PaperlayerView exact revision on arXiv

Source-linked AI summary

Generalised quantum Stein's lemma more robust than ever

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

arXiv:2609.17309v1quant-phcs.ITmath-ph

TL;DR

The paper addresses whether the generalised quantum Stein’s lemma extends beyond i.i.d. and previously studied near-i.i.d. sources. It establishes a Wasserstein-robust version with unchanged asymptotic error exponents and applies it to arbitrarily varying null hypotheses.

  • Problem

    The original lemma assumes perfectly stable, memoryless i.i.d. null sources, whereas realistic procedures may involve correlations and defects among copies.

    Source: §1 · p. 3

  • Method

    The paper establishes a universal generalised quantum Stein’s lemma for null sources uniformly approximating i.i.d. states in normalised quantum Wasserstein distance of order 1.

    Source: §1 · p. 4

  • Results

    Universal tests attain the standard regularised error exponent, while arbitrarily varying null hypotheses yield inf_ρ∈conv(R1) D∞(ρ∥S).

    Source: §1 · p. 4, §1.5 · p. 13

  • Takeaways & Limitations

    Replacing an exact i.i.d. null source with an arbitrary W1 almost i.i.d. source leaves the optimal exponent unchanged.

    Source: §1.5 · p. 12

    The converse requires full-rank support; counterexamples arise when the null support is not contained in the alternative support or for weak sources.

    Source: §4.2.1 · p. 51

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

This section develops a universal generalised quantum Stein’s lemma for Wasserstein almost i.i.d. null sources, using an axiomatic testing framework to establish exponents under broader source and alternative assumptions. It also applies the result to compound and arbitrarily varying null hypotheses.

1.2 Generalised quantum Stein’s lemma: an axiomatic framework

  • The GQSL identifies the optimal type II error exponent against composite alternatives with regularised relative entropy under structural assumptions such as convexity, tensor-product closure, and full-rank-state existence.Composite null constraints are imposed uniformly, while type II error is evaluated in the worst case over the alternative set.Source: §1 · p. 3, §1.2 · pp. 5–6

1 Introduction

  • The work establishes universal tests for every Wasserstein equiconvergent null family that attain the standard GQSL regularised exponent, so unknown source deviations cause no asymptotic performance loss.Individual-source robustness follows as a special case, with further implications for compound i.i.d. and arbitrarily varying hypotheses.Source: §1 · p. 4

1.3 The quantum Wasserstein distance of order 1

  • The quantum Wasserstein distance of order 1 generalises Hamming distance to multipartite quantum states and remains sensitive to local perturbations even when total variation or trace-based distinguishability is large [15].It is built from neighbouring states, coincides with trace distance for one subsystem, and is additive under tensor products [15].Source: §1.3 · pp. 7–8
  • Under (CC) and (REP), relative entropy of resource is continuous in quantum Wasserstein distance, yielding continuity of its regularised version when single-site states converge in trace norm.These continuity properties provide the technical basis for comparing almost i.i.d. null sources with their i.i.d. references.Source: §1.3 · p. 9

1.4 Almost i.i.d. sources

  • Wasserstein almost i.i.d. sources are the paper’s focus because trace-distance convergence is more restrictive and less physically motivated, without necessarily improving robustness guarantees for relevant quantum Shannon protocols [10] [11].The introduction contrasts Wasserstein sources with constant-size defect, MSR, and weakly almost i.i.d. models [6,9,10,30].Source: §1.4 · p. 10
  • Wasserstein almost i.i.d. sources provide a broader robustness guarantee than MSR sources for composite asymmetric hypothesis testing, whereas weakly almost i.i.d. sources offer a less substantial generalisation.The distinction between weakly and Wasserstein almost i.i.d. sources is asymptotic continuity of entropy, and a converse cannot generally be proved for weak sources.Source: §1.4 · p. 11

