Machine Learning-Assisted Screening of Stepped Alloy Surfaces for C1 Catalysis

Design and screening of bimetallic catalysts for nitric oxide reduction by CO: a study of kinetic Monte Carlo simulation based on first-principles calculations

Nitric oxide (NO) emissions pose a significant environmental challenge, and the development of effective catalysts for NO reduction is crucial. This study investigates the potential of striped bimetallic catalysts for NO reduction by CO using kinetic Monte Carlo (KMC) simulations based on first-principles calculations. The simulations reveal that the activity on the striped Ni-Pt-Pt (111) surface is 1-2 orders of magnitude higher than that on the terraced Ni-Pt-Pt (111) surface at the same temperatures, demonstrating the importance of defect engineering. Sensitivity analysis identifies CO oxidation as the rate-determining step, although the 2N* association barrier is higher than CO oxidation, highlighting the need to consider reaction conditions in kinetic simulations. Volcano plots based on the formation energies of NO* and CO* successfully predict the striped Ni-Pd-Pd (111) and Ni-Rh-Rh (111) surfaces as optimal catalysts, which were further validated through DFT calculations and ab initio molecular dynamics simulations. This study offers valuable insights for designing high-performance bimetallic catalysts for NO reduction and underscores the importance of considering specific reaction conditions in kinetic simulations.

Discovery of highly efficient dual-atom catalysts for propane dehydrogenation assisted by machine learning

Propane dehydrogenation (PDH) is a highly efficient approach for industrial production of propylene, and the dual-atom catalysts (DACs) provide new pathways in advancing atomic catalysis for PDH with dual active sites. In this work, we have developed an efficient strategy to identify promising DACs for PDH reaction by combining high-throughput density functional theory (DFT) calculations and the machine-learning (ML) technique. By choosing the γ-Al2O3(100) surface as the substrate to anchor dual metal atoms, 435 kinds of DACs have been considered to evaluate their PDH catalytic activity. Four ML algorithms are employed to predict the PDH activity and determine the relationship between the intrinsic characteristics of DACs and the catalytic activity. The promising catalysts of CuFe, CuCo and CoZn DACs are finally screened out, which are further validated by the whole kinetic reaction calculations, and the highly efficient performance of DACs is attributed to the synergistic effects and interactions between the paired active sites.

Interpreting chemisorption strength with AutoML-based feature deletion experiments

The chemisorption energy of reactants on a catalyst surface, [Formula: see text], is among the most informative characteristics of understanding and pinpointing the optimal catalyst. The intrinsic complexity of catalyst surfaces and chemisorption reactions presents significant difficulties in identifying the pivotal physical quantities determining [Formula: see text]. In response to this, the study proposes a methodology, the feature deletion experiment, based on Automatic Machine Learning (AutoML) for knowledge extraction from a high-throughput density functional theory (DFT) database. The study reveals that, for binary alloy surfaces, the local adsorption site geometric information is the primary physical quantity determining [Formula: see text], compared to the electronic and physiochemical properties of the catalyst alloys. By integrating the feature deletion experiment with instance-wise variable selection (INVASE), a neural network-based explainable AI (XAI) tool, we established the best-performing feature set containing 21 intrinsic, non-DFT computed properties, achieving an MAE of 0.23 eV across a periodic table-wide chemical space involving more than 1,600 types of alloys surfaces and 8,400 chemisorption reactions. This study demonstrates the stability, consistency, and potential of AutoML-based feature deletion experiment in developing concise, predictive, and theoretically meaningful models for complex chemical problems with minimal human intervention.

High-Throughput Screening of Sulfur-Resistant Catalysts for Steam Methane Reforming Using Machine Learning and Microkinetic Modeling

The catalytic activity of bimetallic catalysts for the steam methane reforming (SMR) reaction was extensively studied previously. However, the performance of these materials in the presence of sulfur-containing species is yet to be investigated. In this study, we propose a novel process aided by machine learning (ML) and microkinetic modeling for the rapid screening of sulfur-resistant bimetallic catalysts. First, various ML models were developed to predict atomic adsorption energies (C, H, O, and S) on bimetallic surfaces. Easily accessible physical and chemical properties of the metals and adsorbates were used as input features. The Ensemble learning, artificial neural network, and support vector regression models achieved the best performance with R2 values of 0.74, 0.71, and 0.70, respectively. A microkinetic model was then built based on the elementary steps of the SMR reaction. Finally, the microkinetic model, together with the atomic adsorption energies predicted by the Ensemble model, were used to screen over 500 bimetallic materials. Four Ge-based alloys (Ge3Cu1, Ge3Ni1, Ge3Co1, and Ge3Fe1) and the Ni3Cu1 alloy were identified as promising and cost-effective sulfur-resistant catalysts.

Charge Recombination Dynamics in a Metal Halide Perovskite Simulated by Nonadiabatic Molecular Dynamics Combined with Machine Learning

Nonadiabatic coupling (NAC) plays a central role in driving nonadiabatic dynamics in various photophysical and photochemical processes. However, the high computational cost of NAC limits the time scale and system size of quantum dynamics simulation. By developing a machine learning (ML) framework and applying it to a traditional CH₃N₃PbI₃ perovskite, we demonstrate that the various ML algorithms (XGBoost, LightGBM, and random forest) combined with three descriptors (sine matrix, MBTR, and SOAP) can predict accurate NACs that all agree well with the direct calculations, particularly for the combination of LightGBM and sine matrix descriptor showing the best performance with a high correlation coefficient of ≤0.87. The simulated nonradiative electron-hole recombination time scales agree well with each other between the NACs obtained from direct calculations and ML prediction. The study shows the advantage in accelerating quantum dynamics simulations using ML algorithms.