Practical Deep-Learning Representation for Fast Heterogeneous Catalyst Screening

Hydrogen, Oxygen, and Lead Adsorbates on Al13Co4(100): Accurate Potential Energy Surfaces at Low Computational Cost by Machine Learning and DFT-Based Data

Intermetallic compounds are promising materials in numerous fields, especially those involving surface interactions, such as catalysis. A key factor to investigate their surface properties lies in adsorption energy maps, typically built using first-principles approaches. However, exploring the adsorption energy landscapes of intermetallic compounds can be cumbersome, usually requiring huge computational resources. In this work, we propose an efficient method to predict adsorption energies, based on a Machine Learning (ML) scheme fed by a few Density Functional Theory (DFT) estimates performed on n sites selected through the Farthest Point Sampling (FPS) process. We detail its application on the Al13Co4(100) quasicrystalline approximant surface for several atomic adsorbates (H, O, and Pb). On this specific example, our approach is shown to outperform both simple interpolation strategies and the recent ML force field MACE [arXiv.2206.07697], especially when the number n is small, i.e., below 36 sites. The ground-truth DFT adsorption energies are much more correlated with the predicted FPS-ML estimates (Pearson R-factor of 0.71, 0.73, and 0.90 for H, O and Pb, respectively, when n = 36) than with interpolation-based or MACE-ML ones (Pearson R-factors of 0.43, 0.39, and 0.56 for H, O, and Pb, in the former case and 0.22, 0.35, and 0.63 in the latter case). The unbiased root-mean-square error (ubRMSE) is lower for FPS-ML than for interpolation-based and MACE-ML predictions (0.15, 0.17, and 0.17 eV, respectively, for hydrogen and 0.17, 0.25, and 0.22 eV for lead), except for oxygen (0.55, 0.47, and 0.46 eV) due to large surface relaxations in this case. We believe that these findings and the corresponding methodology can be extended to a wide range of systems, which will motivate the discovery of novel functional materials.

CatEmbed: A Machine-Learned Representation Obtained via Categorical Entity Embedding for Predicting Adsorption and Reaction Energies on Bimetallic Alloy Surfaces

Machine-learning models for predicting adsorption energies on metallic surfaces often rely on basic elemental properties and electronic and geometric descriptors. Here, we apply categorical entity embedding, a featurization method inspired by natural language processing techniques, to predict adsorption energies on bimetallic alloy surfaces using categorical descriptors. Using this method, we develop a machine-learned representation from categorical descriptors (e.g., surface composition, adsorbate type, and site type) of the slab/adsorbate complex. By combining this representation with numerical features (e.g., slab metal stoichiometric ratios), we create the CatEmbed representation. Remarkably, decision tree models trained using CatEmbed, which includes no explicit geometric information, achieve a Mean Absolute Error (MAE) of 0.12 eV. Additionally, we extend this technique to predict reaction energies on bimetallic surfaces, creating the CatEmbed-React representation, which achieves an MAE of 0.08 eV. These findings highlight the effectiveness of categorical entity embedding for predicting adsorption and reaction energies on bimetallic alloy surfaces.

Pourbaix Machine Learning Framework Identifies Acidic Water Oxidation Catalysts Exhibiting Suppressed Ruthenium Dissolution

The demand for green hydrogen has raised concerns over the availability of iridium used in oxygen evolution reaction catalysts. We identify catalysts with the aid of a machine learning-aided computational pipeline trained on more than 36,000 mixed metal oxides. The pipeline accurately predicts Pourbaix decomposition energy (Gpbx) from unrelaxed structures with a mean absolute error of 77 meV per atom, enabling us to screen 2070 new metallic oxides with respect to their prospective stability under acidic conditions. The search identifies Ru0.6Cr0.2Ti0.2O2 as a candidate having the promise of increased durability: experimentally, we find that it provides an overpotential of 267 mV at 100 mA cm-2 and that it operates at this current density for over 200 h and exhibits a rate of overpotential increase of 25 μV h-1. Surface density functional theory calculations reveal that Ti increases metal-oxygen covalency, a potential route to increased stability, while Cr lowers the energy barrier of the HOO* formation rate-determining step, increasing activity compared to RuO2 and reducing overpotential by 40 mV at 100 mA cm-2 while maintaining stability. In situ X-ray absorption spectroscopy and ex situ ptychography-scanning transmission X-ray microscopy show the evolution of a metastable structure during the reaction, slowing Ru mass dissolution by 20× and suppressing lattice oxygen participation by >60% compared to RuO2.

