Machine Learning Models for Predicting Zirconocene Properties and Barriers.

Zr metallocenes have significant potential to be highly tunable polyethylene catalysts through modification of the aromatic ligand framework. Here we report the development of multiple machine learning models using a large library (>700 systems) of DFT-calculated zirconocene properties and barriers for ethylene polymerization. We show that very accurate machine learning models are possible for HOMO-LUMO gaps of precatalysts but the performance significantly depends on the machine learning algorithm and type of featurization, such as fingerprints, Coulomb matrices, smooth overlap of atomic positions, or persistence images. Surprisingly, the description of the bonding hapticity, the number of direct connections between Zr and the ligand aromatic carbons, only has a moderate influence on the performance of most models. Despite robust models for HOMO-LUMO gaps, these types of machine learning models based on structure connectivity type features perform poorly in predicting ethylene migratory insertion barrier heights. Therefore, we developed several relatively robust and accurate machine learning models for barrier heights that are based on quantum-chemical descriptors (QCDs). The quantitative accuracy of these models depends on which potential energy surface structure QCDs were harvested from. This revealed a Hammett-type principle to naturally emerge showing that QCDs from the π-coordination complexes provide much better descriptions of the transition states than other potential-energy structures. Feature importance analysis of the QCDs provides several fundamental principles that influence zirconocene catalyst reactivity.

Efficient Embedded Cluster Density Approximation Calculations with an Orbital-Free Treatment of Environments.

The computational cost of the Kohn-Sham density functional theory (KS-DFT), employing advanced orbital-based exchange-correlation (XC) functionals, increases quickly for large systems. To tackle this problem, we recently developed a local correlation method in the framework of KS-DFT: the embedded cluster density approximation (ECDA). The aim of ECDA is to obtain accurate electronic structures in an entire system. With ECDA, for each atom in a system, we define a cluster to enclose that atom, with the rest atoms treated as the environment. The system's electron density is then partitioned among the cluster and the environment. The cluster's XC energy density is then calculated based on its electron density using an advanced orbital-based XC functional. The system's XC energy is obtained by patching all clusters' XC energy densities in an atom-by-atom manner. In our previous formulation of ECDA, environments were treated by KS-DFT, which makes the following two tasks computationally expensive for large systems. The first task is to partition the system's electron density among a cluster and its environment. The second task is to solve the environments' Sternheimer equations for calculating the system's XC potential. In this work, we remove these two computational bottlenecks by treating the environments with the orbital-free (OF) DFT. The new method is called ECDA-envOF. The performance of ECDA-envOF is examined in two systems: ester and Cl-tetracene, for which the exact exchange (EXX) is used as the advanced XC functional. We show that ECDA-envOF gives results that are very close to the previous formulation in which the environments were treated by KS-DFT. Therefore, ECDA-envOF can be used for future large-scale simulations. Another focus of this work is to examine ECDA-envOF's performance on systems having different bond types. With ECDA-envOF, we calculate the energy curves for stretching/compressing some covalent, metallic, and ionic systems. ECDA-envOF's predictions agree well with the benchmarks by using reasonably large clusters. These examples demonstrate that ECDA-envOF is nearly a black-box local correlation method for investigating heterogeneous materials in which different bond types exist.