Gromov-Hausdorff limits of collapsing Calabi-Yau fibrations
Extracted source text ยท arXiv 2505.14939
\documentclass{amsart}
\usepackage{amssymb, amsmath, esint}
% \usepackage[backref=page]{hyperref}
\usepackage{cite}
\usepackage{hyperref}
%\usepackage{showkeys}
\renewcommand{\[}{\begin{equation}\begin{aligned}}
\renewcommand{\]}{\end{aligned} \end{equation}}
\newcommand{\ddbar}{\sqrt{-1}\partial \bar\partial }
\usepackage{verbatim}
\newtheorem{thm}{Theorem}
\newtheorem{prop}[thm]{Proposition}
\newtheorem{lemma}[thm]{Lemma}
\newtheorem{claim}[thm]{Claim}
\newtheorem{cor}[thm]{Corollary}
\newtheorem{conj}[thm]{Conjecture}
\newtheorem{ques}[thm]{Question}
\newcommand{\tRic}{\widetilde{\mathrm{Ric}}}
\newcommand{\ssubset}{\subset\joinrel\subset}
\theoremstyle{remark}
\newtheorem{remark}[thm]{Remark}
\theoremstyle{definition}
\newtheorem{definition}[thm]{Definition}
\author{G\'abor Sz\'ekelyhidi}
\address{Department of Mathematics, Northwestern University, Evanston,
IL, USA}
\email{gaborsz@northwestern.edu}
\title{Gromov-Hausdorff limits of collapsing Calabi-Yau fibrations}
\date{}
\begin{document}
\begin{abstract}
We study Calabi-Yau metrics on a projective manifold in K\"ahler
classes converging to a semiample class given by a fibration. We show that the
Gromov-Hausdorff limit of the metrics is homeomorphic to the base of
the fibration and in addition the discriminant locus has Hausdorff
codimension at least 2. This resolves conjectures of Tosatti.
\end{abstract}
\maketitle
\section{Introduction}
Let $M$ be a compact K\"ahler manifold with trivial first Chern
class. By Yau's solution of the Calabi conjecture~\cite{Yau78}, every
K\"ahler class on $M$ admits a unique Ricci flat
K\"ahler metric. A natural and important problem, see
e.g. Yau~\cite[Problem 88]{YauProblems}, is to understand how a family
$\omega_t$ of such metrics can degenerate as the corresponding
K\"ahler classes, or even the underlying complex manifold varies. See for
example Tosatti~\cite{Tos20} for a recent survey on the topic.
In this paper we are motivated by the setting where the family $[\omega_t]$
converges to a semiample class on $M$. More precisely we assume that $M$ is a
projective manifold with a semiample line bundle $L$, inducing a
holomorphic map $f: M\to X$. We assume that the
cohomology classes of the Ricci flat metrics $\omega_t$ are given by $[\omega_t] =
c_1(L) + tc_1(A)$, where $A$ is an ample line bundle on $M$, and $t\to
0$. It was shown by Tosatti~\cite{Tos09} and Zhang~\cite{Zhang06}
that the diameters of $(M, \omega_t)$ are uniformly bounded as $t\to
0$, and so we can extract compact Gromov-Hausdorff limits along
sequences $t_k\to 0$. A long standing problem is to identify these
limits, the most interesting case being when $\dim X < \dim M$.
Under the assumptions $f$ is a holomorphic submersion $f: M^\circ \to
X^\circ$ over a Zariski open subset $X^\circ\subset X$. Moreover,
$X^\circ$ admits a certain canonical K\"ahler metric
$\omega_{can}$ constructed by Song-Tian~\cite{ST12} (see
Section~\ref{sec:collapsing} for more details).
It was shown by Hein-Tosatti~\cite{HT25} that as $t\to 0$, the forms
$\omega_t$ converge to $\omega_{can}$ in $C^\infty_{loc}(M^\circ)$
after many partial results, see e.g. \cite{GW00, Tos10, TWY18, HT20}.
As for the global
behavior of the metrics, Song-Tian-Zhang~\cite{STZ19} showed
that the Riemannian manifolds $(M, \omega_t)$ converge to
a compact metric space $(Z, d_Z)$ in the Gromov-Hausdorff
topology, and moreover $(Z,d_Z)$ is the metric completion of the
length metric space $(X^\circ, \omega_{can})$. A basic question that has remained open is
to identify the geometry of this metric completion. The following
conjecture was formulated by Tosatti~\cite{Tos11} (see also
\cite{GTZ13,Tos20}).
\begin{conj}
The Gromov-Hausdorff limit $(Z, d_Z)$ is homeomorphic to $X$, and
the set $X\setminus X^\circ$ has Hausdorff codimension at least two
in $(Z,d_Z)$.
\end{conj}
The conjecture was shown in several special cases, such as when $X$
has dimension one~\cite{GTZ16}, when
$X$ is smooth and the divisor part of $X\setminus X^\circ$ has simple
normal crossings~\cite{LT23}, when $M$ is hyperk\"ahler~\cite{TZ20},
or when $\dim M = \dim X$, see \cite{Song14}.
The homeomorphism between $Z$ and $X$ was
shown by Song-Tian-Zhang~\cite{STZ19} when $X$ has only orbifold
singularities. Our main result is that the conjecture holds in
general.
\begin{thm}\label{thm:GHlimit1}
The Gromov-Hausdorff limit $(Z,d_Z)$ is homeomorphic to $X$. In
addition we have the Hausdorff dimension bounds $\dim_{\mathcal{H}}
(X\setminus X^{reg}) \leq 2n-4$ and $\dim_{\mathcal{H}} (X\setminus
X^{\circ}) \leq 2n-2$ in terms of the metric $d_Z$, where $n =
\dim_{\mathbb{C}}X$. Here $X^{reg}$ is
the complex analytically regular subset of $X$.
\end{thm}
Theorem~\ref{thm:GHlimit1} will follow from a more general statement that we now
describe. Suppose that $(X,\omega_X)$ is an $n$-dimensional compact normal K\"ahler
space, with smooth K\"ahler metric $\omega_X$, and that $\omega =
\omega_X + \ddbar u$ is a (singular) K\"ahler metric on $X$ satisfying
the following conditions:
\begin{enumerate}
\item $u$ is smooth on $X^\circ = X^{reg}\setminus D$ for a divisor $D$,
\item $\omega^n = e^F \omega_X^n$, such that $F\in L^1(\omega_X^n)$ and
$e^F\in L^p(\omega_X^n)$ for some $p > 1$,
\item $\mathrm{Ric}(\omega) > -A(\omega + \omega_X)$, for some $A >
0$ on $X^{reg}$.
\end{enumerate}
Note that beyond normality we are not making assumptions on the
singularities of $X$. This is important since in general $X$ in
Theorem~\ref{thm:GHlimit1} may not be $\mathbb{Q}$-Gorenstein for
instance. We define $\hat{X}$ to be the metric completion of the
metric space $(X^\circ, \omega)$. We will show the following.
\begin{thm}\label{thm:homeo1}
Suppose that the singular K\"ahler space $(X, \omega)$ satisfies
the conditions (1),(2),(3) above, and in addition the metric
completion $\hat{X}$, equipped with the volume measure $\omega^n$,
is an $RCD(K,2n)$-space for some $K\in\mathbb{R}$, in the sense of
\cite{AGS14,EKS15}.
Then $\hat{X}$ is
homeomorphic to $X$, and
we have the dimension bounds $\dim_{\mathcal{H}} (X\setminus
X^{reg}) \leq 2n-4$, $\dim_{\mathcal{H}} (X\setminus X^\circ) \leq
2n-2$ as in Theorem~\ref{thm:GHlimit1}.
\end{thm}
It is natural to conjecture that the RCD assumption in
Theorem~\ref{thm:homeo1} is actually superfluous. This was conjectured
previously in \cite{SzRCD} for singular K\"ahler-Einstein metrics,
where it
was also shown under the assumption that the K\"ahler-Einstein metric can
be approximated by constant scalar curvature K\"ahler metrics on a
resolution. See also Guo-Song~\cite{GS25}, Fu-Guo-Song~\cite{FGS25}
and Chen-Chiu-Hallgren-Sz\'ekelyhidi-T\^o-Tong~\cite{AIM} for
further results in this direction.
\begin{conj}
Suppose that $(X,\omega)$ satisfies conditions (1), (2), (3)
above. Then the metric completion $\hat{X}$, equipped with the
measure $\omega^n$, is an RCD space.
\end{conj}
In order to deduce Theorem~\ref{thm:GHlimit1} from
Theorem~\ref{thm:homeo1} the main additional ingredient is to show that
the metric completion $\hat{X}$ of $(X^\circ, \omega_{can})$, equipped with the
measure $\omega_{can}^n$, is an $RCD(K, 2n)$
space. Note that since $\hat{X}$ is the Gromov-Hausdorff limit of the
$(M,\omega_t)$, it follows by Ambrosio-Gigli-Savar\'e~\cite{AGS14}
that $\hat{X}$, equipped with the
renormalized limit measure, is
automatically an $RCD(K, 2m)$ space, where $m=\dim_{\mathbb{C}}(M)$,
but typically $m > n$ (see also Sturm~\cite{Sturm06} and
Lott-Villani~\cite{LV09} for earlier results about $CD$-spaces).
We will follow arguments of
Gross-Tosatti-Zhang~\cite[Lemma 5.2]{GTZ13} and Fu-Guo-Song~\cite[Proof
of Theorem 1.3]{FGS20} to show that the renormalized limit measure on
$\hat{X}$ actually coincides with $\omega_{can}^n$,
which will then lead to the required $RCD(K, 2n)$-property.
We will now describe the proof of Theorem~\ref{thm:homeo1}. The basic
strategy is based on the author's previous work~\cite{SzRCD}, which in
turn uses the techniques of Donaldson-Sun~\cite{DS14}. These methods
were extended in Guo-Song~\cite{GS25} to a more general
setting of K\"ahler spaces with Ricci curvature bounded below, where
the divisor $D$ is empty. The main
new difficulty in Theorem~\ref{thm:homeo1} in comparison with earlier
works is that there could be points in the divisor $D$ that are
regular in the sense of metric tangent cones, even though the metric $\omega$
is in general singular along $D$. As a result, rescalings of the
metric on $\hat{X}$ do not necessarily have good convergence
properties on the regular sets of its tangent cones. This issue will be dealt with
in Proposition~\ref{prop:coords} below. An important ingredient is a
strict positivity result for singular K\"ahler metrics $\omega$
satisfying the conditions (1),(2),(3), that was shown in \cite{AIM},
generalizing previous work in \cite{SzRCD}. The upshot is that
near almost regular points of $\hat{X}$ in the metric sense we can find
smooth Gromov-Hausdorff approximations with Ricci curvature bounded
below. These approximations can be used to apply the results of
\cite{LiuSz1} on the construction of good local holomorphic charts.
