NEB fixes the first and final images?

LAMMPS LAMMPS Mailing List Mirror
1

Dear LAMMPS users,

Recently I used the NEB function in LAMMPS to calculate the transition path of Silicon atom on amorphous Si surface. All the first/last images were fully optimized using CG method, and it can be verified that they were local stable configurations. For most cases, NEB can provide reasonable results, however, I found some strange cases. After NEB calculations, the first/last images move to a nearby stabler configuration. So my question is, if the NEB implementation in LAMMPS fixes the first and last images? In principle, they should be fixed during NEB and only the intermediate images can be relaxed. I dug into the source code and checked min_fire.cpp, neb.cpp, and fix_neb.cpp, and none of them shows such constraint. Before modifying the source code, can I have any way to realize the fix in NEB?
Furthermore, I found that for some cases NEB completely failed. The total potential energy becomes a very large value which means that the system is melted after adding only one adatom on the surface. Do I have better way to calculate the transition path in LAMMPS?

Regards,
Gang

2

Dear LAMMPS users,

Recently I used the NEB function in LAMMPS to calculate the transition
path of Silicon atom on amorphous Si surface. All the first/last
images were fully optimized using CG method, and it can be verified
that they were local stable configurations. For most cases, NEB can
provide reasonable results, however, I found some strange cases. After
NEB calculations, the first/last images move to a nearby stabler
configuration. So my question is, if the NEB implementation in LAMMPS
fixes the first and last images? In principle, they should be fixed
during NEB and only the intermediate images can be relaxed. I dug into
the source code and checked min_fire.cpp, neb.cpp, and fix_neb.cpp,
and none of them shows such constraint. Before modifying the source
code, can I have any way to realize the fix in NEB?

Your first/last replica might not have been relaxed enough. That happen
often with CG, with the stopping criteria "Line search alpha equal
zero". Since NEB is perform with Quickmin/Fire, you might want to relax
your initial/final structure with the Quickmin/Fire, to a fore norm
lower than the one used by the NEB.

Furthermore, I found that for some cases NEB completely failed. The
total potential energy becomes a very large value which means that the
system is melted after adding only one adatom on the surface. Do I
have better way to calculate the transition path in LAMMPS?

From my personal experience, NEB calculations can be tricky to converge.

I often tune the parameters (spring force, normal spring force,
timestep, number of intermediate configurations, etc.) in order to have
a converge path. A correct set of parameter for a given system, might be
not good with another one (actually, that is often the case in MS/MD...).

Also, quickmin is slower than fire, in some case it i also more stable
(i.e. the system do not "melt" during the "climbing" stage).

Maybe you could paste here the exact command you are using for your NEB?

Julien