ON THE IMAGE OF AUTOMORPHIC GALOIS

REPRESENTATIONS

ALIREZA SHAVALI
arXiv:2502.10799v1 [math.NT] 15 Feb 2025

Abstract. In this paper, we study extra-twists for automorphic representations

of GLn and use them to give a precise description of the image of the Galois
representations associated with regular algebraic cuspidal automorphic represen-
tations of GL3 over totally real fields. We also formulate a conjecture for the
GLn -case and show how it follows from some standard conjectures in the Lang-
lands program. Finally, assuming the existence of a motive associated with the
representation, we study the relation of our constructions with the Mumford-Tate
group.

Contents
1. Introduction 1
2. Extra-Twists for GLn 3
2.1. Inner and outer twists 4
2.2. Constructing forms of algebraic groups using 1-cocycles 6
3. Image of Galois Representations with Extra-Twists 7
3.1. Extra-twists and Galois representations 9
3.2. The Lie algebra of the image 11
4. Image of Automorphic Galois Representations 13
4.1. The GL2 -case 16
4.2. The GL3 -case 17
4.3. The GLn -case 20
5. Relations to the Mumford-Tate Group 21
References 24

1. Introduction
From the viewpoint of the Langlands program, one expects a connection between
three different worlds: Motives, Galois Representations, and Automorphic Repre-
sentations. One expects to be able to translate different features of the objects from
one world to the others. For instance, the motive having many symmetries (hav-
ing a big endomorphism ring) should force the Galois representation to have small
image (as the Mumford-Tate conjecture explains) and these both should translate
to the associated automorphic representation having special properties. This line of
study can be traced back to the work of Serre on Galois representations associated
with rational elliptic curves and his famous open image theorem [21]. This vague
philosophy is our main guide throughout this paper.
We will mostly focus on the connection of the automorphic side and the Galois
side. The picture is well-understood in the GL2 {Q case. Serre’s result can also
1
2 ALIREZA SHAVALI

be viewed as determining the image of the Galois representations associated with

weight 2 modular forms with rational Fourier coefficients. Later, Ribet realized that
for general modular forms of weight 2, the endomorphism of the associated abelian
variety being big, translates into the form having some kinds of symmetries which he
called inner-twists [18] and was able to generalize Serre’s result to this context [17].
Momose then worked out the higher weight case [13] and Nekovář generalized their
work to Hilbert modular forms [14]. The main goal of this paper is to generalize their
work to the GL3 -case. The main observation is that in this case, one should also take
into account the so-called ”outer-twists” of an automorphic representation. This is
due to the fact that not every automorphism of SL3 is inner. Inner- and outer-twists
of a representation together form a group that we call the group of extra-twists.
Let us state our main result. Let K be a totally real field and let π be a regular
algebraic cuspidal automorphic representation of GL3 pAK q that is of general type,
i.e. it neither satisfies π » π b χ for a non-trivial Hecke character χ, nor π » π _ b η
for any Hecke character η. Let Qpπq be its Hecke field with Galois closure E,
ρπ,p : GK Ñ GL3 pQpπq bQ Qp q the p-adic Galois representation attached to π and
Γ Ď AutpEq the group of extra-twists which will be defined later. Here is our main
result:
Theorem 1.1. Let F “ E Γ be the field fixed by all extra-twists of π. Then there
exists a finite extension L{K and a semi-simple algebraic group Hp defined over
Fp :“ F bQ Qp which is a form of SL3 (constructed using the extra-twists), such that
ρπ,p pGL q is contained in Hp pFp q ¨ Qˆ
p Ď GL3 pE bQ Qp q and it is open in the p-adic
topology.
F
In other words, for all p, the algebraic group pResQpp Hp q ¨ Gm,Qp is the connected
component of the Qp -Zariski closure of the image and the image is open in there.
Assuming the functoriality conjectures of Langlands, one can go through the
arguments in the proof of the above and see what assumptions are needed on π to
prove such a result for GLn , i.e. when are extra-twists enough to give a precise
description of the image. This should intuitively mean that π is not coming from
any smaller group via a Langlands transfer. We make this precise in Section 4 and
define automorphic representations of general type and prove a big image theorem
for the Galois representation associated to those, assuming Langlands functoriality.
The conjectures of Clozel [7], predict the existence of a motive Mπ over K (with
coefficients in an extension of Qpπq) attached to π. The Mumford-Tate conjecture for
this motive tells us that the groups Hp should come from a global object H defined
over the field F from Theorem 1.1. Assuming the existence of such a motive, we
will use the action of extra-twists on the Hodge structure to construct a group H8
over F bQ R that should be the Archimedean part of the Mumford-Tate group. We
will also use this action on the rational Hodge structure to construct a global group
H. This group will contain the (special) Mumford-Tate group and in particular its
dimension (which is equal to the dimension of all the groups Hp from the Theorem
1.1) bounds the dimension of the Mumford-Tate group from above.

Methods. We will use the group of extra-twists of the automorphic representation

π to define a 1-cocycle which gives the form H of SL3 . To prove the openness we
need to compare the Lie algebras. We first twist away the determinant and only
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 3

focus on the semi-simple part of the Lie algebra. We use automorphic base change
and induction for degree 3 extensions and regularity of the Galois representation to
show that the Galois representation is strongly irreducible and deduce that the Lie
algebra is reductive. Then, following Ribet, we base change to the algebraic closure
and use Goursat’s lemma. To apply Goursat’s lemma we need two things. First,
we will use the classification of semi-simple Lie subalgebras of sl3 and Langlands
functoriality for sym2 : GL2 Ñ GL3 . In the GL3 -case, being essentially sym2 and
essentially self-dual are equivalent. This is important since it enables us to go back
and forth between the Lie algebras and the groups. Second, we need a description
of the field fixed by inner-twists and the field fixed by all extra-twists of π in terms
of the coefficients of the characteristic polynomial of the Galois representation at
Frobenius elements, which is provided in Lemma 3.5. At the end we can twist
back the determinant and give a complete description of the image of the Galois
representation up to p-adic openness.

The structure of this paper. In Section 2, we define inner- and outer-twists

of automorphic representations and Galois representations for GLn and study their
basic properties. We will also give the construction of the group H from theorem
1.1 in an abstract setting.
In Section 3, we consider a general n-dimensional Galois representation with trivial
determinant satisfying a set of natural properties and compute the Qp -Lie algebra
of its image, assuming that the Qp -Lie algebra is big.
In Section 4, we apply these results to Galois representations coming from certain
automorphic representations and prove our main theorem. We also show that if
one assumes enough conjectures from the Langlands program, then this could be
generalized to a conjecture for the GLn -case.
Finally, in Section 5, assuming the existence of a motive attached to π we study
the relation of the extra-twists with the Mumford-Tate group of this motive and
propose some conjectures.

Acknowledgments. First, I want to thank my advisor Gebhard Böckle for suggest-

ing that I work on this problem and his constant support during my PhD studies.
I am furthermore grateful to Andrea Conti and Judith Ludwig for many helpful
conversations. I would also like to thank Gaëtan Chenevier for suggesting Lemma
4.1, which simplified some of the arguments, and Chun Yin Hui for suggesting part
4 of Remark 3.2, which resulted in removing a density one condition on an earlier
version of this work. This research was supported by the Deutsche Forschungsge-
meinschaft (DFG) through the Collaborative Research Centre TRR 326 Geometry
and Arithmetic of Uniformized Structures, project number 444845124.

2. Extra-Twists for GLn

In this section we define extra-twists for certain automorphic representation and
Galois representations and then give a construction of a form of an algebraic group
from a 1-cocycle.
Let K be a number field and let π be a cuspidal automorphic representation of
GLn pAK q such that:
‚ π is not self-twist, i.e. there is no Hecke character χ ‰ 1 such that π » π b χ.
4 ALIREZA SHAVALI

‚ If n ą 2, π is not essentially self-dual, i.e. there is no Hecke character η such

that π » π _ b η.
These two conditions are analogues to the condition of a modular form being non-
CM in the work of Ribet and Momose. It turns out that these conditions are enough
in the GL3 -case to have a big image theorem for the associated Galois representation
as we will see later, but of course not in the GLn -case.
Let Qpπq Ă C be the Hecke (number) field of π and let E be a number field
containing Qpπq. In what follows we will use strong multiplicity-one for cuspidal
automorphic representations of GLn , many times without mentioning it.

