Bulletin of the American Physical Society
APS March Meeting 2019
Volume 64, Number 2
Monday–Friday, March 4–8, 2019; Boston, Massachusetts
Session X19: Computational Materials Design and Discovery -- Potential Models and Molecular Dynamics |
Hide Abstracts |
Sponsoring Units: DMP DCOMP Chair: Alexander Urban Room: BCEC 156C |
Friday, March 8, 2019 8:00AM - 8:12AM |
X19.00001: Deep neural networks to accelerate and reproduce DFT Linda Hung, Brian Rohr, Kristopher S Brown, Michael Statt, Patrick Herring, Arjun Bhargava, Ha-Kyung Kwon, Santosh Suram, Muratahan Aykol, Jens Hummelshøj Databases such as the Open Quantum Materials Database and the Materials Project contain the results of density functional theory (DFT) calculations for hundreds of thousands of materials structures. Data at this scale allows us to leverage machine learning models to accelerate DFT computations, or to completely replace them. In this talk we present our recent progress in deep neural network models that can accelerate or reproduce DFT predictions, including energies, band gaps, and electron densities. When training for multiple targets, these networks also generate reduced-dimensional latent space representations that may act as materials fingerprints. |
Friday, March 8, 2019 8:12AM - 8:24AM |
X19.00002: MAISE package: Materials prediction accelerated with neural network potentials Aleksey Kolmogorov, Samad Hajinazar, Ernesto D. Sandoval Our recently released Module for Ab Initio Structure Evolution (MAISE) features an evolutionary algorithm for structure prediction and a neural network formalism for modeling interatomic interactions [1]. The open-source code has a simple interface for parallelized local optimization of crystalline or nanosized structures using a library of neural network or traditional classical potentials. Unique capabilities include a symbiotic evolutionary optimization of nanoparticles and a stratified construction of neural network models for multicomponent systems [2]. This presentation will review confirmed MAISE predictions and illustrate the acceleration of global structure searches with neural network models. |
Friday, March 8, 2019 8:24AM - 8:36AM |
X19.00003: Accelerating superalloy discovery using moment tensor potentials Hayden Oliver, Brayden Bekker, Chandramouli Nyshadham, Carlos Alberto Leon Chinchay, Gus Hart Superalloys are used in applications where high-temperature strength is necessary. Their extraordinary mechanical properties rely on the formation of the γ' – L1_{2} phase. The discovery of a new superior superalloy will revolutionize the energy and transit industries. Computational methods such as density functional theory (DFT) are used to predict the γ’ phase, but DFT is expensive and it limits how fast we can discover new superalloys. Moment Tensor Potentials (MTP) create an interatomic potential via machine learning that approximates the quantum mechanical energies of a crystal structure many orders of magnitude faster than DFT. We explore the ternary alloy system Al-Co-W, who’s γ' phase was discovered experimentally in 2006, and show that MTP correctly predicts this phase. We also explore the ternary alloy system Hf-Ni-Si, which is a superalloy candidate that was found to have a lower formation enthalpy and smaller decomposition energy than Al-Co-W, and report the results of the investigation. |
Friday, March 8, 2019 8:36AM - 8:48AM |
X19.00004: Using machine learning interatomic potentials for finding CoNiTi ternaries Carlos Alberto Leon Chinchay, Wiley S Morgan, Gus Hart Superalloys are materials with excellent mechanical properties at extreme temperatures. Superalloys including Co or Ni atoms [1] prompt for a thorough search to improve industry-demanding properties, which requires new computational methods to sweep the huge amounts of possibilities. We search through 200,000 CoNiTi crystal structures to find superalloy phases using machine learning based on interatomic moment tensor potentials (MTP) [2, 3]. We have not only reproduced results reported in the AFLOW database but also predicted stable binary and ternary phases that are not present in the literature. The MTP approach shortens the computational analysis of CoNiTi systems by a factor of 100 compared to a pure DFT methodology. Further analysis will include searching for stable structures at higher temperatures for possible industrial applications. |
Friday, March 8, 2019 8:48AM - 9:00AM |
