Hostname: page-component-54dcc4c588-rz4zl Total loading time: 0 Render date: 2025-09-26T02:46:24.643Z Has data issue: false hasContentIssue false

Foldable fans, cscK surfaces and local K-moduli

Published online by Cambridge University Press:  25 September 2025

Carl Tipler*
Affiliation:
UMR CNRS 6205, Laboratoire de Mathématiques de Bretagne Atlantique, Brest University, Brest, France carl.tipler@univ-brest.fr
Rights & Permissions [Opens in a new window]

Abstract

We study the moduli space of constant scalar curvature Kähler (cscK) surfaces around toric surfaces. To this end, we introduce the class of foldable surfaces: smooth toric surfaces whose lattice automorphism group contains a non-trivial cyclic subgroup. We classify such surfaces and show that they all admit a cscK metric. We then study the moduli space of polarised cscK surfaces around a point given by a foldable surface, and show that it is locally modelled on a finite quotient of a toric affine variety with terminal singularities.

Information

Type
Research Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2025. Published by Cambridge University Press on behalf of the Foundation Compositio Mathematica, in partnership with the London Mathematical Society

1. Introduction

The construction of moduli spaces of varieties is a central problem in complex geometry. Pioneered by Riemann, the case of moduli spaces of curves is now fairly well understood. In higher dimensions, however, the situation becomes much more delicate. Indeed, iterated blow-ups (see e.g. [Reference KollárKol13, Example 4.4]) or the presence of non-discrete automorphisms induce non-separatedness in moduli considerations. It is then remarkable that, despite their very simple combinatorial description, toric surfaces are subject to these two issues, and typically correspond to pathological points in the moduli space of surfaces. In this paper, we will show that when one restricts to constant scalar curvature Kähler (cscK) surfaces, the moduli space enjoys a very nice structure near its toric points.

Our motivation to restrict to cscK surfaces comes from the Yau–Tian–Donaldson conjecture (YTD conjecture, see [Reference YauYau93, Reference TianTia97, Reference DonaldsonDon02]), which predicts that the existence of a cscK metric on a given polarised Kähler manifold should be equivalent to (a uniform version of) K-polystability. In the Fano case, the proof of this conjecture led to the recent construction of the moduli space of K-polystable Fano varieties by Odaka [Reference OdakaOda15] and Li, Wang and Xu [Reference Li, Wang and XuLWX18, Reference Li, Wang and XuLWX19]. See [Reference TianTia97, Reference BermanBer16, Reference Chen, Donaldson and SunCDS15, Reference TianTia15] for the YTD conjecture and also [Reference XuXu21, Part II] and the references therein for a historical survey on the related moduli space. Moreover, from Odaka’s work [Reference OdakaOda12, Reference OdakaOda13], in the canonically polarised case K-stability is equivalent to having semi-log canonical singularities, a class that was successfully introduced to construct the so-called KSBA moduli space (see [Reference KollárKol13] for a survey of the moduli of varieties of general type). While a general algebraic construction of a moduli space for K-polystable varieties seems out of reach at the moment, on the differential geometric side Dervan and Naumann recently built a separated coarse moduli space for compact polarised cscK manifolds [Reference Dervan and NaumannDN20], extending Fujiki and Schumacher’s construction that dealt with the case of discrete automorphism groups [Reference Fujiki and SchumacherFS90] (see also Inoue’s work on the moduli space of Fano manifolds with Kähler–Ricci solitons [Reference InoueIno19]).

However, it seems to the author that the Fano case and its log or pair analogues are the only sources of examples of those moduli spaces, when non–discrete automorphism groups are allowed (see also the discussion in [Reference Dervan and NaumannDN20, Example 4.16], which may lead to further examples via moduli of polystable bundles). Restricting to surfaces, moduli spaces of Kähler–Einstein del Pezzo surfaces were explicitly constructed in [Reference Odaka, Spotti and SunOSS16]. Our focus will then be on the moduli space of polarised cscK surfaces, and its geometry around points corresponding to the toric points. Given that they satisfy the YTD conjecture, by Donaldson’s work [Reference DonaldsonDon02], toric surfaces attract a lot of attention, and they appear as natural candidates on which to test the general machinery (compare also with [Reference Kaloghiros and PetracciKP21, Reference PetracciPet22]). As the full classification of cscK toric surfaces is yet to be carried out, we will consider a subclass whose fans carry further symmetries, which we introduce now.

Definition 1.1. Let $N$ be a rank two lattice and $\Sigma$ a complete smooth fan in $N_{\mathbb{R}}:=N\otimes _{\mathbb{Z}} {\mathbb{R}}$ . The fan $\Sigma$ is called foldable if its lattice automorphism group ${{\rm Aut}}(N,\Sigma )$ contains a non-trivial cyclic subgroup. The associated toric surfaces will be called foldable surfaces.

In order to understand this class of surfaces amongst the toric ones, we show that all crystallographic groups arise as lattice automorphism groups of two dimensional smooth fans (see Section 3.1).

Proposition 1.2. Let $N$ be a rank two lattice and $\Sigma$ a complete smooth fan in $N_{\mathbb{R}}$ . Then ${{\rm Aut}}(N,\Sigma )$ is isomorphic to one of the groups in the following set

\begin{align*} \lbrace C_1, C_2, C_3, C_4,C_6, D_1, D_2, D_3, D_4, D_6 \rbrace . \end{align*}

Moreover, any group in the above list is isomorphic to the lattice automorphism group of some complete two dimensional smooth fan.

In Proposition 1.2, we denote the cyclic group of order $p$ by $C_p$ and the dihedral group of order $2p$ by $D_p$ . We make a distinction between $C_2$ and $D_1$ by assuming that $C_2$ acts via $-{{\rm Id}}$ on $N$ while $D_1$ acts through a reflection (see Section 3.1). Then, from Proposition 1.2, a complete and smooth two dimensional fan $\Sigma$ in $N_{\mathbb{R}}$ is foldable if and only if ${{\rm Aut}}(N,\Sigma )$ is not isomorphic to $C_1$ or $D_1$ . The associated class of toric surfaces then provides a wide class of examples of cscK surfaces (see Proposition 3.14).

Proposition 1.3. Let $X$ be a foldable surface. Then $X$ admits a cscK metric in some Kähler class.

This result relies on the classification of foldable surfaces (Section 3.2) and an application of Arezzo, Pacard, Singer, and Székelyhidi’s blow-up theorem for extremal Kähler metrics [Reference Arezzo, Pacard and SingerAPS11, Reference SzékelyhidiSzé15]. Similar arguments were used in [Reference TiplerTip14] to show that any toric surface admits an iterated toric blow-up with a cscK metric.

Remark 1.4. To our knowledge, all known examples of toric cscK surfaces are foldable (see e.g. [Reference TiplerTip14] for a family of examples with unbounded Picard rank, and [Reference Wang and ZhouWZ11] for a complete classification of polarised cscK toric surfaces up to Picard rank $4$ ). It would be interesting to produce examples of non-foldable toric cscK surfaces.

Our main result shows that the moduli space of polarised cscK surfaces, which is known to be a complex space [Reference Fujiki and SchumacherFS90, Reference Dervan and NaumannDN20], is quite well behaved around a foldable surface, as it is locally modelled on a finite quotient of a toric terminal singularity (see Section 4.3 for a definition of this class of singularities, which originated in the MMP).

Theorem 1.5. Let $(X,[\omega ])$ be a polarised cscK foldable surface with fan $\Sigma$ in $N_{\mathbb{R}}$ , and let $G={{\rm Aut}}(N,\Sigma )$ . Denote by $[X]\in \mathscr{M}$ the corresponding point in the moduli space $\mathscr{M}$ of polarised cscK surfaces. Then there are:

  1. (i) a Gorenstein toric affine $G$ -variety $W$ with at worst terminal singularities, and

  2. (ii) open neighbourhoods $\mathscr{W}$ and $\mathscr{U}$ of, respectively, (the image of) the torus fixed point $x\in \mathscr{W} \subset W/G$ and $[X]\in {\mathscr{U}}\subset \mathscr{M}$ ,

such that $({\mathscr{U}},[X])$ and $(\mathscr{W},x)$ are isomorphic as pointed complex spaces.

This result sheds some light on the singularities that may appear on the moduli space of polarised cscK surfaces. Note that by [Reference Braun, Greb, Langlois and JoaquinBGLJ24, Theorem 4], the good moduli space of smooth K-polystable Fano manifolds of fixed dimension and volume has klt type singularities (see Section 4.3 for definitions). Also, a combination of Braun, Greb, Langlois and Moraga’s work on GIT quotients of klt type singularities [Reference Braun, Greb, Langlois and JoaquinBGLJ24, Theorem 1], together with the construction of Dervan and Naumann’s moduli space for cscK polarised manifolds [Reference Dervan and NaumannDN20, Section 3], directly implies that the moduli space of polarised cscK surfaces has klt type singularities around its toric points (see Proposition 2.3). Hence Theorem1.5 provides a refinement of Proposition 2.3, showing that, up to a finite covering, the toric structure is locally preserved on the moduli space, and that the singularities are terminal, at a foldable point. On the other hand, one might not expect such a nice structure at toric boundary points of a (still hypothetical) compactification $\overline {\mathscr{M}}$ . Indeed, the local structure of the K-moduli space of Fano varieties was studied around singular toric varieties in [Reference Kaloghiros and PetracciKP21, Reference PetracciPet22], where much wilder phenomena were observed, mainly due to the existence of obstructed deformations.

Remark 1.6. It would be interesting to understand whether or not the finite quotient $W/G$ in Theorem 1.5 is still terminal. Note, however, that it is a non-trivial problem to discover whether a finite quotient of a terminal singularity is terminal (see [Reference Kollár, Mori, Clemens and CortiKM98, Remark, p. 161]).

Remark 1.7. Because of cohomology vanishings for toric surfaces, the germ of the moduli space of polarised cscK surfaces at a toric point does not depend on the chosen cscK polarisation. See, however, [Reference Sektnan and TiplerST22] for local wall-crossing type phenomena when the polarisation is pushed away from the cscK locus of the Kähler cone.

Remark 1.8. Theorem 1.5 provides another instance in which canonical Kähler metrics, or K-polystability, become useful in moduli problems. We would like to mention two other approaches to producing moduli spaces for (or around) toric varieties. In [Reference MeerssemanMee19], the notion of an analytic stack is introduced to produce moduli spaces of integrable complex structures modulo diffeomorphisms. Those spaces carry more information, as they classify a wider class of varieties, but are typically non-separated. In another direction, turning from the classical setting to the non-commutative one, quantum toric varieties admit moduli spaces which are orbifolds; see [Reference Katzarkov, Lupercio, Meersseman and VerjovskyKLMV21].

We proceed now to an overview of the proof of Theorem1.5. If $(X,[\omega ])$ is a cscK foldable surface, relying on [Reference SzékelyhidiSzé10, Reference Dervan and NaumannDN20] and an observation in [Reference Rollin and TiplerRT14], the deformation theory of $(X,[\omega ])$ is entirely encoded by the set of unobstructed polystable points in $H^1(X,TX)$ under the action of ${{\rm Aut}}(X)$ , the latter being reductive by Matsushima and Lichnerowicz’s theorem [Reference MatsushimaMat57, Reference LichnerowiczLic58]. Nill’s study of toric varieties with reductive automorphism groups [Reference NillNil06] then shows that either $X$ is isomorphic to ${\mathbb{C}}\mathbb{P}^2$ or ${\mathbb{C}}\mathbb{P}^1\times {\mathbb{C}}\mathbb{P}^1$ , in which case it is rigid, or ${{\rm Aut}}^0(X)\simeq T$ , where $T$ is the torus of $X$ . Then, by Ilten’s work [Reference IltenIlt11], $H^2(X,TX)=0$ , so that, locally, the moduli space we are after is given by (a neighbourhood of the origin in) the GIT quotient $H^1(X,TX)//{{\rm Aut}}(X)$ , where ${{\rm Aut}}(X)\simeq T\rtimes G$ . Using [Reference IltenIlt11] again, the deformation theory of toric surfaces is explicitly described in terms of their fan, and the affine toric variety $W:=H^1(X,TX)//T$ in Theorem1.5 can be explicitly computed. The proof of Theorem1.5 then goes as follows. By a classification result (see Lemma 4.3), $X$ can be obtained by successive $C_p$ -equivariant blow-ups of some ‘minimal’ models in a list of six toric surfaces. We then use the combinatorial description of toric terminal singularities (see e.g. [Reference Cox, Little, Schenck and varietiesCLS11, § 11.4]) to prove the result for surfaces in that list. Then, using convex geometry, we show that the desired properties for the local moduli space are preserved after the $C_p$ -equivariant blow-ups, and hold around $[X]$ .

Remark 1.9. In Section 4.4, we produce an example of a toric surface $X$ such that ${{\rm Aut}}(X)\simeq T$ and the quotient $H^1(X,TX)//T$ is not $\mathbb{Q}$ -Gorenstein. We do not know whether this surface carries a cscK metric. If not, the expectation is that the theory should be extended to smooth pairs $(X,D)$ where $D\subset X$ is a simple normal crossing divisor, considering singular or Poincaré type cscK metrics that are non-singular away from $D$ , such as in [Reference Apostolov, Auvray and SektnanAAS21].

The paper is organised as follows. In Section 2, we settle the necessary background material on cscK manifolds. Section 3 is devoted to the classification of foldable surfaces and the proofs of Proposition 1.2 and Proposition 1.3. In Section 4, we carry over the construction of the local moduli spaces, and prove Theorem1.5. Finally, in Section 5.1, we provide maps between the local moduli spaces and speculate about their Weil–Petersson geometry, while in Section 5.2 we discuss the higher dimensional case.

