Bulletin of the American Physical Society
APS March Meeting 2022
Volume 67, Number 3
Monday–Friday, March 14–18, 2022; Chicago
Session T32: Material Science and Machine Learning IRecordings Available
|
Hide Abstracts |
Sponsoring Units: GDS Chair: William Ratcliff, GDS Room: McCormick Place W-192B |
Thursday, March 17, 2022 11:30AM - 11:42AM |
T32.00001: Deep reinforcement learning for predicting kinetic pathways to surface reconstruction in a ternary alloy Zhonglin Cao, junwoon Yoon, Rajesh Raju, Yuyang Wang, Robert Burnley, Andrew Gellman, Amir Barati Farimani, Zachary Ulissi The screening of material components and alloy composition to optimize selectivity and activity for a given reaction is a major focus of the computational catalyst design. However, predicting the metastability of the alloy catalyst surface at realistic operating conditions requires an extensive sampling of possible surface reconstructions and their associated kinetic pathways. In this work, we propose CatGym, a deep reinforcement learning (DRL) environment for the prediction of the thermal surface reconstruction pathways and their associated kinetic barriers in crystalline solids under reaction conditions. Within the CatGym environment enables the DRL agent to iteratively changes the positions of atoms in the near-surface region to generate kinetic pathways to accessible local minima involving changes in the surface compositions. We showcase our agent by predicting the surface reconstruction pathways of a ternary Ni3Pd3Au2(111) alloy catalyst. Our results show that the DRL agent can not only explore more diverse surface compositions than the conventional minima hopping method, but also generate the kinetic surface reconstruction pathways. We further demonstrate that the kinetic pathway to a global minimum energy surface composition and its associated transition state predicted by our agent is in good agreement with the minimum energy path predicted by nudged elastic band calculations. |
Thursday, March 17, 2022 11:42AM - 11:54AM |
T32.00002: Featurization and Regression Analysis of Stability of Dilute Bimetallic Catalyst Surfaces Isabel Diersen, Cameron J Owen, Steven B Torrisi, Jin Soo Lim, Lixin Sun, Boris Kozinsky Among catalytic materials, dilute transition metal alloys can evince improved catalytic performance compared to their monometallic counterparts. Crucial questions remain surrounding the stability of individual active surface configurations. We attempt to determine which alloy parameters determine configuration stability, using machine learning (ML) models that take as input elemental and structural properties that are quick to compute and easily accessible compared to those calculated using first-principles methods. |
Thursday, March 17, 2022 11:54AM - 12:06PM |
T32.00003: Structure motif–centric machine learning framework for inorganic crystalline systems Qimin Yan, Huta Banjade, Sandro Hauri, Shanshan Zhang, Francesco Ricci, Weiyi Gong, Geoffroy Hautier, Slobodan Vucetic The incorporation of physical principles in a machine learning (ML) architecture is a fundamental step toward the continued development of artificial intelligence for inorganic materials. As inspired by Pauling's rule, we propose that structure motifs in inorganic crystals can serve as a central input to a machine learning framework. We demonstrated that the presence of structure motifs and their connections in a large set of crystalline compounds can be converted into unique vector representations using an unsupervised learning algorithm. To demonstrate the novel use of structure motif information, a motif-centric learning framework is created by combining motif information with the atom-based graph neural networks to form an atom-motif dual graph network (AMDNet), which is more accurate in predicting the electronic structures of metal oxides such as bandgaps. The work illustrates the route toward the fundamental design of graph neural network learning architecture for complex materials by incorporating beyond-atom physical principles. |
Thursday, March 17, 2022 12:06PM - 12:18PM |
