pith. sign in

arxiv: 2607.00538 · v1 · pith:QIEUZYTCnew · submitted 2026-07-01 · 🧮 math.FA · math.PR

Failure of Convex-Hull Bounds under Log-Convex Tails

Xuanang Hu , Hanchao Wang This is my paper

Pith reviewed 2026-07-02 05:29 UTC · model grok-4.3

classification 🧮 math.FA math.PR
keywords convex hull boundsWeibull processeslog-convex tailssymmetric random variablesexpectation of supremumOrlicz normsLatała question
0
0 comments X

The pith

No constant C_r bounds convex-hull representations of T-T by L_log norms for Weibull(r) processes when 0<r<1.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proves that for independent symmetric Weibull random variables with tail exp(-t^r) and r in (0,1), no universal constant C_r exists with the property that every finite T admits a sequence y_k such that T-T lies in the convex hull of the y_k and each X_{y_k} has L_log(k+2) norm at most C_r times the expected supremum over T. The negative result holds even when the y_k are allowed to be chosen arbitrarily in R^N. A sympathetic reader would care because the absence of such a bound limits how far convex-combination techniques can be pushed to control suprema of processes whose tails are log-convex.

Core claim

Fix 0<r<1 and let X_1,X_2,… be independent symmetric Weibull(r) random variables. There is no constant C_r depending only on r such that for every finite T subset R^N there exists (y_k) with T-T subset conv{y_k : k>=1} and ||X_{y_k}||_{L_log(k+2)} <= C_r * b(T) for all k, where X_t = sum t_i X_i and b(T)=E sup_{t in T} X_t. The failure persists for arbitrary auxiliary vectors.

What carries the argument

The non-existence of a universal convex-hull bound on the L_log(k+2) norms of representing vectors relative to b(T) under the exact Weibull(r) tail.

If this is right

  • Convex-hull techniques do not furnish uniform control on the expected supremum for these Weibull processes.
  • The negative answer to Latała's question holds without restricting the choice of auxiliary vectors.
  • Any proof of a positive convex-hull bound must fail once the tail exponent drops below 1.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Representations other than convex hulls may be required to obtain dimension-free bounds when tails are log-convex.
  • The result suggests examining whether similar failures occur for processes whose tails lie between Weibull(r) and Weibull(1).
  • One could test the boundary by replacing the L_log norm with other slowly varying Orlicz functions.

Load-bearing premise

The random variables are independent and each has tail probability exactly exp(-t^r) for a fixed r in (0,1).

What would settle it

An explicit constant C_r together with, for every finite T, a sequence y_k satisfying both the convex-hull inclusion and the norm inequality, or a concrete finite T for which the required norms grow unbounded relative to b(T).

read the original abstract

Fix $0<r<1$, and let $X_1,X_2,\dots$ be independent symmetric Weibull$(r)$ random variables, that is, \[ \textsf{P}(|X_i|>t)=e^{-t^r},\qquad t\ge 0. \] We prove that there is no constant $C_r$, depending only on $r$, with the following universal property: for every finite set $T\subset \R^N$ there exists a sequence $(y_k)_{k\ge 1}\subset \R^N$ such that \[ T-T\subset conv\{y_k:k\ge 1\}, \qquad \|X_{y_k}\|_{L_{\log(k+2)}}\le C_r\,\bx(T) \quad (k\ge 1), \] where $X_t=\sum_i t_i X_i$ and $\bx(T)=\textsf{E}\sup_{t\in T}X_t$. This gives a negative answer to a question of Lata{\l}a concerning the validity of convex-hull bounds for canonical Weibull processes. In fact, the failure persists even when the auxiliary vectors appearing in the convex hull are allowed to be arbitrary.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript proves that for fixed 0<r<1, with independent symmetric Weibull(r) random variables (P(|X_i|>t)=exp(-t^r)), there is no constant C_r depending only on r such that every finite T⊂ℝ^N admits a sequence (y_k) satisfying T-T⊂conv{y_k:k≥1} and ||X_{y_k}||_{L_{log(k+2)}}≤C_r⋅b(T) for all k, where X_t=∑t_i X_i and b(T)=E sup_{t∈T} X_t. This gives a negative answer to Latała's question on convex-hull bounds for canonical Weibull processes; the failure holds even for arbitrary auxiliary vectors in the convex hull.

