[CP2K-user] [CP2K:22448] Difficulty converging all electron magnetic system

[Craig Schwartz]

Hi Thomas and everyone,


I'm copying what Keith wrote to one person to everyone so if we make 
progress everyone can learn from it. Since the post, we have migrated to 
git revision 5976d25a45174d28df9c77845b753e0a0585d6d8 pulled July 1st. 
You’re correct that a multiplicity of 21 doesn’t make a lot of sense for 
that geometry. The multiplicity of 21 was chosen as it is the high spin 
ferromagnetically aligned primitive cell of that R32 structure with the one 
Dy+3 and three Fe+3 each having atomic-like moments of 5 uB with the goal 
to relax_multiplicity after converging a high spin state. For the structure 
in the original post, a multiplicity of 61 makes a bit more sense. In this 
folder https://
drive.google.com/file/d/1AhM6gXSYz9xMYKw2YnaOuBnFdwJ7eV97/view?usp=sharing  
you will find the complete run restart files. This folder 
https://drive.google.com/file/d/1ghf1I0cCaviZyXIrUiD88nzipSnrHmpU/view?usp=sharing 
is a primitive cell version of the run with multiplicity of 21. As you can 
see from the output provided, the plateau is basically the scf oscillating 
around ~three points and that behavior occurred regardless of the &mixing 
parameters, different multiplicities, or cell size we selected.


Our end goal for this project is to run Dy L-edge excitation calculations 
on that system (which is why we need an all-electron basis on the Dy). 
Since posting that we found that neither the &xas or &xas_tdp modules work 
with k-points, and that we could converge that high spin solution for the 
primitive cell with no added_mos and a pseudo for every atom but Dy (see 
here 
https://drive.google.com/file/d/1xbgJgz4L1t8qwLQsOCfbulmj37JLzjfm/view?usp=sharing 
for a complete run folder). However, we still have issues as we need to add 
those mos in for &xas_tdp calculations and that the ot algorithm isn’t 
available to converge transition potentials in an &xas calculation. Any 
help you can provide would be greatly appreciated; also relating to your 
non-relativistic guesses point, are you aware of any way to spin-orbit 
split the 2p states ahead of or during an &xas calculation?


Best,

Craig




On Tuesday, October 13, 2026 at 3:38:54 PM UTC-5 Thomas Kühne wrote:

