Fix_modify energy adds the energy of a particular fix to the total PE,
which is both printed and stored internally for other commands to
use, e.g. a minimizer.
Steve
Hi, Steve,
First, Thank you very much for your previous help. I have another question to ask about NEB. In lammps, the neb calculation will to the energy minimization to the system first to both the initial and final structure, right? the algorithm of energy minimization of initial and final structure are the same or not?
Thank you very much for your help. My lab mates and I appreciate that very much.
Not sure what you are asking. The initial and final states of the NEB chain
are relaxed to their respective minima, using the same minimization
algorithm (e.g. FIRE or QuickMin) that all the states are using. However
the initial and final states do not feel inter-replica forces.
Steve
Hi, Steve,
I am very sorry for my unclear question. Here, let me restate it. Based on my understanding of the manual, I think the first and last states should not feel the spring (inter-replica) forces in the energy minimization process of NEB (either fire or quickmin). Which means the spring constant has no effect of the converged energy in energy minimization process before climbing part. So, I did the test, I run the simulation twice with same initial and final structure and the same input file except the spring constant (for the first case, I set the spring constant to 1 and another case I set it to 0.01). The output give me different converged energy in energy minimization process. For the case of spring constant equals to 1, they are -2413.11 for both initial and final states. While for the case of spring constant equals to 0.001, they are both -2412.93. They are not consistent. I got confused here, I guess I must made some mistake of either my understanding or my coding. So, can you do me a favor to help me to check what’s the problem or give me some hints. My group mates and I struggling for this problem for long time. Thank you very much for your help.
For your reference, I attached both input and out put in this Email. I hope they are helpful. I appreciate your patience and help.
in.spring1 (483 Bytes)
log.lammps1 (48.7 KB)
in.spring001 (486 Bytes)
log.lammps001 (272 KB)
I looked at these 2 outputs. They run for a different number of
timesteps (in the first stage of NEB). So you can’t compare
the energies of the end-point replicas and expect them
to be identical (they are very close). What should be
the case is that they match between the 2 runs up to
the number of timesteps they run. That is nearly the
case. At timestep 12000 in both runs, the PE of tne final
replicas is identical. Then there is a small divergence until
one of the runs ends at step 19000. There may be some
round-off issue I’m not thinking of, but it is a tiny difference.
In both cases, you don’t start with a fully converged end state, since
they are continuing to change during the NEB run. If you don’t
want them to do that, you could just fix the positions of the
atoms in the first and last replica.
Steve
I figured out the other conceptual issue that was
puzzling me. It is correct that the end point replicas
do not feel the spring forces. However, all the replicas
are making a collective decision about how to alter their
velocities while the QuickMin or FIRE minimizer proceeds.
E.g. on what iteration to reset velocities to 0.0. This means
that the velocities of the end point replicas are affected
by the velocities in the middle replicas (which are in turn
affected by the spring forces). So even if the end point
replicas are near a minimum, there can be slight differences
in their potential energies when you make 2 runs with
different spring forces, b/c the velocities in the end point
replicas will be slightly different. That’s what you’re seeing
in your output.
Steve