Page 1 of 1

: "EDIFF and EDIFFG values for relaxing Ti29Co4O66 system (99 atoms) with ISIF=2"

Posted: Tue Oct 01, 2024 10:30 am
by danish_javed

I’m working on a structural relaxation for a Ti29Co4O66 nanocluster containing 99 atoms. My goal is to relax only the atomic positions while keeping the cell fixed, so I’m using the ISIF=2 setting. However, the calculation is taking a long time and has not yet converged.
System composition: Ti29Co4O66 (99 atoms)
Relaxation: Only atomic positions (ISIF=2)
INCAR settings:
EDIFF = 1e-6
EDIFFG = -1e-4
The calculation has been running for a significant amount of time, and convergence hasn’t been reached.
Are the EDIFF and EDIFFG values appropriate for a system of this size?

Should I consider adjusting them, especially given the size of the system (99 atoms)?
Any suggestions for improving convergence or speeding up the calculation? Should I modify other settings like POTIM, ALGO, or any other INCAR parameters to enhance performance?


Re: : "EDIFF and EDIFFG values for relaxing Ti29Co4O66 system (99 atoms) with ISIF=2"

Posted: Tue Oct 01, 2024 11:56 am
by martin.schlipf

In general, this depends on the quantity you are interested in. You should always start from a simpler calculation and measure the quantity of interest (say the bandgap). Continue choosing tighter convergence parameters until that quantity does not change anymore. In your example, you may start with a more lenient EDIFF/EDIFFG value, relax and observe your quantity of interest. Then you tighten EDIFF/EDIFFG and relax again. You could also add selective dynamics to reduce the search space and then slowly increase the number of atoms that is allowed to move.

With these general disclaimers out of the way, EDIFFG of -1e-4 is very tight. Keep in mind that the units of EDIFFG are force units (eV/A) if you set a negative value. So this is much tighter than a EDIFFG = 1e-4 because ions typically don't move a whole Angstrom. Also remember that EDIFF is an extensive quantity, so if you double the size of the cell EDIFF should typically increase by a factor of 2 unless you are interested in an intensive quantity like defect formation energy.