Efficient Global Structure Optimization with a Machine-Learned Surrogate Model

Active-learning for global optimization of Ni-Ceria nanoparticles: The case of Ce4-xNixO8- x (x = 1, 2, 3)

Ni-CeO2 nanoparticles (NPs) are promising nanocatalysts for water splitting and water gas shift reactions due to the ability of ceria to temporarily donate oxygen to the catalytic reaction and accept oxygen after the reaction is completed. Therefore, elucidating how different properties of the Ni-Ceria NPs relate to the activity and selectivity of the catalytic reaction, is of crucial importance for the development of novel catalysts. In this work the active learning (AL) method based on machine learning regression and its uncertainty is used for the global optimization of Ce(4-x)NixO(8-x) (x = 1, 2, 3) nanoparticles, employing density functional theory calculations. Additionally, further investigation of the NPs by mass-scaled parallel-tempering Born-Oppenheimer molecular dynamics resulted in the same putative global minimum structures found by AL, demonstrating the robustness of our AL search to learn from small datasets and assist in the global optimization of complex electronic structure systems.

Going for Gold(-Standard): Attaining Coupled Cluster Accuracy in Oxide-Supported Nanoclusters

The structure of oxide-supported metal nanoclusters plays an essential role in their sharply enhanced catalytic activity over that of bulk metals. Simulations provide the atomic-scale resolution needed to understand these systems. However, the sensitive mix of metal-metal and metal-support interactions, which govern their structure, puts stringent requirements on the method used, requiring calculations beyond standard density functional theory (DFT). The method of choice is coupled cluster theory [specifically CCSD(T)], but its computational cost has so far prevented its application to these systems. In this work, we showcase two approaches to make CCSD(T) accuracy readily achievable in oxide-supported nanoclusters. First, we leverage the SKZCAM protocol to provide the first benchmarks of oxide-supported nanoclusters, revealing that it is specifically metal-metal interactions that are challenging to capture with DFT. Second, we propose a CCSD(T) correction (ΔCC) to the metal-metal interaction errors in DFT, reaching accuracy comparable to that of the SKZCAM protocol at significantly lower cost. This approach forges a path toward studying larger systems at reliable accuracy, which we highlight by identifying a ground-state structure in agreement with experiments for Au20 on MgO, a challenging system where DFT models have yielded conflicting predictions.

Gas-Phase Production of Hydroxylated Silicon Oxide Cluster Cations: Structure, Infrared Spectroscopy, and Astronomical Relevance

The interaction of free cationic silicon oxide clusters, Si x O y + (x = 2-5, y ≥ x), with dilute water vapor, was investigated in a flow tube reactor. Product mass distributions indicate cluster size-dependent dissociative water adsorption. To probe the structure and vibrational spectra of the resulting Si x O y H2 + (x = 2-4) clusters, we employed infrared multiple photon dissociation spectroscopy and density functional theory calculations. The planar rhombic cluster core of the disilicon oxides (x = 2) appears to be retained upon dissociative adsorption of one H2O unit, whereas a significant structural transformation of the tri- and tetra-silicon oxides (x = 3 and 4) is induced, resulting in an increased coordination of the Si atoms and more 3D cluster structures. In an astronomical context, we discuss the potential relevance of Si x O y H z + clusters as seeds for dust nucleation and catalysts for carbon-based chemistry in diffuse or translucent interstellar clouds, where all the necessary conditions for producing these species are found. In the produced clusters, the frequency of the isolated silanol Si-OH stretching vibrational mode is considerably blue-shifted compared to that in hydroxylated bulk silica and small inorganic compounds. This mode has a characteristic frequency range between 1200 cm-1 (8.3 μm) and 1090 cm-1 (9.2 μm) and is associated with the anomalously small Si-OH bond lengths in these ionised species. In infrared observations such high frequency Si-O stretching modes are usually associated with a pure bulk silica component of silicate cosmic dust. The presence of Si x O y H2 + clusters in low silica astrophysical environments could thus potentially be detected via their signature Si-O band using the James Webb space telescope.

Accelerated Bayesian Inference for Molecular Simulations using Local Gaussian Process Surrogate Models

While Bayesian inference is the gold standard for uncertainty quantification and propagation, its use within physical chemistry encounters formidable computational barriers. These bottlenecks are magnified for modeling data with many independent variables, such as X-ray/neutron scattering patterns and electromagnetic spectra. To address this challenge, we employ local Gaussian process (LGP) surrogate models to accelerate Bayesian optimization over these complex thermophysical properties. The time-complexity of the LGPs scales linearly in the number of independent variables, in stark contrast to the computationally expensive cubic scaling of conventional Gaussian processes. To illustrate the method, we trained a LGP surrogate model on the radial distribution function of liquid neon and observed a 1,760,000-fold speed-up compared to molecular dynamics simulation, beating a conventional GP by three orders-of-magnitude. We conclude that LGPs are robust and efficient surrogate models poised to expand the application of Bayesian inference in molecular simulations to a broad spectrum of experimental data.

Surface phase diagrams from nested sampling

Studies in atomic-scale modeling of surface phase equilibria often focus on temperatures near zero Kelvin due to the challenges in calculating the free energy of surfaces at finite temperatures. The Bayesian-inference-based nested sampling (NS) algorithm allows for modeling phase equilibria at arbitrary temperatures by directly and efficiently calculating the partition function, whose relationship with free energy is well known. This work extends NS to calculate adsorbate phase diagrams, incorporating all relevant configurational contributions to the free energy. We apply NS to the adsorption of Lennard-Jones (LJ) gas particles on low-index and vicinal LJ solid surfaces and construct the canonical partition function from these recorded energies to calculate ensemble averages of thermodynamic properties, such as the constant-volume heat capacity and order parameters that characterize the structure of adsorbate phases. Key results include determining the nature of phase transitions of adsorbed LJ particles on flat and stepped LJ surfaces, which typically feature an enthalpy-driven condensation at higher temperatures and an entropy-driven reordering process at lower temperatures, and the effect of surface geometry on the presence of triple points in the phase diagrams. Overall, we demonstrate the ability and potential of NS for surface modeling.

