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

Spherical tropicalization and Berkovich analytification

Desmond Coles
( The University of Texas at Austin
Austin, TX, United States of America
[email protected]
)

Abstract. Let XX be a spherical variety. We show that Tevelev and Vogiannou’s tropicalization map from XX to its tropicalization factors through the Berkovich analytification XanX^{\text{an}}, as in the case for toric varieties. Furthermore we show that the tropicalization is a strong deformation retraction of XanX^{\text{an}}. We also give a strong deformation retraction of Thuillier’s analytification XX^{\beth} onto a subspace described using the colored fan of XX.

Keywords. Spherical variety. Tropicalization. Berkovich analytification. Algebraic Geometry.

Statements and Declarations. The author has no competing interests.

Acknowledgements. The author would like to thank their advisor, Sam Payne, for their support and guidance during this research. The paper also benefited greatly from conversations with Tom Gannon, Sam Raskin, Yixian Wu, Hernán Iriarte, Amy Li, and Logan White. The author was grateful to be funded by NSF DMS–2001502 and NSF DMS–2053261.

1   Introduction

Tropical geometry provides a set of tools for assigning combinatorial objects to algebraic varieties. One then studies the algebraic geometry of the original variety by looking at the combinatorics. One well known example is given by embedding a variety into a toric variety and then tropicalizing the toric variety [Pay09]. In [TV21] Tevelev and Vogiannou introduce tropicalization for spherical homogenous spaces. Let kk be an algebraically closed field. Recall that if GG is a reductive group over kk, and XX is a normal GG-variety, then XX is spherical if there is a Borel subgroup of GG with an open orbit in XX. Spherical varieties can be considered the nonabelian analogue of toric varities. Let K¯=k{{u}}\overline{K}=k\{\!\{u\}\!\} be Puiseux series, the algebraic closure of Laurent series K=k((u))K=k(\!(u)\!). Equip K¯\overline{K} with the uu-adic valuation ||||u|\!|\!\cdot\!|\!|_{u}. Given a spherical homogeneous space G/HG/H, Tevelev and Vogiannou define a set theoretic tropicalization map tropG:G/H(K¯)𝒱\operatorname{trop}_{G}\colon G/H(\overline{K})\rightarrow{\mathcal{V}}, where 𝒱{\mathcal{V}} is the set of G(k)G(k)-invariant valuations on the function field k(G/H)k(G/H). The set 𝒱{\mathcal{V}} in fact has the structure of a finitely generated convex cone [Kno91, Corollary 5.3]. The map tropG\operatorname{trop}_{G} takes xG/H(K¯)x\in G/H(\overline{K}) to the valuation ||||x𝒱|\!|\!\cdot\!|\!|_{x}\in{\mathcal{V}} defined by the formula:

fx=gf(x)ug in general position\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{gf(x)}\right|\!\right|_{u}\quad\text{$g$ in general position}

where general position means gUfg\in U_{f}, UfG(k)U_{f}\subseteq G(k) a nonempty Zariski open such that gf(x)u\left|\!\left|{gf(x)}\right|\!\right|_{u} is constant for gUfg\in U_{f}. For a subvariety WG/HW\subseteq G/H one can then define TropG(W)\operatorname{Trop}_{G}(W) as the closure of tropG(W(K¯))\operatorname{trop}_{G}(W(\overline{K})). This construction agrees with the usual tropicalization map in the case when G=TG=T, TT a torus, and HH trivial.

In [Nas17] Tevelev and Vogiannou’s construction was extended to all spherical varieties, generalizing the extension of tropicalization of tori to toric varities given in [Pay09]. Here the target of the tropicalization map, TropG(X)\operatorname{Trop}_{G}(X), consists of a gluing of the valuation cones associated to GG-orbits in XX. The tropicalization map is defined as a disjoint union of the maps given above. In [KM19] Kaveh and Manon develop tropical geometry for spherical varieties via Gröbner theory and it is shown that this agrees with the above notion.

Let XanX^{\text{an}} be the Berkovich analytification of XX. In the case when XX is a toric variety with dense torus TT, TropT(X)\operatorname{Trop}_{T}(X) is canonically a subspace of XanX^{\text{an}} and the tropicalization map factors through a continuous retraction XanTropT(X)X^{\text{an}}\rightarrow\operatorname{Trop}_{T}(X). It was conjectured in [TV21] that this is true for tropicalizations of spherical homogeneous spaces. In [KM19] this is proved for affine spherical varieties. We show this is the case for all spherical varieties. Let GG^{\beth} denote the kk-analytic space associated to GG as defined in [Thu07, Section 1]. Let 𝐩:XanXan{\mathbf{p}}\colon X^{\text{an}}\rightarrow X^{\text{an}} be the map given by \ast-multiplication by the unique point in the Shilov boundary of GG^{\beth}.

Theorem A.

Let XX be a spherical GG-variety over an algebraically closed and trivially valued field kk, and let K¯\overline{K} denote Puiseux series in kk. Then there is a canonical homeomorphism ι:TropG(X)𝐩(Xan)\iota:\operatorname{Trop}_{G}(X)\rightarrow{\mathbf{p}}(X^{\text{an}}), and 𝐩:XanTropG(X){\mathbf{p}}\colon X^{\text{an}}\rightarrow\operatorname{Trop}_{G}(X) is a retraction of topological spaces such that the following diagram commutes:

X(K¯){X(\overline{K})}Xan{X^{\text{an}}}TropG(X).{\operatorname{Trop}_{G}(X).}tropG\scriptstyle{\operatorname{trop}_{G}}𝐩\scriptstyle{{\mathbf{p}}}
Corollary.

For WW a subvariety of XX, there is a strong deformation retraction H:[0,1]×WanTropG(W)H:[0,1]\times W^{\text{an}}\rightarrow\operatorname{Trop}_{G}(W) such that H(1,p)=𝐩(p)H(1,p)={\mathbf{p}}(p).

The construction of the map 𝐩{\mathbf{p}} was inspired by the work of Thuillier in the toric case, [Thu07, Section 2]. The space GG^{\beth} is a compact analytic subdomain of the kk-analytic space GanG^{\text{an}}, the Shilov boundary of this space is a point with important similarities to the generic point of GG. The notion of \ast-multiplication is an idea originally due to Berkovich which generalizes left-multiplication by points in Gan(k)G^{\text{an}}(k).

Our work in Section 3 extends the notion of tropicalization to other nontrivially valued extensions of kk besides K¯\overline{K}. We show that for any field extension L/kL/k with valuation |||||\!|\!\cdot\!|\!| there is a well-defined tropicalization map tropG,L:X(L)TropG(X)\operatorname{trop}_{G,L}\colon X(L)\rightarrow\operatorname{Trop}_{G}(X). For xX(L)x\in X(L), tropG(x)=||||x\operatorname{trop}_{G}(x)=|\!|\!\cdot\!|\!|_{x} where ||||x|\!|\!\cdot\!|\!|_{x} is the valuation ||||x𝒱|\!|\!\cdot\!|\!|_{x}\in{\mathcal{V}} given by the formula:

fx=gf(x)g in general position\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{gf(x)}\right|\!\right|\quad\text{$g$ in general position}

where general position means gUfg\in U_{f}, UfG(k)U_{f}\subseteq G(k) is a nonempty Zariski open such that gf(x)\left|\!\left|{gf(x)}\right|\!\right| is constant for gUfg\in U_{f}. It follows from Theorem A that tropicalization is invariant under the choice of LL.

Corollary.

Let WW be a subvariety of XX, and LL be a nontrivially valued and algebraically closed extension of kk. Then the closure of tropG,L(W(L))\operatorname{trop}_{G,L}(W(L)) in TropG(X)\operatorname{Trop}_{G}(X) is equal to 𝐩(Wan){\mathbf{p}}(W^{\text{an}}), where WanW^{\text{an}} is considered as a subspace of XanX^{\text{an}} induced by the inclusion WXW\subseteq X.

Our second main result is on the restriction of the map 𝐩{\mathbf{p}} to the subspace XX^{\beth}. This result generalizes Proposition 2.12 of [Thu07] on the toric case.

Theorem B.

There is a strong deformation retraction from XX^{\beth} onto 𝐩(X){\mathbf{p}}(X^{\beth}). If the colored fan 𝔉(X){\mathfrak{F}}(X) satisfies:

(σ,)𝔉(X)σ𝒱.\bigcup_{(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)}\sigma\subseteq{\mathcal{V}}. (\star)

Then 𝐩(X){\mathbf{p}}(X^{\beth}) is homeomorphic to the the canonical compactification of the fan F(X)={σ|(σ,)𝔉(X)}F(X)=\{\sigma\,|\,(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)\}.

Recall that just as toric varieties are classified by fans, spherical varieties are classified by colored fans. For each spherical homogeneous space G/HG/H there is an associated vector space NN_{{\mathbb{R}}}. A colored fan, defined precisely in Section 4, is a finite collection of pairs (σ,)(\sigma,{\mathcal{F}}) where σ\sigma is a strictly convex cone in NN_{{\mathbb{R}}} and {\mathcal{F}} is a finite set. The canonical compactification of a cone σ\sigma is defined as hom(σ,[0,])\hom(\sigma^{\vee},[0,\infty]), where σ\sigma^{\vee} consists of positive linear functionals on σ\sigma and the homomorphisms are semigroup homomorphisms that preserve scaling by 0{\mathbb{R}}_{\geq 0}. The canonical compactification of a fan is the polyhedral complex given by gluing the canonical compactifications of the cones along shared faces. See Remark 6.1 for an explanation of why we impose \starB and some examples where this condition does and does not hold. In Remark 6.5 we discuss how to give a description of 𝐩(X){\mathbf{p}}(X^{\beth}) when \starB does not hold.

Outline of paper.

Section 2 introduces the necessary notions from Berkovich geometry. Section 3 introduces \ast-multiplication and studies the map 𝐩{\mathbf{p}}. Section 4 recalls the Luna-Vust classification of G/HG/H-embeddings, and contains important notation for Sections 5 and 6. Section 5 recalls the work of [TV21] and [Nas17] and gives a proof of Theorem A. Section 6 recalls the work of [Thu07, Section 2] and gives a proof of Theorem B.

Convention Choice.

Though somewhat nonstandard in the literature of spherical varieties we opt to use multiplicative notation for valuations to accommodate the conventions of Berkovich geometry. Our exact conventions are as follows. Let AA be a ring. By a seminorm on AA we mean a function ||||A:A0|\!|\!\cdot\!|\!|_{A}\colon A\rightarrow{\mathbb{R}}_{\geq 0} such that 1A=1\left|\!\left|{1}\right|\!\right|_{A}=1, 0A=0\left|\!\left|{0}\right|\!\right|_{A}=0, fgAfAgA\left|\!\left|{fg}\right|\!\right|_{A}\leq\left|\!\left|{f}\right|\!\right|_{A}\left|\!\left|{g}\right|\!\right|_{A}, and f+gAmax(fA,gA)\left|\!\left|{f+g}\right|\!\right|_{A}\leq\max({\left|\!\left|{f}\right|\!\right|_{A},\left|\!\left|{g}\right|\!\right|_{A}}).111These are non-Archimedean seminorms, which are the only ones of interest because our base field will always be non-Archimedean. We say that ||||A|\!|\!\cdot\!|\!|_{A} is a norm if fA0\left|\!\left|{f}\right|\!\right|_{A}\neq 0 for all f0f\neq 0. A (semi)norm will be called a (semi)valuation if it is multiplicative: fgA=fAgA\left|\!\left|{fg}\right|\!\right|_{A}=\left|\!\left|{f}\right|\!\right|_{A}\left|\!\left|{g}\right|\!\right|_{A}.222More commonly valuations given as functions ν\nu into the additive semigroup {}{\mathbb{R}}\sqcup\{\infty\}, a multiplicative norm is equivalent to such a function ν\nu, set ν=log(||||)\nu=-\log(|\!|\!\cdot\!|\!|). If kk is a field with valuation ||||k|\!|\!\cdot\!|\!|_{k} and AA is a kk-algebra then a seminorm on AA will always be assumed to extend the valuation on kk; λfA=λkfA\left|\!\left|{\lambda f}\right|\!\right|_{A}=\left|\!\left|{\lambda}\right|\!\right|_{k}\left|\!\left|{f}\right|\!\right|_{A} for all λk\lambda\in k. By a valued extension L/kL/k we mean a field extension where LL is equipped with a valuation (extending the valuation on kk by our previous convention).

Acknowledgments.

The author would like to thank their advisor, Sam Payne, for their support and guidance during this research. The paper also benefited greatly from conversations with Tom Gannon, Sam Raskin, Yixian Wu, Hernán Iriarte, Amy Li, and Logan White.

2   Overview of Berkovich geometry

The contents of this Section are an overview of the necessary notions in the theory of Berkovich spaces, in particular we need to work with analytifications of varieties and \beth-spaces. Berkovich spaces provide a framework for doing analytic geometry over valued fields. Given a base field kk with valuation, Berkovich spaces are called kk-analytic spaces. These are locally ringed spaces equipped with a kk-analytic atlas. We will not go into full detail of the theory as we are primarily concerned with the topology of these spaces, and will only introduce the notions we need. The content in this Section is laid out in detail in [Ber90], with the exception of \beth-spaces, for this we refer the reader to [Thu07].

Throughout this Section let kk be a valued field, and let XX be a separated locally finite type kk-scheme.

2.1   Berkovich analytification

Definition 2.1.

The Berkovich analytification of XX, denoted XanX^{\text{an}}, has points given by pairs (x,||||)(x,|\!|\!\cdot\!|\!|) where xx is a scheme point of XX and |||||\!|\!\cdot\!|\!| is a valuation on k(x)k(x). There is a map ρ:XanX\rho\colon X^{\text{an}}\rightarrow X given by (x,||||)x(x,|\!|\!\cdot\!|\!|)\mapsto x. The topology on XanX^{\text{an}} is the coarsest such that ρ\rho is continuous, and for any open UXU\subseteq X and any f𝒪X(U)f\in{\mathcal{O}}_{X}(U), the map ρ1(U)\rho^{-1}(U)\rightarrow{\mathbb{R}} given by (x,||||)||f||(x,|\!|\!\cdot\!|\!|)\mapsto\left|\!\left|{f}\right|\!\right| is continuous.

Any valued extension L/kL/k and map of schemes x:SpecLXx\colon\operatorname{Spec}L\rightarrow X defines a point in XanX^{\text{an}}. The point is given by (x,||||)(x,|\!|\!\cdot\!|\!|) where xx is the point corresponding to the image of SpecL\operatorname{Spec}L and |||||\!|\!\cdot\!|\!| is the valuation on LL restricted to k(x)k(x). There is also a sheaf of analytic functions on XanX^{\text{an}}, however we only need a few details. In particular for pXanp\in X^{\text{an}} the stalk at pp is a local ring. If κ(p)\kappa(p) is the residue field of the stalk at pp then κ(p)\kappa(p) has a norm and we let (p){\mathcal{H}}(p) denote the completion with respect to this norm. If p=(x,||||)p=(x,|\!|\!\cdot\!|\!|) then there is a canonical isometric isomorphism from the completion of k(x)k(x) with respect to |||||\!|\!\cdot\!|\!|, to (p){\mathcal{H}}(p).

