SHTUKAS FOR REDUCTIVE GROUPS AND LANGLANDS

CORRESPONDENCE FOR FUNCTION FIELDS

VINCENT LAFFORGUE

This text gives an introduction to the Langlands correspondence for function

fields and in particular to some recent works in this subject. We begin with a
short historical account (all notions used below are recalled in the text).
The Langlands correspondence [Lan70] is a conjecture of utmost importance,
concerning global fields, i.e. number fields and function fields. Many excellent
surveys are available, for example [Gel84, Bum97, BeGe03, Tay04, Fre07, Art14].
The Langlands correspondence belongs to a huge system of conjectures (Lang-
lands functoriality, Grothendieck’s vision of motives, special values of L-functions,
Ramanujan-Petersson conjecture, generalized Riemann hypothesis). This system
has a remarkable deepness and logical coherence and many cases of these con-
jectures have already been established. Moreover the Langlands correspondence
over function fields admits a geometrization, the “geometric Langlands program”,
which is related to conformal field theory in Theoretical Physics.
Let G be a connected reductive group over a global field F . For the sake of
simplicity we assume G is split.
The Langlands correspondence relates two fundamental objects, of very differ-
ent nature, whose definition will be recalled later,
• the automorphic forms for G,
• the global Langlands parameters , i.e. the conjugacy classes of morphisms
from the Galois group Gal(F /F ) to the Langlands dual group G(Q b ` ).
For G = GL1 we have G b = GL1 and this is class field theory, which describes the
abelianization of Gal(F /F ) (one particular case of it for Q is the law of quadratic
reciprocity, which dates back to Euler, Legendre and Gauss).
Now we restrict ourselves to the case of function fields.
In the case where G = GLr (with r ≥ 2) the Langlands correspondence (in
both directions) was proven by Drinfeld [Dri78, Dri87, Dri88, Dri89] for r = 2
and by Laurent Lafforgue [Laf02] for arbitrary r. In fact they show the “au-
tomorphic to Galois” direction by using the cohomology of stacks of shtukas
and the Arthur-Selberg trace formula, and deduce from it the reverse direction
by using the inverse theorems of Weil, Piatetski-Shapiro and Cogdell [CPS94]
(which are specific to the case of GLr ) as well as Grothendieck’s functional
equation and Laumon’s product formula [Laum87b] (which are specific to the
case of function fields). Other works using the Arthur-Selberg trace formula for
stacks of shtukas are, in chronological order, Laumon [LRS93, Laum96], Ngo

Date: March 10, 2018.

1
2 VINCENT LAFFORGUE

Bao Chau [NBC99, NBC06a, NgNg08], Ngo Dac Tuan [NDT07, NDT09], Lau
[Lau04, Lau07], Kazhdan, Varshavsky [Var04], Badulescu, Roche [BaRo17].
In [Laf12] we show the “automorphic to Galois” direction of the Langlands
correspondence for all reductive groups over function fields. More precisely we
construct a canonical decomposition of the vector space of cuspidal automorphic
forms, indexed by global Langlands parameters. This decomposition is obtained
by the spectral decomposition associated to the action on this vector space of
a commutative algebra B of “excursion operators” such that each character of
B determines a unique global Langlands parameter. Unlike previous works, our
method is independent on the Arthur-Selberg trace formula. We use the following
two ingredients:
• the classifying stacks of shtukas, introduced by Drinfeld for GLr [Dri78,
Dri87] and generalized to all reductive groups and arbitrary number of
“legs” by Varshavsky [Var04] (shtukas with several legs were also consid-
ered in [Lau04, NBC06a]),
• the geometric Satake equivalence, due to Lusztig, Drinfeld, Ginzburg and
Mirkovic-Vilonen [BeDr99, MiVi07] (it is a fundamental ingredient of the
geometric Langlands program, whose idea comes from the fusion of par-
ticles in conformal field theory).
In the last sections we discuss recent works related to the Langlands program
over function fields, notably on the independence on ` and on the geometric
Langlands program. We cannot discuss the works about number fields because
there are too many and it is not possible to quote them in this short text. Let us
only mention that, in his lectures at this conference, Peter Scholze will explain
local analogues of shtukas over Qp .
Acknowledgements. I thank Jean-Benoît Bost, Alain Genestier and Dennis
Gaitsgory for their crucial help in my research. I am very grateful to the Centre
National de la Recherche Scientifique. The Langlands program is far from my
first subject and I would never have been able to devote myself to it without the
great freedom given to CNRS researchers for their works. I thank my colleagues
of MAPMO and Institut Fourier for their support. I thank Dennis Gaitsgory for
his crucial help in writing the part of this text about geometric Langlands. I also
thank Aurélien Alvarez, Vladimir Drinfeld, Alain Genestier, Gérard Laumon and
Xinwen Zhu for their help.

1. Preliminaries
1.1. Basic notions in algebraic geometry. Let k be a field. The ring of
functions on the n-dimensional affine space An over k is the ring k[x1 , ..., xn ] of
polynomials in n variables. For any ideal I, the quotient A = k[x1 , ..., xn ]/I is
the ring of functions on the closed subscheme of An defined by the equations in
I and we obtain in this way all affine schemes (of finite type) over k. An affine
scheme over k is denoted by Spec(A) when A is the k-algebra of functions on it.
It is equipped with the Zariski topology (generated by open subschemes of the
form f 6= 0 for f ∈ A). It is called a variety when A has no non zero nilpotent
 SHTUKAS AND LANGLANDS CORRESPONDENCE 3

element. General schemes and varieties are obtained by gluing. The projective
space Pn over k is the quotient of An+1 \ {0} by homotheties and can be obtained
by gluing n + 1 copies of An (which are the quotients of {(x0 , ..., xn ), xi 6= 0},
for i = 0, ..., n). Closed subschemes (resp. varieties) of Pn are called projective
schemes (resp varieties) over k. Schemes over k have a dimension and a curve is
a variety purely of dimension 1.
1.2. Global fields. A number field is a finite extension of Q, i.e. a field generated
over Q by some roots of a polynomial with coefficients in Q.
A function field F is the field of rational functions on an irreducible curve X
over a finite field Fq .
We recall that if q is a prime number, Fq = Z/qZ. In general q is a power of
a prime number and all finite fields of cardinal q are isomorphic to each other
(although non canonically), hence the notation Fq .
The simplest example of a function field is F = Fq (t), namely the field of
rational functions on the affine line X = A1 . Every function field is a finite
extension of such a field Fq (t).
Given a function field F there exists a unique smooth projective and geometri-
cally irreducible curve X over a finite field Fq whose field of rational functions is
F : indeed for any irreducible curve over Fq we obtain a smooth projective curve
with the same field of rational functions by resolving the singularities and adding
the points at infinity. For example F = Fq (t) is the field of rational functions of
the projective line X = P1 over Fq (we have added to A1 the point at infinity).
For the rest of the text we fix a smooth projective and geometrically irreducible
curve X over Fq . We denote by F the field of functions of X (but F may also
denote a general global field, as in the next subsection).
1.3. Places of global fields and local fields. A place v of a global field F
is a non trivial multiplicative norm F → R≥0 , up to equivalence (where the
equivalence relation identifies k.k and k.ks for any s > 0). The completion Fv of
the global field F for this norm is called a local field. It is a locally compact field
and the inclusion F ⊂ Fv determines v. Therefore a place is “a way to complete
a global field into a local field”.
For any local field there is a canonical normalization of its norm given by the
action on its Haar measure. For any non zero element of a global field the product
of the normalized norms at all places is equal to 1.
For example the places of Q are
• the archimedean place, where the completion is R (with normalized norm
equal to the usual absolute value),
• for every prime number p, the place p where the completion is Qp (the
normalized norm in Qp of a number r ∈ Q× is equal to p−np (r) , where
np (r) ∈ Z is the exponent of p in the decomposition of r as the product
of a sign and powers of the prime numbers).
Thus the local fields obtained by completion of Q are Qp , for all prime numbers
p, and R. A place v is said to be archimedean if Fv is equal to R or C. These
places are in finite number for number fields and are absent for function fields.
4 VINCENT LAFFORGUE

For each non archimedean place v we denote by Ov the ring of integers of Fv ,

consisting of elements of norm ≤ 1. For example it is Zp if Fv = Qp .
In the case of function fields, where we denote by F the field of functions of
X, the places are exactly the closed points of X (defined as the maximal ideals).
The closed points are in bijection with the orbits under Gal(Fq /Fq ) on X(Fq )
(Galois groups are recalled below). For every closed point v of X, we denote by
nv : F × → Z the valuation which associates to a rational function other than
0 its vanishing order at v. We can see Ov as the Fq -algebra of functions on the
formal neighborhood around v in X and Fv as the Fq -algebra of functions on the
punctured formal neighborhood. We denote by κ(v) the residue field of Ov ; it
is a finite extension of Fq , whose degree is denoted by deg(v), therefore it is a
finite field with q deg(v) elements. The normalized norm on F associated to v sends
a ∈ F × to q − deg(v)nv (a) .
In the example where X = P1 = A1 ∪ ∞, the unitary irreducible polynomials
in Fq [t] (which is the ring of functions on A1 ) play a role analoguous to that of
the prime numbers in Z: the places of P1 are
• the place ∞, at which the completion is Fq ((t−1 )),
• the places associated to unitary irreducible polynomials in Fq [t] (the de-
gree of such a place is simply the degree of the polynomial). For example
the unitary irreducible polynomial t corresponds to the point 0 ∈ A1 and
the completion at this place is Fq ((t)).

P recalln that the local field Fq ((t)) consists of Laurent series, i.e.
We sums
n∈Z an t with an ∈ Fq and an = 0 for n negative enough.

1.4. Galois groups. If k is a field, we denote by k an algebraic closure of k. It

is generated over k by the roots of all polynomials with coefficients in k. The
separable closure k sep ⊂ k consists of the elements whose minimal polynomial over
k has a non zero derivative. We denote by Gal(k/k) = Gal(k sep /k) the group of
automorphisms of k (or equivalently of k sep ) which act by the identity on k. It is
a profinite group, i.e. a projective limit of finite groups: an element of Gal(k/k)
is the same as a family, indexed by the finite Galois extensions k 0 ⊂ k of k, of
elements θk0 ∈ Gal(k 0 /k), so that if k 00 ⊃ k 0 , θk00 k0 = θk0 . We recall that k 0 ⊂ k
is said to be a finite Galois extension of k if it is a finite dimensional k-vector
subspace of k sep and is stable under the action of Gal(k/k) = Gal(k sep /k) (and
then Gal(k 0 /k) is a finite group of cardinal equal to the dimension of k 0 over k).
A simple example is given by finite fields: Gal(Fq /Fq ) is equal to the profinite
completion Z b of Z in such a way that 1 ∈ Z b is the Frobenius generator x 7→ xq
(which is an automorphism of Fq equal to identity on Fq ).
We recall that for any Fq -algebra, x 7→ xq is a morphism of Fq -algebras, in
particular (x + y)q = xq + y q . For any scheme S over Fq we denote by FrobS :
S → S the morphism acting on functions by Frob∗S (f ) = f q .
We come back to the function field F of X. Our main object of interest is the
Galois group Γ = Gal(F /F ) = Gal(F sep /F ).
 SHTUKAS AND LANGLANDS CORRESPONDENCE 5

By the point of view of Grothendieck developed in SGA1, we have an equiva-

lence between
• the category of finite sets A endowed with a continuous action of Γ
• the category of finite separable F -algebras
where the functor from the first category to the second one maps A to the finite
separable F -algebra ((F sep )A )Γ (here (F sep )A is the direct sum of copies of F sep
indexed by A and Γ acts on each copy and permutes them at the same time). We
write η = Spec(F ) and η = Spec(F ). Then, for any dense open U ⊂ X, Γ has
a profinite quotient π1 (U, η) such that a continuous action of Γ on a finite set A
factors through π1 (U, η) if and only if Spec(((F sep )A )Γ ) extends (uniquely) to an
étale covering of U . We will not explain the notion of étale morphism in general
and just say that a morphism between smooth varieties over a field is étale if
and only if its differential is everywhere invertible. Thus we have an equivalence
between
• the category of finite sets A endowed with a continuous action of π1 (U, η)
• the category of finite étale coverings of U .
For any place v the choice of an embedding F ⊂ Fv provides an inclusion
Gal(Fv /Fv ) ⊂ Gal(F /F ) (well defined up to conjugation). We denote by Frobv ∈
Gal(F /F ) the image of any element of Gal(F v /Fv ) lifting the Frobenius generator
deg(v)
Frobenius x 7→ xq in Gal(κ(v)/κ(v)) = Z. b When U is open dense in X as
above and v is a place in U , the image of Frobv in π1 (U, η) is well defined up to
conjugation.

