Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adaptive Entanglement Generation for Quantum Routing

Tasdiqul Islam, Md Arifuzzaman, Engin Arslan

TL;DR

The paper tackles the challenge of real-time end-to-end entanglement routing in long-distance quantum networks by replacing compute-heavy ILP link selection with reinforcement learning, and by introducing entanglement caching and proactive swapping to reuse entanglements over time. The proposed Adaptive Entanglement Generation (AEG) framework integrates RL-based link selection into REPS and adds two RL-driven components for proactive swapping and caching, achieving up to ~20× faster decision times while maintaining or improving request satisfaction compared to ILP-based baselines. Empirical results demonstrate that entanglement caching improves performance, and proactive swapping on high-demand segments yields substantial gains (up to ~52.5% in request success) across varying network loads and entanglement probabilities. Overall, AEG offers a scalable, real-time solution that significantly enhances quantum routing efficiency and resource utilization in practical networks.

Abstract

Entanglement generation in long-distance quantum networks is a difficult process due to resource limitations and the probabilistic nature of entanglement swapping. To maximize success probability, existing quantum routing algorithms employ computationally expensive solutions (e.g., linear programming) to determine which links to entangle and use for end-to-end entanglement generation. Such optimization methods, however, cannot meet the delay requirements of real-world quantum networks, necessitating swift yet efficient real-time optimization models. In this paper, we propose reinforcement learning (RL)-based models to determine which links to entangle and proactively swap to meet connection requests. We show that the proposed RL-based approach is 20x faster compared to linear programming. Moreover, we show that one can take advantage of the longevity of entanglements to (i) cache entangled links for future use and (ii) proactively swap entanglement on high-demand path segments, thereby increasing the likelihood of request success. Through comprehensive simulations, we demonstrate that caching unused entanglements leads to a 10-15% improvement in the performance of state-of-the-art quantum routing algorithms. Complementing caching with proactive entanglement swapping further enhances the request success rate by up to 52.55%.

Adaptive Entanglement Generation for Quantum Routing

TL;DR

The paper tackles the challenge of real-time end-to-end entanglement routing in long-distance quantum networks by replacing compute-heavy ILP link selection with reinforcement learning, and by introducing entanglement caching and proactive swapping to reuse entanglements over time. The proposed Adaptive Entanglement Generation (AEG) framework integrates RL-based link selection into REPS and adds two RL-driven components for proactive swapping and caching, achieving up to ~20× faster decision times while maintaining or improving request satisfaction compared to ILP-based baselines. Empirical results demonstrate that entanglement caching improves performance, and proactive swapping on high-demand segments yields substantial gains (up to ~52.5% in request success) across varying network loads and entanglement probabilities. Overall, AEG offers a scalable, real-time solution that significantly enhances quantum routing efficiency and resource utilization in practical networks.

Abstract

Entanglement generation in long-distance quantum networks is a difficult process due to resource limitations and the probabilistic nature of entanglement swapping. To maximize success probability, existing quantum routing algorithms employ computationally expensive solutions (e.g., linear programming) to determine which links to entangle and use for end-to-end entanglement generation. Such optimization methods, however, cannot meet the delay requirements of real-world quantum networks, necessitating swift yet efficient real-time optimization models. In this paper, we propose reinforcement learning (RL)-based models to determine which links to entangle and proactively swap to meet connection requests. We show that the proposed RL-based approach is 20x faster compared to linear programming. Moreover, we show that one can take advantage of the longevity of entanglements to (i) cache entangled links for future use and (ii) proactively swap entanglement on high-demand path segments, thereby increasing the likelihood of request success. Through comprehensive simulations, we demonstrate that caching unused entanglements leads to a 10-15% improvement in the performance of state-of-the-art quantum routing algorithms. Complementing caching with proactive entanglement swapping further enhances the request success rate by up to 52.55%.
Paper Structure (15 sections, 7 figures)

This paper contains 15 sections, 7 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. System Design of AEG
  4. Reinforcement Learning-Based Link-Level Entanglement Generation
  5. Proactive Entanglement Swapping
  6. Evaluations
  7. Evaluation Methodology
  8. Evaluation Results
  9. Entanglement Lifetime
  10. Request Count
  11. Swap Probability
  12. Entanglement Generation Probability
  13. Ablation Study
  14. Execution Time
  15. Conclusion

Figures (7)

  • Figure 1: A simple quantum network with physical links with multiple paths between nodes and entangled photon source (EPS) in the middle of every link.
  • Figure 2: Optimal path selection for request A-D depends on which links are selected to generate entanglement
  • Figure 3: A sample input and output of the Deep Q-Learning. It takes network topology, distances among nodes, requests and entanglement edges as input and output a list of potential links for entanglement generation
  • Figure 4: Impact of entanglement lifetime on the performance of AEG.
  • Figure 5: Performance comparison of REPS and AEG for varying request count, swap probability, and entanglement probability values.
  • ...and 2 more figures