Some New Formulas Involving G
qFunctions.
THOMASERNST(*)
ABSTRACT- In a recent paper we found some new results forq-functions of many variables with the aid of theGqfunction. The Heine notation reminding of the hypergeometric case was used throughout, and some relations between Gq
functions were presented. This paper aims at giving the promised longer ex- position ofGq- revealing also the connection between this and the Jacobi-theta functions which appearin context. We will give a slightly generalized definition of the Heine series with more general tilde operators. 4q-summation formulas of Andrews will be given in the new notation. The close affinity to q-binomial coefficient formulas will be stressed by expressing the finiteq-hypergeometric formulas, the canonical form, in two ways. Two furtherq-analogues of Kummer's
2F1( 1) formula will be given. An ancientq-analogue of the Eulerreflection formula will be used for the proof of a special case of the Bailey-Daum sum- mation formula, conjectured in the previous paper.
Multiple extensions of Gauss' formula will be given by a similar technique. All this will explain the utility of the Heine notation.
1. Introduction.
In this section we present the necessary definitions together with useful related formulas. The notation from [13] will be used whenever possible. The umbral calculus from [15] will play a significant role in the definition of the q-hypergeometric series. The Jackson Gq function fits nicely togetherwith theq-shifted facorial. In Section two, 2q-summation formulas of Andrews are given in terms ofGqfunctions, and 2q-summation formulas of Andrews will be proved by Watson's [46] transformation for- mula fora terminating very-well-poised 8f7 series. In Section three two furtherq-analogues of Kummer's 2F1( 1) formula will be given. We will prove a special case of the Bailey-Daum summation formula, conjectured in
(*) Indirizzo dell'A.: Department of Mathematics, Uppsala University, P.O. Box 480, SE-751 06 Uppsala, Sweden; E-mail: [email protected]
2000 Mathematics Subject Classification: Primary 33D05; Secondary 33D70.
the previous paper [13], by using Jacobi-theta functions. In Section four multiple extensions of Gauss' formula will be given by a similar technique.
1.1 ±Tilde operators.
For preliminary definitions the reader is referred to the paper [13]. In that paperwe defined aq-shifted factorial and a tilde operator.
DEFINITION1. The operator e:C
Z7!C Z is defined by
a7!a pi logq: 1
By (1) it follows that
ha;gqinnY1
m0
(1qam);
2
It turns out that in certain formulas it is not easy to cope with the casea0:
That's why we define
he0;qin he1;qin 1: 3
This means that ifa0 we skip the first factor 11, which is not really aq- shifted facorial, and just compute the lastn 1 factors. The reason is the affinity with the factor1 1, which causes trouble in denominators.
Because
h1 gn;qin h0;eqinq n2 ; 4
we also define
h1 gn;qin h1 gn;qin 1: 5
This formula was already used by Watson [47, p. 64].
To be able to treat a general root of unity, it is desirable to generalize this operator in the following way. In equations (6) to (16) we assume that (m;l)1.
DEFINITION2. The operator e
ml :C Z7!C
Z is defined by
a7!a2pim llogq: 6
This means
hmfla;qinYn 1
m0
(1 e2pimlqam);
7
Furthermore we define
g
mlha;qin hfmla;qin: 8
We will also need another generalization of the tilde operator.
hkea;qinYn 1
m0
Xk 1
i0
qi(am)
: 9
h2ea;qin hea;qin: 10
h1ea;qin1:
11
kha;gqin hkea;qin: 12
The following simple rules follow from (6). Some of them were pre- viously known in othernotation from [17].
THEOREM1.1.
f
mlabml(agb)mod 2pi logq; 13
Xl
k1
g
1lakXl
k1
ak mod 2pi logq; 14
m
l eamg
2lam
l mod 2pi logq; 15
QE(mela)QE(a)e2piml ; 16
where the second equation is a consequence of the fact that we work mod 2pi
logq. Furthermore, THEOREM1.2.
hea;q2in h14ea;34ea;qin: 17
ha;qpin Yp 1
k0
hkpea;qin; where p is an odd prime:
18
ha;qkin ha;qinkha;gqin: 19
This leads to the following q-analogue of [36, p. 22, (2)].
THEOREM1.3.
ha;qiknYk 1
m0
Dam k ;qE
nkD gam k ;qE
n: 20
1.2 ±Heine's series
The q-hypergeometric series was developed by Heine 1846 [24] as a generalization of the hypergeometric series.
