scf convergence error with large systems
41 views
Skip to first unread message
yaoca...@gmail.com
Aug 13, 2025, 3:30:55 AMAug 13
to cp2k
Dear cp2k developers and users:
I find that when I was calculating a large system, in this example, 744 atoms, the scf convergence always failed, no matter how I tried with cutoff, minimizer ...
However, when I use a smaller system (same elements but fewer atoms), the convergence is fine.
My input file and geometry file are pasted below.
Please help me if you have any experience with similar problems, thank you!
Que Zhu
Aug 13, 2025, 3:43:03 AMAug 13
to cp...@googlegroups.com
sorry,I can not open the geom.cell file.Can you find me the cif file and output file ,such log or cp2k.out?
--
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 visit https://groups.google.com/d/msgid/cp2k/eae35ff0-312a-46c8-bb6d-dbcbadf37155n%40googlegroups.com.
Message has been deleted
Konstantin Tokarev
Aug 13, 2025, 4:25:49 AMAug 13
to cp2k
>I find that when I was calculating a large system, in this example,
744 atoms, the scf convergence always failed, no matter how I tried
with cutoff, minimizer ...
>However, when I use a smaller system (same elements but fewer atoms), the convergence is fine.
Such thing is often a sign of (near) linear dependence in the basis set, which is very common in large systems. Everything converges perfectly fine, than you extend your molecular cluster, or increase supercell, and it suddenly crumbles. If reducing the system is not an option, there are a few other ways to proceed:
1) try less diffuse basis set — probably not an option as you are using SR variant already;
2) try smaller basis set — for example, there is SZV-MOLOPT-SR-GTH, or try something from a different basis set family;
3) filter basis set by eigenvalues of overlap matrix — I'm not exactly sure how does it work in CP2K, but changing SCF/EPS_EIGVAL might help. Here is a documentation for a similar feature in ORCA: https://www.faccts.de/docs/orca/6.1/manual/contents/essentialelements/basisset.html#sec-essentialelements-lindep
1) try less diffuse basis set — probably not an option as you are using SR variant already;
2) try smaller basis set — for example, there is SZV-MOLOPT-SR-GTH, or try something from a different basis set family;
3) filter basis set by eigenvalues of overlap matrix — I'm not exactly sure how does it work in CP2K, but changing SCF/EPS_EIGVAL might help. Here is a documentation for a similar feature in ORCA: https://www.faccts.de/docs/orca/6.1/manual/contents/essentialelements/basisset.html#sec-essentialelements-lindep
среда, 13 августа 2025 г. в 10:43:03 UTC+3, Que Zhu:
sorry,I can not open the geom.cell file.Can you find me the cif file
FWIW, it's just XYZ file without number of atoms. It's @INCLUDE'd in NiO-perovskite.inp in COORD section.
Reply all
Reply to author
Forward
0 new messages