{"id":"012a863a-ebce-48fb-9e8d-f5ea907fe0ab","arxiv_id":"2510.25740","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The excess growth rate is the unique functional, up to a constant, satisfying each of three axiom systems; its deterministic maximizer invests only in the best- and worst-performing assets.","lead":"This paper proves that the 'excess growth rate' — the bonus return a portfolio earns when its holdings grow at different rates — is the unique quantity fixed by three natural sets of axioms. It also derives explicit solutions for the portfolio that maximizes this quantity and relates the function to entropy, free energy, and large deviations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.20's converse is the load-bearing step: it invokes [49, Prop 6] and [50, Thm 4.5/4.7] without re-derivation, and if those require hypotheses beyond (3.21), the uniqueness proof collapses.","rationale":"The reader's weakest assumption exactly identifies the same load-bearing concern: Theorem 3.20's converse relies on the authors' earlier theorems rather than a self-contained proof, and the regularity condition (3.21) is the only stated hypothesis ensuring them. I reviewed the proof structure and confirmed that the main derivations in Theorems 3.2 and 3.13 are logically sound, and the optimization results (Lemma 4.2, Theorem 4.3, Theorem 4.11) appear correct. The omitted LDP proof in Theorem 2.15(ii) is a secondary issue. Thus the appropriate verdict remains CONDITIONAL, matching the reader's assessment. The concern is about proof completeness, not a detected error, so no stronger verdict is warranted.","tokens_in":46122,"tokens_out":10117,"duration_ms":90524,"concrete_test":"Re-derive identity (3.25) from Definition 3.17 and the translation invariance of L~ under exponential coordinates, using only the hypotheses in (3.21) (C^4 and strict concavity of Φ in every tangent direction), without citing [50, Theorem 4.7]. If the derivation requires an extra condition—such as uniform strict concavity or a C^4 extension of φ to the closed simplex—then Theorem 3.20's converse is incomplete, and the manuscript should either add that condition or prove it from (3.21).","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is the triple characterization. Theorems 3.2 and 3.13 are self-contained (modulo standard functional-equation results), but Theorem 3.20's converse is not. The paper itself notes this result was previously asserted without proof in [50, Example 3.10]. In the converse, after defining the portfolio map π(p) in (3.24) (which contains a typo: 'x_i' should be 'p_i'), it cites [49, Prop 6] to assert π(p)∈Δ_n° and [50, Thm 4.5/4.7] to obtain positive definiteness of the information metric and the Christoffel identity (3.25). The only stated regularity is (3.21): C^4 and strict concavity of Φ=e^φ in every tangent direction. If those cited theorems require a stronger notion of 'regular exponentially concave'—e.g., a continuous extension to the closed simplex, or a uniform non-degeneracy of the Hessian near the boundary—then the proof does not establish the converse for every function satisfying (3.21). In that case the if-and-only-if reduces to the easy direction. This is a proof-completeness gap, not a detected counterexample, and it does not affect Theorems 3.2/3.13 or the optimization results. The omitted LDP details in Theorem 2.15(ii) are secondary and do not bear on the characterization theorems.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper provides a mathematical study of the excess growth rate Γ(π,R) = log⟨π,R⟩ − Σπ_i log R_i, a quantity used in stochastic portfolio theory. It first collects properties of Γ, including permutation invariance, support dependence, numéraire invariance, a general chain rule, and connections to the Helmholtz free energy, Campbell's coding measure, a large-deviation principle for scaled Dirichlet distributions, and Rényi divergence. The main results are three axiomatic characterizations: Theorem 3.2 shows that Γ is, up to a multiplicative constant, the unique family satisfying measurability, permutation invariance, support dependence, vanishing on constants, and the general chain rule; Theorem 3.13 characterizes Γ among Jensen-gap functions as the unique one satisfying homogeneity (numéraire invariance) plus a constant-mean affinity condition; Theorem 3.20 characterizes Γ as the unique perturbation-invariant logarithmic divergence. The paper also solves the deterministic maximization of γ(π,r) explicitly (Theorem 4.3), gives a variational/perspective interpretation, and derives a first-order condition for maximizing the expected excess growth rate (Theorem 4.11).","tokens_in":46315,"tokens_out":8816,"duration_ms":81928,"significance":"If the results are correct, this is a substantial contribution: it places a finance-motivated functional on the same axiomatic footing as entropy and relative entropy, proves three complementary characterizations, and provides a pleasing link between portfolio theory and information geometry. Theorems 3.2 and 3.13 are proven in the paper with only standard functional-equation ingredients; the