Electronic convergence difficulties for large slab calculations

[jderiseis]

Hello!

I am finalizing and adsorption/incorporation study of adatoms on different surface facets of GaAs and am having difficulties achieving electronic convergence with LREAL = .FALSE. and EDIFF =1E-8. I need accurate energies since I am doing lots of comparisons across facets, with different adatoms, and with different bulk references.

My current procedure is as follows: 1) Converge wavefunction energies to <1E-6 eV with Auto projection scheme and somewhat hot Gaussian smearing. 2) Use converged wavefunctions in calculations with LREAL=.FALSE., ISMEAR = -5, and EDIFF =1E-8 to get final single point energies. This second step is the issue, causing all but one of my calculations to hit the (generous) wall time before convergence. What can I do to improve convergence speed without sacrificing accuracy.

SCF_troubleshooting.zip

I've attached an example of one of the jobs that almost reached convergence as well as all inputs and both sets of INCARs.

[alexey.tal]

Dear jderiseis,

Thank you for your question.
It is hard to say what will work for certain in this case as it can be very much system dependent. However, here are some things you could try:

• enable the symmetry to speed up the calculation (ISYM=2)
• reduce the number of steps stored in the mixer or use the default setting (MAXMIX)
• restart from WAVECAR instead of CHGCAR (ISTART=1 ; ICHARG=0)
• use RMM-DIIS to speed up the calculation (ALGO=Fast)
• use linear mixing by setting BMIX=0.0001 and BMIX_MAG=0.0001.
• restart from a converged calculation with LDIPOL=.FALSE.

You can find our guide on how to improve convergence here.

Best wishes,
Alexey