The attainment of a satisfactory operating point is one of the main problems in the tuning of particle accelerators. These are extremely complex facilities, characterized by the absence of a model that accurately describes their dynamics, and by an often persistent noise which, along with machine drifts, affects their behaviour in unpredictable ways. In this paper, we propose an online iterative Linear Quadratic Regulator (iLQR) approach to tackle this problem on the FERMI free-electron laser of Elettra Sincrotrone Trieste. It consists of a model identification performed by a neural network trained on data collected from the real facility, followed by the application of the iLQR in a Model-Predictive Control fashion. We perform several experiments, training the neural network with increasing amount of data, in order to understand what level of model accuracy is needed to accomplish the task. We empirically show that the online iLQR results, on average, in fewer steps than a simple gradient ascent (GA), and requires a less accurate neural network to achieve the goal.

This study describes the charge‐state evolution of relativistic lead ions passing through a thin aluminum stripper foil. It is motivated by the Gamma Factory project at CERN, where optical laser pulses will be converted into intense gamma‐ray beams with energies up to a few hundred MeV, achieved via excitation of atomic transitions in few‐electron heavy‐ions at highly relativistic velocities. The recently developed BREIT code is employed together with theoretical cross‐sections for single‐electron loss and capture of the projectile ions. All charge‐states starting from Pb^{54 +} up to bare ions are considered at kinetic projectile energies of 4.2 and 5.9 GeV u⁻¹. During recent Gamma Factory beam tests, the calculated preparatory predictions allow to effectively produce Pb^{80 +} and Pb^{81 +} ions from Pb^{54 +} in the transfer beam line between the PS and SPS synchrotron accelerators at CERN and consequently, to store partially stripped ions in the LHC for the very first time. Reasonable agreement is found between the calculations and the very few experimental data available. The study lays the groundwork to optimize the yields of charge states of interest for experiments within the future Gamma Factory project, including the upcoming Gamma Factory Proof‐of‐Principle experiment for which predictions for the production of Pb^{79 +} are presented.

Automated accelerator tuning at GSI and FAIR is achieved using the Geoff framework. Geoff provides real-time optimization of beam parameters and experimental setups, supporting fast deployment and control room integration. It significantly improves performance, reducing SIS18 synchrotron injection losses from 45% to 12% and speeding up fragment separator setup using a classification algorithm. Advanced optimization strategies, such as multi-objective and multi-fidelity Bayesian optimization, were applied to SIS18 injection tuning, while model-predictive control implemented via model-driven reinforcement learning allows fast, constraint-aware adaptation. Using dedicated ion-source setups of the PUMA experiment at TU Darmstadt, automated control and optimization algorithms were used to optimize a hot-cathode electron source and a multi-reflection time-of-flight mass spectrometer, demonstrating the feasibility of real-time tuning of ion sources and particle traps. Geoff’s modular design facilitates easy integration of classical and machine-learning-based algorithms, bridging traditional accelerator operations with modern data-driven optimization.