{"id":"0efbb88d-2e9f-4ad2-8d43-d575ec461a81","arxiv_id":"2608.09105","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For i.i.d. points with a bounded continuous density in a triangle, the longest convex chain grows like c n^{1/3}, with c given by a density-weighted affine arclength variational formula.","lead":"The paper shows that among n independent random points in a triangle, the longest convex chain has length proportional to the cube root of n, and it identifies the constant as a variational maximum over curves. The result generalizes the known uniform-density case and also describes the limiting shape of the longest chain.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; Theorem 1.1's proof is coherent, and the flagged H2 assumption is not load-bearing for the central growth-rate claim.","rationale":"The central result, Theorem 1.1, is proved under (H1) alone by a valid approximation argument, so the reader's flagged assumption (H2) is not load-bearing for the growth-rate statement. I checked the main upper- and lower-bound strategies, the variational analysis, the reduction to regular chains, the piecewise-convex repair step, and the comparison argument for dropping H2; no fatal gap emerged. The proof is long and technical, but the key steps are internally consistent. The only notable imperfections are the omitted proof of Lemma 4.2 and a slightly imprecise invocation of the concentration inequality in Claim 5.21; both are readily fixable and do not affect the validity of the theorems. Consequently, the reader's ACCEPT verdict should stand unchanged.","tokens_in":76403,"tokens_out":52701,"duration_ms":522666,"concrete_test":"Verify Claim 5.21 by re-deriving the concentration step: apply inequality (3.33) directly with s = n^{1/13} N^{1/4}, median m ≤ C N^{1/3}, and the Hoeffding lower bound N ≥ n p_l − n^{2/3}; check that the resulting tail is exp(−c n^{19/78}) (or at least exp(−a n^{1/13})) and that the union bound over F_n, whose size is n^{7L} with L = n^{1/157}, is summable. Also supply the omitted proof of Lemma 4.2 via the multivariate local central limit theorem.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I read the proof of Theorem 1.1 carefully and did not find a gap that would invalidate the central claim. The approximation argument in Section 6 (Claims 6.1–6.3) correctly proves the H1-only statement by comparison with the strictly positive density ep=(1−ε)p+2ε; the errors from L_gain and L_loss are O((ε n)^{1/3}), and Claim 6.1 gives |J~*−J*|=O(ε^{1/3}), so sending ε↓0 recovers the full theorem. The reader's weakest assumption, (H2), is indeed the most fragile premise in the paper, but only for the shape-concentration Theorems 1.2–1.4; Theorem 1.1 itself is proved under (H1) alone. Two minor issues worth noting: Lemma 4.2 is stated without proof, but it is a standard multinomial local-CLT lower bound and is not load-bearing. In Claim 5.21, the text invokes Proposition 3.6(d) with t=n^{1/13}, β=1/4, which directly bounds a deviation of size n^{17/52}, not the stated n^{1/13}N^{1/4}; however, the intended bound follows by applying Talagrand's inequality with s=n^{1/13}N^{1/4} and using the high-probability lower bound N≥np_l−n^{2/3}. This is a repairable exposition issue, not a substantive gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the length L_n of the longest convex chain among n i.i.d. points in the triangle T, for a general continuous density p. The main result, Theorem 1.1, states that L_n/n^{1/3} converges almost surely and in every L^p to (alpha/2) sup_{f in F} ∫_0^1 (f''(x) p(x,f(x)))^{1/3} dx, where alpha is the universal constant from the uniform-case theorem of Ambrus and Bárány. The proof splits into a lower bound (Section 4), an upper bound (Section 5), and an approximation argument (Section 6) that removes the strict-positivity assumption (H2) from the growth-rate theorem. The paper also proves shape concentration of near-longest chains around the maximizers of the variational functional (Theorem 1.2), a corresponding conditional statement when all samples form a convex chain (Theorem 1.3), and an asymptotic for the probability of that event (Theorem 