Hostname: page-component-7dd5485656-wxk4p Total loading time: 0 Render date: 2025-10-27T13:26:26.943Z Has data issue: false hasContentIssue false

Singular Kähler-Einstein metrics and RCD spaces

Published online by Cambridge University Press:  23 October 2025

Gabor Szekelyhidi*
Affiliation:
Department of Mathematics, Northwestern University , Evanston 60208, USA

Abstract

We study Kähler-Einstein metrics on singular projective varieties. We show that under an approximation property with constant scalar curvature metrics, the metric completion of the smooth part is a noncollapsed RCD space, and is homeomorphic to the original variety.

Information

Type
Differential Geometry and Geometric Analysis
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

1 Introduction

A basic idea in complex geometry is to study complex manifolds using canonical Kähler metrics, of which perhaps the most important examples are Kähler-Einstein metrics. Yau’s solution of the Calabi conjecture [Reference Yau61] provides Kähler-Einstein metrics on compact Kähler manifolds with negative or zero first Chern class, while Chen-Donaldson-Sun’s solution of the Yau-Tian-Donaldson conjecture [Reference Chen, Donaldson and Sun17] shows that a Fano manifold admits a Kähler-Einstein metric if and only if it is K-stable. An example of a geometric application of such metrics is Yau’s proof [Reference Yau60] of the Miyaoka-Yau inequality.

Recently there has been increasing interest in Kähler-Einstein metrics on singular varieties. In particular Yau’s theorem was extended to the singular case by Eyssidieux-Guedj-Zeriahi [Reference Eyssidieux, Guedj and Zeriahi29], while the singular case of the Yau-Tian-Donaldson conjecture was finally resolved by Liu-Xu-Zhuang [Reference Liu, Xu and Zhuang44] after many partial results (see, for instance, [Reference Li, Tian and Wang40]). There is now a substantial literature on singular Kähler-Einstein metrics, see, for example, [Reference Berman, Boucksom, Eyssidieux, Guedj and Zeriahi4, Reference Berman3, Reference Greb, Guenancia and Kebekus30, Reference Li, Tian and Wang40, Reference Guo, Phong, Song and Sturm33].

In order to state our main results, suppose that X is an n-dimensional normal compact Kähler space. Let us recall that a singular Kähler-Einstein metric on X can be defined to be a positive current $\omega _{KE}$ that is a smooth Kähler metric on the regular set $X^{reg}$ , has locally bounded potentials, and satisfies the equation $\mathrm {Ric}(\omega _{KE}) = \lambda \omega _{KE}$ on $X^{reg}$ for a constant $\lambda \in \mathbb {R}$ . The metric $\omega _{KE}$ defines a length metric $d_{KE}$ on $X^{reg}$ , and an important problem is to understand the geometry of the metric completion $\overline {(X^{reg}, d_{KE})}$ .

In recent remarkable works, Guo-Phong-Song-Sturm [Reference Guo, Phong, Song and Sturm32, Reference Guo, Phong, Song and Sturm33] showed that this metric completion satisfies many important geometric estimates, such as bounds for their diameters, their heat kernels, as well as Sobolev inequalities, even under far more general assumptions than the Einstein condition. In particular, their results do not assume Ricci curvature bounds. It is natural to expect, however, that singular Kähler-Einstein metrics satisfy sharper results, similar to Riemannian manifolds with Ricci lower bounds. We formulate the following conjecture, which is likely folklore among experts, although we did not find it stated in the literature in this generality.

Conjecture 1. The metric completion $\overline {(X^{reg}, d_{KE})}$ , equipped with the measure $\omega _{KE}^n$ , extended trivially from $X^{reg}$ , is a noncollapsed $RCD(\lambda , 2n)$ -space, homeomorphic to X.

The notion of noncollapsed RCD-space is due to De Philippis-Gigli [Reference De Philippis and Gigli23], building on many previous works on synthetic notions of Ricci curvature lower bounds (see [Reference Sturm53, Reference Lott and Villani45, Reference Ambrosio, Gigli and Savaré1]). The conjecture is already known in several special cases, where in fact $\overline {(X^{reg}, d_{KE})}$ is shown to be a noncollapsed Ricci limit space – these are noncollapsed Gromov-Hausdorff limits of Riemannian manifolds with Ricci lower bounds, studied by Cheeger-Colding [Reference Cheeger and Colding12]. Settings where $\overline {(X^{reg}, d_{KE})}$ is a Ricci limit space are given, for example, by K-stable Fano manifolds with admissible singularities (see Li-Tian-Wang [Reference Li, Tian and Wang40], or Song [Reference Song49] for the case of crepant singularities), or smoothable K-stable Fano varieties, see Donaldson-Sun [Reference Donaldson and Sun27], Spotti [Reference Spotti51].

Our goal in this paper is to move beyond the setting of Ricci limit spaces, and to prove the conjecture in situations where it is not clear whether the singular Kähler-Einstein space $(X, \omega _{KE})$ can be approximated by smooth, or mildly singular, spaces with Ricci curvature bounded below. Instead, our approach is to use an approximation with constant scalar curvature Kähler metrics. The main approximation property that we require is the following.

Definition 2. We say that the singular Kähler-Einstein space $(X, \omega _{KE})$ can be approximated by cscK metrics, if there is a resolution $\pi : Y\to X$ , and a family of constant scalar curvature Kähler metrics $\omega _\epsilon $ on Y satisfying the following:

  1. (a) We have $\omega _\epsilon = \eta _\epsilon + \sqrt {-1}\partial \bar \partial u_\epsilon $ , where $\eta _\epsilon $ converge smoothly to $\pi ^*\eta _X$ and $\eta _\epsilon \geq \pi ^*\eta _X$ . Here $\eta _X\in [\omega _{KE}]$ is a smooth metric on X in the sense that it is locally the restriction of smooth metrics under local embeddings into Euclidean space.

  2. (b) We have the estimates

    (1) $$ \begin{align}\begin{aligned} \sup_Y |u_\epsilon| < C, \quad \frac{\omega_\epsilon^n}{\eta_Y^n}> \gamma, \quad \int_Y \left(\frac{\omega_\epsilon^n}{\eta_Y^n}\right)^p\ \eta_Y^n < C, \end{aligned} \end{align} $$
    for constants $C> 0, p > 1$ independent of $\epsilon $ , where $\eta _Y$ is a fixed Kähler metric on Y, and $\gamma $ is a non-negative continuous function on Y vanishing only along the exceptional divisor, also independent of $\epsilon $ .
  3. (c) The metrics $\omega _\epsilon $ converge locally smoothly on $\pi ^{-1}(X^{reg})$ to $\pi ^*\omega _{KE}$ .

The cscK property of the approximating metrics $\omega _\epsilon $ is used to obtain integral bounds for the Ricci and Riemannian curvatures as in Proposition 14. We expect that such an approximation is possible in all cases of interest; however, at the moment this is only known in limited settings. We have the following result.

Theorem 3. Suppose that $(X, \omega _{KE})$ is a singular Kähler-Einstein space with $\omega _{KE}\in c_1(L)$ for a line bundle over X, and such that X has discrete automorphism group. Assume that X admits a projective resolution $\pi : Y\to X$ for which the anticanonical bundle $-K_Y$ is relatively nef over X. Then $(X, \omega _{KE})$ can be approximated by cscK metrics in the sense of the definition above.

Note that recently Boucksom-Jonsson-Trusiani [Reference Boucksom, Jonsson and Trusiani6] showed the existence of cscK metrics on resolutions in this setting (and even more generally), while Pan-Tô [Reference Pan and Tô47] showed estimates for these approximating cscK metrics closely related to those in Definition 2, in a more general setting.

Our main result on Kähler-Einstein spaces that can be approximated by cscK metrics is the following.

Theorem 4. Suppose that $(X, \omega _{KE})$ can be approximated by cscK metrics, and $\omega _{KE}\in c_1(L)$ for a line bundle L on X. Then Conjecture 1 holds for $\overline {(X^{reg}, d_{KE})}$ . In addition the metric singular set of $\overline {(X^{reg}, d_{KE})}$ agrees with the complex analytic singular set $X\setminus X^{reg}$ , and it has Hausdorff codimension at least 4.

It is natural to expect that Conjecture 1 can also be extended to the setting of singular Kähler-Einstein metrics $\omega $ with cone singularities along a divisor on klt pairs $(X,D)$ . In this case one can hope to approximate $\omega $ using cscK metrics with cone singularities on a log resolution of $(X,D)$ . Some results in this direction were obtained recently by Zheng [Reference Zheng63], but we leave this extension of Theorem 4 for future work.

The RCD property implies important geometric information about the metric completion $\overline {(X^{reg}, d_{KE})}$ , such as the existence of tangent cones (see De Philippis-Gigli [Reference De Philippis and Gigli22]). Moreover, we expect that with only minor modifications the work of Donaldson-Sun [Reference Donaldson and Sun28] and Li-Wang-Xu [Reference Li, Wang and Xu41] on the tangent cones of smoothable Kähler-Einstein spaces can be extended to the setting of Theorem 4, that is, the tangent cones of $\overline {(X^{reg}, d_{KE})}$ are unique, and are determined by the algebraic structure. Knowledge of the tangent cones can then be further leveraged to obtain more refined information about the metric, such as in Hein-Sun [Reference Hein and Sun35], or [Reference Chiu and Székelyhidi18].

Using results of Honda [Reference Honda36], which rely on different equivalent characterizations of RCD spaces by Ambrosio-Gigli-Savaré [Reference Ambrosio, Gigli and Savaré1], the main estimate that we need in order to prove the RCD property in Theorem 4 is that eigenfunctions of the Laplacian are Lipschitz continuous on $(X^{reg}, d_{KE})$ . We will review Honda’s result in Section 2. In order to prove a gradient estimate for eigenfunctions, we use the approximating smooth cscK spaces $(Y, \omega _\epsilon )$ . Note that these do not satisfy uniform gradient estimates, since they do not have uniform Ricci curvature bounds from below. Instead we will prove a weaker estimate on $(Y, \omega _\epsilon )$ , expressed in terms of the heat flow – roughly speaking we obtain an estimate that is valid for times $t> t_\epsilon > 0$ along the heat flow, where $t_\epsilon \to 0$ as $\epsilon \to 0$ . These estimates can be passed to the limit as $\epsilon \to 0$ using the uniform estimates of Guo-Phong-Song-Sturm [Reference Guo, Phong, Song and Sturm32, Reference Guo, Phong, Song and Sturm33] for the heat kernels, and in the limit we obtain the required gradient bound on $(X^{reg}, \omega _{KE})$ . This is discussed in Section 3.

In Section 4 we prove that $\overline {(X^{reg}, d_{KE})}$ is homeomorphic to X, and that the metric singular set has Hausdorff codimension at least 4. Some results of this type were shown by Song [Reference Song49] and La Nave-Tian-Zhang [Reference La Nave, Tian and Zhang39], based on applying Hörmander’s $L^2$ -estimates, following Donaldson-Sun [Reference Donaldson and Sun27]. The main new difficulty in our setting is that a priori we do not have enough control of how large the set $\overline {(X^{reg}, d_{KE})} \setminus X^{reg}$ is in the metric sense. It was shown by Sturm [Reference Sturm52] (see also [Reference Song49]), that this set has capacity zero, which already plays an important role in the RCD property. For the approach of Donaldson-Sun [Reference Donaldson and Sun27] to apply, however, we need a slightly stronger effective bound that can be applied uniformly at all scales. In previous related results this type of estimate relied on showing that the metric regular set in $\overline {(X^{reg}, d_{KE})}$ coincides with $X^{reg}$ , but this is not clear in our setting since our approximating Riemannian manifolds $(Y, \omega _\epsilon )$ do not have lower Ricci bounds.

The new ingredient that we exploit is that the algebraic singular set of X is locally cut out by holomorphic (and therefore harmonic) functions. We show that these functions have finite order of vanishing along the singular set, and therefore we can control the size of their zero sets in any ball that is sufficiently close to a Euclidean ball, using a three annulus lemma argument, somewhat similarly to [Reference Chu and Jiang19]. This leads to the key result that the metric and algebraic regular sets of $\overline {(X^{reg}, d_{KE})}$ coincide. After this the proof follows by now familiar lines from Donaldson-Sun [Reference Donaldson and Sun27] and other subsequent works such as [Reference Liu and Székelyhidi42].

In Section 5 we prove Theorem 3. The proof is based primarily on Chen-Cheng’s existence theorem for cscK metrics [Reference Chen and Cheng15] together with some extensions of their estimates by Zheng [Reference Zheng62]. A similar result, in more general settings, was obtained recently by Boucksom-Jonsson-Trusiani [Reference Boucksom, Jonsson and Trusiani6] and Pan-Tô [Reference Pan and Tô47].

In Section 6, as an example application, we discuss an extension of Donaldson-Sun’s partial $C^0$ -estimate to singular Kähler-Einstein spaces with the cscK approximation property. An additional ingredient that we need is the gap result for the volume densities of (singular) Ricci flat Kähler cone metrics that arise as tangent cones, Theorem 36. This was shown very recently in the more general algebraic setting by Xu-Zhuang [Reference Xu and Zhuang59].

2 Background

2.1 Noncollapsed RCD spaces

By a metric measure space we mean a triple $(Z, d, \mathfrak {m})$ , where $(Z,d)$ is a metric space, and $\mathfrak {m}$ is a measure on Z with $\mathrm {supp}\,\mathfrak {m}=Z$ . By now there are several different, but essentially equivalent, notions of synthetic lower bounds for the Ricci curvature of $(Z,d,\mathfrak {m})$ , due to Sturm [Reference Sturm53], Lott-Villani [Reference Lott and Villani45], and Ambrosio-Gigli-Savaré [Reference Ambrosio, Gigli and Savaré1]. We will be particularly concerned with the notion of noncollapsed RCD( $K,N$ ) space introduced by De Philippis-Gigli [Reference De Philippis and Gigli23]. These should be thought of as the synthetic version of noncollapsed Gromov-Hausdorff limits of N-dimensional manifolds with Ricci curvature bounded below by K.

More specifically we will be concerned with RCD spaces that are the metric completions of smooth Riemannian manifolds. In fact the spaces that we study almost fit into the setting of almost smooth metric measure spaces, studied by Honda [Reference Honda36], except we will use the standard notion of zero capacity set instead of [Reference Honda36, Definition 3.1, 3(b)]. The results of [Reference Honda36] hold with this definition too, as we will outline below. Thus we state the following slight modification of Honda’s definition.

Definition 5. A compact metric measure space $(Z, d, \mathfrak {m})$ is an n-dimensional almost smooth metric measure space, if there is an open subset $\Omega \subset Z$ satisfying the following conditions.

  1. (1) There is a smooth n-dimensional Riemannian manifold $(M,g)$ and a homeomorphism $\phi : \Omega \to M^n$ , such that $\phi $ defines a local isometry between $(\Omega , d)$ and $(M^n, d_g)$ .

  2. (2) The restriction of the measure $\mathfrak {m}$ to $\Omega $ coincides with the n-dimensional Hausdorff measure.

  3. (3) The complement $Z\setminus \Omega $ has measure zero, that is, $\mathfrak {m}(Z\setminus \Omega )=0$ , and it has zero capacity in the following sense: there is a sequence of smooth functions $\phi _i: \Omega \to [0,1]$ with compact support in $\Omega $ such that

    1. (a) For any compact $A\subset \Omega $ we have $\phi _i|_A=1$ for sufficiently large i,

    2. (b) We have

      (2) $$ \begin{align}\begin{aligned} \lim_{i\to\infty} \int_\Omega |\nabla \phi_i|^2\, d\mathcal{H}^n = 0. \end{aligned} \end{align} $$

As a point of comparison we remark that in [Reference Honda36], the condition (b) is replaced by requiring that the $L^1$ -norm of $\Delta \phi _i$ is uniformly bounded. Note that neither of these conditions implies the other one.

In our setting we will have an n-dimensional normal projective variety X equipped with a positive current $\omega $ that is a smooth Kähler metric on $X^{reg}$ . In addition we will assume that $\omega $ has locally bounded Kähler potentials. We use $\omega $ to define a metric structure d on the smooth locus $X^{reg}$ :

(3) $$ \begin{align}\begin{aligned} d(x,y) = \inf\{ \ell(\gamma)\, |\, \gamma\text{ is a smooth curve in } X^{reg} \text{ from } x \text{ to } y\}, \end{aligned} \end{align} $$

where $\ell (\gamma )$ denotes the length of $\gamma $ with respect to $\omega $ . We define $(\hat {X}, d_{\hat {X}})$ to be the metric completion of $(X^{reg}, d)$ , and we extend the volume form $\omega ^n$ to $\hat {X}$ trivially. In this way $(\hat {X}, d_{\hat {X}}, \omega ^n)$ defines a metric measure space. The complement of $X^{reg}$ has zero capacity, by the following result, due to Sturm [Reference Sturm52] (see also Song [Reference Song49, Lemma 3.7]).

Lemma 6. There is a sequence of smooth functions $\phi _i : X^{reg}\to [0,1]$ with compact support, such that we have: for any compact $A\subset X^{reg}$ we have $\phi _i|_A = 1$ for sufficiently large i, and

(4) $$ \begin{align}\begin{aligned} \lim_{i\to\infty} \int_{X^{reg}} |\nabla\phi_i|^2\, \omega^n = 0. \end{aligned} \end{align} $$

From this we have the following.

Lemma 7. $(\hat {X}, d_{\hat {X}}, \omega ^n)$ defines a $2n$ -dimensional almost smooth measure metric space in the sense of Definition 5.

Proof. The open set $\Omega \subset \hat {X}$ is the smooth locus $X^{reg}$ viewed as a subset of its metric completion $\hat {X}$ , equipped with the Kähler metric $\omega $ . The conditions (1) and (2) in Definition 5 are automatically satisfied. The fact that $\hat {X}\setminus X^{reg}$ has capacity zero follows from the existence of good cutoff functions in Lemma 6.

In order to show that $\hat {X}$ is an $RCD$ space, we will use the characterization of $RCD$ spaces in Honda [Reference Honda36, Corollary 3.10] (see also Ambrosio-Gigli-Savaré [Reference Ambrosio, Gigli and Savaré1]). We state this Corollary here in our setting. Note that our notion of almost smooth metric measure space is slightly different from that in [Reference Honda36].

Corollary 8 (See [Reference Honda36]).

The metric completion $(\hat {X}, d_{\hat {X}}, \omega ^n)$ is an $RCD(K, 2n)$ space, where $K\in \mathbb {R}$ , if it is an almost smooth compact metric measure space associated with $X^{reg}$ in the sense of [Reference Honda36, Definition 3.1], and the following conditions hold:

  1. 1. The Sobolev to Lipschitz property holds, that is, any $f\in W^{1,2}(\hat {X})$ , with $|\nabla f|(x)\leq 1$ for $\omega ^n$ -almost every x, has a $1$ -Lipschitz representative;

  2. 2. The $L^2$ -strong compactness condition holds, that is, the inclusion $W^{1,2}(\hat {X})\hookrightarrow L^2(\hat {X})$ is a compact operator;

  3. 3. Any $W^{1,2}$ -eigenfunction of the Laplacian on $\hat {X}$ is Lipschitz;

  4. 4. $\mathrm {Ric}(\omega ) \geq K\omega $ on $X^{reg}$ .

In these conditions the Sobolev space $W^{1,2}(\hat {X})$ is defined by taking the completion of the space of compactly supported smooth functions $C_0^\infty (X^{reg})$ on the Riemannian manifold $(X^{reg}, \omega )$ in the $W^{1,2}$ -norm. By [Reference Honda36, Proposition 3.3] this space coincides with the $H^{1,2}(\hat {X}, d_{\hat {X}}, \omega ^n)$ -space defined using the Cheeger energy.