DEFINITION 3. Generalizing Heine's series, we shall define a q-hy- pergeometric series by (compare [17, p. 4], [22, p. 345]):
pp0frr0(a^1;. . .;a^p; ^b1;. . .;b^rjq;zks1;. . .;sp0;t1;. . .;tr0)
pp0frr0
a^1;. . .;a^p
b^1;. . .;b^rjq;zks1;. . .;sp0
t1;. . .;tr0
" #
X1
n0
h^a1;qin. . .h^ap;qin
h1;qinh^b1;qin. . .h^br;qinh( 1)nq n2 i1rr0 p p0
znYp0
k1
(sk;q)nYr0
k1
(tk;q)n1; 21
whereq60 whenpp0>rr01;and
ba a;
ea;
f
mla;
kea 8>
>>
>>
<
>>
>>
>: 22
In a few cases the parameter^ain (21) will be the real plus infinity (0<jqj<1). They correspond to multiplication by 1. If we want to be formal, we could introduce a symbol1H;with property
h1H;qin h1Ha;qin1; a2C; 0<jqj<1:
23
The symbol1Hcorresponds to parameter 0 in [17, p. 4]. We will denote1H
by1in the rest of the paper.
The terms to the left ofjin (21) are thought to be exponents, they are periodic mod 2pi
logq.
The first term to the right ofjin (21) is the base, and the second is a letter in the spirit of the umbral calculus in [15]. The terms to the right ofk in (21) are Watson q-shifted factorials [17]. There will be a certain re- dundance in notation here, but this is no problem.
1.3 ±TheGqfunction
In the same way as theGfunction plays a basic role in complex analysis, theGqfunction is fundamental forq-calculus. During the last years there has been an increasing interest in the Gq function. Howeverthere are certain restrictions in our knowledge of this function as the following ex- position shows. TheGqfunction is defined in the unit disk 0<jqj<1 by
DEFINITION4.
Gq(x)h1;qi1
hx;qi1(1 q)1 x: 24
Here we deviate from the usual conventionq<1, because we want to work with meromorphic functions of several variables. The simple poles are located atx n 2kpi
logq; n;k2N.
Heine, Ashton and Daum [11] used anotherfunction without the factor (1 q)1 x, which they called the Heine V-function. The main difference between the two functions is thatVhas zeros, in contrast to theGqfunction which has no zeros, and therefore 1
Gqis entire. Ashton [5] in his thesis su- pervised by Lindemann, showed its connection to elliptic functions. Daum [11] tried to find all the basic analogues of Thomae's3F2 transformation formula [42, eq. 11], using a notation analogous to that used by Whipple [48], and by essentially replacing theGqfunction by the HeineV-function.
Daum's work was continued when in 1987 Beyer - Louck - Stein [9] and in 1992 Srinivasa Rao - Van der Jeugt - Raynal - Jagannathan - Rajeswari [38]
showed that certain two-term transformation formulas between hypergeo- metric series easily can be described by means of invariance groups. In other words, they explained Whipple's [48] discovery in group language. In 1999 Van derJeugt - Srinivasa Rao [43] foundq-analogues of these results.
Daum [11] concludes his thesis by sayingIt is hoped, however, that the use of the modified HeineV-function, will serve to emphasize the analogy between hypergeometric series and q-hypergeometric series and simplify the notation generally.
Let's now return to theGq function: To save space, the following no- tation forquotients ofGq functions will often be used.
DEFINITION5. The generalizedGqfunction is given by Gq a1;. . .;ap
b1;. . .;br
Gq(a1). . .Gq(ap) Gq(b1). . .Gq(br): 25
DEFINITION6. This generalizedGqfunction is called balanced if Xp
k1
ak Xr
k1
bk; pr:
26
DEFINITION7. A quotient of infiniteq-shifted factorials Qp
k1hak;qi1 Qr
k1hbk;qi1 27
is called balanced if
Xp
k1
ak Xr
k1
bk; pr:
28
These definitions are easily seen to be equivalent.
Balanced quotients ofGq's occurin many equations ofq-calculus. To see why this is the case, we need to go back 100 years to the work of Mellin [31].
Mellin discovered that every hypergeometric function can be written as a function of generalizedG functions. In the same way, every q-hypergeo- metric function can be written as a function of generalizedGqfunctions [23, p. 258]. In the presence of a balanced quotient of Gq's, limits when the parameters tend to 1usually pose no problem. It would therefore be agreeable if we could find bounds for quotients of balancedGq's. The first step in this direction was made by Ismail and Muldoon [25, p. 320]. This was generalized by Alzer [1].
The following lemmas are just a few examples ofGq formulas.
LEMMA1.4. A relation betweenGq functions with different bases.
Gq2(x)
Gq(x) he1;qi1
hex;qi1(1q)1 x(1q)1 xhe1;qix 1: 29
LEMMA1.5.
