WorksPhysics and Astronomy
Improving the signal-to-noise ratio through postselection of coincident detections in anticounterfeiting and quantum key distribution
Abstract
In quantum communications and other photon-counting applications, reliably detecting weak optical signals against a strong noise background remains a fundamental challenge. In this paper, we develop a simple idea: when a signal is carried by pairs (or more generally multiplets) of photons emitted simultaneously, postselection of coincident detections recorded on two detectors makes it possible to improve-sometimes drastically-the Signal to Noise Ratio. Essentially, this is made possible by the fact that genuine and accidental coincidences scale differently with the width of the postselection window. As long as the window remains wide enough to retain detections from the same pair, the genuine coincidence rate is approximately unchanged. In contrast, coincidences formed from independent clicks decrease linearly as the window is narrowed. The resulting signal-to-noise ratio, therefore, increases inversely with the window width. Our approach makes it possible to reveal photon-pair signals that are otherwise hidden beneath a much stronger uncorrelated background. We analyze here applications of this idea to quantum key distribution, metrology, and anti-counterfeiting.
Introduction.
As a consequence of the projection postulate, it is most often impossible to measure a quantum system without disturbing its state. In particular, if information is encoded in complementary degrees of freedom, only an authorized user who knows the encryption basis has full access to this information, consistent with Heisenberg’s uncertainty principle. The no-cloning theorem also prevents perfect copies of an unknown quantum state [1]. These concepts paved the way for anticounterfeiting and quantum key distribution (QKD), which were respectively conceived by Wiesner in the 70s (though it remained unpublished until 1983 [2]) and by Bennett and Brassard in 1984 [3]. Wiesner proposed the idea of quantum money to create unforgeable banknotes. Bennett and Brassard proposed encoding a random binary key in complementary bases to ultimately serve as a one-time pad. In both cases, the support of the information is a single qubit, and the bases of encryption are so-called mutually unbiased bases (which are associated with orthogonal directions on the Bloch sphere representation of the qubit Hilbert space). In 1991, Artur Ekert proposed a two-qubit scheme in which encryption bases are chosen by two authorized users, Alice and Bob, who share a maximally entangled two-qubit state [4]. In optical implementations, these qubits are commonly encoded in photon pairs generated (see Figure 1) via a nonlinear process such as spontaneous parametric down-conversion [5]. The use of photon pairs introduces an additional experimental task, since Alice and Bob must identify which detection events originate from the same emission process. This identification is usually performed through temporal postselection [6, 7]. After compensating for the relative propagation delay between the two detection channels, two clicks are assigned to the same pair when their time tags fall within a coincidence window of width , which we refer to as the postselection window. Genuine pair detections are retained through this temporal criterion, but independent clicks can also satisfy it by chance, producing accidental coincidences between detections that do not originate from the same emitted pair. Such events can arise from detector dark counts, ambient light, fluorescence in the optical setup, or photons from different emitted pairs [8, 9].

Figure 1. Schematic representation of spontaneous parametric down-conversion (SPDC). A pump photon is converted into two lower-energy photons, denoted and . (a) Momentum conservation (phase-matching condition), . (b) Energy conservation, . Here, and denote the wave vector and angular frequency of photon , with and identifying the pump.
The central observation of this work follows from the different dependence of genuine and accidental coincidences on the postselection window (see II.1). The probability of detecting a genuine pair within a window is proportional to . In contrast, the probability of recording simultaneously one independent background click in each channel is proportional to . Since the number of windows per unit time scales as , the genuine coincidence rate remains independent of the window width, while the accidental coincidence rate increases linearly with it. Temporal postselection can therefore recover a correlated signal even when its contribution to every single record is much weaker than the background. We examine this principle in two settings: the first is entanglement-based quantum key distribution, in which accidental coincidences produce random outcomes, thereby increasing the quantum bit error rate. The second is optical marking (with applications in anticounterfeiting), where a small number of pair-emitting centers form a correlation marker whose presence may remain hidden in the background noise and be measurable through postselected coincidences. This is the “needle-in-a-haystack” approach: temporal correlations reveal a weak quantum-light marker in a strong background.
The remainder of the paper is organized as follows. In Sec. II, we introduce the postselection protocol and derive the counting formulas for single-photon and photon-pair transmission. We then examine how detector inefficiency, transmission loss, multipair emission, optical pollution, and dark counts modify the ideal behavior. Section III applies the same framework to correlation-based optical marking and anticounterfeiting, while Sec. VI presents the experimental results. In Sec. IV, we extend the scaling argument to higher-order photon correlations. Additional experimental plots are provided in the Appendix.
Postselection of coincident photon pairs in quantum key distribution
We consider a source that emits photon pairs at a rate , where is the mean interval between successive emission events. As illustrated in Figure 1, during spontaneous parametric down-conversion (SPDC), a photon of energy (from now on denoted the blue photon) is sent into a non-linear medium and generates a pair of low-energy photons (from now on denoted the red photons or twin photons) of energies of the order of [5].
One photon propagates toward Alice, and the other toward Bob, and the total probability that they produce registered clicks is and , respectively. These quantities include both propagation losses and detector inefficiencies. The registered background rates at the two receivers are denoted by and ; they represent the total rates over the detectors used at each side and may include dark counts, ambient photons, and other random contributions. For symmetric configurations, we write . The detection time tags are divided into windows of width denoted . An event is retained only when exactly one click is registered at Alice’s side and exactly one at Bob’s side.
In contrast, windows containing multiple clicks at either receiver are discarded. Any known difference in propagation times between the two channels is removed by an appropriate delay imposed at the time tagger before this selection is applied, so two detections are considered simultaneous when they fall within the same compensated window.
In what follows, we will consider that the number of pair-emission events within a fixed postselection window is modeled by a Poisson distribution [10]. Other distributions could be considered (e.g., thermal or geometric distributions), but alternative choices of distribution do not substantially alter our analysis; therefore, in the present paper we will, once and for all, assume that the distribution of pairs is Poissonian. The mean number of emitted pairs per window is thus
Accordingly, the probabilities of producing one and two pairs are and . In the sparse regime, where , these expressions reduce to and . The same requirement must hold for the background processes, and the analysis below, therefore, assumes
These inequalities set an upper range on useful postselection windows. In practice, the lower useful range of the postselection window is determined by the detector timing jitter, time-tagger resolution, synchronization uncertainty, and the dependence of the detector response on the background count rate [11]. For the QKD analysis, we consider only the events retained after the measurement bases have been compared (sifting). We denote by the rate of retained events that exhibit the maximal correlation expected for the chosen pair of bases, and by the rate of retained events that give the opposite outcome. Accidental coincidences are uncorrelated and therefore contribute equally, on average, to these two rates. We characterize the quality of the postselected data through the correct-to-error ratio
which measures the quality of the retained bits. The corresponding quantum bit error rate is
Comparison between single-photon and pair protocols
Let us first consider the effect of temporal postselection in an ideal situation: perfect transmission and detection, with no intrinsic state-preparation error (maximal entanglement II.3 Eq. (17)).
