Cap the magnitude of repulsive interactions for closely approaching atoms

LAMMPS LAMMPS Mailing List Mirror
1

Dear Lammps User

I am trying to study different fidelity of MD simulations and apply machine learning. For starter, I am considering MD runs with higher timesteps (10fs or 20fs) as low fidelity run. In this case, if the time-step is too large, then a one time step integration may predict significant overlap between molecules/atoms which will lead to huge (unphysical) repulsive forces and large displacements on the next time step, which in turn will lead to even larger overlap and more unphysical forces on the next time step. To prevent such divergence of integration scheme, I want to cap the magnitude of repulsive interactions for closely approaching atoms. I have search the literature and found this solution:

“For the Lennard-Jones potential, such capping can be implemented straightforwardly by modifying the potential at short distance. As such, for both test cases in this work, a potential energy cap is considered when the ratio of σ/rij exceeds 1.2.”

My question is how can I implement this? Can I do it in my input script? Or do I have to modify the potential and compile it? Obviously I would like to implement it through input script. Is there any way to do it.

Thank you very much
Mahmudul Islam

2

To achieve exactly the model you describe without recompiling LAMMPS, you can use pair_style table or - if available - pair style python.
Pair style python comes with a significant performance penalty, though, however the most convenient way to generate tables for pair style table is by using the pair_write command in LAMMPS, which also works with pair style python. there is an example in examples/python.

Alternate approaches to avoid divergence of the MD integrator due to close contacts without modifying C++ code would be:

  • /soft style soft-core potentials from the USER-FEP package, where the capping of the potential is implemented in a smooth fashion
  • using fix nve/limit to limit large displacements instead
  • using fix setforce with atom style variables for fx fy fz to limit the absolute force value per atom

axel.

3

Thank you very much Shafat and Axel.