{"id":"6ca3549a-e9e1-4f97-a02b-4fdee6448fed","arxiv_id":"2607.29042","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"For every discrete real-valued random variable X with finite entropy, max{H(X+X'), H(XX')} ≥ (8/7)H(X) - O(log H(X)).","lead":"A new mathematical proof shows that for any discrete random variable with finite Shannon entropy, either adding two independent copies or multiplying two independent copies must add at least about one seventh of a bit of entropy per bit. This is the first proof of the entropic sum-product phenomenon and answers a conjecture of Goh.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified","rationale":"The reader's weakest_assumption identified the point-plane incidence bound (Theorem 2.3) and its unpinned constant C_R as the principal caveat. I agree this is the most non-explicit part of the argument, but I do not consider it load-bearing for the central asymptotic claim: the paper proves the 8/7 coefficient with an O(log H) error, and C_R enters only as an additive constant that cannot affect the leading term. I verified the small-doubling chain: the Solymosi-type bound (Theorem 3.1) is correct — the non-obvious equality between the ray-slope count and the multiplicative energy follows from the involution (a,b,a′,b′) ↦ (a,b′,a′,b), which maps ab=a′b′ to b′/a=b/a′. The level-set decomposition (Lemmas 4.2–4.8) and the recombination (Propositions 4.10, 4.12) are internally consistent. The large-doubling chain is also sound: the incidence construction (Lemmas 5.5–5.7) verifies the hypotheses of Theorem 2.3, including the line condition in the unbalanced regime, and the corrected upper bound (Proposition 6.1) fixes a genuine error in [9] without introducing a gap. The conversion from zero-free bounds to the general result (Proposition 7.2) is rigorous, and the truncation (Lemma 7.5) correctly extends to infinite support. No machine-checked proof is provided, so residual risk exists, but no specific flaw was found. Hence the reader's ACCEPT verdict stands unchanged; the C_R caveat merits a note but not a verdict change.","tokens_in":37170,"tokens_out":48487,"duration_ms":382378,"concrete_test":"For the unbalanced regime |G1| ≤ |G2||G3| of Lemma 5.7, construct P and Q as in Lemma 5.5 for random finite sets G1,G2,G3 with |G1| ≈ |G2||G3| and |G2| ≪ |G3|, and brute-force compute I(P,Q) for small sizes to confirm that I(P,Q) ≤ C_R N^{3/2} for a universal absolute constant C_R, and that every line contained in a plane of Q meets P in at most max{|G1|,|G2|} points. If this fails for some family, the large-doubling half and the 8/7 coefficient would be in doubt; if it holds, the remaining unpinned constant is only an additive-explicitness issue.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the proof of Theorem 1.5 in detail. The two-stage argument — small-doubling via the Solymosi energy bound (Sections 3–4) and large-doubling via the Rudnev/de Zeeuw incidence bound (Sections 5–6) — is internally coherent, and the combination in Proposition 7.2 correctly yields δ = 1/7. The only caveat is the unpinned absolute constant C_R in Definition 2.4: no explicit admissible value appears in the literature, so the additive term −(1/2)log C_R in (1.2) is not numerically explicit. However, this does not affect the asymptotic claim (1.3), since C_R is an absolute constant and is absorbed by the O(log H) term. I found no step where an inequality is used outside its stated regime: the line condition in Lemma 5.6 is verified for the unbalanced case, and the key identity in Theorem 3.1's proof (Σ n_s^2 = E×) is correct via the involution (a,b,a′,b′) ↦ (a,b′,a′,b). The corrected upper bound in Proposition 6.1 and the truncation argument in Lemma 7.5 are also sound. Overall, no load-bearing flaw was identified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves the first entropic sum-product phenomenon with an explicit coefficient strictly greater than 1. Specifically, Theorem 1.5 shows that for i.i.d. discrete real-valued random variables X, X' with finite entropy, max{H(X+X'), H(XX')} ≥ (8/7)H(X) - O(log H(X)), answering Goh's Conjecture 1.2 with δ = 1/7. The proof has two main branches: a small-doubling bound (Theorem 