We consider the second-order mock theta function 𝒟5 (q), which Hikami came across in his work on mathematical physics and quantum invariant of three manifold. We give their bilateral form, and show that it is the same as bilateral third-order mock theta function of Ramanujan. We also show that the mock theta function 𝒟5 (q) outside the unit circle is a theta function and also write h1(q) as a coefficient of z0 of a theta series. First writing h1(q) as a coefficient of a theta function, we prove an identity for h1(q).
1. Brief History of Mock Theta Functions
The mock theta functions were introduced and named by Ramanujan and were the subjects of Ramanujan’s last letter to Hardy, dated January 12, 1920, to be specific [1, 2]. Ramanujan gave a list of seventeen functions which he called “mock theta functions.” He divided them into four groups of functions of order 3, 5, 5, and 7. Ramanujan did not rigorously define a mock theta function nor he define the order of a mock theta function. A definition of the order of a mock theta function is given in the Gordon-McIntosh paper on modular transformation of Ramanujan’s fifth and seventh-order mock theta functions [3] Watson [4] while constructing transformation laws for the mock theta function found three further mock theta functions of order 3.
In 1976, Andrews while visiting Trinity college, Cambridge, discovered in the mathematical library of the college a notebook written by Ramanujan towards the end of his life and Andrews called it “Lost” Notebook. In the lost notebook were six more mock theta functions and linear relation between them. Andrews and Hickerson [5] called these mock theta functions of sixth-order and proved the identities.
In the “Lost” Notebook on page 9 appear four more mock theta functions which were called by Choi of tenth-order. Ramanujan also gave eight linear relations connecting these mock theta functions of tenth-order and these relations were proved by Choi [6].
Gordon and McIntosh listed eight functions in their eighth-order paper [7], but later, in their survey paper [8], classified only four of them as eighth-order. The other four are more simple in their modular transformation laws and therefore are considered to be of lower order.
We now come to the second-order mock theta functions. McIntosh [9] considered three second-order mock theta functions and gave transformation formulas for them. Hikami [10] in his work on mathematical physics and quantum invariant of three manifold came across the q-series:
𝒟5(q)=∑n=0∞qn(-q;q)n(q;q2)n+1=1(q;q2)∞2∑n=0∞(q;q2)n2q2n
and proved that 𝒟5(q) is a mock theta function and called it of “2nd” order.
He further showed that 𝒟5(q) is a sum of two mock theta functions h1(q) and ω(q) where h1(q) is of second-order and ω(q) is Ramanujan’s mock theta function of third-order. This 𝒟5(q) will be the basis of our study in this paper.
Before we begin with the study of 𝒟5(q) and h1(q) it will be appropriate to mention the work done earlier.
Gordon and McIntosh in their survey paper [8] have shown that h1(q) is essentially the odd part of the second-order mock theta function B(q), which appears as β(q) in Andrews’ paper on Mordell integrals and Ramanujan’s lost notebook [11] and also in McIntosh paper on second-order mock theta functions [9]. In particular,
h1(q2)=B(q)-B(-q)4q,
where
B(q)=∑n=0∞qn(n+1)(-q2;q2)n(q;q2)n+12=∑n=0∞qn(-q;q2)n(q;q2)n+1.
Since the even part of B(q) is the ordinary theta function
B(q)+B(-q)2=(q4;q4)∞(-q2;q2)∞4,
it follows that the odd part and h1(q) are second-order mock theta functions. Thus 𝒟5(q) is a linear combination of second-order and third-order mock theta function. In some sense, mock theta functions of orders 1, 2, 3, 4, and 6 are all in the same family.
The paper is divided as follows.
In Section 3 we expand 𝒟5(q) as a bilateral q-series and show that it is also a sum of the second-order mock theta function 𝒟5(q) and the third-order mock theta function ω(q). By using Bailey’s transformation we have the interesting result that the bilateral 𝒟5,c(q) is the same as the bilateral ωc(q).
In Section 4, using bilateral transformation of Slater, we write 𝒟5,c(q) as a bilateral series ψ22 series with a free parameter c.
In Section 5, a mild generalization 𝒟5,c(z,α) of 𝒟5,c(q) is given and we show that this generalized function is a Fq-function.
