Gorenstein-Fano polytopes and compactifications of rank 2 polyptych lattices

Abstract.

The notion of polyptych lattices, introduced by Escobar, Harada, and Manon, wraps the data of a collection of lattices related by piecewise-linear bijections together into a single semi-algebraic object, equipped with its own notions of convexity and polyhedra. The main purpose of this manuscript is to construct an explicit family of polyptych lattices, and to illustrate via explicit computations the abstract theory introduced by Escobar-Harada-Manon. Specifically, we first construct a family of rank-$2$ polyptych lattices $\mathcal{M}_{s}$ with $2$ charts, compute their space of points, and prove that they are full and self-dual. We then give a concrete sample computation of a point-convex hull in $\mathcal{M}_{s}\otimes{\mathbb{R}}$ to illustrate that convex geometry in the polyptych lattice setting can exhibit phenomena not seen in the classical situation. We also give multiple examples of $2$-dimensional “chart-Gorenstein-Fano” polytopes, which give rise to pairs of mutation-related $2$-dimensional (classical) Gorenstein-Fano polytopes. Finally, we produce detropicalizations $(\mathcal{A}_{s},\mathfrak{v}_{s})$ of $\mathcal{M}_{s}$, and in the case $s=1$ where the detropicalization is a UFD, and with respect to a certain choice of PL polytope $\mathcal{P}$, we give an explicit generators-and-relations presentation of the (finitely generated) Cox ring of the compactification $X_{\mathcal{A}_{s}}(\mathcal{P})$ of $\textup{Spec}(\mathcal{A}_{s})$ with respect to $\mathcal{P}$.

1. Introduction

We view this manuscript as a companion paper to [3], where the new concept of polyptych lattices is introduced and some of its basic properties explored. The main purpose of this paper is to concretely illustrate several of the abstract constructions given in [3] via a very explicit family of examples.

A polyptych lattice $\mathcal{M}$ is a collection of lattices $\mathcal{S}=\{M_{i}\cong{\mathbb{Z}}^{r}\}$ which are related by piecewise-linear bijections (which we think of as “mutations”) [3, Definition 2.1]. As explained in the introduction of [3], we view the concept of a polyptych lattice as a generalization of a classical lattice $M\cong{\mathbb{Z}}^{r}$ as it appears in toric geometry; in the polyptych lattice world, the classical lattice is the “trivial” case in which the set $\mathcal{S}$ of lattices consists only of a single lattice, and there are no mutations. In this note, we define a family of polyptych lattices which may be considered as the simplest non-trivial case, namely, where $\mathcal{M}$ consists of exactly $2$ lattices $\{M_{1},M_{2}\}$, and each $M_{i}$ is of rank $2$, i.e., $M_{i}\cong{\mathbb{Z}}^{2}$. We will see below that, even in such a simple case, we can already see interesting phenomena.

We now describe the content of this note in some more detail. Let $s$ be a positive integer. After a very brief review in Section 2 of the key definitions of [3], we define in Section 3 a polyptych lattice $\mathcal{M}_{s}$ of rank $2$ over ${\mathbb{Z}}$ consisting of $2$ lattices $M_{1},M_{2}$ (both isomorphic to ${\mathbb{Z}}^{2}$), related by a “shear” mutation where the length of the shear depends on the parameter $s$. The main results of Section 3 explicitly compute the space of points $\textup{Sp}(\mathcal{M}_{s})$ (Proposition 3.3) in the sense of [3, Definition 3.1], and show that $\mathcal{M}_{s}$ is (strictly) self-dual in the sense of [3, Definition 4.1]. We then explore some convex geometry in the polyptych lattice setting in Section 4, where we give a sample computation of a point-convex hull of a finite set $S$ in $\mathcal{M}$ in the sense of [3, Definition 3.22]. This computation shows that convex geometry can be surprising in the PL context; indeed, our example shows that when viewed in one of the lattices $M_{i}$, a point-convex hull $\textup{p-conv}(S)$ of a set $S$ may not be the same as the classical convex hull.

In Section 5 we give multiple examples of what we call chart-Gorenstein-Fano PL polytopes, in the sense of [3, Definitions 5.1, 5.21]. This deserves some discussion, since it is connected to past work in related areas. Since we consider in this paper a family of rank-$2$ polyptych lattices with $2$ charts, our PL polytopes have 2 chart images $P_{1},P_{2}$ which are related by a single mutation. If our PL polytope $\mathcal{P}$ is chart-Gorenstein-Fano (cf. Definition 2.14) then each $P_{i}$ is a classical $2$-dimensional Gorenstein-Fano polytope, and they are related by a piecewise-linear map.

We note that such mutations of polytopes have been studied extensively in the context of, for instance, deformations of toric varieties, and complexity-$1$ $T$-varieties, and we expect our PL theory to be related to this work. More specifically, in the cases considered in this paper, we expect that the compactification $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$ (in the sense of [3, Section 7.2]) is an example of a simultaneous deformation of the toric varieties associated to the two chart images $P_{1},P_{2}$ of the PL polytopes $\mathcal{P}$ (of which we list multiple examples in Section 5). Such deformations have been studied by Petracci [6] and Ilten [5, 4]. In particular, when the mutation between the charts of $\mathcal{M}$ is applied to $\mathcal{P}$, we suspect that it gives an instance of a mutation of polytopes as studied by Ilten [5]; the variety $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$ would then be the general fiber of the total space of the deformation associated to that mutation. Equations which cut out $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$ could then be deduced from work of Petracci [6].

In general, the link between polyptych lattices and algebraic geometry comes from our notion of a detropicalization $\mathcal{A}_{\mathcal{M}}$ of a polyptych lattice $\mathcal{M}$ [3, Definition 6.3] and its associated compactification $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$ with respect to a PL polytope $\mathcal{P}$ [3, Section 7.2]. In addition, in [3, Section 7] we proved some first basic geometric properties of these compactifications, and in particular in [3, Theorem 7.19] we prove that if $\mathcal{A}_{\mathcal{M}}$ is a UFD, then $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$ has finitely generated Cox ring. In this paper, we prove in Section 6 that $\mathcal{M}_{s}$ is detropicalizable, by producing an explicit detropicalization $\mathcal{A}_{s}$ equipped with a valuation $\mathfrak{v}:\mathcal{A}_{s}\to\textup{P}_{\mathcal{M}_{s}}$. (As a sidenote, we remark that this construction also shows that there exist examples of detropicalizations that are not UFDs; indeed, it’s easy to see that for $s=2$, the ring $\mathcal{A}_{s}$ is not a UFD.) Then, in Section 7, by taking advantage of the fact that $\mathcal{A}_{s}$ is a UFD for $s=1$, we take [3, Theorem 7.19] one step further and give an explicit generators-and-relations presentation of the Cox ring of $X_{\mathcal{A}_{s}}(\mathcal{P})$ for a particular choice of $\mathcal{P}$. Finally, we note that in Section 7 we additionally prove a general result that is not limited to the rank-$2$ examples $\mathcal{M}_{s}$ considered in this note. Namely, in Proposition 7.6, we give a computation of the group of units in a detropicalization $\mathcal{A}_{\mathcal{M}}$ for any (finite) polyptych lattice $\mathcal{M}$ over ${\mathbb{Z}}$.

Acknowledgements

Some of the results contained in this note, particularly those in Section 3 and 6, were obtained in the Master’s thesis of the first author, which was supervised by the third author. AC was additionally supported by an NSERC OGS scholarship. LE was supported by NSF CAREER grant DMS-2142656, and a Fields Institute Research Fellowship. MH was supported by a Canada Research Chair Award (Tier 2) and NSERC Discovery Grant 2019-06567. CM is supported by NSF DMS grant 2101911.

2. Background

In this section we briefly recount some of the basic definitions . For details we refer to [3].

We begin with the definition of polyptych lattices. Recall that a polyptych lattice is a generalization of the concept of lattices; a lattice is a free ${\mathbb{Z}}$-module of finite rank, and we often fix an identification of a lattice of rank $r$ with ${\mathbb{Z}}^{r}$. In this note, we restrict to polyptych lattices over ${\mathbb{Z}}$ (in the sense of [3, Definition 2.1]) so we drop the reference to coefficients.

2.1 Definition.

Let $r$ be a positive integer. A polyptych lattice of rank $r$ (over ${\mathbb{Z}}$) is a pair $\mathcal{M}:=(\{M_{\alpha}\}_{\alpha\in\mathcal{I}},\{\mu_{\alpha,\beta}:M_{\alpha}\to M_{\beta}\}_{\alpha,\beta\in\mathcal{I}})$ consisting of a collection $\{M_{\alpha}\}_{\alpha\in\mathcal{I}}$ of free ${\mathbb{Z}}$-modules, each of rank $r$ and indexed by a set $\mathcal{I}$, and a collection of piecewise-linear maps $\mu_{\alpha,\beta}:M_{\alpha}\to M_{\beta}$ for every pair $(\alpha,\beta)$ of indices, satisfying the following conditions:

1. $\mu_{\alpha,\alpha}=\mathrm{Id}_{M_{\alpha}}$ is the identity map for all $\alpha\in\mathcal{I}$,

2. $\mu_{\alpha,\beta}=\mu_{\beta,\alpha}^{-1}$ for all pairs $\alpha,\beta\in\mathcal{I}$, and

3. $\mu_{\beta,\gamma}\circ\mu_{\alpha,\beta}=\mu_{\alpha,\gamma}$ for all triples $\alpha,\beta,\gamma\in\mathcal{I}$.

Note in particular that the requirement (2) above implies that all the maps $\mu_{\alpha,\beta}$ are invertible. We call the maps $\mu_{\alpha,\beta}$ mutations, and we call $M_{\alpha}$ a chart of $\mathcal{M}$. When $\mathcal{I}$ is finite, we say $\mathcal{M}$ is a finite polyptych lattice. ∎

In this note, we focus on a class of examples in which $\lvert\mathcal{I}\rvert=2$, so there are only $2$ charts, and the rank is $2$. In particular, all of the polyptych lattices appearing in this note are finite.

Given a polyptych lattice $\mathcal{M}$, by slight abuse of notation we denote also by $\mathcal{M}$ the quotient space

(2.2) $$\mathcal{M}:=\bigsqcup_{\alpha\in\mathcal{I}}M_{\alpha}\bigg{/}\sim$$

where the equivalence relation is defined by $m_{\alpha}\sim m_{\beta}$, for $m_{\alpha}\in M_{\alpha},m_{\beta}\in M_{\beta}$, precisely when $\mu_{\alpha,\beta}(m_{\alpha})=m_{\beta}$. An element of $\mathcal{M}$ is an equivalence class in the quotient space in (2.2). The $\alpha$-th chart map is $\pi_{\alpha}:\mathcal{M}\to M_{\alpha},\quad m\mapsto m_{\alpha}$ and we call $\pi_{\alpha}(m)$ the $\alpha$-th coordinate of $m\in\mathcal{M}$.

Unlike the situation of a classical lattice, there does not exist in general a well-defined operation of addition in $\mathcal{M}$. Nevertheless, for $m,m^{\prime}\in\mathcal{M}$, and $\alpha\in\pi(\mathcal{M})=\mathcal{I}$, we may define

(2.3) $$m+_{\alpha}m^{\prime}:=\pi_{\alpha}^{-1}(\pi_{\alpha}(m)+\pi_{\alpha}(m^{\prime}))$$

which we think of as “addition in the chart $M_{\alpha}$”. Using this, we can define “points” of $\mathcal{M}$, as below.

2.4 Definition.

Let $\mathcal{M}$ be a polyptych lattice. A point of $\mathcal{M}$ is a function $p:\mathcal{M}\to{\mathbb{Z}}$ such that

(2.5) $$p(m)+p(m^{\prime})=\min\{p(m+_{\alpha}m^{\prime})\,\mid\,\alpha\in\pi(\mathcal{M})\}\,\,\textup{ for all }\,m,m^{\prime}\in\mathcal{M}$$

The set of all such $p:\mathcal{M}\to{\mathbb{Z}}$ is called the space of points of $\mathcal{M}$ and denoted $\textup{Sp}(\mathcal{M})$.

Any point $p\in\textup{Sp}(\mathcal{M})$ induces a function $p_{\alpha}:=p\circ\pi_{\alpha}^{-1}:M_{\alpha}\to{\mathbb{Z}}$ on the lattice $M_{\alpha}$; these are not linear in general.

2.6 Definition.

We let $Sp(\mathcal{M},\alpha)$ denote the subset of points $p$ on $\mathcal{M}$ such that $p_{\alpha}:M_{\alpha}\to{\mathbb{Z}}$ is linear. If $\textup{Sp}(\mathcal{M})=\cup_{\alpha}\textup{Sp}(\mathcal{M},\alpha)$, then we say that $\mathcal{M}$ is full.

We need some polyptych lattice analogues of some classical convex-geometric objects. Given a rank $r$ polyptych lattice $\mathcal{M}$ we may define $\mathcal{M}_{\mathbb{R}}$ by replacing ${\mathbb{Z}}^{r}$ with ${\mathbb{R}}^{r}$ in Definition 2.1 and using the same mutation maps.

2.7 Definition.

Let $\mathcal{M}$ be a polyptych lattice. A PL cone is a subset $\mathcal{C}$ of $\mathcal{M}_{{\mathbb{R}}}$ such that $\pi_{\alpha}(\mathcal{C})\subseteq M_{\alpha}\otimes{\mathbb{R}}$ is a rational polyhedral cone for each $\alpha\in\pi(\mathcal{M})$ (cf. [2, Definition 1.2.1, Definition 1.2.14]).

The dimension of a PL cone $\mathcal{C}$ is the dimension of any chart image $\pi_{\alpha}(\mathcal{C})$. Given a PL cone $\mathcal{C}$, a face $\mathcal{C}^{\prime}$ of $\mathcal{C}$ is a subset of $\mathcal{C}$ such that $\pi_{\alpha}(\mathcal{C}^{\prime})$ is a face of $\mathcal{C}$ for all $\alpha\in\pi(\mathcal{M})$. A facet of a PL cone $\mathcal{C}$ is a face of dimension $\dim(\mathcal{C})-1$. Any face of $\mathcal{C}$ is itself a PL cone.

2.8 Definition.

Let $\mathcal{M}$ be a polyptych lattice. A PL fan in $\mathcal{M}_{{\mathbb{R}}}$ is a finite collection $\Sigma$ of PL cones in $\mathcal{M}_{{\mathbb{R}}}$ such that:

1. for every $\mathcal{C}\in\Sigma$ and every $\alpha\in\pi(\mathcal{M})$, the chart image $\pi_{\alpha}(\mathcal{C})$ is a rational polyhedral cone,

2. for every $\mathcal{C}\in\Sigma$, each face of $\mathcal{C}$ is also in $\Sigma$,

3. for all $\mathcal{C},\mathcal{C}^{\prime}\in\Sigma$, the intersection $\mathcal{C}\cap\mathcal{C}^{\prime}$ is a face of each, (and hence also in $\Sigma$).

The support of a PL fan is $\lvert\Sigma\rvert:=\cup_{\mathcal{C}\in\Sigma}\mathcal{C}$. A PL fan in $\mathcal{M}_{{\mathbb{R}}}$ is complete if $\lvert\Sigma\rvert=\mathcal{M}_{{\mathbb{R}}}$. A PL fan $\Sigma^{\prime}$ refines a PL fan $\Sigma$ if every $\mathcal{C}^{\prime}\in\Sigma^{\prime}$ is contained in a PL cone of $\Sigma$ and $\lvert\Sigma^{\prime}\rvert=\lvert\Sigma\rvert$.

We recall the definition of the PL fan $\Sigma(\mathcal{M})$ associated to a polyptych latice. For any pair $(\alpha,\beta)\in\mathcal{I}^{2}$, there exists a minimal fan $\Sigma(\mathcal{M},\alpha,\beta)$ in $M_{\alpha}\otimes\mathbb{R}$ such that, for each cone $C\in\Sigma(\mathcal{M},\alpha,\beta)$, the restriction $\mu_{\alpha,\beta}|_{C}:C\to\mathbb{R}$ is $\mathbb{R}$-linear. Let $\alpha$ be fixed. Let $\Sigma(\mathcal{M},\alpha)$ denote the common refinement of all $\Sigma(\mathcal{M},\alpha,\beta)$ as $\beta$ ranges over the finite set $\mathcal{I}=\pi(\mathcal{M})$. This is a fan in $M_{\alpha}\otimes{\mathbb{R}}$ which has the property that for any cone $C$ of $\Sigma(\mathcal{M},\alpha)$ and any $\beta\in\mathcal{I}$, the mutation $\mu_{\alpha,\beta}$ restricts to $C$ to be linear. Now let $\mathcal{M}_{\mathbb{R}}=\bigcup_{C\in\Sigma(\mathcal{M},\alpha)}\pi_{\alpha}^{-1}(C)$ be the decomposition of $\mathcal{M}_{\mathbb{R}}$ into preimages of the cones in $\Sigma(\mathcal{M},\alpha)$. We call this decomposition the PL fan of $\mathcal{M}$, and denote it by $\Sigma(\mathcal{M})$. It is shown in [3, Lemma 2.10] that this is indeed a PL fan.

