in this paper
some relevant information can be found about the LAMMPS code implementing REBO potentials: indeed, it is provided a general benchmark solution to test any MD code, and results produced by LAMMPS when adopting second generation Brenner potential have been checked; in its current implementation, the code produces results which are offset from those of the benchmark by a significant amount, and the reason is provided. The paper is endowed with a simple Mathematica code able to furnish the benchmark, and the two LAMMPS files for the original and the modified REBO potential are attached.
I hope it can be useful and practical for the community.
Thanks - we’ll take a look. At first glance it appears that
a param needs to be changed when running AIREBO vs REBO.
In LAMMPS it is the same code that does both. The
latter just turns off some terms in the former. Apparently
an additional param also needs to be changed. Is that
the gist of it?
This seems to be a difference in the definitions of the AIREBO and REBO potentials for C-C bonds. Since these are two different classes, it should be easy to modify the code to handle this automatically, without any user intervention. Were there any other issues with REBO in LAMMPS?
I like this unusual “Note to users” that you included in the paper:
“Oftentimes, numerical instabilities (such as ’exploding’ results) have been reported in internet forums dedicated to LAMMPS simulations. These can be caused by a variety of reasons, not necessarily related to a problem in the potential. At present, we are not aware of any example where the issue we found in the potential causes numerical instabilities.”
this change for REBO (relative to AIREBO) will be in a patch