Helene Sigloch
Homotopy Theory for Rigid Analytic Varieties
Dissertation zur Erlangung des Doktorgrades der Fakultat¨ fur¨ Mathematik und Physik der Albert-Ludwigs-Universitat¨ Freiburg im Breisgau
Marz¨ 2016 Dekan der Fakultat¨ fur¨ Mathematik und Physik: Prof. Dr. Dietmar Kr¨oner
Erster Referent: Dr. Matthias Wendt
Zweiter Referent: Prof. Dr. Joseph Ayoub
Datum der Promotion: 25. Mai 2016 Contents
Introduction 3
1 Rigid Analytic Varieties 11 1.1 Definitions ...... 12 1.2 Formal models and reduction ...... 20 1.3 Flatness, smoothness and ´etaleness ...... 23
2 Vector Bundles over Rigid Analytic Varieties 27 2.1 A word on topologies ...... 28 2.2 Serre–Swan for rigid analytic quasi-Stein varieties ...... 30 2.3 Line bundles ...... 37 2.4 Divisors ...... 38
3 Homotopy Invariance of Vector Bundles 41 3.1 Serre’s problem and the Bass–Quillen conjecture ...... 41 3.2 Homotopy invariance ...... 42 3.3 Counterexamples ...... 44 3.4 Local Homotopy Invariance ...... 48 3.5 The case of line bundles ...... 56 3.5.1 B1-invariance of Pic ...... 57 1 3.5.2 Arig-invariance of Pic ...... 59
4 Homotopy Theory for Rigid Varieties 65 4.1 Model categories and homotopy theory ...... 65 4.2 Sites and completely decomposable structures ...... 70 4.3 Homotopy theories for rigid analytic varieties ...... 71 4.3.1 Relation to Ayoub’s theory ...... 75 4.3.2 A trip to the zoo ...... 76
5 Classification of Vector Bundles 79 5.1 Classification of vector bundles: The classical results ...... 79 5.2 H-principles and homotopy sheaves ...... 82 5.3 The h-principle in A1-homotopy theory ...... 86 5.4 Classifying spaces ...... 89 5.5 Homotopy invariance implies a classification of vector bundles . . . . 95 5.6 Discussion ...... 100
1 2 CONTENTS
Bibliography 103
Index 112 Introduction
The goal of this thesis is to develop tools to classify vector bundles over suitable rigid analytic varieties in a suitable homotopy theory. Our main result is: Theorem 1 (Theorem 5.41). Let X be a smooth rigid analytic variety over a complete, nonarchimedean, nontrivially valued field k of characteristic zero. Let X be quasi-Stein. Then there is a natural bijection
∞ ∼ [X, ] 1 −→ {analytic k-line bundles over X}/ ∼= P Arig between motivic homotopy classes from X into the infinite projective space P∞ over k and isomorphism classes of analytic k-line bundles over X. The assumption on the characteristic of k is superfluous if we use a theorem by Kerz–Saito–Tamme [KST16] which was attained after submission of this thesis.
Theorem 1 can be seen in the context of Morel’s A1-homotopy classification of algebraic vector bundles over a smooth affine variety and of Grauert’s homotopy classification of holomorphic vector bundles over a complex Stein space. The classical model for all of these theorems is Steenrod’s homotopy classification of continuous vector bundles over a paracompact Hausdorff space. Theorem 2 (F. Morel [Mor12], Asok–Hoyois–Wendt [AHW15a]). Let X be a smooth affine algebraic variety over a field K. There is a natural bijection ∼ ∼ [X, Grn/K ]A1 −→ {alg. K-vector bundles of rank n over X}/ =
1 between A -homotopy classes from X into the infinite Grassmannian Grn over K and isomorphism classes of algebraic K-vector bundles of rank n over X. If n = 1, the theorem holds more generally for any smooth scheme X over a regular base [MV99, §4, Proposition 3.8].
Strikingly, if K = C, this classification still holds after analytification. The analyti- fication of a smooth affine complex algebraic variety is a Stein manifold and the analytification of the complex affine line is C. Theorem 3 (Grauert [Gra57a, Gra57b]). Let X be a complex Stein space. There is a natural bijection ∼ ∼ [X, Grn/C] −→ {holomorphic vector bundles of rank n over X}/ = between homotopy classes of maps from X into the infinite Grassmannian Grn over C and isomorphism classes of holomorphic vector bundles of rank n over X.
3 4 CHAPTER 0. INTRODUCTION
Grauert’s theorem is stated in terms of classical (i. e., continuous) homotopy classes and with [0, 1] as an interval. L´arussonshowed that Grauert’s theorem can also be phrased in terms of motivic homotopy theory with C as an interval [L´ar03, L´ar04, L´ar05]. Morel’s theorem and Grauert’s theorem seem to be linked via analytification. But as Morel’s theorem is valid without restrictions on the base field, it seems natural to ask if the same holds for other analytic fields, in particular for a complete nonarchimedean valued field k such as a field Qp of p-adic numbers or a field κ((T )) of formal Laurent series over another field κ.
The starting point for this project was Ayoub’s B1-homotopy theory for rigid analytic varieties. It is a version of Morel–Voevodsky’s homotopy theory of a site with interval where the “unit ball” B1 := Sp khT i figures as an interval. There are several approaches to nonarchimedean analytic geometry. Motivated by Ayoub’s work, we chose to work with rigid analytic varieties, too. His approach was well suited to define and work with motives of rigid analytic varieties, cf. [Ayo15, Vez14a, Vez14b, Vez15].
1 However, it turns out that for our purposes the analytification Arig of the affine line is a better choice of interval. It also fits better with Morel’s theorem.
An example
1 1 Theorem 1 becomes false if one just replaces the interval object Arig by B : Let k = Qp, p ≥ 3 and 2 X = Sp QphT1,T2i/(T1 − T2(T2 − p)(T2 − 2p)) . 1 Consider a line bundle L on X ×B and its restrictions Lr, Ls to X ×{r}, respectively 1 to X × {s} for some r, s ∈ B . Viewing Lr and Ls as line bundles over X, they are 1 ∞ in the same B -homotopy class in [X, P ]B1 . If we are to get a natural bijection ∞ ∼ [X, P ]B1 −→ {analytic k-line bundles over X}/ = , then for each line bundle L on X × B1 and each pair r, s ∈ B1 as above we would need Lr and Ls to be isomorphic. Consequently, we would need every line bundle L on X × B1 to be isomorphic to the pullback of a line bundle over X along 1 pr1 : X × B −→ X. It is a result by Gerritzen that this is not the case. Gerritzen shows that if k is discretely valued then the Picard group of line bundles on a distinguished smooth affinoid variety X whose canonical model is regular and the Picard group of the canonical reduction X˜ are isomorphic [Ger77]. In our example, ˜ 2 3 X = Spec Fp[T1,T2]/(T1 − T2 ) . By a theorem by Bass and Murthy,
1 Pic(X˜) Pic(X˜ × A ) = Pic(X^× B1). Van der Put showed that the same problem occurs if k is algebraically closed instead of discretely valued [vdP82]. 5
The proof
It turns out that homotopy invariance with respect to the interval is really the key question for a homotopy classification of vector bundles over (suitable) rigid analytic 1 1 varieties. For an interval object I ∈ {B , Arig} we say that “vector bundles of rank n are I-invariant” on a subcategory R of the category of rigid analytic varieties if for each object X ∈ R, the projection X × I → X induces a bijection on isomorphism classes of vector bundles of rank n. Let us give an outline of the proof of Theorem 1. Let k be a complete, nonarchimedean, nontrivially valued field and X a smooth rigid analytic variety.
1 a) If char k = 0, then every line bundle over X × Arig has a local trivialisation 1 by subsets of the form {Ui × Arig}i∈I where {Ui}i∈J is an admissible covering of X. This is Corollary 3.35, a corollary of Theorem 3.32 by Tamme. After submission of this thesis, Tamme’s theorem was generalised to k of arbitrary characteristic by Kerz, Saito and Tamme [KST16].
1 b) Consider a line bundle on X × Arig and a local trivialisation of the form 1 1 1 {Ui × Arig}i∈I . Then every transition function from Ui × Arig to Uj × Arig on 1 Ui ∩ Uj × Arig is actually a transition function from Ui to Uj on Ui ∩ Uj, which 1 means that it is constant in the Arig-direction. This is Theorem 3.36. 1 c) Consequently, the projection X × Arig → X induces an isomorphism of groups ∼ 1 1 Pic(X) = Pic(X × Arig), i. e., line bundles are Arig-invariant on smooth rigid analytic varieties. This is Theorem 3.31. d) Every vector bundle over a quasi-Stein or a quasicompact rigid analytic variety admits a finite local trivialisation. These are Theorems 2.16 and 2.17. e) Let R be a subcategory of the category of smooth rigid analytic varieties. As- 1 1 sume that R has a strictly initial object ∅. Let I ∈ {B , Arig} be a representable interval object on R. Theorem 4 (Theorem 5.40). For some n ∈ N, assume that vector bundles of rank n are I-invariant on R. Assume that every vector bundle of rank n over an object X of R admits a finite local trivialisation. Then there is a natural bijection ∼ ∼ [X, Grn/k]I −→ {analytic k-vector bundles of rank n over X}/ = .
1 f) The assumptions of Theorem 4 are satisfied if n = 1, I = Arig, R is the subcategory of quasi-Stein spaces and char k = 0. Again, the assumption on the characteristic is only needed for Tamme’s theorem and can be removed using Kerz–Saito–Tamme [KST16]. Theorem 1 follows.
Outlook
We expect Theorem 1 to generalise to arbitrary characteristic and to vector bundles of higher rank: 6 CHAPTER 0. INTRODUCTION
Conjecture 1. Theorem 1 holds also if char k > 0. Conjecture 2. Let X be a smooth rigid analytic quasi-Stein variety over any com- plete, nonarchimedean, nontrivially valued field k. Then there is a natural bijection
∼ [X, Gr ] 1 −→ {analytic k-vector bundles of rank n over X}/ ∼= n/k Arig between motivic homotopy classes from X into the infinite Grassmannian Grn/k and isomorphism classes of analytic k-vector bundles of rank n over X. Conjecture 1 is now known to be true, after the original submission of this thesis. It follows with Kerz–Saito–Tamme’s generalisation of Tamme’s Theorem [KST16]. Towards a generalisation of Theorem 1 to vector bundles of higher rank, it will probably be necessary to work with a Grothendieck topology on R which is different from the G-topology on rigid analytic varieties, for example with the Nisnevich topology. In fact, we prove more than we stated above as Theorem 4. The theorem we prove allows us to work with a Grothendieck topology different from the G- topology. For a proof of Conjecture 2, it remains to show that vector bundles over 1 smooth quasi-Stein varieties are Arig-invariant. This means, if X is smooth and 1 quasi-Stein, then every vector bundle over X × Arig is isomorphic to the pullback 1 along X × Arig → X of some vector bundle over X. We hope to do this in future work. Furthermore, homotopy invariance should be the only obstruction against generalising the theorem further to nonarchimedean Lie groups other than GLn. In the algebro- geometric setting, Asok–Hoyois–Wendt’s proof also works for principal bundles over an isotropic, reductive, algebraic group over an infinite field [AHW15b, Theorem 4.1.3]. In the complex setting, Grauert’s Oka principle even holds for every complex Lie group [Gra57b]. If Conjecture 2 holds, it should generalise to principal bundles under other suitable nonarchimedean Lie groups. One of our main interests now is to develop geometric applications of Theorem 1, or, more generally, of nonarchimedean motivic homotopy theory. Both Morel’s theorem and Grauert’s theorem set the stage for other theorems that are interesting on their own. Using Morel’s theorem and the theory of algebraic Euler classes also constructed by Morel [Mor12, §8], Asok and Fasel proved splitting theorems for algebraic vector bundles [AF14c, AF14a, AF14b, AF15]. Grauert’s theorem is strongly linked to Gromov’s h-principle, a powerful tool in differential topology. There should be geometric applications of Theorem 1 and Conjecture 2 along the same lines.
