Simple Algorithm
Quantum Computing · Rydberg State · 28 min

Entanglement Swapping & Quantum Repeaters

Building a quantum internet across long distances.

Fiber loses photons exponentially: across 1000 km, even a 10 GHz single-photon source would deliver roughly one photon every 300 years, and no-cloning forbids amplifying them along the way. The fix is to teleport entanglement itself: swap short links into long ones, purify away the errors, and park the halves in quantum memories while you wait.

What you’ll learn

  • Quantify the repeaterless limit: fiber transmittance η=10−αL/10\eta=10^{-\alpha L/10} and the PLOB bound −log⁡2(1−η)-\log_2(1-\eta).
  • Derive entanglement swapping and its action on Werner states, F′=F2+(1−F)2/3F'=F^2+(1-F)^2/3.
  • Show how quantum memories turn a rate ∝η\propto\eta into one ∝η\propto\sqrt\eta with a single intermediate station, and how nesting extends this.
  • Compare repeater architectures: BDCZ nested purification, DLCZ atomic ensembles, and error-corrected generations.
  • Relate end-to-end fidelity to the thresholds of applications: entanglement, CHSH violation and QKD.

The loss wall

Standard telecom fiber attenuates light by about α≈0.2\alpha\approx0.2 dB/km at 1550 nm, so a photon survives a length LL with probability

η=10−αL/10=e−L/Latt,Latt≈22 km.\eta=10^{-\alpha L/10}=e^{-L/L_{\mathrm{att}}},\qquad L_{\mathrm{att}}\approx22\ \text{km}.

Classical networks beat loss with amplifiers. Quantum signals cannot be amplified noiselessly: a perfect amplifier would be a cloner, and phase-insensitive linear amplifiers must add noise (Caves 1982).

Think first A 10 GHz single-photon source feeds 500 km of 0.2 dB/km fiber, with perfect detectors. How many photons arrive per second? What about 1000 km?

500 km means 100 dB of loss, η=10−10\eta=10^{-10}: about one photon per second. At 1000 km the loss is 200 dB and η=10−20\eta=10^{-20}: one photon every 101010^{10} s, roughly once every 300 years.

So every point-to-point link scales linearly in η\eta, that is, exponentially in distance. Beating this requires intermediate quantum nodes. For key distribution specifically, twin-field QKD (Lucamarini, Yuan, Dynes & Shields 2018) already reaches η\sqrt\eta scaling with a single untrusted middle station that merely interferes weak pulses.

Entanglement swapping

Split the distance in two. Node A shares a Bell pair with a middle node C (qubits 1 and 2), and C shares another with node B (qubits 3 and 4). Now teleport C's qubit 2 to B through the second pair:

DerivationThe swapping identity
  1. ∣Φ+⟩12∣Φ+⟩34=12∑i∣i⟩1⊗∣i⟩2∣Φ+⟩34\ket{\Phi^+}_{12}\ket{\Phi^+}_{34}=\tfrac{1}{\sqrt2}\sum_{i}\ket{i}_1\otimes\ket{i}_2\ket{\Phi^+}_{34}

    Expand the first pair.

  2. =12∑x,y∣βxy⟩23⊗(I1⊗XyZx)∣Φ+⟩14=\tfrac12\sum_{x,y}\ket{\beta_{xy}}_{23}\otimes\big(\id_1\otimes X^yZ^x\big)\ket{\Phi^+}_{14}

    Apply the teleportation identity ∣i⟩2∣Φ+⟩34=12∑xy∣βxy⟩23XyZx∣i⟩4\ket i_2\ket{\Phi^+}_{34}=\tfrac12\sum_{xy}\ket{\beta_{xy}}_{23}X^yZ^x\ket i_4, then recombine the sum over ii.

  3. (I⊗XyZx)∣Φ+⟩=∣βxy⟩ ⟹ ∣Φ+⟩12∣Φ+⟩34=12∑x,y∣βxy⟩23∣βxy⟩14(\id\otimes X^yZ^x)\ket{\Phi^+}=\ket{\beta_{xy}}\ \Longrightarrow\ \ket{\Phi^+}_{12}\ket{\Phi^+}_{34}=\tfrac12\sum_{x,y}\ket{\beta_{xy}}_{23}\ket{\beta_{xy}}_{14}

    A Bell measurement at C projects qubits 1 and 4, which never interacted, onto the same Bell state. C announces (x,y)(x,y) and B corrects to Φ+\Phi^+. ■\blacksquare

Noise compounds; purification fights back

DerivationSwapping noisy links
  1. ρW(p)=p ∣Φ+⟩ ⁣⟨Φ+∣+(1−p)I4,F=⟨Φ+∣ρW∣Φ+⟩=1+3p4\rho_W(p)=p\,\dyad{\Phi^+}+(1-p)\tfrac{\id}{4},\qquad F=\bra{\Phi^+}\rho_W\ket{\Phi^+}=\tfrac{1+3p}{4}

    Model each elementary link as a Werner state. Any two-qubit state can be twirled into this form by random bilateral rotations, without changing FF.

