A GIT construction of degenerations of Hilbert schemes of points (Q2415882)

A degeneration $X\rightarrow C$ gives a degeneration of the corresponding Hilbert schemes, and the study of such degenerations is of obvious interest in algebraic geometry. \par New techniques for studying degenerations were introduced by \textit{J. Li} and \textit{B. Wu} [Commun. Anal. Geom. 23, No. 4, 841--921 (2015; Zbl 1349.14014)], with an approach based on the technique of expanded degenerations. The method is very general and can be used to study degenerations of various types of moduli problems such as Hilbert schemes and moduli spaces of sheaves. They also used degenerations of Quot-schemes and coherent systems to obtain degeneration formulae for Donaldson-Thomas invariants and Pandharipande-Thomas stable pairs. \par In the present article, the authors search to understand degenerations of irreducible holomorphic symplectic manifolds. As a start, the authors study degenerations of $K3$ surfaces and their Hilbert schemes, a guiding example being type II degenerations of $K3$ surfaces leading to the investigation of degenerations of the Hilbert scheme of points for simple degenerations $X\rightarrow C$ with no a priori restriction on the type or dimension of the fibre. The assumption of a simple degeneration gives that the total space is smooth, and that the central fibre $X_0$ over the point $0\in C$ of the $1$-dimensional base $C$ has normal crossing along smooth varieties. In this paper, a technique for the construction of degenerations of Hilbert schemes is developed, which allow to control the geometry of the degenerate fibres. \par The authors describe the equalities and differences between their approach and that of Li and Wu [loc. cit.]: First of all, this article considers only Hilbert schemes of points, whereas Li and Wu consider general Hilbert schemes of ideal sheaves with arbitrary Hilbert polynomial, and also Quot schemes. Both approaches apply Li's method of expanded degenerations $X[n]\rightarrow C[n]$ which in the case of constant Hilbert polynomial means the construction of a family whose special fibre over $0$ parametrises length $n$ subschemes of the degenerate fibre $X_0.$ The main problem in this setting is to describe the subschemes whose support meets the singular locus of $X_0,$ and the idea of the construction is that whenever a subscheme approaches a singularity in $X_0,$ a new ruled component is inserted into $X_0$ and it will be sufficient to work with subschemes supported on the smooth loci of the fibres of $X[n]\rightarrow C[n].$ The dimension of the base $C[n]$ is increased at each step of increasing $n,$ and finally one has to take equivalence classes of subschemes supported on the fibres of $X[n]\rightarrow C[n].$ Additionally, the construction of expanded degenerations also includes an action of an $n$-dimensional torus $G[n]\subset\text{SL}(n)$ acting on $X[n]\rightarrow C[n]$ such that $C[n]/\!\!/ G[n]=C.$ \par Li and Wu proceed by constructing the stack $\mathfrak{X}/\mathfrak{C}$ of expanded degenerations associated to $X\rightarrow C,$ giving a notion of equivalence. For fixed Hilbert polynomial $P,$ a definition of stable ideal sheaf is given, and this is used to define a stack $I^P_{\mathfrak{X/C}}$ over $C$ parametrizing those (the Li-Wu stack). In the case where the constant Hilbert polynomial $P=n,$ this gives subschemes of length $n$ supported on the smooth locus of a fibre of an expanded degeneration, having finite automorphism group. In the present article, the method is not to use the Li-Wu stack, but to use GIT with respect to the action of $G.$ \par The body of the article is the construction of a setup allowing to apply GIT-methods. One has to assume that the dual graph $\Gamma(X_0)$ associated to the singular fibre $X_0$ is bipartite, i.e. that it has no cycles of odd length. One can always perform a quadratic base change so that this holds, so the assumption is rather mild. At first, the authors construct a relatively ample line bundle $\mathcal L$ on $X[n]\rightarrow C[n].