Showing papers in "npj computational materials in 2022"

Two-dimensional Stiefel-Whitney insulators in liganded Xenes

Abstract: Two-dimensional (2D) Stiefel-Whitney insulator (SWI), which is characterized by the second Stiefel-Whitney class, is a new class of topological phases with zero Berry curvature. As a novel topological state, it has been well studied in theory but seldom realized in realistic materials. Here we propose that a large class of liganded Xenes, i.e., hydrogenated and halogenated 2D group-IV honeycomb lattices, are 2D SWIs. The nontrivial topology of liganded Xenes is identified by the bulk topological invariant and the existence of protected corner states. Moreover, the large and tunable band gap (up to 3.5 eV) of liganded Xenes will facilitate the experimental characterization of the 2D SWI phase. Our findings not only provide abundant realistic material candidates that are experimentally feasible, but also draw more fundamental research interest towards the topological physics associated with Stiefel-Whitney class in the absence of Berry curvature.

Accelerating materials discovery using artificial intelligence, high performance computing and robotics

Abstract: Abstract New tools enable new ways of working, and materials science is no exception. In materials discovery, traditional manual, serial, and human-intensive work is being augmented by automated, parallel, and iterative processes driven by Artificial Intelligence (AI), simulation and experimental automation. In this perspective, we describe how these new capabilities enable the acceleration and enrichment of each stage of the discovery cycle. We show, using the example of the development of a novel chemically amplified photoresist, how these technologies’ impacts are amplified when they are used in concert with each other as powerful, heterogeneous workflows.

Machine learning in concrete science: applications, challenges, and best practices

Abstract: Abstract Concrete, as the most widely used construction material, is inextricably connected with human development. Despite conceptual and methodological progress in concrete science, concrete formulation for target properties remains a challenging task due to the ever-increasing complexity of cementitious systems. With the ability to tackle complex tasks autonomously, machine learning (ML) has demonstrated its transformative potential in concrete research. Given the rapid adoption of ML for concrete mixture design, there is a need to understand methodological limitations and formulate best practices in this emerging computational field. Here, we review the areas in which ML has positively impacted concrete science, followed by a comprehensive discussion of the implementation, application, and interpretation of ML algorithms. We conclude by outlining future directions for the concrete community to fully exploit the capabilities of ML models.

A review of the recent progress in battery informatics

Abstract: Abstract Batteries are of paramount importance for the energy storage, consumption, and transportation in the current and future society. Recently machine learning (ML) has demonstrated success for improving lithium-ion technologies and beyond. This in-depth review aims to provide state-of-art achievements in the interdisciplinary field of ML and battery research and engineering, the battery informatics. We highlight a crucial hurdle in battery informatics, the availability of battery data, and explain the mitigation of the data scarcity challenge with a detailed review of recent achievements. This review is concluded with a perspective in this new but exciting field.

Magnetic Moment Tensor Potentials for collinear spin-polarized materials reproduce different magnetic states of bcc Fe

Abstract: Abstract We present the magnetic Moment Tensor Potentials (mMTPs), a class of machine-learning interatomic potentials, accurately reproducing both vibrational and magnetic degrees of freedom as provided, e.g., from first-principles calculations. The accuracy is achieved by a two-step minimization scheme that coarse-grains the atomic and the spin space. The performance of the mMTPs is demonstrated for the prototype magnetic system bcc iron, with applications to phonon calculations for different magnetic states, and molecular-dynamics simulations with fluctuating magnetic moments.

Representations of molecules and materials for interpolation of quantum-mechanical simulations via machine learning

Abstract: Computational study of molecules and materials from first principles is a cornerstone of physics, chemistry, and materials science, but limited by the cost of accurate and precise simulations. In settings involving many simulations, machine learning can reduce these costs, often by orders of magnitude, by interpolating between reference simulations. This requires representations that describe any molecule or material and support interpolation. We comprehensively review and discuss current representations and relations between them, using a unified mathematical framework based on many-body functions, group averaging, and tensor products. For selected state-of-the-art representations, we compare energy predictions for organic molecules, binary alloys, and Al-Ga-In sesquioxides in numerical experiments controlled for data distribution, regression method, and hyper-parameter optimization.