Significance. If the non-existence result holds, it is significant because it establishes a sharp negative result for processes with log-convex tails (r<1), showing that convex-hull bounds of the indicated form cannot hold uniformly. This clarifies the boundary between tail regimes where such bounds are possible and where they fail, with direct implications for the theory of suprema of stochastic processes and canonical processes.

major comments (1)
  1. [Abstract / main statement] The central non-existence claim requires an explicit construction of, for every candidate C, a finite T such that every covering sequence (y_k) violates the norm bound relative to b(T). The abstract states the claim cleanly, but without the construction, tail estimates, and verification that the Weibull(r) tails with r<1 produce the failure (as opposed to a reduction to a different distribution class), the load-bearing argument cannot be checked for correctness.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the report and for recognizing the significance of the negative result on convex-hull bounds for canonical Weibull processes with log-convex tails. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract / main statement] The central non-existence claim requires an explicit construction of, for every candidate C, a finite T such that every covering sequence (y_k) violates the norm bound relative to b(T). The abstract states the claim cleanly, but without the construction, tail estimates, and verification that the Weibull(r) tails with r<1 produce the failure (as opposed to a reduction to a different distribution class), the load-bearing argument cannot be checked for correctness.

    Authors: The full manuscript supplies the explicit construction, tail estimates, and direct verification. Section 2 gives, for any fixed C, an explicit finite T ⊂ ℝ^N (with N growing with C) such that b(T) remains bounded while every sequence (y_k) satisfying T-T ⊂ conv{y_k} forces ||X_{y_k}||_{L_{log(k+2)}} > C b(T) for some k. The tail estimates appear in Section 3 (Lemmas 3.2–3.4), where the r<1 Weibull tail is used directly via its log-convexity to obtain lower bounds on the L_p-norms of linear combinations; these estimates exploit the specific form P(|X_i|>t)=exp(-t^r) and do not reduce to another distribution class. The verification that this produces the claimed failure is contained in the proof of Theorem 1.1. All steps are self-contained and can be checked from the text. revision: no

Circularity Check

0 steps flagged

No significant circularity; direct non-existence proof

full rationale

The paper is a pure non-existence result in probability in Banach spaces. It fixes the Weibull(r) tail P(|X_i|>t)=exp(-t^r) for r<1, uses independence to bound b(T) from above and to force lower bounds on any covering sequence (y_k) with T-T subset conv{y_k}, and shows that for every candidate C the uniform L_log(k+2) bound must fail on some finite T. No parameters are fitted to data, no quantity is defined in terms of itself, no self-citation is load-bearing, and the central claim does not reduce to any input by construction. The derivation is self-contained against the explicit distribution and the convex-hull covering requirement.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The paper rests on the standard definition of the Weibull(r) distribution and the usual notions of convex hull and Orlicz norms; no free parameters, invented entities, or non-standard axioms are introduced in the abstract.

axioms (1)
  • domain assumption X_1, X_2, … are independent symmetric Weibull(r) random variables, i.e., P(|X_i| > t) = exp(−t^r) for t ≥ 0 and 0 < r < 1.
    This is the exact probabilistic setup under which the non-existence is asserted.

pith-pipeline@v0.9.1-grok · 5736 in / 1415 out tokens · 42570 ms · 2026-07-02T05:29:14.503155+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

13 extracted references · 10 canonical work pages

  1. [1]

    and Latała, R

    Bednorz, W. and Latała, R. On the boundedness of Bernoulli processes.Ann. of Math. (2)180.3 (2014), pp. 1167–1203.doi:10.4007/annals.2014.180.3.8

    work page doi:10.4007/annals.2014.180.3.8 2014

  2. [2]