Proof. The only place where the difference between our notion of capacity zero in Definition 5 and Honda’s notion plays a role is in the proof of [Reference Honda36, Theorem 3.7] to deduce Equation (3.13), stating that the Hessian of $f_N$ is in $L^2$ (see [Reference Honda36] for the meaning of $f_N$ ). We can also deduce this by using cutoff functions that satisfy our Condition (3b) in Definition 5. To simplify the notation we will write $\Omega = X^{reg}$ . Let us recall Equation (3.12) from [Reference Honda36], which in our notation states

(5) $$ \begin{align}\begin{aligned} \frac{1}{2}\int_{\Omega} |\nabla f_N|^2 \Delta \phi_i^2\, \omega^n \geq \int_{\Omega} \phi_i^2 \Big( |\mathrm{Hess}_{f_N}|^2 + \langle \nabla\Delta f_N, \nabla f_N\rangle + K|\nabla f_N|^2\Big)\, \omega^n, \end{aligned} \end{align} $$

where $\mathrm {Ric}(\omega ) \geq K \omega $ , and we used $\phi _i^2$ as the cutoff function instead of $\phi _i$ . Note that $0 \leq \phi _i^2 \leq 1$ , and $\nabla \phi _i^2 = 2\phi _i \nabla \phi _i$ , so $\phi _i^2$ satisfies the same estimate as $\phi _i$ . In addition $f_N$ is a Lipschitz function such that $f_N, \Delta f_N \in W^{1,2}$ . We have

(6) $$ \begin{align}\begin{aligned} \int_{\Omega} |\nabla f_N|^2 \Delta \phi_i^2\, \omega^n &= -\int_{\Omega} 4|\nabla f_N| \phi_i \langle\nabla |\nabla f_N|, \nabla\phi_i\rangle\, \omega^n \\ &\leq \int_{\Omega} \left( \phi_i^2 |\mathrm{Hess}_{f_N}|^2 + 4|\nabla f_N|^2 |\nabla\phi_i|^2\right)\, \omega^n. \end{aligned} \end{align} $$

It follows using this in (5) that

(7) $$ \begin{align}\begin{aligned} \int_{\Omega} \frac{1}{2}\phi_i^2 |\mathrm{Hess}_{f_N}|^2 \, \omega^n &\leq \int_{\Omega} \Big( 2|\nabla f_N|^2 |\nabla\phi_i|^2 - \phi_i^2 \langle \nabla\Delta f_N, \nabla f_N\rangle \\&\quad \qquad - \phi_i^2 K |\nabla f_N|^2\Big) \, \omega^n. \end{aligned} \end{align} $$

Letting $i\to \infty $ and using that $|\nabla f_N| \in L^\infty $ , we obtain that

(8) $$ \begin{align}\begin{aligned} \int_{\Omega} |\mathrm{Hess}_{f_N}|^2\, \omega^n < \infty. \end{aligned} \end{align} $$

The rest of the argument is the same as in [Reference Honda36, Theorem 3.7].

Note that in our setting we have the following. In Section 3 we will show the remaining Condition (3) in the setting of Theorem 4.

Proposition 9. The metric measure space $(\hat {X}, d_{\hat {X}}, \omega ^n)$ satisfies Conditions (1), (2), and (4) in Corollary 8, for some $K\in \mathbb {R}$ .

Proof. Condition (4) is satisfied by definition. To verify Condition (1), let $f\in W^{1,2}(\hat {X})$ , such that $|\nabla f|(x) \leq 1$ for $\omega ^n$ -almost every x. On $X^{reg}$ the Sobolev to Lipschitz property holds, so we can assume that f is 1-Lipschitz on $X^{reg}$ . By the definition of the distance d, this implies that for any $x,y\in X^{reg}$ we have $|f(x) - f(y)| \leq |x-y|$ . We can then extend f uniquely to the completion $\hat {X}$ so that the same condition continues to hold. Condition (2) follows from the Sobolev inequality shown by Guo-Phong-Song-Sturm [Reference Guo, Phong, Song and Sturm33, Theorem 2.1].

Let us recall from De Philippis-Gigli [Reference De Philippis and Gigli23] that an $RCD(K,N)$ -space $(Z, d, \mathfrak {m})$ is called noncollapsed, if the N-dimensional Hausdorff measure on $(Z, d)$ agrees with $\mathfrak {m}$ . In particular, if an n-dimensional almost smooth metric measure space in Definition 5 satisfies the $RCD(K, n)$ -property, then it is automatically noncollapsed. Noncollapsed RCD spaces satisfy many of the properties enjoyed by noncollapsed Ricci limits spaces studied by Cheeger-Colding [Reference Cheeger and Colding12]. We will now recall some results that we will use.

De Philippis-Gigli [Reference De Philippis and Gigli22] showed that in a noncollapsed $RCD(K,N)$ -space $(Z, d, \mathfrak {m})$ , the tangent cones at every point $z\in Z$ are metric cones. In [Reference De Philippis and Gigli23] they then showed that Z admits a stratification

(9) $$ \begin{align}\begin{aligned} S_0\subset S_1 \subset \ldots \subset S_{N-1} \subset Z, \end{aligned} \end{align} $$

where $S_k$ denotes the set of points $z\in Z$ where no tangent cone splits off an isometric factor of $\mathbb {R}^{k+1}$ , and the strata satisfy the Hausdorff dimension estimate $\dim _{\mathcal {H}} S_k \leq k$ . Note that in contrast with the setting of noncollapsed Ricci limit spaces, it is not necessarily the case that $S_{N-1} = S_{N-2}$ , since a noncollapsed $RCD$ -space can have boundary. In our setting, however, we have the following, which is a consequence of Bruè-Naber-Semola [Reference Bruè, Naber and Semola8, Theorem 1.2].

Proposition 10. Suppose that $(Z, d, \mathfrak {m})$ is a noncollapsed $RCD(K, N)$ -space, and also an N-dimensional almost smooth metric measure space. Then $S_{N-1}=S_{N-2}$ . Moreover any iterated tangent cone $Z'$ of Z also satisfies $S_{N-1}= S_{N-2}$ .

Proof. Using the notation of [Reference Bruè, Naber and Semola8] we define $\partial Z = \overline {S_{N-1}\setminus S_{N-2}}$ to be the boundary of Z. Let $\Omega \subset Z$ denote the smooth Riemannian manifold in the definition of almost smooth metric measure space. For $z\in \Omega $ the tangent cones are all $\mathbb {R}^N$ , so $\partial Z\subset Z\setminus \Omega $ . In particular $\partial Z$ has capacity zero. Using [Reference Bruè, Naber and Semola8, Theorem 1.2(i)] this implies that we must have $\partial Z = \emptyset $ . If an iterated tangent cone $Z'$ satisfied $\partial Z' \not = \emptyset $ , then by [Reference Bruè, Naber and Semola8, Theorem 1.2(i)] we would have $\partial Z\not =\emptyset $ , which is a contradiction as above.

We will be working with harmonic functions on RCD spaces, so we review some basic results. Let us suppose that $(Z, d, \mathfrak {m})$ is a noncollapsed $RCD(K,N)$ -space that is also an N-dimensional almost smooth metric measure space. A function $f : U \to \mathbb {R}$ on an open set $U\subset Z$ is defined to be harmonic if $f\in W^{1,2}_{loc}(U)$ , and for any Lipschitz function $\psi :U\to \mathbb {R}$ with compact support we have

(10) $$ \begin{align}\begin{aligned} \int_U \nabla f\cdot \nabla \psi\, d\mathfrak{m} = 0. \end{aligned} \end{align} $$

Note that in our setting the integration can be taken over $U\cap \Omega $ , where $\Omega \subset Z$ is the dense open set in Definition 5 since $Z\setminus \Omega $ has measure zero. We will use the following result several times.

Lemma 11. Let $u: U \to \mathbb {R}$ for an open set $U\subset Z$ , such that $u\in L^\infty (U)$ . Suppose that $\Delta u = 0$ on $U\cap \Omega $ , using the smooth Riemannian structure on $\Omega $ . Then u is harmonic on U.

Proof. Let $\phi _i$ be functions as in Condition (3) of Definition 5, and $\psi $ a Lipschitz function with compact support in U. We have

(11) $$ \begin{align}\begin{aligned} \int_U \psi^2 \phi_i^2 |\nabla u|^2\, d\mathfrak{m} &= -2\int_U \psi^2 \phi_i u \nabla\phi_i\cdot \nabla u\, d\mathfrak{m} - 2\int_U \phi_i^2 \psi u \nabla u\cdot \nabla\psi\, d\mathfrak{m} \\ &\leq \frac{1}{2}\int_U \psi^2 \phi_i^2 |\nabla u|^2\, d\mathfrak{m} + 4\int_U \psi^2 u^2 |\nabla\phi_i|^2\, d\mathfrak{m} + C_\psi \int_U u^2\, d\mathfrak{m}, \end{aligned} \end{align} $$

where $C_\psi $ depends on $\sup _{U\cap \Omega } |\nabla \psi |$ . Letting $i\to \infty $ , we obtain that $u\in W^{1,2}_{loc}(U)$ .

At the same time we have

(12) $$ \begin{align}\begin{aligned} \int_U \phi_i^2 \nabla u\cdot \nabla\psi\, d\mathfrak{m} &= -2\int_U \phi_i \psi \nabla\phi_i\cdot \nabla u \, d\mathfrak{m} \\ &\leq \int_U |\nabla\phi_i|^2\, d\mathfrak{m} + \int_{\mathrm{supp}(\nabla\phi_i)} \psi^2 |\nabla u|^2\,d\mathfrak{m}.\end{aligned} \end{align} $$

Letting $i\to \infty $ we get $\int _U\nabla u\cdot \nabla \psi = 0$ , so u is harmonic on U.

We will also need the following gradient estimate, generalizing Cheng-Yau’s gradient estimate.

Proposition 12 (Jiang [Reference Jiang37], Theorem 1.1).

Let u be a harmonic function on a ball $B(p, 2R)$ in an $RCD(N, K)$ -space. There is a constant $C=C(R,N,K)$ such that

(13)

Note that a similar estimate holds for solutions of $\Delta u = c$ on $U\subset Z$ for a constant c, by considering $u - ct^2/2$ on the space $U\times \mathbb {R}_t$ .

3 The RCD property of singular Kähler-Einstein spaces

The main result in this section will be that the completion of the Kähler-Einstein metric on $X^{reg}$ in Theorem 4 defines a noncollapsed RCD space. We will first need some estimates for the cscK approximations of $(X,\omega _{KE})$ .

3.1 Constant scalar curvature approximations

Let $(X, \omega _{KE})$ be a singular Kähler-Einstein space, where $\mathrm {Ric}_{\omega _{KE}} = \lambda \omega _{KE}$ . Suppose that $(X, \omega _{KE})$ can be approximated by cscK metrics as in Definition 2. In particular there is a resolution Y of X, that admits a family of cscK metrics $\omega _\epsilon $ in suitable Kähler classes $[\eta _\epsilon ]$ , such that the $\eta _\epsilon $ converge to $\pi ^*\eta _X$ . Here $\eta _X$ is a smooth metric on X in the sense that it is the restriction of a smooth metric under local embeddings into $\mathbb {C}^N$ .

We will need the following, which is immediate from the work of Guo-Phong-Song-Sturm [Reference Guo, Phong, Song and Sturm33, Theorem 2.2].

Theorem 13. Let $H(x,y,t)$ denote the heat kernel on $(Y, \omega _\epsilon )$ . There is a continuous function $\bar {H}:(0,2] \to \mathbb {R}$ , depending on $(X,\omega _{KE})$ , but independent of $\epsilon $ , such that we have the upper bound

(14) $$ \begin{align}\begin{aligned} H(x,y,t) \leq \bar{H}(t), \quad \text{ for } x,y\in Y \text{ and } t\in (0,2]. \end{aligned} \end{align} $$

Note that $\bar {H}(t)\to \infty $ as $t\to 0$ .

In addition the constant scalar curvature metrics $\omega _\epsilon $ satisfy the following integral bounds for their Ricci curvatures. We will use these integral bounds as a replacement for having lower bounds for the Ricci curvature, when we approximate $\omega _{KE}$ with $\omega _\epsilon $ .

Proposition 14. Let us define $\widetilde {\mathrm {Ric}}_\omega = \mathrm {Ric}_\omega - \lambda \omega $ . We have the following estimates:

(15) $$ \begin{align}\begin{aligned} \lim_{\epsilon\to 0} \int_{Y} |\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2 + \frac{|\nabla\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2}{(1 + |\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2)^{1/2}} + |\Delta(1+|\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2)^{1/2}|\, \,\omega_\epsilon^n = 0, \end{aligned} \end{align} $$

and

(16) $$ \begin{align}\begin{aligned}\int_Y |\mathrm{Rm}_{\omega_\epsilon}|^2\, \omega_\epsilon^n < C, \end{aligned} \end{align} $$

for C independent of $\epsilon $ .

Proof. First recall the well-known result of Calabi [Reference Calabi10] relating the $L^2$ -norms of the scalar curvature, the Ricci and Riemannian curvature tensors of a Kähler metric. Let us denote by $R, \mathrm {Ric}, \mathrm {Rm}$ the scalar curvature, the Ricci form and the Riemannian curvature tensor. Since $R_{\omega _\epsilon }$ is constant, we have

(17) $$ \begin{align}\begin{aligned} R_{\omega_\epsilon} = \frac{ 2n \pi c_1(Y) \cup [\omega_\epsilon]^{n-1}}{[\omega_\epsilon]^n}. \end{aligned} \end{align} $$

Note that we have

(18) $$ \begin{align}\begin{aligned} \lim_{\epsilon\to 0} \frac{ 2n \pi c_1(Y) \cup [\omega_\epsilon]^{n-1}}{[\omega_\epsilon]^n} = \frac{2n\pi c_1(X) \cup [\omega_{KE}]^{n-1}}{[\omega_{KE}]^n} = n\lambda, \end{aligned} \end{align} $$

since $[\omega _{KE}]^{n-1}$ vanishes when paired with the exceptional divisor of the map $Y\to X$ . In addition

(19) $$ \begin{align}\begin{aligned} \int_{Y} |\mathrm{Ric}_{\omega_\epsilon}|^2\, \omega_\epsilon^n &= R_{\omega_\epsilon}^2 [\omega_\epsilon]^n - 4n(n-1)\pi^2 c_1(Y)^2\cup [\omega_\epsilon]^{n-2}, \\ \int_Y (|\mathrm{Ric}_{\omega_\epsilon}|^2 - |\mathrm{Rm}_{\omega_\epsilon}|^2)\, \omega_\epsilon^n &= n(n-1)\big( 4\pi^2 c_1(Y)^2 - 8\pi^2 c_2(Y)\big) \cup [\omega_\epsilon]^{n-2}. \end{aligned} \end{align} $$

Since the cohomology classes $[\omega _\epsilon ]=[\eta _\epsilon ]$ are uniformly bounded, and in addition $[\omega _\epsilon ]^n \geq [\eta _X]^n> 0$ , it follows that $R_{\omega _\epsilon }$ , and the $L^2$ norms of $|\mathrm {Ric}_{\omega _\epsilon }|, |\mathrm {Rm}_{\omega _\epsilon }|$ are all uniformly bounded, independently of $\epsilon $ .

To see the first claim in the Proposition, note that

(20) $$ \begin{align}\begin{aligned} \int_Y |\mathrm{Ric}_{\omega_\epsilon} - \lambda \omega_\epsilon|^2\, \omega_\epsilon^n = (R_{\omega_\epsilon} - n\lambda)^2 [\omega_\epsilon]^n - n(n-1) \big( 2\pi c_1(Y) - \lambda[\omega_\epsilon]\big)^2\cup [\omega_\epsilon]^{n-2}. \end{aligned} \end{align} $$

As $\epsilon \to 0$ , this converges to zero by (18) and the fact that $2\pi c_1(X) = \lambda [\omega _{KE}]$ .

To estimate $\nabla \widetilde {\mathrm {Ric}}_{\omega _\epsilon }$ and $\Delta \widetilde {\mathrm {Ric}}_{\omega _\epsilon }$ note that we have the following equation satisfied by any constant scalar curvature metric:

(21) $$ \begin{align}\begin{aligned} \Delta |\widetilde{\mathrm{Ric}}|^2 &= \nabla_k\nabla_{\bar k} (\widetilde{\mathrm{Ric}}_{p\bar q} \widetilde{\mathrm{Ric}}_{q\bar p}) \\ &= 2|\nabla_k \widetilde{\mathrm{Ric}}_{p\bar q}|^2 + \mathrm{Rm}\ast \widetilde{\mathrm{Ric}}\ast \widetilde{\mathrm{Ric}}, \end{aligned} \end{align} $$

where $\ast $ denotes a tensorial contraction. It follows that

(22) $$ \begin{align}\begin{aligned} \Delta (1 + |\widetilde{\mathrm{Ric}}|^2)^{1/2} &= (1+|\widetilde{\mathrm{Ric}}|^2)^{-1/2}\left( |\nabla\widetilde{\mathrm{Ric}}|^2 - |\nabla|\widetilde{\mathrm{Ric}}||^2 + \mathrm{Rm}\ast\widetilde{\mathrm{Ric}}\ast\widetilde{\mathrm{Ric}}\right) \\&\quad + \frac{|\nabla |\widetilde{\mathrm{Ric}}||^2}{(1 + |\widetilde{\mathrm{Ric}}|^2)^{3/2}}. \end{aligned} \end{align} $$

For a constant scalar curvature Kähler metric the form $\widetilde {\mathrm {Ric}}$ is harmonic, so we have the following refined Kato inequality (see Branson [Reference Branson7], Calderbank-Gauduchon-Herzlich [Reference Calderbank, Gauduchon and Herzlich11], or Cibotaru-Zhu [Reference Cibotaru and Zhu20, Theorem 3.8]):

(23) $$ \begin{align}\begin{aligned} |\nabla|\widetilde{\mathrm{Ric}}||^2 \leq \alpha_n |\nabla \widetilde{\mathrm{Ric}}|^2 \end{aligned} \end{align} $$

for a dimensional constant $\alpha _n < 1$ . It follows from (22) that

(24) $$ \begin{align}\begin{aligned} \Delta (1 + |\widetilde{\mathrm{Ric}}|^2)^{1/2} \geq (1-\alpha_n) \frac{ |\nabla\widetilde{\mathrm{Ric}}|^2}{(1 + |\widetilde{\mathrm{Ric}}|^2)^{1/2}} - C |\mathrm{Rm}|\, |\widetilde{\mathrm{Ric}}|. \end{aligned} \end{align} $$

Integrating over Y, we get

(25) $$ \begin{align}\begin{aligned} \int_Y \frac{|\nabla\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2}{(1 + |\widetilde{\mathrm{Ric}}|_{\omega_\epsilon}^2)^{1/2}}\, \omega_\epsilon^n \leq C_1 \Vert \mathrm{Rm}_{\omega_\epsilon}\Vert_{L^2} \Vert \widetilde{\mathrm{Ric}}_{\omega_\epsilon} \Vert_{L^2} \to 0, \end{aligned} \end{align} $$

as $\epsilon \to 0$ . It then follows from (22) that

(26) $$ \begin{align}\begin{aligned} \int_Y |\Delta (1 + |\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2)^{1/2}| \, \omega_\epsilon^n \leq \int_Y \frac{|\nabla\widetilde{\mathrm{Ric}}_{\omega_\epsilon}|^2}{(1 + |\widetilde{\mathrm{Ric}}|_{\omega_\epsilon}^2)^{1/2}}\, \omega_\epsilon^n + C \Vert \mathrm{Rm}_{\omega_\epsilon}\Vert_{L^2} \Vert \widetilde{\mathrm{Ric}}_{\omega_\epsilon} \Vert_{L^2} \to 0,\end{aligned} \end{align} $$

as $\epsilon \to 0$ .

3.2 Proof of the RCD property

In this section we assume that $(X,\omega _{KE})$ is a singular Kähler-Einstein space, with $\mathrm {Ric}_{\omega _{KE}} = \lambda \omega _{KE}$ , that can be approximated with cscK metrics as in Definition 2. Our first result is the following.

Proposition 15. The metric completion $(\hat {X}, d, \omega _{KE}^n)$ is an $RCD(\lambda , 2n)$ space.

