Inverse Iteration is a well known and well studied method for the
calculation of A x =λ x, which requires solves with
A-σ I where σ is the shift.
However for large sparse matrices one is usually forced to use
iterative methods for the linear systems and so
(A-σ I)y=x is satisfied to a residual tolerance τ.
Due to this one obtains an inner-outer method with two parameters
σ and τ.
Questions of convergence and optimal choices for the
parameters need to be answered.
In previous contributions [1] and [2] we provided a general convergence result.
Further we showed that it is beneficial to use the Rayleigh quotient
as shift when Galerkin Krylov solvers are applied to the linear systems.
Recently [3] showed that it is important to change the
right hand side of the linear systems when Cholesky preconditioning is used.
In [1] and [2] we extend this result to arbitrary preconditioners.
This talk will outline theoretical treatment, present numerical result
based on examples from matrix market, including a comparison with LOBPCG,
and discuss in detail implementational issues.
[1] Berns-Mueller, Joerg:
Inexact Inverse Iteration using Galerkin Krylov solvers,
PhD thesis,
University of Bath.
to be submitted.
[2] Berns-Mueller, J., I. G. Graham, A. Spence:
Inverse Iteration and Inexact Solves. to appear.
[3] Simoncini, Valeria and Elden, Lars (2002):
Inexact Rayleigh quotient-type methods for eigenvalue computations,
BIT 42(1), 159-182