1.6) obtained by adapting Solymosi's multiplicative-energy argument via a dyadic uniformization technique, and a large-doubling bound (Theorem 1.7) built on the Rudnev–de Zeeuw point-plane incidence bound and a corrected upper bound for H(X(Y+Z)) (Proposition 6.1). The two branches are combined in Proposition 7.2, after which an atom at zero is reincorporated and the finite-support assumption is removed by truncation. The paper also contains a detailed erratum for a flawed bound in prior work of Gavalakis–Goh–Kontoyiannis.","tokens_in":37473,"tokens_out":5218,"duration_ms":54195,"significance":"If correct, this is a substantial result: it gives the first coefficient strictly larger than 1 for the entropic sum-product phenomenon, resolving an open conjecture. The proof is long but essentially self-contained, and the constants are explicit up to the single absolute incidence constant C_R of Definition 2.4. The dyadic uniformization lemma (Lemma 2.7) and the linear entropic Elekes–Ruzsa theorems are likely to be useful beyond this paper. The paper also provides a careful correction of a previously published upper bound, which increases confidence in both the present proofs and the salvageability of the earlier results. The only caveat—that no explicit admissible value of C_R appears in the literature—is acknowledged by the authors and does not affect the asymptotic claim (1.3), since C_R is an absolute constant absorbed by the O(log H) term.","major_comments":[],"minor_comments":[{"comment":"The constant C_R is defined as the implied constant in Theorem 2.3, but the paper notes that no explicit admissible value is known. Since (1.2) explicitly displays the term −(1/2)log C_R, it may be helpful to add a sentence in the introduction emphasizing that the asymptotic statement (1.3) is fully explicit while the finite-constant version is qualitative in this one additive constant.","section":"Definition 2.4 and Theorem 1.5"},{"comment":"The extension of ϑ(η) to the range 0 < η < 4 via ϑ(η) = ϑ(4) is somewhat ad hoc. Since the main theorems are asymptotic, it might be cleaner to state the theorem for η ≥ 4 and treat small entropy separately, avoiding the artificial constant 93/4.","section":"Corollary 4.14"},{"comment":"The proof of Example 6.5 is described as 'routine but tedious' and some entropy computations are summarized rather tersely, especially the claim that H(X+X' | X>0, X'>0) = log n + O(1). A few additional lines of derivation would improve readability without affecting correctness.","section":"Section 6.3, Example 6.5"},{"comment":"The existence of a finite F with both P(X∉F) ≤ θ and Σ_{x∉F} p(x) log(1/p(x)) ≤ θ follows by convergence of the defining series, but the proof could be more explicit that one takes a sufficiently large finite initial segment after arranging the support in any order.","section":"Lemma 7.5"}],"recommendation":"accept","confidential_remarks":"The manuscript is a strong contribution and well within the journal's scope. The stress-test concern regarding the unbalanced incidence regime does not materialize as a flaw: the hypotheses of Theorem 2.3 are verified in Lemma 5.6 and applied exactly in the regime |G1| ≤ |G2||G3|. The unpinned constant C_R is a limitation only of the explicit finite-constant version, not of the central asymptotic claim. I recommend acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper deserves a serious referee. It proves max{H(X+X'), H(XX')} ≥ (8/7 − o(1))H(X), the first coefficient strictly above 1, answering Goh's conjecture. I read the main line carefully and the argument holds up.\n\nWhat's new: dyadic uniformization to control the entropy loss when splitting into flat pieces, which is what kills the min-entropy bottleneck in [9]; the two 'linear entropic Elekes–Ruzsa' inequalities (small-doubling via a Solymosi-style energy bound, large-doubling via Rudnev/de Zeeuw incidence). The combination of the two lines at u=1/7 is clean. They also fix a real error in [9]'s upper bound; the correction term is exactly the zero-probability term, and their counterexamples are convincing.