A general extension theorem for cohomology classes on non reduced analytic spaces

TL;DR

This paper extends the Ohsawa-Takegoshi L^2 extension theorem to more general settings, including non-reduced subvarieties and non-compact Kähler manifolds, broadening its applicability in complex geometry.

Contribution

It generalizes the L^2 extension theorem to singular hermitian line bundles, non-reduced subvarieties, and non-compact Kähler manifolds, expanding the theorem's scope.

Findings

Extended L^2 theorem to singular hermitian line bundles

Allowed non-reduced subvarieties for extension

Applicable to non-compact Kähler manifolds

Abstract

The main purpose of this paper is to generalize the celebrated L${}^2$ extension theorem of Ohsawa-Takegoshi in several directions : the holomorphic sections to extend are taken in a possibly singular hermitian line bundle, the subvariety from which the extension is performed may be non reduced, the ambient manifold is K{\"a}hler and holomorphically convex, but not necessarily compact.

Equations

$${\Theta }_{E,h}:=i\partial \overline{\partial }\phi =i\partial \overline{\partial }{\phi }_0+i\partial \overline{\partial }u$$

$$\psi (z)=c\log \sum |g_j(z){|}^2+v(z),g_j\in {\mathcal{O}}_X(V),\text{leavevmode}\text{leavevmode}v\in C^{\infty }(V).$$

$$\mathcal{I}(e^{-\varphi})_{x_{0}}=\big{\{}f\in\mathcal{O}_{X,x_{0}}\,;\;\exists U\ni x_{0}\,,\;\int_{U}|f|^{2}e^{-\varphi}d\lambda<+\infty\big{\}}$$

$${\Theta }_{E,h}+(1+\alpha \delta )i\partial \overline{\partial }\psi \ge 0\text{in the sense of currents, for all}\text{leavevmode}\text{leavevmode}\alpha \in [0,1].$$

$$H^q(X,K_X\otimes E\otimes \mathcal{I}(h{e}^{-\psi }))\to H^q(X,K_X\otimes E\otimes \mathcal{I}(h))$$

$$H^q(X,K_X\otimes E\otimes \mathcal{I}(h))\to H^q(X,K_X\otimes E\otimes \mathcal{I}(h)/\mathcal{I}(h{e}^{-\psi }))$$

$$H^0(X,K_X\otimes E)\to H^0(Y,(K_X\otimes E{)}_{|Y}).$$

$$\mathcal{J}_{Y}:=\mathcal{I}(he^{-\psi}):\mathcal{I}(h)=\big{\{}f\in\mathcal{O}_{X}\,;\;f\cdot\mathcal{I}(h)\subset\mathcal{I}(he^{-\psi})\big{\}},$$

$$\overline{\partial }{u}_{\epsilon }+w_{\epsilon }=v$$

$$B=B_{E,h,\omega ,\eta ,\lambda }^{n,q}=[\eta {\Theta }_{E,h}-i\partial \overline{\partial }\eta -i{\lambda }^{-1}d\partial \eta \wedge \overline{\partial }\eta ,{\Lambda }_{\omega }]\in C^{\infty }(X,\mathrm{Herm}({\Lambda }^{n,q}{T}_X^{\ast }\otimes E))$$

$$M(\epsilon ):={\int }_X⟨(B+\epsilon I{)}^{-1}v,v⟩d{V}_{X,\omega }<+\infty ,$$

$${\int }_X(\eta +\lambda {)}^{-1}|u_{\epsilon }{|}^{2}d{V}_{X,\omega }+\frac{1}{\epsilon }{\int }_X|w_{\epsilon }{|}^{2}d{V}_{X,\omega }\le M(\epsilon ).$$

$$H^q({\pi }^{-1}(U),\mathcal{F})\simeq H^0(U,R^q{\pi }_{\ast }\mathcal{F})$$

$$i{\Theta }_{E,h}+(1+\alpha \delta )i\partial \overline{\partial }\psi \ge 0\text{in the sense of currents, for all α∈[0,1]}.$$

$$H^0(X,{\mathcal{O}}_X(K_X\otimes E)\otimes \mathcal{I}(h))\to H^0(Y,{\mathcal{O}}_X(K_X\otimes E)\otimes \mathcal{I}(h)/\mathcal{I}(h{e}^{-\psi }{)}_{|Y})$$

$$0=m_0<m_1<\cdots <m_p<\cdots such\; that\mathcal{I}(m\psi )=\mathcal{I}(m_p\psi )form\in [m_p,m_{p+1}[.$$

$$f\in H^0(Y,{\mathcal{O}}_X(K_X\otimes E{)}_{|Y})=H^0(X,{\mathcal{O}}_X(K_X\otimes E)\otimes {\mathcal{O}}_X/\mathcal{I}(e^{-m_p\psi })).$$

$$0\to \mathcal{I}(e^{-\psi })\to {\mathcal{O}}_X\to {\mathcal{O}}_X/\mathcal{I}(e^{-\psi })\to 0,$$

$${\tilde{f}}_i{|}_{Y\cap U_i}=f{|}_{Y\cap U_i}.$$

$${\tilde{f}}_i-{\tilde{f}}_j\in H^0(U_i\cap U_j,{\mathcal{O}}_X(K_X\otimes E)\otimes \mathcal{I}(e^{-\psi })).$$

$$\int_{X}|\overline{\partial}\widetilde{f}|_{\omega,h}^{2}e^{-\psi}dV_{X,\omega}=\int_{X}\sum_{j}\rho_{j}\Big{|}\sum_{i}(\overline{\partial}\rho_{i})\cdot(\widetilde{f}_{i}-\widetilde{f}_{j})\Big{|}_{\omega,h}^{2}e^{-\psi}dV_{X,\omega}<+\infty.$$

$$R_t$$

$${\Theta }_{E,h}+(1+\delta {\eta }_t^{-1}{\chi }_t^{\prime }(\psi ))i\partial \overline{\partial }\psi \ge 0\text{on }X.$$

$$R_t\ge 0\text{onleavevmodenobreak leavevmodenobreak }X$$

$$R_t\ge \frac{\delta }{16}i\partial \psi \wedge \overline{\partial }\psi \text{onleavevmodenobreak leavevmodenobreak }{W}_t=\{t<\psi <t+1\}.$$

$$\overline{\partial}u_{t,\varepsilon}+w_{t,\varepsilon}=v_{t}:=\overline{\partial}\big{(}\theta(\psi-t)\cdot\widetilde{f}\;\big{)}$$

