Machine Learning Isotropic g Values of Radical Polymers

Area of Science:

• Computational chemistry
• Materials science
• Spectroscopy

Background:

• Density functional theory (DFT) is used for calculating spectroscopic parameters but is computationally expensive for large systems like polymers.
• Organic radical polymers are promising for applications such as batteries, requiring accurate structure-parameter correlations.
• Electron paramagnetic resonance (EPR) spectroscopy provides experimental data on these materials.

Purpose of the Study:

• To develop and evaluate a machine learning (ML) model for predicting isotropic g values (g_iso) of organic radical polymers.
• To compare ML-predicted g_iso values with DFT-calculated and experimentally measured values.
• To assess the ML model's ability to handle varying radical densities and molecular dynamics.

Main Methods:

• Trained a regression tree-based ML model on DFT-calculated g_iso values for poly(2,2,6,6-tetramethylpiperidinyloxy-4-yl methacrylate) (PTMA) structures from molecular dynamics (MD) trajectories.
• Compared ML-predicted g_iso values (g_iso^pred) with DFT-derived (g_iso^calc) and experimental EPR data.
• Evaluated model performance on structures from a separate MD trajectory, assessing sensitivity to radical density and structural equilibration.

Main Results:

• The ML model achieved mean deviations of approximately 0.0001 compared to DFT-calculated g_iso values.
• The model demonstrated sensitivity to radical density, accurately predicting g_iso values even for densities not present in the training set.
• The ML model successfully reproduced changes in g_iso along the MD trajectory, indicating sensitivity to polymer structure equilibration.

Conclusions:

• Machine learning offers a computationally efficient alternative to DFT for predicting spectroscopic parameters in large organic radical polymer systems.
• The developed ML model shows promise for accelerating the design and analysis of materials for applications like organic radical batteries.
• The ML approach can provide insights into structure-property relationships by correlating dynamic structural changes with spectroscopic outputs.