Enhancing Neutron Yield in Cylindrical Target Designs Through Multi-Fidelity Bayesian Optimization

Mitigating hydrodynamic instability growth in inertial confinement fusion implosions is an area of active investigation. Recent efforts have leveraged cylindrical targets, which include the Bell-Plesset effects of convergence that magnify instability growth in imploding systems while maintaining a direct diagnostic line-of-sight.

Optimization techniques can be used to identify designs that minimize hydrodynamic instability and maximize yield. However, such design studies rely upon complex 1D and 2D radiation-hydrodynamics simulations, making optimization procedures expensive. While 1D simulations can significantly reduce costs, they also sacrifice accuracy when capturing implosion dynamics. To enhance the predictive modeling while maintaining low costs, we introduce a multi-fidelity constrained optimization algorithm which fuses data from 1D and 2D simulations in order to identify cylindrical target designs that maximize yield while minimizing instability growth. We compare the costs and predictive capability of the multi-fidelity model with a Gaussian Process surrogate model trained only on 2D data. The optimized design selected by the algorithm is discussed, emphasizing the design choices and implosion physics responsible for the target’s enhanced performance.