Generalized Numerical Framework for Improved Finite-Sized Key Rates with Rényi Entropy

1. Problem Statement

Quantum Key Distribution (QKD) security proofs rely on bounding the secret key rate. In practical, finite-size regimes (where the number of signals $N$ is limited), traditional methods using the von Neumann entropy often yield loose bounds, leading to suboptimal key rates.

• The Gap: While recent theoretical advances (e.g., Dupuis [21]) showed that generalized Rényi entropies could provide tighter bounds via a Rényi leftover hashing lemma, optimizing these quantities is computationally non-trivial.
• The Challenge: Existing numerical frameworks (e.g., Winick et al. [19]) are designed for von Neumann entropy. Adapting them to optimize Rényi entropy requires deriving analytical gradients for Rényi divergence and handling the specific mathematical properties of sandwiched Rényi divergences, which has remained undeveloped.
• Target Application: This is critical for long-distance, satellite-based QKD protocols where channel loss is high and block sizes are small ($N \sim 10^5$), regimes where von Neumann bounds often fail to produce positive key rates.

2. Methodology

The authors propose a generalized numerical framework that extends the semi-definite programming (SDP) approach of Winick et al. [19] to optimize Rényi entropic quantities.

A. Theoretical Formulation

1. Rényi Leftover Hashing Lemma: The secret key length $l$ is formulated using the Rényi leftover hashing lemma (Theorem III.1), which bounds the trace distance between the final key and a uniform distribution using the conditional Rényi sandwiched entropy $H_\alpha$.
2. Entropy-Divergence Bound: To make optimization tractable, the authors derive a lower bound for the conditional Rényi entropy in terms of the sandwiched Rényi divergence (Theorem III.3).
   • They utilize the duality relation to express $H_\alpha(Z_A|E')$ as an infimum of a Rényi divergence $D_\gamma$.
   • Crucially, they relax this to a single-round bound involving $D_\beta$ (where $\beta = \alpha^{-1}$) to remove the infimum, allowing for direct optimization:
     $$H_\alpha(Z_A|E') \geq D_\beta(G(\rho_{AB}) || Z \circ G(\rho_{AB}))$$
3. Objective Function: The optimization goal becomes minimizing the function:
   $$f_\beta(\rho_{AB}) = \text{Tr}(G(\rho_{AB})) \cdot D_\beta(G(\rho_{AB}) || Z \circ G(\rho_{AB}))$$
   subject to constraints on the state space $\mathcal{S}$ (modeling parameter estimation and finite-size effects).

B. Numerical Algorithm

The framework employs the Frank-Wolfe algorithm (conditional gradient method) to minimize the objective function.

• Analytical Gradient Derivation: A key technical contribution is the derivation of the analytical gradient of the sandwiched Rényi divergence (Theorem III.4).
  • The gradient $\nabla D_\beta$ involves complex integral representations of operator functions (using the integral representation of fractional powers).
  • The authors provide explicit formulas for terms $\chi_1, \chi_2, \chi_3$ involving integrals over the spectrum of the operators.
• Two-Step Optimization:
  1. Step 1: Use Frank-Wolfe to find a state $\hat{\rho}$ close to the optimum by iteratively solving SDPs for the descent direction $\Delta \rho$.
  2. Step 2: Linearize the objective function around $\hat{\rho}$ to obtain a rigorous lower bound on the optimal key rate.
• Finite-Size Constraints: The framework incorporates constraints from George et al. [20] to model parameter estimation errors ($\mu$) and statistical deviations, ensuring the bounds hold for finite $N$.

3. Key Contributions

1. Generalized Framework: Successfully generalized the state-of-the-art numerical framework (Winick et al.) to handle generalized Rényi entropies rather than just von Neumann entropy.
2. Analytical Gradient: Derived the first analytical expression for the gradient of the sandwiched Rényi divergence, enabling the use of efficient gradient-based optimization algorithms (Frank-Wolfe) for Rényi key rates.
3. Tighter Bounds: Proved that bounding Rényi entropy directly via Rényi divergence yields tighter lower bounds on the key rate compared to previous methods that relied on von Neumann entropy approximations in the finite-size regime.
4. Open-Source Implementation: Integrated the method into the openQKDsecurity package, making the tool available for the community.

4. Results

The authors conducted numerical simulations on a Prepare-and-Measure (PM) BB84 protocol under depolarizing and lossy channels.

• Optimization of $\alpha$: The key rate depends on the Rényi parameter $\alpha$. The study found that for finite block sizes, there exists an optimal $\alpha$ that maximizes the key rate. As $N \to \infty$, the optimal $\alpha$ converges to 1 (recovering the von Neumann limit).
• Performance in Low Block Sizes ($N < 10^7$):
  • For $N = 10^5$, the Rényi-based key rate is approximately double the von Neumann key rate.
  • The Rényi framework maintains positive key rates in regimes where the von Neumann bound drops to zero.
• Performance in High Loss:
  • In high-loss scenarios (relevant for satellite QKD), the Rényi key rate is significantly more robust. The von Neumann key rate decays rapidly to zero as loss increases, whereas the Rényi bound sustains a non-zero key rate for longer distances.
• Convergence: The numerical results confirm that the improvements are most pronounced in the "finite-size" and "high-loss" regimes, which are the most challenging for practical QKD deployment.

5. Significance and Outlook

• Practical Impact: This work provides a critical tool for evaluating the security of long-distance satellite-based QKD. These systems inherently operate with small block sizes and high channel loss, exactly where the proposed framework offers the most significant performance gains.
• Theoretical Advancement: It bridges the gap between theoretical Rényi entropy bounds and practical numerical optimization, moving beyond the asymptotic limits of von Neumann entropy.
• Future Directions:
  • The authors note that the current method requires perturbing the gradient to ensure continuity (using a depolarizing channel), which introduces slight numerical imprecision. Future work could integrate interior-point methods (e.g., Gaussian interior point solvers) to optimize Rényi divergences directly without perturbation.
  • The framework could be extended to decoy-state protocols under coherent attacks, a direction already being explored by independent researchers.

In summary, this paper establishes a robust, generalized numerical framework that leverages Rényi entropy to significantly improve finite-size key rate estimates, offering a vital advancement for the deployment of secure, long-distance quantum communication networks.