Generalized Entropy Accumulation Theorem

• GEAT is a framework in quantum information that bounds accumulated entropy in sequential processes with quantum side information under a non-signalling condition.
• It generalizes the original EAT by relaxing structural constraints, thereby broadening its applicability to cryptographic protocols like QKD and randomness expansion.
• The theorem employs an iterative chain rule with finite-size corrections and convex optimization techniques to ensure rigorous, operational security guarantees.

The Generalized Entropy Accumulation Theorem (GEAT) is a framework in quantum information theory that bounds accumulated operational entropy in sequential processes with quantum side information under a non-signalling condition. GEAT generalizes the original Entropy Accumulation Theorem (EAT) by relaxing structural constraints on side information registers, broadening the range of cryptographic and information-processing protocols to which entropic security proofs can be rigorously applied.

1. Formal Statement and Setup

The GEAT addresses sequential processes formed by completely positive trace-preserving (CPTP) maps that generate outputs $A_1, \dots, A_n$ and update side-information registers $E_1, \dots, E_n$ via

$$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$

where $R_i$ denotes hidden or memory registers. The initial state $\rho^0_{R_0 E_0}$ evolves to a final state $$\rho_{A^n E_n} = (\mathcal{M}_n \circ \dots \circ \mathcal{M}_1)(\rho^0_{R_0 E_0})$$.

A critical assumption is the non-signalling condition: for each $i$, there exists a CPTP map $$\mathcal{F}_i\colon\mathcal{L}(E_{i-1}) \to \mathcal{L}(E_i)$$ such that

$$\mathrm{Tr}_{A_i R_i}\circ\mathcal{M}_i = \mathcal{F}_i\circ\mathrm{Tr}_{R_{i-1}}.$$

Intuitively, this enforces that, conditionally on discarding output $A_i$ and updated memory $E_1, \dots, E_n$0, the future side information $E_1, \dots, E_n$1 does not reveal information about the discarded memory $E_1, \dots, E_n$2 beyond $E_1, \dots, E_n$3.

The theorem lower bounds the smooth min-entropy accumulated in the output sequence conditional on the final quantum side information,

$E_1, \dots, E_n$4

in terms of single-round conditional von Neumann entropies and complements it with tight finite-size corrections (Metger et al., 2022).

2. Entropic Quantities and Chain Rule Structure

The key elements are the smooth min-entropy $E_1, \dots, E_n$5 (Renner) and the conditional von Neumann entropy $E_1, \dots, E_n$6:

• $E_1, \dots, E_n$7
• $E_1, \dots, E_n$8, with $E_1, \dots, E_n$9

The proof interpolates between sandwiched Rényi entropies,

$$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$0

which allows:

• $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$1
• $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$2

A crucial technical ingredient is an improved chain rule for the sandwiched Rényi divergence, removing certain regularization terms for conditional entropies, which enables the iteration of the chain rule across all rounds (Metger et al., 2022).

3. Theorem Statement, Finite-Size Corrections, and Operational Implications

For a sequence of $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$3 rounds satisfying non-signalling, the GEAT guarantees

$$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$4

where $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$5 is a purification ancilla for $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$6, and $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$7 are explicit constants from the chain rule and smoothing steps (Metger et al., 2022, Metger et al., 2022). Equivalently, in terms of finite-size $$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$8-smoothness and conditioning on test events,

$$\mathcal{M}_i\colon\mathcal{L}(R_{i-1} \otimes E_{i-1}) \to \mathcal{L}(A_i \otimes R_i \otimes E_i),$$9

with $R_i$0 the minimal single-round trade-off (typically optimized via SDP), and $R_i$1 comprising variance and smoothing penalties (Metger et al., 2022, Carceller et al., 2024, Kamin et al., 2024). The significance is that the total min-entropy is, up to $R_i$2, at least the sum of the single-round von Neumann conditional entropies optimized over possible inputs.

4. Comparison to Original EAT and Subsequent Generalizations

The original EAT (Dupuis et al., 2016) required a quantum Markov chain on the (classical+quantum) side information: $R_i$3 where $R_i$4 is fresh side information at each round, and only static registers are updated. The GEAT only necessitates the weaker non-signalling condition, allowing for dynamically updated side information throughout the protocol.

