Next: Shift-Invert Mode
Up: Selecting a Symmetric Driver
Previous: Selecting a Symmetric Driver
Driver dsdrv1 solves the standard eigenvalue problem
This mode only requires matrix vector products with It is most appropriate for computing extremal, non-clustered eigenvalues.
This mode can be successful for clustered eigenvalues, but
convergence is often much slower in this case. Since this
mode is generally the easiest to try, it is recommended to
use it initially and convert to more appropriate modes after
some information about the distribution of the spectrum has been obtained.
It can be particularly useful to run initially
with which = 'BE' with a loose tolerance just to get
an idea of the largest and smallest eigenvalues.