MoS2 in CP2K and CPMD: identical models give different results

[Fominykh Nikita]

Dear colleagues,

I am comparing the MoS2 model with the PBE xc-functional in CP2K and CPMD with the same GTH pseudopotential. I have found some strange discrepancies.

For comparison, I calculate energy differences while changing the lattice constant in the xy-plane (compression and tension). In these tests I use Gamma point only. Please see 3 plots with the Energy-vs-Strain attached.

1) First, let us consider the bulk MoS2 in the 2H structure. CP2K and CPMD with the same GTH pseudopotential give different results. Other CPMD calculations with the ‘CPMD-native’ GTH model and the CPMD MT pseudopotential give similar results to CPMD-withGTHcp2k. Interestingly, VASP results are very close to CP2K.

In order to decipher the puzzle, I have made calculations for pure Mo and for pure S.

2) For the bcc Mo structure all the codes with different pseudopotentials give very close results.

3) However, for the S6 structure of sulphur, the CPMD-withGTHcp2k and CPMD-withMT results agree well with VASP, but do not match CP2K. And the CPMD results for its ‘native’ GTH model show a significant offset.

I also tried changing energy cutoffs, enabling various k-meshes, and enlarging the cell, however I couldn’t find the possible reasons for these discrepancies in CP2K and CPMD with the same GTH pseudopotential of MoS2 and for S6.

There are example files for these comparisons attached.

Please let me know your best judgement about the possible roots of these discrepancies.

Best regards,
Nikita

[Martin Konôpka]

Dear Nikita Fominykh,

The differences may be because of the quite different basis sets used in CP2K/Quickstep on one side and in CPMD and VASP on the other side. I mean basis sets for wavefunctions. In CP2K the standard method is to use gaussian basis sets (and you really used them). In CPMD and VASP plane-wave basis sets are used.

Best regards

Martin Konôpka.

Dátum: piatok 28. marca 2025, čas: 18:13:13 UTC+1, odosielateľ: Fominykh Nikita

[Jürg Hutter]

Hi

I see (or guess) two possible problems:

1) The GTH PP for S from CPMD is not good. You could check this by running that PP in CP2K.
2) GTH PP in CPMD need a high cutoff. Your CPMD(GTH) calculation is not yet converged?

regards
JH

________________________________________
From: cp...@googlegroups.com <cp...@googlegroups.com> on behalf of Fominykh Nikita <nikfo...@gmail.com>
Sent: Friday, March 28, 2025 6:13 PM
To: cp2k
Subject: [CP2K:21309] MoS2 in CP2K and CPMD: identical models give different results

--
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<mailto:cp2k+uns...@googlegroups.com>.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/79e3b838-3ca5-4f31-a606-dbf12dec50cen%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/79e3b838-3ca5-4f31-a606-dbf12dec50cen%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Nikita Fominykh]

Dear Dr. Konôpka, dear Prof. Hutter, 

thank you very much for the helpful hints! After additional tests with the plane wave Sirius implementation, I could find the following answers:

1) In the case of pure S the problem was really in the basis. So with Sirius, it was possible to get an immediate agreement. When I picked TZV2PX, I got the agreement with Quickstep too.

2) In the case of MoS2, where I initially checked the convergence of the basis, it turned out to have no significant impact. 

However, in further tests I noticed a difference in the electrostatic energies. To reproduce the results from CP2K in CPMD, it turned out to be necessary to increase the number of additional supercells included in the real space sum for the Ewald term (the TESR was zero by default).

Best regards,

Nikita

понедельник, 31 марта 2025 г. в 13:05:04 UTC+3, Jürg Hutter: