Page 1 of 1

default value for optforces

Posted: Fri Apr 22, 2011 3:42 pm
by mverstra
Hello all,

proposition: optforces should default to 2 from abinit 6.8. We have wasted countless cpu cycles over the past 20 years calculating useless intermediate forces in the SCF cycle. Could automatically be set to 1 if tolrff or tolrff are set...

barring comment on this post before the release of 6.8, I propose Xavier make the update... This will probably imply changing some test outputs and possibly inputs (although I doubt the latter if we include the toldff => optforces=1 clause)

ciao

Matthieu

Re: default value for optforces

Posted: Fri Apr 22, 2011 3:50 pm
by mverstra
update: the checks on toldff tolrff already exist

Re: default value for optforces

Posted: Fri Apr 29, 2011 9:11 am
by torrent
Hi Matthieu,

1)
Putting optforces=2, you will only gain time for Norm-conserving psps.
Within PAW, the calculation of nonlocal forces (the consuming part) is combined with the calculation of rhoij occupation matrix. In this case, you will probably see a difference, but not so important.
Did you make some tests and checked that optforces=2 is visible on cpu time ?

2)
Yes optforces is set to 1 when the relevant tolerance is selected; and you get a message in log file.

Marc

Re: default value for optforces

Posted: Wed May 04, 2011 11:05 am
by mverstra
ok, thanks. Normally Xavier will incorporate this at the final merge of 6.7.2. NC is still in use, and I have seen some speed up with PAW runs too...

ciao

Matthieu