---
title: Double Heralding for Quantum State Engineering
url: https://www.emergentmind.com/topics/double-heralding-technique
type: topic
---

# Double Heralding for Quantum State Engineering

Searching arXiv for the cited paper and closely related work on double heralding.
Double-heralding is a conditional state-preparation technique in which success is declared only when two heralding detection events satisfy a prescribed pattern. In the cited literature, the term appears in more than one technically distinct sense. In continuous-wave optical state engineering, it denotes conditioning a signal mode on two heralding clicks in the idler arm, with the relative click delay acting as a control parameter for temporal-mode structure [1511.02122]. In spontaneous four-wave mixing, it denotes conditioning on two herald photons in order to prepare a two-photon-number state, with herald-channel noise playing a decisive role [1602.04767]. In recent SPDC source design, it denotes direct detection of two signal photons from independent down-converters so as to project the corresponding idlers into an anti-correlated pair state that is unitarily equivalent to the resource produced by swap-heralded sources [2603.19489]. A broader Gaussian-state framework treats the same idea as photon-number-resolving detection on two heralding modes with prescribed outcomes \((n_1,n_2)\) [2604.25910].

## 1. Definition and terminological scope

In the continuous-wave experiment of Morin, Fabre, and Laurat, a successful preparation event consists of **two heralding detection events**, one on each branch of a split heralding arm, with the two clicks occurring at times \(t_1\) and \(t_2\) [1511.02122]. In the ideal number-correlated picture of a two-mode squeezed vacuum, two detections in the heralding mode project the signal mode into a two-photon state. In that work, this is what “two heralding events” or “double heralding” means.

In the SFWM work of Harder, Silberhorn, and coauthors, “double-heralding” means conditioning the signal-arm state on two click events in the herald/idler arm, with the goal of preparing a two-photon-number state in the signal arm [1602.04767]. Here the central operational claim is direct: if SFWM creates photon pairs, then detection of two herald photons suggests that two paired signal photons were also created.

In the SPDC source architecture of Shapiro and coworkers, the term denotes a protocol in which two single-photon heralds project the corresponding idler photons into an anti-correlated pair \(a_i^\dagger b_i^\dagger|0\rangle\), or its multimode generalization [2603.19489]. That paper presents the method explicitly as a “double-heralding technique” for producing heralded entangled photon pairs from SPDC.

Across adjacent literature, the same detection logic is sometimes present without the term itself. Integrated NOON-state generation on chip uses a **single-stage heralded generation using a two-mode coincidence herald**, not “double heralding” in the standard sequential sense [1005.5119]. Programmable photonic state synthesis distinguishes **one-mode heralding** from **two-mode heralding**, where the latter is a two-click heralding condition \(|1\rangle_{a1}|1\rangle_{a2}\) but is not labeled “double heralding” by the authors [2204.08788]. This suggests that the defining invariant is not a particular physical platform, but conditioning on a joint two-detection signature.

## 2. Continuous-wave double heralding as temporal-mode engineering

The most detailed experimental study of double heralding as a physical mode-engineering problem is the continuous-wave optical implementation based on a **continuous-wave two-mode squeezed vacuum state** generated by a type-II nondegenerate OPO operated far below threshold [1511.02122]. Signal and idler are orthogonally polarized and separated on a PBS. The idler serves as the heralding arm, is split on a balanced beam splitter, and is monitored by two single-photon detectors. Because the source is continuous-wave, the two heralding events need not occur at the same time, so the delay
\[
\Delta t = t_1-t_2
\]
becomes an additional physical parameter.

Each heralding detection at time \(t_i\) defines a normalized trigger temporal mode
\[
g_i(t) = \sqrt{\pi \gamma}\, e^{-\pi \gamma |t-t_i|}, \qquad i=1,2,
\]
where \(\gamma\) is the OPO bandwidth (FWHM) [1511.02122]. The overlap between the two trigger modes,
\[
I = \int g_1(t)g_2(t)\,dt = e^{-\pi \gamma |\Delta t|}\left(1+\pi\gamma |\Delta t|\right),
\]
is the key control parameter. When \(|\Delta t|\to 0\), \(I\to 1\), and the two trigger modes coincide. When \(|\Delta t|\gg 1/\gamma\), \(I\to 0\), and they are effectively orthogonal.

