BosonSampling with realistic single-photon sources

Summary form only given. The extended Church-Turing thesis posits that any computable function can be calculated efficiently by a probabilistic Turing machine. If this thesis held true, the global effort to build quantum computers might ultimately be unnecessary. The thesis would however be strongly contradicted by a physical device that efficiently performs a task believed to be intractable for classical computers. BosonSampling - the sampling from a distribution of n photons undergoing some linear-optical process - is a recently developed, experimentally accessible example of such a task [1].Here we report an experimental verification of one key assumption of BosonSampling: that multi-photon interference amplitudes are given by the permanents of submatrices of a larger unitary describing the photonic circuit. We built a tunable photonic circuit consisting of a central 3x3 fiber beamsplitter (Fig. 1) and exploited orthogonal polarization modes to extend the network to 6x6 modes. We developed a direct characterization method [2] to obtain the unitary description of this network and compared theoretical interference patterns predicted from this unitary with an experimental signature obtained via non-classical interference of three single photons [3].Our results show good agreement with theory, and we can rule out an explanation of the observed interference via classical means. We conclude that small-scale BosonSampling can be performed in the presence of unavoidable optical loss, imperfect photon sources, and inefficient detection [3].To reach a regime of 20 to 30 photons, where BosonSampling experiments are expected to start outperforming modern computers, we need to precisely quantify the contributions of realistic noise created in current photon sources. To this end, we modeled the detrimental effects of spectral and photon-number impurity [4] of independently generated photon pairs on the expected multi-photon interference patterns. We tested our model by fully - apping out three-photon interference as a function of individual temporal delays. While a full-scale demonstration is still out of reach, our results promise that a scaling-up of BosonSampling to single-photon numbers reached in state-of-the-art quantum optics experiments is feasible.