Accelerating the prediction of inorganic surfaces with machine learning interatomic potentials

The surface properties of solid-state materials often dictate their functionality, especially for applications where nanoscale effects become important. The relevant surface(s) and their properties are determined, in large part, by the material's synthesis or operating conditions. These conditions dictate thermodynamic driving forces and kinetic rates responsible for yielding the observed surface structure and morphology. Computational surface science methods have long been applied to connect thermochemical conditions to surface phase stability, particularly in the heterogeneous catalysis and thin film growth communities. This review provides a brief introduction to first-principles approaches to compute surface phase diagrams before introducing emerging data-driven approaches. The remainder of the review focuses on the application of machine learning, predominantly in the form of learned interatomic potentials, to study complex surfaces. As machine learning algorithms and large datasets on which to train them become more commonplace in materials science, computational methods are poised to become even more predictive and powerful for modeling the complexities of inorganic surfaces at the nanoscale.

Accelerated Structural Optimization for the Supported Metal System Based on Hybrid Approach Combining Bayesian Optimization with Local Search

Numerous systematic methods have been developed to search for the global minimum of the potential energy surface, which corresponds to the optimal atomic structure. However, the majority of them still demand a substantial computing load due to the relaxation process that is embedded as an inner step inside the algorithm. Here, we propose a hybrid approach that combines Bayesian optimization (BO) and a local search that circumvents the relaxation step and efficiently finds the optimum structure, particularly in supported metal systems. The hybridization strategy combining the capabilities of BO's effective exploration and the local search's fast convergence expedites structural search. In addition, the formulation of physical constraints regarding the materials system and the feature of screening structure similarity enhance the computational efficiency of the proposed method. The proposed algorithm is demonstrated in two supported metal systems, showing the potential of the proposed method in the field of structural optimization.

Accelerated Lignocellulosic Molecule Adsorption Structure Determination

Here, we present a study combining Bayesian optimization structural inference with the machine learning interatomic potential Neural Equivariant Interatomic Potential (NequIP) to accelerate and enable the study of the adsorption of the conformationally flexible lignocellulosic molecules β-d-xylose and 1,4-β-d-xylotetraose on a copper surface. The number of structure evaluations needed to map out the relevant potential energy surfaces are reduced by Bayesian optimization, while NequIP minimizes the time spent on each evaluation, ultimately resulting in cost-efficient and reliable sampling of large systems and configurational spaces. Although the applicability of Bayesian optimization for the conformational analysis of the more flexible xylotetraose molecule is restricted by the sample complexity bottleneck, the latter can be effectively bypassed with external conformer search tools, such as the Conformer-Rotamer Ensemble Sampling Tool, facilitating the subsequent lower-dimensional global minimum adsorption structure determination. Finally, we demonstrate the applicability of the described approach to find adsorption structures practically equivalent to the density functional theory counterparts at a fraction of the computational cost.

GPAW: An open Python package for electronic structure calculations

We review the GPAW open-source Python package for electronic structure calculations. GPAW is based on the projector-augmented wave method and can solve the self-consistent density functional theory (DFT) equations using three different wave-function representations, namely real-space grids, plane waves, and numerical atomic orbitals. The three representations are complementary and mutually independent and can be connected by transformations via the real-space grid. This multi-basis feature renders GPAW highly versatile and unique among similar codes. By virtue of its modular structure, the GPAW code constitutes an ideal platform for the implementation of new features and methodologies. Moreover, it is well integrated with the Atomic Simulation Environment (ASE), providing a flexible and dynamic user interface. In addition to ground-state DFT calculations, GPAW supports many-body GW band structures, optical excitations from the Bethe-Salpeter Equation, variational calculations of excited states in molecules and solids via direct optimization, and real-time propagation of the Kohn-Sham equations within time-dependent DFT. A range of more advanced methods to describe magnetic excitations and non-collinear magnetism in solids are also now available. In addition, GPAW can calculate non-linear optical tensors of solids, charged crystal point defects, and much more. Recently, support for graphics processing unit (GPU) acceleration has been achieved with minor modifications to the GPAW code thanks to the CuPy library. We end the review with an outlook, describing some future plans for GPAW.

Machine learning predicts atomistic structures of multielement solid surfaces for heterogeneous catalysts in variable environments

Solid surfaces usually reach thermodynamic equilibrium through particle exchange with their environment under reactive conditions. A prerequisite for understanding their functionalities is detailed knowledge of the surface composition and atomistic geometry under working conditions. Owing to the large number of possible Miller indices and terminations involved in multielement solids, extensive sampling of the compositional and conformational space needed for reliable surface energy estimation is beyond the scope of ab initio calculations. Here, we demonstrate, using the case of iron carbides in environments with varied carbon chemical potentials, that the stable surface composition and geometry of multielement solids under reactive conditions, which involve large compositional and conformational spaces, can be predicted at ab initio accuracy using an approach that combines the bond valence model, Gaussian process regression, and ab initio thermodynamics. Determining the atomistic structure of surfaces under working conditions paves the way toward identifying the true active sites of multielement catalysts in heterogeneous catalysis.

