AtomSets – using graph networks as an encoder


Graph networks are an extremely powerful deep learning tool for predicting materials properties. However, a critical weakness is their reliance on large quantities of training data. In this work published in npj Computational Materials, Dr Chi Chen shows that pre-trained MEGNet formation energy models can be effectively used as “encoders” for crystals in what we call the AtomSets framework. The compositional and structural descriptors extracted from graph network deep learning models, combined with standard artificial neural network models, can achieve lower errors than the graph network models at small data limits and other non-deep-learning models at large data limits. AtomSets also transfer better in a simulated materials discovery process where the targeted materials have property values out of the training data limits, require minimal domain knowledge inputs and are free from feature engineering. Check out this work here.

Using ML to Discover New Materials


Our paper on “Accelerating materials discovery with Bayesian optimization and graph deep learning” has just been published in Materials Today! In our group, we are firm advocates of ML models that utilize structure-based features, because only such models can reliably predict property differences between chemically similar but structurally different materials (e.g., diamond vs graphite). However, a bottleneck remains in that obtaining an input structure today still depends on expensive DFT calculations. Here, we show that Bayesian optimization with an accurate MEGNet energy model can be used to obtain sufficiently good input structures for ML model predictions. We demonstrated the power of this approach by screening 400,000 materials for ultra-incompressibility. Two completely novel materials are realized experimentally by Mingde Qin in Prof Jian Luo’s group at UCSD. This work paves the way to ML-accelerated discovery of new materials with exceptional properties. Check out this work here.

Dislocation mobility in refractory high-entropy alloys

Our collaborative paper with the Ritchie and Asta groups on “Atomistic simulations of dislocation mobility in refractory high-entropy alloys (RHEAs) and the effect of chemical short-range order” has been published in Nature Communications! RHEAs are designed for high elevated-temperature strength, with both edge and screw dislocations playing an important role in plastic deformation. Using the highly accurate machine learning interatomic potential developed by MAVRL alum Dr Yunxing Zuo, we investigate mechanisms underlying the mobilities of screw and edge dislocations in the bcc MoNbTaW RHEA over a wide temperature range using MD simulations, and how these mechanisms are affected by the presence of short range order. We show that the mobility of edge dislocations is enhanced by SRO, while the rate of double-kink nucleation in the motion of screw dislocations is reduced. We also found a cross-slip locking mechanism for the motion of screws, which provides for extra strengthening for bcc RHEAs. Check out this work at this link.

AIMD prediction errors in Lithium Superionic Conductors

Congratulations to Ji Qi on his paper on “Bridging the gap between simulated and experimental ionic conductivities in lithium superionic conductors” published in Materials Today Physics. Part of a “Special Issue on Solid state batteries: materials, characterizations, and understandings”, this is Ji’s first first-author paper. Lithium superionic conductors (LSCs) are of major importance as solid electrolytes for next-generation all-solid-state lithium-ion batteries. However, ab initio molecular dynamics (AIMD) often make wrong predictions of their ionic conductivities due to the short time scales and small cell sizes used. Here, we present a strategy to bridge this gap using machine learning interatomic potentials based on the moment tensor potential (MTP) formalism. We show that the DFT functional used to train the MTPs plays a critical role in the accuracy of the predictions. In particular, the van der Waals optB88 functional yield much more accurate lattice parameters, which in turn leads to accurate prediction of ionic conductivities and activation energies for Li0.33La0.56TiO3, Li3YCl6 and Li7P3S11. Nanosecond NPT MD simulations also reveal that all three lithium superionic conductors undergo a transition between two quasi-linear Arrhenius regimes at relatively low temperatures. This transition can be traced to an increase in the number and diversity of diffusion […]

L-edge XANES database

Yiming’s paper on “Database of ab initio L-edge X-ray absorption near edge structure” has just been published in Nature Scientific Data! This work is a collaboration between the Materials Virtual Lab, the Materials Project, Alan Dozier and the groups of Prof John Rehr at the University of Washington and Prof Jordi Cabana at the University of Illinois Chicago. It is a follow-up to our earlier work on a K-edge XANES database, the L-edge XANES database provides instant access to more than 140,000 L-edge spectra for more than 22,000 structures generated using a high-throughput FEFF9 workflow. The L-edge XANES is widely used in the characterization of transition metal compounds. The data is available through the Materials Project XAS app and addresses a critical need for L-edge XANES spectra among the research community. The journal article is available at this link.

