{"id":"92af95a0-d3e4-401b-88c0-707ccfd9a4fe","arxiv_id":"1908.01696","paper_version":3,"verdict":"REJECT","confidence":"HIGH","novelty_score":2.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"The proposed two-parameter entropy reduces, by the paper's own formulas, to the one-parameter Tsallis entropy S_q with q=2k+1, independent of r.","lead":"This paper claims to introduce a two-parameter generalized Tsallis entropy and divergence and to prove their fundamental information-theoretic properties. In its own definitions, the second parameter r cancels identically out of both objects, leaving the one-parameter Tsallis entropy and a standard f-divergence.","discovery_kind":"incremental","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The r parameter cancels identically in S_{k,r} and D_{k,r}; the claimed two-parameter family reduces to Tsallis entropy with q=2k+1.","rationale":"The reader's strongest claim is confirmed by direct algebra. The paper's load-bearing assertion is that S_{k,r} and D_{k,r} constitute a new two-parameter generalized Tsallis entropy and divergence. Since the r parameter cancels identically in both, the advertised two-parameter family does not exist; the surviving expressions are Tsallis-type entropies and divergences with q = 2k+1. This is an internal inconsistency in the paper's own definitions, not merely a disagreement with external consensus. The paper's claim that equation (19) recovers Definition 2 is also only valid for r=k, further confirming the algebraic cancellation. The Hessian computation in Section 5 is likewise incompatible with the r-independence of D_{k,r}; the correct diagonal metric is (1-2k)/p_i, which vanishes at k=1/2, a point within the stated parameter range. These issues undermine the central claim, so the REJECT verdict remains appropriate.","tokens_in":19351,"tokens_out":6679,"duration_ms":53781,"concrete_test":"Evaluate S_{k,r} as defined in Definition 2 for a fixed non-uniform probability vector, e.g., p=(0.2, 0.3, 0.5), at k=0.25 with r=0.5 and r=1.0. If the two numerical values are exactly equal, the r parameter drops out and the 'two-parameter' claim fails. Alternatively, verify the closed-form simplification p^{r+k+1} ln_{k,r}(p) = p(p^{2k}-1)/(2k) by direct substitution of Definition 1 into Definition 2; if the r cancels, the entropy and divergence are one-parameter objects.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Definition 2 defines S_{k,r}(X) = -Σ_x p(x)^{r+k+1} ln_{k,r}(p(x)), with ln_{k,r}(x) = (x^{2k}-1)/(2k x^{r+k}) from Definition 1. Direct substitution yields p^{r+k+1} ln_{k,r}(p) = p(p^{2k}-1)/(2k), so S_{k,r}(X) = (1 - Σ_x p(x)^{2k+1})/(2k), independent of r. Similarly, D_{k,r}(P||Q) = Σ_x p(x) (p(x)/q(x))^{r-k} ln_{k,r}(p(x)/q(x)) = Σ_x p(x)(1-(q(x)/p(x))^{2k})/(2k) = (1 - Σ_x p(x)^{1-2k} q(x)^{2k})/(2k), again independent of r. The paper's assertion that equation (19) with α=1-k+r, β=1+k+r recovers Definition 2 is only correct when r=k; for general r, (p^{1+r-k}-p^{1+r+k})/(2k) differs from p(1-p^{2k})/(2k) by a factor p^{r-k}. Thus the central object is not two-parameter. All subsequent properties (chain rule, sub-additivity, strong sub-additivity, joint convexity, information monotonicity) hold for the one-parameter family with q=2k+1, not for a two-parameter generalization. The Section 5 Hessian calculation is also internally inconsistent: since D_{k,r} is r-independent, the induced Fisher metric cannot depend on r; direct differentiation gives g_ii = (1-2k)/p_i, not (-2k+4r+1)/p_i, and it degenerates at k=1/2, a point inside the stated range.