1.4).","tokens_in":76700,"tokens_out":7098,"duration_ms":78479,"significance":"If correct, the paper gives the first general-density law of large numbers for longest convex chains, extending the uniform result of Ambrus and Bárány and connecting it to an affine-arclength variational formula analogous to the Deuschel–Zeitouni formula for longest monotone chains. The deterministic analysis of the variational problem is substantial and mostly self-contained: compactness and upper semicontinuity are proved in Section 2, the uniform case is solved explicitly, and the maximizer set is shown to be nonempty and compact. The probabilistic input is also carefully organized around tangency triangles, with external ingredients (the Ambrus–Bárány constant, Valtr's formula, and Talagrand's concentration inequality) cited precisely. The proof of Theorem 1.1 under continuity alone via the epsilon-perturbation in Section 6 is a genuine strength, since it shows that the central growth-rate claim does not depend on the strict-positivity assumption (H2); that assumption is used only for the shape-concentration and rare-event theorems 1.2–1.4. I found no load-bearing error in the central derivation.","major_comments":[],"minor_comments":[{"comment":"Lemma 4.2 is stated without proof. Since it supplies the multinomial lower bound used in the proof of Proposition 4.1(b), a citation or a short proof would improve self-containedness; the statement is a standard local central limit theorem and the omission is not a correctness concern.","section":"Section 4, Lemma 4.2"},{"comment":"The invocation of Proposition 3.6(d) is not literally correct as written: with t = n^{1/13} and beta = 1/4, that proposition bounds a deviation of size n^{1/13} n^{1/4} = n^{17/52}, whereas the display involves the random threshold n^{1/13} N_{n,ell}^{1/4}. The intended estimate follows by conditioning on N_{n,ell} and applying the concentration inequality with s = n^{1/13} N_{n,ell}^{1/4}, together with the high-probability lower bound on N_{n,ell} from Hoeffding; this should be stated explicitly.","section":"Section 5.3, Claim 5.21"},{"comment":"There are a few typographical and wording slips: 'we can send delta down to 0 we obtain (4.3)' is missing a 'to', and Claim 5.15 contains 'inequlities'. These do not affect the mathematics.","section":"Section 4, proof of Proposition 4.1"},{"comment":"In the proof of Lemma 5.4, 'togther' appears for 'together'. Also, the reduction from S_n to uniform samples U_n via the density bounds could be phrased more cleanly, though the argument is clear.","section":"Section 5.2, Lemma 5.4"}],"recommendation":"accept","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe one thing to know: this is the real deal. Bates and Sen prove the leading-order law of large numbers for longest convex chains under general continuous densities, with the constant given by a density-weighted affine-arclength variational formula. It generalizes Ambrus-Barany's uniform case and parallels Deuschel-Zeitouni for monotone chains. The paper is long, the proof is heavy, but I read the key parts and the architecture is coherent.\n\nWhat's new: Theorem 1.1 is the first variational formula for the n^{1/3} constant under non-uniform densities. The upper-bound machinery is genuinely novel: piecewise-convex approximants, a finite approximating family of size n^{O(L)} with L = n^{1/157}, and a repair step that recovers convex functions. The deterministic analysis of the variational problem (existence, regularity, approximation by smooth functions) is solid. Theorems 1.2-1.4 give shape concentration and a full-chain rare-event rate; these are natural and the proofs line up. External inputs (Ambrus-Barany's alpha, Valtr's formula, Talagrand's inequality) are cited and used correctly.\n\nSoft spots, in proportion. The (H2) lower-bound assumption is load-bearing for Theorems 1.2-1.4, exactly as the reader says; Theorem 1.1 itself is proved under (H1) alone via the epsilon-approximation in Section 6. That is an honest limitation, not a flaw. Lemma 4.2 is stated without proof, but it is a standard multinomial local-CLT estimate and not central. In Claim 5.21 the text says it applies Proposition 3.6(d) with t = n^{1/13}, beta = 1/4, which would control n^{1/13}N^{1/4} deviations, while the displayed bound uses n^{17/52}. The stress-test note is right: the intended bound follows by applying Talagrand with s = n^{1/13}N^{1/4} and using N >= np_l - n^{2/3}. It is a repairable exposition slip, not a gap.