Predicting Phase Stability at Interfaces

We present the RAFFLE methodology for structural prediction of the interface between two materials and demonstrate its effectiveness by applying it to MgO encapsulated by two layers of graphene. To address the challenge of interface structure prediction, our methodology combines physical insights derived from morphological features observed in related systems with an iterative machine learning technique. This employs physical-based methods, including void-filling and n-body distribution functions to predict interface structures. For the carbon-MgO encapsulated system, we have shown the rocksalt and hexagonal phases of MgO to be the two most energetically stable in the few-layer regime. We demonstrate that monolayer rocksalt is heavily stabilized by interfacing with graphene, becoming more energetically favorable than the graphenelike monolayer hexagonal MgO. The RAFFLE methodology provides valuable insights into interface behavior, and a route to finding new materials at interfaces.

Inferring energy-composition relationships with Bayesian optimization enhances exploration of inorganic materials

Computational exploration of the compositional spaces of materials can provide guidance for synthetic research and thus accelerate the discovery of novel materials. Most approaches employ high-throughput sampling and focus on reducing the time for energy evaluation for individual compositions, often at the cost of accuracy. Here, we present an alternative approach focusing on effective sampling of the compositional space. The learning algorithm PhaseBO optimizes the stoichiometry of the potential target material while improving the probability of and accelerating its discovery without compromising the accuracy of energy evaluation.

Machine-learning-accelerated simulations to enable automatic surface reconstruction

Understanding material surfaces and interfaces is vital in applications such as catalysis or electronics. By combining energies from electronic structure with statistical mechanics, ab initio simulations can, in principle, predict the structure of material surfaces as a function of thermodynamic variables. However, accurate energy simulations are prohibitive when coupled to the vast phase space that must be statistically sampled. Here we present a bi-faceted computational loop to predict surface phase diagrams of multicomponent materials that accelerates both the energy scoring and statistical sampling methods. Fast, scalable and data-efficient machine learning interatomic potentials are trained on high-throughput density-functional-theory calculations through closed-loop active learning. Markov chain Monte Carlo sampling in the semigrand canonical ensemble is enabled by using virtual surface sites. The predicted surfaces for GaN(0001), Si(111) and SrTiO3(001) are in agreement with past work and indicate that the proposed strategy can model complex material surfaces and discover previously unreported surface terminations.

Accessing complex reconstructed material structures with hybrid global optimization accelerated via on-the-fly machine learning

The complex reconstructed structure of materials can be revealed by global optimization. This paper describes a hybrid evolutionary algorithm (HEA) that combines differential evolution and genetic algorithms with a multi-tribe framework. An on-the-fly machine learning calculator is adopted to expedite the identification of low-lying structures. With a superior performance to other well-established methods, we further demonstrate its efficacy by optimizing the complex oxidized surface of Pt/Pd/Cu with different facets under (4 × 4) periodicity. The obtained structures are consistent with experimental results and are energetically lower than the previously presented model.

MAGUS: machine learning and graph theory assisted universal structure searcher

Crystal structure predictions based on first-principles calculations have gained great success in materials science and solid state physics. However, the remaining challenges still limit their applications in systems with a large number of atoms, especially the complexity of conformational space and the cost of local optimizations for big systems. Here, we introduce a crystal structure prediction method, MAGUS, based on the evolutionary algorithm, which addresses the above challenges with machine learning and graph theory. Techniques used in the program are summarized in detail and benchmark tests are provided. With intensive tests, we demonstrate that on-the-fly machine-learning potentials can be used to significantly reduce the number of expensive first-principles calculations, and the crystal decomposition based on graph theory can efficiently decrease the required configurations in order to find the target structures. We also summarized the representative applications of this method on several research topics, including unexpected compounds in the interior of planets and their exotic states at high pressure and high temperature (superionic, plastic, partially diffusive state, etc.); new functional materials (superhard, high-energy-density, superconducting, photoelectric materials), etc. These successful applications demonstrated that MAGUS code can help to accelerate the discovery of interesting materials and phenomena, as well as the significant value of crystal structure predictions in general.

A machine learning potential for simulating infrared spectra of nanosilicate clusters

The use of machine learning (ML) in chemical physics has enabled the construction of interatomic potentials having the accuracy of ab initio methods and a computational cost comparable to that of classical force fields. Training an ML model requires an efficient method for the generation of training data. Here, we apply an accurate and efficient protocol to collect training data for constructing a neural network-based ML interatomic potential for nanosilicate clusters. Initial training data are taken from normal modes and farthest point sampling. Later on, the set of training data is extended via an active learning strategy in which new data are identified by the disagreement between an ensemble of ML models. The whole process is further accelerated by parallel sampling over structures. We use the ML model to run molecular dynamics simulations of nanosilicate clusters with various sizes, from which infrared spectra with anharmonicity included can be extracted. Such spectroscopic data are needed for understanding the properties of silicate dust grains in the interstellar medium and in circumstellar environments.

