Quantum Repeaters

A quantum repeater is the architecture that makes long-distance quantum communication possible at all. The problem it solves is brutal and specific: quantum signals cannot be amplified. A classical telecom repeater simply copies a weak pulse into a strong one, but the no-cloning theorem forbids the deterministic copying of an unknown quantum state, and any physical amplifier injects noise that shreds the very entanglement one is trying to send. Meanwhile loss in optical fibre is exponential — the probability a single photon survives a length $L$ is $\eta(L)=e^{-L/L_{\mathrm{att}}}$ with an attenuation length of only about $22\,\mathrm{km}$ at the $1550\,\mathrm{nm}$ telecom wavelength. Push that to $1000\,\mathrm{km}$ and the survival probability collapses to $\sim 2\times10^{-20}$ ; even a $10\,\mathrm{GHz}$ single-photon source would then wait roughly $170$ years between successes. Direct transmission is not slow, it is impossible. The repeater idea, due to Briegel, Dür, Cirac and Zoller in 1998, refuses the impossible event and buys many possible ones instead. Break the span into $N=2^n$ short elementary links, each short enough that entanglement can be heralded across it with workable probability. Then fuse adjacent entangled pairs into longer-range entanglement by entanglement swapping — a Bell-state measurement on two co-located inner qubits that projects the distant outer qubits into an entangled state they were never near enough to share directly. Nest the swaps in a binary tree and only $n=\log_2 N$ levels span the whole distance. Two demons remain: links arrive at random times, so the scheme needs quantum memories to buffer a completed half-link while its partner is retried; and every swap multiplies infidelities, so it interleaves entanglement purification to haul fidelity back up. The payoff is the whole game — an exponential cost in distance becomes a polynomial one.

Key takeaways

Direct transmission through fibre loses photons exponentially, $\eta(L)=e^{-L/L_{\mathrm{att}}}$ with $L_{\mathrm{att}}\approx22\,\mathrm{km}$ , so point-to-point entanglement over $1000\,\mathrm{km}$ succeeds with probability $\sim10^{-20}$ — utterly hopeless, and no-cloning forbids amplifying the way past it.
A quantum repeater splits the span into $N=2^n$ short elementary links, distributes entanglement across each, and stitches them together with $n=\log_2 N$ nested levels of entanglement swapping.
Because swapping multiplies infidelities and heralded links arrive at random times, the architecture interleaves entanglement purification (to restore fidelity) and quantum memories (to synchronise probabilistic links).
The signature payoff is exponential-to-polynomial: the resource and time cost falls from $e^{L/L_{\mathrm{att}}}$ to roughly $\mathrm{poly}(L)$ , with the polynomial degree set by nesting depth and purification overhead.
To count as a genuine repeater a device must beat the repeaterless PLOB capacity $R\approx1.44\,\eta$ ; even a single memory-assisted midpoint achieves the celebrated $\sqrt{\eta}$ scaling and clears the bound beyond a few attenuation lengths.
Figures of merit are the end-to-end fidelity $F$ , the secret-key rate per channel use, and the memory coherence time relative to the link-and-signalling latency $\sim L_0/c$ ; later generations trade memories for quantum error correction.
Key idea: Replace one exponentially-improbable long transmission by many feasible short links, then fuse them with entanglement swapping while periodically restoring fidelity by purification and buffering timing with quantum memories — converting an exponential cost in distance into a polynomial one.

Where this lives today

The quantum repeater has crossed from theory into the laboratory. The most consequential recent milestone is the experimental defeat of the PLOB bound: twin-field quantum key distribution, which places a phase-referenced interference station at the midpoint of a fibre link, achieves precisely the $\sqrt{\eta}$ scaling of a single-node repeater and has distributed secret keys over more than $500\,\mathrm{km}$ of fibre while beating the repeaterless capacity — the first hardware to clear the wall. On the memory-based front, elementary repeater links have been demonstrated with nitrogen-vacancy centres in diamond (the Delft multi-node network, with heralded entanglement and real-time feedforward), with trapped ions and neutral atoms, and with rare-earth-ion and cold-atomic-ensemble memories now reaching multimode storage and coherence times approaching the seconds needed for continental spans. Satellite links sidestep fibre loss entirely: the Micius satellite distributed entangled photon pairs to ground stations over $1200\,\mathrm{km}$ . The open frontier is integration — combining efficient telecom-wavelength memories, near-deterministic Bell-state measurements, and loss-tolerant photonic codes into a node that beats the bound at metropolitan distances and chains into a genuine multi-hop network: the first router of the quantum internet.

