ⊗ δn : Tn (Qp ) → K × be a continuous character. We say
that ρ is trianguline of parameter δ if D = Drig (V ) admits a filtration
0 = Fil0 D ⊂ Fil1 D ⊂ · · · ⊂ Filn D = D
by (ϕ, Γ)-submodules such that for each i = 1, . . . , n, Fili D is a (ϕ, Γ)-
submodule , direct summand as R-module, of rank i, and there is an iso-
morphism Fili D/ Fili−1 D ∼ = R(δi ). If ρ is trianguline of some parameter,
we simply say that it is trianguline.
This leads to another candidate definition for the ‘finite slope subspace’
Xf s ⊂ X × T : as the Zariski closure of the set of points (t, δ), where t is
the pseudocharacter of a continuous representation ρ : GQ,{p} → GL2 (Qp )
which is trianguline of parameter δ. Since (ϕ, Γ)-modules over R behave well
in families (thanks to the work of Kedlaya–Pottharst–Xiao [KPX14] and
Liu [Liu15]), this space Xf s is well-behaved. In particular, all of its points
correspond to trianguline Galois representations.23
We can now describe the role these ideas play with respect to symmetric
power functoriality. The notion of triangulation plays well with respect to
tensor operations on group representations. In particular, if ρ : GQp →
GL2 (K) is trianguline of parameter δ = (δ1 , δ2 ), then for any m ≥ 1 the
representation Symm ρ : GQp → GLm+1 (K) will be trianguline of parameter
Symm δ := (δ1m , δ1m−1 δ2 , . . . , δ2m ). This construction makes sense in families,
and we obtain a diagram
Ep

 Symm
(X × T )tri / (Xm+1 × Tm+1 )tri ,

where Xm+1 is the pseudodeformation space of tr Symm ρ and we define

Tm+1 = Hom(Tm+1 (Qp ), Gm ).
To go further, we need to introduce another eigenvariety (i.e. a higher
rank analogue of the eigencurve) whose points will give rise to Galois repre-
sentations of rank m + 1 (and which will fit into the empty spot in the top
22The associated (ϕ, Γ)-module is R(δ), which denotes R equipped with the actions
of ϕ, Γ satisfying ϕ(1) = δ(p) and γ · 1 = δ(γ).
23 We are ignoring some technicalities here. One minor point is that the ‘Zariski
closure’ results of [KPX14] require the points taken to be ‘strictly trianguline’, a condition
that is satisfied for the most points on the eigencurve. A more significant problem is that
the results of [KPX14] require us to have a true family of Galois representations, not
just a family of pseudocharacters. In practice, we apply the results of [KPX14] only in
an affinoid neighbourhood of a point (t, δ) where the representation ρ is irreducible, in
which case the pseudocharacter can be lifted to a family of Galois representations in this
neighbourhood in an essentially unique way, by [Che14, Proposition G].
 RECIPROCITY AND FUNCTORIALITY 43

right of the above diagram). Here it is technically convenient to pass to a

soluble CM extension M/Q, and to replace GL2 (AQ ) by G2 (AM + ), where
G2 is a definite unitary group in two variables over M + , and GLm+1 (AQ )
by Gm+1 (AM + ), where Gm+1 is a definite unitary group in m + 1 variables
over M + . This is valid since if π is a regular algebraic cuspidal automorphic
representation of GL2 (AQ ), then we can find an algebraic Hecke charac-
ter χ : M × \A× M → C
× such that the character twist BC
M/Q (π) ⊗ χ of
24
the base change of π to M is RACSDC , hence descends to the group
G2 . (The Galois-theoretic manifestation of this is that the representation
ρ = rι (π)|GM ⊗ rι (χ) satisfies ρc ∼
= ρ∨ ⊗ −1 .)
We lose nothing in doing this, since soluble base change and descent
between GLn (AQ ) and GLn (AM ) is well understood [AC89]25 and base
change and descent between conjugate self-dual, regular algebraic automor-
phic representations of GLm+1 (AM ) and automorphic representations of
Gm+1 (AM + ) is also reasonably well-understood (using e.g. the theorems
proved in [Lab11]) provided we choose our auxiliary data carefully (for ex-
ample, we want the p-adic places of M + to split in M so that G(M + ⊗Q Qp )
looks like a product of general linear groups).26
On the other hand, we gain in that the basic properties of the eigenva-
rieties of definite unitary groups have been very well-studied (in particular,
there are strong classicality theorems that give Galois-theoretic conditions
under which a point is in fact a classical point, i.e. arising from an automor-
phic representation of G(AM + ) and an accessible refinement of said repre-
sentation – see [NT21a, Lemma 2.30], which combines ideas from [Che11]
and [BHS17]). There are also genuine technical simplifications: one expects
that when m > 1 the eigenvariety of GLm+1 (AQ ) does not admit a dense set
of classical points and is not (locally on the source) finite and surjective over
weight space, two properties of the eigencurve Ep that are enjoyed by the
eigenvarieties of definite unitary groups (see e.g. [Che04, Théorème 6.3.6]).
Having done so, we arrive at a diagram

