GPU acceleration for simulations of large-scale identical particles based on path integral molecular dynamics

Area of Science:

• Quantum mechanics
• Computational physics
• Materials science

Background:

• Path Integral Molecular Dynamics (PIMD) is crucial for simulating quantum systems.
• Simulating large numbers of identical particles is computationally intensive.
• GPU acceleration offers potential for overcoming computational bottlenecks.

Purpose of the Study:

• To achieve significant GPU acceleration for quadratic scaling PIMD.
• To develop an open-source PIMD code repository.
• To demonstrate the feasibility of simulating large-scale quantum systems.

Main Methods:

• Implemented quadratic scaling PIMD on GPUs.
• Developed and released an open-source PIMD code.
• Conducted numerical experiments on systems of identical bosons and fermions.

Main Results:

• Achieved significant speedups, simulating 1600 particles in two hours on a single GPU/CPU.
• Demonstrated scalability for simulating tens of thousands of particles.
• GPU acceleration shows a near-linear relationship between computation time and particle number.
• Successfully simulated fermion thermodynamics, overcoming the fermion sign problem.

Conclusions:

• GPU acceleration is foundational for large-scale PIMD simulations (>10,000 particles).
• Simulations of large identical particle quantum systems are now feasible.
• This work paves the way for efficient, accurate simulations of large fermionic systems.