This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

A new axiomatics for masures II

Auguste Hébert
Université de Lorraine, Institut Élie Cartan de Lorraine, F-54000 Nancy, France
UMR 7502, auguste.hebert@univ-lorraine.fr
Abstract

Masures are generalizations of Bruhat-Tits buildings. They were introduced by Gaussent and Rousseau in order to study Kac-Moody groups over valued fields. We prove that the intersection of two apartments of a masure is convex. Using this, we simplify the axiomatic definition of masures given by Rousseau.

1 Introduction

Bruhat-Tits buildings are an important tool in the study of reductive groups over non-Archimedean local fields. They were introduced by Bruhat and Tits in [BT72] and [BT84]. Kac-Moody groups (à la Tits) are interesting infinite dimensional (if not reductive) generalizations of reductive groups. In order to study them over fields endowed with a discrete valuation, Gaussent and Rousseau introduced in [GR08] some spaces similar to Bruhat-Tits buildings, called masures (also known as hovels), on which these groups act. Charignon and Rousseau generalized this construction in [Cha10], [Rou16] and [Rou17]: Charignon treated the almost split case and Rousseau suppressed restrictions on the base field and on the group. Thanks to these works, a masure is now associated with each almost split Kac-Moody group over a valued field (with some additional assumptions on the field in the non-split case, see [Rou17]). Masures enable to obtain results on the representation theory of almost split Kac-Moody groups over non-Archimedean local fields. For example, Bardy-Panse, Gaussent and Rousseau used them to associate with each such group a spherical and an Iwahori-Hecke algebra (see [GR14] and [BPGR16], these algebras were already defined in the split affine case by Braverman, Kazhdan and Patnaik in [BK11] and [BKP16]).

Let GG be a split reductive group over a valued field (𝒦,ω)(\mathcal{K},\omega), where ω:𝒦{}\omega:\mathcal{K}\rightarrow\mathbb{R}\cup\{\infty\} is a valuation. Let TT be a maximal split torus of GG. Let 𝔸=Y\mathbb{A}=Y\otimes\mathbb{R}, where YY is the cocharacter lattice of (G,T)(G,T). As a set, =G×𝔸/\mathcal{I}=G\times\mathbb{A}/\sim, where \sim is some equivalence relation on G×𝔸G\times\mathbb{A}. The definition of \sim is complicated and based on the notion of parahoric subgroups. However, many of the properties of \mathcal{I} can be recovered from the fact that it satisfies the crucial properties (I1) and (I2) below. Let Φ\Phi denote the root system of (G,T)(G,T) and Λ=ω(𝒦){}\Lambda=\omega(\mathcal{K})\setminus\{\infty\}. For αΦ\alpha\in\Phi and kΛk\in\Lambda, let Hα,k={x𝔸|α(x)=k}H_{\alpha,k}=\{x\in\mathbb{A}|\alpha(x)=-k\}. The elements of the form Hα,kH_{\alpha,k}, for αΦ\alpha\in\Phi and kΛk\in\Lambda are called walls. A half-space of 𝔸\mathbb{A} delimited by some wall is called a half-apartment of 𝔸\mathbb{A}. An apartment of \mathcal{I} (resp. a half-apartment of \mathcal{I}) is a set of the form g.𝔸g.\mathbb{A} (resp. g.Dg.D), for some gGg\in G (resp. and some half-apartment DD of 𝔸\mathbb{A}). We call a set enclosed if it is a finite intersection of half-apartments. Then \mathcal{I} satisfies the following properties:

  • (I1)

    for all x,yx,y\in\mathcal{I}, there exists an apartment containing {x,y}\{x,y\},

  • (I2)

    if AA and AA^{\prime} are two apartments of \mathcal{I}, then AAA\cap A^{\prime} is enclosed in AA and there exists gGg\in G such that g.A=Ag.A=A^{\prime} and such that gg fixes AAA\cap A^{\prime}.

Note that (I1) is a building theoretic translation of the Cartan decomposition of GG.

Let GG be a split Kac-Moody group over a valued field (𝒦,ω)(\mathcal{K},\omega). Then similarly to the reductive case, the masure \mathcal{I} of GG is defined as G×𝔸/G\times\mathbb{A}/\sim, for some equivalence relation \sim on G×𝔸G\times\mathbb{A} (see [GR08] and [Rou16]). As the Cartan decomposition does not hold in GG (unless GG is reductive), property (I1) is not necessarily satisfied by \mathcal{I}. It can be replaced by an axiom involving “chimneys”, which are certain objects at the infinity of \mathcal{I} (see (MA iii) in 2.2.2). This axiom corresponds to the Iwasawa and the Birkhoff decompositions in GG. Gaussent and Rousseau proved weak versions of (I2) in [GR08] and [Rou11]. More precisely, let 𝒯\mathcal{T} be the Tits cone of 𝔸\mathbb{A}. If x,yx,y\in\mathcal{I}, one writes xyx\leq y if there exists gGg\in G such that g.x,g.y𝔸g.x,g.y\in\mathbb{A} and g.yg.x𝒯g.y-g.x\in\mathcal{T}. It is proved in [GR08] that if A,AA,A^{\prime} are two apartments, if x,yAAx,y\in A\cap A^{\prime} are such that xyx\leq y, then [x,y]A[x,y]_{A} (the line segment in AA joining xx to yy) is equal to [x,y]A[x,y]_{A^{\prime}} and there exists gGg\in G such that g.A=Ag.A=A^{\prime} and such that gg fixes [x,y]A[x,y]_{A}. This property is called preordered convexity and is fundamental to most of the applications of the theory of masures so far. However, very few was known about convexity properties for pairs of points which are not preordered. In [Héb20], we proved that when \mathcal{I} is associated with an affine Kac-Moody group, then \mathcal{I} satisfies (I2). In general we proved that (I2) is satisfied for pairs of apartments A,AA,A^{\prime} such that AAA\cap A^{\prime} is “large enough” (more precisely, when AAA\cap A^{\prime} contains a generic ray, see [Héb20, Theorem 4.22]). We used this to simplify the definition of masures. In this paper we prove that (I2) is satisfied without assumption on AAA\cap A^{\prime}. More precisely, let \mathcal{I} be a masure in the sense of [Rou11, Définition 2.1], satisfying some mild technical assumption (see assumption 2.1). Then:

Theorem 1.

(see Theorem 3.6) Let AA and AA^{\prime} be two apartments of \mathcal{I}. Then AAA\cap A^{\prime} is enclosed and there exists an apartment isomorphism ϕ:AA\phi:A\rightarrow A^{\prime} fixing AAA\cap A^{\prime}.

Note that in the theorem above, the half-apartments of 𝔸\mathbb{A} are delimited by real roots of (G,T)(G,T) (and not by imaginary roots as it was the case in [GR08]).

We then use this theorem to give a simplified definition of masures equivalent to [Rou11, Définition 2.1]. This also simplifies the definition given in [Héb20, Theorem 1.5] (see Corollary 3.7).

Framework

Actually we do not limit our study to masures associated with Kac-Moody groups: for us a masure is a set satisfying the axioms of [Rou11] and whose apartments are associated with a root generating system (and thus with a Kac-Moody matrix). We do not assume the existence of a group acting strongly transitively on it. Our results apply to the masures associated with split Kac-Moody groups over valued fields constructed in [GR08] and [Rou16] and to masures associated with almost split Kac-Moody groups over valued fields (satisfying some additional conditions, see [Rou17, 6.1]) in [Rou17]. Contrary to [Rou11], we assume that the family of simple coroots is free in 𝔸\mathbb{A} (see assumption 2.1).

Comments on the proof of Theorem 1

In [Héb20], we proved that if AA and AA^{\prime} are apartments containing a common generic ray, then AAA\cap A^{\prime} is enclosed by following the steps below:

  • (1)

    We prove that if B,BB,B^{\prime} are any two apartments, then BBB\cap B^{\prime} can be written as a finite union of enclosed subsets of BB.

  • (2)

    We prove that AAA\cap A^{\prime} is convex. Using (1), we deduce that it is enclosed.

As we already proved (1) in [Héb20], the main difficulty in the proof of Theorem 1 is to prove the convexity of AAA\cap A^{\prime} (without assumption on AA and AA^{\prime}). In this paper, we prove it directly, without using step (2) of [Héb20]. Our proof of the convexity of AAA\cap A^{\prime} is based on the study of some kind of Hecke paths, that is, on the study of the images of line segments by retractions centered at a sector-germ. Our proof of the convexity is actually simpler than the proof of step (2) in [Héb20] (and more general).

Organization of the paper

In section 2, we describe the general framework and recall the definition of masures.

In section 3, we prove Theorem 1.

Funding

The author was supported by the ANR grant ANR-15-CE40-0012.

Acknowledgment

I would like to thank the referee for his/her valuable comments and suggestions.

2 General framework, Masure

In this section, we define our framework and recall the definition of masures. We give the definition of [Héb20].

2.1 Standard apartment

2.1.1 Root generating system

Let AA be a Kac-Moody matrix (also known as generalized Cartan matrix) i.e a square matrix A=(ai,j)i,jIA=(a_{i,j})_{i,j\in I} with integers coefficients, indexed by a finite set II and satisfying:

  1. 1.

    iI,ai,i=2\forall i\in I,\ a_{i,i}=2

  2. 2.