Single-photon protocol
Let denote the mean interval between successive signal photons sent to Bob, so that the signal rate is . During a window of width , the probabilities of receiving a signal photon and a background click are, respectively, and . In the ideal model, a signal photon produces the correct bit, whereas a background click is randomly assigned to either of the two possible outcomes. The correct and erroneous retained-bit rates are therefore
The factor of 2 appears because Bob will interpret a background click as a bit 0 or 1 with equal probability. Using Eq. (3), we obtain
where the final form applies when the signal rate is much larger than the background rate, which is the case in the BB84 protocol, for which security imposes that the quantum bit error rate is upper-bounded by approximately 11 % if we consider coherent attacks and 14.64 % in the case of cloning-based individual attacks [12].
Both the signal and background probabilities are proportional to , so their ratio is independent of the window width. Therefore, modifying the width of the postselection window does not improve the retained-bit fidelity in this case. It is worth noting at this level that we consider a transmission protocol for which we cannot predict in advance the temporal window within which photons sent from Alice to Bob will arrive and/or be detected. This is achieved in practice by using a strongly attenuated laser beam as a source. QKD protocols involving deterministic single-photon guns, and in particular the case where single photons are heralded, will be considered in another section II.5.
Pair protocol
Here, we show that the situation changes when maximally entangled photon pairs carry the signal. In the ideal limit considered here, errors due to mixed signal–background coincidences do not survive the postselection process: both photons of every emitted pair are detected, so an additional background click produces multiple clicks at the level of one receiver, and the corresponding window is discarded. More generally, we will neglect in the rest of the paper situations in which three or more photons are present during the same temporal window, because their contribution is small; it can be considered as a second-order effect which disappears in the limit where the signal is sparse Eq. (2). Mixed signal–background coincidences must be taken into account in nonideal situations (e.g. due to lossy transmission and/or to dark count in inefficient detectors); however, as discussed in the next subsection.
To leading order in the sparse regime, the probability of producing one pair inside a window is , so the genuine coincidence rate (from now on denoted ) remains
By contrast, an accidental coincidence requires one independent background click at each receiver. Its probability is , and the corresponding rate is
A genuine pair yields the expected correlated outcome, whereas an accidental coincidence contributes equally to both correct and erroneous bits. Hence
which yields
The physical origin of the enhancement is thus explicit: genuine coincidences require one pair-emission event and therefore have an approximately window-independent rate, whereas accidental coincidences require two independent detections and grow linearly with . Narrowing the postselection window consequently improves the correct-to-error ratio to , provided the sparse-counting conditions remain satisfied Eq. (2).
Illustration: some orders of magnitude.
To illustrate the contrast between the two scalings, we take equal background rates . For a single-photon signal received at , Eq. (6) gives . On the other hand, for a pair source with and a postselection window , the accidental coincidence rate is
and Eq. (10) yields . Reducing the window to increases this ratio to approximately . Even when the pair-production rate is lowered to , a window of still gives . The same mechanism becomes particularly clear when the pair signal is weaker than the background in each channel. For , the accidental coincidence rate is . A source producing pairs then generates genuine coincidences at a rate one hundred times larger than the accidental rate. However, its contribution to each single record is times smaller than the background. The pair source is therefore almost invisible in the individual detection channels, yet remains clearly identifiable through its temporal correlations. This is our main source of inspiration for the "needle-in-a-haystack" approach to optical marking developed in Sec. III.
Transmission losses and detector efficiencies
In realistic settings, transmission losses and finite detector efficiencies reduce the probability that both photons of an emitted pair are registered. With the total detection probabilities introduced in § 1, the true or genuine coincidence rate becomes
For equal detector efficiencies and lossless channels, this gives , while perfect detectors in lossy channels give . Independent registered backgrounds at Alice’s and Bob’s receivers produce accidental coincidences at the rate
where and . Writing this contribution in terms of the measured rates and is important because environmental background noise and detector dark counts do not respond to detector efficiency in the same way. For example, if the background consists only of noisy single photons with rate at each detector, then and . Dark counts, by contrast, must be included directly in and .
These inefficiencies also permit accidental coincidences absent from the ideal model: a photon from a generated pair may be registered at one receiver while its partner is lost, with an independent background click supplying the second accepted event. Including the symmetric process, we obtain
A second possibility arises when two different pairs are emitted within the same window. One detected photon may then originate from each pair, while the two partners are lost or undetected. Such photons are cousins, not twins, and thus not entangled with each other. When two pairs are emitted by the source inside the same temporal window, and only one photon arrives at each side, there are two ways to form accidental coincidences. Therefore, to leading order in , this cross-pair contribution is thus
Within the sparse-counting approximation, the total accidental rate is therefore
where , , and respectively denote the total rate of accidental pair detections, of simultaneous detection of a background single photon at each side, of a signal photon at one side and a background photon at the other side and, finally, of two (red) signal photons that were not produced by the same blue (pump) photon. All three terms remain proportional to , but their relative weights respond differently on the source brightness, background level, and total detection efficiencies. All other postselected pairs correspond to the effective signal, characterized by the rate .
Intrinsic state error and retained-bit quality
Temporal postselection suppresses coincidences formed from unrelated detections, but it cannot correct errors already present in the photon-pair state. It is therefore useful to separate two physically distinct contributions to the retained-bit error rate. The first arises from imperfect correlations within genuinely detected pairs. The second arises from accidental coincidences between clicks that do not belong to the same pair. We model imperfect pair preparation by the Werner state [13]
where is the maximally entangled target state and quantifies the admixture of the maximally mixed state. The limit describes a perfectly prepared entangled state, whereas gives a completely mixed two-qubit state.
We consider a measurement basis in which produces the expected correlation with unit probability. The ideal component, therefore, contributes only to the correct retained-bit rate, with a contribution . By contrast, the maximally mixed component assigns equal probability to the four possible joint outcomes. Two of these outcomes obey the expected correlation, and two violate it, so this component contributes equally to correct and erroneous bits. The same reasoning applies to the accidental coincidences collected in , since their outcomes are uncorrelated in the selected basis. The retained bit rates are therefore
The corresponding correct-to-error ratio is
while the total retained-bit rate is
The quantum bit error rate consequently becomes
The term represents errors carried by genuine detected pairs because their prepared state is imperfect. In contrast, the term represents errors introduced by background–background, signal–background, and cross-pair coincidences. Both contributions are weighted by because a randomly retained event is erroneous with probability . Several limiting cases clarify the physics and give a pedagogical understanding. For a perfect source and negligible accidentals, and , so that . If the state is perfect but accidental coincidences dominate, then
which is the completely random-bit limit. Most importantly, narrowing the postselection window suppresses through Eq. (16), but it does not modify the state parameter . Hence, once accidental coincidences have been made negligible, the QBER saturates at the intrinsic error floor
Temporal postselection can therefore remove errors caused by incorrect temporal associations, but it cannot restore correlations lost during state preparation, transmission, or measurement. In practice, should therefore be inferred from the measured correlation visibility (as shown explicitly in Table II) or derived from a microscopic noise model. Its dependence on propagation distance is not universal, since different channels may introduce depolarization, dephasing, mode mismatch, or basis-dependent errors.