1.1 Notation

We will use the notation from [Reference Cox, Little, Schenck and varietiesCLS11]. For a toric variety $X$ , ${{\rm Aut}}(X)$ stands for its automorphism group, and ${{\rm Aut}}^0(X)\subset {{\rm Aut}}(X)$ the connected component of the identity. We denote by $T$ the torus of $X$ , and by $N$ the lattice of its one-parameter subgroups, with dual lattice $M$ and pairing $\langle m , u \rangle$ , for $(m,u)\in M\times N$ . The fan of $X$ will be denoted by $\Sigma$ (or $\Sigma _X$ ) and the letters $\tau ,\sigma$ will be used for cones in $\Sigma$ . For $0\leqslant j\leqslant {{\rm dim}}(X)$ , $\Sigma (j)$ is the set of $j$ -dimensional cones in $\Sigma$ . For ${\mathbb{K}}\in \lbrace \mathbb{Q}, {\mathbb{R}},{\mathbb{C}}\rbrace$ , we let $N_{\mathbb{K}}:= N\otimes _{\mathbb{Z}} {\mathbb{K}}$ , and similarly $M_{\mathbb{K}}=M\otimes _{\mathbb{Z}}{\mathbb{K}}$ . Finally, $X_\Sigma$ may be used to refer to the toric variety associated to $\Sigma$ .

2. Background on cscK metrics and their moduli

Let $(X,\Omega )$ be a compact Kähler manifold with Kähler class $\Omega$ . We will give a very brief overview of extremal metrics and their deformations, and refer the reader to [Reference GauduchonGau, Reference SzékelyhidiSzé14] for a more comprehensive treatment.

2.1 Extremal Kähler metrics

Extremal Kähler metrics were introduced by Calabi [Reference Calabi, Bourguignon, Chen and DonaldsonCal21a] and provide canonical representatives of Kähler classes. They are defined as the critical points of the so-called Calabi functional, which assigns to each Kähler metric in $\Omega$ the $L^2$ -norm of its scalar curvature. They include, as special cases of interest, cscK metrics and thus Kähler–Einstein metrics. The following obstruction to the existence of a cscK metric is due to Matsushima and Lichnerowicz [Reference MatsushimaMat57, Reference LichnerowiczLic58].

Theorem 2.1 [Reference MatsushimaMat57, Reference LichnerowiczLic58]. Assume that $X$ carries a cscK metric. Then the automorphism group of $X$ is reductive.

Later on, Futaki discovered another obstruction to the existence of a cscK metric on $(X,\Omega )$ : the vanishing of the so-called Futaki invariant [Reference FutakiFut83, Reference FutakiFut88]. Moreover, from [Reference Calabi, Bourguignon, Chen and DonaldsonCal21b], an extremal Kähler metric in the Kähler class $\Omega$ is cscK precisely when the Futaki invariant ${\rm Fut}_{\Omega }$ vanishes. Together with Arezzo, Pacard, Singer, and Székelyhidi’s results on blow-ups of extremal Kähler metrics, we deduce the following existence result for toric surfaces.

Theorem 2.2 [Reference Arezzo, Pacard and SingerAPS11, Reference SzékelyhidiSzé15]. Let $(X,{\Omega })$ be a smooth compact polarised toric surface with cscK metric ${\omega }\in {\Omega }$ . Let $Z\subset X$ be a finite set of torus fixed points, and $G\subset {{\rm Aut}}(X)$ a finite subgroup such that:

  1. (i) the set $Z$ is $G$ -invariant;

  2. (ii) the class $\Omega$ is $G$ -invariant; and

  3. (iii) the adjoint action of $G$ on ${\rm Lie}({{\rm Aut}}(X))$ has no fixed point but zero.

Then, there is ${\varepsilon }_0 \gt 0$ such that $({\rm Bl}_Z(X), {\Omega }_{\varepsilon })$ carries a cscK metric in the class ${\Omega }_{\varepsilon }:=\pi ^*{\Omega } - {\varepsilon } c_1(E)$ for ${\varepsilon }\in (0,{\varepsilon }_0)$ , where $\pi : {\rm Bl}_Z(X)\to X$ stands for the blow-down map and $E=\sum _{z\in Z} E_z$ is the exceptional locus of $\pi$ .

Proof. From [Reference Arezzo, Pacard and SingerAPS11, Reference SzékelyhidiSzé15], there is ${\varepsilon }_0$ such that $({\rm Bl}_Z(X), {\Omega }_{\varepsilon })$ admits an extremal Kähler metric for ${\varepsilon }\in (0,{\varepsilon }_0)$ . We only need to show the vanishing of the Futaki invariant ${\rm Fut}_{{\Omega }_{\varepsilon }}$ . This will follow from its equivariance, as already used in [Reference Sektnan and TiplerST24, Proposition 2.1]. Define $\tilde X={\rm Bl}_Z(X)$ . By $(i)$ and $(ii)$ , $G$ lifts to a subgroup of ${{\rm Aut}}(\tilde X)$ such that ${\Omega }_{\varepsilon }$ is $G$ -invariant. Then, by equivariance of ${\rm Fut}_{{\Omega }_{\varepsilon }} \in ({\rm Lie} ( {{\rm Aut}}(\tilde X)))^*$ (see e.g. [Reference FutakiFut88, Chapter 3] or [Reference LeBrun and SimancaLS94, § 3.1]), we have, for all $g\in G$ and $v\in ({\rm Lie} ( {{\rm Aut}}(\tilde X)))$ ,

\begin{align*} {\rm Fut}_{{\Omega }_{\varepsilon }}({\rm Ad}_g(v))={\rm Fut}_{{\Omega }_{\varepsilon }}(v). \end{align*}

Let $\tilde w\in {\rm Lie} ( {{\rm Aut}}(\tilde X))$ . Then $\tilde w$ is the lift of an element $w\in {\rm Lie} ( {{\rm Aut}}( X))$ that vanishes on $Z$ . Moreover,

\begin{align*} \tilde w_0 := \sum _{g\in G} {\rm Ad}_g(\tilde w) \end{align*}

is a fixed point under the adjoint action of $G$ on ${\rm Lie} ( {{\rm Aut}}(\tilde X))$ . As $Z$ is $G$ -invariant, we deduce that it is the lift of a fixed point $w_0\in {\rm Lie} ( {{\rm Aut}}( X))$ under the adjoint action of $G$ . From $(iii)$ , we deduce that $\tilde w_0=0$ . Hence, by equivariance of ${\rm Fut}_{{\Omega }_{\varepsilon }}$ ,

\begin{align*} 0={\rm Fut}_{{\Omega }_{\varepsilon }}(\tilde w_0)=\sum _{g\in G} {\rm Fut}_{{\Omega }_{\varepsilon }}({\rm Ad}_g(\tilde w))=\vert G\vert \; {\rm Fut}_{{\Omega }_{\varepsilon }}(\tilde w) \end{align*}

where $\vert G \vert$ is the cardinal of $G$ . This concludes the proof.

2.2 Local moduli of polarised cscK manifolds

A moduli space $\mathscr{M}$ for polarised cscK manifolds was constructed in [Reference Dervan and NaumannDN20], generalising the results in [Reference Fujiki and SchumacherFS90] by taking care of non-discrete automorphisms by means of [Reference SzékelyhidiSzé10, Reference InoueIno19]. This space is Hausdorff and is endowed with a complex space structure. It is a coarse moduli space in the following sense: its points are in bijective correspondence with isomorphism classes of polarised cscK manifolds and, for any complex analytic family $(\mathscr{X},\mathscr{L}) \to {\mathscr{S}}$ of polarised cscK manifolds over some reduced analytic space $\mathscr{S}$ ( $\mathscr{L}$ stands for the relative polarisation of $\mathscr{X}\to {\mathscr{S}}$ ), there is an induced map ${\mathscr{S}} \to \mathscr{M}$ . In what follows, we will restrict to the connected components corresponding to the moduli space of polarised cscK surfaces, still denoted by $\mathscr{M}$ . Roughly, $\mathscr{M}$ is constructed by patching together local moduli spaces $(\mathscr{M}_X)_{X\in \mathfrak{S}}$ parameterised by $\mathfrak{S}$ , the set of polarised cscK surfaces, each $\mathscr{M}_X$ being obtained as some open set in an analytic GIT quotient $W_X//G$ . Our case of interest is when $X$ is a cscK toric surface. The first cohomology group of the tangent sheaf, namely $H^1(X,TX)$ , plays a central role in our study. Indeed, it naturally arises as the space of infinitesimal deformations for $X$ , and will play the role of the tangent space at $X$ to the base of a semi-universal family of deformations of $X$ (see e.g. [Reference KodairaKod05, Chapter 4]). The tangent bundle $TX$ is naturally ${{\rm Aut}}(X)$ -equivariant, meaning that the ${{\rm Aut}}(X)$ -action lifts to an action on $TX$ in such a way that the projection map $TX\to X$ is equivariant (the lift is given by the differential of the action). Using pullbacks, we see that the bundles $\Lambda ^{p,q}TX^*$ are also equivariant. Then, $\Lambda ^{p,q}TX^*\otimes TX$ is equivariant, from which we deduce that ${{\rm Aut}}(X)$ acts naturally on $TX$ -valued $(p,q)$ -forms. More explicitly, for any ${{\rm Aut}}(X)$ -equivariant bundle $E\to X$ , we deduce an action of ${{\rm Aut}}(X)$ on the sections of $E$ by setting, for $s\in \Gamma (E)$ , $g\in {{\rm Aut}}(X)$ and $x\in X$ :

\begin{align*} (g\cdot s)(x):= g( s(g^{-1}x)). \end{align*}

From holomorphicity of the action, we deduce that the ${{\rm Aut}}(X)$ -action on $TX$ -valued $(0,p)$ -forms commutes with the Dolbeault operator, and hence induces a representation of ${{\rm Aut}}(X)$ on

\begin{align*} H^1(X,TX)\simeq \frac {\ker \:( \overline \partial : {\Omega }^{0,1}(TX)\to {\Omega }^{0,2}(TX))}{{\rm Im}\:(\overline \partial : {\Omega }^{0,0}(TX)\to {\Omega }^{0,1}(TX))}. \end{align*}

In a more formal way, we could say that the Dolbeault resolution of the holomorphic tangent bundle is actually a resolution by injectives in the category of ${{\rm Aut}}(X)$ -equivariant sheaves, and thus that the right derived functors of the global sections functor $\Gamma$ , namely the cohomology groups $H^\bullet (X,TX)$ , inherit an ${{\rm Aut}}(X)$ -structure. We then extract from [Reference Dervan and NaumannDN20] the following proposition.

Proposition 2.3. Let $(X,{\Omega })$ be a polarised cscK toric surface. Then the local moduli space $\mathscr{M}_X$ is given by an open neighbourhood of $\pi (0)$ in the GIT quotient

\begin{align*} \pi : H^1(X,TX) \to H^1(X,TX)//{{\rm Aut}}(X). \end{align*}

In particular, its singularities are of klt type.

The above GIT quotient in this affine setting is given by the good categorical quotient that we now recall. Let $R$ be the ring of regular functions on $H^1(X,TX)$ . The ${{\rm Aut}}(X)$ -action on $H^1(X,TX)$ induces an ${{\rm Aut}}(X)$ -module structure on $R$ , given by

\begin{align*}g\cdot f:=f(g\:\cdot \:)\end{align*}

for any $g\in {{\rm Aut}}(X)$ and $f\in R$ . We then introduce $R^{{{\rm Aut}}(X)}$ as the set of ${{\rm Aut}}(X)$ -invariant elements. By a classical result of Hilbert and Nagata, $R^{{{\rm Aut}}(X)}$ is finitely generated as soon as ${{\rm Aut}}(X)$ is reductive. In that case, the GIT quotient is given by

\begin{align*} H^1(X,TX)//{{\rm Aut}}(X):={\rm Spec}(R^{{{\rm Aut}}(X)}), \end{align*}

with quotient map

\begin{align*}\pi : {\rm Spec}(R) \to {\rm Spec}(R^{{{\rm Aut}}(X)})\end{align*}

corresponding to the inclusion of the finitely generated algebras

\begin{align*} R^{{{\rm Aut}}(X)}\to R. \end{align*}

We will then say that a point in $H^1(X,TX)$ is polystable if its ${{\rm Aut}}(X)$ -orbit is closed and that it is stable if, in addition, its stabiliser is discrete. We refer the reader to [Reference Cox, Little, Schenck and varietiesCLS11, § 5.0] for a short account of good categorical quotients and to [Reference Cox, Little, Schenck and varietiesCLS11, § 14.1] for the special case of toric GIT quotients. Finally, we refer to Section 4.3 for the definition of klt type singularities.

