next up previous contents index
Next: Numerical Examples Up: Lanczos Method   A. Previous: Local Reorthogonalization and Detecting   Contents   Index

Software Availability

There are a few software packages available for the Lanczos methods described in this section. One, LANCZOS, was published by Cullum and Willoughby [91,92]. It uses the Lanczos method without reorthogonalization but contains a mechanism to detect and discard spurious eigenvalue approximations. Scott developed LASO, a block Lanczos code with selective orthogonalization, as described in the pioneering paper [363]. The package LANZ was written by Jones and Patrick [249] and includes the SI (4.14) and uses selective (partial) reorthogonalization, as described in §4.4.4 here.

Finally there is a code, LANSO, a FORTRAN package that implements selective orthogonalization as developed by the group led by Parlett at UC Berkeley. The presentation of this section follows the ideas behind LANSO very closely. LANSO makes periodic reorthogonalization as recommended in §4.4.4 and calls a standard tridiagonal QL algorithm to compute eigenvalue approximations (4.11) and estimate residuals (4.13), at appropriate intervals. PLANSO is a parallel version developed by Wu and Simon [462].

For more information about these codes, including how to access them, see the book's homepage, ETHOME.


next up previous contents index
Next: Numerical Examples Up: Lanczos Method   A. Previous: Local Reorthogonalization and Detecting   Contents   Index
Susan Blackford 2000-11-20