Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session A53: AI and Materials I |
Hide Abstracts |
Sponsoring Units: GDS DMP Chair: Rama Vasudervan, Oak Ridge National Lab Room: Room 307 |
Monday, March 6, 2023 8:00AM - 8:12AM |
A53.00001: Screening the unexplored crystal prototype space and inverting XRD patterns with the WREN machine-learning model Rickard Armiento, Abhijith S Parackal, Rhys Goodall, Felix A Faber, Alpha A Lee The WREN (Wyckoff REpresentation regressioN) machine-learning model trained on 300k formation energies across the full chemical space allows near-instant prediction of formation energies of materials just from their element-assigned crystal prototypes (expressed in terms of Wyckoff positions) [1]. This model allows screening for materials with desired properties among structures fundamentally different from those presently catalogued in materials databases. This talk presents the WREN model and demonstrates our recent progress in using it to invert XRD patterns. Our highly efficient implementation enumerates candidate prototypes, uses WREN to order them by formation energy, and then optimizes the remaining degrees of freedom to match the XRD peaks. The approach is shown capable of resolving previously unresolved XRD patterns in the ICDD database. |
Monday, March 6, 2023 8:12AM - 8:24AM |
A53.00002: Semi and Self Supervised approaches to Space Group and Bravais Lattice Determination William Ratcliff, Satvik S Lolla, Ichiro Takeuchi, Aaron Kusne, Haotong Liang During this talk, I will discuss our work [1] to use neural networks to automatically classifiy Bravais lattices and space-groups from neutron powder diffraction data. Our work classifies 14 Bravais lattices and 144 space groups. The novelty of our approach is to use semi-supervised and self-supervised learning to allow for training on data sets with unlabelled data as is common at user facilities. We achieve state of the art results with a semi-supervised approach. Our accuracy for our self-supervised training is comparable to that with a supervised approach. |
Monday, March 6, 2023 8:24AM - 8:36AM Author not Attending |
A53.00003: Machine Learning the Electronic Structure of Phase Change Materials Qunfei Zhou, Suvo Banik, Srilok Srinivasan, Subramanian K Sankaranarayanan, Pierre Darancet Machine learning is becoming a powerful tool to complement current computational techniques in solving materials science problems. While first-principles calculations based on Density Functional Theory (DFT) have demonstrated the numerical accuracy required for most nanoelectronics applications, amorphous and polycrystalline systems represent a particular challenge due to their heterogeneity and the associated size of their structural approximants. Tight-binding methods offer the scalability required, but rely on their prior parametrization, a complex tax for multivalent and phase changing materials used in Beyond-Moore computing. In this work, we use machine learning to parameterize a tight-binding ansatz for the electronic structure of complex phase change materials. Using the DFT results for small unit cells of single and multivalent GexSbyTez(0 |
Monday, March 6, 2023 8:36AM - 8:48AM |
A53.00004: Data-driven studies of topological magnetic vdW materials Romakanta Bhattarai, Peter Minch, Trevor David Rhone First-principles calculations and machine learning techniques are used to investigate monolayers of the form AB2X4 based on the well-known intrinsic topological magnetic vdW material MnBi2Te4. In this study, we consider a very large number of candidate materials (~104) formed by tuning the chemical composition of AB2X4. Investigating this enormous number of candidates by first-principles calculations or experiments is prohibitive. The use of machine learning is a promising way to efficiently explore the entire chemical space thereby accelerating materials discovery. We select an initial subset of 240 structures for investigation using density functional theory (DFT). We calculate the thermodynamic properties, electronic properties, such as the band gap, and magnetic properties, such as the magnetic moment, magnetic order, and the exchange energy. Next, we train a machine learning model to successfully make predictions of various properties that will be useful to accelerate the exploration of the entire chemical space. Our analysis shows that the formation energy and the magnetic moment of the system depend largely on A and B sites, whereas the bandgap depends on all three sites. This study creates avenues for discovering novel materials with desirable properties that are crucial for spintronics, optoelectronics, quantum computing, and quantum communication. |
