A fully parallel method for the singular eigenvalue problem (Q2387361)

The authors consider the generalized real symmetric eigenvalue problem \(Ax=\lambda Bx\), where \(A\) and \(B\) are real symmetric and \(B\) is positive semidefinite. Hence \(A\) can be reduced to a symmetric tridiagonal matrix and \(B\) to a positive semidefinite diagonal matrix with \(b_1>0\) and \(b_i\geq 0\) for \(i=2,3,\ldots,n\). This approach makes the homotopy method a little expensive. A new efficient and accurate homotopy method based on continuation with rank 2 perturbation is presented for finding some or all finite eigenvalues of the given problem directly without first reducing it to a positive definite pencil and without any waste on computing the infinite eigenvalue. It is shown that there are exactly \(m\) disjoint, smooth homotopy paths connecting the trivial eigenvalues to the desired eigenvalues, where \(m\) is the number of finite eigenvalues of the pencil \((A,B)\). The method is a fully parallel scheme since the homotopy curves are monotone and can be easily followed independently.