Structure of the thesis
In the first chapter we introduce rigid analytic varieties and prove two lemmas (1.50 and 1.51) for later use. The second chapter treats vector bundles over rigid analytic varieties. We prove Serre– Swan Theorems for rigid analytic varieties (Theorems 2.12 and 2.14). We conclude that over an affinoid variety, vector bundles with respect to the Zariski topology 7 and vector bundles with respect to the G-topology coincide up to isomorphism (Corollary 2.15). Furthermore, we show that every vector bundle over a quasicompact or quasi-Stein rigid analytic variety admits a finite local trivialisation (Theorems 2.16 and 2.17). The chapter finishes with two short sections on line bundles and divisors. The third chapter treats the question of homotopy invariance with respect to an interval object I, i. e. for which interval I and under which assumptions on X is it true that every vector bundle over X × I is isomorphic to the pullback of some vector bundle over X along X × I −→ X? The first question in this context is the case X = Sp k, the question whether all vector bundles over I are trivial. In the case of algebraic varieties and I = A1, the answer is given by the Quillen–Suslin Theorem. For schemes, the general case is the subject of the Bass–Quillen conjecture. We review the Quillen–Suslin Theorem, the Bass–Quillen Conjecture and possible versions in the rigid analytic setting in the first two sections of chapter three. In the third section we collect examples where I-invariance is violated. Section 4 contains local results on homotopy invariance. The fifth section first collects the known results on B1-invariance by Gerritzen, Bartenwerfer and van der Put. Then we state Tamme’s Theorem on pro-homotopy invariance of the Picard group of a smooth k-affinoid variety if char k = 0 and deduce Theorem 3.31: Line bundles over a smooth k-rigid 1 analytic variety are Arig-invariant if char k = 0. In the fourth chapter we define motivic homotopy theories for rigid analytic varieties in the spirit of Morel–Voevodsky’s A1-homotopy theory and Ayoub’s B1-homotopy 1 1 theory (Proposition 4.15). We show that Arig is contractible in the B -homotopy category (Lemma 4.18). In the fifth and last chapter we prove Theorem 1/Theorem 5.41. The chapter starts out by recalling the classical theorems by Steenrod, Grauert and Morel on homotopy classification of vector bundles. We introduce Gromov’s h-principle and explain how it can be seen as a homotopy sheaf property. This gives the link to excision, which is used both in Morel’s proof and in our proof and which can be seen as an algebraic variant of the h-principle. In the fourth section we give a short account of classifying 1 spaces and prove that the simplicial classifying space of GLn in the Arig-homotopy category is the infinite Grassmannian, a proof we learned from Arndt (Proposition 5.34). In the fifth section we prove Theorem 4/Theorem 5.40, following the proof of the corresponding theorem in A1-homotopy theory by Asok–Hoyois–Wendt. Now we can deduce Theorem 1/Theorem 5.41. The thesis finishes with a brief resum´eand dicussion of open questions.
Acknowledgements
First of all, I want to thank Matthias Wendt for advise and patience, for giving me this intriguing question and making me learn a lot of very beautiful mathematics. Second, I thank Annette Huber-Klawitter for approving of this project and this way making the whole project possible in the first place, for sharing her life experience and being a great boss. Third, I had the opportunity to discuss questions around this thesis with the following mathematicians: Peter Arndt, Aravind Asok, Joseph 8 CHAPTER 0. INTRODUCTION
Ayoub, Federico Bambozzi, Antoine Ducros, Carlo Gasbarri, Fritz H¨ormann,Annette Huber-Klawitter, Marc Levine, Werner L¨utkebohmert, Florent Martin, Vytautas Paˇsk¯unas,J´erˆomePoineau, Dorin Popescu, Shuji Saito, Marco Schlichting, Georg Tamme, Konrad V¨olkel, Matthias Wendt. Thank you! Dear reader, if you do not occur in the list although we had an interesting dicussion, I apologise. I thank Maximilian Schmidtke and Eva Nolden for reading a draft of the first two chapters and pointing out typos and bad language. Once again, I thank Matthias Wendt for reading versions of this thesis carefully and for being severe with it. I thank Jens, Konrad, Matthias, Oliver and Shane for improving the Introduction. Big thanks to J´erˆomePoineau for pointing out a very stupid mistake. I am very grateful to Joseph Ayoub for acting as referee for this thesis, for reading it thoroughly and for his detailed comments. Furthermore, I thank Florian Sigloch for fairness, love and caring. I thank Eva Nolden and Frieder Sigloch for helping out in cases of emergency. Eva Nolden took days off to look after our children, a precious gift to Florian and me. I also thank Elisabeth and Wolfgang Hochmuth for planning their holidays according to our needs. Thanks to Karin Wanzel and Paul Sigloch, to Eva, Lena, Frank, Pablo, Felix, Bj¨orn and Melli. Dear family and friends, I do not think we would have made it without you. Last, I thank the state Baden-W¨urttemberg for a grant in the Brigitte Schlieben- Lange-Programm and the Deutsche Forschungsgemeinschaft (DFG) for funding me in the Graduiertenkolleg 1821. No research without money. Zu allerletzt danke ich Florian, Arthur, Jakoba und Birke daf¨ur,dass wir eine großartige Familie sind. Conventions
We will use the following conventions: All rings are commutative with 1. All algebras are associative with 1. All valuations are nontrivial. All rigid analytic varieties are separated. k is a complete, nonarchimedean, nontrivially valued field, unless stated other- wise. Residue fields are usually denoted by κ. Isomorphisms are denoted by ∼=. Weak equivalences are denoted by '. For other notation we refer to the index at the end of the thesis.
9 10 CHAPTER 0. INTRODUCTION Chapter 1
Rigid Analytic Varieties
The theory of rigid analytic varieties began with a seminar by John Tate at Harvard in 1961. Tate’s purpose was to understand degenerations of elliptic curves. His new approach soon led to a revival of nonarchimedean analysis. With the new theory at hand, complex analysts became interested in nonarchimedean analysis. Notes from Tate’s seminar circulated among them, but it took ten years before they finally were published [Tat71]. The area of nonarchimedean analysis was very active during the sixties and seventies. Several breakthroughs were achieved, for example Kiehl’s version of Cartan’s Theorems A and B for rigid analytic quasi-Stein spaces [Kie67], the Gerritzen–Grauert Theorem on locally closed immersions [GG69] or Raynaud’s theory which views a rigid analytic variety over a field as the generic fibre of a formal scheme over the valuation ring [Ray74].
Tate’s theory of rigid analytic varieties is stronlgy tied to algebraic and arithmetic geometry on the one hand via the analytification functor and GAGA theorems, on the other hand via the reduction functor on affinoid varieties. There are applications in singularity theory via the theory of Mumford curves and Schottky groups (cf. [FvdP04, 5.4]). Harbater proved Abhyankar’s conjecture in inverse Galois theory, using rigid analytic varieties. Stable reduction for curves over a complete discretely valued field can be proved using rigid analytic varieties [BL93a, vdP84].
Besides Tate’s theory, Berkovich’s theory of analytic spaces of seminorms [Ber90] is very fruitful. A third approach, Roland Huber’s theory of adic spaces [Hub93b, Hub93a], had a recent revival sparked by Scholze’s work [Sch12].
In this first chapter we review some aspects of rigid analytic geometry, mostly in order to fix terminology and notation. For a detailed exposition of affinoid algebras and the analytic aspects of the theory, we recommend the book by Bosch–G¨untzer– Remmert [BGR84]. The other classic is Fresnel–van der Put [FvdP04]. It takes a more algebro-geometric point of view, is less detailed and covers more recent material than [BGR84]. In particular, it contains sections on formal models and reduction, on generalised points, on cohomology theories and on Abhyankar’s conjecture.
11 12 CHAPTER 1. RIGID ANALYTIC VARIETIES
1.1 Definitions
Let k be a complete nonarchimedean valued field. Proposition/Definitions 1.1. a) A Banach k-algebra is a complete normed k-algebra. b) The Tate algebra X i1 in khT1,...,Tni := aiT1 ··· Tn all ai ∈ k, |ai| −→ 0 i1,...,in n i=(i1,...,in)∈N is a Banach algebra with respect to the Gauß norm
X i1 in aiT1 ··· Tn := sup |ai|. n i=(i1,...,in)∈N
All ideals of khT1,...,Tni are closed. A detailed exposition can be found in Bosch–G¨untzer–Remmert [BGR84, chapter 5]. c) An affinoid k-algebra is a Banach k-algebra A such that there exists a continuous epimorphism
α: khT1,...,Tni → A from some Tate algebra onto A. As the map α is open, it induces an isomor- phism of Banach k-algebras ∼ A = khT1,...,Tni/ ker α. Choosing different presentations α yields different norms. All of them are equivalent because they induce the same topology on A. Assume that A is reduced, i. e., has no nilpotent elements except zero. Then there is a distinguished norm kfk on A, the spectral norm, see Definition 1.3 below. d) A morphism of affinoid algebras is a continuous k-algebra homomorphism. Remark 1.2. The Tate algebras khT1,...,Tni and, more generally, affinoid algebras are noetherian [BGR84, 5.2.6 Theorem 1]. We can see an affinoid algebra A as an algebra of functions on its maximal spectrum Max A := {m ⊂ A | m maximal ideal} where f ∈ A maps a maximal ideal m to the residue class [f]m of f in A/m. The residue field A/m is a finite field extension of k. Thus, choosing an algebraic closure k¯ of k and embeddings of the residue fields A/m ,→ k,¯ the affinoid algebra A can be seen as an algebra of functions f : Max A → k¯
m 7→ [f]m. The absolute value | · | extends uniquely to the finite field extension A/m. 1.1. DEFINITIONS 13
Definition 1.3 (spectral seminorm [FvdP04, Definition 3.3.1]). The spectral semi- norm on an affinoid algebra A is
kfk := sup |f(x)|, f ∈ A. x∈Max A
Lemma 1.4 ([FvdP04, Corollary 3.4.4]). Let A be a reduced affinoid algebra. Then the spectral seminorm on A is a norm. The category of affinoid algebras has finite coproducts, given by the complete tensor product: Definition 1.5 (Complete tensor product [BGR84, 2.1.7]). Let R = (R, | |) be a normed ring and L, M normed R-modules. The function
| |: L ⊗R M −→ R+ defines a seminorm on L ⊗R M. The completion of L ⊗R M as a seminormed group with respect to this seminorm is denoted by L⊗ˆ RM and called the complete tensor product of L and M over R.
