Transferable Force Fields from Experimental Scattering Data with Machine Learning Assisted Structure Refinement.

Area of Science:

• Condensed matter physics
• Materials science
• Computational physics

Background:

• Deriving transferable pair potentials from scattering data is challenging.
• Current methods yield inaccurate thermodynamic predictions for fluids.
• Accurate potentials are crucial for understanding material behavior.

Purpose of the Study:

• To develop a machine learning assisted structure-inversion method for deriving accurate pair potentials.
• To validate the method using neutron scattering data from noble gases (Ne, Ar, Kr, Xe).
• To assess the transferability and predictive power of the derived potentials.

Main Methods:

• Applied a machine learning assisted structure-inversion technique to neutron scattering data.
• Refined pair potentials to match experimental scattering patterns.
• Validated potentials by simulating microstructure and vapor-liquid equilibria.

Main Results:

• Recovered transferable pair potentials for noble gases.
• Accurately reproduced microstructure and vapor-liquid equilibria from triple to critical point.
• Demonstrated that a single neutron scattering measurement is sufficient for prediction.

Conclusions:

• The machine learning method successfully derives accurate and transferable pair potentials.
• These potentials enable precise prediction of macroscopic thermodynamic properties.
• Provides novel insights into local atomic forces in dense monatomic systems.