---
title: 'Cross-Island Heralding: Multi-Domain Insights'
url: https://www.emergentmind.com/topics/cross-island-heralding
type: topic
---

# Cross-Island Heralding: Multi-Domain Insights

Searching arXiv for the provided papers and related usage of "cross-island heralding".
Cross-island heralding is a term that appears in several technically distinct research contexts, each centered on a success signal that links separated “islands” in the relevant state space, hardware architecture, or physical geography. In spectral-multiplexed quantum optics, it denotes a heralding event in which the \(H\)-click comes from island \(n\) and the \(V\)-click comes from a different island \(m\neq n\), extending heralding beyond same-island events and changing the scaling of entanglement-distribution rates [2507.14427; 2603.06423]. In superconducting quantum networking, a related protocol heralds end-to-end entanglement with one detected photon and then uses teleportation, so that low optical-microwave coupling efficiency is absorbed into a probabilistic heralding step rather than a deterministic transduction path [2012.13408]. The phrase is also used for migration among subpopulations in island-model genetic programming [2606.28381] and, in a speculative hydrodynamic sense, for navigational wave patterns bridging distant islands in Marshallese navigation [1802.09151].

## 1. Terminological scope and core definitions

In the spectral-island SPDC literature, an islands-based source has \(N_I\) spectrally-factorable “islands,” each capable of emitting a signal-idler pair. A same-island heralding event is written \(\mathcal{H}_{nn}\), meaning “one \(H\) and one \(V\) click in island \(n\).” Cross-island heralding is written \(\mathcal{H}_{nm}\), meaning “\(H\) click in island \(n\), \(V\) click in island \(m\),” with \(m\neq n\). Allowing \(n\ne m\) increases the heralding rate by a factor of order \(N_I^2\) instead of \(N_I\), at the cost of needing to frequency-convert and route two different islands’ signals [2603.06423].

In the modified ZALM architecture, the same distinction is expressed as same-island heralds (SIH) and cross-island heralds (CIH). A SIH on island \(n\) consists of exactly two clicks from CWDM-resolved island \(n\), one in an \(H\)-detector and one in a \(V\)-detector. A CIH on islands \((m,n)\) with \(m\neq n\) consists of exactly two clicks, one from island \(m\)’s \(H\)-detector and one from island \(n\)’s \(V\)-detector; these events herald a polarization-Bell state across \(S_{1,m}\) and \(S_{2,n}\) with a known relative phase [2507.14427].

In Evolutional Math, “cross-island heralding” is the system’s label for migration among GP islands. Migration occurs every \(M\) generations on a directed ring, with the top \(K\) individuals sent from island \(i\) to island \((i+1)\bmod N\), replacing the lowest-fitness \(K\) individuals at the destination [2606.28381].

In Harvey’s model of Marshallese navigation, the phrase is attached to the dilep: a signal apparently providing guidance directly between two distant islands. There, the mechanism is not discrete detection but constructive interference between two reflected secondary swells [1802.09151].

| Context | “Island” denotes | Heralding event |
|---|---|---|
| Spectral SPDC / ZALM | Spectrally factorable frequency bin | One \(H\) click and one \(V\) click from different islands |
| Superconducting quantum networking | Remote quantum nodes | One detected optical photon announcing entanglement |
| Island-model GP | Parallel subpopulation | Periodic elite migration between islands |
| Marshallese navigation | Physical islands \(A,B\) | Interference ridge bridging island pairs |

## 2. Spectral-island photonic architectures

The most explicit and mathematically developed use of cross-island heralding is in spectral-multiplexed SPDC. The source model assumes a domain-engineered \(\chi^{(2)}\) crystal producing \(N_I\) well-separated, spectrally factorable islands. The two-photon joint spectral amplitude is
\[
\Psi(\omega_S,\omega_I)=\sum_{n=1}^{N_I}\phi_n(\omega_S)\,\psi_n(\omega_I),
\]
with normalized, non-overlapping island modes satisfying
\[
\int |\phi_n(\omega)|^2 d\omega = \int |\psi_n(\omega)|^2 d\omega = 1,\quad
\langle \phi_n|\phi_m\rangle=\langle \psi_n|\psi_m\rangle=\delta_{nm}.
\]
In the no-pump-depletion regime, each Sagnac source outputs a product of \(N_I\) identical two-mode squeezed-vacuum states, one per island and per polarization, with squeezing parameter \(r\), \(\lambda\equiv\tanh r\), \(G=\cosh^2 r\), and \(G-1=\sinh^2 r\) [2507.14427].