In Section 6 we show that 𝒟5(q), outside the unit circle |q|=1, is a theta function.
In Section 7 we state a generalized Lambert Series expansion for h1(q) as given in [8].
In Section 8 we show that h1(q) is a coefficient of z0 of a theta function.
In Section 9 we prove an identity for h1(q) using h1(q) as a coefficient of z0 of a theta function.
In Section 10 a double series expansion for h1(q) is obtained by using Bailey pair method.
2. Basic Preliminaries
We first introduce some standard notation.
If q and a are complex numbers with |q|<1 and n is a nonnegative integer, then
(a)0=(a;q)0=1,(a)n=(a;q)n=∏k=0n-1(1-aqk),(a)∞=(a;q)∞=∏k=0∞(1-aqk),(a1,…,am)n=(a1,…,am;q)n=(a1;q)n,…,(am;q)n.
Ramanujan’s mock theta function of third-order ω(q) and ν(q) is
ω(q)=∑n=0∞q2n(n+1)(q;q2)n+12,ν(q)=∑n=0∞qn(n+1)(-q;q2)n+1,φ(q)=∑n=-∞∞qn2=(-q;q2)∞2(q2;q2)∞=(-q;-q)∞(q;-q)∞.
We will use the following notations for θ-functions.
Definition 2.1.
If |q|<1 and x≠0, then
j(x,q)=(x,qx,q;q)∞.
If m is a positive integer and a is an integer,
Ja,m=j(qa,qm),J̅a,m=j(-qa,qm),Jm=j(qm,q3m)=(qm;qm)∞,j(qx,q)=j(x,q),j(x,q)=-xj(x-1,q),j(qnx,q)=(-1)nq-n(n-1)/2x-nj(x,q),ifnisaninteger.
By Jacobi’s triple product identity [12, page 282]
j(x,q)=∑n=-∞∞(-1)nqn(n-1)/2xn.
2.1. More Definitions
If z is a complex number with |z|≠1, then
ε(z)={1if|z|<1,-1if|z|>1.
If s is an integer, then
sg(s)={1ifs≥0,-1ifs<0.
Using these definitions,
11-z=ε(z)∑s=-∞sg(s)=ε(z)∞zs.
We shall use the following theorems.
Theorem 2.2 (see [13, Theorem 1.3, page 644]).
Let q be fixed, 0<|q|<1. Let a,b, and m be fixed integers with b≠0 and m≥1. Define
F(z)=1j(qazb,qm).
Then F is meromorphic for z≠0, with simple poles at all points z0 such that z0b=qkm-a for some integer k. The residue of F(z) at such a point z0 is
(-1)k+1qmk(k-1)/2z0bJm3.
Theorem 2.3 (see [13, Theorem 1.8(a), page 647]).
Suppose that
F(z)=∑rFrzr
for all z≠0 and that F(z) satisfies
F(qz)=Cz-nF(z),
where 0<|q|<1 and C≠0. Then
F(z)=∑r=0n-1Frzrj(-C-1qrzn,qn).
Truesdell [14] calls the functions which satisfy the difference equation
∂∂zF(z,α)=F(z,α+1)
as F-function. He unified the study of these F-functions.
The functions which satisfy the q-analogue of the difference equation
Dq,zF(z,α)=F(z,α+1),
where
zDq,zF(z,α)=F(z,α)-F(zq,α)
are called Fq-functions.
3. Bilateral 𝒟5(q) as a Sum of Two Mock Theta Functions of Different Orders
(i) We shall denote the bilateral of 𝒟5(q) by 𝒟5,c(q). We define it as
(q;q2)∞2𝒟5,c(q)=∑n=-∞∞(q;q2)n2q2n.
Now
(q;q2)∞2𝒟5,c(q)=∑n=-∞∞(q;q2)n2q2n=∑n=0∞(q;q2)n2q2n+∑n=-1-∞(q;q2)n2q2n=∑n=0∞(q;q2)n2q2n+∑n=0∞q2n2+2n(q;q2)n+12,
and we use (1.2) in the first summation and (2.2) in the second summation, to write
(q;q2)∞2𝒟5,c(q)=(q;q2)∞2𝒟5(q)+ω(q).