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

• EG2 , EGm+1 are eigenvarieties for the definite unitary groups G2 ,

Gm+1 , respectively. They each contain dense sets of classical points
associated to pairs (π, χ) consisting of an automorphic representa-
tion and accessible refinement, and admit maps κ2 : EG2 → T2,M ,
κm+1 : EGm+1 → Tm+1,M , where e.g. TM,2 is the character variety
of a maximal torus in G2 (M + ⊗Q Qp ).
• X2,M is the rigid space associated parameterizing conjugate self-
dual pseudocharacters lifting the residual pseudocharacter of rι (π)|GM ⊗
rι (χ).
• Xm+1,M is the rigid space associated parameterizing conjugate self-
dual pseudocharacters lifting the residual pseudocharacter of Symm (rι (π)|GM ⊗
rι (χ)).
• The bottom horizontal arrow is ‘passage to symmetric power’.
• The superscript ‘tri’ indicates in each case that we have passed
to the closed subspace which is the Zariski closure of the set of
trianguline points.
• The vertical arrows are given by ‘passage to associated Galois rep-
resentation’ – they are closed immersions.
The arrow we would ultimately like to construct is the broken horizontal
arrow, which would be a kind of overconvergent symmetric power functori-
ality, and from which we would hope to be able to deduce the symmetric
power functoriality in the usual sense (i.e. for classical automorphic repre-
sentations) by applying a classicality theorem to points of EGm+1 which are
images of classical points of EG2 .27
This is asking too much. What we are able to do (and this is enough to
be able to deduce Theorem 3.9) is show that if C ⊂ EG2 is an irreducible
component containing a non-critical classical point xπ,χ such that Symm ∗ (π)
m
exists, then Sym (i2 (C)) is contained in the image of im+1 . Here we use the
useful fact that if Z is an irreducible rigid analytic space and W ⊂ Z is a
Zariski closed subspace which contains a non-empty affinoid open of Z, then
W = Z. This means we need only show that the image of im+1 contains the
image under Symm of an open affinoid neighbourhood of xπ,χ . Ultimately we
need only work with affinoid local version of the above diagram where the
associated pseudocharacters are irreducible; this resolves one of the technical
problems alluded to in footnote 23.

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.

3.15. Vanishing of the adjoint Bloch–Kato Selmer group. Let

M be a CM field, let π be a RACSDC automorphic representation of GLn (AM ),
and let ρ = rι (π). Let us assume (as is conjecturally always the case) that
ρ is irreducible. There is an isomorphism ρc ∼
= ρ∨ ⊗ 1−n . Fixing a choice of
complex conjugation c ∈ GM + , this implies the existence of a non-degenerate
n n
bilnear form h·, ·i : Qp × Qp → Qp satisfying the identity
hρ(g c )v, ρ(g)wi = 1−n (g)hv, wi
n
for all g ∈ GM , v, w ∈ Qp . The adjoint representation ad ρ = Mn (Qp ) is a
Qp [GM ]-module; we can extend the action of GM to an action of GM + by
defining c · X = −X ∗ , where ∗ denotes adjoint with respect to h·, ·i.
The Galois cohomology group H 1 (M + , ad ρ) parameterizes deformations
of ρ to Qp [] which are conjugate self-dual (a variant of the computation of
the tangent space to the universal deformation ring of an absolutely irre-
ducible residual representation, already discussed above). The Bloch–Kato
Selmer group is defined to be
Y
Hf1 (M + , ad ρ) = ker(H 1 (M + , ad ρ) → H 1 (Mv+ , ad ρ)/Hf1 (Mv+ , ad ρ),
v

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

if v|p. Its elements correspond to those deformations of ρ which are poten-

tially semistable. Its vanishing is predicted by the Bloch–Kato conjecture28
and, as we have seen in the previous section, would have consequences for
the geometry of unitary group eigenvarieties.
The vanishing of Hf1 (M + , ad ρ) is in some cases a natural byproduct
of the proof of automorphy lifting theorems. Recall that one approach to
proving an automorphy lifting theorem is to try to prove that a map R →
T from a Galois deformation ring (say with conditions from p-adic Hodge
theory imposed) to a Hecke algebra (say acting on the cuspidal part of the
cohomology of an arithmetic group) is an isomorphism. If our representation
ρ is classified by a homomorphism f : R → Qp , then the Bloch–Kato Selmer
group Hf1 (M + , ad ρ) can be identified with the dual of the Zariski tangent
space29 (ker(f )/ ker(f )2 )[1/p]. On the other hand, if we can show that R →
T is an isomorphism, then this Zariski tangent space will be forced to be
zero, since T[1/p] will be an étale algebra (essentially because the space of
cuspidal automorphic forms is unitary and admissible as a representation
of the adele group, and therefore in particular semisimple). Such vanishing
results appear already in [Wil95].
Recalling the discussion in §3.1, we might expect to need strong condi-
tions on the residual representation ρ (in particular, absolute irreducibility).
However, in our applications we need to be able to treat the case where ρ
is a trivial representation in arbitrary degree (as this is required in order to
be able to prove the existence of Symm ∗ (π∆ ) using the 2-adic eigenvariety).
A slightly different tack is therefore required.30
In the papers [NT20, Tho21a], we prove the following vanishing theo-
rem.
Theorem 3.16. Let M be a CM field and let π be a RACSDC automor-
phic representation of GLn (AM ). Suppose that rι (π)|GM (ζ ∞ ) is irreducible.31
p
Then we have
Hf1 (M + , ad rι (π)) = 0.
28We don’t have space here to discuss further the Bloch–Kato conjecture, which
predicts the order of vanishing at the point s = 1 of the L-function L(ιV, s) of an algebraic
Galois representation V = 0. When V is non-trivial but of weight 0 (such as in the case
V = ad ρ) then the L-function is expected to be non-vanishing at s = 1 (in line with the
non-vanishing results for automorphic L-functions on the line Re(s) = 1 which may be
used to prove the Sato–Tate conjecture. See Fontaine’s survey [Fon92] for more details.
29This is unsurprising, although not quite a formal consequence of the definitions.
See [Kis09c, §2.3] for a justification of a similar statement.
30Another approach is taken in Kisin’s paper [Kis04], which proves the vanishing of
1
Hf (Q, ad rι (π)) in many cases for regular algebriac cuspidal automorphic representations
of GL2 (AQ ). This uses the realisation of these p-adic Galois representations in the étale
cohomology of modular curves in a way which seems difficult to generalise to the higher
rank situation.
31Here M (ζ ∞ ) denotes the subfield of M obtained by adjoining all p-power roots of
p
unity to M .
 RECIPROCITY AND FUNCTORIALITY 47

