Quantum Networks and Entanglement Distribution

A quantum network is the physical and protocol machinery that manufactures, stores, purifies, and routes entanglement between distant nodes — and it is built under a constraint no classical network faces. A classical repeater reads a weak bit, copies it, and retransmits a fresh strong one; a quantum network may do none of that, because the no-cloning theorem forbids duplicating an unknown state. A photon that fades in fibre cannot be reamplified without destroying the very superposition it carries. So the central resource is not the signal but the correlation: a shared Bell pair, most often $|\Phi^+\rangle=(|00\rangle+|11\rangle)/\sqrt{2}$ . Once two nodes hold one high-fidelity pair, any qubit can be moved between them by teleportation — local operations plus two classical bits, nothing quantum on the wire. The enemy is exponential loss. In telecom fibre the probability that a photon survives a length $L$ falls as $e^{-L/L_{\mathrm{att}}}$ with $L_{\mathrm{att}}\approx 22$ km at 1550 nm; over 1000 km that is below $10^{-19}$ , and cloning being outlawed, no amplifier can rescue it. The quantum repeater sidesteps the exponential by refusing to fight it head-on: cut the span into short elementary links that each succeed with decent probability, herald each success into a quantum memory, weld neighbours together by entanglement swapping, and iron out the accumulated noise by distillation. Three primitives — swap, distill, store — turn an exponential wall into a polynomial cost. This article assembles those primitives into one working theory of entanglement distribution, the currency that binds distributed computers, quantum key distribution, and networked sensors into a single fabric: the quantum internet.

Key takeaways

A quantum network distributes entanglement — Bell pairs — between remote nodes; teleportation then moves arbitrary qubits over that shared resource using only local operations and classical communication (LOCC).
Photon loss in fibre is exponential, $e^{-L/L_{\mathrm{att}}}$ with $L_{\mathrm{att}}\approx 22$ km, and no-cloning bans amplification, so direct transmission dies over long distances.
Entanglement swapping fuses two short pairs $A$ – $B$ and $C$ – $D$ into one long $A$ – $D$ pair by a Bell-state measurement at the relay — $A$ and $D$ never interact.
Distillation (purification) burns many noisy pairs to yield fewer purer ones; recurrence purification works if and only if the input fidelity exceeds $1/2$ .
Quantum memories store a heralded link while its neighbour is still being built, decoupling the segment success probabilities — this is what converts $p_0^{-N}$ into a polynomial cost.
The repeaterless PLOB bound $C=-\log_2(1-\eta)$ caps any memoryless scheme; beating it certifies a genuine repeater.
Key idea: entanglement is a distributable, transferable resource — teleportation sends any qubit through a shared Bell pair with only local operations and classical communication, and swapping plus distillation stitch short noisy links into long high-fidelity ones, beating the exponential loss that dooms direct transmission.

The core insight

Entanglement is a distributable, transferable resource: by teleportation any qubit can be sent through a shared Bell pair using only local operations and classical communication, and by swapping plus distillation short noisy links can be combined into long high-fidelity ones, beating the exponential loss of direct transmission.

Where this lives today

The primitives on this page are being welded into real hardware. In 2021 a Delft group operated a three-node entanglement-based network — two end nodes linked through a central node acting as a repeater relay — using nitrogen-vacancy centres in diamond as memory qubits and photons as messengers, delivering heralded, on-demand entanglement between nodes that share no direct link. Cold-atom and trapped-ion nodes have pushed memory coherence times into the second-to-minute range, long enough to clear the classical round-trip for tens of kilometres of fibre. On the theory-benchmark side, several groups have now demonstrated memory-assisted or twin-field protocols that surpass the PLOB repeaterless key-rate bound over hundreds of kilometres, the operational certificate that a genuine repeater advantage has been reached. China's Micius satellite distributed entangled photon pairs to ground stations more than 1000 km apart, sidestepping fibre loss with a free-space vacuum channel and demonstrating entanglement-based key distribution at continental scale. The open frontier is the fault-tolerant "third-generation" repeater — encoding each link in a quantum error-correcting code so that even one-way transmission, with no waiting for classical heralds, can beat loss and gate errors together. That is where quantum networking merges with the rest of this tier: the same teleported gates and encoded links reappear in measurement-based computation, magic-state distillation, and topological memories.