\n\nThis paper is for probabilists working on random geometry, last-passage percolation, and discrete convexity. It deserves a serious referee; I would not desk-reject. The exact value of alpha and the fluctuation theory remain open, but that is not a defect in a paper that settles the first-order law under general densities.\n\nMy recommendation: send it to a strong journal with a referee who can handle 79 pages of analysis. Ask the authors to add the proof of Lemma 4.2 and to fix the scaling statement in Claim 5.21. Both are routine.","headline":"Correct-looking, long, and genuinely new proof of the variational formula for longest convex chains; the additional density assumption (H2) only affects the shape-concentration theorems, not the main growth rate.","tokens_in":77199,"tokens_out":2321,"would_cite":true,"duration_ms":27095,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","60K37","60D05","82B44"],"pacs":[],"model":"deepseek-v4-flash","headline":"For n i.i.d. points in a triangle, the longest convex chain grows as c n^{1/3}, where c is a density-weighted affine arclength maximum and near-longest chains converge to its maximizers.","keywords":["convex chains","i.i.d. points","convex position","last-passage percolation","variational formula","affine arclength","limit shape","large deviations"],"falsifier":"Simulate the longest convex chain for the exponential density $p(x,y)\\propto e^{-4x}$ and for the uniform density on the same triangle at growing $n$, and compare $L_n^{(p)}/L_n^{(unif)}$ with $J_*(p)/2$ computed numerically from the variational formula: the paper predicts convergence to that constant, so a persistent mismatch would refute the claimed density dependence of the growth constant. In the uniform case the companion prediction $(1/n)\\log[(3n)!/n!\\,P(\\text{full chain})]\\to\\log 54$ can be checked directly against the exact formula $P(\\text{full chain})=2^n/(n!(n+1)!)$.","tokens_in":76207,"feed_emoji":"📐","tokens_out":17981,"duration_ms":173253,"temperature":0.7,"pith_summary":"This paper proves the leading-order law of the longest convex chain among $n$ i.i.d. points in a triangle, for any continuous sampling density. The claim is that $L_n/n^{1/3}$ converges to a universal constant times a maximum of a density-weighted affine arclength functional, so the density enters the problem only through a variational constant, and that all near-longest chains concentrate around the maximizing curves. If the paper is right, a natural geometric extremal statistic that was solved only for uniformly distributed points now has a complete first-order theory: the growth scale is $n^{1/3}$ (against $n^{1/2}$ for the analogous longest monotone chain problem), the limiting shape is the optimizer of the variational formula, and even the rare event that all $n$ points form one convex chain has a known rate. A sympathetic reader would care because the proof supplies the same kind of variational description, with optimizer concentration, that has been established for longest monotone chains, and it connects to the affine-perimeter functionals known from limit shapes of random convex lattice polygons.","feed_headline":"Longest convex chain from random points grows as n^{1/3}","feed_subtitle":"The leading constant is a density-weighted arclength maximum; near-longest chains hug the maximizing curve.","key_machinery":"The load-bearing object is the functional $J_p(f)=\\int_0^1 (f''(x)p(x,f(x)))^{1/3}\\,dx$ together with the tangency triangles attached to a candidate curve. For a convex $f$ and a short interval $[x,x+\\varepsilon]$, the tangency triangle is the triangle bounded by the two tangent lines of $f$ at the endpoints and the secant joining them; its area is $\\frac18 f''(x)\\varepsilon^3+o(\\varepsilon^3)$, so the mean number of sample points inside it is about $\\frac{n}{8}f''(x)p(x,f(x))\\varepsilon^3$. The lower bound partitions a curve into