explicit two-support solution of the deterministic maximization problem is useful and clearly stated. Theorem 3.20 is the most ambitious, as it identifies the excess growth rate within the family of logarithmic divergences; however, as discussed below, its converse depends on external results whose hypotheses are not checked against the paper's own regularity condition. The connections to Campbell's measure, Dirichlet large deviations, and Rényi divergence are interesting and broaden the significance beyond finance. Overall the paper is well within the scope of cs.IT and would be a solid contribution once the proof gap in Theorem 3.20 is addressed.","major_comments":[{"comment":"The converse of Theorem 3.20 is the load-bearing step of the third characterization, but its proof invokes [49, Proposition 6] and [50, Theorems 4.5 and 4.7] to assert that the portfolio map π(p) in (3.24) lies in the open simplex, that the metric (g_ij) is strictly positive definite, and that the Christoffel identity (3.25) holds. The regularity condition stated in the paper is (3.21): C^4 and strict concavity of Φ=e^φ in every tangent direction. It is not demonstrated that these cited theorems apply under exactly this hypothesis; if 'regular exponentially concave' in [50] includes additional boundary or nondegeneracy conditions, then the proof does not establish the converse for every function satisfying (3.21), and the if-and-only-if may reduce to the easy direction. Since this is the central claim of Section 3.3, the authors should either prove the needed identities from (3.21) alone","section":"Section 3.3, Theorem 3.20, Eq. (3.24)-(3.25)"},{"comment":"There is a domain inconsistency in the statement and proof of the Jensen-gap characterization. The theorem states g:A_n→R with A_n = Δ_n × Δ_n, but Assumption (D3) quantifies over (π,R)∈D_n, and the proofs of Lemma 3.11 and Theorem 3.13 use R=(u,v,1,...,1) with arbitrary u,v>0, which need not lie in the simplex. For example, in Step 1 of the proof of Theorem 3.13(ii), π=(1−t,t,0,...,0) and R=(u,v,1,...,1) are used, but (π,R)∉A_n unless u=v=...=1. The argument implicitly extends g to D_n via scaling. This extension should be stated explicitly (or g should be defined on D_n throughout), otherwise the derivation of φ=clog is not formally justified as written. This is fixable but needs to be corrected for the proof to be complete.","section":"Section 3.2, Assumption 3.12 and proof of Theorem 3.13"}],"minor_comments":[{"comment":"In the definition of the portfolio map, 'π_i(p) := x_i (1 + ...)' should read 'p_i' instead of 'x_i'; as written, the symbol x_i is undefined. This appears in the load-bearing part of Theorem 3.20 and should be corrected.","section":"Section 3.3, Eq. (3.24)"},{"comment":"The bullet list in the introduction says 'Our first characterization (Theorem 3.20), proved in Section 3.1' but Theorem 3.20 is in Section 3.3 and is the third characterization (via logarithmic divergence). The first characterization is Theorem 3.2 in Section 3.1. The numbering/labeling should be fixed.","section":"Introduction, Section 3.1/3.3"},{"comment":"Just before the exponential-coordinate change, the text writes 'θ=(θ_1,...,θ_n)∈R^{n−1}' and defines θ_i for i=1,...,n−1. The vector should be (θ_1,...,θ_{n−1}), not (θ_1,...,θ_n). This is a typo but it is confusing in a key proof.","section":"Section 3.3, proof of Theorem 3.20"},{"comment":"The proof of the large deviation principle is omitted with 'we omit the details.' While this result is not used later, the uniform convergence in (i) alone does not automatically yield the LDP for all open and closed sets unless goodness of the rate function and exponential tightness are also verified. A brief justification or a precise reference for the implication would make the theorem self-contained.","section":"Theorem 2.15(ii)"},{"comment":"In several places, the notation uses supp(p) where p is not defined (e.g., Eq. (1.8) and the surrounding discussion); these should be supp(π). Also, in the proof of Lemma 2.13, Γπ(y∥x) and Γπ(y|x) are used interchangeably; unify the notation.","section":"Eq. (1.8), Definition 1.1 area"}],"recommendation":"major_revision","confidential_remarks":"The paper is a strong candidate for publication in cs.IT if the authors can close the gap in Theorem 3.20. The concern is not a detected counterexample but a genuine hypothesis mismatch: the paper's 'regular' condition (3.21) is weaker than what may be required by the cited theorems from [49,50]. The other two characterizations and the optimization results appear sound and are valuable. I would not reject, but I would ask the authors to either prove the cited identities under (3.21) or state and verify a stronger regularity condition. The domain issue in Section 3.2 is also easy to fix and should be addressed. The paper's self-contained proofs of Theorems 3.2 and 3.13 and its explicit solutions are notable strengths."