Reductive group schemes over the Fargues-Fontaine curve
Johannes Ansch\"utz

TL;DR
This paper classifies reductive group schemes over the Fargues-Fontaine curve for any non-archimedean local field, extending existing theorems to equal characteristic and providing a classification of torsors via a generalized Kottwitz set.
Contribution
It introduces a classification of reductive group schemes over the Fargues-Fontaine curve using isocrystals and generalizes Kottwitz' set B(G) for torsors, extending Fargues' theorem to equal characteristic.
Findings
Classification of reductive group schemes over the Fargues-Fontaine curve.
Extension of Fargues' theorem to equal characteristic.
Generalization of Kottwitz' set B(G) for torsors.
Abstract
For an arbitrary non-archimedean local field we classify reductive group schemes over the corresponding Fargues-Fontaine curve by group schemes over the category of isocrystals. We then classify torsors under such reductive group schemes by a generalization of Kottwitz' set B(G). In particular, we extend a theorem of Fargues on torsors under constant reductive groups to the case of equal characteristic.
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsAlgebraic Geometry and Number Theory · Advanced Algebra and Geometry · Alkaloids: synthesis and pharmacology
Reductive group schemes over the Fargues-Fontaine curve
Johannes Anschütz
Abstract.
For an arbitrary non-archimedean local field we classify reductive group schemes over the corresponding Fargues-Fontaine curve by group schemes over the category of isocrystals. We then classify torsors under such reductive group schemes by a generalization of Kottwitz’ set . In particular, we extend a theorem of Fargues on torsors under constant reductive groups to the case of equal characteristic.
1. Introduction
Let be a non-archimedean local field and let be a reductive group over . If is -adic L. Fargues classified -torsors on the corresponding Fargues-Fontaine curve (cf. [7]) associated to (and an algebraically closed perfectoid extension of the residue field of ). His result is phrased in terms of R. Kottwitz’ set associated with (cf. [14]). For a general non-archimedean local field let
[TABLE]
be the completion of the maximal unramified extension of and let
[TABLE]
be the Frobenius of . In [14] (cf. [13]) R. Kottwitz defined
[TABLE]
as the set of conjugacy classes in . Our first main theorem is the classification of -torsors on the Fargues-Fontaine curve by this set generalizing L. Fargues’ result for -adic.
Theorem 1.1** (cf. Theorem 3.11).**
There exists a canonical bijection
[TABLE]
We mention that this theorem implies the computation of the Brauer group of (cf. [7, Théorème 2.6.] resp. Theorem 4.2) and that it is a starting point for L. Fargues’ program to geometrize the local Langlands correspondence over (cf. [6]). Similar to [7] we present in Section 4 applications of Theorem 1.1 to the étale and flat cohomology of with locally constant torsion coefficents.
In [7, Remarque 1.2.] L. Fargues asked the question how to generalize Theorem 1.1 about -torsors on to the case where is no longer assumed to be constant, but an arbitrary reductive group scheme over . Examples of possibly non-constant reductive group schemes over the Fargues-Fontaine curve can be constructed as follows. Let be the category of isocrystals over and let be a Hopf algebra in (such a Hopf algebra is also called an affine group scheme over (cf. Definition 5.1)). After fixing an embedding there exists a canonical tensor functor (cf. [8, Section 8.2.3.] resp. [12])
[TABLE]
from the category to the category of vector bundles on . The image of is then the Hopf algebra associated with a flat group scheme over . We call this group scheme the “group scheme associated with the affine group scheme over ”. Our first theorem about reductive group schemes over the Fargues-Fontaine curve is the following classification result.
Theorem 1.2** (cf. Theorem 5.3).**
Every reductive group scheme over the Fargues-Fontaine curve is associated with a, necessarily reductive, group scheme over .
The proof of Theorem 1.2 is straightforward, based on the classification of torsors for reductive groups (cf. Theorem 1.1) and the geometric simply-connectedness of (cf. Theorem 4.1).
However, reductive group schemes over are rather close to being constant (cf. Lemma 6.3). For example, they become constant after removing a closed point of (cf. Lemma 6.3).
Based on Theorem 1.2 we can define for a reductive group scheme over a natural candidate of a set generalizing Kottwitz’ definition in the constant case (cf. Definition 3.9 and Proposition 5.7). Namely, if is associated with the group scheme we can set as the set of -conjugacy classes in where is the reductive group over associated with the Hopf algebra .111Later we call this set .
Our second theorem about reductive group schemes over is then the following classification of torsors.
Theorem 1.3** (cf. Theorem 5.9).**
For a reductive group scheme over there is a canonical bijection
[TABLE]
The proof of Theorem 1.1 and Theorem 1.3 is based on an excessive use of the Tannakian formalism of -graded and -filtered fiber functors, which is an easy adaptation of [18] (cf. Section 2, and the classification of vector bundles on (cf. Theorem 3.5), especially the fact that for a semistable vector bundle of positive slope the cohomology group
[TABLE]
vanishes. The crucial input in our proof is the fact that sending a vector bundle to its Harder-Narasimhan filtration defines a fully faithful, unfortunately non-exact, tensor functor
[TABLE]
from ordinary vector bundles on to -filtered vector bundles on .
Finally, we discuss uniformization results (cf. Section 6) for -torsors under arbitrary reductive group schemes over generalizing the known case that is constant, quasi-split over with -adic (cf. [7, Théorème 7.2.]).
Theorem 1.4** (cf. Theorem 6.5).**
Let be a reductive group scheme over . Then every -torsor becomes trivial after removing a closed point of .
Acknowledgement
The author wants to thank Laurent Fargues, Jochen Heinloth, Michael Rapoport, Peter Scholze and Torsten Wedhorn for answering several questions related to this paper. Especially, the hint of Michael Rapoport to [4, Theorem 5.3.1.] lead to the full proof of Theorem 3.11 in the equal characteristic case.
2. -filtered fiber functors
In this section we want to extend the results of [18] about -filtered fiber functors to -filtered fiber functors. Filtrations on functors indexed by are also discussed in [4], e.g., in Chapter IV.2, or [3].
