Artificial neural networks for predicting charge transfer coupling

Abstract

Quantum chemistry calculations have been very useful in providing many key detailed properties and enhancing our understanding of molecular systems. However, such calculation, especially with ab initio models, can be time-consuming. For example, in the prediction of charge-transfer properties, it is often necessary to work with an ensemble of different thermally populated structures. A possible alternative to such calculations is to use a machine-learning based approach. In this work, we show that the general prediction of electronic coupling, a property that is very sensitive to intermolecular degrees of freedom, can be obtained with artificial neural networks, with improved performance as compared to the popular kernel ridge regression method. We propose strategies for optimizing the learning rate and batch size, improving model performance, and further evaluating models to ensure that the physical signatures of charge-transfer coupling are well reproduced. We also address the effect of feature representation as well as statistical insights obtained from the loss function and the data structure. Our results pave the way for designing a general strategy for training such neural-network models for accurate prediction.

Molecular excited states through a machine learning lens

“…Therefore, ML has been used to learn various intermediate properties such as bandgaps, 95,111,112,[200][201][202] band edges, 203 intramolecular reorganization energies, 180,182,183,204,205 delayed fluorescence rate constant, 181 decay rates of emitters, 206 exciton-transfer times and transfer efficiencies, 177 and electronic couplings. 172,[207][208][209][210][211][212][213][214] All these properties were calculated with QM methods, and ML was later used as a surrogate model for QM to accelerate screening (FIG. 6).…”

“…[180][181][182][183]216 However, because intermediate properties are mostly QM properties, it is natural to use common ML approaches that were specifically designed for learning other molecular QM properties. For example, ML models that use 3D structural descriptors for the calculation of ground-state energies often serve as the basis for models for the calculation of excited-state energies, electronic couplings 172,207,[209][210][211]213 and absorption wavelengths (see the previous sections and Box 3). 215 The main advantage of learning intermediate rather than ultimate properties is that it is usually easier to generate additional reference data for the former, opening much more efficient and economical ways for materials design.…”

Advanced Machine Learning Methods for Learning from Sparse Data in High-Dimensional Spaces: A Perspective on Uses in The Upstream of Development of Novel Energy Technologies

“…One area where ML is gaining more and more traction are novel energy conversion and storage technologies. These techniques are, in particular, intensely explored for application to the development of technologies typically associated with sustainable generation and use of energy such as advanced types (organic and inorganic materials based) of solar cells and LED (light-emitting diodes) [10][11][12][13][14][15][16][17][18][19][20][21][22], inorganic and organic metal ion batteries [23,24], fuel cells and generally heterogeneous catalysis including electro-and photocatalysis [25][26][27][28][29][30][31][32][33][34]. This is natural in the sense that the development of these technologies often passes through optimization and balancing of multiple factors acting simultaneously and to opposite ends; for example, in the case of organic solar cells, there is an optimum to be sought between the donor's bandgap, the band offset between the donor and the acceptor, the reorganization energies of both the donor and the acceptor, the charge transfer integral etc.…”

Stacked Ensemble Machine Learning for Range-Separation Parameters

“…83 Corminboeuf and coworkers designed a ML-based ωPBE functional that recovered the exact DD using the piecewise relationship between the average energy curvature and the electronic number, and reduced the error of the fundamental gap (E g ) of large hole-transporting molecular materials from 0.54 eV (by one version of LC-ωPBE with ω = 0.400a −1 0 ) to 0.15 eV. 84 Compared to direct ML predictions of excited state properties, [96][97][98][99][100] these ML-RSH schemes maintained rigorous solutions of time-(in)dependent Kohn-Sham equations along with valid physical meanings, although they sometime suffered from overfitting and undergeneralization problems due to the limited size and diversity of training sets.…”