arXiv:2610.26900
Concurrent Entanglement Purification Scheduling in Dynamic Quantum Networks
Gongyu Ni, Lester Ho
Abstract
Quantum networks must deliver end-to-end entanglement whose fidelity meets application requirements, which typically requires entanglement purification along multi-hop paths. Existing scheduling approaches either purify at the link level or optimize end-to-end fidelity for a static set of requests, and do not capture concurrent requests competing for network resources over time. This paper studies concurrent entanglement purification scheduling in a time-slotted quantum network with heterogeneous links, in which requests compete for link capacity and quantum memories, and each request may be served by multiple parallel paths for increased reliability. We formulate a mixed-integer linear programming (MILP) scheduler that maximizes the number of scheduled requests, with max-min path allocation as a fairness criterion, and develop reinforcement learning (RL) schedulers based on DQN, PPO, and SAC, whose rewards balance request success against Bell-pair consumption. Simulations under varying network load, topologies and target fidelity thresholds show that the RL schedulers achieve the highest success rates, improving on MILP by up to 6% and on Progressive Filling and Greedy heuristics by 16.5% and 19.5% respectively, at the cost of higher Bell pair consumption and the deferral of a small fraction of requests. MILP, in contrast, achieves the highest fairness and the lowest Bell pair consumption in most scenarios.
AI Summary
This paper studies how to schedule concurrent entanglement requests in a time-slotted quantum network when heterogeneous links have limited channel capacity and associated memories. Requests can receive multiple parallel paths as redundant delivery attempts, while unallocated requests are deferred. The objective is to balance delivery above an end-to-end fidelity threshold against resource consumption, fairness, and latency. Candidate routes and hop-level BBPSSW purification plans are computed before allocation. A MILP maximizes the number of admitted requests, with minimum path allocation as a secondary fairness objective. Double DQN, PPO, and SAC schedulers instead sequentially select paths using graph and path features, with rewards combining stochastic delivery success and Bell-pair consumption. Simulations include loss, channel and memory noise, imperfect operations, purification, and swapping. Evaluation uses one 20-node topology, 1,000 slots per scenario, and variations in traffic, link removal, and fidelity threshold. RL success rates are $0.84$–$0.92$, exceeding MILP by roughly 4–5 percentage points, or $4.9$–$6.0\%$ relatively. RL consumes $8$–$16\%$ more Bell pairs per successful request than MILP and has lower path-allocation fairness. Median latencies are similar, but deferrals raise RL 99th-percentile latency to approximately 8–9 ms versus about 1 ms for non-RL methods. Stated limitations include training and evaluation on the same topology, network-size-dependent path representations, and the need for more detailed hardware validation. Waiting-time-aware rewards and a probability-aware MILP are identified as future work; the latter would help separate learned path-quality benefits from differences in scheduling objectives.
Error found: The documented PPO loss minimizes the clipped policy surrogate instead of its negative, reversing the intended policy-improvement direction. The PPO entropy expression omits the conventional minus sign while retaining subtraction in the minimized total loss, so the stated regularizer penalizes entropy. The purification timing is specified inconsistently as 10 microseconds per round and 10 microseconds per kilometre, and the prose timing model does not reproduce the reported median latencies.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.