REVIEW 2 major objections 5 minor 14 references
Implicit max-stable extremal integrals
T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read This paper extends the f-implicit extremal integral from simple functions to every nonnegative function with finite L^α norm, proving the limit is almost surely independent of the approximating sequence.
desk verdict Genuine and mostly well-built extension of extremal integrals to the f-implicit setting, but the central theorem leans heavily on an unpublished thesis and one lemma has a false degenerate case. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing machinery is the gap behind the attained ∨_f-maximum, quantified in Lemma 3.3: when independent summands have α-Fréchet f-values, the probability that a given component wins but another component comes within a factor 1+γ of it is at most 1-(1+γ)^{-α}. This gap, combined with Lemma 3.7, which turns a uniform gap f(ζ) ≥ (1+δ)f(ζ*) into Lipschitz-type control of the ∨_f-combination under small coefficient perturbations, lets the paper show that approximating integrals form a Cauchy sequence in probability despite the ∨_f operation being neither commutative nor continuous. Egorov's theorem is then used to reduce pointwise convergence of simple integrands to uniform convergence on sets of almost full measure, and a consistency condition on partitions keeps the successive approximations comparable.
What would settle it
Take E=[0,1] with Lebesgue measure, f(x)=|x| on R, α=1, and g≡1; approximate g by two different dyadic step-function sequences and check whether the empirical distributions of the integrals coincide with $Φ^{1}$_{1,ε_1}(1) and do not depend on the approximation. A scheme-dependent limit, or a limit whose f-value is not α-Fréchet with scale 1, would refute Theorem 3.11 and Proposition 4.2.
Extended reading notes
Core claim
The central claim is Theorem 3.11: for a fixed f-implicit α-Fréchet sup-measure M on a σ-finite measure space (E,E,m), the class of M-integrable nonnegative functions is exactly L^α_+(m), and for every g in that class, any sequence of simple functions g_n ↑ g yields a probability limit I(g) that is almost surely independent of the sequence. The resulting map g ↦ I(g) satisfies f(I(g)) ∼ Φ_α(||g||_α), the max-linearity identity I(a g_1 ∨ b g_2) = a I(g_1) ∨_f b I(g_2), independence of I(g_1) and I(g_2) exactly when g_1 g_2 = 0 m-a.e., and monotonicity in the ≤_f order. The paper also proves a convergence theorem identifying convergence in probability of I(g_n) with L^α-convergence of the integrands. The author presents this as solving the open problem from [7] of extending the simple-function integral, and notes that the classical extremal integrals from [14] are contained as a special case.
Load-bearing premise
The argument inherits from the unpublished thesis [7] the existence of f-implicit α-Fréchet sup-measures and the basic properties of the integral for simple functions; if any of those borrowed results is false or inaccessible, the L^α extension and its properties collapse.
Editorial extensions
If this is right
- Every function g with ∫ g^α dm < ∞ is M-integrable, and the value I(g) is almost surely the same for every increasing simple-function approximation; the integral is therefore a well-defined map on L^α_+(m).
- For each integrand g, I(g) is f-implicit α-Fréchet with scale ||g||_α, so f(I(g)) follows a univariate α-Fréchet law with that scale.
- The integral is f-implicit max-linear: I(a g_1 ∨ b g_2) = a I(g_1) ∨_f b I(g_2) almost surely, and I(g_1) and I(g_2) are independent precisely when g_1 g_2 = 0 m-a.e.
- For any family of integrands (g_t), the process X(t) = I(g_t) is f-implicit max-stable, with finite-dimensional ∨_f-combinations having the f-implicit α-Fréchet law with scale ||∨_j a_j g_{t_j}||_α.
- Taking f = |·| on R, the construction reduces to the classical extremal integral of [14], reproducing its max-linearity and convergence theorems.
Reading between the lines
- The gap estimates that make the proof work do not use the particular form of the loss function beyond continuity, zero-set, and 1-homogeneity; the same Cauchy-in-probability argument should therefore survive small perturbations of f, suggesting a stability result the paper does not state.
- Because the integral is linear in the ∨_f sense and independent of approximation scheme, it gives a representation toolkit for f-implicit max-stable processes; a natural next step the paper leaves open is a spectral representation converse, showing every such process can be written as I(g_t) for some sup-measure.
- Theorem 4.3 suggests that L^α-convergence of integrands is the correct topology for the integral; one could test whether the map g ↦ I(g) is continuous in the stronger sense of convergence in probability with respect to the L^α-norm, extending the stated equivalence.
