JPS Journal of Probability and Statistics 1687-9538 1687-952X Hindawi Publishing Corporation 10.1155/2015/972035 972035 Research Article Recurrence Relation and Accurate Value on Inverse Moment of Discrete Distributions Wang ChunYuan Wuyungaowa Psaradakis Zacharias Department of Mathematics College of Sciences and Technology Inner Mongolia University Hohhot 010021 China imu.edu.cn 2015 572015 2015 06 03 2015 07 06 2015 572015 2015 Copyright © 2015 ChunYuan Wang and Wuyungaowa. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Properties of the generalized hypergeometric series functions are employed to get the recurrence relation for inverse moments and inverse factorial moments of some discrete distributions. Meanwhile, with the existence of the recurrence relations, the accurate value for inverse moment of discrete distributions can thus be obtained.

1. Introduction and Preliminaries

Kumar and Consul  develop a recursive relation upon the negative moments of power series distribution. The recurrence relation for the negative moments of the Poisson distribution was first derived by Chao and Strawderman , after which it is shown by Kumar and Consul  as a special case of their result. Using the recurrence relation for the negative moments of the Lagrangian binomial distribution, Kumar and Consul  have established the binomial and negative binomial distributions.

Ahmad and Saboor  proved many properties of the hypergeometric series. Besides, the following series have been provided by Ahmad and Saboor  and (1)Fqpa1,k,a2,k,,ap,k;b1,k,b2,k,,bq,k;z=1+a1k·a2kapkb1k·b2kbqkz+a1a1+1ka2a2+1kapap+1kb1b1+1kb2b2+1kbqbq+1kz22!+is generalized hypergeometric series, where (λik)n=(λik)(λi+1)k(λi+2)k(λi+n-1)k, (λik)0=1, i=1,2,,p, k=1,2,. If k=1, then (2)Fqp=n=0a1na2napnb1nb2nbqnznn!is hypergeometric series.

In this paper, the recurrence relation for negative moments along with negative factorial moments of some discrete distributions can be obtained. These relations have been derived with properties of the hypergeometric series.

In the next part, some necessary definitions have been introduced.

Let X be a generalized negative binomially distributed random variable with parameters θ, β and the probability mass function is (3)Pxθ,β=mm+βxm+βxxθx1-θm+βx-x,x=0,1,2,,where 0<θ<1, |θβ|<1, 1βθ-1, and m>0 and a constant.

Let X be a generalized Poisson distribution with parameters λ, θ and the probability mass function is (4)Pxθ=e-λ1+θxλx1+θxx-1x!,x=0,1,,where 0<λ<, |λθ|<1, and 0<θ<1.

Let the random variable X be equipped with a generalized poisson-negative-binomial distribution with parameters λ, θ, and β; the probability mass function is (5)Pxθ,β=e-λ1+θss=0λs1+θss-1s!mm+βxm+βxxθx1-θm+βx-x,x=0,1,2,,where 0<θ<1, 0<λ<, |λθ|<1, |θβ|<1, 1βθ-1, s0, sZ, and m>0 and a constant.

Let X be a generalized logarithmic distributed random variable with parameters θ and β; the probability mass function is (6)Pxθ,β=1βxβxxθx1-θβx-x-log1-θ,x=1,2,,where 0<θ<1, 0<θβ<1, and β1.

2. The Recurrence Relation for Inverse Moments of Some Discrete Distributions

In this section, some recurrence relations for inverse moments of some discrete distributions can be obtained with the properties of the generalized hypergeometric series functions.

Theorem 1.

Let X be a generalized negative binomial random variable with parameters θ, β, for 0<θ<1, |θβ|<1, 1βθ-1, and probability mass function is defined in (3), and then the inverse moment of first order is given by (7)EX+A-1=mθ1-θm+β-1A+1F32A+1,m+n+1β-n,1;A+2,2;θ1-θ1-β,where A0, n=0,1,2,, and m>0 and a constant.

