Generalised quantum Stein's lemma more robust than ever

Filippo Girardi; Kuan-Yi Lee; Masahito Hayashi; Ludovico Lami

Source-linked AI summary

Generalised quantum Stein's lemma more robust than ever

Filippo Girardi, Kuan-Yi Lee, Masahito Hayashi, Ludovico Lami

arXiv:2609.17309v1Sep 15, 2026quant-phcs.ITmath-ph

TL;DR

The paper addresses how far the i.i.d. null-hypothesis assumption in the generalised quantum Stein's lemma can be relaxed while preserving the Stein exponent. It proves a Wasserstein-robust, universal version and applies it to compound and arbitrarily varying settings, with the latter governed by states in the convex hull of the null base set.

Problem
Prior robust GQSL results covered exact i.i.d. alternatives or the narrower MSR almost-i.i.d. model, leaving broader Wasserstein and weak source classes insufficiently established for composite testing.
Method
The paper establishes a universal GQSL for null sources uniformly approximating an i.i.d. state in normalised quantum Wasserstein distance of order 1, under Assumptions 9 on the alternative families.
Results
For Wasserstein-almost-i.i.d. null sources, the optimal exponent remains D∞(ρ∥S), while arbitrarily varying null hypotheses yield inf_{ρ∈conv(R1)} D∞(ρ∥S).
Takeaways & Limitations
The result provides universal tests without performance loss from ignorance of the precise null source and extends GQSL applications to compound and arbitrarily varying composite testing.
Takeaways & Limitations
The converse requires full-rank support conditions, and the corresponding proposition cannot be extended to weak sources even when the alternative state has full support.

Abstract

from arXiv · show

The generalised quantum Stein's lemma is a key result in quantum hypothesis testing, and connects this fundamental primitive of quantum information processing with quantum resource manipulation, a task that is central for technological applications. Prior works have proved this statement in the idealised setting of independent and identically distributed (i.i.d.) sequences of quantum states, and recent extensions consider also sources that are 'close' to i.i.d., according to the strict notion put forth by Mazzola, Sutter, and Renner. For several applications, however, one would need to consider yet more general sources. We establish a version of the generalised quantum Stein's lemma that is conceptually much simpler and general, as it applies to any source that is asymptotically close to an i.i.d. state with respect to the normalised quantum Wasserstein distance of order 1. As an immediate consequence, we solve the Stein exponent of a scenario where the null hypothesis is arbitrarily varying, expressing it in terms of i.i.d. Stein exponents corresponding to arbitrary states in the convex hull of the null hypothesis base set.

1 Introduction

The paper extends the generalised quantum Stein's lemma from exact or narrowly perturbed i.i.d. null sources to Wasserstein-almost-i.i.d. sources, including universal testing within equiconvergent families. It also applies the result to compound and arbitrarily varying composite hypothesis testing.

1 Introduction: Theorem 11 establishes the GQSL for Wasserstein-almost-i.i.d. null sources while requiring only convexity, closedness, tensor-product closure, and replacer stability of the alternative family.This simultaneously broadens the admissible null sources and weakens the alternative-family assumptions among the robust settings discussed.
1.5 Comparison with previous works: Replacing an exact i.i.d. null source with any Wasserstein-almost-i.i.d. source leaves the optimal type II exponent equal to D∞(ρ∥S), for both achievability and converse.For an individual source, the test may depend on the actual null sequence.
1.5 Comparison with previous works: Universal tests attain D∞(ρ∥S) when the unknown null sequence is any Wasserstein-almost-i.i.d. source along a fixed state ρ within an equiconvergent family.The test can depend on the null family, alternative family, and type I threshold, but not on the particular source in that family.
1 Introduction: The proof develops Wasserstein-stable information-spectrum and block-converse tools, including a padded block-product strong converse and a one-step gap contraction lemma.These propositions replace the i.i.d. block argument and iterative update construction in the relevant proof steps.
1.5 Comparison with previous works: The framework revisits compound i.i.d. testing, removes prior closure-under-partial-traces and permutation assumptions, and extends the result to compound Wasserstein-almost-i.i.d. null hypotheses.Theorem 16 preserves the same asymptotic exponent under Assumption 9, while Theorem 20 handles the compound Wasserstein setting.
1.5 Comparison with previous works: For arbitrarily varying null hypotheses, the exponent is inf_{ρ∈conv(R1)} D∞(ρ∥S), reflecting variation in the product composition across sites.The result allows arbitrary tensor products of states from R1 and their convex mixtures under the stated assumptions.