Table II. Comparison between the relative nominal pump powers and the relative experimentally inferred intrinsic coincidence rates, .
Illustration: Moon-Earth QKD.
In fibered QKD, the transmission decreases exponentially with the distance between the source and the detectors [14]. As the dark-count rate remains approximately constant while the detected signal decreases, the SNR decreases with increasing transmission distance. This loss is a major limitation for conventional single-photon and entanglement-based fiber QKD, for which record distances are typically of the order of a few hundred kilometers [15, 16]. Specialized architectures can extend this range substantially; in particular, twin-field QKD has been demonstrated over more than 1000 km of fiber [17]. In the case of free-space transmission, however, it is possible to realize QKD over much longer distances [18]. In particular, satellite-to-earth communication over 1200 kilometers was achieved by a Chinese team several years ago [19], which, to date, constitutes the longest distance over which violation of Bell’s inequalities has been demonstrated. One reason for the success of these free-space techniques is that atmospheric density, and hence the amount of matter causing absorption and scattering, decreases with altitude. A theoretical study has also investigated Bell tests and entanglement distribution through losses representative of an Earth–Moon link [20]. Accessory, violating Bell’s inequalities over the Moon-Earth distance (390,000 kilometers) would allow a gain of a factor of 300 over the distance of entanglement. Such an experiment would constitute a new fundamental test of Quantum Physics and allow for tighter constraints on alternative theories and interpretations [18].
For such transmission distances, which constitute a regime of extreme loss, the competition between background and multipair events becomes striking2 as we show now. Inspired by Ref. [20], let us consider a source on a satellite that sends entangled pairs to Alice and Bob, with Alice on the Moon and Bob on Earth. Following Ref. [20], we consider an ultrabright source of pairs with . To simplify the treatment, we will also assume here that the registration probabilities are equal and of the order of (this differs from the proposal of Ref. [20], but introducing asymmetric values for and does not drastically modify our analysis). The true coincidence rate is then
We estimate the background rates and to be of the order of , a representative low-noise value for high-quality SPAD systems [8, 21]. As shown in Ref. [20], because , it is sufficient for Alice and Bob to randomly choose their basis of detection at the same rate to realize QKD. Now, this is clearly not sufficient. Supplementary to Alice’s and Bob’s random choice, it is necessary for Alice and Bob to record the time tags of their clicks first and then perform postselection in accordance with the rules described at the beginning of this paper. To be able to differentiate twin photons from cousin photons, a postselection window of at most, say, is thus required. Eqs. (13)–(15) give
The narrow window makes both background contributions negligible, but it does not eliminate coincidences formed from photons belonging to different emitted pairs. Because this contribution grows quadratically with the source rate, it dominates the accidental counts in the present high-brightness regime. For an ideal prepared state, , Eq. (22) yields
Here, the mean number of generated pairs per window is . Note that a 1 ps postselection window is useful only if the source, detectors, synchronization, propagation, compensation, and time tagging preserve true coincidences on that timescale. Our analysis has the merit of emphasizing the technological performance required to realize QKD over such extreme transmission distances.
Comparison with heralded single-photon strategies.
We now compare pair-based protocols with single-photon protocols in which one of the photons emitted by the source (say, Alice’s photon) is detected immediately after emission. At the same time, the second one is sent to Bob. A pair source used in a heralded configuration provides a detection event that announces the photon’s emission time as it travels through the communication channel, allowing background clicks outside the heralding interval to be rejected [9]. Alternatively, Alice could choose to send an attenuated laser pulse with, on average, one photon per pulse, at a pulse rate equal to , using decoy-state methods to detect photon-number-splitting attacks [22] associated with the multiphoton components of weak coherent pulses [23, 24]. The main advantage of such single-photon protocols where the source is a (nearly) deterministic single-photon gun is that the effective bit rate is quite higher than in the case of symmetric QKD à la Ekert, where the transmission factor is squared. For instance, in the case of Moon-Earth transmission, Alice could send bits to Bob at a rate , to compare with the effective rate reached in the case of symmetric QKD à la Ekert (). Let us now assume that Bob and Charles, both on Earth, wish to refresh a shared one-time pad key using photons emitted by a source operated by Alice on a satellite or on the Moon. If Alice acts as a trusted relay and has access to the distributed key, its secrecy depends on her honesty. A separate concern arises if Alice also supplies measurement devices whose operation Bob and Charles cannot verify. Such devices could report correlated outcomes generated by a software program containing preregistered shared "hidden variables" known to Alice [25, 26, 27, 28]. Bob and Charles could then obtain an apparent key whose contents are already known to her. Observing these correlations alone would therefore be insufficient to establish security. Entanglement-based QKD can accommodate an untrusted source when the users’ measurement devices are trusted. If those devices are also untrusted, device-independent security requires assessing a Bell violation without the relevant loopholes, along with the assumptions and security analysis of the chosen protocol.
In the case of QKD à la Ekert, the violation of Bell’s inequalities can support security claims, subject to the protocol’s assumptions and its security analysis. Moreover, in the ideal case of maximally entangled pairs, the owner of the source has no information about the key shared by Bob and Charles, by virtue of the monogamy of entanglement. This property is closely related to the idea of device independence [29, 30]. It is, however, beyond the scope of the present paper to develop this analysis.
Pair-emitting markers for optical authentication
Correlation marking in the needle-in-a-haystack regime
Detecting weak optical markers in background light is one application of temporal postselection that we investigate in the ARTEMIS project [31, 32]. Earlier work has examined strong-coupling effects for optical marking [32] and the storage and revival of Bell nonlocality through emitter geometry and environmental memory [33]. Existing anticounterfeiting labels encode information in several measurable properties. Random fluorescence patterns can identify individual labels, as demonstrated with inkjet-printed quantum dots forming flower-like structures authenticated by image recognition [34]. Measurements under different illumination conditions provide further checks, including the nonlinear dependence of photoluminescence on excitation power for quantum dots encapsulated in a transparent polymer [35], and polarization-dependent patterns from fluorescent nanodiamonds authenticated using deep metric learning [36]. Other markers carry information in their composition. Examples include isotope-labeled glucose added to pharmaceutical formulations and identified by mass spectrometry [37], and engineered DNA mixtures read through hybridization as fluorescent patterns on paper-based tickets [38].
Here, the measured signature is the temporal correlation between photons originating from the same emission event. We consider an optically excited photon-pair-emitting marker whose emission is collected into two detection channels. Selecting coincidences within a short time window reduces accidental associations between independent background detections. Provided that the window still accommodates the timing uncertainty of genuine pairs, narrowing it can reveal the marker’s contribution even when the excess intensity is difficult to resolve in either singles record.