1.5. A lemma of Drinfeld [Dri78]. Let U ⊂ X open dense as above. For any
i ∈ I we denote by Frobi the “partial Frobenius” morphism U I → U I which sends
(xj )j∈I to (x0j )j∈I with x0i = FrobU (xi ) and x0j = xj for j 6= i. For any scheme T
and any morphism T → U I , we say that a morphism a : T → T is “above” Frobi
if the square
T
a / T

 
Frobi
UI / UI
is commutative.
Lemma 1.1. We have an equivalence of categories between
• the category of finite sets A endowed with a continuous action of
(π1 (U, η))I ,
• the category of finite étale coverings T of U I , equipped with partial Frobe-
nius morphisms, i.e. morphisms F{i} above Frobi , commuting with each
other, and whose composition is FrobT .
The functor from the first category to theQsecond one is the following: if the
action of (π1 (U, η))I on A factorizes through i∈I Gal(Ui /U ) where for each i, Ui
is a finite étale Galois covering of U (and Gal(Ui /U ) is its automorphism group),
6 VINCENT LAFFORGUE
Q
then the image by the functor is ( i∈I Ui ) ×Qi∈I Gal(Ui /U ) A, equipped with the
Q

partial Frobenius morphisms F{i} given by FrobUi × j6=i IdUj × IdA .

1.6. Split connected reductive groups and bundles. We denote by Gm =

GL1 the multiplicative group. A split torus over a field k is an algebraic group
T which is isomorphic to Grm for some r.
A connected reductive group over a field k is a connected, smooth, affine alge-
braic group whose extension to k has a trivial unipotent radical (i.e. any normal,
smooth, connected, unipotent subgroup scheme of it is trivial). A connected
reductive group G over k is said to be split if it has a split maximal torus T .
Then (after chosing a Borel subgroup containing T ) the lattices Hom(Gm , T )
and Hom(T, Gm ) are called the coweight and weight lattices of G. The split con-
nected reductive groups over a field k are exactly the quotients by central finite
subgroup schemes of products of Gm , simply-connected split groups in the four
series SLn+1 , Spin2n+1 , Sp2n , Spin2n , and five simply-connected split exceptional
groups.
Let G be a split connected reductive group over a field k, and X a scheme over
k. Then a G-bundle over X is a morphism Y → X, together which an action
of G on the fibers which is simply transitive. A GLr -bundle E gives rise to the
vector bundle of rank r equal to E ×GLr Ar and the notions are equivalent.

2. Reminder on automorphic forms

For the moment we take G = GLr . When the global field is Q, an automorphic
form (without level at finite places) is a function on the quotient GLr (Z)\GLr (R)
(the best known example is the particular case of modular functions, for which
r = 2). This quotient classifies the free Z-modules (or, equivalently, projective
Z-modules) M of rank r equipped with a trivialization M ⊗Z R = Rr (i.e. an
embedding of M as a lattice in Rr ). Indeed if we choose a basis of M over Z its
embedding in Rr is given by a matrix in GLr (R) and the change of the basis of
M gives the quotient by GLr (Z).
Now we come back to our function field F . To explain the analogy with Q we
choose a place v of X (of degree 1 to simplify) playing the role of the archimedean
place of Q (but this choice is not natural and will be forgotten in five lines). An
analogue of a projective Z-module M of rank r equipped with a trivialization
M ⊗Z R = Rr is a vector bundle of rank r over X equipped with a trivialization
on the formal neighborhood around v. Now we forget the trivialization on the
formal neighborhood around v (because we do not want to introduce a level at
v) and then we forget the choice of v.
Thus an automorphic form (without level at any place) for GLr is a function
on the set BunGLr (Fq ) of isomorphism classes of vector bundles of rank r over X.
Now we consider the case of a general group G. From now on we denote by
G a connected reductive group over F , assumed to be split for simplicity. An
automorphic form (without level) for G is a function on the set BunG (Fq ) of
isomorphism classes of G-bundles over X.
 SHTUKAS AND LANGLANDS CORRESPONDENCE 7

Remark 2.1. This remark can be skipped. In fact the G-bundles over X have
finite automorphism groups. Therefore it is more natural to consider BunG (Fq )
as a groupoid, i.e. a category where all arrows are invertible. It is the groupoid of
points over Fq of the Artin stack BunG over Fq whose groupoid of S-points (with
S a scheme over Fq ) classifies the G-bundles over X × S. We refer to [LaMoB99]
for the notion of Artin stack and we say only that examples of Artin stacks are
given by the quotients of algebraic varieties by algebraic groups. In Artin stacks
the automorphism groups of the points are algebraic groups (for example in the
case of a quotient they are the stabilizers of the action). The Quot construction of
Grothendieck implies that BunG is an Artin stack (locally it is even the quotient
of a smooth algebraic variety by a smooth algebraic group). The automorphisms
groups of points in the groupoid BunG (Fq ) are finite, in fact they are the points
over Fq of automorphisms groups of points in BunG , which are algebraic groups
of finite type.

It is convenient to impose a condition relative to the center Z of G. From

now on we fix a subgroup Ξ of finite index in BunZ (Fq ) (for example the trivial
subgroup if Z is finite) and we consider functions on BunG (Fq )/Ξ. However,
except when G is a torus, BunG (Fq )/Ξ is still infinite. To obtain vector spaces
of finite dimension we now restrict ourselves to cuspidal automorphic forms.
For any field E ⊃ Q, we denote by Cccusp (BunG (Fq )/Ξ, E) the E-vector
space of finite dimension consisting of cuspidal functions on BunG (Fq )/Ξ. It
is defined as the intersection of the kernel of all “constant term” morphisms
Cc (BunG (Fq )/Ξ, E) → C(BunM (Fq )/Ξ, E) (which are given by the correspon-
dence BunG (Fq ) ← BunP (Fq ) → BunM (Fq ) and involve only finite sums), for all
proper parabolic subgroups P of G with associated Levi quotient M (defined as
the quotient of P by its unipotent radical). For readers who do not know these
notions, we recall that in the case of GLr a parabolic subgroup P is conjugated
to a subgroup of upper block triangular matrices and that the associated Levi
quotient M is isomorphic to the group of block diagonal matrices. It is legitimate
in the Langlands correspondence to restrict oneself to cuspidal automorphic
forms because all automorphic forms for G can be understood from cuspidal
automorphic forms for G and for the Levi quotients of its parabolic subgroups.
Let ` be a prime number not dividing q. To simplify the notations we assume
that Q` contains a square root of q (otherwise replace Q` everywhere by a finite
extension containing a square root of q). For Galois representations we have to
work with coefficients in Q` and Q` , and not Q, Q and even C (to which Q` is iso-
morphic algebraically but not topologically) because the Galois representations
which are continuous with coefficients in C always have a finite image (unlike
those with coefficients in Q` ) and are not enough to match automorphic forms
in the Langlands correspondence. Therefore, even if the notion of cuspidal auto-
morphic form is (in our case of function fields) algebraic, to study the Langlands
correspondence we will consider cuspidal automorphic forms with coefficients in
E = Q` or Q` .
8 VINCENT LAFFORGUE

3. Class field theory for function fields

It was developped by Rosenlicht and Lang [Ser75]. Here we consider only the
unramified case.
Let Pic be the relative Picard scheme of X over Fq , whose definition is that,
for any scheme S over Fq , Pic(S) (the set of morphisms S → Pic) classifies the
isomorphism classes [E] of line bundles E on X × S (a line bundle is a vector
bundle of rank 1, so it is the same as a Gm -bundle). The relation with BunGL1
is that BunGL1 can be identified with the quotient of Pic by the trivial action of
Gm .
Let Pic0 be the neutral component of Pic, i.e. the kernel of the degree mor-
phism Pic → Z. It is an abelian variety over Fq , also called the jacobian of
X.
Class field theory states (in the unramified case to which we restrict ourselves
in this text) that there is a canonical isomorphism

ab ∼
(3.1) π1 (X, η) ×Zb Z → Pic(Fq )
characterized by the fact that for any place v of X, it sends Frobv to [O(v)],
where O(v) is the line bundle on X whose sections are the functions on X with
a possible pole of order ≤ 1 at v.
The isomorphism (3.1) implies that for any a ∈ Pic(Fq ) of non zero degree
we can associate to any (multiplicative) character χ of the finite abelian group
Pic(Fq )/aZ (with values in any field, e.g. Q` for ` prime to q) a character σ(χ)
of π1 (X, η). We now give a geometric construction of σ(χ), which is in fact the
key step in the proof of the isomorphism (3.1).
The Lang isogeny L : Pic → Pic0 is such that, for any scheme S over Fq and ev-
ery line bundle E on X×S, [E] ∈ Pic(S) is sent by L to [E−1 ⊗(FrobS × IdX )∗ (E)] ∈
Pic0 (S). We note that [(FrobS × IdX )∗ (E)] ∈ Pic(S) is the image by FrobPic of
[E] ∈ Pic(S). The Lang isogeny is surjective and its kernel is Pic(Fq ). For any
I
P
finite set I and any family (ni )i∈I ∈ Z satisfying i∈I ni = 0, we consider
I 0
thePAbel-Jacobi morphism AJ : X → Pic sending (xi )i∈I to the line bundle
O( i∈I ni xi ). We form the fiber product
ChtI,(ni )i∈I / Pic
p  L
 
XI / Pic0
AJ

and see that p is a Galois covering of X I with Galois group Pic(Fq ). Thus,
up to an automorphism group Fq × which we neglect, for any scheme S over Fq ,
ChtI,(ni )i∈I (S) classifies
• morphisms xi : S → X
• a line bundle E on X × S
• an isomorphism E−1 ⊗ (FrobS × IdX )∗ (E) ' O( i∈I ni xi ).
P

Moreover ChtI,(ni )i∈I is equipped with partial Frobenius morphisms F{i} sending
E to E ⊗ O(ni xi ). The morphism F{i} is above Frobi : X I → X I , because
 SHTUKAS AND LANGLANDS CORRESPONDENCE 9

(FrobS × IdX )∗ (O(xi )) = O(FrobS (xi )). Taking the quotient by aZ we obtain a
finite Galois covering

ChtI,(ni )i∈I /aZ

p

XI

with Galois group Pic(Fq )/aZ and equipped with the partial Frobenius morphisms
F{i} . Then Drinfeld’s lemma gives rise to a morphism αI,(ni )i∈I : π1 (X, η)I →
Pic(Fq )/aZ . The character σ(χ) of π1 (X, η) is characterized by the fact that for
any I and (ni )i∈I with sum 0, χ ◦ αI,(ni )i∈I = i∈I σ(χ)ni and this gives in fact a
construction of σ(χ).

