Vanishing theorems for linearly obstructed divisors (Q517072)

The authors study divisors on the blow-up of \(\mathbb{P}^n\) at points in general position that are non-special with respect to the notion of linear speciality. They describe the cohomology groups of their strict transforms via the blow-up of the space along their linear base locus and extend the result to non-effective divisors that sit in a small region outside the effective cone. As an application, they describe linear systems of divisors in \(\mathbb{P}^n\) blown-up at points in star configuration and their strict transforms via the blow-up of the linear base locus. More in detail, denote by \(\mathcal{L}=\mathcal{L}_{n,d}(m_1,\dots, m_s)\) the linear system of hypersurfaces of degree \(d\) in \(\mathbb{P}^n\) passing through a collection of \(s\) points in general position with multiplicities at least \(m_1,\dots, m_s\geq 0\) respectively. The virtual dimension of \(\mathcal{L}\) is denoted by \[ \mathrm{vdim}(\mathcal{L})={n+d \choose n}-\sum_{i=1}^s{n+m_i-1 \choose n} \] and its expected dimension as \[ \mathrm{edim}(\mathcal{L})=\max\{\mathrm{vdim}(\mathcal{L}),0\}. \] If \(D\) is the strict transform of a general divisor in \(\mathcal{L}\) in the blow-up \(X\) of \(\mathbb{P}^n\) at the \(s\) points, \[ D:=dH-\sum_{i=1}^sm_iE_i \tag{1.1} \] then \(\mathrm{vdim}(D):=\mathrm{vdim}(\mathcal{L})\) equals \(\chi(X,\mathcal{O}_X(D))\), the Euler characteristic of the sheaf on \(X\) associated with \(D\), while \(\dim(\mathcal{L})\) is the number of global section of \(\mathcal{O}_X(D)\), namely the dimension of the space \(H^0(X,\mathcal{O}_X(D))\). We call \(\mathcal{L}\) (or \(D\)) nonspecial if its dimension equals the expected dimension, and special if \(\dim(\mathcal{L})>\mathrm{edim}(\mathcal{L})\). The speciality of \(\mathcal{L}\) (\(D\)) is defined to be the difference \(\dim(\mathcal{L}) -\mathrm{edim}(\mathcal{L})\). In the last century the problem of computing the dimension of linear systems was studied with different techniques by many people. In the planar case, the Segre-Harbourne-Gimigliano-Hirschowitz conjectures predicts all special linear systems, while, in the case of \(\mathbb{P}^3\), there is an analogous conjectural classification of special linear systems formulated by Laface and Ugaglia. Due to its complexity and mysterious geometry, the simple question of predicting and computing dimensions of such vector spaces is not even conjectured when \(n\) is four or higher. In the case of \(\mathbb{P}^n\) general results are rare and few things are known. For multiplicities higher than 2, the only general result known so far is a complete cohomological classification of the speciality of only linearly obstructed effective divisors, proved by \textit{M. C. Brambilla} et al. [Trans. Am. Math. Soc. 367, No. 8, 5447--5473 (2015; Zbl 1331.14007)]. One of the goal of this paper is to extend such a classification to the non-effective case. In order to classify the special divisors, one has to understand first what are the obstructions, that is what are the varieties that whenever contained with multiplicity in the base locus of a given divisor force \(\mathcal{L}\) to be special. The authors, in the direction of extending the Segre-Harbourne-Gimigliano-Hirschowitz Conjecture to \(\mathbb{P}^n\) and possibly to other rational projective varieties, pose the following Question 1.1. Consider