IPAC'23 - 14th International Particle Accelerator Conference

Hybrid beamline element ML-training for surrogates in the impactX beam-dynamics code

WEPA101

10 May 2023, 16:30

2h

Salone Adriatico

Poster Presentation MC5.D13: Machine Learning Wednesday Poster Session

Speaker

Ryan Sandberg (Lawrence Berkeley National Laboratory)

Description

The modeling of current and next-generation particle accelerators is a complex endeavour, ranging from the simulation-guided exploration of advanced lattice elements, over design, to commissioning and operations.
This paper explores hybrid beamline modeling, towards coupling s-based particle-in-cell beam dynamics with machine-learning (ML) surrogate models.
As a first example, we train a surrogate model of an advanced accelerator element, a laser-wakefield accelerator stage, via the time-based particle-in-cell code WarpX [1].
A second example trains trains a model for the IOTA nonlinear lens via the s-based code ImpactX [2].

Funding Agency

Work supported by the Exascale Computing Project (17-SC-20-SC), a collaborative effort of the U.S. DOE Office of Science and the NNSA, and by LBNL LDRD under DOE Contract DE-AC02-05CH11231.

Footnotes

• L. Fedeli et al., SC22, 978-1-6654-5444-5, pp. 25-36 (2022)
  ** A. Huebl et al., NAPAC22, arXiv:2208.02382 (2022)