Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Ramanujan's sum of cubes identity is defined by the generating functions,

$$\begin{aligned} \sum_{n=0}^\infty a_n x^n &= \frac{1+53x+9x^2}{R_1}\\ \sum_{n=0}^\infty b_n x^n &= \frac{2-26x-12x^2}{R_1}\\ \sum_{n=0}^\infty c_n x^n &= \frac{2+8x-10x^2}{R_1}\\ d_n &= (-1)^n \end{aligned}\tag{1}$$

(correcting a small typo in Mathworld), where $R_1 = 1-82x-82x^2+x^3$, then,

$$a_n^3+b_n^3 = c_n^3 + d_n^3\tag{2}$$

It turns out the $a_n, b_n, c_n, d_n$ can also be expressed as,

$$\begin{aligned} a_n &=-9p^2+176pq-851q^2 = 1,135,11151,\dots\\ b_n &=4(3p^2-56pq+263q^2) = 2,138,11468,\dots\\ c_n &=2(5p^2-90pq+409q^2) = 2,172,14258,\dots\\ d_n &= -(p^2-85q^2) = 1, -1,1,-1\dots \end{aligned}\tag{3}$$

and {$p,q$} are chosen to satisfy the Pell equation $p^2-85q^2 =\mp 1$. (Actually, $p,q$ are half-integers since one can use $p^2-85q^2 =\mp 4$.) Ramanujan missed using {$p,-q$} which yields the different,

$$\begin{aligned} \sum_{n=0}^\infty a_n x^n &= \frac{9+53x+x^2}{R_1}\\ \sum_{n=0}^\infty b_n x^n &= \frac{-12-26x+2x^2}{R_1}\\ \sum_{n=0}^\infty c_n x^n &= \frac{-10+8x+2x^2}{R_1} \end{aligned}\tag{4}$$

with the same $R_1$. But there are an infinite number of quadratic parametrizations to (2) given by,

$$(ax^2-v_1xy+bwy^2)^3 + (bx^2+v_1xy+awy^2)^3 + (cx^2+v_2xy+dwy^2)^3 + (dx^2-v_2xy+cwy^2)^3 = (a^3+b^3+c^3+d^3)(x^2+wy^2)^3\tag{5}$$

and {$v_1,\, v_2,\, w$} = {$c^2-d^2,\; a^2-b^2,\; (a+b)(c+d)$}. For example, using {$a,b,c,d$} = {$6,-9,8,1$}, and tweaking (5) with a simple linear transformation so the last term transforms from $x^2-45xy+216y^2$ to $p^2-321q^2$ yields,

$$\begin{aligned} a_n &=3(3p^2-104pq+909q^2) = 3753, 693875529,\dots\\ b_n &=-2(4p^2-135pq+1119q^2) = 4528, 837313192,\dots\\ c_n &=6(p^2-37pq+348q^2) = 5262, 972979926,\dots\\ d_n &=p^2-321q^2 = 1,1,1,\dots \end{aligned}\tag{6}$$

and {$p,q$} chosen to satisfy $p^2-321q^2 = 1$. With a few more terms, Mathematica's GeneratingFunction command was able to find,

$$\begin{aligned} \sum_{n=0}^\infty a_n x^n &= \frac{-9(417-5602x+x^2)}{R_2}\\ \sum_{n=0}^\infty b_n x^n &= \frac{8(-566-11315x+x^2)}{R_2}\\ \sum_{n=0}^\infty c_n x^n &= \frac{-6(877+6898x+x^2)}{R_2} \end{aligned}\tag{7}$$

where $R_2 = -1+184899x-184899x^2+x^3$ and,

$$a_n^3+b_n^3 = c_n^3 + 1$$

thus a sum of cubes identity analogous to Ramanujan's. (Using {$p,-q$} will yield the second family.)

Question: Starting with an initial {$a,b,c,d$}, especially the form {$a,b,c,\pm1$}, and using the quadratic parametrization (5) where one term has been equated to $\pm1$, is it always possible to find a generating function analogous to (1)? (I've tried various {$a,b,c,d$} and it seems to be always the case, but a general proof eludes me.) See also Rowlands' survey article here.

share|cite|improve this question
up vote 3 down vote accepted

Yes, this is true. As in your example, you need to start with integers $x,y$ such that the last term of the LHS of (5) is $dx^2+v_2xy+cwy^2=\pm1$. Make a change of variables to eliminate the $xy$ term (i.e., "complete the square"), which yields an equation of the form $p^2-Dq^2=1$. When $D$ is a positive nonsquare, this says that the element $p+q\sqrt{D}$ of $\mathbb{Z}[\sqrt{D}]$ has norm $1$. But it's classical to write down all elements of norm $1$: they are $\pm u^n$ where $n\in\mathbb{Z}$ and $u$ is the smallest element of $\mathbb{Z}[\sqrt{D}]$ such that $N(u)=1$ and $u>1$. Say $$ p+q\sqrt{D} = u^n $$ (the case of $-u^n$ works exactly the same way). Then $p-q\sqrt{D} = 1/u^n$, so $$ p=\frac{u^n+u^{-n}}2 \quad\text{ and }\quad q=\frac{u^n-u^{-n}}{2\sqrt{D}}. $$ Let's just consider nonnegative values of $n$, and write $p_n$ and $q_n$ for the above expressions. Then the corresponding values of $x$ and $y$ (call them $x_n$ and $y_n$) are linear combinations of $p_n$ and $q_n$, and hence of $u^n$ and $u^{-n}$. Writing, for instance, $a_n=ax_n^2−v_1x_ny_n+bwy_n^2$ for the value of the first number you're cubing, it follows that $a_n$ is a linear combination of $u^{2n}$, $1$, and $u^{-2n}$. Therefore $$ \sum_{n=0}^\infty a_n x^n $$ is a linear combination of $$ \sum_{n=0}^\infty u^{2n} x^n,\quad \sum_{n=0}^\infty x^n, \quad\text{and}\quad\sum_{n=0}^\infty u^{-2n} x^n, $$ or equivalently of $$ \frac{1}{1-u^2 x},\quad \frac{1}{1-x},\quad\text{and}\quad\frac{1}{1-u^{-2}x}. $$ Thus the generating function for $a_n$ is a rational function, and we could do the same for $b_n$ and $c_n$, in order to get a result similar to Ramanujan's.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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