Learning Quantum Entanglement Distillation With Noisy Classical Communications

Hari Hara Suthan Chittoor, Osvaldo Simeone

TL;DR

Entanglement distillation is a vital process for cleaning up "noisy" quantum connections, but existing methods break down if the classical messages used to coordinate them are corrupted. This paper introduces NA-LOCCNet, a machine learning-based approach that uses Parameterized Quantum Circuits (PQCs) to design distillation protocols that are explicitly robust to errors in classical communication.

What Problem Does This Paper Address?

To build a "Quantum Internet," we need to share entangled qubits (ebits) across long distances. However, current hardware is imperfect; the entangled pairs produced are often "noisy" mixed states rather than pure, high-quality Bell pairs. Entanglement distillation is the solution: it takes several low-quality entangled pairs and uses Local Operations and Classical Communication (LOCC) to "distill" them into a single pair of much higher quality.

The catch? Almost all existing distillation protocols (like the classic DEJMPS or the newer LOCCNet) assume that the classical part of the communication—the bits Alice and Bob send to each other to coordinate their actions—is perfect. In the real world, classical channels have noise. If a bit flips during transmission, Alice and Bob might mistakenly keep a bad pair or throw away a good one, potentially making the resulting entanglement worse than what they started with.

Methods

The authors propose NA-LOCCNet (Noise Aware-LOCCNet), which treats the design of a distillation protocol as a supervised learning problem.

  1. Parameterized Quantum Circuits (PQCs): Instead of using a hand-designed sequence of gates, Alice and Bob use quantum circuits where the rotation angles ($\theta$) of the gates are tunable parameters.
  2. Novel Circuit Architecture: The authors introduced a specific two-qubit gate, the Pauli $ZY$-rotation, into their PQC. Through experimentation, they found this architecture more flexible and robust for noisy environments than the standard CNOT-based designs used in previous work.
  3. Noise-Aware Objective Function: This is the "secret sauce." Standard methods try to maximize fidelity assuming the classical bits are correct. NA-LOCCNet optimizes for average conditional fidelity. It mathematically accounts for the probability of bit-flips (using a Binary Symmetric Channel model) and ensures the quantum gates are tuned to produce the best possible state even when the classical "success" signal might be a lie.
  4. Optimization: The parameters are optimized using a one-dimensional search (or gradient descent) to find the best $\theta$ for a given level of channel noise.

Datasets

As a theoretical and simulation-based study, the paper does not use external real-world datasets. Instead, it uses:

  • Simulated S-states: A standard model for noisy entangled qubits where the state is a mix of a perfect Bell pair and a "separable" (non-entangled) state.
  • Synthetic Noise Models: Classical communication is simulated using a Binary Symmetric Channel with varying bit-flip probabilities ($p$) ranging from 0 (noiseless) to 0.5 (pure noise).

The authors have made their PyTorch implementation available on GitHub.

Key Insights and Ideas

  • The "Do No Harm" Principle: One of the most striking findings is that standard protocols (DEJMPS and LOCCNet) can actually reduce the fidelity of qubits if classical noise is high enough. NA-LOCCNet is designed to be "noise-aware," meaning it adapts its quantum operations so that the output fidelity never drops below the input fidelity.
  • Architecture is an Ansatz: The choice of quantum gates (the "ansatz") matters deeply. The shift to a $ZY$-rotation gate allowed the system to find robust solutions that simpler architectures missed.
  • Average vs. Instantaneous Success: By optimizing for the average outcome across all possible classical error scenarios, the researchers created a protocol that is "pessimistic" in the right way—it doesn't over-rely on a single measurement outcome.

Key Results

  • Superior Robustness: In a scenario with an input fidelity of 0.6 and a classical bit-flip probability of 0.25, NA-LOCCNet achieved an output fidelity of ~0.85, significantly outperforming DEJMPS (~0.7) and standard LOCCNet (~0.75).
  • Performance at Maximum Entropy: Even when the classical channel was completely random ($p=0.5$), NA-LOCCNet maintained a fidelity of 0.8, while other protocols collapsed to 0.5 (worse than the starting state).
  • No Free Lunch? Interestingly, NA-LOCCNet achieves this higher fidelity without a significant drop in the "probability of success" compared to other machine-learning approaches, proving that you can gain robustness without sacrificing efficiency.

Limitations and Future Work

  • Scaling: The paper focuses on the $k=2$ case (distilling two pairs into one). Real-world applications might require distilling many pairs ($k > 2$) or multi-party entanglement, which increases the complexity of the PQC optimization.
  • Simplified Noise: The study assumes a simple bit-flip model for classical noise and specific "S-state" noise for quantum states. Future work could explore more complex, non-symmetric noise or decoherence effects during the quantum gates themselves.
  • Hardware Implementation: While simulated in PyTorch, the next step would be implementing these optimized PQCs on actual NISQ (Noisy Intermediate-Scale Quantum) hardware to see if the theoretical gains hold up against real-world gate errors.

Summary updated

BibTeX
@inproceedings{chittoor2023learning,
  author = {Hari Hara Suthan Chittoor and Osvaldo Simeone},
  title = {{Learning Quantum Entanglement Distillation With Noisy Classical Communications}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2023},
  doi = {10.1109/icassp49357.2023.10097202},
}

Loading related content...

Figure