This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Minimal extension property of direct images

Chen Zhao [email protected] School of Mathematical Sciences, University of Science and Technology of China, Hefei, 230026, China
Abstract.

Given a projective morphism f:XYf:X\to Y from a complex space to a complex manifold, we prove the Griffiths semi-positivity and minimal extension property of the direct image sheaf f()f_{\ast}(\mathscr{F}). Here, \mathscr{F} is a coherent sheaf on XX, which consists of the Grauert-Riemenschneider dualizing sheaf, a multiplier ideal sheaf, and a variation of Hodge structure (or more generally, a tame harmonic bundle).

1. introduction

Given a holomorphic map between projective manifolds f:XYf:X\to Y and a holomorphic line bundle LL endowed with a singular Hermitian metric hh, the positivity of f(ωX/YL(h))f_{\ast}(\omega_{X/Y}\otimes L\otimes\mathscr{I}(h)) has been a topic of great interest in decades. This positivity problem plays crucial roles in many subjects in complex algebraic geometry such as Iitaka Cn,mC_{n,m} conjecture [Kawamata1985, Viehweg1983, CP2017] and the moduli of projective varieties [Viehweg1995].

A significant breakthrough in recent years has been the recognition of Nakano semi-positivity (also known as the minimal extension property) of f(ωX/YL(h))f_{\ast}(\omega_{X/Y}\otimes L\otimes\mathscr{I}(h)), attributed to Berndtsson [Berdtsson2009], Paun-Takayama [PT2018], Hacon-Popa-Schnell [HPS2018]. This observation has enabled Cao-Paun [CP2017] to solve the Iitaka Cn,mC_{n,m} conjecture over abelian varieties (see also [HPS2018]). The purpose of this article is to demonstrate that the minimal extension property holds for a much broader class of direct images. In this case, the holomorphic line bundle (L,h)(L,h) could be replaced by degenerate bundles, such as a variation of Hodge structure or, a tame harmonic bundle in a broader context (corresponding to certain parabolic Higgs bundles, see Simpson [Simpson1988, Simpson1990] and Mochizuki [Mochizuki20072, Mochizuki20071]).

1.1. Main result

Let f:XYf:X\rightarrow Y be a projective surjective morphism from a complex space XX111All complex space is assumed to be reduced and irreducible. to a complex manifold YY. Let XoXX^{o}\subset X be a dense Zariski open subset and (E,h)(E,h) be a holomorphic vector bundle endowed with a singular Hermitian metric (Definition 2.2). Let SX(E,h)S_{X}(E,h) be the sheaf on XX defined as follows. Let UXU\subset X be an open subset. The space SX(E,h)(U)S_{X}(E,h)(U) consists of holomorphic EE-valued (n,0)(n,0)-forms α\alpha on UXoU\cap X^{o} such that {α,α}\{\alpha,\alpha\} is locally integrable near every point of UU. We define SX/Y(E,h)S_{X/Y}(E,h) as SX(E,h)f(ωY1)S_{X}(E,h)\otimes f^{\ast}(\omega_{Y}^{-1}).

Theorem 1.1.

If the holomorphic vector bundle (E,h)(E,h) is tame on XX (Definition 3.5) and Θh(E)Naks0\Theta_{h}(E)\geq_{\rm Nak}^{s}0 on XoX^{o} (Definition 2.5), then SX/Y(E,h)S_{X/Y}(E,h) is a coherent sheaf on XX and the direct image sheaf

=f(SX/Y(E,h))\mathscr{F}=f_{\ast}(S_{X/Y}(E,h))

has a canonical singular Hermitian metric which is Griffiths semi-positive and satisfies the minimal extension property.

Remark 1.2.

We adopt Caltaldo’s concept of Nakano semi-positivity for singular Hermitian metrics (see Definition 2.5) because it allows for the validity of Hörmander’s estimate ([CataldoAndrea1998, Proposition 4.1.1], also see [Demailly1982, Théorème 5.1]) and the optimal Ohsawa-Takegoshi extension theorem (Guan-Zhou [Guan-Zhou2015], Guan-Mi-Yuan [GMY2023]). Defining ”Nakano semi-positivity” in a way that does not rely on approximations using C2C^{2} metrics, while still ensuring Hörmander’s estimate and the optimal Ohsawa-Takegoshi extension theorem, presents an intriguing challenge. Relevant works addressing this include [KP2021, DNW2021, DNWZ2023, DNZZ2024, In2022, PT2018, Rau2015].

Remark 1.3.

Since we are interested in the case when the vector bundle EE arises from a variation of Hodge structure or a tame harmonic bundle, we do not require EE to have a holomorphic extension to XX. It is possible that there will be some fiber of ff where EE is nowhere defined, and hh may not extend to XX. This presents the main difficulty in this article compared to known works. The primary contribution of this article is the introduction of the ”tame” condition (Definition 3.5), which is motivated by the theory of degeneration of Hodge structure (Schmid [Schmid1973], Cattani-Kaplan-Schmid [Cattani_Kaplan_Schmid1986]) and the theory of tame harmonic bundles (Simpson [Simpson1990], Mochizuki [Mochizuki20072, Mochizuki20071]). Roughly speaking, the tameness of the Hermitian vector bundle (E,h)(E,h) means that the dual metric hh^{\ast} has at most polynomial growth at every point on XX. This condition allows for the use of techniques used in Paun-Takayama [PT2018] and Hacon-Popa-Schnell [HPS2018] on the degenerate loci of (E,h)(E,h). We want to point out that if (E,h)(E,h) is Nakano semi-positive on XoX^{o}, then it is also tame on XoX^{o}. Therefore, the main concern of the tameness condition in Theorem 1.1 is the asymptotic behavior of the metric hh on the boundary X\XoX\backslash X^{o}.

Remark 1.4.

The construction of SX(E,h)S_{X}(E,h) was introduced in [SZ2022] and [SC2021]. It offers a convenient way to combine Hodge-theoretic objects, like the Kollár-Saito SS-sheaf, with transcendental objects, such as the multiplier ideal sheaf. Typically, under certain conditions, SX(E,h)S_{X}(E,h) exhibits good Hodge-theoretic properties, such as Kollár’s package (see [SZ2022]), as well as good transcendental properties, including the strong openness property and the Ohsawa-Takegoshi extension property (see [SC2021]).

1.2. Example: multiplier ideal sheaf

The first example is the case when the bundle EE do not degenerate. When X=XoX=X^{o} (XX is smooth in particular), EE is a holomorphic line bundle with a singular Hermitian metric of semi-positive curvature, the aforementioned theorem implies the positivity result of the direct image sheaf f(ωX/YE(h))f_{\ast}(\omega_{X/Y}\otimes E\otimes\mathscr{I}(h)), as proven by Paun-Takayama [PT2018] and Hacon-Popa-Schnell [HPS2018, Theorem 21.1]. If EE is of higher rank and hh is a metric satisfying conditions in Theorem 1.1, then using Caltaldo’s notation E(h)E(h) (consisting of locally L2L^{2} holomorphic sections in EE), f(ωX/YE(h))f_{\ast}(\omega_{X/Y}\otimes E(h)) has a canonical singular Hermitian metric which is Griffiths semi-positive and satisfies the minimal extension property.

1.3. Example: multiplier SS-sheaf

Let 𝕍\mathbb{V} be a variation of Hodge structure on a regular Zariski open subset of a projective variety XX. Kollár [Kollar1986_2] introduced a coherent sheaf S(ICX(𝕍))S(IC_{X}(\mathbb{V})) that generalizes the dualizing sheaf. He conjectured that S(ICX(𝕍))S(IC_{X}(\mathbb{V})) satisfies Kollár’s package (including the torsion-freeness, the injectivity theorem, Kollár’s vanishing theorem and the decomposition theorem). This conjecture was subsequently proven by Saito [MSaito1991] using the theory of mixed Hodge modules. Saito’s proof is based on the observation that S(ICX(𝕍))S(IC_{X}(\mathbb{V})) represents the highest index Hodge component of the intermediate extension ICX(𝕍)IC_{X}(\mathbb{V}) as a Hodge module. In [SZ2022], the authors provide a new proof of Kollár’s conjecture using the L2L^{2}-method. This is based on their observation that S(ICX(𝕍))S(IC_{X}(\mathbb{V})) is isomorphic to some SX(E,h)S_{X}(E,h) for certain Hermitian bundle (E,h)(E,h) arising from the variation of Hodge structure (see below). The SS-sheaf has played a crucial role in the application of Hodge module theory to complex algebraic geometry (see [Popa2018] for a comprehensive survey).

Let XX be a complex space and XoXregX^{o}\subset X_{\rm reg} a Zariski open subset. Let 𝕍:=(𝒱,,,Q)\mathbb{V}:=(\mathcal{V},\nabla,\mathcal{F}^{\bullet},Q) be a polarized complex variation of Hodge structure on XoX^{o}. To establish the Nadel vanishing theorem for S(ICX(𝕍))S(IC_{X}(\mathbb{V})), the authors of [SC2021] introduce a multiplier SS-sheaf, denoted as S(ICX(𝕍),φ)S(IC_{X}(\mathbb{V}),\varphi), a combination of the SS-sheaf S(ICX(𝕍))S(IC_{X}(\mathbb{V})) and the multiplier ideal sheaf associated with a quasi-plurisubharmonic (quasi-psh) function φ:X[,)\varphi:X\to[-\infty,\infty). Let hQh_{Q} be the Hodge metric defined as (u,v)hQ:=Q(Cu,v¯)(u,v)_{h_{Q}}:=Q(Cu,\overline{v}), where QQ is the polarization of 𝕍\mathbb{V} and CC is the Weil operator. The multiplier SS-sheaf is then defined as

S(ICX(𝕍),φ):=SX(S(𝕍),eφhQ),S(IC_{X}(\mathbb{V}),\varphi):=S_{X}(S(\mathbb{V}),e^{-\varphi}h_{Q}),

where S(𝕍):=max{k|k0}S(\mathbb{V}):=\mathcal{F}^{\max\{k|\mathcal{F}^{k}\neq 0\}} is the top indexed nonzero piece of the Hodge filtration \mathcal{F}^{\bullet}. This type of sheaf possesses several good properties, such as the strong openness property and the Ohsawa-Takegoshi extension property. Additionally, S(ICX(𝕍),0)=S(ICX(𝕍))S(IC_{X}(\mathbb{V}),0)=S(IC_{X}(\mathbb{V})). For more details on the multiplier SS-sheaf and its relation to S(ICX(𝕍))S(IC_{X}(\mathbb{V})) and (φ)\mathscr{I}(\varphi), readers may refer to [SC2021].

