Firefly and PC GAMESS-related discussion club



Learn how to ask questions correctly

Re: How to improve convergence

sanya
sanya@photonics.ru

Are you using HF or DFT? Anyway, your system can hardly be treated by any of these methods. Are you really interested in an open-shell high-spin system? My experience shows that such open-shell systems should be capped by hydrogens to avoid problems with degenerate states that cause poor convergence (believe me, graphite was the subject of my M.Sc. thesis 16 years ago). Real graphite is not paramagnetic, so high-spin models are unappropriate for it. Instead, hydrocarbons with fused rings (such as coronene, etc.) can serve as very good models. They are closed-shell, and you'll have no problems with convergence. However, if you are going to use DFT for geometry optimizations, take dispersion-corrected functionals (implemented in GAMESS-US), because Van der Waals interactions (which are underestimated by conventional DFT) are very important part of plane-plane interaction in graphite.

On the other hand, if you are still going to deal with such unrealistic open-shell models, you can start with the orbitals from the converged run (with mult=5) (GUESS=MOREAD). It may help as likely as not. Changing the value of COORD or INTTYP options has no effect on the SCF convergence. And the reliability of the results obtained for an inappropriate model is questionable.

Alternatively, you may use some plane-wave packages (such as GPAW).

On Fri Dec 18 '09 12:20pm, Vyacheslav wrote
-------------------------------------------
>Hi,
>please, advise me how to improve convergence in my system. My system is a fragment of graphite' structure - three hexagonal rings in one (001) plane and one more ring in the next plane. In point calc of energy for mult=5 run finished succesfully but after ~ 640 iterations only. For mult=3 i cannot achieve convergence. I used different variants including such as INTTYP=huckel/hondo, GUESS=hcore, COORD=unique/zmt (without angles near 180) but always run came to failure.
>In the future the much more complex structures including this fragment wait for me...
>Thanks in advance!
>

[ Previous ] [ Next ] [ Index ]           Sun Dec 20 '09 2:26am
[ Reply ] [ Edit ] [ Delete ]           This message read 1103 times