The heralding hardware is a partial Bell-state-measurement module. The two idler outputs \(I_1\) and \(I_2\) are overlapped on a \(50{:}50\) beamsplitter, then split by a PBS into \(H,V\) ports. A coarse WDM after each port resolves the \(N_I\) islands, and each CWDM output goes to a single-photon detector with quantum efficiency \(\eta_T\) and partial number resolution \((0,1,>1)\). Herald logic accepts any exactly-two-click pattern consisting of one \(H\) and one \(V\) click, and labels the islands and outputs \((+/-)\) to distinguish singlet versus triplet outcomes [2507.14427].

For the dual-Sagnac ZALM analysis, the per-island probability to register one \(H\)-click and one \(V\)-click from island \(n\) is
\[
p_1 \equiv \Pr\{\text{two clicks from island }n\}
= \frac{4[\eta_T(G-1)]^2}{[\eta_T(G-1)+1]^6}.
\]
The same-island herald probability per pulse is
\[
P_{\text{same}}=\tfrac12\left\{1-[1-p_1]^{N_I}\right\},
\]
where the factor \(1/2\) reflects that half of the events are “true” and half are “false.” The cross-island herald probability per pulse is
\[
P_{\text{cross}}=\tfrac12\left\{1-[1-p_1]^{N_I^2}\right\},
\]
because any ordered pair \((m,n)\) can supply \(H\) from \(m\) and \(V\) from \(n\) [2507.14427].

The same general concept is treated in the comparative analysis of islands-based ZALM and islands-based signal-path erasure (SPE). There, each SPDC island has gain \(G\), mean pair number \(\bar n=G-1\), heralding efficiency \(\eta_h\), and signal-path transmissivity \(\eta_R\). The probability of a cross-island herald \(\mathcal{H}_{nm}\) is built from per-polarization click probabilities \(q_Z\) or \(q_S\), depending on whether the architecture is dual-Sagnac ZALM or single-Sagnac SPE [2603.06423].

## 3. Scaling laws, rate formulas, and efficiency thresholds

In the weak-squeezing regime, the essential attraction of cross-island heralding is its scaling. Using
\[
p_1 \approx 4\eta_T^2(G-1)^2 \sim 4\eta_T^2 \sinh^4 r \simeq 4\eta_T^2 r^4,
\]
the same-island probability becomes
\[
P_{\text{same}} \simeq 2\eta_T^2 N_I r^4,
\]
whereas the cross-island probability becomes
\[
P_{\text{cross}} \simeq 2\eta_T^2 N_I^2 r^4.
\]
Thus, same-island heralding scales as \(N_I\), while including cross-island heralds restores \(N_I^2\) scaling in the \(r\ll 1\) regime [2507.14427].

The per-pump-pulse entanglement-distribution rates in the comparative treatment are
\[
R_{ZALM}=R_P\,\mathbb{E}[H(N_M)]\,F_{ZALM}\,\Pr_{ZALM}(\mathrm{Bell}),
\]
\[
R_{SPE}=R_P\,\mathbb{E}[H(N_M)]\,F_{SPE}\,\Pr_{SPE}(\mathrm{Bell}),
\]
and, for unheralded islands-based operation,
\[
R_{unh}=N_M R_P\,F_{unh}\,\Pr_{unh}(\mathrm{Bell}).
\]
Here \(\mathbb{E}[H(N_M)]\) is the expected number of usable heralds capped by the number of memories \(N_M\). In the single-memory limit, \(\mathbb{E}[H(1)]=1-p_0\), giving
\[
R_{ZALM}^{(1)} = R_P(1-p_0)F_{ZALM}\Pr_{ZALM}(\mathrm{Bell}),
\quad
R_{SPE}^{(1)} = R_P(1-p_0)F_{SPE}\Pr_{SPE}(\mathrm{Bell}).
\]
The distribution of total heralds is obtained from independent binomials \(z_H,z_V\sim \mathrm{Binomial}(N_I,q)\), with
\[
p_k=\Pr\{\min(z_H,z_V)=k\}.
\]
All heralds, same or cross, carry the same conditional fidelity \(F\) and Bell-state probability \(\Pr(\mathrm{Bell})\) [2603.06423].

