Bayesian optimization of resonant dispersive wave generation in hollow capillary fibers

Abstract

Resonant dispersive wave (RDW) generation in hollow capillary fibers (HCFs) is a powerful technique for producing ultrashort light pulses in the deep ultraviolet range, which are important for ultrafast spectroscopy and material processing. However, the complex nonlinear dynamics governing this process and the large associated parameter space make it challenging to achieve optimal RDW pulses with the highest peak power. In this study, Bayesian optimization (BO) is coupled with the open source Luna.jl simulation framework to optimize the HCF and pump pulse paramters for less than 5 femtosecond (fs) RDW generation at a target wavelength of 200 nm. Temporally non-structured RDW were consistently identified with peak powers of up to 14 GW, exceeding experimentally published values by up to 70 %. Furthermore, a subset of the RDW optima exhibited an energy stability that is better than that of the pump pulse. Given that this approach can be generalized to other RDW wavelengths, our findings suggest that BO is a valuable tool in developing HCF systems that support RDW generation tailored to a particular experimental need.