PC GAMESS/Firefly-related discussion club

Learn how to ask questions correctly

Alex Granovsky
gran@classic.chem.msu.su

That's really interesting... because just a couple of days ago
we have added the possibility to restart optimization using
approximate Hessian that is stored in the DICTNRY file left by
the previous run... will be available as a part of 7.1.G release.

This requires $contrl irest=1 and works much like the
guess=mosaved (more precisely one would need to specify
$statpt hess=DAF  ). This will reuse old Hessian.
However, gdiis does hold its history in memory only
and this is why it cannot be restarted - but at least
it can be started again using better initial approximate
Hessian.

On a separate note:

Huckel guess is for MOs, not for nuclear Hessians...

Regards,
Alex Granovsky


On Sat Nov 14 '09 7:16pm, Richard wrote
---------------------------------------
>Have a look in the manual for NPUN in the $STATPT group

>Richard

>On Sat Nov 14 '09 6:46pm, Sergey wrote
>--------------------------------------
>>Dear All,

>>As I attempt to restart MP2 optimization job interrupted by power black-out, I find that IREST option is not operable since it reports 'DIIS optimization restarts are not supported'. So, the best option I can do is to copy the latest coordinates set and the last (HF) $VEC group from PUNCH file, and run that as a new job.

>>However, such the way assumes the approximate HESSIAN used in algorithm resets to some initial (Huckel) guess, and hence the lower frequencies in this approximate Hessian are lost. Consequently, for the di-molecular systems I study (e.g. nitrogeneous bases AT, CG stacks in DNA) bound weakly by dispersion interactions, it results as for many optimization cycles after such the 'restart', unless a new good approximate Hessian is elaborated, no relative (intermolecular) motion and optimization occurs.

>>From my expierence, it takes up to 20 cycles for the stacked molecules begin to 'feel' each other, i.e. days on the 4-core Intel Quad @ 4GHz.

>>Question: is there any way to preserve approximate Hessian or to print out only the current approximate Hessian (like it is printed in PUNCH, but - alas - ONLY for the very end of the succesiful optimization jobs)?

>>If this opportunity is implemented, could you pls. give me a hint how do I enable it?
>>If I have approx. Hessian, I can then use also $HESS HESS=READ $END? the problem is, I couldn't find any.

>>Thank you!