Proof of Proposition2.3 . The tangent space at $X$ , denoted $\tilde H^1(TX)$ , of the base of a semi-universal family of complex deformations of $X$ compatible with the polarisation $\Omega$ is constructed in [Reference Chen and SunCS14, Lemma 6.1] (see also [Reference SzékelyhidiSzé10]). In general, the construction of $\tilde H^1(TX)$ goes as follows. First, the usual Kuranishi complex for deformations of complex structures on $X$ is

\begin{align*} \cdots \to {\Omega }^{0,k}(TX) \stackrel {\overline \partial }{\to } {\Omega }^{0,k+1}(TX) \to \cdots , \end{align*}

where the extension of $\overline \partial$ to $(0,p)$ -forms is given in local coordinates, for $\beta = \sum _j \beta _j \otimes \frac {\partial }{\partial z^j}$ , by

\begin{align*} \overline \partial \beta = \sum _j \overline \partial \beta _j \otimes \frac {\partial }{\partial z^j}. \end{align*}

We are interested in deformations that are compatible with some cscK metric ${\omega }\in {\Omega }$ . Following [Reference Fujiki and SchumacherFS90], we define maps

\begin{align*} \begin{array}{cccc} \iota ^k_\bullet : & {\Omega }^{0,k}(TX) & \to & {\Omega }^{0,k+1}\\ & \beta & \mapsto & \iota _\beta {\omega } \end{array} \end{align*}

where $\iota _\beta {\omega }$ is obtained by the composition of first contraction and then alternation operators. For $k\geqslant 1$ , we then set

\begin{align*} {\Omega }_{{\omega }}^{0,k}(TX) := \ker \iota ^k_\bullet , \end{align*}

while we define

\begin{align*} {\Omega }_{{\omega }}^{0,0}(TX):= {\mathscr{C}}^\infty (X,{\mathbb{C}}). \end{align*}

Together with the restriction of $\overline \partial$ , and the map $\overline \partial^0:=\mathscr{D}$ defined, for $f\in {\mathscr{C}}^\infty (X,{\mathbb{C}})$ , by

\begin{align*} \mathscr{D}(f):=\overline \partial (\nabla _{{\omega }}^{1,0} f), \end{align*}

where $\nabla _{{\omega }}^{1,0} f$ is given by $\overline \partial f = {\omega }( \nabla _{{\omega }}^{1,0}f , \cdot )$ , we obtain another elliptic complex $({\Omega }_{{\omega }}^{0,\bullet }(TX), \overline \partial ^\bullet )$ . We denote by $\tilde H^{\bullet }(TX)$ the associated cohomology groups. Then, from this complex, following Kuranishi’s techniques, one can build a semi-universal family $(\mathscr{X},\mathscr{L})\to Z$ of polarised deformations of $(X,\Omega )$ , where $Z\subset B\subset \tilde H^1(TX)$ is an analytic subspace (corresponding to integrable infinitesimal deformations) of some open ball $B$ centred at zero (again, we refer to [Reference Chen and SunCS14, Lemma 6.1] for the detailed construction). As in [Reference DoanDoa22], setting $K:={\rm Aut}(X,{\omega })$ , the group of holomorphic isometries of $(X,\omega )$ , this construction can actually be made $K$ -equivariantly, and even locally $K^{\mathbb{C}}$ -equivariantly on $Z$ , for the natural $K^{\mathbb{C}}$ -action on $\tilde H^1(TX)$ . From [Reference Dervan and NaumannDN20, § 3], $\mathscr{M}_X$ is given by an open neighbourhood of $0$ in the analytic GIT quotient $W_X//G$ , where $G=K^{\mathbb{C}}$ , and where $W_X\subset \tilde H^1(TX)$ is the smallest $G$ -invariant Stein space containing $Z$ . As noted in [Reference Rollin and TiplerRT14, Lemma 2.10], as $X$ is toric, by Bott, Steenbrink and Danilov’s vanishing of $h^{0,1}(X)$ and $h^{0,2}(X)$ (see [Reference Cox, Little, Schenck and varietiesCLS11, Theorem 9.3.2]), the vector space $\tilde H^1(TX)$ is $G$ -equivariantly isomorphic to $H^1(X,TX)$ . Also, for a toric surface, ${{\rm Aut}}(X)$ is a linear algebraic group, so that $G\simeq {{\rm Aut}}(X)$ , as the Lie algebra of ${{\rm Aut}}(X)$ has no parallel vector field (see [Reference GauduchonGau, Chapter 2]). Finally, $H^2(X,TX)=0$ by [Reference IltenIlt11, Corollary 1.5], so that any small element of $H^1(X,TX)$ is integrable, and $Z=B\subset H^1(X,TX)$ . Hence $W_X=H^1(X,TX)$ , and putting all this together, we obtain the first part of Proposition 2.3. The statement about klt type singularities follows directly from [Reference Braun, Greb, Langlois and JoaquinBGLJ24, Theorem 1], which asserts that the GIT quotient of a klt type singularity is of klt type, and which implies in particular that the GIT quotient of $H^1(X,TX)$ (which is smooth) by ${{\rm Aut}}(X)$ is of klt type.

3. Foldable toric surfaces

Let $N$ be a rank two lattice and $\Sigma$ be a fan in $N_{\mathbb{R}}$ with associated toric surface $X$ . We will assume $\Sigma$ to be smooth, that is, that each cone $\sigma \in \Sigma$ is generated by elements in $N$ that form part of a $\mathbb{Z}$ -basis, and to be complete, i.e.

\begin{align*} \bigcup _{\sigma \in \Sigma } \sigma =N_{\mathbb{R}}. \end{align*}

Hence, $X$ is a smooth and compact toric surface.

3.1 Lattice automorphisms and fans

We will be interested in automorphism groups of smooth toric surfaces. Denote by $T:=N\otimes _{\mathbb{Z}}{\mathbb{C}}^*$ the torus of $X$ , and by ${{\rm Aut}}(N,\Sigma )$ the group of lattice automorphisms of $\Sigma$ . Recall that ${{\rm Aut}}(N,\Sigma )$ is the subgroup of ${\rm GL}_{\mathbb{Z}}(N)$ consisting of elements $g\in {\rm GL}_{\mathbb{Z}}(N)$ such that the induced isomorphisms $g\in {\rm GL}_{\mathbb{R}}(N_{\mathbb{R}})$ send $\Sigma$ bijectively to $\Sigma$ . By a result of Demazure [Reference DemazureDem70, Proposition 11], the automorphism group ${{\rm Aut}}(X)$ of $X$ is a linear algebraic group isomorphic to

\begin{align*} {{\rm Aut}}^0(X)\rtimes G, \end{align*}

where $G$ is a quotient of ${{\rm Aut}}(N,\Sigma )$ . We will later on be interested in the GIT quotient of $H^1(X,TX)$ by ${{\rm Aut}}(X)$ , assuming $X$ to admit a cscK metric. A combination of Matsushima and Lichnerowicz’s obstruction together with Nill’s work on toric varieties with reductive automorphism group [Reference NillNil06] implies the following proposition.

Proposition 3.1. Assume that $X$ is a toric surface that admits a cscK metric. Denote by $\Sigma$ its fan and by $N$ its lattice of one-parameter subgroups. Then either

\begin{align*} X\in \lbrace \mathbb{P}^2, \mathbb{P}^1\times \mathbb{P}^1 \rbrace , \end{align*}

or

\begin{align*} {{\rm Aut}}(X)\simeq T\rtimes {{\rm Aut}}(N,\Sigma ). \end{align*}

Proof. By Matsushima and Lichnerowicz’s theorem [Reference MatsushimaMat57, Reference LichnerowiczLic58], ${{\rm Aut}}(X)$ is reductive. The result then follows from Nill’s result [Reference NillNil06, Theorem 1.8] together with Demazure’s structure theorem [Reference DemazureDem70, Proposition 11, p. 581], which we now recall. Demazure’s root system for $(N,\Sigma )$ , which we denote here by $\mathscr{R}$ (it corresponds to ${\rm Rac}(\Sigma )$ in Demazure’s notation [Reference DemazureDem70]), is the set

\begin{align*} {\mathscr{R}}:=\lbrace m\in M\:\vert \: \exists \rho \in \Sigma (1)\:\textrm{with}\: \langle m, u_\rho \rangle=1 \:\textrm{but}\:\forall \rho '\in \Sigma (1),\: \rho '\neq \rho , \langle m , u_{\rho '}\rangle \leqslant 0\rbrace , \end{align*}

where we use the notation $u_\rho \in N$ for the primitive generator of the ray $\rho$ . Demazure’s structure theorem then asserts that the identity component ${{\rm Aut}}(X)^0$ of ${{\rm Aut}}(X)$ is generated by $T$ and some unipotent elements $\lbrace U_m, m\in {\mathscr{R}}\rbrace$ . In particular,

\begin{align*} {{\rm dim}}({{\rm Aut}}(X))={{\rm dim}}(T)+\vert {\mathscr{R}}\vert . \end{align*}

Then, the quotient ${{\rm Aut}}(X)/{{\rm Aut}}(X)^0$ is finite and isomorphic to a quotient of ${{\rm Aut}}(N,\Sigma )$ by a subgroup $W(N,\Sigma )\subset {{\rm Aut}}(N,\Sigma )$ that is the Weyl group of a maximal reductive subgroup of ${{\rm Aut}}(X)$ with root system ${\mathscr{R}}\cap -{\mathscr{R}}$ . Then Nill’s theorem [Reference NillNil06, Theorem 1.8] gives an upper bound for the dimension of the automorphism group of a smooth complete toric variety, when it is reductive. In the case of a smooth toric surface $X$ with reductive automorphism group, the outcome is that either $X$ is a product of projective spaces or ${{\rm dim}}({{\rm Aut}}(X))=2$ . In the latter case, from the previous discussion on Demazure’s result, this implies that the set of Demazure’s roots $\mathscr{R}$ for $(N,\Sigma )$ is empty. Hence, ${{\rm Aut}}(X)^0=T$ and ${{\rm Aut}}(X)/{{\rm Aut}}(X)^0\simeq {{\rm Aut}}(N,\Sigma )$ , which concludes the proof.

As $\mathbb{P}^2$ and $\mathbb{P}^1\times \mathbb{P}^1$ are rigid, we will now focus on the case

\begin{align*} {{\rm Aut}}(X)\simeq T\rtimes {{\rm Aut}}(N,\Sigma ), \end{align*}

and characterise the possible finite groups that arise as lattice automorphism groups of rank two complete fans.

Lemma 3.2. Let $g\in {{\rm Aut}}(N,\Sigma )$ , and denote by $m\in \mathbb{N}$ the order of $g$ . Then

\begin{align*}m\in \lbrace 1, 2, 3, 4, 6\rbrace .\end{align*}

Moreover, the complex linear extension of $g$ to $N_{\mathbb{C}}$ is conjugated (in ${\rm GL}(N_{\mathbb{C}})\simeq {\rm GL}_2({\mathbb{C}})$ ) to the following:

  1. (i) if $m=1$ , then $g \sim {{\rm Id}}$ ,

  2. (ii) if $m=2$ and $\det (g)=1$ , then $g\sim -{{\rm Id}}$ ,

  3. (iii) if $m=2$ and $\det (g)=-1$ , then $g\sim \left [ \begin{array}{cc} 1 &\quad 0 \\ 0 &\quad -1 \end{array} \right ]$ ,

  4. (iv) if $m=3$ , then $g\sim \left [ \begin{array}{cc} j &\quad 0 \\ 0 &\quad j^2 \end{array} \right ]$ ,

  5. (v) if $m=4$ , then $g\sim \left [ \begin{array}{cc} i &\quad 0 \\ 0 &\quad -i \end{array} \right ]$ , and

  6. (vi) if $m=6$ , then $g\sim \left [ \begin{array}{cc} -j &\quad 0 \\ 0 &\quad -j^2 \end{array} \right ]$ ,

where we write $j=e^{i\frac {2\pi }{3}}$ and denote by $\sim$ the equivalence relation given by conjugation, and where we use an isomorphism $N_{\mathbb{C}} \simeq {\mathbb{C}}^2$ .

The proof is an elementary exercise in linear algebra. We include it for convenience of the reader.

Proof. Fix an isomorphism $N\simeq \mathbb{Z}^2$ and identify $g$ with an element of ${\rm GL}_2(\mathbb{Z})$ . The characteristic polynomial of

\begin{align*}g =\left [ \begin{array}{cc} a &\quad b \\ c &\quad d \end{array} \right ]\end{align*}

then reads

\begin{align*} \chi (Y)=Y^2-(a+d)Y+(ad-bc). \end{align*}

On the other hand, Jordan’s normal form for the $\mathbb{C}$ -linear extension of $g$ implies that it is conjugated (in ${\rm M}_2({\mathbb{C}})$ ) to

\begin{align*} \left [ \begin{array}{cc} \alpha &\quad \delta \\ 0 &\quad \beta \end{array} \right ] \end{align*}

for $(\alpha ,\beta )\in {\mathbb{C}}^2$ and $\delta \in \lbrace 0,1 \rbrace$ . As $g$ is both invertible and of finite order, we deduce that $\delta=0$ and $g$ is conjugated to

\begin{align*} \left [ \begin{array}{cc} \alpha &\quad 0 \\ 0 &\quad \beta \end{array} \right ]. \end{align*}

Moreover, $\alpha ^m=\beta ^m=1$ , so that $\alpha$ and $\beta$ are $m$ -th roots of unity in $\mathbb{C}$ . On the other hand, by conjugation invariance, we also have

\begin{align*} \det (g)=\alpha \beta =ad-bc\in \lbrace -1,+1 \rbrace \end{align*}

as $g\in {\rm GL}_2(\mathbb{Z})$ and

\begin{align*} {\rm trace}(g)=\alpha +\beta =a+d \in \mathbb{Z}. \end{align*}

If $\alpha$ and $\beta$ belong to $\lbrace -1, +1 \rbrace$ , that is if $m=2$ , then we are done. If not, at least one of $\alpha$ or $\beta$ is not real, and thus $\alpha$ and $\beta$ must be complex conjugated roots of $\chi (Y)$ . Hence

\begin{align*}{\rm trace}(g)=\alpha +\overline {\alpha }\in \mathbb{Z},\end{align*}

but also

\begin{align*} \vert \alpha +\overline {\alpha } \vert \lt 2\end{align*}

so that

\begin{align*}\alpha +\overline {\alpha } \in \lbrace -1,0, 1 \rbrace . \end{align*}

Hence

\begin{align*}\alpha \in \lbrace \pm i, \pm j, \pm j^2 \rbrace ,\end{align*}

from which the result follows easily.

Recall that we denote by $C_p$ the cyclic group of order $p$ and by $D_p$ the dihedral group of order $p$ . We then have the following proposition.

Proposition 3.3. The group ${{\rm Aut}}(N,\Sigma )$ is isomorphic to one of the groups in the following set

\begin{align*} \lbrace C_1, C_2, C_3, C_4,C_6, D_1, D_2, D_3, D_4, D_6 \rbrace . \end{align*}

Proof. As ${{\rm Aut}}(N,\Sigma )$ is finite, we can fix an ${{\rm Aut}}(N,\Sigma )$ -invariant euclidean metric on $N_{\mathbb{R}}$ (again considering the $\mathbb{R}$ -linear extension of ${{\rm Aut}}(N,\Sigma )$ ). Then, by the classification of finite subgroups of the group of orthogonal transformations of the plane, we deduce that ${{\rm Aut}}(N,\Sigma )$ is isomorphic to $C_p$ or $D_p$ , for some $p\in \mathbb{N}^*$ . As those groups admit elements of order $p$ , by Lemma 3.2, the result follows.