Halide Migration in Lead Halide Perovskites

Manas just published his first paper on “Correlated Octahedral Rotation and Organic Cation Reorientation Assist Halide Ion Migration in Lead Halide Perovskites” in Chemistry of Materials! Halide ion migration is one of the main contributors to instability and hysteresis in lead halide perovskite (LHP) solar cells. In this collaborative work with the Fenning group, we elucidate the effect of octahedral rotation and organic cation rotation on halide ion migration in APbBr3 (A = Cs or methylammonium/MA) LHPs. While both effects lower halide migration barriers, organic cation rotation plays a much bigger role in hybrid organic-inorganic LHPs, which can be linked to changes in H bonding during the halide migration process. We suggest that “locking” the organic cation via chemical and processing means can help mitigate halide migration-induced instability and reduced hysteresis in LHP solar cells. Check out the work at this link.

Metal-Insulator Transition in V2O3 with Intrinsic Defects


Richard’s paper on “Metal-Insulator Transition in V2O3 with Intrinsic Defects” has just been published in Physical Review B! V2O3 is a material of potential interest for neuromorphic computing, i.e., computers that mimic biological brains and have the potential to be far more efficient than traditional von Neumann architectures. A potential implementation utilizes metal insulator transitions (MITs) to implement “leaky, integrate, and fire” to emulate short-term memory. V2O3, which undergo a metal-insulator transition (MIT) at 165K, can be used to implement al for such devices as they exhibit a sudden collapse of insulating behavior under an external stimuli, and they can gradually recover their insulating state over time in the absence of the stimuli. This behavior is known as volatile resistive switching. Here, we show that the PBE + U functional provides the best compromise between accuracy and efficiency in calculating the properties related to the MIT between low-temperature and high-temperature V2O3. We use this functional to explore the various influences that intrinsic point defects will have on the MIT in V2O3. This work is a collaboration with the Schuller group at UCSD as part of the Quantum Materials for Energy Efficient Neuromorphic Computing (QMEEN-C) center, an Energy Frontier Research Center […]

Stable Cathode-Na3-xY1-xZrxCl6 Composite for High Voltage All-Solid-State Na-ion Batteries

Our collaborative work with the Meng (UCSD) and Clement (UCSB) groups on the discovery of the Na3-xY1-xZrxCl6 (NYZC) ion conductor has just been published in Nature Communications. While rechargeable solid-state sodium-ion batteries (SSSBs) promise to bring about safer and more energy-dense energy storage, the poor interfacial stability between existing solid electrolytes and typical oxide cathodes has limited their long-term cycling performance and practicality. Using DFT calculations and MD simulations with a machine learning interatomic potential, Swastika Banerjee and Ji Qi from the Materials Virtual Lab identified NYZC as a promising new ion conductor that is both electrochemically stable up to 3.8 V vs. Na/Na+ and chemically compatible with oxide cathodes. NYZC’s ionic conductivity of 6.6 × 10−5 S/cm at ambient temperature, several orders of magnitude higher than oxide coatings, is due to abundant Na vacancies and cooperative MCl6 rotation. A SSSB comprising a NaCrO2 + NYZC composite cathode, Na3PS4 electrolyte, and Na-Sn anode exhibits an exceptional first-cycle Coulombic efficiency of 97.1% at room temperature and can cycle over 1000 cycles with 89.3% capacity retention at 40 °C. Check out our article at this link.

“Liquid-like” Li sublattice in Mixed-Halide Argyrodites Li6-xPS5-xClBrx

Our collaborative work with Prof Hu’s group at Florida State University on “Tunable Lithium-Ion Transport in Mixed-Halide Argyrodites Li6-xPS5-xClBrx: An Unusual Compositional Space” has been published in Chemistry of Materials. In this work, we report a new compositional space of argyrodite superionic conductors, Li6−xPS5−xClBrx [0 ≤ x ≤ 0.8]. In particular, Li5.3PS4.3ClBr0.7 has a remarkably high ionic conductivity of 24 mS/cm at 25 °C and an extremely low lithium migration barrier of 0.155 eV that makes it highly promising for low-temperature operation. Using NMR and DFT calculations (performed by Swastika Banerjee from the Materials Virtual Lab), we show that bromination leads to co-occupancy of Cl-, Br- , and S2- at 4a/4d sites eventually resulting in a “liquid-like” Li-sublattice with a flattened energy landscape when x approaches 0.7.

Machine learning properties of ordered and disordered materials

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 and the Github repository.