Given two polyptych lattices $\mathcal{M}$ and $\mathcal{N}$, we say that the two are strictly dual to each other if - roughly speaking - we can identify (the elements of) $\mathcal{M}$ with $\textup{Sp}(\mathcal{N})$, and vice versa, and their PL fans are compatible. The precise version is below.

2.9 Definition.

Let $\mathcal{M},\mathcal{N}$ be polyptych lattices and ${\sf v}:\mathcal{M}\to\textup{Sp}(\mathcal{N})$ and ${\sf w}:\mathcal{N}\to\textup{Sp}(\mathcal{M})$ a pair of maps. We say that ${\sf v},{\sf w}$ are a strict dual pairing if:

1. ${\sf v}(m)(n)={\sf w}(n)(m)$ for all $n\in\mathcal{N},m\in\mathcal{M}$,

2. ${\sf v}$ and ${\sf w}$ are both bijections, and

3. the preimages ${\sf v}^{-1}\textup{Sp}_{\mathbb{R}}(\mathcal{N},\beta)$ (respectively ${\sf w}^{-1}\textup{Sp}_{\mathbb{R}}(\mathcal{M},\alpha)$) are precisely the maximal-dimensional faces of $\Sigma(\mathcal{M})$ (respectively $\Sigma(\mathcal{N})$).

If $\mathcal{M}$ has a strict dual pairing with itself with respect to a single map $\mathbf{w}:\mathcal{M}\to\textup{Sp}(\mathcal{M})$, we say $\mathcal{M}$ is (strictly) self-dual.

It is shown in [3, Lemma 3.5] that for any finite polyptych lattice $\mathcal{N}$, any point $p\in\textup{Sp}(\mathcal{N})$ extends naturally to a piecewise linear function, also denoted $p$, on $\mathcal{N}_{\mathbb{R}}$; see [3] for precise definitions. Let $P_{\mathcal{N}}$ denote the set of piecewise linear functions on $\mathcal{N}_{\mathbb{R}}$ generated by $\textup{Sp}(\mathcal{N})$ under the operations $+$ and $\mathrm{min}$; then this set $P_{\mathcal{N}}$ is an idempotent ${\mathbb{Z}}_{\geq 0}$-semialgebra with respect to these operations. We refer to $P_{\mathcal{N}}$ as the point semialgebra of $\mathcal{N}$. We may equip $P_{\mathcal{N}}$ with the partial order defined by $a\geq b$ if and only if $\min\{a,b\}=b$, where here the $\min$ is the pointwise minimum of functions.

For the purposes of this note, we need only define valuations with values in either $P_{\mathcal{N}}$ or ${\mathbb{Z}}$, so we restrict to these cases. A valuation $\mathfrak{v}:\mathcal{A}\to P_{\mathcal{N}}$ (resp. $\mathfrak{v}:\mathcal{A}\to{\mathbb{Z}}$) is an analogue of a classical discrete valuation on a field. We have the following.

2.10 Definition.

Let $\mathcal{A}$ be a Noetherian ${\mathbb{K}}$-algebra which is an integral domain. We say a map $\mathfrak{v}:\mathcal{A}\to P_{\mathcal{N}}$ (resp. $\mathfrak{v}:\mathcal{A}\to{\mathbb{Z}}$) is a valuation with values in $P_{\mathcal{N}}$ (resp. ${\mathbb{Z}}$) if for all $f,g\in\mathcal{A}$ we have:

1. $\mathfrak{v}(fg)=\mathfrak{v}(f)\otimes\mathfrak{v}(g)$,

2. $\mathfrak{v}(f+g)\geq\mathfrak{v}(f)\oplus\mathfrak{v}(g)$,

3. $\mathfrak{v}(cf)=\mathfrak{v}(f)$, for all $c\in{\mathbb{K}}^{*}$, and

4. $\mathfrak{v}(0)=\infty$.

We may now define detropicalizations of polyptych lattices in the case when $\mathcal{M}$ has a strict dual. We restrict to this case since the examples in this note have strict duals. The definition in [3] uses valuations valued in the canonical semialgebra $S_{\mathcal{M}}$ of $\mathcal{M}$ (which we have not defined here), but it is shown in [3, Proposition 4.9] that $S_{\mathcal{M}}\cong P_{\mathcal{N}}$ when $\mathcal{M}$ and $\mathcal{N}$ are strict duals, so here we may take the codomain to be $P_{\mathcal{N}}$.

2.11 Definition.

Let $\mathcal{M}$ be a finite polyptych lattice. Assume that $\mathcal{M}$ has a strict dual $\mathcal{N}$. Let $\mathcal{A}_{\mathcal{M}}$ be a Noetherian ${\mathbb{K}}$-algebra which is an integral domain. Let $\mathfrak{v}:\mathcal{A}_{\mathcal{M}}\to P_{\mathcal{N}}$ be a valuation with values in $P_{\mathcal{N}}$. We say that the pair $(\mathcal{A}_{\mathcal{M}},\mathfrak{v})$ is a detropicalization of $\mathcal{M}$ if every element of $\mathcal{M}\cong\textup{Sp}(\mathcal{N})$ is in the image of $\mathfrak{v}$, and the Krull dimension of $\mathcal{A}_{\mathcal{M}}$ equals the rank $r$ of $\mathcal{M}$. We say that a ${\mathbb{K}}$-vector space basis $\mathbb{B}$ of $\mathcal{A}_{\mathcal{M}}$ is a convex adapted basis for $\mathfrak{v}:\mathcal{A}_{\mathcal{M}}\to\textup{P}_{\mathcal{N}}$ if $\mathfrak{v}(\sum\lambda_{i}\mathbb{b}_{i})=\bigoplus_{i}\mathfrak{v}(\mathbb{b}_{i})=\min_{i}\{\mathfrak{v}(\mathbb{b}_{i})\}$, for any finite collection $\lambda_{i}\in{\mathbb{K}}^{*}$ and $\mathbb{b}_{i}\in\mathbb{B}$, and $\mathfrak{v}(\mathbb{b})\in\textup{Sp}(\mathcal{N})\subset\textup{P}_{\mathcal{N}}$ for all $\mathbb{b}\in\mathbb{B}$.

Let $p\in\textup{Sp}(\mathcal{M})$ be a point of $\mathcal{M}$. Let $a\in{\mathbb{Z}}$. The PL half-space with threshold $a$ associated to $p$ is

(2.12) $$\mathcal{H}_{p,a}:=\{m\in\mathcal{M}_{\mathbb{R}}\,\mid\,p(m)\geq a\}\subset\mathcal{M}_{\mathbb{R}}.$$

A set $\mathcal{P}$ is a PL polytope if it is compact and it is a finite intersection of PL half spaces, i.e.,

$$\mathcal{P}=\bigcap_{i=1}^{\ell}\mathcal{H}_{p_{i},a_{i}}$$

for some collection of points $p_{i}\in\textup{Sp}(\mathcal{M})$ and $a_{i}\in{\mathbb{Z}}$. The set of vertices $V(\mathcal{P})$ of $\mathcal{P}$ is

(2.13) $$V(\mathcal{P}):=\{m\in\mathcal{M}_{{\mathbb{R}}}\,\mid\,\exists\alpha\in\pi(\mathcal{M}),\,\,\pi_{\alpha}(m)\,\textup{ is a vertex of }\,\pi_{\alpha}(\mathcal{P})\}.$$

Vertices need not be elements of $\mathcal{M}$. We say that $\mathcal{P}$ is an integral PL polytope if $\pi_{\alpha}(\mathcal{P})$ is an integral polytope in $M_{\alpha}\otimes{\mathbb{R}}$ (i.e., all its vertices are in $M_{\alpha}$) for every $\alpha\in\pi(\mathcal{M})$.

2.14 Definition.

Let $\mathcal{M}$ be a finite polyptych lattice over ${\mathbb{Z}}$. We say that a PL polytope $\mathcal{P}$ in $\mathcal{M}_{\mathbb{R}}$ is chart-Gorenstein-Fano if $\mathcal{P}$ is a full-dimensional integral PL polytope, and, its PL half-space representation is of the form

$$\mathcal{P}=\bigcap_{i=1}^{\ell}\mathcal{H}_{p_{i},-1}$$

where $p_{i}\in\textup{Sp}(\mathcal{M})$ and $a_{i}=-1$ for all $i\in[\ell]$. ∎

Later in this manuscript, we give multiple explicit examples of chart-Gorenstein-Fano PL polytopes. Moreover, following [3] and in the setting when $\mathcal{M}$ possesses a strict dual, we also have a theory of dual polytopes. Indeed, in the presence of a strict dual $\mathcal{N}$ to $\mathcal{M}$ we define the support function $\psi_{\mathcal{P}}:\mathcal{N}_{{\mathbb{R}}}\to{\mathbb{R}}$ of $\mathcal{P}$ as

(2.15) $$\psi_{\mathcal{P}}(-):=\min\{{\sf v}(u)(-)\mid u\in\mathcal{P}\}.$$

Then the dual PL polytope $\mathcal{P}^{\vee}$ to $\mathcal{P}$ (with respect to the strict dual $\mathcal{N}$) is

(2.16) $$\mathcal{P}^{\vee}:=\{n\in\mathcal{N}_{\mathbb{R}}\,\mid\,\psi_{\mathcal{P}}(n)\geq-1\}\subset\mathcal{N}_{\mathbb{R}}.$$

It is shown in [3, Lemma 5.16] that $\mathcal{P}^{\vee}$ can be expressed as

(2.17) $$\mathcal{P}^{\vee}=\bigcap_{m\in V(\mathcal{P})}\mathcal{H}_{{\sf v}(m),-1}$$

and $\mathcal{P}^{\vee}$ is compact in $\mathcal{N}_{\mathbb{R}}$.

2.18 Remark.

It should be emphasized here that, in our setting of PL polytopes, it is not necessarily true that the dual of a chart-Gorenstein-Fano polytope is an integral polytope. See Example 5.6. We intend to explore these subtleties of PL convex geometry in future work.

3. The rank-$2$ polyptych lattices $\mathcal{M}_{s}$ and its space of points

As mentioned in the introduction, one of the goals of this note is to construct a concrete family of rank-$2$ polyptych lattices which serve to illustrate the abstract theory introduced in [3]. In this section, we will define our family of polyptych lattices, compute the associated spaces of points, and show that they are full and strictly self-dual.

Let $s$ be a non-negative integer. We define a polyptych lattice $\mathcal{M}_{s}$ associated to $s$ as follows. There are two coordinate charts $M_{1}$ and $M_{2}$, both isomorphic to ${\mathbb{Z}}^{2}$, so the rank $r$ is $2$ and the set of charts $\mathcal{I}=\{1,2\}$. We fix once and for all identifications of $M_{1}$ and $M_{2}$ with ${\mathbb{Z}}^{2}$, and use coordinates $(x,y)\in{\mathbb{Z}}^{2}$ on $M_{1}$, and $(u,v)\in{\mathbb{Z}}^{2}$ on $M_{2}$. To specify the mutations, it suffices to describe the piecewise-linear mutation map $\mu_{1,2}:M_{1}\rightarrow M_{2}$ as follows:

(3.1) $$\begin{split}\mu_{1,2}(x,y)&=(\min\{0,s\,y\}-x,y)\\ &=\begin{cases}(-x,y)\quad\quad\textup{ if }y\geq 0\\ (s\,y-x,y)\quad\textup{ if }y<0.\end{cases}\end{split}$$

It is straightforward that $(\mu_{1,2})_{{\mathbb{R}}}$ is continuous and we can see that the domains of linearity of $\mu_{1,2}$ are the upper- and lower-half spaces $\{y\geq 0\}$ and $\{y\leq 0\}$ of $M_{1}$, where the mutation may be represented, respectively, by the matrices $\begin{bmatrix}-1&0\\ 0&1\end{bmatrix}$ and $\begin{bmatrix}-1&s\\ 0&1\end{bmatrix}$. It is also straightforward to compute that the inverse mutation $\mu_{2,1}:M_{2}\to M_{1}$ is given by the same formula,

$$\mu_{2,1}(u,v)=(\min\{0,s\,v\}-u,v)$$

and thus also has two domains of linearity, $\{v\geq 0\}$ and $\{v\leq 0\}$. Since there are only 2 charts, we will also refer to $(x,y)$ and $(u,v)$ as the first and second coordinates respectively (of an element of $\mathcal{M}$), and $M_{1}$ as the first chart, $M_{2}$ as the second chart. The maximal cones of the PL fan $\Sigma(\mathcal{M})$ consists of the two disjoint subsets $H_{+}:=\pi_{1}^{-1}(\{y\geq 0\})=\pi_{2}^{-1}(\{v\geq 0\})\subset\mathcal{M}_{s}$ and $H_{-}:=\pi_{1}^{-1}(\{y\leq 0\})=\pi_{2}^{-1}(\{v\leq 0\})\subset\mathcal{M}_{s}$.

The following lemma is proven in [3].

3.2 Lemma.

Let $\mathcal{M}$ be a finite polyptych lattice and let $p\in\textup{Sp}(\mathcal{M})$. Let $C$ be a cone in the PL fan $\Sigma(\mathcal{M})$ of $\mathcal{M}$. Then $p$ is linear when restricted to $C$.

Using the above lemma, we can explicitly compute the space of points $\textup{Sp}(\mathcal{M}_{s})$ of $\mathcal{M}_{s}$. Indeed, by the lemma, we know that for any point $p\in\textup{Sp}(\mathcal{M}_{s})$, the induced functions $p_{i}=p\circ\pi_{i}^{-1}$ must be linear on the upper-half and lower-half spaces of $M_{i}$, so both $p_{1}$ and $p_{2}$ are completely specified by two linear functions on these two half-spaces. With this in mind, we set the following notation. Let $\{\mathbf{e}_{1},\mathbf{e}_{2}\}$ denote the standard basis for ${\mathbb{Z}}^{2}$. Consider the following elements of $\mathcal{M}_{s}$: ${\mathfrak{e}}_{1}:=\pi_{1}^{-1}(\mathbf{e}_{1})$, $\mathfrak{e}_{2}:=\pi_{1}^{-1}(\mathbf{e}_{2})$, and $\mathfrak{e}^{\prime}_{2}:=\pi_{1}^{-1}(-\mathbf{e}_{2})$. Note also that, since an element $m\in\mathcal{M}$ is completely determined by its first coordinate $\pi_{1}(m)$, any function $p:\mathcal{M}\to{\mathbb{Z}}$ is uniquely determined by the induced function $p_{1}:=p\circ\pi_{1}^{-1}$. We take advantage of this observation in the proposition below. We have the following.

3.3 Proposition.

Let $p(\mathfrak{e}_{1})$, $p(\mathfrak{e}_{2})$, and $p(\mathfrak{e}^{\prime}_{2})$ denote integers chosen such that $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=\min\{0,s\cdot p(\mathfrak{e}_{1})\}$. Let $p:\mathcal{M}_{s}\to{\mathbb{Z}}$ be the function uniquely specified by

(3.4) $$p_{1}(x,y):=p\circ\pi_{1}^{-1}(x,y)=\begin{cases}x\cdot p(\mathfrak{e}_{1})-y\cdot p(\mathfrak{e}^{\prime}_{2}),&y\leq 0\\ x\cdot p(\mathfrak{e}_{1})+y\cdot p(\mathfrak{e}_{2}),&y\geq 0\end{cases}.$$

Then $p$ is a point on $\mathcal{M}_{s}$, and, any point in $\textup{Sp}(\mathcal{M}_{s})$ is of this form. In particular, $\mathcal{M}_{s}$ is full, and $\textup{Sp}(\mathcal{M}_{s})\otimes\mathbb{R}$ can be identified with the subset $\mathcal{T}_{s}$ of $\mathbb{R}^{3}$ defined as

(3.5) $$\mathcal{T}_{s}:=\{(a,b,c)\in{\mathbb{R}}^{3}\,\mid\,a+b=\min\{0,s\cdot c\}\}\subset{\mathbb{R}}^{3}.$$

Proof.

We first show that a function $p:\mathcal{M}_{s}\to{\mathbb{Z}}$ defined by (3.4) is in $\textup{Sp}(\mathcal{M}_{s})$. To prove this, we must check the condition (2.5). We may compute in terms of $p_{1}$ instead of $p$, where the requirement becomes that for all $(x,y)(x^{\prime},y^{\prime})\in M_{1}$, we have

(3.6) $$p_{1}(x,y)+p_{1}(x^{\prime},y^{\prime})=\min\{p_{1}(x+x^{\prime},y+y^{\prime}),p_{1}(\min\{0,s(y+y^{\prime})\}-\min\{0,sy\}-\min\{0,sy^{\prime}\}+x+x^{\prime},y+y^{\prime})\}$$

where the second expression in the minimum is equal to $\mu_{2,1}(\mu_{1,2}(x,y)+\mu_{1,2}(x^{\prime},y^{\prime}))$. (This is the first coordinate of the addition of $\pi_{1}^{-1}(x,y)$ and $\pi_{1}^{-1}(x^{\prime},y^{\prime})$ in the chart $M_{2}$ as in (2.3).)

