Themed collection Insightful Machine Learning for Physical Chemistry

This themed collection includes a collection of articles on Insightful Machine Learning for Physical Chemistry.

A basis expansion view of popular ML methods is useful and can explain their properties and pitfalls, in particular in high-dimensional spaces and under low density, uneven data distribution.

Machine learning can simultaneously infer multiple physics-consistent material properties from electroanalytical tests, as well as describe underlying field variations.

Predicting drop coalescence based on process parameters is crucial for experimental design in chemical engineering.

We investigated the various factors impacting the performance of Δ-machine learning (Δ-ML) solution phase molecular properties.

Molecular fragments of metal–organic frameworks can be used to construct high-dimensional neural network potentials. Here we provide a recipe of how the smallest possible fragments can be chosen that still provide a HDNNP transferable to the bulk crystal.

A fast machine learning based framework is introduced for the prediction of solubility parameters and selection of green solvents for small molecular donor-based organic solar cells.

Artificial neural networks trained on 23 density functional approximations (DFAs) from multiple rungs of “Jacob's ladder” enable the prediction of where each DFA has zero curvature for chemical discovery.

A deep learning potential distinct from the empirical potential is developed for the study of thermal transport across solid–liquid interfaces.

We study the capability of transfer learning for efficiently generating chemically accurate interatomic neural network potentials.

A prototypical Eley–Rideal reaction between incident H/D atoms and pre-covered D/H atoms on Cu (111) is studied by molecular dynamics simulations using a neural network potential with first-principles accuracy.

We develop Bayesian Chemical Reaction Neural Network (B-CRNN), a method to infer chemical reaction models and provide the associated uncertainty purely from data without prior knowledge of reaction templates.

We present an approach to predict the group-interaction parameters of thermodynamic group contribution (GC) methods based on the machine-learning concept of matrix completion and thereby substantially extend the scope of GC methods.

We present explainable machine learning approaches for understanding and predicting free energies, enthalpies, and entropies of ion pairing in different solvents.

We present machine learning models trained on experimental data to predict room-temperature solubility for any polymer–solvent pair.

We proposed a “hierarchical” protocol based on the unsupervised machine learning algorithms (principal component analysis and clustering approaches) to automatically analyze the ring deformation in the nonadiabatic molecular dynamics.

Deep reinforcement learning can be used as an efficient artificial intelligence approach to control time-dependent quantum dynamical systems.

The use of graph convolutional neural networks for mixed (QM)ML/MM molecular dynamics simulations of condensed-phase systems is investigated and benchmarked. We find that a Δ-learning scheme using DFTB as a baseline achieves the best performance.