{"id":"cb089c0d-a899-44cd-9fbf-956140107120","arxiv_id":"2607.05195","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"The f_0-divergence defined via Jordan decomposition integrals coincides with Araki's relative entropy on arbitrary von Neumann algebras, extending Frenkel's finite-dimensional formula.","lead":"The paper defines hockeystick and f-divergences for general von Neumann algebras using Jordan decompositions, and proves that the f_0-divergence (with f_0(t) = t ln t) equals Araki's relative entropy for all von Neumann algebras. This matters because it provides a new, simpler route to relative entropy in settings like quantum field theory where infinite-dimensional algebras are unavoidable.","discovery_kind":"unclear","skeptic_critique":{"model":"glm-5.2","headline":"Prop 4.8 (operator integral equality) is applied to non-invertible cutoff densities h_{ψ,n} in Thm 3.17, but is proven only under Assumption 4.2 requiring A^{-1} bounded. Gap is minor and fixable by A→A+ε approximation.","rationale":"The reader correctly identified the application of Prop 4.8 in Thm 3.17 as the most fragile step, but framed the concern as extension to unbounded operators. The specific gap is more precise: even the cutoff h_{ψ,n} is not invertible in M (it has a kernel on the complement of E_{h_ψ}[n⁻¹,n]), so Assumption 4.2 is violated for every finite n, not just in the limit. The proof of Prop 4.7/4.8 uses invertibility of A via Lemma 4.13 (resolvent bounds) and the contour deformation argument. However, this is a minor gap: the A→A+ε approximation is standard, the necessary continuity estimates are already in Lemma 4.1, and the scalar computation on the kernel of A confirms the integrals agree there (both reduce to B·∫f''_a). The rest of the proof is sound: the f_a→f_0 monotone convergence, the norm-continuity of D_{f_a} (Thm 2.13b), the dominated convergence for relative entropy (Prop 3.4d), and the Haagerup reduction (Thm 3.18→3.19) are all correctly applied. The independent proof by [59] provides external corroboration. The verdict of ACCEPT with HIGH confidence is appropriate.","tokens_in":50524,"tokens_out":16150,"duration_ms":302355,"concrete_test":"Verify that for bounded A ≥ 0 (not necessarily invertible) and bounded B ∈ L¹(M,τ), both I_f(A+ε,B) → I_f(A,B) and J_f(A+ε,B) → J_f(A,B) as ε→0⁺, using the L¹-norm continuity estimates from Lemma 4.1. Specifically, check that ‖(B−t(A+ε))⁺ − (B−tA)⁺‖_{L¹} → 0 uniformly in t on compact sets (by continuity of the positive-part map on L¹(M,τ) ≅ M*), and that ‖C_{A+ε,B}(s) − C_{A,B}(s)‖_{L¹} → 0 for each s > 0 (by resolvent identities and Hölder). If both converge, dominated convergence closes the gap and Prop 4.8 extends to non-invertible A.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Thm 3.17, the paper applies Prop 4.8 to A = h_{ψ,n} = E_{h_ψ}[n⁻¹,n] h_ψ and B = h_{φ,n}. The paper states \"h_{ψ,n}⁻¹ is bounded as well,\" but h_{ψ,n} vanishes on the complement of E_{h_ψ}[n⁻¹,n], so it is not invertible in M. Assumption 4.2 (under which all of Sec 4.2, including Prop 4.8, is proven) requires A, A⁻¹ positive and bounded — i.e., A ≥ a > 0. The proof of Prop 4.7 (the f=1 base case) uses this invertibility in an essential way: Lemma 4.13 requires C > 0 boundedly invertible for resolvent bounds, and the contour deformation in Prop 4.7 relies on these bounds to control the arcs at infinity and near the origin. When A has a kernel, these bounds fail. The gap is that Prop 4.8 is invoked outside its stated hypotheses. The fix is straightforward: replace A by A+ε (which is invertible since A ≥ 0), apply Prop 4.8, and take ε→0 using the L¹-continuity estimates already established in Lemma 4.1. Both I_f(A+ε,B)→I_f(A,B) and J_f(A+ε,B)→J_f(A,B) follow from continuity of the positive-part map on L¹(M,τ) and of multiplication by bounded elements. This approximation step is standard but not written out. The concern does not threaten the central claim: the gap is in a technical lemma, the fix uses machinery already in the paper, and the result has independent corroboration from [59].","agreement_with_reader":"partial"},"referee_report":{"model":"glm-5.2","summary":"This paper defines and analyzes hockeystick divergences and f-divergences for normal positive functionals on general von Neumann algebras. The f-divergences are defined via an integral representation over hockeystick divergences (Def. 2.11), which