Extensions for finite Chevalley groups. III: Rational and generic cohomology. (Q2251894)

Let \(G\) be a simple algebraic group defined and split over the prime field \(\mathbb F_p\) with \(p>0\). Choose the usual split Borel subgroup \(B\) and maximal torus \(T\). Let \(M\) be a \(G\)-module. We have the discrete subgroups \(G(\mathbb F_q)\) and the Frobenius kernels \(G_r\). The authors have developed a method to compare the cohomology groups \(H^n(G,M)\), \(H^n(G(\mathbb F_q),M)\), \(H^n(G_r,M)\) [for part I see Adv. Math. 183, No. 2, 380-408 (2004; Zbl 1061.20041); for part II see Trans. Am. Math. Soc. 354, No. 11, 4421-4454 (2002; Zbl 1050.20031)]. They start this paper with extensive weight calculations giving information on possible weights of \(H^m(B_{s+f},p^s\mu+\lambda)\), where \(\lambda\), \(\mu\) are weights. This then serves to study vanishing behavior of \(H^n(B,p^s\mu+\lambda)\). And then their method allows them to revisit earlier work of \textit{H. L. Hiller} [J. Pure Appl. Algebra 16, 259-263 (1980; Zbl 0425.20043)] and of \textit{E. Cline, B. Parshall, L. Scott} and \textit{W. van der Kallen} [Invent. Math. 39, 143-163 (1977; Zbl 0336.20036)]. They get new proofs for the existence of generic cohomology and improved bounds for when it is reached, at least for the simply connected \(G\) that they consider.