Proof. From Proposition 9 it follows that it is sufficient to check condition (3) in Corollary 8, that is, to show that the eigenfunctions of the Laplacian on $\hat {X}$ are bounded. More precisely, suppose that $u\in W^{1,2}(\hat {X})$ satisfies $\Delta u = -b u$ on $X^{reg}$ for a constant b. We will show that then $|\nabla u| \in L^\infty (X^{reg})$ .

For simplicity we can assume that $\Vert u\Vert _{L^2} = 1$ . Using that $u\in W^{1,2}(\hat {X})$ , and also [Reference Guo, Phong, Song and Sturm33, Lemma 11.2], we have

(27) $$ \begin{align}\begin{aligned} \sup |u| + \int_{X^{reg}} |\nabla u|^2 \omega_{KE}^n < C, \end{aligned} \end{align} $$

where C could depend on u (in particular on b).

Next we will use the approximating cscK metrics $\omega _\epsilon $ on the resolution Y of X. Let us fix a large i, and let $f = \phi _iu$ for the cutoff function $\phi _i$ in Lemma 6. We can view f as a function on Y, supported away from the exceptional divisor, where the metrics $\omega _\epsilon $ converge smoothly to $\omega _{KE}$ . Note that we have a uniform bound $\sup |f| < C$ , and also

(28) $$ \begin{align}\begin{aligned} \int_Y |\nabla f|^2 \omega_\epsilon^n \leq \int_Y 2( |u \nabla\phi_i|^2 + |\phi_i \nabla u|^2)\, \omega_\epsilon^n < 2C, \end{aligned} \end{align} $$

for sufficiently small $\epsilon $ .

Let us fix a point $x_0\in X$ where $\phi _i(x_0)=1$ . We can view $x_0\in Y$ too. We will do the following calculation on Y, using the metric $\omega _\epsilon $ for sufficiently small $\epsilon $ . To simplify the notation we will omit the subscript $\epsilon $ . All geometric quantities are defined using the metric $\omega _\epsilon $ . We will write $\rho _t = H(x_0,y,t)$ for the heat kernel centered at $x_0$ on $(Y, \omega _\epsilon )$ , and let $f_t$ denote the solution of the heat equation on $(Y, \omega _\epsilon )$ with initial condition f. We will also omit the volume form $\omega _\epsilon ^n$ in the integrals below. We have the following.

(29) $$ \begin{align}\begin{aligned} \partial_s \int_Y \frac{1}{2}|\nabla f_{t-s}|^2\, \rho_s &= \int_Y -\langle \nabla f_{t-s}, \nabla \Delta f_{t-s}\rangle \rho_s + \frac{1}{2}|\nabla f_{t-s}|^2\, \Delta \rho_s \\ &= \int_Y \Big( |\nabla^2 f_{t-s}|^2 + \mathrm{Ric}(\nabla f_{t-s}, \nabla f_{t-s}) \Big) \rho_s. \end{aligned} \end{align} $$

In order to compensate for the Ricci term, we let $\psi ^2 = (1 + |\widetilde {\mathrm {Ric}}|^2)^{1/2}$ , where $\widetilde {\mathrm {Ric}} = \mathrm {Ric}_{\omega _\epsilon } - \lambda \omega _\epsilon $ as in Proposition 14. We have

(30) $$ \begin{align}\begin{aligned} \partial_s \int_Y \psi^2 f_{t-s}^2\, \rho_s &= \int_Y -2\psi^2 f_{t-s} \Delta f_{t-s} \,\rho_s + \psi^2 f_{t-s}^2 \Delta \rho_s \\ &= \int_Y \Big(\Delta(\psi^2) f_{t-s}^2 + 2\langle \nabla\psi^2, \nabla f^2_{t-s}\rangle + 2\psi^2 |\nabla f_{t-s}|^2\Big)\, \rho_s \\ &\geq -C\int_Y (|\Delta \psi^2| + |\nabla\psi|^2)\, \rho_s + \int_Y \psi^2 |\nabla f_{t-s}|^2\, \rho_s, \end{aligned} \end{align} $$

where the constant C depends on the uniform supremum bound for $f_{t-s}$ .

Note that $\psi ^2 \geq |\mathrm {Ric}| - n|\lambda |$ , so if we combine (29) and (30), we get

(31) $$ \begin{align}\begin{aligned} \partial_s \int_Y \left( \frac{1}{2} |\nabla f_{t-s}|^2 + \psi^2 f_{t-s}^2\right)\, \rho_s \geq -C\int_Y (|\Delta\psi^2| + |\nabla\psi|^2)\, \rho_s - \int_Y n|\lambda| |\nabla f_{t-s}|^2\, \rho_s. \end{aligned} \end{align} $$

At this point, let us fix $s_0> 0$ , and only work with $s \in [s_0, 2]$ . From Proposition 14 we know that $\Vert \Delta \psi ^2 \Vert _{L^1}, \Vert \nabla \psi \Vert _{L^2} \to 0$ as $\epsilon \to 0$ . From Theorem 13 we have a uniform upper bound for $\rho _s$ , depending on $s_0$ , but independent of $\epsilon $ . Therefore, if we choose $\epsilon $ sufficiently small, say $\epsilon < \epsilon _{s_0}$ , then we have

(32) $$ \begin{align}\begin{aligned} \partial_s \int_Y \left( \frac{1}{2} |\nabla f_{t-s}|^2 + \psi^2 f_{t-s}^2\right)\, \rho_s \geq -1 - n|\lambda| \int_Y |\nabla f_{t-s}|^2\, \rho_s, \end{aligned} \end{align} $$

and so

(33) $$ \begin{align}\begin{aligned} \partial_s\, e^{2n|\lambda|s} \int_Y \left( \frac{1}{2} |\nabla f_{t-s}|^2 + \psi^2 f_{t-s}^2\right)\, \rho_s \geq -C. \end{aligned} \end{align} $$

Applying this with $t=1+s_0$ and integrating from $s=s_0$ to $s=1+s_0$ , it follows that for such $\epsilon $ we have

(34) $$ \begin{align}\begin{aligned} e^{2n|\lambda| s_0} \int_Y \left(\frac{1}{2} |\nabla f_1|^2 + \psi^2 f_1^2\right)\, \rho_{s_0} \leq C + e^{2n|\lambda|(s_0+1)} \int_Y \left(\frac{1}{2}|\nabla f|^2 + \psi^2 f^2\right)\, \rho_{1+s_0}. \end{aligned} \end{align} $$

Using the uniform upper bound for $\rho _{1+s_0}$ , together with the integral bound for $|\widetilde {\mathrm {Ric}}|^2$ from Proposition 14, we obtain that

(35) $$ \begin{align}\begin{aligned} \int_Y |\nabla f_1|^2\, \rho_{s_0} \leq C, \end{aligned} \end{align} $$

where C is independent of $\epsilon , s_0$ . As $\epsilon \to 0$ , the heat kernels $\rho _{s_0}$ converge locally smoothly on $X^{reg}$ to the heat kernel on $(\hat {X}, \omega _{KE})$ , and so in the limit we obtain the estimate

(36) $$ \begin{align}\begin{aligned} \int_{X^{reg}} |\nabla f_1|^2\, \rho_{s_0} \leq C, \end{aligned} \end{align} $$

where all the quantities are computed using $\omega _{KE}$ , and recall that $f_1$ is simply the solution $f_t$ of the heat flow with initial condition f at time $t=1$ . Note that the constant C does not depend on $s_0$ , so in fact, by letting $s_0\to 0$ , we obtain the pointwise estimate

(37) $$ \begin{align}\begin{aligned} |\nabla f_1|^2(x_0) \leq C, \end{aligned} \end{align} $$

and this holds uniformly for any $x_0\in X^{reg}$ .

Recall that $f= \phi _i u$ , where u is the eigenfunction that we want to estimate, and $\phi _i$ is a cutoff function from Lemma 6. To keep track of the dependence on i, let us now write $f^{(i)}= \phi _i u$ , and write $f^{(i)}_1$ for the corresponding solutions of the heat equation at time $1$ . Since $f^{(i)} \to u$ in $L^2$ , it follows that for any compact set $K\subset X^{reg}$ the solutions $f^{(i)}_1$ converge smoothly to $u_1$ on K. But $u_1 = e^{-b}u$ , so we obtain the required pointwise bound $|\nabla u|^2(x_0) \leq e^{2b} C$ for any $x_0\in X^{reg}$ .

Next we show that singular Kähler-Einstein metrics on projective varieties, that can be approximated by cscK metrics, define Kähler currents. This result was previously shown by Guedj-Guenancia-Zeriahi [Reference Guedj, Guenancia and Zeriahi31] for singular Kähler-Einstein metrics that are either globally smoothable, or that only have isolated smoothable singularities.

Theorem 16. Let $\omega _{KE}$ denote a singular Kähler-Einstein metric on a normal projective variety X, which can be approximated by cscK metrics as in Definition 2. Let $\eta _{FS}$ denote the pullback of the Fubini-Study metric to X under a projective embedding of X. Then there is a constant $\delta> 0$ such that $\omega _{KE}> \delta \eta _{FS}$ .

Proof. By assumption we have cscK metrics $\omega _\epsilon = \eta _\epsilon + \sqrt {-1}\partial \bar \partial u_\epsilon $ on a resolution $\pi :Y \to X$ , where $\eta _\epsilon \to \pi ^*\eta _X$ for a smooth metric $\eta _X$ on X, where $\eta _\epsilon \geq \pi ^*\eta _X$ . We apply the Chern-Lu inequality to the map $\pi : Y \to X$ , away from the exceptional divisor E, where on Y we use the metric $\omega _\epsilon $ and on X we use the pullback $\eta _{FS}$ of the Fubini-Study metric under a projective embedding of X. For simplicity we write $\eta _{FS}$ for $\pi ^*\eta _{FS}$ , and we write $g_{i\bar j}$ and $h_{i\bar j}$ for the metric components of $\omega _\epsilon $ and $\eta _{FS}$ , respectively. On $Y\setminus E$ we then have $|\partial \pi |^2 = \mathrm {tr}_{\omega _\epsilon } \eta _{FS}$ , and (see, e.g., [Reference Jeffres, Mazzeo and Rubinstein46])

(38) $$ \begin{align}\begin{aligned} \Delta_{\omega_\epsilon} \log \mathrm{tr}_{\omega_\epsilon} \eta_{FS} \geq \frac{ g^{i\bar l}g^{k\bar j} \mathrm{Ric}(\omega_\epsilon)_{i\bar j} h_{k\bar l}}{\mathrm{tr}_{\omega_\epsilon} \eta_{FS}} - A \mathrm{tr}_{\omega_\epsilon} \eta_{FS}, \end{aligned} \end{align} $$

where A is independent of $\epsilon $ , using that $\eta _{FS}$ has bisectional curvature bounded from above. It follows that

(39) $$ \begin{align}\begin{aligned} \Delta_{\omega_\epsilon} \log \mathrm{tr}_{\omega_\epsilon} \eta_{FS} &\geq \frac{ g^{i\bar l}g^{k\bar j} (\mathrm{Ric}(\omega_\epsilon)_{i\bar j} - \lambda g_{i\bar j})h_{k\bar l}}{\mathrm{tr}_{\omega_\epsilon} \eta_{FS}} + \lambda - A \mathrm{tr}_{\omega_\epsilon} \eta_{FS} \\ &\geq -|\mathrm{Ric}_{\omega_\epsilon} - \lambda\omega_\epsilon| +\lambda - A\mathrm{tr}_{\omega_\epsilon} \eta_X. \end{aligned} \end{align} $$

We also have

(40) $$ \begin{align}\begin{aligned} \Delta_{\omega_\epsilon} (-u_\epsilon) = \mathrm{tr}_{\omega_\epsilon} \eta_\epsilon - n \geq \mathrm{tr}_{\omega_\epsilon} \eta_X - n \geq C_1^{-1} \mathrm{tr}_{\omega_\epsilon} \eta_{FS} - n, \end{aligned} \end{align} $$

for some $C_1> 0$ , using that locally both $\eta _{FS}$ and $\eta _X$ are given by pullbacks of smooth metrics under embeddings of X. This implies that

(41) $$ \begin{align}\begin{aligned} \Delta_{\omega_\epsilon} ( \log \mathrm{tr}_{\omega_\epsilon} \eta_{FS} - AC_1u_\epsilon) &\geq -|\mathrm{Ric}_{\omega_\epsilon} - \lambda\omega_\epsilon| +\lambda - AC_1n \\ &\geq -|\mathrm{Ric}_{\omega_\epsilon} - \lambda\omega_\epsilon| - C_2, \end{aligned} \end{align} $$

for some $C_2> 0$ . Let us define

(42) $$ \begin{align}\begin{aligned} F = \max\{0, \log \mathrm{tr}_{\omega_\epsilon} \eta_{FS} - AC_1u_\epsilon\}. \end{aligned} \end{align} $$

Since $\omega _\epsilon $ is a Kähler metric, F is bounded from above, and by definition F is also bounded below. In addition F satisfies the differential inequality

(43) $$ \begin{align}\begin{aligned} \Delta_{\omega_\epsilon} F \geq - |\mathrm{Ric}_{\omega_\epsilon} - \lambda\omega_\epsilon| - C_2 \end{aligned} \end{align} $$

in a distributional sense on all of Y. To see this, note first that the differential inequality is satisfied in the distributional sense on $Y\setminus E$ by the definition of F as a maximum of two functions satisfying the inequality. Then the differential inequality can be extended across E using that F is bounded, by an argument similar to Lemma 11.

Fix $x\in Y\setminus E$ , and let $H(x,y,t)$ denote the heat kernel on $(Y, \omega _\epsilon )$ . Fix some $t_0> 0$ . For $t\in [t_0, 1]$ we have

(44) $$ \begin{align}\begin{aligned} \partial_t \int_Y F(y)\, H(x,y,t)\, dy &= \int_Y F(y)\, \Delta_y H(x,y,t)\, dy \\ &= \int_Y \Delta_y F(y) \, H(x,y,t)\, dy \\ &\geq \int_Y (-|\mathrm{Ric}_{\omega_\epsilon} - \lambda\omega_{\epsilon}|(y) -C_2) H(x,y,t)\, dy. \end{aligned} \end{align} $$

Using the uniform upper bound for H (see Theorem 13), together with Proposition 14, we find that there exists an $\epsilon _0 = \epsilon _0(t_0)$ , depending on $t_0$ , such that if $\epsilon < \epsilon _0$ , then

(45) $$ \begin{align}\begin{aligned} \partial_t \int_Y F(y)\, H(x,y,t)\, dy \geq -2C_2, \end{aligned} \end{align} $$

and so for $\epsilon < \epsilon _0$ we have

(46) $$ \begin{align}\begin{aligned} \int_Y F(y) H(x,y,t_0)\, dy &\leq \int_Y F(y) H(x,y,1)\, dy + 2C_2. \end{aligned} \end{align} $$

Note that

(47) $$ \begin{align}\begin{aligned} F \leq e^{-AC_1u_\epsilon} \mathrm{tr}_{\omega_\epsilon} \eta_{FS}, \end{aligned} \end{align} $$

so we have (using the uniform upper bound for the heat kernel as well),

(48) $$ \begin{align}\begin{aligned} \int_Y F(y) H(x,y,t_0)\, dy &\leq C_3 e^{AC_1\sup |u_\epsilon|} \int_Y \mathrm{tr}_{\omega_\epsilon}\eta_{FS}\, \,\omega_\epsilon^n + 2C_2 \\ &\leq C_4. \end{aligned} \end{align} $$

Here we also used that we have a uniform bound for $\sup |u_\epsilon |$ , and the cohomology classes $[\omega _\epsilon ]$ are uniformly bounded. Crucially, the constant $C_4$ is independent of $t_0$ .

Note that as $\epsilon \to 0$ , the heat kernels $H(x,y,t)$ for $(Y,\omega _\epsilon )$ converge locally smoothly on $Y\setminus E$ to the heat kernel for $(X, \omega _{KE})$ . At the same time, the function $F(y)$ converges locally uniformly on $Y\setminus E$ to

(49) $$ \begin{align}\begin{aligned}\max\{0,\log \mathrm{tr}_{\omega_{KE}}\eta_{FS} - AC_1u_{KE}\}.\end{aligned} \end{align} $$

It follows that in the limit, for any $t> 0$ , we have

(50) $$ \begin{align}\begin{aligned} \int_{X^{reg}} (\log\mathrm{tr}_{\omega_{KE}}\eta_{FS}- AC_1u_{KE})(y)\, H_{\omega_{KE}}(x,y,t) \, \omega_{KE}^n(y) \leq C_4. \end{aligned} \end{align} $$

Letting $t\to 0$ we obtain a pointwise bound $\mathrm {tr}_{\omega _{KE}}\eta _{FS} < C_5$ , as required.

4 Homeomorphism with the underlying variety

In this section our goal is to show that the metric completion $\hat {X}$ of the smooth locus of a singular Kähler-Einstein metric $(X, \omega _{KE})$ is homeomorphic to X, under suitable assumptions. These assumptions hold in the setting of Theorem 4, where $(X,\omega _{KE})$ can be approximated with cscK metrics.

We assume that X is a normal projective variety of dimension n, and we have a Kähler current $\omega $ on X with bounded local potentials, such that $\omega \in c_1(L)$ for a line bundle L on X. We will write $\eta _{FS}$ for the pullback of the Fubini-Study metric to X under a projective embedding. We make the following assumptions:

  1. (1) The Ricci form of $\omega $ , as a current, satisfies $\mathrm {Ric}(\omega ) = \lambda \omega $ for a constant $\lambda \in \mathbb {R}$ on the regular part $X^{reg}$ of X.

  2. (2) $\omega $ is a Kähler current, that is, $\omega \geq c \eta _{FS}$ on X for some $c> 0$ .

  3. (3) The metric completion $(\hat {X}, d_{\hat {X}})$ of $(X^{reg}, \omega )$ is a noncollapsed $RCD(2n, \lambda )$ space, where the measure on $\hat {X}$ is the pushforward of $\omega ^n$ from $X^{reg}$ .

  4. (4) We have $\omega ^n = F \eta _{FS}^n$ , where $F \in L^p(X, \eta _{FS}^n)$ for some $p> 1$ .

We have seen that Conditions (1)–(3) are satisfied for singular Kähler-Einstein metrics $(X,\omega _{KE})$ , with $\omega _{KE}\in c_1(L)$ , that can be approximated with cscK metrics in the sense of Definition 2. For Condition (4), see Eyssidieux-Guedj-Zeriahi [Reference Eyssidieux, Guedj and Zeriahi29, Section 7].

The main result of this section is the following, and the proof will be completed after Proposition 27 below.

Theorem 17. Let $(X,\omega )$ satisfy the conditions (1)–(4) above. Then the metric completion $\hat {X}$ is homeomorphic to X.

Rescaling the metric $\omega $ we can assume that L is a very ample line bundle on X. The sections of L define a holomorphic embedding $\Phi _X : X\to \mathbb {CP}^N$ , and we can identify the image of this embedding with X. By the assumption that $\omega $ is a Kähler current, we have that the map

(51) $$ \begin{align}\begin{aligned} \Phi_X : (X^{reg}, \omega) \to (X, \eta_{FS}) \subset \mathbb{CP}^N \end{aligned} \end{align} $$

is Lipschitz continuous, where we use the length metric as defined in (3). In particular $\Phi _X$ extends to a Lipschitz continuous map

(52) $$ \begin{align}\begin{aligned} \hat{\Phi}_X : \hat{X} \to (X, \eta_{FS}). \end{aligned} \end{align} $$

Note that $\hat {\Phi }_X$ is surjective, since the image of $X^{reg}$ is dense in X, so our task is to prove that $\hat {\Phi }_X$ is injective, that is, to show that the sections of L separate points of $\hat {X}$ . In fact we will work with $L^k$ for large k, however since L is very ample, the map defined by section of $L^k$ is obtained by composing the map defined by sections of L with an embedding of $\mathbb {CP}^N$ into a larger projective space.

