REVIEW 1 major objections 2 minor 23 references
Poisson hyperplane processes and approximation of convex bodies
T0 review · 1 major / 2 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read Random halfspace-cut polytopes approximate convex bodies at sharp, dimension-dependent rates.
desk verdict Solid extension of Schneider's isotropic K-cell result to non-isotropic processes; exact constants and facet bounds are new, and the only real blemish is a possibly overbroad Theorem 2. 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 central object is the $K$-cell $Z_K^{(n)}$, defined as the intersection of all closed halfspaces bounded by hyperplanes of the process that do not meet $K$. The proofs run through three mechanisms. A metric entropy estimate for the space of convex bodies gives a deviation inequality for the hitting functional $\Phi$, from which all moment bounds of order $n^{-2k/(d+1)}$ follow. A dualization of the floating-body construction maps the wet part of the polar body $K^\circ$ to the family of hyperplanes separating $K$ from a point, so cap-volume lower bounds for the polar body become lower bounds for the mean-width error. A Poissonization lemma transfers asymptotic relations known for $k$ independent random hyperplanes to the Poisson process, and the Slivnyak–Mecke identity $E f_{d-1}(Z_K^{(n)})=2n\,E[\Phi(Z_K^{(n)})-\Phi(K)]$ turns the hitting-functional estimates into facet-number estimates.
What would settle it
For a numerical check, take $K$ as a disk in $\mathbb{R}^2$ and an isotropic Poisson line process, simulate the $K$-cell for increasing $n$, and verify that $n^{2/3}(E[W(Z_K^{(n)})]-W(K))$ approaches the constant from Theorem 5 computed for the circle; if it does not settle on that value, the sharp upper-order claim fails. The logarithmic lower order can be tested by repeating the simulation for a square: $(n/\log n)(E[W(Z_K^{(n)})]-W(K))$ should approach the stated constant from Theorem 6.
Extended reading notes
Core claim
At the center is a pair of order-sharp bounds for the mean width. If the directional distribution $\varphi$ satisfies $\varphi\le a_0\sigma$, then $E[W(Z_K^{(n)})-W(K)]\gg n^{-1}\log^{d-1}n$; if $\varphi\ge a_1\sigma$, then $E[W(Z_K^{(n)})-W(K)]\ll n^{-2/(d+1)}$. Here $\sigma$ is normalized spherical Lebesgue measure and $\gg,\ll$ mean up to constants depending only on $d,K,\varphi$. Under stronger smoothness assumptions the paper evaluates the limits: for positive continuous density $q$, $n^{2/(d+1)}E[W(Z_K^{(n)})-W(K)]$ converges to $2^{-2/(d+1)}F(K,q)$, where $F(K,q)$ is an explicit boundary integral of the density and the Gauss–Kronecker curvature; and for an isotropic process with simplicial polytopes, $n/\log^{d-1}n$ times the same error converges to $rd(\log 2/(d+1))^{d-1}$. It also proves that the $k$-th moment of the facet number of the $K$-cell is at most $c(k)n^{k(d-1)/(d+1)}$ and that, for smooth $K$, $n^{-(d-1)/(d+1)}E f_{d-1}(Z_K^{(n)})$ converges to $2^{(d-1)/(d+1)}G(K,q)$.
Load-bearing premise
The exact constants in the asymptotic limits rest on prior limiting results for the mean width of polytopes circumscribed around $K$ by $k$ independent random tangent hyperplanes; if either prior result is incorrect, the corresponding constants in this paper would not follow, and the lower bound also relies on a cap-volume lower bound for the polar body.
Editorial extensions
If this is right
- The mean-width error of the $K$-cell is bounded below by $n^{-1}\log^{d-1}n$ whenever the directional distribution is not too concentrated, so no isotropy assumption is needed for the lower order.
- For any body and any directional distribution, the expected hitting-functional error is $O(n^{-2/(d+1)})$, even in cases where the $K$-cell does not converge almost surely to $K$.
- For smooth bodies with a positive continuous directional density, the limiting constant is given explicitly as a curvature- and density-weighted boundary integral, showing that the exponent $2/(d+1)$ is sharp.
- For isotropic processes and simplicial polytopes, the logarithmic lower order is attained, so both bounds in the main theorem are optimal.
- The expected facet number of the $K$-cell grows like $n^{(d-1)/(d+1)}$ up to constants, with a matching asymptotic constant for smooth $K$.
