Experimental High-Dimensional Quantum Overlapping Tomography

AI summary

Overview

Research area: Quantum information science — specifically quantum state tomography and entanglement certification for multipartite, high-dimensional (qudit) systems, demonstrated on a photonic platform.

Technical level: Advanced. The paper assumes familiarity with density matrices, mutually unbiased bases (MUBs), Schmidt numbers, partial transpose/negativity, and graph-theoretic covering problems.

Scope (one sentence): The authors formulate high-dimensional quantum overlapping tomography (QOT) as a graph/covering-array problem and demonstrate it experimentally on a four-photon hybrid qudit–qubit state with local dimensions (4, 4, 2, 2), reconstructing all six two-body marginals with 25 MUB settings.

What This Paper Is About

Characterizing a full quantum state becomes prohibitively expensive as the number of particles and the local dimension grow, yet most applications only need few-body (local) information such as two-body reduced density matrices. Quantum overlapping tomography (QOT) extracts all such marginals with far fewer measurement settings than independent tomography of each marginal, but previous QOT protocols and experiments were essentially restricted to qubits. This paper generalizes QOT to arbitrary (prime-power) local dimensions using a graph-theoretic clique-cover / covering-array formulation, and then realizes the protocol on a photonic four-party entangled state to reconstruct all six two-body marginals.

Key Contributions

1. A general high-dimensional QOT protocol. The authors map the problem of reconstructing all k-body marginals of an n-party system with local dimensions dᵢ onto a covering-array problem with alphabet size v = d + 1 (for MUB-based measurements), and, for k = 2, onto a clique-cover problem on the complete multipartite graph K_{d₁+1, d₂+1, …, dₙ+1}. Independent measurements per marginal would need (d+1)^k · C(n,k) settings; the construction scales as (d+1)^k · O(log n) for fixed k.

2. First experimental demonstration of high-dimensional QOT. The protocol is implemented on a photonic four-photon state in a 4 × 4 × 2 × 2 system (two ququarts, two qubits), reconstructing all six two-body marginals from only 25 MUB projective measurement settings (1600 outcomes), versus 94 settings (6016 outcomes) for independent tomography of every marginal and 225 settings (14400 outcomes) for full state tomography.

3. Revealing a layered entanglement structure. The reconstructed marginals show that only ρ_AB carries significant off-diagonal coherence and statistically significant negativity, with a structure corresponding to a classical mixture of two Bell pairs in orthogonal subspaces — a resource the authors connect to layered quantum communication networks.

4. A noise-robust certification route. Using the QOT marginals, the authors apply an inner-product criterion (IPC) based on global-to-local overlap ratios to lower-bound the Schmidt-number vector, and show experimentally that IPC continues to certify entanglement dimensionality at noise levels where the fidelity-based criterion (FBC) fails.

Main Findings

• Setting reduction: All six two-body marginals were reconstructed with 25 MUB settings for the K_{5,5,3,3} graph, compared with 94 for independent two-body tomography and 225 for full state tomography. The graph example given in the paper (three qutrits, K_{4,4,4}) requires 16 settings versus 48 for independent marginal tomography.

• Marginal reconstruction quality: Average fidelity of the six marginals with respect to the ideal marginals is 0.876 ± 0.014. Individual values: F_AB = 0.812 ± 0.012, F_AC = 0.900 ± 0.015, F_AD = 0.863 ± 0.017, F_BC = 0.856 ± 0.015, F_BD = 0.936 ± 0.012, F_CD = 0.887 ± 0.014.

• State preparation quality: The prepared state has fidelity 0.809 ± 0.009 with the target |Λ⟩, determined from 35 measurement settings. The maximum overlap achievable by any state with a smaller Schmidt number (for at least one bipartition) is F_max = 0.75, so the observed fidelity certifies the Schmidt-number vector [4 4 2 2 4 4 2] across the seven bipartitions.

• Entanglement located in one pair: Only ρ_AB shows significant non-zero off-diagonal terms and a negativity that is non-zero with statistical significance; the other five two-body marginals display only classical correlations. ρ_AB corresponds to a classical mixture of the Bell pairs (|00⟩ + |22⟩) and (|11⟩ + |33⟩) in orthogonal subspaces.

• Remote Bell-state distribution: Projecting parties C and D onto |00⟩_CD prepares the Bell state (|11⟩ + |33⟩)_AB, while projecting onto |11⟩_CD prepares (|00⟩ + |22⟩)_AB — different levels of the high-dimensional system, a feature the authors highlight for flexible quantum networks.

• Noise-resilient certification: With the noise model ρ(p) = (1 − p)ρ + p·I/64, the IPC still certifies the entanglement dimensionality vector as white noise increases, in a regime where the FBC fails once the experimental fidelity drops below F < 0.75. Experimentally, the IPC lower-bounds r_{A|BCD}, r_{B|ACD}, r_{AC|BD}, r_{AD|BC} > 3 and r_{AB|CD}, r_{C|ABD}, r_{D|ABC} > 1.

• Scaling and sample complexity: The number of settings grows as (d+1)^k · O(log n), derived both by a probabilistic union-bound argument and via the Stein–Lovász–Johnson covering-array bound. The number of samples scales as N = O[(σ²_max/ε²) log(|K|/δ)]; the settings used here have σ_max = 19 (compared with σ_Pauli = 5 for the standard nine two-body Pauli settings for two-qubit tomography). In the homogeneous case this becomes O[d^{2k}/ε² · (k log n + log(1/δ))].

• Experimental parameters: Four-photon coincidence rate of 0.8 Hz with interference visibility 0.893 ± 0.005; coincidence counts collected for 600 seconds per setting; error bars from Monte Carlo sampling with 1000 samples following a Poisson distribution, representing 99% confidence intervals (Fig. 4 error bars are one standard deviation).

