ML-MIX: A LAMMPS Package for Force-Mixing Machine-Learned Interatomic Potentials

Machine-learned interatomic potentials (MLIPs) can offer near first-principles accuracy but are computationally expensive, limiting their application to large-scale molecular dynamics simulations. Inspired by quantum mechanics/molecular mechanics (QM/MM) methods this talk will present ML-MIX, a CPU and GPU compatible LAMMPS package to accelerate simulations by spatially force-mixing interatomic potentials of different complexities. By restricting the use of costly MLIPs to only the necessary regions of the simulation domain, ML-MIX enables researchers with constrained computational resources to overcome traditional cost–accuracy trade-offs. Its capabilities will be demonstrated through two case studies in tungsten; screw dislocation mobility and helium implantation.