To check (3.6), we take cases. Note that we already know that $p$ is linear when restricted to $H_{+}$ or $H_{-}$ so we only need to check the cases in which the $m$ and $m^{\prime}$ are contained in distinct cones of linearity. Consider first the case when $m\in H_{+},m^{\prime}\in H_{-}$, and $m+_{i}m^{\prime}\in H_{+}$ for $i=1,2$. The LHS of (3.6) is then

(3.7) $$xp(\mathfrak{e}_{1})+yp(\mathfrak{e}_{2})+x^{\prime}p(\mathfrak{e}_{1})-y^{\prime}p(\mathfrak{e}^{\prime}_{2}).$$

The RHS of (3.6) can be simplified using that $y\geq 0,y^{\prime}\leq 0,y+y^{\prime}\geq 0$, and we obtain

$$\min\{(x+x^{\prime})p(\mathfrak{e}_{1})+(y+y^{\prime})p(\mathfrak{e}_{2}),(-sy^{\prime}+x+x^{\prime})p(\mathfrak{e}_{1})+(y+y^{\prime})p(\mathfrak{e}_{2})\}$$

which is in turn equal to

(3.8) $$(x+x^{\prime})p(\mathfrak{e}_{1})+(y+y^{\prime})p(\mathfrak{e}_{2})+\min\{0,-sy^{\prime}\,p(\mathfrak{e}_{1})\}=(x+x^{\prime})p(\mathfrak{e}_{1})+(y+y^{\prime})p(\mathfrak{e}_{2})-sy^{\prime}\min\{0,p(\mathfrak{e}_{1})\}$$

where the last equality follows because $s\geq 0,y^{\prime}\leq 0$ implies $-sy^{\prime}\geq 0$. Setting (3.7) equal to (3.8) the condition becomes

$$-y^{\prime}\,p(\mathfrak{e}^{\prime}_{2})=y^{\prime}\,p(\mathfrak{e}_{2})-sy^{\prime}\min\{0,p(\mathfrak{e}_{1})\}$$

where this equality must hold for any $y^{\prime}\leq 0$. This is true if and only if $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=s\cdot\min\{0,p(\mathfrak{e}_{1})\}$. Checking the other case when $m\in H_{+},m^{\prime}\in H_{-}$ and $m+_{i}m\in H_{-}$ is similar and is left to the reader. In this case we also obtain that the condition of being a point is satisfied if and only if $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=s\cdot\min\{0,p(\mathfrak{e}_{1})\}$. Thus we conclude that $p$ is a point in $\mathcal{M}_{s}$, and moreover, if $p$ is a point in $\mathcal{M}_{s}$, then the values $p(\mathfrak{e}_{1}),p(\mathfrak{e}_{2}),p(\mathfrak{e}^{\prime}_{2})$, which correspond go the values of $p$ on the elements $\mathfrak{e}_{1},\mathfrak{e}_{2},\mathfrak{e}^{\prime}_{2}$ respectively, must satisfy $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=s\cdot\min\{0,p(\mathfrak{e}_{1})\}$. This proves the first statatement of the proposition.

To see that $\mathcal{M}_{s}$ is full, it suffices to show that any point $p$ in $\textup{Sp}(\mathcal{M}_{s})$ is linear in either the first chart or the second chart. We know that $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=s\cdot\min\{0,p(\mathfrak{e}_{1})\}$, so let us take cases. Suppose $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=0$. Then $p(\mathfrak{e}_{2})=-p(\mathfrak{e}^{\prime}_{2})$. From the definition of $p$ from (3.4) it follows immediately that, in this case, $p_{1}=p\circ\pi_{1}^{-1}$ is linear on all of $M_{1}$, so $p\in\textup{Sp}(\mathcal{M},1)$. On the other hand, suppose that $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=s\cdot p(\mathfrak{e}_{1})<0$. Then $p(\mathfrak{e}_{2})=s\cdot p(\mathfrak{e}_{1})-p(\mathfrak{e}^{\prime}_{2})$ so that we may write

$$p((x,y),\mu_{12}(x,y))=\begin{cases}x\cdot p(\mathfrak{e}_{1})-y\cdot p(\mathfrak{e}^{\prime}_{2}),&y\leq 0\\ (x+sy)\cdot p(\mathfrak{e}_{1})-y\cdot p(\mathfrak{e}^{\prime}_{2}),&y\geq 0\end{cases}.$$

Rewriting this in the coordinates for the second chart, we have

(3.9) $$p(\mu_{21}(u,v),(u,v))=\begin{cases}(sv-u)\cdot p(\mathfrak{e}_{1})-v\cdot p(\mathfrak{e}^{\prime}_{2}),&v\leq 0\\ (sv-u)\cdot p(\mathfrak{e}_{1})-v\cdot p(\mathfrak{e}^{\prime}_{2}),&v\geq 0\end{cases}$$

which shows that, in this case, $p_{2}:=p\circ\pi_{2}^{-1}$ is linear, i.e. $p$ is linear in the second chart, and $p\in\textup{Sp}(\mathcal{M}_{s},2)$. Thus any point in $\textup{Sp}(\mathcal{M}_{s})$ is linear in one of the coordinate charts, so $\textup{Sp}(\mathcal{M}_{s})=\textup{Sp}(\mathcal{M}_{s},1)\cup\textup{Sp}(\mathcal{M}_{s},2)$ and $\mathcal{M}_{s}$ is full. Finally, it follows from the above that the space of points $\textup{Sp}(\mathcal{M}_{s})\otimes{\mathbb{R}}$ may be identified with the set of parameters $\{(a,b,c):a+b=\min\{0,s\cdot c\}\}\subset{\mathbb{R}}^{3}$, given by the choices of the values $p(\mathfrak{e}_{1}),p(\mathfrak{e}_{2}),p(\mathfrak{e}^{\prime}_{2})$, so the last claim follows. ∎

3.10 Remark.

In the proposition above, we express a point $p$ in $\textup{Sp}(\mathcal{M}_{s})$ as a function of the variables of the first coordinate chart $M_{1}$. For later computations, it will also be useful to express $p$ in terms of the $M_{2}$ coordinates. It is straightforward to compute that, given the parameters $p(\mathfrak{e}_{2}),p(\mathfrak{e}^{\prime}_{2}),p(\mathfrak{e}_{1})$ as in Proposition 3.3, $p\in\textup{Sp}(\mathcal{M}_{s})$ expressed in $M_{2}$ coordinates $(x^{\prime},y^{\prime})$ is given by

(3.11) $$p(\mu_{2,1}(z^{\prime},y^{\prime}),(x^{\prime},y^{\prime}))=\begin{cases}-x^{\prime}\cdot p(\mathfrak{e}_{1})+y^{\prime}\cdot(s\cdot p(\mathfrak{e}_{1})-p(\mathfrak{e}^{\prime}_{2})),\,\quad\textup{ if }\,y^{\prime}\leq 0\\ -x^{\prime}\cdot p(\mathfrak{e}_{1})+y^{\prime}\cdot p(\mathfrak{e}_{2}),\quad\quad\quad\quad\quad\textup{ if }\,y^{\prime}\geq 0.\end{cases}$$

∎

We next claim that $\mathcal{M}_{s}$ is self-dual in the sense of Definition 2.9, i.e., there exists a strict dual pairing of $\mathcal{M}_{s}$ with itself. This means that we seek a bijective mapping $\mathbf{w}_{s}:\mathcal{M}_{s}\to\textup{Sp}(\mathcal{M}_{s})$ such that for any $m,m^{\prime}\in\mathcal{M}_{s}$, we have

$$\mathbf{w}_{s}(m)(m^{\prime})=\mathbf{w}_{s}(m^{\prime})(m)$$

and such that preimages of $\textup{Sp}(\mathcal{M}_{s},i)$ for $i=1,2$ land precisely on the maximal-dimensional faces of $\Sigma(\mathcal{M}_{s})$. Recall that by Proposition 3.3 we know that a point $p$ in $\textup{Sp}(\mathcal{M}_{s})$ is completely determined by a triple $(p(\mathfrak{e}_{2}),p(\mathfrak{e}^{\prime}_{2}),p(\mathfrak{e}_{1}))$ of integers in $\mathcal{T}_{s}$. More precisely we have a bijection

(3.12) $$\psi:\textup{Sp}(\mathcal{M}_{s})\to\mathcal{T}_{s}\cap{\mathbb{Z}}^{3},\quad\quad p\mapsto(p(\mathfrak{e}_{2}),p(\mathfrak{e}^{\prime}_{2}),p(\mathfrak{e}_{1})).$$

For the remainder of this discussion we identify $\textup{Sp}(\mathcal{M}_{s})$ with $\mathcal{T}_{s}$ via $\psi$ and as such, we will define below a function $\mathbf{w}_{s}:\mathcal{M}_{s}\to\mathcal{T}_{s}$ and interpret this as a mapping to $\textup{Sp}(\mathcal{M}_{s})$.

Let $m=\pi_{1}^{-1}(x,y)$. We then define

(3.13) $$\mathbf{w}_{s}(m)=\mathbf{w}_{s}(\pi_{1}^{-1}(x,y))=\begin{cases}(x,-x,y)\quad\quad\textup{ if }\,\,y\geq 0\\ (x,sy-x,y)\quad\textup{ if }\,\,y\leq 0.\end{cases}$$

3.14 Lemma.

The map $\mathbf{w}_{s}$ of (3.13) defines a strict self-dual pairing of $\mathcal{M}_{s}$ with itself.

Proof.

We must check the conditions (1),(2),(3) of Definition 2.9 for $\mathcal{M}=\mathcal{N}=\mathcal{M}_{s}$ and $\mathbf{v}=\mathbf{w}$, where $\mathbf{w}$ is defined in (3.13).

We first prove (1). We take cases. First suppose $m,m^{\prime}\in H_{+}$. Then $m=\pi_{1}^{-1}(x,y),m^{\prime}=\pi_{1}^{-1}(x^{\prime},y^{\prime})$, where $y,y^{\prime}\geq 0$. To check that $\mathbf{w}_{s}(m)(m^{\prime})=\mathbf{w}_{s}(m^{\prime})(m)$, we compute both sides. The LHS is

$$(\mathbf{w}_{s}(\pi_{1}^{-1}(x,y))(\pi_{1}^{-1}(x^{\prime},y^{\prime}))=y\cdot x^{\prime}+x\cdot y^{\prime}$$

because $\mathbf{w}_{s}(\pi_{1}^{-1}(x,y))$ is defined to be $(x,-x,y)$, i.e. $p(\mathfrak{e}_{2})=x,p(\mathfrak{e}^{\prime}_{2})=-x,p(\mathfrak{e}_{1})=y$, so the computation follows from (3.4). The RHS may similarly computed to be $y^{\prime}\cdot x+x^{\prime}\cdot y$, and hence the equality holds.

Next suppose that $m,m^{\prime}\in H_{-}$. In this case we have $m=\pi_{1}^{-1}(x,y),m^{\prime}=\pi_{1}^{-1}(x^{\prime},y^{\prime}))$ with $y\leq 0$, and by definition $\mathbf{w}_{s}(\pi_{1}^{-1}(x,y))=(x-sy,-x,y)\in\mathcal{T}_{s}$. It follows from (3.4) that we have

$$\mathbf{w}_{s}(\pi_{1}^{-1}(x,y))(\pi_{1}^{-1}(x^{\prime},y^{\prime}))=yx^{\prime}-syy^{\prime}+xy^{\prime}$$

The RHS may be computed similarly to be

$$\mathbf{w}_{s}((x^{\prime},y^{\prime}),(sy^{\prime}-x^{\prime},y^{\prime}))((x,y),(sy-x,y))=y^{\prime}x-sy^{\prime}\,y+x^{\prime}y$$

so the two sides are equal, as desired. Finally, for the case $m\in H_{+},m^{\prime}\in H_{-}$, similar computations show that the LHS is equal to $yx^{\prime}+xy^{\prime}$ and the RHS is equal to $y^{\prime}x+x^{\prime}y$, so they are again equal. By symmetry, the equality holds also for the case $m\in H_{-},m^{\prime}\in H_{+}$. This concludes the proof of (1).

The condition (2) of Definition 2.9 follows immediately since the map is evidently injective, since the three parameters $p(\mathfrak{e}_{1}),p(\mathfrak{e}_{2}),p(\mathfrak{e}^{\prime}_{s})$ completely determine $p$, and is surjective by the claim of Proposition 3.3.

It remains to prove the condition (3). We have seen in the discussion above that $\Sigma(\mathcal{M}_{s})$ consists of the two cones of linearity $H_{+}$ and $H_{-}$. Moreover, in the proof of Proposition 3.3 we saw that $\textup{Sp}(\mathcal{M}_{s},1)=\{p\,\mid\,p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=0=\min\{0,s\cdot c\}\}$. In other words, in terms of coordinates on $\mathcal{T}_{s}$, the subset $\textup{Sp}(\mathcal{M}_{s},1)$ corresponds to $\{c\geq 0\}=\{a+b=\min\{0,s\cdot c\}=0\}$. Now from (3.13) it follows that the preimage under $\mathbf{w}_{s}$ of the subset $\{c\geq 0\}$ precisely $\{y\geq 0\}=H_{+}$. By a similar argument, $\textup{Sp}(\mathcal{M}_{s},2)$ is identified with $\{c\leq 0\}=\{a+b=\min\{0,s\cdot c\}=s\cdot c\}$, which again from (3.13) can be seen to have preimage $H_{-}$. Thus the preimages of $\textup{Sp}(\mathcal{M}_{s},i)$ for $i=1,2$ correspond precisely to the maximal-dimensional cones of $\Sigma(\mathcal{M}_{s})$, as desired. This concludes the proof. ∎

4. Example: a point-convex hull in $\mathcal{M}_{s}\otimes{\mathbb{R}}$

In this section, we take a moment to illustrate via one sample computation that convex geometry in the context of polyptych lattices can exhibit phenomena that are not intuitive from the classical perspective. First we recall some definitions from [3]. Given a subset $S\subset\mathcal{M}_{\mathbb{R}}$ we define the point-convex hull of $S$, denoted $\textup{p-conv}_{\mathbb{R}}(S)$, to be

(4.1) $$\textup{p-conv}_{\mathbb{R}}(S):=\bigcap_{S\subset\mathcal{H}_{p,\lambda}}\mathcal{H}_{p,\lambda}$$

where $p\in\textup{Sp}(\mathcal{M}),\lambda\in{\mathbb{Z}}$, and the intersection ranges over those choices $p,\lambda$ with $S\subset\mathcal{H}_{p,\lambda}$. Point-convexity is a natural polyptych-lattice analogue of the classical notion of convexity.

For this discussion we fix $s=1$, so the mutation $\mu_{1,2}:M_{1}\to M_{2}$ is given by $\mu_{1,2}(x,y)=(\min\{0,y\}-x,y)$. Now fix the (finite) set $S:=\{\pi_{1}^{-1}(0,0),\pi_{1}^{-1}(0,1),\pi_{1}^{-1}(0,-1)\}\subset\mathcal{M}_{s}$. Then $\pi_{2}(S)=\{(-1,-1),(0,0),(0,1)\}\subset M_{2}$. We illustrate $S$ in each of the coordinate charts in Figure 1 below.

Figure 1. The two chart images of the set $S$. On the left is $\pi_{1}(S)$ and on the right is $\pi_{2}(S)$. In what follows, we compute the point-convex hull of $S$.

We now compute $\textup{p-conv}_{\mathbb{R}}(S)$ and we will see that it can happen that $\pi_{i}(\textup{p-conv}_{\mathbb{R}}(S))$ is not the same as the classical convex hull of $\pi_{i}(S)$ in $M_{i}\otimes{\mathbb{R}}$.

To compute $\textup{p-conv}_{\mathbb{R}}(S)$, by its definition (4.1), we must first identify those PL half-spaces $\mathcal{H}_{p,\lambda}$ with the property that $S\subset\mathcal{H}_{p,\lambda}$, and then we must take the intersection of all of them. We have already seen above that any $p\in\textup{Sp}(\mathcal{M}_{s})$ is of the form

$$p_{1}(x,y)=p\circ\pi_{1}^{-1}(x,y)=\begin{cases}cx-by,\quad\textup{ if }y\leq 0\\ cx+ay,\quad\textup{ if }y\geq 0\end{cases}$$

for a triple $(a,b,c)\in{\mathbb{R}}^{3}$ satisfying $a+b=\min\{0,c\}$. To analyze the behavior of various pairs of $p$ and $\lambda$, we take cases.

