Machine Learning-Accelerated Path Integral Molecular Dynamics and ¹³C NMR Simulations Unlock New Insights into...

Area of Science:

• Computational Chemistry
• Materials Science
• Machine Learning Applications

Background:

• Accurate simulation of negative thermal expansion (NTE) and ¹³C nuclear magnetic resonance (NMR) in C₆₀ fullerenes is crucial.
• Previous quantum-mechanical methods are computationally expensive and limited in applicability.
• Existing alternative methods like ab initio path integral molecular dynamics (PIMD) are also computationally demanding for NMR parameter calculations.

Purpose of the Study:

• To develop an accurate and efficient computational approach for simulating NTE and ¹³C NMR signatures in C₆₀.
• To overcome the limitations of traditional methods, particularly for weakly bound systems and complex quantum phenomena.
• To investigate the influence of dispersion effects and basis set choices on NTE magnitude.

Main Methods:

• Introduction of a novel neural network-based approach combining machine learning interatomic potentials (MLIPs) with an NMR machine learning (NMR-ML) model.
• Implementation of machine learning PIMD (MLPIMD) simulations using MLIPs.
• Direct computation of ¹³C isotropic magnetic shielding (σ_iso) from MLPIMD snapshots using the NMR-ML model.

Main Results:

• Temperature-dependent MLPIMD simulations consistently show negative thermal expansion (NTE).
• The study reveals the significant influence of dispersion effects and atomic basis set choices on NTE magnitude.
• NTE is confirmed as a quantum-mechanical phenomenon, unachievable with classical molecular dynamics (MD) simulations.

Conclusions:

• The developed ML-accelerated simulation approach provides accurate and efficient modeling of thermally activated quantum mechanical phenomena.
• The method accurately reproduces experimental data for secondary isotope shifts in ¹³C NMR magnetic shielding.
• This work validates the power of machine learning in advancing computational studies of complex molecular systems.