in turn are built on the Jordan decomposition of selfadjoint normal functionals (Def. 2.5). This approach yields clean proofs of standard properties—monotonicity, convexity, lower semicontinuity, data processing, Pinsker's inequality—collected in Theorem 2.13. The central result is Theorem 3.19, proving that the f_0-divergence (f_0(t) = t ln t) coincides with Araki's relative entropy for arbitrary von Neumann algebras. The proof proceeds in two stages: first for semifinite algebras (Thm. 3.17), where the key technical ingredient is an equality of two operator integrals (Prop. 4.8), and then for general algebras via Haagerup reduction (Thm. 3.18). A complementary result (Thm. 3.10) establishes the multi-shot regularization limit via modular bounds. The paper extends Frenkel's finite-dimensional formula and unifies several recent results [25–29, 35, 40, 41].","tokens_in":51426,"tokens_out":1623,"duration_ms":79459,"significance":"The main contribution—establishing D^M_{f_0} = S^M_{rel} for general von Neumann algebras—is a significant result that connects the Jordan-decomposition-based approach to divergences with Araki's relative modular theory. The integral representation (1.5) offers a new, operationally motivated perspective on relative entropy, particularly relevant for quantum field theory where type III algebras arise. The proofs of the operator integral equality (Sec. 4.2) via analytic continuation and contour deformation, and the extension to general algebras via Haagerup reduction, are substantial technical achievements. The authors appropriately acknowledge independent simultaneous work by Koßmann, Schwonnek, Liu, and Cheng [59] and credit the Haagerup reduction argument to van Luijk. The paper also provides a clear and self-contained development of the f-divergence framework, with the data processing inequality and Pinsker's inequality derived efficiently from the hockeystick representation.","major_comments":[{"comment":"Theorem 3.17, p. 34 (proof): The proof applies Proposition 4.8 to A = h_{ψ,n} = E_{h_ψ}[n⁻¹,n] h_ψ and B = h_{φ,n}. The text states 'h_{ψ,n}⁻¹ is bounded as well,' but h_{ψ,n} vanishes on the complement of E_{h_ψ}[n⁻¹,n], so it is not invertible in M. Assumption 4.2 (under which Prop. 4.8 is proven) requires A, A⁻¹ positive and bounded, i.e., A ≥ a > 0. The proof of Prop. 4.7 (the f=1 base case) uses this invertibility in an essential way: Lemma 4.13 requires C > 0 boundedly invertible for resolvent bounds, and the contour deformation in Prop. 4.7 relies on these bounds to control arcs at infinity and near the origin. When A has a kernel, these bounds fail. The gap is that Prop. 4.8 is invoked outside its stated hypotheses. The fix is straightforward: replace A by A+ε (which is invertible since A ≥ 0), apply Prop. 4.8, and take ε→0 using the L¹-continuity estimates already established in","section":null},{"comment":"Lemma 4.1. Both I_f(A+ε,B)→I_f(A,B) and J_f(A+ε,B)→J_f(A,B) follow from continuity of the positive-part map on L¹(M,τ) and of multiplication by bounded elements. This approximation step is standard but not written out. The concern does not threaten the central claim: the gap is in a technical lemma, the fix uses machinery already in the paper, and the result has independent corroboration from [59]. Nevertheless, the authors should either (a) add a remark after Prop. 4.8 stating that the result extends to positive A with nontrivial kernel by the A→A+ε approximation, with a brief justification, or (b) adjust the proof of Thm. 3.17 to explicitly perform this approximation. As it stands, there is a gap between the stated hypotheses of Prop. 4.8 and its application in the proof of the paper's central theorem.","section":null}],"minor_comments":[{"comment":"Section 1, p. 5: The display equations for J(h₁,h₂) and I(h₁,h₂) use notation that is not fully defined at that point (e.g., the square-root notation √_s/(h₁+s)). A forward reference to Section 4 or a brief explanatory remark would help the reader.","section":null},{"comment":"Definition 2.11, Eq. (2.13): The integral is stated as a Lebesgue integral with values in [0,∞], but the correction term ϕ₁(1)−ϕ₂(1) can make the overall value negative. The text notes this, but the intermediate claim that the integral itself takes values in [0,∞] could be stated more precisely to avoid momentary confusion.","section":null},{"comment":"Proposition 2.7, item c): The