First suppose $c=0$. Then $a+b=0$ and $p_{1}(x,y)=-by$ for all $(x,y)\in M_{1}$. Now if $S\subset\mathcal{H}_{p,\lambda}$ then $p_{1}(0,1)=-b\geq\lambda,p_{1}(0,0)=0\geq\lambda$, and $p_{1}(0,-1)=b\geq\lambda$. So $\lambda\leq 0$ and $b\in[\lambda,-\lambda]$. If $b=0$ then $p_{1}\equiv 0$ and $\pi_{1}(\mathcal{H}_{p,\lambda})=M_{1}\otimes{\mathbb{R}}$ so this case is trivial and we may omit it from consideration. For $b\neq 0$, it can be seen that for such a $p$ and $\lambda$, we can describe $\pi_{1}(\mathcal{H}_{p,\lambda})$ as follows. If $b<0$ then

$$\pi_{1}(\mathcal{H}_{p,\lambda})=\{(x,y)\in M_{1}\otimes{\mathbb{R}}\,\mid\,y\geq-\frac{\lambda}{b}\}$$

and if $b>0$ then

$$\pi_{1}(\mathcal{H}_{p,\lambda})=\{(x,y)\in M_{1}\otimes{\mathbb{R}}\,\mid\,y\leq\frac{\lambda}{b}\}.$$

Note that from the condition $b\in[\lambda,-\lambda]$ it follows that $\lvert\frac{\lambda}{b}\rvert\geq 1$. Second, suppose $c>0$. A similar computation shows that for $S\subset\mathcal{H}_{p,\lambda}$ to hold we again must have $\lambda\leq 0$ and $b\in[\lambda,-\lambda]$ and thus, for such $p$ and $\lambda$, we have

(4.2) $$\begin{split}\pi_{1}(\mathcal{H}_{p,\lambda})&=\{(x,y)\in M_{1}\otimes{\mathbb{R}}\,\mid\,cx-by\geq\lambda\}\\ &=\{(x,y)\in M_{1}\otimes{\mathbb{R}}\,\mid\,x-b^{\prime}y\geq\lambda^{\prime}\}\textup{ where }\,\lambda^{\prime}=\frac{\lambda}{c}\leq 0,b^{\prime}=\frac{b}{c}\in[\lambda^{\prime},-\lambda^{\prime}].\end{split}$$

We depict the different possibilities when $c>0$ in Figure 2.

Figure 2. We illustrate $\pi_{1}(\mathcal{H}_{p,\lambda})$ for the cases when $c>0$, divided into cases according to whether $b^{\prime}$ is $0$, $>0$ or $<0$.

We have noted in Figure 2 that $\lvert\frac{\lambda^{\prime}}{b^{\prime}}\rvert\geq 1$ due to the relation between $\lambda^{\prime}$ and $b^{\prime}$. Moreover, it is not difficult to see that there are choices of $a,b,c,\lambda$ such that we can obtain any value $\lambda^{\prime}\leq 0,-\lambda^{\prime}/b^{\prime}\geq 1$.

Next we consider the case $c<0$. Then

$$p_{1}(x,y)=p\circ\pi_{1}^{-1}(x,y)=\begin{cases}cx-by,\quad\textup{ if }y\leq 0\\ cx-by+cy,\quad\textup{ if }y\geq 0.\end{cases}$$

For $\lambda\in{\mathbb{Z}}$ and $p_{1}$ as above, a computation similar to those above shows that the condition $S\subset\mathcal{H}_{p,\lambda}$ is equivalent to the conditions $\lambda\leq 0,\lambda\leq b,\lambda\leq c-b$. Since the computations are similar, we omit details and record the results in Figure 3 (the cases $b=0$ and $b>0,c<0$) and Figure 4 (the cases $b=c<0$, $c<b<0$, and $b<c<0$).

Figure 3. Case (i): on the left we show the $b=0$ case. Case (ii): on the right we illustrate the case $b>0,c<0$, in which we have $\lambda\leq c-b<c<0<b$. For $x\leq\frac{\lambda}{c}$, the boundary is defined by the equation $x=(\frac{b}{c}-1)y+\frac{\lambda}{c}$. For $x\geq\frac{\lambda}{c}$, the boundary is given by $x=\frac{b}{c}y+\frac{\lambda}{c}$. The absolute value of $\frac{\lambda}{c-b}$ is larger than that of $\frac{\lambda}{c}$ since $c-b<c$.

Figure 4. Case (iii): $b=c<0$. Here we know $\lambda/c\geq 1$. Case (iv): $c<b<0$. In this case we know $\lambda/c\geq 1/2$. Case (v): $b<c<0$. In the region $x\leq\lambda/c$, the boundary is given by a linear function of slope $\frac{c}{b}$, while on the region $x\geq\lambda/c$, the boundary is given by a linear function of slope $\frac{c}{b-c}>\frac{c}{b}$.

For case (iv) in Figure 4 we remark that the lower bound $\lambda/c\geq 1/2$ is found by observing that since $\lambda\leq b$ and $\lambda\leq c-b$, we know $\frac{\lambda+b}{c}\geq 1$ and hence $\lambda/c\geq 1-b/c$. On the other hand we also know $\lambda\leq b$ so $\lambda/c\geq b/c$, so $\lambda/c\geq\max\{1-b/c,b/c\}$. By the hypotheses we know $b/c\geq 0$ so $\lambda/c$ must be larger than the min of the function $\max\{1-x,x\}$ on $[0,\infty)$ which is $1/2$. It is possible to achieve this min by selecting $2b=c$ and $\lambda=c/2$.

The above computations determine the set of $p,\lambda$ with $S\subset\mathcal{H}_{p,\lambda}$ and it now follows that the image under $\pi_{1}$ of the intersection of all such $\mathcal{H}_{p,\lambda}$ is as depicted on the left in Figure 5. By mutating, it is also immediate that $\pi_{1}(\textup{p-conv}_{\mathbb{R}}(S))$ is given by the figure on the right.

Figure 5. The image of $\textup{p-conv}_{\mathbb{R}}(S)$ in the two charts, with $\pi_{1}(S)$ on the left and $\pi_{2}(S)$ on the right.

Thus we see that the point-convex hull of the set $S$ with $\pi_{1}(S)$ equal to $3$ collinear points in $M_{1}\otimes{\mathbb{R}}$ is a $2$-dimensional triangle, and in particular is not equal to the classical convex hull of $\pi_{1}(S)$, which is the $1$-dimensional line segment connecting the points in $\pi_{1}(S)$. However, the image of $S$ under $\pi_{2}$ is not collinear, and $\pi_{2}(\textup{p-conv}_{\mathbb{R}}(S))$ is in fact equal to the classical convex hull of $\pi_{2}(S)$.

5. Examples: chart-Gorenstein-Fano polytopes in $\mathcal{M}_{s}$

We now build, by way of example, several PL polytopes in $(\mathcal{M}_{s})_{{\mathbb{R}}}$ which are chart-Gorenstein-Fano in the sense of Definition 2.14. Since $\mathcal{M}_{s}$ is rank $2$, by [3, Lemma 5.21] we expect the coordinate chart images of such a PL polytope to be classical $2$-dimensional Gorenstein-Fano polytopes. As mentioned in the Introduction, we expect these examples to be related to past work of e.g. Petracci, Ilten, and Christophersen on deformations of toric varieties and complexity-$1$ $T$-varieties.

We begin with an example for $s=1$, where we give full details. We have seen from Proposition 3.3 that a point in $\textup{Sp}(\mathcal{M}_{s})$ is specified by $3$ parameters, namely $(a,b,c)\in\mathcal{T}_{s}$ such that $a+b=s\cdot\min\{0,c\}$. Under this identification, our $3$ points $\mathsf{p},\mathsf{q},\mathsf{r}$ are specified by the choices

$$\mathsf{p}=(-2,2,1),\quad\mathsf{q}=(0,-1,-1),\quad\mathsf{r}=(1,-1,1)$$

where the triples are interpreted as elements of $\mathcal{T}_{s}$. More concretely, this means that, for example, the point $\mathsf{p}$ expressed in $M_{1}$ coordinates $(x,y)$ and $M_{2}$ coordinates $(x’,y’)$ respectively, is

$$\mathsf{p}(x,y)=x-2y,\quad\mathsf{p}(x’,y’)=\begin{cases}-x^{\prime}-2y’\quad\textup{ if }\,y\geq 0\\ -x’-y’\quad\textup{ if }\,y<0.\end{cases}$$

Note that $\mathsf{p}$ is linear on $M_{1}$. The associated PL half-space $\mathcal{H}_{\mathsf{p},-1}$ is depicted in both charts in Figure 6.

Figure 6. The two chart images of the PL half-space $\mathcal{H}_{\mathsf{p},-1}$. On the left is $\pi_{1}(\mathcal{H}_{\mathsf{p},-1})$ and on the right is $\pi_{2}(\mathcal{H}_{\mathsf{p},-1})$.

Similarly, the point $\mathsf{q}$ expressed in $M_{1}$ and $M_{2}$ coordinates is

$$\mathsf{q}(x,y)=\begin{cases}-x\quad\quad\textup{ if }\,y\geq 0\\ -x+y\quad\textup{ if }y\leq 0\end{cases}\quad\quad\quad\textup{ and }\quad\quad\quad\mathsf{q}(x’,y’)=x’$$

So $\mathsf{q}$ is linear on $M_{2}$ and not linear on $M_{1}$. The point $\mathsf{r}$ is given by

$$\mathsf{r}(x,y)=x+y,\quad\mathsf{r}(x;,y’)=\begin{cases}-x’+y’,\quad\textup{ if }\,y’\geq 0\\ -x’+2y’,\quad\textup{ if }y’\leq 0\end{cases}$$

so $\mathsf{r}$ is linear on $M_{1}$. The associated PL half-spaces $\mathcal{H}_{\mathsf{q},-1}$ and $\mathcal{H}_{\mathsf{r},-1}$ are depicted, in both charts, in Figure 7 and Figure 8 respectively.

Figure 7. The two chart images of the PL half-space $\mathcal{H}_{\mathsf{q},-1}$. On the left is $\pi_{1}(\mathcal{H}_{\mathsf{q},-1})$ and on the right is $\pi_{2}(\mathcal{H}_{\mathsf{q},-1})$.

Figure 8. The two chart images of the PL half-space $\mathcal{H}_{\mathsf{r},-1}$. On the left is $\pi_{1}(\mathcal{H}_{\mathsf{r},-1})$ and on the right is $\pi_{2}(\mathcal{H}_{\mathsf{r},-1})$.

The intersection of these $3$ PL half-spaces is then a PL polytope $\mathcal{P}$. We depict both coordinate chart images of $\mathcal{P}$ in Figure 9. We note that the two coordinate chart images $\pi_{1}(\mathcal{P})$ and $\pi_{2}(\mathcal{P})$ are both Gorenstein-Fano polytopes in the classical sense. Indeed, in the list of all $2$-dimensional Gorenstein-Fano polytopes (up to lattice isomorphism) given in [2, p. 382], the two polytopes are of type 4b and 4c respectively. These are related by the piecewise-linear mutation $\mu_{1,2}:M_{1}\to M_{2}$.

Figure 9. The two chart images of the PL polytope $\mathcal{P}=\mathcal{H}_{\mathsf{p},-1}\cap\mathcal{H}_{\mathsf{q},-1}\cap\mathcal{H}_{\mathsf{r},-1}$. On the left is $\pi_{1}(\mathcal{P})$ and on the right is $\pi_{2}(\mathcal{P})$. Both chart images are Gorenstein-Fano polytopes in the classical sense.

In the classical setting, the dual of a Gorenstein-Fano polytope is again a lattice polytope. We now explicitly compute $\mathcal{P}^{\vee}$ for the example above to see that this is also the case in this specific example. Note that since $\mathcal{M}_{s}$ is self-dual, both $\mathcal{P}$ and $\mathcal{P}^{\vee}$ are PL polytopes in $(\mathcal{M}_{s})_{{\mathbb{R}}}$. A computationally effective method of computing $\mathcal{P}^{\vee}$ is given in [3, Lemma 5.16] which states

(5.1) $$\mathcal{P}^{\vee}=\bigcap_{m\in\mathrm{V}(\mathcal{P})}\mathcal{H}_{\mathbf{v}(m),-1}$$

where $\mathrm{V}(\mathcal{P})$ denotes the set of vertices of $\mathcal{P}$ as in (2.13) and $\mathbf{v}$ is the strict dual pairing. To take advantage of this characterization, we need the vertices of $\mathcal{P}$ in our example. It is straightforward to compute

$$\mathrm{V}(\mathcal{P})=\{(1,1),(-1,1)),((1,0),(-1,0)),((0,-1),(-1,-1)),((-1,0),(1,0))\}.$$

In order to interpret the vertices as points in $\textup{Sp}(\mathcal{M}_{s})$, we must take their images under the strict dual pairing $\mathbf{w}_{s}$ of $\mathcal{M}_{s}$, as given in (3.13), which takes values in $\mathcal{T}_{s}\cong\textup{Sp}(\mathcal{M}_{s})$. We can then use these images to define the PL half-spaces in the RHS of (5.1). The relevant data is summarized in the table below.

$m\in\mathrm{V}(\mathcal{P})$	$p(\mathfrak{e}_{2})$	$p(\mathfrak{e}’_{2})$	$p(\mathfrak{e}_{1})$	on $M_{1}$	on $M_{2}$
((1,1),(-1,1))	1	-1	1	$x+y$	$\begin{cases}-x’+y’\,\textup{ if }y’\geq 0\\ -x’+2y’\,ify’\leq 0\end{cases}$
((1,0),(-1,0))	1	-1	0	$y$	$y’$
((0,-1),(-1,-1))	0	-1	-1	$\begin{cases}-x+y\,\textup{ if }\,y\leq 0\\ -x\,\textup{ if }\,y\geq 0\end{cases}$	$x’$
((-1,0),(1,0))	-1	1	0	$-y$	$-y’$

The intersection of the $4$ PL half-spaces $\mathcal{H}_{\mathbf{w}_{s}(m),-1}$ for the $4$ vertices in $\mathrm{V}(\mathcal{P})$ is depicted in both coordinate charts in Figure 10. It is not difficult to see that the two chart images are equivalent up to a transformation in $SL(2,{\mathbb{Z}})$, hence are lattice-equivalent. In the list of 2-dimensional Gorenstein-Fano polytopes given in [2], these two polytopes correspond to type 7b.

Figure 10. The two chart images of the PL dual polytope $\mathcal{P}^{\vee}$. On the left is $\pi_{1}(\mathcal{P}^{\vee})$ and on the right is $\pi_{2}(\mathcal{P}^{\vee})$.

We now proceed to record several more examples of chart-Gorenstein-Fano polytopes in $\mathcal{M}_{s}$. Since the computations are similar to those given above, we do not give details. We should emphasize here that we do not claim any general classification results or existence results; we leave this open for future exploration.

In the examples below, we record the representatives in $\mathcal{T}_{s}$ of the points with respect to which we define the half-spaces $\mathcal{H}_{p,-1}$ whose intersection is the PL polytope $\mathcal{P}$. We also illustrate the coordinate chart images of $\mathcal{P}$ in $M_{1}$ and $M_{2}$, as well as the coordinate chart images of the dual polytope $\mathcal{P}^{\vee}$.

5.2 Example.

We begin with another example in the $s=1$ case. We represent the five points in $\textup{Sp}(\mathcal{M}_{1})$ as elements in $\mathcal{T}_{1}$ using the identification given in Section 3. With this understood, the points are

$$\mathsf{p}=(-1,0,-1),\quad\mathsf{q}=(1,-1,0),\quad\mathsf{r}=(-1,1,0),\quad\mathsf{s}=(0,0,1),\quad\mathsf{t}=(1,-1,1).$$

So the polytope $\mathcal{P}$ is the intersection

$$\mathcal{P}=\mathcal{H}_{\mathsf{p},-1}\cap\mathcal{H}_{\mathsf{q},-1}\cap\mathcal{H}_{\mathsf{r},-1}\cap\mathcal{H}_{\mathsf{s},-1}\cap\mathcal{H}_{\mathsf{t},-1}$$

where we have chosen all parameters $a$ in the definition of the half-spaces to be equal to $-1$, since we wish to describe a chart-Gorenstein-Fano PL polytope. It is then straightforward to compute that the chart images of $\mathcal{P}$ are as given in Figure 11.

Figure 11. We illustrate the chart images of the chart-Gorenstein-Fano polytope $\mathcal{P}$ defined by the five given points. The image $\pi_{1}(\mathcal{P})$ in $M_{1}\otimes{\mathbb{R}}$ is illustrated on the left, $\pi_{2}(\mathcal{P})\subset M_{2}\otimes{\mathbb{R}}$ is on the right.

The vertices of this PL polytope can be seen to have chart image $\{(1,0),(0,1),(-1,1),(-1,0),(0,-1),(1,-1)\}$. Using the formula for ${\sf w}_{s}$ given in (3.13) we may then compute its associated points and the corresponding dual polytope $\mathcal{P}^{\vee}$. We illustrate the resulting dual PL polytope $\mathcal{P}^{\vee}$ in Figure 12.

Figure 12. The dual PL polytope $\mathcal{P}^{\vee}$ of $\mathcal{P}$. We depict its chart image in $M_{1}$ on the left, and the image in $M_{2}$ on the right.

We now begin an exploration of examples for the cases when $s>1$.

5.3 Example.

In this example we take $s=2$, so the mutation is now $\mu_{1,2}(x,y)=(\min\{0,2y\}-x,y)$. In this case, the set $\mathcal{T}_{2}$ parametrizing the set of points $\textup{Sp}(\mathcal{M}_{2})$ is $\mathcal{T}_{2}:=\{(a,b,c)\in{\mathbb{Z}}^{3}\,\mid\,a+b=\min\{0,2c\}\}$. We consider the following set of four points in $\mathcal{T}_{2}$:

$$\mathsf{p}=(-1,-1,-1),\quad\textsf{q}=(1,-1,1),\quad\mathsf{r}=(0,0,1),\quad\mathsf{s}=(-1,1,0).$$

Then the chart-Gorenstein-Fano polytope $\mathcal{P}$ is given by the half-spaces defined by the above points, with parameter $-1$. We illustrate its chart images in Figure 13.

