Pith. sign in

REVIEW 4 minor 39 references

Facets of spherical random polytopes

T0 review · 0 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper derives explicit asymptotic formulas for the number and heights of the facets of the convex hull of $n$ uniform points on the sphere $S^{d-1}$, across every regime of $n$ and $d$.

desk verdict Complete asymptotic regime classification for spherical random polytopes; solid and citable, with one minor algebraic slip that does not affect the main results. read the letter →

arxiv 1908.04033 v1 pith:IQ6LKYTY submitted 2019-08-12 math.PR math.MG

Gilles Bonnet , Eliza O'Reilly This is my paper · ORCID
classification math.PRmath.MG MSC 60D0552A2260F05
keywords randompolytopessphericalpointsconvexhullfacetheightsexpectednumberoffacetshigh-dimensionalasymptoticsstochasticgeometry
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper works out, in every asymptotic regime of $n$ (number of points) and $d$ (dimension), what the facets of the convex hull of $n$ uniform points on the unit sphere $S^{d-1}$ look like. It gives explicit asymptotic formulas for the expected number of facets and for the distribution of the typical facet height, showing that behavior splits into slow regimes, subexponential, exponential, and super-exponential regimes. If correct, these formulas give a complete classification of facet statistics for spherical random polytopes as $n\to\infty$ and $d$ either fixed or growing, and they extend known fixed-dimension results to high dimension. The results come from analyzing one exact integral representation, not from simulation.

What carries the argument

The engine is the exact integral representation of the expected number of facets with height in $[h_1,h_2]$, namely $F[h_1,h_2]=\binom{n}{d}2c_{(d^2-2d-1)/2}\int_{h_1}^{h_2}(1-h^2)^{(d^2-2d-1)/2}\left(c_{(d-3)/2}\int_{-1}^{h}(1-s^2)^{(d-3)/2}ds\right)^{n-d}dh$. The inner integral is a distribution function that becomes a rescaled normal CDF as $d$ grows, so the integrand is approximated by $\exp(d f_\rho(r))$ with $f_\rho(r)=\rho\ln\Phi(r)-r^2/2$. The standard asymptotic method for integrals with a single dominating peak then turns the integrals into explicit asymptotics. A Gamma substitution and tail bounds isolate the typical height in the fast regimes, giving total variation convergence to a Gamma variable when $\ln n\gg d\ln d$.

What would settle it

For an increasing sequence such as $d_j=j^2$ and $n_j=d_j+5j$ (so $(n-d)/\sqrt{d}\to 5$ while $n-d\ll d$), evaluate the integral (5) by numerical quadrature and compare its logarithm with the asymptotic from Theorem 9, $F[-1,1]=\binom{n}{d}2^{d-n+1}e^{(n-d)^2/(\pi d)+o(1)}$. A persistent relative mismatch in the exponent would refute the paper's central claim; the same check can be repeated for one sequence in each regime.

Watch

Extended reading notes

Core claim

The paper's central claim is that the expected number of facets $F[-1,1]$ and the typical facet height $H_{\mathrm{typ}}$ are governed by a handful of explicit formulas depending on how $\ln n$ compares with $d$. In the slow sub-linear regime $n-d\ll d$, $F[-1,1]=\binom{n}{d}2^{d-n+1}e^{(n-d)^2/(\pi d)+o(1)}$. In the exponential regime $(\ln n)/d\to\rho$, $F[-1,1]=[2\pi(e^{2\rho}-1)d(1+o(1))]^{(d-1)/2}$, and the typical height converges in probability to $\sqrt{1-e^{-2\rho}}$. In the super-exponential regime $\ln n\gg d$, the expected number grows as $n K_d h_*^{d-1}$ with $h_*=\sqrt{1-d^{3/(d-1)}n^{-2/(d-1)}}$. The same analysis identifies nearly deterministic bounds $[h_1,h_2]$ that contain all facet heights with probability tending to one.

