Skip to main content
The Scientific World Journal logoLink to The Scientific World Journal
. 2014 May 7;2014:940358. doi: 10.1155/2014/940358

New Proofs of Some q-Summation and q-Transformation Formulas

Xian-Fang Liu 1, Ya-Qing Bi 1, Qiu-Ming Luo 1,*
PMCID: PMC4033541  PMID: 24895675

Abstract

We obtain an expectation formula and give the probabilistic proofs of some summation and transformation formulas of q-series based on our expectation formula. Although these formulas in themselves are not the probability results, the proofs given are based on probabilistic concepts.

1. Introduction

The probabilistic method is an important tool to derive results in combinatorics, theory of numbers, and other fields (see [115]). There have been many applications in the basic hypergeometric series (or q-series). For example, Fulman [3] presented a probabilistic proof of Rogers-Ramanujan identity using Markov chain. Chapman [2] proved the Andrews-Gordon identity by using extended Fulman's methods. Kadell [4] gave a probabilistic proof of Ramanujan's 1 ψ 1 summation based on the order statistics.

Recently, Wang [13, 14] constructed a random variable X and introduced a new probability distribution W(x; q):

P(X=xnqk)=pn,k(x;q)=(x)nqk(xn1qk+1,xnqk+1;q)(q,q/x,x;q), (1)

where

pn,k(x;q)>0,pn,k(x;q)=1,x<0,0<q<1,n=0,1,k=0,1,2,. (2)

By applying the above probability distribution, Wang proved the q-binomial theorem and q-Gauss summation formula and also obtained some new summation formulas and transformation formulas.

One of the most important concepts in probability theory is that of the expectation of a random variable. If X is a discrete random variable having a probability mass function p(x), then the expectation, or the expected value, or the expectation operator of X, denoted by E[X], is defined by (e.g., [9, page 125])

E[X]=p(x)>0xp(x). (3)

In the following section we introduce some notations, definitions, and formulas of q-series. Throughout this paper we suppose qC, |q| < 1.

The q-shifted factorials are defined by

(a;q)0=1,(a;q)n=k=0n1(1aqk),(a;q)=k=0(1aqk),n1. (4)

Clearly,

(a;q)n=(a;q)(aqn;q),(a2;q2)n=(a;q)n(a;q)n. (5)

The following are compact notations for the multiple q-shifted factorials:

(a1,a2,,am;q)n=(a1;q)n(a2;q)n(am;q)n,(a1,a2,,am;q)=(a1;q)(a2;q)(am;q). (6)

The basic hypergeometric series or q-series r ϕ s are defined by (see [16, 17])

ϕsr(a1,a2,,arb1,b2,,bs;q,z)=n=0(a1,a2,,ar;q)n(q,b1,b2,,bs;q)n[(1)nq(n2)]1+srzn. (7)

Heine introduced the r + 1 ϕ r basic hypergeometric series which is defined by

ϕrr+1(a1,a2,,ar+1b1,b2,,br;q,z)=n=0(a1,a2,,ar;q)n(q,b1,b2,,br;q)nzn. (8)

Jackson defined the q-integral (see [17, 18]):

0df(t)dqt=d(1q)n=0f(dqn)qn, (9)
cdf(t)dqt=0df(t)dqt0cf(t)dqt. (10)

The following is the Andrews-Askey integral (see [19]) which can be derived from Ramanujan's 1 ψ 1:

cd(qt/c,qt/d;q)(at,bt;q)dqt=d(1q)(q,dq/c,c/d,abcd;q)(ac,ad,bc,bd;q), (11)

provided that there are no zero factors in the denominator of the integrals. Recently, Liu and Luo [20] further generalized the above Andrews-Askey integral in the following more general form.

Lemma 1 (see [21, page 5. (2.5)] [20, Theorem 1]) —

One has

ab(qt/a,qt/b,ct;q)(et,ft,ht;q)dqt=b(1q)(q,bq/a,a/b,bc,abef;q)(ae,af,be,bf,bh;q)ϕ23×(be,bf,chabef,bc;q,ah), (12)

provided |ah| < 1 and |q| < 1, provided that there are no zero factors in the denominator of the integrals.

Lemma 2 (see [21, page 5. (2.7)]) —

One has

ab(qt/a,qt/b,ct;q)(et,ht;q)dqt=b(1q)(q,bq/a,a/b,bc;q)(ae,be,bh;q)ϕ12(be,chbc;q,ah), (13)

provided |ah| < 1 and |q| < 1, provided that there are no zero factors in the denominator of the integrals.

The aim of the present paper is to give an expectation formula and introduce some probabilistic proofs of the corresponding summation and transformation formulas of q-series based on an expectation formula. In Section 2 we give an expectation formula of the random variables (dX;q)/(aX,bX,cX;q). In Section 3 we show the probabilistic proofs of transformation formulas of 3 ϕ 2. In Section 4 we give probabilistic proof of Heine's transformations and Jackson's transformations. In Section 5 we give probabilistic proof of some formulas of q-series, for example, q-binomial theorem, q-Chu-Vandermonde sum formulas, q-Gauss sum formula, q-Kummer sum formula, Bailey sum formula, and so forth.

2. Main Theorem

In this section we obtain the expectation formulas of some random variables which are very useful to prove the summation and transformation formulas of q-series.

Theorem 3 —

Let X denote a random variable with probability distribution W(x; q), −1 < x < 0. Then one has

E[(dX;q)(aX,bX,cX;q)]=(d,abx;q)(a,ax,b,bx,c;q)ϕ23(a,b,dcabx,d;q,cx), (14)

provided that max⁡(|a|, |b|, |c|, |d|) < 1, |cx| < 1, and |q| < 1.