When X=SpecAX=\operatorname{Spec}A, AA a kk-algebra, we can alternatively describe XanX^{\text{an}} as being the semivaluations on AA. The topology on XanX^{\text{an}} in this case will be the coarsest such that evaluation at an element of AA is continuous. One can see that this description of XanX^{\text{an}} is equivalent to the original as follows. Let the kernel of a semivaluation |||||\!|\!\cdot\!|\!| on AA, be ker(||||):={fA|||f||=0}\ker(|\!|\!\cdot\!|\!|):=\{f\in A\,|\,\left|\!\left|{f}\right|\!\right|=0\}. Then |||||\!|\!\cdot\!|\!| defines a valuation on the fraction field of A/ker(||||)A/\ker(|\!|\!\cdot\!|\!|), which we also denote |||||\!|\!\cdot\!|\!|. This valuation is characterized by ||f+ker(||||)||:=||f||\left|\!\left|{f+\ker(|\!|\!\cdot\!|\!|)}\right|\!\right|:=\left|\!\left|{f}\right|\!\right|.

Berkovich analytification is functorial. For any morphism of kk-schemes f:XYf\colon X\rightarrow Y there is a morphism of Berkovich spaces over kk, fan:XanYanf^{\text{an}}\colon X^{\text{an}}\rightarrow Y^{\text{an}}. In particular for a point (x,||||)Xan(x,|\!|\!\cdot\!|\!|)\in X^{\text{an}} one has that fan((x,||||))=(f(x),||fx()||)f^{\text{an}}((x,|\!|\!\cdot\!|\!|))=(f(x),\left|\!\left|{f_{x}(\cdot)}\right|\!\right|), where fx:k(f(x))k(x)f_{x}\colon k(f(x))\rightarrow k(x) is the induced map on function fields.

2.2   The Berkovich spectrum and \beth-spaces

We need to introduce one other type of kk-analytic space. Let 𝒜{\mathcal{A}} be a Banach kk-algebra, i.e. 𝒜{\mathcal{A}} is a kk-algebra equipped with a norm ||||𝒜|\!|\!\cdot\!|\!|_{{\mathcal{A}}}, and 𝒜{\mathcal{A}} is complete with respect to this norm.

Definition 2.2.

The Berkovich spectrum, (𝒜){\mathcal{M}}({\mathcal{A}}), is the set of bounded semivaluations on 𝒜{\mathcal{A}}, i.e. semivaluations |||||\!|\!\cdot\!|\!| such that where ||||||||𝒜|\!|\!\cdot\!|\!|\leq|\!|\!\cdot\!|\!|_{{\mathcal{A}}}. The topology is the coarsest such that evaluation at an element of 𝒜{\mathcal{A}} is continuous.

Example 2.3.

The one-point spaces in the category of kk-analytic spaces are given by (L){\mathcal{M}}(L) where LL is a complete field over kk.

Let 𝒜{\mathcal{A}} and {\mathcal{B}} be Banach kk-algebras with norms ||||𝒜|\!|\!\cdot\!|\!|_{{\mathcal{A}}} and |||||\!|\!\cdot\!|\!|_{{\mathcal{B}}}. One defines the completed tensor product, 𝒜^k{\mathcal{A}}\hat{\otimes}_{k}{\mathcal{B}} to be the completion of 𝒜k{\mathcal{A}}\otimes_{k}{\mathcal{B}}, with respect to the norm:

f𝒜,:=inf{maxi||gi||𝒜fi|ifigi𝒜k such thatf=ifigi}.|\!|f|\!|_{{\mathcal{A}},{\mathcal{B}}}:=\inf\left\{\max_{i}|\!|g_{i}|\!|_{{\mathcal{A}}}|\!|f_{i}|\!|_{{\mathcal{B}}}\,\bigg{|}\,\text{$\sum\limits_{i}f_{i}\otimes g_{i}\in{\mathcal{A}}\otimes_{k}{\mathcal{B}}$ such that}\,f=\sum\limits_{i}f_{i}\otimes g_{i}\right\}.

The function ||||𝒜k|\!|\!\cdot\!|\!|_{{\mathcal{A}}{\otimes}_{k}{\mathcal{B}}} is clearly a seminorm, that it is a norm and that 𝒜k{\mathcal{A}}{\otimes}_{k}{\mathcal{B}} embeds isometrically into its completion follows from part 4 of Theorem 1 in [Gru66, Section 3].

Proposition 2.4.

[Ber90, Chapter 3] The norm ||||𝒜,|\!|\!\cdot\!|\!|_{{\mathcal{A}},{\mathcal{B}}} makes (𝒜^k){\mathcal{M}}({\mathcal{A}}\hat{\otimes}_{k}{\mathcal{B}}) the product (𝒜)×(){\mathcal{M}}({\mathcal{A}})\times{\mathcal{M}}({\mathcal{B}}) in the category of kk-analytic spaces.

Notation.

To avoid cluttered subscripts, when considering pp in some analytic space will choose to write ||||p|\!|\!\cdot\!|\!|_{p} for the norm on (p){\mathcal{H}}(p) instead of ||||(p)|\!|\!\cdot\!|\!|_{{\mathcal{H}}(p)}. Similarly for a pair of points pp and qq we will write ||||p,q|\!|\!\cdot\!|\!|_{p,q} for the norm on (p)^k(q){\mathcal{H}}(p)\hat{\otimes}_{k}{\mathcal{H}}(q) instead of ||||(p),(q)|\!|\!\cdot\!|\!|_{{\mathcal{H}}(p),{\mathcal{H}}(q)}.

Consider when kk is trivially valued. Let AA be a finitely generated kk-algebra, AA equipped with the trivial valuation is complete. In this case we can identify (A){\mathcal{M}}(A) with seminorms |||||\!|\!\cdot\!|\!|, that are bounded by 11. Recall that if XX is a variety over kk, L/kL/k is a valued field extension, and RR is the valuation ring of LL, then the map of schemes SpecLX\operatorname{Spec}L\rightarrow X has center if it factors through SpecLSpecR\operatorname{Spec}L\rightarrow\operatorname{Spec}R. We say the center is the image of the closed point of SpecR\operatorname{Spec}R. Furthermore if X=SpecAX=\operatorname{Spec}A and AA is an integral domain, then a valuation |||||\!|\!\cdot\!|\!| has center in XX if and only if |||||\!|\!\cdot\!|\!| is bounded by 1 on AA. From this it follows that for any trivially valued finitely generated kk-algebra AA, (A){\mathcal{M}}(A) can be taken to be pairs (x,||||)(x,|\!|\!\cdot\!|\!|), xx a scheme point of AA, and |||||\!|\!\cdot\!|\!| a valuation on k(x)k(x) having center in SpecA\operatorname{Spec}A. For a general locally finite type scheme XX over a trivially valued field kk, we define XX^{\beth} to be given by pairs of points (x,||||)(x,|\!|\!\cdot\!|\!|) where |||||\!|\!\cdot\!|\!| is a valuation on k(x)k(x), and Speck(x)X\operatorname{Spec}k(x)\rightarrow X has center in XX. So (A)=(SpecA){\mathcal{M}}(A)=\left(\operatorname{Spec}A\right)^{\beth}.

Proposition 2.5.

[Thu07] The space XX^{\beth} is a compact analytic subdomain of XanX^{\text{an}}, and they are equal when XX is proper.

2.3   The reduction map

Let 𝒜:={f𝒜|f𝒜1}{\mathcal{A}}^{\circ}:=\{f\in{\mathcal{A}}\,|\,\left|\!\left|{f}\right|\!\right|_{{\mathcal{A}}}\leq 1\} and 𝒜:={f𝒜|f𝒜<1}{\mathcal{A}}^{\circ\circ}:=\{f\in{\mathcal{A}}\,|\,\left|\!\left|{f}\right|\!\right|_{{\mathcal{A}}}<1\}. Then the reduction of 𝒜{\mathcal{A}} is said to be 𝒜~:=𝒜/𝒜\widetilde{{\mathcal{A}}}:={\mathcal{A}}^{\circ}/{\mathcal{A}}^{\circ\circ}. Any map of Banach kk-algebras 𝒜{\mathcal{A}}\rightarrow{\mathcal{B}} induces a map of rings 𝒜~~\widetilde{{\mathcal{A}}}\rightarrow\widetilde{{\mathcal{B}}}. In particular for p(𝒜)p\in{\mathcal{M}}({\mathcal{A}}) there is a map of rings hp:𝒜~(p)~h_{p}\colon\widetilde{{\mathcal{A}}}\rightarrow\widetilde{{\mathcal{H}}(p)}. Then there is a map of sets called the reduction map, r:(𝒜)Spec(𝒜~)r\colon{\mathcal{M}}({\mathcal{A}})\rightarrow\operatorname{Spec}(\widetilde{{\mathcal{A}}}) defined by pker(hp)p\mapsto\ker(h_{p}).

When AA is a finitely generated algebra with the trivial norm we have A=AA^{\circ}=A and A=0A^{\circ\circ}=0, so reduction gives a map of sets r:(A)SpecAr\colon{\mathcal{M}}(A)\rightarrow\operatorname{Spec}A. More generally we have a map r:XXr\colon X^{\beth}\rightarrow X. When XX^{\beth} is viewed as pairs (x,||||)(x,|\!|\!\cdot\!|\!|), reduction maps (x,||||)(x,|\!|\!\cdot\!|\!|) to its center. This map is anticontinuous [Ber90, Corollary 2.4.2], and surjective [Ber90, Proposition 2.4.4]. Recall that anticontinuous means that the preimage of an open set is closed.

2.4   The Shilov boundary

A boundary of (𝒜){\mathcal{M}}({\mathcal{A}}) is a closed subset of (𝒜){\mathcal{M}}({\mathcal{A}}), such that for every f𝒜f\in{\mathcal{A}} the function evf:(𝒜)\text{ev}_{f}\colon{\mathcal{M}}({\mathcal{A}})\rightarrow{\mathbb{R}} given by ||||||f|||\!|\!\cdot\!|\!|\mapsto|\!|f|\!|, attains its maximum. A unique minimal boundary is said to be the Shilov boundary.

Proposition 2.6.

[Ber90, Proposition 2.4.4] The reduction map r:(𝒜)Spec(𝒜~)r\colon{\mathcal{M}}({\mathcal{A}})\rightarrow\operatorname{Spec}(\widetilde{{\mathcal{A}}}) gives a bijection between the Shilov boundary of (𝒜){\mathcal{M}}({\mathcal{A}}) and the generic points of the components of Spec(𝒜~)\operatorname{Spec}(\widetilde{{\mathcal{A}}}).

3   The retraction map

In this Section we assume kk is a trivially valued and algebraically closed field, GG is a connected linear algebraic group over kk, and XX is a GG-variety. Let mm to denote the action map m:G×XGm\colon G\times X\rightarrow G. The objective is to introduce \ast-multiplication, use it construct a retraction of topological spaces 𝐩:XanXan{\mathbf{p}}\colon X^{\text{an}}\rightarrow X^{\text{an}}, and give a description of the image of 𝐩{\mathbf{p}}. Furthermore we show that when GG is reductive we can construct a strong deformation from XanX^{\text{an}} onto 𝐩(Xan){\mathbf{p}}(X^{\text{an}}) using \ast-multiplication.

The notion of \ast-multiplication is taken directly from [Ber90, Chapter 5]. Properties specific to \ast-multiplication by the unique point in the Shilov boundary of GG^{\beth} are directly inspired by the results on tori from [Thu07, Section 2].

3.1   \ast-multiplication

Definition 3.1.

We say that a commutative Banach kk-algebra 𝒜{\mathcal{A}} is peaked if for any valued extension L/kL/k the norm on L^k𝒜L\hat{\otimes}_{k}{\mathcal{A}} is a valuation. We say a point pp of a kk-analytic space is peaked if the completed residue field (p){\mathcal{H}}(p) is peaked.

Proposition 3.2.

All the the points of GanG^{\text{an}} are peaked.

Proof.

This is a particular case of [Poi13, Corollaire 3.14]. ∎

Definition 3.3.

Let gGang\in G^{\text{an}}, because gg is peaked the norm ||||g,p|\!|\!\cdot\!|\!|_{g,p} is a valuation and is thus an element of ((g))×((p)){\mathcal{M}}({\mathcal{H}}(g))\times{\mathcal{M}}({\mathcal{H}}(p)). There is a map of sets XanXanX^{\text{an}}\rightarrow X^{\text{an}} given by pgpp\mapsto g\ast p; where gpg\ast p is the image of ||||g,p|\!|\!\cdot\!|\!|_{g,p} under the composition of maps of kk-analytic spaces:

((g))×((p))Gan×XanmanXan.{\mathcal{M}}({\mathcal{H}}(g))\times{\mathcal{M}}({\mathcal{H}}(p))\hookrightarrow G^{\text{an}}\times X^{\text{an}}\xrightarrow{m^{\text{an}}}X^{\text{an}}.

This is called \ast-multiplication by gg.

Example 3.4.

Let gGan(k)=G(k)g\in G^{\text{an}}(k)=G(k). The map XanXanX^{\text{an}}\rightarrow X^{\text{an}} given by pgpp\mapsto g\ast p is equal to the map induced by left multiplication by gg.

Below are some general facts about \ast-multiplication.

Proposition 3.5.

[Ber90, Proposition 5.2.8] Let GG and HH be algebraic groups acting on varieties XX and YY, respectively. Let ϕ:GH\phi\colon G\rightarrow H be a morphism of algebraic groups, and ψ:XY\psi\colon X\rightarrow Y a morphism of varieties.

  1. 1.

    The map XanXanX^{\text{an}}\rightarrow X^{\text{an}} given by \ast-multiplication by a point gGang\in G^{\text{an}} is continuous.

  2. 2.

    If gg and hh are points of GanG^{\text{an}} and HanH^{\text{an}}, respectively, and pp a point of XanX^{\text{an}}, then g(hp)=(gh)pg\ast(h\ast p)=(g\ast h)\ast p

  3. 3.

    Given a commutative diagram:

    Gan×Xan{G^{\text{an}}\times X^{\text{an}}}Xan{X^{\text{an}}}Han×Yan{H^{\text{an}}\times Y^{\text{an}}}Yan{Y{\text{an}}}ϕan×ψan\scriptstyle{\phi^{\text{an}}\times\psi^{\text{an}}}ψan\scriptstyle{\psi^{\text{an}}}

    one has that ψan(gp)=ϕan(g)ψan(p)\psi^{\text{an}}(g\ast p)=\phi^{\text{an}}(g)\ast\psi^{\text{an}}(p).

Let 𝐠Gan{\mathbf{g}}\in G^{\text{an}} be the point given by the trivial valuation. This is the unique point in the Shilov boundary of GG^{\beth}. Let 𝐩:XanXan{\mathbf{p}}\colon X^{\text{an}}\rightarrow X^{\text{an}} be given by p𝐠pp\mapsto{\mathbf{g}}\ast p. By Proposition 3.5 part 1, 𝐩{\mathbf{p}} is a continuous map of topological spaces. The remainder of this section will be devoted to studying this map.