We follow the definitions and notations of [18, Chapter 2], but for us fiber functors will take values in the category of locally free sheaves of finite ranks and not in the category of arbitrary quasi-coherent modules.222which is not a restriction as fiber functors take their image in vector bundles
Let be a field and let be a Tannakian category over . Let be a totally ordered abelian group. Mainly, we will be interested in the case that is a subgroup. We start by recalling the definition of a -graded fiber functor (cf. [16, Chapitre IV.1]). For this we denote by the tensor category of -graded vector bundles on a -scheme . Equivalently, if denotes the constant multiplicative group (over ) with character group , then the category is equivalent to the category of representations of over on locally free sheaves of finite rank.
Definition 2.1**.**
Let be a scheme over . A -graded fiber functor of over is an exact tensor functor
[TABLE]
Equivalently, a -graded fiber functor on over consists of a usual fiber functor
[TABLE]
i.e., is an exact tensor functor, together with a homomorphism of group schemes over where denotes the group scheme of tensor automorphisms of (cf. [16, Chapitre IV.1]). If is non-empty, then each -graded fiber functor is automatically faithful (cf. [5, Corollarie 2.10.ii)]).
Now we define -filtered fiber functors. We start by defining the category of -filtered vector bundles on a -scheme .
Definition 2.2**.**
Let be a scheme over . A -filtered vector bundle on is a vector bundle on together with subbundles, i.e., locally direct summands,
[TABLE]
for such that
[TABLE]
if and for resp. for . Morphisms of -filtered vector bundles are morphisms
[TABLE]
of vector bundles respecting the subsheaves , i.e.,
[TABLE]
The category of -filtered vector bundles on is denoted by .
If is a -filtered bundle we denote by for the vector bundle
[TABLE]
of . We moreover denote by
[TABLE]
where
[TABLE]
the associated -graded vector bundle on . In this way we obtain a functor
[TABLE]
We note that for a -filtered vector bundle there are, locally on , only finitely many such that
[TABLE]
If and are two -filtered vector bundles on then we define their tensor product
[TABLE]
by setting
[TABLE]
Moreover, we make into an exact category (in the sense of Quillen) by requiring that a sequence
[TABLE]
is exact if and only if the associated sequence
[TABLE]
is exact. In particular, the functor
[TABLE]
is then an exact functor. Moreover, is even a tensor functor. We can also go into the other direction and associate to a -graded vector bundle a filtered one. Namely, for a -graded vector bundle
[TABLE]
we can define the -filtered vector bundle
[TABLE]
by setting
[TABLE]
In this way we obtain an exact tensor functor
[TABLE]
We are now able to define what a -filtered fiber functor on is (cf. [18, Section 4.1]).
Definition 2.3**.**
Let be a scheme over .
i) A -filtered fiber functor of over is an exact tensor functor
[TABLE]
ii) A splitting of a -filtered fiber functor is a -graded fiber functor
[TABLE]
such that .
We remark that if a -filtered fiber functor admits a splitting then also -filtered fiber functors isomorphic to do. Moreover, if admits a splitting , then .
Clearly, for every morphism of schemes over the pullback
[TABLE]
of vector bundles induces exact tensor functors
[TABLE]
resp.
[TABLE]
Moreover,
[TABLE]
and analogous . Let
[TABLE]
resp.
[TABLE]
be the fibered categories of -filtered resp. -graded fiber functors for . As in [18, Lemma 4.32, Lemma 3.12.] it follows that both fibered categories are stacks for the fpqc-topology. Moreover, resp. define morphisms
[TABLE]
resp.
[TABLE]
of stacks. Finally, for a -filtered fiber functor
[TABLE]
we denote by
[TABLE]
the functor on -schemes sending to the set of splittings of the -filtered fiber functor . The functor is a sheaf for the fpqc-topology.
The main result of [18, Main theorem 4.14] is the following (in the case we are interested in, i.e., the band of is reductive, also cf. [3]).
Theorem 2.4**.**
Every -filtered fiber functor is splittable, i.e. admits a splitting, fpqc-locally on .
We want to extend Theorem 2.4 to more general groups . As in [18, Definition 4.8.] we make the following definitions.
Definition 2.5**.**
Let
[TABLE]
be a -filtered fiber functor on the Tannakian category . We define the following group sheaves over .
i)
ii)
iii)
Moreover, we define a canonical filtration
[TABLE]
as follows. For , we set as the subgroup of consisting of elements which act trivially on
[TABLE]
for all and .
Clearly, for the group
[TABLE]
is normal. We denote by
[TABLE]
the union in of all subgroups for and by
[TABLE]
the quotient
[TABLE]
We will use the following argument to deduce results for -filtered fiber functors where is a subgroup from results about -filtered fiber functors. Let be a subgroup and let be a -filtered fiber functor. Define
[TABLE]
Then is a subgroup as and are tensor functors. If has a tensor generator, then (by exactness of and ) the group is finitely generated by the such that for a tensor generator . Hence, is isomorphic to or in this case because . Moreover, there are fully faithful embeddings
[TABLE]
resp.
[TABLE]
and factors through a -filtered fiber functor
[TABLE]
Moveover,
[TABLE]
etc. Thus all data for is defined by a -filtered fiber functor to which we can apply the known results. In particular, we can conclude by Theorem 2.4 that if has a tensor generator, then every -filtered fiber functor admits a splitting, fpqc-locally on . We record the following theorem collecting results about -filtered fiber functors.
To state it let be the fundamental group of (cf. [5, Definition 8.13]), an affine group scheme in the Tannakian category represented by a Hopf algebra in Ind-. For every fiber functor it has the property
[TABLE]
as affine group schemes over . More precisely, the -Hopf algebra representing is isomorphic to the Hopf algebra . If admits a tensor generator we define the Lie algebra
[TABLE]
where is the augmentation ideal of . If for some affine group scheme over , then
[TABLE]
with acting on by conjugation and we see that we recover the usual notion of the Lie algebra of with its adjoint action.