The heralded state conditioned on the two clicks is written as
\[
\left| \Psi_2 \right\rangle = \frac{1}{\sqrt{1+I^2}} \iint dt\,dt'\, g_1(t)g_2(t')\, \hat a^\dagger(t)\hat a^\dagger(t')\,|0\rangle,
\]
with the normalization factor \(1/\sqrt{1+I^2}\) arising because the two created excitations may occupy nonorthogonal temporal modes [1511.02122]. The paper then introduces the orthogonal symmetric and antisymmetric temporal modes
\[
f_1(t) = \frac{g_1(t)+g_2(t)}{\sqrt{2(1+I)}}, \qquad
f_2(t) = \frac{g_1(t)-g_2(t)}{\sqrt{2(1-I)}},
\]
which diagonalize the modal structure of the heralded state.

In that basis, the two-photon fidelity in either \(f_1\) or \(f_2\) is
\[
F = \frac{(1\pm I)^2}{2(1+I^2)} = \frac{1}{2}\pm \frac{I}{1+I^2},
\]
with \(+\) for \(f_1\) and \(-\) for \(f_2\), so \(f_1\) is the optimal mode for maximizing the two-photon fidelity [1511.02122]. This is one of the central conceptual points of continuous-wave double heralding: the target state is determined not only by the source and heralding pattern, but also by the choice of temporal mode used for state reconstruction or downstream processing.

Two limiting regimes organize the physics. In the small-delay limit \(\gamma\Delta t\to 0\), for the optimal symmetric mode \(f_1(t)\),
\[
F \simeq 1 - \left(\frac{\pi \gamma \Delta t}{2}\right)^4.
\]
The correction is fourth order in delay, which explains the observed plateau of high fidelity for short delays [1511.02122]. In the opposite limit \(\Delta t \gg 1/\gamma\), the overlap vanishes, and even in the optimal mode the two-photon fidelity saturates at
\[
F \to \frac{1}{2}.
\]
In that regime, the more natural interpretation is two separated single-photon wavepackets rather than a single-mode two-photon Fock state.

Experimentally, the source is a **continuous-wave type-II OPO below threshold**, pumped at \(532\) nm and operated at about \(2\%\) of threshold; the cavity bandwidth is measured as \(\gamma \simeq 53\,\mathrm{MHz}\), and the idler arm is detected with high-efficiency WSi SNSPDs whose detector system efficiency is \(85\%\) with dark counts below \(10\) counts/s [1511.02122]. The heralded state is characterized by homodyne tomography with temporal-mode selection performed in post-processing via
\[
x = \int \xi(t)\,x(t)\,dt.
\]
For nearly simultaneous clicks, with an acceptance window of \(0.8\,\mathrm{ns}\), the uncorrected two-photon fidelity reaches \(58\%\) at a heralding rate of \(200\,\mathrm{Hz}\); corrected for detection losses it becomes \(79\%\), limited mainly by the OPO escape efficiency [1511.02122]. For large delay, specifically \(\Delta t=40\,\mathrm{ns}\), the experiment observes approximately single-photon behavior in modes \(g_1\) and \(g_2\), with fidelity around \(76\%\), while the two-photon component in \(f_1\) and \(f_2\) drops to about half the zero-delay value, in agreement with the \(I\to 0\) limit [1511.02122].

A further distinction is drawn between an **optimal adapted temporal mode** and a **fixed temporal mode**. If one uses a fixed mode aligned with the first click, \(h_1(t)=g_1(t)\), then after tracing over all other modes the photon-number probabilities in that fixed mode are
\[
P_2 = \frac{2I^2}{1+I^2}, \qquad
P_1 = \frac{1-I^2}{1+I^2}, \qquad
P_0 = 0.
\]
For small delay, the two-photon fidelity in the fixed mode behaves as
\[
F \simeq 1 - \left(\frac{\pi\gamma\Delta t}{\sqrt{2}}\right)^2,
\]
so the degradation is second order rather than fourth order [1511.02122]. The conclusion is operationally sharp: in continuous-wave double heralding, temporal-mode matching is not optional.