Reading between the lines
- Because the asymptotic constant $F(K,q)$ weights the boundary by $q^{-2/(d+1)}\kappa^{d/(d+1)}$, the paper's result implies that for a fixed process the body's local curvature, not just its volume, controls the leading approximation error; this dependence could be tested by comparing bodies with the same width but different curvature.
- The same entropy-net plus Poissonization route is the natural template for the open problem the paper names: exact rates for intrinsic volumes $V_i$ with $i\ne 1,d$.
- The exact identity linking expected facet count to the hitting-functional error means that in applications one can estimate approximation quality by counting facets of the $K$-cell, a quantity directly observable from the hyperplane arrangement.
- For anisotropic processes, Theorem 5 predicts that the mean-width error's leading constant changes by the factor $q^{-2/(d+1)}$ integrated over $\partial K$; tilting an otherwise isotropic process and measuring the error would provide a clean test of that prediction.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies the K-cell Z_K^(n) of a stationary Poisson hyperplane process in R^d, defined as the intersection of all closed halfspaces containing a convex body K whose boundary hyperplanes do not meet K. The central object is the expected deviation of a functional evaluated at Z_K^(n) from its value at K, as the intensity n tends to infinity. For the mean width W, Theorem 1 establishes the two-sided rate n^{-1} log^{d-1} n ≪ E W(Z_K^(n)) - W(K) ≪ n^{-2/(d+1)} under the respective assumptions φ ≤ a0 σ and φ ≥ a1 σ on the directional distribution φ. For the hitting functional Φ, Theorem 2 gives a general upper bound of order n^{-2/(d+1)}, with moment versions in Theorem 4 and facet-number moment bounds in Theorem 7. Under a positive continuous density of φ, Theorems 5 and 8 provide exact asymptotic constants for the mean width difference and the expected facet number; in the isotropic simplicial-polytope case, Theorems 6 and 9 give exact logarithmic asymptotics. The proofs combine a dual wet-part argument based on Bárány-Larman, Bronshtein's entropy bound for convex bodies, and Poissonization lemmas that transfer finite-k asymptotics to Poisson intensities.
Significance. If correct, the paper significantly extends the isotropic results of [21] to general directional distributions, resolves the sharp order of the mean-width approximation for the K-cell model, and provides exact constants in smooth and simplicial cases. The hitting-functional bounds and the facet-number estimates are of independent interest and give a Poisson-dual counterpart to classical random-polytope results. The proofs are detailed and largely cross-checkable, and the main external dependencies are published theorems (Bárány-Larman, [4], [5]) rather than new claims. The paper does not provide machine-checked proofs or code, but the analytic arguments are transparent and the key inequalities are explicitly derived.
major comments (1)
- [Section 2, Lemma 1] Lemma 1 as stated is false without a non-degeneracy assumption on φ. For example, in d=2 take φ=(δ_{e1}+δ_{-e1})/2; then every hyperplane of the process is vertical and the K-cell is almost surely an unbounded vertical strip, so Ro(Z_K^(n))=∞ and the claimed bound P(Ro(Z_K^(n))>b(Ro+x)) ≤ 2d e^{-a n x} fails for large x. The condition that φ is not concentrated on a great subsphere is used later in Section 3 but is not assumed in Section 2, where inequalities (6)-(7) and the proof of Theorem 4 rely on Lemma 1. Consequently Theorem 2 and Theorem 7 are proved only under an implicit extra assumption as they stand. Since the hypotheses (3) and (4) of Theorem 1 imply the needed non-degeneracy, the main theorem is not affected, but the statements of Theorems 2-4 and 7 should either explicitly assume φ is not concentrated on a great subsphere (or a comparable condition) or provide a separate argument that handles degenerate φ.
minor comments (2)
- [Section 3, Eq. (17)] If ν is defined as the image measure of Lebesgue measure under the map η(ru)=H(u,1/r), then the displayed formula should contain the factor d in front of ω_d, i.e., dω_d ∫ τ^{-(d+1)} dτ σ(du), because the normalized spherical measure σ is related to the surface measure by dS = dω_d σ. The missing factor is harmless for the constant comparisons c5,c6 that follow, but the displayed equation should be corrected.
- [Section 3, proof of Theorem 1] The identity W(Z)-W(K)=∫_{H\H_K} 1_{H∩Z≠∅} μ(dH) deserves a one-sentence justification: the integration over the two families τ>h(K,u) and τ<-h(K,-u) produces the factor 2 that appears in W=2∫ h dσ. As written, the reader must infer this from the definition of μ in (14).