Load-bearing premise

All of the paper's conclusions rest on a single previously derived exact formula for the expected number of facets of the spherical hull, which the paper uses as its starting point and never re-derives.

Editorial extensions

If this is right

  • In the exponential regime $(\ln n)/d\to\rho$, the expected facet count grows as $[2\pi(e^{2\rho}-1)d]^{(d-1)/2}$, so the polytope has super-exponentially many facets in $d$ while typical facet height stays bounded away from $0$ and $1$.
  • In the linear regime $n-d=\rho d+o(d)$, the classical threshold puts the origin inside the hull with probability going to one for $\rho>1$, but the expected number of negative-height facets still diverges for $\rho$ up to about $3.4$, an intermediate regime the paper identifies.
  • The Hausdorff distance from the polytope to the ball converges in probability to $1-\sqrt{1-e^{-2\rho}}$ in the exponential regime, to $0$ in the super-exponential regime, and to $1$ when $\ln n\ll d$.
  • When $\ln n\gg d\ln d$, the rescaled quantity $n\Gamma(d/2)(2\sqrt{\pi}\Gamma((d+1)/2))^{-1}(1-H_{\mathrm{typ}}^2)^{(d-1)/2}$ converges in total variation to a gamma variable with shape parameter $d-1$, which also recovers the distribution of circumscribed cap radii in spherical Delaunay triangulations.
  • The same formulas answer the three questions posed: the distribution of the typical facet, a tight range containing all facet heights, and the expected number of facets, in every regime considered.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper gives first-order asymptotics only; the same integral representation could be expanded to yield finite-$n$ corrections and rates of convergence in each regime.
  • Because the starting integral formula exists for the broader class of beta polytopes, the regime classification and the main formulas may extend with modified constants to beta-distributed points, a testable reading of the cited formulas.
  • The agreement between these regimes and the known angle-extremum regimes for $n$ random spherical vectors suggests facet heights and angular coherence are two views of the same high-dimensional geometry; a sharper finite-$n$ correspondence between minimum angle and maximum facet height could be checked directly from the hull of $n$ points.
  • The threshold near $\rho\approx3.4$ for negative-height facets in expectation suggests a large-deviation regime for the origin being outside the hull that is not captured by the probability threshold alone.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

0 major / 4 minor

Summary. The paper studies the random polytope P_{n,d} formed as the convex hull of n i.i.d. uniform points on S^{d-1}, with n→∞ and d either fixed or tending to ∞. It identifies several asymptotic regimes for the pair (n,d) — sublinear, linear, subexponential, exponential, and super-exponential — and, in each regime, establishes the limiting distribution of the height of a typical facet, a tight interval for the heights of all facets, and asymptotic formulas for the expected number of facets. The results are derived from the exact integral representation (5) of the expected number of facets with height in an interval, which is cited from earlier work, followed by Laplace's method or dominated convergence. The paper also recovers and extends fixed-dimension results of Buchta–Müller–Tichy and Kabluchko–Thäle–Zaporozhets, and connects the facet-height asymptotics to the Hausdorff distance and to spherical Delaunay triangulations.

Significance. This is a significant contribution to high-dimensional stochastic geometry. The paper provides a systematic, regime-by-regime asymptotic analysis of facet heights and expected facet numbers for spherical random polytopes as d grows, with explicit constants and no free parameters. The derivation is rigorous and self-contained beyond the well-established integral formula (5), and the estimates are standard dominated-convergence and Laplace arguments. The consistency of the results with known fixed-dimension limits and with Gaussian analogues, such as Theorem 12 matching the form of results in [9], gives additional confidence in the conclusions. The paper also demonstrates the versatility of the exact integral representation (5) as a tool for dimension-dependent asymptotics.

