Return to Article Details Nonlinear random extrapolation estimates of π under Dirichlet distributions

Nonlinear Random Extrapolation Estimates of π under Dirichlet Distributions

Shasha Wang, Zecheng Li\ddag Wen-Qing Xu

September 17, 2023; accepted: December 9, 2023; published online: December 22, 2023.

Department of Mathematics and Physics, Shijiazhuang Tiedao University, Shijiazhuang, Hebei 050043, China, e-mail: wshasha@stdu.edu.cn.
\ddagQuestrom School of Business, Boston University, Boston, MA 02215, USA, e-mail: lizc0816@bu.edu
Beijing Institute for Scientific and Engineering Computing, Beijing University of Technology, Beijing 100124, China. Department of Mathematics and Statistics, California State University Long Beach, CA 90840, USA, e-mail: wxu@csulb.edu (corresponding author).

We construct optimal nonlinear extrapolation estimates of π based on random cyclic polygons generated from symmetric Dirichlet distributions. While the semiperimeter Sn and the area An of such random inscribed polygons and the semiperimeter (and area) Sn of the corresponding random circumscribing polygons are known to converge to π w.p.1 and their distributions are also asymptotically normal as n, we study in this paper nonlinear extrapolations of the forms Wn=SnαAnβSnγ and Wn(p)=(αSnp+βAnp+γSnp)1/p where α+β+γ=1 and p0. By deriving probabilistic asymptotic expansions with carefully controlled error estimates, we show that Wn and Wn(p) also converge to π w.p.1 and are asymptotically normal. Furthermore, to minimize the approximation error associated with Wn and Wn(p), the parameters must satisfy the optimality condition α+4β2γ=0. Our results generalize previous work on nonlinear extrapolations of π which employ inscribed polygons only and the vertices are also assumed to be independently and uniformly distributed on the unit circle.

MSC. Primary 65C20, 60F05; Secondary 05C80, 60D05.

Keywords. Random polygons; Nonlinear extrapolations; Dirichlet distribution; Central limit theorems; Cramér’s theorem.

1 Introduction

Given a convex set KRd, the stochastic properties of the convex hull Kn generated by n independent random points on K, such as the area, volume and number of vertices of Kn, their probability distributions and asymptotic behavior have attracted extensive attention (see, e.g., [ 7 , 10 , 11 , 16 , 17 , 20 ] ). In the case of n points randomly selected on a unit circle in R2, the resulting convex hull is a random n-gon inscribed in the circle which is obtained by connecting all adjacent vertices on the circle. Using the same set of random points, one may also construct (w.p.1) a random circumscribing n-gon which is tangent to the circle at each of the n random points. In the simplest case when the vertices are independent and uniformly distributed on the circle, it is known that the semiperimeter Sn and area An of such random inscribed polygons, and the semiperimeter (or area) Sn of the random circumscribing polygons all converge to π w.p.1 as n and their distributions are also asymptotically Gaussian [ 4 , 24 ] . Furthermore, by using extrapolation techniques [ 12 , 15 ] originating exactly from the famous Archimedean approximations of π based on regular polygons [ 3 , 13 , 19 ] , it has been shown [ 22 , 23 , 25 ] that simple weighted averages such as 43Sn13An, 23Sn+13Sn and 1615Sn15An+215Sn, etc., provide much more accurate approximations of π, and at the same time also satisfy similar central limit theorems as n. We note that extrapolation methods are useful in many important applications such as numerical evaluation of integrals, numerical solution of differential equations, and polynomial interpolations, etc. To accelerate the convergence associated with existing low-precision approximations, extrapolation seeks to combine them in a way such that the leading order error terms are cancelled out as much as possible. For example, in the simpler case of the Archimedean approximation of π, while both Sn=nsin(π/n) and An=12nsin(2π/n)=Sn/2 converge to π with errors of order O(n2), a closer look reveals that Snππ36n2 and Anπ2π33n2. This implies the error of An roughly quadruples that of Sn. Consequently, the weighted average 43Sn13An, or equivalently, 43Sn13Sn/2, exactly cancels out the leading order error terms in Sn and An to yield an improved estimate of π with a reduced error now of order O(n4).

More recently, in [ 26 ] , the authors have initiated the study of novel nonlinear extrapolation estimates of π in the forms Xn=SnαAnβ and Yn(p)=(αSnp+βAnp)1/p where α+β=1 and p0. By deriving probabilistic asymptotic expansions with carefully controlled error estimates, it is shown that, for both Xn and Yn(p), the same choice α=4/3, β=1/3 minimizes the approximation error with Xn=π+n3+δo(1), Yn(p)=π+n3+δo(1) where δ>0 is any positive number and o(1) represents a random variable which converges to 0 w.p.1 as n. Furthermore, Xn and Yn(p) are also asymptotically normal with XnAN(π2π5/n4,2496π10/n9), Yn(p)AN(π2(p+1)π5/n4,(160p2+960p+2496)π10/n9) where for a sequence of random variables {Zn} and μnR, σn>0, the notation ZnAN(μn,σn2) means (Znμn)/σnLN(0,1). In particular, for p=1, Yn(1) reduces to the optimal linear extrapolation estimate 43Sn13An. Moreover, Xn may be viewed as the limit of Yn(p) when p0, a reflection of the relation limp0(αxp+βyp)1/p=xαyβ for any x,y>0 and α+β=1.

In this paper, we aim to further develop nonlinear random extrapolation methods for approximating π. On the one hand, it would be natural to include also random circumscribing polygons in the approximation process. Motivated by the work in [ 26 ] , we study nonlinear functions of Sn, An and Sn in the forms Wn=SnαAnβSnγ, Wn(p)=(αSnp+βAnp+γSnp)1/p where α+β+γ=1 and p0. On the other hand, we are also interested in extending the theory to more general random cyclic polygons whose vertices are not independently and uniformly distributed on the circle. While this is a very challenging problem in general, as a first step, we focus on the particular case of random cyclic polygons generated from symmetric Dirichlet distributions with an arbitrary concentration parameter a>0. We note that in such cases, it has been proved [ 21 ] that the respective semiperimeters and areas, again denoted by Sn, An and Sn, satisfy similar convergence estimates and central limit theorems as in [ 4 , 24 ] for the uniform case, which in fact corresponds to the special case a=1 of the Dirichlet distribution.

Clearly, as in [ 26 ] , the case p=1 reduces to linear extrapolations based on Sn, An and Sn, and due to the relation limp0(αxp+βyp+γzp)1/p=xαyβzγ, we expect to recover Wn from Wn(p) in the limit p0. More importantly, based on similar asymptotic expansion results established in [ 21 ] for Sn, An and Sn in terms of various power sums of the underlying Dirichlet distribution, we derive rigorous probabilistic asymptotic expansions with carefully controlled error estimates for various nonlinear functions of Sn, An and Sn, particularly Wn and Wn(p). Such probabilistic asymptotic expansions resemble the well-known Taylor series expansions in many deterministic approximation problems and provide a cornerstone for establishing the corresponding probability convergence estimates and central limit theorems.