Remark 3.4. This proposition simply recovers the well-known classification of crystallographic groups in dimension $2$ .

We will now show that all the groups listed above arise as lattice automorphism groups of rank two complete smooth fans. Note that $C_2$ and $D_1$ will be distinguished by the fact that their generator is $-{{\rm Id}}$ in the first case and a reflection in the second case. Consider then the following fans in $\mathbb{Z}^2$ , where we denote by $\lbrace e_1, e_2 \rbrace$ the standard basis of $\mathbb{Z}^2$ .

Example 3.5. Let ${\mathbb{F}}_n$ be the $n$ th Hirzebruch surface, that is, the total space of the fibration

\begin{align*}{\mathbb{P}}(\mathscr{O}_{\mathbb{P}^1}\oplus \mathscr{O}_{\mathbb{P}^1}(n)) \rightarrow {\mathbb{C}}{\mathbb{P}}^1.\end{align*}

Note that ${\mathbb{F}}_0={\mathbb{C}}{\mathbb{P}}^1\times {\mathbb{C}}{\mathbb{P}}^1$ . Then, up to isomorphism, the fan $\Sigma _{{\mathbb{F}}_n}$ of ${\mathbb{F}}_n$ is described by

\begin{align*}\Sigma _{{\mathbb{F}}_n}(1)=\lbrace {\mathbb{R}}_+\cdot (0,-1),\:{\mathbb{R}}_+\cdot (1,0),\:{\mathbb{R}}_+\cdot (0,1),\:{\mathbb{R}}_+\cdot (-1,-n) \rbrace .\end{align*}

As an example, ${\mathbb{F}}_2$ admits the fan description of Figure 1, which we will denote by $\Sigma _1'$ . In this figure, and those that follow, the dots represent the lattice points, and we only represent the ray generators of the complete two dimensional fan.

Figure 1. Fan $\Sigma _1'$ of ${\mathbb{F}}_2$ .

Example 3.6. In Figures 2 and 3, respectively, the fans $\Sigma _{\mathbb{P}^2}=\Sigma _3'$ and $\Sigma _{\mathbb{P}^1\times \mathbb{P}^1}=\Sigma _4'$ of $\mathbb{P}^2$ and $\mathbb{P}^1\times \mathbb{P}^1$ are depicted (the numbering of the fans is motivated by Proposition 3.9 below).

Figure 2. Fan $\Sigma _3'$ of $\mathbb{P}_2$ .

Figure 3. Fan $\Sigma _4'$ of ${\mathbb{C}}{\mathbb{P}}^1\times {\mathbb{C}}{\mathbb{P}}^1$ .

Example 3.7. Recall that blowing up a smooth toric surface along a torus fixed point produces a new smooth toric surface. By the orbit cone correspondence, such a fixed point corresponds to a two dimensional cone

\begin{align*}\sigma ={\mathbb{R}}_+\cdot e_i + {\mathbb{R}}_+\cdot e_{i+1}\end{align*}

in the fan of the blown-up surface, and the fan of the resulting surface is obtained by adding the ray generated by $e_i + e_{i+1}$ to the set of rays of the initial surface (see [Reference Cox, Little, Schenck and varietiesCLS11, Chapter 3, § 3]). By iterated blow-ups, we obtain the following fans. Figure 4 represents an iterated blow-up of $\mathbb{P}^1\times \mathbb{P}^1$ . The following Figure 5 is a two points blow-up of $\mathbb{P}^1\times \mathbb{P}^1$ and, at the same time, a three points blow-up of ${\mathbb{C}}\mathbb{P}^2$ .

Example 3.8. Here are further examples to complete our classification of lattice automorphism groups of rank two complete smooth fans (Figures 6, 7, 8, 9 and 10). The first is a single blow-up of ${\mathbb{F}}_2$ , which cancels the symmetry of $\Sigma _1'=\Sigma _{{\mathbb{F}}_2}$ (Figure 6). The next (Figure 7) is obtained from $\Sigma _2'$ by blowing up in a $C_2$ -equivariant way (recall that $C_2=\langle -{{\rm Id}} \rangle$ ). The third example is obtained from the fan of ${\mathbb{P}}^2$ by three successive blow-ups of three distinct fixed points, in a symmetric way (Figure 8). The next example is obtained by two successive blow-ups of ${\mathbb{P}}^1\times {\mathbb{P}}^1$ at four fixed points (Figure 9). The last is obtained from $\Sigma _6'$ by two successive blow-ups at six fixed points (Figure 10).

Figure 4. Fan $\Sigma _2'$ : iterated blow-up of $\mathbb{P}^1\times \mathbb{P}^1$ .

Figure 5. Fan $\Sigma _6'$ : blow-up of ${\mathbb{P}}^2$ along its three fixed points.

Figure 6. Fan $\Sigma _1$ : one-point blow-up of ${\mathbb{F}}_2$ .

Figure 7. Fan $\Sigma _2$ .

Figure 8. Fan $\Sigma _3$ .

Figure 9. Fan $\Sigma _4$ .

Figure 10. Fan $\Sigma _6$ .

Proposition 3.9. Let $p\in \lbrace 1,2,3,4,6 \rbrace$ . Then we have

\begin{align*}{{\rm Aut}}(\mathbb{Z}^2,\Sigma _p)\simeq C_p\end{align*}

and

\begin{align*}{{\rm Aut}}(\mathbb{Z}^2,\Sigma _p')\simeq D_p.\end{align*}

Proof. The proof is straightforward, although a bit tedious, so we will only give a sketch of it. Fix an orientation of ${\mathbb{R}}^2$ , and pick a two dimensional cone $\sigma$ of $\Sigma _p$ (the proof is the same for $\Sigma _p'$ ). Let $g\in {{\rm Aut}}(\mathbb{Z}^2,\Sigma _p)$ . Then $g\cdot \sigma \in \Sigma _p(2)$ . Moreover, $g$ maps the two facets of $\sigma$ to the two facets of $g\cdot \sigma$ . As the fan $\Sigma _p$ is explicitly described, for each possible $\tau =g\cdot \sigma \in \Sigma _p(2)$ , we deduce the possible matrix coefficients for $g$ ,

\begin{align*} g=\left [ \begin{array}{cc} a_\tau &\quad b_\tau \\ c_\tau &\quad d_\tau \end{array} \right ]\in {\rm GL}_2(\mathbb{Z}). \end{align*}

For any possible $\tau$ , one can then explicitly compute the images of the rays of $\Sigma _p$ by $g$ . They should all belong to $\Sigma _p(1)$ , and this enables us to select the allowed cones $\tau$ for $g$ to be a lattice automorphism of $\Sigma _p$ . The result then follows easily.

Remark 3.10. One can show similarly that for any Hirzebruch surface ${\mathbb{F}}_n$ with $n\geqslant 1$ and with fan $\Sigma _{{\mathbb{F}}_n}$ , one has ${{\rm Aut}}(N,\Sigma _{{\mathbb{F}}_n})\simeq D_1$ .

3.2 Classification of foldable surfaces

We first recall the following definition.

Definition 3.11. We say that a smooth and complete two dimensional fan is foldable if its lattice automorphism group contains a non-trivial cyclic group. A foldable surface is a toric surface whose fan is foldable.

We now proceed to a partial classification of foldable toric surfaces. Notice that if $G\subset {{\rm Aut}}(N,\Sigma )$ , then we can find a $G$ -action on $X$ that is faithful as soon as ${{\rm Aut}}(X)\simeq T\rtimes {{\rm Aut}}(N,\Sigma )$ .

Definition 3.12. A $G$ -equivariant blow-up of $X$ is a blow-up of $X$ along a $G$ -orbit of torus fixed points. We say that a smooth toric surface $\tilde X$ is obtained from $X$ by successive $G$ -equivariant blow-ups if there is a sequence of $G$ -equivariant blow-ups:

\begin{align*}\tilde X=X_k \to X_{k-1} \to \cdots \to X_1 \to X_0=X.\end{align*}

Denote by ${\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)$ the toric surface associated to the fan of Figure 5, that is, the blow-up of $\mathbb{P}^2$ at its three torus fixed points.

Proposition 3.13. Let $X$ be a foldable surface with fan $\Sigma$ . Assume that ${{\rm Aut}}(N,\Sigma )$ contains a subgroup isomorphic to $C_p$ , $p \geqslant 2$ .

  1. (i) If $p\in \lbrace 2, 4 \rbrace$ , then $X$ is obtained from $\mathbb{P}^1\times \mathbb{P}^1$ by successive $C_p$ -equivariant blow-ups.

  2. (ii) If $p=3$ , then $X$ is obtained from $\mathbb{P}^2$ by successive $C_p$ -equivariant blow-ups.

  3. (iii) If $p=6$ , then $X$ is obtained from ${\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)$ by successive $C_p$ -equivariant blow-ups.

Proof. Through its representation

\begin{align*}C_p \to {\rm GL}_{\mathbb{R}}(N_{\mathbb{R}}), \end{align*}

the subgroup $C_p$ of ${{\rm Aut}}(N,\Sigma )$ acts on $N_{\mathbb{R}}$ by rotations (once an invariant euclidean metric is fixed), and hence has no fixed points. We deduce that the action of $C_p$ on $\Sigma (1)$ is free, and then $p$ divides $\vert \Sigma (1)\vert$ , the number of rays of $\Sigma$ .

On the other hand, from the classification of toric surfaces (see e.g. [Reference Cox, Little, Schenck and varietiesCLS11, Chapter 10]), we must have $ \vert \Sigma (1)\vert \geqslant 3$ . Moreover, if $\vert \Sigma (1)\vert=3$ , then $X\simeq \mathbb{P}^2$ , while if $\vert \Sigma (1)\vert=4$ , then $X\simeq {\mathbb{F}}_n$ for some $n\in \mathbb{N}$ . By Remark 3.10, we see that $X\simeq \mathbb{P}^1\times \mathbb{P}^1$ in that case.

So we may assume that $\vert \Sigma (1)\vert \geqslant 5$ , and that $p$ divides $\vert \Sigma (1)\vert$ . By the classification of toric surfaces, we know that $X$ is obtained by successive blow-ups from ${\mathbb{F}}_n$ , $n\geqslant 0$ , or from $\mathbb{P}^2$ . Then, there is a ray $\rho _i$ in $\Sigma (1)$ generated by the sum of two primitive elements

\begin{align*}u_i=u_{i-1}+u_{i+1}\end{align*}

generating adjacent rays $\rho _{i-1}$ and $\rho _{i+1}$ in $\Sigma (1)$ (that is both $\rho _i+\rho _{i-1}$ and $\rho _i+\rho _{i+1}$ belong to $\sigma (2)$ ). Note that

\begin{align*} \det (u_{i+1},u_{i-1})=\det (u_i-u_{i-1},u_{i-1})=\det (u_i,u_{i-1})=1 \end{align*}

so the contraction to a point of the divisor associated to the ray $\rho _i$ is smooth. By linearity, for any $g\in C_p$ we have

\begin{align*}g\cdot u_i=g\cdot u_{i-1}+g\cdot u_{i+1}, \end{align*}

and hence we can contract to points the $C_p$ -orbit of the torus invariant divisor $D_{\rho _i}$ associated to $\rho _i$ and obtain a smooth toric surface whose fan still admits $C_p$ as a subgroup of its lattice symmetry group. By an inductive argument on $\vert \Sigma (1)\vert$ , we obtain the result.

From the classification, we obtain the following proposition.

Proposition 3.14. Let $X$ be a foldable surface. Then $X$ admits a cscK metric.

Proof. This is simply an induction on the number of $C_p$ -equivariant blow-ups in Proposition 3.13, using Theorem2.2 at each step, and the fact that $\mathbb{P}^1\times \mathbb{P}^1$ , $\mathbb{P}^2$ and ${\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)$ admit cscK metrics in $C_p$ -equivariant classes (for ${\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)$ , this follows again from Theorem2.2 applied to $\mathbb{P}^2$ ).

4. The local cscK moduli around a foldable surface

Let $X$ be a smooth and complete toric surface with fan $\Sigma$ and one-parameter subgroups lattice $N$ . We will denote by $(u_i)_{0\leqslant i \leqslant r}$ the primitive elements of $N$ that generate the rays

\begin{align*}\rho _i={\mathbb{R}}_+\cdot u_i \in \Sigma (1),\end{align*}

labelled in counterclockwise order, so that, for each $i$ , $(u_i, u_{i+1})$ is a positively oriented $\mathbb{Z}$ -basis of $N$ , and $\rho _i+\rho _{i+1}\in \Sigma (2)$ , with the convention $u_0=u_r$ .

