---
title: Quantum Relative Entropy
url: https://www.emergentmind.com/topics/quantum-relative-entropy-qre
type: topic
---

# Quantum Relative Entropy

Quantum relative entropy (QRE), originally introduced by Umegaki, is the principal quantum generalization of the classical Kullback–Leibler (KL) divergence. It is the fundamental information-theoretic measure of state distinguishability, governing error exponents in quantum hypothesis testing, the geometry of quantum state spaces, operational rates in quantum resource theories, and non-asymptotic and asymptotic bounds for quantum protocols. QRE is ubiquitous across quantum information, quantum computing, statistical mechanics, and quantum field theory. Its mathematical structure supports a rich convex-analytic, operational, and algebraic theory, with deep connections to thermodynamics, convex optimization, and quantum channel theory.

## 1. Definition, Properties, and Operational Meaning

For density operators $\rho$, $\sigma$ on a finite-dimensional Hilbert space $\mathcal{H}$, the Umegaki quantum relative entropy is defined as
\[
D(\rho\Vert \sigma) =
\begin{cases}
\Tr\big[\rho(\log\rho - \log\sigma)\big], & \mathrm{supp}(\rho)\subseteq\mathrm{supp}(\sigma),\\
+\infty, & \text{otherwise}.
\end{cases}
\]
The function satisfies:
- **Nonnegativity**: $D(\rho\Vert\sigma)\ge 0$ with equality if and only if $\rho = \sigma$.
- **Monotonicity/Data-processing inequality (DPI)**: For any completely positive trace-preserving (CPTP) map $\mathcal{M}$,
  \[
  D(\rho\Vert \sigma) \ge D(\mathcal{M}(\rho)\Vert \mathcal{M}(\sigma)).
  \]
- **Joint convexity**: $D(\cdot\Vert\cdot)$ is convex in $(\rho, \sigma)$.
- **Classical limit**: For commuting states, $D(\rho\Vert\sigma)$ reduces to the classical KL divergence.

QRE quantifies asymptotic state distinguishability in quantum hypothesis testing (Stein’s Lemma); it is the sharp error exponent for Type-II errors under a fixed Type-I constraint, and thus serves as the rate function for large deviations in quantum statistical mechanics and state estimation [1202.2943].

## 2. Axiomatic and Operational Characterizations

Matsumoto’s reverse test establishes QRE as the unique monotone, additive, and lower semi-continuous extension of classical relative entropy:
- **Axioms**: (i) Monotonicity under CPTP, (ii) Normalization to KL for commuting states, (iii) Additivity, and (iv) Asymptotic continuity [1010.1030].
- **Reverse test**: The minimal classical divergence needed to simulate a given quantum pair $(\rho,\sigma)$ via CPTP encoding is
  \[
  D^R(\rho\Vert\sigma) = \Tr[\rho\ln(\sqrt{\rho}\,\sigma^{-1}\sqrt{\rho})],
  \]
  which universally lower-bounds any monotone quantum relative entropy.
- **Asymptotic uniqueness**: Under natural operational scenarios, Umegaki’s QRE is the unique (up to constant factors) quantum extension—mirroring resource conversion rates and optimal costs in quantum Shannon theory and resource resource theories [1010.1030].

## 3. Chain Rules, Data Processing, and Recovery

In the classical setting, KL divergence obeys exact chain rules decomposing divergence across components or subsystems. QRE captures an analogous structure only asymptotically:
- **Quantum chain rule (asymptotic, regularized)** [1909.05826]:
  \[
  D(\mathcal{E}(\rho_{A_1A_2})\Vert \mathcal{F}(\sigma_{A_1A_2})) \le D(\rho_{A_2}\Vert \sigma_{A_2}) + \bar D^{\mathrm{reg}}(\mathcal{E}\Vert\mathcal{F})
  \]
  with $\bar D^{\mathrm{reg}}$ the regularized channel divergence.
- **Single-letter bounds** [2510.16918]:
  - **POVM decomposition**:
    \[
    D(\rho\Vert\sigma) - D(\mathcal{M}(\rho)\Vert\mathcal{N}(\sigma)) \ge - \mathbb{E}_{j\sim P^G_\rho} D(\mathcal{M}(\rho_j)\Vert\mathcal{N}(\sigma_j)),
    \]
    where $G$ is a POVM and $\rho_j$ are “conditioned” states.
  - **Projector-based and semiclassical decompositions**: For spectral projectors $\Pi_j$ of $\rho$,
    \[
    D(\rho\Vert\sigma) - D(\mathcal{M}(\rho)\Vert\mathcal{N}(\sigma)) \ge - \sum_j p_j D(\mathcal{M}(\Pi_j)\Vert\mathcal{N}(\Pi_j)),
    \]
    with tightness for commuting states (recovering the classical chain rule).