Thus 𝒟5,c(q) is a sum of a second-order mock theta function and a third-order mock theta function.
(ii) Transformation of Bilateral 𝒟5,c(q) into bilateral ωc(q) is as follows.
It is very interesting that the bilateral 𝒟5,c(q) can be written as bilateral third-order mock theta function ωc(q).
We use Bailey’s bilateral transformation [15, 5.20(ii), page 137]:
ψ22[a,bc,d;q,z]=(az,bz,cq/abz,dq/abz;q)∞(q/a,q/b,c,d;q)∞×ψ22[abz/c,abz/daz,bz;q,cd/abz].
Letting q→q2, and setting a=b=q, c=d=0, and z=q2 in (3.4), we get
Slater [15, (5.4.3), page 129] gave the following transformation formula, and we have taken r=2:
(b1,b2,q/a1,q/a2,dz,q/dz;q)∞(c1,c2,q/c1,q/c2;q)∞ψ22[a1,a2b1,b2;q,z]=qc1(c1/a1,c1/a2,qb1/c1,qb2/c1,dc1z/q,q2/dc1z;q)∞(c1,q/c1,c1/c2,qc2/c1;q)∞ψ22[qa1/c1,qa2/c1qb1/c1,qb2/c1;q,z]+idem(c1;c2),
where d=a1a2/c1c2, |b1b2/a1a2|<|z|<1, and idem(c1;c2) after the expression means that the preceding expression is repeated with c1 and c2 interchanged.
In the transformation it is interesting that the c’s are absent in the ψ22 series on the left side of (4.1). This gives us the freedom to choose the c’s in a convenient way.
Letting q→q2 and setting, a1=a2=q, b1=b2=0, and z=q2 in (4.1), so d=q2/c1c2 and 0<|z|<1, to get
(q;q2)∞4(q4/c1c2;q2)∞(c1c2/q2;q2)∞(c2;q2)∞(q2/c2;q2)∞𝒟5,c(q)=q2c1(c1/q;q2)∞2(q2/c2;q2)∞(c2;q2)∞(c1/c2;q2)∞(q2c2/c1;q2)∞∑n=-∞∞(q3/c1;q2)n2q2n+idem(c1;c2).
By choosing c1 suitably we can have different expansion identities. Moreover (4.2) can be seen as a generalization of (3.3).
5. Mild Generalization of 𝒟5,c(q)
We define the bilateral generalized function 𝒟5,c(z,α) as
(q;q2)∞2𝒟5,c(z,α)=1(z)∞∑n=-∞∞(q;q2)n2(z)nqnα+n.
For α=1, z=0, 𝒟5,c(z,α) reduce to 𝒟5,c(q).
Now
Dq,z[𝒟5,c(z,α)]=1z[𝒟5,c(z,α)-𝒟5,c(zq,α)]=1z(q;q2)∞2[1(z)∞∑n=-∞∞(q;q2)n2(z)nqnα+n-1(zq)∞∑n=-∞∞(q;q2)n2(zq)nqnα+n]=1z(q;q2)∞21(z)∞∑n=-∞∞(q;q2)n2(z)nqnα+n(1-(1-zqn))=1(q;q2)∞2(z)∞∑n=-∞∞(q;q2)n2(z)nqnα+2n=𝒟5,c(z,α+1).
So 𝒟5,c(z,α+1) is an Fq-function.
Being Fq-function it has unified properties of Fq-functions. For example, one has the following.
(i) The inverse operator Dq,x-1 of q-differentiation is related to q-integration as
Dq,x-1ϕ(x)=(1-q)-1∫ϕ(x)dqx.
See Jackson [16].
(ii) Dq,znFq(z,α)=Fq(z,α+n), where n is a nonnegative integer.
6. Behaviour of 𝒟5(q) outside the Unit Circle
By definition (1.1)
𝒟5(q)=∑n=0∞(-q;q)n(q;q2)n+1qn.
Replacing q by 1/q and writing 𝒟5*(q) for 𝒟5(1/q) [10],
𝒟5*(q)=∑n=0∞(-1)nq(n2+n)/2(-q;q)n(q;q2)n+1=1-q2+q6-q12+q20-q30+⋯=∑n=0∞(-1)nqn2+n,
which is a θ-function.