The resulting comparison is not monotone in favor of heralding. For \(90\%\) or lower heralding efficiencies, ZALM’s per-pump-pulse entanglement-distribution rate exceeds that of the signal-path erasure source, but both are inferior to unheralded operation when all three systems employ \(N_I\) spectral islands and allocate \(N_M=N_I\) quantum memories to each pump pulse. In the large-\(N_I\), large-\(N_M\) limit, the comparison hinges on \(F\Pr(\mathrm{Bell})\). For \(\bar n=G-1\approx 0.02\), \(\eta_h=0.9\), and \(\eta_R=0.01\), the reported numerical products are
\[
F_{unh}\Pr_{unh}(\mathrm{Bell}) \approx 2.3\times 10^{-4},
\]
\[
F_{ZALM}\Pr_{ZALM}(\mathrm{Bell}) \approx 1.3\times 10^{-4},
\]
\[
F_{SPE}\Pr_{SPE}(\mathrm{Bell}) \approx 1.1\times 10^{-4}.
\]
Under these conditions, unheralded operation wins by roughly a factor of two over ZALM, which in turn beats SPE by \(\sim 20\%\). By contrast, in the single-memory regime with \(N_I\sim 50\), \(R_P=10^9/\mathrm{s}\), and the same fidelity constraints, \(R_{ZALM}^{(1)}\approx 2.5\times 10^4/\mathrm{s}\), while unheralded one-island operation yields \(\sim 10^3/\mathrm{s}\) [2603.06423].

A central practical threshold is the heralding efficiency. Because \(\Pr(\mathcal{H}_{nm})\propto \eta_h^2\), the crossover is reported around \(\eta_h\approx 90\%\). With \(N_I=N_M=20\), \(R_P=10^9/\mathrm{s}\), and \(\bar n\) chosen for \(F=0.99\), the ratios are
\[
\eta_h=95\%:\ R_{ZALM}/R_{unh}\approx 2.1,
\]
\[
\eta_h=90\%:\ R_{ZALM}/R_{unh}\approx 1.0,
\]
\[
\eta_h=80\%:\ R_{ZALM}/R_{unh}\approx 0.4.
\]
This is the reported break-even point [2603.06423].

The modified ZALM proposal also gives explicit quasi-deterministic examples. For \(\eta_T=0.9\) and \(\eta_R=0.01\), choosing \(G-1=0.0173\) yields \(\mathcal{F}=0.99\) and \(\mathcal{B}=0.99986\); with SPCI heralding, \(N_I=28\) islands suffice to get \(P_{\text{herald}}>0.25\), and at \(R_P=10^{10}\,\mathrm{s}^{-1}\), \(R\approx 2.5\times 10^7\,\mathrm{s}^{-1}\). For \(\eta_T=0.8\) and \(\eta_R=0.01\), choosing \(G-1=8.59\times 10^{-3}\) yields \(\mathcal{F}=0.99\) and \(\mathcal{B}=0.99986\); \(N_I=62\) islands then give \(P_{\text{herald}}>0.25\) and \(R\approx 2.5\times 10^5\,\mathrm{s}^{-1}\) [2507.14427].

