Entanglement swapping

Two Bell pairs sharing one node can be swapped into a single Bell pair spanning the two outer endpoints — without those endpoints ever interacting and without measuring the qubits being networked. The trick is a Bell-state measurement at the shared node: the BSM consumes the two qubits at the BSA, and what's left is a Bell pair between the two distant qubits, up to a Pauli correction that's announced via a classical channel ^{Żukowski et al. 1993}.

Swapping is the primitive that extends the reach of a quantum network. A direct Bell pair between Alice and Bob in standard fibre runs out of usable rate after a few tens of kilometres. At 0.20 dB/km, photon survival halves every 15 km — so only ~1 % make it through 100 km. Current heralded entanglement demonstrations sit in the 10–50 km range over deployed fibre ^{Knaut et al. 2024}. Chains of swaps stitch shorter pairs together so the end-to-end pair can span much further. Every memory-based quantum repeater — and every link architecture in distribution — is built on this primitive ^{Kumar et al. 2025}.

The setup and protocol

Three parties: Alice, a Bell-state analyser (BSA), and Bob. Two Bell pairs already exist: A at Alice paired with A′ at R, and B′ at the BSA paired with B at Bob. The BSA holds A′ and B′, one qubit from each pre-shared pair; the qubits at Alice and Bob are never touched until the final Pauli correction.

Three steps follow, structurally identical to teleportation — except that what's being "teleported" is Alice's half of the first pair onto Bob's qubit, with the BSA performing the BSM in the middle.

BSM at the BSA. The BSA performs a Bell-state measurement on its two qubits $R_{1}$ and $R_{2}$ . Both pre-shared pairs ( $A \leftrightarrow R_{1}$ and $R_{2} \leftrightarrow B$ ) are consumed by the measurement. The outcome is two classical bits identifying which of the four Bell states the BSM landed on, which in turn determines which Bell state $A$ and $B$ are now in.
Classical communication. The BSA sends the 2 bits to Bob (or Alice — either endpoint will do).
Pauli correction. Bob applies one of $I$ , $X$ , $Z$ , $Z X$ to his qubit $B$ based on the bits. After the correction, $A$ and $B$ are in the Bell state $∣ Φ^{+} ⟩_{A B}$ — a freshly minted Bell pair across the full distance.

1. Two pre-shared Bell pairs: A↔A′ at Alice and B′↔B at Bob, meeting at the BSA
2. R performs a Bell-state measurement on A′ and B′
3. The 2-bit BSM outcome travels to Bob over a classical channel
4. Bob applies the matching Pauli correction (I, X, Z, or XZ)
5. A and B share a fresh Bell pair — they never interacted

Top row: two short Bell pairs (magenta wavy lines) sit on either side of the BSA. Its BSM consumes A′ and B′ — their clouds turn grey the moment they enter the apparatus. Two classical bits emerge on the channel and travel to Bob, who reads them and applies the matching Pauli. Bottom row: the resulting long A↔B Bell pair traces in as the correction lands. The cycle iterates through the four Bell outcomes.

Chained swapping for reach

One swap doubles the reach of an unaided link. A chain of $n$ swaps links $n + 1$ short pairs into a single Bell pair spanning the full distance. Each BSA in the chain performs its own BSM and broadcasts its 2-bit result; the corrections compose at one of the endpoints. Memory-based repeaters use this pattern as their core operation, with quantum memories at each node holding the qubits while pairs are generated and swaps are coordinated ^{Jones et al. 2016}.

The canonical 1G repeater

The four primitives so far — heralded entanglement generation (EG), Bell-state measurement, swapping, and quantum memory — compose into one workflow. This is the first-generation repeater design proposed by Briegel, Dür, Cirac and Zoller ^{Briegel et al. 1998} and surveyed in ^{Sangouard et al. 2011}, ^{Azuma et al. 2023}. The diagram below traces a 4-hop chain from Alice through three quantum repeaters (R₁, R₂, R₃) to Bob from start to finish: green dots are live matter qubits in memory, blue dots are flying photons used only inside the EG round, grey dots are matter qubits already spent by a Bell measurement, and blue wavy lines are entanglement.