Proof —

A random variable X has the distribution W(x; q). From definitions (9) we have

01(qt/x,qt,dt;q)(at,bt,ct;q)dqt=(1q)k=0(qk+1/x,qk+1,dqk;q)qk(aqk,bqk,cqk;q),0x(qt/x,qt,dt;q)(at,bt,ct;q)dqt=x(1q)k=0(qk+1,xqk+1,dxqk;q)qk(axqk,bxqk,cxqk;q). (15)

From definitions (10) and combining (15) we have

x1(qt/x,qt,dt;q)(at,bt,ct;q)dqt=(1q)k=0(qk+1/x,qk+1,dqk;q)qk(aqk,bqk,cqk;q)x(1q)k=0(qk+1,xqk+1,dxqk;q)qk(axqk,bxqk,cxqk;q). (16)

By using the probability distribution W(x; q) and noting (16) and (12) of Lemma 1, we calculate the expectation of the random variable (dX;q)/(aX,bX,cX;q) as follows:

E[(dX;q)(aX,bX,cX;q)]=n=01k=0(x)n(xn1qk+1,xnqk+1;q)qk(q,q/x,x;q)×(dxnqk;q)(axnqk,bxnqk,cxnqk;q)=1(1q)(q,q/x,x;q)×((1q)k=0(qk+1/x,qk+1,dqk;q)qk(aqk,bqk,cqk;q)x(1q)k=0(qk+1,xqk+1,dxqk;q)qk(axqk,bxqk,cxqk;q))=1(1q)(q,q/x,x;q)x1(qt/x,qt,dt;q)(at,bt,ct;q)dqt=1(1q)(q,q/x,x;q)(1q)(q,q/x,x,d,abx;q)(ax,bx,a,b,c;q)ϕ23.×(a,b,dcabx,d;q,cx). (17)

Hence, we obtain

E[(dX;q)(aX,bX,cX;q)]=(d,abx;q)(ax,a,bx,b,c;q)ϕ23(a,b,dcabx,d;q,cx). (18)

The proof is complete.

Theorem 4 —

Let X denote a random variable with probability distribution W(x; q), −1 < x < 0. Then one has

E[(dX;q)(aX,bX;q)]=(d,abx;q)(a,ax,b,bx;q)ϕ22(a,babx,d;q,dx), (19)

provided that max⁡(|a|, |b|, |d|) < 1, |dx| < 1, and |q| < 1.

Proof —

By (7) and (8) we have

ϕ23(a,b,dcabx,d;q,cx)=n=0(a;q)n(b;q)n(d/c;q)n(cx)n(q;q)n(abx;q)n(d;q)n=n=0(a;q)n(b;q)nxn(q;q)n(abx;q)n(d;q)n(dc;q)ncn. (20)

By (4) we have

(dc;q)ncn=(1dc)(1dcq)(1dcqn1)cn=(cd)(cdq)(cdqn1). (21)

Substituting (20) and (21) into the right-hand side of (14), we obtain

E[(dX;q)(aX,bX,cX;q)]=(d,abx;q)(a,ax,b,bx,c;q)n=0(a;q)n(b;q)nxn(q;q)n(abx;q)n(d;q)n×(cd)(cdq)(cdqn1). (22)

Next, let us replace c by λc, respectively, and let λ → 0 in (22); we get

E[(dX;q)(aX,bX;q)]=(d,abx;q)(a,ax,b,bx;q)n=0(a;q)n(b;q)nxn(q;q)n(abx;q)n(d;q)n×(d)(dq)(dqn1)=(d,abx;q)(a,ax,b,bx;q)n=0(a;q)n(b;q)nxn(q;q)n(abx;q)n(d;q)n×dn(1)nqn(n1)/2=(d,abx;q)(a,ax,b,bx;q)n=0(a;q)n(b;q)n(dx)n(q;q)n(abx;q)n(d;q)n(1)nq(n2)=(d,abx;q)(a,ax,b,bx;q)ϕ22(a,babx,d;q,dx). (23)

The proof is complete.

Theorem 5 —

Let X denote a random variable with probability distribution W(x; q), −1 < x < 0. Then one has

E[(cxX;q)(aX,bX,cX;q)]=(cx,abx;q)(ax,bx,a,b,c;q)ϕ23(a,b,xabx,cx;q,cx), (24)

provided that max⁡(|a|, |b|, |c|) < 1, |cx| < 1, and |q| < 1.

Proof —

Letting d = cx in (14) of Theorem 3, we obtain (24).

Corollary 6 (see [13, page 463, Theorem 1]) —

Let X denote a random variable with probability distribution W(x; q), −1 < x < 0. Then one has

E[(cxX;q)(aX,bX,cX;q)]=(acx,bcx,x;q)(a,ax,b,bx,c,cx;q)ϕ23(ab,c,cxacx,bcx;q,x), (25)

provided that max⁡(|a|, |b|, |c|) < 1.

Proof —

Using (24) of Theorem 5, we deduce

E[(cxX;q)(aX,bX,cX;q)]=(cx,abx;q)(ax,bx,a,b,c;q)ϕ23(a,b,xabx,cx;q,cx). (26)

Using (31) of Theorem 8, we have

ϕ23(a,b,xabx,cx;q,cx)=ϕ23(a,x,babx,cx;q,cx)=(x,bcx,acx;q)(abx,cx,cx;q)ϕ23(ab,c,cxbcx,acx;q,x). (27)

Substituting (27) into the right-hand sides of (26), we have

E[(cxX;q)(aX,bX,cX;q)]=(cx,abx;q)(ax,bx,a,b,c;q)(x,bcx,acx;q)(abx,cx,cx;q)ϕ23×(ab,c,cxbcx,acx;q,x)=(acx,bcx,x;q)(a,ax,b,bx,c,cx;q)ϕ23(ab,c,cxacx,bcx;q,x). (28)

The proof is complete.

