Comparison of methods for metals

Systems

6 layers Ni

3 layers Ni - 3 vacuum

MG – Mauri Galli functional GCGM – Grassmann_metal

PK – Pulay_kerker mixing PTF – Pulay_tf mixing

Ex. 1 6 layers of Ni in (100) direction at ideal positions - 2x4x4 k-point mesh

GCGM is the new algorithm to handle the partial occupancies of Ni properly. It is very evident from the plot that the MG functional ( with no inclusion of occupancies) runs into problems around the 4^th SCF cycle. At this system size and its homogeneity, there is little difference between the PK and PTF mixing methods.

Ex. 2 3 layers of Ni in (100) and 3 vacuum at ideal positions - 2x4x4 k-point mesh

Again we see problems with the MG functional. The PK and PTF mixing methods are still comparable.

Ex. 3 5 layers of Ni in (100) and 5 vacuum at ideal positions - 1x2x2 k-point mesh

Ex. 4 6 layers of Al in (100) and 6 vacuum at ideal positions - 2x4x4 k-point mesh

Again we see the benefit of the GCGM algorithm.

Even at this small system, we can see the benefit of the PTF algorithm over the PK method. A maximum of 5 max_iter_diag also seems preferable.

Similar behavior as above.

We see some poor convergence for the PTF algorithm here. A benefit is seen by doing PK mixing for the first 2 steps and from then on doing a PTF mixing. This protocol has been adopted for the pulay_tf algorithm. Further benefit was achieved by modifying the way GCGM algorithm uses the occupations.