Entanglement Swapping

You want to share a Bell pair between a lab in Bengaluru and a lab in Mumbai. The problem is not the 980 km. The problem is the optical fibre. Every kilometre of silica fibre attenuates light by roughly 0.2 decibels — which means that after 100 km, only about 1% of the photons you sent survive; after 500 km, about one in 10^{10}; after 1000 km, essentially zero. You cannot just "turn up the laser," because the thing you are trying to send is a single photon carrying a single qubit of quantum information, and classical amplification would destroy the quantum state (the no-cloning theorem forbids copying an unknown state before the loss, so there is no quantum analogue of a classical repeater amplifier).

The clean classical solution — a repeater station every 50 km that receives the signal, amplifies it, and re-transmits — does not work. If you measure the photon at the repeater to read it, you collapse the quantum state and the qubit is gone. If you try to copy it without measuring, you violate no-cloning. So the direct approach is out.

Here is the idea that rescues the whole field. Instead of sending one qubit over 980 km, share a Bell pair over the first 450 km — Bengaluru to Pune. Share a second Bell pair over the remaining 530 km — Pune to Mumbai. Each pair is short enough that photon survival is manageable. Then do one clever operation at Pune that stitches the two pairs together, leaving Alice in Bengaluru and Bob in Mumbai holding a single fresh Bell pair. That operation is entanglement swapping.

No qubit ever travels from Bengaluru to Mumbai. Alice and Bob never communicate directly during the stitching. And yet at the end of it, their two qubits — which have no shared history, have never been in the same city, have never interacted with each other — are in one of the four Bell states |\Phi^+\rangle, |\Phi^-\rangle, |\Psi^+\rangle, |\Psi^-\rangle, each of which is maximally entangled. Two classical bits from Pune tell Alice and Bob which one. A Pauli correction on Bob's qubit nails it down to exactly |\Phi^+\rangle if that is what they wanted.

By the end of this chapter you should be able to: draw the four-qubit setup, write out the initial joint state of the two independent Bell pairs, rewrite that state in the Bell basis of the two middle qubits, read off the four measurement outcomes and their associated Alice–Bob Bell states, and explain why chaining many swaps in series is the standard design for a long-distance quantum repeater network.

The four-qubit setup

Pictures before formulas. You have four qubits living in three places.

Writing the joint state down, with qubit order (A, M_1, M_2, B):

Expanding the tensor product term by term:

where the four indices are (A, M_1, M_2, B) in that order.

Why four terms: each of the two Bell pairs contributes two terms (|00\rangle + |11\rangle), and the tensor product multiplies them out into 2 \times 2 = 4 combined terms, each with amplitude \tfrac{1}{2}. The prefactor is \tfrac{1}{\sqrt 2} \times \tfrac{1}{\sqrt 2} = \tfrac{1}{2}.

Notice two things about |\Psi_{\mathrm{init}}\rangle. First, Alice's qubit is not entangled with Bob's qubit. You can verify this by tracing out the middle qubits — Alice's reduced state and Bob's reduced state are each completely mixed, and their joint reduced state is the product of those two mixed states. There is no A–B correlation beyond what two independent coin flips would give. Second, the middle node holds two qubits that are each entangled with someone else — but q_{M_1} and q_{M_2} are not entangled with each other yet. They are the communication bridge the Pune node has access to.

The key move is to make a Bell measurement on (q_{M_1}, q_{M_2}) — the two middle qubits. That measurement projects the pair onto one of the four Bell states, consumes both qubits in the process, and — here is the magic — leaves (q_A, q_B) in a Bell state that is precisely correlated with whichever outcome came up.

The circuit — what actually happens at Pune

The Bell measurement at Pune is a standard three-step thing, identical to the one you met in the Bell-states chapter: CNOT with q_{M_1} as control and q_{M_2} as target, then Hadamard on q_{M_1}, then measure both middle qubits in the computational basis. The two classical bits that come out — call them (m_1, m_2) — identify which Bell state the middle pair collapsed to.

This is the full swap. Four wires, one CNOT, one Hadamard, two measurements. What remains on the hardware is Alice's qubit in Bengaluru and Bob's qubit in Mumbai. What travels over a classical channel to them is the two-bit outcome (m_1, m_2), which Pune tweets to both labs.