Theorem 2.6**.**
Let be a subgroup. Let be a Tannakian category over and let
[TABLE]
be a -filtered fiber functor. Define
[TABLE]
as the group scheme over defined by the usual fiber functor
[TABLE]
*of over . Let be the fundamental group of . Then
i) For , , the group sheaf is representable by a group scheme, affine and faithfully flat over .
ii) The affine group schemes and for are pro-unipotent and for the group is abelian and pro-unipotent.
iii)
iv) If admits a tensor generator, then for *
[TABLE]
and
[TABLE]
*where is the Lie algebra of the fundamental group of .
v) If is of finite presentation over (or equivalently, if has a tensor generator), then*
[TABLE]
*for are of finite presentation.
vi) If is smooth over , then , , and for are smooth.
vii) If is reductive over , then is a parabolic subgroup scheme of and is its unipotent radical. For the groups*
[TABLE]
are vector bundles, isomorphic to
[TABLE]
*viii) The sheaf of splittings of is an -torsor over with respect to the fpqc-topology, in particular, admits a splitting fpqc-locally on and is represented by a scheme affine and faithfully flat over .
ix) If*
[TABLE]
is the associated -graded fiber functor of , then for
[TABLE]
Proof.
i) If admits a tensor generator this follows from [18, Lemma 4.20] and the discussion preceeding this theorem. In the general case, writing as a union of sub-Tannakian categories admitting a tensor generator, shows that is an inverse limit of schemes, affine and faithfully flat over and hence itself affine and faithfully flat over .
ii) The pro-unipotence of and for follows after taking the limit from the case that has a tensor generator, say . Then resp. embeds into as a subgroup of the upper triangular matrices, showing the unipotence. As in [18, Lemma 4.21] it follows that is abelian. As is a quotient of it is also pro-unipotent.
iii) This follows as in [18, Lemma 4.23] from viii).
iv) This follows from [16, Proposition IV.2.1.4.1].
v) This follows from [16, Proposition IV.2.1.4.1].
vi) This follows from [18, Lemma 4.20].
vii) This follows from [18, Lemma 4.40] and [18, Proposition 4.25].
viii) By Theorem 2.4, and the discussion preceeding this theorem, the statement is known if admits a tensor generator. Moreover, the sheaf is represented by a scheme, affine and faithfully flat over by [18, Lemma 4.20]. The general case follows by the usual limit argument.
ix) For there exists a canonical map
[TABLE]
and thus the statement that this is an isomorphism is fpqc-local on and we may assume that
[TABLE]
is split in which case the statement is trivial. ∎
3. Admissible fiber functors over the Fargues-Fontaine curve
Fix a local field with residue field , i.e., either or is a finite extension of . Moreover, let be a perfectoid algebraically closed extension of an algebraic closure of . Let be the schematic Fargues-Fontaine curve associated with and (cf. [6, Chapter 1]). In the notation we will omit the field and we will just write for . Our results will not depend on the field (assuming that it is algebraically closed).
Let be a Tannakian category over . The basic example will be the category of representations of an affine group scheme over . We record the following lemma from the Tannakian formalism (cf. [5]).
Lemma 3.1**.**
Let be a field and let be an affine group scheme. Then for every scheme over the groupoid of fiber functors
[TABLE]
is equivalent to the groupoid of -torsors over for the fpqc-topology. If is locally of finite presentation, the same holds with “fpqc” replaced by “fppf”. If is locally smooth, the same holds with “fpqc” replaced by “étale”.
The following lemma is very important for our proof of the classification of the later defined admissible fiber functors
[TABLE]
of over . First we recall the Harder-Narasimhan filtration for a vector bundle on (cf. [8, Section 8.2.4.] resp. [12, Corollary 11.7.]). Namely, there exists a canonical -filtration on by subbundles
[TABLE]
such that for the graded piece
[TABLE]
is semistable of slope .
Lemma 3.2**.**
Sending a vector bundle to the -filtered vector bundle
[TABLE]
defines a fully faithful tensor functor
[TABLE]
Proof.
The Harder-Narasimhan filtration is preserved by every morphism of vector bundles. Hence the functor is well-defined and fully faithful. The classification of vector bundles on (cf. [8, Chapter 8], [12] resp. Theorem 3.5) implies that the tensor product of two semistable vector bundles of slope resp. is again semistable of slope . This implies that is moreover a tensor functor. ∎
However, note that the functor is not exact, because semistable vector bundles are successive extensions of line bundles which are possibly of different slopes. Therefore we make the following definition.333In [16] Saavedra-Rivano calls filtered fiber functors admissible if they are fpqc-locally splittable. By [18] this notion is obsolete and thus we think that our terminology is not very confusing. Also our admissible fiber functors are not equipped with a filtration as would be the case in Saavedra-Rivano’s notation.
Definition 3.3**.**
A fiber functor is called admissible if the composition
[TABLE]
is again exact, i.e., a -filtered fiber functor.
For example, if and for a reductive group over , then every fiber functor is admissible as is semisimple in this case.
Admissibility is a convenient notion as it can be checked after base change along a finite field extension .
Lemma 3.4**.**
Let be a finite field extension and let be a fiber functor. Then is admissible if and only if the composition
[TABLE]
is admissible where is the canonical morphism.
Proof.
Because is faithfully flat it suffices to show that the diagram
[TABLE]
commutes, i.e., if is a vector bundle on , then the pullback of the Harder-Narasimhan filtration of on is the Harder-Narasimhan filtration of on . We assume first that over is separable. Let be a vector bundle on . If is the slope of , then is the slope of because preserves ranks, but multiplies degrees by . In particular, if is semistable, then is semistable, because for a destabilizing subbundle of the pullback would again be destabilizing. This implies that it suffices to prove the claim in the case that is Galois. Then the Harder-Narasimhan filtration of must be stable under the Galois action and hence descends to . Moreover, this descended filtration must be the Harder-Narasimhan filtration of as its graded pieces are semistable of strictly decreasing slopes because this holds after pullback to . In particular, the pullback of the Harder-Narasimhan filtration of is the Harder-Narasimhan filtration of .
Now assume that is purely inseparable. Again it is clear that a vector bundle is semistable if is semistable. Conversely, let be a semistable vector bundle. We want to show that is semistable. For this we may assume that is simple. In this case there exists and , s.t.
[TABLE]
where is an unramified extension of of degree and
[TABLE]
is the canonical morphism. As is purely inseparable, is isomorphic to the unramified extension of of degree . In particular,
[TABLE]
where is the projection, is again semistable. ∎
Let
[TABLE]
be the completion of the maximal unramified extension of and let
[TABLE]
be the category of isocrystals over . As we have fixed an embedding we obtain a canonical exact tensor functor
[TABLE]
(cf. [8, Section 8.2.3.] if resp. [12, Chapter 8] if ).
We recall properties of this functor.