It turns out that, for both Wn and Wn(p), the optimal approximation occurs when α+4β2γ=0 with Wn=π+n3+δo(1) and Wn(p)=π+n3+δo(1). Such results are comparable with those obtained in [ 26 ] for the case γ=0 and are actually weaker than the corresponding optimal linear extrapolation estimate Wn(1)=1615Sn15An+215Sn=π+n5+δo(1), see Theorems 1113 below for details. Note that together with α+β+γ=1, the condition α+4β2γ=0 implies that α=4/32γ, β=1/3+γ where γ is an arbitrary constant. Due to complicated nonlinear effects, however, the extra “free" parameter γ can no longer be used to further improve the approximation associated with Wn and Wn(p). Nevertheless, by further combining such nonlinear extrapolation estimates with different values of γ, it is possible achieve additional improvements better than the linear case.

Finally, it is interesting to note that in the case of the classical Archimedean polygons, for both Wn and Wn(p), the optimal estimates also occur when α+4β2γ=0 with Wn=π+145π5n4+O(n6), Wn(p)=π+1180π5n4[45γp10p+4]+O(n6). In fact, for Wn(p), by choosing γ=10p445p and p=p±=21±72170, we can further obtain Wn(p±)=π119±721661500π9n8+O(n10), which is two orders of magnitude higher than the optimal linear extrapolation estimate Wn(1)=1615Sn15An+215Sn=π+π7105n6+O(n8). However, for Wn, the result turns out to be completely independent of γ. This is due to the relation AnSn=Sn2, a variant of Archimedes’s celebrated geometric mean relation, which implies Wn=SnαAnβSnγ=Snα+2γAnβγ.

The remainder of the paper is organized as follows. In section 2, we present some useful preliminary results related to the Dirichlet distribution and its various power sums. section 3 is devoted to the study of nonlinear extrapolation estimates of π based on random inscribed and circumscribing polygons generated from symmetric Dirichlet distributions. Finally, we offer several additional remarks in section 4 and some concrete numerical simulation results in section 5 to conclude our study on nonlinear random extrapolation approximations.

2 PRELIMINARIES

2.1 Basic properties of Dirichlet distributions

Recall that a random vector Y=(Y1,,Yn1)Rn1, n2, is said to have Dirichlet distribution [ 2 ] with parameters a=(a1,,an1;an)Rn+ if it has joint probability density function

fY1,,Yn1(y1,,yn1)=Γ(a1++an)Γ(a1)Γ(an)y1a11yn1an11ynan1

where yi>0, i=1n1yi<1, yn=1i=1n1yi, and Γ(a)=0va1evdv is the gamma function defined for all a>0. Let Yn=1i=1n1Yi. With slight abuse of notation, we also refer to Y=(Y,Yn)Rn as Dirichlet distribution and write YDir(a), YDir(a).

In this paper, we focus on symmetric Dirichlet distributions, that is, ai=a>0 for all 1in. In such cases, all YiBeta(a,(n1)a) have identical Beta distribution.

Lemma 1 Tail probability, [ 21 ]

Let Δn=max1inYi and Zn any measurable function of Y1,Y2,,Yn. Then for any t(0,1),

  1. Pr(Δnt) decays exponentially as n.

  2. Zn1{Δnt}0 w.p.1 as n.

Lemma 2 Dirichlet integrals, [ 2 ]

Let k=(k1,,kn)Rn such that a+kj>0 for all 1jn. Then

E(j=1nYjkj)=Γ(|a|)Γ(|a+k|)j=1nΓ(a+kj)Γ(a)=Γ(na)Γ(na+j=1nkj)j=1nΓ(a+kj)Γ(a).
1

Let Dn,k=i=1nYik, kN. Then n(k1)Dn,k1 w.p.1. Furthermore, for large n, by using the above Dirichlet integrals, it is easy to compute E(Dn,k)=nγk(a)/γk(na)n(k1)mk and Var(Dn,k)n(2k1)σk2 where γk(a)=Γ(a)1Γ(a+k)=a(a+1)(a+k1), mk=akγk(a), and σk2=m2k(1+k2/a)mk2.

Lemma 3 Asymptotic convergence of Dn,k, [ 21 ]

Let kN. Then

  1. nk1δDn,k0 in probability for all δ>0 as n.

  2. nk2δDn,k0 w.p.1 for all δ>0 as n.

  3. Dn,kAN(n(k1)mk,n(2k1)σk2), that is, n(nk1Dn,kmk)LN(0,σk2) as n.

The following two lemmas provide additional asymptotic convergence results for various nonlinear expressions of Dn,k. Their proofs are slightly lengthy and are deferred to the Appendix.

Lemma 4

For any l1, k1,k2,,kl and p1,p2,,plN, then for any δ>0, it holds that

  1. n(i=1l(ki1)pi)δj=1lDn,kjpj0 in probability.

  2. n(i=1l(ki1)pi)1δj=1lDn,kjpj0 w.p.1 as n.

Lemma 5

Let α,βR, α2+β20 and Tn=αDn,32+βDn,5. Then TnAN(n4μT(α,β),n9σT2(α,β)) where μT(α,β)=αm32+βm5=a4(a+1)(a+2)[α(a+1)(a+2)+β(a+3)(a+4)] and

σT2(α,β)=4α2m32σ32+β2σ52+4αβ(m3m8(1+15/a)m32m5)=8a9(a+1)(a+2)[3α2(a+1)2(a+2)2(3a+7)+30αβ(a+1)(a+2)(a+3)2(a+4)+5β2(a+3)(a+4)(5a3+60a2+250a+363)].

Remark 6

The underlying matrix A=(aij) associated with the quadratic form in σT2(α,β) is strictly positive definite with

a11=3(a+1)2(a+2)2(3a+7),a12=a21=15(a+1)(a+2)(a+3)2(a+4),a22=5(a+3)(a+4)(5a3+60a2+250a+363),detA=15(a+1)2(a+2)2(a+3)(a+4){20a3+225a2+814a+921}>0.

This implies that σT2(α,β) is non-degenerate unless α=β=0.

