Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session E22: Computational Materials Design and Discovery -- Machine LearningFocus
|
Hide Abstracts |
Sponsoring Units: DMP DCOMP Chair: Xavier Gonze, Universite catholique de Louvain Room: BCEC 157C |
Tuesday, March 5, 2019 8:00AM - 8:12AM |
E22.00001: Physics-Based Machine Learning Models for Discovery of Novel Scintillator Chemistries Ghanshyam Pilania, Christopher R. Stanek, Blas Pedro Uberuaga Applications of inorganic scintillators—activated with lanthanide dopants, such as Ce—are found in diverse fields. As a strict requirement to exhibit scintillation, the 4f ground state and 5d1 lowest excited state levels induced by the activator must lie within the host bandgap. This talk will discuss a new machine learning (ML) based screening strategy that relies on a high throughput prediction of the lanthanide dopants’ ground and excited state energy levels with respect to the host valance and conduction band edges for efficient chemical space explorations to discover novel inorganic scintillators. Using a set of perovskite oxides and elpasolite halides as examples, it will be demonstrated that the developed approach is able to (i) capture systematic chemical trends across host chemistries and (ii) effectively screen promising compounds in a high-throughput manner. While a number of other application-specific performance requirements need to be considered for a viable scintillator, the present scheme can be a practically useful tool to systematically down-select the most promising candidate materials in a first line of screening for a subsequent in-depth investigation. |
Tuesday, March 5, 2019 8:12AM - 8:24AM |
E22.00002: Finding Novel Fast Ionic Conductors Using Combined Techniques from Density Functional Theory and Materials Informatics Randy Jalem, Kenta Kanamori, Ichiro Takeuchi, Yoshitaka Tateyama, Masanobu Nakayama Computational new material search is an ongoing hot topic for research in various fields of applications. In here, we show our works related to efficient computational methods for finding novel fast ionic conductors for potential use in solid-state batteries. One topic deals with our proposed search framework based on a Bayesian optimization algorithm with a kernel definition that is general for high dimension of material descriptors, coupled to a DFT method to calculate ion migration energy barriers (Eb) over chemical search spaces of oxides (Eb as a search criterion) (Sci. Rep. 2018, 8, 5845). The next part shows our formulation of descriptors for crystalline solids which are derived from literature data and real feature values from atom-centered Voronoi partitioning (STAM 2018, 19, 231). We validated the scheme in terms of machine learning of various DFT-calculated properties such as cohesive energy, material density, electronic band gap energy, and convex hull decomposition energy. |
Tuesday, March 5, 2019 8:24AM - 8:36AM |
E22.00003: Crystal structure prototype database based on machine learning classification of existing inorganic material structures Shulin Luo, Bangyu Xing, Jian Lv, Lijun Zhang Combining high-throughput calculations with database construction and data mining offers opportunities for computational material scientists to discover new materials. Candidate materials considered in high-throughput calculations are usually from chemical substitution or structure variation based on known crystal structures. So, knowledge of crystal structure prototypes is vital for the validity of high-throughput calculations. We herein built a high quality crystal structure prototype database with the aid of machine learning classification of existing inorganic materials structures. The structure data were classified by the hierarchical clustering approach and the state-of-the-art structure fingerprinters including the bond order parameters and the smooth overlap of atomic positions were used for structure characterization. The database can generate sub-databases dynamically based on new criteria. We have integrated the database into the in-house developed infrastructure of JUMP2, a python framework for materials discovery via high-throughput calculations, aiming at creating an automatic and high-performance computational materials discovery platform. |
Tuesday, March 5, 2019 8:36AM - 8:48AM |
E22.00004: Development of linearly independent descriptor generation method for sparse and interpretable modeling in materials science Hitoshi Fujii, Tetsuya Fukushima, Tamio Oguchi In recent years, researches using techniques of machine-learning have been considerably activated in the field of materials science and we focus on research for empirical law discovery to elucidate a mechanism of physical properties of target materials. We propose linearly independent descriptor generation method for increasing the expression capability of linear regression model without generating any multicollinearity and strong near-multicollinearity which are a major problem in linear regression analysis. Our method is expected to be an essential preprocessing technique for sparse and interpretable modeling in materials science. |
Tuesday, March 5, 2019 8:48AM - 9:00AM |