The complete tensor product L⊗ˆ RM of normed R-modules L, M is a normed R- module and a normed Rˆ-module [BGR84, p. 71]. Example 1.6. Let A be an affinoid algebra. Then X i1 in AhT1,...,Tni := aiT1 ··· Tn all ai ∈ A, kaikA −→ 0 i1,...,in n i=(i1,...,in)∈N is affinoid and ∼ AhT1,...,Tni = A⊗ˆ kkhT1,...,Tni [BGR84, p. 224].
For more details about the complete tensor product, see [BGR84, section 2.1.7 and p. 224]. Proposition 1.7 ([FvdP04, Lemma 3.7.1]). The category of affinoid k-algebras has finite coproducts, given by the complete tensor product over k. Definitions 1.8 (A◦, A◦◦). Let A be an affinoid algebra. a) The subring of power-bounded elements of A is ◦ r A = f ∈ A sup kf k < ∞ . r∈N
b) The set of topologically nilpotent elements of A is n o A◦◦ = f ∈ A f r −→ 0 . r→∞
The set A◦◦ is an ideal in A◦, cf. [BGR84, 1.2.4, 1.2.5]. 14 CHAPTER 1. RIGID ANALYTIC VARIETIES
Example 1.9. For the field Qp of p-adic numbers, the ring of power-bounded elements ◦ ◦◦ is the ring Qp = Zp of p-adic integers with maximal ideal Qp = pZp. Definitions 1.10 (affinoid subset, rational subset [FvdP04, 4.1.4, 4.1.1]). Let A be an affinoid algebra. A subset U ⊂ Max A is called affinoid if there is an affinoid algebra B and a morphism of affinoid algebras
ϕ: A → B such that the induced morphism
ϕ∗ : Max B → Max A maps ϕ∗(Max B) ⊂ U and is universal with respect to this property: For every affinoid algebra C and morphism
ψ : A → C with ψ∗(Max C) ⊂ U there is a unique morphism of affinoid algebras τ : B → C such that ψ = τ ◦ ϕ.
A subset U ⊂ Max A is called rational if there are elements f0, f1, . . . , fs ∈ A generating the unit ideal of A such that
U = {x ∈ Max A | |fi(x)| ≤ |f0(x)| for i = 1, . . . , s}.
Rational subsets are in particular affinoid and if U ⊂ Max A is rational as above, we have ∼ U = Max AhT1,...,Tsi/(f1 − f0T1, . . . , fs − f0Ts) .
Example 1.11 (complete localisation). If A is an affinoid k-algebra and f ∈ A with |f| ≤ 1, then
Ahf −1i := AhT i/(1 − fT )
−1 is called the complete localisation of A at f. The algebraic localisation Af =: A[f ] is dense in Ahf −1i [Bos14, §3.3]. Definition 1.12 (G-topology [FvdP04, Definition 2.4.1]). A G-topology on a set X consists of a family F of subsets of X (admissible subsets) and for each admissible subset U ∈ F a set Cov(U) of coverings by elements of F (admissible coverings) such that: a) ∅,X ∈ F . b) U, V ∈ F ⇒ U ∩ V ∈ F . c) U ∈ F ⇒ {U} ∈ Cov(U). d) U, V ∈ F,V ⊂ U, U ∈ Cov(U) ⇒ U ∩ V ∈ Cov(V ) where U ∩ V := {U 0 ∩ V | U 0 ∈ U}. 1.1. DEFINITIONS 15
e) Let U ∈ F, {Ui}i∈I ∈ Cov(U) and for each i ∈ I let Ui ∈ Cov(Ui). S S S 0 0 Then i∈I Ui ∈ Cov(U) where i∈I Ui := i∈I {U | ∃i ∈ I,U ∈ Ui}. In short: A G-topology on X is a Grothendieck pretopology T on a category C that has certain subsets of X as objects and inclusions as morphisms. Remark 1.13. Every topology induces a a G-topology: Every open set is admissible and every covering is admissible. A G-topology does not necessarily arise from a topology: First, an arbitrary union of admissible subsets does not need to be admissible. Second, a jointly surjective union of admissible subsets is not necessarily an admissible covering. Definition 1.14 ((very) weak G-topology [FvdP04, Definition 4.2.1], [BGR84, 9.1.4]). Let A be an affinoid algebra. The very weak G-topology on the set Max A of maximal ideals of A is defined as follows: • The admissible subsets are rational subsets of Max A. • Let U be an admissible subset. An admissible covering of U is a covering {Ui}i∈I by admissible subsets Ui for which there exists a finite subset J ⊂ I S such that U = i∈J Ui. The weak G-topology on Max A • has as admissible subsets finite unions of rational subsets of Max A. • Let U be an admissible subset. An admissible covering of U is a covering {Ui}i∈I by admissible subsets Ui for which there exists a finite subset J ⊂ I S such that U = i∈J Ui. Definition 1.15 (slightly finer ([FvdP04, p. 26], [BGR84, 9.1.2 Definition 1]). Let T and T 0 be G-topologies on a set X. Then T 0 is called slightly finer than T if a) every T -admissible set is T 0-admissible, b) every T -covering is a T 0-covering, c) every T 0-admissible set U has a T 0-covering by T -admissible sets, d) every T 0-covering of a T -admissible set can be refined by a T -covering. Definition 1.16 (strong G-topology [FvdP04, pp. 80f], [BGR84, 9.1.4]). Let A be an affinoid algebra. The strong G-topology TA on Max A is given by: The admissible subsets are the affinoid subsets of X. Let U be an admissible subset. An admissible covering of U is a covering {Ui}i∈I by admissible subsets Ui for which there exists a S finite subset J ⊂ I such that U = i∈J Ui. The strong G-topology is the finest G-topology which is slightly finer than the very weak G-topology [BGR84, 9.1.4]. For most purposes it is enough to keep in mind the weak G-topology. Besides those G-topologies, the set Max A carries two honest topologies: The topology Tk·k induced by the norm and the Zariski topology TZar. The Zariski topology is very coarse and ignores the analytic structure given by the norm. The topology induced by the norm is very fine: The topological space (Max A, Tk·k) is totally disconnected. The G-topology TA generated by rational subsets and finite coverings is coarser than Tk·k and finer than TZar. 16 CHAPTER 1. RIGID ANALYTIC VARIETIES
A G-topology (or more generally: a site) is enough to define the glueing condition for sheaves. Definition 1.17. Let (X, T ) be a set with a G-topology. Define X to be the category whose objects are the admissible subsets of X and whose morphisms are the inclusions. A presheaf of sets (or of rings, or of simplicial sets, or whatever) is a contravariant functor from X to the category of sets (or of rings, or of simplicial sets, or whatever). Let C be a category that has finite products. A presheaf F : X → C is a sheaf if sections over admissible coverings glue uniquely, i. e., for each admissible subset U ⊂ X and each admissible covering U of U, the diagram of restrictions Y Y F(U) → F(Ui) ⇒ F(Ui ∩ Uj) Ui∈U Ui,Uj ∈U is an equaliser diagram. Definition 1.18. A locally G-ringed space is a triple (X, TX , OX ) where X is a set, TX a G-topology on X and OX a presheaf of rings on X which is a sheaf with respect to TX . Definition 1.19 (affinoid variety). The affinoid variety associated to an affinoid algebra A is the locally G-ringed space
Sp A := (Max A, TA, OSp A) where the structure sheaf OSp A is the sheaf of Banach algebras on (Max A, TA) defined by
OSp A(U) := B if U = Sp B admissible and
O(Sp B,→ Sp B0) = (res: B0 → B).
See [BGR84, 7.3.2] for more details. A morphism
Sp A → Sp B of affinoid varieties is the morphism of locally G-ringed spaces induced by a morphism B → A of affinoid algebras. For more details, see [BGR84, 7.1.4 and 7.2.2]. Remark 1.20. The category of affinoid varieties with the very weak G-topology and the category of affinoid varieties with the strong G-topology are equivalent [FvdP04, pp. 80f]. Definition 1.21. The affinoid unit ball B1 is defined as
1 B := Sp(khT i). Lemma 1.22 (Fibre product of affinoid varieties [BGR84, 7.1.4 Proposition 4]). The category of affinoid k-varieties has fibre products. Suppose we are given affinoid maps Sp A → Sp C and Sp B → Sp C. Then the fibre product of Sp A with Sp B over Sp C is given by
Sp A ×Sp C Sp B := Sp(A⊗ˆ C B). 1.1. DEFINITIONS 17
Definition 1.23 (rigid analytic variety [FvdP04, Definition 4.3.1]). A k-rigid analytic variety is a locally G-ringed space (X, TX , OX ) where:
a) The structure sheaf OX is a sheaf of k-algebras.
b) There exists an admissible covering {Xi}i∈I ∈ Cov(X) such that for each i ∈ I,
the space (Xi, TX |Xi , OX |Xi ) is an affinoid variety.
c) Admissibility of subsets can be tested on affinoids: U ∈ TX if and only if
U ∩ Xi ∈ TXi for all i.
The sheaf OX is called the structure sheaf of X. n Example 1.24 (The affine n-space Arig [BGR84, 9.3.4 Example 1]). Let η ∈ k with |η| > 1 and define
−i −i Ai := khη T1, . . . , η Tni. n i This means that Sp(Ai) = B|η|i is the n-dimensional ball of polyradius |η| . The algebras Ai form a chain
khT1,...,Tni = A0 ) A1 ) A2 ) ··· ) k[T1,...,Tn] and the inclusions Ai ⊃ Ai+1 correspond to inclusions of rational subsets n n B|η|i ⊂ B|η|i+1 . n The affine n-space Arig is defined as the colimit
n [ n Arig := B|η|i i∈N of all those inclusions of rational subsets. It does not depend on the choice of the n element η. Affine space Arig is a rigid analytic variety which is not affinoid. It is an unbounded Stein space, see Definition 1.30 below. Definition 1.25 (Fibre product of rigid varieties [BGR84, 9.3.5 Theorem 2]). Let X → S and Y → S be morphisms of rigid analytic varieties. Then the fibre product X ×S Y of X with Y over S exists in the category of rigid analytic varieties. Furthermore, if X, Y and S are affinoid, then this fibre product coincides with the fibre product of X with Y over S in the category of affinoid varieties constructed in Lemma 1.22. Definition 1.26 (closed immersion [BGR84, 7.1.4 Definition 3 and 9.5.3, p. 388]). An affinoid map Sp A → Sp B is called a closed immersion if the corresponding map B → A on affinoid algebras is an epimorphism. A morphism ϕ: X → Y of rigid analytic varieties is called a closed immersion if there exists an admissible affinoid covering {Ui}i∈I of the target Y such that for all i ∈ I, the induced map −1 ϕ (Ui) → Ui is a closed immersion of affinoid varieties. Definitions 1.27. a) An affinoid algebra is reduced if it has no nonzero nilpotent elements. An affinoid variety Sp A is reduced if A is. b) A rigid analytic variety X is separated if the diagonal morphism
∆: X → X ×Sp k X is a closed immersion [BGR84, 9.6.1, Definition 1]. 18 CHAPTER 1. RIGID ANALYTIC VARIETIES
c) A rigid analytic variety is quasiseparated if the intersection of any two open affinoid domains is a finite union of open affinoid domains [FvdP04, Defini- tions 7.3.4]. Remark 1.28. Every affinoid variety is quasi-separated as the intersection of two affinoid subdomains of an affinoid variety is again an affinoid subdomain [BGR84, 7.2.2 Corollary 5]. There are three special classes of rigid analytic varieties that we are particularly interested in: Affinoid varieties, quasicompact varieties and quasi-Stein varieties. Definition 1.29 (quasicompact rigid analytic variety [FvdP04, Definitions 7.3.4]). A rigid analytic variety is called quasicompact if it has an admissible covering by finitely many affinoid subsets. Quasi-Stein and Stein varieties were defined by Kiehl in [Kie67] where he also proved the analogues of Cartan’s Theorem A and Theorem B for them. They are analogues of complex Stein spaces in the nonarchimedean world. Definition 1.30 ((quasi-)Stein [Kie67, Definition 2.3]). a) A rigid analytic space X is quasi-Stein if there is an admissible covering by open affinoid subspaces
U1 ⊂ U2 ⊂ U3 ⊂ · · ·
such that for all i, the image of OX (Ui+1) is dense in OX (Ui).