    (i,j)I2|ij,ai,j0\forall(i,j)\in I^{2}|i\neq j,\ a_{i,j}\leq 0

  3. 3.

    (i,j)I2,ai,j=0aj,i=0\forall(i,j)\in I^{2},\ a_{i,j}=0\Leftrightarrow a_{j,i}=0.

A root generating system of type AA is a 55-tuple 𝒮=(A,X,Y,(αi)iI,(αi)iI)\mathcal{S}=(A,X,Y,(\alpha_{i})_{i\in I},(\alpha_{i}^{\vee})_{i\in I}) made of a Kac-Moody matrix AA indexed by II, of two dual free \mathbb{Z}-modules XX (of characters) and YY (of cocharacters) of finite rank rk(X)\mathrm{rk}(X), a family (αi)iI(\alpha_{i})_{i\in I} (of simple roots) in XX and a family (αi)iI(\alpha_{i}^{\vee})_{i\in I} (of simple coroots) in YY. They have to satisfy the following compatibility condition: ai,j=αj(αi)a_{i,j}=\alpha_{j}(\alpha_{i}^{\vee}) for all i,jIi,j\in I.

Assumption 2.1.

We assume that (αi)iI(\alpha_{i})_{i\in I} is free in 𝔸\mathbb{A}^{*} and that (αi)iI(\alpha_{i}^{\vee})_{i\in I} is free in 𝔸\mathbb{A}.

Let 𝔸=Y\mathbb{A}=Y\otimes\mathbb{R}. Every element of XX induces a linear form on 𝔸\mathbb{A}. We consider XX as a subset of the dual 𝔸\mathbb{A}^{*} of 𝔸\mathbb{A}: the αi\alpha_{i}, iIi\in I are viewed as linear forms on 𝔸\mathbb{A}. For iIi\in I, we define an involution rir_{i} of 𝔸\mathbb{A} by ri(v)=vαi(v)αir_{i}(v)=v-\alpha_{i}(v)\alpha_{i}^{\vee} for all v𝔸v\in\mathbb{A}. Its space of fixed points is kerαi\ker\alpha_{i}. The subgroup of GL(𝔸)\mathrm{GL}(\mathbb{A}) generated by the αi\alpha_{i} for iIi\in I is denoted by WvW^{v} and is called the vectorial Weyl group of 𝒮\mathcal{S}. Then (Wv,{ri|iI})(W^{v},\{r_{i}|i\in I\}) is a Coxeter system.

One defines an action of the group WvW^{v} on 𝔸\mathbb{A}^{*} as follows: if x𝔸x\in\mathbb{A}, wWvw\in W^{v} and α𝔸\alpha\in\mathbb{A}^{*}, then (w.α)(x)=α(w1.x)(w.\alpha)(x)=\alpha(w^{-1}.x). Let Φ={w.αi|(w,i)Wv×I}\Phi=\{w.\alpha_{i}|(w,i)\in W^{v}\times I\} be the set of (real) roots. Then ΦQ\Phi\subset Q, where Q=iIαiQ=\bigoplus_{i\in I}\mathbb{Z}\alpha_{i}. Let Q+=iIαiQ^{+}=\bigoplus_{i\in I}\mathbb{N}\alpha_{i}, Φ+=Q+Φ\Phi^{+}=Q^{+}\cap\Phi and Φ=(Q+)Φ\Phi^{-}=(-Q^{+})\cap\Phi. Then Φ=Φ+Φ\Phi=\Phi^{+}\sqcup\Phi^{-}.

We set Q=iIαiQ^{\vee}_{\mathbb{R}}=\bigoplus_{i\in I}\mathbb{R}\alpha_{i}^{\vee} and Q+=iI+αiQ^{\vee}_{\mathbb{R}_{+}}=\bigoplus_{i\in I}\mathbb{R}_{+}\alpha_{i}^{\vee}. For x,y𝔸x,y\in\mathbb{A}, we write xQyx\leq_{Q^{\vee}_{\mathbb{R}}}y if yxQ+y-x\in Q^{\vee}_{\mathbb{R}_{+}}.

2.1.2 Vectorial faces and Tits cone

Define Cfv={v𝔸|αi(v)>0,iI}C_{f}^{v}=\{v\in\mathbb{A}|\ \alpha_{i}(v)>0,\ \forall i\in I\}. We call it the fundamental vectorial chamber. For JIJ\subset I, one sets Fv(J)={v𝔸|αi(v)=0iJ,αi(v)>0iJ\I}F^{v}(J)=\{v\in\mathbb{A}|\ \alpha_{i}(v)=0\ \forall i\in J,\alpha_{i}(v)>0\ \forall i\in J\backslash I\}. Then the closure Cfv¯\overline{C_{f}^{v}} of CfvC_{f}^{v} is the union of the Fv(J)F^{v}(J) for JIJ\subset I. The positive (resp. negative) vectorial faces are the sets w.Fv(J)w.F^{v}(J) (resp. w.Fv(J)-w.F^{v}(J)) for wWvw\in W^{v} and JIJ\subset I. A vectorial face is either a positive vectorial face or a negative vectorial face. We call positive chamber (resp. negative) every cone of the form w.Cfvw.C_{f}^{v} for some wWvw\in W^{v} (resp. w.Cfv-w.C_{f}^{v}). For all xCfvx\in C_{f}^{v} and for all wWvw\in W^{v}, w.x=xw.x=x implies that w=1w=1. In particular the action of ww on the positive chambers is simply transitive. The Tits cone 𝒯\mathcal{T} is defined by 𝒯=wWvw.Cfv¯\mathcal{T}=\bigcup_{w\in W^{v}}w.\overline{C^{v}_{f}}. One defines a WvW^{v}-invariant preorder \leq on 𝔸\mathbb{A}, the Tits preorder by:

(x,y)𝔸2,xyyx𝒯.\forall(x,y)\in\mathbb{A}\mathrm{}^{2},\ x\leq y\ \Leftrightarrow\ y-x\in\mathcal{T}.

2.1.3 Affine Weyl group of 𝔸\mathbb{A}

We now define the Weyl group WW of 𝔸\mathbb{A}. If 𝒳\mathcal{X} is an affine subspace of 𝔸\mathbb{A}, one denotes by 𝒳\vec{\mathcal{X}} its direction. One equips 𝔸\mathbb{A} with a family \mathcal{M} of affine hyperplanes called real walls such that:

  1. 1.

    For all MM\in\mathcal{M}, there exists αMΦ\alpha_{M}\in\Phi such that M=ker(αM)\vec{M}=\ker(\alpha_{M}).

  2. 2.

    For all αΦ\alpha\in\Phi, there exists an infinite number of hyperplanes MM\in\mathcal{M} such that α=αM\alpha=\alpha_{M}.

  3. 3.

    If MM\in\mathcal{M}, we denote by rMr_{M} the reflection with respect to the hyperplane MM whose associated linear map is rαMr_{\alpha_{M}}. We assume that the group WW generated by the rMr_{M} for MM\in\mathcal{M} stabilizes \mathcal{M}.

The group WW is the Weyl group of 𝔸\mathbb{A}. We assume that 0 is special (i.e we assume that kerα\ker\alpha\in\mathcal{M} for every αΦ\alpha\in\Phi) and thus WWvW\supset W^{v}.

For α𝔸\alpha\in\mathbb{A}^{*} and kk\in\mathbb{R}, set M(α,k)={v𝔸|α(v)+k=0}M(\alpha,k)=\{v\in\mathbb{A}|\alpha(v)+k=0\}. Then for all MM\in\mathcal{M}, there exists αΦ\alpha\in\Phi and kMk_{M}\in\mathbb{R} such that M=M(α,kM)M=M(\alpha,k_{M}). For αΦ\alpha\in\Phi, set Λα={kM|MandM=ker(α)}\Lambda_{\alpha}=\{k_{M}|\ M\in\mathcal{M}\mathrm{\ and\ }\vec{M}=\ker(\alpha)\}. Then Λw.α=Λα\Lambda_{w.\alpha}=\Lambda_{\alpha} for all wWvw\in W^{v} and αΦ\alpha\in\Phi.

If αΦ\alpha\in\Phi, one denotes by Λ~α\tilde{\Lambda}_{\alpha} the subgroup of \mathbb{R} generated by Λα\Lambda_{\alpha}. By (3), Λα=Λα+2Λ~α\Lambda_{\alpha}=\Lambda_{\alpha}+2\tilde{\Lambda}_{\alpha} for all αΦ\alpha\in\Phi. In particular, Λα=Λα\Lambda_{\alpha}=-\Lambda_{\alpha} and when Λα\Lambda_{\alpha} is discrete, Λ~α=Λα\tilde{\Lambda}_{\alpha}=\Lambda_{\alpha} is isomorphic to \mathbb{Z}.

One sets Q=αΦΛ~ααQ^{\vee}=\bigoplus_{\alpha\in\Phi}\tilde{\Lambda}_{\alpha}\alpha^{\vee}. This is a subgroup of 𝔸\mathbb{A} stable under the action of WvW^{v}. Then one has W=WvQW=W^{v}\ltimes Q^{\vee}.

