Probabilistic solving of NP -hard problems with bistable nonlinear optical networks

Abstract

We study theoretically a lattice of locally bistable driven-dissipative nonlinear cavities. The system is found to resemble the classical Ising model and enables its effective simulation. First, we benchmark the performance of driven-dissipative nonlinear cavities for spin-glass problems, and study the scaling of the ground-state-energy deviation and success probability as a function of system size. Next, we show how an effective bias field can be included in an optical model and use it for probabilistic solving of optimization problems. As particular examples we consider NP-hard problems embedded in the Ising model, namely graph partitioning and the knapsack problem. Finally, we confirm that locally bistable polariton networks act as classical optimizers and can potentially provide an improvement within the exponential complexity class.