SCF run not converged-wB97X-V

[farzaneh sarrami]

Hi Cp2k experts:

I am new to Cp2k and trying to run FPMD calculation for my system using wB97X-V  as XC_FUNCTIONAL and NVE for MD part. But the SCF run not converged after 100 cycles.

I did increase the number of cycles and Cutoff as well but none of them works. Would you please help me? I have attached my input file to have a look if sth wrong?

Thanks

Farzaneh

[Mohammad Salem]

Hi Farzaneh,

I am new to CP2K as well, but I took a quick look on the SCF subsection: 

"         MINIMIZER DIIS

         PRECONDITIONER FULL_SINGLE_INVERSE

         ENERGY_GAP 0.01

“

How big is your system? 

Please check the tips on setting these parameters here. You could try “CG” for the MINIMIZER, “FULL_ALL” for the PRECONDITIONER together with a small energy gap.

Regards,

Mohammad

--
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/eb9b8714-3782-458c-9737-cde20c3f3f88%40googlegroups.com.
<PEDOT-Group.inp>

[Frederick Stein]

Dear Farzaneh,

You have to set up a Hartree-Fock section and use a different vdW potential. You need basically the same setup like with its metaGGA version wB97XM-V (https://groups.google.com/forum/#!searchin/cp2k/wb97x%7Csort:date/cp2k/ARya5uM-pA4/2wt5PgtkAQAJ).


Regards

Frederick

[farzaneh sarrami]

Thanks Frederick for your contribution.

Actually I did some changes based on your comment and now is running but very slowly. I am wondering if this might be the reason?

I attached the modified input file. would you please have a look and give me your comment?

Cheers

Farzaneh

[Frederick Stein]

Dear Farzaneh,

Without further information this is not easy to tell. How have you compiled CP2K and do you run your calculations in parallel? How many atoms do you have in your system?

Do you have the timing report of a single point calculation?

EDIT: When the calculation was faster before, it might be the calculation of the Hartree-Fock integrals (there are simply a lot). You can try a higher EPS_SCHWARZ (lets say 1e-8).

Best regards

Frederick

[farzaneh sarrami]

Thanks for your quick response, I have compiled CP2K and I do the calculation in parallel.

I have 86 atom in my system currently. I just realised that the SCF convergence problem still exist.

Actually I am not sure that my input file is correct completely. I would appreciate you if you double check it for me.

Cheers

Farzaneh

On Thursday, August 8, 2019 at 3:36:25 PM UTC+2, Frederick Stein wrote:

Dear Farzaneh,

Without further information this is not easy to tell. How have you compiled CP2K and do you run your calculations in parallel? How many atoms do you have in your system?

Do you have the timing report of a single point calculation?

[Frederick Stein]

The input file looks correct to me. When you still have convergence issues, you might try the conjugate gradient minimizer (MINIMIZER CG). How many MPI ranks do you use? How much time does one step need?

Concerning the costs you can try to use a higher EPS_SCHWARZ (if it really is Hartree-Fock and the results do not get worse) or a different preconditioner like FULL_KINETIC or FULL_SINGLE.

[farzaneh sarrami]

Each step will take about 10min.

I will try with MINIMIZER CG and let you know.

Thanks for your advices.

Farzaneh

[Stanislav Šimko]

Dear Farzaneh,

I had only a very brief look at your input, however, according to "latest" literature, you're not using "recommended" setup. I suggest to use "PARAMETERS 6.2 0.01" for the rVV10 section, as can be found at 10.1021/acs.jpclett.6b02527

I know it's quite a bit late but hopefully you read it. Besides this, my "latest" setup is rather similar to yours, however, I do not have got "benchmark" for the cp2k results yet, so I am still "unsure" whether we don't miss something important. Please, let me know if you're interested in helping with the "benchmark".

Best regards,

Stanislav