Journal of Combinatorial Theory, Series A 130 (2015) 1–14

Contents lists available at ScienceDirect

Journal of Combinatorial Theory,

Series A
www.elsevier.com/locate/jcta

A truncated Jacobi triple product theorem

Ae Ja Yee 1
Department of Mathematics, The Pennsylvania State University, University Park,
PA 16802, USA

a r t i c l e i n f o a b s t r a c t

Article history: Recently, G.E. Andrews and M. Merca considered a truncated

Received 28 March 2014
Available online xxxx
Keywords:
Partitions
Euler’s pentagonal number theorem
Jacobi’s triple product identity
version of Euler’s pentagonal number theorem and obtained
a nonnegativity result. They asked the same question on a
truncated Jacobi triple product identity, which can be found
as a conjecture in a paper of V.J.W. Guo and J. Zeng. In this
paper, we provide an answer to the question, which is purely
combinatorial. We also provide a combinatorial proof of the
main theorem in the paper of Andrews and Merca.
© 2014 Elsevier Inc. All rights reserved.

1. Introduction

One of the well-known theorems in the partition theory is the following pentagonal
number theorem.

Theorem 1.1 (Euler’s pentagonal number theorem). We have

∞

(−1)n q n(3n−1)/2 = (q; q)∞ .
n=−∞

E-mail address: auy2@psu.edu.

1
This work was partially supported by a grant (#280903) from the Simons Foundation.

http://dx.doi.org/10.1016/j.jcta.2014.10.005
0097-3165/© 2014 Elsevier Inc. All rights reserved.
2 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

This identity leads to

∞
1
(−1)n q n(3n−1)/2 = 1. (1.1)
(q; q)∞ n=−∞

Here and throughout this paper, we use the following customary q-series notation:

(a; q)0 := 1,
∞
  
(a; q)∞ := 1 − aq k ,
k=0

(a; q)∞
(a; q)n := for any n,
(aq n ; q)∞
    
n n 0, if k < 0 or k > n,
:= := (q;q)n
k k q (q;q)k (q;q)n−k , otherwise.

Recently, G.E. Andrews and M. Merca considered a truncated version of (1.1) and
obtained the following result:

1 
k−1
 
(−1)n q n(3n+1)/2 1 − q 2n+1
(q; q)∞ n=0
∞

k  
k−1 q 2 +(k+1)n
n−1
= 1 + (−1) , (1.2)
n=1
(q; q)n k−1

from which they deduced the following partition theorem.

Theorem 1.2. (See [4, Theorem 1.1].) For n > 0, k ≥ 1


k−1
    
(−1)k−1 (−1)j p n − j(3j + 1)/2 − p n − j(3j + 5)/2 − 1 = Mk (n)
j=0

where Mk (n) is the number of partitions of n in which k is the least integer that is not
a part and there are more parts > k than there are < k.

Theorem 1.1 has the following generalization.

Theorem 1.3 (The Jacobi triple product identity). For z = 0,

∞
 n
(−z)n q 2 = (z; q)∞ (q/z; q)∞ (q; q)∞ .
n=−∞
 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14 3

For positive integers S and R with 1 ≤ S < R/2, replace z by q S and q by q R in the
Jacobi triple product to obtain

∞ n
1
S R R−S R R R
(−1)n q R 2 +Sn = 1.
(q ; q )∞ (q ; q )∞ (q ; q )∞ n=−∞

Andrews and Merca considered the truncated version of the left hand side: for
k ≥ 1,

1 
k−1
 
(−1)n q Rn(n+1)/2−Sn 1 − q (2n+1)S , (1.3)
(q S ; q R )∞ (q R−S ; q R )∞ (q R ; q R )∞ n=0

and at the end of their paper, they asked the following question:

There is a substantial amount of numerical evidence to conjecture that (1.3) has

nonnegative coefficients if k is odd and nonpositive coefficients if k is even. Note that
the corollary implies the case R = 3 and S = 1.

