Non-physical pressure behavior for non-local vdW functionals

[Luke Gibson]

Dear developers and users,

We are looking at the equilibrium density for an AlCl3 molten salt melt by measuring pressure values over 10 ps in NVT at a variety of volumes. We are comparing the pressures for a variety of functionals with various vdW corrections: PBE-D3, optB86b-vdW, optB88-vdW, SCAN-D3, and SCAN-RVV10. The data are shown in the attached plot with the mean pressure (± 1 standard deviation) for each volume (the plot for PBE-D3-smooth can be ignored).

It seems optB86b-vdW, optB88-vdW, and SCAN-RVV10 are reporting unexpected behavior in the stress tensors. Both optB86b and optB88 seem to be the most "wrong" with negative pressures across the board and an unusual increase in the pressure as the system's volume is expanded (all systems have the same number of particles). SCAN-RVV10 at least follows a believable trend and isn't too far off, so it may just not be appropriate for this system.

We are wondering if the stress tensor calculation is properly implemented for these non-local vdW functionals, since both PBE-D3 and SCAN-D3 are producing believable trends.

For reference, I have included the input files for the 3 problematic functionals I mentioned for the volumes corresponding to V_exp in the attached plots. The versions used for each of those runs are listed below:

• optB86b-vdW: CP2K-8.2 
• optB88-vdW: CP2K-6.1
• SCAN-RVV10: CP2K-2022.2

Any help or insight is appreciated!

Thank you,

Luke Gibson

[Jürg Hutter]

Hi

difficult to say what goes wrong here. No atomic coordinates for further testing available.
But I would suggest to make the following test:
Debug the stress tensor using finite differences and avoid
STRESS_TENSOR DIAGONAL_ANALYTICAL
use
STRESS_TENSOR ANALYTICAL

Use the DEBUG run_type:

RUN_TYPE DEBUG

with

&DEBUG
DEBUG_FORCES .FALSE.
DEBUG_STRESS_TENSOR .TRUE.
DEBUG_DIPOLE .FALSE.
DEBUG_POLARIZABILITY .FALSE.
DE 0.001
STOP_ON_MISMATCH .FALSE.
EPS_NO_ERROR_CHECK 5.e-5
&END

regards
JH

________________________________________
From: cp...@googlegroups.com <cp...@googlegroups.com> on behalf of Luke Gibson <ldgib...@gmail.com>
Sent: Monday, November 28, 2022 4:19 PM
To: cp2k
Subject: [CP2K:18105] Non-physical pressure behavior for non-local vdW functionals

* optB86b-vdW: CP2K-8.2
* optB88-vdW: CP2K-6.1
* SCAN-RVV10: CP2K-2022.2

Any help or insight is appreciated!

Thank you,

Luke Gibson

--
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 on the web visit https://groups.google.com/d/msgid/cp2k/afc955e0-daad-40f4-8bce-87b08f87cb8dn%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/afc955e0-daad-40f4-8bce-87b08f87cb8dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Luke Gibson]

Hi Prof. Hutter,

Thank you for your suggestion. We ran that DEBUG job with the input you provided for optB86b-vdW. In the output, it compares the stress tensor computed numerically and analytically, which showed that the two methods gave the same result (within error). I have attached the output file from that DEBUG job for reference. I have also included an xyz file for the system we are looking at for any testing.

We also discovered that we were running the SCAN-RVV10 calculations without actually including the non-local vdW corrections from RVV10. So in the figure I previously attached, the plot labeled "SCAN-RVV10" was actually showing the pressures from SCAN without any dispersion corrections. Therefore, we only see this non-physical behavior when using non-local vdW functionals and showed that it is not attributed to the method of computing the stress tensor (i.e., numerical vs analytical) based on the DEBUG job we ran.

We appreciate your help in identifying the source of this problem. If you need any other information from us for testing, we are happy to provide it.

Thank you,

Luke

To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe@googlegroups.com<mailto:cp2k+unsubscribe@googlegroups.com>.

[Vyacheslav Bryantsev]

Dear CP2K Developers,

I would like to follow up on unphysical stress tensor from AIMD using the van der Waals density functionals (e.g., optB88-vdW, optB86b-vdW).

The input file we use (attached in the previous emails) works perfectly fine for PBE-D3, SCAN, and SCAN-D3, which implies some problems with stress tensor computations in CP2K using the van der Waals density functionals.

Thank you,

Slava

Vyacheslav Bryantsev

To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com<mailto:cp2k+uns...@googlegroups.com>.