Accessing the usefulness of atomic adsorption configurations in predicting the adsorption properties of molecules with machine learning

We present a systematic study into the effect of adding atomic adsorption configurations into the training and validation dataset for a neural network's predictions of the adsorption energies of small molecules on single metal and bimetallic, single crystal surfaces. Specifically, we examine the efficacy of models trained with and without H and X atomic adsorption configurations, where X is C, N, or O, to predict XHn adsorption energies. In addition, we compare our machine learning models to traditional simple scaling relationships. We find that models trained with the atomic adsorption configurations outperform models trained with only molecular adsorption configurations, with as much as a 0.37 eV decrease in the MAE. We find that models trained with the atomic adsorption configurations slightly outperform traditional scaling relationships. In general, these results suggest it may be possible to vastly reduce the number of adsorption configurations one needs for training and validation datasets by supplementing said data with the adsorption configurations of composite atoms or smaller molecular fragments.

Unravelling CO2 Reduction Reaction Intermediates on High Entropy Alloy Catalysts: An Interpretable Machine Learning Approach to Establish Scaling Relations

Establishment of a scaling relation among the reaction intermediates is highly important but very much challenging on complex surfaces, such as surfaces of high entropy alloys (HEAs). Herein, we designed an interpretable machine learning (ML) approach to establish a scaling relation among CO2 reduction reaction (CO2 RR) intermediates adsorbed at the same adsorption site. Local Interpretable Model-Agnostic Explanations (LIME), Accumulated Local Effects (ALE), and Permutation Feature Importance (PFI) are used for the global and local interpretation of the utilized black box models. These methods were successfully applied through an iterative way and validated on CuCoNiZnMg and CuCoNiZnSnbased HEAs data. Finally, we successfully predicted adsorption energies of *H2 CO (MAE: 0.24 eV) and *H3 CO (MAE: 0.23 eV) by using the *HCO training data. Similarly, adsorption energy of *O (MAE: 0.32 eV) is also predicted from *H training data. We believe that our proposed method can shift the paradigm of state-of-the-art ML in catalysis towards better interpretability.

Effect of Framework Composition and NH3 on the Diffusion of Cu+ in Cu-CHA Catalysts Predicted by Machine-Learning Accelerated Molecular Dynamics

Cu-exchanged zeolites rely on mobile solvated Cu+ cations for their catalytic activity, but the role of the framework composition in transport is not fully understood. Ab initio molecular dynamics simulations can provide quantitative atomistic insight but are too computationally expensive to explore large length and time scales or diverse compositions. We report a machine-learning interatomic potential that accurately reproduces ab initio results and effectively generalizes to allow multinanosecond simulations of large supercells and diverse chemical compositions. Biased and unbiased simulations of [Cu(NH3)2]+ mobility show that aluminum pairing in eight-membered rings accelerates local hopping and demonstrate that increased NH3 concentration enhances long-range diffusion. The probability of finding two [Cu(NH3)2]+ complexes in the same cage, which is key for SCR-NOx reaction, increases with Cu content and Al content but does not correlate with the long-range mobility of Cu+. Supporting experimental evidence was obtained from reactivity tests of Cu-CHA catalysts with a controlled chemical composition.

