Computational Thermodynamics and Kinetics: Data Methods, Tools and High Throughput
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS: Chemistry and Physics of Materials Committee, TMS: Computational Materials Science and Engineering Committee
Program Organizers: Nana Ofori-Opoku, Canadian Nuclear Laboratories; Eva Zarkadoula, Oak Ridge National Laboratory; Enrique Martinez Saez, Clemson University; Vahid Attari, Texas A&M University; Jorge Munoz, University of Texas at El Paso
Thursday 8:30 AM
March 18, 2021
Room: RM 54
Location: TMS2021 Virtual
Session Chair: Rodrigo Freitas, Stanford University; Dehao Liu, Georgia Institute of Technology; Arunima Singh, Arizona State University; Raymundo Arroyave, Texas A&M University
8:30 AM Invited
Data-driven Discovery of Materials for Photocatalytic Energy Conversion: Arunima Singh1; 1Arizona State University
Photocatalytic fuel production, e.g. generation of hydrogen from water splitting or the conversion of the greenhouse gas carbon dioxide to chemical fuels, promise us alternative energy sources that are clean, environmentally friendly, and renewable. In this talk, I will show how we designed first-principles simulations based descriptors for thermodynamic stability, synthesizability, corrosion resistance, visible-light absorption, and compatibility of the electronic structure for the catalytic reaction and used them for a rational data-driven photocatalytic materials screening. I will present two examples here, first where we have performed the largest CO2 photocathode search to date, starting with 68,860 candidate materials and found that only 52 materials meet the stringent requirements for CO2 reduction photocatalysts. In the other example, I will show how we employed data-mining and high-throughput simulations to design two-dimensional forms of the discovered photocathodes that promise high-efficiency fuel production due to their large specific surface area.
9:00 AM
High-throughput Density-functional Theory Methods for Discovery of Actinide Materials: Matthew Christian1; Erin Johnson2; Theodore Besmann1; 1University of South Carolina; 2Dalhousie University
Density-functional theory has become a key tool for modeling and understanding material properties under a wide range of conditions. High-throughput adaptations using various approximations have been developed for main-group elements, but have not been extended to actinide elements. High-throughput methods for actinide elements can be problematic because of the complication of relativistic effects and electron occupations that can result in metastable electronic effects. This talk discusses the role of functional, occupational and relativistic effects on a benchmark set of uranium compounds to discuss key factors for accurately developing a high-throughput method for actinide material discovery.
9:20 AM Invited
The High Entropy Alloy Space is Not as Big as We Think It is: Raymundo Arroyave1; Tanner Kirk1; 1Texas A&M University
Over the past decades, the concept of "high entropy alloys" has become a source of inspiration for the field of metallurgy as we try to identify yet to be explored regions in the metal alloy space with properties that can potentially surpass those of alloys currently in use in a number of applications. The "behind entropy" premise of much of the HEA program in the early years has given way to the argument that the HEA space is vast and therefore there are boundless opportunities for further discovery. While strictly speaking the HEA alloy+process space indeed is infinite, in this work we present some recent investigations that suggest that, while big, the feasible HEA space in any given sub-sector (e.g. FCC HEAs, RHEAs, etc) is severely constrained by typical alloy design considerations. Combining CALPHAD, physics-based models, machine learning, search/optimization algorithms we present a more nuanced view of the HEA space.
9:50 AM Invited
Uncovering Atomistic Mechanisms of Crystallization Using Machine Learning: Rodrigo Freitas1; Evan Reed2; 1Massachusetts Institute of Technology; 2Stanford University
Solid-liquid interfaces have notoriously haphazard atomic environments. While essentially amorphous, the liquid has short-range order and heterogeneous dynamics. The crystal, albeit ordered, contains a plethora of defects ranging from adatoms to dislocation-created spiral steps. All these elements are of paramount importance in the crystal growth process, which makes the crystallization kinetics challenging to describe concisely in a single framework. In this seminar I will introduce a novel data-driven approach to systematically detect, encode, and classify all atomic-scale mechanisms of crystallization. I will also show how this approach naturally leads to a predictive kinetic model of crystallization that takes into account the entire zoo of microstructural elements present at solid-liquid interfaces. The result is an approach that blends prevailing scientific methods with data-science tools to produce physically-consistent models and novel conceptual knowledge.
10:20 AM Invited
A Data-driven Approach to Long-Time Molecular Dynamics: Danny Perez1; Nithin Mathew1; Enrique Martinez1; 1Los Alamos National Laboratory
One of the most stringent limitation of conventional molecular dynamics is the very short timescales that can be simulated, even when massively-parallel resources are leveraged. In contrast, the Parallel Replica Dynamics (ParRep) method allows for an effective parallelization of the problem in the time-domain, potentially allowing for millisecond-long simulations. However, in order use ParRep, one needs to identify long-lived metastable states and estimate of the time required to relax to local equilibrium within each state. Both these problems can be readily addressed on smooth energy landscapes typical of hard materials, but become extremely challenging for complex, soft, systems.We show how both these problems can be tackled simultaneously using modern data analysis techniques. In this method, states are defined and characterized on-the-fly by analyzing the trajectories produced by the ParRep simulation itself. We demonstrate the new approach through a variety of examples, including obstacle bypassing by dislocation in metals.
10:50 AM Invited
Dendritic Growth Prediction in Metal Additive Manufacturing with Physics-constrained Neural Networks: Dehao Liu1; Yan Wang1; 1Georgia Institute of Technology
The lack of complete process-structure-property (P-S-P) relationships for metal additive manufacturing is still the bottleneck in producing defect-free, structurally sound, and reliable parts. To alleviate the curse of dimensionality in constructing P-S-P relationships, a physics- constrained machine learning approach is proposed to construct surrogates in a high-dimensional parameter space with reduced amount of training data. In this work, the physics-constrained neural network with the minimax architecture (PCNN-MM) is developed to predict the dendritic growth of Ti-6Al-4V alloy during the rapid solidification process. The training of the PCNN-MM is to solve a minimax problem by searching the saddle points of the objective function with a Dual-Dimer saddle point search algorithm. The results show that the PCNN-MM can provide fast online predictions of phase field, composition field, and temperature field after offline training. The PCNN-MM has the potential of accelerating materials design and process design where simulation and experimental data are sparse.