D1a
2 ;1 ba 2;q2E h1a b;q2i1 1
Gq 1a b;1a 2 1a;1a
2 b 2
64
3 75
D g1a 2 b;qE g 1
D1a
2;1ga b;qi1 : 30
PROOF. Use the Bailey-Daum theorem. p
We are going to find q-analogues of hypergeometric summation for- mulas in this paper. One example is the Whipple formula [48] for a ter- minating series, which was first given by Watson.
31 3F2 a; n;c
2;1a n 2 ;c;1
G
1a n 2 ;1c
2 ;1c an
2 ;1
2 1a
2 ;1 n
2 ;1 ac
2 ;1cn 2 2
66 4
3 77 5
2. Fourq-summation formulas of Andrews.
In recent years, combinatorics has developed quickly. Combinatorial identities can be expressed either as hypergeometric formulas - the ca-
nonical form-, or as binomial coefficient identities. Combinatorial results are often expressed asq-formulas. Here we have a similar duality:q-hy- pergeometric formulas - the canonical form-, or q-binomial coefficient identities. Nowq-binomial coefficient computations are easier to handle by hand, orby computer. Some mathematicians, like forexample Catalan, Bateman and Gould, have tried (in vain) to make systematic treatments of binomial coefficient identities. Nonetheless, q-hypergeometric formulas and q-binomial coefficient identities belong together. Therefore we re- present many theorems of this chapter both ways.
As a basis forthe following calculations, we are going to use 4 q-for- mulas by Andrews. The first is
THEOREM2.1. Andrews's q-analogue of Kummer's formula[28, (2), p. 134]from[2, 1.8 p. 526].
32
2f2
a;b 1ab
2 ;1gab 2
jq; q 2
4
3 5Gq
1ab 2 ;1
2 1b
2 ;1a 2 2
66 4
3 77 5
g1b 2 ;q
a2
e1 2;q
a2
Gq2
1ab 2 ;1
2 1b
2 ;1a 2 2
66 4
3 77 5:
PROOF. By [2, 1.8 p. 526]
LHSby (30) Gq
1ab 2 ;1a
2 1a;1b
2 2
66 4
3 77 5
g1b 2 ;e1;q
g1ab 1
2 ; g
1a 2;q
1 by (29)
Gq
1ab 2 1b
2 ;1a 2
66 4
3 77
5Gq2 1a 2 2 4
3 5(1q)a2
g1b 2 ;q
g1ab 1
2 ;q
1
33
Gq
1ab 2 1b
2 2 66 4
3 77 5Gq2
1 2 1a
2 2 66 4
3 77
5(1q) a2
g1b 2 ;q
g1ab 1
2 ;q
1
by (29) RHS;
where we have used theq-analogue of the Legendre duplication formula in
the penultimate step. p
REMARK 1. The name Gauss second summation theorem for Kummer's formula [28, (2), p. 134] was coined by Slater. This formula has recently received a lot of attention. We will generalize the q- analogue later.
Ifbis a negative integer, (32) may be reformulated in the form
THEOREM2.2.
2f2
a; 2N 1a 2N
2 ;1ga 2N 2
jq; q 2
4
3 5
X2N
k0
2N k
q q2 k2k(1 2N) ( 1)kha;qik D1a 2N
2 ;q2E
k
D1 2;q2E
Nq Na D1 a
2 ;q2E
N
D1 2N 2 ;q2E D1a 2N N
2 ;q2E
N
: 34
2f2
a; N 1a N
2 ;1ga N 2
jq; q 2
4
3 5
XN
k0
N
k q q2 k2 k(1 N) ( 1)kha;qik D1a N
2 ;q2E
k
0;N odd:
35
The following corollary was influenced by [27]. The proof is the same.
This idea to use the contiguity relations goes back to Kummer [28, p.
134-36].
COROLLARY2.3.
