Highly efficient path-integral molecular dynamics simulations with GPUMD using neuroevolution potentials: Case...

Area of Science:

• * Computational materials science
• * Quantum mechanics
• * Statistical mechanics

Background:

• * Path-integral molecular dynamics (PIMD) is vital for modeling nuclear quantum effects.
• * High computational cost of PIMD hinders simulations of finite-size effects.
• * Integrating advanced potentials with efficient algorithms is needed.

Purpose of the Study:

• * To implement and validate a GPU-accelerated PIMD approach.
• * To combine neuroevolution potentials (NEP) with ring-polymer molecular dynamics (RPMD) and thermostatted ring-polymer molecular dynamics (TRPMD).
• * To enable large-scale, accurate simulations of materials with nuclear quantum effects.

Main Methods:

• * Developed specialized GPU implementation of PIMD within the GPUMD package.
• * Integrated highly accurate and efficient machine-learned neuroevolution potential (NEP) models.
• * Applied the combined NEP-PIMD approach to diverse materials: LiH, MOFs, water, and aluminum.

Main Results:

• * Achieved accuracy comparable to first-principles calculations with empirical potential efficiency.
• * Successfully simulated isotope effects in LiH and captured quantum effects in water's structure.
• * Demonstrated accurate thermal expansion and phonon properties for aluminum using TRPMD.
• * Highlighted the necessity of considering quantum effects and dispersive interactions for MOFs.

Conclusions:

• * The GPU-accelerated NEP-PIMD method offers an accessible, accurate, and scalable tool.
• * This approach effectively overcomes finite-size limitations in materials simulations.
• * Facilitates exploration of complex material properties influenced by nuclear quantum effects across various applications.