{"id":"2bcb852e-82e5-4bc6-a5a3-a95a2072590e","arxiv_id":"1908.00584","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Fractional Weibull, Fréchet, and Gumbel distributions are characterized as unique solutions to fractional differential equations, with representations via α-stable subordinators and new analytic properties for the associated special functions.","lead":"The paper defines fractional versions of the three classical extreme value distributions, solving fractional differential equations with unique distribution functions and explicit stochastic representations in terms of α-stable subordinators. It then uses these representations to prove new analytic properties of Kilbas-Saigo and Le Roy special functions, including a complete monotonicity result that settles an open question.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Appendix A.1.3 defines D^α_+ with a minus sign, but Theorem 1.3's proof and the α=1 limit require D^α_+ = d/dx I^{1-α}_+; the claimed solution fails (1.9) under the paper's own definition.","rationale":"The reader's weakest assumption concerns reliance on external identities from [25]. Those identities are load-bearing, but the paper explicitly notes the required condition q=ρ-α>-α (Remark 3.1(c) and Remark 3.3(a)), and they come from a published ESAIM paper by the same group; independent re-derivation would be desirable but there is no concrete indication of failure. By contrast, the definition of D^α_+ in Appendix A.1.3 is internally inconsistent with the proof of Theorem 1.3 and with the stated α=1 boundary case. A direct calculation shows the claimed solution satisfies (1.9) only if D^α_+ = d/dx I^{1-α}_+, not -d/dx I^{1-α}_+. This is a concrete, verifiable inconsistency in one of the three central theorems, affecting every result that uses D^α_+. It is easily corrected by removing the minus sign in A.1.3, but as written the paper cannot be read unambiguously. Secondary omissions in Propositions 4.7, 4.15, and 4.17 also support a conditional verdict. I therefore keep the reader's CONDITIONAL verdict unchanged, while identifying a different, more immediate load-bearing concern.","tokens_in":35940,"tokens_out":25941,"duration_ms":219141,"concrete_test":"Compute D^α_+F for F=1-L_α(-e^{λx}) using the literal definition in Appendix A.1.3. If the result is -λ^α e^{λx}\\bar F, then (1.9) is not solved. Recompute with the convention D^α_+ = d/dx I^{1-α}_+, which should give (1.9) and reduce to F'=λ e^{λx}\\bar F at α=1. This single check settles whether the operator definition or the theorem statement needs correction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Appendix A.1.3 defines the progressive Liouville derivative on R as D^α_+ f = -d/dx(I^{1-α}_+ f). The proof of Theorem 1.3 sets \\bar F(x)=L_α(-e^{λx}) and asserts D^α_+ F = λ^α e^{λx}\\bar F. Computing directly with F=1-\\bar F gives I^{1-α}_+F = Σ_{n≥1} (-1)^{n+1}(λn)^{α-1}e^{λnx}/(n!)^α, hence d/dx I^{1-α}_+F = λ^α e^{λx}\\bar F. Thus under A.1.3 one obtains D^α_+F = -λ^α e^{λx}\\bar F, contradicting (1.9). The intended identity holds only with the opposite sign, D^α_+ = d/dx I^{1-α}_+, which is also the convention consistent with the boundary case α=1 (F'=λ e^{λx}\\bar F). This internal sign inconsistency makes Theorem 1.3, one of the three central existence theorems, ill-posed as written: a reader following the appendix definition cannot verify the stated equation, and the existence proof is not compatible with the operator definition. This is more load-bearing than the cited identities from [25], because it is an internal inconsistency rather than a matter of external verification, and it directly affects the Gumbel case of the main claim.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces fractional analogues of the three classical extreme value distributions. For α ∈ [0,1], λ,ρ > 0, Theorems 1.1, 1.2, and 1.3 assert that the fractional differential equations (1.5), (1.7), and (1.9) each have a unique distribution function, and that the corresponding random variables admit explicit stochastic representations involving an independent α-stable subordinator and an exponential variable. The proofs combine uniqueness via alternating-series fixed-point arguments (Section 2), existence via Kilbas-Saigo and Le Roy functions and their moment determinacy (Section 3), and a substantial body of analytic consequences (Section 4): complete monotonicity, Mellin transforms and density asymptotics, optimal hyperbolic bounds, infinite divisibility properties, and