relaxation of structure EDIFFG!!!!!!!

Message

ouarda · #1 Post by **ouarda** » Mon Dec 11, 2006 3:29 pm

Dear all
if we set EDIFFG to his default value in incar (EDIFF*10). so my question is what the corresponding tolerance for the forces in the unit cell.
thanks for your time

#2 Post by **admin** » Tue Dec 12, 2006 1:39 pm

there is no criterium on the forces at all if EDIFFG>0, only the differences in total energy are checked in that case.

ouarda · #3 Post by **ouarda** » Tue Dec 12, 2006 2:38 pm

thanks admin for reply. but what is better setting in order to save time computting, EDIFFG>0 or EDIFFG< 0 ?
thanks

tjf · #4 Post by **tjf** » Tue Dec 12, 2006 3:28 pm

[quote="'smallblacktext'>[ Edited Tue Dec 12 2006, 06:51PM "]</span>

graeme · #5 Post by **graeme** » Tue Dec 12, 2006 6:09 pm

I would like to add a vote for requiring the use of the force-based criteria (ediff<0) in publications. The energy-based criterion (ediff>0) is not reproducible in the sense that it is based upon the details of the optimizer and the optimizer path (starting configuration). For example, a slow optimizer that takes small steps will converge at a larger force than an aggressive optimizer. The force-based convergence criterion, on the other hand, is well defined, and independent of the optimizer details.

ouarda · #6 Post by **ouarda** » Thu Dec 14, 2006 9:50 am

thanks for response and discussion, in my opinion I think the setting EDFFG<0 is better when the study concerns the relaxation of system in transition state ( diffusion energy....... etc), i.e for neb calculation. but for other local minumum, its better to set EDFFG>0.
so its still my opinion.
thanks

tjf · #7 Post by **tjf** » Thu Dec 14, 2006 9:55 am

Hi ouarda

On what do you base that opinion?