2.2 Random cyclic polygons under symmetric Dirichlet distributions

The Dirichlet distribution YDir(a) is naturally associated with the (non-uniform) random division 0=X0<X1<<Xn1<Xn=1 of the unit interval where X0=0 and Xi=j=1iYj for 1in. In the special case a=1, this corresponds to the classical uniform random division [ 6 , 14 ] generated by n1 independent and uniformly distributed random points on (0,1). With the rescaling Xiθi=2πXi, this can be further mapped to a random division of the unit circle, separated by points Pi(cosθi,sinθi), 0in, in counterclockwise direction where Pn represents the same point as P0. By connecting these points consecutively, we obtain an inscribed random n-gon with its semiperimeter Sn and area An given by Sn=i=1nsinπ(XiXi1)=i=1nsinπYi, An=12i=1nsin2π(XiXi1)=12i=1nsin2πYi. Similarly, using the same random vertices Pi, we can also construct w.p.1 a circumscribing random n-gon which is tangent to the circle at each point Pi with its semiperimeter and area both given by Sn=i=1ntanπ(XiXi1)=i=1ntanπYi. Note that in the event (which has probability 0) all vertices are equally spaced, that is, Yi=1/n for all 1in, these random n-gons happen to be regular n-gons inscribed in or circumscribed about the circle with Sn=nsin(π/n), An=12nsin(2π/n)=Sn/2 and Sn=ntan(π/n). Additionally, such random cyclic polygons generated from the symmetric Dirichlet distribution YDir(a) also degenerate to regular n-gons in the limit as a.

By using the Taylor series expansion of the sine and tangent functions, it is easy to obtain, at least formally, the following probabilistic asymptotic expansions for Sn, An and Sn:

Sn=j=1(1)j1(2j1)!π2j1Dn,2j1=π13!π3Dn,3+15!π5Dn,517!π7Dn,7+,An=j=1(1)j122j2(2j1)!π2j1Dn,2j1=π43!π3Dn,3+165!π5Dn,5647!π7Dn,7+,Sn=j=1B2j(4)j(14j)(2j)!π2j1Dn,2j1=π+13π3Dn,3+215π5Dn,5+17315π7Dn,7+.

where Bj is the jth Bernoulli number. Note that by Lemma 3, the random infinitesimal terms Dn,k in the above expansions decrease progressively in order of magnitude. The validity of these asymptotic expansions is rigorously justified by the following lemma.

Lemma 7 [ 21 ]

Let m be any positive integer and δ>0. Then

Sn=j=1m(1)j1(2j1)!π2j1Dn,2j1+n(2m1)+δo(1),An=j=1m(1)j122j2(2j1)!π2j1Dn,2j1+n(2m1)+δo(1),Sn=j=1mB2j(4)j(14j)(2j)!π2j1Dn,2j1+n(2m1)+δo(1).

In particular, this implies that Sn, An and Sn all converge to π w.p.1 and their distributions are also asymptotically normal with

SnAN(π16n2m3π3,136n5σ32π6),AnAN(π23n2m3π3,49n5σ32π6),SnAN(π+13n2m3π3,19n5σ32π6)

where m3=a2(a+1)(a+2) and σ32=6a5(a+1)(a+2)(3a+7).

3 Nonlinear extrapolation estimates

3.1 Probabilistic asymptotic expansions for nonlinear functions of Sn, An and Sn

In this section, we study nonlinear random extrapolation estimates of π based on the semiperimeters and areas of both inscribed and circumscribed random polygons with an aim to construct more accurate nonlinear extrapolation estimates than in [ 26 ] . To facilitate the derivation of asymptotic expansions for nonlinear functions of Sn, An and Sn in the forms Wn=SnαAnβSnγ and Wn(p)=(αSnp+βAnp+γSnp)1/p where α+β+γ=1 and p0, we follow the development in [ 26 ] and write

Wn,1=13!Dn,3,Wn,2=12(3!)2Dn,3215!Dn,5,Wn,3=13(3!)3Dn,3313!5!Dn,3Dn,5+17!Dn,7,Un,p,1=13!(p1)Dn,3,Un,p,2=1(3!)2(p2)Dn,32+15!(p1)Dn,5,Un,p,3=1(3!)3(p3)Dn,33+23!5!(p2)Dn,3Dn,5+17!(p1)Dn,7.

Lemma 8

Let δ>0. Then it holds that

log(Sn/π)=π2Wn,1π4Wn,2π6Wn,3+n7+δo(1),log(An/π)=4π2Wn,116π4Wn,264π6Wn,3+n7+δo(1),(Sn/π)p=1Un,p,1π2+Un,p,2π4Un,p,3π6+n7+δo(1),(An/π)p=14Un,p,1π2+16Un,p,2π464Un,p,3π6+n7+δo(1).

We mention that while the analysis in [ 26 ] is carried out for uniform random divisions only, with ??, it is straightforward to verify that exactly the same asymptotic expansion results in fact extend to the case of symmetric Dirichlet distributions YDir(a) for arbitrary a>0. Note also that, in view of ??, we have, for any δ>0, Wn,1=n1+δo(1), Un,p,1=n1+δo(1), Wn,2=n3+δo(1), Un,p,2=n3+δo(1), Wn,3=n5+δo(1), Un,p,3=n5+δo(1). Thus, the above probabilistic asymptotic expansions in lemma 8 imply, log(Sn/π)=π2Wn,1+n3+δo(1)=π2Wn,1π4Wn,2+n5+δo(1) and (Sn/π)p=1Un,p,1π2+n3+δo(1)=1Un,p,1π2+Un,p,2π4+n5+δo(1).

Next, to derive asymptotic expansions for nonlinear functions of Sn such as log(Sn/π) and Snp, we apply the same Taylor series expansions of log(1+x) and (1+x)p on Sn/π1=13π2Dn,3+215π4Dn,5+17315π6Dn,7+n7+δo(1). Similar to lemma 8, we now obtain

Lemma 9

As n, it holds that, for any δ>0,

log(Sn/π)=Mn,1π2+Mn,2π4+Mn,3π6+n7+δo(1),(Sn/π)p=1+Vn,p,1π2+Vn,p,2π4+Vn,p,3π6+n7+δo(1)

where

Mn,1=13Dn,3,Mn,2=118Dn,32+215Dn,5,Mn,3=181Dn,33245Dn,3Dn,5+17315Dn,7,Vn,p,1=13(p1)Dn,3,Vn,p,2=19(p2)Dn,32+215(p1)Dn,5,Vn,p,3=127(p3)Dn,33+445(p2)Dn,3Dn,5+17315(p1)Dn,7.

Proof â–¼
Let 0<t<1/2 and τ=πt. From Lemma 1, it is clear that log(Sn/π)1{Δn>t}=nko(1) for all k0. Next we consider log(Sn/π)1{Δnt}. For Δnt, since Sn=i=1ntanπYi, by using the uniform estimate T2m1tanxT2m1+Cm,τx2m+1 for 0xτ where m1 and T2m1=j=1m(1)j14j(4j1)B2j(2j)!x2j1 is the mth Taylor polynomial of the tangent function and Cm,τ is some positive constant which depends on m and τ, we obtain
Tn,2m1SnTn,2m1+Cm,τDn,2m+1

where

Tn,2m1=j=1m(1)j14j(4j1)B2j(2j)!Dn,2j1.

