Counting real characters after Gallagher (Q6114004)

Let \(G\) be a finite group, \(N\) a normal subgroup and \(\theta\) an irreducible (complex) character of \(N\). \textit{P. X. Gallagher} [Proc. Sympos. Pure Math. 21, 51--52 (1971; Zbl 0257.20007)] gave a characterisation of the number of irreducible characters of \(G\) lying over \(\theta\) (i.e., that contain \(\theta\) as a constituent when restricted to \(N\)) in terms of so called \(\theta\)-good conjugacy classes. Let \(G_{\theta}=\{g\in G\mid\theta^{g}=\theta\}\) denote the stabiliser of \(\theta\) in \(G\) (also called the inertia group of \(\theta\)). Thus \((G_{\theta},N,\theta)\) is a character triple. For any \(x\in G_{\theta}\) it is known that \(\theta\) extends to a character \(\theta_{x}\) (necessarily irreducible) of \(N\langle x\rangle\) (because the quotient \(N\langle x\rangle/N\) is cyclic). The conjugation action of \(G\) on itself induces an action of the centraliser \(C_{G}(xN)\) on \(N\langle x\rangle\). A coset \(xN\) (or the element \(x\)) is said to be \textit{good} for \(\theta\) (or \textit{\(\theta\)-good}) if \(\theta_{x}\) is fixed by every element of \(C_{G}(xN)\). This property does not depend on the choice of extension \(\theta_{x}\) of \(\theta\). Moreover, \(xN\) is good for \(\theta\) if and only if all of its \(G_{\theta}/N\)-conjugates are good for \(\theta\), so we have a notion of \(\theta\)-good conjugacy classes of \(G_{\theta}/N\). A theorem of Gallagher then says that the number of irreducible characters of \(G\) lying over \(\theta\) is equal to the number of \(\theta\)-good conjugacy classes of \(G_{\theta}/N\). The goal of the paper under review is to prove a modified version of Gallagher's theorem, counting only real-valued characters. To this end, the author considers the ``extended inertial group'' \[ G_{\theta}^{*}:=\{g\in G\mid\theta^{g}=\theta\text{ or }\overline{\theta}\}, \] where \(\overline{\theta}\) is the dual character and, for any \(\theta\)-good \(xN\in G_{\theta}/N\), an ``indicator'' \(\sigma(xN)\in\{0,\pm1\}\) as follows. Suppose first that there is a \(g\in G_{\theta}^{*}\) such that \((xN,\theta)^{g}=(x^{-1}N,\overline{\theta})\) (simultaneous conjugation by \(g\)). Then set \(\sigma(xN)=+1\) if \(\theta_{x}^{g}=\overline{\theta}_{x}\) and \(\sigma(xN)=-1\) otherwise. If no such \(g\) exists (for instance when \(xN\) is not real in \(G_{\theta}^{*}/N\), that is, not conjugate to its inverse), then set \(\sigma(xN)=0\). Then \(\sigma(xN)=\sigma(yN)\) whenever \(yN\) is \(G_{\theta}/N\)-conjugate to \(xN\). We thus have a notion of \(\theta\)-good conjugacy classes of \(G_{\theta}/N\) as being of type \(\sigma_{+},\sigma_{-}\) or \(\sigma_{0}\). The first main result of the paper under review is: Theorem. The number of real irreducible characters of \(G\) lying over \(\theta\) is equal to the number of \(\theta\)-good conjugacy classes of \(G_{\theta}/N\) of type \(\sigma_{+}\) minus the number of conjugacy classes of type \(\sigma_{-}\). The Frobenius-Schur indicator of \(\theta\) is \[ \epsilon(\theta):=\frac{1}{|N|}\sum_{g\in G}\theta(g^{2}), \] so that \(\epsilon(\theta)=0,+1\) or \(-1\) as \(\theta\) is not real-valued, the character of a real representation or real-valued but not the character of a real representation, respectively. For involutions in \(G/N\), the author considers a related indicator (the Gow-Frobenius-Schur indicator) which also has values \(\pm1\) or \(0\) and which thereby gives rise to the notion of an involution in \(G/N\) being of type \(+1,-1\) or \(0\). The second main result of the paper is then Theorem. Let \(\chi\) be the character of \(G\) induced by \(\theta\). Then \(\epsilon(\chi)\) is equal to the number of involutions in \(G_{\theta}^{*}/N\) of type \(+1\) minus the number of involutions of type \(-1\).