{"id":"ed0541a3-fe17-4b0c-a9f7-2edd2002c32e","arxiv_id":"2607.17907","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"The paper claims complexity-maximizing distributions are always a single dominant probability value plus a uniform halo, but the continuous version is unbounded and the proof has a boundary gap.","lead":"This paper claims that probability distributions maximizing a generalized entropy-based 'complexity' always have a two-level structure: one dominant core and a large background halo, in both discrete and continuous settings. The result would unify much of the statistical-complexity literature, but the continuous claim is false and the discrete proof has a gap.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Continuous case is unbounded above: a shrinking core with fixed nonzero mass makes ln C_q diverge, so Theorem 2's global maximizer does not exist.","rationale":"The reader's verdict is REJECT with high confidence, and my stress-test supports rejection. The reader's stated weakest assumption is the Section 6.1 boundary claim used to exclude the zero-root case in the discrete proof; that is a genuine gap: the simplex boundary contains nonuniform distributions with zero components and positive Φ, such as (0.6,0.4,0) for N=3,q=2. However, the single most load-bearing issue is the continuous case, which is falsified outright. The two-level density family above is fully admissible and makes ln C_q diverge for every q>1. The paper's own concentrated-density calculation in Section 5.4 uses the δ-limit with core mass tending to 1; keeping the core mass fixed at α<1 while shrinking its measure changes the asymptotic and produces unboundedness. This does not depend on any questionable exclusion principle or boundary analysis, so it is stronger than the discrete proof gap. The reader also mentions unboundedness in the rationale, though not as the weakest assumption, so agreement is partial. The correct verdict remains REJECT; since the reader already reached that verdict, I mark the outcome UNCHANGED.","tokens_in":42500,"tokens_out":8768,"duration_ms":87578,"concrete_test":"Take q=2, Ω=[0,1], α=1/2, and evaluate ln C_q for the two-level density with a=10^{-1}, 10^{-2}, 10^{-3}, 10^{-4}. The asymptotic formula gives ln C_q ≈ 0.693 + (1/2) ln(1/a), i.e. approximately 1.844, 2.996, 4.147, 5.298, increasing without bound. If the computed values follow this growth, the continuous maximization problem has no finite maximum, directly falsifying Theorem 2.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive problem is the continuous half. On Ω=[0,1], fix α∈(0,1) and define the normalized two-level density ρ_a(x)=α/a on [0,a] and ρ_a(x)=(1−α)/(1−a) on (a,1]. For q>1, the optimized quantity is ln C_q = H1 + ln P_{2q} − 2 ln P_q, with P_r=∫ρ^r dx. As a→0, P_q ∼ α^q a^{1−q} and P_{2q} ∼ α^{2q} a^{1−2q}, while H1 = −α ln(α/a) − (1−α)ln((1−α)/(1−a)). Hence ln C_q = H_bin(α) + (α−1) ln a + O(1). Since α<1 and ln a→−∞, this diverges to +∞. Thus the complexity functional is unbounded above on the admitted class of densities, so no global maximizer exists and Theorem 2 is false as stated. Section 5.4 only checks the uniform density and the fully concentrated δ-limit, which corresponds to core mass α=1; the missing family with 0<α<1 makes the 'minimal core measure, M→0+' conclusion an unattained infimum rather than a maximum. This is a counterexample, not merely a proof gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces a generalized statistical complexity C_q = e^{H_1} P_{2q}/(P_q)^2, equivalently Φ = H_1 + (2q-2)H_q - (2q-1)H_{2q}, for q>1, and studies its maximization over finite discrete probability simplices and over continuous probability densities. The central claims are: (i) every stationary solution of the Euler–Lagrange equation has exactly two probability levels (a core and a halo); (ii) the global maximum is attained by a single core in the discrete case (m=1) and by an infinitesimal core in the continuous case (M→0^+); and (iii) this gives an exact unified discrete–continuous theory. The proofs proceed by analyzing the transcendental stationary equation f(x)=ln x + A_q x^{q-1} - B_q x^{2q-1} + C=0, excluding zero- and one-root cases, reducing to two-level distributions, and then applying an envelope-theorem monotonicity argument to a universal rational kernel.","tokens_in":42831,"tokens_out":5229,"duration_ms":46594,"significance":"If correct, the result would be a clean and sweeping structural statement: complexity maximizers of this family are always core–halo distributions, independent of the probability space. The manuscript has some genuine strengths: the variational derivation is