With Tn,1=π and Dn,kΔnk1tk1, it is clear that we may choose t suitably small such that 0<Sn/π11/2 and 0Tn,2m1/π11/2. By the mean value theorem, we thus obtain

log(Sn/π)1{Δnt}log(Tn,2m1/π)1{Δnt}Sn/πTn,2m1/π =n(2m1)+δo(1).

We now take m=4. By inserting Tn,7/π1=13π2Dn,3+215π4Dn,5+17315π6Dn,7=n1+δo(1) into the Taylor series approximation log(1+x)=x12x2+13x3+O(1)x4 for |x|1/2 and keeping only terms at order n7+δ and below, we obtain log(Tn,7/π)1{Δnt}=13π2Dn,3+(118Dn,32+215Dn,5)π4+(181Dn,33245Dn,3Dn,5+17315Dn,7)π6+n7+δo(1)=Mn,1π2+Mn,2π4+Mn,3π6 +n7+δo(1). Consequently, we have log(Sn/π)=Mn,1π2+Mn,2π4+Mn,3π6+n7+δo(1). Similarly, we can verify (Sn/π)p=1+Vn,p,1π2+Vn,p,2π4+Vn,p,3π6+n7+δo(1).

Proof â–¼

3.2 Nonlinear extrapolations of the form Wn=SnαAnβSnγ

With the above preparations, we are now ready to derive probabilistic asymptotic expansions for the nonlinear extrapolation estimates Wn=SnαAnβSnγ and Wn(p)=(αSnp+βAnp+γSnp)1/p where α+β+γ=1 and p0. By taking the exponential of the linear combination of log(Sn/π), log(An/π) and log(Sn/π) in 0, 1 and 4, or by multiplying Snα, Anβ and Snγ directly from 2, 3, and 5, we obtain

Lemma 10

For any δ>0, it holds that

log(Wn/π)=[(α+4β)Wn,1γMn,1]π2[(α+16β)Wn,2γMn,2]π4[(α+64β)Wn,3γMn,3]π6+n7+δo(1),Wn=ππ3[(α+4β)Wn,1γMn,1]+π5{12[(α+4β)Wn,1γMn,1]2(α+16β)Wn,2+γMn,2}+π7{13![(α+4β)Wn,1γMn,1]3  +[(α+4β)Wn,1γMn,1][(α+16β)Wn,2γMn,2]  [(α+64β)Wn,3γMn,3]}+n7+δo(1),

where

(α+4β)Wn,1γMn,1=16(α+4β2γ)Dn,3,(α+16β)Wn,2γMn,2=172(α+16β+4γ)Dn,321120(α+16β+16γ)Dn,5,(α+64β)Wn,3γMn,3=1648(α+64β8γ)Dn,331720(α+64β32γ)Dn,3Dn,5+15040(α+64β272γ)Dn,7.

Let η=α+4β2γ. Then with α+β+γ=1, we may write α=432γ13η, β=13+γ+13η. Clearly if η0, we have Wn=π+n1+δo(1)=π16ηπ3Dn,3+n3+δo(1). Then by Slutsky’s theorem, it follows that Wn(p)AN(π16n2ηπ3m3,136n5η2π6σ32).

However, if η=0, that is, α=4/32γ, β=1/3+γ and γ is an arbitrary constant, it is then possible to eliminate the leading error term involving Dn,3 in 5 to obtain

Wn=π14π5[(γ2/9)Dn,32+(2/15γ)Dn,5]14536π7[14(27γ10)Dn,3363(3γ2)Dn,3Dn,59(21γ+2)Dn,7]+n7+δo(1).

In particular, this implies Wn=π14π5[(γ29)Dn,32+(215γ)Dn,5]+n5+δo(1). By lemma 5 and remark 6, it is clear that Tn=Tn(γ29,215γ)=(γ29)Dn,32+(215γ)Dn,5 is nondegenerate for all γR and is asymptotically normal with TnAN(n4μT(γ29,215γ),n9σT2(γ29,215γ)).

By further applying Slutsky’s theorem, a related central limit theorem can be established for the optimal nonlinear extrapolation estimate Wn= Sn4/32γAn1/3+γSnγ. Note that in such cases, it is impossible to take advantage of γ to further eliminate the leading order error term in 6 to achieve Wn=π+n5+δo(1).

Theorem 11
  1. If η=0, then Wn=π+n3+δo(1) for any δ>0,
    \noindentWnAN(ππ54n4μT(γ29,215γ),π1016n9σT2(γ29,215γ)).

  2. If η0, then Wn=π+n1+δo(1) for any δ>0 and WnAN(π16n2ηπ3m3,136n5η2π6σ32).

3.3 Nonlinear extrapolations of the form Wn(p)=(αSnp+βAnp+γSnp)1/p

Next, we consider Wn(p)=(αSnp+βAnp+γSnp)1/p. By taking the linear combination of (Sn/π)p, (An/π)p and (Sn/π)p in 2, 3 and 5, and applying Newton’s generalized binomial formula for (1+x)1/p, we derive

Lemma 12

For any δ>0, it holds that

(Wn(p)/π)p=1[(α+4β)Un,p,1γVn,p,1]π2+[(α+16β)Un,p,2+γVn,p,2]π4[(α+64β)Un,p,3γVn,p,3]π6+n7+δo(1),Wn(p)=ππ3(1/p1)[(α+4β)Un,p,1γVn,p,1]+π5{(1/p2)[(α+4β)Un,p,1γVn,p,1]2+(1/p1)[(α+16β)Un,p,2+γVn,p,2]}+π7{(1/p3)[(α+4β)Un,p,1+γVn,p,1]3+2(1/p2)[(α+4β)Un,p,1+γVn,p,1][(α+16β)Un,p,2+γVn,p,2](1/p1)[(α+64β)Un,p,3γVn,p,3]}+n7+δo(1),

where

(α+4β)Un,p,1γVn,p,1=16(p1)(α+4β2γ)Dn,3,(α+16β)Un,p,2+γVn,p,2=136(p2)(α +16β+4γ)Dn,32+1120(p1)(α+16β+16γ)Dn,5,(α+64β)Un,p,3γVn,p,3=1216(p3)(α+64β8γ)Dn,33   +1360(p2)(α+64β32γ)Dn,3Dn,5   +15040(p1)(α+64β272γ)Dn,7.

