Experimental Verification of Multipartite Entanglement in Quantum Networks

INTRODUCTION

• The protocol is designed to verify GHZ-state fidelity and genuine multipartite entanglement despite untrusted parties and imperfections, extending earlier work that could not tolerate loss above 50%.It also aims to quantify fidelity when some parties are untrusted and to address loss and other inefficiencies.Source: INTRODUCTION · pp. 1–2

The verification protocol

• The Verifier distributes random angles whose sum is a multiple of π; parties measure in angle-dependent bases and return outcomes tested against an even- or odd-parity rule.For honest parties, the fidelity bound is F(ρ) ≥ 2P(ρ) − 1, and GME is verified when P(ρ) > 3/4.Source: The verification protocol · pp. 2–3
• With dishonest parties, the θ-protocol’s GME threshold is approximately 0.818, compared with approximately 0.854 for XY; loss can let cheaters skew pass statistics.The stated thresholds are 1/2 + 1/π for θ and cos²(π/8) for XY; the loss analysis allows dishonest parties to declare loss selectively.Source: The verification protocol · p. 3

Experimental setup

• The setup generates three- and four-photon GHZ states using photon-pair sources and a postselected fusion operation, then characterizes the states by tomography.The measured fidelities are 0.80 ± 0.01 for three photons and 0.70 ± 0.01 for four photons; the experiment uses postselection and considers its impact, including loss.Source: Experimental setup · pp. 3–4, Figure 2 · p. 4

Three-party verification

• In three-party experiments, XY passed at 0.838 ± 0.005, below the dishonest-party GME bound of 0.854; θ passed at 0.834 ± 0.005, exceeding the 0.818 bound by over three standard deviations and enabling verification under distrust.Source: Entanglement verification · pp. 4–5
• 0.834 pass probability exceeded the θ-protocol’s 0.818 dishonest-party GME bound by more than three standard deviations, verifying three-party GME with an untrusted party.The XY result, 0.838, did not exceed its 0.854 bound.Source: Entanglement verification · p. 5

Dishonest strategies

• The θ-protocol’s optimal non-GME cheating strategy uses a rotated Bell pair and loss declarations for angles with the lowest pass probability.For the XY-protocol, the optimal state is a probabilistic mixture of a Bell pair and a π/4-rotated Bell pair; its GME bound reaches 1 at 50% loss, versus 100% for θ.Source: Entanglement verification · p. 6

Verification with loss

• The three-party θ-protocol verified GME up to about 5% loss, while the four-party experiment could not verify GME at any loss.The three-party loss tolerance depended on high state fidelity; the four-party honest pass probability remained below the GME bound.Source: Entanglement verification · pp. 6–7

Four-party verification

• For four parties, the all-honest pass probabilities were 0.776 ± 0.005 for XY and 0.767 ± 0.005 for θ, just sufficient to verify GME in the honest case.With dishonest parties, neither protocol confirmed GME because the pass probabilities remained below the 0.854 and 0.818 bounds, respectively.Source: Entanglement verification · p. 6

Dishonest strategies

• The four-party lower-noise biseparable resource raised θ pass probability from 0.669 ± 0.005 to 0.698 ± 0.008, but remained below the GME bound.The alternative resource used a three-qubit GHZ state for honest parties and an unentangled photon for the dishonest party.Source: Entanglement verification · p. 7, Figure 4 · p. 7

DISCUSSION

• The demonstrated three-party protocol verified GME with untrusted parties, including the source, for up to 5% loss; higher-fidelity states could improve loss tolerance and support more qubits.The discussion places this result in distributed photonic communication networks.Source: DISCUSSION · p. 7
• The verification tests also link their outcome to the state used by honest parties, beyond merely detecting entanglement.The discussion notes that network tasks may use multipartite states other than GHZ, such as stabiliser states, to which the authors expect the protocols to apply.Source: DISCUSSION · p. 7

Supplementary Figures

• The θ-protocol performs better than the XY-protocol in the supplementary loss comparison and remains viable after 50% loss.Source: Supplementary Figures · p. 9

Supplementary Note 1

• The supplementary note provides further θ-protocol details and treats the XY-protocol proofs as a special case with two measurement settings.Source: Supplementary Note 1 · p. 9

• For an ideal n-qubit GHZ state, the θ-protocol passes with unit probability.Source: Correctness of the θ-protocol · p. 10
• The parties’ X-Y-plane measurements are equivalent to Z-axis rotations followed by Pauli X-basis measurements.Source: Correctness of the θ-protocol · p. 10
• When the summed rotation angle is 0 modulo 2π, the X-basis outcomes have even parity; when it is π modulo 2π, they have odd parity.Source: Correctness of the θ-protocol · p. 10

