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 A60: AI Materials Design and Discovery IFocus Live
|
Hide Abstracts |
Sponsoring Units: GDS DCOMP Chair: Cheng-Chien Chen, Univ of Alabama at Birmingham; Sergei Kalinin, Oak Ridge National Lab |
Monday, March 15, 2021 8:00AM - 8:36AM Live |
A60.00001: Network Theory Meets Materials Science Invited Speaker: Christopher Wolverton One of the holy grails of materials science, unlocking structure-property relationships, has largely been pursued via bottom-up investigations of how the arrangement of atoms and interatomic bonding in a material determine its macroscopic behavior. Here we consider a complementary approach, a top-down study of the organizational structure of networks of materials, based on the interaction between materials themselves. We demonstrate the utility of applying network theory to materials science in two applications: First, we unravel the complete “phase stability network of all inorganic materials” as a densely-connected complex network of 21,000 thermodynamically stable compounds (nodes) interlinked by 41 million tie-lines (edges) defining their two-phase equilibria, as computed by high-throughput density functional theory. Using the connectivity of nodes in this phase stability network, we derive a rational, data-driven metric for material reactivity, the “nobility index”, and quantitatively identify the noblest materials in nature. Second, we apply network theory to the problem of synthesizability of inorganic materials, a grand challenge for accelerating their discovery using computations. We use machine-learning of our network to predict the likelihood that hypothetical, computer generated materials will be amenable to successful experimental synthesis. ** In collaboration with V. Hegde, M. Aykol, S. Kirklin, L. Hung, S. Suram, P. Herring, and J. Hummelshoj |
Monday, March 15, 2021 8:36AM - 8:48AM Live |
A60.00002: Neural network – assisted search for active site ensembles in dilute bimetallic nanoparticle catalysts Nicholas Marcella, Steven Torrisi, Jin Soo Lim, Boris Kozinsky, Anatoly Frenkel The activity of dilute bimetallic nanoparticle catalysts depend on the gas and temperature treatment of the catalyst prior to reaction. The pretreatment drives the redistribution of the reactive dilute species between the surface, subsurface, and bulk, resulting in significant changes to the distribution of atoms. As a result these catalysts enable many reaction pathways. Current methods of reaction modeling require the assumption of an active configuration a priori, but tools for the verification of such assumptions are presently lacking. Here we present a method that combines the in situ measurements of X-ray absorption fine structure spectroscopy (XAFS) and catalytic activity, by way of neural network modeling, to create starting configurations for theoretical reaction modeling. The focus of this method is to facilitate the search for catalytically active structures that provide agreement between experimentally measured and theoretically calculated activities. We demonstrate its utility by using a series of theoretically simulated structures and their corresponding reaction pathways for hydrogen dissociation over dilute Pd-Au catalysts. |
Monday, March 15, 2021 8:48AM - 9:00AM Live |
A60.00003: Accelerating Finite-Temperature Kohn-Sham Density Functional Theory with Deep Neural Networks Attila Cangi, J. A. Ellis, Normand Arthur Modine, J. Adam Stephens, Aidan Thompson, Sivasankaran Rajamanickam We present a numerical modeling workflow based on deep neural networks that reproduce spatially-resolved, energy-resolved, and integrated quantities of Kohn-Sham density functional theory at finite electronic temperature to within chemical accuracy. We demonstrate the efficacy of this approach for both solid and liquid metals. Our machine-learning density functional theory framework opens up the path towards multiscale materials modeling for matter under ambient and extreme conditions at a computational scale and cost that is unattainable with current algorithms. |
Monday, March 15, 2021 9:00AM - 9:12AM Live |
A60.00004: Graph Neural Network for Metal Organic Framework Potential Energy Approximation: Energy Landscape Database and Rigidity Christopher Owen, Shehtab Zaman Metal Organic Frameworks (MOFs) are molecules that consist of metal ion clusters with organic ligands. Despite their widespread potential usage ranging from hydrogen storage to gas purification, MOFs are not predominately used in these sectors since they are mechanically unstable. To design these mechanical properties, a universal forcefield specific for MOFs is needed as well as a theory capable of assessing the mechanical stability given these forcefields. The forcefield for a single MOF is captured by a database of ground state energies vs atomic configurations. We have generated such a database using density functional theory(DFT) for the MOF, ‘FIGXAU’. In a second talk by Shehtab Zaman (Binghamton University), we will discuss the generation of a preliminary universal forcefield from this database. Finally, we show how a rigidity matrix formalism is ideally suited to studying the stability of MOFs and that MOFs in general are fine-tuned so that their nearest neighbor bonds, viewed as springs, are near the isostatic point in Maxwell constraint counting. We conclude with implications of our work for the design of MOFs for application purposes. |
Monday, March 15, 2021 9:12AM - 9:24AM Live |