4.1 Deformation theory of toric surfaces

We recall here some results from [Reference IltenIlt11] regarding the deformation theory of toric surfaces. As $X$ is toric, dualising the generalised Euler sequence yields (see e.g. [Reference Cox, Little, Schenck and varietiesCLS11, Theorem 8.1.6])

\begin{align*} 0\to {\rm Pic}(X)^*\otimes _{\mathbb{Z}}{\mathscr{O}}_X\to \bigoplus _{i=1}^r{\mathscr{O}}_X(D_{\rho _i})\to TX\to 0, \end{align*}

where ${\rm Pic}(X)$ stands for the Picard group of $X$ and $D_{\rho _i}$ is the torus invariant divisor associated to the ray $\rho _i\in \Sigma (1)$ . This sequence is actually a short exact sequence of ${{\rm Aut}}(X)$ -equivariant bundles, and induces a long exact sequence of ${{\rm Aut}}(X)$ -modules in cohomology. From a classical vanishing theorem (see e.g. [Reference Cox, Little, Schenck and varietiesCLS11, Theorem 9.3.2]), we have $H^1(X,{\mathscr{O}}_X)=0$ and $H^2(X,{\mathscr{O}}_X)=0$ , and this long exact sequence induces an isomorphism of ${{\rm Aut}}(X)$ -modules:

(1) \begin{align} \bigoplus _{i=1}^r H^1(X,{\mathscr{O}}_X(D_{\rho _i}))\simeq H^1(X,TX). \end{align}

In particular, the action of the torus $T$ on $X$ induces a representation of $T$ on

\begin{align*} V:=H^1(X,TX) \end{align*}

compatible with the above isomorphism. The complete reducibility theorem for torus actions (see [Reference Cox, Little, Schenck and varietiesCLS11, Proposition 1.1.2] and the references therein) provides a weight space decomposition that we will denote by

\begin{align*} V=\bigoplus _{m\in M} V_m. \end{align*}

More precisely, $V$ splits into $T$ -invariant subspaces $V_m\subset V$ where the $T$ -action on $V_m$ is given by

\begin{align*} t\cdot x= \chi ^m(t)\,x \end{align*}

for $t\in T$ , $x\in V_m$ , and where $\chi ^m : T \to {\mathbb{C}}^*$ is the character associated to the weight $m\in M$ . Using the isomorphism (1), Ilten obtained, for each weight $m\in M={{\rm Hom}}_{\mathbb{Z}}(N,\mathbb{Z})$ ([Reference IltenIlt11, Corollary 1.5]),

(2) \begin{align} {{\rm dim}}(V_m)=\sharp \displaystyle \left \{ \rho _i\in \Sigma (1)\: \left \vert \begin{array}{c} \langle m, u_i \rangle=-1 \\[3pt] \langle m, u_{ i\pm 1} \rangle \lt 0 \end{array} \right . \right \}. \end{align}

where $\sharp$ stands for the cardinal.

Figure 11. Fan of $Y_4$ .

Figure 12. Fan of $Y_3$ .

Example 4.1. We will provide explicit examples of weight space decompositions for $V=H^1(X,TX)$ , when $X$ is the smooth toric surface associated to a foldable fan. First, $\mathbb{P}^1 \times \mathbb{P}^1$ , $\mathbb{P}^2$ and ${\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)$ are rigid, as can be checked directly using Formula (2). In those cases, $V=0$ . We denote by $Y_2$ the foldable toric surface associated to the fan $\Sigma _2'$ of Figure 4. We also introduce the foldable toric surfaces $Y_4$ and $Y_3$ associated to the fans in Figures 11 and 12.

One can compute directly that, for $i\in \lbrace 2,3,4\rbrace$ ,

\begin{align*}{{\rm Aut}}(N,\Sigma _{Y_i})\simeq D_i.\end{align*}

Set $V_i=H^1(Y_i,TY_i)$ . Denote by $(e_1,e_2)$ the standard basis of $N=\mathbb{Z}^2$ and by $(e_1^*,e_2^*)$ the dual basis of $M=(\mathbb{Z}^2)^*\simeq \mathbb{Z}^2$ . Testing the conditions in Formula (2) for each ray generator gives

(3) \begin{align} V_2 & = V^1_{+e_2^*}\oplus V^1_{-e_2^*}\oplus V^1_{-e_1^*+e_2^*}\oplus V^1_{e_1^*-e_2^*},\nonumber\\[5pt] V_3 & = V^2_{-e_2^*}\oplus V^2_{e_1^*}\oplus V^2_{-e_1^*+e_2^*}, \nonumber\\[5pt] V_4 & = V^1_{+e_1^*}\oplus V^1_{-e_1^*}\oplus V^1_{+e_2^*}\oplus V^1_{-e_2^*}, \end{align}

where $V^j_m$ stands for a $j$ -dimensional weight $m$ representation of $T$ .

Remark 4.2. One can check, by unravelling the isomorphisms used in [Reference IltenIlt11, § 1], that the $C_3$ -action on $Y_3$ induces an action on the coordinates $\chi _j^m$ of $V_3$ , with two orbits given by

\begin{align*}\left\lbrace \chi _j^{-e_2^*},\chi _j^{e_1^*}, \chi _j^{-e_1^*+e_2^*}\right\rbrace , \quad j\in \lbrace 1,2\rbrace \end{align*}

where $\chi _j^m$ , for $1\leqslant j\leqslant {{\rm dim}}(V_m)$ , stand for the generators of the weight $m$ space $V_m$ . Similarly, the $D_2$ (respectively $D_4$ ) action on $Y_2$ (respectively $Y_4$ ) induces a transitive action on the coordinates of $V_2$ (respectively $V_4$ ).

4.2 The toric GIT quotient

We will assume from now on that $X$ carries a cscK metric, so, by Proposition 3.1, we have

\begin{align*} {{\rm Aut}}(X)\simeq T \rtimes {{\rm Aut}}(N,\Sigma ). \end{align*}

We will set

\begin{align*} G:= {{\rm Aut}}(N,\Sigma ). \end{align*}

From Proposition 2.3, the local moduli space $\mathscr{M}_X$ of polarised cscK surfaces around $[X]\in \mathscr{M}$ is given by a neighbourhood of the origin in the GIT quotient of $V$ by $T\rtimes G$ . Assume now that $X$ is foldable. To prove Theorem1.5, it is then enough to show that the GIT (or categorical) quotient

\begin{align*} W:=V//T \end{align*}

is toric, Gorenstein and terminal. We will end this section by showing that $W$ is indeed an affine toric variety. We first need the following lemma, where by $C_p\subset G$ we mean that there is an injection of $C_p$ in $G$ .

Lemma 4.3. Let $X$ be a foldable toric surface. Then one of the following holds:

  1. (i) If $X$ is rigid, then $X\in \lbrace \mathbb{P}^2, \mathbb{P}^1\times \mathbb{P}^1 , {\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)\rbrace$ .

  2. (ii) If $X$ is not rigid and $C_2\subset G$ , $X\in \lbrace Y_2, Y_4 \rbrace$ or ${\rm X} $ is obtained from $Y_2$ or $Y_4$ by successive $C_2$ -equivariant blow-ups.

  3. (iii) If $X$ is not rigid and $C_3\subset G$ , $X=Y_3$ or ${\rm X} $ is obtained from $Y_3$ by successive $C_3$ -equivariant blow-ups.

Recall that $Y_2$ , $Y_3$ and $Y_4$ were defined in Example 4.1.

Proof. Note that, by Example 4.1, $\mathbb{P}^2$ , $\mathbb{P}^1\times \mathbb{P}^1$ and ${\rm Bl}_{p_1,p_2,p_3}(\mathbb{P}^2)$ are rigid. If $X$ is not one of those three foldable surfaces, then, from Proposition 3.13, it either belongs to $\lbrace Y_2, Y_3, Y_4 \rbrace$ or is obtained from $Y_j$ by successive $C_p$ -equivariant blow-ups, where $p=2$ for $j\in \lbrace 2,4 \rbrace$ and $p=3$ for $j=3$ . From [Reference IltenIlt11, Corollary 1.6], $H^1(Y_j,TY_j)$ injects in $H^1(X,TX)$ . Hence, from the computations of Example 4.1, we see that $X$ is not rigid, and the result follows.

Proposition 4.4. The GIT quotient $W$ inherits the structure of a normal toric affine variety.

Although the proof is straightforward, we should warn the reader that our setting is slightly different from the standard quotient construction of toric varieties (see e.g. [Reference Cox, Little, Schenck and varietiesCLS11, Chapter 5]), as the weight spaces $V_m$ may have dimension greater than $1$ , and the quotient map $\tilde \sigma \to \sigma$ (see the proof below) may send several rays to the same ray. This detailed proof will also settle the necessary notation for the following section.

Proof. If $V=0$ , then there is nothing to prove. We then assume that $X$ is not rigid. From Lemma 4.3, $X$ is obtained from $Y_j$ by successive $C_p$ -equivariant blow-ups, where $j\in \lbrace 2, 3, 4\rbrace$ , for suitable $p\in \lbrace 2,3 \rbrace$ .

Denote by $d={{\rm dim}}(V)$ and by $\tilde N\simeq \mathbb{Z}^d$ the lattice of one-parameter subgroups of $\tilde T \simeq ({\mathbb{C}}^*)^d$ acting on $V\simeq {\mathbb{C}}^d$ by multiplication on each coordinate:

\begin{align*} \forall (t,x)\in \tilde T \times V,\: t\cdot x =(t_i x_i)_{1\leqslant i \leqslant d}. \end{align*}

The $T$ -module structure of $V$ is then equivalent to the injection $T \to \tilde T$ , or the lattice monomorphism

\begin{align*} B : N \to \tilde N \end{align*}

defined by

\begin{align*}B(u)=(\langle m_i , u \rangle )_{1\leqslant i \leqslant d}\end{align*}

where the weights $m_i$ run through all the weights in the weight space decomposition of $V$ (with multiplicities when ${{\rm dim}}(V_m)\geqslant 2$ ). The fact that $B$ is injective comes again from the injection $V_j \subset V$ (cf. [Reference IltenIlt11, Corollary 1.6]) and the explicit computation of $V_j$ in Example 4.1. We then consider the quotient

\begin{align*} N':=\tilde N / N, \end{align*}

where we identify $N$ with $B(N)$ by abuse of notation. We claim that $N'$ is again a lattice, that is, that $N$ is saturated in $\tilde N$ . If $X=Y_j$ , for $j\in \lbrace 2,3,4\rbrace$ , then this can be checked directly using the description of $V_j$ in Example 4.1. If $X$ is a blow-up of $Y_j$ , then $V_j \subset V$ , so that $B$ can be written $B=(B_j, B_+)$ , where

\begin{align*}B_j : N \to \tilde N_j\end{align*}

is the injection corresponding to the weight space decomposition of $V_j$ , with

\begin{align*}\tilde N_j\simeq \mathbb{Z}^{d_j},\end{align*}

for $d_j={{\rm dim}}(V_j)$ . Then, the fact that $B_j(N)$ is saturated in $\tilde N_j\subset \tilde N$ implies that $B(N)$ is saturated in $\tilde N$ . Thus, we have a short exact sequence of lattices

(4) \begin{align} 0\longrightarrow N \stackrel {B}{\longrightarrow }\tilde N \stackrel {A}{\longrightarrow } N' \longrightarrow 0 \end{align}

where

\begin{align*}A : \tilde N \to \tilde N /N\end{align*}

denotes the quotient map.

Introduce $(\tilde e_i)_{1\leqslant i\leqslant d}$ , the basis of $V$ dual to the coordinates $\chi ^{m_i}$ of the weight spaces $V_{m_i}$ . This corresponds to a $\mathbb{Z}$ -basis of $\tilde N$ , still denoted $(\tilde e_i)_{1\leqslant i \leqslant d}$ . The cone

\begin{align*}\tilde \sigma =\sum _{i=1}^d {\mathbb{R}}_+\cdot \tilde e_i\subset \tilde N_{\mathbb{R}}\end{align*}

satisfies

\begin{align*} {\rm Spec}({\mathbb{C}}[\tilde \sigma ^\vee \cap \tilde M])=V, \end{align*}

where

\begin{align*}\tilde M={{\rm Hom}}_{\mathbb{Z}}(\tilde N, \mathbb{Z}). \end{align*}

Consider the cone of $N'_{\mathbb{R}}$

\begin{align*} \sigma = A(\tilde \sigma )= \sum _{i=1}^d {\mathbb{R}}_+\cdot A(\tilde e_i) \end{align*}

(we will still denote by $A$ and $B$ their $\mathbb{R}$ -linear extensions). Our goal is then to show that $W$ is isomorphic to the affine toric variety defined by $\sigma$ . We first need to prove that $\sigma$ has the required properties to define such a variety. The cone $\sigma$ is clearly rational, convex and polyhedral. We now prove that $\sigma$ is strictly convex, that is

\begin{align*}-\sigma \cap \sigma =\lbrace 0 \rbrace .\end{align*}

So let $v\in -\sigma \cap \sigma$ . Then there is $(u_+, u_-)\in (\tilde \sigma )^2$ such that $A (u_\pm )=\pm v$ . We deduce that

\begin{align*}u_+ + u_-\in \ker (A)\cap \tilde \sigma = {{\rm Im}} (B)\cap \tilde \sigma .\end{align*}

We will then show that

(5) \begin{align} {{\rm Im}} (B) \cap \tilde \sigma = \lbrace 0 \rbrace , \end{align}

which implies $u_+=u_-=0$ by strict convexity of $\tilde \sigma$ . Notice that it is enough to show (5) for $X=Y_j$ . Indeed, using the decomposition $B=(B_j,B_+)$ as before, if

\begin{align*}u=B(x)\in \tilde \sigma ,\end{align*}

then

\begin{align*}u_j=B_j(x)\in \tilde \sigma _j,\end{align*}

where $\tilde \sigma _j \subset (\tilde N_j)_{\mathbb{R}}$ is the cone corresponding to $V_j$ . If

(6) \begin{align} {{\rm Im}}(B_j)\cap \tilde \sigma _j =\lbrace 0 \rbrace , \end{align}

then by injectivity of $B_j$ we deduce that $x=0$ and then $u=0$ . It remains to prove (6), which follows again from the explicit description of the weights of the $T$ action on $V_j$ . For example, if $j=2$ , and if

\begin{align*}B_2(x_1,x_2)=(x_2,-x_2,-x_1+x_2,x_1-x_2) \in \tilde \sigma _2,\end{align*}

then by definition of $\tilde \sigma _2$ we deduce

\begin{align*} \left \{ \begin{array}{ccc} x_2 & \geqslant & 0, \\ - x_2 & \geqslant & 0, \\ -x_1 +x_2 & \geqslant & 0, \\ x_1 - x_2 & \geqslant & 0, \end{array} \right . \end{align*}

and thus $x_1=x_2=0$ . The cases $j\in \lbrace 3, 4 \rbrace$ are similar. We have just proved that $\sigma$ is a strongly convex rational polyhedral cone, and it thus defines an affine toric variety.

