A collaboration between the Beijing Academy of Quantum Information Sciences (BAQIS) and IFIMAC at the Universidad Autónoma de Madrid has reported a Bell inequality violation from an unlikely source: a single quantum dot that emits no photon pairs at all. Instead of generating two photons and entangling them, the team worked with a superposition of vacuum and one photon — a state in which a single photon is, in a precise quantum-mechanical sense, both present and absent — and split that superposition between two remote nodes. Measured with Franson interferometers, the two nodes showed correlations that no classical model can reproduce, with a CHSH parameter of S = 2.675 ± 0.050, more than thirteen standard deviations above the classical limit of 2. The work is published in Optica as “Bell Inequality Violation with Vacuum-One-Photon Number Superposition States,” and was highlighted by IFIMAC on October 30, 2026.
The headline is tempting to misread, so the first task is to state it correctly. This is not entanglement across time in the sense of time travel, and it is not entanglement between a photon and its past self. It is time-bin entanglement between two spatial modes: the entangled parties are two nodes, conventionally Alice and Bob, and the entanglement lives in the early-versus-late arrival-time structure of a single-photon superposition shared between them. The “time” in the result is a measurement basis, not a journey. Once that is clear, the result is genuinely interesting — not because it breaks a record, but because it removes a piece of apparatus that entangled-photon sources have historically required: the pair itself.
What the experiment actually did
The source is a self-assembled InAs quantum dot coupled to a micropillar cavity, driven resonantly by a continuous-wave laser. Under resonant excitation, the dot’s resonance fluorescence — the light it scatters when driven coherently — contains a component that is a pure superposition of vacuum and one photon. Write it schematically as a superposition with amplitudes for zero photons and for one photon, where the one-photon part carries the phase of the driving laser. That phase inheritance matters: it is what makes the state coherent rather than a classical mixture of “sometimes a photon, sometimes nothing.”
The team then sent this superposition through a fiber beamsplitter, delocalizing it between two spatial modes. Neither node receives a photon in the ordinary sense; each receives part of a superposition that spans both. Alice and Bob each measured their share with an asymmetric Mach-Zehnder interferometer of the Franson type, which interferes the early and late time bins against each other. By stepping the interferometer phases through settings including 0, π/2, π, and 3π/2, and recording coincidence counts between detectors at the two nodes, the team mapped out the correlation function whose combination gives the CHSH parameter.
Two engineering choices made the measurement possible at all. First, the micropillar cavity was engineered to be almost non-reflective at the cavity resonance, so that the quantum dot’s response, rather than reflected laser light, dominated the collected signal. Second, the drive was kept very weak: the laser driving power was set at n = 0.01, a low mean photon number at which the vacuum-one-photon character of the fluorescence is cleanest. Both choices are about the same enemy, laser background, and the paper’s power-dependence data show why. As the drive was increased from n = 0.01 toward n = 9.81, the measured second-order correlation rose, the CHSH parameter fell, and at the highest powers it dropped to about S = 1.344, below the violation threshold. Entanglement here is a low-power phenomenon, and the experiment is designed around keeping it there.
The numbers to quote, with their conditions
The headline number is S = 2.675 ± 0.050, against a classical bound of 2. The margin — more than thirteen standard deviations — is the strength of the claim, and it should be quoted with its uncertainty attached, not rounded away. A second number characterizes the quality of the time-bin entanglement itself: a reported visibility of 92.8 ± 2.6 percent in the time-entangled state, indicating strong correlation between the two nodes’ measurement outcomes.
A third number sets the single-photon character of the source: a Hanbury Brown–Twiss second-order correlation of approximately g²(0) = 0.037 at the operating point, far below the value of 1 expected for classical light and consistent with a source that essentially never emits two photons at once. That number is doing real work in the argument. The whole scheme depends on the state being a superposition of vacuum and exactly one photon; a significant two-photon component would contaminate it, and the low g²(0) is the evidence that the contamination is small.
The power-dependence data also give the result its most instructive limit. The Bell violation disappeared once the unwanted laser component of the collected light reached approximately 43 percent of the total average photon number, corresponding to a second-order correlation of roughly g²(0) ≈ 0.2. That crossover is the experiment’s error budget stated as a threshold: it tells any group trying to reproduce the work exactly where the scheme breaks.
One further reported figure deserves a hedge rather than a headline. Coverage of the paper describes the source’s brightness as exceeding previous implementations of similar entangled states by more than two orders of magnitude, even under the weakest excitation conditions. That comparison is plausible given the efficiency of a cavity-coupled dot source, but the institute highlight itself does not publish the underlying rates or a full loss budget, and this article does not invent them. Until the Optica paper’s device figures — count rates, collection and detection efficiencies, and the loss budget of the interferometer arms — are checked against the published text, brightness should be treated as the authors’ reported comparison, not as an independently verified specification. The same restraint applies to any claim about transmission distance: the highlight describes the nodes as remote, but gives no fiber length, and no distance figure is quoted here.
Why skipping the pair matters
To see why this result is more than a curiosity, compare it with the two established routes to entangled photons. Spontaneous parametric down-conversion produces entangled pairs probabilistically: pairs appear at random times, and one photon must be detected to herald the other. The biexciton cascade in a quantum dot, covered on this desk before, produces pairs on demand — but it requires preparing a two-pair state, controlling the fine-structure splitting so the two decay paths are indistinguishable, and managing the timing jitter between the two sequential emissions. Both routes pay a price for the fact that they must create two photons and then entangle them.