limit lim_{t→∞} E^M_t(ϕ₁|ϕ₂) = ϕ₁(s₂⊥) is proven, but the proof uses a subsequential compactness argument. It would help to explicitly note that the full limit exists (not just subsequential) because E^M_t is monotone decreasing in t.","section":null},{"comment":"Theorem 3.10, Eq. (3.7): The lower bound involves δ_r for r∈(0,1) and the optimization over T₁, T₂. The choice r(t) is piecewise constant; a brief remark on how sharp this bound is, or whether smoother choices of r(t) yield improvements, would add context.","section":null},{"comment":"Section 4.2: The standing Assumption 4.2 lists A, B, C, D ∈ L¹(M,τ) and x ∈ M, but C and D do not appear in the statements of Prop. 4.4–4.8. If they are used in auxiliary estimates, this should be clarified; otherwise they can be removed from the assumption.","section":null},{"comment":"Reference [59] is listed as 'To appear' with title 'Device-independent Quantum Key Distribution in the commuting operator framework.' This title does not match the content (equality of f₀-divergence and relative entropy). The authors should verify that the reference is correct and complete.","section":null},{"comment":"Page 3, line 2: 'hockeystick' is written as one word in some places and 'hockey stick' (two words) might be more standard. This is purely cosmetic; the authors should pick one convention and use it consistently.","section":null}],"recommendation":"minor_revision","confidential_remarks":"The independent simultaneous result by Koßmann, Schwonnek, Liu, and Cheng [59] is acknowledged. The authors should ensure that the final published version coordinates with [59] to present both contributions fairly. The Haagerup reduction argument is credited to van Luijk; the authors should confirm that appropriate acknowledgment or co-authorship arrangements are in place. The paper is well-suited for a serious operator algebras or mathematical physics journal."},"author_rebuttal":{"model":"glm-5.2","summary":"We thank the referee for the careful reading and the identification of a genuine gap in the application of Proposition 4.8. The referee's diagnosis is correct, and we will implement the suggested fix.","responses":[{"response":"We thank the referee for identifying this gap, which is entirely correct. In the proof of Theorem 3.17, the operator $h_{ψ,n} = E_{h_ψ}[n^{-1},n] h_ψ$ is not invertible in $M$ because it vanishes on the complement of $E_{h_ψ}[n^{-1},n]$. Proposition 4.8 is proven under Assumption 4.2, which requires $A ≥ a > 0$ (i.e., $A$ is boundedly invertible), so invoking it with $A = h_{ψ,n}$ is indeed outside the stated hypotheses. The sentence claiming that $h_{ψ,n}^{-1}$ is bounded is misleading: while $h_{ψ,n} ≥ n^{-1}$ on its support, it is not invertible in $M$ itself. We will correct this in the revision. The fix proposed by the referee is the natural one and uses only machinery already present in the paper. Specifically, we will replace $A = h_{ψ,n}$ by $A_ε = h_{ψ,n} + ε$ (which satisfies $A_ε ≥ ε > 0$ and is thus boundedly invertible), apply Proposition 4.8 with $A_ε$, and then pass to the limit $ε → 0$. The convergence $I_f(A_ε, B) → I_f(A, B)$ and $J_f(A_ε, B) → J_f(A, B)$ in $L^1(M, τ)$ follow from the continuity of the positive-part map $X ↦ X^+$ on $L^1(M, τ)$ (noted in the proof of Lemma 4.1, using the isometric isomorphism $L^1(M,τ) ≅ M_*$ and the norm continuity of $ϕ ↦ |ϕ|$ from [56]) and the norm continuity of multiplication by bounded elements of $M$. We will implement option (b): the proof of Theorem 3.17 will be adjusted to explicitly perform this $ε$-approximation, and a brief remark will be added after Proposition 4.8 noting that the result extends to positive $A$ with nontrivial kernel by this approximation argument. We agree with the referee that this does not affect the validity of the central result, which also has independent corroboration from [59].","revision_made":"yes","referee_comment":"Theorem 3.17, p. 34 (proof): The proof applies Proposition 4.8 to A = h_{ψ,n} = E_{h_ψ}[n⁻¹,n] h_ψ and B = h_{φ,n}. The text states 'h_{ψ,n}⁻¹ is bounded as well,' but h_{ψ,n} vanishes on the complement of E_{h_ψ}[n⁻¹,n], so it is not invertible in M. Assumption 4.2 (under which Prop. 4.8 is proven) requires A, A⁻¹ positive and bounded, i.e., A ≥ a > 0. ... The gap is that Prop. 4.8 is invoked outside its stated hypotheses. The fix is straightforward: replace A by A+ε (which is invertible since A ≥ 0), apply Prop. 4.8, and take ε→0 using the L¹-continuity estimates already established in Lemma 4.1."