\n\nSoft spots: the constant C_R in the point-plane incidence bound has no explicit admissible value in the literature, so the additive constants in (1.2) are not numerically explicit. That's annoying but doesn't touch the asymptotic claim. The proof is long and not machine-checked, so I'd set confidence moderate rather than high, though I found no load-bearing gap. The Solymosi half is restricted to R because of the ordering; the field remark at the end is appropriately cautious. The conjecture δ=1/3 remains open.\n\nWho it's for: anyone working in additive combinatorics, entropy, or incidence bounds. It's a genuine structural contribution, not just a constant improvement.\n\nMy verdict: send it to referees. It deserves careful reading; referees should double-check the incidence application in the unbalanced case (Lemma 5.7) and the constant-tracking in Sections 3–4, but I'd be surprised if the main theorem breaks.","headline":"First real progress on Goh's entropic sum-product conjecture: δ=1/7 with a long but sound proof.","tokens_in":37916,"tokens_out":1452,"would_cite":true,"duration_ms":16283,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["94A17","11B75","60E05"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves the entropic sum-product phenomenon: for any i.i.d. discrete real-valued random variables with finite entropy, either the sum or the product has entropy at least 8/7 of the original entropy, up to a logarithmic correction.","keywords":["entropic sum-product phenomenon","Shannon entropy","sum-product conjecture","point-plane incidence bound","dyadic uniformization","multiplicative energy","few-sums-many-products","additive doubling"],"falsifier":"Run a search over finite sets G1,G2,G3 ⊂ R satisfying |G1| ≤ |G2||G3| and the line condition, computing the collision count Col = #{(a,b,c,a',b',c'): a(b+c)=a'(b'+c')}; if any family has Col growing faster than a constant times (|G1||G2||G3|)^{3/2}, Theorem 5.1 and hence the 8/7 coefficient collapse. Since the constant C_R is unpinned, an explicit calculation of Col for small asymmetric sets (e.g., |G1|≈|G2||G3|) would also pin down whether the stated constants are realistic.","tokens_in":37093,"feed_emoji":"🔢","tokens_out":9813,"duration_ms":85860,"temperature":0.7,"pith_summary":"This paper proves the entropic analog of the sum-product phenomenon: for any two independent copies of a discrete real-valued random variable with finite Shannon entropy, either their sum or their product must have entropy at least 8/7 of the original entropy, up to an O(log H) correction. This answers a conjecture in the literature that merely asked for a coefficient strictly larger than 1. A sympathetic reader should care because it shows a fundamental tension: a probability distribution cannot be nearly closed under both addition and multiplication, just as a set of numbers cannot. The previous tools were blocked by distributions with a large gap between Shannon entropy and min-entropy; the paper removes this blockage by slicing the distribution into near-uniform pieces, at a cost of only O(log H) bits. Two complementary linear bounds are then combined, one effective at small additive doubling and one at large doubling, yielding the 8/7 coefficient.","feed_headline":"Sum or product of a random variable exceeds 8/7 of its entropy","feed_subtitle":"First result proving a coefficient above 1 for the entropic sum-product phenomenon, answering the conjecture with δ = 1/7.","key_machinery":"Three tools carry the argument. A dyadic uniformization lemma cuts any finite-entropy law into near-uniform pieces of size powers of two, at a total loss of O(log H) bits, replacing the min-entropy bottleneck of earlier work. Two linear entropic few-sums-many-products bounds are then proved: one from a multiplicative-energy estimate on sign-pure sets, giving product entropy ≥ 2 − 6eu; the other from a point-plane incidence bound, giving product entropy ≥ 5/4 − 3eu/4. Finally, a recombination proposition converts the envelope of these two zero-free bounds into a 8/7 coefficient for all laws, including those with an atom at zero.","core_discovery":"Central claim: max{H(X+X'), H(XX')} ≥ (8/7)H(X) − O(log H(X)) for any i.i.d. discrete real-valued variables with finite entropy; asymptotically this is (1 + 1/7 − o(1))H(X). The proof cuts the law into