The interpretation of mixed events, however, differs from the QKD case. A coincidence formed by one marker photon and one background click carries no reliable qubit correlation and therefore contributes to the QKD error rate. In an on–off marker measurement, however, the same event remains associated with the presence of the illuminated marker and contributes to the measured coincidence excess. A false marker indication instead arises from coincidences that persist in the marker-absent reference, such as background–background events. This interpretation assumes that the background statistics remain unchanged between the marker-present and marker-absent measurements.
In anticounterfeiting, it is assumed that we excite a sample containing markers and that the emitted photons are detected in the vicinity of the sample. As always, we include the short propagation path between the sample and the detectors in the total registration probabilities and . Moreover, as the signal is carried by the presence of a photon and not by its qubit degree of freedom (polarisation, time bins and so on), it suffices, to reveal the presence of pair emitters, to estimate the difference between the (simultaneous pair) count rate in the presence of emitters and the count rate in their absence.
The detected genuine-pair contribution is, as always
while coincidences containing one detected marker photon and one background click occur at the rate
In the absence of the marker, the reference coincidence rate is
We consider markers that are sufficiently dilute for their emission to remain weak. In this regime, higher-order pair emission can be neglected, and the marker-dependent excess is
and we quantify its visibility through the rate contrast
We adopt here the definition of introduced in Sec. II through the registered background rate. Therefore, the detected background rates are and . Fluorescence clicks and environmental noise may be included through the general registered rates and , having in mind that they do not follow the optical-efficiency scaling in and as the dark count. Hence,
Substituting these rates into Eq. (32) gives
The first term originates from genuine marker pairs and displays the same inverse-window enhancement as the pair contribution in the QKD analysis. The second term arises from mixed marker–background events. Because both the mixed contribution and the marker-absent reference rate are proportional to , their ratio is independent of the window width.
The agreement between the theoretical prediction of Eq. (34) and the experimental results is examined in Sec. V and in the appendix. The relative deviation between the theoretical and experimental values is approximately 10% for the contribution associated with the first term of Eq. (34), as discussed in connection with Eq. (39), and approximately 4.1% for the mixed marker–background contribution described by the second term, as shown by Eqs. (51) and (52).
Extension to photon multiplets
The temporal advantage becomes stronger when an accepted event requires more than two correlated detections [39]. Let us now consider a source that produces simultaneous -photon events at a detection rate , with one photon registered in each of the channels. The probability of a genuine event within a window of width is . By contrast, an accidental -fold coincidence requires, in first approximation (neglecting here corrections due to departures from ideality), one independent background click in every channel. For registered single photons background rates , its probability is
and the corresponding accidental rate is
As long as the postselection window retains the genuine multiplet, the ratio between genuine and accidental event rates therefore scales as
A true triplet source3 consequently yields a enhancement, while a genuine four-photon source yields a enhancement. A fourfold event arising from two independent photon pairs scales differently because the desired signal itself requires two emission events within the same time window. Its probability is proportional to , whereas four independent background detections occur with probability proportional to . After conversion to rates, their ratio scales as , rather than the scaling of a genuine four-photon emission event. More generally, the temporal advantage is determined by the difference between the number of independent events required to generate the desired signal and the number required to reproduce an uncorrelated background.
Higher-order postselection can therefore provide increasingly strong rejection of an independent background. It is beyond the scope of the present paper to study in depth the final gain in SNR. One should, however, keep in mind that the potential gain for photon multiplets with larger than 2 is accompanied by a rapidly decreasing detected signal, since the registration probability is the product of the creation, transmission, and detection efficiencies of all participating photons. Candidate multiplet sources include cascaded radiative transitions and higher-order nonlinear optical processes [41, 42, 43], which are generally characterized by very small coupling constants; consequently, we expect the creation rate to decrease drastically for photon triplets, quadruplets, and so on.
Experimental demonstration of optical marking with pairs of simultaneous photons.
Experimental setup and measurement protocol
To implement our ideas on marking and anti-counterfeiting, we used the Thorlabs EDU-QOP1 pedagogical kit to generate pairs of identical photons that are uncorrelated in polarization but correlated in time [44]. The experimental arrangement is shown in Figure 2. The down-converted photons are emitted near 810 nm into two spatially separated arms.

Figure 2. Simplified schematic of the experimental setup used to generate and detect temporally correlated photon pairs. A continuous-wave laser at 405 nm pumps a type-I (-BBO) crystal, producing photon pairs near 810 nm. The photons are collected in two separate arms and detected by single-photon counting modules A and B. Electrical pulses from the detectors are recorded on separate channels of a time-tagger, and the acquisition data are transferred to a computer for temporal postselection analysis.
The electrical pulses produced by the two detectors are sent to separate inputs of a time tagger, denoted channels and , and the resulting detection times are transferred to a computer for post-processing. The measurement consists of recording the singles and coincidence rates reported by the software as the width of the temporal postselection window is varied from 3 ns to 300 µs. Before applying the coincidence criterion, the relative delay between the two detection channels is compensated so that detections originating from the same photon pair are centered at zero relative delay. A detection in channel is then associated with a detection in channel when the two compensated time tags fall within the selected window. It should be, however, noted that the pedagogical setup has several limitations. For instance, the dark count rate is high compared to more sophisticated and precise devices. The quantitative fits presented below use window widths of 3 ns and above. Moreover, we had no access to statistics on the timing of individual clicks, which remained hidden in the “black box” of the time-tagger. It was thus not possible to apply the strategy described in § 1, that is to say, to retain a window only when exactly one click is registered at Alice’s side, and exactly one click is registered at Bob’s side. This is because the kit’s time tagger systematically counts and integrates, for each detector, the number of successive clicks within a given window. When there is at least one click in each detector within the same temporal window, it also accumulates the number of clicks across both detectors. Finally, the software delivers the rates of single clicks, as well as the rates of coincidences established in this manner. At long windows, the software counting procedure differs from the coincidence protocol modeled in Sec. II; its results cannot be interpreted by the sparse-window equations. However, for short windows, multiple clicks per channel within one window are rare. The measured rates approach the predictions of that model. Now, it is precisely in the sparse regime that we expect our approach to be useful for increasing the SNR, so that in the regime of short temporal windows, the experiment can be seen as a proof of principle, which validates the theoretical predictions of previous sections.
An overview of the four experimental configurations, together with the fit results analyzed below, is provided in Table I. The first measurement is performed with strong ambient illumination and the pump laser switched off (background noise only). The second is performed with the pump operated at 35 mA and the ambient illumination removed (signal plus dark counts only). The remaining two measurements are performed with the pump and ambient illumination present simultaneously, at pump currents of 25.4 mA (weak signal) and 35 mA (strong signal). For each configuration, the singles and coincidence rates are recorded across different postselection-window settings, without changing the optical alignment or detector settings. The acquisition software reports the singles rates and in the two detection channels and a quantity labeled , which we denote by . This quantity includes contributions from genuine photon pairs and accidental temporal associations between unrelated detections. We retain this software-reported rate throughout the experimental analysis.
| Configuration | (kcps) | (kcps/ns) | (kcps/ns) | RMSE (kcps) | |
| Ambient on, pump off | 0.01866 | 0.01001 | 0.99981 | 0.0221 | |
| Ambient off, pump 35 mA | 74.713 | 0.51308 | 0.99479 | 3.583 | |
| Ambient on, pump 25.4 mA | 9.7989 | 0.07884 | 0.99586 | 0.523 | |
| Ambient on, pump 35 mA | 84.468 | 0.69327 | 0.99619 | 4.572 |
Table I. Quadratic fits of the retained-click rate over the interval from 3 ns to 300 ns. The rate is expressed in kcps and in ns. The coefficients provide a phenomenological description of the short-window data, while the RMSE quantifies the residual deviation from the fitted curve.
