Machine Learning Isotropic g Values of Radical Polymers

Davis Thomas Daniel; Souvik Mitra; Rüdiger-A Eichel; Diddo Diddens; Josef Granwehr

doi:10.1021/acs.jctc.3c01252

Machine Learning Isotropic g Values of Radical Polymers

J Chem Theory Comput. 2024 Mar 26;20(6):2592-2604. doi: 10.1021/acs.jctc.3c01252. Epub 2024 Mar 8.

Authors

Davis Thomas Daniel^{1

2}, Souvik Mitra³, Rüdiger-A Eichel^{1

4}, Diddo Diddens⁵, Josef Granwehr^{1

2}

Affiliations

¹ Institute of Energy and Climate Research (IEK-9), Forschungszentrum Jülich GmbH, 52425 Jülich, Germany.
² Institute of Technical and Macromolecular Chemistry, RWTH Aachen University, 52056 Aachen, Germany.
³ Institute of Physical Chemistry, University of Münster, 48149 Münster, Germany.
⁴ Institute of Physical Chemistry, RWTH Aachen University, Aachen 52056, Germany.
⁵ Helmholtz Institute Münster (IEK-12), Forschungszentrum Jülich GmbH, 48149 Münster, Germany.

Abstract

Methods for electronic structure computations, such as density functional theory (DFT), are routinely used for the calculation of spectroscopic parameters to establish and validate structure-parameter correlations. DFT calculations, however, are computationally expensive for large systems such as polymers. This work explores the machine learning (ML) of isotropic g values, g_iso, obtained from electron paramagnetic resonance (EPR) experiments of an organic radical polymer. An ML model based on regression trees is trained on DFT-calculated g values of poly(2,2,6,6-tetramethylpiperidinyloxy-4-yl methacrylate) (PTMA) polymer structures extracted from different time frames of a molecular dynamics trajectory. The DFT-derived g values, g_iso^calc, for different radical densities of PTMA, are compared against experimentally derived g values obtained from in operando EPR measurements of a PTMA-based organic radical battery. The ML-predicted g_iso values, g_iso^pred, were compared with g_iso^calc to evaluate the performance of the model. Mean deviations of g_iso^pred from g_iso^calc were found to be on the order of 0.0001. Furthermore, a performance evaluation on test structures from a separate MD trajectory indicated that the model is sensitive to the radical density and efficiently learns to predict g_iso values even for radical densities that were not part of the training data set. Since our trained model can reproduce the changes in g_iso along the MD trajectory and is sensitive to the extent of equilibration of the polymer structure, it is a promising alternative to computationally more expensive DFT methods, particularly for large systems that cannot be easily represented by a smaller model system.