4. The Langlands correspondence for split tori

Split tori are isomorphic to Grm , so there is nothing more than in the case of
Gm = GL1 explained in the previous section. Nevertheless the isomorphism of
a split torus with Grm is not canonical (because the automorphism group of Grm
is non trivial, equal to GLr (Z)). Let T be a split torus over F . To obtain a
canonical correspondence we introduce the Langlands dual group Tb, defined as
the split torus over Q` whose weights are the coweights of T and reciprocally.
In other words the lattice Λ = Hom(Tb, Gm ) is equal to Hom(Gm , T ). Then the
Langlands correspondence gives a bijection χ 7→ σ(χ) between
×
• characters BunT (Fq ) → Q` with finite image
• continuous morphisms π1 (X, η) → Tb(Q` ) with finite image
characterized by the fact that for any place v of X and any λ ∈ Λ the image
λ × λ
− Q` is equal to the image of O(v) by Pic(Fq ) →
of σ(χ)(Frobv ) by Tb(Q` ) → −
χ ×
BunT (Fq ) →
− Q` (this condition is the particular case for tori of the condition of
“compatibility with the Satake isomorphism” which we will consider later for all
reductive groups).
The construction of σ(χ) works as in the previous section, except that aZ has
to be replaced by a subgroup Ξ of BunT (Fq ) of finite index which is included
in the kernel of χ, and we now have to use schemes of T -shtukas, defined using
T -bundles instead of line bundles.

5. Reminder on the dual group

Let G be a split reductive group over F . We denote by G b the Langlands dual
group of G. It is the split reductive group over Q` characterized by the fact that
its roots and weights are the coroots and coweights of G, and reciprocally. Here
10 VINCENT LAFFORGUE

are some examples:

G G
b
GLn GLn
SLn P GLn
SO2n+1 Sp2n
Sp2n SO2n+1
SO2n SO2n
and if G is one of the five exceptional groups, G
b is of the same type. Also the
dual of a product of groups is the product of the dual groups.
Definition 5.1. A global Langlands parameter is a conjugacy class of morphisms
σ : Gal(F /F ) → G(Q
b ` ) factorizing through π1 (U, η) for some open dense U ⊂ X,
defined over a finite extension of Q` , continuous and semisimple.
We say that σ is semisimple if for any parabolic subgroup containing its image
there exists an associated Levi subgroup containing it. Since Q` has characteristic
0 this means equivalently that the Zariski closure of its image is reductive [Ser05].
We now define the Hecke operators (the spherical ones, also called unramified,
i.e. without level). They are similar to the Laplace operators on graphs.
Let v be a place of X. If G and G0 are two G-bundles over X we say that (G0 , φ)
is a modification of G at v if φ is an isomorphism between the restrictions of G
and G0 to X \ v. Then the relative position is a dominant coweight λ of G (in the
case where G = GLr it is the r-uple of elementary divisors). Let λ be a dominant
coweight of G. We get the Hecke correspondence
Hv,λ

y h← h→ %
BunG (Fq ) BunG (Fq )

where Hv,λ is the groupoid classifying modifications (G, G0 , φ) at v with relative

position λ and h← and h→ send this object to G0 and G. Then the Hecke operator
acts on functions by pullback by h← followed by pushforward (i.e. sums in the
fibers) by h→ . In other words
Tλ,v : Cccusp (BunG (Fq )/Ξ, Q` ) → Cccusp (BunG (Fq )/Ξ, Q` )
X
f 7→ G 7→ f (G0 )
 

(G0 ,φ)

where the finite sum is taken over all the modifications (G0 , φ) of G at v with
relative position λ.
These operators form an abstract commutative algebra Hv , the so-
called spherical (or unramified) Hecke algebra at v, and this algebra acts on
Cccusp (BunG (Fq )/Ξ, Q` ). This algebra Hv is equal to Cc (G(Ov )\G(Fv )/G(Ov ), Q` )
and it is possible to write its action with the help of adèles. The actions of these
algebras Hv for different v commute with each other.
 SHTUKAS AND LANGLANDS CORRESPONDENCE 11

The Satake isomorphism [Sat63, Car79] can be viewed [Gro98] as a canonical

isomorphism
[V ] 7→ TV,v
from the Grothendieck ring of representations of G b (with coefficients in Q` ) to the
unramifed Hecke algebra Hv , namely we have TV ⊕V 0 ,v = TV,v +TV 0 ,v and TV ⊗V 0 ,v =
TV,v TV 0 ,v . If V is an irreducible representation of G,
b TV,v is a combination of the
Tλ,v for λ a weight of V .

6. Presentation of the main result of [Laf12]

We now explain the construction in [Laf12] of a canonical decomposition of
Q` -vector spaces
M
(6.1) Cccusp (BunG (Fq )/Ξ, Q` ) = Hσ
σ

where the direct sum is taken over global Langlands parameters σ : π1 (X, η) →
G(Q
b ` ). This decomposition is respected by and compatible with the action of
Hecke operators. In fact we construct a commutative Q` -algebra
B ⊂ End(Cccusp (BunG (Fq )/Ξ, Q` ))
containing the image of Hv for all places v and such that each character ν of B
with values in Q` corresponds in a unique way to a global Langlands parameter
σ.
Since B is commutative we deduce a canonical spectral decomposition
M
Cccusp (BunG (Fq )/Ξ, Q` ) = Hν
ν

where the direct sum is taken over characters ν of B with values in Q` and Hν
is the generalized eigenspace associated to ν. By associating to each ν a global
Langlands parameter σ we deduce the decomposition (6.1) we want to construct.
We show in [Laf12] that any σ obtained in this way factorizes through π1 (X, η),
and that the decomposition (6.1) is compatible with the Satake isomorphism at
every place v of X, in the sense that for every representation V of G, b TV,v acts
on Hσ by multiplication by TrV (σ(Frobv )).
The elements of B are constructed with the help of the `-adic cohomology of
stacks of shtukas and are called excursion operators.
In the case of GLr , since every semisimple linear representation is determined
up to conjugation by its character and since the Frobenius elements Frobv are
dense in Gal(F /F ) by the Chebotarev theorem, the decomposition (6.1) is
uniquely determined by its compatibility with the Satake isomorphism.
On the contrary, for some groups G other than GLr , according to Blasius and
Lapid [Bla94, Lap99] it may happen that different global Langlands parameters
correspond to the same characters of Hv for every place v. This comes from
the fact that it is possible to find finite groups Γ and couples of morphisms
σ, σ 0 : Γ → G(Q
b ` ) such that σ and σ 0 are not conjugated but that for any γ ∈ Γ,
0
σ(γ) and σ (γ) are conjugated [Lar94, Lar96].
12 VINCENT LAFFORGUE

Thus for a general group G, the algebra B of excursion operators may not be
generated by the Hecke algebras Hv for all places v and the compatibility of the
decomposition (6.1) with Hecke operators may not characterize it in a unique
way. Therefore we wait for the construction of the excursion operators (done in
section 8) before we write the precise statement of our main result, which will be
theorem 8.4.

7. The stacks of shtukas and their `-adic cohomology

The `-adic cohomology of a variety (over any algebraically closed field of char-
acteristic 6= `) is very similar to the Betti cohomology of a complex variety, but
it has coefficients in Q` (instead of Q for the Betti cohomology). For its defi-
nition Grothendieck introduced the notions of site and topos, which provide an
extraordinary generalization of the usual notions of topological space and sheaf
of sets on it.
To a topological space X we can associate the category whose
• objects are the open subsets U ⊂ X
• arrows U → V are the inclusions U ⊂ V
and we have the notion of a covering of an open subset by a family of open subsets.
A site is an abstract category with a notion of covering of an object by a family
of arrows targetting to it, with some natural axioms. A topos is the category of
sheaves of sets on a site (a sheaf of sets F on a site is a contravariant functor of
“sections of F” from the category of the site to the category of sets, satisfying, for
each covering, a gluing axiom). Different sites may give the same topos.
To define the étale cohomology of an algebraic variety X we consider the étale
site
• whose objects are the étale morphisms
U


X
• whose arrows are given by commutative triangles of étale morphisms,
U / V

~
X
• with the obvious notion of covering.
The étale cohomology is defined with cefficients in Z/`n Z, whence Z` by passing
to the limit, and Q` by inverting `.
The stacks of shtukas, introduced by Drinfeld, play a role analoguous to
Shimura varieties over number fields. But they exist in a much greater generality.
Indeed, while the Shimura varieties are defined over the spectrum of the ring of
integers of a number field and are associated to a minuscule coweight of the
dual group, the stacks of shtukas exist over arbitrary powers of the curve X, and
 SHTUKAS AND LANGLANDS CORRESPONDENCE 13

can be associated to arbitrary coweights, as we will see now. One simple reason
for this difference between function fields and number fields is the following: in
the example of the product of two copies, the product X × X is taken over Fq
whereas nobody knows what the product Spec Z × Spec Z should be, and over
what to take it.
Let I be a finite set and W = i∈I Wi be an irreducible Q` -linear representation
bI (in other words each Wi is an irreducible representation of G).
of G b
We define ChtI,W as the reduced Deligne-Mumford stack over X I whose points
over a scheme S over Fq classify shtukas, i.e.
• points (xi )i∈I : S → X I , called the legs of the shtuka (“les pattes du
chtouca” in French),
• a G-bundle G over X × S,
• an isomorphism
φ:G → (IdX × FrobS )∗ (G)
∼
S S
(X×S)r( i∈I Γxi ) (X×S)r( i∈I Γxi )

(where Γxi denotes the graph of xi ), such that

(7.1) the relative position at xi of the modification φ is bounded
by the dominant coweight of G corresponding to the dominant weight of Wi .
The notion of Deligne-Mumford stack is in algebraic geometry what corresponds
to the topological notion of orbifold. Every quotient of an algebraic variety by a
finite étale group scheme is a Deligne-Mumford stack and in fact ChtI,W is locally
of this form.
Remark 7.1. Compared to the notion of Artin stacks mentioned in remark 2.1,
a Deligne-Mumford stack is a particular case where the automorphism groups of
geometric points are finite groups (instead of algebraic groups).
Remark 7.2. In the case of GL1 , resp. split tori, we had defined schemes of
shtukas. With the above definition, the stacks of shtukas are the quotients of
these schemes by the trivial action of Fq × , resp. T (Fq ).
We denote by HI,W the Q` -vector space equal to the “Hecke-finite” subspace of
the `-adic intersection cohomology with compact support, in middle degree, of the
fiber of ChtI,W /Ξ over a generic geometric point of X I (or, in fact equivalently,
over a generic geometric point of the diagonal X ⊂ X I ). To give an idea of
intersection cohomology, let us say that for a smooth variety it is the same as
the `-adic cohomology and that for (possibly singular) projective varieties it is
Poincaré self-dual. An element of this `-adic intersection cohomology is said to
be Hecke-finite if it belongs to a sub-Z` -module of finite type stable by all Hecke
operators Tλ,v (or equivalently by all Hecke operators TV,v ). Hecke-finiteness is a
technical condition but Cong Xue has proven [Xue17] that HI,W can equivalently
be defined by a cuspidality condition (defined using stacks of shtukas for parabolic
subgroups of G and their Levi quotients) and that it has finite dimension over
Q` .
Drinfeld has constructed “partial Frobenius morphisms” between stacks of
shtukas. To define them we need a small generalization of the stacks of shtukas
14 VINCENT LAFFORGUE

where we require a factorization of φ as a composition of several modifications.