X19.00005: Exploring Materials Space with Machine Learning Brayden Bekker, Hayden Oliver, Chandramouli Nyshadham, Alexander Shapeev, Gus Hart Electronic structure calculations are too computationally expensive to thoroughly explore the composition space of any system. We use a surrogate model approach that constructs an interatomic potential from a small training set of electronic structure calculations. The surrogate model, called the Moment Tensor Potential (MTP)[1], automates and optimizes the creation of the training set for the interatomic potentials. The potential is then used to explore materials space and predict stable structures including those with new geometries not contained in the training set[2]. We use MTP to study the six promising candidates from a recent high-throughput search for ternary superalloys [3], namely MnNiSb, NiSbTi, NiSbSi, HfNiSi, CoTaV, and CoNbV. We analyze the phase diagram of each ternary system using a pool of 1.2 million structures to show the accuracy of these machine learned surfaces and predict new stable phases at a fraction of the computational time. |
Friday, March 8, 2019 9:00AM - 9:12AM |
X19.00006: Combined cluster and atomic displacement expansion for solid solutions and magnetism Kevin Garrity Finite temperature disordered solid solutions and magnetic materials are difficult to study directly using first principles calculations, due to the large unit cells and many independent samples that are required. In this work, we develop a combined cluster expansion and atomic displacement expansion, which we fit to first principles energies, forces, and stresses. We then use the expansion to calculate thermodynamic quantities at nearly first principles levels of accuracy. We demonstrate that by treating all the relevant degrees of freedom (DOF) explicitly, we can in some cases achieve better convergence than a simple cluster expansion, and we can naturally treat coupling between structural DOF and chemical or magnetic DOF. As examples, we use our expansion to calculate properties of Si_{1-x}Ge_{x}, magnetic MnO, Al with vacancies, and K_{x}Bi_{1-x}TiO_{3}. |
Friday, March 8, 2019 9:12AM - 9:24AM |
X19.00007: High-order rotational invariants for structural representations: Application to linearized machine learning interatomic potential Atsuto Seko, Atsushi Togo, Isao Tanaka Machine-learning interatomic potential (MLIP) has been of growing interest as a useful method to describe the energetics of systems of interest. Firstly, we examine the accuracy of linearized pairwise MLIPs and angular-dependent MLIPs for 31 elemental metals [1]. They correspond to generalizations of the embedded atom method (EAM) and modified EAM potentials, respectively. Building the optimal MLIPs for the 31 elemental metals, we show the robustness of the linearized frameworks, the general trend of the predictive power of MLIPs and the limitation of pairwise MLIPs. We also introduce higher-order rotational invariants for improving the accuracy of linearized MLIPs.In this study, a set of rotational invariants up to six-order is derived by the general process of reducing Kronecker products of irreducible representations for SO(3) group. The use of high-order invariants significantly improves the prediction error for a wide range of structures generated from many structure types. |
Friday, March 8, 2019 9:24AM - 9:36AM |
X19.00008: The second-principles MULTIBINIT software project Fabio Ricci, Alexandre Martin, Marcus Schmitt, Jordan Bieder, Xu He, Eric Bousquet, Matthieu J. Verstraete, Philippe Ghosez Density Functional Theory calculations are limited to relatively small spatial- and time-scales. The purpose of the MULTIBINIT project is to extend the capabilities of first-principles codes to predict properties at the meso-scale, accounting for external constraints (temperature, pressure and fields), while retaining most of the first-principles predictive power and accuracy. MULTIBINIT is distributed within the ABINIT package [1] exploiting first-principles data by a “second-principles” approach. In its initial form, it relies on effective atomic potentials [2] for lattice dynamics, including also explicit coupling with homogeneous strains. Moreover, current developments involve the construction of a spin model and its coupling with the lattice. MULTIBINIT integrates efficient tools for the (i) automatic generation of the models, (ii) automatic fit of the coefficients from first-principles data, (iii) finite temperature simulations and (iv) efficient analysis of results. The power of the method will be illustrated on the full-Heusler Fe2VAl compound to predict finite temperature lattice dynamics and the influence on the thermoelectric properties. |
Friday, March 8, 2019 9:36AM - 9:48AM |