b) The space X is Stein if additionally, for each i there exists ai+1 ∈ k with (i+1) (i+1) 0 < |ai+1| < 1 and a system of topological generators f1 , . . . , fni+1 of O(Ui+1) such that
(i+1) Ui = {x ∈ Ui+1 | |fj (x)| ≤ |ai+1| for j = 1, . . . , ni+1}. Definition 1.31 (coherent sheaf [FvdP04, Definition 4.5.1]). Let X be a rigid analytic variety and F a sheaf of OX -modules. The sheaf F is called coherent if there exists an admissible covering by affinoids {Xi}i∈I ∈ Cov(X) and for every i a finitely generated OX (Xi)-module Mi such that the restriction of F to Xi is isomorphic to ˜ ˜ Mi as a sheaf of OXi -modules. Here the sheaf Mi is defined by
U 7→ Mi ⊗OX (Xi) OX (U) for U ⊂ Xi rational.
Theorem 1.32 (Kiehl [Kie67, Satz 2.4]). Let X be a quasi-Stein space and {Ui}i∈N an admissible covering as in Definition 1.30a). Let G be a coherent sheaf on X. Then the following hold:
a) The image of G(X) is dense in G(Ui) for all i. b) The cohomology groups Hi(X, G) vanish for i > 0 (Theorem B).
c) For each x ∈ X, the image of G(X) in the stalk Gx generates this stalk as an OX,x-module (Theorem A). i Theorem 1.33 (L¨utkebohmert [L¨ut73]). a) All fj can be chosen to be global i functions: fj ∈ OX (X) by [L¨ut73,Korollar 4.2]. b) Stein varieties are separated [L¨ut73,Satz 3.4]. 1.1. DEFINITIONS 19
Remark 1.34. a) L¨utkebohmert proved moreover that a Stein variety X over an algebraically closed field k can be embedded in some kN if X is smooth or if the local embedding dimension of X is bounded [L¨ut73, Theorem 4.21]. Here, an embedding is a finite, injective, proper analytic function onto a closed subvariety of kN such that its differentials generate the universal differential module of X. For precise definitions and statements we refer to L¨utkebohmert’s article.
b) There is also a weaker notion of Steinness, introduced by Liu in [Liu88] and investigated further in [Liu89]. Liu calls a rigid analytic space X Stein if for every coherent sheaf F on X all higher cohomology groups Hi(X, F), i ≥ 1, vanish. Definition 1.35 (Stein algebra). Let X be a rigid analytic (quasi-)Stein variety. The k-algebra OX (X) of global functions on X is called a (quasi-)Stein algebra. (Quasi-)Stein algebras are Fr´echet algebras, but in general not Banach. They are complete, but their topology does not necessarily arise from a norm. As the following example shows, (quasi-)Stein algebras are in general not noetherian. Example 1.36 (compare [GR04, Remark 3 on p. 179]). Let X be a (quasi-)Stein space that contains an infinite discrete subset D which does not have a limit point in X. Let A = OX (X) be the corresponding (quasi-)Stein algebra. The nonarchimedean analogue of Weierstrass’ product theorem holds: Given an infinite discrete subset D0 ⊂ X which does not have a limit point in X and an assignment
0 D −→ N d 7−→ m(d)
0 there exists a function f ∈ OX (X) such that for all d ∈ D , the function f has a zero in d of multiplicity m(d). The proof is the same as in the complex case, e. g. as in [Rud87]. We now construct a non-finitely generated ideal m ⊂ A. The elements of the ideal
I = {f ∈ A | f(x) = 0 for almost all x ∈ D} have no common zeroes. So a maximal ideal m containing I cannot be finitely generated.
In complex analysis, Stein algebras were introduced by Forster [For64, For67] and studied by several mathematicians since then. In nonarchimedean analysis they appear implicitly in L¨utkebohmert’s 1973 article [L¨ut73]. The more general Fr´echet– Stein algebras and their coadmissible modules (corresponding to coherent sheaves on the Stein varieties) were introduced by Schneider and Teitelbaum [ST03] in the context of p-adic analytic groups and Langlands theory. There are approaches to unify the archimedean and the nonarchimedean theory. Poineau constructed Berkovich spaces over Z, capturing both the archimedean and the nonarchimedean theory, cf. [Poi13]. Poineau uses Liu’s definition of Stein spaces. Bambozzi–Ben Bassat–Kremnizer investigate the topology of Stein spaces over any valued base field [BBK15]. They use Kiehl’s definition of nonarchimedean Stein spaces. 20 CHAPTER 1. RIGID ANALYTIC VARIETIES
1.2 Formal models and reduction
Definition 1.37 (reduction of an affinoid algebra [BGR84, 6.3]). Let A be an affinoid k-algebra. We define its reduction by
A˜ := A◦/A◦◦ where A◦ ⊂ A is the subring of power-bounded elements and A◦◦ the set of topologi- cally nilpotent elements of A. The set A◦◦ is an ideal in A◦. The reduction A˜ is an affine k˜-algebra, that is, a quotient of a polynomial ring over k˜ in finitely many variables. In fact, reduction defines a functor from the category of affinoid k-algebras to the category of affine k˜-algebras. Examples 1.38. a) The reduction of the p-adic numbers Qp is ∼ Qfp = Zp/pZp = Fp,
the field with p elements. b) The power-bounded elements of the affinoid algebra
A = khT1i
are the elements whose coefficients are all power-bounded, i. e.,
◦ ◦ A = k hT1i.
Similarly, the topologically nilpotent elements are
◦◦ ◦◦ ◦ A = k · k hT1i.
Thus, the reduction of A is
˜ ˜ A = k[T1].
−1 c) Let A be affinoid and f ∈ A with kfksup = 1. Then Ahf i ⊂ A is an affinoid subset. Its reduction is
−1 ˜ ˜−1 ˜ A^hf i = A[f ] = Af˜
the ring-theoretic localisation of A˜ at f˜ [BGR84, 7.2.6 Proposition 3]. d) Assume char k˜ > 3. Let
2 A1 = khT1,T2i/(T1 − T2(T2 − 1)(T2 − 2)).
Its reduction is
˜ ˜ 2 A1 = k[T1,T2]/(T1 − T2(T2 − 1)(T2 − 2)). 1.2. FORMAL MODELS AND REDUCTION 21
e) Assume char k˜ > 3 and choose π ∈ k with 0 < |π| < 1. Let
2 A2 = khT1,T2i/(T1 − T2(T2 − π)(T2 − 2π)). Its reduction is
˜ ˜ 2 3 A2 = k[T1,T2]/(T1 − T2 ).
Although A2 is smooth, its reduction has a cuspidal singularity. f) Again, assume char k˜ > 3 and choose π ∈ k with 0 < |π| < 1. Let
2 A3 = khT1,T2i/(T1 − T2(T2 − 1)(T2 − 2π)). Its reduction is
˜ ˜ 2 2 A3 = k[T1,T2]/(T1 − T2 (T2 − 1)).
This time, A3 is smooth and its reduction has a nodal singularity. Remark 1.39. Examples e) and f) show that a smooth affinoid algebra can have a singular reduction. Assume that π ∈ k has a sixth root ζ in k. Then we can see the affinoid varieties corresponding to the algebras in examples d) and e) as pieces of different size cut out from the Stein variety
[ n 2 m m m X = Sp khT1,T2i/(η T1 − δ T2(δ T2 − 1)(δ T2 − 2)) n,m∈N for arbitrary η, δ ∈ k with |η|, |δ| > 1. Obviously, X does not depend on the choice 2 of η and δ as long as |η|, |δ| > 1. The Stein variety X embeds into Arig. Let
2 −n −m Bn,m = Sp khη T1, δ T2i
n m 2 −1 be the polydisc of polyradius (η , δ ) in Arig. Choosing η = δ = ζ gives
2 Sp A1 = X ∩ B0,0 2 Sp A2 = X ∩ B3,2.
Let A be an affinoid k-algebra and Ae its reduction. Then A◦ corresponds to a formal k◦-scheme X that has Sp A as its generic fibre and Spec Ae as its special fibre. Raynaud [Ray74] realised that in many cases one can work with formal schemes instead of rigid analytic varieties and hence apply tools from algebraic geometry. Definition 1.40. The formal scheme X is called the canonical model of X. The special fibre of X is an affine algebraic k˜-variety and denoted by X˜ c. It is called the canonical reduction of X. Remark 1.41. Obviously, X˜ c = Spm A˜. Here we denote by Spm R the algebraic variety defined by a ring R, that is, the maximal spectrum of R together with the Zariski topology and the structure sheaf. For nonaffinoid rigid analytic varieties, formal models can still exist, but usually there is no canonical choice any more. 22 CHAPTER 1. RIGID ANALYTIC VARIETIES
Definition 1.42 (formal model, analytic reduction). If the rigid analytic k-variety X is the generic fibre of the formal k◦-scheme X , then X is called a formal model of X.
If a rigid analytic variety has a locally finite covering by affinoids, then it has a formal model. Raynaud’s approach was pursued further in the “Formal and rigid geometry” series [BL93a, BL93b, BLR95a, BLR95b] by Bosch, L¨utkebohmert and Raynaud. Among other things, they develop the technique of admissible formal blowup in [BL93a]. Admissible formal blowup gives resolution of singularities in the case of equal characteristic zero, cf. Nicaise [Nic09, Proposition 2.43]. This follows from Temkin’s resolution of singularities for noetherian quasi-excellent schemes of characteristic zero [Tem08]. Compare also Hartl [Har03]. Resolving singularities in the special fibre, it leads to: Theorem 1.43 (Ayoub [Ayo15, Proposition 1.1.63]). Let k = κ((T )) for κ a field of char κ = 0. Every smooth k-affinoid variety has a semistable formal model (see Definition 1.46). Definition 1.44 (admissible formal scheme). a) A k◦-algebra A is called admis- sible if it has no k◦◦-torsion, i. e., if {a ∈ A | ∃c ∈ k◦◦ : ca = 0} = {0} [BL93a, p. 293].
b) An affine formal k◦-scheme Spf A is admissible if A is admissible [BL93a, p. 296].
c) A formal k◦-scheme X is admissible if it has a covering by affine admissible formal schemes. Lemma 1.45 ([BL93a, Proposition 1.7]). Let A be a k◦-algebra that is complete and ◦◦ separated with respect to the k -adic topology and let {Spf Bi}i∈I be an affine open covering of Spf A. Then A is an admissible k◦-algebra if and only if for all i ∈ I, the ◦ k -algebra Bi is admissible. In this case the canonical maps A → Bi are flat.
