[charged calculations on metal oxides]

[Miguel RP]

Hi everyone,

I’m trying to run some charged CP2K calculations for an optimal-tuning test of a CAM-like range-separated hybrid functional on isolated TiO2 nanoclusters.

The neutral CHARGE 0 calculations converge fine, but the charged systems (CHARGE +1  (-1), UKS .TRUE., MULTIPLICITY 2) are very unstable and do not converge. The SCF energy keeps jumping by several Hartree even after many iterations.

The setup is a 100% long-range CAM-like functional, using for example:

SCALE_COULOMB 0.19
SCALE_LONGRANGE 0.81
OMEGA 0.15

I also generated a neutral UKS singlet restart file (CHARGE 0, UKS .TRUE., MULTIPLICITY 1) and used it to restart the charged calculations, which solved the previous nspin mismatch issue. However, the SCF itself remains unstable.

Has anyone dealt with charged open-shell hybrid calculations in CP2K for this type of systems? Any suggestions for a more stable SCF protocol, or whether this may indicate that the extra electron/hole is not well localized/bound in this setup?

So far, I have tried / am trying:
- restarting charged UKS doublets from neutral UKS singlet .wfn
- diagonalization + Broyden mixing
- lowering ALPHA in the mixing

Thanks!