Machine Learning-Enabled Development of Accurate Force Fields for Refrigerants

Hydrofluorocarbon (HFC) refrigerants with zero ozone-depleting potential have replaced chlorofluorocarbons and are now ubiquitous. However, some HFCs have high global warming potential, which has led to calls by governments to phase out these HFCs. Technologies to recycle and repurpose these HFCs need to be developed. Therefore, thermophysical properties of HFCs are needed over a wide range of conditions. Molecular simulations can help understand and predict the thermophysical properties of HFCs. The prediction capability of a molecular simulation is directly tied to the accuracy of the force field. In this work, we applied and refined a machine learning-based workflow to optimize the Lennard-Jones parameters of classical HFC force fields for HFC-143a (CF₃CH₃), HFC-134a (CH₂FCF₃), R-50 (CH₄), R-170 (C₂H₆), and R-14 (CF₄). Our workflow involves liquid density iterations with molecular dynamics simulations and vapor-liquid equilibrium (VLE) iterations with Gibbs ensemble Monte Carlo simulations. Support vector machine classifiers and Gaussian process surrogate models save months of simulation time and can efficiently select optimal parameters from half a million distinct parameter sets. Excellent agreement as evidenced by low mean absolute percent errors (MAPEs) of simulated liquid density (ranging from 0.3% to 3.4%), vapor density (ranging from 1.4% to 2.6%), vapor pressure (ranging from 1.3% to 2.8%), and enthalpy of vaporization (ranging from 0.5% to 2.7%) relative to experiments was obtained for the recommended parameter set of each refrigerant. The performance of each new parameter set was superior or similar to the best force field in the literature.

A Machine-Learning-Based Approach for Solving Atomic Structures of Nanomaterials Combining Pair Distribution Functions with Density Functional Theory

Determination of crystal structures of nanocrystalline or amorphous compounds is a great challenge in solid-state chemistry and physics. Pair distribution function (PDF) analysis of X-ray or neutron total scattering data has proven to be a key element in tackling this challenge. However, in most cases, a reliable structural motif is needed as a starting configuration for structure refinements. Here, an algorithm that is able to determine the crystal structure of an unknown compound by means of an on-the-fly trained machine learning model, which combines density functional theory calculations with comparison of calculated and measured PDFs for global optimization in an artificial landscape, is presented. Due to the nature of this landscape, even metastable configurations and stacking disorders can be identified.

Data-Driven Machine Learning for Understanding Surface Structures of Heterogeneous Catalysts

The design of heterogeneous catalysts is necessarily surface-focused, generally achieved via optimization of adsorption energy and microkinetic modelling. A prerequisite is to ensure the adsorption energy is physically meaningful is the stable existence of the conceived active-site structure on the surface. The development of improved understanding of the catalyst surface, however, is challenging practically because of the complex nature of dynamic surface formation and evolution under in-situ reactions. We propose therefore data-driven machine-learning (ML) approaches as a solution. In this Minireview we summarize recent progress in using machine-learning to search and predict (meta)stable structures, assist operando simulation under reaction conditions and micro-environments, and critically analyze experimental characterization data. We conclude that ML will become the new norm to lower costs associated with discovery and design of optimal heterogeneous catalysts.

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

Adsorption of carbon monoxide (CO) on transition-metal surfaces is a prototypical process in surface sciences and catalysis. Despite its simplicity, it has posed great challenges to theoretical modeling. Pretty much all existing density functionals fail to accurately describe surface energies and CO adsorption site preference as well as adsorption energies simultaneously. Although the random phase approximation (RPA) cures these density functional theory failures, its large computational cost makes it prohibitive to study the CO adsorption for any but the simplest ordered cases. Here, we address these challenges by developing a machine-learned force field (MLFF) with near RPA accuracy for the prediction of coverage-dependent adsorption of CO on the Rh(111) surface through an efficient on-the-fly active learning procedure and a Δ-machine learning approach. We show that the RPA-derived MLFF is capable to accurately predict the Rh(111) surface energy and CO adsorption site preference as well as adsorption energies at different coverages that are all in good agreement with experiments. Moreover, the coverage-dependent ground-state adsorption patterns and adsorption saturation coverage are identified.

Dual-Level Training of Gaussian Processes with Physically Inspired Priors for Geometry Optimizations

Gaussian process (GP) regression has been recently developed as an effective method in molecular geometry optimization. The prior mean function is one of the crucial parts of the GP. We design and validate two types of physically inspired prior mean functions: force-field-based priors and posterior-type priors. In this work, we implement a dual-level training (DLT) optimizer for the posterior-type priors. The DLT optimizers can be considered as a class of optimization algorithms that belong to the delta-machine learning paradigm but with several major differences compared to the previously proposed algorithms in the same paradigm. In the first level of the DLT, we incorporate the classical mechanical descriptions of the equilibrium geometries into the prior function, which enhances the performance of the GP optimizer as compared to the one using a constant (or zero) prior. In the second level, we utilize the surrogate potential energy surfaces (PESs), which incorporate the physics learned in the first-level training, as the prior function to refine the model performance further. We find that the force-field-based priors and posterior-type priors reduce the overall optimization steps by a factor of 2-3 when compared to the limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) optimizer as well as the constant-prior GP optimizer proposed in previous works. We also demonstrate the potential of recovering the real PESs with GP with a force-field prior. This work shows the importance of including domain knowledge as an ingredient in the GP, which offers a potentially robust learning model for molecular geometry optimization and for exploring molecular PESs.

Stability of oxidized states of free-standing and ceria-supported PtOx particles

Nanostructured materials based on CeO2 and Pt play a fundamental role in catalyst design. However, their characterization is often challenging due to their structural complexity and the tendency of the materials to change under reaction conditions. In this work, we combine calculations based on the density functional theory, a machine-learning assisted global optimization method (GOFEE), and ab initio thermodynamics to characterize stable oxidation states of ceria-supported PtyOx particles in different environments. The collection of global minima structures for different stoichiometries resulting from the global optimisation effort is used to assess the effect of temperature, oxygen pressure, and support interactions on the phase diagrams, oxidation states, and geometries of the PtyOx particles. We thus identify favoured structural motifs and O:Pt ratios, revealing that oxidized states of free-standing and ceria-supported platinum particles are more stable than reduced ones under a wide range of conditions. These results indicate that studies rationalizing activity of ceria-supported Pt clusters must consider oxidized states, and that previous understanding of such materials obtained only with fully reduced Pt clusters may be incomplete.