Let SX/Y(ICX(𝕍),φ):=S(ICX(𝕍),φ)f(ωY1)S_{X/Y}(IC_{X}(\mathbb{V}),\varphi):=S(IC_{X}(\mathbb{V}),\varphi)\otimes f^{\ast}(\omega_{Y}^{-1}). As a consequence of Theorem 1.6, we obtain the following.

Theorem 1.5.

Let (E,eφh)(E,e^{-\varphi}h) be a Hermitian vector bundle on XX such that φ\varphi is a quasi-psh function on XX and hh is a smooth metric. Assume that 1¯(φ)+1Θh(E)Nak0\sqrt{-1}\partial\bar{\partial}(\varphi)+\sqrt{-1}\Theta_{h}(E)\geq_{\rm Nak}0. Let f:XYf:X\to Y be a surjective projective morphism to a complex manifold YY. Then

f(SX/Y(ICX(𝕍),φ)E)f_{\ast}\left(S_{X/Y}(IC_{X}(\mathbb{V}),\varphi)\otimes E\right)

has a singular Hermitian metric which is Griffiths semi-positive and satisfies the minimal extension property.

In particular, if EE is a holomorphic vector bundle on XX endowed with a smooth Hermitian metric with Nakano semi-positive curvature (i.e., φ=0\varphi=0 in Theorem 1.5), then f(SX/Y(ICX(𝕍))E)f_{\ast}\left(S_{X/Y}(IC_{X}(\mathbb{V}))\otimes E\right) has a singular Hermitian metric that is Griffiths semi-positive and satisfies the minimal extension property. This generalizes the result of Schnell-Yang [SY2023] to the relative case.

1.4. Example: parabolic Higgs bundle

The concept of the multiplier SS-sheaf, as shown in the previous example, can be extended to the framework of non-abelian Hodge theory. This extension of the multiplier SS-sheaf is elaborated on in [SZ2022].

Let XX be a smooth, projective variety, and let DD be a reduced simple normal crossing divisor on XX. Let’s consider a locally abelian parabolic Higgs bundle (H,{HE}EDivD(X),θ)(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)},\theta) on (X,D)(X,D). This bundle consists of the following data:

  • A locally abelian parabolic vector bundle (H,{HE}EDivD(X))(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)}) with parabolic structures on DD. Here, the filtration {HE}\{{{}_{E}}H\} is indexed by the set DivD(X){\rm Div}_{D}(X), which consists of \mathbb{R}-divisors whose support lies in DD.

  • A Higgs field θ:H|X\DH|X\DΩX\D\theta:H|_{X\backslash D}\to H|_{X\backslash D}\otimes\Omega_{X\backslash D}, which has regular singularity along DD.

This parabolic Higgs bundle is required to have vanishing parabolic Chern classes and to be polystable with respect to an ample line bundle AA on XX.

The main focus of this study is to examine a specific extension, denoted as PE,(2)(H)P_{E,(2)}(H), of H|X\DH|_{X\backslash D}. To define this extension, let EE be an \mathbb{R}-divisor supported on DD. We denote H<E{{}_{<E}}H as E<EHE\cup_{E^{\prime}<E}{{}_{E^{\prime}}}H. The coherent sheaf PE,(2)(H)P_{E,(2)}(H) is determined by the following conditions.

  1. (1)

    H<EPE,(2)(H)HE{{}_{<E}}H\subset P_{E,(2)}(H)\subset{{}_{E}}H.

  2. (2)

    Let xx be a point in DD and (U;z1,,zn)(U;z_{1},\dots,z_{n}) be holomorphic local coordinates on some open neighborhood UU of xx in XX, such that D={z1zr=0}D=\{z_{1}\cdots z_{r}=0\}. We denote Di={zi=0}D_{i}=\{z_{i}=0\} for i=1,,ri=1,\dots,r. Now, let {Wm,i(HE)}m\{W_{m,i}({{}_{E}}H)\}_{m\in\mathbb{Z}} be the monodromy weight filtration on HE|U{{}_{E}}H|_{U} at xx, with respect to the nilpotent part of the residue map ResDi(θ){\rm Res}_{D_{i}}(\theta) of the Higgs field along DiD_{i}. Then, we have:

    (1.1) PE,(2)(H)|U=H<E+i=1rW2,i(HE).\displaystyle P_{E,(2)}(H)|_{U}={{}_{<E}}H+\bigcap_{i=1}^{r}W_{-2,i}({{}_{E}}H).

When E=DE=D, PD,(2)(H)P_{D,(2)}(H) represents the sheaf of L2L^{2}-holomorphic sections with coefficients in HH. This construction was originally introduced by S. Zucker [Zucker1979] for algebraic curves, and it involves HH arising from a variation of Hodge structure. Consequently, it is a significant subject of study in the context of L2L^{2}-cohomology of a variation of Hodge structure.

To extend Zucker’s construction [Zucker1979] to higher-dimensional bases and non-canonical indexed extensions, we introduce PE,(2)(H)P_{E,(2)}(H). Specifically, when E0E\geq 0, PDE,(2)(H)P_{D-E,(2)}(H) combines elements from both PD,(2)(H)P_{D,(2)}(H) and the multiplier ideal sheaf associated with EE. This aspect makes PE,(2)(H)P_{E,(2)}(H) more convenient in applications where EDE\neq D. It can be proven that PE,(2)(H)P_{E,(2)}(H) is always locally free.

According to the non-abelian Hodge theory of Simpson [Simpson1988, Simpson1990] and Mochizuki [Mochizuki2006, Mochizuki20071], a μA\mu_{A}-polystable regular parabolic flat bundle (V,{VE}EDivD(X),)(V,\{{{}_{E}}V\}_{E\in{\rm Div}_{D}(X)},\nabla) is associated with the parabolic Higgs bundle (H,{HE}EDivD(X),θ)(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)},\theta). Moreover, there exists an isomorphism between the CC^{\infty} complex bundles:

ρ:H|X\D𝒪X\D𝒞X\D=V|X\D𝒪X\D𝒞X\D.\rho:H|_{X\backslash D}\otimes_{\mathscr{O}_{X\backslash D}}\mathscr{C}^{\infty}_{X\backslash D}=V|_{X\backslash D}\otimes_{\mathscr{O}_{X\backslash D}}\mathscr{C}^{\infty}_{X\backslash D}.

In particular, the CC^{\infty} complex bundle associated with H|X\DH|_{X\backslash D} has two complex structures: ¯\bar{\partial}, the complex structure of the Higgs bundle H|X\DH|_{X\backslash D}, and 0,1\nabla^{0,1}, the (0,1)(0,1)-part of \nabla in the flat bundle V|X\DV|_{X\backslash D}.

In this setting, Theorem 1.1 implies the following result.

Theorem 1.6.

Let KK be a locally free subsheaf of H|X\DH|_{X\backslash D} satisfying the following conditions:

  • Holomorphicity: 0,1(K)=0\nabla^{0,1}(K)=0, meaning that KK is holomorphic with respect to both the complex structures ¯\bar{\partial} and 0,1\nabla^{0,1}.

  • Weak transversality222This condition is referred to as weak transversality due to Griffiths’s transversality when HH arises from a variation of Hodge structure with {Fp}p\{F^{p}\}_{p\in\mathbb{Z}} as the Hodge filtration and K=FpK=F^{p} for some pp.: (θ)(K)K𝒜X\D1,0(\nabla-\theta)(K)\subset K\otimes\mathscr{A}^{1,0}_{X\backslash D}.

Let LL be a line bundle on XX such that LB+NL\simeq_{\mathbb{R}}B+N, where BB is a semi-positive \mathbb{R}-divisor and NN is an \mathbb{R}-divisor on XX supported on DD. Let FF be a Nakano semi-positive vector bundle on XX. Let j:X\DXj:X\backslash D\to X be the immersion, and let f:XYf:X\to Y be a surjective projective morphism to a complex manifold YY.

Then, the sheaf f(ωX/Y(PDN,(2)(H)jK)FL)f_{\ast}\left(\omega_{X/Y}\otimes(P_{D-N,(2)}(H)\cap j_{\ast}K)\otimes F\otimes L\right) has a singular Hermitian metric which is Griffiths semi-positive and satisfies the minimal extension property.

In [SZtwisted], the authors demonstrate that ωX(PDN,(2)(H)jK)FL\omega_{X}\otimes(P_{D-N,(2)}(H)\cap j_{\ast}K)\otimes F\otimes L satisfies Kollár’s package. In particular, they establish that f(ωX/Y(PDN,(2)(H)jK)FL)f_{\ast}\left(\omega_{X/Y}\otimes(P_{D-N,(2)}(H)\cap j_{\ast}K)\otimes F\otimes L\right) is weakly positive in the sense of Viehweg.

Remark 1.7.

Let (H,{HE}EDivD(X),θ)(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)},\theta) be the parabolic Higgs bundle associated with a variation of Hodge structure and let K=S(𝕍)K=S(\mathbb{V}). Let φN\varphi_{N} be the weight quasi-psh function associated with the divisor NN. Then P(H)DN,(2)jKP{{}_{D-N,(2)}}(H)\cap j_{\ast}K is coincide with the multiplier SS-sheaf S(ICX(𝕍),φN)S(IC_{X}(\mathbb{V}),\varphi_{N}).

1.5. Example: parabolic bundle

Let XX be a smooth projective variety and DXD\subset X be a simple normal crossing divisor on XX. Let (H,{HE}EDivD(X))(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)}) be a locally abelian parabolic bundle on (X,D)(X,D) with vanishing parabolic Chern classes. This bundle is also polystable with respect to an ample line bundle AA on XX. In this case, we can consider (H,{HE}EDivD(X))(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)}) as a parabolic Higgs bundle with a vanishing Higgs field. Consequently, PE,(2)(H)=H<EP_{E,(2)}(H)={{}_{<E}}H. By selecting K=H|X\DK=H|_{X\backslash D} in Theorem 1.1, the conditions of holomorphicity and weak transversality are satisfied for KK. Therefore, Theorem 1.6 implies the following.