Monday, March 6, 2023 8:48AM - 9:00AM |
A53.00005: Data-driven Study of Magnetic Anisotropy in Transition Metal Dichalcogenide Monolayers Peter Minch, Romakanta Bhattarai, Trevor David Rhone We investigate the magnetic and thermodynamic properties of transition metal dichalcogenides of the form AX2, based on monolayer MnSe2 using data analytics. That is, we combine first-principles calculations with machine learning methods to elucidate the microscopic origins of the magnetocrystalline anisotropy in these materials. We explore a large number of candidate transition metal dichalcogenides by varying the chemical compositions of the transition metal (A) and chalcogen (X) sites. We investigate the transition between in-plane and out-of-plane magnetization in addition to the magnetocrystalline anisotropy. We demonstrate that the interplay between the spin-orbit interactions of the chalcogen and transition metal atoms gives rise to a diverse array of magnetic behavior. |
Monday, March 6, 2023 9:00AM - 9:12AM |
A53.00006: Using chemical-formula-based generalizable models to expand the search space for viable interconnect materials Akash Ramdas, Evan J Reed, Felipe H da Jornada The identification of viable interconnect materials with lower effective resistivity is critical in ensuring the further scaling of transistor CMOS technology. Previously, we curated a set of key metrics for over 15,000 previously synthesized systems, obtained from the Materials Project, to identify compounds that are promising metallic bulk conductors. Using this approach, we identify around 20 systems that can potentially outperform traditional interconnect materials such as Cu or Ru. Our selection criteria involves a multi-objective optimization of Fermi velocities, scalability and stability. For our stability metrics, we use the energy with respect to the convex hull, and 0K thermodynamic reaction energies with air, water and SiO2. . To identify more viable candidate materials, we train classification models based on the chemical formulas of these 15,000 systems to predict those with large Fermi velocity and stability metrics. We apply this approach to over 12,039 charge-balanced binary formulas (AxBy) and identify over 500 previously unconsidered chemical formulas with the potential to outperform Ru or Cu interconnects, and validate some of our entries with first-principles transport calculations. Our approach can be broadly utilized to identify suitable materials for specific transport applications. |
Monday, March 6, 2023 9:12AM - 9:24AM |
A53.00007: How to Search for Stable Inorganic Compounds More Efficiently Sean D Griesemer, Ruijie Zhu, Koushik Pal, Cheol Park, Logan Ward, Christopher M Wolverton The computational search for new stable inorganic compounds remains highly expensive, due to the combinatorically explosive number of hypothetical compounds to consider. To guide the search towards the most likely stable compounds, several recommendation engines have been developed, varying in their strategy from data mining to machine learning. We conduct a systematic comparison of the performance of previously developed recommendation engines in recovering stable hypothetical compounds in the Open Quantum Materials Database (OQMD), and develop workflows to execute these methods in a highly efficient manner. For example, we find that crystal graph convolution neural networks outperform methods based on substitution of chemically similar elements into existing compounds; that employing a feedback loop (where method parameters are periodically retrained during execution) greatly improves the performance of recommendation engines in identifying already-calculated stable Heusler compounds; and that design of training set is crucial for the performance of neural networks. We also examine the status of stable compound searches in OQMD and find that, while thousands of non-experimentally-known stable compounds have already been identified, there are evidently at least thousands more that remain to be found, and hence the quest for materials discovery is far from over. |
Monday, March 6, 2023 9:24AM - 9:36AM |