Remark 3.6.

We can alternatively describe 𝐩{\mathbf{p}} in the following way. We have that 𝐩(p){\mathbf{p}}(p) is the image of the unique point in the Shilov boundary of G×((p))G^{\beth}\times{\mathcal{M}}({\mathcal{H}}(p)) under the map of kk-analytic spaces:

G×((p))G×XanXan.G^{\beth}\times{\mathcal{M}}({\mathcal{H}}(p))\hookrightarrow G^{\beth}\times X^{\text{an}}\rightarrow X^{\text{an}}.

So when G=TG=T is a torus, 𝐩{\mathbf{p}} is the retraction map defined in [Thu07, Section 2].

Example 3.7.

Let X=T=Speck[t1±1,tn±1]X=T=\operatorname{Spec}k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}] be the nn-dimensional torus, and equip k[t1±1,tn±1]k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}] with the trivial valuation. Let λ[0,1]\lambda\in[0,1], and let 𝐠λ{\mathbf{g}}_{\lambda} be the point of TT^{\beth} corresponding to the semivaluation:

IaItI=maxIλ|I||\!|\sum\limits_{I}a_{I}t^{I}|\!|=\max_{I}\lambda^{|I|}

where II denotes a multi-index and if I=(I1,In)I=(I_{1},\ldots I_{n}) then |I|=I1+In|I|=I_{1}+\cdots I_{n}. Then one can see via computation that for any valued extension LL, the norm ||||𝐠λ,L|\!|\!\cdot\!|\!|_{{\mathbf{g}}_{\lambda},L} on (𝐠λ)^kL{\mathcal{H}}({\mathbf{g}}_{\lambda})\hat{\otimes}_{k}L is multiplicative. For λ=0\lambda=0 this is because (𝐠λ)=k{\mathcal{H}}({\mathbf{g}}_{\lambda})=k. For λ(0,1]\lambda\in(0,1] 𝐠λ{\mathbf{g}}_{\lambda} defines a norm on k[t1±1,tn±1]k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}]. For f=IaItIk[t1±1,tn±1]kL=L[t1±1,tn±1]f=\sum_{I}a_{I}t^{I}\in k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}]\otimes_{k}L=L[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}] we compute that:

f𝐠λ,L=maxIaILλ|I||\!|f|\!|_{{\mathbf{g}}_{\lambda},L}=\max_{I}|\!|a_{I}|\!|_{L}\lambda^{|I|}

and this is a valuation.

Let pTanp\in T^{\text{an}} be given by x=(x1,xn)T(L)x=(x_{1},\ldots x_{n})\in T(L), where LL is a valued extension of kk, so p=(x,||||L)p=(x,|\!|\!\cdot\!|\!|_{L}). Let mxm_{x} the composition T×Speck(x)T×T𝑚TT\times\operatorname{Spec}k(x)\hookrightarrow T\times T\xrightarrow{m}T. Then for f=IaItISpeck[t1±1,tn±1]f=\sum_{I}a_{I}t^{I}\in\operatorname{Spec}k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}] seminorm 𝐠λp{\mathbf{g}}_{\lambda}\ast p is given by:

f𝐠λp=mxf𝐠λ,p=IaItIxI𝐠λ,p=maxIaIxILλ|I|.|\!|f|\!|_{{\mathbf{g}}_{\lambda}\ast p}=|\!|m_{x}^{\ast}f|\!|_{{\mathbf{g}}_{\lambda},p}=|\!|\sum\limits_{I}a_{I}t^{I}\otimes x^{I}|\!|_{{\mathbf{g}}_{\lambda},p}=\max_{I}|\!|a_{I}x^{I}|\!|_{L}\lambda^{|I|}.
Corollary 3.8.

The map: 𝐩:XanXan{\mathbf{p}}\colon X^{\text{an}}\rightarrow X^{\text{an}} is idempotent, i.e. 𝐠𝐠=𝐠{\mathbf{g}}\ast{\mathbf{g}}={\mathbf{g}}.

Proof.

One has that (𝐠)=k(G){\mathcal{H}}({\mathbf{g}})=k(G) and the norm on k(G)k(G) defined by 𝐠{\mathbf{g}} is the trivial valuation so we have that for f(𝐠)k(𝐠)f\in{\mathcal{H}}({\mathbf{g}})\otimes_{k}{\mathcal{H}}({\mathbf{g}}):

f𝐠,𝐠=inf{maxi||gi||𝐠fi𝐠|f=igifi}|\!|f|\!|_{{\mathbf{g}},{\mathbf{g}}}=\inf\{\max_{i}|\!|g_{i}|\!|_{{\mathbf{g}}}|\!|f_{i}|\!|_{{\mathbf{g}}}\,|\,f=\sum\limits_{i}g_{i}\otimes f_{i}\}

which is 0 or 1, and 0 if and only if f=0f=0. Therefore it is the trivial norm. so 𝐠𝐠=𝐠{\mathbf{g}}\ast{\mathbf{g}}={\mathbf{g}}. So by Proposition 3.5 𝐩{\mathbf{p}} is idempotent. ∎

Corollary 3.9.

The map 𝐩:XanXan{\mathbf{p}}\colon X^{\text{an}}\rightarrow X^{\text{an}} is a retraction of topological spaces.

Proposition 3.10.

Let p=(x,||||)Xanp=(x,|\!|\!\cdot\!|\!|)\in X^{\text{an}}, then 𝐩(p)=(ηY,||||¯){\mathbf{p}}(p)=(\eta_{Y},\overline{|\!|\!\cdot\!|\!|}). Where ηY\eta_{Y} is the generic point of YY, YY is the Zariski closure of the image of the composition:

G×Spec(k(x))G×X𝑚XG\times\operatorname{Spec}(k(x))\hookrightarrow G\times X\xrightarrow{m}X

and ||||¯\overline{|\!|\!\cdot\!|\!|} is a G(k)G(k)-invariant valuation on k(Y)k(Y).

The proof is similar to the proofs of [Thu07, Proposition 2.5 (2)] and [Thu07, Proposition 2.3 (ii)], which prove the above for G=TG=T a torus.

Proof.

Recall that there is a map ρ:XanX\rho:X^{\text{an}}\rightarrow X given by (x,||||)x(x,|\!|\!\cdot\!|\!|)\mapsto x. To see that ρ(𝐩(p))=ηY\rho({\mathbf{p}}(p))=\eta_{Y} notice that there is a commutative diagram:

((𝐠))×((p)){{\mathcal{M}}({\mathcal{H}}({\mathbf{g}}))\times{\mathcal{M}}({\mathcal{H}}(p))}Speck(ηG)×Speck(x){\operatorname{Spec}k(\eta_{G})\times\operatorname{Spec}k(x)}Gan×Xan{G^{\text{an}}\times X^{\text{an}}}G×X{G\times X}Xan{X^{\text{an}}}X{X}ρ\scriptstyle{\rho}man\scriptstyle{m^{\text{an}}}ρ\scriptstyle{\rho}m\scriptstyle{m}ρ\scriptstyle{\rho}

where ηG\eta_{G} is the generic point of GG. In this diagram the image of the generic point of Speck(ηG)×Speck(x)\operatorname{Spec}k(\eta_{G})\times\operatorname{Spec}k(x) in XX is ηY\eta_{Y}. But then because ρ(||||𝐠,p)\rho(|\!|\!\cdot\!|\!|_{{\mathbf{g}},p}) is the generic point of Speck(ηG)×Speck(x)\operatorname{Spec}k(\eta_{G})\times\operatorname{Spec}k(x) we have the first claim.

To see that ||||¯\overline{|\!|\!\cdot\!|\!|} is G(k)G(k)-invariant notice that for any gG(k)g\in G(k), g𝐠=𝐠g\ast{\mathbf{g}}={\mathbf{g}}. Thus by Proposition 3.5 part 2 we have that g(𝐠p)=𝐠pg\ast({\mathbf{g}}\ast p)={\mathbf{g}}\ast p. But pgpp\mapsto g\ast p is the map induced by left multiplication by gg, which for points of the form (ηY,||||)(\eta_{Y},|\!|\!\cdot\!|\!|) is given by precomposing |||||\!|\!\cdot\!|\!| with the left-multiplication by gg. ∎

Furthermore we have the following.

Proposition 3.11.

The image of 𝐩{\mathbf{p}} is exactly the set of points (ηY,||||)(\eta_{Y},|\!|\!\cdot\!|\!|), where YY is a GG-invariant subvariety of XX and |||||\!|\!\cdot\!|\!| is G(k)G(k)-invariant.

This proof is the same as the proof of [Thu07, Proposition 2.3 (ii)], which proves this fact for G=TG=T a torus. We include the argument, except for one lemma, for completeness.

Proof.

By Proposition 3.10 we have one containment. For the other notice that points of the form (ηY,||||)(\eta_{Y},|\!|\!\cdot\!|\!|), where YY is a GG-invariant subvariety and |||||\!|\!\cdot\!|\!| is G(k)G(k)-invariant, are exactly the points (x,||||)Xan(x,|\!|\!\cdot\!|\!|)\in X^{\text{an}} fixed by the action of G(k)G(k). Let p=(x,||||)Xanp=(x,|\!|\!\cdot\!|\!|)\in X^{\text{an}} be a point fixed by the group action G(k)G(k). The morphism:

G×((p))G×XanmanXanG^{\beth}\times{\mathcal{M}}({\mathcal{H}}(p))\hookrightarrow G^{\beth}\times X^{\text{an}}\xrightarrow{m^{\text{an}}}X^{\text{an}}

maps the subset G(k)×((p))G^{\beth}(k)\times{\mathcal{M}}({\mathcal{H}}(p)) onto xx. The map is continuous and xx is closed as Berkovich spaces are locally hausdorff and thus T1, and so every point in the closure of G(k)×((p))G^{\beth}(k)\times{\mathcal{M}}({\mathcal{H}}(p)) will have image pp. Because 𝐩(p){\mathbf{p}}(p) is the image in XanX^{\text{an}} of the unique Shilov boundary point of G×((p))G^{\beth}\times{\mathcal{M}}({\mathcal{H}}(p)) under

G×((p))G×XanmanXanG^{\beth}\times{\mathcal{M}}({\mathcal{H}}(p))\hookrightarrow G^{\beth}\times X^{\text{an}}\xrightarrow{m^{\text{an}}}X^{\text{an}}

we need to show that the unique point in the Shilov boundary of G×((p))G^{\beth}\times{\mathcal{M}}({\mathcal{H}}(p)) is in the closure of G(k)×((p))G^{\beth}(k)\times{\mathcal{M}}({\mathcal{H}}(p)). The field kk is algebraically closed so G(k)×Spec(p)~G(k)\times\operatorname{Spec}\widetilde{{\mathcal{H}}(p)} is a dense subset of G×Spec(p)~G\times\operatorname{Spec}\widetilde{{\mathcal{H}}(p)}. The following lemma will complete the proof. ∎

Lemma 3.12.

Let L/kL/k be a nonarchimedean field extension, and let Σ\Sigma be a subset of G×k(L)G^{\beth}\times_{k}{\mathcal{M}}(L) with r(Σ)r(\Sigma) dense in G×kSpec(L~)G\times_{k}\operatorname{Spec}(\tilde{L}). Then the Shilov boundary is in the closure of Σ\Sigma.

Proof.

The case when GG is a torus is proved in [Thu07, lemma 2.4], and the proof proceeds the same for any connected linear algebraic group GG. ∎

3.2   The action of 𝐩{\mathbf{p}} on homogeneous spaces

If GG is reductive and XX is a spherical GG-variety, then any GG-invariant subvariety is in fact the closure of the GG-orbit of some closed point yXy\in X ([Kno91]). Thus we will restrict our attention to when GG acts on an orbit. In this case we will give a formula for 𝐩{\mathbf{p}} in terms of valuations that will immediately relate it to the tropicalization map. We begin by recalling the construction of Luna and Vust which is used in [TV21] to define tropicaliziation.

Let K¯=k{{u}}\overline{K}=k\{\!\{u\}\!\} denote Puiseux series over kk, equipped with the uu-adic valuation ||||u|\!|\!\cdot\!|\!|_{u}. In the case that G/HG/H is spherical for any K¯\overline{K}-point xx, Luna and Vust define a G(k)G(k)-invariant valuation ||||x|\!|\!\cdot\!|\!|_{x} ([LV83, Section 4]). For any fk(G/H)f\in k(G/H) there is a nonempty open subset of UG(k)U\subseteq G(k) such that if gUg\in U then gf(x)gf(x) is well defined. Luna and Vust show that for any fk(G/H)f\in k(G/H) there is a dense open UfG(k)U_{f}\subseteq G(k) where gf(x)u\left|\!\left|{gf(x)}\right|\!\right|_{u} is constant as gg varies over UfU_{f}. The valuation ||||x|\!|\!\cdot\!|\!|_{x} is given by fx=gf(x)u\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{gf(x)}\right|\!\right|_{u} for gUfg\in U_{f}.

We will now explain how this generalizes to other valued extensions L/kL/k. Let x:SpecLGx\colon\operatorname{Spec}L\rightarrow G and denote the valuation on LL by |||||\!|\!\cdot\!|\!|. The proof of the following lemma uses the same technique as the proof of [Kno91, Lemma 1.4].

Lemma 3.13.

For each fk(G)f\in k(G) there is a nonempty Zariski open UfUU_{f}\subseteq U such that for any gUfg\in U_{f}, gf(x)\left|\!\left|{gf(x)}\right|\!\right| is constant. For fk[G]f\in k[G] we have gf(x)=maxhG(k)hf(x)\left|\!\left|{gf(x)}\right|\!\right|=\max_{h\in G(k)}\left|\!\left|{hf(x)}\right|\!\right| when gUfg\in U_{f}.

Proof.

Without loss of generality we can assume fk[G]f\in k[G]. Let MM be the GG-module generated by ff, this is a finite dimensional vector space and thus has basis g1f,grfg_{1}f,\ldots g_{r}f. Let v=maxigif(x)v=\max_{i}\left|\!\left|{g_{i}f(x)}\right|\!\right|. Let V(v)={fk[G]|f(x)<v}V(v)=\{f^{\prime}\in k[G]\,|\,\left|\!\left|{f^{\prime}(x)}\right|\!\right|<v\}. Then set Uf={gG(k)|gfV(v)M}U_{f}=\{g\in G(k)\,|\,gf\notin V(v)\cap M\}. ∎

Let ||||x:k(G)(0,)|\!|\!\cdot\!|\!|_{x}:k(G)\rightarrow(0,\infty) be given by:

fgf(x)gUf.f\mapsto\left|\!\left|{gf(x)}\right|\!\right|\quad g\in U_{f}. (1)
Corollary 3.14.

The function ||||x|\!|\!\cdot\!|\!|_{x} defines a G(k)G(k)-invariant valuation on k(G)k(G).

