Computational Methods and Experimental Approaches for Uncertainty Quantification and Propagation, Model Validation, and Stochastic Predictions: Uncertainty Quantification and Model Validation for Classical Force Fields
Sponsored by: TMS: Computational Materials Science and Engineering Committee, TMS: Chemistry and Physics of Materials Committee, TMS: Integrated Computational Materials Engineering Committee
Program Organizers: Francesca Tavazza, National Institute of Standards and Technology; Richard Hennig, University of Florida; Li Ma, NIST; Shawn Coleman, ARL; Jeff Doak, QuesTek Innovations, LLC; Fadi Abdeljawad, Sandia naional Laboratory
Thursday 8:30 AM
March 2, 2017
Location: San Diego Convention Ctr
Session Chair: Shawn Coleman, U.S. Army Research Laboratory; Lucas Hale, National Institute of Standards and Technology
8:30 AM Invited
Advancements in Parameterization and Validation of Empirical Potentials: Tao Liang1; Kamal Choudhary2; Susan Sinnott1; 1Pennsylvania State University; 2NIST
Molecular dynamics (MD) simulations are often used in conjunction with experimental data, theory, and simulations at other length scales to address problems in materials science and engineering. The most challenging question that MD simulations have faced is: “What is the confidence level associated with the simulation predictions?” The answer to this requires improvement in the transferability of empirical potentials as well as developing the toolbox to quantify the uncertainty of these potentials. Coupling classical MD simulations with existing materials databases and cyberinfrastructures, we are building a toolbox to validate the potentials that are available in LAMMPS. In particular, we have started with validating selected compounds, such as Al2O3, and molecules, such as of hydrocarbons. The results are used to identify the origin of errors associated with the empirical potentials and to propose solutions to ultimately improve the transferability and design of empirical potentials.
Development of Semi-Empirical Potentials Suitable for Simulation of Phase Transformations in Titanium: Mikhail Mendelev1; Tom Underwood2; Graeme Ackland3; 1Ames Laboratory; 2University of Bath; 3University of Edinburgh
Titanium is among the most important metals for use as a structural material. A Ti semi-empirical potential should reproduce the two main features: the existence of several crystal phases and the very high hcp basal stacking fault energy which leads to deformation via prism and twinning slip. We will present an EAM potential which satisfies these prerequisites. A special attention will be paid to determination of the hcp-bcc transformation temperature: we will demonstrate that the values obtained from the molecular dynamics simulation and the lattice-switch Monte Carlo method agree with each other within 2 K. We will also present another Ti potential which provides an excellent agreement with ab initio data on the point defect formation energies in the hcp Ti but dramatically underestimates the melting temperature. We will discuss the reason why all properties cannot be described by a single EAM potential and propose a temperature dependent EAM potential.
Evaluation and Comparison of Classical Interatomic Potentials through a User-friendly Interactive Web-interface: Kamal Choudhary1; Faical Congo1; Francesca Tavazza1; 1National Institute of Standards and Technology
Classical empirical potentials/force-fields provide atomistic insights into material phenomena through molecular dynamics and Monte Carlo simulations. Despite their wide applicability, a systematic evaluation of materials properties using such potentials and, especially, an easy-to-use user-interface for their comparison is still lacking. To address this deficiency, we computed energetics and elastic properties of variety of materials such as metals and ceramics using a wide range of empirical potentials and compared them to density functional theory (DFT) as well as to experimental data, where available. The database currently consists of 3128 entries among energetics and elastic property calculations, and it is still increasing. The data covers 1471 materials and 116 force-fields. A major feature of this database is that the web interface offers easy look up tables to compare at a glance the results from different potentials (for the same system).
9:40 AM Invited
Evaluation of Atomistic Potentials for Silicon: Ganga P. Purja Pun1; Yuri Mishin1; 1George Mason University
We present a systematic comparison of several most popular atomistic potentials for Si, pointing to their strengths and weaknesses with respect to various properties. To address some of the shortcomings of the existing potentials, a new Si potential has been developed by fitting to experimental and first-principles data by utilizing a combination of a genetic algorithm and simulated annealing. The potential has a modified Tersoff format with additional parameters. Extensive tests indicate that the potential accurately reproduces a wider spectrum of properties than other potentials, including the elastic constants, phonon frequencies, point defect formation energies, surface energies and reconstructions, melting properties, formation energies of a large set of alternate crystal structures and a number of other properties. General strategies of potential development for atomistic simulations are discussed.
10:10 AM Break
10:30 AM Invited
Uncertainty Quantification of Classical Interatomic Potentials: Eugene Ragasa1; Christopher O'Brien2; Richard Hennig1; Stephen Foiles2; Simon Phillpot1; 1University of Florida; 2Sandia National Laboratories
The materials fidelity of classical interatomic potentials has increased significantly over the last few decades. It is thus now meaningful to assess the uncertainty in the predictions of specific potentials. Here briefly review some well-known ideas in the economic theory of investment portfolio management and suggest that similar approaches may prove fruitful in uncertainty quantification of interatomic potentials. In particular, we show that the analysis of a potential in terms of the Pareto surface allows the parameterization with high materials fidelity and with high robustness. The efficacy of this approach is illustrated for the simple example of a Buckingham potential for MgO. The analysis of the Pareto surface to compare the potential materials fidelity of various functional form for interatomic potentials is discussed.
11:00 AM Invited
Molecular Dynamics, Dislocation Interactions and Uncertainty: Lucas Hale1; Zachary Trautt1; Chandler Becker1; 1National Institute of Standards and Technology
Classical atomistic simulations are uniquely suited for studying dislocation interactions since the simulations provide the necessary atomic level description of the defects involved, along with the capability of observing and measuring dynamic behaviors. However, molecular dynamics predictions are dependent on the choice of interatomic potential used. This is especially relevant for defect interaction simulations as the complex conditions observed likely have features outside any potential’s fitted phase space. To help address this, high-throughput tools developed for the NIST Interatomic Potentials Repository are used to perform a variety of molecular dynamics simulations across a number of different potentials to investigate the interaction of bcc dislocations and vacancies. This methodology allows for predictions across potentials to be compared, as well as investigations of how basic materials properties influence the predictions of more complex behaviors.