One-Shot Entropies: Finite Regime Analysis

• One-shot entropies are specialized measures that quantify uncertainty in finite, non-asymptotic settings by incorporating a smoothing parameter for bounded error probabilities.
• They leverage frameworks like smooth min-/max-entropies, Rényi entropy variants, and hypothesis testing divergences to replace classical entropy measures in operational tasks.
• Their applications span quantum cryptography, channel coding, and thermodynamics, precisely characterizing achievable rates, resource costs, and finite-size error bounds.

A one-shot entropy is an entropic quantity formulated to characterize operational tasks—such as source coding, channel coding, randomness extraction, privacy amplification, quantum thermodynamics, and resource manipulations—in non-asymptotic (single-use or finite blocklength) settings. Unlike traditional information measures derived in the independent and identically distributed (i.i.d.) limit such as the Shannon or von Neumann entropy, one-shot entropies precisely capture finite-size and non-i.i.d. effects, often by introducing a smoothing parameter to allow bounded error probabilities. Major classes include the smooth min- and max-entropies, smooth hypothesis testing divergence, information spectrum divergences, and smooth Rényi entropies (including sandwiched and Petz-type). Their operational content pervades modern quantum information theory, non-equilibrium thermodynamics, quantum cryptography, and resource theories.

1. Formal Definitions and Families of One-Shot Entropies

The core mathematical structure underlying most one-shot entropies is the generalization and smoothing of Rényi-type quantities. For a quantum state ρ and reference σ (positive semi-definite operators), the principal objects are:

• Smooth max-relative entropy:

$$D_{\max}^{\varepsilon}(\rho \| \sigma) := \min_{ \tilde\rho:\,P(\tilde\rho, \rho)\leq\varepsilon } \inf\{ \lambda:\, \tilde\rho\leq 2^{\lambda}\,\sigma \}$$

where P(⋅,⋅) is the purified distance.

• Smooth min-entropy and smooth max-entropy:

$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$

$$H_{\max}^{\varepsilon}(A|B)_{\rho} = \inf_{ \rho' \in B^{\varepsilon}(\rho) } H_{\max}(A|B)_{\rho'}$$

where the non-smooth forms are

$$H_{\min}(A|B)_\rho = \sup_{ \sigma_B } \{ \lambda : 2^{-\lambda} I_A \otimes \sigma_B \geq \rho_{AB} \}$$

$$H_{\max}(A|B)_\rho = \sup_{ \sigma_B } \log F(\rho_{AB}, I_A \otimes \sigma_B)^2$$

These duals encapsulate the min/max extractable information and recovery cost in a one-shot scenario (Vitanov et al., 2012).

• Smooth hypothesis testing divergence:

$$D_{H}^{\varepsilon}(\rho \| \sigma) = -\log \inf \{ \operatorname{Tr}[ \Lambda \sigma ] : 0 \leq \Lambda \leq I, \operatorname{Tr}[ \Lambda \rho ] \geq 1-\varepsilon \}$$

• Smooth information-spectrum divergence:

$$D_{s}^{\varepsilon}(\rho \| \sigma) = \sup \{ \gamma : \operatorname{Tr}[ \{ \rho>2^{\gamma} \sigma \} \rho ] > 1-\varepsilon \}$$

(Spectral projections and tail bounds.)

• Smooth Rényi entropy (classical case):

$$H_{\alpha}^{\varepsilon}(X) = \frac{1}{1-\alpha} \log \inf_{Q \in B^\varepsilon(P_X)} \sum_x Q(x)^\alpha$$

with the smoothing set $B^\varepsilon(P_X)$ defined via an $\varepsilon$-mass truncation (Sakai et al., 2020, Warsi, 2015).

• Sandwiched Rényi relative entropy (quantum):

$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$0

with the corresponding (conditional) sandwiched Rényi entropy (Cheng et al., 2024).

2. Operational Interpretations and Roles

One-shot entropies directly characterize achievable rate regions, resource costs, and finite-blocklength effects in diverse operational settings:

• Channel and source coding: Smooth min- and max-entropies replace Shannon/von Neumann entropy in single-use or finite-n regimes, determining compression, transmission, and randomness extraction rates with bounded error probabilities (Datta et al., 2011, Wakakuwa et al., 2020).
• Randomness extraction/Privacy amplification: The extractable number of uniform bits in the presence of quantum side information is tightly characterized by smooth min-entropy, with improved bounds using measurement-smooth Rényi divergences (Regula et al., 4 Mar 2026, Wang et al., 2024).
• Quantum decoupling and state merging: Achievable error in decoupling protocols, required for state merging and quantum communication, decays exponentially with a sum of smooth conditional entropies, typically min- and max-entropy (or sandwiched Rényi extensions) (Dupuis et al., 2010, Cheng et al., 2024).
• Resource theories: Minimum one-shot conversion cost and maximal distillation yield of quantum resources are governed by smooth max-/min-relative entropy and hypothesis testing divergence, modulated by theory-dependent coefficients when using a "currency" state (Liu et al., 2019).
• Thermodynamics/Statistical Mechanics: One-shot min/max and Rényi entropies quantify the worst-case and $$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$1-guaranteed work in nonequilibrium fluctuation relations and small-system thermodynamics (Garner, 2018, Weilenmann et al., 2015).