many such triangles, builds a longest chain inside each by comparison with the uniform case, and glues them with a concatenation lemma. The upper bound shows conversely that every convex chain is captured by the tangency triangles of some piecewise-convex function $\\tilde f$ drawn from a family of only $e^{o(n)}$ possibilities, so the chain length is controlled by $J(\\tilde f)$ up to negligible error, and a repair step converts $\\tilde f$ into a genuinely convex function with nearly the same $J$-value. Upper semicontinuity of $J$ on a compact metric space of convex functions guarantees that maximizers exist; the separation penalty (Proposition 2.15) then converts 'staying $\\varepsilon$ away from every maximizer' into a uniform loss of $\\theta$ in the growth constant, which is the mechanism behind the concentration theorems.","core_discovery":"At the center of the paper is Theorem 1.1: if $S_n$ is a set of $n$ independent samples from a triangle $T$ with continuous density $p$, and $L_n$ is the length of the longest convex chain in $S_n$, then $L_n/n^{1/3}\\to \\frac{\\alpha}{2}\\sup_{f\\in\\mathcal{F}}J_p(f)$ almost surely and in $L^p$ for every $p\\in[1,\\infty)$, where $\\mathcal{F}$ is the class of continuous convex functions $f:[0,1]\\to[0,1]$ with $f(0)=0$, $J_p(f)=\\int_0^1 (f''(x)p(x,f(x)))^{1/3}\\,dx$, and $\\alpha$ is the universal constant of the uniform case established in [4]. The functional $J_p$ is a density-weighted equi-affine arclength, and the supremum $J_*$ is attained because $J$ is upper semicontinuous on a compact metrization of $\\mathcal{F}$. Theorem 1.2 states that every convex chain whose length is within $\\delta n^{1/3}$ of $L_n$ lies within distance $\\varepsilon$ of the maximizer set of $J$, for a $\\delta$ depending on the density; Theorem 1.3 proves the same concentration for the conditional law with all $n$ samples forming one convex chain, at a faster exponential rate; and Theorem 1.4 identifies the rate of that rare event as $(1/n)\\log[(3n)!/n!\\,P(S_n\\text{ is a convex chain})]\\to \\log(27J_*^3/4)$. The growth-rate theorem holds with only continuity of $p$, obtained by perturbing the density with uniform mass; the shape statements assume in addition that $p$ stays bounded away from zero.","pith_inferences":["Because the universal constant $\\alpha$ is unknown, ratios are the cleanest test of the formula: the theory predicts $L_n^{(p)}/L_n^{(q)}\\to J_*(p)/J_*(q)$ for two densities, so taking $q$ uniform (where $J_*=2$) isolates the density dependence that the paper predicts, without needing the value of $\\alpha$.","The positivity assumption on the density is likely not intrinsic: for densities supported on a proper subtriangle the same proof scheme should give a version of the theorem on that subtriangle, whereas densities that vanish smoothly at the boundary would require a different control of maximizer regularity and of points near the boundary.","For separable densities $p(x,y)=u(x)v(y)$ the variational integral may reduce to an Euler–Lagrange equation with explicit solutions, which would supply the first computed non-uniform maximizers and a sharp quantitative test of the shape-concentration statement.","The proof bounds fluctuations of $L_n$ by $n^{1/6+o(1)}$ but leaves their true order open; whether the typical fluctuations actually reach the $n^{1/6}$ scale, and with what distribution, is a question the paper explicitly leaves to future work."],"forward_implications":["The leading-order length of the longest convex chain is $(\\alpha/2)J_*(p)\\,n^{1/3}$ almost surely: the sampling density changes only the multiplicative constant, not the $n^{1/3}$ scale.","Near-longest chains have a deterministic limit shape: any chain of length at least $L_n-\\delta n^{1/3}$ is $\\varepsilon$-close to the set of maximizers of $J_p$, and when the maximizer is unique the limit is a single explicit curve.","The probability that all $n$ points lie in convex position obeys $(1/n)\\log[(3n)!/n!