Foldable fans, cscK surfaces and local K-moduli

Carl Tipler Univ Brest, UMR CNRS 6205, Laboratoire de Mathématiques de Bretagne Atlantique, France [email protected]

Abstract.

We study the moduli space of constant scalar curvature Kähler surfaces around the toric ones. To this aim, we introduce the class of foldable surfaces : smooth toric surfaces whose lattice automorphism group contain a non trivial cyclic subgroup. We classify such surfaces and show that they all admit a constant scalar curvature Kähler metric (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 modeled on a finite quotient of a toric affine variety with terminal singularities.

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 dimension however, the situation becomes much more delicate. Indeed, iterated blow-ups (see e.g. [25, 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 those 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 [53, 48, 13]) which predicts that the existence of a constant scalar curvature Kähler metric (cscK metric for short) 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 and Li–Wang–Xu [38, 30, 31] (see [48, 4, 9, 49] for the YTD conjecture and also [52, Part II] and reference therein for a historical survey on the related moduli space). Moreover, from Odaka’s work [36, 37], 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 [25] for a survey on 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 [12], extending Fujiki and Schumacher’s construction that dealt with the case of discrete automorphism groups [15] (see also Inoue’s work on the moduli space of Fano manifolds with Kähler–Ricci solitons [22]).

However, it seems to the author that the Fano case, and its log or pair analogues, are the only sources of examples for those moduli spaces, when non–discrete automorphism groups are allowed (see also the discussion in [12, Example 4.16] that may lead to further examples via moduli of polystable bundles). Restricting to surfaces, moduli spaces of Kähler–Einstein Del Pezzos were explicitly constructed in [39]. Our focus will then be on the moduli space of polarised cscK surfaces, and its geometry around points corresponding to the toric ones. Given that they satisfy the YTD conjecture by Donaldson’s work [13], toric surfaces draw a lot of attention (see e.g. [14, 29, 27, 50, 41, 54, 2, 55, 8, 42, 1]), and they appear as natural candidates to test the general machinery on (compare also with [23, 40, 19]). 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 $\mathrm{Aut}(N,\Sigma)$ 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 $\mathrm{Aut}(N,\Sigma)$ is isomorphic to one of the groups in the following set

\{C_{1},C_{2},C_{3},C_{4},C_{6},D_{1},D_{2},D_{3},D_{4},D_{6}\}.

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 $-\mathrm{Id}$ 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 $\mathrm{Aut}(N,\Sigma)$ 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 3.13) :

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 [3, 47]. Similar arguments were used in [50] 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. [50] for a family of examples with unbounded Picard rank, and [51] 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 [15, 12], is quite well behaved around a foldable one, as it is locally modeled on a finite quotient of a toric terminal singularity (see Section 4.3 for a definition of this class of singularities that 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 $G=\mathrm{Aut}(N,\Sigma)$ . Denote by $[X]\in\mathscr{M}$ the corresponding point in the moduli space $\mathscr{M}$ of polarised cscK surfaces. Then, there are :

(i)

A Gorenstein toric affine $G$ -variety $W$ with at worst terminal singularities,
(ii)

Open neighborhoods $\mathscr{W}$ and $\mathscr{U}$ of respectively (the image of) the torus fixed point $x\in\mathscr{W}\subset W/G$ and of $[X]\in\mathscr{U}\subset\mathscr{M}$ ,

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 [5, 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 [5, Theorem 1], together with the construction of Dervan and Naumann’s moduli space for cscK polarised manifolds [12, 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, Theorem 1.5 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 may 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 [23, 40, 19], where much wilder phenomena were observed, mainly due to the existence of obstructed deformations.

Remark 1.6.

It would be interesting to understand whether the final quotient $W/G$ in Theorem 1.5 still is terminal or not. Note however that it is a non-trivial problem to settle when a finite quotient of a terminal singularity is terminal (see [26, Remark page 161]).

Remark 1.7.

Due to 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 [43] 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 where canonical Kähler metrics, or K-polystability, turns useful in moduli problems. We would like to mention two other approaches to produce moduli spaces for (or around) toric varieties. In [34], a notion of 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, leaving the classical setting to the non-commutative one, quantum toric varieties admit moduli spaces which are orbifolds, see [24].

We proceed now to an overview of the proof of Theorem 1.5. If $(X,[\omega])$ is a cscK foldable surface, relying on [45, 12] and an observation in [41], 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 $\mathrm{Aut}(X)$ , the latter being reductive by Matsushima and Lichnerowicz’s theorem [33, 32]. Nill’s study of toric varieties with reductive automorphism groups [35] 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 $\mathrm{Aut}^{0}(X)\simeq T$ , where $T$ is the torus of $X$ . Then, by Ilten’s work [21], $H^{2}(X,TX)=0$ , so that locally, the moduli space we are after is given by (a neighborhood of the origin in) the GIT quotient $H^{1}(X,TX)//\mathrm{Aut}(X)$ , where $\mathrm{Aut}(X)\simeq T\rtimes G$ . Thanks again to [21], 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 Theorem 1.5 can be explicitly computed. The proof of Theorem 1.5 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 $6$ toric surfaces. We then use the combinatorial description of toric terminal singularities (see e.g. [10, Section 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 $\mathrm{Aut}(X)\simeq T$ 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 snc divisor, considering singular or Poincaré type cscK metrics, such as in [1].

The paper is organised as follows. In Section 2, we settle the necessary background material on cscK manifolds. Then, 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 Theorem 1.5. 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.

Notations

We will use the notations from [10]. For a toric variety $X$ , $\mathrm{Aut}(X)$ stands for its automorphism group, and $\mathrm{Aut}^{0}(X)\subset\mathrm{Aut}(X)$ the connected component of the identity. We denote $T$ the torus of $X$ , $N$ 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 $\Sigma$ (or $\Sigma_{X}$ ) and letters $\tau,\sigma$ will be used for cones in $\Sigma$ . For $0\leq j\leq\mathrm{dim}(X)$ , $\Sigma(j)$ is the set of $j$ -dimensional cones in $\Sigma$ . For $\mathbb{K}\in\{\mathbb{Q},\mathbb{R},\mathbb{C}\}$ , 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 $\Sigma$ .

Acknowledgments

The author would like to thank Ruadhaí Dervan and Cristiano Spotti for answering his questions on moduli of cscK manifolds and for several helpful comments, as well as Ronan Terpereau for stimulating discussions on the topic. The author is partially supported by the grants MARGE ANR-21-CE40-0011 and BRIDGES ANR–FAPESP ANR-21-CE40-0017.

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 on extremal metrics and their deformations, and refer the reader to [18, 46] for a more comprehensive treatment.

2.1. Extremal Kähler metrics

Extremal Kähler metrics were introduced by Calabi [6] and provide canonical representatives of Kähler classes. They are defined as the critical points of the so-called Calabi functional, that assign 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 [33, 32].

Theorem 2.1 ([33, 32]).

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 [16, 17]. Moreover, from [7], an extremal Kähler metric in the Kähler class $\Omega$ is cscK precisely when the Futaki invariant $\mathrm{Fut}_{\Omega}$ 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 ([3, 47]).

Let $(X,\Omega)$ be a smooth polarised toric surface with cscK metric $\omega\in\Omega$ . Let $Z\subset X$ be a set of torus fixed points, and $G\subset\mathrm{Aut}(X)$ a finite subgroup such that :

(i)

The set $Z$ is $G$ -invariant,
(ii)

The class $\Omega$ is $G$ -invariant,
(iii)

The adjoint action of $G$ on $\mathrm{Lie}(\mathrm{Aut}(X))$ has no fixed point but zero.