The general strategy for showing that sections of $L^k$ separate points of $\hat {X}$ is similar to the work of Donaldson-Sun [Reference Donaldson and Sun27]. We will apply the following form of Hörmander’s estimate (see, e.g., [Reference Demailly25, Theorem 6.1]):

Theorem 18. Let $(P,h_P)$ be a Hermitian holomorphic line bundle on a Kähler manifold $(M, \omega _M)$ , which admits some complete Kähler metric. Suppose that the curvature form of $h_P$ satisfies $\sqrt {-1}F_{h_P} \geq c \omega _M$ for some constant $c> 0$ . Let $\alpha \in \Omega ^{n,1}(P)$ be such that $\bar \partial \alpha = 0$ . Then there exists $u\in \Omega ^{n,0}(P)$ such that $\bar \partial u = \alpha $ , and

(53) $$ \begin{align}\begin{aligned} \Vert u\Vert_{L^2}^2 \leq \frac{1}{c} \Vert \alpha\Vert^2_{L^2}, \end{aligned} \end{align} $$

provided the right hand side is finite.

We will apply this result to $M=X^{reg}$ , with the metric $\omega _M=k\omega $ . Note that it follows from Demailly [Reference Demailly24, Theorem 0.2], that $X^{reg}$ admits a complete Kähler metric. For the line bundle P we will take $P = L^k \otimes K_M^{-1}$ , so that an $(n,0)$ -form valued in P is simply a section of $L^k$ . For the metric on P we take the metric induced by the metric $h^k$ on $L^k$ whose curvature is $k\omega $ , together with the metric given by $\omega ^n$ on $K_{M}$ . The curvature of $h_P$ then satisfies

(54) $$ \begin{align}\begin{aligned} \sqrt{-1}F_{h_P} = k\omega + \mathrm{Ric}_{\omega} = (k+\lambda)\omega> \frac{1}{2}\omega_M, \end{aligned} \end{align} $$

for large enough k.

We will need the following $L^\infty $ and gradient estimates for holomorphic sections of $L^k$ .

Proposition 19. Let f be a holomorphic section of $L^k$ over $M = X^{reg}$ . We then have the following estimates

(55) $$ \begin{align}\begin{aligned} \sup_{M} |f|_{h^k} + |\nabla f|_{h^k, \omega_M} \leq K_1\Vert f\Vert_{L^2(M, h^k, \omega_M)}, \end{aligned} \end{align} $$

where we emphasize that we are using the metrics $h^k$ and $\omega _M = k\omega $ to measure the various norms, and $K_1$ does not depend on k.

Proof. Note first that f extends to a holomorphic section of $L^k$ over X, using that X is normal. Using that $\omega $ has locally bounded potentials, we have that $\sup _X |f|_{h^k} < \infty $ .

Next we show that $|\nabla f|_{h^k, \omega _M} < \infty $ . For any $\hat {x}\in \hat {X}$ , let $x = \hat {\Phi }_X(\hat x)\in X$ . We can find a section $s\in H^0(X, L)$ and some $r> 0$ such that $s(y) \not =0$ for $y\in B_{\eta _{FS}}(x,r)$ . The assumption that $\omega $ is a Kähler current implies that we have constants $r'> 0$ and $C> 0$ (depending on $\hat {x}$ ) such that if we write $|s|^2_h = e^{-u}$ , then $|u| < C$ on $X^{reg} \cap B_{\omega _M}(\hat {x}, r')$ . We have $\Delta _{\omega _M} u = n$ on $X^{reg}\cap B_{\omega _M}(\hat {x}, r'/2)$ , and since u is bounded, this equation extends to $B_{\omega _M}(\hat {x}, r'/2)$ by Lemma 11 and Lemma 6. The gradient estimate in Proposition 12 then implies that $|\nabla u| < C_1$ on $B_{\omega _M}(\hat {x}, r'/2)$ . This implies that $|\nabla s| < C_2$ on $B_{\omega _M}(\hat {x}, r'/2)$ . If f is any holomorphic section of $L^k$ , then on $B_{\omega _M}(\hat {x}, r'/2)$ the ratio $f/s^k$ is a bounded harmonic function, so using the gradient estimate again, together with the bounds for s, we find that $|\nabla f| < C_3$ on $B_{\omega _M}(\hat {x}, r'/4)$ . We can cover $\hat {X}$ with finitely many balls of this type, showing that $|\nabla f|_{h^k, \omega _M} < \infty $ globally.

We can obtain the effective estimates claimed in the proposition as follows. Since the curvature of $h^k$ is $\omega _M$ , on M we have

(56) $$ \begin{align}\begin{aligned} \Delta_{\omega_M} |f|^2_{h^k} = |\nabla f|^2_{h^k, \omega_M} - n |f|^2_{h^k}. \end{aligned} \end{align} $$

Let $\phi _i$ denote cutoff functions as in Lemma 6. We have, omitting the subscripts,

(57) $$ \begin{align}\begin{aligned} \int_{M} \phi_i^2 |\nabla f|^2 \omega_M^n &= \int_{M} \phi_i^2 (\Delta |f|^2 + n |f|^2)\, \omega_M^n \\ &= \int_{M} (-4\phi_i |f| \nabla \phi_i\cdot \nabla |f| + \phi_i^2n |f|^2)\, \omega_M^n \\ &\leq \int_{M} \left( \frac{1}{2} \phi_i^2 |\nabla f|^2 + 8 |\nabla\phi_i|^2 |f|^2 + \phi_i^2 n |f|^2\right) \,\omega_M^n. \end{aligned} \end{align} $$

Letting $i\to \infty $ , and using that $|f|\in L^\infty $ , we get

(58) $$ \begin{align}\begin{aligned} \int_{M} |\nabla f|^2\, \omega_M^n \leq 2n\int_{M} |f|^2\, \omega_M^n. \end{aligned} \end{align} $$

We also have the following Bochner-type formula on M (see, e.g., La Nave-Tian-Zhang [Reference La Nave, Tian and Zhang39, Lemma 3.1]):

(59) $$ \begin{align}\begin{aligned} \Delta |\nabla f|^2 \geq \mathrm{Ric}_{\omega_M}(\nabla f, \nabla f) - (n+2) |\nabla f|^2 \geq -(n+2+|\lambda|) |\nabla f|^2, \end{aligned} \end{align} $$

where we are using the metrics $h^k, \omega _M$ as above.

Both (56) and (59) are of the form

(60) $$ \begin{align}\begin{aligned} \Delta v \geq - Av, \end{aligned} \end{align} $$

where v is a smooth $L^\infty $ function on M. We can argue using the cutoff functions $\phi _i$ , as in the proof of Lemma 11, to show that v satisfies this differential inequality on all of $\hat {X}$ in a weak sense, that is, for any Lipschitz test function $\rho \geq 0$ we have

(61) $$ \begin{align}\begin{aligned} \int_{M} (-\nabla \rho\cdot \nabla v + A\rho v)\, \omega_M^n \geq 0. \end{aligned} \end{align} $$

Using this, together with estimates for the heat kernel on $\hat {X}$ , we can obtain the required $L^\infty $ bound for $v = |f|^2$ and $v = |\nabla f|^2$ . More precisely, using [Reference Jiang, Li and Zhang38, Theorem 1.2], together with the RCD property in Proposition 15, we obtain an $L^2$ -bound for the heat kernel $H(x,y,1)$ on M, independently of k. Using (60), for any $x\in M$ we have

(62) $$ \begin{align}\begin{aligned} \frac{d}{dt} \int_M v(y) H(x, y,t)\, \omega_M^n(y) = \int_M v(y) \Delta_y H(x,y,t)\, \omega_M^n(y) \geq -A \int_M v(y)H(x,y,t)\, \omega_M^n(y), \end{aligned} \end{align} $$

so

(63) $$ \begin{align}\begin{aligned} v(x) \leq e^A \int_M v(y) H(x,y,1)\, \omega_M^n(y) \leq e^A C \Vert v\Vert_{L^2}, \end{aligned} \end{align} $$

as required.

In order to show that sections of $L^k$ separate points of $\hat {X}$ for large k (and therefore also for $k=1$ ), we follow the approach of Donaldson-Sun [Reference Donaldson and Sun27], constructing suitable sections of $L^k$ using Hörmander’s $L^2$ -estimate. For this the basic ingredient in [Reference Donaldson and Sun27] is to consider a tangent cone Z of $\hat {X}$ at x, and use that the regular part of Z is a Kähler cone, while at the same time the singular set can be excised by a suitable cutoff function. The main new difficulty in our setting is that along the pointed convergence of a sequence of rescalings

(64) $$ \begin{align}\begin{aligned} (\hat{X}, \lambda_i d_{\hat{X}}, x) \to (Z, d_Z, o),\end{aligned} \end{align} $$

with $\lambda _i\to \infty $ , we do not know that compact subsets $K\subset Z^{reg}$ of the (metric) regular set in Z are obtained as smooth limits of subsets of the (complex analytic) regular set $X^{reg}$ . For example, a priori it may happen that along the convergence in (64), even if $Z=\mathbb {R}^{2n}$ , the singular set $X\setminus X^{reg}$ converges to a dense subset of Z. This is similar to the issue dealt with in Chen-Donaldson-Sun [Reference Chen, Donaldson and Sun17], but in that work it is used crucially that the singular spaces considered are limits of smooth manifolds with lower Ricci bounds.

To deal with this issue in our setting, we exploit the fact that $X\setminus X^{reg}$ is locally contained in the zero set of holomorphic functions, which also define harmonic functions on the RCD space $\hat {X}$ . Crucially, these functions have a bound on their order of vanishing (Lemma 20), which can be used to control the size of the zero set at different scales, at least on balls that are sufficiently close to a Euclidean ball. This can be used to show that balls in $\hat {X}$ that are almost Euclidean are contained in $X^{reg}$ (Proposition 24). This is the main new ingredient in our argument. Given this, we can closely follow the arguments in Donaldson-Sun [Reference Donaldson and Sun27] or [Reference Liu and Székelyhidi42] to construct holomorphic sections of $L^k$ .

Let us write $\Gamma = X\setminus X^{reg}$ for the algebraic singular set. Observe that $\Gamma $ can locally be cut out by holomorphic functions. Therefore, we can cover X with open sets $U_k'$ and we have nonzero holomorphic functions $s_k$ on $U_k'$ such that $\Gamma \cap U_k' \subset s_k^{-1}(0)$ . We can assume that the $s_k$ are bounded, and that we have relatively compact open sets that still cover X. We let $\hat {U}_k, \hat {U}_k'$ be the corresponding open sets pulled back to $\hat {X}$ . Using Lemma 6, we can extend the $s_k$ to complex valued harmonic functions on $\hat {X}$ , which vanish along $\Gamma $ . Our first task will be to show that we have a bound for the order of vanishing of the $s_k$ at each point. Note first that by the assumption that $\omega $ is a Kähler current, there exists an $r_0> 0$ such that if $p\in \hat {U}_k$ , then $B(p, r_0) \subset \hat {U}_k'$ . Here, and below, a ball $B(p,r)$ always denotes the metric ball using the metric $d_{\hat X}$ on $\hat {X}$ induced by $d_{\omega }$ on $X^{reg}$ .

Lemma 20. There are constant $c_1, N> 0$ , depending on $(X, \omega )$ , such that for any $\hat {x} \in \hat {U}_k$ and $r\in (0,r_0)$ , we have

(65) $$ \begin{align}\begin{aligned} \int_{B(\hat{x}, r)} |s_k|^2\, \omega^n \geq c_1 r^N\!,\end{aligned} \end{align} $$

for all $r < r_0$ .

Proof. First note that since $\hat {X}$ is a noncollapsed RCD space, we have a constant $\nu> 0$ such that $\mathrm {vol}\, B(\hat {x}, r)> \nu r^{2n}$ for all $r < 1$ . At the same time we can bound the volume of sublevel sets $U_k'\cap \{|s_k| < t\}$ from above, using the assumptions on $\omega $ . Indeed, on $U_k'$ we have $\omega ^n = F\eta ^n_{FS}$ , and $F\in L^p(X, \eta ^n_{FS})$ for some $p> 1$ . It follows that for any $t> 0$ we have

(66) $$ \begin{align}\begin{aligned} \mathrm{vol} (U_k'\cap \{ |s_k| < t\}, \omega^n) &= \int_{U_k'\cap \{|s_k| < t\}} \omega^n \\ &= \int_{U_k'\cap \{|s_k| < t\}} F\, \eta_{FS}^n \\ &\leq C_1 \mathrm{vol}(U_k'\cap \{|s_k| < t\}, \eta_{FS}^n)^{1/p'} \left(\int_{U_k'} F^p\, \eta_{FS}^n\right)^{1/p} \\ &\leq C_2 \mathrm{vol}(U_k'\cap \{|s_k| < t\}, \eta_{FS}^n)^{1/p'}\!, \end{aligned} \end{align} $$

for suitable constants $C_1, C_2$ independent of t, and $p'$ is the conjugate exponent of p. Since $|s_k|^{-\epsilon }\, \eta _{FS}^n$ is integrable for some $\epsilon> 0$ , it follows that we have a bound

(67) $$ \begin{align}\begin{aligned} \mathrm{vol}(U_k'\cap \{|s_k| < t\}, \eta_{FS}^n) \leq C_3 t^\epsilon, \end{aligned} \end{align} $$

and so in sum we have

(68) $$ \begin{align}\begin{aligned} \mathrm{vol} (U_k'\cap \{ |s_k| < t\}, \omega^n) \leq C_4 t^{\alpha},\end{aligned} \end{align} $$

for some $C_4, \alpha> 0$ independent of t. Given a small $r> 0$ such that $B(\hat {x}, r)\subset \hat {U}_k'$ , choose $t_r$ such that

(69) $$ \begin{align}\begin{aligned} C_4 t_r^\alpha = \frac{1}{2} \nu r^{2n}, \end{aligned} \end{align} $$

that is,

(70) $$ \begin{align}\begin{aligned} t_r = \left(\frac{\nu}{2C_4}\right)^{1/\alpha} r^{2n/\alpha} = c_5 r^{2n\alpha^{-1}}, \end{aligned} \end{align} $$

for suitable $c_5> 0$ . By our estimates for the volumes, we then have

(71) $$ \begin{align}\begin{aligned} \mathrm{vol}( B(\hat{x}, r) \cap \{|\hat{s}_k| \geq t_r\}) \geq \frac{1}{2}\nu r^{2n}, \end{aligned} \end{align} $$

and so

(72) $$ \begin{align}\begin{aligned} \int_{B(\hat{x}, r)} |\hat{s}_k|^2\, \omega^n &\geq \frac{c_5^2 r^{4n\alpha^{-1}}}{2} \nu r^{2n} = c_1 r^N,\end{aligned} \end{align} $$

for some $c_1, N> 0$ , independent of r, as required.

Next we need a version of the three annulus lemma for almost Euclidean balls, similar to [Reference Ding26, Theorem 0.7].

Lemma 21. For any $\mu> 0$ , $\mu \not \in \mathbb {Z}$ , there is an $\epsilon> 0$ depending on $\mu , n$ with the following property. Suppose that $B(p,1)$ is a unit ball in a noncollapsed $RCD(-1,2n)$ -space such that

(73) $$ \begin{align}\begin{aligned} d_{GH}(B(p,1), B(0_{\mathbb{R}^{2n}},1)) < \epsilon, \end{aligned} \end{align} $$

where $0_{\mathbb {R}^{2n}}$ denotes the origin in Euclidean space. Let $u : B(p,1)\to \mathbb {C}$ be a harmonic function such that

(74)

Then

(75)

Proof. The proof is by contradiction, similarly to [Reference Ding26], based on the fact that on the Euclidean space $\mathbb {R}^{2n}$ every homogeneous harmonic function has integer degree.

Combining the previous two results, we have the following, controlling the decay rate of the defining functions $\hat {s}_k$ around almost regular points.

Lemma 22. There exists an $\epsilon _0, r_0> 0$ , depending on $(X, \omega )$ , such that if $\hat {x}\in \hat {U}_k$ and for some $r_1\in (0,r_0)$ we have

(76) $$ \begin{align}\begin{aligned} d_{GH}(B(\hat{x}, r_1), B(0_{\mathbb{R}^{2n}}, r_1)) < r_1\epsilon_0, \end{aligned} \end{align} $$

then

(77)

for the N in Lemma 20.

Proof. Fix $\mu \in (N/2, N)$ such that $\mu \not \in \mathbb {Z}$ . If $\epsilon _0$ and $r_0$ are sufficiently small (depending on $\mu $ ), then the inequality (76) implies that for any $r \leq r_1$ we have

(78) $$ \begin{align}\begin{aligned} d_{GH}(B(\hat{x}, r), B(0_{\mathbb{R}^{2n}}, r)) < r\epsilon, \end{aligned} \end{align} $$

for the $\epsilon $ in Lemma 21, and so the conclusion of that Lemma holds. It follows that if

(79)

for some $r\leq r_1$ , then applying Lemma 21 inductively, we have

(80)

as long as $2^jr \leq r_1$ . Given any $r \leq r_1$ , if we let $\bar j$ denote the largest j such that $2^jr\leq r_1$ , then we obtain

(81)

where C is independent of r, but depends on the $L^2$ -norm of $\hat {s}_k$ on $B(\hat {x}, r_1)$ . Applying Lemma 20 we then have

(82) $$ \begin{align}\begin{aligned} 2^{-\bar j \mu} C\geq c_1^{1/2} (r/2)^{N/2}. \end{aligned} \end{align} $$

Since $2^{\bar j+1} r> r_1$ , it follows that $2^{-\bar j \mu } < (2r/r_1)^\mu $ , so

(83) $$ \begin{align}\begin{aligned} c_1^{1/2}(r/2)^{N/2} < (2r/r_1)^\mu C. \end{aligned} \end{align} $$

Since $\mu> N/2$ , this inequality implies a lower bound for r satisfying (79). The required conclusion (77) follows.

Using this result, we will show that almost Euclidean balls are contained in the complex analytically regular set $X^{reg} \subset \hat {X}$ . Note that the assumption (85) will hold on sufficiently small balls around a given point, by the previous lemma.

Proposition 23. There exists an $\epsilon _2> 0$ , depending on $(X, \omega )$ , with the following property. Suppose that $\hat {x}\in \hat {U}_j$ and $k> 0$ is a large integer such that $\epsilon _2^{-1}k^{-2} < \epsilon _2$ . Suppose in addition that

(84) $$ \begin{align}\begin{aligned} d_{GH}\Big( B(\hat{x}, \epsilon_2^{-1}k^{-2}), B_{\mathbb{R}^{2n}}(0, \epsilon_2^{-1}k^{-2})\Big) < \epsilon_2 k^{-2}, \end{aligned} \end{align} $$

and that

(85)

for the N in Lemma 20. Then $\hat {x}\in X^{reg}$ , where $X^{reg}$ is the complex analytically regular set of X, viewed as a subset of $\hat {X}$ .

Proof. We will argue by contradiction, similarly to [Reference Liu and Székelyhidi42, Proposition 3.1] which in turn is based on Donaldson-Sun [Reference Donaldson and Sun27]. Suppose that no suitable $\epsilon _2$ exists. Then we have a sequence of points $\hat {x}_i$ , and integers $k_i> i$ such that the hypotheses are satisfied (with $\epsilon _2 = 1/i$ ). We will show that for sufficiently large i we have $\hat {x}_i \in X^{reg}$ by constructing holomorphic coordinates in a neighborhood of $\hat {x}_i$ .

