converging condition

Message

lmiao · #1 Post by **lmiao** » Tue Feb 21, 2006 11:17 pm

Hi,
I was using EDIFFG=-0.05 to as a converging condiction to limit the force under 0.05ev/A when I relaxed my systems. I have seen Admin's previous post saying that 0.01eV/A is a good choice. so does it mean the number I use is too large to decide whether the system is relaxed or not?

Also, after the system is totally relaxed under EDIFFG=-0.05, I did a SCF calculation, but I found that the FORCE on a few atoms are larger than 0.05ev/A. Why this happened? I have tried both PREC=High and Accurate, but some atoms still have forces larger than 0.05, which is the converging condition I used for relaxation. (the total drift is small though) Can someone help answer me this question?

#2 Post by **admin** » Fri Feb 24, 2006 12:43 pm

ad 1) yes.
ad 2) I'm sure you continued from CONTCAR (?)
--) Please have a look whether the electronic structure was really converged for that follow-up step
--) also, if you change the k-mesh (and especially if the k-mesh was not fully converged for the relaxation runs), the forces may change.

lmiao · #3 Post by **lmiao** » Fri Feb 24, 2006 5:16 pm

1) I have seen many published papers where 0.05eV/A is used to relax their systems. of course some other people use 0.01eV/A. but if 0.05 ev/A is considered to be too large, does it mean those who use 0.05 are wrong?

2)good answer. I did change k-mesh.

Thank you

konglt · #4 Post by **konglt** » Fri Feb 24, 2006 7:32 pm

To what extent you want to relax the lattice depends the problem you are studying. If your problem does not require a very high precision, you can loose the criterion. There is no wrong or right for 0.05 eV/A or 0.01 eV/A, it depends on the situation.

lmiao · #5 Post by **lmiao** » Fri Feb 24, 2006 7:42 pm

That was what I am thinking too: the interested force cirterion may depend on the system being studied.