Again, let η=α+4β2γ so that α=432γ13η, β=13+γ+13η. Thus as in the case of Wn, if η0, then Wn(p)=π16ηπ3Dn,3+n3+δo(1) and Wn(p)AN(π16n2ηπ3m3,136n5η2π6σ32). However, if η=0, that is, α=4/32γ, β=1/3+γ, a further improvement as in theorem 11 is possible. In such cases, we have

Wn(p)=π+π54[(p1)(γ2/9)Dn,32+(γ2/15)Dn,5]π74536[7(p1)(p2)(27γ10)Dn,33+63(p1)(3γ2)Dn,3Dn,59(21γ+2)Dn,7]+n9+δo(1).

Theorem 13
  1. If η=0, then Wn(p)=π+n3+δo(1) for any δ>0 and Wn(p)AN(π+14n4π5μT((p1)(γ29),γ215),
    116n9π10σT2((p1)(γ29),γ215)).

  2. If η0, then Wn(p)=π+n1+δo(1) for any δ>0 and Wn(p)AN(π16n2ηπ3m3,136n5η2π6σ32).

Remark 14

The asymptotic estimates for Wn in theorem 11 can be recovered from those for Wn(p) in theorem 13 by setting p=0.

Remark 15

When p=1, we may choose γ=215 to further eliminate the leading order error term in -1. This yields α=1615, β=15 and the optimal linear extrapolation estimate Wn(1)=1615Sn15An+215Sn=π+n5+δo(1) in [ 21 ] , which satisfies Wn(1)AN(π+1105n6π7m7,111025n13π14σ72). However this is not possible if p1 since, as in the case of Wn, that would require γ=29 and γ=215 simultaneously.

4 Additional Remarks

We offer some additional remarks to conclude our study on the nonlinear random extrapolation estimates Wn=SnαAnβSnγ and Wn(p)=(αSnp+βAnp+γSnp)1/p. For brevity, we address for both Wn and Wn(p) the optimal case only with α+4β2γ=0, that is, α=4/32γ, β=1/3+γ.

4.1 Special cases of α=0, β=0, or γ=0

Note that for γ=0, γ=1/3 and γ=2/3, Wn reduces to Xn=Sn4/3An1/3, Yn=Sn2/3Sn1/3 and Zn=An1/3Sn2/3 respectively with Xn=π+π590(5Dn,323Dn,5)+n5+δo(1), Yn=ππ5180(5Dn,329Dn,5)+n5+δo(1), Zn=ππ545(5Dn,326Dn,5)+n5+δo(1).

Corollary 16

For any δ>0, it holds that

  1. Xn=π+n3+δo(1),

    XnAN(π+190n4π5μT(5,3),18100n9π10σT2(5,3)).

  2. Yn=π+n3+δo(1),

    YnAN(π1180n4π5μT(5,9),132400n9π10σT2(5,9)).

  3. Zn=π+n3+δo(1),

    ZnAN(π145n4π5μT(5,6),12025n9π10σT2(5,6)).

Similarly, for Wn(p), the same choices γ=0, γ=1/3, γ=2/3 yield Xn(p)=(43Snp13Anp)1/p=π190π5[5(p1)Dn,32+3Dn,5]+n5+δo(1), Yn(p)=(23Snp+13Snp)1/p=π+1180π5[5(p1)Dn,32+9Dn,5]+n5+δo(1), Zn(p)=(13Anp+23Snp)1/p=π+145π5[5(p1)Dn,32+6Dn,5]+n5+δo(1).

Corollary 17

For any δ>0, it holds that

  1. Xn(p)=π+n3+δo(1),

    Xn(p)AN(ππ590n4μT(5(p1),3),π108100n9σT2(5(p1),3)).

  2. Yn(p)=π+n3+δo(1),

    Yn(p)AN(π+π5180n4μT(5(p1),9),π1032400n9σT2(5(p1),9)).

  3. Zn(p)=π+n3+δo(1),

    Zn(p)AN(π+π545n4μT(5(p1),6),π102025n9σT2(5(p1),6)).

4.2 Uniform spacings

In the special case of a=1, the symmetric Dirichlet distribution corresponds to the uniform spacings generated by n1 independent and uniformly distributed random points on the unit interval. Thus, by setting a=1, we immediately obtain optimal nonlinear extrapolation estimates for random polygons generated by independent and uniformly distributed random points on the unit circle.

Theorem 18 Uniform spacings
  1. Wn=π+n3+δo(1), Wn(p)=π+n3+δo(1) for any δ>0.

  2. WnAN(π+n4π5(21γ2),48n9π10(3405γ2840γ+52)).

  3. Wn(p)AN(π+n4π5μp,γ,n9π10σp,γ2) where μp,γ=(9p+21)γ2(p+1), σp,γ2=116σT2((p1)(γ2/9),γ2/15)=360(9p2+102p+454)γ2480(3p2+26p+84)γ+32(5p2+30p+78).

  4. For p=1 and γ=2/15, Wn(1)=1615Sn15An+215Sn satisfies Wn(1)=π+1105π7Dn,7+n7+δo(1)=π+n5+δo(1), Wn(1)AN(π+48n6π7,7792128n13π14).

4.3 Regular polygons

Finally, we remark that in the case of regular polygons, with Yi=1/n and Dn,k=n(k1), it is straightforward to check that the optimal estimate for both Wn and Wn(p) occurs also when α+4β2γ=0 with

Wn=π+145π5n4+4567π7n6+1405π9n8+O(n10),Wn(p)=π+π5180n4[45γp10p+4]π74536n6[189p(p2)γ70p2+84p32]π925920n8[810p2(p1)γ245(13p220p+20)γ +2(55p3120p2+100p32)]+O(n10).

Note that the result for Wn actually does not depend on the parameter γ at all. This is due to the fact AnSn=Sn2, a variant of Archimedes’s celebrated geometric mean relation, which implies Wn=SnαAnβSnγ=Snα+2γAnβγ. In particular, in such cases, Xn=Sn4/3An1/3, Yn=An2/3Sn1/3 and Zn=An1/3Sn2/3 all yield exactly the same result. While this relation no longer holds for random polygons, it helps explain why throwing in an extra term does not always increase the accuracy of the extrapolation estimates.

For Wn(p), however, by taking γ=10p445p, it is possible to eliminate the leading error term O(n4) to obtain Wn(p)=π+15670π7n6(35p2+21p2)132400π9n8(25p375p252p+12)+O(n10). Thus if we choose p±=21±72170, we can further obtain Wn(p±)=π119±721661500π9n8+O(n10), which is two orders of magnitude higher than the optimal linear estimate Wn(1)=1615Sn15An+215Sn=π+1105π7n6+1360π9n8+O(n10).