## 4. Optically heralded entanglement of superconducting systems

In superconducting quantum networking, the closely related protocol replaces cascaded direct transduction by optical networking via heralding end-to-end entanglement with one detected photon and teleportation. The underlying electro-optic transducer couples an optical mode \(\hat a\) and a microwave mode \(\hat b\) through a strongly pumped \(\chi^{(2)}\) interaction. With pump mode \(\hat p\), the interaction-picture Hamiltonian is
\[
\hat H_{\text{total}}=\hbar g_0(\hat p^\dagger \hat a^\dagger \hat b + \hat p \hat a \hat b^\dagger)+\hat H_{\text{free}},
\]
and under a large coherent pump \(\langle n_p\rangle\equiv |\langle \hat p\rangle|^2\gg 1\), one defines \(g\equiv g_0\sqrt{\langle n_p\rangle}\). Red detuning gives a beam-splitter interaction,
\[
\hat H_{bs}=\hbar g(\hat a^\dagger \hat b+\hat b^\dagger \hat a),
\]
whereas blue detuning gives two-mode squeezing,
\[
\hat H_{sq}=\hbar g(\hat a\hat b+\hat a^\dagger \hat b^\dagger).
\]
The optical mode decays to the output waveguide at rate \(\gamma_e\), has intrinsic loss \(\gamma_i\), and total linewidth \(\gamma=\gamma_e+\gamma_i\). The microwave mode is taken to have \(\Gamma_{mw}\approx 0\) on the entanglement timescale. Photon detection is described by collapse operator \(\hat c=\sqrt{\gamma_e}\hat a\) and effective no-click Hamiltonian
\[
\hat H_{\text{eff}}=\hat H_{sq}-\frac{i\hbar}{2}\gamma_e \hat a^\dagger \hat a
\]
[2012.13408].

In the weak-coupling limit \(g\ll \gamma\), continuous blue-detuned pumping in pulses of duration \(\Delta t\) yields a single-node pair-generation rate
\[
r_0=\frac{4g^2\gamma_e}{(\gamma_e+\gamma_i)^2}.
\]
Defining internal transducer efficiency
\[
\eta_{om}\equiv \frac{\gamma_e}{\gamma_e+\gamma_i},
\]
channel efficiency \(\eta_{ch}\), and detector efficiency \(\eta_{det}\), the one-way heralding success probability per attempt is
\[
P_{\text{herald}}\approx r_0\Delta t\cdot \eta_{om}\cdot \eta_{ch}\cdot \eta_{det},
\]
and the corresponding continuous-time rate is
\[
R_{\text{herald}}=r_0\cdot \eta_{om}\cdot \eta_{ch}\cdot \eta_{det}.
\]
For two-node path erasure at a beamsplitter, the entanglement-generation rate becomes
\[
R_e = 2R_{\text{herald}}\,e^{-R_{\text{herald}}\Delta t}\,\frac{\Delta t}{\Delta t+t_r},
\]
where \(t_r\) is the microwave-resonator reset time [2012.13408].

The proposal’s main claim is that heralding breaks the usual rate-fidelity trade-off. In deterministic red-detuned transduction, increasing \(g/\gamma\) boosts rate but adds noise and degrades fidelity. In the heralded SPDC approach, infidelity \(\propto (g/\gamma)\) arises from double excitation or higher-order SPDC and is suppressed for \(g\ll \gamma\), while conditioning on exactly one click purifies out vacuum and multi-photon components, making fidelity effectively independent of channel loss. The resulting Bell-pair fidelity is summarized as
\[
F \approx 1-\alpha(g/\gamma)-\beta(R_eT_1),
\]
with the first term attributed to double-emission error and the second to memory decoherence [2012.13408].

