Research article

The exponential non-uniform bound on the half-normal approximation for the number of returns to the origin

  • Received: 10 March 2024 Revised: 13 May 2024 Accepted: 29 May 2024 Published: 06 June 2024
  • MSC : 60F05

  • This research explored the number of returns to the origin within the framework of a symmetric simple random walk. Our primary objective was to approximate the distribution of return events to the origin by utilizing the half-normal distribution, which is chosen for its appropriateness as a limit distribution for nonnegative values. Employing the Stein's method in conjunction with concentration inequalities, we derived an exponential non-uniform bound for the approximation error. This bound signifies a significant advancement in contrast to existing bounds, encompassing both the uniform bounds proposed by Döbler [1] and polynomial non-uniform bounds presented by Sama-ae, Chaidee, and Neammanee [2], and Siripraparat and Neammanee [3].

    Citation: Tatpon Siripraparat, Suporn Jongpreechaharn. The exponential non-uniform bound on the half-normal approximation for the number of returns to the origin[J]. AIMS Mathematics, 2024, 9(7): 19031-19048. doi: 10.3934/math.2024926

    Related Papers:

    [1] Najmeddine Attia . Some remarks on recursive sequence of fibonacci type. AIMS Mathematics, 2024, 9(9): 25834-25848. doi: 10.3934/math.20241262
    [2] Chunwei Wang, Shaohua Li, Jiaen Xu, Shujing Wang . Dividend problem of an investment risk model under random observation. AIMS Mathematics, 2024, 9(9): 24039-24057. doi: 10.3934/math.20241169
    [3] Chunwei Wang, Jiaen Xu, Naidan Deng, Shujing Wang . Two-sided jumps risk model with proportional investment and random observation periods. AIMS Mathematics, 2023, 8(9): 22301-22318. doi: 10.3934/math.20231137
    [4] Abd El-Raheem M. Abd El-Raheem, Mona Hosny . Saddlepoint p-values for a class of nonparametric tests for the current status and panel count data under generalized permuted block design. AIMS Mathematics, 2023, 8(8): 18866-18880. doi: 10.3934/math.2023960
    [5] Xintao Li . Wong-Zakai approximations and long term behavior of second order non-autonomous stochastic lattice dynamical systems with additive noise. AIMS Mathematics, 2022, 7(5): 7569-7594. doi: 10.3934/math.2022425
    [6] Essam A. Ahmed, Laila A. Al-Essa . Inference of stress-strength reliability based on adaptive progressive type-Ⅱ censing from Chen distribution with application to carbon fiber data. AIMS Mathematics, 2024, 9(8): 20482-20515. doi: 10.3934/math.2024996
    [7] Young Joon Ahn . $ G^2/C^1 $ Hermite interpolation of offset curves of parametric regular curves. AIMS Mathematics, 2023, 8(12): 31008-31021. doi: 10.3934/math.20231587
    [8] Fiaz Ahmad Bhatti, G. G. Hamedani, Mashail M. Al Sobhi, Mustafa Ç. Korkmaz . On the Burr XII-Power Cauchy distribution: Properties and applications. AIMS Mathematics, 2021, 6(7): 7070-7092. doi: 10.3934/math.2021415
    [9] Chunting Ji, Hui Liu, Jie Xin . Random attractors of the stochastic extended Brusselator system with a multiplicative noise. AIMS Mathematics, 2020, 5(4): 3584-3611. doi: 10.3934/math.2020233
    [10] Naveed Iqbal, Imran Khan, Rasool Shah, Kamsing Nonlaopon . The fuzzy fractional acoustic waves model in terms of the Caputo-Fabrizio operator. AIMS Mathematics, 2023, 8(1): 1770-1783. doi: 10.3934/math.2023091
  • This research explored the number of returns to the origin within the framework of a symmetric simple random walk. Our primary objective was to approximate the distribution of return events to the origin by utilizing the half-normal distribution, which is chosen for its appropriateness as a limit distribution for nonnegative values. Employing the Stein's method in conjunction with concentration inequalities, we derived an exponential non-uniform bound for the approximation error. This bound signifies a significant advancement in contrast to existing bounds, encompassing both the uniform bounds proposed by Döbler [1] and polynomial non-uniform bounds presented by Sama-ae, Chaidee, and Neammanee [2], and Siripraparat and Neammanee [3].