Theorem 3.5**.**
The functor
[TABLE]
is an exact faithful tensor functor inducing a bijection on isomorphism classes. Moreover, for it induces an equivalence between semistable isocrystals of slope with semistable vector bundles of slope .
Proof.
Cf. [8, Section 8.2.4.] resp. [12, Proposition 8.6., Theorem 11.1.]. ∎
One has to be a bit careful when comparing the slope of some isocrystal
[TABLE]
with the slope of the vector bundle . If is a uniformizer then for the isocrystal
[TABLE]
is of slope and sent to the line bundle and not to . This explains the appearance of the sign in Theorem 3.5.
Theorem 3.5 implies that the functor
[TABLE]
preserves duals and symmetric resp. exterior powers because this is true for the functor sending an isocrystal to its decomposition into isoclinic components.
Moreover, the category is canonically -graded by decomposing an isocrystal into its isoclinic components. We thus obtain a functor
[TABLE]
Lemma 3.6**.**
The functor identifies the category with the full subcategory of consisting of -graded vector bundles
[TABLE]
such that for the vector bundle is semistable of slope .
Proof.
This follows from the classification of vector bundles on and their homomorphisms (cf. Theorem 3.5). ∎
We moreover see that the functor
[TABLE]
is fully faithful and exact as it is isomorphic to the functor . In the following we will sometimes just write instead of .
We now start to classify admissible fiber functors.
Lemma 3.7**.**
Let be an exact tensor functor. Then the fiber functor
[TABLE]
is admissible.
Proof.
It suffices to show that the tensor functor
[TABLE]
is exact. But as was noted above the functor
[TABLE]
is exact, which implies the claim. ∎
Let be a Tannakian category and let
[TABLE]
be a fiber functor of over a non-empty scheme . If the group scheme of tensor automorphisms is smooth over , then this is true for every fiber functor of as fiber functors are fpqc-locally isomorphic. Hence, this property of smoothness is in fact intrinsic to (and can be rephrased by saying that the band of the gerbe associated to is smooth).
We now prove our main theorem about admissible functors.
Theorem 3.8**.**
Let be a Tannakian category admitting a tensor generator and let
[TABLE]
be an admissible fiber functor such that is represented by a reductive group scheme over . Then factors as
[TABLE]
for an exact tensor functor
[TABLE]
Proof.
We consider the tensor functor
[TABLE]
By assumption this tensor functor is a -filtered fiber functor, i.e., it is exact. Therefore it gives rise to the -torsor
[TABLE]
of splittings of (cf. Theorem 2.6). Moreover, the group scheme over is smooth (by Theorem 2.6) as is assumed to be smooth, unipotent and admits a filtration
[TABLE]
for . Moreover, for the associated graded pieces are vector bundles with
[TABLE]
where is the fundamental group of (cf. Theorem 2.6). In particular,
[TABLE]
is semistable of slope . As only finitely many are non-zero we can conclude
[TABLE]
as for every semistable vector bundle of slope on the cohomology group
[TABLE]
vanishes (cf. [8, Chapter 8] resp. [12]). In particular, the -torsor -torsor is trivial and we see that we can choose a splitting
[TABLE]
of , i.e.,
[TABLE]
By construction for every and every the locally free sheaf
[TABLE]
must be semistable of slope . In other words (using Lemma 3.6), the functor
[TABLE]
factors as
[TABLE]
for an exact tensor functor
[TABLE]
Forgetting the grading shows , which proves the claim. ∎
Definition 3.9**.**
Let be a Tannakian category over . We define as the groupoid of exact tensor functors
[TABLE]
and as the set of isomorphism classes of such.
By [4, Lemma 9.1.4.] (cf. Proposition 5.7) this definition agrees with Kottwitz original definition of the set as -conjugacy classes in if for a connected reductive group over .
Recall that we denote by
[TABLE]
the category of fiber functors of over . For we let
[TABLE]
be the full subcategory of admissible fiber functors of over . From Theorem 3.8 we obtain the following classification of admissible fiber functors.
Theorem 3.10**.**
The composition of the functors
[TABLE]
and
[TABLE]
is naturally isomorphic to the identity. These functors induce a bijection of with the set of isomorphism classes in .
Proof.
The first statement is clear by Lemma 3.6. By Theorem 3.8 the canonical map from to isomorphism classes of admissible fiber functors is surjective, hence bijective by the first statement. ∎
Let be a representation. We will need to consider the symmetric tensors of ([15, Chapitre 3]), i.e., is the set of invariants for the permutation action of on . By [15, Théorème IV.1., Proposition IV.5.] the vector space has the following universal property: For every homogenous polynomial
[TABLE]
of degree there exists a unique linear form
[TABLE]
such that for all . If is -invariant, then will be -invariant as well.
From a different view point the module is isomorphic to the -the divided power of ([15, Proposition IV.5.]). It is clear from the definition that the module of symmetric tensors of the dual of is canonically isomorphic (as a -representation) to the dual of the -th symmetric power of .
Using these symmetric tensors we obtain the following result, generalizing the main theorem of [7] if .
Theorem 3.11**.**
Let be a reductive group over . Then there is a canonical bijection
[TABLE]
In other words, every fiber functor
[TABLE]
is admissible.
Proof.
