Uhlmann’s Monotonicity Theorem

• Uhlmann’s Monotonicity Theorem is a core result establishing that quantum relative entropy does not increase under CPTP and unital Schwarz maps.
• It employs a sophisticated interpolation of sesquilinear forms and operator-algebraic techniques to address even non-invertible density operators.
• The theorem underpins the quantum Data Processing Inequality and inspires further advances in quantum divergence studies and recovery map analyses.

Uhlmann’s Monotonicity Theorem establishes that the quantum relative entropy is monotonically non-increasing under completely positive, trace-preserving (CPTP) maps, including the broader class of unital Schwarz maps. This result provides a conceptual and technically robust proof of the quantum Data Processing Inequality for the Umegaki (or quantum) relative entropy, applicable even in cases with non-invertible density operators. The theorem’s significance lies in its highly abstract, operator-algebraic approach using the interpolation of sesquilinear forms, which directly leverages the structure of quantum states and maps without recourse to operator convexity or traditional contractive inequalities (Pérez-Pardo, 2022, Matheus et al., 14 Sep 2025).

1. Operator-Algebraic Setting and Interpolation of Quadratic Forms

Let $A$ be a unital C*-algebra over $\mathbb{C}$ with unit $e$. A state $\omega$ on $A$ is a linear functional $\omega : A \to \mathbb{C}$ adhering to:

• $\omega(a^* a) \geq 0$ for all $a \in A$,
• $\omega(e)=1$,
• $\omega(a^*) = \overline{\omega(a)}$ (reality on self-adjoint elements).

Given two states $\omega, v$ on $A$, two positive Hermitian quadratic forms are defined on $A$:

• $\omega_R(a, b) = \omega(b a^*)$,
• $v_L(a, b) = v(a^* b)$.

For any pair $p, q$ of positive Hermitian quadratic forms on a complex vector space $V$, the Pusz–Woronowicz interpolation yields a family of quadratic forms. There exists a Hilbert space $H$ and commuting positive operators $P, Q \in B(H)$ with $P + Q = I$, such that for $t \in [0, 1]$: $$Y_{p,q}^t(a, b) = \langle [a], P^{1-t} Q^t [b] \rangle_H,$$ where $[a]$ denotes the equivalence class of $a$ in $H$. This construction is pivotal in Uhlmann’s formulation (Pérez-Pardo, 2022).

Key facts used from the interpolation theory include:

• Monotonicity under form-order: If $p \geq p'$ and $q \geq q'$, then $Y_{p, q}^t(a, a) \geq Y_{p', q'}^t(a, a)$ for all $t$.
• Pull-back dominance: For a linear map $\Phi: V' \to V$, the interpolation of pull-backs $$p \circ (\Phi \times \Phi), q \circ (\Phi \times \Phi)$$ dominates the pulled-back interpolation.

2. Uhlmann’s Definition of Quantum Relative Entropy

Given states $w, v$ on a unital C*-algebra $A$ and the interpolating quadratic forms $Y_{w_R, v_L}^t$, define: $$D_{w, v}^t(a, b) := Y_{w_R, v_L}^t(a, b) - w_R(a, b),$$

$$D_{w, v}(a, b) := -\liminf_{t \to 0^+} \frac{D_{w, v}^t(a, b)}{t}.$$

The (quantum) relative entropy is then $S[w, v] = D_{w, v}(e, e)$. In the finite-dimensional case $A = M_n(\mathbb{C})$ with $w(a) = \operatorname{Tr}(\rho a)$, $v(a) = \operatorname{Tr}(\sigma a)$, this reduces to

$$S[w, v] = \operatorname{Tr}(\rho \log \rho) - \operatorname{Tr}(\rho \log \sigma).$$

(Pérez-Pardo, 2022, Matheus et al., 14 Sep 2025)

3. Statement and Proof of Uhlmann’s Monotonicity Theorem

Let $A, B$ be unital C*-algebras, and $\Phi: A \to B$ a unital Schwarz map (in particular any CPTP map). For the pull-back states $w \circ \Phi, v \circ \Phi$, the theorem asserts: $S[w, v] \geq S[w \circ \Phi, v \circ \Phi].$

The proof exploits the structure of the quadratic form interpolation:

• By definition,

$$S[w, v] = -\liminf_{t \to 0^+} \frac{Y_{w_R, v_L}^t(e, e) - w_R(e, e)}{t},$$

and

$$S[w \circ \Phi, v \circ \Phi] = -\liminf_{t \to 0^+} \frac{Y_{(w \circ \Phi)_R, (v \circ \Phi)_L}^t(\hat{e}, \hat{e}) - (w \circ \Phi)_R(\hat{e}, \hat{e})}{t},$$

with $\hat{e}$ the unit of $A$ and $\Phi(\hat{e}) = e$.

