Abstract
We study the number of facets of the convex hull of n independent standard Gaussian points in \({\mathbb {R}}^d\). In particular, we are interested in the expected number of facets when the dimension is allowed to grow with the sample size. We establish an explicit asymptotic formula that is valid whenever \(d/n\rightarrow 0\). We also obtain the asymptotic value when d is close to n.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The convex hull \([X_1,\ldots ,X_n]\) of n independent standard Gaussian samples \(X_1,\ldots ,X_n\) from \({\mathbb {R}}^d\) is the Gaussian polytope \(P^{(d)}_n\). For fixed dimension d, the face numbers and intrinsic volumes of \(P_n^{(d)}\) as n tends to infinity are well understood by now. For \(i=0\ldots ,d\) and polytope Q, let \(f_i(Q)\) denote the number of i-faces of Q and let \(V_i(Q)\) denote the ith intrinsic volume of Q. The asymptotic behavior of the expected value of the number of facets \(f_{d-1}(P^{(d)}_n)\) as \(n \rightarrow \infty \) was provided by Rényi and Sulanke [22] if \(d=2\), and by Raynaud [21] if \(d\ge 3\). Namely, they proved that, for any fixed d,
as \(n \rightarrow \infty \). For \(i=0,\ldots ,d\), expected value of \(V_i(P^{(d)}_n)\) as \(n \rightarrow \infty \) was computed by Affentranger [1], and that of \(f_i(P^{(d)}_n)\) was determined Affentranger and Schneider [2] and Baryshnikov and Vitale [3], see Hug et al. [15] and Fleury [12] for a different approach. More recently, Kabluchko and Zaporozhets [18, 19] proved explicit expressions for the expected value of \(V_d(P^{(d)}_n)\) and the number of k-faces \(f_k(P^{(d)}_n)\). Yet these formulas are complicated and it is not immediate how to deduce asymptotic results for large n high dimensions d.
After various partial results, including the variance estimates of Calka and Yukich [6] and Hug and Reitzner [16], central limit theorems were proved for \(f_i(P^{(d)}_n)\) and \(V_d(P^{(d)}_n)\) by Bárány and Vu [5], and for \(V_i(P^{(d)}_n)\) by Bárány and Thäle [4]. These results have been strengthened considerably by Grote and Thäle [14]. The interesting question whether \({\mathbb {E}}f_{d-1}(P^{(d)}_n)\) is an increasing function in n was answered in the positive by Kabluchko and Thäle [17]. It would be interesting to investigate the monotonicity behavior of the facet number if n and d increases simultaneously.
The “high-dimensional” regime, that is, when d is allowed to grow with n, is of interest in numerous applications in statistics, signal processing, and information theory. The combinatorial structure of \(P^{(d)}_n\), when d tends to infinity and n grows proportionally with d, was first investigated by Vershik and Sporyshev [23], and later Donoho and Tanner [11] provided a satisfactory description. For any \(t>1\), Donoho and Tanner [11] determined the optimal \(\varrho (t)\in (0,1)\) such that if n/d tends to t, then \(P^{(d)}_n\) is essentially \(\varrho (t)d\)-neighbourly (if \(0<\eta <\varrho (t)\) and \(0\le k\le \eta d\), then \(f_k(P^{(d)}_n)\) is asymptotically \({n \atopwithdelims ()k+1}\)). See Donoho [10], Candés et al. [7], Candés and Tao [8, 9], Mendoza-Smith et al. [20].
In this note, we consider \(f_{d-1}(P^{(d)}_n)\), the number of facets, when both d and n tend to infinity. Our main result is the following estimate for the expected number of facets of the Gaussian polytope. The implied constant in \(O(\cdot )\) is always some absolute constant. We write \({{\,\mathrm{{lln}}\,}}x\) for \(\ln (\ln x)\).
Theorem 1.1
Assume \(P^{(d)}_n\) is a Gaussian polytope. Then for \(d \ge 78\) and \(n \ge e^e d\), we have
with \(\theta =\theta (n,d) \in [-34,2]\).
When n/d tends to infinity as \(d \rightarrow \infty \), Theorem 1.1 provides the asymptotic formula
If \(n/(d e^d)\rightarrow \infty \), then we have \(\frac{d}{\ln \frac{n}{d}} \rightarrow 0 \) and hence
as \(d \rightarrow \infty \). In the case when n grows even faster such that \((\ln n)/(d\ln d) \rightarrow \infty \), the asymptotic formula simplifies to the result (1) of Rényi and Sulanke [22] and Raynaud [21] for fixed dimension.