Why it works — the term-by-term derivation

The claim is: after the Bell measurement on the middle pair, the outer pair is in a Bell state determined by the outcome. The clean way to see this is to rewrite |\Psi_{\mathrm{init}}\rangle in the Bell basis of (q_{M_1}, q_{M_2}) and then just read off the structure.

The Bell-basis decomposition of the two-qubit computational basis is the same in either direction. Recall from the Bell-states chapter:

Why: these four identities come from inverting the definitions |\Phi^\pm\rangle = \tfrac{1}{\sqrt 2}(|00\rangle \pm |11\rangle) and |\Psi^\pm\rangle = \tfrac{1}{\sqrt 2}(|01\rangle \pm |10\rangle). Add and subtract to isolate each computational basis vector.

Now we apply these to the middle-pair part of |\Psi_{\mathrm{init}}\rangle. Start with the expansion

Regroup so the middle two qubits are together in each term:

Now substitute each middle-pair computational basis state with its Bell-basis form:

• |00\rangle_{M_1 M_2} = \tfrac{1}{\sqrt 2}(|\Phi^+\rangle + |\Phi^-\rangle)
• |01\rangle_{M_1 M_2} = \tfrac{1}{\sqrt 2}(|\Psi^+\rangle + |\Psi^-\rangle)
• |10\rangle_{M_1 M_2} = \tfrac{1}{\sqrt 2}(|\Psi^+\rangle - |\Psi^-\rangle)
• |11\rangle_{M_1 M_2} = \tfrac{1}{\sqrt 2}(|\Phi^+\rangle - |\Phi^-\rangle)

Substituting and pulling out a \tfrac{1}{\sqrt 2} for an overall \tfrac{1}{2\sqrt 2} prefactor:

Why the prefactor: the original \tfrac{1}{2} is the product of two \tfrac{1}{\sqrt 2} Bell-pair prefactors. Each substitution above pulls out one more \tfrac{1}{\sqrt 2}, so the global coefficient becomes \tfrac{1}{2} \cdot \tfrac{1}{\sqrt 2} = \tfrac{1}{2\sqrt 2}.

Now group by Bell state of the middle pair — collect all terms multiplying each of |\Phi^+\rangle, |\Phi^-\rangle, |\Psi^+\rangle, |\Psi^-\rangle:

Why this regrouping works: you are just collecting like terms in the Bell basis. |\Phi^+\rangle_{M_1 M_2} appears with |0\rangle_A|0\rangle_B (from the first line) and |1\rangle_A|1\rangle_B (from the fourth line), giving the combination |00\rangle_{AB} + |11\rangle_{AB}. The signs come from which of |\Phi^+\rangle or |\Phi^-\rangle appears in the substitution above. Do the same accounting for the other three.

Factor the Bell-state prefactors out of the outer-pair combinations, and recognise that each outer-pair combination is itself (up to a \tfrac{1}{\sqrt 2}) a Bell state:

Why the prefactor collapsed to \tfrac{1}{2}: each outer-pair combination |00\rangle_{AB} \pm |11\rangle_{AB} equals \sqrt 2 \cdot |\Phi^\pm\rangle_{AB}. Factoring out that \sqrt 2 turns the overall \tfrac{1}{2\sqrt 2} into \tfrac{1}{2}.

This is the key identity. Two independent |\Phi^+\rangle Bell pairs, in the computational basis they look like a four-term superposition of product states with no A–B correlation. In the Bell basis of the middle pair, they are an equal superposition over four perfectly correlated terms: whichever Bell state the middle pair is in, the outer pair is in the same Bell state.

Check the probability normalisation: there are four terms, each with squared amplitude \tfrac{1}{4}, summing to 1. Good.

The outcome table

A Bell measurement on (q_{M_1}, q_{M_2}) picks out one of the four terms with probability \tfrac{1}{4} each, and projects the outer pair onto the corresponding Bell state. Summarising:

Why the outcome bits label this way: the Bell measurement circuit (CNOT + H + computational measurement) was derived in the Bell-states chapter to output (0,0) for |\Phi^+\rangle, (1,0) for |\Phi^-\rangle, (0,1) for |\Psi^+\rangle, (1,1) for |\Psi^-\rangle. The same labelling applies here because Pune's measurement is a textbook Bell measurement.