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a two-parameter deformed logarithm ln_{k,r}(x) and uses it to define a generalized Tsallis entropy S_{k,r}(X) and a generalized Tsallis relative entropy D_{k,r}(P||Q). It derives pseudo-additivity, chain rules, sub-additivity, strong sub-additivity, joint convexity, information monotonicity, and a Hessian information-geometric structure for these objects. The central claim is that S_{k,r} and D_{k,r} form a genuine two-parameter extension of Tsallis and Shannon entropy.","tokens_in":19728,"tokens_out":9536,"duration_ms":89975,"significance":"If the central claim were correct, the paper would provide a useful two-parameter entropy/divergence family with the standard information-theoretic toolbox. The manuscript is self-contained, with explicit definitions and mostly checkable algebraic proofs. However, direct substitution shows that the parameter r cancels identically from both S_{k,r} and D_{k,r}, so the proposed objects are actually one-parameter families. Consequently, the main contribution is not realized, and several derived statements are internally inconsistent with the paper's own definitions. The explicit, machine-checkable algebra is a strength, but it is used here to disprove the paper's central claim.","major_comments":[{"comment":"The parameter r cancels identically from the proposed entropy. Substituting ln_{k,r}(x)=(x^{2k}-1)/(2k x^{r+k}) into Definition 2 gives S_{k,r}(X)=Σ_x [p(x)-p(x)^{2k+1}]/(2k)=(1-Σ_x p(x)^{2k+1})/(2k), which is independent of r and is the Tsallis entropy with q=2k+1. The assertion that α=1-k+r and β=1+k+r in Eq. (19) recover Definition 2 is incorrect: Eq. (19) then gives Σ_x (p(x)^{1-k+r}-p(x)^{1+k+r})/(2k), which equals the Definition 2 expression only when r=k. The two-parameter claim is therefore false by construction.","section":"Section 3, Definition 2 and Eq. (19)"},{"comment":"The same cancellation occurs in the divergence. Substituting ln_{k,r} into D_{k,r}(P||Q)=Σ_x p(x)(p(x)/q(x))^{r-k} ln_{k,r}(p(x)/q(x)) gives D_{k,r}(P||Q)=Σ_x [p(x)-p(x)^{1-2k} q(x)^{2k}]/(2k), which is independent of r. Hence Theorems 6-8, Lemma 9, and the other divergence properties are properties of a one-parameter family, not of the claimed two-parameter generalization. The parameter r enters Definition 5 only through the deformed logarithm and cancels before any information-theoretic statement is made.","section":"Section 4, Definition 5"},{"comment":"The Hessian computation is internally inconsistent. Since D_{k,r} is r-independent, the induced metric cannot depend on r. Direct differentiation of the simplified expression D_{k,r}(P||Q)=Σ_i [p_i-p_i^{1-2k}q_i^{2k}]/(2k) gives ∂²D/∂p_i²|_{Q=P}=(1-2k)/p_i, not the reported (-2k+4r+1)/p_i. Thus Eq. (85), the metric in Eq. (86), and the Hessian-manifold claim in Theorem 9 are not supported by the definitions. Moreover, at k=1/2, which lies in the stated range 0<k≤1/2, the metric degenerates.","section":"Section 5, Eq. (85) and Theorem 9"},{"comment":"At the boundary k=1/2 of the allowed parameter range, the divergence collapses identically: D_{1/2,r}(P||Q)=Σ_x [p(x)-q(x)]/(1)=0 for all probability distributions P and Q. This contradicts the positivity claim in Lemma 9, which states that equality holds only for P=Q, and it violates the divergence axioms listed in Section 5. The one-parameter object itself therefore has a genuine internal consistency problem at an admissible parameter value.","section":"Section 4, Lemma 9"}],"minor_comments":[{"comment":"The term 'Hassian' should be 'Hessian' in Theorem 9 and in the concluding discussion; this is a typographical issue but appears repeatedly.","section":"Theorem 9 and Conclusion"},{"comment":"Equation (8) defines Ln_{k,r}(x) with x^r in the numerator, while Definition 1 defines ln_{k,r}(x) with x^r in the denominator. The two functions are different, and the notation is easy to confuse; an explicit remark distinguishing them would improve clarity.","section":"Eq. (8) and Definition 1"},{"comment":"The displayed joint convexity inequality in Eq. (4) appears to contain a typo: the right-hand side should be (1-λ)D_{k,r}(P^{(1)}||Q^{(1)})+λD_{k,r}(P^{(2)}||Q^{(2)}), not D_{k,r}(P^{(1)}||Q^{(1)})+λD_{k,r}(P^{(1)}||Q^{(1)}).","section":"Introduction, Eq. (4)"},{"comment":"The proof relies on a product-of-convex-functions fact with a citation but does not state the required hypotheses precisely; adding a short proof or an exact reference statement would make the argument fully self-contained.","section":"Theorem 1"}],"recommendation":"reject","confidential_remarks":"The central flaw is decisive: the proposed two-parameter entropy and divergence are one-parameter objects by the paper's own definitions, so the main contribution cannot be repaired by local revision. The Section 5 Hessian and Lemma 9 positivity issues are additional independent problems. The manuscript's extensive self-citation is not itself a technical concern, but it does not affect the correctness assessment."