}],"tokens_in":50481,"tokens_out":2273,"duration_ms":42073,"standing_objections":[]},"desk_editor":{"model":"glm-5.2","letter":"The main result is Theorem 3.19: for any von Neumann algebra M and any two normal positive functionals, the f_0-divergence D^M_{f_0} equals Araki's relative entropy S^M_{rel}. This extends Frenkel's integral formula from matrix algebras to the general case, including type III algebras relevant to quantum field theory. The result is genuinely new — prior work covered matrices, B(H), and approximately finite-dimensional algebras, but not the general setting. The paper also provides a clean, self-contained development of hockeystick and f-divergences for general von Neumann algebras (Section 2), with all standard properties — monotonicity, convexity, data processing, state discrimination — derived efficiently from Jordan decomposition. The data processing inequality proof is notably short. The modular bounds in Section 3.1 and the regularization result (Theorem 3.10) are solid and give a first route connecting the two quantities, even though that route alone doesn't yield equality. The equality proof goes through semifinite algebras (Theorem 3.17) via an operator integral identity (Proposition 4.8), then extends to general M by Haagerup reduction. The operator integral equality I_f(A,B) = J_f(A,B) in Section 4 is the technical heart, and the proof for constant f=1 (Proposition 4.7) via analytic continuation and contour deformation is carefully done. The extension to general continuous f by polynomial approximation is standard. The stress-test concern about Proposition 4.8 being applied to non-invertible cutoff densities h_{ψ,n} in Theorem 3.17 is real but minor. The spectral cutoff h_{ψ,n} = E_{h_ψ}[n^{-1},n] h_ψ vanishes on the complement of the spectral projection, so it is not invertible in M, while Assumption 4.2 requires A ≥ a > 0. The fix is the standard A → A+ε approximation, and the L^1-continuity estimates needed for the limit are already in Lemma 4.1. This step should be written out explicitly, but it does not threaten the main result. The simultaneous independent proof by another group (reference [59]) provides external corroboration. This paper is for operator algebraists and mathematical physicists working with quantum information in infinite-dimensional settings. It deserves a serious referee who can check the contour deformation arguments in Section 4.2 carefully.","headline":"D_{f_0} = S_{rel} for general von Neumann algebras, including type III, via Jordan decomposition and operator integral equality — a genuine extension of Frenkel's formula.","tokens_in":51635,"tokens_out":610,"would_cite":true,"duration_ms":72178,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["46L10","46L53","81P45","94A15"],"pacs":[],"model":"glm-5.2","headline":"f-divergence equals relative entropy in general von Neumann algebras","keywords":["f-divergence","hockeystick divergence","relative entropy","von Neumann algebra","Araki relative entropy","Jordan decomposition","data processing inequality","quantum hypothesis testing"],"falsifier":"Exhibit a von Neumann algebra and two normal positive functionals for which the f_0-divergence integral formula yields a value different from Araki's relative entropy, which would require a failure of the operator integral equality or the approximation step for some unbounded density configuration.","tokens_in":50699,"feed_emoji":"∫","tokens_out":883,"duration_ms":78910,"temperature":0.7,"pith_summary":"This paper defines hockeystick divergences and f-divergences for normal positive functionals on arbitrary von Neumann algebras, building them from the Jordan decomposition of selfadjoint functionals rather than from relative modular theory. The hockeystick divergence at likelihood ratio t is simply the norm of the positive part of the difference phi_1 - t*phi_2, which directly generalizes the classical total-variation-based quantity and has an operational meaning in quantum hypothesis testing as the excess success probability of the optimal test. General f-divergences are then defined as integrals of hockeystick divergences weighted by the second derivative of a convex function f, mirroring the classical integral representation of Csiszár divergences. This construction yields