Data-driven discovery of electrocatalysts for CO2 reduction using active motifs-based machine learning

The electrochemical carbon dioxide reduction reaction (CO2RR) is an attractive approach for mitigating CO2 emissions and generating value-added products. Consequently, discovery of promising CO2RR catalysts has become a crucial task, and machine learning (ML) has been utilized to accelerate catalyst discovery. However, current ML approaches are limited to exploring narrow chemical spaces and provide only fragmentary catalytic activity, even though CO2RR produces various chemicals. Here, by merging pre-developed ML model and a CO2RR selectivity map, we establish high-throughput virtual screening strategy to suggest active and selective catalysts for CO2RR without being limited to a database. Further, this strategy can provide guidance on stoichiometry and morphology of the catalyst to researchers. We predict the activity and selectivity of 465 metallic catalysts toward four expected reaction products. During this process, we discover previously unreported and promising behavior of Cu-Ga and Cu-Pd alloys. These findings are then validated through experimental methods.

Data-Driven Prediction of Configurational Stability of Molecule-Adsorbed Heterogeneous Catalysts

The design of new heterogeneous catalysts that convert small molecules into valuable chemicals is a key challenge for constructing sustainable energy systems. Density functional theory (DFT)-based design frameworks based on the understanding of molecular adsorption on the catalytic surface have been widely proposed to accelerate experimental approaches to develop novel catalysts. In addition, a machine learning (ML)-combined design framework was recently proposed to further reduce the inherent time cost of DFT-based frameworks. However, because of the lack of prior information on chemical interactions between arbitrary surfaces and adsorbates, the efficacy of the computational screening approaches would be reduced by obtaining unexpected structural anomalies (i.e., abnormally converged surface-adsorbate geometries after the DFT calculations) during an exhaustive exploration of chemical space. To overcome this challenge, we propose an ML framework that directly predicts the configurational stability of a given initial surface-adsorbate geometry. Our benchmark experiments with the Open Catalysts 20 (OC20) dataset show promising performance on classifying stable geometry (i.e., F1-score of 0.922, the area under the receiver operating characteristics (AUROC) of 0.906, and Matthews correlation coefficient (MCC) of 0.633) with a high precision of 0.921 by utilizing an ensemble approach. We further interpret the generalizability and domain applicability of the trained model in terms of the chemical space of the OC20 dataset. Furthermore, from an experiment on the training set size dependence of model performance, we found that our ML model could be practically applicable to classify stable configurations even with a relatively small number of training data.

Iterative multiscale and multi-physics computations for operando catalyst nanostructure elucidation and kinetic modeling

Modern heterogeneous catalysis has benefitted immensely from computational predictions of catalyst structure and its evolution under reaction conditions, first-principles mechanistic investigations, and detailed kinetic modeling, which are rungs on a multiscale workflow. Establishing connections across these rungs and integration with experiments have been challenging. Here, operando catalyst structure prediction techniques using density functional theory simulations and ab initio thermodynamics calculations, molecular dynamics, and machine learning techniques are presented. Surface structure characterization by computational spectroscopic and machine learning techniques is then discussed. Hierarchical approaches in kinetic parameter estimation involving semi-empirical, data-driven, and first-principles calculations and detailed kinetic modeling via mean-field microkinetic modeling and kinetic Monte Carlo simulations are discussed along with methods and the need for uncertainty quantification. With these as the background, this article proposes a bottom-up hierarchical and closed loop modeling framework incorporating consistency checks and iterative refinements at each level and across levels.

Fast evaluation of the adsorption energy of organic molecules on metals via graph neural networks

Modeling in heterogeneous catalysis requires the extensive evaluation of the energy of molecules adsorbed on surfaces. This is done via density functional theory but for large organic molecules it requires enormous computational time, compromising the viability of the approach. Here we present GAME-Net, a graph neural network to quickly evaluate the adsorption energy. GAME-Net is trained on a well-balanced chemically diverse dataset with C1-4 molecules with functional groups including N, O, S and C6-10 aromatic rings. The model yields a mean absolute error of 0.18 eV on the test set and is 6 orders of magnitude faster than density functional theory. Applied to biomass and plastics (up to 30 heteroatoms), adsorption energies are predicted with a mean absolute error of 0.016 eV per atom. The framework represents a tool for the fast screening of catalytic materials, particularly for systems that cannot be simulated by traditional methods.