> Dear Craig,
>
> I have examined your input and basis files and run several short 
> diagnostic calculations with recent CP2K development builds.
>
> Your Dy ALL entry has the correct electron count. The conventional cell 
> contains 780 electrons, and MULTIPLICITY 21 gives 
> 400 alpha and 380 beta electrons. However, MULTIPLICITY specifies the 
> spin, not the experimentally measured total magnetic 
> moment. For Dy, the orbital contribution matters, and the scalar-ZORA 
> setup does not include spin–orbit coupling. The measured 
> 7 μB per formula unit therefore does not uniquely determine the 
> appropriate multiplicity.
>
> One concrete source-level observation is that calculate_atomic_orbitals 
> generates a nonrelativistic atomic guess even when the 
> crystal calculation uses scaled ZORA. A diagnostic-only change to use a 
> matching scalar-relativistic atomic guess lowered the 
> initial energy by approximately 450 Hartree in my tests. Nevertheless, it 
> did not produce SCF convergence, so this is not an 
> established explanation or a production-ready fix.
>
> I also tested finer atom-centred grids, locally spin-polarized ionic 
> starting densities, coupled GAPW mixing with GMIX_P, and 
> alternative minimization settings. Unfortunately, none yielded a verified 
> converged solution within the tested runs. I would 
> therefore avoid recommending these parameter combinations as a solution at 
> this stage.
>
> Could you share the .kp restart from near the end of your 2000-step run, 
> together with the complete output and exact Git revision? 
> A file-transfer link would be suitable if the restart is large. This would 
> let me investigate the reported plateau directly. It would also 
> help to know which magnetic arrangement you intend to describe.
>
> Best regards,
> Thomas
>
> > Am 28.05.2026 um 00:58 schrieb Craig Schwartz <craig.s... at UNLV.EDU>:
> > 
> > Hello,
> > I am trying, and struggling, to converge an all electron calculation on 
> Dysprosium ferroborate. An example input and basis set are attached. The 
> system is challenging in that its magnetic and we need all the electrons to 
> do certain x-ray calculations at the heavy element’s low-lying edges (so we 
> can’t just use the MOLOPT and GTH PBE pseudopotentials for instance). The 
> basis sets are all taken directly from basis set exchange (cp2k format) and 
> the ‘ALL’ potential for Dy is roughly a copy of I from the standard cp2k 
> ‘POTENTIAL’ file. 
> > Dy ALLELECTRON ALL
> > 12 24 20 10
> > 0.50000000 0
> > 
> > I’ve been using a 2026.1 release from the git (pulled April 29, 2026) 
> which includes Gaussian smearing, but I’ve also tried Fermi-Dirac smearing 
> as well as the 2025.2 development version. The system is a bit too big for 
> hybrid functionals, and things like PBE have not worked.
> > I’ve messed around with Boyden, Pulay, Kerker and direct mixing and 
> nothing has worked. The best convergence I’ve seen has been with 
> DIRECT_P_MIXING and a small ALPHA 0.1, although even this does not 
> converge. Broyden, Kerker, and Pulay mixing attempts with different values 
> of ALPHA and BETA terminate with fluctuations on the order of E+02. 
> > 1990 P_Mix/Diag. 0.10E+00 80.9 21.05789673 -50737.0152956767 2.02E-01
> > 1991 P_Mix/Diag. 0.10E+00 80.9 12.46218230 -50736.8642879388 1.51E-01
> > 1992 P_Mix/Diag. 0.10E+00 80.8 11.07681641 -50737.0034034056 -1.39E-01
> > 1993 P_Mix/Diag. 0.10E+00 80.8 12.96840862 -50736.7868106099 2.17E-01
> > 1994 P_Mix/Diag. 0.10E+00 80.7 11.69891226 -50736.9641743435 -1.77E-01
> > 1995 P_Mix/Diag. 0.10E+00 80.8 15.49182193 -50737.0700617575 -1.06E-01
> > 1996 P_Mix/Diag. 0.10E+00 80.9 20.70203774 -50737.1124326006 -4.24E-02
> > 1997 P_Mix/Diag. 0.10E+00 80.8 14.68011752 -50737.0046223170 1.08E-01
> > 1998 P_Mix/Diag. 0.10E+00 80.7 11.80777138 -50737.1114555977 -1.07E-01
> > 1999 P_Mix/Diag. 0.10E+00 80.7 14.55452265 -50736.6616467933 4.50E-01
> > 2000 P_Mix/Diag. 0.10E+00 80.9 10.58192107 -50736.8174171267 -1.56E-01
> > 
> > Leaving inner SCF loop after reaching 2000 steps.
> > 
> > Experimentally this structure has a net magnetization of 7 bohr magnetor 
> per formula unit when in field, so I’ve been using both an unspecified 
> multiplicity low-spin initial guess and a high spin initial guess that is 
> close to per 7 bohr mangeton/formula unit (i.e. the Dy f electrons align 
> with the field). 
> > Does anyone have any suggestions on how to converge this problem? It’s 
> annoying because of the combination of all electron, magnetization, and 
> rare-earth f-electrons that aren’t in the pseudopotential. This is the 
> standard R-centered cell for this structure, but I’ve also tried the 
> primitive cell with a denser k-grid and have had similar issues.
> > Thanks for any help,
> > Craig
> > 
> > -- 
> > 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... at googlegroups.com.
> > To view this discussion visit 
> https://groups.google.com/d/msgid/cp2k/903bc731-5df3-4bc1-91b8-62b2620f8797n%40googlegroups.com
> .
> > <BASIS_CCPVDZ-DK3><BASIS_CCPVDZ><test.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/04a529f1-27d2-4791-bcd5-b35220b69aecn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20261019/bf8ea5fd/attachment.htm>