Efficient classical algorithms for linear optical circuits

Abstract: We present efficient classical algorithms to approximate expectation values and probability amplitudes in linear optical circuits. Specifically, our classical algorithm efficiently approximates the expectation values of observables in linear optical circuits for arbitrary product input states within an additive error under a mild condition. This result suggests that certain applications of linear optical circuits relying on expectation value estimation, such as photonic variational algorithms, may face challenges in achieving quantum advantage. In addition, the (marginal) output probabilities of boson sampling with arbitrary product input states can be efficiently approximated using our algorithm, implying that boson sampling can be efficiently simulated if its output probability distribution is polynomially sparse. Moreover, our method generalizes Gurvits's algorithm, originally designed to approximate the permanent, to also approximate the hafnian of complex symmetric matrices with an additive error. The algorithm also solves a molecular vibronic spectra problem for arbitrary product input states as precisely as boson samplers. Finally, our method extends to near-Clifford circuits, enabling the classical approximation of their expectation values of any observables and (marginal) output probabilities.