$${\int }_X({\eta }_t+{\lambda }_t{)}^{-1}|u_{t,\epsilon }{|}_{\omega ,h}^2{e}^{-\psi }d{V}_{X,\omega }+\frac{1}{\epsilon }{\int }_X|w_{t,\epsilon }{|}_{\omega ,h}^2{e}^{-\psi }d{V}_{X,\omega }\le {\int }_X⟨(R_t+\epsilon I{)}^{-1}{v}_t,v_t⟩e^{-\psi }d{V}_{X,\omega },$$

$$v_{t}=\overline{\partial}\big{(}\theta(\psi-t)\widetilde{f}\;\big{)}=\theta^{\prime}(\psi-t)\,\overline{\partial}\psi\wedge\widetilde{f}+\theta(\psi-t)\,\overline{\partial}\widetilde{f}.$$

$${\int }_X|w_{t,\epsilon }{|}_{\omega ,h}^2{e}^{-\psi }d{V}_{X,\omega }\le \frac{128\epsilon }{\delta }{\int }_{\{t<\psi <t+1\}}|\tilde{f}{|}_{\omega ,h}^2{e}^{-\psi }d{V}_{X,\omega }+2{\int }_{\{\psi <t+1\}}|\overline{\partial }\tilde{f}{|}_{\omega ,h}^2{e}^{-\psi }d{V}_{X,\omega }.$$

$${\int }_{\{t<\psi <t+1\}}|\tilde{f}{|}_{\omega ,h}^2{e}^{-\psi }d{V}_{X,\omega }\le C_1{e}^{-t},$$

Full text

This work was supported by the Agence Nationale de la Recherche grant “Convergence de Gromov-Hausdorff en géométrie kählérienne” (ANR-GRACK), the European Research Council project “Algebraic and Kähler Geometry” (ERC-ALKAGE, grant No. 670846 from September 2015), the Japan Society for the Promotion of Science Grant-in-Aid for Young Scientists (B) (JSPS, grant No. 25800051) and the JSPS Program for Advancing Strategic International Networks to Accelerate the Circulation of Talented Researchers.

In memory of Professor Lu QiKeng (1927–2015)

1. Introduction and preliminaries

The purpose of this paper is to generalize the celebrated $L^{2}$ extension theorem of Ohsawa-Takegoshi [OT87] under the weakest possible hypotheses, along the lines of [Dem15b] and [Mat16b]. Especially, the ambient complex manifold $X$ is a Kähler manifold that is only assumed to be holomorphically convex, and is not necessarily compact; by the Remmert reduction theorem, this is the same as a Kähler manifold $X$ that admits a proper holomorphic map $\pi:X\to S$ onto a Stein complex space $S$. This allows in particular to consider relative situations over a Stein base. We consider a holomorphic line bundle $E\to X$ equipped with a singular hermitian metric $h$, namely a metric which can be expressed locally as $h=e^{-\varphi}$ where $\varphi$ is a quasi-psh function, i.e. a function that is locally the sum $\varphi=\varphi_{0}+u$ of a plurisubharmonic function $\varphi_{0}$ and of a smooth function $u$. Such a bundle admits a curvature current

[TABLE]

which is locally the sum of a positive $(1,1)$-current $i\partial\overline{\partial}\varphi_{0}$ and a smooth $(1,1)$-form $i\partial\overline{\partial}u$. Our goal is to extend sections that are defined on a (non necessarily reduced) complex subspace $Y\subset X$, when the structure sheaf $\mathcal{O}_{Y}:=\mathcal{O}_{X}/\mathcal{I}(e^{-\psi})$ is given by the multiplier ideal sheaf of a quasi-psh function $\psi$ with neat analytic singularities, i.e. locally on a neighborhood $V$ of an arbitrary point $x_{0}\in X$ we have

[TABLE]

Let us recall that the multiplier ideal sheaf $\mathcal{I}(e^{-\varphi})$ of a quasi-psh function $\varphi$ is defined by

[TABLE]

with respect to the Lebesgue measure $\lambda$ in some local coordinates near $x_{0}$. As usual, we also denote by $K_{X}=\Lambda^{n}T^{*}_{X}$ the canonical bundle of an $n$-dimensional complex manifold $X$. As is well known, $\mathcal{I}(e^{-\varphi})\subset\mathcal{O}_{X}$ is a coherent ideal sheaf (see e.g. [Dem-book]). Our main result is given by the following general statement.

Theorem 1.1**.**

Let $E$ be a holomorphic line bundle over a holomorphically convex Kähler manifold $X$. Let $h$ be a possibly singular hermitian metric on $E$, $\psi$ a quasi-psh function with neat analytic singularities on $X$. Assume that there exists a positive continuous function $\delta>0$ on $X$ such that

[TABLE]

Then the morphism induced by the natural inclusion $\mathcal{I}(he^{-\psi})\to\mathcal{I}(h)$

[TABLE]

is injective for every $q\geq 0$. In other words, the morphism induced by the natural sheaf surjection $\mathcal{I}(h)\to\mathcal{I}(h)/\mathcal{I}(he^{-\psi})$

[TABLE]

is surjective for every $q\geq 0$.

Remark 1.2**.**

If $h$ is smooth, we have $\mathcal{I}(h)=\mathcal{O}_{X}$ and $\mathcal{I}(h)/\mathcal{I}(he^{-\psi})=\mathcal{O}_{X}/\mathcal{I}(e^{-\psi}):=\mathcal{O}_{Y}$ where $Y$ is the zero subvariety of the ideal sheaf $\mathcal{I}(e^{-\psi})$. Then for $q=0$, the surjectivity statement can be interpreted an extension theorem for holomorphic sections, with respect to the restriction morphism

[TABLE]

In general, the quotient sheaf $\mathcal{I}(h)/\mathcal{I}(he^{-\psi})$ is supported in an analytic subvariety $Y\subset X$, which is the zero set of the quotient ideal

[TABLE]

and (6) can be considered as a restriction morphism to $Y$.∎**

The crucial idea of the proof is to prove the results (say, in the form of the surjectivity statement), only up to approximation. This is done by solving a $\overline{\partial}$-equation

[TABLE]

where the right hand side $v$ is given and $w_{\varepsilon}$ is an error term such that $\|w_{\varepsilon}\|=O(\varepsilon^{a})$ as $\varepsilon\to 0$, for some constant $a>0$. A twisted Bochner-Kodaira-Nakano identity introduced by Donnelly and Fefferman [DF83], and Ohsawa and Takegoshi [OT87] is used for that purpose, with an additional correction term. The version we need can be stated as follows.