By a slight abuse of notation we will write $\hat {U}_i, \hat {s}_i$ instead of $\hat {U}_{j_i}$ and $\hat {s}_{j_i}$ to simplify the notation. The assumptions imply that the rescaled balls

(86) $$ \begin{align}\begin{aligned} k_i^{1/2} B(\hat{x}_i, i k_i^{-1/2}) \to \mathbb{R}^{2n}, \end{aligned} \end{align} $$

in the pointed Gromov-Hausdorff sense. Using Lemma 21 together with the condition (85), we can extract a nontrivial limit of the normalized functions

(87)

Indeed, we have

(88)

and using Lemma 21 with some $\mu \in (N, 2N)$ , together with (85), implies that for sufficiently large i we have

(89)

for all $j \geq 0$ . In particular, viewed as functions on the rescaled balls $k_i^{1/2} B(\hat {x}_i, ik_i^{-1/2})$ , the $L^2$ norms of the $\tilde {s}_i$ are bounded independently of i on any R-ball. Using the gradient estimate, Proposition 12, it follows that up to choosing a subsequence, the functions $\tilde {s}_i$ converge locally uniformly to a harmonic function $\tilde {s}_\infty : \mathbb {R}^{2n}\to \mathbb {C}$ . As a consequence, $\tilde {s}_\infty $ is smooth, and because of the normalization (88), $\tilde {s}_\infty $ is nonzero.

Note that if we take a sequence of rescalings of $\mathbb {R}^{2n}$ with factors going to infinity, and consider the corresponding pullbacks of $\tilde {s}_\infty $ , normalized to have unit $L^2$ -norm on the unit balls, then this new sequence of harmonic functions will converge to the leading order homogeneous piece of the Taylor expansion of $\tilde {s}_\infty $ at the origin (up to a constant factor). This means that in the procedure above, up to replacing the integers $k_i$ by suitable larger integers, we can assume that the limit $\tilde {s}_\infty $ is in fact homogeneous.

Let us write $\Sigma = \tilde {s}_\infty ^{-1}(0)$ . Our next goal is to show that under the convergence in (86), the set $\mathbb {R}^{2n}\setminus \Sigma $ is the locally smooth limit of subsets of $X^{reg}$ , and that $\tilde {s}_\infty $ is actually a holomorphic function under an identification $\mathbb {R}^{2n} = \mathbb {C}^n$ . Then we will be able to follow the argument in the proof of [Reference Liu and Székelyhidi42, Proposition 3.1] with the cone $V = \mathbb {C}^n$ , but treating $\Sigma $ as the singular set.

Note that since $\tilde {s}_\infty $ is a nonzero harmonic function, the set $\mathbb {R}^{2n}\setminus \Sigma $ is open and dense in $\mathbb {R}^{2n}$ . Suppose that $V\subset \mathbb {R}^{2n}$ is an open subset such that $\bar {V}$ is compact and $|\tilde {s}_\infty |> 0$ on $\bar {V}$ . Then, because of the local uniform convergence of $\tilde {s}_i$ to $\tilde {s}_\infty $ , and the fact that the sets $\tilde {s}_i\not =0$ are contained in $X^{reg}$ , it follows that we have open subsets $V_i \subset \subset k_i^{-2}B(\hat {x}_i, ik_i^{-1/2})\cap X^{reg}$ , which converge in the Gromov-Hausdorff sense to V. The metrics on the $V_i$ are smooth noncollapsed Kähler-Einstein metrics, so using Anderson’s $\epsilon $ -regularity result [Reference Anderson2], up to choosing a subsequence, the complex structures on $V_i$ converge to a complex structure on V with respect to which the Euclidean metric is Kähler. Note that we do not yet know that $\mathbb {R}^{2n}\setminus \Sigma $ is connected, and in principle we may get different complex structures on different connected components. Our next goal is to show that the Hausdorff dimension of $\Sigma $ is at most $2n-2$ , which will show that the complement of $\Sigma $ is connected.

We can assume that the holomorphic functions $\tilde {s}_i$ on $V_i$ converge to a holomorphic function $\tilde {s}_\infty $ on V. Writing $\tilde {s}_\infty = u_\infty + \sqrt {-1} v_\infty $ , we therefore have $\langle \nabla u_\infty , \nabla v_\infty \rangle = 0$ and $|\nabla v_\infty | = |\nabla u_\infty |$ on $\mathbb {R}^{2n}\setminus \Sigma $ , and by density these relations extend to all of $\mathbb {R}^{2n}$ . We can assume that $u_\infty $ is nonconstant. Let $\alpha> 2n-2$ , and suppose that the Hausdorff measure $\mathcal {H}^\alpha (\Sigma )> 0$ . By Caffarelli-Friedman [Reference Caffarelli and Friedman9] (see also Han-Lin [Reference Han and Lin34]) we know that $\mathcal {H}^\alpha (\Sigma \cap |\nabla u_\infty |^{-1}(0)) =0$ , and so we can find an $\alpha $ -dimensional point of density q of $\Sigma \setminus |\nabla u_\infty |^{-1}(0)$ . Since $\nabla u_\infty (q)\not =0$ , it follows that $\nabla v_\infty (q)\not =0$ and $\langle \nabla v_\infty (q), \nabla u_\infty (q)\rangle = 0$ . Therefore in a neighborhood of q the set $\Sigma $ is a smooth $2n-2$ -dimensional submanifold, contradicting that q is an $\alpha $ -dimensional point of density. In conclusion $\dim _{\mathcal {H}} \Sigma \leq 2n-2$ , and so $\mathbb {R}^{2n-2}\setminus \Sigma $ is connected.

We can therefore assume that in the argument above the complex structure that we obtain on $\mathbb {R}^{2n}\setminus \Sigma $ agrees with the standard structure on $\mathbb {C}^n$ , and $\tilde {s}_\infty $ is a holomorphic function on $\mathbb {C}^n\setminus \Sigma $ , but since it is smooth, it is actually holomorphic on $\mathbb {C}^n$ . In particular $\tilde {s}_\infty ^{-1}(0)$ is a complex hypersurface defined by a homogeneous holomorphic function.

At this point we can closely follow the proof of [Reference Liu and Székelyhidi42, Proposition 3.1]), treating the zero set $\tilde {s}_\infty ^{-1}(0)$ as the singular set $\Sigma $ in [Reference Liu and Székelyhidi42]. The properties of the set $\Sigma $ that are used are that the tubular $\rho $ -neighborhood $\Sigma _\rho $ satisfies the volume bounds $\mathrm {vol}(\Sigma _\rho \cap B(0,R))\leq C_R \rho ^2$ , where the constant $C_R$ in our setting could depend on $R, \tilde {s}_\infty $ . In addition if $B(p, 2r) \in \mathbb {R}^{2n}\setminus \Sigma $ , then $B(p, r)$ is the Gromov-Hausdorff limit of balls $B(p_i, r) \subset (M, k_i\omega )$ in Kähler-Einstein manifolds, and so by Anderson’s result [Reference Anderson2] we have good holomorphic charts on the $B(p_i, r)$ for sufficiently large i, analogous to those in [Reference Liu and Székelyhidi42, Theorem 1.4]. The rest of the proof is then identical to the argument in the proof of [Reference Liu and Székelyhidi42, Proposition 3.1] (see also Donaldson-Sun [Reference Donaldson and Sun27]) to show that for sufficiently large i we can construct holomorphic sections $s_0, \ldots , s_n$ of $L^{k_i'}$ for suitable powers $k_i'$ , such that $\frac {s_1}{s_0}, \ldots , \frac {s_n}{s_0}$ define a generically one-to-one map from a neighborhood of $x_i = \hat {\Phi }_X(\hat {x}_i)$ in X to a subset of $\mathbb {C}^n$ . Since X is normal, it follows that the map is one-to-one, and so $x_i\in X^{reg}$ . Therefore $\hat {x}_i\in X^{reg}$ as claimed.

For any $\epsilon> 0$ , let us define the $\epsilon $ -regular set $\mathcal {R}_\epsilon (Y)$ in a noncollapsed RCD space Y to be the set of points p that satisfy

(90) $$ \begin{align}\begin{aligned} \lim_{r\to 0} r^{-2n} \mathrm{vol}(B(p,r))> \omega_{2n} - \epsilon, \end{aligned} \end{align} $$

where $\omega _{2n}$ is the volume of the $2n$ -dimensional Euclidean unit ball. Then $\mathcal {R}_\epsilon (Y)$ is an open set, and from the previous result we obtain the following.

Proposition 24. There exists an $\epsilon _3> 0$ , depending on $(X, \omega )$ , such that the $\epsilon _3$ -regular set $\mathcal {R}_{\epsilon _3}(\hat {X})\subset \hat {X}$ coincides with the complex analytically regular set $X^{reg}$ .

Proof. It is clear that $X^{reg}\subset \mathcal {R}_{\epsilon _3}(\hat {X})$ . To see the reverse inclusion, note that by Cheeger-Colding [Reference Cheeger and Colding12], and De Philippis-Gigli [Reference De Philippis and Gigli23] in the setting of noncollapsed RCD spaces, given the $\epsilon _2> 0$ in Proposition 23, there exists an $\epsilon _3> 0$ such that if $\hat {x} \in \mathcal {R}_{\epsilon _3}$ , then for all sufficiently large k (depending on $\hat {x}$ ), we have

(91) $$ \begin{align}\begin{aligned} d_{GH}\Big( B(\hat{x}, \epsilon_2^{-1}k^{-2}), B_{\mathbb{R}^{2n}}(0, \epsilon_2^{-1}k^{-2})\Big) < \epsilon_2 k^{-2}. \end{aligned} \end{align} $$

Using also Lemma 22 (and choosing $\epsilon _3$ smaller if necessary), we have the growth estimate (85). Proposition 23 then implies that $\hat {x}\in X^{reg}$ .

This has the following immediate corollary.

Corollary 25. There is an $\epsilon> 0$ , depending on $(X, \omega )$ , such that the $\epsilon $ -regular set $\mathcal {R}_\epsilon (\hat {X})$ coincides with the metric regular set of $\hat {X}$ , that is, the points $\hat {x}\in \hat {X}$ where the tangent cone is $\mathbb {R}^{2n}$ .

Given these preliminaries, we have the following result, analogous to [Reference Liu and Székelyhidi42, Proposition 3.1] in our setting.

Proposition 26. Let $(V,o)$ be a metric cone, such that for any $\epsilon> 0$ the singular set $V\setminus \mathcal {R}_\epsilon (V)$ has zero capacity (in the sense of (3) in Definition 5). Let $\zeta> 0$ . There are $K, \epsilon , C> 0$ , depending on $\zeta , (X, \omega ), V$ satisfying the following property. Suppose that k is a large integer such that $\epsilon ^{-1}k^{-1/2} < \epsilon $ and for some $\hat {x}\in \hat {X}$

(92) $$ \begin{align}\begin{aligned} d_{GH}\Big(B(\hat{x}, \epsilon^{-1}k^{-1/2}), B(o, \epsilon^{-1} k^{-1/2})\Big) < \epsilon k^{-1/2}.\end{aligned} \end{align} $$

Then for some $m < K$ the line bundle $L^{mk}$ admits a holomorphic section s over $M=X^{reg}\setminus D$ such that $\Vert s\Vert _{L^2(h^{mk}, mk \omega )} < C$ and

(93) $$ \begin{align}\begin{aligned} \Big| |s(z)| - e^{-mk d(z, \hat{x})^2/2}\Big| < \zeta \end{aligned} \end{align} $$

for $z\in M$ .

Given the results above, the argument is essentially the same as that in [Reference Liu and Székelyhidi42] (see also Donaldson-Sun [Reference Donaldson and Sun27]). One main difference is that in the setting of noncollapsed RCD spaces the sharp estimates of Cheeger-Jiang-Naber [Reference Cheeger, Jiang and Naber14] do not yet seem to be available in the literature. However, the proof of [Reference Liu and Székelyhidi42, Proposition 3.1] applies under the assumption that for any $\epsilon> 0$ the singular set $\Sigma = V\setminus \mathcal {R}_{\epsilon }(V)$ has zero capacity.

We can rule out nonflat (iterated) tangent cones that split off a Euclidean factor of $\mathbb {R}^{2n-2}$ , following the approach of Chen-Donaldson-Sun [Reference Chen, Donaldson and Sun16, Proposition 12] (see also [Reference Liu and Székelyhidi42, Proposition 3.2]).

Proposition 27. Suppose that $\hat {x}_j\in \hat {X}$ and for a sequence of integers $k_j\to \infty $ the rescaled pointed sequence $(\hat {X}, k_j^2 d_{\hat {X}}, \hat {x}_j)$ converges to $\mathbb {R}^{2n-2}\times C(S^1_\gamma )$ in the pointed Gromov-Hausdorff sense. Here $C(S^1_\gamma )$ is the cone over a circle of length $\gamma $ . Then $\gamma = 2\pi $ , that is, $C(S^1_\gamma ) = \mathbb {R}^2$ .

Proof. If $V =\mathbb {R}^{2n-2}\times C(S^1_\gamma )$ , then the singular set of V has capacity zero, and so Proposition 26 can be applied. Then, as in [Reference Chen, Donaldson and Sun16, Proposition 12], it follows that for sufficiently large j, we can find a biholomorphism $F_j$ from a neighborhood $\Omega _j$ of $\hat {x}_j$ to the unit ball $B(0,1)\subset \mathbb {C}^n$ . In particular $B(\hat {x}_j, \frac {1}{2}k_j^{-2}) \subset X^{reg}$ , and then the limit $\mathbb {R}^{2n-2}\times C(S^1_\gamma )$ of $(\hat {X}, k_j^2d_{\hat {X}}, \hat {x}_j)$ must be smooth at the origin. Therefore $\gamma =2\pi $ .

As a consequence of this result we can prove Theorem 17.

Proof of Theorem 17.

Using Propositions 10 and 27, and De Philippis-Gigli’s dimension estimate [Reference De Philippis and Gigli23] for the singular set (extending Cheeger-Colding [Reference Cheeger and Colding12]), it follows that the singular set of any iterated tangent cone of $\hat {X}$ has Hausdorff codimension at least 3. Using Proposition 24 we know that the singular set is closed, and so as in Donaldson-Sun [Reference Donaldson and Sun27, Proposition 3.5] we see that the singular set of any iterated tangent cone has capacity zero. In particular Proposition 26 can be applied to any $(V,o)$ that arises as a rescaled limit of $\hat {X}$ .

Suppose that $p\not = q$ are points in $\hat {X}$ . Applying Proposition 26 to tangent cones at $p, q$ , we can find sections $s_p$ and $s_q$ of some powers $L^{m_p}, L^{m_q}$ , such that $|s_p(p)|> |s_p(q)|$ , and $|s_q(q)|> |s_q(p)|$ . Taking powers we find that the sections $s_p^{m_q}$ and $s_q^{m_p}$ of $L^{m_pm_q}$ separate the points $p,q$ , and so the map $\hat {\Phi }_X$ is injective as required.

To complete the proofs of Theorem 4, it remains to show the codimension bounds for the singular set of $\hat {X}$ . By the dimension estimate of [Reference De Philippis and Gigli23], it suffices to show the following. Note that this result would follow from a version of Cheeger-Colding-Tian [Reference Cheeger, Colding and Tian13, Theorem 9.1] for RCD spaces, but in our setting we can give a more direct proof.

Proposition 28. In the setting of Theorem 4, suppose that a tangent cone $\hat {X}_p$ at $p\in \hat {X}$ splits off an isometric factor of $\mathbb {R}^{2n-3}$ . Then $\hat {X}_p = \mathbb {R}^{2n}$ . In particular in the stratification of the singular set of $\hat {X}$ we have $S_{2n-1} = S_{2n-4}$ , and so $\dim _{\mathcal {H}} S \leq 2n-4$ .

Proof. Suppose that $\hat {X}$ has a tangent cone of the form $\hat {X}_p = C(Z)\times \mathbb {R}^{2n-3}$ , where Z is two-dimensional. If Z had a singular point, necessarily with tangent cone $C(S^1_\gamma )$ for some $\gamma < 2\pi $ , then $\hat {X}$ would have an iterated tangent cone of the form $\mathbb {R}^{2n-2}\times C(S^1_\gamma )$ . This is ruled out by Proposition 27. Therefore Z is actually a smooth two-dimensional Einstein manifold with metric satisfying $\mathrm {Ric}(h) = h$ . This implies that Z is the unit 2-sphere, and it follows that $\hat {X}_p = \mathbb {R}^{2n}$ so that p is a regular point. Therefore the singular set of $\hat {X}$ coincides with $S_{2n-4}$ , as required.

5 CscK approximations

In this section we will prove Theorem 3. Thus, let $(X, \omega _{KE})$ be an n-dimensional singular Kähler-Einstein space, such that the automorphism group of X is discrete and $\omega _{KE} \in c_1(L)$ for an ample $\mathbb {Q}$ -line bundle on X. On the regular part we have $\mathrm {Ric}(\omega _{KE}) = \lambda \omega _{KE}$ for a constant $\lambda \in \mathbb {R}$ . We will assume that $\lambda \in \{0,-1,1\}$ . In the latter two cases we have $L = \pm K_X$ . We first recall the properness of the Mabuchi K-energy in this singular setting. This has been well studied in the Fano setting (see Darvas [Reference Darvas21] for example), but we were not able to find the corresponding much easier result in the literature for singular varieties in the case when $\lambda \leq 0$ .

First recall the definitions of certain functionals (see Darvas [Reference Darvas21] or Boucksom-Eyssidieux-Guedj-Zeriahi [Reference Berman, Boucksom, Guedj and Zeriahi5] for instance). We choose a smooth representative $\omega \in c_1(L)$ . This means that $m\omega $ is the pullback of the Fubini-Study metric under an embedding using sections of $L^m$ for large m. In general we define a function $f : U\to \mathbb {R}$ on an open set $U\subset X$ to be smooth, and write $f\in C^\infty (U)$ , if it is the restriction of a smooth function under an embedding $U\subset \mathbb {C}^N$ . We let

(94) $$ \begin{align}\begin{aligned} \mathcal{H}_{\omega}(X) &= \{ u\in C^\infty(X)\,:\, \omega_u := \omega + \sqrt{-1}\partial \bar\partial u> 0\}, \\ PSH_{\omega}(X) &= \{ u\in L^1(X)\,:\, \omega_u:= \omega + \sqrt{-1}\partial \bar\partial u \geq 0\}. \end{aligned} \end{align} $$

We define the $\mathcal {J}_\omega $ functional on $PSH_\omega (X)\cap L^\infty $ by setting $\mathcal {J}_\omega (0)=0$ and the variation

(95) $$ \begin{align}\begin{aligned} \delta\mathcal{J}_\omega(u) = n\int_{X^{reg}} \delta u (\omega - \omega_u)\wedge \omega_u^{n-1}. \end{aligned} \end{align} $$

Let us choose a smooth metric h on $K_X$ , that is, if $\sigma $ is a local nonvanishing section of $K_X^r$ , then the norm $|\sigma |^2_{h^r}$ is a smooth function. The adapted measure $\mu $ is defined using such local trivializing sections to be (see [Reference Eyssidieux, Guedj and Zeriahi29, Section 6.2])

(96) $$ \begin{align}\begin{aligned} \mu = (i^{rn^2} \sigma\wedge \bar\sigma)^{1/r} |\sigma|_{h^r}^{-2/r} \text{ on } X^{reg}, \end{aligned} \end{align} $$

extended trivially to X. Recall that if X has klt singularities, then $\mu $ has finite total mass. Moreover, if $\pi : Y \to X$ is a resolution, and $\Omega $ is a smooth volume form on Y, then we have

(97) $$ \begin{align}\begin{aligned} \pi^*\mu = F \Omega\, \text{ on } \pi^{-1}(X^{reg}), \end{aligned} \end{align} $$

where $F \in L^p(\Omega )$ for some $p> 1$ (see [Reference Eyssidieux, Guedj and Zeriahi29, Lemma 6.4]). In our three cases $\lambda \in \{0,-1,1\}$ we can choose the metric h in such a way that the curvature of h is given by $-\lambda \omega $ for the smooth metric $\omega $ .

We define the Mabuchi K-energy, for $u\in PSH_\omega (X)\cap L^\infty $ , by

(98) $$ \begin{align}\begin{aligned} M_\omega(u) = \int_{X^{reg}} \log\left(\frac{\omega_u^n}{\mu}\right) \omega_u^n -\lambda \mathcal{J}_\omega(u). \end{aligned} \end{align} $$

The first term (the entropy) is defined to be $\infty $ , unless $\omega _u^n = f\mu $ and $f\log f$ is integrable with respect to $\mu $ . We have the following result.

