Can joint photon measurements improve closure-phase accuracy in interferometric imaging?

Closure phase cancels piston noise but costs photons

Ground-based interferometric arrays sample the complex coherence of a scene, and imaging normally proceeds by Fourier transforming those coherences. Atmospheric fluctuations add an unknown telescope-dependent phase, called a piston phase, to each measured visibility, so simple Fourier inversion of complex visibilities is not robust. The paper works in the photon-starved regime, where each temporal mode across the array is dominated by vacuum and holds at most one photon with small probability.

For any triangle of telescopes, the bispectrum is the product of the three complex visibilities around that triangle, and its phase is the closure phase. Because each piston phase enters once with a plus sign and once with a minus sign around the cycle, the closure phase is invariant under telescope-dependent phase shifts and retains only source information. The paper notes that closure phase still constrains scene properties such as spatial extent and asymmetry, even though two degrees of freedom of the intrinsic visibilities are lost when only the closure phase is estimated.

The catch is that the piston phases are unknown and vary in time. The paper models this with the atmospheric coherence time: within one such window the piston phases are assumed fixed, and across windows they are treated as independently uniform on [0, 2π). Photons in the same window but with significantly different frequencies would experience different piston phases, so the analysis is restricted to a small spectral band, typically about 10^11 Hz wide.

Three photons in one window are the minimum

The paper first asks how many photons from the same atmospheric stability window are needed for closure-phase information to survive the unknown piston noise. It establishes an information-theoretic lower bound: at least three photons are required. With fewer than three, the noise-invariant quantity cannot be accessed.

This is a statement about the joint state of the photons, not about simply counting detections. The paper describes the n-photon state in one window as a classical mixture over all possible piston-phase vectors, and shows that the closure-phase information lives in coherences that require three photons to be present. The authors also note that three-photon noisy states can be constructed from a larger n-photon state by partitioning the photons into disjoint triples, which is how the later large-array protocol reuses the same building block even when many photons arrive per window.

Joint three-photon measurement beats single-photon strategies

In the minimal three-telescope setting, the paper characterizes the fundamental limits of collective three-photon measurements and of strategies that measure photons individually. Collective measurements allow the total photon number at each telescope to be measured nondestructively, so the informative block carrying closure-phase information can be identified quickly and noisy measurements avoided.

Guided by this comparison, the authors construct a quantum-memory-assisted three-photon protocol and prove that it estimates the closure phase more efficiently than any single-photon measurement strategy. The advantage is in photon cost at fixed accuracy, not in a different observable: both approaches target the same closure phase, but the joint measurement extracts more information per detected photon.

The paper also presents a simple-to-implement measurement scheme for this minimal case, labeled Protocol 2, and generalizes it to more photons and sites. The practical requirement is that photons arriving within one atmospheric coherence time be stored coherently in quantum memory, or equivalently in long optical delay lines that can be rapidly and accurately tuned, before a quantum circuit and measurement.

Large arrays carry far less bispectrum information than they seem to

An m-telescope array contains Θ(m²) independent triangle bispectra, which naively suggests that estimating all of them should scale with the number of triangles. The paper shows that physical constraints on the visibilities make the total information content grow only nearly linearly with m at fixed accuracy. In other words, the full set of complex bispectra is highly compressible in astronomically relevant parameter regimes.

This is an information-content result, not a claim that the bispectra themselves are redundant as observables. The authors use it to argue that the number of photons sufficient to estimate all bispectra may be much lower than the triangle count would suggest. The compression is what makes an efficient protocol possible.

Photon cost scales linearly with array size

Exploiting the compressible structure, the paper constructs a protocol that uses at most three-photon joint measurements and estimates all bispectra with photon cost Õ(m). By contrast, any sequential single-photon protocol requires Ω(m²) photons. The comparison is between two ways of processing the same incoming light: joint measurements on up to three photons versus measuring photons one at a time and keeping only classical records.

The paper states the consequence in terms of observing time: when the number of detected photons per temporal stability window scales linearly with the array size, the required exposure time remains independent of m up to logarithmic factors. This is a statement about the scaling of exposure time under that photon-rate assumption, not a claim that any particular array already achieves it.

The authors also analyze the complementary high-precision regime using multiparameter quantum metrology and find a similar advantage for collective measurements. In that regime, the gap between the Holevo and Nagaoka-Hayashi Cramér-Rao bounds grows linearly in m for estimating the vector of closure phases, meaning a general measurement on all photons gives an extensive reduction in estimation variance as the array grows. For three-photon packets, by contrast, the two bounds coincide, so there is no additional advantage to measuring multiple three-photon packets together.

Where the advantage does not transfer

The linear-photon-cost result depends on the physical constraints on the visibilities that make the bispectrum information compressible; it is derived under a global mean-square error criterion for estimating all bispectra, not for an arbitrary subset or for a single closure phase in a large array. The three-photon minimum is tied to the atmospheric coherence time: photons must come from the same stability window, and the analysis assumes a small spectral band so that piston phases are unchanged across the band.

The paper also assumes |g_i| = 1, so atmospheric fluctuations are treated as pure phase errors and visibility amplitudes are unaffected. Adaptive optics or other calibration techniques are outside this model. The high-precision metrology results are asymptotic statements about Cramér-Rao bounds, and the paper notes that the gap between collective and individual measurement bounds vanishes for three-photon packets, so the collective advantage in that regime comes from measuring all photons together rather than from grouping them into triples.

The protocol requires coherent storage of photons for up to one atmospheric coherence time while waiting for the other photons to arrive. That is a hardware requirement, and the paper frames it as enabled by quantum memories or rapidly tunable long optical delay lines followed by a quantum circuit and measurement.

What to watch when comparing protocols

The decision-relevant comparison in this paper is not whether joint measurements are more accurate in the abstract, but how photon cost scales with array size at fixed accuracy. For a three-telescope triangle, the relevant threshold is three photons from the same atmospheric stability window: below that, closure-phase information does not survive the unknown piston noise, and at that level joint measurement already outperforms single-photon strategies.

For an m-telescope array, the relevant split is between Õ(m) photons for the three-photon joint protocol and Ω(m²) photons for any sequential single-photon protocol. If the detected photon rate per stability window grows linearly with m, that gap translates into exposure time that stays flat in m up to logarithmic factors. Readers comparing imaging strategies should therefore check which regime they are in: the minimal triangle, where the three-photon advantage is already established, or a large array, where the advantage depends on the compressibility of the bispectrum information and on the photon rate per stability window.

The paper's own framing is that quantum information processing can substantially improve the sensitivity and robustness of optical interferometry. The specific priority it establishes is that the minimum unit of joint processing is three photons, and that this unit is enough to change the scaling of the whole bispectrum estimation problem from quadratic to nearly linear in the number of telescopes.

Sources used in this answer