1.5 Comparison with previous works

  • Theorem 11 simultaneously allows Wasserstein almost i.i.d. perturbations of the null hypothesis and requires only Assumptions 9—CC, TR, and REP—on the alternative family.REP automatically implies FR, so the result broadens admissible null sources while weakening alternative-structure requirements.Source: §1.5 · p. 12
  • Theorem 11 preserves the GQSL exponent D∞(ρ∥S) for arbitrary W1 almost i.i.d. null sources and also constructs tests universal over equiconvergent source families.The individual-source result holds for tests depending on the source sequence, while the universal test need not depend on which source in F(n) occurs.Source: §1.5 · p. 12
  • The proof requires new information-spectrum and block-converse statements stable under W1 almost i.i.d. perturbations because standard i.i.d. equipartition and global pinching do not apply.The paper also strengthens an earlier Wasserstein result from convergence in expectation to convergence in probability in Proposition 24.Source: §1.5 · pp. 12–13
  • Theorem 16 generalises the compound i.i.d. null setting of [31, Theorem 1] under Assumption 9, while Theorem 20 extends it to compound W1 almost i.i.d. null hypotheses with base set R1.The generalisation removes the earlier closure assumptions under partial traces and permutations.Source: §1.5 · p. 13
  • For arbitrarily varying null hypotheses, Theorem 21 gives exponent inf_{ρ∈conv(R1)} D∞(ρ∥S) under Assumption 9 and permutation-twirl closure of the alternative sets.The convex-hull optimisation reflects freedom to vary the product composition across tensor sites.Source: §1.5 · p. 13

2 A Wasserstein-robustgeneralisedquantumStein’slemma

The section develops a Wasserstein-robust generalised quantum Stein’s lemma for sources approaching i.i.d. states, using universal tests under equiconvergence and individual-source perturbations. The result applies under structural conditions on the alternative hypothesis families without requiring permutational symmetry or partial-trace closure.

Robust source framework

  • The framework does not require permutational symmetry or partial-trace closure, accommodating noise-perturbed states whose exact characterisation may be unavailable.Instead, the perturbed states are controlled through their normalised Wasserstein distance from the ideal i.i.d. state.Source: §2 · p. 14
  • Wasserstein equiconvergence is introduced to formalise families of null hypotheses whose members remain uniformly close to an i.i.d. state at every blocklength.The associated universal tests are defined through a general sequence of convex null-hypothesis families and the corresponding testing exponent.Source: §2 · pp. 14–15

Main theorem

  • Theorem 11 establishes the generalised quantum Stein’s lemma for W1 almost i.i.d. sources under Assumption 9 on the alternative hypothesis sequence.The assumptions require convexity and closedness, tensor-product closure, and a full-rank state enabling replacement of individual tensor factors.Source: §2 · pp. 14–15
  • For compact, convex Wasserstein equiconvergent sources along ρ, the theorem guarantees W1-robust universal tests achieving the optimal type II error exponent.These tests remain effective when the source states are uniformly close to ρ⊗n in normalised Wasserstein distance.Source: §2 · p. 15
  • The theorem also covers individual Wasserstein almost i.i.d. sources along ρ, extending idealised i.i.d. testing to perturbed sequences.The perturbation distance is required to converge to zero, while the testing sequence must perform at least as well as in the idealised setting.Source: §2 · pp. 14–15

3 Applications

This section applies the W1-robust generalised quantum Stein’s lemma to compound, almost-i.i.d., and arbitrarily varying null hypotheses. It develops the arbitrarily varying case for finite and infinite sets and clarifies the limitations of the MSR almost-i.i.d. result.