2.1. Inner and outer twists.

Definition 2.1. An (E-)extra-twist of the automorphic representation π is either
of the following two:
(1) (An inner-twist) A pair pσ, χq where σ P AutpEq and χ : Aˆ ˆ ˆ
K {K Ñ C is a
σ
Hecke character, such that π – π b χ.
(2) (An outer-twist) A pair pτ, ηq where τ P AutpEq and η : Aˆ ˆ ˆ
K {K Ñ C is a
τ _
Hecke character, such that π – π b η.
Remark 2.2. We make three remarks about this definition.
(a) The role of E might seem a bit auxiliary and one might think it should be
enough to take E “ Qpπq. But, making this slightly more general definition
will help us on the Galois side when dealing with issues regarding the field
of definition of automorphic Galois representations. Also it will be more
convenient in sections 3 and 4 to assume that E is Galois over Q. Since we
will always fix E to begin with, we will usually drop it from the notation.
(b) Notice that the Galois action in the above definition is on the coefficients. In
particular, do not confuse an outer-twist with an essential conjugate self-dual
of an automorphic representation over a CM field (e.g. as in [2]).
(c) For a general reductive group, there should be a class of extra-twists for every
automorphism of a (fixed) based root datum. This would also make sense
on the Galois side since the automorphism group of the dual root datum is
clearly the same.
One can similarly define the notion of extra-twists for Galois representations.
ś Let
E be a number field, p a (rational) prime number and let Ep “ E bQ Qp – p|p Ep .
Assume we have a continuous irreducible Galois representation:
ź ź
ρp “ ρ : GK Ñ GLn pEp q “ GLn pEp q
p|p p|p

unramified outside a finite set of places. We also assume that ρ is neither self-twist
nor essentially self dual in the n ą 2 case, i.e. it neither satisfies ρ » ρ b χ for
a non-trivial Galois character χ, nor ρ » ρ_ b η for any Galois character η in the
n ą 2 case.
Definition 2.3. An inner-twist of ρ is a pair pσ, χq where σ P AutpEq and χ : GK Ñ
Ep ˆ is a (continuous) Galois character, such that σ ρ – ρ b χ. An outer-twist of ρ is
a pair pτ, ηq where τ P AutpEq and η a Galois character, such that τ ρ – ρ_ b η. An
extra-twist of ρ is either an inner- or an outer-twist.
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 5

Remark 2.4. Note that ρ_ is just isomorphic to the representation ρ´T and hence
has coefficients in Ep (and not just E bQ Qp ). This easily implies that the characters
χ appearing in the extra-twists are forced to have values in Epˆ and we do not lose
any generality by making this assumption in the definition.
From now on, we assume that K is totally real and π is a regular algebraic
cuspidal automorphic representation. Then it is known ([10], [20]) that there exists
a compatible family of Galois representations ρπ,p associated with π. We will see
in lemma 4.1 that this compatible family can be defined over a coefficient field E
of finite degree and Galois over Q. Then we get a bijection between the set of E-
extra-twists of π and E-extra-twists of ρ :“ ρπ,p . Therefore, we usually identify the
two.

The most basic properties of the extra-twists of ρ (or π) are summarized in the next
lemma:
Lemma 2.5. Let K be totally real and ρ : GK Ñ GLn pEp q be a p-adic Galois
representation that is neither self-twist nor essentially self-dual in the n ą 2 case.
Then extra-twists of ρ satisfy the following properties:
(i) For an extra-twist pσ, χq, the automorphism σ uniquely determines the char-
acter χ.
(ii) If pσ, χq is an inner-twist and pτ, ηq an outer-twist, then σ ‰ τ .
(iii) Extra-twists form a group under the operation pσ, χq ˝ pτ, ηq :“ pσ ˝ τ, χ ¨ σ ηq.
(iv) Inner-twists form a subgroup of the group of all extra-twists. If at least one
outer-twist exists then this is an index 2 subgroup.
(v) If ρ is associated with an algebraic automorphic representation π then for
any inner-twist pσ, χq, the character χ is finite.
Proof. First assume that pσ, χ1 q and pσ, χ2 q are both inner-twists. Then ρ » ρ b
χ1 χ´1
2 which implies χ1 “ χ2 by our assumptions on ρ. A similar argument proves
the other cases of (i) and also (ii). For part (iii) let assume that both pσ, χq and
pτ, ηq are inner-twists. Then
σ τ
p ρq – σ pρ b ηq “ σ ρ b σ η – ρ b χ b σ η
The other cases are similar. For part (iv) one just needs to notice that the product
of two outer-twists is clearly an inner-twist. Finally, to see (v), note that since
K is totally real, the central character of π must be of the form | ¨ |m ω for some
m P Z and finite order character ω. Hence detpρq “ ǫm p ω where ǫp is the p-adic
cyclotomic character and we are viewing ω as a finite Galois character. Now, taking
the determinant of both sides of σ ρ – ρ b χ we get
σ detpρq σω
χn “ “
detpρq ω
which implies that χn (and hence χ) is a finite order character. 
We will denote the group of all extra-twists of a Galois representation (or an
automorphic representation) by Γ and the subgroup of inner-twists by Γinn and the
set of outer-twists by Γout , if there is any. The last lemma shows that we can identify
Γ with a subgroup of AutpEq by forgetting the character and we will do so from now
inn
on. Let F :“ E Γ be the field fixed by all the extra-twists and F inn :“ E Γ be the
6 ALIREZA SHAVALI

field fixed by the inner-twists. In particular, Γ “ GalpE{F q and Γinn “ GalpE{F inn q.
If there is at least one outer-twist then rF inn : F s “ 2, otherwise F “ F inn .
2.2. Constructing forms of algebraic groups using 1-cocycles. In this section
we explain the abstract construction of a form of an algebraic group from a 1-cocycle
of a Galois action of the automorphisms of the group. This will be used later for
the group SLn and a cocycle that comes from the group of extra-twists of a Galois
representation.
Let E{F either be a finite Galois extension of fields or the semi-local Galois
extension Ep {Fp “ pE{F qbQ Qp for a finite Galois extension E “ F pαq{F of number
fields and let Φ be the minimal polynomial of α. Let G{E be an algebraic group
and let Γ “ GalpE{F q. Let f : Γ Ñ AutE pGq be a 1-cocycle and write fσ for the
image of σ. We will construct a form of G defined over F using this cocycle.
Let G1 “ ResEF G, so for every F -algebra R we have
G1 pRq “ GpE bF Rq
Therefore, G1 pRq is equipped with an action of Γ (where it acts on the first com-
ponent of the tensor product) which is clearly functorial. Hence the collection of
morphisms σ : G1 pRq Ñ G1 pRq is a natural transformation and so it is induced from
a morphism σ : G1 Ñ G1 of algebraic groups. Therefore, we can define H to be the
subgroup of G1 satisfying fσ pσ gq “ g for every σ P Γ. In other words H “ pG1 qtwf pΓq
where we define the f -twisted action of σ on G1 to be given by twf pσq g “ fσ pσ gq.
This gives a closed subgroup of G1 .
By base changing H Ď ResE F G to E and then projecting to the id-component one
gets ź πid
HE ãÑ pResE F GqE “ GE ÝÝÑ G
Γ
We prove that this is an isomorphsim by checking thisśon points. First, we need
to give a description of the algebraic action of Γ on Γ G via the identification
ś
pResE
F GqE “
E
Γ G. Let R be an E-algebra. Note that ResF GpRq “ GpE bF Rq
and the algebraic action of Γ is just the action on the E component. Now
GpE bF Rq “ GpE bF E bE Rq
and the action is only on the first E component. Then
ˆ ˙ ˆ ˙
F rxs Erxs
GpE bF Rq “ GpE bF E bE Rq “ G E bF bE R “ G bE R
Φpxq Φpxq
where the action is only on the coefficients of the first component. So
ˆ ˙ ˜˜ ¸ ¸ ˜ ¸
Erxs ź ź ź
“G ś bE R “ G E bE R “ G R “ GpRq
Γ px ´ σαq Γ Γ Γ
where the action of γ P Γ is given by
paσ qσ ÞÑ pγaγ ´1 σ qσ
Proposition 2.6. With the notation as above, for any E-algebra R the map
ź
HpRq ãÑ ResE F GpRq “ GpRq Ñ GpRq
Γ
in an isomorphism of groups, so the algebraic group H is a form of G.
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 7

Proof. By definition, HpRq is theś subgroup of the elements invariant under the
twisted action of Γ. Let pgσ qσ P GpRq be invariant under the twisted action:
twf pγq
pgσ qσ “ pfγ pγ gγ ´1 σ qqσ
By looking at the component σ “ γ we get
gγ “ fγ pγ g1 q
So the g1 component determines all other gγ ’s. This shows that the map
ź π
HpRq ãÑ GpRq ÝÝid Ñ GpRq
Γ

is injective.
To prove the surjectivety, we need to show that for every g1 P GpRq the element
pfσ pσ g1 qqσ is invariant under the twisted action:
twf pγq ´1 σ
pfσ pσ g1 qqσ “ pfγ pγ pfγ ´1 σ pγ g1 qqqqσ “ ppfγ ˝ γ fγ ´1 σ qpσ g1 qqσ
Now, by the cocycle condition
fγ ˝ γ fγ ´1 σ “ fσ
therefore
twpγq
pfσ pσ g1 qqσ “ pfσ pσ g1 qqσ
which is exactly what we needed to prove. So the map is an isomorphism for all the
R-points and hence an isomorphism of affine algebraic groups over E. 

We can generalize this theorem to understand the behavior of H under any Galois
base change of F .
Corollary 2.7. Let F Ă F0 Ă E be an intermediate field that is Galois over F and
let Γ0 “ GalpE{F0 q and f0 : Γ0 Ñ AutE pGq be the restriction of f to Γ0 . Then
twf0 pΓ0 q
H ˆF F0 » pResE
F0 Gq

twf0 pΓ0 q
Proof. Apply proposition 2.6 to the group pResE
F0 Gq for the twisted action of
the group Γ{Γ0 . 