We now claim that

\begin{align*} W\simeq {\rm Spec}({\mathbb{C}}[\sigma ^\vee \cap M']) \end{align*}

with

\begin{align*}M'={{\rm Hom}}_{\mathbb{Z}}(N', \mathbb{Z}). \end{align*}

It is equivalent to show that

\begin{align*} {\mathbb{C}}[\sigma ^\vee \cap M']\simeq {\mathbb{C}}[\chi ^{m_1},\ldots ,\chi ^{m_d}]^T. \end{align*}

The latter equality follows easily from the definitions. Indeed, considering the dual sequence to (4),

(7) \begin{align} 0\longrightarrow M' \stackrel {A^*}{\longrightarrow }\tilde M \stackrel {B^*}{\longrightarrow } M \longrightarrow 0 \end{align}

for $m'\in M'$ , if $\tilde m=A^* (m')$ , then

\begin{align*} \begin{array}{ccc} \chi ^{m'} \in \sigma ^\vee & \Longleftrightarrow & \forall i\in [\![ 1 , d ]\!],\: \langle m', A(\tilde e_i) \rangle \geqslant 0\\ & & \\ & \Longleftrightarrow & \forall i\in [\![ 1 , d ]\!],\: \langle A^*( m'), \tilde e_i \rangle \geqslant 0\\ & & \\ & \Longleftrightarrow & \left \{ \begin{array}{c} B^*(\tilde m)=0 \\ \forall i\in [\![ 1 , d ]\!],\: \langle \tilde m, \tilde e_i \rangle \geqslant 0\end{array} \right . \\ & & \\ & \Longleftrightarrow & \chi ^{\tilde m}\in {\mathbb{C}}[\chi ^{m_1},\ldots ,\chi ^{m_d}]^T, \end{array} \end{align*}

where in the last equivalence we used the facts that, from the definition of the basis $(\tilde e_i)_{1\leqslant i\leqslant d}$ ,

\begin{align*} \chi ^{\tilde m}\in {\mathbb{C}}[\chi ^{m_1},\ldots ,\chi ^{m_d}] \Longleftrightarrow \forall i\in [\![ 1 , d ]\!],\: \langle \tilde m, \tilde e_i \rangle \geqslant 0, \end{align*}

and the $T$ -module structure on ${\mathbb{C}}[\chi ^{m_1},\ldots ,\chi ^{m_d}]$ reads, for any one-parameter subgroup $\lambda _u:{\mathbb{C}}^*\to T$ generated by some $u\in N$ ,

\begin{align*} \lambda _u(t)\cdot \chi ^{\tilde m}=t^{\langle \tilde m, B(u)\rangle }\;\chi ^{\tilde m}. \end{align*}

This proves the claim, and the result follows.

4.3 Singularities

We keep the notation from the previous section (see, in particular, the proof of Proposition 4.4). Our goal here is to conclude the proof of Theorem1.5, by showing that $W$ is Gorenstein and terminal. Let us first recall the definitions of these notions, and their combinatorial characterisation in the toric case (we refer the reader to [Reference Kollár, Mori, Clemens and CortiKM98, Introduction and Chapter 5] for a more general treatment of these notions).

Definition 4.5. Let $Z$ be a normal toric variety.Footnote 1 Then $Z$ is Gorenstein if the canonical divisor $K_Z$ is Cartier ( $\mathbb{Q}$ -Gorenstein if $K_Z$ is $\mathbb{Q}$ -Cartier). In that case, $Z$ has terminal singularities if there is a resolution of singularities $\pi : \tilde Z \to Z$ such that if we set

\begin{align*} K_{\tilde Z}=\pi ^*K_Z + \sum _i a_i E_i, \end{align*}

where the divisors $E_i$ are distinct and irreducible, then for all $i$ , we have $a_i \gt 0$ .

One can check that the above definition does not depend on the choice of resolution. Terminal singularities play an important role in the minimal model programme, being the singularities of minimal models. Their logarithmic version has also turned out to be very useful. Recall that a log pair is a normal variety $Z$ together with an effective $\mathbb{Q}$ -divisor $D$ with coefficients in $[0,1]\cap \mathbb{Q}$ . A log resolution for a log pair $(Z,D)$ is a resolution of singularities $\pi : \tilde Z \to Z$ such that the exceptional locus ${\rm Exc}(\pi )$ of $\pi$ is a divisor and such that $\pi ^{-1}({\rm Supp}(D))\cup {\rm Exc}(\pi )$ is a simple normal crossing divisor.

Definition 4.6. Let $(Z,D)$ be a log pair such that $K_Z+D$ is $\mathbb{Q}$ -Cartier (the pair is then called $\mathbb{Q}$ -Gorenstein). In that case, $(Z,D)$ has klt singularities if there is a log resolution $\pi : \tilde Z \to Z$ such that if we set

\begin{align*} K_{\tilde Z}=\pi ^*(K_Z+D) + \sum _i a_i E_i, \end{align*}

where the divisors $E_i$ are distinct and irreducible, then for all $i$ , we have $a_i \gt -1$ . Following [Reference Braun, Greb, Langlois and JoaquinBGLJ24], we say that a normal variety $Z$ is of klt type if there exists an effective $\mathbb{Q}$ -divisor $D$ such that $(Z,D)$ is a $\mathbb{Q}$ -Gorenstein log pair with klt singularities.

At this stage, from [Reference Braun, Greb, Langlois and JoaquinBGLJ24, Theorem 1], we know that $W$ is of klt type. Also, by [Reference Cox, Little, Schenck and varietiesCLS11, Corollary 11.4.25], if $W$ is Gorenstein, then it has log terminal singularities, meaning that the pair $(W,0)$ is klt. We will give a direct proof that it is actually Gorenstein with terminal singularities, using the following characterisation from [Reference Cox, Little, Schenck and varietiesCLS11, Proposition 11.4.12] (where we use the notation $\sigma (1)$ to denote the set of rays in a cone $\sigma$ and $u_\rho$ to denote the primitive generator of a ray $\rho$ in $\sigma (1)$ ).

Proposition 4.7. Consider $W={\rm Spec}({\mathbb{C}}[\sigma ^\vee \cap M'])$ , the affine toric variety associated to the rational strictly convex polyhedral cone $\sigma \subset N'_{\mathbb{R}}$ (with lattice $N'$ ). Then

  1. (i) $W$ is Gorenstein if and only if there exists $m\in M'$ such that for all $\rho \in \sigma (1)$ , $\langle m, u_\rho \rangle=1$ ; and

  2. (ii) in that case, $W$ has terminal singularities if and only if the only lattice points in

    \begin{align*} \Pi _\sigma := \left \{ \sum _{\rho \in \sigma (1)} \lambda _\rho u_\rho \:\vert \: \sum _{\rho \in \sigma (1)} \lambda _\rho \leqslant 1,\: 0\leqslant \lambda _\rho \leqslant 1 \right \}\subset N'_{\mathbb{R}} \end{align*}
    are given by its vertices.

From the discussion in Section 4.2, the following proposition concludes the proof of Theorem1.5.

Proposition 4.8. The affine toric variety $W$ is Gorenstein and has terminal singularities.

Proof. We will exclude the rigid cases, and, as in the proof of Proposition 4.4, assume that $X$ is obtained from $Y_j$ by successive $C_p$ -equivariant blow-ups, where $j\in \lbrace 2, 3, 4\rbrace$ , for suitable $p\in \lbrace 2,3 \rbrace$ .

We use the characterisation in Proposition 4.7, and first need to describe the rays of $\sigma$ and their primitive generators. Note that, by construction, the rays of $\sigma$ belong to the set $\lbrace {\mathbb{R}}_+\cdot A(\tilde e_i),\: 1\leqslant i\leqslant d \rbrace$ . Let $\rho _i={\mathbb{R}}_+\cdot A(\tilde e_i)$ be such a ray. We claim that $A(\tilde e_i)$ is primitive in $N'$ . The argument is similar to the one used to prove strict convexity in the proof of Proposition 4.4. Suppose, for a contradiction, that there is $a\in \mathbb{N}$ , $a\geqslant 2$ , and $\tilde e \in \tilde \sigma \cap \tilde N$ such that $A(\tilde e_i)=a A(\tilde e)$ . Note that there is $x\in N$ such that

(8) \begin{align} B(x)=(B_j(x),B_+(x))=\tilde e_i - a \tilde e. \end{align}

By injectivity of $B_j$ , if $B_j(x)=0$ , then $x=0$ hence $\tilde e_i = a\tilde e$ . This is absurd as $\tilde e_i$ is primitive. So we may assume $B_j(x)\neq 0$ . Similarly, using Equation (6), we may assume as well that $B_j(x)\notin - \tilde \sigma _j$ and $B_j(x)\notin \tilde \sigma _j$ . Hence $B_j(x)$ must have at least one positive and one negative coordinate in the basis $(\tilde e_k)_{1\leqslant k \leqslant d_j}$ . As $\tilde e \in \tilde \sigma \cap \tilde N$ , its coordinates in the basis $(\tilde e_k)_{1\leqslant k\leqslant d}$ are non-negative integers, and, as $a\geqslant 2$ , we can write

\begin{align*} a\tilde e = \sum _{k=1}^d a_k \tilde e_k \end{align*}

with $a_k=0$ or $a_k\geqslant 2$ . Then, from Equation (8), $B_j(x)$ has exactly one coordinate equal to $1$ , while its other non-zero coordinates are all less than or equal to $-2$ . A case-by-case analysis using the description of $V_j$ in Equation (3) (cf. Example 4.1) shows that this is impossible. Indeed, for $V_2$ , the map $B_2$ is

\begin{align*} B_2(x_1,x_2)=(x_2,-x_2, -x_1+x_2, x_1-x_2), \end{align*}

with $x=(x_1,x_2)\in {\mathbb{R}}^2=N_{\mathbb{R}}$ , so if one coordinate of $B_2(x)$ equals $1$ , there is another coordinate that equals $-1$ . A similar argument leads to the same conclusion for $B_4$ . For $B_3$ , we can simply use the fact that the weight spaces are two dimensional, and we have

\begin{align*} B_3(x)=(-x_2,-x_2,x_1,x_1,-x_1+x_2,-x_1+x_2), \end{align*}

so the value $1$ would appear at least twice in the coordinates of $B_3(x)$ , if it ever does.

We will then prove that there is $m' \in M'$ such that

\begin{align*} \forall i\in [\![1, d ]\!]\: ,\:\langle m', A(\tilde e_i) \rangle=1 ,\end{align*}

which implies that $W$ is Gorenstein. So let $m'\in M'$ . Set

\begin{align*}\tilde m=A^*(m')\in \ker (B^*).\end{align*}

Then,

\begin{align*} \begin{array}{ccc} \forall i\in [\![1, d ]\!]\:,\: \langle m', A(\tilde e_i) \rangle=1 & \Longleftrightarrow & \forall i\in [\![1, d ]\!]\:,\: \langle \tilde m, \tilde e_i \rangle=1 \\ & & \\ & \Longleftrightarrow & \tilde m = (1, \ldots ,1) \end{array} \end{align*}

where we used the basis $(\tilde e_i)$ to produce the coordinates of $\tilde m$ . Hence, it is equivalent to show that

\begin{align*} (1, \ldots , 1)\in \ker (B^*), \end{align*}

which, by definition of $B$ , is equivalent to

\begin{align*} \sum _{i=1}^d m_i=0\in M \end{align*}

where the weights $m_i$ describe the $T$ -action on $V$ . This is where we use the fact that $X$ is foldable. Recall that $G={{\rm Aut}}(N,\Sigma )$ . The $G$ -action on $N$ naturally induces a $G$ -action on $M$ by duality, in such a way that the duality pairing is $G$ -invariant. Explicitly, for $m\in M={{\rm Hom}}(N,\mathbb{Z})$ , and $g\in G$ , we have

\begin{align*} g\cdot m := m(g^{-1}\,\cdot ) \end{align*}

and thus

\begin{align*}\langle g\cdot m , g\cdot u \rangle =\langle m , u\rangle \end{align*}

for any $u\in N$ . By the characterisation of the weights $(m_i)$ in Equation (2), we see that this $G$ -action preserves the set of weights appearing in the decomposition $V=\bigoplus _{m\in M} V_m$ . As $G$ contains a non-trivial cyclic group, there is an element $g\in {{\rm Aut}}(N)$ of order $p$ with no fixed point but $0$ . This element generates a cyclic group that acts freely on $\lbrace m_i,\: 1\leqslant i \leqslant d \rbrace$ (freeness comes from the fact that the action is free on the whole of $M_{\mathbb{R}}$ ). Hence

\begin{align*} \sum _{i=1}^d m_i=\sum _{i'} \sum _{k=0}^{p-1} g^k\cdot m_i' \end{align*}

where we picked a single element $m_i'$ in each orbit under this action. As $g\neq {{\rm Id}}$ , we have

\begin{align*} {{\rm Id}} + g + \cdots + g^{p-1}=0 \end{align*}

so that for each orbit

\begin{align*} \sum _{k=0}^{p-1} g^k\cdot m_i'=0. \end{align*}

We then deduce the existence of the required $m'$ , and that $W$ is Gorenstein.

