High-Pressure Inelastic Neutron Spectroscopy: Experimental Validation of Machine-Learned Interatomic Potential Energy...

Area of Science:

• Computational Materials Science
• Spectroscopy
• Machine Learning

Background:

• Machine-learned interatomic potentials (MLIPs) offer accurate and cost-effective atomistic modeling.
• Experimental validation of MLIP reliability beyond training data is challenging.

Purpose of the Study:

• To use pressure-dependent inelastic neutron spectroscopy (INS) to experimentally probe MLIP transferability.
• To validate the predictive accuracy of MLIPs under varying thermodynamic conditions.

Main Methods:

• Measurement of INS spectra of crystalline 2,5-diiodothiophene at 10 K under atmospheric pressure and 1.5 GPa using a high-pressure clamp cell.
• Development and fine-tuning of MACE-based MLIPs using density-functional theory (DFT) data.
• Finite-temperature molecular dynamics simulations.

Main Results:

• MLIPs accurately reproduced experimental INS spectra across 0-1200 cm^-1 at both atmospheric pressure and 1.5 GPa.
• MLIPs remained stable in molecular dynamics simulations at 300 K.
• The models successfully captured pressure-induced spectral shifts, including blue shifts from steric effects and an anomalous red shift due to intermolecular interactions.

Conclusions:

• Pressure-dependent INS serves as a practical experimental method for validating MLIPs.
• This technique assesses not only equilibrium structure but also the pressure-dependent potential energy surface.
• High-pressure INS is essential for establishing the transferability and reliability of MLIPs for molecular materials.