Proposition 1.3**.**

(see [Dem15b, Prop. 3.12])* Let $X$ be a complete Kähler manifold equipped with a $($non necessarily complete$)$ Kähler metric $\omega$, and let $(E,h)$ be a Hermitian vector bundle over $X$. Assume that there are smooth and bounded functions $\eta,\,\lambda>0$ on $X$ such that the curvature operator*

[TABLE]

satisfies $B+\varepsilon I>0$ for some $\varepsilon>0$ $($so that $B$ can be just semi-positive or even slightly negative; here $I$ is the identity endomorphism$)$. Given a section $v\in L^{2}(X,\Lambda^{n,q}T^{*}_{X}\otimes E)$ such that $\overline{\partial}v=0$ and

[TABLE]

there exists an approximate solution $f_{\varepsilon}\in L^{2}(X,\Lambda^{n,q-1}T^{*}_{X}\otimes E)$ and a correction term $w_{\varepsilon}\in L^{2}(X,\Lambda^{n,q}T^{*}_{X}\otimes E)$ such that $\overline{\partial}u_{\varepsilon}=v-w_{\varepsilon}$ and

[TABLE]

Moreover, if $v$ is smooth, then $u_{\varepsilon}$ and $w_{\varepsilon}$ can be taken smooth.

In our situation, the main part of the solution, namely $u_{\varepsilon}$, may very well explode as $\varepsilon\to 0$. In order to show that the equation $\overline{\partial}u=v$ can be solved, it is therefore needed to check that the space of coboundaries is closed in the space of cocycles in the Fréchet topology under consideration (here, the $L^{2}_{\rm loc}$ topology), in other words, that the related cohomology group $H^{q}(X,\mathcal{F})$ is Hausdorff. In this respect, the fact of considering $\overline{\partial}$-cohomology of smooth forms equipped with the $C^{\infty}$ topology on the one hand, or cohomology of forms $u\in L^{2}_{\rm loc}$ with $\overline{\partial}u\in L^{2}_{\rm loc}$ on the other hand, yields the same topology on the resulting cohomology group $H^{q}(X,\mathcal{F})$. This comes from the fact that both complexes yield fine resolutions of the same coherent sheaf $\mathcal{F}$, and the topology of $H^{q}(X,\mathcal{F})$ can also be obtained by using Čech cochains with respect to a Stein covering $\mathcal{U}$ of $X$. The required Hausdorff property then comes from the following well known fact.

Lemma 1.4**.**

Let $X$ be a holomorphically convex complex space and $\mathcal{F}$ a coherent analytic sheaf over $X$. Then all cohomology groups $H^{q}(X,\mathcal{F})$ are Hausdorff with respect to their natural topology $($induced by the Fréchet topology of local uniform convergence of holomorphic cochains$)$.111 It was pointed out to us by Prof. Takeo Ohsawa that this result does not hold under the assumption that $X$ is weakly pseudoconvex, i.e., if we only assume that $X$ admits a smooth psh exhaustion. A counter-example can be derived from [Kaz84]. As a consequence, it is unclear whether the results of the present paper extend to the Kähler weakly pseudoconvex case, although the main $L^{2}$ estimates are still valid in that situation.

In fact, the Remmert reduction theorem implies that $X$ admits a proper holomorphic map $\pi:X\to S$ onto a Stein space $S$, and Grauert’s direct image theorem shows that all direct images $R^{q}\pi_{*}\mathcal{F}$ are coherent sheaves on $S$. Now, as $S$ is Stein, Leray’s theorem combined with Cartan’s theorem B tells us that we have an isomorphism $H^{q}(X,\mathcal{F})\simeq H^{0}(S,R^{q}\pi_{*}\mathcal{F})$. More generally, if $U\subset S$ is a Stein open subset, we have

[TABLE]

and when $U\Subset S$ is relatively compact, it is easily seen that this a topological isomorphism of Fréchet spaces since both sides are $\mathcal{O}_{S}(U)$ modules of finite type and can be seen as a Fréchet quotient of some direct sum $\mathcal{O}_{S}(U)^{\oplus N}$ by looking at local generators and local relations of $R^{q}\pi_{*}\mathcal{F}$. Therefore $H^{q}(X,\mathcal{F})\simeq H^{0}(S,R^{q}\pi_{*}\mathcal{F})$ is a topological isomorphism and the space of sections in the right hand side is a Fréchet space. In particular, $H^{q}(X,\mathcal{F})$ is Hausdorff.∎The isomorphism (8) shows that it is enough to prove Theorem 1.1 locally over $X$, i.e., we can replace $X$ by $X^{\prime}=\pi^{-1}(S^{\prime})\Subset X$ where $S^{\prime}\Subset S$. Therefore, we can assume that $\delta>0$ is a constant rather than a continuous function.

2. Proof of the extension theorem

In this section, we give a proof of Theorem 1.1 based on a generalization of the arguments of [Dem15b, Th. 2.14]. We start by proving the special case of the extension result for holomorphic sections ($q=0$).

Theorem 2.1**.**

Let $(X,\omega)$ be a holomorphically convex Kähler manifold and $\psi$ be a quasi-psh function with neat analytic singularities. Let $E$ be a line bundle with a possibly singular metric $h$, and $Y$ the support of the sheaf $\mathcal{I}(h)/\mathcal{I}(he^{-\psi})$, along with the structure sheaf $\mathcal{O}_{Y}:=\mathcal{I}(he^{-\psi}):\mathcal{I}(h)$. Assume that there is a continuous function $\delta>0$ such that

[TABLE]

Then the restriction morphism

[TABLE]

is surjective.

Proof.

(a) Let us first assume for simplicity that $h$ is smooth. We will explain the general case later. Then $\mathcal{I}(h)=\mathcal{O}_{X}$ and $\mathcal{I}(h)/\mathcal{I}(he^{-\psi})=\mathcal{O}_{Y}=\mathcal{O}_{X}/\mathcal{I}(e^{-\psi})$. After possibly shrinking $X$ into a relatively compact holomorphically convex open subset $X^{\prime}=\pi^{-1}(S^{\prime})\Subset X$, we can suppose that $\delta>0$ is a constant and that $\psi\leq 0$, after subtracting a large constant to $\psi$. Also, without loss of generality, we can assume that $\psi$ admits a discrete sequence of “jumping numbers”

[TABLE]

Since $\psi$ is assumed to have analytic singularities, this follows from using a log resolution of singularities, thanks to the Hironaka desingularization theorem (by the much deeper result of [GZ15] on the strong openness conjecture, one could even possibly eliminate the assumption that $\psi$ has analytic singularities). We fix here $p$ such that $m_{p}\leq 1<m_{p+1}$, and in the notation of [Dem15b], we let $Y=Y^{(m_{p})}$ be defined by the non necessarily reduced structure sheaf $\mathcal{O}_{Y}=\mathcal{O}_{X}/\mathcal{I}(e^{-\psi})=\mathcal{O}_{X}/\mathcal{I}(e^{-m_{p}\psi})$.

