\(L ^{2}\)-\(\overline{\partial}\)-cohomology groups of some singular complex spaces (Q2377346)

The topic of the present, significant paper is \(L^2\)-theory for the \(\overline\partial\)-operator on complex spaces with isloated singularities. It contains a complete description of the \(L^2\)-\(\overline\partial\)-cohomology for \((0,q)\)-forms at isolated singularities in terms of a resolution of singularities. This leads particularly to the solution of a long-standing conjecture of Pardon and Stern (Corollary 1.5). For a very long time, the \(\overline\partial\)-theory was essentially restricted to (smooth) complex manifolds. Efforts to study the \(\overline\partial\)-operator also on singular complex spaces started around 1990. Of particular importance are two papers of \textit{W. L. Pardon} [Topology 28, No. 2, 171--195 (1989; Zbl 0682.32024)] and \textit{W. L. Pardon} and \textit{M. Stern} [J. Am. Math. Soc. 4, No. 3, 603--621 (1991; Zbl 0751.14011)] who studied the \(L^2\)-theory of the \(\overline\partial\)-operator on projective varieties. It turned out that it might be possible to create a profound \(L^2\)-\(\overline\partial\)-theory, whereas regularity of the \(\overline\partial\)-operator in other function spaces seems to be a tricky topic and not so promising. The \(L^2\)-theory is pretty interesting as it has tight connections to the \(L^2\)-theory for the DeRham-operator \(d\) and intersection (co-)homology. However, in the 90s the development got stuck at a certain point and a central conjecture of Pardon and Stern remained unsolved. The topic was then revitalized by K. Diederich , J. E. Fornæss, N. Øvrelid and S. Vassiliadou in a series of papers [\textit{K. Diederich} et al., Math. Scand. 92, No. 2, 269--294 (2003; Zbl 1033.32005); \textit{J. E. Fornæss} et al. Int. J. Math. 16, No. 4, 387--418 (2005; Zbl 1081.32024); \textit{N. Øvrelid} and {S. Vassiliadou}, Am. J. Math. 131, No. 1, 129--151 (2009; Zbl 1172.32013)], just to name the most important. They achieved already the following see [Zbl 1081.32024]: Let \(X\) be an analytic set in \(\mathbb{C}^n\) of pure dimension \(n\) with an isolated singularity at the origin, \(X_r=X\cap B_r(0)\) the intersection with a small ball (so that \(X_r\) contains no other singularities) and \(X_r^*=X_r \setminus\{0\}\). Let \(X_r^*\) carry the Hermitian metric induced from the metric of the ambient space, \(\mathbb{C}^n\). Then: \[ H^{p,q}_{(2)}(X_r^*) = 0 \;\;\;\text{ for }\;\;p+q>n \] and \(H^{p,q}_{(2)}(X_r^*)\) is finite-dimensional for \(p+q<n\) and \(q>0\) (see also [\textit{W. Pardon} and \textit{M. Stern}, J. Reine Angew. Math. 533, 55--80 (2001; Zbl 0960.14009)]). Here, \(H^{p,q}_{(2)}\) stands for the \(L^2\)-cohomology of the \(\overline\partial\)-operator in the sense of distributions. The next question was then to see whether it is possible to determine the \(L^{p,q}_{(2)}\)-cohomlogy explicitly, maybe in terms of a resolution of the singularity (as in [Zbl 0751.14011)]). First steps in that direction have been made in [\textit{J. Ruppenthal}, Int. J. Math. 20, No. 4, 459--489 (2009; Zbl 1180.32015); Math. Z. 263, No. 2, 447--472 (2009; Zbl 1187.32007)]. Whereas the case \(p>0\) is still open, the present paper gives a complete answer to the question for \((0,q)\)-forms, \(q>0\) (at isloated singularities). Let \(\pi: X_r' \rightarrow X_r\) be a resolution of singularities. Then pushforward of forms under \(\pi\) induces an isomorphism \[ \pi_*: H^q(X_r',\mathcal{O}) \cong H^{0,q}_{(2)}(X_r') \longrightarrow H^{0,q}_{(2)}(X_r^*) \] for \(1\leq q \leq n-2\) (Theorem 1.1). On the other hand, it is not hard to see that \(H^{0,n}_{(2)}(X_r^*)=0\). It remains to study the most interesting but also most difficult cohomology group \(H^{0,n-1}_{(2)}(X_r^*)\). Assume from now on that \(\dim X=n\geq 2\). Following ideas from [Zbl 0682.32024], [Zbl 0751.14011)] and [\textit{J. Ruppenthal}, ``L2-theory for the dbar-operator on compact complex spaces'', \url{arXiv:1004.0396}], the authors obtain the following: Let \(\pi: X_r' \rightarrow X_r^*\) be a resolution of singularities with only normal crossings. Then there exists an effective divisor \(D\) with support on the exceptional set such that \(\pi\) induces an epimorphism \[ \pi_*: H^{0,n-1}\big(X_r', \mathcal{O}(D)\big) \rightarrow H^{0,n-1}_{(2)}(X_r^*) \] whose kernel is naturally isomorphic to \(H^{n-1}_E(X_r',\mathcal{O}(D))\) (Theorem 1.2). Here \(H^{n-1}_E(X_r',\mathcal{O}(D))\) is the cohomology with support on the exceptional set \(E\) of the resolution. In case \(n=\dim X=2\), on can take \(D=E-|E|\) and one has \(H^1_E(X_r',\mathcal{O}(E-|E|))=0\), making \(\pi_*\) an isomorphism (this solves the conjecture of Pardon and Stern, see Corollary 1.5). The claim that \(D=E-|E|\) is also eligible for \(n>2\) is still open because there was a gap in the argument given by the reviewer in [arXiv:1004.0396] from where some statements are cited by the authors. This seems to be an intersting open question. However, already the case \(n=2\) solves the conjecture of Pardon-Stern in the affirmative. In the rest of the paper, the authors use their local theory to prove global analogous statements: Let \(X\) be a projective variety with only isolated singularities, \(X^*\) the regular set, carrying the restriction of the Fubini-Study-metric, and \(\pi: X'\rightarrow X\) a suitable resolution of singularities. Then the statements above hold with \(X^*\) and \(X'\) in place of \(X_r^*\) and \(X_r'\), respectively. Let us conclude with a few words on the techniques used in the paper. There are two essential ingredients. First, the results from [Zbl 1081.32024], which are achieved by putting a complete metric on \(X_r^*\) (i.e., moving the singularity to infinity). This leads to Theorem 1.1. Second, for Theorem 1.2, some new ideas are needed in addition. Besides the \(\overline\partial\)-operator in the sense of distributions (i.e., the maximal closed extension), one has to study other closed \(L^2\)-extensions of the \(\overline\partial\)-operator, particularly the minimal closed extension, and to explain the relation between the minimal and the maximal operator (see also [\url{arXiv:1004.0396}]).