Learning two-phase microstructure evolution using neural operators and autoencoder architectures

Abstract: Abstract Phase-field modeling is an effective but computationally expensive method for capturing the mesoscale morphological and microstructure evolution in materials. Hence, fast and generalizable surrogate models are needed to alleviate the cost of computationally taxing processes such as in optimization and design of materials. The intrinsic discontinuous nature of the physical phenomena incurred by the presence of sharp phase boundaries makes the training of the surrogate model cumbersome. We develop a framework that integrates a convolutional autoencoder architecture with a deep neural operator (DeepONet) to learn the dynamic evolution of a two-phase mixture and accelerate time-to-solution in predicting the microstructure evolution. We utilize the convolutional autoencoder to provide a compact representation of the microstructure data in a low-dimensional latent space. After DeepONet is trained in the latent space, it can be used to replace the high-fidelity phase-field numerical solver in interpolation tasks or to accelerate the numerical solver in extrapolation tasks.

A machine-learned interatomic potential for silica and its relation to empirical models

Abstract: Abstract Silica (SiO 2 ) is an abundant material with a wide range of applications. Despite much progress, the atomistic modelling of the different forms of silica has remained a challenge. Here we show that by combining density-functional theory at the SCAN functional level with machine-learning-based interatomic potential fitting, a range of condensed phases of silica can be accurately described. We present a Gaussian approximation potential model that achieves high accuracy for the thermodynamic properties of the crystalline phases, and we compare its performance (and performance–cost trade-off) with that of multiple empirically fitted interatomic potentials for silica. We also include amorphous phases, assessing the ability of the potentials to describe structures of melt-quenched glassy silica, their energetic stability, and the high-pressure structural transition to a mainly sixfold-coordinated phase. We suggest that rather than standing on their own, machine-learned potentials for silica may be used in conjunction with suitable empirical models, each having a distinct role and complementing the other, by combining the advantages of the long simulation times afforded by empirical potentials and the near-quantum-mechanical accuracy of machine-learned potentials. This way, our work is expected to advance atomistic simulations of this key material and to benefit further computational studies in the field.

Design high-entropy carbide ceramics from machine learning

Abstract: Abstract High-entropy ceramics (HECs) have shown great application potential under demanding conditions, such as high stresses and temperatures. However, the immense phase space poses great challenges for the rational design of new high-performance HECs. In this work, we develop machine-learning (ML) models to discover high-entropy ceramic carbides (HECCs). Built upon attributes of HECCs and their constituent precursors, our ML models demonstrate a high prediction accuracy (0.982). Using the well-trained ML models, we evaluate the single-phase probability of 90 HECCs that are not experimentally reported so far. Several of these predictions are validated by our experiments. We further establish the phase diagrams for non-equiatomic HECCs spanning the whole composition space by which the single-phase regime can be easily identified. Our ML models can predict both equiatomic and non-equiatomic HECs based solely on the chemical descriptors of constituent transition-metal-carbide precursors, which paves the way for the high-throughput design of HECCs with superior properties.

Discovering plasticity models without stress data

Abstract: Abstract We propose an approach for data-driven automated discovery of material laws, which we call EUCLID (Efficient Unsupervised Constitutive Law Identification and Discovery), and we apply it here to the discovery of plasticity models, including arbitrarily shaped yield surfaces and isotropic and/or kinematic hardening laws. The approach is unsupervised , i.e., it requires no stress data but only full-field displacement and global force data; it delivers interpretable models, i.e., models that are embodied by parsimonious mathematical expressions discovered through sparse regression of a potentially large catalog of candidate functions; it is one-shot , i.e., discovery only needs one experiment. The material model library is constructed by expanding the yield function with a Fourier series, whereas isotropic and kinematic hardening is introduced by assuming a yield function dependency on internal history variables that evolve with the plastic deformation. For selecting the most relevant Fourier modes and identifying the hardening behavior, EUCLID employs physics knowledge, i.e., the optimization problem that governs the discovery enforces the equilibrium constraints in the bulk and at the loaded boundary of the domain. Sparsity promoting regularization is deployed to generate a set of solutions out of which a solution with low cost and high parsimony is automatically selected. Through virtual experiments, we demonstrate the ability of EUCLID to accurately discover several plastic yield surfaces and hardening mechanisms of different complexity.