Circularity Check
No significant circularity: the central derivation is self-contained and rests on independent external estimates; same-author citations supply prior lemmas, not the target result.
full rationale
The paper's central Theorem 1 is derived from two independent external ingredients: the Bronshtein epsilon-net bound (Lemma 2, [6]) for the upper estimate and the Barany-Larman wet-part estimate (13) applied to the polar body for the lower estimate. The intermediate identity (15) and the inclusion eta(K^o(t)) subset H^psi(const*t) are proved inside the paper and do not assume the desired asymptotic. Lemma 1 is imported from Schneider [21], but it is a tail estimate on the radius of the K-cell, not the mean-width difference, and it is used only to confine the K-cell to a bounded ball; it is not the target theorem. Theorem 2 and the upper bound in Theorem 1 follow from the deviation estimate Theorem 3, whose proof is self-contained apart from that external tail lemma. The exact asymptotic relations in Theorems 5, 6, 8, and 9 use [4, Thms. 5.2, 5.3] and [5, Thm. 1.3] by overlapping authors; those are previously published finite-random-hyperplane limit theorems with different non-Poisson distributions, and Lemmas 5 and 8 supply the Poissonization. Thus the exact constants are inherited from published theorems rather than assumed, and no fitted parameter or definitional identity is renamed as a prediction. No circular step was found.
Assumptions & free parameters
assumptions (6)
- domain assumption Stationary Poisson hyperplane process with intensity n and even directional distribution phi on S^{d-1} has independent counts and the given intensity measure representation.
- standard math Bárány-Larman lower bound lambda_d(K(epsilon)) >= const * epsilon log^{d-1}(1/epsilon) for sufficiently small epsilon (their Theorem 2).
- standard math Bronshtein epsilon-net theorem for convex bodies in a fixed ball: N_epsilon <= c1 exp(epsilon^{-(d-1)/2}).
- standard math Limits for random hyperplane intersections: [4, Thm 5.2] gives lim_{k->infty} k^{2/(d+1)} g_k = F(K,q); [5, Thm 1.3] gives the simplicial polytope limit; [4, Thm 5.3] gives lim k^{-(d-1)/(d+1)} p_k = G(K,q).
- standard math Slivnyak-Mecke formula for Poisson hyperplane processes.
- standard math Lemma 1 from [21]: P(Ro(Z_K)> b(Ro+x)) <= 2d e^{-a n x}.
Cite this review
Pith. "Pith review of Poisson hyperplane processes and approximation of convex bodies." pith.science (2026). https://pith.science/paper/7AIJ74FI
@misc{pith2026190809498,
author = {Pith},
title = {Pith review of: Poisson hyperplane processes and approximation of convex bodies},
year = {2026},
howpublished = {\url{https://pith.science/paper/7AIJ74FI}},
note = {Machine review of arXiv:1908.09498}
}
abstract
A natural model for the approximation of a convex body $K$ in $\mathbb{R}^d$ by random polytopes is obtained as follows. Take a stationary Poisson hyperplane process in the space, and consider the random polytope $Z_K$ defined as the intersection of all closed halfspaces containing $K$ that are bounded by hyperplanes of the process not intersecting $K$. If $f$ is a functional on convex bodies, then for increasing intensities of the process, the expectation of the difference $f(Z_K)-f(K)$ may or may not converge to zero. If it does, then the order of convergence and possible limit relations are of interest. We study these questions if $f$ is either the hitting functional or the mean width.