any non-empty linear system \(\mathcal{L}\) in \(\mathbb{P}^n\). Let \(\tilde{X}\) be the smooth composition of blow-ups of \(\mathbb{P}^n\) along the (strict transforms of the) cycles of the base locus of \(\mathcal{L}\), ordered in increasing dimension. We denote by \(D\) a general divisor of the linear system \(\mathcal{L}\), and by \(\tilde{\mathcal{D}}\) the strict transform of \(D\) in \(\tilde{X}\). Does \(h^i(\tilde{X},\mathcal{O}_{\tilde X}(\tilde{\mathcal{D}}))\) vanish for all \(i\geq 1\)? An affirmative answer to this question implies that \(\dim(\mathcal{L}) = \chi(\tilde{X},\mathcal{O}_{\tilde X}(\tilde{\mathcal{D}}))\), translating the classical dimensionality problem for linear systems into a Riemann-Roch formula for divisors living in subsequently blown-up spaces. Denote now by \(\tilde{D}\) the strict transform of \(D\) in \(\tilde{X}\), the blow-up of \(X\) along the linear cycles of the base locus of \(D\). Clearly \(\tilde{D}\) is different from \(\tilde{\mathcal{D}}\) of Question 1.1 since the second one denotes the strict transform of \(D\) in \(\tilde{X}\), the blow-up of \(X\) along all -- linear and non linear -- cycles of the base locus of \(D\). In this paper the authors use the concept of linear exprcted dimension instead of the one of expected dimension. Definition 1.2. Given a linear system \(\mathcal{L}=\mathcal{L}_{n,d}(m_1,\dots, m_s)\), for any integer \(-1 \leq r \leq s - 1\) and for any multi-index \(I(r) = \{i_1,\dots,i_{r+1}\}\subseteq \{1,\dots, s\}\), define the integer \[ k_{I(r)}:=\max(m_{i_1}+\cdots +m_{i_{r+1}}-rd,0). \] The linear virtual dimension of \(\mathcal{L}\) (or of \(D\)), denoted by \(\mathrm{ldim}(\mathcal{L})\), is the number \[ \sum_{r=-1}^{s-1}\sum_{I(r)\subseteq\{1,\dots, s\}} (-1)^{r+1} \binom{n+k_{I(r)}-r-1}{n}, \] where we set \(I(-1)=\emptyset\). The linear expected dimension of \(\mathcal{L}\) is defined as follows: it is 0 if \(\mathcal{L}\prec_s\mathcal{L}'\) and \(\mathrm{ldim}(\mathcal{L}') \leq 0\), otherwise it is the maximum between \(\mathrm{ldim}(\mathcal{L})\) and \(0\). Here we write \(\mathcal{L}\prec_s\mathcal{L}'\), for two linear systems \(\mathcal{L}_{n,d}(m_1,\dots,m_s)\) and \(\mathcal{L}_{n,d}(m'_1,\dots,m'_s)\) with the same degree, if \(m_i \geq m'_i\) for all \(i \in \{1,\dots,s\}\). It is important to notice that this notion is well-defined not only for all effective divisors but also for non-effective ones. In this light, asking whether the dimension of a given linear system equals its linear expected dimension can be thought of as a refinement of the classical question of asking whether the dimension equals the expected dimension. If the answer to this question is affirmative, then \(\mathcal{L}\) is said to be a only linearly obstructed. Obviously, non-special linear systems are always only linearly obstructed. There exist linear systems that are linearly obstructed without being only linearly obstructed. For instance \(\mathcal{L}_{4,10}(6^7)\) contains all lines \(L_{ij}\), \(i,j \in \{1, \dots, 7\}\) with multiplicity two in its base locus as well as the rational normal curve through the seven points. In [loc. cit.] it is proved that Theorem 1.3. All non-empty linear systems of the form \(\mathcal{L}=\mathcal{L}_{n,d}(m_1,\dots, m_s)\) with \(s \leq n+2\) base points are only linearly obstructed. Moreover, if \(s \geq n + 3\) and \[ \sum_{i=1}^s m_i \leq