Dear Prof. Hekelman,
For antiferromagnetic system, I found that the barrier is dependent on the initial magnetic moments I assigned. For example, I have an interstitial with initial moment up and final moment down, so it should have changed the moment at some point. However, the CINEB results are found to depend on the initial moment. That is to say, when I assign the moment of initial state or final state, the calculated barrier is different and the magnetic moment is also totally different. In particular, when using 5 images, I found that the transition of moment always occurs at last image. The moment of last image is somewhat similar to initial state but totally flipped with respect to final state.
What do you think to this problem? It seems a serious problem with NEB calculations.
Thank you!
Shijun
Initial magnetic moment dependent barrier
Moderator: moderators
Re: Initial magnetic moment dependent barrier
The magnetic moment of images along an NEB calculation can certainly be very important for convergence of the band. I would not say, however, that this is a problem with the NEB algorithm. Rather, it is a problem with electronic structure methods automatically finding their lowest energy electronic configuration.
One thing you can do is to do a single-point energy calculation for each image, using MAGMOM (or any trick you have) to get the correct electronic structure. Then, you can use the resulting WAVECAR and CHGCAR to start from the correct electronic structure in your NEB calculation.
Another trick is to used the damped dynamics algorithm for the electronic structure (e.g. IALGO=54). While slower than the default algorithms, this does seem to improve convergence to the global electronic structure minimum.
One thing you can do is to do a single-point energy calculation for each image, using MAGMOM (or any trick you have) to get the correct electronic structure. Then, you can use the resulting WAVECAR and CHGCAR to start from the correct electronic structure in your NEB calculation.
Another trick is to used the damped dynamics algorithm for the electronic structure (e.g. IALGO=54). While slower than the default algorithms, this does seem to improve convergence to the global electronic structure minimum.