We demonstrate the robustness of BosonSampling against imperfections in the linear optical network that cause a small deviation in the matrix it implements. We show that applying a noisy matrix $\stackrel{\ifmmode \tilde{}\else \~{}\fi{}}{U}$ that is within $\ensuremath{\epsilon}$ of the desired matrix $U$ in operator norm leads to an output distribution that is within $\ensuremath{\epsilon}n$ of the desired distribution in variation distance, where $n$ is the number of photons. This lets us derive a sufficient tolerance for beam splitters and phase shifters in the network. This result only concerns errors that result from the network encoding a different unitary than desired and not from other sources of noise such as photon loss and partial distinguishability.
We demonstrate the robustness of BosonSampling against imperfections in the linear optical network that cause a small deviation in the matrix it implements. We show that applying a noisy matrix U\ifmmode~\else˜\fi that is within \ensuremathϵ of the desired matrix U in operator norm leads to an output distribution that is within \ensuremathϵn of the desired distribution in variation distance, where n is the number of photons. This lets us derive a sufficient tolerance for beam splitters and phase shifters in the network. This result only concerns errors that result from the network encoding a different unitary than desired and not from other sources of noise such as photon loss and partial distinguishability.
BosonSampling is an intermediate model of quantum computation where linear-optical networks are used to solve sampling problems expected to be hard for classical computers. Since these devices are not expected to be universal for quantum computation, it remains an open question of whether any error-correction techniques can be applied to them, and thus it is important to investigate how robust the model is under natural experimental imperfections, such as losses and imperfect control of parameters. Here, we investigate the complexity of BosonSampling under photon losses, more specifically, the case where an unknown subset of the photons is randomly lost at the sources. We show that if k out of n photons are lost, then we cannot sample classically from a distribution that is $1/{n}^{\mathrm{\ensuremath{\Theta}}(k)}$ close (in total variation distance) to the ideal distribution, unless a BPPNP machine can estimate the permanents of Gaussian matrices in nO(k) time. In particular, if k is constant, this implies that simulating lossy BosonSampling is hard for a classical computer, under exactly the same complexity assumption used for the original lossless case.
BosonSampling, which we proposed three years ago, is a scheme for using linear-optical networks to solve sampling problems that appear to be intractable for a classical computer. In arXiv:1306.3995, Gogolin et al. claimed that even an ideal BosonSampling device's output would be "operationally indistinguishable" from a uniform random outcome, at least "without detailed a priori knowledge"; or at any rate, that telling the two apart might itself be a hard problem. We first answer these claims---explaining why the first is based on a definition of "a priori knowledge" so strange that, were it adopted, almost no quantum algorithm could be distinguished from a pure random-number source; while the second is neither new nor a practical obstacle to interesting BosonSampling experiments. However, we then go further, and address some interesting research questions inspired by Gogolin et al.'s mistaken arguments. We prove that, with high probability over a Haar-random matrix A, the BosonSampling distribution induced by A is far from the uniform distribution in total variation distance. More surprisingly, and directly counter to Gogolin et al., we give an efficient algorithm that distinguishes these two distributions with constant bias. Finally, we offer three "bonus" results about BosonSampling. First, we report an observation of Fernando Brandao: that one can efficiently sample a distribution that has large entropy and that's indistinguishable from a BosonSampling distribution by any circuit of fixed polynomial size. Second, we show that BosonSampling distributions can be efficiently distinguished from uniform even with photon losses and for general initial states. Third, we offer the simplest known proof that FermionSampling is solvable in classical polynomial time, and we reuse techniques from our BosonSampling analysis to characterize random FermionSampling distributions.
BosonSampling, which we proposed three years ago, is a scheme for using linear-optical networks to solve sampling problems that appear to be intractable for a classical computer. \ In a recent manuscript, Gogolin et al.\ claimed that even an ideal BosonSampling device's output would be operationally indistinguishable\textquotedblright\ from a uniform random outcome, at least \textquotedblleft without detailed a priori knowledge; or at any rate, that telling the two apart might itself be a hard problem. We first answer these claims---explaining why the first is based on a definition of a priori knowledge such that, were it adopted, almost no quantum algorithm could be distinguished from a pure random-number source; while the second is neither new nor a practical obstacle to interesting BosonSampling experiments.However, we then go further, and address some interesting research questions inspired by Gogolin et al.'s arguments. We prove that, with high probability over a Haar-random matrix A, the BosonSampling distribution induced by A is far from the uniform distribution in total variation distance. More surprisingly, and counter to Gogolin et al., we give an efficient algorithm that distinguishes these two distributions with constant bias. Finally, we offer three bonus results about BosonSampling. First, we report an observation of Fernando Brandao: that one can efficiently sample a distribution that has large entropy and that's indistinguishable from a BosonSampling distribution by any circuit of fixed polynomial size. Second, we show that BosonSampling distributions can be efficiently distinguished from uniform even with photon losses and for general initial states. Third, we offer the simplest known proof that Fermion Sampling is solvable in classical polynomial time, and we reuse techniques from our Boson Sampling analysis to characterize random FermionSampling distributions.
