PC GAMESS-related discussion club

Alex Granovsky
gran@classic.chem.msu.su

just a couple of suggestions from my side

1. Most of the options you are using are on by default.

2. Most of the default PC GAMESS settings are selected to deliver the best performance

>$CONTRL SCFTYP=RHF DFTTYP=B3LYP RUNTYP=OPTIMIZE COORD=UNIQUE ICHARG=0 MULT=1 gencon=.t. fstint=.t. reordr=.t. $END

3. Your $contrl group looks fine to me. I would suggest in addition to use DLC internal coordinates for optimization:

 $contrl nzvar=1 $end
 $zmat dlc=.t. auto=.t. $end

4.
>$SYSTEM MWORDS=30 TIMLIM=600000 kdiag=0 nojac=100 aoints=dist mpisnc=.t. $END

I would suggest to drop mpisnc and aoints options.


5.

>6. For $SCF group, and $INTGRL group, QFMM is turned on:
>$SCF    dirscf=.t. $END
>$INTGRL schwrz=.t. $END

Actually, this does not turn QFMM on. To turn it on, you need to set additional flag in the $intgrl group. It is generally bad idea to use QFMM for small molecules. For larger systems, it is better to try first whether QFMM speeds up things or not for your particular molecules, then to use or not to use QFMM.

6.
>7. For $FMM group, I am using full direct option with PCGAMESS routines for Coulomb:
>$FMM    method=fulldrct nearj=2 $END

I would suggest just to use default settings.

Other parts of your input look reasonable to me.

Best regards,
Alex



On Tue Feb 26 '08, Tae-Yun Park wrote
-------------------------------------
>Dear PCGAMESS users,

>Although everyone knows the excellent performance of PCGAMESS in terms of calculation speed, it looks to me that it is not difficult to miss some of useful options to enhance the performance even more since there are lots of possible option combinations that users may overlook.

>Considering the possibility that I may be one of those users, I would like to list out the options I’m currently using at the moment, and make these options to be reviewed by the PCGAMESS experts.

>Currently, I am performing geometry optimization and frequency calculation followed by raman calculation using a single AMD quad core CPU(Phenom9500) based machine with 4-core parallelization enabled by mpich under CentOS 5.1. Typical molecule to be calculated consists of 25 heavy atoms, e.g., C, N, O, S, suitably connected with hydrogen.

>Here are the options for geometry optimization and frequency calculation:

>1. For $CONTRL group, I am using unique coordinate with exact symmetry, and turning on the fastints options:
>$CONTRL SCFTYP=RHF DFTTYP=B3LYP RUNTYP=OPTIMIZE COORD=UNIQUE ICHARG=0 MULT=1 gencon=.t. fstint=.t. reordr=.t. $END

>2. For $SYSTEM group, fastdiag option is turned on with the options for Jacobi diagonalization, fast networking, and buffer overflow:
>$SYSTEM MWORDS=30 TIMLIM=600000 kdiag=0 nojac=100 aoints=dist mpisnc=.t. $END




>3. For $GUESS group, the followings are used to speed up Huckel guess:
>$GUESS  kdiag=0 guess=huckel $END

>4. The basis set used for the calculation is:
>$BASIS  NGAUSS=6 GBASIS=N31 NDFUNC=1 $END

>5. For $STATPT group, the options are:
>$STATPT hssend=.t. NSTEP=50 $END

>6. For $SCF group, and $INTGRL group, QFMM is turned on:
>$SCF    dirscf=.t. $END
>$INTGRL schwrz=.t. $END

>7. For $FMM group, I am using full direct option with PCGAMESS routines for Coulomb:
>$FMM    method=fulldrct nearj=2 $END

>8. For $SMP and $P2P, the following options are used:
>$SMP    csmtx=.t. call64=.t. $END
>$P2P    p2p=.t. dlb=.t. $END

>For raman calculation, only run type is changed to RAMAN, deleting the hssend, i.e.,
> $CONTRL SCFTYP=RHF DFTTYP=B3LYP RUNTYP=RAMAN COORD=UNIQUE
>         ICHARG=0 MULT=1 gencon=.t. fstint=.t. reordr=.t. $END
> $SYSTEM MWORDS=30 TIMLIM=600000
>         kdiag=0 nojac=100 aoints=dist mpisnc=.t. $END
> $GUESS  kdiag=0 guess=huckel $END
> $BASIS  NGAUSS=6 GBASIS=N31 NDFUNC=1 $END
> $STATPT NSTEP=50 $END
> $SCF    dirscf=.t. $END
> $INTGRL schwrz=.t. $END
> $FMM    method=fulldrct nearj=2 $END
> $SMP    csmtx=.t. call64=.t. $END
> $P2P    p2p=.t. dlb=.t. $END

>These are the combinations of PCGAMESS options I found so far to improve the calculation speed of geometry optimization and frequency calculation followed by raman calculation.

>I wonder if there is any option that I’m missing or incorrectly assigned to improve the calculation speed of PCGAMESS for my run type.

>Any suggestion or comment related to this subject is greatly appreciated.

>Many thanks in advance,

>Tae-Yun
>