The physical picture

Picture entanglement as a perfectly correlated quantum wire you must lay one segment at a time. You cannot stretch a single segment across the whole distance, because photons vanish in the fibre; instead you build a short segment, latch it into a memory the instant it is heralded good, and then weld it to its neighbour. The weld is entanglement swapping: a Bell-state measurement at the junction node collapses two short pairs into one long pair, exactly as if the wire had been continuous. Repeat the weld up a nested hierarchy and the wire spans a continent. The first figure lays out this architecture and puts a real number on why a single segment is already hard.

Figure

A three-link quantum repeater: nodes A and D are joined through relays R₁, R₂. Each 100 km elementary link has photon survival e^(−100/22) ≈ 1.0×10⁻² so it must be heralded and stored in a memory; Bell-state measurements at the relays swap the three short pairs into one A–D pair spanning 300 km, and the two-bit outcomes are broadcast classically so the ends can apply the Pauli correction

Why can we not simply amplify, the way an erbium-doped fibre amplifier boosts a fading laser pulse? Because amplification is copying, and no-cloning forbids copying an unknown quantum state. The only legal moves on the network are local unitaries, local measurements, and classical messages — LOCC. Remarkably, that is enough: teleportation proves a shared Bell pair plus two classical bits transmits a qubit exactly, so the network's entire job reduces to manufacturing good Bell pairs at a distance. But swapping erodes fidelity and memories decohere, so the deep payoff is quantitative, not just conceptual. The second figure is the punchline every network engineer stares at: secret-key rate versus distance, direct fibre against a repeater.

Figure

Secret-key rate versus distance on a log scale. Direct fibre obeys the repeaterless PLOB bound C = −log₂(1−η) with η = e^(−L/22 km), so for small η the rate ≈ 1.44 η falls exponentially — a straight line dropping about one decade every 11 km. A quantum repeater pays a fixed overhead but scales polynomially, so it overtakes direct transmission near 180 km and by 1000 km leads by roughly fourteen orders of magnitude

Distillation deserves one more sentence of intuition, because it is where the fidelity is bought back. It is error correction applied to a resource you never look at directly. Two parties each hold several imperfect pairs; they run local CNOTs and measure the sacrificial pairs, learning only parity information about the survivors. When the parities agree — a heralded success — the surviving pairs are statistically purer. You pay in yield (many noisy pairs in, few clean pairs out) but you buy back the fidelity that every swap keeps eroding. The whole discipline of repeater design is interleaving swaps that extend distance with distillations that restore quality, level by level up the nesting tree.

The core equations

The network's currency is a single maximally entangled two-qubit state. Every protocol below either produces this state at a distance, consumes it, or measures how far a noisy laboratory state falls short of it.

|\Phi^+\rangle=\frac{1}{\sqrt{2}}\big(|00\rangle+|11\rangle\big) \qquad (1)

where: $|\Phi^+\rangle$ — maximally entangled ("Bell") state shared by two nodes, one ebit of entanglement; $|00\rangle,|11\rangle$ — computational-basis product states of the two qubits. Reading: measure either qubit and the other instantly agrees, in every basis — the perfect correlation that teleportation cashes in. Teleportation is the reason a Bell pair is worth having. A sender holding an unknown qubit $|\psi\rangle$ and one half of a shared pair performs a joint Bell measurement; the algebra rewrites the three-qubit state so the receiver's qubit is already $|\psi\rangle$ up to a Pauli fixed by the outcome.

|\psi\rangle_C\,|\Phi^+\rangle_{AB}=\frac12\sum_{j=0}^{3}|\Phi_j\rangle_{CA}\,\big(\sigma_j|\psi\rangle_B\big) \qquad (2)

where: $|\psi\rangle_C$ — unknown qubit at the sender; $|\Phi_j\rangle_{CA}$ — the four Bell states of the measured qubits; $j$ — two-bit measurement outcome, broadcast classically; $\sigma_j\in\{I,X,Y,Z\}$ — Pauli correction the receiver applies. Reading: each outcome occurs with probability $1/4$ ; after the correction $B$ holds $|\psi\rangle$ exactly, with no quantum information ever crossing the wire. Entanglement swapping is teleportation with an entangled state as the "unknown" input: teleporting node $B$ 's half of the $A$ – $B$ pair through the $C$ – $D$ pair leaves $A$ entangled with $D$ . A Bell-state measurement (BSM) at the relay does it.

\Phi^+_{AB}\otimes\Phi^+_{CD}\;\xrightarrow{\ \text{BSM on }B,C\ }\;\Phi^{(j)}_{AD} \qquad (3)

where: $\Phi^+_{AB},\Phi^+_{CD}$ — independent pairs on adjacent links; BSM — Bell-state measurement at the relay on its two local qubits $B,C$ ; $\Phi^{(j)}_{AD}$ — the resulting pair across the joined span, equal to $|\Phi^+\rangle$ up to a Pauli set by outcome $j$ . Reading: $A$ and $D$ end up entangled although they never met — the relay's entanglement was swapped onto the long link. The obstacle all this fights is loss, which in fibre is exponential in distance. The survival probability sets both the raw entangling rate and, through it, everything downstream.

