Result: Systematic investigations of anthraquinone derivative redox potentials

Anthraquinone (AQ) and its derivatives show promise as battery materials. However, their limited commercial availability and subsequent measurement of redox potential pose a considerable challenge. This study presents an automated pipeline that can: (i) generate an arbitrary number of anthraquinone derivatives; (ii) compute their one- and two-electron redox potentials using density functional theory (DFT) or density functional theory–based tight binding (DFTB) methods; and (iii) analyze the resulting dataset using common machine learning algorithms. Approximately 40,000 anthraquinone derivatives were generated and their redox potentials in acetonitrile were calculated. Three multiple linear regression (MLR) models and two feed-forward neural network (NN) models were trained using quantum-mechanical redox potential data. A small sample set was used to compare quantum-mechanical methods with MLR and NN regression models against cyclic-voltammetry measurements. All methods showed good agreement with the measured data. The efficacy of the MLR method shows that a simple, additive treatment of functional groups can reasonably approximate the calculated redox potential values. We used principal component analysis to compare selected functionalizations of the anthraquinone molecule. The presented workflow is built using the Python programming language and common libraries. This workflow can easily be expanded to include other quinones, such as benzoquinones and naphthoquinones.