Theorem 1.8.

Let XX be a smooth, projective variety and DD a simple normal crossing divisor on XX. Let (H,{HE}EDivD(X))(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)}) be a locally abelian parabolic bundle on (X,D)(X,D) with vanishing parabolic Chern classes, which is polystable with respect to an ample line bundle AA on XX. Let LL be a line bundle on XX such that LB+NL\simeq_{\mathbb{R}}B+N, where BB is a semi-positive \mathbb{R}-divisor and NN is an \mathbb{R}-divisor on XX supported on DD. Let FF be an arbitrary Nakano semi-positive vector bundle on XX.

Let f:XYf:X\to Y be a surjective projective morphism to a complex manifold YY. Then the direct image sheaf f(ωX/YH<DNFL)f_{\ast}\left(\omega_{X/Y}\otimes{{}_{<D-N}}H\otimes F\otimes L\right) has a singular Hermitian metric which is Griffiths semi-positive and satisfies the minimal extension property.

This article is structured as follows. In Section 2, we provide a review of basic concepts such as a singular metric on a torsion-free sheaf, Caltaldo’s notion of Nakano semi-positivity, Hacon-Popa-Schnell’s notion of minimal extension property, and Guan-Mi-Yuan’s optimal Ohsawa-Takegoshi extension theorem. Section 3 introduces and examines SX(E,h)S_{X}(E,h), while also explaining its connection to significant transcendental and Hodge theoretic objects. The main result is demonstrated in Section 4.

2. preliminary

2.1. Positivity of singular Hermitian metrics on a torsion free coherent sheaf

In this subsection, we review the notion of Nakano/Griffiths positivity for singular Hermitian metrics.

Throughout this subsection, let XX be a complex manifold of dimension nn and EE be a vector bundle of rank rr on XX.

Definition 2.1 (Nakano positivity and Griffiths positivity).

A C2C^{2} smooth Hermitian metric hh on EE defines the Chern curvature and associated Hermitian form:

(2.1) 1ΘhC0(X,Λ1,1End(E))and1Θ~hC0(X,Herm(TXE)).\displaystyle\sqrt{-1}\Theta_{h}\in C^{0}(X,\Lambda^{1,1}\otimes{\rm End}(E))\quad\textrm{and}\quad\sqrt{-1}\tilde{\Theta}_{h}\in C^{0}(X,{\rm Herm}(T_{X}\otimes E)).

The metric hh is said to be

  • Nakano semi-positive, denoted as Θh(E)Nak0\Theta_{h}(E)\geq_{\rm Nak}0, if for every uTXEu\in T_{X}\otimes E, it holds that 1Θ~h(u,u)0\sqrt{-1}\tilde{\Theta}_{h}(u,u)\geq 0.

  • Griffiths semi-positive, if for all ξTX\xi\in TX and sEs\in E, it holds that 1Θ~h(ξs,ξs)0\sqrt{-1}\tilde{\Theta}_{h}(\xi\otimes s,\xi\otimes s)\geq 0.

We review the singular version of the Griffiths positivity and Nakano positivity of a singular hermitian metric on a vector bundle. First, the concept of a singular Hermitian metric, as introduced by Berndtsson-Paun [BP2008], Paun-Takayama [PT2018], and Hacon-Popa-Schnell [HPS2018], is defined as follows.

Definition 2.2.

A singular Hermitian metric on a vector bundle EE is a function hh that associates to every point xXx\in X a singular hermitian inner product ||h,x:Ex[0,+]|-|_{h,x}:E_{x}\rightarrow[0,+\infty] on the complex vector space ExE_{x}, subject to the following two conditions:

  • hh is finite and positive definite almost everywhere, meaning that for all xx outside a set of measure zero, ||h,x|-|_{h,x} is a singular hermitian inner product on ExE_{x}.

  • hh is measurable, meaning that the function

    |s|h:U[0,+],x|s(x)|h,x|s|_{h}:U\rightarrow[0,+\infty],x\mapsto|s(x)|_{h,x}

    is measurable whenever UXU\subset X is open and sH0(U,E)s\in H^{0}(U,E).

Definition 2.3.

Let \mathscr{F} be a torsion free coherent sheaf on XX. Denote by X()XX(\mathscr{F})\subset X the maximal open subset where \mathscr{F} is locally free and let E:=|X()E:=\mathscr{F}|_{X(\mathscr{F})}. A singular Hermitian metric on \mathscr{F} is a singular Hermitian metric hh on the holomorphic vector bundle EE.

A singular Hermitian metric hh on EE induces a dual singular Hermitian metric hh^{\ast} on EE^{\ast}.

Definition 2.4.

A singular hermitian metric hh on a vector bundle EE is called Griffiths semi-positive if log|u|h2\log|u|^{2}_{h^{\ast}} is plurisubharmonic for any local holomorphic section uu of EE^{\ast}. A singular hermitian metric hh on a torsion free coherent sheaf \mathscr{F} is called Griffiths semi-positive if (|X(),h)(\mathscr{F}|_{X(\mathscr{F})},h) is Griffiths semi-positive.

When hh is a smooth Hermitian metric on EE, the above definition coincides with the classical Griffiths positivity.

The following definition, which can be regarded as the singular version of Nakano positivity, was introduced by Cataldo [CataldoAndrea1998] and Guan-Mi-Yuan [GMY2023].

Definition 2.5.

Let ω\omega be a Hermitian form on XX and θ\theta be a continuous real (1,1)(1,1)-form on XX. A singular Hermitian metric hh on EE is called θ\theta-Nakano semi-positive in the sense of approximations, denoted by

Θh(E)NaksθIdE\Theta_{h}(E)\geq_{\textrm{Nak}}^{s}\theta\otimes Id_{E}

if there is a collection of data (Σ,Xj,hj,s)(\Sigma,X_{j},h_{j,s}) satisfying that

  1. (1)

    ΣX\Sigma\subset X is a closed set of measure zero;

  2. (2)

    {Xj}j=1+\{X_{j}\}_{j=1}^{+\infty} is an open cover of XX made of sequence of relatively compact subsets of XX such that X1X2XjXj+1X_{1}\Subset X_{2}\Subset\cdots\Subset X_{j}\Subset X_{j+1}\Subset\cdots;

  3. (3)

    For each XjX_{j}, there exists a sequence of C2C^{2} Hermitian metrics {hj,s}s=1+\{h_{j,s}\}_{s=1}^{+\infty} on XjX_{j} such that

    (2.2) lims+hj,s=hpoint-wisely onXjΣ\displaystyle\lim_{s\rightarrow+\infty}h_{j,s}=h\quad\textrm{point-wisely on}\quad X_{j}\setminus\Sigma

    and for each xXjx\in X_{j} and eExe\in E_{x} we have

    (2.3) |e|hj,s|e|hj,s+1ass;\displaystyle|e|_{h_{j,s}}\nearrow|e|_{h_{j,s+1}}\quad\textrm{as}\quad s\nearrow\mathbb{N};
  4. (4)

    For each XjX_{j}, there exists a sequence of continuous functions λj,s\lambda_{j,s} on XjX_{j} and a continuous function λj\lambda_{j} on XjX_{j} subject to the following requirements:

    • Θhj,s(E)Nakθλj,sωIdE\Theta_{h_{j,s}}(E)\geq_{\rm Nak}\theta-\lambda_{j,s}\omega\otimes{\rm Id}_{E} on XjX_{j};

    • λj,s0\lambda_{j,s}\rightarrow 0 almost everywhere on XjX_{j};

    • 0λj,sλj0\leq\lambda_{j,s}\leq\lambda_{j} on XjX_{j} for any ss\in\mathbb{N}.

Especially, when θ=0\theta=0, the singular Hermitian metric hh is called singular Nakano semi-positive, denoted by Θh(E)Naks0\Theta_{h}(E)\geq_{\rm Nak}^{s}0.

2.2. Ohsawa-Takegoshi extension theorem and the minimal extension property

Let XX be a complex manifold of dimension nn and let XoXX^{o}\subset X be a Zariski open subset. Let (E,h)(E,h) be a holomorphic vector bundle on XoX^{o} with a singular Hermitian metric such that Θh(E)Naks0\Theta_{h}(E)\geq_{\rm Nak}^{s}0 (Definition 2.5). Let α=αω\alpha=\alpha^{\prime}\otimes\omega be an EE-valued (n,0)(n,0)-form, where α\alpha^{\prime} is a section of EE and ω\omega is an (n,0)(n,0)-form. We use the notation

{α,α}h:=cn|α|h2ωω¯,cn=2n(1)n22.\{\alpha,\alpha\}_{h}:=c_{n}|\alpha^{\prime}|^{2}_{h}\omega\wedge\overline{\omega},\quad c_{n}=2^{-n}(-1)^{\frac{n^{2}}{2}}.

The L2L^{2}-norm of αH0(Xo,ωXoE)\alpha\in H^{0}(X^{o},\omega_{X^{o}}\otimes E) is defined as

αh2=Xo{α,α}[0,+].\|\alpha\|_{h}^{2}=\int_{X^{o}}\{\alpha,\alpha\}\in[0,+\infty].

We follow [HPS2018] to use the scaling cnc_{n} in the L2L^{2}-norm.

Suppose f:XBf:X\rightarrow B is a projective map to the open unit ball BrB\subset\mathbb{C}^{r}, where 0B0\in B is a regular value of ff. Then the central fiber X0=f1(0)X_{0}=f^{-1}(0) is a projective manifold of dimension nrn-r. We assume that X0XoX_{0}\cap X^{o}\neq\emptyset and denote by (E0,h0)(E_{0},h_{0}) the restriction of (E,h)(E,h) to X0XoX_{0}\cap X^{o}. The L2L^{2} extension theorem for vector bundles equipped with a singular Hermitian metric, originally developed by Ohsawa-Takegoshi [OT1987], has been further elaborated by Guan-Zhou [GZ2015] and Guan-Mi-Yuan [GMY2023]. This theorem plays a crucial role in the proof of the main theorem of this article. For more results on this direction, we refer the readers to [Blocki2013, CPB2024, BP2008, Berndtsson1996, Demailly2000, DHP2013, OT1988] and the references therein.