P_{\mathrm{survive}}(L)=e^{-L/L_{\mathrm{att}}},\qquad L_{\mathrm{att}}=\frac{10\log_{10}e}{\alpha_{\mathrm{dB}}}\approx 22\ \mathrm{km} \qquad (4)

where: $L$ — fibre length; $L_{\mathrm{att}}$ — attenuation length; $\alpha_{\mathrm{dB}}\approx0.2$ dB/km — loss at the 1550 nm telecom window. Reading: with $10\log_{10}e=4.34$ , $L_{\mathrm{att}}=4.34/0.2\approx22$ km, so over 1000 km the survival probability is $e^{-1000/22}\approx10^{-20}$ — hopeless for direct transmission. Real distributed pairs are noisy, and the standard isotropic model is the Werner state: a Bell pair mixed with white noise, parametrized by a single fidelity.

\rho_W=F\,|\Phi^+\rangle\langle\Phi^+|+\frac{1-F}{3}\big(\mathbb{1}-|\Phi^+\rangle\langle\Phi^+|\big) \qquad (5)

where: $\rho_W$ — noisy two-qubit state; $F=\langle\Phi^+|\rho_W|\Phi^+\rangle$ — fidelity with the target Bell state; the remaining weight $1-F$ is split equally over the other three Bell states. Reading: $F=1$ is a perfect pair, $F=1/4$ is the maximally mixed state, and — as the next equation shows — only $F>1/2$ is distillable. Recurrence purification takes two Werner pairs of fidelity $F$ , applies a bilateral CNOT, measures the target pair, and keeps the survivor when parities agree. One round maps $F$ to a higher $F'$ whenever $F>1/2$ .

F'=\frac{F^2+\tfrac19(1-F)^2}{F^2+\tfrac23 F(1-F)+\tfrac59(1-F)^2} \qquad (6)

where: $F$ — input fidelity of each pair; $F'$ — output fidelity of the surviving pair on a heralded success; the denominator is the success probability $p_{\mathrm{succ}}$ . Reading: the map has a fixed point at $F=1/2$ ; above it iteration flows to $F\to1$ , below it collapses to $F\to1/4$ , which is exactly the distillability threshold. No memoryless scheme can beat a fundamental capacity: the repeaterless PLOB bound is the maximum two-way entanglement (or secret-key) rate of a pure-loss channel of transmissivity $\eta$ .

C_{\mathrm{direct}}(\eta)=-\log_2(1-\eta)\ \ \text{ebits per channel use} \qquad (7)

where: $\eta=e^{-L/L_{\mathrm{att}}}$ — channel transmissivity; $C_{\mathrm{direct}}$ — the ultimate rate without a repeater. Reading: for small $\eta$ , $-\log_2(1-\eta)\approx\eta/\ln2\approx1.44\,\eta$ , so the rate inherits the fibre's exponential decay — beating this line is the operational definition of a working repeater. Finally, how much pure entanglement a given noisy state actually contains is bounded below by the coherent information — the hashing bound, achievable by a one-way LOCC protocol.

E_D(\rho)\ge S(\rho_A)-S(\rho_{AB})\equiv I_{\mathrm{coh}}(A\rangle B) \qquad (8)

where: $E_D$ — distillable entanglement, ebits per copy extractable by LOCC; $S(\cdot)$ — von Neumann entropy; $I_{\mathrm{coh}}$ — coherent information. Reading: a positive coherent information guarantees distillable entanglement; the bound is tight for many channels and sets the yield ceiling a repeater segment can hope for.

Derivation AEntanglement swapping fuses two links into one

The goal is to prove the single line that makes long-distance entanglement possible: that projecting the two inner qubits of two independent Bell pairs onto a Bell state leaves the two outer qubits — which have never interacted and may be hundreds of kilometres apart — in a Bell state of their own. We start from the product of two maximally entangled pairs, expand in the computational basis, apply the Bell projector at the relay, and read off the outer state and its Pauli correction, so that "swapping" is exposed as nothing more than teleporting one half of a pair through the other.

