Jigyen Bhavsar, Rajni Bala, and Siddhartha Santra of the Indian Institute of Technology Bombay formulated and systematically analyzed entanglement distillation as a Markov decision process and obtained optimal deterministic policies that minimize the expected waiting time required to achieve a target fidelity. The researchers formulated entanglement distillation as a Markov decision process and used a value iteration algorithm to identify optimal strategies that minimize the time needed to achieve a target level of entanglement fidelity.

Their results show the expected waiting time decreases as the probability of generating entanglement increases, and with a larger number of quantum memories. This approach enables the systematic design of policies for quantum information tasks in realistic, resource-constrained settings.

Markov Decision Process Formulation of Entanglement Distillation

Researchers formulated a Markov decision process (MDP) to systematically identify the optimal sequence of operations needed to boost entanglement fidelity, minimizing the time required to reach a target level. This approach moves beyond existing policies like entanglement pumping and banded purification by offering a framework adaptable to realistic resource limitations. The study models entanglement distillation as an MDP defined by a four-tuple encompassing system states, possible actions, transition probabilities and reward functions. This allows for a systematic computation of the optimal distillation policy.

The team’s formulation represents the interaction between an agent and a system comprising two parties, Alice and Bob, each possessing at least two quantum memories. This MDP framework allows the researchers to use the Bellman equation and a value iteration algorithm to determine the policy that minimizes expected waiting time, an important metric for practical quantum communication.

The analysis reveals that the expected waiting time under the optimal policy decreases with both increasing generation probability, denoted as ‘p’ and the number of quantum memories, ‘m’. The performance advantage of the optimal policy is dependent on the difference in fidelity between the initial and target states, as well as the system parameters of ‘p’, ‘m’ and the initial fidelity ‘ $f_{0}$ ’. The team acknowledges that accounting for operational imperfections, such as measurement errors, represents an important next step.

The data supporting these findings are available upon request, allowing other researchers to validate and build upon this work. The team’s work provides a pathway toward designing policies for quantum information tasks under realistic, resource-constrained conditions, potentially accelerating the development of practical quantum technologies.

Entanglement Distillation and Fidelity Improvement with LOCC

The work demonstrates that, contrary to typical expectations, adding resources does not necessarily increase complexity in this scenario, but instead directly accelerates the process of fidelity improvement. This approach focuses on upgrading existing entanglement, beginning with states possessing an initial fidelity, $f_{0}$ , to a target fidelity, $f_{T}$ , where $f_{T}$ is greater than $f_{0}$ . The study considered scenarios where entanglement generation attempts occur with probability ‘p’, resulting in Werner states of raw fidelity $f_{0}$ , and where parallel generation operations are permitted at each step.

The researchers constructed a distillation matching greedily, pairing links of initial fidelity to maximize fidelity enhancement, and even pairing low-fidelity links amongst themselves when higher-fidelity partners were unavailable. A key finding concerns the relationship between system resources and performance. Figures show the expected waiting time of the optimal policy, $T_{o} p t$ , decreases monotonically with increasing ‘m’.

For a fixed generation probability, a larger ensemble of memories expands the state space, providing more opportunities for simultaneous entanglement generation and parallel distillation, ultimately enabling faster achievement of the target fidelity. The expected waiting time exhibits non-monotonic behavior with respect to $f_{0}$ for a fixed fidelity gap, (Δ f = f_T-f_0). The research also highlights the limitations of a simple “pumping” policy, where low-fidelity pairs are repeatedly used to incrementally improve higher-fidelity links.

This method, while memory-efficient, becomes increasingly ineffective when the difference between the initial and target fidelities is large. The team also explored a “nested” policy, where distillation is performed only between pairs of identical fidelity, as an alternative strategy. This complex interplay between parameters highlights the subtle nature of optimizing entanglement distillation and the benefits of a systematic approach to policy design.