## 3. Double heralding for two-photon-number-state preparation in SFWM

In the SFWM implementation, the physical objective is conditional preparation of a two-photon-number state by detecting two herald photons from a non-degenerate, third-order nonlinear-optical photon-pair source [1602.04767]. The source is a non-degenerate spontaneous four-wave-mixing source in birefringent fiber: single-mode birefringent optical fiber, Fibercore HB800, pumped by a mode-locked Ti:sapphire laser at \(802\) nm with \(36\) ps transform-limited pulses and repetition rate \(76\) MHz. The generated wavelengths are \(686\) nm for the signal and \(965\) nm for the idler/herald [1602.04767].

The herald arm is split by a \(50/50\) beamsplitter and detected by two multimode-fiber-coupled APDs. A double-herald event is declared when both herald detectors click for the same pump pulse [1602.04767]. The signal arm is then analyzed by a spatially multiplexed three-detector apparatus, and the photon-number distribution is inferred by inverting the conditional click-probability matrix
\[
\begin{pmatrix} q_1 \\ q_2 \\ q_3 \end{pmatrix}
=
\begin{pmatrix}
P(1|1) & P(1|2) & P(1|3) \\
0 & P(2|2) & P(2|3) \\
0 & 0 & P(3|3)
\end{pmatrix}
\begin{pmatrix} p_1 \\ p_2 \\ p_3 \end{pmatrix},
\]
with vacuum inferred from
\[
p_0=1-(p_1+p_2+p_3).
\]
This gives reconstructed conditional photon-number probabilities rather than a homodyne-reconstructed density matrix [1602.04767].

The main technical issue is herald-channel noise. If the probability that a herald click is due to the target process is \(\eta_r\), and due to noise is \(1-\eta_r\), then conditioned on observing \(k_r\) herald events the signal distribution is
\[
P(n_s|k_r) =\sum_{m=0}^{k_r}\binom{k_r}{m}\eta_r^{m}\left(1-\eta_r\right)^{k_r-m}\bar{P}_{sr}(n_s|m),
\]
where \(\bar{P}_{sr}(n_s|m)\) is the noise-free conditional distribution [1602.04767]. For double heralding, the terms \(m=2\), \(m=1\), and \(m=0\) correspond respectively to two true heralds, one true herald plus one noise click, and two noise clicks.

The paper’s central result is that noise photons degrade conditional preparation of two-photon number states more than they degrade conditional single-photon states [1602.04767]. In the low-gain scaling argument, if \(\epsilon_{Pair}=\epsilon_{Noise}=1/100\), then for single heralding the conditional probability of having a signal photon is \(0.5\), but for double heralding the probability of having two signal photons given two herald detections is \(0.25\), while the conditioned single-photon probability is \(0.5\) [1602.04767]. The dominant contamination is therefore not merely extra vacuum, but a substantial one-photon component.

The paper derives the equivalence between herald-channel noise and effective loss in the signal channel. The noisy-herald conditional distribution can be written in the same form as a fictitious signal-loss model with
\[
\eta_r = 1-\rho_{Loss},
\]
so false heralds behave as though the signal channel suffered an effective loss [1602.04767]. This provides a practical route to correction through the Klyshko heralding efficiency
\[
\eta_{K} = \frac{P(H,S)}{P(H)},
\]
which in the noise-free case equals the signal efficiency, while with herald noise becomes
\[
\eta_K = \eta_S \frac{\gamma}{\gamma+\sigma}.
\]
Fitting the measured power- and length-dependent Klyshko efficiencies, the authors obtain
\[
\eta_{System} = 41.95\%, \qquad \beta = 5.324\times10^{-6}\ \text{mW}, \qquad \alpha=344\ \text{mW-meters},
\]
and conclude that Raman contribution is negligible compared to leaked pump photons [1602.04767].