all standard properties—monotonicity, convexity, semicontinuity, data processing inequality, state discrimination—in a unified and comparatively simple way across all types of von Neumann algebras, including the type III algebras that arise in quantum field theory.","feed_headline":"Quantum relative entropy equals an integral of hockeystick divergences","feed_subtitle":"The equality holds for all von Neumann algebras, including type III, unifying state distinguishability measures with modular theory.","key_machinery":"Hockeystick divergence defined as the norm of the positive part of (phi_1 - t*phi_2) via Jordan decomposition; f-divergence as an integral of hockeystick divergences weighted by f''; equality of two operator integrals I_f(A,B) and J_f(A,B) proven via analytic continuation and contour deformation; Haagerup reduction extending from semifinite to general von Neumann algebras","core_discovery":"The central result is that the f_0-divergence (the f-divergence for the information function f_0(t) = t ln t) coincides exactly with Araki's relative entropy for any von Neumann algebra M and any two normal positive functionals. This is proven in two stages: first for semifinite algebras by establishing the equality of two operator integrals—one built from the positive part of density differences, the other from resolvent-based sandwiched expressions—via analytic continuation and contour deformation, then approximating unbounded densities by spectral cutoffs; second, for general algebras, by passing through finite subalgebras of a crossed product (Haagerup reduction) and using martingale con","pith_inferences":[],"forward_implications":["Relative entropy in quantum field theory (where type III von Neumann algebras are standard) can be computed via the integral formula involving hockeystick divergences, potentially sidestepping direct modular theory calculations.","The data processing inequality for relative entropy follows from a few lines of Jordan decomposition arguments, offering a simpler proof route than existing approaches.","The multi-shot regularization of the f_0-divergence equals Araki's relative entropy, connecting the integral representation to asymptotic error rates in quantum hypothesis testing.","Pinsker's inequality and other bounds on relative entropy follow directly from the hockeystick divergence framework, providing computationally accessible estimates."],"fun_headline_variants":["Araki relative entropy equals f-divergence integral on von Neumann algebras","f-divergence integrals on von Neumann algebras match Araki relative entropy","Integral f-divergences on von Neumann algebras recover Araki relative entropy","General von Neumann algebras: f-divergence integrals equal Araki entropy","Araki entropy recovered via f-divergence integrals on von Neumann algebras"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The equality of two operator integrals is first established for bounded operators with bounded inverses and then extended to unbounded L^1 densities by spectral cutoff approximation; if the norm continuity of the regularized divergence or the monotone convergence step fails for pathological unbounded operators, the semifinite case—and hence the general case—would not hold.","fun_headline_variants_meta":{"raw":{"variants":["Araki relative entropy equals f-divergence integral on von Neumann algebras","f-divergence integrals on von Neumann algebras match Araki relative entropy","Integral f-divergences on von Neumann algebras recover Araki relative entropy","General von Neumann algebras: f-divergence integrals equal Araki entropy","Araki entropy recovered via f-divergence integrals on von Neumann algebras"]},"model":"glm-5.2","effort":"high","cost_usd":0.0,"raw_usage":{"total_tokens":1278,"prompt_tokens":509,"completion_tokens":769,"prompt_tokens_details":null},"tokens_in":509,"tokens_out":769,"duration_ms":29310,"temperature":1.0,"reasoning_tokens":686,"cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-07T23:56:32.168324+00:00","model_set":{"reader":"glm-5.2"},"falsifier":"Exhibit a von Neumann algebra and two normal positive functionals for which the f_0-divergence integral formula yields a value different from Araki's relative entropy, which would require a failure of the operator integral equality or the approximation step for some unbounded density configuration.","supporting_citations":[],"review_version":1}