new asymptotics for the Le Roy function. The appendix fixes the conventions for fractional integrals and derivatives and collects the needed properties of Barnes' double Gamma function.","tokens_in":36263,"tokens_out":7284,"duration_ms":66819,"significance":"If the main theorems are correct, the paper gives a coherent and elegant family of 'fractional extreme' laws interpolating between the classical Weibull, Fréchet, and Gumbel distributions and Pareto/logistic-type laws, with the stable subordinator playing a structurally natural role. The analytic spin-offs are also valuable: the complete monotonicity characterization of Kilbas-Saigo functions resolves a question raised in earlier literature, and the uniform hyperbolic bounds and Le Roy asymptotics are new and cleanly derived from stochastic representations. The paper's reliance on prior results from [25] for the beta-product identities is acceptable practice for a research article, though it means the stochastic representations are not self-contained. However, one load-bearing internal inconsistency in the definition of the progressive Liouville derivative on the line makes Theorem 1.3 ill-posed as written; this is a local sign error but it directly affects one of the three central existence theorems.","major_comments":[{"comment":"Appendix A.1.3 defines the progressive Liouville derivative on R by D^α_+ f = −d/dx(I^{1−α}_+ f), with the boundary case D^1_+ described as the usual derivative. Under this definition, the claimed identity D^α_+F(x) = λ^α e^{λx}\\bar F(x) in Theorem 1.3 is false. For \\bar F(x)=L_α(−e^{λx}) = Σ_{n≥0}(−1)^n e^{λnx}/(n!)^α and F=1−\\bar F, a direct computation gives I^{1−α}_+F = Σ_{n≥1}(−1)^{n+1}(λn)^{α−1}e^{λnx}/(n!)^α and hence d/dx I^{1−α}_+F = λ^α e^{λx}\\bar F(x). The appendix definition therefore yields D^α_+F = −λ^α e^{λx}\\bar F(x), contradicting Eq. (1.9). The proof of Theorem 1.3 is compatible only with the opposite sign convention, D^α_+ = +d/dx(I^{1−α}_+·), which is also the convention consistent with the classical α=1 case F'=e^x\\bar F. This is an internal inconsistency in a central existence theorem, not merely a typo; it needs a local but substantive correction, either in the sign of the operator in A.1.3 or in the sign of the right-hand side of (1.9).","section":"Appendix A.1.3 and Theorem 1.3 (§3.3, Eq. (1.9))"}],"minor_comments":[{"comment":"Several secondary results are stated with proofs that omit essential details: Proposition 4.7 says 'We omit details' for the residue computation of the density asymptotics at zero, and Propositions 4.15 and 4.17 leave the single-intersection stochastic-order verification to the reader. These are not blocking for the central existence claims, but for publication the omitted arguments should be supplied or replaced by precise references.","section":"Propositions 4.7, 4.15, 4.17"},{"comment":"The displayed formula for log E[e^{sG_{α,λ}}] contains a sign typo: the second term should be (1−α) log Γ(1−sλ^{-1}) rather than (1−α) log Γ(1+sλ^{-1}); the Lévy measure expression written immediately afterward is consistent with the corrected version.","section":"Proposition 4.21"},{"comment":"The phrase 'anxiolytic extension' in the sentence accompanying Eq. (4.20) is presumably a typo for 'analytic extension', since the intended meaning is that the right-hand side is understood analytically when z is a non-positive integer.","section":"Appendix A.2, Eq. (4.20)"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper introduces a genuinely new class of 'fractional extreme' distributions, proves existence and uniqueness for the associated fractional hazard equations, and uses the stochastic representations to settle an open question about complete monotonicity of E_{α,m,m-1}(-x). The main probabilistic content looks solid: Theorems 1.1 and 1.2 are proved in detail via infinite beta products, moment determinacy, and the homogeneous functional identities from [25]. I did not machine-check the algebra, but the chain is coherent and the transfer from moments to law is standard.\n\nWhere it gets shaky: Theorem 1.3, the Gumbel case, conflicts with the paper's own definition of D^α_+ in Appendix A.1.3. The appendix defines D^α_+ f = -d/dx(I^{1-α}_+ f) (the regressive convention), but the proof asserts D^α_+ F = λ^α e^{λx}\\bar F(x), which requires the opposite sign. The α=1 boundary case in the same appendix says D^1_+ is the usual derivative, which also contradicts the minus sign. So as written, a reader following the definitions cannot verify (1.9). The intended identity is almost certainly salvageable by changing the sign in the appendix—that is the standard progressive derivative on R—but it is a load-bearing inconsistency, not a typo in a footnote, because it affects one of the three central theorems.