Prof Ong gave an invited seminar talk at the National University of Singapore on Jul 5 2023. In this talk, Prof Ong discusses the different ways in which machine learning (ML) can be used to improve or accelerate the various steps of in silico materials design. The general goal is to preserve the universality and accuracy of ab initio approaches as far as possible while achieving orders of magnitude speed-ups and improved scaling. Prof Ong shared his view that graph deep learning models trained on large diverse materials datasets, such as the M3GNet universal potential, are the “foundation” models for materials science. He further argues that the most robust approach is to replace the smallest, most expensive step in the materials design workflow with ML and preserve as much as the physics of thermodynamics, kinetics, etc. in the computation of materials properties.
The Materials Virtual Lab is proud to be part of a collaborative work titled “Compositionally complex perovskite oxides: Discovering a new class of solid electrolytes with interface-enabled conductivity improvements” recently published in Cell Press Matter! This work was performed under the Materials Research Science and Engineering Center (MRSEC) at UC Irvine together with the groups of Prof Xiaoqing Pan @ UCI and Prof Jian Luo @ UCSD. Compositionally complex ceramics (CCCs), including high-entropy ceramics, offer a vast, unexplored compositional space for materials discovery. Here, we propose non-equimolar compositionally complex perovskite oxide (CCPO) solid electrolytes with improved lithium ionic conductivities beyond the limit of conventional doping. For example, we demonstrate that the ionic conductivity can be improved by >60% in (Li0.375Sr0.4375)(Ta0.375Nb0.375Zr0.125Hf0.125)O3-d. MAVRL group member, Ji Qi, developed a machine learning interatomic potential (MLIP) for this 7-component CCPO using an active learning protocol, and demonstrated this enhanced ionic conductivity can be attributed to enhanced GB diffusitivity that is related to the absence of Li depletion at GB regions, which is observed in the resistive GB of the LLTO. This work suggests new routes for discovering and tailoring CCCs for energy storage and many other applications. Check out this work here.
We are excited to announce that Materials Graph Library (matgl), our Deep Graph Library/PyTorch reimplementation of the MatErials Graph Network (MEGNet) and Materials 3-body Graph Network (M3GNet) models, is now ready for widespread beta testing! We finally achieved near-feature parity with the original implementations in Tensorflow after months of hard work. The new MatGL includes retrained models of the M3GNet universal potential and the MEGNet formation energy and multi-fidelity band gap models. We have also taken the trouble to include example notebooks to get users started quickly. We believe this new implementation will be more future-proof and extensible. Feedback/issue reports are definitely welcome. This is a collaborative effort between the Materials Virtual Lab and Intel Labs.
Prof Shyue Ping Ong gave a talk at the 243rd Electrochemical Society (ECS) Meeting in Boston on May 30 2003 on “Machine Learning for Solid State Batteries: Progress vs Hype:. Machine learning in materials science and solid-state batteries are two topics that have captured the imagination of researchers in recent years. Therefore, it is unsurprising that many researchers have attempted to apply the advances in ML to the discovery and study of materials for solid-state batteries. In this talk, Prof Ong discusses the importance of going back to fundamentals in the application of ML to materials for solid-state batteries. I will highlight areas where ML has had a transformative impact on our understanding and discovery of solid electrolytes, and what are some of the remaining challenges that remain to be surmounted.
Hui’s swansong collaborative work on “Multi-scale investigation of short-range order and dislocation glide in MoNbTi and TaNbTi multi-principal element alloys” is now out in npj Computational Materials! This is a really exciting work that showcases an electron (DFT) to atom (machine learning interatomic potentials) to continuum (phase field dislocation dynamics) approach to the study of materials. It is a collaborative work between Hui and Lauren Fey of the group of Irene Beyerlein at UC Santa Barbara. Refractory multi-principal element alloys (RMPEAs) are promising materials for high-temperature structural applications. In this work, we performed an electron-to-atom-to-continuum study of the role of short-range ordering (SRO) on dislocation glide in the MoNbTi and TaNbTi RMPEAs. Monte carlo/molecular dynamics simulations with a moment tensor potential show that MoNbTi exhibits a much greater degree of SRO than TaNbTi and the local composition has a direct effect on the unstable stacking fault energies (USFEs). From mesoscale phase-field dislocation dynamics simulations, we find that increasing SRO leads to higher mean USFEs and stress required for dislocation glide. The gliding dislocations experience significant hardening due to pinning and depinning caused by random compositional fluctuations, with higher SRO decreasing the degree of USFE dispersion and hence, amount of hardening. […]
