[CP2K-user] [CP2K:22251] Calcite crystal RPBE force convergence

[Thomas Kühne]

Dear Kit,

I agree with your interpretation. The important point is that the residual is now almost entirely parallel to the applied periodic field. In your z-field calculation,

FORCES| Sum -6.11045615E-09 7.87254948E-09 2.04962359E-04

the x and y components remain at the zero-field accuracy, whereas the z component is of the order of the applied field strength. This strongly suggests that 
we are no longer looking at the same XC/grid egg-box issue that was cured by GAPW_XC + GAPW_ACCURATE_XCINT. Rather, the remaining term seems 
to originate from the finite-field/Berry-phase force contribution itself.

For a neutral periodic system I would still expect the total force to vanish, also in a homogeneous Berry-phase electric field, up to numerical noise. The 
residual is also not simply a physical net-charge force. The system is neutral in terms of the GTH valence charges, and the residual is approximately 0.82 
times the field strength rather than an obvious integer-charge contribution.

I do not think there is a Wannier-localization setting that should fix this. The PERIODIC_EFIELD implementation uses the Berry-phase formulation directly; 
the printed Wannier/localization controls are not convergence parameters for this force term. Likewise, increasing the XC fine grid should only affect the 
ordinary XC/grid part. Since the transverse components are already small, and the residual tracks the field direction, I would not expect FINE_GRID_FACTOR 
alone to remove it completely.

Could you please do two quick checks, if convenient?

1. Repeat the same calculation with the field reversed, i.e. INTENSITY -2.5e-4 or POLARISATION 0 0 -1.
2. Repeat with half the field strength, e.g. 1.25e-4 a.u.

If the force-sum residual changes sign with the field and scales linearly with its magnitude, that would be very strong evidence that the issue is in the periodic 
electric-field force contribution rather than in the GAPW/XC integration.

At first sight, finite-field force code path looks like a missing or imperfect cancellation between the ionic and electronic Berry-phase force terms under a uniform 
translation. In other words, it may be a real CP2K bug or at least a limitation of the present PERIODIC_EFIELD force implementation, rather than an 
input-convergence issue.

As a practical workaround for testing internal forces, subtracting the average force from all atoms would remove the spurious center-of-mass acceleration, but 
I would not present that as a formally clean solution until we understand the origin of the term.