The BAQIS–IFIMAC scheme pays a different, smaller price. It creates one superposition and splits it. There is no multiphoton state to generate, no cascade timing to engineer, and no post-selection of multiphoton components to throw away good events — the step that, in earlier vacuum-one-photon approaches, reduced efficiency. The entanglement is generated directly from the natural fluorescence of a single dot under resonant drive, with what the authors describe as neither complex excitation sequences nor detailed signal post-processing. The source is, in that sense, simpler than the sources it competes with: one emitter, one beamsplitter, two interferometers.
That simplicity is also the source’s limitation, and the two should be stated together. A split single-photon superposition is an entangled state of two modes, and it can serve protocols that need entanglement between nodes — tests of nonlocality, and in principle entanglement distribution for quantum networks. But it is not a two-photon resource: it cannot supply two independent photons to two independent downstream operations the way a pair source can, and the “photon” at each node is not a stored qubit. The entanglement is carried by propagating light and measured on arrival. Any application that needs storage, or that needs to hold entanglement while a computation proceeds, still needs matter qubits and the interfaces that couple them to photons. This result simplifies the source; it does not replace the network.
What this does not show
The usual cautions apply, and they are worth stating explicitly because the headline invites overreading. First, this is a Bell inequality violation measured by Franson interferometry on coincidence counts between two nodes. It is a demonstration of nonlocal correlations in this system, under the experiment’s conditions. It is not presented as a loophole-free Bell test, and it should not be cited as one: the highlight and the paper’s abstract claim a clear CHSH violation, not the closure of the locality, detection, or other loopholes that a loophole-free test requires. The distinction matters for cryptography, where device-independent protocols demand loophole-free violations, and this result does not supply that.
Second, the result is a laboratory demonstration, not a network deployment. No deployed link, no repeater, and no multi-node network is reported. The nodes are two measurement stations on an optical table fed by one source, and the honest description of the achievement is a source-and-measurement demonstration with a quantified violation and a quantified breaking point.
Third, the entanglement is between spatial modes in the time-bin basis. It is not entanglement between two quantum dots, not entanglement between two stored qubits, and not evidence that excitonic qubits themselves have been entangled over distance. Applied here, the checklist from this desk’s earlier article on entanglement gives: two optical modes at two nodes; Franson interferometry and a CHSH test; S = 2.675 ± 0.050 with 92.8 percent visibility; deterministic generation of the superposition with coincidence-based measurement; at low drive power, n = 0.01, with the violation surviving only while laser background stays below about 43 percent of the collected light. That is a strong, well-specified claim — and its boundaries are part of the claim.
What to watch next
The most concrete follow-up is already sketched by the authors: replacing the fiber beamsplitter with a tritter or a 1-to-3 fiber coupler to distribute the vacuum-one-photon superposition across three spatial modes, creating a tripartite entangled state. If that succeeds with a comparable violation, it would show that the scheme scales in the number of modes without scaling in the complexity of the source — one dot, more outputs. That is the experiment to look for, and the number to demand from it is the same one reported here: a multipartite Bell parameter with its uncertainty, measured as a function of drive power, with the background crossover stated.
A second follow-up concerns integration. The present demonstration uses a micropillar cavity dot, fiber beamsplitters, and bulk-style interferometers feeding superconducting nanowire single-photon detectors. Whether the same vacuum-one-photon scheme survives on-chip routing, with its additional losses and phase instability, will determine whether it is a network component or a laboratory technique. The loss budget — the figure this article could not verify from the highlight — will be the deciding document: every decibel lost between the dot and the detectors eats directly into the coincidence rates and, through background admixture, into the violation itself.
A third question is reproducibility across devices. Quantum dots are famously individual: each dot’s wavelength, brightness, and coupling differ, and a result on one well-chosen device is a demonstration, not a platform. The scheme’s claim to scalability rests on the argument that it needs only one good dot and standard fiber components, rather than a matched pair or a tuned cascade. That argument will be tested when a second group, with a second dot, reports the same violation with the same power dependence.
None of these caveats diminishes the core observation. A single quantum dot, driven gently and coherently, produces a state whose very emptiness — the vacuum part of the superposition — is an active ingredient in entanglement. Splitting nothing-and-something between two places, and finding correlations that classical physics forbids, is a reminder that in quantum optics the vacuum is not the absence of a resource. It is, under the right drive, half of one.
Sources and further reading
- Zi-Qi Zeng, Jian Wang, Xiu-Bin Liu, Xu-Jie Wang, Li Liu, Hanqing Liu, Haiqiao Ni, Zhichuan Niu, Carlos Antón-Solanas, Bang Wu, and Zhiliang Yuan, “Bell Inequality Violation with Vacuum-One-Photon Number Superposition States,” Optica. DOI: 10.1364/OPTICA.601049. Preprint: arXiv:2511.15413 (submitted November 19, 2025).
- IFIMAC–UAM research highlight, “Testing the Bell Inequality with Time-Entanglement of Single Photons Emitted from a Quantum Dot,” published October 30, 2026.
- Prior literature on vacuum-one-photon entanglement and Franson interferometry, and on resonance fluorescence from single quantum dots, as cited in the Optica paper.
- This desk’s earlier articles on entanglement with excitons and on entangled photon pairs from the biexciton cascade, for the comparison between pair sources and single-photon-superposition sources.
- Note on verification: the CHSH value S = 2.675 ± 0.050, the classical limit of 2, the low-drive setting n = 0.01, the InAs quantum dot in a micropillar cavity under continuous-wave resonant drive, the phase settings, and the October 30, 2026 highlight date are taken from the IFIMAC highlight and the paper’s abstract and reporting. The visibility, g²(0), and crossover figures are as reported in coverage of the paper; device-level rates and the full loss budget were not available in the highlight and are deliberately not quoted above.