11 Classification g Bundles

The Classification of G-bundles
Notes by Tony Feng

for a talk by Michael Rapoport
April 7, 2016
We now want to generalize the preceding classification of vector bundles to G-bundles

(recovering the old results when G = GLn ).
1 Background
1.1 Notation
We fix the notation for this talk: let
• E be a local field (of characteristic 0 or p),
• $E the uniformizer,
• Fq the residue field,
• Ĕ the completion of the maximal unramified extension, and
• F an algebraically closed perfectoid field of characteristic p.
1.2 Classical G-bundles

Definition 1.1. Let G be a connected linear algebraic group over E. A connected G-bundle
on X can be defined in either of the following two ways:
1. (“internal”) A principal homogeneous space T under G on X which is locally trivial
for the (étale or fppf) topology.
2. (“external” ) An exact faithful E-linear ⊗−functor RepE G → VectX .
Example 1.2. Why are the two definitions equivalent? We sketch one direction. Given a
G-torsor T , we can define the functor
VT ((V, ρ)) = T ×G,ρ V.
Definition 1.3. We denote by | BunG | the set of isomorphism classes of connected G-bundles
on X.
Example 1.4. If G = GLn then | BunG | = VectX,n .
1
1.3 The classification of VectX
We have a functor
E : ϕ − ModĔ → VectX
sending
ϕ=$dE
B+E,F ⊗Ĕ V
M
(V, ϕ) 7→ .
d≥0
Theorem 1.5. This E is a faithful exact E-linear ⊗-functor, which is essentially surjective
(but not fully faithful, see Warning 1.6).
It also induces an equivalence of categories
(isoclinic ϕ-isocrystals) ↔ (semi-stable vector bundles)
and a bijection of objects

|ϕ − ModĔ | = | VectX |.
Warning 1.6. The functor is not fully faithful because End(Triv ⊕ Triv(1))
! is E ⊕ E in
E BC
the category of isocrystals but a “Banach-Colmez-like object” in the category of
E
vector bundles.
This theorem is what we want to generalize, from vector bundles to G-bundles.
2 G-isocrystals (following Kottwitz)

2.1 The definition
Definition 2.1. Let G be a connected linear algebraic group over E. A G-isocrystal can be
defined in either of the following two ways:
1. (external) An exact faithful E-linear ⊗-functor
N : RepE G → ϕ − ModĔ .
2. (internal) An element b ∈ G(Ĕ). These form a category via
Hom(b, b0 ) = {g ∈ G(Ĕ) | gbσ(g)−1 = b0 }.
We denote by B(G) the set of G-isocrystals up to isomorphism.

Example 2.2. Why are the internal and external versions equivalent? Given b ∈ G(Ĕ), we
can associate the functor Nb defined by
Nb (V, ρ) = (V ⊗E Ĕ, ρ(b) ◦ (Id ⊗σ))
Example 2.3. For G = GLn , the classical isocrystal description of an element b ∈ G(Ĕ) is
(Ĕ ⊕n , b ◦ σ).
2
2.2 The Newton and Kottwitz invariants
Let G be reductive. We construct two invariants associated to G-bundles.
The Newton Invariant. Let b ∈ G(Ĕ). Then we can associate a homomorphism
νb : DĔ → G Ĕ
where D is the split torus over E with X ∗ (D) = Q. This homomorphism νb is characterized
by the property that for all (V, ρ), the morphism
ρ ◦ νb : DĔ → GL(VĔ )
has induced Q-grading on VĔ equal to the slope filtration of (VĔ , bσ).
The cocharacter group X∗ (G) has an action of G, and we set
X∗ (G)Q /G = HomĔ (DĔ , G Ĕ )/G(Ĕ).
There is an action of σ on X∗ (G)Q , and one can show that νb ∈ (X∗ (G)Q /G)σ only depends
on [b], thus inducing a well-defined map
ν : B(G) → (X∗ (G)Q /G)σ . (1)
This is the Newton invariant.

