the system is in chaos

[weiyidan]

Dear professor,
I am doing the movement of silicon vacancy. However, it seems systems are chaos during the movement (see in 01 02 03) and forces are hard to converge. Can you help me to tell me the reason?
Thanks for your help and support.
Yidan Wei

[graeme]

Your initial structures are completely crazy. Start with a nice smooth path and use a conservative optimizer such as IBRION=3, POTIM=1 until the forces drop below 1 eV/Ang. Check then that the path makes sense. After that, you can try a more aggressive optimizer to increase the rate of convergence.

If you look in this forum, you can see that I have given this advice many many times, and provided examples where I redid peoples calculations in this way and seen smooth convergence to sensible paths. See if you can follow one of those examples and get to the bottom of this yourself. If you try and really feel that there is something strange going on, I'll take a look.

[weiyidan]

Dear professor,
Thanks for your help.
I follow your advice today,here are forces of images,
$ ./nebef.pl
0 0.017443 -1139.628900 0.000000
1 1.104287 -1137.882100 1.746800
2 0.935224 -1137.566200 2.062700
3 0.810268 -1138.500700 1.128200
4 0.018839 -1139.870200 -0.241300
$ ./nebef.pl
0 0.017443 -1139.628900 0.000000
1 9.026600 -1130.582600 9.046300
2 2.885696 -1116.948200 22.680700
3 2.162512 -1131.660000 7.968900
4 0.018839 -1139.870200 -0.241300
$ ./nebef.pl
0 0.017443 -1139.628900 0.000000
1 9.206077 -1111.400300 28.228600
2 5.642115 -1116.948200 22.680700
3 3.583494 -1131.660000 7.968900
4 0.018839 -1139.870200 -0.241300
we can see forces begin to rise after several steps. And here is my INCAR,
SYSTEM = Si-Diamond
ENCUT = 400
ISTART = 0
ICHARG = 2
PREC = Normal
EDIFF = 1.0E-5
EDIFFG = -1
ISYM = 0
NSW = 200
ISIF = 2
LREAL = .TRUE.
LWAVE = .FALSE.
LCHARG = .FALSE.
NELECT = 853
NELMIN = 5

#NEB
IBRION = 3
POTIM = 1
IOPT = 0
#ICHAIN = 0
#LCLIMB = .TRUE.
#SPRING = -5
IMAGES = 3
I donot know why systems begin to be chaos, can you help me?
Yidan Wei

[graeme]

Ah, sorry typo in my post; my bad. I meant to say POTIM=0.1, but you might even go 0.01 - 0.05. Just make sure that the forces drop systematically in the initial iterations and get below 1 eV/Ang.

The initial path in this calculation looks fine.

[weiyidan]

Dear professor,
Thanks for your advice, systems converged with Ediffg=-1. Please see attachment (Ediffg=-1)
However, when i continued to calculate with Ediffg=-0.02. Please see attachment (Ediffg=-0.02). It was hard to converge. If i reduce the INVCURV parameter, can this criterion be reach?
Yidan Wei

[weiyidan]

here is the one with INVCURV=0.002. However, forces are still hard to converge......

[graeme]

Attached is what would be my standard NEB calculation for this system. The run1 calculation uses IOPT=3 to get the forces below 1 eV/Ang; run2 uses IOPT=1 which rapidly converges to the ediffg specification.

[weiyidan]

Thanks for your help. Does the number of images make it hard to converge?

[graeme]

If you have too few images to resolve the path, it can make convergence harder. I don't know if you need more than 3 images - I just ran what I consider a standard calculation for your system. There is no problem with convergence. You can play around with different settings to see which are important.