By Theorem 3.10 it suffices to prove the second statement that every fiber functor
[TABLE]
is automatically admissible because the isomorphism classes of fiber functors of over are in bijection with -torsors (for the étale topology) over (cf. Lemma 3.1). Let
[TABLE]
be a fiber functor. We want to show that the composite
[TABLE]
is still exact. Equivalently, we can check this after taking the associated graded, i.e., we can check that the functor
[TABLE]
is exact. As was remarked after Theorem 3.5 all of the functors and are compatible with duals and symmetric resp. exterior powers. Moreover, they preserve modules of symmetric tensors, i.e., -invariants in the -th tensor powers. Using this a slightly modified proof as in [4, Theorem 5.3.1.] works. We fill out the details. Let
[TABLE]
be an exact sequence in . Set
[TABLE]
It suffices to prove that the morphism
[TABLE]
is injective. Indeed, dualizing the sequence shows that then the morphism
[TABLE]
has a torsion cokernel. But this morphism is a direct sum of morphisms between semistable vector bundles of the same slope, which implies that this morphism is already surjective. Hence the morphism
[TABLE]
is surjective, which then implies that the sequence
[TABLE]
is exact, because
[TABLE]
Hence, we are left with proving that the morphism
[TABLE]
is injective. But is injective if and only this is true after taking the exterior power . As the functors and preserve exterior powers this reduces the claim to the case that is of dimension one. Tensoring with the dual of reduces to the case that is a trivial -representation. It suffices to proof that is non-zero. By Haboush’s theorem (cf. [11]) there exists some and a -equivariant homogenous polynomial
[TABLE]
such that . Using the universal property of we obtain a -equivariant linear form
[TABLE]
such that is non-zero when restricted to . In particular, the morphism
[TABLE]
is split. This implies, using the compatibility of with symmetric tensors, that the morphism
[TABLE]
is split as well. In particular,
[TABLE]
is non-zero. This finishes the proof. ∎
We want to explain why our proof differs from the one in [4, Theorem 5.3.1]. In fact the proof there is wrong and we give a counter example to their argument. However, using the above argument with symmetric tensors instead of symmetric powers, the proof in [4, Theorem 5.3.1.] can be fixed. The basic problem is that in positive characteristic the dual of a symmetric power is not canonically isomorphic to the -th symmetric power of the dual but to the -th module of symmetric tensors of . As was remarked before Theorem 3.11 the module of symmetric tensors is canonically isomorphic to the -th divided power of . We will rather speak about divided powers in the following.
To give the counter example we assume that is some field of characteristic . Let and take as the standard representation of with standard basis . Note that is a selfdual representation using the paring
[TABLE]
Let be the second divided power of . Then there is a short (non-split) exact sequence
[TABLE]
where is spanned by the element and where denotes the Frobenius twist of . Note that acts trivially on . We claim that there does not exist some such that the morphism is split. In fact, we prove that for every the restriction morphism
[TABLE]
from -invariants in to is zero. Using selfduality of and the fact that symmetric and divided powers are dual to each other it suffices to show that the morphism
[TABLE]
is zero where denotes the quotient (concretely
[TABLE]
Set . A general matrix
[TABLE]
maps these elements to
[TABLE]
It suffices to show that there does not exist a -invariant element of the form
[TABLE]
with only involving divided power monomials with . Namely, elements like span the kernel of the restriction map to . Let be the standard torus. The element
[TABLE]
maps , and . Therefore we can conclude that every -invariant element must be of the form
[TABLE]
i.e. in each divided power monomial occuring in the divided powers of match the divided powers of . Now apply the element
[TABLE]
to . This yields
[TABLE]
by -invariance of . Now we collect the coefficient of using the rules for calculating with divided powers. Namely, this coefficient is given by
[TABLE]
where . By -equivariance of we know that . We claim moreover that
[TABLE]
which implies , i.e., our claim about . It suffices to see that if the number
[TABLE]
is divisible by if as we are over a field of characteristic . But
[TABLE]
and is divisible by if by Pascal’s identity
[TABLE]
4. Applications
Theorem 3.11 has applications to local class field theory and to the étale and flat cohomology of with finite coefficients (cf. [7, Section 3]). In this section we present these applications to handle, similarly to [7] for , the case . However, the methods for are very similar to the one for and we present the statements uniformly for both cases.
We keep the notations from the last section. Moreover, we denote by a separable closure of . First we recall the calculation of the étale fundamental group of .
Theorem 4.1**.**
For every geometric point of the canonical morphism
[TABLE]
is an isomorphism.
Proof.
This is proven in [8, Théorème 8.6.1.] for , but the same proof applies to using Theorem 3.5. ∎
Let be the Brauer group of . Because is noetherian of Krull dimension [10, Corollaire 2.2.] implies
[TABLE]
Theorem 4.2**.**
We have
[TABLE]
Proof.
By definition, every element in is in the image of
[TABLE]
for some . By Theorem 3.11 there is a commutative diagram
[TABLE]
Moreover, using Proposition 5.7 the top horizontal arrow is trivially surjective. In particular, every element in maps to [math] in . Thus,
[TABLE]
∎
Let
[TABLE]
be the canonical morphism. As for every finite extension of we have
[TABLE]
Using the Leray spectral sequence for and Theorem 4.2 one obtains an exact sequence
[TABLE]
But
[TABLE]
and
[TABLE]
with trivial -action. In particular, we obtain the computation of the Brauer group of
[TABLE]
in a rather complicated way. One should note that in order to deduce this theorem we implicitly used Steinbergs theorem that
[TABLE]
for a (connected) reductive group over the maximal unramified extension of to deduce the concrete description of (cf. Proposition 5.7) and therefore the surjectivity of
[TABLE]
Theorem 4.3**.**
Let be a discrete torsion module for with for some prime to the characteristic of . Then for the canonical morphism
[TABLE]
is an isomorphism.
Proof.
Let be the canonical morphism. It suffices to prove
[TABLE]
for and prime to . Because and are coprime the Kummer sequence
[TABLE]
is an exact sequence of étale sheaves. But
[TABLE]
[TABLE]
and
[TABLE]
(cf. Theorem 4.2). This implies the claim. ∎
Now we discuss the case with -torsion coefficients using flat cohomology.
Theorem 4.4**.**
Let be a finite multiplicative group scheme over . Then the canonical morphism
[TABLE]
is an isomorphism for . Moreover, if and , then
[TABLE]
for . In particular,
[TABLE]
for .
Proof.
Let again be the canonical morphism. As in Theorem 4.3 it suffices to prove for every prime
[TABLE]
resp.
[TABLE]
for resp. all . Let be an algebraic closure of . It suffices to prove
[TABLE]
for resp.
[TABLE]
for . For coefficients the statement follows as in Theorem 4.3 using that for the Kummer sequence is exact for the flat topology. The second statement follows using the Artin-Schreier sequence
[TABLE]
together with the facts that
[TABLE]
for and . ∎
In particular, we obtain that
[TABLE]
is canonically isomorphic to the -torsion in the Brauer group of .
Finally, we record the calculation of for an arbitrary torus over (cf. [7, Théoréme 2.7.]).
Theorem 4.5**.**
Let be a torus over . Then
[TABLE]
Proof.
