Lost photon enhances superresolution

Diffraction sets classical resolution limits, but various super-resolution techniques (STED, SOFI, antibunching-based methods, structured illumination, quantum imaging) can overcome them. Entanglement can enhance resolution and visibility; measuring nth-order correlations with an n-photon entangled state narrows the effective PSF by sqrt(n) relative to coherent imaging. Intuitively, one might expect that using the highest available correlation order (n) is optimal. However, effective PSF narrowing and resolution enhancement can occur with classical correlations as well, and the maximal correlation order is not necessarily optimal. This work shows that, given an n-photon entangled illumination, measuring only (n-1)-photon coincidences (i.e., ignoring one photon) can outperform nth-order measurements: (n-1)th-order correlations yield an effectively narrower PSF and higher Fisher-information-based resolution than nth-order detection. Despite entanglement fragility, losing a photon can still increase informativeness through state modification of the remaining photons. The effect is related to ghost imaging and imaging with undetected photons: a photon outside the imaging aperture imparts a position-dependent phase on the remaining photons, effectively shaping their wavefunction (akin to structured illumination or PSF shaping), thus improving resolution. The paper further proposes a practical conditional scheme: placing a bucket detector outside the main imaging aperture to post-select events where one photon is detected off-axis while the remaining (n-1) reach a position-sensitive detector, achieving additional resolution enhancement over plain (n-1)-photon detection. Fisher information is employed to quantify resolution gains in a realistic multi-parameter imaging task.

Imaging model and signals:

• Object: transmission amplitude A(s), with real 0 ≤ A(s) ≤ 1; s is transverse position at the object plane.
• Illumination: n-photon entangled quantum state |Ψ_n⟩ with spatial correlations across transverse wavevectors k_i.
• Optical system: point-spread function (PSF) h(s, r) mapping object-plane position s to image-plane coordinate r.
• Measurement: coincidence photo-counting of intensity correlations.

Correlation signals:

• nth-order coincidence rate (at image position r): G^(n)(r) ∝ ∫ d^2s |A(s)|^2 |h(s, r)|^(2n). This includes the nth power of the PSF, giving an effective PSF narrowing compared to coherent imaging.
• (n-1)th-order coincidence (ignoring one photon): G^(n−1)(r) ∝ ∫ d^2s |A(s)|^(2(n−1)) |h(s, r)|^(2(n−1)). For n > 2, this produces a sharper image than G^(n)(r) because the integrand contains a higher power of the PSF and modulates object transmission more strongly.

Effective state modification (conditional pathways for the nth photon):

• Decompose n-photon detection into sequential steps: detection (or not) of one photon modifies the remaining (n−1)-photon state, which is then detected.
• Cases for the nth photon when (n−1)-photon coincidences are registered:
  1. Transmitted through the object and inside the imaging aperture: if detected at r′, the remaining state is |Ψ_(n−1)(r′)⟩ ∝ ∫ d^2s A(s) h(s, r′) [a†(s)]^(n−1) |0⟩. This biases remaining photons through regions mapped to r′.
  2. Transmitted but outside the imaging aperture with transverse momentum k: the remaining state acquires a phase factor exp(i k·s), effectively imprinting a periodic phase modulation akin to structured illumination.
  3. Absorbed by the object: modeled by an additional mode (beamsplitter model), contributing absorption-weighted terms.
• Averaging over these possibilities yields an effective separable mixed state of (n−1) photons (a mixture of (n−1)-photon excitations of spatial modes), yet with modified statistics that can enhance sensitivity.

Conditional post-selection (hybrid near-field/far-field scheme):

• Place a bucket detector in the far field outside the lens aperture to detect the nth photon in a region Ω of k-space (e.g., Ω = {k: k_max ≤ |k| ≤ 2 k_max}).
• Measure the joint coincidence signal G^(n−1,1)(r, Ω) for (n−1) photons at image-plane position r and the nth photon anywhere in Ω: G^(n−1,1)(r, Ω) ∝ ∫ d^2s ∫ d^2s′ A^n(s) A^n(s′) h^(n−1)(s, r) h^(n−1)(s′, r) g(s − s′), where g(s − s′) = ∫_{k ∈ Ω} d^2k exp[i k·(s − s′)].
• This selectively enhances destructive interference between object features (e.g., two slits), boosting image contrast and resolution.

Model examples and simulations:

• Two-pinhole object (separation 2d) and NOON-like input for intuition: nth-order detection includes constructive interference cross-terms that blur; (n−1)-order removes that cross-term, yielding sharper peaks for n > 2; conditioning on specific off-axis k can induce destructive interference for maximum contrast.
• Practical multi-slit objects with semitransparent slits; imaging system passes transverse momenta up to k_max; compare three strategies: G^(n), G^(n−1), G^(n−1,1) with Ω choices (e.g., k_max ≤ |k| ≤ 1.5 k_max or ≤ 2 k_max). Simulations show enhanced visual contrast and resolution for (n−1) and further improvement with conditioning.