• Using the Schwarz property of $\Phi$, monotonicity and pull-back properties of interpolation yield:

$$Y_{w_R, v_L}^t(\Phi(\hat{e}), \Phi(\hat{e})) \leq Y_{(w \circ \Phi)_R, (v \circ \Phi)_L}^t(\hat{e}, \hat{e}),$$

and

$(w \circ \Phi)_R(\hat{e}, \hat{e}) = w_R(e, e).$

The inequality passes to the limit, establishing the theorem (Pérez-Pardo, 2022).

4. Abstract Operator-Theoretic Formulation and Extensions

Uhlmann’s strategy can be recast in the finite-dimensional setting with density operators $\rho, \sigma$ and quantum channel (CPTP map) $\Phi: B(\mathcal{H}) \to B(\mathcal{K})$. The quantum relative entropy is defined as

$$D(\rho\|\sigma) = \begin{cases} \operatorname{Tr}[\rho (\log \rho - \log \sigma)], & \operatorname{supp} \rho \subseteq \operatorname{supp} \sigma, \ +\infty, & \text{otherwise.} \end{cases}$$

and Uhlmann’s Monotonicity is

$$D(\Phi(\rho) \| \Phi(\sigma)) \leq D(\rho \| \sigma).$$

The interpolation method uses positive sesquilinear forms $\rho_L, \sigma_R$ on $B(\mathcal{H})$ and their $t$-interpolant: $$\gamma^t_{\rho_L \to \sigma_R}(A, B) = \operatorname{Tr}[\rho^{1-t} A^\dagger \sigma^t B].$$ This yields a log-convex function in $t$ and accommodates the singular case naturally, as the definition and chain of inequalities do not require invertibility of $\rho$ or $\sigma$. The proof generalizes directly to all (not necessarily invertible) density operators and any positive unital map (Matheus et al., 14 Sep 2025).

5. Conditions for Equality and Illustrative Examples

Equality $S[w, v] = S[w \circ \Phi, v \circ \Phi]$ is attained if and only if all interpolation-induced inequalities are saturated. Specifically,

$$(w \circ \Phi)_R(a, a) = w_R(a, a) \quad \text{and} \quad (v \circ \Phi)_L(a, a) = v_L(a, a) \quad \forall a \in A,$$

which holds precisely when $\Phi$ is a $*$-homomorphism on the support of the relevant forms. In finite dimensions, this restricts equality to reversible channels on the joint support of $\rho, \sigma$.

A nontrivial example arises when restricting to a unital $*$-subalgebra $B \subset A$ via the inclusion map $\Phi: A \to B$ (a $*$-homomorphism, hence CPTP). For any two states $w, v$ on $B$, their extensions $w \circ \Phi, v \circ \Phi$ on $A$ satisfy

$S[w \circ \Phi, v \circ \Phi] \leq S[w, v],$

i.e., restricting to a smaller algebra cannot increase relative entropy. This encompasses partial trace (marginalization) in quantum systems, emphasizing that tracing out a subsystem strictly decreases the relative entropy between joint states (Pérez-Pardo, 2022).

6. Comparative Perspective: Uhlmann versus Petz

Uhlmann’s proof employs interpolation of sesquilinear forms and does not resort to the operator convexity of $-\log(\cdot)$ or Jensen-type contractive inequalities, unlike the method of Petz. The Uhlmann approach is

• Manifestly symmetric in $\rho, \sigma$ and robust to singularities,
• Applicable with only the Cauchy–Schwarz inequality and extremal properties of the geometric mean,
• Invariant under passage to any positive unital map (Schwarz map), not only CPTP.

In contrast, Petz’s proof, while more direct, relies on Jensen’s inequality and explicitly requires the Choi–Kraus representation of quantum channels. Uhlmann’s interpolation technique is thus more abstract, encompassing, and directly reveals the underlying operator-algebraic structure of quantum relative entropy monotonicity (Matheus et al., 14 Sep 2025).

7. Broader Significance and Applications

Uhlmann’s Monotonicity Theorem underpins the quantum Data Processing Inequality and admits generalization beyond completely positive maps, facilitating extensions to other quantum divergences and recovery analyses. The sesquilinear-form calculus underlying the proof has found applications in the study of quantum divergences, transpose-channel recovery, and interpolation theory.

The approach emphasizes the structural properties of quantum information, highlighting the role of quadratic-form geometry rather than specific operator inequalities. It thus provides a versatile and robust framework for theoretical developments in quantum information theory and operator algebras (Pérez-Pardo, 2022, Matheus et al., 14 Sep 2025).