Bryce Meredig

A hybrid computational–experimental approach for automated crystal structure solution

Crystal structure solution from diffraction experiments is one of the most fundamental tasks in materials science, chemistry, physics and geology. Unfortunately, numerous factors render this process labour intensive and error prone. Experimental conditions, such as high pressure or structural metastability, often complicate characterization. Furthermore, many materials of great modern interest, such as batteries and hydrogen storage media, contain light elements such as Li and H that only weakly scatter X-rays. Finally, structural refinements generally require significant human input and intuition, as they rely on good initial guesses for the target structure. To address these many challenges, we demonstrate a new hybrid approach, first-principles-assisted structure solution (FPASS), which combines experimental diffraction data, statistical symmetry information and first-principles-based algorithmic optimization to automatically solve crystal structures. We demonstrate the broad utility of FPASS to clarify four important crystal structure debates: the hydrogen storage candidates MgNH and NH(3)BH(3); Li(2)O(2), relevant to Li-air batteries; and high-pressure silane, SiH(4).

Perspective: Web-based machine learning models for real-time screening of thermoelectric materials properties

Chemistries Michael W. Gaultois, a) Anton O. Oliynyk, Arthur Mar, Taylor D. Sparks, Gregory J. Mulholland, and Bryce Meredig b) Materials Research Laboratory and the Department of Chemistry and Biochemistry, University of California, Santa Barbara, California, 93106, USA Department of Chemistry, University of Alberta, Edmonton, Alberta, T6G 2G2, Canada Department of Materials Science and Engineering, University of Utah, Salt Lake City, Utah, 84112, USA Citrine Informatics, Redwood City, California, 94061, USAThe experimental search for new thermoelectric materials remains largely confined to a limited set of successful chemical and structural families, such as chalcogenides, skutterudites, and Zintl phases. In principle, computational tools such as density functional theory (DFT) offer the possibility of rationally guiding experimental synthesis efforts toward very different chemistries. However, in practice, predicting thermoelectric properties from first principles remains a challenging endeavor [J. Carrete et al., Phys. Rev. X 4, 011019 (2014)], and experimental researchers generally do not directly use computation to drive their own synthesis efforts. To bridge this practical gap between experimental needs and computational tools, we report an open machine learning-based recommendation engine (http://thermoelectrics.citrination.com) for materials researchers that suggests promising new thermoelectric compositions based on pre-screening about 25 000 known materials and also evaluates the feasibility of use...

High-Dimensional Materials and Process Optimization Using Data-Driven Experimental Design with Well-Calibrated Uncertainty Estimates

The optimization of composition and processing to obtain materials that exhibit desirable characteristics has historically relied on a combination of domain knowledge, trial and error, and luck. We propose a methodology that can accelerate this process by fitting data-driven models to experimental data as it is collected to suggest which experiment should be performed next. This methodology can guide the practitioner to test the most promising candidates earlier and can supplement scientific and engineering intuition with data-driven insights. A key strength of the proposed framework is that it scales to high-dimensional parameter spaces, as are typical in materials discovery applications. Importantly, the data-driven models incorporate uncertainty analysis, so that new experiments are proposed based on a combination of exploring high-uncertainty candidates and exploiting high-performing regions of parameter space. Over four materials science test cases, our methodology led to the optimal candidate being found with three times fewer required measurements than random guessing on average.

An improved interatomic potential for xenon in UO2: a combined density functional theory/genetic algorithm approach

We have created an improved xenon interatomic potential for use with existing UO2 potentials. This potential was fit to density functional theory calculations with the Hubbard U correction (DFT + U) using a genetic algorithm approach called iterative potential refinement (IPR). We examine the defect energetics of the IPR-fitted xenon interatomic potential as well as other, previously published xenon potentials. We compare these potentials to DFT + U derived energetics for a series of xenon defects in a variety of incorporation sites (large, intermediate, and small vacant sites). We find the existing xenon potentials overestimate the energy needed to add a xenon atom to a wide set of defect sites representing a range of incorporation sites, including failing to correctly rank the energetics of the small incorporation site defects (xenon in an interstitial and xenon in a uranium site neighboring uranium in an interstitial). These failures are due to problematic descriptions of Xe-O and/or Xe-U interactions of the previous xenon potentials. These failures are corrected by our newly created xenon potential: our IPR-generated potential gives good agreement with DFT + U calculations to which it was not fitted, such as xenon in an interstitial (small incorporation site) and xenon in a double Schottky defect cluster (large incorporation site). Finally, we note that IPR is very flexible and can be applied to a wide variety of potential forms and materials systems, including metals and EAM potentials.

Materials Data Infrastructure and Materials Informatics

Data-driven materials research requires two key supporting components: data infrastructure and informatics. In this chapter, we review the state of the art in materials data infrastructure, focusing in detail on four infrastructure projects spanning academia, government, and industry. We also discuss data standards as an enabling step on the path to community-scale materials data infrastructure. We then introduce materials informatics as a potent accelerator of materials development and highlight specific application areas, including polymer dielectrics and dielectric breakdown.

A Formation Energy Predictor for Crystalline Materials Using Ensemble Data Mining

Formation energy is one of the most important properties of a compound that is directly related to its stability. More negative the formation energy, the more stable the compound is likely to be. Here we describe the development and deployment of predictive models for formation energy, given the chemical composition of the material. The data-driven models described here are built using nearly 100,000 Density Functional Theory (DFT) calculations, which is a quantum mechanical simulation technique based on the electron density within the crystal structure of the material. These models are deployed in an online web-tool that takes a list of material compositions as input, generates over hundred composition-based attributes for each material and feeds them into the predictive models to obtain the predictions of formation energy. The online formation energy predictor is available at http://info.eecs.northwestern.edu/FEpredictor.

Can machine learning identify the next high-temperature superconductor? Examining extrapolation performance for materials discovery

Traditional machine learning (ML) metrics overestimate model performance for materials discovery. We introduce (1) leave-one-cluster-out cross-validation (LOCO CV) and (2) a simple nearest-neighbor benchmark to show that model performance in discovery applications strongly depends on the problem, data sampling, and extrapolation. Our results suggest that ML-guided iterative experimentation may outperform standard high-throughput screening for discovering breakthrough materials like high-Tc superconductors with ML.