1. EG. Adjacent matter qubits emit photons along the quantum channel; the midpoint BSA performs the Bell-state measurement and heralds a Bell pair across the link.
2. 1st ES. R₁ and R₃ run their swaps in parallel; outcomes go forward to R₂ and B.
3. 2nd ES. R₂ swaps and ships its outcome to B for the final Pauli correction.
4. Result. One long A↔B Bell pair — ready for teleportation or QKD.

Classical bits flow forward, one hop at a time. All pulses travel at the same horizontal speed, so the long R₂→B leg takes visibly longer than the short ones — the latency the memories wait through. Each row is a repeat-until-success loop (currently shown deterministic; loss retries can be toggled on).

Pauli frame tracking — one correction at the end, not one per swap

Each BSM along the chain produces two classical bits that determine which of the four Bell states the swap landed on. The naive thing to do is apply the corresponding Pauli correction (one of ${I, X, Z, X Z}$ ) physically at each step. The better thing is to not apply any correction during the chain and instead XOR each new bit pair into a 2-bit Pauli frame held classically at one endpoint — metadata that travels alongside the Bell pair but never touches the qubits.

For a 4-hop chain that means four heralding-BSM outcomes plus six swap-BSM bits (two per BSA across three BSAs) all accumulate into a single 2-bit frame at B. When the pair is finally consumed — at teleportation, QKD, or a remote gate — the frame is XOR'd with the consuming measurement's outcome bits and one Pauli correction is applied. One physical gate at the end, not seven along the chain. Each avoided physical correction is one less infidelity contribution.

Why fidelity drops with chain length

Each Bell pair starts at some fidelity below 1 — heralded entanglement is never perfect. Each BSM at a BSA is also imperfect, and each Pauli correction sees noise from the classical-message latency relative to the memory's coherence time. The end-to-end Bell pair's fidelity is roughly the product of the per-link fidelities, multiplied by the BSM and memory contributions, so chains of swaps degrade fidelity multiplicatively. This is why purification matters at scale — it counters the multiplicative loss by spending throughput on cleaner pairs.

The first multi-node demo with this pattern is Pompili et al. 2021: a three-node NV-centre network — Alice, Bob, and Charlie — all sitting in the same lab at QuTech (Delft), with a maximum inter-node distance of 7 metres. Charlie sits between Alice and Bob and acts as the swapping BSA, demonstrating on-demand entanglement distribution between the non-neighbouring endpoints. The point isn't reach — it's that the protocol stack (entanglement generation, swap, fidelity reporting, classical signalling) ran end-to-end on a real platform ^{Pompili et al. 2021}.

Successive milestones from the same era:

Hermans et al. 2022 — same Delft 3-node hardware, now running qubit teleportation through the middle node, not just entanglement distribution ^{Hermans et al. 2022}.
Krutyanskiy et al. 2023 — trapped-ion entanglement between two nodes over 230 m of deployed urban fibre in Innsbruck, the first outdoor-fibre heralded demo ^{Krutyanskiy et al. 2023}.
Knaut et al. 2024 — two SiV-centre nodes at Harvard with telecom-O QFC, generating entanglement through a 35 km loop of deployed Boston-area fibre at 17 dB total loss ^{Knaut et al. 2024}.

None of these are >3 nodes or km-scale across more than two; multi-node swapping chains over real fibre are still the open engineering frontier.

Network role

Swapping is the primitive that turns a two-node Bell-pair distribution into a full network. The link-layer service ("deliver Bell pair (A, B) on demand") inside a single link is a generation problem; getting that pair across more than one link is a swapping problem. A repeater is, mechanically, a station that holds memories long enough to align swaps — the network engineering is all about which pairs to generate when, who fires first, and how to recover if any single link's heralding fails.

Three follow-on subjects expand on this:

Distribution covers the link architectures (MM, SR, MS) that get a Bell pair onto a single link in the first place. See the distribution subject.
Purification trades two noisy Bell pairs for one cleaner pair, restoring fidelity that chained swapping erodes. See the purification subject.
Repeaters integrate generation, swapping, and purification into the 1G / 2G / 3G architectural families. See the repeaters subject.