Zhi Deng is the lead author in our recently published work in npj Computational Materials on a machine-learned (ML) electrostatic Spectral Neighbor Analysis Potential (eSNAP) for Li3N, a prototypical superionic conductor. By incorporating long-ranged electrostatics, we developed a highly accurate eSNAP model for Li3N that far outperforms traditional potentials in the prediction of energies, forces and properties such as lattice constants, elastic constants, and phonon dispersion curves. Most importantly, we demonstrate that the eSNAP enables long-time, large-scale Li diffusion studies in Li3N, computing the Haven ratio and simulating GB diffusion in Li3N for the first time to excellent agreement with experimental values.
Our group members are also co-authors in several recently published works.
Congratulations to Mahdi and Manas for passing their Master Thesis Defense!
We are pleased to announce that Richard’s follow-up work on the anisotropic work functions of the elements has been published in Surface Science. The work function is a fundamental electronic property of a solid that varies with the facets of a crystalline surface. It is a crucial parameter in spectroscopy as well as materials design, especially for technologies such as thermionic electron guns and Schottky barriers. In this work, we present the largest database of calculated work functions for elemental crystals to date. This well-validated database contains the anisotropic work functions of more than 100 polymorphs of about 72 elements. One significant advance is the development of an improved model for the work function of metals from atomic parameters such as the electronegativity and metallic radius based on Gauss’ law.
The work function database can be accessed at the Crystalium website together with other surface properties.
The Materials Virtual Lab is proud to announce the launch of crystals.ai, a website of curated models, software and datasets for AI in materials science. Here, you will find web applications implementing on-the-fly prediction of properties using our MEGNet and other models, open-source software frameworks for building your own AI models, as well as curated datasets for reproducible materials AI research.
Our paper on MatErials Graph Networks (MEGNet) for machine learning in crystals and molecules have been published in Chemistry of Materials. The article is available here. Key advances include the incorporation of state variables such as temperature, pressure and entropy, transfer learning from models with large data (e.g., formation energies) to models with smaller data (e.g., elastic constants) and extraction of chemical trends from learned elemental embeddings. These advances address key limitations in ML in materials science, such as data size limitations and physical interpretability.
We have also released all our codes and data in our open Github repo at https://github.com/materialsvirtuallab/megnet to enable others to reproduce and improve on our models.
Prof Ong is the Feature Editor in Mar 2019’s MRS Energy Quarterly article on “Artificial intelligence is aiding the search for energy materials”. In this article written by Prachi Patel, we interview various leaders in the field on their perspectives on how AI is being applied in energy materials design, from discovering entirely novel materials to enabling large-scale complex simulations to providing insights into how to synthesize materials. Check it out at https://doi.org/10.1557/mrs.2019.51.
Paul’s collaborative paper on “Rational Synthesis and Electrochemical Performance of LiVOPO4 Polymorphs” as part of the NorthEast Center for Chemical Energy Storage has just been published in Journal of Materials Chemistry A. Here, we have carried out a comprehensive experimental and DFT study of the α-I, β and ε polymorphs of LiVOPO4 and demonstrated how selectivity for each polymorph can be tuned through manipulating the precursor and oxidation environment. Further, we discuss how synthesis conditions may be used to improve the rate performance of β-LiVOPO4.
Prof Ong’s review article as a finalist for the Rising Stars in Computational Materials Science is now available. This review outlines the efforts in the Materials Virtual Lab to integrate software automation, data generation and curation and machine learning to (i) design and optimize technological materials for energy storage, energy efficiency and high-temperature alloys; (ii) develop scalable quantum-accurate models, and (iii) enhance the speed and accuracy in interpreting characterization spectra.
Our collaborative work with the Hu group @ Florida State University on functional defects in the Cl-doped Na3PS4 solid electrolyte has just been published in Advanced Functional Materials! Through tuning of Cl and Na vacancies in Cl-doped Na3PS4, we show that conductivities as high as 1.96 mS/cm can be achieved, with excellent full-cell performance in Na/Na3.0PS3.8Cl0.2/Na3V2(PO4)3 with a reversible capacity of 100 mAh/g at room temperature.
Chi has recently developed MatErials Graph Networks (MEGNet) based on DeepMind’s graph networks approach.We show that MEGNet models are a universal approach to machine learning for both crystals and molecules, outperforming prior ML models on a broad array of properties. We also demonstrate the incorporation of state (e.g., temperature, pressure) into MEGNet models, and how transfer learning can be used to accelerate and improve the accuracy of models trained on smaller data sets.
A preprint of our paper is published on arXiv and the models are available at https://github.com/materialsvirtuallab/megnet. We have also posted a useful data set of 69,000 crystals from the Materials Project on figshare for ML purposes.