The banks of the cohomology river (Q1950280)

Let \(\mathcal F\) be a coherent sheaf on the projective \(n\)-space \({\mathbb P}^n\) over a field \(\mathbb K\). Let \(\Gamma_\ast(\mathcal F)\) denote the graded \(S := {\mathbb K}[X_0, \dots ,X_n]\)-module \(\bigoplus_j\Gamma({\mathbb P}^n, {\mathcal F}(j))\). As a consequence of the results of \textit{I. N. Bernstein, I. M. Gelfand} and \textit{S. I. Gelfand} [Funkts. Anal. Prilozh. 12, No. 3, 66--67 (1978; Zbl 0402.14005)], there exists a bounded complex \({\mathcal M}^\bullet\) of sheaves on \({\mathbb P}^n\) of the form: \[ \cdots \longrightarrow {\mathcal O}_{\mathbb P}(-i)^{b_{-i}} \longrightarrow \cdots \longrightarrow {\mathcal O}_{\mathbb P}^{b_0} \longrightarrow \cdots \longrightarrow {\mathcal O}_{\mathbb P}(i)^{b_i} \longrightarrow \cdots \] such that \(H^0({\mathcal M}^\bullet) \simeq {\mathcal F}\) and \(H^i({\mathcal M}^\bullet) = 0\) for \(i \neq 0\). Moreover, \({\mathcal M}^\bullet\) is unique if one assumes that it has no direct summand of the form \(\text{T}^{-i}{\mathcal K}^\bullet(i)\), where \({\mathcal K}^\bullet\) is the tautological Koszul complex on \({\mathbb P}^n\) and ``T'' is the translation functor. Let us call this minimal \({\mathcal M}^\bullet\) the \textit{BGG linear monad} of \(\mathcal F\). \textit{D. Eisenbud, G. Fløystad} and \textit{F.-O. Schreyer} [Trans. Am. Math. Soc. 355, No. 11, 4397--4426 (2003; Zbl 1063.14021)] found a method of getting \({\mathcal M}^\bullet\) from certain parts of the cohomology modules of \(\mathcal F\) by computing syzygies over exterior algebras. It follows, however, directly from the BGG correspondence that: \[ \begin{gathered} H^i(\Gamma_\ast({\mathcal M}^\bullet)) \simeq {\bigoplus_{j\geq -i}} \text{H}^i({\mathcal F}(j)) \;\;\text{and}\\ H^{n-i}(\Gamma_\ast({\mathcal M}^{\bullet \ast})) \simeq {\bigoplus_{j\geq -n+i}}\text{Ext}^{n-i}_{{\mathcal O}_{\mathbb P}} (\mathcal F,{\mathcal O}_{\mathbb P}(j)) \simeq \left({\bigoplus_{k< -i}}\text{H}^i({\mathcal F}(k))\right)^\ast(n+1) \, . \end{gathered} \] Let \(p\) be a nonnegative integer. One deduces immediately that \(b_i = 0\) for \(i > p\) iff \(\text{H}^i({\mathcal F}(-i)) = 0\) for \(i > p\). Moreover, if these conditions are fulfilled, then \(b_p = \text{h}^p({\mathcal F}(-p))\), \(\text{H}^i({\mathcal F}(1 - i)) = 0\) for \(i > p\), and the multiplication map: \[ \text{H}^p({\mathcal F}(-p)) \otimes \text{H}^0({\mathcal O}_{\mathbb P}(1)) \longrightarrow \text{H}^p({\mathcal F}(1 - p)) \] is surjective. For \(p = 0\), this is the Castelnuovo-Mumford lemma. Similarly, \(b_{-i} = 0\) for \(i > p\) iff \(\text{H}^j({\mathcal F}(-1-j)) = 0\) for \(j < n-p\) and, in this case, \(b_{-p} = \text{h}^{n-p}({\mathcal F}(-1-n+p))\). Based on this kind of observations, the authors of the paper under review introduce, for \(0 \leq p < n\), the following \textit{(co)regularity indices}, generalizing the Castelnuovo-Mumford regularity: \[ \begin{gathered} \text{reg}^p{\mathcal F} := \inf\{m\in {\mathbb Z}\mid \text{H}^i({\mathcal F}(m-i)) = 0 \;\text{for} \;i>p\}\\ \text{coreg}^p{\mathcal F} := \sup\{m \in {\mathbb Z}\mid \text{H}^j({\mathcal F}(m-j)) = 0 \;\text{for} \;j < n-p\}\, . \end{gathered} \] Notice that \(\text{(co)reg}^p{\mathcal F}(m) = \text{(co)reg}^p{\mathcal F} - m\) and that if \(\mathcal F\) is locally free then, by Serre duality, \(\text{coreg}^p{\mathcal F} = - \text{reg}^p{\mathcal F}^\ast - 1\). The first result of the paper under review asserts that if \(\mathcal F\) and \(\mathcal G\) are coherent sheaves on \({\mathbb P}^n\) with \(\dim \text{Supp}\, {\mathcal T}or_1({\mathcal F}, {\mathcal G}) \leq p + 2\) then: \[ \text{reg}^p({\mathcal F}\otimes {\mathcal G}) \leq \min{k+l = p} (\text{reg}^k{\mathcal F} + \text{reg}^l{\mathcal G})\, . \] Moreover, if \(\dim \text{Supp}\, {\mathcal T}or_1({\mathcal F}, {\mathcal G}) \leq 1\) then: \[ \text{coreg}^p({\mathcal F}\otimes {\mathcal G}) \geq 1 + \underset{k+l = p}{\text{max}} (\text{coreg}^k{\mathcal F} + \text{coreg}^l{\mathcal G})\, . \] The first assertion generalizes a result of G. Caviglia [Proc. Am. Math. Soc. 135, No. 7, 1949--1957 (2007; Zbl 1183.13024)] who considered the case \(p = 0\). The proof is similar to that of Caviglia's, with a modification consisting of the replacement of the linear free resolution of a convenient twist of \(\mathcal F\) by its BGG linear monad. Then the authors of the paper under review show that the above inequalities are sharp in a strong sense. More precisely, let \(Q \simeq \text{T}_{\mathbb P}(-1)\) be the universal quotient bundle on \({\mathbb P}^n\) and let \(\lambda = (\lambda_{n-1} \geq \cdots \geq \lambda_0 \geq 0)\) be a partition with \(n\) parts. Applying the Schur functor \(S_\lambda\) to \(Q\) one gets a homogeneous vector bundle \(S_\lambda Q\) on \({\mathbb P}^n\) and, in characteristic 0, Bott's theorem implies that \(\text{reg}^pS_\lambda Q = - \lambda_p\). As a byproduct of their proof of the Boij-Söderberg conjectures, \textit{D. Eisenbud} and \textit{F.-O. Schreyer} [J. Am. Math. Soc. 22, No. 3, 859--888 (2009; Zbl 1213.13032)] showed that, in characteristic 0, if \(\text{reg}^0{\mathcal F} \leq 0\) then there exists a chain of partitions \(\lambda^0 \leq \cdots \leq \lambda^s\) such that an integer multiple of the cohomology table of \(\mathcal F\) equals the cohomology table of \(\bigoplus_{j=0}^sS_{\lambda^j}Q\). Now, the second result of the paper under review asserts that if \({\mathcal F} = \bigoplus_{j=0}^sS_{\lambda^j}Q\) and \({\mathcal G} = \bigoplus_{j=0}^tS_{\mu^j}Q\) then: \[ \text{reg}^p({\mathcal F}\otimes {\mathcal G}) = \min_{k+l = p} (\text{reg}^k{\mathcal F} + \text{reg}^l{\mathcal G})\, ,\;\;0\leq p < n\, . \] The proof of this fact is a matter of representation theory. As an application of their first result, the authors of the paper under review construct families of unobstructed vector bundles on \({\mathbb P}^n\) with some special kind of cohomology tables. This application uses the following observation: let \({\mathcal B}^\bullet\) be the \textit{Beilinson monad} of \(\mathcal F\) (cf. \textit{A. A. Beilinson} [Funkts. Anal. Prilozh. 12, No. 3, 68--69 (1978; Zbl 0402.14006)]). It has terms: \[ {\mathcal B}^m = \begin{cases} \bigoplus_{i=m}^n\text{H}^i({\mathcal F}(m-i))\otimes \Omega^{i-m}(i-m),& m\geq 0\\ \bigoplus_{j=0}^{n+m}\text{H}^j({\mathcal F}(m-j))\otimes \Omega^{j-m}(j-m),& m<0\, . \end{cases} \] One sees that if \(\text{reg}^p{\mathcal F} \leq p + 1\) then \({\mathcal B}^m = 0\) for \(m > p\), and if \(\text{coreg}^p{\mathcal F} \geq -p - 1\) then \({\mathcal B}^m = 0\) for \(m < -p\).