About static calculcation for stiffness-bond length approach

Alloy Theoretic Automated Toolkit (ATAT)
1

Hi all,

I have a question about static calculation in stiffness-bond length approach used for fitsvsl code. In order to do this, I generate supercell containing >35 atoms and use this INCAR:
[INCAR]
encut = 500
IREAL = Auto
KPPRA = 2000
POTUSE = PAWPBE

in this case, the default NSW should be 0 and EDIFF = 1e-4. However it takes a long long time to run (more than 24 hours). Note that my supercell contains 35 atoms becuase I set fitsvsl -er=9 -dr=0.2 -ns=2 -ms=0.01. I see it is said in manual that this calculation should be very fast and now I suspect where goes wrong. Do I need to lower the encut or kpoints? I am testing that but wonder whether this will influence the accuracy of force.

2

Supercell calculations do take a while (relative to the simple structures used for maps).
You can indeed speed them up a bit by lowering ENCUT and KPOINTS, after testing how much of an effect it has. Keep in mind that differences in free energies are what matters, not absolute values. So you can exploit systematic error cancellations to reduce the accuracy without too much negative impact.
Sorry that I am not telling anything you don’t already know! Your approach is valid.