The physical picture

Picture carrying a fragile message across an ocean by ferry. One direct ferry that almost always sinks is worthless. So you build a chain of islands and run a short, reliable ferry between each pair; at every island you hand the message across. Entanglement swapping is that handoff — a measurement at the island that fuses two short-range entangled pairs into one longer-range pair. The catch is that the quantum message frays a little at each handoff, and the ferries never run on schedule: entanglement is heralded at random moments. So you also need a photocopier for fidelity (purification) and waiting rooms that keep the message intact between ferries (quantum memories). The first figure shows the architecture and the numbers that make it win.

Figure

The nested quantum repeater. Direct transmission of a photon over 1000 km survives with probability η ≈ 2×10⁻²⁰. Instead the span is cut into N = 8 elementary links of L₀ = 125 km, each heralding entanglement with probability η(L₀) ≈ 3.4×10⁻³, then fused by log₂8 = 3 nested levels of entanglement swapping into one end-to-end Bell pair

The exponential-to-polynomial conversion is the entire point. Direct transmission costs $\sim e^{L/L_{\mathrm{att}}}$ attempts; the nested repeater needs only $\log_2 N$ swap levels, and the per-level overhead from probabilistic heralding and purification multiplies into a mere power of $N$ , that is a power of $L$ . You trade one impossible event for many possible ones plus bookkeeping. Two subtleties decide whether the bookkeeping is affordable. First, fidelity multiplies under swapping and is restored only above a hard threshold: two Werner pairs of fidelity $F$ yield a better pair only when $F>1/2$ , so a repeater must swap until fidelity sags toward the threshold, then purify back up. Second, time: establishing a link, waiting for its partner, and broadcasting the classical outcomes each cost $\sim L_0/c$ , and if a memory decoheres before the rounds finish, no amount of nesting helps. The second figure shows the reward for getting this right — beating the fundamental repeaterless capacity.

Figure

Secret-key rate versus distance in optical fibre on a log scale. The repeaterless PLOB bound R ≈ 1.44 η falls with the full loss exponent −L/L_att, while a memory-assisted midpoint repeater scales as √η, halving the exponent. The lines cross near 46 km, and at 500 km the midpoint beats the repeaterless bound by a factor of about 30000

The core equations

The starting point is the exponential loss law of an optical fibre — the wall that every repeater exists to defeat. Loss is quoted in decibels per kilometre; converting to a natural exponential fixes the attenuation length.

\eta(L)=e^{-L/L_{\mathrm{att}}},\qquad L_{\mathrm{att}}=\frac{10}{\alpha\ln 10},\quad \alpha\approx0.2\,\mathrm{dB/km}\Rightarrow L_{\mathrm{att}}\approx22\,\mathrm{km} \qquad (1)

where: $\eta(L)$ — single-photon transmission probability over fibre length $L$ ; $L_{\mathrm{att}}$ — attenuation length, about $22\,\mathrm{km}$ at $1550\,\mathrm{nm}$ ; $\alpha$ — attenuation coefficient in dB/km, the standard telecom value being $0.2$ . Read it as: the survival probability decays by a factor $e$ every $22\,\mathrm{km}$ , so distance enters the cost exponentially. The repeater cuts the total span into $N=2^n$ elementary links. The key observation is that each short link is far more transmissive than the whole, and — crucially — heralding plus memory let each link succeed independently rather than all at once.

N=2^n,\qquad L_0=\frac{L}{N},\qquad \eta(L_0)=e^{-L/(N L_{\mathrm{att}})} \qquad (2)

where: $N$ — number of elementary links; $n$ — nesting depth, equal to the number of swap levels, $n=\log_2 N$ ; $L_0=L/N$ — elementary-link length. Read it as: shrink each hop by a factor $N$ and its survival probability climbs from $e^{-L/L_{\mathrm{att}}}$ to $e^{-L/(N L_{\mathrm{att}})}$ , exponentially closer to one. Adjacent entangled pairs are fused by entanglement swapping. A Bell-state measurement on the two inner qubits of $A\!-\!B$ and $B'\!-\!C$ (with $B,B'$ co-located) projects the distant outer qubits $A,C$ into an entangled state, up to a Pauli correction fixed by the two classical outcome bits.

