Computational Thermodynamics and Kinetics: Data and High Throughput Methods I
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Nana Ofori-Opoku, Canadian Nuclear Laboratories; Jorge Munoz, University of Texas at El Paso; Sara Kadkhodaei, University Of Illinois Chicago; Vahid Attari, Texas A&M University; James Morris, Ames Laboratory
Thursday 8:30 AM
February 27, 2020
Room: 33C
Location: San Diego Convention Ctr
Session Chair: Timothy Hartnett, University of Virginia
8:30 AM Invited
Exploring 2D Materials using Machine Learning Force Fields: Xiaofeng Qian1; 1Texas A&M University
Physical behavior of materials often involves complex interactions of electrons and ions, from ultrafast processes at atomistic scale to slow dynamics at mesoscale. First-principles density-functional theory is often accurate, however it is limited to small system and short timescale. In contrast, classical molecular dynamics is a powerful tool for simulating systems at much larger length-scale and longer time-scale, its application is constrained by the availability of accurate force fields. Here I will present our recent development of machine learning force fields (MLFF) from first-principles theory for large-scale long-time simulations. We applied it to study phase transition and thermal properties of 2D materials. The generated force fields are able to capture essential physics both qualitatively and quantitatively such as energetic, structural, and dynamic properties. Our MLFF framework can be generally applied to complex multinary materials, opening up unprecedented avenues for understanding and predicting physical behaviors of complex materials.
9:00 AM
Bayesian Interference of Solid-liquid Interface Properties Out of Equilibrium Based on Phase-field and Molecular Dynamics Simulations: Munekazu Ohno1; Yukimi Oka1; Shinji Sakane2; Tomohiro Takaki2; Yasushi Shibuta3; 1Hokkaido University; 2Kyoto Institute of Technology; 3The University of Tokyo
The phase-field model is an effective tool for simulating microstructural evolution during solidification in alloy systems. It is important for prediction of solidification microstructure to obtain accurate values for solid-liquid interfacial properties such as interfacial energy, interfacial mobility and their anisotropy parameters. Rapidly growing high-performance computing techniques bring the reachable scale of molecular dynamics (MD) simulations to the level of small microstructures, enabling the “on-the-fly” use of information from MD and phase-field simulations in collaboration with data science. In this study, the ensemble Kalman filter (EnKF), a method of data assimilation, is applied to estimation of interfacial parameters for the phase-field simulation from microstructural process obtained by the MD simulation for isothermal solidification in pure Fe. It will be shown that the solid-liquid interfacial energy, interfacial mobility and their anisotropy parameters can be simultaneously estimated based on EnKF.
9:20 AM
Thermal Phenomena in Covalently Bonded Systems Modeled via Physically Informed Neural Network Potentials: James Hickman1; Ganga Purja Pun2; Francesca Tavazza1; Yuri Mishin2; 1National Institute of Standards and Technology; 2George Mason University
The present work focuses on the development and application of a novel class of interatomic potentials known as a “physically-informed neural network” (PINN) potentials. This potential format combines the high level of flexibility inherent to artificial neural networks (ANNs) with the transferability associated with physically inspired analytic potential models. Currently we focus on single component, covalently bonded systems including silicon (Si), germanium (Ge), and carbon (C), however, the PINN model can generally be applied to any multicomponent metallic or covalent system. The comparison between these newly developed force fields and existing classical and ANN potentials demonstrates the increased accuracy and transferability of the PINN model. Finally we discuss the application of these new potentials to the study of various thermal properties including thermal stability, expansion, and conductivity in low dimensional phases such as Silicene and Germanene.
9:40 AM Invited
Materials Design in High Dimensional Chemical and Structural Configuration Spaces: Joerg Neugebauer1; Jan Janssen1; Liam Huber1; Yuji Ikeda1; Fritz Koermann1; Blazej Grabowski2; Tilmann Hickel1; Alexander Shapeev3; 1Max-Planck-Institut fuer Eisenforschung; 2University of Stuttgart; 3Skoltech
Modern engineering materials have evolved from chemically and structurally simple alloys with well-established design rules to chemically, thermodynamically and structurally highly complex materials. Ab initio approaches provide perfect tools to explore and identify design strategies but face serious challenges to efficiently handle the vast high-dimensional configuration spaces resulting from this complexity. To address this complexity, we have developed a python based framework called pyiron that combines (i) rapid prototyping of the complex simulation protocols as needed to achieve computational efficiency and accuracy, (ii) a seamless integration to big data analytics and machine learning tools, (iii) a simple upscaling from interactive prototyping to high-throughput calculations on supercomputer frameworks. The flexibility and the predictive power of this approach will be discussed for examples ranging from the design of ductile Mg alloys, over describing the finite temperature thermodynamics of high entropy alloys, to the discovery of general rules for interstitials in metals.
