On the supercuspidal representations of \(GL_ 4\). I (Q1084522)

This is the first of two papers whose purpose is to show that if F is a p-adic field, then all supercuspidal representations of \(G=GL_ 4(F)\) are induced from representations on open, compact (mod the center) subgroups. When p, the residual characteristic of F, is odd, the result is known from theorems of \textit{R. E. Howe} [Pac. J. Math. 73, 437-460 (1977; Zbl 0404.22019)] and \textit{A. Moy} [Am. J. Math. 108, 863-929 (1986; Zbl 0597.12019)]. When \(p=2\), however, the situation is much more complicated. The general idea is this: let K be one of the maximal compact open (mod center) subgroups of G. Associated with K are congruence subgroups \(K^ r\). Suppose that \(\pi\) is irreducible admissible then \(\pi| K\) decomposes into finite-dimensional representations. Let \(\sigma\) be one such. Then for some r, \(\sigma\) is trivial on \(K^{r+1}\) and not on \(K^ r\). When \(r\geq 1\), \(K^ r/K^{r+1}\) is abelian, and \(\sigma | K^ r\) is a sum of characters \(\chi\). The authors show that it is possible to choose K and r so that \(\chi\) takes on one of a rather small number of forms. For some \(\chi\), called very cuspidal representations, any irreducible summand of \(Ind(K^ r\to K,\chi)\) induces to a supercuspidal of G; in this case, \(\pi\) is necessarily supercuspidal. For some other \(\chi\), called principal, no supercuspidal contains a copy of \(\chi\) ; in this case, \(\pi\) is not supercuspical. There is a third class of K, called alfalfa, for which further analysis is necessary. In this paper, the authors show that if \(\pi\) is supercuspidal then it is induced from some K except possibly when \(\chi\) is a certain kind of alfalfa representation (called here a ''wild alfalfa representation''). Their analysis also yields a formula for the \(\epsilon\)-factors of supercuspidal representation when combined with results of \textit{C. F. Bushnell} and \textit{A. Fröhlich} [Proc. Lond. Math. Soc., III. Ser. 50, 207-264 (1985; Zbl 0558.12007)].