The upshot: the outer pair is always in a Bell state after the swap. Which Bell state is random — but it is one of exactly four, and two classical bits from the middle node tell Alice and Bob which. If they wanted |\Phi^+\rangle specifically, they can apply a Pauli correction keyed by (m_1, m_2): no correction for (0,0), Z on Bob for (1,0), X on Bob for (0,1), and XZ on Bob for (1,1). After that correction, the outer pair is deterministically in |\Phi^+\rangle — Alice and Bob have a shared, freshly prepared Bell pair as if they had made it themselves.

Quantum repeaters — chaining the swap

One swap connects two hops into one. The real payoff is chaining. Put N intermediate stations between Alice and Bob, each connected to its two neighbours by a Bell pair of manageable length. Then pipeline N-1 Bell measurements — each at a different station — and you have stitched the chain end to end. Alice and Bob share a single long-distance Bell pair even though no qubit ever traversed more than one hop.

The practical constraint is probabilistic. Each link has some chance of succeeding (photons survive, memories hold), and each swap succeeds conditionally on both its input Bell pairs being ready. If you need N links all working simultaneously, your success probability scales as p^N, which crashes fast. Real repeater architectures (DLCZ-style, and the more modern "second-generation repeaters" with error correction) mitigate this using quantum memories that store successful Bell pairs until the adjacent link is ready — so you only pay a geometric waiting time, not an exponential failure rate.

The engineering details are forbidding. Building a quantum memory that holds a Bell pair for milliseconds without decoherence is hard. Converting between the flying photon qubit you need for the fibre and the stationary spin or atom qubit you need for the memory is hard. Doing a heralded Bell measurement with photons alone is only 50% efficient without auxiliary entanglement (the Lütkenhaus-Calsamiglia bound). All of these are active research areas.

Why this matters: entanglement swapping is the reason an intercontinental quantum internet is not ruled out by basic physics. Direct fibre loss kills single-photon transmission past about 500 km. Chain enough swaps and enough memories and the loss budget becomes tractable. The Indian National Quantum Mission (launched 2023, ₹6003 crore over 8 years) has "long-distance secure quantum communication" as an explicit pillar, and every proposed architecture — whether fibre-based repeater trunks between Indian metros, or entangled satellites relaying between RRI Bengaluru and Ahmedabad PRL — rests on the primitive you just derived.

Common confusions

• "The swap teleports entanglement from one pair to the other." Not quite. The swap redistributes the entanglement: the two middle qubits go from being entangled with the outer qubits to being entangled with each other, while the outer qubits go from being independent to being entangled. The total entanglement in the 4-qubit system is conserved (entropy-wise); what changes is which pairs carry it. Calling this "teleportation of entanglement" is a common loose phrasing, but the more accurate word is swapping — the existing entanglement gets rerouted, not moved bodily.

• "A qubit travels from Alice to Bob during the swap." No. No qubit traverses the end-to-end distance. What travels is classical information: the two bits (m_1, m_2) from the middle node to whoever needs them. The "quantum" part of the swap happens only at the middle node, where the two middle qubits meet and are measured jointly. Alice and Bob each hold their original qubits the whole time.

• "Entanglement swapping sends information faster than light." It does not. Before the classical bits arrive, Alice sees a completely random state on her side, and so does Bob. Their local statistics carry zero information about each other. Only when the classical bits from the middle arrive (at subluminal speeds) can they apply the correction and end up with a known, correlated Bell pair. The no-communication theorem is safe.

• "Without the Pauli correction, Alice and Bob aren't really entangled." They are really entangled. The outer pair is in a specific Bell state after the swap, determined by the middle outcome. Alice and Bob's qubits are in a maximally entangled state — their joint correlations violate Bell inequalities maximally. What the Pauli correction does is relabel which Bell state it is, to get a specific one (usually |\Phi^+\rangle) that the downstream protocol expects. The entanglement was already there; the correction just makes the labelling deterministic.

• "Swapping requires both input Bell pairs to be in |\Phi^+\rangle." No. The algebra goes through for any pair of input Bell states, and more generally for any two maximally entangled pairs. If the inputs are |\Phi^+\rangle_{A M_1} and |\Psi^-\rangle_{M_2 B}, the outer pair after a (0, 0) measurement outcome ends up in a different specific Bell state, which you can derive by the same Bell-basis expansion. The four-branch structure always holds; only the labelling shifts.