Lemma 3.15.

Let x:SpecLGx:\operatorname{Spec}L\rightarrow G be as above and let p=(x,||||)p=(x,|\!|\!\cdot\!|\!|) be the corresponding point in GanG^{\text{an}}. Then if 𝐩(p)=(ηG,||||¯){\mathbf{p}}(p)=(\eta_{G},\overline{|\!|\!\cdot\!|\!|}), we have that ||||¯=||||x\overline{|\!|\!\cdot\!|\!|}=|\!|\!\cdot\!|\!|_{x}.

Proof.

We will show that ||||x|\!|\!\cdot\!|\!|_{x} and ||||¯\overline{|\!|\!\cdot\!|\!|} agree for all regular functions fk[G]f\in k[G]. Let mxm_{x} be the composition G×Speck(x)G×G𝑚GG\times\operatorname{Spec}k(x)\hookrightarrow G\times G\xrightarrow{m}G. Then if mxf=igifi(x)k[G]k(x)m_{x}^{*}f=\sum_{i}g_{i}\otimes f_{i}(x)\in k[G]\otimes k(x) we have ||f||x=||gf(x)||=||igi(g)fi(x)||maxi(||fi(x))||)\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{gf(x)}\right|\!\right|=\left|\!\left|{\sum_{i}g_{i}(g)\otimes f_{i}(x)}\right|\!\right|\leq\max_{i}(\left|\!\left|{f_{i}(x))}\right|\!\right|), where gUfg\in U_{f}, UfU_{f} the open subset of G(k)G(k) from Lemma 3.13. But then f¯=mxf𝐠,p\overline{\left|\!\left|{f}\right|\!\right|}=\left|\!\left|{m_{x}^{*}f}\right|\!\right|_{{\mathbf{g}},p} and

mxf𝐠,p=inf{maxi||gi||𝐠fi(x)p|igifi(x)=mxf}.\left|\!\left|{m_{x}^{*}f}\right|\!\right|_{{\mathbf{g}},p}=\inf\left\{\max_{i}\left|\!\left|{g_{i}}\right|\!\right|_{{\mathbf{g}}}\left|\!\left|{f_{i}(x)}\right|\!\right|_{p}\,\,\bigg{|}\,\,\sum\limits_{i}g_{i}\otimes f_{i}(x)=m_{x}^{*}f\right\}.

Because ||||𝐠|\!|\!\cdot\!|\!|_{{\mathbf{g}}} is trivial we have that fxf¯\left|\!\left|{f}\right|\!\right|_{x}\leq\overline{\left|\!\left|{f}\right|\!\right|}.

Conversely, to show that fxf¯\left|\!\left|{f}\right|\!\right|_{x}\geq\overline{\left|\!\left|{f}\right|\!\right|} it suffices to show that there is one expression mxf=igifi(x)m_{x}^{*}f=\sum_{i}g_{i}\otimes f_{i}(x) such that fx=maxifi(x)\left|\!\left|{f}\right|\!\right|_{x}=\max_{i}\left|\!\left|{f_{i}(x)}\right|\!\right|. Let mxf=igifi(x)m_{x}^{*}f=\sum_{i}g_{i}\otimes f_{i}(x), by manipulating the sum we can assume all the gig_{i} and fi(x)f_{i}(x) are nonzero and the functions gig_{i} are kk-linearly independent. We can also assume that maxi(fi(x))=f1(x)\max_{i}(\left|\!\left|{f_{i}(x)}\right|\!\right|)=\left|\!\left|{f_{1}(x)}\right|\!\right|. Let VV be the kk-span in LL of fi(x)f_{i}(x). Fix α>1\alpha>1. It follows from [BGR84, Subsection 2.6.2, Proposition 3]333While [BGR84] works with kk nontrivially valued, this result holds for trivially valued kk as well, the proof is effectively the same. Such bases are called α\alpha-Cartesian bases. there is a kk-basis of VV, v1,vrv_{1},\ldots v_{r} such that v1v_{1} is a multiple of f1(x)f_{1}(x), and for any λ1,λrk\lambda_{1},\ldots\lambda_{r}\in k

||jλjvj||α1maxj(||vj||).\left|\!\left|{\sum\limits_{j}\lambda_{j}v_{j}}\right|\!\right|\geq\alpha^{-1}\max_{j}(\left|\!\left|{v_{j}}\right|\!\right|).

Then we have for each ii, fi(x)=jλijvjf_{i}(x)=\sum_{j}\lambda^{i}_{j}v_{j}. So then mxf=j(λijgi)vjm_{x}^{*}f=\sum_{j}(\sum\limits\lambda^{i}_{j}g_{i})\otimes v_{j}, by assumption for each jj, the function iλijgik[G]\sum_{i}\lambda^{i}_{j}g_{i}\in k[G] is not identically 0. Take gUfg\in U_{f} and such that iλi1gi(g)0\sum_{i}\lambda^{i}_{1}g_{i}(g)\neq 0, which exists because UfU_{f} is nonempty and open and so is the set of points such that iλi1gi(g)0\sum_{i}\lambda^{i}_{1}g_{i}(g)\neq 0. Now we have that:

||f||x=||j(iλijgi(g))vj||α1maxj(||vj||)=α1||f1(x)||.\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{\sum\limits_{j}(\sum\limits_{i}\lambda^{i}_{j}g_{i}(g))\otimes v_{j}}\right|\!\right|\geq\alpha^{-1}\max_{j}(\left|\!\left|{v_{j}}\right|\!\right|)=\alpha^{-1}\left|\!\left|{f_{1}(x)}\right|\!\right|.

But this holds for all α>1\alpha>1. So ||f||x=||f1(x)||=maxi(||fi(x))||)\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{f_{1}(x)}\right|\!\right|=\max_{i}(\left|\!\left|{f_{i}(x))}\right|\!\right|). So ||f||x||f||¯\left|\!\left|{f}\right|\!\right|_{x}\geq\overline{\left|\!\left|{f}\right|\!\right|}, and thus the valuations are equal on k[G]k[G], and thus are equal on all of k(G)k(G). ∎

Let YY be the GG-orbit in XX of some closed point yy. The morphism GYG\rightarrow Y given by ggyg\mapsto gy defines an inclusion k(Y)k(G)k(Y)\rightarrow k(G). Given some valued extension L/kL/k and xG(L)x\in G(L), one can restrict ||||x|\!|\!\cdot\!|\!|_{x} to k(Y)k(Y). Let xx^{\prime} be the image of xx under the map G(L)Y(L)G(L)\rightarrow Y(L). Then ||||x|\!|\!\cdot\!|\!|_{x} is in fact given by

||f||x=||gf(x)||gUf.\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{gf(x^{\prime})}\right|\!\right|\quad g\in U_{f}. (2)

We denote the restriction of ||||x|\!|\!\cdot\!|\!|_{x} to k(Y)k(Y) by ||||x|\!|\!\cdot\!|\!|_{x^{\prime}}.

Corollary 3.16.

The map 𝐩:YanYan{\mathbf{p}}\colon Y^{\text{an}}\rightarrow Y^{\text{an}} is given by (x,||||)(ηY,||||x)(x,|\!|\!\cdot\!|\!|)\mapsto(\eta_{Y},|\!|\!\cdot\!|\!|_{x^{\prime}}). Furthermore all the G(k)G(k)-invariant valuations on k(Y)k(Y) have the form ||||x|\!|\!\cdot\!|\!|_{x^{\prime}} for some xYx^{\prime}\in Y.

Proof.

We have a surjection of sets GanYanG^{\text{an}}\rightarrow Y^{\text{an}} and a commutative diagram of maps of analytic spaces:

Gan×Gan{G^{\text{an}}\times G^{\text{an}}}Gan{G^{\text{an}}}Gan×Yan{G^{\text{an}}\times Y^{\text{an}}}Yan{Y^{\text{an}}}

Thus by Proposition 3.5 part 3 we see that 𝐩((x,||||))=(ηY,||||x){\mathbf{p}}((x,|\!|\!\cdot\!|\!|))=(\eta_{Y},|\!|\!\cdot\!|\!|_{x^{\prime}}).

To see that this is all G(k)G(k)-invariant valuations on k(Y)k(Y), notice that it follows from Proposition 3.11 that all G(k)G(k)-invariant valuations on k(G)k(G) are of the form ||||x|\!|\!\cdot\!|\!|_{x} for some xX(L)x\in X(L) and it follows from [Kno91, Corollary 1.5] that any G(k)G(k)-invariant valuation on k(Y)k(Y) is the restriction of one on k(G)k(G). ∎

3.3   A strong deformation retraction when GG is reductive

Assume GG is a reductive group. We construct a strong deformation retraction from XanX^{\text{an}} onto 𝐩(Xan){\mathbf{p}}(X^{\text{an}}), using the same method as the proof of [Ber90, Theorem 6.2.1]. Fix a maximal torus TT in GG, Borel subgroup B+B_{+} containing TT, and let BB_{-} be the opposite Borel i.e. the unique Borel subgroup of GG such that B+B=TB_{+}\cap B_{-}=T. Let U+U_{+} be the unipotent radical of B+B_{+}, and let UU_{-} be the unipotent radical of BB_{-}. Then the map:

U×T×U+GU_{-}\times T\times U_{+}\rightarrow G

given by the multiplication map, is in fact an open immersion [GP11, Exposé XXII]. Denote the image by Ω\Omega. The groups UU_{-} and U+U_{+} are isomorphic affine NN-space for some NN. Let U+=Speck[u1,uN]U_{+}=\operatorname{Spec}k[u_{1},\ldots u_{N}], and U=Speck[u1,uN]U_{-}=\operatorname{Spec}k[u_{-1},\ldots u_{-N}], and let T=Spec(k[t1±1,tn±1)]T=\operatorname{Spec}(k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1})]. Then the coordinate ring of Ω\Omega, k[Ω]k[\Omega], is given by Speck[ti±1,uj]\operatorname{Spec}k[t_{i}^{\pm 1},u_{j}]. Then set Ti=ti1T_{i}=t_{i}-1. For each λ[0,1]\lambda\in[0,1] there is a point 𝐠λk[Ω]{\mathbf{g}}_{\lambda}\in k[\Omega]^{\beth} given by the semivaluation ||||𝐠λ|\!|\!\cdot\!|\!|_{{\mathbf{g}}_{\lambda}}, which is defined by:

||I,JαI,JTIuJ||𝐠λ=minI,J||αI,J||λ|I|+|J|\left|\!\left|{\sum_{I,J}\alpha_{I,J}T^{I}u^{J}}\right|\!\right|_{{\mathbf{g}}_{\lambda}}=\min_{I,J}\left|\!\left|{\alpha_{I,J}}\right|\!\right|\lambda^{|I|+|J|}

where II and JJ denote multi-indices.Then 𝐠1=𝐠{\mathbf{g}}_{1}={\mathbf{g}} and 𝐠0=e{\mathbf{g}}_{0}=e, the identity of GG. Thus we have a map of sets:

H:[0,1]×XanXanH\colon[0,1]\times X^{\text{an}}\rightarrow X^{\text{an}}

(λ,p)𝐠λp(\lambda,p)\mapsto{\mathbf{g}}_{\lambda}\ast p. Furthermore H(0,p)=pH(0,p)=p and H(1,p)=𝐩(Xan)H(1,p)={\mathbf{p}}(X^{\text{an}}). Notice that because 𝐠λX{\mathbf{g}}_{\lambda}\in X^{\beth} for any λ[0,1]\lambda\in[0,1], we can restrict HH to a map [0,1]×XX[0,1]\times X^{\beth}\rightarrow X^{\beth}, with H(0,p)=pH(0,p)=p and H(1,p)=𝐩(p)H(1,p)={\mathbf{p}}(p).

Proposition 3.17.

The map HH is a homotopy, and in fact a strong deformation retraction of XanX^{\text{an}} onto 𝐩(Xan){\mathbf{p}}(X^{\text{an}}). Similarly HH defines a strong deformation retraction from XX^{\beth} onto 𝐩(X){\mathbf{p}}(X^{\beth}).

Proof.

That HH is continuous follows directly from [Ber90, Corollary 6.1.2]. It follows from Proposition 3.5 part 2 that HH is strong deformation retract because we have that 𝐠λ𝐠1=𝐠1{\mathbf{g}}_{\lambda}\ast{\mathbf{g}}_{1}={\mathbf{g}}_{1}. Indeed, for fk[Ω]f\in k[\Omega] we will show that ||mf||𝐠λ,𝐠1=1\left|\!\left|{m^{\ast}f}\right|\!\right|_{{\mathbf{g}}_{\lambda},{\mathbf{g}}_{1}}=1. Let mf=aI,JL,MtIuJtLuMm^{\ast}f=\sum a_{I,J}^{L,M}t^{I}u^{J}\otimes t^{L}u^{M}. Without loss of generality we can assume that aI,JL,M=0a_{I,J}^{L,M}=0 whenever II or LL is less than 0, because otherwise we can factor out a power of tRtSt^{R}\otimes t^{S}, and ||tRtS||𝐠λ,𝐠1=1|\!|t^{R}\otimes t^{S}|\!|_{{\mathbf{g}}_{\lambda},{\mathbf{g}}_{1}}=1. But then we can rewrite the sum by substituting TiT_{i} for ti1t_{i}-1, and we have that mf=bI,JL,MTIuJTLuMm^{\ast}f=\sum b_{I,J}^{L,M}T^{I}u^{J}\otimes T^{L}u^{M}. Note that TIT^{I} and uJu^{J} both evaluate to 0 on the identity. If f0f\neq 0 then e×idmf0e^{\ast}\times\text{id}\circ m^{\ast}f\neq 0, so there must be some term b0,0L,M0b_{0,0}^{L,M}\neq 0. But then:

||mf||𝐠λ,𝐠1=||bI,JL,MTIuJTLuM||𝐠λ,𝐠1=maxI,J,L,M(||bI,JL,M||)(λ|I|+|J|)=1\left|\!\left|{m^{\ast}f}\right|\!\right|_{{\mathbf{g}}_{\lambda},{\mathbf{g}}_{1}}=\left|\!\left|{\sum b_{I,J}^{L,M}T^{I}u^{J}\otimes T^{L}u^{M}}\right|\!\right|_{{\mathbf{g}}_{\lambda},{\mathbf{g}}_{1}}=\max_{I,J,L,M}(\left|\!\left|{b_{I,J}^{L,M}}\right|\!\right|)(\lambda^{|I|+|J|})=1

as the maximum is achieved when |I|+|J|=0|I|+|J|=0.

4   The Luna-Vust Classification of Spherical varieties

Here we review the classification of embeddings of spherical homogeneous spaces via colored fans. The classification was first given by Luna and Vust in [LV83], though we will follow the exposition of [Kno91]. Much of the notation for spherical varieties is different than that of toric varieties, we will attempt to align our notation with that of toric varieties, similar to the notation in [Nas17].