\n\nOther weaknesses are minor. Several secondary propositions (4.7, 4.15, 4.17) say 'details omitted'; in a referee version at least the residue computation in 4.7 and the single-intersection claims in 4.15 and 4.17 should be written out or deferred to an appendix. The reliance on the cited identities from [25] is legitimate—they are prior published results with their own proofs—but the authors should state exactly which parameter ranges are needed, since ρ>0, α∈(0,1) is the whole range.\n\nBottom line: this is a serious paper with a real contribution and a fixable but central sign error. It deserves a careful referee, not a desk reject. If I were editing, I would send it out with a request that the authors correct the definition of D^α_+, reprove Theorem 1.3 under the corrected convention, and fill in the omitted details in those three propositions.","headline":"A new family of fractional extreme distributions with solid main theorems, undercut by a sign error in the definition of the progressive derivative that breaks Theorem 1.3 as written.","tokens_in":36859,"tokens_out":2880,"would_cite":true,"duration_ms":26810,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["26A33","33E12","45E10","60E05","60E15","60G52"],"pacs":[],"model":"deepseek-v4-flash","headline":"Fractional-order versions of the defining equations of Weibull, Fréchet, and Gumbel laws have unique solutions with explicit stable-subordinator representations.","keywords":["fractional differential equations","extreme value distributions","alpha-stable subordinator","Kilbas-Saigo function","Le Roy function","Mittag-Leffler function","complete monotonicity","infinite divisibility"],"falsifier":"Simulate an $\\alpha$-stable subordinator for a fixed $\\alpha\\in(0,1)$ and $\\rho>0$, compute the integral $\\int_0^\\infty((1-\\sigma_t^{(\\alpha)})_+)^{\\rho-\\alpha}dt$, and check whether the resulting survival function matches the series $E_{\\alpha,\\rho/\\alpha,\\rho/\\alpha-1}(-\\lambda x^\\rho)$ at several values of $x$; a mismatch at any single point would falsify the Weibull-type representation. Equivalently, test the Mellin transform identity $E[W_{\\alpha,\\lambda,\\rho}^s]=(\\rho\\alpha/\\lambda)^{s/\\rho}\\Gamma(1+s/\\rho)[\\rho+(1-\\alpha);\\rho]_{-s}/[\\rho;\\rho]_{-s}$ for $s\\in(-\\rho,\\rho)$, since the right-hand side must be analytic in that strip.","tokens_in":35693,"feed_emoji":"📈","tokens_out":15373,"duration_ms":140402,"temperature":0.7,"pith_summary":"This paper extends the three classical extreme-value distributions—Weibull, Fréchet, and Gumbel—to fractional order. For every order $\\alpha\\in[0,1]$ and positive parameters $\\lambda,\\rho$, the authors prove that the fractional differential equations obtained by replacing the ordinary derivative with a Liouville fractional derivative have exactly one distribution-function solution, and they identify the solving random variables as the classical extreme laws multiplied or shifted by functionals of an independent $\\alpha$-stable subordinator. The power-law examples are solved by Kilbas-Saigo functions in the Weibull and Fréchet cases and by Le Roy functions in the Gumbel case, and the stochastic representations yield new analytic properties of those special functions. The family traces a continuous arc from Pareto-type and logistic laws at $\\alpha=0$ to the classical extreme families at $\\alpha=1$.","feed_headline":"Fractional derivatives bridge Pareto and Weibull extreme laws","feed_subtitle":"A single order α takes the defining equations of Weibull, Fréchet, and Gumbel laws into Pareto and logistic laws.","key_machinery":"The load-bearing object is the infinite $\\beta$ product $T(a,b,c)=\\prod_{n\\ge0}\\frac{a+nb+c}{a+nb}B_{a+nb,c}$, together with the known identities that identify it, up to Gamma factors, with integral functionals of the $\\alpha$-stable subordinator; for example $\\int_0^\\infty((1-\\sigma_t^{(\\alpha)})_+)^{\\rho-\\alpha}dt$ has the law of $\\frac{\\Gamma(\\rho+1-\\alpha)}{\\Gamma(\\rho+1)}T(1,\\rho^{-1},(1-\\alpha)\\rho^{-1})$. These identities convert the formal series solutions of the fractional equations into explicit multiplicative or additive stochastic representations. The analytic counterparts are the Kilbas-Saigo functions $E_{\\alpha,m,l}$, three-parameter Mittag-Leffler-type series, and the Le Roy functions $L_\\alpha(z)=\\sum_{n\\ge0}z^n/(n!)