Step 1** (Construction of a smooth extension).**

Take

[TABLE]

Let $\mathcal{U}=(U_{i})$ be a Stein covering of $X$ and let $(\rho_{i})$ be a partition of unity subordinate to $(U_{i})$. Thanks to the exact sequence

[TABLE]

we can find a $\widetilde{f}_{i}\in H^{0}(U_{i},\mathcal{O}_{X}(K_{X}\otimes E))$ such that

[TABLE]

Then (10) implies that

[TABLE]

As a consequence, the smooth section $\widetilde{f}:=\sum\limits_{i}\rho_{i}\cdot\widetilde{f}_{i}$ is a smooth extension of $f$ and satisfies $\overline{\partial}\widetilde{f}=\sum\limits_{i}(\overline{\partial}\rho_{i})\cdot(\widetilde{f}_{i}-\widetilde{f}_{j})$ on $U_{j}$, hence

[TABLE]

Step 2** ($L^{2}$-estimates).**

We follow here the arguments of [Dem15b, proof of th. 2.14, p. 217]. Let $t\in\mathbb{Z}^{-}$ and let $\chi_{t}$ be the negative convex increasing function defined in [Dem15b, (5.8$*$), p. 211]. Put $\eta_{t}:=1-\delta\cdot\chi_{t}(\psi)$ and $\lambda_{t}:=2\delta\frac{(\chi_{t}^{2}(\psi))^{2}}{\chi_{t}^{\prime\prime}(\psi)}$. We set

[TABLE]

Note that $\chi_{t}^{\prime\prime}(\psi)\geq\frac{1}{8}$ on $W_{t}=\{t<\psi<t+1\}$. The curvature assumption (4) implies

[TABLE]

As in [Dem15b], we find

[TABLE]

and

[TABLE]

Let $\theta:[-\infty,+\infty[{}\rightarrow[0,1]$ be a smooth non increasing real function satisfying $\theta(x)=1$ for $x\leq 0$, $\theta(x)=0$ for $x\geq 1$ and $|\theta^{\prime}|\leq 2$. By applying the $L^{2}$ estimate (Proposition 1.3), for every $\varepsilon>0$ we can find sections $u_{t,\varepsilon}$, $w_{t,\varepsilon}$ satisfying

[TABLE]

and

[TABLE]

where

[TABLE]

Combining (13), (14), (16) and (17), we get $\int_{X}|u_{t,\varepsilon}|_{\omega,h}^{2}e^{-\psi}dV_{X,\omega}<+\infty$ and

[TABLE]

We now estimate the right hand side of (18). Since $\widetilde{f}$ is smooth, we have an obvious upper bound of the first term

[TABLE]

where $C_{1}$ is the $C^{0}$ norm of $\widetilde{f}$. For the second term, thanks to (9), (11) and (12), we have

[TABLE]

for any $\alpha\in{}]0,m_{p+1}-1[$. As a consequence, we get

[TABLE]

for some constant $C_{2}$ depending only on $\alpha$. By taking $\varepsilon=e^{(1+\alpha)t}$, (18), (19) and (21) imply

[TABLE]

for some constant $C_{3}$, whence the error tends to [math] as $t\to-\infty$ and $\varepsilon\to 0$.

Step 3** (Final conclusion).**

Putting everything together and redefining $u_{t}=u_{t,\varepsilon}$, $w_{t}=w_{t,\varepsilon}$ for simplicity of notation, we get

[TABLE]

and

[TABLE]

After shrinking $X$, we can assume that we have a finite Stein covering $\mathcal{U}=(U_{i})$ where the $U_{i}$ are biholomorphic to bounded pseudoconvex domains. The standard Hörmander $L^{2}$ estimates then provide $L^{2}$ sections $s_{t,j}$ on $U_{j}$ such that $\overline{\partial}s_{t,j}=w_{t}$ on $U_{j}$ and

[TABLE]

Then

[TABLE]

As $\overline{\partial}(s_{t,j}-s_{t,i})=0$ on $U_{i}\cap U_{j}$, the difference is holomorphic and the right hand side of (26) is smooth. Moreover, (25) shows that these differences converge uniformly to [math], hence the right hand side of (26) converges to [math] in $C^{\infty}$ topology. The left hand side implies that this is a coboundary in the $C^{\infty}$ Dolbeault resolution of $\mathcal{O}_{X}(K_{X}\otimes E)$. By applying Lemma 1.4, we conclude that there is a $C^{\infty}$ section $\sigma_{t}$ of $K_{X}\otimes E$ converging uniformly to [math] on compact subsets of $X$ as $t\to-\infty$, such that $\overline{\partial}\sigma_{t}=\sum\limits_{j}(\overline{\partial}\rho_{j})\cdot s_{t,j}$ on $X$. This implies that

[TABLE]

is holomorphic on $X$. Hörmander’s $L^{2}$ estimates also produce local smooth solutions $\sigma_{t,i}$ on $U_{i}$ with the additional property that $\lim\limits_{t\to-\infty}\int_{U_{i}}|\sigma_{t,i}|_{\omega,h}^{2}e^{-\psi}dV_{X,\omega}=0$. Therefore

[TABLE]

is holomorphic on $U_{i}$ and $\widetilde{f}_{t}-\widetilde{f}_{t,i}$ converges uniformly to [math] on compact subsets of $U_{i}$. However, by construction, $\widetilde{f}_{t,i}-\widetilde{f}_{i}$ is a holomorphic section on $U_{i}$ that satisfies the $L^{2}$ estimate with respect to the weight $e^{-\psi}$, hence $\widetilde{f}_{t,i}-\widetilde{f}_{i}$ is a section of $\mathcal{O}_{X}(K_{X}\otimes E)\otimes\mathcal{I}(e^{-\psi})$ on $U_{i}$, in other words the image of $\widetilde{f}_{t,i}$ in

[TABLE]

coincides with $f_{|U_{i}}$. As a consequence, the image of $\widetilde{f}_{t}$ in

[TABLE]

converges to $f$. By the direct image argument used in the preliminary section, this density property implies the surjectivity of the restriction morphism to $Y$.