Proof.

Since X is a generalized negative binomial random variable with parameters θ, β, then (8)EX+A-1=x=1mm+βxm+βxxθx1-θm+βx-x1x+A=mθ1-θm+β-1A+11+A+1m+2β-1A+2·2!θ1-θ1-β+A+1m+3β-1m+3β-2A+3·3!θ21-θ21-β+=mθ1-θm+β-1A+11+A+1m+2β-1·1A+2·2·1!θ1-θ1-β+A+1A+2m+3β-1m+3β-2·1·2A+2A+3·2·3·2!θ1-θ1-β2+=mθ1-θm+β-1A+1F32A+1,m+n+1β-n,1;A+2,2;θ1-θ1-β.

Theorem 2.

Let X be a generalized negative binomially distributed random variable with parameters θ and β, and probability mass function is defined in (3). Then the following relation holds: (9)Azp-A-1EX+A-1=A1-AEX+A-1-1-P11-z-p+1+Q0,where Q0=[A+(p-A-1)z]F21[1,p;2;z], p=m+(n+1)β-n, n=0,1,2,, z=θ/(1-θ)1-β, and P1=P(X=1)=mθ(1-θ)m+β+1.

Proof.

From Theorem 1, we have (10)EX+A-1=P1A+1F32A+1,p,1;A+2,2;z,using the identity (see [4, page 85]) (11)1-xF32α1,α2,α3;β1,β2;x=F32α1-1,α2,α3;β1,β2;x+x·α2-β1α3-β1β1β2-β1F32α1,α2,α3;β1+1,β2;x+xα2-β2α3-β2β2β1-β2·F32α1,α2,α3;β1,β2+1;x.For α1=A+1, α2=p, α3=1, β1=2, β2=A+1, and x=z we have (12)1-zF32A+1,p,1;A+1,2;z=F32A,p,1;A+1,2;z+2-pz2A-1F32A+1,p,1;A+1,3;z+Azp-A-1A2-1F32A+1,p,1;A+2,2;z,and we know that (13)F32A+1,p,1;A+1,2;z=F211,p;2;z,F32A+1,p,1;A+1,3;z=F211,p;3;z.Then (12) becomes (14)1-zF211,p;2;z=F32A,p,1;A+1,2;z+2-pz2A-1F211,p;3;z+Azp-A-1A2-1F32A+1,p,1;A+2,2;z.Using another identity (see [4, page 71]) (15)a+b-czF21a,b;c;z=a1-zF21a+1,b;c;z-c-ac-bzcF21a,b;c+1;z,and for a=1, b=p, c=2, and z=z we have (16)1+p-2zF211,p;2;z=1-zF212,p;2;z-2-pz2F211,p;3;zas (17)F212,p;2;z=F10p;-;z=1-z-p.From (14) and (16) we have (18)1-zF211,p;2;z=F32A,p,1;A+1,2;z+Azp-A-1A2-1F32A+1,p,1;A+2,2;z-1+p-2zA-1F211,p;2;z+1-z-p+1A-1.Rearranging we get (19)A+p-A-1zA-1F211,p;2;z-1-z-p+1A-1=AP1EX+A-1-1+Azp-A-1P1A-1EX+A-1.By collating, we get the result (9).

Theorem 3.

Let the random variable X be equipped with a generalized Poisson distribution with parameters θ and λ, and probability mass function is defined in (4). Then (20)AzEX+A-1=A1-AEX+A-1-1+AP1F111;2;z-P1,and P1=P(X=1)=λe-λ(1+θ), z=λ(1+(n+1)θ)/eλθ, and n=0,1,2,.

Proof.

Using the identity (see [4, page 84]) (21)F22α1,α2;β1,β2;x=F22α1-1,α2;β1,β2;x+xα2-β1β1β2-β1F22α1,α2;β1+1,β2;x+xα2-β2β2β1-β2F22α1,α2;β1,β2+1;x.By the same as Theorem 2 calculating, we get the result (20).

