Multi-Indexed Schatten Quasi- and Anti-Norms and Additivity of Conditional Rényi Entropies

AI summary

Overview

Research area: Quantum information theory and operator space theory — specifically non-commutative $\ell_p$-type (Schatten) spaces, mixed norms, and Rényi entropic quantities.

Technical level: Advanced. The paper is written for readers comfortable with Schatten norms, operator spaces, completely bounded maps, and the sandwiched Rényi divergence. The prose is accessible in motivation but the results are theorem-heavy.

Scope (one sentence): The paper introduces multi-indexed Schatten quasi-norms and anti-norms for arbitrarily many indices (positive or negative) satisfying a mutual compatibility condition, and uses them to prove additivity and chain rules for conditional Rényi entropies of quantum channels.

What This Paper Is About

Classical information theory describes many Rényi quantities as mixed $\ell_{q_1}[\ell_{q_2}[\cdots]]$ norms — for example, Arimoto's conditional entropy is an $\ell_1[\ell_\alpha]$ quasi-norm of a joint distribution. Quantum information theory cannot simply copy this construction, because non-commutativity and entanglement remove the coordinate structure that makes mixed norms easy to define; the naïve guess involving partial traces is not even convex for $p > 2$. Pisier's operator-valued Schatten spaces gave a tractable answer only for indices $q,p \geq 1$ and became analytically unwieldy beyond two or three indices, while a recent two-indexed quasi-norm extension reached $\alpha \geq 1/2$ but no further.

This paper lifts both restrictions at once: it builds a full theory of multi-indexed Schatten spaces with arbitrarily many indices, including negative ones (anti-norms), under the compatibility condition $|1/q_i - 1/q_j| \leq 1$, and then applies the resulting toolbox to multiplicativity and chain-rule questions for conditional entropies of quantum channels.

Key Contributions

1. Definition and structural theory of multi-indexed Schatten spaces. An inductive construction assigns an index $q_i$ to each system $\mathcal{H}_i$ in an $n$-fold tensor product, yielding quasi-norms for positive indices and anti-norms for negative indices, with proofs of triangle inequality, Hölder duality, reverse triangle/reverse Hölder duality, collapse on block-diagonal operators, multiplicativity over product operators, and partial-trace behavior of a trailing index $1$.

2. A manifestly multi-indexed factorization formula (Theorem 2.3), the technical main result. It factors $|\cdot|_{(q_1,\dots,q_n)}$ through a single Schatten quasi-norm with weighted operators carrying indices $1/s_i = 1/q_i - 1/\bar{q}$, generalizing Pisier's factorization formula (also extended here in Theorem 2.2) and the two-indexed formulas of prior work.

3. Lifting of the Devetak–Junge–King–Ruskai toolbox. A non-commutative Minkowski inequality (Theorem 3.1) for swapping adjacent decreasing indices, now allowing non-trivial inner indices, plus two identity-removal equalities (Theorems 3.2 and 3.3) for indices in $[1,\infty]$ and $[1/2,1]$ respectively.

4. Entropic consequences. Additivity under tensor products of channels for the completely bounded minimal output conditional sandwiched Rényi entropy and for the maximal output conditional entropy, for all $\alpha \geq 1/2$; additivity of the completely bounded $\alpha$-Rényi coherent information; and chain rules for compositions of possibly signaling channels for $\alpha \in [1/2,1)$.

Main Findings

• Compatibility is the right condition. A tuple $\mathcal{Q} = (q_1,\dots,q_n)$ of indices of the same sign is called compatible when $|1/q_i - 1/q_j| \leq 1$ for all $i,j$. This is automatic when all $q_i \geq 1$, and for the pair $(1,\alpha)$ it holds exactly when $\alpha \geq 1/2$ — the same range in which the sandwiched Rényi divergence satisfies data processing.

• Factorized weights. Theorem 2.3 expresses the multi-indexed quasi-norm as an infimum (or supremum) over operators $A,B$ acting on all $n$ systems, with weights indexed by $1/s_i = 1/q_i - 1/\bar{q}$ and $1/t_i = 1/\underline{q} - 1/q_i$; for self-adjoint $X$ one may take $A = B^* \geq 0$.

• Reduction to known cases. For a single index the formulas recover the standard variational formulas for Schatten quasi-norms; for two indices they recover the two-indexed lemmas of the author's prior work with Fawzi and Rouzé.

• Quasi-norm control. For positive indices the construction gives $\min{1, \min_i q_i}$-normable quasi-norms (Theorem 2.5) satisfying the triangle inequality when all $q_i \geq 1$, together with multi-indexed Hölder duality (Proposition 2.2); for compatible $0 < q_i \leq 1$ one gets a reverse triangle inequality on positive operators (Theorem 2.6) and reverse Hölder duality (Corollary 2.4).

• Block-diagonal collapse. $|\oplus_i X_i|{(q,\mathcal{Q})} = \left(\sum_i |X_i|{(\mathcal{Q})}^q\right)^{1/q}$ (Corollary 2.6), extending Pisier's corollary to quasi- and anti-norms.

