Skip to main content

A note to the convergence rates in precise asymptotics

Abstract

Let {X, X n ,n1} be a sequence of i.i.d. random variables with zero mean. Set S n = k = 1 n X k , E X 2 = σ 2 >0, and λ r , p (ϵ)= n = 1 n r / p 2 P(| S n | n 1 / p ϵ). In this paper, the author discusses the rate of approximation of p r p E | N | 2 ( r p ) / ( 2 p ) by ϵ 2 ( r p ) / ( 2 p ) λ r , p (ϵ) under suitable moment conditions, where N is normal with zero mean and variance σ 2 >0, which improves the results of Gut and Steinebach (J. Math. Anal. Appl. 390:1-14, 2012) and extends the work He and Xie (Acta Math. Appl. Sin. 29:179-186, 2013). Specially, for the case r=2 and p= 1 β + 1 , β> 1 2 , the author discusses the rate of approximation of σ 2 2 β + 1 by ϵ 2 λ 2 , 1 / ( β + 1 ) (ϵ) under the condition E X 2 I(|X|>t)=O( t δ l(t)) for some δ>0, where l(t) is a slowly varying function at infinity.

MSC:60F15, 60G50.

1 Introduction

Let {X, X n ,n1} be a sequence of i.i.d. random variables. Set S n = k = 1 n X k and λ r , p (ϵ)= n = 1 n r / p 2 P(| S n | n 1 / p ϵ). Heyde [1] proved that

lim ϵ 0 ϵ 2 λ 2 , 1 (ϵ)= σ 2 ,

whenever EX=0 and E X 2 = σ 2 <. Klesov [2] studied the rate of the approximation of σ 2 by ϵ 2 λ 2 , 1 (ϵ) under the condition E | X | 3 <. He and Xie [3] improved the results of Klesov [2]. Gut and Steinebach [4] extended the results of Klesov [2] and obtained the following Theorem A. Gut and Steinebach [5] studied the general idea of proving precise asymptotics.

Theorem A Let {X, X n ,n1} be a sequence of i.i.d.random variables with zero mean and 0<p<2, r2.

  1. (1)

    If E X 2 = σ 2 >0 and E | X | q < for some r<q3, then

    ϵ 2 ( r p ) / ( 2 p ) λ r , p (ϵ) p r p E | N | 2 ( r p ) / ( 2 p ) =o ( ϵ p ( q 2 ) ( r p ) ( q p ) ( 2 p ) ) .
  2. (2)

    If E X 2 = σ 2 >0 and E | X | q < for some q3 with q> 2 r 3 p 2 p , then

    ϵ 2 ( r p ) / ( 2 p ) λ r , p (ϵ) p r p E | N | 2 ( r p ) / ( 2 p ) =o ( ϵ 2 p ( r p ) ( 2 p ) ( p + 2 q p q ) ) ,

where N is normal with mean 0 and variance σ 2 >0.

The purpose of this paper is to strengthen Theorem A and extend the theorem of He and Xie [3] under suitable moment conditions. In addition, we shall discuss the rate at which ϵ 2 λ 2 , 1 / ( β + 1 ) (ϵ) converges to σ 2 2 β + 1 under the condition T(t)=O( t δ l(t)) for some δ>0, where T(t)=E X 2 I(|X|>t), l(t) is a slowly varying function at infinity. Throughout this paper, C represents a positive constant, though its value may change from one appearance to the next, and [x] denotes the integer part of x. Φ(x) is the standard normal distribution function, Φ(x)= 1 2 π x e t 2 / 2 dt, φ(x)= Φ (x).

2 Main results

From Gut and Steinebach [6], it is easy to obtain the following lemma.