We proceed to the proof of the fact that $W$ has terminal singularities. Let

\begin{align*} \Pi _\sigma := \left \{ \sum _{\rho \in \sigma (1)} \lambda _\rho u_\rho \:\vert \: \sum _{\rho \in \sigma (1)} \lambda _\rho \leqslant 1,\: 0\leqslant \lambda _\rho \leqslant 1 \right \}. \end{align*}

As described above, we may fix a subset of

\begin{align*}\lbrace A(\tilde e_i),\: 1\leqslant i\leqslant d \rbrace \end{align*}

as a set of generators for the rays in $\sigma (1)$ . Let $u'\in \Pi _\sigma \cap N'$ be given by

\begin{align*} u'=\sum _{i=1}^d \lambda _i A(\tilde e_i), \end{align*}

with

\begin{align*}0\leqslant \lambda _i \leqslant 1,\end{align*}

and

\begin{align*} \lambda _1 +\cdots +\lambda _d \leqslant 1,\end{align*}

assuming $\lambda _i=0$ when $A(\tilde e_i)$ is not in our fixed chosen set of ray generators. Assume that there is $i_0$ with $\lambda _{i_0}\neq 0$ . We need to show that $\lambda _{i_0}=1$ and $\lambda _i=0$ for $i\neq i_0$ . By the assumptions on the $\lambda _i$ , it is enough to show that $\lambda _i\in \mathbb{Z}$ for all $i$ . As $u'\in N'$ , there exists $\tilde u\in \tilde N$ such that

\begin{align*} A (\tilde u )= A\left(\sum _{i=1}^d \lambda _i \tilde e_i\right), \end{align*}

and then there is $x\in N_{\mathbb{R}}$ such that

\begin{align*} \tilde u - \sum _{i=1}^d \lambda _i \tilde e_i=B(x)=(B_j(x), B_+(x)), \end{align*}

where we recall that we still denote by $A$ and $B$ their $\mathbb{R}$ -linear extensions to $\tilde N_{\mathbb{R}}$ and $N_{\mathbb{R}}$ . By injectivity of $B_j$ again, we only need to consider the case when $B_j(x)\neq 0$ and $1\leqslant i_0\leqslant d_j$ . Indeed, if $B_j(x)=0$ then by injectivity $x=0$ , and $\tilde u$ being in $\tilde N$ forces the $\lambda _i$ to be integral, whereas if $\lambda _i=0$ for all $i\leqslant d_j$ , then $B_j(x)\in \tilde N$ (recall that $B_j$ maps to the first $d_j$ -coordinates) and so by saturation of $B_j(N)$ we have $x\in N$ and thus $B(x)\in \tilde N$ which, together with $\tilde u\in \tilde N$ , implies that the $\lambda _i$ are integral. Hence, we have reduced the problem to the case when $X=Y_j$ . We will again use the explicit descriptions of the spaces $V_j$ from Example 4.1. For $V_3$ , we find the system

\begin{align*} \left \{ \begin{array}{l} \tilde u_1 - \lambda _1 = -x_2,\\ \tilde u_2 - \lambda _2 = -x_2,\\ \tilde u_3 - \lambda _3 = x_1,\\ \tilde u_4 - \lambda _4 = x_1,\\ \tilde u_5 - \lambda _5 = -x_1+x_2,\\ \tilde u_6 - \lambda _6 = -x_1+x_2, \end{array} \right . \end{align*}

where $x=(x_1,x_2)\in {\mathbb{R}}^2=N_{\mathbb{R}}$ and $\tilde u=(\tilde u_1,\ldots ,\tilde u_6)\in \mathbb{Z}^6\simeq \tilde N$ (using the basis $(\tilde e_i)_{1\leqslant i \leqslant 6}$ ), from which we obtain

\begin{align*} ( \lambda _1 + \lambda _3+\lambda _5, \lambda _1 + \lambda _3+\lambda _6,\lambda _1 + \lambda _4+\lambda _6,\lambda _1 + \lambda _4+\lambda _5)\in \mathbb{N}^4 \end{align*}

and

\begin{align*} (\lambda _2 + \lambda _3+\lambda _5, \lambda _2 + \lambda _3+\lambda _6,\lambda _2 + \lambda _4+\lambda _6,\lambda _2 + \lambda _4+\lambda _5)\in \mathbb{N}^4, \end{align*}

where we used the fact that the $\tilde u_i$ are integers and $\lambda _i\geqslant 0$ for all $i$ . Together with

(9) \begin{align} 0 \leqslant \lambda _1 + \cdots + \lambda _6 \leqslant 1, \end{align}

we deduce that actually

\begin{align*} ( \lambda _1 + \lambda _3+\lambda _5, \lambda _1 + \lambda _3+\lambda _6,\lambda _1 + \lambda _4+\lambda _6,\lambda _1 + \lambda _4+\lambda _5)\in \lbrace 0,1\rbrace^4 \end{align*}

and

\begin{align*} (\lambda _2 + \lambda _3+\lambda _5, \lambda _2 + \lambda _3+\lambda _6,\lambda _2 + \lambda _4+\lambda _6,\lambda _2 + \lambda _4+\lambda _5)\in \lbrace 0,1\rbrace^4. \end{align*}

Let us then assume that $i_0=1$ , to fix our ideas, so $\lambda _1\gt 0$ , the other cases being similar. We must then have $\lambda _1 + \lambda _3+\lambda _5=1$ , and summing with $\lambda _2 + \lambda _4+\lambda _6$ , we see with (9) that $\lambda _2+\lambda _4+$ $\lambda _6=0$ . Hence $\lambda _2=\lambda _4=\lambda _6=0$ . Then, $\lambda _1+\lambda _4+\lambda _6=\lambda _1\gt 0$ so $\lambda _1=1$ . Considering $\lambda _1 + \lambda _3+\lambda _6=1$ and $\lambda _1 + \lambda _4+\lambda _5=1$ yields $\lambda _3=\lambda _5=0$ , and the result follows in that case. The cases of $V_2$ and $V_4$ are similar, so we will only treat $V_4$ . In that case, in the coordinates $\tilde x=(\tilde x_1,\ldots ,\tilde x_4)\in {\mathbb{R}}^4\simeq \tilde N_{\mathbb{R}}$ given by $(\tilde e_1,\ldots ,\tilde e_4)$ , the map $A$ is given by

\begin{align*} A(\tilde x_1,\tilde x_2,\tilde x_3,\tilde x_4)=(\tilde x_1+\tilde x_2,\tilde x_3+\tilde x_4). \end{align*}

Hence

\begin{align*} \left \{ \begin{array}{ccc} A(\tilde e_1) & = & A(\tilde e_2), \\ A(\tilde e_3) & = & A(\tilde e_4), \end{array} \right . \end{align*}

so we can pick the ray generators $(A(\tilde e_1), A(\tilde e_3))$ for the two dimensional cone $\sigma$ . The description of $V_4$ then implies

\begin{align*} \left \{ \begin{array}{ccc} \tilde u_1 - \lambda _1 & = & x_1,\\ \tilde u_2 & = & -x_1,\\ \tilde u_3 - \lambda _3 & = & x_2,\\ \tilde u_4 & = & -x_2, \end{array} \right . \end{align*}

hence $x\in N$ and the result follows.

4.4 Examples

We first provide two examples of local moduli spaces and then an example of a quotient

\begin{align*}H^1(X,TX)//T\end{align*}

that is not $\mathbb{Q}$ -Gorenstein, for a smooth toric surface $X$ .

4.4.1 A smooth example: $Y_4$

We consider the toric variety $W_4$ which is the quotient of $V_4$ by $T$ (recall the definition of $Y_4$ in Example 4.1, and that $V_4=H^1(Y_4,TY_4)$ ). As seen in the proof of Proposition 4.8, the generators of $\sigma$ in that case can be taken to be $A(\tilde e_1)$ and $A(\tilde e_3)$ , so that $\sigma$ is isomorphic to the cone

\begin{align*} {\mathbb{R}}_+\cdot (1,0)+{\mathbb{R}}_+\cdot (0,1) \subset {\mathbb{R}}^2. \end{align*}

Then,

\begin{align*} W_4\simeq {\mathbb{C}}^2, \end{align*}

and the $G$ -action on $V_4$ (see Remark 4.2) descends to a $D_1$ -action on $W_4\simeq {\mathbb{C}}^2$ generated by a reflection

\begin{align*} (x,y)\mapsto (y,x). \end{align*}

Hence, we conclude that the local moduli space is modelled on

\begin{align*} W_4//G \simeq {\mathbb{C}}^2. \end{align*}

4.4.2 A singular example: $Y_3$

Let us now consider $W_3$ , given by the GIT quotient of $V_3$ by $T$ . We can pick isomorphisms $ N\simeq \mathbb{Z}^2,$ $ \tilde N\simeq \mathbb{Z}^6$ and

\begin{align*} N'= \tilde N/N\simeq \mathbb{Z}^4 \end{align*}

such that the map $B : N\to \tilde N$ is given by

\begin{align*} B=\left [ \begin{array}{cc} 0 &\quad -1 \\ 0 &\quad -1 \\ 1 &\quad 0 \\ 1 &\quad 0 \\ -1 &\quad 1 \\ -1 &\quad 1 \end{array} \right ] \end{align*}

and the map $A : \tilde N \to N'$ by

\begin{align*} A=\left [ \begin{array}{cccccc} -1 &\quad 1 &\quad 0 &\quad 0 &\quad 0 &\quad 0 \\ 0 &\quad 0 &\quad -1 &\quad 1 &\quad 0 &\quad 0 \\ 1 &\quad 0 &\quad 1 &\quad 0 &\quad 1 &\quad 0 \\ 1 &\quad 0 &\quad 1 &\quad 0 &\quad 0 &\quad 1 \end{array} \right ]. \end{align*}

Hence, $W_3$ is the toric variety associated to the cone

\begin{align*} \sigma = \sum _{i=1}^6 {\mathbb{R}}_+\cdot e_i'\subset {\mathbb{R}}^4 \end{align*}

where the $(e_i')_{1\leqslant i\leqslant 6}$ are given by the columns of the matrix $A$ . This toric affine variety is singular. It is not even simplicial, as the $(e_i')_{1\leqslant i\leqslant 6}$ , which are the primitive generators of the six rays in $\sigma (1)$ , do not form a $\mathbb{Q}$ -basis for $N'_{\mathbb{Q}}$ . The $G$ -action descends to a $D_3$ -action on $W_3$ . Indeed, as explained in the proof of Proposition 4.8, $G\simeq D_3$ acts on the set of weights $m_i$ that appear in the weight space decomposition of $V_3\simeq {\mathbb{C}}^6$ . This action results in permutations of the $\chi ^{m_i}$ , and thus in permutations of the elements of the basis $(\tilde e_i)_{1\leqslant i\leqslant 6}$ (we use the notations from the previous proofs of Propositions 4.4 and 4.8). Then, this $D_3$ -action descends to $N'\simeq \mathbb{Z}^4$ , by permutations of the $e_i'$ . Explicitly, we compute that the associated representation

\begin{align*} D_3\to {\rm GL}_4(\mathbb{Z}) \end{align*}

is generated by

\begin{align*} \left [ \begin{array}{cccc} 0 &\quad 0 &\quad -1 &\quad 1 \\ 1 &\quad 0 &\quad 0 &\quad 0 \\ 0 &\quad 0 &\quad 1 &\quad 0 \\ 0 &\quad 1 &\quad 1 &\quad 0 \end{array} \right ] \quad \textrm{and} \quad \left [ \begin{array}{cccc} -1 & 0 & 0 & 0 \\ 0 &\quad -1 &\quad 0 &\quad 0 \\ 1 &\quad 1 &\quad 0 &\quad 1 \\ 1 &\quad 1 &\quad 1 &\quad 0 \end{array} \right ] \end{align*}

where the first matrix represents an element of order $3$ and the second one a reflection. The final quotient $W_3//D_3$ provides a singular example of a local moduli space.

4.5 A non–Gorenstein example

Here we produce an example of a toric surface $X$ with ${{\rm Aut}}(X)\simeq T$ and $H^1(X,TX)//T$ a non $\mathbb{Q}$ -Gorenstein toric variety (see Figure 13). For this, simply blow up $Y_4$ in a single point to produce the fan within Figure 13.

Figure 13. Fan of $X$ .

As $X$ is blown up from $Y_4$ , we have ${{\rm Aut}}^0(X)\simeq T$ . We can also check that we actually have ${{\rm Aut}}(X)\simeq T$ . Moreover, the arguments in Section 4.2 go through, and we find that, if $X$ were cscK, its local moduli space would be modelled on the GIT quotient of $H^1(X,TX)$ by $T$ . A direct computation again, using the method in Section 4.1, provides the weight space decomposition

\begin{align*} H^1(X,TX)= V^1_{+e_1^*}\oplus V^2_{-e_1^*}\oplus V^1_{+e_2^*}\oplus V^1_{-e_2^*}\oplus V^1_{e_1^*-e_2^*}. \end{align*}

Given that the sum of the weights that appear in this decomposition does not vanish, the quotient $H^1(X,TX)//T$ is not $\mathbb{Q}$ -Gorenstein (see the proof of Proposition 4.7).

5. Discussion and perspectives

5.1 Relations between local moduli and Weil–Petersson metrics

In this section we will discuss the fact that the local moduli spaces we considered are related by toric fibrations. Assume that $\pi : X \to X_0$ is a $G$ -equivariant blow-up between two foldable toric surfaces. We keep the notation from the previous sections, using the subscript $0$ to refer to the spaces associated to $X_0$ . From [Reference IltenIlt11, Corollary 1.6], the corresponding space $V_0$ injects in $V$ . It is a straightforward exercise to check that we have the following commutative diagram:

\begin{align*} \begin{array}{ccccccccc} 0 &\longrightarrow & N &\stackrel {B}{\longrightarrow } &\tilde N &\stackrel {A}{\longrightarrow } & N'& \longrightarrow & 0 \\ & & \downarrow & & \downarrow & & \downarrow & & \\ 0 & \longrightarrow & N & \stackrel {B_0}{\longrightarrow } & \tilde N_0 & \stackrel {A_0}{\longrightarrow } & N_0' & \longrightarrow & 0 \end{array} \end{align*}

where the first vertical arrow is the identity and the last two are surjective. By construction, one sees that the surjective map $N'\to N_0'$ is compatible with $\sigma$ and $\sigma _0$ , and induces a toric locally trivial fibration $W\to W_0$ whose fibre is itself an affine toric variety. The whole construction is $G$ -equivariant, and provides maps between the associated local moduli spaces (up to shrinking the neighbourhoods we consider).

