Distribution of entanglement in large-scale quantum networks

1. Introduction

• Quantum entanglement supports cryptographic, communication, and computation protocols, yet quantum information cannot be copied perfectly, complicating its distribution.Entangled correlations can exceed those possible in classical theories.Source: §1 · p. 4
• Remote entanglement is difficult to create: photon transmission through optical fibres falls exponentially with distance because of noise, scattering, and absorption.Transferring photon states to quantum memories for manipulation and storage is another experimental challenge.Source: §1 · p. 6
• One-dimensional quantum repeaters distribute entanglement over long distances, but reasonable communication rates require entanglement per link to increase with distance, beyond present technologies.Repeaters use entanglement purification and swapping.Source: §1 · p. 6
• Percolation protocols can distribute entanglement over infinite distances for pure states and, under specific noise conditions, some mixed states.With realistic noisy operations, network-based error correction instead uses information from all paths between stations and requires no quantum memory.Source: §1 · pp. 6–7
• The review extends entanglement-distribution work from regular lattices to complex networks, where communication quality depends on topology and available quantum operations.This extension reflects the complex structure of present communication networks.Source: §1 · p. 7

2. Concepts and methods

• The section provides pedagogical background on entanglement and network operations, and directs familiar readers to the later discussions of quantum repeaters or percolation.It refers readers seeking fuller treatment to.Source: §2 · p. 8

2.1. Quantum states

• A qubit is described in a two-state basis, and measurement probabilities depend on the basis; measurements use projectors or generalized positive operators, while evolution is deterministic and unitary.The review's distribution procedures combine measurements, unitaries, and classical communication.Source: §2.1 · p. 8, §2.1.1 · pp. 8–9

2.1.2. Entanglement

• A pure bipartite state is entangled when it cannot be written as a product of subsystem states; Bell states are maximally entangled and enable perfect teleportation of one qubit.The review mostly considers bipartite systems, whose entanglement is better understood than multipartite entanglement.Source: §2.1.2 · pp. 9–10
• Mixed states are represented by density operators, and a mixed bipartite state is entangled when its joint density operator cannot be written as a mixture of product states.The ensemble decomposition of a mixed state is not unique.Source: §2.1.2 · p. 10
• For pure two-qubit states, Schmidt coefficients characterize separability and maximal entanglement; concurrence and conversion probability provide entanglement measures.The review also describes Werner states, whose depolarisation can reduce entanglement, and notes that no general algorithm determines separability for arbitrary density operators.Source: §2.1.2 · p. 11

2.2. Entanglement manipulation in quantum networks

• Entanglement is fragile and deteriorates during manipulation or storage, motivating network-based approaches to distributing it.Sending photons through fibres is limited by photon loss and loss of coherence; intermediate nodes allow links to be joined through entanglement swapping.Source: §2.2 · p. 12

2.2.1. Structure of quantum networks

• A quantum network consists of nodes holding qubits and links representing entanglement between qubits at different nodes; its structure can be viewed as a graph.A regular graph with local connections is called a lattice.Source: §2.2.1 · pp. 12–13

2.2.2. Local operations and classical communication

• The review restricts allowed operations to LOCC: quantum operations occur within nodes, while only classical messages pass between nodes.Qubits at different nodes cannot interact quantum mechanically.Source: §2.2.2 · p. 13

2.2.3. Purification of weakly entangled states

• Purification concentrates entanglement from multiple weakly entangled states into a more entangled state using LOCC; for pure states, majorisation characterizes deterministic conversions.A two-qubit pure state can be converted to a Bell pair with optimal probability 2ϕ1 using the Procrustean method.Source: §2.2.3 · pp. 13–15
• For Werner states, at least two copies are needed for finite-probability LOCC purification to a higher-fidelity state, and perfect Bell pairs require infinitely many copies.The described purification step improves the target-state fidelity only under specified input conditions and succeeds probabilistically.Source: §2.2.3 · p. 15
• Iterating purification can make a state arbitrarily close to a pure state, but this asymptotic procedure has vanishing yield; positive yield requires switching to one-way purification.The initial protocol assumes perfect operations.Source: §2.2.3 · p. 15

