3

Ramanujan's sum of cubes identity is defined by the generating functions (easily calculated by Wolfram),

\begin{aligned} \sum_{n=0}^\infty a_{n} x^n &= \frac{1+53x+9x^2}{1-82x-82x^2+x^3} = 1,135,11151,\dots\\ \sum_{n=0}^\infty b_{n} x^n &= \frac{2-26x-12x^2}{1-82x-82x^2+x^3} = 2,138,11468,\dots\\ \sum_{n=0}^\infty c_{n} x^n &= \frac{2+8x-10x^2}{1-82x-82x^2+x^3} = 2,172,14258,\dots\\ \end{aligned}

then $a_n^3+b_n^3 = c_n^3 + (-1)^n$. In this 2013 post, I found the similar,

\begin{aligned} \sum_{n=0}^\infty a_n x^n &= \frac{-9(417-5602x+x^2)}{-1+184899x-184899x^2+x^3} = 3753, 693875529,\dots\\ \sum_{n=0}^\infty b_n x^n &= \frac{8(-566-11315x+x^2)}{-1+184899x-184899x^2+x^3} = 4528, 837313192,\dots\\ \sum_{n=0}^\infty c_n x^n &= \frac{-6(877+6898x+x^2)}{-1+184899x-184899x^2+x^3} = 5262, 972979926,\dots \end{aligned}

then $a_n^3+b_n^3 = c_n^3 + 1,$ and stated there should be infinitely many. Seeing this 2020 post which connected the topic to unimodular matrices made me revisit it.


I. Generating Functions

From the well-known identity,

$$(9k^4)^3+(1-9k^3) = (9k^4-3k)^3+1$$

so $(a,b,c) = (9k^4,\; 1-9k^3,\; 9k^4-3k),\;$ define,

\begin{aligned} \sum_{n=0}^\infty a_n x^n &= \frac{-a (x - 1)^2 + 144 k^4 (351 k^6 - 1) (x - 1) + 6^6 k^{10}}{x^3 - 3m x^2 + 3m x - 1}\\[4pt] \sum_{n=0}^\infty b_n x^n &= \frac{-b (x - 1)^2 - 216 k^6 (432 k^6 - 18 k^3 + 5) (x - 1) - 12^3k^6 (54 k^{6} + 1)}{x^3 - 3m x^2 + 3m x - 1}\\[4pt] \sum_{n=0}^\infty c_n x^n &= \frac{-c (x - 1)^2 - 72 k^4 (594 k^6 - 18 k^3 - 1) (x - 1) - 6^6 k^{10}}{x^3 - 3m x^2 + 3m x - 1} \end{aligned}

where $m=(432k^6 - 1)(144k^6 - 1)$ then,

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

and the 2013 post was just the case $k=1$. If $k = -1$, then we get terms of Ramanujan's famous taxicab number $9^3+10^3 = 12^3+1 = 1729$,

\begin{aligned} \sum_{n=0}^\infty a_n x^n &= \frac{-9(417-5602x+x^2)}{-1+184899x-184899x^2+x^3} = \; 3753, 693875529,\dots\\ \sum_{n=0}^\infty b_n x^n &= \frac{-10(-323 + 9826x + x^2)}{-1+184899x-184899x^2+x^3} = -3230, -597125510,\dots\\ \sum_{n=0}^\infty c_n x^n &= \frac{-12(223 + 3664x + x^2)}{-1+184899x-184899x^2+x^3} = \; 2676, 494833692,\dots \end{aligned}

thus, $\color{blue}{3753^3 + (-3230)^3 = 2676^3 + 1}$ and so on.