It is worth noting that a similar difficulty appears in the dimension
estimate for $X\setminus X^\circ$, namely that this set could contain
points that are regular in the sense of metric tangent cones. Because
of this, Cheeger-Colding's dimension estimates~\cite{CC1}, and their
extension to the RCD setting by De Philippis-Gigli~\cite{DPG18}, do
not immediately lead to the Hausdorff codimension 2 result. Instead
we exploit the fact that locally $X\setminus X^\circ$ is cut out
by holomorphic functions.
To close the introduction we briefly mention another setting where
Theorem~\ref{thm:homeo1} can be applied. Here $M$ is a projective
manifold of dimension $m$, whose canonical line bundle $K_M$ is semiample, and we assume
that the Kodaira dimension is $n \in (0, m]$. Given an ample line
bundle $A$ on $M$, La Nave-Tian~\cite{LNT16} showed that there are
unique metrics $\omega_t\in c_1(tA+K_M)$ satisfying the equations
\[ \mathrm{Ric}(\omega_t) = -\omega_t + t\omega_A, \]
for a K\"ahler metric $\omega_A\in c_1(A)$. As above, $K_M$ induces a
holomorphic map $f: M\to X$ which is a submersion over $X^\circ$. It was shown in
Song-Tian-Zhang~\cite{STZ19} that similarly to the collapsing
Calabi-Yau setting considered above, there is a canonical metric
$\omega_{can}$ on $X^\circ$, and $(M, \omega_t)$ converge in the
Gromov-Hausdorff sense to the metric completion $\hat{X}$ of $(X^\circ,
\omega_{can})$as $t\to 0$. Essentially the same arguments as in the proof of
Theorem~\ref{thm:homeo1} imply that $\hat{X}$ is
homeomorphic to $X$, and we have the same Hausdorff dimension bounds
as in Theorem~\ref{thm:GHlimit1}. We will discuss this setting in more
detail at the end of Section~\ref{sec:collapsing}.
\subsection*{Acknowledgements}
I would like to thank Valentino Tosatti, Jian Song and S{\l }awomir Dinew
for helpful discussions. This work was supported in part by NSF grant DMS-2203218.
\section{Homeomorphism with the underlying variety}
Suppose that $(X,\omega)$ is a normal K\"ahler space satisfying the
conditions (1), (2), (3) in the introduction for a divisor $D\subset
X$. In addition, we assume that the metric completion $(\hat{X}, d)$ of
$(X^{reg}\setminus D, \omega)$ is an $RCD(K, 2n)$-space with the
volume form $\omega^n$. Note that by \cite[Theorem 1.1]{AIM} we know
that $\omega > c_0 \omega_X$ for a constant $c_0 > 0$, where
$\omega_X$ is a smooth K\"ahler metric on $X$. Our first goal in this section is
to prove the following, generalizing the author's earlier result in
\cite{SzRCD}.
\begin{thm}\label{thm:homeo2}
The metric space $\hat{X}$ is homeomorphic to the variety $X$.
\end{thm}
First, observe
that the inclusion map $(X^{reg}\setminus D, \omega) \to
(X, \omega_X)$ is Lipschitz because of the lower bound $\omega > c_0
\omega_X$. It follows that
this map has a Lipschitz extension
\[ \label{eq:Phidefn} \Phi : (\hat{X}, d_{\hat{X}}) \to (X, \omega_X) \]
to the metric completion, which is surjective by the density of
$X^{reg}\setminus D$ in $X$. To prove Theorem~\ref{thm:homeo2}
we need to show that $\Phi$ is injective.
Following the strategy of \cite{SzRCD}, we show that if $p_1 \not=
p_2$ in $\hat{X}$, with $\Phi(p_1) = \Phi(p_2) = p \in X$, then in a
neighborhood of $p$ we can construct a bounded holomorphic function $f$ on
$X^{reg}\setminus D$ whose extension to $\hat{X}$ (as complex harmonic
functions) has different values
at $p_1, p_2$. This is a contradiction, since $f$ extends continuously
as a holomorphic function to $X$ as well, showing that $\Phi$ is in fact
injective. In comparison with the proof in \cite{SzRCD}
the presence of the divisor $D$ causes substantial new technical
difficulties (see Proposition~\ref{prop:coords}), and note also that here
we are not assuming that the space is projective.
Instead of working with a global line bundle on $X$ as
in \cite{SzRCD}, we will work with a cover of $X$ with Stein
neighborhoods. Specifically, for any point $q\in X$ we can choose a Stein
neighborhood $\mathcal{U}$ of $q$, for instance the intersection of $X$ with a
Euclidean ball under a local embedding of $X$ into $\mathbb{C}^N$. We
then consider $\mathcal{V} = \mathcal{U}\cap (X^{reg}\setminus D)$. We
also choose a smaller neighborhood $q\in \mathcal{U}'\subset\subset
\mathcal{U}$ and let $\mathcal{V}' = \mathcal{V}\cap \mathcal{U}'$. We can
assume that on $\mathcal{V}$ we have $\omega = \ddbar u$ for a bounded
function $u$, and we let $L$
denote the trivial line bundle over $\mathcal{V}$, equipped with the (smooth)
Hermitian metric $h=e^{-u}$. Note that $\mathcal{V}$ admits a complete
K\"ahler metric, using Demailly~\cite{Dem82}, and so we can apply the
following H\"ormander estimate (see \cite[Theorem 4.1]{Dem82})
to the line bundle $P = L^k\otimes K_X^{-1}$
over $\mathcal{V}$, for large $k$.
\begin{thm}\label{lem:Hormander}
Let $(P,h_P)$ be a Hermitian holomorphic
line bundle on a K\"ahler manifold $(M, \omega_M)$, which
admits a complete K\"ahler 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
\[ \Vert u\Vert_{L^2}^2 \leq \frac{1}{c} \Vert
\alpha\Vert^2_{L^2}, \]
provided the right hand side is finite.
\end{thm}
We will let $M=\mathcal{V}$ with the metric $\omega_M = k\omega$ for
some $k > 0$. On $P=L^k\otimes K_X^{-1}$ we use the metric
$h^k$ on $L^k$ and $\omega^n$ on $K_X$.
We have the following estimate for sections of $L^k$, analogous to
\cite[Proposition 19]{SzRCD}.
\begin{prop}
Suppose that $f$ is a holomorphic section of $L^k$ over
$\mathcal{V}$. Then we have
\[ \sup_{\mathcal{V}'} |f|_{h^k} + |\nabla f|_{h^k, \omega_M} \leq
K_1 \Vert f\Vert_{L^2(\mathcal{V}, h^k, \omega_M)}, \]
where $K_1$ does not depend on $k$.
\end{prop}
\begin{proof}
The proof is fairly standard, similar to that in \cite{SzRCD}, once
we extend the section to $\mathcal{U}$.
Since $\omega > c_0\omega_X$, and $h = e^{-u}$ for a bounded
function $u$, a section $f$ of $L^k$ with bounded $L^2$-norm with
respect to $h^k, \omega$ can be identified with a holomorphic
function on $\mathcal{V}$ that is $L^2$ with respect to
$\omega_X$. It follows that first $f$ extends across $D$ to a
holomorphic function on $\mathcal{U}\cap X^{reg}$. Then using that
$X$ is normal we can extend $f$ to $\mathcal{U}$. In particular $f$
is bounded on compact subsets of $\mathcal{U}$. Using the
gradient estimate for RCD spaces from Jiang~\cite{Jiang14} it also
follows that $|\nabla f|_{h^k}$ is bounded on compact sets, using
again that the metric $h$ is $e^{-u}$ for bounded $u$.
To obtain the estimates for $|f|_{h^k}, |\nabla f|_{h^k, \omega_M}$,
note that both quantities satisfy differential inequalities of the
form $\Delta v \geq -Cv$ in terms of the metric $\omega_M$. We have
an $r_0 > 0$ such that for any point $x\in \mathcal{V}'$ the ball
$B_{\omega_M}(x,r_0) \subset \mathcal{U}$, no matter what $k$
is. Using this, together with the mean value inequality and gradient
estimate for harmonic functions, we obtain the required bounds for
$|f|, |\nabla f|$.
\end{proof}
Our next goal is to show that the almost regular set in $\hat{X}$ in
the metric sense is contained in $X^{reg}$. More precisely, for any
$\epsilon > 0$ we define the $\epsilon$-regular set
$\mathcal{R}_\epsilon(Y)$ in a noncollapsed $2n$-dimensional
RCD space $Y$ to consist of the set of points $p$ that satisfy
\[ \lim_{r\to 0} r^{-2n} \mathrm{vol}(B(p,r)) > \omega_{2n} -
\epsilon, \]
where $\omega_{2n}$ is the volume of the $2n$-dimensional Euclidean
unit ball. We have the following, similar to \cite[Proposition
24]{SzRCD}.
\begin{prop}\label{prop:epsilonreg}
There is an $\epsilon_3 > 0$, depending on $(X,\omega)$, such that
the map $\Phi$ in \eqref{eq:Phidefn} is injective on
$\mathcal{R}_{\epsilon_3}(\hat{X})$, and its image is contained in $X^{reg}$.
\end{prop}
\begin{proof}
The proof mainly follows along the lines of the proof in
\cite{SzRCD}, and we will only focus on the differences here. Note
that there may be additional metric singular points contained in the
divisor $D$, so $\mathcal{R}_{\epsilon_3}(\hat{X})$ may be smaller
than $X^{reg}$, and in addition it is not immediate that $\Phi$ is
injective over the preimage $\Phi^{-1}(X^{reg})$.
Let us define $\Gamma = (X\setminus X^{reg})\cup D$, which is the
set where $\omega$ is not assumed to be smooth. We suppose that we
have Stein neighborhoods $\mathcal{U}_k' \subset \mathcal{U}_k$ as
above, such that the $\mathcal{U}_k'$ cover $X$. In addition we can
assume that we have bounded holomorphic functions $s_k$ on
$\mathcal{U}_k$ such that $\Gamma \cap \mathcal{U}_k \subset
s_k^{-1}(0)$. Let us write $\hat{\mathcal{U}}_k, \hat{\mathcal{U}}_k'$
for the preimages of $\mathcal{U}_k, \mathcal{U}_k'$ under the map
$\Phi$. The $s_k$ define bounded complex valued harmonic functions
on the $\hat{\mathcal{U}}_k$.
Using $\omega > c_0\omega_X$
we also have $r_0 > 0$ such that $B(q, r_0)\subset
\hat{\mathcal{U}}_k$ for any $q\in \hat{\mathcal{U}}_k'$.