Window dependence and saturation
The results of table I are plotted in Figure 3 and Figure 8. The rates of individual detectors (single photons) remain approximately constant over the explored range of , whereas increases as a progressively larger fraction of temporally separated clicks is retained. At short time windows, the ambient-only configuration is strongly suppressed, while the pump-driven configurations retain a substantial contribution, reflecting the presence of temporally correlated photon-pair events.

Figure 3. Measured singles rates and , together with the retained-click rate , as functions of the postselection window . The four panels correspond to (a) ambient illumination on with the pump off, (b) ambient illumination on with the pump operated at 25.4 mA, (c) ambient illumination on with the pump operated at 35 mA, and (d) ambient illumination off with the pump operated at 35 mA.

Figure 8. Postselection-window dependence of the retained-click signal for the four experimental configurations. (a) Normalized fraction , where and are the measured singles rates in the two detection channels. The dashed horizontal line marks the saturation limit , reached when the postselection window becomes sufficiently large that essentially every detected click can be associated with a click in the opposite channel. (b) Same as the left figure but this time with absolute, not normalized pair rates for the same four configurations.
For large , the measured rate approaches the limiting value . In this limit, nearly all clicks are accepted by the software’s poor-matching procedure. This artificial saturation is outside the sparse-window regime of our theoretical model; the short-window measurements are the appropriate basis for comparison. The contrast between ambient-only and pump-driven measurements is already apparent in the regime of short temporal windows: below 100 ns (left portions of Figs. 3 and 8. With the pump switched off, is strongly suppressed at short windows despite singles rates of approximately 77 kcps in each channel. Activating the pump produces a much larger retained-click contribution at the shortest measured windows, consistent with significant production of twin photons. The persistence of this contribution under simultaneous pump and ambient illumination provides the experimental basis for detecting a weak source of simultaneous pairs hidden in a strong noisy background.
Short time-window behavior
To quantify the pre-saturation behavior, in the sparse regime of interest, we fitted the five measurements between 3 ns and 300 ns with the phenomenological expansion
QKD with pairs of maximally entangled simultaneous photons.
Experimental setup and measurement protocol
The coefficient captures the portion of the measured rate that varies weakly over the fitted interval and is therefore phenomenologically associated with the genuine pair production rate. The coefficient contains contributions from all kinds of accidental pairs. The quadratic term empirically accounts for the curvature away from the linear sparse-window behavior due to the aforementioned saturation effects. The measured short-window data and the corresponding quadratic fits are shown in Figure 7.

Figure 7. Pair detection rate in the sparse regime for the four experimental configurations. The symbols show the measured data, while the solid curves are phenomenological quadratic fits of the form over the interval .
The fitted parameters are summarized in Table I. In its column, kcps corresponds to ambient illumination only (pump off), kcps to the 35 mA pump only (ambient off), and kcps to that pump with ambient illumination. The weak-pump (25.4 mA) row is a separate configuration and is not included in the comparison in the Appendix. For ambient illumination alone, the estimate of the coincidence rate in the limit where the window width goes to zero is kcps, whereas the pump-driven measurements give values between 9.8 kcps and 84.5 kcps. This separation is consistent with the short-window behavior visible in Figure 7a: the ambient-only signal is dominated by coincidences that increase with the postselection window.
A more direct test is obtained from the data measured with the laser turned off (subpanel (a) in Figure 7 and first line of Table I). For short coincidence windows, we compare the measured rate with the value calculated from
For ns, we measured kcps and kcps, yielding kcps. The corresponding measured rate is 0.037 kcps. The relative difference is approximately 2.5%. This agreement supports our prediction that the measured rate results from accidental matching between statistically independent detections Eq. (39).
To generate polarization-entangled photon pairs, we used the polarization-entanglement extension [44] of the device described in the previous section.
The experimental arrangement is shown in Figure 4. A continuous-wave laser at 405 nm is directed by mirrors M1 and M2 through a half-wave plate (HWP) and a temporal compensation crystal (TCC) onto two crossed type-I -barium borate (-BBO) crystals. The polarization of the pump is adjusted with a half-wave plate oriented close to , so that the two crystals are excited coherently and generate the two-photon state:

Figure 4. Experimental setup for generating and measuring polarization-entangled photon pairs. A continuous-wave laser at 405 nm is directed by mirrors M1 and M2 through a half-wave plate (HWP) and a temporal compensation crystal (TCC) onto two crossed type-I (-BBO) crystals, producing photon pairs near 810 nm. The HWP sets the pump polarization, and the TCC and spatial compensation crystals (SCC) reduce temporal and spatial distinguishability between the two pair-generation processes. Each photon passes through a polarizer, an 810 nm band-pass filter, and an iris before reaching a single-photon counting module. The detector pulses are recorded on separate channels of a time tagger, with the acquisition data transferred to a computer for polarization-correlation and temporal postselection analysis.
where the phase depends on the orientation of the compensation crystals, TCC and SCC [5].
The twin photons are now correlated in polarization. We adjusted the phase to to approach the target state . The detection arrangement selects the measurement basis and the optical channels through which the photon state is probed [45]. We distinguish two analyzer configurations. Measurements with both analyzers set to probe the correlated outcome and the recovery of its intrinsic pair contribution relative to the ambient background. Separate measurements with analyzer set to and analyzer set to probe the outcome.
We model state-preparation imperfections by the Werner state of Eq. (17), with . The corresponding intrinsic probabilities are and . The separately added ambient illumination produces accidental coincidences and does not, by itself, change the state parameter . Thus, the intrinsic coincidence rate is a correlated-outcome rate, whereas the intrinsic rate probes an error outcome.
Correlated-outcome measurements in the configuration
We performed measurements in the configuration, with both polarization analyzers set to . Finite timing resolution prevents direct access to the zero-window limit. We therefore fit the measured in the sparse-window range using and extrapolate the fitted lines to . This intercept estimates the window-independent contribution, under the assumption that genuine pair detections are retained over the fitted range, while accidental associations scale approximately linearly with .
