Speaker
Description
The evolution of high peak power laser technology especially in repetition rate and pulse-to-pulse stability, as well as the development of the suitable targetry enabled LPIAs to operate quasi-continuously, at a repetition rate of around 1 Hz [1-2]. This allowed the use of optimisation algorithms based on deep learning to improve the performance of LPIA. One of the most commonly integrated algorithms is the Bayesian, where a surrogate model is created. The model is refined by sampling the parameter space, which makes it possible to predict the performance and may help to explore the physics. This optimisation method has been demonstrated to be effective in enhancing the performance of up to 2 parameters (for example wavefront control [3-5] or spectral phase [6]) and a few hundred samples in the parameter space. Besides, it performs poorly with higher sample number and optimisation for more than a very few parameters is prohibitively time consuming.
State-of-the-art ion accelerators driven by kHz repetition rate laser system open new opportunities for optimisation processes due to the larger amount of data. The Light Energy Ion Accelerator (LEIA) beamline in ELI-ALPS is driven by the Sylos3 laser system. Pulses with 8.5 fs duration and 80 mJ energy on target accelerate protons and deuterons up to 2.5 MeV cut-off energies. The system has been demonstrated to be capable of producing low-energy ions with a kHz repetition rate, utilising a thin liquid sheet target, resulting in a laser accelerated ion beam with an average power close to 10 W [7]. The LEIA beamline is equipped with two Thomson Parabola Spectrometers (TPS), which provide real-time information about the ion beam and the acceleration performance. These properties led to the use of another optimisation algorithm, namely the Markov-Chain Monte Carlo (MCMC) method. MCMC algorithm is able to manage a substantially larger number of free parameters, performs well with large number of samples, and last but not least natively converges to the optimal parameters.
To prove the feasibility of the MCMC algorithm, we have optimised the performance of the ion accelerator with the use different liquids and thicknesses. The spectral phase of the laser pulses was optimized real-time to reach the highest proton/deuteron cut-off or bunch energy at a given target material and thickness. Additionally, due to the tunability of the MCMC model we were able to scan the parameter space simultaneously, revealing the real distribution of the appointed parameters (cut-off or bunch energy) as a function of the phase derivatives. For adequate optimisation and result, 900-1000 sample points in the parameter space were necessitated, which took 15 minutes at most cases, made it possible to find the optimal parameters for different applications just before the real experiment.
References:
[1] Lelievre et al., Phys. Plasmas 31, 093106 (2024)
[2] Streeter et al., Nat Commun 16, 1004 (2025)
[3] B. Loughran et al., HPLSE, 11, e35 (2023).
[4] Catrix, E. et al., Appl. Phys. Lett. 126, 254104 (2025)
[5] Glenn et al., Phys. Rev. Research 8, 013101 (2026)
[6] Torrance et al., HPLSE 13, e105 (2025)
[7] Osvay et al, A laser-plasma accelerator with a high average power ion beam for applications, HPLSE Conference, Chengdu, China, 2026.
| Working group | WG2 |
|---|