  2. ρW(p)=(id⊗Dp)(∣Φ+⟩ ⁣⟨Φ+∣),Dp(ρ)=pρ+(1−p)I2\rho_W(p)=(\mathrm{id}\otimes\mathcal D_p)\big(\dyad{\Phi^+}\big),\qquad\mathcal D_p(\rho)=p\rho+(1-p)\tfrac{\id}{2}

    A Werner state is Φ+\Phi^+ with one half sent through a depolarizing channel.

  3. swap(ρW(p1),ρW(p2))=ρW(p1p2)\mathrm{swap}\big(\rho_W(p_1),\rho_W(p_2)\big)=\rho_W(p_1p_2)

    Swapping teleports one half of the first link through the second. Teleportation through ρW(p2)\rho_W(p_2) acts as Dp2\mathcal D_{p_2}, and Dp2∘Dp1=Dp1p2\mathcal D_{p_2}\circ\mathcal D_{p_1}=\mathcal D_{p_1p_2}. (A perfect Bell measurement and Pauli correction are assumed.)

  4. F′=1+3p1p24 =F1=F2=F F2+(1−F)23F'=\tfrac{1+3p_1p_2}{4}\ \overset{F_1=F_2=F}{=}\ F^2+\tfrac{(1-F)^2}{3}

    With nn equal links the end-to-end parameter is pnp^n: it decays exponentially and F→14F\to\tfrac14. Swapping alone cannot cover long distances.

Predict before you look

Two Werner links, each with fidelity F=0.9F=0.9, are swapped with a perfect Bell measurement. What is the fidelity of the resulting long link?

Show the answer

0.813

F′=F2+(1−F)23=0.81+0.013≈0.813F'=F^2+\tfrac{(1-F)^2}{3}=0.81+\tfrac{0.01}{3}\approx0.813. Equivalently p=4F−13=1315p=\tfrac{4F-1}{3}=\tfrac{13}{15}, and p2≈0.751p^2\approx0.751 gives F′=1+3p24≈0.813F'=\tfrac{1+3p^2}{4}\approx0.813. Each swap multiplies the Werner parameters, so the noise compounds.

Entanglement purification (Bennett, Brassard, Popescu, Schumacher, Smolin & Wootters 1996; Deutsch et al. 1996) uses local operations and two-way classical communication to turn noisy pairs into fewer, better ones. In the recurrence protocol both parties apply CNOTs between two pairs and measure the target pair, keeping the source pair only if their outcomes agree. For Werner inputs of fidelity FF:

F′=F2+19(1−F)2F2+23F(1−F)+59(1−F)2F'=\frac{F^2+\frac19(1-F)^2}{F^2+\frac23F(1-F)+\frac59(1-F)^2}
The denominator is the success probability. F′>FF'>F whenever 12<F<1\tfrac12<F<1: for example F=0.9↦0.926F=0.9\mapsto0.926, with success probability 0.8760.876.

Nested purification (Briegel, Dür, Cirac & Zoller 1998): split the channel into N=2kN=2^k segments; generate and purify elementary pairs; swap neighbors to double the distance; purify again; repeat kk times. If local operations err below a threshold at the percent level, the final fidelity can be held fixed while the resources grow only polynomially with distance. With "entanglement pumping", purifying one pair repeatedly with fresh elementary pairs, each node needs a number of memory qubits that grows only logarithmically with distance.

Memories and repeater architectures

Why do repeaters need memories? Put one station at the midpoint. Each half-link of length L/2L/2 is heralded with probability p≈ηp\approx\sqrt\eta per attempt (times detector and coupling efficiencies). Without memories, both halves must succeed in the same attempt, with probability p2=ηp^2=\eta: no better than direct transmission. With memories, a successful half simply waits for the other.

DerivationWaiting time with memories
  1. T1,T2∼Geom(p),Emax⁡(T1,T2)=ET1+ET2−Emin⁡(T1,T2)T_1,T_2\sim\mathrm{Geom}(p),\qquad\mathbb E\max(T_1,T_2)=\mathbb ET_1+\mathbb ET_2-\mathbb E\min(T_1,T_2)

    Count the attempts until each half succeeds; the maximum is the sum minus the minimum.

