PC GAMESS/Firefly-related discussion club

Alex Granovsky
gran@classic.chem.msu.su

In addition to points which Pedro already mentioned (thanks!),
I would like to make some general comments regarding your
(and perhaps other users) questions. I think this is important
as some things are either poorly or not documeted at all.

1. The only PCM model implemented in PC GAMESS is D-PCM. This is
because the PC GAMESS was forked out from GAMESS (US) at the time
when it was the only PCM model available in GAMESS (US) as well.

2. There are multiple internal problems specific to this solvent model, namely:

• It is not very numerically stable with respect to particular
  details of tesselation procedure. Especially, energy gradients are
  not stable.
  The purpose of PCMAnalyze code was an attempt to improve stability
  by deleting and joining some "dangerous" triangles. Normally, the use
  of PCMAnalyze does not seriously affect energies, while helps improve
  gradients. However, in some cases PCMAnalyze tends to enter into
  quasi-infinite loop. The reason is it fails to reach self-consistent
  answer. To avoid this problem, one can limit the number of passes for
  PCMAnalyze by setting $pcm analyz $end to the desired maximum allowed
  number of passes.


• The energy gradients are slow for large systems! Lots of work has
  alredy been done to speed them up as much as possible, however, due
  to the nature of the D-PCM model itself, they cannot be made as fast
  as for other formulations of PCM.


• The energy gradients are inexact and includes some semi-numerical
  contributions. Thus one often needs to loosen opttol to 1d-3 or so
  for geometry optimization.

I think in the future we'll completely replace D-PCM model in the
PC GAMESS/Firefly by the fast, efficient, and stable implementation
of COSMO model.

Regards,
Alex Granovsky


On Sat Nov 8 '08 4:47am, sanya wrote
------------------------------------
>Dear Alex,

>What is the model of PCM implemented in PCGAMESS? It is not the same as in GAMESS-US or G03.

>In addition, it exhibits strange behavior: for one molecule PCMAnalyze procedure finishes after several steps, and for an almost similar molecule (just a different conformer or a protonated form) PCMAnalyze makes many hundreds of iterations. And it works so slowly that I've already put the idea of geometry optimization in PCM out of my head :( I need only CIS or TDDFT spectra.