Our paper on “Learning properties of ordered and disordered materials from multi-fidelity data” has just been published in the inaugural issue of Nature Computational Science! In this work, we address two major impediments to ML for materials science. The first impediment is that valuable accurate data is much more expensive to obtain than less accurate data. Using multi-fidelity materials graph networks (MEGNet), we show that we can use the lower quality data to improve underlying structural representations in models, and in the process significantly improve predictions on smaller, more valuable data (e.g., experimental measurements). The second impediment is that making predictions on disordered materials, which is the vast majority of known materials, is much more difficult than on ordered materials. We show that the elemental representations (embeddings) learned by our MEGNet models can be used to directly model disordered materials. The article is available here. For an independent perspective on the findings, check out the Nature News & Views article. All data and code are available from http://crystals.ai and the Github repository.
Chen Zheng’s and Chi Chen’s paper on “Random Forest Models for Accurate Identiﬁcation of Coordination Environments from X-Ray Absorption Near-Edge Structure” has just been published in the second issue of the Cell Press journal Patterns! Analyzing coordination environments using X-ray absorption spectroscopy has broad applications in solid-state physics and material chemistry. Here, we show that random forest models trained on 190,000 K-edge XANES can directly identify the main atomic coordination environment with a high accuracy of 85.4%. More importantly, we show that the random forest models can be used to predict coordination environments from experimental K-edge XANES with minimal loss in accuracy. Check out the work here.
Weike’s paper on “Deep Neural Networks for Accurate Predictions of Crystal Stability” is now out in Nature Communications. Predicting the stability of crystals is one of the central problems in materials science. Here, we show that deep neural networks, i.e., algorithms that mimic the animal brain, utilizing just the electronegativity and ionic radii as inputs can predict formation energies of crystals with extremely high accuracy. We also demonstrate how these models can be generalized for mixed crystals using a binary encoding scheme, and use it to identify thousands of potentially stable new compositions. We have published a web app (http://crystals.ai) that enables anyone to use these models. News: UCSD News: Scientists use artificial neural networks to predict new stable materials Paper: W. Ye, C. Chen, Z. Wang, I.-H. Chu, S.P. Ong, Deep neural networks for accurate predictions of crystal stability, Nat. Commun. 9 (2018) 3800. doi:10.1038/s41467-018-06322-x.
Chen’s paper on “Automated generation and ensemble-learned matching of X-ray absorption spectra” has been published in npj Computational Materials. In this work, we developed XASdb, a large database of computed reference X-ray absorption spectra (XAS), and a novel Ensemble-Learned Spectra IdEntification (ELSIE) algorithm for the matching of spectra. XASdb currently hosts more than 800,000 K-edge X-ray absorption near-edge spectra (XANES) for over 40,000 materials from the open-science Materials Project database. We will demonstrate that the ELSIE algorithm, which combines 33 weak “learners” comprising a set of preprocessing steps and a similarity metric, can achieve up to 84.2% accuracy in identifying the correct oxidation state and coordination environment. The XASdb with the ELSIE algorithm has been integrated into a web application in the Materials Project, providing an important new public resource for the analysis of XAS to all materials researchers. Finally, the ELSIE algorithm itself has been made available as part of Veidt, an open source machine learning library for materials science.
Prof Ong has been appointed to the Editorial Board of iScience. iScience is a new interdisciplinary, open-access journal by Cell Press that publishes basic and applied research that advances a specific field across life, physical, and earth sciences. Check out their latest articles at http://www.cell.com/iscience/home.
Congratulations to Zhenbin Wang for successfully defending his PhD thesis on Mar 6! Zhenbin is the first PhD graduate from the Materials Virtual Lab. Zhenbin joined the Materials Virtual Lab in Sep 2014 from the University of Science and Technology of China. During his PhD, Zhenbin has done ground-breaking work on the design and discovery of phosphor materials for white light-emitting diodes. He has devised new ways to screen for narrow-band red-emitting phosphors, provided useful optimization insights for the β-SiAlON green phosphor, and discovered a completely novel, earth-abundant phosphor host Sr2LiAlO4 that has been confirmed experimentally. Zhenbin is also an outstanding mentor to his fellow group members, having helped guide many to their own research findings. Check out Zhenbin’s tribute video from group members and photos of the defense and celebration by clicking on the full post!
Our paper on “Mining Unexplored Chemistries for Phosphors for High-Color-Quality White-Light-Emitting Diodes” has been published in Joule. Using supercomputers and data mining, we identified Sr2LiAlO4, the first known Sr-Li-Al-O quaternary crystal, as a highly promising phosphor material in low-cost, high-color-quality white LEDs. Eu2+ and Ce3+-activated Sr2LiAlO4 phosphors exhibit broad green-yellow and blue emissions, respectively, with excellent thermal quenching resistance of > 88% intensity at 150oC. A prototype phosphor-converted white LED utilizing Sr2LiAlO4-based phosphors yields an excellent color rendering index exceeding 90. This work is a collaboration between the Materials Virtual Lab (UCSD), McKittrick group (UCSD) and Im group (Chonnam University). The lead authors are Zhenbin Wang, Jungmin Ha and Yoon Hwa Kim. More information about this work can be found in the Jacobs School of Engineering News as well as Science Daily, Phys.org, etc.
Professor Ong recently gave a plenary talk on “Creating It from Bit – Designing Materials by Integrating Quantum Mechanics, Informatics and Computer Science” at the 57th Sanibel Symposium held on St Simon’s Island in Georgia, USA. The slides of this talk at available on SlideShare. In this talk, he discussed two emerging trends that holds the promise to continue to push the envelope in computational design of materials. The first trend is the development of robust software and data frameworks for the automatic generation, storage and analysis of materials data sets. The second is the advent of reliable central materials data repositories, such as the Materials Project, which provides the research community with efficient access to large quantities of property information that can be mined for trends or new materials. The talk showed how we have leveraged on these new tools to accelerate discovery and design in energy and structural materials as well as our efforts in contributing back to the community through further tool or data development, and provide perspectives on future challenges in high-throughput computational materials design.
Professor Ong has been appointed to the Editorial Board of Computational Materials Science. The goal of Computational Materials Science is to report on results that provide new insights into, or significantly expand our understanding of, the properties of materials or phenomena associated with their design, synthesis, processing, characterization, and utilization. All aspects of modern materials modeling are of interest, including quantum chemical methods, density functional theory (DFT), semi-empirical and classical approaches, statistical mechanics, atomic-scale simulations, mesoscale modeling, and phase-field techniques.