Corollary 1.2
Assume \(P^{(d)}_n\) is a Gaussian polytope. If \((\ln n)/(d\ln d) \rightarrow \infty \), we have
There is a (simpler) counterpart of our main results stating the asymptotic behavior of the expected number of facets of \(P_n^{(d)}\), if \(n-d\) is small compared to d, that is, if n/d tends to one.
Theorem 1.3
Assume \(P^{(d)}_n\) is a Gaussian polytope. Then for \(n - d = o( d)\), we have
as \(d \rightarrow \infty \).
This complements a result of Affentranger and Schneider [2] stating the number of k-dimensional faces for \(k \le n-d\) and \(n-d\) fixed,
as \(d \rightarrow \infty \).
In the next section we sketch the basic idea of our approach, leaving the technical details to later sections. In Sect. 3 we provide asymptotic approximations for the tail of the normal distribution. In Sect. 4 concentration inequalities are derived for the \(\beta \)-distribution. Finally, in Sects. 5 and 6, Corollary 1.2 and Theorem 1.3 are proven.
2 Outline of the Argument
For \(z\in {\mathbb {R}}\), let
Our proof is based on the approach of Hug, Munsonius, and Reitzner [15]. In particular, [15, Theorem 3.2] states that if \(n\ge d+1\) and \(X_1,\ldots ,X_n\) are independent standard Gaussian points in \({\mathbb {R}}^d\), then
where \(Y,Y_{1},\dots ,Y_{n-d}\) are independent real-valued random variables with \(Y {\mathop {=}\limits ^{d}} N\left( 0,\frac{1}{2d}\right) \) and \(Y_{i} {\mathop {=}\limits ^{d}} N\left( 0,\frac{1}{2}\right) \) for \(i=1,\ldots ,n-d\). This gives
Note that similar integrals appear in the analysis of the expected number of k-faces for values of k in the entire range \(k=0,\ldots ,d-1\). In our case, the analysis boils down to understanding the integral of \(\Phi (y)^{n-d} {\phi }(y)^d\) over the real line. By substituting \( (1-u) = \Phi (y)\), we obtain
Clearly, \(n \ge d+2\) is the nontrivial range. When \(n/d \rightarrow \infty \), \((1-u)^{n-d}\) is dominating, and we need to investigate the asymptotic behavior of \({\phi }(\Phi ^{-1}(1-u))\) as \(u \rightarrow 0\). We show that the essential term is precisely 2u. Hence, it makes sense to rewrite the integral as
For \(x,y>0\), the Beta-function is given by \({\varvec{B}}(x,y)= \int _0^1 (1-u)^{x-1} u^{y-1} du \). It is well known that for \(k,l \in {\mathbb {N}}\) we have \({\varvec{B}}(k,l)= \frac{(k-1)!(l-1)!}{(k+l-1)!}\). A random variable U is \({\varvec{B}}_{(x,y)}\) distributed if its density is given by \({\varvec{B}}(x,y)^{-1} (1-u)^{x-1} u^{y-1}\). With this, we have established the following identity:
Proposition 2.1
where
and U is a \({\varvec{B}}(n-d+1,d)\) random variable.
In Lemma 3.3 below we show that
as \(u \rightarrow 0\). Because the Beta function is concentrated around \(\frac{d}{n}\), see Lemma 4.1 and Lemma 4.2, this yields
which implies our main result.
