[CP2K-user] [CP2K:22076] Suggestions for Improving SCF Convergence in CP2K‘s Diagonalization and K-Point Calculations

[yis...@163.com]

Dear CP2K Developers,

Hello!

         I  am a beginner in the field of theoretical calculations. My main 
background is in experimental science, and I have taught myself the 
computational aspects. CP2K is an excellent first-principles calculation 
software, and its computational speed is very impressive—whether dealing 
with large systems or small unit cells and k-point calculations, it 
performs much faster than other software like VASP.

         However, I have noticed that compared to VASP, when CP2K uses 
diagonalization methods for k-point calculations, the convergence of the 
self-consistent field (SCF) iterations is relatively slow, especially for 
magnetic systems. VASP offers multiple strategies to improve convergence in 
complex magnetic systems, such as:
1）Performing a non-magnetic calculation first, then using the CHGCAR file 
as an initial guess for the magnetic calculation;
2）Using the NELMDL keyword to set a delay in the SCF steps to promote 
convergence;
3）Selecting the RMM-DIIS or DAVIDSON algorithm via the ALGO keyword for SCF 
iterations.
     Unfortunately, similar flexible options are currently lacking in CP2K.

             Of course, the OT (orbital transformation) algorithm in CP2K 
is excellent, but it is mainly suitable for large cell systems, while for 
small cells the computational cost remains high. Therefore, optimizing the 
SCF convergence efficiency in diagonalization methods and k-point 
calculations would significantly enhance the user experience and make CP2K 
more advantageous in a wider range of research scenarios.

            I would like to kindly request the development team to consider 
improving and optimizing the relevant algorithms in future versions to 
enhance convergence stability and speed in various systems. With progress 
in these areas, CP2K would undoubtedly become the preferred tool for more 
computational researchers and play an even greater role in broader 
applications.

            Thank you for your continuous contributions to the scientific 
computing community with such outstanding software!

Best regards,
A CP2K User

-- 
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/84aabbf8-dce2-43b4-bf3a-c64205aaf7dan%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260202/31149c09/attachment.htm>