Let kk be an algebraically closed and trivially valued field444In the literature on spherical varieties one does not generally specify a valuation on kk, but it is assumed the valuations on k(X)k(X) are trivial when restricted to kk. and let GG be a reductive group over kk. Let HH be a closed subgroup such that G/HG/H has an open BB-orbit, for some Borel subgroup BGB\subseteq G. A G/HG/H-embedding is a normal GG-variety XX with a GG-equivariant open embedding G/HXG/H\hookrightarrow X. Note that if XX is a spherical variety with open BB-orbit B0B_{0}, then it is a G0G_{0}-embedding where G0G_{0} is the open GG-orbit GB0GB_{0}. The Luna-Vust theory is a generalization of the toric case which puts embeddings of a torus in bijection with fans. Here G/HG/H-embeddings are put in bijection with colored fans. To define colored fans and state the classification we will need to introduce some data associated to G/HG/H, primarily a lattice MM, its dual lattice NN, a convex cone 𝒱{\mathcal{V}} in NN\otimes_{{\mathbb{Z}}}{\mathbb{R}}, and a finite set of colors 𝒟(G/H){\mathcal{D}}(G/H). We refer the reader to Example 4.8 for an explicit demonstration of the theory.

4.1   Data associated to G/HG/H

Let 𝔛(B){\mathfrak{X}}(B) be the character lattice of BB and define:

k(G/H)(B)={fk(G/H)×|there exists χf𝔛(B) such that for any bB(k), bf=χf(b)f}.k(G/H)^{(B)}=\{f\in k(G/H)^{\times}\,|\,\text{there exists $\chi_{f}\in{\mathfrak{X}}(B)$ such that for any $b\in B(k),$ }b\cdot f=\chi_{f}(b)f\}.

We say elements of k(G/H)(B)k(G/H)^{(B)} are BB semi-invariant. There is a map k(G/H)(B)𝔛(B)k(G/H)^{(B)}\rightarrow{\mathfrak{X}}(B) given by:

fχff\mapsto\chi_{f}

and the kernel is exactly k×k^{\times}. Let the image of this map be MM, set N=hom(M,)N=\hom(M,{\mathbb{Z}}), M=NM_{{\mathbb{R}}}=N\otimes_{{\mathbb{Z}}}{\mathbb{R}}, and N=NN_{{\mathbb{R}}}=N\otimes_{{\mathbb{Z}}}{\mathbb{R}}. We say MM is the weight lattice of G/HG/H. Notice that every valuation |||||\!|\!\cdot\!|\!| on k(G/H)k(G/H) defines an element ϱ(||||)N\varrho(|\!|\!\cdot\!|\!|)\in N_{{\mathbb{R}}} by:

ϱ(||||)(χf)=log(||f||).\varrho(|\!|\!\cdot\!|\!|)(\chi_{f})=-\log(\left|\!\left|{f}\right|\!\right|).

In particular if 𝒱{\mathcal{V}} is the set of G(k)G(k)-invariant valuations on k(G/H)k(G/H) then the map ϱ\varrho restricted to 𝒱{\mathcal{V}} is an injection by [Kno91, Corollary 1.8] and the image is a finitely generated convex cone [Kno91, Corollary 5.3]. In general though the map ϱ\varrho is not injective: if ||||1|\!|\!\cdot\!|\!|_{1} and ||||2|\!|\!\cdot\!|\!|_{2} are distinct valuations, but not G(k)G(k)-invariant, then it is possible that ϱ(||||1)=ϱ(||||2)\varrho(|\!|\!\cdot\!|\!|_{1})=\varrho(|\!|\!\cdot\!|\!|_{2}). We say 𝒱{\mathcal{V}} is the valuation cone of G/HG/H and identify it with its image in NN_{{\mathbb{R}}}. The cone 𝒱{\mathcal{V}} spans NN_{{\mathbb{R}}} [Kno91, Corollary 5.3] and is cosimplicial, i.e. there exist linearly independent elements χ1,χr\chi_{1},\ldots\chi_{r} in the dual space NN_{{\mathbb{R}}}^{\ast} such that 𝒱={vN|χi(v)0, for all i}{\mathcal{V}}=\{v\in N_{{\mathbb{R}}}\,\,|\,\,\chi_{i}(v)\geq 0,\text{ for all $i$}\} [Kno91, Theorem 5.4].

Now let XX be a G/HG/H-embedding, and let YY be a GG-orbit. Then define:

𝒟(X):={D|D a B-stable prime divisor in X}\displaystyle{\mathcal{D}}(X):=\{D\,|\,\text{$D$ a $B$-stable prime divisor in $X$}\}
𝒟Y(X):={D|D a B-stable prime divisor in X and DY}\displaystyle{\mathcal{D}}_{Y}(X):=\{D\,|\,\text{$D$ a $B$-stable prime divisor in $X$ and $D\supseteq Y$}\}

We say that the colors of G/HG/H are 𝒟(G/H){\mathcal{D}}(G/H). Notice that every prime divisor DD, in particular every color, defines a valuation on k(G/H)k(G/H), which we denote ||||D|\!|\!\cdot\!|\!|_{D}. These valuations then define an element of NN_{{\mathbb{R}}}, ϱ(||||D)\varrho(|\!|\!\cdot\!|\!|_{D}), so in particular each color defines an element of NN_{{\mathbb{R}}}. For convenience we may write ϱ(D)\varrho(D) when we mean ϱ(||||D)\varrho(|\!|\!\cdot\!|\!|_{D}).

When we say ‘the data associated to G/HG/H’ we mean the vector space NN_{{\mathbb{R}}}, the valuation cone 𝒱{\mathcal{V}} in NN_{{\mathbb{R}}}, and the set of colors 𝒟(G/H){\mathcal{D}}(G/H).

4.2   Colored fans

We can now define colored fans and colored cones. By a rational point of NN_{{\mathbb{R}}} we mean a point in NN\otimes_{{\mathbb{Z}}}{\mathbb{Q}}.

Definition 4.1.

A colored cone is a pair (σ,)(\sigma,{\mathcal{F}}) where σ\sigma is a strictly convex cone in NN_{{\mathbb{R}}}, and {\mathcal{F}} is a subset of 𝒟(G/H){\mathcal{D}}(G/H), such that the following are satisfied:

  1. CC1

    σ\sigma is generated by ϱ()\varrho({\mathcal{F}}) and finitely many rational points of 𝒱{\mathcal{V}}

  2. CC2

    The relative interior of σ\sigma intersects 𝒱{\mathcal{V}}

  3. CC3

    0ϱ()0\notin\varrho({\mathcal{F}}).

We say (σ,)(\sigma^{\prime},{\mathcal{F}}^{\prime}) is a colored face of a colored cone (σ,)(\sigma,{\mathcal{F}}) if σ\sigma^{\prime} is a face of the cone σ\sigma, the relative interior of σ\sigma^{\prime} intersects 𝒱{\mathcal{V}}, and =ϱ1(σ){\mathcal{F}}^{\prime}={\mathcal{F}}\cap\ \varrho^{-1}(\sigma). A colored fan, 𝔉{\mathfrak{F}}, is a finite set of colored cones containing all the colored faces of any colored cone in 𝔉{\mathfrak{F}}, and for any ||||𝒱|\!|\!\cdot\!|\!|\in{\mathcal{V}} there is at most one colored cone (σ,)𝔉(\sigma,{\mathcal{F}})\in{\mathfrak{F}} with |||||\!|\!\cdot\!|\!| in the relative interior of σ\sigma.

We associate a colored cone to each GG-orbit in XX, YY, as follows. First define:

Y(X)={D𝒟Y(X)|D is G-stable}\displaystyle{\mathcal{B}}_{Y}(X)=\{D\in{\mathcal{D}}_{Y}(X)\,|\,D\text{ is $G$-stable}\}
Y(X)={D𝒟Y(X)|D is not G-stable}\displaystyle{\mathcal{F}}_{Y}(X)=\{D\in{\mathcal{D}}_{Y}(X)\,|\,D\text{ is not $G$-stable}\}

and let σY(X)\sigma_{Y}(X) be the cone in NN_{{\mathbb{R}}} generated by ρ(Y(X))\rho({\mathcal{B}}_{Y}(X)) and ρ(Y(X))\rho({\mathcal{F}}_{Y}(X)). We identify the elements of Y(X){\mathcal{F}}_{Y}(X) with elements of 𝒟(G/H){\mathcal{D}}(G/H) by intersecting each element DD with G/HG/H, as DG/HD\cap G/H is dense in DD when DD isn’t GG-stable. Then we get a colored cone (σY(X),Y(X))(\sigma_{Y}(X),{\mathcal{F}}_{Y}(X)). Just as affine toric varieties correspond to cones, we have a particular type of G/HG/H-embedding corresponding to colored cones. A G/HG/H-embedding is simple if it has a unique closed GG-orbit.

Theorem 4.2.

[Kno91, Theorem 3.1] The map

X(σY(X),Y(X))X\mapsto(\sigma_{Y}(X),{\mathcal{F}}_{Y}(X))

gives a bijection between isomorphism classes of simple G/HG/H-embeddings and colored cones.

We can construct all simple subembeddings of XX in the following way. Let YY be a GG-orbit in XX, then define:

X0:=X(D𝒟(X)𝒟Y(X)D).X_{0}:=X\setminus\left(\bigcup_{D\in{\mathcal{D}}(X)\setminus{\mathcal{D}}_{Y}(X)}D\right). (3)

Note that 𝒟(X)𝒟Y(X){\mathcal{D}}(X)\setminus{\mathcal{D}}_{Y}(X) consists of all divisors DD such that ||||D|\!|\!\cdot\!|\!|_{D} is not an element of σY(X)\sigma_{Y}(X).

Theorem 4.3.

[Kno91, Theorem 2.1] The open subvariety X0X_{0} is a BB-stable, affine open subset of XX. The intersection YX0Y\cap X_{0} is an open BB-orbit, and GX0GX_{0} is a simple embedding with closed orbit YY.

When XX is simple we have that GX0=XGX_{0}=X. The following theorem describes the cone σY(X)\sigma_{Y}(X) in this case.

Theorem 4.4.

[Kno91, Theorem 2.5] Let XX be a simple embedding with closed orbit YY, and let X0X_{0} be as in 3.

  1. 1.

    k[X0](B)={fk(G/H)(B)|χf(σY(X))}k[X_{0}]^{(B)}=\{f\in k(G/H)^{(B)}\,|\,\chi_{f}\in(\sigma_{Y}(X))^{\vee}\}.

  2. 2.

    The center of |||||\!|\!\cdot\!|\!| exists if and only if ||||σY(X)|\!|\!\cdot\!|\!|\in\sigma_{Y}(X).

  3. 3.

    The center of |||||\!|\!\cdot\!|\!| is YY if and only if |||||\!|\!\cdot\!|\!| is in the relative interior of σY(X)\sigma_{Y}(X).

Furthermore we have the following about the faces of σY(X)\sigma_{Y}(X) and the orbits of XX.

Lemma 4.5.

[Kno91, Lemma 3.2] The map

Z(σZ(X),Z(X))Z\mapsto(\sigma_{Z}(X),{\mathcal{F}}_{Z}(X))

gives a bijection between orbits ZZ whose closure contains YY and the colored faces of (σY(X),Y(X))(\sigma_{Y}(X),{\mathcal{F}}_{Y}(X)).

Due to the above, and the fact that there are finitely many GG-orbits in XX ([Kno91]), we have that the set 𝔉(X):={(σY(X),Y(X))|Y is a G-orbit}{\mathfrak{F}}(X):=\{(\sigma_{Y}(X),{\mathcal{F}}_{Y}(X))\,|\,\text{$Y$ is a $G$-orbit}\} is a colored fan. We can now state the main theorem in the classification.

Theorem 4.6.

[Kno91, Theorem 3.3] The map:

X𝔉(X)X\mapsto{\mathfrak{F}}(X)

is a bijection between isomorphism classes of G/HG/H-embeddings and colored fans.

Example 4.7.

The classification of toric varieties by fans is a special case of the Luna-Vust theory. Let G=TG=T be a torus, let HH be trivial. Then 𝒟(T)={\mathcal{D}}(T)=\emptyset, NN is the cocharacter lattice of TT, and TT-embeddings are classified by fans in NN_{{\mathbb{R}}}.

Example 4.8.

Let G=SL2G=\text{SL}_{2} and let HH be the subgroup of upper triangular matrices with 11s on the diagonal. Let OO denote the origin of 𝔸2{\mathbb{A}}^{2}, there is an isomorphism G/H𝔸2kOG/H\rightarrow{\mathbb{A}}^{2}_{k}\setminus O given by mapping a matrix to its first column. The complement of the xx-axis is an open orbit of the Borel subgroup of upper triangular matrices, BB. We have that 𝔛(B){\mathfrak{X}}(B)\cong{\mathbb{Z}}, where the generator is the character χ\chi given by:

[ab0a1]a.\left[{\begin{array}[]{cc}a&b\\ 0&a^{-1}\\ \end{array}}\right]\mapsto a.

Furthermore k(G/H)(B)/k×={yn|n}k(G/H)^{(B)}/k^{\times}=\{y^{n}\,|\,n\in{\mathbb{Z}}\} and χyn=χn\chi_{y^{n}}=\chi^{-n}, so M=𝔛(B)M={\mathfrak{X}}(B). Thus NN_{{\mathbb{R}}}\cong{\mathbb{R}}. Let DD be the xx-axis in 𝔸2{\mathbb{A}}^{2}. Consider the embedding 𝔸22{\mathbb{A}}^{2}\hookrightarrow{\mathbb{P}}^{2} given by (a,c)[a:c:1](a,c)\mapsto[a:c:1]. Extend the SL2\text{SL}_{2}-action to 2{\mathbb{P}}^{2} by letting SL2\text{SL}_{2} act on the first two homogeneous coordinates while preserving the third.

Let LL be the line at infinity in 2{\mathbb{P}}^{2}. Then both ||||D|\!|\!\cdot\!|\!|_{D} and ||||L|\!|\!\cdot\!|\!|_{L} are G(k)G(k)-invariant and furthermore ϱ(||||D)=1\varrho(|\!|\!\cdot\!|\!|_{D})=1 and ϱ(||||L)=1\varrho(|\!|\!\cdot\!|\!|_{L})=-1, thus 𝒱=N{\mathcal{V}}=N. There is one color, given by the xx-axis. The table below consists of a complete list of all G/HG/H-embeddings. The hollow circle denotes the color DD, and EE denotes the exceptional divisor in the blow ups at OO: Bl0𝔸2\text{Bl}_{0}{\mathbb{A}}^{2} and BlO2\text{Bl}_{O}{\mathbb{P}}^{2}.