Fisher information analysis (multi-parameter estimation):

• Decompose A(s) into basis functions f_μ(s) (slit-like pixels), A(s) = Σ_μ θ_μ f_μ(s).
• For sampled signal S(r_i), define the Fisher information matrix per coincidence event: F_{μν} = Σ_i [1/S(r_i)] (∂S(r_i)/∂θ_μ)(∂S(r_i)/∂θ_ν).
• Cramér-Rao bound on total parameter variance: Σ_μ Var(θ_μ) ≥ Tr(F^{-1})/N, where N is the number of registered coincidence events.
• Define resolution by requiring Tr(F^{-1}) ≤ N_max (e.g., N_max = 10^5) and determining minimal feature size d (normalized by Rayleigh limit d_R = 3.83/k_max) satisfying this criterion.
• Compare methods at equal single-event normalization; also account for reduced coincidence rates of conditioned events by comparing overall detection probabilities p = ∫ S(r) and using ratios P_{n−1,1}/P_n.

Event rates:

• Conditioned (n−1,1) events occur 3–20 times less frequently than n-photon coincidences, depending on Ω and object. Nonetheless, due to rapid Fisher information loss below the resolution threshold in multi-parameter tasks, the rate penalty minimally affects the resolved limit in the studied regimes.

• Counterintuitive advantage: For an n-photon entangled source, measuring (n−1)-photon coincidences (ignoring one photon) yields an effectively narrower PSF and higher Fisher information per event than nth-order detection for n > 2. This arises because losing/detecting the nth photon modifies the remaining (n−1)-photon state, increasing its sensitivity to object features.
• Mechanism: If the nth photon exits the aperture with transverse momentum k, the remaining state acquires a phase exp(i k·s), effectively implementing structured-illumination-like phase modulation that can induce destructive interference and enhance contrast.
• Conditional scheme: Post-selecting events with the nth photon detected in a far-field region Ω outside the aperture (bucket detector) provides additional contrast and information gain beyond plain (n−1)-photon detection.
• Fisher information gains (multi-slit objects, single-event normalization): For n = 3 and 4, (n−1)-order correlations yield about 10–20% better resolution than nth-order; conditioning on Ω yields an additional ~10–15% improvement that diminishes in deep superresolution (d ≤ 0.2 d_R).
• Biphoton case: For n = 2, (n−1) does not outperform n in principle, but conditioning can still yield modest benefits for specific Ω choices.
• Quantitative example (n = 4): Minimal resolvable slit width d (threshold Tr F^{-1} ≤ 10^5) improves from 0.212 d_R (G^(n)) to 0.170 d_R (G^(n−1,1) with Ω = {k: k_max ≤ |k| ≤ 2 k_max}); accounting for reduced detection rate, 0.177 d_R, still significantly below classical limit.
• Event-rate trade-off: Conditioned (n−1,1) coincidences are 3–20× less frequent than n-photon coincidences, but the overall resolution threshold is largely unaffected in the studied multi-parameter regime.
• Practical significance: Comparable resolution gains to increasing photon number (e.g., from n = 3 to 4) can be achieved by adding a simple bucket detector, avoiding the complexity of generating higher-order entanglement.

The work addresses whether maximal correlation order is optimal for superresolution with entangled photons. It demonstrates that, due to quantum-state modification upon loss or off-axis detection of one photon, the remaining (n−1) photons can carry more actionable information for resolving object features than measuring all n photons, for n > 2. The mechanism connects to ghost imaging and imaging with undetected photons, where undetected paths imprint phase information, effectively shaping the illumination and suppressing blurring due to constructive interference. Fisher information analysis in a realistic multi-parameter setting confirms that (n−1)-photon detection enhances information per event and achievable resolution, and that conditioning on off-aperture detection further improves performance while incurring a tolerable rate penalty. These findings suggest a practical route to resolution enhancement without increasing photon-number entanglement complexity, with potential benefits for low-photodose imaging (e.g., biological samples) and for systems where generating triphoton or higher entangled states is challenging.

The study reveals that, for n > 2, measuring (n−1)-photon correlations with an n-photon entangled source can surpass traditional nth-order coincidence detection in resolution and information per event. The advantage stems from effective state modification of the remaining photons when one photon is lost or detected outside the imaging aperture, imparting structured phase modulations that enhance contrast. A practical hybrid scheme using a bucket detector to post-select off-aperture events achieves further gains with modest experimental overhead. Fisher information analysis in multi-parameter imaging corroborates 10–20% resolution improvements for n = 3, 4, with additional 10–15% from conditioning, and provides concrete thresholds well beyond the classical Rayleigh limit. Future research directions include experimental demonstrations with biphotons and triphotons, optimization of the conditioning region Ω and detection geometry for maximal information gain versus rate, robustness analysis to losses and detector inefficiencies, and extensions to more complex objects and imaging modalities.

• Fundamental advantage requires n > 2; for biphotons (n = 2), (n−1) does not inherently outperform nth-order detection, though conditioning can provide limited benefits.
• Generating n ≥ 3 entangled photon states remains experimentally demanding (e.g., cascaded SPDC, four-wave mixing, third-order SPDC).
• Conditioned schemes reduce coincidence rates by factors of ~3–20 relative to nth-order detection, impacting acquisition time.
• The additional gain from conditioning diminishes in deep superresolution (small feature sizes d where k⊥ d < 1), limiting benefits in that regime.
• Modeling assumptions (e.g., real-valued A(s), idealized PSFs, neglect of various experimental imperfections) may limit direct quantitative transfer to all practical systems; detector efficiencies and noise can alter the balance between (n−1) and n-photon strategies.