Underestimation of SOC splittings

[Nisarg Trivedi]

Dear CP2K community,

I have been trying to benchmark/test the spin orbit coupling capabilities in CP2K, by measuring the spin orbit splitting of the valence bands of Silicon at the gamma point. 
These calculations were performed on CP2K v2025.1 using the GTH_SOC_POTENTIALS and TZV2P MOLOPT basis.

The values I get are much smaller than the reported theoretical literature and experimental values. From this run, CP2K gives 22 meV of splitting. I have attached the input file (si.inp) and the output bandstructure file (bandstructure_si). 

The same structure ran through QE with fully relativistic pseudopotentials gives 48 meV of splitting; which is closer to the experimental value of ~43 meV. 

It would really be helpful if someone can clear it up if I have made a mistake in the input/calculation or this is a known limitation of the pseudopotential/method used in CP2K?

Best regards,

Nisarg.

[BIBEK SAMAL]

Hi Nisarg,

I would suggest switching to a newer version of CP2K. There have been several fixes and updates to the SOC implementation since v2025.1.

Please note that the SOC section of the input has also changed slightly in the newer versions; the corresponding syntax and options are described in the current CP2K manual.

With the updated implementation, you should be able to obtain a spin-orbit splitting much closer to the expected value.

Best regards,
Bibek