Hi There,
I am performing a CINEB calculation. I have set
EDIFFG = 1E05
but when the ionic convergence in energy goes below this value the calculation does not stop:
e.g:
1 F= .44618348E+03 E0= .44618348E+03 d E =.446183E+03
2 F= .44618353E+03 E0= .44618353E+03 d E =.468724E04
3 F= .44618356E+03 E0= .44618356E+03 d E =.286150E04
4 F= .44618357E+03 E0= .44618357E+03 d E =.164662E04
5 F= .44618359E+03 E0= .44618359E+03 d E =.211903E04
6 F= .44618362E+03 E0= .44618362E+03 d E =.267459E04
7 F= .44618369E+03 E0= .44618369E+03 d E =.718887E04
8 F= .44618382E+03 E0= .44618382E+03 d E =.129830E03
9 F= .44618397E+03 E0= .44618397E+03 d E =.145161E03
10 F= .44618409E+03 E0= .44618409E+03 d E =.116766E03
11 F= .44618412E+03 E0= .44618412E+03 d E =.333419E04
12 F= .44618417E+03 E0= .44618417E+03 d E =.470692E04
13 F= .44618436E+03 E0= .44618436E+03 d E =.196732E03
14 F= .44618454E+03 E0= .44618454E+03 d E =.179372E03
15 F= .44618460E+03 E0= .44618460E+03 d E =.568035E04
16 F= .44618468E+03 E0= .44618468E+03 d E =.804223E04
17 F= .44618472E+03 E0= .44618472E+03 d E =.416295E04
18 F= .44618474E+03 E0= .44618474E+03 d E =.178415E04
19 F= .44618476E+03 E0= .44618476E+03 d E =.184930E04
20 F= .44618477E+03 E0= .44618477E+03 d E =.131951E04
21 F= .44618485E+03 E0= .44618485E+03 d E =.833533E04
22 F= .44618492E+03 E0= .44618492E+03 d E =.648123E04
23 F= .44618496E+03 E0= .44618496E+03 d E =.410043E04
24 F= .44618501E+03 E0= .44618501E+03 d E =.479206E04
25 F= .44618499E+03 E0= .44618499E+03 d E =0.170913E04
26 F= .44618499E+03 E0= .44618499E+03 d E =.378032E05
27 F= .44618500E+03 E0= .44618500E+03 d E =.522079E05
28 F= .44618504E+03 E0= .44618504E+03 d E =.431281E04
29 F= .44618508E+03 E0= .44618508E+03 d E =.349731E04
30 F= .44618514E+03 E0= .44618514E+03 d E =.599824E04
31 F= .44618522E+03 E0= .44618522E+03 d E =.810415E04
32 F= .44618529E+03 E0= .44618529E+03 d E =.672010E04
33 F= .44618537E+03 E0= .44618537E+03 d E =.849066E04
34 F= .44618545E+03 E0= .44618545E+03 d E =.832445E04
35 F= .44618544E+03 E0= .44618544E+03 d E =0.135523E04
36 F= .44618545E+03 E0= .44618545E+03 d E =.852695E05
37 F= .44618545E+03 E0= .44618545E+03 d E =.600605E06
38 F= .44618545E+03 E0= .44618545E+03 d E =.133630E06
39 F= .44618545E+03 E0= .44618545E+03 d E =.284472E05
40 F= .44618544E+03 E0= .44618544E+03 d E =0.117636E04
41 F= .44618544E+03 E0= .44618544E+03 d E =.430231E05
Any suggestions on how to successfully terminate this calculation would be welcome.
Thanks!
NEB calc doesn't stop when convergence reached,
Moderator: moderators

 Posts: 5
 Joined: Thu Sep 12, 2013 4:18 am
Re: NEB calc doesn't stop when convergence reached,
Check to see if the convergence criterion is met for all of the images in the NEB calculation.
As an aside, it is much better to use a forcebased convergence criterion (EDIFFG<0).
As an aside, it is much better to use a forcebased convergence criterion (EDIFFG<0).

 Posts: 5
 Joined: Thu Sep 12, 2013 4:18 am
Re: NEB calc doesn't stop when convergence reached,
Thank you for your reply.
As suggested I have switched to a forcebased convergence criterion.
I have seen in another thread that you have said: "The true forces on the NEB images will not, in general, drop below EDIFFG. Remember, the images along a minimum energy path climb up over a barrier. It is the NEB forces that matter for convergence.", which very much makes sense and was the reason I had been using EDIFFG > 0.
However when I look at the forces it seems to me that the convergence criterion is based upon the true forces.