Hence, Definition 1.44c) makes sense. Definition 1.46 (semistable model, [HL00, Definition 1.1]). An admissible formal scheme X is called strictly semistable if
a) its generic fibre Xη is smooth over k,
b) its special fibre Xσ is geometrically reduced, i. e., reduced and still so after base change to the separable closure k˜sep of k˜,
c) the special fibre Xσ has (at most) normal crossings singularities.
∼ If X is a strictly semistable formal scheme with Xη = X, then X is called a strictly semistable formal model of X. Remark 1.47 ([Har03, Remark 1.1.1]). Such a formal scheme is regular. 1.3. FLATNESS, SMOOTHNESS AND ETALENESS´ 23
1.3 Flatness, smoothness and ´etaleness
Definition 1.48 (flat morphism). a) A morphism ϕ: A → B of affinoid algebras is flat if the functor
ModA −→ ModB
M 7−→ B ⊗A M
from A-modules to B-modules is exact. b) A morphism φ: Sp B → Sp A of affinoid varieties is flat if the corresponding morphism ϕ: A → B is flat. c) A morphism φ: X → Y of rigid analytic varieties is flat if for each x ∈ X there exists an affinoid subset U 3 x such that φ|U is flat. Lemma 1.49. Let A be an affinoid algebra. Then A → AhT i is flat. If |f| ≤ 1, then A → Ahf −1i is flat. Inclusion of an affinoid subdomain is flat.
Proof. The morphisms A → A[T ] and A → Af are flat. Completion is faithfully flat [Bou61, §3, no. 4 Th´eor`eme3 and no. 5 Proposition 9]. Inclusion of an affinoid subdomain is flat [Bos14, 4.1 Corollary 5].
We extend this to the coefficient rings appearing in the definition of a quasi-Stein variety. Lemma 1.50. Let
X = colim(U1 ⊂ U2 ⊂ U3 ⊂ · · · ) be a quasi-Stein variety with Ui = Sp Ai. Then A := OX (X) → Ai is flat for all i.
Proof. Using Kiehl’s Theorem B [Kie67, Satz 2.4], the proof works just as Gruson’s proof of [Gru68, V, Corollaire 1]: Let I ⊂ A be any finitely generated ideal. We want to show that the morphism
I ⊗A Ai −→ A ⊗A Ai (1.1) is injective. Then Ai is flat over A by [sta, Tag 00H9, Lemma 10.38.5].
The ideal I gives rise to a coherent sheaf I on X. For each j ≥ i, Aj is flat over Ai by Lemma 1.49. Therefore, for each j ≥ i, the morphism
I(Uj) ⊗Aj Ai −→ Aj ⊗Aj Ai
1 is injective. By Kiehl’s Theorem B [Kie67, Satz 2.4], the derived limit limj (I(Uj)) vanishes and hence the morphism 1.1 is injective, too.
We want to use Lemma 1.50 in combination with the following lemma. It is the Banach version of a well-known theorem in commutative algebra. 24 CHAPTER 1. RIGID ANALYTIC VARIETIES
φ Lemma 1.51. Let A −→ A0 be a flat morphism of commutative Banach algebras. Let M be a finitely presented complete normed A-module and N a finitely generated complete normed A-module. Then the natural map
0 0 0 σˆ : A ⊗ˆ A HomA(M,N) −→ HomA0 (A ⊗ˆ AM,A ⊗ˆ AN) is an isomorphism. To prove this, we want to use the fact that there is no essential difference between the category of finitely generated complete normed A-modules with continuous A-linear maps as morphisms and the category of finitely generated A-modules with A-linear maps as morphisms without demanding continuity [BGR84, §3.7.3]. During the proof, Hom-sets in the first named category are denoted by Homcont, morphisms being both linear and continuous. On the other hand, Hom-sets in the second named category consist of morphisms that only have to be linear and will be denoted by Homlin.
Proof. We only need to prove thatσ ˆ is bijective. There is a commutative diagram
0 ˆ cont σˆ cont 0 ˆ 0 ˆ A ⊗A HomA (M,N) / HomA0 (A ⊗AM,A ⊗AN) . O O o α o β
0 lin ∼ lin 0 0 A ⊗A HomA (M,N) σ / HomA0 (A ⊗A M,A ⊗A N)
The maps α and β are bijections by [BGR84, §3.7.3 Proposition 2, Proposition 6]. The map σ is an isomorphism by [Lam06, Proposition I.2.13] or [Eis95, Proposition 2.10]. Thus,σ ˆ is bijective, too. Definition 1.52 ([Ayo15]). a) An affinoid k-variety X = Sp A is called regular in x ∈ X if OX,x is regular. Furthermore, X is called smooth in x if for every 0 0 0 finite field extension k ⊂ k , the affinoid k -variety Sp(A⊗ˆ kk ) is regular in each point x0 lying above x. b) A rigid analytic variety X is called regular (smooth) if it is regular (smooth) in all x ∈ X. c) A morphism f : X → Y of rigid k-varieties is smooth if it is flat and for each y ∈ Y the morphism restricted to the fibre f −1(y) → κ(y) is smooth. d) The morphism f is ´etale if additionally all its non-empty fibres are of dimension zero. e) The morphism f is ´etalein x ∈ X if x has an affinoid neighbourhood U 3 x such that the restriction f|U is ´etale. Definition 1.53 (smRig). The category of smooth rigid analytic k-varieties with rigid analytic morphisms is denoted by smRig k. The subscript k will be suppressed in the notation except when there is a chance of confusion. As for schemes, there are several ways to define ´etaleness.Several characterisations of smooth and ´etalemaps between rigid analytic varieties can be found in the articles 1.3. FLATNESS, SMOOTHNESS AND ETALENESS´ 25 by de Jong–van der Put [dJvdP96, section 3] and Ayoub [Ayo15]. For example, if f : Y → X is a morphism, one can define the sheaf Ωf of relative differentials as in algebraic geometry. It is a coherent OY -module. Smoothness can as usual be characterised by looking at Ωf or the Jacobian matrix. Berkovich spaces and Huber’s adic spaces are very similar to rigid analytic varieties, especially if we only care for affinoid varieties. So we find further characterisations of ´etale,smooth, flat, unramified and finite maps in Berkovich’s paper [Ber93, section 3.3] and Roland Huber’s book [Hub96, section 1.7]. Theorem 1.54 (Berkovich [Ber93, Theorem 1.6.1]). The category of paracompact strictly k-analytic Berkovich spaces and the category of quasiseparated rigid analytic k-varieties that have an admissible covering of finite type are equivalent. An affinoid covering is said to be of finite type if each of its members intersects only finitely many of the other members. Remark 1.55. Berkovich’s definition of ´etaleness[Ber93, Definition 3.3.4] is equivalent to ours by [Ayo15, Theor`eme1.1.47]. Lemma 1.56 ([Ber93, Proposition 3.3.10]). Let ϕ: X → Y be a finite morphism of affinoid varieties. Then the set of points x ∈ X in which ϕ is ramified, respectively non-´etale,is Zariski closed.
Proof. This is true for a quasifinite morphism of Berkovich spaces by [Ber93, Propo- sition 3.3.10]. By Theorem 1.54 it also holds for affinoid varieties. 26 CHAPTER 1. RIGID ANALYTIC VARIETIES Chapter 2
Vector Bundles over Rigid Analytic Varieties
A vector bundle over a space X can be thought of as a space E together with a projection to X such that, locally on X, the total space E looks like the product of X with a vector space. The definition depends on the choice of topology on X. In the first section we define different kinds of vector bundles.
In the second section we prove a Serre–Swan theorem for vector bundles over rigid analytic varieties. There are several situations in various fields of geometry where vector bundles are completely determined by their global sections. In those situations, the global sections of a vector bundle form a projective module over the ring of global sections of the structure sheaf of the underlying geometric object. The first instance of this principle was discovered by Serre [Ser55, §I.50, p. 242] and it was maybe the starting point to ask his famous question about projective modules. Serre showed that vector bundles of finite rank over an affine variety correspond to finitely generated projective modules over the coefficient ring. Swan [Swa62, Theorem 2] proved the same for topological vector bundles over compact Hausdorff spaces. Serre’s Theorem, Swan’s Theorem and other theorems of this type are usually referred to as Serre–Swan Theorems. A Serre–Swan Theorem for complex Stein spaces follows from Forster’s work [For64, For67]. Morye extracted the greatest common divisor of Serre–Swan Theorems, i. e., which features are responsible for a Serre–Swan Theorem to hold. We use her nice and very clear exposition [Mor09] to prove that a Serre–Swan Theorem also holds for vector bundles over rigid analytic quasi-Stein varieties. As a corollary we get that on an affinoid variety vector bundles with respect to the G-topology and vector bundles with respect to the Zariski topology coincide up to isomorphism.
The chapter concludes with a few remarks on line bundles and divisors.
27 28 CHAPTER 2. VECTOR BUNDLES OVER RIGID ANALYTIC VARIETIES
2.1 A word on topologies
Let us first discuss the concept of a vector bundle very generally and, for the beginning, informally. A vector bundle on a geometric object X is a projection from another geometric object p: E → X whose fibres have the structure of isomorphic vector spaces and such that there exists a local trivialisation. One often identifies the vector bundle p with its sheaf of local sections on X. Adopting this point of view, a vector bundle on a geometric object X is a locally free OX -module where OX is the sheaf of functions on X. We explain now what this means. Definition 2.1. a) Let C be a geometric category: A geometric object X ∈ Ob(C) is supposed to come with • a site X whose objects are morphisms (U → X) ∈ Mor(C) and whose morphisms are commutative triangles
V
U / X
of morphisms in C. By abuse of notation, we write U ∈ Ob(X) and (V → U) ∈ Mor(X).
• a sheaf OX of rings on the site X, the structure sheaf of X. b) A sheaf F on X is by definition a presheaf from X that satisfies the cocycle condition. Assuming that the category X has fibre products and that the presheaf has values in a category which admits products, the cocycle condition reads: For all U ∈ Ob(X) and all {Ui → U}i∈I ∈ Cov(U), the diagram of restrictions Y Y F(U) → F(Ui) ⇒ F(Ui ×X Uj) i∈I i,j∈I
is an equaliser diagram. c) By a sheaf on X we mean a sheaf on X.
d) An OX -module is a sheaf of modules such that for each U ∈ Ob(X), its value F(U) is an OX (U)-module.
e) It is locally free if there exists {Ui}i∈I ∈ Cov(X) such that the F(Ui) are free OX (Ui)-modules for all i.
f) A covering {Ui}i∈I such that all F(Ui) are free is called a local trivialisation of F. g) If K is a field, a locally free sheaf F of K-algebras of rank n is called a K-vector bundle of rank n. h) Two vector bundles are called isomorphic if they are isomorphic as sheaves. 2.1. A WORD ON TOPOLOGIES 29
A local trivialisation {Ui}i∈I together with cocycle maps
i n ϕj ∈ Aut (OX (Ui ×X Uj)) = GLn(OX (Ui ×X Uj)), i, j ∈ I fulfilling the cocycle conditions
i ϕi = id i j i ϕjϕl = ϕl is enough to determine the isomorphism type of the vector bundle. To conclude, let us note: When we speak of a vector bundle on X, we make implicit choices: The choice of site and the choice of structure sheaf. We will be careful to always specify which (Grothendieck) topology we are referring to. On the contrary, there will be no chance of confusion about the structure sheaf.