Impact of Generation Probability and Memory Pairs on Distillation Time

The efficiency of entanglement distillation improves as the number of available quantum memories increases, a finding that challenges the expectation that more resources invariably complicate quantum processes. Parallel entanglement generation operations use these additional memories, allowing for simultaneous attempts across unused pairs, and accelerating the overall distillation process. The ability to perform these generation attempts in parallel across all inactive links, alongside multiple simultaneous 2-to-1 entanglement distillation operations, is central to minimizing the overall waiting time.

Each quantum memory, however, can only participate in one operation per step, establishing a key resource constraint. The formulation of this problem as a Markov decision process allows for systematic optimization of distillation policies, identifying the sequence of entanglement generation and distillation operations that minimizes expected waiting time.

Non-Monotonic Behavior of Initial Fidelity on Waiting Time

The expected waiting time exhibits a non-monotonic relationship with the initial fidelity, peaking at both low and high starting points. However, this benefit is most pronounced when the initial fidelity falls within a specific range, rather than consistently improving with higher initial quality. This complex interplay between parameters highlights the detailed nature of optimizing entanglement distillation and the benefits of a systematic approach to policy design.

Optimal Policies Outperform Baseline Distillation Strategies

The performance of optimal entanglement distillation policies is intricately linked to the fidelity gap, Δf, though the exact relationship depends on both the baseline policy employed and system parameters like generation probability and the number of quantum memories. The study demonstrates that the advantage gained by employing an optimal policy isn’t uniform across all fidelity gaps, requiring detailed consideration of system configuration.

Computational challenges arise when exploring regimes with larger fidelity gaps due to the rapid expansion of the state space during the value iteration algorithm, detailed in Appendix A. To benchmark the effectiveness of these newly developed optimal policies, the researchers compared their performance against several established heuristic strategies, including greedy, entanglement pumping and nested purification. These baseline policies, while simpler to implement, exhibit strong performance only within specific parameter ranges encompassing the number of memories, generation probability, initial fidelity and target fidelity.

The team found that with an unlimited supply of quantum memories, the optimal policy converges to the well-known nested purification protocol, a standard technique for enhancing entanglement quality. The investigation extends beyond simply identifying optimal policies; it also explores how these policies manifest in practice. Figure 2(d) illustrates a possible history of an optimal policy for parameters of p = 0.95, $f_{0}$ = 0.648, and $f_{T}$ = 0.710, providing a visual representation of the decision-making process.

The work employs a dynamic programming approach, specifically the value iteration algorithm, to determine these policies, with a more detailed explanation of the method available in Appendix An and a referenced chapter. Is a strategy that the optimal policy can surpass, depending on the specific system parameters. This approach contrasts with relying on established heuristics that may be suboptimal outside their intended operating regimes.

The researchers formalized the entanglement distillation policy optimization problem within this framework, building upon a model of entanglement generation and distillation where two parties, Alice and Bob, each use multiple quantum memories connected by parallel quantum channels. The availability of these channels allows for probabilistic entanglement generation, an important element in the distillation process.

Entanglement Generation and Distillation Operation Constraints

The efficiency of entanglement distillation is directly linked to the number of quantum memories available at each node, a key constraint in practical implementations. Parallel execution of entanglement generation and distillation operations is possible, provided no single memory participates in more than one operation simultaneously, maximizing resource use. This allows for simultaneous entanglement generation across all inactive links, alongside multiple distillation operations performed on stored entangled pairs, a strategy that accelerates the process.

The time cost of distillation extends beyond local operations, incorporating the time needed for measurement outcome exchange between parties and reporting results, adding a communication overhead of 2l/c to each distillation action. The formulation of this problem as a Markov decision process allows for systematic optimization of distillation policies, identifying the sequence of entanglement generation and distillation operations that minimizes expected waiting time.

Source: https://iopscience.iop.org/article/10.1088/2058-9565/aeb8ee

More like this