Let (I1 , ..., Ik ) be an ordered partition of I. An example is the coarse partition
(I)
(I) and in fact the stack ChtI,W previously defined is equal to ChtI,W in the
following definition.
1 (I ,...,I )
k
Definition 7.3. We define ChtI,W as the reduced Deligne-Mumford stack
whose points over a scheme S over Fq classify
φ1 φ2 φk−1 φk
(xi )i∈I , G0 −→ G1 −→ · · · −−−→ Gk−1 −→ (IdX × FrobS )∗ (G0 )


(7.2)
with
• xi ∈ (X r N )(S) for i ∈ I,
• for i ∈ {0, ..., k − 1}, Gi is a G-bundle over X × S and we write Gk =
(IdX × FrobS )∗ (G0 ) to prepare the next item,
• for j ∈ {1, ..., k}
φj : Gj−1 → Gj
∼
S S
(X×S)r( i∈I Γxi ) (X×S)r( i∈I Γxi )
j j

is an isomorphism such that the relative position of Gj−1 with respect to Gj

at xi (for i ∈ Ij ) is bounded by the dominant coweight of G corresponding
to the dominant weight of Wi .
1 (I ,...,I )
We can show that the obvious morphism ChtI,W k
→ ChtI,W (which forgets
the intermediate modifications G1 , ..., Gk−1 ) gives an isomorphism at the level of
(I1 ,...,Ik )
intersection cohomology. The interest of ChtI,W is that we have the partial
(I1 ,...,Ik ) (I2 ,...,Ik ,I1 )
Frobenius morphism FrobI1 : ChtI,W → ChtI,W which sends (7.2) to
φ2 φk−1 φ
(x0i )i∈I , G1 −→ · · · −−−→ Gk−1 −→
k
(IdX × FrobS )∗ (G0 )
(IdX × FrobS )∗ (φ1 )
−−−−−−−−−−−→ (IdX × FrobS )∗ (G1 )

where x0i = Frob(xi ) if i ∈ I1 and x0i = xi otherwise. Taking I1 to be a singleton

we get the action on HI,W of the partial Frobenius morphisms. Thanks to an
extra work (using the Hecke-finiteness condition and Eichler-Shimura relations),
we are able in [Laf12] to apply Drinfeld’s lemma, and this endows the Q` -vector
space HI,W with a continuous action of Gal(F /F )I .
For I = ∅ and W = 1 (the trivial representation), we have
(7.3) H∅,1 = Cccusp (BunG (Fq )/Ξ, Q` ).
Indeed the S-points over Cht∅,1 classify the G-bundles G over X × S, equipped
with an isomorphism
φ : G → (IdX × FrobS )∗ (G).
∼

If we see G as a S-point of BunG , (IdX × FrobS )∗ (G) is its image by FrobBunG .

Therefore Cht∅,1 classifies the fixed points of FrobBunG and it is dicrete (i.e.
of dimension 0) and equal to BunG (Fq ). Therefore the `-adic cohomology of
Cht∅,1 /Ξ is equal to Cc (BunG (Fq )/Ξ, Q` ) and in this particular case it is easy to
see that Hecke-finiteness is equivalent to cuspidality, so that (7.3) holds true.
 SHTUKAS AND LANGLANDS CORRESPONDENCE 15

Up to now we defined a vector space HI,W for every isomorphism class of

bI . A construction based on the
irreducible representation W = i∈I Wi of G
geometric Satake equivalence enables to
a) define HI,W functorialy in W
b) understand the fusion of legs
as explained in the next proposition.
Proposition 7.4. a) For every finite set I,
W 7→ HI,W , u 7→ H(u)
is a Q` -linear functor from the category of finite dimensional representations
of GbI to the category of finite dimensional and continuous representations of
Gal(F /F )I .
This means that for every morphism
u : W → W0
bI , we have a morphism
of representations of G
H(u) : HI,W → HI,W 0
of representations of Gal(F /F )I .
b) For each map ζ : I → J between finite sets, we have an isomorphism
∼
χζ : HI,W → HJ,W ζ
which is
• functorial in W , where W is a representation of G bI and W ζ denotes the
J
representation of Gb on W obtained by composition with the diagonal
morphism
bJ → G
G bI , (gj )j∈J 7→ (gζ(i) )i∈I
• Gal(F /F )J -equivariant, where Gal(F /F )J acts on the LHS by the diag-
onal morphism
Gal(F /F )J → Gal(F /F )I , (γj )j∈J 7→ (γζ(i) )i∈I ,
ζ η
• and compatible with composition, i.e. for every I →
− J →
− K we have
χη◦ζ = χη ◦ χζ .
The statement b) is a bit complicated, here is a basic example of it. For every
finite set I we write ζI : I → {0} the tautological map (where {0} is an arbitrary
choice of notation for a singleton). If W1 and W2 are two representations of G, b
the statement of b) provides a canonical isomorphism
∼
(7.4) χζ{1,2} : H{1,2},W1 W2 → H{0},W1 ⊗W2
associated to ζ{1,2} : {1, 2} → {0}. We stress the difference between W1  W2
b 2 and W1 ⊗ W2 which is a representation of G.
which is a representation of (G) b
16 VINCENT LAFFORGUE

Another example of b) is the isomorphism on the left in

χ−1
ζ (7.3)
∅
(7.5) H{0},1 −−
∼→H∅,1 = Cccusp (BunG (Fq )/Ξ, Q` )
which is associated to ζ∅ : ∅ → {0} (the idea of the isomorphism χζ∅ is that
H∅,1 resp. H{0},1 is the cohomology of the stack of shtukas without legs, resp.
with a inactive leg, and that they are equal). Thanks to (7.5) we are reduced to
construct a decomposition
M
(7.6) H{0},1 ⊗Q` Q` = Hσ .
σ

Idea of the proof of proposition 7.4. We denote by ChtI the inductive

limit of Deligne-Mumford stacks over X I , defined as ChtI,W above, but without
the condition (7.1) on the relative position. In other words, and with an extra
letter G0 to prepare the next definition, the points of ChtI over a scheme S over
Fq classify
• points (xi )i∈I : S → X I ,
• two G-bundles G and G0 over X × S,
• a modification φ at the xi , i.e. an isomorphism
φ:G → G0
∼
S S
(X×S)r( i∈I Γxi ) (X×S)r( i∈I Γxi )

• an isomorphism θ : G0 → (IdX × FrobS )∗ (G).

∼

We introduce the “prestack” MI of “modifications on the formal neighborhood of

the xi ”, whose points over a scheme S over Fq classify
• points (xi )i∈I : S → X I ,
• two G-bundles G and G0 on the formal completion X \ × S of X × S in the
neighborhood of the union of the graphs Γxi ,
• a modification φ at the xi , i.e. an isomorphism
φ:G → G0
∼
(X×S)r(
S S .
i∈I Γxi ) (X×S)r( i∈I Γxi )
\ \

The expert reader will notice that for any morphism S → X I , MI ×X I S is the
quotient of the affine grassmannian of Beilinson-Drinfeld over S by Γ(X
\ × S, G).
We have a formally smooth morphism I : ChtI → MI given by restricting G and
G0 to the formal neighborhood of the graphs of the xi and forgetting θ.
The geometric Satake equivalence, due to Lusztig, Drinfeld, Ginzburg and
Mirkovic-Vilonen [BeDr99, MiVi07], is a fundamental statement which constructs
Gb from G and is the cornerstone of the geometric Langlands program. It is a
canonical equivalence of tensor categories between
• the category of perverse sheaves on the fiber of M{0} above any point of
X (where {0} is an arbitrary notation for a singleton)
• the tensor category of representations of G.
b
For the non expert reader we recall that perverse sheaves, introduced in
[BBD82], behave like ordinary sheaves and have, in spite of their name, very
good properties. An example is given by intersection cohomology sheaves of
 SHTUKAS AND LANGLANDS CORRESPONDENCE 17

closed (possibly singular) subvarieties, whose total cohomology is the intersection

cohomology of this subvarieties.
The tensor structure on the first category above is obtained by “fusion of legs”,
thanks to the fact that M{1,2} is equal to M{0} × M{0} outside the diagonal of X 2
and to M{0} on the diagonal. The first category is tannakian and G b is defined as
the group of automorphisms of a natural fiber functor.
This equivalence gives, for every representation W of G bI , a perverse sheave
SI,W on MI , with the following properties:
• SI,W is functorial in W ,
• for every surjective map I → J, SJ,W ζ is canonically isomorphic to the
restriction of SI,W to MI ×X I X J ' MJ , where X J → X I is the diagonal
morphism,
• for every irreducible representation W , SI,W is the intersection cohomology
sheaf of the closed substack of MI defined by the condition (7.1) on the
relative position of the modification φ at the xi .
Then we define HI,W as the “Hecke-finite” subspace of the cohomology with com-
pact support of ∗I (SI,W ) on the fiber of ChtI /Ξ over a geometric generic point
of X I (or, in fact equivalently, over a geometric generic point of the diagonal
X ⊂ X I ). The first two properties above imply a) and b) of the proposition.
The third one and the smoothness of I ensure that, for W irreducible, ∗I (SI,W )
is the intersection cohomology sheaf of ChtI,W and therefore the new definition
of HI,W generalizes the first one using the intersection cohomology of ChtI,W .

