Evaluation of Distimation's Real-world Performance on a Superconducting Quantum Computer

Quantum state estimation plays a crucial role in ensuring reliable creation of entanglement within quantum networks, yet conventional Quantum State Tomography (QST) methods remain resource-intensive and impractical for scaling. To address these limitations, we experimentally validate Distimation, a novel distillation-based protocol designed for efficient Bell-diagonal state estimation. Using IBM Quantum simulators and hardware, we demonstrate that Distimation accurately estimates Bell parameters under simulated and real-world noise conditions, but also demonstrating limitations with operational noise and number of available shots. Additionally, we simulate an asymmetric-fidelity Bell pair scenario via Measurement-Based Quantum Computation (MBQC) to further validate Distimation under realistic network conditions. Our results establish Distimation as a viable method for scalable, real-time entanglement monitoring in practical quantum networks.

💡 Research Summary

The paper presents an experimental validation of “Distimation,” a distillation‑based protocol for estimating Bell‑diagonal quantum states, and assesses its performance on both IBM quantum simulators and a real superconducting quantum processor (ibm_kawasaki). The motivation stems from the need for efficient state‑characterisation in quantum networks, where traditional quantum state tomography (QST) is prohibitively resource‑intensive (requiring up to 4ⁿ‑1 measurement settings for n qubits) and disrupts network operation. Distillation, already a routine error‑mitigation step in repeater architectures, can be repurposed to infer the underlying state parameters directly from its success probability, thereby offering a low‑overhead alternative to QST.

The authors first review the theoretical framework. A Bell‑diagonal state ρ_BD = Σ_i q_i |Φ_i⟩⟨Φ_i| arises from Pauli noise on an ideal Bell pair. The special case of a Werner state reduces to a single depolarisation parameter ω. Three distillation circuits—(a) CNOT + Z‑basis measurement (detects X‑type errors), (b) reversed CNOT + X‑basis measurement (detects Z‑type errors), and (c) pre‑rotation + CNOT (detects both X and Z, but not simultaneous Y)—provide three independent success probabilities ˆp(i). Simple algebraic inversions (Eqs. 3‑5) map these probabilities to estimates of ω or the full set {q₁,…,q₄}.

In simulation, the authors use Qiskit AerSimulator configured to mimic the physical layout of ibm_kawasaki (qubits 41‑44). Pauli error channels are injected via identity gates, and large shot numbers (2.7×10⁵ for Werner, 9×10⁴ per Bell‑diagonal scenario) ensure negligible statistical noise. The trace distance between the true and estimated density matrices remains essentially zero across the entire parameter sweep, confirming that Distimation works perfectly under ideal Pauli‑noise conditions. Moreover, the Werner protocol requires only one of the three circuits, while full Bell‑diagonal reconstruction needs all three.

Hardware experiments reveal the protocol’s limitations. The same circuits are run on the real device with only 9×10⁴ shots per circuit (≈5.4×10⁵ Bell pairs total). Measured outcome frequencies (Table I) lead to estimated Werner parameters ω_a = 0.505, ω_b = 0.181, ω_c = 0.593. When compared to a QST‑derived density matrix (obtained via ParallelExperiment with measurements in Z, X, Y bases), the trace distances are 0.636, 0.417, and 0.700 respectively—substantially larger than in simulation. The authors attribute this degradation to two factors: (1) the device’s non‑Pauli noise (notably amplitude‑damping) violates the Bell‑diagonal assumption, and (2) the limited shot budget amplifies statistical fluctuations, especially in the presence of gate errors. The original Distillation paper suggested a bisection‑based estimator that is more robust to noise but requires many more samples; the present work adopts a simpler inversion method, which is more sample‑efficient but also more sensitive to imperfections.

To explore a more realistic network scenario, the authors simulate asymmetric Bell‑pair generation using measurement‑based quantum computation (MBQC). Here one Bell pair decoheres while waiting for the second to be successfully distributed. The asymmetric case yields slightly lower trace distances (≈0.55) than the symmetric case, indicating that Distimation can still provide useful information when the two copies experience different noise histories.

Overall, the study demonstrates that Distimation can dramatically reduce the number of required measurements compared with QST and can accurately recover state parameters under controlled Pauli noise. However, on current superconducting hardware its performance is limited by non‑Pauli errors and by the practical constraints on shot numbers. The authors suggest three avenues for making Distimation viable in operational quantum networks: (i) increase the number of shots (or aggregate data over time), (ii) incorporate pre‑characterised noise models to correct the inversion formulas, and (iii) employ more robust post‑processing such as the bisection algorithm or Bayesian inference. With these enhancements, Distimation could become a cornerstone tool for real‑time entanglement monitoring and error management in future large‑scale quantum communication infrastructures.