    A symmetric simple random walk is a discrete-time stochastic process applicable in various fields, including physics, finance, biology, and probability theory. It is used to represent the movement of a particle involving randomness. Let X1,X2,,Xn be independent, identically distributed, random variables with

    P(Xi=1)=P(Xi=1)=12,i=1,2,,n.

    The symmetric simple random walk is a process (Sn)n0 defined by

    S0=0 and Sn=ni=1Xi for n1.

    Here, Sn represents the position of the walk in the nth step.

    For the sake of convenience, let us assume that n=2m for natural number m. We are interested in the number of returns to the origin, which is defined by

    Kn=|{kN|1knandSk=0}|

    with a probability mass function

    P(Kn=r)=(nrn2)2n+r (1.1)

    for r=0,1,2,,m (see [1, p. 178]).

    From the point probability formula (1.1), we are able to compute the probability distribution of the statistic Kn directly, particularly for cases where n is relatively small. For example, we have

    P(K21)=1i=0P(K2=i)=122(21)+12(11)=1.

    In situations where n takes on large values, such as in this work where we assume that n4, the computation of the probability distribution becomes a time-consuming task and requires the utilization of a high-performance computing system. Consequently, this leads us to consider the approximation of the probability distribution of the statistic Kn. In [4], it is shown that

    Kn=Knn

    converges in distribution to a half-normal distribution denoted by H(z) as n, where H(z) is defined by

    H(z)={2Φ(z)1,ifz0,0,ifz<0,

    and Φ is a distribution function of a standard normal random variable. This implies that we can approximate the distribution of Kn using the half-normal distribution.

    In the context of the approximation problem, it is imperative to establish a rigorous error bound stemming from the approximation. To this end, let ϵn(z) be the distance between the probability distribution of Kn and a half-normal distribution, i.e.,

    ϵn(z)=|P(Knz)H(z)|,

    and let

    ϵn=supz0ϵn(z).

    A bound on ϵn is termed a uniform bound, while a bound on ϵn(z) is referred to as a non-uniform bound. Döbler [1] showed in 2015 that when n is an even positive integer, the following uniform bound holds:

    ϵn1n(3.07521+1.5n). (1.2)

    Subsequently, Sama-ae, Chaidee, and Neammanee [2] further refined the error bounds by presenting polynomial non-uniform bounds of degree 3, which exhibit greater precision compared to the uniform bound (1.2). Their result states that if z is a nonnegative real number and n is an even positive integer, then,

    ϵn(z)1(1+z)3n(107.56185+73.75519n+43.14923n+13.97885nn+2n2). (1.3)

    More recently, Siripraparat and Neammanee [3] improved the bound for the number of returns to the origin by introducing polynomial non-uniform bounds with an arbitrary degree k for any positive integer k. Presented below is their resultant finding. If z1, kN and n is an even positive integer such that n4, then,

    ϵn(z)1n{2.0918e7z232+0.8946zez22+1zk[2.0958+(2.9166(43)k+32k)EKk+1n]}, (1.4)

    where EKlnl21i=0(2l2i1) for l=2,3,4, and l2 is the greatest integer less than or equal to l2.

    Notice that the bound (1.4) decreases as k increases due to the term 1zk. However, the bound also incorporates the term EKk+1n, which increases with k. Consequently, in this study, we present a more precise bound in the form of an exponential non-uniform bound. The following represents our primary result.

    Theorem 1. Let z be a nonnegative real number. For any even positive integer n such that n4, we have

    ϵn(z)1n(2.9469e7z232+2.3874ez22+31.4793ez+10.4408e3z4). (1.5)

    The rest of this paper is structured as follows: Section 2 introduces Stein's method for half-normal approximation, while Section 3 presents the moment bounds for Kn and Kn. Section 4 is dedicated to proving a concentration inequality. Section 5 provides the proof of the main result. In Section 6, we present the application of Kn, and finally, Section 7 gives a conclusion.