2.2.4. Entanglement swapping

• Entanglement swapping uses a Bell-basis measurement at an intermediate station to connect previously unconnected stations; for Werner inputs with parameters x and x′, the output is ρW(xx′).For pure inputs, outcomes may be Bell pairs or weakly entangled states, while average output entanglement is not less than that of the inputs; equal inputs give the conserved entanglement described in.Source: §2.2.4 · pp. 15–16
• Repeated swapping creates increasingly distant connections, but long-distance entanglement decreases exponentially with the number of swaps.This loss motivates protocols designed to efficiently entangle stations across a network.Source: §2.2.4 · pp. 16–17

2.3. The quantum repeaters

• Repeaters store several states with fidelity above Fmin on short links, purify them, swap purified states across neighboring links, and iterate to extend entanglement over long distances.Noisy swapping reduces fidelity, requiring additional memories and purification steps.Source: §2.3 · p. 18
• Reliable quantum memories remain a major challenge to realizing repeaters, despite proposals that avoid increasing qubits per station with distance.The cited review covers proposed improvements; the constant-qubit result is attributed to.Source: §2.3 · p. 18

2.3.1. Implementations

• Teleportation over 150 km using entangled photons distributed through free space has been demonstrated, alongside teleportation using atom-photon interfaces and entanglement swapping in different scenarios.The cited demonstrations are,, and, respectively.Source: §2.3.1 · p. 18
• The DARPA Quantum Network distributes quantum keys across sites tens of kilometres apart, while Vienna's six-node SECOQC network has links ranging from 6 to 85 km.The network examples are cited as [66] and, respectively.Source: §2.3.1 · p. 18

• In higher-dimensional networks, information may reach its destination through alternative paths despite defective links, unlike the one-dimensional case where any defective link destroys the procedure.Percolation theory describes conditions under which two nodes in a square lattice remain connected by a path with non-vanishing probability.Source: §3 · p. 19
• Percolation strategies initially addressed pure states and were later generalized to some special cases of mixed states.The discussion points to sections 3.2 and 3.3 for these strategies and generalizations.Source: §3 · p. 19

3.1.1. Hierarchical graphs

• The hierarchical protocol swaps central links ρ, σ, ρ into ˜σ, then purifies three such connections into σ′ and iterates while E(σ′) ≥ E(σ).Figure 5 depicts the two infinite ternary trees and the swapping-then-purification sequence.Source: §3.1.1 · p. 19, Figure 5 · p. 20
• Hierarchical graphs propagate two-qubit entanglement when µ > µc = 1/3; for µ ≥ µ∗ ≈0.655, the propagated state is a perfect Bell pair.For µ < µc, entanglement decreases for every E; µ = C4(ρ).Source: §3.1.1 · p. 20, Figure 6 · p. 21
• The pure-state hierarchical recursion uses µ = C4(ρ), and entanglement cannot propagate when E(ρ) < Ec ≈0.35.For µ > µc, a stable non-trivial fixed point appears.Source: §3.1.1 · p. 20
• For mixed-state hierarchical graphs with Werner connections ρ=ρW(x) and σ=ρW(y), two consecutive central-state swaps yield ˜σ=ρW(˜y), where ˜y=x²y; purification of three ˜σ states is then attempted to form one connection σ′.Purification of mixed states is intrinsically probabilistic.Source: §3.1.1 · p. 20

3.1.2. Regular graphs

• The deterministic hierarchical strategy has a physical scaling limit: either bond length or qubits per node grows exponentially with iteration depth.Source: §3.1.2 · pp. 21–22
• For square lattices with node degree Z ≥4, the centipede protocol yields perfect Bell pairs after finitely many steps when E(ϕ) > Ec ≈0.649, enabling Bell pairs between spine nodes regardless of distance.Below that threshold, the fixed point is strictly below unity and entanglement along the one-dimensional spine decreases exponentially with length.Source: Figure 7 · p. 21, §3.1.2 · p. 22

3.2. Percolation of partially entangled pure states

• The purified-backbone method requires pure links, backbone nodes with at least four neighbours, and link entanglement above a geometry-dependent critical value; percolation relaxes some of these conditions.The backbone method uses a finite-width strip, while percolation exploits a genuinely multidimensional setting and can also apply to certain rank-less-than-four mixed states.Source: §3.2 · pp. 22–23

3.2.1. Classical entanglement percolation

• Classical bond percolation has a critical probability pc: above it, an infinite cluster appears, whereas below it clusters remain bounded.CEP uses this connectivity transition to determine when distant entanglement can be generated.Source: §3.2.1 · p. 23, Figure 8 · p. 24
• CEP converts each partially entangled link to a Bell pair with probability E(ϕ); above the lattice threshold, a random Bell-pair path can be swapped to entangle distant nodes.The path exists when both stations belong to the infinite cluster.Source: §3.2.1 · p. 24