By [9, Théoréme (11.7)] étale and flat cohomology for tori agree because tori are smooth. Let be a finite extension of splitting . Then we obtain an exact sequence
[TABLE]
of tori, where denotes the Weil restriction of to . By Theorem 3.11 and Theorem 4.2 we obtain a commutative diagram
[TABLE]
where the lower horizontal arrow is surjective by Steinberg’s theorem and Proposition 5.7. This implies that
[TABLE]
as desired. ∎
Theorem 4.5 is related to Tate-Nakayama duality for tori (cf. [17]). The proétale covering yields a spectral sequence
[TABLE]
But
[TABLE]
and
[TABLE]
which implies by Theorem 4.5 that there is an exact sequence
[TABLE]
where
[TABLE]
with (cf. [14]).
5. Classification of reductive group schemes
We continue with the notations from the previous two sections. We now want to classify reductive group schemes over the Fargues-Fontaine curve. First we recall the definition of an affine group scheme over (cf. [4, Definition 9.1.8.]) as they provide the key examples of group schemes over .
Definition 5.1**.**
An affine group scheme over is a Hopf algebra object in the category of Ind-objects of .
In other words, an affine group scheme over consists of an affine group scheme over and an isomorphism
[TABLE]
such that the Hopf algebra of over is the increasing union of -stable subspaces. Therefore we recover [4, Definition 9.1.8.]. An affine group scheme over is called reductive if the corresponding group scheme over is reductive.
If is an affine group scheme over , then gives natural rise to an affine group scheme over by considering the base change of to with its canonical isomorphism . Equivalently, the Hopf algebra underlying is given by with Frobenius acting on .
Let be an affine group scheme over . Applying the tensor functor (cf. Theorem 3.5)
[TABLE]
to the Hopfalgebra yields a Hopf algebra
[TABLE]
over . Taking the relative Spec of this Hopf algebra
[TABLE]
defines a flat group scheme over . We call the group scheme (over ) associated with and write
[TABLE]
if we want to make this more precise. If is reductive, then also is reductive (over ) as this can be tested fiberwise and then over a large field extension of .
We record the following general lemma.
Lemma 5.2**.**
Let be a field, let be an affine group scheme over , let be its adjoint quotient and let be a -scheme. Let be a -torsor over with corresponding inner form of over . Let
[TABLE]
be the fiber functor of over associated with (cf. Lemma 3.1). Then
[TABLE]
where is considered as a Hopf algebra in the category of Ind-objects in via the adjoint action of and where denotes the -Hopf algebra of .
Proof.
The fiber functor of is given by
[TABLE]
Moreover, the inner form of is by definition given by the group scheme
[TABLE]
over . Equivalently, this twisting can be done on the Hopf algebra of . This shows
[TABLE]
and the proof is finished. ∎
We can now prove the following classification of reductive group schemes over the Fargues-Fontaine curve.
Theorem 5.3**.**
Let be a reductive group scheme over the Fargues-Fontaine curve . Then there exists a reductive group scheme over , unique up to isomorphism, such that is isomorphic to the group scheme associated with . There exists moreover a quasi-split group over such that is an inner form (over ) of .
Proof.
Let be the split reductive group over such that is a form of . Consider the canonical exact sequence
[TABLE]
of group schemes over . It gives rise to an exact sequence of pointed sets
[TABLE]
But
[TABLE]
because is constant and
[TABLE]
by Theorem 4.1. The choice of a pinning of defines a splitting of and the image of
[TABLE]
under this splitting consists precisely of the classes of constant reductive group schemes
[TABLE]
for a quasi-split form of over . Moreover, the elements in a fiber of are all inner forms of each other. In particular, we can see that is an inner form (over ) of a quasi-split form (over ) of .
Let be the -torsor over and let
[TABLE]
be the corresponding fiber functor. By Lemma 5.2 we obtain
[TABLE]
where is considered as a -representation via the adjoint action. By Theorem 3.11 we know that the fiber functor is admissible and we can apply theorem Theorem 3.8. This yields an exact tensor functor
[TABLE]
such that
[TABLE]
In particular, we see that is isomorphic to the group scheme associated with the group scheme given by the Hopf algebra
[TABLE]
over . Moreover, as is reductive, is reductive. As
[TABLE]
we see that is determined, up to isomorphism, by . ∎
In short, the functor from reductive group schemes over to reductive group schemes over is faithful and induces a bijection on isomorphism classes. But it is not an equivalence, for and the global sections depend on the chosen algebraically closed perfectoid field .
We can moreover obtain the following result.
Lemma 5.4**.**
Let be an algebraic extension of such that has cohomological dimension and, if , contains the perfection of . Let be a reductive group scheme over . Then there exists a finite subextension such that the base change is a pure inner form of a quasi-split group over .
Proof.
By Theorem 5.3 we already know that is an inner form of for a quasi-split group over . There exists an exact sequence (in the flat topology)
[TABLE]
yielding an exact sequence
[TABLE]
Thus it suffices to show that every element in maps to zero under a finite extension contained in . Let be a maximal torus and consider the exact sequence
[TABLE]
Using the associated long exact sequence and Theorem 4.5 it suffices to prove that
[TABLE]
As has cohomological dimension (and contains the perfection of if ) the degrees of finite subextensions get divisible by arbitrary large integers . We can conclude
[TABLE]
and
[TABLE]
As is surjective the morphism
[TABLE]
is surjective. This finishes the proof of the lemma. ∎
For example, looking at the proof shows that one can take also to be the maximal unramified extension of . But one can also take as the composition of a totally ramified extension and the perfection of , if .
We now start to classify torsors under reductive group schemes over .
Definition 5.5**.**
Let be a group scheme over . We define as the category of representations of on isocrystals, i.e., as the category of finite dimensional comodules of the Hopf algebra in the category .
In other words, an object consists of an isocrystal and a coaction which is moreover a morphism of Ind-isocrystals. Clearly, the category is Tannakian over .444We mention the following possible cause of confusion. If is a reductive group over with associated group scheme over , then the category of finite dimensional -representations of is the full subcategory of given by representations of whose underlying isocrystal is semistable of slope [math]. We denote by
[TABLE]
the canonical exact tensor functor sending a -representation to its underlying isocrystal. We define
[TABLE]
and call it the canonical fiber functor of over .