Experimentally, the reconstructed distributions show that with double heralding the two-photon contribution increases, but a strong one-photon contribution remains [1602.04767]. The paper therefore establishes a general constraint on double-heralded Fock-state preparation with pair sources: if herald noise is not much smaller than true pair generation, the conditioned state can be far from a true two-photon state.

## 4. Double heralding for SPDC entangled-pair sources and multiplexing

A newer use of the term appears in SPDC source engineering, where the aim is not a two-photon-number state in one mode but a heralded entangled-photon-pair resource [2603.19489]. The central claim is that the useful resource hidden inside earlier swap-heralded schemes is a heralded **anti-correlated two-photon idler state**, and that once this is recognized it can be generated more directly: use independent SPDC/TMSV sources, detect two signal photons, and thereby herald the corresponding two idler photons in orthogonal modes.

In the minimal implementation, one TMSV source emits modes \((a_s,a_i)\), the other emits \((b_s,b_i)\), with detectors placed directly on \(a_s\) and \(b_s\). Success is a **single-photon detection in each signal arm**, and conditioned on that event the idler state is projected onto
\[
a_i^\dagger b_i^\dagger |0\rangle,
\]
an anti-correlated pair in the dual-rail basis [2603.19489]. The paper shows that this state is unitarily equivalent to the resource produced by prior swap-heralded sources.

For a single TMSV source, the pair-number distribution is
\[
P_1(n)=(1-\lambda)\lambda^n.
\]
For the basic double-heralded TMSV source, the probability that each of two independent TMSV sources emits one pair is
\[
P_1(1)^2 \le \left(\frac12\right)^4=0.0625.
\]
By contrast, the prior swap-heralded dual-TMSV source yields a maximum heralding probability from a valid two-pair event of
\[
P_2^{(2)} \le \left(\frac{2}{3}\right)^6 \approx 0.088.
\]
The paper interprets this advantage as a multiplexing-like gain purchased with roughly greater hardware overhead, since the double-heralded source uses fewer heralding modes, fewer mode-sorting optics, and fewer detectors on the heralding path [2603.19489].

The method is then generalized to an \(M\)-TMSV array. The probability of producing exactly one pair across the array is
\[
P_M = M(1-\lambda)\lambda,
\]
optimized at
\[
\lambda=\frac{1}{M+1}, \qquad \mu=\frac{1}{M},
\]
and the maximal double-herald probability from two \(M\)-TMSV arrays is
\[
P_M^{(2)} = \left(\frac{M}{M+1}\right)^{2M+2},
\]
which tends to \(1/e^2\approx 0.135\) as \(M\to\infty\) [2603.19489].

The detector model is explicit. A detector on a heralding mode has POVM
\[
E^{(m)}=\sum_{n=0}^\infty p(m|n)\,|n\rangle\langle n|,
\]
with a quasi-PNR realization based on a balanced \(1\times K\) splitter feeding \(K\) ideal SPDs. The corresponding parameters are
\[
\alpha=\frac{K-1}{K}, \qquad
\delta_0 = 1-(1-p_d)^K, \qquad
\delta_1 = K p_d(1-p_d)^{K-1}, \qquad
\delta_2 = 1-(1-p_d)^{K-1},
\]
where \(\alpha\) quantifies “PNR availability” [2603.19489]. In this model, the fidelity of the heralded state to the desired anti-correlated pair is
\[
F = \frac{P_M^{(1,1)}}{p_M^{(1,1)}},
\]
and with ideal PNR and no excess noise it simplifies to
\[
F_{\mathrm{PNR}} = \left(\frac{1+\eta\mu}{1+\mu}\right)^{2+2M}.
\]
With excess noise, even at \(\eta=1\), the maximum fidelity is bounded by
\[
F \le \left(\frac{1-\delta_2}{\delta_1+1-\delta_2}\right)^2 \approx 1-2\delta_1.
\]
These expressions formalize the expected tradeoff between brightness, loss, dark counts, and partial number resolution [2603.19489].