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the headline claim is false. Substituting Definition 1 into Definition 2 gives S_{k,r}(X) = (1 - Σ p^{2k+1})/(2k), and D_{k,r}(P||Q) = (1 - Σ p^{1-2k} q^{2k})/(2k). Both are independent of r; the first is Tsallis entropy with q=2k+1. The paper's own equations cancel the parameter it advertises.\n\nThere is real work here, and I want to credit it. The product rule in Lemma 1 is a clean algebraic identity, and it genuinely drives the chain rule, sub-additivity, strong sub-additivity, joint convexity, and information monotonicity proofs. Those proofs are self-contained and mostly correct for the surviving one-parameter family on 0<k≤1/2 (i.e., Tsallis q in (1,2]). If the paper were framed as an exposition of those properties using a deformed logarithm, the technical core would be acceptable, if not novel.\n\nThe soft spots are not minor. The claimed reduction to Furuichi's (α,β)-entropy, equation (19), is only correct when r=k; for general r there is a missing factor p^{r-k}. The Fisher metric calculation in Section 5 cannot be right because D does not depend on r; direct differentiation gives g_ii=(1-2k)/p_i, not (-2k+4r+1)/p_i. Worse, at k=1/2, which is inside the stated parameter range, the divergence is identically zero, so the metric degenerates. There is also a notational whiplash in Section 2: the original Sharma-Mittal logarithm uses x^r in the numerator and r<0, then Definition 1 moves x^r to the denominator and requires r>0. That switch is what makes the product rule work, but it is chosen for exactly that reason and has no external justification.\n\nThe citations are fine: some self-citations appear, but nothing load-bearing depends on them.\n\nWho is this for? Someone who wants a self-contained re-derivation of standard Tsallis entropy properties with a particular deformed logarithm might extract some pedagogical value from the proof sections. But the paper as written does not deliver the two-parameter generalization it promises. I would not send this to peer review in its current form. The honest recommendation is to reject and invite the authors to resubmit a much shorter note, stripped of the spurious r parameter, with the Hessian corrected.","headline":"The paper's own algebra cancels the r parameter from its entropy and divergence, so the advertised two-parameter family is Tsallis entropy with q=2k+1; the remaining one-parameter proofs are largely correct but standard.","tokens_in":20374,"tokens_out":4759,"would_cite":false,"duration_ms":46741,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A17","94A15"],"pacs":[],"model":"deepseek-v4-flash","headline":"A two-parameter deformed logarithm is claimed to yield a generalized Tsallis entropy with a full information-theoretic toolkit.","keywords":["two-parameter entropy","deformed logarithm","Tsallis entropy","Tsallis relative entropy","chain rule","sub-additivity","information monotonicity","Hessian metric"],"falsifier":"Take $P=(1/2,1/3,1/6)$ and compute $S_{1/4,1}(P)$ and $S_{1/4,2}(P)$ directly from Definition 2: because $S_{k,r} = (1-\\sum_x p^{2k+1})/(2k)$, both numbers are identical, which would contradict the claim that $S_{k,r}$ depends on two parameters; the same check on $D_{1/4,1}(P\\|Q)$ and $D_{1/4,2}(P\\|Q)$ gives equal values.","tokens_in":19040,"feed_emoji":"📊","tokens_out":10387,"duration_ms":96893,"temperature":0.7,"pith_summary":"This paper sets out to build a two-parameter generalization of Tsallis entropy and relative divergence from a deformed logarithm $\\ln_{k,r}(x) = (x^k - x^{-k})/(2k x^r)$. Its aim is to give this entropy the same working toolkit that Shannon and Tsallis entropies have: a chain rule, sub-additivity, strong sub-additivity, joint convexity, pseudo-additivity for independent systems, and information monotonicity under coarse-graining. The authors choose the deformed logarithm so that the weighted expression $x^{r+k}\\ln_{k,r}(x)$ simplifies to $(x^{2k}-1)/(2k)$, which makes a product rule hold and lets all these properties be derived by convexity