• Multiplicativity along ordered cuts. $|X_{\mathcal{Q}} \otimes Y_{\mathcal{P}}|{(\mathcal{Q},\mathcal{P})} = |X{\mathcal{Q}}|{(\mathcal{Q})}|Y{\mathcal{P}}|_{(\mathcal{P})}$ (Corollary 2.5), and a mixed tensor-product splitting lemma (Lemma 3.3) whose special case $q=1$, $p=\alpha \in [1/2,\infty]$ reproduces additivity of the optimized sandwiched Rényi conditional entropy under tensor-product states.

• Trailing index 1 is a partial trace. $|X|{(q_1,\dots,q_n,1)} = |\mathrm{tr}{n+1}[X]|_{(q_1,\dots,q_n)}$ for $X \geq 0$ (Corollary 2.3).

• Non-commutative Minkowski inequality. For compatible $\mathcal{E}, p, q, \mathcal{R}$ with $q \leq p$, swapping adjacent systems is non-contractive: $|X_{EQPR}|{(E:\mathcal{E},Q:q,P:p,R:\mathcal{R})} \geq |X{EPQR}|_{(E:\mathcal{E},P:p,Q:q,R:\mathcal{R})}$ (Theorem 3.1). Unlike Pisier's corollary, it permits inner indices $\mathcal{R}$, which the author states had not been considered even in the normed operator-space setting.

• Identity removal. $|\Phi \otimes \mathrm{id}T|{(Q:\mathcal{Q},T:\mathcal{T}) \to (P:\mathcal{P},T:\mathcal{T})} = |\Phi|_{(Q:\mathcal{Q}) \to (P:\mathcal{P})}$ when all indices lie in $[1,\infty]$ (Theorem 3.2), with a co-quasi-norm analogue for indices in $[1/2,1]$ (Theorem 3.3). The inner space must be normed: the equality fails otherwise, by a counterexample in the two-indexed literature.

• CP maps attain their mixed norms on positive operators (Corollary 3.2), $|\Phi| = |\Phi|^+$, generalizing earlier results for mixed Schatten norms, $cb,q\to p$ norms, and $cb,q\to(r,s)$ norms to arbitrary numbers of positive indices.

• Tensorization of (reverse) hypercontractivity. The $cb,q\to p$ quasi-norm is multiplicative under tensor products of linear maps for all compatible $q \leq p$ (Lemma 3.4) and the $cb,co,q\to p$ co-quasi-norm for all compatible $q \geq p$ (Theorem 3.4), both including negative indices. The former extends a result restricted to $1 \leq q \leq p$ and CP maps; the latter upgrades super-multiplicativity for $0 < p \leq q$ to full multiplicativity.

• Additivity of conditional entropies. For all $\alpha \in [1/2,1) \cup (1,\infty]$, the completely bounded minimal output conditional sandwiched Rényi entropy is additive under tensor products of channels (Theorem 4.1), extending the $\alpha > 1$ case to the whole range $\alpha \geq 1/2$. The maximal output conditional version is also additive (Theorem 4.2) and, unlike the minimal one, needs no complete boundedness — it coincides with its completely bounded version, which the author states has not appeared explicitly in the literature.

• Coherent information. The completely bounded $\alpha$-Rényi coherent information of a channel is additive under tensor products (Theorem 4.3).

• Signaling chain rule. For channels $\Phi: Q \to A_1B_1Q_1$ and $\Psi: Q_1 \to A_2B_2$ and any $\alpha \in [1/2,1]$, the worst-case conditional entropy of the composition is bounded by the sum of the single-round worst cases (Theorem 4.4), with a companion statement in Theorem 4.5.

Methodology in Plain English

The construction proceeds by induction, absorbing one system at a time. To combine system 1 with system 2, one inserts an optimal positive operator $a$ on system 1 — dividing it out (an infimum) when $q_1 \leq q_2$, or multiplying it in (a supremum) when $q_1 \geq q_2$ — with the weight measured in a Schatten norm of order $2r$, where $1/r = |1/q_1 - 1/q_2|$. The two systems are then treated as a single system carrying index $q_2$, and the procedure repeats until one ordinary Schatten quasi-norm remains. Negative indices use the same recursion restricted to positive operators $X > 0$ and $a = b^* > 0$.

The pivotal move is to prove a factorization formula, in the spirit of a multi-indexed Hölder inequality, that rewrites the whole nested object as a single Schatten quasi-norm sandwiched between weighted operators. Once this is available, nearly every structural property — triangle inequality, duality, block-diagonal collapse, multiplicativity, partial-trace behavior — follows by combining operator-space arguments with matrix-analytic techniques.

For the entropic applications, the key observation is that conditional output entropies of channels are $(1,\alpha)$-type mixed norms of the channel's Choi-type representation. A tensor product of two channels therefore produces four indices $(1,\alpha,1,\alpha)$, and sequential compositions produce nested structures with systems buried deep inside; both require the ability to swap systems at inner positions, which is exactly what the new Minkowski inequality provides. The additivity and chain-rule proofs then take only a few lines once the toolbox is in place.