We can argue exactly like in \cite[Lemma 20]{SzRCD} to show that
there are constants $c_1, N > 0$ such that
\[ \int_{B(q, r)} |s_k|^2\, \omega^n \geq c_1 r^N, \]
for all $r\in (0,r_0)$ and $q\in \hat{\mathcal{U}}_k'$. Using this,
we obtain the same result as \cite[Lemma 22]{SzRCD}. Namely, there
are $\epsilon_0, r_0 > 0$, such that whenever $q\in
\hat{\mathcal{U}}_k'$ and
\[ d_{GH} (B(q, r_1) , B(0_{\mathbb{R}^{2n}}, r_1)) <
r_1\epsilon_0, \]
for some $r_1\in (0,r_0)$, then we have
\[ \limsup_{r\to 0} \frac{ \fint_{B(q, r)} |s_k|^2\,
\omega^n}{\fint_{B(q, r/2)} |s_k|^2\, \omega^n} \leq 2^{2N}. \]
Suppose that we have a sequence of points $p_j \in \hat{X}$, and
$r_j \to 0$ such that the pointed Gromov-Hausdorff limit of the rescaled
sequence $(\hat{X}, r_j^{-1}d_{\hat{X}}, p_j)$ is
$\mathbb{R}^{2n}$. Without loss of generality we can assume that
each $p_j$ is in one of our neighborhoods $\hat{\mathcal{U}}_k'$,
which we denote for simplicity by $\hat{\mathcal{U}}' \subset
\hat{\mathcal{U}}$. We will show that for sufficiently large $j$ we
have $\Phi(p_j)\in X^{reg}$. As above, we let $L$ denote the trivial line
bundle over $\mathcal{V} = \mathcal{U}\setminus \Gamma$,
with metric $e^{-u}$, where $\ddbar u = \omega$.
Let us denote by $\Omega_j$ the domains $\mathcal{U}$,
equipped with the metric $\omega_j = r_j^{-2}\omega$, and the trivial
line bundle $L_j$ with metric $e^{-r_j^{-2}u}$. Write
$\hat{\Omega}_j$ for the preimage of $\Omega_j$ in $\hat{X}$, with
the rescaled metric $r_j^{-1}d_{\hat{X}}$. Let us denote by
$q_j\in\hat{\Omega}_j$ the points corresponding to the $p_j$. By decreasing the $r_j$,
we can assume that on $\Omega_j$ the set $\Gamma$ is
contained in the zero set of a holomorphic function $f_j$ (a
suitable scaling of one of the $s_k$ above) such that
$\Vert f_j\Vert_{L^2(B(q_j, 1))} = 1$, and in addition for any $R >
0$ if $j$ is sufficiently large we have
\[ \frac{\fint_{B(q_j, r)} |f_j|^2}{\fint_{B(q_j, r/2)} |f_j|^2}
\leq 2^{2N+1}, \text{ for all } r < R. \]
Using that the $f_j$ define complex valued harmonic functions on
$\hat{\Omega}_j$, the above properties imply that up to choosing a
subsequence we can extract a limit $f_\infty :
\mathbb{R}^{2n}\to\mathbb{C}$, which is a non-zero harmonic
function. By further decreasing the $r_j$ we can assume that
$f_\infty$ is homogeneous. Let us write $\Sigma =
f_\infty^{-1}(0)$. We will argue somewhat similarly
to the argument in \cite[Proposition 23]{SzRCD} to show that
$f_\infty$ is actually a holomorphic function, so that $\Sigma$ has
codimension two.
Note that if $V\subset\mathbb{R}^{2n}\setminus
\Sigma$ is a relatively compact open set, then under the pointed convergence
of $(\hat{\Omega}_j, q_j)$ to $(\mathbb{R}^{2n}, 0)$, $V$ is the
limit of subsets $V_j\subset \hat\Omega_j$ where the metric
$\omega_j$ is a smooth K\"ahler metric with Ricci curvature bounded
below. In \cite{SzRCD} we worked with K\"ahler-Einstein metrics
and relied on Anderson's $\epsilon$-regularity theorem, while here
we can use \cite[Theorem 1.4]{LiuSz1} instead. The latter result implies
that the complex structures on the $V_j$ converge to a complex
structure on $V$ with respect to which the Euclidean metric is
K\"ahler (see \cite[Proof of Proposition 2.5]{LiuSz1}), and for
which $f_\infty$ is holomorphic. Using this we can
argue exactly as in \cite[Proposition 23]{SzRCD} to show that
$\mathbb{R}^{2n}\setminus \Sigma$ is connected, $f_\infty$ is
holomorphic for a complex structure on $\mathbb{R}^{2n}$, and
therefore $\Sigma$ is a complex hypersurface defined by a
homogeneous function.
At this point we can closely follow the proof of \cite[Proposition
3.1]{LiuSz1} to prove the following: there is a $C > 0$, and for
any $\zeta > 0$ there is a $K > 0$ (depending on $\zeta$) with the
following property. For sufficiently large $j$ the line bundle $L_j^m$ for some $m < K$
admits a holomorphic section $s^{(j)}$ over $\hat{\Omega}_j$
satisfying $\Vert s^{(j)}\Vert_{L^2} \leq C$ and $\Big| |s^{(j)}(x)| - e^{-m
d(x, q_j)^2/2}\Big| < \zeta$ for $x\in \hat{\Omega}_j$. Note that $s^{(j)}$
here is simply a holomorphic function on $\Omega_j$, which naturally
extends to a complex valued harmonic function on $\hat{\Omega}_j$,
and whose norm is weighted by
$e^{-mu}$.
The section $s^{(j)}$ corresponds to the constant function $1$
as a section of the trivial line bundle over $\mathbb{C}^n$ with
metric $e^{-|z|^2}$. By considering the coordinate functions
$z_1,\ldots, z_n$ on
$\mathbb{C}^n$ we can similarly construct holomorphic sections
$s^{(j)}_1,\ldots, s^{(j)}_n$, and the ratios $s^{(j)}_i / s^{(j)}$
converge to $z_1, \ldots, z_n$ under the Gromov-Hausdorff
convergence. It follows from this that for sufficiently large $j$
the point $\Phi(q_j)$, and so the corresponding $\Phi(p_j)$ as well, lies in
$X^{reg}$. Using this, and a contradiction argument, it follows that for
sufficiently small $\epsilon_3$ if $p\in
\mathcal{R}_{\epsilon_3}(\hat{X})$, then $\Phi(p)\in X^{reg}$.
We
can see that $\Phi$ is injective on $\mathcal{R}_{\epsilon_3}$, after
possibly shrinking $\epsilon_3$, as follows. Suppose that $\Phi(p_1)
= \Phi(p_2)$. We can apply the argument above to the rescaled
pointed spaces $(\Omega_j, \omega_j, q_j)$, where $\Omega_j=
\mathcal{U}$, $\omega_j =
r_j^{-2}\omega_j$ for a sequence $r_j \to 0$, and $q_j$ corresponding
to $p_1$. Fixing $\zeta = \frac{1}{10}$, once $j$ is
sufficiently large, we will find a holomorphic function $s$ such
that $|s(p_1)|e^{-mu(p_1)} > 9/10$, while $|s(p_2)|e^{-mu(p_2)} <
2/10$. Without loss of generality we can assume that $u(p_1) \geq
u(p_2)$, in which case the holomorphic function $s$ on $\mathcal{U}$
separates $p_1$ and $p_2$. This means that in fact $\Phi(p_1)\not=
\Phi(p_2)$.
\end{proof}
The following is the crucial new ingredient that we need in comparison
with the arguments in \cite{SzRCD}.
\begin{prop}\label{prop:coords}
Given any $\tau > 0$, there exists an $\epsilon > 0$ depending on
$(X,\omega)$ and $\tau$ with the following property. Suppose that
$r < \epsilon^2$, and for $p\in \hat{X}$
\[ \label{eq:d10} d_{GH}(B(p, \epsilon^{-1}r), B_{\mathbb{C}^n}(0, \epsilon^{-1}r))
< r\epsilon. \]
Then there is a holomorphic chart $F : B(p,r) \to \mathbb{C}^n$,
which is an $r\tau$-Gromov-Hausdorff approximation to its
image. In addition we can write $\omega = \ddbar \phi$ on $B(p,r)$
with $|\phi - d_p^2| < r^2\tau$, where $d_p$ denotes the distance
to $p$.
\end{prop}
\begin{proof}
If the metric $\omega$ on $B(p, \epsilon^{-1}r)$ were smooth, then
this would be a direct application of \cite[Theorem
1.4]{LiuSz1}. In order to apply this result, we will first
approximate $\omega$ with smooth metrics in the Gromov-Hausdorff
sense.
Note that by Proposition~\ref{prop:epsilonreg}, if $\epsilon$
is chosen sufficiently small, then \eqref{eq:d10} implies that $B(p,
\epsilon^{-1}r/2)$ can be identified with a subset of $X^{reg}$. By the assumption for
$\omega$ we have $\mathrm{Ric}(\omega) > -A(\omega + \omega_X)$, and
by the lower bound $\omega > c_0\omega_X$ it follows that
$\mathrm{Ric}(\omega) > -A_1\omega$ for some $A_1$ on $X^{reg}$. Let
us write $\mathrm{Ric}(\omega) = -A_1\omega + \alpha$. The basic idea
is to try to regularize $\alpha$ using a sequence of $\alpha_k \geq 0$ that are
smooth on $X^{reg}$, and then approximate $\omega$ with solutions of
$\mathrm{Ric}(\omega_k) = -A_1\omega_k + \alpha_k$. This is not
quite what we achieve, but rather we end up regularizing $\alpha +
A'\omega$ for a sufficiently large $A'$. A technical difficulty
arises from the fact that $\alpha$ may not be locally $\ddbar$-exact
around the singularities of $X$, which is why we end up only regularizing on
compact subsets of $X^{reg}$.
Consider a resolution $\pi : Y\to X$, which admits a K\"ahler metric
$\omega_Y$. We can write the equation $\mathrm{Ric}(\omega) =
-A_1\omega + \alpha$ as a degenerate Monge-Amp\`ere equation on
$Y$. To be more precise we will work on $Y\setminus E$, where
$E\subset Y$ is the exceptional divisor. Throughout we will identify
$Y\setminus E \cong X^{reg}$. On $Y\setminus E$
we have $\omega = \omega_X + \ddbar u$, where
\[ (\omega_X + \ddbar u)^n = e^{A_1u + F}\omega_X^n. \]
It follows that on $Y\setminus E$ we have
\[ \mathrm{Ric}(\omega) = -A_1\ddbar u - \ddbar F + \mathrm{Ric}(\omega_X), \]
and so
\[ \label{eq:ddbF1} \ddbar (-F) = \alpha - A_1\omega_X - \mathrm{Ric}(\omega_X). \]
Note that $F$ is bounded below, since $u$ is bounded, and $\omega >
c_0\omega_X$, and in fact $-F$ extends as a quasi-psh function on $Y$
(see the proof of \cite[Proposition 3.1]{AIM}).
