Stochastic optimization problems are powerful models for the operation of systems under uncertainty and are in general computationally intensive to solve. Two-stage stochastic optimization is one such problem, where the objective function involves calculating the expected cost of future decisions to inform the best decision in the present. In general, even approximating this expectation value is a #P-Hard problem. We provide a quantum algorithm to estimate the expected value function in a two-stage stochastic optimization problem in time complexity largely independent from the complexity of the random variable. Our algorithm works in two steps: (1) By representing the random variable in a register of qubits and using this register as control logic for a cost Hamiltonian acting on the primary system of qubits, we use the quantum alternating operator ansatz (QAOA) with operator angles following an annealing schedule to converge to the minimal decision for each scenario in parallel. (2) We then use Quantum Amplitude Estimation (QAE) to approximate the expected value function of the per-scenario optimized wavefunction. We show that the annealing time of the procedure in (1) is independent of the number of scenarios in the probability distribution. Additionally, estimation error in QAE converges inverse-linear in the number of "repetitions" of the algorithm, as opposed to converging as the inverse of the square root in traditional Monte Carlo sampling. Because both QAOA and QAE are expected to have polynomial advantage over their classical computing counterparts, we expect our algorithm to have polynomial advantage over classical methods to compute the expected value function. We implement our algorithms for a simple optimization problem inspired by operating the power grid with renewable generation and uncertainty in the weather, and give numerical evidence to support our arguments.
Comment: 20 pages, 5 figures