The cscK metric on $X$ lives in the class $\pi ^*[{\omega }_0]-{\varepsilon } E_i$ , where ${\omega }_0$ is cscK on $X_0$ , $\varepsilon$ is small and the $E_i$ stand for the exceptional divisors of the blow-up. It would be interesting to understand the behaviour of the associated Weil–Petersson metrics $({\Omega }^{WP}_{{\varepsilon }})_{0 \lt {\varepsilon } \lt {\varepsilon }_0}$ on the local moduli spaces $\mathscr{W}_{\varepsilon }\subset W/G$ as constructed in [Reference Dervan and NaumannDN20] when $\varepsilon$ goes to zero. It seems natural to expect that the volume of the fibres of the fibration $W/G\to W_0/G$ would go to zero, so that $\mathscr{W}_{\varepsilon }$ would converge in a Gromov–Hausdorff sense to $\mathscr{W}_0$ .

5.2 Higher dimensional case

Many features that hold for toric surfaces fail in a higher dimensions. First, even when it is reductive, the identity component of the automorphism group of a non-rigid toric variety will not a priori be isomorphic to the torus (see [Reference NillNil06]). Then, from dimension $3$ , toric varieties may be obstructed (see [Reference Ilten and TuroIT20]). Finally, the toric MMP (which produces the classification of toric surfaces) in general produces singular varieties. Our main ingredients for proving Theorem1.5 therefore do not generalise, a priori, in higher dimensions.

Nevertheless, it would be interesting to study what survives the new difficulties. We expect the right extension of the notion of foldable fans in higher dimension to be fans whose lattice automorphism group admits a subgroup that acts with no fixed point. This raises several questions:

  1. (i) Do all crystallographic groups arise as lattice automorphism groups of smooth complete fans?

  2. (ii) Do all foldable toric varieties admit a cscK metric?

  3. (iii) Are foldable toric varieties unobstructed?

  4. (iv) What are the singularities of the moduli space of cscK metrics around cscK foldable toric varieties?

Acknowledgements

The author would like to thank Ruadhaí Dervan and Cristiano Spotti for answering his questions on the moduli of cscK manifolds and for several helpful comments, as well as Ronan Terpereau for stimulating discussions on the topic.

Conflicts of Interest

None.

Financial Support

The author is partially supported by the grants MARGE ANR-21-CE40-0011 and BRIDGES ANR–FAPESP ANR-21-CE40-0017, and beneficiated from the France 2030 framework programme, Centre Henri Lebesgue ANR-11-LABX-0020-01.

Journal Information

Moduli is published as a joint venture of the Foundation Compositio Mathematica and the London Mathematical Society. As not-for-profit organisations, the Foundation and Society reinvest $100\%$ of any surplus generated from their publications back into mathematics through their charitable activities.

Footnotes

1 Being normal is not required in the most general definition of Gorenstein singularities. However, we will only deal with normal toric varieties, which are rational [Reference Cox, Little, Schenck and varietiesCLS11, Theorem 11.4.2], and hence Cohen–Macaulay, which is usually required to define Gorenstein singularities.

References

Apostolov, V., Auvray, H. and Sektnan, L. M., Extremal Kähler Poincaré type metrics on toric varieties, J. Geom. Anal. 31 (2021), 12231290.10.1007/s12220-019-00263-7CrossRefGoogle Scholar
Arezzo, C., Pacard, F. and Singer, M., Extremal metrics on blowups, Duke Math. J. 157 (2011), 151. MR 2783927.10.1215/00127094-2011-001CrossRefGoogle Scholar
Berman, R. J., K-polystability of Q-Fano varieties admitting Kähler-Einstein metrics, Invent. Math. 203 (2016), 9731025.10.1007/s00222-015-0607-7CrossRefGoogle Scholar
Braun, L., Greb, D., Langlois, K. and Joaquin, M., Reductive quotients of klt singularities, Invent. Math. 237 (2024), 16431682.10.1007/s00222-024-01280-2CrossRefGoogle Scholar
Calabi, E., Extremal Kähler metrics , in Collected works, eds Bourguignon, J.-P., Chen, X. and Donaldson, S. (Springer, Berlin, 2021), 549580.Google Scholar
Calabi, E., Extremal Kähler metrics. II , in Collected works, eds Bourguignon, J.-P., Chen, X. and Donaldson, S. (Springer, Berlin, 2021, 617636.Google Scholar
Chen, X., Donaldson, S. and Sun, S., Kähler-Einstein metrics on Fano manifolds, I, II, III, J. Am. Math. Soc. 28 (2015), 183278.CrossRefGoogle Scholar
Cox, D. A., Little, J. B., Schenck, H. K., varieties, T. and Grad, Toric varieties, Graduate Studies in Mathematics, vol. 124 (American Mathematical Society (AMS), Providence, RI, 2011).Google Scholar
Chen, X. and Sun, S., Calabi flow, geodesic rays, and uniqueness of constant scalar curvature Kähler metrics, Ann. Math. (2) 180 (2014), 407454.10.4007/annals.2014.180.2.1CrossRefGoogle Scholar
Demazure, M., Sous-groupes algébriques de rang maximum du groupe de Cremona. (Algebraic subgroups of maximal rank in the Cremona group), Ann. Sci. Éc. Norm. Supér. (4) 3 (1970), 507588, (French).10.24033/asens.1201CrossRefGoogle Scholar
Dervan, R.í and Naumann, P., Moduli of polarised manifolds via canonical Kähler metrics, Preprint (2020), arXiv:1810.02576.Google Scholar
Doan, A.-K., Equivariant Kuranishi family of complex compact manifolds, Manuscr. Math. 167 (2022), 793808.10.1007/s00229-021-01289-4CrossRefGoogle Scholar
Donaldson, S. K., Scalar curvature and stability of toric varieties, J. Differ. Geom. 62 (2002), 289349.Google Scholar
Fujiki, A. and Schumacher, G., The moduli space of extremal compact Kähler manifolds and generalized Weil-Petersson metrics, Publ. Res. Inst. Math. Sci. 26 (1990), 101183.10.2977/prims/1195171664CrossRefGoogle Scholar
Futaki, A., An obstruction to the existence of Einstein Kaehler metrics, Invent. Math. 73 (1983), 437443.10.1007/BF01388438CrossRefGoogle Scholar
Futaki, A. Kähler-Einstein metrics and integral invariants, Lecture Notes in Mathematics, vol. 1314 (Springer-Verlag, Berlin, 1988).10.1007/BFb0078084CrossRefGoogle Scholar
Gauduchon, P., Calabi’s extremal Kähler metrics: An elementary introduction.Google Scholar
Ilten, N. O., Deformations of smooth toric surfaces, Manuscr. Math. 134 (2011), 123137.10.1007/s00229-010-0386-9CrossRefGoogle Scholar
Inoue, E., The moduli space of Fano manifolds with Kähler-Ricci solitons, Adv. Math. 357 (2019), 65.CrossRefGoogle Scholar
Ilten, N. and Turo, C., Deformations of smooth complete toric varieties: obstructions and the cup product, Algebra Number Theory 14 (2020, 907926.10.2140/ant.2020.14.907CrossRefGoogle Scholar
Katzarkov, L., Lupercio, E., Meersseman, L. and Verjovsky, A., Quantum (non-commutative) toric geometry: foundations, Adv. Math. 391 (2021), 110, Id/No 107945.CrossRefGoogle Scholar
Kollár, János, Mori, S., Clemens, H. and Corti, A., Birational geometry of algebraic varieties. With the collaboration of C. H. Clemens and A. Corti, Cambridge Tracts in Mathematics, vol. 134 (Cambridge University Press, Cambridge, 1998).Google Scholar
Kodaira, K., Complex manifolds and deformation of complex structures. Transl. from the Japanese by Kazuo Akao. Reprint of the 1986 edition Classics in Mathematics (Springer, Berlin, 2005).Google Scholar
Kollár, J., Moduli of varieties of general type, in Handbook of moduli, vol. II (International Press, Higher Education Press, Somerville, MA, Beijing, 2013), 131157.Google Scholar
Kaloghiros, A.-S. and Petracci, A., On toric geometry and K-stability of Fano varieties, Trans. Am. Math. Soc., Ser. B 8 (2021), 548577.CrossRefGoogle Scholar
Lichnerowicz, A., Géométrie des groupes de transformations, Travaux et Recherche Mathématiques 3, (Dunod. ix, Paris, 1958), 193 (French).Google Scholar
LeBrun, C. and Simanca, S. R., Extremal Kähler metrics and complex deformation theory, Geom. Funct. Anal. 4 (1994), 298336.CrossRefGoogle Scholar
Li, C., Wang, X. and Xu, C., Quasi-projectivity of the moduli space of smooth Kähler-Einstein Fano manifolds, Ann. Sci. Éc. Norm. Supér. (4) 51 (2018), 739772.Google Scholar
Li, C., Wang, X. and Xu, C., On the proper moduli spaces of smoothable Kähler-Einstein Fano varieties, Duke Math. J. 168 (2019), 13871459.10.1215/00127094-2018-0069CrossRefGoogle Scholar
Matsushima, Y., Sur la structure du groupe d’homéomorphismes analytiques d’une certaine variété kaehlérinne, Nagoya Math. J. 11 (1957), 145150.CrossRefGoogle Scholar
Meersseman, L., The Teichmüller and Riemann moduli stacks, J. Éc. Polytech., Math 6 (2019), 879945.10.5802/jep.108CrossRefGoogle Scholar
Nill, B., Complete toric varieties with reductive automorphism group, Math. Z. 252 (2006), 767786.10.1007/s00209-005-0880-zCrossRefGoogle Scholar
Odaka, Y., The Calabi conjecture and K-stability, Int. Math. Res. Not. 2012 (2012), 22722288.Google Scholar
Odaka, Y., The GIT stability of polarized varieties via discrepancy, Ann. Math. (2) 177 (2013), 645661.10.4007/annals.2013.177.2.6CrossRefGoogle Scholar
Odaka, Y., Compact moduli spaces of Kähler-Einstein Fano varieties, Publ. Res. Inst. Math. Sci. 51 (2015), 549565.CrossRefGoogle Scholar
Odaka, Y., Spotti, C. and Sun, S., Compact moduli spaces of del Pezzo surfaces and Kähler-Einstein metrics, J. Differ. Geom. 102 (2016), 127172.Google Scholar
Petracci, A., On deformation spaces of toric singularities and on singularities of K-moduli of Fano varieties, Trans. Am. Math. Soc. 375 (2022), 56175643.10.1090/tran/8636CrossRefGoogle Scholar
Rollin, Y. and Tipler, C., Deformations of extremal toric manifolds, J. Geom. Anal. 24 (2014), 19291958.10.1007/s12220-013-9403-zCrossRefGoogle Scholar
Sektnan, L. M. and Tipler, C., Analytic k-semistability and wall-crossing, Preprint (2022), arXiv:2212.08383.Google Scholar
Sektnan, L. M. and Tipler, C., On the Futaki invariant of Fano threefolds, Ann. Univ. Ferrara 70 (2024), 811837.10.1007/s11565-024-00503-xCrossRefGoogle Scholar
Székelyhidi, G., The Kähler-Ricci flow and K-polystability, Am. J. Math. 132 (2010), 10771090.CrossRefGoogle Scholar
Székelyhidi, G., An introduction to extremal Kähler metrics, Graduate Students in Mathematics, vol. 152 (American Mathematical Society (AMS), Providence, RI, 2014).CrossRefGoogle Scholar
Székelyhidi, G., Blowing up extremal Kähler manifolds, II, Invent. Math. 200 (2015), 925977.10.1007/s00222-014-0543-yCrossRefGoogle Scholar
Tian, G., Kähler-Einstein metrics with positive scalar curvature, Invent. Math. 130 (1997), 137.10.1007/s002220050176CrossRefGoogle Scholar
Tian, G., K-stability and Kähler-Einstein metrics, Commun, Pure Appl. Math. 68 (2015), 10851156.CrossRefGoogle Scholar
Tipler, C., A note on blow-ups of toric surfaces and csc Kähler metrics, Tôhoku Math. J. (2) 66 (2014), 1529.10.2748/tmj/1396875660CrossRefGoogle Scholar
Wang, X.-J. and Zhou, B., On the existence and nonexistence of extremal metrics on toric Kähler surfaces, Adv. Math. 226 (2011), 44294455.10.1016/j.aim.2010.12.008CrossRefGoogle Scholar
Xu, C., K-stability of Fano varieties: an algebro-geometric approach, EMS Surv. Math. Sci. 8 (2021), 265354.CrossRefGoogle Scholar
Yau, S. T., Open problems in geometry, Differential geometry. Part 1: Partial differential equations on manifolds, in Proceedings of a Summer Research Institute, held at the University of California, Los Angeles, CA, USA, July 8-28, 1990 (American Mathematical Society, Providence, RI, 1993, 128.Google Scholar
Figure 0

Figure 1. Fan $\Sigma _1'$ of ${\mathbb{F}}_2$.

Figure 1

Figure 2. Fan $\Sigma _3'$ of $\mathbb{P}_2$.

Figure 2

Figure 3. Fan $\Sigma _4'$ of ${\mathbb{C}}{\mathbb{P}}^1\times {\mathbb{C}}{\mathbb{P}}^1$.

Figure 3

Figure 4. Fan $\Sigma _2'$: iterated blow-up of $\mathbb{P}^1\times \mathbb{P}^1$.

Figure 4

Figure 5. Fan $\Sigma _6'$: blow-up of ${\mathbb{P}}^2$ along its three fixed points.

Figure 5

Figure 6. Fan $\Sigma _1$: one-point blow-up of ${\mathbb{F}}_2$.

Figure 6

Figure 7. Fan $\Sigma _2$.

Figure 7

Figure 8. Fan $\Sigma _3$.

Figure 8

Figure 9. Fan $\Sigma _4$.

Figure 9

Figure 10. Fan $\Sigma _6$.

Figure 10

Figure 11. Fan of $Y_4$.

Figure 11

Figure 12. Fan of $Y_3$.

Figure 12

Figure 13. Fan of $X$.