Lemma 5.6**.**
Let be a group scheme over . Let
[TABLE]
be a fiber functor of over and let
[TABLE]
be the corresponding group scheme over . Let
[TABLE]
be the Hopf algebra underlying considered as a representation of via the adjoint action. Then there is a natural isomorphism
[TABLE]
of Hopf algebras. In other words, with the adjoint action by is the fundamental group of .
Proof.
Let be a quasi-coherent -algebra and let
[TABLE]
be a morphism of -algebras. Then for the composition
[TABLE]
where the first morphism is induced by the comultiplication of , which is a morphism of -representations if is equipped with the adjoint action, and the second by and multiplication in , is natural in and a naturally a tensor automorphism, hence defines a -valued point of . Conversely, if
[TABLE]
is a natural tensor automorphism, then evaluating on the Ind-object of defines a morphism of -algebras
[TABLE]
where the first morphism is induced by the unit of and the last by the counit of , hence an -valued point of . One checks that both maps are inverse homomorphisms of groups. Hence, the claim follows. ∎
Applying this lemma to
[TABLE]
shows that
[TABLE]
is isomorphic to the group scheme associated with the group scheme over . A similar proof shows moreover, that
[TABLE]
where is the functor sending an algebra object in to the group of tensor automorphisms of . In particular, we can conclude that the band of the gerbe of the Tannakian category is reductive if is.
We can now give a concrete classification of the category (cf. Definition 3.9) of a connected group scheme over (cf. [4, Lemma 9.1.4]). We let be the following category: Its objects are elements (more precisely, if is the group scheme over underlying ) and the set of morphisms from to are the elements such that
[TABLE]
where
[TABLE]
is induced by the given isomorphism of lying over the Frobenius of .
Given an element we can define an exact tensor functor
[TABLE]
by sending a -representation to the isocrystal
[TABLE]
where the second morphism denotes the action of on .
Proposition 5.7**.**
Let be a connected affine group scheme over . Sending to defines an equivalence of categories
[TABLE]
Proof.
Let resp.
[TABLE]
be the canonical exact tensor functors. If denotes the underlying connected group scheme of over then by Lemma 5.6
[TABLE]
Let
[TABLE]
be an exact tensor functor. By Steinberg’s theorem the functors and are isomorphic. Let
[TABLE]
be an -linear tensor isomorphism. Moreover, both functors come equipped with canonical -semilinear tensor automorphisms . Then
[TABLE]
is an -linear tensor automorphism of . Therefore, by the Tannakian formalism, there exists an element such that is given by the multiplication by . But then is isomorphic to . A similar reasoning shows that the functor is fully faithful. ∎
Let be a connected affine group scheme and let be the connected affine group scheme over . Using [4, Lemma 9.1.4] (i.e., the same reasoning as in the proof above) we can conclude that there is a canonical chain of equivalences
[TABLE]
where is defined exactly as .
We will need the following lemma. If is a Tannakian category over a field and a finite extension we denote by the base change of to (cf. [18, Construction 2.12]).
Lemma 5.8**.**
Let be a composition of a finite totally ramified and a finite purely inseparable extension and let be an affine group scheme over . Then the base change of to is equivalent to the category where is the base change of to (i.e., the Hopf algebra of is given by the Hopf algebra in where is the completion of the maximal unramified extension of ).
Proof.
An object of is, by definition, given by a triple
[TABLE]
where is an isocrystal over , an -algebra homomorphism and a coaction (in ) of on . Moreover, is -linear as maps into the -endomorphisms of . We can map this triple to the -representation where is considered as an isocrystal over with acting via on and being the given -semilinear automorphism (which is also -linear). Here we have written resp. for the Frobenius of resp. . Finally, is defined as the composition
[TABLE]
which is a morphism of isocrystals over . Conversely, a -representation
[TABLE]
can be sent to the triple
[TABLE]
where is considered as an isocrystal over , is the given action of on and the given coaction (using ). These two functors define inverse equivalences of categories. ∎
The same proof shows that for the Tannakian category of representations of an affine group scheme over a field and a finite extension of the base change is equivalent to the Tannakian category .
Theorem 5.9**.**
Let be a reductive group scheme over and let be an affine group scheme over such that is associated to , i.e.,
[TABLE]
Then the canonical morphism
[TABLE]
is bijective.
Proof.
Let be a quasi-split group over such that is an inner form of (cf. Theorem 5.3). We first assume that is a pure inner form of . Then there exists a -torsor over such that
[TABLE]
By Theorem 3.11, Proposition 5.7 and Lemma 5.6 we can see that there exists an element such that
[TABLE]
with given by the composition
[TABLE]
where
[TABLE]
denotes the Frobenius on and be adjoint action of on . Let be the group scheme over associated with . One can deduce that
[TABLE]
by mapping a representation of to the -representation (cf. [4, Example 9.1.22]). In particular, we can conclude that every fiber functor
[TABLE]
for is admissible because this holds true for by Theorem 3.11 (and the fact these fiber functors identify with fiber functors for , cf. Proposition 5.7). By Theorem 3.10 we can conclude in this case.
Now assume that is an arbitrary inner form of . Let be a composition of a finite totally ramified extension of and a finite purely inseparable extension of such that is a pure inner form of (the existence of such a field extension is guaranteed by Lemma 5.4 and the discussion following it). We already know (from Theorem 3.10) that the map
[TABLE]
is injective and, again by Theorem 3.10, it suffices to check that every fiber functor
[TABLE]
is admissible. Let be such a fiber functor. By [18, Construction 2.12] the composition
[TABLE]
factors over a fiber functor
[TABLE]
where denotes the base change of the category to . By Lemma 5.8 the categories and are equivalent. As is a pure inner form of (and associated with ) the fiber functor is admissible by the case already proven. By Lemma 3.4 this implies that is admissible. In particular, this finishes the proof of the classification of -torsors. ∎
6. Uniformization results
In this section we establish uniformization results for -torsors over the Fargues-Fontaine curve.
First, we want to prove that reductive group schemes over the Fargues-Fontaine curve become constant after removing a closed point.
We will need the following lemma.
Lemma 6.1**.**
Let be a quasi-split reductive group over . Then every element in admits a reduction to some torus .
Proof.
