Dear Wannier-ists, I am currently attempting to wannierise a set of isolated valence bands for a spin-polarised system. I have successfully obtained MLWFs for the spin-up channel - these have very low Re/Im ratios and interpolate the band structure very well - but I cannot replicate this for the spin-down channel. For the spin-up channel, I used the SCDM method for initial projections, and ran wannier90 to reach a convergence threshold of 1.0e-10 on the spread functional. Applying the same procedure to the spin-down channel yields MLWFs which are almost as well localised (the total spread is marginally higher) but the Re/Im ratio is very large, suggesting that I am getting stuck in a local minimum.
What I have tried so far: * SCDM + random noise during minimisation * Starting from projections on appropriately-chosen atomic orbitals * Starting from totally random projections In each case, I end up with more or less the same total spread and the same large Re/Im ratios. Perhaps my intuition is incorrect, but this is all the wrong way round in my head. The spin-up channel contains d-bands at similar energy to p-bands, and yet this wannierises very well. The spin-down channel contains only the p-bands (the d-bands are above the Fermi level), and yet I cannot seem to escape the local minimum. Any advice as to how I could further troubleshoot this issue would be much appreciated (and forgive my ignorance if I am missing obvious steps - I am relatively new to the world of Wannier functions). Thanks, Patrick Patrick J. Taylor Doctoral Researcher Morgan Group 1S 1.15, Department of Chemistry, University of Bath, BA2 7AY
_______________________________________________ Wannier mailing list [email protected] https://lists.quantum-espresso.org/mailman/listinfo/wannier