    The primary technique employed to establish the main result, which provides a half-normal approximation, is Stein's method combined with concentration inequalities, as demonstrated by Döbler [1], Sama-ae, Chaidee, and Neammanee [2], and Siripraparat and Neammanee [3]. Stein [5] introduced a method to establish the bounds in the normal approximation for random variables, a technique known as Stein's method. This approach has been extended to various other distributions, including the Poisson distribution [6], binomial distribution [7], negative binomial distribution [8], beta distribution [9], variance-gamma distribution [10], Laplace distribution [11,12], and exponential distribution [13,14,15]. Moreover, the Stein's method can also be extended to work with random vectors as well [16].

    We introduce Stein's method as applied to the half-normal distribution, which is employed to approximate the distribution of any random variable Döbler [1] utilized this approach and presented Stein's equation for the standard half-normal approximation, outlined as follows:

    f(x)xf(x)=h(x)H(z), (2.1)

    where f and h are continuous, piecewise, differentiable functions on [0,).

    To derive an equation for the distribution function from Eq (2.1), we define a function hz:[0,)R as follows for z0:

    hz(x)={1, if 0xz,0, if x>z. (2.2)

    Consequently, for any random variable W, we obtain

    E(fz(W))E(Wfz(W))=P(Wz)H(z), (2.3)

    where fz is the Stein solution of the differential Eq (2.1) with hz in (2.2) given by

    fz(x)={2πex22(1Φ(z))(2Φ(x)1), if xz,2πex22(1Φ(x))(2Φ(z)1), if x>z, (2.4)

    for z0. Note that

    fz(x)={x2πex22(1Φ(z))(2Φ(x)1)+2[1Φ(z)], if xz,x2πex22(1Φ(x))(2Φ(z)1)2[Φ(z)1], if x>z, (2.5)

    and

    |fz(x)|1for allxR, (2.6)

    (see Döbler [1,p. 177]). From (2.3), we can bound |E(fz(W))E(Wfz(W))| instead of |P(Wz)H(z)|. This technique is called Stein's method.

    In order to prove our main result, we need the following properties of fz and fz.

    Proposition 1. Let x,z>0.

    1) 0<fz(x)<ex2z22 for xz.

    2) 0<fz(x)<min(1,1z).

    3) 0fz(x)ex2z22+1.65ez22 for xz.

    Proof. 1) Let xz. By (2.4) and the fact that

    1Φ(x)ex222πxforx>0 ([17,p.23]), (2.7)

    we obtain

    fz(x)2πex22ez222πz(2Φ(1)1)ex2z22forz1.

    Next, we consider z<1. By recalling (2.4) and (2.7), we get

    fz(x)2πex22ez222πz(2Φ(x)1)ex2z22x(2Φ(x)1)ex2z22, (2.8)

    where we use the fact that

    2Φ(x)1x1forx>0 (2.9)

    in the last inequality.

    2) By Sama-ae, Chaidee, and Neammanee ([2, p. 781]), we have

    0<fz(x)<1z,

    for x,z>0 and z1. In the case that z<1, we divide the proof into two cases.

    Case 1: x>z and z<1.

    By (2.4), (2.7), and (2.9), we get

    fz(x)2πex22ex222πx(2Φ(z)1)zx<1forx>0.

    Case 2: xz and z<1.

    We get immediately from (2.8) that fz(x)<1.

    3) Let xz. By Sama-ae, Chaidee, and Neammanee ([2,p. 785]), we obtain

    0fz(x)xzex2z22+2π1zez22ex2z22+0.7979ez22,

    for z1. By (2.5), (2.7), and z<1, we get

    fz(x)x2πex22ez222πz(2Φ(1)1)+2(1Φ(0))ex2z22+1ex2z22+1.65ez22.

    In this section, we consider the moments of Kn, which play a crucial role in establishing the exponential non-uniform bound (1.5).

    Let n=2m, where m is a natural number. It is known that the number of returns to the origin Kn with support [0,m]Z has the following characterization

    E[(2mX+1)(g(X)g(X1))(X+1)g(X)]=0, (3.1)

    for all function g:[1,m]ZR such that g(1)=0 ([1, p. 178]).

    Following Lemma 3.1 in [1,p. 178], we obtain that

    EKn2π. (3.2)

    Using (3.1), Sama-ae, Chaidee, and Neammanee ([2, p. 783]) showed that

    EK2n1,EK3n1.6,andEK4n3. (3.3)

    Siripraparat and Neammanee ([3,p. 46]) improved the moments of Kn to the general case by using the fact that

    EKkn=kEKk1n+(n+1)[k2l=1(k1l)(1)klEKln+(1)k(1P(Kn=0))]k3l=0(k1l)(1)klEKl+1n, (3.4)

    and obtained that

    EKknk21i=0(2k2i1)for k=2,3,4,, (3.5)

    where k2 is the largest integer less than or equal to k2.