A60.00005: Symmetry incorporated graph convolutional neural networks for solid-state materials Weiyi Gong, Hexin Bai, Peng Chu, Haibin Ling, Qimin Yan Recently, graph convolutional neural network (GCN) has been applied in crystal structures with a crystal graph representation to achieve an accurate prediction of material properties. However, graph convolutions used in previous work are mostly performed in real space based on the geometric information of crystal structures. The lack of space group symmetry information in real and reciprocal space limits the prediction accuracy of electron structure related properties. In this talk, we will demonstrate the development of a graph convolutional neural network with global and local symmetries in both real and reciprocal spaces incorporated. The newly proposed model gives accurate predictions, compared to the state-of-the-art atom-based graph neural network models, and inspiring physical insights in the correlation between orbital symmetries and electronic structure properties of solid-state crystalline systems. |
Monday, March 15, 2021 9:24AM - 9:36AM Live |
A60.00006: CCDCGAN: Inverse design of crystal structures Teng Long, Nuno Fortunato, Yixuan Zhang, Chen Shen, Oliver Gutfleisch, Hongbin Zhang Autonomous materials discovery with desired properties is one of the ultimate goals of materials science. We have developed constrained crystal deep convolutional generative adversarial networks (CCDCGAN), which can be used to design unreported (meta-)stable crystal structures using encoded 2D latent space.1 Such a latent space can also be applied to forwardly predict physical properties like the formation energy. Correspondingly, it is demonstrated that the optimization of physical properties in the latent space can be integrated into the generative model as on-top screening or backwards propagator, both with their own advantages. The CCDCGAN has been successfully applied on a specific binary (i.e., Bi-Se) system and multicomponent systems (i.e., all binary and ternary compounds in the Materials Project database). It is observed that the crystal structures distinct from the known cases can be obtained covering the whole composition range. We suspect that CCDCGAN can be extended to multi-objective optimization, such as band gap and mechanic properties, which paves the way to achieve the inverse design of crystalline materials with optimal properties. |
Monday, March 15, 2021 9:36AM - 9:48AM Live |
A60.00007: Network-based representation and analysis of materials space Alexander Veremyev, Laalitha Liyanage, Marco Fornari, Vladimir Boginski, Stefano Curtarolo, Sergiy Butenko, Marco Buongiorno Nardelli Modeling and analysis of the materials universe is an emerging area of research with many important applications in materials science. The main goal is to create a map of materials which allows not only to visualize and navigate the materials space, but also reveal complex relationships and "connections" among materials and potentially find clusters of materials with similar properties. In this talk, we consider the problem of mapping and exploring the materials universe using network science tools and concepts. The networks are based on the open-source materials data repository AFLOW.org where each material is represented as a node, and each pair of nodes is connected by a link if the respective materials exhibit a high level of similarity between their Density of States (DOS) functions. We discuss the importance of similarity measure selection, investigate basic structural properties of the resulting networks, and demonstrate advantages and limitations of the proposed approaches. |
Monday, March 15, 2021 9:48AM - 10:00AM Live |
A60.00008: Uncovering the Relationship Between Thermal Conductivity and Anharmonicity with Symbolic Regression Thomas Alexander Reichmanis Purcell, Matthias Scheffler, Luca M. Ghiringhelli, Christian Carbogno Quantitatively understanding the link between anharmonicity and thermal conductivity, κ, is pivotal to the search for better thermal insulators. While it is qualitatively known that more anharmonic materials have a lower κ, until recently, no quantitative measure of anharmonicity existed. Here we present descriptors of κ based on our new measure of anharmonicity, σA [1]. We find the analytical expressions with symbolic regression, via the sure-independence screening and sparsifying operator (SISSO) method [2]. To better capture the nonlinearities in the correlation between κ and σA, we introduce an automatic scaling and shifting of the input data when generating new features like exp(α x + a). Using our new strategy, we generate expressions that are competitive with those previously reported in the literature using only a third of primary the features [3], and reduce the models test error by a third when compared to traditional SISSO. Finally, we discuss the implications of the new models on future materials design. |
Monday, March 15, 2021 10:00AM - 10:12AM Live |
A60.00009: Enhanced Machine Learning Models for Structure-Property Mapping with Principal Covariates Regression Rose K. Cersonsky, Benjamin A. Helfrecht, Guillaume Fraux, Edgar Engel, Michele Ceriotti Data analyses based on linear methods constitute the simplest, most robust, and transparent approaches to the automatic processing of large amounts of data for building supervised or unsupervised machine learning models. Principal covariates regression (PCovR) is an underappreciated method that interpolates between principal component analysis and linear regression, and can be used to conveniently reveal structure-property relations in terms of simple-to-interpret, low-dimensional maps. Here we introduce a kernelized version of PCovR and demonstrate the performance of this approach in revealing and predicting structure-property relations in chemistry and materials science. Additionally, we demonstrate the improved performance resulting from incorporating PCovR into two popular data selection methodologies, CUR and Farthest Point Sampling, which iteratively identify the most diverse samples and discriminating features. |
Monday, March 15, 2021 10:12AM - 10:24AM Live |
A60.00010: Graph Neural Network for Metal-Organic Framework Potential Energy Approximation Shehtab Zaman, Christopher Owen, Kenneth Chiu, Michael Lawler Metal-organic frameworks (MOFs) are nanoporous compounds composed of metal ions and organic linkers. Due to the flexibility of combining hundreds of organic ligands with tens of metal ions in thousands of network geometries, the configuration space of possible MOFs is almost infinite. The mechanical properties of MOFs can be tuned to produce desirable characteristics, so rapidly quantifying the properties is key. The potential energy is a fundamental calculation needed to design MOFs for many applications. The potential energy is currently computed via techniques such as density functional theory (DFT). Although DFT provides accurate results, it is computationally very costly. We propose a machine learning approach for estimating the potential energy of candidate MOFs, decomposing it into separate pair-wise atomic interactions using a graph neural network. Our modified graph convolutional neural network predicts energies of MOFs and learns bond level contributions to the energy. |
Monday, March 15, 2021 10:24AM - 10:36AM Live |
A60.00011: Towards Inverse Design of Metal-Organic Frameworks to Maximize Hydrogen Storage using Deep Learning Kevin Phillips, Shehtab Zaman, Kenneth Chiu, Michael Lawler Metal-organic frameworks (MOFs) are a class of crystalline porous materials consisting of metal nodes and organic linkers. MOFs have applications in gas separation, gas purification, and electrolytic catalysis, among other fields. Consequently, the creation of better MOFs for these purposes represents a multibillion-dollar engineering challenge. Using machine learning can help exponentially accelerate the research and discovery of suitable MOFs for these applications. We implement Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) that utilize scaled-down voxel representations of real MOFs for the inverse design of new MOFs with maximal hydrogen adsorption. High hydrogen adsorption MOFs offer a potentially safe and efficient storage method of hydrogen gas for use in fuel cells. This could be critical to environmentally friendly transportation or UAVs industries as well useful to scientists studying materials science, machine learning, and their intersection |
Monday, March 15, 2021 10:36AM - 10:48AM Live |
A60.00012: Predicting geometric properties of metal-organic frameworks by fusing 3D and graph convolutional neural networks Jacob Barkovitch, Musen Zhou, Shehtab Zaman, Kenneth Chiu, Michael Lawler, Jianzhong Wu Metal-organic frameworks (MOFs) have emerged in recent years as a substantial class of crystalline structures with extremely high porosity, inner surface area, and variability of the organic and inorganic components. Calculating geometric properties of MOFs is done through Monte Carlo simulations which are both time-consuming and tedious. Fast and accurate prediction of these is a first step to enabling the synthesis of new and novel structures. We propose a fusion model that combines a 3D convolutional neural network and a graph convolutional neural network to predict geometric properties of MOFs such as Henry’s constant, surface area, pore limiting diameter, and largest cavity diameter. The model utilizes both 3D grid and graph-structured representations of MOFs to predict the geometric properties. We used the CoRE MOF 2019 dataset with expanded geometric properties such as Henry's constant and surface area. Our model quickly predicts the geometric properties of MOFs and will aid in the high-throughput characterization of MOFs. |
Monday, March 15, 2021 10:48AM - 11:00AM Live |
A60.00013: Generating Multiscale Amorphous Molecular Structures Using Deep Learning: A Study in 2D Lena Simine, Michael Kilgour, Nicolas Gastellu, David Yu-Tung Hui, Yoshua Bengio Amorphous molecular assemblies appear in a vast array of systems: from living cells to chemical plants and from everyday items to new devices. The absence of long-range order in amorphous materials implies that precise knowledge of their underlying structures throughout is needed to rationalize and control their properties at the mesoscale. Standard computational simulations suffer from exponentially unfavorable scaling of the required compute with system size. We present a method based on deep learning that leverages the finite range of structural correlations for an autoregressive generation of disordered molecular aggregates up to arbitrary size from small-scale computational or experimental samples. We benchmark performance on self-assembled nanoparticle aggregates and proceed to simulate monolayer amorphous carbon with atomistic resolution. This method bridges the gap between the nanoscale and mesoscale simulations of amorphous molecular systems. |
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