X19.00009: Structural Deformations in Graphene under Laser Ablation Mohammad Alaghemandi, Michelle Y. Sander, Sahar Sharifzadeh Graphene, with its extraordinary properties, is an excellent compound for a wide range of applications. Moreover, the ability of designing, controlling, and fabricating 3D structures based on graphene would be a breakthrough for manufacturing the new advanced nano-structures. Laser ablation is currently the only technique that allows patterning of free-standing substrates. In this study, we investigate the ablation of a single layer graphene under high energy pulses by using first-principles density functional theory and reactive molecular dynamics (RMD) simulations. To mimic the laser pulse irradiation, we locally heat the selected area of the graphene layer using a Nosé-Hoover thermostat and considered a range of thermally-heated areas with a radius from 2 to 100 Å, and temperature from 1000 to 10000 K. RMD studies indicate that the shape of the ablated area is not only a function of the pulse energy, but also the radius of the pulse beam. When the radius of pulse beam is smaller than 10 Å, no deformation in graphene is observed for pulses with temperature lower than 8000 Å. Additionally, our predicted trends in the size and shape of the ablated areas coincide well with the experimental results carried out using femtosecond laser beams on a micrometer scale. |
Friday, March 8, 2019 9:48AM - 10:00AM |
X19.00010: Atomistic investigation of a carbonization process for C/H/O/N-based polymers with use of the reactive potentials: ReaxFF Malgorzata Kowalik, Chowdhury M. Ashraf, Adri C. T. van Duin During a carbonization process of the raw polymer precursors, the graphitic structures evolve and are responsible for improved mechanical properties of the carbonized carbon fibers. To gain a deeper understanding of a chemistry behind an evolution of these graphitic structures, we perform atomistic simulations using a reactive force field: ReaxFF. We considered three different polymers as a precursor: idealized ladder PAN, proposed oxidized PAN and PBO to understand how underlying molecular details of polymers direct final carbon fibers structure. Since these are C/H/O/N-based polymers, firstly we proposed an improved force field for C/N/H chemistry based on new DFT data. Then, with use of this improved force field, we perform atomistic simulations of the carbonization process for the considered polymers. Based on these simulations we were able to determine small molecules, as well, all carbon rings productions, and analyzed the graphitic structures evolutions. We also performed the stress-strain simulations on the initially carbonized samples and were able to assess how a presence of the graphitic structure affects the mechanical responds. |
Friday, March 8, 2019 10:00AM - 10:12AM |
X19.00011: Molecular Dynamics Investigations of the Mechanical Properties of Heterogeneous Structures Composed of Graphene Sheets, Graphene Ribbons, and Boron Nitride Sheets Cuiying Jian, Nicola Ferralis, Jeffrey C Grossman It is well known that monolayer/multilayer graphene possesses ultra-high elastic modulus and critical stress with intrinsic low toughness. In order to explore ways to reduce its brittleness, in this work we employ atomistic modeling to examine the mechanical properties of mono- and multilayers of graphene, nanoribbons, boron nitride and a range of combinations. Regardless of size, edge type exhibits a significant effect on the critical stress/strain, evidenced by the higher mechanical strength of zigzag compared to armchair ribbons, as well as the armchair direction in graphene sheets. Furthermore, our calculations show that with increasing number of layers the mechanical strength of graphene deviates from that of the monolayer sheet, while in contrast multilayer boron nitride sheets preserve the fracture point of the single layer. Based on these results, we explored a range of heterogeneous structures composed of graphene sheets, graphene nanoribbons, and boron nitride sheets. Under tensile stretch, components in these heterogeneous structures show asynchronous cracking behavior helping to improve the overall toughness. By analyzing interfacial binding/sliding between heterogeneous components, the underlying mechanisms are explored and improved compositions can be proposed. |
Friday, March 8, 2019 10:12AM - 10:24AM |