3. Mathematical Structure and Properties

Key mathematical features of one-shot entropies include:

• Smoothing: To robustly model errors or fluctuations, entropic quantities are optimized over a ball (in trace, purified or variational distance) centered at the actual state. This ensures operational relevance for tasks with nonzero failure probability (Vitanov et al., 2012).
• Monotonicity (Data Processing Inequality): For any CPTP map, $$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$2, and similarly for max-entropy (Wakakuwa et al., 2019, Wakakuwa et al., 2020).
• Duality: For any pure tripartite state $$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$3,

$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$4

which enables sharp reductions and chain rules (Wakakuwa et al., 2019, Dupuis et al., 2010).

• Chain rules and additivity: Precise addition/subadditivity inequalities relate the joint entropy of subsystems to that of the parts; equalities often become inequalities or have additive corrections in one-shot regimes (Vitanov et al., 2012, Wakakuwa et al., 2020).
• Asymptotic Expansion: In the i.i.d. limit with vanishing error parameter, smooth one-shot entropies converge to standard information measures (Shannon/von Neumann entropy), with optimal second-order (normal approximation) corrections available via central limit theorem analogues (Sakai et al., 2020).

4. Core Technical Results and Inequalities

Several pivotal technical advances underpin the theory:

• Minimax reduction: A variational characterization of smoothed max-divergences allows "commuting" quantum smoothing with classical test optimization. This yields tight inequalities linking smooth max-relative, hypothesis testing, and information spectrum divergences (Anshu et al., 2019).
• Direct = Converse up to smoothing: In most one-shot settings, achievability and converse (impossibility) bounds match modulo logarithmic additive corrections in the smoothing parameter. This enables practically tight finite-blocklength analysis (Anshu et al., 2017, Dupuis et al., 2010).
• Exponential error decay without smoothing: Using Rényi exponents (e.g., sandwiched Rényi entropy), one obtains strong one-shot error exponent bounds, realizing strict exponential decoupling/coding convergence even without explicit smoothing (Cheng et al., 2024).
• Measurement-based smoothing: Measurement-lifted smoothing yields the tightest known privacy amplification and decoupling bounds, outperforming prior purified distance smoothing (Regula et al., 4 Mar 2026).

5. Examples and Applications

A selection of domains where one-shot entropies play a definitive role:

Domain	One-Shot Entropy Used	Operational Role/Result
Quantum key distribution (BB84)	$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$5	Secure key rate in finite-blocklength regime (Wang et al., 2024)
Coherence/entanglement theory	$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$6	Minimum formation cost, distillation yield (Liu et al., 2019)
Quantum channel coding	$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$7	Achievable rate and exponential error in one-shot (Wakakuwa et al., 2020, Cheng et al., 2024, Yuan, 2018)
Fluctuation theorems (thermo)	$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$8	$$H_{\min}^{\varepsilon}(A|B)_{\rho} = \sup_{ \rho' \in B^{\varepsilon}(\rho) } H_{\min}(A|B)_{\rho'}$$9-guaranteed worst-case work (Garner, 2018)
Randomness extraction/privacy amp	$$H_{\max}^{\varepsilon}(A|B)_{\rho} = \inf_{ \rho' \in B^{\varepsilon}(\rho) } H_{\max}(A|B)_{\rho'}$$0, smoothed Rényi	Tight extractor bounds against quantum side info (Regula et al., 4 Mar 2026)

Each is mapped precisely to application-specific error criteria and protocol constraints; the table demonstrates both the universality and nuance provided by the one-shot paradigm.

6. Extensions and Research Directions

Current developments and open questions:

• Strong converse exponents: Understanding conditions under which one-shot exponent bounds yield strong converses for transmission and secrecy (Regula et al., 4 Mar 2026).
• Generalized smoothing methods: Lifting smoothing to Hermitian operators, super-operators, or measurement sets broadens applicability and sharpens bounds (Regula et al., 4 Mar 2026, Cheng et al., 2024).
• Resource theories of quantum channels: Channel-based one-shot entropies (e.g., $$H_{\max}^{\varepsilon}(A|B)_{\rho} = \inf_{ \rho' \in B^{\varepsilon}(\rho) } H_{\max}(A|B)_{\rho'}$$1 for channels) extend operational quantification from states to general quantum processes (Yuan, 2018).
• Thermodynamic and beyond-i.i.d. analysis: One-shot entropies provide single-shot analogues of free energy, majorization, and non-equilibrium resource conversion rates, supporting the modern resource-theory approach to quantum thermodynamics (Weilenmann et al., 2015, Garner, 2018).
• Explicit asymptotic expansions: Second-order expansions and refined normal approximations for various smooth Rényi entropies enable tightly calibrated coding theorems in both average- and maximum-error formalisms (Sakai et al., 2020, Warsi, 2015).

7. Foundational and Conceptual Perspective

One-shot entropies represent the pivot of the operational information-theoretic approach in the single-use and finite-blocklength regime. They reinterpret classical and quantum information measures as quantities directly meaningful for protocols under finite resources, uncertainty, and non-asymptotic constraints. Their canonical status is reinforced by their appearance both as optimal protocol rates and as the unique monotones in resource theories and nonequilibrium thermodynamics, heralding a unified mathematical language for quantum, classical, and hybrid tasks (Weilenmann et al., 2015, Liu et al., 2019).