Perfect stratifications and weight theory (Q2366852)

Let \(k\) be an algebraically closed field of any characteristic, and let \(X\) be a \(k\)-scheme, i.e. a separated scheme of finite type over \(k\). Denote by \(H^ \bullet(X)\) (and \(H^ \bullet_ c(X))\) the \(\ell\)-adic étale cohomology (with compact support) of \(X\). Thus if \(k\) has characteristic zero and is a subfield of \(\mathbb{C}\) one may also take rational singular cohomology. \(H^ \bullet(X)\) and \(H^ \bullet_ c(X)\) carry a weight filtration \(W_ \bullet\), and one defines the (pure) Euler characteristics of \(X\) by \(\chi_ m(X)=\sum_ i(-1)^ i\dim Gr^ W_ mH^ i(X)\), and similarly \(\chi^ m_ c(X)\) for \(Gr^ W_ mH^ i_ c(X)\). The (pure) Poincaré polynomial of \(X\) is defined as \(P_{\text{pur}}(X,t)=\sum_ m\chi_ m(X)t^ m\), and similarly \(P^ c_{\text{pur}}(X,t)=\sum_ m\chi^ m_ c(X)t^ m\). For a \(k\)-scheme \(X\) one can define a stratification or, a little more generally, a (filterable) decomposition. Such a stratification (or filterable decomposition) \(X=\bigcup X_ \alpha\) is called perfect if \(X\) and also the cells \(X_ \alpha\) are smooth and connected and one has the following relation between the Betti numbers of \(X\) and the \(X_ \alpha:b_ i(X)=\sum_ \alpha b_{i-2d_ \alpha}(X_ \alpha)\), where \(d_ \alpha\) is the codimension of \(X_ \alpha\) in \(X\). As a first result it is shown that for a smooth, connected, complete \(k\)-scheme \(X\) with Białynicki-Birula decomposition \(X=\bigcup X_ \alpha\) and with an action of an algebraic torus \(T\), one has \(b_ i(X)-\sum_ \alpha b_{i-2d_ \alpha}(X^ T_ \alpha)\) for all \(i\), where \(X^ T=\bigcup X^ T_ \alpha\) is the decomposition of the \(T\)-fixed part \(X^ T\) of \(X\) into connected components. Next it is shown that one can define, for a linear \(k\)-algebraic group acting on a \(k\)-scheme \(X\), pure Euler characteristics for the equivariant cohomology \[ H^ i_ G(X): \chi^ m_ G(X)=\sum_ i(- 1)^ i\dim Gr^ W_ mH^ i_ G(X). \] Similarly for the equivariant cohomology with compact support \(H^ i_{G,c}(X)\) one defines \(\chi^ m_{G,c}(X)\). Also, the pure equivariant Poincaré series of \(X\) is defined: \(P^ G_{\text{pur}}(X,t)=\sum_ m\chi^ m_ G(X)t^ m\). One can define (filterable) \(G\)-decompositions and \(G\)-stratifications of the \(G\)-scheme \(X\) in a straightforward way. The final result says that for a smooth connected \(G\)-scheme \(X\) with \(G\)-decomposition \(X=\bigcup X_ \alpha\) such that the \(X_ \alpha\) are smooth, equidimensional of codimension \(d_ \alpha\) in \(X\), one has \(\chi^ m_ G(X)=\sum_ \alpha\chi_ G^{m-2d_ \alpha}(X_ \alpha)\) and \(P^ G_{\text{pur}}(X,T)=\sum_ \alpha t^{2d_ \alpha}P^ G_{\text{pur}}(X_ \alpha,t)\). This is applied to obtain inductively the Betti numbers of a symplectic or a geometric quotient of a variety. This leads to formulas obtained before by \textit{F. Kirwan}. A second application concerns the calculation of the Betti numbers of geometric quotients as done by Białynicki-Birula and Sommese. The last section of the paper gives a motivic version of the preceding results. A motivic Euler characteristic \(\chi_{\text{mot}}(X)\) is defined by means of the system of mixed realizations \(SRM^ \bullet(X)\) defined by \(H^ \bullet(X)\). Similarly for compact support. In the situation of the Białynick-Birula decomposition of \(X\) with torus action as sketched above, one finds a relation for the Hodge numbers of \(X\) and the \(X^ T_ \alpha:h^{p,q}(X)=\sum_ \alpha h^{p-d_ \alpha,q-d_ \alpha}(X^ T_ \alpha)\). Also, for the \(G\)-scheme \(X\) with \(G\)-decomposition \(X=\bigcup X_ \alpha\) one finds for the motivic Poincaré series, \(P^ G_{\text{mot,pur}}(X,t)=\sum_ \alpha t^{2d_ \alpha}P^ G_{\text{mot,pur}}(X_ \alpha,t)(-d_ \alpha)\), where \((-d_ \alpha)\) means Tate twist. One deduces a formula for Hodge numbers.