Step 1. Write the two independent pairs, with $A$ far to the left, relay qubits $B$ and $C$ co-located, and $D$ far to the right: $|\Psi\rangle=|\Phi^+\rangle_{AB}\otimes|\Phi^+\rangle_{CD}$ . Why: the relay physically holds $B$ and $C$ , so it is the one place a joint measurement on the inner qubits can be performed.

Step 2. Expand both pairs in the computational basis, $|\Psi\rangle=\tfrac12\sum_{a,d}|a\rangle_A|a\rangle_B|d\rangle_C|d\rangle_D$ with $a,d\in\{0,1\}$ . Why: writing everything in the same basis makes the action of the Bell projector on $B,C$ a matter of picking out surviving terms.

Step 3. Apply the projector for the $|\Phi^+\rangle$ outcome, $\langle\Phi^+|_{BC}=\tfrac{1}{\sqrt2}(\langle00|+\langle11|)_{BC}$ . Only terms with $a=d$ survive, giving $\tfrac12\cdot\tfrac{1}{\sqrt2}\sum_a|a\rangle_A|a\rangle_D=\tfrac12|\Phi^+\rangle_{AD}$ . Why: the projector keeps exactly the correlated terms, welding $A$ 's label to $D$ 's and discarding the rest.

Step 4. Handle the other three outcomes. Projecting onto $|\Phi^-\rangle,|\Psi^+\rangle,|\Psi^-\rangle$ yields $|\Phi^-\rangle_{AD},|\Psi^+\rangle_{AD},|\Psi^-\rangle_{AD}$ respectively; each amplitude is $\tfrac12$ , so each occurs with probability $1/4$ and is undone by a known Pauli on $D$ once the two-bit outcome is broadcast. Why: the four outcomes partition the probability evenly, and a deterministic Pauli always maps them back to $|\Phi^+\rangle$ , so the swap succeeds every time — no post-selection loss.

Step 5. Conclude that $A$ and $D$ now share a Bell pair though they never interacted; the relay converted its local entanglement into a longer link. Why: this is the inductive step of every repeater — a swap at each junction doubles the reachable distance while consuming one intermediate pair. *Verified: $\langle\Phi^+|_{BC}(|\Phi^+\rangle_{AB}\otimes|\Phi^+\rangle_{CD})=\tfrac12|\Phi^+\rangle_{AD}$ ; the four outcome probabilities sum to $4\times(1/2)^2=1$ .*

Go deeper · Derivation B · 2 minRecurrence purification raises fidelity above the one-half threshold

Derivation BRecurrence purification raises fidelity above the one-half threshold

The goal is to derive the fidelity-recurrence map for a single round of bilateral-CNOT distillation and to locate its fixed points, showing precisely why a noisy pair can be purified toward a perfect Bell state if and only if its fidelity already exceeds one half. We model the input as a Werner state, track how the bilateral CNOT permutes the Bell-state populations, post-select on coincident parity, renormalize the surviving weight, and then solve $F'=F$ to find where the flow reverses — the operational entanglement threshold of the whole repeater programme.

Step 1. Parametrize each input pair as a Werner state: weight $F$ on $|\Phi^+\rangle$ and $(1-F)/3$ on each of $|\Phi^-\rangle,|\Psi^+\rangle,|\Psi^-\rangle$ . Why: the Werner form is the worst-case isotropic noise, and a single number $F$ then controls the entire round, making the recurrence closed.

Step 2. Apply the bilateral CNOT — both parties use their kept pair as control and their sacrificial pair as target — and measure the target pair in the computational basis. The CNOTs permute the joint Bell labels so that the target's measured parity reports on the correlated-versus-anticorrelated character of the two inputs. Why: the measurement extracts parity information about the survivor without ever measuring the survivor itself, the essence of purifying a resource you must not collapse.

Step 3. Keep only the coincident (even-parity) outcomes. Their total probability is the success probability $p_{\mathrm{succ}}=F^2+\tfrac23F(1-F)+\tfrac59(1-F)^2$ , and the surviving $|\Phi^+\rangle$ weight is $F^2+\tfrac19(1-F)^2$ . Why: post-selection is the herald — a classical flag that the round worked — and dividing the good weight by $p_{\mathrm{succ}}$ renormalizes the kept state.

Step 4. Form the ratio $F'=\big[F^2+\tfrac19(1-F)^2\big]/p_{\mathrm{succ}}$ and solve $F'=F$ . The relevant fixed points are $F=1/2$ (unstable) and $F=1$ (stable), with $F=1/4$ the sink below threshold. Why: the sign of $F'-F$ across $F=1/2$ is what separates pairs that can be cleaned up from pairs that only get worse — the distillability line.

