Measured f-Divergences in Quantum Info

• Measured f-divergences are a quantum generalization of classical f-divergences that capture maximal statistical distinguishability via measurement.
• They admit dual variational formulations and form part of a hierarchy with standard and maximal divergences, emphasizing their operational significance.
• Applications span quantum hypothesis testing, channel reversibility, and state estimation, underpinning data-processing inequalities and resource quantification.

A measured $f$-divergence is a quantum generalization of classical $f$-divergence, capturing the maximal classical statistical distinguishability obtainable between two quantum states via measurement. Given its critical role in quantum information theory, measured $f$-divergence constitutes a cornerstone in resource quantification, hypothesis testing, reversibility analysis of quantum operations, and the foundational study of data-processing inequalities.

1. Fundamental Definition and Dual Variational Formulations

Let $f : (0,\infty) \to \mathbb{R}$ be a convex function with $f(1)=0$. Given two density operators $\rho$ and $\sigma$ on a finite-dimensional Hilbert space $\mathcal{H}$, a measurement (POVM) $\mathcal{M} = \{ M_i \}_{i \in I}$ induces probability distributions $$P^\mathcal{M}_\rho(i) = \operatorname{Tr}[M_i \rho]$$ and $f$0. The measured quantum $f$1-divergence is then

$f$2

where $f$3 denotes the classical $f$4-divergence. Rank-one projective measurements are sufficient for the supremum, i.e.,

$f$5

For operator-convex $f$6, the conjugate of $f$7, the supremum reduces to a tractable dual form over Hermitian operators: $f$8 This parallels the classical Fenchel dual for $f$9-divergences and is central to recent advances in quantum statistical estimation and recovery theory (Matsumoto, 2014, Fang et al., 11 Feb 2025).

2. Structural Properties and Hierarchies of Quantum $f$0-Divergences

Measured $f$1-divergence forms part of a hierarchy: $f$2 where $f$3 is Petz's "standard" $f$4-divergence and $f$5 is the maximal Petz-Ruskai extension. Equality $f$6 holds if and only if $f$7 and $f$8 commute or if $f$9 is at most quadratic; otherwise, strict inequality is generic. These hierarchies distinguish operational tasks by the quantumness of accessible information (Hiai et al., 2016, Matsumoto, 2014).

Key attributes:

• Monotonicity: For any CPTP map $f : (0,\infty) \to \mathbb{R}$0, $f : (0,\infty) \to \mathbb{R}$1.
• Convexity: Joint convexity in $f : (0,\infty) \to \mathbb{R}$2.
• Attainment: Under operator-convex $f : (0,\infty) \to \mathbb{R}$3, the supremum is attained at a Hermitian operator $f : (0,\infty) \to \mathbb{R}$4 solving a stationary point equation (Matsumoto, 2014).

3. Relations to Classical and Quantum Hypothesis Testing

For $f : (0,\infty) \to \mathbb{R}$5 (relative entropy), the measured divergence coincides with the Umegaki quantum relative entropy, yielding

$f : (0,\infty) \to \mathbb{R}$6

In the case of classical probability distributions or commuting quantum states, all quantum $f : (0,\infty) \to \mathbb{R}$7-divergences reduce to their classical analogs.

For Rényi divergences $f : (0,\infty) \to \mathbb{R}$8, several variants arise, including the measured (projective), sandwiched, and maximal versions, with precise relations quantifying their ordering and operational significance. Notably,

$f : (0,\infty) \to \mathbb{R}$9

Measured Rényi divergences govern the error exponents in quantum hypothesis testing under measurement constraints (Hiai et al., 2016, Matsumoto, 2014).

4. Convex-Analytic, Variational, and Operational Characterizations

Recent results provide a general convex-optimization formulation for measured $f(1)=0$0-divergence, extending variational formulas for wide classes of $f(1)=0$1 (Fang et al., 11 Feb 2025): $f(1)=0$2 Under suitable operator-convexity, optimization over POVMs reduces to a single Hermitian parameter, allowing formulation as a convex program.

For measured Rényi divergences, setting $f(1)=0$3, the variational form simplifies explicitly. For $f(1)=0$4, the measured divergence recovers the squared fidelity: $f(1)=0$5 which exactly aligns with Uhlmann's theorem. The framework extends this minimax property to general $f(1)=0$6 (with conditions on $f(1)=0$7), yielding Uhlmann-type theorems for measured divergences and connecting duals of divergence minimization to linearly constrained convex sets of states or measurements (Fang et al., 11 Feb 2025).

5. Additivity, Regularization, and Open Problems

The additivity of measured $f(1)=0$8-divergences generally fails:

$f(1)=0$9

Regularized limits, $\rho$0, are studied to remedy this for asymptotic analyses.

Open questions include:

• Characterizing all $\rho$1 for which $\rho$2 in full generality (settled only for Umegaki and Rényi cases).
• Sharpening necessary and sufficient conditions for supremum attainment in the dual formulation.
• Fully developing sharp upper and lower bounds on measured $\rho$3-divergences under constraints, relevant for finite-copy or resource-constrained tasks (Matsumoto, 2014).

6. Applications in Quantum Information Theory

Measured $\rho$4-divergences impact several domains:

• Quantum channel reversibility: $\rho$5 is preserved by a CPTP map $\rho$6 if and only if $\rho$7 is reversible on the input pair $\rho$8, i.e., there exists a CPTP Recovery map $\rho$9 with $\sigma$0, $\sigma$1 (Hiai et al., 2016).
• Hypothesis testing: The trade-off between Type-I and Type-II error rates under restricted observables is characterized by the measured Rényi divergences.
• Quantum thermodynamics: Coarse-grained distinguishability and resource monotones can be modeled by measured $\sigma$2-divergences.
• Statistical estimation: Variational and dual representations of measured $\sigma$3-divergences enable their numerical and algorithmic implementation in quantum state or process tomography (Fang et al., 11 Feb 2025).

7. Example Table: Comparison of Classical and Measured $\sigma$4-Divergences

Divergence	Classical Probability	Quantum (Measured $\sigma$5-divergence)
Kullback-Leibler	$\sigma$6	$\sigma$7
Rényi $\sigma$8	$\sigma$9	$\mathcal{H}$0
Trace distance	$\mathcal{H}$1	$\mathcal{H}$2

This comparison highlights how measured $\mathcal{H}$3-divergence extends the operational meaning of classical divergences to the quantum regime, with supremization over all measurements delivering the maximal classical information extractable from quantum systems (Matsumoto, 2014, Hiai et al., 2016, Fang et al., 11 Feb 2025).

---

Measured $\mathcal{H}$4-divergences thus unify convex-analytic, variational, and operational perspectives on quantum statistical distances and provide a universal bridge between quantum and classical theories of statistical distinguishability. Their dual characterizations, monotonicity under quantum operations, and centrality in quantum reversibility and resource theory underscore their broad foundational and applied significance (Hiai et al., 2016, Matsumoto, 2014, Fang et al., 11 Feb 2025).