For example, in an NEB calc that successfully terminated with EDIFFG = 0.03 the forces on each image are:
1********:
NEB: forces: par spring, perp REAL, dneb 0.008044 0.041039 0.000000
2********:
NEB: forces: par spring, perp REAL, dneb 0.004864 0.054856 0.000000
3********:
NEB: forces: par spring, perp REAL, dneb 0.002813 0.040098 0.000000
4********:
NEB: forces: par spring, perp REAL, dneb 0.001306 0.048815 0.000000
5********:
NEB: forces: par spring, perp REAL, dneb 0.000559 0.026384 0.000000
6********:
NEB: forces: par spring, perp REAL, dneb 0.001107 0.046936 0.000000
7********:
NEB: forces: par spring, perp REAL, dneb 0.002724 0.045223 0.000000
8********:
NEB: forces: par spring, perp REAL, dneb 0.004766 0.066533 0.000000
9********:
NEB: forces: par spring, perp REAL, dneb 0.007901 0.078301 0.000000
I am guessing it is the perp REAL component I should be looking at for convergence here  which is higher than the convergence criterion.
while the true forces have dropped below the criterion.
1********:
FORCES: max atom, RMS 0.013222 0.005639
2********:
FORCES: max atom, RMS 0.021925 0.007426
3********:
FORCES: max atom, RMS 0.028241 0.005420
4********:
FORCES: max atom, RMS 0.021828 0.006585
5********:
FORCES: max atom, RMS 0.014334 0.003558
6********:
FORCES: max atom, RMS 0.021778 0.006331
7********:
FORCES: max atom, RMS 0.029203 0.006109
8********:
FORCES: max atom, RMS 0.026683 0.008994
9********:
FORCES: max atom, RMS 0.029560 0.010612
I wonder if you could tell me which forces I should be looking at for convergence? Also what you would suggest if the forces remain fairly high over the barrier ( ~ 0.1 ev/A)?
Many thanks for your time.
As suggested I have switched to a forcebased convergence criterion.
I have seen in another thread that you have said: "The true forces on the NEB images will not, in general, drop below EDIFFG. Remember, the images along a minimum energy path climb up over a barrier. It is the NEB forces that matter for convergence.", which very much makes sense and was the reason I had been using EDIFFG > 0.
However when I look at the forces it seems to me that the convergence criterion is based upon the true forces.
For example, in an NEB calc that successfully terminated with EDIFFG = 0.03 the forces on each image are:
1********:
NEB: forces: par spring, perp REAL, dneb 0.008044 0.041039 0.000000
2********:
NEB: forces: par spring, perp REAL, dneb 0.004864 0.054856 0.000000
3********:
NEB: forces: par spring, perp REAL, dneb 0.002813 0.040098 0.000000
4********:
NEB: forces: par spring, perp REAL, dneb 0.001306 0.048815 0.000000
5********:
NEB: forces: par spring, perp REAL, dneb 0.000559 0.026384 0.000000
6********:
NEB: forces: par spring, perp REAL, dneb 0.001107 0.046936 0.000000
7********:
NEB: forces: par spring, perp REAL, dneb 0.002724 0.045223 0.000000
8********:
NEB: forces: par spring, perp REAL, dneb 0.004766 0.066533 0.000000
9********:
NEB: forces: par spring, perp REAL, dneb 0.007901 0.078301 0.000000
I am guessing it is the perp REAL component I should be looking at for convergence here  which is higher than the convergence criterion.
while the true forces have dropped below the criterion.
1********:
FORCES: max atom, RMS 0.013222 0.005639
2********:
FORCES: max atom, RMS 0.021925 0.007426
3********:
FORCES: max atom, RMS 0.028241 0.005420
4********:
FORCES: max atom, RMS 0.021828 0.006585
5********:
FORCES: max atom, RMS 0.014334 0.003558
6********:
FORCES: max atom, RMS 0.021778 0.006331
7********:
FORCES: max atom, RMS 0.029203 0.006109
8********:
FORCES: max atom, RMS 0.026683 0.008994
9********:
FORCES: max atom, RMS 0.029560 0.010612
I wonder if you could tell me which forces I should be looking at for convergence? Also what you would suggest if the forces remain fairly high over the barrier ( ~ 0.1 ev/A)?
Many thanks for your time.
Re: NEB calc doesn't stop when convergence reached,
The NEB force is the real force perpendicular to the band (perp real) plus the spring forces parallel to the band (par spring). This NEB force is passed back to vasp and so when you are looking at FORCES: max atom, that is the NEB force. What does not converge, is the total real force which includes the component parallel to the band. If you think about a minimum energy path which has images climbing up a hill, the force down the hill will not get below your convergence criterion, but that is ok, because is the only the real force perpendicular to the path that matters for convergence (and the spring force to keep the images equally spaced).