2f2
a;b 2ab
2 ;2gab 2
jq; q 2
4
3
5qa(1 qb) qa qb Gq
2ab 2 ;1 2b 2
2 ;1a 2 2
66 4
3 77 5
g2b 2 ;q
a2
e1 2;q
a2
qb(1 qa) qa qb Gq
2ab 2 ;1 1b 2
2 ;2a 2 2
66 4
3 77 5
g1b 2 ;q
a12
e1 2;q
a12
: 36
2f2
a;b 3ab
2 ;3gab 2
jq; q 2
4
3 5
qa(1 qb) qa qb
qa(1 qb1) qa qb1 Gq
3ab 2 ;1
2 3b
2 ;1a 2 2
66 4
3 77 5
g3b 2 ;q
a2
e1 2;q
a2
2 66 66 4
qb1(1 qa) qa qb1 Gq
3ab 2 ;1
2 2b
2 ;2a 2 2
66 4
3 77 5
g2b 2 ;q
a12
e1 2;q
a12
3 77 77 5
qb(1 qa) qa qb
qa1(1 qb) qa1 qb Gq
3ab 2 ;1
2 2b
2 ;2a 2 2
66 4
3 77 5
g2b 2 ;q
a12
e1 2;q
a12
2 66 66 4
qb(1 qa1) qa1 qb Gq
3ab 2 ;1 1b 2
2 ;3a 2 2
66 4
3 77 5
g1b 2 ;q
a22
e1 2;q
a22
3 77 77 5: 37
THEOREM2.4. Andrews's q-analogue of Kummer's formula[28, p. 134]
from[2, 1.8 p. 526]and[39, (15), p. 9].
2f2 a;1 a c;e1 jq; qc
" #
Gq
c 2;1c
2 1c a
2 ;ac 2 2
66 4
3 77 5
g1c a 2 ;q
a2
ec 2;q
a2
Gq2 c 2;1c 1c a2
2 ;ac 2 2
66 4
3 77 5: 38
PROOF. By [2, 1.9 p. 526]
LHS
D1c a 2 ;ac
2 ;q2
hc;qi1 1 D1c a
2 ;1gc a 2 ;ac
2 ;agc 2 ;q
1
1c 2 ;1gc
2 ;c 2;ec
2;q
1
RHS:
39
p Ifais a negative integer, (38) may be reformulated in the form THEOREM2.5.
2f2 2N;12N c;e1 jq; qc
" # X2N
k0
2N k
q ( 1)kq2 k2 k(c 2N)h12N;qik hc;e1;qik
Dc 2N 2 ;q2E D1c N
2 ;q2E
N
: 40
2f2 N;1N c;e1 jq; qc
" #
Dc N
2 ;q2E
1N2
Dc 2;q2E
1N2
;N odd:
41
We have expressed two of Andrew's formulas by theGq function, and also expressed the corresponding finite sums by quotients of q-shifted factorials. We will continue with generalizations of these two formulas.
We need a lemma forthe following proof.
LEMMA2.6. A q-analogue of the Legendre duplication formula for the G-function written in q-shifted factorial form is
h1 n;q2in2 h1 gn;qin 1D1
2;q2E
n2
( 1)n2 QE n2 4
;neven:
42
THEOREM2.7. Compare[18, (II 17), p. 355].A q-analogue of[45], [8, (1.2), p. 237].
4f3
c 2;ec
2;a; 2N 2N1a
2 ; 2Ng1a
2 ;c
jq;q 2
66 64
3 77 75
D1
2;1c a 2 ;q2E D1 a N
2 ;1c 2 ;q2E
N
: 43
4f3
c 2;ec
2;a; N N1a
2 ; Ng1a
2 ;c
jq;q 2
66 4
3 77
50; N odd:
44
PROOF. The proof of (44) is relegated to the next proof, i.e. (54). We will use the method in Andrews [3, p. 334]. For convenience, we will denote the
LHS of (43)4f3
c 2;ec
2;a; n n1a
2 ; ng1a 2 ;c
jq;q 2
66 4
3 77 5:
By Watson's [46] transformation formula for a terminating very-well- poised8f7 series, denoting
(a)(a;b;c;d;11 2a; g
11
2a;e; n);
45
(b)(1a b;1a c;1a d;1a e;1 2a;1f
2a;1an);
46
this formula takes the following form
8f7 (a)
(b)jq;q2a2n b c d e
h1a;1a d e;qin h1a d;1a e;qin
4f3 d;e;1a b c; n
1a b;1a c;de n ajq;q
: 47
Now make the substitution a!gn;b! n1 a
2 ;c! ng1 a 2 ;d!c
2;e!ec 2; 48
to obtain
8f7 (a0)
(b0)jq;q1a c
g
n1 c
2; n1 c 2;q
gn1; n1 c;q n
n
4f3
c 2;ec
2;a; n n1a
2 ; ng1a 2 ;c
jq;q 2
66 4
3 77 5: 49
where
(a0) gn; n1 a
2 ; ng1 a 2 ;c
2;ec 2;1 g
41 1
2n;3 g
41 1
2n; n
!
