Nonsingularity/singularity criteria for nonstrictly block diagonally dominant matrices. (Q1864967): Difference between revisions

This is an extensive study of equivalent conditions for singularity of block-matrices \(A= [A_{ij}]^N_{i,j=1}\) which have invertible diagonal blocks, are block irreducible, and satisfy some diagonal dominance criteria. For example, suppose there exists a positive vector \(v\) such that \[ \sum^N_{j\neq i,j=1}\| A^{-1}_{ii} A_{ij}\| v_j\leq v_i \] for each \(i= 1,\dots, N\). Then \(A\) is singular if and only if all those inequalities are equalities and there exist nonzero vectors \(y_1,\dots, y_N\) such that \[ {y_i* A^{-1}_{ii} A_{ij} y_j\over\| y_i\|\,\| y_j\|}= -\| A^{-1}_{ii}\|\,\| A_{ij}\|\quad\text{for each }i\neq j. \] Here \(\| X\|\) denotes spectral norm, i.e., the largest singular value of \(X\). Let \(\rho(X)\) be the spectral radius when \(X\) is square. A fundamental tool is the following block generalization of Wielandt's Lemma: If \(A= [A_{ij}]^N_{i,j=1}\) is a block matrix with square diagonal blocks and \(P\) is an \(N\times N\) no nonnegative matrix such that \(\| A_{ij}\|\leq p_{ij}\) for each \(i\), \(j\), then \(\rho(A)\leq \rho(P)\). If \(P\) is also irreducible, then \(\rho(A)= \rho(P)\) if and only if \(p_{ij}= \| A_{ij}\|\) for each \(i\), \(j\) and there exist nonzero vectors \(y_1,\dots, y_N\) such that for each \(i\), \(j\), \({y_i* A_{ij} y_j\over\| y_i\|\,\| y_j\|}= {\lambda\over \rho(P)}\| A_{ij}\|\) for some dominant eigenvalue \(\lambda\) of \(A\). When \(A\) is irreducible, necessary and sufficient conditions are proved for a boundary point of the union of the Gerschgorin circles to be an eigenvalue of \(A\), and similar equivalent conditions are proved for a boundary point of the union of the ovals of Cassini to be an eigenvalue of \(A\).