Design of an Accurate Machine Learning Algorithm to Predict the Binding Energies of Several Adsorbates on Multiple Sites of Metal Surfaces

Praveen, C. S.; Comas‐Vives, Aleix

doi:10.1002/cctc.202000517

Cited by 30 publications

(33 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To inspect the ML predictions in greater detail, we also utilized SHapley Additive exPlanations (SHAP) analysis [71–74] . As illustrated in Figure 6, SHAP is a method in which the predicted value is decomposed into the additive sum of contributions from individual feature values.…”

Section: Resultsmentioning

confidence: 99%

“…Technically speaking, SHAP is grounded in optimal credit allocation in cooperative game theory and is computed as the difference between the predicted values with and without the feature value averaged over all possible combinations for other feature values. In general, this quantity is computationally intractable (NP‐hard), but for tree models such as RFR, ETR, and XGB, efficient (polynomial‐time) exact algorithms are available [71–74] …”

Section: Resultsmentioning

confidence: 99%

“…The results were similar to those obtained using the feature importance score analysis, confirming the high reliability of the method. As described above, SHAP analysis can also be used to visualize the dependence of the predicted C 2 yield on the value of each descriptor [71–74] . For example, high values (red color in Figure 7) for the feature “reaction temperature” are correlated to high C 2 yield (SHAP value).…”

Section: Resultsmentioning

confidence: 99%

See 2 more Smart Citations

Analysis of Updated Literature Data up to 2019 on the Oxidative Coupling of Methane Using an Extrapolative Machine‐Learning Method to Identify Novel Catalysts

et al. 2021

View full text Add to dashboard Cite

We have constructed and analyzed an updated dataset consisting of 4759 experimental datapoints for the oxidative coupling of methane (OCM) reaction based on literature data reported before 2020 (∼2019) using machine learning (ML) methods. Several ML methods, including random forest regression (RFR), extra trees regression (ETR), and gradient boosting regression with XGBoost (XGB), were used in conjunction with our proposed approach, in which elemental features are used as input representations rather than inputting the catalyst compositions directly. A recent research trend, namely, the extensive exploration of Mn/Na2WO4/SiO2 catalyst systems in recent years due to their high activity and durability, was clearly reflected in the dataset analysis. An ML model for the prediction of the reaction outcome (C2 yield) was successfully developed, and feature importance scores and SHapley Additive exPlanations (SHAP) values were calculated based on ETR and XGB, respectively, to identify the input variables with the greatest influence on the catalyst performance and observe how these important variables affect the C2 yield in the OCM. The discovery and optimization of catalytic processes using ML as a “surrogate” model were explored, and promising catalytic system candidates for the OCM reaction were identified. Notably, the developed ML model predicted catalysts containing elements that do not appear in the OCM dataset. This clearly demonstrates desirably high potential of our ML model to enable extrapolative predictions for ML‐aided future catalysis research.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Analysis of Updated Literature Data up to 2019 on the Oxidative Coupling of Methane Using an Extrapolative Machine‐Learning Method to Identify Novel Catalysts

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Highthroughput studies can be accelerated by exploiting also surrogate models, i.e., efficient, empirical models that can produce property predictions such as adsorption energies, albeit less accurately than a first-principles-based model such as density functional theory (DFT) [110]. Surrogate models can be scaling relationships [111][112][113], physical descriptors [114][115][116][117], or machine learning (ML) models trained on physical or structural descriptors [118][119][120][121][122][123][124][125][126][127][128][129][130][131][132][133][134]. Furthermore, they can be enhanced by stability analysis to save computing time on unstable materials [135].…”

Section: Computational Catalysis and Mechanism Explorationmentioning

confidence: 99%

Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis

Steiner

Reiher

2022

Top Catal

View full text Add to dashboard Cite

Autonomous computations that rely on automated reaction network elucidation algorithms may pave the way to make computational catalysis on a par with experimental research in the field. Several advantages of this approach are key to catalysis: (i) automation allows one to consider orders of magnitude more structures in a systematic and open-ended fashion than what would be accessible by manual inspection. Eventually, full resolution in terms of structural varieties and conformations as well as with respect to the type and number of potentially important elementary reaction steps (including decomposition reactions that determine turnover numbers) may be achieved. (ii) Fast electronic structure methods with uncertainty quantification warrant high efficiency and reliability in order to not only deliver results quickly, but also to allow for predictive work. (iii) A high degree of autonomy reduces the amount of manual human work, processing errors, and human bias. Although being inherently unbiased, it is still steerable with respect to specific regions of an emerging network and with respect to the addition of new reactant species. This allows for a high fidelity of the formalization of some catalytic process and for surprising in silico discoveries. In this work, we first review the state of the art in computational catalysis to embed autonomous explorations into the general field from which it draws its ingredients. We then elaborate on the specific conceptual issues that arise in the context of autonomous computational procedures, some of which we discuss at an example catalytic system. Graphical Abstract

show abstract

“…Recently, Praveen and Comas-Vives [24] designed a highly accurate ML algorithm, able to predict the adsorption energies of several adsorbates binding either via C, N, O, or H on the surface sites of (100), (111), and (211) facets of transition-metals simultaneously. They combined electronic and structural descriptors (such as CN, GCN, and others) of free adsorbates on clean metal surfaces (10-13 features in total) and applied extragradient boost regression in combination with a tree booster to obtain the best performance.…”

Section: Introductionmentioning

confidence: 99%

Adsorption Sites on Pd Nanoparticles Unraveled by Machine-Learning Potential with Adaptive Sampling

et al. 2022

View full text Add to dashboard Cite

Catalytic properties of noble-metal nanoparticles (NPs) are largely determined by their surface morphology. The latter is probed by surface-sensitive spectroscopic techniques in different spectra regions. A fast and precise computational approach enabling the prediction of surface–adsorbate interaction would help the reliable description and interpretation of experimental data. In this work, we applied Machine Learning (ML) algorithms for the task of adsorption-energy approximation for CO on Pd nanoclusters. Due to a high dependency of binding energy from the nature of the adsorbing site and its local coordination, we tested several structural descriptors for the ML algorithm, including mean Pd–C distances, coordination numbers (CN) and generalized coordination numbers (GCN), radial distribution functions (RDF), and angular distribution functions (ADF). To avoid overtraining and to probe the most relevant positions above the metal surface, we utilized the adaptive sampling methodology for guiding the ab initio Density Functional Theory (DFT) calculations. The support vector machines (SVM) and Extra Trees algorithms provided the best approximation quality and mean absolute error in energy prediction up to 0.12 eV. Based on the developed potential, we constructed an energy-surface 3D map for the whole Pd55 nanocluster and extended it to new geometries, Pd79, and Pd85, not implemented in the training sample. The methodology can be easily extended to adsorption energies onto mono- and bimetallic NPs at an affordable computational cost and accuracy.

show abstract

Design of an Accurate Machine Learning Algorithm to Predict the Binding Energies of Several Adsorbates on Multiple Sites of Metal Surfaces

Cited by 30 publications

References 51 publications

Analysis of Updated Literature Data up to 2019 on the Oxidative Coupling of Methane Using an Extrapolative Machine‐Learning Method to Identify Novel Catalysts

Analysis of Updated Literature Data up to 2019 on the Oxidative Coupling of Methane Using an Extrapolative Machine‐Learning Method to Identify Novel Catalysts

Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis

Adsorption Sites on Pd Nanoparticles Unraveled by Machine-Learning Potential with Adaptive Sampling

Contact Info

Product

Resources

About