minor comments (4)
  1. [§6.1.1, Lemma 18] In the case r < sqrt(2/π), the endpoint Laplace prefactor in (15) and (16) should be proportional to 1/[2^{n-d} (n-d) sqrt(d) |sqrt(2/π)-r|], not (n-d)/[2^{n-d} sqrt(d) |sqrt(2/π)-r|]; the current expression is too large by a factor (n-d)^2. This error does not affect the proofs of Theorems 2 and 9 because those conclusions depend on the exponential order and on the r > sqrt(2/π) case, but the lemma statement as written is incorrect and should be corrected.
  2. [§6.2.4, Proof of Theorem 5] After defining A1 = [0,b_{n,d}] and A2 = [b_{n,d},∞), the sentence "Therefore, for any [a,b] ⊆ A2" appears in the middle of the argument for A1; it should read A1, since the stronger inequality (37) is claimed for subsets of A1.
  3. [Theorem 5] The phrase "For k ∈ N" introduces a symbol k that is never used; the random variable is X_{d-1}, so this phrase should be removed or replaced with a sentence that does not introduce unused notation.
  4. [Theorem 8, equation (1)] The displayed formula for h2 is garbled by the typesetting ("/radicaltp /radicalvertex /radicalvertex"); it should be h2 = sqrt(1 - (r2 d/n)^{2(d+1)/(d-1)^2}) as defined later in (28).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all asymptotics are deduced from the externally cited exact integral representation (5).

full rationale

The derivation chain is self-contained given the exact expected-facet formula (5), which is cited from Kabluchko, Thale and Zaporozhets [26, Thm 1.2] and Bonnet et al. [8]. The paper then defines I[h1,h2] in (7) and F[h1,h2] in (8) directly from that formula, and every theorem in Sections 2-4 is obtained by asymptotic evaluation of that integral using dominated convergence, Laplace's method, and Gamma-distribution concentration. No parameter is fitted to any target quantity, and no simulated or empirical quantity is renamed as a prediction. The 'typical height' Htyp is defined probabilistically, not as the output of the asymptotic analysis; its representation as I[h1,h2]/I[-1,1] is the external exact formula, so deriving its limit from that representation is not circular. The self-citations [7,8] are not load-bearing for the central claims: [8] is cited alongside [26] for detailed proofs of the same exact representation, and [7] is contextual related work. The only internal issue noticed is a non-central prefactor slip in Lemma 18's r < sqrt(2/pi) branch, which affects no theorem because Theorem 2 is governed by exponential order and I[-1,1] is computed from the correct r > r* case; this is a correctness matter, not circularity. Therefore the paper's claims do not reduce to their inputs by construction.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claim is built on an exact integral formula from the literature and on standard asymptotic tools; there are no free parameters fitted to data and no invented entities. The only non-elementary input beyond classical analysis is the integral representation (5), which is the single external theorem.

assumptions (6)
  • domain assumption Integral representation of expected number of facets, Eq. (5): F[h1,h2] = binom(n,d) * 2*c * integral from h1 to h2 of (1-h^2)^((d^2-2d-1)/2) * (c * integral from -1 to h of (1-s^2)^((d-3)/2) ds)^(n-d) dh.
    Taken from [26, Thm 1.2] and [8]; the exact starting point for all proofs in Section 6.
  • standard math Laplace's method, Eqs. (9)-(10)
    Used to evaluate integrals with exponents growing in d; quoted from [39].
  • standard math Gautschi's inequality, Eq. (11)
    Gives c_alpha ~ sqrt(d/(2*pi)); used repeatedly for normalizing constants.
  • standard math Stirling's formula for factorial and binomial asymptotics
    Used in proofs of Theorems 10, 11, 13 and Lemma 25.
  • domain assumption Wendel's theorem on origin containment
    Used in Remark 1 to interpret the regime threshold; not needed for main formulas.
  • standard math Gamma(d-1) concentration: X_{d-1}/(d-1) tends to 1 in probability
    Used in proofs of Theorems 5, 11, 12, 13 to show tail probabilities vanish.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Facets of spherical random polytopes." pith.science (2026). https://pith.science/paper/IQ6LKYTY