XXClosed GG-orbits𝔉(X){\mathfrak{F}}(X)𝔸2O{\mathbb{A}}^{2}\setminus OBlO𝔸2\text{Bl}_{O}{\mathbb{A}}^{2}2O{\mathbb{P}}^{2}\setminus OBlO2\text{Bl}_{O}{\mathbb{P}}^{2}𝔸2{\mathbb{A}}^{2}2{\mathbb{P}}^{2}EELLL,EL,\,EOOL,OL,\,O

4.3   Orbits and Orbit Closures

Let ZZ be the GG-orbit of a closed point in XX. Notice that from Theorem 4.3 it follows that ZZ is spherical. Let Z¯\overline{Z} be the closure of ZZ in XX, so Z¯\overline{Z} is a ZZ-embedding. Let M(Z)M(Z) be the weight lattice of ZZ, N(Z)=hom(M(Z),)N_{{\mathbb{R}}}(Z)=\hom(M(Z),{\mathbb{R}}), 𝒱(Z){\mathcal{V}}(Z) be the valuation cone, and 𝒟(Z){\mathcal{D}}(Z) be the BB-stable prime divisors. We will relate the above to MM, NN_{{\mathbb{R}}}, 𝒱{\mathcal{V}}, and 𝒟(G/H){\mathcal{D}}(G/H). Furthermore we will give the relationship between the colored fan of XX and that of Z¯\overline{Z}. See [Kno91, Section 4] for further discussion of the following.

Given zZz\in Z there is a map φ:G/HZ\varphi\colon G/H\rightarrow Z given by gHgzgH\mapsto gz. This induces a map of function fields k(Z)k(G/H)k(Z)\hookrightarrow k(G/H). The map k(Z)k(G/H)k(Z)\hookrightarrow k(G/H) gives an inclusion M(Z)MM(Z)\hookrightarrow M and after dualizing, a surjection φ:NN(Z)\varphi_{*}\colon N_{{\mathbb{R}}}\rightarrow N_{{\mathbb{R}}}(Z). Let τ=σZ(X)\tau=\sigma_{Z}(X), N(τ)=N/span(τ)N(\tau)=N_{{\mathbb{R}}}/\operatorname{span}(\tau), and let M(τ)M(\tau) be the dual vector space to N(τ)N(\tau). Consider the following subset of 𝒟{\mathcal{D}}:

φ={D𝒟|D maps dominantly to Z}.{\mathcal{F}}_{\varphi}=\{D\in{\mathcal{D}}\,|\,\text{$D$ maps dominantly to $Z$}\}.

Let ϕc=𝒟ϕ{\mathcal{F}}_{\phi}^{\text{c}}={\mathcal{D}}\setminus{\mathcal{F}}_{\phi}.

Theorem 4.9.

[Kno91, Theorem 4.4] We have that N(τ)=N(Z)N(\tau)=N(Z), the map φ\varphi_{*} is the projection NN(τ)N\rightarrow N(\tau), M(τ)=M(Z)M(\tau)=M(Z), φ(𝒱)=𝒱(Z)\varphi_{*}({\mathcal{V}})={\mathcal{V}}(Z), and there is a bijection φc𝒟(Z){\mathcal{F}}_{\varphi}^{\text{c}}\rightarrow{\mathcal{D}}(Z) given by φ\varphi.

To state the relationship between 𝔉(X){\mathfrak{F}}(X) and 𝔉(Z¯){\mathfrak{F}}(\overline{Z}) define the following colored fan:

Star(τ):={(φ(σ),φ(φc))|(σ,)𝔉(X) and (τ,Z(X)) is a colored face of (σ,).}.\text{Star}(\tau):=\{(\varphi_{*}(\sigma),\varphi_{*}({\mathcal{F}}\cap{\mathcal{F}}_{\varphi}^{\text{c}}))\,|\,(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)\text{ and $(\tau,{\mathcal{F}}_{Z}(X))$ is a colored face of $(\sigma,{\mathcal{F}})$.}\}.

Then as a result of [Kno91, Theorem 4.5] we have that.

Theorem 4.10.

Let XX be a G/HG/H-embedding, and let ZXZ\subseteq X be a GG-orbit. Let the Zariski closure of ZZ in XX be Z¯\overline{Z}. Then Star(σZ(X))=𝔉(Z¯)\text{Star}(\sigma_{Z}(X))={\mathfrak{F}}(\overline{Z}).

5   Spherical Tropicalization

In this Section we review tropicalization for toric varieties, tropicalization of spherical varieties, and prove Theorem A. We also discuss applications of Theorem A. Let kk be an algebraically closed and trivially valued field, let GG be a reductive group over kk, let XX be a spherical GG-variety with open GG-orbit G/HG/H.

5.1   Tropicalization of toric varieties

Let K¯=k{{u}}\overline{K}=k\{\!\{u\}\!\} be the field of Puiseux series, equipped with the uu-adic valuation which we denote ||||u|\!|\!\cdot\!|\!|_{u}. Let T=Speck[t1±1,tn±1]T=\operatorname{Spec}k[t_{1}^{\pm 1},\ldots t_{n}^{\pm 1}] and let be WW a subvariety of TT. There is a map of sets tropT:W(K¯)n\operatorname{trop}_{T}\colon W(\overline{K})\rightarrow{\mathbb{R}}^{n} given by:

(t1,tn)(log(||t1||u),log(||tn||u)).(t_{1},\ldots t_{n})\mapsto(-\log(\left|\!\left|{t_{1}}\right|\!\right|_{u}),\ldots-\log(\left|\!\left|{t_{n}}\right|\!\right|_{u})).

We say the closure of tropT(W(K¯))\operatorname{trop}_{T}(W(\overline{K})) is the tropicalization of WW, denoted TropT(W)\operatorname{Trop}_{T}(W).

This construction extends to subvarieties of toric varieties, as follows. Let XX be a toric variety, with dense torus TT. Let NN be the cocharacter lattice of TT, and let F(X)F(X) be the fan in N=NN_{{\mathbb{R}}}=N\otimes_{{\mathbb{Z}}}{\mathbb{R}} associated to XX. Given a cone σF(X)\sigma\in F(X), consider the vector space N(σ)=N/span(σ)N(\sigma)=N_{{\mathbb{R}}}/\operatorname{span}(\sigma). As a set we define we the tropicalization of XX to be:

TropT(X):=σF(X)N(σ).\operatorname{Trop}_{T}(X):=\bigsqcup_{\sigma\in F(X)}N(\sigma).

To define a topology on TropT(X)\operatorname{Trop}_{T}(X) we first consider the case when XX is affine; so XX corresponds to a cone σF(X)\sigma\in F(X). Let MM_{{\mathbb{R}}} be the dual vector space to NN_{{\mathbb{R}}}; let σM\sigma^{\vee}\subseteq M_{{\mathbb{R}}} denote linear functions on NN_{{\mathbb{R}}} that are nonnegative on σ\sigma. Let ¯={}\overline{{\mathbb{R}}}={\mathbb{R}}\cup\{\infty\}. If hom(σ,¯)\hom(\sigma^{\vee},\overline{{\mathbb{R}}}) denotes semigroup homomorphisms σ¯\sigma^{\vee}\rightarrow\overline{{\mathbb{R}}}, then there is a natural bijection TropT(X)hom(σ,¯)\operatorname{Trop}_{T}(X)\rightarrow\hom(\sigma^{\vee},\overline{{\mathbb{R}}}) defined as follows. The fan F(X)F(X) consists of the faces of σ\sigma. Let τ\tau be a face of σ\sigma and let M(τ)M(\tau) be the dual of N(τ)N(\tau). Recall that M(τ)=τMM(\tau)=\tau^{\perp}\subseteq M_{{\mathbb{R}}}. So αN(τ)\alpha\in N(\tau) defines a semigroup homomorphism α~:σ¯\tilde{\alpha}\colon\sigma^{\vee}\rightarrow\overline{{\mathbb{R}}}

α~(v)={α(v),vτσ,else\tilde{\alpha}(v)=\left\{\begin{array}[]{cc}\alpha(v),&v\in\tau^{\perp}\cap\sigma^{\vee}\\ \infty,&\text{else}\\ \end{array}\right.

This then defines a bijection TropT(X)hom(σ,¯)\operatorname{Trop}_{T}(X)\rightarrow\hom(\sigma^{\vee},\overline{{\mathbb{R}}}) and we give TropT(X)\operatorname{Trop}_{T}(X) the topology hom(σ,¯)\hom(\sigma^{\vee},\overline{{\mathbb{R}}}) inherits from ¯σ\overline{{\mathbb{R}}}^{\sigma^{\vee}}. For a general toric variety XX we glue the tropicalizations of affine toric subvareities according to the shared faces of their respective cones in F(X)F(X). For further discussion of tropicalization of toric varieties and their subvarieties see [Pay09].

Recall that XX is a disjoint union of tori which are in bijection with the cones of F(X)F(X). Thus there is a tropicalization map X(K¯)TropT(X)X(\overline{K})\rightarrow\operatorname{Trop}_{T}(X) which is defined by taking a disjoint union of the tropicalization maps of the tori T(σ)N(σ)T(\sigma)\rightarrow N(\sigma), where T(σ)T(\sigma) is the torus in XX corresponding to σF(X)\sigma\in F(X). Thus for any subvariety WW of a toric variety XX we have a tropicalization map from W(K¯)W(\overline{K}) into TropT(X)\operatorname{Trop}_{T}(X).

5.2   Tropicalization of spherical varieties

In [TV21] Tevelev and Vogiannou define a tropicalization map tropG:G/H(K¯)𝒱\operatorname{trop}_{G}\colon G/H(\overline{K})\rightarrow{\mathcal{V}}. The map is given as follows. For each xG/H(K¯)x\in G/H(\overline{K}) define ||||x|\!|\!\cdot\!|\!|_{x} to be the valuation given by:

||f||x=||gf(x)||ugUf\left|\!\left|{f}\right|\!\right|_{x}=\left|\!\left|{gf(x)}\right|\!\right|_{u}\quad g\in U_{f}

as in equation 1. Then define tropG(x)=||||x\operatorname{trop}_{G}(x)=|\!|\!\cdot\!|\!|_{x}, this gives a map G/H(K¯)𝒱G/H(\overline{K})\rightarrow{\mathcal{V}}. For a subvariety WG/HW\subseteq G/H the tropicalization is the closure of tropG(W(K¯))\operatorname{trop}_{G}(W(\overline{K})), in particular the tropicalization of G/HG/H is 𝒱{\mathcal{V}}.

In [Nas17] the above is extended to construct a tropicalization for any G/HG/H-embedding XX, this is done similarly to the toric case. As a set the tropicalization, denoted TropG(X)\operatorname{Trop}_{G}(X), is defined to be:

TropG(X):=(τ,)𝔉(X)𝒱(τ)\operatorname{Trop}_{G}(X):=\bigsqcup_{(\tau,{\mathcal{F}})\in{\mathfrak{F}}(X)}{\mathcal{V}}(\tau)

where 𝒱(τ){\mathcal{V}}(\tau) is the valuation cone of the GG-orbit associated to τ\tau.

Remark 5.1.

If GG is a torus and XX is a toric variety then the cones 𝒱(τ){\mathcal{V}}(\tau) are exactly the vector spaces N(τ)N(\tau) as in the previous subsection.

We now put a topology on TropG(X)\operatorname{Trop}_{G}(X). Consider first the case when XX is simple with closed orbit YY; so XX corresponds to the colored cone (σY(X),Y(X))(\sigma_{Y}(X),{\mathcal{F}}_{Y}(X)). Let σ=σY(X)\sigma=\sigma_{Y}(X). Recall that hom(σ,¯)\hom(\sigma^{\vee},\overline{{\mathbb{R}}}) is in bijection with the disjoint union of the vector spaces N(τ)=N/span(τ)N(\tau)=N_{{\mathbb{R}}}/\operatorname{span}(\tau), where τ\tau ranges over the faces of σ\sigma. The bijeciton is given by αα~\alpha\mapsto\tilde{\alpha} where

α~(v)={α(v),vτσ,else\tilde{\alpha}(v)=\left\{\begin{array}[]{cc}\alpha(v),&v\in\tau^{\perp}\cap\sigma^{\vee}\\ \infty,&\text{else}\\ \end{array}\right.

The colored faces of (σ,Y(X))(\sigma,{\mathcal{F}}_{Y}(X)) are exactly the faces of σ\sigma whose relative interior intersects 𝒱{\mathcal{V}}. Given such a face τ\tau, N(τ)N(\tau) contains the valuation cone 𝒱(τ){\mathcal{V}}(\tau). Thus we have that TropG(X)hom(σ,¯)¯σ\operatorname{Trop}_{G}(X)\subseteq\hom(\sigma^{\vee},\overline{{\mathbb{R}}})\subseteq\overline{{\mathbb{R}}}^{\sigma^{\vee}} and we give TropG(X)\operatorname{Trop}_{G}(X) the induced topology. For XX not necessarily simple, TropG(X)\operatorname{Trop}_{G}(X) is obtained by tropicalizing the simple spherical subvarieties and gluing the valuation cones of shared orbits.

Remark 5.2.

Let XX be simple with closed orbit YY, and σ=σY(X)\sigma=\sigma_{Y}(X). In general TropG(X)\operatorname{Trop}_{G}(X) is not equal to hom(σ,¯)\hom(\sigma^{\vee},\overline{{\mathbb{R}}}) as we do not necessarily have that σ𝒱\sigma\subseteq{\mathcal{V}} or that 𝒱(τ)=N(τ){\mathcal{V}}(\tau)=N(\tau), for a given colored face (τ,)(\tau,{\mathcal{F}}).

Example 5.3.

Let X=BlO2X=\text{Bl}_{O}{\mathbb{P}}^{2} be the embedding of SL2/H\text{SL}_{2}/H as described in example 4.8. Then the tropicalization is given by

The left most point representing the valuation cone of LL, the right most of EE, and the interior representing the valuation cone of SL2/H\text{SL}_{2}/H. One can find further illustrated examples of this in Sections 4 and 5 of [Nas17].

We have a tropicalization map TropG:X(K¯)TropG(X)\operatorname{Trop}_{G}\colon X(\overline{K})\rightarrow\operatorname{Trop}_{G}(X) extending the one defined in [TV21]. This is because all GG-orbits of XX are spherical homogeneous spaces so we can take the disjoint union of the tropicalization maps. Explicitly, for xX(K¯)x\in X(\overline{K}) if xx is in the GG-orbit YY then set TropG(x)\operatorname{Trop}_{G}(x) to be the valuation ||||x|\!|\!\cdot\!|\!|_{x} on k(Y)k(Y). Then for a subvariety WXW\subseteq X we can again define TropG(W):=tropG(W(K¯))\operatorname{Trop}_{G}(W)\colon=\operatorname{trop}_{G}(W(\overline{K}))

Remark 5.4.

In [Nas17] Nash actually defines a colored tropicalization. This consists of attaching the finite set of colors associated to a colored face, to the corresponding valuation cone. However we do not deal with the colors here so this was not discussed.

We now prove Theorem A. Recall that in Section 3 we showed that we have a retraction of topological spaces 𝐩:XanXan{\mathbf{p}}\colon X^{\text{an}}\rightarrow X^{\text{an}}, given by \ast-multiplication by the unique point of the Shilov boundary of GG^{\beth} (Corollary 3.9).

Theorem A.