Then, there is $\varepsilon_{0}>0$ such that $(\mathrm{Bl}_{Z}(X),\Omega_{\varepsilon})$ carries a cscK metric in the class $\Omega_{\varepsilon}:=\pi^{*}\Omega-\varepsilon c_{1}(E)$ for $\varepsilon\in(0,\varepsilon_{0})$ , where $\pi:\mathrm{Bl}_{Z}(X)\to X$ stands for the blow-down map and $E=\sum_{z\in Z}E_{z}$ is the exceptional locus of $\pi$ .

Proof.

From [3, 47], there is $\varepsilon_{0}$ such that $(\mathrm{Bl}_{Z}(X),\Omega_{\varepsilon})$ admits an extremal Kähler metric for $\varepsilon\in(0,\varepsilon_{0})$ . We only need to show the vanishing of the Futaki invariant $\mathrm{Fut}_{\Omega_{\varepsilon}}$ . This will follow from its equivariance, as already used in [44, Proposition 2.1]. Denote $\tilde{X}=\mathrm{Bl}_{Z}(X)$ . By $(i)$ and $(ii)$ , $G$ lifts to a subgroup of $\mathrm{Aut}(\tilde{X})$ , such that $\Omega_{\varepsilon}$ is $G$ -invariant. Then, by equivariance of $\mathrm{Fut}_{\Omega_{\varepsilon}}\in(\mathrm{Lie}(\mathrm{Aut}(\tilde{X})))^{*}$ (see e.g. [17, Chapter 3] or [28, Section 3.1]), together with $(iii)$ , we deduce $\mathrm{Fut}_{\Omega_{\varepsilon}}=0$ , which concludes the proof. ∎

2.2. Local moduli of polarised cscK manifolds

A moduli space $\mathscr{M}$ for polarised cscK manifolds has been constructed in [12], generalising the results in [15] by taking care of non-discrete automorphisms by mean of [45, 22]. This space is Hausdorff and 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 ( $\mathscr{L}$ stands for the relative polarisation of $\mathscr{X}\to\mathscr{S}$ ), there is an induced map $\mathscr{S}_{\mathrm{red}}\to\mathscr{M}$ where $\mathscr{S}_{\mathrm{red}}$ is the reduction of $\mathscr{S}$ . 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. We then extract from [12] 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 neighborhood of $\pi(0)$ in the GIT quotient

\pi:H^{1}(X,TX)\to H^{1}(X,TX)//\mathrm{Aut}(X).

In particular, its singularities are of klt type.

By GIT quotient in this affine setting, we mean the good categorical quotient

H^{1}(X,TX)//\mathrm{Aut}(X):=\mathrm{Spec}(R^{\mathrm{Aut}(X)}),

where $R$ stands for the ring of regular functions on $H^{1}(X,TX)$ . We refer to Section 4.3 for the definition of klt type singularities.

Proof.

From [12, Section 3], $\mathscr{M}_{X}$ is given by an open neighborhood of $0$ in the analytic GIT quotient $W_{X}//G$ , where $G$ stands for the complexification of the isometry group of the cscK metric in $\Omega$ , and where $W_{X}\subset\tilde{H}^{1}$ is the smallest $G$ -invariant Stein space containing small integrable infinitesimal polarised deformations of $(X,\Omega)$ (see [45] for the construction of the space $\tilde{H}^{1}$ ). As noticed in [41, Lemma 2.10], by Bott–Steenbrink–Danilov’s vanishing of $h^{0,1}(X)$ and $h^{0,2}(X)$ (see [10, Theorem 9.3.2]), the vector space $\tilde{H}^{1}$ is $G$ -equivariantly isomorphic to $H^{1}(X,TX)$ . Also, for a toric surface, $\mathrm{Aut}(X)$ is a linear algebraic group, so that $G\simeq\mathrm{Aut}(X)$ , as the Lie algebra of $\mathrm{Aut}(X)$ has no parallel vector field (see [18, Chapter 2]). Finally, $H^{2}(X,TX)=0$ by [21, Corollary 1.5], so that any small element of $H^{1}(X,TX)$ is integrable. Altogether, we obtain the first part of Proposition 2.3. The statement about klt type singularities follows directly from [5, Theorem 1]. ∎

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 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.

\bigcup_{\sigma\in\Sigma}\sigma=N_{\mathbb{R}}.

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 $\mathrm{Aut}(N,\Sigma)$ the group of lattice automorphisms of $\Sigma$ . Recall that $\mathrm{Aut}(N,\Sigma)$ is the subgroup of $\mathrm{GL}_{\mathbb{Z}}(N)$ consisting in elements $g\in\mathrm{GL}_{\mathbb{Z}}(N)$ such that the induced isomorphisms $g\in\mathrm{GL}_{\mathbb{R}}(N_{\mathbb{R}})$ send bijectively $\Sigma$ to $\Sigma$ . By a result of Demazure [11, Proposition 11], the automorphism group $\mathrm{Aut}(X)$ of $X$ is a linear algebraic group isomorphic to

\mathrm{Aut}^{0}(X)\rtimes G,

where $G$ is a quotient of $\mathrm{Aut}(N,\Sigma)$ . We will later on be interested in the GIT quotient of $H^{1}(X,TX)$ by $\mathrm{Aut}(X)$ , 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 [35] implies :

Proposition 3.1.

Assume that $X$ admits a cscK metric. Then either

X\in\{\mathbb{P}^{2},\mathbb{P}^{1}\times\mathbb{P}^{1}\},

\mathrm{Aut}(X)\simeq T\rtimes\mathrm{Aut}(N,\Sigma).

Proof.

By Matsushima and Lichnerowicz theorem [33, 32], $\mathrm{Aut}(X)$ is reductive. The result then follows from Nill’s result [35, Theorem 1.8] together with Demazure’s structure theorem [11, Proposition 11]. ∎

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

\mathrm{Aut}(X)\simeq T\rtimes\mathrm{Aut}(N,\Sigma),

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

Lemma 3.2.

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

m\in\{1,2,3,4,6\}.

Moreover, the complex linear extension of $g$ to $N_{\mathbb{C}}$ is conjugated (in $\mathrm{GL}(N_{\mathbb{C}})\simeq\mathrm{GL}_{2}(\mathbb{C})$ ) to the following :

(1)

If $m=1$ , then $g\sim\mathrm{Id}$ ,
(2)

If $m=2$ and $\det(g)=1$ , then $g\sim-\mathrm{Id}$ ,
(3)

If $m=2$ and $\det(g)=-1$ , then $g\sim\left[\begin{array}[]{cc}1&0\\ 0&-1\end{array}\right]$ ,
(4)

If $m=3$ , then $g\sim\left[\begin{array}[]{cc}j&0\\ 0&j^{2}\end{array}\right]$ ,
(5)

If $m=4$ , then $g\sim\left[\begin{array}[]{cc}i&0\\ 0&-i\end{array}\right]$ ,
(6)

If $m=6$ , then $g\sim\left[\begin{array}[]{cc}-j&0\\ 0&-j^{2}\end{array}\right]$ ,

where we denoted by $j=e^{i\frac{2\pi}{3}}$ and by $\sim$ the equivalence relation given by conjugation, and where we used 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 and isomorphism $N\simeq\mathbb{Z}^{2}$ and identify $g$ with an element of $\mathrm{GL}_{2}(\mathbb{Z})$ . The characteristic polynomial of

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

then reads

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

On the other hand, Jordan’s normal form for the $\mathbb{C}$ -linear extension of $g$ implies that it is conjugated (in $\mathrm{M}_{2}(\mathbb{C})$ ) to

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

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

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

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

\det(g)=\alpha\beta=ad-bc\in\{-1,+1\}

as $g\in\mathrm{GL}_{2}(\mathbb{Z})$ and

\mathrm{trace}(g)=\alpha+\beta=a+d\in\mathbb{Z}.