3 Asymptotics of the \(\Phi \)-Function
To estimate \(\Phi (z)\), we need a version of Gordon’s inequality [13] for the Mill’s ratio:
Lemma 3.1
For any \(z>1\) there exists \(\theta \in (0,1)\), such that
Proof
It follows by partial integration that
which yields the lemma. \(\square \)
Lemma 3.2
For any \(u \in (0,e^{-1}]\) there is a \({\delta }={\delta }(u) \in (0,16)\) such that
Proof
It is useful to prove (5) for the transformed variable \(u=e^{-t}\). We define
which exists for \(t > 0\). In a first step we prove that this is the asymptotic expansion of \(z=\Phi ^{-1}(1-e^{-t})\) as \(z,t \rightarrow \infty \) with a suitable function \({\delta }={\delta }(t)=O(1)\). In a second step we show the bound on \(\delta \). Observe that \(z \ge 1\) implies \(t \ge \ln \Phi (-1))=-2,54\dots \). By Lemma 3.1, for \( z \ge 1\)
as \(z \rightarrow \infty \) with some \(\theta (z) \in (0,1)\), which immediately implies that \( z=z(t) \rightarrow \infty \) as \( t \rightarrow \infty \). Equation (7) shows that \(e^t \ge 2 \sqrt{\pi }z e^{z^2}\) and thus
for \(z \ge 1\). The function \(z=z(t)\) is the inverse function we are looking for, if it satisfies
We plug (6) into this equation. This leads to
and shows \( - \ln (2\sqrt{\pi }) + o(1)=- 2 \delta (t) - O(1) \). Thus the function z(t) given by (6) in fact satisfies (7) and therefore it is the asymptotic expansion of the inverse function.
The desired estimate for \({\delta }\) follows from some more elaborate but elementary calculations. First we prove that \({\delta }\ge 0\). By (8) and because \(e^x \ge 1+x\),
which is equivalent to
for \(t \ge 1\). On the other hand, again by (8),
and using \(e^x \ge 1+x\) implies
\(\square \)
An asymptotic expansion for \({\phi }(\Phi ^{-1} (1-u))\) follows immediately:
Lemma 3.3
For any \(u \in (0,e^{-1}]\) there is a \({\delta }={\delta }(u) \in (0,16)\) such that
4 Concentration of the \(\beta \)-Distribution
A basic integral for us is the Beta-integral
Let \(U \sim {\varvec{B}}({\alpha }, {\beta })\) distributed. Then \({\mathbb {E}}U=\frac{{\beta }}{{\alpha }+{\beta }}\) and \(\textrm{var}(U)= \frac{{\alpha }{\beta }}{({\alpha }+{\beta })^2 ({\alpha }+{\beta }+1)}\) Next we establish concentration inequalities for a Beta-distributed random variable around its mean. Observe that if \(U \sim \textbf{B}({\alpha },{\beta })\), then \(1-U \sim \textbf{B}({\beta }, {\alpha })\). Hence we may concentrate on the case \({\alpha }\ge {\beta }\).
Lemma 4.1
Let \(U \sim {\varvec{B}}(a+1, b+1)\) distributed with \(a \ge b\) and set \(n=a+b\). Then
Proof
We have to estimate the integral
For an estimate from above we substitute \(x= \frac{b}{n} - \frac{y}{n} \sqrt{\frac{ab}{n}}\).
It is well known that
for \(x \in (-1,1]\). Since \(a \ge b\), we have
which implies
In the last step we use Stirling’s formula,
to see that
\(\square \)
Lemma 4.2
Let \(U \sim {\varvec{B}}(a+1, b+1)\) distributed with \(a \ge b\) and set \(n=a+b\). Then for \({\lambda }\ge 2\),
Proof
We assume that \(a \ge b\) and thus \(a \ge \frac{n}{2}\). We have to estimate the probability
We substitute \(x \rightarrow \frac{1}{a} x + {\lambda }\frac{b}{n}\) and obtain
The use of the binomial formula and the Gamma functions yields
because \( b \le {\lambda }\frac{ab}{n}\) for \(a \ge \frac{n}{2} \ge b\) and \({\lambda }\ge 2\), and \(\frac{1}{k!} \left( {\lambda }\frac{ab}{n}\right) ^k \) is increasing for \( k\le \left( {\lambda }\frac{ab}{n}\right) \). Using (11) this gives
and with \((1+x) \le e^x\) the lemma. \(\square \)
5 The Case \(n-d \) Large
In this section we combine Lemma 3.3 which gives the asymptotic behavior of \( g_d(u) \) as \(u \rightarrow 0\), with the concentration properties of the Beta function just obtained. We split our proof in two Lemmata.