self-contained, the discrete and continuous equations are shown to have the same algebraic form, no empirically fitted parameters enter, and the envelope-theorem reduction to a ratio formulation is explicitly carried out with symbolic verification noted for the algebra. These features make the paper easy to read and the intended argument transparent. However, the central mathematical claims are not supported. The continuous theorem is false as stated because the functional is unbounded above on the admitted class of densities, and the discrete proof relies on an incorrect boundary analysis. These are not local gaps but problems with the main theorems, despite the paper's expository virtues.","major_comments":[{"comment":"The exclusion of the zero-root case is invalid. §6.1 argues that if no interior stationary point exists, then the maximum lies on the boundary and, because §5.4 computed only the uniform and Dirac configurations, concludes Φ_max=0. But the boundary of the simplex contains many other points. For N=3, q=2, W=(0.6,0.4,0) gives Φ≈0.118>0, greater than both boundary values computed in §5.4. Thus the assertion 'one obtains Φ_max=0' is false, and the contradiction used to prove N_r≠0 collapses. The same type of boundary oversight affects the one-root exclusion in §6.2, because a boundary point can have some zero coordinates and only one positive root among the non-zero components. This gap is load-bearing for the two-root theorem and hence for Theorem 1.","section":"§5.4–§6.1"},{"comment":"The continuous theorem is false as stated. On Ω=[0,1], fix α∈(0,1) and define the normalized two-level density ρ_a(x)=α/a on [0,a], ρ_a(x)=(1-α)/(1-a) on (a,1]. For q>1, as a→0, P_q ∼ α^q a^{1-q}, P_{2q} ∼ α^{2q} a^{1-2q}, and H_1 = H_bin(α) + α ln a + O(1). Therefore ln C_q = H_bin(α) + (α-1) ln a + O(1). Since α<1, this tends to +∞. Hence the functional is unbounded above on the class of normalized densities, so no global maximizer exists. The paper's conclusion in §7.6 and Theorem 2 that the maximum is attained at 'minimal core measure, M→0^+' describes an unattained infimum, not a maximizer. This is a counterexample to the theorem, not merely a missing detail.","section":"§8, Theorem 2"},{"comment":"Even for the discrete problem, the proof that global maximizers have the two-level m=1 form is incomplete. Section 7 maximizes only over the interior stationary two-level family after the two-root theorem. If the boundary analysis in §6.1 fails, a global maximizer could sit on a boundary face and need not satisfy the interior stationary equation at all. The monotonicity argument for Φ^*(m) therefore only concerns a restricted family and does not by itself rule out boundary maxima. A complete proof would need to handle boundary faces, or show that a boundary maximizer can be perturbed into the interior with higher Φ. As written, Theorem 1 relies on an unproved and, with the current boundary treatment, false premise.","section":"§7–§8"}],"minor_comments":[{"comment":"The entropy representation is correct only if H_{2q} denotes the Rényi entropy of order 2q, not the Rényi entropy of order 2 evaluated on P_q. The paper should state this explicitly in the notation section, because a reader can easily mistake H_{2q} for a second-order entropy.","section":"§3.3 Eq. (3.3.4)"},{"comment":"The table of contents says Section 6 addresses the continuous case, but Section 6 is primarily about the discrete exclusion principle; the continuous analogue appears later in Appendix B.5. This mismatch should be corrected.","section":"Table of Contents / §6"},{"comment":"The 'concentrated density limit' discussion says the divergences 'cancel exactly' as ρ_ε→δ. The claim is not generally true: the limit depends on the shape of the approaching sequence. For the two-level family in the major comment above, the limit is +∞ when α<1. The text should either avoid this assertion or restrict it to a precise class of approximating sequences.","section":"§5.4"},{"comment":"Several equations contain stray commas in integrals, e.g. '∫ ρ(x) lnρ(x), dx' in §4.2. These are harmless but should be cleaned up.","section":"Throughout"}],"recommendation":"reject","confidential_remarks":"The paper has a clear and readable structure, and the algebraic reduction to a single rational kernel is attractive, but the two main theorems are not established as stated. The continuous theorem is directly contradicted by an explicit unbounded sequence of admissible densities, and the discrete proof contains a false boundary premise. These are load-bearing, not presentation issues. I see no route to acceptance without a substantial reformulation of the claims and a new proof strategy for the discrete boundary cases; the current manuscript should be rejected."