If $\alpha$ and $\beta$ belong to $\{-1,+1\}$ , 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

\mathrm{trace}(g)=\alpha+\overline{\alpha}\in\mathbb{Z},

but also

|\alpha+\overline{\alpha}|<2

so that

\alpha+\overline{\alpha}\in\{-1,0,1\}.

Hence

\alpha\in\{\pm i,\pm j,\pm j^{2}\},

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 :

Proposition 3.3.

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

\{C_{1},C_{2},C_{3},C_{4},C_{6},D_{1},D_{2},D_{3},D_{4},D_{6}\}.

Proof.

As $\mathrm{Aut}(N,\Sigma)$ is finite, we can fix an $\mathrm{Aut}(N,\Sigma)$ -invariant euclidean metric on $N_{\mathbb{R}}$ (again considering the $\mathbb{R}$ -linear extension of $\mathrm{Aut}(N,\Sigma)$ ). Then, by the classification of finite subgroups of the group of orthogonal transformations of the plane, we deduce that $\mathrm{Aut}(N,\Sigma)$ 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 above listed groups 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 $-\mathrm{Id}$ in the first case and a reflection in the second case. Consider then the following fans in $\mathbb{Z}^{2}$ , where we denote by $\{e_{1},e_{2}\}$ 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

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

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

\Sigma_{\mathbb{F}_{n}}(1)=\{\mathbb{R}_{+}\cdot(0,-1),\>\mathbb{R}_{+}\cdot(1,0),\>\mathbb{R}_{+}\cdot(0,1),\>\mathbb{R}_{+}\cdot(-1,-n)\}.

As an example, $\mathbb{F}_{2}$ admits the fan description of Figure 1, that we will denote by $\Sigma_{1}^{\prime}$ . In this figure, and the one 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}^{\prime}

\mathbb{F}_{2}

Example 3.6.

Here are the fans $\Sigma_{\mathbb{P}^{2}}=\Sigma_{3}^{\prime}$ and $\Sigma_{\mathbb{P}^{1}\times\mathbb{P}^{1}}=\Sigma_{4}^{\prime}$ of $\mathbb{P}^{2}$ and $\mathbb{P}^{1}\times\mathbb{P}^{1}$ respectively (the numbering of the fans is motivated by Proposition 3.8 below).

Figure 2. Fan

\Sigma_{3}^{\prime}

\mathbb{P}_{2}

Figure 3. Fan

\Sigma_{4}^{\prime}

\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

\sigma=\mathbb{R}_{+}\cdot e_{i}+\mathbb{R}_{+}\cdot e_{i+1}

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 [10, Chapter 3, Section 3]). By iterated blow-ups, we obtain the following fans.

Figure 4. Fan

\Sigma_{2}^{\prime}

: iterated blow-up of

\mathbb{P}^{1}\times\mathbb{P}^{1}

The following 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 $\mathbb{C}\mathbb{P}^{2}$ .

Figure 5. Fan

\Sigma_{6}^{\prime}

: blow-up of

\mathbb{P}^{2}

along its three fixed points.

Here are further examples to complete our classification of lattice automorphism groups of rank two complete smooth fans. The first one is a single blow-up of $\mathbb{F}_{2}$ , that cancels the symmetry of $\Sigma_{1}^{\prime}=\Sigma_{\mathbb{F}_{2}}$ .

Figure 6. Fan

\Sigma_{1}

: one-point blow-up of

\mathbb{F}_{2}

The next one is obtained from $\Sigma_{2}^{\prime}$ by blowing-up in a $C_{2}$ -equivariant way (recall that $C_{2}=\langle-\mathrm{Id}\rangle$ ).

Figure 7. Fan

\Sigma_{2}

The third one 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. Fan

\Sigma_{3}

The next one is obtained by two successive blow-ups of $\mathbb{P}^{1}\times\mathbb{P}^{1}$ at four fixed points.

Figure 9. Fan

\Sigma_{4}

The last one is obtained from $\Sigma_{6}^{\prime}$ by two successive blow-ups at six fixed points.

Figure 10. Fan

\Sigma_{6}

Proposition 3.8.

Let $p\in\{1,2,3,4,6\}$ . Then we have

\mathrm{Aut}(\mathbb{Z}^{2},\Sigma_{p})\simeq C_{p}

and

\mathrm{Aut}(\mathbb{Z}^{2},\Sigma_{p}^{\prime})\simeq D_{p}.

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 goes the same for $\Sigma_{p}^{\prime}$ ). Let $g\in\mathrm{Aut}(\mathbb{Z}^{2},\Sigma_{p})$ . 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 $g$ :

g=\left[\begin{array}[]{cc}a_{\tau}&b_{\tau}\\ c_{\tau}&d_{\tau}\end{array}\right]\in\mathrm{GL}_{2}(\mathbb{Z}).

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 to select the allowed $\tau$ ’s for $g$ to be a lattice automorphism of $\Sigma_{p}$ . The result then follows easily. ∎

Remark 3.9.

One can also show similarly that for any Hirzebruch surface $\mathbb{F}_{n}$ with $n\geq 1$ , and with fan $\Sigma_{\mathbb{F}_{n}}$ , one has $\mathrm{Aut}(N,\Sigma_{\mathbb{F}_{n}})\simeq D_{1}$ .

3.2. Classification of foldable surfaces

We first recall the definition :

Definition 3.10.

We will 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 $G\subset\mathrm{Aut}(N,\Sigma)$ , then we can find a $G$ -action on $X$ , faithfull as soon as $\mathrm{Aut}(X)\simeq T\rtimes\mathrm{Aut}(N,\Sigma)$ .

Definition 3.11.

A $G$ -equivariant blow-up of $X$ is a blow-up of $X$ along a $G$ -orbit of torus fixed points. We will 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 :

\tilde{X}=X_{k}\to X_{k-1}\to\ldots\to X_{1}\to X_{0}=X.

Denote by $\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ 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 3.12.

Let $X$ be a foldable surface with fan $\Sigma$ . Assume that $\mathrm{Aut}(N,\Sigma)$ contains a subgroup isomorphic to $C_{p}$ , $p\geq 2$ .

•

If $p\in\{2,4\}$ , then $X$ is obtained from $\mathbb{P}^{1}\times\mathbb{P}^{1}$ by successive $C_{p}$ -equivariant blow-ups,
•

If $p=3$ , then $X$ is obtained from $\mathbb{P}^{2}$ by successive $C_{p}$ -equivariant blow-ups,
•

If $p=6$ , then $X$ is obtained from $\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ by successive $C_{p}$ -equivariant blow-ups.

Proof.

Through its representation

C_{p}\to\mathrm{GL}_{\mathbb{R}}(N_{\mathbb{R}}),

the subgroup $C_{p}$ of $\mathrm{Aut}(N,\Sigma)$ acts on $N_{\mathbb{R}}$ by rotations (once an invariant euclidean metric is fixed), hence has no fixed points. We deduce that the action of $C_{p}$ on $\Sigma(1)$ is free, and then $p$ divides $|\Sigma(1)|$ , the number of rays of $\Sigma$ . If $|\Sigma(1)|=p$ , then from the classification of toric surfaces (see e.g. [10, Chapter 10]) we must have

p\in\{3,4,6\}.