The corollary referred to in the question is our Eq. (1.2). V.J.W. Guo and J. Zeng [5]
found two further incidences of this nature and conjectured the problem of Andrews and
Merca as well.
Another question of Andrews and Merca is to provide a combinatorial proof of The-
orem 1.2 hopefully characterizing the partitions remaining after a sieving process.
The purpose of this paper is to provide answers to these questions of Andrews and
Merca. Recently, R. Mao [7] has proved the nonnegativity conjecture of Andrews and
Merca along with a conjecture of Guo and Zeng that arises from a truncated series of
the following Jacobi identity:

∞

(q; q)3∞ = (−1)j (2j + 1)q j(j+1)/2 .
j=0

Mao’s proofs are based on q-series manipulations. The proof presented in this paper is
combinatorial.
A variety of truncated Euler’s pentagonal number theorem and truncated Jacobi’s
triple product theorem can be found in literature, for instance in [1,3,8–10]. L. Kolitsch
and M. Burnette [6] have found two further partition functions that are related to the
partition function Mk (n) of Andrews and Merca.
This paper is organized as follows. In Section 2, we first sketch the combinatorial
proof of the Jacobi triple product identity given by E.M. Wright [11] and then answer
the question on the truncated Jacobi triple product identity. That is:
4 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

Theorem 1.4. For k ≥ 1 and positive integers S and R with 1 ≤ S < R/2, let
k−1 ∞
j=0 (−1) q
j Rj(j+1)/2−Sj
(1 − q (2j+1)S ) 
=1+ ak (n)q n . (1.4)
(q S ; q R )∞ (q R−S ; q R )∞ (q R ; q R )∞ n=1

Then (−1)k−1 ak (n) ≥ 0 for all n ≥ 1.

The generating function of ak (n) is given explicitly in Section 3. In Section 4, we prove

Theorem 1.2 combinatorially.

2. Proof of Theorem 1.4

Before we prove Theorem 1.4, we first sketch Wright’s combinatorial proof of the
Jacobi triple product identity. Throughout this paper, for a partition λ, |λ| denotes the
sum of its parts and (λ) denotes the number of its parts. We also denote the largest
part of λ by l(λ).

2.1. Modular Ferrers diagram

Let m and s be positive integers with m ≥ s. For a partition π into parts πi congruent
to s mod m, its m-modular Ferrers diagram is the diagram in which the i-th row has
λi /m boxes, the first box of each row has s, and the other boxes have m. Then the sum
of the numbers in the boxes equals the number that π partitions. If the parts of π are all
distinct, we can draw its modular Ferrers diagram in the form of a staircase. Moreover,
when s < m, if necessary, we may use triangles for the boxes on the main diagonal. For
instance, the following is an m-modular Ferrers diagram in the form of a staircase.

@s m m m m m
@
@s m m m
@
@s m
@
@s
@

2.2. Wright’s bijection ϕ

Let us replace q by q R and z by z −1 q −S in the Jacobi triple product identity. Note

that
   −1 R−S R  
(−z)(μ )−(μ ) q |μ |+|μ | ,
1 2 1 2
zq S ; q R ∞
z q ;q ∞ = (2.1)
(μ1 ,μ2 )
 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14 5

where the sum is over all pairs of partitions (μ1 , μ2 ) into distinct parts, and the parts
of μ1 and μ2 are congruent to S and −S, respectively, mod R. We now consider pairs
(μ1 , μ2 ) with (μ1 ) − (μ2 ) = d only. Then the generating function is

 q R(d −d)/2+Sd
2

q |μ |+|μ2 |
1
= . (2.2)
(q R ; q R )∞
(μ1 ,μ2 )
(μ1 )−(μ2 )=d

This can be shown as follows. Let π be a partition into parts congruent to R modulo R,
and let Δ be the following modular partition of R(d2 − d)/2 + Sd in the triangular
form:

((d − 1)R + S, (d − 2)R + S, . . . , R + S, S), if d ≥ 0,

Δ=
(|d|R − S, (|d| − 1)R − S, . . . , 2R − S, R − S), if d < 0.

