Python Tools for Structure-Informed Prediction of Formation Energy using Neural Networks — pySIPFENN (March 2, 2023)¶
Adam M. Krajewski (Pennsylvania State University)
Richard Otis (Materials Genome Foundation)
Zi-Kui Liu (Pennsylvania State University)
Where: Virtual (Zoom)
When: 12 pm - 1:30 pm EST (6 - 7:30 pm CET / 9 - 10:30 am PST) on March 2, 2023
Registration deadline: February 28, 2023
This workshop will provide an introduction to the pySIPFENN tool for structure-informed prediction of formation energy. It will include short introductory lectures followed by hands-on demonstrations in an interactive cloud environment. In the end, all attendees will be able to (1) install the software on their machine, (2) run all default neural network models on a set of structure files (e.g., POSCAR), and (3) incorporate pySIPFENN into their research as a surrogate for density-functional theory (DFT) calculation.
Recently Machine Learning (ML) is becoming an increasingly important tool for material discovery, thanks to its ability to predict results of time and cost-intensive calculations like first-principles calculations based on the density functional theory (DFT), allowing researchers to guide them efficiently. The default pySIPFENN models shipped with the software and included in this workshop are trained on large (hundreds of thousands) open DFT datasets using structure-informed features to predict the formation energy of an arbitrary atomic configuration.
Workshop attendees will gain the ability to perform formation energy predictions on large sets of atomic structures with high accuracy and low computational cost. After the workshop, they will be able to independently install pySIPFENN, fetch models from repositories, and run them on both personal machines and high-performance computers (HPC), as needed. Advanced users will also be able to set up pySIPFENN for parallel computation and utilize customized models. All attendees will also gain helpful insight into manipulating crystal structures using the pymatgen and Spglib libraries.
Adam M. Krajewski is the lead developer of pySIPFENN and a 4th year PhD student working with Prof. Zi-Kui Liu at The Pennsylvania State University. Adam's research focuses on a wide variety of databases and software tools for computational thermodynamics. Prof. Liu has over 20 years of experience in the prediction of fundamental thermodynamic properties such as formation enthalpy, and for the last four years, works with Adam on machine learning (ML) surrogates to guide experiments and more expensive ab-initio methods.
- GitHub repository: https://github.com/PhasesResearchLab/pySIPFENN
- Journal article: Adam M. Krajewski, Jonathan W. Siegel, Jinchao Xu, Zi-Kui Liu, "Extensible Structure-Informed Prediction of Formation Energy with improved accuracy and usability employing neural networks", Computational Materials Science, Volume 208, 2022, 111254 (https://doi.org/10.1016/j.commatsci.2022.111254) (preprint: https://arxiv.org/abs/2008.13654)
How to register¶
Supported by the National Science Foundation POSE project “A Path to Sustaining a New Open-Source Ecosystem for Materials Science (OSEMatS)”, Materials Genome Foundation in collaboration with Penn State is pleased to offer this workshop free of charge, with cloud computing resources provided by IBM Cloud.
Interested graduate students, postdoctoral or early-career researchers, and other enthusiasts are encouraged to register prior to the deadline on February 28, 2023.
Please be sure to use a valid e-mail address, as this is how the organizers will communicate with you regarding workshop logistics.
All times are in Eastern Standard Time.
Day 1 (March 2, 2023)¶
- 12:00pm Opening
- 12:05 Introduction to data science for materials research
- 12:20 pySIPFENN Introduction
- 12:30 pySIPFENN Demonstration using Jupyter
- 1:15 Q&A
- 1:30 Closing / Discussion Room