About SCF convergence：Can first perform a pre-optimization using a larger EPS_SCF?

[Charlotte Lai]

Dear CP2K community,

While optimizing the structure of a triangular MoS₂ monolayer, I find that the SCF convergence cannot be pushed below the originally set EPS_SCF = 1 × 10⁻⁶. Could experienced users advise whether it is acceptable to first carry out a coarse geometry optimization with EPS_SCF = 5 × 10⁻⁶, and then restart the calculation on the resulting structure with the tighter criterion EPS_SCF = 1 × 10⁻⁶ for a more accurate refinement?

Thank you in advance for any suggestions.

[Konstantin Tokarev]

среда, 3 сентября 2025 г. в 17:14:05 UTC+3, Charlotte Lai:

Dear CP2K community,

While optimizing the structure of a triangular MoS₂ monolayer, I find that the SCF convergence cannot be pushed below the originally set EPS_SCF = 1 × 10⁻⁶. Could experienced users advise whether it is acceptable to first carry out a coarse geometry optimization with EPS_SCF = 5 × 10⁻⁶, and then restart the calculation on the resulting structure with the tighter criterion EPS_SCF = 1 × 10⁻⁶ for a more accurate refinement?

Thank you in advance for any suggestions.

In my opinion, EPS_SCF 1E-6 is already very coarse (I normally use 1E-8). If you are having convergence troubles with 1E-6, it likely indicates numeric instability in SCF, which, for example, may be caused by basis set being too diffuse for the system under investigation.