Lemma 2.1 Let {X, X n ,n1} be a sequence of i.i.d. normal distribution random variables with zero mean and variance σ 2 >0. Set 0<p<2 and r2, then

ϵ 2 ( r p ) / ( 2 p ) n = 1 n ( r / p ) 2 P ( | S n | n 1 / p ϵ ) p r p E | N | 2 ( r p ) / ( 2 p ) = { O ( ϵ 2 ( r p ) / ( 2 p ) ) , r < 3 p , O ( ϵ 4 p / ( 2 p ) ) , r 3 p .
(2.1)

Lemma 2.2 (Bingham et al. [7])

Let l(t) be a slowly varying function. We have

  1. (1)

    for any η>0,

    lim t t η l(t)=, lim t t η l(t)=0;
  2. (2)

    if 0<δ<1, then

    a t s δ l(s)ds 1 1 δ t 1 δ l(t),t;
  3. (3)

    if δ>1, then

    t s δ l(s)ds 1 1 δ t 1 δ l(t),t;
  4. (4)

    if δ=1, then L(t)= t l ( s ) s ds, m(t)= a t l ( s ) s ds are slowly varying functions; and

    lim t l ( t ) L ( t ) =0, lim t l ( t ) m ( t ) =0.

Theorem 2.1 Let {X, X n ,n1} be a sequence of i.i.d.random variables with zero mean and 0<p<2, r2.

  1. (1)

    If E X 2 = σ 2 >0 and E | X | 3 < for some r<3, then

    ϵ 2 ( r p ) / ( 2 p ) λ r , p (ϵ) p r p E | N | 2 ( r p ) / ( 2 p ) ={ O ( ϵ 2 ( r p ) / ( 2 p ) ) , 2 r < 3 p 2 , O ( ϵ p / ( 2 p ) log 1 ϵ ) , r = 3 p 2 , O ( ϵ p / ( 2 p ) ) , 3 p 2 < r < 3 .
    (2.2)
  2. (2)

    If E X 2 = σ 2 >0 and E | X | 2 + δ < for some 0<δ<1, r<2+δ, then

    ϵ 2 ( r p ) / ( 2 p ) λ r , p ( ϵ ) p r p E | N | 2 ( r p ) / ( 2 p ) = { O ( ϵ 2 ( r p ) / ( 2 p ) ) , 2 r < ( 1 + δ / 2 ) p , O ( ϵ p δ / ( 2 p ) log 1 ϵ ) , r = ( 1 + δ / 2 ) p , o ( ϵ p δ / ( 2 p ) ) , ( 1 + δ / 2 ) p < r < 2 + δ .
    (2.3)
  3. (3)

    If E X 2 = σ 2 >0 and E | X | q < for some q3 with q> 2 r 3 p 2 p , then

    ϵ 2 ( r p ) / ( 2 p ) λ r , p (ϵ) p r p E | N | 2 ( r p ) / ( 2 p ) ={ O ( ϵ 2 ( r p ) / ( 2 p ) ) , 2 r < 3 p 2 , O ( ϵ p / ( 2 p ) log 1 ϵ ) , r = 3 p / 2 , O ( ϵ p / ( 2 p ) ) , r > 3 p / 2 ,
    (2.4)

where N is normal with mean 0 and variance σ 2 >0.

Remark 2.1 Clearly, Theorem 1 and Theorem 2 in He and Xie [3] are special cases of Theorem 2.1, by taking r=2 and p=1.

Remark 2.2 If 0<p<2, r2, we have min( 2 ( r p ) 2 p , p δ 2 p )> p δ ( r p ) ( 2 + δ p ) ( 2 p ) for r<2+δ=q3 and min( 2 ( r p ) 2 p , p 2 p )> 2 ( r p ) p ( 2 p ) ( p + 2 q p q ) for some q3 with q> 2 r 3 p 2 p . So, the results of Theorem 2.1 are stronger than those of Theorem A.

Theorem 2.2 Let {X, X n ;n1} be a sequence of i.i.d random variables with zero mean, and let T(t)=O( t δ l(t)) for some δ>0, where l(t) is a slowly varying function at infinity. Set E X 2 = σ 2 >0 and β> 1 2 .