Example 2.4. If G is quasi-split, say with Borel B, maximal torus T ⊂ B, and maximal split
torus A ⊂ T ⊂ B then the right side of (1) can be identified with X∗ (A)+Q .
Remark 2.5. There is also an internal definition of the Newton invariant. Given b, there
exists b0 with b ∼ b0 such that s 0 such that
(b0 σ) s = s · νb0 ($E ) · σ s
with the equality taking place in G(Ĕ) o hσi. This characterizes ν[b] = ν[b0 ] (since ν is
supposed to be defined on isomorphism classes).
The Kottwitz invariant. Consider
π1 (G) = X∗ (T )/X∗ (T sc ).
This is canonically and functorially associated to G, and admits an action of Γ. The Kottwitz
invariant is described in terms of this fundamental group, as a map
κ : B(G) → π1 (G)Γ .
This is not so easy to define, but we will try to give some feeling for it. Roughly B(G) is
similar to π0 (LG) (but not quite on the nose) and π0 (LG) = π1 (G)Γ .
Theorem 2.6. The map B(G) → (X∗ (G)Q /G)σ × π1 (G)Γ is injective.
3
The description of the image is not easy in general, but in the quasi-split case it is fairly
easy to describe it.
Example 2.7. Let G = GLn . Then X∗ (A)+Q = (Qn )+ and π1 (G)Γ = Z. In this case the first
component of the map gives the slopes of the Newton polygon, and the second component
gives the endpoint of the Newton polygon. So in this case the 1st component determines
the second, since the endpoint can be determined from the slopes via the formula
X
(λi ) ∈ (Qn )+ 7→ λi .
Therefore, the image can be characterized as the tuples whose break points are integers.
Example 2.8. Let G = T . Then X(A)+Q = X∗ (T )Γ ⊗ Q. (There are no positivity conditions
because there are no roots.) The second component is π1 (T )Γ = X∗ (T )Γ . In this case the
second component determines the first, via
γ ∈ X∗ (T )Γ → X∗ (T )Γ ⊗ Q.
(And the first component determines the second up to torsion.)
2.3 More structure to B(G)

First, there is an analogue of the semistable/isoclinic set.
Definition 2.9. Let B(G)basic = {[b] | νb = central homomorphism}.
Example 2.10. For GLn , this means isoclinic.
Inside B(G)basic there is the subset B(G)0basic = {[b] | νb = trivial}. This is the analogue
of the unit root isocrystals.
These form a section to the Kottwitz invariant. In other words, κ induces bijections
B(G)basic → π1 (G)Γ
and
B(G)0basic → π1 (G)Γ,tors H 1 (E, G).
In this sense B(G) is a generalization of Galois cohomology.
2.4 The automorphism group

Another piece of structure is the automorphism group. For b ∈ G(Ĕ), we can associate a a
group
Jb (R) := {g ∈ G(Ĕ ⊗ R) | gbσ(g)−1 = b}.
Then Jb (E) = Aut(b). This turns out to always be a reductive group over E.
Remark 2.11. The Jb (E) are Levis if G is quasiplit.
4
Some facts.
• An element b ∈ G(Ĕ) is basic if and only if Jb is an inner form of G.
• If Z(G) is connected then every inner form comes from some basic b.
• If G is quasisplit, then B(G) can be described in terms B(M)basic for standard Levi
subgroups M ⊂ G.
• B(G) is a partially ordered set and its basic elements are the minimal ones.
3 G-bundles on the Fargues-Fontaine Curve

3.1 Semistable G-bundles
We want to define a functor
EG : G − isocrystals → BunG .
There are again two definitions.
1. (external) Given a G-isocrystal RepE → ϕ − ModĔ in the external sense, composing

with E gives
E
RepE → ϕ − ModĔ −
→ VectX .
This is a G-bundle in the external sense.
2. (internal) Given b ∈ G(Ĕ), form G Ĕ ×Ĕ YE /ϕZ with ϕ acting diagonally by ϕ on YE

and by g 7→ bσ(g) on G Ĕ .
Theorem 3.1. Assume that ch E = 0. Then this functor EG is faithful and induces a bijec-
tion
B(G) → | BunG |.
Furthermore, EG induces an equivalence of categories between B(G)basic and the category
of semi-stable G-bundles.
Definition 3.2. A G-bundle T is semi-stable if
1. (half-external) T (Lie G, Ad) is a semi-stable vector bundle.
2. (external) T (V, ρ) is a semi-stable vector bundle if ρ is homogeneous. (Remark:

we are using here that tensor of semistable is semistable, which follows from the
classification of vector bundles.)
5
3. (internal) Let P ⊂ G be a power-bounded subgroup. Let AP be the split part of the
center of P. We have dually A0P the split part of the cocenter of P. Then the map
AP → A0P
is an isogeny, identifying the rational cocharacter groups. Let T be a G-bundle and
suppose TP is a P-structure on T . Then we define the slope cocharacter µ(TP ) ∈
X∗ (AP )Q which is characterized by the property
hµ(TP ), λi = deg λ∗ (TP ) for all λ ∈ X ∗ (A0P ).
Finally, we define T to be semi-stable if and only if
hµ(TP ), αi ≤ 0∀α ∈ Lie NP .
3.2 The two invariants

