Page 1 of 1

ionic minimization with force based optimizers

Posted: Tue Jul 27, 2021 12:26 pm
by tg459335
My system consists of ZnGaO quantum dot (672 atoms) passivated with pseudo hydrogen atoms (H.5 for O and H1.5 for Zn) with 25% Zn atoms replaced with Ga.The H atoms were fixed with selective dynamics. I tried to minimize first with IBRION = 1 and IBRION=2 found that IBRION=1 optimizes better. However, near the potential minimum, it keeps on oscillating despite reducing EDIFFG to -.001 eV/A. So, I copied the CONTCAR of a configuration that gives the minimum energy to POSCAR and applied IBRION=3 with IOPT =7 and EDIFFG = -0.01. The electronic self consistent loop after the first ionic run is just two iterations with EDIFF= .0001. I ran the optimization for NSW=15 to see how it goes, and it does not converge despite the total energies being almost identical and the total drift due to forces being reasonable. Here is a summary of the output:

1 F= -.17080555E+04 E0= -.17079025E+04 d E =-.170806E+04
2 F= -.17080554E+04 E0= -.17079027E+04 d E =0.386897E-04
3 F= -.17080554E+04 E0= -.17079026E+04 d E =0.111773E-04
4 F= -.17080554E+04 E0= -.17079025E+04 d E =0.580936E-05
5 F= -.17080554E+04 E0= -.17079025E+04 d E =-.961358E-06
6 F= -.17080554E+04 E0= -.17079024E+04 d E =0.188628E-05
7 F= -.17080554E+04 E0= -.17079025E+04 d E =-.168919E-06
8 F= -.17080554E+04 E0= -.17079024E+04 d E =0.824220E-06
9 F= -.17080554E+04 E0= -.17079025E+04 d E =0.143074E-06
10 F= -.17080554E+04 E0= -.17079024E+04 d E =0.306754E-06
11 F= -.17080554E+04 E0= -.17079024E+04 d E =0.136439E-06
12 F= -.17080554E+04 E0= -.17079024E+04 d E =0.250293E-07
13 F= -.17080554E+04 E0= -.17079024E+04 d E =0.143191E-07
14 F= -.17080554E+04 E0= -.17079024E+04 d E =-.142027E-07
15 F= -.17080554E+04 E0= -.17079024E+04 d E =-.291038E-08

and

total drift: -0.003656 -0.015646 0.007522
total drift: -0.001932 -0.065164 -0.052542
total drift: -0.002071 -0.030570 -0.027956
total drift: 0.008793 0.007088 0.000588
total drift: 0.006038 -0.006442 -0.003008
total drift: 0.016039 0.022247 0.011494
total drift: 0.010947 0.006623 0.005820
total drift: 0.018981 0.027611 0.015146
total drift: 0.013966 0.013287 0.010150
total drift: 0.019687 0.027809 0.016072
total drift: 0.020843 0.030094 0.017100
total drift: 0.019713 0.027501 0.016195
total drift: 0.019808 0.027812 0.016385
total drift: 0.019732 0.027440 0.016189
total drift: 0.019776 0.027526 0.016226

Does this make sense? The EDIFF criteria has been more than met (the energies do not correspond to forces according to your description?). The total drift term, for instance, the fourth one seems to satisfy .01 eV/A.

Does this seem okay or shall I increase NSW and hope it converges?

Re: ionic minimization with force based optimizers

Posted: Tue Jul 27, 2021 12:55 pm
by graeme
Well, I would need to see the forces. The drift is the sum of the forces, which is irrelevant to the EDIFFG convergence criterion. If you look at the forces, I expect they will be above 0.01 eV/A and that you will need to continue optimization.

Note that IBRION=1 is a second order optimizer which relies on curvatures from differences in the forces. For that method to work well, you need accurate forces. Your value of EDIFF=1e-4 is likely insufficient. Try EDIFF=1e-6 and I expect IBRION=1 will converge nicely.

If you remain stuck, you can upload your input files and I should be able to find settings which give smooth and rapid convergence.

Re: ionic minimization with force based optimizers

Posted: Tue Jul 27, 2021 4:16 pm
by tg459335
Thanks. I did try EDIFF = 1e-6 and it did not converge. I will try again and get back to you if it doesn't converge.