Methodology in Plain English

The core idea is to avoid measuring every pair of particles separately. Each particle of local dimension dᵢ has dᵢ + 1 possible mutually unbiased basis settings. The authors draw a graph in which each particle is a group of dᵢ + 1 vertices, and every vertex of one particle is connected to every vertex of another — a complete multipartite graph. Choosing one setting per particle corresponds to picking a clique (a fully connected subgroup) in this graph, and one global measurement then simultaneously delivers measurement data for several two-body marginals at once. Finding the fewest global measurements that still cover every pairwise combination is exactly the minimum clique-cover problem; for k > 2 body marginals it becomes a covering-array problem.

Experimentally, the team produced polarization-entangled photon pairs by spontaneous parametric down-conversion in β-barium borate (BBO) crystals, pumped by ultraviolet femtosecond pulses (390 nm, 80 MHz, 230 mW average power) and emitting at 780 nm. Beam displacers convert polarization into path, so that the states |H_u⟩, |H_l⟩, |V_u⟩, |V_l⟩ encode the four levels |0⟩, |1⟩, |2⟩, |3⟩ — this path–polarization encoding is how the two ququarts are built while the other two photons remain qubits. Half-wave plates at 22.5° after the beam displacers, postselection on fourfold coincidences after a polarizing beam splitter, and 3 nm narrow-band filtering produce the target state. The two ququarts are analyzed with modules of beam displacers, wave plates, polarizing beam splitters, and single-photon avalanche diodes, with fourfold coincidences recorded by time-correlated single-photon counting. Maximum likelihood estimation enforces positivity of the reconstructed marginals.

For the noise-robustness test, independent noise sources are placed in front of each optical coupler to add white noise, and the two entanglement-dimensionality criteria (inner-product versus fidelity-based) are compared across noise levels.

Why This Matters

Impact on research. Full tomography of multipartite qudit states scales exponentially and quickly becomes impractical; QOT replaces that with a logarithmic-in-n scaling for the measurement-setting count, making characterization of larger and higher-dimensional systems feasible. The work connects high-dimensional quantum information to mature combinatorial designs (covering arrays, clique covers) and extends the QOT toolbox beyond qubits, following earlier qubit demonstrations and Pauli/Gell-Mann formulations. It also shows that marginal data alone can support entanglement certification that outperforms fidelity-based witnesses under noise.

Real-world applications (as framed by the paper):

• Layered quantum communication networks: the asymmetric pairwise entanglement structure allows parties with higher security levels (the paper's example: banks) to share extra secret keys inaccessible to parties at lower levels (e.g., ATMs), with QOT certifying security levels across every pair simultaneously.
• Multipartite secret-key distribution: the reconstruction supports the conclusion that keys can be shared across all four parties while A and B additionally share keys regardless of what C and D measure.
• Entanglement certification for device characterization: Schmidt-number vector certification with noise tolerance, useful wherever high-dimensional entanglement is a resource.
• Benchmarking quantum gates and noise processes: the authors note that the scheme can be extended to quantum process overlapping tomography for qudits, with potential use in benchmarking gates and noise in quantum computing devices.

Industry relevance. The demonstration uses photonic hardware compatible with on-chip optical platforms, a platform the paper cites as an active route to high-dimensional multipartite entanglement. The measurement-overhead reduction matters directly for any organization that must validate quantum hardware — quantum networking, quantum communication security, and quantum computing benchmarking — where characterization cost is a practical bottleneck.

Future Directions

• Scaling to more parties and higher dimensions. The logarithmic-in-n setting count is proven theoretically and demonstrated at four parties; whether the protocol remains practical for substantially larger n and larger d under real noise and finite coincidence rates is untested here.
• Process tomography of qudit channels. Extending the framework to quantum process overlapping tomography is proposed but not demonstrated; it would open benchmarking of qudit gates and noise processes.
• Fermionic and other generalizations. The introduction notes that graph-based QOT has been generalized to fermionic systems, while the high-dimensional formulation here is restricted, for simplicity, to prime-power dimensions with MUBs; non-prime-power dimensions require less symmetric informationally complete settings.
• Improving sample complexity. The number of measurement settings is reduced, but the sample count still scales with σ²_max/ε²; the reported σ_max = 19 for these settings (versus 5 for the standard two-qubit Pauli set) indicates room for optimizing the measurement design against statistical cost, not just setting count.

Target Audience

Quantum information researchers working on state characterization, entanglement certification, and high-dimensional/multipartite entanglement; experimental photonic quantum optics groups; combinatorial design theorists interested in covering arrays and clique covers; and quantum networking or quantum computing engineers who need practical, low-overhead benchmarking protocols for multi-qudit devices. Readers without a graduate-level background in quantum information will find the protocol and certification sections demanding.

Authors’ abstract

Large-scale quantum systems have advanced rapidly via the exploration of more particles and higher dimensions, offering great potential for developing quantum technologies. However, their characterization becomes prohibitive with increasing local dimensionality and particle number. Here we propose high-dimensional quantum overlapping tomography based on a graph-theoretic formulation, which allows one to efficiently reconstruct few-body marginals of multipartite high-dimensional quantum systems. We experimentally realize it on a photonic four-party entangled state in a $4 \times 4 \times 2 \times 2$ system. Using measurements in mutually unbiased bases, we reconstruct all six two-body marginals with only 25 projective measurement settings, compared with 94 and 225 settings for independent tomography of all two-body reduced states and full state tomography, respectively. The reconstructed marginals reveal a layered entanglement structure vital for high-dimensional quantum networks. We further show that these marginals enable more noise-resilient certification of multipartite high-dimensional entanglement than the fidelity-based criterion. Our work thus offers a scalable route for learning multidimensional quantum systems.