Bulletin of the American Physical Society
APS March Meeting 2018
Volume 63, Number 1
Monday–Friday, March 5–9, 2018; Los Angeles, California
Session X34: Machine Learning in Condensed Matter Physics V |
Hide Abstracts |
Sponsoring Units: DCOMP DCMP Chair: Eun-Ah Kim, Cornell Univ Room: LACC 409A |
Friday, March 9, 2018 8:00AM - 8:12AM |
X34.00001: Classifying Surface Probe Images in Strongly Correlated Electronic Systems Via Machine Learning Erica Carlson, Lukasz Burzawa, Shuo Liu Scanning probe experiments such as scanning tunneling microscopy (STM) and atomic force microscopy (AFM) on strongly correlated electronic systems often reveal complex pattern formation on multiple length scales. By studying the universal scaling in these images, we have shown in several distinct correlated electronic systems that the pattern formation is driven by proximity to a disorder-driven critical point,[1, 2] revealing a unification of the pattern formation in these materials. As an alternative approach to this image classification problem of novel materials, here we report the first investigation of the machine learning method to determine which underlying physical model is driving pattern formation in a system. Using a neural network architecture, we are able to achieve 97% accuracy on classifying configuration images from three models with Ising symmetry. By adding machine learning algorithms to our previously developed cluster techniques, we expect the complementary nature of the two techniques to further facilitate our understanding of correlated materials. [1] B. Phillabaum et al., Nat. Commun. 3, 915 (2012); [2] S. Liu et al., Phys. Rev. Lett. 116, 036401 (2016). |
Friday, March 9, 2018 8:12AM - 8:24AM |
X34.00002: AI Identification of the Intertwined Electronic Ordered State Hidden in Complex Electronic Structure Images Andrej Mesaros, Kelvin Chng, Kazuhiro Fujita, Stephen Edkins, Mohammad Hamidian, Hiroshi Eisaki, Shin-ichi Uchida, James Davis, Ehsan Khatami, Eun-Ah Kim With artificial intelligence(AI) revolutionizing data science across disciplines, it is natural to ask whether AI can help us improve the understanding of quantum electronic matter. However, quantum mechanical imaging of electronic behavior, for instance using scanning tunneling microscopy(STM), is probabilistic and hence its interpretation is highly non-trivial. Guided by recent success in training AI to recognize key defining features of many-body states from simulation data, we introduce a general protocol for revealing driving forces of emergent intertwined order in experimental data for quantum matter through an AI-human coalition. Following this protocol, we build and train artificial neural networks to differentiate simulated STM data associated with different theoretical hypotheses. We then employ a group of trained networks to test experimental data obtained from high Tc cuprates over a range of doping. Remarkably the AIs report a feature in the patterns of symmetry breaking in the STM data that points to a universal real-space based mechanism. Hence, we establish that the proposed AI-human coalition can drive future discoveries of quantum phenomena. |
Friday, March 9, 2018 8:24AM - 8:36AM |
X34.00003: Compact representation of crystal structures using three-dimensional diffraction patterns and deep learning Angelo Ziletti, Matthias Scheffler, Luca Ghiringhelli Big data is emerging as a new paradigm in materials science. A vast amount of three-dimensional structural data is provided by both computational repositories (e.g. https://nomad-coe.eu) and experiments (e.g. atom probe tomography). Computational methods that automatically and efficiently detect long-range order are of paramount importance for materials characterization and analytics. Current methods are either not stable with respect to defects, or base their representation on local atomic neighbourhoods, which in turn makes it difficult to detect "average" longe-range order. In the proposed approach, for a given crystal structure we first calculate its diffraction pattern, expand it on spherical harmonics, and then use a neural-network model to obtain a compact, low-dimensional representation. We apply this workflow to a subset of materials from the Novel Materials Discovery (NOMAD) Archive, and show that our deep-learning-based approach compactly encodes structural information, is robust to defects (e.g. point defects, and/or strain), and allows to build easily interpretable structural-similarity maps. |
Friday, March 9, 2018 8:36AM - 8:48AM |
X34.00004: Stochastic Replica Voting Machine Prediction of Stable Perovskite, Double Perovskite and Binary Alloys Tahereh Mazaheri Kouhani
|
Friday, March 9, 2018 8:48AM - 9:00AM |