Step 5. Read off the cost. The yield per round is at most $p_{\mathrm{succ}}/2$ , so reaching $F\approx1$ consumes many raw pairs; in practice a few rounds are interleaved with each swap to hold $F$ above threshold at every nesting level. Why: distillation buys fidelity in units of yield, and the repeater's total resource budget is the running product of these per-level costs. *Verified: at $F=1$ , $F'=1$ ; at $F=1/2$ , numerator $=1/4+1/36=10/36$ equals $p_{\mathrm{succ}}=1/4+1/6+5/36=10/36$ , so $F'=1/2$ ; numerically $F=0.7\to F'=0.735$ with $p_{\mathrm{succ}}=0.68$ .*

Go deeper · Derivation C · 2 minWhy heralded memories give polynomial scaling

Derivation CWhy heralded memories give polynomial scaling

The goal is to explain the repeater's decisive advantage quantitatively: to show that without quantum memories the $N$ elementary links must all succeed in the same instant, giving an exponential $p_0^{-N}$ cost identical to direct transmission, whereas storing each heralded link decouples the trials and collapses the cost to a polynomial in $N$ . We treat each link as a Bernoulli trial, compare the joint-success and stored-success waiting times, build the nested-doubling hierarchy, and flag the coherence condition that makes it all hold.

Step 1. Without memory, all $N=L/L_0$ links must be entangled simultaneously, so the joint probability is $P_{\mathrm{all}}=p_0^{N}$ with $p_0=e^{-L_0/L_{\mathrm{att}}}$ , and the expected number of attempts is $\langle T_{\mathrm{no\,mem}}\rangle\sim p_0^{-N}=e^{(L/L_0)(L_0/L_{\mathrm{att}})}=e^{L/L_{\mathrm{att}}}$ . Why: forcing simultaneity multiplies the per-link probabilities, reconstructing exactly the direct-transmission exponential — no advantage at all.

Step 2. With memory, each link is retried independently and its heralded success is held while neighbours catch up; the bottleneck is only the slowest of $N$ links. For a two-link segment the exact expected time is $\langle T_2\rangle=\tfrac{3-2p_0}{2-p_0}\,p_0^{-1}$ , which for small $p_0$ is $\approx\tfrac{3}{2}p_0^{-1}$ — not $p_0^{-2}$ . Why: storage turns a product of failure probabilities into a maximum of independent waiting times, the single structural change that defeats the exponential.

Step 3. Nest the construction by doubling, $L=2^n L_0$ : swap $2^n$ elementary links in $n$ levels, distilling at each level to hold fidelity above threshold. The total resource — qubits and time — then grows only polynomially in $L$ , the hallmark of the Briegel–Dür–Cirac–Zoller repeater. Why: each doubling level multiplies the cost by a bounded factor rather than by $p_0^{-1}$ , so $n\sim\log_2(L/L_0)$ levels give a polynomial rather than exponential total.

Step 4. Impose the coherence condition: the memory time $T_2$ must exceed the classical round-trip $2L_0/c$ needed to herald and signal each success, or the stored fidelity decays toward $1/4$ before its neighbour arrives and the exponential returns. Why: the whole advantage is borrowed against memory coherence, which is why long-lived quantum memories are the experimental linchpin of every repeater. *Verified: with $p_0=e^{-100/22}\approx0.010$ , the two-link time $\langle T_2\rangle=(3-2p_0)/[(2-p_0)p_0]\approx150$ against the memoryless $p_0^{-2}\approx10^{4}$ — a sixty-fold saving already at two links.*

Limiting cases

Repeater performance is a contest of competing exponentials — in loss, in fidelity, and in memory time. The regimes that matter:

Case	Result	Meaning
No repeater (direct fibre)	$P_{\mathrm{survive}}(L)=e^{-L/L_{\mathrm{att}}},\ L_{\mathrm{att}}\approx22$ km	Rate falls exponentially; over 1000 km the survival probability is $\approx10^{-20}$ , ruling out direct transmission.
Heralded repeater, ideal memory	$R_{\mathrm{rep}}\sim\mathrm{poly}(L)\cdot e^{-L_0/L_{\mathrm{att}}}$	With $N=L/L_0$ segments and perfect memories only the short length $L_0$ enters the exponent; total cost is polynomial in $L$ .
Repeaterless bound (PLOB)	$C_{\mathrm{direct}}\le-\log_2(1-\eta),\ \eta=e^{-L/L_{\mathrm{att}}}$	No memoryless scheme beats this two-way capacity; for small $\eta$ , $C\approx1.44\,\eta$ , linear in transmissivity.
Finite memory coherence	$F(t)\to\tfrac14$ as $t\gg T_2$	If a memory decoheres faster than neighbours are built, its fidelity decays to the mixed value and the advantage vanishes.
Swap-limited fidelity	$F_{\mathrm{end}}\approx F^{\,N}$ (schematic, depolarizing)	Each swap compounds error; without interleaved distillation, end-to-end fidelity degrades with the number of nested levels $N$ .
Distillation threshold	purify iff $F>1/2$	Below the recurrence fixed point noisy pairs flow to $F\to1/4$ ; only $F>1/2$ can be driven toward $1$ .