  1. (1)

    If δ>1, then

    ϵ 2 λ 2 , 1 / ( β + 1 ) (ϵ) σ 2 2 β + 1 ={ O ( ϵ 2 ) , 1 2 < β < 1 4 , O ( ϵ 2 log 1 ϵ ) , β = 1 4 , O ( ϵ 1 / ( 2 β + 1 ) ) , β > 1 4 .
    (2.5)
  2. (2)

    If 0<δ<1, then

    ϵ 2 λ 2 , 1 / ( β + 1 ) (ϵ) σ 2 2 β + 1 ={ O ( ϵ 2 ) , 1 2 < β < 1 2 + δ 4 , O ( ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) ) , β 1 2 + δ 4 .
    (2.6)
  3. (3)

    If δ=1, then

    ϵ 2 λ 2 , 1 / ( β + 1 ) (ϵ) σ 2 2 β + 1 ={ O ( ϵ 2 + ϵ 2 1 ϵ 1 / ( 2 β + 1 ) l ( t ) t d t ) , 1 2 < β < 1 4 , O ( ϵ 2 ( 1 + 1 ϵ 5 l ( t ) t d t ) log 1 ϵ ) , β = 1 4 , O ( ϵ 1 / ( 2 β + 1 ) ( 1 + 1 ϵ ( 2 β + 3 ) / ( 2 β + 1 ) l ( t ) t d t ) ) , β > 1 4 .
    (2.7)

Remark 2.3 For r=2, p= 1 β + 1 . If l(t)=1, then the result of Theorem 2.2 is weaker than that of Theorem 2.1 for 0<δ<1, β 1 2 + δ 4 , and weaker than that of Theorem 2.1 for δ=1. But the condition T(t)=O( t δ ) is weaker than the condition E | X | 2 + δ <. If l(t)0 as t, then the result of Theorem 2.2 is the same as that of Theorem 2.1 for 0<δ<1.

Remark 2.4 For δ>0, the condition E | X | 2 + δ < is neither sufficient nor necessary for the condition T(t)=O( t δ l(t)). Here are some suitable examples.

Example 1 Let X be a random variable with density f(x)= C ( 1 + δ ln | x | ) | x | 3 + δ ln 2 | x | I(|x|>e), where C is a normalizing constant, and 0<δ<1, then EX=0 and T(t)= C t δ ln t I(t>e), l(t)= 1 ln t is a slowly varying function at infinity. But E | X | 2 + δ =C | x | > e 1 + δ ln | x | | x | ln 2 | x | dx=.

Example 2 Let X be a random variable with density f(x)= C ( δ ln 2 | x | + | x | ( ln | x | 1 ) ) | x | δ + 3 ln 2 | x | e | x | / ln | x | I(|x|>e), where 0<δ<1, then EX=0 and T(t)= C t δ e t / ln ( t ) I(t>e), h(t)= 1 e t / ln t , E | X | 2 + δ <. But h(t)= 1 e t / ln t is not a slowly varying function at infinity.

In fact, we have the following result.

Theorem 2.3 Suppose X is a real random variable and δ>0. Then E | X | 2 + δ < if and only if t δ T(t)0 and t s δ 1 T(s)ds0 as t.

Remark 2.5 If t δ T(t) is bounded as t for some δ>0, then we have E | X | 2 + α < for every α(0,δ) from Theorem 2.3.

Remark 2.6 Let X be a random variable with zero mean. If there exist positive constants C 1 and C 2 such that C 1 l(t) t δ T(t) C 2 l(t) for sufficiently large t and some δ>0, where l(t) is a slowly varying function at infinity, then from Lemma 2.2(4) and Theorem 2.3, we have

E | X | 2 + δ < t l ( s ) s ds0as t.

3 Proofs of the main results

Proof of Theorem 2.1 Without loss of generality, we suppose that σ 2 =1, 0<ϵ<1. Since

P ( | S n | n 1 / p ϵ ) =2 ( 1 Φ ( n ( 2 p ) / 2 p ϵ ) ) + R n ,
(3.1)

where

R n =P ( S n n 1 / p ϵ ) Φ ( n 1 / p 1 / 2 ϵ ) +Φ ( n 1 / p 1 / 2 ϵ ) P ( S n n 1 / p ϵ ) .

From (3.1), we have

ϵ 2 ( r p ) / ( 2 p ) λ r , p ( ϵ ) p r p E | N | 2 ( r p ) / ( 2 p ) = 2 ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 ( 1 Φ ( n ( 2 p ) / 2 p ϵ ) ) p r p E | N | 2 ( r p ) / ( 2 p ) + ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n .
(3.2)

By Lemma 2.1, in order to prove Theorem 2.1, we only need to estimate ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n .