@misc{pith2026190804033,
  author       = {Pith},
  title        = {Pith review of: Facets of spherical random polytopes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IQ6LKYTY}},
  note         = {Machine review of arXiv:1908.04033}
}
abstract

Facets of the convex hull of $n$ independent random vectors chosen uniformly at random from the unit sphere in $\mathbb{R}^d$ are studied. A particular focus is given on the height of the facets as well as the expected number of facets as the dimension increases. Regimes for $n$ and $d$ with different asymptotic behavior of these quantities are identified and asymptotic formulas in each case are established. Extensions of some known results in fixed dimension to the case where dimension tends to infinity are described.

Figures

Figures reproduced from arXiv: 1908.04033 by the authors.

Figure 1
Figure 1. Plot of the function (0, ∞) ∋ ρ 7→ gρ(0). 6.2 Fast regimes We now turn to the proofs of results in the regime where n ≫ d. The following lemma gives the asymptotic behavior of a height depending on n and d in a particular way that will be used in the approximations in this regime. Lemma 22. Let f(n, d) be a function of n and d such that ln f(n, d) = o(ln(n/d)). Assume that n ≫ d and let h := h(n, d) = s 1 −  d n f(… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 39 canonical work pages

  1. [9]

    Facets of high-dimensional Gaussian polytopes

    K´ aroly J. B¨ or¨ oczky, G´ abor Lugosi, and Matthias Reitzner.Facets of high-dimensional Gaussian polytopes. arXiv:1808.01431, 2018

    work page Pith review arXiv 2018

  2. [1]

    Volume degeneracy of t he typical cell and the chord length distribution for Poisson-Voronoi tessellations in high dimensions

    Kasra Alishahi and Mohsen Sharifitabar. Volume degeneracy of t he typical cell and the chord length distribution for Poisson-Voronoi tessellations in high dimensions. Advances in Applied Probability , 40:919–938, 2008

    work page 2008

  3. [2]

    On the isotropy constant of random convex sets

    David Alonso-Guti´ errez. On the isotropy constant of random convex sets. Proceedings of the American Math- ematical Society, 136(9):3293–3300, 2008

    work page 2008

  4. [3]

    Litvak, and Nicole Tomcz ak-Jaegermann

    David Alonso-Guti´ errez, Alexander E. Litvak, and Nicole Tomcz ak-Jaegermann. On the isotropic constant of random polytopes. Journal of Geometric Analysis , 26(1):645–662, 2016

    work page 2016

  5. [4]

    David Alonso-Guti´ errez, Florian Besau, Julian Grote, Zakhar K abluchko, Mattias Reitzner, Christoph Th¨ ale, Beatrice-Helen Vritsiou, and Elisabeth M. Werner. Asymptotic norm ality for random simplices and convex bodies in high dimensions. arXiv:1906.02471, 2019

    work page Pith review arXiv 1906

  6. [5]

    Random polytopes, convex bodies, and approximation

    Imre B´ ar´ any. Random polytopes, convex bodies, and approximation. In Stochastic geometry, volume 1892 of Lecture Notes in Math. , pages 77–118. Springer, Berlin, 2007

    work page 2007

  7. [6]

    Central limit theorems for gaussian polytopes

    Imre B´ ar´ any and Van Vu. Central limit theorems for gaussian polytopes. The Annals of Probability, 35(4):1593– 1621, 2007

    work page 2007

  8. [7]

    Threshold phenomena for high-dimensional random polytopes

    Gilles Bonnet, Giorgos Chasapis, Julian Grote, Daniel Temesvari, a nd Nicola Turchi. Threshold phenomena for high-dimensional random polytopes. Communications in Contemporary Mathematics , 21(05):1850038, 2019

    work page 2019

Show all 39 references
  1. [8]