T32.00004: Prediction of optical spectra of BeZnO alloys using machine learning Cindy Wong, Andre Schleife The selection of materials with specific, desired optical properties is essential to improve optical and photonic devices. Specifically, BeZnO alloys have been a promising candidate due to their wide and tunable bandgaps. In order to represent the alloy, either cluster expansion methods are used, that are limited to small supercell sizes on the order of 10-20 atoms, or special quasirandom structures are used to describe random alloys by means of somewhat larger cell sizes. We explore whether machine learning can be used to accelerate the computation of optical spectra of large BeZnO alloy supercells. To understand the relationship between alloying and optical properties, structural descriptors are used as input for machine learning models. The optical spectra dataset is created by using density functional theory of several hundreds of representations from a cluster expansion. We train and apply random forest regression to our dataset. Evaluation of these models depends on the following metrics: mean absolute error, root mean squared error, and coefficient of determination. We use these models, once trained, to predict optical spectra of a larger number of cluster classes with high accuracy and at a lower computational cost. |
Thursday, March 17, 2022 12:18PM - 12:30PM |
T32.00005: Stability of copper-based alloys investigated through active learning Angel Diaz Carral, Maria Fyta Copper-based alloys, due to their high electrical conductivity and high strength, are of great interest for electric and electronic applications such as connectors and lead frames. Here, we investigate the stability of Cu-Ni-Si-Cr alloys for different types and concentrations of alloy elements through computational means. Quantum-mechanical simulations are performed on a large number of structural candidates. We use a relaxation-on-the-fly active learning algorithm in order to relax the large number of n-ary candidates to their minimum energy, and target those with the lowest enthalpy of formation. The stability of alloy phases is then assessed based on their convex hull and the phonon density of states analysis. In the end, we propose copper alloy candidates with robust structural properties for practical applications. |
Thursday, March 17, 2022 12:30PM - 12:42PM |
T32.00006: Bias-imbalance in data-driven materials science: a case study on MODNet Pierre-Paul De Breuck, Matthew L Evans, Gian-Marco Rignanese As the number of novel data-driven approaches to material science continues to grow, it is crucial to perform consistent quality, reliability and applicability assessments of model predictions. In this respect, an important task is the uncertainty assessment of a model towards a target domain. Significant variations in test errors can be observed, depending on the imbalance and bias in the training set (i.e., similarity between training and application space). To illustrate this, the Materials Optimal Descriptor Network (MODNet), a method for small datasets is used as a case study on MatBench v0.1, a curated test suite of materials datasets. By using an ensemble MODNet model, confidence intervals can be built and the uncertainty on individual predictions can be quantified. Imbalance and bias issues are often overlooked, and yet are important for successful real-world applications of machine learning in materials science and condensed matter. |
Thursday, March 17, 2022 12:42PM - 12:54PM |
T32.00007: Ligand optimization of exchange interaction in Co(II) dimer single molecule magnet by machine learning Sijin Ren, Eric C Fonseca, William Perry, Hai-Ping Cheng, Xiaoguang Zhang, Richard G Hennig Designing single-molecule magnets (SMMs) for potential applications in quantum computing and high-density data storage requires tuning their magnetic properties, es-pecially the strength of the magnetic interaction. These properties can be characterized by first-principles calculation based on density functional theory (DFT). In this work, we study the experimentally synthesized Co(II) dimer SMM with the goal to control the exchange energy, ΔEJ, between the Co atoms through tuning of the capping ligands. The experimentally synthesized Co(II) dimer molecule has avery small ΔEJ<1 meV. We assemble a DFT dataset of 1081 ligand-substitutions for the Co(II) dimer. The ligand exchange provides abroad range of exchange energies, ΔEJ, from+50 meV to -200 meV, with 80% of the ligands yielding a small ΔEJ<10 meV. We identify descriptors for the classification and regression of ΔEJ using gradient boosting machine learning models. We compare structure-based, one-hot encoded, and chemical descriptors consisting of the HOMO/LUMO energies of the individual ligands and the maximum electronegativity difference and bond order for the ligand atom connecting to Co. We observe a similar overall performance with the chemical descriptors out-performing the other descriptors. We show that the exchange coupling, ΔEJ, is correlated to the difference in the average bridging angle between the ferromagnetic and antiferromagnetic states, similar to the Goodenough–Kanamori rules. |