  1. (1)

    On account of a non-uniform estimate of the central limit theorem by Nagaev [8], for every xR,

    | P ( S n n < x ) Φ ( x ) | C E | X | 3 n ( 1 + | x | ) 3 .
    (3.3)

By (3.3), | R n | C E | X | 3 n ( 1 + ϵ n ( 2 p ) / 2 p ) 3 .

  1. (a)

    If r<3p/2, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n C ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 5 / 2 =O ( ϵ 2 ( r p ) / ( 2 p ) ) .
    (3.4)
  2. (b)

    If 3p/2<r<3, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n C ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 n ( 1 + ϵ n ( 2 p ) / 2 p ) 3 C ϵ 2 ( r p ) / ( 2 p ) ( n = 1 [ ϵ 2 p / ( 2 p ) ] n r / p 2 n + ϵ 3 n = [ ϵ 2 p / ( 2 p ) ] + 1 n r / p 5 / 2 ( 6 3 p ) / 2 p ) = O ( ϵ p / ( 2 p ) ) .
    (3.5)
  3. (c)

    If r=3p/2, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n C ϵ p / ( 2 p ) ( n = 1 [ ϵ 2 p / ( 2 p ) ] 1 n + ϵ 3 n = [ ϵ 2 p / ( 2 p ) ] + 1 n 1 ( 6 3 p ) / 2 p ) = O ( ϵ p / ( 2 p ) log 1 ϵ ) .
    (3.6)

From (2.1), (3.2), (3.4), (3.5) and (3.6), we obtain (2.2). This completes the proof of part (1).

  1. (2)

    By the inequality in Osipov and Petrov [9], there exists a bounded and decreasing function ψ(u) on the interval (0,) such that lim u ψ(u)=0 and

    | P ( 1 n σ S n < x ) Φ ( x ) | ψ ( n ( 1 + | x | ) ) n δ / 2 ( 1 + | x | ) 2 + δ .

Let x= n ( 2 p ) / 2 p ϵ, we have | R n | 2 ψ ( n ( 1 + n ( 2 p ) / 2 p ϵ ) ) n δ / 2 ( 1 + n ( 2 p ) / 2 p ϵ ) 2 + δ , so that:

  1. (a)

    If 2<r<(1+δ/2)p, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 δ / 2 =O ( ϵ 2 ( r p ) / ( 2 p ) ) .
    (3.7)
  2. (b)

    If (1+δ/2)p<r<2+δ, then by noticing that lim u ψ(u)=0 for any η>0, there exists a natural number N 0 such that ψ( n )<η whenever n> N 0 . We conclude that

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n C ϵ 2 ( r p ) / ( 2 p ) n = 1 2 n r / p 2 ψ ( n ( 1 + n ( 2 p ) / 2 p ϵ ) ) n δ / 2 ( 1 + ϵ n ( 2 p ) / 2 p ) 2 + δ C ϵ 2 ( r p ) / ( 2 p ) ( n = 1 N 0 n r / p 2 δ / 2 ψ ( n ) + η n = N 0 + 1 [ ϵ 2 p / ( 2 p ) ] n r / p 2 δ / 2 ) + C ϵ 2 ( r p ) / ( 2 p ) 2 δ ψ ( ϵ p / ( 2 p ) ) n = [ ϵ 2 p / ( 2 p ) ] + 1 n r / p 2 δ / 2 ( 1 / p 1 / 2 ) ( 2 + δ ) ϵ 2 ( r p ) / ( 2 p ) N 0 r / p 1 δ / 2 + C η ϵ p δ / ( 2 p ) + C ψ ( ϵ p / ( 2 p ) ) ϵ p δ / ( 2 p ) = o ( ϵ p δ / ( 2 p ) ) .
    (3.8)
  3. (c)

    If r=(1+δ/2)p, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n =O ( ϵ p δ / ( 2 p ) log 1 ϵ ) .
    (3.9)

By (2.1) and combining with (3.2), (3.7), (3.8) and (3.9), we obtain (2.3), which completes the proof of part (2).

  1. (3)

    We make use of the following large deviation estimate in Petrov [10]:

    | P ( 1 n σ S n < x ) Φ ( x ) | C n ( 1 + | x | ) q ,x>0.

So, | R n | C n ( 1 + ϵ n ( 2 p ) / 2 p ) q . Hence we have the following.

