PC GAMESS-related discussion club

Igor Polyakov
polyakoviv@gmail.com

>is there a way to lift SCF convergence from 2x10-5 up? I have particularly tough to converge systems and trying to preoptimize with MINI. Problem is that it's oscillating near 1x10-4 and not converging.

In manual it is said:

         NCONV  = n  SCF density convergence criteria.
                  Convergence is reached when the density change
                  between two consecutive SCF cycles is less than
                  10.0**(-n) in absolute value.  One more cycle
                  is executed after reaching convergence.   Less
                  accuracy in NCONV gives questionable gradients.
                  (default is n=5, except CI or MP2 gradients n=6)

And this one really works- i've just checked. So add $scf nconv=4 $end In case of difficult convergence u can also try to change the convergers ( switch to DIIS from SOSCF for example). Also sometimes u can converge the case with higher multiplicity and use this guess as a start for ur original calculation or if u use HF u can compute DFT (as from my experience DFT converges better) orbitals and use them as a guess. There is also a lot of trickier ways.

>That's odd, but changing keyword CONV has not been recognized and simply aborted calculation as if the keyword was illegal. So did trying to change NOCONV to true. It's aborting as if this keyword was illegal. Have some of the GAMESS keywords been changed in PCGAMESS?

there is no such option NOCONV for PC GAMESS - and also it is not described in manual. Please use input.doc from default PC GAMESS distribution and for some specific options- manuals from "readmes" directory. Also on this webpage u can find some additional manuals.

Best luck, Igor