Proposition 29. The functional $M_\omega $ is proper in the sense that there are constants $\delta , B> 0$ such that for all $u\in PSH_\omega (X)\cap L^\infty $ we have

(99) $$ \begin{align}\begin{aligned} M_{\omega}(u)> \delta \mathcal{J}_\omega(u) - B. \end{aligned} \end{align} $$

Proof. The case when $\lambda =1$ is well known, going back to Tian [Reference Tian56] in the smooth setting, who proved a weaker version of properness. The properness in the form (99) was shown by Phong-Song-Sturm-Weinkove [Reference Phong, Song, Sturm and Weinkove48]. In the singular setting the result was shown in Darvas [Reference Darvas21, Theorem 2.2]. Note that we are assuming that X has discrete automorphism group and admits a Kähler-Einstein metric.

The cases $\lambda =0, -1$ are much easier (see Tian [Reference Tian57] or Song-Weinkove [Reference Song and Weinkove50, Theorem 1.2] for a similar result). For this, note that $\mathcal {J}_\omega \geq 0$ , and so when $\lambda \leq 0$ , we have

(100) $$ \begin{align}\begin{aligned} M_\omega(u) \geq \frac{1}{V} \int_{X^{reg}} \log\left(\frac{\omega_u^n}{\mu}\right) \omega_u^n. \end{aligned} \end{align} $$

At the same time, using Tian [Reference Tian54], we know that there are $\alpha , C_1> 0$ such that for all $u\in PSH_\omega (X)$ with $\sup _X u=0$ we have

(101) $$ \begin{align}\begin{aligned} \int_{Y} e^{-\alpha \pi^*u}\, \Omega < C_1,\end{aligned} \end{align} $$

and so with $p^{-1} + q^{-1} =1$ (such that F in (97) is in $L^p$ ) we have

(102) $$ \begin{align}\begin{aligned} \int_{X^{reg}} e^{-\alpha q^{-1} u}\, d\mu &= \int_{\pi^{-1}(X^{reg})} e^{-\alpha q^{-1} \pi^*u}\, \pi^*\mu \\ &= \int_{\pi^{-1}(X^{reg})} e^{-\alpha q^{-1} \pi^*u }\, F \Omega \\ &\leq \left(\int_{\pi^{-1}(X^{reg})} e^{-\alpha \pi^*u}\, \Omega\right)^{1/q} \left(\int_{\pi^{-1}(X^{reg})} F^p\Omega\right)^{1/p} \\ &\leq C_2. \end{aligned} \end{align} $$

Using the convexity of the exponential function we then have, as in [Reference Song and Weinkove50, Lemma 4.1],

(103) $$ \begin{align}\begin{aligned} \int_{X^{reg}} \log\left(\frac{\omega_u^n}{\mu}\right)\, \omega_u^n \geq \alpha q^{-1}\int_{X^{reg}} (-u)\, \omega_u^n - C_3, \end{aligned} \end{align} $$

for all $u\in PSH_\omega (X)$ with $\sup _X u =0$ . As the same time, if $\sup _X u =0$ and $u\in L^\infty $ , then we have

(104) $$ \begin{align}\begin{aligned} \int_{X^{reg}} (-u)\, \omega_u^n \geq \mathcal{J}_\omega(u). \end{aligned} \end{align} $$

To see this, note that

(105) $$ \begin{align}\begin{aligned} \int_{X^{reg}} (-u)\, \omega_u^n &= \int_0^1 \frac{d}{dt} \int_{X^{reg}} (-tu)\, \omega_{tu}^n\, dt \\ &= \int_0^1 \int_{X^{reg}} (-u)\, \omega_{tu}^n - n tu \sqrt{-1}\partial \bar\partial u\wedge \omega_{tu}^{n-1}\, dt \\ &\geq \int_0^1 n \int_{X^{reg}} u (\omega - \omega_{tu}) \wedge \omega_{tu}^{n-1}\, dt \\ &= \int_0^1 \frac{d}{dt}\mathcal{J}_\omega(tu)\, dt = \mathcal{J}_\omega(u). \end{aligned} \end{align} $$

So combining the estimates above we obtain (99).

Suppose that $\pi :Y \to X$ is a projective resolution such that the anticanonical bundle $-K_Y$ is relatively nef. Let us write E for the exceptional divisor. The relatively nef assumption implies (see Boucksom-Jonsson-Trusiani [Reference Boucksom, Jonsson and Trusiani6]), that we have a smooth volume form $\Omega $ on Y, whose Ricci form $\mathrm {Ric}(\Omega )$ satisfies

(106) $$ \begin{align}\begin{aligned} \mathrm{Ric}(\Omega) \geq -C \pi^*\omega \end{aligned} \end{align} $$

for suitable $C> 0$ . Let us fix a smooth Kähler metric $\eta _Y$ on Y, with volume form $\Omega $ , and we let $\eta _\epsilon = \pi ^*\omega + \epsilon \eta _Y$ , which is a smooth Kähler metric on Y. For any closed $(1,1)$ -form $\alpha $ on Y, we define the functional $\mathcal {J}_{\eta _\epsilon , \alpha }$ on $PSH_{\eta _\epsilon }(Y)\cap L^\infty $ by letting $\mathcal {J}_{\eta _\epsilon , \alpha }(0)=0$ and its variation

(107) $$ \begin{align}\begin{aligned} \delta\mathcal{J}_{\eta_\epsilon, \alpha}(u) = n\int_Y \delta u \big(\alpha - c_\alpha\eta_{\epsilon, u}\big) \wedge \eta_{\epsilon, u}^{n-1}. \end{aligned} \end{align} $$

Here $c_\alpha $ is the constant determined by

(108) $$ \begin{align}\begin{aligned} \int_Y \big(\alpha - c_\alpha\eta_{\epsilon, u}\big) \wedge \eta_{\epsilon, u}^{n-1} = 0, \end{aligned} \end{align} $$

and $\eta _{\epsilon , u} = \eta _\epsilon + \sqrt {-1}\partial \bar \partial u$ .

We write $\mathcal {J}_{\eta _\epsilon } = \mathcal {J}_{\eta _\epsilon , \eta _\epsilon }$ , which is consistent with the earlier definition. The twisted Mabuchi K-energy in the class $[\eta _\epsilon ]$ is defined, for $u\in PSH_{\eta _\epsilon }(Y)\cap L^\infty $ by

(109) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon,s}(u) = \int_Y \log \left(\frac{\eta_{\epsilon, u}^n}{\Omega}\right)\, \eta_{\epsilon, u}^n + \mathcal{J}_{\eta_\epsilon, s\eta_\epsilon - \mathrm{Ric}(\Omega)}. \end{aligned} \end{align} $$

Note that

(110) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon,s}(u) \geq M_{\eta_\epsilon} := M_{\eta_\epsilon, 0}\end{aligned} \end{align} $$

for $s \geq 0$ . The critical points of this functional are the twisted cscK metrics $\eta _{\epsilon , u}\in [\eta _\epsilon ]$ , satisfying

(111) $$ \begin{align}\begin{aligned} R(\eta_{\epsilon, u}) - s\, \mathrm{tr}_{\eta_{\epsilon, u}} \eta_\epsilon= \mathrm{const.} \end{aligned} \end{align} $$

The following result uses our assumption that $-K_Y$ is relatively nef.

Lemma 30. Assuming that $-K_Y$ is relatively nef, there is a constant $C_2> 0$ such that $\mathcal {J}_{\eta _\epsilon , -\mathrm {Ric}(\Omega )} \geq - C_2 \mathcal {J}_{\eta _\epsilon }$ on $PSH_{\eta _\epsilon }(Y)\cap L^\infty $ . In particular there are constants $s_0, \epsilon _0> 0$ (depending on $(X, \omega _{KE})$ ) such that for $s \geq s_0$ and $\epsilon < \epsilon _0$ the twisted K-energy is proper:

(112) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon, s}(u) \geq \mathcal{J}_{\eta_\epsilon}(u), \end{aligned} \end{align} $$

for all $u\in PSH_{\eta _\epsilon }(Y)\cap L^\infty $ .

Proof. For $u\in PSH_{\eta _\epsilon }(Y)\cap L^\infty $ with $\sup _Yu = 0$ , we have

(113) $$ \begin{align}\begin{aligned} -\mathcal{J}_{\eta_\epsilon, \mathrm{Ric}(\Omega)}(u) &= n \int_0^1 \int_Y (-u) (\mathrm{Ric}(\Omega) - c \eta_{\epsilon, tu})\wedge \eta_{\epsilon, tu}^{n-1} \\ &\geq - n\int_0^1 \int_Y (-u) (C \pi^*\omega + c \eta_{\epsilon, tu})\wedge \eta_{\epsilon, tu}^{n-1} \\ &\geq - C_1 n\int_0^1 \int_Y (-u) (\eta_{\epsilon} + \eta_{\epsilon, tu})\wedge \eta_{\epsilon, tu}^{n-1} \\ &\geq -C_2 J_{\eta_\epsilon}(u). \end{aligned} \end{align} $$

Note that since the entropy term is nonnegative, we have $M_{\eta _\epsilon , s} \geq \mathcal {J}_{\eta _\epsilon , s\eta _\epsilon - \mathrm {Ric}(\Omega )} $ and also

(114) $$ \begin{align}\begin{aligned} \mathcal{J}_{\eta_\epsilon, s\eta_\epsilon - \mathrm{Ric}(\Omega)} = s \mathcal{J}_{\eta_\epsilon, \eta_\epsilon} - \mathcal{J}_{\eta_\epsilon,\mathrm{Ric}(\Omega)}. \end{aligned} \end{align} $$

It follows that for $s> C_2 + 1$ ,

(115) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon, s}(u) \geq J_{\eta_\epsilon}(u). \end{aligned} \end{align} $$

It follows from this result, using the work of Chen-Cheng [Reference Chen and Cheng15], that if $\epsilon < \epsilon _0$ and $s> s_0$ , then there exists a twisted cscK metric $\eta _{\epsilon , u} \in [\eta _{\epsilon }]$ satisfying

(116) $$ \begin{align}\begin{aligned} R(\eta_{\epsilon,u}) - s\, \mathrm{tr}_{\eta_{\epsilon,u}} \eta_\epsilon = \mathrm{const.} \end{aligned} \end{align} $$

We will use a continuity method to construct twisted cscK metrics in $[\eta _\epsilon ]$ for sufficiently small $\epsilon $ , that satisfy (116) for $s\in [0,s_0]$ , and so in particular we obtain a cscK metric in $[\eta _\epsilon ]$ . For this we will need a refinement of Chen-Cheng’s estimates, which are uniform in the degenerating cohomology classes $[\eta _\epsilon ]$ as $\epsilon \to 0$ . Such a refinement was shown by Zheng [Reference Zheng62] who worked in the more complicated setting of cscK metrics with cone singularities. See also Pan-Tô [Reference Pan and Tô47].

Note that in Zheng’s work the cscK metrics are expressed relative to metrics with a fixed volume form, rather than metrics of the form $\eta _\epsilon $ . Let us write $\tilde {\eta _\epsilon } \in [\eta _\epsilon ]$ for the metrics with $\tilde {\eta _\epsilon }^n = c_\epsilon \Omega $ provided by Yau [Reference Yau61], where the $c_\epsilon $ are bounded above and below uniformly. Note that we have $\tilde {\eta _\epsilon } = \eta _\epsilon + \sqrt {-1}\partial \bar \partial v_\epsilon $ with a uniform bound on $\sup |v_\epsilon |$ , independent of $\epsilon $ , so it does not matter whether we obtain $L^\infty $ bounds for potentials relative to $\eta _\epsilon $ or relative to $\tilde {\eta _\epsilon }$ .

In order to state the estimates in a form that we will use, we make the following definition.

Definition 31. Fix an exhaustion $K_1 \subset K_2 \subset \ldots \subset \pi ^{-1}(X^{reg})$ of $\pi ^{-1}(X^{reg})$ by compact sets. Let $a_0, a_1, \ldots $ be a sequence of positive numbers, and $p> 1$ . We say that a potential $u \in PSH_{\eta _\epsilon }(Y)$ is $\{p,a_j\}_{j\geq 0}$ -bounded, if we have

(117) $$ \begin{align}\begin{aligned} \left\Vert \frac{\eta_{\epsilon, u}^n}{\Omega}\right\Vert_{L^p(\Omega)} + \sup_Y |u| \leq a_0, \qquad \sup_{K_j} \left|\log \frac{\eta_{\epsilon, u}^n}{\Omega}\right| + \Vert u\Vert_{C^4(K_j, \eta_Y)} \leq a_j. \end{aligned} \end{align} $$

In other words such a potential is uniformly bounded globally, has volume form in $L^p$ , is locally bounded in $C^4$ , and its volume form is locally bounded above and below away from the exceptional divisor E.

We then have the following.

Proposition 32. Suppose that $\epsilon \in (0,\epsilon _0), s\in (0,s_0]$ , and $\eta _{\epsilon , u} := \eta _\epsilon + \sqrt {-1}\partial \bar \partial u$ satisfies the twisted cscK equation

(118) $$ \begin{align}\begin{aligned} R(\eta_{\epsilon, u}) - s\, \mathrm{tr}_{\eta_{\epsilon, u}} \eta_\epsilon = c_{s, \epsilon}, \end{aligned} \end{align} $$

where $c_{s,\epsilon }$ is a constant determined by $s, \epsilon $ through cohomological data. Assume that $\sup u = 0$ . Let $\phi = \log |s_E|^2$ , where $s_E$ is a section of $\mathcal {O}(E)$ vanishing along E, and we are using a smooth metric on $\mathcal {O}(E)$ to compute the norm. There are constants $C, a> 0$ , $p> 1$ , depending on $Y, \eta _Y, \eta _0, s_0$ , as well as on the entropy $\int _Y \log \left (\frac {\eta _{\epsilon , u}^n}{\Omega }\right )\, \eta _{\epsilon , u}^n$ , but not on $\epsilon , s$ , such that we have the following estimates:

  1. 1.

    (119) $$ \begin{align}\begin{aligned}\sup_Y\left( \log\frac{\eta_{\epsilon,u}^n}{\Omega} + a\phi\right) + \left\Vert \frac{\eta_{\epsilon,u}^n}{\Omega} \right\Vert_{L^p(\eta_Y)} + \sup_Y |u| < C, \end{aligned} \end{align} $$
  2. 2.

    (120) $$ \begin{align}\begin{aligned} \inf_Y \left( \log\frac{\eta_{\epsilon,u}^n}{\Omega} - a\phi\right)> C, \end{aligned} \end{align} $$
  3. 3.

    (121) $$ \begin{align}\begin{aligned} \Vert e^{a\phi}\mathrm{tr}_{\eta_Y} \eta_{\epsilon, u}\Vert_{L^q(\eta_Y)} < C, \text{ for any } q> 1. \end{aligned} \end{align} $$

In particular there exist $p> 1$ and $a_j> 0$ such that u is $\{p, a_j\}_{j\geq 0}$ -bounded.

Proof. The estimates (1) are shown in [Reference Zheng62, Proposition 5.12], the estimate (2) is in [Reference Zheng62, Proposition 5.15], and the estimate (3) is [Reference Zheng62, Proposition 5.18]. Since the notation in [Reference Zheng62] is quite different, and they consider a more general situation including conical singularities along a divisor, we recall their setup. In [Reference Zheng62, Section 5], the author considers the equations

(122) $$ \begin{align}\begin{aligned} F &= \log \frac{\eta_{\epsilon, u}^n}{\Omega}, \\ \Delta_{\eta_{\epsilon, u}} F &= \mathrm{tr}_{\eta_{\epsilon, u}} \Theta - c_{s,\epsilon}. \end{aligned} \end{align} $$

Here $\Omega $ is a smooth volume form on Y as above, with Ricci curvature $\mathrm {Ric}(\Omega ) = \theta $ , and we define $\Theta = \theta - s \eta _\epsilon $ . The coupled equations then imply

(123) $$ \begin{align}\begin{aligned} -R(\eta_{\epsilon, u}) + \mathrm{tr}_{\eta_{\epsilon, u}} \theta = \mathrm{tr}_{\eta_{\epsilon, u}} (\theta - s\eta_{\epsilon, u}) - c_{s,\epsilon}, \end{aligned} \end{align} $$

which is (118). Note that in [Reference Zheng62] the resolution is called X and the singular variety is Y, which is the opposite of our notation. There is also an additional function f which we take to be zero. Zheng considers a semipositive form $\omega _{sr}$ on Y, which we can take to be $\pi ^*\omega $ , and a Kähler form $\omega _K$ on Y, which we take to be $\eta _Y$ , so $\eta _\epsilon = \pi ^*\omega + \epsilon \eta _Y$ is what Zheng calls $\omega _\epsilon $ .

In order to deal with the degeneracy of $\eta _\epsilon $ as $\epsilon \to 0$ , Zheng uses the technique of Tsuji [Reference Tsuji58], relying on the fact that if we choose a suitable smooth metric on the line bundle $\mathcal {O}(E)$ for a divisor supported on the exceptional set of $\pi $ , then for any $a> 0$ the current $\pi ^*\omega + a \sqrt {-1}\partial \bar \partial \log |s_E|^2$ dominates a Kähler form on Y, where $s_E$ vanishes along E. We can assume that on $Y\setminus E$ we have

(124) $$ \begin{align}\begin{aligned} \eta_Y = \pi^*\omega + a \sqrt{-1}\partial \bar\partial \log |s_E|^2. \end{aligned} \end{align} $$

We apply [Reference Zheng62, Proposition 5.12], to deduce the estimates (1). For this we need to check the condition that $e^{-\phi _l} \in L^{p_0}$ for some $p_0> 1$ , where $\phi _l$ in Zheng’s notation is defined in his Lemma 5.6. Since in that Lemma $\phi _{\theta _\epsilon }$ is uniformly bounded, it is enough to check the integrability of $e^{-p_0 \tilde {\phi }_l}$ where we define

(125) $$ \begin{align}\begin{aligned} \tilde{\phi}_l = (\inf_{(X, \eta_Y)}\Theta) (- a\log |s_E|^2). \end{aligned} \end{align} $$

We claim that the infimum $\inf _{(X, \eta _Y)}\Theta $ is bounded below, independently of the choice of small $a> 0$ (note that the choice of a affects the definition of $\eta _Y$ and so also $\Theta $ ). To see this, we use the condition $\theta = \mathrm {Ric}(\Omega ) \geq - C\pi ^*\omega $ , so that we have

(126) $$ \begin{align}\begin{aligned} \Theta &= \theta - s\eta_\epsilon \geq -(C + s)\pi^*\omega - s\epsilon \eta_Y \\ &= - C' \eta_Y, \end{aligned} \end{align} $$

for $C'$ depending on C and $s_0$ . Here we also used that if $a> 0$ is sufficiently small, then $\eta _Y> \frac {1}{2}\pi ^*\omega _X$ . It follows from this that

(127) $$ \begin{align}\begin{aligned} - \tilde{\phi}_l \leq - C'a \log |s_E|^2, \end{aligned} \end{align} $$

and so if $a> 0$ is sufficiently small, then $e^{- \tilde {\phi }_l} \in L^{p_0}$ for $p_0> 1$ , as required in Zheng’s Proposition 5.12. The conclusion is the estimates (1). Note that the quantities that the estimate in Proposition 5.12 depends on are all uniformly bounded in $s, \epsilon $ in our setting. Similarly, Propositions 5.15 and 5.18 imply the estimates (2) and (3).

Note that the $L^p$ -bound on the trace of $\eta _{\epsilon , u}$ implies higher order estimates for u on compact sets away from E, using Chen-Cheng’s local estimate [Reference Chen and Cheng15, Proposition 6.1]. This leads to the $\{p, a_j\}$ -boundedness of u. See also [Reference Pan and Tô47, Theorem C] for similar estimates.

