A Castelnuovo-Mumford regularity bound for threefolds with rational singularities (Q2131767)

Let \(X\subset \mathbb P^r\) be a non-degenerate complex projective variety of degree \(d\) and codimension \(e\); then, \(X\) is said to be \textit{\(m\)-regular in the sense of Castelnuovo-Mumford} if \(H^i(\mathbb P^r, \mathcal I {X|\mathbb P^r} (m- i)) = 0\) for all \(i > 0\). It is equivalent to say that \(X\) is \((m-1)\)-normal and \(\mathcal O_X\) is \((m-1\)-regular, i.e., the natural restriction map \(H^0(\mathbb P^r,\mathcal O_{\mathbb P^r} (m - 1)) \to H^0(X,\mathcal O_X (m - 1))\) is surjective and \(H^i(X,\mathcal O_X (m - 1 - i)) = 0\) for \(i > 0\). The minimal such number \(m\) is called the \textit{Castelnuovo-Mumford regularity} of \(X\) and is denoted by reg\((X)\). One of the fundamental problems in projective algebraic geometry is to find a bound for reg\((X)\) in terms of algebraic and geometric data such as the degree \(d\) and the codimension \(e\). A sharp bound was suggested by \textit{D. Eisenbud} and \textit{S. Goto} [J. Algebra 88, 89--133 (1984; Zbl 0531.13015)] (cf. [\textit{L. Gruson} et al., Invent. Math. 72, 491--506 (1983; Zbl 0565.14014)]) in the regularity conjecture, which says that reg\((X) \le d - e + 1\). There is also a folklore conjecture which predicts that the maximal degree, denoted by max deg\((X)\), of minimal generators of the defining ideal \(\mathcal I_{X|\mathbb P^r}\) of \(X\subset \mathbb P^r\) is at most \(d\). Since max deg\((X)\le\) reg\((X)\), the regularity conjecture implies this folklore conjecture. The regularity conjecture has only been proved for a few cases up to dimension \(3\), but it turns out that the conjecture is not true: \textit{J. McCullough} and \textit{I. Peeva} [J. Am. Math. Soc. 31, No. 2, 473--496 (2018; Zbl 1390.13043)] constructed counterexamples to the regularity conjecture, although their examples are highly singular. It is then a natural problem to make a distinction between mildly singular varieties satisfying the regularity conjecture and highly singular varieties not satisfying the regularity conjecture. The threefold case is particularly interesting because one of the counterexamples to the regularity conjecture is a projective threefold \(X\) in \(\mathbb P^5\) such that \(\dim X_{\mathrm{sing}}=1\), \(\deg(X) = 31\), and \(\max\deg(X)=\mathrm{reg}(X)=38\) (Example 4.7 of [loc. cit.]) while the regularity conjecture holds for a smooth projective threefold in \(\mathbb P^5\) by \textit{S.-J. Kwak} [J. Reine Angew. Math. 509, 21--34 (1999; Zbl 0978.14042)]. Notice also that the singular threefold \(X\) in \(\mathbb P^5\) does not even fulfill the folklore conjecture aforementioned. The first main result of this paper (Theorem 1.1) is to establish the following regularity bound for threefolds with rational singularities: Let \(X \subset \mathbb P^r\) be a non-degenerate normal projective threefold of degree \(d\) and codimension \(e\). Suppose that \(X\) has rational singularities. Then we have reg\((X)\leq d - e + 2\). Observe that bound is only off by \(1\) from the bound in the regularity conjecture, and in particular, the maximal degree can be bounded by the degree. The codimension two case is specially important as one of the counterexamples to regularity conjecture by McCullough-Peeva appeared in this case. The second main result of this paper (Theorem 1.2) is to establish a sharp regularity bound for codimension two normal projective threefolds with Cohen-Macaulay Du Bois singularities and to classify the extremal cases. To prove these results, the author consider a general projection \(\pi\colon X \dashrightarrow \overline{X}\subset \mathbb P^4\) and analyze the complexity of the fibers of \(\pi\). Following ideas of Castelnuovo and Lazarsfeld, it can be shown known that if one has a smaller Castelnuovo-Mumford regularity bound for all fibers of \(\pi\), then one would get a better Castelnuovo-Mumford regularity bound for \(X\). The main novelty of this paper is to overcome the difficulty by computing the dimension of the secant variety swept out by secant lines passing through the singular locus of \(X\), when this has positive dimension (see Theorem 1.3). This implies that the support of the fiber \(X_y:=\pi^{-1}(y)\) has at most two points (Proposition 4.2), and this result leads the authors to the desired control on the regularity of the fiber \(X_y\).