In that situation, if $p=3$ , then $X\simeq\mathbb{P}^{2}$ . If $p=4$ , then $X\simeq\mathbb{F}_{n}$ for some $n\in\mathbb{N}$ . By Remark 3.9, we see that $X\simeq\mathbb{P}^{1}\times\mathbb{P}^{1}$ . We will see shortly that if $p=6$ , then $X\simeq\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ .

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

u_{i}=u_{i-1}+u_{i+1}

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 $\sigma(2)$ ). By linearity, for any $g\in C_{p}$ we have

g\cdot u_{i}=g\cdot u_{i-1}+g\cdot u_{i+1},

hence we can contract 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 induction, we obtain the result. ∎

From the classification, we obtain the following proposition.

Proposition 3.13.

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 3.12, using Theorem 2.2 at each step, and the fact that $\mathbb{P}^{1}\times\mathbb{P}^{1}$ , $\mathbb{P}^{2}$ and $\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ admit cscK metrics in $C_{p}$ -equivariant classes (for $\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ , it follows again from Theorem 2.2 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 $(u_{i})_{0\leq i\leq r}$ the primitive elements of $N$ that generate the rays

\rho_{i}=\mathbb{R}_{+}\cdot u_{i}\in\Sigma(1),

labeled in the 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

The action of the torus $T$ on $X$ naturally induces a representation of $T$ on

V:=H^{1}(X,TX).

We will denote by

V=\bigoplus_{m\in M}V_{m}

the associated weight space decomposition. From [21, Corollary 1.5], for each weight $m\in M=\mathrm{Hom}_{\mathbb{Z}}(N,\mathbb{Z})$ :

(1)

\mathrm{dim}(V_{m})=\sharp\displaystyle\left\{\rho_{i}\in\Sigma(1)\>\left|\begin{array}[]{c}\langle m,u_{i}\rangle=-1\\ \langle m,u_{i\pm 1}\rangle<0\end{array}\right.\right\}.

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 $\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ are rigid, as can be checked directly using Formula (1). In those cases, $V=0$ . Then, denote by $Y_{2}$ the foldable toric surface associated to the fan $\Sigma_{2}^{\prime}$ of Figure 4. We also introduce the foldable toric surfaces $Y_{4}$ and $Y_{3}$ associated to the following fans :

Figure 11. Fan of

Y_{4}

Figure 12. Fan of

Y_{3}

Set $V_{i}=H^{1}(Y_{i},TY_{i})$ . Testing the conditions in Formula (1) for each ray generator gives

(2)

\begin{array}[]{ccc}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}^{*}}\\ &&\\ V_{3}&=&V^{2}_{-e_{2}^{*}}\oplus V^{2}_{e_{1}^{*}}\oplus V^{2}_{-e_{1}^{*}+e_{2}^{*}}\\ &&\\ V_{4}&=&V^{1}_{+e_{1}^{*}}\oplus V^{1}_{-e_{1}^{*}}\oplus V^{1}_{+e_{2}^{*}}\oplus V^{1}_{-e_{2}^{*}}\\ &&\end{array}

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

Remark 4.2.

One can check by unraveling the isomorphisms used in [21, Section 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

\{\chi_{j}^{-e_{2}^{*}},\chi_{j}^{e_{1}^{*}},\chi_{j}^{-e_{1}^{*}+e_{2}^{*}}\},\>j\in\{1,2\}

where $\chi_{j}^{m}$ , for $1\leq j\leq\mathrm{dim}(V_{m})$ , stand for generators of the weight $m$ space $V_{m}$ . Similarly, the $D_{2}$ (resp. $D_{4}$ ) action on $Y_{2}$ (resp. $Y_{4}$ ) induces a transitive action on the coordinates of $V_{2}$ (resp. $V_{4}$ ).

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

\mathrm{Aut}(X)\simeq T\rtimes\mathrm{Aut}(N,\Sigma).

We will set

G:=\mathrm{Aut}(N,\Sigma).

From Proposition 2.3, the local moduli space $\mathscr{M}_{X}$ of polarised cscK surfaces around $[X]\in\mathscr{M}$ is given by a neighborhood of the origin in the GIT quotient of $V$ by $T\rtimes G$ . Assume now that $X$ is foldable. To prove Theorem 1.5, it is then enough to show that the GIT (or categorical) quotient

W:=V//T

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.

One of the following holds :

•

If $X$ is rigid, then $X\in\{\mathbb{P}^{2},\mathbb{P}^{1}\times\mathbb{P}^{1},\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})\}$ .
•

If $X$ is not rigid and $C_{2}\subset G$ , $X\in\{Y_{2},Y_{4}\}$ or is obtained from $Y_{2}$ or $Y_{4}$ by successive $C_{2}$ -equivariant blow-ups.
•

If $X$ is not rigid and $C_{3}\subset G$ , $X=Y_{3}$ or 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 $\mathrm{Bl}_{p_{1},p_{2},p_{3}}(\mathbb{P}^{2})$ are rigid. If $X$ is not one of those three foldable surfaces, then from Proposition 3.12, either it belongs to $\{Y_{2},Y_{3},Y_{4}\}$ , or is obtained from $Y_{j}$ by successive $C_{p}$ -equivariant blow-ups, where $p=2$ for $j\in\{2,4\}$ and $p=3$ for $j=3$ . From [21, 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. [10, 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 one. This detailed proof will also settle the necessary notations 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 $j\in\{2,3,4\}$ , for suitable $p\in\{2,3\}$ .

Denote by $d=\mathrm{dim}(V)$ 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 :

\forall(t,x)\in\tilde{T}\times V,\>t\cdot x=(t_{i}x_{i})_{1\leq i\leq d}.

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

B:N\to\tilde{N}

defined by

B(u)=(\langle m_{i},u\rangle)_{1\leq i\leq d}

where the $m_{i}$ ’s run through all the weights in the weight space decomposition of $V$ (with multiplicities when $\mathrm{dim}(V_{m})\geq 2$ ). The fact that $B$ is injective comes again from the injection $V_{j}\subset V$ (cf [21, Corollary 1.6]) and the explicit computation of $V_{j}$ in Example 4.1. We then consider the quotient

N^{\prime}:=\tilde{N}/N,

where we identify $N$ with $B(N)$ by abuse of notation. We claim that $N^{\prime}$ is again a lattice, that is $N$ is saturated in $N^{\prime}$ . If $X=Y_{j}$ , for $j\in\{2,3,4\}$ , 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

B_{j}:N\to\tilde{N}_{j}

is the injection corresponding to the weight space decomposition of $V_{j}$ , with

\tilde{N}_{j}\simeq\mathbb{Z}^{d_{j}},

for $d_{j}=\mathrm{dim}(V_{j})$ . 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

(3)

0\longrightarrow N\stackrel{{\scriptstyle B}}{{\longrightarrow}}\tilde{N}\stackrel{{\scriptstyle A}}{{\longrightarrow}}N^{\prime}\longrightarrow 0

where

A:\tilde{N}\to\tilde{N}/N

denote the quotient map.

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

\tilde{\sigma}=\sum_{i=1}^{d}\mathbb{R}_{+}\cdot\tilde{e}_{i}\subset\tilde{N}_{\mathbb{R}}

satisfies

\mathrm{Spec}(\mathbb{C}[\tilde{\sigma}^{\vee}\cap\tilde{M}])=V,

where

\tilde{M}=\mathrm{Hom}_{\mathbb{Z}}(\tilde{N},\mathbb{Z}).

