Artificial intelligence for online characterization of ultrashort X-ray free-electron laser pulses - Scientific Reports

X-ray free-electron lasers (XFELs) are the world's fastest X-ray cameras, providing ultrashort exposure times in combination with a spatial resolution limit down to the sub-nanometer range, which allows for time-resolved experiments'freezing' the motion of atoms and molecules. In fact, XFELs have revolutionized several fields of science enabling us to observe the role of transient structures and resonances in atoms1 as well as single-molecule or cluster imaging2, investigations of ultrafast processes at element-specific observer sites3, and the study of nonlinear light–matter interaction in the X-ray regime4. Over the past decade, further development of the underlying machine operation techniques has enabled increasingly sophisticated control over the photon pulse parameters. One of the most recent major upgrades is the increased repetition rate of XFELs that is anticipated to initiate a leap from proof-of-principle experiments to advanced applications of interdisciplinary importance, thus representing a cornerstone of modern XFEL science5. Most of the FELs and in fact all XFELs worldwide are currently based on the principle of self-amplification of spontaneous emission (SASE)6. More precisely, their pulses are formed stochastically through the interplay between the relativistically accelerated electron bunches themselves and the spontaneously emitted synchrotron radiation, caused by their sinusoidal trajectories inside magnetic structures with periodically changing polarity, so-called undulators.