Speaker
Description
At LCLS, attosecond xFEL pulses (“XLEAP” mode) allow users to probe electronic process on their natural timescale.* However, XLEAP is much more sensitive than conventional SASE operation to drifts and jitter in the RF phases, laser timing, bunch charge, and beam orbit. As a result, XLEAP setup and pulse energy optimization is highly variable and difficult to replicate even minute-to-minute, leading to delays in user experiments and unpredictable time commitments for operators.
In this work, we explore archival data to identify sources of variation driving XLEAP performance. This allows us to frame XLEAP as an optimization problem in high-dimensional control space with conflicting objectives, few diagnostics, and time dependence on scales from the sampling rate to minutes and hours. We consider how machine learning techniques can be used to effectively tune the machine in this complicated landscape.
Funding Agency
Supported by the Basic Energy Sciences Accelerator and Detector Research program under Contract No. DE-AC02-76SF00515, and the Office of Basic Energy Science Accelerator and Detector Research Program.
Footnotes
*Duris, J., Li, S., Driver, T. et al. Tunable isolated attosecond X-ray pulses with gigawatt peak power from a free-electron laser. Nat. Photonics 14, 30–36 (2020).
| I have read and accept the Privacy Policy Statement | Yes |
|---|