How are the two invariants expressed in terms of the corresponding G-bundles?
Newton invariant. First assume that G is quasi-split, with A ⊂ T ⊂ B as before. Let T be

a G-bundle. The Harder-Narasimhan reduction theorem says that there exists a unique pair
(P, TP ) with P a standard parabolic subgroup and TP a P-bundle such that
1. TP ×P MP is a semistable MP -bundle, and
2. µ(TP ) ∈ X∗ (AP )++
Q .
Now the maximal split subtorus AP ⊂ A gives a map from X∗ (AP )++ +
Q → X∗ (A)Q , sending
+
µ(TP ) 7→ νT ∈ X∗ (A)Q .
Proposition 3.3. We have [νb ] = −νT (b) .
Why the minus sign? It came up already in Dospinescu’s talk: a minus sign was taken
to get compatibility of endomorphisms.
Kottwitz invariant. We know that

| BunG | = Hét1 (X, G).
Fargues defines a G-equivariant Chern class
1 : Hét (X, G) → π1 (G)Γ .

1
cG
Proposition 3.4. We have
κ(b) = cG
1 (EG (b)).
3.3 What’s wrong in characteristic p?

In the book of Fargues and Fontaine, they construct various categories of ϕ-isocrystals
which give vector bundles. One of these functors is not exact. When you want to apply
the external definition of G-bundles you need an exact functor; this uses the fact that in
characteristic 0 the representation theory is semisimple.

11 Classification g Bundles

Uploaded by

Copyright:

Available Formats

11 Classification g Bundles

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

11 Classification g Bundles

Uploaded by

Copyright:

Available Formats

The Classification of G-bundles

Notes by Tony Feng

We now want to generalize the preceding classification of vector bundles to G-bundles

1.2 Classical G-bundles

(isoclinic ϕ-isocrystals) ↔ (semi-stable vector bundles)

and a bijection of objects

2 G-isocrystals (following Kottwitz)

2. (internal) An element b ∈ G(Ĕ). These form a category via

Hom(b, b0 ) = {g ∈ G(Ĕ) | gbσ(g)−1 = b0 }.

We denote by B(G) the set of G-isocrystals up to isomorphism.

Nb (V, ρ) = (V ⊗E Ĕ, ρ(b) ◦ (Id ⊗σ))

The Newton Invariant. Let b ∈ G(Ĕ). Then we can associate a homomorphism

X∗ (G)Q /G = HomĔ (DĔ , G Ĕ )/G(Ĕ).

ν : B(G) → (X∗ (G)Q /G)σ . (1)

This is the Newton invariant.

(b0 σ) s = s · νb0 ($E ) · σ s

(And the first component determines the second up to torsion.)

2.3 More structure to B(G)

2.4 The automorphism group

• An element b ∈ G(Ĕ) is basic if and only if Jb is an inner form of G.

3 G-bundles on the Fargues-Fontaine Curve

There are again two definitions.

1. (external) Given a G-isocrystal RepE → ϕ − ModĔ in the external sense, composing

2. (internal) Given b ∈ G(Ĕ), form G Ĕ ×Ĕ YE /ϕZ with ϕ acting diagonally by ϕ on YE

Definition 3.2. A G-bundle T is semi-stable if

1. (half-external) T (Lie G, Ad) is a semi-stable vector bundle.

2. (external) T (V, ρ) is a semi-stable vector bundle if ρ is homogeneous. (Remark:

3.2 The two invariants

Newton invariant. First assume that G is quasi-split, with A ⊂ T ⊂ B as before. Let T be

Kottwitz invariant. We know that

1 : Hét (X, G) → π1 (G)Γ .

3.3 What’s wrong in characteristic p?

You might also like