2 A Wasserstein-robustgeneralisedquantumStein’slemma

The paper establishes a generalised quantum Stein’s lemma robust to Wasserstein almost i.i.d. null sources, including universal tests for equiconvergent families. The proof extends information-spectrum results and derives individual-source robustness as a special case.

Main result: Theorem 11 establishes the GQSL for W1 almost i.i.d. sources when the alternative family satisfies Assumption 9.The stated assumptions include convexity and closedness, tensor-product closure, and stability under replacing any site with a full-rank state.
Universal robustness: Universal tests exist for compact, convex null families that converge uniformly to ρ^⊗n in normalised Wasserstein distance.These W1-robust tests achieve the optimal type II error exponent under uniformly small Wasserstein perturbations.
Universal robustness: The universal exponent matches the regularised exponent of the standard GQSL, so unknown null-source identity causes no asymptotic exponent loss.The universal test must satisfy the type I constraint for every state in the prescribed equiconvergent family.
Individual sources: Individual Wasserstein almost i.i.d. sources inherit the robustness result as a special case.The theorem covers sequences ρ_n that are Wasserstein-close to an i.i.d. state ρ^⊗n, without requiring a perfect characterization of the perturbation.
Proof strategy: The proof extends the information-spectrum ingredients used previously and then proves individual and universal robustness in separate stages.Sections 4.1–4.3 respectively address the technical extension, individual-source robustness, and the universal equiconvergent-source result.

3 Applications

The applications extend generalised quantum Stein results to broader composite-null settings, including compound almost-i.i.d. and arbitrarily varying sources. The proofs use a Wasserstein-robust lemma and weaker assumptions on alternative hypotheses than earlier approaches.

3.1 Compound i.i.d. null hypothesis: Theorem 16 gives a strict generalisation of the compound i.i.d. result under less restrictive assumptions on the alternative-hypothesis families.The same proof technique also extends to compound almost-i.i.d. null hypotheses.
3.1 Compound i.i.d. null hypothesis: The Wasserstein-robust generalised quantum Stein’s lemma is the novel tool making the compound-null proof more general, elegant, and simple.An alternative simplified proof requires closure of the alternative families under permutation twirling.
Proof strategy: The proof strategy combines a converse bound, quasi-concavity estimates, subsequence constructions, and continuity of resource relative entropy under normalised W1 distance.These steps support the required optimisation identities for the compound almost-i.i.d. setting.
3.1 Compound i.i.d. null hypothesis: Compound almost i.i.d. null hypotheses allow each base state in R1 to have an associated W1 almost i.i.d. source, with the exact i.i.d. case as a special instance.Theorem 20 treats these sources when the alternative family is closed under permutation twirl.

4 Proof of Theorem 11

The proof of Theorem 11 adapts information-spectrum, pinching, and likelihood-ratio techniques to Wasserstein almost-i.i.d. null sources. Three propositions provide entropy concentration, pinching stability, and a strong converse for block-product alternatives.

Proof ingredients: Proposition 24 proves an intrinsic information-spectrum asymptotic equipartition property for W1 almost-i.i.d. sources.Under the spectral measure of ρ_n, -1/n log ρ_n converges in probability to the von Neumann entropy of ρ.
Proof ingredients: Proposition 25 shows that this information-spectrum result remains stable under pinching maps with a sub-exponential number of outcomes.Pinching is used to reduce the quantum argument to a classical one.
Proof ingredients: Proposition 26 combines the preceding input with a padded block-product log-likelihood moment estimate to establish a W1-stable strong converse.This replaces the corresponding ingredient in the Hayashi–Yamasaki proof.