8. Excursion operators and the main theorem of [Laf12]

Let I be a finite set. Let (γi )i∈I ∈ Gal(F /F )I . Let W be a representation of
bI and x ∈ W and ξ ∈ W ∗ be invariant by the diagonal action of G.
G b We define
the endomorphism SI,W,x,ξ,(γi )i∈I of (7.5) as the composition

H(x) χ−1
ζ (γi )i∈I χζ H(ξ)
I I
(8.1) H{0},1 −−→ H{0},W ζI −−
∼→ HI,W −−−−→ HI,W −−
→
∼ H{0},W ζI −−→ H{0},1

b and x : 1 → W ζI and ξ : W ζI → 1
where 1 denotes the trivial representation of G,
are considered as morphisms of representations of Gb (we recall that ζI : I → {0}
ζI
is the obvious map and that W is simply the vector space W equipped with the
diagonal action of G).
b
Paraphrasing (8.1) this operator is the composition
• of a creation operator associated to x, whose effect is to create legs at the
same (generic) point of the curve,
• of a Galois action, which moves the legs on the curve independently from
each other, then brings them back to the same (generic) point of the curve,
• of an annihilation operator associated to ξ.
It is called an “excursion operator” because it moves the legs on the curve (this
is what makes it non trivial).
18 VINCENT LAFFORGUE

To W, x, ξ we associate the matrix coefficient f defined by

(8.2) f ((gi )i∈I ) = hξ, (gi )i∈I · xi.
We see that f is a function on G bI invariant by left and right translations by
the diagonal G.b In other words f ∈ O(G\ b G bI /G),
b where G\b GbI /G
b denotes the
coarse quotient, defined as the spectrum of the algebra of functions f as above.
Unlike the stacky quotients considered before, the coarse quotients are schemes
and therefore forget the automorphism groups of points.
For every function f ∈ O(G\b G bI /G)
b we can find W, x, ξ such that (8.2) holds.
We show easily that SI,W,x,ξ,(γi )i∈I does not depend on the choice of W, x, ξ satis-
fying (8.2), and therefore we denote it by SI,f,(γi )i∈I .
The conjectures of Arthur and Kottwitz on multiplicities in vector spaces of
automorphic forms and in the cohomologies of Shimura varieties [Art89, Kot90]
give, by extrapolation to stacks of shtukas, the following heuristics.
Remark 8.1. Heuristically we conjecture that for every global Langlands pa-
rameter σ there exists a Q` -linear representation Aσ of its centralizer Sσ ⊂ Gb
I
(factorizing through Sσ /ZGb ), so that we have a Gal(F /F ) -equivariant isomor-
phism
?
M
Sσ
(8.3) HI,W ⊗Q` Q` = Aσ ⊗Q` WσI
σ

where WσI is the Q` -linear representation of Gal(F /F )I obtained by composition

of the representation W of G bI with the morphism σ I : Gal(F /F )I → G(Q b ` )I ,
and Sσ acts diagonally. We conjecture that (8.3) is functorial in W , compatible
to χζ and that it is equal to the decomposition (7.6) when W = 1 (so that
Hσ = (Aσ )Sσ ).
In the heuristics (8.3) the endomorphism SI,f,(γi )i∈I = SI,W,x,ξ,(γi )i∈I of
χ−1
ζ ?
∅
M
H{0},1 ⊗Q` Q` −−
∼→H∅,1 ⊗Q` Q` = (Aσ )Sσ
σ
Sσ
acts on (Aσ ) by the composition
χ ζ∅ IdA ⊗x (σ(γi ))i∈I
(Aσ )Sσ −−
→(Aσ ⊗ 1)Sσ −−−
∼
σ
−→ (Aσ ⊗ WσI )Sσ −−−−−→ (Aσ ⊗ WσI )Sσ

IdAσ ⊗ξ χ−1
ζ
Sσ ∅
−−−−→ (Aσ ⊗ 1) −−
∼→(Aσ )Sσ
i.e. by the scalar

hξ, (σ(γi ))i∈I · xi = f (σ(γi ))i∈I .
In other words we should have
?

(8.4) Hσ = eigenspace of the SI,f,(γi )i∈I with the eigenvalues f (σ(γi ))i∈I .
The heuristics (8.4) clearly indicates the path to follow. We show in [Laf12] that
the SI,f,(γi )i∈I generate a commutative Q` -algebra B and satisfy some properties
implying the following proposition.
 SHTUKAS AND LANGLANDS CORRESPONDENCE 19

Proposition 8.2. For each character ν of B with values in Q` there exists a

unique global Langlands parameter σ such that for all I, f and (γi )i∈I , we have
(8.5) ν(SI,f,(γi )i∈I ) = f ((σ(γi ))i∈I ).
The unicity of σ in the previous proposition comes from the fact that, for any
integer n, taking I = {0, ..., n}, the coarse quotient G\ bI /G
b G b identifies with the
n
coarse quotient of (G) b by diagonal conjugation by G, b and therefore, for any
n
(γ1 , ..., γn ) ∈ (Gal(F /F )) , (8.5) applied to (γi )i∈I = (1, γ1 , ..., γn ) determines
(σ(γ1 ), ..., σ(γn )) up to conjugation and semisimplification (thanks to [Ric88]).
The existence and continuity of σ are justified thanks to relations and topological
properties satisfied by the excursion operators.
Since B is commutative we obtain a canonical spectral decomposition H{0},1 ⊗Q`
ν Hν where the direct sum is taken over characters ν of B with values
L
Q` =
in Q` . Associating to each ν a unique global Langlands parameter σ as in the
previous proposition, we deduce the decomposition (7.6) we wanted to construct.
We do not know if B is reduced.
Moreover the unramified Hecke operators are particular cases of excursion op-
erators: for every place v and for every irreducible representation V of G b with
character χV , the unramified Hecke operator TV,v is equal to the excursion oper-
ator S{1,2},f,(Frobv ,1) where f ∈ O(G\( b is given by f (g1 , g2 ) = χV (g1 g2−1 ),
b 2 /G)
b G)
and Frobv is a Frobenius element at v. This is proven in [Laf12] by a geometric
argument (essentially a computation of the intersection of algebraic cycles in a
stack of shtukas). It implies the compatibility of the decomposition (7.6) with
the Satake isomorphism at all places.
Remark 8.3. By the Chebotarev density theorem, the subalgebra of B generated
by all the Hecke algebras Hv is equal to the subalgebra generated by the excursion
operators with ]I = 2. The remarks at the end of section 6 show that in general
it is necessary to consider excursion operators with ]I > 2 to generate the whole
algebra B.
Finally we can state the main theorem.
Theorem 8.4. We have a canonical decomposition of Q` -vector spaces
M
(8.6) Cccusp (BunG (Fq )/Ξ, Q` ) = Hσ ,
σ
where the direct sum in the RHS is indexed by global Langlands parameters ,
b ` )-conjugacy classes of morphisms σ : Gal(F /F ) → G(Q
i.e. G(Q b ` ) factorizing
through π1 (X, η), defined over a finite extension of Q` , continuous and semisim-
ple.
This decomposition is uniquely determined by the following property : Hσ is
equal to the generalized eigenspace associated to the character ν of B defined by
(8.7) ν(SI,f,(γi )i∈I ) = f ((σ(γi ))i∈I .
This decomposition is respected by the Hecke operators and is compatible with
the Satake isomorphism at all places v of X.
20 VINCENT LAFFORGUE

Everything is still true with a level (a finite subscheme N of X). We denote

by BunG,N (Fq ) the set of isomorphism classes of G-bundles over X trivialized on
N . Then we have a canonical decomposition
M
(8.8) Cccusp (BunG,N (Fq )/Ξ, Q` ) = Hσ ,
σ

where the direct sum is taken over global Langlands parameters σ : π1 (X r

N, η) → G(Q
b ` ). This decomposition is respected by all Hecke operators and
compatible with the Satake isomorphism at all places of X r N . If G is split we
have, by [NQT11],
(8.9) BunG,N (Fq ) = G(F )\G(A)/KN
(where A is the ring of adèles, O is the ring of integral adèles, ON the ring of
functions on N and KN = Ker(G(O) → G(ON ))). When G is non necessarily
split the RHS of (8.9) must be replaced by a direct sum, indexed by the finite
group ker1 (F, G), of adelic quotients for inner forms of G and in the definition of
global Langlands parameters we must replace G b by the L-group (see [Bor79] for
L-groups).
We have a statement similar to theorem 8.4 with coefficients in F` instead of
Q` .
We can also consider the case of metaplectic groups thanks to the metaplec-
tic variant of the geometric Satake equivalence due to Finkelberg and Lysenko
[FiLy10, Lys14, GaLy16].
Remark 8.5. Drinfeld gave an idea to prove something like the heuristics (8.3)
but it is a bit difficult to formulate the result. Let Reg be the left regular
representation of Gb with coefficients in Q` (considered as an inductive limit of
finite dimensional representations). We can endow the Q` -vector space H{0},Reg
(of infinite dimension in general) with
a) a structure of module over the algebra of functions on the “affine space S
of morphisms σ : Gal(F /F ) → G b with coefficients in Q` -algebras”,
b) an algebraic action of G
b (coming from the right action of Gb on Reg) which
is compatible with conjugation by G b on S.
The space S is not rigorously defined and the rigorous definition of structure a)
is the following. For any finite dimensional Q` -linear representation V of G,
b with
underlying vector space V , H{0},Reg ⊗ V is equipped with an action of Gal(F /F ),
making it an inductive limit of finite dimensional continuous representations of
Gal(F /F ), as follows. We have a G-equivariant
b isomorphism
θ : Reg ⊗V ' Reg ⊗V
f ⊗ x 7→ [g 7→ f (g)g.x]

where G
b acts diagonally on the RHS, and to give a meaning to the formula the
b → V . Therefore
RHS is identified with the vector space of algebraic functions G
 SHTUKAS AND LANGLANDS CORRESPONDENCE 21

we have an isomorphism
θ
H{0},Reg ⊗ V = H{0},Reg ⊗V −−
→H{0},Reg ⊗V ' H{0,1},Reg V
∼
where the first equality is tautological (since V is just a vector space) and the
last isomorphism is the inverse of the fusion isomorphism χζ{0,1} of (7.4). Then
the action of Gal(F /F ) on the LHS is defined as the action of Gal(F /F ) on the
RHS corresponding to the leg 1. If V1 and V2 are two representations of G, b the
two actions of Gal(F /F ) on H{0},Reg ⊗ V1 ⊗ V2 associated to the actions of G b on
V1 and V2 commute with each other and the diagonal action of Gal(F /F ) is the
action associated to the diagonal action of G b on V1 ⊗ V2 . This gives a structure
as we want in a) because if V is as above, x ∈ V , ξ ∈ V ∗ , f is the function on G
b
defined as the matrix coefficient f (g) = hξ, g.xi, and γ ∈ Gal(F /F ) then we say
that Ff,γ : σ 7→ f (σ(γ)), considered as a “function on S”, acts on H{0},Reg by the
composition
Id ⊗x γ Id ⊗ξ
H{0},Reg −−−→ H{0},Reg ⊗ V →
− H{0},Reg ⊗ V −−−→ H{0},Reg .

