https://wse.zoom.us/j/91752450849
Meeting ID: 917 5245 0849 | Passcode: 605594
“Hybrid Physics and Machine Learning Framework: Achieving High Fidelity Modeling While Reducing the Computational Cost for First Principle Density Functional Theory Calculations”
Presented by Alhassan S. Yasin, Ph.D.
Research Scientist and Senior Professional Staff, Johns Hopkins University Applied Physics Laboratory
Lecturer, Department of Computer Science, Johns Hopkins University Engineering for Professionals
During the past two decades, first-principle calculations based on density-functional theory (DFT) unfolded as a successful approach to solve the electronic structure of matter. DFT is a widely used computational quantum mechanical modeling method that helps investigate the electronic structure and properties of many-body systems. The theory can reduce the many-body Schrödinger equation to an effective single-electron problem by relying on the Hohenberg-Kohn theorem and Kohn-Sham method, thus making material property predictions computationally feasible. The renowned success of DFT for describing ground-state properties for vast classes of materials such as semiconductors, insulators, half metals, semimetals, transition metals, etc., at the nanostructure scale makes it one of the most used methods for modern electronic structure analyses. Due to the extreme computational costs of most theoretical studies, limitations can and do arise when using approximation methods because accuracy is compromised in exchange for speedup time.
In this talk, I will describe a novel machine learning-based technique to reduce the computational cost required to explore large design spaces of substitutional alloys. The first advancement is based on a neural network (NN) approach to predict the initial position of minority and majority ions prior to DFT relaxation. The second advancement is to allow the NN to predict the total energy for every possible minority ion position and select the most stable configuration in the absence of relaxing each trial minority configuration. The third advancement is to use the machine learning approach to make future predictions for candidate ions. A bismuth oxide materials system, (BixLayYbz)2MoO6, was used as the model system to demonstrate the developed methods and quantify the resulting computational speedup. Compared to a brute force method that requires the calculation of every permutation of minority configuration and subsequent DFT relaxation, a significant speedup was realized if the NN predicted the initial configuration of ions prior to relaxation. Implementation of the second advancement allowed the NN to predict the total energy for all possible trial configurations and down select the most stable configurations prior to relaxation. Finally, the third advancement allowed the NN to predict thermodynamic reaction barriers during nitrogen fixation to further reduce the computational cost and make future predictions for candidate ions. Validation was done by comparing the NN and DFT predictions for the position, energy, and reaction pathways. The presentation will conclude with a brief discussion of current and future research within the different projects and groups I am privileged to be part of.
Dr. Alhassan S. Yasin is a research scientist at Johns Hopkins University (JHU) Applied Physics Laboratory (APL) and a Lecturer in the Department of Computer Science at JHU Engineering for Professionals. He received his graduate degrees in physics and engineering in 2018, where he studied and developed the physics and machine learning framework described during this presentation. Some of his current research span topics related to quantum computing, artificial intelligence, machine learning, hypersonics, theoretical and computational physics, and mechanical engineering. His current research involves developing hybrid machine learning and physics modeling frameworks, developing and evaluating algorithms for quantum computing, DFT analysis, resilience modeling and analysis, high throughput analysis, and developing biologically inspired approaches to resolve complex systems. Some of his research looks at developing algorithms that allow computer programs to automatically improve through experience. He also has an interest in developing the science and engineering of making computer systems behave in ways that until recently was thought to require human intelligence. He is part of the diverse teams that are driving cutting edge work in research and development at JHU-APL. His research aims to leverage and understand how multidisciplinary perspectives can further advance novel approaches to system design and analysis.