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this paper's core is the three axiomatic characterizations of the excess growth rate, and they are new and mostly hold up. The Jensen-gap characterization (Theorem 3.13) is the cleanest — the functional equation genuinely forces φ = c·log plus affine, as the reader says. The first characterization (Theorem 3.2) is also self-contained: the relative-entropy link in Lemma 3.4 and the boundary arguments in Lemmas 3.9–3.10 check out. The optimization results are solid too; the two-point support claim for the deterministic maximizer is a nice, correct surprise.\n\nThe real soft spot is Theorem 3.20. The converse leans on the authors' earlier results — [49, Prop 6] for the portfolio map landing in the simplex and [50, Thm 4.5/4.7] for positive definiteness and the Christoffel identity. Those are not re-derived, and the stated regularity (3.21) may or may not imply what the cited theorems need. If they require extra boundary control, the converse collapses to the easy direction. This is a proof-completeness gap, not a detected counterexample, but it should be fixed before publication: either state the stronger hypotheses up front or prove the needed identities in the appendix.\n\nMinor issues: the LDP in Theorem 2.15(ii) is asserted without proof, though it's secondary and doesn't touch the characterizations. And there's a typo in (3.24) — 'x_i' should be 'p_i'.\n\nThe authors are honest about provenance. They explicitly say Theorem 3.20 was claimed without proof in [50, Example 3.10] and that some 'connections' are transparent rewrites. That honesty counts for something.\n\nWho gets value from this: people in stochastic portfolio theory, information geometry, and axiomatic characterization of information measures. It's a serious contribution, not a desk-reject. Send it to a referee who knows the Pal–Wong line of work and ask them to scrutinize Theorem 3.20's assumptions. My take: engage with it.","headline":"Three genuinely new characterization theorems, the first two self-contained and sound; the third leans on the authors' prior results and needs a patch, but the paper deserves refereeing.","tokens_in":46956,"tokens_out":1621,"would_cite":true,"duration_ms":18120,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A17","91G10","60F10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Three axiomatic characterizations force the excess growth rate — the Jensen gap of the logarithm — to be the unique measure of diversification return up to a multiplicative constant.","keywords":["excess growth rate","axiomatic characterization","relative entropy","Jensen gap","logarithmic divergence","portfolio theory","large deviations","Rényi divergence"],"falsifier":"Produce a regular exponentially concave φ (C⁴ with Φ=e^φ strictly concave in every tangent direction) whose logarithmic divergence satisfies Lφ(q⊕h∥p⊕h)=Lφ(q∥p) for all p,q,h in the open simplex but whose portfolio map π(p) is nonconstant; the paper's Theorem 3.20 predicts no such φ exists. A direct calculation of the portfolio map and Christoffel symbols for any candidate φ would settle the claim.","tokens_in":45842,"feed_emoji":"📈","tokens_out":4992,"duration_ms":47004,"temperature":0.7,"pith_summary":"The paper aims to show that the excess growth rate — the Jensen gap log(ΣπᵢRᵢ) − Σπᵢ log Rᵢ used in portfolio theory as a measure of diversification return — is not one ad hoc functional among many. It gives three axiomatic characterizations: any measurable, permutation-invariant, support-dependent, constant-vanishing functional satisfying a general chain rule must be a constant multiple of Γ; any measurable generator satisfying the gap axioms plus numeraire invariance must be c·log; and any regular logarithmic divergence invariant under perturbations must come from negative cross-entropy and hence equal Γ. The paper also proves that the deterministic maximizer of Γ puts positive weight only on the largest and smallest log-returns, with an explicit closed form, and derives a first-order condition for the expected version. These results pull together portfolio theory, information theory, geometry, and large deviations around a single logarithmically defined object.","feed_headline":"Axioms pin the excess growth rate to log, up to scale","feed_subtitle":"Portfolio theory's diversification measure is the only functional satisfying measurability, the chain rule, and numeraire invariance.","key_machinery":"The central object is the family Γn(π,R)=log Σ_{i∈supp π} πᵢRᵢ − Σ πᵢ log Rᵢ, a Jensen gap for the logarithm and the difference between exponential and arithmetic means. Three mechanisms carry the argument: (i) the algebraic identity Γ(π,r)=H(π∥π⊕π r) that rewrites Γ as a relative entropy under the simplex's multiplicative perturbation structure; (ii) the general chain rule Γ(π∘p,a∘R)=Γ(π,a⟨⟨p,R⟩⟩)+ΣπᵢΓ(pᵢ,Rᵢ), which lets the proof reduce to a known characterization of relative entropy; (iii) for the divergence characterization, the portfolio map π(p)=pᵢ(1+∂_{eᵢ−p}φ(p)) and the information-geometric identity Γᵏᵢⱼ(θ)=δᵢⱼₖ−δᵢₖπⱼ−δⱼₖπᵢ, which turns perturbation invariance into constancy of the","core_discovery":"On its own terms, the paper establishes that the family Γn is the unique (up to a common multiplicative constant) solution to three separate sets of natural axioms. First, from