Any function f on G b can be written as such a matrix coefficient, and the functions
Ff,γ when f and γ vary are supposed to “generate topologically all functions
on S”. The property above P with V1 and V2 implies relations among the Ff,γ ,
namely
P α that F f,γ1 γ2 = F
α f1 ,γ1 Ff2 ,γ2 if the image of f by comultiplication is
α α
α
α f1 ⊗ f2 . In [Zhu17] Xinwen Zhu gives an equivalent construction of the
structure a). Structures a) and b) are compatible in the following sense: the
conjugation gFf,γ g −1 of the action of Ff,γ on H{0},Reg by the algebraic action of
g∈G b is equal to the action of Ff g ,γ where f g (h) = f (g −1 hg).
The structures a) and b) give rise to a “O-module on the stack S/G b of global
Langlands parameters” (such that the vector space of its “global sections on S” is
H{0},Reg ). For any morphism σ : Gal(F /F ) → G(Q b ` ), we want to define Aσ as the
fiber of this O-module at σ (considered as a “Q` -point of S whose automorphism
group in the stack S/G b is Sσ ”). Rigorously we define Aσ as the biggest quotient
of H{0},Reg ⊗Q` Q` on which any function Ff,γ as above acts by multiplication by
the scalar f (σ(γ)), and Sσ acts on Aσ . If the heuristics (8.3) is true it is the
same as Aσ from the heuristics. When σ is elliptic (i.e. when Sσ /ZGb is finite),
σ is “isolated in S” in the sense that it cannot be deformed (among continuous
morphisms whose composition with the abelianization of G b is of fixed finite order)
and, as noticed by Xinwen Zhu, heuristics (8.3) is true when we restrict on both
sides to the parts lying over σ. In general due to deformation of some non elliptic
σ there could a priori be nilpotents, and for example we don’t know how to prove
that B is reduced so we don’t know how to prove the heuristics (8.3).
We can see the heuristics (8.3), and the structures a) and b) above, as an
illustration of the general idea that, in a spectral decomposition, when the points
of the spectrum naturally have automorphism groups, the multiplicities should
be associated to representations of these groups. By contrast the algebra B of
22 VINCENT LAFFORGUE

excursion operators gives the spectral decomposition with respect to the coarse
quotient associated to S/G,
b where we forget the automorphism groups Sσ .

Remark 8.6. The previous remark makes sense although S was not defined.
To define a space like S rigorously it may be necessary to consider continuous
morphisms σ : Gal(F /F ) → G b with coefficients in Z` -algebras where ` is nilpotent
(such σ have finite image), and S would be an ind-scheme over Spf Z` . Then to
define structure a) we would need to consider Reg with coefficients in Z` , and,
for any representation W of G bI with coefficients in Z` , to construct HI,W as a
Z` -module.

9. Local aspects: joint work with Alain Genestier

In [GeLa17], Alain Genestier and I construct the local parameterization up to
semisimplification and the local-global compatibility.
Let G be a reductive group over a local field K of equal characteristics. We
recall that the Bernstein center of G(K) is defined, in two equivalent ways, as
• the center of the category of smooth representations of G(K),
• the algebra of central distributions on G(K) acting as multipliers on the
algebra of locally constant functions with compact support.
On every Q` -linear irreducible smooth representation of G(K), the Bernstein
center acts by a character.
The main result of [GeLa17] associates to any character ν of the Bernstein
center of G(K) with values in Q` a local Langlands parameter σK (ν) up to
semisimplification , i.e. (assuming G split to simplify) a conjugacy class of mor-
phisms Weil(K/K) → G(Q b ` ) defined over a finite extension of Q` , continuous
and semisimple.
We show in [GeLa17] the local-global compatibility up to semisimplication,
whose statement is the following. Let X be a smooth projective and geometrically
irreducible curve over FN q and let N be a level. Then if σ is a global Langlands
parameter and if π = πv is an irreducible representation of G(A) such that
π KN is non zero and appears in Hσ in the decomposition (8.8), then for every
place v de X we have equality between
• the semisimplification of the restriction of σ to Gal(Fv /Fv ) ⊂ Gal(F /F ),
• the semisimple local parameter σK (ν) where ν is the character of the
Bernstein center by which it acts on the irreducible smooth representation
πv of G(K).
We use nearby cycles on arbitrary bases (Deligne, Laumon, Gabber, Illusie,
Orgogozo), which are defined on oriented products of toposes and whose proper-
ties are proven in [Org06] (see also [Ill06] for an excellent survey). Technically we
show that if all the γi are in Gal(Fv /Fv ) ⊂ Gal(F /F
) then the global excursion
operator SI,f,(γi )i∈I ∈ End Cccusp (BunG,N (Fq )/Ξ, Q` ) acts by multiplication by
an element zI,f,(γi )i∈I of the `-adic completion of the Bernstein center of G(Fv )
which depends only on the local data at v. We construct zI,f,(γi )i∈I using stacks
of “restricted shtukas”, which are analogues of truncated Barsotti-Tate groups.
 SHTUKAS AND LANGLANDS CORRESPONDENCE 23

Remark 9.1. In the case of GLr the local correspondence was known by Laumon-
Rapoport-Stuhler [LRS93] and the local-global compatibility (without semisim-
plification) was proven in [Laf02]. Badulescu and Henniart explained us that in
general we cannot hope more that the local-global compatibility up to semisim-
plication.

10. Independence on `
Grothendieck motives (over a given field) form a Q-linear category and unify
the `-adic cohomologies (of varieties over this field) for different `: a motive is “a
factor in a universal cohomology of a variety”. We consider here motives over F .
We conjecture that the decomposition
M
Cccusp (BunG (Fq )/Ξ, Q` ) = Hσ
σ

we have constructed is defined over Q (instead of Q` ), indexed by motivic Lang-

lands parameters σ, and independent on `. This conjecture seems out of reach
for the moment.
The notion of motivic Langlands parameter is clear if we admit the standard
conjectures. A motivic Langlands parameter defined over Q would give rise to a
“compatible” family of morphisms σ`,ι : Gal(F /F ) → G(Q b ` ) for any ` not dividing
q and any embedding ι : Q ,→ Q` . When G = GLr , the condition of compatibility
is straightforward (the traces of the Frobenius elements should belong to Q and
be the same for all ` and ι) and the fact that any irreducible representation (with
determinant of finite order) for some ` belongs to such a family (and has therefore
“compagnons” for other ` and ι) was proven as a consequence of the Langlands
correspondence in [Laf02]. It was generalized in the two following independent
directions
• Abe [Abe13] used the crystalline cohomology of stacks of shtukas to con-
struct crystalline compagnons,
• when F is replaced by the field of rational functions of any smooth variety
over Fq , Deligne proved that the traces of Frobenius elements belong to a
finite extension of Q and Drinfeld constructed these compatible families
[Del12, Dri12, EsKe12].
For a general reductive group G the notion of compatible family is subtle
(because the obvious condition on the conjugacy classes of the Frobenius elements
is not sufficient). In [Dri15] Drinfeld gave the right conditions to define compatible
families and proved that any continuous semisimple morphism Gal(F /F ) →
b ` ) factorizing through π1 (U, η) for some open dense U ⊂ X (and such that
G(Q
the Zariski closure of its image is semisimple) belongs to a unique compatible
family.

11. Conjectures on Arthur parameters

We hope that all global Langlands parameters σ which appear in this decompo-
sition come from elliptic Arthur parameters, i.e. conjugacy classes of continuous
24 VINCENT LAFFORGUE

semisimple morphisms Gal(F /F ) × SL2 (Q` ) → G(Qb ` ) whose centralizer is finite

modulo the center of G.b This SL2 should be related to the Lefschetz SL2 acting
on the intersection cohomology of compactifications of stacks of shtukas. We even
hope a parameterization of the vector space of discrete automorphic forms (and
not only cuspidal ones) indexed by elliptic Arthur parameters.
Moreover we expect that generic cuspidal automorphic forms appear exactly
in Hσ such that σ is elliptic as a Langlands parameter (i.e. that it comes from
an elliptic Arthur parameter with trivial SL2 action). This would imply the
Ramanujan-Petersson conjecture (an archimedean estimate on Hecke eigenval-
ues).
By [DrKe16] the conjectures above would also imply p-adic estimates on Hecke
eigenvalues which would sharper than those in [Laf11].

12. Recent works on the Langlands program for function fields

in relation with shtukas
In [BHKT16] G. Böckle, M. Harris, C. Khare, and J. Thorne apply the results
explained in this text together with Taylor-Wiles methods to prove (in the split
and everywhere unramified situation) the potential automorphy of all Langlands
parameters with Zariski-dense image. Thus they prove a weak form of the “Galois
to automorphic” direction.
In [YuZh15] Zhiwei Yun and Wei Zhang proved analogues of the Gross-Zagier
formula, namely equality between the intersection numbers of some algebraic
cycles in stacks of shtukas and special values of derivatives of L-functions (of
arbitrary order equal to the number of legs).
In [XiZh17] Liang Xiao and Xinwen Zhu construct algebraic cycles in special
fibers of Shimura varieties. Their construction was inspired by the case of the
stacks of shtukas and is already new in this case (it gives a conceptual setting for
the Eichler-Shimura relations used in [Laf12]).

13. Geometric Langlands program

The results explained above are based on the geometric Satake equivalence
[MiVi07], and are inspired by the factorization structures studied by Beilinson and
Drinfeld [BD04]. The geometric Langlands program was pioneered by Drinfeld
[Dri83] and Laumon [Laum87a], and then developped itself in two variants, which
we will discuss in turn.

13.1. Geometric Langlands program for `-adic sheaves. Let X be a smooth

projective curve over an algebraically closed field of characteristic different from
`.
For any representation W of GbI the Hecke functor
φI,W : Dcb (BunG , Q` ) → Dcb (BunG ×X I , Q` )
is given by
φI,W (F) = q1,! q0∗ (F) ⊗ FI,W


 SHTUKAS AND LANGLANDS CORRESPONDENCE 25

q0 q1
where BunG ← − HeckeI − → BunG ×X I is the Hecke correspondence classifying
modifications of a G-bundle at the xi , and FI,W is defined as the inverse image
of SI,W by the natural formally smooth morphism HeckeI → MI .
Let σ be a G-local
b system on X. Then F ∈ Dcb (BunG , Q` ) is said to be an
b I,
eigensheaf for σ if we have, for any finite set I and any representation W of (G)
an isomorphism φI,W (F) → F  Wσ , functorial in W and compatible to exterior
∼

tensor products and fusion. The conjecture of the geometric Langlands program
claims the existence of an σ-eigensheaf F (it should also satisfy a Whittaker
normalization condition which in particular prevents it to be 0). For G = GLr
this conjecture was proven by Frenkel, Gaitsgory, Vilonen in [FGV02, Gai04]
When X, BunG , σ and F are defined over Fq (instead of Fq ), a construction of
Braverman and Varshavsky [BrVa06] produces subspaces of cohomology classes in
the stacks of shtukas and this allows to show that the function given by the trace
of Frobenius on F belongs to the factor Hσ of decomposition (8.6), as explained
in section 15 of [Laf12].
The `-adic setting is truly a geometrization of automorphic forms over func-
tion fields, and many constructions were geometrized: Braverman and Gaits-
gory geometrized Eisenstein series [BG02], and Lysenko geometrized in particular
Rankin-Selberg integrals [Lys02], theta correspondences [Lys04, Lys11, LaLy13],
and several constructions for metaplectic groups [Lys15, Lys17].

