Tailoring Laser-Generated Supercontinuum with Machine Learning
Rob HolcombUniversity of Rochester’s Laboratory for Laser Energetics
AbstractBroadband supercontinuum sources are widely used in spectroscopy, metrology, imaging, and as seed sources for multipetawatt laser systems. However, continuum generation arises from coupled nonlinear optical processes, making performance highly sensitive to input conditions and difficult to model and optimize. We generate a configurable supercontinuum in a solid-state material using tunable subpicosecond 1-µm pulses from a home-built Yb:YAG thin-disk regenerative amplifier with nonlinear postcompression in a gas-filled hollow-core fiber. Characterization with a photodiode-based spectral diagnostic reveals key continuum features, including spectral tilt and threshold-dependent stability regimes that correlate with side-view fluorescence imaging. We demonstrate a tandem neural network framework for inverse modeling of white-light continuum generation and show that it reproduces desired outputs and enables system optimization. These results provide insight into coupled nonlinear dynamics and highlight the potential of neural-network–based inverse design for stable, configurable supercontinuum sources.
Bio
Parking and locationThe talk will be held at UR River Campus, Goergen 101. Parking is available in the lot across the street in Intercampus Drive Lot, and is free for talk attendees (no pass needed).