Theorem 2.6.

[GMY2023] Notations as above. Suppose that Θh(E)Naks0\Theta_{h}(E)\geq_{\rm Nak}^{s}0 and h0+h_{0}\not\equiv+\infty. Then for every αH0(X0Xo,ωX0XoE|X0Xo)\alpha\in H^{0}(X_{0}\cap X^{o},\omega_{X_{0}\cap X^{o}}\otimes E|_{X_{0}\cap X^{o}}) with αh02<\|\alpha\|_{h_{0}}^{2}<\infty, there exists βH0(Xo,ωXoE)\beta\in H^{0}(X^{o},\omega_{X^{o}}\otimes E) with

(2.4) β|X0=αdfandβh2μ(B)αh02.\displaystyle\beta|_{X_{0}}=\alpha\wedge df\quad\textrm{and}\quad\|\beta\|_{h}^{2}\leq\mu(B)\cdot\|\alpha\|_{h_{0}}^{2}.

The minimal extension property for singular Hermitian metrics, which is closely related to the Ohsawa-Takegoshi extension theorem, was introduced by Hacon, Popa, and Schnell [HPS2018]. This property allows for the extension of sections across a bad locus while maintaining control over the norm of the section.

Let us still denote by BnB\subset\mathbb{C}^{n} the open unit ball.

Definition 2.7 (minimal extension property).

A singular Hermitian metric hh on a torsion-free coherent sheaf \mathscr{F} is said to have the minimal extension property if there exists a nowhere dense closed analytic subset ZZ with the following two properties:

  • \mathscr{F} is locally free on XZX\setminus Z.

  • For every embedding : ι:BX\iota:B\rightarrow X with x=ι(0)XZx=\iota(0)\in X\setminus Z, and every vExv\in E_{x} with |vh,x|=1|v_{h,x}|=1, there is a holomorphic section sH0(B,)s\in H^{0}(B,\mathscr{F}) such that s(0)=vs(0)=v and

    1μ(B)B|s|h2𝑑μ1,\frac{1}{\mu(B)}\int_{B}|s|_{h}^{2}d\mu\leq 1,

    where (E,h)(E,h) denotes the restriction to the open subset X()X(\mathscr{F}).

According to [DNWZ2023], the minimal extension property of a C2C^{2} metric is equivalent to the Nakano semi-positivity of its curvature form.

3. SX(E,h)S_{X}(E,h) and its basic properties

Now, let’s turn our attention to the main object of this paper, SX(E,h)S_{X}(E,h). This concept builds upon the same idea introduced in [SZ2022]. The main difference here is that we allow for the metric hh to be singular.

Let XX be a complex space of dimension nn and XoXregX^{o}\subset X_{\rm reg} a dense Zariski open subset of the regular locus XregX_{\rm reg}. Let (E,h)(E,h) be a vector bundle on XoX^{o} with a singular Hermitian metric.

Definition 3.1.

SX(E,h)S_{X}(E,h) is a sheaf defined as follows: for an open subset UXU\subset X, the space SX(E,h)(U)S_{X}(E,h)(U) consists of holomorphic EE-valued (n,0)(n,0)-forms α\alpha on UXoU\cap X^{o} such that {α,α}\{\alpha,\alpha\} is locally integrable near every point of UU.

Let f:XYf:X\to Y be a holomorphic morphism to a complex manifold YY. We define SX/Y(E,h)S_{X/Y}(E,h) as SX(E,h)fωY1S_{X}(E,h)\otimes f^{\ast}\omega_{Y}^{-1}.

If X=XoX=X^{o} (in particular, XX is smooth) and EE is a holomorphic line bundle, then SX(E,h)ωXE(h)S_{X}(E,h)\simeq\omega_{X}\otimes E\otimes\mathscr{I}(h). The sheaf SX(E,h)S_{X}(E,h) is a torsion-free 𝒪X\mathscr{O}_{X}-module with properties described in [SZ2022]. The proofs of these properties are analogous and will be omitted here.

Lemma 3.2.

If UXoU\subset X^{o} be a dense Zariski open subset, then SX(E,h)=SX(E|U,h|U)S_{X}(E,h)=S_{X}(E|_{U},h|_{U}).

Proposition 3.3 (Functoriality Property).

Let π:XX\pi:X^{\prime}\to X be a proper holomorphic map between complex spaces which is biholomorphic over XoX^{o}. Then

πSX(πE,πh)=SX(E,h).\pi_{\ast}S_{X^{\prime}}(\pi^{\ast}E,\pi^{\ast}h)=S_{X}(E,h).
Lemma 3.4.

Let (F,hF)(F,h_{F}) be a Hermitian vector bundle on XX where hFh_{F} is a smooth metric. Then

SX(EF|Xo,hhF)SX(E,h)F.S_{X}(E\otimes F|_{X^{o}},hh_{F})\simeq S_{X}(E,h)\otimes F.

We generalize the tameness condition introduced in [SZ2022] to include singular Hermitian metrics. The concept of ”tameness” is inspired by the theory of degeneration of Hodge structures [Schmid1973, Cattani_Kaplan_Schmid1986] and the theory of tame harmonic bundles [Simpson1988, Simpson1990, Mochizuki20072, Mochizuki20071].

Definition 3.5.

Let XX be a complex space and XoXregX^{o}\subset X_{\rm reg} a dense Zariski open subset. A vector bundle (E,h)(E,h) on XoX^{o} with a singular Hermitian metric is called tame on XX if, for every point xXx\in X, there is an open neighborhood UU of xx, a proper bimeromorphic morphism π:U~U\pi:\widetilde{U}\to U which is biholomorphic over UXoU\cap X^{o}, and a vector bundle QQ endowed with a smooth metric hQh_{Q} on U~\widetilde{U} such that the following conditions hold.

  1. (1)

    πE|π1(XoU)Q|π1(XoU)\pi^{\ast}E|_{\pi^{-1}(X^{o}\cap U)}\subset Q|_{\pi^{-1}(X^{o}\cap U)} as a subsheaf.

  2. (2)

    There is a singular Hermitian metric hQh^{\prime}_{Q} on Q|π1(XoU)Q|_{\pi^{-1}(X^{o}\cap U)} so that hQ|πEπhh^{\prime}_{Q}|_{\pi^{\ast}E}\sim\pi^{\ast}h on π1(XoU)\pi^{-1}(X^{o}\cap U) and

    (3.1) (i=1rπfi2)chQhQ\displaystyle(\sum_{i=1}^{r}\|\pi^{\ast}f_{i}\|^{2})^{c}h_{Q}\lesssim h^{\prime}_{Q}

    for some cc\in\mathbb{R}. Here {f1,,fr}\{f_{1},\dots,f_{r}\} is an arbitrary set of local generators of the ideal sheaf defining U~\π1(Xo)U~\widetilde{U}\backslash\pi^{-1}(X^{o})\subset\widetilde{U}.

Remark 3.6.

The following are typical examples of tame Hermitian metrics.

  • A continuous Hermitian metric.

  • Any singular Hermitian metric of type eφhe^{-\varphi}h on a vector bundle is tame. Here hh is a smooth metric and φ\varphi is a quasi-psh function.

  • The Hodge metric of a variation of Hodge structure is tame at its boundary points. This is a consequence of the norm estimate for the Hodge metric, which was established by Schmid [Schmid1973] and Cattani-Kaplan-Schmid [Cattani_Kaplan_Schmid1986].

  • A tame harmonic metric on a harmonic bundle remains tame at boundary points. This conclusion is drawn from the norm estimate of the tame harmonic metric, as established by Simpson [Simpson1990] and Mochizuki [Mochizuki20072, Mochizuki20071].

Proposition 3.7.

Assume that the holomorphic vector bundle (E,h)(E,h) with a singular Hermitian metric satisfies the following conditions.

  1. (1)

    For every point xXx\in X there is a neighborhood UU of xx, a bounded CC^{\infty} function φ\varphi on UXoU\cap X^{o} such that 1Θeφh(E)Naks0\sqrt{-1}\Theta_{e^{-\varphi}h}(E)\geq_{\rm Nak}^{s}0 holds on UXoU\cap X^{o}.

  2. (2)

    The holomorphic vector bundle (E,h)(E,h) is tame on XX.

Then SX(E,h)S_{X}(E,h) is a coherent sheaf.

Notice that Condition (1) implies that (E,h)(E,h) is tame on XoX^{o}.

Proof.

Since the problem is local, we assume that XX is a germ of complex space. By replacing hh by eφhe^{-\varphi}h for some smooth bounded function φ\varphi (this does not alter SX(E,h)S_{X}(E,h)) we may assume that (E,h)(E,h) is Nakano semi-positive. Let π:X~X\pi:\widetilde{X}\to X be a desingularization so that π\pi is biholomorphic over XoX^{o} and D:=π1(X\Xo)D:=\pi^{-1}(X\backslash X^{o}) is a simple normal crossing divisor. For the sake of convenience, we will consider XoX~X^{o}\subset\widetilde{X} as a subset. Since (E,h)(E,h) is tame, we assume the existence of a Hermitian vector bundle (Q,hQ)(Q,h_{Q}) on X~\widetilde{X} such that EE is a subsheaf of Q|XoQ|_{X^{o}} and there exists an integer mm\in\mathbb{N} satisfying

(3.2) |z1zr|2mhQhQ\displaystyle|z_{1}\cdots z_{r}|^{2m}h_{Q}\lesssim h_{Q}^{\prime}

where z1,,znz_{1},\cdots,z_{n} are local coordinates on X~\widetilde{X} with respect to which D={z1zr=0}D=\{z_{1}\cdots z_{r}=0\}, and where hQh_{Q}^{\prime} denotes a singular Hermitian metric on Q|XoQ|_{X^{o}} such that hQ|Ehh_{Q}^{\prime}|_{E}\sim h. It follows from Proposition 3.3 that there is an isomorphism

SX(E,h)π(SX~(E,h)).\displaystyle S_{X}(E,h)\simeq\pi_{\ast}\left(S_{\widetilde{X}}(E,h)\right).