Consider the cone of $N^{\prime}_{\mathbb{R}}$ :

\sigma=A(\tilde{\sigma})=\sum_{j=i}^{d}\mathbb{R}_{+}\cdot A(\tilde{e}_{i})

(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 $\sigma$ clearly is rational, convex and polyhedral. We now prove that $\sigma$ is strictly convex, that is

-\sigma\cap\sigma=\{0\}.

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

u_{+}+u_{-}\in\ker(A)\cap\tilde{\sigma}=\mathrm{Im}(B)\cap\tilde{\sigma}.

We will then show that

(4)

\mathrm{Im}(B)\cap\tilde{\sigma}=\{0\},

which implies $u_{+}=u_{-}=0$ by strict convexity of $\tilde{\sigma}$ . Notice that it is enough to show (4) for $X=Y_{j}$ . Indeed, using the decomposition $B=(B_{j},B_{+})$ as before, if

u=B(x)\in\tilde{\sigma},

then

u_{j}=B_{j}(x)\in\tilde{\sigma}_{j},

where $\tilde{\sigma}_{j}\subset(\tilde{N}_{j})_{\mathbb{R}}$ is the cone corresponding to $V_{j}$ . If

(5)

\mathrm{Im}(B_{j})\cap\tilde{\sigma}_{j}=\{0\},

by injectivity of $B_{j}$ , we deduce $x=0$ and then $u=0$ . Remains to prove (5), which follows again from the explicit descrition of the weights of the $T$ action on $V_{j}$ . For example, if $j=2$ , and if

B_{2}(x_{1},x_{2})=(x_{2},-x_{2},-x_{1}+x_{2},x_{1}-x_{2})\in\tilde{\sigma}_{2},

then by definition of $\tilde{\sigma}_{2}$ we deduce

\left\{\begin{array}[]{ccc}x_{2}&\geq&0\\ -x_{2}&\geq&0\\ -x_{1}+x_{2}&\geq&0\\ x_{1}-x_{2}&\geq&0\end{array}\right.

and thus $x_{1}=x_{2}=0$ . The cases $j\in\{3,4\}$ are similar. We just proved that $\sigma$ is a strongly convex rational polyhedral cone, and thus defines an affine toric variety.

We now claim that

W\simeq\mathrm{Spec}(\mathbb{C}[\sigma^{\vee}\cap M^{\prime}])

with

M^{\prime}=\mathrm{Hom}_{\mathbb{Z}}(N^{\prime},\mathbb{Z}).

It is equivalent to show

\mathbb{C}[\sigma^{\vee}\cap M^{\prime}]\simeq\mathbb{C}[\chi^{m_{i}}]^{T}.

The latter equality follows easily from the definitions. Indeed, considering the dual sequence to (3),

(6)

0\longrightarrow M^{\prime}\stackrel{{\scriptstyle A^{*}}}{{\longrightarrow}}\tilde{M}\stackrel{{\scriptstyle B^{*}}}{{\longrightarrow}}M\longrightarrow 0

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

\begin{array}[]{ccc}\chi^{m^{\prime}}\in\sigma^{\vee}&\Longleftrightarrow&\forall i\in[\![1,d]\!],\>\langle m^{\prime},A(\tilde{e}_{i})\rangle\geq 0\\ &&\\ &\Longleftrightarrow&\forall i\in[\![1,d]\!],\>\langle A^{*}(m^{\prime}),\tilde{e}_{i}\rangle\geq 0\\ &&\\ &\Longleftrightarrow&\left\{\begin{array}[]{c}B^{*}(\tilde{m})=0\\ \forall i\in[\![1,d]\!],\>\langle\tilde{m},\tilde{e}_{i}\rangle\geq 0\end{array}\right.\\ &&\\ &\Longleftrightarrow&\chi^{\tilde{m}}\in\mathbb{C}[\chi^{m_{i}}]^{T}.\end{array}

This proves the claim, and the result follows. ∎

4.3. Singularities

We keep the notations from the previous section (see in particular the proof of Proposition 4.4). Our goal here is to conclude the proof of Theorem 1.5, by showing that $W$ is Gorenstein and terminal. Let us first recall the definitions of those notions, and their combinatorial characterisation in the toric case.

Definition 4.5.

Let $Z$ be a normal toric variety¹¹1Being normal is not required in the most general definition of Gorenstein singularities. However, we will only deal with normal toric varieties, that are rational [10, Theorem 11.4.2], hence Cohen–Macaulay, which is usually required to define Gorenstein singularities.. 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

K_{\tilde{Z}}=\pi^{*}K_{Z}+\sum_{i}a_{i}E_{i},

where the $E_{i}$ ’s are distinct irreducible divisors, then for all $i$ , we have $a_{i}>0$ .

One can check that the above definition doesn’t depend on the choice of resolution. Terminal singularities play an important role in the minimal model program, being the singularities of minimal models. Their logarithmic version turned out very useful as well. 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 $\mathrm{Exc}(\pi)$ of $\pi$ is of pure codimension $1$ and the divisor $\pi_{*}^{-1}D+\mathrm{Exc}(\pi)$ has simple normal crossing.

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

K_{\tilde{Z}}=\pi^{*}(K_{Z}+D)+\sum_{i}a_{i}E_{i},

where the $E_{i}$ ’s are distinct irreducible divisors, then for all $i$ , we have $a_{i}>-1$ . Following [5], we will 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 that stage, from [5, Theorem 1], we know that $W$ is of klt type. Also, by [10, 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 [10, Proposition 11.4.12] (where we use the notation $u_{\rho}$ for the primitive generator of a ray $\rho$ in $\sigma(1)$ ).

Proposition 4.7.

Consider $W=\mathrm{Spec}(\mathbb{C}[\sigma^{\vee}\cap M^{\prime}])$ the affine toric variety associated to the rational strictly convex polyhedral cone $\sigma$ . Then

•

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

•

In that case, $W$ has terminal singularities if and only if the only lattice points in

\Pi_{\sigma}:=\left\{\sum_{\rho\in\sigma(1)}\lambda_{\rho}u_{\rho}\>|\>\sum_{\rho\in\sigma(1)}\lambda_{\rho}\leq 1,\>0\leq\lambda_{\rho}\leq 1\right\}

are given by its vertices.

From the discussion in Section 4.2, the following proposition concludes the proof of Theorem 1.5.

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 $j\in\{2,3,4\}$ , for suitable $p\in\{2,3\}$ .

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 $\{\mathbb{R}_{+}\cdot A(\tilde{e}_{i}),\>1\leq i\leq d\}$ . Let $\rho_{i}=\mathbb{R}_{+}\cdot A(\tilde{e}_{i})$ be such a ray. We claim that $A(\tilde{e}_{i})$ is primitive in $N^{\prime}$ . The argument is similar to the one used to prove strict convexity in the proof of Proposition 4.4. Suppose by contradiction that there is $a\in\mathbb{N}$ , $a\geq 2$ , and $\tilde{e}\in\tilde{\sigma}\cap\tilde{N}$ such that $A(\tilde{e}_{i})=aA(\tilde{e})$ . Notice that there is $x\in N$ such that

(7)

B(x)=(B_{j}(x),B_{+}(x))=\tilde{e}_{i}-a\tilde{e}.

