{"id":"bf03be3c-33fc-4521-9007-851ca0f80574","arxiv_id":"2509.07232","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"For any copula, Spearman's footrule cannot exceed the square root of Chatterjee's ξ, and for stochastically increasing copulas the exact attainable region is ξ ≤ ψ ≤ √ξ.","lead":"This paper maps the joint range of Chatterjee's rank correlation and Spearman's footrule over copulas. It proves that the upper edge of this range is the curve ψ=√ξ, attained only by Fréchet copulas, and pins down the exact region for stochastically increasing copulas.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.4's exact SI region rests on the external pointwise Markov-product inequality [24, Thm. 4.2]; independent verification of Eq. (12) is the decisive check.","rationale":"The central claims — Theorem 2.1 (max ψ given ξ) and Theorem 2.4 (exact SI region) — are proved with checkable KKT arguments, and the convex-optimization relaxations are handled carefully. My reading confirms the reader's identification of Eq. (12), i.e., [24, Thm. 4.2], as the load-bearing external assumption for Theorem 2.4: without it, both ξ≤ψ and the equality characterization in Prop. 2.2 fail. The inequality itself is plausible and can be reduced to an elementary statement about non-increasing h_v(t); my spot checks did not produce a counterexample. I therefore treat this as a verification concern, not a proven flaw, and keep the verdict unchanged. A secondary observation: Theorem 3.3's proof of the inclusion (25) explicitly treats only copulas with ψ(C)≤0, while the theorem states the bound for all y∈[-1/2,1]; for y>0 a supporting argument for the extended algebraic family would be needed. Since the paper explicitly disclaims exactness of the full lower boundary, this does not affect the headline SI-region result.","tokens_in":17926,"tokens_out":49547,"duration_ms":405042,"concrete_test":"Take the non-symmetric SI copula from Remark 2.3 with ∂1C(t,v)=1_{t<v/2}+v1_{(v/2,(v+1)/2)} and, on a fine grid of v (e.g., 10^4 points), compute both (C⊤∗C)(v,v)=∫_0^1 (∂1C(t,v))^2 dt and C(v,v)=∫_0^v ∂1C(t,v) dt; verify the difference C(v,v)-(C⊤∗C)(v,v) is nonnegative to numerical precision. Repeat for a family of ordinal sums of Π with several intervals. If any violation appears, [24, Thm. 4.2] is false and Theorem 2.4 fails; if none appears, the external pillar is corroborated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The exact SI region in Theorem 2.4 and the equivalence in Prop. 2.2 both hinge on the pointwise Markov-product inequality (C⊤∗C)(v,v)≤C(v,v) for stochastically increasing copulas, taken from [24, Thm. 4.2] and used as Eq. (12). This is an external theorem not re-derived in the paper; if it failed for any SI copula, the inequality ξ≤ψ and hence the characterization R_SI={(x,y):x≤y≤√x} would collapse. The internal proof of Prop. 2.2 treats the inequality as given and only derives equality cases, so confidence in the main result is exactly confidence in that external theorem. The same external source supplies [24, Thm. 5.1] for idempotence of ordinal sums of Π, which is used to attain the diagonal; the proof of Theorem 2.4 does not otherwise depend on the author's preprint [4].","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the attainable joint range of Chatterjee's rank correlation ξ and Spearman's footrule ψ over bivariate copulas. The main results are: (i) Theorem 2.1 proves that for each x∈[0,1] the maximum possible ψ given ξ=x is √x, uniquely attained by the Fréchet copula C^Fr_α with α=√x; the proof uses a convex optimization reformulation over partial derivatives and verifies the KKT conditions as well as a quadratic-growth second-order condition. (ii) Theorem 2.4 characterizes the exact (ξ,ψ)-region for stochastically increasing copulas as R_SI = {(x,y): x≤y≤√x}, with the lower boundary attained by ordinal sums of the independence copula and the upper boundary uniquely by the