Next we show that by Proposition 29, the Mabuchi energy $M_{\eta _\epsilon }$ is proper on $\{p, a_j\}$ -bounded classes of potentials, when $\epsilon $ is sufficiently small.

Proposition 33. Given $p> 1$ and a sequence $\{a_j\}_{j \geq 0}$ , let $V \subset PSH_{\eta _\epsilon }(Y)$ denote the $\{p, a_j\}_{j\geq 0}$ -bounded potentials. Then for sufficiently small $\epsilon $ , depending on the $p, a_j$ , the K-energy $M_{\eta _\epsilon }$ is proper on V in the sense that

(128) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon}(u)> \delta \mathcal{J}_{\eta_\epsilon}(u)- B_2, \text{ for all } u\in V. \end{aligned} \end{align} $$

Here $\delta $ is the same constant as in Proposition 29, while $B_2$ is a constant depending on $(X, \omega )$ and $\Omega $ , but not on the $p, a_j$ .

Proof. We argue by contradiction. Suppose that we have a sequence $\epsilon _i \to 0$ , and $u_i \in PSH_{\eta _{\epsilon _i}}(Y)$ that are $\{p, a_j\}_{j\geq 0}$ -bounded, such that

(129) $$ \begin{align}\begin{aligned} M_{\eta_{\epsilon_i}}(u_i) \leq \delta \mathcal{J}_{\eta_\epsilon}(u) - B_2, \end{aligned} \end{align} $$

for $B_2$ to be determined below. Up to choosing a subsequence we can assume that $u_i \to u_\infty $ in $L^1$ and also in $C^{3,\alpha }$ on compact sets away from the exceptional divisor E. We have $u_\infty \in PSH_{\pi ^*\omega }(Y)$ , and we have an identification $PSH_{\pi ^*\omega }(Y) = PSH_{\omega }(X)$ . We will next show that in terms of F in (97) we have

(130) $$ \begin{align}\begin{aligned} M_{\eta_{\epsilon_i}}(u_i) &\to M_{\omega}(u_\infty) + \int_Y \log F\, \eta_0^n, \\ \mathcal{J}_{\eta_{\epsilon_i}}(u_i) &\to \mathcal{J}_\omega(u_\infty). \end{aligned} \end{align} $$

Let us first consider the relevant entropy terms. Note that

(131) $$ \begin{align}\begin{aligned} \int_Y \log \left(\frac{\eta_{{\epsilon_i}, u_i}^n}{\Omega}\right) \eta_{\epsilon_i, u_i}^n = \int_Y \log\left(\frac{ \eta^n_{\epsilon_i, u_i}}{\Omega}\right) \frac{\eta_{\epsilon_i, u_i}^n}{\Omega}\, \Omega. \end{aligned} \end{align} $$

Our assumptions mean that the integrand has a uniform $L^p(\Omega )$ -bound for some $p> 1$ . Using this, and the $C^{3,\alpha }$ -convergence $u_i\to u_\infty $ on compact sets away from E, it follows that

(132) $$ \begin{align}\begin{aligned} \int_Y \log \left(\frac{\eta_{{\epsilon_i}, u_i}^n}{\Omega}\right) \eta_{\epsilon_i, u_i}^n &\to \int_Y \log \left(\frac{\eta_{0, u_\infty}^n}{\Omega}\right) \eta_{0, u_\infty}^n. \end{aligned} \end{align} $$

Using (97) we have

(133) $$ \begin{align}\begin{aligned} \int_Y \log \left(\frac{\eta_{0,u_\infty}^n}{\Omega}\right) \eta_{0, u_\infty}^n &= \int_{X^{reg}} \log \left(\frac{\omega_{u_\infty}^n}{\mu}\right) \omega_{u_\infty}^n + \int_Y \log F\, \eta_{0,u_\infty}^n. \end{aligned} \end{align} $$

The last term can be computed by writing

(134) $$ \begin{align}\begin{aligned} \int_Y \log F\, \eta_{0,u_\infty}^n &= \int_Y \log F\, \eta_0^n + \int_0^1 \frac{d}{dt} \int_Y \log F\, \eta_{0, tu_\infty}^n\, dt \\ &= \int_Y \log F\, \eta_0^n + \int_0^1 n \int_Y u_\infty \sqrt{-1}\partial \bar\partial \log F\, \wedge \eta_{0,t u_\infty}^{n-1}\,dt \\ &= \int_Y \log F\, \eta_0^n + \int_0^1 n \int_Y u_\infty (\mathrm{Ric}(\Omega) - \mathrm{Ric}(\pi^*\mu))\wedge \eta_{0,u_\infty}^{n-1}\,dt \\ &= \int_Y \log F\, \eta_0^n + \mathcal{J}_{\eta_0, \mathrm{Ric}(\Omega)}(u_\infty) -\lambda \mathcal{J}_{\omega}(u_\infty). \end{aligned} \end{align} $$

For the last step note that $\eta _0$ vanishes along E, so although $\mathrm {Ric}(\pi ^*\mu )$ has current contributions along E, the only part that survives in the integral is $\mathrm {Ric}(\mu ) = \lambda \omega $ on X. In conclusion we have that

(135) $$ \begin{align}\begin{aligned} \int_Y \log \left(\frac{\eta_{{\epsilon_i}, u_i}^n}{\Omega}\right) \eta_{\epsilon_i, u_i}^n &\to \int_{X^{reg}} \log\left(\frac{\omega^n_{u_\infty}}{\mu}\right)\, \omega_{u_\infty}^n + \int_Y \log F\, \eta_0^n \\ &\qquad + \mathcal{J}_{\eta_0, \mathrm{Ric}(\Omega)}(u_\infty) - \lambda \mathcal{J}_\omega(u_\infty). \end{aligned} \end{align} $$

Next we consider the $\mathcal {J}$ -functional terms. Consider a general smooth, closed (1,1)-form $\alpha $ on Y. We claim that we have $\mathcal {J}_{\eta _{\epsilon _i}, \alpha }(u_i) \to \mathcal {J}_{\eta _0, \alpha }(u_\infty )$ . Using the variational definition of $\mathcal {J}$ , the local $C^{3,\alpha }$ -convergence, and the uniform $L^\infty $ -bound for the $u_i$ , it is enough to show that for every $\kappa> 0$ there is a compact set $K\subset Y\setminus E$ , such that

(136) $$ \begin{align}\begin{aligned} \int_{Y\setminus K} \eta_1 \wedge \eta_{\epsilon_i, u_i}^{n-1} + \int_{Y\setminus K} \eta_{\epsilon_i, u_i}^n < \kappa, \text{ for all } i. \end{aligned} \end{align} $$

To see this, let $h= -\log |s_E|^2$ , where $s_E$ is a section of the line bundle $\mathcal {O}(E)$ over Y vanishing along the exceptional divisor E, and we use a smooth metric on $\mathcal {O}(E)$ . We have

(137) $$ \begin{align}\begin{aligned} \sqrt{-1}\partial \bar\partial h = \chi - [E], \end{aligned} \end{align} $$

where $\chi $ is a smooth form on Y. We can assume that $h\geq 0$ , and note that $h\to \infty $ along E. We show by induction that for each $k=0, \ldots , n$ there is a constant $C_k> 0$ , independent of i, such that

(138) $$ \begin{align}\begin{aligned} \int_Y h \eta_1^{n-k}\wedge \eta_{\epsilon_i, u_i}^k \leq C_k. \end{aligned} \end{align} $$

For $k=0$ this is clear since h has logarithmic singularities. Suppose that the bound has been established for a value of k. Then

(139) $$ \begin{align}\begin{aligned} \int_Y h \eta_1^{n-k-1}\wedge \eta_{\epsilon_i, u_i}^{k+1} &= \int_Y h \eta_1^{n-k-1} \wedge (\eta_{\epsilon_i} + \sqrt{-1}\partial \bar\partial u_i)\wedge \eta_{\epsilon_i, u_i}^k \\ &= \int_Y h \eta_{1}^{n-k-1}\wedge \eta_{\epsilon_i}\wedge \eta_{\epsilon_i, u_i}^k + \int u_i\sqrt{-1}\partial \bar\partial h \wedge \eta_1^{n-k-1}\wedge \eta_{\epsilon_i, u_i}^k \\ &\leq \int_Y h \eta_{1}^{n-k}\wedge \eta_{\epsilon_i, u_i}^k + \int_Y u_i \chi\wedge \eta_1^{n-k-1}\wedge \eta_{\epsilon_i, u_i}^k - \int_E u_i \eta_1^{n-k-1}\wedge \eta_{\epsilon_i, u_i}^k \\ &\leq C_k(1+C) - \int_E u_i \eta_1^{n-k-1}\wedge \eta_{\epsilon_i, u_i}^k \\ &\leq C_k(1+ C) + C', \end{aligned} \end{align} $$

where $C, C'$ depend on $\chi $ and the uniform $L^\infty $ bound for $u_i$ .

Since $h\to \infty $ along E, it follows from (138) that for any $\kappa> 0$ we can find a compact set $K\subset Y\setminus E$ such that (136) holds. It follows that

(140) $$ \begin{align}\begin{aligned} \mathcal{J}_{\eta_{\epsilon_i, -\mathrm{Ric}(\Omega)}}(u_i) \to \mathcal{J}_{\eta_{0, -\mathrm{Ric}(\Omega)}}(u_\infty), \end{aligned} \end{align} $$

and also

(141) $$ \begin{align}\begin{aligned} \mathcal{J}_{\eta_{\epsilon_i}}(u_i) \to \mathcal{J}_{\omega}(u_\infty). \end{aligned} \end{align} $$

From this, together with (135), we have

(142) $$ \begin{align}\begin{aligned} M_{\eta_{\epsilon_i}}(u_i) &\to \int_{X^{reg}} \log\left( \frac{\omega_{u_\infty}^n}{\mu}\right)\, \omega_{u_\infty}^n -\lambda \mathcal{J}_\omega(u_\infty) + \int_Y \log F\, \eta_0^n \\ &= M_{\omega}(u_\infty) + \int_Y \log F\, \eta_0^n. \end{aligned} \end{align} $$

From (129) we therefore get

(143) $$ \begin{align}\begin{aligned} M_{\omega}(u_\infty) + \int_Y \log F\, \eta_0^n \leq \delta\mathcal{J}_\omega(u_\infty) - B_2. \end{aligned} \end{align} $$

Choosing $B_2 = B - \int _Y \log F\, \eta _0^n$ for the B in Proposition 29, we get a contradiction.

We are now ready to combine the different ingredients to prove the main result of this section.

Proof of Theorem 3.

We will choose suitable $p> 0, a_j > 0$ shortly. By Proposition 33, for a given $p, a_j$ we have some $\epsilon _1> 0$ such that once $\epsilon < \epsilon _1$ and for any $s\geq 0$ , we have

(144) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon, s}(u) \geq M_{\eta_\epsilon}(u)> \delta \mathcal{J}_{\eta_\epsilon}(u)- B_2, \end{aligned} \end{align} $$

for $\{p. a_j\}$ -bounded potentials u. Recall that $\delta , B_2$ do not depend on $\{p, a_j\}$ . For small $\kappa> 0$ we have

(145) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon, s}(u) &\geq \kappa \int_Y \left(\frac{\eta_{\epsilon, u}^n}{\Omega}\right)\, \eta_{\epsilon, u}^n + \kappa \mathcal{J}_{\eta_\epsilon, s\eta_\epsilon-\mathrm{Ric}(\Omega)}(u) + (1-\kappa)\delta \mathcal{J}_{\eta_\epsilon}(u) - (1-\kappa)B_2 \\ &= \kappa \int_Y \left(\frac{\eta_{\epsilon, u}^n}{\Omega}\right)\, \eta_{\epsilon, u}^n + (\kappa s + (1-\kappa)\delta) \mathcal{J}_{\eta_\epsilon}(u) + \kappa \mathcal{J}_{\eta_\epsilon, -\mathrm{Ric}(\Omega)}(u) - (1-\kappa)B_2. \end{aligned} \end{align} $$

If $\kappa $ is chosen sufficiently small (depending on $\delta $ ), then by Lemma 30 we find that

(146) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon, s}(u) &\geq \kappa \int_Y \log \left(\frac{\eta_{\epsilon, u}^n}{\Omega}\right)\, \eta_{\epsilon, u}^n - B_2. \end{aligned} \end{align} $$

We also have

(147) $$ \begin{align}\begin{aligned} M_{\eta_\epsilon, s}(0) = \int_Y \log \left(\frac{\eta_{\epsilon}^n}{\Omega}\right)\, \eta_{\epsilon}^n < C_3, \end{aligned} \end{align} $$

for a constant $C_3> 0$ independent of $\epsilon $ . Since twisted cscK metrics minimize the twisted Mabuchi K-energy, it follows that if $\eta _{\epsilon , u} \in [\eta _\epsilon ]$ is a twisted cscK metric, then we have $M_{\eta _\epsilon , s}(u) < C_3$ . From (146) we get

(148) $$ \begin{align}\begin{aligned} \int_Y \log \left(\frac{\eta_{\epsilon, u}^n}{\Omega}\right)\, \eta_{\epsilon, u}^n \leq \kappa^{-1}(C_3 + B_2), \end{aligned} \end{align} $$

and in particular the entropy of $\eta _{\epsilon , u}$ is bounded independently of $\epsilon $ . We apply Proposition 32. As long as $s\leq s_0$ , for the $s_0$ determined by Lemma 30, we find that if $\eta _{\epsilon , u} = \eta _\epsilon + \sqrt {-1}\partial \bar \partial u$ is a solution of the twisted cscK equation

(149) $$ \begin{align}\begin{aligned} R(\eta_{\epsilon, u}) - s\,\mathrm{tr}_{\eta_{\epsilon, u}} \eta_\epsilon = \mathrm{const.}, \end{aligned} \end{align} $$

then u is $\{p, a_j\}$ -bounded, for suitable $p, a_j$ , determined by $s_0$ and the entropy bound (148). From now we fix this choice of $p, a_j$ .

We can now use a continuity method to show that if $\epsilon < \epsilon _1$ , for the $\epsilon _1$ determined by $\{p, a_j\}$ , for all $s\in [0,s_0]$ we can solve the twisted cscK equation (149). To see this, let us fix $\epsilon < \epsilon _1$ , and set

(150) $$ \begin{align}\begin{aligned} S = \{ s\in [0,s_0]\, :\, \text{ the equation (149) has a solution}\}. \end{aligned} \end{align} $$

We have $s_0\in S$ , and it follows from the implicit function theorem that S is open. To see that it is closed, note that the twisted cscK metrics for $s\in S$ automatically satisfy the entropy bound (148). Using the main estimates of Chen-Cheng [Reference Chen and Cheng15], we find that the potentials of the corresponding twisted cscK metrics satisfy a priori $C^k$ -estimates, and the metrics are bounded below uniformly (these estimates depend on $\epsilon $ , but now $\epsilon $ is fixed). It follows that S is closed.

It follows that for sufficiently small $\epsilon> 0$ the classes $[\eta _\epsilon ]$ on Y admit cscK metrics. The estimates required by Definition 2 follow from Proposition 32.

Remark 34. To conclude this section we give an example where the assumption that $-K_Y$ is relatively nef is satisfied. Let M be a smooth Fano manifold, and suppose that P is a line bunde over M such that $P^r = -K_M$ for some $r> 0$ . We let V denote the total space of $P^{-1}$ , with the zero section blown down to a point o. Suppose that X has one isolated singularity p, and a neighborhood of p is isomorphic to the neighborhood of $o\in V$ . In this case we can consider a resolution $\pi : Y\to X$ , obtained by blowing up the singular point. Then

(151) $$ \begin{align}\begin{aligned} K_Y = \pi^*K_X + rE, \end{aligned} \end{align} $$

where the exceptional divisor E isomorphic to M, and is in particular irreducible. It follows that in this case $-K_Y$ is relatively nef (in fact relatively ample). Note that this family of examples does not fit into the framework of admissible singularities studied by Li-Tian-Wang [Reference Li, Tian and Wang40].

6 Partial $C^0$ -estimate

An important result of Donaldson-Sun [Reference Donaldson and Sun27] is the partial $C^0$ -estimate for smooth Kähler-Einstein manifolds, conjectured by Tian [Reference Tian55]. More precisely, suppose that $(X, \omega _{KE})$ is a smooth Kähler-Einstein manifold, with $\omega _{KE}\in c_1(L)$ for an ample line bundle, and such that for some constant $D> 0$ we have

  1. 1. noncollapsing: $\mathrm {vol}\, B_{\omega _{KE}}(p, 1)> D^{-1}$ for a basepoint $p\in X$ ,

  2. 2. bounded volume: $\mathrm {vol} (X, \omega _{KE}) < D$ ,

  3. 3. bounded Ricci curvature: $\mathrm {Ric}(\omega _{KE}) = \lambda \omega _{KE}$ for $|\lambda | < D$ .

For any integer $k> 0$ the density of states function $\rho _{k, \omega _{KE}}$ is defined by

(152) $$ \begin{align}\begin{aligned} \rho_{k, \omega_{KE}}(x) = \sum_j |s_j|^2(x), \end{aligned} \end{align} $$

where the $s_j$ form an $L^2$ -orthonormal basis of $H^0(X, L^k)$ in terms of the metric induced by $k\omega _{KE}$ . Then, by Donaldson-Sun [Reference Donaldson and Sun27], there is a power $k_0 = k_0(n, D)$ , and $b=b(n,D)> 0$ , depending on the dimension and the constant D, such that $\rho _{k_0, \omega _{KE}}> b$ . In this section we show the following extension of this result to singular Kähler-Einstein spaces that admit good cscK approximations.

Theorem 35. Given $n, D> 0$ there are constants $k_0(n,D), b(n,D)> 0$ with the following property. Suppose that $(X,\omega _{KE})$ is a singular Kähler-Einstein variety of dimension n, such that $\omega _{KE}\in c_1(L)$ for a line bundle L. Assume that $(X, \omega _{KE})$ can be approximated by cscK metrics, and in addition the conditions (1), (2), (3) above hold. Then the corresponding density of states function satisfies $\rho _{k, \omega _{KE}}> b$ .

The proof of the result follows the same strategy as Donaldson-Sun [Reference Donaldson and Sun27], arguing by contradiction. We suppose that the sequence $(X_i, \omega _{KE, i})$ satisfies the bounds (1)–(3), but no fixed power $L_i^k$ of the corresponding line bundles is very ample. The corresponding metric completions $\hat {X}_i$ are noncollapsed RCD spaces by Proposition 15, and we can pass to the Gromov-Hausdorff limit $\hat {X}_\infty $ along a subsequence. We would then like to use the structure of the tangent cones of $\hat {X}_\infty $ to construct suitable holomorphic sections of a suitable power $L_i^k$ for large i, leading to a contradiction.

The difficulty in executing this strategy is that we do not have good control of the convergence of $\hat {X}_i$ to $\hat {X}_\infty $ on the regular set of $\hat {X}_\infty $ , because in Corollary 25 the constant $\epsilon $ depends on the singular Kähler-Einstein space X that we are considering. As such it is a priori possible that the singular set of $\hat {X}_\infty $ , consisting of points where the tangent cone is not given by $\mathbb {R}^{2n}$ , is dense. In order to rule this out, we prove the following. Note that recently this result was shown in the more general algebraic setting by Xu-Zhuang [Reference Xu and Zhuang59] (see also Liu-Xu [Reference Liu and Xu43] for the three-dimensional case).

Theorem 36. There is an $\epsilon> 0$ , depending only on the dimension n, with the following property. Suppose that $\hat {X}$ is the metric completion of a singular Kähler-Einstein space as in Theorem 17, that is, one that can be approximated by cscK metrics. Let $(\hat {X}_p, o)$ be a tangent cone of $\hat {X}$ , such that $\hat {X}_p \not = \mathbb {R}^{2n}$ . Then

(153) $$ \begin{align}\begin{aligned} \mathrm{vol} B(o, 1) < \omega_{2n} - \epsilon, \end{aligned} \end{align} $$

where $\omega _{2n}$ is the volume of the Euclidean unit ball in $\mathbb {R}^{2n}$ .