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper has a clean variational setup and a nice algebraic reduction, but the continuous theorem is false and the discrete proof has a boundary gap. The stress-test counterexample lands.\n\nWhat is new: the q-generalized complexity Cq = e^{H1} P_{2q}/(P_q)^2, the common stationary equation, and the reduction of the two-level family to a one-parameter multiplicity problem. The envelope-theorem step is elegant; conditional on having exactly two interior roots, the R_k monotonicity argument is correct and gives dPhi/dm < 0, hence m=1 among interior two-level candidates.\n\nThe soft spots are load-bearing. First, Theorem 2 as stated is false. On [0,1], take rho_a with mass alpha on [0,a] and 1-alpha on (a,1], with 0<alpha<1. As a -> 0, ln C_q = H_bin(alpha) + (alpha-1) ln a + O(1) -> +infinity. So the functional is unbounded above; no global maximizer exists. Section 5.4 only computes the delta-limit with alpha=1; the missing mixed-core family destroys the claim. This is a counterexample, not a proof gap.\n\nSecond, the discrete proof's exclusion of zero roots in Section 6.1 is invalid. It assumes the only boundary configurations are uniform and Dirac. The simplex boundary includes faces; e.g., (0.6,0.4,0) for N=3, q=2 has Phi ~ 0.118, while uniform and Dirac both have Phi = 0. So the argument that no-root implies Phi_max = 0 fails. The two-root theorem can at best hold for interior stationary points, not for all global maximizers.\n\nThe paper is not a waste of time: the unified stationary equation and the multiplicity reduction are worth reading, and the discrete theorem might be salvageable with a proper boundary analysis. But as submitted, the stated theorems are not established. A serious referee would need to rewrite the boundary treatment and either reformulate the continuous problem (e.g., bounded densities or a supremum statement) or accept that the continuous functional has no maximizer.\n\nI would not cite this version in my own work, and I would not plan a reading group around it. But it deserves a real referee if the authors are willing to revise; the core question—whether one core maximizes Cq—is still plausible in the discrete setting, and the algebraic machinery is reusable.\n\nRecommendation: send to peer review, but expect major revision or rejection; the counterexample to Theorem 2 should be decisive unless the authors change the problem statement.","headline":"The continuous theorem is false (the functional is unbounded), the discrete proof misses boundary faces, but the variational machinery is clean and potentially salvageable.","tokens_in":43255,"tokens_out":4760,"would_cite":false,"duration_ms":49313,"reading_group":"no","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A17"],"pacs":[],"model":"deepseek-v4-flash","headline":"Maximizing generalized statistical complexity forces a two-level core–halo form on every probability distribution, discrete or continuous.","keywords":["statistical complexity","Rényi entropy","core–halo structure","variational principle","escort distribution","information energy","complexity maximization","discrete and continuous probability"],"falsifier":"Compute Φ for the boundary vector (0.6, 0.4, 0) with N = 3, q = 2. If Φ = H_1 − 2 ln P_2 + ln P_4 exceeds the value at the claimed one-core maximizer, then the theorem's assertion that every global maximizer has the two-level all-positive form is refuted for this case.","tokens_in":42389,"feed_emoji":"🎯","tokens_out":5673,"duration_ms":51688,"temperature":0.7,"pith_summary":"The paper defines a generalized statistical complexity C_q = e^{H_1} P_{2q}/(P_q)^2 built from Shannon and Rényi entropies and asks which probability distributions maximize it. Its central claim is that every global maximizer has exactly two probability levels: a single large core and a uniform halo. In finite discrete spaces the maximum is one dominant component over N−1 equal smaller ones; in continuous spaces it is a minimal-measure high-density core over a uniform sea. The same stationary equation governs both settings, so the core–halo structure is proposed as a universal law of complexity maximization.","feed_headline":"Every complexity maximizer is one core over a uniform halo","feed_subtitle":"Discrete and continuous probability spaces obey the same two-level law, and the smallest core always wins.","key_machinery":"The load-bearing object is the unified stationary equation f(x) = ln x + A_q x^{q−1} − B_q x^{2q−1} + C = 0, identical for discrete weights and continuous densities, with A_q and B_q