Speaker
Description
Since the 80s, the BNL LINAC has delivered pulses of polarized protons for physics and proton beams for isotope production. Over the years, the LINAC has been steadily upgraded and optimized to ensure maximum output and efficiency. With over 300 independent quadrupoles and equally many rf cavities, there are obvious challenges with direct optimization especially due to the associated costs. Most tuning progress has been done with repetitive, careful, and incremental simulation of each sector using qualitative figures of merit. The historical success of this approach opens the door for the usage of constrained Bayesian optimization techniques on principal components of this large set of beam optics controls. By porting the existing simulation infrastructure into our differentiable physics ecosystem, we are able to qualitatively train a GP surrogate to predict optimal emittance transport.
Funding Agency
Work funded by U.S. DOE grant DE-SC0024287
| In which format do you inted to submit your paper? | LaTeX |
|---|