Local convergence of Newton-like methods for degenerate eigenvalues of nonlinear eigenproblems: II. Accelerated algorithms (Q2514251)

To speed up the convergence rates of single-vector Newton-like methods for defective eigenvalues, an accelerated inverse iteration is proposed and studied, showing that it exhibits quadratic local convergence towards defective eigenvalues at the cost of solving two linear systems per iteration. A detailed analysis for a defective \(\lambda\) with a single Jordan chain is performed, but it can be extended without difficulty to the case involving multiple Jordan chains. An accelerated Jacobi-Davidson method that bears a close connection with the accelerated inverse iteration is also proposed. Numerical experiments confirm the quadratic convergence of the accelerated methods. The most efficient up to now methods for solving a defective eigenvalue of algebraic eigenproblems of the form \(T(\lambda)v = 0\), including the standard linear problem \(Av = \lambda v\), polynomial, rational, and truly nonlinear eigenproblems is received. For Part I see [ibid. 129, No. 2, 353--381 (2015; Zbl 1309.65059)].