arguments. If the construction works as claimed, it would place the Sharma-Mittal entropy family inside classical information theory and connect it to an information-geometric Hessian metric, with Tsallis and Shannon entropies recovered at special parameter values.","feed_headline":"Two-parameter entropy obeys chain rule and Tsallis limits","feed_subtitle":"A single product rule for the deformed logarithm drives sub-additivity, convexity, and monotonicity.","key_machinery":"The load-bearing object is the two-parameter deformed logarithm $\\ln_{k,r}(x) = (x^k - x^{-k})/(2k x^r)$, with $x^r$ placed in the denominator and $0<k\\le \\tfrac12$. This placement makes $x^{r+k}\\ln_{k,r}(x) = (x^{2k}-1)/(2k)$, a simple power, and it is exactly what makes the product rule of Lemma 1 hold. That product rule is the mechanism: it turns the deformed logarithm into a calculus in which joint entropies decompose into marginals and conditionals, so the chain rule, sub-additivity, strong sub-additivity, joint convexity, and information monotonicity all follow from Jensen-type convexity arguments applied to that one identity.","core_discovery":"On its own terms, the paper's central claim is that the pair $S_{k,r}(X) = -\\sum_x p(x)^{r+k+1} \\ln_{k,r}(p(x))$ and $D_{k,r}(P\\|Q) = \\sum_x p(x)(p(x)/q(x))^{r-k} \\ln_{k,r}(p(x)/q(x))$ constitutes a genuine two-parameter generalized Tsallis entropy and divergence. The engine is the product rule $(xy)^{r+k}\\ln_{k,r}(xy) = x^{r+k}\\ln_{k,r}(x) + y^{r+k}\\ln_{k,r}(y) + 2k x^{r+k} y^{r+k} \\ln_{k,r}(x) \\ln_{k,r}(y)$, which makes the joint entropy split into a marginal plus a conditional term. From this product rule the paper derives the chain rule, sub-additivity, strong sub-additivity, pseudo-additivity, joint convexity, and information monotonicity for the divergence, and it identifies parameter choices that recover Tsallis entropy and Tsallis relative entropy. The main asserted payoff is not any single formula but the infrastructure: conditional entropies satisfy the same inequalities as classical ones, and the divergence induces a Hessian metric on the probability simplex.","pith_inferences":["Direct algebra from Definition 2 shows the paper's $r$ drops out: $S_{k,r}(X) = (1-\\sum_x p(x)^{2k+1})/(2k)$ and $D_{k,r}(P\\|Q) = (1-\\sum_x p(x)^{1-2k}q(x)^{2k})/(2k)$, so the two-parameter label is not supported by the definitions; the family is the one-parameter Tsallis family reindexed with $q=2k+1$.","The chain rule, sub-additivity, and convexity proved in the paper are therefore properties of that one-parameter family; a genuine two-parameter theory needs a deformation in which the second parameter survives the entropy weighting.","One testable route to a real second parameter would be to weight the deformed logarithm by $p^\\alpha$ with an exponent independent of the one inside the logarithm, so that no telescoping cancellation occurs; the resulting product rule would be more complex but would keep $r$ visible.","The paper's Hessian calculation gives $g_{ii} = (1-2k+4r)/p_i$, which depends on $r$, even though the divergence it comes from does not; recomputing the second derivative from the simplified divergence suggests that $r$-dependence is an artifact of differentiating the un-simplified form."],"forward_implications":["If correct, the chain rule $S_{k,r}(X,Y) = S_{k,r}(X) + S_{k,r}(Y|X)$ holds, and with it sub-additivity $S_{k,r}(X,Y) \\le S_{k,r}(X) + S_{k,r}(Y)$ for arbitrary random variables.","The divergence $D_{k,r}$ is nonnegative, vanishes only for $P=Q$, is permutation invariant, admits a zero-probability extension, and is pseudo-additive under tensor products.","$D_{k,r}$ is jointly convex in $(P,Q)$ and monotone under stochastic maps, so it qualifies as an information-theoretic divergence and can support data-processing-type inequalities.","The divergence induces a Hessian metric on the probability simplex, giving the family a dually flat information-geometric structure analogous to that of the Kullback-Leibler divergence.","The special parameter limits recover Tsallis entropy and divergence, with Shannon entropy reached in the $k\\to 0$ limit, making the construction a potential interpolation tool."],"supporting_citations":[{"why":"Supplies