Functionality
|
This is a SPECIALIZED thick restart
implementation of Lanczos. It is for
the case where one wants a few eigenvalues and eigenvectors of a large
hermitian eigenproblem for which shift and invert is not possible.
This method aims to reduce both storage space and
computation time in three ways.
The thick restart technique is used, and the user chooses how many Lanczos
vectors are generated. The implementation of matrix-vector multiplication is
left to the user so that it can be custom made for the problem being solved.
To reduce the time spent in reorthogonalization, it is only performed when
necessary. This method performs less arithmetic operations than the comparable
implicitly restarted Arnoldi and thick restart Davidson methods. Most
arithmetic computation is done through calls to BLAS and LAPACK.
The program is highly modularized so as to be easily
modifiable. It is written in Fortran 90 which can be easily called from C
or C++. This code can run on a single address machine or in a
distributed parallel environment, which requires MPI.
|