Fréchet family. (iii) Section 3 provides a lower bound for the minimal ψ given ξ, based on Jensen's inequality, proves that at ψ=-1/2 the minimal ξ is 1/2 (uniquely attained by a checkerboard copula), and constructs a two-parameter copula family that approximately realizes the lower boundary. The paper is clearly written and explicitly acknowledges that the lower boundary for all copulas is not exactly characterized.","tokens_in":18199,"tokens_out":43086,"duration_ms":348625,"significance":"If correct, the upper-bound theorem and the exact SI region are substantive, elegant results: they give a sharp, simple geometric description of the joint range of two popular dependence measures, and the uniqueness of the Fréchet maximizer is non-obvious. The KKT proof of Theorem 2.1 is detailed and rigorous, including stationarity, primal/dual feasibility, complementarity, positive-definite Hessian, and convexity-based global minimality. The paper also gives a clean construction (ordinal sums of Π) attaining the full diagonal in the SI region, and it is honest about the fact that the global lower boundary is only bounded, not exactly reached. A notable caveat is that the SI-region result rests on the external pointwise inequality (C^T*C)(v,v) ≤ C(v,v) for stochastically increasing copulas, cited from [24, Thm. 4.2]; the manuscript does not re-derive this inequality, so the exactness of R_SI is contingent on that published theorem. Additionally, several supporting results in Section 3 are quoted from the author's own preprints [4] and [23], which are not yet peer-reviewed.","major_comments":[{"comment":"The proof that the full attainable region R is convex is incomplete. The argument mixes two lower-boundary copulas C0 and C1 and observes that x_λ := ξ(C_λ) ≤ (1−λ)x0+λx1 and y_λ := ψ(C_λ) = (1−λ)φ(x0)+λφ(x1). However, membership of (x_λ,y_λ) in R gives only y_λ ≥ φ(x_λ), and since x_λ may be strictly smaller than the weighted average, this does not imply the convexity inequality φ((1−λ)x0+λx1) ≤ (1−λ)φ(x0)+λφ(x1). The subsequent sentence that the point 'lies weakly below the straight line' does not establish that it lies above the lower boundary. Moreover, the assertion that 'R is only limited by the upper and lower boundaries' requires a proof that vertical slices are intervals; the paper only establishes that horizontal slices R_y are intervals. The closure proof depends on the same convexity claim. Please provide a complete proof or weaken the statement of Theorem 3.3 by omitting the convexity/closedness assertion, which is not needed for the inclusion (25).","section":"Section 3.3, proof of Theorem 3.3 (convexity of R)"},{"comment":"The displayed inclusion in Theorem 3.3 is not well-defined for y>0. The parameter µ(y) is introduced as the unique solution in [0,2] to the cubic equation, but as the proof itself notes, the equation has such a solution only for y∈[−1/2,0], because ψ(C↘_µ) ranges from 0 to −1/2 on that interval. For y>0, the expression ξ(C↘_µ(y)) is therefore undefined, yet the set in (25) ranges over y∈[−1/2,1]. Please restrict the lower-bound inequality to y≤0 and state separately that for y>0 the only bound used is y≤√x (equivalently x≥y^2).","section":"Theorem 3.3, statement (25)"},{"comment":"The proof of Theorem 3.3 relies on [4, Prop. 2.4 & Cor. 2.5] (the existence of an SD rearrangement preserving ξ and not increasing ψ) and on [2, Cor. 3.6] (continuity of ξ on C_SD). Both are results from unpublished preprints by the author and coauthors. Since this reduction is load-bearing for the lower-bound inclusion (25), the manuscript should either state these results explicitly with proofs or cite published versions; otherwise a reader cannot verify the key step that inf{ψ(C): ξ(C)=x} equals the infimum restricted to SD copulas. If the results are correct, this is a strength, but the current manuscript makes the verification difficult.","section":"Section 3.3, reduction to SD copulas