5 Numerical simulations

In this section, we present numerical simulation results to confirm the main probabilistic convergence estimates obtained in this paper. For this purpose, we use the MATLAB command gamrnd(a,1) to first generate n independent gamma random variables V=(V1,V2,Vn) with shape parameter a>0. The symmetric Dirichlet random vector YDir(a1) is then obtained by normalization Y=V/V1. Next, we compute Sn=i=1nsinπYi, An=12i=1nsin2πYi, Sn=i=1ntanπYi, and subsequently, Wn=SnαAnβSnγ, Wn(p)=(αSnp+βAnp+γSnp)1/p for n=96×2k, k=4,5,6,7. For simplicity, we consider two p values: p=2, p=1 and choose α=16/15, β=1/5, γ=2/15 which clearly satisfies the optimality condition η=α+4β2γ=0. We repeat these simulations for m=100,000 times. For each of these random samples, we compute its empirical mean μ^(Xn) and empirical standard deviation σ^(Xn) with normalization X~n=(Xnμ^(Xn))/σ^(Xn). The histograms (with bin size 400) for the empirical PDFs of normalized Sn, An, Sn, Wn, Wn(p) are displayed in figure 1 and figure 2 below for specified parameter values.

To effectively compare the empirical data with their theoretical asymptotic values, suitable scaling factors are used for μ^(Xn) and σ^(Xn) in the tables below. For Wn and Wn(p), by ??, it is clear that μ^(Xn)=π+O(1)n4π5, and σ^(Xn)=O(1)n9/2π5. Thus, for easy numerical comparison, we display μ^n=n4π5(μ^(Xn)π) and σ^n=n9/2π5σ^(Xn) instead, together with similarly scaled limiting values. For Sn, An and Sn, the scaling factors for the mean and standard deviation are n2π3 and n5/2π3 respectively.

Finally, we note that for p=1, the optimal linear extrapolation Wn(1)=1615Sn15An+215Sn converges most rapidly with an asymptotical mean μ^(Wn(1)) =π+O(1)n6 and standard deviation σ^(Wn(1))=O(1)n13/2. At such “atomic" scales, however, for n in the range 103104, the usual double precision computation may not be enough to prevent severe loss of significant digits. In such cases, the distribution of the rescaled simulated data would appear to be more “discrete" with unusually large variance. As a compromise, partial numerical evidence of the convergence results may be witnessed by using relatively smaller values of n instead. See figure 3 and table 7. Such phenomena also occur to Wn and Wn(p), but to a much lesser extent.

\includegraphics[width=6.44cm,height=4cm]{mydirfig1a1.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig1a2.jpg}
\includegraphics[width=6.44cm,height=4cm]{mydirfig2a1.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig2a2.jpg}
\includegraphics[width=6.44cm,height=4cm]{mydirfig3a1.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig3a2.jpg}

Figure 1 Empirical PDF of normalized Sn, An, Sn: a=1, a=2 and n=12288.

\includegraphics[width=6.44cm,height=4cm]{mydirfig5a1.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig5a2n.jpg}
\includegraphics[width=6.44cm,height=4cm]{mydirfig6a1.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig6a2n.jpg}
\includegraphics[width=6.44cm,height=4cm]{mydirfig7a1.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig7a2n.jpg}

Figure 2 Empirical PDF of normalized Wn, Wn(2), Wn(1): a=1, n=3072 and a=2, n=1536.

\includegraphics[width=6.44cm,height=4cm]{mydirfig4a1n1536.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig4a2n768.jpg}
\includegraphics[width=6.44cm,height=4cm]{mydirfig4a1n3072.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig4a2n1536.jpg}
\includegraphics[width=6.44cm,height=4cm]{mydirfig4a1n6144.jpg} \includegraphics[width=6.44cm,height=4cm]{mydirfig4a2n3072.jpg}

Figure 3 Empirical PDF of normalized Wn(1)=1615Sn15An+215Sn: a=1, n=1536,3072,6144 and a=2, n=768,1536,3072.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

1536

-0.9980

-1.0000

0.9980

3.1522

3.1623

0.9968

a=1

3072

-0.9989

-1.0000

0.9989

3.1388

3.1623

0.9926

 

6144

-0.9996

-1.0000

0.9996

3.1692

3.1623

1.0022

 

12288

-0.9998

-1.0000

0.9998

3.1627

3.1623

1.0001

 

1536

-0.4997

-0.5000

0.9993

0.8996

0.9014

0.9980

a=2

3072

-0.4998

-0.5000

0.9996

0.9018

0.9014

1.0005

 

6144

-0.4999

-0.5000

0.9997

0.9021

0.9014

1.0008

 

12288

-0.5000

-0.5000

0.9999

0.9026

0.9014

1.0013

Table 1 Adjusted means and standard deviations of Sn: μ^(Sn)=π+π3n2μ^n, σ^(Sn)=π3n5/2σ^n. The first four rows are for a=1, and the next four rows for a=2.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

1536

-3.9921

-4.0000

0.9980

12.6082

12.6491

0.9968

a=1

3072

-3.9955

-4.0000

0.9989

12.5550

12.6491

0.9926

 

6144

-3.9984

-4.0000

0.9996

12.6769

12.6491

1.0022

 

12288

-3.9992

-4.0000

0.9998

12.6510

12.6491

1.0001

 

1536

-1.9987

-2.0000

0.9993

3.5983

3.6056

0.9980

a=2

3072

-1.9993

-2.0000

0.9996

3.6072

3.6056

1.0004

 

6144

-1.9995

-2.0000

0.9997

3.6085

3.6056

1.0008

 

12288

-1.9998

-2.0000

0.9999

3.6102

3.6056

1.0013

Table 2 Adjusted means and standard deviations of An: μ^(An)=π+π3n2μ^n, σ^(An)=π3n5/2σ^n for a=1 and a=2.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

1536

1.9962

2.0000

0.9981

6.3053

6.3246

0.9970

a=1

3072

1.9978

2.0000

0.9989

6.2778

6.3246

0.9926

 

6144

1.9992

2.0000

0.9996

6.3385

6.3246

1.0022

 

12288

1.9996

2.0000

0.9998

6.3255

6.3246

1.0002

 

1536

0.9994

1.0000

0.9994

1.7993

1.8028

0.9981

a=2

3072

0.9996

1.0000

0.9996

1.8036

1.8028

1.0005

 

6144

0.9997

1.0000

0.9997

1.8042

1.8028

1.0008

 

12288

0.9999

1.0000

0.9999

1.8051

1.8028

1.0013

Table 3 Adjusted means and standard deviations of Sn: μ^(Sn)=π+π3n2μ^n, σ^(Sn)=π3n5/2σ^n for a=1 and a=2.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

1536

0.8022

0.8000

1.0027

5.1913

5.0596

1.0260

a=1

3072

0.8008

0.8000

1.0009

5.0906

5.0596

1.0061

 

6144

0.8001

0.8000

1.0001

5.1441

5.0596

1.0167

 

12288

0.7905

0.8000

0.9881

11.5189

5.0596

2.2766

 