the Sharma-Mittal deformed logarithm and entropy that the paper modifies to form $\\ln_{k,r}$.","marker":"[12]"},{"why":"Gives the product rule for the original deformed logarithm whose failure motivates the new placement of $x^r$ in the denominator.","marker":"[4]"},{"why":"Defines Tsallis entropy and its information-theoretic properties that the new entropy generalizes.","marker":"[8]"},{"why":"Provides the Tsallis relative entropy properties, including pseudo-additivity and convexity, that the new divergence extends.","marker":"[7]"},{"why":"Introduces the two-parameter entropy $S_{\\alpha,\\beta}$ whose form Definition 2 recovers.","marker":"[15]"},{"why":"Supplies the standard information-theoretic chain rule and divergence framework that motivates the construction.","marker":"[14]"},{"why":"Frames divergences as Hessian metrics on statistical manifolds, the framework used in Section 5.","marker":"[1]"}],"fun_headline_variants":["Product rule for deformed log gives entropy chain rule","Two-parameter entropy splits joint entropy via product rule","Deformed log product rule yields entropy inequalities","Two-parameter entropy: chain rule, sub-additivity, convexity"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on the choice to put $x^r$ in the denominator of the deformed logarithm, $\\ln_{k,r}(x) = (x^k - x^{-k})/(2k x^r)$, because that placement is what makes the product rule of Lemma 1 hold, and the chain rule, sub-additivity, and all later properties are derived from that product rule alone.","fun_headline_variants_meta":{"raw":{"variants":["Product rule for deformed log gives entropy chain rule","Two-parameter entropy splits joint entropy via product rule","Deformed log product rule yields entropy inequalities","Two-parameter entropy: chain rule, sub-additivity, convexity"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000699,"raw_usage":{"total_tokens":3129,"prompt_tokens":892,"completion_tokens":2237,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":508,"completion_tokens_details":{"reasoning_tokens":2173}},"tokens_in":508,"tokens_out":2237,"duration_ms":18058,"temperature":1.0,"reasoning_tokens":2173,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:08:45.112756+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $P=(1/2,1/3,1/6)$ and compute $S_{1/4,1}(P)$ and $S_{1/4,2}(P)$ directly from Definition 2: because $S_{k,r} = (1-\\sum_x p^{2k+1})/(2k)$, both numbers are identical, which would contradict the claim that $S_{k,r}$ depends on two parameters; the same check on $D_{1/4,1}(P\\|Q)$ and $D_{1/4,2}(P\\|Q)$ gives equal values.","supporting_citations":[{"cited_title":"Deformed logarithms an d en- tropies","cited_arxiv_id":null,"evidence_quote":"Supplies the Sharma-Mittal deformed logarithm and entropy that the paper modifies to form $\\ln_{k,r}$."},{"cited_title":"Legendre structure of the thermostatistics th eory based on the sharma–taneja–mittal entropy","cited_arxiv_id":null,"evidence_quote":"Gives the product rule for the original deformed logarithm whose failure motivates the new placement of $x^r$ in the denominator."},{"cited_title":"Information theoretical properties of tsallis e ntropies","cited_arxiv_id":null,"evidence_quote":"Defines Tsallis entropy and its information-theoretic properties that the new entropy generalizes."},{"cited_title":"Fundamenta l properties of tsallis relative entropy","cited_arxiv_id":null,"evidence_quote":"Provides the Tsallis relative entropy properties, including pseudo-additivity and convexity, that the new divergence extends."},{"cited_title":"An axiomatic characterization of a two-param eter ex- tended relative entropy","cited_arxiv_id":null,"evidence_quote":"Introduces the two-parameter entropy $S_{\\alpha,\\beta}$ whose form Definition 2 recovers."},{"cited_title":"Elements of information theory","cited_arxiv_id":null,"evidence_quote":"Supplies the standard information-theoretic chain rule and divergence framework that motivates the construction."},{"cited_title":"Methods of information geometry , volume 191","cited_arxiv_id":null,"evidence_quote":"Frames divergences as Hessian metrics on statistical manifolds, the framework used in Section 5."}],"review_version":1}