An Accurate Machine-Learned Potential for Krypton under Extreme Conditions.

Area of Science:

• Computational physics and chemistry
• Materials science
• Quantum mechanics

Background:

• Accurate interatomic potentials are crucial for simulating materials.
• The Lennard-Jones model is widely used but has limitations for krypton.
• Quantum chemical calculations provide high-fidelity data for potential development.

Purpose of the Study:

• To develop and validate novel machine-learned pair potentials for krypton.
• To assess the performance of these potentials against experimental data and existing models.
• To improve the accuracy of molecular dynamics simulations for krypton.

Main Methods:

• Machine learning was used to create two pair potentials for krypton.
• Potentials were trained on coupled cluster with single, double, and perturbative triple excitations (CCSD(T)) quantum chemical calculations.
• Extensive molecular dynamics simulations were performed for validation.

Main Results:

• The developed potentials accurately reproduce experimental data for krypton's equation of state, melting point, and neutron scattering in the fluid phase.
• Both machine-learned potentials performed comparably to the Lennard-Jones model at low pressures.
• At high pressures (up to 30 GPa), the machine-learned potentials showed significantly better agreement with experimental solid-state data than the Lennard-Jones model.

Conclusions:

• Machine-learned potentials offer a significant improvement over traditional models like Lennard-Jones for krypton, especially at high pressures.
• These new potentials provide a more reliable tool for simulating krypton's properties across a wider range of conditions.
• The study highlights the power of combining quantum chemistry with machine learning for developing accurate interatomic potentials.