First-principles search of hot superconductivity in La-X-H ternary hydrides

Abstract: Abstract Motivated by the recent claim of hot superconductivity with critical temperatures up to 550 K in La + x hydrides, we investigate the high-pressure phase diagram of compounds that may have formed in the experiment, using first-principles calculations for evolutionary crystal structure prediction and superconductivity. Starting from the hypothesis that the observed T c may be realized by successive heating upon a pre-formed LaH 10 phase, we examine plausible ternaries of lanthanum, hydrogen and other elements present in the diamond anvil cell: boron, nitrogen, carbon, platinum, gallium, gold. We find that only boron and, to a lesser extent, gallium form metastable superhydride-like structures that can host high- T c superconductivity, but the predicted T c ’s are incompatible with the experimental reports. Our results indicate that, while the claims of hot superconductivity should be reconsidered, it is very likely that unknown H-rich ternary or multinary phases containing lanthanum, hydrogen, and possibly boron or gallium may have formed under the reported experimental conditions, and that these may exhibit superconducting properties comparable, or even superior, to those of currently known hydrides.

High-throughput predictions of metal–organic framework electronic properties: theoretical challenges, graph neural networks, and data exploration

Abstract: Abstract With the goal of accelerating the design and discovery of metal–organic frameworks (MOFs) for electronic, optoelectronic, and energy storage applications, we present a dataset of predicted electronic structure properties for thousands of MOFs carried out using multiple density functional approximations. Compared to more accurate hybrid functionals, we find that the widely used PBE generalized gradient approximation (GGA) functional severely underpredicts MOF band gaps in a largely systematic manner for semi-conductors and insulators without magnetic character. However, an even larger and less predictable disparity in the band gap prediction is present for MOFs with open-shell 3 d transition metal cations. With regards to partial atomic charges, we find that different density functional approximations predict similar charges overall, although hybrid functionals tend to shift electron density away from the metal centers and onto the ligand environments compared to the GGA point of reference. Much more significant differences in partial atomic charges are observed when comparing different charge partitioning schemes. We conclude by using the dataset of computed MOF properties to train machine-learning models that can rapidly predict MOF band gaps for all four density functional approximations considered in this work, paving the way for future high-throughput screening studies. To encourage exploration and reuse of the theoretical calculations presented in this work, the curated data is made publicly available via an interactive and user-friendly web application on the Materials Project.

Phase classification of multi-principal element alloys via interpretable machine learning

Abstract: There is intense interest in uncovering design rules that govern the formation of various structural phases as a function of chemical composition in multi-principal element alloys (MPEAs). In this paper, we develop a machine learning (ML) approach built on the foundations of ensemble learning, post hoc model interpretability of black-box models and clustering analysis to establish a quantitative relationship between the chemical composition and experimentally observed phases of MPEAs. The novelty of our work stems from performing instance-level (or local) variable attribution analysis of ML predictions based on the breakdown method, and then identifying similar instances based on k-means clustering analysis of the breakdown results. We also complement the breakdown analysis with Ceteris Paribus profiles that showcase how the model response changes as a function of a single variable, when the values of all other variables are fixed. Results from local model interpretability analysis uncover key insights into variables that govern the formation of each phase. Our developed approach is generic, model-agnostic, and valuable to explain the insights learned by the black-box models. An interactive web application is developed to facilitate model sharing and accelerate the design of novel MPEAs with targeted properties

Artificial neural network approach for multiphase segmentation of battery electrode nano-CT images