Since π\pi is a proper map, it suffices to show that SX~(E,h)S_{\widetilde{X}}(E,h) is a coherent sheaf on X~\widetilde{X}. Since the problem is local and X~\widetilde{X} is smooth, we may assume that X~n\widetilde{X}\subset\mathbb{C}^{n} is the unit ball, such that D={z1zr=0}D=\{z_{1}\cdots z_{r}=0\}. Without loss of generality we assume that QQ admits a global holomorphic frame {e1,,el}\{e_{1},\dots,e_{l}\} and h0h_{0} is the trivial metric associated with this frame, i.e.,

(3.3) ei,ejh0={1,i=j0,ij.\displaystyle\langle e_{i},e_{j}\rangle_{h_{0}}=\begin{cases}1,&i=j\\ 0,&i\neq j\end{cases}.

Since QQ is coherent, the space Γ(X~,SX~(E,h))\Gamma(\widetilde{X},S_{\widetilde{X}}(E,h)) generates a coherent subsheaf 𝒥\mathscr{J} of QQ by strong Noetherian property for coherent sheaves. We have the inclusion 𝒥SX~(E,h)\mathscr{J}\subset S_{\widetilde{X}}(E,h) by the construction. It remains to prove the converse. By Krull’s theorem ([Atiyah1969, Corollary 10.19]), it suffices to show that

(3.4) 𝒥x+SX~(E,h)xmX~,xk+1Q=SX~(E,h)x,k0,xX~.\displaystyle\mathscr{J}_{x}+S_{\widetilde{X}}(E,h)_{x}\cap m_{\widetilde{X},x}^{k+1}Q=S_{\widetilde{X}}(E,h)_{x},\quad\forall k\geq 0,\quad\forall x\in\widetilde{X}.

Let αSX~(E,h)x\alpha\in S_{\widetilde{X}}(E,h)_{x} be defined in a precompact neighborhood VV of xx. Choose a CC^{\infty} cut-off function λ\lambda such that λ1\lambda\equiv 1 near xx and suppλV{\rm supp}\lambda\subset V. Let

ψk(z):=2(n+k+rm)log|zx|+|z|2\displaystyle\psi_{k}(z):=2(n+k+rm)\log|z-x|+|z|^{2}

and hψk:=eψkhh_{\psi_{k}}:=e^{-\psi_{k}}h, where |z|2:=i=1n|zi|2|z|^{2}:=\sum_{i=1}^{n}|z_{i}|^{2}. Let ω0:=1¯|z|2\omega_{0}:=\sqrt{-1}\partial\bar{\partial}|z|^{2}. Then

1Θhψk(E)=1¯ψk+1Θh(E)Naksω0.\displaystyle\sqrt{-1}\Theta_{h_{\psi_{k}}}(E)=\sqrt{-1}\partial\bar{\partial}\psi_{k}+\sqrt{-1}\Theta_{h}(E)\geq^{s}_{\rm Nak}\omega_{0}.

Since supp(λα)V{\rm supp}(\lambda\alpha)\subset V and ¯(λα)=0\bar{\partial}(\lambda\alpha)=0 near xx, we know that

¯(λα)ω0,hψk2¯(λα)ω0,h2¯λL2αω0,h2+|λ|2¯αω0,h2<\displaystyle\|\bar{\partial}(\lambda\alpha)\|^{2}_{\omega_{0},h_{\psi_{k}}}\sim\|\bar{\partial}(\lambda\alpha)\|^{2}_{\omega_{0},h}\leq\|\bar{\partial}\lambda\|^{2}_{L^{\infty}}\|\alpha\|^{2}_{\omega_{0},h}+|\lambda|^{2}\|\bar{\partial}\alpha\|^{2}_{\omega_{0},h}<\infty

Since there is a complete Kähler metric on XoX^{o} by [SZ2022, Lemma 2.14], [CataldoAndrea1998, Proposition 4.1.1] (see also [Demailly1982, Théorème 5.1]) gives a solution to the equation ¯β=¯(λα)\bar{\partial}\beta=\bar{\partial}(\lambda\alpha) so that

(3.5) βω0,h2Xo|β|ω0,h2|zx|2(n+k+rm)volω0¯(λα)ω0,hψk2<.\displaystyle\|\beta\|^{2}_{\omega_{0},h}\lesssim\int_{X^{o}}|\beta|^{2}_{\omega_{0},h}|z-x|^{-2(n+k+rm)}{\rm vol}_{\omega_{0}}\lesssim\|\bar{\partial}(\lambda\alpha)\|^{2}_{\omega_{0},h_{\psi_{k}}}<\infty.

Thus γ=βλα\gamma=\beta-\lambda\alpha is holomorphic and γΓ(X~,SX~(E,h))\gamma\in\Gamma(\widetilde{X},S_{\widetilde{X}}(E,h)).

Notice that ¯β=0\bar{\partial}\beta=0 near xx. We may shrink X~\widetilde{X} and assume that

β=i=1lfieidz1dzn\beta=\sum_{i=1}^{l}f_{i}e_{i}dz_{1}\wedge\cdots\wedge dz_{n}

for some holomorphic functions f1,,fl𝒪X~(Xo)f_{1},\dots,f_{l}\in\mathscr{O}_{\widetilde{X}}(X^{o}). Noticing that hQh0h_{Q}\sim h_{0}, we can deduce from (3.2), (3.3) and (3.5) that

i=1lXo|fi|2|z1zr|2m|zx|2(n+k+rm)volω0\displaystyle\sum_{i=1}^{l}\int_{X^{o}}|f_{i}|^{2}|z_{1}\cdots z_{r}|^{2m}|z-x|^{-2(n+k+rm)}{\rm vol}_{\omega_{0}}
=\displaystyle= Xo|β|ω0,h02|z1zr|2m|zx|2(n+k+rm)volω0\displaystyle\int_{X^{o}}|\beta|^{2}_{\omega_{0},h_{0}}|z_{1}\cdots z_{r}|^{2m}|z-x|^{-2(n+k+rm)}{\rm vol}_{\omega_{0}}
\displaystyle\lesssim Xo|β|ω0,h2|zx|2(n+k+rm)volω0<.\displaystyle\int_{X^{o}}|\beta|^{2}_{\omega_{0},h}|z-x|^{-2(n+k+rm)}{\rm vol}_{\omega_{0}}<\infty.

This implies that for every i=1,,li=1,\dots,l, we have z1mzrmfimX~,xk+1+rmz_{1}^{m}\cdots z_{r}^{m}f_{i}\in m_{\widetilde{X},x}^{k+1+rm} ([Demailly2012, Lemma 5.6]). Consequently, βx\beta_{x} belongs to mX~,xk+1Qm_{\widetilde{X},x}^{k+1}Q, and we establish the validity of (3.4). ∎

3.1. Example: parabolic Higgs bundle

We use the notations in §1.4. Let XX be a smooth projective variety and DD a reduced simple normal crossing divisor on XX. Consider (H,{HE}EDivD(X),θ)(H,\{{{}_{E}}H\}_{E\in{\rm Div}_{D}(X)},\theta) as a locally abelian parabolic Higgs bundle on (X,D)(X,D), which is polystable with respect to an ample line bundle AA on XX. Let hh be a tame harmonic metric on H|X\DH|_{X\backslash D}, which is compatible with the parabolic structure. The existence of such a metric is ensured by Simpson [Simpson1990] for algebraic curves and by Mochizuki [Mochizuki2006] in higher dimensions. Let θ¯\overline{\theta} be the adjoint of θ\theta, and \partial be the unique (1,0)(1,0)-connection such that +¯\partial+\bar{\partial} is compatible with hh. Consequently, (H|X\D𝒪X\D𝒜X\D0,:=+¯+θ+θ¯)(H|_{X\backslash D}\otimes_{\mathscr{O}_{X\backslash D}}\mathscr{A}^{0}_{X\backslash D},\nabla:=\partial+\bar{\partial}+\theta+\overline{\theta}) specifies a meromorphic flat connection that remains regular along the divisor DD. Let =1,0+0,1\nabla=\nabla^{1,0}+\nabla^{0,1} be the decomposition with respect to the bi-degree. Notice that 1,0=+θ\nabla^{1,0}=\partial+\theta and 0,1=¯+θ¯\nabla^{0,1}=\bar{\partial}+\overline{\theta}.

Let KH|X\DK\subset H|_{X\backslash D} be a locally free subsheaf such that the following conditions hold:

  • 0,1(K)=0\nabla^{0,1}(K)=0, i.e. KK is holomorphic with respect to both the complex structures ¯\bar{\partial} and 0,1\nabla^{0,1}.

  • (θ)(K)K𝒜X\D1,0(\nabla-\theta)(K)\subset K\otimes\mathscr{A}^{1,0}_{X\backslash D}.

Let (F,hF)(F,h_{F}) be an arbitrary Nakano semi-positive Hermitian vector bundle on XX. Let LL be a line bundle on XX such that LB+NL\simeq_{\mathbb{R}}B+N, where BB is a semi-positive \mathbb{R}-divisor and NN is an \mathbb{R}-divisor on XX which is supported on DD. Let φN\varphi_{N} be a weight function associated with NN. By [SZtwisted, Lemma 3.8], there is a singular Hermitian metric hLh_{L} on LL such that the following conditions hold:

  1. (1)

    hLh_{L} is smooth over X\supp(N)X\backslash{\rm supp}(N).

  2. (2)
    (3.6) 1ΘhL(L|X\supp(N))=1ΘhB(B)|X\supp(N)0;\displaystyle\sqrt{-1}\Theta_{h_{L}}(L|_{X\backslash{\rm supp}(N)})=\sqrt{-1}\Theta_{h_{B}}(B)|_{X\backslash{\rm supp}(N)}\geq 0;
  3. (3)
    (3.7) |e|hLexp(φN)\displaystyle|e|_{h_{L}}\sim\exp(-\varphi_{N})

    for a local generator ee of LL.