    Monotonic- ity of facet numbers of random convex hulls

    Gilles Bonnet, Julian Grote, Daniel Temesvari, Christoph Th¨ ale, Nicola Turchi, and Florian Wespi. Monotonic- ity of facet numbers of random convex hulls. Journal of Mathematical Analysis and Applications , 455(2):1351– 1364, 2017

    2017

  2. [10]

    M¨ uller, and Robert F

    Christian Buchta, Josef S. M¨ uller, and Robert F. Tichy. Stoch astical approximation of convex bodies. Mathe- matische Annalen , 271(2):225–235, 1985

    1985

  3. [11]

    Distribution of angles in random packing on spheres

    Tony Cai, Jianqing Fan, and Tiefeng Jiang. Distribution of angles in random packing on spheres. Journal of Machine Learning Research, 14:1837–1864, 2013

    2013

  4. [12]

    Limiting laws of coherence of random matrices with applications to testing covariance structure and construction of compressed sensing m atrices

    Tony Cai and Tiefeng Jiang. Limiting laws of coherence of random matrices with applications to testing covariance structure and construction of compressed sensing m atrices. The Annals of Statistics , 39:1496–1525, 2011

    2011

  5. [13]

    Near-optimal recovery f rom random projections: Universal encoding strategies? IEEE Transactions on Information Theory , 12:5406–5425, 2006

    Emmanuel Candes and Terence Tao. Near-optimal recovery f rom random projections: Universal encoding strategies? IEEE Transactions on Information Theory , 12:5406–5425, 2006. 27

    2006

  6. [14]

    The Dantzig selector: Sta tistical estimation when p is much larger than n

    Emmanuel Candes and Terence Tao. The Dantzig selector: Sta tistical estimation when p is much larger than n. The Annals of Statistics , 35(6):2313–2351, 2007

    2007

  7. [15]

    Giannopoulos, and Olivier Gu´ edon

    Nikos Dafnis, Apostolos A. Giannopoulos, and Olivier Gu´ edon. On the isotropic constant of random polytopes. Adv. Geom., 10(2):311–322, 2010

    2010

  8. [16]

    David L. Donoho. Compressed sensing. IEEE Transactions on Information Theory , 52:1289–1306, 2006

    2006

  9. [17]

    Random inscribed p olytopes have similar radius functions as Poisson-Delaunay mosaics

    Herbert Edelsbrunner and Anton Nikitenko. Random inscribed p olytopes have similar radius functions as Poisson-Delaunay mosaics. The Annals of Applied Probability , 28(5):3215–3238, 2018

    2018

  10. [18]

    Network Information Theory

    Abbas El Gamal and Young-Han Kim. Network Information Theory . Cambridge University Press, 2011

    2011

  11. [19]

    Some elementary inequalities relating to the gam ma and incomplete gamma function

    Walter Gautschi. Some elementary inequalities relating to the gam ma and incomplete gamma function. Studies in Applied Mathematics , 38:77–81, 1959

    1959

  12. [20]

    Asymptotic approximation of smooth convex bodies by polytopes

    Stefan Glasauer and Rolf Schneider. Asymptotic approximation of smooth convex bodies by polytopes. Forum Mathematicum, 8, 1996

    1996

  13. [21]

    Limit theorems for random simplices in high dimensions

    Julian Grote, Zakhar Kabluchko, and Christoph Th¨ ale. Limit theorems for random simplices in high dimensions. ALEA Latin American Journal of Probability and Mathematica l Statistics , 16:141–177, 2019

    2019

  14. [22]

    Random polytopes

    Daniel Hug. Random polytopes. In Stochastic geometry, spatial statistics and random fields , volume 2068 of Lecture Notes in Math. , pages 205–238. Springer, Heidelberg, 2013

    2013

  15. [23]

    Poisson polyhedra in high dimensions

    Julia H¨ orrman, Daniel Hug, Matthias Reitzner, and Christoph T h¨ ale. Poisson polyhedra in high dimensions. Advances in Mathematics , 281:1–39, 2015