$ Then the bipartite assumption allows to construct a $G[n]$-linearisation on $\mathcal L$ which is proven to be applicable to Hilbert schemes. The choice of the correct $G[n]$- linearisation is the most important technical tool of the present article. Using $\mathcal L$, the authors construct an ample line bundle $\mathcal M_l$ on the relative Hilbert scheme defined by $\mathbf{H}^n:=\text{Hilb}^n(X[n]/C[n]),$ with a natural $G[n]$-linearisation. In this explicit situation, GIT stability can be analysed using a relative version of the Hilbert-Mumford numerical criterion: It is proved that (semi-)stability of a point $[Z]\subset\mathbf{H}^n$ only depends on the degree $n$-cycle associated to $Z.$ \par Fixing the $G[n]$-linearised sheaf $\mathcal L,$ the construction depends on several choices: The orientation of the dual graph $\Gamma(X_0)$ has two possible choices as it is bipartite, and both give isomorphic GIT quotients. A suitable $l$ is chosen in the construction of $\mathcal M_l,$ but the characterisation of stable $n$-cycles shows that the final result is independent of this. \par Let $Z\subset X[n]_q$ for some point $q\in C[n].$ Using a local étale coordinate $t$ it is obtained coordinates $t_1,\dots,t_{n+1}$ on $C[n]$ and then $\{a_1,\dots,a_r\}$ is defined to be the subset indexing coordinates with $t_i(q)=0.$ Put $a_0=1,\;a_{r+1}=n+1,$ then $\mathbf{a}=(a_0,\dots,a_{r+1})\in\mathbb Z^{r+2}$ determines a vector $\mathbf{v_a}\in\mathbb Z^{r+1}$ with $i$-th component $a_i-a_{i-1}.$ Now, by definition, $Z$ has smooth support if it is supported in the smooth part of the fibre $X[n]_q.$ Then each point $P_i$ in the support of $Z$ is contained in a unique component of $X[n]_q$ with multiplicity say $n_i.$ This leads to the definition of the numerical support $\mathbf{v}(Z)\in\mathbb Z^{r+1}.$ \par The first main result then describes the stable locus in $\mathbf{H}^n$ with respect to $\mathcal M_l$ when $l\gg 2n^2:$ If $[Z]\in\mathbf{H}^n$ has smooth support, then $[Z]\in\mathbf{H}^n(\mathcal M_l)^{\text{ss}}$ if and only if $\mathbf{v}(Z)=\mathbf{v_a}$ (then also $[Z]\in\mathbf{H}^n_{\text{GIT}}:=\mathbf{H}^n(\mathcal M_l)^s$), if $[Z]\in\mathbf{H}^n$ does not have smooth support, then $[Z]\notin\mathbf{H}^n(\mathcal M_l)^{\text{ss}}.$ \par Given this result, the authors define the main object of study in the paper, the GIT quotient $I_{X/C}=\mathbf{H}_{\text{GIT}}^n/G[n].$ The advantage is that the GIT stable points can be controlled very explicitly so that the geometry of the fibres of the degenerate Hilbert-schemes can be controlled in detail. On the other side, the authors define the stack quotient $\mathcal I_{X/C}=[\mathbf{H}^n_{\text{GIT}}/G[n]].$ The next main result gives the relation between the two concepts, saying that the GIT quotient $I^n_{C/X}$ is projective over $C,$ and that the stack $\mathcal I^n_{X/C}$ is a Deligne-Mumford stack, proper and of finite type over $C$ with $I_{X/C}$ as coarse moduli space. In addition, the morphism $f:\mathcal I^n_{X/C}\rightarrow T^n_{\mathfrak{X/C}}$ is an isomorphism of Deligne-Mumford stacks. \par The above results prove that the GIT approach and the Li-Wu construction of degenerations of Hilbert schemes of points are equivalent. One advantage are the tools to explicitly describe the degenerate Hilbert schemes, and this is thoroughly illustrated with an example of degree $n$ Hilbert schemes on two components. \par As mentioned, a main objective of the article is to construct good degenerations of Hilbert schemes of $K3$ surfaces. It turns out that the technique with relative Hilbert schemes cannot be applied directly, but the situation is analysed based on the results in the paper, and a sketch of further work is given. \par This is a very deep and good article, showing the force of GIT theory compared to stack-theory. This proves that techniques of classical algebraic geometry are appropriate for solving algebraic geometric problems, and gives the theory for further studies (in particular of degenerations of Hilbert schemes of $K3$ surfaces). In addition, the article is very well written, and can be used as a guide for authors in the field.