Parallel variational quantum algorithms with gradient-informed restart to speed up optimisation in the presence of barren plateaus

Inspired by the Fleming-Viot stochastic process, we propose a parallel implementation of variational quantum algorithms with the aim of reducing the time spent by the algorithm in barren plateaus, where optimization direction is unclear. In the Fleming-Viot tradition, parallel searches are called particles. In the proposed approach, the search by a Fleming-Viot particle is stopped when it encounters a region where the gradient is too small or noisy, suggesting a barren plateau area. The stopped particle continues the search after being regenerated at another location of the parameter space, potentially taking the exploration away from barren plateaus. We first analyze the behavior of the Fleming-Viot particles from a theoretical standpoint. We show that, when simulated annealing optimizers are used as particles, the Fleming-Viot system is expected to find the global optimum faster than a single simulated annealing optimizer, with a relative efficiency that increases proportionally to the percentage of barren plateaus in the domain. This result is corroborated by numerical experiments carried out on synthetic problems as well as on instances of the Max-Cut problem, which show that our method performs better than plain simulated annealing when large barren plateaus are present in the domain.