; 50
(b0)( n1a
2 ; ng1a
2 ;1 n c
2; g
1 n c
2;1g
4 n
2 ;3g
4 n
2 ;e1);
51
Now
hea;q2in e14a;34ea;q
n: 52
By theq-Dixon theorem we have, assumingneven
4f3
c 2;ec
2;a; n n1a
2 ; ng1a 2 ;c
jq;q 2
66 4
3 77 5
4f3
n1 a
2 ; g
1 n 2;c
2; n 1 na
2 ;1 n c
2; gn 2
jq2;q1a c 2
66 4
3 77 5
g n1 c
2; n1 c 2;q
gn1; n1 c;q n
n
g n1 c
2; n1 c 2;q
gn1; n1 c;q n
n
1 n;1 na c 2 ;q2
n2
n1 c
2;1 na 2 ;q2
n2 by (42)
D1
2;1c a 2 ;q2E
n2
D1 a 2 ;1c
2 ;q2E
n2
: 53
p
Equation (43) may be reformulated in the form THEOREM2.8. Compare[3, p. 334].
4f3
c 2;ec
2;a; n n1a
2 ; ng1a 2 ;c
jq;q 2
66 4
3 77 5
e1;1 c;q g1 n; n1 1c;q
1
D 2n2
2 ;1a
2 ; n2 c
2 ;a n1 c 2 ;q2
D2 c 1
2 ;1a n
2 ; n2
2 ;a1 c 2 ;q2
1
: 54
PROOF. LHS4f3
n1 a
2 ; g
1 n 2;c
2; n
1 na
2 ;1 n c
2; gn 2
jq2;q1a c 2
66 4
3 77 5
g n1 c
2; n1 c 2;q
gn1; n1 c;q n
n
e1;1 c;q
1
Da1
2 ; n1; n2 c
2 ; na1 c 2 ;q2
1
gn1; n1 c;q
1
Da n1 2 ; n2
2 ;2 c
2 ;a1 c 2 ;q2
1
e1;q
1
Da1 2 ; n1
2 ;1 c
2 ; na1 c 2 ;q2
Da n1 1
2 ; n1 c
2 ;a1 c 2 ;q2
1
e1;1 c;q g1 n; n1 1c;q
1
D 2n2 2 ;1a
2 ; n2 c
2 ;a n1 c 2 ;q2
D2 c 1
2 ;1a n 2 ; n2
2 ;a1 c 2 ;q2
1
: 55
p REMARK2. The equivalent expression
Da1
2 ; n1 2 ;1 c
2 ; na1 c
2 ;q2
D1 1
2;a n1
2 ; n1 c
2 ;a1 c 2 ;q2
1
56
forthe RHS of (54) was given in [21, 1.4.]. The simple proof uses the Euler formula [16, p. 271]
e1;q
1
1 2;q2
11:
57
This was the beginning of the investigation that led to the current paper.
Now letc! 1in (54) and (43) to arrive at (32) and (34).
COROLLARY2.9.
3f2
a; 2N;1 1a 2N
2 ;1ga 2N 2
jq;q 2
4
3 5
D1 2;q2E D1 a N
2 ;q2E
N
: 58
PROOF. Letc! 1in (43). p
COROLLARY2.10.
3f2 c 2;ec
2; 2N
c;1 jq; q
" #
X2N
k0
2N k
qq k2 k(1 2N)( 1)kDc 2;q2E hc;qik k
D1
2;q2E
N qNc D1c
2 ;q2E
N
: 59
3f2 c 2;ec
2; N c;1 jq;q
" #
XN
k0
N
k q q k2k(1 N)( 1)kDc 2;q2E
hc;qik k0;N odd:
60
PROOF. Leta! 1in (43) and (44). p
COROLLARY2.11. Compare[18, ex. 3.4, p. 101].
3f1 c 2;ec
2; 2N
c
jq; q2N 2
4
3 5X2N
k0
2N k
q
( 1)kDc 2;q2E hc;qik k
D1
2;q2E D1c N
2 ;q2E
N
: 61
3f1 c 2;ec
2; N c
jq; qN 2
4
3 5XN
k0
N k q
( 1)kDc 2;q2E
hc;qik k0;Nodd:
62
PROOF. Leta! 1in (43) and (44). p
THEOREM2.12. [3, p. 333]
4f3 c 2;ec
2;1n; n c1 e;e1;e
jq;q 2
64
3 75
ee;e c;q 1
g
e n; ne c;q
1
De1n
2 ; ne1 c
2 ; ne;q2 D ne1 1
2 ;n1e c 2 ;e;q2
1
: 63
PROOF. We will again use Watson's 1929 [46] transformation formula fora terminating very-well-poised8f7series. Now make the substitution
a! nge 1;b! ne 1;c!gn;d!c 2;e!ec
2: 64
to obtain
8f7 (a0)
(b0)jq;qe1 a c
g
ne c
2; ne c 2;q
nge; ne c;q n
n
4f3 c 2;ec
2;1n; n e;e1;c1 e
jq;q 2
64
3 75:
65
where
(a0) fn; ne 1; nge 1;c 2;ec
2;1 g
41e n
2 ;3 g
41e n
2 ; n
!