Lemma 5.1
For \(d \ge d_0=78\) and \(n \ge e^e d\) we have
Lemma 5.2
For \(d \ge d_0=78\) and \(n \ge e^e d\) we have
These two bounds prove Theorem 1.1. The idea is to split the expectation into the main term close to \(\frac{d}{n}\) and two error terms,
Proof of Lemma 5.2
Recall that U is \({\varvec{B}}(n-d+1,d)\)-distributed. Lemma 4.2 with \(a=n -d \) and \(b= d-1 \) shows that
because \( \frac{d-1}{n-1}< \frac{d}{n}\). For \({\lambda }=2\) this gives
for \(n \ge 10 d\). The probability that U is small is estimated by Lemma 4.1 with \(s= (1- e^{-2})\sqrt{\frac{(d-1)(n-1)}{n-d}}\),
for \(d \ge 6\). Combining both estimates and using
for \(x \in [0, \frac{1}{2}]\), we have
for \(d \ge d_0=78\). (Observe that \( \frac{2e^6}{\pi } \sqrt{d}_0 e^{ - \frac{1}{10} d_0} \le \frac{1}{2} \).) In the last step we compute
for \(n \ge e^e d\). Here, note that \(\frac{{{\,\mathrm{{lln}}\,}}x}{\ln x}\) is decreasing for \(x \ge e^e\). Now using
and
for \(n \ge e^{e} d \), we have
with \(\delta '=\frac{3 \ln 2}{2}+2 \max \delta \in [0, 34]\). Combining this estimate with (14) we obtain
for \(d \ge d_0\) and \(n \ge e^e d\).
Proof of Lemma 5.1
As an upper bound we have
since \(\delta \ge 0\), and where the last term follows from (12). For the first term we use that \({\phi }(\Phi ^{-1}(\cdot ))\) is a symmetric and concave function and thus increasing on \([0, e^{-2}\frac{d}{ n}]\), and that \({\delta }\ge 0\).
Now the remaining integration is trivial. We use Stirling’s formula (11) to estimate the Beta-function and obtain
e.g. for \(n \ge e^e d\) and \(d \ge 78\). Combining our results gives
In a similar way as above, we get rid of the involved constant \(e^2\) by using
and
This yields
\(\square \)
6 The Case \(n-d\) Small
Finally, it remains to prove Theorem 1.3. The starting point here is again formula (2), together with the substitution \(y \rightarrow \frac{y}{\sqrt{d}}\).
The Taylor expansion of \( \Phi \) at \(y=0\) is given by
with some \( \theta _1, \theta _2 \in {\mathbb {R}}\) depending on y. Since \(\Phi (y)\) is above its tangent at 0 for \(y>0\) and below it for \(y<0\), we have \(0 \le 1- \theta _2 y \le 1\). Further,
Hence an expression for \(\ln \Phi \) at \(y=0\) is given by
We need again estimates for the logarithm, namely \(\ln (1+x) = x - \theta _3 x^2 < x\) with some \(\theta _3 = \theta _3 (x) \ge 0\). In addition, there exists \(c_3 \in {\mathbb {R}}\) such that \(\theta _3 < c_3 \) if x is bounded away from \(-1\), for example, for \(x \ge 2 \Phi (-1)-1\). This gives
and
with \(\theta _3 < c_3 \) for \(y \ge -1\). Thus the Taylor expansion of \(\ln \Phi \) at \(y=0\) is given by
with some \( \theta _4= \theta _4(y) > - \frac{1}{2}\), and there exists a \(c_4 \in {\mathbb {R}}\) with \(\theta _4 \le c_4\) for \(y \ge -1\). We plug this into (16) and obtain
Since \(\frac{n-d}{d} \rightarrow 0\) we assume that \( 1+\theta _4 \frac{n-d}{d} \ge 1- \frac{1}{2} \frac{n-d}{d} > 0.\) As an estimate from above we have
The estimate from below is slightly more complicated. For \(y \ge -\sqrt{d}\) there is an upper bound \(c_4\) for \(\theta _4\). Using this we have
Now we use \((a-b)^2 \le 2a^2 + 2b^2\) which shows that
Recall the estimate for \(\Phi (z) \) from Lemma 3.1,
We combine Eqs. (17), (18) and (19) and obtain
which yields Theorem 1.3.