We now concatenate π and Δ as follows. If d ≥ 0, then we put the diagram of Δ to the

right of the diagram of π with the largest part of Δ adjacent to the largest part of π;
if d < 0, then we put the modular diagram of Δ on the top of the modular diagram
of π with the largest part of Δ right above the largest part of π. See the pictures
below. Each box has weight R and the triangles in the first picture and the second
picture have weight S and R − S, respectively. Thus, the first picture is the diagram of
π = (4R, 4R, 3R, 3R, 2R, 1R) combined with Δ = (2R + S, R + S, S), and the second
picture is the diagram of π = (6R, 5R, 2R) combined with Δ = (3R − S, 2R − S, R − S).

@ @
@ @
@ @
@ @
@ @
@ @

We now extend the diagonal to divide the concatenated partition into two diagrams.
We assign S to each triangle below the diagonal and R − S to each triangle above the
diagonal. Along the diagonal, we read off the number of boxes with the triangle to the
right of the diagonal in each row to form a partition μ1, and read off the number of boxes
with the triangle above the diagonal in each column to form a partition μ2 . With the same
examples above, we obtain (μ1 , μ2 ) = ((6R+S, 5R+S, 3R+S, 2R+S, S), (3R−S, R−S))
from the first picture and (μ1 , μ2 ) = ((2R+S, S), (6R−S, 5R−S, 3R−S, 2R−S, R−S)).
6 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

@ @
@ @
@ @
@ @
@ @
@ @
@ @
@ @
@ @
@ @

In either case, it can be easily checked that μ1 and μ2 are partitions into distinct
parts congruent to S and −S modulo R, respectively, and |π| + |Δ| = |μ1 | + |μ2 | and
(μ1 ) − (μ2 ) = d. Also, this process is reversible. Thus, by (2.1) and (2.2), we have

∞

   −1 R−S R  1
(−z)d q R(d −d)/2+Sd ,
2
zq S ; q R ∞
z q ;q ∞ =
(q R ; q R )∞
d=−∞

which is the Jacobi triple product identity.

2.3. Proof of Theorem 1.4

Let us take the left hand side of (1.4): for k ≥ 1,

1 
k−1
 
(−1)j q Rj(j+1)/2−Sj 1 − q (2j+1)S . (2.3)
(q S ; q R )∞ (q R−S ; q R )∞ (q R ; q R )∞ j=0

We now show that the coefficients are all nonnegative if k is odd, and the coefficients are
all nonpositive if k is even.
We first note that

1 
q |λ |+|λ2 |
1
= , (2.4)
(q S ; q R )∞ (q R−S ; q R )∞
(λ1 ,λ2 )

where the sum is over all partition pairs (λ1 , λ2 ), λ1 has parts congruent to S mod R,
and λ2 has parts congruent to −S mod R; and

1 
k−1
 
R R
(−1)j q Rj(j+1)/2−Sj 1 − q (2j+1)S
(q ; q )∞ j=0

1 
k
= (−1)j q Rj(j−1)/2+Sj
(q ; q R )∞
R
j=−k+1
 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14 7


k 
(−1)d q |μ |+|μ2 |
1
= , (2.5)
d=−k+1 1 2
(μ ,μ )
(μ1 )−(μ2 )=d

where the inner sum is over all partition pairs (μ1 , μ2 ) with (μ1 ) − (μ2 ) = d, μ1 is a
partition into distinct parts congruent to S mod R and μ2 is a partition into distinct
parts congruent to −S mod R.
In view of the analysis in (2.4) and (2.5), we define a partition set whose weighted
generating function is (2.3), and then we will construct an injection from the subset with
negative (positive) weight to the subset with positive (negative) weight for k odd (even,
resp.). Let Pk (n) be the set of quadruples (λ1 , λ2 , μ1 , μ2 ) such that