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 (5), 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 $(\tilde{e}_{k})_{1\leq k\leq d_{j}}$ . As $\tilde{e}\in\tilde{\sigma}$ , we can write

a\tilde{e}=\sum_{k=1}^{d}a_{k}\tilde{e}_{k}

with $a_{k}=0$ or $a_{k}\geq 2$ . Then, from Equation (7), $B_{j}(x)$ has exactly one coordinate equal to $1$ , while its other non-zero coordinates are all less or equal to $-2$ . A case by case analysis using the description of $V_{j}$ in Equation (2) (cf Example 4.1) shows that this is impossible (for $V_{3}$ , use the fact that the weight spaces are $2$ -dimensional).

We will then prove that there is $m^{\prime}\in M^{\prime}$ such that

\forall i\in[\![1,d]\!]\>,\>\langle m^{\prime},A(\tilde{e}_{i})\rangle=1,

which implies that $W$ is Gorenstein. So let $m^{\prime}\in M^{\prime}$ . Set

\tilde{m}=A^{*}(m^{\prime})\in\ker(B^{*}).

Then,

\begin{array}[]{ccc}\forall i\in[\![1,d]\!]\>,\>\langle m^{\prime},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}

where we used the basis $(\tilde{e}_{i})$ to produce the coordinates of $\tilde{m}$ . Hence, it is equivalent to show that

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

which by definition of $B$ is equivalent to

\sum_{i=1}^{d}m_{i}=0\in M

where $m_{i}$ ’s are the weights describing the $T$ -action on $V$ . This is where we use that $X$ is foldable. The $G$ -action on $N$ naturally induces a $G$ -action on $M$ , and on the weights $(m_{i})$ ’s. As $G$ contains a non-trivial cyclic group, there is an element $g\in\mathrm{Aut}(N)$ of order $p$ with no fixed point but $0$ . This element generates a cyclic group that acts freely on $\{m_{i},\>1\leq i\leq d\}$ . Hence

\sum_{i=1}^{d}m_{i}=\sum_{i^{\prime}}\sum_{k=0}^{p-1}g^{k}\cdot m_{i}^{\prime}

where we picked a single element $m_{i}^{\prime}$ in each orbit under this action. As $g\neq\mathrm{Id}$ , we have

\mathrm{Id}+g+\cdots+g^{p-1}=0

so that for each orbit

\sum_{k=0}^{p-1}g^{k}\cdot m_{i}^{\prime}=0.

We then deduce the existence of the required $m^{\prime}$ , and that $W$ is Gorenstein.

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

\Pi_{\sigma}:=\left\{\sum_{\rho\in\sigma(1)}\lambda_{\rho}u_{\rho}\>|\>\sum_{\rho\in\sigma(1)}\lambda_{\rho}\leq 1,\>0\leq\lambda_{\rho}\leq 1\right\}.

As described above, we may fix a subset of

\{A(\tilde{e}_{i}),\>1\leq i\leq d\}

as a set of ray generators for $\sigma$ . Let $u^{\prime}\in\Pi_{\sigma}\cap N^{\prime}$ given by

u^{\prime}=\sum_{i=1}^{d}\lambda_{i}A(\tilde{e}_{i}),

with

0\leq\lambda_{i}\leq 1,

and

\lambda_{1}+\ldots+\lambda_{d}\leq 1,

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 assumptions on the $(\lambda_{i})$ ’s, it is enough to show that $\lambda_{i}\in\mathbb{Z}$ for all $i$ . As $u^{\prime}\in N^{\prime}$ , there exists $\tilde{u}\in\tilde{N}$ such that

A(\tilde{u})=A(\sum_{i=1}^{d}\lambda_{i}\tilde{e}_{i}),

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

\tilde{u}-\sum_{i=1}^{d}\lambda_{i}\tilde{e}_{i}=B(x)=(B_{j}(x),B_{+}(x)).

By injectivity of $B_{j}$ again, we only need to consider the case when $B_{j}(x)\neq 0$ and $1\leq i_{0}\leq d_{j}$ (as if $\lambda_{i}=0$ for all $i\leq d_{j}$ then $x\in N$ and $B(x)\in\tilde{N}$ ). Hence, we reduced ourselves to study the case when $X=Y_{j}$ . We will use again the explicit descriptions of the $V_{j}$ ’s from Example 4.1. For $V_{3}$ , we find the system

\left\{\begin{array}[]{ccc}\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.

from which we obtain

(\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{Z}^{4}

and

(\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{Z}^{4}.

Together with

0\leq\lambda_{1}+\ldots+\lambda_{6}\leq 1,

we deduce that all those linear combinations of the $\lambda_{i}$ ’s equal $0$ or $1$ , with at least one that is non zero. Using $\lambda_{i}\geq 0$ , we then deduce that $\lambda_{i_{0}}=1$ and the result follows in that case. The cases $V_{2}$ and $V_{4}$ are similar, so we will only treat $V_{4}$ . In that case, one checks that

\left\{\begin{array}[]{ccc}A(\tilde{e}_{1})&=&A(\tilde{e}_{2})\\ A(\tilde{e}_{3})&=&A(\tilde{e}_{4})\end{array}\right.

so we can pick the ray generators $(A(\tilde{e}_{1}),A(\tilde{e}_{3}))$ for $\sigma(1)$ . The description of $V_{4}$ then implies

\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.

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 $H^{1}(X,TX)//T$ that is not $\mathbb{Q}$ -Gorenstein, for $X$ a smooth toric surface.

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

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

Then,

W_{4}\simeq\mathbb{C}^{2},

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

(x,y)\mapsto(y,x).

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

W_{4}//G\simeq\mathbb{C}^{2}.

4.4.2. A singular example : $Y_{3}$

Let’s consider now $W_{3}$ , 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

N^{\prime}=\tilde{N}/N\simeq\mathbb{Z}^{4}

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

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

and the map $A:\tilde{N}\to N^{\prime}$ by

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

Hence, $W_{3}$ is the toric variety associated to the cone

\sigma=\sum_{i=1}^{6}\mathbb{R}_{+}\cdot e_{i}^{\prime}\subset\mathbb{R}^{4}

where $(e_{i}^{\prime})_{1\leq i\leq 6}$ are given by the columns of the matrix $A$ . Using Sage Math, we find

W_{3}\simeq\mathrm{Spec}(\mathbb{C}[z_{0},\ldots,z_{7}]/I_{3})

with $I_{3}$ the ideal whose set of generators is given by

\begin{array}[]{ccc}I_{3}&=&\langle-z_{1}z_{4}+z_{0}z_{7},-z_{1}z_{5}+z_{0}z_{6},-z_{4}z_{6}+z_{5}z_{7},\\ &&z_{0}z_{4}-z_{3}z_{5},-z_{2}z_{4}+z_{3}z_{7},-z_{1}z_{4}+z_{3}z_{6},\\ &&z_{0}z_{2}-z_{1}z_{3},z_{2}z_{6}-z_{1}z_{7},-z_{1}z_{4}+z_{2}z_{5}\rangle.\end{array}

This toric affine variety is singular (it is not even simplicial). The $G$ -action descends to a $D_{3}$ -action on $W_{3}$ . This action is induced by a representation

D_{3}\to\mathrm{GL}_{4}(\mathbb{Z}),

and a set of generators for this action is given by

\left[\begin{array}[]{cccc}0&0&-1&1\\ 1&0&0&0\\ 0&0&1&0\\ 0&1&1&0\end{array}\right]\mathrm{and}\left[\begin{array}[]{cccc}-1&0&0&0\\ 0&-1&0&0\\ 1&1&0&1\\ 1&1&1&0\end{array}\right]

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 local moduli space.