How machine learning can accelerate electrocatalysis discovery and optimization

Advances in machine learning (ML) provide the means to bypass bottlenecks in the discovery of new electrocatalysts using traditional approaches. In this review, we highlight the currently achieved work in ML-accelerated discovery and optimization of electrocatalysts via a tight collaboration between computational models and experiments. First, the applicability of available methods for constructing machine-learned potentials (MLPs), which provide accurate energies and forces for atomistic simulations, are discussed. Meanwhile, the current challenges for MLPs in the context of electrocatalysis are highlighted. Then, we review the recent progress in predicting catalytic activities using surrogate models, including microkinetic simulations and more global proxies thereof. Several typical applications of using ML to rationalize thermodynamic proxies and predict the adsorption and activation energies are also discussed. Next, recent developments of ML-assisted experiments for catalyst characterization, synthesis optimization and reaction condition optimization are illustrated. In particular, the applications in ML-enhanced spectra analysis and the use of ML to interpret experimental kinetic data are highlighted. Additionally, we also show how robotics are applied to high-throughput synthesis, characterization and testing of electrocatalysts to accelerate the materials exploration process and how this equipment can be assembled into self-driven laboratories.

Graph neural networks for materials science and chemistry

Machine learning plays an increasingly important role in many areas of chemistry and materials science, being used to predict materials properties, accelerate simulations, design new structures, and predict synthesis routes of new materials. Graph neural networks (GNNs) are one of the fastest growing classes of machine learning models. They are of particular relevance for chemistry and materials science, as they directly work on a graph or structural representation of molecules and materials and therefore have full access to all relevant information required to characterize materials. In this Review, we provide an overview of the basic principles of GNNs, widely used datasets, and state-of-the-art architectures, followed by a discussion of a wide range of recent applications of GNNs in chemistry and materials science, and concluding with a road-map for the further development and application of GNNs.

Screening of bimetallic electrocatalysts for water purification with machine learning

Electrocatalysis provides a potential solution to [Formula: see text] pollution in wastewater by converting it to innocuous N2 gas. However, materials with excellent catalytic activity are typically limited to expensive precious metals, hindering their commercial viability. In response to this challenge, we have conducted the most extensive computational search to date for electrocatalysts that can facilitate [Formula: see text] reduction reaction, starting with 59 390 candidate bimetallic alloys from the Materials Project and Automatic-Flow databases. Using a joint machine learning- and computation-based screening strategy, we evaluated our candidates based on corrosion resistance, catalytic activity, N2 selectivity, cost, and the ability to synthesize. We found that only 20 materials will satisfy all criteria in our screening strategy, all of which contain varying amounts of Cu. Our proposed list of candidates is consistent with previous materials investigated in the literature, with the exception of Cu–Co and Cu–Ag based compounds that merit further investigation.

Predicting binding motifs of complex adsorbates using machine learning with a physics-inspired graph representation

Computational screening in heterogeneous catalysis relies increasingly on machine learning models for predicting key input parameters due to the high cost of computing these directly using first-principles methods. This becomes especially relevant when considering complex materials spaces such as alloys, or complex reaction mechanisms with adsorbates that may exhibit bi- or higher-dentate adsorption motifs. Here we present a data-efficient approach to the prediction of binding motifs and associated adsorption enthalpies of complex adsorbates at transition metals and their alloys based on a customized Wasserstein Weisfeiler-Lehman graph kernel and Gaussian process regression. The model shows good predictive performance, not only for the elemental transition metals on which it was trained, but also for an alloy based on these transition metals. Furthermore, incorporation of minimal new training data allows for predicting an out-of-domain transition metal. We believe the model may be useful in active learning approaches, for which we present an ensemble uncertainty estimation approach.

