fix pimd/langevin: An efficient implementation of path integral molecular dynamics in LAMMPS

Area of Science:

• Computational Chemistry
• Materials Science
• Quantum Mechanics

Background:

• Path Integral Molecular Dynamics (PIMD) is crucial for simulating nuclear quantum effects.
• Accurate PIMD requires many beads, making it computationally expensive.
• Existing PIMD software can be a bottleneck, especially with efficient machine learning potentials like Deep Potential (DP).

Purpose of the Study:

• To develop a highly efficient PIMD implementation within the LAMMPS framework.
• To leverage massively parallel supercomputing architectures for PIMD simulations.
• To accelerate PIMD calculations, particularly when using machine learning interatomic potentials.

Main Methods:

• Implementation of `fix pimd/langevin` in LAMMPS.
• Utilizing the Message Passing Interface (MPI) for parallelization.
• Validation using liquid water simulations and benchmarking against i-PI.
• Performance evaluation through strong and weak scaling tests.

Main Results:

• The new LAMMPS implementation (`fix pimd/langevin`) provides high computational efficiency.
• The code achieves several-fold acceleration compared to the i-PI software package for DP simulations of water.
• Demonstrated favorable parallel performance through strong and weak scaling.

Conclusions:

• The `fix pimd/langevin` module offers an efficient and scalable solution for PIMD simulations in LAMMPS.
• This implementation effectively captures nuclear quantum effects, especially when combined with machine learning potentials.
• The enhanced computational efficiency opens possibilities for more complex molecular simulations on modern supercomputers.