Pairwise Fusion Gates in Quantum Optics

• Pairwise Fusion Gates are probabilistic optical operations that fuse smaller photonic entangled resource states into larger graph or cluster states using heralded measurements.
• Recent advancements have introduced boosted fusion schemes that exceed the traditional 50% success limit by employing auxiliary states, achieving up to 75% efficiency and surpassing critical percolation thresholds.
• Extensions into high-dimensional qudit fusion and applications in topological quantum computing and state construction demonstrate the versatility and scalability of pairwise fusion techniques across diverse quantum platforms.

Pairwise fusion gates are operations that act on two subsystems—most often two photonic qubits or qudits extracted from larger entangled resource states—and herald a projection that either joins the parent resources into a larger entangled object or isolates a Bell-type link in a prescribed subspace. In linear-optical quantum computing, they are a core primitive because large graph or cluster states are difficult to generate directly but can be assembled from smaller resource states by repeated probabilistic fusions. Recent work has broadened the concept from standard qubit type-I and type-II fusion to boosted Bell measurements, high-dimensional qudit fusion, and recovery of structured failure sectors, while other literatures use closely related terminology for anyonic fusion primitives and for decision-level fusion outside quantum information (Guo et al., 2024, Yamazaki et al., 2024).

1. Operational definition and basic variants

In the photonic setting, a fusion gate is a probabilistic optical operation applied to two input resource states, such as two Bell pairs or two small graph states. Selected photons are interfered on beam splitters and the outputs are measured; successful detection patterns herald that the two smaller states have been fused into a larger entangled state. This operational picture underlies graph-state growth, cluster-state construction, and entanglement-swapping-like protocols in linear optics (Guo et al., 2024).

A standard distinction is between type-I and type-II fusion. For path-encoded qubits in modes $\hat a_{0,1}$ and $\hat b_{0,1}$, a type-I fusion gate mixes $\hat a_1$ and $\hat b_1$ on a beam splitter, and a single detection event heralds success, leaving a new photonic qubit in modes $\hat a_0$ and $\hat b_0$ while entangling the larger registers $A$ and $B$. A type-II fusion gate mixes both $\hat a_1,\hat b_1$ and $\hat a_0,\hat b_0$; two clicks in different beam splitters herald success and entangle the larger registers $\hat b_{0,1}$0 and $\hat b_{0,1}$1. In dual-rail notation, type-I implements the entangling map

$\hat b_{0,1}$2

and, without ancillary resources, both type-I and type-II are limited to success probability $\hat b_{0,1}$3 (Aqua et al., 2024, Melkozerov et al., 28 Mar 2026).

This $\hat b_{0,1}$4 ceiling is not merely a technical nuisance. In photonic architectures built from repeated pairwise fusions, the asymptotic behavior of the entire construction depends on whether local fusion succeeds often enough to compensate for resource consumption and fragmentation. That dependence is most explicit in percolation-based analyses of graph-state growth.

2. Scalable graph-state generation and the percolation threshold

The clearest experimentally established role of pairwise fusion gates is in scalable photonic graph-state generation. The relevant setting is the repeated fusion of small entangled resource states into a larger, fully connected graph state. Percolation theory provides the threshold criterion: for three-photon GHZ states used as the resource state, scalable graph-state generation requires the fusion success probability to exceed $\hat b_{0,1}$5. The same work also mentions a related numerical threshold of $\hat b_{0,1}$6 in simulations of 2D cluster-state growth without photon loss, underscoring that the precise threshold depends on the lattice and resource assumptions (Guo et al., 2024).

A conventional Bell-state measurement can reliably identify only some Bell states, so standard fusion gates are limited to at most $\hat b_{0,1}$7 success. The boosted type-II fusion gate demonstrated in 2024 overcomes that limit by using deterministically generated auxiliary states to improve Bell-state discrimination. In the reported implementation, photons 2 and 3 are interfered on a beam splitter, and auxiliary two-photon N00N-like states are introduced so that $\hat b_{0,1}$8 are distinguished at the first beam splitter while $\hat b_{0,1}$9 are resolved through additional interference on BS2 or BS3. The resulting theoretical success probability is $\hat a_1$0, and the experimentally measured value is $\hat a_1$1, which exceeds the $\hat a_1$2 percolation threshold by $\hat a_1$3 standard deviations. The same experiment fused two Bell states with output fidelity

$\hat a_1$4

exceeding the classical threshold of $\hat a_1$5 by more than $\hat a_1$6 standard deviations and providing a direct verification that the gate functions as an entangling fusion primitive rather than only as a statistical Bell discriminator (Guo et al., 2024).

This result fixes an important misconception in the literature on probabilistic photonic entangling gates. A success probability above the $\hat a_1$7 Bell-measurement limit is not by itself the relevant scalability statement; the experimentally meaningful benchmark is whether the gate crosses the percolation threshold associated with the chosen resource-state architecture. In the GHZ-based setting, that benchmark is $\hat a_1$8, not an abstract improvement over $\hat a_1$9.