Bimetallic Gold-Silver Nanostructures Drive Low Overpotentials for Electrochemical Carbon Dioxide Reduction

Alloy formation is an advanced approach to improve desired properties that the monoelements cannot achieve. Alloys are usually designed to tailor intrinsic natures or induce synergistic effects by combining materials with distinct properties. Indeed, unprecedented properties have emerged in many cases, superior to a simple sum of pure elements. Here, we present Au-Ag alloy nanostructures with prominent catalytic properties in an electrochemical carbon dioxide reduction reaction (eCO₂RR). The Au-Ag hollow nanocubes are prepared by galvanic replacement of Au on Ag nanocubes. When the Au-to-Ag ratio is 1:1 (Au₁Ag₁), the alloy hollow nanocubes exhibit maximum Faradaic efficiencies of CO production in a wide potential range and high mass activity and CO current density superior to those of the bare metals. In particular, overpotentials are estimated to be similar to or lower than that of the Au catalyst under various standard metrics. Density functional theory calculations, machine learning, and a statistical consideration demonstrate that the optimal configuration of the *COOH intermediate is a bidentate coordination structure where C binds to Au and O binds to Ag. This active Au-Ag neighboring configuration has a maximum population and enhanced intrinsic catalytic activity on the Au₁Ag₁ surface among other Au-to-Ag compositions, in good agreement with the experimental results. Further application of Au₁Ag₁ to a membrane electrode assembly cell at neutral conditions shows enhanced CO Faradaic efficiency and current densities compared to Au or Ag nanocubes, indicating the possible extension of Au-Ag alloys to larger electrochemical systems. These results give a new insight into the synergistic roles of Au and Ag in the eCO₂RR and offer a fresh direction toward a rational design of bimetallic catalysts at a practical scale.

Atomic Structure-Free Representation of Active Motifs for Expedited Catalyst Discovery

To discover new catalysts using density functional theory (DFT) calculations, binding energies of reaction intermediates are considered as descriptors to predict catalytic activities. Recently, machine learning methods have been developed to reduce the number of computationally intensive DFT calculations for a high-throughput screening. These methods require several steps such as bulk structure optimization, surface structure modeling, and active site identification, which could be time-consuming as the number of new candidate materials increases. To bypass these processes, in this work, we report an atomic structure-free representation of active motifs to predict binding energies. We identify binding site atoms and their nearest neighboring atoms positioned in the same layer and the sublayer, and their atomic properties are collected to construct fingerprints. Our method enabled a quicker training (200-400 s using CPU) compared to the previous deep-learning models and predicted CO and H binding energies with mean absolute errors (MAEs) of 0.120 and 0.105 eV, respectively. Our method is also capable of creating all possible active motifs without any DFT calculations and predicting their binding energies using the trained model. The predicted binding energy distributions can suggest promising candidates to accelerate catalyst discovery.

Deep Learning-Based Prediction of Material Properties Using Chemical Compositions and Diffraction Patterns as Experimentally Accessible Inputs

We report a deep learning (DL) model that predicts various material properties while accepting directly accessible inputs from routine experimental platforms: chemical compositions and diffraction data, which can be obtained from the X-ray or electron-beam diffraction and energy-dispersive spectroscopy, respectively. These heterogeneous forms of inputs are treated simultaneously in our DL model, where the novel chemical composition vector is proposed by developing element embedding with the normalized composition matrix. With 1524 binary samples available in the Materials Project database, the model predicts formation energies and band gaps with mean absolute errors of 0.29 eV/atom and 0.66 eV, respectively. According to the weighing test between these two inputs, the properties tend to be more influenced by the chemical composition than the crystal structure. This work intentionally avoids using inputs that are not directly accessible (e.g., atomic coordinates) in experimental platforms, and thus is expected to substantially improve the practical use of DL models.

Low-Cost Pt Alloys for Heterogeneous Catalysis Predicted by Density Functional Theory and Active Learning

Pt is a key high-performing catalyst for important chemical conversions, such as biomass conversion and water splitting. Limited Pt reserves, however, demand that we identify more sustainable alternative catalyst materials for these processes. Here, we combine state-of-the-art graph neural networks and crystal graph machine learning representations with active learning to discover new, low-cost Pt alloy catalysts for biomass reforming and hydrogen evolution reactions. We identify 12 Pt-based alloys which have comparable catalytic activity to that of the exemplar Pt(111) surface. Notably, Cu₃Pt and FeCuPt₂ exhibit near identical catalytic performance as that of Pt(111). These results demonstrate the potential of machine learning for predicting new catalytic materials without recourse to expensive DFT geometry optimizations, the current bottleneck impeding high-throughput materials discovery. We also examine the performance of d-band theory in elucidating trends in binary and ternary Pt alloys.

