Variations on the Baer-Suzuki theorem. (Q2339677)

The Baer-Suzuki theorem says that if \(p\) is a prime, \(x\) is a \(p\)-element in a finite group \(G\) and \(\langle x,x^g\rangle\) is a \(p\)-group for all \(g\in G\), then the normal closure of \(x\) in \(G\) is a \(p\)-group. The authors consider the case where \(x^g\) is replaced by \(y^g\) for some other \(p\)-element \(y\). In a previous joint paper with \textit{P. H. Tiep} [Adv. Math. 234, 618-652 (2013; Zbl 1277.20029)] they provided examples showing that the analog of Baer-Suzuki is not true; the most one could expect is that \(x^G\) and \(y^G\) commute modulo \(O_p(G)\) and this can fail. However they proved some variation: if \(C\) and \(D\) are normal subsets of a finite group \(G\) with \(\langle C\rangle=\langle D\rangle\) and \(\langle c,d\rangle\) is a \(p\)-group for a prime \(p\geq 5\) and for all \((c,d)\in C\times D\), then \(\langle C\rangle\) is a normal \(p\)-subgroup of \(G\). The conclusion fails for \(p=2,3\) and fails without the assumption that \(\langle C\rangle=\langle D\rangle\). In the present paper, the author prove that under a stronger assumption on the structure of \(\langle c,d\rangle\), a new version of Baer-Suzuki holds: let \(C\) and \(D\) be normal subsets of a finite group \(G\) such that \(\langle c,d\rangle\) is a \(p\)-group with no section isomorphic to \(Z_p\wr Z_p\) for all \((c,d)\in C\times D\), then \([C,D]\leq O_p(G)\). They also classify all pairs of conjugacy classes \(C,D\) of elements of order \(p\) in finite almost simple groups such that \(\langle c,d\rangle\) is a \(p\)-group for all \((c,d)\in C\times D\). The authors also answer a closely related question of Pavel Shumyatsky on commutators of conjugacy classes of \(p\)-elements and propose the following conjecture: let \(5\neq p\) be a prime and let \(C\) be a conjugacy class of \(p\)-elements in the finite group \(G\); if \([c,d]\) is a \(p\)-element for all \(c,d\in C\), then \(C\subseteq O_p(G)\). They notice that their methods can be used to reduce to showing the conjecture for \(G={^2G_2(q^2)}\) with \(q^2=3^{2a+1}>27\) and \(p\) a primitive prime divisor of \(q^2+\sqrt 3q+1\).