2.1.4 Filters

Definition 2.2.

A filter on a set EE is a nonempty set FF of nonempty subsets of EE such that, for all subsets SS, SS^{\prime} of EE, if SS, SFS^{\prime}\in F then SSFS\cap S^{\prime}\in F and, if SSS^{\prime}\subset S, with SFS^{\prime}\in F, then SFS\in F.

If FF is a filter on a set EE, and EE^{\prime} is a subset of EE, one says that FF contains EE^{\prime} if every element of FF contains EE^{\prime}. If EE^{\prime} is nonempty, the set FEF_{E^{\prime}} of subsets of EE containing EE^{\prime} is a filter. We will sometimes regard EE^{\prime} as a filter on EE by identifying FEF_{E^{\prime}} and EE^{\prime}. If FF is a filter on EE, its closure F¯\overline{F} (resp. its convex hull) is the filter of subsets of EE containing the closure (resp. the convex envelope) of some element of FF. A filter FF is said to be contained in an other filter FF^{\prime}: FFF\subset F^{\prime} (resp. in a subset ZZ in EE: FZF\subset Z) if and only if any set in FF^{\prime} (resp. if ZZ) is in FF.

If x𝔸x\in\mathbb{A} and Ω\Omega is a subset of 𝔸\mathbb{A} containing xx in its closure, then the germ of Ω\Omega in xx is the filter germx(Ω)germ_{x}(\Omega) of subsets of 𝔸\mathbb{A} containing a neighborhood of xx in Ω\Omega.

A sector in 𝔸\mathbb{A} is a set of the form 𝔰=x+Cv\mathfrak{s}=x+C^{v} with Cv=±w.CfvC^{v}=\pm w.C_{f}^{v} for some x𝔸x\in\mathbb{A} and wWvw\in W^{v}. A point uu such that 𝔰=u+Cv\mathfrak{s}=u+C^{v} is called a base point of 𝔰\mathfrak{s} and CvC^{v} is its direction. The intersection of two sectors of the same direction is a sector of the same direction.

The sector-germ of a sector 𝔰=x+Cv\mathfrak{s}=x+C^{v} is the filter 𝔖\mathfrak{S} of subsets of 𝔸\mathbb{A} containing an 𝔸\mathbb{A}-translate of 𝔰\mathfrak{s}. It only depends on the direction CvC^{v}. We denote by ++\infty (resp. -\infty) the sector-germ of CfvC_{f}^{v} (resp. of Cfv-C_{f}^{v}).

A ray δ\delta with base point xx and containing yxy\neq x (or the interval ]x,y]=[x,y]\{x}]x,y]=[x,y]\backslash\{x\} or [x,y][x,y] or the line containing xx and yy) is called preordered if xyx\leq y or yxy\leq x and generic if yx±𝒯̊y-x\in\pm\mathring{\mathcal{T}}, the interior of ±𝒯\pm\mathcal{T}.

For αΦ\alpha\in\Phi, and k{+}k\in\mathbb{R}\cup\{+\infty\}, let D(α,k)={v𝔸|α(v)+k0}D(\alpha,k)=\{v\in\mathbb{A}|\alpha(v)+k\geq 0\} (and D(α,+)=𝔸D(\alpha,+\infty)=\mathbb{A}\mathrm{}) and D(α,k)={v𝔸|α(v)+k>0}D^{\circ}(\alpha,k)=\{v\in\mathbb{A}|\ \alpha(v)+k>0\} (for αΦ\alpha\in\Phi and k{+}k\in\mathbb{R}\cup\{+\infty\}).

Let \mathscr{L} be the set of families (Λα)αΦ(\Lambda^{\prime}_{\alpha})_{\alpha\in\Phi} such that ΛαΛα\Lambda_{\alpha}\subset\Lambda^{\prime}_{\alpha}\subset\mathbb{R} and Λα=Λα\Lambda_{\alpha}^{\prime}=-\Lambda_{-\alpha}^{\prime}, for αΦ\alpha\in\Phi.

An apartment is a root generating system equipped with a Weyl group WW (i.e with a set \mathcal{M} of real walls, see 2.1.3) and a family Λ\Lambda^{\prime}\in\mathscr{L}. Let 𝔸¯=(𝒮,W,Λ)\underline{\mathbb{A}}=(\mathcal{S},W,\Lambda^{\prime}) be an apartment and 𝔸\mathbb{A} be the underlying affine space. A set of the form M(α,k)M(\alpha,k), with αΦ\alpha\in\Phi and kΛαk\in\Lambda^{\prime}_{\alpha} is called a wall of 𝔸\mathbb{A} and a set of the form D(α,k)D(\alpha,k), with αΦ\alpha\in\Phi and kΛαk\in\Lambda^{\prime}_{\alpha} is called a half-apartment of 𝔸\mathbb{A}. A subset XX of 𝔸\mathbb{A} is said to be enclosed if there exist kk\in\mathbb{N}, β1,,βkΦ\beta_{1},\ldots,\beta_{k}\in\Phi and (λ1,,λk)i=1kΛβi(\lambda_{1},\ldots,\lambda_{k})\in\prod_{i=1}^{k}\Lambda^{\prime}_{\beta_{i}} such that X=i=1kD(βi,λi)X=\bigcap_{i=1}^{k}D(\beta_{i},\lambda_{i}).

2.2 Masure

In this section, we define masures.

2.2.1 Definitions of faces, chimneys and related notions

Let 𝔸¯=(𝒮,W,Λ)\underline{\mathbb{A}}=(\mathcal{S},W,\Lambda^{\prime}) be an apartment and 𝔸\mathbb{A} be the underlying affine space.

Let x𝔸x\in\mathbb{A} and FvF^{v} be a vectorial face. The local-face F(x,Fv)=germx(x+Fv)F^{\ell}(x,F^{v})=germ_{x}(x+F^{v}) is the filter defined as the intersection of x+Fvx+F^{v} with the the filter of neighborhoods of xx in 𝔸\mathbb{A}. The face F(x,Fv)F(x,F^{v}) is the filter consisting of the subsets containing a finite intersection of half-spaces D(α,λα)D(\alpha,\lambda_{\alpha}) or D(α,λα)D^{\circ}(\alpha,\lambda_{\alpha}), with λαΛα{+}\lambda_{\alpha}\in\Lambda^{\prime}_{\alpha}\cup\{+\infty\} for all αΦ\alpha\in\Phi (at most one λαΛα\lambda_{\alpha}\in\Lambda_{\alpha} for each αΦ\alpha\in\Phi). We say that FF is positive (or negative) if FvF^{v} is.

Let F,FF,F^{\prime} be two faces. We say that FF^{\prime} dominates FF if FF¯F\subset\overline{F^{\prime}}. The dimension of a face FF is the smallest dimension of an affine space generated by some SFS\in F. Such an affine space is unique and is called its support. A face is said to be spherical if the direction of its support meets the open Tits cone 𝒯̊\mathring{\mathcal{T}}; then its pointwise stabilizer WFW_{F} in WvW^{v} is finite.

A chamber (or alcove) is a face of the form F(x,Cv)F(x,C^{v}) where x𝔸x\in\mathbb{A} and CvC^{v} is a vectorial chamber of 𝔸\mathbb{A}.

A panel is a face which is maximal (for the domination relation) among the faces contained in at least one wall.

Let F=F(x,Fv)F=F(x,F^{v}) be a face (x𝔸x\in\mathbb{A} and FvF^{v} is a vectorial face). The chimney 𝔯(F,Fv)\mathfrak{r}(F,F^{v}) is the filter consisting of the sets containing an enclosed set containing F+FvF+F^{v}. The face FF is the basis of the chimney and the vectorial face FvF^{v} its direction. A chimney is splayed if FvF^{v} is spherical.

A shortening of a chimney 𝔯(F,Fv)\mathfrak{r}(F,F^{v}), with F=F(x,F0v)F=F(x,F_{0}^{v}) is a chimney of the form 𝔯(F(x+ξ,F0v),Fv)\mathfrak{r}(F(x+\xi,F_{0}^{v}),F^{v}) for some ξFv¯\xi\in\overline{F^{v}}. The germ of a chimney 𝔯\mathfrak{r} is the filter of subsets of 𝔸\mathbb{A} containing a shortening of 𝔯\mathfrak{r} (this definition of shortening is slightly different from the one of [Rou11] 1.12 but follows [Rou17] 3.6) and we obtain the same germs with these two definitions).

