Quantum Logarithmic Relative Entropy

• Quantum logarithmic relative entropy is defined for density matrices as D(ρ∥σ)=Tr[ρ(logρ - logσ)] and generalizes the classical Kullback–Leibler divergence.
• It underpins operational results in hypothesis testing, channel capacity, and many-body physics through its data-processing inequality and monotonicity properties.
• Recent computational methods leverage semidefinite programming relaxations and variational algorithms to efficiently estimate this divergence on modern quantum devices.

Quantum logarithmic relative entropy, frequently referred to as Umegaki relative entropy, is a fundamental quantum information theoretic divergence that generalizes the classical Kullback–Leibler divergence to the non-commutative setting. It quantifies the distinguishability of two quantum states (density matrices) and underpins operational and structural results throughout quantum information science, mathematical physics, and quantum statistical mechanics. The logarithmic form, $D(\rho\|\sigma)=\Tr[\rho(\log\rho - \log\sigma)]$, possesses rich convex analytic, operator-algebraic, and functional-inequality properties that enable its use in hypothesis testing, channel capacity theory, many-body physics, and the mathematics of open quantum systems.

1. Mathematical Definition and Monotonicity Structure

Quantum logarithmic relative entropy is defined for a pair of density matrices $\rho,\sigma$ (positive semidefinite, unit trace) on a Hilbert space $\mathcal{H}$ with $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$ as

$D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$

and $D(\rho\|\sigma)=+\infty$ if this support condition is violated (Lewin et al., 2013, Okamura, 2012, Kudler-Flam, 2021).

This quantity is a specialization of the general quantum $f$-divergence

$H_\phi(A,B) = \Tr\left[\phi(A) - \phi(B) - \phi'(B)(A - B)\right]$

to the case $\phi(x) = x\log x$, where the operator-monotonicity of $\phi'$ ensures monotonicity of $\rho,\sigma$0 under completely positive trace-preserving (CPTP) maps. Concretely, for any CPTP map $\rho,\sigma$1, the data-processing inequality holds: $\rho,\sigma$2 The functional $\rho,\sigma$3 is jointly convex in its arguments, lower semicontinuous, and continuous on sets of states with spectra bounded away from $\rho,\sigma$4 (Lewin et al., 2013, Okamura, 2012).

In the infinite-dimensional setting, $\rho,\sigma$5 is defined by approximation via finite-rank projections, and the key monotonicity and lower-semicontinuity properties carry over (Lewin et al., 2013, Wirth, 12 May 2025).

2. Statistical Interpretation and Rate Functions

Quantum relative entropy functions as the sharp rate function in quantum hypothesis testing and large deviations (Okamura, 2012). For quantum analogs of classical Stein's lemma, the optimal exponential decay rate for the error of the second kind in asymmetric hypothesis testing between $\rho,\sigma$6 and $\rho,\sigma$7 is given by $\rho,\sigma$8. In the context of quantum Sanov's theorem, $\rho,\sigma$9 appears in large deviations for empirical distributions of repeated quantum measurements.

The Hiai–Ohya–Tsukada theorem reduces quantum relative entropy between states on a $\mathcal{H}$0-algebra to the classical relative entropy between their barycentric probability central measures, enabling a transfer of large-deviation and information-criteria results to the quantum regime (Okamura, 2012).

Pinsker-type inequalities, such as $\mathcal{H}$1, quantify non-commutative distinguishability and underlie quantum versions of information-theoretic bounds (Guo et al., 9 May 2026).

3. Functional Inequalities, Markov Dynamics, and Gradient Flows

Quantum logarithmic relative entropy is central to the modified logarithmic Sobolev inequality (MLSI) for quantum Markov semigroups (Wirth, 12 May 2025). For a primitive quantum dynamical semigroup with generator $\mathcal{H}$2 and invariant state $\mathcal{H}$3, MLSI with constant $\mathcal{H}$4 asserts: $\mathcal{H}$5 MLSI is equivalent to the exponential decay of $\mathcal{H}$6 at rate $\mathcal{H}$7 along the semigroup evolution, a quantum generalization of classical entropy dissipation (Wirth, 12 May 2025).

The quantum HWI inequality interpolates among quantum relative entropy, a non-commutative Wasserstein-2 distance (constructed via quantum Markov generators), and quantum Fisher information. Under a quantum Ricci curvature bound $\mathcal{H}$8, for faithful $\mathcal{H}$9 (Rouzé et al., 2017): $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$0 This machinery underpins convergence and concentration properties of quantum dynamical systems and connects entropy to geometry (Rouzé et al., 2017).