    In this paper, we need to bound EeKn by some constant. If we used (3.5), then,

    EeKn=k=01k!k21i=0(2k2i1)=,

    which is divergent. Our aim in this section is to enhance the precision of (3.5), as in the following proposition.

    Proposition 2. For the even positive integer n, we have

    1)EKknnk22l=0(k12l)EK2ln, where k is even and k2;

    2)EKknnk32l=0(k12l+1)EK2l+1n, where k is odd and k3.

    Proof. By (3.4), we have

    EKkn=A1+A2, (3.6)

    where

    A1=kEKk1n+(k1k2)EKk2n+(k1k3)EKk2nn(k1k3)EKk3n, (3.7)

    and

    A2=n(k1)EKk2n+nk4l=1(k1l)(1)klEKln+k3l=1(k1l)(1)klEKlnk4l=0(k1l)(1)klEKl+1n+(n+1)(1)k(1P(Kn=0)). (3.8)

    By the facts that

    KlnKl+1n for lN, (3.9)
    Kl+1nnEKln for lN, (3.10)

    and (3.7), we obtain

    A1=kEKk1n+(k2k2)EKk2nn(k23k+22)EKk3nkEKk2n+(k2k2)EKk2n(k23k+22)EKk2n0. (3.11)

    1) If k is even, then by the fact that (n+1)P(Kn=0)=EKn+1 ([1,p. 178]) and (3.8), we obtain that

    A2=n(k1)EKk2n+nk4l=1(k1l)(1)klEKln+k3l=1(k1l)(1)klEKlnk4l=0(k1l)(1)klEKl+1n+nEKn.

    Next, we utilize the facts (3.9) and (3.10) to eliminate the odd moment in the second term and the backward terms. This results in the remaining terms being even moments as follows:

    A2nk22l=0(k12l)EK2ln. (3.12)

    2) Suppose that k is odd. By (3.8), we establish

    A2=n(k1)EKk2n+nk4l=1(k1l)(1)klEKln+k3l=1(k1l)(1)klEKlnk4l=0(k1l)(1)klEKl+1n(n+1)(1P(Kn=0)).

    Using a similar technique as in (3.12), we retain the odd moments while eliminating even moments, and thus, we establish

    A2nk32l=0(k12l+1)EK2l+1n. (3.13)

    From (3.6) and (3.11)–(3.13), we complete the proof.

    Note that we can bound additional moments by employing Proposition 2, transforming Kn into Kn, and utilizing the initial moments presented in (3.2), (3.3), and taking into account the condition n4. Below are the fifth, sixth, and seventh moments for Kn:

    EK5n1nn[(41)EKn+(43)EK3n]=1n(41)EKn+(43)EK3n7.1979, (3.14)
    EK6n1n2(50)+1n(52)EK2n+(54)EK4n17.5625, (3.15)

    and

    EK7n1n2(61)EKn+1n(63)EK3n+(65)EK5n51.4867. (3.16)

    From the demonstration above, it is evident that one can calculate all moments dependent on the forward moments. However, these calculations can be straightforward in contrast to complex. In the next proposition, we offer a bound for the moments of Kn that relies solely on the parameter k and does not depend on other moments. The technique used to derive this proposition is mathematical induction.

    Proposition 3. Let n4. Then,

    EKkn(k1)!(k4)(k5),

    for kN and k6.

    Proof. Let kN with k6. The proof is divided into two cases.

    Case 1: k is even and k6.

    By (3.15), we see that

    EK6n17.5625(k1)!(k4)(k5)fork=6.