  2. =2p−11−(1−p)2=3−2pp(2−p) ≈ 32p=\frac2p-\frac{1}{1-(1-p)^2}=\frac{3-2p}{p(2-p)}\ \approx\ \frac{3}{2p}

    The minimum of two independent geometric variables is geometric with success probability 1−(1−p)21-(1-p)^2.

  3. Rmem≈23 p=23ηvs.Rno mem=p2=ηR_{\mathrm{mem}}\approx\tfrac23\,p=\tfrac23\sqrt\eta\qquad\text{vs.}\qquad R_{\text{no mem}}=p^2=\eta

    Rate per attempt, assuming a deterministic swap. The square root is the whole game. With 2k2^k segments the elementary success probability becomes η1/2k\eta^{1/2^k}, at the price of waiting-time and swap-failure factors that grow only polynomially with the number of segments.

Check your understanding

A single midpoint station heralds entanglement in both halves of the link, but has no quantum memories: a pair is distributed only if both halves succeed in the same attempt. How does the end-to-end rate scale with the total transmittance η\eta?

  1. As η\eta: no better than direct transmission
  2. As η\sqrt\eta
  3. As η2\eta^2
  4. Independently of η\eta
Show the answer

A. As η\eta: no better than direct transmission

Each half succeeds with probability η\sqrt\eta, and both must succeed simultaneously: (η)2=η(\sqrt\eta)^2=\eta. Memories turn simultaneous success into asynchronous success, which is what gives η\sqrt\eta.

GenerationLoss handled byOperation errors handled byRate limited byExamples
1stHeralded entanglement generationHeralded purification (two-way communication)Classical signalling over ever longer distancesBDCZ 1998; DLCZ 2001
2ndHeralded entanglement generationQuantum error correction (one-way)Heralding time of one link, ∼L0/c\sim L_0/cJiang et al. 2009; Munro et al. 2010
3rdQuantum error correction (loss-tolerant codes)Quantum error correctionLocal operation timeMunro et al. 2012; Muralidharan et al. 2014
The classification of Muralidharan, Li, Lütkenhaus, Lukin & Jiang (2016). Third-generation schemes need very low loss per hop, hence closely spaced stations and high-fidelity local gates. Memoryless "all-photonic" repeaters based on photonic graph states (Azuma, Tamaki & Lo 2015) are another route.

Fidelity thresholds and the frontier

ApplicationWerner conditionFidelity needed
Entangled (and hence distillable)p>13p>\tfrac13F>12F>\tfrac12
Teleportation beats the classical Fˉ=23\bar F=\tfrac23p>13p>\tfrac13F>12F>\tfrac12
Violates the CHSH inequalityp>12p>\tfrac{1}{\sqrt2}F>1+3/24≈0.78F>\tfrac{1+3/\sqrt2}{4}\approx0.78
BB84-type key with one-way post-processing (Shor–Preskill: Q<11%Q<11\%)p≳0.78p\gtrsim0.78F≳0.835F\gtrsim0.835
For Werner states F=1+3p4F=\tfrac{1+3p}{4}, and the bit-error rate in either basis is Q=1−p2=23(1−F)Q=\tfrac{1-p}{2}=\tfrac23(1-F). Two-way post-processing tolerates higher error rates.
Think first Elementary links have F=0.99F=0.99 and there is no purification. How many links can be chained by swapping before the end-to-end pair stops violating CHSH?

p=4(0.99)−13≈0.9867p=\tfrac{4(0.99)-1}{3}\approx0.9867, and nn links give pnp^n. CHSH violation needs pn>1/2p^n>1/\sqrt2, i.e. n<ln⁡(1/2)/ln⁡p≈25.8n<\ln(1/\sqrt2)/\ln p\approx25.8: at most 25 links (24 swaps). Even excellent links run out after a few dozen hops, which is why long chains need purification or error correction.

