Billiards and tilting characters for \(\mathrm{SL}_3\) (Q1743725)

Let \(G\) be a split simple and simply connected algebraic group over a field \(k\) of prime characteristic \(p\). A long-standing problem has been the determination of the characters of simple \(G\)-modules. For a dominant weight \(\lambda\), let \(\chi_{\lambda}\) denote the character of the simple module with highest weight \(\lambda\). In an effort to reformulate his earlier conjecture on these characters, in [Represent. Theory 19, 3--8 (2015; Zbl 1316.20049)], the first author introduced an infinite sequence of approximations (referred to as ``generations'') \(\chi_{\lambda}^{0}, \chi_{\lambda}^{1}, \chi_{\lambda}^2, \dots, \chi_{\lambda}^{\infty}\) to \(\chi_{\lambda}\). Here, \(\chi_{\lambda}^{0}\) is the character in characteristic zero and \(\chi_{\lambda}^{\infty} = \chi_{\lambda}\). One can inductively determine higher generations from lower generations via Kazhdan-Lusztig polynomials, and these approximations eventually stabilize (depending on the prime) for a given \(\lambda\). The focus of this paper is on a similar concept for characters of indecomposable tilting modules. The key task is to understand the coefficients that would allow one to move from one generation to the next. Here, the authors consider those coefficients relating the second generation characters back to the first generation. The problem is translated to one of identifying \(p\)-canonical basis elements for the associated anti-spherical module (obtained from the Iwahori-Hecke algebra), where one can also talk about generations for \(p\)-canonical basis elements. The authors discuss an algorithm for potentially computing \(p\)-canonical basis elements and computer computations that were done. The algorithm makes use of work of \textit{S. Riche} and the second author [Tilting modules and the \(p\)-canonical basis. Paris: Société Mathématique de France (SMF) (2018; Zbl 1437.20001)] and \textit{N. Libedinsky} and the second author [``The anti-spherical category'', Preprint, \url{arXiv:1702:00459}]. Specifically, the authors consider the case when \(G = \mathrm{SL}_3\) and the prime \(p\) is odd. They describe the construction of a multiset of labelled points in the weight lattice. This multiset is constructed via a dynamical process akin to billiards bouncing in the alcoves of the dominant chamber. Based on some computer calculations for small primes, the authors conjecture formulas, using this multiset, for certain \(p\)-canonical basis elements and second generation \(p\)-canonical basis elements. Still for \(\mathrm{SL}_3\), let \(\varpi\) be a fundamental dominant weight and consider the indecomposable tilting module \(T_{3pk\varpi}\) (of highest weight \(3pk\varpi\)) for a positive integer \(k\). The authors observe that, if their conjecture holds, the multiplicities of Weyl modules in \(T_{3pk\varpi}\) grow at least exponentially in \(k\). Given known connections with symmetric groups, one could also make new computations of decomposition numbers for symmetric groups. In particular, it would follow that decomposition numbers for the symmetric group \(S_n\) grow exponentially in \(n\).