Bulletin of the American Physical Society
APS March Meeting 2021
Volume 66, Number 1
Monday–Friday, March 15–19, 2021; Virtual; Time Zone: Central Daylight Time, USA
Session E60: AI Materials Design and Discovery IIIFocus Live
|
Hide Abstracts |
Sponsoring Units: GDS DCOMP Chair: William Ratcliff, NIST; Cheng-Chien Chen, University of Alabama at Birmingham |
Tuesday, March 16, 2021 8:00AM - 8:36AM Live |
E60.00001: Capturing and Leveraging Computational and Experimental Data in Materials Physics Invited Speaker: Maria Chan In order to use artificial intelligence and machine learning for scientific advances, access to complex, multimodal, and accurate data is critical. In this talk, we will discuss efforts in generating, capturing, and leveraging computational and experimental data, with examples in generation of computational defect properties datasets, capturing microscopy data, and combining streams of computational and experimental data. We will also discuss concerted efforts at US Department of Energy Scientific User Facilities in data infrastructure. |
Tuesday, March 16, 2021 8:36AM - 8:48AM Live |
E60.00002: Physics-Informed Data-Driven Approach for Optimizing Electrocaloric Cooling Jie Gong, Rohan Mehta, Alan McGaughey Electrocaloric (EC) cooling offers great potential to build efficient solid-state cooling devices that are quiet, low weight, and compact. However, the physics behind the EC temperature change are elusive. The current search for effective EC materials relies on the instincts of experts and brute force experimental synthesis of ceramics, polymers, and composites. There is no robust physical model to predict the EC temperature change based on material properties. |
Tuesday, March 16, 2021 8:48AM - 9:00AM Live |
E60.00003: First-Principles Prediction of Substrate Induced Changes in Layered Nanomaterials via Physics-Based Machine Learning Sanghamitra Neogi, Artem Pimachev Advances in nanofabrication techniques offer remarkable control over epitaxial growth of atomically thin 2-dimensional (2D) semiconductors. However, the growing substrates could strongly influence their optical, electrical, mechanical, and chemical properties. First-principles studies revealed that substrate induced strain, in particular, can tune the electronic transport properties of layered materials, offering routes to modulate properties by strain engineering. It remains a challenge to model electronic properties of layered nanomaterials scanning the vast fabrication dependent parameter space. It is highly desirable to formulate a method that is capable to learn the information from high-cost calculations and predict propertie of a wide range of configurations. We use a machine learning model to capture the relationships between local atomic environments and electronic properties of small test systems and apply the model to predict the interface states and band dispersions of 2D semiconductors grown on different substrates. Our prediction is validated against measurements obtained from high precision techniques such as angle-resolved photoemission spectroscopy. |
Tuesday, March 16, 2021 9:00AM - 9:12AM Live |
E60.00004: Featureless adaptive optimization accelerates functional electronic materials design Yiqun Wang, James M Rondinelli Electronic materials exhibiting multiple phase transitions between metastable states with distinct physical properties are challenging to decoding using conventional machine learning methods owing to data scarcity and absence of physically meaningful materials descriptors. We demonstrate a discovery strategy based on multi-objective Bayesian optimization to directly circumvent these bottlenecks by utilizing latent variable Gaussian processes combined with high-fidelity electronic structure calculations for validation in the chalcogenide lacunar spinel family. We directly and simultaneously learn phase stability and bandgap tunability from chemical composition alone to efficiently discover all superior compositions on the design Pareto front. Previously unidentified electronic transitions also emerge from our featureless adaptive optimization engine. Our methodology readily generalizes to optimization of multiple properties, enabling co-design of complex multifunctional materials, especially where prior data is sparse. |
Tuesday, March 16, 2021 9:12AM - 9:24AM Live |
E60.00005: Benchmarking Coordination Number Prediction Algorithms on Inorganic Crystal Structures Hillary Pan, Alex Ganose, Matthew Horton, Muratahan Aykol, Kristin Persson, Nils E.R. Zimmermann, Anubhav Jain Coordination numbers play a fundamental role in describing materials and conceptualizing their properties. On a large scale, algorithms to determine coordination numbers are useful for applications in machine learning and automatic structure analysis. We have developed a benchmarking framework called MaterialsCoord, an open-source software package for comparing algorithms on how well they determine coordination environments as described in the literature. A total of eight algorithms—seven of which are well-established and a novel algorithm, CrystalNN—are benchmarked on a diverse set of prototypical crystal structures. Apart from performance on the benchmark, we provide other analyses that may be important for implementation of these algorithms such as computational demand and sensitivity towards small perturbations that mimic thermal motion. |