    Assume that

    EKkn(k1)!(k4)(k5) (3.17)

    is true for k=6,8,10,. By Proposition 2(1), and the fact that

    Krn=Krn(n)rforrN, (3.18)

    we have

    EKk+2n1(n)kk2l=0(k+12l)EK2ln1(n)k(k+10)+1(n)k2(k+12)EK2n+1(n)k4(k+14)EK4n+1nk22l=3(k+12l)EK2ln+(k+1k)EKkn=Bk+2+Ck+2+Dk+2, (3.19)

    where

    Bk+2=1(n)k(k+10)+1(n)k2(k+12)EK2n+1(n)k4(k+14)EK4n, (3.20)
    Ck+2=1nk22l=3(k+12l)EK2ln, (3.21)

    and

    Dk+2=(k+1k)EKkn. (3.22)

    To derive a bound for Bk+2, we apply the initial moments bound (3.3) while considering the conditions k6 and n4. This results in the bound (3.20) taking the form (k+1)!(k2)(k3) as follows:

    Bk+21(n)6+1(n)4(k+1)!(k1)!2!EK2n+1(n)2(k+1)!(k3)!4!EK4n(k+1)![126(k+1)!+124(k1)!2!+322(k3)!4!](k+1)![0.0157(k+1)!+0.0313(k1)!+0.0313(k3)!](k+1)![113125(k2)(k3)+1156(k2)(k3)+115(k2)(k3)](k+1)![0.0732(k2)(k3)]. (3.23)

    Next, to obtain a bound for Ck+2, one can prove that, for a fixed lN such that l3, we have k(k4)(k5)(k2l+1)!(2l)(2l4)(2l5)1 for k2l. From this fact, (3.21), and n4, we obtain that

    Ck+21nk22l=3(k+1)!(k2l+1)!(2l)!(2l1)!(2l4)(2l5)(k+1)!4k22l=31(k2l+1)!(2l)(2l4)(2l5)(k+1)!k22l=314k(k4)(k5)=(k+1)!k681k(k4)(k5)(k+1)![0.4268(k2)(k3)]. (3.24)

    By (3.22), we obtain

    Dk+2(k+1)(k1)!(k4)(k5)=(k+1)!k(k4)(k5)(k+1)!2(k2)(k3). (3.25)

    From (3.19), (3.8), (3.24), and (3.25), we conclude that

    EKk+2n(k+1)!(0.0732(k2)(k3)+0.4268(k2)(k3)+12(k2)(k3))=(k+1)!(k2)(k3).

    By mathematical induction, we have (3.17) when k is an even positive integer and k6.

    Case 2: k is odd and k7.

    By (3.16), we observe that

    EK7n(k1)!(k4)(k5)fork=7.

    Therefore, the basic step is true.

    To use mathematical induction, we assume that (3.17) is true for k=7,9,11,.

    By Proposition 2(1) and (3.18), we obatin

    EKk+2n1(n)kk12l=0(k+12l+1)EK2l+1n1(n)k1(k+11)EKn+1(n)k3(k+13)EK3n+1(n)k5(k+15)EK5n+1nk32l=3(k+12l+1)EK2l+1n+(k+1k)EKkn=Ek+2+Fk+2+Gk+2, (3.26)

    where

    Ek+2=1(n)k1(k+11)EKn+1(n)k3(k+13)EK3n+1(n)k5(k+15)EK5n,Fk+2=1nk32l=3(k+12l+1)EK2l+1n,

    and

    Gk+2=(k+1k)EKkn.

    To bound Ek+2,Fk+2, and Gk+2, we can directly modify the technique in (3.8), (3.24), and (3.25) to obtain the following results:

    Ek+20.075(k+1)!(k2)(k3), (3.27)
    Fk+20.425(k+1)!(k2)(k3), (3.28)

    and

    Gk+20.5(k+1)!(k2)(k3). (3.29)

    By (3.26)–(3.29), we conclude that

    EKk+2n(k+1)![0.075(k2)(k3)+0.425(k2)(k3)+12(k2)(k3)]=(k+1)!(k2)(k3).

    From these two cases, we have completed the proof.

    To prove our main theorem, we establish a concentration inequality for Kn. Notably, Döbler [1] was the first mathematician providing a uniform concentration inequality for Kn. His result is

    P(z<Knz+1n)2πnforz>0, (4.1)

    (see [1,p. 181]). The term "uniform concentration inequality" indicates that the obtained bound is independent of z. Subsequently, the concentration inequality (4.1) is extended to a non-uniform concentration inequality in terms of zk for kN, as detailed in [2,3]. In this section, we enhance the concentration inequality for Kn in terms of ez, presented in Proposition 4.

    Proposition 4. For z0 and n4, we have

    P(z<Knz+1n)31.4793ezn.