3.2.2. Quantum entanglement percolation

• On the honeycomb lattice, swapping at half the nodes maps the graph to a triangular lattice without changing average bond entanglement, enabling QEP below CEP’s threshold.QEP succeeds with positive probability in the range where CEP cannot generate long-distance entanglement.Source: §3.2.2 · pp. 25–26, Figure 9 · p. 26

3.2.3. Multi-partite entanglement percolation

• Incomplete generalized swapping and concentration create GHZ states that form occupied sites on a new lattice, where a spanning cluster can be joined into a giant GHZ state.Measuring all but two qubits of a GHZ state in the X basis yields a Bell pair between the remaining qubits.Source: §3.2.3 · pp. 26–27
• The multipartite protocol lowers the honeycomb threshold to 0.650 from CEP’s 0.677 and improves thresholds and connection probabilities across all Archimedean lattices.For the other Archimedean lattices, the connection-probability improvement holds for ˆEc < E(ϕ) < 1.Source: §3.2.3 · pp. 27–28

3.3. Towards noisy quantum networks

• Two disjoint paths of purifyable mixed states (PMSs) are required for a nonzero Bell-pair probability, and the final stage must purify multiple PMSs.For two PMSs, a pure-state conversion measurement uses local CNOT gates and target measurements; identical inputs yield a Bell pair on success.Source: §3.3 · p. 28
• Swapping PMSs leaves only two useful Bell-basis outcomes, and the average output fidelity decays exponentially with the number of links.Source: §3.3 · p. 29
• For two PMSs, Bell-pair conversion succeeds with maximum probability 1/2, allowing percolation on a double-bond triangular lattice but not a double-bond square lattice.For three or more bonds, conversion rates above 1/2 are possible, but the optimal protocol is unknown.Source: §3.3 · p. 29
• With two PMSs per link, the described QEP strategy performs pure-state conversion on each link, swaps the resulting states, and delays Bell-pair conversion.For identical input states, this reduces to CEP because successful conversion already produces Bell pairs.Source: §3.3 · p. 29
• When PMS parameters differ within a link, pure-state conversion maps the problem to pure-state percolation with probabilistically deleted links; rank-three alternatives can leave separable states on failure.For rank-three links, the resulting lattice can be treated with error-correction methods, with binary-state pairs deleted probabilistically.Source: §3.3 · pp. 29–30

3.4. Open problems

• Percolation protocols are not optimal at every link entanglement: deterministic Bell-pair generation is possible above 0.649, while entanglement percolation requires E(ϕ) = 1.Purification fails below 0.649, whereas CEP can succeed for E(ϕ) between 1/2 and 0.649.Source: §3.4 · p. 30
• For pure-state networks, the minimum link entanglement for connecting infinitely distant qubits remains unknown, and multipartite percolation is efficient but not known to be optimal.Other protocols, including error-correction techniques on the square lattice, may also achieve long-distance quantum correlations.Source: §3.4 · p. 30
• For square-lattice Werner links, p below the bond-percolation threshold 1/2 precludes long-distance quantum correlations, even though links are entangled for p above 1/3.Source: §3.4 · p. 30
• In the cubic lattice, the known lower bound is approximately 0.2488, but it is not useful because Werner states are separable for p ≤ 1/3.More generally, a network whose best-separable-approximation separable weight falls below the network percolation threshold cannot distribute long-distance entanglement.Source: §3.4 · pp. 30–31
• Determining whether a lower bound Emin exists and designing protocols that bring Ec close to it remain open problems.Source: §3.4 · p. 31

4. Network-based error-correction

• Full-rank mixed-state links pose a different problem because a finite number cannot be purified into maximally entangled states, unlike pure or restricted mixed-state links.The section focuses on Werner states, which it says can represent mixed states of the same fidelity after depolarisation.Source: §4 · p. 32