7. Lambert Series Expansion for h1(q)
For the double series expansion, we first require the generalized Lambert series expansion for h1(q).
By Entry 12.4.5, of Ramanujan’s Lost Notebook [17, page 277], Hikami [10] noted that
𝒟5(q)=2h1(q)-(-q;q)∞2ω(q),
where
h1(q)=∑n=0∞(-q;q)2n(q;q2)n+12qn.
There is a slight misprint in the definition h1(q) in Hikami’s paper [10] which has been corrected and Gordon and McIntosh have also pointed out in their survey [8].
In the following theorem of Hickerson [13, Theorem 1.4, page 645],
∑r=-∞∞xr1-qry=J13j(xy,q)j(x,q)j(y,q)
let q→q2, and then put y=q, to get
∑r=-∞∞xr1-q2r+1=J23j(qx,q2)j(x,q2)j(q,q2).
For |q|<1, and z≠0 and not an integral power of q, let
A(z)=12θ4(0,q)J23j(qz,q2)j(z,q2)j(q,q2)j(zq,q2).
Theorem 8.1.
Let q be fixed with 0<|q|<1. Then h1(q) is the coefficient of z0 in the Laurent series expansion of A(z) in the annulus |q|<|z|<1.
Proof.
By (7.3)
2θ4(0,q)h1(q)=∑n=-∞∞(-1)nqn(n+2)1-q2n+1=coefficientofz0in∑n=-∞∞zn1-q2n+1∑s=-∞∞(-1)sqs2+s(zq)-s=coefficientofz0inJ23j(qz,q2)j(z,q2)j(q,q2)j(zq,q2)
dividing by 2θ4(0,q) gives the theorem.
9. An Identity for h1(q)Theorem 9.1.
If 0<|q|<1 and z is neither zero nor an integral power of q, then
A(z,q)=j(z,q2)h1(q)-(-12)∑r=-∞∞(-1)rqr2+3r-1zr+21-q2r+2z-12∑r=-∞∞(-1)rqr2+3r+1z-r-11-q2r+2z-1.
Define
L(z)=-12∑r=-∞∞(-1)rqr2+3r-1zr+21-q2r+2zM(z)=12∑r=-∞∞(-1)rqr2+3r+1z-r-11-q2r+2z-1,F(z)=A(z)+L(z)+M(z).
The scheme will be first to show that F(z) satisfies the functional relation:
F(q2z)=-z-1F(z).
One considers the poles of L(z) and M(z) and shows that the residue of F(z) at these poles is zero. So F(z) is analytic at these points. One then shows that the coefficients of z0 in L(z) and M(z) are zero and equating the coefficient of z0 in (9.4) one has the theorem.
Proof.
We show that
F(q2z)=-z-1F(z).
We shall show that each of A(z), L(z), and M(z) satisfies the functional equation:
A(z)=12(-q;q)∞(q;q)∞J23j(qz,q2)j(z,q2)j(q,q2)j(zq,q2),
and so
A(q2z)=12(-q;q)∞(q;q)∞J23j(q3z,q2)j(q2z,q2)j(q,q2)j(zq,q2).
We employ (2.11) on the right-hand side to get
A(q2z)=12(-q;q)∞(q;q)∞J23(-1)z-1q-1j(zq,q2)(-1)qz-1(-1)z-1j(z,q2)j(q,q2)j(zq,q2)A(q2z)=-z-1A(z).
We now take L(z):
L(q2z)=12∑r=-∞∞(-1)rqr2+3r-1(zq2)r+21-q2r+2(zq2).
Writing r-1 for ron the right-hand side we have
L(q2z)=-z-1L(z).
Similarly only writing r+1 for r we have
M(q2z)=-z-1M(z).
Hence the functional equation (9.4) is proved.
Obviously L(z) and M(z) are meromorphic for z≠0. L(z) has simple poles at z=q2k-2 and M(z) has simple poles at z=q2k+2. Hence F(z) is meromorphic for z≠0 with, at most, simple poles at z=q2k±2.
Taking r=0 in (9.2), we calculate the residue of L(z) at the point z=1/q2:
ResidueofL(z)=limz→1/q212(z-1q2)z2q-1(z-(1/q2))q2=12q-5.