This strictly generalizes the original EAT: the Markov condition always implies non-signalling, but not vice versa. The relaxation is essential for cryptographically relevant protocols where an adversary can adaptively update their quantum memory—such as prepare-and-measure QKD and randomness expansion—directly in the sequential model (Metger et al., 2022, Carceller et al., 2024). As a technical consequence, GEAT enables direct, rather than artificially symmetrized, security derivations for these settings.

Marginal-constrained EAT (MEAT) (Arqand et al., 4 Feb 2025) and recent generalized Rényi GEATs (Arqand et al., 2024) employ further generalizations, including fully adaptive tradeoff functions and convex-program-based rate characterizations, eliminating affine min-tradeoff function construction altogether.

5. Applications in Quantum Cryptography

GEAT’s principal operational utility is in quantum cryptography, most notably:

• Device-independent and semi-device-independent QKD: GEAT yields security bounds without IID or symmetry assumptions, and its flexible structure supports direct analysis of prepare-and-measure and decoy-state QKD protocols (Metger et al., 2022, Kamin et al., 2024).
• Blind randomness expansion: For protocols involving untrusted quantum devices, GEAT provides the first multi-round quantitative security proofs, enabling exponential randomness expansion rates as a function of observed nonlocal game statistics (Metger et al., 2022).
• Finite-size analysis: Tight finite-size corrections permit security against fully general attacks with composable security parameters—critical for practical implementations (Kamin et al., 2024).

The process involves, for each round, numerically optimizing (typically via SDP) the conditional von Neumann entropy lower bound compatible with observed classical statistics (e.g., error rates, test outcomes), then applying the GEAT to accumulate these across rounds.

6. Key Technical Insights and Methodologies

GEAT’s proof and application pipeline is summarized as:

1. Non-signalling structural assumption: Formally encode that side information can be adaptively updated but not retro-causally signal past outputs.
2. Single-round trade-off optimization: For each round, the worst-case conditional entropy is minimized over allowed channel inputs and side information, indexed by observed data.
3. Chain rule iteration: Using a novel sandwiched Rényi entropy chain rule, lower bound total entropy as a sum of single-round contributions with regularization.
4. Smoothing and finite-size analysis: Convert Rényi bounds to smooth min-entropy, paying a penalty vanishing as $R_i$5; corrections are explicit, enabling parameter optimization.
5. Convexity and duality: In advanced formulations (Arqand et al., 2024, Kamin et al., 2024), the entire accumulation and trade-off process is cast directly as a convex optimization, eliminating the need for affine trade-off construction and admitting per-batch or per-round adaptivity.

No virtual tomography or repetition-rate limitation arises, and convex-dual interfaces allow for efficient numerical implementations for cryptographic analysis (Kamin et al., 2024).

7. Extensions, Related Work, and Outlook

Recent works build on the GEAT to handle full Rényi entropy accumulation, marginal constraints ("MEAT"), and adaptive quantum probability estimation (Arqand et al., 4 Feb 2025, Arqand et al., 2024). These extensions further generalize the framework, providing optimal finite-$R_i$6 rates, fully adaptive entropy estimation, and applicability to high-rate or arbitrarily correlated sequential protocols by leveraging $R_i$7-weighted Rényi techniques and source-replacement mappings.

Open questions include extending fully adaptive trade-off optimization into the $R_i$8 Rényi regime, optimizing rate constants by tightening embedding steps, and unifying GEAT with factorization-based estimation frameworks (Arqand et al., 2024). Empirical analyses indicate that GEAT-based security proofs already closely match the ideal IID asymptotic rates at moderate $R_i$9 (typically $\rho^0_{R_0 E_0}$0–$\rho^0_{R_0 E_0}$1 rounds) (Arqand et al., 2024, Kamin et al., 2024).

---

References:

• "Generalised entropy accumulation" (Metger et al., 2022)
• "Security of quantum key distribution from generalised entropy accumulation" (Metger et al., 2022)
• "Entropy accumulation" (Dupuis et al., 2016)
• "Finite-size analysis of prepare-and-measure and decoy-state QKD via entropy accumulation" (Kamin et al., 2024)
• "Improving semi-device-independent randomness certification by entropy accumulation" (Carceller et al., 2024)
• "Generalized Rényi entropy accumulation theorem and generalized quantum probability estimation" (Arqand et al., 2024)
• "Marginal-constrained entropy accumulation theorem" (Arqand et al., 4 Feb 2025)