Corollary 7 (see [14, page 245, Lemma 2.4]) —

Let X denote a random variable with probability distribution W(x; q), −1 < x < 0. Then one has

E[1(aX,bX;q)]=(abx;q)(ax,bx,a,b;q), (29)

provided that max⁡(|a|, |b|) < 1.

Proof —

Letting d = c or d = a = c in (14) of Theorem 3, then we have

E[1(aX,bX;q)]=(abx;q)(ax,bx,a,b;q)ϕ23(a,b,1abx,d;q,cx)=(abx;q)(ax,bx,a,b;q). (30)

The proof is complete.

3. Probabilistic Proofs of Transformation Formulas of 3 ϕ 2

Sears' 3 ϕ 2 transformation formula is widely applied to the special functions. In this section we will introduce probabilistic proofs of transformation of 3 ϕ 2.

Theorem 8 (see [17, page 359. III. 9, III. 10]) —

One has

ϕ23(a,b,cd,e;q,deabc)=(e/a,de/bc;q)(e,de/abc;q)ϕ23(a,db,dcd,debc;q,ea) (31)
=(b,de/ab,de/bc;q)(d,e,de/abc;q)ϕ23(db,eb,deabcdeab,debc;q,b). (32)

Proof —

Interchanging b and c in (14), then we have

E[(dX;q)(aX,bX,cX;q)]=(d,acx;q)(ax,a,cx,c,b;q)ϕ23(a,c,dbacx,d;q,bx). (33)

Interchanging a and c in (14), then we have

E[(dX;q)(aX,bX,cX;q)]=(d,bcx;q)(bx,b,cx,c,a;q)ϕ23(c,b,dabcx,d;q,ax). (34)

By (14) and (33), we obtain

ϕ23(a,b,dcabx,d;q,cx)=(bx,acx;q)(cx,abx;q)ϕ23(a,c,dbacx,d;q,bx), (35)

and, replacing (a, b, d/c, d, abx) by (a, b, c, d, e) in (35), we obtain a 3 ϕ 2 transformation formula

ϕ23(a,b,cd,e;q,deabc)=(e/a,de/bc;q)(e,de/abc;q)ϕ23(a,db,dcd,debc;q,ea). (36)

By (14) and (34) and then replacing (b, a, d/c, d, abx) by (a, b, c, d, e), we obtain (31).

By (33) and (34), we have

ϕ23(c,b,dabcx,d;q,ax)=(bx,acx;q)(ax,bcx;q)ϕ23(a,c,dbacx,d;q,bx), (37)

and, replacing (c, b, d/a, d, b cx) by (a, b, c, d, e) in (37), we obtain (32). The proof is complete.

4. Probabilistic Proof of Heine and Jackson's Transformations

Heine [22] derived transformation formulas for 2 ϕ 1 and also proved Euler's transformation formula. A basic hypergeometric representation for a given function is by no means unique. There are groups of transformation between various hypergeometric representations of the same function. We will first prove the classical Heine's transformation formula which will be useful in proving many other formulas. In this section we give the probabilistic proofs of Heine and Jackson's transformations.

Theorem 9 (see [17, page 359, III. 1, III. 2, III. 3]) —

Heine's transformation formulas for 2 ϕ 1 are

ϕ12(a,bc;q,z)=(b,az;q)(c,z;q)ϕ12(cb,zaz;q,b) (38)
=(c/b,bz;q)(c,z;q)ϕ12(abzc,bbz;q,cb) (39)
=(abz/c;q)(z;q)ϕ12(ca,cbc;q,abzc). (40)

Proof —

Comparing (24) of Theorem 5 and (25) of Corollary 6, we obtain

(acx,bcx,x;q)(a,ax,b,bx,c,cx;q)ϕ23(ab,c,cxacx,bcx;q,x)=(cx,abx;q)(ax,bx,a,b,c;q)ϕ23(a,b,xabx,cx;q,cx), (41)

or, equivalently, that

ϕ23(ab,c,cxacx,bcx;q,x)=(cx,abx,cx;q)(acx,bcx,x;q)ϕ23(a,b,xabx,cx;q,cx). (42)

Setting b = 0 in (42), we have

ϕ12(c,cxacx;q,x)=(cx,cx;q)(acx,x;q)ϕ12(a,xcx;q,cx). (43)

Replacing (c, cx, a cx, x) by (a, b, c, z) in (43), we get

ϕ12(a,bc;q,z)=(b,az;q)(c,z;q)ϕ12(cb,zaz;q,b)for  |b|<1,|z|<1, (44)

which is just (38).

Setting d = 0 and a = b and replacing c by b in (14), we have

E[1(aX,aX,bX;q)]=(a2x;q)(ax,ax,a,a,b;q)ϕ12(a,aa2x;q,bx). (45)

Setting d = 0 and a = c in (14) of Theorem 3, we have

E[1(aX,aX,bX;q)]=(abx;q)(ax,bx,a,a,b;q)ϕ12(a,babx;q,ax). (46)

Comparing (45) and (46), we obtain

ϕ12(b,aabx;q,ax)=(bx,a2x;q)(abx,ax;q)ϕ12(a,aa2x;q,bx). (47)

Replacing (a, b, x) by (b, a, z/b), we get

ϕ12(a,baz;q,z)=(az/b,bz;q)(az,z;q)ϕ12(b,bbz;q,azb). (48)

Letting c = az in (48) gives

ϕ12(a,bc;q,z)=(c/b,bz;q)(c,z;q)ϕ12(abzc,bbz;q,cb). (49)

We get (39). From (39) we can deduce (40).

