Schur complements and the Weinstein-Aronszajn theory for modified matrix eigenvalue problems (Q1112913): Difference between revisions

Motivated by the computational success of intermediate-problem techniques since about 1960, the authors consider the Weinstein-Aronszajn theory from the computational viewpoint in connection with eigenvalue problems for an Hermitian \(m\times m\) matrix B and an \(n\times n\) matrix A that is (i) an extension \((n>m)\), \(A=\left[\begin{smallmatrix} B&C\\C^*&D\end{smallmatrix}\right]\), (ii) a restriction \((n<m)\), \(A=P^*BP\), P having orthonormal columns, (iii) a modification \((n=m)\), \(A=B+H_ k\), \(H_ k\) Hermitian and of rank k. After a brief survey of relevant recent work (Sec. 1) and an explanation of the problems (Sec. 2) the authors derive (Sec. 3) the so-called Aronszajn formulas and spectrum-slicing formulas involving the Weinstein- Aronszajn matrices W(\(\lambda)\) corresponding to the three types of problems, and (Sec. 4) consider the duality obtained by interchanging the roles of B and A, and finally (Sec. 5) give eigenvalue inclusions and exclusions involving \(\| W(\lambda)^{-1}\|\), the norm being the Euclidean vector or matrix operator norm. The further study of practical issues, e.g., those arising from fast methods for intermediate problems is promised for later papers.