and external dependencies"}],"minor_comments":[{"comment":"The formula α(µ) = 1/2 − 2/5 µ for µ>2 yields negative values (e.g., µ=2.5 gives α=−0.5), contradicting the standing assumption α∈[0,0.5). This makes the path C_µ undefined for µ>2. If the intention was to continue from the point (0.3,0.5) toward (0.5,0.5) along β=0.5, a different expression (possibly 1/2 − 2/(5µ)) is needed.","section":"Section 3.2, Eq. (33)"},{"comment":"The displayed objective 'minimize µvh 1 + vh 2 1 + (1−v)h 2 2' is typographically confusing; it should read µ v h_1 + v h_1^2 + (1−v) h_2^2. Please clarify the notation.","section":"Proof of Theorem 3.2, pointwise problem (22)"},{"comment":"Several load-bearing results are cited from the author's preprints [4] and [23], as well as from other very recent preprints ([1], [2], [14]). It would help the reader if the introduction or a remark explicitly stated which external results are required for each main theorem and which of these are not yet published.","section":"Introduction and references"},{"comment":"The caption refers to a 'solid line from Π to C#' and a 'dotted line from Π to C#' as lower-bound curves. In black-and-white print these two curves may be difficult to distinguish; please use more distinct line styles or colors and add explicit labels in the figure itself.","section":"Figure 1 caption"}],"recommendation":"major_revision","confidential_remarks":"The core results (Theorems 2.1 and 2.4) appear sound, and the SI-region characterization is attractive. The main weakness is the proof of Theorem 3.3: the convexity/closedness argument is not rigorous and the statement of the inclusion (25) is ill-posed for y>0. The author should also address the heavy reliance on unpublished preprints [4], [23] and [2] for load-bearing steps. These are fixable within the manuscript's scope, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read Rockel's paper. The stuff worth knowing: Theorem 2.1 gives the upper boundary ψ ≤ √ξ for all copulas, with unique maximizer the Fréchet family. That's a clean result, and the proof is unusually complete for this literature: stationarity, primal/dual feasibility, complementary slackness, positive Hessian, and convexity are all spelled out. I checked the algebra in the KKT system; it works. Theorem 2.4 characterizes the full (ξ,ψ) region over stochastically increasing copulas as x ≤ y ≤ √x, with the diagonal attained by ordinal sums of Π and the upper curve by Fréchet copulas. If true, this is the right extension of the exact-region program from Spearman's ρ to footrule. The paper is honest that the lower boundary of the full copula region is not exact: Jensen gives a bound, a non-copula family C↘_µ is used to evaluate that bound, and a two-parameter copula family gets close numerically. That is a real limitation but stated openly, not a hidden flaw.\n\nThe main soft spot is exactly the one flagged in the stress test: the SI region hinges on [24, Thm. 4.2], the pointwise Markov-product inequality (C⊤∗C)(v,v) ≤ C(v,v). The paper invokes it twice without proof. I haven't gone through [24] myself, so I can't certify it from the paper alone. But it is a published theorem in J. Math. Anal. Appl. from 2021, not a private communication, and the usage looks faithful. If I were refereeing, I'd ask the author to state the result explicitly in a self-contained way or at least quote it in full. This is a prop, not a reason to reject. The paper also leans on the author's own preprint [4] for the copula-derivative characterization; that lemma is plausible and was presumably already vetted for the ρ-paper, but a referee may want it stated rather than cited.\n\nTheorem 3.4 (minimal ξ = 1/2 when ψ = −1/2, uniquely via checkerboard) is a nice bonus. It uses external results from [13] and [23], but those are also legitimate published checks.