ABSTRACT
Due to the limitation of solvent models, quantum chemistry calculation of solution-phase molecular properties often deviates from experimental measurements. Recently, Δ-machine learning (Δ-ML) was shown to be a promising approach to correcting errors in the quantum chemistry calculation of solvated molecules. However, this approach's applicability to different molecular properties and its performance in various cases are still unknown. In this work, we tested the performance of Δ-ML in correcting redox potential and absorption energy calculations using four types of input descriptors and various ML methods. We sought to understand the dependence of Δ-ML performance on the property to predict the quantum chemistry method, the data set distribution/size, the type of input feature, and the feature selection techniques. We found that Δ-ML can effectively correct the errors in redox potentials calculated using density functional theory (DFT) and absorption energies calculated by time-dependent DFT. For both properties, the Δ-ML-corrected results showed less sensitivity to the DFT functional choice than the raw results. The optimal input descriptor depends on the property, regardless of the specific ML method used. The solvent-solute descriptor (SS) is the best for redox potential, whereas the combined molecular fingerprint (cFP) is the best for absorption energy. A detailed analysis of the feature space and the physical foundation of different descriptors well explained these observations. Feature selection did not further improve the Δ-ML performance. Finally, we analyzed the limitation of our Δ-ML solvent effect approach in data sets with molecules of varying degrees of electronic structure errors.
ABSTRACT
Semiclassical electrons (aka Lewis dots) have been a mainstay of chemists' thinking about molecular structure, polarizability, and reactivity for over a century. This utility has motivated the development of a corresponding quantitative description. Here we devise pairwise potentials that describe the behavior of valence electron pairs in hydrocarbons, including those in single, double, bridge, and bent bonds of linear, branched, and cyclic compounds, including anionic and cationic states. Beyond predicting structures and energies, the new subatomistic force field, dubbed LEWIS-B, efficiently simulates carbocation addition to a double bond and cation migration to a neighboring carbon. A crucial feature of the semiclassical electrons is variable spread, a fourth degree of freedom in addition to three Cartesian coordinates. In spontaneously adapting to different environments, the spread provides a signature of electron stability, with more contracted clouds where the electron interactions are favorable and expanded clouds where electrons are less tightly held. In addition, the pair potentials provide insight into the subtle trade-offs that govern isomerizations and reactions.
ABSTRACT
As very light fermions, electrons are governed by antisymmetric wave functions that lead to exchange integrals in the evaluation of the energy. Here we use the localized representation of orbitals to decompose the electronic energy in a fashion that isolates the enigmatic exchange contributions and characterizes their distinctive control over electron distributions. The key to this completely general analysis is considering the electrons in groups of three, drawing attention to the curvatures of pair potentials, rather than just their amplitudes and slopes. We show that a positive curvature at short distances is essential for the mutual distancing of electrons and a negative curvature at longer distances is essential to account for the influence of lone pairs on bond torsion. Neither curvature is available in the absence of the exchange contributions. Thus, although exchange energies are much shorter range than Coulomb energies, their influence on molecular geometry is profound and readily understood.