Common trapA quantum repeater amplifies the signal the way a classical repeater does

It does not, and it cannot. No-cloning forbids copying an unknown quantum state, so there is no quantum analogue of the classical "read, amplify, retransmit" loop. A repeater works only because each short elementary link is heralded — a classical flag that a good Bell pair now exists — and then stored in a quantum memory while its neighbour is built. Strip out the heralding and the storage and every segment must succeed at the same instant; the joint probability $p_0^{N}$ multiplies straight back into the direct-transmission exponential. Two further traps ride along. First, swapping degrades fidelity, so distillation must be interleaved at every nesting level; ignoring this makes naive end-to-end fidelity estimates wildly optimistic. Second, the advantage is borrowed against memory coherence: if $T_2$ is shorter than the classical round-trip $2L_0/c$ , the stored pair rots to the maximally mixed state before it can be used, and all the polynomial scaling evaporates.

Worked examples

1 · Direct fibre versus a repeater over 1000 km (fully worked)

A telecom-fibre link at 1550 nm has loss $0.2$ dB/km ( $L_{\mathrm{att}}\approx22$ km). Compare the entangling rate of direct transmission over 1000 km with a repeater built from ten 100 km elementary links.

Step 1. Direct survival. $P_{\mathrm{survive}}(1000)=e^{-1000/22}=e^{-45.5}$ . Numerically this is $\approx1.6\times10^{-20}$ ; equivalently the loss budget is $0.2\times1000=200$ dB, and $10^{-200/10}=10^{-20}$ . Reading either way, fewer than one photon in $10^{19}$ survives — at a gigahertz source rate that is one photon every few centuries.

Step 2. The PLOB ceiling. With $\eta=P_{\mathrm{survive}}$ the best possible memoryless rate is $C=-\log_2(1-\eta)\approx1.44\,\eta\approx1.4\times10^{-20}$ ebits per channel use. No repeaterless trick — not squeezing, not clever coding — can raise this; it is a fundamental capacity.

Step 3. Segment the span. Each 100 km link has $p_0=e^{-100/22}\approx1.0\times10^{-2}$ , roughly a 1% survival per attempt — vastly more workable than $10^{-20}$ . The exponential now involves only $L_0=100$ km, not the full 1000 km.

Step 4. The memory payoff. Without memories the ten links would need to fire together, cost $\sim p_0^{-10}\approx10^{20}$ — no better than direct. With memories the cost grows only polynomially; already a two-link segment costs $\langle T_2\rangle\approx150$ attempts against the memoryless $p_0^{-2}\approx10^4$ . Heralded storage is the whole game: it turns $10^{20}$ into a polynomial.

Step 5. Verdict. Direct transmission is astronomically hopeless ( $\sim10^{-20}$ ), the PLOB bound proves no memoryless scheme escapes it, and a repeater with $\sim100$ km segments plus memories restores a usable rate. Direct: $\approx10^{-20}$ ; repeater: polynomial in $L$ — the crossover sits near 180 km.

2 · One round of entanglement distillation (fully worked)

Two nodes share Werner pairs of fidelity $F=0.7$ . Apply one recurrence-purification round. Find the output fidelity, the success probability, and confirm the direction of the flow.

Step 1. Success probability (the denominator of the map): $p_{\mathrm{succ}}=F^2+\tfrac23F(1-F)+\tfrac59(1-F)^2$ . With $F=0.7$ : $0.49+\tfrac23(0.7)(0.3)+\tfrac59(0.3)^2=0.49+0.14+0.05=\mathbf{0.68}$ . So a heralded success comes about 68% of the time.

Step 2. Surviving good weight (the numerator): $F^2+\tfrac19(1-F)^2=0.49+\tfrac19(0.09)=0.49+0.01=0.50$ .

Step 3. Output fidelity: $F'=0.50/0.68=\mathbf{0.735}$ . The pair improved from $0.700$ to $0.735$ in one round — consistent with $F>1/2$ driving the flow upward.