Abstract: Abstract The segmentation of tomographic images of the battery electrode is a crucial processing step, which will have an additional impact on the results of material characterization and electrochemical simulation. However, manually labeling X-ray CT images (XCT) is time-consuming, and these XCT images are generally difficult to segment with histographical methods. We propose a deep learning approach with an asymmetrical depth encode-decoder convolutional neural network (CNN) for real-world battery material datasets. This network achieves high accuracy while requiring small amounts of labeled data and predicts a volume of billions voxel within few minutes. While applying supervised machine learning for segmenting real-world data, the ground truth is often absent. The results of segmentation are usually qualitatively justified by visual judgement. We try to unravel this fuzzy definition of segmentation quality by identifying the uncertainty due to the human bias diluted in the training data. Further CNN trainings using synthetic data show quantitative impact of such uncertainty on the determination of material’s properties. Nano-XCT datasets of various battery materials have been successfully segmented by training this neural network from scratch. We will also show that applying the transfer learning, which consists of reusing a well-trained network, can improve the accuracy of a similar dataset.

Explainable machine learning in materials science

Abstract: Abstract Machine learning models are increasingly used in materials studies because of their exceptional accuracy. However, the most accurate machine learning models are usually difficult to explain. Remedies to this problem lie in explainable artificial intelligence (XAI), an emerging research field that addresses the explainability of complicated machine learning models like deep neural networks (DNNs). This article attempts to provide an entry point to XAI for materials scientists. Concepts are defined to clarify what explain means in the context of materials science. Example works are reviewed to show how XAI helps materials science research. Challenges and opportunities are also discussed.

Low-overhead distribution strategy for simulation and optimization of large-area metasurfaces

Abstract: Abstract Fast and accurate electromagnetic simulation of large-area metasurfaces remains a major obstacle in automating their design. In this paper, we propose a metasurface simulation distribution strategy which achieves a linear reduction in the simulation time with the number of compute nodes. Combining this distribution strategy with a GPU-based implementation of the Transition-matrix method, we perform accurate simulations and adjoint sensitivity analysis of large-area metasurfaces. We demonstrate ability to perform a distributed simulation of large-area metasurfaces (over 600 λ × 600 λ ), while accurately accounting for scatterer-scatterer interactions significantly beyond the locally periodic approximation.

Hybrid magnetorheological elastomers enable versatile soft actuators

Abstract: Abstract Recent advances in magnetorheological elastomers (MREs) have posed the question on whether the combination of both soft- and hard-magnetic particles may open new routes to design versatile multifunctional actuators. Here, we conceptualise ultra-soft hybrid MREs (≈1–10 kPa stiffness) combining experimental and computational approaches. First, a comprehensive experimental characterisation is performed. The results unravel that the magneto-mechanical performance of hybrid MREs can be optimised by selecting an adequate mixing ratio between particles. Then, a multi-physics computational framework provides insights into the synergistic magneto-mechanical interactions at the microscale. Soft particles amplify the magnetisation and hard particles contribute to torsional actuation. Our numerical results suggest that the effective response of hybrid MREs emerges from these intricate interactions. Overall, we uncover exciting possibilities to push the frontiers of MRE solutions. These are demonstrated by simulating a bimorph beam that provides actuation flexibility either enhancing mechanical bending or material stiffening, depending on the magnetic stimulation.

MatSciBERT: A materials domain language model for text mining and information extraction

Abstract: Abstract A large amount of materials science knowledge is generated and stored as text published in peer-reviewed scientific literature. While recent developments in natural language processing, such as Bidirectional Encoder Representations from Transformers (BERT) models, provide promising information extraction tools, these models may yield suboptimal results when applied on materials domain since they are not trained in materials science specific notations and jargons. Here, we present a materials-aware language model, namely, MatSciBERT, trained on a large corpus of peer-reviewed materials science publications. We show that MatSciBERT outperforms SciBERT, a language model trained on science corpus, and establish state-of-the-art results on three downstream tasks, named entity recognition, relation classification, and abstract classification. We make the pre-trained weights of MatSciBERT publicly accessible for accelerated materials discovery and information extraction from materials science texts.