QRE monotonicity is intimately linked to the Petz recovery map. The difference $D(\rho\Vert \sigma) - D(\mathcal{N}(\rho)\Vert \mathcal{N}(\sigma))$ quantitatively bounds the fidelity between $\rho$ and its Petz recovery $\mathcal{R}_{\sigma,\mathcal{N}}(\mathcal{N}(\rho))$: this underpins results in strong subadditivity, convexity, and conditional entropy [1412.4067].

## 4. Information Geometry and Metric Responses

QRE generates an information-geometric structure on quantum state space:
- **Metric tensor (QRE-susceptibility)**: For a smooth one-parameter family $\rho(\lambda)$,
  \[
  S(\rho(\lambda)\|\rho(\lambda+\delta\lambda)) = \Sigma_{\lambda\lambda} (\delta\lambda)^2 + O(\delta\lambda^3), \quad
  \Sigma_{ij} = \frac{1}{2}\Tr[\rho^{-1}\partial_i\rho\partial_j\rho].
  \]
- **Quantum critical points (QCPs)**: In quantum many-body systems, the QRE metric susceptibility diverges at QCPs, encoding universal scaling features. For the transverse-field Ising model, the divergence is logarithmic, whereas in non-integrable three-spin chains the divergence is quadratic in system size, reflecting different universality classes [2509.22515].
- **Relation to fidelity susceptibility and Rényi divergences**: QRE susceptibility generalizes fidelity susceptibility to mixed-state spaces, and preserves connections to the Petz–Rényi spectrum [2509.22515].

## 5. Quantum Resource Theories and Optimization

Resource measures in quantum information—such as entanglement, coherence, and “magic”—are most generally cast as QRE distances to convex sets of “free” states:
- **Relative entropy of entanglement**:
  \[
  E_R(\rho) = \min_{\sigma \in \mathrm{Sep}} D(\rho \Vert \sigma).
  \]
- **Resource-theoretic tasks**: The regularized QRE, $D^{(\infty)}(\rho\Vert\sigma)$, determines rates in resource conversion, such as entanglement cost and distillable entanglement, and magic-state distillation [2502.15659].
- **QRE programming and convex optimization**: Efficient semidefinite programming (SDP) and self-concordant barrier techniques support practical computation of QRE-based quantities in high dimensions [2312.07438, 1705.06671]. Rational approximations of the matrix logarithm enable embedding QRE constraints in optimization frameworks critical for quantum communication, key distribution, and state certification.

## 6. Extensions: Rényi and Measured Relative Entropies

QRE generalizes to parametrized $\alpha$-divergences:
- **Petz–Rényi and sandwiched Rényi relative entropy**:
  \[
  S_\alpha(\rho\Vert\sigma) = \frac{1}{\alpha-1}\log\Tr\left[\left(\sigma^{\frac{1-\alpha}{2\alpha}}\rho\sigma^{\frac{1-\alpha}{2\alpha}}\right)^\alpha\right],
  \]
  interpolating between trace distance (for $\alpha = 1/2$), QRE ($\alpha\to 1$), and max-divergence ($\alpha\to\infty$) [1512.02615, 2411.09696, 1404.3216].
- **Measured relative entropy $D_m$**: The maximal classical relative entropy over all POVMs, $D_m(\rho\Vert\sigma)\le D(\rho\Vert\sigma)$, with equality if and only if $[\rho,\sigma]=0$. Sandwiched Rényi divergences dominate measured counterparts for $\alpha > 1/2$ [1512.02615].
- **QFT and modular theory**: In algebraic QFT, Petz–Rényi relative entropy is constructed via modular operators and encompasses genuinely quantum fluctuations beyond the classical symplectic structure [2411.09696].

## 7. Applications and Contemporary Algorithms

QRE and its generalizations underpin quantum channel discrimination, state estimation, and statistical inference:
- **Large deviation theory and information criteria**: QRE is the rate function for quantum large deviation principles, directly justifying information criteria such as AIC and WAIC in the quantum context, with classical accuracies [1202.2943].
- **Quantum algorithms**: Recent variational quantum algorithms efficiently estimate $D(\rho\Vert\sigma)$ using quadrature techniques and operator ansätze, with polynomial resource scaling and direct applicability to distributed quantum architectures [2501.07292].
- **Thermodynamics and coherence**: QRE measures deviation from equilibrium, quantifying quantum coherence consumption and its separation from classical entropy production in thermalization near black holes [2501.00229].

---

A plausible implication of this corpus is that QRE serves as the universal non-classical divergence in quantum statistical mechanics, quantum computing, and information processing, bestowing both foundational and practical structure on the geometry and dynamics of quantum information [2510.16918, 2502.15659, 2509.22515, 1412.4067, 2312.07438]. Open problems focus on single-letterization of chain rules, extension to other divergences, modular-theoretic/geometric interpretations, and further computational refinements.

Source: https://www.emergentmind.com/topics/quantum-relative-entropy-qre