4. Computation, Optimization, and Algorithms

Efficient computation of $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$1 underlies both theoretical study and practical applications. In finite dimensions, matrix logarithms are approximated by rational operator-concave functions, leading to semidefinite programming (SDP) relaxations (Fawzi et al., 2017). For instance, the Fawzi–Saunderson–Parrilo rational approximation method expresses $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$2 as a weighted sum of resolvents, enabling the construction of SDP constraints converging exponentially fast to $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$3. This approach is implemented in the CvxQuad toolbox and is tractable for systems up to moderate dimension.

On quantum hardware, estimation of $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$4 has recently advanced via variational-quadrature algorithms that combine operator functional calculus with hardware-efficient parameterizations. These algorithms approximate the logarithmic divergence by a sum over kernel-based $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$5-divergences, each estimated by solving a variational minimization involving only measurement of traces against $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$6 and $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$7, and requiring only $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$8 qubits for $\mathrm{supp}\,\rho\subseteq\mathrm{supp}\,\sigma$9-qubit states (Lu et al., 13 Jan 2025). This resolves long-standing open questions about the tractable quantum estimation of logarithmic relative entropy and enables distributed quantum computation scenarios.

5. Applications in Quantum Information, Many-Body, and Physics

Quantum relative entropy is a universal resource metric in quantum information. It underpins the definitions of the relative entropy of entanglement, secret key rates, and classical, quantum, and entanglement-assisted channel capacities via variational expressions within convex sets of states (Fawzi et al., 2017, Okamura, 2012). Many capacity expressions reduce to optimizing $D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$0 over input ensembles or output states.

In many-body systems and statistical physics, quantum relative entropy is the core tool in proving mean-field limits, entropy production identities, and propagation of chaos for quantum dynamics (Guo et al., 9 May 2026). It quantifies the proximity of the many-body density matrix to effective tensorized mean-field states, and its block-subadditivity ensures propagation of entropy bounds to reduced marginals.

For random quantum states, analytical large-$D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$1 techniques (e.g., diagrammatics and replica trick) yield closed-form expressions for expected quantum relative entropy between independently sampled states from Wishart ensembles. This underlies precise statements of subsystem eigenstate thermalization (sETH) and allows for rigorous quantification of black hole microstate distinguishability in AdS/CFT (Kudler-Flam, 2021).

6. Refined Inequalities, Stability, and Quasi-Entropy Frameworks

Quantum logarithmic relative entropy admits refinements and stability bounds expressed via remainder terms quantifying the saturation of key inequalities:

• Explicit norm bounds relating the data-processing deficit $D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$2 to recovery map errors (e.g., Petz recovery) and Hilbert–Schmidt or trace-norm power differences.
• Remainder terms for strong subadditivity and generalized operator inequalities, connecting the gaps in entropy inequalities to distances in operator powers (Vershynina, 2018).
• Extensions within the quasi-relative entropy framework ($D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$3), where $D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$4 and $D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$5 recovers Umegaki entropy, yield joint convexity, data-processing, and subadditivity stability quantifications (Vershynina, 2018).

Belavkin–Staszewski (BS) relative entropy, a logarithmic $D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$6-divergence associated to $D(\rho\|\sigma) = \Tr\left[\rho\left(\log\rho - \log\sigma\right)\right]$7, admits weak quasi-factorization theorems analogous to Umegaki entropy, which are central in analysis of log-Sobolev inequalities for quantum Markov semigroups (Bluhm et al., 2021).

7. Perspectives and Further Directions

Quantum logarithmic relative entropy is established as the universal divergence-object in non-commutative probability, unifying concepts in quantum statistical inference, operational information theory, open-system dynamics, and mathematical physics. Its compatibility with operator-algebraic structures, monotone functional calculus, and convex-analytic frameworks positions it as the reference measure for quantum statistical distances.

Ongoing research extends its functional-analytic properties to type III von Neumann algebras, explores sharp constants in functional inequalities, and investigates its geometry via quantum optimal transport. Quantum algorithms now offer scalable routes for experimental estimation on distributed and NISQ-era devices, promising further cross-disciplinary impacts (Lu et al., 13 Jan 2025, Wirth, 12 May 2025, Rouzé et al., 2017).