Figure 13. Here we depict the chart-Gorenstein-Fano PL polytope $\mathcal{P}$ in $\mathcal{M}_{2}\otimes{\mathbb{R}}$ given by the above points. The image $\pi_{1}(\mathcal{P})$ is on the left and $\pi_{2}(\mathcal{P})$ is on the right.

The set of vertices of this PL polytope can be seen to have image in $\pi_{1}$ equal to $\{(0,1),(1,0),(0,-1),(-1,0),(-1,1)\}$. Following the procedure already established we may compute the dual PL polytope $\mathcal{P}^{\vee}$; we depict the result in Figure 14.

Figure 14. The two chart images of the dual PL polytope $\mathcal{P}^{\vee}$ in $\mathcal{M}_{2}\otimes{\mathbb{R}}$, with the chart image in $M_{1}$ on the left and $M_{2}$ on the right.

5.4 Example.

We continue with an example for $s=3$. Since the details are similar as for the previous cases, we will be brief. We choose points in $\mathcal{T}_{3}$ as follows:

$$\mathsf{p}=(-2,-1,-1),\quad\textsf{q}=(1,-1,1),\quad\mathsf{r}=(0,0,1).$$

The corresponding $\mathcal{P}$ is depicted in Figure 15.

Figure 15. A chart-Gorenstein-Fano polytope $\mathcal{P}$ in $\mathcal{M}_{3}\otimes{\mathbb{R}}$ corresponding to the given $3$ points. The $M_{1}$ image is on the left and $M_{2}$ on the right.

The vertices of $\mathcal{P}$ have image in $M_{1}$ given by $\{(1,0),(-1,0),(-1,1),(0,-1)\}$. The dual PL polytope $\mathcal{P}^{\vee}$ is depicted in Figure 16.

Figure 16. The two chart images of the dual PL polytope $\mathcal{P}^{\vee}$ in $\mathcal{M}_{3}\otimes{\mathbb{R}}$. The image in $M_{1}$ is on the left and $M_{2}$ on the right.

5.5 Example.

Finally, we give an example for the $s=4$ case. The points chosen are

$$\mathsf{p}=(-2,-2,-1),\quad\mathsf{q}=(0,0,1).$$

It is useful to note that this exhibits different behavior of the PL situation from the classical one, since we may define a bounded PL polytope with only two PL half-spaces. The resulting PL polytope $\mathcal{P}$ is shown in Figure 17. The vertices $V(\mathcal{P})$ have chart image in $M_{1}$ given by $\{(1,0),(-1,1),(-1,0),(-1,-1)\}$. The dual PL polytope $\mathcal{P}^{\vee}$ is given in Figure 18.

Figure 17. The two chart images of the chart-Gorenstein-Fano PL polytope $\mathcal{P}$ in $\mathcal{M}_{4}\otimes{\mathbb{R}}$. The image in $M_{1}$ is on the left and $M_{2}$ on the right.

Figure 18. The two chart images of the dual PL polytope $\mathcal{P}^{\vee}$. The chart image in $M_{1}$ is on the left and $M_{2}$ on the right, though in fact they are the same.

We finish with an example of a chart-Gorenstein-Fano PL polytope $\mathcal{P}$ in the case $s=1$ that has the property that its dual PL polytope is not an integral polytope. This example, together with [3, Example 5.17], suggest that the convex geometry of dual PL polytopes is subtle.

5.6 Example.

Here we choose $s=1$ so we are working in $\mathcal{M}_{1}$. The points chosen are

$$\mathsf{p}=(0,0,1),\quad\mathsf{q}=(2,-2,1),\quad\mathsf{r}=(-1,0,-1).$$

As usual we choose all parameters $a_{i}$ defining the half-spaces to be equal to $-1$. The resulting chart-Gorenstein-Fano PL polytope $\mathcal{P}$ is shown in Figure 19. The vertices $V(\mathcal{P})$ have chart image in $M_{1}$ given by $\{(1,0),(1,-1),(-1,0),(-1,2)\}$. The dual PL polytope $\mathcal{P}^{\vee}$ is given in Figure 20. As we can see from Figure 20, the dual PL polytope $\mathcal{P}^{\vee}$ is not an integral polytope due to the presence of the vertex $(0.5,0)$.

Figure 19. The two chart images of the chart-Gorenstein-Fano PL polytope $\mathcal{P}$ in $\mathcal{M}_{1}\otimes{\mathbb{R}}$. The image in $M_{1}$ is on the left and $M_{2}$ on the right.

Figure 20. The two chart images of the dual PL polytope $\mathcal{P}^{\vee}$. The chart image in $M_{1}$ is on the left and $M_{2}$ on the right. Note that the chart image in $M_{2}$ is not an integral polytope.

6. The polyptych lattice $\mathcal{M}_{s}$ is detropicalizable

Having explored convex geometry in Sections 4 and 5, we now return to the algebra and algebraic geometry. Our main goal in this section is to exhibit a detropicalization of the polyptych lattice $\mathcal{M}_{s}$, thus showing that $\mathcal{M}_{s}$ is detropicalizable. To accomplish this goal, we need some preliminaries.

We first identify the coordinate charts $M_{1}$ and $M_{2}$ of $\mathcal{M}_{s}$ with sublattices of ${\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}$. Define

$$M_{s}(1):=\{(a_{1},a_{2},b_{1},b_{2})\in{\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}\,\mid\,a_{2}=b_{1}=0\}=\{(a_{1},0,0,b_{2})\}$$

and

$$M_{s}(2):=\{(c_{1},c_{2},d_{1},d_{2})\in{\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}\,\mid\,c_{2}=d_{2}=0\}=\{(c_{1},0,d_{1},0)\}.$$

Now we choose identifications

(6.1) $$\Theta_{1}:M_{1}\to M_{s}(1),\quad(x,y)\mapsto(y,0,0,x)$$

and

(6.2) $$\Theta_{2}:M_{2}\to M_{s}(2),\quad(u,v)\mapsto(v,0,u,0).$$

If we define a mutation map

(6.3) $$\tilde{\mu}_{1,2}:M_{s}(1)\to M_{s}(2),\quad(a,0,0,b)\mapsto(a,0,\min\{0,s\cdot a\}-b,0)$$

from $M_{s}(1)\to M_{s}(2)$ then it is straightforward to check that the following diagram commutes:

so we may realize $\mathcal{M}_{s}$ in terms of the coordinate charts $M_{s}(1),M_{s}(2)$ in place of $M_{1},M_{2}$. This will be convenient for some of our arguments below. We also specify a subset $\mathbb{M}_{s}$ of ${\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}$ which we will identify with (the set of elements of) $\mathcal{M}_{s}$ as follows:

$$\mathbb{M}_{s}:=\left\{(w_{1},w_{2},z_{1},z_{2})\in{\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}\,\mid\,\min\{w_{1},w_{2}\}=0,\,-s\cdot w_{2}=z_{1}+z_{2}\right\}.$$

We then define maps (which we may think of as projections to coordinate charts) $\Psi_{i}:\mathbb{M}_{s}\to M_{s}(i)$ by

$$\Psi_{1}(\bar{w},\bar{z}):=\tilde{\pi}_{1}\left(\bar{w}+\frac{1}{s}\langle\bar{z},\mathbf{1}\rangle\mathbf{1},\bar{z}\right)\in M_{s}(1)$$

and

$$\Psi_{2}(\bar{w},\bar{z}):=\tilde{\pi}_{2}\left(\bar{w}+\frac{1}{s}\langle\bar{z},\mathbf{1}\rangle\mathbf{1},\bar{z}\right)\in M_{s}(2)$$

where $\bar{w}=(w_{1},w_{2}),\bar{z}=(z_{1},z_{2})$, $\mathbf{1}=(1,1)$, the pairing $\langle\cdot,\cdot\rangle$ denotes the standard inner product, and $\tilde{\pi}_{i}$ denotes the map which sets the $z_{i}$ coordinate equal to $0$. Note that $\frac{1}{s}\langle\bar{z},\mathbf{1}\rangle=\frac{1}{s}(z_{1}+z_{2})$ is an integer by the hypotheses on the vectors in $\mathbb{M}_{s}$. In particular, it follows that we may rewrite $\Psi_{1}$ and $\Psi_{2}$ as follows:

(6.4) $$\Psi_{1}(w_{1},w_{2},z_{1},z_{2})=(w_{1}-w_{2},0,0,z_{2})$$

and

(6.5) $$\Psi_{2}(w_{1},w_{2},z_{1},z_{2})=(w_{1}-w_{2},0,z_{1},0).$$

It is straightforward to compute the inverse of $\Psi_{1}$ to be

(6.6) $$\Psi_{1}^{-1}(a,0,0,b)=\begin{cases}(a,0,-b,b)\,\,\quad\textup{ if }a\geq 0\\ (0,-a,sa-b,b)\,\,\textup{ if }a\leq 0\end{cases}$$

from which it immediately follows that

$$\Psi_{2}\circ\Psi_{1}^{-1}(a,0,0,b)=(a,0,\min\{0,s\cdot a\}-b,0).$$

Note that this is the same as the mutation map $\tilde{\mu}_{1,2}$ of (6.3).

Just as we have identified a subset of ${\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}$ with $\mathcal{M}_{s}$, we now define a subset of ${\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}$ which corresponds to the space of points $\textup{Sp}(\mathcal{M}_{s})$ as follows. We define

(6.7) $$\mathbb{T}_{s}:=\{(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2})\in{\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}\,\mid\,\alpha_{1}+\alpha_{2}=s\cdot\min\{\beta_{1},\beta_{2}\},\,\,\beta_{2}=0\}.$$

We also define

(6.8) $$\mathbb{T}_{s}(1):=\{(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2})\in\mathbb{T}_{s}\,\mid\,\alpha_{1}+\alpha_{2}=s\cdot\beta_{1}\}$$

and

(6.9) $$\mathbb{T}_{s}(2):=\{(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2})\in\mathbb{T}_{s}\,\mid\,\alpha_{1}+\alpha_{2}=s\cdot\beta_{2}=0\}$$

so $\mathbb{T}_{s}(1)$ consists of those tuples $(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2})\in\mathbb{T}_{s}$ where $\beta_{1}\leq\beta_{2}=0$, whereas $\mathbb{T}_{s}(2)$ consists of the tuples where $\beta_{2}=0\leq\beta_{1}$.

Given a tuple $(\bar{\alpha},\bar{\beta})=(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2})\in\mathbb{T}_{s}$, we may define an associated function $f_{(\bar{\alpha},\bar{\beta})}$ on $\mathbb{M}_{s}$ as follows:

(6.10) $$f_{(\bar{\alpha},\bar{\beta})}(w_{1},w_{2},z_{1},z_{2}):=\alpha_{1}w_{1}+\alpha_{2}w_{2}+\beta_{1}z_{1}+\beta_{2}z_{2}=\alpha_{1}w_{1}+\alpha_{2}w_{2}+\beta_{1}z_{1}$$

where the last equality holds since $\beta_{2}=0$ by assumption. Notice that $f_{(\bar{\alpha},\bar{\beta})}$ is simply the restriction to $\mathbb{M}_{s}$ of the usual standard inner product pairing with $(\bar{\alpha},\bar{\beta})$, but since $\mathbb{M}_{s}$ is not a linear subspace (additive subgroup) of ${\mathbb{Z}}^{2}\times{\mathbb{Z}}^{2}$, we cannot discuss linearity on $\mathbb{M}_{s}$, and in particular, it is not a linear map. Using a sequence of bijections $\pi_{1}:\mathcal{M}_{s}\to M_{1}$, $\Theta_{1}:M_{1}\to M_{s}(1)$, and $\Psi_{1}^{-1}:M_{s}(1)\to\mathbb{M}_{s}$, we may pullback the function $f_{(\bar{\alpha},\bar{\beta})}$ to $\mathcal{M}_{s}$, thus defining a function on $\mathcal{M}_{s}$. The following lemma shows that this association gives a bijection from $\mathbb{T}_{s}$ to $\textup{Sp}(\mathcal{M}_{s})$.

6.11 Lemma.

Let $\Phi$ denote the association $(\bar{\alpha},\bar{\beta})\in\mathbb{T}_{s}\mapsto p_{(\bar{\alpha},\bar{\beta})}:=f_{(\bar{\alpha},\bar{\beta})}\circ\Psi_{1}^{-1}\circ\Theta_{1}\circ\pi_{1}:\mathcal{M}_{s}\to{\mathbb{Z}}$ where $f_{(\bar{\alpha},\bar{\beta})}$ is the function on $\mathbb{M}_{s}$ defined above. Then

1. $\Phi$ has image $\textup{Sp}(\mathcal{M}_{s})$, and

2. $\Phi$ defines a bijection from $\mathbb{T}_{s}$ to $\textup{Sp}(\mathcal{M}_{s})$, and

3. $\Phi$ respects addition when restricted to $\mathbb{T}_{s}(i)$ for $i=1$ or $i=2$. More precisely, for fixed $i$ and for $(\bar{\alpha},\bar{\beta}),(\bar{\alpha}^{\prime},\bar{\beta}^{\prime})\in\mathbb{T}_{s}(i)$, we have $\Phi((\bar{\alpha},\bar{\beta})+(\bar{\alpha}^{\prime},\bar{\beta}^{\prime}))=\Phi(\bar{\alpha},\bar{\beta})+\Phi(\bar{\alpha}^{\prime},\bar{\beta}^{\prime})$ as functions on $\mathcal{M}_{s}$.

Proof.

Since $\pi_{1}$ is a bijection, a point in $\textup{Sp}(\mathcal{M}_{s})$ can be uniquely described by its values on $M_{1}$, so we may consider instead the function $f_{(\bar{\alpha},\bar{\beta})}\circ\Psi_{1}^{-1}\circ\Theta_{1}:M_{1}\to{\mathbb{Z}}$ and verify that it has the form given in Proposition 3.3. From the formulas for $\Theta_{1}$ and $\Psi_{1}^{-1}$ it is straightforward to compute that

(6.12) $$f_{(\bar{\alpha},\bar{\beta})}\circ\Psi_{1}^{-1}\circ\Theta_{1}(x,y)=\begin{cases}-\beta_{1}x+(s\cdot\beta_{1}-\alpha_{2})y\,\,\,\textup{ if }y\leq 0,\\ -\beta_{1}x+\alpha_{1}y\,\,\quad\quad\quad\quad\textup{ if }y\geq 0.\\ \end{cases}$$

In notation motivated by Proposition 3.3, set $p(\mathfrak{e}_{1})=-\beta_{1},p(\mathfrak{e}_{2})=\alpha_{1}$, and $p(\mathfrak{e}^{\prime}_{2})=-(s\cdot\beta_{1}-\alpha_{2})$. Now by Proposition 3.3, the function (6.12) is an element of $\textup{Sp}(\mathcal{M}_{s})$ if and only if $p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=\min\{0,s\cdot p(\mathfrak{e}_{1})\}$. We have

$$\begin{split}p(\mathfrak{e}_{2})+p(\mathfrak{e}^{\prime}_{2})=\min\{0,s\cdot p(\mathfrak{e}_{1})\}&\iff\alpha_{1}+\alpha_{2}-s\cdot\beta_{1}=\min\{0,-s\cdot\beta_{1}\}\\ &\iff\alpha_{1}+\alpha_{2}=\min\{s\cdot\beta_{1},0\}\\ &\iff\alpha_{1}+\alpha_{2}=s\min\{\beta_{1},0\}\,\,\textup{ since }s\geq 0.\\ \end{split}$$

From the above reasoning, it follows that if $(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2}=0)$ is in $\mathbb{T}_{s}$, then (6.12) is in $\textup{Sp}(\mathcal{M}_{s})$. Moreover, for any $p\in\textup{Sp}(\mathcal{M}_{s})$ with corresponding values of $p(\mathfrak{e}_{1}),p(\mathfrak{e}_{2})$ and $p(\mathfrak{e}^{\prime}_{2})$, we can take $\alpha_{1}=p(\mathfrak{e}_{2}),\beta_{1}=-p(\mathfrak{e}_{1}),\alpha_{2}=p(\mathfrak{e}^{\prime}_{2})+s\cdot\beta_{1}$ and $\beta_{2}=0$ to obtain $p$ as $\Phi(\alpha_{1},\alpha_{2},\beta_{1},\beta_{2})=p$, so $\Phi$ is a bijection as claimed.

