8
$\begingroup$

"Failure to locate stationary point, SCF has not coverged"

This is the most annoying message after Optimization run ( even with a good primary start geometry like UFF we ay see that) . The only solution I know is to use the output log file for another run but it is very time consuming. Is there any way to force GAMESS to automatically do that ?( I mean makes another automatic run on current output results ).

$\endgroup$
14
$\begingroup$

There can be many reasons for the SCF to not converge.

  • The initial guess wasn't great, and the convergence is slow. Increasing the number of SCF iterations can help here.
  • The SCF oscillates. This can happen if there are two orbitals close in energy. Using damping, mixing (i.e., the next iteration is 50% the old and 50% the updated iteration), Fermi broadening, etc. can help here.
  • The SCF strategy (e.g., DIIS) is causing problems. Usually DIIS can greatly accelerate convergence, but in some hard-to-converge systems, you need a lot of iterations with another strategy (e.g., Newton-Raphson)

Without more information from your particular file, it's impossible to know how to help. Many times, you may simply want to increase the maximum number of SCF iterations and it'll work fine.

$\endgroup$
  • $\begingroup$ Thank you Geoff your suggestion for using DAMP=.T. solved my current problem. Sometimes I have seen using Direct SCF +/- DIIS helps also, but not always. What I am looking for is a systematic approach for non-convergence problem. Can we reach a conclusion on it for future reference ? $\endgroup$ – Aug Mar 7 '15 at 4:28
  • $\begingroup$ Also I don't know if there is somewhere in Avogadro for setting Fermi broadening. $\endgroup$ – Aug Mar 7 '15 at 4:30
  • $\begingroup$ I don't think GAMESS has an implementation of Fermi broadening. Different QM packages have different SCF convergence tricks. $\endgroup$ – Geoff Hutchison Mar 7 '15 at 12:20
  • $\begingroup$ There is also no "systematic approach" because the approach will vary depending on the program and depending on the type of SCF convergence issue. You have to take a look at the SCF iterations and diagnose for yourself (or ask others). $\endgroup$ – Geoff Hutchison Mar 7 '15 at 12:21
0
$\begingroup$

Moreover, the problem may depend on the basis set chosen, the geometry, and the molecule properties. Some basis sets may be inappropriate for some molecules (and switching to other basis set may fix the problem), some geometries may be more prone to problems (and changing bond lengths may solve the issue), if you do not consider the correct multiplicity or electron distribution it may also fail, and you should change those, allowing orbital mixing may also help in the same way. Sometimes, computing the structure of the same molecule with less electrons will facilitate convergence and then you may use the result for a final full-electron calculation. The same may happen if you include (or avoid) excited states. In some cases I've seen "naturally unsolvable" problems were the structure was oscillating between two states with very similar energy, maybe tautomers. And it may also happen that there is some chemistry going on with bonds breaking and forming (I saw this on some long-chain radicals which might break down into pieces)... A possible approach may be sometimes to build the molecule incrementally.

The point is that there are too many things that may be going on and that each case is special, so you'll have to look into your problem, see the evolution of energy and error, visualize the evolution of the geometry with energy, look at the values of the basis set for your atoms (some of them may lead to non-orthogonality if a coefficient for one of your atoms is too small) and decide on your specific problem.

Right now, there is no silver bullet that can solve all problems.

$\endgroup$
  • $\begingroup$ How can tautomers occur during a SCF calculation? The coordinates of the nuclei do not change so how can there be different tautomers? How can bonds break or be formed, long chains breaking down? That should be impossible during a single point calculation. Do you maybe refer to a geometry optimization here? Because that is something completely different! $\endgroup$ – logical x 2 Oct 17 '17 at 18:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.