|\Phi^{+}\rangle_{AB}\,|\Phi^{+}\rangle_{B'C}\ \xrightarrow{\ \text{BSM on }B,B'\ }\ |\Phi^{+}\rangle_{AC}\quad(\text{up to Pauli correction}) \qquad (3)

where: $|\Phi^{+}\rangle=\tfrac{1}{\sqrt2}(|00\rangle+|11\rangle)$ — a Bell state; BSM — Bell-state measurement on the two co-located inner qubits; $A,C$ — distant outer qubits, entangled afterward despite never having interacted. Read it as: measurement, not interaction, is what creates the long-range entanglement — the defining trick of the repeater. Swapping is not free: it multiplies the infidelities of its inputs. Modelling each pair as a Werner state, the output fidelity is dominated by the product of the input fidelities.

F_{\mathrm{swap}}=F_1 F_2+\frac{(1-F_1)(1-F_2)}{3} \qquad (4)

where: $F_1,F_2$ — fidelities of the two pairs entering the swap; $F_{\mathrm{swap}}$ — fidelity of the pair leaving it. Read it as: perfect inputs ( $F_1=F_2=1$ ) give $F_{\mathrm{swap}}=1$ , but repeated swapping drives fidelity toward the $1/4$ floor of a two-qubit maximally mixed state — which is exactly why purification must be interleaved. Recurrence purification reverses the decay by consuming two low-fidelity pairs to produce one better pair, but only above a threshold. The DEJMPS/BBPSSW update for Werner input is:

F'=\frac{F^2+\big(\tfrac{1-F}{3}\big)^2}{\,F^2+\tfrac{2}{3}F(1-F)+\tfrac{5}{9}(1-F)^2\,},\qquad p_{\mathrm{succ}}=F^2+\tfrac{2}{3}F(1-F)+\tfrac{5}{9}(1-F)^2 \qquad (5)

where: $F$ — Werner fidelity of each of the two consumed pairs; $F'$ — fidelity of the kept pair on success; $p_{\mathrm{succ}}$ — probability the bilateral-CNOT-and-measure step succeeds. Read it as: $F'>F$ if and only if $F>1/2$ ; below the threshold purification makes matters worse, above it iteration converges toward $F\to1$ . Any device claiming to be a repeater must beat the best possible repeaterless rate. The PLOB bound is the proven secret-key capacity of the pure-loss bosonic channel — the exponential wall itself.

R_{\mathrm{PLOB}}=-\log_2(1-\eta)\ \xrightarrow{\ \eta\ll1\ }\ \frac{\eta}{\ln 2}\approx1.44\,\eta\quad(\text{bits/mode}) \qquad (6)

where: $R_{\mathrm{PLOB}}$ — Pirandola–Laurenza–Ottaviani–Banchi capacity of a lossy channel with no repeater; $\eta$ — end-to-end transmittance. Read it as: the best possible repeaterless key rate falls off linearly in $\eta$ , hence exponentially in $L$ — the benchmark a genuine repeater must surpass. Finally, the throughput of a real repeater is throttled by probabilistic gates and by latency, and the whole scheme rests on quantum memories staying coherent long enough.

R_{\mathrm{rep}}\sim\frac{P_{\mathrm{link}}\,P_{\mathrm{swap}}^{\,n}}{T_0},\qquad T_0\gtrsim\frac{L_0}{c},\qquad \tau_{\mathrm{mem}}\gtrsim\frac{L}{c}\cdot\frac{1}{P_{\mathrm{eff}}} \qquad (7)

where: $P_{\mathrm{link}}$ — per-attempt elementary-link probability (includes $\eta(L_0)$ and detector efficiency); $P_{\mathrm{swap}}$ — success probability of one swap (e.g. $1/2$ for a linear-optics BSM); $T_0$ — per-round time, bounded below by the classical-signalling latency $L_0/c$ ; $\tau_{\mathrm{mem}}$ — memory coherence time; $P_{\mathrm{eff}}$ — effective per-round success probability. Read it as: rate drops with each swap level and each round-trip, and a stored half-link must stay coherent for the average $\sim(L/c)/P_{\mathrm{eff}}$ it waits for its partner — the central experimental bottleneck.

Derivation AWhy direct transmission is impossible and segmentation is not

The goal is to quantify, in hard numbers, why a thousand kilometres of fibre defeats direct transmission by an astronomical margin, and to see precisely which ingredient of the repeater — segmentation, heralding, or memory — rescues tractability. We start from the attenuation law, count the attempts a single end-to-end success demands, then segment the channel and count again, watching a number near $10^{20}$ shrink to a number near $10$ .