Key takeaways

  • Direct transmission is capped at −log⁡2(1−η)≈1.44 η-\log_2(1-\eta)\approx1.44\,\eta (PLOB): exponentially small in distance. Only intermediate quantum nodes beat it.
  • Swapping: ∣Φ+⟩∣Φ+⟩=12∑∣βxy⟩23∣βxy⟩14\ket{\Phi^+}\ket{\Phi^+}=\tfrac12\sum\ket{\beta_{xy}}_{23}\ket{\beta_{xy}}_{14}. For Werner links the parameters multiply, F′=F2+(1−F)2/3F'=F^2+(1-F)^2/3, so purification is unavoidable.
  • Memories turn simultaneous success (η\eta) into asynchronous success (≈23η\approx\tfrac23\sqrt\eta per middle station); nested swapping and purification give polynomial overhead (BDCZ).
  • Architectures trade two-way heralding and purification for quantum error correction (three generations); DLCZ showed how far ensembles and linear optics can go.
  • Applications set fidelity thresholds: 12\tfrac12 for entanglement, ≈0.78\approx0.78 for CHSH, ≈0.835\approx0.835 for one-way BB84-type keys.

Check yourself

In the interactive lesson these are scored with your confidence, and what you miss comes back for review.

Question 1

At node C, a Bell measurement on qubits 2 and 3 of ∣Φ+⟩12∣Φ+⟩34\ket{\Phi^+}_{12}\ket{\Phi^+}_{34} returns Ψ−\Psi^-. What state do qubits 1 and 4 now share?

  1. Φ+\Phi^+
  2. A product state
  3. Ψ−\Psi^-
  4. Φ−\Phi^-
Show the answer

C. Ψ−\Psi^-

From ∣Φ+⟩12∣Φ+⟩34=12∑xy∣βxy⟩23∣βxy⟩14\ket{\Phi^+}_{12}\ket{\Phi^+}_{34}=\tfrac12\sum_{xy}\ket{\beta_{xy}}_{23}\ket{\beta_{xy}}_{14}, the outcome β11=Ψ−\beta_{11}=\Psi^- leaves qubits 1 and 4 in Ψ−\Psi^-. A Pauli correction at B then turns it into Φ+\Phi^+.

Question 2

Two Werner links with p1=0.9p_1=0.9 and p2=0.8p_2=0.8 are swapped with a perfect Bell measurement. What is the Werner parameter of the result?

  1. 0.850.85
  2. 0.720.72
  3. 0.800.80
  4. 0.980.98
Show the answer

B. 0.720.72

Swapping composes the two depolarizing channels, and Dp2∘Dp1=Dp1p2\mathcal D_{p_2}\circ\mathcal D_{p_1}=\mathcal D_{p_1p_2}: p=0.72p=0.72, so F=1+3(0.72)4=0.79F=\tfrac{1+3(0.72)}{4}=0.79.

Question 3

For a pure-loss channel with transmissivity η=10−3\eta=10^{-3}, the PLOB bound on secret-key bits per channel use is about:

  1. 1.4×10−31.4\times10^{-3}
  2. 3.2×10−23.2\times10^{-2}, that is, η\sqrt\eta
  3. 0.50.5
  4. 10−610^{-6}
Show the answer

A. 1.4×10−31.4\times10^{-3}

−log⁡2(1−η)≈η/ln⁡2≈1.44×10−3-\log_2(1-\eta)\approx\eta/\ln2\approx1.44\times10^{-3}. Beating this linear scaling requires intermediate quantum nodes; η\sqrt\eta-type scaling is what a single middle station with memories (or twin-field QKD) can reach.

Question 4

What distinguishes third-generation quantum repeaters?

  1. They use satellites instead of fiber
  2. They need no entanglement at all
  3. They amplify the quantum signal at each station
  4. They correct both loss and operation errors with quantum error correction, so only one-way classical communication is needed
Show the answer

D. They correct both loss and operation errors with quantum error correction, so only one-way classical communication is needed

Encoding qubits in loss-tolerant codes and correcting at every station removes the heralding round trips; the rate is then limited by local operation speed, at the price of very low loss per hop and high-fidelity local gates.

Further reading

  • M. Żukowski, A. Zeilinger, M. A. Horne & A. K. Ekert, "'Event-ready-detectors' Bell experiment via entanglement swapping", Physical Review Letters 71, 4287 (1993).
  • H.-J. Briegel, W. Dür, J. I. Cirac & P. Zoller, "Quantum repeaters: The role of imperfect local operations in quantum communication", Physical Review Letters 81, 5932 (1998).
  • L.-M. Duan, M. D. Lukin, J. I. Cirac & P. Zoller, "Long-distance quantum communication with atomic ensembles and linear optics", Nature 414, 413 (2001).
  • N. Sangouard, C. Simon, H. de Riedmatten & N. Gisin, "Quantum repeaters based on atomic ensembles and linear optics", Reviews of Modern Physics 83, 33 (2011).
  • S. Pirandola, R. Laurenza, C. Ottaviani & L. Banchi, "Fundamental limits of repeaterless quantum communications", Nature Communications 8, 15043 (2017).