This is rather general. Indeed, there is no condition at all on the residual

representation or on the prime p. If π arises by symmetric power lifting and
base change of a regular algebraic cuspidal automorphic representation of
GL2 (AQ ) which does not have CM, then rι (π)|GM (ζ ∞ ) is irreducible, so
p
we obtain the required vanishing. (This follows from the stronger fact that
the Zariski closure of the image of rι (π) in GLm+1 contains a conjugate of
the the image of the unique irreducible representation SL2 → GLm+1 , see
[NT20, Example 2.34].)
In proving this theorem we make use of some new ideas in the theory
of pseudocharacters and in our understanding of the Taylor–Wiles method.
We briefly sketch the main ones. Instead of considering a map R → T,
we consider a map P → T in the category CO , where P is a universal
pseudodeformation ring of tr ρ and T is a Hecke algebra acting on a space
of algebraic modular forms, with integral structure, on the definite unitary
group Gn . We need to impose p-adic Hodge theoretic conditions on the
pseudocharacters that appear in P . It is far from obvious how to do this,
since a pseudocharacter by its very nature does not give rise to a module
with a Galois action on which conditions can be imposed. This problem
was solved by Wake–Wang-Erickson [WWE19], who show that one can
get around this by using the notion of Cayley–Hamilton representation of
a pseudocharacter defined by Chenevier [Che14]. Another result we need
concerns deformations of pseudocharacters with integral structure. More
precisely, we prove the following theorem ([NT20, Proposition 2.9]):

Theorem 3.17. Let Γ be a profinite group and let ρ : Γ → GLn (O)

be a continuous homomorphism such that ρ ⊗O K is absolutely irreducible.
Let Θ = tr ρ. Consider the map ρ̃ 7→ tr ρ̃ from the set of continuous ho-
momorphisms ρ0 : Γ → GLn (O ⊕ K/O) lifting ρ to the set of continuous
pseudocharacters Θ0 : Γ over O ⊕ K/O lifting Θ. Then there is a constant
C ∈ Z≥0 , depending only on ρ(Γ), such that this map is ‘bijective up to
$C -torsion’, in the following sense:
(1) For any Θ0 , there exists a representation ρ0 such that tr ρ0 = αC ◦ θ,
where αC is the ring endomorphism of O⊕K/O which sends x+y
to x + $C y.
(2) For any pair ρ0 , ρ00 of representations such that tr ρ0 = tr ρ00 , αC ◦ ρ0
and αC ◦ ρ00 are conjugate under the group 1 + Mn (K/O).

The proof of this theorem uses the definition of pseudocharacter intro-

duced by Lafforgue in [Laf18], and is in some sense an integral refinement
of the proof there that pseudocharacters biject with semisimple represen-
tations over an algebraically closed field. It is interesting to note that the
different but equivalent definitions of pseudocharacter given in the papers
[Che14, Laf18] both have a role to play.
To proceed, we consider the map f : P → Qp associated to rι (π) (which
factors through T, essentially by definition), and try to show that the map
48 J. A. THORNE

Pker f → Tker f induced after localisation and completion at ker f is an iso-

morphism. This will suffice to prove the vanishing of the adjoint Bloch–Kato
Selmer group, for the same reasons as outlined above: the (completed and lo-
calised) Hecke algebra is étale (and even a field in this case) while the Zariski
tangent space of Pker f may be identified with the Bloch–Kato Selmer group
(here we crucially use the results of [Liu07] to match up the Bloch–Kato
Selmer condition with the one arising from the p-adic Hodge theoretic con-
dition imposed in [WWE19], cf. [WWE19, Remark 5.2.2]).
To show that Pker f → Tker f is an isomorphism, we use a variant of the
Taylor–Wiles method, which requires patching the action of certain auxiliary
rings PQ on auxiliary spaces HQ of automorphic forms on the definite unitary
group. In the classical Taylor–Wiles method, Q is a set of auxiliary places of
the base number field at which additional ramification is allowed. There is
an finite abelian group ∆Q (a quotient of the product of the inertia groups at
the places of Q) and a homomorphism O[∆Q ] → PQ such that PQ ⊗O[∆Q ] O
may be naturally identified with P . The space of HQ of automorphic forms
with additional level at Q is supposed to be a PQ -module which is finite free
as O[∆Q ]-module, and which satisfies the analogous property HQ ⊗O[∆Q ] O ∼ =
H, where H is the space of automorphic forms on which the Hecke algebra
T at minimal level acts faithfully. (This freeness of HQ over the ring O[∆Q ]
is what provides the lower bound on the growth of the Q-ramified Hecke
algebra, mentioned in our sketch of the Taylor–Wiles method in §3.1.)
In order to be able to define HQ (and in particular, construct an isomor-
phism HQ ⊗O[∆Q ] O ∼ = H), one typically needs the residual representation
ρ to be sufficiently non-degenerate; the places v ∈ Q are selected so that
e.g. ρ is unramified at v and ρ(Frobv ) has no repeated eigenvalues. In our
situation, where ρ can even be trivial, this is no longer possible, and we need
lose control over the auxiliary spaces HQ . In particular, although they are
O[∆Q ]-modules, they may no longer be of bounded rank. Since the Taylor–
Wiles method traditionally used a kind of diagonalization or compactness
argument to construct the patched objects, this looks like a serious problem!
In his work on the Fontaine–Mazur conjecture for residually reducible
representations ρ : GQ → GL2 (Qp ), Pan [Pan21] contended with exactly
this difficulty. He showed how to use ultrafilters to construct very large
‘ultra-patched’ modules. This may not have reasonable properties but it
turns out that after localization and completion at the prime ideal corre-
sponding to ρ, enough finiteness properties hold that one can still make the
Taylor–Wiles method work in a neighbourhood of ρ. The same technique
applies equally well in our case. It seems an interesting problem to clarify
the role played by ultrafilters in this argument.