\n\nBottom line: solid mathematical paper, well executed, clear scope. The exact SI region is the kind of result people will cite. It deserves a serious referee. My only real request is transparency about the cited external inequalities.","headline":"A clean, well-executed proof of the upper boundary ψ≤√ξ and the exact (ξ,ψ) region for stochastically increasing copulas; the main caveat is heavy reliance on one external theorem that is not re-derived.","tokens_in":18703,"tokens_out":1915,"would_cite":true,"duration_ms":17822,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62H20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The attainable (ξ, ψ) region for stochastically increasing copulas is exactly x ≤ y ≤ √x, with the Fréchet family uniquely tracing the upper curve and ordinal sums of the independence copula tracing the diagonal.","keywords":["Chatterjee's rank correlation","Spearman's footrule","copula","stochastically increasing","attainable region","Fréchet copula","Markov product","ordinal sum"],"falsifier":"Take any copula $C$ and compute, on a fine grid of $v\\in[0,1]$, the diagonal difference $D(v)=C(v,v)-(C^{\\top}\\ast C)(v,v)$. If $C$ is stochastically increasing and $D(v)<0$ for even one $v$, then $\\xi(C)>\\psi(C)$ and the lower boundary of $R_{\\mathrm{SI}}$ fails; if for any copula $\\psi(C)>\\sqrt{\\xi(C)}$, the upper-bound theorem fails. The theorems predict no such point exists and that the Fréchet copula is the unique maximizer at each level, so a single numerical counterexample would settle the claim.","tokens_in":17726,"feed_emoji":"📈","tokens_out":11768,"duration_ms":99657,"temperature":0.7,"pith_summary":"Chatterjee's rank correlation $\\xi$ measures how close $Y$ is to a function of $X$, and Spearman's footrule $\\psi$ is a classical rank-association measure; this paper asks which pairs $(\\xi,\\psi)$ a bivariate copula can produce. It proves that $\\psi$ can never exceed $\\sqrt{\\xi}$, and for any fixed $\\xi=x$ the ceiling is reached only by the Fréchet copula with parameter $\\alpha=\\sqrt{x}$. For the subclass of stochastically increasing copulas—where larger values of one variable make the other stochastically larger—the paper obtains the exact region $x\\le y\\le\\sqrt{x}$: every point is attainable, the diagonal comes from ordinal sums of the independence copula, and the upper curve comes only from the Fréchet family. The full copula region is shown to be convex and closed, with a Jensen-based lower bound and a two-parameter family approaching it, although the exact lower boundary remains open. If correct, these results give a sharp geometric description of how much rank association can accompany a given degree of directed dependence.","feed_headline":"Chatterjee's ξ and Spearman's footrule obey ψ ≤ √ξ","feed_subtitle":"For stochastically increasing copulas, the pair (ξ, ψ) fills exactly the band ξ ≤ ψ ≤ √ξ.","key_machinery":"The argument runs through three linked devices. First, the Markov-product identity $\\xi(C)=\\psi(C^{\\top}\\ast C)$, where $C^{\\top}(u,v)=C(v,u)$ and $(C_1\\ast C_2)(u,v)$ is the Markov product; this rewrites Chatterjee's coefficient as Spearman's footrule of a derived copula. Second, Lemma 1.1 characterizes a copula by its partial derivative $h_v(t)=\\partial_1 C(t,v)$, subject to $\\int_0^1 h_v(t)\\,dt=v$ and monotonicity in $v$; in terms of $h$, $\\psi$ is a linear functional while $\\xi$ is a quadratic functional, turning the region problem into convex optimization over square-integrable functions. Third, for stochastically increasing copulas the diagonal inequality $(C^{\\top}\\ast C)(v,v)\\le C(v,v)$ yields $\\xi\\le\\psi$; the upper boundary then comes from a KKT analysis of the convex program, and the lower boundary of the SI region comes from this diagonal inequality together with continuity of $\\xi$ along the convex segment between an ordinal-sum copula and a Fréchet copula.","core_discovery":"The paper's central claim is Theorem 2.4: the attainable $(\\xi,\\psi)$-region for stochastically increasing copulas