Comprehensive classification of the algebra generated by two idempotent matrices (Q6650596)

Let \(P\) and \(Q\) be two idempotent \(n\times n\) complex matrices and consider the unital algebra \(\mathcal{A}:=\) \textrm{alg}\((I,P,Q)\) which they generate. The authors give a complete description of all such algebras. The proof begins by observing that \((P-Q)^{2}\) always lies in the center of the algebra, and therefore the eigenspaces of \((P-Q)^{2}\) define a direct decomposition of \(\mathcal{A}\). This reduces the classification problem to the case for which \((P-Q)^{2}=\lambda I\) for some \(\lambda \in \mathbb{C}\) and each of the cases where \(\lambda =0,\) \(\lambda =1\) and \(\lambda \neq 0\) or \(1 \) is investigated separately. The proof uses direct calculations and a theorem from [\textit{A. Böttcher} and \textit{I. M. Spitkovsky}, Linear Algebra Appl. 432, No. 6, 1412--1459 (2010; Zbl 1189.47073)], showing that under these hypotheses each matrix in \(\mathcal{A}\) has at most two distinct eigenvalues. The final section of the paper gives a criterion for an element \(A\) of \(\mathcal{A}\) to have a group inverse \(A^{g}\) and a precise formula for \(A^{g}\) in terms of \(A\).