I recently stumbled across the two seemingly similar identities $$ \prod_{i\geq 1}\frac{1}{1-xq^i}=\sum_{n\geq 0}\frac{x^nq^n}{(1-q)(1-q^2)\cdots(1-q^n)} $$ and $$ \prod_{i\geq 1}(1+xq^i)=\sum_{n\geq 0}\frac{x^nq^{\binom{n+1}{2}}}{(1-q)(1-q^2)\cdots(1-q^n)}. $$

Out of curiosity, is there some combinatorial interpretation for these identities, to understand why they hold? Thanks.

If this would better be asked as two questions, I am glad to split it into two.


The LHS of the first identity is a generating function $$\prod_{i \ge 1} \frac{1}{1 - xq^i} = \sum p_{m,n} q^m x^n$$

where $p_{m,n}$ counts the number of ways to partition the number $m$ into a sum of $n$ positive integers. In other words, $p_{m,n}$ counts the number of Ferrers diagrams with $m$ dots and $n$ rows.

For fixed $n$, given such a Ferrers diagram slice off the leftmost column. The remaining columns form a partition into parts of size at most $n$, and such partitions have generating function $\frac{1}{(1 - q)...(1 - q^n)}$. The leftmost column contributes a factor of $q^n$, and the fact that we started with a Ferrers diagram with $n$ rows contributes a factor of $x^n$. This gives the RHS of the first identity.

The LHS of the second identity counts the number of ways to partition the number $m$ into a sum of $n$ distinct positive integers. To get the RHS, instead of slicing off the leftmost column of the corresponding Ferrers diagram, you can slice off a right triangle with side lengths $n$ and $n$ because the number of dots in each row are distinct (draw a diagram to convince yourself of this). This triangle has ${n+1 \choose 2}$ dots and the rest of the argument is the same as above.

  • $\begingroup$ +1: I knew of the first interpretation, but the second was new to me. Any reference? Sagan or Stanely, perhaps? $\endgroup$ – Alex Youcis Dec 18 '11 at 22:33
  • $\begingroup$ I wouldn't be surprised if the second was in Stanley somewhere but I think I first saw it in a course that didn't work from a textbook. $\endgroup$ – Qiaochu Yuan Dec 18 '11 at 22:53

I'll somewhat expound the Qiaochu Yuan's wonderful solution, and show that how it can be translated to algebraic expressions.

Second identity

We start from the second identity \begin{align} \prod_{i = 1}^\infty (1 + q^i x) = \sum_{n = 0}^\infty p^F_{n} x^n. \tag{1} \end{align}

Expanding the left hand side and comparing the coefficient of $x^n$, we find $$ \sum_{1 \le a_1 < \dots < a_n} q^{a_1 + \cdots + a_n} = p^F_n, \tag{2} $$ where the sum is carried over all the combinations of $a_1, \dots, a_n$, satisfying the condition $1 \le a_1 < \dots < a_n$.

Each term in the sum for $p^F_n$ can be represented by a Ferrers diagram, in which, there are $n$ rows, each with a different number $a_k$ ($1 \le k \le n)$ of dots. If the rows are labelled from bottom to top, a row with fewer dots lies below a row with more dots.

Now consider the change of variables \begin{align} b_k = a_k - a_{k - 1}, \end{align} with $a_0\equiv 0$, then the condition of $a_k > a_{k-1}$ becomes $$ b_k \ge 1, $$ for any $k = 1, \dots, n$. The inverse relation is \begin{align} a_1 &= b_1, \\ a_2 &= a_1 + b_2 = b_1 + b_2, \\ a_3 &= a_2 + b_3 = b_1 + b_2 + b_3, \\ \cdots \\ a_n &= a_{n - 1} + b_n = b_1 + b_2 + \cdots + b_n. \end{align} The sum on the exponent of the left hand side of (2) becomes $$ a_1 + \cdots + a_n = n \, b_1 + (n-1) \, b_2 + \cdots + b_n. $$ and (2) becomes \begin{align} p^F_n &= \sum_{b_1 \ge 1, \; b_2 \ge 1, \; \cdots, \; b_n \ge 1} q^{n \, b_1 + (n-1) \, b_2 + \cdots + b_n} \\ &= \sum_{b_1 \ge 1} q^{n \, b_1} \sum_{b_2 \ge 1} q^{(n - 1) \, b_2} \cdots \sum_{b_n \ge 1} q^{b_n} \\ &= \frac{q^n}{1-q^n} \frac{q^{n-1}}{1-q^{n-1}} \cdots \frac{q}{1-q} \\ &= \frac{ q^{n(n+1)/2} } {(1-q) (1-q^2) \cdots (1-q^n) }. \end{align}

This is the desired result.

First identity

The case for the first identity is similar \begin{align} \prod_{i = 1}^\infty \left[1 + q^i x + (q^i x)(q^i x) + (q^i x)(q^i x)(q^i x) + \cdots \right] = \sum_{n = 0}^\infty p^B_{n} x^n. \tag{3} \end{align}

Comparing the coefficient of $x^n$, we find $$ \sum_{1 \le a_1 \le \dots \le a_n} q^{a_1 + \cdots + a_n} = p^B_n, \tag{4} $$ Note that “$<$” is replaced by “$\le$” because of the $(q^i x)(q^i x), (q^i x)(q^i x)(q^i x), \dots$ terms. In terms of the Ferrers diagram, we now allow multiple rows with the same number of dots. For example, if we choose the term $(q^i x) (q^i x) (q^i x)$ for the $i$th factor, then there are three rows with $i$ dots in the Ferrer diagram.

Then, following the same procedure of changing variables, we now get the condition $$ b_k \ge 0, $$ for $k > 1$, and $b_1 \ge 1$. and \begin{align} p^B_n &= \sum_{b_1 \ge 1, \; b_2 \ge 0, \; \cdots, \; b_n \ge 0} q^{n \, b_1 + (n-1) \, b_2 + \cdots + b_n} \\ &= \sum_{b_1 \ge 1} q^{n \, b_1} \sum_{b_2 \ge 0} q^{(n - 1) \, b_2} \cdots \sum_{b_n \ge 0} q^{b_n} \\ &= \frac{q^n}{1-q^n} \frac{1}{1-q^{n-1}} \cdots \frac{1}{1-q} \\ &= \frac{ q^n } {(1-q) (1-q^2) \cdots (1-q^n) }. \end{align} Q.E.D.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.