Experimental Validation of a Computational Screening Approach to Predict Redox Potentials for a Diverse Variety of Redox-Active Organic Molecules

Abstract

Organic redox flow batteries are currently the focus of intense scientific interest because they have the potential to be developed into low-cost, environmentally sustainable solutions to the energy storage problem that stands in the way of widespread uptake of renewable power generation technologies. Because the search space of suitable redox-active electrolytes is large, computational screening is increasingly being employed as a tool to identify promising candidates. It is well known in the computational chemistry literature that redox potentials for organic molecules can be accurately calculated on a class-by-class basis, but the general utility and accuracy of the relatively low-cost quantum chemical methods used in high-throughput screening are currently unclear. In this work, we measure the redox potentials of 24 commonly available but chemically diverse redox-active organic molecules in acetonitrile, carefully controlling experimental errors by using an internal reference (a ferrocene/ferrocenium redox couple), and compare these with redox potentials computed at B3LYP/6–31+G­(d,p) using a polarizable continuum model to account for solvation. Unlike previous large-scale computational screening studies, this work carefully establishes the accuracy of the computational procedure by benchmarking against experimental results. While previous small-scale computational studies have been carried out on structurally homologous compounds, this work assesses the accuracy of the computational model across a variety of compound classes, without applying class-dependent empirical corrections. We find that redox potential differences for coupled one-electron transfer processes can be computed to within 0.4 V and two-electron redox potential differences can usually be computed to within 0.15 V.

A Computational Protocol Combining DFT and Cheminformatics for Prediction of pH-Dependent Redox Potentials

“…The main goal of this contribution is the establishment and validation of a standard procedure for the computational prediction of redox potentials of organic molecules undergoing a proton-coupled electron transfer. One of the main motivations for benchmarking such predictions is the computational screening of the vast chemical space of organic molecules to identify electrolytes for redox flow batteries (RFBs) [ 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 ]. At some point in any computational workflow for materials discovery, one needs a reliable method to predict with reasonable accuracy and moderate computational cost the properties of an already pre-selected candidate pool.…”

Bridging the Experiment-Calculation Divide: Machine Learning Corrections to Redox Potential Calculations in Implicit and Explicit Solvent Models

“…6 Although low unsigned errors of less than 100 mV were observed in small batch studies, more realistic errors can be much larger when more diverse and larger test sets are considered. 26,61 In our calculation of the ROP313 data set with Compared to experimental results, the calculated redox potentials of the OROP data set show a systematic underestimation, especially in the higher value range with high data density (Figure 3). This type of systematic bias is commonly seen in implicit solvent redox potential calculations and is often corrected with simple linear fitting between the calculated and experimental redox potentials.…”

“…73 Further investigation in the performance differences between SHE and Fc + /Fc references will allow a better comparison of the results. McNeill and co-workers recently benchmarked another chemically diverse organic data set, 61 reaching an overall MAE of 0.4 V after a linear regression correction. Their work also included some two-electron transfer processes, which are more complicated than the oneelectron transfer processes investigated in this work.…”

Experimental Protocols for Studying Organic Non-aqueous Redox Flow Batteries

“…In many regards, redox potential is the first property of interest to be computed for ROMs for benchmarking the computational method. DFT has been shown to be robust for predicting the redox potentials of anolytes and catholytes, with mean absolute errors being about 0.10 V or less. ,− In order to compute the redox potential of a molecule, the geometries of the neutral and reduced/oxidized states must be first optimized. Frequency calculations are subsequently carried out on those optimized geometries to obtain corresponding Gibbs free energies ( G ) at 298 K. The reduction ( E red ) and oxidation ( E ox ) potentials, in V with respect to the Normal Hydrogen Electrode (NHE), are then calculated as follows: where G red , G ox , and G neu are the computed Gibbs free energies of the molecule in the reduced, oxidized, and neutral state of charge, respectively.…”