Variations on a theme of Cline and Donkin. (Q1956357)

Let \(G\) be a reductive algebraic group over an algebraically closed field \(k\) of prime characteristic \(p\) with \(G\) being split over \(\mathbb F_p\). Let \(G_r\) denote the \(r\)-th Frobenius kernel of \(G\). A long-standing conjecture of Humphreys and Verma is that the projective indecomposable modules for \(G_r\) admit a compatible structure as \(G\)-modules. If \(p\) is sufficiently large (depending on \(G\)), this conjecture was confirmed by \textit{S. Donkin} [Math. Z. 212, No. 1, 39-60 (1993; Zbl 0798.20035)]. More recently, [in Adv. Math. 226, No. 3, 2065-2088 (2011; Zbl 1234.20056)], the authors proved a stable version of the conjecture with no condition on the prime: that a direct sum of copies of a PIM admits a \(G\)-structure. In this paper, the authors investigate this stability result in greater generality. Let \(Q\) be a finite dimensional rational \(G_r\)-module such that \(Q\) is a direct summand of a rational \(G\)-module \(M\) in such a way that the \(G_r\)-socle of \(Q\) is a \(G\)-submodule of \(M\). Then it is shown that some direct sum of copies of \(Q\) admits a compatible \(G\)-module structure. More generally, let \(G\) denote a connected affine algebraic group scheme over an algebraically closed field \(k\) (of arbitrary characteristic) and \(N\) be a closed, normal \(k\)-subgroup scheme of \(G\) such that \(G/N\) is reductive. Further, let \(Q\) be a finite dimensional rational \(N\)-module and \(V\) be a rational \(G\)-submodule of \(Q\) containing the \(N\)-socle of \(Q\). For example, in the above setting, \(N\) can be taken as \(G_r\) and \(V\) can be taken as the \(G_r\)-socle of \(Q\). The authors introduce three notions of \(G\)-stability for such a pair \((Q,V)\): \textit{strong}, \textit{tensor}, and \textit{numerical}. Numerical \(G\)-stability is precisely the desired condition that a direct sum of copies of \(Q\) admits a compatible \(G\)-structure. Tensor \(G\)-stability requires the existence of a \(G/N\)-module \(Y\) such that \(Q\otimes Y\) admits a compatible \(G\)-structure (with \(V\otimes Y\) being a \(G\)-submodule). It is fairly readily shown that tensor and numerical stability are equivalent. Strong \(G\)-stability requires the existence of a \(k\)-morphism \(\alpha\colon G\to GL_k(Q)\) such that the composition \(\alpha\circ\text{mult}\colon G\times N\to G\to GL_k(Q)\) is equal to the composition \(\text{mult}\circ(1\times\rho)\circ(\alpha\times 1_N)\colon G\times N\to GL_k(Q)\times N\to GL_k(Q)\times GL_k(Q)\to GL_k(Q)\), where \(\rho\colon N\to GL_k(Q)\) is the map defining the \(N\)-module structure of \(Q\). The main theorem of the paper is that all three notions of \(G\)-stability are in fact equivalent. In the special case of prime characteristic with \(N=G_r\) and \(V\) being the \(G_r\)-socle of \(Q\), the claimed numerical stability then follows by showing that the pair \((Q,V)\) is in fact strongly \(G\)-stable. The key lemma giving this strong stability uses ideas going back to much earlier work of \textit{E. Cline} [J. Algebra 22, 350-364 (1972; Zbl 0282.16020)] and \textit{S. Donkin} [Math. Scand. 51, 142-150 (1982; Zbl 0499.20026)]. A fundamental concept used in this work is the notion of a Schreier system on a pair of groups \(G\) and \(U\) involving extensions of \(G\) by \(U\) and the construction of a group structure on the set \(U\times G\). In the context of the main theorem, the group \(G\) will be as above and the group \(U\) will be a certain unipotent subgroup of \(GL_k(Q)\) determined by the submodule \(V\). If one has a strongly \(G\)-stable pair \((Q,V)\), the authors also show that the graded module associated to \(Q\) arising from the socle series relative to \(N\) has a compatible rational \(G\)-module structure. Further, the authors also obtain two cohomological obstructions to extending the \(G\)-structure on \(V\) to \(Q\), and they give examples to show that one can have a strongly \(G\)-stable pair \((Q,V)\) for which \(Q\) does not admit a compatible \(G\)-structure.