Modular intersection cohomology complexes on flag varieties (Q1925746)

The authors study the intersection cohomology of Schubert varieties. Let \(\mathsf{k}\) be a field of characteristic \(p\). Let \(G\) be a connected reductive algebraic group over \(\mathbb{C}\), let \(B\subset G\) be a Borel subgroup of \(G\) and let \((W, S)\) be the corresponding Weyl group and its simple reflections. Let \(G/B\) be the flag variety with its classical topology and consider the intersection cohomology sheaves \(\mathbf{IC}(w)\) which belong to the bounded derived category of sheaves of \(\mathsf{k}\)-vector spaces on \(G/B\) constructible along \(B\)-orbits. Let \(\mathcal{H}\) be the Hecke algebra of \((W, S)\) over \(\mathbb{Z}[v, v^{-1}]\) and let \(\{H_w | w\in W\}\) be the Kazhdan-Lusztig basis of \(\mathcal{H}\). If \(\mathsf{k}\) is of characteristic zero, a theorem of Kazhdan and Lusztig says that \(\mathrm{ch}(\mathbf{IC}(w)) = H_w\), where \(\mathrm{ch}(\mathbf{IC}(w))\) is the character sheaf of \(\mathbf{IC}(w)\). Thus the Poincaré polynomials of the stalks of the intersection cohomology sheaves are given by Kazhdan-Lusztig polynomials. It then follows that the same is true in almost all characteristics, however for any given characteristic almost nothing is known. It has been known since the original papers of Kazhdan and Lusztig that in non-simply laced cases the intersection cohomology complexes may have a different character in characteristic 2. In 2002 Braden discovered examples of Schubert varieties in simply laced types \(A_7\) and \(D_4\) where the character of the intersection cohomology sheaf in characteristic 2 is different to all other characteristics. The authors define a new basis \(\{B_w | w\in W\}\) using the information encoded in the \( W\)-graph and say that \(x\in W\) is separated if \(H_x=B_x\). The first main result states that \(ch(\mathbf{IC}(w)) = H_w\) for any field \(\mathsf{k}\) if \(w\) is separated. The determination of the characters of \(\mathbf{IC}(w)\) is closely related to the decomposition theorem. Given a simple reflection \(s\in S\) let \(P_s\) be the corresponding standard minimal parabolic subgroup and consider the associated quotient map \(\pi_s\) over \(G/P_s\). If \(\mathsf{k}\) is of characteristic zero, the decomposition theorem implies that \(\pi_{s*}(\mathbf{IC}(w))\) is a direct sum of shifts of intersection cohomology sheaves. The second main result states some conditions, using the previous definition of separateness, so that the same holds in positive characteristic. Unfortunately, the calculation of the sets \(\sigma(W)\) of separated elements can be very complicated. The authors use Fokko du Cloux's program Coxeter to prove that \(\sigma(W)=W\) if \(G\) is of type \(A_n\) for \(n<7\). By results of Soergel, this implies a part of Lusztig's conjecture for \(SL(n)\) with \(n<8\). The authors say also that their techniques are not as affective for the other groups. Finally, in the appendix Braden shows that, both in type \(D_4\) and \(A_7\), \(ch(\mathbf{IC}(w)) \neq H_w\) if \(w\) is a minimal element in \(W \;\sigma(W)\) and \(\mathsf{k}\) has characteristic 2. The main theoretical tool in the demonstration of the main results is the existence and uniqueness of parity sheaves.