II. Unimodular Matrices

To be consistent with the above, we start with $n=0$.

\begin{array}{}\begin{bmatrix} a_n \\\ b_n \\\ c_n \end{bmatrix} & = & {\begin{bmatrix} u_1v_1 & 3\times6^3\, k^7v_1 & -24\, k^3w_1 \\\ 6^3k^5u_1 & 3\times6^6\, k^{12} + 1 & - 6^3k^5u_2 \\\ 24k^3w_2 & 3\times6^3\, k^7v_2 & - u_2v_2\end{bmatrix}}^{n+1} \cdot \begin{bmatrix} 9k^4\\\ 1 - 9k^3\\\ 9k^4 - 3k\end{bmatrix}\end{array}

\begin{aligned} u_1 &= (108k^6 + 18k^3 -1)(6k^3 + 1)\\ u_2 &= (108k^6 - 18k^3 -1)(6k^3 - 1)\\ v_1 &= 216k^6 - 36k^3 -1\\ v_2 &= 216k^6 + 36k^3 -1\\ w_1 &= 81k^6(72k^6 - 36k^3 + 5)-2\\ w_2 &= 81k^6(72k^6 + 36k^3 + 5)-2\end{aligned}

which, for $n\geq0$, should yield the same values as the above so,

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

The determinant of the $(3\times3)$ matrix, of course, is $\text{det}=1.$ For example, again let $k = -1$, so $9^3+10^3 = 12^3+1 = 1729$,

\begin{array}{}\begin{bmatrix} a_n \\\ b_n \\\ c_n \end{bmatrix} & = & {\begin{bmatrix} -111695 & -162648 & 219624 \\\ \quad 96120 & \quad 139969 & -189000\quad \\\ -79656& -115992 & 156625\end{bmatrix}}^{n+1} \cdot \begin{bmatrix} 9\\\ 10\\\ 12\end{bmatrix}\end{array}

and for $n=0$, Wolfram gives $(3753, -3230, 2676)$ thus, $\color{blue}{3753^3 + (-3230)^3 = 2676^3 + 1}$ same as above. And so on for other $n$.


III. Question

I found these relations empirically and they apparently hold true. So how do we prove them rigorously, especially the correspondence between the generating functions and the unimodular matrix?

(P.S. Note that the matrix also yields valid solutions for negative exponents $n$.)

  • 1
    I didn't check the details, but overall it seems straightforward :on your right-hand sides you have rational fractions $\frac{P(x)}{Q(x)}$. By the fundamental theorem of algebra, you can factorize $Q(x)=\prod_{k}(x-\lambda_k)$ over $\mathbb C$. Then, by the reduction of rational fraction into sum of simple elements, you can write $\frac{P(x)}{Q(x)}$ as a sum of terms of the form $\frac{A(x)}{(x-\lambda}^j}$ , whose generating function are well-known. [To be continued] – Ewan Delanoy Jan 02 '24 at 20:25
  • 1
    [Continued] Finally, the coefficients of the generating function thus computed can clearly be expressed in terms of the powers of $M$, where $M$ is any matrix with characteristic polynomial equal to $Q$. – Ewan Delanoy Jan 02 '24 at 20:27
  • @EwanDelanoy I see. I had to do a crash course on matrix mathematics. Are there many examples of (3x3) matrices with polynomial entries and integer coefficients such that determinant $\text{det}=1$? – Tito Piezas III Jan 02 '24 at 22:31
  • Sure, there are many. You can start with transvection matrices (matrices who have ones on the diagonal and zero everywhere else except for one place), and remember that integer matrices with det=1 form a group under multiplication. – Ewan Delanoy Jan 03 '24 at 05:13
  • @EwanDelanoy In my crash course, I did came across those. I should have been more specific. Are there well-known (3x3) matrices such that ALL entries are non-zero and non-unity but with determinant $\text{det}=1$? – Tito Piezas III Jan 03 '24 at 05:19
  • Yes, you can find a family of solutions parametrized by 6 integer variables : consider the matrix $A(a_1,a_2,a_3,a_4,a_5,a_6)$ whose columns are $(2,2a_1+1,a_3)$, $(2a_2+1,1+a_1+a_2+2a_1a_2,a_4)$, and $(a_5,a_6,2a_5a_3a_2 + (a_5a_3 - 2a_5a_4))a_1 + ((a_5 - 2a_6)a_3a_2 + ((a_5 - a_6)a_3 + ((-a_5 + 2a_6)a_4 + 1)$. This matrix has determinant $1$, and for most uples, all the entries of $A$ will be non-zero and non-unity. – Ewan Delanoy Jan 03 '24 at 05:29

0 Answers0