The term “vector bundle” means “locally free OX -module” throughout this thesis. The two points of view, i. e. the projection map versus the sheaf of sections, reappear in section 5.2 in the context of fibre bundles and Gromov’s h-principle. We will see that if a partial differential relation on a manifold M satisfies Gromov’s h-principle, then this means that a presheaf of certain sections of the jet bundle of M satisfies the sheaf conditions up to homotopy. Such a presheaf is called a homotopy sheaf . In the world of rigid analytic varieties, there are in particular the following two types of vector bundles: Definitions 2.2 (analytic/Zariski vector bundle). a) Vector bundles with respect to the G-topology of a rigid analytic variety are called analytic vector bundles. b) Vector bundles with respect to the Zariski topology of an affinoid variety are called Zariski vector bundles. In fact the two notions are the same: We will see in Corollary 2.15 that the iso- morphism classes of Zariski vector bundles and the isomorphism classes of analytic vector bundles coincide. The reason is that both Zariski and analytic vector bundles on affinoid varieties are determined by their global sections, and those coincide by definition. Remark 2.3. The definition of an analytic vector bundle coincides with the usual definition of a vector bundle on a rigid analytic variety [FvdP04, p. 87]. Furthermore, every quasiseparated rigid analytic variety can be seen as a Berkovich space by Theorem 1.54 by Berkovich. Berkovich spaces are topological spaces and thus come with a notion of continuous vector bundle. But often one wants to take the analytic structure into account. Definition 2.4 (Vector bundle on a Berkovich space). Let X be a Berkovich space.
a) Let OX be the structure sheaf of analytic functions on X. An analytic vector bundle on X is a locally free OX -module. b) Let C0(X) be the sheaf of continuous functions on X.A continuous vector bundle on X is a locally free C0(X)-module. 30 CHAPTER 2. VECTOR BUNDLES OVER RIGID ANALYTIC VARIETIES
Let X be a quasiseparated rigid analytic space and XB the corresponding Berkovich space. Do their categories of analytic vector bundles coincide? The answer depends on X. Proposition 2.5 (Berkovich). If every point x ∈ XB has an affinoid neighbourhood, then the category of analytic vector bundles on X and the category of analytic vector bundles on XB are equivalent.
Proof. Berkovich calls an analytic space Y good if every point x ∈ XB has an affinoid neighbourhood [Ber93, Remark 1.2.16]. If Y is good, the category of analytic vector bundles with respect to the Berkovich topology on Y and the category of vector bundles with respect to the G-topology on Y are equivalent by [Ber93, Proposition 1.3.4]. The category of vector bundles with respect to the G-topology on Y and the category of analytic vector bundles on the associated rigid analytic space are equivalent by [Ber93, bottom of p. 37]. Berkovich’s G-topology on Y is defined on [Ber93, p. 25]. The G-topology on the corresponding rigid analytic space coincides with the one in Definition 1.16.
Example 2.6. If X is an affinoid space, the corresponding Berkovich space is called strictly affinoid. Strictly affinoid Berkovich spaces are obviously good. Each strictly affinoid Berkovich space is of the form XB for an affinoid rigid analytic space X by construction of the correspondence 1.54. Berkovich spaces corresponding to rigid analytic quasi-Stein spaces are good by construction. Remark 2.7. It is an open question under which conditions the category of analytic vector bundles and the category of continuous vector bundles on a Berkovich space are equivalent.
2.2 Serre–Swan for rigid analytic quasi-Stein va- rieties
In this section we prove the rigid analytic version of the Serre–Swan theorem: Isomorphism classes of vector bundles of rank n over a rigid analytic quasi-Stein variety X correspond to isomorphism classes of projective modules of rank n over the ring of global functions OX (X). There is a general principle for proving Serre–Swan theorems. This was analysed by A. Morye [Mor09] and we give a short account of her argument. Definition 2.8. Let (X, T , OX ) be a locally G-ringed space. A coherent sheaf of OX -modules is called finitely generated by global sections if there is a finite set of global sections whose images in the stalk at any point x ∈ X generate that stalk as an OX,x-module. The following notion will only be needed in this section: Definition 2.9 (admissible subcategory [Mor09, Definition 1.1]). Let (X, OX ) be a locally G-ringed space. A subcategory C of the category of OX -modules is called an admissible subcategory if 2.2. SERRE–SWAN FOR RIGID ANALYTIC QUASI-STEIN VARIETIES 31
C1 C is a full abelian subcategory of the category of OX -modules with internal
Hom, i. e., for F, G ∈ Ob(C) the sheaf HomOX (F, G) is also an object of C, C2 every sheaf in C is acyclic and generated by global sections, and
C3 C contains the category of locally free OX -modules. Theorem 2.10 (General Serre–Swan [Mor09, Theorem 2.1]). Let (X, OX ) be a locally G-ringed space such that a sequence of coherent sheaves
0 → E → F → G → 0 on X is exact if and only if for each x ∈ X the sequence of OX,x-modules
0 → Ex → Fx → Gx → 0 is exact (“coherent sheaves may be tested on stalks”).
Assume that the category of OX -modules contains an admissible subcategory C and that every locally free sheaf of bounded rank on X is finitely generated by global sections. Then the Serre–Swan theorem holds for (X, OX ), i. e., the category of locally free sheaves of bounded rank on X and the category of finitely generated projective OX (X)-modules are equivalent via the global sections functor. Morye’s Theorem is formulated for a locally ringed space, but she actually proves the slightly stronger statement given here. The site associated to a locally ringed space always has enough points, so all sheaves may be tested on stalks (see [MLM94, Chapter IX, §3, Proposition 3 on p. 480]).
Proof. Let (X, OX ) be a locally G-ringed space such that coherent sheaves may be tested on stalks. The proof proceeds in the following steps:
a) If D is a full subcategory of the category of OX -modules such that OX ∈ Ob(D) and such that every sheaf in D is generated by global sections, then the global sections functor is fully faithful on D [Mor09, Proposition 2.2].
b) Under the same assumptions, the composition of the “associated sheaf” functor
S : OX (X)-modules −→ OX -modules
M 7−→ (U 7→ M ⊗OX (X) OX (U)
with the global sections functor is essentially surjective: Every F ∈ Ob(D) is isomorphic to S(F(X)) [Mor09, Proposition 2.5].
c) Assume that the category of OX -modules contains an admissible subcategory. Let F be locally free sheaf of bounded rank which is finitely generated by global sections. Then F(X) is a finitely generated projective OX (X)-module [Mor09, Proposition 2.8]. 32 CHAPTER 2. VECTOR BUNDLES OVER RIGID ANALYTIC VARIETIES
Proof. By assumption there exist an n ∈ N and an epi- n morphism u: OX F. As HomOX (F, ker u) is acyclic, Ext1 (F, ker u) = 0 and hence the exact sequence OX
n u 0 → ker u → OX → F → 0
splits and the exact sequence induced on global sections
n u 0 → (ker u)(X) → (OX (X)) → F(X) → 0
splits, too.
Recall the definition of quasi-Stein spaces and Stein spaces from 1.30. S Lemma 2.11. Let X = i(Sp(Ai)) be a rigid analytic quasi-Stein variety of bounded dimension, i. e., supx∈X dim OX,x =: d < ∞. Let E be a vector bundle on X. Then E is finitely generated by global sections.
Proof. By Theorem A [Kie67, Satz 2.4.3], every coherent sheaf on a quasi-Stein space is generated by global sections. Let A := OX (X). We have to show that E(X) is finitely generated as an A-module. Gruson does this for Stein spaces which are unions of polydiscs [Gru68, V, Th´eor`eme1] and the same proof works for general quasi-Stein spaces. We give it now.
Let n the rank of the Ai-module E(Sp Ai). Both are independent of i. Let F := An(d+1). Of course there exist homomorphisms F → E(X). We show that the set E := {F E(X)} of epimorphisms from F onto E(X) is dense in HomA(F, E(X)). Then E is in particular nonempty and we know that E(X) is finitely generated.
The map Ai+1 → Ai is flat by Lemma 1.49. For each i, set Fi := F ⊗ˆ AAi and ∼ ∼ Mi := E(Sp Ai). We have that Fi = Fi+1⊗ˆ AAi and Mi = Mi+1⊗ˆ AAi for all i. Moreover, Fi and Mi are finitely presented for all i. Using Lemma 1.51, we see that the projective system
· · · ← Ai ← Ai+1 ← · · · induces a projective system
· · · ←− HomAi (Fi, E(Sp Ai)) ←− HomAi+1 (Fi+1, E(Sp Ai+1)) ←− · · ·
The limit is HomA(F, E(X)) by Kiehl’s Theorem B [Kie67, Satz 2.4.2]. As flat base change maps epimorphisms to epimorphisms, we can look at the projective system
· · · ←− Ei ←− Ei+1 ←− · · · where Ei := {Fi E(Sp Ai)}. We claim that its limit is E. It is clear that E ⊂ lim(Ei)i∈N. We have to show that the limit of a system (φi)i ∈ (Ei)i of epimorphisms is an epimorphism: lim(φi)i ∈ E. As the category of coherent sheaves 2.2. SERRE–SWAN FOR RIGID ANALYTIC QUASI-STEIN VARIETIES 33 on a rigid analytic variety is abelian [FvdP04, p. 87], the sheaf cokernel D of F → E is coherent. By Kiehl’s Theorem B [Kie67, Satz 2.4.2], it is enough to test surjectivity of a map of coherent sheaves on an affinoid covering. But we have D(Sp Ai) = 0 for all i, because there the morphisms F(Sp Ai) E(Sp Ai) are epi. Thus, the cokernel sheaf D is zero and we have lim(φi)i ∈ E. A lemma by Gruson [Gru68, V, Lemme 2] states that if B is any Banach algebra of topologically finite type of Krull dimension d and if P is a projective B-module of rank p and L is the free B-module B(d+1)p, then the set of epimorphisms L → P is open and dense in the topological B-module Hom(L, P ).
This applies in particular to the affinoid algebras Ai. We get that Ei is open and dense in HomAi (Fi, E(Sp Ai)) for each i. By the general Mittag-Leffler Theorem [Bou66, II.3.5, Th´eor`eme1], E is dense in HomA(F, E(X)) with respect to the limit topologies. Hence, E is nonempty and E(X) is finitely generated by global sections. S Theorem 2.12 (rigid analytic Serre–Swan). Let X = i(Sp(Ai)) be a quasi-Stein variety of bounded dimension, i. e., supx∈X dim OX,x < ∞. Set A := OX (X). Then the global sections functor
fin,proj Γ: VBX −→ A-Mod from vector bundles over X to finitely generated projective A-modules is an equivalence of categories.
