Performance analysis of derandomized evolution strategies in quantum control experiments

Genetic Algorithms (GAs) are historically the most commonly used optimization method in Quantum Control (QC) experiments. We transfer specific Derandomized Evolution Strategies (DES) that have performed well on noise-free theoretical Quantum Control calculations, including the Covariance Matrix Adaptation (CMA-ES) algorithm, into the noisy environment of Quantum Control experiments. We study the performance of these DES variants in laboratory experiments, and reveal the underlying strategy dynamics of first- versus second-order landscape information. It is experimentally observed that global maxima of the given QC landscapes are located when only first-order information is used during the search. We report on the disruptive effects to which DES are exposed in these experiments, and study covariance matrix learning in noisy versus noise-free environments. Finally, we examine the characteristic behavior of the algorithms on the given landscapes, and draw some conclusions regarding the use of DES in QC laboratory experiments.