Dartmouth Events

Physics & Astronomy - Senior Honor Thesis - Timothy Strang, Dartmouth College

Title: "Neural Nets for Hartree-Fock"

Friday, May 27, 2022
2:00pm – 1:00pm
Silsby Hall 28
Intended Audience(s): Public
Categories: Lectures & Seminars

Abstract: Hartree-Fock is one of the cornerstone numerical methods used to approximate the ground-state quantum wavefunctions of atomic many-body systems, utilizing several physical approximations before iteratively refining an ansatz function to achieve minimal energy. Multiple traditionally coded programs exist to carry out these calculations, but convergence to a global minimum is not guaranteed and run times scale poorly with system size. One solution may be to use a quantum computer to simulate quantum many-body systems. As part of the Whitfield Group, one long-term goal is to identify and benchmark areas such as these that may present opportunities for a quantum speedup. However, to conclusively identify a quantum computational advantage, all possible methods of classical computation must be shown to be less efficient than the quantum calculations. The goal of this thesis is to probe several neural net based approaches to the Hartree-Fock algorithm. In addition to laying the groundwork for future quantum computational benchmarks, all methodologies employed are assessed for immediate practical applicability.

A variety of neural nets were trained to reproduce the results of traditionally-coded techniques, then checked for fidelity and speedups. In the end, the project time frame did not permit research into nearly all relevant avenues of interest. However, a net was successfully trained to predict traditionally-calculated Hartree-Fock energies on a database of real, small molecules with high fidelity and drastically reduced CPU time. Additional nets were trained on a database of randomly generated molecules to predict Hartree-Fock and CISD energies, yielding reasonable fidelities and a similarly drastic speedup, at the front-end expense of a lengthy training process. Overall, these results should be viewed as a promising initial probe into the much broader problem space.

*************************************************************************************************************

Join Zoom Meeting
https://dartmouth.zoom.us/j/99643713340?pwd=b09yOUNtVEpldjFoVDJsQlZUdE9hUT09

Meeting ID: 996 4371 3340
Passcode: Strang

For more information, contact:
Tressena Manning
603-646-2854

Events are free and open to the public unless otherwise noted.