13.2. Geometric Langlands program for D-modules. Now let X be a

smooth projective curve over an algebraically closed field of characteristic 0.
A feature of the setting of D-modules is that one can upgrade the statement
of Langlands correspondence to a conjecture about an equivalence between
categories on the geometric and spectral sides, respectively. See [Gai15a] for a
precise statement of the conjecture and [Gai17] for a survey of recent progress.
Such an equivalence can in principle make sense due to the fact that Galois
representations into G,
b instead of being taken individually, now form an algebraic
stack LocSysGb classifying G-local
b systems, i.e. G-bundles
b with connection (by
contrast one does not have such an algebraic stack in the `-adic setting).
On the geometric side, one considers the derived category of D-modules on
BunG , or rather a stable ∞-category enhancing it. It is denoted D-mod(BunG )
and is defined and studied in [DrGa15]. The category on the spectral side is a
certain modification of QCoh(LocSysGb ), the (derived or rather ∞-) category of
quasi-coherent sheaves on the stack LocSysGb . The modification in question is
denoted IndCohNilp (LocSysGb ), and it has to do with the fact that LocSysGb is not
smooth, but rather quasi-smooth (a.k.a. derived locally complete intersection).
The difference between IndCohNilp (LocSysGb ) and QCoh(LocSysGb ) is measured
by singular support of coherent sheaves, a theory developed in [ArGa15]. The
introduction of Nilp in [ArGa15] was motivated by the case of P1 [Laf09] and
the study of the singular support of the geometric Eisenstein series. In terms of
Langlands correspondence, this singular support may also be seen as accounting
for Arthur parameters. More precisely the singularities of LocSysGb are controlled
by a stack Sing(LocSysGb ) over LocSysGb whose fiber over a point σ is the H −1
26 VINCENT LAFFORGUE

of the cotangent complex at σ, equal to HdR 2

(X, bgσ )∗ ' HdR
0
g∗σ ) ' HdR
(X, b 0
(X, b
gσ )
where the first isomorphism is Poincaré duality and the second depends on the
choice of a non-degenerate ad-invariant symmetric bilinear form on b g. Therefore
Sing(LocSysGb ) is identified to the stack classifying (σ, A), with σ ∈ LocSysGb and
A an horizontal section of the local system b gσ associated to σ with the adjoint
representation of G. Then Nilp is the cone of Sing(LocSysGb ) defined by the
b
condition that A is nilpotent. By the Jacobson-Morozov theorem, any such A
is the nilpotent element associated to a morphism of SL2 to the centralizer of
σ in G,
b i.e. it comes from an Arthur parameter. The singular support of a co-
herent sheaf on LocSysGb is a closed substack in Sing(LocSysGb ). The category
IndCohNilp (LocSysGb ) (compared to QCoh(LocSysGb )) corresponds to the condi-
tion that the singular support of coherent sheaves has to be included in Nilp
(compared to the zero section where A = 0). The main conjecture is that there
is an equivalence of categories

(13.1) D-mod(BunG ) ' IndCohNilp (LocSysGb ).

Something weaker is known: by [Gai15a], D-mod(BunG ) “lives" over LocSysGb

in the sense that QCoh(LocSysGb ), viewed as a monoidal category, acts natu-
rally on D-mod(BunG ). Note that QCoh(LocSysGb ) acts on IndCohNilp (LocSysGb )
(one can tensor a coherent complex by a perfect one and obtain a new coherent
complex) and the conjectured equivalence (13.1) should be compatible with the
actions of QCoh(LocSysGb ) on both sides.
Theorem 8.4 (refined in remark 8.5) can be considered as an arithmetic ana-
logue of the fact that D-mod(BunG ) “lives" over LocSysGb (curiously, due to the
lack of an `-adic analogue of LocSysGb , that result does not have an analogue in
the `-adic geometric Langlands program, even if the vanishing conjecture proven
by Gaitsgory [Gai04] goes in this direction). And the fact that Arthur multiplici-
ties formula is still unproven in general is parallel to the fact that the equivalence
(13.1) is still unproven.
When G = T is a torus, there is no difference between QCoh(LocSysTb ) and
IndCohNilp (LocSysTb ). In this case, the desired equivalence QCoh(LocSysTb ) '
D-mod(BunT ) is a theorem, due to Laumon [Laum96].
The formulation of the geometric Langlands correspondence as an equivalence
of categories (13.1), and even more the proofs of the results, rely on substantial
developments in the technology, most of which had to do with the incorporation
of the tools of higher category theory and higher algebra, developed by J. Lurie
in [Lur09, Lur17]. Some of the key constructions use categories of D-modules and
quasi-coherent sheaves on algebro-geometric objects more general than algebraic
stacks (a typical example is the moduli space of G-bundles on X equipped with
a reduction to a subgroup at the generic point of X).

13.3. Work of Gaitsgory and Lurie on Weil’s conjecture on Tamagawa

numbers over function fields. In [GaLu14, GaLu17] (see also [Gai15b]) Gaits-
gory and Lurie compute the cohomology with coefficients in Z` of the stack BunG
 SHTUKAS AND LANGLANDS CORRESPONDENCE 27

when X is any smooth projective curve over an algebraically closed field of char-
acteristic other than `, and G is a smooth affine group scheme over X with con-
nected fibers, whose generic fiber is semisimple and simply connected. They use
in particular a remarkable geometric ingredient, belonging to the same framework
of factorization structures [BD04] (which comes from conformal field theory) as
the geometric Satake equivalence. The Ran space of X is, loosely speaking, the
prestack classifying non-empty finite subsets Z of X. The affine grassmannian
GrRan is the prestack over the Ran space classifying such a Z, a G-bundle G on X,
and a trivialization α of G on X \ Z. Then the remarkable geometric ingredient is
that the obvious morphism GrRan → BunG , (Z, G, α) 7→ G has contractible fibers
in some sense and gives an isomorphism on homology. Note that when k = C
and G is constant on the curve, their formula implies the well-known Atiyah-Bott
formula for the cohomology of BunG , whose usual proof is by analytic means.
Now assume that the curve X is over Fq . By the Grothendieck-Lefschetz trace
formula their computation of the cohomology of BunG over Fq gives a formula for
| BunG (Fq )|, the number of Fq -points on the stack BunG . Note that since BunG
is a stack, each isomorphism class y of points is weighted by Auty1(Fq ) , where Auty
is the algebraic group of automorphisms of y, and Auty (Fq ) is the finite group of
its Fq -points. Although the set of isomorphism classes y of points is infinite, the
weighted sum converges. Gaitsgory and Lurie easily reinterpret | BunG (Fq )| as
the volume (with respect to some measure) of the quotient G(A)/G(F ) (where F
is the function field of X and A is its ring of adèles) and prove in this way, in the
case of function fields, a formula for the volume of G(A)/G(F ) as a product of
local factors at all places. This formula, called the Tamagawa number formula,
had been conjectured by Weil for any global field F .
Over number fields BunG does not make sense, only the conjecture of Weil on
the Tamagawa number formula remains and it had been proven by Kottwitz after
earlier works of Langlands and Lai by completely different methods (residues of
Eisenstein series and trace formulas).

14. Homage to Alexandre Grothendieck (1928-2014)

Modern algebraic geometry was built by Grothendieck, together with his stu-
dents, in the realm of categories: functorial definition of schemes and stacks, Quot
construction for BunG , tannakian formalism, topos, étale cohomology, motives.
His vision of topos and motives already had tremendous consequences and others
are certainly yet to come. He also had a strong influence outside of his school,
as testified by the rise of higher categories and the work of Beilinson, Drinfeld,
Gaitsgory, Kontsevich, Lurie, Voevodsky (who, sadly, passed away recently) and
many others. He changed not only mathematics, but also the way we think about
it.

References
[Abe13] Tomoyuki Abe. Langlands correspondence for isocrystals and existence of crystalline
companion for curves. Preprint, arXiv:1310.0528 (2013)
28 VINCENT LAFFORGUE

[ArGa15] D. Arinkin and D. Gaitsgory. Singular support of coherent sheaves and the geometric
Langlands conjecture. Selecta Math. 21 (2015), 1–199.
[Art89] J. Arthur. Unipotent automorphic representations: conjectures. Orbites unipotentes et
représentations, II. Astérisque 171-172 13–71 (1989)
[Art14] J. Arthur. L-functions and automorphic representations. Proceedings of ICM 2014, 1,
171–197 (2014)
[BaRo17] A. Badulescu and Ph. Roche. Global Jacquet-Langlands correspondence for division
algebras in characteristic p. Int Math Res Notices 7, 2172–2206 (2017)
[BBD82] A. Beilinson, J. Bernstein et P. Deligne. Faisceaux pervers. Astérisque 100, Société
Mathématique de France (1982)
[BeDr99] A. Beilinson and V. Drinfeld. Quantization of Hitchin’s integrable system and Hecke
eigensheaves (1999), available at the address
http://math.uchicago.edu/~mitya/langlands.html
[BD04] A. Beilinson and V. Drinfeld. Chiral algebras. American Mathematical Society Collo-
quium Publications 51 (2004)
[BeGe03] J. Bernstein and S. Gelbart editors. An Introduction to the Langlands Program.
Birkhäuser (2003)
[Bla94] D. Blasius. On multiplicities for SL(n). Israel J. Math. 88, 237–251 (1994)
[Bor79] A. Borel. Automorphic L-functions. Automorphic forms, representations and L-
functions (Corvallis), Part 2, Proc. Sympos. Pure Math., XXXIII, AMS, 27–61 (1979)
[BHKT16] G. Böckle, M. Harris, C. Khare, and J. Thorne. G-localb systems on smooth projec-
tive curves are potentially automorphic. Preprint, arXiv:1609.03491 (2016)
[BG02] A. Braverman and D. Gaitsgory. Geometric Eisenstein series. Invent. Math. 150 (2)
287–384 (2002)
[BrVa06] A. Braverman and Y. Varshavsky. From automorphic sheaves to the cohomology of
the moduli spaces of F -bundles. Unpublished preprint (2006)
[Bum97] D. Bump. Automorphic Forms and Representations. Cambridge University Press
(1997)
[Car79] P. Cartier. Representations of p-adic groups: a survey. Automorphic forms, representa-
tions and L-functions, Corvallis, Proc. Sympos. Pure Math. XXXIII, Amer. Math. Soc.,
111–155 (1979)
[CPS94] J.W. Cogdell and I.I. Piatetski-Shapiro. Converse theorems for GLn . Publ. Math. IHES
79, 157–214 (1994)
[Del12] P. Deligne. Finitude de l’extension de Q engendrée par des traces de Frobenius, en
caractéristique finie. Mosc. Math. J. 12 (3) 497–514 (2012)
[Dri78] V. G. Drinfeld. Langlands’ conjecture for GL(2) over functional fields. Proceedings of the
International Congress of Mathematicians (Helsinki, 1978), 565–574, Acad. Sci. Fennica,
Helsinki, 1980.
[Dri83] V. G. Drinfeld. Two-dimensional `-adic representations of the fundamental group of
a curve over a finite field and automorphic forms on GL(2). Amer. J. Math. 105 (1),
85–114 (1983)
[Dri87] V. G. Drinfeld. Moduli varieties of F -sheaves. Func. Anal. and Appl. 21, 107–122 (1987)
[Dri88] V. G. Drinfeld. Proof of the Petersson conjecture for GL(2) over a global field of char-
acteristic p. Funct. Anal. Appl. 22 (1), 28–43 (1988)
[Dri89] V. G. Drinfeld. Cohomology of compactified manifolds of modules of F-sheaves of rank
2. J. Soviet Math. 46, 1789–1821 (1989)
[Dri12] V. G. Drinfeld. On a conjecture of Deligne. Mosc. Math. J. 12 (3) 515–
542 (2012). See http://www.ihes.fr/jsp/site/Portal.jsp?document_id=3084&
portlet_id=1140 for a lecture at IHES.
[Dri15] V. G. Drinfeld. On the pro-semisimple completion of the fundamental group of a smooth
variety over a finite field. Preprint, arXiv:1509.06059 (2015)
 SHTUKAS AND LANGLANDS CORRESPONDENCE 29

