Transition metal oxide SCF Convergence

[Börries von Seggern]

Dear Cp2k Forum,

For a while now, I have been trying to calculate XAS Spectra of transition metal oxides using the LR-TDDFT method. Unfortunately, I wasn`t able to make the preceeding Hybrid-DFT  SCF calculation converge for open-shell systems (CuO, CoO, NiO), while other closed shell systems (MgO, ZnO) show rapid convergence quite easily. So far, I've tried using the OT methods, which do seem to converge until reaching a sort of plateau, where the energy starts increasing in miniscule amounts. Further, Iǜe been trying to use diaganolization, however these calculations seem to oscillate even when including very little amounts (1e-2) of the new density.

Concerning the OT methods, I ve tried both DIIS and CG (with 2pnt and 3pnt linesearch), however, all calculations show the same behaviour.

Could anyone please help me with these calculations?

Best regards,

Börries

PS: Please find an example input, output and structure files attached, I am using cp2k - 8.2. As even this calculation using the CG optimizer with linesearch 3pnt reached the plateau and was very expensive, I unfortunately cancelled the calculation. (But could restart it if necessary)

[Nicholas Winner]

Some suggestions for what to try:

(1) Precision

 Your value of EPS_PGF_ORB is too large and should be substantially decreased. I think the default value is too large for a lot of systems. Most numerical issues stem from the overlap matrix precision in my opinion. Set it to something like 1e-16 and see what happens. That might be too fine, but its not very much extra computational cost and it can make a huge difference.

(2) Try these settings

&OT

 algorithm irac # do not enforce strict orthogonality

 minimizer cg # safer than DIIS, though often its too safe and makes the calculation slow.

 linesearch 2pnt # 3pnt is not compatible with irac+rotation

 rotation true # allow for subspace rotations (fractional occupations)

 preconditioner full_single_inverse # full_all not compatible with rotations

&end

If that doesn't work, decrease STEPSIZE in &OT to be 0.05 

Hope that helps,

Nick

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/149a0cc5-5ea3-4a94-b3ea-b893b725b9ddn%40googlegroups.com.

[Börries von Seggern]

Hey,

Thanks a lot for your help!

Unfortunately, the issue persists. :/

Do you have any other ideas of what could be going wrong?

Please find some outputfiles attached.

Kind regards,

Börries