References
Affentranger, F.: The convex hull of random points with spherically symmetric distributions. Rend. Sem. Mat. Univ. Politec. Torino 49, 359–383 (1991)
Affentranger, F., Schneider, R.: Random projections of regular simplices. Discrete Comput. Geom. 7, 219–226 (1992)
Baryshnikov, Y.M., Vitale, R.A.: Regular simplices and Gaussian samples. Discrete Comput. Geom. 11, 141–147 (1994)
Bárány, I., Thäle, C.: Intrinsic volumes and Gaussian polytopes: the missing piece of the jigsaw. Doc. Math. 22, 1323–1335 (2017)
Bárány, I., Vu, V.H.: Central limit theorems for Gaussian polytopes. Ann. Probab. 35, 1593–1621 (2007)
Calka, P., Yukich, J.E.: Variance asymptotics and scaling limits for Gaussian polytopes. Probab. Theory Relat. Fields 163, 259–301 (2015)
Candés, E.J., Romberg, J., Tao, T.: Robust uncertainty principles: exact signal reconstruction from highly incomplete frequency information. IEEE Trans. Inf. Theory 52, 489–509 (2006)
Candés, E.J., Tao, T.: Decoding via linear programming. IEEE Trans. Inf. Theory 51, 4203–4215 (2005)
Candés, E.J., Tao, T.: Near optimal signal recovery from random projections and universal encoding strategies. IEEE Trans. Inf. Theory 52, 5406–5425 (2006)
Donoho, D.L.: High-dimensional centrally symmetric polytopes with neighborliness proportional to dimension. Discrete Comput. Geom. 35, 617–652 (2006)
Donoho, D.L., Tanner, J.: Counting faces of randomly projected polytopes when the projection radically lowers dimension. J. Am. Math. Soc. 22, 1–53 (2009)
Fleury, B.: Poincaré inequality in mean value for Gaussian polytopes. Probab. Theory Relat. Fields 152, 141–178 (2012)
Gordon, R.D.: Values of Mills’ ratio of area to bounding ordinate and of the normal probability integral for large values of the argument. Ann. Math. Stat. 12, 364–366 (1941)
Grote, J., Thäle, C.: Gaussian polytopes: a cumulant-based approach. J. Complex. 47, 1–41 (2018)
Hug, D., Munsonius, G.O., Reitzner, M.: Asymptotic mean values of Gaussian polytopes. Beitr. Algebra Geom. 45, 531–548 (2004)
Hug, D., Reitzner, M.: Gaussian polytopes: variances and limit theorems. Adv. Appl. Probab. 37, 297–320 (2005)
Kabluchko, Z., Thäle, C.: Monotonicity of expected \(f\)-vectors for projections of regular polytopes. Proc. Am. Math. Soc. 146, 1295–1303 (2018)
Kabluchko, Z., Zaporozhets, D.: Absorption probabilities for Gaussian polytopes and regular spherical simplices. Adv. Appl. Probab. 52, 588–616 (2020)
Kabluchko, Z., Zaporozhets, D.: Expected volumes of Gaussian polytopes, external angles, and multiple order statistics. Trans. Am. Math. Soc. 372, 1709–1733 (2019)
Mendoza-Smith, R., Tanner, J., Wechsung, F.: A robust parallel algorithm for combinatorial compressed sensing. IEEE Trans. Signal Process. 66, 2167–2177 (2018)
Raynaud, H.: Sur lénveloppe convexe des nuages de points aléatoires dans \({\mathbb{R} }^n\). J. Appl. Probab. 7, 35–48 (1970)
Rényi, A., Sulanke, R.: Über die konvexe Hülle von \(n\) zufällig gewählten Punkten. Z. Wahrsch. Verw. Gebiete 2, 75–84 (1963)
Vershik, A.M., Sporyshev, P.V.: Asymptotic behavior of the number of faces of random polyhedra and the neighborliness problem. Sel. Math. Soviet. 11, 181–201 (1992)
Acknowledgements
We would like to thank Imre Bárány and Gergő Nemes for fruitful discussions.
Funding
Open access funding provided by ELKH Alfréd Rényi Institute of Mathematics.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
K. Böröczky research was supported by Hungarian Grant NKFIH 132002. G. Lugosi research was supported by the Spanish grant PID2022-138268NB-I00, financed by MCIN/AEI/10.13039/501100011033, FSE+MTM2015-67304-P, and FEDER, EU. M. Reitzner research was supported by the German Research Foundation DFG-GRK 1916.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Böröczky, K.J., Lugosi, G. & Reitzner, M. Facets of High-Dimensional Gaussian Polytopes. J Geom Anal 34, 69 (2024). https://doi.org/10.1007/s12220-023-01440-5
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s12220-023-01440-5