; 66
(b0)(e;ee;e n c
2; g
e n c
2;1 g
4 ne 1
2 ;3 g
4 ne 1
2 ;e1):
67
Now
hea;q2in e14a;34ea;q
n: 68
And we have
4f3 c 2;ec
2;1n; n e;e1;c1 e
jq;q 2
4
3 5
4f3 n 1e;1 gne 2 ;c
2; n e;e n c
2; nge 1 2
jq2;qe1n c 2
66 64
3 77 75
g ne c
2; ne c 2;q
n
gne; ne c;qin
hee;e c;qi1Dne1
2 ; ne; ne1 c 2 ;q2
gne; ne c;q 1
1
e; ne1
2 ;ne1 c 2 ;q2
1
hee;e c;qi1 g
e n; ne c;q
1
2n2e c
2 ;e1n
2 ; ne1 c 2 ;2e c
2 ; ne;q2 1
2e c
2 ; 2n2e c
2 ; ne1
2 ;n1e c 2 ;e;q2
1
RHS:
69
p Equation (63) may be reformulated in the form
THEOREM2.13. A q-analogue of the Whipple formula[48, p. 114], [8, (1.3), p. 237].
4f3 c 2;ec
2;12N; 2N c1 e;e1;e
jq;q 2
4
3 5
X2N
k0
2N k
q q k2 k(1 2N)( 1)kh12N;qikDc 2;q2E
k
hc1 e;e1;e;qik
( 1)N QE(N2NNc Ne)De 2N
2 ;e1 c 2 ;q2E De1 N
2 ;2c e 2 ;q2E
N
: 70
4f3 c 2;ec
2;22N; 2N 1 c1 e;e1;e
jq;q 2
4
3 5
2N1X
k0
2N1 k
q q k2 2kN( 1)kh22N;qikDc 2;q2E
k
hc1 e;e1;e;qik
( 1)N1 QE((N1)2(c e)(N1))De 2N 1 2 ;e c
2 ;q2E De N1
2;1c e 2 ;q2E
N1
: 71
Now letc! 1in (70) and (71) to arrive at (40) and (41).
COROLLARY2.14. Anotherq-analogue of Kummers formula.
3f2 2N;12N;1 c;e1 jq;q
X2N
k0
2N k
q ( 1)kq k2k 2Nkh12N;qik hc;e1;qik
Dc 2N 2 ;q2E D1c N
2 ;q2E
N
q2N2N: 72
3f2 N;1N;1 c;e1 jq;q
XN
k0
N
k q ( 1)kq k2k Nkh1N;qik hc;e1;qik
Dc N
2 ;q2E
1N2
Dc 2;q2E
1N2
q N12 ;N odd:
73
PROOF. Letc! 1in (70) and (71). p
3. Kummer's 2F1( 1)formula and Jacobi's theta function.
In [13] we used aq-analogue of the Dixon-Schafheitlin theorem
4f3 a;b;c; g 11 2a 1a b;1a c;1f
2a
jq;q1a2 b c 2
66 64
3 77 75
Gq
1a b;1a c;1a 2;1a
2 b c
1a;1a b c;1a
2 b;1a 2 c 2
66 4
3 77 5 74
to prove a q-analogue of Kummer's 2F1( 1) formula [28], the Bailey- Daum summation formula,
75 2f1(a;b;1a bjq; q1 b)Gq
1a b;1a 2 1a;1a
2 b 2
4
3 5
1ga 2 b;e1;q
g 1
1a
2;1gb;q
1
;
where 1a b60; 1; 2. . ..
We can easily find two related formulas.
THEOREM3.1. A second q-analogue of Kummer's 2F1( 1)formula.
3f3
a;b; g 1a
2;1a b;ea
2;1jq;q1a2 b
Gq
1a b;1a 2 1a;1a
2 b 2
4
3 5; 76
where1a b60; 1; 2. . ..
PROOF. Letc! 1in (74). This proof is aq-analogue of [7, p. 13]. p THEOREM3.2. A third q-analogue of Kummer's 2F1( 1)formula.
4f2
a;b; g 1a
2;1;1a b;ea
2jq;q a2 b
Gq 1a b;1a 2 1a;1a
2 b 2
4
3 5qab2; 77
where1a b60; 1; 2. . ..