    Proof. Let f:RR be defined by

    f(t)={0,ift<z1n,et+1n(tz+1n),ifz1ntz+1n,2net+1n,ift>z+1n.

    Then, we have f(t)ez>0 for z1n<t<z+1n, which implies that f is increasing.

    We follow the argument of Sama-ae, Chaidee, and Neammanee ([2,pp. 784-785]) to obtain that

    P(z<Knz+1n)1ez(H1+H2), (4.2)

    where

    S|H1|=|2EKnf(Kn)|4nE|KneKn+1n|=4ne1nEKneKn, (4.3)

    and

    |H2|=1n|Ef(Kn)|2n|EeKn+1n|=2ne1nEeKn. (4.4)

    By Proposition 3, we have

    k=6EKknk!=k=61k![(k1)!(k4)(k5)]16,

    and

    k=6EKk+1nk!=k=61k![k!(k3)(k4)]12.

    From these facts, (3.2), (3.3), (3.14), and (3.15), we obtain:

    EeKn=k=0EKknk!2.9163, (4.5)

    and

    EKneKn=k=0EKk+1nk!4.0442. (4.6)

    By (4.2)–(4.6), we conclude that for n4,

    P(z<Knz+1n)31.4793ezn.

    In this section, we give an exponential non-uniform bound for Kn. From this point forward, we use f to denote fz, which is the unique solution of (2.4).

    Proof of Theorem 1: By (1.2), we see that Theorem 1 is true for z=0. Now, we assume z>0 and n4. Döbler ([1,p. 179]) and Siripraparat and Neammanee ([3,p. 51]) showed that

    |E[f(Kn)]E[Knf(Kn)]||J1|+|J2|+|J3|+|J4|, (5.1)

    where

    J1=E[Kn(f(Kn)f(Kn1n))],J2=1nE[f(Kn)],J3=nE[KnKn1nKnt(f(s)+sf(s))dsdt],

    and

    J4=P(z<Knz+1n).

    Bounding |J1|: Applying the fundamental theorem of calculus and employing a truncation technique, we partition |J1| into two terms. By utilizing Proposition 1(3) in the first term and applying (2.6) in the second term, we can then obtain the following:

    |J1|EKn[KnKn1n|f(t)|I(Kn<3z4)dt]+EKn[KnKn1n|f(t)|I(Kn3z4)dt]1n(1e7z232+1.65ez22)EKn+1nEKnI(Kn3z4)1n(1e7z232+1.65ez22)EKn+1nEKneKne3z4.

    By (3.2) and (4.6), we obtain

    |J1|1n(1e7z232+1.65ez22)2π+4.0442ne3z41n(0.7979e7z232+1.3166ez22+4.0442e3z4). (5.2)

    Bounding |J2|: By Markov's inequality, we obtain that

    P(Kn3z4)EeKne3z4. (5.3)

    By employing a truncation technique together with the argument of |J1|, and utilizing Proposition 1(1), Proposition 1(2), (4.5), and (5.3), we establish

    |J2|1nE|f(Kn)|I(Kn<3z4)+1nE|f(Kn)|I(Kn3z4)1ne7z232P(Kn<3z4)+1nP(Kn3z4)1n(1e7z232+EeKne3z4)1n(1e7z232+2.9163e3z4). (5.4)

    Bounding |J3|: Through the application of a truncation technique again, we represent J3 in the following form:

    |J3||J31|+|J32|,

    where

    J31=nE[KnKn1nKnt|f(s)+sf(s)|I(Kn<3z4)dsdt],

    and

    J32=nE[KnKn1nKnt|f(s)+sf(s)|I(Kn3z4)dsdt].

    From Proposition 1(1), Proposition 1(3), and (3.2), we obtain

    |J31|nez22E[KnKn1nKntes22I(Kn<3z4)dsdt]+n(1e7z232+1.65ez22)E[KnKn1nKntmax{1n,Kn}I(Kn<3z4)dsdt]ne7z232(12n)+n2(1e7z232+1.65ez22)(1nn+1nEKn)1n(1.1490e7z232+1.0708ez22),

    where we use (3.2) and n4 in the last inequality.