(i) λ1 is a partition into parts congruent to S mod R;

(ii) λ2 is a partition into parts congruent to −S mod R;
(iii) μ1 is a partition into distinct parts congruent to S mod R;
(iv) μ2 is a partition into distinct parts congruent to −S mod R;
(v) |λ1 | + |λ2 | + |μ1 | + |μ2 | = n;
(vi) −k + 1 ≤ (μ1 ) − (μ2 ) ≤ k,

and Pk (n; e) and Pk (n; o) be the subsets of Pk (n) with (μ1 ) − (μ2 ) even and odd,
respectively. We also denote the size of Pk (n; e) and Pk (n; o) by pk (n; e) and pk (n; o),
respectively. Then, by (2.3) and (2.4),

∞
  
pk (n; e) − pk (n; o) q n
n=0

1 
k−1
 
= S R R−S R R R
(−1)n q Rn(n+1)/2−Sn 1 − q (2n+1)S .
(q ; q )∞ (q ; q )∞ (q ; q )∞ n=0

Thus the question of Andrews and Merca is equivalent to showing that for n ≥ 1,
 
(−1)k pk (n; o) − pk (n; e) ≥ 0. (2.6)

We prove this by constructing an injection from Pk (n; o) to Pk (n; e) for k odd, and an
injection from Pk (n; e) to Pk (n; o) for k even.

2.4. The odd k case

Let (λ1 , λ2 , μ1 , μ2 ) ∈ Pk (n, o) and d = (μ1 ) − (μ2 ). Since n ≥ 1, at least one of these
partitions cannot be empty.

Case 1: (λ1 , λ2 , μ1 , μ2 ) = (∅, λ2 , ∅, μ2 ). Since d is odd, μ1 and μ2 cannot be both empty.

Also, since k is odd and μ1 is empty, −k + 2 ≤ d ≤ −1.
8 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

(1) If l(λ2 ) > l(μ2 ), then we move l(λ2 ) to μ2 ;

(2) if l(λ2 ) ≤ l(μ2 ), then we move l(μ2 ) to λ2 .
In either case, after the moving, the number of parts of μ2 increases or decreases
by 1, so d becomes even and −k + 1 ≤ d ≤ 0. Therefore, the resulting partition
quadruple belongs to Pk (n, e). In addition, moving the largest part defines an
injection since the inverse can be defined by comparing the largest parts of λ2
and μ2 .
Case 2: λ1 and μ1 are not both empty. Since k is odd, −k + 2 ≤ d ≤ k.
(1) If l(λ1 ) > l(μ1 ), then we move l(λ1 ) to μ1 ;
(2) if l(λ1 ) ≤ l(μ1 ), then we move l(μ1 ) to λ1 .
In either case, after the moving, the number of parts of μ1 increases or decreases
by 1, so d becomes even and −k + 1 ≤ d ≤ k + 1. That is,
(1) l(λ1 ) > l(μ1 ) with −k+2 ≤ d ≤ k =⇒ l(λ1 ) ≤ l(μ1 ) with −k+3 ≤ d ≤ k+1;
(2) l(λ1 ) ≤ l(μ1 ) with −k+2 ≤ d ≤ k =⇒ l(λ1 ) > l(μ1 ) with −k+1 ≤ d ≤ k−1.
Therefore, the resulting partition quadruple belongs to Pk (n, e) unless d = k +1.
We now modify our map for the d = k +1 case. This happens when l(λ1 ) > l(μ1 )
and d = k before the moving. As seen above, no (λ1 , λ2 , μ1 , μ2 ) with l(λ1 ) ≤
l(μ1 ) with d = −k + 1 has a pre-image. So, we modify our map by constructing
an injection from (λ1 , λ2 , μ1 , μ2 ) with l(λ1 ) > l(μ1 ) and d = k to (λ1 , λ2 , μ1 , μ2 )
with l(λ1 ) ≤ l(μ1 ) and d = −k + 1.
Suppose that l(λ1 ) > l(μ1 ) and d = k. We apply the inverse ϕ−1 of Wright’s
map to (μ1 , μ2 ) to obtain π and Δ. Note that l(π) = l(μ1 ) − (k − 1)R − S. So
   
l λ1 − kR − S ≥ l μ1 − (k − 1)R − S = l(π). (2.7)