It can be shown that hhFhLhh_{F}h_{L} has Nakano semi-positive curvature on X\DX\backslash D^{\prime} and is tame on XX [SZtwisted, Lemma 3.10]. Moreover, one has the following.

Theorem 3.8.

[SZtwisted, Corollary 3.13] ωX(P(H)DN,(2)jK)FLSX(KF|X\DL|X\D,hhFhL)\omega_{X}\otimes(P{{}_{D-N,(2)}}(H)\cap j_{\ast}K)\otimes F\otimes L\simeq S_{X}(K\otimes F|_{X\backslash D^{\prime}}\otimes L|_{X\backslash D^{\prime}},hh_{F}h_{L}).

Thus Theorem 1.1 implies Theorem 1.6.

3.2. Example: multiplier SS-sheaf

We use the notation in §1.3. Note that Griffiths’s curvature formula ensures that (S(𝕍),hQ)(S(\mathbb{V}),h_{Q}) is Nakano semi-positive ([SC2021, Theorem 2.3], see also [Schmid1973, Lemma 7.18]). Thus (S(𝕍)F,eφhQh)(S(\mathbb{V})\otimes F,e^{-\varphi}h_{Q}h) is Nakano semi-positive. The tameness of (S(𝕍),hQ)(S(\mathbb{V}),h_{Q}) follows from the theory of degeneration of Hodge structure (see [SZ2022, Proposition 5.4]). Therefore (S(𝕍)F,eφhQh)(S(\mathbb{V})\otimes F,e^{-\varphi}h_{Q}h) is tame on XX. By Lemma 3.4 one has the following.

Theorem 3.9.

S(ICX(𝕍),φ)FSX(S(𝕍)F,eφhQh)S(IC_{X}(\mathbb{V}),\varphi)\otimes F\simeq S_{X}(S(\mathbb{V})\otimes F,e^{-\varphi}h_{Q}h).

Thus Theorem 1.1 implies Theorem 1.5.

4. proof of the main theorem

The proof of the main theorem is strongly influenced by the one in [HPS2018].

4.1. Construction of the metric on its locally free part

To define the singular Hermitian metric HH on =f(SX/Y(E,h))\mathscr{F}=f_{\ast}(S_{X/Y}(E,h)), we first construct the metric on some Zariski open subset YZY\setminus Z. Then, we extend it over ZZ using the L2L^{2} extension theorem 2.6. The constructions and proofs follow a similar approach as [HPS2018]. It is worth noting that the tameness condition allows the arguments in [HPS2018] to apply to those degenerate (E,h)(E,h) as well. Thanks to Proposition 3.3, we can choose a resolution of singularity for XX and assume that XX is a complex manifold throughout the proof. Notice that SX/Y(E,h)S_{X/Y}(E,h) is a torsion-free coherent sheaf (Proposition 3.7). We begin by selecting a closed, nowhere dense analytic subset ZYZ\subset Y that satisfies the following conditions:

  1. (1)

    The morphism ff is submersive over YZY\setminus Z.

  2. (2)

    XyXoX_{y}\cap X^{o}\neq\emptyset for every yYZy\in Y\setminus Z.

  3. (3)

    The sheaf \mathscr{F} is locally free on YZY\setminus Z.

  4. (4)

    \mathscr{F} has the base change property on YZY\setminus Z, that is, the natural morphism yH0(Xy,SX/Y(E,h)|Xy)\mathscr{F}_{y}\to H^{0}(X_{y},S_{X/Y}(E,h)|_{X_{y}}) is an isomorphism for every yY\Zy\in Y\backslash Z.

Consequently, when restricted to the open subset YZY\setminus Z, the sheaf \mathscr{F} forms a holomorphic vector bundle FF with a rank of r1r\geq 1. Conditions (3) and (4) ensure that whenever yYZy\in Y\setminus Z and Xy:=f1(y)X_{y}:=f^{-1}(y), we have Fy=|y=f(SX/Y(E,h))|y=H0(Xy,SX/Y(E,h)|Xy)F_{y}=\mathscr{F}|_{y}=f_{\ast}(S_{X/Y}(E,h))|_{y}=H^{0}(X_{y},S_{X/Y}(E,h)|_{X_{y}}).

Let (Ey,hy)(E_{y},h_{y}) denote the restriction of (E,h)(E,h) to XyXoX_{y}\cap X^{o}. Then Theorem 2.6 implies the following lemma.

Lemma 4.1.

For any yYZy\in Y\setminus Z, we have the inclusion

H0(Xy,SXy(Ey,hy))Fy=H0(Xy,SX/Y(E,h)|Xy).H^{0}(X_{y},S_{X_{y}}(E_{y},h_{y}))\subset F_{y}=H^{0}(X_{y},S_{X/Y}(E,h)|_{X_{y}}).
Proof.

If hy+h_{y}\equiv+\infty, then H0(Xy,SXy(Ey,hy))H^{0}(X_{y},S_{X_{y}}(E_{y},h_{y})) is trivial. Therefore, we only need to consider the case when hyh_{y} is not identically equal to ++\infty. A small neighborhood UU of yy can be chosen such that it is biholomorphic to the open unit ball BrB\in\mathbb{C}^{r}, and ωY\omega_{Y} is trivial on it. Given αH0(Xy,SXy(Ey,hy))\alpha\in H^{0}(X_{y},S_{X_{y}}(E_{y},h_{y})), Theorem 2.6 provides a section βH0(U,SX(E,h))\beta\in H^{0}(U,S_{X}(E,h)) such that β|Xy=αdf\beta|_{X_{y}}=\alpha\wedge df. ∎

For each yYZy\in Y\setminus Z, a singular Hermitian metric on FyF_{y} can be defined as

|α|H,y2=Xy{α,α}hy[0,+],|\alpha|_{H,y}^{2}=\int_{X_{y}}\{\alpha,\alpha\}_{h_{y}}\in[0,+\infty],

and it is finite on H0(Xy,SXy(Ey,hy))H^{0}(X_{y},S_{X_{y}}(E_{y},h_{y})). This definition is valid because XyXoX_{y}\setminus X^{o} has a measure of zero, so it does not cause any issues for the integral.

In order to patch the singular Hermitian metric ||H,y|-|_{H,y} together on YZY\setminus Z, we select a point yYZy\in Y\setminus Z and an open neighborhood UYZU\subset Y\setminus Z that is biholomorphic to the open unit ball BrB\subset\mathbb{C}^{r}. By pulling everything back to UU, we can assume that Y=BY=B, Z=Z=\emptyset, and y=0y=0. Let’s denote by t1,,trt_{1},\dots,t_{r} the standard coordinates on BB. Then the canonical bundle ωB\omega_{B} is trivialized by the global section dt1dtrdt_{1}\wedge\cdots\wedge dt_{r}, and the volume form on BB is given by

dμ=cr(dt1dtr)(dt¯1dt¯r).d\mu=c_{r}(dt_{1}\wedge\cdots dt_{r})(d\bar{t}_{1}\wedge\dots d{\bar{t}}_{r}).

We fix a holomorphic section sH0(B,F)s\in H^{0}(B,F), and denote by β=s(dt1dtr)H0(B,ωBF)H0(X,SX(E,h))\beta=s\wedge(dt_{1}\wedge\cdots\wedge dt_{r})\in H^{0}(B,\omega_{B}\otimes F)\simeq H^{0}(X,S_{X}(E,h)) the corresponding holomorphic nn-form on XX with coefficients in EE. Given that f:XBf:X\rightarrow B is smooth, Ehresmann’s fibration theorem implies that XX is diffeomorphic to the product B×X0B\times X_{0}. After selecting a Kähler metric ω0\omega_{0} on X0X_{0}, we can express

(4.1) |β|h2=Gdμω0nr(nr)!.\displaystyle|\beta|_{h}^{2}=G\cdot d\mu\wedge\frac{\omega_{0}^{n-r}}{(n-r)!}.

Since hh is an increasing limit of C2C^{2} metrics near every point, the function G:B×X0[0,+)G:B\times X_{0}\rightarrow[0,+\infty) is both lower semi-continuous and locally integrable.

At every point yBy\in B, we then have, by construction,

(4.2) |s(y)|H,y2=X0G(y,)ω0nr(nr)!.\displaystyle|s(y)|_{H,y}^{2}=\int_{X_{0}}G(y,-)\frac{\omega_{0}^{n-r}}{(n-r)!}.

According to Fubini’s theorem, the function y|s(y)|H,yy\mapsto|s(y)|_{H,y} is measurable. Furthermore, since X0X_{0} is compact and GG is locally integrable, |s(y)|H,y<|s(y)|_{H,y}<\infty for almost every yBy\in B. As FF is coherent, it is generated over BB by finite many global sections. Therefore, the singular Hermitian inner product ||H,y|-|_{H,y} is finite and positive definite for almost every yBy\in B, and thus for almost every yYZy\in Y\setminus Z. The above discussion ensures that the Hermitian inner products satisfy the conditions in Definition 2.2, making them a singular Hermitian metric on FF on YZY\setminus Z.

Proposition 4.2.

On YZY\setminus Z, the singular Hermitian inner products ||H,y|-|_{H,y} determine a singular Hermitian metric on the holomorphic vector bundle FF.

4.2. Extend the metric to the whole YY

The aim of this subsection is to extend the singular Hermitian metric from YZY\setminus Z to the entire YY. To achieve this, we will examine the measurable function ψ:=log|g|H:YZ[,+]\psi:=\log|g|_{H^{\ast}}:Y\setminus Z\rightarrow[-\infty,+\infty], where HH^{\ast} is the induced singular Hermitian metric on FF^{\ast} and gH0(Y,)g\in H^{0}(Y,\mathscr{F}^{\ast}). Our goal is to demonstrate that the function ψ\psi is a plurisubharmonic function on YZY\setminus Z and is locally bounded near every point of ZZ, which will allow us to extend it as a plurisubharmonic function across the entire YY using the Riemann extension theorem for holomorphic functions [Demailly2012, Theorem I.5.24].

First, let us restate the Ohsawa-Takegoshi theorem as given in Theorem 2.6 in a form that is more suitable for our subsequent purposes.

Lemma 4.3.

