Velocity Jumps for Molecular Dynamics

Area of Science:

• Computational Chemistry
• Molecular Dynamics Simulations
• Statistical Mechanics

Background:

• Classical molecular dynamics simulations often face computational bottlenecks due to long-range interactions.
• Existing methods like Langevin dynamics can be computationally expensive for large systems.
• Multi-time-step methods offer speedups but can suffer from resonance issues.

Purpose of the Study:

• To introduce a new class of molecular dynamics integrators, the Velocity Jumps (JUMP) approach.
• To develop a method that accelerates simulations while maintaining accuracy in sampling and dynamics.
• To enhance computational efficiency in molecular dynamics by replacing expensive calculations with random velocity resampling.

Main Methods:

• Developed the Velocity Jumps (JUMP) approach, a hybrid model combining Langevin diffusion and a piecewise deterministic Markov process.
• Replaced the computation of long-range pairwise interactions with velocity resampling at random intervals.
• Integrated JUMP with classical multi-time-step methods (JUMP-RESPA, JUMP-RESPA1).
• Implemented the JUMP integrators in the GPU-accelerated Tinker-HP package.

Main Results:

• The JUMP approach significantly accelerates molecular dynamics simulations.
• Preservation of key sampling and dynamical properties, including the diffusion constant, was demonstrated.
• Integration with multi-time-step methods further enhanced computational speed.
• The random nature of velocity jumps helped avoid resonance issues common in other methods.
• JUMP integrators showed superior performance compared to their BAOAB counterparts on GPU.

Conclusions:

• The Velocity Jumps (JUMP) approach offers a computationally efficient alternative for molecular dynamics simulations.
• JUMP provides a robust framework for accelerating simulations without compromising accuracy.
• The implementation in Tinker-HP demonstrates practical applicability and significant performance gains on GPUs.