(b) We now prove the theorem for the general case when $h=e^{-\varphi}$ is not necessarily smooth. We can reduce ourselves to the case when $\psi$ has divisorial singularities (see [Dem15b] or the next section for a more detailed argument). Let us pick a section

[TABLE]

By using the same reasoning as in Step 1, we can find a smooth extension $\widetilde{f}\in\mathcal{C}^{\infty}(X,K_{X}\otimes E)$ of $f$ such that

[TABLE]

For every $t\in\mathbb{Z}^{-}$ fixed, as $\psi$ has divisorial singularities, we still have

[TABLE]

where $(i\partial\overline{\partial}\psi)_{{\rm ac}}$ is the absolutely continuous part of $i\partial\overline{\partial}\psi$. The regularization techniques of [DPS01] and [Dem15a, Th. 1.7, Remark 1.11] (cf. also the next section) produce a family of singular metrics $\{h_{t,\varepsilon}\}_{k=1}^{+\infty}$ which are smooth in the complement $X\smallsetminus Z_{t,\varepsilon}$ of an analytic set, such that $\mathcal{I}(h_{t,\varepsilon})=\mathcal{I}(h)$, $\mathcal{I}(h_{t,\varepsilon}e^{-\psi})=\mathcal{I}(he^{-\psi})$ and

[TABLE]

The additional error term $-\frac{1}{2}\varepsilon\omega$ is irrelevant when we use Proposition 1.3, as it is absorbed by taking the hermitian operator $B+\varepsilon I$. Therefore for every $t\in\mathbb{Z}^{-}$, with the adjustment $\varepsilon=e^{\alpha t}$, $\alpha\in{}]0,m_{k+1}-1[$, we can find a singular metric $h_{t}=h_{t,\varepsilon}$ which is smooth in the complement $X\setminus Z_{t}$ of an analytic set, such that $\mathcal{I}(h_{t})=\mathcal{I}(h)$, $\mathcal{I}(h_{t}e^{-\psi})=\mathcal{I}(he^{-\psi})$ and $h_{t}\uparrow h$ as $t\rightarrow-\infty$, and approximate solutions of the $\overline{\partial}$-equation such that

[TABLE]

and

[TABLE]

Proposition 1.3 can indeed be applied since $X\smallsetminus Z_{t}$ is complete Kähler (at least after we shrink $X$ a little bit as $X^{\prime}=\pi^{-1}(S^{\prime})$, cf. [Dem82]). The theorem is then proved by using the same argument as in Step 3; it is enough to notice that the holomorphic sections $s_{t,j}-s_{t,i}$ and $\widetilde{f}_{t,i}-\widetilde{f}_{i}$ satisfy the $L^{2}$-estimate with respect to $(h_{t},\psi)$ [instead of the expected $(h,\psi)$], but the multiplier ideal sheaves involved are unchanged. The Hausdorff property is applied to the cohomology group $H^{1}(X,\mathcal{O}_{X}(K_{X}\otimes E)\otimes\mathcal{I}(h))$ instead of $H^{1}(X,K_{X}\otimes E)$, and the density property to the morphism of direct image sheaves

[TABLE]

over the Stein space $S$. ∎

Proof of the extension theorem for degree q cohomology classes. The reasoning is extremely similar, so we only explain the few additional arguments needed. In fact, Proposition 1.3 can be applied right away to arbitrary $(n,q)$-forms with $q\geq 1$, and the twisted Bochner-Kodaira-Nakano inequality yields exactly the same estimates. Any cohomology class in

[TABLE]

is represented by a holomorphic Čech $q$-cocycle with respect to the Stein covering $\mathcal{U}=(U_{i})$, say

[TABLE]

By the standard sheaf theoretic isomorphisms with Dolbeault cohomology (cf. e.g. [Dem-e-book]), this class is represented by a smooth $(n,q)$-form

[TABLE]

by means of a partition of unity $(\rho_{i})$ subordinate to $(U_{i})$. This form is to be interpreted as a form on the (non reduced) analytic subvariety $Y$ associated with the ideal sheaf $\mathcal{J}=\mathcal{I}(he^{-\psi}):\mathcal{I}(h)$ and the structure sheaf $\mathcal{O}_{Y}=\mathcal{O}_{X}/\mathcal{J}$. We get an extension as a smooth (no longer $\overline{\partial}$-closed) $(n,q)$-form on $X$ by taking

[TABLE]

where $\widetilde{c}_{i_{0}\ldots i_{q}}$ is an extension of $c_{i_{0}\ldots i_{q}}$ from $U_{i_{0}}\cap\ldots\cap U_{i_{q}}\cap Y$ to $U_{i_{0}}\cap\ldots\cap U_{i_{q}}$. Again, we can find approximate $L^{2}$ solutions of the $\overline{\partial}$-equation such that

[TABLE]

and

[TABLE]

The difficulty is that $L^{2}$ sections cannot be restricted in a continuous way to a subvariety. In order to overcome this problem, we play again the game of returning to Čech cohomology by solving inductively $\overline{\partial}$-equations for $w_{t}$ on $U_{i_{0}}\cap\ldots\cap U_{i_{k}}$, until we reach an equality

[TABLE]

with holomorphic sections $s_{t,I}=s_{t,i_{0}\ldots i_{q}}$ on $U_{I}=U_{i_{0}}\cap\ldots\cap U_{i_{q}}$, such that

[TABLE]

Then the right hand side of (28) is smooth, and more precisely has coefficients in the sheaf $\mathcal{C}^{\infty}\otimes_{\mathcal{O}}\mathcal{I}(he^{-\psi})$, and $\widetilde{w}_{t}\to 0$ in $C^{\infty}$ topology. A priori, $\widetilde{u}_{t}$ is an $L^{2}$ $(n,q)$-form equal to $u_{t}$ plus a combination $\sum\rho_{i}s_{t,i}$ of the local solutions of $\overline{\partial}s_{t,i}=w_{t}$, plus $\sum\rho_{i}s_{t,i,j}\wedge\overline{\partial}\rho_{j}$ where $\overline{\partial}s_{t,i,j}=s_{t,j}-s_{t,i}$, plus etc $\ldots$ , and is such that

[TABLE]

Since $H^{q}(X,\mathcal{O}_{X}(K_{X}\otimes E)\otimes\mathcal{I}(he^{-\psi}))$ can be computed with the $L^{2}_{\rm loc}$ resolution of the coherent sheaf, or alternatively with the $\overline{\partial}$-complex of $(n,{\scriptstyle\bullet})$-forms with coefficients in $\mathcal{C}^{\infty}\otimes_{\mathcal{O}}\mathcal{I}(he^{-\psi})$, we may assume that $\widetilde{u}_{t}\in\mathcal{C}^{\infty}\otimes_{\mathcal{O}}\mathcal{I}(he^{-\psi})$, after playing again with Čech cohomology. Lemma 1.4 yields a sequence of smooth $(n,q)$-forms $\sigma_{t}$ with coefficients in $\mathcal{C}^{\infty}\otimes_{\mathcal{O}}\mathcal{I}(h)$, such that $\overline{\partial}\sigma_{t}=\widetilde{w}_{t}$ and $\sigma_{t}\to 0$ in $C^{\infty}$-topology. Then $\widetilde{f}_{t}=\theta(\psi-t)\cdot\widetilde{f}-\widetilde{u}_{t}-\sigma_{t}$ is a $\overline{\partial}$-closed $(n,q)$-form on $X$ with values in $\mathcal{C}^{\infty}\otimes_{\mathcal{O}}\mathcal{I}(h)\otimes\mathcal{O}_{X}(E)$, whose image in $H^{q}(X,\mathcal{O}_{X}(K_{X}\otimes E)\otimes\mathcal{I}(h)/\mathcal{I}(he^{-\psi}))$ converges to $\{f\}$ in $C^{\infty}$ Fréchet topology. We conclude by a density argument on the Stein space $S$, by looking at the coherent sheaf morphism

[TABLE]

3. An alternative proof based on injectivity theorems

We give here an alternative proof based on injectivity theorems, in the case when $X$ is compact Kähler. The case of a holomorphically convex manifold is entirely similar, so we will content ourselves to indicate the required additional arguments at the end.

