PHYSICS DISSERTATION DEFENSE: Kareem Hegazy
Ph.D. Candidate: Kareem Hegazy
Research Advisors: Ryan Coffee and Phil Bucksbaum
Time: 3:00 pm
Location: PAB 102/103
Zoom Link: https://stanford.zoom.us/j/91709478270?
Zoom Password: email nickswan [at] stanford.edu for password
Title: Merging ultrafast gas-phase diffraction experiment, theory, and machine learning for a new look at molecular dynamics
Abstract: Understanding fundamental molecular dynamics from first principles is crucial in developing a more complete understanding of chemistry and chemical reactions. Chemical reactions are generally defined as a change in molecular structure. Ultrafast gas phase diffraction is a primary probe of such fundamental structural dynamics that provides sufficient spatio-temporal resolution to produce “molecular movies”. Although this measurement has been around for 50 years, the absence of a general methodology to retrieve molecular geometries from the data alone often requires us to employ complex ab initio molecular dynamics simulations to fully describe our measurement. Therefore, the capabilities of ultrafast gas-phase diffraction are often limited to the domain of dynamics that can be simulated, keeping much of the interesting and crucial molecular behavior out of reach.
In this defense, I introduce a novel analysis technique that removes this simulation barrier to retrieve the molecular geometry distribution (|Ψ(r, t)|2). This is accomplished through a statistical re-framing of the measurement and employing fundamental machine learning techniques. The result of which is the molecular distribution (|Ψ(r, t)|2) with an atomic pair-wise resolution of ~100 fm, an improvement by 100X over traditional approaches. This method’s greatest potential is to transform ultrafast gas-phase diffraction into a discovery-oriented technique capable of probing molecular dynamics that are intractable to current simulation methods.