Step 1. Evaluate the direct cost. Over $L=1000\,\mathrm{km}$ with $L_{\mathrm{att}}=22\,\mathrm{km}$ the survival probability is $\eta(L)=e^{-1000/22}\approx2\times10^{-20}$ , so the expected number of attempts per success is $\eta(L)^{-1}=e^{L/L_{\mathrm{att}}}\approx5\times10^{19}$ . At a $10\,\mathrm{GHz}$ source the mean wait is $\sim1/(10^{10}\times2\times10^{-20})\approx5\times10^{9}\,\mathrm{s}\approx170$ years per photon. Why: this fixes the scale of the disaster — the failure is exponential in distance, not merely large, so no faster source or better detector can close a twenty-order-of-magnitude gap.

Step 2. Segment the span. Choose $N=16=2^4$ links of $L_0=62.5\,\mathrm{km}$ ; each now transmits with $\eta(L_0)=e^{-62.5/22}\approx0.058$ , a manageable few percent instead of $10^{-20}$ . Why: shrinking the hop moves the exponent from $-45$ to $-2.8$ , and a few-percent event is one you can actually wait for and retry.

Step 3. Herald and store each link independently. Because a completed link is heralded and parked in a quantum memory, you pay $\sim\eta(L_0)^{-1}\approx17$ attempts per link, and the links do not all have to succeed on the same shot. Naively demanding simultaneous success across all $N$ links would give $\eta(L_0)^N=e^{-L/L_{\mathrm{att}}}$ — the original catastrophe back again. Why: this is the crux. Segmentation alone is worthless; it is heralding-plus-memory that decouples the links, replacing a product of tiny probabilities by a sum of independent, retryable waits.

Step 4. Stitch with nested swaps. Only $n=\log_2 16=4$ levels of Bell-state measurement fuse the 16 short links into one end-to-end pair. The one $10^{-20}$ event has been replaced by $\mathcal{O}(N)$ few-percent events plus $\log_2 N$ swap rounds — polynomial, not exponential. Why: the logarithmic swap depth is what keeps the overhead polynomial; doubling the distance adds just one swap level, not a multiplicative loss factor. *Quick check: $\eta=e^{-L/22\,\mathrm{km}}$ gives $e^{-1000/22}=1.8\times10^{-20}$ and $e^{-62.5/22}=0.058$ , confirming the $170$ -year wait and $n=\log_2 16=4$ swap levels quoted above.*

Go deeper · Derivation B · 2 minFidelity decay across a swap chain and the purification threshold

Derivation BFidelity decay across a swap chain and the purification threshold

The goal is to show, cleanly, that ideal entanglement swapping multiplies the depolarisation parameter of Werner states, so that fidelity decays exponentially in the number of segments toward a hard floor, and to establish why purification must be interleaved rather than saved for the end. We parameterise a noisy pair by a single number, track it through a chain of swaps, and then invoke the purification threshold to see how the decay is arrested.

Step 1. Parameterise. Write each noisy pair as a Werner state $\rho(p)=p\,|\Phi^{+}\rangle\langle\Phi^{+}|+\tfrac{1-p}{4}\mathbb{1}$ , whose fidelity is $F=\tfrac{1+3p}{4}$ ; here $p=1$ is a perfect Bell pair ( $F=1$ ) and $p=0$ is maximally mixed ( $F=1/4$ ). Why: collapsing the state to one parameter $p$ makes the effect of swapping a simple arithmetic law rather than a density-matrix computation.

Step 2. Swap multiplies $p$ . An ideal Bell-state measurement fusing $\rho(p_1)$ and $\rho(p_2)$ yields a Werner state with $p_{\mathrm{out}}=p_1 p_2$ — equivalently the fidelity relation of equation (4). Why: the measurement keeps only the correlated component of each input, so the surviving order parameter is the product, never the sum, of the inputs.

Step 3. Chain the swaps. Fusing $k$ identical links gives $p_{\mathrm{end}}=\prod_i p_i=p^{k}$ , so $F_{\mathrm{end}}=\tfrac{1+3p^{k}}{4}\to1/4$ as $k$ grows. Concretely, $p=0.95$ across a $16$ -link chain gives $p^{16}=0.44$ and $F_{\mathrm{end}}=0.58$ — already grazing the threshold. Why: exponential decay in the number of segments means end-to-end fidelity would be worthless at continental scale without intervention, no matter how good each link is.