Why This Matters

The paper effectively completes the quantum analogue of the classical theory of mixed $\ell_{q_1}[\ell_{q_2}[\cdots]]$ spaces, and does so in the regime $\alpha \geq 1/2$ that matters information-theoretically: $\alpha = 1/2$ is the max-entropy with direct operational meaning, and $\alpha \in [1/2,1)$ supplies strong converse exponents. Additivity and chain rules answer the operational question of whether an adversary with access to environments across many rounds of a parallel or sequential strategy can do better than the sum of single-round bounds — and whether entangled inputs across rounds help. The results say no, so multi-round protocols reduce to single-round quantities.

Real-world applications:

• Device-independent entanglement certification: the signaling chain rule extends the smooth max-entropy chain rule used there (the single point $\alpha = 1/2$) to all $\alpha \in [1/2,1)$, so certification bounds can in principle be optimized over $\alpha$.
• Quantum key distribution security proofs: the $\alpha > 1$ additivity of completely bounded minimal output entropies was previously applied to time-varying channels; this work extends that range down to $\alpha = 1/2$.
• Finite-resource information theory: one-shot and finite-block-length bounds, error and strong converse exponents, and security proofs with finitely many rounds all depend on $\alpha < 1$ Rényi quantities.
• Dissipative open quantum system dynamics: the tensorization of complete hypercontractivity and complete reverse hypercontractivity bears on modified logarithmic Sobolev inequalities and mixing-time analyses.

Industry relevance: the results feed most directly into quantum cryptography — QKD security analysis under realistic, time-dependent device behavior — and into the theoretical certification layer behind device-independent protocols. They are foundational rather than immediately implementable.

Future Directions

• Pushing past compatibility. The condition $|1/q_i - 1/q_j| \leq 1$ is described as sufficient and essentially necessary for two indices; whether a meaningful multi-indexed theory survives outside it is left open.
• Optimizing device-independent bounds over $\alpha$. The chain rule now holds on the whole interval $[1/2,1)$; quantifying the gain from optimizing $\alpha$ in DI entanglement distillation certification is a natural next step, analogous to earlier rate improvements from $\alpha \in [1,\infty]$ chain rules.
• Longer signaling chains. The chain rules are stated for compositions of two maps ($\Phi$ then $\Psi$); extending to $n$-round compositions of the type sketched in the paper's protocol figure is a clear open direction.
• Further multi-indexed applications. Channel uncertainty relations and other quantities that naturally carry one index per system may benefit from the three- and four-indexed norms that previously required ad hoc generalizations of Pisier's factorization formula.

Target Audience

Quantum information theorists working on Rényi entropies, channel capacities, and one-shot bounds; operator algebraists and operator-space researchers interested in non-commutative $L_p$ structures; mathematical physicists studying hypercontractivity and logarithmic Sobolev inequalities in open quantum systems; and security researchers who need additivity or chain-rule statements for multi-round QKD and device-independent protocols. A background in Schatten norms and completely bounded maps is assumed throughout.

Authors’ abstract

We introduce and study multi-indexed Schatten spaces which lift the commutative $\ell_{q_1}\![\ell_{q_2}\![\ell_{q_3}\![\dots]]]$ -spaces to the quantum toolbox. They extend Pisier's operator-valued Schatten norms and the recently introduced two-indexed Schatten quasi-norms. For every ordered tuple $\mathcal{Q}=(q_1,\dots,q_n)$ of non-zero indices we define a quasi-norm (for positive indices) or an anti-norm (for negative indices) on operators on an $n$-fold tensor product of Hilbert spaces. We do this for indices of the same sign satisfying the mutual compatibility condition $|\frac{1}{q_i}-\frac{1}{q_j}|\leq1$. Beyond establishing desirable properties of these quasi- and anti-norms via factorization formulas that combine operator space and matrix-analytic techniques, we lift several central tools and results of the seminal [Devetak, Junge, King, Ruskai, CMP 2006] to the multi-indexed and quasi-normed setting. These include a general non-commutative Minkowski inequality relating swapped systems and two identity-removal equalities. As a consequence, complete hypercontractivity and complete reverse hypercontractivity tensorize for all compatible indices, including negative ones. Concerning quantum information theory we prove that the completely bounded minimal and the maximal output conditional entropies are additive under tensor products of quantum channels for all $α\geq\frac12$. This extends the case $α&gt;1$ of [Fawzi, Kochanowski, Rouzé, Van Himbeeck, CMP 2026] and the unconditional results of [Devetak, Junge, King, Ruskai, CMP 2006]. Beyond tensor products, we establish chain rules for $\tilde H^\uparrow_α$, $α\in[\frac12,1)$, for compositions of possibly signaling channels, extending upon the smooth max-entropy chain rule used in device-independent entanglement certification to the whole interval $α\in[1/2,1)$.