Wavelet Poisson solver - Huge change in total energy when increasing the plane-wave cut-off energy from 1200 to 1260 Ry

[Torstein Fjermestad]

Dear all, 

As part of a basis set convergence test, I gradually increase the plane wave basis set cut-off energy. The atomic system is a cluster model of methanol adsorbed in a zeolite. I isolate the system from its neighboring unit cells by using the wavelet Poisson solver. 

I use a cubic box of 30 Å. The size of the system is about 21 Å in the longest direction. The atom-centered basis set is TZV2P. 

When I increase the cut-off from 900 Ry to 1200 Ry, the total energy varies in a range less than 5 kJ/mol. However, when I increase the cut-off from 1200 Ry to 1260 Ry, the total energy is after  6 SCF iterations  1.38E+07 kJ/mol lower! 

Because of the massively different total energy, I stopped the calculation.   

I have attached the output files of the two calculations. The complete input can be found at the start of the output files. 

What could be the cause of this behavior? 

Thank you very much for your help. 

Please tell me if you need more information to understand the problem better.

Best regards,

Torstein Fjermestad

 

[Frederick Stein]

Dear Torstein,

You not only need to adjust the CUTOFF but also EPS_DEFAULT (and if set explicitly the other thresholds) in the QS section. How did you obtain the input file? I would never set all parameters. Especially EPS_DEFAULT sets a variety of other parameters if these are not set explicitly.

Best,

Frederick

[Torstein Fjermestad]

Dear Frederick, 

Thanks for the piece of advice.

I generated the input file by setting the name of the variable and its value like this:

@SET <name> <value>

In the QS section, I explicitly set the following variables:

@SET eps_def 1.0E-10
@SET extrapol PS
@SET precOM 1.0E-5

    &QS

      EPS_DEFAULT ${eps_def}
      EXTRAPOLATION ${extrapol}
      EPS_PGF_ORB ${precOM}
      EXTRAPOLATION_ORDER 3 
    &END QS

 
I followed your advice, and I decreased the value of  EPS_DEFAULT to  1.0E-12 and the value of  EPS_PGF_ORB to 1.0E-6.

This did not have any notable effect on the results. 

(For the calculation with a cut-off of 1260 Ry, the energy after the first SCF iteration was -4394.5092346069 Hartree with the new settings and -4394.5090808424 Hartree with the previous settings. However, these values are already a few thousand Hartree below the converged value of the calculation with a cut-off of 1200 Ry (-1389.249071606565622 Hartree))

Should I decrease EPS_DEFAULT and EPS_PGF_ORB further, or are other variables I should change?

Thanks. 

Best regards,

Torstein

[Jürg Hutter]

Hi

yes, I can reproduce your problem using a small molecule in a 30x30x30 box with
cutoffs 1200 and 1260.

These settings result in internal FFT grids of size 625 (works) and 648 (fails).
Some tests show that up to 640 (cutoff 1230) it works, everything larger fails.

This needs a deeper look into the wavelet poisson solver. Maybe we should open
a bug report at the CP2K page on Github.

regards
JH

________________________________________
From: cp...@googlegroups.com <cp...@googlegroups.com> on behalf of Torstein Fjermestad <tfjer...@gmail.com>
Sent: Monday, December 29, 2025 10:18 AM
To: cp2k
Subject: [CP2K:22012] Re: Wavelet Poisson solver - Huge change in total energy when increasing the plane-wave cut-off energy from 1200 to 1260 Ry

--
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/9cbaa759-1d2e-41c9-a4ad-5e376f6398f3n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/9cbaa759-1d2e-41c9-a4ad-5e376f6398f3n%40googlegroups.com?utm_medium=email&utm_source=footer>.