Two multiplexing strategies are analyzed. In complete active multiplexing, if one elementary block heralds a single photon with probability \(p_M\), the probability that at least two out of \(N\) blocks succeed is
\[
P_M^{(2)}(N) = 1-(1-p_M)^N - N p_M (1-p_M)^{N-1}.
\]
In bipartite multiplexing, appropriate to a ZALM-like architecture, the success probability becomes
\[
P_M^{(2)}(N) = \left[1-(1-p_M)^{N/2}\right]^2.
\]
The paper’s bottom-line design conclusion is that, although increasing \(M\) can reduce active switching burden, the total source/detector count \(MN\) is minimized at \(M=1\); thus the most resource-efficient implementation is plain double-heralded TMSV [2603.19489].

## 5. General multimode Gaussian formulation with two heralding modes

A general theoretical treatment of double heralding is given for multimode Gaussian states with photon-number-resolving detection on two heralding modes [2604.25910]. The resource is a pure multimode Gaussian state with zero coherent displacement,
\[
|G\rangle =Z^{1/4} \exp\left(\frac{1}{2} \sum_{j=0}^m \sum_{k=0}^m A_{jk} \hat{a}_j^{\dagger}\hat{a}_k^\dagger\right) |\mathrm{vac}\rangle,
\]
with
\[
Z=\det(I-AA^\dagger), \qquad I-AA^\dagger >0.
\]
When there are two heralding modes, the successful event is a joint outcome \((n_1,n_2)\) on those two modes, and the remaining one or two signal modes are projected into a non-Gaussian target state [2604.25910].

For a single-signal-mode output, the total herald count
\[
N=\sum_{j=1}^m n_j
\]
sets the stellar rank and upper Fock cutoff of the conditional output,
\[
|\psi_N\rangle=\sum_{k=0}^N c_k |k\rangle.
\]
Because the input Gaussian state has zero displacement and hence even total parity, the heralded state has parity fixed by \(N\): if \(N\) is even then only even Fock components survive, and if \(N\) is odd then only odd components survive [2604.25910]. This parity constraint is especially consequential for double heralding because the two-mode detection pattern \((n_1,n_2)\) fixes the output parity immediately.

The resource matrix is decomposed as
\[
A= \Lambda B \Lambda,
\]
where \(\Lambda=\mathrm{diag}(1,x_1,\dots,x_m)\), \(x_j>0\), and \(B\) contains the state-shaping parameters [2604.25910]. The heralding probability for a pattern \((n_1,\dots,n_m)\) is
\[
p_S=\frac{N!}{|c_N|^2} \left[\det(I-AA^\dagger)\right]^{1/2}\prod_{j=1}^m \frac{x_j^{2n_j}}{n_j!}.
\]
Introducing \(X_j=x_j^2\) and \(K=\Lambda^2\), the optimization reduces to maximizing
\[
\tilde{p}=\det\left(I-B K B^\dagger K\right ) \prod_{j=1}^m X_j^{2n_j},
\]
subject to polynomial stationarity equations
\[
\frac{\partial\tilde{p}}{\partial X_j}=0.
\]
This is the formal reason the paper can treat double-herald optimization as a system of polynomial equations [2604.25910].

The paper’s main explicit worked examples are double-heralding cases. For \(m=2\), the Gaussian resource is a three-mode state with one signal mode and two heralding modes,
\[
|G\rangle=Z^{1/4}\exp\left( x_1\hat{a}_0^\dagger \hat{a}_1^\dagger+x_2\hat{a}_0^\dagger \hat{a}_2^\dagger+x_1x_2\nu \hat{a}_1^\dagger \hat{a}_2^\dagger +\frac{s_1 x_1^2}{2}\hat{a}_1^{\dagger 2}+ \frac{s_2 x_2^2}{2}\hat{a}_2^{\dagger 2}  \right) |0,0,0\rangle.
\]
For the balanced target
\[
|\psi_6\rangle=\frac{1}{2}\left(|0\rangle+|2\rangle+|4\rangle+|6\rangle\right),
\]
the optimal success probabilities reported are \(p_S=0.013201\) for pattern \((3,3)\) and \(p_S=0.010586\) for pattern \((4,2)\) [2604.25910]. For
\[
|\psi_7\rangle=\frac{1}{2}\left(|1\rangle+|3\rangle+|5\rangle+|7\rangle\right),
\]
the reported optima are \(p_S=0.006458\) for pattern \((4,3)\) and \(p_S=0.008903\) for pattern \((5,2)\) [2604.25910]. A major implication is that different two-herald patterns with the same total herald count can have substantially different optimal probabilities.

