Purification Strategy Optimization for Entanglement Routing in Quantum Networks

Abstract: Quantum networks rely on the efficient distribution of entanglement to enable long-distance quantum communication and information processing. A key challenge in these networks is the design of routing protocols capable of maintaining high quality entanglement in the presence of noise, decoherence, and imperfect operations, which progressively degrade the fidelity of entangled states through entanglement swapping. Entanglement purification provides an effective mechanism to mitigate this degradation at the cost of additional resources. In this work, we study purification-aware quantum routing and formulate the problem of selecting optimal purification strategies as an optimization task. By employing dynamic programming techniques, we identify strategies that optimally balance resource consumption and end-to-end fidelity, demonstrating the effectiveness of our approach across different scenarios.

• The paper formulates purification-aware routing as a constrained integer optimization problem and uses memoization-based dynamic programming to evaluate paths of up to 200 nodes instead of exhaustively searching an exponential strategy space.
• The results show that early, aggressive purification maximizes end-to-end fidelity at low decoherence, while increasing decoherence makes purification counterproductive because its time costs outweigh fidelity gains.
• The paper identifies an apparent transition near γ·t ≈ 1, where optimal strategies stabilize, while noting that probabilistic operations, alternative protocols, path selection, and joint time–fidelity optimization remain unresolved.

Overview and motivation

This paper addresses a specific gap in quantum network routing research: while prior work has optimized entanglement generation and swapping strategies, the role of entanglement purification within routing decisions has largely been ignored. The authors, from the atlanTTic Research Centre at Universidade de Vigo, formulate purification-aware routing as an integer optimization problem: given a source–destination (SD) pair, a time threshold $\hat{T}$, and a fidelity threshold $\hat{F}$, select the purification strategy — a vector specifying how many purification rounds to apply to each entangled pair along a path — that maximizes end-to-end (E2E) fidelity subject to memory, time, and fidelity constraints. Their central contribution is a dynamic programming (DP) approach based on memoization that makes this optimization tractable for paths of up to 200 nodes, where brute-force search would require evaluating $(n_{max}+1)^{2N-3}$ candidate strategies.

The work positions itself explicitly against Huang et al.'s decoherence-aware framework (Huang et al., 16 Oct 2025), which optimizes entangling and swapping strategies but deliberately avoids computing overall fidelity through auxiliary functions that do not depend on it. That simplification, the authors argue, is incompatible with purification: the E2E fidelity under purification protocols takes values that cannot be bounded by the max/min arguments used there. This paper therefore restricts attention to the purification subproblem, adopting the known result that skewed (source-first) swapping strategies maximize fidelity.

System model

The network is modeled as a graph $G=(V,E)$ with per-node memory limits, discrete time slots, and a fixed path found by an arbitrary path-finding algorithm. Entangled links are described by Werner states whose fidelity degrades via decoherence following $F_d(F,t) = F \cdot e^{-(\gamma t)^\kappa}$ with $\kappa=1$. Swapping fidelity follows the standard model $F_s(F_1,F_2) = F_1 F_2 + \frac{1}{3}(1-F_1)(1-F_2)$, and purification uses the recursive BBPSSW protocol [Bennett et al., 1996], which improves fidelity only for inputs above 0.5:

$F'(F) = \frac{10F^2 - 2F + 1}{8F^2 - 4F + 5}.$

A notable structural result is that the purification strategy vector $S$ has dimension $2N-3$ for an $\hat{F}$0-node path: $\hat{F}$1 components for the elementary links plus $\hat{F}$2 for the long-range pairs created by swapping. The authors prove this by decomposing $\hat{F}$3 into symmetric-base and asymmetric parts. This dimensionality is what drives the exponential blow-up of the strategy space: $\hat{F}$4 candidates must be enumerated exhaustively.

The optimization problem is stated as an integer program over binary selection variables $\hat{F}$5 choosing one numerology $\hat{F}$6 (a resource-allocation tuple pairing each entangled pair with its purification count and time slots) per request, subject to memory capacity, single-strategy selection, deadline, and fidelity-threshold constraints. The authors note the problem is NP-hard and that the relaxations used in prior work are not applicable here because purification breaks the fidelity bounds those relaxations rely on.