Thursday, March 17, 2022 12:54PM - 1:06PM |
T32.00008: Neural-Network Predictive Modeling of Physical Properties in Binary Magnetic and Non-Magnetic Alloys Sairam Tangirala, Massimiliano L Pasini, Markus Eisenbach, Ying-Wai Li We present a deep learning (DL) approach to reproduce the first principles Density Functional Theory (DFT) based calculations pertaining to macroscopic physical properties of a non-magnetic (CuAu) and a magnetic (FePt) binary alloys. In this study, a neural network (NN) is developed and trained using thousands of theoretically possible lattice configurations obtained from the Locally Self-Consistent Multiple Scattering (LSMS) DFT code [1]. The intrinsic physical properties of alloys like composition ratio, unit-cell structure, spatial charge distributions, Coulombic interactions, etc. are inputted into the NN model structured by the “bag-of-bonds” representation [2]. The NN regression model is trained to capture the relationship between intrinsic parameters and the total energy of the alloys. Although NNs are complex and computationally expensive to train, they are flexible and can effectively pick up nonlinear relationships between inputs and outputs. Our results show that the trained NN model is orders-of-magnitude faster than DFT in inferring the total energy with comparable accuracy [3]. This demonstrates the potential of applying the NN formalism in accelerating the computational studies of condensed matter systems. |
Thursday, March 17, 2022 1:06PM - 1:18PM |
T32.00009: A thorough descriptor search to machine learn the lattice thermal conductivity of half-Heusler alloys Dipanwita Bhattacharjee, Krishnaraj Kundavu, Parul R Raghuvanshi, Deepanshi Saraswat, Amrita Bhattacharya Predicting the lattice thermal conductivity (KL) of compounds prior to synthesis is an extremely challenging task because of the complexity associated with determining the phonon scattering lifetimes for underlying normal and Umklapp processes. An accurate ab-initio prediction is extremely expensive computationally, seeking data-driven alternatives. We perform machine learning (ML) on theoretically computed KL of half-Heusler (HH) compounds. An exhaustive descriptor list comprising of elemental and compound descriptors is used to build several ML models. We find that ML models built with compound descriptors can reach high accuracy with a fewer number of descriptors, while a set of a large number of elemental descriptors may be used to tune the performance of the model as accurately. Thereby, using only the elemental descriptors, we build a model with exceptionally high accuracy (with an R2 score of ~0.95) using one of the compress sensing techniques. This work, while unfolding the complex interplay of the descriptors in different dimensions, reveals the competence of the simple low-level elemental descriptors in building a robust model for predicting the KL. |
Thursday, March 17, 2022 1:18PM - 1:30PM |
T32.00010: Bayesian optimization for the traversal of molecular properties William Perry, Sijin Ren, Eric C Fonseca, Hai-Ping Cheng, Richard G Hennig, Xiaoguang Zhang In this work, we apply a Bayesian Optimization (BOpt) process to two sets of DFT-generated data of molecules containing transition metals to determine the resource savings that the process would generate when optimizing certain molecular properties. We minimize the HOMO-LUMO energy gap in a database of 641 Manganese based coordination complexes with elements consisting of various combinations of planar-arranged ligands and the ferromagnetic-antiferromagnetic (FM-AFM) ground state energy difference in a database of 1081 Cobalt dimer based Single-Molecule magnets (SMM) with elements consisting of a static core region and different combinations of capping ligands. We compare the behavior of several acuisition functions: probability of improvement (PI), a modified PI, and a custom acquisition function that estimates outcome probability via posterior smoothness. We also employ a non-constant mean function which is based on a Bayesian posterior generated by a partial data set and a Fourier-Transform based descriptor for descriptors with cyclical symmetries. We find that for these discrete data sets the greedy PI acquisition functions reliably enable sampling of molecules with low HOMO-LUMO gap and FM-AFM energies with much less resource expenditure than Monte-Carlo or brute force methods. |
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