{"id":"c7bce0fb-f486-45cb-aafc-836b8231d670","arxiv_id":"2502.05627","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"If a convex function is operator convex along every line, its epigraph's natural log-barrier is self-concordant, giving optimal barriers for sandwiched Rényi entropies.","lead":"This paper proves a new shortcut for checking when a convex optimization problem can be solved by fast interior-point methods. The shortcut turns a property called operator convexity into a proof that the standard log-barrier works, and it yields the first such barrier for sandwiched Rényi entropies from quantum information.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified. The proof is internally coherent and the main external input, Hiai's analytic-continuation lemma, is cited with precise parameters; a re-derivation is still worth doing.","rationale":"The reader's weakest-assumption identification points to Lemma 4.2, and I agree that this is the least independently verified input of the paper. However, I do not regard it as a genuine correctness risk: the citation is specific, the parameter substitution is stated explicitly, and the surrounding argument in Section 4.1 correctly leverages the resulting operator monotonicity of Fhat to obtain operator concavity of F on (0,1) and (-1,0), then extends to the full interval through Lemma 4.4. I also checked the composition step in Theorem 3.1: the integral representation of operator concave functions and the exchange of differentiation and integration are legitimate, and the sign handling is correct because F''(0) <= 0 flips the inequality. The perspective argument for alpha in [1,2] is more lengthy but internally consistent: the block-matrix representation and the positive linear map Phi correctly recover -Psi_alpha(X,Y), and the compatibility of the composed perspective follows from Theorem 4.5. The optimality claims are backed by the diagonal restriction argument and the cited lower-bound results. Since the central claim is mathematically coherent and the external dependency is a precise citation rather than an unsupported assertion, the appropriate verdict remains ACCEPT with no change.","tokens_in":21695,"tokens_out":27506,"duration_ms":269826,"concrete_test":"Independently re-derive Lemma 4.2: set Phi = Psi = I, p = (1-alpha)/(2alpha), q = 1, and s = alpha in Hiai (2013, Theorem 2.1 and its proof), and verify that Im Fhat(z) > 0 for all z in C+ whenever X±H >= 0 and Y±V >= 0. If this derivation cannot be completed, then the alpha in [1/2,1] barrier claim needs a separate proof rather than the citation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the proof of Theorem 3.1, the gluing argument in Lemma 4.4, the perspective constructions in Section 4.2, and the compatibility-to-barrier composition via Lemma 2.4, and I did not find an internal inconsistency or a gap that threatens the central claim. The least independently verified step is Lemma 4.2: the hypograph barrier for alpha in [1/2,1] depends on a Pick-function analytic continuation of Fhat that the paper cites as an intermediate result in the proof of Hiai's Theorem 2.1 rather than reproducing. This is a standard external dependency, and I see no evidence of a mismatch; however, it is load-bearing in the precise sense that if Hiai's theorem did not cover the semidefinite endpoint constraints X±H >= 0 and Y±V >= 0, or the full interval alpha in [1/2,1], then Theorem 1.2(i) would not follow. The rest of the argument, including the alpha in [1,2] epigraph construction and the optimality lower bounds, appears sound.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a sufficient condition for the natural logarithmic barrier associated with the epigraph (or hypograph) of a function to be self-concordant. The main technical result, Theorem 3.1, shows that if every scalarized restriction of a function along lines is operator concave, then the function is (K,1)-compatible; combining this with Nesterov’s compatibility theorem yields self-concordant barriers. The authors apply this to the sandwiched Rényi quasi-relative entropy Psi_alpha, obtaining a (1+2n)-logarithmically homogeneous self-concordant barrier for the hypograph when alpha in [1/2,1] and for the epigraph when alpha in [1,2], together with optimality of the barrier parameter. They also obtain a barrier for the epigraph of the perspective D_alpha