We now take all the parts of λ1 greater than l(λ1 ) − kR, subtract S from each of
these parts, and place them to the top of the Ferrers diagram of π in decreasing
order from top to bottom. It follows from (2.7) that the resulting diagram is a
partition. We call the resulting partition π̄. We now remove the smallest part
of Δ and subtract 2S from each of the remaining parts of Δ to make its parts
congruent to −S mod R, and then put the resulting partition Δ̄ on the top of
the modular diagram of π and apply Wright’s map ϕ to obtain (μ1 , μ2 ). It is
trivial that μ1 and μ2 have parts congruent to S and −S mod R, respectively.
We claim that (μ1 ) − (μ2 ) = −k + 1 and l(λ1 ) ≤ l(μ1 ). First of all, since the
smallest part of Δ is removed, the number of parts of Δ is reduced by 1, so it is
clear that (μ1 ) − (μ2 ) = −k + 1. To show that l(λ1 ) ≤ l(μ1 ), we first note that
   
l μ1 = a − kR ≥ l λ1 ,

where a is the largest part of the old λ1 and the inequality follows since we
moved all the parts of the old λ1 greater than a − kR to π. Finally, we add to λ1
as parts the removed S’s while moving the parts to π. Therefore, the resulting
 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14 9

quadruple (λ1 , λ2 , μ1 , μ2 ) falls into Case 1 with l(λ1 ) ≤ l(μ1 ) and d = −k + 1.

In addition, since k is odd, (λ1 , λ2 , μ1 , μ2 ) belongs to Pk (n, e).
We now show that the map is an injection. From the construction, it can be
easily seen that the map is injective except for the case when l(λ1 ) > l(μ1 ) and
d = k. So it suffices to verify this exceptional case only. Even in this case, each
step but one can be easily shown to be invertible. The step that is uncertain is
whether we can identify the parts moved from λ1 to π. Indeed, we can because
(i) we moved all the parts of λ1 that are greater than l(λ1 ) − kR, so we move
all the parts of π greater than l(π) − kR;
(ii) and by the inequality in (2.7), the part of π corresponding to l(μ1 ) will not
be moved.

We illustrate the injection for the case when d = k with an example.

Example 1. Let R = 5, S = 1, and k = 3. Let

λ1 = (7R + S, 5R + S, 5R + S, 4R + S, 4R + S, R + S)
λ2 = ∅
μ1 = (5R + S, 4R + S, 3R + S, R + S, S)
μ2 = (3R − S, R − S).

Then |λ1 | + |λ2 | + |μ1 | + |μ2 | = 43R + 9S, d = (μ1 ) − (μ2 ) = 3 and l(λ1 ) = 7R + S >
5R + S = l(μ1 ). Following the steps, we

(1) combine μ1 and μ2 to get π and Δ, namely

π = (3R, 3R, 3R, 2R, 2R, R) and Δ = (2R + S, R + S, S); (2.8)

(2) move the parts 7R + S, 5R + S, 5R + S that are greater than (7 − 3)R + S to π after
subtracting S from each to obtain

π̄ = (7R, 5R, 5R, 3R, 3R, 3R, 2R, 2R, R);

(3) remove the smallest part of Δ and subtract 2S from each of the remaining parts of
Δ to obtain

Δ̄ = (2R − S, R − S);

(4) combine this Δ̄ with π̄ to obtain

μ1 = (4R + S, R + S, S) and μ2 = (11R − S, 9R − S, 6R − S, 2R − S, R − S);

10 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

(5) add the removed 8 copies of S to the remaining parts of λ1 , namely

λ1 = (4R + S, 4R + S, R + S, S, S, S, S, S, S, S, S).

