In-situ characterization of few-fs laser pulses by learning from first-principles calculations

The field of ultrafast spectroscopy is based on lasers being able to produce pulses that are as short as a few femtoseconds. Due to their broad bandwidth, these ultrashort light transients are strongly affected by propagation through materials. Therefore, a careful characterization of their temporal profile is required before any application. We propose a scheme for their characterization in-situ, ensuring that the pulse parameters are measured in the region where the interaction with the sample takes place. Our method is based on first-principles calculations for strong-field ionization of rare-gas atoms and autocorrelation. We introduce a machine-learning algorithm (VSNIC) that uses the results from the first-principles calculations as input and reconstructs from a strong-field autocorrelation pattern for an unknown pulse the pulse length and spectral width by narrow margins.