Next: Test Matrices for the
Up: Testing the Nonsymmetric Eigenvalue
Previous: Testing the Nonsymmetric Eigenvalue
The driver routines for the nonsymmetric eigenvalue problem are
- eigenvalue/eigenvector driver,
- expert version of xGEEV (includes condition estimation),
- Schur form driver, and
- expert version of xGEES (includes condition estimation).
For these subroutines,
some tests are done by generating random matrices
of a dimension and type chosen by the user, and computing error bounds
similar to those used for the nonsymmetric eigenvalue computational
routines. Other tests use a file of precomputed matrices and condition
numbers, identical to that used for testing the nonsymmetric
eigenvalue/vector condition estimation routines.
The parameters that can be varied in the random matrix tests are:
- the order 13#13 of the matrix 16#16
- the type of test matrix 16#16
- five numerical parameters: NB (the block size),
NBMIN (minimum block size),
NX (minimum dimension for blocking),
NS (number of shifts in xHSEQR), and
NBCOL (minimum column dimension for blocking).