Step 4. Iterate to see the flow. Feeding $F'=0.735$ back gives $F''\approx0.77$ , then $\approx0.82$ , converging toward $1$ ; had we started at $F=0.4<1/2$ the same map would have driven the fidelity down toward $0.25$ . As a metro-scale reference, a 50 km link has $\eta=e^{-50/22}\approx0.10$ and PLOB rate $C=-\log_2(1-0.10)\approx0.16$ ebits/use — about six channel uses per delivered secret bit. One round: $F=0.70\to F'=0.735$ at $p_{\mathrm{succ}}=0.68$ ; iteration flows to $F\to1$ because $0.70>1/2$ .

Physics of the portrait

Entanglement distribution is a young idea with an old lineage: Einstein's "spooky action" of 1935 became, sixty years later, an engineering resource to be piped between buildings. The portrait belongs to the physicist who named the goal and built the first nodes toward it.

Origins · 1 minH. Jeff Kimble and how this was discovered

H. Jeff Kimble

b. 1949, American physicist · cavity QED and the quantum internet

H. Jeff Kimble is the experimentalist who turned "quantum network" from a metaphor into a research programme. Working at Caltech from the 1980s onward, he pioneered cavity quantum electrodynamics — trapping single atoms in high-finesse optical cavities so that one atom and one photon interact strongly and reversibly, the physical basis of a network node that can hold, emit, and absorb a flying qubit on demand. His group performed landmark demonstrations of the strong-coupling regime, single-photon sources, and atom–photon entanglement, showing that a stationary matter qubit could be entangled with a photon sent down a fibre — precisely the memory-plus-messenger primitive a repeater needs. In 2008 he crystallized the field's ambition in a widely cited Nature article, "The quantum internet," which framed distributed entanglement as a universal resource and set out the node-and-channel architecture — quantum memories at the nodes, photons on the links, entanglement as the currency — that organizes the subject to this day. Kimble's arc mirrors the physics itself: from the microscopic reversibility of a single atom coupled to a single mode, up through atom–photon entanglement, to the vision of a continental web of entangled nodes. The theory of swapping, distillation, and repeaters that fills this article is the protocol layer that rides on the hardware layer he did so much to build.

History

1935

— Einstein, Podolsky, Rosen frame entanglement; Schrödinger names it

1993

— Bennett and co-workers give the quantum teleportation protocol

1996

— Bennett–DiVincenzo–Smolin–Wootters and Deutsch et al. formalize entanglement distillation

1998

— Briegel, Dür, Cirac, Zoller propose the quantum repeater

2008

— Kimble lays out the quantum-internet architecture

2017

— Pirandola, Laurenza, Ottaviani, Banchi prove the repeaterless PLOB bound

Going further

Texts. Nielsen & Chuang, Quantum Computation and Quantum Information, ch. 1, 10, 12 (teleportation, distillation, capacities) · Preskill, Lecture Notes on Quantum Computation, ch. 4 (quantum channels and information) · Wilde, Quantum Information Theory, ch. 19–24 (distillation and channel capacities) · Watrous, The Theory of Quantum Information (LOCC and entanglement transformations) · Kitaev, Shen, Vyalyi, Classical and Quantum Computation (measurement and gate primitives).

Papers. Bennett, Brassard, Crépeau, Jozsa, Peres, Wootters, Phys. Rev. Lett. 70, 1895 (1993) — teleportation · Bennett, DiVincenzo, Smolin, Wootters, Phys. Rev. A 54, 3824 (1996) and Deutsch et al., Phys. Rev. Lett. 77, 2818 (1996) — distillation · Briegel, Dür, Cirac, Zoller, Phys. Rev. Lett. 81, 5932 (1998) — quantum repeaters · Kimble, Nature 453, 1023 (2008) — the quantum internet · Pirandola, Laurenza, Ottaviani, Banchi, Nat. Commun. 8, 15043 (2017) — PLOB bound.

Provenance. Teleportation corrections (each of four Paulis, probability $1/4$ ): Nielsen & Chuang §1.3.7 and Bennett et al. 1993 — verify. Swapping identity $\langle\Phi^+|_{BC}(|\Phi^+\rangle_{AB}|\Phi^+\rangle_{CD})=\tfrac12|\Phi^+\rangle_{AD}$ : direct computation. Fibre $L_{\mathrm{att}}=22$ km from $\alpha=0.2$ dB/km via $10\log_{10}e/\alpha=4.34/0.2$ — recompute. Recurrence map and $F=1/2$ fixed point: set $F'=F$ (checked numerically: $F=0.7\to0.735$ , $p_{\mathrm{succ}}=0.68$ ). PLOB $-\log_2(1-\eta)$ and small- $\eta$ limit $1.44\eta$ : Pirandola et al. 2017. Hashing bound: Wilde ch. 21–24.