Theorem 4.

Let the random variable X be equipped with a generalized Poisson-negative-binomial distribution with parameters λ, θ, and β, and probability mass function is defined in (5); then (22)Azp-A-1EX+A-1=A1-AEX+A-1-1-P11-z-p+1+Q1,where P1=P(X=1)=e-λ1+θss=0(λs(1+θs)s-1/s!)mθ(1-θ)m+β-1, p=m+(n+1)β-n, n=0,1,2,, Q1=P1[A+(p-A-1)z]F21[1;p;2;z], and z=θ/(1-θ)1-β.

Proof.

Using identities (11) and (15), by the same as Theorem 2 calculating, we get the result (22).

Theorem 5.

Let X be a generalized logarithmic distributed random variable with parameters θ and β, and probability mass function is defined in (6). Then the following recurrence relation holds: (23)Azp-A-1EX+A-1=A1-AEX+A-1-1-P11-z-p+1+Q2,where P1=P(X=1)=θ(1-θ)β-1/[-log(1-θ)], Q2=P1[A+(p-A-1)z]F21[1,p;2;z], p=(n+1)β-n, n=0,1,2,, and z=θ/(1-θ)1-β.

Proof.

Using identities (11) and (15), by the same as Theorem 2 calculating, we get the result (23).

3. Recurrence Relation for Inverse Factorial Moments of Discrete Distributions

In this section, some recurrence relations for inverse factorial moments of some discrete distributions can be obtained with the properties of the generalized hypergeometric series functions.

Theorem 6.

Let X be a generalized negative binomial random variable with parameters θ, β, for 0<θ<1, |θβ|<1, 1βθ-1, and probability mass function is defined in (3), and then the factorial inverse moment of first order is given by (24)Ei=1lx+i-1=mθ1-θm+β-1l+1!F322,m+n+1β-n,1;l+2,2;θ1-θ1-β,where l1 and lZ+, n=0,1,2,.

Proof.

Since X is a generalized negative binomial random variable with parameters θ, β, then (25)Ei=1lx+i-1=x=1lmm+βxm+βxxθx1-θm+βx-xi=1l1x+i=x=1lmm+βxm+βxxθx1-θm+βx-xx!l+x!=mθ1-θm+β-1l+1!1+2m+2β-1·1l+2·2·1!θ1-θ1-β+2·3m+3β-1m+3β-2·1·2l+2l+3·2·3·2!θ1-θ1-β2+=mθ1-θm+β-1l+1!F21m+n+1β-n,1;l+2;θ1-θ1-β.

Theorem 7.

Let X equipped with generalized negative binomial probability distribution be defined in (3) with parameters θ and β; suppose μ-[l] is the lth negative factorial moment of X. Then the relation (26)l+1p-l-2zμ-l+1=p-2l-2z+l+1μ-l-1-zμ-l-1holds for l=2,3,4,, and p=m+(n+1)β-n, n=0,1,2,, z=θ/(1-θ)1-β.

Proof.