Jackson's transformations formula is an important formula in basic hypergeometric series, and now we give a probabilistic proof of Jackson's transformation formulas for 2 ϕ 1 and 2 ϕ 2.

Theorem 10 (see [17, page 359, III.4]) —

Jackson's transformations of 2 ϕ 1, 2 ϕ 2 series are

ϕ12(a,bc;q,z)=(az;q)(z;q)ϕ22(a,cbc,az;q,bz). (50)

Proof —

This includes employing two different forms of E[(dX;q)/(aX,bX;q)].

Letting b = 0 in (14) of Theorem 3 and then replacing c by b, we get

E[(dX;q)(aX,bX;q)]=(d;q)(a,ax,b;q)ϕ12(a,dbd;q,bx). (51)

Comparing (51) and (19) of Theorem 4 gives

(d;q)(a,ax,b;q)ϕ12(a,dbd;q,bx)=(d,abx;q)(a,ax,b,bx;q)ϕ22(a,babx,d;q,dx). (52)

Then we obtain

ϕ12(a,dbd;q,bx)=(abx;q)(bx;q)ϕ22(a,babx,d;q,dx). (53)

Replacing (a, d/b, d, bx) by (a, b, c, z) gives

ϕ12(a,bc;q,z)=(az;q)(z;q)ϕ22(a,cbc,az;q,bz). (54)

This completes the proof.

5. Probabilistic Proofs of Some Formulas of q-Series

The q-binomial theorem is an important mathematical result which has been widely applied in the special functions, physics, quantum algebra, and quantum statistics. The q-binomial theorem was derived by Cauchy [23], Heine [22], and Jacobi [24] concerning the nonterminating form. There are many proofs of the q-binomial theorem to show the corresponding references; for example, a better and simpler proof, by using the method of the finite difference, was obtained by Askey (see [25]); a nice proof of the q-binomial theorem based on combinatorial considerations was given by Joichi and Stanton (see [26]). In 1847, Heine [22] derived a q-analogue of Gauss's summation formula which is important in q-series. Joichi and Stanton [26] gave a bijective proof of the q-Gauss summation formula based on combinatorial considerations. Rahman and Suslov [27] used the method of the first order linear difference equations to prove the q-Gauss summation formula. By analytic continuation, the terminating case, when a = q n, reduces to q-analogues of Vandermonde's formula. Bailey and Daum independently discovered the q-Kummer summation formula.

In this section we will introduce probabilistic proof of some formulas of q-series, for example, q-binomial theorem, q-Chu-Vandermonde, q-Gauss summation formula and q-Kummer summation formula, and so forth.

Theorem 11 (see [16, page 488, Theorem 10.2.1] [17, page 354. II. 3]) —

The q-binomial theorem is

ϕ01(a;q,z)=n=0(a;q)n(q;q)nzn=(az;q)(z;q)for  |z|<1,|q|<1. (55)

Proof —

Below we give two proofs of (55).

Setting d = a = 0 and replacing b and c by a and b in (14), we obtain

E[1(aX,bX;q)]=1(ax,a,b;q)ϕ01(a;q,bx). (56)

Comparing (56) and (29) of Corollary 7, we have

1(ax,a,b;q)ϕ01(a;q,bx)=(abx;q)(ax,bx,a,b;q). (57)

Then we obtain

ϕ01(a;q,bx)=(abx;q)(bx;q). (58)

Replacing bx by z, we can get

ϕ01(a;q,z)=(az;q)(z;q); (59)

that is,

ϕ01(a;q,z)=n=0(a;q)n(q;q)nzn=(az;q)(z;q)for|z|<1. (60)

Another proof of the q-binomial theorem is as follows.

Setting d = a = 0 and b = c and replacing b by a in (14), we obtain

E[1(aX;q)2]=1(ax,a,a;q)ϕ01(a;q,ax). (61)

Letting a = b = c = d or a = b and d = c in (14) of Theorem 3, we obtain

E[1(aX;q)2]=(a,a2x;q)(ax,ax,a,a,a;q)ϕ23(a,a,1a2x,a;q,ax)=(a,a2x;q)(ax,ax,a,a,a;q)=(a2x;q)(ax,a;q)2. (62)

Comparing (61) and (62) gives

1(ax,a,a;q)ϕ01(a;q,ax)=(a2x;q)(ax,a;q)2. (63)

Then we obtain

ϕ01(a;q,ax)=(a2x;q)(ax;q). (64)

Replacing ax by z gives

ϕ01(a;q,z)=(az;q)(z;q); (65)

that is,

ϕ01(a;q,z)=n=0(a;q)n(q;q)nzn=(az;q)(z;q)for|z|<1. (66)

This proof is complete.

Theorem 12 (see [17, page 354, II. 7]) —

The q-Chu-Vandermonde sums are

ϕ12(a,qnc;q,cqna)=(c/a;q)n(c;q)n. (67)

Proof —

The below are two proofs of the q-Chu-Vandermonde.

