Random Party Distillation on a Superconducting Processor

Alexander C. B. Greenwood, Jackson Russett, Hoi-Kwong Lo, Qian LiarXiv 2025 · DOI · arXiv

TL;DR

Researchers have successfully implemented a multi-round "random party distillation" protocol on IBM’s superconducting quantum processors, achieving a record-breaking distillation rate of 0.81 EPR pairs per W-state. By utilizing weak measurements and advanced error mitigation, they demonstrated that multipartite entanglement can be efficiently converted into useful bipartite entanglement even on noisy, intermediate-scale quantum (NISQ) hardware.

What Problem Does This Paper Address?

Entanglement is the "fuel" of quantum communication and computation. While bipartite entanglement (between two parties, like an EPR pair) is the standard requirement for tasks like teleportation, multipartite entanglement (between three or more parties) is often easier to generate in certain physical systems.

The challenge lies in distillation: how do you take a single copy of a multipartite state, like a three-qubit "W-state," and extract a high-quality bipartite EPR pair from it? A simple "strong" measurement on one qubit works but has a limited success probability (2/3 for a W-state). "Random party distillation" (RPD) offers a better way: by performing a series of "weak" measurements, you can keep the state alive across multiple rounds, increasing the total probability of success toward 100%. However, implementing this on real hardware is difficult because mid-circuit measurements are slow and introduce significant noise (dephasing) into the system.

Methods

The authors adapted a theoretical protocol originally designed for high-dimensional "qudits" into a version compatible with standard qubits.

  1. Weak Measurements via Ancillas: Instead of measuring the system qubits directly, each party (Alice, Bob, and Charlie) couples their qubit to an auxiliary "ancilla" qubit using a controlled rotation gate ($CRY(\theta)$).
  2. Dynamic Circuits: They used "dynamic circuits," where the results of mid-circuit measurements on the ancillas determine whether the protocol continues to the next round or exits with a distilled pair.
  3. Error Mitigation: To combat the ~1µs delay caused by mid-circuit measurements—a lifetime in the quantum world—the team employed XY4 dynamical decoupling (flipping qubits to cancel out environment noise) and Matrix-free Measurement error Mitigation (M3) to clean up the readout data.
  4. Hardware: The experiments were primarily conducted on the 127-qubit ibm_quebec processor, with additional tests on the 156-qubit ibm_fez (Heron r2) processor.

Datasets

The study does not use traditional external datasets. Instead, it generates experimental data by:

  • Initializing a three-qubit W-state with a measured fidelity of 0.958 ± 0.007.
  • Running the RPD protocol for up to four rounds.
  • Benchmarking the qubits' coherence ($T_1$) and dephasing ($T_2$) times, which ranged from ~160µs to ~480µs.

Key Insights and Ideas

  • The "Random" Advantage: The protocol is "random" because it doesn't matter which two parties end up with the EPR pair. This flexibility is what allows the success rate to climb higher than specific-party distillation.
  • Measurement is the Bottleneck: The authors identify that the dominant source of error isn't the logic gates themselves, but the latency of mid-circuit measurements. While a gate takes 20-60ns, a measurement takes ~1000ns, giving the qubits ample time to lose their quantum "magic" (decohere).
  • Ancilla-Heavy Alternative: The paper proposes a secondary protocol that avoids mid-circuit measurements entirely by using a "cascade" of CNOT gates to transfer information to a larger number of ancilla qubits. This suggests a trade-off: use more qubits to save time, or use fewer qubits and risk dephasing.

Key Results

  • Record Distillation Rate: The 4-round protocol achieved a distillation rate of 0.81 pairs per W-state, significantly higher than the previous state-of-the-art (~0.75) and the simple 1-round limit (0.66).
  • Theoretical Alignment: For the first two rounds, the experimental success rate tracked the theoretical prediction almost perfectly within 1-$\sigma$ error.
  • Entanglement Quality: The average "entanglement of formation" (a measure of how "entangled" the resulting pair actually is) remained high, though it naturally degraded as the number of rounds increased due to accumulated noise.

Limitations and Future Work

The primary limitation is readout hardware. The long duration of current mid-circuit measurements limits how many rounds can be performed before the state becomes too noisy to be useful.

The authors suggest that future improvements will come from:

  • Faster Readout: New architectures like "junction readout" could drastically reduce measurement time.
  • Staggered Decoupling: More sophisticated ways to shield qubits from crosstalk during measurements.
  • Scaling: Extending these protocols to larger multipartite systems (more than three parties) to build complex quantum networks.

Summary updated

Abstract

Random party distillation refers to the process by which Einstein-Podolsky-Rosen pairs are randomly extracted from a single copy of a multipartite entangled state after multiple rounds of performing positive operator value measure operations. In this work, we propose a qubit-based implementation of a random party distillation protocol and demonstrate its efficacy on the superconducting hardware device, ibm_quebec. We demonstrate a 4-round implementation of the protocol, showing distillation rates superior (0.81 pairs/ W state) to the state of the art. Finally, we explore the dynamical properties of the protocol when implemented on superconducting hardware, and how errors introduced by mid-circuit measurements can be mitigated.
BibTeX
@misc{greenwood2025random,
  author = {Alexander C. B. Greenwood and Jackson Russett and Hoi-Kwong Lo and Qian Li},
  title = {{Random Party Distillation on a Superconducting Processor}},
  year = {2025},
  eprint = {2508.09110},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2508.09110},
  howpublished = {arXiv},
}

Loading related content...

Figure