X19.00012: Toward Sustainable Asphaltic Materials: A Molecular Dynamics (MD) Investigation of Bio-oil Modified Asphalt Iskinder Arsano, Kshitij C Jha, Mesfin Tsige We have investigated the behavior of a novel asphaltic material obtained by adulterating a model asphalt system^{1} with a rationally selected foreign species of bio-oil. A wide range of systems is modeled by MD simulations featuring a systematic combination of modifier type, modifier concentration, and temperature. The mobility and structural integrity of the proposed materials were investigated in addition to mechanical response prediction by use of moduli calculations. The need to resolve issues of processability such as phase separation of commonly used modifiers, for example scrap tire^{2}, represents an important impetus for the current project. A case is made whereby the molecular observations made in the proposed modified asphalt systems indicate a maintenance of or improvement over essential macro functionalities of pristine asphalt. The replacement of a significant proportion of traditional asphalt by bio-oils constitutes a move toward green construction materials by allowing for less use of bitumen, a byproduct in heavy oil refineries implicated in pollution. |
Friday, March 8, 2019 10:24AM - 10:36AM |
X19.00013: Computational design of organic molecules for reducing friction at the nanoscale Jing Yang, Jon Paul Janet, Fang Liu, Heather J Kulik Computational modeling has the promise to enable atom-by-atom design of nanoscale properties that give rise to essential changes in macroscale properties. In the quest for increasing energy efficiency and resource utilization, energy losses remain an outstanding challenge that can be solved in part through computational materials design. Friction reducers (FRs) are molecular additives that can minimize friction loss in these engines by reducing friction between moving parts at the nanoscale. Traditional FRs contain metals, sulfur, and phosphorus, which can poison exhaust system catalysts and diesel particulate filters. Thus, if suitably designed, organic friction reducers (OFRs) present a promising alternative solution. Here, we apply non-equilibrium molecular dynamics simulations together with density functional theory methods to enable the atom-by-atom design of OFRs. We directly compute the coefficients of friction of OFRs on model engine iron oxide surfaces with varying coverage and temperature, and explore a number of conditions not easily probed during experiments. These studies allow us to build a quantitative structural property relationship for predicting good OFR characteristics, enabling an iteratively improving materials design workflow. |
Friday, March 8, 2019 10:36AM - 10:48AM |
X19.00014: Machine-Learning Provides New Insights into the Coil-to-Globule Transitions of Thermosensitive Polymers Karteek Kumar Bejagam, Yaxin An, Samrendra Singh, Sanket Deshmukh First of its kind, a temperature-independent coarse-grained (CG) model of poly(N-isopropylacrylamide) (PNIPAM) that can accurately predict its experimental lower critical solution temperature (LCST) in the presence of explicit water model is developed. This extensively validated CG model by conducting MD simulations by changing the radius of gyration of initial structure, the chain length, and the angle between the adjacent monomers of the initial configuration of PNIPAM. The model could retain PNIPAM’s tacticity and thereby predict its LCST, which is consistent with experiments and all-atom simulations. A data-driven machine-learning (ML) approach, non-metric multidimensional scaling (NMDS) method, was used to analyze these CG MD simulation trajectories. This analysis suggest that PNIPAM chain undergoes a coil-to-globule transition above the LCST via multiple metastable states. |
Friday, March 8, 2019 10:48AM - 11:00AM |
X19.00015: Bayesian inference of grain growth prediction via multi-phase-field models Hiromichi Nagao, Shin-ichi Ito, Takashi Kurokawa, Tadashi Kasuya, Junya Inoue We propose a Bayesian inference methodology to evaluate unobservable parameters involved in multi-phase-field models with the aim of accurately predicting the observed grain growth, such as in metals and alloys. This approach integrates models and a set of observational image data of grain structures. Since the set of image data is not a time series, directly applying conventional inference techniques that require time series as the input data is difficult. Our key idea is to construct a time series with an appropriate statistic that characterizes static image data of grain structures. The empirical Bayes method estimates not only a probability density function of the parameters but also an initial phase-field, which is generally unobservable in real experiments. The proposed method is confirmed to estimate, from real experimental images of grain structures in a steel alloy, unobservable parameters together with their uncertainties, and successfully selects the initial phase-field that best explains the experimental data from among candidate initial phase-fields. |
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. |
© 2020 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
1 Research Road, Ridge, NY 11961-2701
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700