Best wishes,
Thomas

> Am 26.05.2026 um 16:35 schrieb 'Kit Joll' via cp2k <cp2k at googlegroups.com>:
> 
> Dear Thomas,
> 
> Thank you again for your help. I was able to reproduce your zero-field results to approximately the same accuracy with a master version cloned on Friday 22nd May. I presume the remaining small differences are due to build/library differences.
> 
> However, while the zero-field force sum is now well converged, I am not seeing the same behaviour once a finite periodic electric field is applied.
> 
> For example, using
> 
> METHOD GAPW_XC
> GAPW_ACCURATE_XCINT
> USE_FINER_GRID T
> FINE_GRID_FACTOR 2.0
> 
> together with a periodic Berry-phase electric field of intensity 2.5e-4 a.u. along z, the force sum becomes
> 
> FORCES| Sum -6.11045615E-09 7.87254948E-09 2.04962359E-04
> FORCES| Total atomic force 2.04962360E-04
> 
> So the transverse components remain well converged, but the component along the applied field direction returns to the order of 1e-4 Ha/Bohr. I see similar behaviour when applying the field along other directions. The issue also persists when increasing FINE_GRID_FACTOR from 2 to 8: the force-sum error remains orders of magnitude larger along the field axis (however, all components sum do decrease by ~1 order of magnitude relative to FINE_GRID_FACTOR 2). 
> 
> Do you know whether there are additional grid, Berry-phase, or Wannier-related settings that need to be converged for finite-field force calculations? Or is there another setting that should be used together with GAPW_XC and GAPW_ACCURATE_XCINT in the presence of a periodic electric field?
> 
> Please find attached a minimal input (start.inp) /output (output.out) pair that reproduces the issue.
> 
> Thank you again for your help and advice.
> 
> All the best,
> Kit
> 
> On Thursday, May 21, 2026 at 10:48:00 AM UTC+1 Kit Joll wrote:
>> Dear Thomas,
>> 
>> Many thanks for the very quick and detailed reply! This is extremely helpful, and your diagnosis makes a lot of sense given the behaviour we were seeing.
>> 
>> I will try compiling the current CP2K master using the new make_cp2k.sh workflow, run the relevant regression tests, and then first reproduce your zero-field force-sum results on the same calcite frame. After that, I will check whether the same settings also resolve the sensitivity we observe in the finite-field response tensors.
>> 
>> Thank you also for pointing out the PRINT_LEVEL LOW workaround and the relevant GAPW_XC / GAPW_ACCURATE_XCINT settings. I will report back once I have tested these on our side.
>> 
>> 
>> Best wishes,
>> Kit
>> 
>> 
>> On Wednesday, May 20, 2026 at 9:00:47 PM UTC+1 Thomas Kühne wrote:
>>> Dear Kit,
>>> Thank you for the detailed report and for the input/output files. I was able to
>>> reproduce the force-sum residual for the provided calcite frame:
>>> GPW baseline:
>>>   Sum F = ( 3.48675295E-05, 1.12518883E-04, -5.39277596E-05 ) Ha/Bohr
>>>   |Sum F| = 1.29554803E-04 Ha/Bohr
>>> The behavior originate from the XC/grid treatment:
>>> GPW + USE_FINER_GRID T, FINE_GRID_FACTOR 2.0:
>>>   |Sum F| = 2.26659720E-05 Ha/Bohr
>>> 
>>> GPW + finer/smoothed XC grid:
>>>   |Sum F| = 1.10078684E-06 Ha/Bohr
>>> 
>>> GAPW_XC:
>>>   |Sum F| = 4.20880179E-05 Ha/Bohr
>>> 
>>> GAPW_XC + GAPW_ACCURATE_XCINT:
>>>   |Sum F| = 1.68750841E-07 Ha/Bohr
>>> 
>>> GAPW + GAPW_ACCURATE_XCINT + USE_FINER_GRID T, FINE_GRID_FACTOR 2.0:
>>>   |Sum F| = 3.85198029E-08 Ha/Bohr
>>> 
>>> GAPW_XC + GAPW_ACCURATE_XCINT + USE_FINER_GRID T, FINE_GRID_FACTOR 2.0:
>>>   |Sum F| = 1.04105369E-08 Ha/Bohr
>>> This strongly suggests that the residual is dominated by an egg-box/grid-coupling
>>> effect in the XC/grid integration rather than by pseudopotentials, basis sets, or
>>> SCF noise.
>>> For this system, the best tested setting was:
>>> &QS
>>>   EPS_DEFAULT 1.0E-18
>>>   METHOD GAPW_XC
>>>   GAPW_ACCURATE_XCINT
>>> &END QS
>>> 
>>> &XC
>>>   DENSITY_CUTOFF  1.0E-10
>>>   GRADIENT_CUTOFF 1.0E-10
>>>   TAU_CUTOFF      1.0E-10
>>>   &XC_GRID
>>>     FINE_GRID_FACTOR 2.0
>>>     USE_FINER_GRID T
>>>   &END XC_GRID
>>>   &XC_FUNCTIONAL NO_SHORTCUT
>>>     &GGA_C_PBE
>>>     &END GGA_C_PBE
>>>     &GGA_X_RPBE
>>>     &END GGA_X_RPBE
>>>   &END XC_FUNCTIONAL
>>> &END XC
>>> METHOD GAPW with the same GAPW_ACCURATE_XCINT and fine XC grid also performs
>>> very well, but in this test GAPW_XC was slightly better and should be the less
>>> intrusive change relative to the original GPW calculation.
>>> One practical note: GAPW_ACCURATE_XCINT has been improved recently, so I would
>>> recommend testing with a recent CP2K master or a release that includes those
>>> changes. Also, for this calcite input current master hit a post-SCF output crash
>>> with PRINT_LEVEL MEDIUM; using PRINT_LEVEL LOW while keeping explicit
>>> &FORCES ON avoided that and still printed the force block.
>>> Best wishes,
>>> Thomas
> 
> 
> -- 
> 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+unsubscribe at googlegroups.com <mailto:cp2k+unsubscribe at googlegroups.com>.
> To view this discussion visit https://groups.google.com/d/msgid/cp2k/5caec0c1-cdad-455b-815f-dd7f993d0230n%40googlegroups.com <https://groups.google.com/d/msgid/cp2k/5caec0c1-cdad-455b-815f-dd7f993d0230n%40googlegroups.com?utm_medium=email&utm_source=footer>.
> <output.out><start.inp>

-- 
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+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/49D23516-73C9-47CC-BFF6-A1250A7F1BAD%40gmail.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260526/d4aec9a4/attachment-0001.htm>