A Generalized quantum Stein lemma on von Neumann algebras

Technical Summary: A Generalized Quantum Stein Lemma on von Neumann Algebras

Problem Statement
The paper addresses the problem of hypothesis testing for quantum states within the framework of arbitrary von Neumann algebras. Specifically, it seeks to determine the optimal asymptotic type-II error exponent for distinguishing a fixed i.i.d. null state $\rho^{\otimes n}$ against a composite alternative hypothesis defined by a convex, tensor-stable family of states $\mathcal{F} = (F_n)_{n \ge 1}$.

In finite-dimensional systems, the Quantum Stein Lemma establishes that the optimal error exponent is the relative entropy $D(\rho\|\sigma)$. The "Generalized Quantum Stein Lemma" (GQSL) extends this to composite hypotheses (families of states). While previous work by Hayashi and Yamasaki established this for finite-dimensional systems under specific closure and invariance assumptions, this paper aims to extend the result to infinite-dimensional systems and general von Neumann algebras (including types II and III), removing assumptions regarding norm-closedness, permutation invariance, or closure under partial traces.

Methodology
The proof strategy relies on a combination of integral representations of relative entropy, modular theory, and convex analysis:

1. Integral Representation of Relative Entropy: The author utilizes Frenkel's integral representation of relative entropy, recently established for von Neumann algebras. This expresses the relative entropy $D(\omega\|\sigma)$ as an integral of "hockey-stick" divergences $H_t(\omega\|\sigma)$:
   $$D(\omega\|\sigma) = \int_1^\infty H_t(\omega\|\sigma) \frac{dt}{t} + \int_1^\infty H_t(\sigma\|\omega) \frac{dt}{t^2}$$
   This allows the asymptotic analysis of relative entropy rates to be translated into the analysis of hockey-stick quantities.

2. Modular Testing Bounds: The proof employs a modular testing bound derived from Ogata's generalized Powers–Størmer inequality. This relates the hockey-stick divergence to the quadratic form of the relative modular operator $\Delta_{\sigma,\omega}$, specifically utilizing the quantity $q_{\omega,\sigma}(s) = \|\Delta_{\sigma,\omega}^{s/2}\xi_\omega\|^2$.

3. Convex Minimax Arguments: To handle the composite nature of the alternative hypothesis (where the test must work for all $\sigma \in F_n$), the author applies Sion's minimax theorem. This establishes a duality between the infimum of hockey-stick divergences over the convex set $F_n$ and the existence of a single optimal test operator $T$.

4. Fixed-Block Convergence: The author constructs a sequence of "block comparator" states using a finite-entropy reference state $\tau$ and a specific block state $\sigma_k$. They demonstrate that the testing profiles (hockey-stick quantities) for these constructed states converge in $L^1$ to a step function determined by the entropy rate $R$.

Key Results
Under the assumption that there exists a one-copy alternative state $\tau \in F_1$ such that the relative entropy $D(\rho\|\tau)$ is finite, the paper proves the following:

1. Existence of the Regularized Relative Entropy Rate: The limit
   $$R := \lim_{n \to \infty} \frac{1}{n} \inf_{\sigma \in F_n} D(\rho^{\otimes n} \| \sigma)$$
   exists and equals the infimum over all $n$.

2. Sharp Threshold for Hockey-Stick Quantities: The composite hockey-stick quantity $Q_n(r) = \inf_{\sigma \in F_n} H_{e^{nr}}(\rho^{\otimes n} \| \sigma)$ exhibits a sharp threshold behavior:
   $$\lim_{n \to \infty} Q_n(r) = \begin{cases} 1 & \text{if } r < R \\ 0 & \text{if } r > R \end{cases}$$

3. Generalized Stein Lemma: For any type-I error tolerance $\varepsilon \in (0, 1)$, the hypothesis testing relative entropy converges to the regularized relative entropy rate:
   $$\lim_{n \to \infty} \frac{1}{n} D_H^\varepsilon(\rho^{\otimes n} \| F_n) = R$$

4. Strong Converse: If a sequence of tests $(T_n)$ achieves a type-II error exponent strictly greater than $R$ (i.e., $\liminf -\frac{1}{n} \log B_n(T_n) > R$), then the type-I error $\alpha_n(T_n)$ necessarily converges to 1.

Significance and Claims
The paper claims to extend the Generalized Quantum Stein Lemma to the most general setting of von Neumann algebras without requiring:

• Finite dimensionality.
• Norm-closedness of the family $F_n$.
• Permutation invariance of the family.
• Closure under restrictions to tensor factors.
• Faithfulness of the null state $\rho$ or the reference state $\tau$.
• $\sigma$-finiteness of the algebra.

The primary contribution is the removal of structural assumptions (like permutation invariance and norm-closure) that were present in previous finite-dimensional proofs, while simultaneously handling infinite-dimensional systems (Type II and III algebras). The author notes that the finite relative entropy condition $D(\rho\|\tau) < \infty$ is a necessary constraint for the proof, analogous to the full-rank assumption in finite dimensions, but they explicitly state that the theorem does not apply to families where no such finite-entropy reference exists.

The proof strategy is distinct from previous works, relying on the integral representation of relative entropy via hockey-stick divergences rather than the specific techniques used in finite-dimensional settings. The results recover the Hayashi–Yamasaki theorem for finite-dimensional systems with full-rank references as a special case.