For the residue of A(z) at z=1/q2, take b=1, k=-1, m=2, a=0 in (2.16) to get
ResidueofA(z)=12(-q;q)∞(q;q)∞J23j(1/q,q2)j(1/q3,q2)j(q,q2)1J23=12(-q;q)∞(q;q)∞j(q,q2)j(q3,q2)q4j(q,q2)=12(-q;q)∞(q;q)∞1q4(q3;q2)∞(1q;q2)∞(q2;q2)∞=12(-q;q)∞(q;q)∞1q4(1-(1/q))(1-q)(q;q2)∞(q;q2)∞(q2;q2)∞=-12(-q;q)∞(q;q)∞1q5(q;q)∞(-q;q)∞=-12q-5.
So the residue of F(z) at z=1/q2 is -(1/2)q5+0+(1/2)q5=0.
Now we calculate the residue at z=q2:
ResidueofM(z)=limz→q212(z-q2)qz-1(1-q2z-1)=limz→q212(z-q2)q(z-q2)=q2,
and for the residue of A(z) at z=q2, taking b=1, k=1, m=2, and a=0 in (2.16), so
ResidueofA(z)=12(-q;q)∞(q;q)∞J23j(q3,q2)j(q,q2)j(q,q2)q2J23=12(-q;q)∞(q;q)∞j(q3,q2)q2=12(-q;q)∞(q;q)∞(q3;q2)∞(1q;q2)∞(q2;q2)∞q2=12(-q;q)∞(q;q)∞(1-(1/q))(1-q)(q;q2)∞(q;q2)∞(q2;q2)∞q2=-12q.
Hence the residue of F(z) at z=q2 is 0+(1/2)q-(1/2)q=0. Hence F(z) is analytic at z=q2.
Since F(z) satisfies (9.4), so F(z) is analytic at all points of the form z=q2k±2 and hence for all z≠0.
We now apply (2.20) with n=1 and c=-1 and q replaced by q2 to get
F(z)=F0j(z,q2),
where F0 is the coefficient of z0 in the Laurent expansion of F(z),z≠0.
Now for |q|<|z|<1, by Theorem 8.1, the coefficient of z0 in A(z) is h1(q).
For such z, |q2r+2z|<1 if and only if r≥0.
That is,
ε(q2r+2z)=sg(r).
Hence by (2.15)
11-q2r+2z=sg(r)∑r=-∞sg(r)=sg(s)∞q(2r+2)szs.
So
L(z)=-12∑sg(r)=sg(s)∞sg(r)(-1)rqr2+3r-1+(2r+2)szr+2+s.
If sg(r)=sg(s), then r+s+2 is either ≥1 or ≤-1; so coefficient of z0 in L(z) is 0. Similarly the coefficient of z0 in M(z) is 0 and so the coefficient of z0 in F(z) is h1(q).
Hence by (9.17), we have
F(z)=h1(q)j(z,q2),
which gives the theorem.
10. Double Series Expansion
Now we derive the double series expansion for h1(q). We shall use the Bailey pair method, as used by Andrews [18] for fifth and seventh-order mock theta functions and by Andrews and Hickerson [5] for sixth-order mock theta functions.
We define Bailey pair.
Two sequences {αn} and {βn}, n≥0, form a Bailey pair relative to a number a if
βn=∑r=0nαr(q)n-r(aq)n+r,
for all n≥0.
Corollary 10.1 (see [5, Corollary. 2.1, page 70]).
If {αn} and {βn} form a Bailey pair relative to a, then
∑n=0∞(ρ1)n(ρ2)n(aq/ρ1ρ2)nαn(aq/ρ1)n(aq/ρ2)n=(aq)∞(aq/ρ1ρ2)∞(aq/ρ1)∞(aq/ρ2)∞∑n=0∞(ρ1)n(ρ2)n(aqρ1ρ2)nβn
provided that both sums converge absolutely.
We state the theorem of Andrews and Hickerson [5, Theorem 2.3, pages 72-73].
