Relation of EDIFFG to stress

[dsiegel]

Can someone explain how a particular setting of EDIFFG < 0 (say -1E-2), in conjunction with ISIF = 3, translates into a minimum stress convergence criterion?

In other words, if I have
EDIFFG = -0.01
ISIF = 3
in my INCAR, I know that the ionic relaxation will stop when the forces on all of the atoms are each less than 0.01 eV/Ang. However, with ISIF = 3 the cell shape and volume are also optimized based on the stress tensor, so there must be an equivalent convergence criterion for the components of the stress tensor (that appears to be dependent upon the value of EDIFFG).

So, does anyone know how setting EDIFFG determines the stress convergence criterion? This does not appear to be spelled out in the VASP manual.

Thanks,
Don

[admin]

there is a double check for convergence if ions and cell shape/volume are relaxed: both quantities (the forces on the ions and all stress tensor components) have to be less than |EDIFFG| for the convergence criterium to be fulfilled.

[rgc]

Isn't this check too arbitrary, considering that force and stress are measured in different units?

Is there any reason why a convergence to, e.g., 0.01 of all forces is somehow equivalent to a convergence of 0.01 for the stress? For example, would they lead to similar degrees of convergence of the total energy?

[rgc]

According to your experiences, for a given negative value of EDIFFG, what is typically reached first: stress convergence or force convergence?

[admin]

hmmm, depends on the system. If the Bravais matrix you start from is a good choice for the unit/super cell and the atoms lie in low-symmetry positions, the stress tensors usually converge first. If you give a bad starting guess for the unit cell but the atoms are in high-symmetry positions (fixed by the space group of the lattice), the forces on the atoms will converge faster