Optimization in optical systems revisited: Beyond genetic algorithms

Designing integrated photonic devices such as waveguides, beam-splitters and beam-shapers often requires optimization of a cost function over a large solution space [1]. Metaheuristics -- algorithms based on empirical rules for exploring the solution space -- are specifically tailored to those problems. One of the most widely used metaheuristics is the standard genetic algorithm (SGA), based on the evolution of a population of candidate solutions. However, the stochastic nature of the SGA sometimes prevents access to the optimal solution. Our goal is to show that a parallel tabu search (PTS) algorithm is more suited to optimization problems in general, and to photonics in particular. PTS is based on several search processes using a pool of diversified initial solutions. To assess the performance of both algorithms (SGA and PTS), we consider an integrated photonics design problem, the generation of \emph{arbitrary} beam profiles using a two-dimensional waveguide-based dielectric structure [2].\\[4pt] [1] A. Vukovic, P. Sewell, and T. M. Benson, J. Opt. Soc. Am. A \textbf{27} (2010), no.~10, 2156--2168. \newline [2] D. Gagnon, J. Dumont, and L. J. Dub\'{e}, J. Opt. Soc. Am. A \textbf{29} (2012), no.~12, 2673--2678.