4.1 Ouverture. 𝑾1-information-spectrum stability and consequences

This section develops the W1-information-spectrum tools underlying the robust generalised quantum Stein’s lemma, including stability under Wasserstein almost-i.i.d. perturbations and subexponential pinching. It also establishes the W1-stable strong converse and supporting geometric estimates.

Information-spectrum rates remain stable when a source is asymptotically W1-close to an i.i.d. state.This is the first key stability statement used for Wasserstein almost-i.i.d. sources.
Subexponential pinching preserves the information-spectrum rate, enabling logarithmic spectral rounding in the W1 setting.The strengthened pinching result is identified as crucial for the main theorem.
The W1-stable strong converse applies to padded block-product alternatives and remains valid for every η∈(0,1).Its key point is uniform control beyond the η→0 limit.
Lipschitz cutoff operators for fattened subspaces have quantum Lipschitz constant at most 1, making them admissible in the dual W1 formulation.The operators D_V, A_V,r, and B_V,r provide the observables used to obtain lower bounds.
Fattening controls subspace dimension growth and supports uniform small-subspace bounds for Wasserstein almost-i.i.d. sources.These estimates combine geometric growth bounds with typical-subspace arguments.

4.2 First act. Proof of Theorem 11 for individual sources

The individual-source proof adapts the i.i.d. GQSL skeleton to null sources that are only normalised W1-close to an i.i.d. sequence. It replaces the i.i.d. block converse with a W1-stable strong converse and closes achievability through one-step gap contraction and W1 continuity.

4.2.1 Converse: The W1 proof replaces the i.i.d. block converse with a W1-stable padded block-product strong converse.This replacement is required because the null sequence is non-i.i.d., despite being normalised W1-close to an i.i.d. sequence.
4.2.2 Achievability: The achievability construction uses logarithmic rounding, pinching, two testing bounds, and a three-region decomposition.These steps are carried out using the actual source ρ_n before replacement by ρ^⊗n after minimisation over S(n).
4.2.1 Converse: The converse attains the regularised GQSL exponent by testing against padded block-product alternatives drawn from the composite alternative sets.A nearly optimal block state is made full rank, padded into a product alternative, and combined with Proposition 26.
4.2.1 Converse: The strong converse is new even for full-rank i.i.d. alternatives, while full rank is essential and the result cannot extend generally to weak sources.A counterexample exists when supp ρ_n is not contained in supp σ^⊗n, and another rules out the weak-source extension.
4.2.2 Achievability: For achievability, a single gap-contraction lemma replaces the iterative gap-closing construction used in the i.i.d. proof.The argument combines the gap contraction lemma with Wasserstein continuity of the relative entropy of resource.

4.3 Second act. Universality for equiconvergent sources

The section establishes universal tests for equiconvergent null sources by reducing the composite problem to the worst individual null state. This yields a universal sequence of POVMs and completes the proof of Theorem 11.

4.3 Second act. Universality for equiconvergent sources: A minimax identity reduces the composite null problem to the worst individual null state.
4.3 Second act. Universality for equiconvergent sources: The reduction establishes a sequence of tests that is universal within the sequence of families F = (F(n))_n≥1.
4.3 Second act. Universality for equiconvergent sources: The proof completes Theorem 11 by showing the existence of a universal sequence of POVMs.

A Alternative proof of Theorem 16

This section gives an alternative proof of Theorem 16 for permutation-twirl-closed families. It replaces the quantum weak quasi-concavity ingredient with a classical result and uses pinching in the Schur–Weyl decomposition.

A Alternative proof of Theorem 16: The alternative proof applies when the sequence of families is closed under the permutation twirl.
A Alternative proof of Theorem 16: The proof follows the structure of Theorem 16 while replacing Lemma 18 with Lemma 49.
A Alternative proof of Theorem 16: Lemma 49 uses a classical weak quasi-concavity result instead of the quantum formulation used in Lemma 19.
A Alternative proof of Theorem 16: Pinching in the Schur–Weyl basis forces outputs to commute with σ_n and uses only polynomially many projectors, with |P| ≤ (n + 1)^(d^2−1).
A Alternative proof of Theorem 16: The proof combines data processing, Lemma 50, weak/strong-converse duality, smoothing conversion, and the Li–Yao pinching inequality.