4.5. A non–Gorenstein example

We produce here an example of a toric surface $X$ with $\mathrm{Aut}(X)\simeq T$ , and $H^{1}(X,TX)//T$ a non $\mathbb{Q}$ -Gorenstein toric variety. For this, simply blow-up $Y_{4}$ in a single point to produce the fan on the following page :

Figure 13. Fan of

X

As $X$ is blown-up from $Y_{4}$ , we have $\mathrm{Aut}^{0}(X)\simeq T$ . One can also check that we actually have $\mathrm{Aut}(X)\simeq T$ . Moreover, the arguments in Section 4.2 go through, and we find that if $X$ were cscK, then its local moduli space would be modeled 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

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}^{*}}.

Given that the sum of the weights that appear in this decomposition doesn’t 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

We will discuss in this section 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 notations from previous sections, using the subscript $0$ to refer the the spaces associated to $X_{0}$ . From [21, 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{array}[]{ccccccccc}0&\longrightarrow&N&\stackrel{{\scriptstyle B}}{{\longrightarrow}}&\tilde{N}&\stackrel{{\scriptstyle A}}{{\longrightarrow}}&N^{\prime}&\longrightarrow&0\\ &&\downarrow&&\downarrow&&\downarrow&&\\ 0&\longrightarrow&N&\stackrel{{\scriptstyle B_{0}}}{{\longrightarrow}}&\tilde{N}_{0}&\stackrel{{\scriptstyle A_{0}}}{{\longrightarrow}}&N_{0}^{\prime}&\longrightarrow&0\end{array}

where the first vertical arrow is the identity and the last two are surjective. By construction, one sees that the surjective map $N^{\prime}\to N_{0}^{\prime}$ is compatible with $\sigma$ and $\sigma_{0}$ , and induces a toric locally trivial fibration $W\to W_{0}$ whose fiber 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 neighborhoods we considered).

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 $E_{i}$ ’s stand for the exceptional divisors of the blow-up. It would be interesting to understand the behaviour of the associated Weil–Petersson metrics $(\Omega^{WP}_{\varepsilon})_{0<\varepsilon<\varepsilon_{0}}$ on the local moduli spaces $\mathscr{W}_{\varepsilon}\subset W/G$ as constructed in [12] when $\varepsilon$ goes to zero. It seems natural to expect that the volume of the fibers of the fibration $W/G\to W_{0}/G$ go to zero, so that $\mathscr{W}_{\varepsilon}$ would converge in Gromov–Hausdorff sense to $\mathscr{W}_{0}$ .

5.2. Higher dimensional case

Many features that hold for toric surfaces fail in higher dimension. 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 [35]). Then, from dimension $3$ , toric varieties may be obstructed (see [20]). Finally, the toric MMP (that produces the classification of toric surfaces) produces singular varieties in general. So our main ingredients to prove Theorem 1.5 do not generalise, a priori, in higher dimension.

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

(1)

Do all crystallographic groups arise as lattice automorphism groups of smooth complete fans?
(2)

Do all foldable toric varieties admit a cscK metric?
(3)

Are foldable toric varieties unobstructed?
(4)

What are the singularities of the moduli space of cscK metrics around cscK foldable toric varieties?