A53.00008: Integrating Machine Learning with Mechanistic Models for Predicting the Yield Strength of High Entropy Alloys Shunshun Liu, Kyungtae Lee, Prasanna V Balachandran Accelerating the design of materials with targeted properties is one of the key materials informatics tasks. The most common approach takes a data-driven motivation, where the underlying knowledge is incorporated in the form of domain-inspired input features. Machine learning (ML) models are then built to establish the input-output relationships. An alternative approach involves leveraging mechanistic models, where the domain knowledge is incorporated in a predefined functional form. In this work, we demonstrate a computational approach that integrates mechanistic models with phenomenological and ML models to rapidly predict the temperature-dependent yield strength of high entropy alloys (HEAs) that form in the single-phase face-centered cubic (FCC) structure. This allows us to improve the treatment of elastic constant mismatch to the solid solution strengthening effect in the mechanistic model, which is important for the reliable prediction of yield strength as a function of temperature in single-phase FCC-based HEAs. We accomplish this by combining Bayesian inference with ensemble ML methods. The outcome is a probability distribution of elastic constants which, when propagated through the mechanistic model, yields a prediction of temperature-dependent yield strength, along with the uncertainties. |
Monday, March 6, 2023 9:36AM - 9:48AM |
A53.00009: Statistics on the magnetism of cobalt compounds: A database approach to discovering new Co-based ferromagnets Journey K Byland, Yunshu Shi, David S Parker, Jingtai Zhao, Shaoqing Ding, Rogelio Mata, Haley E Magliari, Andriy Palasyuk, Sergey L Bud'ko, Paul C Canfield, Peter Klavins, Valentin Taufour Discovery of novel ferromagnetic materials can often be hampered by the lack of a comprehensive repository of magnetic data, and it can be easy to overlook the less well-known families of compounds. We propose a method to facilitate the search for novel ferromagnets by creating a database of cobalt-based compounds. Based on a comprehensive literature survey of the magnetic properties of compounds with more than 33 at. % Co, we classify more than 13,000 compounds by structure types, cobalt content, ordering type and temperature, and easy magnetization direction. We focus our analysis on data trends of Curie temperatures, structure families favoring ferromagnetism, and uniaxial anisotropy. We identify compounds TaCo2Ga, La6Co13Bi, and Nd2Co3 as potential ferromagnets, and use first-principles calculations and experimental synthesis and measurements to confirm their magnetic properties. From the subset of known ferromagnets with unknown anisotropy, we select the compound MgCo2 as potentially having uniaxial magnetic anisotropy at room temperature, which we also confirm experimentally. |
Monday, March 6, 2023 9:48AM - 10:00AM |
A53.00010: Navigating materials design space with variational autoencoders to learn materials thermodynamics Vahe Gharakhanyan, Dallas R Trinkle, Snigdhansu Chatterjee, Alexander Urban Recent advances in artificial intelligence for materials design and discovery target the screening of entire libraries of materials for desirable properties as well as the prediction of novel materials for target properties. A particular challenge has been the materials design for thermodynamic properties far away from zero Kelvin and ambient pressure because of the lack of public thermodynamic data. |
Monday, March 6, 2023 10:00AM - 10:12AM |
A53.00011: Automatic, physical data extraction from scientific publications for application to generative molecular design in computational materials discovery Ronaldo Giro, Mohab Elkaref, Hsianghan Hsu, Nathan Herr, Geeth de Mel, Mathias B Steiner One of the major barriers for the application of artificial intelligence (AI) in materials design and discovery is the lack of training data for machine-learning models. Despite the recent emergence of public data repositories in materials sciences, the data formats are not standardized and searchability of application specific data sets is limited. This contrasts with the vast amounts of structured data tables available in published papers nowadays. In this contribution, we will present a method and research tool that allows the annotation and automatic extraction of physical and chemical data tables from document files. The necessary configuration steps include: (i) defining a corpus of papers which are relevant to the discovery application of interest; (ii) reviewing and selecting the extracted tables and converting the files, and (iii) transforming the materials’ names into a machine-readable string format. With the above steps completed, we obtain an integrated data table with materials properties that is used for training the AI models. In our research, we have used the above method to collect about 500 data entries with the following polymer properties: CO2 permeability and CO2/N2 selectivity. Currently, the amount of data entries we have extracted is limited by the number of documents in the corpus. Finally, we discuss our initial results obtained with AI models trained on the extracted data tables for designing high-performance membranes for carbon dioxide capture and separation. |