is exactly $R_{\\mathrm{SI}}=\\{(x,y)\\in[0,1]^2\\mid x\\le y\\le \\sqrt{x}\\}$. The lower edge $x=y$ is attained by ordinal sums of the independence copula, and the upper edge $y=\\sqrt{x}$ is traced uniquely by the Fréchet copula family $C^{\\mathrm{Fr}}_\\alpha=(1-\\alpha)\\Pi+\\alpha M$, $\\alpha\\in[0,1]$. This upper edge is not special to the SI class: Theorem 2.1 states that for every copula $C$, $\\psi(C)\\le\\sqrt{\\xi(C)}$, and for fixed $\\xi=x$ the only maximizer is the Fréchet copula with $\\alpha=\\sqrt{x}$. The paper also characterizes equality $\\xi=\\psi$ inside the SI class by a two-point form of the conditional density, and for the full copula region it proves convexity and closedness, gives a Jensen-derived lower boundary curve, and shows that at $\\psi=-1/2$ the smallest possible $\\xi$ is $1/2$, attained only by the $2\\times2$ checkerboard copula with mass on the off-diagonal squares.","pith_inferences":["The paper leaves implicit that the upper ceiling $\\psi\\le\\sqrt{\\xi}$ is class-independent: any convex family of copulas containing the Fréchet family will share the same square-root upper boundary, so only the lower boundary distinguishes one subclass from another.","Because the Jensen lower-bound curve is generated by functions that are not true copulas, the exact lower boundary of the full region remains open; a concrete next target is to identify whether the two-parameter family described in Section 3.2, or some refinement of it, actually attains it.","Proposition 2.2 suggests a quantitative stability version: for an SI copula, the gap $\\xi-\\psi$ should measure the distance from the two-point conditional-density form, which could be turned into a diagnostic for near-functional dependence.","Remark 2.6(c) identifies the attainable region for lower semilinear copulas with the SI region; if the same convexity argument applies to other subclasses that contain the Fréchet family and are closed under mixtures, exact regions for those classes would follow from their equality cases alone."],"forward_implications":["Knowing $\\xi$ for an SI copula determines $\\psi$ up to the band $\\xi\\le\\psi\\le\\sqrt{\\xi}$, and both endpoints of the band are attainable at every level.","For any copula whatsoever, $\\psi\\le\\sqrt{\\xi}$ with a unique extremizer at each level: only the Fréchet copula with parameter $\\sqrt{x}$ reaches $\\psi=\\sqrt{x}$ when $\\xi=x$.","Equality $\\xi=\\psi$ is possible at every value from 0 to 1, realized by ordinal sums of the independence copula, so the two measures coincide on a full diagonal of structures.","The corollary $\\xi\\le\\frac{3}{4}\\tau+\\frac{1}{4}$ for SI copulas connects the square-root band to Kendall's tau.","The full attainable region is convex and closed, so mixtures of copulas with the same $\\psi$ fill out intervals of possible $\\xi$ values between the two endpoint copulas."],"supporting_citations":[{"why":"Define Chatterjee's rank correlation as the integral of the squared partial derivative of the copula, the objective being optimized.","marker":"[7, 9]"},{"why":"Establishes that the Markov product of two copulas is again a copula, which makes the identity $\\xi(C)=\\psi(C^{\\top}\\ast C)$ legitimate.","marker":"[8]"},{"why":"Supplies the characterization of copulas by conditional densities and the convex-optimization technique used for both the upper boundary and the lower-bound estimate.","marker":"[4]"},{"why":"Provides the Banach-space KKT and quadratic-growth lemmas that certify the proposed optimizers.","marker":"[5]"},{"why":"Gives the closed-form value of $\\xi$ for Fréchet copulas, used to verify that the upper-bound candidate has $\\xi=x$.","marker":"[12]"},{"why":"Supplies the diagonal inequality for Markov products of stochastically increasing copulas that yields $\\xi\\le\\psi$ and the SI-region lower boundary.","marker":"[24]"},{"why":"Provides the Kendall-tau bound applied to $\\xi\\le\\psi$ in