Algorithm Development in Materials Science and Engineering: Atomistic Simulations, Interatomic Potential, and Computer Science Models
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Functional Materials Division, TMS Structural Materials Division, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee, TMS: Phase Transformations Committee, TMS: Solidification Committee, TMS: Chemistry and Physics of Materials Committee
Program Organizers: Adrian Sabau, Oak Ridge National Laboratory; Ebrahim Asadi, University of Memphis; Enrique Martinez Saez, Clemson University; Garritt Tucker, Colorado School of Mines; Hojun Lim, Sandia National Laboratories; Vimal Ramanuj, Oak Ridge National Laboratory

Thursday 2:00 PM
March 23, 2023
Room: Cobalt 502B
Location: Hilton

Session Chair: Garritt Tucker, Colorado School of Mines; Vimal Ramanuj, Oak Ridge National Laboratory


2:00 PM  
EAM-X: Simple Parameterization of Embedded Atom Method Potentials for FCC Metals and Alloys: Murray Daw1; Michael Chandross2; 1Clemson University; 2Sandia National Laboratory
    We report on a simple parametric form for interatomic potentials of the Embedded Atom Method (EAM-X) for FCC metals and alloys. With this model, we deviate from the usual approach of fitting a set of functions to basic properties from experiments and/or DFT calculations, and then using those functions to investigate more complex properties. Instead, we illustrate here what we term the “inside out” approach, which seeks to understand generically how complex properties are dependent on the EAM-X parameters themselves. This method enables the identification of regions of parameter space that correspond to desirable attributes, and then possibly matching that neighborhood to real elements. Within the model we find useful the idea of “spread alloys”, where the constituent elements are defined as parametric perturbations from a central, “average” FCC metal, and where different alloys are quantified by a measure of the magnitude of the perturbation. Several applications are presented. [SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525 (SAND2022-1056 A).]

2:20 PM  
EAM-X: Universal trends in FCC Grain Boundary Energies: Yasir Mahmood1; Murray Daw1; Michael Chandross2; Fadi Abdeljawad1; 1Clemson University; 2Sandia National Laboratories
     Traditionally, inter-atomic potentials are developed using a complex set of functions fitted to a material’s basic properties and are used to explore complex properties. Recently, Daw & Chandross [1] introduced a simple, parametric functional form of the Embedded Atom Method for FCC metals (EAM-X) that allows for an inside-out approach where complex properties are explored within parameter space followed by mapping of real elements to the corresponding parameters. Here we explore FCC grain boundary (GB) properties using a representative set of symmetric tilt, asymmetric tilt and twist GBs. We observe that GB energies can be described as a smooth function of the EAM-X parameters and are also correlated with the shear moduli. This approach demonstrates the possibility of rapid exploration of GB properties for FCC metals.[1] M. S. Daw & M. Chandross, “Simple Parameterization of Embedded Atom Method Potentials for FCC Metals” (submitted).

2:40 PM  Invited
Enabling Long Timescale Molecular Dynamics Simulation with ab initio Precision: Jan Janssen1; Danny Perez1; 1Los Alamos National Laboratory
    Classical molecular dynamics (MD) is in principle an ideal tool to investigate the long-time evolution of materials, as ab initio-based MD simulations remain limited to very short time. While modern machine learning MD potentials report errors on the order 1 meV/atom, these errors are only typical of configurations that are similar to those found in the training set, and transferability remains limited. This poses a challenge to the accuracy of long-time MD simulations for two reasons: transition rates are exponentially sensitive to energy barriers, and saddle configurations form a very small subset of the whole configuration space and so are very unlikely to appear in traditional datasets. We propose an automated workflow to develop and validate transferable machine learning potentials for long-time simulations. Starting from an information-entropy optimized training set with over 7 million atomic environments, fitted potentials are benchmarked on a set of transition states to characterize their transferability.