Proof of Theorem 1.1..

First of all, we reduce the proof of Theorem 1.1 to the case when $\psi$ has divisorial singularities. Since $\psi$ has analytic singularities, there exists a modification $\pi\colon X^{\prime}\to X$ such that the pull-back $\pi^{*}\psi$ has divisorial singularities. For the singular hermitian line bundle $(E^{\prime},h^{\prime}):=(\pi^{*}E,\pi^{*}h)$ and the quasi-psh function $\psi^{\prime}:=\pi^{*}\psi$, we can easily check that

[TABLE]

Hence we obtain the following commutative diagram $:$

[TABLE]

where $f$, $g$ are the morphisms induced by the natural inclusions and $\pi^{*}$ is the natural edge morphism. It follows that the left edge morphism $\pi^{*}$ is an isomorphism since the curvature of the singular hermitian metric $h^{\prime}e^{-\psi^{\prime}}$ on $E^{\prime}$ is semi-positive by the assumption. Indeed, even if $h^{\prime}$ does not have analytic singularities, we can see that

[TABLE]

by [Mat16b, Corollary 1.5]. (In the case of $X$ being a projective variety, a relatively easy proof can be found in [FM16].) If Theorem 1.1 can be proven when $\psi$ has divisorial singularities, it follows that the morphism $g$ in the above diagram is injective since $(E^{\prime},h^{\prime})=(\pi^{*}E,\pi^{*}h)$ and $\psi^{\prime}=\pi^{*}\psi$ satisfy the assumptions in Theorem 1.1 and $\psi^{\prime}$ has divisorial singularities. Therefore the morphism $f$ is also injective by the commutative diagram.

Now we explain the idea of the proof of Theorem 1.1. If we can obtain equisingular approximations $h_{\varepsilon}$ of $h$ satisfying the following properties :

[TABLE]

then a proof similar to [FM16] works, where $\omega$ is a fixed Kähler form on $X$. In the case when $\psi$ has divisorial singularities, we can attain either of the above curvature properties, but we do not know whether we can attain them at the same time. For this reason, we will look for an essential curvature condition arising from the assumptions on the curvatures in Theorem 1.1, in order to use the “twisted” Bochner-Kodaira-Nakano identity.

From now on, we consider a quasi-psh $\psi$ with divisorial singularities. Then there exist an effective $\mathbb{R}$-divisor $D$ and a smooth $(1,1)$-form $\gamma$ on $X$ such that

[TABLE]

in the sense of $(1,1)$-currents, where $[D]$ denotes the current of the integration over $D$. For the irreducible decomposition $D=\sum_{i=1}^{N}a_{i}D_{i}$ and the defining section $t_{i}$ of $D_{i}$, we can take a smooth hermitian metric $b_{i}$ on $D_{i}$ such that

[TABLE]

For a positive number $0<c\ll 1$, we define the continuous functions $\sigma$ and $\eta$ on $X$ by

[TABLE]

where $\chi(t):=t-\log(-t)$.

Remark 3.1**.**

(i) We may assume that $|s|_{b}^{2}<1/5$ by subtracting a positive constant from $\psi$. Further we may assume that $\sigma<\log(1/5)$ and $1<\chi^{\prime}(\sigma)<7/4$ by choosing a sufficiently small $c>0$.

(ii) Further, the function $\eta$ is a continuous function on $X$ with $\eta>1/c$. The function $\eta$ is smooth on $X\setminus D$, but it need not be smooth on $X$ since $|s|_{b}^{2}$ is not smooth in the case when $0<a_{i}<1$ for some $i$. **

Throughout the proof, we fix a Kähler form $\omega$ on $X$. The following proposition gives a suitable approximation of a singular hermitian metric $h$ on $E$, which enables us to use the twisted Bochner-Kodaira-Nakano identity. The proof is based on the argument in [Ohs04], [Fuj13] and the equisingular approximation theorem in [DPS01, Theorem 2.3].

Proposition 3.2**.**

There exist singular hermitian metrics $\{h_{\varepsilon}\}_{0<\varepsilon\ll 1}$ on $E$ with the following properties $:$

• (a)

$h_{\varepsilon}$* is smooth on $X\setminus Z_{\varepsilon}$, where $Z_{\varepsilon}$ is a proper subvariety on $X$.*

• (b)

$h_{\varepsilon^{\prime}}\leq h_{\varepsilon^{\prime\prime}}\leq h$* holds on $X$ for $\varepsilon^{\prime}>\varepsilon^{\prime\prime}>0$.*

• (c)

$\mathcal{I}(h)=\mathcal{I}(h_{\varepsilon})$* and $\mathcal{I}(he^{-\psi})=\mathcal{I}(h_{\varepsilon}e^{-\psi})$ on $X$.*

• (d)

$\eta(\Theta_{h_{\varepsilon}}(E)+\gamma)-i\partial\overline{\partial}\eta-\eta^{-2}i\partial\eta\wedge\overline{\partial}\eta\geq-\varepsilon\omega$* on $X\setminus D$.*

• (e)

For arbitrary $t>0$, by taking a sufficiently small $\varepsilon>0$, we have

[TABLE]

*where $\phi$ *$($*resp. \phi_{\varepsilon}$$) is a local weight of $h$ *$($resp. h_{\varepsilon}$$).

Proof.

