increasing precision of calculation

oasd001 · Post by **oasd001** » Thu Mar 02, 2006 11:35 pm

Dear developpers and users,

When running the 'bader' code on a pure vanadium 2x2x2 supercell (16 atom with ideal BCC positions), I get fluctuations in the charge of the vanadium atoms depending on their positions in the supercell. Typical variations are of order 0.05 in 13.00 electrons.
Is there a parameter I can change to decrease the magnitude of these unphysical variations?
thanks in advance for your advice.
Regards,
Olivier.

Post by **graeme** » Fri Mar 03, 2006 1:57 am

Yes, one way to improve the precision is to use a finer charge density grid. You can do this by specifying NGXF, NGYF and NGZF in the INCAR file. I usually look in the OUTCAR or CHGCAR file to see what it is by default, and then multiply the default values by 1.5 or 2.

andri · Post by **andri** » Fri Mar 03, 2006 4:57 pm

A good thing to try a well is to converge the electronic structure using a smaller than usual EDIFF value, something like 1E-8.

Post by **graeme** » Fri Mar 03, 2006 5:46 pm

Andri, have you seen a smaller ediff value (going from a normal value of 1e-5 to to a smaller value of 1e-8) effect the bader charges?

andri · Post by **andri** » Fri Mar 03, 2006 6:57 pm

I can't recall any specific instances but I'm pretty sure that decreasing EDIFF as helped me in the past to reduce noise in the charges.