[Pw_forum] PW taskgroups and a large run on a BG/P

Lorenzo Paulatto paulatto at sissa.it
Thu Feb 12 09:20:12 CET 2009


On Wed, 11 Feb 2009 21:08:14 +0100, David Farrell <davidfarrell2008 at u.northwestern.edu> wrote:

> Indeed it isn't - so that was a dead end. Guess I'll have to look into
> why the call to the lapack lib is failing.

That subroutine is usually where error originated in the most remotes parts of the code finally emerge. There are dozens of possible errors that will not cause any apparent problem for a while, until the errors accumulate enough to break positive definiteness of the matrix.

Some possible reasons (in not very strict order of plausibility):
1. especially with USPP, atoms too close together will break several approximations, this may happen after a few ionic optimization iterations
2. bad pseudopotential, with a ghost state at high energy
3. smearing is too low, especially with mexfeld-paxton or marzari-vanderbilt smearings
4. cutoff is too low
5. bad interpolation of wavefunctions over ionic dynamic steps
6. some error in the algorithm

cheers

-- 
Lorenzo Paulatto
SISSA  &  DEMOCRITOS (Trieste)
phone: +39 040 3787 511
skype: paulatz
www:   http://people.sissa.it/~paulatto/

     *** save italian brains ***
 http://saveitalianbrains.wordpress.com/



More information about the users mailing list