From Theorem 6, we have (27)μ-l=mθ1-θm+β-1l+1!F211,p;l+2;z.Using the identity (see [4, page 71]) (28)a-c+1F21a,b;c;z=aF21a+1,b;c;z-c-1F21a,b;c-1;z,and for a=1, b=p, c=l+3, and z=z we have (29)F211,p;l+3;z=l+2l+1F211,p;l+2;z-1l+1F212,p;l+3;z.Using the identity (see [4, page 71]) (30)1-zF21a1,b1;c1;z=F21a1-1,b1;c1;z-c1-b1c1zF21a1,b1;c1+1;z,for a1=2, b1=p, c=l+2, and  z=z; rearranging we have (31)F212,p;l+3;z=l+21-zp-l-2zF212,p;l+2;z-l+2p-l-2zF211,p;l+2;z.Substituting in (29), we get (32)F211,p;l+3;z=l+2p-l-2z+1l+1p-l-2zF211,p;l+2;z-l+21-zl+1p-l-2zF212,p;l+2;z.Consider F21[2,p;l+2;z] and using (28), for a=1, b=p, c=l+2, and z=z we have (33)F212,p;l+2;z=l+1F211,p;l+1;z-lF211,p;l+2;z.Substituting in (32), we get (34)F211,p;l+3;z=l+2p-l-2z+1l+1p-l-2zF211,p;l+2;z-l+21-zp-l-2zF211,p;l+1;z+ll+21-zl+1p-l-2zF211,p;l+2;z=l+2p-2l-2z+l+1l+1p-l-2zF211,p;l+2;z-l+21-zp-l-2zF211,p;l+1;z.Rearranging we have (35)l+2!mθ1-θm+β-1μ-l+1=l+2p-2l-2z+l+1l+1p-l-2zl+1!mθ1-θm+β-1μ-l-l+21-zl+1p-l-2zl+1!mθ1-θm+β-1μ-l-1,and by collating, we get the result (26).

Theorem 8.

Let the random variable X equipped with a generalized Poisson distribution be defined in (4) with parameters λ and θ, and μ-[l] is the lth negative factorial moment of X. Then the relation (36)l+1zμ-l+1=l+z+1μ-l-μ-l-1holds for l=2,3,4,, z=(1+(n+1)θ)λ/eλθ, λ,θ>0, n=0,1,2,.

Proof.

Using the identity (see [4, pages 82 and 84]) (37)α1-β1+1F22α1,α2;β1,β2;x=α1F22α1+1,α2;β1,β2;x-β1-1F22α1,α2;β1-1,β2;x,F22α1,α2;β1,β2;x=F22α1-1,α2;β1,β2;x+xα2-β1β1β2-β1F22α1,α2;β1+1,β2;x+xα2-β2β2β1-β2F22α1,α2;β1,β2+1;x,by the same as Theorem 7 calculating, we get the result (36).

Theorem 9.

Let X equipped with generalized logarithmic series distribution be defined in (6) with parameters θ and β, and μ-[l] is the lth inverse factorial moment of X. Then the relation (38)l+1p-l-2zμ-l+1=p-2l-2z+l+1μ-l-1-zμ-l-1,holds for l=2,3,4,, p=(n+1)β-n, n=0,1,2,, z=θ/(1-θ)1-β.

Proof.

Using the identities (28) and (30), by the same as Theorem 7 calculating, we get the result (38).

4. The Accurate Value for Inverse Moments of Some Discrete Distributions

In this section, some accurate values for inverse moments of some discrete distributions can be obtained with their recurrence relations.

Theorem 10.

Let X be a generalized negative binomially distributed random variable with parameters θ and β, and probability mass function is defined in (3). Then the accurate value for inverse moment of first order is given by (39)AEX+A-1=i=-1A-3-1iA+1-ii·P1p-A-1i+1·zi+11-z-p+1+-1A-13A-2·P1p-A-1A-1·zA-1F211,p;3;z+i=0A-2-1i·A+1-iiA-i+p-A+1-izp-A-1i+1·zi+1F211,p;2;z,where p=m+(n+1)β-n, n=0,1,2,, z=θ/(1-θ)1-β, P1=P(X=1)=mθ(1-θ)m+β+1.

Proof.