for alpha in [1/2,1), prove a general compatibility result for free mappings (Theorem 4.5), and provide derivative oracles and numerical experiments in the QICS solver.","tokens_in":21872,"tokens_out":65144,"duration_ms":594862,"significance":"The central result is significant: it gives the first self-concordant barriers for the sandwiched Rényi entropy cones, thereby enabling interior-point methods for optimization problems involving these quantum-information-theoretic functions. The line-operator-concavity criterion in Theorem 3.1 is simple, broadly applicable, and unifies previous compatibility results for operator convex functions and their perspectives. The paper is commendable for providing machine-verifiable derivative formulas, an open-source implementation, and numerical verification against known fixed-point and rate-distortion formulas. The main proof is generally clean; the principal external dependency is Lemma 4.2, quoted from Hiai, which I found no evidence to doubt, though it is a verification burden that should be made more explicit.","major_comments":[],"minor_comments":[{"comment":"Lemma 4.2 is load-bearing for the alpha in [1/2,1] hypograph barrier, but its proof is only a citation to an intermediate step in the proof of Hiai's Theorem 2.1. I did not find a mismatch with the stated ranges or endpoint conditions, but the dependence would be much easier to verify if the authors either reproduced the Pick-function argument or stated the precise theorem in Hiai's notation and explicitly checked the parameter mapping (p=(1-alpha)/(2alpha), q=1, s=alpha) and the semidefinite endpoint constraints X±H >= 0 and Y±V >= 0.","section":"Section 4.1, Lemma 4.2"},{"comment":"The proof of Proposition 6.1 states that (alpha-2)xhat - (alpha+1)yhat <= 2alpha-1, but the bound needed to derive beta=(2alpha-1)/3 is the lower bound (alpha-2)xhat - (alpha+1)yhat >= -(2alpha-1). The displayed inequality is true but does not by itself imply the required estimate on the third derivative; please correct this step.","section":"Appendix A.5"},{"comment":"Lemma 2.2 states that the measure mu is a unique positive finite Borel measure, while the proof of Theorem 3.1 refers to a unique Borel probability measure. These statements are inconsistent when g''(0)=0; please clarify the normalization and the uniqueness claim.","section":"Lemma 2.2 and Theorem 3.1"},{"comment":"The statement of Lemma 4.4 ends with 'then f is also H^n_+-convex (a,c)', which should read 'H^n_+-convex on (a,c)'.","section":"Lemma 4.4"},{"comment":"The proof says the result follows directly from Theorem 3.1 and Lemma 2.4. A reader may be confused because Lemma 2.4 contains a factor beta/3; the argument becomes explicit if one applies Lemma 2.4 to the 3-scaled domain barrier 3G, which is a valid (3nu)-self-concordant barrier. Please spell this out.","section":"Theorem 1.1 proof"}],"recommendation":"minor_revision","confidential_remarks":"This is a well-written and substantial paper. The only point that gave me pause is the external dependence of Theorem 1.2(i) on Hiai's analytic-continuation lemma; I saw no indication of an error, and the requested clarification is modest. The paper is a good fit for the journal and I support publication after minor revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Have you seen this? It actually settles something concrete: no self-concordant barrier was known for the sandwiched Rényi entropy, and here they build optimal-parameter barriers for the hypograph for α∈[1/2,1] and the epigraph for α∈[1,2]. The route is a new sufficient condition that is strictly weaker than the operator-convexity assumptions in earlier compatibility results. That alone makes it worth reading.\n\nThe main theorem, 3.1, is cleaner than I expected from the abstract. If every scalar restriction t↦⟨z,f(x+th)⟩ is operator concave on the relevant line segment, then f is (K,1)-compatible. The proof uses the integral representation of operator concave functions and a simple bound on ξ''' in terms of ξ''; I rechecked the constant and it's right. Combining with Nesterov's barrier composition gives the self-concordance claim. The application sections are more involved but the logic holds together. The α∈[1,2] case uses a clever Kronecker-product representation of the trace function and a positive linear map to reduce to concavity of a perspective; I did not find a gap. The α∈[1/2,1] case leans on an analytic-continuation result of Hiai (Lemma 4.2) that is quoted as an intermediate step in the proof of Hiai's Theorem 2.1. That is the softest point. It is load-bearing for the hypograph barrier, and the paper does not reproduce the proof. I could not see a mismatch with the stated matrix ranges or the α interval, but a referee should insist on a precise statement or a short proof in an appendix. This is not a fatal flaw, but it is the one place where I would want more support.