Conformational Sampling over Transition-Metal-Catalyzed Reaction Pathways: Toward Revealing Atroposelectivity.

The Py-Conformational-Sampling (PyCoSa) technique is introduced as a systematic computational means to sample the configurational space of transition-metal-catalyzed stereoselective reactions. When applied to atroposelective Suzuki-Miyaura coupling to create axially chiral biaryl products, the results show a range of mechanistic possibilities that include multiple low-energy channels through which C-C bonds can be formed.

Advances in Parallel Heat Bath Configuration Interaction.

Heat-bath configuration interaction (HCI) is a deterministic method that approaches the full CI limit at greatly reduced computational cost. In this work, computational improvements to the HCI algorithm are introduced targeting speed, parallel efficiency, and memory requirements. The new implementation introduces a hash function to distribute determinants and takes advantage of MPI and OpenMP for parallelism allowing for a (22e,168o) active space to be studied, which explicitly includes 2.39 × 107 variational determinants and 8.95 × 1010 perturbative determinants. Benchmarks show up to 86% parallel efficiency of the perturbative step on 32 nodes (4096 cores) and a total efficiency of 74%. The new HCI implementation is benchmarked for accuracy against prior results and applied to study the triplet-quintet gap in the challenging [FeO(NH3)5]2+ complex.

Predicting reaction conditions from limited data through active transfer learning.

Transfer and active learning have the potential to accelerate the development of new chemical reactions, using prior data and new experiments to inform models that adapt to the target area of interest. This article shows how specifically tuned machine learning models, based on random forest classifiers, can expand the applicability of Pd-catalyzed cross-coupling reactions to types of nucleophiles unknown to the model. First, model transfer is shown to be effective when reaction mechanisms and substrates are closely related, even when models are trained on relatively small numbers of data points. Then, a model simplification scheme is tested and found to provide comparative predictivity on reactions of new nucleophiles that include unseen reagent combinations. Lastly, for a challenging target where model transfer only provides a modest benefit over random selection, an active transfer learning strategy is introduced to improve model predictions. Simple models, composed of a small number of decision trees with limited depths, are crucial for securing generalizability, interpretability, and performance of active transfer learning.

What Does the Machine Learn? Knowledge Representations of Chemical Reactivity.

In a departure from conventional chemical approaches, data-driven models of chemical reactions have recently been shown to be statistically successful using machine learning. These models, however, are largely black box in character and have not provided the kind of chemical insights that historically advanced the field of chemistry. To examine the knowledgebase of machine-learning models-what does the machine learn-this article deconstructs black-box machine-learning models of a diverse chemical reaction data set. Through experimentation with chemical representations and modeling techniques, the analysis provides insights into the nature of how statistical accuracy can arise, even when the model lacks informative physical principles. By peeling back the layers of these complicated models we arrive at a minimal, chemically intuitive model (and no machine learning involved). This model is based on systematic reaction-type classification and Evans-Polanyi relationships within reaction types which are easily visualized and interpreted. Through exploring this simple model, we gain deeper understanding of the data set and uncover a means for expert interactions to improve the model's reliability.

Discovery of conical intersection mediated photochemistry with growing string methods.

Conical intersections (CIs) are important features of photochemistry that determine yields and selectivity. Traditional CI optimizers require significant human effort and chemical intuition, which typically restricts searching to only a small region of the CI space. Herein, a systematic approach utilizing the growing string method is introduced to locate multiple CIs. Unintuitive MECI are found using driving coordinates that can be generated using a combinatorial search, and subsequent optimization allows reaction pathways, transition states, products, and seam-space pathways to be located. These capabilities are demonstrated by application to two prototypical photoisomerization reactions and the dimerization of butadiene. In total, many reaction pathways were uncovered, including the elusive stilbene hula-twist mechanism, and a previously unidentified product in butadiene dimerization. Overall, these results suggest that growing string methods provide a predictive strategy for exploring photochemistry.

Estimating the Derivative Coupling Vector Using Gradients.

The difference gradient and derivative coupling vectors span the branching planes of conical intersections between electronic states. While gradients are commonly available in many electronic structure methods, the derivative coupling vectors are not always implemented and ready for use in characterizing conical intersections. This Letter shows how the derivative coupling vectors can be computed to high accuracy (direction and magnitude) using energy and gradient information. The new method is based on the combination of a linear-coupling two-state Hamiltonian and a finite-difference Davidson approach for computing the branching plane. Benchmark cases are provided showing these vectors can be efficiently computed near conical intersections.