Then we can easily check that |λ1 | + |λ2 | + |μ1 | + |μ2 | = 43R + 9S, d = (μ1 ) − (μ2 ) =
−2 = −k + 1, and l(λ1 ) = 4R + S = l(μ1 ).
Conversely, with μ1 = (4R + S, R + S, S) and μ2 = (11R − S, 9R − S, 6R − S, 2R −
S, R − S), we easily obtain

π̄ = (7R, 5R, 5R, 3R, 3R, 3R, 2R, 2R, R) and Δ̄ = (2R − S, R − S).

Since 5R > 7R − 3R, by removing the parts 7R, 5R, 5R of π̄, we obtain

π = (3R, 3R, 3R, 2R, 2R, R),

which matches the partition π in (2.8).

2.5. The even k case

The injection defined for the odd k case works for the even k case, too. The only
difference is we apply the injection from Pk (n; e) to Pk (n; o). Thus we omit the details.

3. The generating function of the fixed points

In this section, we provide the generating function of the fixed points of our injection
defined in Section 2. We will consider the odd k case only.
Before we proceed, we recall the partition interpretation of the q-binomial coefficients
m
and the Euler expansion formula for 1/(zq; q)∞ . It is known that k q is the generating
function of partitions into at most k parts less than or equal to m − k and 1/(zq; q)∞
is the generating function of partitions in which the exponent of z keeps track of the
number of parts in the partitions (see in [2]).
From the construction of the injection, two sets of fixed points arise. First of all, in
Case 1: (λ1 , λ2 , μ1 , μ2 ) = (∅, λ2 , ∅, μ2 ), the difference d is equal to −(μ2 ) and after the
injection, d changes as follows:

−k + 2 ≤ d ≤ −1 =⇒ −k + 1 ≤ d ≤ 0.

We note that (∅, λ2 , ∅, μ2 ) has mo pre-image when d = 0 with l(λ2 ) ≤ l(μ2 ) or d = −k +1

with l(λ2 ) > l(μ2 ). If d = 0 with l(λ2 ) ≤ l(μ2 ), then λ2 = μ2 = ∅. For d = −k + 1 with
l(λ2 ) > l(μ2 ), let l(λ2 ) = Rm + (R − S) for some m ≥ 0. Then μ2 has exactly k − 1
distinct parts congruent to −S mod R and l(μ2 ) < Rm + (R − S). Such λ2 and μ2 are
generated by
 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14 11

rm+1 /s
(r/s; r)m+1

and

2  
r(k−1) /2 m
.
sk−1 k−1 r

Here and throughout this section, we replace q R and q S by r and s for brevity.
Thus the generating function of those fixed points is

∞ 2  
rm+1+(k−1) /2 /sk m
1+ , (3.1)
m=0
(r/s; r)m+1 k−1 r

where 1 accounts for (∅, ∅, ∅, ∅).

We now consider the second case: λ1 or μ1 is not empty. The injection changes d as
follows:

(1) l(λ1 ) > l(μ1 ) with −k + 2 ≤ d ≤ k =⇒ l(λ1 ) ≤ l(μ1 ) with −k + 3 ≤ d ≤ k + 1;

(2) l(λ1 ) ≤ l(μ1 ) with −k + 2 ≤ d ≤ k =⇒ l(λ1 ) > l(μ1 ) with −k + 1 ≤ d ≤ k − 1.

We recall that

• (λ1 , λ2 , μ1 , μ2 ) ∈ Pk (m; o) with l(λ1 ) > l(μ1 ) and d = k has no image, and
• (λ1 , λ2 , μ1 , μ2 ) ∈ Pk (m; e) with l(λ1 ) ≤ l(μ1 ) and d = −k + 1 has no pre-image.