We fix a a sufficiently small $c$ with $7c/4\leq\delta$. Then, by Remark 3.1, we can easily check that

[TABLE]

In particular, it follows that

[TABLE]

since $\psi$ has divisorial singularities and satisfies the assumptions in Theorem 1.1. By applying the equisingular approximation theorem ([DPS01, Theorem 2.3]) to $h$, we can take singular hermitian metrics $\{h_{\varepsilon}\}_{0<\varepsilon\ll 1}$ on $E$ satisfying properties (a), (b), (e), the former conclusion of (c), and the following curvature property:

[TABLE]

Now we check property (d) from the above curvature property. The function $\eta$ may not be smooth on $X$, but it is smooth on $X\setminus D$. Therefore the same computation as in [Ohs04] and [Fuj13] works on $X\setminus D$. In particular, from a complicated but straightforward computation, we obtain

[TABLE]

on $X\setminus D$ (see [Fuj13] for the precise computation). Then, by $-\gamma=\Theta_{b}(D)$ on $X\setminus D$, we can see that

[TABLE]

on $X\setminus D$. A straightforward computation yields that $\chi^{\prime\prime}(\sigma)/\chi^{\prime}(\sigma)^{2}\geq 1/\eta^{2}$, and thus the first term is semi-positive. Since $\eta$ is bounded above, we infer that property (d) holds.

Finally we check the last conclusion of property (c) by proving the following lemma, which can be obtained from the strong openness theorem (see [GZ15], [Lem14], [Hie14]) and property (e).

Lemma 3.3**.**

For a quasi-psh function $\varphi$, we have $\mathcal{I}(he^{-\varphi})=\mathcal{I}(h_{\varepsilon}e^{-\varphi})$. In particular, we obtain the last conclusion of property (c).

Proof.

We have the inclusion $\mathcal{I}(he^{-\varphi})\subset\mathcal{I}(h_{\varepsilon}e^{-\varphi})$ by $h_{\varepsilon}\leq h$. To get the converse inclusion, we consider a local holomorphic function $g$ such that $|g|^{2}e^{-\varphi-\phi_{\varepsilon}}$ is integrable, where $\phi_{\varepsilon}$ (resp. $\phi$) is a local weight of $h_{\varepsilon}$ (resp. $h$). Then H$\rm{\ddot{o}}$lder’s inequality yields

[TABLE]

where $p$, $q$ are real numbers such that $1/p+1/q=1$ and $p>1$. By the strong openness theorem, the function $|g|^{2p}e^{-p(\varphi+\phi_{\varepsilon})}$ is integrable when $p$ is sufficiently close to one. On the other hand, we have

[TABLE]

The right hand side is finite for a sufficiently small $\varepsilon$ by property (e). ∎

This concludes the proof of Proposition 3.2. ∎

From now on, we proceed to prove Theorem 1.1 by using Proposition 3.2. In the same way as in [FM16, Section 5], one constructs a family of complete Kähler forms $\{\omega_{\varepsilon,\delta}\}_{0<\delta\ll 1}$ on $Y_{\varepsilon}:=X\setminus(Z_{\varepsilon}\cup D)$ with the following properties :

• (A)

$\omega_{\varepsilon,\delta}$ is a complete Kähler form on $Y_{\varepsilon}:=X\setminus(Z_{\varepsilon}\cup D)$ for every $\delta>0$.

• (B)

$\omega_{\varepsilon,\delta}\geq\omega$ on $Y_{\varepsilon}$ for every $\delta\geq 0$.

• (C)

For every point $p$ in $X$, there exists a bounded function $\Psi_{\varepsilon,\delta}$ on an open neighborhood $B_{p}$ such that $\omega_{\varepsilon,\delta}=i\partial\overline{\partial}\Psi_{\varepsilon,\delta}$ on $B_{p}$ and $\Psi_{\varepsilon,\delta}$ converges uniformly to a bounded function that is independent of $\varepsilon$.

For simplicity, we put $H:=he^{-\psi}\text{ and }H_{\varepsilon}:=h_{\varepsilon}e^{-\psi}$. We consider a cohomology class $\beta\in H^{q}(X,K_{X}\otimes E\otimes\mathcal{I}(H))$ such that $\beta=0\in H^{q}(X,K_{X}\otimes E\otimes\mathcal{I}(h))$. By the De Rham-Weil isomorphism

[TABLE]

the cohomology class $\beta$ can be represented by a $\overline{\partial}$-closed $E$-valued $(n,q)$-form $u$ with $\|u\|_{H,\omega}<\infty$ (that is, $\beta=\{u\}$). Here $L^{n,\bullet}_{(2)}(E)_{H,\omega}$ is the $L^{2}$-space of $E$-valued $(n,\bullet)$-forms on $X$ with respect to the $L^{2}$-norm $\|\bullet\|_{H,\omega}$ defined by

[TABLE]

where $dV_{\omega}:=\omega^{n}/n!$ and $n:=\dim X$. For the $L^{2}$-norm $\|\bullet\|_{H_{\varepsilon},\omega_{\varepsilon,\delta}}$ defined by

[TABLE]

one can easily check that

[TABLE]

Indeed, the first inequality is obtained from property $(b)$, and the second inequality is obtained from property $(B)$ for $\omega_{\varepsilon,\delta}$ (for example see [FM16, Lemma 2,4]). In particular, we see that $u$ belongs to the $L^{2}$-space

[TABLE]

of $E$-valued $(n,q)$-forms on $Y_{\varepsilon}$ (not $X$) with respect to $\|\bullet\|_{\varepsilon,\delta}$. By the orthogonal decomposition (see for example [Mat16a, Proposition 5.8])

[TABLE]

the $E$-valued form $u$ can be decomposed as follows $:$

[TABLE]

Here $\overline{\partial}^{*}_{\varepsilon,\delta}$ is (the maximal extension of) the formal adjoint of the $\overline{\partial}$-operator and $\mathcal{H}^{n,q}_{\varepsilon,\delta}(E)$ is the space of harmonic forms on $Y_{\varepsilon}$, that is,

[TABLE]

Proposition 3.4 (resp. Proposition 3.5) can be proved by the same method as in [FM16, Proposition 5.4, 5.6, 5.7] (resp. [FM16, Proposition 5.9, 5.10]), so we omit the proofs here.

Proposition 3.4**.**

If we have

[TABLE]

for every relatively compact set $K\Subset X\setminus D$, then the cohomology class $\beta$ is zero in $H^{q}(X,K_{X}\otimes E\otimes\mathcal{I}(H))$. Here $\|\bullet\|_{K,h_{\varepsilon},\omega_{\varepsilon,\delta}}$ denotes the $L^{2}$-norm on $K$ with respect to $h_{\varepsilon}$ $($not H_{\varepsilon}$$) and $\omega_{\varepsilon,\delta}$.