For every embedding ι:BY\iota:B\rightarrow Y from the unit ball BdimYB\subset\mathbb{C}^{\dim Y} with y=ι(0)YZy=\iota(0)\in Y\setminus Z, and for every αFy\alpha\in F_{y} with |α|H,y=1|\alpha|_{H,y}=1, there is a holomorphic section sH0(B,ι)s\in H^{0}(B,\iota^{\ast}\mathscr{F}) with s(0)=αs(0)=\alpha and

1μ(B)B|s|H2𝑑μ1.\frac{1}{\mu(B)}\int_{B}|s|_{H}^{2}d\mu\leq 1.
Proof.

After pulling everything back to BB, we can assume that Y=BY=B and y=0y=0. By Theorem 2.6, since |α|H,0=1|\alpha|_{H,0}=1, there exists an element βH0(X,SX(E,h))\beta\in H^{0}(X,S_{X}(E,h)) such that β|X0=αdf\beta|_{X_{0}}=\alpha\wedge df and βh2μ(B)\|\beta\|_{h}^{2}\leq\mu(B). We can trivialize the canonical bundle ωB\omega_{B} using dt1dtrdt_{1}\wedge\cdots\wedge dt_{r}, which allows us to consider β\beta as a holomorphic section sH0(B,ι)s\in H^{0}(B,\iota^{\ast}\mathscr{F}) with s(0)=αs(0)=\alpha. Additionally, 1μ(B)B|s|H2𝑑μ1\frac{1}{\mu(B)}\int_{B}|s|_{H}^{2}d\mu\leq 1, as dμ=cr(dt1dtr)(dt¯1dt¯r)d\mu=c_{r}(dt_{1}\wedge\cdots dt_{r})\wedge(d\bar{t}_{1}\wedge\cdots d\bar{t}_{r}). ∎

Proposition 4.4.

Every point in YY has an open neighborhood UYU\subset Y such that ψ=log|g|H\psi=\log|g|_{H^{\ast}} is bounded from above by a constant on UZU\setminus Z.

Proof.

Given an arbitrary point xYx\in Y, we select two small open neighborhoods UVYU\subset V\subset Y of xx, where V¯\overline{V} is compact, U¯V\overline{U}\subset V, and for every point yUy\in U, there is an embedding ι:BY\iota:B\rightarrow Y of the unit ball BrB\in\mathbb{C}^{r} with ι(0)=y\iota(0)=y and ι(B)V\iota(B)\subset V. Now, we aim to prove the existence of a constant CC such that ψC\psi\leq C on UZU\setminus Z.

Let yUZy\in U\setminus Z be a fixed point. If ψ(y)=\psi(y)=-\infty, then there is nothing to prove. However, assuming ψ(y)\psi(y)\neq-\infty, we can use the definition of the metric on the dual bundle to find a vector α\alpha in FyF_{y} that satisfies |α|H,y=1|\alpha|_{H,y}=1 and ψ(y)=log|g(α)|\psi(y)=\log|g(\alpha)|. We choose an embedding ι:BY\iota:B\rightarrow Y such that ι(0)=y\iota(0)=y and ι(B)V\iota(B)\subset V. According to Lemma 4.3, there exists a holomorphic section sH0(V,)s\in H^{0}(V,\mathscr{F}) with s(0)=αs(0)=\alpha and

1μ(B)V|s|H2𝑑μ1.\frac{1}{\mu(B)}\int_{V}|s|_{H}^{2}d\mu\leq 1.

Therefore, we have ψ(y)=log|g(s)|y\psi(y)=\log|g(s)|_{y}, and the desired upper bound can be obtained from Proposition 4.5 below. ∎

Proposition 4.5.

Fix a constant K0K\geq 0, and consider the set

SK={sH0(V,)V|s|H2𝑑μK}.S_{K}=\left\{s\in H^{0}(V,\mathscr{F})\mid\int_{V}|s|_{H}^{2}d\mu\leq K\right\}.

Then

  1. (1)

    every sequence {sk}SK\{s_{k}\}\in S_{K} has a subsequence that converges uniformly on compact subsets;

  2. (2)

    there is a constant C0C\geq 0 such that, for every section sSKs\in S_{K}, the holomorphic function g(s)g(s) is uniformly bounded by CC on the compact set U¯\overline{U}.

Proof.

For each section sSKs\in S_{K}, we define

β=s(dt1dtr)H0(V,ωY)=H0(f1(V),SX(E,h)),\beta=s\otimes(dt_{1}\wedge\cdots\wedge dt_{r})\in H^{0}(V,\omega_{Y}\otimes\mathscr{F})=H^{0}(f^{-1}(V),S_{X}(E,h)),

the corresponding holomorphic section of SX(E,h)S_{X}(E,h). It satisfies that βh2=V|s|H2𝑑μK\|\beta\|_{h}^{2}=\int_{V}|s|_{H}^{2}d\mu\leq K. Because V¯\overline{V} is compact and ff is proper, we can cover f1(V)f^{-1}(V) with a finite number of open sets WW that are biholomorphic to the open unit ball in n\mathbb{C}^{n}.

Let π:W~W\pi:\widetilde{W}\to W be a desingularization such that π\pi is biholomorphic over Wo:=WXoW^{o}:=W\cap X^{o}, and D:=π1(W\Wo)D:=\pi^{-1}(W\backslash W^{o}) is a simple normal crossing divisor. For the sake of simplicity, we consider WoW^{o} as a subset of W~\widetilde{W}. Since (E,h)(E,h) is tame, we assume the existence of a CC^{\infty} Hermitian vector bundle (Q,hQ)(Q,h_{Q}) on W~\widetilde{W} such that EE is a subsheaf of Q|WoQ|_{W^{o}}. Additionally, there exists an mm\in\mathbb{N} that satisfies the inequality

(4.3) |z1zr|2mhQh\displaystyle|z_{1}\cdots z_{r}|^{2m}h_{Q}\lesssim h

where z1,,znz_{1},\cdots,z_{n} are local coordinates on W~\widetilde{W} and D={z1zr=0}D=\{z_{1}\cdots z_{r}=0\}.

To continue, we choose a set of holomorphic local frames s1,,srs_{1},\dots,s_{r} of QQ on some open subset WW~W^{\prime}\subset\widetilde{W}. Let h0h_{0} be the trivial metric associated with these frames, i.e., (si,sj)h0=δij(s_{i},s_{j})_{h_{0}}=\delta_{ij}. Using this, we can write β=1irbisidz1dzn\beta=\sum_{1\leq i\leq r}b_{i}s_{i}\otimes dz_{1}\wedge\cdots\wedge dz_{n}, where bib_{i} are holomorphic functions on WWoW^{\prime}\cap W^{o}. Notice that the two Hermitian metrics h0h_{0} and hQh_{Q} are quasi-isometric, i.e., there exists some positive constant C1C_{1} such that 1C1hQh0C1hQ\frac{1}{C_{1}}h_{Q}\leq h_{0}\leq C_{1}h_{Q}. This leads to the following inequality:

(4.4) W1ir|bi|2|z1zr|2m(dx1dy1)(dxndyn)\displaystyle\int_{W^{\prime}}\sum_{1\leq i\leq r}|b_{i}|^{2}|z_{1}\cdots z_{r}|^{2m}(dx_{1}\wedge dy_{1})\wedge\cdots\wedge(dx_{n}\wedge dy_{n})
C1W1ir|bi|2|z1zr|2m|si|hQ2(dx1dy1)(dxndyn)(4.3)C2W|β|h2C2K,\displaystyle\leq C_{1}\int_{W^{\prime}}\sum_{1\leq i\leq r}|b_{i}|^{2}|z_{1}\cdots z_{r}|^{2m}|s_{i}|^{2}_{h_{Q}}(dx_{1}\wedge dy_{1})\wedge\cdots\wedge(dx_{n}\wedge dy_{n})\stackrel{{\scriptstyle(\ref{align_tame_2})}}{{\leq}}C_{2}\int_{W}|\beta|_{h}^{2}\leq C_{2}K,

for some positive constant C2C_{2}. In particular, the functions (z1zr)mbi(z_{1}\cdots z_{r})^{m}b_{i} are holomorphic on WW^{\prime}.

Let {sn}\{s_{n}\} be an arbitrary sequence in SKS_{K}. We denote by βn\beta_{n} the corresponding holomorphic sections in SX(E,h)S_{X}(E,h) on WW^{\prime}. We can write βn\beta_{n} as βn=1irbn,isidz1dzn\beta_{n}=\sum_{1\leq i\leq r}b_{n,i}s_{i}\otimes dz_{1}\wedge\cdots\wedge dz_{n}. According to [HPS2018, Proposition 12.5], by replacing it with a subsequence, we can assume that the sequence {(z1zr)mbn,i}n\{(z_{1}\cdots z_{r})^{m}b_{n,i}\}_{n\in\mathbb{N}} converges uniformly on compact subsets in H0(W,𝒪W)H^{0}(W^{\prime},\mathscr{O}_{W^{\prime}}). Notice that the natural injective morphism 𝒪W𝒪W(mD)\mathscr{O}_{W^{\prime}}\rightarrow\mathscr{O}_{W^{\prime}}(mD) induces an injective continuous mapping

ι:=×(z1zr)m:H0(W,𝒪W)H0(W,𝒪W(mD))\iota:=\times(z_{1}\cdots z_{r})^{-m}:H^{0}(W^{\prime},\mathscr{O}_{W^{\prime}})\rightarrow H^{0}(W^{\prime},\mathscr{O}_{W^{\prime}}(mD))

between Fréchet spaces ([GR2009, Ch. VIII,§A]). Therefore, the sequence {bn,i=ι((z1zr)mbn,i)}\{b_{n,i}=\iota((z_{1}\cdots z_{r})^{m}b_{n,i})\} converges to b,iH0(W,𝒪W(mD))b_{\infty,i}\in H^{0}(W^{\prime},\mathscr{O}_{W^{\prime}}(mD)).