Adsorption Enthalpies for Catalysis Modeling through Machine-Learned Descriptors

Heterogeneous catalysts are rather complex materials that come in many classes (e.g., metals, oxides, carbides) and shapes. At the same time, the interaction of the catalyst surface with even a relatively simple gas-phase environment such as syngas (CO and H₂) may already produce a wide variety of reaction intermediates ranging from atoms to complex molecules. The starting point for creating predictive maps of, e.g., surface coverages or chemical activities of potential catalyst materials is the reliable prediction of adsorption enthalpies of all of these intermediates. For simple systems, direct density functional theory (DFT) calculations are currently the method of choice. However, a wider exploration of complex materials and reaction networks generally requires enthalpy predictions at lower computational cost.The use of machine learning (ML) and related techniques to make accurate and low-cost predictions of quantum-mechanical calculations has gained increasing attention lately. The employed approaches span from physically motivated models over hybrid physics-ΔML approaches to complete black-box methods such as deep neural networks. In recent works we have explored the possibilities for using a compressed sensing method (Sure Independence Screening and Sparsifying Operator, SISSO) to identify sparse (low-dimensional) descriptors for the prediction of adsorption enthalpies at various active-site motifs of metals and oxides. We start from a set of physically motivated primary features such as atomic acid/base properties, coordination numbers, or band moments and let the data and the compressed sensing method find the best algebraic combination of these features. Here we take this work as a starting point to categorize and compare recent ML-based approaches with a particular focus on model sparsity, data efficiency, and the level of physical insight that one can obtain from the model.Looking ahead, while many works to date have focused only on the mere prediction of databases of, e.g., adsorption enthalpies, there is also an emerging interest in our field to start using ML predictions to answer fundamental science questions about the functioning of heterogeneous catalysts or perhaps even to design better catalysts than we know today. This task is significantly simplified in works that make use of scaling-relation-based models (volcano curves), where the model outcome is determined by only one or two adsorption enthalpies and which consequently become the sole target for ML-based high-throughput screening or design. However, the availability of cheap ML energetics also allows going beyond scaling relations. On the basis of our own work in this direction, we will discuss the additional physical insight that can be achieved by integrating ML-based predictions with traditional catalysis modeling techniques from thermal and electrocatalysis, such as the computational hydrogen electrode and microkinetic modeling, as well as the challenges that lie ahead.

Compact atomic descriptors enable accurate predictions via linear models

We probe the accuracy of linear ridge regression employing a three-body local density representation derived from the atomic cluster expansion. We benchmark the accuracy of this framework in the prediction of formation energies and atomic forces in molecules and solids. We find that such a simple regression framework performs on par with state-of-the-art machine learning methods which are, in most cases, more complex and more computationally demanding. Subsequently, we look for ways to sparsify the descriptor and further improve the computational efficiency of the method. To this aim, we use both principal component analysis and least absolute shrinkage operator regression for energy fitting on six single-element datasets. Both methods highlight the possibility of constructing a descriptor that is four times smaller than the original with a similar or even improved accuracy. Furthermore, we find that the reduced descriptors share a sizable fraction of their features across the six independent datasets, hinting at the possibility of designing material-agnostic, optimally compressed, and accurate descriptors.