The last claim follows from the fact that $f_{(\overline{\alpha},\overline{\beta})}$ is defined as the restriction of the standard inner product, which is linear in both variables, and the fact that $\mathbb{T}_{s}(i)$ is closed under addition for both $i=1$ and $i=2$. ∎

Motivated by the above proof, we define the following bijection between $\mathcal{T}_{s}$ and $\mathbb{T}_{s}$:

(6.13) $$\Upsilon:\mathcal{T}_{s}\to\mathbb{T}_{s},\quad(a,b,c)\mapsto(a,b-sc,-c,0).$$

Next, we wish to translate the self-dual pairing $\mathbf{w}_{s}:\mathcal{M}_{s}\to\textup{Sp}(\mathcal{M}_{s})$ from Section 3 into the language of $\mathbb{M}_{s}$ and $\mathbb{T}_{s}$. Using the identifications $\Psi_{1}:\mathbb{M}_{s}\to M_{s}(1),\Theta^{-1}:M_{s}(1)\to M_{1}$, and $\Upsilon:\mathcal{T}_{s}\to\mathbb{T}_{s}$, it is straightforward to compute that when the map $\tilde{\mathbf{w}}_{s}:=\Upsilon\circ\mathbf{w}_{s}\circ\Theta^{-1}\circ\Psi_{1}:\mathbb{M}_{s}\to\mathbb{T}_{s}$ given by composition of $\mathbf{w}_{s}$ with these identifications is given by

(6.14) $$\tilde{\mathbf{w}}_{s}(w_{1},w_{2},z_{1},z_{2})=\begin{cases}(z_{2},-z_{2}-s(w_{1}-w_{2}),-(w_{1}-w_{2}),0)\quad\quad\textup{ if }\,w_{1}-w_{2}\geq 0\\ (z_{2},-z_{2},-(w_{1}-w_{2}),0)\quad\quad\quad\quad\quad\quad\textup{ if }\,w_{1}-w_{2}\leq 0.\end{cases}$$

Recall that tuples $(w_{1},w_{2},z_{1},z_{2})$ in $\mathbb{M}_{s}$ satisfy $\min\{w_{1},w_{2}\}=0$. Thus the conditions $w_{1}-w_{2}\geq 0$ and $w_{1}-w_{2}\leq 0$ can be rephrased as $w_{2}=0$ and $w_{1}=0$ respectively, and this reformulation is also used below.

Let us now consider the following algebra:

(6.15) $$\mathcal{A}_{s}=\mathbb{C}[x_{1},x_{2},y_{1},y_{2},y_{1}^{-1},y_{2}^{-1}]/\langle x_{1}x_{2}-y_{1}^{s}-y_{2}^{s},~{}y_{2}-1\rangle.$$

It is straightforward to see that $\mathcal{A}_{s}$ is a Noetherian ${\mathbb{C}}$-algebra and an integral domain. Our main goal of this section is to prove that $\mathcal{A}_{s}$ can be equipped with a valuation $\mathfrak{v}_{s}$ in such a way that the pair $(\mathcal{A}_{s},\mathfrak{v}_{s})$ is a detropicalization of $\mathcal{M}_{s}$. To do this, we first identify an additive basis of $\mathcal{A}_{s}$. Consider the following set:

(6.16) $$\mathbb{B}_{s}:=\{x_{1}^{w_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}~{}|~{}(w_{1},w_{2},z_{2},z_{2})\in\mathbb{M}_{s}\}\subset\mathcal{A}_{s}.$$

To see that $\mathbb{B}_{s}$ forms an additive basis for $\mathcal{A}_{s}$, we may argue in two steps. First suppose that the defining ideal consists of the single relation $x_{1}x_{2}-y_{1}^{s}-y_{2}^{2}$. Then, since there is a monomial ordering $<$ such that the initial term of this relation is $x_{1}x_{2}$, it is immediate from standard results of Gröbner bases [7, Proposition 1.1] that the monomials $x_{1}^{w_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}$ with $\min\{w_{1},w_{2}\}=0$ form a basis for ${\mathbb{C}}[x_{1},x_{2},y_{1},y_{2}]/\langle x_{1}x_{2}-y_{1}^{s}-y_{2}^{s}\rangle$. For $\mathcal{A}_{s}$, however, we also have the additional defining relation $y_{2}=1$. This means that we may take as additive basis a set of monomials where the exponent $z_{2}$ of $y_{2}$ is uniquely determined by the exponents on the other variables. (It would be conventional simply to pick $z_{2}=0$ at all times, but it will be more convenient for us to pick $z_{2}$ to be a function of $w_{1},w_{2},z_{1}$.) In our setting, we choose $z_{2}=-z_{1}-sw_{2}$. This argument shows that $\mathbb{B}_{s}$ is an additive basis of $\mathcal{A}_{s}$. We record this statement in the following.

6.17 Lemma.

The image of the set $\mathbb{B}_{s}$ under the projection ${\mathbb{C}}[x_{1},x_{2},y_{1}^{\pm},y_{2}^{\pm}]\to\mathcal{A}_{s}$ is an additive basis for $\mathcal{A}_{s}$. In particular, there is a one-to-one correspondence between $\mathbb{B}_{s}$ and $\mathbb{M}_{s}$, given by taking the exponent vector of a monomial in $\mathbb{B}_{s}$.

We are now ready to define a valuation $\mathfrak{v}_{s}$ which realizes $\mathcal{A}_{s}$ as a detropicalization of $\mathcal{M}_{s}$. Let $\mathfrak{v}_{s}:\mathbb{B}_{s}\rightarrow Sp(\mathcal{M}_{b})$ be the function defined as follows. For any $(w_{1},w_{2},z_{1},z_{2})\in\mathbb{M}_{s}$, we have just seen that the monomial $x_{1}^{w_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}$ is in $\mathbb{B}_{s}$ and we define:

(6.18) $$\mathfrak{v}_{s}(x_{1}^{w_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}):=\Phi\circ\tilde{\mathbf{w}}_{s}(w_{1},w_{2},z_{1},z_{2})$$

where $\Phi:\mathbb{T}_{s}\to\textup{Sp}(\mathcal{M}_{s})$ is the bijection constructed in Lemma 6.11.

Since $\mathbb{B}_{s}$ is a basis of $\mathcal{A}_{s}$, we may then extend $\mathfrak{v}_{s}$ to a function on the algebra $\mathcal{A}_{s}$ by defining

(6.19) $$\mathfrak{v}_{s}(\sum\lambda_{i}\mathbb{b}_{i}):=\bigoplus\mathfrak{v}(\mathbb{b}_{i})\in P_{\mathcal{M}_{s}}$$

for any linear combination $\sum_{i}\lambda_{i}\mathbb{b}_{i}$ of elements $\mathbb{b}_{i}\in\mathbb{B}_{i}$ where $\lambda_{i}\in\mathbb{C}$ are scalars. By definition we also set $\mathfrak{v}_{s}(0):=\infty$. Recall that we think of elements of $\textup{Sp}(\mathcal{M}_{s})$ as piecewise-linear functions on $\mathcal{M}_{s}$ and the operation $\oplus$ is the min-combination of functions.

We are now ready to state and prove the main theorem.

6.20 Theorem.

Let $s$ be a positive integer and let $\mathcal{M}_{s}$ be the polyptych lattice asociated to $s$ defined in Section 3. Let $\mathcal{A}_{s}$ be the ${\mathbb{C}}$-algebra defined in (6.15) and let $\mathfrak{v}_{s}:\mathcal{A}_{s}\to P_{\mathcal{M}_{s}}$ be the map defined in (6.19). Then:

1. $\mathfrak{v}_{s}:\mathcal{A}_{s}\to P_{\mathcal{M}_{s}}$ is a valuation of on $\mathcal{A}_{s}$ with values in the idempotent semialgebra $P_{\mathcal{M}_{s}}$ in the sense of Definition 2.10, and

2. the pair $(\mathcal{A}_{s},\mathfrak{v}_{s})$ is a detropicalization of $\mathcal{M}_{s}$ in the sense of Definition 2.11.

Proof of Theorem 6.20.

We begin with the claim (1). To prove it, we must check the conditions for a valuation as listed in Definition 2.10. Suppose that $f,g\in\mathcal{A}_{s}$. We wish to prove that $\mathfrak{v}_{s}(fg)=\mathfrak{v}(f)\odot\mathfrak{v}(g)$. Recalling that the $\odot$ operation in $P_{\mathcal{M}_{s}}$ is given by pointwise addition of functions, this is equivalent to showing that $\mathfrak{v}_{s}(fg)=\mathfrak{v}_{s}(f)+\mathfrak{v}_{s}(g)$ as functions on $\mathcal{M}_{s}$.

We take cases. First suppose that $f=\mathbb{b}=x_{1}^{w_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}\in\mathbb{B}_{s},g=\mathbb{b}^{\prime}=x_{1}^{w^{\prime}_{1}}x_{2}^{w^{\prime}_{2}}y_{1}^{z^{\prime}_{1}}y_{2}^{z^{\prime}_{2}}\in\mathbb{B}_{s}$, so both $f$ and $g$ are monomials with exponent vectors contained in $\mathbb{M}_{s}$, and additionally assume that the product monomial $\mathbb{b}\mathbb{b}^{\prime}$ is in $\mathbb{B}_{s}$, i.e., $(w_{1}+w^{\prime}_{1},w_{2}+w^{\prime}_{2},z_{1}+z^{\prime}_{1},z_{2}+z^{\prime}_{2})$ is contained in $\mathbb{M}_{s}$. By definition, $\mathfrak{v}_{s}(\mathbb{b}\mathbb{b}^{\prime})=\Phi\circ\tilde{\mathbf{w}}_{s}(\mathbb{b}\mathbb{b}^{\prime})$, and if $\mathbb{b},\mathbb{b}^{\prime}\in\mathbb{B}_{s}$ then this implies that either $w_{1}=w^{\prime}_{1}=0$ or $w_{2}=w^{\prime}_{2}=0$. In either case, the definition of $\tilde{\mathbf{w}}_{s}$ in (6.14) implies that $\tilde{\mathbf{w}}_{s}(w_{1}+w^{\prime}_{1},w_{2}+w^{\prime}_{2},z_{1}+z^{\prime}_{1},z_{2}+z^{\prime}_{2})=\tilde{\mathbf{w}}_{s}(w_{1},w_{2},z_{1},z_{2})+\tilde{\mathbf{w}}_{s}(w^{\prime}_{1},w^{\prime}_{2},z^{\prime}_{1},z^{\prime}_{2})$, and moreover, all three images under $\tilde{\mathbf{w}}_{s}$ lie in $\mathbb{T}_{s}(i)$ for some $i$. Then Lemma 6.11(3) implies that $\Phi\circ\tilde{\mathbf{w}}_{s}$ is also additive on $(w_{1},w_{2},z_{1},z_{2})+(w^{\prime}_{1},w^{\prime}_{2},z^{\prime}_{1},z^{\prime}_{2})$, so by definition $\mathfrak{v}_{s}(\mathbb{b}\mathbb{b}^{\prime})=\mathfrak{v}_{s}(\mathbb{b})+\mathfrak{v}_{s}(\mathbb{b}^{\prime})$ in this case.

Next we consider the case $f=\mathbb{b}=x_{1}^{w_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}\in\mathbb{B}_{s},g=\mathbb{b}^{\prime}=x_{1}^{w^{\prime}_{1}}x_{2}^{w^{\prime}_{2}}y_{1}^{z^{\prime}_{1}}y_{2}^{z^{\prime}_{2}}\in\mathbb{B}_{s}$, where this time we suppose that $\mathbb{b}\mathbb{b}^{\prime}$ is not in $\mathbb{B}_{s}$. This means that $w_{1}+w^{\prime}_{1}>0$ and $w_{2}+w^{\prime}_{2}>0$. Since we know that $\min\{w_{1},w_{2}\}=0=\min\{w^{\prime}_{1},w^{\prime}_{2}\}$ by assumption, we may assume without loss of generality that $w_{1}=0,w_{2}>0,w^{\prime}_{1}>0,w^{\prime}_{2}=0$. In order to prove $\mathfrak{v}(\mathbb{b}\mathbb{b}^{\prime})=\mathfrak{v}(\mathbb{b})+\mathfrak{v}(\mathbb{b}^{\prime})$, we compute both sides as functions on $\mathcal{M}_{s}$. Since $\mathbb{b},\mathbb{b}^{\prime}\in\mathbb{B}_{s}$, the RHS may be computed from the definitions to be

$$\Phi\circ\tilde{\mathbf{w}}_{s}(\mathbb{b})+\Phi\circ\tilde{\mathbf{w}}_{s}(\mathbb{b}^{\prime})=f_{(z_{2},-z_{2},w_{2},0)}+f_{(z^{\prime}_{2},-z^{\prime}_{2}-sw^{\prime}_{1},-w^{\prime}_{1},0)}.$$

(Here by slight abuse of notation we view functions on $\mathbb{M}_{s}$ as functions on $\mathcal{M}_{s}$ via the identifications we established above.) For the LHS, we must first express $\mathbb{b}\mathbb{b}^{\prime}$ as a linear combination of monomials in $\mathbb{B}_{s}$. Consider the case $w^{\prime}_{1}\leq w_{2}$. Then we have

(6.21) $$\begin{split}x_{1}^{w^{\prime}_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}+z^{\prime}_{1}}y_{2}^{z_{2}+z^{\prime}_{2}}&=(x_{1}x_{2})^{w^{\prime}_{1}}x_{2}^{w_{2}-w^{\prime}_{1}}y_{1}^{z_{1}+z^{\prime}_{1}}y_{2}^{z_{2}+z^{\prime}_{2}}\\ &=(y_{1}^{s}+y_{2}^{s})^{w^{\prime}_{1}}x_{2}^{w_{2}-w^{\prime}_{1}}y_{1}^{z_{1}+z^{\prime}_{1}}y_{2}^{z_{2}+z^{\prime}_{2}}\end{split}$$

and the expansion of $(y_{1}^{s}+y_{2}^{s})^{w^{\prime}_{1}}$ contains monomials of the form $y_{1}^{sk}y_{2}^{s(w^{\prime}_{1}-k)}$, for $0\leq k\leq w^{\prime}_{1}$, so we conclude that $x_{1}^{w^{\prime}_{1}}x_{2}^{w_{2}}y_{1}^{z_{1}+z^{\prime}_{1}}y_{2}^{z_{2}+z^{\prime}_{2}}$ can be expressed as a linear combination of the monomials

$$x_{2}^{w_{2}-w^{\prime}_{1}}y_{1}^{z_{1}+z^{\prime}_{1}+sk}y_{2}^{z_{2}+z^{\prime}_{2}+s(w^{\prime}_{1}-k)}\quad\quad\textup{ for }\,0\leq k\leq w^{\prime}_{1}.$$

We claim that the above monomials are in $\mathbb{B}_{s}$. Indeed, by assumption we have $z_{1}+z_{2}=-sw_{2}$ and $z^{\prime}_{1}+z^{\prime}_{2}=0$, so $z_{1}+z^{\prime}_{1}+sk+z_{2}+z^{\prime}_{2}+s(w^{\prime}_{1}-k)=-sw_{2}+sw^{\prime}_{1}$ as required. Thus by the definition of $\mathfrak{v}_{s}$ we compute $\mathfrak{v}_{s}(\mathbb{b}\mathbb{b}^{\prime})$ by taking the minimum

(6.22) $$\begin{split}\mathfrak{v}_{s}(\mathbb{b}\mathbb{b}^{\prime})&=\min\left\{\mathfrak{v}_{s}(x_{2}^{w_{2}-w^{\prime}_{1}}y_{1}^{z_{1}+z^{\prime}_{1}+sk}y_{2}^{z_{2}+z^{\prime}_{2}+s(w^{\prime}_{1}-k)})\,\mid\,0\leq k\leq w^{\prime}_{1}\right\}\\ &=\min\left\{\Phi(z_{2}+z^{\prime}_{2}+s(w^{\prime}_{1}-k),-z_{2}-z^{\prime}_{2}-s(w^{\prime}_{1}-k),w_{2}-w^{\prime}_{1},0)\,\mid\,0\leq k\leq w^{\prime}_{1}\right\}\\ &=\min\left\{f_{(z_{2}+z^{\prime}_{2}+s(w^{\prime}_{1}-k),-z_{2}-z^{\prime}_{2}-s(w^{\prime}_{1}-k),w_{2}-w^{\prime}_{1},0)}\,\mid\,0\leq k\leq w^{\prime}_{1}\right\}\\ &=\min\left\{f_{(z_{2},-z_{2},w_{2},0)}+f_{(z^{\prime}_{2},-z^{\prime}_{2}-sw^{\prime}_{1},-w^{\prime}_{1},0)}+g_{(s(w^{\prime}_{1}-k),sk,0,0)}\,\mid\,0\leq k\leq w^{\prime}_{1}\right\}\\ &=f_{(z_{2},-z_{2},w_{2},0)}+f_{(z^{\prime}_{2},-z^{\prime}_{2}-sw^{\prime}_{1},-w^{\prime}_{1},0)}+\min\{g_{(s(w^{\prime}_{1}-k),sk,0,0)}\,\mid\,0\leq k\leq w^{\prime}_{1}\}\end{split}$$

where $g_{(c,d,0,0)}$ denotes the function on $\mathbb{M}_{s}$ defined by $g_{(c,d,0,0)}(w_{1},w_{2},z_{1},z_{2})=cw_{1}+dw_{2}$. For any element $(w_{1},w_{2},z_{1},z_{2})\in\mathbb{M}_{s}$, we have

$$\min\{g_{(s(w^{\prime}_{1}-k),sk,0,0)}(w_{1},w_{2},z_{1},z_{2})\,\mid\,0\leq k\leq w^{\prime}_{1}\}=\min\{s(w^{\prime}_{1}-k)w_{1}+skw_{2}\,\mid\,0\leq k\leq w^{\prime}_{1}\}=0$$

because $\min\{w_{1},w_{2}\}=0$ for an element in $\mathbb{M}_{s}$. Hence the function $\min\{g_{(s(w^{\prime}_{1}-k),sk,0,0)}\,\mid\,0\leq k\leq w^{\prime}_{1}\}$ is identically $0$ on $\mathbb{M}_{s}$, and we conclude

$$\mathfrak{v}_{s}(\mathbb{b}\mathbb{b}^{\prime})=f_{(z_{2},-z_{2},w_{2},0)}+f_{(z^{\prime}_{2},-z^{\prime}_{2}-sw^{\prime}_{1},-w^{\prime}_{1},0)}=\mathfrak{v}_{s}(\mathbb{b})+\mathfrak{v}_{s}(\mathbb{b}^{\prime})$$

as desired. The case $w_{2}<w^{\prime}_{1}$ follows similarly. This proves that $\mathfrak{v}_{s}(\mathbb{b}\mathbb{b}^{\prime})=\mathfrak{v}_{s}(\mathbb{b})+\mathfrak{v}_{s}(\mathbb{b}^{\prime})$ for any two elements $\mathbb{b},\mathbb{b}^{\prime}\in\mathbb{B}_{s}$.