Thus, instead of trying to characterize the fixed points, we will proceed by taking the
difference of the generating functions of (λ1 , λ2 , μ1 , μ2 ) with l(λ1 ) > l(μ1 ) and d = k,
and (λ1 , λ2 , μ1 , μ2 ) with l(λ1 ) ≤ l(μ1 ) and d = −k + 1.

• When d = k, let l(λ1 ) = Rm + S for some m ≥ 0. Then, l(μ1 ) < Rm + S and

l(π) = l(μ1 ) − (k − 1)R − S ≤ R(m − k), where π is the partition resulting from the
application of ϕ−1 to μ1 and μ2 . Thus the generating function of such λ1 is

rm s
,
(s; r)m+1

and the generating function of such μ1 and μ2 , equivalently Δ and π is

rk(k−1)/2 sk
.
(r; r)m−k
12 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

Therefore, the generating function of these (λ1 , λ2 , μ1 , μ2 ) is

∞
1 rm+k(k−1)/2 sk+1
.
(r/s; r) ∞ m=0 (s; r)m+1 (r; r)m−k

• When d = −k + 1, let l(μ1 ) = Rm + S for some m ≥ 0. Then, l(λ1 ) ≤ Rm + S and

l(π) = l(μ1 ) + (k − 1)R + R − S = R(m + k). Thus the generating function of such
λ1 is

1
,
(s; r)m+1

and the generating function of such μ1 and μ2 , equivalently Δ and π is

r(m+k)+k(k−1)/2 /sk−1
.
(r; r)m−k

Therefore, the generating function of these (λ1 , λ2 , μ1 , μ2 ) is

∞
1 r(m+k)+k(k−1)/2 /sk−1
.
(r/s; r) ∞ m=0 (s; r)m+1 (r; r)m+k

Now combining these generating functions, we obtain the generating function of the
fixed point arising in the second case:

∞ ∞
1 r(n+k)+k(k−1)/2 /sk−1 1 rm+k(k−1)/2 sk+1
−
(r/s; r) ∞ m=0 (s; r)m+1 (r; r)m+k (r/s; r) ∞ m=0 (s; r)m+1 (r; r)m−k
∞ ∞

1  r(m+k)+k(k−1)/2 sk−1  r(m+k)+k(k−1)/2 sk+1
= −
(r/s; r) ∞ m=0 (s; r)m+1 (r; r)m+k m=0
(s; r)m+k+1 (r; r)m

∞  
1 r(m+k)+k(k−1)/2 /sk−1 1 s2k
= − m+1
(r/s; r) ∞ m=0 (s; r)m+1 (r; r)m (r(m+1) ; r)k (r s; r)k

∞ ∞  
1 r(m+k)+k(k−1)/2 /sk−1  (m+1)j   j+k−1
= r 1−s(j+2k)
(r/s; r) ∞ m=0 (s; r)m+1 (r; r)m k−1 r
j=0

∞ ∞  
1 r(m+k)+k(k−1)/2 /sk−1  (m+1)j j + k − 1 1 − sj+2k
= r . (3.2)
(r/s; r) ∞ m=0 (rs; r)m (r; r)m k−1 r 1−s
j=0

Therefore, it follows from (3.1) and (3.2) that the generating function of the fixed
points is
 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14 13

∞ 2  
rm+1+(k−1) /2 /sk m
1+
m=0
(r/s; r)m+1 k−1 r

∞ ∞  
1 r(m+k)+k(k−1)/2 /sk−1  (m+1)j j + k − 1 1 − sj+2k
+ r ,
(r/s; r) ∞ m=0 (rs; r)m (r; r)m k−1 r 1−s
j=0

where (1 − sj+2k )/(1 − s) is a polynomial with positive coefficients.

