Turn-key constrained parameter space exploration for particle accelerators using Bayesian active learning.

Area of Science:

• Physical Sciences
• Chemical Sciences
• Biological Sciences

Background:

• Particle accelerators are crucial for scientific discovery.
• Characterizing beam response to input parameters is essential for experiments.
• Current methods like parameter scans are inefficient for complex, high-dimensional spaces.

Purpose of the Study:

• To adapt Bayesian optimization for efficient exploration of accelerator input parameter spaces.
• To minimize the need for prior knowledge of beam behavior and measurement constraints.
• To enable autonomous, adaptive, and cost-effective experimental characterization.

Main Methods:

• Adaptation of a popular Bayesian optimization algorithm.
• Autonomous exploration of multi-dimensional input parameter spaces.
• Integration with single-shot, constrained beam phase-space measurements.

Main Results:

• The algorithm enables turn-key exploration of parameter spaces, replacing traditional scans.
• Experimental demonstration of orders-of-magnitude speed improvement over conventional methods.
• Successful autonomous exploration with minimal prior information and complex constraints.

Conclusions:

• The developed Bayesian optimization approach significantly enhances the efficiency of accelerator-based experiments.
• This method offers a powerful tool for autonomous experimental design across various scientific disciplines.
• It addresses limitations of traditional parameter sampling in complex research environments.