2.2.2 Masure

An apartment of type 𝔸\mathbb{A} is a set A¯\underline{A} with a nonempty set Isom(𝔸,A)\mathrm{Isom}(\mathbb{A},A) of bijections (called Weyl-isomorphisms) such that if f0Isom(𝔸,A)f_{0}\in\mathrm{Isom}(\mathbb{A},A) then fIsom(𝔸,A)f\in\mathrm{Isom}(\mathbb{A},A) if and only if there exists wWw\in W satisfying f=f0wf=f_{0}\circ w. We will say isomorphism instead of Weyl-isomorphism in the sequel. An isomorphism between two apartments ϕ:AA\phi:A\rightarrow A^{\prime} is a bijection such that (fIsom(𝔸,A)f\in\mathrm{Isom}(\mathbb{A},A) if, and only if, ϕfIsom(𝔸,A)\phi\circ f\in\mathrm{Isom}(\mathbb{A},A^{\prime})). We extend all the notions that are preserved by WW to each apartment. Thus sectors, enclosures, faces and chimneys are well defined in any apartment of type 𝔸\mathbb{A}. If AA and BB are two apartments, and ϕ:AB\phi:A\rightarrow B is an apartment isomorphism fixing some set XX, we write ϕ:A𝑋B\phi:A\overset{X}{\rightarrow}B.

Definition 2.3.

A masure of type 𝔸¯=(𝒮,W,Λ)\underline{\mathbb{A}}=(\mathcal{S},W,\Lambda^{\prime}) is a set \mathcal{I} endowed with a covering 𝒜\mathcal{A} of subsets called apartments such that:

(MA i) Any A𝒜A\in\mathcal{A} is equipped with the structure of an apartment of type 𝔸\mathbb{A}.

(MA ii) : if two apartments A,AA,A^{\prime} contain a generic ray, then AAA\cap A^{\prime} is enclosed and there exists an apartment isomorphism ϕ:AAAA\phi:A\overset{A\cap A^{\prime}}{\rightarrow}A^{\prime}.

(MA iii): if \mathfrak{R} is the germ of a splayed chimney and if FF is a face or a germ of a chimney, then there exists an apartment containing \mathfrak{R} and FF.

In this definition, we say that an apartment contains a germ of a filter if it contains at least one element of this germ. We say that a map fixes a germ if it fixes at least one element of this germ.

The main examples of masures are the masures associated with (almost) split Kac-Moody group over valued fields constructed in [Rou16] and [Rou17].

By [Héb20, Theorem 5.1], definition 2.3 is equivalent to [Rou11, Définition 2.1] (at least under Assumption 2.1).

2.2.3 Example: masure associated with a split Kac-Moody group over a valued field

Let AA be a Kac-Moody matrix and 𝒮\mathcal{S} be a root generating system of type AA. We consider the group functor 𝐆\mathbf{G} associated with the root generating system 𝒮\mathcal{S} in [Tit87] and in [Rém02, Chapitre 8]. This functor is a functor from the category of rings to the category of groups satisfying axioms (KMG1) to (KMG 9) of [Tit87]. When RR is a field, 𝐆(R)\mathbf{G}(R) is uniquely determined by these axioms by Theorem 1’ of [Tit87]. This functor contains a toric functor 𝐓\mathbf{T}, from the category of rings to the category of commutative groups (denoted 𝒯\mathcal{T} in [Rém02]) and two functors 𝐔+\mathbf{U^{+}} and 𝐔\mathbf{U^{-}} from the category of rings to the category of groups.

Let 𝒦\mathcal{K} be a field equipped with a non-trivial valuation ω:𝒦{+}\omega:\mathcal{K}\rightarrow\mathbb{R}\cup\{+\infty\}, 𝒪\mathcal{O} be its ring of integers and G=𝐆(𝒦)G=\mathbf{G}(\mathcal{K}) (and U+=𝐔+(𝒦)U^{+}=\mathbf{U^{+}}(\mathcal{K}), …). For all ϵ{,+}\epsilon\in\{-,+\}, and all αΦϵ\alpha\in\Phi_{\epsilon}, we have an isomorphism xαx_{\alpha} from 𝒦\mathcal{K} to a group UαU_{\alpha}. For all kk\in\mathbb{R}, one defines a subgroup Uα,k:=xα({u𝒦|ω(u)k})U_{\alpha,k}:=x_{\alpha}(\{u\in\mathcal{K}|\ \omega(u)\geq k\}). Let \mathcal{I} be the masure associated with GG constructed in [Rou16]. Then for all αΦ\alpha\in\Phi, Λα=Λα=ω(𝒦)\Lambda_{\alpha}=\Lambda^{\prime}_{\alpha}=\omega(\mathcal{K}^{*}). If moreover ω(𝒦)\omega(\mathcal{K}^{*}) is discrete, one has (up to renormalization) Λα=\Lambda_{\alpha}=\mathbb{Z} for all αΦ\alpha\in\Phi. Moreover, we have:

  • -

    the fixator of 𝔸\mathbb{A} in GG is H=𝐓(𝒪)H=\mathbf{T}(\mathcal{O}) (by remark 3.2 of [GR08])

  • -

    for all αΦ\alpha\in\Phi and kk\in\mathbb{Z}, the fixator of D(α,k)D(\alpha,k) in GG is H.Uα,kH.U_{\alpha,k} (by 4.2 7) of [GR08])

  • -

    for all ϵ{,+}\epsilon\in\{-,+\}, H.UϵH.U^{\epsilon} is the fixator of ϵ\epsilon\infty (by 4.2 4) of [GR08])

  • -

    when moreover the residue field of (𝒦,ω)(\mathcal{K},\omega) contains \mathbb{C}, the fixator of {0}\{0\} in GG is Ks=𝐆(𝒪)K_{s}=\mathbf{G}(\mathcal{O}) (by example 3.14 of [GR08]).

If moreover, 𝒦\mathcal{K} is local, with residue cardinal qq, each panel is contained in 1+q1+q chambers.

The group GG is reductive if and only if WvW^{v} is finite. In this case, \mathcal{I} is the usual Bruhat-Tits building of GG and one has 𝒯=𝔸\mathcal{T}=\mathbb{A}.

2.2.4 Tits preorder on \mathcal{I}

As the Tits preorder \leq on 𝔸\mathbb{A} is invariant under the action of WvW^{v}, one can equip each apartment AA with A\leq_{A}. Let AA be an apartment of \mathcal{I} and x,yAx,y\in A be such that xAyx\leq_{A}y. Then by [Rou11, Proposition 5.4], if BB is an apartment containing xx and yy, [x,y]A=[x,y]B[x,y]_{A}=[x,y]_{B} and there exists an apartment isomorphism ψ:A[x,y]B\psi:A\overset{[x,y]}{\rightarrow}B. In particular, xByx\leq_{B}y. This defines a relation \leq on \mathcal{I}. By Théorème 5.9 of [Rou11], \leq is a preorder on \mathcal{I}. It is invariant by apartment isomorphisms: if A,BA,B are apartments, ϕ:AB\phi:A\rightarrow B is an apartment isomorphism and x,yAx,y\in A are such that xyx\leq y, then ϕ(x)ϕ(y)\phi(x)\leq\phi(y). We call it the Tits preorder on \mathcal{I}.

2.2.5 Retractions centered at sector-germs

Let 𝔰\mathfrak{s} be a sector-germ of \mathcal{I} and AA be an apartment containing it. Let xx\in\mathcal{I}. By (MA iii), there exists an apartment AxA_{x} of \mathcal{I} containing xx and 𝔰\mathfrak{s}. By (MA ii), there exists an apartment isomorphism ϕ:AxA\phi:A_{x}\rightarrow A fixing 𝔰\mathfrak{s}. By [Rou11, 2.6], ϕ(x)\phi(x) does not depend on the choices we made and thus we can set ρA,𝔰(x)=ϕ(x)\rho_{A,\mathfrak{s}}(x)=\phi(x).

The map ρA,𝔰:A\rho_{A,\mathfrak{s}}:\mathcal{I}\rightarrow A is the retraction onto AA centered at 𝔰\mathfrak{s}.

We denote by ρ+:𝔸\rho_{+\infty}:\mathcal{I}\twoheadrightarrow\mathbb{A} (resp. ρ\rho_{-\infty}) the retraction onto 𝔸\mathbb{A} centered at ++\infty (resp. -\infty).

3 Intersection of two apartments in a masure

Let A,BA,B be two apartments of \mathcal{I}. We prove below that ABA\cap B is enclosed and that there exists ϕ:AABB\phi:A\overset{A\cap B}{\rightarrow}B (see Theorem 3.6). Let us sketch our proof. We assume that 𝔸=B\mathbb{A}=B. By results of [Héb20], the main difficulty is to prove that A𝔸A\cap\mathbb{A} is convex. We first assume that A𝔸A\cap\mathbb{A} has nonempty interior. Then using [Héb21, Proposition 9] or [Héb18, Proposition 4.2.8], we write A=i=1kPiA=\bigcup_{i=1}^{k}P_{i}, where kk\in\mathbb{N}, the PiP_{i} are enclosed and if i1,ki\in\llbracket 1,k\rrbracket, there exists an apartment AiA_{i} containing Pi,P_{i},-\infty and an apartment isomorphism APiAiA\overset{P_{i}}{\rightarrow}A_{i} (note that in [Héb21], the masure \mathcal{I} is moreover assumed to be semi-discrete, which means that Λα=\Lambda^{\prime}_{\alpha}=\mathbb{Z} for all αΦ\alpha\in\Phi, but this assumption is easily dropped for the mentioned results, see [Héb18]). Let \mathcal{H} be a finite set of walls delimiting the PiP_{i}. If a,bAa,b\in A we denote by τa,b:[0,1]A\tau_{a,b}:[0,1]\rightarrow A the affine parametrization of [a,b]A[a,b]_{A} such that τa,b(0)=a\tau_{a,b}(0)=a and τa,b(1)=b\tau_{a,b}(1)=b. Let a,bAa,b\in A be such that at each time tt such that τ(t)\tau(t) is in some wall of \mathcal{H}, then this wall is unique (as we shall see, almost every pair (a,b)A2(a,b)\in A^{2} satisfies this property). We prove that π=ρτa,b\pi=\rho_{-\infty}\circ\tau_{a,b} is a piecewise linear path whose left-hand and right-hand derivatives satisfy some growth property with respect to Q\leq_{Q^{\vee}_{\mathbb{R}}} (see Lemma 3.4). We deduce that if bA𝔸b\in A\cap\mathbb{A}, then for almost all a𝔸Aa\in\mathbb{A}\cap A, π+(0)Qba\pi^{\prime}_{+}(0)\leq_{Q^{\vee}_{\mathbb{R}}}b-a. Applying the analogous inequality to ρ+τ\rho_{+\infty}\circ\tau, we deduce that ρτ=ρ+τ\rho_{-\infty}\circ\tau=\rho_{+\infty}\circ\tau. We then deduce that τ𝔸\tau\subset\mathbb{A} and we conclude by using a density argument.