    Using (2.6), (3.3), Proposition 1(2), (4.5), and (5.3), we have

    |J32|nE[KnKn1nKnt(1+Kn)I(Kn3z4)dsdt]=12nE[(1+Kn)I(Kn3z4)]12nE(1+Kn)eKne3z4=1n(0.5EeKn+0.5EKneKn)e3z43.4803ne3z4.

    Hence, we obtain that

    |J3|1n(1.1490e7z232+1.0708ez22+3.4803e3z4). (5.5)

    Bounding |J4|: It follows immediately from Proposition 4 that

    |J4|31.4793nez. (5.6)

    By Stein's equation (2.3), (5.1), and (5.2)–(5.6), we conclude that

    |P(Knz)H(z)|1n(2.9469e7z232+2.3874ez22+31.4793ez+10.4408e3z4).

    In this section, we provide an application of Theorem 1. Consider an option pricing following the binomial model (see [18,19] for more details) where the possible price of an option called "premium" is either increasing or decreasing. Let the initial premium be S0=0, which means that there is no change in the price. Let a random variable Xi be the change in the premium with distribution

    P(Xi=1)=P(Xi=1)=12,i=1,2,,n.

    Then, Sn=ni=1Xi represents the total change of the premium at period n. The premium is the same as at the initial state if Sn=0 for some period of time n. If we need to forecast the chance that the premium is the same as the initial state in a fixed period of time, we can approximate this by the half-normal distribution.

    By applying Theorem 1, we obtain an error bound for the half-normal approximation. We present numerical results for (1.2)–(1.5) to emphasize the sharpness of our result compared to other bounds. The results are displayed in Table 1. It is worth noting that our exponential non-uniform bound rapidly decreases, especially when z is large.

    Table 1.  Comparison of the constants C in uniform and non-uniform bounds in the form of Cn for large n.
    Bounds k z=10 z=50 z=500 z=1000
    (1.2) kN 3.07521 3.07521 3.07521 3.07521
    (1.3) kN 0.08082 8.10864×104 8.55353×107 1.07239×107
    (1.4) k=3 0.49672 0.00398 3.97368×106 4.96711×107
    k=7 2.65978 3.40452×105 3.40452×1012 2.65978×1014
    k=11 4269.57975 8.74409×105 8.74409×1016 4.26958×1019
    k=13 1.39042×106 0.00114 1.13903×1016 1.39042×1020
    (1.5) kN 0.00721 5.40375×1016 1.43981×10162 1.98552×10325

     | Show Table
    DownLoad: CSV

    In addition, when k=7 is fixed and the value of z varies from 30 to 200, we obtain the constants C in the form of Cn for each error bound, as shown in Figure 1. The graph is plotted on a semi-log scale on the vertical axis. We observe that the constant C for the bound (1.3) is approximately 105, and the bound (1.4) is approximately 109. However, our constant C for (1.5) ranges between 1065 and 109, steadily decreasing as z increases. Indeed, the constant C for (1.5) provides the best error bound.

    Figure 1.  Comparison of the constant C for error bounds (1.3), (1.4) when k=7, and (1.5) in the form of Cn.

    By utilizing Stein's method, this study derived an exponential non-uniform bound for the difference between the number of returns to the origin and a half-normal distribution in a symmetric simple random walk. Comparing our exponential non-uniform bound with (1.2), (1.3), and (1.4), it is evident that our bound of this study is sharper as shown in Table 1. Consequently, Theorem 1 is more suitable for evaluating the accuracy of this approximation. We finally provided an example, an option pricing, that supported our research and illustrated the significance of the result. In future work, we will attempt to generalize these criteria to the scenario involving an asymmetric random walk.

    All authors contributed equally to this work. They have read and approved the final version of the manuscript for publication.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    The authors would like to thank the referees for all of the valuable comments and suggestions that helped to improve the quality of the manuscript.

    All authors declare that they have no conflicts of interest.