  1. (a)

    If r<3p/2, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 5 / 2 =O ( ϵ 2 ( r p ) / ( 2 p ) ) .
    (3.10)
  2. (b)

    If r>3p/2, then r p 5 2 2 q p q 2 p <1. By noting that q> 2 r 3 p 2 p , we obtain

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n C ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 ( 1 + ϵ n ( 2 p ) / 2 p ) q n C ϵ 2 ( r p ) / ( 2 p ) n = 1 [ ϵ 2 p / ( 2 p ) ] n r / p 2 1 / 2 + C ϵ 2 ( r p ) / ( 2 p ) q n = [ ϵ 2 p / ( 2 p ) ] + 1 n r / p 2 1 / 2 ( 2 p ) q / 2 p = O ( ϵ p / ( 2 p ) ) .
    (3.11)
  3. (c)

    If r=3p/2, then

    ϵ 2 ( r p ) / ( 2 p ) n = 1 n r / p 2 R n C ϵ p / ( 2 p ) n = 1 [ ϵ 2 p / ( 2 p ) ] 1 n + C ϵ p / ( 2 p ) q n = [ ϵ 2 p / ( 2 p ) ] + 1 n 1 ( 2 p ) q / 2 p = O ( ϵ p / ( 2 p ) log 1 ϵ ) .
    (3.12)

By (2.1), from (3.2), (3.10), (3.11) and (3.12), we have (2.4), which completes the proof of part (3). □

Proof of Theorem 2.2 We write

ϵ 2 λ 2 , 1 / ( β + 1 ) ( ϵ ) 1 2 β + 1 = ( 2 ϵ 2 2 π n = 1 n 2 β ϵ n β + 1 / 2 e t 2 / 2 d t 1 2 β + 1 ) + ϵ 2 ( n = 1 [ ϵ 4 / ( 2 β + 1 ) ] + n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 ) n 2 β ( P ( | S n | ϵ n β + 1 ) 2 2 π ϵ n β + 1 / 2 e t 2 / 2 d t ) = : I 1 + I 2 + I 3 .
(3.13)

First, according to Lemma 2.1, we have

I 1 ={ O ( ϵ 2 ) , 1 2 < β < 1 2 , O ( ϵ 4 / ( 2 β + 1 ) ) , β 1 2 .
(3.14)

For I 3 , applying Lemma 2.3 of Xie and He [11], and letting x=2y= n β + 1 ϵ, we obtain

P ( | S n | n β + 1 ϵ ) nP ( | X | 1 2 n β + 1 ϵ ) +8 e 2 ϵ 4 n 4 β 2 .
(3.15)

Observing the following fact

2 2 π ϵ n β + 1 / 2 e t 2 / 2 dt=2 ( 1 Φ ( n β + 1 2 ϵ ) ) 2 φ ( n β + 1 / 2 ϵ ) n β + 1 / 2 ϵ =O ( ϵ 5 n 5 β 5 / 2 ) ,
(3.16)

from (3.15) and (3.16), we have

I 3 ϵ 2 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n 2 β P ( | S n | ϵ n β + 1 ) + ϵ 2 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 2 n 2 β 2 π ϵ n β + 1 / 2 e t 2 / 2 d t ϵ 2 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n 2 β + 1 P ( | X | > ϵ n β + 1 2 ) + C ϵ 2 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n 2 β 2 + C ϵ 3 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n 3 β 5 / 2 ϵ 2 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n 2 β + 1 | x | 1 2 n β + 1 ϵ d F ( x ) + O ( ϵ 2 ) + O ( ϵ 3 ) ϵ 2 n = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n 2 β + 1 k = n 1 2 k β + 1 ϵ x < 1 2 ( k + 1 ) β + 1 ϵ d F ( x ) + O ( ϵ 2 ) ϵ 2 k = [ ϵ 4 / ( 2 β + 1 ) ] + 1 n = 1 k n 2 β + 1 1 2 k β + 1 ϵ x < 1 2 ( k + 1 ) β + 1 ϵ d F ( x ) + O ( ϵ 2 ) C ϵ 2 k = [ ϵ 4 / ( 2 β + 1 ) ] + 1 k 2 β + 2 1 2 k β + 1 ϵ x < 1 2 ( k + 1 ) β + 1 ϵ d F ( x ) + O ( ϵ 2 ) C k = [ ϵ 4 / ( 2 β + 1 ) ] + 1 1 2 k β + 1 ϵ x < 1 2 ( k + 1 ) β + 1 ϵ x 2 d F ( x ) + O ( ϵ 2 ) C x 1 2 ϵ ( 2 β + 3 ) / ( 2 β + 1 ) x 2 d F ( x ) + O ( ϵ 2 ) = C T ( ϵ ( 2 β + 3 ) / ( 2 β + 1 ) ) + O ( ϵ 2 ) .