• "The middle node needs to know what the two input Bell pairs are." It doesn't. The middle node does the same Bell measurement regardless. Its only output is the classical bits (m_1, m_2). Whoever knows what the input pairs were (this is usually a classical labelling agreed upon by the protocol designer, not the nodes themselves) uses those bits plus the known input pairs to work out what Bell state the outer pair is in, and what correction to apply. The measurement itself is input-agnostic.

Going deeper

The formal protocol and the full correction group

Written as a quantum channel, entanglement swapping sends the initial four-qubit state |\Phi^+\rangle_{AM_1} \otimes |\Phi^+\rangle_{M_2 B} to the two-qubit output |\Phi^+\rangle_{AB} after tracing out the measured middle qubits and conditioning on the outcome. The correction rule can be written as a controlled Pauli:

The correction group is the single-qubit Pauli group \{I, X, Y, Z\} (ignoring phases) — the same group that labels the four Bell states via the Pauli-Bell correspondence. This is not a coincidence: Bell-state labelling, Pauli corrections in teleportation, and the syndrome structure of stabiliser error-correcting codes are all different faces of the same Pauli-group algebra. Once you notice the pattern, swapping, teleportation, and superdense coding all read as variations on one theme.

Entanglement swapping \equiv teleportation of half a Bell pair

Here is a structural observation that ties this chapter to the teleportation chapter. In quantum teleportation, Alice wants to send an unknown single-qubit state |\psi\rangle to Bob. She uses a Bell pair she shares with Bob and a Bell measurement on |\psi\rangle and her half of the pair.

Now imagine Alice's "unknown state" |\psi\rangle is actually one half of another Bell pair, with its twin sitting at a faraway node — call that node Charlie. From Alice's point of view, her qubit q_A looks like a single qubit in some state (mixed, from her local view). She teleports it to Bob using the Alice–Bob Bell pair. After teleportation, Bob holds Alice's original qubit — which means Bob is now entangled with Charlie, because that qubit was one half of a Bell pair with Charlie.

Relabel: Alice = M_1, her unknown state = q_A (one half of the original (q_A, q_{M_1}) pair — hold on, this relabelling is getting confusing, but the punchline is clean). Teleportation of one half of a Bell pair is entanglement swapping. The two protocols are the same circuit seen from different angles. This equivalence is the cleanest way to remember how swapping works: it is just teleportation applied to "half an entangled resource" rather than "a free-standing qubit."

The DLCZ protocol and quantum memories

The single most influential paper on practical quantum repeaters is the 2001 proposal by Duan, Lukin, Cirac, and Zoller (DLCZ) — Nature 414, 413 (2001), arXiv:quant-ph/0105105. Their idea: use atomic ensembles (clouds of billions of atoms) as quantum memories that can both store a quantum state and emit an entangled photon on demand. The entanglement-generation and entanglement-swapping steps both use single photons heralded by detector clicks, so loss events are detected rather than silently destroying the protocol.

A DLCZ link works like this. At each node, you have an atomic ensemble that can be excited by a laser. When the laser pulses, with small probability a single collective excitation is created in the ensemble, and a single photon is emitted. The photon goes down the fibre to a beam splitter halfway between two nodes, interferes with the photon from the other side, and a detector click heralds a Bell pair between the two ensembles. Because the click comes after the photons have traversed the fibre, fibre loss just reduces the heralding probability — it does not corrupt the Bell pair, only delay it. After heralding, the Bell pair lives in the two atomic ensembles until you are ready to swap.

Modern variants (second-generation repeaters; all-photonic repeaters; Bunny-ear-style memory architectures) improve DLCZ in various ways, but the swap primitive you derived in this chapter is the same. The research frontier is the memory lifetime and the single-photon efficiency, not the protocol logic.

Photonic Bell-measurement efficiency

A central bottleneck is the Bell measurement itself. On superconducting or trapped-ion hardware, where the two qubits are in the same fridge, a Bell measurement is a textbook CNOT + H + computational readout — close to deterministic. But on photon-based platforms, where each qubit is a single polarised photon, CNOT gates are hard. The standard linear-optical Bell measurement (Lütkenhaus-Calsamiglia, 1999) uses beam splitters and photon detectors, and can distinguish only two of the four Bell states — |\Psi^+\rangle and |\Psi^-\rangle. The other two (|\Phi^\pm\rangle) produce coincidences that are indistinguishable from one another, and the protocol has to discard those events. The efficiency cap is 50%.

