Hochschild cohomology of generalised Grassmannians (Q6177366)

Due to their nature, partial flag varieties are among the most common guiding examples for studying algebro-geometric invariants of projective varieties, as their geometry is often ruled out by representation-theoretic properties. The invariant on which this work focuses on is the Hochschild cohomology of a variety \(X\), namely \(HH^i(X):=\bigoplus_{p+q=i}H^q(X,\bigwedge^pT_X)\) (cf. eq. 1.1) for \(T_X\) its tangent bundle. As varieties, the authors consider the subclass of partial flag varieties defined by a simple algebraic group (over an algebraically closed field in zero-characteristic) and a maximal parabolic subgroups, namely the generalised Grassmannians. The whole paper is motivated by the question of which partial flag variety \(G/P\) is Hochschild global, in the sense \(H^q(G/P,\bigwedge^pT_{G/P})=0\) for any \(q\geq 1\) (cf. eq. 1.2). The expectation among experts was that such a property held for any partial flag variety, but the authors prove that this is not the case. More precisely, first they prove that both (co)minuscule varieties, (co)adjoint generalised Grassmannians and \(A\)-type adjoint flag varieties are Hochschild global and compute their Hochschild cohomology as \(\mathfrak{g}\)-representations (Theorems A-B-C). Next they show that not all generalised Grassmannians satisfy the Hochschild property, by exhibiting an example with the (\(C\)-type) symplectic Grassmannian \(SGr(3,2n)\) (Proposition D). The above \(C\)-type example also allows the authors to prove that the Bott vanishing (cf. Definition 49) fails for \(SGr(3,2n)\) (Corollary E), giving the first example in the literature of the failure of Bott vanishing for generalised Grassmannians which are not cominuscule. The authors concludes the work by conjecturing that Theorem A (partially) is an ''if and only if'', that is any generalised Grassmannian which is neither (co)minuscule or (co)adjoint is not Hochschild global (Conjecture F). All topics are well introduced, with a very clear crash course in representation theory for making the paper accessible to any algebraic geometer.