Jean Fan’s recent column in Nature on “Why it’s worth making computational methods easy to use” is an excellent article on a topic close to my heart. The following quote rings especially true. “We probably spent as many hours making STdeconvolve accessible as we did in developing it. Some of my colleagues have been surprised by this effort, as those hours won’t lead to new publications.” Our group is the maintainer of pymatgen, maml, matgl and a few other software used extensively by the materials science community. Colleagues frequently asked me the same question – “Why do I do it? Surely a professor can spend the time writing proposals, papers, etc.?” I disagree. Our group’s code is a critical avenue in which we contribute and engage with the community. A well written and maintained code can probably 10-100x the impact of a work beyond that one publication. I argue the Materials Virtual Lab, in open collaboration with thousands of other researchers, have saved millions of hours in research hours because some graduate student or postdoc was able to do their research faster and more accurately. That may not appear in my CV, but it makes me motivated to continue to […]
Ji Qi’s co-first author paper on “Atomic-scale origin of the low grain-boundary resistance in perovskite solid electrolyte Li0.375Sr0.4375Ta0.75Zr0.25O3 (LSTZ0.75)” has been published in Nature Communications! This is a highly-collaborative work under the Center for Complex and Active Materials (CCAM), an NSF MRSEC. Perovskite solid electrolytes for all-solid-state lithium-ion batteries are often plagued by grain boundary (GB) resistance. In this work, the CCAM team use aberration-corrected scanning transmission electron microscopy and spectroscopy, along with an active learning moment tensor potential, to reveal the atomic scale structure and composition of LSTZ0.75 grain boundaries. A key finding is that Li depletion in the GB is mitigated in LSTZ0.75 compared to the typical LLTO peroskite SE. Instead, a nanoscale defective cubic perovskite interfacial structure that contained abundant vacancies is formed. Ji’s contribution is the development of an accurate machine learning interatomic potential to study the highly complex LLTO and LSTZ perovskites, including the GB structures. Using MC and MC simulations, we demonstrate that Li enrichment and Sr vacancies in the GBs of LSTZ play a key role in fast diffusion in LSTZ. Check out this work here.
Jasleen’s first-author paper on “Polaron-induced metal-to-insulator transition in vanadium oxides from density functional theory calculations” is out in Physical Review B! Vanadium oxides are promising phase-change memory units for neuromorphic computing due to their metal-insulator transitions (MIT) at or near room temperature. In this work, we show that V3O5 exhibits very low hole and electron polaron migration barriers (< 100 meV) compared to V2O3 and VO2, leading to much higher estimated polaronic conductivity. The relative migration barriers are found to be related to the amount of distortion that has to travel when the polaron migrate from one site to another. Polarons in V3O5 also have smaller binding energies to vanadium and oxygen vacancy defects. These results explain recent experiment studies showing the injection of charge carriers into vanadium oxides as an alternative switching mechanism and also potentially as a means to tune the MIT temperature. Check out this work here.
Xingyu’s swansong work in the Materials Virtual Lab, “Intercalation Chemistry of the Disordered Rocksalt Li3V2O5 Anode from Cluster Expansions and Machine Learning Interatomic Potentials” has been published in Chemistry of Materials! We revisited the intercalation chemistry of the highly promising DRX-Li3V2O5 using machine learning-based computational techniques that enable much larger scale simulations. DRX Li3V2O5 is a promising anode candidate for rechargeable lithium-ion batteries because of its low voltage, high rate capability, and good cycling stability. In contrast to previous DFT studies, we show that insertion of Li primarily occurs in the tetrahedral sites and that the voltage profile depends critically on the initial Li/V disorder. MD simulations also show that DRX-Li3V2O5 has a fast Li diffusivity, which depends on the concentration of Li. We propose tuning the Li:V ratio as a means of trading off increased lithiation capacity and decreased anode voltage in this system. This work provides in-depth insights into the high-performance DRX-Li3V2O5 anode and paves the way for the discovery of other disordered anode materials. Check out the work here.
Dr Chi Chen’s swansong work in our group, “A Universal Graph Deep Learning Potential for the Periodic Table” is now published in Nature Computational Science! Interatomic potentials (IAPs), which describe the potential energy surface of atoms, are a fundamental input for atomistic simulations. However, existing IAPs are either fitted to narrow chemistries or too inaccurate for general applications. In this work, we combine graph neural networks with traditional 3-body interactions to develop a flexible, yet accurate architecture for machine learning of materials properties. Using the massive database of structural relaxations performed by the Materials Project over the past ten years, we train a universal IAP for 89 elements of the periodic table with broad applications in structural relaxation, dynamic simulations and property prediction of materials across diverse chemical spaces. Using the new capabilities of the M3GNet universal IAP, we are proud to launch matterverse.ai, a ML database of yet-to-be-synthesized materials. Matterverse.ai currently contains about 31 million hypothetical crystal structures, of which about 1.8 million materials were identified to be potentially stable. The database also provides ML properties using state-of-the-art multi-fidelity MEGNet models, such as experimental, HSE and PBE band gaps, bulk and shear moduli, etc. Check out the article here. […]