We must next prove that $\mathfrak{v}_{s}(fg)=\mathfrak{v}_{s}(f)+\mathfrak{v}_{s}(g)$ for arbitrary $f,g\in\mathcal{A}_{s}$. However, the argument is the same as that given in [3, Lemma 8.10] so we do not reproduce it here. The remaining properties of valuations in Definition 2.10 are straightforward to verify and are left to the reader.

We now prove that the pair $(\mathcal{A}_{s},\mathfrak{v}_{s}:\mathcal{A}_{s}\to P_{\mathcal{M}_{s}})$ is a detropicalization of $\mathcal{M}_{s}$ in the sense of Definition 2.11. We have already shown that $\mathfrak{v}_{s}:\mathcal{A}_{s}\to P_{\mathcal{M}_{s}}$ is a valuation (with values in the idempotent semialgebra $P_{\mathcal{M}_{s}}$), so it remains only to show that every element of $\textup{Sp}(\mathcal{M}_{s})$ is in the image of $\mathfrak{v}_{s}$, i.e. that $\mathfrak{v}_{s}$ is surjective onto $\textup{Sp}(\mathcal{M}_{s})$, and that the Krull dimension of $\mathcal{A}_{s}$ is equal to the rank of $\mathcal{M}_{s}$. The first claim follows immediately from the fact that $\mathfrak{v}_{s}$ restricted to $\mathbb{B}_{s}$ is a bijection from $\mathbb{B}_{s}$ to $\textup{Sp}(\mathcal{M}_{s})$, as was seen above. The second claim follows from the fact that $Spec$ of $\mathcal{A}_{s}$ is an affine variety of dimension $2$, so the Krull dimension of $\mathcal{A}_{s}$ is $2$, which is the rank of $\mathcal{M}_{s}$, as required. ∎

7. Example: a Cox ring of a compactification $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$

In [3, Section 7], the authors establish a general framework for constructing a compactification of $Spec(\mathcal{A}_{\mathcal{M}})$ (where $\mathcal{A}_{\mathcal{M}}$ is a detropicalization of a polyptych lattice $\mathcal{M}$) with respect to a choice of PL polytope $\mathcal{P}\subset\mathcal{M}_{{\mathbb{R}}}$. Moreover, in the case when $\mathcal{A}_{\mathcal{M}}$ is a UFD, it is shown that the Cox ring of the compactification is finitely generated. The main purpose of this section is to illustrate the general theory outlined in [3] by working out, in detail, the Cox ring of the compactification of $Spec(\mathcal{A}_{s})$ with respect to a PL polytope $\mathcal{P}$. More specifically, we showed in [3, Theorem 7.19] that both the class group and the Cox ring of the compactification is finitely generated. Here, for the rank-$2$ example $\mathcal{M}_{s}$ for $s=1$ and for a specific PL polytope $\mathcal{P}$, we take a step further: we give a concrete presentation of the Cox ring in terms of generators and relations.

Let $s=1$. We note first that it is straightforward to check, from the explicit generators-and-relation presentation of $\mathcal{A}_{s}$ for the case $s=1$, that $\mathcal{A}_{s}$ is a UFD. Therefore, [3, Theorem 7.19] applies. Next, we specify the PL polytope $\mathcal{P}$ in question. As in Section 5, we specify points under the identification $\textup{Sp}(\mathcal{M}_{s})\cong\mathcal{T}_{s}$. With this understanding, we consider the $3$ points

$$\mathsf{p}=(1,-1,1),\quad\mathsf{q}=(-2,2,1),\quad\mathsf{r}=(1,-3,-2)$$

and define

(7.1) $$\mathcal{P}:=\mathcal{H}_{\mathsf{p},-1}\cap\mathcal{H}_{\mathsf{q},-1}\cap\mathcal{H}_{\mathsf{r},-1}.$$

It is not hard to check that this is compact, and hence a PL polytope. It is not an integral PL polytope, hence not chart-Gorenstein-Fano; however, the computation of its Cox ring is still useful to illustrate the general theory.

We now briefly recall the definition of the compactification. For details we refer to [3, Section 7]. For $k$ a positive integer, we define the polytope $k\mathcal{P}$ by scaling the parameters in the defining inequalities, so in our case

$$k\mathcal{P}:=\mathcal{H}_{\mathsf{p},-k}\cap\mathcal{H}_{\mathsf{q},-k}\cap\mathcal{H}_{\mathsf{r},-k}.$$

We also define

(7.2) $$\Gamma(\mathcal{A}_{s},k\mathcal{P}):=\{f\in\mathcal{A}_{s}\mid\mathfrak{v}(f)\geq\psi_{k\mathcal{P}}\}$$

where $\psi_{k\mathcal{P}}:\mathcal{N}=\mathcal{M}_{s}\to F$ denotes the support function of the PL polytope $k\mathcal{P}$ and the inequality is with respect to the partial order on $P_{\mathcal{N}}=P_{\mathcal{M}_{s}}$ (i.e. pointwise inequality of functions). Recall also that the support of $f\in\mathcal{A}_{s}$ is defined as follows. If $f=\sum\lambda_{i}b_{i}$ for $\lambda_{i}\in{\mathbb{K}}$ and $b_{i}\in\mathbb{B}_{s}$ is an element in $\mathcal{A}_{s}$ expressed uniquely as a linear combination of elements of $\mathbb{B}_{s}$, the support of $f$, denoted by $\mathrm{supp}(f)$, is the point-convex hull of $\{{\sf v}^{-1}(\mathfrak{v}(b_{i}))\mid\lambda_{i}\neq 0\}$ in $\mathcal{M}_{s}\otimes{\mathbb{Q}}$.

It is shown in [3, Lemma 7.4] that the space $\Gamma(\mathcal{A}_{s},k\mathcal{P})$ can be equivalently described as

(7.3) $$\Gamma(\mathcal{A}_{s},k\mathcal{P})=\{f\in\mathcal{A}_{s}\,\mid\,\mathrm{supp}(f)\subseteq k\mathcal{P}\}.$$

We will use this characterization. Then the PL polytope algebra $\mathcal{A}_{s}^{\mathcal{P}}$ is defined as

(7.4) $$\mathcal{A}_{s}^{\mathcal{P}}:=\bigoplus_{k\geq 0}\Gamma(\mathcal{A}_{s},k\mathcal{P})\cdot t^{k}=\bigoplus_{k\geq 0}\{f\in\mathcal{A}_{s}\,\mid\,\mathrm{supp}(f)\subseteq k\mathcal{P}\}\cdot t^{k}$$

where the last equality is by (7.3). The algebra $\mathcal{A}_{s}^{\mathcal{P}}$ is evidently ${\mathbb{Z}}_{\geq 0}$-graded by the degree of $t$, and we define the compactification of $\textup{Spec}(\mathcal{A}_{s})$ with respect to $\mathcal{P}$ as

$$X_{\mathcal{A}_{s}}(\mathcal{P}):=\textup{Proj}(\mathcal{A}_{\mathcal{M}}^{\mathcal{P}}).$$

In preparation for the computation of the Cox ring of $X_{\mathcal{A}_{s}}(\mathcal{P})$, it will be useful to prove some general results. The utility of these results in relation to the Cox ring computation will become apparent below when we explain the general method of computation, which is derived from [1, Construction 1.4.2.1]. We emphasize that Lemma 7.5 and Proposition 7.6 apply to any finite polyptych lattice $\mathcal{M}$ over ${\mathbb{Z}}$, not just the rank-$2$ $\mathcal{M}_{s}$ case.

7.5 Lemma.

Let $\mathcal{M}$ be a finite polyptych lattice of rank $r$ over ${\mathbb{Z}}$ with a fixed choice of strict dual (${\mathbb{Z}}$-)pair $(\mathcal{M},\mathcal{N},{\sf v},{\sf w})$. Let $\mathbb{K}$ be an algebraically closed field and $(\mathcal{A}_{\mathcal{M}},\mathfrak{v})$ a detropicalization of $\mathcal{M}$ with convex adapted basis $\mathbb{B}=\{\mathbb{b}_{m}\}_{m\in\mathcal{M}}$ (with $\mathfrak{v}(\mathbb{b}_{m})=m\in\mathcal{M}\subset S_{\mathcal{M}}\cong P_{\mathcal{N}}$). Suppose $f,g\in\mathcal{A}_{\mathcal{M}}$ and $f\cdot g=\mathbb{b}_{m_{0}}$ for some $m_{0}\in\mathcal{M}$. Let $\beta\in\pi(\mathcal{N})$ such that $m_{0}\in C_{\beta}:={\sf v}^{-1}(\textup{Sp}(\mathcal{N},\beta))$. Then $\mathfrak{v}(f)=m,\mathfrak{v}(g)=m^{\prime}$ for some $m,m^{\prime}\in C_{\beta}\cap\mathcal{M}$, and $f=c\,\mathbb{b}_{m},g=c^{\prime}\,\mathbb{b}_{m^{\prime}}$ for some $c,c^{\prime}\neq 0,c,c^{\prime}\in\mathbb{K}$.

Proof.

If $f\cdot g=\mathbb{b}_{m_{0}}$ then $\mathfrak{v}(f)+\mathfrak{v}(g)=\mathfrak{v}(\mathbb{b}_{m_{0}})=m_{0}$. Since $m_{0}$ is assumed to be in $C_{\beta}={\sf v}^{-1}(\textup{Sp}(\mathcal{N},\beta))$, this implies that, interpreted as a function on $\mathcal{N}$ via ${\sf v}$, $m_{0}$ induces a linear function on the coordinate chart $N_{\beta}$. By definition of $\mathfrak{v}$, the images $\mathfrak{v}(f),\mathfrak{v}(g)$ are convex piecewise-linear functions on $N_{\beta}$, and their sum is linear on $N_{\beta}$. This can occur only if both $\mathfrak{v}(f)$ and $\mathfrak{v}(g)$ are linear on $N_{\beta}$. Thus $\mathfrak{v}(f),\mathfrak{v}(g)$ are contained in ${\sf v}^{-1}(\textup{Sp}(\mathcal{N},\beta))=C_{\beta}$, and by definition of detropicalizations are also contained in $\mathcal{M}$. Hence $\mathfrak{v}(f),\mathfrak{v}(g)\in C_{\beta}\cap\mathcal{M}$. In particular there exist $m,m^{\prime}\in C_{\beta}\cap\mathcal{M}$ such that $\mathfrak{v}(f)=m,\mathfrak{v}(g)=m^{\prime}$. Since $\mathbb{B}$ is a convex adapted basis for $\mathfrak{v}$, it now follows that $f=c\mathbb{b}_{m}+\sum_{i}c_{i}\mathbb{b}_{m_{i}}$ and $g=c^{\prime}\mathbb{b}_{m^{\prime}}+\sum_{j}c_{j}\mathbb{b}_{m_{j}}$ for $c,c^{\prime}\in\mathbb{K}^{*}$. Moreover, since $\mathfrak{v}(f)=\min(\{m\}\cup\{m_{i}\})=m$, we must have that $\mathfrak{v}(\mathbb{b}_{m_{i}})=m_{i}\geq m$ as functions on $\mathcal{N}$ for all $i$, and similarly, $m_{j}\geq m^{\prime}$ as functions on $\mathcal{N}$ for all $j$. But $m_{i},m_{j}$ are convex functions obtained as a minimum of a finite set of linear functions, so $m_{i}\geq m,m_{j}\geq m^{\prime}$ are only possible if $m_{i}=m,m_{j}=m^{\prime}$. In other words, $f=c\mathbb{b}_{m}$ and $g=c^{\prime}\mathbb{b}_{m^{\prime}}$. This concludes the proof. ∎

We can now compute the group of units in $\mathcal{A}_{\mathcal{M}}$.

7.6 Proposition.

Let the notation and assumptions be as in Lemma 7.5. Let $u\in\mathcal{A}_{\mathcal{M}}$. Then $u$ is a unit in $\mathcal{A}_{\mathcal{M}}$ if and only if $u=c\,\mathbb{b}_{m}$ for $c\in\mathbb{K}^{*}$ and $m\in\bigcap_{C_{\beta}\in\Sigma(\mathcal{M})}C_{\beta}$ where the intersection is over all maximal-dimensional cones in $\Sigma(\mathcal{M})$.

Proof.

Suppose first that $u$ is a unit. Then there exists $v\in\mathcal{A}_{\mathcal{M}}$ with $u\cdot v=1$, and $1=\mathbb{b}_{0}$ for $0\in\mathcal{M}$. Note also that $0\in C_{\beta}$ for all maximal-dimensional cones $C_{\beta}$ in $\Sigma(\mathcal{M})$, since $C_{\beta}$ is a cone. By Lemma 7.5, it follows that $\mathfrak{v}(v)=\mathbb{b}_{m}$ for $m\in\left(\bigcap_{C_{\beta}\in\Sigma(\mathcal{M})}C_{\beta}\right)\cap\mathcal{M}$ and that $u=c\,\mathbb{b}_{m}$ for $c\in\mathbb{K}^{*}$.

Now for the opposite implication, suppose that $m\in\left(\bigcap_{C_{\beta}\in\Sigma(\mathcal{M})}C_{\beta}\right)\cap\mathcal{M}$. The PL fan $\Sigma(\mathcal{M})$ is a complete fan, so the intersection of all its cones must be a linear subspace (in particular, addition is well-defined), and thus $\left(\bigcap_{C_{\beta}\in\Sigma(\mathcal{M})}C_{\beta}\right)\cap\mathcal{M}$ is a lattice. Hence for any $m\in\left(\bigcap_{C_{\beta}\in\Sigma(\mathcal{M})}C_{\beta}\right)\cap\mathcal{M}$ there exists $m^{\prime}\in\left(\bigcap_{C_{\beta}\in\Sigma(\mathcal{M})}C_{\beta}\right)\cap\mathcal{M}$ with $m+m^{\prime}=0$. Now consider $\mathbb{b}_{m}\cdot\mathbb{b}_{m^{\prime}}$ which has $\mathfrak{v}(\mathbb{b}_{m}\cdot\mathbb{b}_{m^{\prime}})=0=\mathfrak{v}(1)$. Then $\mathbb{b}_{m}\cdot\mathbb{b}_{m^{\prime}}=c^{\prime}\,1+\sum_{i}c_{i}\mathbb{b}_{m_{i}}$ for $c\in\mathbb{K}^{*}$ and $m_{i}\geq 0$ as functions on $\mathcal{N}$ for all $i$. The same argument as in the proof of Lemma 7.5 shows $m_{i}\equiv 0$ on $\mathcal{N}$, i.e. $m_{i}=0$, and so $\mathbb{b}_{m}\cdot\mathbb{b}_{m^{\prime}}=c^{\prime}\,1$. An inverse of $c\cdot\mathbb{b}_{m}$ is therefore given by $\frac{1}{c\,c^{\prime}}\mathbb{b}_{m^{\prime}}$ and the claim is proved. ∎

Turning back to our concrete rank-$2$ example $\mathcal{M}_{s}$ for $s=1$, an application of Proposition 7.6 immediately yields the following.

7.7 Corollary.

Let $s=1$ and consider $\mathcal{M}_{s},\mathcal{A}_{s}$ as above. The group of units of $\mathcal{A}_{s}$ is generated by $y_{1}y_{2}^{-1}$.

Proof.

By Proposition 7.6 we must find $m$ in the intersection of the maximal cones of $\Sigma(\mathcal{M}_{s})$. The (lattice points inside the) intersection of the two maximal cones in $\Sigma(\mathcal{M}_{s})$, viewed as a subset of $\mathbb{M}_{s}$, is the set $\{w_{1}=w_{2}=0\}\subset\mathbb{M}_{s}$. In this subset we must have $z_{1}+z_{2}=0$, so $z_{2}=-z_{1}$ and we see that the monomials – i.e. the convex adapted basis elements – corresponding to these lattice points are exactly $y_{1}^{k}y_{2}^{-k}$ for $k\in{\mathbb{Z}}$. These are generated as a group by the single generator $y_{1}y_{2}^{-1}$. ∎

We now explain our general method for computing the Cox ring of $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$, which is also applicable in general, not just our specific rank $2$ examples. It is based on [1, Construction 1.2.4.1] as well as the proof of [3, Theorem 7.16]. First, in [1] it is explained that the Cox ring of $X$ (for $X$ an irreducible, normal prevariety with $\Gamma(X,\mathcal{O}^{*})=\mathbb{K}^{*}$ and finitely generated class group) can be described as

(7.8) $$\Gamma(X,\mathcal{S})/\Gamma(X,\mathcal{I})$$

where $\mathcal{S}$ is a certain sheaf of divisorial algebras (and $\Gamma(X,\mathcal{S})$ the ring of its global sections)) and $\mathcal{I}$ is a sheaf of ideals of $\mathcal{S}$. In this exposition, we do not give a detailed description of either $\mathcal{S}$ and $\mathcal{I}$ because we are able to give another, more concrete, description of both of these rings. Indeed, in the course of the proof of [3, Theorem 7.16] we show the following.