Group and Period-Based Representations for Improved Machine Learning Prediction of Heterogeneous Alloy Catalysts

Machine learning has recently emerged as an efficient and powerful alternative to density functional theory for studying heterogeneous catalysis. Machine learning methods rely on a geometrical representation of the chemical environment around the catalytic adsorption site based on physical or chemical descriptors. Here, we show that replacing the atomic number in geometrical representations with elemental groups and periods (GP) yields significant improvements in predicted adsorption energies on bimetallic alloy surfaces. Notably, the GP-based Labeled Site Crystal Graph representation reported here achieves mean absolute error (MAE) ∼0.05 eV (near chemical accuracy) in predicting hydrogen adsorption and MAE ∼0.10 eV for other strong binding adsorbates such as carbon, nitrogen, oxygen, and sulfur. We also show GP-based representations to be robust in predicting adsorption on surface facets, elements, and alloys that are not included in the initial training set. This reliability makes GP-based representations an ideal basis for high-throughput approaches and materials discovery based on active learning techniques, which often involve limited training sets.

Perspective on computational reaction prediction using machine learning methods in heterogeneous catalysis

Heterogeneous catalysis plays a significant role in the modern chemical industry. Towards the rational design of novel catalysts, understanding reactions over surfaces is the most essential aspect. Typical industrial catalytic processes such as syngas conversion and methane utilisation can generate a large reaction network comprising thousands of intermediates and reaction pairs. This complexity not only arises from the permutation of transformations between species but also from the extra reaction channels offered by distinct surface sites. Despite the success in investigating surface reactions at the atomic scale, the huge computational expense of ab initio methods hinders the exploration of such complicated reaction networks. With the proliferation of catalysis studies, machine learning as an emerging tool can take advantage of the accumulated reaction data to emulate the output of ab initio methods towards swift reaction prediction. Here, we briefly summarise the conventional workflow of reaction prediction, including reaction network generation, ab initio thermodynamics and microkinetic modelling. An overview of the frequently used regression models in machine learning is presented. As a promising alternative to full ab initio calculations, machine learning interatomic potentials are highlighted. Furthermore, we survey applications assisted by these methods for accelerating reaction prediction, exploring reaction networks, and computational catalyst design. Finally, we envisage future directions in computationally investigating reactions and implementing machine learning algorithms in heterogeneous catalysis.

Autobifunctional Mechanism of Jagged Pt Nanowires for Hydrogen Evolution Kinetics via End-to-End Simulation

The extraordinary mass activity of jagged Pt nanowires can substantially improve the economics of the hydrogen evolution reaction (HER). However, it is a great challenge to fully unveil the HER kinetics driven by the jagged Pt nanowires with their multiscale morphology. Herein we present an end-to-end framework that combines experiment, machine learning, and multiscale advances of the past decade to elucidate the HER kinetics catalyzed by jagged Pt nanowires under alkaline conditions. The bifunctional catalysis conventionally refers to the synergistic increase in activity by the combination of two different catalysts. We report that monometals, such as jagged Pt nanowires, can exhibit bifunctional characteristics owing to its complex surface morphology, where one site prefers electrochemical proton adsorption and another is responsible for activation, resulting in a 4-fold increase in the activity. We find that the conventional design guideline that the sites with a 0 eV Gibbs free energy of adsorption are optimal for HER does not hold under alkaline conditions, and rather, an energy between -0.2 and 0.0 eV is shown to be optimal. At the reaction temperatures, the high activity arises from low-coordination-number (≤7) Pt atoms exposed by the jagged surface. Our current demonstration raises an emerging prospect to understand highly complex kinetic phenomena on the nanoscale in full by implementing end-to-end multiscale strategies.

Semi-grand canonical Monte Carlo simulation of the acrolein induced surface segregation and aggregation of AgPd with machine learning surrogate models

The single atom alloy of AgPd has been found to be a promising catalyst for the selective hydrogenation of acrolein. It is also known that the formation of Pd islands on the surface will greatly reduce the selectivity of the reaction. As a result, the surface segregation and aggregation of Pd on the AgPd surface under reaction conditions of selective hydrogenation of acrolein are of great interest. In this work, we lay out a workflow that can predict the surface segregation and aggregation of Pd on a FCC(111) AgPd surface with and without the presence of acrolein. We use machine learning surrogate models to predict the AgPd bulk energy, AgPd slab energy, and acrolein adsorption energy on AgPd slabs. Then, we use the semi-grand canonical Monte Carlo simulation to predict the surface segregation and aggregation under different bulk Pd concentrations. Under vacuum conditions, our method predicts that only trace amount of Pd will exist on the surface at Pd bulk concentrations less than 20%. However, with the presence of acrolein, Pd will start to aggregate as dimers on the surface at Pd bulk concentrations as low as 6.5%.

