Phase Space Reconstruction from Accelerator Beam Measurements Using Neural Networks and Differentiable Simulations

Area of Science:

• Accelerator physics
• Computational physics
• Data analysis

Background:

• Accurate characterization of particle beam phase space is crucial for accelerator performance.
• Traditional methods struggle with high-dimensional distributions or require specialized equipment.
• Inferring beam properties often involves simplifying assumptions or complex diagnostics.

Purpose of the Study:

• To develop a general-purpose algorithm for efficient reconstruction of high-dimensional particle beam phase space distributions.
• To overcome limitations of conventional methods in characterizing beam dynamics.
• To enable accurate measurement of correlated phase spaces without specialized diagnostics.

Main Methods:

• Integration of neural networks with differentiable particle tracking.
• Development of a general-purpose algorithm for phase space reconstruction.
• Utilizing limited measurements from a single focusing quadrupole and diagnostic screen.

Main Results:

• Accurate reconstruction of detailed 4D phase space distributions.
• Successful demonstration in both simulation and experimental data.
• Provision of confidence intervals for the reconstructed distributions.
• Capability to measure multiple correlated phase spaces simultaneously.

Conclusions:

• The novel algorithm efficiently reconstructs high-dimensional phase space distributions.
• This technique simplifies the measurement of beam properties without specialized diagnostics.
• The method paves the way for future simplified 6D phase space distribution reconstructions.