Fast atomic structure optimization with on-the-fly sparse Gaussian process potentials. JOURNAL OF PHYSICS. CONDENSED MATTER : AN INSTITUTE OF PHYSICS JOURNAL 2022;34:344007. [PMID: 35675808 DOI: 10.1088/1361-648x/ac76ff]

We apply on-the-fly machine learning potentials (MLPs) using the sparse Gaussian process regression (SGPR) algorithm for fast optimization of atomic structures. Great acceleration is achieved even in the context of a single local optimization. Although for finding the exact local minimum, due to limited accuracy of MLPs, switching to another algorithm may be needed. For random gold clusters, the forces are reduced to ∼0.1 eV Å^-1within less than ten first-principles (FP) calculations. Because of highly transferable MLPs, this algorithm is specially suitable for global optimization methods such as random or evolutionary structure searching or basin hopping. This is demonstrated by sequential optimization of random gold clusters for which, after only a few optimizations, FP calculations were rarely needed.

Atomistic Global Optimization X: A Python package for optimization of atomistic structures

Modelling and understanding properties of materials from first principles require knowledge of the underlyingatomistic structure. This entails knowing the individual chemical identity and position of all atoms involved.Obtaining such information for macro-molecules, nano-particles, clusters, and for the surface, interface, andbulk phases of amorphous and solid materials represents a difficult high-dimensional global optimizationproblem. The rise of machine learning techniques in materials science has, however, led to many compellingdevelopments that may speed up structure searches. The complexity of such new methods has prompted aneed for an efficient way of assembling them into global optimization algorithms that can be experimentedwith. In this paper, we introduce the Atomistic Global Optimization X (AGOX) framework and code, asa customizable approach that enables efficient building and testing of global optimization algorithms. Amodular way of expressing global optimization algorithms is described and modern programming practicesare used to enable that modularity in the freely available AGOX python package. A number of examplesof global optimization approaches are implemented and analyzed. This ranges from random search andbasin-hopping to machine learning aided approaches with on-the-fly learnt surrogate energy landscapes. Themethods are show-cased on problems ranging from supported clusters over surface reconstructions to largecarbon clusters and metal-nitride clusters incorporated into graphene sheets.

Structure of an Ultrathin Oxide on Pt3 Sn(111) Solved by Machine Learning Enhanced Global Optimization

Determination of the atomic structure of solid surfaces typically depends on comparison of measured properties with simulations based on hypothesized structural models. For simple structures, the models may be guessed, but for more complex structures there is a need for reliable theory-based search algorithms. So far, such methods have been limited by the combinatorial complexity and computational expense of sufficiently accurate energy estimation for surfaces. However, the introduction of machine learning methods has the potential to change this radically. Here, we demonstrate how an evolutionary algorithm, utilizing machine learning for accelerated energy estimation and diverse population generation, can be used to solve an unknown surface structure-the (4×4) surface oxide on Pt₃ Sn(111)-based on limited experimental input. The algorithm is efficient and robust, and should be broadly applicable in surface studies, where it can replace manual, intuition based model generation.

Structure of an Ultrathin Oxide on Pt3Sn(111) Solved by Machine Learning Enhanced Global Optimization

Determination of the atomic structure of solid surfaces typically depends on comparison of measured properties with simulations based on hypothesized structural models. For simple structures, the models may be guessed, but for more complex structures there is a need for reliable theory-based search algorithms. So far, such methods have been limited by the combinatorial complexity and computational expense of sufficiently accurate energy estimation for surfaces. However, the introduction of machine learning methods has the potential to change this radically. Here, we demonstrate how an evolutionary algorithm, utilizing machine learning for accelerated energy estimation and diverse population generation, can be used to solve an unknown surface structure-the (4×4) surface oxide on Pt3Sn(111)-based on limited experimental input. The algorithm is efficient and robust, and should be broadly applicable in surface studies, where it can replace manual, intuition based model generation.

Systematic Comparison of Genetic Algorithm and Basin Hopping Approaches to the Global Optimization of Si(111) Surface Reconstructions

We present a systematic study of two widely used material structure prediction methods, the Genetic Algorithm and Basin Hopping approaches to global optimization, in a search for the 3 × 3, 5 × 5, and 7 × 7 reconstructions of the Si(111) surface. The Si(111) 7 × 7 reconstruction is the largest and most complex surface reconstruction known, and finding it is a very exacting test for global optimization methods. In this paper, we introduce a modification to previous Genetic Algorithm work on structure search for periodic systems, to allow the efficient search for surface reconstructions, and present a rigorous study of the effect of the different parameters of the algorithm. We also perform a detailed comparison with the recently improved Basin Hopping algorithm using Delocalized Internal Coordinates. Both algorithms succeeded in either resolving the 3 × 3, 5 × 5, and 7 × 7 DAS surface reconstructions or getting “sufficiently close”, i.e., identifying structures that only differ for the positions of a few atoms as well as thermally accessible structures within k_BT/unit area of the global minimum, with T = 300 K. Overall, the Genetic Algorithm is more robust with respect to parameter choice and in success rate, while the Basin Hopping method occasionally exhibits some advantages in speed of convergence. In line with previous studies, the results confirm that robustness, success, and speed of convergence of either approach are strongly influenced by how much the trial moves tend to preserve favorable bonding patterns once these appear.