1536

0.2002

0.2000

1.0008

0.7266

0.7211

1.0076

a=2

3072

0.2000

0.2000

1.0002

0.7258

0.7211

1.0065

 

6144

0.1995

0.2000

0.9976

0.9367

0.7211

1.2990

 

12288

0.1926

0.2000

0.9631

10.2946

0.7211

14.2760

Table 4 Adjusted means and standard deviations of Wn: μ^(Wn)=π+π5n4μ^n, σ^(Wn)=π5n9/2σ^n for a=1 and a=2.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

1536

-0.8018

-0.8000

1.0022

5.1782

5.0596

1.0234

a=1

3072

-0.8007

-0.8000

1.0009

5.0872

5.0596

1.0055

 

6144

-0.8010

-0.8000

1.0012

5.1421

5.0596

1.0163

 

12288

-0.8075

-0.8000

1.0094

11.2584

5.0596

2.2251

 

1536

-0.2001

-0.2000

1.0007

0.7261

0.7211

1.0070

a=2

3072

-0.2001

-0.2000

1.0004

0.7258

0.7211

1.0065

 

6144

-0.2003

-0.2000

1.0013

0.9326

0.7211

1.2932

 

12288

-0.2025

-0.2000

1.0123

10.1136

0.7211

14.0251

Table 5 Adjusted means and standard deviations of Wn(2): μ^(Wn(2))=π+π5n4μ^n, σ^(Wn(2))=π5n9/2σ^n for a=1 and a=2.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

1536

1.6042

1.6000

1.0026

10.3763

10.1193

1.0254

a=1

3072

1.6016

1.6000

1.0010

10.1794

10.1193

1.0059

 

6144

1.6016

1.6000

1.0010

10.2350

10.1193

1.0114

 

12288

1.6028

1.6000

1.0017

14.5945

10.1193

1.4422

 

1536

0.4003

0.4000

1.0008

1.4529

1.4422

1.0074

a=2

3072

0.4002

0.4000

1.0004

1.4503

1.4422

1.0056

 

6144

0.4004

0.4000

1.0011

1.5646

1.4422

1.0849

 

12288

0.4046

0.4000

1.0115

10.5186

1.4422

7.2933

Table 6 Adjusted means and standard deviations of Wn(1): μ^(Wn(1))=π+π5n4μ^n, σ^(Wn(1))=π5n9/2σ^n for a=1 and a=2.
 

n

μ^n

μ

μ^n/μ

σ^n

σ

σ^n/σ

 

768

46.6014

48

0.9709

2330.9783

2791.4383

0.8351

a=1

1536

46.8429

48

0.9759

2704.3826

2791.4383

0.9688

 

3072

9.3958

48

0.1957

34597.5684

2791.4383

12.3942

 

768

2.9572

3

0.9857

82.3636

83.7250

0.98374

a=2

1536

2.5341

3

0.8447

467.2516

83.7250

5.5808

 

3072

25.8888

3

-8.6296

34440.6661

83.7250

411.3546

Table 7 Adjusted means and standard deviations of Wn(1)=1615Sn15An+215Sn: μ^(Wn(1))=π+π7n6μ^n, σ^(Wn(1))=π7n13/2σ^n for a=1 and a=2.

Appendix

Proof for lemma 4

We use similar ideas in [ 21 ] to prove ??. First, by using the multinomial expansion formula

(x1+x2++xn)p=m(pm)x1m1x2m2xnmn=m(pm)j=1nxjmj

where p is a positive integer, m=(m1,m2,,mn) is a multi-index with each mj0 and |m|=j=1nmj=p, and (pm) is given by

(pm)=(pm1,m2,,mn)=p!m1!m2!mn!,

we obtain for each 1il

Dn,kipi=(j=1nYjki)pi=mi(pimi)j=1nYjkimi,j

where mi=(mi,1,mi,2,,mi,n) such that |mi|=j=1nmi,j=pi. Then for all 1il, we obtain

i=1lDn,kipi=mi[i=1l(pimi)]j=1nYji=1lkimi,j.

By using 1 and j=1ni=1lkimi,j=i=1lj=1nkimi,j=i=1lkipi, we obtain

E(i=1lDn,kipi)=mi[i=1l(pimi)]Γ(na)Γ(na+j=1ni=1lkimi,j)j=1nΓ(a+i=1lkimi,j)Γ(a)=mi[i=1l(pimi)]Γ(na)Γ(na+i=1lkipi)j=1nΓ(a+i=1lkimi,j)Γ(a).

The key is to estimate j=1nΓ(a+qj)Γ(a) with qj=i=1lkimi,j. Note that

j=1nΓ(a+qj)Γ(a)={j:qj1γqj(a),if   qj1,1,if   qj=0.

Since

j=1nqj=j=1ni=1lkimi,j=i=1lj=1nkimi,j=i=1lkipi=m,

the number r of indices #{1jn:qj1} is at most m. Then if qj1, we have

j:qj1Γ(a+qj)Γ(a)=j:qj1γqj(a)j:qj1γqj(a^)=(a^+qi11)!(a^+qi21)!(a^+qir1)!(m(a^1)+s=1rqis)!(m(a^1)+j=1nqj)!=(ma^)!=O(1),

where a^=a is the smallest integer that is greater than or equal to a, O(1) is some positive constant independent of n. By using Stirling’s formula [ 1 ]

Γ(z)ezzz1/2(2π)1/2[1+112z+1288z213951840z35712488320z4+]

as z, we obtain

Γ(na)Γ(na+i=1lkipi)ena(na)na1/2e(na+i=1lkipi)(na+i=1lkipi)(na+i=1lkipi)1/2=ei=1lkipi(na+i=1lkipi)i=1lkipi(1i=1lkipina+i=1lkipi)na1/2ni=1lkipi.

Substituting 3 and 4 into 0, and using

mii=1l(pimi)=i=1lmi(pimi)=i=1lnpi=ni=1lpi,

we have

E(i=1lDn,kipi)=O(1)ni=1l(ki1)pi,for large n.

By applying Markov inequality, for any δ>0 and ε>0, we obtain

Pr(n(i=1l(ki1)pi)δi=1lDn,kipi>ε)ε1E[n(i=1l(ki1)pi)δi=1lDn,kipi]=O(1)ε1nδ0

as n, which implies n(i=1l(ki1)pi)δj=1lDn,kjpj0 in probability as n. In addition, we have

n3Pr(n(i=1l(ki1)pi)1δi=1lDn,kipi>ε)O(1)ε1n3n1δ<.

By Borel-Cantelli Lemma, it follows that n(i=1l(ki1)pi)1δi=1lDn,kipi0 with probability 1 as n. This completes the proof of lemma 4.