  • (i)
    First proof: setting d = a = b and replacing c by b in (14), we have
    E[1(aX,bX;q)]=(a2x;q)(ax,ax,a,b;q)ϕ12(a,aba2x;q,bx). (68)

Replacing (a, b, x) by (a, aq n, c/a 2) in (68), then we have

P(Y=(ca2)nqk)  =pn,k(ca2;q)=(c/a2)nqk((c/a2)n1qk+1,(c/a2)nqk+1;q)(q,a2q/c,c/a2;q), (69)

where

pn,k(ca2;q)>0,  pn,k(ca2;q)=1,ca2<0,0<q<1,n=0,1,k=0,1,2,. (70)

Hence,

E[1(aY,aqnY;q)]=(c;q)(c/a,c/a,a,aqn;q)ϕ12(a,qnc;q,cqna). (71)

By using the probability distribution W(c/a 2; q) and employing Andrews-Askey q-integral (11), now we calculate the expectation of the random variables 1/(aY,aq n Y;q) as follows:

E[1(aY,aqnY;q)]=n=01k=0(c/a2)nqk((c/a2)n1qk+1,(c/a2)nqk+1;q)(q,a2q/c,c/a2,a(c/a2)nqk,(aqn)(c/a2)nqk;q)=1(1q)(q,a2q/c,c/a2;q)×((1q)k=0(qk+1/(c/a2),qk+1;q)qk(aqk,(aqn)qk;q)(ca2)(1q)×k=0(qk+1,(c/a2)qk+1;q)qk(a(c/a2)qk,(aqn)(c/a2)qk;q))=1(1q)(q,a2q/c,c/a2;q)c/a21(qt/(c/a2),qt;q)(at,aqnt;q)dqt=1(1q)(q,a2q/c,c/a2;q)(1q)(q,a2q/c,c/a2,cqn;q)(c/a,cqn/a,a,aqn;q)=(cqn;q)(c/a,cqn/a,a,aqn;q). (72)

Comparing (71) and (72) gives

(c;q)(c/a,c/a,a,aqn;q)ϕ12(a,qnc;q,cqna)=(cqn;q)(c/a,cqn/a,a,aqn;q). (73)

Then we obtain

ϕ12(a,qnc;q,cqna)=(cqn;q)(c/a,cqn/a,a,aqn;q)(c/a,c/a,a,aqn;q)(c;q)=(cqn;q)(c/a;q)(c;q)(cqn/a;q)=(c/a;q)n(c;q)n, (74)

which is just q-Vandermonde sums (67).

  • (ii)
    Second proof: replacing (a, b, x) by (a, aq n, c/a 2) in (29), we have
    E[1(aY,aqnY;q)]=(cqn;q)(c/a,cqn/a,a,aqn;q). (75)

Comparing (71) and (75), we obtain

(c;q)(c/a,c/a,a,aqn;q)ϕ12(a,qnc;q,cqna)=(cqn;q)(c/a,cqn/a,a,aqn;q). (76)

Hence,

ϕ12(a,qnc;q,cqna)=(cqn;q)(c/a;q)(c;q)(cqn/a;q)=(c/a;q)n(c;q)n, (77)

which is just q-Vandermonde sums (67).

Theorem 13 (see [16, page 522, Corollary 10.9.2] or [17, page 354, II. 8]) —

The q-Gauss sum is

ϕ12(a,bc;q,cab)=(c/a,c/b;q)(c,c/ab;q). (78)

Proof —

Letting d = a = b and replacing c by b in (14), we obtain

E[1(aX,bX;q)]=(a2x;q)(ax,ax,a,b;q)ϕ12(a,aba2x;q,bx). (79)

Comparing (29) and (79) gives

(a2x;q)(ax,ax,a,b;q)ϕ12(a,aba2x;q,bx)=(abx;q)(ax,bx,a,b;q); (80)

hence we get

ϕ12(a,aba2x;q,bx)=(abx,ax;q)(a2x,bx;q). (81)

Replacing (a, a/b, a 2 x) by (a, b, c) in the above formula, we obtain

ϕ12(a,bc;q,cab)=(c/a,c/b;q)(c,c/ab;q), (82)

which is just the q-Gauss sum (78).

Theorem 14 (see [17, page 354, II. 9]) —

The q-Kummer sum formula is

ϕ12(a,baqb;q,qb)=(q;q)(  aq,aq2/b2;q2)(q/b,aq/b;q). (83)

Proof —

Letting b = 0 in (14) and then replacing c by b, we have

E[(dX;q)(aX,bX;q)]=(d;q)(a,ax,b;q)ϕ12(a,dbd;q,bx). (84)

Replacing (a, b, d, x) by (a, aq/b 2, aq/b, −b/a) in (84), we write

P(Z=(ba)nqk)=pn,k(ba;q)=(b/a)nqk((b/a)n1qk+1,(b/a)nqk+1;q)(q,aq/b,b/a;q), (85)

where

pn,k(ba;q)>0,pn,k(ba;q)=1,ba<0,0<q<1,n=0,1,k=0,1,2,. (86)

Hence, we obtain

E[((aq/b)Z;q)(aZ,(aq/b2)Z;q)]=(aq/b;q)(a,b,aq/b2;q)ϕ12(a,baqb;q,qb). (87)

By using the probability distribution W(−b/a; q) and Lemma 2, we calculate the expectation of the random variables ((aq/b)Z;q)/(aZ,(aq/b 2)Z;q) as follows:

E[((aq/b)Z;q)(aZ,(aq/b2)Z;q)]=E[((aq/b)Z;q)((aq/b2)Z,aZ;q)]=n=01k=0((ba)nqk((ba)n1qk+1,(ba)nqk+1,(aqb)(ba)nqk;q)×((q,aqb,ba,(aqb2)(ba)nqk,a(ba)nqk;q))1)=1(1q)(q,aq/b,b/a;q)×((1q)k=0(qk+1/(b/a),qk+1,(aq/b)qk;q)qk((aq/b2)qk,aqk;q)(ba)(1q)×k=0(qk+1,(b/a)qk+1,(aq/b)(b/a)qk;q)qk((aq/b2)(b/a)qk,a(b/a)qk;q))=1(1q)(q,aq/b,b/a;q)×b/a1(qt/(b/a),qt,(aq/b)t;q)((aq/b2)t,at;q)dqt=(1q)(q,aq/b,b/a,aq/b;q)(1q)(q,aq/b,b/a,q/b,aq/b2,a;q)ϕ12×(aqb2,qbaqb;q,b)=(aq/b;q)(q/b,aq/b2,a;q)ϕ12(aqb2,qbaqb;q,b). (88)