We have $\omega_X^n = e^g \omega_Y^n$ for a function $g$ given by
\[ g = \sum_i a_i \log |s_{E_i}|_{h_i}^2, \]
where $a_i > 0$, and the $s_{E_i}$ are defining sections of the line bundles corresponding to the components $E_i$ of the exceptional divisor, with smooth Hermitian metrics $h_i$. This implies that on $Y\setminus E$ we have $\mathrm{Ric}(\omega_X) > -A_2 \omega_Y$ for some $A_2 > 0$. At the same time we can find a function $\rho$ on $Y$, such that $\rho$ is bounded above, smooth on $Y\setminus E$, and $\rho\to -\infty$ along $E$, such that for some $A_3 > 0$ we have
\[ A_3(\omega_X + \ddbar \rho) > \omega_Y. \]
In particular this implies that on $Y\setminus E$ we have
\[ \label{eq:ddbrho1} A_2A_3\ddbar\rho > -A_2A_3\omega_X - \mathrm{Ric}(\omega_X). \]
Let us define the functions $F_k$ by
\[ -F_k = \max \{ -F, A_2A_3\rho - k\}. \]
Note that $-F \leq -F_k \leq C$ for a uniform $C$, since $-F$ and
$\rho$ are bounded above.
Using \eqref{eq:ddbF1} and \eqref{eq:ddbrho1} we get
\[ \ddbar(-F_k) > -A_4\omega_X - \mathrm{Ric}(\omega_X), \]
on $Y\setminus E$, for some $A_4 > 0$.
Since $F_k$ is bounded away from $E$, we can further regularize using
the technique of Blocki-Kolodziej~\cite{BK07} to obtain $\tilde{F}_k$
such that on any compact subset of $Y\setminus E$ the $\tilde{F}_k$
are smooth for sufficiently large $k$. More precisely, suppose that
$K\subset Y\setminus E$ is compact. For small $s > 0$ depending on
$K$, we can cover $K$ with small balls
$B_{\omega_Y}(x_i, s)$, such that on the balls $B_{\omega_Y}(x_i,
10s) \subset\subset Y\setminus E$ we have K\"ahler potentials $\phi_i$
for $\omega_Y$ satisfying
\[ -100s^2 &< \phi_i < 100s^2, \\
\inf_{\partial B_{\omega_Y}(x_i, 10s)} \phi_i &> 50s^2, \\
\sup_{B_{\omega_Y}(x_i, s)} \phi_i &< -s^2. \]
For this note that $\phi_i$ can be chosen to be approximately equal
to $d_{\omega_Y}(x_i, \cdot)^2 - 5s^2$. Let $\delta > 0$ be small, to
be chosen. In each
ball $B_{\omega_Y}(x_i, 10s)$ we can use a mollifier to construct a
smooth function $F_{k,\delta,i}$, satisfying $|F_{k, \delta,i} - F_k| < \delta$,
and
\[ \ddbar(-F_{k,\delta,i}) > -A_4\omega_X - \mathrm{Ric}(\omega_X). \]
Let $F_{k,\delta,i}' = F_{k, \delta,i} + \delta^{1/2}\phi_i$. Then
\[ \label{eq:Fkdeltalower} \ddbar(-F_{k, \delta,i}') > -A_4\omega_X - \delta^{1/2}\omega_Y -
\mathrm{Ric}(\omega_X), \text{ on } B_{\omega_Y}(x_i, 10s). \]
We define $-F_{k,\delta}$ to be a regularized maximum of the function
$-F_k$ together with all of the $-F_{k,\delta, i}'$.
For sufficiently small $\delta >
0$ this will be a well defined function on $Y\setminus E$, since if $y\in \partial
B_{\omega_Y}(x_i, 10s)$ for some $i$, then necessarily
\[ -F_{k,\delta, i}'(y) &= ( -F_{k,\delta, i}(y) + F_k(y) )+
(F_{k,\delta, i}(y) - F_{k,\delta, i}'(y)) - F_k(y) \\
&< \delta - 50 \delta^{1/2} s^2 - F_k(y) \\
&< -F_k(y) - \delta, \]
if $\delta$ is sufficiently small (depending on $s$). In addition
$-F_{k,\delta}$ is smooth on $K$, since if $y\in B_{\omega_Y}(x_i,
s)$ for some $i$, then
\[ -F_{k,\delta, i}'(y) &> -F_{k,\delta, i}(y) + \delta^{1/2} s^2 \\
&> -F_k(y) - \delta + \delta^{1/2}s^2 > -F_k(y) + \delta \]
for sufficiently small $\delta$, and the $-F_{k, \delta, i}'$ are
smooth in $B_{\omega_Y}(x_i, 10s)$. Finally note that outside of the union of the balls
$B_{\omega_Y}(x_i, 10s)$ we have $F_{k,\delta} = F_k$, so
\[ \label{eq:Fkdeltalower2} \ddbar(-F_{k, \delta}) > -A_4\omega_X - \mathrm{Ric}(\omega_X),
\text{ on } Y\setminus \left( E \bigcup B_{\omega_Y}(x_i, 10s)\right). \]
We can define the sequence of regularizations $\tilde{F}_k$ by
choosing an exhaustion of $Y\setminus E$ with compact sets $K$ as
above, and for each $K$ choosing $\delta$ sufficiently small so that
\eqref{eq:Fkdeltalower} and \eqref{eq:Fkdeltalower2} imply that
\[ \label{eq:ddbarF2}
\ddbar(-\tilde{F}_k) > -2A_4\omega_X - \mathrm{Ric}(\omega_X), \]
At the same time we can ensure that we have $-C \leq \tilde{F}_k
\leq F$, for a uniform $C$, and on
any compact subset of $\pi^{-1}(X^{reg}\setminus D)$ we have $\tilde{F}_k \to F$ smoothly.
Define $\omega_k = \pi^*\omega_X + \ddbar u_k$ to be solutions of the Monge-Amp\`ere equations
\[ \omega_k^n = e^{A_1u_k + \tilde{F}_k}\omega_X^n. \]
Since $\tilde{F}_k \leq F$, and $e^{pF}\omega_X^n$ is integrable for some $p > 1$, it follows from \cite[Theorem 4.1]{EGZ09} that the equation has a unique solution, and we have a uniform $L^\infty$ bound $|u_k| < C$ independent of $k$. Using the argument in \cite[Proposition 3.1]{AIM}, adapted from an argument in Guedj-Lu~\cite{GL25}, it follows that on any compact subset of $\pi^{-1}(X^{reg}\setminus D)$ the $u_k$ converge smoothly to $u$.
The Ricci curvature of $\omega_k$ on $Y\setminus E$ satisfies
\[ \mathrm{Ric}(\omega_k) = A_1(\omega_X - \omega_k) - \ddbar \tilde{F}_k + \mathrm{Ric}(\omega_X), \]
and so by \eqref{eq:ddbarF2} we get
\[ \mathrm{Ric}(\omega_k) \geq - A_1\omega_k - 2A_4\omega_X. \]
Using \cite[Theorem 1.1]{AIM} it follows that $\omega_k > c_1
\omega_X$, for some $c_1 > 0$ independent of $k$. The upshot is that
on $B_{\omega}(p, \epsilon^{-1}r/2)$ we have a sequence of smooth
K\"ahler metrics $\omega_k$, converging locally smoothly to $\omega$
on $B_{\omega}(p, \epsilon^{-1}r/2)\setminus D$, satisfying a uniform
Ricci lower bound $\mathrm{Ric}(\omega_k) > -A\omega_k$, and uniform
lower bounds $\omega_k > c_1\omega_X$.
For simplicity let us write $\Omega = B_\omega(p, \epsilon^{-1/2}r)$.
We claim that $\Omega$, equipped with the metrics $\omega_k$
converges in the Gromov-Hausdorff sense to $\Omega$ equipped with the
metric $\omega$. More precisely
we show that for $q_1, q_2 \in \Omega$ we have $d_{\omega_k}(q_1,q_2) \to
d_{\omega}(q_1, q_2)$, and this convergence is uniform in $q_1,
q_2$. Recall that since $\omega$ may be singular along $D$, the
distance $d_\omega(q_1,q_2)$ is defined using curves lying in
$B_\omega(p, \epsilon^{-1}r/2) \setminus D$, joining sequences of
points $q_1^j \to q_1$ and $q_2^j\to q_2$.
We can closely follow arguments in Song-Tian-Zhang~\cite{STZ19}
for instance.
For $\delta > 0$ let us denote by $D_\delta$
the set of points $q\in X$ such that $d_{\omega_X}(q, D) >
\delta$. Let us also choose $\delta_0 > 0$ such that the
$\omega_k$-distance from $\Omega$ to $\partial
B_{\omega}(p, \epsilon^{-1}r)$ is at least $2\delta_0$ for all $k$.
Such $\delta_0$ exists because of the lower bound $\omega_k >
c_1\omega_X$.
The main ingredient is the following ``almost convexity'' result for the set
$D_\delta\cap \Omega$. We claim that
for any $\delta > 0$ there exists a $\rho > 0$ such that for all $k$ if
$q_1,q_2\in \Omega \cap D_\delta$ and
$d_{\omega_k}(q_1,q_2) < \delta_0$, then
there is a smooth path $\gamma_k \subset B_\omega(p, \epsilon^{-1}r)\cap
D_\rho$ such that its length with respect to $\omega_k$ satisfies
\[ \ell_{\omega_k}(\gamma_k) \leq d_{\omega_k}(q_1, q_2) + \delta. \]
This closely follows the argument in \cite[Lemma 2.7]{STZ19}, or
Datar~\cite[Proposition 1.1]{Dat16}, based on relative volume
comparison. Note that our assumption means that if
$\ell_{\omega_k}(\gamma_k) < 2\delta_0$, then $\gamma_k$ is contained
in $B_\omega(p, \epsilon^{-1}r/2)$.