3 Applications

  • The applications build on smoothed max-relative entropy, its duality with hypothesis-testing relative entropy, and weak quasi-concavity.These tools support both the compound and arbitrarily varying extensions, with type methods handling the latter.Source: §3.1 · p. 17, §3.2 · p. 24

3.1 Compound i.i.d. null hypothesis

  • Theorem 16 strictly generalises the compound i.i.d. result of [31, Theorem 1] under the weaker Assumption 9 on alternative hypotheses.Its proof is enabled by the W1 robust generalised quantum Stein’s lemma, Theorem 11, rather than the more restrictive earlier assumptions.Source: §3.1 · p. 17

3.1.1 Proof of Theorem 16

  • The proof of Theorem 16 combines a converse bound, a lower bound under the weaker alternative-hypothesis assumptions, and a new infimum-commutation argument.The infimum commutation is established in Lemma 19 under Assumption 9, replacing the full Brandão–Plenio axioms used in [31].Source: §3.1.1 · p. 18

3.1.2 An extension to the case of a compound almost i.i.d. null hypothesis

  • Theorem 20 extends Theorem 16 to compound almost-i.i.d. null hypotheses whose component sources are asymptotically close to i.i.d. states in normalised W1 distance.The result assumes Assumption 9 and closure of the alternative family under permutation twirl.Source: §3.1.2 · p. 21

3.2 Arbitrarily varying null hypothesis

  • Theorem 21 solves the arbitrarily varying null-hypothesis setting by expressing its Stein exponent through i.i.d. exponents for states in conv(R1).The proof uses permutation twirling, type decompositions, and weak quasi-concavity of the smoothed max-relative entropy.Source: §3.2 · pp. 23–25
  • Theorem 21 is completed for arbitrarily varying null hypotheses, including both finite and infinite null-hypothesis sets.The finite-set case follows from weak/strong converse duality, while the infinite-set case is reduced to finite approximations and completed using continuity of regularised relative entropy of resource.Source: §3.2 · pp. 27–28
  • For infinite null-hypothesis sets, depolarising regularisation moves the set into the interior of the state space, enabling finite outer approximations of its convex hull.The argument fixes 0 < p < 1, constructs a finite outer approximation for every 0 < δ < 1, applies the finite-set theorem, and then takes δ → 0 followed by p → 0.Source: §3.2 · pp. 27–28
  • The robust generalised quantum Stein’s lemma for MSR almost-i.i.d. sources [8, Theorem 4.3] cannot generally complete the computation because the relevant source need not be MSR almost i.i.d.This non-applicability is identified using [10, Proposition 25].Source: §3.2 · p. 28

4 Proof of Theorem 11

The section develops three technical ingredients that extend the proof of the generalised quantum Stein’s lemma from i.i.d. to W1 almost i.i.d. null sources. These ingredients establish entropy concentration, pinching stability, and a W1-stable strong converse for block-product alternatives.

4 Proof of Theorem 11

  • The proof treats an arbitrary W1 almost i.i.d. null source (ρ_n)_n converging toward a fixed state ρ, requiring direct information-spectrum, pinching, and likelihood-ratio arguments.This isolates the technical tools needed to adapt the proof in Ref. to a non-i.i.d. null source.Source: §4 · p. 28
  • Proposition 24 proves an intrinsic information-spectrum asymptotic equipartition property: −1/n log ρ_n converges in probability to the von Neumann entropy of ρ.The result strengthens the convergence in expectation of [10, Proposition 21].Source: §4 · p. 29
  • Proposition 25 shows that the information-spectrum result remains stable under any pinching map with a sub-exponential number of outcomes.This enables the proof to reduce the quantum setting to the classical case.Source: §4 · p. 29
  • Proposition 26 combines the information-spectrum input with a padded block-product log-likelihood moment estimate to establish a W1-stable strong converse for block-product alternatives.This replaces the strong-converse ingredient used in the Hayashi–Yamasaki proof [3, Lemma S6].Source: §4 · p. 29

4.1 Ouverture. 𝑾1-information-spectrum stability and consequences

This section develops the W1-information-spectrum, block-product, moment, and testing tools for Wasserstein almost i.i.d. sources. It establishes stability and robustness results and combines pinching with classical hypothesis-testing arguments to prove the main proposition.