3. Indistinguishability, exchange symmetry, and boosted qubit fusion

A second development concerns the physical requirement usually summarized as “photon indistinguishability.” The temporal quantum eraser work argues that ideal two-photon fusion-gate operation depends not on literal identity of the photons but on the exchange symmetry of the two-photon wavefunction. The temporal wavefunction is decomposed as

$\hat b_1$0

with $\hat b_1$1 for symmetric and $\hat b_1$2 for antisymmetric states. In this picture, HOM interference is controlled by symmetry: symmetric states bunch and antisymmetric states antibunch. A temporal quantum eraser projects the two-photon state into a purely symmetric or purely antisymmetric exchange sector, thereby recovering ideal HOM-type interference and the correct interpretation of type-I and type-II fusion outcomes even when the photons were initially distinguishable. With the temporal quantum eraser, the fusion-gate success probability remains the usual $\hat b_1$3, but the infidelity caused by distinguishability is removed because the symmetry sector is heralded (Aqua et al., 2024).

This symmetry-based reformulation also clarifies failure mechanisms. In the nonlinear pair-source platform, double-pair emission can produce a symmetry-defined heralding event while still introducing number impurity, because both signal photons may occupy the same waveguide. In the $\hat b_1$4-emitter platform, the heralding relies on dark-port parity, so photon loss and detector inefficiency directly spoil the symmetry classification. The paper therefore separates two issues that are often conflated: modal impurity and exchange symmetry on the one hand, and number impurity or loss sensitivity on the other (Aqua et al., 2024).

Boosting has also been developed for type-I fusion. A 2026 proposal achieves total success probability $\hat b_1$5 using only four ancillary single photons, passive linear optics, and photon-number-resolving detectors. The direct success probability is $\hat b_1$6: the standard odd-parity branch contributes $\hat b_1$7, and a useful four-photon branch contributes an additional $\hat b_1$8. A two-photon partially entangled branch occurs with probability $\hat b_1$9 and can be distilled into further successful events; in the asymptotic limit this adds another $\hat a_0$0, giving the total

$\hat a_0$1

With one distillation stage and balanced beam splitters, the total becomes $\hat a_0$2. The stated novelty is that earlier $\hat a_0$3-efficient type-I schemes relied on ancillary Bell pairs, whereas this scheme uses only ancillary single photons (Melkozerov et al., 28 Mar 2026).

4. High-dimensional pairwise fusion gates

High-dimensional generalizations recast pairwise fusion as a measurement on two $\hat a_0$4-rail single-photon qudits. In this setting, each logical qudit is a single photon delocalized over $\hat a_0$5 orthogonal modes, and the gate does not attempt a full $\hat a_0$6-label Bell discrimination. Instead, it heralds Bell projections onto definite two-rail subspaces. The relevant pairwise Bell states are

$\hat a_0$7

for $\hat a_0$8. In the passive, ancilla-free setting, success occurs with probability

$\hat a_0$9

because the off-diagonal sector is fully convertible into successful pairwise Bell projections and all failures are confined to the diagonal logical subspace. Ancilla-assisted circuits raise the success probability to

$\hat b_0$0

using $\hat b_0$1 ancilla photons in products of high-dimensional GHZ states. The same framework is proposed as a fast quantum repeater primitive with three-qudit GHZ states and quantum memories, and the paper notes that entanglement swapping must be above roughly $\hat b_0$2 for linear scaling in repeater time (Yamazaki et al., 2024).

A notable conceptual point is that the passive failure sector is structured rather than random. For the diagonal input $\hat b_0$3, the beam splitter produces a bunched two-mode output,

$\hat b_0$4

which reveals the occupied rail $\hat b_0$5. The failure therefore acts like a same-rail projection, not an erasure channel (Laha et al., 28 Jun 2026).

That structure enables active recovery by output squeezing. In the squeezing-enhanced scheme, identical single-mode squeezers are applied to all $\hat b_0$6 interferometer outputs before photon-number-resolving detection. Odd-parity signatures preserve the successful off-diagonal $\hat b_0$7 outcomes, while selected all-even count patterns yield POVM elements proportional to definite diagonal Bell projectors. The paper proves that an all-even pattern is accepted if and only if its photon-number-imbalance vector has exactly two nonzero components of equal magnitude. Numerically, the ideal success probability rises from $\hat b_0$8 to $\hat b_0$9 for $A$0, and from $A$1 to $A$2 for $A$3. With detector saturation threshold $A$4, the certified values remain $A$5 and $A$6, respectively (Laha et al., 28 Jun 2026).

