Smooth Hilbert schemes: their classification and geometry (Q2112367)

Hilbert schemes \(\mathrm{Hilb}^{p(t)} (\mathbb P^m)\) parametrizing closed subschemes \(X \subset \mathbb P^m\) with Hilbert polynomial \(p(t)\) have received much attention from algebraic geometers since their construction in the early 1960s. Although connected [\textit{R. Hartshorne}, Publ. Math., Inst. Hautes Étud. Sci. 29, 5--48 (1966; Zbl 0171.41502)], \(\mathrm{Hilb}^{p(t)} (\mathbb P^m)\) exhibits many bad behaviors, for example it can have non-reduced components [\textit{D. Mumford}, Am. J. Math. 84, 642--648 (1962; Zbl 0114.13106)] and every singularity type appears in some Hilbert scheme [\textit{R. Vakil}, Invent. Math. 164, No. 3, 569--590 (2006; Zbl 1095.14006)]. In the paper under review, the authors classify all smooth Hilbert schemes. Recall that a Hilbert polynomial of a closed subscheme \(X \subset \mathbb P^m\) can be uniquely written in the form \(p(t) = \sum_{i=1}^r \binom{t+\lambda_i-i}{\lambda_i-1}\) where \(\lambda = (\lambda_1, \dots, \lambda_r)\) is a partition of integers satisfying \(\lambda_1 \geq \dots \geq \lambda_r \geq 1\) [\textit{G. Gotzmann}, Math. Z. 158, 61--70 (1978; Zbl 0352.13009)]. The authors prove that the partitions corresponding to a smooth Hilbert scheme are precisely those in the following list: (1) \(m = 2 \geq \lambda_1\). (2) \(m \geq \lambda_1\) and \(\lambda_r \geq 2\). (3) \(\lambda = (1)\) or \(\lambda = (m^{r-2}, \Lambda_{r-1},1)\), where \(r \geq 2\) and \(m \geq \lambda_{r-1} \geq 1\). (4) \(\lambda = (m^{r-s-3}, \lambda_{r-s-s}^{s+2},1)\), where \(r-3 \geq s \geq 0\) and \(m-1 \geq \lambda_{r-s-2} \geq 3\). (5) \(\lambda = (m^{r-s-5}, 2^{s+4},1)\), where \(r-5 \geq s \geq 0\). (6) \(\lambda = (m^{r-3},1^3)\), where \(r \geq 3\). (7) \(\lambda = (m+1)\) or \(r=0\). Moreover, the authors describe the schemes parametrized by each family listed. For example, the general member of family (3) is a union of a hypersurface of degree \(r-2\), a linear subspace of dimension \(\lambda_{r-1}\) and a point while the general member of family (5) is a union of a hypersurface of degree \(r-s-3\), a hypersurface of degree \(s+2\) of a linear subspace of dimension \(\lambda_{r-s-2}\) and a point. The families in (7) correspond to the one-point Hilbert scheme parametrizing \(\mathbb P^m\) and the empty scheme. All families in the theorem were previously known to be smooth: smoothness for families (1) and (6) follows from work of \textit{J. Fogarty} [Am. J. Math. 90, 511--521 (1968; Zbl 0176.18401)], families (2) and (3) were shown smooth by \textit{A. P. Staal} [Math. Z. 296, No. 3--4, 1593--1611 (2020; Zbl 1451.14010)] and families (4) and (5) were shown smooth by \textit{R. Ramkumar} [J. Algebra 617, 17--47 (2023; Zbl 1503.13007)] in his work on Hilbert schemes with at most two Borel-fixed ideals. The main contribution here is that this list is complete. As to the strategy of the proof, it is known from work of \textit{A. Reeves} and \textit{M. Stillman} that the lexicographical point is always smooth on the Hilbert scheme and determines a unique irreducible component of \(\mathrm{Hilb}^{p(t)} (\mathbb P^m)\) of computable dimension [J. Algebr. Geom. 6, No. 2, 235--246 (1997; Zbl 0924.14004)]. In theory one could attempt to show smoothness by computing the dimension of the Zariski tangent space at the other Borel-fixed points, but this is unwieldy. Instead, the authors construct families of subschemes corresponding to points on Hilbert schemes that necessarily singular. To describe these points, define a \textit{residual inclusion} \(X \subset Y \subset \mathbb P^m\) to be a closed immersion such that there is a linear subspace \(\Lambda \subset \mathbb P^m\) containing \(X\) and a hypersurface \(D \subset \Lambda\) with \(Y\) the residual scheme of \(D \subset X\) in \(\Lambda\). A \textit{residual flag} is a flag \(\emptyset \subset X_e \subset X_{e-1} \subset \dots \subset X_1\) of residual inclusions. For most Hilbert polynomials not in the list, the authors produce such an \(X_1\) near the lexicographic point which corresponds to a singular point on the Hilbert scheme. The others not on the list are handled with three other singular families. The proof is valid over \(\mathrm{Spec}\,\mathbb Z\). The authors credit \texttt{Macaulay2} [\url{http://www.math.uiuc.edu/Macaulay2/}] for many experimental computations that were indispensable in discovering their results.