Step 4. Purify to arrest it. Recurrence purification (equation 5) raises fidelity whenever $F>1/2$ , so interleaving a purification round each time fidelity approaches the threshold pins the working fidelity in a stable window well above $1/4$ . Let $F$ fall below $1/2$ first and distillation only makes it worse. Why: this is the defining loop of the BDCZ scheme — swap to extend range and lower $F$ , purify to spend pairs and restore $F$ — and it must be interleaved because the threshold is a cliff, not a slope. *Quick check: the multiplicativity $p_{\mathrm{out}}=p_1p_2$ under ideal swapping and the $F>1/2$ recurrence threshold are well-established results; $0.95^{16}=0.440$ and $F=(1+3\times0.440)/4=0.580$ confirm the numbers quoted above.*

Go deeper · Derivation C · 1 minBeating the repeaterless PLOB bound with a single midpoint

Derivation CBeating the repeaterless PLOB bound with a single midpoint

The goal is to establish the benchmark that separates a true repeater from a mere relay, and to show that even the simplest memory-assisted station — one node at the midpoint — already breaks the fundamental repeaterless scaling, achieving the famous $\sqrt{\eta}$ improvement. We write down the wall, write down the midpoint rate, and solve for the crossover distance beyond which the station wins.

Step 1. The wall. The proven secret-key capacity of a pure-loss channel is $R_{\mathrm{PLOB}}=-\log_2(1-\eta)\approx1.44\,\eta=1.44\,e^{-L/L_{\mathrm{att}}}$ for $L\gg L_{\mathrm{att}}$ — the best any repeaterless scheme can do, falling with the full loss exponent. Why: without this ceiling "beating loss" is undefined; the bound turns "is it a real repeater?" into a sharp inequality.

Step 2. The midpoint rate. Place one memory-assisted station at $L/2$ : each half-link is established with probability $\sim\eta(L/2)$ , and because the two halves are heralded and buffered independently, the end-to-end rate scales as $R_{\mathrm{rep}}\sim\tfrac12\,\eta(L/2)=\tfrac12 e^{-L/(2L_{\mathrm{att}})}$ . Why: the memory lets each half succeed on its own clock, so the exponent is halved — this is exactly the $\sqrt{\eta}$ scaling, since $\eta(L/2)=\sqrt{\eta(L)}$ .

Step 3. Crossover. Setting $\tfrac12 e^{-L/(2L_{\mathrm{att}})}=1.44\,e^{-L/L_{\mathrm{att}}}$ and solving gives $L\approx46\,\mathrm{km}\approx2\,L_{\mathrm{att}}$ ; beyond this the midpoint wins, and the margin grows exponentially. At $L=500\,\mathrm{km}$ the bound gives $R_{\mathrm{PLOB}}\approx1.9\times10^{-10}$ while the midpoint gives $\sim5.8\times10^{-6}$ — an advantage near $3\times10^{4}$ . Why: the half-exponent is not a constant-factor win but a change of scaling, so the gap widens without limit, which is why repeaters matter precisely at long distance.

Step 4. Multi-node scaling. Adding $N$ stations shrinks the exponent by $N$ , $R\sim e^{-L/(N L_{\mathrm{att}})}$ , so nesting $\log$ -many levels approaches the polynomial regime — provided memories and swap fidelities keep up. Why: the single midpoint is only the first rung; the full nested repeater is this argument iterated, which is what turns $\sqrt{\eta}$ into genuine polynomial scaling. *Quick check: at $500\,\mathrm{km}$ , $\eta=e^{-500/22}=1.35\times10^{-10}$ gives $R_{\mathrm{PLOB}}=1.9\times10^{-10}$ and midpoint rate $5.8\times10^{-6}$ , a ratio of $\approx3\times10^{4}$ , with the crossover at $46\,\mathrm{km}$ — consistent with the results derived above.*

Limiting cases

Different physical regimes select different repeater generations and scalings:

Case	Result	Meaning
No repeater (direct transmission)	$R_{\mathrm{direct}}\propto\eta(L)=e^{-L/L_{\mathrm{att}}}$	Rate falls exponentially with distance; the PLOB bound caps the repeaterless key rate at $-\log_2(1-\eta)\approx1.44\,\eta$ bits/use for small $\eta$ .
First generation (BDCZ: swap + purify + memory)	$T(L)\sim L\,(\log L)^{c},\ c>0$	Polynomial time and resource scaling, but needs two-way classical communication and long memory coherence; throttled by the $L_0/c$ latency.
Second generation (encoded qubits, QEC for gate errors)	$R\propto\eta(L_0),\ \ \text{rate}\sim1/(L_0/c)$	Error correction replaces purification, so it is one-way for operation errors, but loss is still heralded, so memories remain; demands high gate fidelity.
Third generation (one-way / all-photonic, loss-tolerant QEC)	$R\sim1/t_{\mathrm{local}}$ , distance-independent	Codes handle both loss and operation errors; no two-way signalling and no long-lived memory, at the price of large photon and qubit overhead per node.