4.1 Ouverture. 𝑾1-information-spectrum stability and consequences

  • Proposition 24 establishes information-spectrum stability for any source asymptotically close to an i.i.d. state in quantum Wasserstein distance W1.Its proof separately controls upper and lower spectral tails using typical-subspace dimension bounds and a uniform small-subspace estimate.Source: §4.1 · p. 30, §4.1.3 · pp. 37–38

4.1.4 Proof of Proposition 25

  • Proposition 25 shows that pinching a Wasserstein almost i.i.d. source into a subexponential number of blocks preserves its information-spectrum rate.This handles the polynomially many bins produced by logarithmic spectral rounding, despite W1 contractivity being guaranteed only for single-system channels.Source: §4.1 · p. 30, §4.1.4 · pp. 38–40

4.1 Ouverture. 𝑾1-information-spectrum stability and consequences

  • Proposition 26 provides the W1-stable strong converse for padded block-product alternatives needed in both converse and achievability arguments.The result applies to a fixed block length and full-rank block alternative, and its bound remains valid for every η∈(0,1), not only as η→0+.Source: §4.1 · p. 31

4.1.1 The dual formulation of the quantum Wasserstein distance of order 1

  • The dual formulation characterizes quantum W1 through site dependence and Lipschitz constants, yielding positive-operator bounds used throughout the section.The section defines the site dependence of self-adjoint operators and invokes the corresponding dual formula for the quantum Wasserstein distance.Source: §4.1.1 · pp. 31–32

4.1.2 Fattening, distance cutoffs and blowing-up

  • Fattening and distance cutoffs produce 1-Lipschitz observables, blowing-up estimates, dimension-growth bounds, and uniform small-subspace control for Wasserstein almost i.i.d. sources.These tools support typical-projector arguments and ensure that fattening by linearly growing radii does not retain asymptotically significant mass outside the relevant subspaces.Source: §4.1.2 · pp. 33, 35, 36

4.1.5 Technical ingredients needed for the proof of Proposition 26

  • The strong-converse proof combines binary-rounded pinching, a noncommutative law of large numbers for padded block products, projector tests, and an extension to general POVMs.Binary rounding yields only polynomially many spectral values, making the noncommuting problem reducible to a commuting pair before the law-of-large-numbers and POVM steps.Source: §4.1.5 · pp. 40–41
  • 4.1.5 establishes padded block-product and moment estimates for W1-almost-i.i.d. sources, using blocking, negligible padding, and [12, Lemma 2].The construction relies on the source remaining Wasserstein almost i.i.d. along ρ^⊗m by [11, Lemma 31] and weakly almost i.i.d. by [10, Corollary 16].Source: §4.1.5 · pp. 42, 44
  • 4.1.5 derives a padded block-product strong converse by converting optimal tests into threshold projectors and applying the preceding moment bounds.The proof uses compactness to select an optimal test, thresholds it into a projector, and then invokes the hypothesis of Lemma 39.Source: §4.1.5 · pp. 44–45

4.1.6 Proof of Proposition 26

  • 4.1.6 proves Proposition 26 by combining binary pinching, refined commuting projectors, operator monotonicity, and the moment estimate.The refined pinching construction has at most 2(n+1) projectors, enabling classical Markov and likelihood-ratio bounds after commuting reductions.Source: §4.1.6 · pp. 46–47
  • 4.1.6 translates the quantum projector conditions into classical events and applies [43, Lemma 4.1.2], essentially equivalent asymptotically to the Neyman–Pearson lemma [44].The resulting implication is obtained by identifying the pinched quantum states with classical random variables and taking δ arbitrarily small.Source: §4.1.6 · pp. 48–49

4.2 First act. Proof of Theorem 11 for individual sources

For a Wasserstein almost i.i.d. source, the proof establishes Theorem 11 by combining a W1-stable padded block-product strong converse with a one-step gap-contraction achievability argument and Wasserstein continuity. The argument adapts the i.i.d. generalised quantum Stein lemma proof to non-i.i.d. null sequences while preserving its core testing structure.

