Frobenius semisimplicity for convolution morphisms (Q1650171)

The paper at hand concerns the action of the Frobenius automorphism on mixed $\ell$-adic complexes on varieties over a finite field $k$, and in particular a fiberwise criterion for the semisimplicity and Frobenius-semisimplicity of the direct image of such a complex under a proper morphism $f: X\to Y$ between varieties over $k$. \par Let $P$ be a mixed and simple perverse sheaf on $X$. After passage to an algebraic closure $\overline{k}$ of $k$, the decomposition theorem of Beilinson, Bernstein, Deligne and Gabber ensures that the direct image complex of the base change of $P$ decomposes as a direct sum of certain shifted intersection cohomology complexes. It is not known whether $Rf_*P$, the direct image for the morphism $f$ over $k$ is already semisimple. A related question is whether $Rf_*P$ is Frobenius semisimple in the sense that the Frobenius action on the stalks of the sheaves $R^i f_*P$ is semisimple. \par The authors conjecture that the direct image of the intersection complex on the domain of a proper morphism is always semisimple and Frobenius semisimple. This would imply a strong form of the decomposition theorem over finite fields. In [Int. Math. Res. Not. 2015, No. 24, 13106--13121 (2015; Zbl 1349.14091)], \textit{M. A. A. De Cataldo} has shown the validity of this conjecture for proper toric morphisms between toric varieties. \par In the paper at hand, the conjecture is proved for ``generalized convolution morphisms'' for partial affine flag varieties of split connected reductive groups over finite fields. (See also [\textit{A. Beilinson} et al., J. Am. Math. Soc. 9, No. 2, 473--527 (1996; Zbl 0864.17006)] and [\textit{P. N. Achar} and \textit{S. Riche}, Ann. Inst. Fourier 63, No. 4, 1511--1612 (2013; Zbl 1348.14041)], [\textit{R. Bezrukavnikov} and \textit{Z. Yun}, Represent. Theory 17, 1--98 (2013; Zbl 1326.20051)] for related results in the setting of finite-dimensional flag varieties and of affine flag varieties.) As one of the tools, the authors develop a new schematic theory of the big cell for loop groups.