From (9), we have (40)Azp-A-1EX+A-1=A1-AEX+A-1-1-P11-z-p+1+A+p-A-1zF211,p;2;z,and let A=A-1; we get (41)A-1zp-AEX+A-1-1=A-12-AEX+A-2-1-P11-z-p+1+A-1+p-AzF211,p;2;z,and hence (42)AEX+A-1=AA-1A-2p-A-1p-A·z2EX+A-2-1+A·P11-z-p+1p-A-1p-A·z2-AA-1+p-Azp-A-1p-A·z2F211,p;2;z-P11-z-p+1p-A-1·z+A+p-A-1zp-A-1·zF211,p;2;z,and so on; repeat the above steps; we can get (43)AEX+A-1=-1A-12A-1p-A-1A-1·zA-1·EX+1-1+-1A-1·3A-2·P11-z-p+1p-A-1A-1·zA-1+-1A-2·4A-3·P11-z-p+1p-A-1A-2·zA-2++-12A1·P11-z-p+1p-A-12·z2+-11·A+10·P11-z-p+1p-A-11·z+-1A-2·3A-22+p-3zp-A-1A-1·zA-1·F211,p;2;z+-1A-3·4A-33+p-4zp-A-1A-2·zA-2·F211,p;2;z++-11·A1A-1+p-Azp-A-12·z2·F211,p;2;z+-10·A+10A+p-A-1zp-A-11·z·F211,p;2;z,and, from (10), we have (44)EX+1-1=P12F211,p;3;z,so we obtain the result (39).

Theorem 11.

Let the random variable X be equipped with a generalized Poisson distribution with parameters θ and λ, and probability mass function is defined in (4), and then the accurate value for inverse moment of first order is given by (45)AEX+A-1=i=-1A-3-1iA-ii+1·P1zi+2+-1A-13A-2·P1zA-1F111;3;z+i=0A-2-1iA-ii+1·P1zi+1F111;2;z,where P1=P(X=1)=λe-λ(1+θ), z=λ(1+(n+1)θ)/eλθ, and n=0,1,2,.

The proof is the same as Theorem 10, omitted here.

Theorem 12.

Let the random variable X be equipped with a generalized poisson-negative-binomial distribution with parameters λ, θ, and β, and probability mass function is defined in (5), and then the accurate value for inverse moment of first order is given by (46)AEX+A-1=i=-1A-3-1iA-ii+1·P1p-A-1i+2·zi+21-z-p+1+-1A-13A-2·P1p-A-1A-1·zA-1F211,p;3;z+i=0A-2-1i·A+1-iiA-i-p-A+1-iz·P1p-A-1i+1·zi+1F211,p;2;z,where P1=P(X=1)=e-λ1+θss=0(λs(1+θs)s-1/s!)mθ(1-θ)m+β-1, p=m+(n+1)β-n, n=0,1,2,, and z=θ/(1-θ)1-β.

The proof is the same as Theorem 10, omitted here.

Theorem 13.

Let X be a generalized logarithmic distributed random variable with parameters θ and β, and probability mass function is defined in (6). Then the accurate value for inverse moment of first order is given by (47)AEX+A-1=i=-1A-3-1iA-ii+1·P1p-A-1i+2·zi+21-z-p+1+-1A-13A-2·P1p-A-1A-1·zA-1F211,p;3;z+i=0A-2-1i·A+1-iiA-i+p-A+1-iz·P1p-A-1i+1·zi+1F211,p;2;z,where P1=P(X=1)=θ(1-θ)β-1/[-log(1-θ)], p=(n+1)β-n, n=0,1,2,, and z=θ/(1-θ)1-β.

The proof is the same as Theorem 10, omitted here.

Conflict of Interests

The authors declare that there is no conflict of interests regarding the publication of this paper.

Acknowledgments

The research is supported by the National Natural Science Foundation of China under Grant 11461050 and Natural Science Foundation of Inner Mongolia 2012MS0118.

Kumar A. Consul P. C. Negative moments of a modified power series distribution and bias of the maximum likelihood estimator Communications in Statistics 1979 8 2 151 166 Chao M. T. Strawderman W. E. Negative moments of positive random variables Journal of the American Statistical Association 1972 67 338 429 431 10.2307/2284399 Ahmad M. Saboor A. Properties of a newly defined hypergeometric power series function Pakistan Journal of Statistics 2009 25 2 227 233 MR2668158 2-s2.0-67651174839 Rainville E. D. Special Functions 1960 New York, NY, USA Chelsea MR0107725