RiPSOGM = Rotation Invariant Particle Swarm Optimization with Gaussian Mutations.

RiPSOGM was designed to be a simple and straightforward optimization platform based on the Particle Swarm Optimization (PSO) algorithm for ReaxFF reactive force fields parameterization tasks. It enables an efficient parameterization of ReaxFF-based reactive force fields to be carried out in an automated fashion exploiting recent advances in powerful meta-heuristic optimization strategies.

A well-parametrized ReaxFF force field allows prediction of molecular properties with accuracy and transferability over a wide range of materials. RiPSOGM is currently interfaced to the "gold standard" stand-alone Fortran version of ReaxFF (In the future, we aim to interface it to LAMMPS as well).

While RiPSOGM is still in its development stage, we release it to public use because we find it helpful for the process of ReaxFF force field parameterization.


Features:

  • A fast C++ interface with hooks to the "gold standard" Fortran implementation of ReaxFF
  • Implicit multi-objective fitness function specification including: partial charges, relative energies, lattice parameters, molecular geometries and heats of formation
  • Flexible parameter space exploration
  • Training set support for both finite-size and periodic systems


Current developments:

  • Integration of the code to github to assist community-based development
  • MPI Parallelization of search agents to significantly reduce development time
  • Algorithmic improvements


Download and use:

  • Please register using the form below and we will email you a link to a compiled RiPSOGM executable.
  • If you find RiPSOGM useful for your project, please cite: David Furman, Benny Carmeli, Yehuda Zeiri and Ronnie Kosloff, "Enhanced Particle Swarm Optimization Algorithm: Efficient Training of ReaxFF Reactive Force Fields", J. Chem. Theory Comput., 2018, 14 (6), pp 3100–3112.
  • An up-to-date manual (pdf) and example files can be downloaded from here
  • Please note that we do not distribute the stand-alone ReaxFF (or ReaxFF-lg) fortran sources. If you would like to obtain a copy, please contact the main developer of ReaxFF, Prof. Adri van Duin (acv13@psu.edu ).