\n\nThe numerical experiments are a modest but real plus: they solve the Rényi mutual information and a rate-distortion problem, and the residuals match the fixed-point characterization (for one case) and the closed-form limit (for the other). The code is available through QICS, though not pinned to a specific commit; that is a minor reproducibility nit.\n\nThe open questions at the end are honest. The conjecture for α∈(2,∞) with compatibility parameter (2α−1)/3 is clearly stated and the scalar case is proven tight. That is a useful pointer for follow-up work.\n\nWho should read this: anyone working on self-concordant barriers, conic optimization with quantum information constraints, or nonsymmetric cones. It deserves a serious referee. I would send it out and expect that a careful referee will ask for a proof or precise statement of Lemma 4.2 and a pinned software version, but nothing more structural.","headline":"Optimal self-concordant barriers for sandwiched Rényi entropy cones are constructed via a clean generalization of compatibility; the main caveat is a load-bearing cited lemma from Hiai that a referee should ask to be stated precisely.","tokens_in":22441,"tokens_out":3329,"would_cite":true,"duration_ms":32807,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90C25","90C51","47A63","81P45"],"pacs":[],"model":"deepseek-v4-flash","headline":"A one-line convexity test yields optimal self-concordant barriers for sandwiched Rényi entropies.","keywords":["self-concordant barrier","operator convexity","operator concavity along lines","sandwiched Rényi entropy","quantum relative entropy","interior-point methods","logarithmically homogeneous barrier","compatibility"],"falsifier":"Take $n=2$, $\\alpha=3/4$, $X=Y=I$, $H=\\operatorname{diag}(1/2,-1/2)$, $V=0$, and evaluate $\\hat F(z)=\\operatorname{tr}[((zI)^{(1-\\alpha)/(2\\alpha)}(zI+H)(zI)^{(1-\\alpha)/(2\\alpha)})^\\alpha]$ for $z$ with positive imaginary part; if the imaginary part of $\\hat F(z)$ is ever nonpositive, Lemma 4.2 is false and the paper's hypograph barrier for $\\alpha\\in[1/2,1]$ no longer follows. A numerical scan over many boundary-feasible $H,V$ and $\\alpha$ values could settle whether the quoted continuation holds on the exact stated domain.","tokens_in":21472,"feed_emoji":"🧮","tokens_out":10431,"duration_ms":93887,"temperature":0.7,"pith_summary":"The paper establishes a short bridge between operator convexity and convex optimization: if a sufficiently smooth cone-valued concave function is operator concave along every one-dimensional slice of its domain, then the natural logarithmic barrier of its hypograph is self-concordant. The proof runs through Nesterov and Nemirovskii's compatibility concept, and the one-dimensional condition supplies the required third-derivative bound with parameter one. The main application is the sandwiched Rényi quasi-relative entropy from quantum information: for $\\alpha\\in[1/2,1]$ the logarithmically homogeneous barrier for the hypograph is self-concordant with the optimal parameter $1+2n$, and for $\\alpha\\in[1,2]$ the same holds for the epigraph. These are the first self-concordant barriers known for this family, making the corresponding entropy optimization problems accessible to interior-point methods.","feed_headline":"One-line convexity test yields Rényi entropy barriers","feed_subtitle":"The test yields optimal barriers that open Rényi entropy problems to interior-point solvers.","key_machinery":"The load-bearing object is the scalar line restriction $F(t)=\\langle z,f(x+th)\\rangle$. Operator concavity of $F$ on $(-1,1)$ gives the integral representation $F(t)=F(0)+F'(0)t+\\tfrac12 F''(0)\\int_{-1}^{1}\\frac{t^2}{1-st}\\,d\\mu(s)$ for a probability measure $\\mu$. The kernel $\\xi_s(t)=t^2/(1-st)$ satisfies $\\xi_s'''(0)=6s\\ge -6=-3\\xi_s''(0)$ for every $s\\in[-1,1]$; combined with $F''(0)\\le 0$, this yields $\\langle z,D^3f(x)[h,h,h]\\rangle\\le -3\\langle z,D^2f(x)[h,h]\\rangle$ for every $z$, which is $(K,1)$-compatibility. For the Rényi application, the proof that $F$ is operator concave for $\\alpha\\in[1/2,1]$ goes through the transposed function $\\hat F(t)=tF(1/t)$ and a quoted analytic-continuation result identifying it as a Pick function; the $\\alpha\\in[1,2]$ case is assembled from noncommutative perspectives of operator concave functions using direct sums and unitary covariance.","core_discovery":"The central claim is Theorem 3.1: let $f$ be a $C^3$ function from an open convex domain into a finite-dimensional space ordered by a proper cone $K$. If, for every dual functional $z\\in K^*$ and every line $x+th$ staying within the closure of the domain, the scalar function $t\\mapsto\\langle z,f(x+th)\\rangle$ is operator concave on $(-1,1)$, then $f$ is $(K,1)$-compatible with its closed domain. Compatibility is exactly the third-derivative inequality $D^3f(x)[h,h,h]\\preceq_K -3D^2f(x)[h,h]$ needed for Nesterov and Nemirovskii's composition lemma to turn a self-concordant barrier of the domain into a self-concordant barrier of the hypograph. Applying this to the sandwiched Rényi quasi-relative entropy $\\Psi_\\alpha$ yields logarithmic barriers for the hypograph when $\\alpha\\in[1/2,1]$ and for the epigraph when $\\alpha\\in[1,2]$, each with the optimal barrier parameter $1+2n$; the perspective of the sandwiched Rényi entropy also obtains a $(2+2n)$-barrier for its epigraph when $\\alpha\\in[1/2,1)$.","pith_inferences":["Extending the paper's logic, any future function whose one-dimensional slices can be shown operator concave through Loewner-type arguments would inherit compatibility without a separate third-derivative calculation.","The same strategy could plausibly cover the general trace family $\\Psi_{p,q,s}$ in the parameter ranges listed in the conclusion, once an analogue of the quoted continuation lemma is available for those parameters.","If the paper's conjecture for $\\alpha\\in(2,\\infty)$ is true, epigraph barriers with parameter growing like $O(\\alpha^3)$ would follow from the same composition lemma, giving an interior-point route for a range that is currently open."],"forward_implications":["For every $\\alpha\\in[1/2,1]$, the natural logarithmic barrier for $\\operatorname{cl}\\operatorname{hypo}\\Psi_\\alpha$ is a $(1+2n)$-logarithmically homogeneous self-concordant barrier with optimal parameter, so maximizing $\\Psi_\\alpha$ over convex sets can be handled by interior-point methods.","For every $\\alpha\\in[1,2]$, the analogous barrier for $\\operatorname{cl}\\operatorname{epi}\\Psi_\\alpha$ is also $(1+2n)$-logarithmically homogeneous and optimal, covering the convex minimization range of $\\Psi_\\alpha$.","For $\\alpha\\in[1/2,1)$, the perspective of the sandwiched Rényi entropy has a $(2+2n)$-self-concordant barrier for its epigraph, again optimal.","The compatibility results for noncommutative perspectives of operator concave functions previously established in the quantum relative entropy literature follow as corollaries of the one-dimensional criterion.","The barrier functions are implemented in the open-source interior-point solver QICS and are exposed through PICOS; numerical experiments on Rényi mutual information and quantum rate-distortion problems reach residuals between $10^{-8}$ and $10^{-10}$."],"supporting_citations":[{"why":"Supplies the composition lemma that turns compatibility of a concave function into a self-concordant barrier for its hypograph, together with the definitions of self-concordance and compatibility.","marker":"[31]"},{"why":"Provides the integral representation of operator convex functions on $(-1,1)$ used to convert linewise operator concavity into the third-derivative compatibility inequality.","marker":"[29]"},{"why":"Is the source of the analytic continuation to the upper half-plane that makes the transposed Rényi trace function operator monotone for $\\alpha\\in[1/2,1]$.","marker":"[12]"},{"why":"Loewner's theorem, quoted as Lemma 2.1, connects the analytic continuation to operator monotonicity of the transposed function.","marker":"[30]"},{"why":"Earlier work establishing compatibility and optimal barrier parameters for quantum relative entropies; several of its results are recovered as corollaries here.","marker":"[4]"},{"why":"Introduces