Suggestions for Improving SCF Convergence in CP2K‘s Diagonalization and K-Point Calculations
29 views
Skip to first unread message
yis...@163.com
Feb 2, 2026, 8:24:25 AM (yesterday) Feb 2
to cp2k
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
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
Reply all
Reply to author
Forward
0 new messages