• The section establishes a lower bound on shared-state fidelity as a function of the honest test’s pass probability.Source: Security in the Honest Model · p. 10
• The test for a rotated GHZ state requires the parties’ angle sum to match its phase modulo π.Source: Security in the Honest Model · p. 10
• For an honest shared state ρ, the pass probability obeys P(ρ) ≤ 1/2 + F(ρ)/2, yielding the fidelity bound F(ρ) ≥ 2P(ρ) − 1.Source: Security in the Honest Model · p. 11

Security in the Dishonest Model

• F′(ρ) ≥ 4P(ρ) − 3 lower-bounds the fidelity after optimizing over operations on the dishonest parties’ share.The analysis allows dishonest parties to collaborate with the source and apply operations to their part of the state.Source: Security in the Dishonest Model · p. 11
• P(ρ|θ-protocol) ≤ 1/2 + 1/π and P(ρ|XY-protocol) ≤ cos²(π/8) bound dishonest parties’ pass probabilities in the stated corollary.The respective bounds are approximately 0.818 and 0.854.Source: Security in the Dishonest Model · p. 13
• For the XY-protocol, the optimal dishonest strategy reaches pass probability cos²(π/8) ≈ 0.854 by choosing the state-dependent angle α = −π/4.The bound follows from maximizing the probability over the angle the source can choose.Source: Security in the Dishonest Model · p. 14

Loss

• To bound non-GME states, the analysis enlarges the dishonest party to include honest players on its side of a separable cut, then considers product states.The source and dishonest parties are treated as a single system with control of the source’s purification.Source: Loss · p. 14
• For the θ-protocol, dishonest players can post-select a λ fraction of angles, targeting those near π/2 + θ′.The passage identifies these as the worst angles for the honest state.Source: Loss · p. 15
• At zero loss, the XY-protocol’s optimal pass probability is cos²(π/8) ≈ 0.854; at 50% loss, it reaches 1 by declaring loss on Pauli Y requests.For loss between those cases, the optimal strategy is a probabilistic mixture of the two pure strategies.Source: Loss · p. 15
• For intermediate loss, the XY source mixes the θ′ = 0 loss-declaring strategy with the θ′ = π/4 strategy.The mixture uses probabilities 2λ and 1 − 2λ, respectively.Source: Loss · p. 15

State generation

• Photon pairs are generated by spontaneous four-wave mixing in strongly birefringent, microstructured fibers, with narrowband, spectrally uncorrelated photons as the target.The source is pumped with Ti-Sapphire pulses at 726 nm; the signal and idler wavelengths are 623 nm and 871 nm.Source: State generation · p. 15
• In a Sagnac loop, bidirectional pumping produces pairs that coherently share a mode and form a polarization-entangled pair state.The state is 1/√2(|H⟩s|H⟩i + e^{iθ}|V⟩s|V⟩i), up to phase θ.Source: State generation · p. 16
• Post-selected fusion produces three- and four-photon GHZ states, with source temperature tuning used to improve signal-photon spectral indistinguishability.The second source was tuned to 23.7°C relative to an ambient 17.6°C; residual spectral differences limit maximum fidelity.Source: State generation · p. 16
• Wave-plate rotations and polarizing beam splitters implement local projective measurements, but chromatic deviations mean some rotations are only approximated.Approximate input states were chosen to reach the Pauli bases and X-Y equator.Source: State generation · p. 16

4-qubit GHZ

• The fused state contains the desired four-photon GHZ terms at order α² and non-ideal higher-order terms at order α³.The two-source state is transformed at a polarizing beam splitter, retaining events with at least one photon in each mode.Source: 4-qubit GHZ · p. 16
• H/V-basis postselection can discard some higher-order terms, but this additional filtering is not available in all measurement bases.The text notes that this basis-dependent postselection was not included in the stated fidelity model.Source: 4-qubit GHZ · p. 17
• At 7 mW pump power per fiber per direction, higher-order emissions reduce the modeled GHZ fidelity by 11%, to 0.89.Reducing pump power to 1 mW reduces the fidelity impact to 2%, at the expense of the overall four-fold rate.Source: 4-qubit GHZ · p. 17

3-qubit GHZ

• The source state includes vacuum and one- and two-pair horizontal-polarization terms, with higher-order contributions represented by O(α3).The state is written as a superposition weighted by α and α2.Source: 3-qubit GHZ · p. 17
• The fusion starts with one source in the product configuration and the other in the entangled configuration, then applies an HWP and PBS transformations.Terms with at least one photon in each mode are retained, conditioned on detecting one or more photons in mode i1.Source: 3-qubit GHZ · p. 17
• Terms proportional to α2 produce the desired GHZ state, while higher-order α3 terms make it non-ideal.The derivation does not include further postselection based on measurement basis.Source: 3-qubit GHZ · p. 17
• At P = 1 mW, the impact of higher-order terms on fidelity is reduced to 2%.For n̄ = 0.05 and α = 0.22, the fidelity is F = 0.88; emissions up to α3 reduce state quality by 12%.Source: 3-qubit GHZ · p. 17