Machine learning-based framework to cover optimal Pareto-front in many-objective optimization

One of the crucial challenges of solving many-objective optimization problems is uniformly well covering of the Pareto-front (PF). However, many the state-of-the-art optimization algorithms are capable of approximating the shape of many-objective PF by generating a limited number of non-dominated solutions. The exponential increase of the population size is an inefficient strategy that increases the computational complexity of the algorithm dramatically—especially when solving many-objective problems. In this paper, we introduce a machine learning-based framework to cover sparse PF surface which is initially generated by many-objective optimization algorithms; either by classical or meta-heuristic methods. The proposed method, called many-objective reverse mapping (MORM), is based on constructing a learning model on the initial PF set as the training data to reversely map the objective values to corresponding decision variables. Using the trained model, a set of candidate solutions can be generated by a variety of inexpensive generative techniques such as Opposition-based Learning and Latin Hypercube Sampling in both objective and decision spaces. Iteratively generated non-dominated candidate solutions cover the initial PF efficiently with no further need to utilize any optimization algorithm. We validate the proposed framework using a set of well-known many-objective optimization benchmarks and two well-known real-world problems. The coverage of PF is illustrated and numerically compared with the state-of-the-art many-objective algorithms. The statistical tests conducted on comparison measures such as HV, IGD, and the contribution ratio on the built PF reveal that the proposed collaborative framework surpasses the competitors on most of the problems. In addition, MORM covers the PF effectively compared to other methods even with the aid of large population size.

Automated exploitation of the big configuration space of large adsorbates on transition metals reveals chemistry feasibility

Mechanistic understanding of large molecule conversion and the discovery of suitable heterogeneous catalysts have been lagging due to the combinatorial inventory of intermediates and the inability of humans to enumerate all structures. Here, we introduce an automated framework to predict stable configurations on transition metal surfaces and demonstrate its validity for adsorbates with up to 6 carbon and oxygen atoms on 11 metals, enabling the exploration of ~10⁸ potential configurations. It combines a graph enumeration platform, force field, multi-fidelity DFT calculations, and first-principles trained machine learning. Clusters in the data reveal groups of catalysts stabilizing different structures and expose selective catalysts for showcase transformations, such as the ethylene epoxidation on Ag and Cu and the lack of C-C scission chemistry on Au. Deviations from the commonly assumed atom valency rule of small adsorbates are also manifested. This library can be leveraged to identify catalysts for converting large molecules computationally.

The discovery of heterogeneous catalysts for large molecule conversion has been lagging due to the combinatorial inventory of intermediates. Here, the author presents an automated framework to explore the chemical space of reaction intermediates.

Geometry meta-optimization

Recent work has demonstrated the promise of using machine-learned surrogates, in particular, Gaussian process (GP) surrogates, in reducing the number of electronic structure calculations (ESCs) needed to perform surrogate model based (SMB) geometry optimization. In this paper, we study geometry meta-optimization with GP surrogates where a SMB optimizer additionally learns from its past "experience" performing geometry optimization. To validate this idea, we start with the simplest setting where a geometry meta-optimizer learns from previous optimizations of the same molecule with different initial-guess geometries. We give empirical evidence that geometry meta-optimization with GP surrogates is effective and requires less tuning compared to SMB optimization with GP surrogates on the ANI-1 dataset of off-equilibrium initial structures of small organic molecules. Unlike SMB optimization where a surrogate should be immediately useful for optimizing a given geometry, a surrogate in geometry meta-optimization has more flexibility because it can distribute its ESC savings across a set of geometries. Indeed, we find that GP surrogates that preserve rotational invariance provide increased marginal ESC savings across geometries. As a more stringent test, we also apply geometry meta-optimization to conformational search on a hand-constructed dataset of hydrocarbons and alcohols. We observe that while SMB optimization and geometry meta-optimization do save on ESCs, they also tend to miss higher energy conformers compared to standard geometry optimization. We believe that further research into characterizing the divergence between GP surrogates and potential energy surfaces is critical not only for advancing geometry meta-optimization but also for exploring the potential of machine-learned surrogates in geometry optimization in general.

Dimerization of dehydrogenated polycyclic aromatic hydrocarbons on graphene

Dimerization of polycyclic aromatic hydrocarbons (PAHs) is an important, yet poorly understood, step in the on-surface synthesis of graphene (nanoribbon), soot formation, and growth of carbonaceous dust grains in the interstellar medium (ISM). The on-surface synthesis of graphene and the growth of carbonaceous dust grains in the ISM require the chemical dimerization in which chemical bonds are formed between PAH monomers. An accurate and cheap method of exploring structure rearrangements is needed to reveal the mechanism of chemical dimerization on surfaces. This work has investigated the chemical dimerization of two dehydrogenated PAHs (coronene and pentacene) on graphene via an evolutionary algorithm augmented by machine learning surrogate potentials and a set of customized structure operators. Different dimer structures on surfaces have been successfully located by our structure search methods. Their binding energies are within the experimental errors of temperature programmed desorption measurements. The mechanism of coronene dimer formation on graphene is further studied and discussed.

Automatic structural elucidation of vacancies in materials by active learning

The artificial intelligence method based on active learning for the automatic structural elucidation of vacancies in materials. This is implemented in the quantum machine learning software/agent for material design and discovery (QMLMaterial).

Data-efficient iterative training of Gaussian approximation potentials: Application to surface structure determination of rutile IrO2 and RuO2

Machine-learning interatomic potentials, such as Gaussian Approximation Potentials (GAPs), constitute a powerful class of surrogate models to computationally involved first-principles calculations. At a similar predictive quality but significantly reduced cost, they could leverage otherwise barely tractable extensive sampling as in global surface structure determination (SSD). This efficiency is jeopardized though, if an a priori unknown structural and chemical search space as in SSD requires an excessive number of first-principles data for the GAP training. To this end, we present a general and data-efficient iterative training protocol that blends the creation of new training data with the actual surface exploration process. Demonstrating this protocol with the SSD of low-index facets of rutile IrO₂ and RuO₂, the involved simulated annealing on the basis of the refining GAP identifies a number of unknown terminations even in the restricted sub-space of (1 × 1) surface unit cells. Particularly in an O-poor environment, some of these, then metal-rich terminations, are thermodynamically most stable and are reminiscent of complexions as discussed for complex ceramic materials.

Dynamics of Heterogeneous Catalytic Processes at Operando Conditions

The rational design of high-performance catalysts is hindered by the lack of knowledge of the structures of active sites and the reaction pathways under reaction conditions, which can be ideally addressed by an in situ/operando characterization. Besides the experimental insights, a theoretical investigation that simulates reaction conditions-so-called operando modeling-is necessary for a plausible understanding of a working catalyst system at the atomic scale. However, there is still a huge gap between the current widely used computational model and the concept of operando modeling, which should be achieved through multiscale computational modeling. This Perspective describes various modeling approaches and machine learning techniques that step toward operando modeling, followed by selected experimental examples that present an operando understanding in the thermo- and electrocatalytic processes. At last, the remaining challenges in this area are outlined.

Entropy-based active learning of graph neural network surrogate models for materials properties

Graph neural networks trained on experimental or calculated data are becoming an increasingly important tool in computational materials science. Networks once trained are able to make highly accurate predictions at a fraction of the cost of experiments or first-principles calculations of comparable accuracy. However, these networks typically rely on large databases of labeled experiments to train the model. In scenarios where data are scarce or expensive to obtain, this can be prohibitive. By building a neural network that provides confidence on the predicted properties, we are able to develop an active learning scheme that can reduce the amount of labeled data required by identifying the areas of chemical space where the model is most uncertain. We present a scheme for coupling a graph neural network with a Gaussian process to featurize solid-state materials and predict properties including a measure of confidence in the prediction. We then demonstrate that this scheme can be used in an active learning context to speed up the training of the model by selecting the optimal next experiment for obtaining a data label. Our active learning scheme can double the rate at which the performance of the model on a test dataset improves with additional data compared to choosing the next sample at random. This type of uncertainty quantification and active learning has the potential to open up new areas of materials science, where data are scarce and expensive to obtain, to the transformative power of graph neural networks.

Active Site Fluxional Restructuring as a New Paradigm in Triggering Reaction Activity for Nanocluster Catalysis

The rationale of the catalytic activity observed in experiments is a crucial task in fundamental catalysis studies. Efficient catalyst design relies on an accurate understanding of the origin of the activity at the atomic level. Theoretical studies have been widely developed to reach such a fundamental atomic scale understanding of catalytic activity. Current theories ascribe the catalytic activity to the geometric and electronic structure of the active site, in which the geometrical and electronic structure effects are derived from the equilibrium geometry of active sites characterizing the static property of the catalyst; however catalysts, especially in the form of nanoclusters, may present fluxional and dynamic structure under reaction conditions, and the effect of this fluxional behavior is not yet widely recognized. Therefore, this Account will focus on the fluxionality of the active sites, which is driven by thermal fluctuations under finite temperature.Under reaction conditions, nanocluster catalysts can readily restructure, either being promoted to another metastable isomer (named as plastic fluxionality) or presenting ample deformations around their equilibrium geometry (named as elastic fluxionality). This Account summarizes our recent studies on the fluxionality of the nanoclusters and how plastic and elastic fluxionalities play roles in highly efficient reaction pathways. Our results show that the low energy metastable isomers formed by plastic fluxionality can manifest high reactivity despite their minor occurrence probability in the mixture of catalyst isomers. In the end, the highly active metastable isomer may dominate the total observed reactivity. In addition, the isomerization between the global minimum structure and the highly active metastable isomer can be a central step in catalytic transformations in order to circumvent some difficult reaction steps and may govern the overall mechanism. In addition, the thermal fluctuation driven elastic fluxionality is also found to play a key role, complementary to plastic fluxionality. The elastic fluxionality creates substantial structural deformations of the active site, and these deformed geometries enable low activation energies and high catalytic activity, which cannot be found from the static equilibrium geometry of the catalyst. A dedicated global activity search algorithm is proposed to search for the optimal reaction pathway on fluxional nanoclusters. In summary, our studies demonstrate that thermal-driven fluxionality provides a different paradigm for understanding the high activity of nanoclusters under reaction conditions beyond the static description of geometric and electronic structure. We first summarize our previous results and then provide a perspective for further studies on how to investigate and take the advantage of the fluxional geometry of nanoclusters. We will defend in this Account that the static picture for the active site is not complete and might miss critical reaction pathways that are highly efficient and only open after thermally induced restructuring of the active site.

Atomic Structure Optimization with Machine-Learning Enabled Interpolation between Chemical Elements

We introduce a computational method for global optimization of structure and ordering in atomic systems. The method relies on interpolation between chemical elements, which is incorporated in a machine-learning structural fingerprint. The method is based on Bayesian optimization with Gaussian processes and is applied to the global optimization of Au-Cu bulk systems, Cu-Ni surfaces with CO adsorption, and Cu-Ni clusters. The method consistently identifies low-energy structures, which are likely to be the global minima of the energy. For the investigated systems with 23-66 atoms, the number of required energy and force calculations is in the range 3-75.

