Computational Thermodynamics and Kinetics: Data and High Throughput Methods II
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 2:00 PM
February 27, 2020
Room: 33C
Location: San Diego Convention Ctr
Session Chair: Praytush Tiwary, University of Maryland; Maryam Ghazisaeidi, Ohio State University
2:00 PM Invited
Autonomous Efficient Experiment Design for Materials Discovery: A Case Study on MAX Phases: Anjana Talapatra1; Raymundo Arroyave2; Shahin Boluki2; Xiaoning Qian2; Edward Dougherty2; 1Los Alamos National Laboratory; 2Texas A&M University
The goal–oriented discovery of materials necessitates the identification of the composition and process history necessary to achieve specific multi–scale structural features that in turn bring about desired/targeted properties. In this talk, we present a framework capable of optimally exploring the materials design space in order to attain an optimal material response. Specifically, we use variants of the Efficient Global Optimization algorithm to deploy an autonomous computational ç platform capable of performing optimal sequential computational experiments to discover optimal materials - applied to the class of pure MAX phases. We demonstrate single and multi–objective optimization and we also show how this framework can be made robust against selection of non–informative features by using Bayesian Model Averaging approaches. The complete framework thus demonstrates the possibility of attaining a robust and autonomousplatform for computer–driven materials discovery.
2:30 PM
Artificial Intelligence for Predicting Phase Stability on High Entropy Alloys: Anus Manzoor1; Dilpuneet Aidhy1; 1University of Wyoming
Using a combination of artificial intelligence (AI) and density functional theory (DFT), we elucidate the contributions of various entropies, i.e., vibrational, electronic and configurational towards predicting the phase stability of HEAs. We show that the entropy contributions could be quantitatively comparable to the mixing enthalpy; as a result, including various entropy contributions is important for correctly predicting the alloy phase stability. We also show that while the configurational entropy always favors phase stability, the role of vibrational entropy is not predictable. The configurational and vibrational entropies can either compete to destabilize or can collectively contribute to stabilize the solid solutions. As a result, even those systems that have negative mixing enthalpy can show phase instability; conversely, systems with positive mixing enthalpy can have stable phases due to the vibrational entropy contributions. Finally, we discuss our AI database that allows circumventing expensive DFT calculations towards predicting the phase stability of alloys.
2:50 PM Invited
Autonomous Scanning Droplet Cell for On-demand Alloy Electrodeposition and Characterization: Brian DeCost1; Howie Joress1; Stephen Ambrozik1; Trevor Braun1; Zachary Trautt1; Aaron Kusne1; Jason Hattrick-Simpers1; 1National Institute of Standards and Technology
We are developing an autonomous scanning droplet cell (ASDC) capable of on-demand electrodeposition and real-time electrochemical characterization for investigating the corrosion-resistance properties of multicomponent alloys. The ASDC is a millimeter-scale electrochemical cell with a programmable pump array that can electrodeposit an alloy film and immediately acquire polarization curves to obtain electrochemical quantities of interest, such as passive current density and oxide breakdown potential. We develop active machine learning algorithms for experimentally mapping Pourbaix diagrams. We demonstrate this approach to multicomponent Pourbaix diagrams, benchmarking on well-characterized unary systems and prefabricated combinatorial composition spread thin films. We will also discuss the ongoing development of on-demand alloy electrodeposition techniques to enable exploration of higher-order Pourbaix diagrams. Finally, we will discuss the incorporation of prior knowledge in the form of theoretical and data-driven predictive models, such as the Materials Project computed Pourbaix diagrams.
3:20 PM
Bond-order Bond Energy Model for Alloys: Wolfgang Windl1; Christian Oberdorfer1; Maryam Ghazisaeidi1; 1Ohio State University
We introduce a novel way to parameterize alloy energies in the form of a bond-order bond energy model. There, a bond order function models the transition between competing phases and switches their respective bond energies on and off. We demonstrate this on the example of the Ni-Cr-Mo alloy system, which has both face- and body-centered cubic phases. We show that the bond-order bond energy model can predict phase diagrams with excellent accuracy in a simple fashion. We also show that bond-energies define quantitative, composition-dependent chemical potentials in a natural way, allowing to efficiently calculate configuration-optimized alloy vacancy formation energies. As proposed by the concept of the extended Gibbs adsorption isotherm, alloying decreases formation energies, where values smaller than zero indicate thermodynamic instability of the underlying crystal. With that, the bond-order bond energy model provides an intuitive holistic picture that unites defect and phase stability.
3:40 PM Break
4:00 PM Invited
Efficient Navigation of the Search Space for Accelerated Materials Discovery: Prasanna V. Balachandran1; 1University of Virginia
Computational strategies that enable efficient navigation of the vast materials search space have the potential to accelerate the discovery of new materials. This is especially critical when brute-force evaluation of the search space is prohibitively expensive. In this talk, I will focus on examples where we have shown the synergistic integration of density functional theory (DFT) calculations, machine learning (ML), optimal learning and experimental data to rationally guide future experiments in search of new materials with targeted properties. The role of ML is two-fold: (i) to establish a relationship between the features and property of interest and (ii) to quantify prediction uncertainties. The optimal design, on the other hand, uses the ML outcome to recommend the next promising experiment for validation and feedback. Data from DFT calculations are used to construct descriptors for ML or to validate ML predictions to provide confidence in the data-driven approach, prior to performing experiments.
4:30 PM Invited
From Molecular Dissociation to Crystal Nucleation: Next Generation Methods for Sampling Rare Events in All-atom Resolution: Pratyush Tiwary1; 1University of Maryland
Processes such as the nucleation of a crystal or the dissociation of molecular complexes are so slow that they can not be studied using straightforward all-atom simulation techniques. Thankfully, several enhanced sampling algorithms have been proposed that can perform sampling of rare events in an accelerated but controllable manner. However, a large class of these methods need an a priori low-dimensional reaction coordinate (RC) even before performing the sampling. In order to deal with this cyclic problem where one needs extensive sampling to know the RC, but also needs to know the RC to perform sampling, it is thus desirable to construct methods that learn the RC as they perform the sampling. In this talk we will describe two such recent methods, SGOOP and RAVE that use statistical mechanics and deep learning to solve this problem of simultaneously identifying the RC and obtaining converged free energy surfaces and kinetics.
5:00 PM
Using Machine-learning Potentials for Free Energy Calculations of Multicomponent Alloys: Prashanth Srinivasan1; Yuji Ikeda2; Blazej Grabowski3; Jan Janssen2; Alexander Shapeev4; Jörg Neugebauer2; Fritz Körmann2; 1Delft University of Technology; 2Max-Planck-Institut für Eisenforschung; 3University of Stuttgart; 4Skolkovo Institute of Science and Technology
Accurate and efficient computational schemes to predict parameter-free free energies are crucial in computational materials design. A large contribution stems from the vibrational free energy, which determination, including anharmonic contributions is, in general, challenging. We present a scheme which is equally applicable from unaries to concentrated solid solutions (also high entropy alloys) to derive highly accurate vibrational free energies including explicit anharmonic contributions. A machine-learnt moment tensor potential [Shapeev, 2016] is built serving as a highly efficient reference potential for sampling the atomic phase space. The potential is then used as a part of a thermodynamic integration to compute accurate free energies. The scheme is applied to 17 refractory element-based systems ranging from unaries up to five-component high entropy alloys to study the impact of configurational entropy on the total free energy and the vibrational (and anharmonic) contribution. The workflow is implemented in http://pyiron.org to enhance its dissemination and reuse.