[Pw_forum] Parallelization of diaghg

[Paolo Giannozzi]

On Jan 6, 2007, at 23:49 , Alexander Shaposhnikov wrote:

> By the way, i have found that diago_david_ndim=2 gives better
> times and also saves alot of memory comparing with default =4.

it saves a lot of memory by reducing the workspace where auxiliary
vectors (H-e)psi are stored during the diagonalization. This typically
leads to a slight increase in execution times, because each time
the workspace is filled one has to recalculate a new estimate of
the eigenvectors. Maybe in your case the advantage of a reduced
memory footprint more than compensate the cost of such operation
(which has to be done anyway when the diagonalization is converged).

Paolo
---
Paolo Giannozzi, Democritos and University of Udine, Italy