\\,P(\\text{full chain})]\\to\\log(27J_*^3/4)$, pinning the rare-event rate up to subexponential factors.","Conditioning on the full-sample chain changes the regime — fluctuations of $L_n$ vanish — but the same maximizing curves describe the shape, now with concentration at the faster $e^{-cn}$ rate in Theorem 1.3.","The functional appearing in the constant is the affine perimeter that already governs limit shapes of random convex lattice polygons, so the result connects random-point chains to that existing universality."],"supporting_citations":[{"why":"supplies the uniform-case constant alpha (Theorem 1.6) on which the variational formula and all comparisons rest","marker":"[4]"},{"why":"the longest-monotone-chain analogue whose variational formula and optimizer concentration this paper extends to convex chains","marker":"[26]"},{"why":"gives the exact probability (3.32) that n uniform points in a triangle lie in convex position, used for full-chain likelihood and tail estimates","marker":"[10]"},{"why":"provides the convex-position probability for uniform points in a parallelogram that controls chains confined to small squares in Lemma 5.9","marker":"[39]"},{"why":"reproduced as Lemma 3.9, it supplies the concentration inequality behind every exponential tail in the argument","marker":"[31]"},{"why":"the doubling-measure localisation lemma used in Lemma 2.17 to prove maximizers have absolutely continuous derivatives","marker":"[38]"},{"why":"its display (5.3) gives the uniform convex-position probability used alongside (3.32) in Proposition 3.6(b)","marker":"[9]"}],"fun_headline_variants":["Longest convex chain from i.i.d. points: n^{1/3} asymptotics","Random triangle points: longest convex chain is ~n^{1/3}","Scaling law for longest convex chain from i.i.d. points: n^{1/3}","n^{1/3} growth for longest convex chain in random triangle"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the sampling density stays bounded away from zero everywhere in the triangle: without that, the exponential estimates behind shape concentration, the regularity of maximizers, and the full-chain rare-event rate are not proved, and only the $n^{1/3}$ growth rate itself survives on continuity alone.","fun_headline_variants_meta":{"raw":{"variants":["Longest convex chain from i.i.d. points: n^{1/3} asymptotics","Random triangle points: longest convex chain is ~n^{1/3}","Scaling law for longest convex chain from i.i.d. points: n^{1/3}","n^{1/3} growth for longest convex chain in random triangle"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000695,"raw_usage":{"total_tokens":3221,"prompt_tokens":1098,"completion_tokens":2123,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":714,"completion_tokens_details":{"reasoning_tokens":2032}},"tokens_in":714,"tokens_out":2123,"duration_ms":16208,"temperature":1.0,"reasoning_tokens":2032,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T23:28:11.881106+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate the longest convex chain for the exponential density $p(x,y)\\propto e^{-4x}$ and for the uniform density on the same triangle at growing $n$, and compare $L_n^{(p)}/L_n^{(unif)}$ with $J_*(p)/2$ computed numerically from the variational formula: the paper predicts convergence to that constant, so a persistent mismatch would refute the claimed density dependence of the growth constant. In the uniform case the companion prediction $(1/n)\\log[(3n)!/n!\\,P(\\text{full chain})]\\to\\log 54$ can be checked directly against the exact formula $P(\\text{full chain})=2^n/(n!(n+1)!)$.","supporting_citations":[{"cited_title":"https://doi.org/10.1002/rsa.20269","cited_arxiv_id":null,"evidence_quote":"supplies the uniform-case constant alpha (Theorem 1.6) on which the variational formula and all comparisons rest"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the exact probability (3.32) that n uniform points in a triangle lie in convex position, used for full-chain likelihood and tail estimates"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the convex-position probability for uniform points in a parallelogram that controls chains confined to small squares in Lemma 5.9"},{"cited_title":"InNonlinear Analysis, Function Spaces and Applications","cited_arxiv_id":null,"evidence_quote":"the doubling-measure localisation lemma used in Lemma 2.17 to prove maximizers have absolutely continuous derivatives"}],"review_version":1}