4.1. A critical phenomenon in lattices

• Syndromes are dual-lattice vertices whose value records the parity of adjacent defective links, revealing nearby errors without directly testing the links.Links are defective with probability p and valid with probability 1 − p.Source: §4.1 · p. 32
• Below the critical error rate, syndrome patterns allow most defective links to be detected, supporting long-range entanglement in mixed-state lattices.On an infinite square lattice, local detection failures do not affect correction macroscopically when p is at or below pc.Source: §4.1 · p. 32, Figure 11 · p. 33
• The square-lattice example illustrates a critical transition: below pc most errors can be corrected, while above it error correction fails macroscopically.The figure identifies pc as approximately 0.11.Source: Figure 11 · p. 33, §4.1 · p. 33

4.2. Correction of local errors from a global syndrome pattern

• The general protocol creates a noisy multipartite state, measures all but two distant qubits to reveal error information, then globally analyzes outcomes to choose corrections.The final operations are applied to the remaining two qubits to obtain useful remote entanglement.Source: §4.2 · p. 33

4.2.2. Syndrome pattern

• Noisy links create a multipartite state with errors, so measuring all but the target qubits immediately leaves the final Bell-pair correction basis correct only about one quarter of the time.Without knowing which of four Bell pairs is present, the target qubits are separable.Source: §4.2.2 · p. 34
• Local measurements check symmetries of the multipartite state and reveal whether an adjacent link introduced an error without damaging long-distance quantum correlations.A mismatch at a node is called a syndrome; determining the responsible link is the next recovery problem.Source: §4.2.2 · p. 34

4.2.3. Error recovery

• Syndromes mark endpoints of error chains on the dual lattice, but their positions alone leave recovery ambiguous because different chains can produce the same pattern.The observed pattern supplies no other information about the chains.Source: §4.2.3 · p. 34
• For an infinite square lattice, recovery is possible below pc ≈ 10.94%; minimum-weight perfect matching efficiently approximates the otherwise infeasible most-likely error pattern.The threshold is obtained by mapping to the random-bond Ising model; Figure 12 illustrates syndrome matching and failure across the lattice.Source: Figure 12 · p. 35, §4.2.3 · p. 35

4.3.1. Independent bit-flip and phase errors

• With bit-flip errors only, a giant GHZ state on an N × N lattice corrects most errors when p ≤ pc ≈ 10.9%.The links create the GHZ state by generalized entanglement swapping; Figure 14 shows the syndrome pattern and matching.Source: §4.3.1 · p. 36, Figure 14 · p. 37
• Adding phase-error correction encodes each logical qubit in n = 2t + 1 qubits, corrects up to t phase errors by majority vote, and requires links between neighbours to grow logarithmically with distance.Operations can be applied simultaneously, so this approach needs no quantum memory.Source: §4.3.1 · p. 37

4.3.2. Surface codes

• The planar surface-code communication protocol spreads a logical state across an array through repeated stabiliser measurements, with code depth d improving gate-error correction.Repeated measurements trace error chains through space and time; long-distance repeater communication uses Bell pairs at each of d steps.Source: Figure 15 · p. 38, §4.3.2 · p. 39
• For the repeater-based surface code, average time to failure grows non-negligibly with depth when link fidelity is F ≳ 0.92, and transmission loss below 0.45 can be handled efficiently under ideal gates and no-loss assumptions elsewhere.The number of qubits per repeater grows with d, which scales as log(N/pc).Source: Figure 16 · p. 39, §4.3.2 · p. 39
• The network-based surface-code scheme uses entanglement between neighbouring nodes to measure stabilisers, tolerating approximately 1.67% channel error, corresponding to Werner links with x ≈ 0.98.Red and blue nodes perform X and Z stabiliser measurements on neighbouring black-node processing qubits; Figure 17 depicts the node layout.Source: §4.3.2 · p. 40, Figure 17 · p. 41

4.3.3. Entanglement distribution with constant resources

• Unlike the preceding error-correction protocols, the proposals in [30] achieve long-distance entanglement with a constant number of Werner-state connections between neighbours.The section describes a three-dimensional cluster-state protocol and a teleportation-based approach for two- and three-dimensional lattices.Source: §4.3.3 · p. 40
• The first constant-connection protocol generates entanglement across an arbitrarily large cubic lattice when link fidelity exceeds Fc, assuming perfect local operations.The protocol creates a cluster state using noisy connections; local Z errors produce syndromes that are paired for recovery, and the measurement pattern leaves opposite-face qubits maximally entangled ideally.Source: §4.3.3 · pp. 40, 42
• In the three-dimensional cluster-state protocol, distant qubits remain entangled for p < pc ≈ 2.3%, despite imperfect correlations from unobserved syndromes on opposite faces.Monte Carlo simulations support this threshold; the lattice is three-dimensional and some boundary syndromes cannot be assigned values.Source: §4.3.3 · p. 42