Once a microwave Bell pair
\[
|\Phi^+\rangle_{AB}=\frac{|0\rangle_A|1\rangle_B+|1\rangle_A|0\rangle_B}{\sqrt 2}
\]
has been established, an arbitrary qubit \(|\psi\rangle_C=\alpha|0\rangle_C+\beta|1\rangle_C\) can be teleported by a Bell-basis measurement on \(C\) and \(A\), transmission of two classical bits, and conditional recovery at \(B\):
\[
|\Phi^+\rangle \rightarrow I,\quad
|\Phi^-\rangle \rightarrow Z,\quad
|\Psi^+\rangle \rightarrow X,\quad
|\Psi^-\rangle \rightarrow XZ.
\]
In density-matrix form,
\[
\rho_C\otimes |\Phi^+\rangle_{AB}\langle \Phi^+| \rightarrow \rho_B.
\]
Using \(g_0\approx 1\,\mathrm{kHz}\), \(\gamma_e=\gamma_i\approx 100\,\mathrm{MHz}\), \(T_{1,mw}\approx 1\,\mathrm{ms}\), \(t_r\approx 1\,\mu\mathrm{s}\), and \(P\approx 1\,\mu\mathrm{W}\), the reported estimates are \(R_e\approx 10^5\,\mathrm{s}^{-1}\), \(F_0\ge 0.99\), and, after one stage of superconducting purification, \(F\approx 0.999\) with \(R_e\rightarrow \sim 50\,\mathrm{kHz}\). The protocol is described as one in which losses only reduce the rate, not the fidelity [2012.13408].

## 5. Island-model genetic programming

In symbolic regression on small, wide datasets, “cross-island heralding” names a migration mechanism rather than a photonic detector event. Evolutional Math uses \(N=4\) GP islands running in parallel, each seeded with a different operator family: algebraic, log-exp, trigonometric, and generalist/full. Each island’s initial population has size \(|P|\), typically \(300\), and is generated at random subject to the island’s operator constraint [2606.28381].

Migration occurs every \(M=25\) generations on a directed ring. At each migration step, island \(i\) sorts its population \(P_i\) by the same fitness used throughout GP, namely \(k\)-fold cross-validated \(R^2\) with a complexity penalty,
\[
\mathcal{L}_{cv}(f;X,y)= -\frac{1}{k}\sum_{j=1}^k \mathrm{clip}_{[-1,1]}\!\left(R^2(f,X_{val}^{(j)},y_{val}^{(j)})\right)+\lambda\cdot cx(f),
\]
with \(k=3\) and \(\lambda=0.005\). The top \(K=3\) individuals are chosen as migrants,
\[
M_i \leftarrow \mathrm{TopK}(P_i,f,K),
\]
and sent to island \((i+1)\bmod N\), where they replace the bottom \(K\) individuals [2606.28381].

The migration mechanism is embedded in a broader diversity-preserving system. Structural deduplication defines a canonical signature \(\sigma(f)\) by replacing every constant node in the prefix serialization with the placeholder \(C\), for example
\[
\sigma(x_0 + 1.985\cdot \sin(x_1)) = \text{“(+ x0 (* C (sin x1)))”}.
\]
The elite archive admits at most one formula per signature, and a new candidate displaces the archived one only if its fitness is strictly higher. Separately, a global seen-set keyed by the full prefix string avoids reevaluating identical trees [2606.28381].

Constant refinement occurs every \(K'=25\) generations, in practice the same interval as migration. The top \(T=10\) individuals within each island undergo L-BFGS-B optimization of their numeric constants, up to \(50\) iterations, using
\[
\Phi(\mathbf{c}) = \frac{1}{n}\sum_{i=1}^n \bigl(y_i-f_{\mathbf c}(x_i)\bigr)^2,
\qquad
\mathbf{c}^*=\arg\min_{\mathbf c}\Phi(\mathbf c).
\]
The refined individual replaces the original only if its cross-validated \(R^2\) fitness is strictly improved. The paper states that the interplay of seeding, heralding, deduplication, and refinement prevents collapse to a single motif and sustains diversity throughout the run [2606.28381].

## 6. Interference ridges between physical islands

In the literature on Marshallese navigation, Harvey proposes a speculative explanation for the dilep in terms of simple wave interference. Two small islands \(A\) and \(B\), separated by distance \(D\), are idealized as point scatterers reflecting a single monochromatic ocean wave of amplitude \(A\), angular frequency \(\omega\), and wavelength \(\lambda\). With the primary swell removed from the model, the observation-point amplitude is
\[
\Psi(x,y,t)=A e^{i[kr_1-\omega t]} + A e^{i[kr_2-\omega t]},
\]
where
\[
r_1=|(x,y)-(x_A,y_A)|,\qquad
r_2=|(x,y)-(x_B,y_B)|,\qquad
k=2\pi/\lambda.
\]
The physical elevation is \(\eta(x,y,t)=\mathrm{Re}[\Psi(x,y,t)]\) [1802.09151].