Comparing (87) and (88), we have

ϕ12(a,baqb;q,qb)=(a,b,aq/b2;q)(aq/b;q)(aq/b;q)(q/b,aq/b2,a;q)ϕ12×(aqb2,qbaqb;q,b)=(b;q)(q/b;q)ϕ12(aqb2,qbaqb;q,b). (89)

Using Heine's transformation and q-binomial theorem, we have

ϕ12(a,baqb;q,qb)=(b;q)(q/b;q)(a,q;q)(aq/b,b;q)ϕ12(qb,qbq;q,a)=(a,q;q)(q/b,aq/b;q)n=0(q2/b2;q2)n(q2;q2)nan=(a,q;q)(q/b,aq/b;q)(aq2/b2;q2)(a;q2)=(q;q)(aq,aq2/b2;q2)(q/b,aq/b;q). (90)

Hence, we obtain (83).

Theorem 15 (see [17, page 354, II. 10]) —

Bailey's sum formula is

ϕ22(a,qaq,b;q,b)=(ab,bq/a;q2)(b;q). (91)

Proof —

By (19), we have

E[(dX;q)(aX,bX;q)]=(d,abx;q)(a,ax,b,bx;q)ϕ22(a,babx,d;q,dx). (92)

Replacing (a, b, d, x) by (a, q/a, −q, b/q) in (92) gives

P(R=(bq)nqk)=pn,k(bq;q)=(b/q)nqk((b/q)n1qk+1,(b/q)nqk+1;q)(q,q2/b,b/q;q), (93)

where

pn,k(bq;q)>0,pn,k(bq;q)=1,bq<0,0<q<1,n=0,1,k=0,1,2,. (94)

Hence, we have

E[(qR;q)(aR,(q/a)R;q)]=(q,b;q)(a,ab/q,q/a,b/a;q)ϕ22(a,qab,q;q,b). (95)

By using the probability distribution W(b/q; q) and Lemma 2, we calculate the expectation of the random variables (−qR; q)/(aR,(q/a)R;q) as follows:

E[(qR;q)(aR,(q/a)R;q)]=n=01k=0((bq)nqk((bq)n1qk+1,(bq)nqk+1,(q)(bq)nqk;q)×((q,q2b,bq,a(bq)nqk,(qa)(bq)nqk;q))1)=1(1q)(q,q2/b,b/q;q)×((1q)k=0(qk+1/(b/q),qk+1,(q)qk;q)qk(aqk,(q/a)qk;q)(bq)(1q)×k=0(qk+1,(b/q)qk+1,(q)(b/q)qk;q)qk(a(b/q)qk,(q/a)(b/q)qk;q))=1(1q)(q,q2/b,b/q;q)×b/q1(qt/(b/q),qt,qt;q)(at,(q/a)t;q)dqt=1(1q)(q,q2/b,b/q;q)(1q)(q,q2/b,b/q,q;q)(ab/q,a,q/a;q)ϕ12×(a,aq;q,ba)=(q;q)(ab/q,a,q/a;q)ϕ12(a,aq;q,ba). (96)

Comparing (95) and (96), we have

ϕ22(a,qaq,b;q,b)=(b/a;q)(b;q)ϕ12(a,aq;q,ba)=(b/a;q)(b;q)n=0(a,a;q)n(q,q;q)n(ba)n=(b/a;q)(b;q)n=0(a2;q2)n(q2;q2)n(ba)n=(b/a;q)(b;q)(ab;q2)(b/a;q2)=(ab,bq/a;q2)(b;q). (97)

Hence, we get (91).

Theorem 16 (see [17, page 354, II. 11]) —

The Gauss sum formula is

ϕ22(a2,b2abq1/2,abq1/2;q,q)=(a2q,b2q;q2)(q,a2b2q;q2). (98)

Proof —

By (14), we have

E[(dX;q)(aX,bX;q)]=(d,abx;q)(a,ax,b,bx;q)ϕ22(a,babx,d;q,dx). (99)

Replacing (a, b, d, x) by (a 2, b 2, −abq 1/2, q 1/2/ab) in (99) gives

P(S=(q1/2ab)nqk)=pn,k(q1/2ab;q)=(q1/2/ab)nqk((q1/2/ab)n1qk+1,(q1/2/ab)nqk+1;q)(q,abq1/2,q1/2/ab;q), (100)

where

pn,k(q1/2ab;q)>0,pn,k(q1/2ab;q)=1,q1/2ab<0,0<q<1,n=0,1,k=0,1,2,. (101)

Hence,

E[(abq1/2S;q)(a2S,b2S;q)]=(abq1/2,abq1/2;q)(a2,aq1/2/b,b2,bq1/2/a;q)ϕ22×(a2,b2abq1/2,abq1/2;q,q). (102)

By using the probability distribution W(q 1/2/ab; q) and employing Andrews-Askey q-integral (13) of Lemma 2, we calculate the expectation of the random variables (−abq 1/2 S; q)/(a 2 S,b 2 S;q) as follows:

E[(abq1/2S;q)(a2S,b2S;q)]=n=01k=0((q1/2ab)nqk((q1/2ab)n1qk+1,(q1/2ab)nqk+1,(abq1/2)(q1/2ab)nqk;q)×((q,abq1/2,q1/2ab,a2(q1/2ab)nqk,b2(q1/2ab)nqk;q))1)=1(1q)(q,abq1/2,q1/2/ab;q)×((1q)k=0(qk+1/(q1/2/ab),qk+1,(abq1/2)qk;q)qk(a2qk,b2qk;q)(q1/2ab)(1q)×k=0(qk+1,(q1/2/ab)qk+1,(abq1/2)(q1/2/ab)qk;q)qk(a2(q1/2/ab)qk,b2(q1/2/ab)qk;q))=1(1q)(q,abq1/2,q1/2/ab;q)×q1/2/ab1(qt/(q1/2/ab),qt,abq1/2t;q)(a2t,b2t;q)dqt=(1q)(q,abq1/2,q1/2/ab,abq1/2;q)(1q)(q,abq1/2,q1/2/ab,aq1/2/b,a2,b2;q)ϕ12×(a2,aq1/2babq1/2;q,bq1/2a)=(abq1/2;q)(aq1/2/b,a2,b2;q)ϕ12(a2,aq1/2babq1/2;q,bq1/2a). (103)

Comparing (102) and (103), we have

ϕ22(a2,b2abq1/2,abq1/2;q,q)=(a2,aq1/2/b,b2,bq1/2/a;q)(abq1/2,abq1/2;q)(abq1/2;q)(aq1/2/b,a2,b2;q)ϕ12×(a2,aq1/2babq1/2;q,bq1/2a)=(bq1/2/a;q)(abq1/2;q)ϕ12(a2,aq1/2babq1/2;q,bq1/2a). (104)

Using Heine's transformation formula, we have

ϕ12(a2,aq1/2babq1/2;q,bq1/2a)=(b2,q;q)(abq1/2,bq1/2/a;q)ϕ12(aq1/2b,aq1/2bq;q,b2). (105)

Substituting (105) into (104) yields

ϕ22(a2,b2abq1/2,abq1/2;q,q)=(bq1/2/a;q)(abq1/2;q)(b2,q;q)(abq1/2,bq1/2/a;q)ϕ12×(aq1/2b,aq1/2bq;q,b2)=(b2,q;q)(abq1/2,abq1/2;q)n=0(aq1/2/b,aq1/2/b;q)n(q,q;q)n(b2)n=(b2,q;q)(abq1/2,abq1/2;q)n=0(a2q/b2;q2)n(q2;q2)n(b2)n=(b2,q;q)(abq1/2,abq1/2;q)(a2q;q2)(b2;q2). (106)

Noting that

(a2;q2)=(a;q)(a;q),(aq;q2)=(a;q)(a;q2), (107)

we have

(q;q)=(q;q)(q;q)(q;q)=(q2;q2)(q;q)=1(q;q2),(b2;q)(b2;q2)=(b2q;q2),(abq1/2;q)(abq1/2;q)=(a2b2q;q2). (108)

Substituting (108) into (106) yields

ϕ22(a2,b2abq1/2,abq1/2;q,q)=(a2q,b2q;q2)(q,a2b2q;q2). (109)

Hence, we obtain (98).

Theorem 17 (see [17, page 354, II. 5]) —

A sum formula of 1 ϕ 1 is

ϕ11(ac;q,ca)=(c/a;q)(c;q). (110)

Proof —

Letting d = a = b in (19) of Theorem 4, we obtain

E[1(aX;q)]=(a2x;q)(ax,ax,a;q)ϕ11(aa2x;q,ax). (111)

Letting d = a = c and b = 0 in (14) of Theorem 3 or setting d = c and a = 0 and replacing b by a in (14) of Theorem 3, we obtain

E[1(aX;q)]=1(ax,a;q). (112)

Comparing (111) and (112) gives

(a2x;q)(ax,ax,a;q)ϕ11(aa2x;q,ax)=1(ax,a;q); (113)

that is,

ϕ11(aa2x;q,ax)=(ax;q)(a2x;q). (114)

Replacing (a, a 2 x) by (a, c), we get

ϕ11(ac;q,ca)=(c/a;q)(c;q). (115)

This proof is complete.

Theorem 18 (see [17, page 354, II. 1, II. 2]) —

The two q-exponential functions are

eq(z)=n=0zn(q;q)n=1(z;q)for|z|<1,Eq(z)=n=0q(n2)zn(q;q)n=(z;q). (116)

Proof —

Setting a = b = d = 0 and then replacing c by a in (14), we obtain

E[1(aX;q)]=1(a;q)ϕ01(0;q,ax). (117)

Letting d = a = c and b = 0 in (14) or letting d = c and a = 0 and replacing b by a in (14), we obtain

E[1(aX;q)]=1(ax,a;q). (118)

Comparing (117) and (118) gives

1(a;q)ϕ01(0;q,ax)=1(ax,a;q). (119)

From the above formula, we obtain

ϕ01(0;q,ax)=1(ax;q). (120)

Replacing ax by z gives

ϕ01(0;q,z)=1(z;q); (121)

that is,

eq(z)=n=0zn(q;q)n=1(z;q)for|z|<1. (122)

Similarly, setting d = a and b = 0 in (19) of Theorem 4, we obtain

E[1]=1(ax;q)ϕ00(;q,ax). (123)

And obviously letting d = c and a = b = 0 in (14) of Theorem 3, we obtain

E[1]=1. (124)

Comparing (123) and (124), we obtain

1(ax;q)ϕ00(;q,ax)=1. (125)

Replacing ax by −z, we get

ϕ00(;q,z)=(z;q); (126)

that is,

Eq(z)=n=0q(n2)zn(q;q)n=(z;q). (127)

Remark 19 —

In the present paper we obtain the part transformation and sum formulas of the q-series by applying the probabilistic method. We hope to find and construct another probability distribution in order to prove the transformation and sum formulas of the bilateral basic hypergeometric series, for example, Ramanujan and Bailey sum formulas and so forth.

