Page 1 of 1

cNEN convergence problem

Posted: Mon Dec 04, 2017 4:15 am
by hakuna
Dear Prof. Graeme:
we use the cNEB to find the transition state, 5 images are used in 3 tests and the 3 tests are the same work.

in test 3, 5 images were produced by using nebmake.pl and modified manually to remove the short bond length, then we use CG optimizer to follow the path relexition. at the beginning phase about 100 steps, all the things seem OK, hereafter the images begin to go mad gradually......

in test 1, 5 images are generated as test 1, but LBFGS optimizer was used to pursue the path, the situation is also like test 1 and the job was interrupted after several hundred steps since the chain becomes disordered, then we choose the structures after 100 steps as the beginning to restart the job, unfortunately it goes wrong, like that in test 1 .....

in test 2, as in test 1, we choose the structures after 100 steps as the beginning and use the CG optimizer to restart the job, it was very regret that the images go crazy.....

This jobs try to test the works reported in Phys. Chem. Chem. Phys., 2017, 19, 15444--15453, there the workers used the dmol3.

I attached the related files here. my questions are:
1. what caused the chaos of the images? how to deal with this situation?
2. in the zipped files, an .err file, 44351.err or like that, provides some error messages. generally theses file formed after several hundred steps, but the job continues. in other TS search jobs, similar situation are also occurred but at the end the TS are really cached. is these error special to the current job? how to solve these problems?

thanks

Re: cNEN convergence problem

Posted: Mon Dec 04, 2017 1:21 pm
by graeme
First, make sure that the same atoms are frozen in every image. It looks like you have frozen the top layer for the NEB images but not the endpoints.

Second, start with a conservative optimizer until the force on every image drops below about 0.5 eV/Ang. Using IBRION=3 and POTIM=0.1 for about 100 steps is typically good. Then, check to make sure that your path is smooth, continuous, and that the forces are dropping systematically. Then you can switch to any optimizer (except CG). You may have to adjust POTIM for IBRION=1, or if you are using our optimizers, then INVCURV using IOPT=1.

Re: cNEN convergence problem

Posted: Wed Dec 06, 2017 7:44 am
by hakuna
yest-day and before, my reply was dropped out when submit, I am not sure what was wrong
so the thanks is late^-^
thank you Professor, thank you for so quickly a reply!
I change the strategy in searching the TS according to you suggestion and the test is running.....

by the way, I prefer your optimizer, in this case, the POTIM should always be 0, right?
I have used the early version of VTST with VASP 4.6, cNEB with CG optimizer is impressive, why did you not recommend this optimizer now?