Using the assumption on T(t) and Lemma 2.2(1), we can obtain

I 3 ={ O ( ϵ 2 ) , 1 2 < β min ( δ , 1 ) 4 1 2 , O ( ϵ 1 / ( 2 β + 1 ) ) , β 1 4 , δ 1 , O ( ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) ) , β 1 2 + δ 4 , 0 < δ < 1 .
(3.17)

For I 2 , by Bikelis’s inequality (see [12]), we have

I 2 ϵ 2 n = 1 [ ϵ 4 / ( 2 β + 1 ) ] C n 2 β ( 1 + ϵ n β + 1 / 2 ) 3 n 0 ( 1 + ϵ n β + 1 / 2 ) n T ( v ) d v ϵ 2 n = 1 [ ϵ 2 / ( 2 β + 1 ) ] n 2 β 1 / 2 0 ( 1 + ϵ n β + 1 / 2 ) n T ( v ) d v + ϵ 1 n = [ ϵ 2 / ( 2 β + 1 ) ] + 1 [ ϵ 4 / ( 2 β + 1 ) ] n β 2 0 ( 1 + ϵ n β + 1 / 2 ) n T ( v ) d v .

Now, the proof of Theorem 2.2 will be divided into the following cases.

Case 1 of δ>1.

Noting that T(t)E X 2 =1, let δ 1 be a real number such that 1< δ 1 <δ, by Lemma 2.2(1), lim t t δ 1 δ l(t)=0. Therefore, there is a real number T 0 >0 such that | l ( t ) t δ δ 1 |<1 whenever t> T 0 . Then

0 T(t)dt 0 1 T(t)dt+ 1 T(t)dtC+ T 0 1 t δ 1 dt<.

We have

I 2 C ϵ 2 n = 1 [ ϵ 2 / ( 2 β + 1 ) ] n 2 β 1 / 2 + C ϵ 1 n = [ ϵ 2 / ( 2 β + 1 ) ] + 1 [ ϵ 4 / ( 2 β + 1 ) ] + 1 n β 2 = { O ( ϵ 2 ) , 1 2 < β 1 4 , O ( ϵ 2 log 1 ϵ ) , β = 1 4 , O ( ϵ 1 / ( 2 β + 1 ) ) , β > 1 4 .
(3.18)

From (3.13), (3.14), (3.17) and (3.18), we obtain (2.5).

Case 2 of 0<δ<1.

  1. (a)

    If 1 2 <β< 1 2 + δ 4 , then n = 1 n 2 β 1 / 2 < and 1 t 4 β + 1 δ l(t)dt<. Making use of Lemma 2.2(2)-(3), we have

    I 2 C ϵ 2 n = 1 [ ϵ 2 / ( 2 β + 1 ) ] n 2 β 1 / 2 ( 1 + 1 2 n T ( t ) d t ) + C ϵ 1 n = [ ϵ 2 / ( 2 β + 1 ) ] + 1 [ ϵ 4 / ( 2 β + 1 ) ] n β 2 ( 1 + 1 2 ϵ n β + 1 T ( t ) d t ) C ϵ 2 n = 1 [ ϵ 2 / ( 2 β + 1 ) ] n 2 β 1 / 2 ( n ) 1 δ l ( n ) + O ( ϵ 2 ) + C ϵ 1 / ( 2 β + 1 ) + C ϵ 1 n = [ ϵ 2 / ( 2 β + 1 ) ] + 1 [ ϵ 4 / ( 2 β + 1 ) ] n β 2 ( 2 n β + 1 ϵ ) 1 δ l ( 2 n β + 1 ϵ ) C ϵ 2 1 ϵ 2 / ( 2 β + 1 ) x 2 β 1 / 2 ( x ) 1 δ l ( x ) d x + C ϵ δ ϵ 2 / ( 2 β + 1 ) x β 2 l ( 2 x β + 1 ϵ ) x ( β + 1 ) ( 1 δ ) d x + O ( ϵ 2 ) C ϵ 2 1 ϵ 1 / ( 2 β + 1 ) t 4 β + 1 δ l ( t ) d t + C ϵ 1 / ( 2 β + 1 ) l ( t ) t 1 + δ d t + O ( ϵ 2 ) C ϵ 2 + C ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) = O ( ϵ 2 ) .
    (3.19)
  2. (b)