Van der Waals force-induced intralayer ferroelectric-to-antiferroelectric transition via interlayer sliding in bilayer group-IV monochalcogenides

Abstract: Two-dimensional materials with ferroelectric properties break the size effect of conventional ferroelectric materials and unlock unprecedented potentials of ferroelectric-related application at small length scales. In this work, using density functional theory (DFT) calculations, we discover a tribo-ferroelectricity behavior in a group of bilayer group-IV monochalcogenides (MX, with M = Ge, Sn and X = S, Se). Upon interlayer sliding over an in-plane unit cell length, the top layer exhibits a reversible intralayer ferroelectric switching, leading to a reversible transition between the ferroelectric (electric polarization of 40$\mu$C/cm$^2$) and antiferroelectric states in the bilayer MXs. Our results show that the interlayer van der Waals interaction, which is usually considered to be weak, can actually generate an in-plane lattice distortion and thus cause the breaking/forming of intralayer covalent bonds in the top layer, leading to the observed tribo-ferroelectricity phenomenon. This unique property has several advantages for energy harvesting over existing piezoelectric and triboelectric nanogenerators. The interlayer sliding-induced polarization change is as high as 40$\mu$C/cm$^2$, which can generate an open-circuit voltage two orders of magnitude higher than that of MoS$_2$-based nanogenerators. The polarization change occurs over a time period for interlayer sliding over a unit-cell length, leading to an ultrahigh polarization changing rate and thus an ultrahigh short-circuit current. The theoretical prediction of power output for the tribo-ferroelectric bilayer MXs at a moderate sliding speed 1 m/s is four orders of magnitude higher than the MoS$_2$ nanogenerator, indicating great potentials in energy harvesting applications.

Composition design of high-entropy alloys with deep sets learning

Abstract: Abstract High entropy alloys (HEAs) are an important material class in the development of next-generation structural materials, but the astronomically large composition space cannot be efficiently explored by experiments or first-principles calculations. Machine learning (ML) methods might address this challenge, but ML of HEAs has been hindered by the scarcity of HEA property data. In this work, the EMTO-CPA method was used to generate a large HEA dataset (spanning a composition space of 14 elements) containing 7086 cubic HEA structures with structural properties, 1911 of which have the complete elastic tensor calculated. The elastic property dataset was used to train a ML model with the Deep Sets architecture. The Deep Sets model has better predictive performance and generalizability compared to other ML models. Association rule mining was applied to the model predictions to describe the compositional dependence of HEA elastic properties and to demonstrate the potential for data-driven alloy design.

Automated pipeline for superalloy data by text mining

Abstract: Abstract Data provides a foundation for machine learning, which has accelerated data-driven materials design. The scientific literature contains a large amount of high-quality, reliable data, and automatically extracting data from the literature continues to be a challenge. We propose a natural language processing pipeline to capture both chemical composition and property data that allows analysis and prediction of superalloys. Within 3 h, 2531 records with both composition and property are extracted from 14,425 articles, covering γ ′ solvus temperature, density, solidus, and liquidus temperatures. A data-driven model for γ ′ solvus temperature is built to predict unexplored Co-based superalloys with high γ ′ solvus temperatures within a relative error of 0.81%. We test the predictions via synthesis and characterization of three alloys. A web-based toolkit as an online open-source platform is provided and expected to serve as the basis for a general method to search for targeted materials using data extracted from the literature.

