19–21 Oct 2026
Frascati
UTC timezone
The deadline for the abstracts submission has been delayed to July 20, 2026

Plasma density diagnostics and machine-learning jitter analysis for stable PWFA operation at SPARC_LAB

21 Oct 2026, 10:50
10m
Sala Bruno Brunelli (Frascati)

Sala Bruno Brunelli

Frascati

Building F23 Via Enrico Fermi 45, 00044 Frascati, Rome

Speaker

Romain Demitra (INFN_-LNF)

Description

Plasma-wakefield accelerators sustain GV/m-scale gradients, but shot-to-shot reproducibility remains the main obstacle to user-oriented operation. Diagnosing and controlling this instability requires measuring the plasma itself, not just the beam.
At the SPARC_LAB plasma-based accelerator facility at INFN-LNF, Stark broadening spectroscopy of the Hβ line provided shot-resolved plasma density measurements during live acceleration runs. Laser-triggered discharge operation reduced discharge fluctuations from ~10% to 1%, cutting empty shots from ~45% to 0.5% and witness-energy jitter by a factor of five.
Residual jitter was then studied on a dataset of ~941 shots across six working points and 19 machine features, combining Stark-derived density with upstream beam diagnostics. A heteroscedastic neural network quantifies predictive uncertainty shot-by-shot, while permutation importance and conditional mutual information isolate the contribution of each feature. Injected charge is the dominant driver of witness-energy fluctuations. Witness energy-spread jitter, by contrast, cannot be predicted from the measured machine state, setting a quantitative bound on what diagnostics and closed-loop controls can realistically achieve.

Authors

Angelo Biagioni (INFN-LNF) Lucio Crincoli (Istituto Nazionale di Fisica Nucleare) Massimo Ferrario (INFN-LNF) Riccardo Pompili (INFN-LNF) Romain Demitra (INFN_-LNF)

Presentation materials