Bulletin of the American Physical Society
2018 Annual Meeting of the APS Four Corners Section
Volume 63, Number 16
Friday–Saturday, October 12–13, 2018; University of Utah, Salt Lake City, Utah
Session J04: Computational Physics 3 |
Hide Abstracts |
Chair: Remi Dingreville, Sandia National Lab Room: CSC 10/12 |
Saturday, October 13, 2018 8:00AM - 8:24AM |
J04.00001: Machine learning for materials design: a case study in fluorescent metal clusters Invited Speaker: Stacy Copp Advances in experimental science and data management are allowing researchers to amass larger and larger data sets. In the context of materials physics, how can we explore the ever-growing data to better understand the systems we study and to guide discovery of new materials? This talk presents a case study for data-driven materials design: fluorescent silver clusters stabilized by DNA. Composed of just ~10-30 silver atoms, these clusters exhibit fluorescence colors spanning the visible to near-infrared spectrum that are selected by the sequence of the stabilizing DNA strand. Exactly how sequence selects cluster size and thus color is unknown, limiting promising applications of these materials for biosensing and photonics. I will discuss how we are combining high-throughput experiments and machine learning to solve the silver cluster “genome” and to design new DNA template sequences that are selective for cluster size and color. This approach to a soft-matter-inorganic hybrid system characterized by an extremely large parameter space exhibits the potential of machine learning and data mining for materials research. |
Saturday, October 13, 2018 8:24AM - 8:36AM |
J04.00002: General machine learning models for materials prediction Chandramouli Nyshadham, Matthias Rupp, Brayden Bekker, Alexander Shapeev, Tim Mueller, Conrad Rosenbrock, Gabor Csanyi, David Wingate, Gus Hart Machine learning tools applied to problems in materials science are transforming the way we predict properties of materials. These tools enable us to compute properties of materials with the accuracy of quantum mechanics at a fraction of the time. We present five general machine learning based models which were used to simultaneously predict formation energies of 10 different materials (AgCu, AlFe, AlMg, AlNi, AlTi, CoNi, CuFe, CuNi, FeV, NbNi). We show that the results of using machine learning for materials prediction are independent of the particular model used. Prediction errors of all five models were found to qualitatively agree, with errors of the order of 1, meV/atom.
|
Saturday, October 13, 2018 8:36AM - 8:48AM |
J04.00003: Study of CoNiTi alloys using MTP machine Carlos Leon, Wiley S Morgan, Gus L.W. Hart Co and Ni are the most common base elements for alloys with excellent mechanical properties [1]. Here we explore the CoNiTi ternary system for a possible superalloy phase by computing the total energy of more than 200,000 crystal structures. We use a systematically improvable machine learning interatomic potential called Moment Tensor Potential (MTP) [2, 3] to estimate the formation energy of each structure. By analyzing the formation energies, we identify the stable CoNiTi structures. We then compare our results with CoNiTi ternary system in the AFLOW database. Our work will shorten the time for analysis and experimental synthesis of promising CoNiTi alloys with enhanced physical properties. References [1] C. Nyshadham et al. Acta Mater. 122, 438 (2017).
[2] A. Shapeev. Multiscale Model. Simul., 14, 1153 (2016).
[3] K. Gubaev et al. ArXiv:1806.10567 [cond-mat.mtrl-sci].
|
Saturday, October 13, 2018 8:48AM - 9:00AM |
J04.00004: Optimal Data Collection for Machine Learning Mark Transtrum, Kent L Gee, Katrina Pedersen, Brooks Butler, Casie Gaza Machine learning refers to a collection of computational techniques for identifying or learning patterns in data. Although existing techniques are most effective on large data sets, there is growing interest in applying methods on smaller ones. We consider the application of machine learning to predicting ambient sound levels in the contiguous United States from GIS data. The challenge is limited availability of training data from which to construct a model--data collection in this case is both cost and time expensive. This leads us to consider two questions: First, how to best validate a machine learning model with limited training data and two, given additional data can we measurably improve the accuracy of the model. We create an ensemble of models that perform equally well as measured by leave-one-out cross validation on our initial training set. However, these models give wildly different predictions for areas in the central region of the country. By collecting additional data in cropland areas in Utah, we were able to improve the predictions of our machine learning model to other, geographically similar regions of the country. |
Saturday, October 13, 2018 9:00AM - 9:12AM |
J04.00005: Studying Cobalt Based Superalloys with Machine Learning Brayden D Bekker, Chandramouli Nyshadham, Gus L.W. Hart Materials discovery is a catalyst for human progression. Modern computational approaches seek to predict new superalloys, high-performance materials to extend and improve human potential. A recent high throughput search for ternary superalloys revealed six promising candidates including CoTaV and CoNbV. Further experimental investigation on these two ternary systems confirmed the superalloy phase but found them to be metastable. We explored the theoretical phase diagrams of these two ternary systems by computing total energies of more than 200k structures for exploring the stable phases. We do this using a class of systematically improvable potentials called the Moment Tensor Potential (MTP). |
Saturday, October 13, 2018 9:12AM - 9:24AM |
J04.00006: Machine Learning and Materials Discovery Gus LW Hart, Chandramouli Nyshadham, Brayden Bekker, Matthias Rupp, Gabor Csanyi, Alexander Shapeev, Conrad W Rosenbrock, Tim Mueller, David Wingate The relative accuracy and speed of density functional calculations have transformed computational materials science. But true ”materials by design” or in-silico materials discovery has not yet been realized, though there are isolated success stories. To make computational discovery of new materials possible, or to discover materials engineering routes to improve already-deployed materials, a brute force approach will not be practical—some other paradigm will be required. Machine learning, so successful in some other application areas, is an intriguing idea, but there are hurdles to overcome. There are two important differences between the standard machine learning problems of image recognition, voice recognition, etc., and materials prediction. In the first instance, we cannot afford the typical accuracy tradeoff—materials predictions are not useful without meeting a high accuracy target; the energy difference of competing phases is often very small, requiring high fidelity in the models. The second difference is the amount of training data—we don’t have ”big data”. How do we move forward? I will review the state of the art in this emerging discipline and show some results from BYU’s Materials Simulation Group efforts in this area. |
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