The framework also treats two-signal-mode outputs. For the single-rail entangled target
\[
|\Phi(w)\rangle=\frac{1}{\sqrt{1+w^2}}\left(|11\rangle+w|00\rangle\right),
\]
the paper states explicitly that one heralding mode is insufficient and that **two heralding modes with heralding pattern \((1,1)\) are required** [2604.25910]. This extends double heralding beyond single-mode Fock-superposition synthesis to two-mode entangled-state generation.

## 6. Related schemes, misconceptions, and practical boundaries

A recurrent source of confusion is that not every two-click conditional protocol is called double heralding in its native literature. The integrated-photonics work on path-entangled NOON-state generation uses detection of one photon in each of two auxiliary modes \(i,l\) as a **single joint event**, and the authors stress that this is better described as **single-stage heralded generation using a two-mode coincidence herald** than as standard double heralding [1005.5119]. Likewise, the programmable linear-optics work compares **one-mode heralding** \(|2\rangle_a\) and **two-mode heralding** \(|1\rangle_{a1}|1\rangle_{a2}\), with the latter being the closest analogue of double heralding inside passive state-synthesis networks [2204.08788].

A second misconception is to equate all heralding with click heralding. “Heralding on zero photons” shows that vacuum-conditioning can also be implemented with ordinary non-number-resolving detectors and time-tagging. Its core formulas are
\[
p(NC|n)=(1-d)(1-\eta)^n,
\]
\[
P_s=\sum_{n=0}^\infty(1-d)(1-\eta)^n P_n,
\]
and
\[
P(\psi_{out}|NC)=\frac{P_0}{\sum_{n=0}^\infty(1-\eta)^n P_n},
\]
so inefficiency, rather than dark counts, becomes the main fidelity limitation for no-click heralding [2107.03468]. This is directly relevant to any double-heralding architecture that combines click and no-click conditions.

A third boundary concerns canonical versus adjacent uses. A cascaded SPDC source based on entanglement swapping employs **one composite heralding condition**: a successful Bell-state measurement in the swap stage, typically a specific two-click coincidence signature [2107.14360]. The paper itself says this is adjacent to, but not an instance of, canonical double heralding. A similar caveat applies to the 2026 rate analysis of heralded versus unheralded SPDC entanglement distribution: both heralded architectures there condition on two idler detections per successful event, but the work is about coincidence-heralded entanglement distribution rather than “double heralding” as a named protocol [2603.06423].

Finally, practical lessons recur across all versions of the technique. Conditional acceptance improves state quality but imposes a success-probability tax [1511.02122; 1602.04767; 2603.19489]. Loss, dark counts, and nonideal number resolution create false heralds or erase the benefit of conditioning [1602.04767; 2603.19489]. In continuous-wave operation, the relevant control variable is often the temporal separation of the two heralds and the associated overlap \(I\) [1511.02122]. In multimode Gaussian engineering, the decisive variables are the heralding pattern \((n_1,n_2)\), the state-shaping parameters in the Gaussian core, and the available squeezing budget [2604.25910]. Taken together, these results indicate that double heralding is best understood not as a single protocol family, but as a general two-event conditional-measurement strategy whose precise meaning depends on whether the target is a Fock state, a temporally engineered state, or an entangled dual-rail resource.

Source: https://www.emergentmind.com/topics/double-heralding-technique