Common trapA quantum repeater amplifies or copies the signal like a classical one

It does neither, and it cannot. The no-cloning theorem forbids the deterministic copying that a classical optical amplifier performs, and any physical amplifier adds noise that destroys entanglement. A quantum repeater never touches the unknown message state — it distributes pre-shared entanglement, which is then consumed by teleportation and swapping to move quantum information without cloning it. Two further traps ride along. First, swapping multiplies fidelities, so without interleaved purification (or error correction) the end-to-end fidelity decays toward the $1/4$ floor as the number of segments grows — you cannot swap first and clean up once at the end. Second, finite memory coherence imposes a hard ceiling on distance: a stored half-link must survive the average $\sim(L/c)/P_{\mathrm{eff}}$ it waits for its partner, and once that exceeds $\tau_{\mathrm{mem}}$ the success probability collapses no matter how many nodes you add.

Worked examples

1 · Fidelity across a swap chain and the need to purify (fully worked)

Eight elementary links are each prepared as Werner pairs with fidelity $F_0=0.96$ . They are fused by three nested swap levels into one end-to-end pair, with no purification. What is the end-to-end fidelity, and does it clear the purification threshold?

Step 1. Convert fidelity to the Werner parameter. From $F=\tfrac{1+3p}{4}$ we solve $p_0=\tfrac{4F_0-1}{3}=\tfrac{4(0.96)-1}{3}=\tfrac{2.84}{3}=\mathbf{0.9467}$ . Each pair carries depolarisation parameter $p_0\approx0.947$ .

Step 2. Multiply through the chain. Ideal swapping multiplies the parameters, and fusing $8$ links into one end-to-end pair takes the product of all eight: $p_{\mathrm{end}}=p_0^{8}=0.9467^{8}=\mathbf{0.649}$ . Equivalently the three nested levels multiply $8=2^3$ links.

Step 3. Convert back to fidelity. $F_{\mathrm{end}}=\tfrac{1+3p_{\mathrm{end}}}{4}=\tfrac{1+3(0.649)}{4}=\tfrac{2.947}{4}=\mathbf{0.737}$ . The chain has fallen from $0.96$ to $0.74$ .

Step 4. Check the threshold and the trend. $F_{\mathrm{end}}=0.74$ still exceeds the $F>1/2$ purification threshold, so purification can still rescue it — but only just, and pushing to $16$ links gives $p_0^{16}=0.421$ and $F=0.566$ , a hair above threshold. End-to-end fidelity is F ≈ 0.74 for eight links; purification remains possible because F > 1/2, but the margin vanishes as the chain lengthens — which is why real repeaters purify at every level rather than once at the end.

2 · A single midpoint repeater versus the PLOB wall (fully worked)

Compare, at $L=400\,\mathrm{km}$ of standard fibre ( $L_{\mathrm{att}}=22\,\mathrm{km}$ ), the best possible repeaterless secret-key rate with the rate of one ideal memory-assisted midpoint station. By what factor does the midpoint win?

Step 1. End-to-end transmittance. $\eta(400)=e^{-400/22}=e^{-18.18}=\mathbf{1.27\times10^{-8}}$ .

Step 2. The repeaterless ceiling. For this tiny $\eta$ , $R_{\mathrm{PLOB}}\approx1.44\,\eta=1.44\times1.27\times10^{-8}=\mathbf{1.83\times10^{-8}}$ bits per channel use.

Step 3. The midpoint rate. Each $200\,\mathrm{km}$ half transmits with $\eta(200)=e^{-200/22}=e^{-9.09}=1.13\times10^{-4}$ , and the buffered midpoint rate scales as $R_{\mathrm{rep}}\sim\tfrac12\,\eta(200)=\mathbf{5.6\times10^{-5}}$ — note this is $\tfrac12\sqrt{\eta(400)}$ , the $\sqrt{\eta}$ scaling made explicit.