Proof. Let us verify the assumptions of Theorem 2.10 (general Serre–Swan). • A sequence of coherent sheaves
0 → E → F → G → 0
on a rigid analytic variety X is exact if and only if for each x ∈ X the sequence of OX,x-modules
0 → Ex → Fx → Gx → 0 is exact [FvdP04, p. 192].
• The category of coherent OX -modules on X is abelian [FvdP04, p. 87] and an
admissible subcategory of ModOX [Kie67, Satz 2.4]. • Vector bundles are finitely generated by global sections. This is true for affinoid varieties by Kiehl [Kie67, Satz 2.2] and for general quasi-Stein varieties of bounded dimension by Lemma 2.11. Consequently the Serre–Swan Theorem holds for rigid analytic quasi-Stein varieties.
Remark 2.13. If X = Sp A is affinoid itself, this follows directly from Kiehl [Kie67], as pointed out by Gruson [Gru68] and L¨utkebohmert [L¨ut77].If
[ n [ −1 −1 X = Bs = khη T1, . . . , η Tni s alg fin,proj Γ: VBX −→ A-Mod from Zariski vector bundles over X to finitely generated projective A-modules is an equivalence of categories. Proof. Again we need to verify the assumptions of Theorem 2.10 (general Serre– Swan). • A sequence of coherent sheaves 0 → E → F → G → 0 on an affinoid variey X is exact if and only if for each x ∈ X the sequence of the Zariski stalks Zar Zar Zar 0 → Ex → Fx → Gx → 0 is exact [sta, Tag 00EN, Lemma 10.23.1(3)]. Zar • As (X, Zar, OX ) is a locally ringed space (not only locally G-ringed), the category of coherent Zariski OX -modules on X is abelian by the classical arguments [sta, Tag 01BU, Lemma 17.12.4]. It is an admissible subcategory of ModOX : C1 Internal Hom between coherent sheaves is coherent [sta, Tag 01CM, Lemma 17.20.5]. C2 Every coherent sheaf is acyclic by a theorem by Grothendieck (see [Har77, Theorem 2.7]) and generated by global sections by noetherianity (cf. the argument in the next point below). C3 is clear. • Vector bundles are finitely generated by global sections. Let E be a vector bundle of rank n on X and dim X = d. As A is noetherian, E has a finite local trivialisation of the form {Sp(Afi )}i=1,...,N with fi ∈ A for all i. That is, {Sp(Afi )}i=1,...,N is a finite covering of X by affinoid domains Xi := Sp(Afi ) such that for all i the restriction E(Xi) is trivial. In other words, ∼ n E(Xi) = (Afi ) . Let fi1 , . . . , fin ∈ Afi be generators of E(Xi) as an Afi -module. Now if some fij is not in the image of the localisation A ⊂ Afi , this means that fij has s1 sn a finite power sj of fi in the denominator. The elements fi1 fi , . . . , fin fi 2.2. SERRE–SWAN FOR RIGID ANALYTIC QUASI-STEIN VARIETIES 35 are now in the image of the localisation A ⊂ Afi . As fi is a unit in Afi , the s1 sn fi1 fi , . . . , fin fi still generate E(Xi) as an Afi -module. They lift to global ˜ ˜ sections fi1 ,..., fin ∈ A and hence E is generated by the global sections ˜ ˜ ˜ ˜ f11 ,..., f1n ,..., fN1 ,..., fNn . By induction over the dimension we can even show that N can be chosen less than or equal to d + 1. Let E be a Zariski vector bundle on X. We need to show that E is finitely generated by global sections. As A is noetherian, there is a local trivialisation of E by affinoid Zariski-open subspaces Xi = Sp(Afi ), i = 1, . . . , r with fi ∈ O(X) for i ∈ {1, . . . , r}. By Kiehl’s Theorem A, that is, Theorem 1.32c) in this thesis, every stalk Ex is generated by global sections as an OX,x-module. Set n := dim X. For each x ∈ X, choose global sections fx1, . . . , fxn whose images generate Ex as an OX,x-module. For each i ∈ {1, . . . , r}, the images of the global sections from Si := {fxj ∈ G(X) | x ∈ Xi, j = 1, . . . , n} in G(Xi) generate G(Xi) as an OX (Xi)-module. As G(Xi) is a free OX (Xi)-module of dimension n, already n of those sections are sufficient. Choose fi1, . . . , fin ∈ Si such that their images in G(Xi) generate it as an OX (Xi)-module. In particular, for any x ∈ Xi, the images of the fij in Gx generate it as an OX,x-module. The set S := {fij ∈ G(X) | i = 1, . . . , r; j = 1, . . . , n} of global sections of G(X) is finite. The images of its elements generate each stalk of G. Corollary 2.15. The categories of analytic vector bundles and of Zariski vector bundles on an affinoid variety are equivalent. Recall from Definition 1.29 that a rigid analytic variety is quasicompact if it can be covered by finitely many affinoids. Proposition 2.16. Every vector bundle on a quasicompact rigid analytic variety has a finite local trivialisation. Proof. Let X be a quasicompact variety and let E be a vector bundle on X. Assume that X is affinoid. Then E is isomorphic to a Zariski vector bundle E 0 on X. It has a finite local trivialisation because X is noetherian. If X is not affinoid, it has a finite admissible covering {U1,...,Ur} by affinoid subsets. The restrictions of E to those affinoid subsets each have a finite local trivialisation. Putting them together yields a finite local trivialisation of E. The same holds for quasi-Stein varieties: Theorem 2.17 (Ben’s Theorem). Every vector bundle on a rigid analytic quasi-Stein variety has a finite local trivialisation. 36 CHAPTER 2. VECTOR BUNDLES OVER RIGID ANALYTIC VARIETIES The author asked for a proof or counterexample on mathoverflow.net (question 219140). The following proof was given by the user Ben. He proved that, over a locally ringed space, any vector bundle which is finitely generated by global sections admits a finite local trivialisation. This easily carries over to a rigid analytic variety with the G-topology. Proof. Let X be a quasi-Stein variety and E be a vector bundle of rank r on X. By Lemma 2.11, it is generated by finitely many global sections s1, . . . , sd. Let XB be the Berkovich space associated to X. As X is quasi-Stein, every x ∈ XB has an affinoid neighbourhood (Example 2.6). By Proposition 2.5, the category of vector bundles with respect to the Berkovich topology on XB and the category of analytic vector bundles on X are equivalent. The proof is done by explicit construction. For 1 ≤ j ≤ r, let e be the section of Or that is given by the constant function j XB (0,..., 0, 1, 0,..., 0) where the 1 is located in the j-th entry. Clearly, e1, . . . , er are generators of Or . For each XB I = {i1, . . . , ir} ⊂ {1, . . . , d} define the morphism ϕ : Or −→ E I XB ej 7−→ sij . Then the subsets UI = {x ∈ XB | (ϕI )x → Ex is an iso } form a finite trivialisation of E. By definition of UI , the map ϕ | : Or (U ) → E(U ) I UI XB I I induces isomorphisms on stalks. As E is coherent, it has to be an isomorphism. We still need to check that {UI | I = {i1, . . . , ir} ⊂ {1, . . . , d}} is a covering of XB. Let x ∈ XB. The stalk Ex is generated by the residue classes of s1, . . . , sd. As it is of dimension r, already r of them suffice. Thus, there is an I such that x ∈ UI . Hence, all the UI together cover XB. Each UI is open because the set of x ∈ XB for which the residue classes of si1 , . . . , sir in Ex satisfy a nontrivial relation is closed. We have shown that every vector bundle over the Berkovich space associated to a quasi-Stein rigid analytic variety has a finite local trivialisation with respect to the Berkovich topology. We need a finite local trivialisation with respect to the G-topology. Let X1 ⊂ X2 ⊂ · · · be a strictly affinoid covering of XB. Let V ∈ Cov(XB) be an admissible G-covering which is a local trivialisation of E such that for any V ⊂ V there exists an I with V ⊂ UI . Let W ⊂ V be a subcovering whose restriction to each Xi is finite. This exists because the Xi are affinoid and hence compact. Set ˜ [ UI := V. V ∈W, V ⊂UI 2.3. LINE BUNDLES 37 ˜ Now {UI | I = (i1, . . . , ir) ⊂ {1, . . . , d}} is a finite admissible G-covering of X which trivialises E. It gives rise to a finite local trivialisation of the original vector bundle over the rigid analytic space X. 2.3 Line bundles Definition 2.18 (Line bundle). Let X be a rigid analytic variety. A locally free OX -module of rank one is called a line bundle. As usual, line bundles form a group, the Picard group. Proposition/Definition 2.19 (Picard group, cf. [Ger77, 1.1, 1.2]). Let X be a rigid analytic variety. The set of isomorphism classes of line bundles on X forms a group under ⊗. It is called the Picard group and denoted by Pic(X). For an affinoid algebra A we write Pic(A) := Pic(Sp(A)) for the group of isomorphism classes of invertible A-modules of rank one. Proof. Compare [Ger70]. Let L, L0 be line bundles on X. Then their tensor product 0 0 L ⊗ L is again a line bundle on X. Let {Ui}i∈I and {Uj}j∈J be local trivialisations 0 0 of L and L , respectively. Then a common refinement {Ui ∩ Uj | i ∈ I, j ∈ J} yields a local trivialisation of L ⊗ L0. Hence, L ⊗ L0 is a line bundle on X. Let L be a line −1 bundle on X. Then the sheaf L := Hom(L, OX ) is a line bundle on X and ∼ ∼ Hom(L, OX ) ⊗ L = Hom(L, L) = OX . ∼ Obviously, L ⊗ OX = L. The Picard group of an affinoid variety is strongly linked to the variety’s canonical formal model and canonical reduction: Theorem 2.20 (Gerritzen [Ger77, §4]). Let k be a complete discretely, nontrivially valued field. Let A be a reduced k-affinoid algebra whose ring of integers A◦ is regular. Assume kAk = |k|. Then there is an isomorphism Pic(A) ∼= Pic(A˜). More generally, Urs Hartl and Werner L¨utkebohmert proved the following. Theorem 2.21 (Hartl–L¨utkebohmert [HL00]). Let k be a complete discretely, non- trivially valued field. Let X be a smooth, proper, connected rigid analytic k-variety that admits a strict semistable formal model over k◦. Assume furthermore that there exists a k-rational point x: Sp k → X. Denote by smRig the category of smooth, quasicompact, separated rigid analytic k-varieties. Then the Picard functor smRig → Sets, n o ∼ ∗ ∼ V 7→ (L, λ) L line bundle over X × V, λ: OV → (x, id) L iso / = is representable. 