Ultrafast measurements using x-ray free-electron lasers

X-ray free-electron lasers (XFELs) are unique tools for ultrafast science, allowing one to generate tunable pairs of few-fs x-ray pulses with photon energies between ~10 eV and ~10 keV. XFEL machines are however much larger and much more complex than table-top optical lasers, spanning from several hundreds of metres in length up to a few kilometres. This increased size and complexity makes them very prone to fluctuations and single-shot jitter in the parameters of the generated x-ray pulses such as photon energy, intensity and time delay, of up to ~1%, ~100% and ~15 fs, respectively. Using data from different experiments carried out at the Linac Coherent Light Source (LCLS) and the SPring-8 Angstrom Compact Free Electron Laser (SACLA), the author will show examples of the influence that these fluctuations can have on the experimental signals, proposing solutions to separate this influence from actual meaningful experimental results. The thesis starts with a brief introduction to x-ray science and the working principles of XFELs followed by an extensive description of the common configurations that allow generating pairs of pulses for time-resolved experiments. The first experimental results are then presented for two successful experiments measuring nuclear dynamics in the 10-100 fs timescales in the acetylene and C60 molecules. While successful, these experiments already show the effects of the fluctuations, effects that, as shown in the following chapter, are amplified when attempting to use few-fs pulses to measure few-fs dynamics. Although the presence of the fluctuations may be unavoidable, this thesis shows, using data from three additional ultrafast experiments, that the fluctuations can be circumvented by performing a full single-shot characterization of all x-ray pulses and implementing sophisticated data analysis techniques to sort the experimental data. Finally, having exposed the importance of single-shot x-ray characterization, the author proposes and demonstrates a technique based on machine learning to enable such full x-ray characterization at the next generation of high-repetition-rate XFELs.Open Acces