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

such representation π. In order to do show the existence of such a π, we use

an argument in which many of the same themes appear (p-adic automorphic
forms, residually reducible Galois representations) but in slightly different
form, in a spirit closer to our previous work with Clozel on symmetric power
functoriality in small degrees [CT14, CT15, CT17].
To explain the idea, suppose given a regular algebraic cuspidal automor-
phic representation π of GL2 (AQ ) with the following properties:
• π does not have CM.
• There exists a prime number p and an isomorphism ι : Qp → C
G
such that rι (π) ∼
= Ind Q χ, for some imaginary quadratic extension
GK
×
K/Q and continuous character χ : GK → Fp .
It is easy to construct examples of such automorphic representations. For
example, we could start with an elliptic curve E over Q which does have CM,
and is therefore associated to a CM regular algebraic cuspidal automorphic
representation πE of GL2 (AQ ). Then choose a prime number p such that
ρE,p is irreducible and a prime number l 6= p such that E has good reduction
at l and ρE,p |GQl is the trivial representation. Ribet’s level-raising theorem
then implies the existence of another regular algebraic automorphic repre-
sentation π such that rι (π) ∼= rι (πE ) and πl is the Steinberg representation
of GL2 (Ql ) (i.e. the unique irreducible admissible representation of trivial
central character and conductor l). The local condition at l implies that π
does not have CM.
We now consider the problem of establishing the existence of the mth
symmetric power lifting of π. Restricting to K, we see that the residual
representation of Symm rι (π) has the simple form
m
∼
M
m
Sym rι (π)|GK = χm−i χc,i−1 .
i=0
m
Although Sym rι (π) is irreducible, its residual representation is highly re-
ducible. Automorphy lifting theorems are one of the most powerful tools
that exist to prove Galois representations are automorphic, but establishing
the residual automorphy of the residual representation is often a sticking
point. Here we might hope that the relatively simple form of the residual
representation makes it easier to check the residual automorphy, and that an
automorphy lifting theorem might be available which applies in this residu-
ally reducible situation.
Such an automorphy lifting theorem is proved in [ANT20], generalising
the one proved in [Tho15] that was applied to the symmetric power functo-
riality problem in the papers [CT14, CT15, CT17]. Here is the statement.

Theorem 3.19. Let M be a CM number field, let ι : Qp → C be an

isomorphism, and let ρ : GM → GLn (Qp ) be a conjugate self-dual Galois
representation which satisfies the following conditions:
50 J. A. THORNE

×
(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

Then ρ is ordinarily automorphic, in the sense that there is an ι-ordinary

RACSDC automorphic representation Π of GLn (AM ) such that ρ ∼ = rι (Π).
We do not define the ι-ordinary condition on π here32, but only note that
it generalies the well-known ‘slope of Up –eigenvalue equals 0’ condition on
classical Hecke eigenforms and corresponds, under local-global compatibility
at the p-adic places of M , to the requirement that rι (π) is ordinary of weight
corresponding to that of π.
The proof of this theorem relies on yet another class of p-adic automor-
phic forms, namely the ordinary p-adic automorphic forms, which may be
thought of as overconvergent modular forms of slope 0. Starting in the 1980’s
with the paper [Hid86], Hida developed a powerful theory of ordinary p-
adic modular forms, using it to construct families of Galois representations
with values in Hecke algebras which are finite covers of the Iwasawa alge-
bra Λ whose rigid generic fibre is the usual weight space. Geraghty [Ger19]
32But see [Ger19, Definition 5.3].
 RECIPROCITY AND FUNCTORIALITY 51

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

so propagate the automorphy of any irreducible component of Spec P ord to

any other component.
The key step in this argument of finding ‘generic’ primes in the intersec-
tions Wi ∩ Wi+1 is possible only if the locus of reducible pseudocharacters
in P ord is ‘not too large’. Unfortunately, it can definitely be too large in
practice. An important form of functoriality relevant for the unitary group
Gn is given by the theory of endoscopy, which implies (roughly speaking)
the existence of functorial liftings from groups Gk × Gn−k to Gn . Since the
dimension of the ordinary Hecke algebra for Gk is k[M + : Q], we would ex-
pect to see components in Spec Tord of dimension k[M + : Q] + (n − k)[M + :
Q] = n[M + : Q] = dim Tord arising by endoscopic transfer (and therefore
corresponding to Galois representations which are reducible in GLn ). We
therefore need to impose additional conditions to rule out the existence of
entire components of Spec Tord corresponding to reducible Galois represen-
tations. (This is one point where we depart from the context of [SW99],
where there are no endoscopic forms and the arguments needed are substan-
tially different.)
This explains the presence of the final condition (6) in the statement of
Theorem 3.19 – it is a local condition that can be used effectively to elimi-
nate the contribution of endoscopic lifts, since the Steinberg representation
(corresponding as it does under the local Langlands correspondence to an
indecomposable Weil–Deligne representation) cannot be the local compo-
nent of a lift from an endoscopic group of Gn . We have dwelled on this
point because it ties into the problem of verifying the residual automorphy
of the residual representation Symm rι (π)|GK (or rather a character twist
of this representation which is conjugate self-dual). We can hope to verify
the residual automorphy by, informally speaking, taking an endoscopic lift
from the group U (1)n , where automorphic representations can be written
down ‘by hand’ in terms of conjugate self-dual Hecke characters. However,
this automorphic representation of Gn (or its base change to GLn (AM )) is
not suitable for application of Theorem 3.19 – we need to modify it so that
its local component at the place v0 is an unramified twist of the Steinberg
representation. In other words, we need to raise the level by finding a con-
gruence to another automorphic representation of Gn which has the same
residual representation but this specified local behaviour at the place v0 .
This level-raising argument is one of the most technically intricate part
of the paper [NT21a], where it is carried out using an inductive argument,
which treats the case of even m using the theory of types and the case
of odd m by reference to the paper [AT20], which establishes a version of
Ihara’s lemma for unitary groups of the form U (2, D) (D a division algebra).
The forthcoming paper [Tho22] will give a direct approach (not based on
induction on m).
 RECIPROCITY AND FUNCTORIALITY 53

