Abstract:Microkinetic models for catalytic systems require estimation of many thermodynamic and kinetic parameters that can be calculated for isolated species and transition states using ab initio methods. However, the presence of nearby co-adsorbates on the surface can dramatically alter these thermodynamic and kinetic parameters causing them to be dependent on species coverage fractions. As there are combinatorially many co-adsorbed configurations on the surface, computing the coverage dependence of these parameters is far less straightforward. We present a framework for generating and applying machine learning models to predict coverage dependent parameters for microkinetic models. Our toolkit enables automatic calculation and evaluation of co-adsorbed configurations allowing us to sample 2000 co-adsorbed adsorbates and transition states (TSs) for a diverse set of 9 reactions on Cu111, a challenging surface, with four possible co-adsorbates. This dataset was then used to train subgraph isomorphic decision trees (SIDTs) to predict the stability and association energy of configurations. With which we were able to achieve mean absolute errors (MAEs) of 0.106 eV on adsorbates, 0.172 eV on TSs, and due to natural error cancellation in SIDTs for relative properties 0.130 eV on reaction energies and 0.180 eV on activation barriers. We then explain how to use these models to predict coverage dependent corrections for arbitrary adsorbates and TSs and demonstrate on H∗, HO∗ and O∗ comparing the generated SIDT model with an iteratively refined version.

What problem does this paper attempt to address?

### Problems Addressed by the Paper This paper aims to address the issue of coverage dependence in surface reaction kinetics. Specifically, microkinetic models in catalytic systems require the estimation of numerous thermodynamic and kinetic parameters, which can be obtained by calculating isolated species and transition states using ab initio methods. However, these thermodynamic and kinetic parameters change significantly when neighboring co-adsorbates are present on the surface, thus depending on the species' coverage. Due to the presence of multiple co-adsorption configurations on the surface, calculating the coverage dependence of these parameters becomes very complex. ### Main Challenges 1. **Computational Complexity**: Even when considering only one type of co-adsorbate, calculating the minimum binding energy at a given coverage is very expensive. Considering multiple co-adsorbates is computationally infeasible. 2. **Limitations of Existing Methods**: - **Periodic Unit Cell Size Adjustment**: This method assumes that the minimum energy coverage pattern is periodic, but in many cases, the lowest energy configuration cannot be captured with a reduced unit cell. - **Cluster Expansion Model (CE)**: CE parameters are specific to the computed adsorbates and interaction terms, thus can only be used for specific combinations of adsorbates and co-adsorbates, and cannot predict interactions of new adsorbates or co-adsorbates. - **Deep Neural Networks (DNNs)**: Although DNNs can learn from diverse datasets, they require a large amount of training data and lack interpretability, making them difficult to analyze and improve. ### Solution The authors propose a framework to predict coverage-dependent parameters in microkinetic models by generating and applying machine learning models. The specific steps include: 1. **Automated Calculation of Co-adsorption Configurations**: Using an automated quantum chemistry workflow to generate 2000 co-adsorption adsorbates and transition states (TSs), involving 9 different reactions and 4 possible co-adsorbates. 2. **Training Subgraph Isomorphic Decision Trees (SIDTs)**: Using the generated dataset to train SIDTs to predict stability and interaction energies. SIDTs can handle small datasets, are easy to scale and retrain, and have inherent visualization and interpretability. 3. **Error Cancellation Properties**: SIDTs exhibit significant error cancellation properties in predicting relative properties such as reaction energies and activation barriers, improving prediction accuracy. ### Experimental Results - **Stability Prediction**: Classification accuracy on the test set is 86%, with a precision of 84%. - **Interaction Energy Prediction**: Overall mean absolute error (MAE) is 0.126 eV, with an MAE of 0.106 eV for adsorbates and 0.172 eV for transition states. - **Activation Barrier and Reaction Energy Correction Prediction**: MAE for activation barriers is 0.180 eV, and for reaction energies, it is 0.130 eV, showing significant error cancellation effects. ### Conclusion By using SIDTs, the authors successfully addressed the computational problem of coverage-dependent parameters in surface reaction kinetics, providing a high-accuracy prediction method and demonstrating its effectiveness in practical applications.

Resolving the Coverage Dependence of Surface Reaction Kinetics with Machine Learning and Automated Quantum Chemistry Workflows

Computing Surface Reaction Rates by Adaptive Multilevel Splitting Combined with Machine Learning and Ab Initio Molecular Dynamics

Comprehensive sampling of coverage effects in catalysis by leveraging generalization in neural network models

Combining Machine Learning and Many-Body Calculations: Coverage-Dependent Adsorption of CO on Rh(111)

Coverage-Dependent Microkinetics in Heterogeneous Catalysis Powered by the Maximum Rate Analysis

A Machine Learning Based Approach to Reaction Rate Estimation

Controlling neural network extrapolation enables efficient and comprehensive sampling of coverage effects in catalysis

Understanding Surface Catalyzed Decomposition Reactions Using a Chemical Pathway Analysis

Achieving Theory–Experiment Parity for Activity and Selectivity in Heterogeneous Catalysis Using Microkinetic Modeling

Machine-learning driven global optimization of surface adsorbate geometries

Influence of coverage dependence on the thermophysical properties of adsorbates and its impact on microkinetic models

Toward accurate and efficient dynamic computational strategy for heterogeneous catalysis: Temperature-dependent thermodynamics and kinetics for the chemisorbed on-surface CO

Data-efficient modeling of catalytic reactions via enhanced sampling and on-the-fly learning of machine learning potentials

Adsorption Enthalpies for Catalysis Modeling through Machine-Learned Descriptors

Ab initio surface chemistry with chemical accuracy

Beyond Mean-Field Microkinetics: Toward Accurate and Efficient Theoretical Modeling in Heterogeneous Catalysis

An Atomistic Machine Learning Package for Surface Science and Catalysis

Predicting Chemical Reaction Barriers with a Machine Learning Model

Benchmarking of machine learning interatomic potentials for reactive hydrogen dynamics at metal surfaces

A general method for studying reactive surface species, CT-SKAn: Charge-transfer spectrokinetic analysis