Relaxation not stopping when EDIFFG is reached.

[jonathan_schmidt]

I was running the attached relaxation and using grep -A 23 TOTAL-FORCE OUTCAR it seemed that the forces were already quite a bit smaller than EDIFFG = -0.0001 without the relaxation stopping.
Anything I can do to change this or any setting is wrong? It's seems to just be wasting computing time.
thank you so much.

e.g.
POSITION TOTAL-FORCE (eV/Angst)
-----------------------------------------------------------------------------------
0.00000 -0.00000 1.95875 -0.000000 0.000000 0.000000
2.73442 2.73442 1.95875 -0.000000 0.000000 -0.000000
2.73442 2.73442 5.87624 -0.000000 0.000000 -0.000000
0.00000 -0.00000 5.87624 -0.000000 0.000000 -0.000000
2.73442 0.00000 3.91750 -0.000000 -0.000000 0.000000
0.00000 2.73442 3.91750 -0.000000 0.000000 0.000000
0.00000 2.73442 7.83499 -0.000000 -0.000000 0.000000
2.73442 0.00000 7.83499 -0.000000 0.000000 0.000000
2.73442 0.00000 1.95875 -0.000000 0.000000 -0.000000
0.00000 2.73442 1.95875 -0.000000 0.000000 0.000000
3.90368 3.90368 0.00000 0.000002 0.000002 0.000000
4.29958 4.29958 3.91749 -0.000002 -0.000002 0.000000
1.56516 3.90368 3.91749 -0.000002 0.000002 0.000000
1.16926 4.29958 0.00000 0.000002 -0.000002 0.000000
0.00000 2.73442 5.87624 -0.000000 0.000000 -0.000000
2.73442 0.00000 5.87624 -0.000000 0.000000 0.000000
1.16926 1.16926 3.91749 0.000002 0.000002 0.000000
1.56516 1.56516 0.00000 -0.000002 -0.000002 0.000000
4.29958 1.16926 0.00000 -0.000002 0.000002 0.000000
3.90368 1.56516 3.91749 0.000002 -0.000002 0.000000
-----------------------------------------------------------------------------------
total drift: 0.000000 0.000000 0.000000

[manuel_engel1]

The settings are correct. There is nothing wrong with your calculation per se. However, the EDIFFG setting is a bit more complicated than the current VASP wiki entry would lead you to believe. For ISIF=3, the cell shape/volume is allowed to change. In this case, the break condition specified by EDIFFG also includes the stress tensor, not just the forces. This is not properly documented online and we are sorry for the confusion. We are currently exploring possibilities of how to improve the situation.

[jonathan_schmidt]

Thank you very much for the reply, do you have some stopping equation in terms of stress components and EDIFFG that you can share.