Speaker
Description
The design optimization of RFQ and DTL cavities involve expensive black-box simulations, difficult constraints coming from simulation failures and the absence of reliable derivative information. These characteristics make derivative-free optimization methods particularly interesting, yet their practical performance in a conjoint RFQ-DTL cavity design optimization remains insufficiently studied. This contribution presents a prototype optimization framework connecting the execution of established accelerator codes, automating post-processing and derivative-free optimization methods. This work discusses the performance of the derivative-free method MADS comparing it to commonly used Bayesian approaches in black-box optimization. The framework under development also aims to study an inverse-design formulation: instead of only evaluating a proposed design, we study a way for the optimizer to search for feasible input configurations that reproduce a selected target output while still trying to preserve the optimality of the point.
| I have read and accept the Privacy Policy Statement | Yes |
|---|