Suppose now that $q_1,q_2\in \Omega$ and
$d_{\omega_k}(q_1, q_2) < \delta_0$. Since we have a uniform
$L^p$-bound for the volume forms of $\omega_k$, we can use the work of
Guo-Phong-Song-Sturm~\cite[Theorem 3.2]{GPSS2} to get a uniform
volume non-collapsing
bound $\mathrm{Vol}_{\omega_k}(B_{\omega_k}(q, r)) \geq \kappa
r^{2\mu}$ for some $\kappa,\mu > 0$, while a similar bound holds for $\omega$ by the
non-collapsed RCD property. Using the $L^p$ bound for the volume forms
of $\omega, \omega_k$ we also know that the $\omega_k$-volume of the
$\omega_X$-tubular neighborhood
$B_{\omega_X}(D, \delta)$ satisfies
\[ \mathrm{Vol}_{\omega_k}(B_{\omega_X}(D, \delta)) < \Psi(\delta), \]
where $\Psi(\delta)$ denotes a function satisfying $\Psi(\delta) \to
0$ as $\delta \to 0$, and which we may change below, but which is
independent of $k$. The same estimate also
holds for the $\omega$-volume. This implies that for any
$q\in \Omega$, if $\delta > 0$, and we fix $k$, then
we can find some $q' \in D_\delta$ such that $d_{\omega}(q, q') <
\Psi(\delta)$ and $d_{\omega_k}(q,q') < \Psi(\delta)$. We apply this to
$q_1, q_2$ to obtain points $q_1', q_2'$. The almost convexity result
above, together with the smooth convergence of $\omega_k$ to $\omega$
on $D_\rho$ implies that for sufficiently large $k$ we have
\[ |d_{\omega}(q_1, q_2) - d_{\omega_k}(q_1,q_2)| < \Psi(\delta). \]
The claimed Gromov-Hausdorff convergence follows from this.
We have therefore shown that under the assumption $\eqref{eq:d10}$,
for sufficiently small $\epsilon$, we can realize $B_\omega(p,
\epsilon^{-1/2}r)$, equipped with the metric $\omega$,
as a Gromov-Hausdorff limit of smooth K\"ahler
manifolds with Ricci curvature bounded below. We can then apply
\cite[Theorem 1.4]{LiuSz1} to obtain holomorphic
charts $F_k : B(p_k,r)\to \mathbb{C}^n$, and K\"ahler potentials
$\phi_k$ on balls $B(p_k, r)$ in this sequence of smooth K\"ahler
manifolds. As in \cite[Proposition 2.4]{LiuSz1}, we can pass to a
limit to obtain the required holomorphic chart $F$. We can also pass
the K\"ahler potentials $\phi_k$ to the limit to get a
plurisubharmonic function $\phi$, and the argue as in \cite[Claim 2.1
of Proposition 2.5]{LiuSz1} to see that $\phi$ is a K\"ahler potential
for $\omega$ on $B(p, r)$ satisfying the required bound.
\end{proof}
Given this result, we can closely follow the arguments in
\cite[Proposition 3.1]{LiuSz1}. The main difference is that in the
setting of noncollapsed RCD spaces the sharp volume estimates of
Cheeger-Jiang-Naber~\cite{CJN21} do not seem to be available in the
literature. Instead, as described in \cite[Proposition 5.1]{LiuSz1},
we can use the approach of Chen-Donaldson-Sun~\cite{CDS2}, which
deals separately with the codimension two singular stratum.
As above, we focus on one of the Stein neighborhoods $\mathcal{U}$,
and $\mathcal{U}'\subset\subset \mathcal{U}$, with the trivial bundle
$L$ equipped with the metric $e^{-u}$. The proof of the following is
analogous to \cite[Proposition 3.1]{LiuSz1}.
\begin{prop}\label{prop:LSz2}
Let $(V, o)$ be an iterated tangent cone of $\hat{X}$, with the
property that the set $V\setminus \mathcal{R}_{\epsilon_3}(V)$ has
capacity zero, for the $\epsilon_3$ in
Proposition~\ref{prop:epsilonreg}. There is a constant $C > 0$, and
for any $\zeta > 0$ there are $K, \epsilon > 0$ depending on $\zeta,
(X, \omega), V$, with the following property. Suppose that $r > 0$
satisfies $\epsilon^{-1}r < \epsilon$, and for some $p \in
\hat{\mathcal{U}}'$ we have
\[ d_{GH}(B(p, \epsilon^{-1}r), B(o, \epsilon^{-1}r)) < \epsilon
r. \]
Then for some $m < K$ there is a holomorphic function $s$ on
$\mathcal{U}$ such that
\[ \int_{\mathcal{U}} |s|^2 e^{-r^{-2}mu}\, (r^{-2}m \omega)^n < C, \]
and
\[ \Big| |s(z)| e^{-mr^{-2}u} - e^{-mr^{-2}d(p, z)^2/2}\Big| <
\zeta, \]
for $z\in \mathcal{U}$.
\end{prop}
If $V = \mathbb{R}^{2n-2}\times C(S^1_\gamma)$, where $S^1_\gamma$ is
a circle of length $\gamma < 2\pi$, then the singular set
$\mathbb{R}^{2n-2}\times \{0\}$ has capacity zero, so the conclusion
of Proposition~\ref{prop:LSz2} holds. As a result we obtain the
following, whose proof is essentially identical to \cite[Proposition
3.3]{LiuSz1}.
\begin{prop}\label{prop:codim2cone}
Suppose that we have a sequence $p_j\in \hat{\mathcal{U}}'$ and
$\lambda_j\to\infty$ such that the rescaled pointed sequence $(\hat{X},
\lambda_j d_{\hat{X}}, p_j)$ converges to
$\mathbb{R}^{2n-2}\times C(S^1_\gamma)$ in the pointed
Gromov-Hausdorff sense. Then for sufficiently large $j$
the map $\Phi$ in \eqref{eq:Phidefn} is an injection on
$B(p_j, \lambda_j^{-1})$, its image is in
$X^{reg}$, and in addition
\[ \lim_{j\to\infty} \int_{B(p_j, \lambda_j^{-1})} n\,
\mathrm{Ric}(\omega)\wedge \omega^{n-1} = \lambda_j^{2-2n}
\omega_{2n-2}(2\pi - \gamma), \]
where $\omega_{2n-2}$ is the volume of the unit ball in
$\mathbb{R}^{2n-2}$.
\end{prop}
Given this, we can argue exactly as in \cite[Proposition
5.1]{LiuSz1} to show that the singular set $V\setminus
\mathcal{R}_{\epsilon_3}(V)$ has capacity zero in any iterated tangent
cone of $\hat{X}$. Therefore the conclusion of
Proposition~\ref{prop:LSz2} holds for any iterated tangent cone. As a
consequence we can complete the proof of Theorem~\ref{thm:homeo2}.
\begin{proof}[Proof of Theorem~\ref{thm:homeo2}]
We can argue similarly to the end of the proof of
Proposition~\ref{prop:epsilonreg}.
Suppose that $\Phi(p_1) = \Phi(p_2)$. Then $p_1, p_2$ are contained in
the same neighborhood $\mathcal{U}' \subset\subset \mathcal{U}$. We can
assume that the K\"ahler potential $u \leq 0$, and $u(p_1) =
0$. Applying Proposition~\ref{prop:LSz2} to the tangent cone at $p_1$
with $\zeta = \frac{1}{10}$, and sufficiently small $r$, we can
construct a holomorphic function $s$ on
$\mathcal{U}$ such that $s(p_1) > 9/10$, while $s(p_2) < 2/10$. This
means that $\Phi(p_1) \not= \Phi(p_2)$.
\end{proof}
We will now prove the Hausdorff dimension estimates for $X\setminus
X^{reg}$ and $X\setminus X^\circ$, which will complete the proof of
Theorem~\ref{thm:homeo1}. Note that at this point we know that
$\hat{X}$ is homeomorphic to $X$ and so we identify the two spaces.
\begin{prop}
We have $\mathrm{dim}_{\mathcal{H}}(X\setminus X^{reg}) \leq 2n-4$,
and $\mathrm{dim}_{\mathcal{H}}(X\setminus X^\circ) \leq 2n-2$.
\end{prop}
\begin{proof}
Let us first consider $X\setminus
X^{reg}$. Proposition~\ref{prop:codim2cone} implies that if $p\in
X\setminus X^{reg}$, then no tangent cone at $p$ can split a factor
of $\mathbb{R}^{2n-2}$. We claim that if a tangent cone $V$ at $p$
splits a factor of $\mathbb{R}^{2n-3}$, then it also follows that
$p\in X^{reg}$. We expect more generally that in analogy with
Cheeger-Colding-Tian~\cite[Theorem 9.1]{CCT02}, if a tangent cone at
$p$ splits a factor of $\mathbb{R}^k$ for odd $k$, then in fact it splits
$\mathbb{R}^{k+1}$, however here we will just
deal with our special situation where $k=2n-3$.
Suppose that a tangent cone at $p\in \hat{X}$ splits a factor of
$\mathbb{R}^{2n-3}$, so is of the form $V = \mathbb{R}^{2n-3}\times
C(Y)$, where $Y$ is 2-dimensional.
Using the techniques of
Donaldson-Sun~\cite{DS17} and \cite{LiuSz2}, one can show that $V$
has the structure of a normal affine algebraic variety. We will
first show that in fact $V$ is smooth as a complex variety.
Away from
$\mathbb{R}^{2n-3}\times\{o\}$, where $o\in C(Z)$ is the vertex,
each tangent cone of $V$ splits at least a factor of
$\mathbb{R}^{2n-2}$, and so using Proposition~\ref{prop:codim2cone}
this part of $V$ has the structure of a smooth complex manifold. At
the same time, we claim that if $q\in \mathbb{R}^{2n-3}\times\{o\}$
were a singular point, then $V$ would be singular along all of
$\mathbb{R}^{2n-3}\times\{o\}$. This would contradict that by
normality of $V$ the singular set should have real codimension at
least 4. To see this, it is enough to show that the isometries of $V$ given
by translations in the $\mathbb{R}^{2n-3}$ factor preserve the
complex structure.
Let $f : V\to\mathbb{R}$ be the projection to one of the
$\mathbb{R}$ factors, and let $V' = V \setminus
(\mathbb{R}^{2n-3}\times \{o\})$. Let $\epsilon > 0$, $r < \epsilon^2$ and $B(q,
\epsilon^{-1}r)\subset V'$ satisfying \eqref{eq:d10} as in
Proposition~\ref{prop:coords}. As in the proof of that proposition,
we can realize $B(q, r)$ as the Gromov-Hausdorff limit
of a sequence of balls $B(q_k, \epsilon^{-1} r)$ in (incomplete) K\"ahler
manifolds $(M_k, g_k)$ with Ricci
curvature bounded below uniformly, and we can assume that the Ricci
curvature of $g_k$ is bounded below by $k^{-1}$. Using
\cite[Theorem 1.4]{LiuSz1} we can find holomorphic coordinates on the
$B(q_k, r)$, which converge to holomorphic coordinates on $B(q,r)$
under the Gromov-Hausdorff convergence
$B(q_k, r)\to B(q,r)$. Using these coordinates we can identify the
balls $B(q_k, r)$ with $B(q,r)$. Since the
function $f$ defines a splitting function on $B(q, r)$, it follows
that we have corresponding harmonic almost splitting functions $f_k$
on $B(q_k, r)$ converging to $f$, and satisfying
\[ \label{eq:intD2f} \fint_{B(q_k, r)} |\nabla^2 f_k|^2 \to 0. \]
We now work in analogy with the argument in \cite[Proposition 15]{LiuSz2}. Let
$v_k = \nabla f_k$, and $\sigma_{k, t}$ the diffeomorphisms
generated by $v_k$. Write $v_k^{(1,0)}$ for the $(1,0)$-part of
$v_k$. Arguing as in \cite{LiuSz2}, we can approximate the $v_k$ with
holomorphic vector fields $\tilde{v}_k$, generating biholomorphisms
$\tilde{\sigma}_{k, t}$, and in the limit as $k\to \infty$, the
limiting diffeomorphisms $\sigma_t$ of the $\sigma_{k,t}$ coincide with the
biholomorphisms $\tilde{\sigma}_t$ given by the limit of
$\tilde{\sigma}_{k,t}$.