Acknowledgments

The authors express their sincere gratitude to the referee for many valuable comments and suggestions. The present investigation was supported by the Natural Science Foundation Project of Chongqing, China, under Grant CSTC2011JJA00024, Research Project of Science and Technology of Chongqing Education Commission, China, under Grant KJ120625, and Fund of Chongqing Normal University, China, under Grants 10XLR017 and 2011XLZ07.

Conflict of Interests

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

References

  • 1.Chang G, Xu C. Generalization and probabilistic proof of a combinatorial identity. The American Mathematical Monthly. 2011;118(2):175–177. [Google Scholar]
  • 2.Chapman R. A probabilistic proof of the Andrews-Gordon identities. Discrete Mathematics. 2005;290(1):79–84. [Google Scholar]
  • 3.Fulman J. A probabilistic proof of the Rogers-Ramanujan identities. Bulletin of the London Mathematical Society. 2001;33(4):397–407. [Google Scholar]
  • 4.Kadell KWJ. A probabilistic proof of Ramanujan's 1 ψ 1 sum. SIAM Journal on Mathematical Analysis. 1987;18(6):1539–1548. [Google Scholar]
  • 5.Lee PA. Probability distribution and a self-reciprocal relation involving confluent hypergeometric functions. Journal of the Franklin Institute. 1997;334(1):19–22. [Google Scholar]
  • 6.Ong SH. Probabilistic proof of a Hankel transform of Laguerre polynomials. Journal of the Franklin Institute. 1999;336(6):1007–1011. [Google Scholar]
  • 7.Pascu MN. A probabilistic proof of the fundamental theorem of algebra. Proceedings of the American Mathematical Society. 2005;133(6):1707–1711. [Google Scholar]
  • 8.Rosalsky A. A simple and probabilistic proof of the binomial theorem. The American Statistician. 2007;61(2):161–162. [Google Scholar]
  • 9.Ross SM. A First Course in Probability. 8th edition. Upper Saddle River, NJ, USA: Prentice Hall; 2010. [Google Scholar]
  • 10.Srivastava HM, Vignat C. Probabilistic proofs of some relationships between the Bernoulli and Euler polynomials. European Journal of Pure and Applied Mathematics. 2012;5(2):97–107. [Google Scholar]
  • 11.Miller SJ. A probabilistic proof of Wallis's formula for π . The American Mathematical Monthly. 2008;115(8):740–745. [Google Scholar]
  • 12.Sun P. Moment representation of Bernoulli polynomial, Euler polynomial and Gegenbauer polynomials. Statistics & Probability Letters. 2007;77(7):748–751. [Google Scholar]
  • 13.Wang M. An expectation formula with applications. Journal of Mathematical Analysis and Applications. 2011;379(1):461–468. [Google Scholar]
  • 14.Wang M. A new probability distribution with applications. Pacific Journal of Mathematics. 2010;247(1):241–255. [Google Scholar]
  • 15.Xia A. A probabilistic proof of Stein's factors. Journal of Applied Probability. 1999;36(1):287–290. [Google Scholar]
  • 16.Andrews GE, Askey R, Roy R. Special Functions. Cambridge, UK: Cambridge University Press; 1999. [Google Scholar]
  • 17.Gasper G, Rahman M. Basic Hypergeometric Series. Cambridge, UK: Cambridge University Press; 2004. [Google Scholar]
  • 18.Jackson FH. On q-definite integrals. The Quarterly Journal of Pure and Applied Mathematics. 1910;50:101–112. [Google Scholar]
  • 19.Andrews GE, Askey R. Another q-extension of the beta function. Proceedings of the American Mathematical Society. 1981;81(1):97–100. [Google Scholar]
  • 20.Liu DF, Luo QM. An extension for Andrews-Askey integral formula. In press. [Google Scholar]
  • 21.Rahman M, Verma A. A q-integral representation of Rogers' q-ultraspherical polynomials and some applications. Constructive Approximation. 1986;2(1):1–10. [Google Scholar]
  • 22.Heine E. Untersuchungen über die Reihe. Journal für die reine und angewandte Mathematik. 1847;1847(34):285–328. [Google Scholar]
  • 23.Cauchy A-L. Mémoire sur les fonctions dont plusieurs valeurs sont liées entre elles par une équation linéaire, et sur diverses transformations de produits composés d’un nombre indéfini de facteurs. Comptes Rendus de l'Académie des Sciences. 1843;17:p. 523. in Oeuvres de Cauchy, vol. 8, 1st série, pp. 42–50, Gauthier-Villars, Paris, France, 1893. [Google Scholar]
  • 24.Jacobi CGJ. Über einiger der Binomialreihe analoge Reihen. Journal für die reine und angewandte Mathematik. 1846;32:197–204. reprinted in Gesammelte Werke, vol. 1, pp. 49–239, Reimer, Berlin, Germany, 1881. [Google Scholar]
  • 25.Askey R. Ramanujan's extensions of the gamma and beta functions. The American Mathematical Monthly. 1980;87(5):346–359. [Google Scholar]
  • 26.Joichi JT, Stanton D. Bijective proofs of basic hypergeometric series identities. Pacific Journal of Mathematics. 1987;127(1):103–120. [Google Scholar]
  • 27.Rahman M, Suslov SK. A unified approach to the summation and integration formulas for q-hypergeometric functions I. Journal of Statistical Planning and Inference. 1996;54(1):101–118. [Google Scholar]

Articles from The Scientific World Journal are provided here courtesy of Wiley

RESOURCES