Proof. We will argue by contradiction. If the stated result is not true, then we can find a sequence $\hat {X}_i$ , and a sequence of singular points $p_i\in \hat {X}_i$ with tangent cones $V_{p_i}$ such that $V_{p_i}\to \mathbb {R}^{2n}$ in the pointed Gromov-Hausdorff sense.

We will prove a more general statement about almost smooth metric measure spaces in the sense of Definition 5, of any dimension, which satisfy the following conditions.

Definition 37. We say that an almost smooth metric measure space V satisfies Condition ( $\ast $ ) if the following conditions hold:

  1. 1. For some $\epsilon> 0$ (possibly depending on V), the $\epsilon $ -regular set $\mathcal {R}_\epsilon \subset V$ , defined by (90), can be chosen to be the set $\Omega $ in Definition 5.

  2. 2. The Riemannian metric on $\Omega $ is Ricci flat.

  3. 3. If a tangent cone $V'$ of V is of the form $C(S^1_\gamma )\times \mathbb {R}^{2n-2}$ , then $V' = \mathbb {R}^{2n}$ .

Note that by Propositions 24 and 27, the (iterated) tangent cones of the spaces $\hat {X}_i$ satisfy Condition ( $\ast $ ). Moreover, if a space $V = W \times \mathbb {R}^j$ satisfies Condition ( $\ast $ ), then so does W, and so do the tangent cones of V.

We argue by induction on the dimension to show that if a sequence of k-dimensional cones $V_j$ satisfies Condition ( $\ast $ ), and $V_j \to \mathbb {R}^k$ in the pointed Gromov-Hausdorff sense, then $V_j = \mathbb {R}^k$ for sufficiently large i. For $k=2$ this follows directly from Condition ( $\ast $ ).

Assuming $k> 2$ , suppose first that for all sufficiently large j the cones $V_j$ have smooth link (i.e., the singular set consists of only the vertex). In this case $V_j = C(Y_j)$ , where the $(Y_j, h_j)$ are $(k-1)$ -dimensional smooth Einstein manifolds satisfying $\mathrm {Ric}(h_j) = (k-2)h_j$ . Moreover the $(Y_j, h_j)$ converge in the Gromov-Hausdorff sense to the unit $(k-1)$ -sphere. As long as $k-1> 1$ , it follows that for sufficiently large j we have $\mathrm {vol}(Y_j, h_j) = \mathrm {vol}(S^{k-1}, g_{S^{k-1}})$ , using that Einstein metrics are critical points of the Einstein-Hilbert action. The Bishop-Gromov comparison theorem then implies that in fact $(Y_j, h_j)$ is isometric to the unit $(k-1)$ -sphere for sufficiently large j, so that $V_j = \mathbb {R}^{k}$ . If $k-1=1$ , then $V_j$ is a cone over a circle, so by Condition ( $\ast $ ) we have $V_j = \mathbb {R}^2$ . Either way, we have a contradiction.

We can therefore assume, up to choosing a subsequence, that the $V_j$ all have singularities $q_j$ away from the vertex. By taking tangent cones at the $q_j$ , we obtain a new sequence of cones, $V_j'$ , which still satisfy the Condition ( $\ast $ ), they converge to $\mathbb {R}^k$ , and they all split off an isometric factor of $\mathbb {R}$ , that is, $V_j' = W_j \times \mathbb {R}$ . The cones $W_j$ are then $k-1$ dimensional, they also satisfy Condition ( $\ast $ ), and $W_j \to \mathbb {R}^{k-1}$ . We can then apply the inductive hypothesis. It follows that $W_j = \mathbb {R}^{k-1}$ for large j, so $V_j'= \mathbb {R}^k$ , contradicting that the $q_j$ are singular points.

Given this result, we can follow the argument of Donaldson-Sun [Reference Donaldson and Sun27] to prove Theorem 35.

Proof of Theorem 35.

We argue by contradiction. Suppose that there are singular Kähler-Einstein spaces $(X_i, \omega _{KE, i})$ , that can be approximated by cscK metrics, with $\omega _{KE,i}\in c_1(L_i)$ , satisfying the conditions (1)–(3) before the statement of Theorem 35, but such that there is no fixed power $L_i^k$ of the line bundles $L_i$ whose density of states functions are bounded away from zero uniformly. Up to choosing a subsequence, we can assume that the corresponding RCD spaces $\hat {X}_i$ converge to $\hat {X}_\infty $ in the Gromov-Hausdorff sense. Theorem 36 implies that for some $\epsilon> 0$ , the $\epsilon $ -regular subset of $\hat {X}_\infty $ coincides with the regular set $\mathcal {R}\subset \hat {X}_\infty $ (given by the points with tangent cone $\mathbb {R}^{2n}$ ). Therefore the set $\mathcal {R}$ is open, and by Theorem 36 together with Proposition 24, it follows that the convergence $\hat {X}_i\to \hat {X}_\infty $ is locally smooth on $\mathcal {R}$ . In addition, using the argument in Proposition 27, we know that no iterated tangent cone of $\hat {X}_\infty $ is given by $C(S^1_\gamma )\times \mathbb {R}^{2n-2}$ with $\gamma < 2\pi $ . This means that we are in essentially the same setting as Donaldson-Sun [Reference Donaldson and Sun27], and can closely follow their arguments to show that there is a $k_0> 0$ , such that the density of states functions of the sections of $L_i^{k_0}$ are bounded away from zero for all sufficiently large i.

Acknowledgements

I would like to thank Aaron Naber, Max Hallgren, Yuchen Liu, Tamás Darvas, Valentino Tosatti, Jian Song, Yuji Odaka, Mattias Jonsson, Sebastien Boucksom, and Antonio Trusiani for helpful discussions. In addition I’m grateful to Chung-Ming Pan and Tat Dat Tô for sharing their preprint [Reference Pan and Tô47]. This work was supported in part by NSF grant DMS-2203218.

Competing interests

The authors have no competing interests to declare.

References

Ambrosio, L., Gigli, N. and Savaré, G., ‘Bakry–Émery curvature-dimension condition and Riemannian Ricci curvature bounds’, Ann. Probab. 43(1) (2015), 339404.10.1214/14-AOP907CrossRefGoogle Scholar
Anderson, M. T., ‘Convergence and rigidity of manifolds under Ricci curvature bounds’, Invent. Math. 102(2) (1990), 429445.10.1007/BF01233434CrossRefGoogle Scholar
Berman, R., ‘K-polystability of Q-Fano varieties admitting Kähler–Einstein metrics’, Invent. Math. 203(3) (2016), 9731025.10.1007/s00222-015-0607-7CrossRefGoogle Scholar
Berman, R., Boucksom, S., Eyssidieux, P., Guedj, V. and Zeriahi, A., ‘Kähler–Einstein metrics and the Kähler–Ricci flow on log Fano varieties’, J. Reine Angew. Math. 751 (2019), 2789.10.1515/crelle-2016-0033CrossRefGoogle Scholar
Berman, R., Boucksom, S., Guedj, V. and Zeriahi, A., ‘A variational approach to complex Monge–Ampère equations’, Publ. Math. Inst. Hautes Études Sci. 117 (2013), 179245.10.1007/s10240-012-0046-6CrossRefGoogle Scholar
Boucksom, S., Jonsson, M. and Trusiani, A., ‘Weighted extremal Kähler metrics on resolutions of singularities’, preprint (2024), arXiv:2412.06096.Google Scholar
Branson, T., ‘Kato constants in Riemannian geometry’, Math. Res. Lett. 7(2–3) (2000), 245261.10.4310/MRL.2000.v7.n3.a1CrossRefGoogle Scholar
Bruè, E., Naber, A. and Semola, D., ‘Boundary regularity and stability for spaces with Ricci bounded below’, Invent. Math. 228(2) (2022), 777891.10.1007/s00222-021-01092-8CrossRefGoogle Scholar
Caffarelli, L. A. and Friedman, A., ‘Partial regularity of the zero-set of solutions of linear and superlinear elliptic equations’, J. Differential Equations 60(3) (1985), 420433.10.1016/0022-0396(85)90133-0CrossRefGoogle Scholar
Calabi, E., ‘Extremal Kähler metrics’, Seminar on Differential Geometry, Ann. of Math. Stud. 102 (1982), 259290.Google Scholar
Calderbank, D. M. J., Gauduchon, P. and Herzlich, M., ‘Refined Kato inequalities and conformal weights in Riemannian geometry’, J. Funct. Anal. 173(1) (2000), 214255.10.1006/jfan.2000.3563CrossRefGoogle Scholar
Cheeger, J. and Colding, T. H., ‘On the structure of spaces with Ricci curvature bounded below I’, J. Differential Geom. 46(3) (1997), 406480.10.4310/jdg/1214459974CrossRefGoogle Scholar
Cheeger, J., Colding, T. H. and Tian, G., ‘On the singularities of spaces with bounded Ricci curvature’, Geom. Funct. Anal. 12(5) (2002), 873914.10.1007/PL00012649CrossRefGoogle Scholar
Cheeger, J., Jiang, W. and Naber, A., ‘Rectifiability of singular sets of noncollapsed limit spaces with Ricci curvature bounded below’, Ann. of Math. (2) 193(2) (2021), 407538.10.4007/annals.2021.193.2.2CrossRefGoogle Scholar
Chen, X. and Cheng, J., ‘On the constant scalar curvature Kähler metrics (I)—A priori estimates’, J. Amer. Math. Soc. 34(4) (2021), 909936.10.1090/jams/967CrossRefGoogle Scholar
Chen, X., Donaldson, S. and Sun, S., ‘Kähler–Einstein metrics on Fano manifolds. II: Limits with cone angle less than 2π’, J. Amer. Math. Soc. 28(1) (2015), 199234.10.1090/S0894-0347-2014-00800-6CrossRefGoogle Scholar
Chen, X., Donaldson, S. and Sun, S., ‘Kähler–Einstein metrics on Fano manifolds. III: Limits as cone angle approaches 2π and completion of the main proof’, J. Amer. Math. Soc. 28(1) (2015), 235278.10.1090/S0894-0347-2014-00801-8CrossRefGoogle Scholar
Chiu, S.-K. and Székelyhidi, G., ‘Higher regularity for singular Kähler–Einstein metrics’, Duke Math. J. 172(18) (2023), 35213558.10.1215/00127094-2022-0107CrossRefGoogle Scholar
Chu, J. and Jiang, W., ‘A note on nodal sets on manifolds with lower Ricci bounds’, Methods Appl. Anal. 27(4) (2020), 359374.10.4310/MAA.2020.v27.n4.a3CrossRefGoogle Scholar
Cibotaru, D. and Zhu, P., ‘Refined Kato inequalities for harmonic fields on Kähler manifolds’, Pacific J. Math. 256(1) (2012), 5166.10.2140/pjm.2012.256.51CrossRefGoogle Scholar
Darvas, T., ‘Metric geometry of normal Kähler spaces, energy properness, and existence of canonical metrics’, Int. Math. Res. Not. IMRN (2017), no. 22, 67526777.Google Scholar
De Philippis, G. and Gigli, N., ‘From volume cone to metric cone in the nonsmooth setting’, Geom. Funct. Anal. 26(6) (2016), 15261587.10.1007/s00039-016-0391-6CrossRefGoogle Scholar
De Philippis, G. and Gigli, N., ‘Non-collapsed spaces with Ricci curvature bounded from below’, J. Éc. polytech. Math. 5 (2018), 613650.10.5802/jep.80CrossRefGoogle Scholar
Demailly, J.-P., ‘Estimations L2 pour l’opérateur $\overline{\partial}$ d’un fibré vectoriel holomorphe semi-positif au-dessus d’une variété kählérienne complète’, Ann. Sci. École Norm. Sup. (4) 15(3) (1982), 457511.10.24033/asens.1434CrossRefGoogle Scholar
Demailly, J.-P., Complex analytic and differential geometry. (1997), Université de Grenoble I, Grenoble.Google Scholar
Ding, Y., ‘The gradient of certain harmonic functions on manifolds of almost nonnegative Ricci curvature’, Israel J. Math. 129 (2002), 241251.10.1007/BF02773166CrossRefGoogle Scholar
Donaldson, S. and Sun, S., ‘Gromov–Hausdorff limits of Kähler manifolds and algebraic geometry’, Acta Math. 213(1) (2014), 63106.10.1007/s11511-014-0116-3CrossRefGoogle Scholar
Donaldson, S. and Sun, S., ‘Gromov–Hausdorff limits of Kähler manifolds and algebraic geometry, II’, J. Differential Geom. 107(2) (2017), 327371.10.4310/jdg/1506650422CrossRefGoogle Scholar
Eyssidieux, P., Guedj, V. and Zeriahi, A., ‘Singular Kähler–Einstein metrics’, J. Amer. Math. Soc. 22(3) (2009), 607639.10.1090/S0894-0347-09-00629-8CrossRefGoogle Scholar
Greb, D., Guenancia, H. and Kebekus, S., ‘Klt varieties with trivial canonical class: holonomy, differential forms, and fundamental groups’, Geom. Topol. 23(4) (2019), 20512124.10.2140/gt.2019.23.2051CrossRefGoogle Scholar
Guedj, V., Guenancia, H. and Zeriahi, A., ‘Strict positivity of Kähler–Einstein currents’, preprint (2023), arXiv:2305.12422.Google Scholar
Guo, B., Phong, D. H., Song, J. and Sturm, J., ‘Diameter estimates in Kähler geometry’, preprint (2022), arXiv:2209.09428.Google Scholar
Guo, B., Phong, D. H., Song, J. and Sturm, J., ‘Sobolev inequalities on Kähler spaces’, preprint (2023), arXiv:2311.00221.Google Scholar
Han, Q. and Lin, F.-H., ‘On the geometric measure of nodal sets of solutions’, J. Partial Differential Equations 7(2) (1994), 111131.Google Scholar
Hein, H.-J. and Sun, S., ‘Calabi–Yau manifolds with isolated conical singularities’, Publ. Math. Inst. Hautes Études Sci. 126 (2017), 73130.10.1007/s10240-017-0092-1CrossRefGoogle Scholar
Honda, S., ‘Bakry–Émery conditions on almost smooth metric measure spaces’, Anal. Geom. Metr. Spaces 6(1) (2018), 129145.10.1515/agms-2018-0007CrossRefGoogle Scholar
Jiang, R., ‘Cheeger-harmonic functions in metric measure spaces revisited’, J. Funct. Anal. 266(3) (2014), 13731394.10.1016/j.jfa.2013.11.022CrossRefGoogle Scholar
Jiang, R., Li, H. and Zhang, H., ‘Heat kernel bounds on metric measure spaces and some applications’, Potential Anal. 44(3) (2016), 601627.10.1007/s11118-015-9521-2CrossRefGoogle Scholar
La Nave, G., Tian, G. and Zhang, Z., ‘Bounding diameter of singular Kähler metric’, Amer. J. Math. 139(6) (2017), 16931731.10.1353/ajm.2017.0042CrossRefGoogle Scholar
Li, C., Tian, G. and Wang, F., ‘On the Yau–Tian–Donaldson conjecture for singular Fano varieties’, Comm. Pure Appl. Math. 74(8) (2021), 17481800.10.1002/cpa.21936CrossRefGoogle Scholar
Li, C., Wang, X. and Xu, C., ‘Algebraicity of metric tangent cones and equivariant K-stability’, J. Amer. Math. Soc. 34(4) (2021), 11751214.10.1090/jams/974CrossRefGoogle Scholar
Liu, G. and Székelyhidi, G., ‘Gromov–Hausdorff limits of Kähler manifolds with Ricci curvature bounded below’, Geom. Funct. Anal. 32(2) (2022), 236279.10.1007/s00039-022-00594-8CrossRefGoogle Scholar
Liu, Y. and Xu, C., ‘K-stability of cubic threefolds’, Duke Math. J. 168(11) (2019), 20292073.10.1215/00127094-2019-0006CrossRefGoogle Scholar
Liu, Y., Xu, C. and Zhuang, Z., ‘Finite generation for valuations computing stability thresholds and applications to K-stability’, Ann. of Math. (2) 196(2) (2022), 507566.10.4007/annals.2022.196.2.2CrossRefGoogle Scholar
Lott, J. and Villani, C., ‘Ricci curvature for metric-measure spaces via optimal transport’, Ann. of Math. (2) 169 (2009), 903991.10.4007/annals.2009.169.903CrossRefGoogle Scholar
Jeffres, T., Mazzeo, R. and Rubinstein, Y. A., ‘Kähler–Einstein metrics with edge singularities’, Ann. of Math. (2) 183(1) (2016), 95176.10.4007/annals.2016.183.1.3CrossRefGoogle Scholar
Pan, C.-M. and , T. D., ‘Singular weighted cscK metrics on Kähler varieties’, preprint (2024)Google Scholar
Phong, D. H., Song, J., Sturm, J. and Weinkove, B., ‘The Moser–Trudinger inequality on Kähler–Einstein manifolds’, Amer. J. Math. 130(4) (2008), 10671085.10.1353/ajm.0.0013CrossRefGoogle Scholar
Song, J., ‘Riemannian geometry of Kähler–Einstein currents’, preprint (2014), arXiv:1404.0445.Google Scholar
Song, J. and Weinkove, B., ‘On the convergence and singularities of the J-flow with applications to the Mabuchi energy’, Comm. Pure Appl. Math. 61(2) (2008), 210229.10.1002/cpa.20182CrossRefGoogle Scholar
Spotti, C., ‘Kähler–Einstein metrics on Q-smoothable Fano varieties, their moduli and some applications’, Springer INdAM Ser. 21 (2017), Springer, Cham, 211229.Google Scholar
Sturm, J., Unpublished notes.Google Scholar
Sturm, K.-T., ‘On the geometry of metric measure spaces. I’, Acta Math. 196(1) (2006), 65131.10.1007/s11511-006-0002-8CrossRefGoogle Scholar
Tian, G., ‘On Kähler–Einstein metrics on certain Kähler manifolds with ${C}_1(M)>0$ ’, Invent. Math. 89(2) (1987), 225246.10.1007/BF01389077CrossRef0$+’,+Invent.+Math.+89(2)+(1987),+225–246.>Google Scholar
Tian, G., ‘Kähler–Einstein metrics on algebraic manifolds’, Proc. Int. Congr. Math. (Kyoto, 1990) I (1990), 587598.Google Scholar
Tian, G., ‘Kähler–Einstein metrics with positive scalar curvature’, Invent. Math. 130(1) (1997), 137.10.1007/s002220050176CrossRefGoogle Scholar
Tian, G., ‘Canonical metrics in Kähler geometry’, Lectures Math. ETH Zürich, Birkhäuser Verlag, Basel (2000), vi+101 pp.10.1007/978-3-0348-8389-4CrossRefGoogle Scholar
Tsuji, H., ‘Existence and degeneration of Kähler–Einstein metrics on minimal algebraic varieties of general type’, Math. Ann. 281(1) (1988), 123133.10.1007/BF01449219CrossRefGoogle Scholar
Xu, C. and Zhuang, Z., ‘Boundedness of log Fano cone singularities and discreteness of local volumes’, preprint (2024), arXiv:2404.17134.Google Scholar
Yau, S.-T., ‘Calabi’s conjecture and some new results in algebraic geometry’, Proc. Nat. Acad. Sci. U.S.A. 74(5) (1977), 17981799.10.1073/pnas.74.5.1798CrossRefGoogle ScholarPubMed
Yau, S.-T., ‘On the Ricci curvature of a compact Kähler manifold and the complex Monge–Ampère equation. I’, Comm. Pure Appl. Math. 31(3) (1978), 339411.10.1002/cpa.3160310304CrossRefGoogle Scholar
Zheng, K., ‘Existence of constant scalar curvature Kähler cone metrics, properness and geodesic stability’, preprint (2018), arXiv:1803.09506.Google Scholar
Zheng, K., ‘Singular scalar curvature equations’, preprint (2022), arXiv:2205.14726.Google Scholar