The four series in Figure 5 correspond to different nominal pump powers. With the pump off Figure 5(a), we measured a dark-room control and two additional controls with weak and strong thermal illumination. At 11.6 mW, Figure 5(b), measurements were performed in a dark room with weak and strong thermal illumination. The fitted intercepts are close to one another, with a mean of approximately 11.91 kcps. At 3.8 mW, Figure 5(c), the two runs used weak and strong thermal illumination. Their mean fitted intercept is approximately 3.99 kcps. At 1.3 mW, Figure 5(d), only one run was recorded, with strong thermal illumination. Its fitted intercept is approximately 1.38 kcps. For each background condition, a separate ordinary least-squares fit was performed using the three points at . When several runs were available at a given pump power, the quantity reported in Table II is the arithmetic mean of their fitted intercepts; for 1.3 mW, it is the intercept of the single fit. The fits were not constrained to share a common intercept.

Figure 5. Measured rates versus postselection-window width , with both polarization analyzers set to , at nominal pump powers of (a) 0, (b) 1.3, (c) 3.8, and (d) 11.6 mW. Symbols show measurements at , 10, and 30 ns under the background conditions indicated in each legend. Solid lines are separate linear fits to these three points, extrapolated to . Dashed vertical lines mark the shortest fitted window, 3 ns, and grey arrows indicate the extrapolation interval. The annotations report the mean fitted intercept across the background conditions at each power; panel (b) contains a single fit. Quoted uncertainties are formal fit-derived standard errors, propagated when averaging intercepts.
Standard errors of the individual intercepts were estimated from the residual covariance of the three-point unweighted fits. For independent runs, the standard error of their arithmetic mean was calculated as
The resulting regression uncertainties are approximately 0.060, 0.146, 0.072, and 0.232 kcps at 0, 11.6, 3.8, and 1.3 mW, respectively.
The normalized intrinsic rates approximately follow the relative pump powers, as shown in Table II. This supports
the recovery of a pump-dependent correlated-pair contribution despite varying ambient background. Combined with the separate measurements, they allow an estimate of the intrinsic error under the symmetry assumptions described below.
Error-outcome measurements in the configuration
A separate measurement series used in channel and in channel . In this series, all recorded pump-on runs used the same nominal power of 11.6 mW, while the ambient illumination was varied: dark room, small neon lights, large neon lights, and large neon lights supplemented by thermal illumination. The pump-off control used weak thermal illumination.
The pump-off control is shown in Figure 6(a). With the pump switched off and weak thermal illumination present, the measured rate increases approximately linearly with the postselection window. The control fit is constrained to pass through zero.

Figure 6. Measured rates versus postselection-window width , with the polarization analyzers set to and . Symbols show measurements at , 10, and 30 ns; solid lines are linear least-squares fits extrapolated to . (a) Pump-off control under weak thermal background, with the fitted intercept constrained to zero. (b) Measurements at a nominal pump power of 11.6 mW under the five background conditions indicated in the legend. A joint fit with a common intercept and a separate slope for each background gives kcps. The common intercept is imposed by the fitting model, and the quoted uncertainty is its formal fit-derived standard error.
Figure 6(b) shows the pump-on measurements at the fixed nominal power of 11.6 mW. Different ambient-background conditions produce different slopes, whereas the plotted lines share the intrinsic intercept reported by the analysis, kcps. The short-window dependence is represented by
where labels the ambient-background condition.
Let denote the intrinsic coincidence-event rate for outcome , inferred from short-window extrapolation. If all four rates are obtained under matched source conditions and the relative detection efficiencies are accounted for, the Werner-state model gives
To estimate from the available and measurements, we assume symmetry between the two correlated outcomes, , and between the two error outcomes, . We also assume comparable source brightness and effective detection efficiencies for the two measurement series at the same nominal pump power. Under these assumptions, Eq. (43) reduces to
At a nominal pump power of 11.6 mW, the fits give kcps, while the analysis in Figure 6(b) gives kcps. Consequently,
The estimated intrinsic probability of a single outcome is therefore approximately 1.3%. The intrinsic QBER includes both error outcomes, and , and is
Conclusions.
To conclude, experimental data confirmed our theoretical analysis: reducing the width of the postselection window increases the signal-to-noise ratio because the intrinsic signal is insensitive to this postselection, whereas the probability of accidental pair detections scales proportionally to the window width. For entanglement-based quantum key distribution, for instance, the extrapolated rates yield an estimated intrinsic QBER of approximately 2.6%. This information can guide strategies for improving retained-bit quality. Our analysis estimates the contributions of different noise sources and the potential benefit of improving experimentally controllable parameters.
Suppose that the intrinsic QBER in the Earth–Moon example of § 1.4 is 8%. From Eqs. (25) and (26), and . An accidental coincidence gives a random bit, which is wrong with probability ; its contribution to the total QBER for an ideal state is therefore . Including the 8% error, we get approximately 12%. The result exceeds the 11% upper bound mentioned above [12]. The security of the entanglement-based protocol requires an analysis appropriate to it. If the detection system could resolve while retaining genuine pairs, Eq. (15) predicts a tenfold reduction of the cross-pair rate; under unchanged source and channel parameters, the same simplified model gives a total QBER of approximately 8.4%, still within the safety threshold.
Regarding anti-counterfeiting, our ideas are also useful for experimentally revealing the presence of a very weak source of simultaneously emitted photon pairs. Even if, due to the presence of strong ambient noise and various sources of accidental simultaneous clicks (dark count, fluorescence, and so on), the genuine signal remains hidden within a noisy background, our method makes it possible to estimate the intrinsic pair-production rate. Our extrapolation method thus makes it possible to artificially eliminate the aforementioned sources of noise that contaminate our measurements.
Of course, the experimental results described in § 3.1 constitute only a proof of principle. The present experiment relies on a bright bulk-crystal SPDC source [5, 44], whereas practical anticounterfeiting and tagging schemes will likely require compact and integrable quantum-light platforms. Candidate approaches investigated within the ARTEMIS consortium include molecular lanthanide emitters produced by the University of Eastern Piedmont team in Vercelli [46], nonlinear molecular materials developed in Cagliari [47], and plasmonically engineered nanostructures explored by the Vigo team [48]. Other promising platforms include ultrathin nonlinear metasurfaces [49], resonant metasurfaces [50] and semiconductor quantum dots [51]. To upgrade our technique to the prototype level requires to characterize and to identify sufficiently bright yet hidden and discrete sources of simultaneous photon pairs (or even triplets or quadruplets, as explained in § 3). Obviously, the crystals used in our experiment are not easy to hide, in contrast to nanocomponents, which are smaller to begin with. Phase-matching - see Figure 1 is more difficult to realize in this case, and we expect that fewer pairs will be produced and that they will be scattered over a large range of directions. Actually, second harmonic generation has been observed in many nonlinear nanocomponents [52], but its inverse process, parametric down-conversion, is much less documented due to its intrinsic weakness [50]. Therefore, we also expect the signal to be hidden in a consequent background noise. As a first step, the extrapolation method of § 3.1 will be used to estimate the intrinsic pair rate while suppressing window-dependent accidental contributions. From this perspective, our approach can be considered a cornerstone of anticounterfeiting and marking based on paired sources.
Acknowledgments
We thank Benoit Chalopin for helpful discussions. Ce travail a bénéficié d’une aide de l’État gérée par l’Agence Nationale de la Recherche au titre de France 2030 et par l’Union européenne NextGenerationEU portant la référence ANR-22-CMAS-0001 (QuanTEdu-France). This work was carried out within the framework of the ARTEMIS project, “Molecular Materials for On-Chip Integrated Quantum Light Sources,” and was supported by the European Union through the EIC Pathfinder Challenges 2022 programme under Grant Agreement No. 101115149. The content reflects only the authors’ views, and the European Commission is not responsible for any use that may be made of the information it contains.
References
Mixed accidental pairs.
The results plotted in Figure 9 also provide another direct test of our theoretical model of § 2. They clearly establish that the rates of simultaneous events collected with the pump and the external source of ambient light turned on simultaneously is higher than the sum of their separate contributions, demonstrating the presence of accidental pairs involving one ambient-induced click and one pump-induced click. As always, at sufficiently large , all curves approach their respective saturation values as the matching procedure retains nearly all detected clicks.

Figure 9. Test of additivity for the correlation rate . The dashed blue and purple curves show the measurements obtained with ambient illumination alone and with the 35 mA pump alone, respectively. The green curve is the pointwise sum of these two separate measurements, while the red curve corresponds to the simultaneous presence of ambient illumination and the pump. The red curve systematically exceeds the green curve.
We denote the ambient-only singles rates by and , the pump-only rates by and , and the rates measured under simultaneous illumination by and . Because the independent retained-click slope is proportional to the product of the singles rates, the excess slope expected in the simultaneous acquisition is (34)
where the numerical prefactor converts singles rates in kcps and window widths in ns into a slope in kcps/ns. When the simultaneous singles equal the sums of the separately measured pump and ambient contributions, this expression reduces to the two mixed products proportional to .
Averaging the singles rates over the fitted interval gives
Equation (A1) then predicts
The same quantity extracted from the fitted slopes is
which differs from the singles-product prediction by approximately 4.1%. The corresponding nonadditivity over the full postselection-window range is shown in Figure 9. The agreement provides a quantitative consistency check that the excess window-dependent contribution under simultaneous illumination is governed predominantly by mixed associations between pump-induced and ambient detections.
This excess has different operational meanings in the two applications considered here. In QKD, a coincidence formed by one surviving pair of photons and one background click carries a random bit value and contributes to the error rate. In the marker measurement, the same term is absent when the pump-driven source is removed, thereby contributing to the marker-dependent on-off signal.
Detailed description of the data analysis
References
- [1]W. K. Wootters and W. H. Zurek, A single quantum cannot be cloned, Nature 299, 802 (1982).DOI
- [2]S. Wiesner, Conjugate coding, ACM SIGACT News 15, 78 (1983).DOI
- [3]C. H. Bennett and G. Brassard, Quantum Cryptography: Public key distribution and coin tossing, in Proceedings of the IEEE International Conference on Computers, Systems and Signal Processing (Bangalore, India, 1984) pp. 175–179, arXiv:2003.06557.
- [4]A. K. Ekert, Quantum cryptography based on Bell’s theorem, Physical Review Letters 67, 661 (1991).pubmed.ncbi.nlm.nih.gov/10044956
- [5]P. G. Kwiat, E. Waks, A. G. White, I. Appelbaum, and P. H. Eberhard, Ultrabright source of polarization-entangled photons, Physical Review A 60, R773 (1999).DOI
- [6]D. C. Burnham and D. L. Weinberg, Observation of simultaneity in parametric production of optical photon pairs, Physical Review Letters 25, 84 (1970).DOI
- [7]C. Holloway, J. A. Doucette, C. Erven, J.-P. Bourgoin, and T. Jennewein, Optimal pair-generation rate for entanglement-based quantum key distribution, Physical Review A 87, 022342 (2013).
- [8]R. H. Hadfield, Single-photon detectors for optical quantum information applications, Nature Photonics 3, 696 (2009).DOI
- [9]M. D. Eisaman, J. Fan, A. Migdall, and S. V. Polyakov, Invited review article: Single-photon sources and detectors, Review of Scientific Instruments 82, 071101 (2011).DOI
- [10]S. P. Neumann, T. Scheidl, M. Selimovic, M. Pivoluska, B. Liu, M. Bohmann, and R. Ursin, Model for optimizing quantum key distribution with continuous-wave pumped entangled-photon sources, Physical Review A 104, 022406 (2021).DOI
- [11]H. Kuniyil, S. Al-Kuwari, A. Ali, A. Czerwinski, and S. M. Arsian, Optimizing continuous-wave pumped entanglement-based quantum key distribution in a noisy environment, Physical Review A 113, 052613 (2026).DOI
- [12]N. J. Cerf, M. Bourennane, A. Karlsson, and N. Gisin, Security of quantum key distribution using d-level systems, Physical Review Letters 88, 127902 (2002).pubmed.ncbi.nlm.nih.gov/11909502
- [13]R. F. Werner, Quantum states with Einstein–Podolsky–Rosen correlations admitting a hidden-variable model, Physical Review A 40, 4277 (1989).pubmed.ncbi.nlm.nih.gov/9902666
- [14]S. Pirandola, R. Laurenza, C. Ottaviani, and L. Banchi, Fundamental limits of repeaterless quantum communications, Nature Communications 8, 15043 (2017).
- [15]A. Boaron et al., Secure quantum key distribution over 421 km of optical fiber, Physical Review Letters 121, 190502 (2018).DOI
- [16]S. P. Neumann, A. Buchner, L. Bulla, M. Bohmann, and R. Ursin, Continuous entanglement distribution over a transnational 248 km fiber link, Nature Communications 13, 6134 (2022).DOI
- [17]Y. Liu et al., Experimental twin-field quantum key distribution over 1000 km fiber distance, Physical Review Letters 130, 210801 (2023).DOI
- [18]T. Durt and J. Schneider, Quantum correlation tests at cosmic distances, Annales de la Fondation Louis de Broglie (2026), arXiv:2604.14252 [quant-ph].arxiv.org/abs/2604.14252
- [19]J. Yin et al., Satellite-based entanglement distribution over 1200 kilometers, Science 356, 1140 (2017).arxiv.org/abs/1707.01339
- [20]Y. Cao et al., Bell test over extremely high-loss channels: Towards distributing entangled photon pairs between earth and the moon, Physical Review Letters 120, 140405 (2018).DOI
- [21]M. A. Wayne, J. C. Biemfang, and A. L. Migdall, Low-noise photon counting above 100×106 counts per second with a high-efficiency reach-through single-photon avalanche diode system, Applied Physics Letters 118, 134002 (2021).DOI
- [22]T. Durt, Comment on “practical free-space quantum key distribution over 1 km”, Physical Review Letters 83, 2476 (1999).DOI
- [23]H.-K. Lo, X. Ma, and K. Chen, Decoy state quantum key distribution, Physical Review Letters 94, 230504 (2005).