Machine learning potentials for complex aqueous systems made simple

Understanding complex materials, in particular those with solid–liquid interfaces, such as water on surfaces or under confinement, is a key challenge for technological and scientific progress. Although established simulation approaches have been able to provide important atomistic insight, ab initio techniques struggle with the required time and length scales, while force field methods can often be limited in terms of their accuracy. Here we show how these limitations can be overcome in a simple and automated machine learning procedure to provide accurate models of interactions at the ab initio level, as illustrated for a variety of complex aqueous systems. These developments open up the prospect of the straightforward exploration of many technologically relevant systems by molecular simulations.

Simulation techniques based on accurate and efficient representations of potential energy surfaces are urgently needed for the understanding of complex systems such as solid–liquid interfaces. Here we present a machine learning framework that enables the efficient development and validation of models for complex aqueous systems. Instead of trying to deliver a globally optimal machine learning potential, we propose to develop models applicable to specific thermodynamic state points in a simple and user-friendly process. After an initial ab initio simulation, a machine learning potential is constructed with minimum human effort through a data-driven active learning protocol. Such models can afterward be applied in exhaustive simulations to provide reliable answers for the scientific question at hand or to systematically explore the thermal performance of ab initio methods. We showcase this methodology on a diverse set of aqueous systems comprising bulk water with different ions in solution, water on a titanium dioxide surface, and water confined in nanotubes and between molybdenum disulfide sheets. Highlighting the accuracy of our approach with respect to the underlying ab initio reference, the resulting models are evaluated in detail with an automated validation protocol that includes structural and dynamical properties and the precision of the force prediction of the models. Finally, we demonstrate the capabilities of our approach for the description of water on the rutile titanium dioxide (110) surface to analyze the structure and mobility of water on this surface. Such machine learning models provide a straightforward and uncomplicated but accurate extension of simulation time and length scales for complex systems.

On application of deep learning to simplified quantum-classical dynamics in electronically excited states

Deep learning (DL) is applied to simulate non-adiabatic molecular dynamics of phenanthrene, using the time-dependent density functional based tight binding (TD-DFTB) approach for excited states combined with mixed quantum–classical propagation. Reference calculations rely on Tully’s fewest-switches surface hopping (FSSH) algorithm coupled to TD-DFTB, which provides electronic relaxation dynamics in fair agreement with various available experimental results. Aiming at describing the coupled electron-nuclei dynamics in large molecular systems, we then examine the combination of DL for excited-state potential energy surfaces (PESs) with a simplified trajectory surface hopping propagation based on the Belyaev–Lebedev (BL) scheme. We start to assess the accuracy of the TD-DFTB approach upon comparison of the optical spectrum with experimental and higher-level theoretical results. Using the recently developed SchNetPack (Schütt et al 2019 J. Chem. Theory Comput. 15 448–55) for DL applications, we train several models and evaluate their performance in predicting excited-state energies and forces. Then, the main focus is given to the analysis of the electronic population of low-lying excited states computed with the aforementioned methods. We determine the relaxation timescales and compare them with experimental data. Our results show that DL demonstrates its ability to describe the excited-state PESs. When coupled to the simplified BL scheme considered in this study, it provides reliable description of the electronic relaxation in phenanthrene as compared with either the experimental data or the higher-level FSSH/TD-DFTB theoretical results. Furthermore, the DL performance allows high-throughput analysis at a negligible cost.

Gaussian Process Regression for Materials and Molecules

We provide an introduction to Gaussian process regression (GPR) machine-learning methods in computational materials science and chemistry. The focus of the present review is on the regression of atomistic properties: in particular, on the construction of interatomic potentials, or force fields, in the Gaussian Approximation Potential (GAP) framework; beyond this, we also discuss the fitting of arbitrary scalar, vectorial, and tensorial quantities. Methodological aspects of reference data generation, representation, and regression, as well as the question of how a data-driven model may be validated, are reviewed and critically discussed. A survey of applications to a variety of research questions in chemistry and materials science illustrates the rapid growth in the field. A vision is outlined for the development of the methodology in the years to come.

Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems

Machine learning models are poised to make a transformative impact on chemical sciences by dramatically accelerating computational algorithms and amplifying insights available from computational chemistry methods. However, achieving this requires a confluence and coaction of expertise in computer science and physical sciences. This Review is written for new and experienced researchers working at the intersection of both fields. We first provide concise tutorials of computational chemistry and machine learning methods, showing how insights involving both can be achieved. We follow with a critical review of noteworthy applications that demonstrate how computational chemistry and machine learning can be used together to provide insightful (and useful) predictions in molecular and materials modeling, retrosyntheses, catalysis, and drug design.

Perspective on integrating machine learning into computational chemistry and materials science

Machine learning (ML) methods are being used in almost every conceivable area of electronic structure theory and molecular simulation. In particular, ML has become firmly established in the construction of high-dimensional interatomic potentials. Not a day goes by without another proof of principle being published on how ML methods can represent and predict quantum mechanical properties-be they observable, such as molecular polarizabilities, or not, such as atomic charges. As ML is becoming pervasive in electronic structure theory and molecular simulation, we provide an overview of how atomistic computational modeling is being transformed by the incorporation of ML approaches. From the perspective of the practitioner in the field, we assess how common workflows to predict structure, dynamics, and spectroscopy are affected by ML. Finally, we discuss how a tighter and lasting integration of ML methods with computational chemistry and materials science can be achieved and what it will mean for research practice, software development, and postgraduate training.