MAX_ITER_DIAG

Ex.1 2 atom cell of GaAs 10 k-points

Fig 1

From this plot, one can see that the value of 5-10 takes approximately the same amount of time. 40 iteration on the other hand takes significantly longer especially as one goes to more k-points.

Ex.2 12 atoms. 3 layers GaAs – 3 layers InAs

Here we can see problems with the Broyden mixing scheme. One must use a maximum of 40 iterations in order to get good convergence. This does not happen for the other mixing schemes.

Ex.3 3 atoms 30 electrons: 3 layers of Ni 3 vacuum in (100)

The big difference is that the 10 Ryd submatrix diagonalization and 5 iterations actually has not converged yet. Overall we do not see much difference between 20 Ryd submatrix diagonalization and 5 iterations and 20 Ryd submatrix diagonalization and 10 iterations.

Ex.4 5 atoms 50 electrons: 5 layers of Ni 5 vacuum in (100)

This plot again points out that there is no advantage in having more that a maximum of 10 iterations.

Ex.5 6 atoms 60 electrons: 6 layers of Ni 6 vacuum in (100): 3 k-points

For this system, we encounter a situation where it seems advantageous to use 5 iterations. This is true provided that one uses 20 Ryd for the submatrix diagonalization. Remember that we saw problems in Ex.3 for 10 Ryd and 5 iterations.

Ex.6 6 atoms 18 electrons: 6 layers of Al 6 vacuum in (100): 3 k-points