Solution method

Two computational approaches are presented. First, a brute-force Python implementation enumerates all strategies for small instances, establishing baseline behavior and confirming exponential scaling. Second, and more importantly, a DP scheme exploits the observation that extending a path from $\hat{F}$7 to $\hat{F}$8 nodes adds only two new entanglements without disturbing previously computed optimal strategies. Memoization therefore allows optimal strategies for larger paths to be built incrementally from smaller ones, avoiding redundant recomputation in the recursive structure of repeater chains.

Simulation parameters are drawn from experimental literature: initial link fidelities sampled uniformly from $\hat{F}$9, 1 ms time slots consistent with solid-state network demonstrations, operation probabilities set to 1.0 (justified by reported real-world values of 0.9–0.95), and decoherence rates $(n_{max}+1)^{2N-3}$0 spanning from near zero to well above 1000 Hz. Paths were computed from $(n_{max}+1)^{2N-3}$1 up to $(n_{max}+1)^{2N-3}$2 nodes.

Results

The principal empirical findings are as follows:

• Purify early and aggressively when decoherence permits. For low decoherence rates ($(n_{max}+1)^{2N-3}$3 Hz or below), the optimal strategy applies the maximum number of rounds to the initial entanglements, then progressively fewer rounds on later ones. A representative run at $(n_{max}+1)^{2N-3}$4 Hz yields strategies of the form $(n_{max}+1)^{2N-3}$5: full purification on the first three-link segment, one round on the next 45 components, and no purification thereafter.
• Decoherence dominates purification gains. As $(n_{max}+1)^{2N-3}$6 increases toward 20–100 Hz, purification becomes counterproductive because BBPSSW rounds consume time during which exponential decoherence erodes more fidelity than purification restores; the optimum shifts toward minimal or no purification.
• A phase transition near $(n_{max}+1)^{2N-3}$7. When the product of decoherence rate and slot duration approaches unity, fidelity drops by a factor $(n_{max}+1)^{2N-3}$8 per slot, and the optimal strategy vectors "freeze" — runs at $(n_{max}+1)^{2N-3}$9 and $G=(V,E)$0 Hz produce essentially identical strategies. The authors interpret this freezing of strategy components across parameter variation as evidence of a phase transition in the optimization landscape, drawing an analogy to thermodynamic critical points.

An important caveat the authors state plainly: different strategies can achieve equal or near-equal E2E fidelities, so there is no unique absolute best strategy in each regime. They also acknowledge that the choice to maximize fidelity alone means final completion time $G=(V,E)$1 is not jointly optimized, which they deem impossible within their formulation.

Limitations and open questions

Several limitations constrain the generality of these results. The model assumes deterministic operation success probabilities of 1.0, whereas real swapping and purification succeed probabilistically (0.9–0.95); the authors show only trivially that lower probabilities reduce performance. Only the BBPSSW protocol is evaluated, though the framework is claimed to be protocol-agnostic. The analysis is restricted to a single fixed path with a fixed skewed swapping schedule, so joint optimization over path selection, swapping order, and purification remains open. The phase transition phenomenon is identified empirically but not characterized analytically. Finally, the authors concede that realistic environmental interactions are more complex than the single-parameter decoherence model, and that practical networks involve few nodes with improving quantum memories — regimes where the interesting high-$G=(V,E)$2 behavior may be less relevant.

Conclusion

This paper formulates purification strategy selection in entanglement routing as a constrained integer optimization problem and solves it at scale using memoization-based dynamic programming, reaching 200-node paths that would be infeasible by exhaustive search. The consistent qualitative conclusion is that purification should occur as early as possible in an entanglement's lifetime, with intensity governed by the decoherence rate, and that beyond a critical $G=(V,E)$3 regime the optimal strategy undergoes an apparent phase transition. The results provide actionable guidance for purification-aware routing protocol design, while leaving open the integration of path finding, probabilistic operations, alternative purification protocols, and an analytical treatment of the observed transition.