For this is [7, Proposition 7.2.]. Using [13, Proposition 13.1] we can extend this argument once we establish that in the case every element in admits a reduction to some basic element in for some Levi subgroup . Let and consider the filtered fiber functor
[TABLE]
It defines a parabolic subgroup scheme
[TABLE]
If is a Borel subgroup, then there exists a unique standard parabolic
[TABLE]
such that for every the groups and are conjugated. Using [1, Exposé XXVI, Proposition 1.3.] we can conclude that there exists a -torsor over such that
[TABLE]
with acting on via conjugation (in [7, Section 5.1.] this -torsor would be called the canonical reduction of the torsor associated with ). As in [7, Proposition 5.16.] this -torsor admits a reduction to a Levi subgroup because for the vector bundles
[TABLE]
(cf. Definition 2.5) are semistable of slope . By construction this -torsor is semistable which implies that reduces to a basic element in (cf. [7, Proposition 5.12.]). ∎
As in [7] this implies the following theorem.
Theorem 6.2**.**
Let be a quasi-split reductive group over and let be a closed point. Then every -torsor over is trivial over .
Proof.
Using Lemma 6.1 and Theorem 4.5 the same proof as in [7, Théorème 7.1.] works. ∎
Lemma 6.3**.**
Let be a reductive group scheme over and let be a closed point. Then is isomorphic to a constant quasi-split reductive group.
Proof.
By Theorem 5.3 there exists a quasi-split group over such that is an inner form of . Let be a -torsor such that
[TABLE]
Then is again quasi-split as the image of a Borel subgroup is a Borel subgroup of . By Theorem 6.2 every -torseur over is trivial over the punctured curve . In particular, is trivial which shows that
[TABLE]
is isomorphic to a constant quasi-split reductive group. ∎
In fact, reductive group schemes over are not too far from being constant. Let be a reductive group scheme over and write (cf. Theorem 5.3)
[TABLE]
for a quasi-split reductive group over and a -torsor over . Let be an element giving rise to (cf. Proposition 5.7). Assume first that is basic (cf. [14, 5.1.]). We claim that in this case is already constant. Namely, the group is semisimple and it suffices to prove the following lemma showing that is isomorphic to the pullback of a -torsor over in this case. We denote by the set of basic elements in if is a reductive group over (cf. [14, 5.1.]).
Lemma 6.4**.**
Let be a semisimple group. Then the canonical map
[TABLE]
(cf. [14, 1.8.]) is bijective.
Proof.
By [14, 4.5.] an element lies in the image of the canonical injection
[TABLE]
if and only if the associated morphism is trivial where is the constant pro-torus with character group (cf. [14, 4.2.] for the construction of ). Moreover, by definition is basic if and only if factors through the center of ([14, 5.1.]). But if is semisimple the center of is finite which implies that every central homomorphism must be trivial as is connected. ∎
Now assume that is arbitrary. By [14, Proposition 6.2.] resp. the proof of Lemma 6.1 the element is -conjugate to some for some Levi subgroup such that is basic. In particular,
[TABLE]
if denotes the -torsor corresponding to (cf. Proposition 5.7). Let be the preimage of under the canonical map . As is connected the image of in must be contained in the adjoint group of . Let be the push forward of the -torsor to . The element corresponding to will again be basic. Namely, if corresponds to the -torseur over , then
[TABLE]
because the homomorphism factors through the center of and
[TABLE]
By Lemma 6.4 we can conclude that contains the constant reductive group
[TABLE]
which is of the same rank.
In general reductive group schemes can be non-constant. For example, let be a non-semistable vector bundle on . Then the reductive group scheme
[TABLE]
is non-constant.
Finally we record the following “uniformization result” generalizing [7, Théoème 7.1.] (resp. Theorem 6.2).
Theorem 6.5**.**
Let be a reductive group scheme over , let be a closed point and let be a -torsor. Then is isomorphic to the trivial -torsor.
Proof.
By Lemma 6.3 the group scheme
[TABLE]
is isomorphic to a constant quasi-split reductive group scheme. Let be the -torseur corresponding to under such an isomorphism. Then admits an extension to a -torsor over . Namely, by Beauville-Laszlo glueing (cf. [2] and Lemma 3.1) it suffices to construct a -torsor over together with an isomorphism to over . But abstractly
[TABLE]
is isomorphic to a power series ring by Cohen’s structure theorem. In particular,
[TABLE]
is of cohomological dimension as is algebraically closed in our case. Steinberg’s theorem finally implies that every -torsor over is trivial and, in particular, admits an extension to . Thus, let be a -torsor extending . By Theorem 6.2 we can conclude that
[TABLE]
is isomorphic to the trivial torsor because is quasi-split. ∎
In fact, in Theorem 6.5 we have shown
[TABLE]
for every reductive group over .
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1[1] M. Artin, J. E. Bertin, M. Demazure, P. Gabriel, A. Grothendieck, M. Raynaud, and J.-P. Serre. Schémas en groupes. Fasc. 7: Exposés 23 à 26 , volume 1963/64 of Séminaire de Géométrie Algébrique de l’Institut des Hautes Études Scientifiques . Institut des Hautes Études Scientifiques, Paris, 1965/1966.
- 2[2] Arnaud Beauville and Yves Laszlo. Un lemme de descente. C. R. Acad. Sci. Paris Sér. I Math. , 320(3):335–340, 1995.
- 3[3] Christophe Cornut. Filtrations and buildings. preprint on webpage at https://webusers.imj-prg.fr/~christophe.cornut/papers/Fil Buiv 3.1.pdf .
- 4[4] Jean-François Dat, Sascha Orlik, and Michael Rapoport. Period domains over finite and p 𝑝 p -adic fields , volume 183 of Cambridge Tracts in Mathematics . Cambridge University Press, Cambridge, 2010.
- 5[5] P. Deligne. Catégories tannakiennes. In The Grothendieck Festschrift, Vol. II , volume 87 of Progr. Math. , pages 111–195. Birkhäuser Boston, Boston, MA, 1990.
- 6[6] L. Fargues. Geometrization of the local Langlands correspondence: an overview. Ar Xiv e-prints , February 2016.
- 7[7] Laurent Fargues. G-torseurs en théorie de hodge p-adique. preprint on webpage at https://webusers.imj-prg.fr/~laurent.fargues/Prepublications.html .
- 8[8] Laurent Fargues and Jean-Marc Fontaine. Courbes et fibrés vectoriels en théorie de hodge p-adique. preprint on webpage at https://webusers.imj-prg.fr/~laurent.fargues/Prepublications.html .