Proposition 3.5**.**

There exists $v_{\varepsilon,\delta}\in L^{n,q-1}_{(2)}(E)_{h_{\varepsilon},\omega_{\varepsilon,\delta}}$ satisfying the following properties $:$

[TABLE]

Remark 3.6**.**

In general, we have

[TABLE]

and thus $v_{\varepsilon,\delta}$ may not be $L^{2}$-integrable with respect to $H_{\varepsilon}$. **

For the above solution $v_{\varepsilon,\delta}$ of the $\overline{\partial}$-equation, by using the density lemma, we can take a family of smooth $E$-valued forms $\{v_{\varepsilon,\delta,k}\}_{k=1}^{\infty}$ with the following properties $:$

[TABLE]

Now we consider the level set $X_{c}:=\{x\in X\,|\,-|s|_{b}^{2}<c\}\Subset X\setminus D$ for a negative number $c$. The set of the critical values of $|s|_{b}^{2}$ is of Lebesgue measure zero from Sard’s theorem. Hence, for a given relatively compact $K\Subset X\setminus D$, we can choose $-1\ll c<0$ such that

[TABLE]

Then, by [Mat16b, Proposition 2.5, Remark 2.6] (see also [FK, (1.3.2) Proposition]), we obtain

[TABLE]

for almost all $d\in{}]c-a,c+a[$, where $a$ is a sufficiently small positive number. Here $\overline{\partial}^{*}_{h_{\varepsilon},\omega_{\varepsilon,\delta}}$ is the formal adjoint of the $\overline{\partial}$-operator in $L^{n,\bullet}_{(2)}(E)_{h_{\varepsilon},\omega_{\varepsilon,\delta}}$ and ${(\!(}\bullet,\bullet{)\!)}_{\partial X_{d},h_{\varepsilon},\omega_{\varepsilon,\delta}}$ is the inner product on the boundary $\partial X_{d}$ defined by

[TABLE]

for smooth $E$-valued forms $a$, $b$, where $dS_{\varepsilon,\delta}$ denotes the volume form on $\partial X_{d}$ defined by dS_{\varepsilon,\delta}:=-*d|s|_{b}^{2}/\big{|}d|s|_{b}^{2}\big{|}_{h_{\varepsilon},\omega_{\varepsilon.\delta}} and $*$ denotes the Hodge star operator with respect to $\omega_{\varepsilon,\delta}$. Note that $dV_{\varepsilon,\delta}=dS_{\varepsilon,\delta}\wedge d|s|_{b}^{2}$. One can easily see that

[TABLE]

by $(\ref{a4})$, and thus it is sufficient to show that the right hand side of equality (33) converges to zero. For this purpose, we first prove the following proposition.

Proposition 3.7**.**

[TABLE]

Proof of Proposition 3.7.

By property (d) and property (B), we have

[TABLE]

Since $u_{\varepsilon,\delta}$ is harmonic with respect to $H_{\varepsilon}$ and $\omega_{\varepsilon,\delta}$, we have $\overline{\partial}^{*}_{\varepsilon,\delta}u_{\varepsilon,\delta}=0$ and $\overline{\partial}u_{\varepsilon,\delta}=0$. Further we have $\gamma=i\partial\overline{\partial}\psi$ on $X\setminus D$. Therefore we obtain

[TABLE]

from the twisted Bochner-Kodaira-Nakano identity (see [Ohs04, Lemma 2.1] or [Fuj13, Proposition 2.20. 2.21]). On the other hand, one can easily check that

[TABLE]

By the above arguments, we conclude that

[TABLE]

It follows that the left hand side converges to zero from $\|u\|_{H,\omega}\geq\|u_{\varepsilon,\delta}\|_{\varepsilon,\delta}$. Further the function $\chi^{\prime}(\sigma)/\eta(|s|_{b}^{2}+c)$ is bounded below since we have

[TABLE]

for some constant $C$. This completes the proof. ∎

Finally we prove the following proposition by using Proposition 3.7.

Proposition 3.8**.**

(i)* For a relatively compact set $K\Subset X\setminus D$, we have*

[TABLE]

(ii)* For almost all $d\in{}]c-a,c+a[$, we have*

[TABLE]

Proof of Proposition 3.8.

In general, we have the formula $D^{\prime}_{g}u=G^{-1}\partial(Gu)$ for a smooth hermitian metric $g$, where $G$ is a local function representing $g$. Let $G_{\varepsilon}$ be a local function representing $h_{\varepsilon}$. We remark that $G_{\varepsilon}e^{-\psi}$ is a local function representing $H_{\varepsilon}$. By the definition of $\overline{\partial}^{*}_{\varepsilon,\delta}=\overline{\partial}^{*}_{H_{\varepsilon},\omega_{\varepsilon,\delta}}$, we have

[TABLE]

and thus we obtain

[TABLE]

Now we have

[TABLE]

Since $\varlimsup_{\delta\to 0}\|v_{\varepsilon,\delta}\|_{K,h_{\varepsilon},\omega_{\varepsilon,\delta}}$ can be bounded by a constant that is independent of $\varepsilon$, it is sufficient to show that $\lim_{\varepsilon\to 0}\lim_{\delta\to 0}\|\overline{\partial}^{*}_{h_{\varepsilon},\omega_{\varepsilon,\delta}}u_{\varepsilon,\delta}\|_{K,h_{\varepsilon},\omega_{\varepsilon,\delta}}=0$. We have $e^{-\psi/2}=1/|s|_{b}<C_{K}$ on $K$ for some constant $C_{K}>0$, since $K$ is a relatively compact set in $X\setminus D$. Hence we see that

[TABLE]

We obtain the first statement (i) since the right hand side converges to zero by Proposition 3.7.

Now we prove statement (ii). By the Cauchy-Schwarz inequality, we have

[TABLE]

By Fubini’s theorem, we obtain

[TABLE]

By Fatou’s lemma, we see that

[TABLE]

Therefore the integrand of the left hand side is finite for almost all $d\in(c-a,c+a)$. On the other hand, by the same argument, we see that

[TABLE]

Therefore the integrand of the left hand side is zero for almost all $d\in(c-a,c+a)$. This completes the proof. ∎

Theorem 1.1 is now a consequence of Proposition 3.4, Proposition 3.8, and equation (33). ∎

Remark 3.9**.**

In the case of a holomorphically convex manifold, a proof based on injectivity theorems can be obtained by a slight modification of the above proof. The only problem is that an $E$-valued differential form $u$ representing a given cohomology class is not necessarily $L^{2}$-integrable but just locally $L^{2}$-integrable. Since $X$ admits a holomorphic map $\pi:X\to S$ to a Stein space $S$, the form $u$ is $L^{2}$-integrable with respect to the metric $he^{-\psi}e^{-\Phi}$ for a suitable psh exhaustion function $\Phi$ on $X$. Then it is not hard to check that our arguments still work by replacing $h$ with $he^{-\Phi}$.**

Remark 3.10**.**

It would be interesting to know whether the hypothesis that $\psi$ has analytic singularities is really needed. The main statement still makes sense when $\psi$ has arbitrary analytic singularities, and one may thus guess that the result can be extended by performing a further regularization of $\psi$.**