Reference graph
Works this paper leans on
-
[21]
Schneider, R., Interaction of Poisson hyperplane proc esses and convex bodies. J. Appl. Prob. (to appear), arXiv:1812.08443
-
[4]
B¨ or¨ oczky, K.J., Fodor, F., Hug, D., The mean width of random polytopes circumscribed around a convex body. J. London Math. Soc. 81 (2010), 499–523
work page 2010
-
[5]
B¨ or¨ oczky, K.J., Schneider, R., The mean width of circu mscribed random polytopes. Canad. Math. Bull. 53 (2010), 614–628
work page 2010
-
[1]
B´ ar´ any, I., Random polytopes, convex bodies, and appr oximation. In: A. Baddeley, I. B´ ar´ any, R. Schneider, W. Weil,Stochastic Geometry, pp. 77–118, Lecture Notes in Math. 1892, Springer, Berlin, 2007
work page 2007
-
[2]
B´ ar´ any, I., Random points and lattice points in convex bodies. Bull. Amer. Math. Soc. 45 (2008), 339–356
work page 2008
-
[3]
Mathematika 35 (1988), 274–291
B´ ar´ any I., Larman, D.G., Convex bodies, economic cap c overings, random polytopes. Mathematika 35 (1988), 274–291
work page 1988
-
[6]
Bronshtein, E.M., ǫ-entropy of convex sets and functions. Siberian Math. J. 17 (1976), 393–398
work page 1976
-
[7]
Brunel, V.–E., Uniform deviation and moment inequaliti es for random polytopes with general densities in arbitrary convex bodies. arXiv:1704. 01620
Show all 23 references
-
[8]
Buchta, C., Zuf¨ allige Polyeder – Eine ¨Ubersicht. In: E. Hlawka (ed), Zahlentheoretische Analysis, pp. 1–13, Lecture Notes in Math. 1114. Springer, Berlin, 1985
1985
-
[9]
Calka, P., Schreiber, T., Large deviation probabilitie s for the number of vertices of random polytopes in the ball. Adv. Appl. Prob. (SGSA) 38 (2006), 47–58
2006
-
[10]
Hug, D., Random polytopes. In: E. Spodarev (ed), Stochastic Geometry, Spatial Statis- tics and Random Fields—Asymptotic Methods, pp. 205–238, Lecture Notes in Math. 2068, Springer, Berlin, 2013
2013
-
[11]
Hug, D., Schneider, R., Approximation properties of ra ndom polytopes associated with Poisson hyperplane processes. Adv. Appl. Prob. 46 (2014), 919–936
2014
-
[12]
Doctoral The- sis, Albert-Ludwigs-Universit¨ at, Freiburg i
Kaltenbach, F.J., Asymptotisches Verhalten zuf¨ alliger konvexer Polyeder. Doctoral The- sis, Albert-Ludwigs-Universit¨ at, Freiburg i. Br, 1990
1990
-
[13]
Korostelev, A.P., Simar, L., Tsybakov, A.B., On estima tion of monotone and convex boundaries. Publ. Inst. Stat. Univ. Paris 39 (1995), 3–18
1995
-
[14]
Randomized Algorithms and Probabilistic Analysis
Mitzenmacher, M., Upfal, E., Probability and Computing. Randomized Algorithms and Probabilistic Analysis. Cambridge University Press, Cambridge, 2005
2005
-
[15]
Reitzner, M., Central limit theorems for random polyto pes. Probab. Theory Relat. Fields 133 (2005), 483–507
2005
-
[16]
Reitzner, M., Random polytopes. In: W.S. Kendall, I. Mo lchanov (eds), New Perspec- tives in Stochastic Geometry, pp. 45–76, Oxford University Press, Oxford, 2010
2010
-
[17]
Aequationes Math
Schneider, R., Approximation of convex bodies by rando m polytopes. Aequationes Math. 32 (1987), 304–310. 19
1987
-
[18]
Schneider, R., Random approximation of convex sets. J. Microscopy 151 (1988), 211– 227
1988
-
[19]
Schneider, R., Discrete aspects of stochastic geometr y. In: J.E. Goodman, J. O’Rourke, C.D. T´ oth (eds),Handbook of Discrete and Computational Geometry, 3rd edn., pp. 299– 329, CRC Press, Boca Raton, 2018
2018
-
[20]
2nd edn., Encyclopedia of Mathematics and Its Applications, vol
Schneider, R., Convex Bodies: The Brunn–Minkowski Theory. 2nd edn., Encyclopedia of Mathematics and Its Applications, vol. 151, Cambridge University Press, Cambridge, 2014
2014
-
[22]
Springer, Berlin, 2008
Schneider, R., Weil, W., Stochastic and Integral Geometry. Springer, Berlin, 2008
2008
-
[23]
Weil, W., Wieacker, J.A., Stochastic geometry. In: P.M . Gruber, J.M. Wills (eds), Handbook of Convex Geometry, pp. 1391–1438, North-Holland, Amsterdam, 1993. Authors’s addresses: Daniel Hug Karlsruhe Institute of Technology, Department of Mathemat ics D-76128 Karlsruhe, Germa...
1993
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.