At the same time the
diffeomorphisms $\sigma_t$ are given by translation in the
$\mathbb{R}$-factor corresponding to the splitting function $f$.
It follows that $V$ cannot have any singularities, since otherwise
the singular set would have real codimension at most 3.
So, as claimed, $V$ is a smooth complex variety. Let
$z_1,\ldots, z_n$ be a coordinate chart in a neighborhood of the
vertex in $V$. Using the H\"ormander $L^2$-estimates we can use
these to construct a holomorphic chart in a sufficiently small
neighborhood of $p\in X$, and so $p\in X^{reg}$ as claimed. It
follows that tangent cones at points $p\in X\setminus X^{reg}$
can split at most a factor of $\mathbb{R}^{2n-4}$, and so by the
dimension estimates of De Philippis-Gigli~\cite{DPG18}, the
Hausdorff dimension of $X\setminus X^{reg}$ is at most $2n-4$.
Let us now consider the claim that
$\mathrm{dim}_{\mathcal{H}}(X\setminus X^\circ) \leq 2n-2$. Note
that $X\setminus X^\circ \subset (D\cap
\mathcal{R}) \cup (\hat{X}\setminus
\mathcal{R})$, where $\mathcal{R}$ denotes the metric regular set (where the
tangent cones are $\mathbb{R}^{2n}$). The dimension estimate $\dim_{\mathcal{H}}(\hat{X}\setminus
\mathcal{R}) \leq 2n-2$ follows from De
Philippis-Gigli~\cite{DPG18}. Suppose that we have $\dim_{\mathcal{H}}(D\cap
\mathcal{R}) > 2n-2$, so that there is some $\ell>
2n-2$ and a point of density $x\in (D\cap
\mathcal{R})$ for the $\ell$-dimensional
Hausdorff measure. As in the proof of
Proposition~\ref{prop:epsilonreg} we can now exploit that $D$ is cut out
by a holomorphic function of finite order of vanishing. Let $r_k \to
0$, and consider the rescaled balls $r_k^{-1}B(x, r_k)$ converging
to the Euclidean unit ball $B(0, 1)$ in the tangent cone at $x$. We
can assume that $D\cap B(x,r_1) = s^{-1}(0)$ for a holomorphic
function $s$ on $B(x,r_1)$. Let us denote by $s_k$ the holomorphic
function $s$ normalized to have $L^2$-norm 1 on the rescaled ball
$r_k^{-1}B(x, r_k)$. As in the proof of Proposition~\ref{prop:epsilonreg} we can assume that
up to choosing a subsequence, after suitable normalizations the
$s_k$ converge to a non-zero holomorphic function $s_\infty$ on
$B(0,1)\subset \mathbb{C}^n$, and under the Gromov-Hausdorff
convergence the points of $D\cap B(x,r_k)$ converge to the zero set
$s_{\infty}^{-1}(0)$. Using that $x$ was a point of density for the
$\ell$-dimensional Hausdorff measure, it follows that $s_{\infty}^{-1}(0)$ has Hausdorff
dimension at least $\ell$, if the $r_k$ are chosen suitably. But this contradicts that the zero set of
a non-zero holomorphic function has Hausdorff dimension at most $2n-2$.
\end{proof}
\section{Collapsing Calabi-Yau metrics}\label{sec:collapsing}
In this section our goal is to prove Theorem~\ref{thm:GHlimit1},
by showing that the setting of collapsing
Calabi-Yau metrics fits into the
assumptions of Theorem~\ref{thm:homeo1}. Recall that $M$ is a (smooth)
projective manifold of dimension $m$ with trivial first Chern class,
together with a semiample line bundle $L$, giving rise
to a holomorphic map $f : M\to X$, where $\dim X = n$. We let $A$ denote an ample line
bundle on $M$, and consider the Ricci flat metrics $\omega_t \in
c_1(L) + t c_1(A)$ for $t > 0$, whose existence is provided by Yau's
Theorem~\cite{Yau78}.
More precisely, we let $\Omega$ be a smooth
volume form on $M$ which satisfies $\mathrm{Ric}(\Omega)=0$, and
$\int_M \Omega = 1$. Choosing suitable metrics $\omega_M\in c_1(A),
\omega_X\in c_1(L)$ on
$M, X$, we can write $\omega_t = f^*\omega_X + t\omega_M + \ddbar\psi_t$,
satisfying the Monge-Amp\`ere equation
\[ \omega_t^m = t^{m-n} c_t \Omega. \]
Integrating, we find that the normalizing constants $c_t$ satisfy
\[c_0 := \lim_{t\to 0} c_t = \binom{m}{n} c_1(L)^n \cup
c_1(A)^{m-n}. \]
There is a complex subvariety $B\subset X$, such that $f$ is a
holomorphic submersion over $X^\circ = X \setminus B$, and $X^\circ$
is smooth. It was shown in Song-Tian~\cite{ST12} that $X$ admits a
twisted K\"ahler-Einstein metric $\omega_{can} = \omega_X + \ddbar u$
with $u\in L^\infty$,
satisfying
\[ \mathrm{Ric}(\omega_{can}) = \omega_{WP} \]
on $X^\circ$. Here $\omega_{WP} \geq 0$ is the Weil-Petersson metric given
by the variation of the Calabi-Yau fibers of $f$ over $X^\circ$. On $X^\circ$ the metric
$\omega_{can}$ satisfies the Monge-Amp\`ere equation
\[ \label{eq:canMA} \omega_{can}^n = d_1 f_*\Omega = e^F \omega_X^n, \]
where $d_1 = \int_X \omega_X^n$.
It is shown \cite[Lemma 2.3]{GS21} that $F$ is bounded below, and
from \cite[Proposition 3.2]{ST12} we have that $e^F \in
L^p(\omega_X^n)$ for some $p > 1$ (note the differences between
our definition for $F$ and the ones used in these references).
It follows that on $X^{reg}$ we can
locally write $\mathrm{Ric}(\omega_{can}) = \ddbar v$, where $v\in
L^1_{loc}$, and $v$ is locally bounded from above. Since
$\mathrm{Ric}(\omega_{can}) \geq 0$ on $X^{\circ}$, it follows from
this that $\mathrm{Ric}(\omega_{can}) \geq 0$ on $X^{reg}$ in the
distributional sense. As a consequence, $(X, \omega_{can})$ satisfies
the conditions (1), (2), (3) in Theorem~\ref{thm:homeo1}, taking $D$
to be a divisor containing the subvariety $B$.
In order to prove Theorem~\ref{thm:GHlimit1}, it remains to show that
the metric completion $\hat{X}$ is an $RCD(0, 2n)$-space. Note
that $\hat{X}$ is an almost smooth metric measure space in the sense
of \cite[Definition 5]{SzRCD}, which is a slight modification of the
notion due to Honda~\cite{Honda18}. Just as in \cite[Proposition
9]{SzRCD}, by using the work of Honda~\cite{Honda18}, and
Guo-Phong-Song-Sturm~\cite{GPSS2}, we have the following.
\begin{prop}\label{prop:gradestimate}
Suppose that the $W^{1,2}$-eigenfunctions of the Laplacian on
$(X^\circ, \omega_{can})$ are Lipschitz. Then $\hat{X}$ is a
(non-collapsed) $RCD(0, 2n)$-space.
\end{prop}
In order to show that the eigenfunctions are Lipschitz, we will
exploit the fact that the metric space $\hat{X}$ is the
Gromov-Hausdorff limit of the Ricci flat manifolds $(M, \omega_t)$ as
$t\to 0$, according to Song-Tian-Zhang~\cite[Proposition
3.1]{STZ19}. As shown by Cheeger-Colding~\cite{CC1}, the
Gromov-Hausdorff limit can be equipped with a renormalized limit
measure $\nu$ which we now recall.
We fix basepoints $p_t\in (M,
\omega_t)$, and $p\in \hat{X}$, such that $p_t \to p$ under the
Gromov-Hausdorff convergence. Note that we can take $p\in X^\circ$,
and $p_t \in f^{-1}(p)$ to be independent of $t$, but we write $p_t$
to emphasize that we use the varying metrics $\omega_t$. Then for any $x\in
(M, \omega_t)$ and $r > 0$ we define the renormalized volume function to be
\[ \underline{V}(x,r) =
\frac{\mathrm{Vol}(B_{\omega_t}(x,r))}{\mathrm{Vol}(B_{\omega_t}(p_t,
1))}. \]
By \cite[Theorem 1.6]{CC1} we have a continuous function
$\underline{V}_\infty : \hat{X}\times \mathbb{R}_+ \to \mathbb{R}_+$
such that along a sequence $t_j\to 0$ we have the following: for every
$q_j \in (M, \omega_{t_j})$ with $q_j \to x\in \hat{X}$, and every $R
> 0$, we have
\[ \underline{V}(q_j, R) \to \underline{V}_\infty(x, R). \]
By \cite[Theorem 1.10]{CC1} the renormalized limit measure $\nu$ on
$\hat{X}$ is then uniquely defined by the property that $\nu(B(x, R))
= \underline{V}_\infty(x, R)$ for all $x\in \hat{X}$ and $R > 0$.
We next show that $\nu = v \omega_{can}^n$ for a constant $v > 0$. For this we can
follow arguments similar to Gross-Tosatti-Zhang~\cite[Lemma 5.2]{GTZ13} and
Fu-Guo-Song~\cite[Proof of Theorem 1.3]{FGS20}.
\begin{prop}\label{prop:nulimit}
Up to normalizing the measure, the metric measure space $(\hat{X},
d_{\hat{X}}, \omega_{can}^n)$ is the measured Gromov-Hausdorff
limit of the $(M, d_{\omega_t}, \omega_t^m)$.