4.2 First act. Proof of Theorem 11 for individual sources

  • The converse extends the i.i.d. proof of Hayashi–Yamasaki by replacing its block argument with W1 information-spectrum, pinching-stability, and block-law-of-large-numbers ingredients.Full-rank alternatives are essential; counterexamples show the result need not extend without the support condition or to weak sources, even when the alternative has full support.Source: §4.2 · pp. 49–50, §4.2.1 · p. 51

4.2.1 Converse

  • The converse bounds the exponent by testing against full-rank padded block-product alternatives in S(n), using the W1-stable strong converse of Proposition 26.A nearly optimal block alternative is full-rank approximated, padded into a block product, and used because the composite alternative cannot have a larger exponent than this particular sequence.Source: §4.2.1 · pp. 50–52

4.2.2 Achievability

  • Achievability replaces the iterative update construction of with one application of the gap contraction lemma, followed by Wasserstein continuity of the relative entropy of resource.The proof retains logarithmic rounding, pinching, threshold testing, and a three-region decomposition, while Proposition 26 replaces the i.i.d. block strong converse.Source: §4.2 · p. 49, §4.2.2 · pp. 52–53

4.3 Second act. Universality for equiconvergent sources

The section proves a universal sequence of tests for equiconvergent sources by reducing the composite null problem to the worst individual null state through a minimax identity. This establishes universal POVMs within the sequence of families F = (F(𝑛))𝑛≥1.

4.3 Second act. Universality for equiconvergent sources

  • A minimax identity reduces the composite null problem to the worst individual null state, enabling universal testing for equiconvergent sources.The proof’s key idea is this reduction.Source: §4.3 · p. 59
  • For each n, the construction uses compact and convex families F(𝑛) and selects 𝜌∗.These properties are assumed by hypothesis.Source: §4.3 · p. 59
  • Applying the individual-source statement to an individual Wasserstein almost i.i.d. source yields the universal testing conclusion.The source is considered along 𝜌.Source: §4.3 · p. 59
  • The proof establishes a sequence of POVMs (𝐸𝑛)𝑛≥1 that is universal within the family sequence F = (F(𝑛))𝑛≥1.This completes the proof of Theorem 11.Source: §4.3 · p. 59

A Alternative proof of Theorem 16

Under permutation-twirl closure, the section gives a simpler alternative proof of Theorem 16 by replacing the quantum weak quasi-concavity step with a classical one. The proof uses permutation invariance, Schur–Weyl decomposition, and polynomial-size pinching.

Alternative proof

  • Permutation-twirl closure enables a simpler alternative proof of Theorem 16 using a classical formulation of weak quasi-concavity.Theorem 48 assumes Assumption 9 together with closure of every family under the permutation twirl.Source: A Alternative proof of Theorem 16 · p. 63
  • The proof follows Theorem 16’s structure, replacing Lemma 18 with Lemma 49 and invoking the classical weak quasi-concavity result of.The classical result constructs a pointwise-maximum acceptance function from optimal tests for the component distributions.Source: A Alternative proof of Theorem 16 · pp. 63–64

Proof of Lemma 49

  • Permutation invariance is imposed without loss of generality on the alternative states by the data-processing inequality and permutation-twirl closure.The proof considers states in the convex hull of i.i.d. states, whose components are permutation invariant.Source: A Alternative proof of Theorem 16 · p. 64
  • The Schur–Weyl decomposition expresses relevant states as convex combinations of at most (n + 1)^(d^2−1) + 1 = poly(n) i.i.d. terms.The proof then diagonalizes Schur–Weyl blocks and applies a pinching channel that commutes with the alternative state.Source: A Alternative proof of Theorem 16 · pp. 64–65
  • The pinching channel has polynomially many projectors, with |P| ≤ (n + 1)^(d^2−1), enabling the entropy bounds that complete Lemma 49.The proof combines data processing, Lemma 50, weak/strong-converse duality, smoothing conversion, and the Li–Yao pinching inequality [47, Proposition 13].Source: A Alternative proof of Theorem 16 · pp. 65–66
Loading 2609.17309v1…