Step 4. The advantage. The ratio is $R_{\mathrm{rep}}/R_{\mathrm{PLOB}}=5.6\times10^{-5}/1.83\times10^{-8}\approx\mathbf{3.1\times10^{3}}$ . Even one ideal midpoint beats the fundamental repeaterless bound by roughly three thousand times at 400 km, and because the win is a change of scaling (√η versus η) the margin grows without limit at longer distances.

Physics of the portrait

The quantum repeater is a young idea with a sharp origin: it was born the moment three separate quantum-information primitives — teleportation, entanglement purification, and entanglement swapping — were assembled into one architecture that turned an exponential wall into a polynomial staircase.

Origins · 1 minHans Briegel and how this was discovered

Hans Briegel

born 1962 · quantum repeaters, entanglement purification, and measurement-based quantum computation

Hans Jürgen Briegel is a German-born theoretical physicist whose 1998 paper with Wolfgang Dür, Ignacio Cirac and Peter Zoller — universally abbreviated BDCZ — created the field of long-distance quantum communication by naming and solving its central problem. Working in Innsbruck, the group saw that entanglement swapping and entanglement purification, each already known, could be nested together so that fidelity is restored at every level of a binary swapping tree, converting the exponential cost of direct transmission into a polynomial one and making a transcontinental quantum link conceivable for the first time. The following year, with Dür, Cirac and Zoller, Briegel worked out the nested-purification bookkeeping in full, giving the first explicit polynomial-resource protocol and fixing the $F>1/2$ purification threshold that all later "first-generation" repeaters inherit. Briegel's reach extends well beyond repeaters: with Robert Raussendorf in 2001 he introduced the one-way (measurement-based) quantum computer built on cluster states, one of the most influential models of quantum computation. But the repeater is the idea that bears his initial most visibly — every discussion of the quantum internet still opens with the BDCZ scheme, and every milestone in memory-assisted entanglement distribution is measured against the polynomial scaling he first showed was attainable.

History

1982

— Wootters, Zurek and Dieks prove the no-cloning theorem, ruling out the amplifiers that classical repeaters depend on

1993

— Bennett and co-workers introduce quantum teleportation, and soon after entanglement purification, supplying the repeater's two key primitives

1998

— Briegel, Dür, Cirac and Zoller propose the quantum repeater with nested swapping and purification

2011

— Sangouard, Simon, de Riedmatten and Gisin review atomic-ensemble repeaters, turning the architecture into concrete device targets

2017

— Pirandola and co-workers prove the repeaterless PLOB secret-key capacity, fixing the benchmark every repeater must beat

Going further

Texts and papers. Briegel, Dür, Cirac & Zoller, "Quantum Repeaters: The Role of Imperfect Local Operations in Quantum Communication," Phys. Rev. Lett. 81, 5932 (1998) · Dür, Briegel, Cirac & Zoller, "Quantum repeaters based on entanglement purification," Phys. Rev. A 59, 169 (1999) · Sangouard, Simon, de Riedmatten & Gisin, "Quantum repeaters based on atomic ensembles and linear optics," Rev. Mod. Phys. 83, 33 (2011) · Muralidharan et al., "Optimal architectures for long distance quantum communication," Sci. Rep. 6, 20463 (2016) · Pirandola, Laurenza, Ottaviani & Banchi, "Fundamental limits of repeaterless quantum communications," Nat. Commun. 8, 15043 (2017) · Nielsen & Chuang, Quantum Computation and Quantum Information, chapters on teleportation and entanglement distillation · Preskill, Lecture Notes on Quantum Computation, Ch. 4 (quantum channels) and Ch. 7 (quantum error correction). The attenuation numbers ( $\alpha=0.2\,\mathrm{dB/km}$ , $L_{\mathrm{att}}\approx22\,\mathrm{km}$ ) and the $e^{-1000/22}\approx2\times10^{-20}$ survival probability are standard telecom values, following directly from the dB-to-fraction conversion in equation (1). The entanglement-swapping projection identity, the Werner-parameter multiplicativity under swapping, and the DEJMPS/recurrence purification threshold $F>1/2$ are well-established results; the $F\to1/4$ floor for two-qubit Werner states is exact. The repeaterless PLOB bound $-\log_2(1-\eta)$ is the proven pure-loss secret-key capacity, and the $\sqrt{\eta}$ midpoint scaling is standard. Key worked values: $0.9467^{8}=0.649$ and the $\sqrt{\eta}$ ratios at $500\,\mathrm{km}$ .