38 CHAPTER 2. VECTOR BUNDLES OVER RIGID ANALYTIC VARIETIES 2.4 Divisors Definition 2.22 (Divisor in a ring [Eis95, p. 262]). Let R be a ring. A divisor in R is an element of the free abelian group Div(R) generated by those prime ideals of R that are of height one. Example 2.23. If R is of dimension one, then Div(R) = Max(R)Z. In other words, a divisor on a curve X is a formal sum of closed points of X. Definition 2.24 (Divisor class group of a ring). The divisor of a principal ideal is called a principal divisor, cf. [Bou06, p. 196]. The set of principal divisors of a ring R is denoted by P (R). The divisor class group of a ring R is the quotient Cl(R) := Div(R)/P (R). Definition 2.25 (Divisor class group of an affinoid variety). For an affinoid variety X = Sp(A) we denote Div(X) := Div(A) and Cl(X) := Cl(A). S Proposition 2.26 ([Gru68, p. 87, Remarque 2]). If X = s The corresponding statement for closed discs follows from Bourbaki [Bou06], cf. [Gru68, Remarque 2 on p. 87]. We name a few results on the class group. Definition 2.27 (spherically complete, see [Laz62, D´efinition 5.2] or [BGR84, 2.4.4 Definition 1]). A complete valued field k is called spherically complete (also: max- imally complete) if it satisfies the embedded discs principle: For r ∈ R and x ∈ k denote the disc of radius r with centre x by Br(x) ⊂ k. Now look at sequences of discs Br (xn) such that for all n ∈ we have Br (xn+1) ⊂ Br (xn). The n n∈N N n+1 n embedded discs principle is satisfied if each such sequence of discs has a common point: T B (x ) 6= ∅. n∈N rn n Examples 2.28. a) Complete discretely valued fields are spherically complete [Sch02, Lemma 1.6]. b) The field Cp, defined as the completion of the algebraic closure of Qp, is not spherically complete [Sch02, p. 4f]. S Theorem 2.29 (Lazard [Laz62, Th´eor`eme2 in §7]). Let X = s Proposition 2.30 (Denneberg [Den69, Korollar 1]). Let X = Sp A be an affinoid curve over a complete, algebraically closed, nonarchimedean field k, i. e., a k-affinoid variety of dimension one. Let A be reduced. If the reduction X˜ is nonsingular and (f) ∈ Div(X˜) is a principal divisor, then every preimage of (f) under the reduction is a principal divisor in Div(X). Corollary 2.31 (Denneberg [Den69, Korollar 2]). Let X be an affinoid curve over a complete, algebraically closed, nonarchimedean field k. If its reduction X˜ is irreducible, ˜ then the natural epimorphism Div(X) Div(X) induces an epimorphism of the class groups ˜ ϕ: Cl(X) Cl(X). If additionally X˜ is nonsingular, then ϕ is an isomorphism. In the context of the class group, van der Put proved among other things: Theorem 2.32 (Van der Put [vdP80, Theorem 2.1]). Let k be algebraically closed and X = Sp A a connected affinoid curve. Then A is a unique factorisation domain 1 if and only if X is an affinoid subspace of Pk. Furthermore he asserts that for a normal, connected, affinoid curve X over a complete, algebraically closed, nonarchimedean field k, 1 ∗ ∼ H (X, OX ) = Div(X)/ P(X). This assertion is used in the proof of Theorem 2.33 (Van der Put). Let X be a normal, connected, affinoid curve over a complete, algebraically closed, nonarchimedean field k. a) There exists a complete, nonsingular algebraic curve C over k such that X is an affinoid subspace of the analytification of C. b) Such an embedding into an algebraic curve C fixed, there is an isomorphism 1 ∗ ∼ H (X, OX ) = J(C)/H where J(C) denotes the Jacobian of C and H is the subgroup consisting of the images of those divisors of degree zero on C whose support lies in C \ X. Proof. a) This is [vdP80, Theorem 1.1]. b) This is [vdP80, Proposition 3.1]. Van der Put used this theorem to compute the Picard groups of some affinoid curves [vdP80, pp. 163f]. 40 CHAPTER 2. VECTOR BUNDLES OVER RIGID ANALYTIC VARIETIES Chapter 3 Homotopy Invariance of Vector Bundles 3.1 Serre’s problem and the Bass–Quillen conjec- ture In [Ser55], Serre asked whether all finitely generated projective modules over a polynomial ring over a field are free. The question was answered affirmatively and independently in 1976 by Quillen [Qui76] and by Suslin [Sus76]. The theorem is now known as the Quillen–Suslin Theorem. We warmly recommend Lam’s wonderful book [Lam06] about Serre’s problem. The analogue of the Quillen–Suslin Theorem in rigid analytic geometry is also true: Theorem 3.1 (L¨utkebohmert, Kedlaya). Every finitely generated projective khT1,...,Tni-module is free. Proof. L¨utkebohmert [L¨ut77, Theorem 1] uses Weierstraß preparation and GAGA to reduce the proof to the original Quillen–Suslin Theorem. Kedlaya [Ked04, Proposition 6.6] proves a more general version of the theorem. He uses Weierstraß preparation to reduce the proof to classical statements about unimodular vectors and finite free resolution of modules. Definition 3.2 (extended module). Let R be a ring. An R[T ]-module M is called extended from R if there exists an R-module N such that ∼ M = N ⊗R R[T ]. When Serre’s problem was still open, Bass generalised the question to the following: Question 3.3 (Bass [Bas73, Problem IX on page 21]). Let R be a commutative regular ring. Is every finitely generated projective R[T ]-module extended from R? For his proof of Serre’s problem [Qui76], Quillen developed a method to patch finitely presented modules. With Quillen patching at hand, Quillen reformulated 41 42 CHAPTER 3. HOMOTOPY INVARIANCE OF VECTOR BUNDLES Bass’ question: For an affirmative answer to question 3.3, it is enough to prove the statement for a commutative regular local ring R. The problem became known as the Bass–Quillen Conjecture. Hartmut Lindel showed in [Lin82] that the Bass–Quillen Conjecture is true if R is a regular algebra over a field K such that R is essentially of finite type over K. Here, essentially of finite type means that R is a localisation of a K-algebra R0 which is of finite type over K. Lindel’s theorem was generalised by Dorin Popescu [Pop89, Corollary 4.4] and others, e. g. Sankar Prasad Dutta [Dut95, Theorem 3.4]. For the most general form of the theorem, see [AHW15a, Theorem 5.2.1]. Theorem 3.4 (Lindel–Popescu, [Lin82, Pop89]). The Bass–Quillen Conjecture is true if R is a regular local ring containing a field. We are concerned with the corresponding question in the rigid analytic world: Question 3.5. Let A be a smooth affinoid k-algebra and M a finitely generated projective AhT i-module. Is M extended from A? That is, does there exist a finitely generated projective A-module N such that ∼ M = N⊗ˆ AAhT i? One motivation to be interested in the Bass–Quillen conjecture is its geometric meaning. By the Serre–Swan theorem, the Bass–Quillen conjecture is a conjecture about vector bundles. 3.2 Homotopy invariance By the Serre–Swan Theorem 2.12, Question 3.5 becomes: Question 3.6. Let X = Sp A be a smooth affinoid k-variety, B1 = Sp khT i the unit ball and 1 pr1 : X × B −→ X the projection onto the first factor. Is every vector bundle E over X × B1 the pullback along pr1 of some vector bundle F over X? This immediately becomes a subquestion of the following question: Question 3.7. Let X be a smooth rigid analytic k-variety and 1 pr1 : X × B −→ X the projection onto the first factor. Is every vector bundle E over X × B1 the pullback along pr1 of some vector bundle F over X? In the sequel we will refer to this as B1-invariance. We will see that the answer is no in general, for example if X = P1. So we will have to refine the question. The corresponding questions in algebraic geometry and in complex analysis are well understood. In algebraic geometry, affineness is a necessary condition for homotopy 3.2. HOMOTOPY INVARIANCE 43 invariance of vector bundles of rank ≥ 2. In complex analysis, the space X has to be Stein. In both cases, the projective line P1 provides a counterexample for vector bundles of rank at least two. Examples using Grothendieck’s classification of line bundles on P1 and Serre’s GAGA are well known. We will see one of those examples in 3.10, adapted to the rigid analytic situation. A complex Stein space is a complex space for which Cartan’s Theorem B holds, see also Definition 5.4. For a detailed exposition on complex Stein spaces we recommend Grauert–Remmert [GR04]. For now, let us just say that a complex Stein space has many global functions. As a consequence every holomorphic vector bundle over a complex Stein space is entirely determined, up to isomorphism, by its global sections [For64, For67]. In algebraic geometry, the varieties that have many global functions are exactly the affine varieties. By Serre’s Serre–Swan Theorem [Ser58] every vector bundle on an affine variety is entirely determined, up to isomorphism, by its global sections. In our case, rigid analytic quasi-Stein varieties have many global functions. By Theorem 2.12, the isomorphism type of a vector bundle over a quasi-Stein variety is completely determined by its global sections. Thus rigid analytic quasi-Stein varieties should be the right class of spaces to look at. Another obvious generalisation of question 3.6 is 1 Question 3.8. Let X be a smooth rigid analytic quasi-Stein variety, Arig the rigid analytic affine line (see Example 1.24) and 1 pr1 : X × Arig −→ X 1 the projection onto the first factor. Is every vector bundle E over X ×Arig the pullback along pr1 of some vector bundle F over X? 1 In the sequel we will refer to this as Arig-invariance. The first thing to check is whether the Quillen–Suslin Theorem still holds. That is, 1 are all vector bundles over Arig trivial? Proposition 3.9 ([Gru68, V, Proposition 2]). Let Xr be an open polydisc of polyra- dius r ∈ (|k| ∪ {∞})n, that is [ −1 −1 Xr = Sp(khη1 T1, . . . , ηn Tni). |ηi|=si The Picard group Pic(Xr) of line bundles over Xr is trivial if and only if one of the following holds: a) the field k is spherically complete (see Definition 2.27) or n b) the polyradius is r = (∞,..., ∞), that is, Xr = Arig is the analytic affine space. The “only if” direction was proven by Lazard [Laz62, Proposition 6]: Let k be spherically incomplete and r ∈ |k|. Lazard constructs a divisor on Xr which is not a principal divisor. This implies that open discs Xr which are bounded in at least one direction have nontrivial line bundles. The reason is that the divisor class group and the Picard group coincide on open polydiscs D by Gruson’s Proposition 2.26. 44 CHAPTER 3. HOMOTOPY INVARIANCE OF VECTOR BUNDLES 3.3 Counterexamples Homotopy invariance cannot hold in full generality. In order to stake the limits we assemble examples where homotopy invariance is violated in this section. The examples are: 1 1 • Neither B -invariance nor Arig-invariance hold for vector bundles of rank at 1 least two over Prig (Example 3.10). 1 1 • Neither B -invariance nor Arig-invariance hold for vector bundles over an affinoid variety with a cusp singularity (Example 3.12 by Gerritzen). • B1-invariance can fail for a smooth affinoid variety whose reduction has a cusp singularity (Example 3.14 by Gerritzen or Bosch). 1 1 Example 3.10. We show that neither B - nor Arig-invariance hold for vector bundles 1 1 of rank at least two over Prig. Every counterexample to A -invariance of algebraic vector bundles over algebraic k-varieties works. The example we present here is the analytified version of Asok–Morel [AM11, Example 3.2.9]. 1 1 Introducing coordinates (t, x) on (Prig \{0, ∞}) × Arig, let E be the vector bundle on 1 1 1 1 1 1 Prig × Arig which is free on (Prig \{0}) × Arig and free on (Prig \ {∞}) × Arig and has the transition matrix t2 xt 0 1 1 1 1 on (Prig \{0, ∞}) × Arig. We claim that it is not the pullback of any bundle on Prig. 1 1 Restricting E to x ∈ A yields a bundle Ex on Prig. For x = 0 we get E0, described by t2 0 . 0 1