Vous n'êtes pas autorisé à supprimer des messages dans ce groupe
Copier le lien
Signaler le message
Afficher le message d'origine
Soit les adresses e-mail sont anonymes pour ce groupe, soit vous devez disposer de l'autorisation "Afficher les adresses e-mail des membres" pour afficher le message d'origine.
à cp2k
Dear all,
are there plans already to implement a working NON_LOCAL dispersion correction for the &ENERGY_CORRECTION part?
It appears to me that currently only the PAIR_POTENTIAL type delivers a finite Edisp value in the CP2K output file, although input files with a NON_LOCAL &VDW_POTENTIAL specifications get accepted with no further warning from the code, but do return Edisp = 0.d0 in the EC part of the output.
An inspections of the SUBROUTINE ec_disp(qs_env, ec_env, calculate_forces) in the file energy_corrections.F with a hard-wired call to the function calculate_dispersion_pairpot() further supports that assumption.
Thanks in advance for any clarification on that subject!
Regards,
Ralf
Jürg Hutter
non lue,
22 févr. 2024, 05:12:0022 févr.
Répondre à l'auteur
Connectez-vous pour répondre à l'auteur
Transférer
Connectez-vous pour transférer le message
Supprimer
Vous n'êtes pas autorisé à supprimer des messages dans ce groupe
Copier le lien
Signaler le message
Afficher le message d'origine
Soit les adresses e-mail sont anonymes pour ce groupe, soit vous devez disposer de l'autorisation "Afficher les adresses e-mail des membres" pour afficher le message d'origine.
à cp...@googlegroups.com
Hi
thank you for raising this issue. I have opened a new feature request issue on
github (#3277) and have disabled nl-vdW functionals in EC calculation in the
current cp2k version.
The main problem is the required 2nd derivative of the functionals for
the gradient calculation.
regards
JH
________________________________________
From: cp...@googlegroups.com <cp...@googlegroups.com> on behalf of Ralf Frischmann <ralf.p.f...@gmail.com>
Sent: Thursday, February 22, 2024 5:57 AM
To: cp2k
Subject: [CP2K:19950] NON_LOCAL dispersion in ENERGY_CORRECTION ?
Vous n'êtes pas autorisé à supprimer des messages dans ce groupe
Copier le lien
Signaler le message
Afficher le message d'origine
Soit les adresses e-mail sont anonymes pour ce groupe, soit vous devez disposer de l'autorisation "Afficher les adresses e-mail des membres" pour afficher le message d'origine.
à cp2k
Hi JH,
thanks for the quick reply and the clarification. Wasn't 100% sure whether my assumption above were actually correct.
Another possible option for getting NL dispersion in these cases could be to apply the dispersion NL functional in a post-SCF-type fashion (quantum chemistry codes ORCA and psi4 allow this as an option). Then it should be possible to take RHO from a converged HFX-SCF procedure and "stuff it" into the existing NL energy+gradient routines prior to the DFT response calculation, so with that no response calculation should be needed for the NL dispersion part.
Having the post-SCF NL option in general would also be a nice and "fairly easy" to do CP2K feature. Relative errors introduced by that are usually very small (~0.0x kcal/mol) in gas phase calculations, and substantial computational savings might result from it.
Regards,
Ralf
Jürg Hutter
non lue,
22 févr. 2024, 10:33:4022 févr.
Répondre à l'auteur
Connectez-vous pour répondre à l'auteur
Transférer
Connectez-vous pour transférer le message
Supprimer
Vous n'êtes pas autorisé à supprimer des messages dans ce groupe
Copier le lien
Signaler le message
Afficher le message d'origine
Soit les adresses e-mail sont anonymes pour ce groupe, soit vous devez disposer de l'autorisation "Afficher les adresses e-mail des membres" pour afficher le message d'origine.
à cp...@googlegroups.com
Hi
Yes, the procedure you described is available in CP2K as the
ENERGY_FUNCTION DCDFT
option in the ENERGY_CORRECTION method. If the nl vdW is only
in the energy correction functional, only first derivatives are needed.
Second derivatives would be needed only for the option
ENERGY_FUNCTION HARRIS.
Unfortunately, both options will need the update of the XC functional
call in the energy correction routines. This is no big deal, but still needs a
couple of days work (including testing).