3. Image of Galois Representations with Extra-Twists

The extra-twists of an algebraic automorphic representation induce extra-twists
on the associated Galois representation. In this section we fix a p-adic Galois rep-
resentation satisfying a list of natural properties (including the property that the
Qp -Lie algebra of the image is big) and use these extra-twists to determine the Qp -
Lie algebra of the image of this Galois representation. In the next section, we will
apply the results of this section to Galois representation associated to certain auto-
morphic representations. We will need to use some automorphic input to show that
in the 3-dimensional case this list of properties is satisfied.
Let K be a number field as usual and GK its absolute Galois group. Let E be
another number field that is assumed to be Galois over Q and let Ep “ E bQ Qp –
8 ALIREZA SHAVALI
ś
p|p Ep .
Assume that for each finite place p|p of E, we have a continuous semi-
simple Galois representation ρp : GK Ñ GLn pEp q. It is usually more convenient to
work with the product of all these Galois representations:
ź ź
ρp “ ρp : GK Ñ GLn pEp q “ GLn pEp q
p|p p|p

or equivalently a free Ep -module Vp of rank n, with a continuous Galois action on

À space Vp “ Vp bEp Ep over Ep with a
it and for each p|p an n-dimensional vector
continuous Galois action such that Vp “ p|p Vp as Qp -vector spaces.
Each embedding λ : E ãÑ Qp induces an absolute value and hence gives a finite
place p of E above p. Therefore, λ extends to an embedding λ : Ep ãÑ Qp by
continuity. Now we define
Vλ :“ Vp bEp ,λbid Qp “ Vp bEp ,λ Qp
that is an n-dimensional vector space over Qp with a continuous Galois action. We
denote this representation by ρλ . Note that ρλ : GK Ñ GLn pQp q is essentially the
same object as ρp : GK Ñ GLn pEp q, it is just considered with coefficients in Qp
instead of Ep via λ.
Now, we need to make the following list of natural assumptions of our Galois repre-
sentations to be able to compute the Lie algebra of the image in the next subsection.
These properties are expected to hold for Galois representations attached to regular
cuspidal algebraic automorphic representations of general type, after possibly some
finite base change and twist by a character.
Definition 3.1. Keeping the above notations, the Galois representation ρp : GK Ñ
GLn pEp q is called valid if
‚ Each ρλ is continuous and unramified outside a finite set S of places of K
containing the Archimedean places and all places above p.
ppq
‚ fv pxq :“ CharPolypρp pFrobv qq has coefficients in E, for each place v R S.
‚ Each ρλ is neither self-twist nor essentially self-dual for n ą 2.
‚ detpρp q is trivial.
‚ For each λ the Qp -Lie algebra of the p-adic Lie group ρλ pGK q is equal to
sln pQp q.
Remark 3.2. We make the following four remarks about this definition:
(1) Note that by the last condition, each ρλ is strongly irreducible, i.e. the
restriction of ρλ to GL , the absolute Galois group of L, is irreducible for any
finite extension L{K. This is simply because going to a finite extension does
not affect the Lie-algebra. In practice, we will usually need to prove this first
in order to show that a Galois representation is valid.
(2) Note that for n “ 2 we are not excluding essential self-duality but we are
excluding being self-twist.
(3) The condition on the determinant is not very restrictive because we can
trivialize the determinant after a finite extension of K and a twist. Since
our first goal is to compute the semisimple part of the Lie algebra of the
image, it doesn’t change anything if we restrict to an open subgroup and
also twist with a character.
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 9

(4) In the case where ρλ ’s come from a compatible family of semi-simple Galois
representations, it is enough to check the last condition at only one λ. More
precisely, by Theorem 3.19 and Remark 3.22 of [11], the semi-simple rank
and the formal character of the tautological representations of the algebraic
monodromy group are independent of λ. Then [12, Theorem 4] implies that
this uniquely determines the Lie algebra in the type An case hence if we have
sln at one place λ, we should have sln at every place.
3.1. Extra-twists and Galois representations. From now on we assume that ρp
is a valid Galois representation. There are two cases that we have to deal with. ρp
either has an outer-twist or it does not. Notice that in the n “ 2 case we only have
inner-twists since every representation is essentially self-dual. We try to deal with
both cases at the same time but at some places it is easier to make a distinction.
We need a few lemmas:
Lemma 3.3. For every extra-twist pσ, χσ q of ρp , one has that χσ is a finite character.
Proof. This is always true for inner-twists as we saw in lemma 2.5. Let pσ, χσ q be
an outer-twist. Then
σ
ρp – ρ_p b χσ
Looking at the determinants of both side we see that χnσ “ 1 which implies the
result. 
Lemma 3.4. Let L be a finite Galois extension of K and let V1 and V2 be p-
adic finite dimensional GK -representations such that the restriction of V2 to GL is
absolutely irreducible. If V1 » V2 as representations of GL , then V1 » V2 b φ as
representations of GK for some character φ of GK .
Proof. we can choose two bases for V1 and V2 such that the representations ρ1 :
GK Ñ GLn pQp q and ρ2 : GK Ñ GLn pQp q associated with V1 and V2 are isomorphic
when restricted to GL . Now define
φpgq :“ ρ´1
1 pgqρ2 pgq
A priori φ is just a map φ : GK Ñ GLn pQp q which is trivial on GL . We want to
prove that it is actually a homomorphism with values in the center (hence actually
a character) on all of GK .
Let g P GK and h P GL . Note that ρ1 phq “ ρ2 phq and ρ1 pghg ´1 q “ ρ2 pghg ´1 q
since GL is normal in GK . Now the following computation shows that φpgq “
ρ´1
1 pgqρ2 pgq commutes with ρ2 phq:
ρ´1 ´1 ´1
1 pgqρ2 pgqρ2 phq “ ρ1 pgqρ2 pghq “ ρ1 pgqρ2 pghg
´1
qρ2 pgq
“ ρ1 pg´1 qρ1 pghg´1 qρ2 pgq “ ρ1 phqρ´1 ´1
1 pgqρ2 pgq “ ρ2 phqρ1 pgqρ2 pgq
Now since V2 is absolutely irreducible when restricted to GL , we have EndGL pV2 q “
Qp and we are done. 
Now fix a valid Galois representation ρp . Note that by the last assumption in
definition 3.1, it makes sense to consider the group Γ Ď AutpEq of all the extra-
inn
twists of ρp . Let Γinn , F “ E Γ , and F inn “ E Γ be as in section 2.1. By lemma
2.5, the character χ in an extra-twist pσ, χq is uniquely determined by σ so we use
the notation χσ for this character.
10 ALIREZA SHAVALI

We assumed in definition 3.1 that we know the Qp -Lie-algebra of the image. The
next two lemmas are our main tool to compute the Qp -Lie-algebra. The next lemma
is the only place that we will use the assumption that E{Q is Galois.
Lemma 3.5. Let GL “ XσPΓ kerpχσ q and let L1 {L be a finite extension that is
Galois over K. For every finite unramified place v of L1 let av , bv P E such that
fv pxq :“ charPolypρp pFrv qq “ X n ´ av X n´1 ` ¨ ¨ ¨ ` p´1qn´1 bv X ` p´1qn . Then,
F inn “ Qptav uv q and F “ Qptav ` bv uv q.
Proof. We first prove that fv pxq P F inn rxs. This is because for any inner-twist pσ, χq,
the character χ becomes trivial after restricting to GL1 so ρp |GL1 – σ ρp |GL1 which
means fv is invariant under the action of Γinn which implies the result. Now if pτ, ηq
is an outer-twist, after restriction to GL1 one has ρp |GL1 – τ ρp´T |GL1 . Looking at
the characteristic polynomials of Frv of both sides one gets X n ´ av X n´1 ` ¨ ¨ ¨ `
p´1qn´1 bv X ` p´1qn “ X n ´ τ bv X n´1 ` ¨ ¨ ¨ ` p´1qn´1τ av X ` p´1qn . In particular,
av ` bv is invariant under the outer-twists as well. This implies av ` bv P F .
Now, let F 1 “ Qptav uv q Ď F inn . We want to prove that F 1 is the field fixed
by the inner-twists. This suffices because E was assumed to be Galois over Q and
hence it is Galois over F 1 . So it’s enough to construct an inner-twist of ρp for every
σ P GalpQ{F 1 q. Now, note that ρp |GL1 has traces in F 1 so ρp |GL1 and σ ρp |GL1 have
the same traces and since they are semisimple, they must be conjugate. Now, by
lemma 3.4 there exists a character such that ρp b χ » σ ρp so we are done.
At last, let F 2 “ Qptav `bv uv q Ď F inn . We want to prove that F 2 is the field fixed
by all extra-twists. So it’s enough to construct an inner-twist or an outer-twist of ρp
for every σ P GalpQ{F 2 q. Now, note that pρp ‘ρ´T 2
p q|GL1 has traces in F because the
trace of the image of Frv would be av ` bv . So pρp ‘ ρ´T σ ´T
p q|GL1 and pρp ‘ ρp q|GL1
have the same traces and since they are semisimple, should be conjugate. Now
by the strong irreducibility assumption, we must either have ρp |GL1 » σ ρp |GL1 or
ρp |GL1 » σ ρ´Tp |GL1 . Then by lemma 3.4 we either get an inner-twist or an outer-
twist. 
Lemma 3.6. Let λ, µ : E ãÑ Qp be two places of E above p and let GL “
XσPΓ kerpχσ q and L1 a finite extension of L that is Galois over K. Then in the
case that there are outer-twists one has:
i. Vλ » Vµ as representations of GL1 if and only if λ|F inn “ µ|F inn
ii. Vλ » Vµ_ as representations of GL1 if and only if λ|F “ µ|F but λ|F inn ‰
µ|F inn
and in the case that there are no outer-twists one has that F “ F inn “ Qptav uv q
and part (i) of the above is true and part (ii) never occurs.
Proof. Since all our representations are semi-simple, it is enough to check equality
on characteristic polynomials and since they are continuous it is enough to check
this on a dense subset. We check this on the Frobenius elements of finite places of
L1 at which ρp is unramified.
Keeping the notation of lemma 3.5, the characteristic polynomial of Frv acting on
Vλ is X n ´λpav qX n´1 `¨ ¨ ¨`p´1qn´1 λpbv qX `p´1qn and on Vµ is X n ´µpav qX n´1 `
¨ ¨ ¨ ` p´1qn´1 µpbv qX ` p´1qn . Since av ’s generate F inn be lemma 3.5, part i follows.
For part ii, notice that the characteristic polynomial of Frv acting on Vµ_ is X n ´
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 11

µpbv qX n´1 ` ¨ ¨ ¨ ` p´1qn´1 µpav qX ` p´1qn . So if Vλ » Vµ_ then λpav q “ µpbv q and
λpbv q “ µpav q which implies that λpav ` bv q “ µpav ` bv q and hence λ|F “ µ|F .
On the other hand, Vλ is not essentially self-dual so by lemma 3.4 its restriction to
GL1 cannot be self-dual. So Vλ and Vµ are not allowed to be isomorphic as GL1 -
representations in this case, hence λ|F inn ‰ µ|F inn . The other direction also follows
easily since the characteristic polynomials of Frobenius elements at unramified places
clearly match. 

3.2. The Lie algebra of the image. Now, we want to compute the Qp -Lie-algebra
of the image of ρp . First, we need to use the results of section 2.2 to construct the
right algebraic group which contains the image and then compare the Lie-algebra of
the image with the (algebraic) Lie-algebra of this group.
Recall that we assumed that ρp has trivial determinant. Therefore we have

ρp : GK Ñ SLn pEp q

We first define a 1-cocycle f : Γ Ñ AutEp pSLn q using extra-twists. For every inner-
twist σ P Γinn one has that ρp |GL and σ ρp |GL have the same trace. Since each
ρλ is strongly irreducible, this means that they are isomorphic and there exists
ασ P SLn pEp q such that ρp |GL “ ασ ¨ σ ρp |GL ¨ α´1 σ . For the inner-twist pσ, χσ q
we define fσ “ adpασ q. If τ P Γout is an outer-twist (if there exist any) then
ρp |GL and τ ρ´T
p |GL have the same traces and there exists ατ P SLn pEp q such that
ρp |GL “ ατ ¨ ρ´T
τ ´1
p |GL ¨ ατ . For the outer-twist pτ, χτ q we define fτ “ adpατ q ˝ p¨q
´T .

One can easily check that f : Γ Ñ AutEp pSLn q defined above is a 1-cocycle.
Now, as in section 2.2 we can define the twisted action of Γ by this cocycle on
SLn . From the construction of f , it is clear that every matrix in the image of GL is
E
invariant under this twisted action.Let H “ pResFpp SLn qtwf pΓq that is an algebraic
group over Fp . Then it follows:

Corollary 3.7. The representation ρp |GL factors through HpFp q Ď SLn pEp q, i.e.
ρp pGL q Ď HpFp q.

Note that by proposition 2.6, H is a form of SLn and in particular is a semi-

simple group. Also, note that since GL is open in GK , the Lie-algebra of the p-adic
Lie groups ρp pGK q and ρp pGL q are the same. Let g be the Lie algebra (over Qp )
of ρp pGL q and h be the Lie algebra of the algebraic group H{Fp , both viewed as
Lie-subalgebras of SLn pEp q. Our next goal is to show that these two Lie-algebras
are in fact equal.

Proposition 3.8. With the notation as above, g “ h.

Proof. First note that g Ď h by the last corollary. Since h is semi-simple, it suffices
to prove that gder “ gder bQp Qp is equal to h “ h bQp Qp .
For every embedding σ : F ãÑ Qp fix an extension σ r : E ãÑ Qp of σ. All of the
other extensions of σ can be obtained by composing with different elements of the
Galois group Γ “ GalpE{F q, i.e. are of the form σrτ for τ P Γ. Now we base change
our representation ρp |GL to Qp to get:
12 ALIREZA SHAVALI

“
ρp : GL SLn pEp q SLn pEp q ResE
F pSLn qpFp q

“
HpFp q HpFp q ResE
F pSLn q
twpΓq pF q
p

ś
where Ep :“ Ep bQp Qp and Fp :“ Fp bQp Qp “ σ:F ãÑQp Qp . Note that we have
ź ź
ResE
F pSLn qpFp q » SLn pE bF,σ Qp q » SLn pE bF E bE,rσ Qp q
σ:F ãÑQp σ:F ãÑQp
ź ź ź
» SLn pQp q » SLn pQp q
σ:F ãÑQp Γ λ:EãÑQp

where λ “ σ rτ for τ P Γ. By proposition 2.6, projection to the id-component of Γ

gives the isomorphism of the form ResEF pSLn q
twpΓq of SL with SL over E so we
n n p
have:
ś ś “ ś
ρp : GL SLn pEp q σ:F ãÑQp Γ SLn pQp q λ:EãÑQp SLn pQp q

ś
HpFp q σ:F ãÑQp SLn pQp q

For each embedding σ : F ãÑ Qp the composition

ź prσ
ρσ : GL Ñ HpFp q ãÑ SLn pQp q ÝÝÑ SLn pQp q
σ:F ãÑQp

corresponded to the action of GL on the vector space Vλ for some embedding λ :

E ãÑ Qp extending σ. Note that by lemma 3.6 these Vσ ’s are neither isomorphic
nor dual to each other after any finite extension. This is the main point of the rest
of the argument.
On the level of Lie algebras this gives the embedding:
»
ź
gder Ď h ÝÑ sln pQp q
σ:F ãÑQp

Let gder
σ Ď sln pQp q be the projection of g
der to the σ-component in of the above map.

This is the Qp -Lie-algebra of the image of the representation ρσ (=ρλ for some λ
extending σ), so by our assumption on ρp being valid, we have gder σ “ sln pQp q.
Now we can apply [17, Lemma 4.6] to
ź
gder Ď h Ñ sln pQp q
σ:F ãÑQp

We only need to prove that for every σ, τ : F ãÑ Qp the projections pprσ ˆ prτ qpgder q
and pprσ ˆ prτ qphq are equal. We follow the arguments of [21, §6.2].
Clearly it’s enough to show that pprσ ˆ prτ qpgder q “ sln pQp q ˆ sln pQp q. Note that
the first factor corresponds to the representation Vσ and the second to Vτ . Now by
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 13

the Lie algebra version of the Goursat’s theorem [16, Lemma 5.2.1] if pprσ ˆprτ qpgder q
is not equal to sln pQp q ˆ sln pQp q then it has to be the graph of an isomorphism.
Let’s call this automorphism φ : sln pQp q Ñ sln pQp q. Since sln is simple the group
of its outer automorphisms is the group of automorphisms of its Dynkin diagram
Z
which is trivial for n “ 2 and is isomorphic to 2Z in the n ą 2 case. In this case, the
class of this non-trivial outer automorphism is given by the map X ÞÑ ´X T . So φ
is either an inner automorphism or a conjugate of this outer automorphism.
First assume φ is an inner automorphism and is given by conjugation by some
matrix α. In other words, we have the following diagram:

sln pQp q
prσ

g φ“adpαq
prτ

sln pQp q

which means that Vσ and Vτ are isomorphic as representations of g. This implies that
they are indeed isomorphic as representations of some open subgroup of GL which
contradicts Lemma 3.6. Now assume that φ is a conjugate of X ÞÑ ´X T . Similarly,
this means that Vσ – Vτ_ as representations of some small enough normal open
subgroups of GL which again contradicts lemma 3.6. This implies the result. 
Corollary 3.9. The image of GL under ρp is an open subgroup of the p-adic Lie
group HpFp q.
If ρp has no outer-twists, then the cocycle f is always defined by an inner au-
tomorphism and H is an inner-form of SLn . If there is at least one outer-twist
then this is not true anymore, but H becomes an inner-form after a degree two
extension. In fact, the restriction of f to the index two subgroup Γinn factors
through InnEp pSLn q Ă AutEp pSLn q. Then by corollary 2.7, the base change of
H to Fpinn “ F inn bQ Qp is an inner-form. In any case we have:
Corollary 3.10. The group HFpinn , the base change of H to Fpinn , is an inner-form
of the group SLn which splits over Ep . Moreover, if p splits in the extension Fpinn {Fp
(p splits in F inn ”more” than it does it F ) then H is an inner-form of SLn .
Proof. The first part follows from the discussion above. For the second part, if there
are no outer-twists then there is nothing to prove. Otherwise, notice that when p
splits, Fpinn » Fp ˆ Fp so if H becomes an inner-form over Fpinn , it was already an
inner-form over F inn . 

4. Image of Automorphic Galois Representations

In this section we will apply the results of the last section to Galois representations
attached to certain automorphic representations. In the case of GL2 all the results
are already known but we include this case as well, for the sake of completeness.
Throughout this section, we assume that K is a totally real number field and π is a
regular cuspidal algebraic automorphic representation of GLn pAK q. It is known by
14 ALIREZA SHAVALI

the work of Harris-Lan-Taylor-Thorne [10] or Scholze [20], that there is a compatible

family of p-adic Galois representations associated with π. Our goal is to understand
the image of these representations.
Let |.|m ω be the central character of π where ω is a finite order Hecke character
and m ě 1 an integer. Then for each embedding λ : Qpπq ãÑ Qp there exist a
continuous semi-simple Galois representation
ρπ,λ : GK Ñ GLn pQp q
that is an unramified Galois representaton for all unramified places of π not above
p and at all these places like v, the characteristic polynomial of Frobv is deter-
mined by the Satake parameters of π at v. In other words, these representations
form a compatible family of Galois representations in the sense of Serre. Moreover,
detpρπ,λ q “ pλ ˝ ωq ¨ ǫm
p where ǫp is the (Global p-adic) cyclotomic character and we
are regarding the finite order character ω as a Galois character via class field theory.
It is not known if we can conjugate these Galois representations to have values
in the completions of the Hecke field Qpπq. Nevertheless, we will show that we can
do this for a finite extension E of Qpπq. This is exactly the reason why we defined
extra-twists for any Galois coefficient field containing the Hecke field in definition
2.1.
The Galois representations ρπ,λ are expected to be irreducible since π is cuspidal.
Let tπ,λ : GK Ñ Qp be the trace of ρπ,λ . This is an irreducible pseudo-representation
and it clearly takes values in Qpπqλ . Then by a result of Rouquier [19, Theorem
5.1] (or more generally Chenevier [6, corollary 2.23]) there exists a central simple
algebra Dλ over Qpπqλ of dimension n2 such that this pseudo-representation can be
realized as the reduced trace of a representation GK Ñ Dλˆ . The base change of this
representation to Qp clearly gives back ρπ,λ because of the Brauer-Nesbitt theorem.
In other words, the image of ρπ,λ is in fact in Dλˆ :
ρπ,λ : GK Ñ Dλˆ Ă Dλˆ bQpπqλ Qp » GLn pQp q
Lemma 4.1 (Chenevier). Assume that ρπ,λ is irreducible for all λ and regular for
at least one λ. Then, there exists a finite extension E 1 {Qpπq that is Galois over Q,
such that for all but finitely many finite places λ of Qpπq and any place µ of E 1 above
λ, the central simple algebra Dλ splits over Eµ1 . In particular, there exists a finite
extension E{Qpπq such that all representations ρπ,λ can be defined over E (can be
conjugated to have values in completions of E).
Proof. First recall that a central simple algebra Dλ Ă Mn pQp q splits in an extension
M of Eλ if and only if it contains an element with n pairwise distinct eigenvalues
in M . Let v be a place of K at which π is unramified and f pvq pxq P Qpπqrxs be the
characteristic polynomial of the Frobenius element at v, which is independent of λ.
Then as in the proof of [2, Lemma 5.3.1], choosing a λ for which ρπ,λ is regular, we
get that for infinitely many places v one has that f pvq pxq has distinct roots. This
shows that if E 1 is the splitting field of f pvq pxq, then Dλ splits over the completion
of E 1 at any finite place coprime to v and the level of π. Since it clearly splits over
some finite extension of Eλ as well, we can find a number field E which splits all
Dλ ’s at the same time. At the end we take the Galois closure over Q.

 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 15

Remark 4.2. A natural question that arises after this lemma is if one should expect
Dλ ’s to come from a global object D{Qpπq. We will discuss this more in the section
5. In particular in the special case of n ď 3 this follows from our results, existence
of a motive associated to π and the Mumford-Tate conjecture for that motive.
From now on we take E to be the number field coming from the last lemma and
we take our Galois representations to have values in GLn pEp q for finite primes p of
E. So we are in the setting of section 3 and we define:
ź ź
ρπ,p “ ρπ,p : GK Ñ GLn pEp q “ GLn pEp q
p|p p|p
ś
where Ep “ E bQ Qp – p|p Ep as usual.
From now on, we assume that π is neither self-twist nor essentially self-dual in the
n ą 2 case. Then it makes sense to talk about E-extra-twists of π. Since E is fixed
we will drop it from the notation from now on. By multiplicity one, inner-twists of
π and ρπ,p agree (we are using class field theory to identify the characters). So let
Γ, Γinn , Γout , F and F inn be as in section 2.
The determinant of ρp is given by ω ¨ ǫm p . To apply the results of the last section,
we first need to kill the determinant. This is always possible after a finite extension.
In fact, after a finite extension the cyclotomic character will have values in 1 ` pZp
and then we can use the p-adic logarithm. Therefore, there exists a finite extension
M {K such that ǫp |GM has an n’th root. We fix one of these characters and denote
1{n
it by ǫp . Now, we enlarge M to trivialize ω if necessary. Then
ρ1π,p :“ ρπ,p |GM b ǫ´m{n
p : GM Ñ GLn pEp q
has trivial determinant. This is the Galois representation that we will apply our
results from the last section to. Notice that extra-twists of ρ1π,p and ρπ,p are the
same. More precisely, the characters might have changed after the twist but the
group Γ Ď AutpEq has not. It is also not hard to see how the Lie algebra of the
image changes:
Lemma 4.3. Assume that ρ1π,p is valid. Let g1 :“ Liepρ1π,p pGM qq and g “ Liepρπ,p pGL qq.
Then gder “ g1 .
Proof. Let G “ ρπ,p pGM q and G1 “ ρ1π,p pGM q and let Z be the center of GLn pEp q.
Then we clearly have G Ď G1 ¨ Z and G1 Ď G ¨ Z. Taking the Lie algebras we find
that g Ď g1 ` z and g1 Ď g ` z where z is the Lie algebra of the center. Since g1 is
semi-simple by proposition 3.8, this implies that g is reductive and gder “ g1 . 
In the case that ρ1π,p is valid, let Hp {Fp be the semi-simple group from corollary
3.7 applied to ρ1π,p . Then we have:
Proposition 4.4. If ρ1π,p is valid then there exists a finite extension L of K such
that such that ρπ,p pGL q is contained and p-adically open in Hp pFp q ¨ Qˆ
p Ď GLn pEp q.

Proof. Since the image of ρ1π,p is contained in Hp pFp q after a finite extension, and
1{n
the image of ǫp is in Zˆ ˆ
p , the image of ρπ,p is contained in Hp pFp q ¨ Qp after a finite
extension. The image is open in Hp Ď SLn pEp q by lemma 4.3 and the image of the
determinant is open in Qˆp , so we are done. 
16 ALIREZA SHAVALI

This in particular implies that the connected component of the Qp -Zariski closure
F
of the image is the algebraic group pResQpp Hp q ¨ Gm,Qp . Now, we only need to check
validity of ρ1π,p to deduce that the image is big.

4.1. The GL2 -case. As we mentioned in the introduction, essentially everything is

known in this case by the work of Ribet [17], Momose [13] and Nekovar [14]. We
repeat the arguments for the sake of completeness. In this case all representations
are essentially self dual so there are no outer-twists and Γ “ Γinn . One can in fact
take E “ Qpπq (then Γ would be abelian), but it is not necessary for our discussion.
Recall the we assumed that π is not self-twist, i.e. does not satisfy π » π b χ
for χ ‰ 1. In this case it is more common to say that π does not have complex
multiplication (CM).
Proposition 4.5. Assume that n “ 2 and π does not have CM. Then ρ1π,p is valid.
Proof. Most of the properties are clear from the analogous properties for ρπ,p . We
only need to check strong irreducibility and compute the Qp -Lie algebra.
It is known that ρπ,λ is irreducible by a result of Ribet [18]. It is also known
that it is de Rham and the Hodge-Tate weights are distinct. Strong irreducibility of
ρ1π,λ and ρπ,λ are clearly equivalent. Assume that ρπ,λ |GL is reducible for some finite
Galois extension L{K. Since ρπ,λ is semi-simple, so is its restriction to GL and we
have ρπ,λ |GL » χ1 ‘ χ2 . Since ρπ,λ is Hodge-Tate with distinct Hodge-Tate weights,
so is ρπ,λ |GL . This implies that χ1 ‰ χ2 . Let K 1 be the fixed field of the stabilizer
of χ1 . Then it is clearly a degree 2 extension of K and if GalpK 1 {Kq “ t1, σu then
χ2 “ σχ1 (otherwise χ1 would be a direct summand of ρπ,λ ). This means that
1
ρπ,λ » IndK K pχ1 q which implies that π is CM. Therefore ρπ,λ and hence ρπ,λ are
1

strongly irreducible.
Now let gder
λ Ď sl2 pQp q be the derived part of the Qp -Lie algebra of the image
1
of ρπ,λ : GM Ñ SL2 pQp q. Since ρ1π,λ is strongly irreducible, the irreducibility holds
infinitesimally, i.e. gλ Ď gl2 pQp q is an irreducible representation. This means that
the centralizer of g and hence its center is in the center of gl2 pQp q which implies that
gder
λ Ď sl2 pQp q is also irreducible. The only irreducible semi-simple Lie-subalgebra
of sl2 is itself so we are done.


Now from proposition 4.4 it follows:

Corollary 4.6. Let π be a regular, algebraic, cuspidal automorphic representation
of GL2 pAK q that doesn’t have complex multiplication and let F be the field fixed
by inner-twists. Then there exists an inner form Hp of SL2 over Fp and a finite
extension L of K such that the image of ρπ,p pGL q is contained and p-adically open
in Hp pFp q ¨ Qˆ
p Ď GLn pEp q.

In the work of Ribet, Momose and Nekovar, they construct an Azumaya algebra
Dp {Fp which contains the image. The relation to the last corollary is that if Dpˆ is
the algebraic group of units of Dp then pDpˆ qder “ Hp . In the case that π has parallel
weight 2, so we expect an abelian variety to be associated to π, this algebra Dp is
closely related to the endomorphism ring of that abelian variety. We will explain
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 17

the relation of these results to the Mumford-Tate conjecture for that abelian variety
in section 5.
4.2. The GL3 -case. In this section we prove our main result. The CM-case for
n “ 2 can be thought of as the form essentially coming from GL1 by induction.
Similarly, in the n “ 3 case we need to first exclude all the cases that π comes from
smaller groups via a Langlands transfer, in which case the image would be easy to
describe by previous results. It turns out that we only need to exclude the following
two cases to be able to describe the image:
(1) π is essentially sym2 , i.e. there exists an automorphic representation θ of
GL2 pAK q and a Hecke character η such that π “ sym2 pθq b η.
(2) π is an induction of a character, i.e. there exist a degree 3 extension L{K
and a Hecke character η of AL such that π “ IndL K pηq.
2
Notice that Langlands functoriality is known for sym : GL2 Ñ GL3 by [9] and
automorphic base change is known for prime degree extensions by [1]. In the first
case above, determining the image reduces to the GL2 case and in the second case
to the GL1 case. The next two lemmas give equivalent classifications of the above
cases and show that they follow from our primary assumptions on π (not essentially
self-dual and not self-twist) that are needed to define extra-twists to begin with.
Lemma 4.7. π is essentially sym2 if and only if there exist a Hecke character χ
such that π “ π _ b χ.
Proof. Since GL2 representation are essentially self-dual the ”only if” part is clear.
Now assume π “ π _ b χ and let ω be the central character of π. Taking the central
characters of both sides we have χ3 “ ω 2 . So χ “ pωχ´1 q2 has a square root and
by twisting out this square root we can assume that π is self-dual. Now the result
follows from [15]. 
Remark 4.8. Since we assumed K is totally real and π is cuspidal regular algebraic
and hence the existence of the associated Galois representations is known, one can
equivalently work with the associated Galois representation since strong multiplicity
one is known for GLn . Then one can give a different proof in the Galois side by
investigating the projective image of the representation.
We also need the following lemma from [1, §3]
Lemma 4.9. π is an induction of a character if and only if there exists a Hecke
character χ such that π “ π b χ.
Definition 4.10. An automorphic representation π of GL3 pAK q is said to be of
general type if π is neither essentially self-dual nor self-twist. Equivalently, π is
neither essentially sym2 nor an induction of a character.
Since we know strong multiplicity one for GLn and Langlands functoriality for
sym2 : GL2 Ñ GL3 and automorphic induction for degree three extensions, these
assumptions are equivalent to the similar assumption on each of the Galois repre-
sentations ρπ,λ .
In [4], Böckle and Hui prove that (in the n “ 3 case), ρπ,λ is irreducible for all λ.
They also prove that for a density 1 set of rational primes P, ρπ,λ is de Rham with
distinct Hodge-Tate weights for all λ : E ãÑ Qp and p P P. We will use these results
18 ALIREZA SHAVALI

to check the validity of ρ1π,p (in fact we only need regularity at one place λ by part
4 of Remark 3.2).
Proposition 4.11. Assume that π is not self-twist. Then for each p P P and
λ : E ãÑ Qp one has that ρ1π,λ is strongly irreducible.
Proof. Assume that ρπ,λ |GL is reducible for some finite Galois extension L{K. Since
ρπ,λ is semi-simple (in fact irreducible), so is its restriction to GL and we have
that ρπ,λ |GL decomposes into the sum of irreducible direct summands. If it de-
composes into a two irreducible summand, then the action of GK cannot switch
the two summands for dimension reasons and hence each summand is actually a
sub-representation of ρπ,λ which is a contradiction. So we must have ρπ,λ |GL »
χ1 ‘ χ2 ‘ χ3 . Since ρπ,λ is Hodge-Tate with distinct Hodge-Tate weights, so is
ρπ,λ |GL . This implies that the three characters are distinct. Let K 1 be the fixed
field of the stabilizer of χ1 . The action of GK on these characters must be transitive
so K 1 is a degree 3 extension of K and if GalpK 1 {Kq “ t1, σ, σ 2 u then χ2 “ σχ1
1
and χ3 “ σ 2 χ1 (or the other way around). This means that ρπ,λ » IndK K pχ1 q which
implies that π is also a degree three automorphic induction of a character and hence
is self-twist which contradicts the assumption. Therefore ρπ,λ and hence ρ1π,λ are
strongly irreducible. 
Proposition 4.12. Assume that π is of general type. Then for each prime number
p and embedding λ : E ãÑ Qp one has that the Qp -Lie algebra of the image of ρ1π,λ
is sl3 pQp q.
Proof. First assume that p P P. Let Vλ be the underlying vector space of ρ1π,λ and
let gλ Ď sl3 pQp q be the Qp -Lie algebra of the image of ρ1π,λ : GM Ñ SL3 pQp q. Since
ρ1π,λ is strongly irreducible, for every finite extension N of M we have EndGN pVλ q “
Qp , hence this should be true infinitesimally and we have Endgλ pVλ q “ Qp . This
means that the standard representation gλ ãÑ EndQp pVλ q » gl3 pQp q is an irreducible
faithful representation which implies that gλ is reductive. Let gder λ be its derived
subgroup and hence a semi-simple subgroup of sl3 pQp q. Since the center of gl3 acts
by scaler multiplication, any gder λ -invariant subspace is automatically gλ -invariant as
well which implies that the centralizer of g and hence its center are in the center of
gl3 pQp q which implies that gder λ Ď gl3 pQp q is also irreducible.
The only irreducible, semi-simple Lie-subalgebras of sl3 up to conjugation are sl3
or sl2 embedded into sl3 by sym2 . We need to show that the latter does not happen.
Assume that gder λ is the image of sym2 : sl2 Ñ sl3 . Since sym2 is an irreducible
representation, its centralizer in gl3 is the center which means that gλ “ gder λ ‘ zpgλ q
is in the image of sym2 : gl2 Ñ gl3 . This means that there is an open subgroup
of GM whose image under ρ1π,p is in the image of sym2 : GL2 Ñ GL3 . So there
exists a finite Galois extension M 1 of M such that ρ1π,p pGM 1 q is in the image of
sym2 : GL2 pEλ q Ñ GL3 pEλ q and hence is essentially self-dual. Since the determinant
of ρ1π,p is trivial, there exist a finite extension N of M 1 (which clearly can be taken to
be Galois over M ) such that the restriction to GN is in fact self-dual. Now applying
lemma 3.4 to two representations ρ1π,p |GN and ρ1π,p |_ GN , there exists a character φ
1 1 _
such that ρπ,p » pρπ,p q b φ which contradicts non-essential-self-duality of π. This
contradiction implies the result in the case where p P P.
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 19

Now by [4, Theorem 3.2], the irreducible type of ρπ,λ is independent of λ. This in
particular means that if the Qp -Lie algebra of the image of ρπ,λ contains sl3 for one
λ (irreducible type A2 ), it contains sl3 for every λ. This clearly implies the result in
general. 
Now we can easily deduce our main result:
Proof of Theorem 1.1. The last two propositions imply that for π of general type,
ρ1π,p is valid for any prime number p. Then proposition 4.4 implies the result. 
Example 4.13. In [8], van Geeman and Top construct a 3-dimensional Qpiq-rational
compatible family of (motivic) Galois representations of GQ which is not self-twist or
essentially self-dual and an automorphic representation of GL3 pAQ q that should cor-
respond to it. We can apply our results to the Galois representations they construct.
For each prime p they construct a Galois representation
ρp : GQ Ñ GL3 pQpiq b Qp q
which has the property that ρp » ρp _ b ǫp where p¨q indicates complex conjugation.
For each unramified p, the characteristic polynomial of Frobp looks like
X 3 ´ bp X 2 ` pbp X ´ p3
and they give a list of values of bp P Qpiq for small primes.
Now in our notation, the coefficient field is E “ Qpiq. There is one outer twist
p¨, ǫp q and there can’t be any more non-trivial extra-twists since AutpEq » Z{2Z.
Therefore F inn “ Qpiq as well and F “ Q. Then for each prime p we can construct a
form Hp of SL3 over the field Qp as in section 3 and the image of ρp is contained and
Qpiqp
open in Hp pQp q ¨ Qˆp . So we get an algebraic group Hp ¨ Gm Ď ResQp GL3 whose
Qp points describe the image. We know that Hp is a form of SL3 . Recall that it is
constructed as
Qpiq
Hp “ pResQp p SL3 qtwpGalpQpiq{Qqq
Similar to corollary 3.10 if p is a prime that splits in Qpiq, a prime that is congruent
to 1 modulo 4, then Hp is in fact isomorphic to SL3 over Qpiqp » Qp ˆQp . Otherwise,
it is not an inner-form and since it splits over Qpiq it is isomorphic to the the special
unitary group SU3 for the degree two field extension Qpiqp {Qp . So for half of the
primes (primes of the form p “ 4k ` 1) the (Zariski closure of the) image is GL3 and
for the other half (primes of the form p “ 4k ` 3) it is SU3 ¨ Gm .
Example 4.14. In [22], Upton constructs a 3-dimensional Qpζ3 q-rational compat-
ible family of (motivic) Galois representations of GQpζ3 q which is not self-twist or
essentially self-dual and gives a precise description of its image. It is clear from her
construction that these Galois representations have an outer-twist. She also observes
the similar phenomenon as in the last example. Namely, that for half of the primes
the image is GL3 and for the other half is a unitary group. Although, we believe
there is a slight error in her conclusion and the image in the latter case should be
SU3 ¨ Gm as above, rather than the general unitary group GU3 as she claims. In
fact since in the split case the image is 9-dimensional (as it is GL3 ) if one believes in
the Mumford-Tate conjecture, the image cannot be the 10-dimensional group GU3
in the non-split case.
20 ALIREZA SHAVALI