Prerequisites. Bell States and Teleportation · LOCC and Entanglement Measures · Cluster States and Graph States · Quantum Error Correction (stabilizer codes). Leads to. Measurement-Based Quantum Computation · Magic State Distillation · Toric Code and Topological Order.

Ask the tutor: Ask me to walk through a nested repeater's rate-versus-distance budget, or to show step by step why a Bell measurement at the relay swaps entanglement onto qubits that never met.

If no-cloning forbids copying, how does teleportation "send" a qubit without violating it?

Teleportation never copies. The Bell measurement at the sender destroys the original qubit — after it, the sender's copy is gone, collapsed into a random Bell outcome that carries none of the state's information by itself. Only after the two classical bits arrive and the receiver applies the matching Pauli does the state reappear, now solely at the receiver. At no instant do two copies coexist, so no-cloning is respected. What crosses the channel is two classical bits and the pre-shared entanglement, never the quantum amplitudes themselves — which is also why teleportation cannot signal faster than light: without the classical bits the receiver's qubit is maximally mixed.

What exactly is heralding, and why is it the linchpin of a repeater?

Heralding is a classical signal that says "a good entangled pair now exists here," produced by a measurement whose outcome flags success without revealing or destroying the delivered state. In a typical scheme two photons, one from each node, interfere at a beamsplitter and a detector click announces that the two remote memories are now entangled. The power of heralding is that it lets you retry a link until it works and only then commit it to memory. Because each link is retried and stored independently, the segments no longer need to succeed simultaneously — the joint probability $p_0^{N}$ becomes a maximum of independent waiting times. Remove heralding and storage and the exponential scaling of direct transmission comes straight back.

Why can distillation only improve pairs with fidelity above one half?

Because the recurrence map has an unstable fixed point at $F=1/2$ . Above it, one round of bilateral-CNOT purification returns a higher fidelity, and iterating flows to $F\to1$ ; below it, the same round returns a lower fidelity and the flow collapses to the maximally mixed value $F=1/4$ . The deeper reason is that $F>1/2$ is the point at which a Werner state becomes entangled and, for two qubits, distillable: a state at or below $1/2$ is separable or bound in a sense that no LOCC protocol can concentrate. So $F=1/2$ is not an artifact of this particular protocol but the genuine distillability threshold for the isotropic noise model.

How does a repeater turn an exponential cost into a polynomial one?

The exponential comes from demanding that all $N$ elementary links succeed at once, which multiplies their probabilities to $p_0^{N}$ . A repeater breaks that demand in two ways. First, quantum memories let each link be attempted independently and stored the moment it is heralded, so the cost is set by the slowest single link, not the product of all of them. Second, a nested-doubling hierarchy joins $2^n$ links in $n\sim\log(L/L_0)$ swap levels, and distillation at each level keeps fidelity above threshold. Each level multiplies the cost by a bounded factor rather than by $p_0^{-1}$ , so a logarithmic number of levels yields a total resource that is polynomial in the distance $L$ .

What does it take to actually beat the PLOB bound in the lab?

You must demonstrate an end-to-end entanglement or secret-key rate over a lossy channel that exceeds $-\log_2(1-\eta)$ for the channel's transmissivity $\eta$ . Since that bound falls exponentially with distance, at short range it is easy to stay below and hard to beat — direct transmission is already near-optimal there. The bound bites at long range, where you need a genuine memory-assisted middle node: photons interfere at a central station, a heralded event entangles two memories, and the stored entanglement is later swapped out. Recent twin-field QKD and memory-assisted demonstrations have crossed the bound over hundreds of kilometres, which is the community's accepted certificate that a repeater segment — not merely a better point-to-point link — is working.

Where do the three generations of quantum repeaters differ?

They differ in how they fight the two enemies, loss and operation errors. First-generation repeaters use heralded entanglement generation, memories, two-way classical signaling for entanglement swapping, and recurrence distillation to fight both — simple but slow, because every level waits for a classical round-trip. Second-generation repeaters keep memories but replace distillation with quantum error correction against operation errors, still using two-way signaling for loss. Third-generation repeaters encode each link in a loss-tolerant quantum code and transmit one-way, with no waiting for heralds at all, so the rate is limited only by local gate speed — at the price of many physical qubits per link. The progression trades memory time and classical latency for qubit overhead.