nd+\min(n-s(d),s-n-2), \quad 1\leq m_i \leq d, \] where \(s(d) \geq 0\) is the number of points of multiplicity \(d\), then \(\mathcal{L}\) is non-empty and only linearly obstructed. For every effective divisor \(D\), let \(D_{(r)}\) denote the strict transform of \(D\) in the space \(X^n_{(r)}\) obtained as a sequence of blow-ups of \(\mathbb{P}^n\) along the linear base locus of \(D\) up to dimension \(r\), with \(r \leq n - 1\): \[ D_{(r)}:=D- \sum_{\rho=1}^{r}\sum_{I(\rho)\subseteq\{1,\dots, s\}} k_{I(\rho)}E_{I(\rho)}. \] where \(E_{I(\rho)}\) denotes the (strict transform of the) exceptional divisor of the linear subspace of \(\mathbb{P}^n\) of dimension \(\rho\) spanned by the points parametrised by the multi-index \(I(\rho)\) and \(k_{I(\rho)}\) is the multiplicity with which the aforementioned subspace is contained in the base locus. Set \(\tilde{D}:=D_{(\overline{r})}\), where \(\overline{r}\) is the maximum dimension of the linear base locus. In [loc. cit.] it is also proved that Theorem 1.4. Given integers \(d\), \(m_1, \dots,m_s\), if \(s \leq n + 2\) and \(D\) is effective, the following statements hold. \((a)\) \(h^0(D)=\mathrm{ldim}(D)\) and \(h^i(\tilde{D})=0\) for every \(i\geq 1\). \((b)\) For any \(0\leq r\leq n-1\), \(h^i(D_{(r)})=0\) for every \(i\geq 1\) and \(i\not=r+1\), while \[ h^{r+1}(D_{(r)})=\sum_{\rho=r+1}^{s-1}\sum_{I(r)\subseteq\{1,\dots, s\}} (-1)^{\rho -r-1} \binom{n+k_{I(\rho)}-\rho-1}{n}. \] One of the goal of this paper is to show that the same type of results as in Theorem 1.4 holds for larger classes of divisors, such as effective divisors with arbitrary number of general base points and non-effective divisors. The authors also extend the formula in Theorem 1.4 to the case of any effective divisor, not necessarily only linearly obstructed, interpolating an arbitrary collection of general multiple points. Theorem 1.5. Given integers \(d\), \(m_1, \dots,m_s\), consider the divisor \(D\) in (1.1). if \(D\) is effective, then for any \(0\leq r\leq n-1\) we have \[ h^{r+1}(D_{(r)})=\sum_{\rho=r+1}^{s-1}\sum_{I(r)\subseteq\{1,\dots, s\}} (-1)^{\rho -r-1} \binom{n+k_{I(\rho)}-\rho-1}{n}+\sum_{\rho=r+1}^n(-1)^{\rho -r-1}h^\rho(\tilde{D}). \] In particular, \[ h^0(D)=\mathrm{ldim}(D)+\sum_{\rho=r+1}^n(-1)^{\rho -r-1}h^\rho(\tilde{D}). \] Moreover, if \(h^i(\tilde{D})=0\), for all \(i\geq1\), then \(h^i(D_{(r)})=0\) for all \(i\not= r+1\). The geometric interpretation is that for any effective divisor \(D\), every \(\rho\)-dimensional linear cycle \(L_{I(\rho)}\), for which \(k_{I(\rho)}\geq1\) and \(\rho\geq r+1\), gives a contribution with sign, \((-1)^{r-\rho+1}\), equal to \[ \binom{n+k_{I(\rho)}-\rho-1}{n} \] to \(h^{r+1}(D_{(r)})\) and to the formula for \(\mathrm{ldim}(D)\). Moreover, such a contribution is zero when \(k_{I(\rho)}\leq \rho\). The main result of this paper is a complete cohomological description of \(D_{(r)}\) in the following cases, where \(b:=b(D)=\sum_{i=1}^sm_i-nd\). Theorem 1.6. Fix \(d\), \(m_1, \dots,m_s\). Statements \((a)\) and \((b)\) of Theorem 1.4 hold for all divisors \(D\) of the form (1.1) with \(m_i \leq d+1\) under the following hypothesis: \(s\leq n+1\) and \(b\leq n\), or \(s=n+2\) and \(b\leq 1\), or \(s\geq n+3\) and \(b\leq \min(n-s(d),s-n-2)\). Moreover, if \(s\leq n+1\) then \(h^i(\tilde{D})=0\), for all \(i\leq n-1\), and \(h^n(\tilde{D})=\binom{b-1}{n}\) for \(b\geq n+1\) and zero otherwise.