Let XX be a spherical GG-variety over an algebraically closed and trivially valued field kk, and let K¯\overline{K} denote Puiseux series in kk. Then there is a canonical homeomorphism ι:TropG(X)𝐩(Xan)\iota:\operatorname{Trop}_{G}(X)\rightarrow{\mathbf{p}}(X^{\text{an}}), and 𝐩:XanTropG(X){\mathbf{p}}\colon X^{\text{an}}\rightarrow\operatorname{Trop}_{G}(X) is a retraction of topological spaces such that the following diagram commutes:

X(K¯){X(\overline{K})}Xan{X^{\text{an}}}TropG(X).{\operatorname{Trop}_{G}(X).}tropG\scriptstyle{\operatorname{trop}_{G}}𝐩\scriptstyle{{\mathbf{p}}}
Proof.

First observe that as sets we have an inclusion ι:TropG(X)Xan\iota:\operatorname{Trop}_{G}(X)\hookrightarrow X^{\text{an}} given as follows. For each orbit ZZ of XX, and ||||𝒱(Z)|\!|\!\cdot\!|\!|\in{\mathcal{V}}(Z), |||||\!|\!\cdot\!|\!| defines an element of XanX^{\text{an}} by:

||||(ηZ,||||)|\!|\!\cdot\!|\!|\mapsto(\eta_{Z},|\!|\!\cdot\!|\!|)

where ηZ\eta_{Z} is the generic point of ZZ. All the GG-invariant subvarieties of a spherical variety are closures of GG-orbits ([Kno91]); thus by Proposition 3.11 we have that 𝐩(Xan)=ι(TropG(X)){\mathbf{p}}(X^{\text{an}})=\iota(\operatorname{Trop}_{G}(X)) as sets. That the diagram commutes as maps of sets follows from Corollary 3.16. Finally, 𝐩:Xan𝐩(Xan){\mathbf{p}}:X^{\text{an}}\rightarrow{\mathbf{p}}(X^{\text{an}}) is continuous by Proposition 3.5. So it remains to show that the topology on TropG(X)\operatorname{Trop}_{G}(X) agrees with the topology of 𝐩(Xan){\mathbf{p}}(X^{\text{an}}), i.e. that ι\iota is a homeomorphism onto its image. We will need to temporarily emphasize the difference between TropG(X)\operatorname{Trop}_{G}(X) and its image in XanX^{\text{an}}. Let {\mathcal{I}} be the set ι(TropG(X))\iota(\operatorname{Trop}_{G}(X)), and give Xan{\mathcal{I}}\subseteq X^{\text{an}} the subspace topology.

Consider the case when XX is simple with closed orbit YY. Let X0X_{0} be the BB-stable affine open defined by equation 3. The open X0X_{0} meets every GG-orbit in XX by Theorem 4.3. Thus X0anX_{0}^{\text{an}} contains {\mathcal{I}}. The topology on X0anX_{0}^{\text{an}} is generated by the functions evf:k[X0]an¯\text{ev}_{f}\colon k[X_{0}]^{\text{an}}\rightarrow\overline{{\mathbb{R}}}, where evf(||||)=||f||\text{ev}_{f}(|\!|\!\cdot\!|\!|)=\left|\!\left|{f}\right|\!\right|, fk[X0]f\in k[X_{0}]. The functions evf\text{ev}_{f} for fk[X0](B)f\in k[X_{0}]^{(B)} define the topology on TropG(X)\operatorname{Trop}_{G}(X) by part 1 of Theorem 4.4. So the inclusion ι:TropG(X)\iota\colon\operatorname{Trop}_{G}(X)\rightarrow{\mathcal{I}} is open. Given that for a general G/HG/H-embedding TropG(X)\operatorname{Trop}_{G}(X) is given by gluing the tropicalizations of simple subembeddings we have an open bijection ι:TropG(X)\iota\colon\operatorname{Trop}_{G}(X)\rightarrow{\mathcal{I}} for all XX.

Consider the case then when XX is proper, then XanX^{\text{an}} is compact. We know {\mathcal{I}} is a closed subset of XanX^{\text{an}}, and therefore {\mathcal{I}} is compact. Thus ι:TropG(X)\iota\colon\operatorname{Trop}_{G}(X)\rightarrow{\mathcal{I}} is a homeomorphism onto its image. Then for a general XX there exists a proper spherical GG-variety X¯\overline{X} and an open GG-equivariant embedding XX¯X\hookrightarrow\overline{X} by [Sum74, Theorem 3]. But then we have a commutative diagram:

Xan{X^{\text{an}}}X¯an{\overline{X}^{\text{an}}}TropG(X){\operatorname{Trop}_{G}(X)}TropG(X¯){\operatorname{Trop}_{G}(\overline{X})}

where the composition TropG(X)TropG(X¯)X¯an\operatorname{Trop}_{G}(X)\hookrightarrow\operatorname{Trop}_{G}(\overline{X})\hookrightarrow\overline{X}^{\text{an}} is a homeomorphism onto its image by the above. Furthermore the map XanX¯anX^{\text{an}}\hookrightarrow\overline{X}^{\text{an}} is an open embedding so the inclusion TropG(X)Xan\operatorname{Trop}_{G}(X)\hookrightarrow X^{\text{an}} is a homeomorphism onto {\mathcal{I}}. ∎

Remark 5.5.

In [KM19] it is shown that tropicalization factors through XanX^{\text{an}} when XX is affine. It follows from Corollary 3.16 that 𝐩{\mathbf{p}} and the map XanTropG(X)X^{\text{an}}\rightarrow\operatorname{Trop}_{G}(X) given in Definition 5.18 of [KM19] agree.

Corollary 5.6.

For WW a subvariety of XX, TropG(W)\operatorname{Trop}_{G}(W) is a strong deformation retract of WanW^{\text{an}}.

Proof.

We have that TropG(W)=𝐩(Wan)\operatorname{Trop}_{G}(W)={\mathbf{p}}(W^{\text{an}}) so this follows from Proposition 3.17. ∎

Remark 5.7.

Notice that as a result of Proposition 3.5 one has that TropG\operatorname{Trop}_{G} defines a functor from the category of spherical GG-varieties to the category of topological spaces. One can describe the morphism as follows. If XXX\rightarrow X^{\prime} is a morphism of GG-varieties, then for a ZXZ\subseteq X a GG-orbit, ZZ maps dominantly onto a GG-orbit ZXZ^{\prime}\subseteq X^{\prime}. Then ||||𝒱(Z)|\!|\!\cdot\!|\!|\in{\mathcal{V}}(Z) is mapped to its restriction to k(Z)k(Z^{\prime}) under the inclusion k(Z)k(Z)k(Z^{\prime})\hookrightarrow k(Z). This is the same as the construction of tropicalization of morphisms give in Section 2 of [Nas20].

From Theorem A we can see that tropicalization did not depend on our choice to use Puiseux series. Let L/kL/k be an algebraically closed, valued extension, with nontrivial valuation. Define tropG,L:X(L)𝒱\operatorname{trop}_{G,L}\colon X(L)\rightarrow{\mathcal{V}} to be the map given x||||xx\mapsto|\!|\!\cdot\!|\!|_{x}, where xx is contained in some GG-orbit YY and ||||x|\!|\!\cdot\!|\!|_{x} is the valuation on k(Y)k(Y) defined in 2.

Corollary 5.8.

The closure of tropG,L(W(L))\operatorname{trop}_{G,L}(W(L)) in TropG(X)\operatorname{Trop}_{G}(X) is equal to 𝐩(Wan)=TropG(W){\mathbf{p}}(W^{\text{an}})=\operatorname{Trop}_{G}(W), where WanW^{\text{an}} is considered as a subspace of XanX^{\text{an}} induced by the inclusion WXW\subseteq X.

Proof.

For a subvariety VXV\subseteq X, W(L)W(L) is a dense subset of WanW^{\text{an}}, and we have a commutative diagram

X(L){X(L)}Xan{X^{\text{an}}}TropG(X).{\operatorname{Trop}_{G}(X).}tropG,L\scriptstyle{\operatorname{trop}_{G,L}}𝐩\scriptstyle{{\mathbf{p}}}

So the closure of tropG(W(L))\operatorname{trop}_{G}(W(L)) in TropG(X)\operatorname{Trop}_{G}(X) for any subvariety WXW\subseteq X must be equal to 𝐩(Wan){\mathbf{p}}(W^{\text{an}}). ∎

In [Nas20] Nash gives a description of tropicalization by embedding XX into a toric variety and then applying the usual tropicalization process for toric varieties. See [Nas20, Section 6] for details. In particular this process is used to prove the following result, for which we can give an alternate proof.

Theorem 5.9.

[Nas20, Theorem 6.7] If WG/HW\subseteq G/H is a subvariety, and W¯\overline{W} is the closure of WW in XX. Then we have that:

TropG(W¯)=TropG(W)¯\operatorname{Trop}_{G}(\overline{W})=\overline{\operatorname{Trop}_{G}(W)}

where TropG(W)¯\overline{\operatorname{Trop}_{G}(W)} is the closure taken in TropG(X)\operatorname{Trop}_{G}(X).

Proof.

This follows from Theorem A, as 𝐩{\mathbf{p}} is a retraction. ∎

6   Compactification of the colored fan

Recall from subsection 2.2 that when working over a trivially valued field XanX^{\text{an}} has a compact analytic subdomain XX^{\beth}. The space XX^{\beth} is defined as the subspace whose points are pairs (x,||||)Xan(x,|\!|\!\cdot\!|\!|)\in X^{\text{an}}, where |||||\!|\!\cdot\!|\!| has center in XX. In this Section we study 𝐩(X){\mathbf{p}}(X^{\beth}) and prove Theorem B. In the case of toric varieties it is shown in [Thu07, Section 2] that 𝐩(X){\mathbf{p}}(X^{\beth}) is the canonical compactification of the fan associated to XX. We begin by reviewing the work of Thuillier on the toric case.

Let XX be a toric variety over kk with dense torus TT. Let F(X)F(X) denote the fan associated to XX. Let σ\sigma be a cone in F(X)F(X) and UσU_{\sigma} the TT-stable open affine associated to σ\sigma. We will assume all the cones of F(X)F(X) are strictly convex. In [Thu07] Thuillier defines a map of topological spaces XXX^{\beth}\rightarrow X^{\beth}; which is equal to the map 𝐩{\mathbf{p}} given by \ast-mutliplication by the unique point in the Shilov boundary of TT^{\beth}. Let σ\sigma^{\vee} denote positive linear functionals on σ\sigma. Thuillier shows that ρ1(T)𝐩(Uσ)\rho^{-1}(T)\cap{\mathbf{p}}(U_{\sigma}^{\beth}) is naturally homeomorphic to hom(σ,[0,))\hom(\sigma^{\vee},[0,\infty)), where hom\hom denotes semigroup homomorphisms that preserve scaling by 0{\mathbb{R}}_{\geq 0} and [0,)[0,\infty) has the additive semigroup structure. The space hom(σ,[0,))\hom(\sigma^{\vee},[0,\infty)) is naturally isomorphic to σ\sigma, as σ\sigma is strictly convex. Furthermore 𝐩(U){\mathbf{p}}(U^{\beth}) is compact, and in fact is canonically homeomorphic to hom(σ,[0,])\hom(\sigma^{\vee},[0,\infty]), the canonical compactification of σ\sigma. We will denote the canonical compactification of a strictly convex cone by σ¯\overline{\sigma}. In this way 𝐩(X){\mathbf{p}}(X^{\beth}) is the canonical compactification of the fan F(X)F(X).

Remark 6.1.

Let XX be a spherical GG-variety with dense GG-orbit G/HG/H. Observe that 𝐩{\mathbf{p}} has image consisting only of G(k)G(k)-invariant valuations, and colored fans may contain points of NN_{{\mathbb{R}}} lying outside 𝒱{\mathcal{V}}. Consider Example 6.2. Because of this fact we can give a more concise description of 𝐩(X){\mathbf{p}}(X^{\beth}) when the following holds:

(σ,)𝔉(X)σ𝒱\bigcup_{(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)}\sigma\subseteq{\mathcal{V}} (\star)

Horospherical varieties, for example, satisfy this condition. However for a symmetric homogeneous space G/HG/H there will be colors whose image in NN_{{\mathbb{R}}} lies outside 𝒱{\mathcal{V}} and thus symmetric varieties may not satisfy the above condition (see [Vus90] or [Per14, Section 3.4]). In Remark 6.5 we discuss how to describe 𝐩(X){\mathbf{p}}(X^{\beth}) when \star6.1 is not satisfied.

Example 6.2.

As in [Nas17, Example 4.4], let G=GL2×GL2G=\text{GL}_{2}\times\text{GL}_{2} act on GL2\text{GL}_{2} via (g,h)x=gxh1(g,h)\cdot x=gxh^{-1}. Note that GL2G/H\text{GL}_{2}\cong G/H where HH is the diagonal subgroup. Let BB be the Borel subgroup given by elements of the form (b,b)(b,b^{\prime}) where bb is an upper triangular matrix and bb^{\prime} is lower triangular; then BB has an open orbit in GL2\text{GL}_{2}. Let XX be the GL2\text{GL}_{2}-embedding given by 2×22\times 2 matrices over kk, and let XX^{\prime} be the GL2\text{GL}_{2}-embedding given by 2×22\times 2 matrices of rank at least one over kk. The data associated to GL2\text{GL}_{2} and the colored cones associated to XX and XX^{\prime} are depicted in Figure 1. In particular there is a single color, DD, given by matrices where the bottom right entry is 0. Neither variety is horospherical [Kno91, Corollary 6.2] though both are symmetric varieties and \star6.1 does not hold for XX but it does for XX^{\prime}.

𝒱{\mathcal{V}}
Figure 1: To the left is the vector space NN_{{\mathbb{R}}}, the valuation cone 𝒱{\mathcal{V}}, and the one color DD, for the homogeneous space GL2\text{GL}_{2}. The blue circle depicts ϱ(||||D)\varrho(|\!|\!\cdot\!|\!|_{D}), and the shaded triangle depicts 𝒱{\mathcal{V}}. In the middle is the colored cone for the GL2\text{GL}_{2}-embedding XX. To the right is the colored cone for XX^{\prime}.
Lemma 6.3.

Let XX be a G/HG/H-embedding satisfying \star6.1. Let ZZ be a GG-orbit of XX, and Z¯\overline{Z} be the closure of ZZ in XX. Then the ZZ-embedding Z¯\overline{Z} satisfies \star6.1.

Proof.

