[Pw_forum] Increase in number of diagonalization errors

[Paolo Giannozzi]

On Jan 14, 2009, at 19:20 , Heather Whitley wrote:

> I am, however, curious if there is some reason why I would see these
> errors more frequently with the new version?

not that I know, but since the reason for these errors is usually  
hard to
find, it is also hard to give a more definite answer. At a given time  
the
algorithm for subspace diagonalization has been replaced by a better
performing one. Maybe it is also less stable.

>       from  pdpotf  : error #       422
>        problems computing cholesky decomposition

this is the parallel version of subspace diagonalization. In the CVS  
version
there is another one available, by the way, based on SCALAPACK

>      startingwfc='random'

there is usually no need to start with random wavefunctions

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