References

[1] Vestislav Apostolov, Hugues Auvray, and Lars Martin Sektnan, Extremal Kähler Poincaré type metrics on toric varieties, J. Geom. Anal. 31 (2021), no. 2, 1223–1290 (English).
[2] Vestislav Apostolov, David M. J. Calderbank, and Paul Gauduchon, Ambitoric geometry. II: Extremal toric surfaces and Einstein 4-orbifolds, Ann. Sci. Éc. Norm. Supér. (4) 48 (2015), no. 5, 1075–1112 (English).
[3] Claudio Arezzo, Frank Pacard, and Michael Singer, Extremal metrics on blowups, Duke Math. J. 157 (2011), no. 1, 1–51. MR 2783927
[4] Robert J. Berman, K-polystability of Q-Fano varieties admitting Kähler-Einstein metrics, Invent. Math. 203 (2016), no. 3, 973–1025 (English).
[5] Lukas Braun, Daniel Greb, Kevin Langlois, and Moraga Joaquin, Reductive quotients of klt singularities, (2021), Arxiv preprint 2111.02812.
[6] Eugenio Calabi, Extremal Kähler metrics, Collected works. Edited by Jean-Pierre Bourguignon, Xiuxiong Chen and Simon Donaldson, Berlin: Springer, 2021, pp. 549–580 (English).
[7] by same author, Extremal Kähler metrics. II, Collected works. Edited by Jean-Pierre Bourguignon, Xiuxiong Chen and Simon Donaldson, Berlin: Springer, 2021, pp. 617–636 (English).
[8] Bohui Chen, An-Min Li, and Li Sheng, Extremal metrics on toric surfaces, Adv. Math. 340 (2018), 363–405 (English).
[9] Xiuxiong Chen, Simon Donaldson, and Song Sun, Kähler-Einstein metrics on Fano manifolds. I, II, III, J. Am. Math. Soc. 28 (2015), no. 1, 183–278 (English).
[10] David A. Cox, John B. Little, and Henry K. Schenck, Toric varieties, Grad. Stud. Math., vol. 124, Providence, RI: American Mathematical Society (AMS), 2011 (English).
[11] Michel Demazure, 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), 507–588 (French).
[12] Ruadhaí Dervan and Philipp Naumann, Moduli of polarised kähler manifolds via canonical kähler metrics, 2020, Arxiv preprint 1810.02576.
[13] S. K. Donaldson, Scalar curvature and stability of toric varieties., J. Differ. Geom. 62 (2002), no. 2, 289–349 (English).
[14] Simon K. Donaldson, Constant scalar curvature metrics on toric surfaces, Geom. Funct. Anal. 19 (2009), no. 1, 83–136 (English).
[15] Akira Fujiki and Georg Schumacher, The moduli space of extremal compact Kähler manifolds and generalized Weil-Petersson metrics, Publ. Res. Inst. Math. Sci. 26 (1990), no. 1, 101–183 (English).
[16] Akito Futaki, An obstruction to the existence of Einstein Kaehler metrics, Invent. Math. 73 (1983), 437–443 (English).
[17] by same author, Kähler-Einstein metrics and integral invariants, Lect. Notes Math., vol. 1314, Berlin etc.: Springer-Verlag, 1988 (English).
[18] Paul Gauduchon, Calabi’s extremal kähler metrics: An elementary introduction.
[19] Liana Heuberger and Andrea Petracci, On k-moduli of fano threefolds with degree 28 and picard rank 4, 2023, ArXiv preprint 2303.12562.
[20] Nathan Ilten and Charles Turo, Deformations of smooth complete toric varieties: obstructions and the cup product, Algebra Number Theory 14 (2020), no. 4, 907–926 (English).
[21] Nathan Owen Ilten, Deformations of smooth toric surfaces, Manuscr. Math. 134 (2011), no. 1-2, 123–137 (English).
[22] Eiji Inoue, The moduli space of Fano manifolds with Kähler-Ricci solitons, Adv. Math. 357 (2019), 65 (English), Id/No 106841.
[23] Anne-Sophie Kaloghiros and Andrea Petracci, On toric geometry and K-stability of Fano varieties, Trans. Am. Math. Soc., Ser. B 8 (2021), 548–577 (English).
[24] Ludmil Katzarkov, Ernesto Lupercio, Laurent Meersseman, and Alberto Verjovsky, Quantum (non-commutative) toric geometry: foundations, Adv. Math. 391 (2021), 110 (English), Id/No 107945.
[25] János Kollár, Moduli of varieties of general type, Handbook of moduli. Volume II, Somerville, MA: International Press; Beijing: Higher Education Press, 2013, pp. 131–157 (English).
[26] János Kollár and Shigefumi Mori, Birational geometry of algebraic varieties. With the collaboration of C. H. Clemens and A. Corti, Camb. Tracts Math., vol. 134, Cambridge: Cambridge University Press, 1998 (English).
[27] Claude LeBrun, Calabi energies of extremal toric surfaces, Geometry and topology. Lectures given at the geometry and topology conferences at Harvard University, Cambridge, MA, USA, April 29–May 1, 2011 and Lehigh University, Bethlehem, PA, USA, May 25–27, 2012, Somerville, MA: International Press, 2013, pp. 195–226 (English).
[28] Claude LeBrun and Santiago R. Simanca, Extremal Kähler metrics and complex deformation theory, Geom. Funct. Anal. 4 (1994), no. 3, 298–336 (English).
[29] Eveline Legendre, Toric geometry of convex quadrilaterals, J. Symplectic Geom. 9 (2011), no. 3, 343–385 (English).
[30] Chi Li, Xiaowei Wang, and Chenyang Xu, Quasi-projectivity of the moduli space of smooth Kähler-Einstein Fano manifolds, Ann. Sci. Éc. Norm. Supér. (4) 51 (2018), no. 3, 739–772 (English).
[31] by same author, On the proper moduli spaces of smoothable Kähler-Einstein Fano varieties, Duke Math. J. 168 (2019), no. 8, 1387–1459 (English).
[32] André Lichnérowicz, Géométrie des groupes de transformations, 1958 (French).
[33] Yozô Matsushima, Sur la structure du groupe d’homéomorphismes analytiques d’une certaine variété kaehlérinne, Nagoya Mathematical Journal 11 (1957), 145–150.
[34] Laurent Meersseman, The Teichmüller and Riemann moduli stacks, J. Éc. Polytech., Math. 6 (2019), 879–945 (English).
[35] Benjamin Nill, Complete toric varieties with reductive automorphism group, Math. Z. 252 (2006), no. 4, 767–786 (English).
[36] Yuji Odaka, The Calabi conjecture and K-stability, Int. Math. Res. Not. 2012 (2012), no. 10, 2272–2288 (English).
[37] by same author, The GIT stability of polarized varieties via discrepancy, Ann. Math. (2) 177 (2013), no. 2, 645–661 (English).
[38] by same author, Compact moduli spaces of Kähler-Einstein Fano varieties, Publ. Res. Inst. Math. Sci. 51 (2015), no. 3, 549–565 (English).
[39] Yuji Odaka, Cristiano Spotti, and Song Sun, Compact moduli spaces of del Pezzo surfaces and Kähler-Einstein metrics, J. Differ. Geom. 102 (2016), no. 1, 127–172 (English).
[40] Andrea Petracci, On deformation spaces of toric singularities and on singularities of K-moduli of Fano varieties, Trans. Am. Math. Soc. 375 (2022), no. 8, 5617–5643 (English).
[41] Yann Rollin and Carl Tipler, Deformations of extremal toric manifolds, J. Geom. Anal. 24 (2014), no. 4, 1929–1958 (English).
[42] Lars Martin Sektnan, An investigation of stability on certain toric surfaces, New York J. Math. 24 (2018), 317–354 (English).
[43] Lars Martin Sektnan and Carl Tipler, Analytic k-semistability and wall-crossing, (2022), Arxiv preprint 2212.08383.
[44] by same author, On the futaki invariant of fano threefolds, (2023), Arxiv preprint 2307.02258.
[45] Gábor Székelyhidi, The Kähler-Ricci flow and K-polystability, Am. J. Math. 132 (2010), no. 4, 1077–1090 (English).
[46] by same author, An introduction to extremal Kähler metrics, Grad. Stud. Math., vol. 152, Providence, RI: American Mathematical Society (AMS), 2014 (English).
[47] by same author, Blowing up extremal Kähler manifolds. II, Invent. Math. 200 (2015), no. 3, 925–977 (English).
[48] Gang Tian, Kähler-Einstein metrics with positive scalar curvature, Invent. Math. 130 (1997), no. 1, 1–37 (English).
[49] by same author, K-stability and Kähler-Einstein metrics, Commun. Pure Appl. Math. 68 (2015), no. 7, 1085–1156 (English).
[50] Carl Tipler, A note on blow-ups of toric surfaces and csc Kähler metrics, Tôhoku Math. J. (2) 66 (2014), no. 1, 15–29 (English).
[51] Xu-jia Wang and Bin Zhou, On the existence and nonexistence of extremal metrics on toric Kähler surfaces, Adv. Math. 226 (2011), no. 5, 4429–4455 (English).
[52] Chenyang Xu, K-stability of Fano varieties: an algebro-geometric approach, EMS Surv. Math. Sci. 8 (2021), no. 1-2, 265–354 (English).
[53] Shing Tung Yau, Open problems in geometry, Differential geometry. Part 1: Partial differential equations on manifolds. Proceedings of a summer research institute, held at the University of California, Los Angeles, CA, USA, July 8-28, 1990, Providence, RI: American Mathematical Society, 1993, pp. 1–28 (English).
[54] Bin Zhou, Extremal metrics on toric manifolds – existence and K-stability, Sci. Sin., Math. 44 (2014), no. 1, 1–11 (Chinese).
[55] by same author, Variational solutions to extremal metrics on toric surfaces, Math. Z. 283 (2016), no. 3-4, 1011–1031 (English).

Foldable fans, cscK surfaces and local K-moduli

Abstract.

1. Introduction

Definition 1.1.

Proposition 1.2.

Proposition 1.3.

Remark 1.4.

Theorem 1.5.

Remark 1.6.

Remark 1.7.

Remark 1.8.

Remark 1.9.

Notations

Acknowledgments

2. Background on cscK metrics and their moduli

2.1. Extremal Kähler metrics

Theorem 2.1 ([33, 32]).

Theorem 2.2 ([3, 47]).

Proof.

2.2. Local moduli of polarised cscK manifolds

Proposition 2.3.

Proof.

3. Foldable toric surfaces

3.1. Lattice automorphisms and fans

Proposition 3.1.

Proof.

Lemma 3.2.

Proof.

Proposition 3.3.

Proof.

Remark 3.4.

Example 3.5.

Example 3.6.

Example 3.7.

Proposition 3.8.

Proof.

Remark 3.9.

3.2. Classification of foldable surfaces

Definition 3.10.

Definition 3.11.

Proposition 3.12.

Proof.

Proposition 3.13.

Proof.

4. The local cscK moduli around a foldable surface

4.1. Deformation theory of toric surfaces

Example 4.1.

Remark 4.2.

4.2. The toric GIT quotient

Lemma 4.3.

Proof.

Proposition 4.4.

Proof.

4.3. Singularities

Definition 4.5.

Definition 4.6.

Proposition 4.7.

Proposition 4.8.

Proof.

4.4. Examples

4.4.1. A smooth example : Y4Y_{4}

4.4.2. A singular example : Y3Y_{3}

4.5. A non–Gorenstein example

5. Discussion and perspectives

5.1. Relations between local moduli and Weil–Petersson metrics

5.2. Higher dimensional case

References

4.4.1. A smooth example : $Y_{4}$

4.4.2. A singular example : $Y_{3}$