Energy oscillations in vdW-DF calculations

111 views
Skip to first unread message

Michael Fischer

unread,
Jun 23, 2021, 9:33:35 AM6/23/21
to cp2k
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

MOR_acetaminophen_1.inp

Krack Matthias (PSI)

unread,
Jun 23, 2021, 10:59:17 AM6/23/21
to cp...@googlegroups.com

Hi Michael

 

you can also try to reduce EPS_DEFAULT to 1.0E-12 or less for a better numerical accuracy.

 

HTH

 

Matthias

 

--
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/45540fc6-9eb6-4a4b-91e0-037e3e918eaen%40googlegroups.com.

Michael Fischer

unread,
Jun 23, 2021, 12:07:22 PM6/23/21
to cp2k
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

Krack Matthias (PSI)

unread,
Jun 23, 2021, 12:21:14 PM6/23/21
to cp...@googlegroups.com

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

 

Michael Fischer

unread,
Jun 24, 2021, 3:03:51 AM6/24/21
to cp2k
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
Reply all
Reply to author
Forward
0 new messages