Monday, March 6, 2023 10:12AM - 10:24AM |
A53.00012: Machine Learned Synthesizability Predictions Aided by Density Functional Theory Andrew Lee, Suchismita Sarker, James E Saal, Logan Ward, Christopher Borg, Apurva Mehta, Christopher M Wolverton Accurately predicting a material's synthesizability remains a grand challenge in materials science. From early heuristics like Pauling's Rules to density functional theory (DFT) calculations, there are a wide variety of approaches to solving this challenge. Machine learning and data-driven approaches have recently made significant progress, yet some works do not account for phase stability. Here, we demonstrate that stability calculated from DFT plays a crucial role in enabling a machine learning model to accurately predict half-Heusler synthesizability. Our model takes ternary 1:1:1 compositions and predicts synthesizabilities in the half-Heusler structure, achieving a precision of 0.82 and recall of 0.82. Our model identifies 121 synthesizable candidates out of 4141 unreported compositions. 39 stable compositions are predicted unsynthesizable while 62 unstable compositions are predicted synthesizable; these findings otherwise cannot be made using DFT alone. |
Monday, March 6, 2023 10:24AM - 10:36AM |
A53.00013: Hypothesis-driven active learning over the chemical space Ayana Ghosh, Sergei V Kalinin, Maxim Ziatdinov From applications in identifying potential drug targets to designing electronics, catalysts, photovoltaics and chemical reactions, efforts to discover molecular candidates has risen steeply over the years. The rapid exploration of chemical space targeting desired functionalities is performed by high-throughput screening combined with computational simulations and synthesis. Here, we introduce a novel approach for active learning of a wide chemical space based on hypothesis learning. The study is conducted on ~130,000 molecules present in the QM9 dataset to actively learn about formation enthalpy of all molecules. We construct multiple hypotheses based on the possible relationships between structures and functionalities of interest and introduce these as mean functions for Gaussian Process. This approach then combines the elements from the symbolic regression methods such as SISSO and Bayesian Optimization in a single framework. Although demonstrated for the QM9 dataset, this method is expected to be universally applicable for other datasets containing information on molecules to solid-state materials. |
Monday, March 6, 2023 10:36AM - 10:48AM |
A53.00014: Artificial intelligence guided materials discovery of van der Waals magnets Trevor David Rhone, Bethany A Lusch, Misha Salim, Haralambos Gavras, Vaishnavi Neema, Daniel T Larson, Efthimios Kaxiras The discovery of van der Waals (vdW) materials with intrinsic magnetic order in 2017 has given rise to new avenues for the study of emergent phenomena in two dimensions. In particular, monolayer CrI3 was found to be ferromagnet. Other vdW transition metal halides were later found to have different magnetic properties. How many vdW magnetic materials exist in nature? What are their properties? How do these properties change with the number of layers? A conservative estimate for the number of candidate vdW materials (including monolayers, bilayers and trilayers) exceeds ~106. A recent study showed that artificial intelligence (AI) can be harnessed to discover new vdW Heisenberg ferromagnets based on Cr2Ge2Te6 [1]. In this talk, we will harness AI to efficiently explore the large chemical space of vdW transition metal halides and to guide the discovery of magnetic vdW materials with desirable spin properties. That is, we investigate crystal structures based on monolayer Cr2I6 of the form A2X6, which are studied using density functional theory (DFT) calculations and AI. Magnetic properties, such as the magnetic moment are determined. The formation energy is also calculated and used as a proxy for the chemical stability. We show that AI, combined with DFT, can provide a computationally efficient means to predict the thermodynamic and magnetic properties of vdW materials. This study paves the way for the rapid discovery of chemically stable magnetic vdW materials with applications in spintronics and data storage. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700