Bayesian learning of chemisorption for bridging the complexity of electronic descriptors

Building upon the d-band reactivity theory in surface chemistry and catalysis, we develop a Bayesian learning approach to probing chemisorption processes at atomically tailored metal sites. With representative species, e.g., *O and *OH, Bayesian models trained with ab initio adsorption properties of transition metals predict site reactivity at a diverse range of intermetallics and near-surface alloys while naturally providing uncertainty quantification from posterior sampling. More importantly, this conceptual framework sheds light on the orbitalwise nature of chemical bonding at adsorption sites with d-states characteristics ranging from bulk-like semi-elliptic bands to free-atom-like discrete energy levels, bridging the complexity of electronic descriptors for the prediction of novel catalytic materials.

Developing a generalizable model to describe adsorption processes at metal surfaces can be extremely challenging due to complex phenomena involved. Here the authors introduce a Bayesian learning approach based on ab initio data and the d-band model to capture the essential physics of adsorbate–substrate interactions.

Differentiable Optimization for the Prediction of Ground State Structures (DOGSS)

Ground state or relaxed inorganic structures are the starting point for most computational materials science or surface science analyses. Many of these structure relaxations represent systematic changes to the structure, but there are currently no general methods to improve the initial structure guess based on past calculations. Here we present a method to directly predict the ground state configuration using differentiable optimization and graph neural networks to learn the properties of a simple harmonic force field that approximates the ground state structure and properties. We demonstrate this flexible open source tool for improving the initial configurations for large datasets of inorganic multicomponent surface relaxations across 32 elements and the relaxation of adsorbates (H and CO) on these surfaces. Using these improved initial configurations reduces the expensive adsorbate-covered surface relaxations by approximately 50% and is complementary to other approaches to accelerate the relaxation process.

Electrocatalysis as the Nexus for Sustainable Renewable Energy: The Gordian Knot of Activity, Stability, and Selectivity

The use of renewable energy by means of electrochemical techniques by converting H2 O, CO2 and N2 into chemical energy sources and raw materials, is the basis for securing a future sustainable "green" energy supply. Some weaknesses and inconsistencies in the practice of determining the electrocatalytic performance, which prevents a rational bottom-up catalyst design, are discussed. Large discrepancies in material properties as well as in electrocatalytic activity and stability become obvious when materials are tested under the conditions of their intended use as opposed to the usual laboratory conditions. They advocate for uniform activity/stability correlations under application-relevant conditions, and the need for a clear representation of electrocatalytic performance by contextualization in terms of functional investigation or progress towards application is emphasized.

Rational Prediction of Single Metal Atom Supported on Two-Dimensional Metal Diborides for Electrocatalytic N2 Reduction Reaction with Integrated Descriptor

Nitrogen reduction reaction (NRR) plays an important role in chemical industry, so it is significant to develop low-cost and efficient electrocatalysts for nitrogen fixation instead of the traditional Haber-Bosch process. In this paper, the electrocatalytic performance of various single atoms doped on two-dimensional metal diborides with a B vacancy for N₂ reduction to ammonia is calculated and predicted. By screening numerous catalysts, we find that Ti@VB₂ is the most active catalyst for NRR, and the limiting potential of Ti@VB₂ for NRR is -0.61 V. Through high-throughput search and LASSO regression, an integrated descriptor combining the intrinsic properties of the single transition metal atom (TM) and the substrate (MB₂) is proposed, which can fit the relationship between intrinsic properties of catalysts and NRR activity well. Therefore, this study not only discovers a promising electrocatalyst for nitrogen fixation but also provides a strategy for predicting the activity of catalysts.