3.20. The case of level N > 1. Let π be a regular algebraic cuspidal

automorphic representation of GL2 (AQ ), without CM. The arguments de-
scribed so far suffice to prove the existence of the symmetric power liftings
of π in every degree provided that π is everywhere unramified (therefore
lifted from a newform of level 1), or more generally provided that π deter-
mines a point of the 2-adic, tame level 1 Coleman–Mazur eigencurve E2 . We
can interpret this latter condition in terms of the local component π2 : it
means that π2 has an accessible refinement.34 We can phrase this equiva-
lently as the condition that π2 has non-trivial Jacquet module, or that π2
is not supercuspidal, or that the Weil–Deligne representation recQp (π2 ) is
reducible.
In order to establish the existence of the symmetric power liftings of the
automorphic representation π without any conditions on its ramification,
we essentially induct on the level. This kind of strategy is familiar from
other situations in the literature. For example, in the original work of Wiles
[Wil95], a modularity lifting theorem is proved first at ‘minimal level’ (rela-
tive to the ramification of the residual representation ρ). This minimal level
modularity lifting theorem is then used as a stepping stone to an R = T
theorem where additional ramification is allowed on both the Galois and
automorphic sides. Another example which is particularly relevant to us is
the proof of Serre’s modularity conjecture [Ser87] by Khare–Wintenberger
[KW09a, KW09b]. Recall that Serre’s conjecture asserts that any repre-
sentation ρ : GQ → GL2 (Fp ) which is of ‘S-type’, in the sense that it is
is absolutely irreducible and odd, arises as the residual representation of a
newform f . Serre associated to any such representation ρ a weight k(ρ) and
tame conductor N (ρ), and conjectured moreover that it should be possible
to choose f to be of weight k(ρ) and level N (ρ). The approach taken in
[KW09a] is essentially to induct on the number of primes dividing N (ρ),
the base case where N (ρ) = 1 (in other words, where ρ is ramified only
above the prime p) having been treated already in [Kha06]. We will come
to the technique of ‘killing ramification’, introduced in [KW09a], and which
makes this induction argument possible, below.
In our case it is natural to divide the proof up slightly differently into two
parts, as follows. Let π again be a regular algebraic cuspidal automorphic
representation of GL2 (AQ ), without CM.
• First, we establish the existence of the symmetric power liftings of
π in every degree under the assumption that π has no supercuspidal
local components; in other words, that for every prime p, πp admits
an accessible refinement. In this part of the proof, we induct on the

34In fact, we require that π admits an accessible refinement χ = χ ⊗χ such that χ

2 1 2 1
is unramified. However, if π2 admits an accessible refinement χ we can always twist π so
that χ1 is unramified, and twisting is easily seen to preserve the existence (or otherwise)
of the symmetric power lifting in any given degree.
54 J. A. THORNE

number of primes p such that πp is ramified, the base case where π

is everywhere unramified having been taken care of already.
• Then, we establish the existence of the symmetric power liftings of
π in complete generality. In this part of the proof, we induct on
the number of primes p such that πp is supercuspidal, the base case
where there are no such primes being the content of the previous
bullet point.
The first bullet point takes us to the end of the paper [NT21a], and rep-
resents the natural reach of the methods that have been introduced so far
in this article. The second bullet point is covered by the follow-up paper
[NT21b] and requires a new idea, that of the ‘functoriality lifting theorem’.
We address these in turn.
3.20.1. The everywhere locally refineable case. Suppose then that π is a
regular algebraic cuspidal automorphic representation of GL2 (AQ ), without
CM, and that πp is ramified but admits an accessible refinement χ. Let
N = M pr be the conductor of π, where (p, M ) = 1. Then (π, χ) defines a
classical point of the tame level M , p-adic eigencurve Ep (M ) (say as defined
in [Buz07]). Using the accumulation property of the eigencurve, it is possible
to find an affinoid neighbourhood U of xπ,χ in Ep (M ) and a dense set of
classical points Ep (M ) in U corresponding to automorphic representations
which are unramified at p, and for which the symmetric power liftings are
therefore known to exist, by our inductive hypothesis. To conclude, we invoke
our ‘analytic continuation of functoriality’ principle. In our first statement
of this above (as Theorem 3.9), we restricted to the case of the tame level
1 eigencurve Ep = Ep (1), but in fact it is valid without restriction on the
tame level (and even for automorphic representations of unitary groups over
CM fields – the main technical ingredient is the vanishing of the adjoint
Bloch–Kato Selmer group, which is indeed proved for RACSDC automorphic
representations over an arbitrary CM field). Here is such a statement.
Theorem 3.21. Fix m ≥ 1. Let p be a prime number and let (π, χ) and
(π 0 , χ0 )
be pairs consisting of regular algebraic automorphic representations,
without CM, and equipped with accessible refinements of the components at
p. Suppose that π, π 0 have prime-to-p conductors dividing M , and that the
following conditions are satisfied:
(1) The refinement χ has small slope.
(2) Both refinements χ, χ0 are (m + 1)-regular.
(3) Symm∗ (π) exists.
If the classical points xπ,χ , xπ0 ,χ0 of Ep (M ) lie on a common irreducible
component of Ep (M ), then Symm ∗ (π) exists too.

One wrinkle is the (m + 1)-regularity condition on the refinements. This

condition does not trouble us in the level 1 case as one can show that any
for any everywhere unramified regular algebraic cuspidal automorphic rep-
resentation π of GL2 (AQ ), every refinement of π2 is (m+1)-regular for every
 RECIPROCITY AND FUNCTORIALITY 55

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.

In order to remove the (m + 1)-regularity condition, we use a trick based

on the Taylor–Wiles method. This appears to be a novel application of the
Taylor–Wiles method, which we apply to prove a kind of density statement
for the local components of automorphic representations in a given congru-
ence class. One can draw an analogy with equidistribution theorems such as
those proved in [Ser97, Shi12], which prove that the set of local components
of automorphic representations of varying weight and level is equidistributed
with respect to Plancherel measure; we obtain a weaker statement, namely
Zariski density of the corresponding points in the spectrum of a local lifting
ring, but with the crucial difference that we are considering the set of au-
tomorphic representations with a given residual representation with respect
to a fixed isomorphism ι : Qp → C. The precise statement we prove is as
follows.
Theorem 3.23. Fix m ≥ 1, and let π be a cuspidal automorphic rep-
resentation of GL2 (AQ ) of weight k, without CM. Suppose that for each
prime number l, πl admits an accessible refinement. Then we can find a
prime number p, an isomorphism ι : Qp → C, and another cuspidal auto-
morphic representation π 0 of weight k with the following properties:
m 0
(1) Symm∗ (π) exists if and only if Sym∗ (π ) does.
(2) For each prime number l, πl0 admits an accessible refinement. If
moreover πl0 is ramified, then this refinement is (m + 1)-regular.
This theorem, combined with Proposition 3.22, suffices to prove the exis-
tence of Symm ∗ (π) when π has no supercuspidal local components. We sketch
the proof. First, we will choose the prime p to be large relative to m, k, and
π 35. We will construct π 0 to have the same weight k, the same residual rep-
resentation rι (π) ∼
= rι (π 0 ) with respect to some isomorphism ι : Qp → C,
and also to be unramified at p. We may then apply [BLGGT14, Theorem
4.2.1] to conclude that the first point in the theorem statement is true (in the

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

equivalent form that the Galois representation Symm rι (π) is automorphic

if and only if Symm rι (π 0 ) is).
To construct π 0 , we use Kisin’s refinement of the Taylor–Wiles method.
We need to introduce some relevant lifting and deformation rings. Let ψ =
det rι (π), and let S denote the set of primes at which rι (π) is ramified. Fixing
a large enough coefficient field K, we let Rρ,S,ψ ∈ CO denote the object
representing the functor of deformations of ρ of determinant ψ, unramified
outside S, and Fontaine–Laffaille of Hodge–Tate weights {0, k − 1} on the
decomposition group GQp . If l ∈ S, we write Rρ,l,ψ ∈ CO for the object
representing the functor of liftings of ρ|GQl of determinant ψ (which are
moreover Fontaine–Laffaille of Hodge–Tate weights {0, k − 1} if l = p).
Choosing p sufficiently large relative to p, we can assume that each of the
rings Rρ,l,ψ is formally smooth over O of relative dimension 3.
In order to compare the global deformation ring with the local lifting
rings, we need to introduce the device of framed liftings. Accordingly, we
S
let Rρ,S,ψ ∈ CO denote the object representing the functor of orbits of pairs
(ρ, {αl }l ∈ S), where ρ is a lifting of type Rρ,S,ψ and for each l ∈ S, αl ∈
ker(GL2 (A) → GL2 (k)). The orbits are taken under the action of the group
ker(GL2 (A) → GL2 (k)), which acts by the formula g · (ρ, {αl }l ∈ S) =
S
(gρg −1 , {gαl }l∈S ). There is a natural ring homomorphism Rρ,S,ψ → Rρ,S,ψ ,
which corresponds (under the Yoneda lemma) to the functor of ‘forgetting
the matrices αl ’. This ring homomorphism is formally smooth and in fact
S
has a section, so that Rρ,S,ψ is non-canonically isomorphic to a power series
ring over Rρ,S,ψ .
Having introduced the framed version of the global deformation ring,
we introduce also the completed tensor product Rρ,S,ψ loc =⊗ b l∈S Rρ,l,ψ ∈ CO ,
which represents the functor of tuples ({ρl }l∈S ) of liftings of the local repre-
sentations ρ|GQl (these liftings supposed to be all of determinant ψ|GQl ). A
tangent space calculation generalising the one described in §3.1 implies that
there is a presentation
loc S
Rρ,S,ψ JX1 , . . . , Xg K  Rρ,S,ψ .

