NEB analysis, convergence?

aakbarz · Post by **aakbarz** » Mon Apr 23, 2007 6:58 pm

hello there,
I am trying to run NEB, after completely relaxing the initial and final configurations I started some calculations with 8 images. It is now 324 SCF iterations and it is still not converged and some of the forces on some images are large, I included some part of it. The slab I have has got 48 atoms of 3 different species. in this problem I try to remove to hydrogen from the surface and put it somewhere in the vacuum far from surface. I am using the new optimizer modules, here I used CG optimizer.
Is there anything wrong? or should I be patient and run the calculations until it reaches the required accuracy? what about the distance between the springs are they reasonable? Is there any detailed example somewhere so that I can read it and try to redo it?
Any suggestions are greatly appreciated.

Sincerely,

***************************
INCAR parameters for NEB settings:

setting for NEB method
IMAGES = 8
ISPRING = 3
SPRING = -5
SPOWER = 1
EFIRST = -193.73952
ELAST = -189.32986
EMIN = 0.0
EMAX = 0.0
LTANGENT = .TRUE.
*************************************
example of part of the forces on atoms:

POSITION TOTAL-FORCE (eV/Angst)
-----------------------------------------------------------------------------------
1.91073 1.45919 1.91401 1.089326 -0.330677 0.403312
2.21525 1.15312 7.31898 -1.069266 0.252300 1.013549
1.35811 2.53507 6.62996 0.302029 -0.623795 0.053179
-0.43595 3.68854 11.19067 0.295138 0.642428 0.818986
5.37657 3.82355 4.29037 -0.411248 -0.150652 -0.069419
4.77358 4.18659 10.77121 1.575371 -0.581984 1.318317
6.40221 0.70262 4.00827 -0.552841 0.371571 -1.588166
5.31891 1.89441 7.63279 -0.347612 -0.233819 -1.383367
5.71146 1.26090 6.45935 0.210705 0.352647 2.558503
-0.24490 4.55232 12.08406 0.421606 -0.554632 -1.144896
4.53535 3.48986 3.31041 0.117391 0.413968 0.230821
5.18378 2.73857 6.63549 -0.042301 -0.044685 1.869471
7.49894 0.28011 3.59551 -0.557851 -0.277060 -0.405421
-0.09616 3.68442 10.01027 -0.564431 0.925539 0.153082

--------------------------------------
NEB: Tangent
----------------------------------------------
-0.08903 -0.01353 0.03378
0.13493 -0.05679 -0.15222
-0.03887 0.04913 -0.00153
-0.01314 -0.04207 -0.07772
0.05474 0.00439 -0.00440
-0.17621 0.04413 -0.13655
0.06146 -0.04777 0.18996
-0.00153 0.07385 0.17383
-0.05809 -0.06786 -0.29392
-0.07779 0.03874 0.15474
-0.00531 -0.03691 -0.03632
0.03647 -0.00224 -0.20377
0.08319 0.02319 0.07533
0.06426 -0.07293 0.03329
0.12004 -0.00101 0.12272
-0.04342 0.03924 0.20129

NEB: forces: par spring, perp REAL, dneb 0.941261 3.590879 0.000000
NEB: distance to prev, next image, angle between 10.124088 10.312341 106.491196
NEB: projections on to tangent (spring, REAL) 0.941261 -7.893029

FORCES: max atom, RMS 2.030303 0.535809
FORCE total and by dimension 3.712194 2.019348

OPT: Flag T
OPT: CG Step
OPT: CG fdstep
OPT: CG gam 2.072299

*********************************

andri · Post by **andri** » Mon Apr 23, 2007 9:22 pm

What is the output from *nebef.pl*?

aakbarz · Post by **aakbarz** » Mon Apr 23, 2007 10:06 pm

[quote="andri"]What is the output from *nebef.pl*?[/quote]

hello andri,

here is the result after nebef.pl, as you see the forces on imgae 2, 3 and 4 are hugh?!
**********
0 0.00908900 -193.73952000 0.00000000
1 0.23138300 -187.94415600 5.79536400
2 1.49967100 -177.61582300 16.12369700
3 4.12039600 -165.41490600 28.32461400
4 3.34229200 -170.60620500 23.13331500
5 0.79876100 -181.82248600 11.91703400
6 0.94269900 -181.10564200 12.63387800
7 0.41859400 -186.79028800 6.94923200
8 0.25759800 -191.72255600 2.01696400
9 0.00974300 -189.32986500 4.40965500
************
FYI, I used EDIFFG=-0.01 and EDIFF=1E-6 in the INCAR file.
Shall I comment the line for EDIFF? that is just stick to the force criteria?

thanks,

andri · Post by **andri** » Mon Apr 23, 2007 10:31 pm

Using for instance the *dist.pl* script, what is the distance between the images? Also, have you tried to generate a .xyz movie (nebmovie.pl) from the the current CONTCARs? Is there something funny going in with the path?

I would perhaps decrease EDIFF to 1.0E-4.

Post by **graeme** » Tue Apr 24, 2007 2:23 am

The forces and energies are very large. I agree with Andri that it would be best to start by looking at the path.

If your initial path is somewhat unphysical (a long way from the minimum energy path) it is best to start with a conservative optimizer (IBRION=3 or IOPT=3 (quickmin) or IOPT=7 (fire)) using a small time step (e.g. POTIM=0.01, or TIMESTEP=0.01 for the new optimizers) until the forces get to around 1 eV/Ang, and the barrier is lower than 10 eV. This should happen fairly quickly, usually within 20 iterations. If it doesn't, you should look to see (visually) what's going wrong with the pathway.

aakbarz · Post by **aakbarz** » Wed Apr 25, 2007 6:23 pm

graeme,

Thanks for the input,
I wonder by increasing the number of images, and or decreasing the distance between the images this will help. Right now the images have an average distance of 5.6 to 6.7 from each other (for 8 images).
one more thing, if I set LCLIMB=.TRUE., does this help?

thanks

andri · Post by **andri** » Thu Apr 26, 2007 8:22 pm

If the path is curved then 5 A is probably too large of a distance between the images. Remember that the neighboring images are used in a 'finite' difference way to estimate the local tangent. Personally, I prefer never to have the distance exceed 1 A, and try very hard to have it close to 0.5 A. So perhaps you should try to increase the number of images.

Maybe you can try the dimer method to find the saddle point and then slide down from it on each side of the saddle? The saddle point along your current path could be used to generate the dimer.