Corollary 2.5.","marker":"[17]"},{"why":"Characterizes copulas with $\\psi=-1/2$ as being supported on off-diagonal $2\\times2$ squares, the starting point for Theorem 3.4.","marker":"[13]"},{"why":"Supplies the checkerboard lower bound on $\\xi$ used to prove that the $2\\times2$ checkerboard copula uniquely minimizes $\\xi$ at $\\psi=-1/2$.","marker":"[23]"}],"fun_headline_variants":["Exact region: ξ ≤ ψ ≤ √ξ for stochastically increasing","ψ ≤ √ξ always; equality only for Fréchet copulas","Exact band for (ξ, ψ): ξ ≤ ψ ≤ √ξ","For SI copulas, (ξ, ψ) fills ξ ≤ ψ ≤ √ξ exactly"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the known diagonal inequality $(C^{\\top}\\ast C)(v,v)\\le C(v,v)$ for stochastically increasing copulas; the full-region lower bound additionally rests on a rearrangement lemma from the author's companion preprint, so either failure would collapse the corresponding boundary.","fun_headline_variants_meta":{"raw":{"variants":["Exact region: ξ ≤ ψ ≤ √ξ for stochastically increasing","ψ ≤ √ξ always; equality only for Fréchet copulas","Exact band for (ξ, ψ): ξ ≤ ψ ≤ √ξ","For SI copulas, (ξ, ψ) fills ξ ≤ ψ ≤ √ξ exactly"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001167,"raw_usage":{"total_tokens":4868,"prompt_tokens":1024,"completion_tokens":3844,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":640,"completion_tokens_details":{"reasoning_tokens":3759}},"tokens_in":640,"tokens_out":3844,"duration_ms":25067,"temperature":1.0,"reasoning_tokens":3759,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:14:23.213956+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take any copula $C$ and compute, on a fine grid of $v\\in[0,1]$, the diagonal difference $D(v)=C(v,v)-(C^{\\top}\\ast C)(v,v)$. If $C$ is stochastically increasing and $D(v)<0$ for even one $v$, then $\\xi(C)>\\psi(C)$ and the lower boundary of $R_{\\mathrm{SI}}$ fails; if for any copula $\\psi(C)>\\sqrt{\\xi(C)}$, the upper-bound theorem fails. The theorems predict no such point exists and that the Fréchet copula is the unique maximizer at each level, so a single numerical counterexample would settle the claim.","supporting_citations":[{"cited_title":"Quantifying directed dependence via dimension reduction.J","cited_arxiv_id":null,"evidence_quote":"Gives the closed-form value of $\\xi$ for Fréchet copulas, used to verify that the upper-bound candidate has $\\xi=x$."},{"cited_title":"Darsow, Bao Nguyen, and Elwood T","cited_arxiv_id":null,"evidence_quote":"Establishes that the Markov product of two copulas is again a copula, which makes the identity $\\xi(C)=\\psi(C^{\\top}\\ast C)$ legitimate."},{"cited_title":"The exact region and an inequality between Chatterjee's and Spearman's rank correlations","cited_arxiv_id":"2506.15897","evidence_quote":"Supplies the characterization of copulas by conditional densities and the convex-optimization technique used for both the upper boundary and the lower-bound estimate."},{"cited_title":"Fr´ ed´ eric Bonnans and Alexander Shapiro.Perturbation Analysis of Optimization Problems","cited_arxiv_id":null,"evidence_quote":"Provides the Banach-space KKT and quadratic-growth lemmas that certify the proposed optimizers."},{"cited_title":"Stochastic monotonicity and the Markov product for copulas.J","cited_arxiv_id":null,"evidence_quote":"Supplies the diagonal inequality for Markov products of stochastically increasing copulas that yields $\\xi\\le\\psi$ and the SI-region lower boundary."},{"cited_title":"On the exact regions determined by Kendall’s tau and other concordance measures.Mediterr","cited_arxiv_id":null,"evidence_quote":"Provides the Kendall-tau bound applied to $\\xi\\le\\psi$ in Corollary 2.5."},{"cited_title":"On the lower bound of Spearman’s footrule.Depend","cited_arxiv_id":null,"evidence_quote":"Characterizes copulas with $\\psi=-1/2$ as being supported on off-diagonal $2\\times2$ squares, the starting point for Theorem 3.4."}],"review_version":2}