E22.00005: Important descriptors and descriptor groups of Curie temperatures of rare-earth transition-metal binary alloys Hiori Kino We analyze Curie temperatures of rare-earth transition metal binary alloys with machine learning method. In order to select important descriptors and descriptor groups, we introduce newly developed subgroup relevance analysis and adopt the hierarchical clustering in the representation. We execute the exhaustive search and illustrate that our approach indeed leads to the successful} selection of important descriptors and descriptor groups. It helps us to choose the combination of the descriptors and to understand the meaning of the selected combination of descriptors. |
Tuesday, March 5, 2019 9:00AM - 9:12AM |
E22.00006: Supervised learning and prediction of electronic properties: Discovery and Design of Materials and Interfaces for back-end-of-line interconnects Ganesh Hegde, Harsono Simka, Chris Bowen Supervised machine learning based techniques have found notable success in the recent past in the fields of atomic structure prediction and interatomic potential generation. |
Tuesday, March 5, 2019 9:12AM - 9:24AM |
E22.00007: Machine-Learning-Assisted Accurate Band Gap Predictions of Functionalized MXene Arunkumar Rajan, Avanish Mishra, Swanti Satsangi, Rishabh Vaish, Abhishek Kumar Singh MXene is a recent addition to the ever-growing family of 2D-materials, promising for optical, electronic, energy storage and photocatalytic applications. Utilizing statistical learning-based approach, we electronically characterize this vast class of materials by predicting their band gaps with GW level accuracy. Using a classification model, MXene having finite band gaps are filtered out and few of them are selected to build a machine learning model. The model is built by correlating the easily available elemental and computed properties as features with respect to calculated GW band gaps of selected MXene. Depending upon feature combinations, Gaussian process regression method resulted in optimized model yielding low root-mean-squared-error of 0.14 eV, which can be employed to estimate the accurate GW band gaps of tens of thousands of MXenes [1,2] within minutes. Our results demonstrate that machine learning model can bypass band gap underestimation problem of local and semi-local functionals used in DFT calculations, without subsequent correction using time-consuming GW approach. |
Tuesday, March 5, 2019 9:24AM - 9:36AM |
E22.00008: Accelerating inorganic discovery with meta-calculation filtering via a decision classifier Chenru Duan, Jon Paul Janet, Aditya Nandy, Fang Liu, Heather J Kulik Machine learning (ML) has the capacity to revolutionize materials discovery with accurate property estimation at negligible computational cost. Still, most discovery workflows require computationally-demanding simulation to generate data to feed in an ML model. However, two key challenges remain at the stage of data generation: i) materials may not form a stable complex and ii) calculations may fall outside the domain of applicability of the chosen method. Usually, these two failure modes can only be detected after calculations finished, leading to a massive waste of computational resources. To address this problem, we trained a classifier to estimate the success rate of a calculation directly from topological, heuristic features prior to simulation. Inspired by the data distribution in the latent space, we designed a model confidence metric specifically for classification tasks, lowering the risk of terminating jobs that are actually fruitful. A dynamical classifier that utilizes the information generated during simulation is also developed, which directs the on-the-fly decision of whether to abandon an in-progress calculation. Our classifiers are useful in dataset generation with first-principles calculations to accelerate the ML-driven design of novel inorganic materials. |
Tuesday, March 5, 2019 9:36AM - 9:48AM |
E22.00009: Multi-fidelity Information Fusion with Machine Learning: A Case Study of Dopant Formation Energies in Hafnia Rohit Batra, Ghanshyam Pilania, Blas Pedro Uberuaga, Ramamurthy Ramprasad Cost versus accuracy trade-offs are frequently encountered in materials science, where a particular property of interest can be measured at different levels of accuracy or fidelity. Naturally, the most accurate measurement is also the most resource-intensive, while the inexpensive quicker alternatives tend to be noisy. In such situations, machine learning strategies, such as multi-fidelity information fusion (MFIF), can be employed to fuse information accessible from sources with varying levels of fidelity, and allow for accelerated property predictions. In this work, we use a dataset consisting of dopant formation energies of 42 dopants in hafnia—each studied in six different hafnia phases—computed at two levels of fidelity. The performance of traditional single fidelity (SF) and three MFIF models, namely, Δ-learning, low-fidelity as a feature, and multi-fidelity (MF) co-kriging are compared. We find that the MF based learning scheme not only outperforms the traditional SF machine learning methods, such as Gaussian process regression, but also provides an accurate, inexpensive and flexible alternative to other MFIF strategies. The learning approach is expected to be general and can be readily applied to a much wider spectrum of materials discovery and optimization problems. |
Tuesday, March 5, 2019 9:48AM - 10:00AM |
