Jiuzhang Gaussian boson sampling experiments

Short answer: Gaussian boson sampling is a task in which photons are sent through an interferometer and the output is sampled, and the Jiuzhang photonic experiments claim quantum computational advantage on it. A Science paper reports a 2020 experiment with 50 squeezed states and 100 modes, and later papers report Jiuzhang 4.0 with 1,024 squeezed states. The authors present these as advantage claims against classical simulation, and the Science text itself says the competition with classical methods is ongoing.

The first experiment

The Science paper describes Zhong et al. sending 50 indistinguishable single-mode squeezed states into a 100-mode ultralow-loss interferometer and sampling the output with 100 high-efficiency single-photon detectors. The optical setup is phase-locked, with full connectivity and a random matrix. The samples were validated against hypotheses using thermal states, distinguishable photons and a uniform distribution. The photonic device, called Jiuzhang, generates up to 76 output photon clicks, a state-space dimension of about 10 to the 30. The authors report a sampling rate about 10 to the 14 times faster than state-of-the-art classical simulation strategies and supercomputers.

The same page gives a balanced framing: it says recent work claimed quantum computational advantage but subsequent work sped up classical simulation and pointed to a sample-size-dependent loophole, and that advantage will be the result of a long-term competition between quantum devices and classical simulation, not a one-shot proof.

Jiuzhang 4.0

A 2025 arXiv preprint and a Nature paper describe Jiuzhang 4.0, a programmable photonic processor with 1,024 high-efficiency squeezed states in a hybrid spatial-temporal encoded 8,176-mode circuit. The Nature abstract reports 92% source efficiency, 51% overall system efficiency, detection events up to 3,050 photons, and sampling in a Hilbert space of dimension about 10 to the 2,461. The authors say photon loss has been the main barrier to scaling and that the results are validated against all current classical simulation methods, including matrix product state algorithms designed to exploit photon loss.

The preprint adds the authors' estimate that the best known matrix product state approach on a named supercomputer would need more than 10 to the 42 years to construct the required tensor network, while the processor produces a sample in 25.6 microseconds. These are the authors' own estimates and comparisons.

Limits

These are claims of advantage on a sampling task, validated by the authors against the classical algorithms they considered. The Science page notes that classical simulation of the earlier experiment was later sped up. The pages used here do not describe a practical application of the sampling task, and the authors' statements about fault-tolerant photonic computing are about future directions.

FAQ

What is Gaussian boson sampling?

In the first experiment, squeezed photons enter an interferometer and the output photon patterns are sampled.

What did the first Jiuzhang experiment use?

50 squeezed states, a 100-mode interferometer and 100 detectors, with up to 76 clicks.

What did Jiuzhang 4.0 add?

1,024 squeezed states in 8,176 modes, with detection events up to 3,050 photons, according to the authors.

Is the advantage settled?

The Science page frames it as an ongoing competition with classical simulation, noting that earlier work was followed by faster classical simulation methods.

Is there a practical application?

None is described in the pages used here.