Notice that (4.4) implies that SW(πE,πh)ωWi=1r𝒪W(mD)siS_{W^{\prime}}(\pi^{\ast}E,\pi^{\ast}h)\subset\omega_{W^{\prime}}\otimes\oplus_{i=1}^{r}\mathscr{O}_{W^{\prime}}(mD)s_{i}, we conclude that H0(W,SW(πE,πh))H^{0}(W^{\prime},S_{W^{\prime}}(\pi^{\ast}E,\pi^{\ast}h)) is a closed subspace of H0(W,ωWi=1r𝒪W(mD)si)H^{0}(W^{\prime},\omega_{W^{\prime}}\otimes\oplus_{i=1}^{r}\mathscr{O}_{W^{\prime}}(mD)s_{i}) ([GR2009, Proposition VIII. A.2]). Therefore, the argument on the previous paragraph shows that βn\beta_{n} converges to β=1irb,isidz1dznH0(W,SW(πE,πh))\beta_{\infty}=\sum_{1\leq i\leq r}b_{\infty,i}s_{i}\otimes dz_{1}\wedge\cdots\wedge dz_{n}\in H^{0}(W^{\prime},S_{W^{\prime}}(\pi^{\ast}E,\pi^{\ast}h)). Since we are dealing with finitely many open sets, the corresponding sequence {βn}\{\beta_{n}\} of every sequence {sn}\{s_{n}\} in SKS_{K} has a subsequence that converges uniformly on compact subsets to some βH0(f1(V),SX(E,h))\beta_{\infty}\in H^{0}(f^{-1}(V),S_{X}(E,h)). Let sH0(V,)s_{\infty}\in H^{0}(V,\mathscr{F}) be the unique section such that

β=s(dt1dtr).\beta_{\infty}=s_{\infty}\otimes(dt_{1}\wedge\cdots\wedge dt_{r}).

Based on [GR2009, Proposition VIII. A.2], the sequence sks_{k} converges to ss_{\infty} in the Fréchet space topology on H0(V,)H^{0}(V,\mathscr{F}). Thus (1) is proved.

We will prove second claim of the lemma by contradiction. Let us assume that g(s)g(s) is not uniformly bounded on the compact set U¯\overline{U} for all sSKs\in S_{K}. This implies that there exists a sequence s0,s1,s2,SKs_{0},s_{1},s_{2},\dots\in S_{K} such that the maximum value of |g(sk)||g(s_{k})| on the compact set U¯\overline{U} is at least kk. It follows from (1) that, by passing to a subsequence if necessary, the sequence sks_{k} converges to some section sH0(V,)s_{\infty}\in H^{0}(V,\mathscr{F}). As the map g:H0(V,)H0(V,𝒪Y)g:H^{0}(V,\mathscr{F})\rightarrow H^{0}(V,\mathscr{O}_{Y}) is a continuous map, the holomorphic functions g(sk)g(s_{k}) then converge uniformly on compact subsets to g(s)g(s_{\infty}). Consequently, |g(sk)||g(s_{k})| must be uniformly bounded on U¯\overline{U}, contradicting the previous assumption.

Proposition 4.6.

For every gH0(Y,)g\in H^{0}(Y,\mathscr{F}^{\ast}), the function ψ=log|g|H\psi=\log|g|_{H^{\ast}} is upper semi-continuous on YZY\setminus Z.

Proof.

Given an arbitrary point yYZy\in Y\setminus Z, we can assume that Y=BY=B, Z=Z=\emptyset, and y=0y=0 by choosing a sufficiently small open neighborhood of yy. In this case, gH0(B,F)g\in H^{0}(B,F^{\ast}), and we just need to show the upper semi-continuity of ψ=log|g|H\psi=\log|g|_{H^{\ast}} at the origin. Equivalently, we need to show that

(4.5) lim supk+ψ(yk)ψ(0)\displaystyle\limsup_{k\rightarrow+\infty}\psi(y_{k})\leq\psi(0)

for every sequence {yk}B\{y_{k}\}\in B converging to the origin. We may assume that ψ(yk)\psi(y_{k})\neq-\infty for all kk\in\mathbb{N}, and that the sequence ψ(yk)\psi(y_{k}) converges. By the definition of the metric on the dual bundle, there is a holomorphic section skH0(B,F)s_{k}\in H^{0}(B,F) for each kk\in\mathbb{N}, such that ψ(yk)=log|g(sk)|yk\psi(y_{k})=\log|g(s_{k})|_{y_{k}}. By Lemma 4.3, we can choose these sections such that |sk(yk)|H,yk=1|s_{k}(y_{k})|_{H,y_{k}}=1 and

B|sk|H2𝑑μK\int_{B}|s_{k}|^{2}_{H}d\mu\leq K

for some constant K0K\geq 0. If necessary, we can pass to a subsequence and let sks_{k} converge uniformly on compact subsets to some sH0(B,F)s\in H^{0}(B,F) using Proposition 4.5. Then, the holomorphic functions g(sk)g(s_{k}) uniformly converge on compact subsets to g(s)g(s). To prove (4.5), we must show log|g(s(0))|ψ(0)\log|g(s(0))|\leq\psi(0). The definition of the dual metric HH^{\ast} implies that

ψ=log|g|Hlog|g(s)|log|s|H.\psi=\log|g|_{H^{\ast}}\geq\log|g(s)|-\log|s|_{H}.

Therefore, it is equivalent to proving that |s(0)|H1|s(0)|_{H}\leq 1. According to the discussions in §3.1, there is a lower semi-continuous function Gk:B×X0[0,+)G_{k}:B\times X_{0}\rightarrow[0,+\infty) associated to each sks_{k}, such that

1=|sk(yk)|H,yk2=X0Gk(yk,)ω0nr(nr)!.1=|s_{k}(y_{k})|_{H,y_{k}}^{2}=\int_{X_{0}}G_{k}(y_{k},-)\frac{\omega_{0}^{n-r}}{(n-r)!}.

Similarly, the section ss determines a lower semi-continuous function G:B×X0[0,+)G:B\times X_{0}\rightarrow[0,+\infty). By Fatou’s lemma, we can deduce that

(4.6) |s(0)|H2=X0G(0,)ω0nr(nr)!X0lim infk+Gk(yk,)ω0nr(nr)!lim infk+X0Gk(yk,)ω0nr(nr)!=1.\displaystyle|s(0)|_{H}^{2}=\int_{X_{0}}G(0,-)\frac{\omega_{0}^{n-r}}{(n-r)!}\leq\int_{X_{0}}\liminf_{k\rightarrow+\infty}G_{k}(y_{k},-)\frac{\omega_{0}^{n-r}}{(n-r)!}\leq\liminf_{k\rightarrow+\infty}\int_{X_{0}}G_{k}(y_{k},-)\frac{\omega_{0}^{n-r}}{(n-r)!}=1.

Therefore, the proof is finished.

By referring to Proposition 4.4 and Proposition 4.6, verifying that ψ\psi satisfies the mean value inequalities, as derivable from Lemma 4.3, is all that’s required to show that ψ\psi extends to a plurisubharmonic function on YY.

Proposition 4.7.

For every holomorphic mapping γ:ΔYZ\gamma:\Delta\rightarrow Y\setminus Z, the function ψ=log|g|H\psi=\log|g|_{H^{\ast}} satisfies the mean-value inequality

(ψγ)(0)1πΔ(ψγ)𝑑μ.(\psi\circ\gamma)(0)\leq\frac{1}{\pi}\int_{\Delta}(\psi\circ\gamma)d\mu.

Here Δ\Delta denotes the unit disc in \mathbb{C}.

Proof.

Since the inequality holds when h+h\equiv+\infty, we can assume that hh is not identically equal to ++\infty. As the mapping f:XYf:X\rightarrow Y is a submersion over YZY\setminus Z, we can simplify the problem by considering the case when Y=ΔY=\Delta. If ψ(0)=\psi(0)=-\infty, then the mean-value inequality is always true. Assuming that ψ(0)\psi(0)\neq-\infty, we can select an element αF0\alpha\in F_{0}, where |α|H,0=1|\alpha|_{H,0}=1, such that

ψ(0)=log|g|H,0=log|g(α)|.\psi(0)=\log|g|_{H^{\ast},0}=\log|g(\alpha)|.

By Lemma 4.3, there exists a holomorphic section sH0(Δ,F)s\in H^{0}(\Delta,F) with s(0)=αs(0)=\alpha and 1πΔ|s|H2𝑑μ1\frac{1}{\pi}\int_{\Delta}|s|_{H}^{2}d\mu\leq 1. With the definition of the metric HH^{\ast} on the dual bundle, we can derive the inequality:

|g|H|g(s)||s|H.|g|_{H^{\ast}}\geq\frac{|g(s)|}{|s|_{H}}.

This inequality yields 2ψlog|g(s)|2log|s|H22\psi\geq\log|g(s)|^{2}-\log|s|^{2}_{H}. Integrating both sides leads to:

1πΔ2ψ𝑑μ1πΔlog|g(s)|2dμ1πΔlog|s|H2dμ.\frac{1}{\pi}\int_{\Delta}2\psi d\mu\geq\frac{1}{\pi}\int_{\Delta}\log|g(s)|^{2}d\mu-\frac{1}{\pi}\int_{\Delta}\log|s|_{H}^{2}d\mu.

Because log|g(s)|2\log|g(s)|^{2} follows the mean-value inequality, the first term on the right side is at least log|g(α)|2=2ψ(0)\log|g(\alpha)|^{2}=2\psi(0). Further, the function xlogxx\mapsto-\log x is convex, and the function |s|H2|s|_{H}^{2} is integrable. Hence, we can bound the second term using Jensen’s inequality as:

log(1πΔ|s|H2𝑑μ)log1=0.-\log\left(\frac{1}{\pi}\int_{\Delta}|s|_{H}^{2}d\mu\right)\geq-\log 1=0.

Combining both results, we we reach the conclusion:

1πΔψ𝑑μψ(0),\frac{1}{\pi}\int_{\Delta}\psi d\mu\geq\psi(0),

establishing the mean-value inequality. ∎

To summarize, the function ψ\psi is plurisubharmonic on YZY\setminus Z and is locally bounded on YY. Consequently, it it extends to a plurisubharmonic function on the entire space YY. Furthermore, the singular Hermitian metric HH extends to the torsion-free coherent sheaf \mathcal{F}, and the minimal extension property is a direct consequence of Lemma 4.3. With this, the proof of Theorem 1.1 is concluded.

References