Insights into the Spin-Lattice Dynamics of Organic Radicals Beyond Molecular Tumbling: A Combined Molecular Dynamics and Machine-Learning Approach

Abstract

The prediction of spin-lattice dynamics is a challenging computational and theoretical problem due to the complex interplay among atomic motion and spin interactions. Here, we employ machine-learning paradigms to generate a first-principles-accurate and computationally inexpensive model that predicts both conformational energy and spin Hamiltonian parameters of an organic radical as function of its molecular distortions. This model is applied to the study of the g tensor correlation function during molecular dynamics in solution and it is shown that it is possible to access the time dependence of the spin Hamiltonian parameters without making assumptions on the type of molecular motion or spin-lattice dynamics’ time-scales involved. We further use these approach to disentangle inter- and intra-molecular contributions to the g tensor correlation function and provide new insights into the nature of vibronic-coupling in organic semiconductors.