We equip every apartment with the topology defined by its structure of a finite dimensional real-affine space. If EE is a subset of an apartment AA, we denote by Int(E)\mathrm{Int}(E) or by E̊\mathring{E} its interior, depending on the context. This does not depend on the choice of an apartment containing EE by [Héb20, Proposition 3.26]. Let AA be an apartment. For a,bAa,b\in A, aba\neq b, we denote by (a,b)=A(a,b)\mathcal{L}(a,b)=\mathcal{L}_{A}(a,b) the line containing aa and bb in AA. Let x,yx,y\in\mathcal{I} be such that xyx\leq y and AA be an apartment containing xx and yy. We write x<̊yx\mathring{<}y if there exists an apartment isomorphism ϕ:A𝔸\phi:A\rightarrow\mathbb{A} such that ϕ(y)ϕ(x)𝒯̊\phi(y)-\phi(x)\in\mathring{\mathcal{T}}. This does not depend on the choices of AA and ϕ\phi.

Lemma 3.1.

Let xx\in\mathcal{I} and A,BA,B be two apartments containing xx. We assume that there exists a neighborhood VV of xx in AA and a half-apartment DD of AA such that ABDVA\cap B\supset D\cap V. Then:

  1. 1.

    Either xInt(AB)x\in\mathrm{Int}(A\cap B) or there exists a neighborhood V~\tilde{V} of xx in AA such that ABV~=DV~A\cap B\cap\tilde{V}=D\cap\tilde{V}.

  2. 2.

    If xInt(AB)x\notin\mathrm{Int}(A\cap B), there exists an apartment BB^{\prime} such that BBB\cap B^{\prime} is a half-apartment and such that xInt(AB)x\in\mathrm{Int}(A\cap B^{\prime}).

Proof.

Let MM be the wall of AA delimiting DD and PP be a positive panel of MM based at xx. Then PABP\subset A\cap B. Let CC be the chamber of AA dominating PP and not contained in DD. We can assume, reducing VV if necessary, that VV is convex and open. By [Héb20, Proposition 3.26], there exists ϕ:ADVB\phi:A\overset{D\cap V}{\rightarrow}B. Let DB=ϕ(D)D_{B}=\phi(D). We have

ϕ(DV)=DVϕ(D)V=DBV.\phi(D\cap V)=D\cap V\subset\phi(D)\cap V=D_{B}\cap V. (1)

Let BB^{\prime} be an apartment containing DBD_{B} and CC (the existence of such an apartment is provided by [Rou11, Proposition 2.9 1)]). Then BAB^{\prime}\cap A contains DBVD_{B}\cap V. Let QQ be the sector of AA based at xx and containing CC. Reducing VV if necessary, we may assume that VQBV\cap Q\subset B^{\prime}. Let yVQy\in V\cap Q. Then y>̊xy\mathring{>}x. Let V1VV_{1}\subset V be a neighborhood of xx in AA such that y>̊V1y\mathring{>}V_{1}. Let z(x,y)Int(DV)z\in\mathcal{L}(x,y)\cap\mathrm{Int}(D\cap V). Let V2V1V_{2}\subset V_{1} be a neighborhood of zz in AA such that V2DVV_{2}\subset D\cap V. Then V3:=vV2[y,v]AV_{3}:=\bigcup_{v\in V_{2}}[y,v]_{A} contains xx in its interior. By [Rou11, Proposition 5.4],

V3AB.V_{3}\subset A\cap B^{\prime}. (2)

Suppose BCB\supset C. By applying the result above with B=BB^{\prime}=B we deduce that xInt(AB)x\in\mathrm{Int}(A\cap B). Consequently ABA\cap B contains CC if and only if xInt(AB)x\in\mathrm{Int}(A\cap B). Suppose now xInt(AB)x\notin\mathrm{Int}(A\cap B). Then BCB\nsupseteq C. Then BBB^{\prime}\cap B contains a half-apartment and thus it contains a generic ray. Thus we have:

  • BBB\cap B^{\prime} is enclosed in BB (by (MA ii)),

  • BBCB\cap B^{\prime}\nsupseteq C,

  • BBDBB\cap B^{\prime}\supset D_{B}.

Therefore BB=DBB\cap B^{\prime}=D_{B}.

Let V~V3\tilde{V}\subset V_{3} be a convex neighborhood of xx in AA. Let us prove the equality

DV~=DBV~.D\cap\tilde{V}=D_{B}\cap\tilde{V}. (3)

By [Héb20, Proposition 3.26], there exists an apartment isomorphism f:BV~Af:B^{\prime}\overset{\tilde{V}}{\rightarrow}A. By (MA ii), there exists ψ:BDBB\psi:B^{\prime}\overset{D_{B}}{\rightarrow}B. By (1), ϕ1ψ:BA\phi^{-1}\circ\psi:B^{\prime}\rightarrow A fixes pointwise DVDV~D\cap V\supset D\cap\tilde{V}. As DV~D\cap\tilde{V} has nonempty interior, we have f=ϕ1ψf=\phi^{-1}\circ\psi. Let now yDBV~y^{\prime}\in D_{B}\cap\tilde{V}. Then

f(y)=y=ϕ1(ψ(y))=ϕ1(y)f(y^{\prime})=y^{\prime}=\phi^{-1}\left(\psi(y^{\prime})\right)=\phi^{-1}(y^{\prime})

and thus ϕ1\phi^{-1} fixes DBV~D_{B}\cap\tilde{V} pointwise. Since V~V\tilde{V}\subset V, ϕ\phi also fixes pointwise DV~D\cap\tilde{V}. Similarly to (1), we deduce that DBV~DV~D_{B}\cap\tilde{V}\subset D\cap\tilde{V} and DBV~DV~D_{B}\cap\tilde{V}\supset D\cap\tilde{V}, which proves (3).

By (2), we have V~AB\tilde{V}\subset A\cap B^{\prime}. Therefore:

ABV~\displaystyle A\cap B\cap\tilde{V} =ABBV~\displaystyle=A\cap B^{\prime}\cap B\cap\tilde{V}
=ADBV~\displaystyle=A\cap D_{B}\cap\tilde{V}
=ADV~ (by (3))\displaystyle=A\cap D\cap\tilde{V}\text{ (by }\eqref{eqSecond_equation})
=DV~,\displaystyle=D\cap\tilde{V},

which completes the proof of the lemma. ∎

An element rWvr\in W^{v} is called a reflection if r=wriw1r=wr_{i}w^{-1}, for some wWvw\in W^{v} and iIi\in I.

Let BB be an apartment containing -\infty. We denote by ρ,B\rho_{-\infty,B} the retraction onto BB centered at -\infty. If x,yBx,y\in B we denote xQ,Byx\leq_{Q^{\vee}_{\mathbb{R},B}}y if ρ(x)Qρ(y)\rho_{-\infty}(x)\leq_{Q^{\vee}_{\mathbb{R}}}\rho_{-\infty}(y).

For tt\in\mathbb{R}, we denote by ]t,t]]t^{-},t] the filter on \mathbb{R} consisting of the sets containing a set of the form [tϵ,t][t-\epsilon,t], for some ϵ>0\epsilon>0.

Lemma 3.2.

Let AA be an apartment and xAx\in A. Let BB be an apartment containing -\infty. We assume that there exists a half-apartment DD of AA and a neighborhood VV of xx in AA such that VAB=VDV\cap A\cap B=V\cap D. Let τ:A\tau:\mathbb{R}\rightarrow A be a non-constant affine map such that τ(0)=x\tau(0)=x and such that τ(]0,0])D\tau(]0^{-},0])\subset D. Let π=ρ,Bτ\pi=\rho_{-\infty,B}\circ\tau. Then π+(0)Qπ(0)\pi^{\prime}_{+}(0)\geq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{-}(0). If moreover π+(0)π(0)\pi^{\prime}_{+}(0)\neq\pi^{\prime}_{-}(0), then there exists a reflection r\vec{r} of WvW^{v} such that π+(0)=r.π(0)\pi^{\prime}_{+}(0)=\vec{r}.\pi^{\prime}_{-}(0).