    If β 1 2 + δ 4 , then we have

    I 2 C ϵ 2 ϵ ( 4 β + 1 ) / ( 2 β + 1 ) ( 1 + 1 2 ϵ 1 / ( 2 β + 1 ) T ( t ) d t ) + C ϵ 1 / ( 2 β + 1 ) + C ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) C ϵ 1 / ( 2 β + 1 ) ( 1 + ( 2 ϵ ( 1 δ ) / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) ) + C ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) C ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) .
    (3.20)

Therefore

I 2 ={ O ( ϵ 2 ) , 1 2 < β 1 2 + δ 4 , O ( ϵ δ / ( 2 β + 1 ) l ( ϵ 1 / ( 2 β + 1 ) ) ) , β 1 2 + δ 4 .
(3.21)

Combining the estimate with (3.11) and (3.14), by (3.10), this implies that (2.6) follows.

Case 3 of δ=1.

  1. (a)

    If 1 2 <β< 1 4 , then n = 1 n 2 β 1 2 <. We have

    I 2 C ϵ 2 ( 1 + 1 ϵ 1 / ( 2 β + 1 ) T ( t ) d t ) n = 1 [ ϵ 2 / ( 2 β + 1 ) ] n 2 β 1 / 2 + C ϵ 1 ( 1 + 1 ϵ 2 β + 3 2 β + 1 T ( t ) d t ) n = [ ϵ 2 / ( 2 β + 1 ) ] + 1 [ ϵ 4 / ( 2 β + 1 ) ] n β 2 C ϵ 2 ( 1 + 1 1 / ( 2 β + 1 ) l ( t ) t d t ) + ϵ 1 / ( 2 β + 1 ) ( 1 + 1 ϵ 2 β + 3 2 β + 1 l ( t ) t d t ) C ϵ 2 ( 1 + 1 ϵ 1 / ( 2 β + 1 ) l ( t ) t d t ) .
    (3.22)
  2. (b)

    If β> 1 4 , then we have

    I 2 C ϵ 2 ϵ 2 2 β + 1 ( 2 β + 1 / 2 ) ( 1 + 1 ϵ 1 / ( 2 β + 1 ) l ( t ) t d t ) + C ϵ 1 / ( 2 β + 1 ) ( 1 + 1 ϵ ( 2 β + 3 ) / ( 2 β + 1 ) l ( t ) t d t ) C ϵ 1 / ( 2 β + 1 ) ( 1 + 1 1 / ( 2 β + 1 ) l ( t ) t d t ) + ϵ 1 / ( 2 β + 1 ) ( 1 + 1 ϵ ( 2 β + 3 ) / ( 2 β + 1 ) l ( t ) t d t ) C ϵ 1 / ( 2 β + 1 ) ( 1 + 1 ϵ ( 2 β + 3 ) / ( 2 β + 1 ) l ( t ) t d t ) .
    (3.23)
  3. (c)

    If β= 1 4 , then we have

    I 2 C ϵ 2 log 1 ϵ ( 1 + 1 ϵ 2 l ( t ) t d t ) + C ϵ 2 ( 1 + 1 ϵ 5 l ( t ) t d t ) C ϵ 2 log 1 ϵ ( 1 + 1 ϵ 5 l ( t ) t d t )

so that

I 2 ={ O ( ϵ 2 ( 1 + 1 ϵ 1 / ( 2 β + 1 ) l ( t ) t d t ) ) , 1 2 < β 1 4 , O ( ϵ 2 ( 1 + 1 ϵ 5 l ( t ) t d t ) log 1 ϵ ) , β = 1 4 , O ( ϵ 1 / ( 2 β + 1 ) ( 1 + 1 ϵ ( 2 β + 3 ) / ( 2 β + 1 ) l ( t ) t d t ) ) , β > 1 4 .
(3.24)