Without loss of generality we may assume XX is simple. Denote the closed orbit by YY. Let τ=σZ(X)\tau=\sigma_{Z}(X), the face of σY(X)\sigma_{Y}(X) corresponding to ZZ. For zZz\in Z let φ:G/HZ\varphi\colon G/H\rightarrow Z be the map given by gHgzgH\mapsto gz, and let φ:NN(τ)\varphi_{*}\colon N_{{\mathbb{R}}}\rightarrow N_{{\mathbb{R}}}(\tau) be the induced map. It follows from Theorem 4.10 that φ(σY(X))=σY(Z¯)\varphi_{*}(\sigma_{Y}(X))=\sigma_{Y}(\overline{Z}), and we know from Theorem 4.9 that φ(𝒱)=𝒱(Z)\varphi_{*}({\mathcal{V}})={\mathcal{V}}(Z). So σY(Z¯)𝒱(τ)\sigma_{Y}(\overline{Z})\subseteq{\mathcal{V}}(\tau). ∎

Because colored cones correspond to simple embeddings, we will first consider simple embeddings. Let XX be simple with closed GG-orbit YY. Let σ=σY(X)\sigma=\sigma_{Y}(X). We begin by defining a map Φ:𝐩(X)σ¯\Phi\colon{\mathbf{p}}(X^{\beth})\rightarrow\overline{\sigma}. The set ρ1(G/H)𝐩(X)\rho^{-1}(G/H)\cap{\mathbf{p}}(X^{\beth}) consists of G(k)G(k)-invariant valuations on k(G/H)k(G/H) having center in XX. Given such a valuation |||||\!|\!\cdot\!|\!| we know it uniquely defines a semigroup homomorphism ϱ(||||):M\varrho(|\!|\!\cdot\!|\!|)\colon M\rightarrow{\mathbb{R}} by:

ϱ(||||)(χf)=log(||f||).\varrho(|\!|\!\cdot\!|\!|)(\chi_{f})=-\log(\left|\!\left|{f}\right|\!\right|).

If p=(ηG/H,||||)Xp=(\eta_{G/H},|\!|\!\cdot\!|\!|)\in X^{\beth} then denote the restriction of ϱ(||||)\varrho(|\!|\!\cdot\!|\!|) to σ\sigma^{\vee} by ϕp\phi_{p}. By Theorem 4.4 part 2, we have that ϕp\phi_{p} takes values in [0,)[0,\infty). Thus we have a map of sets ρ1(G/H)𝐩(X)hom(σ,[0,))\rho^{-1}(G/H)\cap{\mathbf{p}}(X^{\beth})\rightarrow\hom(\sigma^{\vee},[0,\infty)) given by pϕpp\mapsto\phi_{p}.

A general point of 𝐩(X){\mathbf{p}}(X^{\beth}) is a point p=(ηZ,||||)p=(\eta_{Z},|\!|\!\cdot\!|\!|) where ηZ\eta_{Z} is the generic point of a GG-orbit ZZ and |||||\!|\!\cdot\!|\!| is a G(k)G(k)-invariant valuation having center in Z¯\overline{Z}, where Z¯\overline{Z} is the Zariski closure of ZZ in XX. Let τ\tau be the face of σ\sigma corresponding to ZZ. Then |||||\!|\!\cdot\!|\!| defines an element of hom(M(τ),)=hom(τσ,)\hom(M(\tau),{\mathbb{R}})=\hom(\tau^{\perp}\cap\sigma^{\vee},{\mathbb{R}}). This extends uniquely to a morphism of semigroups ϕp:σ¯\phi_{p}\colon\sigma^{\vee}\rightarrow\overline{{\mathbb{R}}} by setting ϕp(u)=\phi_{p}(u)=\infty when uτu\notin\tau^{\perp}. But by Theorem 4.9 and 4.10 we have that τσ=σY(Z¯)\tau^{\perp}\cap\sigma^{\vee}=\sigma_{Y}(\overline{Z}), thus by Theorem 4.4 part 2 we see that ϕp\phi_{p} only takes values in [0,][0,\infty]. So we have a map Φ:𝐩(X)σ¯\Phi\colon{\mathbf{p}}(X^{\beth})\rightarrow\overline{\sigma}, given by Φ(p)=ϕp\Phi(p)=\phi_{p}.

Proposition 6.4.

Let XX be a simple G/HG/H-embedding corresponding to the colored cone (σ,)(\sigma,{\mathcal{F}}), and satisfying \star6.1. Then there is a commutative diagram:

𝐩(X){{\mathbf{p}}(X^{\beth})}σ¯{\overline{\sigma}}𝐩(X)ρ1(G/H){{\mathbf{p}}(X^{\beth})\cap\rho^{-1}(G/H)}σ{\sigma}\scriptstyle{\sim}\scriptstyle{\sim}

where the horizontal arrows are homeomorphisms and are given by Φ\Phi.

Proof.

Notice that

σ¯=τhom(τσ,[0,))\overline{\sigma}=\bigsqcup_{\tau}\hom(\tau^{\perp}\cap\sigma^{\vee},[0,\infty))

where τ\tau runs over the faces of σ\sigma. But then because XX satisfies \star6.1 we have that:

τhom(τσ,[0,))=Zhom(σY(Z¯),[0,))\bigsqcup_{\tau}\hom(\tau^{\perp}\cap\sigma^{\vee},[0,\infty))=\bigsqcup_{Z}\hom(\sigma_{Y}(\overline{Z})^{\vee},[0,\infty))

where ZZ runs over the orbits of XX, and Z¯\overline{Z} is the Zariski closure of ZZ in XX. Because XX satisfies \star6.1 all elements of hom(σY(Z¯),[0,))\hom(\sigma_{Y}(\overline{Z})^{\vee},[0,\infty)) are uniquely given by G(k)G(k)-invariant valuations on k(Z)k(Z), and by Theorem 4.4 part 2, hom(σY(Z¯),[0,)\hom(\sigma_{Y}(\overline{Z})^{\vee},[0,\infty) is exactly those G(k)G(k)-invariant valuations with center in Z¯\overline{Z}. Thus it follows that Φ\Phi defines a bijection 𝐩(X)σ¯{\mathbf{p}}(X^{\beth})\rightarrow\overline{\sigma} and 𝐩(X)ρ1(G/H)σ{\mathbf{p}}(X^{\beth})\cap\rho^{-1}(G/H)\rightarrow\sigma.

It remains to show that Φ\Phi is a homeomorphism, and because σ¯\overline{\sigma} is compact it suffices to show Φ\Phi is open. Let X0X_{0} be the BB-stable affine open from equation 3, note that X0X_{0} intersects all GG-orbits of XX. So 𝐩(X)X0{\mathbf{p}}(X^{\beth})\subseteq X_{0}^{\beth}. Recall from Theorem 4.4 part 1 that k[X0](B)={fk(G/H)(B)|χf(σY(X))}k[X_{0}]^{(B)}=\{f\in k(G/H)^{(B)}\,|\,\chi_{f}\in(\sigma_{Y}(X))^{\vee}\}. But then evaluation at the functions fk[X0](B)f\in k[X_{0}]^{(B)} generates the topology on σ¯\overline{\sigma}, so Φ\Phi is open. ∎

As a corollary we have the following.

Theorem B.

There is a strong deformation retraction from XX^{\beth} onto 𝐩(X){\mathbf{p}}(X^{\beth}). If the colored fan 𝔉(X){\mathfrak{F}}(X) satisfies:

(σ,)𝔉(X)σ𝒱.\bigcup_{(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)}\sigma\subseteq{\mathcal{V}}.

Then 𝐩(X){\mathbf{p}}(X^{\beth}) is homeomorphic to the the canonical compactification of the fan F(X)={σ|(σ,)𝔉(X)}F(X)=\{\sigma\,|\,(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)\}.

Proof.

That XX^{\beth} strong deformation retracts onto 𝐩(X){\mathbf{p}}(X^{\beth}) space follows from Corollary 3.17. By Proposition 6.4 if XX satisfies \star6.1 then 𝐩(X){\mathbf{p}}(X^{\beth}) is homeomorphic to the canonical compactification of the fan F(X)={σ|(σ,)𝔉(X)}F(X)=\{\sigma\,|\,(\sigma,{\mathcal{F}})\in{\mathfrak{F}}(X)\}. ∎

Remark 6.5.

Because the map 𝐩{\mathbf{p}} maps into the closure of 𝒱{\mathcal{V}} in XX^{\beth} by construction, one will not have the above for a general spherical variety. This is because a colored cone need not be contained in 𝒱{\mathcal{V}}. However given a simple embedding XX, one will have that 𝐩(X)=σ𝒱¯{\mathbf{p}}(X^{\beth})=\overline{\sigma\cap{\mathcal{V}}}, where the closure is taken in hom(σ,[0,])\hom(\sigma^{\vee},[0,\infty]). Consider for example the GL2\text{GL}_{2}-embedding 𝐩(X){\mathbf{p}}(X^{\beth}) as in Figure 1, one can see 𝐩(X){\mathbf{p}}(X^{\beth}) depicted in Figure 2.

Figure 2: The canonical compactification of the cone σ\sigma associated to XX, along with 𝐩(X){\mathbf{p}}(X^{\beth}) depicted as the pink shaded region to the right.
Remark 6.6.

Because of this difference between spherical varieties that satisfy \star6.1 and those that do not, it may be interesting to describe \ast-multiplication by 𝐛{\mathbf{b}}, the unique point in the Shilov boundary of BB^{\beth}. All of the results described in Section 3 regarding 𝐠{\mathbf{g}} apply to 𝐛{\mathbf{b}}, including the construction of the homotopy from XX^{\beth} to the image of \ast-multiplication by 𝐛{\mathbf{b}}.

References

  • [Ber90] Vladimir Berkovich “Spectral theory and analytic geometry over non-Archimedean fields” 33, Mathematical Surveys and Monographs American Mathematical Society, Providence, RI, 1990, pp. x+169 DOI: http://doi.org/10.1090/surv/033
  • [BGR84] Siegfried Bosch, Ulrich Güntzer and Reinhold Remmert “Non-Archimedean analysis” A systematic approach to rigid analytic geometry 261, Grundlehren der mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences] Springer-Verlag, Berlin, 1984, pp. xii+436 DOI: http://doi.org/10.1007/978-3-642-52229-1
  • [BLV86] M. Brion, D. Luna and Th. Vust “Espaces homogènes sphériques” In Invent. Math. 84.3, 1986, pp. 617–632 DOI: 10.1007/BF01388749
  • [BP87] Michel Brion and Franz Pauer “Valuations des espaces homogènes sphériques” In Comment. Math. Helv. 62.2, 1987, pp. 265–285 DOI: 10.1007/BF02564447
  • [CLS11] David Cox, John Little and Henry Schenck “Toric varieties” 124, Graduate Studies in Mathematics American Mathematical Society, Providence, RI, 2011, pp. xxiv+841 DOI: http://doi.org/10.1090/gsm/124
  • [Ful93] William Fulton “Introduction to toric varieties” The William H. Roever Lectures in Geometry 131, Annals of Mathematics Studies Princeton University Press, Princeton, NJ, 1993, pp. xii+157 DOI: http://doi.org/10.1515/9781400882526
  • [Gag17] Giuliano Gagliardi “Spherical varieties with the AkA_{k}-property” In Math. Res. Lett. 24.4, 2017, pp. 1043–1065 DOI: 10.4310/MRL.2017.v24.n4.a6
  • [GP11] “Schémas en groupes (SGA 3). Tome III. Structure des schémas en groupes réductifs” Séminaire de Géométrie Algébrique du Bois Marie 1962–64. [Algebraic Geometry Seminar of Bois Marie 1962–64], A seminar directed by Michel Demazure and Alexander Grothendieck with the collaboration of Michael Artin, Bertin, Jean-Étienne Bertin, Pierre Gabriel, Michèle Raynaud and Jean-Pierre Serre, Revised and annotated edition of the 1970 French original 8, Documents Mathématiques (Paris) [Mathematical Documents (Paris)] Société Mathématique de France, Paris, 2011, pp. lvi+337 DOI: https://doi.org/10.1007/BFb0059027
  • [Gru66] Laurent Gruson “Théorie de Fredholm pp-adique” In Bull. Soc. Math. France 94, 1966, pp. 67–95 DOI: https://doi.org/10.24033/bsmf.1635
  • [KM19] Kiumars Kaveh and Christopher Manon “Gröbner theory and tropical geometry on spherical varieties” In Transform. Groups 24.4, 2019, pp. 1095–1145 DOI: http://doi.org/10.1007/s00031-019-09536-5
  • [Kno91] Friedrich Knop “The Luna-Vust theory of spherical embeddings” In in Proc. of the Hyderabad Conf. on Algebraic Groups, Manoj Prakashan, 1991, pp. 225–249
  • [LV83] Domingo Luna and Thierry Vust “Plongements d’espaces homogènes” In Comment. Math. Helv. 58.2, 1983, pp. 186–245 DOI: https://doi.org/10.1007/BF02564633
  • [Mil17] James Milne “Algebraic groups” The theory of group schemes of finite type over a field 170, Cambridge Studies in Advanced Mathematics Cambridge University Press, Cambridge, 2017, pp. xvi+644 DOI: http://doi.org/10.1017/9781316711736
  • [MS15] Diane Maclagan and Bernd Sturmfels “Introduction to tropical geometry” 161, Graduate Studies in Mathematics American Mathematical Society, Providence, RI, 2015, pp. xii+363 DOI: http://doi.org/10.1090/gsm/161
  • [Nas17] Evan Nash “Tropicalizing Spherical Embeddings”, 2017 arXiv:1609.07455 [math.AG]
  • [Nas20] Evan Nash “Global spherical tropicalization via toric embeddings” In Math. Z. 294.1-2, 2020, pp. 615–637 DOI: http://doi.org/10.1007/s00209-019-02284-y
  • [Pay09] Sam Payne “Analytification is the limit of all tropicalizations” In Math. Res. Lett. 16.3, 2009, pp. 543–556 DOI: http://doi.org/10.4310/MRL.2009.v16.n3.a13
  • [Per14] Nicolas Perrin “On the geometry of spherical varieties” In Transform. Groups 19.1, 2014, pp. 171–223 DOI: 10.1007/s00031-014-9254-0
  • [Poi13] Jérôme Poineau “Les espaces de Berkovich sont angéliques” In Bull. Soc. Math. France 141.2, 2013, pp. 267–297 DOI: http://doi.org/10.24033/bsmf.2648
  • [Sum74] Hideyasu Sumihiro “Equivariant completion” In Journal of Mathematics of Kyoto University 14.1 Duke University Press, 1974, pp. 1–28 DOI: http://doi.org/10.1215/kjm/1250523277
  • [Thu07] Amaury Thuillier “Géométrie toroïdale et géométrie analytique non archimédienne. Application au type d’homotopie de certains schémas formels” In Manuscripta Math. 123.4, 2007, pp. 381–451 DOI: http://doi.org/10.1007/s00229-007-0094-2
  • [Tim11] Dmitry Timashev “Homogeneous spaces and equivariant embeddings” Invariant Theory and Algebraic Transformation Groups, 8 138, Encyclopaedia of Mathematical Sciences Springer, Heidelberg, 2011, pp. xxii+253 DOI: http://doi.org/10.1007/978-3-642-18399-7
  • [TV21] Jenia Tevelev and Tassos Vogiannou “Spherical tropicalization” In Transform. Groups 26.2, 2021, pp. 691–718 DOI: http://doi.org/10.1007/s00031-021-09641-4
  • [Vus90] Thierry Vust “Plongements d’espaces symétriques algébriques: une classification” In Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4) 17.2, 1990, pp. 165–195 URL: http://www.numdam.org/item?id=ASNSP_1990_4_17_2_165_0