4. Proof of Theorem 1.2

We now consider the following: for k > 0,

∞
     
p n − k(3k + 1)/2 − p n − k(3k + 5)/2 − 1 q n
n=1

q k(3k+1)/2 (1 − q 2k+1 )
=
(q; q)∞
q k(3k+1)/2 1 q 3k(k+1)/2 1
= k+1
+
(q; q)k−1 (q ; q)∞ (q; q)k (q k+2 ; q)∞
∞ ∞
q k(3k+1)/2+(k+1)n  q 3k(k+1)/2 q (k+2)n
= +
n=0
(q; q)k−1 (q; q)n n=0
(q; q)k (q; q)n

∞    ∞  
q k(k−1)/2+(k+1)(n+k) n + k − 1 q k(k−1)/2 q (k+2)(n+k) n + k
= +
n=0
(q; q)n+k−1 k−1 n=0
(q; q)n+k k

∞    ∞  
q k(k−1)/2+(k+1)(n+k) n + k − 1 q k(k−1)/2 q (k+2)(n+k) n + k − 1
= +
n=0
(q; q)n+k−1 k−1 n=0
(q; q)n+k k−1

∞  
q k(k+1)/2 q (k+2)(n+k) n + k − 1
+
n=1
(q; q)n+k k

∞    ∞  
q k(k−1)/2+(k+1)(n+k) n + k − 1 q k(k+1)/2 q (k+2)(n+k) n + k − 1
= +
n=0
(q; q)n+k k−1 n=1
(q; q)n+k k

∞    ∞  
q k(k−1)/2+(k+1)(n+k) n + k − 1 q k(k+1)/2 q (k+2)(n+k+1) n + k
= +
n=0
(q; q)n+k k−1 n=0
(q; q)n+k+1 k
∞
  
= Mk (n) + Mk+1 (n) q n ,
n=1

from which we obtain

   
p n − k(3k + 1)/2 − p n − k(3k + 5)/2 − 1 = Mk (n) + Mk+1 (n).
14 A.J. Yee / Journal of Combinatorial Theory, Series A 130 (2015) 1–14

Therefore, Theorem 1.2 follows. We note that each equality above can be obtained com-
binatorially. This yields a combinatorial proof of the theorem.

References

[1] K. Alladi, A. Berkovich, New polynomial analogues of Jacobi’s triple product and Lebesgue’s iden-
tities, Adv. in Appl. Math. 32 (2004) 801–824.
[2] G.E. Andrews, The Theory of Partitions, Addison–Wesley, Reading, MA, 1976; reissued: Cambridge
University Press, Cambridge, 1998.
[3] G.E. Andrews, Euler’s “Exemplum memorabile inductionis fallacis” and q-trinomial coefficients,
J. Amer. Math. Soc. 3 (1990) 653–669.
[4] G.E. Andrews, M. Merca, The truncated pentagonal number theorem, J. Combin. Theory Ser. A
119 (2012) 1639–1643.
[5] V.J.W. Guo, J. Zeng, Two truncated identities of Gauss, J. Combin. Theory Ser. A 120 (2013)
700–707.
[6] L.W. Kolitsch, M. Burnette, Interpreting the truncated pentagonal number theorem using partition
pairs, preprint.
[7] R. Mao, Proofs of two conjectures on truncated series, preprint.
[8] I. Schur, Ein Beitrag zur additiven Zahlentheorie und zur theorie der Kettenbrüche, S.-B. Preuss.
Akad. Wiss. Phys.-Math. Kl., 1917, pp. 302–321; reprinted in: Gesammelte Abhandlungen, vol. 2,
Springer, Berlin, 1973, pp. 117–136.
[9] D. Shanks, A short proof of an identity of Euler, Proc. Amer. Math. Soc. 2 (1951) 747–749.
[10] S.O. Warnaar, The generalized Borwein conjecture. II. Refined q-trinomial coefficients, Discrete
Math. 272 (2003) 215–258.
[11] E.M. Wright, An enumerative proof of an identity of Jacobi, J. Lond. Math. Soc. 40 (1965) 55–57.