4.4. Conclusion

• Long-distance entanglement can be generated in mixed-state networks by collecting information about where bit-flip and phase errors enter.Source: §4.4 · p. 44
• Parity checks at network nodes create syndrome patterns, which are paired to determine local-unitary error corrections.Source: §4.4 · p. 44
• The protocols rely on theoretical results for infinite lattices, so their efficiency in small or medium realistic networks remains to be investigated.Source: §4.4 · p. 44

5.1. Random graphs

• For random graphs, edge probability p defines the Erdős–Rényi model, whose small-world paths and critical giant-component transition make it a useful, analytically tractable starting point despite missing some real-network features.Real complex networks are also characterized by clustering and scale-free behavior, which the Erdős–Rényi graph lacks.Source: §5 · p. 45

5.1.1. Appearance of subgraphs

• In random graphs, many properties emerge at critical probabilities; for p ≲ N^-1 the graph fragments into small clusters, while above that threshold a giant component forms.The thresholds for specific subgraphs depend on how p scales with N; increasing z in p(N) ∼ N^z allows increasingly complex subgraphs to appear.Source: §5.1.1 · pp. 46–47, Table 1 · p. 47

5.2. Quantum random graphs

• A quantum random graph replaces each possible edge’s probability with a shared two-qubit state; converting each connection individually to a Bell pair maps the remaining-state problem to the classical graph model.The graph state coherently superposes simple graphs, with connection entanglement scaled with the number of nodes.Source: §5.2 · pp. 47–48

5.2.1. A complete collapse of the critical exponents

• In quantum random graphs, every finite subgraph can be generated with positive LOCC probability at p ∼ N^-2, collapsing all classical critical exponents to z = −2.A joint measurement at each node creates a highly entangled state shared by a random subset, from which local operations extract the desired subgraph.Source: §5.2.1 · p. 48

5.3. Percolation

• On tree-like networks, q-swap lowers the entanglement-percolation threshold; on Erdős–Rényi networks, its optimal use near average degree 4 reduces the threshold by 20%.The optimal degree-k strategy is to always or never apply q-swap, depending on the lattice and k; on Watts–Strogatz networks, the threshold falls, but q-swap can lose to classical percolation at high initial entanglement.Source: §5.3 · pp. 49–50

5.4. Mixed state distribution

• Many short alternate paths outperform a few long ones in serial multiple purification, while the advantageous region approaches a limit as the number of paths grows.The protocol uses n pairs of paths covering a fraction α of the shortest path, with paired paths taken to have equal length.Source: §5.4 · p. 52, Figure 24 · p. 52
• At the critical Erdős–Rényi point Np = 1, the single-purification protocol’s asymptotic concurrence gain scales as Δ̄C ∼ A N^-2(1 − x)^-4, with A ≈ 6.5 × 10^-5.The expression’s divergence as x → 1 lies outside the asymptotic regime, where purified paths must remain much shorter than the giant-cluster radius; the protocol is also noise-sensitive, with its advantage destroyed above noise levels of a few percent.Source: §5.4 · pp. 52–53

5.5. Optimal path for distributing entanglement on networks

• For swapping Werner states, assigning edge weights −log(x_i) makes the maximum-entanglement path the classical shortest path; optimal-path methods also apply to a restricted teleportation-fidelity problem, but added destinations can break optimal substructure.The teleportation result assumes θ = 0 and partially entangled pure links; the mixed-state extension requires two independent weights per link.Source: §5.5 · pp. 53–54

6. Conclusion

• The application of direct transmission, swapping, purification, and error correction to real networks remains in an exploratory phase.The review frames these developments as theoretical progress driven by experimental results on entanglement-distribution building blocks.Source: §6 · p. 55
• Entanglement distribution has been demonstrated in several physical systems, including quantum cryptographic key distribution and teleportation, but exponential fidelity decay remains unsolved.The review describes higher-dimensional network geometry as promising for concentrating entanglement, alongside classical error correction and quantum connections such as weak-measurement syndromes.Source: §6 · p. 55
• Future work includes mixed-state networks, protocols for small networks, entanglement percolation, error correction, and complex-network theory as node numbers grow.The review expects quantum repeaters to become more robust and enable small networks to form.Source: §6 · p. 55