E22.00010: Machine Learning for Energetic Material Detonation Performance Brian Barnes We present advances in accurate, extremely rapid prediction of detonation performance for energetic molecules. These models may be integrated into a larger effort for high-throughput virtual screening or rapid pre-screening of molecules before any hazardous synthesis is attempted. Our workflow utilizes (a) a reference dataset generated from quantum mechanical calculations and a thermochemical code, (b) a cheminformatics approach to molecular descriptors, and (c) neural network and kernel-based algorithms for nonlinear regression. This data-driven approach leverages modern “machine learning” techniques for prediction of molecular properties. |
Tuesday, March 5, 2019 10:00AM - 10:12AM |
E22.00011: Machine learning study of two-dimensional magnetic materials Trevor David Rhone, Wei Chen, Shaan Desai, Amir Yacoby, Efthimios Kaxiras When the dimensionality of an electron system is reduced, new behavior emerges. This has been demonstrated in GaAs quantum Hall systems since the 1980’s, and more recently in van der Waals (vdW) materials. We discuss the behavior of electrons in reduced dimensions with a focus on their spin properties. We study vdW materials with intrinsic magnetic order, materials at the forefront of physics research. We use materials informatics (machine learning applied to materials science) to study the magnetic and thermodynamic properties of vdW materials. Crystal structures based on monolayer Cr2Ge2Te6, of the form A2B2X6, are studied using density functional theory (DFT) calculations and machine learning tools. Magnetic properties, such as the magnetic moment are determined. The formation energies are also calculated and used to estimate chemical stability. We show that machine learning, combined with DFT, provides a computationally efficient means to predict properties of two-dimensional (2D) magnets. In addition, data analytics provides insights into the microscopic origins of magnetic ordering in 2D. This novel approach to materials research paves the way for the rapid discovery of chemically stable 2D magnets. |
Tuesday, March 5, 2019 10:12AM - 10:24AM |
E22.00012: Stochastic Discovery of Variance Mechanisms in Heterogeneous Dielectric Coatings Venkatesh Meenakshisundaram, David Yoo, Andrew Gillman, James Deneault, Nicholas Glavin, Philip Buskohl Microscale spatial and material heterogeneities in 3D printed electrical devices present significant challenges to predictable electrical performance and device reliability. Dielectric particles are often added to dielectric inks to tailor the macro level permittivity of printed dielectric substrates and coatings. However, the combined role of particle morphology, discrete spatial arrangement and material properties on variance is difficult to distinguish experimentally, due to the large parameter space of processing variables and electrical sensitivity to local heterogeneities. We address this challenge by combining a finite element capacitor model with a neural network guided genetic algorithm (GA) to optimize the volume fraction, particle size and permittivity distributions of dielectric particles to identify systems with high capacitance variance. Classification-based machine learning techniques were also applied to the GA-created database to extract correlations between the spatial/material distributions of the dielectric particles and the capacitance variance. Collectively, the study provides a useful framework to correlate electrical performance with both macro- and microstructural variation sources, which is key to accelerating the development of 3D printing materials. |
Tuesday, March 5, 2019 10:24AM - 10:36AM |
E22.00013: Ligand Optimization for the Spin-Lattice Coupling of Single-Molecule Magnets Mn3 Jie Gu, William Perry, Maher Yazbak, Dianteng Chen, Mark E. Turiansky, Hai-Ping Cheng, Xiaoguang Zhang Single molecule magnets (SMM) exhibit a pseudo-multiferroicity which arises from structural changes that occur during spin state transitions. This spin-lattice coupling leads directly to magnetocapacitance which may also allow the spin state of the ground state to be tuned via strain. We propose that this tuning can be performed via replacement of the ligands which surround the SMM core atoms, and we attempt to search for a ligand which maximizes the spin-lattice coupling in the SMM Mn3 as a test case. We use density functional theory (DFT) calculations, Bayesian optimization, and slightly modified atomic environment vectors (AEVs) to perform the search. These techniques are employed in order to minimize the cost of searching through a large number of candidate ligands from the PubChem database. Following this procedure, we have obtained evidence that the spin-lattice coupling can be affected through a judicious choice of ligand. |
Tuesday, March 5, 2019 10:36AM - 10:48AM |
E22.00014: Identification of stable Cu-Pd-Ag nanoparticles using neural network interatomic potentials Samad Hajinazar, Ernesto D. Sandoval, Aiden J. Cullo, Aleksey Kolmogorov A neural network potential constructed with a stratified training scheme available in the MAISE package [1,2] has been used to find low-energy structures of elemental, binary and ternary Cu-Pd-Ag clusters. The efficiency of the employed unbiased global ground state evolutionary search for elemental nanoparticles was improved by co-evolving clusters across a range of sizes. We systematically compared the stability of the clusters found with the neural network model against previously reported structures found with the Gupta potential. Predictions made with the neural network show a consistent improvement in nanoparticle stability at the density functional theory level. |
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