3:00 PM  
Investigating Magnetic Phase Transitions with Ising Models Accounting for Long-range Spin Interactions: Ender Eger1; Arulmurugan Senthilnathan1; Mahmudul Hasan1; Pinar Acar1; 1Virginia Tech
    In the past, the Ising model was used to address the problem of ferromagnetic to paramagnetic phase transition. However, to reduce the computing cost, several assumptions were made. These assumptions include the consideration of only nearest neighbor spin interactions, and the exclusion of an external magnetic field and uncertainty effects. Furthermore, the Ising model is constructed without the presence of defects. The goal of this study is to use the mean field solution for exploring the magnetic phase transition problem using a higher order Ising model involving crystallographic defects in both 2D and 3D directions. The created model incorporates long-range interactions of spins while accounting for an external magnetic field and the effects of the uncertainty in the external field and temperature. The results of this work are expected to contribute to the robust design of magneto-mechanical materials used in extreme environments.

3:20 PM Break

3:40 PM  
Modular and Scalable Solutions for Training Machine Learned Interatomic Potentials: Mitchell Wood1; Andrew Rohskpof1; Charles Sievers2; Danny Perez3; Aidan Thompson1; 1Sandia National Laboratories; 2Boeing; 3Los Alamos National Lab
    Model development that utilizes machine learning must define feature sets, model forms, and ultimately if said models are efficient to use for the desired accuracy. Specialized software to develop machine learned interatomic potentials utilized in MD is highlighted herein. The FitSNAP code has evolved quickly in the last few years to accept numerous descriptor sets, model forms (and associated regression techniques), all while ensuring portability into LAMMPS for efficient use. This talk will overview the user friendly FitSNAP code and its integration into the Exascale Computing Project EXAALT software stack with a focus on the challenges and advances made to tackle exascale sized training sets needed to construct robust and truly transferable interatomic potentials. A new method of training set generation that is applicable beyond interatomic potentials will be demonstrated.

4:00 PM  Invited
PyEBSDIndex: Fast Indexing of EBSD data: David Rowenhorst1; Patrick Callahan1; Håkon Wiik Ånes2; 1The US Naval Research Laboraory; 2Norwegian University of Science and Technology
    There have been significant advances in the ability to collect and index electron back-scattered diffraction (EBSD) data, with collection speeds increasing an order of magnitude in the last decade, and advanced indexing method have greatly increased the accuracy and precision of indexing methods, but often require significant computational resources. Meanwhile new data collection modes, such as in-situ studies or 3D serial-sectioning methods would greatly benefit from fast indexing methods that can be tightly integrated into real-time analysis, something difficult with current commercial software packages. Here, we present a new indexing algorithm and software package that combines the traditional Hough/Radon-based methods and modified band voting with modern GPU processing. The current implementation indexes EBSD patterns >20,000 patterns/sec on typical desktop hardware, with indexing accuracy <0.1° misorientation. The algorithm is presented as a Python package with an open source license, making it available for the community to utilize and advance.

4:20 PM  
Training Machine-learned Interatomic Potentials for Chemical Complexity - Application to Refractory CCAs: Megan McCarthy1; Jacob Startt1; Remi Dingreville1; Mitchell Wood1; 1Sandia National Laboratories
    Though machine-learned interatomic potentials (MLIAPs) have greatly improved the accuracy of molecular dynamics, there is still much to be learned in training models for chemical complexity. One important example is found in complex concentrated alloys (CCAs), which contain high concentrations of three or more metallic elements. While excellent progress has been made in generating CCA MLIAPs for single compositions, far less is understood about creating generalized transferable potentials for a range of compositions. This capability is critical to accurate large-scale modeling of CCAs, as chemical complexity can result in large variability in local properties. In this talk, we discuss development of MLIAPs for MoNbTaTi refractory CCAs designed for cross-compositional modeling, using a spectral neighbor analysis potential (SNAP). We describe new ways of quantifying chemical diversity in CCA data sets and explore how it affects mechanical and local-ordering phenomena. SNL is managed and operated by NTESS under DOE NNSA contract DE-NA0003525.