How coherence is governing diffuson heat transfer in amorphous solids

Abstract: Abstract Thermal transport in amorphous materials has remained one of the fundamental questions in solid state physics while involving a very large field of applications. Using a heat conduction theory incorporating coherence, we demonstrate that the strong phase correlation between local and non-propagating modes, commonly named diffusons in the terminology of amorphous systems, triggers the conduction of heat. By treating the thermal vibrations as collective excitations, the significant contribution of diffusons, predominantly relying on coherence, further reveals interesting temperature and length dependences of thermal conductivity. The propagation length of diffuson clusters is found to reach the micron, overpassing the one of propagons. The explored wavelike behavior of diffusons uncovers the unsolved physical picture of mode correlation in prevailing models and further provides an interpretation of their ability to transport heat. This work introduces a framework for understanding thermal vibrations and transport in amorphous materials, as well as an unexpected insight into the wave nature of thermal vibrations.

Inverse design of truss lattice materials with superior buckling resistance

Abstract: Abstract Manipulating the architecture of materials to achieve optimal combinations of properties (inverse design) has always been the dream of materials scientists and engineers. Lattices represent an efficient way to obtain lightweight yet strong materials, providing a high degree of tailorability. Despite massive research has been done on lattice architectures, the inverse design problem of complex phenomena (such as structural instability) has remained elusive. Via deep neural network and genetic algorithm, we provide a machine-learning-based approach to inverse-design non-uniformly assembled lattices. Combining basic building blocks, our approach allows us to independently control the geometry and topology of periodic and aperiodic structures. As an example, we inverse-design lattice architectures with superior buckling performance, outperforming traditional reinforced grid-like and bio-inspired lattices by ~30–90% and 10–30%, respectively. Our results provide insights into the buckling behavior of beam-based lattices, opening an avenue for possible applications in modern structures and infrastructures.

Antiferromagnetic second-order topological insulator with fractional mass-kink

Abstract: Abstract Generally, the topological corner state in two-dimensional (2D) second-order topological insulator (SOTI) is equivalent to the well-known domain wall state, which is originated from the mass-inversion between two adjacent edges with phase shift of π. In this work, go beyond this conventional physical picture, we report a fractional mass-kink induced 2D SOTI in monolayer FeSe with canted checkerboard antiferromagnetic (AFM) order by analytic model and first-principles calculations. The canted spin associated in-plane Zeeman field can gap out the quantum spin Hall edge state of FeSe, forming a fractional mass-kink with phase shift of π/2 at the rectangular corner, and generating an in-gap topological corner state with fractional charge of e/4. Moreover, the topological corner state is robust to a finite perturbation, existing in both naturally and non-naturally cleaved corners, regardless of the edge orientation. Our results not only demonstrate a material system to realize the unique 2D AFM SOTI, but also pave a way to design the higher-order topological states from fractional mass-kink with arbitrary phase shift.

Long-range current-induced spin accumulation in chiral crystals

Abstract: Chiral materials, similarly to human hands, have distinguishable right-handed and left-handed enantiomers which may behave differently in response to external stimuli. Here, we use for the first time an approach based on the density functional theory (DFT)+PAOFLOW calculations to quantitatively estimate the so-called collinear Rashba-Edelstein effect (REE) that generates spin accumulation parallel to charge current and can manifest as chirality-dependent charge-to-spin conversion in chiral crystals. Importantly, we reveal that the spin accumulation induced in the bulk by an electric current is intrinsically protected by the quasi-persistent spin helix arising from the crystal symmetries present in chiral systems with the Weyl spin-orbit coupling. In contrast to conventional REE, spin transport can be preserved over large distances, in agreement with the recent observations for some chiral materials. This allows, for example, the generation of spin currents from spin accumulation, opening novel routes for the design of solid-state spintronics devices.

2D spontaneous valley polarization from inversion symmetric single-layer lattices

Abstract: Abstract 2D spontaneous valley polarization attracts great interest both for its fundamental physics and for its potential applications in advanced information technology, but it can only be obtained from inversion asymmetric single-layer crystals, while the possibility to create 2D spontaneous valley polarization from inversion symmetric single-layer lattices remains unknown. Here, starting from inversion symmetric single-layer lattices, a general design principle for realizing 2D spontaneous valley polarization based on van der Waals interaction is mapped out. Using first-principles calculations, we further demonstrate the feasibility of this design principle in a real material of T-FeCl 2 . More remarkably, such design principle exhibits the additional exotic out-of-plane ferroelectricity, which could manifest many distinctive properties, for example, ferroelectricity-valley coupling and magnetoelectric coupling. The explored design-guideline and phenomena are applicable to a vast family of 2D materials. Our work not only opens up a platform for 2D valleytronic research but also promises the fundamental research of coupling physics in 2D lattices.