5. Pairwise fusion as a state-construction primitive

In linear optical cluster-state generation, pairwise fusion is not always best understood as a universal two-qubit gate. The analysis of linear optical cluster states argues that the standard stochastic CZ gate, with success probability $A$7, is suboptimal because cluster growth only requires the correct transformation on a specific input subspace rather than on the full two-qubit computational space. The optimal operation is therefore a hybrid state transformation on a subspace. For sequential linear-cluster growth, the maximal success probability for adding one qubit is $A$8, giving total success $A$9 for a length-$B$0 cluster built by sequential single-qubit addition. For Bell-pair addition, the maximal success probability is $B$1, and the corresponding Bell-pair growth scaling is $B$2. The paper further reports numerical success rates $B$3 for $B$4, about $B$5 for $B$6, and about $B$7 for grafting a qubit in the middle of a linear cluster (Uskov et al., 2014).

A different but related fusion literature concerns multipartite W states. In the Fredkin-enhanced W-state fusion scheme, one photon from $B$8 and one photon from $B$9 are sent through a setup that inserts a single Fredkin gate and an ancillary $\hat a_1,\hat b_1$0-polarized photon before the original polarization-based fusion device. The key modification is that the former destructive $\hat a_1,\hat b_1$1 case is converted into a success by the controlled swap

$\hat a_1,\hat b_1$2

The resulting success probability becomes

$\hat a_1,\hat b_1$3

improving on the original

$\hat a_1,\hat b_1$4

and the successful output is $\hat a_1,\hat b_1$5 rather than $\hat a_1,\hat b_1$6. The same setup can fuse Bell states, $\hat a_1,\hat b_1$7, into $\hat a_1,\hat b_1$8 with success probability $\hat a_1,\hat b_1$9 (Bugu et al., 2013).

These constructions illustrate a general pattern: “pairwise fusion gate” need not mean a universal entangling operator. In several architectures it instead denotes a task-specific, heralded primitive tailored to a fixed resource-state family and a fixed growth rule.

6. Anyonic, trapped-ion, and non-quantum uses of the term

Outside linear optics, the same language appears in topological quantum computation, where fusion itself becomes a computational primitive. In the $\hat a_0,\hat b_0$0 quantum double, braiding alone is not universal, so the computational model is augmented by fusion and charge measurement. Pairwise fusion of $\hat a_0,\hat b_0$1 flux pairs and $\hat a_0,\hat b_0$2-charge probe pairs implements logical projection and comparison operations: an ancilla $\hat a_0,\hat b_0$3 pair braided around a logical $\hat a_0,\hat b_0$4 pair and then fused realizes an $\hat a_0,\hat b_0$5-basis measurement, while a $\hat a_0,\hat b_0$6-charge pair braided around a data qutrit and a reference qutrit and then fused realizes a $\hat a_0,\hat b_0$7-basis comparison. Together with the pull-through braid

$\hat a_0,\hat b_0$8

these fusion-based measurements form a universal topological gate set on logical qutrits (Lo et al., 28 Jan 2026).

A related fusion-and-measurement construction appears for SU(2)$\hat a_0,\hat b_0$9 or JK$\hat b_{0,1}$00 anyons. There, braiding alone is again non-universal, but braiding plus fusion plus topological charge measurement becomes universal. Pairwise fusion with charge-4 ancillas implements a logical NOT on the 1221 encoding, the maps $\hat b_{0,1}$01 and $\hat b_{0,1}$02 convert the 1111 encoding into the 1221 encoding, and topological qubit fusion maps $\hat b_{0,1}$03 and $\hat b_{0,1}$04 compress a two-qubit state into a one-qubit state. Those primitives are then used to generate an exact irrational phase gate and an approximate controlled-$\hat b_{0,1}$05 gate (Levaillant et al., 2015).

The phrase also appears in contexts that are only terminologically related. In trapped-ion hardware, “pairwise-parallel” entangling gates refer to simultaneous Mølmer–Sørensen gates executed on orthogonal motional mode sets, yielding a three-qubit GHZ state with fidelity $\hat b_{0,1}$06 and effectively extending available gate depth by up to two times; this is an entangling-gate scheduling method rather than a fusion measurement (Zhu et al., 2023). In continual learning, “Glocal Pairwise Fusion” denotes an online decision-level fusion mechanism in which a 3D matrix $\hat b_{0,1}$07 accumulates counts over true labels, global predictions, and local predictions, and inference uses

$\hat b_{0,1}$08

to choose the final class label (Loo et al., 2023).

Taken together, these usages show that pairwise fusion gates are best understood as a family of pairwise, heralded combination rules whose exact meaning depends on the host theory. In photonic quantum information they are primarily probabilistic entangling measurements for resource-state growth; in topological models they are fusion-and-measurement primitives that supplement non-universal braiding; and in other fields the term can denote a structurally analogous pairwise combination mechanism without any quantum-measurement content.