PROOF. Letc! 1in (74). This proof is aq-analogue of [4, p. 144].
p
In [13] we found the following special case of (75)
2f1( 2N;b;1 2N bjq; q1 b) X2N
k0
2N k
q q k2k(1 2N b) hb;qik h1 2N b;qik 2XN 1
k0
2N k
q( 1)k hb;qik
h2Nb k;qik 2N N
q( 1)N hb;qiN hNb;qiN heb;qiN
hNb;qiN D1
2;q2E
N: 78
2f1( N;b;1 N bjq; q1 b)0;Nodd:
79
The aim of the rest of this section is to give a proof of (78).
The proof forq1 [35, p. 43] uses the Euler reflection formula G(x)G(1 x) p
sinpx: 80
The following crucial formula has an interesting history. It was known in the literature twice before it was given in English in 2001. It first ap- peared 1873 in a book by Thomae about among other things theta func- tions. Then it appeared in an Italian paper 1923 by Pia Nalli. The inter- esting thing is that both Thomae and Pia Nalli used notations fortheta functions which are clearly different from the modern notation. The Tho- mae notation is reminiscent of Riemann theta functions and the Pia Nalli notation was probably influenced by nineteenth-century Italian books about elliptic functions. It is also rumoured that Andrews and Askey have known this formula for many years.
THEOREM3.3. The q-analogue of(80)is[41, p. 183 (168a)], [6, p. 1326], [32, p. 338]
Gq(x)Gq(1 x) i q1=8(1 q)(h1;qi1)3 qx=2u1
ix
2 logq;q1=2 ; 81
where the first Jacobi theta function is given by u1(z;q)2X1
n0
( 1)nQE n1
2 2
sin (2n1)z:
82
This function has period2pand quasiperiod i 2 logq.
For the proof of (78) we need two further lemmata.
LEMMA3.4. Another q-analogue of the Legendre duplication formula written in q-shifted factorial form is
hN;qiN h1 N;gqiN 1D1
2;q2E
N
QE N 2
: 83
LEMMA3.5.
u1 iNlogq;q1=2 u1
iN
2 logq;q1=2
( 1)Nq (N=2):
84
PROOF. Use the quasiperiodicity ofu1. p
PROOF. If we use the same method of proof as in [35, p. 43], but use (81) instead of (80), we arrive at
2f1( 2N;b;1 2N bjq; q1 b)
1 gN b;e1;q g1 N;1gb;q1 1
h1 N b;N;qi1
h1 2N b;2N;qi1QEN 2
u1 iNlogq;q1=2 u1
iN
2 logq;q1=2
1 gN b;N;qiN g1 N;1 2N b;q
N
QE N
2
u1 iNlogq;q1=2 u1
iN
2 logq;q1=2
eb;N;q g1 N;bN;Nq
N
QE
N2N 2
( 1)N u1 iNlogq;q1=2 u1
iN
2 logq;q1=2 : 85
On the otherhand, by (5), (83) and (84) we have
86 hN;qiN g1 N;q
N
QE
N2N 2
( 1)N u1 iNlogq;q1=2 u1
iN
2 logq;q1=2 D1
2;q2E
N:
The last two formulas together prove (78). p
4. Multiple extensions of Gauss' formula.
In this section we will look at several multipleq-formulas, so we start with some notation.
DEFINITION8. The notationP
~
m denotes a multiple summation with the indices m1;. . .;mn running over all non-negative integer values. In this connection we putjmj Pn
j1mj.
Ifm~and~kare two arbitrary vectors withnelements, theirq-binomial coefficient is defined as
m~
~k qYn
j1
mj kj q: 87
Iffxjgnj1andfyjgnj1are two arbitrary sequences of complex numbers, then theirscalarproduct is defined by
~ xyXn
j1
xjyj: 88
In the same way we define the following vectorversions of powers, etc.
~x~aYn
j1
xajj; 89
dm;~~k 1; if m~~k;
0; otherwise:
( 90
q ~k2 Yn
j1
q kj2
; 91
(~u;q)~s Yn
j1
(uj;q)sj; ( 1)~k ( 1)jkj: 92
The following theorem, forn1 forms the basis ofq-analysis.
According to Jensen [26, p. 30], Ward [44, p. 255] and Kupershmidt [29, p. 244], this theorem was obtained by Euler. It was also obtained by Gauss and published posthumously in 1876 [19]. It is proved by induction [12].
THEOREM4.1.
X~k
m~ ~0
( 1)~m ~k m~ !
q
q m~2 ~um~ (~u;q)~k
93
COROLLARY4.2.
X~k
~ m~0
( 1)m~ ~k
~ m !
q
q m~2 d~k;~0
94
Ouraim is now to find some furthermultidimensional variations of (94) and a couple of related formulas. For the proof we need some other for- mulas, starting with
THEOREM4.3. The Jackson q-derivative expressed as q-difference of a q-shifted factorial.