Tuesday, March 16, 2021 9:24AM - 9:36AM Live |
E60.00006: Prediction of atomization energies using entropic data representation and machine learning Michael De La Rosa, Jorge Munoz Calculations of the atomization energies of molecules can be computationally expensive, but machine learning techniques have been used as an effective and accurate method for predicting these energies using the positions and charges of atoms within the molecule as features. This information is encoded in the Coulomb matrix, but the disparity in the number of atoms and lack of a well-defined ordering system means that it is necessary to use another method of data representation to apply machine learning methods effectively. Previous methods include an eigenspectrum representation, sorting the Coulomb matrices, or using randomly sorted Coulomb matrices (Hansen et al., 2013). We introduce a new method of data representation using a novel information entropy metric that is unaffected by the size or order of the Coulomb matrix. We tested this approach with the QM7 dataset which includes structural information and atomization energies of 7165 molecules. A raw application of our representation produces a correlation between the atomization energy and the graph information entropy of up to 0.97, and predictions close to state-of-the-art are achieved based on other statistical metrics when combined with well-established learning algorithms such as neural networks and k-nearest neighbors. |
Tuesday, March 16, 2021 9:36AM - 9:48AM Live |
E60.00007: Highly Accurate Machine Learning Point Group Classifier for Crystals Abdulmohsen Alsaui, Saad Alqahtani, Faisal Mumtaz, Ibrahim Alsayoud, Mohammed Al Ghadeer, Ali Muqaibel, Sergey Rashkeev, Ahmer Baloch, Fahhad Alharbi Inspired by the remarkable ongoing progress of the data-driven science approach, a predictive model is developed for the crystallographic point group classification of the ternary compounds using machine learning. In this work, the first step is to generate a space of all possible ternary compounds based on the common and uncommon oxidation states of 77 elements. The total number of possible elemental combinations has surpassed 600 million materials. The structures of more than 10 million of these materials were obtained from the NOMAD material database. Finally and starting only from the chemical formula, the elemental properties are utilized to develop an accurate predictive model for the crystallographic point group classification. The average balanced accuracy of the predictive model has exceeded 90%. The success of this work will contribute effectively to the advancement of materials discovery. |
Tuesday, March 16, 2021 9:48AM - 10:00AM Live |
E60.00008: CRYSPNet: Machine Learning Tool for Crystal Structure Predictions haotong liang, Valentin Stanev, Aaron Kusne, Ichiro Takeuchi Structure is the most basic and important property of crystalline solids; it determines directly or indirectly most other materials characteristics. However, predicting the crystal structure of solids remains a formidable problem; standard theoretical tools for the task are computationally expensive and not always reliable. As an alternative, we developed a tool, CRYSPNet, that can predict the Bravais lattice, space group, and lattice parameters of a material based on its chemical formula. It consists of a bag of neural network models with predictors based on aggregate features of the elements constituting the compound. The tool was trained and validated on more than 100,000 entries from the Inorganic Crystal Structure Database (ICSD). It demonstrates good predictive power and significantly outperforms naive strategies. CRYSPNet is easy to use and can be combined with tools for generating Wyckoff positions to create candidate structures for further exploration or refinement. Furthermore, activations from the hidden layers of the model can be used to measure the chemical and structural similarity between materials, which in turn can be used for predictions of materials with new functionalities. |
Tuesday, March 16, 2021 10:00AM - 10:12AM Live |
E60.00009: Machine learning materials properties for small datasets Pierre-Paul De Breuck, Geoffroy Hautier, Gian-Marco Rignanese In order to make accurate predictions of material properties, current machine-learning approaches generally require large amounts of data, which are often not available in practice. In this work, a novel all-round framework is presented which relies on a feedforward neural network and the selection of physically-meaningful features. Next to being faster in terms of training time, this approach is shown to outperform current graph-network models on small datasets. In particular, the vibrational entropy at 305K of crystals is predicted with a mean absolute test error of 0.009 meV/K/atom (four times lower than previous studies). Furthermore, the proposed framework enables the prediction of multiple properties, such as temperature functions, by using joint-transfer learning. Finally, the selection algorithm highlights the most important features and thus helps understanding the underlying physics. |
Tuesday, March 16, 2021 10:12AM - 10:24AM Live |