- Remark 4.6 mentions signed or matrix-valued integrands; the difficulties there hint that a ∨_f-based definition using positive and negative parts, rather than subtraction, is the more coherent extension, and the B-homogeneous case would need new estimates because Lemma 3.7 fails.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper constructs an f-implicit extremal integral for nonnegative deterministic functions with respect to an f-implicit alpha-Frechet sup-measure M. The main result, Theorem 3.11, asserts that the class of M-integrable functions is exactly L^alpha_+(m), that for any g in this class the integral is obtained as the probability limit of I(g_n) for simple g_n increasing to g, and that this limit is almost surely independent of the approximating sequence. Section 4 then derives the main calculus properties: f-implicit alpha-Frechet marginals with scale ||g||_alpha, max-linearity, the independence criterion g1 g2 = 0 m-a.e., monotonicity, and a convergence theorem characterizing convergence of integrals by L1-convergence of g^alpha. The paper also recovers the classical max-stable extremal integral of Stoev and Taqqu as a special case and constructs f-implicit max-stable processes from integrals.
Significance. Assuming the quoted results from [7] are valid, the paper delivers a complete integral calculus for f-implicit sup-measures, solving the open extension problem raised in [7] and closely paralleling the Stoev-Taqqu theory. The proof of Theorem 3.11 is the technical heart of the paper and is carried out in detail through the gap lemmas, Proposition 3.6, Proposition 3.9, and Lemma 3.10; this is an original and non-obvious technique. The paper also gives a clean statement of useful properties (Proposition 4.2) and an application to f-implicit max-stable processes (Proposition 4.9). The main caveat is that several foundational inputs are quoted from an apparently unpublished thesis, so the significance is conditional on the correctness and accessibility of that source.
major comments (2)
- [Section 2, Proposition 2.3; Section 3, Theorem 3.11] The central theorem is conditional on several results from [7] that are not proved in the paper and apparently are not publicly available: Proposition 2.3 (existence of the f-implicit sup-measure M, cited as Theorem 3.1.12 of [7]), Proposition 3.2.3/3.2.4 and Lemma 3.1.14 of [7] (unique representation, distribution and monotonicity of the simple-function integral, and the implication f(X_n - Y_n) -> 0 implies X_n - Y_n -> 0). These are used at load-bearing steps in Proposition 3.6, Proposition 3.9, Lemma 3.10, Lemma 4.1, Lemma 4.5, and parts of Proposition 4.2. Since [7] is cited as a PhD thesis with no year and no URL, a reader cannot currently certify these foundations. The author should either provide the missing proofs in an appendix or point to a freely accessible version of [7] where all quoted results appear.
- [Section 3, Lemma 3.3] Lemma 3.3 is false as stated. If all sigma_j = 0, then each f(X_j) = 0 a.s. and, because f(x) = 0 only for x = 0, each X_j = 0 a.s.; hence f(vee_f X_j) = 0 = f(vee^*_f X_j) a.s., so the left-hand probability is 1, while the right-hand side 1 - (1+gamma)^(-alpha) is strictly less than 1. The proof begins 'Without loss of generality we can assume that sigma_j > 0', which is not a valid reduction. This lemma is used in the proof of Proposition 3.6; in that application the all-zero case is excluded by the positivity of the limit Y^*, so the argument is likely repairable, but the statement and proof need a corrected hypothesis (for example, that not all sigma_j are zero) or a separate treatment of the degenerate case.
minor comments (5)
- [Section 3, Lemma 3.7] In the proof of Lemma 3.7, the displayed line 'the assertion would follow if we could show that f(beta_1 x_1) vee_f ... vee_f f(beta_k x_k) = beta_{j0} x_{j0}' should read 'beta_1 x_1 vee_f ... vee_f beta_k x_k = beta_{j0} x_{j0}', since the operation vee_f is defined on R^d, not on R.
- [Section 4, Proposition 4.2] In the proof of Proposition 4.2(ii), the decomposition near the end writes 'g_i = g_i 1_{g1 <= g2} vee 1_{g1 > g2}'; the second indicator should carry the factor g_i, i.e. the term should be g_i 1_{g1 > g2}.
- [Section 4, Theorem 4.3] The proof of the first implication in Theorem 4.3 is not self-contained: it says 'we can mostly follow the proof of Theorem 2.1 in [14]' and 'The details are left to the reader.' Since this is a stated theorem, either provide the full argument or state the reduction to Theorem 2.1 of [14] more explicitly, with the modifications needed in the f-implicit setting.
- [References] Reference [7] is a PhD thesis with no year and no repository or URL; please add this information so that readers can verify the results quoted from it.
- [Section 3, Lemma 3.10] The proof of Lemma 3.10 switches between E_k and E_l in equations (3.18) and (3.19); the subscripts should be consistent. The typesetting of the norms in (3.19) also has an extra alpha that should be cleaned up.