^\\alpha$; moment determinacy via a classical convergence criterion and Mellin transforms carry the analytic consequences.","core_discovery":"The paper claims that each of the three classical extreme-value distributions can be deformed continuously through a fractional order $\\alpha\\in[0,1]$: replacing the ordinary derivative in their defining equations by a Liouville fractional derivative yields, for every $\\lambda,\\rho>0$, a unique distribution function. The solving laws are explicit. If $W_\\rho$, $F_\\rho$, and $G$ denote the usual Weibull, Fréchet, and Gumbel variables and $\\sigma^{(\\alpha)}$ an independent $\\alpha$-stable subordinator, then $W_{\\alpha,\\lambda,\\rho}=W_\\rho(\\lambda\\int_0^\\infty((1-\\sigma_t^{(\\alpha)})_+)^{\\rho-\\alpha}dt)^{-1/\\rho}$, $F_{\\alpha,\\lambda,\\rho}=F_\\rho(\\lambda\\int_0^\\infty(1+\\sigma_t^{(\\alpha)})^{-\\rho-\\alpha}dt)^{1/\\rho}$, and $G_{\\alpha,\\lambda}=\\lambda^{-1}(G-G_\\alpha)$ with $G_\\alpha=\\log\\int_0^\\infty e^{-\\sigma_t^{(\\alpha)}}dt$. At $\\alpha=1$ the equations reduce to the classical ones, while at $\\alpha=0$ the solutions become Pareto-type and logistic laws, so the family traces an arc between those distributions. The same explicit laws are then used to prove analytic properties of the Kilbas-Saigo and Le Roy special functions.","pith_inferences":["Inference: The single parameter $\\alpha$ could serve as a tail-interpolation index for real data; one could estimate it from the slope of the log-survival function or from ratios of sample moments, a statistical use the paper does not develop.","Inference: The convex-order and peacock structure observed for the underlying beta products may extend to the fractional extreme variables themselves, implying monotonicity in $\\alpha$ of quantiles or risk measures; this is not established in the paper.","Inference: The identity $W_{\\alpha,1,\\rho}\\,F_{1-\\alpha,\\rho,\\rho}^{-1}=(L_1L_2/L_3)^{1/\\rho}$ with independent unit exponentials $L_i$ suggests a duality $\\alpha\\leftrightarrow1-\\alpha$ in the family that could support symmetric estimation or model selection for the fractional order, a direction the paper leaves implicit."],"forward_implications":["For every $\\alpha\\in[0,1]$, $\\lambda,\\rho>0$, each of the three fractional equations has exactly one distribution-function solution, so the fractional-extreme family is a well-defined interpolation from Pareto and logistic laws at $\\alpha=0$ to the Weibull, Fréchet, and Gumbel laws at $\\alpha=1$.","The explicit representations yield Mellin transforms and exact tail asymptotics for the new laws, such as $f^W_{\\alpha,\\lambda,\\rho}(x)\\sim(\\rho/(\\lambda\\Gamma(1-\\alpha)))x^{-\\rho-1}$ as $x\\to\\infty$ and $f^F_{\\alpha,\\lambda,\\rho}(x)\\sim C\\,x^{\\rho+\\alpha-1}$ as $x\\to0$ for an explicit constant $C$.","The Kilbas-Saigo function $E_{\\alpha,m,m-1}(-x)$ is completely monotone for $\\alpha\\in(0,1]$ and $m>0$, with an explicit Bernstein representation; this settles a previously open question.","Optimal uniform hyperbolic bounds are obtained for the Kilbas-Saigo functions and for the generalized Mittag-Leffler functions $\\Gamma(\\beta)E_{\\alpha,\\beta}(-x)$, extending the classical two-sided bounds for $E_\\alpha(-x)$.","The fractional extreme laws are infinitely divisible in certain parameter regimes: $W_{\\alpha,\\lambda,\\rho}$ is a generalized Gamma convolution when $\\rho\\le1$, $W$ and $F$ are hyperbolically completely monotone when $\\rho\\le1-\\alpha$, and $G_{\\alpha,\\lambda}$ belongs to the extended Thorin class for all $\\alpha$."],"supporting_citations":[{"why":"Supplies the core law identities equating the infinite beta product with integral functionals of the α-stable subordinator; the stochastic representations in Theorems 1.1 and 1.2 rest on them.","marker":"[25]"},{"why":"Provides the fractional-calculus definitions, inversion formulas, and Mittag-Leffler function properties used to solve and verify the fractional equations.","marker":"[17]"},{"why":"Gives the generalized stable-law identification used to rewrite the Weibull-side integral functional and to derive density asymptotics.","marker":"[21]"},{"why":"Supplies the moment identity for the Gumbel-side exponential functional and related subordinator results used in the proof of Theorem 1.3.","marker":"[11]"},{"why":"Establishes the Mittag-Leffler survival representation that motivates