determined by escort moments. Its unimodal shape (exactly one critical point) limits it to at most two positive roots; the paper's root-count and boundary arguments then force exactly two. The second key tool is the envelope-theorem reduction: after eliminating normalization, only the core multiplicity m (or measure M) is free, and the derivative dΦ*/dm collapses to the kernel R_1 + 2(q−1)R_q − (2q−1)R_{2q}, where R_k(τ) = (τ^k − 1)/(m τ^k + N − m) and τ = w_c/w_h > 1. Since R_k increases with k, this kernel is","core_discovery":"The paper constructs C_q from the Hill number e^{H_1} and the escort information energy P_{2q}/(P_q)^2, takes its logarithm to obtain Φ = H_1 + (2q−2)H_q − (2q−1)H_{2q}, and derives a common Euler–Lagrange equation ln x + A_q x^{q−1} − B_q x^{2q−1} + C = 0 for both discrete masses and continuous densities. Shape analysis shows this equation has at most two positive roots; the paper argues that the zero-root and one-root cases are impossible, so exactly two levels w_h < w_c (or ρ_h < ρ_c) are forced on every stationary solution. The remaining freedom is the multiplicity m of the core. Applying the envelope theorem, the optimized complexity decreases monotonically with m, so the global maximum","pith_inferences":["The same two-level logic likely extends to other complexity measures built from a concave entropy multiplied by a homogeneous convex concentration term; the paper only demonstrates it for the H_1–Rényi family.","The minimal-measure continuous core suggests that in finite numerical approximations the maximizer will appear as a single grid cell spike; the paper leaves this discretization effect unexplored.","The monotonic decrease in core multiplicity is consistent with a majorization-ordering characterization: distributions with a more spread core are always less complex, a lattice structure the paper does not develop.","If the variational framework is applied to Tsallis or other generalized entropies, a natural conjecture is that a two-level or two-scale maximizer persists; this extension is not tested here."],"forward_implications":["If correct, every discrete maximum-complexity state is exactly one dominant probability over N−1 equal smaller ones, never a richer hierarchy.","The continuous analogue says maximum-complexity densities are piecewise constant with two levels: a concentrated core of vanishing measure sitting on a uniform halo.","The two-root theorem unifies discrete and continuous optimization into a single variational problem, so structural results transfer between the settings.","The deformation parameter q acts as a heterogeneity control: larger q widens the core–halo gap while preserving the two-level form.","Because the ordinary LMC complexity appears as a particular member of the same family, its known core–halo maximizer becomes part of a continuous family with identical geometry."],"fun_headline_variants":["Complexity max always means a core over a uniform halo","Smallest core always wins the complexity maximum","Core-halo shape is universal for complexity maximizers","Discrete or continuous, complexity peaks at two levels","Complexity max: one dominant state plus uniform rest"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The argument excluding the zero-root case assumes that, without interior stationary points, the maximum of Φ must lie on a boundary point with Φ = 0 because the only boundary configurations considered are the uniform and Dirac distributions; boundary points with a zero component and unequal nonzero weights are not analyzed and can have positive Φ.","fun_headline_variants_meta":{"raw":{"variants":["Complexity max always means a core over a uniform halo","Smallest core always wins the complexity maximum","Core-halo shape is universal for complexity maximizers","Discrete or continuous, complexity peaks at two levels","Complexity max: one dominant state plus uniform rest"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000524,"raw_usage":{"total_tokens":2377,"prompt_tokens":763,"completion_tokens":1614,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":507,"completion_tokens_details":{"reasoning_tokens":1540}},"tokens_in":507,"tokens_out":1614,"duration_ms":10524,"temperature":1.0,"reasoning_tokens":1540,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T16:39:46.499587+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute Φ for the boundary vector (0.6, 0.4, 0) with N = 3, q = 2. If Φ = H_1 − 2 ln P_2 + ln P_4 exceeds the value at the claimed one-core maximizer, then the theorem's assertion that every global maximizer has the two-level all-positive form is refuted for this case.","supporting_citations":[],"review_version":1}