Circularity Check
No circularity: the extension result is a genuine derivation from definitions, external foundations, and prior results; the main theorem does not reduce to its inputs.
full rationale
The central claim (Theorem 3.11) is not obtained by fitting or renaming: it defines the integral first for simple functions and then proves, via Cauchy-in-probability and Egorov arguments (Propositions 3.6, 3.9, Lemma 3.10), that the limit for gn↑g exists and is independent of the approximating sequence. The properties in Proposition 4.2 are proved from this limit, not assumed. The paper does rely on external results: Proposition 2.3 and the simple-function calculus (Proposition 3.2.4, Lemma 3.1.14) come from Goldbach's thesis [7], and the scalar comparison results come from Stoev-Taqqu [14]. That reliance is a verifiability and accessibility issue, since [7] is unpublished and the paper does not re-prove those foundations, but it is not circularity, because those results are inputs rather than the theorem being derived. The only self-citation, [9], appears in historical context ("See [9], [11] and [12]", Section 2) and is not load-bearing for Theorem 3.11 or Proposition 4.2. No equation in the paper is shown to be equivalent by construction to a fitted parameter or to a self-cited uniqueness claim.
Assumptions & free parameters
assumptions (4)
- domain assumption The loss function f: R^d→R_+ is continuous, vanishes only at 0, and is 1-homogeneous (properties (i)-(iii), Section 1).
- domain assumption Existence of f-implicit α-Fréchet sup-measures M with M(A) ~ Φ^f_{α,κ}(m(A)^{1/α}) (Proposition 2.3, attributed to [7]).
- domain assumption Properties of the simple-function integral I(g) from [7]: well-definedness (Prop. 3.2.3), f(I(g)) ~ Φ_α(||g||_α) (Prop. 3.2.4), and Lemma 3.1.14 that f(x_n)→0 implies x_n→0.
- standard math Standard measure-theoretic tools: Egorov's theorem, dominated and monotone convergence, and the convergence-in-probability Cauchy criterion (Corollary 6.15 in [8]).
Cite this review
Pith. "Pith review of Implicit max-stable extremal integrals." pith.science (2026). https://pith.science/paper/MHDA44NJ
@misc{pith2026190806840,
author = {Pith},
title = {Pith review of: Implicit max-stable extremal integrals},
year = {2026},
howpublished = {\url{https://pith.science/paper/MHDA44NJ}},
note = {Machine review of arXiv:1908.06840}
}
abstract
Recently, the notion of implicit extreme value distributions has been established, which is based on a given loss function $f \ge 0$. From an application point of view, one is rather interested in extreme loss events that occur relative to $f$ than in the corresponding extreme values itself. In this context, so-called $f$-implicit $\alpha$-Fr\'{e}chet max-stable distributions arise and have been used to construct independently scattered sup-measures that possess such margins. In this paper we solve an open problem in [7] by developing a stochastic integral of a deterministic function $g\ge 0$ with respect to implicit max-stable sup-measures. The resulting theory covers the construction of max-stable extremal integrals (see [14]) and, at the same time, reveals striking parallels.
Reference graph
Works this paper leans on
- [7]
-
[1]
H. Bierm´ e, M. M. Meerschaert, and H.-P. Scheffler. Operator s caling stable random fields. Stochastic Processes and their Applications , 117(3):312–332, 2007
work page 2007
- [2]
-
[3]
R. de Fondeville and A. C. Davison. High-dimensional peaks-over- threshold inference. Biometrika, 105(3):575–592, 2018
work page 2018
-
[4]
L. De Haan and A. Ferreira. Extreme value theory: an introduction . Springer Science & Business Media, 2007
work page 2007
-
[5]
C. Dombry and M. Ribatet. Functional regular variations, paret o processes and peaks over threshold. Statistics and its Interface , 8(1):9–17, 2015
work page 2015
- [6]
-
[8]
A. Klenke. Wahrscheinlichkeitstheorie. Springer, 2006
work page 2006
Show all 14 references
-
[9]
Kremer and H.-P
D. Kremer and H.-P. Scheffler. Multivariate stochastic integrals w ith respect to independently scattered random measures on δ-rings. Publicationes Mathematicae Debrecen, 2019. Accepted
2019
-
[10]
Li and Y
Y. Li and Y. Xiao. Multivariate operator-self-similar random field s. Stochastic Processes and their Ap- plications, 121(6):1178–1200, 2011
2011
-
[11]
B. S. Rajput and J. Rosinski. Spectral representations of infi nitely divisible processes. Probability Theory and Related Fields , 82(3):451–487, 1989
1989
-
[12]
Samoradnitsky and M
G. Samoradnitsky and M. S. Taqqu. Stable non-Gaussian random processes: stochastic models w ith infinite variance . CRC press, 1994
1994
-
[13]
Scheffler and S
H.-P. Scheffler and S. Stoev. Implicit extremes and implicit max–st able laws. Extremes, 20(2):265–299, 2017
2017
-
[14]
S. A. Stoev and M. S. Taqqu. Extremal stochastic integrals: a parallel between max-stable processes and α -stable processes. Extremes, 8(4):237–266, 2005. 25 Dustin Kremer, Department Mathematik, Universit ¨at Siegen, 57068 Siegen, Germany E-mail address : kremer@mathematik.un...
2005
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.