Boson sampling is a candidate for quantum protocols to truly realize the quantum computation advantage and to be used in advanced fields where complex computations are needed, such as quantum chemistry. However, this proposal is hard to achieve due to the existence of noise sources such as photon losses. In order to quantificationally analyze the influences of photon losses in optical networks, boson sampling is classically simulated based on the equivalent beam splitter mechanism, where the photon loss happening in optical units is equivalent to the photon transmission into the environmental paths through a virtual beam splitter. In our simulation, networks corresponding to random unitary matrices are made up, considering both the Reck structure and the Clements structure. The photon loss probability in an optical unit is well controlled by adjusting the parameters of the virtual beam splitter. Therefore, to simulate boson sampling with photon losses in optical networks is actually to simulate ideal boson sampling with more modes. It is found that when the photon loss probability is constant, boson sampling with Clements structures distinctly performs much better than that with Reck structures. Furthermore, the photon loss probability is also set to follow the normal distribution, which is thought to be closer to the situation in reality. It is found that when the mean value of photon loss probability is constant, for both network structures, errors of outputs become more obvious with the increase of standard deviation. It can be inferred that the increase of error rate can be explained by the network depth and the conclusion is suitable for larger-scale boson sampling. Finally, the number of output photons is taken into consideration, which is directly related to the classical computation complexity. It is found that with the photon loss probability, the ratio of output combinations without photon losses decreases sharply, implying that photon losses can obviously affect the quantum computation advantage of boson sampling. Our results indicate that photon losses can result in serious errors for boson sampling, even with a stable network structure such as that of Clements. This work is helpful for boson sampling experiments in reality and it is desired to develop a better protocol, for example, a well-designed network or excellent optical units, to well suppress photon losses.
Sampling experiments provide a viable route to show quantum advantages of quantum devices over classical computers in well-defined computational tasks. However, quantum devices such as boson samplers are susceptible to various errors, including stochastic errors due to fabrication imperfections. These cause the implemented unitary operations to deviate randomly from their intended targets, following distributions with finite variance. Whilst full-scale quantum error correction remains challenging in the near term, quantum error mitigation schemes have been devised to estimate expectation values, but it is unclear how these schemes would work for sampling experiments. In this work, we demonstrate that, given access to multiple stochastic unitaries, it is possible to mitigate the effect of these errors in sampling experiments. We adopt the unitary averaging protocol which employs multiple stochastic boson samplers to generate a distribution that approximates the ideal boson sampler distribution as the number of samplers increases. We derive a rigorous upper bound on the trace distance between the output probability distributions induced by invertible vacuum-heralded networks based on the Schur-Weyl duality. This result can be seen concretely as an error mitigation scheme in sampling experiments against stochastic errors. On a broader level, it suggests a path towards understanding error mitigation for sampling experiments and developing analysis tools for photonic circuits incorporating measurements and feed-forward. We further provide other applications of unitary averaging, including its use in implementing the linear combination of unitaries and benchmarking fabrication repeatability in linear optics.
For near-term quantum devices, an important challenge is to develop efficient methods to certify that noise levels are low enough to allow potentially useful applications to be carried out. We present such a method tailored to photonic quantum devices consisting of single photon sources coupled to linear optical circuits coupled to photon detectors. It uses the output statistics of BosonSampling experiments with input size n (n input photons in the ideal case). We propose a series of benchmark tests targetting two main sources of noise, namely photon loss and distinguishability. Our method results in a single-number metric, the Photonic Quality Factor, defined as the largest number of input photons for which the output statistics pass all tests. We provide strong evidence that passing all tests implies that our experiments are not efficiently classically simulable, by showing how several existing classical algorithms for efficiently simulating noisy BosonSampling fail the tests. Finally we show that BosonSampling experiments with average photon loss rate per mode scaling as o(1) and average fidelity of $ (1-o(\frac{1}{n^6}))^2$ between any two single photon states is sufficient to keep passing our tests. Unsurprisingly, our results highlight that scaling in a manner that avoids efficient classical simulability will at some point necessarily require error correction and mitigation.