Combining the estimate with (3.14) and (3.17), by (3.13), this implies that (2.7) follows, and hence Theorem 2.2 is proved. □

Proof of Theorem 2.3 Set T 1 (t)=E | X | 2 + δ I(|X|>t). First, note that

E | X | 2 + δ I ( | X | > t ) = | x | > t | x | 2 + δ d F ( x ) = | x | > t x 2 ( t | x | δ y δ 1 d y ) d F ( x ) + t δ | x | > t x 2 d F ( x ) = t δ y δ 1 ( | x | > y x 2 d F ( x ) ) d y + t δ T ( t ) = δ t s δ 1 T ( s ) d s + t δ T ( t ) .

We have

T 1 (t)=δ t s δ 1 T(s)ds+ t δ T(t).

Since t s δ 1 T(s)ds0, t δ T(t)0, we have

T 1 (t)0 t δ T(t)0and t s δ 1 T(s)ds0as t.

Next, it is easy to get

E | X | 2 + δ < T 1 (t)0as t.

From the above facts, the proof of Theorem 2.3 is complete. □

References

  1. Heyde CC: A supplement to the strong law of large numbers. J. Appl. Probab. 1975, 12: 903–907.

    MathSciNet  Article  Google Scholar 

  2. Klesov OI: On the convergence rate in a theorem of Heyde. Theory Probab. Math. Stat. 1994, 49: 83–87.

    MathSciNet  Google Scholar 

  3. He JJ, Xie TF: Asymptotic property for some series of probability. Acta Math. Appl. Sin. 2013, 29: 179–186.

    MathSciNet  Article  Google Scholar 

  4. Gut A, Steinebach J: Convergence rates in precise asymptotics. J. Math. Anal. Appl. 2012, 390: 1–14. 10.1016/j.jmaa.2011.11.046

    MathSciNet  Article  Google Scholar 

  5. Gut A, Steinebach J: Precise asymptotics-a general approach. Acta Math. Hung. 2013, 138: 365–385. 10.1007/s10474-012-0236-1

    MathSciNet  Article  Google Scholar 

  6. Gut, A, Steinebach, J: Correction to ‘Convergence rates in precise asymptotics’. Preprint (http://www2.math.uu.se/~allan/90correction.pdf) (2012).

  7. Bingham NH, Goldie CM, Teugels JL: Regular Variation. Cambridge University Press, Cambridge; 1987.

    Book  Google Scholar 

  8. Nagaev SV: Some limit theorems for large deviation. Theory Probab. Appl. 1965, 10: 214–235. 10.1137/1110027

    MathSciNet  Article  Google Scholar 

  9. Osipov LV, Petrov VV: On an estimate of remainder term in the central limit theorem. Theory Probab. Appl. 1967, 12: 281–286. 10.1137/1112030

    MathSciNet  Article  Google Scholar 

  10. Petrov VV: Limit Theorems of Probability Theory. Oxford University Press, Oxford; 1995.

    Google Scholar 

  11. Xie TF, He JJ: Rate of convergence in a theorem of Heyde. Stat. Probab. Lett. 2012, 82: 1576–1582. 10.1016/j.spl.2012.03.034

    MathSciNet  Article  Google Scholar 

  12. Bikelis A: Estimates of the remainder in the central limit theorem. Liet. Mat. Rink. 1966, 6: 323–346.

    MathSciNet  Google Scholar 

Download references

Acknowledgements

The author would like to thank the referee for helpful comments.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Jianjun He.

Additional information

Competing interests

The author declares that they have no competing interests.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 2.0 International License (https://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Reprints and Permissions

About this article

Cite this article

He, J. A note to the convergence rates in precise asymptotics. J Inequal Appl 2013, 378 (2013). https://doi.org/10.1186/1029-242X-2013-378

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/1029-242X-2013-378

Keywords

  • convergence rate
  • precise asymptotics
  • slowly varying function