Even though in her case K is not totally real, we can still directly apply corollary
3.9 to a twist of the Galois representations she constructs (after a finite extension)
and then deduce openness. It is easy to check the validity of this twisted Galois
representation.
4.3. The GLn -case. In this section we discuss the GLn -case. Everything we say
here is conjectural. We assume the irreducibility conjecture (Galois representations
associated with cuspidal automorphic representations are irreducible), Langlands
Functoriality and the expected p-adic Hodge theoretic properties (de Rham with
distinct Hodge-Tate weights) of our Galois representations. We want to see, as-
suming all these, when we can apply proposition 4.4 to a regular cuspidal algebraic
automorphic representation π of GLn pAK q. First of all we need to assume π is nei-
ther self-twist nor essentially self-dual (in n ą 2 case). Then we only need to check
that ρπ,λ is strongly irreducible for each λ and that the Qp -Lie algebra of the image
of ρ1π,λ is sln .
Assume that ρπ,λ is reducible after restricting to GL for a finite Galois extension
L of K. The irreducible direct summands of ρπ,λ |GL are distinct since the Hodge-
Tate weights are distinct and the action of GK on them is transitive since ρπ,λ
is irreducible. This easily implies that ρπ,λ is an induction of a representation of a
proper subgroup. This means that the automorphic representation π is an induction,
assuming that the automorphic induction is true. So in order to make sure that ρπ,λ
is strongly irreducible, we only need to assume that it is not an induction.
Now let g be the Qp -Lie algebra of the image of ρ1π,λ . Since ρπ,λ and hence ρ1π,λ are
strongly irreducible, g is an irreducible Lie sub-algebra of gln and hence reductive.
We only need to show that gder “ sln . It is well-known that since gder is semi-simple,
there exists a semi-simple (connected) algebraic subgroup G1 of GLn (over Qp ) such
that LiepG1 q “ gder . This implies that after a finite Galois extension M {K, the
image of GM under ρπ,p lies in GpQp q Ĺ GLn pQp q for the (connected) reductive
group G0 “ G1 ¨ Gm Ĺ GLn . This is exactly the connected component of the Qp -
Zariski closure of the image of ρπ,λ . Let the whole image be G. Then the stabilizer of
G0 would give a finite Galois extension L{K and the component group is isomorphic
to GalpL{Kq. The nicest situation would be if GpQp q “ G0 pQp q ¸ GalpL{Kq. But
it is not clear if this should happen. Nevertheless, one has the following result of
Brion [5]:
Lemma 4.15 (Brion). Let G be an algebraic group over a field k and let
1ÑN ÑGÑQÑ1
be a short exact sequence (of algebraic groups over k) such that Q is finite. Then
there exists a finite subgroup F of G, such that G “ N ¨F . In other words, F surjects
to Q and G is a quotient of N ¸ F where F acts on N by conjugation.
Now using this result one can at least find a finite Galois extension M {K such
that G is the quotient of G0 pQp q ¸ GalpM {Kq. Now we can form the L-group
G0 pQp q ¸ GK where the GK action factors through the GalpM {Kq action from
above. Fix a maximal torus and a Borel subgroup of G0 containing it. The above
action of GalpM {Kq on G0 pQp q gives an action on the based root datum which in
turn gives an action on the dual root datum. This finally gives a reductive group
H over K which splits over M whose Langlands L-group is G0 pQp q ¸ GK with the
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 21

above action. Now the Langlands functoriality for the L-map L H Ñ L GLn implies
that π should come from an automorphic representation of the non-split reductive
group H via the Langlands transfer induced by the above L-map. This motivates
the following definition:
Definition 4.16. A regular, cuspidal, automorphic representation π of GLn pAK q
is said to be of general type, if it is neither self-twist, nor essentially self-dual (in
the n ą 2 case) and there does not exists any reductive group H over K that is
a form of a proper subgroup of GLn such that π is the image of an automorphic
representation of HpAK q under the Langlands transfer attached to the L-map
L
H Ñ L GLn,K
For instance, in the GL2 case this just means that π is not CM and in the GL3 case
it agrees with the definition 4.10. Note that the condition above also automatically
includes that π is not an induction since it would be in the image of the following
L-map then:
L
ResL L
K GLd Ñ GLdrL:Ks
The discussion above show that if one believes in Langlands functoriality, the
irreducibility conjecture and that ρπ,λ is de Rham with distinct Hodge-Tate weights
then ρ1π,p is valid. In conclusion we make the following conjecture:
Conjecture 4.17. Let K be totally real and π be a regular, cuspidal, algebraic
automorphic representation of GLn pAK q of general type. Let E “ Qpπq and let F
be the field fixed by the E-extra-twists of π. Then there exists a semi-simple group
H over F which is a form of SLn and there exists a finite extension L{K such that
for any prime p the image of ρπ,p pGL q is contained and open in HpFp q ¨ Qˆ
p.
Remark 4.18. If π is not of general type then it comes from a smaller group H. Since
the dimension of the group is getting smaller, there should be an optimal choice for
H. Loosely speaking, π should be of general type for some group. Then one has
to study the image inside this smaller group, via the extra-twists for the Langlands
dual of this group. Then it might be possible to give a precise description of the
image as above, using the extra-twists.

5. Relations to the Mumford-Tate Group

In this section we study the relation of our results from the last section to the
Mumford-Tate group and the Mumford-Tate conjecture. Almost everything in this
section is conjectural, but it could give an idea of why one should believe in the
conjectures presented here.
Clozel predicts that there should be a correspondence between algebraic automor-
phic representations of GLn pAK q and motives over K with coefficients in number
fields. Let K be totally real as before and π be a regular cuspidal algebraic auto-
morphic representation of GLn pAK q. Then Clozel predicts the existence of a motive
M “ Mπ over K with coefficients in a number field E containing Qpπq that is asso-
ciated to π in the way explained in [7]. Here, we are thinking about a motive as a
collection of realizations whose different structures are compatible through a set of
comparison isomorphism. From now on, we assume that such a motive exists. Let
HB pM q, HdR pM q and Hp pM q be the Betti, de Rham and p-adic realizations of M .
Note that the first two are E-vector spaces and the last one is an Ep -module.
22 ALIREZA SHAVALI

Let V “ HB pM q. The real and complex

ś Betti cohomology V bQ R and V bQ
C have an E bQ R and E bQ C “ λ:EãÑC C structure, respectively. śSimilarly
the complex de Rham cohomology HdR pM q bQ C has an E bQ C “ λ:EãÑC C
structure. The (E bQ C-modules) comparison isomorphism between Betti and de
Rham cohomologies, VC » HdR pM q bQ C, equips V with a rational Hodge structure.
We denote this Hodge structure with
hπ : S Ñ GLpVR q
where S “ ResCR Gm is the Deligne torus. Fixing an E-basis for VQ which in turn
gives an E b R basis for VR enables us to write this as
hπ,8 : S Ñ GLn pE bQ Rq
This representation should be thought of as the analogue of our p-adic Galois rep-
resentations ρπ,p : GK Ñ GLn pE bQ Qp q from section 4, associated to the prime at
infinity. Note that this is equipped with an action of AutpEq on the coefficients.
Recall that the Mumford-Tate group of the motive M is defined to be the smallest
algebraic group G Ď GLpVQ q over Q that contains the image of hπ (the map hπ
factors through GR ). The Mumford-Tate conjecture then states that the connected
component of the Qp -Zariski closure of the image of the Galois representation Hp pM q
is equal to G ˆQ Qp .
Now, let π be of general type and Γ be the group of E-extra-twists of π. Let
|.|m ω be the central character of π where ω is a finite order Hecke character. From
now on, for simplicity we assume that m is divisible by n. So let m “ nd. Then
the outer-twists of π are of the form pτ, | ¨ |2d ηq for a finite character η and hence
the outer-twists of ρπ,p are of the form pτ, ǫ2d p ηq where ǫp is the p-adic cyclotomic
character and we think of η as a finite Galois character. The extra-twist of π then
induce extra-twists on the motive Mπ . An inner-twist pσ, χq induces an isomorphism
σ M » M b χ where χ is the Artin motive associated with the finite character χ.
π π
The outer-twist pτ, | ¨ |2d ηq induces an isomorphism τ Mπ » Mπ_ b Qp2dq b η.
In particular, the extra-twists also induce symmetries on the Hodge-structure
since E acts on the motive M via endomorphisms. Twisting with finite characters
does not affect the Hodge structure and twisting with the 2d’th power of the cy-
clotomic character amounts to twisting with the Tate Hodge structure Qp2dq. This
means that for each inner-twist σ P Γinn one has σ hπ » hπ and for each outer-
twist τ P Γout one has τ hπ » h_ π bQ Qp2dq. Now, if we twist hπ with Qp´dq we
still get hπ p´dq » hπ p´dq for each inner-twist and τ hπ p´dq » hπ p´dq_ for each
σ

outer-twist. This is analogous to the representation ρ1π,p from section 4.