    2015

  16. [24]

    On th e isotropic constant of random polytopes with vertices on an ℓp-sphere

    Julia H¨ orrmann, Joscha Prochno, and Christoph Th¨ ale. On th e isotropic constant of random polytopes with vertices on an ℓp-sphere. The Journal of Geometric Analysis , 28(1):405–426, 2017

    2017

  17. [25]

    Ex pected intrinsic volumes and facet numbers of random beta-polytopes

    Zakhar Kabluchko, Daniel Temesvari, and Christoph Th¨ ale. Ex pected intrinsic volumes and facet numbers of random beta-polytopes. Mathematische Nachrichten , 292(1):79–105, 2019

    2019

  18. [26]

    Beta polytopes and poisson polyhedra: f-vectors and angles

    Zakhar Kabluchko, Christoph Th¨ ale, and Dmitry Zaporozhets. Beta polytopes and poisson polyhedra: f-vectors and angles. arXiv:1805:01338, 2018

    2018

  19. [27]

    On the hyperplane conjecture f or random convex sets

    Bo’az Klartag and Gady Kozma. On the hyperplane conjecture f or random convex sets. Israel Journal of Mathematics, 170:253–268, 2009

    2009

  20. [28]

    Thin-shell concentration for zero cells of stationar y poisson mosaics

    Eliza O’Reilly. Thin-shell concentration for zero cells of stationar y poisson mosaics. arXiv:1809.04134, 2018

    2018 arXiv

  21. [29]

    The stochastic geometry of un constrained one-bit data compression

    Eliza O’Reilly and Fran¸ cois Baccelli. The stochastic geometry of un constrained one-bit data compression. arXiv:1810.06095, 2018

    2018 arXiv

  22. [30]

    The isotr opic constant of random polytopes with vertices on convex surfaces

    Joscha Prochno, Christoph Th¨ ale, and Nicola Turchi. The isotr opic constant of random polytopes with vertices on convex surfaces. Journal of Complexity , 54, 2019

    2019

  23. [31]

    Central limit theorems for random polytope s

    Matthias Reitzner. Central limit theorems for random polytope s. Probability and Related Fields , 133(4):483– 507, 2005

    2005

  24. [32]

    Richardson, Van H

    Ross M. Richardson, Van H. Vu, and Lei Wu. An inscribing model f or random polytopes. Discrete & Compu- tational Geometry , 39(1):469–499, 2008

    2008

  25. [33]

    Stochastic and Integral Geometry

    Rolf Schneider and Wolfgang Weil. Stochastic and Integral Geometry . Springer, 2008

    2008

  26. [34]

    Claude E. Shannon. A mathematical theory of communication. Bell System Technical Journal , 27:79–423 and 623–656, 1948

    1948

  27. [35]

    Central limit theorem for the volume of rando m polytopes with vertices on the bound- ary

    Christoph Th¨ ale. Central limit theorem for the volume of rando m polytopes with vertices on the bound- ary. Discrete & Computational Geometry. An International Journ al of Mathematics and Computer Science , 59(4):990–1000, 2018

    2018

  28. [36]

    Limit theorems for random polyto pes with vertices on convex surfaces

    Nicola Turchi and Florian Wespi. Limit theorems for random polyto pes with vertices on convex surfaces. Advances in Applied Probability , 50(4):1227–1245, 2018. 28

    2018

  29. [37]

    Central limit theorems for random polytopes in a smooth convex set

    Van Vu. Central limit theorems for random polytopes in a smooth convex set. Advances in Mathematics , 207(1):221–243, 2006

    2006

  30. [38]

    James G. Wendel. A problem in geometric probability. Mathematica Scandinavica, 11:109–112, 1962

    1962

  31. [39]

    Asymptotic approximations of integrals , volume 34

    Roderick Wong. Asymptotic approximations of integrals , volume 34. SIAM, 2001. 29

    2001

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.