X34.00005: Predictions of New ABO3 Perovskite Compounds by Combining Machine Learning and Density Functional Theory Prasanna Balachandran, Antoine Emery, James Gubernatis, Turab Lookman, Christopher Wolverton, Alex Zunger We apply machine-learning (ML) methods to a training set of 390 syntheisized ABO3 compounds (254 perovskites (22 cubic and 232 non-cubic) and 136 not perovskites. After classifying the 390 compounds, we construct statistical models that predict out of 625 ABO3 compounds not included in the training set, 235 possible new perovskite materials with 20 new cubic perovskites. We find that the new perovskites are most likely to occur when the A and B atoms are a lanthanide or actinide, when the A atom is an alkali, alkali earth, or late transition metal atom, or when the B atom is a p-block atom. We compare the ML predictions of compounds formed in a given structure with the DFT predictions of compounds stable within 100 meV/atom of the convex hull in these structures. We find that DFT convex hull within OQMD predicts only 87 of the new 235 ML-predicted perovskite compounds to be thermodynamically stable, including 6 cubics. We suggest these 87 as the most promising candidates for future synthesis of novel perovskites. This study clarifies the roles of ML vs DFT predictions of new compounds. |
Friday, March 9, 2018 9:00AM - 9:12AM |
X34.00006: Deep Learning of Perovskite Octahedral Rotations from Electron Microscopy Nouamane Laanait, Albina Borisevich The interdependency between magnetic order, electronic phase transitions, and lattice distortions in strongly correlated systems is well-known. In particular, local distortions in epitaxial perovskites heterostructures, in the form of oxygen octahedral rotations, are of immense interest. Scanning Transmission Electron microscopy (STEM) is one of the few probes with atomic resolution of local distortions, yet determining the 3-dimensional octahedral configuration from a 2-dimensional STEM projection of the lattice is challenging. In this talk, we will present a deep learning model that accurately predicts the 3-dimensional octahedral configuration from STEM experimental studies of a variety of heterostructures such as rare-earth perovskite LaFeO3/EuFeO3 superlattices. Our approach consists of co-training deep convolutional neural networks and deep autoencoders on electron scattering simulations and experiments, respectively. This combination of supervised and unsupervised learning produces a robust and general model capable of extracting both the symmetry and magnitude of octahedral distortions from STEM data, unit cell-by-unit cell, unassisted, and in real time. |
Friday, March 9, 2018 9:12AM - 9:24AM |
X34.00007: Continuous Representation of Chemical Environments for the Prediction of Material Properties Tian Xie, Arthur France-Lanord, Yanming Wang, Jeffrey Grossman Machine learning (ML) methods are becoming increasingly popular for accelerating the design of new materials by predicting material properties with computational speeds orders of magnitudes faster than ab-initio methods. Previously, we developed a generalized crystal graph convolutional neural networks (CGCNN) framework to directly learn structure-property relations from the connectivity of atoms in crystals, providing an accurate and interpretable representation of crystalline materials. Despite its success in the prediction of crystal properties, it fails to extend to a broader range of materials like polymers and glasses, where connectivity alone cannot completely describe the system due to their amorphous nature. In this work, we develop a continuous representation of materials that captures arbitrary configurational and compositional features to predict their properties. We demonstrate the improvement of prediction performances compared with CGCNN on crystalline materials, as well as its application on amorphous materials. Finally, several examples illustrating how this method can be applied to the design of new materials will be presented. |
Friday, March 9, 2018 9:24AM - 9:36AM |
X34.00008: Neural network potentials for disordered carbon and silicon systems. Jorge Hernandez Zeledon, James Lewis Neural network potentials have been used as an alternative method for calculating energies and forces in systems of atoms. In the present work, we improved the classical Behler and Parrinello neural network potential; and, applied our methodologies to systems with different combinations of carbon and silicon, up to several hundreds of particles, arranged in ordered and disordered structure. We will demonstrate how our neural network potential can handle many distinct configurations of disordered materials. Our improvements primarily include - 1) changing the atomic-like potentials for structure-like potentials; and 2) adding four- and five-body interactions to increase the information stored in the features that feed the neural network. With our neural network potential, we can calculate energies and forces for systems of hundreds of atoms, with accuracies close to density functional theory codes, but within the time-frame of a force-field calculation. We use Google’s framework for artificial intelligence, called Tensor Flow - for the implementation of the neural network potential - this enables us to speedup the calculations while keeping the error low. |
Friday, March 9, 2018 9:36AM - 9:48AM |