Since an isomorphism of rational Hodge structures comes from an isomorphism
over Q between the underlying rational vector spaces, and since everything is com-
patible with the E-structures, we can find matrices ασ , ατ P GLn pEq that give the
isomorphisms above by conjugation. so we get:
"
hπ,8 p´dq “ ασ σ hπ,8 p´dqα´1
σ
hπ,8 p´dq “ ατ τ h´T ´1
π,8 p´dqατ

Note that the determinant of hπ,8 p´dq is trivial so it has values in SLn . Now define
the 1-cocycle f : Γ Ñ AutE pSLn q by sending an inner-twist σ to adpασ q and an
outer-twist τ to adpατ q ˝ p¨q´T . Then as in section 2.2, we can define the twisted
action of Γ on ResE F SLn and the matrices in the image of hπ,8 p´dq are clearly
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 23

invariant under this action. We define the groups H8 :“ pResEbR F bR SLn q

twpΓq and

H :“ pResE F SLn q
twpΓq . Note that H
8 is the base change of H to R and it is the
Archimedean analogue of the groups Hp from section 4.
Remark 5.1. We could also define the group H8 without assuming the existence of
the motive, only from the real Hodge structure coming from the Archimedean part
of π. But then the connection to the Mumford-Tate group is of course less clear.
Lemma 5.2. The Mumford-Tate group of the motive Mπ is contained in the group
ResFQ pHq ¨ Gm .
Proof. The image of hπ,8 p´dq lies in H8 by the definition of H8 . This implies that
the map hπ,8 factors through H8 ¨ Gm,R and therefore the map hπ factors through
the base change of the group ResFQ pHq ¨ Gm,Q to R. This implies the result. 
Note that the dimension of the group H is equal to the dimension of all groups Hp
from section 4. Our results in that section make it reasonable to make the following
conjecture.
Conjecture 5.3. If π is of general type then the Mumford-Tate group of Mπ is
equal to ResFQ pHq ¨ Gm,Q .
We showed that the group H (or any of the Hp ’s from section 4) gives an upper
bound for the Mumford-Tate group. On the other hand, in the special case that M
is an abelian motive, Deligne shows that the Mumford-Tate group MTpMπ q, after
base changing to Qp , always contains the connected component of the image of the
p-adic Galois representations. So if we known that the image is open in Hp even for
one prime, the conjecture above follows. In particular we have the following result
in the n “ 2 case (which was probably known to Nekovar):
Corollary 5.4. Let f be a non-CM Hilbert modular newform of parallel weight 2
over the totally real field K and assume there is an abelian variety Af associated to
it. Then conjecture 5.3 and the Mumford-Tate conjecture hold for Af .
Proof. Note that MTpMπ q bQ Qp is contained in pResFQ pHq ¨ Gm q b Qp by lemma
5.2 and contains the connected component of the Zariski-closure of the image of ρf,p
F
which is equal to ResQpp Hp pQp q ¨ Qˆ
p by 4.6. Since the dimensions match, both must
be equality. 
At the end we want to come back to remark 4.2. Recall that by [6], for each
prime p there exist an Azumaya algebra Dp over Qpπqp :“ Qpπq bQ Qp such that the
Galois representation ρπ,p factors through Dp Ă Dp bQpπq E » Mn pEp q, assuming
irreducibility of the Galois representation. We are interested to see if the local object
Dp should be a global object D defined over Qpπq. In the n “ 1 case this is clear.
In the n “ 2 case, since ρπ,p is odd, the residual representation is multiplicity free
and a result of Bellaı̈che and Chenevier [3] shows that in this case every pseudo-
representation can be defined over its trace field and hence Dp » GL2 for all primes
p. So D » GL2 works.
Now assume that π is of general type and assume conjecture 5.3 and the Mumford-
Tate conjecture. First notice that for every σ P GalpE{Qpπqq we have an inner-twist
π » σ π therefore F Ď Qpπq. Now we know by our assumptions that for some finite
24 ALIREZA SHAVALI

Galois extension L{K the image of ρπ,p |GL is open in Hp pFp q ¨ Qˆ p . This is the Qp -
Zariski closure of the image hence the Fp -Zariski closure of the image is Hp pFp q ¨ Fpˆ .
Note that the inner product is happening in GLn so since Hp is a form of SLn we
deduce that Gp :“ Hp ¨ Gm,Fp is a form of GLn . Since the image is contained in Dpˆ
which is an algebraic group over Qpπqp and Fp Ď Qpπqp , we have
Gp pFp q Ď Dpˆ pQpπqp q Ď GLn pEp q
and the base change of either Gp or Dpˆ to Ep is equal to GLn . This implies that Dpˆ
is the Qpπqp -Zariski closure of the image of the Galois representation. In particular,
by the Mumford-Tate conjecture, the groups Dpˆ should come from a Global group,
namely the Qpπq-Zariski closure of the image of the map hπ,8 . This is an inner-form
of GLn since all Gp ’s become inner forms over Fpinn so it is the group of units of a
central simple algebra D over Qpπq.
In particular in the n “ 3 case, π is either an induction of a character from a
degree 3 extension, essentially sym2 or of general type. In the first two cases Dp
being global reduces to the n “ 1 and n “ 2 case and in the third case follows from
the discussion above (assuming all the above conjectures). This makes it reasonable
to make the following conjecture:
Conjecture 5.5. Let π be a regular cuspidal algebraic automorphic representation
of GLn pAK q and ρπ,p the associated Galois representation and Dp the Azumaya
algebra coming from [6]. Then there exists a central simple algebra D over Qpπq
such that Dp » D bQpπq Qpπqp for every p.
In particular this conjecture implies that for all but finitely many primes, the
representation ρπ,p can be defined over its trace field.

References
1. James Arthur and Laurent Clozel, Simple algebras, base change, and the advanced theory of the
trace formula, no. 120, Princeton University Press, 1989.
2. Thomas Barnet-Lamb, Toby Gee, David Geraghty, and Richard Taylor, Potential automorphy
and change of weight, Annals of Mathematics (2014), 501–609.
3. Joël Bellaıche and Gaëtan Chenevier, Families of galois representations and selmer groups,
Astérisque 324 (2009), 1–314.
4. Gebhard Böckle and Chun-Yin Hui, Weak abelian direct summands and irreducibility of galois
representations, arXiv preprint arXiv:2404.08954 (2024).
5. Michel Brion, On extensions of algebraic groups with finite quotient, Pacific J. Math 279 (2015),
no. 1-2, 135–153.
6. Gaëtan Chenevier, The p-adic analytic space of pseudocharacters of a profinite group and
pseudorepresentations over arbitrary rings, Automorphic Forms and Galois Representations
1 (2014), 221–285.
7. Laurent Clozel, Représentations galoisiennes associées aux représentations automorphes auto-
duales de GLpnq, Publications Mathématiques de l’IHÉS 73 (1991), 97–145.
8. Bert van Geemen and Jaap Top, A non-selfdual automorphic representation of GL3 and a galois
representation, Invent. math 117 (1994), no. 3, 391–401.
9. Stephen Gelbart and Hervé Jacquet, A relation between automorphic representations of GL2
and GL3 , Annales Scientifiques de l’École Normale Supérieure, vol. 11, 1978, pp. 471–542.
10. Michael Harris, Kai-Wen Lan, Richard Taylor, and Jack Thorne, On the rigid cohomology of
certain shimura varieties, Research in the Mathematical Sciences 3 (2016), 1–308.
11. Chun Yin Hui, Monodromy of galois representations and equal-rank subalgebra equivalence,
Ph.D. thesis, Indiana University, 2013.
 ON THE IMAGE OF AUTOMORPHIC GALOIS REPRESENTATIONS 25

12. Michael Larsen and Richard Pink, Determining representations from invariant dimensions,
Inventiones mathematicae 102 (1990), no. 1, 377–398.
13. Fumiyuki Momose, On the ℓ-adic representations attached to modular forms, J. Fac. Sci. Univ.
Tokyo Sect. IA Math 28 (1981), no. 1, 89–109.
14. Jan Nekovář, Level raising and anticyclotomic selmer groups for hilbert modular forms of weight
two, Canadian Journal of Mathematics 64 (2012), no. 3, 588–668.
15. Dinakar Ramakrishnan, An exercise concerning the selfdual cusp forms on GLp3q, Indian J.
Pure Appl. Math 45 (2014), no. 5, 777–785.
16. Kenneth A Ribet, Galois action on division points of abelian varieties with real multiplications,
American Journal of Mathematics 98 (1976), no. 3, 751–804.
17. , Twists of modular forms and endomorphisms of abelian varieties, Mathematische An-
nalen 253 (1980), no. 1, 43–62.
18. , Galois representations attached to eigenforms with nebentypus, Modular Functions of
One Variable V: Proceedings International Conference, University of Bonn, Sonderforschungs-
bereich Theoretische Mathematik July 2–14, 1976, Springer, 2006, pp. 18–52.
19. Raphaël Rouquier, Caractérisation des caracteres et pseudo-caracteres, Journal of Algebra 180
(1996), no. 2, 571–586.
20. Peter Scholze, On torsion in the cohomology of locally symmetric varieties, Annals of Mathe-
matics (2015), 945–1066.
21. Jean-Pierre Serre, Propriétés galoisiennes des points d’ordre fini des courbes elliptiques, Invent.
math 15 (1972), 259–331.
22. Margaret G Upton, Galois representations attached to picard curves, Journal of Algebra 322
(2009), no. 4, 1038–1059.