Introducing sets Q of Taylor–Wiles primes (valid since rι (π) is absolutely

irreducible, although its symmetric power might not be), we obtain auxiliary
diagrams
loc S S
Rρ,S,ψ JX1 , . . . , Xg K  Rρ,S∪Q,ψ  Rρ,S,ψ .
We may introduce spaces of automorphic forms that these deformation rings
act on (through a map to an appropriate Hecke algebra). The Taylor–Wiles–
Kisin patching argument can then be used to ‘patch’ these together to
obtain a ‘patched module’ of automorphic forms H∞ , which is a faithful
loc JX , . . . , X K-module. Typically at this point in the argument, one
Rρ,S,ψ 1 g
S
would ‘return to level 1’ to find that Rρ,S,ψ acts faithfully (or at least has
 RECIPROCITY AND FUNCTORIALITY 57

nilpotent annihilator) on a space of automorphic forms at base level, which

in turn implies an automorphy lifting theorem.
Here however we want to do something slightly different. Recalling the
statement of Theorem 3.23, let us suppose for contradiction that for every set
Q of Taylor–Wiles primes for ρ there is no automorphic representation π 0 of
weight k which is unramified away from S ∪ Q − {p}, satisfies rι (π 0 ) ∼ = rι (π),
and with the property that πl0 is (m + 1)-regular for each l ∈ S. This implies
that for every such π 0 , the local representations rι (π 0 )|GQl (l ∈ S) must lie
in the ‘non-(m + 1)-regular’ locus of Spec Rρ,S,ψloc , a proper Zariski closed

subset. This leads to a contradiction because this property is inherited by

the patched module H∞ , implying in particular that H∞ cannot be a faithful
loc JX , . . . , X K-module. We deduce that there must be at least one set
Rρ,S,ψ 1 g
Q of Taylor–Wiles primes and one automorphic representation π 0 with the
desired (m+1)-regular refinements at primes l ∈ S. We’re now done: we just
need to check that if πq0 is ramified for some q ∈ Q then πq0 is also (m + 1)-
regular, but this is automatic since the image rι (π 0 )(IQq ) is generated by a
non-scalar matrix of p-power order and p is large relative to m.
3.23.1. The case of supercuspidal local components. We now treat the
general case of our main Theorem 2.28. To reduce this case to the previous
one, we use the the following ‘functoriality lifting theorem’:
Theorem 3.24. Let π, π 0 be cuspidal automorphic representations of
GL2 (AQ ) of weight 2. Let m ≥ 1, let p be a prime number, and let ι :
Qp → C be an isomorphism such that the following conditions are satisfied:
(1) There is an isomorphism rι (π) ∼= rι (π 0 ).
0
(2) Neither π nor π is ι-ordinary.
(3) For each prime l 6= p, πl is a twist of the Steinberg representation
if and only if πl0 is.
(4) There exists a ≥ 1 such that pa > max(5, 2m + 1) and there is a
sandwich
PSL2 (Fpa ) ⊂ Proj rι (π)(GQ ) ⊂ PGL2 (Fpa )

up to conjugacy in PGL2 (Fp ). (Here Proj rι (π) denotes the projec-

tive representation associated to rι (π).)
m 0
Then Symm
∗ (π) exists if and only if Sym∗ (π ) does.

Let us explain why we call this a functoriality lifting theorem. It could

be viewed as a kind of automorphy lifting theorem: it says that the au-
tomorphy of the Galois representation Symm rι (π) implies the automorphy
of the Galois representation Symm rι (π 0 ), which has the same associated
residual representation. However, because we impose more stringent condi-
tions, namely that these Galois representations arise as functorial transfers
of 2-dimensional Galois representations which are known already to be au-
tomorphic, we are able to obtain a more refined statement that we would
58 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.

36This definition requires 3 prime numbers to be in a good position with respect to

π. The word is supposed to suggest that, having been sprinkled with a few well-chosen
primes, the automorphic representation π is especially palatable. An additional subtlety
over the case considered in [KW09a] is that we need control of the residual representation
of π including at the prime q where additional ramification is imposed.
 RECIPROCITY AND FUNCTORIALITY 59

We are now ready to carry out the induction step. We choose ι : Qp → C

such that rι (π) has large image and πp is supercuspidal. Using well-known
principles (in particular, the theory of types as in [Kha01]), we can find
another cuspidal automorphic representation π 0 of weight 2 such that rι (π) ∼=
0 0
rι (π ), such that for each prime l 6= p, πl , πl have the same inertial type,
and (crucially) such that πp0 is not supercuspidal. Thus Symm 0
∗ (π ) exists, by
m
induction, and Theorem 3.24 implies that Sym∗ (π) exists too.
3.24.2. Proving Theorem 3.24. Let us now describe the modification of
the Taylor–Wiles–Kisin method that leads to the proof of Theorem 3.24. The
case p = 2 requires additional complications, analogous to those appearing
in the article [Kis09b] relative to [Kis09c], so we concentrate on the case
where p is odd. In fact, we adopt the following simplifying hypotheses:
• After base change to a soluble totally real extension F/Q, rι (π)|GF
is unramified away from p but crystalline (hence Barsotti–Tate) at
the p-adic places of F .
• det rι (π) = −1 .
• [F : Q] is even.
• Let Sp denote the set of p-adic places of F . If v ∈ Sp , then rι (π)|GFv
is the trivial representation.
Let ρ = rι (π)|GF , let Sp denote the set of p-adic places of F , and let
Rρ,Sp ,−1 ∈ CO denote the object representing the functor of deformations of
ρ which are unramified outside Sp , of determinant −1 , and Barsotti–Tate
non-ordinary in the sense of [Kis09c]. The hypothesis that [F : Q] is even
implies that there is a quaternion algebra D over F which is split at fi-
nite places and definite at the infinite places; the deformation ring Rρ,Sp ,−1
acts on a module HD of non-ordinary algebraic modular forms with integral
structure on D.
The Taylor–Wiles–Kisin method, as exhibited in [Kis09c] and sketched
in the previous section, makes it possible to construct an object S∞ ∈ CO ,
an S∞ -algebra R∞ and an R∞ -module HD,∞ with the following properties:
• There is a number q such that S∞ ∼ = OJS1 , . . . , Sq K.
• There is an isomorphism R∞ ⊗S∞ O ∼ = Rρ,Sp ,−1 and a compatible
isomorphism HD,∞ ⊗S∞ O ∼ = HD .
• HD,∞ is finite free as S∞ -module and R∞ is an integral domain of
the same Krull dimension as S∞ .
These facts together imply that R∞ acts faithfully on HD,∞ , hence that the
annihilator of HD as Rρ,Sp ,−1 -module is nilpotent – and this directly implies
that any deformation of ρ corresponding to a homomorphism Rρ,Sp ,−1 → O
is automorphic.
We now introduce the structures associated to Symm ρ. If p is larger
than m then this representation will be reducible, so we need to use pseu-
dodeformation rings. Accordingly, we introduce a quadratic CM extension
60 J. A. THORNE

