Stochastic optimization of conventional and holey double-clad fibres

摘要

Stochastic optimization is proposed for the pump absorption enhancement in double-clad fibres (DCFs). Monte Carlo forms of the Nelder–Mead (NM) method are proposed and benchmarked against a genetic algorithm (GA) and generalized pattern search (GPS) and mesh adaptive direct search (MADS) methods. A simple sampling NM variant offered the highest efficiency while MADS achieved global convergence with the highest probability in low dimensions. All algorithms examined are described within the pattern search frame while detailed pseudocode descriptions are given in structured functions. Smoothly perturbed patterns are proposed for implicitly constrained optimization in high dimensions. An improved Monte Carlo NM method, based on importance sampling and stochastically perturbed patterns, exhibited global convergence with higher efficiency than MADS in high dimensions. Many polymer and air outer cladding designs are optimized. A circular holey DCF with an optimally offset core and a single large solid-state hole with optimum offset, size and ellipticity offered 180% absorption enhancement over a conventional circular DCF with centred core.