Supports for linear degenerations of flag varieties (Q2066971)

The supports of a projective map are a collection of irreducible closed subvarieties of the target space, whose local intersection cohomology control the variation of cohomology of the fibers of the map. The authors study supports of a flat family called linear degenerations of flag varieties. Rather than degenerating flag varieties by relaxation of the containment relation between the subspaces restricted by the flag, they consider new degenerations, which share favorable geometric properties. The determination of the supports of this flat family is through the observation, featuring varieties and maps arising in Lusztig's geometric realization of quantized enveloping algebras as a special case. This allowed the authors to reduce the determination of supports to an algebraic problem, expanding a certain monomial in Chevalley generators of a quantized enveloping algebra into elements of the canonical basis. Letting \(f:X\rightarrow Y\) be a projective morphism of complex algebraic varieties, where \(X\) is irreducible and smooth, let \(X_y:=f^{-1}(y)\) be the family of fibers of \(y\). One is interested in the cohomology \(\text{H}^*(X_y;\mathbb{Q})\) of the fiber as \(y\) varies along \(Y\). Since the cohomology of a fiber may be identified with a stalk of its cohomology sheaves \(\text{H}^*(X_y;\mathbb{Q})\simeq \mathcal{H}_y^*(\text{R}f_*(\mathbb{Q}_X))\), where \(\text{R}f_*(\mathbb{Q}_X)\in \mathcal{D}^b(Y)\), \(\text{R}f_*(\mathbb{Q}_X)\) is isomorphic to a direct sum \(\text{R}f_*(\mathbb{Q}_X)\simeq \bigoplus_{i=1}^n \text{IC}(\overline{S_i},\mathcal{L}_i)[s_i]\) of shifts of intersection cohomology complexes by the Decomposition Theorem by Beilinson-Bernstein-Deligne, where \(\{ (S_i,\mathcal{L}_i,s_i)\}\) is a finite set of data consisting of smooth, locally closed subvariety \(S_i\subseteq Y\), a nonzero local system \(\mathcal{L}_i\) on \(S_i\), and an integer \(s_i\). The set \(\{ \overline{S_i}: 1\leq i\leq n\}\) is called the set of supports of \(f\). One then concludes that the cohomology of the fiber \(X_y\) is essentially controlled by the local intersection cohomology of the supports \(\overline{S_i}\), i.e., there exists a decomposition \(\text{H}^*(X_y;\mathbb{Q}) \simeq \bigoplus_{i=1}^n \mathcal{H}_y^*(\text{IC}(\overline{S_i},\mathcal{L}_i))\). The \(\text{GL}_{n+1}(\mathbb{C})\)-flag variety description is given in Section 2.2 on page 1983, with the set of Motzkin paths explained therein. One of the authors' main results is written in terms of Motzkin paths, showing the reader that the combinatorics appearing via Motzkin is rather astounding, i.e., the set of supports is highly nontrivial yet completely controllable (Theorem 2.3, page 1984): the intersection complex \(\text{IC}(\overline{\mathcal{O}(\mathbf{r})})\) appears, up to a shift, as a direct summand of \(\text{R}\pi_*\mathbb{Q}_{\mathcal{F}_U}\) if and only if \(\mathbf{r}\) is of the form \(\mathbf{r}(\mathbf{x})\) for a Motzkin path \(\mathbf{x}\in\mathcal{M}_n\). This also shows that the set of supports is asymptotically very small, compared to the set of all potential supports.