The intensity is
\[
|\Psi|^2 = 2A^2[1+\cos(k(r_1-r_2))].
\]
Constructive interference occurs when \(\Delta \phi = k(r_1-r_2)=2\pi m\), so the maxima satisfy
\[
r_1-r_2 = m\lambda.
\]
These loci are a family of hyperbolae with \(A\) and \(B\) as foci. The \(m=0\) branch gives \(r_1=r_2\), the straight-line “backbone” between the islands, while \(m=\pm 1,\pm 2,\ldots\) form parallel curves on either side. Harvey also notes that when \(AB\) happens to be an integer number of wavelengths, the same condition can be restated in terms of sums and identified with very elongated ellipses, but the key point is the difference-of-distance rule [1802.09151].

The model yields quantitative predictions. Writing \(N=D/\lambda\), the transverse spacing between adjacent central lobes is estimated as
\[
\Delta y \simeq \sqrt{N}\,\lambda \quad \text{or} \quad \Delta y \simeq \sqrt{N/2}\,\lambda,
\]
typically in the \(0.7\)–\(1.0\sqrt N\) range. For \(N=1000\), corresponding to \(D=100\,\mathrm{km}\) and \(\lambda=100\,\mathrm{m}\), this gives \(2\)–\(3\,\mathrm{km}\) between central dileps. The “shoulder” of each standing-wave ridge extends roughly half the spacing, so central widths are of order \(1\,\mathrm{km}\) [1802.09151].

The paper emphasizes that there is no agreed causal explanation for the dilep, and the proposed mechanism is explicitly speculative. It also lists empirical tests: SAR or sun-glint imagery looking for near-parallel bright lines spaced \(\sim 2\,\mathrm{km}\) apart; shallow-water or Boussinesq simulations with realistic bathymetry; small-boat trials using GPS and pitch-and-heave accelerometers; and ethnographic interviews about “booj” spacing in time and distance [1802.09151].

## 7. Comparative interpretation and recurrent design logic

Across these usages, cross-island heralding consistently denotes a condition in which success is established not within a single island but through relations among distinct islands. In spectral SPDC, that relation is an ordered pair of frequency bins producing one \(H\) and one \(V\) click; in superconducting networking, it is a remote entanglement event announced by one optical click; in GP, it is an elite structure crossing a ring-topology boundary; in Harvey’s navigation model, it is an interference ridge extending between island pairs [2507.14427; 2012.13408; 2606.28381; 1802.09151].

Several misconceptions are explicitly corrected by the cited work. First, cross-island heralding is not universally better than unheralded operation: when all systems employ \(N_I\) spectral islands and \(N_M=N_I\) memories, unheralded operation can outperform heralded schemes, especially at \(90\%\) or lower heralding efficiencies [2603.06423]. Second, heralding does not imply deterministic transfer. In the superconducting proposal, the protocol is probabilistic, and the point is that inefficiency is absorbed into a heralding overhead so that fidelity can remain high [2012.13408]. Third, in the Marshallese case the model predicts not a single privileged line but a small family of near-parallel paths [1802.09151]. Fourth, in evolutionary computation, “heralding” is not detection but coordinator-driven migration, with explicit parameters \(N=4\), \(M=25\), and \(K=3\) [2606.28381].

A plausible unifying implication is that the phrase marks a design strategy in which cross-island structure is exploited to improve search coverage, rate scaling, or signal identifiability. In the quantum-optical setting, that strategy can turn \(O(N_I)\) scaling into \(O(N_I^2)\) scaling in the weak-squeezing limit [2507.14427]. In superconducting networking, it can decouple fidelity from channel loss by conditioning on exactly one click [2012.13408]. In GP, it can prevent collapse into one region of formula space [2606.28381]. In Harvey’s model, it identifies the only robust, scale-invariant signals bridging distant islands as standing-wave ridges created by constructive interference [1802.09151].

Source: https://www.emergentmind.com/topics/cross-island-heralding