Dnq;qahga;qik ( 1)nfk n1gn;qhgan;qik nq n2 ng: 95
We will only need oneq-Lauricella function, which is defined by DEFINITION9.
F(n)D (a;b1;. . .;bn;cjq;x1;. . .;xn)
X
~ m
ha;qim1...mnhb1;qim1. . .hbn;qimnQn
j1xmj j hc;qim1...mnQn
j1h1;qimj : 96
The following lemma is aq-analogue of Chaundy [10, p. 164].
LEMMA4.4.
XR
r0
XS
s0
( 1)rsh1;qiRh1;qiS h1;qirh1;qiR rh1;qish1;qiS s hc 1;qirshc;qi2r2s
hc 1;qi2r2shc;qiRSrs QE r
2 s 2 Sr
1; if RS0;
0; otherwise:
97
PROOF. LHSXR
r0
XS
s0
h R;qirh S;qis h1;qirh1;qisDc 1
2 ;c 2;q2E
rs
hc 1;qirsDc1 2 ;c
2;q2E
hc;qiRShcRS;qirsrsQE(RrS(rs))
1 hc;qiRS
X
RS k0
hc 1;qikDc1 2 ;c
2;q2E
k
hcRS;qikDc 1 2 ;c
2;q2ik X
rsk
h R;qirh S;qis
h1;qirh1;qis QE(RrSk) 98
1 hc;qiRS
X
RS k0
hc 1;qikDc1 2 ;c
2;q2E
k
hcRS;qikDc 1 2 ;c
2;q2E
k
Xk
r0
h R; k;qirh S;qik
h1;1S k;qirh1;qikQE(RrSkr(1S))
1 hc;qiRS
X
RS k0
hc 1;qikDc1 2 ;c
2;q2E
k
hcRS;qikDc 1 2 ;c
2;q2E
k
h S;qik
h1;qik 2f1( k; R;1S kjq;q1SR)
1
hc;qiRS X
RS k0
hc 1;qikDc1 2 ;c
2;q2E
k
hcRS;qikDc 1 2 ;c
2;q2E
k
h S;qik
h1;qik Gq 1S k;1SR 1S;1S kR
1
hc;qiRS X
RS k0
hc 1;qikDc1 2 ;c
2;q2E
k
hcRS;qikDc 1 2 ;c
2;q2E
k
(98)
h S;qik h1;qik
h1S;qi k h1RS;qi k
1
hc;qiRS X
RS k0
Dc 1; S R;c1 2 ;cg1
2 ;qE
k qk(SR) hcRS;1;c 1
2 ;cg1 2 ;qE
k
1
hc;qiRS 4f3(c 1; R S;c1 2 ;cg1
2 ;cRS;c 1 2 ;cg1
2 jq;qSR)
1
hc;qiRSGq
RS;cRS;c1 2 ;c 1
2 c1
2 RS;c 1
2 RS;0;c 2
66 4
3 77 5:
Result 0, except forRS0. Then the 3f2equals 1 and we get 1. p COROLLARY4.5.
Xs
m0
Xt
n0
( 1)mn s m q
t
n q q m2 n2 mt 1; if st0;
0; otherwise:
99
The following result from [13] leads to the same formula.
F1(a;b;b0;cjq;qc a b;qc a b b0)
F1(a;b;b0;cjq;qc a b b0;qc a b0)Gq c;c a b b0 c a;c b b0
100 ;
wherejqc a b b0j<1. Putb s;b0 tto obtain Xs
m0
Xt
n0
( 1)mn s m q
t
n qha;qimnhcmn;qist m n
QE m
2 n
2 m(c a)n(c as)
hc a;qist: 101
Now multiply withqa(st)and applyDstq;qa to both sides to obtain Xs
m0
Xt
n0
s m q
t
n qhcmn;qist m n QE mn
2
m
2 n
2 c(mn)ns
1 102
Finally, applyDstq;qcto both sides to obtain Xs
m0
Xt
n0
( 1)mn s m q
t
n qQE mn 2
m
2 n
2 ns
QE st m n 2
(st m n)(mn)
0 103
The following important formula is aq-analogue of [34] and [37].
THEOREM 4.6. [14]. If fCng1n0; fang1n0 are bounded sequences of complex numbers then
X
~ m
Cm1...mnQn
j1xmjhaj;qimj Qn
j1h1;qimj QE Xn
k1
mkXk
l2
al
!
X1
N0
CNxND Pn
k1ak;qE
N
h1;qiN QE NXn
l2
al
! : 104