E60.00010: Identifying "materials genes" by symbolic regression: The hierarchical SISSO approach Lucas Foppa, Thomas Alexander Reichmanis Purcell, Sergey V. Levchenko, Matthias Scheffler, Luca M. Ghiringhelli A key goal in materials science is the identification of interpretable and basic physical features that correlate with materials properties and functions. These correlations reflect the actuators, i.e. the facilitators or obstructors of the different governing processes, for a given property, and have thus been referred to as "materials genes". Here, we illustrate how to find these "materials genes", even in the limit of small datasets, using symbolic regression [1]. In particular, we discuss a new strategy for discovering more complicated relationships between the features and properties by exploiting the learning of simpler, but related properties: the hierarchical sure-independence screening and sparsifying operator (hiSISSO) approach. We demonstrate this strategy by using models for the lattice constants of ABO3 perovskites to learn their bulk moduli. We show that the hierarchical approach not only outperforms traditional machine-learning methods when trained on small datasets, but also provides exploitable models, which are suitable for materials optimization and design. |
Tuesday, March 16, 2021 10:24AM - 10:36AM Live |
E60.00011: A massive dataset of synthesis-friendly hypothetical polymers Arunkumar Rajan, Chiho Kim, Christopher Kuenneth, Deepak Kamal, Rishi Gurnani, Rohit Batra, Rampi Ramprasad Polymer informatics is an emerging field in materials science. It aims to build data-driven models to instantaneously predict the properties of polymers, and use this capability to screen a large candidate set of polymers to identify promising ones based on their predicted properties. However, it is important for this candidate set to include synthesizable polymers. By utilizing ~13k experimentally known polymers, we identified two distinct pathways to generate a dataset of synthesis-friendly hypothetical polymers. These pathways comprise a combinatorial assembly of retrosynthetic fragments obtained from the ~13k polymers, and a framework that treats polymers are graphs followed by graph-to-graph translations. This has resulted in a massive dataset of 100 million hypothetical but synthesis-friendly polymers. Additionally, we quantify the synthetic feasibility of each polymer as a score and demonstrate that a large portion of the generated polymers are synthesis-ready. This massive database can be used (1) for direct screening purposes using available property prediction models, and (2) within unsupervised approaches to train of generative models to enable and accelerate polymer discovery. |
Tuesday, March 16, 2021 10:36AM - 10:48AM Live |
E60.00012: Bayesian Optimization Approach for Discovery of High-Capacity Small-Molecule Adsorption in Metal-Organic Frameworks Eric Taw, Jeffrey Neaton Metal-organic frameworks, due to their highly porous structures, have emerged as a promising class of small-molecule adsorbent materials for a variety of separations, storage, and usage applications. While it is possible to construct viable hypothetical MOFs (hMOFs) from known metal nodes and organic linkers, it is computationally expensive to calculate at a high accuracy the small molecule uptake capacity of MOF structures. Using ~51,000 hypothetical MOF structures and data calculated from [1] for CH4, we show it is possible to identify candidates for high-performance CH4 adsorbents by calculating uptake capacities for <1% of the database using Bayesian optimization. Furthermore, we show that building chemical intuition into the surrogate model and including structural characteristics dramatically improves the performance and interpretability of the optimization process. The applicability of our Bayesian approach and workflow to molecules beyond CH4 and other hypothetical adsorbents is discussed. |
Tuesday, March 16, 2021 10:48AM - 11:00AM Live |
E60.00013: Data-driven studies of the magnetic anisotropy of two-dimensional magnetic materials Yiqi Xie, Trevor Rhone, Georgios Tritsaris, Oscar Grånäs, Efthimios Kaxiras Two-dimensional materials with intrinsic ferromagnetic order are at the forefront of condensed matter research. How many of these materials exist in nature? What is the relationship between thier crystal structure and magnetic properties? Remarkably, atomically-thin magnetic structures can exhibit novel spin properties which do not exist in the corresponding bulk materials. We use first-principles calculations, based on density functional theory, and machine learning to study the magnetocrystalline anisotropy of monolayer transition metal trichalcogenides of the form A2B2X6. That is, we created permutations of the chemical composition of the ferromagnetic semiconductor Cr2Ge2Te6. Specifically, we identify trends in their magnetocrystalline anisotropy data. We find that the X site dominates the machine learning prediction of the magnetocrystalline anisotropy of an A2B2X6 monolayer. Our data-driven study aims to uncover physical insights into the microscopic origins of magnetism in reduced dimensions and to demonstrate the success of a high-throughput computational approach for the targeted design of quantum materials with potential applications from sensing to 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