    [1] C. Döbler, Stein's method for the half-normal distribution with applications to limit theorems related to simple random walk, ALEA, Lat. Am. J. Probab. Math. Stat., 12 (2015), 171–191.
    [2] A. Sama-ae, N. Chaidee, K. Neammanee, Half-normal approximation for statistics of symmetric simple random walk, Commun. Stat.-Theor. M., 47 (2018), 779–792. https://doi.org/10.1080/03610926.2016.1139129 doi: 10.1080/03610926.2016.1139129
    [3] T. Siripraparat, K. Neammanee, A non uniform bound for half-normal approximation of the number of returns to the origin of symmetric simple random walk, Commun. Stat.-Theor. M., 47 (2018), 42–54. https://doi.org/10.1080/03610926.2017.1300286 doi: 10.1080/03610926.2017.1300286
    [4] W. Feller, An introduction to probability theory and its applications, 3 Eds., New York: Wiley, 1968.
    [5] C. Stein, A bound for the error in the normal approximation to the distribution of a sum of dependent random variables, In: Proceeding of the sixth Berkeley symposium on mathematical statistics and probability, Berkeley: University of California Press, 1972,583–602.
    [6] L. H. Y. Chen, Poisson approximation for dependent trials, Ann. Probab., 3 (1975), 534–545. https://doi.org/10.1214/aop/1176996359 doi: 10.1214/aop/1176996359
    [7] A. N. Kumar, P. Vellaisamy, Binomial approximation to locally dependent collateralized debt obligations, Methodol. Comput. Appl. Probab., 25 (2023), 81. https://doi.org/10.1007/s11009-023-10057-8 doi: 10.1007/s11009-023-10057-8
    [8] A. N. Kumar, P. Kumar, A negative binomial approximation to the distribution of the sum of maxima of indicator random variables, Stat. Probabil. Lett., 208 (2024), 110040. https://doi.org/10.1016/j.spl.2024.110040 doi: 10.1016/j.spl.2024.110040
    [9] C. Döbler, Stein's method of exchangeable pairs for the beta distribution and generalizations, Electron. J. Probab., 20 (2014), 1–34. https://doi.org/10.1214/EJP.v20-3933 doi: 10.1214/EJP.v20-3933
    [10] R. Gaunt, Variance-gamma approximation via Stein's method, Electron. J. Probab., 19 (2014), 1–33. https://doi.org/10.1214/EJP.v19-3020 doi: 10.1214/EJP.v19-3020
    [11] J. Pike, H. Ren, Stein's method and the Laplace distribution, ALEA, Lat. Am. J. Probab. Math. Stat., 11 (2014), 571–587.
    [12] K. Barman, N. S. Upadhye, On Stein factors for Laplace approximation and their application to random sums, Stat. Probabil. Lett., 206 (2024), 109996. https://doi.org/10.1016/j.spl.2023.109996 doi: 10.1016/j.spl.2023.109996
    [13] J. Fulman, N. Ross, Exponential approximation and Stein's method of exchangeable pairs, ALEA, Lat. Am. J. Probab. Math. Stat., 10 (2013), 1–13.
    [14] S. Chatterjee, J. Fulman, A. Röllin, Exponential approximation by Stein's method and spectral graph theory, ALEA, Lat. Am. J. Probab. Math. Stat., 8 (2011), 197–223.
    [15] E. A. Peköz, A. Röllin, New rates for exponential approximation and the theorems of Rényi and Yaglom, Ann. Probab., 39 (2011), 587–608. https://doi.org/10.1214/10-AOP559 doi: 10.1214/10-AOP559
    [16] B. S. Zuo, C. C. Yin, Stein's lemma for truncated generalized skew-elliptical random vectors, AIMS Mathematics, 5 (2020), 3423–3433. https://doi.org/10.3934/math.2020221 doi: 10.3934/math.2020221
    [17] C. Stein, Approximate computation of expectations, Stanford University: Institute of Mathematical Statistics, 1986. https://doi.org/10.1214/lnms/1215466568
    [18] Z. Ahmad, R. Browne, R. Chowdhury, R. Das, Y. S. Huang, Y. M. Zhu, Fast American option pricing using nonlinear stencils, In: Proceedings of the 29th ACM SIGPLAN annual symposium on principles and practice of parallel programming, New York: Association for Computing Machinery, 2024,316–332. https://doi.org/10.1145/3627535.3638506
    [19] Y. Ratibenyakool, K. Neammanee, Rate of convergence of binomial formula for option pricing, Commun. Stat.-Theor. M., 49 (2020), 3537–3556. https://doi.org/10.1080/03610926.2019.1590600 doi: 10.1080/03610926.2019.1590600
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1035) PDF downloads(63) Cited by(0)

Figures and Tables

Figures(1)  /  Tables(1)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog