Crystal graph attention networks for the prediction of stable materials

Accelerated Discovery of Halide Perovskite Materials via Computational Methods: A Review

Halide perovskites have gained considerable attention in materials science due to their exceptional optoelectronic properties, including high absorption coefficients, excellent charge-carrier mobilities, and tunable band gaps, which make them highly promising for applications in photovoltaics, light-emitting diodes, synapses, and other optoelectronic devices. However, challenges such as long-term stability and lead toxicity hinder large-scale commercialization. Computational methods have become essential in this field, providing insights into material properties, enabling the efficient screening of large chemical spaces, and accelerating discovery processes through high-throughput screening and machine learning techniques. This review further discusses the role of computational tools in the accelerated discovery of high-performance halide perovskite materials, like the double perovskites A2BX6 and A2BB'X6, zero-dimensional perovskite A3B2X9, and novel halide perovskite ABX6. This review provides significant insights into how computational methods have accelerated the discovery of high-performance halide perovskite. Challenges and future perspectives are also presented to stimulate further research progress.

Synergy of machine learning and density functional theory calculations for predicting experimental Lewis base affinity and Lewis polybase binding atoms

Investigation of Lewis acid-base interactions has been conducted by ab initio calculations and machine learning (ML) models. This study aims to resolve two critical tasks that have not been quantitatively investigated. First, ML models developed from density functional theory (DFT) calculations predict experimental BF3 affinity with Pearson correlation coefficients around 0.9 and mean absolute errors around 10 kJ mol-1. The ML models are trained by DFT-calculated BF3 affinity of more than 3000 adducts, with input features readily obtained by rdkit. Second, the ML models have the capability of predicting the relative strength of Lewis base binding atoms in Lewis polybases, which is either an extremely challenging task to conduct experimentally or a computationally expensive task for ab initio methods. The study demonstrates and solidifies the potential of combining DFT calculations and ML models to predict experimental properties, especially those that are scarce and impractical to empirically acquire.

Generative learning facilitated discovery of high-entropy ceramic dielectrics for capacitive energy storage

Dielectric capacitors offer great potential for advanced electronics due to their high power densities, but their energy density still needs to be further improved. High-entropy strategy has emerged as an effective method for improving energy storage performance, however, discovering new high-entropy systems within a high-dimensional composition space is a daunting challenge for traditional trial-and-error experiments. Here, based on phase-field simulations and limited experimental data, we propose a generative learning approach to accelerate the discovery of high-entropy dielectrics in a practically infinite exploration space of over 1011 combinations. By encoding-decoding latent space regularities to facilitate data sampling and forward inference, we employ inverse design to screen out the most promising combinations via a ranking strategy. Through only 5 sets of targeted experiments, we successfully obtain a Bi(Mg0.5Ti0.5)O3-based high-entropy dielectric film with a significantly improved energy density of 156 J cm-3 at an electric field of 5104 kV cm-1, surpassing the pristine film by more than eight-fold. This work introduces an effective and innovative avenue for designing high-entropy dielectrics with drastically reduced experimental cycles, which could be also extended to expedite the design of other multicomponent material systems with desired properties.

Sampling the Materials Space for Conventional Superconducting Compounds

A large scale study of conventional superconducting materials using a machine-learning accelerated high-throughput workflow is performed, starting by creating a comprehensive dataset of around 7000 electron-phonon calculations performed with reasonable convergence parameters. This dataset is then used to train a robust machine learning model capable of predicting the electron-phonon and superconducting properties based on structural, compositional, and electronic ground-state properties. Using this machine, the transition temperatures (Tc ) of approximately 200 000 metallic compounds are evaluated, all of which are on the convex hull of thermodynamic stability (or close to it) to maximize the probability of synthesizability. Compounds predicted to have Tc values exceeding 5 K are further validated using density-functional perturbation theory. As a result, 541 compounds with Tc values surpassing 10 K, encompassing a variety of crystal structures and chemical compositions, are identified. This work is complemented with a detailed examination of several interesting materials, including nitrides, hydrides, and intermetallic compounds. Particularly noteworthy is LiMoN2 , which is predicted to be superconducting in the stoichiometric trigonal phase, with a Tc exceeding 38 K. LiMoN2 has previously been synthesized in this phase, further heightening its potential for practical applications.