Let a, b, c, and q be complex numbers with a≠1, b≠0, c≠0, q≠0, and none a/b, a/c, qb, qc of the form q-k with k≥0. For n≥0, define
An′(a,b,c,q)=qn2(bc)n(1-aq2n)(a/b)n(a/c)n(1-a)(qb)n(qc)n×∑j=0n(-1)j(1-aq2j-1)(a)j-1(b)j(c)jq(2j)(bc)j(q)j(a/b)j(a/c)j,Bn′(a,b,c,q)=1(qb)n(qc)n.
Then the sequences {An′(a,b,c,q)} and {Bn′(a,b,c,q)} form a Bailey pair relative to a.
Letting q→q2 and then taking a=q2, b=c=q, in (10.3), we get
An′(q2,q,q2)=q2n2+2n(1-q4n+2)(q;q2)n2(1-q2)(q3;q2)n2×∑j=0n(-1)j(1-q4j)(q2;q2)j-1(q;q2)j2qj2+j(q2;q2)j(q;q2)j2=(1-q)2(1+q2n+1)q2n2+2n(1-q2)(1-q2n+1)∑j=0n(-1)jq-j2-j(1+q2j)=(1-q)2(1+q2n+1)q2n2+2n(1-q2)(1-q2n+1)[1+∑j=-nn(-1)jq-j2-j],Bn′(q2,q,q,q2)=1(q3;q2)n2.
Now letting q→q2 and then setting ρ1=-q, ρ2=-q2, a=q2 in (10.2) we get
∑n=0∞qn(-q;q2)n(-q3;q2)nαn=(q;q2)∞(q4;q2)∞(-q2;q2)∞(-q3;q2)∞∑n=0∞(-q;q2)n(-q2;q2)nqnβn.
Taking An′ and Bn′ for αn′ and βn′, respectively, in (10.5) and using the definition of h1(q), we get
(q;q2)∞(q2;q2)∞(-q;q2)∞(-q2;q2)∞h1(q)=∑n=0∞q2n2+3n1-q2n+1[1+∑j=-nn(-1)jq-j2-j]
or
h1(q)=(-q;q2)∞(-q2;q2)∞(q;q2)∞(q2;q2)∞∑n=0∞q2n2+3n1-q2n+1[1+∑j=-nn(-1)jq-j2-j],
which is the double series expansion for h1(q).
This double series expansion can be used to get more properties of 𝒟5(q).
RamanujanS.1927Cambridge, UKCambridge University PressRamanujanS.1962New York, NY, USAChelseaGordonB.McIntoshR. J.Modular transformations of Ramanujan's fifth and seventh order mock theta functions20037193222WatsonG. N.The final problem: an account of the mock theta functions1936115580AndrewsG. E.HickersonD.Ramanujan's “lost” notebook. VII. The sixth order mock theta functions199189160105MR112309910.1016/0001-8708(91)90083-JZBL0739.11042ChoiY.-S.Tenth order mock theta functions in Ramanujan's lost notebook. IV2002354270573310.1090/S0002-9947-01-02861-6MR1862564ZBL1043.33012GordonB.McIntoshR. J.Some eighth order mock theta functions200062232133510.1112/S0024610700008735MR1783627ZBL1031.11007GordonB.McIntoshR. J.A survey of classical mock theta
functionspreprintMcIntoshR. J.Second order mock theta functions2007502284290MR2317449ZBL1133.11013HikamiK.Transformation formula of the “second” order mock theta function2006751939810.1007/s11005-005-0039-1MR2207051ZBL1109.33020AndrewsG. E.Mordell integrals and Ramanujan's lost notebook1981899BerlinSpringer1048HardyG. H.WrightE. M.19684thLondon, UKOxford University PressHickersonD.A proof of the mock theta conjectures1988943639660MR96924710.1007/BF01394279ZBL0661.10059TruesdellC.1948Princeton, NJ, USAPrinceton University Pressiv+182Annals of Mathematics Studies, no. 18MR0023960GasperG.RahmanM.199035Cambridge, UKCambridge University Pressxx+287Encyclopedia of Mathematics and Its ApplicationsMR1052153JacksonF. H.Basic integration19512116AndrewsG. E.BerndtB. C.2005New York, NY, USASpringerxiv+437MR2135178AndrewsG. E.The fifth and seventh order mock theta functions19862931113134MR81491610.2307/2000275ZBL0593.10018