Physics-informed deep learning for solving phonon Boltzmann transport equation with large temperature non-equilibrium

Abstract: Abstract Phonon Boltzmann transport equation (BTE) is a key tool for modeling multiscale phonon transport, which is critical to the thermal management of miniaturized integrated circuits, but assumptions about the system temperatures (i.e., small temperature gradients) are usually made to ensure that it is computationally tractable. To include the effects of large temperature non-equilibrium, we demonstrate a data-free deep learning scheme, physics-informed neural network (PINN), for solving stationary, mode-resolved phonon BTE with arbitrary temperature gradients. This scheme uses the temperature-dependent phonon relaxation times and learns the solutions in parameterized spaces with both length scale and temperature gradient treated as input variables. Numerical experiments suggest that the proposed PINN can accurately predict phonon transport (from 1D to 3D) under arbitrary temperature gradients. Moreover, the proposed scheme shows great promise in simulating device-level phonon heat conduction efficiently and can be potentially used for thermal design.

Efficient and universal characterization of atomic structures through a topological graph order parameter

Abstract: A new graph-based order parameter is introduced for the characterization of atomistic structures. The order parameter is universal to any material/chemical system, and is transferable to all structural geometries. Three sets of data are used to validate both the generalizability and accuracy of the algorithm: (1) liquid lithium configurations spanning up to 300 GPa, (2) condensed phases of carbon along with nanotubes and buckyballs at ambient and high temperature, and (3) a diverse set of aluminum configurations including surfaces, compressed and expanded lattices, point defects, grain boundaries, liquids, nanoparticles, all at non-zero temperatures. The aluminum configurations are also compared to existing characterization methods for both speed and accuracy. Our order parameter uniquely classifies every configuration and outperforms all crystalline order parameters studied here, opening the door for its use in a multitude of complex application spaces that can require fine configurational characterization of materials.

Calibration after bootstrap for accurate uncertainty quantification in regression models

Abstract: Abstract Obtaining accurate estimates of machine learning model uncertainties on newly predicted data is essential for understanding the accuracy of the model and whether its predictions can be trusted. A common approach to such uncertainty quantification is to estimate the variance from an ensemble of models, which are often generated by the generally applicable bootstrap method. In this work, we demonstrate that the direct bootstrap ensemble standard deviation is not an accurate estimate of uncertainty but that it can be simply calibrated to dramatically improve its accuracy. We demonstrate the effectiveness of this calibration method for both synthetic data and numerous physical datasets from the field of Materials Science and Engineering. The approach is motivated by applications in physical and biological science but is quite general and should be applicable for uncertainty quantification in a wide range of machine learning regression models.

Data-augmentation for graph neural network learning of the relaxed energies of unrelaxed structures

Abstract: Abstract Computational materials discovery has grown in utility over the past decade due to advances in computing power and crystal structure prediction algorithms (CSPA). However, the computational cost of the ab initio calculations required by CSPA limits its utility to small unit cells, reducing the compositional and structural space the algorithms can explore. Past studies have bypassed unneeded ab initio calculations by utilizing machine learning to predict the stability of a material. Specifically, graph neural networks trained on large datasets of relaxed structures display high fidelity in predicting formation energy. Unfortunately, the geometries of structures produced by CSPA deviate from the relaxed state, which leads to poor predictions, hindering the model’s ability to filter unstable material. To remedy this behavior, we propose a simple, physically motivated, computationally efficient perturbation technique that augments training data, improving predictions on unrelaxed structures by 66%. Finally, we show how this error reduction can accelerate CSPA.