near-uniform dyadic pieces at O(log H) cost, proving two zero-free linear bounds: product entropy ≥ (2 − 6eu − o(1))H and ≥ (5/4 − 3eu/4 − o(1))H, where eu is relative additive doubling. Their envelope crosses at eu = 1/7; a recombining step preserves the coefficient when there is an atom at zero, and truncation covers all discrete laws.","pith_inferences":["The same framework is likely to yield the optimal δ = 1/3 if a single linear bound with intercept 4/3 can be proved; the paper's Conjecture 1.10 states exactly the inequality that would imply it, so the bottleneck is sharpening the slope-6 coefficient in the small-doubling line.","Since no explicit value of the incidence constant C_R is known, the theorem's stated constants c1=18, c2=63 should be treated as schematic; the asymptotic 8/7 is unaffected, but effective versions depend on a future explicit incidence bound.","The order structure of the real line is used essentially in the multiplicative-energy half; the authors note the large-doubling half extends to characteristic-0 fields, suggesting a testable extension: prove an entropic sum-product bound over Q or number fields with a smaller δ, showing what loss the absence of ordering imposes.","Because the paper shows adding an atom at zero merely contracts the difference in Conjecture 1.10, we expect future improvements can focus on zero-free laws; the hard case is not the zero atom."],"forward_implications":["This is the first proof that the entropic sum-product coefficient exceeds 1, settling the conjecture with δ = 1/7; the known upper bound is 4/3, so the optimal constant lies in [8/7, 4/3].","The dyadic uniformization lemma with O(log H) loss is a reusable tool: any entropy inequality currently limited by min-entropy should be re-examinable in Shannon-entropy form.","The corrected product-sum upper bound (Proposition 6.1) repairs an error in earlier published bounds; the paper shows the earlier main theorems survive with small modifications.","Uniformizing a finite set A recovers a combinatorial sum-product statement with exponent 1/7 from the entropic theorem, though the combinatorial record remains higher.","The small-doubling bound gives a genuinely linear entropic 'few sums force many products' inequality, stronger than the additive-constant Freiman-type bounds previously available."],"fun_headline_variants":["Entropic sum-product: max(H(X+Y),H(XY)) ≥ 8/7 H(X)","Sum or product entropy exceeds 8/7 of original entropy","New bound: entropy of sum or product ≥ 8/7 H(X)","8/7 entropy guarantee for sum or product of a variable"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The 8/7 coefficient rests on the point-plane incidence bound holding for arbitrary finite point and plane sets in real three-dimensional space with an absolute constant, in the unbalanced regime |G1| ≤ |G2||G3|; no explicit admissible value of that constant appears in the literature, so the stated additive constants are not numerically pinned.","fun_headline_variants_meta":{"raw":{"variants":["Entropic sum-product: max(H(X+Y),H(XY)) ≥ 8/7 H(X)","Sum or product entropy exceeds 8/7 of original entropy","New bound: entropy of sum or product ≥ 8/7 H(X)","8/7 entropy guarantee for sum or product of a variable"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000187,"raw_usage":{"total_tokens":1189,"prompt_tokens":794,"completion_tokens":395,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":538,"completion_tokens_details":{"reasoning_tokens":322}},"tokens_in":538,"tokens_out":395,"duration_ms":4486,"temperature":1.0,"reasoning_tokens":322,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T14:44:59.490989+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a search over finite sets G1,G2,G3 ⊂ R satisfying |G1| ≤ |G2||G3| and the line condition, computing the collision count Col = #{(a,b,c,a',b',c'): a(b+c)=a'(b'+c')}; if any family has Col growing faster than a constant times (|G1||G2||G3|)^{3/2}, Theorem 5.1 and hence the 8/7 coefficient collapse. Since the constant C_R is unpinned, an explicit calculation of Col for small asymmetric sets (e.g., |G1|≈|G2||G3|) would also pin down whether the stated constants are realistic.","supporting_citations":[],"review_version":1}