Proof.

We assume that π+(0)π(0)\pi^{\prime}_{+}(0)\neq\pi^{\prime}_{-}(0), since otherwise there is nothing to prove. Then xInt(AB)x\notin\mathrm{Int}(A\cap B). By Lemma 3.1, there exists an apartment BB^{\prime} such that BBB^{\prime}\cap B is a half-apartment and such that BAB^{\prime}\cap A contains a neighborhood of xx. Maybe reducing VV, we may assume that VV is open and convex, that it is contained in ABA\cap B^{\prime} and that ABV=DVA\cap B\cap V=D\cap V. By [Héb20, Proposition 3.26], there exists ϕ:AB\phi:A\rightarrow B^{\prime} fixing VV. Therefore there is no loss of generality in assuming that A=BA=B^{\prime} and thus that ABA\cap B contains a half-apartment. Then AB=DA\cap B=D (by (MA ii)). We identify BB and 𝔸\mathbb{A}.

Let ψ:AA𝔸𝔸\psi:A\overset{A\cap\mathbb{A}}{\rightarrow}\mathbb{A}. As π(0)π+(0)\pi^{\prime}_{-}(0)\neq\pi^{\prime}_{+}(0), AA does not contain -\infty. Let DAD_{A} be the half-apartment of AA opposite to DD and D𝔸D_{\mathbb{A}} be the half-apartment of 𝔸\mathbb{A} opposite to DD. By [Rou11, Proposition 2.9 2)], A~:=DAD𝔸\tilde{A}:=D_{A}\cup D_{\mathbb{A}} is an apartment. Let ψ~:A~A~𝔸𝔸\tilde{\psi}:\tilde{A}\overset{\tilde{A}\cap\mathbb{A}}{\rightarrow}\mathbb{A}. Then as A~\tilde{A}\supset-\infty, we have ρ(y)=ψ~(y)\rho_{-\infty}(y)=\tilde{\psi}(y) for every yDAy\in D_{A}. Let rr be the reflection of the affine Weyl group WW with respect to the wall of DD. By [Héb21, Lemma 6 2)], ψ(τ(t))=r.ψ~(τ(t))\psi\big{(}\tau(t)\big{)}=r.\tilde{\psi}\big{(}\tau(t)\big{)} for every t+t\in\mathbb{R}_{+}. Therefore π(0)=r.π+(0)\pi^{\prime}_{-}(0)=\vec{r}.\pi^{\prime}_{+}(0), where r\vec{r} is the linear map associated with rr. Let αΦ+\alpha\in\Phi_{+} be such that r=rα\vec{r}=r_{\alpha}. Let D𝔸=α1()\vec{D}_{\mathbb{A}}=\alpha^{-1}(\mathbb{R}_{-}). Then D𝔸\vec{D}_{\mathbb{A}} is parallel to D𝔸D_{\mathbb{A}}. Then as τ(]0,0])D\tau(]0^{-},0])\subset D we have π(0)D𝔸\pi^{\prime}_{-}(0)\in\vec{D}_{\mathbb{A}} and thus α(π(0))<0\alpha\big{(}\pi^{\prime}_{-}(0)\big{)}<0. Consequently π+(0)=π(0)α(π(0))α>Qπ(0)\pi^{\prime}_{+}(0)=\pi^{\prime}_{-}(0)-\alpha\big{(}\pi^{\prime}_{-}(0)\big{)}\alpha^{\vee}>_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{-}(0). ∎

Let AA be an apartment and 𝔮\mathfrak{q} be a sector-germ of \mathcal{I}. Then by [Héb21, Proposition 9] or [Héb18, Proposition 4.2.8], there exist kk\in\mathbb{N} and enclosed subsets P1,,PkP_{1},\ldots,P_{k} such that:

  • A=i=1kPiA=\bigcup_{i=1}^{k}P_{i},

  • for every i1,ki\in\llbracket 1,k\rrbracket, there exists an apartment AiA_{i} containing PiP_{i} and 𝔮\mathfrak{q},

  • for every i1,ki\in\llbracket 1,k\rrbracket, there exists an apartment isomorphism ϕi:APiAi\phi_{i}:A\overset{P_{i}}{\rightarrow}A_{i},

  • P̊i\mathring{P}_{i}\neq\emptyset, for i1,ki\in\llbracket 1,k\rrbracket.

The last condition follows from [Héb20, Lemma 3.10] applied with U=X=AU=X=A.

For i1,ki\in\llbracket 1,k\rrbracket, we write Pi=j=1kiDi,jP_{i}=\bigcap_{j=1}^{k_{i}}D_{i,j}, where kik_{i}\in\mathbb{N} and Di,jD_{i,j} is a half-apartment of AA. We denote by Mi,jM_{i,j} the wall of Di,jD_{i,j}. Then we set (A,𝔮)={Mi,j|i1,k,j1,ki}\mathcal{H}(A,\mathfrak{q})=\{M_{i,j}|i\in\llbracket 1,k\rrbracket,j\in\llbracket 1,k_{i}\rrbracket\}. For bAb\in A, we then set

E(A,𝔮,b)=AM,M(A,𝔮),MMxMM(x,b).E(A,\mathfrak{q},b)=A\setminus\bigcup_{M,M^{\prime}\in\mathcal{H}(A,\mathfrak{q}),M\neq M^{\prime}}\bigcup_{x\in M\cap M^{\prime}}\mathcal{L}(x,b).

Note that (A,𝔮)\mathcal{H}(A,\mathfrak{q}) and E(A,𝔮,b)E(A,\mathfrak{q},b) depend on the choices we made (we can for example artificially increase the number of PiP_{i}). However for our purpose, this dependency will not be important and we will not be specified: for any choice, E(A,𝔮,b)E(A,\mathfrak{q},b) is obtained from AA by removing finitely many hyperplanes (which are in general not walls). It is in particular dense in AA. By definition, we have the following lemma:

Lemma 3.3.

Let AA be an apartment, 𝔮\mathfrak{q} be a sector-germ of \mathcal{I}, bAb\in A and aE(A,𝔮,b)a\in E(A,\mathfrak{q},b). Let τ:[0,1][a,b]A\tau:[0,1]\rightarrow[a,b]_{A} be an affine parametrization of [a,b]A[a,b]_{A}. Then for every t[0,1]t\in[0,1], we have either τ(t)M(A,𝔮)M\tau(t)\notin\bigcup_{M\in\mathcal{H}(A,\mathfrak{q})}M or there exists a unique M(A,𝔮)M\in\mathcal{H}(A,\mathfrak{q}) such that τ(t)M\tau(t)\in M.

A piecewise linear continuous path of 𝔸\mathbb{A} is a continuous path π:[0,1]𝔸\pi:[0,1]\rightarrow\mathbb{A} such that there exist nn\in\mathbb{N} and 0<t1<<tn<10<t_{1}<\ldots<t_{n}<1 such that π|[ti,ti+1]\pi|_{[t_{i},t_{i+1}]} is a line segment, for every i1,n1i\in\llbracket 1,n-1\rrbracket. In [GR08, 6.1], Gaussent and Rousseau study the image π\pi by ρ\rho_{-\infty} of a segment τ:[0,1]\tau:[0,1]\rightarrow\mathcal{I} such that τ(0)τ(1)\tau(0)\leq\tau(1). They prove that this image is a Hecke path (see [GR08, Definition 5.2 and Theorem 6.2]). Roughly speaking, this means that π\pi is a piecewise linear continuous path such that for all tt such that π(t)π+(t)\pi^{\prime}_{-}(t)\neq\pi^{\prime}_{+}(t), π+(t)\pi^{\prime}_{+}(t) “is farther from -\infty” than π(t)\pi^{\prime}_{-}(t). More precisely this means that there exists a sequence ξ1=π(t)\xi_{1}=\pi^{\prime}_{-}(t), ξ2,,ξn=π+(t)\xi_{2},\ldots,\xi_{n}=\pi^{\prime}_{+}(t) such that for all i1,n1i\in\llbracket 1,n-1\rrbracket, ξi+1=wi.ξi\xi_{i+1}=w_{i}.\xi_{i}, where wiw_{i} is a reflection of WvW^{v} with respect to some wall MiM_{i}, such that if DiD_{i} is the half-apartment of 𝔸\mathbb{A} delimited by MiM_{i} and containing -\infty, then ξiDi\xi_{i}\in D_{i}. In the lemma below, we study the image π\pi by ρ\rho_{-\infty} of a segment τ:[0,1]\tau:[0,1]\rightarrow\mathcal{I} (satisfying some technical properties). Here we do not assume that τ(0)τ(1)\tau(0)\leq\tau(1). We prove that π\pi is a piecewise linear continuous path such that for all tt such that π(t)π+(t)\pi^{\prime}_{-}(t)\neq\pi^{\prime}_{+}(t), π+(t)\pi^{\prime}_{+}(t) “is farther from -\infty” than π(t)\pi^{\prime}_{-}(t). Thanks to our technical assumption on τ\tau, we have n=2n=2 at each folding. We deduce an inequality similar to that of [BPGR16, 1.8 Lemma].

