k-points convergence for spin polarised vs non-polarised DFT calculations

Spin-polarised calculations come at a computational cost and I tend to take ISPIN = 2 only after I'm done with a proper initial relaxation of the structure, so the k-mesh, ENCUT, and $$\sigma$$ are already converged before magnetism is accounted for. Is this approach alright?

Does spin-coupling affect the k-mesh (or $$\sigma$$) convergence?

2 Answers

If you compare calculations with and without spin polarization and your system is magnetic, the orbitals that are occupied will change. Strictly speaking, this means that everything would need to be rechecked for convergence.

In practice, the convergence threshold for the energy cutoff and k-points will probably be close and there’s no way to know without testing whether the spin polarized version would require more k-points or less k-points than the calculations which are not spin polarized.

I would suggest that you do your convergence testing with the calculation settings you are using for your production calculations.

1 Spin polarization needs to be open (ISPIN=2) at structure relaxation.

2 K-mesh can be generated with VASPKIT according to your calculation accuracy.

3 Spin-polarization/spin-orbit coupling has no influence on K-mesh/smearing convergence. You can test these tags with scalar calculation.