the sandwiched Rényi entropy and the perspective/composition viewpoint used for the $\\alpha\\in[1,2]$ convexity proof.","marker":"[5]"},{"why":"Supplies joint concavity of noncommutative perspectives, used in Corollary 4.8 to assemble compatibility for $\\alpha\\in[1,2]$.","marker":"[34]"}],"fun_headline_variants":["Operator convexity along lines yields Rényi entropy barriers","Rényi entropy barriers from a one-line convexity test","Line-based operator convexity gives self-concordant Rényi barriers","Convexity test unlocks Rényi entropy optimization","Self-concordant barriers for Rényi entropies via line convexity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole $\\alpha\\in[1/2,1]$ half of the main theorem rests on a quoted analytic-continuation result, not proved in the paper: a certain trace function formed from the two matrices and their perturbation directions must extend to the upper half-plane in a positivity-preserving way, and if that extension fails on the exact boundary of the stated domain, the hypograph barrier for this range collapses.","fun_headline_variants_meta":{"raw":{"variants":["Operator convexity along lines yields Rényi entropy barriers","Rényi entropy barriers from a one-line convexity test","Line-based operator convexity gives self-concordant Rényi barriers","Convexity test unlocks Rényi entropy optimization","Self-concordant barriers for Rényi entropies via line convexity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000613,"raw_usage":{"total_tokens":2901,"prompt_tokens":1044,"completion_tokens":1857,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":660,"completion_tokens_details":{"reasoning_tokens":1778}},"tokens_in":660,"tokens_out":1857,"duration_ms":14592,"temperature":1.0,"reasoning_tokens":1778,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T18:37:32.634022+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $n=2$, $\\alpha=3/4$, $X=Y=I$, $H=\\operatorname{diag}(1/2,-1/2)$, $V=0$, and evaluate $\\hat F(z)=\\operatorname{tr}[((zI)^{(1-\\alpha)/(2\\alpha)}(zI+H)(zI)^{(1-\\alpha)/(2\\alpha)})^\\alpha]$ for $z$ with positive imaginary part; if the imaginary part of $\\hat F(z)$ is ever nonpositive, Lemma 4.2 is false and the paper's hypograph barrier for $\\alpha\\in[1/2,1]$ no longer follows. A numerical scan over many boundary-feasible $H,V$ and $\\alpha$ values could settle whether the quoted continuation holds on the exact stated domain.","supporting_citations":[{"cited_title":"Nesterov, Lectures on Convex Optimization","cited_arxiv_id":null,"evidence_quote":"Supplies the composition lemma that turns compatibility of a concave function into a self-concordant barrier for its hypograph, together with the definitions of self-concordance and compatibility."},{"cited_title":"Matrix analysis: matrix monotone functions, matrix mea ns, and majorization,","cited_arxiv_id":null,"evidence_quote":"Provides the integral representation of operator convex functions on $(-1,1)$ used to convert linewise operator concavity into the third-derivative compatibility inequality."},{"cited_title":"Concavity of certain matrix trace and norm functions,","cited_arxiv_id":null,"evidence_quote":"Is the source of the analytic continuation to the upper half-plane that makes the transposed Rényi trace function operator monotone for $\\alpha\\in[1/2,1]$."},{"cited_title":"Simon, Loewner’s theorem on monotone matrix functions","cited_arxiv_id":null,"evidence_quote":"Loewner's theorem, quoted as Lemma 2.1, connects the analytic continuation to operator monotonicity of the transposed function."},{"cited_title":"Optimal self-concordant barrie rs for quantum relative en- tropies,","cited_arxiv_id":null,"evidence_quote":"Earlier work establishing compatibility and optimal barrier parameters for quantum relative entropies; several of its results are recovered as corollaries here."},{"cited_title":"On quantum R´ enyi entropies: A new generalization and some properties,","cited_arxiv_id":null,"evidence_quote":"Introduces the sandwiched Rényi entropy and the perspective/composition viewpoint used for the $\\alpha\\in[1,2]$ convexity proof."},{"cited_title":"Perspective s of matrix convex functions,","cited_arxiv_id":null,"evidence_quote":"Supplies joint concavity of noncommutative perspectives, used in Corollary 4.8 to assemble compatibility for $\\alpha\\in[1,2]$."}],"review_version":1}