Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

High-Pressure Inelastic Neutron Spectroscopy: A true test of Machine-Learned Interatomic Potential energy landscapes

Jeff Armstrong, Adam Jackson, Alin Elena

Abstract

Machine-learned interatomic potentials (MLIPs) promise to provide near density-functional theory accuracy at a fraction of the computational cost, offering a transformative route toward genuinely predictive chemistry. Yet their predictive validity beyond the training regime remains largely untested experimentally. Here we use pressure-dependent broadband inelastic neutron spectroscopy (INS) as a direct experimental probe of MLIP transferability. Employing a newly developed high-pressure superalloy clamp cell, we measure INS spectra of crystalline 2,5-diiodothiophene at 10~K under ambient conditions and at 1.5~GPa. A MACE-based MLIP, fine-tuned on targeted DFT data, reproduces the experimental spectra across 0--1200~cm$^{-1}$ at both pressures and remains thermodynamically stable under rigorous molecular dynamics validation at 300~K. The model captures systematic pressure-induced blue shifts arising from steric stiffening and reproduces an anomalous red shift at 453~cm$^{-1}$ driven by pressure-modified intermolecular interactions, providing direct validation of its many-body character. This constitutes the first experimental demonstration of MLIP transferability across distinct thermodynamic states using neutron spectroscopy, and establishes high-pressure INS as a stringent benchmark for predictive machine-learned potentials.

High-Pressure Inelastic Neutron Spectroscopy: A true test of Machine-Learned Interatomic Potential energy landscapes

Abstract

Machine-learned interatomic potentials (MLIPs) promise to provide near density-functional theory accuracy at a fraction of the computational cost, offering a transformative route toward genuinely predictive chemistry. Yet their predictive validity beyond the training regime remains largely untested experimentally. Here we use pressure-dependent broadband inelastic neutron spectroscopy (INS) as a direct experimental probe of MLIP transferability. Employing a newly developed high-pressure superalloy clamp cell, we measure INS spectra of crystalline 2,5-diiodothiophene at 10~K under ambient conditions and at 1.5~GPa. A MACE-based MLIP, fine-tuned on targeted DFT data, reproduces the experimental spectra across 0--1200~cm at both pressures and remains thermodynamically stable under rigorous molecular dynamics validation at 300~K. The model captures systematic pressure-induced blue shifts arising from steric stiffening and reproduces an anomalous red shift at 453~cm driven by pressure-modified intermolecular interactions, providing direct validation of its many-body character. This constitutes the first experimental demonstration of MLIP transferability across distinct thermodynamic states using neutron spectroscopy, and establishes high-pressure INS as a stringent benchmark for predictive machine-learned potentials.
Paper Structure (2 sections, 3 figures)

This paper contains 2 sections, 3 figures.

Table of Contents

  1. Data Availability
  2. Acknowledgements

Figures (3)

  • Figure 1: Experimental INS spectra (top blue panel) compared with MLIP-derived neutron-weighted spectra for 1$^{\text{st}}$ and 2$^{\text{nd}}$ generation models (yellow and green panels respectively) at ambient pressure (black lines) and 1.5 GPa (blue lines). Illustrative blue and red shifts are shown as colored arrows.
  • Figure 2: Left: MLIP-relaxed packing motif for the MLIP, illustrating bilayer formation and herringbone arrangement. Right: Molecular depictions/assignments of vibrational modes across the 80-500 cm$^{-1}$ range in ascending frequency.
  • Figure 3: Validation of the 2nd Gen. MLIP (FT-MH-1-PBE-D3) simulation at 300 K. Top left: The mean-squared displacement of each atomic component. Top right: Fluctuations of the components of the stress tensor. Bottom left: Fluctuation of the potential energy. 1000-point running averages are shown for each plot as dark lines, and std are shown as shaded windows as a guide for the eye. Bottom right: Radial distribution functions for C--C, H--H, and C--H, sampled between 0.1--0.3 ps (solid lines) and between 0.9--1.0 ps (shaded plots).