[CP2K-user] [CP2K:15617] Energy oscillations in vdW-DF calculations
Michael Fischer
doktor... at gmail.com
Thu Jun 24 07:03:51 UTC 2021
Hello Matthias,
Thank you once again for the explanation. If the wavefunction is so well
converged and the energy still changes, is it reasonable to suspect that
the behaviour is somehow related to the non-local contribution that is
computed in the vdW-DF framework (possibly some numerical noise?)? I also
observed that the numbers in the "Change" column that are printed in the
output are often the same (usually 1.21E-06 Ha).
Best regards,
Michael
Matthias Krack schrieb am Mittwoch, 23. Juni 2021 um 18:21:14 UTC+2:
> Hello Michael
>
>
>
> The OT convergence criterion is defined by the wavefunction gradient
> printed in the column “Convergence” which reached already 6.8E-10 and not
> by the energy change between two consecutive SCF iteration steps. You can
> also try an alternative OT minimizer like CG or restart using
> diagonalization with DIIS instead of OT. On the other hand, your system
> seems to be well converged already.
>
>
>
> Best
>
>
>
> Matthias
>
>
>
> *Von:* c... at googlegroups.com <c... at googlegroups.com> *Im Auftrag von *Michael
> Fischer
> *Gesendet:* Mittwoch, 23. Juni 2021 18:07
> *An:* cp2k <c... at googlegroups.com>
> *Betreff:* Re: [CP2K:15617] Energy oscillations in vdW-DF calculations
>
>
>
> Dear Matthias,
>
> Thank you very much for the fast reply. Even when setting EPS_DEFAULT to
> 10-14, I get some of these oscillations, as shown below:
>
> Step Update method Time Convergence Total energy
> Change
>
>
> ------------------------------------------------------------------------------
>
> 1 OT DIIS 0.15E+00 5.1 0.00000073 -5301.9181906664
> -5.30E+03
>
> 2 OT DIIS 0.15E+00 2.5 0.00000029 -5301.9181922654
> -1.60E-06
>
> 3 OT DIIS 0.15E+00 2.5 0.00000005 -5301.9181886933
> 3.57E-06
>
> 4 OT DIIS 0.15E+00 2.5 0.00000001 -5301.9181899089
> -1.22E-06
>
> 5 OT DIIS 0.15E+00 2.5 4.4595E-09 -5301.9181899089
> -5.09E-11
>
> 6 OT DIIS 0.15E+00 2.5 1.8983E-09 -5301.9181886941
> 1.21E-06
>
> 7 OT DIIS 0.15E+00 2.5 6.8365E-10 -5301.9181899090
> -1.21E-06
>
>
>
> *** SCF run converged in 7 steps ***
>
>
> I am actually wondering why the SCF cycle converges at all, despite the
> change in total energy of 1.21E-06 Ha between the 6th and 7th iteration.
>
> Best regards,
> Michael
>
>
>
> Matthias Krack schrieb am Mittwoch, 23. Juni 2021 um 16:59:17 UTC+2:
>
> Hi Michael
>
>
>
> you can also try to reduce EPS_DEFAULT to 1.0E-12 or less for a better
> numerical accuracy.
>
>
>
> HTH
>
>
>
> Matthias
>
>
>
> *Von:* c... at googlegroups.com <c... at googlegroups.com> *Im Auftrag von *Michael
> Fischer
> *Gesendet:* Mittwoch, 23. Juni 2021 15:34
> *An:* cp2k <c... at googlegroups.com>
> *Betreff:* [CP2K:15615] Energy oscillations in vdW-DF calculations
>
>
>
> Dear CP2K community,
>
> I am seeing a somewhat strange behaviour in CP2K calculations using vdW-DF
> functionals (version 7.1): Within one SCF cycle, the calculation sometimes
> reaches a point where it jumps back and forth between different energy
> values. Such behaviour is illustrated in a portion of a log file appended
> to this message. This calculation used the vdW-DF-b86r functional, but I
> have seen the same behaviour with other vdW-DF functionals.
> I should add the following points:
> 1) The compound under consideration, a zeolite with an adsorbed molecule,
> contains only main group elements (Si, O, C, H, N).
> 2) Analogous calculations using DFT-D methods (like PBE-D3) show no
> convergence issues whatsoever.
> 3) A change of preconditioner (FULL_ALL/FULL_KINETIC) seems to affect the
> behaviour, but does not resolve the issue.
>
> I should also note that I tested the "CUTOFF" parameter within &XC
> &vdW_POTENTIAL &NON_LOCAL part of the input and found that it should be
> increased with regard to the default setting to get converged adsorption
> energies (I use 400 Ry).
>
> Any help or suggestions would be highly appreciated.
> Best regards,
> Michael
>
> Portion of output - oscillations start in the 8th iteration
>
> Step Update method Time Convergence Total energy
> Change
>
>
> ------------------------------------------------------------------------------
>
> 1 OT DIIS 0.15E+00 5.0 0.00004571 -5301.9144127012
> -5.30E+03
>
> 2 OT DIIS 0.15E+00 2.4 0.00001858 -5301.9158989932
> -1.49E-03
>
> 3 OT DIIS 0.15E+00 2.4 0.00001214 -5301.9161596822
> -2.61E-04
>
> 4 OT DIIS 0.15E+00 2.5 0.00000142 -5301.9161913868
> -3.17E-05
>
> 5 OT DIIS 0.15E+00 2.4 0.00000043 -5301.9161930782
> -1.69E-06
>
> 6 OT DIIS 0.15E+00 2.4 0.00000011 -5301.9161943445
> -1.27E-06
>
> 7 OT DIIS 0.15E+00 2.4 0.00000005 -5301.9161967775
> -2.43E-06
>
> 8 OT DIIS 0.15E+00 2.4 0.00000001 -5301.9161955633
> 1.21E-06
>
> 9 OT DIIS 0.15E+00 2.4 4.0079E-09 -5301.9161967782
> -1.21E-06
>
> 10 OT DIIS 0.15E+00 2.4 1.7554E-09 -5301.9161955634
> 1.21E-06
>
> 11 OT DIIS 0.15E+00 2.4 6.2727E-10 -5301.9161967782
> -1.21E-06
>
>
>
> --
> 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 cp... at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/45540fc6-9eb6-4a4b-91e0-037e3e918eaen%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/45540fc6-9eb6-4a4b-91e0-037e3e918eaen%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
> --
> 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 cp... at googlegroups.com.
>
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/de44ca45-8343-440f-b8c5-e3996cd18c32n%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/de44ca45-8343-440f-b8c5-e3996cd18c32n%40googlegroups.com?utm_medium=email&utm_source=footer>
> .
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210624/7fccb89e/attachment.htm>
More information about the CP2K-user
mailing list