Lemma 3.4.

Let AA be an apartment. Let a,bAa,b\in A. We assume that: aE(A,,b)a\in E(A,-\infty,b).

Let τ:[0,1][a,b]A\tau:[0,1]\rightarrow[a,b]_{A} be the affine parametrization such that τ(0)=a\tau(0)=a and τ(1)=b\tau(1)=b. Let π=ρτ\pi=\rho_{-\infty}\circ\tau. Then π\pi is a piecewise linear continuous path such that:

  1. 1.

    for all t[0,1[t\in[0,1[ (resp. t]0,1]t\in]0,1]), π+(t)Wv.π+(0)\pi^{\prime}_{+}(t)\in W^{v}.\pi^{\prime}_{+}(0) (resp. π(t)Wv.π+(0)\pi^{\prime}_{-}(t)\in W^{v}.\pi^{\prime}_{+}(0)),

  2. 2.

    for every t,t[0,1]t,t^{\prime}\in[0,1] such that 0<t<t<10<t<t^{\prime}<1, we have:

    π+(0)Qπ(t)Qπ+(t)Qπ(t)Qπ+(t)Qπ(1).\pi^{\prime}_{+}(0)\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{-}(t)\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{+}(t)\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{-}(t^{\prime})\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{+}(t^{\prime})\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{-}(1).

We have ρ(b)ρ(a)Qπ+(0)\rho_{-\infty}(b)-\rho_{-\infty}(a)\geq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{+}(0) and if π\pi is not a line segment, then this inequality is strict. In particular, if a,b𝔸a,b\in\mathbb{A}, then π+(0)Qba\pi^{\prime}_{+}(0)\leq_{Q^{\vee}_{\mathbb{R}}}b-a.

Proof.

As ρ\rho_{-\infty} is an affine morphism on every PiP_{i}, i1,ki\in\llbracket 1,k\rrbracket, π\pi is a piecewiese linear continuous path. Let nn\in\mathbb{N} and 0<t1<tn<10<t_{1}<\ldots t_{n}<1 be such that π|[ti,ti+1]\pi|_{[t_{i},t_{i+1}]} is a line segment, for every i1,n1i\in\llbracket 1,n-1\rrbracket.

We now prove that for all i1,ni\in\llbracket 1,n\rrbracket, π(ti)Qπ+(ti)\pi^{\prime}_{-}(t_{i})\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{+}(t_{i}). Let i1,ni\in\llbracket 1,n\rrbracket. If π(ti)=π+(ti)\pi^{\prime}_{-}(t_{i})=\pi^{\prime}_{+}(t_{i}), there is nothing to prove. We assume that π(ti)π+(ti)\pi^{\prime}_{-}(t_{i})\neq\pi^{\prime}_{+}(t_{i}). Let PP be an enclosed subset of AA with non-empty interior such that:

  1. 1.

    PP is delimited by walls of :=(A,)\mathcal{H}:=\mathcal{H}(A,-\infty),

  2. 2.

    τ(]ti,ti])P\tau(]t_{i}^{-},t_{i}])\subset P,

  3. 3.

    there exist an apartment BB containing -\infty and PP and an apartment isomorphism ϕ:B𝑃A\phi:B\overset{P}{\rightarrow}A.

As π(ti)π+(ti)\pi^{\prime}_{-}(t_{i})\neq\pi^{\prime}_{+}(t_{i}), there exists MM\in\mathcal{H} such that τ(ti)M\tau(t_{i})\in M and by Lemma 3.3, this wall is unique. Let VV be a convex neighborhood of τ(ti)\tau(t_{i}) in AA such that VM{M}M=V\cap\bigcup_{M^{\prime}\in\mathcal{H}\setminus\{M\}}M^{\prime}=\emptyset. Let DD be the half-apartment of AA delimited by MM and containing PP. Then VABVPV\cap A\cap B\supset V\cap P. Let us prove that VP=VDV\cap P=V\cap D. As DD and PP are delimited by MM, we have PDP\subset D. Suppose VDVPV\cap D\neq V\cap P. Let y(VD)Py\in(V\cap D)\setminus P. Let xP̊Vx\in\mathring{P}\cap V. Then there exists z[x,y]z\in[x,y] such that zz is in the boundary Fr(P)\mathrm{Fr}(P) of PP. Then zVz\in V and by assumption on VV, we have zMz\in M. As xD̊x\in\mathring{D}, we have y=zMy=z\in M. But then [x,y]P[x,y]\subset P: a contradiction. Therefore VP=VDV\cap P=V\cap D. Then by Lemma 3.1, maybe replacing VV by a smaller neighborhood of τ(ti)\tau(t_{i}), we can assume that VAB=VDV\cap A\cap B=V\cap D. Set πB=ρ,Bτ\pi_{B}=\rho_{-\infty,B}\circ\tau. By Lemma 3.2, we have πB,(ti)Q,BπB,+(ti)\pi^{\prime}_{B,-}(t_{i})\leq_{Q^{\vee}_{\mathbb{R},B}}\pi_{B,+}^{\prime}(t_{i}) and π(ti)Wv.π+(ti)\pi^{\prime}_{-}(t_{i})\in W^{v}.\pi^{\prime}_{+}(t_{i}). Therefore π(ti)Qπ+(ti)\pi^{\prime}_{-}(t_{i})\leq_{Q^{\vee}_{\mathbb{R}}}\pi^{\prime}_{+}(t_{i}). By induction we deduce that π\pi satisfies (1) and (2). By integrating π\pi^{\prime} between 0 and 11, we deduce the lemma. ∎

Lemma 3.5.

Let A,B𝒜A,B\in\mathcal{A} be two apartments such that Int(AB)\mathrm{Int}(A\cap B)\neq\emptyset. Then ABA\cap B is convex.

Proof.

Using isomorphisms of apartments, we may assume that B=𝔸B=\mathbb{A}. Let aInt(A𝔸)a\in\mathrm{Int}(A\cap\mathbb{A}) and bA𝔸b\in A\cap\mathbb{A}. We begin by proving that [a,b]A𝔸[a,b]_{A}\subset\mathbb{A}. We choose a sequence (an)Int(A𝔸)(a_{n})\in\mathrm{Int}(A\cap\mathbb{A})^{\mathbb{N}} such that anaa_{n}\rightarrow a and such that for all nn\in\mathbb{N}, anE(A,,b)E(A,+,b)a_{n}\in E(A,-\infty,b)\cap E(A,+\infty,b) (this is possible since E(A,,b)E(A,-\infty,b) and E(A,+,b)E(A,+\infty,b) are obtained from AA by removing finitely many hyperplanes). Let nn\in\mathbb{N}. Let τ:[0,1][an,b]A\tau:[0,1]\mapsto[a_{n},b]_{A} be the affine parametrization of [an,b][a_{n},b] such that τ(0)=an\tau(0)=a_{n} and τ(1)=b\tau(1)=b. Let π=ρτ\pi=\rho_{-\infty}\circ\tau. Then by Lemma 3.4, π+(0)Qban\pi^{\prime}_{+}(0)\leq_{Q^{\vee}_{\mathbb{R}}}b-a_{n}. Let π~=ρ+τ:[0,1]𝔸\tilde{\pi}=\rho_{+\infty}\circ\tau:[0,1]\rightarrow\mathbb{A}. Then similarly to Lemma 3.4, we have π~+(0)Qban\tilde{\pi}_{+}(0)\geq_{Q^{\vee}_{\mathbb{R}}}b-a_{n}. Moreover τ([0,ϵ)𝔸\tau([0,\epsilon)\subset\mathbb{A} for ϵ>0\epsilon>0 small enough. Therefore ρ+τ|[0,ϵ]=ρτ|[0,ϵ]=τ|[0,ϵ]\rho_{+\infty}\circ\tau|_{[0,\epsilon]}=\rho_{-\infty}\circ\tau|_{[0,\epsilon]}=\tau|_{[0,\epsilon]}. Consequently

π+(0)=π~+(0)QbanQπ~+(0)=π+(0).\pi^{\prime}_{+}(0)=\tilde{\pi}^{\prime}_{+}(0)\leq_{Q^{\vee}_{\mathbb{R}}}b-a_{n}\leq_{Q^{\vee}_{\mathbb{R}}}\tilde{\pi}^{\prime}_{+}(0)=\pi^{\prime}_{+}(0).

As (αi)iI(\alpha_{i}^{\vee})_{i\in I} is free, we deduce that π+(0)=π~+(0)=ban\pi^{\prime}_{+}(0)=\tilde{\pi}_{+}(0)=b-a_{n}. The path π\pi is thus a line segment, since otherwise we would have ban=π(1)π(0)>Qbanb-a_{n}=\pi(1)-\pi(0)>_{Q^{\vee}_{\mathbb{R}}}b-a_{n} by Lemma 3.4. By symmetry, π~\tilde{\pi} is the line segment joining ana_{n} to bb. Therefore π=π~\pi=\tilde{\pi} and thus ρτ(t)=ρ+τ(t)\rho_{-\infty}\circ\tau(t)=\rho_{+\infty}\circ\tau(t) for t[0,1]t\in[0,1]. By [Héb20, Proposition 3.7], τ(t)𝔸\tau(t)\in\mathbb{A} for every t[0,1]t\in[0,1]. Therefore [an,b]𝔸[a_{n},b]\subset\mathbb{A}. Moreover

n[an,b]A¯[a,b]A.\overline{\bigcup_{n\in\mathbb{N}}[a_{n},b]_{A}}\supset[a,b]_{A}.

By [Héb20, Proposition 3.9], A𝔸A\cap\mathbb{A} is closed and hence [a,b]A𝔸[a,b]_{A}\subset\mathbb{A}.

Let VV be an open neighborhood of aa contained in A𝔸A\cap\mathbb{A}. Then by what we proved, vV[v,b]AA𝔸\bigcup_{v\in V}[v,b]_{A}\subset A\cap\mathbb{A}. Moreover if nn\in\mathbb{N}, then (11n)b+1nV(1-\frac{1}{n})b+\frac{1}{n}V is a neighborhood of (11n)b+1na(1-\frac{1}{n})b+\frac{1}{n}a contained in vV[v,b]AA𝔸\bigcup_{v\in V}[v,b]_{A}\subset A\cap\mathbb{A}. Therefore bInt(A𝔸)¯b\in\overline{\mathrm{Int}(A\cap\mathbb{A})}. Therefore Int(A𝔸)\mathrm{Int}(A\cap\mathbb{A}) is dense in A𝔸A\cap\mathbb{A}. We also proved that Int(A𝔸)\mathrm{Int}(A\cap\mathbb{A}) is convex and as A𝔸A\cap\mathbb{A} is closed, the lemma follows. ∎

Theorem 3.6.

Let AA and BB be two apartments. Then ABA\cap B is enclosed and there exists an apartment isomorphism ϕ:AB\phi:A\rightarrow B fixing ABA\cap B.

Proof.

We assume that ABA\cap B is non-empty. Let a,bABa,b\in A\cap B. Let CaC_{a} (resp. Cb)C_{b}) be a chamber of AA (resp. of BB) based at aa (resp. bb). By [Héb20, Proposition 5.17 (ii)], there exists an apartment A~\tilde{A} containing CaC_{a} and CbC_{b}. By Lemma 3.5, AA~A\cap\tilde{A} and A~B\tilde{A}\cap B are convex. By [Héb20, Proposition 3.26], there exists ϕA:AAA~A~\phi_{A}:A\overset{A\cap\tilde{A}}{\rightarrow}\tilde{A} and ϕB:A~A~BB\phi_{B}:\tilde{A}\overset{\tilde{A}\cap B}{\rightarrow}B. Set fa,b=ϕBϕAf_{a,b}=\phi_{B}\circ\phi_{A}. Then [a,b]A=[a,b]A~=[a,b]B[a,b]_{A}=[a,b]_{\tilde{A}}=[a,b]_{B} and fa,bf_{a,b} fixes [a,b]A[a,b]_{A}. Therefore ABA\cap B is convex.

Let HH be the support of ABA\cap B, that is, HH is the smallest affine subspace of AA containing ABA\cap B. If EHE\subset H, we denote by IntH(E)\mathrm{Int}_{H}(E) its interior in HH. By [Héb20, Proposition 3.14], there exist kk\in\mathbb{N}, enclosed subsets P1,,PkP_{1},\ldots,P_{k} of AA such that AB=i=1kPiA\cap B=\bigcup_{i=1}^{k}P_{i} and such that for all i1,ki\in\llbracket 1,k\rrbracket, there exists an apartment isomorphism ϕi:APiB\phi_{i}:A\overset{P_{i}}{\rightarrow}B. As ABA\cap B and the PiP_{i} are convex, there exists i1,ki\in\llbracket 1,k\rrbracket such that IntH(Pi)\mathrm{Int}_{H}(P_{i})\neq\emptyset.

Let us prove that ϕi\phi_{i} fixes ABA\cap B. Let aIntH(Pi)a\in\mathrm{Int}_{H}(P_{i}). Let bABb\in A\cap B. Let f=ϕi1fa,b:AAf=\phi_{i}^{-1}\circ f_{a,b}:A\rightarrow A. Then ff fixes a neighborhood of aa in [a,b][a,b] and as it is an affine isomorphism, it fixes (a,b)\mathcal{L}(a,b). In particular f(b)=bf(b)=b and hence ϕi(b)=fa,b(b)=b\phi_{i}(b)=f_{a,b}(b)=b, which proves that ϕi\phi_{i} fixes ABA\cap B. By [Héb20, Proposition 3.22], ABA\cap B is enclosed, which completes the proof of the theorem. ∎

Corollary 3.7.

Let \mathcal{I} be a set endowed with a covering 𝒜\mathcal{A} of subsets called apartments. We assume that (αi)iI(\alpha_{i})_{i\in I} is free in 𝔸\mathbb{A}^{*} and that (αi)iI(\alpha_{i}^{\vee})_{i\in I} is free in 𝔸\mathbb{A}. Then \mathcal{I} is a masure in the sense of [Rou11, Définition 2.1] if and only if it satisfies the following axioms:

(MA I)=(MA i): Any A𝒜A\in\mathcal{A} is equipped with the structure of an apartment of type 𝔸\mathbb{A}.

(MA II) : Let A,AA,A^{\prime} be two apartments. Then AAA\cap A^{\prime} is enclosed and there exists an apartment isomorphism ϕ:AAAA\phi:A\overset{A\cap A^{\prime}}{\rightarrow}A^{\prime}.

(MA III)=(MA iii): if \mathfrak{R} is the germ of a splayed chimney and if FF is a face or a germ of a chimney, then there exists an apartment containing \mathfrak{R} and FF.

Remark 3.8.

By [Héb20, Proposition 4.25 and Remark 4.26], if \mathcal{I} is thick (which means that each panel of \mathcal{I} is dominated by at least three chambers) and if there exists a group acting strongly transitively on \mathcal{I} (see [Héb18, 3.1.5 and Corollary 4.4.40] for the definition of such an action), then for every enclosed subset PP of 𝔸\mathbb{A} such that P̊\mathring{P}\neq\emptyset, there exists an apartment AA such that A𝔸=PA\cap\mathbb{A}=P. This is in particular true when \mathcal{I} is the masure associated with a split Kac-Moody group over a valued field.

References

  • [BK11] Alexander Braverman and David Kazhdan. The spherical Hecke algebra for affine Kac-Moody groups I. Annals of mathematics, 174 (2011) pages 1603–1642.
  • [BKP16] Alexander Braverman, David Kazhdan, and Manish M. Patnaik. Iwahori-Hecke algebras for pp-adic loop groups. Invent. Math., 204(2):347–442, 2016.
  • [BPGR16] Nicole Bardy-Panse, Stéphane Gaussent, and Guy Rousseau. Iwahori-Hecke algebras for Kac-Moody groups over local fields. Pacific J. Math., 285(1):1–61, 2016.
  • [BT72] François Bruhat and Jacques Tits. Groupes réductifs sur un corps local. Publications Mathématiques de l’IHÉS, 41(1):5–251, 1972.
  • [BT84] François Bruhat and Jacques Tits. Groupes réductifs sur un corps local. Publications Mathématiques de l’IHÉS, 60(1):5–184, 1984.
  • [Cha10] Cyril Charignon. Immeubles affines et groupes de Kac-Moody. PhD thesis, université Henri Poincaré Nancy 1, 2010.
  • [GR08] Stéphane Gaussent and Guy Rousseau. Kac-Moody groups, hovels and Littelmann paths. In Annales de l’institut Fourier, volume 58, pages 2605–2657, 2008.
  • [GR14] Stéphane Gaussent and Guy Rousseau. Spherical Hecke algebras for Kac-Moody groups over local fields. Annals of Mathematics, 180(3):1051–1087, 2014.
  • [Héb18] Auguste Hébert. Study of masures and of their applications in arithmetic. PhD thesis. English version. hal.archives ouvertes tel-01856620v1, June 2018.
  • [Héb20] Auguste Hébert. A New Axiomatics for Masures. Canad. J. Math., 72(3):732–773, 2020.
  • [Héb21] Auguste Hébert. Distances on a masure. working paper or preprint, August 2021, accepted by Transformation groups. https://hal.archives-ouvertes.fr/hal-01397819v3/file/distances_on_a_masure.pdf
  • [Rém02] Bertrand Rémy. Groupes de Kac-Moody déployés et presque déployés. Astérisque, (277):viii+348, 2002.
  • [Rou11] Guy Rousseau. Masures affines. Pure and Applied Mathematics Quarterly, 7(3):859–921, 2011.
  • [Rou16] Guy Rousseau. Groupes de Kac-Moody déployés sur un corps local II. Masures ordonnées. Bull. Soc. Math. France, 144(4):613–692, 2016.
  • [Rou17] Guy Rousseau. Almost split Kac–Moody groups over ultrametric fields. Groups Geometry, and Dynamics, 11:891–975, 2017.
  • [Tit87] Jacques Tits. Uniqueness and presentation of Kac-Moody groups over fields. J. Algebra, 105(2):542–573, 1987.