- [24]B.-Y. Qiu et al., 7 km free-space time-bin quantum key distribution, Optics Express 33, 35176 (2025).DOI
- [25]C. H. Brans, Bell’s theorem does not eliminate fully causal hidden variables, International Journal of Theoretical Physics 27, 219 (1988).DOI
- [26]T. Durt, Three interpretations of the violation of bell’s inequalities, Foundations of Physics 27, 415 (1997).DOI
- [27]T. Durt, Comment on “pulsed energy-time entangled twin-photon source for quantum communication”, Physical Review Letters 86, 1392 (2001).
- [28]D. Aktas, S. Tanzilli, A. Martin, G. Pütz, R. Thew, and N. Gisin, Demonstration of quantum nonlocality in the presence of measurement dependence, Physical Review Letters 114, 220404 (2015).pubmed.ncbi.nlm.nih.gov/26196606
- [29]A. Acín, N. Brunner, N. Gisin, S. Massar, S. Pironio, and V. Scarani, Device-independent security of quantum cryptography against collective attacks, Physical Review Letters 98, 230501 (2007).DOI
- [30]B.-W. Lu, C.-W. Yang, R.-Q. Wang, B.-F. Gao, Y.-Z. Zhen, Z.-G. Wang, J.-K. Shi, Z.-Q. Ren, T. A. Hahn, E. Y.-Z. Tan, X.-P. Xie, M.-Y. Zheng, X. Jiang, J. Zhang, F. Xu, Q. Zhang, X.-H. Bao, and J.-W. Pan, Device-independent quantum key distribution over 100 km with single atoms, Science 391, 592 (2026).DOI
- [31]European Commission, ARTeMIS: Molecular materials for on-chip integrated quantum light sources, CORDIS, Grant Agreement No. 101115149 (2023).
- [32]M. Hatifi, D. Mara, B. Bokic, R. Van Deun, B. Stout, E. Lassalle, B. Kolaric, and T. Durt, Fluorimetry in the strong-coupling regime: From a fundamental perspective to engineering new tools for tracing and marking materials and objects, Applied Sciences 12, 9238 (2022).DOI
- [33]M. Hatifi, Geometry-controlled freezing and revival of Bell nonlocality through environmental memory, Physical Review A 113, 022204 (2026).
- [34]Y. Liu et al., Inkjet-printed unclonable quantum dot fluorescent anti-counterfeiting labels with artificial intelligence authentication, Nature Communications 10, 2409 (2019).DOI
- [35]M. J. Fong, C. S. Woodhead, N. M. Abdelazim, D. C. Abreu, A. Lamantia, E. M. Ball, K. Longmate, D. Howarth, B. J. Robinson, P. Speed, and R. J. Young, Using intrinsic properties of quantum dots to provide additional security when uniquely identifying devices, Scientific Reports 12, 16919 (2022).
- [36]L. Wang, X. Yu, T. Zhang, Y. Hou, D. Lei, X. Qi, and Z. Chu, High-dimensional anticounterfeiting nanodiamonds authenticated with deep metric learning, Nature Communications 15, 10602 (2024).
- [37]L. A. Felton, P. P. Shah, Z. Sharp, V. Atudorei, and G. S. Timmins, Stable isotope-labeled excipients for drug product identification and counterfeit detection, Drug Development and Industrial Pharmacy 37, 88 (2011).DOI
- [38]J. Li et al., Hybridization-encoded DNA tags with paper-based readout for anti-forgery raw material tracking, Nature Communications 16, 5832 (2025).
- [39]K. Laiho, T. Dirmier, M. Schmidt, S. Reitzenstein, and C. Marquardt, Measuring higher-order photon correlations of faint quantum light: A short review, Physics Letters A 435, 128059 (2022).DOI
- [40]Z. Chen and Y. Fainman, Coincidence detection theory for time-correlated photon sources, Optics Express 32, 43500 (2024).DOI
- [41]H. Hübel, D. R. Hamel, A. Fedrizzi, S. Ramelow, K. J. Resch, and T. Jennewein, Direct generation of photon triplets using cascaded photon-pair sources, Nature 466, 601 (2010).DOI
- [42]M. Corona, K. Garay-Palmett, and A. B. U’Ren, Experimental proposal for the generation of entangled photon triplets by third-order spontaneous parametric downconversion in optical fibers, Optics Letters 36, 190 (2011).
- [43]J.-W. Pan, Z.-B. Chen, C.-Y. Lu, H. Weinfurter, A. Zeilinger, and M. Żukowski, Multiphoton entanglement and interferometry, Reviews of Modern Physics 84, 777 (2012).DOI
- [44]Thorlabs, EDU-QOPA1 Quantum Optics Educational Kit and EDU-QOPA1 Polarization-Entanglement Extension Kit (2026).
- [45]M. Hatifi and B. Stout, Basis- and Channel-Selective Quantum Photodetection (2026).arxiv.org/abs/2605.16886
- [46]S. Bonabello, R. A. Nesta Silva, D. Mara, F. Travagin, M. Petreti, M. Saccone, M. Melegari, L. Marchiò, L. Pilia, G. B. Giovenzana, and F. Artizzu, Design of molecular lanthanide-based quantum light sources for on-chip integration, Advanced Optical Materials 14, e03824 (2026).DOI
- [47]S. S. Attar, F. Artizzu, L. Pilia, A. Serpe, A. Colombo, C. Dragnetti, F. Fagnani, D. Roberto, D. Marinotto, and P. Deplano, HX-linear and nonlinear optical responsiveness of rationally designed heteroleptic d8-metallo-dithiolene complexes, Molecules (Basel, Switzerland) 30, 4004 (2025).
- [48]E. Y. Santiagoo, M. Irfan, O. Ávalos-Ovando, A. O. Govorov, M. A. Correa-Duarte, and L. V. Besteiro, Light-to-matter chirality transfer in plasmonics, Materials Horizons 10, 1039/D5MH00179 (2025).DOI
- [49]J. Zhang, J. Ma, M. Parry, M. Cai, R. Camacho-Morales, L. Xu, D. N. Neshev, and A. A. Sukhorukov, Spatially entangled photon pairs from lithium niobate nonlocal metasurfaces, Science Advances 8, 10.1126/sciadv.abq4240 (2022).DOI
- [50]T. Santiago-Cruz, A. Fedotova, V. Sultanov, M. A. Weissflog, D. Arslan, M. Younesi, T. Pertsch, I. Staude, F. Setzpfandt, and M. Chekhova, Photon pairs from resonant metasurfaces, Nano Letters 21, 4423 (2021).
- [51]A. O. C. Hör, M. A. M. Versteegh, K. D. Jöns, and S. Ducci, Semiconductor devices for entangled photon pair generation: A review, Reports on Progress in Physics 80, 076001 (2017).
- [52]G. Bachelier, J. Butet, I. Russier-Antoine, C. Jonin, E. Benichou, and P.-F. Brevet, Origin of optical second-harmonic generation in spherical gold nanoparticles: Local surface and nonlocal bulk contributions, Physical Review B 82, 235403 (2010).DOI