the fractional equations and gives the classical limiting case.","marker":"[27]"},{"why":"Provides the classical hyperbolic bounds and convex-ordering method that the paper extends to Kilbas-Saigo functions.","marker":"[30]"},{"why":"Defines the classes of generalized Gamma convolutions and hyperbolically completely monotone laws used for the infinite divisibility results.","marker":"[9]"},{"why":"Supplies the asymptotic estimate for the relevant density that yields the logarithmic behaviour of the Le Roy function at large negative arguments.","marker":"[4]"}],"fun_headline_variants":["Fractional order deforms extreme laws into Pareto and logistic","One alpha bridges Weibull, Fréchet, Gumbel to Pareto and logistic","Stable subordinators reveal fractional extreme distributions","Fractional extreme laws unify classical distributions via alpha","Kilbas-Saigo and Le Roy functions arise from fractional extremes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction rests on previously established equalities in law between certain infinite products of $\\beta$-distributed random variables and integral functionals of an $\\alpha$-stable subordinator; if those equalities fail for the parameter ranges used here ($\\rho>0$, $\\alpha\\in(0,1)$), the explicit formulas for the fractional extreme laws and the analytic results built on them would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Fractional order deforms extreme laws into Pareto and logistic","One alpha bridges Weibull, Fréchet, Gumbel to Pareto and logistic","Stable subordinators reveal fractional extreme distributions","Fractional extreme laws unify classical distributions via alpha","Kilbas-Saigo and Le Roy functions arise from fractional extremes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000227,"raw_usage":{"total_tokens":1495,"prompt_tokens":993,"completion_tokens":502,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":415}},"tokens_in":609,"tokens_out":502,"duration_ms":5244,"temperature":1.0,"reasoning_tokens":415,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:45:36.783995+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Simulate an $\\alpha$-stable subordinator for a fixed $\\alpha\\in(0,1)$ and $\\rho>0$, compute the integral $\\int_0^\\infty((1-\\sigma_t^{(\\alpha)})_+)^{\\rho-\\alpha}dt$, and check whether the resulting survival function matches the series $E_{\\alpha,\\rho/\\alpha,\\rho/\\alpha-1}(-\\lambda x^\\rho)$ at several values of $x$; a mismatch at any single point would falsify the Weibull-type representation. Equivalently, test the Mellin transform identity $E[W_{\\alpha,\\lambda,\\rho}^s]=(\\rho\\alpha/\\lambda)^{s/\\rho}\\Gamma(1+s/\\rho)[\\rho+(1-\\alpha);\\rho]_{-s}/[\\rho;\\rho]_{-s}$ for $s\\in(-\\rho,\\rho)$, since the right-hand side must be analytic in that strip.","supporting_citations":[{"cited_title":"Letemplier and T","cited_arxiv_id":null,"evidence_quote":"Supplies the core law identities equating the infinite beta product with integral functionals of the α-stable subordinator; the stochastic representations in Theorems 1.1 and 1.2 rest on them."},{"cited_title":"Gorenflo, A","cited_arxiv_id":null,"evidence_quote":"Provides the fractional-calculus definitions, inversion formulas, and Mittag-Leffler function properties used to solve and verify the fractional equations."},{"cited_title":"Jedidi, T","cited_arxiv_id":null,"evidence_quote":"Gives the generalized stable-law identification used to rewrite the Weibull-side integral functional and to derive density asymptotics."},{"cited_title":"Carmona, F","cited_arxiv_id":null,"evidence_quote":"Supplies the moment identity for the Gumbel-side exponential functional and related subordinator results used in the proof of Theorem 1.3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the Mittag-Leffler survival representation that motivates the fractional equations and gives the classical limiting case."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the classical hyperbolic bounds and convex-ordering method that the paper extends to Kilbas-Saigo functions."},{"cited_title":"Bondesson","cited_arxiv_id":null,"evidence_quote":"Defines the classes of generalized Gamma convolutions and hyperbolically completely monotone laws used for the infinite divisibility results."},{"cited_title":"Berg and J","cited_arxiv_id":null,"evidence_quote":"Supplies the asymptotic estimate for the relevant density that yields the logarithmic behaviour of the Le Roy function at large negative arguments."}],"review_version":1}