Supervisors: Manuel Kirchen (DESY) & Kevin Cassou (IJCLab)
DESY, a world-leading research center with over 2900 employees at its Hamburg and Zeuthen sites, is dedicated to uncovering the structure and function of matter—from fundamental particles to the building blocks of life. In addressing key scientific, societal, and industrial challenges, DESY pioneers research that bridges fundamental physics and practical applications.
Plasma accelerators offer a disruptive technology for generating compact, high-energy electron beams with potential uses in science, healthcare, and industry. The plasma acceleration group at DESY—comprising more than 50 researchers, engineers, and students—is advancing next-generation laser- and beam-driven plasma accelerators.
Our flagship project, KALDERA, focuses on overcoming the limitations of current laser-plasma acceleration (LPA) techniques. While previous experiments have validated LPA as a promising alternative to conventional RF technology, the low repetition rates (typically a few Hz) of high-power lasers limit the average power and prevent the adoption of fast feedback systems necessary to control beam instabilities. KALDERA aims to change that by developing a laser system delivering 100 TW peak power at up to 1 kHz, enabling active stabilization and enhanced performance of LPA beams.
Working within the KALDERA project, you will employ data-driven methods to improve the quality and stability of laser-plasma accelerated beams. This includes the application of machine learning techniques (e.g., neural networks, Gaussian processes) to build surrogate models that reveal hidden correlations between laser and electron beam parameters. The models will be applied for autonomous optimization and active stabilization, targeting sub-percent level stability in critical properties like electron beam energy. You will also contribute to the design of advanced diagnostics, controls, and online processing tools for managing high-repetition-rate data streams.
Expected Results
- Assist in the development of advanced diagnostics and controls for KALDERA
- Develop efficient online processing and analysis tools for high-repetition-rate data streams
- Apply machine learning to build surrogate models from experimental and simulated LPA data
- Use the models for optimization and active stabilization of the laser-plasma accelerator
Planned secondments
- 6-month at ICJLab (distributed in 2–3 visits) to gain experience in an alternative lab environment and broaden the applicability of your machine learning tools
A high-intensity laser pulse generates a plasma wave that accelerates electron bunches. Utilizing machine learning, one can analyse and control the properties of these bunches. Credit: Science Communication Lab for DESY