\end{prop}
\begin{proof}
To show that $\nu = v\omega_{can}^n$ for a constant $v > 0$, it
is enough to show the following:
\begin{itemize}
\item[(1)] For any compact set $K\subset X^\circ$ there is an $r_K >
0$ such that
\[ \nu(B(x, r)) = v\omega_{can}^n(B(x,r))\, \quad \text{ for
} x\in K, r < r_K. \]
\item[(2)] We have $\nu(\hat{X}) = v\omega_{can}^n(X)$.
\end{itemize}
Indeed, from (1) it follows that for any compact set $K\subset
X^\circ$ we have $\nu(K) = v \omega_{can}^n(K)$. At the same
time we also have that the measure $\nu$ on $\hat{X}$ is obtained by
extending its restriction to $X^\circ$ trivially. This is because
for any $\epsilon > 0$ we can find a compact set $K\subset X^\circ$
such that $\omega_{can}^n(K) > \omega_{can}^n(X) -
\epsilon$. The conditions (1) and (2) above imply that $\nu(K) >
\nu(\hat{X}) - v\epsilon$.
To see the condition (1) above, we use the $C^0$-convergence of the
metrics $\omega_t$ to $f^*\omega_{can}$ on $f^{-1}(K)$ for any
compact $K\subset X^\circ$ (see \cite{TWY18}). For any $x\in
X^\circ$, we can choose $q_j = q \in f^{-1}(x)$ so that $q_j \to x$
under the Gromov-Hausdorff convergence. Using the $C^0$ convergence
above, we can find an $r_0 > 0$ and a function $h(t)\to 0$ such that
for any $r < r_0$ we have
\[ \label{eq:Bcontain} f^{-1}(B_{\omega_{can}}(x, r-h(t_j))) \subset
B_{\omega_{t_j}}(q_j, r) \subset f^{-1}(B_{\omega_{can}}(x, r +
h(t_j))). \]
For any $R > 0$ we also have
\[ \mathrm{Vol}(f^{-1}(B(x, R), \omega_t)) &=
\int_{f^{-1}(B(x,R))} \omega_t^m \\
&= t^{m-n}c_t \int_{f^{-1}(B(x,R))} \Omega \\
&= t^{m-n} c_t \int_{B(x,R)} f_*\Omega. \]
It follows that
\[ \lim_{t\to 0} t^{n-m} \mathrm{Vol}(f^{-1}(B(x,R)), \omega_t)
= c_0 \int_{B(x,R)} f_*\Omega = c_0 d_1^{-1} \mathrm{Vol}(B(x,R),
\omega_{can}). \]
Note that the ball $B(x,R)$ denotes the metric ball of radius $R$ in the metric
space $\hat{X}$.
It follows from this, together with \eqref{eq:Bcontain}, that for $r
< r_0$ we have
\[ \label{eq:limVB1} \lim_{j\to\infty} \frac{\mathrm{Vol}(B(q_j, r),
\omega_{t_j})}{t_j^{m-n}} = c_0d_1^{-1} \mathrm{Vol}(B(x,r),
\omega_{can}). \]
Note that the diameters of $(M, \omega_t)$ are uniformly bounded
from above by a constant $D > 0$ (see Tosatti~\cite{Tos09}, Zhang~\cite{Zhang06}), and at
the same time
\[ \frac{\mathrm{Vol}(M, \omega_t)}{t^{m-n}} \to c_0, \]
as $t\to 0$. It follows from volume comparison that up to choosing a
subsequence we have
\[ \label{eq:volcomp} \frac{\mathrm{Vol}(B(p_{t_j}, 1))}{\mathrm{Vol}(M, \omega_t)} \to
d_2\]
for some $d_2 > 0$. Using \eqref{eq:limVB1}, this implies that for
$r < r_0$ we have
\[ \nu(B(x,r)) = \lim_{j\to\infty} \frac{\mathrm{Vol}(B(q_j, r),
\omega_{t_j})}{\mathrm{Vol}(B(p_{t_j}, 1))} = \frac{1}{d_1d_2}
\mathrm{Vol}(B(x,r), \omega_{can}). \]
In addition, using \eqref{eq:volcomp} and \eqref{eq:canMA}, we have
\[ \nu(\hat{X}) = \lim_{j\to\infty} \frac{\mathrm{Vol}(B(q_j, D),
\omega_{t_j})}{\mathrm{Vol}(B(p_{t_j}, 1))} = \frac{1}{d_2} =
\frac{1}{d_1d_2} \mathrm{Vol}(M, \omega_{can}^n). \]
This completes the proof of properties (1) and (2) above, with $v=
(d_1d_2)^{-1}$.
\end{proof}
As a consequence we can complete the proof of
Theorem~\ref{thm:GHlimit1}.
\begin{proof}[Proof of Theorem~\ref{thm:GHlimit1}]
As shown by Song-Tian-Zhang~\cite{STZ19}, the Gromov-Hausdorff limit
$(Z, d_Z)$ coincides with the metric completion of $(X^\circ,
\omega_{can})$. We have seen above that $(X, \omega_{can})$ satisfies
the conditions (1), (2), (3) in the introduction. In addition $(Z,
d_Z)$, equipped with the measure $\omega_{can}^n$ is a measured
Gromov-Hausdorff limit of the Ricci flat manifolds $(M_t, \omega_t)$
as $t\to 0$ by Proposition~\ref{prop:nulimit}. It follows by
Cheeger-Colding~\cite{CC3} that the $W^{1,2}$ eigenfunctions of the Laplacian
on $(X^\circ, \omega_{can})$ are Lipschitz, and therefore by
Proposition~\ref{prop:gradestimate} the space $(Z, d_Z,
\omega_{can}^n)$ is an $RCD(0, 2n)$-space. It follows that we can
apply Theorem~\ref{thm:homeo1}, completing the proof.
\end{proof}
We now consider a closely related setting, namely the continuity
method introduced by La Nave-Tian~\cite{LNT16}. Suppose that $M$ is
projective, of dimension $m$, with semi ample canonical bundle
$K_M$. Let $X$ denote the canonical model of $M$, and $f: M\to X$ the
map induced by a power of $K_M$. Write $n = \dim X$, and assume
that $n\in (0,m]$. Similarly to the above, $f$ is a holomorphic
submersion over $X^\circ = X\setminus B$ for a subvariety $B$.
Let $A$ denote an ample line bundle
on $M$, and $\omega_M\in c_1(A)$ a metric. By \cite{LNT16}, for any
$t\in (0,1]$ we can find a unique $\omega_t\in c_1(K_X) + tc_1(A)$
satisfying the equation
\[ \label{eq:LNT} \mathrm{Ric}(\omega_t) = -\omega_t + t\omega_M. \]
To write this in terms of a Monge-Amp\`ere equation, we can choose a
smooth volume form $\Omega$ on $M$ such that $\mathrm{Ric}(\Omega) =
-f^*\omega_X$ for a smooth K\"ahler metric on $X$. We can then write
\[ \omega_t = f^*\omega_X + t\omega_M + \ddbar \psi_t, \]
where the $\psi_t$ satisfy the equations
\[ \omega_t^m = t^{m-n} e^{\psi_t}\Omega. \]
It was shown in Fu-Guo-Song~\cite{FGS20} that the diameter of $(M,
\omega_t)$ is uniformly bounded, so along a subsequence the $(M,
\omega_t)$ converge in the Gromov-Hausdorff sense to a compact metric
space $(Z,d_Z)$ as $t\to 0$. From Song-Tian-Zhang~\cite[Proposition
2.2]{STZ19} we know that, similarly to the collapsing Calabi-Yau
setting, $(Z,d_Z)$ is isometric to the metric
completion of $(X^\circ, \omega_{can})$, where $\omega_{can}$ is a
twisted K\"ahler-Einstein metric constructed by Song-Tian~\cite{ST12},
satisfying
\[ \mathrm{Ric}(\omega_{can}) = -\omega_{can} + \omega_{WP}. \]
As before, $\omega_{WP} \geq 0$ on $X^{\circ}$, determined by the
variation of the Calabi-Yau fibers of $f$. Similarly to the above,
$\omega_{can} = \omega_X + \ddbar\psi$, where $\psi$ is a bounded
solution of the Monge-Amp\`ere equation
\[ \omega_{can}^n = d_1 e^\psi f_*\Omega, \]
and $f^*\psi$ is the uniform limit of the $\psi_t$. The arguments from
the collapsing Calabi-Yau setting can easily be modified to obtain the
following.
\begin{thm}
The metric completion $\hat{X}$ of $(X^\circ, \omega_{can})$ defines
an $RCD(-1, 2n)$-space, and satisfies Conditions (1), (2), (3)
in Theorem~\ref{thm:homeo1}. As a consequence $\hat{X}$ is
homeomorphic to $X$, and satisfies the dimension estimates
$\dim_{\mathcal{H}}(X \setminus X^{reg}) \leq 2n-4$, and
$\dim_{\mathcal{H}}(X\setminus X^\circ) \leq 2n -2$.
\end{thm}
\bibliography{mybib}{}
\bibliographystyle{plain}
\end{document}
\begin{thebibliography}{10}
\bibitem{AGS14}
Luigi Ambrosio, Nicola Gigli, and Giuseppe Savar\'{e}.
\newblock Metric measure spaces with {R}iemannian {R}icci curvature bounded
from below.
\newblock {\em Duke Math. J.}, 163(7):1405--1490, 2014.
\bibitem{BK07}
Zbigniew B{\l }ocki and S{\l }awomir Ko{\l }odziej.
\newblock On regularization of plurisubharmonic functions on manifolds.
\newblock {\em Proc. Amer. Math. Soc.}, 135(7):2089--2093, 2007.
\bibitem{CCT02}
J.~Cheeger, T.~H. Colding, and G.~Tian.
\newblock On the singularities of spaces with bounded {R}icci curvature.
\newblock {\em Geom. Funct. Anal.}, 12(5):873--914, 2002.
\bibitem{CC1}
Jeff Cheeger and Tobias~H. Colding.
\newblock On the structure of spaces with {R}icci curvature bounded below. {I}.
\newblock {\em J. Differential Geom.}, 46(3):406--480, 1997.
\bibitem{CC3}
Jeff Cheeger and Tobias~H. Colding.
\newblock On the structure of spaces with {R}icci curvature bounded below.
{III}.
\newblock {\em J. Differential Geom.}, 54(1):37--74, 2000.
\bibitem{CJN21}
Jeff Cheeger, Wenshuai Jiang, and Aaron Naber.
\newblock Rectifiability of singular sets of noncollapsed limit spaces with
{R}icci curvature bounded below.
\newblock {\em Ann. of Math. (2)}, 193(2):407--538, 2021.
\bibitem{CDS2}
Xiuxiong Chen, Simon Donaldson, and Song Sun.
\newblock K\"{a}hler-{E}instein metrics on {F}ano manifolds. {II}: {L}imits
with cone angle less than {$2\pi$}.