[DrGa15] V. Drinfeld and D. Gaitsgory. Compact generation of the category of D-modules on

the stack of G-bundles on a curve. Camb. J. Math. 3 (1-2) 19–125 (2015)
[DrKe16] V. G. Drinfeld, K. Kedlaya. Slopes of indecomposable F-isocrystals. Preprint,
arXiv:1604.00660 (2016)
[EsKe12] H. Esnault and M. Kerz. A finiteness theorem for Galois representations of function
fields over finite fields (after Deligne). Acta Math. Vietnam. 37 (4) 531–562 (2012)
[FiLy10] M. Finkelberg and S. Lysenko. Twisted geometric Satake equivalence. J. Inst. Math.
Jussieu 9 (4) 719–739 (2010)
[FGV02] E. Frenkel, D. Gaitsgory, K. Vilonen. On the geometric Langlands conjecture. J. Amer.
Math. Soc. 15 (2), 367–417 (2002)
[Fre07] E. Frenkel. Lectures on the Langlands program and conformal field theory. Frontiers in
number theory, physics, and geometry. II, 387–533, Springer (2007)
[Gai04] D. Gaitsgory. On a vanishing conjecture appearing in the geometric Langlands corre-
spondence. Ann. of Math. 160 (2), 617–682 (2004)
[Gai15a] D. Gaitsgory. Outline of the proof of the geometric Langlands conjecture for GL(2).
De la Géométrie Algébrique aux Formes Automorphes (II): Une collection d’articles en
l’honneur du soixantième anniversaire de Gérard Laumon. Astérisque 370 (2015)
[Gai15b] D. Gaitsgory. The Atiyah-Bott formula for the cohomology of the moduli space of
bundles on a curve. arXiv:1505.02331 (2015)
[Gai17] D. Gaitsgory, Progrès récents dans la théorie de Langlands géométrique. Séminaire
Bourbaki. Exp. No. 1109. Astérisque No. 390, 139–168 (2017)
[GaLu14] D. Gaitsgory and J. Lurie. Weil’s Conjecture for Function Fields, Preprint
available at http://math.harvard.edu/~lurie
[GaLu17] D. Gaitsgory and J. Lurie. Weil’s Conjecture for Function Fields-I, Preprint
available at http://math.harvard.edu/~lurie
[GaLy16] D. Gaitsgory and S. Lysenko. Parameters and duality for the metaplectic geometric
Langlands theory. Preprint, arXiv:1608.00284 (2016)
[Gel84] S. Gelbart. An elementary introduction to the Langlands program. Bull. Amer. Math.
Soc. 10 (2), 177–219 (1984)
[GeLa17] A. Genestier et V. Lafforgue. Chtoucas restreints pour les groupes réductifs et
paramétrisation de Langlands locale. Preprint, arXiv:1709.00978 (2017)
[Gro98] B. Gross. On the Satake isomorphism. Galois Representations in Arithmetic Algebraic
Geometry, Cambridge University Press, 223–237, 1998, also available at the address
http://www.math.harvard.edu/~gross/preprints/sat.pdf
[Ill06] L. Illusie. Vanishing cycles over general bases, after P. Deligne, O. Gabber, G. Laumon
and F. Orgogozo. Preprint. http://www.math.u-psud.fr/~illusie/vanishing1b.pdf
(2006)
[Kot90] R. E. Kottwitz. Shimura varieties and λ-adic representations. Automorphic forms,
Shimura varieties, and L-functions (Ann Arbor), Vol. I, 161–209, Perspect. Math., 10,
Academic Press (1990)
[Laf02] L. Lafforgue. Chtoucas de Drinfeld et correspondance de Langlands. Invent. Math. 147
(1), 1–241 (2002)
[Laf09] V. Lafforgue. Quelques calculs reliés à la correspondance de Langlands géométrique
pour P1 . Available from http://vlafforg.perso.math.cnrs.fr/geom.pdf
[Laf11] V. Lafforgue. Estimées pour les valuations p-adiques des valeurs propes des opérateurs
de Hecke. Bull. Soc. Math. France 139 (4), 455–477 (2011)
[Laf12] V. Lafforgue. Chtoucas pour les groupes réductifs et paramétrisation de Langlands
globale. Preprint http://arxiv.org/abs/1209.5352 (2012)
[LaLy13] V. Lafforgue and S. Lysenko. Geometrizing the minimal representations of even or-
thogonal groups. Represent. Theory 17 263–325 (2013)
[Lan70] R. P. Langlands. Problems in the theory of automorphic forms. Lectures in modern
analysis and applications, III, 18–61. Lecture Notes in Math. 170 Springer (1970)
30 VINCENT LAFFORGUE

[Lap99] E. Lapid. Some results on multiplicities for SL(n). Israel J. Math. 112, 157–186 (1999)
[Lar94] M. Larsen. On the conjugacy of element-conjugate homomorphisms. Israel J. Math.
88, 253–277 (1994)
[Lar96] M. Larsen. On the conjugacy of element-conjugate homomorphisms. II. Quart. J. Math.
Oxford Ser. (2) 47 (185), 73–85 (1996)
[Laum87a] G. Laumon Correspondance de Langlands géométrique pour les corps de fonctions.
Duke Math. J. 54 (2), 309–359 (1987)
[Laum87b] G. Laumon. Transformation de Fourier, constantes d’équations fonctionnelles et
conjecture de Weil. Publ. Math. IHES 65, 131–210 (1987)
[Laum96] G. Laumon Transformation de Fourier generalisée. Preprint, arXiv:alg-geom/9603004
[LaMoB99] G. Laumon et L. Moret-Bailly. Champs algébriques. Ergebnisse der Math. 39,
Springer (1999)
[LRS93] G. Laumon, M. Rapoport, U. Stuhler. D-elliptic sheaves and the Langlands correspon-
dence. Invent. Math. 113 (2), 217–338 (1993)
[Laum96] G. Laumon. Cohomology of Drinfeld modular varieties. Part I,II. Cambridge Studies
in Advanced Mathematics, 41 and 56. Cambridge University Press (1997)
[Lau04] E. Lau. On generalised D-shtukas. Dissertation, Rheinische Friedrich-Wilhelms-
Universität Bonn, 2004. Bonner Mathematische Schriften 369. Universität Bonn, Math-
ematisches Institut, Bonn (2004), available at the address
http://www.math.uni-bielefeld.de/~lau/publ.html
[Lau07] E. Lau. On degenerations of D-shtukas. Duke Math. J. 140 (2), 351–389 (2007)
[Lys02] S. Lysenko. Local geometrized Rankin-Selberg method for GL(n). Duke Math. J. 111
(3), 451–493 (2002)
[Lys04] S. Lysenko. Moduli of metaplectic bundles on curves and theta-sheaves. Ann. Sci. École
Norm. Sup. 39 (3), 415–466 (2006)
[Lur09] J. Lurie. Higher Topos Theory, Princeton University Press, 2009.
[Lur17] J. Lurie. Higher Algebra, available at http://math.harvard.edu/~lurie
[Lys11] S. Lysenko. Geometric theta-lifting for the dual pair SO2m ,Sp2n . Ann. Sci. École Norm.
Sup. 44 (3), 427–493 (2011)
[Lys14] S. Lysenko. Twisted geometric Satake equivalence: reductive case. preprint,
arXiv:1411.6782 [math.RT] (2014)
[Lys15] S. Lysenko. Twisted geometric Langlands correspondence for a torus. Int. Math. Res.
Not. 18, 8680–8723 (2015)
[Lys17] S. Lysenko. Twisted Whittaker models for metaplectic groups. Geom. Funct. Anal. 27
(2), 289–372 (2017), with erratum in arXiv:1509.02433
[MiVi07] I. Mirkovic and K. Vilonen. Geometric Langlands duality and representations of al-
gebraic groups over commutative rings. Annals of Math. 166, 95–143 (2007)
[NBC99] Ngô Bao Châu. Faisceaux pervers, homomorphisme de changement de base et lemme
fondamental de Jacquet et Ye. Ann. Sci. École Norm. Sup. 32 (5), 619–679 (1999)
[NBC06a] Ngô Bao Châu. D-chtoucas de Drinfeld à modifications symétriques et identité de
changement de base. Ann. Sci. École Norm. Sup. 39 (2), 197–243 (2006)
[NgNg08] Ngô Bao Châu et Ngo Dac Tuan. Comptage de G-chtoucas: la partie régulière ellip-
tique. J. Inst. Math. Jussieu 7 (1), 181–203 (2008)
[NDT07] Ngo Dac Tuan. Compactification des champs of chtoucas et théorie géométrique des
invariants. Astérisque 313 (2007)
[NDT09] Ngo Dac Tuan. Sur le développement spectral de la formule des traces d’Arthur-
Selberg pour les corps of fonctions. Bulletin de la Soc. Math. France 137, 545–586
(2009)
[NQT11] Nguyen Quoc Thang. On Galois cohomology and weak approximation of connected
reductive groups over fields of positive characteristic. Proc. Japan Acad. Ser. A Math.
Sci. 87 (10), 203–208 (2011)
 SHTUKAS AND LANGLANDS CORRESPONDENCE 31

[Org06] F. Orgogozo. Modifications et cycles proches sur une base générale. Int. Math. Res.
Not. 2006, 1–38 (2006)
[Sat63] I. Satake. Theory of spherical functions on reductive algebraic groups over p-adic fields.
Publ. Math. IHES 18, 5–69 (1963)
[Ric88] R. W. Richardson. Conjugacy classes of n-tuples in Lie algebras and algebraic groups.
Duke Math. J. 57, 1–35 (1988)
[Ser75] J.-P. Serre. Groupes algébriques et corps de classes. (1975)
[Ser05] J.-P. Serre. Complète réductibilité. Séminaire Bourbaki. Vol. 2003/2004. Astérisque No.
299 (2005)
[Tay04] R. Taylor. Galois representations (expanded version of ICM 2002 lecture) Annales de
la Faculté des Sciences de Toulouse, Vol. XIII, 1, 73–119 (2004)
[Var04] Y. Varshavsky. Moduli spaces of principal F -bundles. Selecta Math. (N.S.) 10 (1),
131–166 (2004)
[XiZh17] Liang Xiao and Xinwen Zhu. Cycles on Shimura varieties via geometric Satake.
Preprint. arXiv:1707.05700 (2017)
[Xue17] Cong Xue. Cohomologie cuspidale des champs de chtoucas. Thèse de doctorat, Uni-
versité Paris-Saclay (2017)
[YuZh15] Zhiwei Yun and Wei Zhang. Shtukas and the Taylor expansion of L-functions.
Preprint. arXiv:1512.02683 (2015)
[Zhu17] Xinwen Zhu. Pseudo representations for reductive groups Unpublished notes (2017)

Vincent Lafforgue: CNRS et Institut Fourier, UMR 5582, Université Greno-

ble Alpes, 100 rue des Maths, 38610 Gières, France.