Prerequisites. Entanglement Swapping · Bell-State Measurement · Quantum Teleportation · No-Cloning Theorem in Cryptography · E91 Protocol. Leads to. E91 Protocol · BB84 Protocol · Information Reconciliation and Privacy Amplification · Quantum Bit Error Rate and Sifting.

Ask the tutor: Want to see how the nested purification-and-swapping schedule is optimised to hit a target end-to-end fidelity within a given memory coherence time?

Is a quantum repeater the same thing as a classical signal repeater

No, and the difference is the whole subject. A classical repeater receives a weak, attenuated signal, reads it, and copies it into a fresh strong one — copying is the essence of the operation. For quantum information that route is closed: the no-cloning theorem forbids deterministically copying an unknown state, and any amplifier adds noise that destroys the fragile entanglement. A quantum repeater instead never reads or copies the message. It pre-distributes entanglement over short links, purifies and swaps that entanglement into a long-range shared Bell pair, and then the quantum information is teleported end-to-end using that pair. The repeater's job is to manufacture and deliver entanglement, not to relay a signal.

What exactly does entanglement swapping accomplish in a repeater

Swapping fuses two short entangled links into one longer link without the two end qubits ever meeting. If Alice shares a Bell pair with a midpoint node and the node shares another Bell pair with Bob, a Bell-state measurement on the node's two qubits projects Alice's and Bob's qubits into an entangled state — even though they have never interacted and may be hundreds of kilometres apart. The measurement outcome is two classical bits that tell Alice and Bob which Bell state they now hold, so they apply a Pauli correction. Nesting this in a binary tree lets $\log_2 N$ swap levels join $N$ short links into one continental-scale pair.

Why does the repeater need quantum memories at all

Because entanglement across each elementary link is heralded probabilistically — it succeeds at random, unpredictable moments. To swap two adjacent links you need both to be ready at the same time, but if each succeeds independently with small probability, the chance that all links are simultaneously ready is vanishingly small — exactly the exponential catastrophe the repeater is trying to escape. A quantum memory breaks the deadlock: once a link succeeds, its entangled qubit is stored and held while the neighbouring links keep retrying. Only when both partners of a swap are in memory does the swap fire. The memory's coherence time must exceed the average wait, which is why memory quality is the central bottleneck of first-generation repeaters.

Why is there a fidelity threshold of one-half for purification

Recurrence purification takes two imperfect Werner pairs and, on a successful bilateral-CNOT-and-measurement, outputs one pair whose fidelity $F'$ obeys the update in equation (5); solving $F'>F$ shows the improvement holds only when $F>1/2$ . The intuition: purification amplifies whatever majority component the pairs already have. Above $F=1/2$ the correct Bell component is the majority, and distillation reinforces it toward $F\to1$ ; below $1/2$ the noise is the majority, and the same operation reinforces the noise instead. That is why a repeater must purify before fidelity crosses the threshold, and why swapping — which always lowers fidelity — cannot run unchecked between purification rounds.

What is the PLOB bound and why does it matter

The PLOB bound (Pirandola–Laurenza–Ottaviani–Banchi, 2017) is the proven maximum secret-key rate of a pure-loss optical channel with no repeater: $R=-\log_2(1-\eta)\approx1.44\,\eta$ bits per channel use, where $\eta$ is the end-to-end transmittance. It matters because it draws a sharp line: any device that produces a key rate above this bound at a given distance is genuinely doing something a repeater does — extracting more than loss allows — while a device below it is just a relay that could be replaced by better fibre. Because $\eta$ falls exponentially with distance, so does the bound, and repeaters break it by achieving $\sqrt{\eta}$ or better scaling. Beating PLOB is now the standard experimental milestone for claiming true repeater advantage.

How do second- and third-generation repeaters differ from the original BDCZ scheme

The generations are classified by which errors they fight with two-way versus one-way methods. First-generation (BDCZ) repeaters use entanglement purification and heralding for both loss and operation errors, which demands slow two-way classical communication and long-lived memories. Second-generation repeaters replace purification with quantum error correction on encoded logical qubits to handle operation errors one-way, but still herald loss, so memories remain. Third-generation (one-way or all-photonic) repeaters use loss-tolerant quantum codes to correct both loss and operation errors on the fly, eliminating two-way signalling and long-lived memory entirely — at the cost of very large numbers of high-quality qubits or photons per node. Which generation is best depends on the scarcest resource: memory coherence, gate fidelity, or qubit count.