Lebesgue measurability, permutation invariance, dependence only on the support, vanishing on constant returns, and a general chain rule for composite portfolios, the only functionals are cΓ. Second, among gap functions φ(⟨π,R⟩)−Σπᵢφ(Rᵢ) that vanish on constant returns and are affine on constant-mean slices, numeraire invariance forces φ = c log plus an affine term, hence the gap is cΓ. Third, within the class of regular exponentially concave functions on the open simplex, the logarithmic divergence is perturbation inv","pith_inferences":["Editorial inference: Because all three uniqueness theorems only pin Γ up to the same multiplicative constant, the constant carries no information; in practice it can be absorbed into the length of the rebalancing period. This suggests that any empirical calibration of an 'excess growth' parameter is testing the time scale, not the functional form.","Editorial inference: The two-point support result points toward a testable portfolio rule: when one asset has the best expected log return but also the highest volatility, the EGR-maximizing portfolio deliberately holds the worst asset as a risk offset. One could backtest whether such max-min portfolios empirically harvest rebalancing premia better than equal-weighting.","Editorial inference: The perturbation-invariance characterization could be turned into a model diagnostic: given a candidate divergence on returns, check numerically whether it depends on initial prices or only on returns; any non-logarithmic divergence that passes the test would refute the paper's uniqueness, while a log divergence that fails would suggest the regularity conditions are needed.","Editorial inference: The large-deviation connection suggests using the excess growth rate as a divergence for compositional data beyond finance; a straightforward extension is to test the scaled Dirichlet rate function empirically on simplex-valued data such as market shares or species abundances."],"forward_implications":["Any quantity that satisfies the five properties in Assumption 3.1 must be proportional to Γ; so the rebalancing premium of a constant-rebalanced portfolio is, up to scale, the only possible such measure.","The deterministic maximizer of Γ is supported on the two assets with the largest and smallest log returns, and the max value has the closed form in Theorem 4.3; this gives an explicit 'volatility harvesting' portfolio.","For the expected version, a portfolio maximizes E[γ(π,r)] iff it satisfies (4.14); in the special case where all expected log returns are equal, the EGR-maximizing portfolio coincides with the growth optimal portfolio.","Γ emerges as the rate function for a large-deviation principle for scaled Dirichlet distributions and equals a Rényi divergence between members of that family, so information-theoretic tools such as Sanov-type bounds and Rényi divergences can be applied to portfolio volatility questions.","The gap axioms plus numeraire invariance imply that among all gap generators, only the logarithm yields a gap that is invariant under rescaling returns; hence the functional form is forced by economics rather than by convention."],"fun_headline_variants":["Log is the only 'excess growth' that fits three axiom sets","Excess growth rate pinned: it's log, up to scale","Three axiom sets force excess growth to be log times a constant","Uniqueness of excess growth: only log survives the axioms","Why excess growth must be log: an axiomatic proof"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that the C⁴, strictly-concave-Φ regularity assumption (3.21) is enough to make two cited prior results hold — the information-geometry identity for the Christoffel symbols and the fact that a constant portfolio map forces φ = −H×(π∥·)+c; if those cited results need extra boundary hypotheses beyond (3.21), the 'if and only if' in Theorem 3.20 collapses to the easy direction.","fun_headline_variants_meta":{"raw":{"variants":["Log is the only 'excess growth' that fits three axiom sets","Excess growth rate pinned: it's log, up to scale","Three axiom sets force excess growth to be log times a constant","Uniqueness of excess growth: only log survives the axioms","Why excess growth must be log: an axiomatic proof"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000165,"raw_usage":{"total_tokens":1059,"prompt_tokens":690,"completion_tokens":369,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":434,"completion_tokens_details":{"reasoning_tokens":295}},"tokens_in":434,"tokens_out":369,"duration_ms":4092,"temperature":1.0,"reasoning_tokens":295,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T07:26:19.874272+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Produce a regular exponentially concave φ (C⁴ with Φ=e^φ strictly concave in every tangent direction) whose logarithmic divergence satisfies Lφ(q⊕h∥p⊕h)=Lφ(q∥p) for all p,q,h in the open simplex but whose portfolio map π(p) is nonconstant; the paper's Theorem 3.20 predicts no such φ exists. A direct calculation of the portfolio map and Christoffel symbols for any candidate φ would settle the claim.","supporting_citations":[],"review_version":1}