10:10 AM Break
10:25 AM Invited
Multi Cell Monte Carlo Method for Phase Prediction: Maryam Ghazisaeidi1; You Rao1; Wolfgang Windl1; Changning Niu2; 1Ohio State University; 2QuesTek Innovations LLC
We present a Multi-Cell Monte Carlo algorithm, or (MC)^2, for predicting stable phases of arbitrary alloys. Free atomic transfer among cells is achieved via the application of the lever rule, where an assigned molar ratio virtually controls the percentage of each cell in the overall simulation. All energies are computed using density functional theory. We test the method by successful prediction of the stable phases of known binary systems. We then apply the method to a quaternary high entropy alloy. The method is particularly robust in predicting stable phases of multi-component systems for which phase diagrams do not exist.
10:55 AM
Prediction of Formable Apatites using Machine Learning and Density Functional Theory: Timothy Hartnett1; Mukil Ayyasamy1; Prasanna Balachandran1; 1University of Virginia
Materials with apatite structure have applications ranging from biomaterials to fuel cells. Their chemical flexibility and structural diversity provide a fertile ground to tune functionalities as potential candidates for many applications. In this work, Random Forest and Gradient Tree Boosting methods are applied to search for formable compounds in the apatite structure-type. The results are compared with a traditional convex hull analysis of formation energy predicted from density functional theory (DFT) calculations. We assert that data-driven methods have the potential to capture more formable compounds than the hull method since these are based on experimental results rather than DFT simulations. By comparing formable apatites to their predicted hull energies we are able to derive a “Degree of Metastabilty” where the Hull analysis predicts the compound is unable to form but formation is still observed. Thus, using machine learning, we are able to identify compounds which would be missed by DFT.
11:15 AM
First-principles Methods to Elucidate the High-temperature Thermodynamics of Multicomponent Alloys: Anirudh Raju Natarajan1; Pavel Dolin1; Anton Van der Ven1; 1University of California, Santa Barbara
Materials with several alloying elements are attractive candidates in high-performance engineering applications. First-principles based computational methods have emerged as the tool of choice in identifying promising alloying elements and discerning optimal processing routes. However, navigating this high-dimensional space in a systematic manner has proved challenging. This problem is exacerbated by complex structural and order-disorder phase transformations that occur when alloying multiple elements. In this talk, we will employ high-throughput first-principles calculations to inform the parameterization of accurate cluster expansion Hamiltonians and derive finite-temperature properties through statistical mechanics techniques. We will also describe how structural phase transformation mechanisms may be discerned from such high-throughput first-principles calculations. These techniques will be used to elucidate design principles in multicomponent lightweight and refractory alloys.
11:35 AM Invited
MS-CRADLE: A Tool for Developing Corrosion Resistant HEAs for Molten Salt Technologies: Thien Duong1; Xiaoli Yan2; Santanu Chaudhuri1; 1Argonne National Laboratory; 2University of Illinois at Chicago
HEAs are potential container materials for molten salt (MS) technologies. Although their aqueous corrosion have been studied, little is known about how they interact with MS. Interesting questions are: how do thermodynamics of alloys and salts affect corrosion? More specifically, are high-energy surfaces/GBs susceptible to MS attack? Would depletion happen and alter HEAs' stability? What role ionic impurities play? Since the composition space of HEAs are higher than conventional alloys, their R&D cost is higher. To reduce the cost, we present our initial results from a high-throughput modeling framework for Molten Salt Corrosion Resistant Alloy Design and Lifetime Evaluations (MS-CRADLE). MS-CRADLE's focus is on high-throughput (HT) calibration of thermodynamic descriptors relevant to above questions using DFT calculations. Specifically, surface/GB energies, work functions, adsorption and formation energies. Modeled properties will be coupled with HT experiments and subsequent corrosion testing to build a machine-learning model that helps to make synthesis decision.