7.9 Lemma.

Let $\mathcal{M}$ be a finite polyptych lattice of rank $r$ over ${\mathbb{Z}}$ with a fixed choice of strict dual ${\mathbb{Z}}$-pair $(\mathcal{M},\mathcal{N},{\sf v},{\sf w})$. Let ${\mathbb{K}}$ be an algebraically closed field and $(\mathcal{A}_{\mathcal{M}},\mathfrak{v})$ a detropicalization of $\mathcal{M}$ with convex adapted basis $\mathbb{B}$. Let $\mathcal{P}=\cap_{i=1}^{\ell}\mathcal{H}_{{\sf w}(n_{i}),a_{i}}\subset\mathcal{M}_{\mathbb{R}}$ be a full-dimensional PL polytope. Suppose $n_{i}\in\mathcal{N},n_{i}\neq 0$, the $n_{i}$ are pairwise distinct, and $a_{i}\in{\mathbb{Z}}_{<0}$ for all $i\in[\ell]$. Suppose also that for each $n_{i}$ there exists a coordinate chart $\alpha_{i}\in\pi(\mathcal{M})$ on which $n_{i}$ is linear, and, the intersection of the boundary of $\pi_{\alpha_{i}}(\mathcal{H}_{{\sf w}(n_{i}),a_{i}})$ with $\pi_{\alpha_{i}}(\mathcal{P})$ is a facet of $\pi_{\alpha_{i}}(\mathcal{P})$. Let $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P}):=\mathrm{Proj}(\mathcal{A}^{\mathcal{P}}_{\mathcal{M}})$ be the compactification of $Spec(\mathcal{A}_{\mathcal{M}})$ constructed in [3, Section 7]. Then, for $X=X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$, the ring $\Gamma(X,\mathcal{S})$ can be described as

(7.10) $$\bigoplus_{\overline{r}\in{\mathbb{Z}}^{\ell}}\mathcal{A}_{\mathcal{M}}(\overline{r})t_{1}^{r_{1}}\cdots t_{\ell}^{r_{\ell}}\subset\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]$$

where for $\overline{r}=(r_{1},\cdots,r_{\ell})\in{\mathbb{Z}}^{\ell}$ we define

(7.11) $$\mathcal{A}_{\mathcal{M}}(\overline{r}):=\mathrm{span}_{\mathbb{K}}\{\mathbb{b}_{m}\,\mid\,\langle n_{i},m\rangle+r_{i}\geq 0\,\,\textup{ for all }i\in[\ell]\}$$

and $\langle n_{i},m\rangle$ denotes the dual pairing between $\mathcal{N}$ and $\mathcal{M}$.

Continuing this line of reasoning, we can also describe $\Gamma(X,\mathcal{I})$ more concretely in terms of $\mathcal{A}_{\mathcal{M}}$ as follows. For $f\in\mathcal{A}_{\mathcal{M}}$, let $\overline{d}_{f}\in{\mathbb{Z}}^{\ell}$ denote the integer vector $\overline{d}_{f}:=(\mathrm{ord}_{D_{1}}(f),\cdots,\mathrm{ord}_{D_{\ell}}(f))$ given by the orders of vanishing of $f$ along the divisors $D_{i}$ corresponding to the PL half-spaces $\mathcal{H}_{{\sf w}(n_{i}),a_{i}}$ (as described in [3, Section 7]). Then we have the corresponding monomial $t^{\overline{d}_{f}}:=t_{1}^{\mathrm{ord}_{D_{1}}(f)}t_{2}^{\mathrm{ord}_{D_{2}}}\cdots t_{\ell}^{\mathrm{ord}_{D_{\ell}}(f)}$.

7.12 Lemma.

Let the assumptions and notation be as in Lemma 7.9. Under the identification of $\Gamma(X,\mathcal{S})$ with (7.10) given in Lemma 7.9, the ideal $\Gamma(X,\mathcal{I})$ in $\Gamma(X,\mathcal{S})$ is contained in the ideal in (7.10) generated by $u-t^{\overline{d}_{u}}$, as $u$ ranges over the group of units in $\mathcal{A}_{\mathcal{M}}$.

Proof.

We interpret the objects cited in [1, Construction 1.2.4.1] in the same way as in [3, Proof of Theorem 7.14] so we will only sketch the argument. First, it is explained in [1] that $\Gamma(X,\mathcal{I})$ is generated by sections of the form $1-\chi(E)$ where $1$ is homogeneous of degree $0$, $E$ ranges over elements of the kernel $K^{0}$ of the surjection $\oplus_{i}{\mathbb{Z}}\cdot D_{i}\to\mathrm{Cl}(X)$ onto the class group of $X$, $\chi:K^{0}\to\mathbb{K}(X)^{*}$ is a character, and $\chi(E)$ is homogeneous of degree $-E$. Since $K^{0}$ consists of divisors $E=\sum_{i}a_{i}D_{i}$ such that there exists a rational function $f\in\mathbb{K}(X)^{*}$ with $\mathrm{div}(f)=E$, and since the $D_{i}$ form the boundary of the compactification $X_{\mathcal{A}_{\mathcal{M}}}(\mathcal{P})$ of $Spec(\mathcal{A}_{\mathcal{M}})$, it follows that $K^{0}$ consists of $\sum_{i}a_{i}D_{i}$ for which there exists a unit $u\in\mathcal{A}_{\mathcal{M}}$ with $\mathrm{div}(u)=\sum_{i}a_{i}D_{i}$. In particular, in the notation of [1], $\chi(\sum_{i}a_{i}D_{i})=u$. Moreover, our conventions on the homogeneous degree (encoded by the $t_{i}$ variables) imply that the relation $1-\chi(E)$ is equivalent to $u-t^{\overline{d}_{u}}=0$. This proves the claim. ∎

With these results in place we can now explain our method of computation. Let $\mathcal{J}$ denote the ideal in $\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]$ generated by the elements $u-t^{\overline{d}_{u}}=0$ as $u$ ranges over the group of units of $\mathcal{A}_{\mathcal{M}}$. Then it follows from [1, Construction 1.2.4.1], Lemma 7.9, and Lemma 7.12 that there is an injective ring homomorphism

$$\Gamma(X,\mathcal{S})/\Gamma(X,\mathcal{I})\hookrightarrow\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]/\mathcal{J}.$$

The map is induced by the natural inclusion of (7.10) into $\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]$. Thus, in order to give an explicit presentation of $\mathcal{R}(X)$, it suffices to determine a finite list of generators of (7.10), which we denote as $\{X_{1},\cdots,X_{n}\}$, and then define a surjective homomorphism

$$\varphi:{\mathbb{C}}[u_{1},u_{2},\cdots,u_{n}]\rightarrow\bigoplus_{\overline{r}\in{\mathbb{Z}}^{\ell}}\mathcal{A}_{\mathcal{M}}(\overline{r})t_{1}^{r_{1}}\cdots t_{\ell}^{r_{\ell}},\quad u_{i}\mapsto X_{i}.$$

Composing $\varphi$ with the inclusion

$$\bigoplus_{\overline{r}\in{\mathbb{Z}}^{\ell}}\mathcal{A}_{\mathcal{M}}(\overline{r})t_{1}^{r_{1}}\cdots t_{\ell}^{r_{\ell}}\hookrightarrow\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]$$

and the quotient map

$$\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]\to\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},\cdots,t_{\ell}^{\pm}]/\mathcal{J}$$

then gives a surjective map from ${\mathbb{C}}[u_{1},\cdots,u_{n}]$ to a ring isomorphic to $\Gamma(X,\mathcal{S})/\Gamma(X,\mathcal{I})$. Computing the kernel $\kappa$ of this map then gives the desired presentation, namely

$$\Gamma(X,\mathcal{S})/\Gamma(X,\mathcal{I})\cong{\mathbb{C}}[u_{1},\cdots,u_{n}]/\kappa.$$

We now implement the above strategy in our case of $\mathcal{M}_{s}$ for $s=1$, the detropicalization $\mathcal{A}_{s}$ of Section 6, and the PL polytope $\mathcal{P}$ of (7.1). In Corollary 7.7 we already computed the generator of the group of units to be $y_{1}y_{2}^{-1}$. By mapping the corresponding element of $\mathbb{M}_{s}$ to $M_{1}$ and then evaluating the point on the image, it is straightforward from the explicit description of the three points $\mathsf{p}=(1,-1,1),\mathsf{q}=(-2,2,1),\mathsf{r}=(1,-3,2)$ (thought of as elements of $\mathcal{T}_{s}$) that the order of vanishing of $y_{1}y_{2}^{-1}$ along their corresponding divisors are given by $-1,-1,2$ respectively. Thus in our example we have

$$\mathcal{J}=\langle y_{1}y_{2}^{-1}-t_{1}^{-1}t_{2}^{-1}t_{3}^{2}\rangle.$$

Following the general method outlined above, our next step is to find a set of generators for (7.10). To accomplish this, we take the following approach. The description (7.11) of the $\overline{r}$-graded piece of (7.10) makes it clear that (7.10) is spanned by $\mathbb{b}_{m}t^{\overline{r}}$ where $m\in\mathcal{M}$ and $\overline{r}\in{\mathbb{Z}}^{\ell}$ satisfy certain inequalities. We have seen in Section 3 that the PL fan $\Sigma(\mathcal{M}_{s})$ of $\mathcal{M}_{s}$ has two maximal cones, each of which are half-spaces. Using the identification $\mathcal{M}_{s}\cong\mathbb{M}_{s}$, (the lattices within) these half-spaces may be identified with

$$\mathbb{M}_{s}(1):=\{(0,w_{2},z_{1},z_{2})\,\mid\,w_{2}\geq 0,-w_{2}=z_{1}+z_{2}\}$$

(here we have used $s=1$) and

$$\mathbb{M}_{s}(2):=\{(w_{1},0,z_{1},z_{2})\,\mid\,w_{1}\geq 0,0=z_{1}+z_{2}\}.$$

Note that in both cases, $z_{2}$ is completely determined by the other variables, so $\mathbb{M}_{s}(i)\cong{\mathbb{Z}}_{\geq 0}\times{\mathbb{Z}}$ for both $i=1,2$.

Now we consider the cases $i=1,2$ separately. First suppose $i=1$. Then for each $m=(0,w_{2},z_{1},z_{2})\in\mathbb{M}_{s}(1)$ (here we are implicitly using the identification $\mathcal{M}_{s}\cong\mathbb{M}_{s}$) the corresponding convex adapted basis element $\mathbb{b}_{m}$ is $x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{z_{2}}=x_{2}^{w_{2}}y_{1}^{z_{1}}y_{2}^{-w_{2}-z_{1}}$. Here we have used the conditions for a vector to be in $\mathbb{M}_{s}(1)$ and have also used that $s=1$. Note that $\mathbb{M}_{s}(1)$ is closed under addition. Now it follows from (7.11) that the monomials $\mathbb{b}_{m}t^{\overline{r}}$, with $m=(0,w_{2},z_{1},z_{2})\in\mathbb{M}_{s}(1)$, lying in (7.10) are precisely those satisfying

$$\mathsf{p}(0,w_{2},z_{1},z_{2})+r_{1}\geq 0,\quad\mathsf{q}(0,w_{2},z_{1},z_{2})+r_{2}\geq 0,\quad\mathsf{r}(0,w_{2},z_{1},z_{2})+r_{3}\geq 0.$$

In general we have

$$\mathsf{p}(w_{1},w_{2},z_{1},z_{2})=z_{2}+w_{1}-w_{2},\quad\mathsf{q}(w_{1},w_{2},z_{1},z_{2})=z_{2}-2w_{1}+2w_{2},$$

and

$$\mathsf{r}(w_{1},w_{2},z_{1},z_{2})=\begin{cases}-2z_{2}+w_{1}-w_{2}\,\textup{ if }\,w_{1}-w_{2}\geq 0\\ -2z_{2}+3(w_{1}-w_{2})\,\textup{ if }\,w_{1}-w_{2}\leq 0,\end{cases}.$$

For $i=1$ we have the relation $z_{2}=-w_{2}-z_{1}$ so we conclude that the basis elements of (7.10) corresponding to $m\in\mathbb{M}_{s}(1)$ are in bijective correspondence with

$$T_{1}=\{(w_{2},z_{1},r_{1},r_{2},r_{3})\in{\mathbb{Z}}^{5}\,\mid\,w_{2}\geq 0,-2w_{2}-z_{1}+r_{1}\geq 0,w_{2}-z_{1}+r_{2}\geq 0,-w_{2}+2z_{1}+r_{3}\geq 0\}\subset{\mathbb{Z}}^{5}.$$

A similar computation shows that for $i=2$ there is a bijection between

$$T_{2}=\{(w_{1},z_{1},r_{1},r_{2},r_{3})\in{\mathbb{Z}}^{5}\,\mid\,w_{1}\geq 0,w_{1}-z_{1}+r_{1}\geq 0,-2w_{1}-z_{1}+r_{2}\geq 0,w_{1}+2z_{1}+r_{3}\geq 0\}$$

and another subset of basis elements of (7.10), corresponding to $\mathbb{M}_{s}(2)$. Together, the union of these basis elements for $i=1$ and $i=2$ span all of (7.10). Notice that since $\mathbb{M}_{s}(i)$ is closed under addition for $i=1,2$, the monomials (in the variables $w_{1},w_{2},z_{1},z_{2},t_{1},t_{2},t_{3}$) are closed under multiplication. Thus, if we find generators for the affine semigroups $T_{1}$ and $T_{2}$, then the union of the corresponding monomials will generate (7.10). A Macaulay2 computation reveals that $T_{1}$ is generated by

$$\left\{\begin{pmatrix}1\\ 0\\ 2\\ -1\\ 1\end{pmatrix},\begin{pmatrix}0\\ 0\\ 1\\ 0\\ 0\end{pmatrix},\begin{pmatrix}0\\ 0\\ 0\\ 1\\ 0\end{pmatrix},\begin{pmatrix}0\\ 0\\ 0\\ 0\\ 1\end{pmatrix},\begin{pmatrix}0\\ -1\\ -1\\ -1\\ 2\end{pmatrix},\begin{pmatrix}0\\ 1\\ 1\\ 1\\ -2\end{pmatrix}\right\}$$

where the vector entries correspond to the variables $w_{2},z_{1},r_{1},r_{2},r_{3}$ respectively, while $T_{2}$ is generated by

$$\left\{\begin{pmatrix}1\\ 0\\ -1\\ 2\\ -1\end{pmatrix},\begin{pmatrix}0\\ 0\\ 1\\ 0\\ 0\end{pmatrix},\begin{pmatrix}0\\ 0\\ 0\\ 1\\ 0\end{pmatrix},\begin{pmatrix}0\\ 0\\ 0\\ 0\\ 1\end{pmatrix},\begin{pmatrix}0\\ -1\\ -1\\ -1\\ 2\end{pmatrix},\begin{pmatrix}0\\ 1\\ 1\\ 1\\ -2\end{pmatrix}\right\}$$

where the vector entries correspond to $w_{1},z_{1},r_{1},r_{2},r_{3}$ respectivey. It follows that the following seven monomials generate (7.10):

$$x_{2}y_{1}^{-1}t_{1}^{2}t_{2}^{-1}t_{3},x_{1}t_{1}^{-1}t_{2}^{2}t_{3}^{-1},t_{1},t_{2},t_{3},y_{1}y_{2}^{-1}t_{1}t_{2}t_{3}^{-2},y_{1}^{-1}y_{2}t_{1}^{-1}t_{2}^{-1}t_{3}^{2}.$$

Temporarily labelling the above monomials as $X_{1},\cdots,X_{7}$ from left to right, this implies that the ring homomorphism defined as

$$\tilde{G}:{\mathbb{C}}[W_{1},\cdots,W_{7}]\rightarrow\mathcal{A}_{\mathcal{M}}[t_{1}^{\pm},t_{2}^{\pm},t_{3}^{\pm}],\quad\quad W_{i}\mapsto X_{i}\,\,\textup{ for all }i,1\leq i\leq 7$$

is surjective onto (7.10). Thus, by computing the kernel of $\tilde{G}$ yields a presentation of the Cox ring. We see that $\mathrm{ker}(\tilde{G})=\langle W_{6}-1`,W_{5}-1,W_{2}W_{3}-W_{1}W_{7}+W_{4}\rangle$, and thus we obtain the following.

7.13 Proposition.

The Cox ring of $X_{\mathcal{A}_{s}}(\mathcal{P})$ is isomorphic to

$${\mathbb{C}}[W_{1},W_{2},W_{3},W_{4},W_{5}]/\langle W_{2}W_{3}-W_{1}W_{5}+W_{4}\rangle.$$