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

product functoriality – itself a non-trivial problem. More speculatively, non-

abelian (or non-soluble) descent for algebraic automorphic representations
would allow one to upgrade potential automorphy, as established in great
generality in the papers [BLGGT14, PT15] to true automorphy. We look
forward to finding out what the future holds.

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

[Fon94b] Jean-Marc Fontaine. Représentations l-adiques potentiellement semi-stables.

Number 223, pages 321–347. 1994. Périodes p-adiques (Bures-sur-Yvette,
1988).
[Fon94c] Jean-Marc Fontaine. Représentations p-adiques semi-stables. Number 223,
pages 113–184. 1994. With an appendix by Pierre Colmez, Périodes p-adiques
(Bures-sur-Yvette, 1988).
[Gau66] Carl Friedrich Gauss. Disquisitiones arithmeticae. Yale University Press, New
Haven, Conn.-London, 1966. Translated into English by Arthur A. Clarke, S.
J.
[Gel75] Stephen S. Gelbart. Automorphic forms on adèle groups. Annals of Mathe-
matics Studies, No. 83. Princeton University Press, Princeton, N.J.; Univer-
sity of Tokyo Press, Tokyo, 1975.
[Ger19] David Geraghty. Modularity lifting theorems for ordinary Galois representa-
tions. Math. Ann., 373(3-4):1341–1427, 2019.
[GJ72] Roger Godement and Hervé Jacquet. Zeta functions of simple algebras. Lec-
ture Notes in Mathematics, Vol. 260. Springer-Verlag, Berlin-New York, 1972.
[GK14] Toby Gee and Mark Kisin. The Breuil-Mézard conjecture for potentially
Barsotti-Tate representations. Forum Math. Pi, 2:e1, 56, 2014.
[Han17] David Hansen. Universal eigenvarieties, trianguline Galois representations,
and p-adic Langlands functoriality. J. Reine Angew. Math., 730:1–64, 2017.
With an appendix by James Newton.
[Hid86] Haruzo Hida. Galois representations into GL2 (Zp [[X]]) attached to ordinary
cusp forms. Invent. Math., 85(3):545–613, 1986.
[HSBT10] Michael Harris, Nick Shepherd-Barron, and Richard Taylor. A family
of Calabi-Yau varieties and potential automorphy. Ann. of Math. (2),
171(2):779–813, 2010.
[JS81] H. Jacquet and J. A. Shalika. On Euler products and the classification of
automorphic representations. I. Amer. J. Math., 103(3):499–558, 1981.
[JT20] Christian Johansson and Jack A. Thorne. On subquotients of the étale co-
homology of Shimura varieties, page 306–334. London Mathematical Society
Lecture Note Series. Cambridge University Press, 2020.
[Ked04] Kiran S. Kedlaya. A p-adic local monodromy theorem. Ann. of Math. (2),
160(1):93–184, 2004.
[Kha01] Chandrashekhar Khare. A local analysis of congruences in the (p, p) case. II.
Invent. Math., 143(1):129–155, 2001.
[Kha06] Chandrashekhar Khare. Serre’s modularity conjecture: the level one case.
Duke Math. J., 134(3):557–589, 2006.
[Kis03] Mark Kisin. Overconvergent modular forms and the Fontaine-Mazur conjec-
ture. Invent. Math., 153(2):373–454, 2003.
[Kis04] Mark Kisin. Geometric deformations of modular Galois representations. In-
vent. Math., 157(2):275–328, 2004.
[Kis09a] Mark Kisin. The Fontaine-Mazur conjecture for GL2 . J. Amer. Math. Soc.,
22(3):641–690, 2009.
[Kis09b] Mark Kisin. Modularity of 2-adic Barsotti-Tate representations. Invent.
Math., 178(3):587–634, 2009.
[Kis09c] Mark Kisin. Moduli of finite flat group schemes, and modularity. Ann. of
Math. (2), 170(3):1085–1180, 2009.
[KMSW14] Tasho Kaletha, Alberto Minguez, Sug Woo Shin, and Paul-James White.
Endoscopic classification of representations: Inner forms of unitary groups,
2014.
[Kot92] Robert E. Kottwitz. On the λ-adic representations associated to some simple
Shimura varieties. Invent. Math., 108(3):653–665, 1992.
66 J. A. THORNE

[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.

DPMMS, Wilberforce Road, Cambridge, CB3 0WB, UK

Email address: [email protected]