X34.00009: Electronic Band Structure Prediction with Machine Learning Bart Olsthoorn, Stanislav Borysov, Richard Geilhufe, Alexander Balatsky Within the emerging field of materials informatics, machine learning-based methods are in the forefront of recent research. Although these models usually require a large amount of training data, they can be used for fast and accurate predictions and quantum-mechanical insights. In this work, we explore various machine learning methods to predict an electronic band structure bypassing computationally demanding ab initio calculations. We investigate the problem on a vast amount of randomly generated effective tight-binding band structures. We also estimate the applicability of the developed methods to real materials contained within the Organic Materials Database (OMDB) [http://omdb.diracmaterials.org] using different crystal structure representations. Finally, we briefly discuss the inverse problem of prediction of the of the impact to the initial crystal structure given a slight modification of the corresponding electronic structure. |
Friday, March 9, 2018 9:48AM - 10:00AM |
X34.00010: Active Machine Learning for Combinatorial Exploration of Metal-Insulator Transitions Brian DeCost, Jason Hattrick-Simpers, Yangang Liang, Ichiro Takeuchi, Aaron Kusne We apply a combination of unsupervised and active machine learning algorithms to dynamically map structural phases associated with metal-insulator transitions using sparse X-ray diffraction (XRD) measurements of composition spread thin film libraries. As each XRD pattern is collected, we jointly perform matrix factorization and clustering to identify distinct phases. Semi-supervised learning algorithms extend the clustering results to obtain a probabilistic prediction of the phase diagram, to which we apply an active learning criterion to select a sequence of measurements to maximally increase the confidence of the model for the predicted phase diagram. We demonstrate that our method is nearly an order of magnitude more efficient than dense measurements for rapidly identifying the metal-insulator transition in the VO2-NbO2 system, even when restricting the sampling order to monotonic increases in temperature. Active sampling promises to enable efficient exploration of chemical systems with more than three components with adaptive resolution in composition and temperature. Future work will focus on expanding to other experimental workflows and integrating the synthesis and measurement steps into online experimental systems for optimizing functional properties of materials. |
Friday, March 9, 2018 10:00AM - 10:12AM |
X34.00011: Epitaxial angle of MoS2 grown on h-BN: A first principle and machine learning study Talat Rahman, Duy Le We will report our density functional theory and machine learning study of the epaxial angle of MoS2 grown on h-BN. We will show that the preferred epitaxial angle is zero degree. Interestingly, in the lowest energy alignment, the lattice mismatch between MoS2 and h-BN supercells is not zero. Instead, the MoS2 is found to be slightly larger than BN with a mismatch of about -0.01 %. The dimension of supercell is 72.78 Å. The small mismatch suggests that the resultant MoS2/h-BN is not completely flat but does not experience any significant surface corrugation. The negative mismatch suggests that larger dimension of MoS2 is for accommodating a small corrugation of the MoS2 layer. |
Friday, March 9, 2018 10:12AM - 10:24AM |
X34.00012: A machine-driven hunt for global reaction coordinates of azobenzene photoisomerization James Lewis, Pedram Tavadze, Guillermo Avendaño-Franco, Pengju Ren, Xiaodong Wen, Yongwang Li Azobenzene is an important system which is often studied to better understanding light-activated mechanical transformations via photoisomerization. The C-N=N-C dihedral angle is widely recognized as the primary reaction coordinate for isomerization. We report on a global reaction coordinate to thoroughly describe the reaction mechanism. Our global reaction coordinate includes all of the internal coordinates of azobenzene contributing to the photoisomerization reaction coordinate. We quantify the contribution of each internal coordinate of azobenzene to the overall reaction mechanism. Finally, we provide a detailed mapping on how each significantly contributing internal coordinate changes throughout the energy profile. In our results, the central C-N=N-C dihedral remains the primary internal coordinate responsible for the reaction coordinate; however, we also conclude that the disputed inversion-assisted rotation is half as important to the overall reaction mechanism and the inversion-assisted rotation is driven by four adjacent dihedral angles C-C-N=N with very little change to the adjacent C-C-N angles. |
Friday, March 9, 2018 10:24AM - 10:36AM |
X34.00013: A Data Driven Statistical Model to Predict Critical Temperature of Superconducting Material Kam Hamidieh We estimate a statistical model to predict the critical temperature of superconducting materials based on the features extracted from the constituent elements. The statistical model gives reasonable out of sample predictions: +/- 10 on average based on root-mean-squared-error. The model does not provide a simple picture of the relationships between the features and critical temperature. However, we are able to examine the feature importance in prediction accuracy. It is also crucial to note that our model does not predict whether a material is a superconductor or not; it only gives predictions for superconducting materials. |
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