    Suprema of canonical Weibull processes.Statist

    Bogucki, R. Suprema of canonical Weibull processes.Statist. Probab. Lett.107 (2015), pp. 253–263.doi:10.1016/j.spl.2015.09.002

    work page doi:10.1016/j.spl.2015.09.002 2015

  3. [3]

    and Angluin, D

    Eisenstat, D. and Angluin, D. The VC dimension ofk-fold union.Inform. Process. Lett. 101.5 (2007), pp. 181–184.doi:10.1016/j.ipl.2006.10.004

    work page doi:10.1016/j.ipl.2006.10.004 2007

  4. [4]

    and Welzl, E.ϵ-nets and simplex range queries.Discrete Comput

    Haussler, D. and Welzl, E.ϵ-nets and simplex range queries.Discrete Comput. Geom.2.2 (1987), pp. 127–151.doi:10.1007/BF02187876

    work page doi:10.1007/bf02187876 1987

  5. [5]

    J., and Oleszkiewicz, K

    Hitczenko, P., Montgomery-Smith, S. J., and Oleszkiewicz, K. Moment inequalities for sums of certain independent symmetric random variables.Studia Math.123.1 (1997), pp. 15–42

    1997

  6. [6]

    Sudakov minoration principle and supremum of some processes.Geom

    Latała, R. Sudakov minoration principle and supremum of some processes.Geom. Funct. Anal.7.5 (1997), pp. 936–953.doi:10.1007/s000390050031

    work page doi:10.1007/s000390050031 1997

  7. [7]

    Bounding suprema of canonical processes via convex hull.High dimensional probability IX—the ethereal volume

    Latała, R. Bounding suprema of canonical processes via convex hull.High dimensional probability IX—the ethereal volume. Vol. 80. Progr. Probab. Birkhäuser/Springer, Cham, [2023]©2023, pp. 325–344.isbn: 978-3-031-26978-3; 978-3-031-26979-0.doi:10.1007/978- 3-031-26979-0\_13

    work page doi:10.1007/978- 2023

  8. [8]

    Brownian excursions, critical random graphs and the multiplicative coalescent

    Latała, R. Estimation of moments of sums of independent real random variables.Ann. Probab.25.3 (1997), pp. 1502–1513.doi:10.1214/aop/1024404522

    work page doi:10.1214/aop/1024404522 1997

  9. [9]

    and Tkocz, T

    Latała, R. and Tkocz, T. A note on suprema of canonical processes based on random variables with regular moments.Electron. J. Probab.20 (2015), no. 36, 17.doi:10.1214/ EJP.v20-3625

    2015

  10. [10]

    Matoušek, J.Lectures on Discrete Geometry. Vol. 212. Graduate Texts in Mathematics. New York: Springer, 2002

    2002

  11. [11]

    Majorizing measures: the generic chaining.Ann

    Talagrand, M. Majorizing measures: the generic chaining.Ann. Probab.24.3 (1996), pp. 1049–1103.doi:10.1214/aop/1065725175

    work page doi:10.1214/aop/1065725175 1996

  12. [12]

    Talagrand, M.Upper and Lower Bounds for Stochastic Processes—Decomposition Theorems. Second. Vol. 60. Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics. Springer, Cham, [2021]©2021, pp. xviii+726.isbn: 978-3-030-82594-2; 978-3-030-82595-9.doi:10.1007/978-3-030-82595-9

    work page doi:10.1007/978-3-030-82595-9 2021

  13. [13]

    Vershynin, R.High-dimensional probability. Vol. 47. Cambridge Series in Statistical and Probabilistic Mathematics. Cambridge University Press, Cambridge, 2018, pp. xiv+284. isbn: 978-1-108-41519-4.doi:10.1017/9781108231596. 19

    work page doi:10.1017/9781108231596 2018