Proof for lemma 5

By applying the equivalent representation [ 2 ]

(X1X0,X2X1,,XnXn1)=L(V1i=1nVi,V2i=1nVi,,Vni=1nVi),

where V1,V2,,Vn are independent gamma random variables with ViΓ(a,0,1), i=1,2,,n, for a>0, we may rewrite

Tn=αDn,32+βDn,5=α(i=1nVi3)2(i=1nVi)6+βi=1nVi5(i=1nVi)5.

With the above reformulation of Tn, we may consider the joint asymptotic distribution of the three sums i=1nVi,i=1nVi3,i=1nVi5. By using the multivariate central limit theorem, we obtain, as n,

n((1ni=1nVi1ni=1nVi31ni=1nVi5)(μ1μ2μ3))LN(0,Σ)

where μ1=E(Vi)=a, μ2=E(Vi3)=γ3(a), μ3=E(Vi5)=γ5(a), and Σ is the covariance matrix of the random vector (Vi,Vi3,Vi5) with

Σ=(a3γ3(a)5γ5(a)3γ3(a)γ6(a)γ32(a)γ8(a)γ3(a)γ5(a)5γ5(a)γ8(a)γ3(a)γ5(a)γ10(a)γ52(a)).

Next, we apply Cramér’s theorem [ 9 ] to obtain

n(f(1ni=1nVi,1ni=1nVi3,1ni=1nVi5)f(μ))LN(0,σT2)

where f is a mapping: R3R such that f(μ) is continuous in a neighborhood of μ=(μ1,μ2,μ3)R3, and σF2=f(μ)Σ(f(μ))T. To do so, we choose f(x,y,z)=αx6y2+βx5z with fx=6αx7y25βx6z, fy=2αx6y, fz=βx5. Then we have

f(1ni=1nVi,1ni=1nVi3,1ni=1nVi5)=n4Tn

with f(μ)=αm32+βm5=a4(a+1)(a+2)[α(a+1)(a+2)+β(a+3)(a+4)], f(μ)=(6αa1m325βa1m5,2αa6γ3(a),βa5). Then σT2=f(μ)Σ(f(μ))T=4α2m32σ32+4αβ(m3m8(1+15/a)m32m5)+β2σ52=8a9(a+1)(a+2)[3α2(a+1)2(a+2)2(3a+7)+5β2(a+3)(a+4)(5a3+60a2+250a+363)+30αβ(a+1)(a+2)(a+3)2(a+4)]. This completes the proof of lemma 5.

1

M. Abramowitz, I.A. Stegun, Handbook of Mathematical Functions, Dover, 1970.

2

N. Balakrishnan, V.B. Nevzorov, A Primer on Statistical Distributions, Wiley, 2003, https:doi.org/10.1002/0471722227.

3

P. Beckmann, A History of π, Fifth edition, The Golem Press, 1982.

4

C. Bélisle, On the polygon generated by n random points on a circle, Statist. Probab. Lett., 81 (2011), pp. 236–242, https://doi.org/10.1016/j.spl.2010.11.012.

5

P. Billingsley, Probability and Measure, Third edition, Wiley, 1995.

6

D.A. Darling, On a class of problems related to the random division of an interval, Ann. Math. Statistics, 24 (1953), pp. 239–253, https:doi.org/10.1214/aoms/1177729030.

7

B. Efron, The convex hull of a random set of points, Biometrika, 52 (1965), pp. 331–343, https://doi.org/10.1093/biomet/52.3-4.331.

8

W. Feller, An Introduction to Probability Theory and its Applications, vol. II, Wiley, 1966.

9

T.S. Ferguson, A Course in Large Sample Theory, Chapman & Hall, New York, 1996.

10

T. Hsing, On the asymptotic distribution of the area outside a random convex hull in a disk, Ann. Appl. Probab, 4 (1994) no. 2, pp. 478–493.

11

I. Hueter, The convex hull of a normal sample, Adv. Appl. Probab, 26 (1994) no. 4, pp. 855–875, https://doi.org/10.2307/1427894.

12

D.C. Joyce, Survey of extrapolation processes in numerical analysis, SIAM Rev., 13 (1971), pp. 435–490, https://doi.org/10.1137/1013092.

13

T. Li, Essays about π, Higher Education Press, 2007 (Chinese).

14

R. Pyke, Spacings, J. Roy. Statist. Soc. Ser. B, 27 (1965), pp. 395–449, https://doi.org/10.1111/j.2517-6161.1965.tb00602.x.

15

P. Rabinowitz, Extrapolation methods in numerical integration, Numer. Algor., 3 (1992), pp. 17–28, https://doi.org/10.1007/BF02141912.

16

A.Rényi, R. Sulanke, Über die konvexe Hülle von n zufällig gewählten Punkten I, Z. Wahrscheinlichkeitstheorie und Verw. Gebiete, 2 (1963), pp. 75–84, https://doi.org/10.1007/BF00535300.

17

A. Rényi, R. Sulanke, Über die konvexe Hülle von n zufällig gewählten Punkten II, Z. Wahrscheinlichkeitstheorie und Verw. Gebiete, 3 (1964), pp. 138–147.

18

R.J. Serfling, Approximation Theorems of Mathematical Statistics, Wiley, 1980, https://doi.org/10.1002/9780470316481.

19

T. Tang, Computational mathematics originated from the computation of π, Higher Education Press, 2018 (Chinese).

20

V. Vu, Central limit theorems for random polytopes in a smooth convex set, Adv. Math., 207 (2006) no. 1, pp. 221–243, https://doi.org/10.1016/j.aim.2005.11.011.

21

S. Wang, W.-Q. Xu, Random cyclic polygons from Dirichlet distributions and approximations of π, Statist. Probab. Lett., 140 (2018), pp. 84–90, https://doi.org/10.1016/j.spl.2018.05.007.

22

S. Wang, W.-Q. Xu, J. Liu, Random polygons and optimal extrapolation estimates of π, Stat. Optim. Inf. Comput., 9 (2021), pp. 241–249, https://doi.org/10.19139/soic-2310-5070-1193.

23

W.-Q. Xu, Extrapolation methods for random approximations of π, J. Numer. Math. Stoch., 5 (2013), pp. 81–92.

24

W.-Q. Xu, Random circumscribing polygons and approximations of π, Statist. Probab. Lett., 106 (2015), pp. 52–57, https://doi.org/10.1016/j.spl.2015.06.026.

25

W.-Q. Xu, L. Meng, Y. Li, Random polygons and estimations of π, Open Math., 17 (2019), pp. 575–581, https://doi.org/10.1515/math-2019-0049.

26

W.-Q. Xu, S. Wang, D. Xu, Nonlinear extrapolation estimates of π, Acta Math. Sin. Appl. Engl. Ser., 40 (2024), no. 1, pp. 91–108.