\newblock {\em J. Amer. Math. Soc.}, 28(1):199--234, 2015.
\bibitem{AIM}
Yifan Chen, Shih-Kai Chiu, Max Hallgren, G{\'a}bor Sz{\'e}kelyhidi, Tat~Dat
T{\^o}, and Freid Tong.
\newblock On {K}\"ahler-{E}instein currents.
\newblock {\em arXiv:2502.09825}, 2025.
\bibitem{Dat16}
Ved~V. Datar.
\newblock On convexity of the regular set of conical {K}\"{a}hler-{E}instein
metrics.
\newblock {\em Math. Res. Lett.}, 23(1):105--126, 2016.
\bibitem{DPG18}
Guido De~Philippis and Nicola Gigli.
\newblock Non-collapsed spaces with {R}icci curvature bounded from below.
\newblock {\em J. \'{E}c. polytech. Math.}, 5:613--650, 2018.
\bibitem{Dem82}
Jean-Pierre Demailly.
\newblock Estimations {$L^{2}$} pour l'op\'{e}rateur {$\bar \partial $} d'un
fibr\'{e} vectoriel holomorphe semi-positif au-dessus d'une vari\'{e}t\'{e}
k\"{a}hl\'{e}rienne compl{\`e}te.
\newblock {\em Ann. Sci. \'{E}cole Norm. Sup. (4)}, 15(3):457--511, 1982.
\bibitem{DS14}
Simon Donaldson and Song Sun.
\newblock Gromov-{H}ausdorff limits of {K}\"{a}hler manifolds and algebraic
geometry.
\newblock {\em Acta Math.}, 213(1):63--106, 2014.
\bibitem{DS17}
Simon Donaldson and Song Sun.
\newblock Gromov-{H}ausdorff limits of {K}\"{a}hler manifolds and algebraic
geometry, {II}.
\newblock {\em J. Differential Geom.}, 107(2):327--371, 2017.
\bibitem{EKS15}
Matthias Erbar, Kazumasa Kuwada, and Karl-Theodor Sturm.
\newblock On the equivalence of the entropic curvature-dimension condition and
{B}ochner's inequality on metric measure spaces.
\newblock {\em Invent. Math.}, 201(3):993--1071, 2015.
\bibitem{EGZ09}
Philippe Eyssidieux, Vincent Guedj, and Ahmed Zeriahi.
\newblock Singular {K}\"{a}hler-{E}instein metrics.
\newblock {\em J. Amer. Math. Soc.}, 22(3):607--639, 2009.
\bibitem{FGS20}
Xin Fu, Bin Guo, and Jian Song.
\newblock Geometric estimates for complex {M}onge-{A}mp{\`e}re equations.
\newblock {\em J. Reine Angew. Math.}, 765:69--99, 2020.
\bibitem{FGS25}
Xin Fu, Bin Guo, and Jian Song.
\newblock {RCD structures on singular K\"ahler spaces of complex dimension
three}.
\newblock {\em arXiv:2503.08865}, 2025.
\bibitem{GTZ13}
Mark Gross, Valentino Tosatti, and Yuguang Zhang.
\newblock Collapsing of abelian fibered {C}alabi-{Y}au manifolds.
\newblock {\em Duke Math. J.}, 162(3):517--551, 2013.
\bibitem{GTZ16}
Mark Gross, Valentino Tosatti, and Yuguang Zhang.
\newblock Gromov-{H}ausdorff collapsing of {C}alabi-{Y}au manifolds.
\newblock {\em Comm. Anal. Geom.}, 24(1):93--113, 2016.
\bibitem{GW00}
Mark Gross and P.~M.~H. Wilson.
\newblock Large complex structure limits of {$K3$} surfaces.
\newblock {\em J. Differential Geom.}, 55(3):475--546, 2000.
\bibitem{GL25}
Vincent Guedj and Chinh~H. Lu.
\newblock Degenerate complex {H}essian equations on compact {H}ermitian
manifolds.
\newblock {\em Pure Appl. Math. Q.}, 21(3):1171--1194, 2025.
\bibitem{GPSS2}
Bin Guo, Duong~H. Phong, Jian Song, and Jacob Sturm.
\newblock {Sobolev inequalities on K\"ahler spaces}.
\newblock {\em arXiv:2311.00221}, 11 2023.
\bibitem{GS21}
Bin Guo and Jian Song.
\newblock Positivity of {W}eil-{P}etersson currents on canonical models.
\newblock {\em Pure Appl. Math. Q.}, 17(3):1045--1059, 2021.
\bibitem{GS25}
Bin Guo and Jian Song.
\newblock Nash entropy, {C}alabi energy and geometric regularization of
singular {K}\"ahler metrics.
\newblock {\em arXiv:2502.02041}, 2025.
\bibitem{HT20}
Hans-Joachim Hein and Valentino Tosatti.
\newblock Higher-order estimates for collapsing {C}alabi-{Y}au metrics.
\newblock {\em Camb. J. Math.}, 8(4):683--773, 2020.
\bibitem{HT25}
Hans-Joachim Hein and Valentino Tosatti.
\newblock Smooth asymptotics for collapsing {C}alabi-{Y}au metrics.
\newblock {\em Comm. Pure Appl. Math.}, 78(2):382--499, 2025.
\bibitem{Honda18}
Shouhei Honda.
\newblock Bakry-\'{E}mery conditions on almost smooth metric measure spaces.
\newblock {\em Anal. Geom. Metr. Spaces}, 6(1):129--145, 2018.
\bibitem{Jiang14}
Renjin Jiang.
\newblock Cheeger-harmonic functions in metric measure spaces revisited.
\newblock {\em J. Funct. Anal.}, 266(3):1373--1394, 2014.
\bibitem{LNT16}
Gabriele La~Nave and Gang Tian.
\newblock A continuity method to construct canonical metrics.
\newblock {\em Math. Ann.}, 365(3-4):911--921, 2016.
\bibitem{LT23}
Yang Li and Valentino Tosatti.
\newblock On the collapsing of {C}alabi-{Y}au manifolds and
{K}\"{a}hler-{R}icci flows.
\newblock {\em J. Reine Angew. Math.}, 800:155--192, 2023.
\bibitem{LiuSz2}
Gang Liu and G\'{a}bor Sz\'{e}kelyhidi.
\newblock Gromov-{H}ausdorff limits of {K}\"{a}hler manifolds with {R}icci
curvature bounded below {II}.
\newblock {\em Comm. Pure Appl. Math.}, 74(5):909--931, 2021.
\bibitem{LiuSz1}
Gang Liu and G\'{a}bor Sz\'{e}kelyhidi.
\newblock Gromov-{H}ausdorff limits of {K}\"{a}hler manifolds with {R}icci
curvature bounded below.
\newblock {\em Geom. Funct. Anal.}, 32(2):236--279, 2022.
\bibitem{LV09}
John Lott and C\'{e}dric Villani.
\newblock Ricci curvature for metric-measure spaces via optimal transport.
\newblock {\em Ann. of Math. (2)}, 169(3):903--991, 2009.
\bibitem{Song14}
Jian Song.
\newblock {Riemannian geometry of K\"ahler-Einstein currents}.
\newblock {\em arXiv:1404.0445}, 04 2014.
\bibitem{ST12}
Jian Song and Gang Tian.
\newblock Canonical measures and {K}\"{a}hler-{R}icci flow.
\newblock {\em J. Amer. Math. Soc.}, 25(2):303--353, 2012.
\bibitem{STZ19}
Jian Song, Gang Tian, and Zhenlei Zhang.
\newblock Collapsing behavior of {R}icci-flat {K}\"ahler metrics and long time
solutions of the {K}\"ahler-{R}icci flow.
\newblock {\em arXiv:1904.08345}, 2019.
\bibitem{Sturm06}
Karl-Theodor Sturm.
\newblock On the geometry of metric measure spaces. {I}.
\newblock {\em Acta Math.}, 196(1):65--131, 2006.
\bibitem{SzRCD}
G{\'a}bor Sz{\'e}kelyhidi.
\newblock Singular {K}\"ahler-{E}instein metrics and {RCD} spaces.
\newblock {\em arXiv:2408.10747}, 2024.
\bibitem{Tos09}
Valentino Tosatti.
\newblock Limits of {C}alabi-{Y}au metrics when the {K}\"{a}hler class
degenerates.
\newblock {\em J. Eur. Math. Soc. (JEMS)}, 11(4):755--776, 2009.
\bibitem{Tos10}
Valentino Tosatti.
\newblock Adiabatic limits of {R}icci-flat {K}\"{a}hler metrics.
\newblock {\em J. Differential Geom.}, 84(2):427--453, 2010.
\bibitem{Tos11}
Valentino Tosatti.
\newblock Degenerations of {C}alabi-{Y}au metrics.
\newblock In {\em Geometry and Physics in Cracow}, volume~4 of {\em Acta Phys.
Polon. B Proc. Suppl.}, pages 495--505. 2011.
\bibitem{Tos20}
Valentino Tosatti.
\newblock Collapsing {C}alabi-{Y}au manifolds.
\newblock In {\em Surveys in differential geometry 2018. {D}ifferential
geometry, {C}alabi-{Y}au theory, and general relativity}, volume~23 of {\em
Surv. Differ. Geom.}, pages 305--337. Int. Press, Boston, MA, [2020]
\copyright 2020.
\bibitem{TWY18}
Valentino Tosatti, Ben Weinkove, and Xiaokui Yang.
\newblock The {K}\"{a}hler-{R}icci flow, {R}icci-flat metrics and collapsing
limits.
\newblock {\em Amer. J. Math.}, 140(3):653--698, 2018.
\bibitem{TZ20}
Valentino Tosatti and Yuguang Zhang.
\newblock Collapsing {H}yperk\"{a}hler manifolds.
\newblock {\em Ann. Sci. \'{E}c. Norm. Sup\'{e}r. (4)}, 53(3):751--786, 2020.
\bibitem{Yau78}
Shing~Tung Yau.
\newblock On the {R}icci curvature of a compact {K}\"{a}hler manifold and the
complex {M}onge-{A}mp{\`e}re equation. {I}.
\newblock {\em Comm. Pure Appl. Math.}, 31(3):339--411, 1978.
\bibitem{YauProblems}
Shing-Tung Yau.
\newblock Open problems in geometry.
\newblock In {\em Differential geometry: partial differential equations on
manifolds ({L}os {A}ngeles, {CA}, 1990)}, volume~54 of {\em Proc. Sympos.
Pure Math.}, pages 1--28. Amer. Math. Soc., Providence, RI, 1993.
\bibitem{Zhang06}
Yuguang Zhang.
\newblock Convergence of {K}\"ahler manifolds and calibrated fibrations.
\newblock {\em PhD Thesis}, 2006.
\end{thebibliography}