Grassmann algebras as Hilbert spaces (Q2531312)

Let \(A= (A_{ij})\) \((1\le i, j \le m)\) be an \((mn)\)-square positive definite Hermitian matrix, where each \(A_{ij}\) is an \(n\)-square matrix. Let \(A_{(k)} = (A_{ij})_{1\le i, j \le k}\) be the upper left \((nk)\)-square submatrix of \(A\), and let \(\tilde A_{(k)} = (\det A_{ij})_{1\le i, j \le k}\) denote the \(k\)-square matrix obtained by taking the determinant of each block in \(A_{(k)}\). A result due to \textit{M. Marcus} and \textit{W. R. Gordon} [Duke Math. J. 31, 691--696 (1964; Zbl 0123.00403)] asserts that \[ \frac{\det A_{(k+1)}}{\det \tilde A_{(k+1)}} \le \frac{\det A_{(k)}}{\det \tilde A_{(k)}}\,. \] Here \(k= 1,2,3,\ldots,m - 1\). Moreover, equality holds (for fixed \(k)\) if and only if \(A_{(k+1)}) = A_{(k)}\oplus A_{(k+1,k+1)}\). In the paper under review this result is proved anew. The proof involves an analysis of projections in certain Grassmann spaces, and has some features in common with the Marcus-Gordon proof. It seems to the reviewer that the organization of the proof in the paper under review would have involved less technical detail than the Marcus-Gordon proof were it not for the fact that at a crucial stage it is necessary to employ a ``deus ex machina'' in the form of a result of \textit{K. Vala} [C. R. Acad. Sci., Paris 242, 2499--2500 (1956; Zbl 0071.32704), resp. Ann. Acad. Sci. Fennicae, Ser. A I 233, 36 p. (1956; Zbl 0071.32705)].