This looks crippling but it has a workaround: pad the system with extra entangled photons ("ancillary entanglement") and you can boost the efficiency arbitrarily close to 100%, at the cost of needing more resources per swap. Alternatively, use hybrid platforms — photon for communication, stationary qubit (atom or NV centre) for the Bell measurement — which sidesteps the linear-optics bound. Most current repeater proposals are hybrid.

Experimental lineage and India's programme

The original theory is due to Zukowski, Zeilinger, Horne, and Ekert — Event-ready-detectors Bell experiment via entanglement swapping, Phys. Rev. Lett. 71, 4287 (1993). The first experimental demonstration was the Pan, Bouwmeester, Weinfurter, Zeilinger paper — Experimental entanglement swapping: entangling photons that never interacted, Phys. Rev. Lett. 80, 3891 (1998), arXiv:quant-ph/9803032. They prepared two independent polarisation-entangled photon pairs from parametric down-conversion, performed a partial Bell measurement on one photon from each pair, and verified via polarisation-correlation experiments that the remaining two photons (which had never met) were themselves entangled.

Subsequent experiments have extended this to atomic ensembles (the DLCZ-style Kimble group at Caltech), trapped ions (Innsbruck, Oxford), solid-state colour centres (NV centres in diamond, Delft), and satellite links (Chinese Micius, 2017). The current record for longest entanglement swap via fibre is around 100 km per link (as of 2024); satellite-to-ground links extend the effective range much further by using free-space propagation above the atmosphere.

In India, Raman Research Institute (Bangalore) runs an active quantum-optics programme producing polarisation-entangled photon pairs; ISRO's Space Applications Centre (Ahmedabad) has demonstrated free-space entanglement distribution (2022 satellite QKD demonstration); IIT-Bombay's Center of Excellence in Quantum Information, Computation, Science and Technology works on memory-based repeater architectures; and the National Quantum Mission explicitly funds long-distance quantum communication as one of its four verticals. The ambition is a metro-scale and eventually national-scale quantum network with Bengaluru, Mumbai, Hyderabad, and Delhi as initial hubs — an architecture in which the protocol derived in this chapter is the load-bearing primitive at every repeater station.

Multipartite swapping — distributing GHZ and cluster states

You can swap more than two Bell pairs. A three-party variant: Alice holds one half of pair 1, Bob holds one half of pair 2, Charlie holds one half of pair 3; the other halves all sit at a central node. The central node performs a GHZ measurement (the three-qubit analogue of a Bell measurement) on its three qubits. Outcome-dependent corrections leave Alice, Bob, and Charlie sharing a three-qubit GHZ state — all three of them entangled with each other even though no pair of them was connected before.

This generalisation is how cluster states and graph states are distributed in measurement-based quantum computing networks. You grow large entangled states by stitching small ones together at a central "stitching" node. The elementary primitive is entanglement swapping; the structure is the graph you want to build.

Where this leads next

• Quantum teleportation — the one-qubit protocol that entanglement swapping generalises.
• Superdense coding — the other direction of the teleportation–coding duality.
• Quantum-repeater architectures — how chains of swaps power long-distance quantum links.
• Quantum key distribution — the immediate application of distributed Bell pairs over long distances.
• GHZ and W states — the multipartite entanglement that multi-swapping distributes.
• No-communication theorem — the hard proof that no swap, no matter how clever, beats the speed of light.

References

1. J.-W. Pan, D. Bouwmeester, H. Weinfurter, A. Zeilinger, Experimental Entanglement Swapping: Entangling Photons That Never Interacted (1998) — arXiv:quant-ph/9803032. The first experimental demonstration.
2. Wikipedia, Entanglement swapping — concise overview with the key identities.
3. John Preskill, Lecture Notes on Quantum Computation, Ch. 4 — theory.caltech.edu/~preskill/ph229. The cleanest pedagogical derivation.
4. L.-M. Duan, M. Lukin, J. I. Cirac, P. Zoller, Long-distance quantum communication with atomic ensembles and linear optics (2001) — arXiv:quant-ph/0105105. The DLCZ repeater paper.
5. Government of India, National Quantum Mission mission document — the ₹6003-crore, eight-year programme whose long-distance-communication pillar is built on entanglement swapping.
6. Nielsen and Chuang, Quantum Computation and Quantum Information (2010), §1.3.7 and §12 — Cambridge.