HP MLIB User's Guide Vol. 2 7th Ed.

Chapter 16 Introduction to BCSLIB-EXT 1083
BCSLIB-EXT functionality
• Standard eigenvalue problems
• Generalized eigenvalue problems
BCSLIB-EXT solves these problems in cases where the coefficient matrices are
sparse and large, perhaps extremely large.
Sparse matrices are those for which almost all of the entries are zero; only a few
entries in any given row or column are not zero. For example, a routine
structural engineering application of BCSLIB-EXT might involve a coefficient
matrix K of order 1,000,000 but on average, a row of K would have only 60
nonzero entries. The sparsity of the coefficient matrix has two implications.
BCSLIB-EXT provides out-of-core functionality: it is unique in its capabilities
for problems that are too large to fit conveniently in main memory. It has
extensive secondary storage capabilities to enable the user to solve extremely
large problems.
BCSLIB-EXT provides four high-level drivers to solve sparse linear equations
and symmetric eigenvalue problems for matrices stored in either triples or
compressed column representation. These subroutines, XDSLFS, XDSLFT,
XDSEE1, and XDSEET provide the simplest ways to use BCSLIB-EXT. In
total, there are five options for communicating the coefficient matrix(ces) to the
package:
• Single point entry
• Vector of triples input
• Dense submatrix imput
• Column input
• Compressed column matrix input
BCSLIB-EXT allows the user to mix and match all of these methods within the
application. For additional information on sparse matrix storage format, refer
to Chapter 4, “Sparse BLAS Operations”.
An electronic copy of the BCSLIB-EXT documentation can be found in the HP
MLIB installation tree in /opt/mlib/doc.