Polytypic Two-Dimensional FeAs with High Anisotropy

In the realm of two-dimensional (2D) crystal growth, the chemical composition often determines the thermodynamically favored crystallographic structures. This relationship poses a challenge in synthesizing novel 2D crystals without altering their chemical elements, resulting in the rarity of achieving specific crystallographic symmetries or lattice parameters. We present 2D polymorphic FeAs crystals that completely differ from bulk orthorhombic FeAs (Pnma), differing in the stacking sequence, i.e., polytypes. Preparing polytypic FeAs outlines a strategy for independently controlling each symmetry operator, which includes the mirror plane for 2Q-FeAs (I4/mmm) and the glide plane for 1Q-FeAs (P4/nmm). As such, compared to bulk FeAs, polytypic 2D FeAs shows highly anisotropic properties such as electrical conductivity, Young's modulus, and friction coefficient. This work represents a concept of expanding 2D crystal libraries with a given chemical composition but various crystal symmetries.

Accelerating the prediction of stable materials with machine learning

Despite the rise in computing power, the large space of possible combinations of elements and crystal structure types makes large-scale high-throughput surveys of stable materials prohibitively expensive, especially for complex materials and materials subject to environmental conditions such as finite temperature. When physics-based computational methods and labor-intensive experiments are not feasible, machine learning (ML) methods can be a rapid and powerful alternative. Owing to a wealth of experimental and first-principles data as well as improved ML frameworks designed for materials modeling, ML is shown to be effective in predicting stability parameters and accelerating the discovery of new stable materials. In this Review, we summarize the most recent advancements in applying ML methodologies in predicting materials stability, focusing particularly on predictions of zero- and finite-temperature stability. We also highlight the need for more ML development in predictions of other thermodynamic knobs, such as pressure and surface/interfacial energy, which practically impact materials stability.

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.

Interference recommendation for the pump sizing process in progressive cavity pumps using graph neural networks

Pump sizing is the process of dimensional matching of an impeller and stator to provide a satisfactory performance test result and good service life during the operation of progressive cavity pumps. In this process, historical data analysis and dimensional monitoring are done manually, consuming a large number of man-hours and requiring a deep knowledge of progressive cavity pump behavior. This paper proposes the use of graph neural networks in the construction of a prototype to recommend interference during the pump sizing process in a progressive cavity pump. For this, data from different applications is used in addition to individual control spreadsheets to build the database used in the prototype. From the pre-processed data, complex network techniques and the betweenness centrality metric are used to calculate the degree of importance of each order confirmation, as well as to calculate the dimensionality of the rotors. Using the proposed method a mean squared error of 0.28 is obtained for the cases where there are recommendations for order confirmations. Based on the results achieved, it is noticeable that there is a similarity of the dimensions defined by the project engineers during the pump sizing process, and this outcome can be used to validate the new design definitions.

Quantum Graph Neural Network Models for Materials Search

Inspired by classical graph neural networks, we discuss a novel quantum graph neural network (QGNN) model to predict the chemical and physical properties of molecules and materials. QGNNs were investigated to predict the energy gap between the highest occupied and lowest unoccupied molecular orbitals of small organic molecules. The models utilize the equivariantly diagonalizable unitary quantum graph circuit (EDU-QGC) framework to allow discrete link features and minimize quantum circuit embedding. The results show QGNNs can achieve lower test loss compared to classical models if a similar number of trainable variables are used, and converge faster in training. This paper also provides a review of classical graph neural network models for materials research and various QGNNs.

Machine-Learning-Assisted Determination of the Global Zero-Temperature Phase Diagram of Materials

Crystal-graph attention neural networks have emerged recently as remarkable tools for the prediction of thermodynamic stability. The efficacy of their learning capabilities and their reliability is however subject to the quantity and quality of the data they are fed. Previous networks exhibit strong biases due to the inhomogeneity of the training data. Here a high-quality dataset is engineered to provide a better balance across chemical and crystal-symmetry space. Crystal-graph neural networks trained with this dataset show unprecedented generalization accuracy. Such networks are applied to perform machine-learning-assisted high-throughput searches of stable materials, spanning 1 billion candidates. In this way, the number of vertices of the global T = 0 K phase diagram is increased by 30% and find more than ≈150 000 compounds with a distance to the convex hull of stability of less than 50 meV atom^-1 . The discovered materials are then accessed for applications, identifying compounds with extreme values of a few properties, such as superconductivity, superhardness, and giant gap-deformation potentials.

Geometric Deep Learning for Molecular Crystal Structure Prediction

We develop and test new machine learning strategies for accelerating molecular crystal structure ranking and crystal property prediction using tools from geometric deep learning on molecular graphs. Leveraging developments in graph-based learning and the availability of large molecular crystal data sets, we train models for density prediction and stability ranking which are accurate, fast to evaluate, and applicable to molecules of widely varying size and composition. Our density prediction model, MolXtalNet-D, achieves state-of-the-art performance, with lower than 2% mean absolute error on a large and diverse test data set. Our crystal ranking tool, MolXtalNet-S, correctly discriminates experimental samples from synthetically generated fakes and is further validated through analysis of the submissions to the Cambridge Structural Database Blind Tests 5 and 6. Our new tools are computationally cheap and flexible enough to be deployed within an existing crystal structure prediction pipeline both to reduce the search space and score/filter crystal structure candidates.

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.

Machine learning-enabled high-entropy alloy discovery

High-entropy alloys are solid solutions of multiple principal elements that are capable of reaching composition and property regimes inaccessible for dilute materials. Discovering those with valuable properties, however, too often relies on serendipity, because thermodynamic alloy design rules alone often fail in high-dimensional composition spaces. We propose an active learning strategy to accelerate the design of high-entropy Invar alloys in a practically infinite compositional space based on very sparse data. Our approach works as a closed-loop, integrating machine learning with density-functional theory, thermodynamic calculations, and experiments. After processing and characterizing 17 new alloys out of millions of possible compositions, we identified two high-entropy Invar alloys with extremely low thermal expansion coefficients around 2 × 10^-6 per degree kelvin at 300 kelvin. We believe this to be a suitable pathway for the fast and automated discovery of high-entropy alloys with optimal thermal, magnetic, and electrical properties.

SepNet: A neural network for directionally correlated data

Multi-dimensional tensor data appear in diverse settings, including multichannel signals, spectrograms, and hyperspectral data from remote sensing. In many cases, these data are directionally correlated, i.e. the correlation between variables from different dimensions is significantly weaker than the correlation between variables from the same dimension. Convolutional neural networks are readily applicable to directionally correlated data but are often inefficient, as they impose many unnecessary connections between neurons. Here we propose a novel architecture, SepNet, specifically for directionally correlated datasets. SepNet uses directional operators to extract directional features from each dimension separately, followed by a linear operator along the depth to generate higher-level features from the directional features. Experiments on two representative directionally correlated datasets showed that SepNet improved network efficiency up to 100-fold while maintaining high accuracy comparable with state-of-the-art convolutional neural network models. Furthermore, SepNet can be flexibly constructed with minimal restriction on the output shape of each layer. These results reveal the potential of data-specific architecting of neural networks.

Formula Graph Self-Attention Network for Representation-Domain Independent Materials Discovery

The success of machine learning (ML) in materials property prediction depends heavily on how the materials are represented for learning. Two dominant families of material descriptors exist, one that encodes crystal structure in the representation and the other that only uses stoichiometric information with the hope of discovering new materials. Graph neural networks (GNNs) in particular have excelled in predicting material properties within chemical accuracy. However, current GNNs are limited to only one of the above two avenues owing to the little overlap between respective material representations. Here, a new concept of formula graph which unifies stoichiometry-only and structure-based material descriptors is introduced. A self-attention integrated GNN that assimilates a formula graph is further developed and it is found that the proposed architecture produces material embeddings transferable between the two domains. The proposed model can outperform some previously reported structure-agnostic models and their structure-based counterparts while exhibiting better sample efficiency and faster convergence. Finally, the model is applied in a challenging exemplar to predict the complex dielectric function of materials and nominate new substances that potentially exhibit epsilon-near-zero phenomena.

A dataset of 175k stable and metastable materials calculated with the PBEsol and SCAN functionals

In the past decade we have witnessed the appearance of large databases of calculated material properties. These are most often obtained with the Perdew-Burke-Ernzerhof (PBE) functional of density-functional theory, a well established and reliable technique that is by now the standard in materials science. However, there have been recent theoretical developments that allow for increased accuracy in the calculations. Here, we present a dataset of calculations for 175k crystalline materials obtained with two functionals: geometry optimizations are performed with PBE for solids (PBEsol) that yields consistently better geometries than the PBE functional, and energies are obtained from PBEsol and from SCAN single-point calculations at the PBEsol geometry. Our results provide an accurate overview of the landscape of stable (and nearly stable) materials, and as such can be used for reliable predictions of novel compounds. They can also be used for training machine learning models, or even for the comparison and benchmark of PBE, PBEsol, and SCAN.

Measurement(s)	optimized geometry (PBESol) • total energy (PBESol, Scan) • bandgap (PBESol, Scan)
Technology Type(s)	Density functional theory (VASP)
Factor Type(s)	Exchange correlation functional • Crystal structure