Machine-Learned Solid-State 13C Chemical Shifts for Structure-Resolved Analysis of Polymer Simulations
Abstract
Solid-state 13C nuclear magnetic resonance (NMR) spectroscopy is among the most powerful probes of polymer structure, yet predicting solid-state 13C NMR spectra from molecular simulations has remained infeasible due to intractable computational costs. Here we introduce PNGWNN, a framework coupling molecular dynamics and machinelearned chemical shifts to efficiently generate solid-state 13C NMR spectra from molecular simulation. Trained on three polymer chemistries, PNGWNN achieves ∼1 ppm chemical shift prediction accuracy against density functional theory and transfers quantitatively to six chemically distinct test polymers across amorphous, crystalline, surface, multiphase, and copolymer morphologies. Moreover, predicted spectral features can be decomposed into atomically detailed contributions associated with specific conformations, hydrogen-bond motifs, and phase compositions, revealing structure–spectrum relationships without prior structural hypotheses. PNGWNN makes solid-state 13C NMR a computationally accessible observable, opening routes to simulation-driven structural inference and experimental design in polymers.
Related
- A Transferable Machine-Learned Potential Enabling Large-Scale Copper-Carbon Simulations from Equilibrium to Extreme Conditions
- What Rate Theories Compute: Chemical kinetics from Arrhenius to machine-learned potentials
- Accelerating Mechanistic Understanding and Prediction of Chemical Selectivity with Foundation Machine-Learned Interatomic Potentials
- Enhanced Quantum Simulations Reveal Conformation-Controlled, Site-Resolved Proton Binding in Biliverdin
- TS Second Opinion: catching, repairing and triaging mirror-image errors in machine-learned transition states
BibTeX
@misc{zhang2026machinelearned,
title = {{Machine-Learned Solid-State 13C Chemical Shifts for Structure-Resolved Analysis of Polymer Simulations}},
author = {Hang Zhang and Nicholas E. Jackson},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15001687/v2},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15001687/v2},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Machine-Learned Solid-State 13C Chemical Shifts for Structure-Resolved Analysis of Polymer Simulations
AU - Hang Zhang
AU - Nicholas E. Jackson
PY - 2026
DA - 2026/09/06
PB - ChemRxiv
DO - 10.26434/chemrxiv.15001687/v2
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15001687/v2
AB - Solid-state 13C nuclear magnetic resonance (NMR) spectroscopy is among the most powerful probes of polymer structure, yet predicting solid-state 13C NMR spectra from molecular simulations has remained infeasible due to intractable computational costs. Here we introduce PNGWNN, a framework coupling molecular dynamics and machinelearned chemical shifts to efficiently generate solid-state 13C NMR spectra from molecular simulation. Trained on three polymer chemistries, PNGWNN achieves ∼1 ppm chemical shift prediction accuracy against density functional theory and transfers quantitatively to six chemically distinct test polymers across amorphous, crystalline, surface, multiphase, and copolymer morphologies. Moreover, predicted spectral features can be decomposed into atomically detailed contributions associated with specific conformations, hydrogen-bond motifs, and phase compositions, revealing structure–spectrum relationships without prior structural hypotheses. PNGWNN makes solid-state 13C NMR a computationally accessible observable, opening routes to simulation-driven structural inference and experimental design in polymers.
ER -