The equation,

$$ (ax_1^2+by_1^2)(ax_2^2+by_2^2) = ax_0^2+by_0^2\tag1$$

has the well-known solution when $a=b=1$,

$$ (x_1^2+y_1^2)(x_2^2+y_2^2) = (x_1 y_2 + x_2 y_1)^2 + (x_1 x_2 - y_1 y_2)^2$$

Hence the product of the sum of two squares is itself the sum of two squares. If we use one more factor, then it has an identity for general $a,b$, namely,

$$ (ax_1^2+by_1^2)(ax_2^2+by_2^2)(ax_3^2+by_3^2) = ax_0^2+by_0^2\tag2$$


$$x_0 =a \color{blue}{x_1 x_2 x_3} + b \big(-\color{blue}{x_1} y_2 y_3 + \color{blue}{x_2} y_1 y_3 + \color{blue}{x_3} y_1 y_2 \big)$$

$$y_0 =a \big(-x_2 x_3 \color{brown}{y_1} + x_1 x_3 \color{brown}{y_2} + x_1 x_2 \color{brown}{y_3}\big) + b \color{brown}{y_1 y_2 y_3}$$

High-lighted this way, one can immediately see the pattern.

Q: Is there a similar identity for ternary quadratic forms,

$$ (ax_1^2+by_1^2+cz_1^2)(ax_2^2+by_2^2+cz_2^2)(ax_3^2+by_3^2+cz_3^2) = ax_0^2+by_0^2+cz_0^2\tag3$$

such that $x_0, y_0, z_0$ are integer functions in terms of the other $x_i, y_i, z_i$ just like for $(2)$?

  • $\begingroup$ Well, first you have to solve the equation. $$(a_1x_1^2+b_1y_1^2+c_1z_1^2)(a_2x_2^2+b_2y_2^2+c_2z_2^2)=ax^2+by^2+cz^2$$ Then you can move on to a more complicated formula. $\endgroup$ – individ Dec 11 '15 at 14:22
  • $\begingroup$ @individ: The problem with that approach is that there might be no solution. For example, the equation $(1)$, namely, $$(ax_1^2+by_1^2)(ax_2^2+by_2^2) = ax_0^2+by_0^2$$ has no known solution for general $a,b$, but becomes solvable when a third factor is used like in $(2)$. $\endgroup$ – Tito Piezas III Dec 11 '15 at 14:37
  • $\begingroup$ Isn't it a known theorem that there are such formulas for 2,4,8, and 16 terms, but no others? I can't remember where I saw that… $\endgroup$ – Kieren MacMillan Aug 6 '16 at 11:54
  • 1
    $\begingroup$ @KierenMacMillan:That is for $n$-nary quadratic forms with unit coefficients for $=2,4,8$ (The case $n=16$ and higher already involves rational functions.) This post asks for more general coefficients. $\endgroup$ – Tito Piezas III Aug 6 '16 at 15:36
  • $\begingroup$ In (2) and what follows, are there $\pm$ options like the two different options in (1) and what follows? $\endgroup$ – Kieren MacMillan Oct 16 '16 at 0:20

No, there cannot be a completely general identity. There cannot even be such an identity for $(a,b,c)=(1,1,1)$, since $$ (1+0+0)(1+1+1)(0+1+4) = 15 $$ and by Legendre's three-square theorem $15$ cannot be written as the sum of three squares.

Since we can add factors of $(1+0+0)$ without changing the product, it is clear that additional factors cannot resolve the problem in this case.

  • $\begingroup$ Hm, that is rather deflating. I guess the next case would be, $$\prod_{i=1}^n (ax_i^2+by_i^2+cz_i^2+dt_i^2) = ax_0^2+by_0^2+cz_0^2+dt_0^2$$ for either $n=4$ or $5$. $\endgroup$ – Tito Piezas III Dec 16 '15 at 5:03
  • $\begingroup$ @TitoPiezasIII, works for $n=4$ because we get quaternion multiplication with $w^2 + a x^2 + b y^2 + ab z^2.$ Pretty sure $w^2 + ac x^2 + bc y^2 + ab z^2$ also works. In both cases it is a multiplication of pairs of coefficient quadruples, $x_0 \, 1 + i x_1 \sqrt a + j x_2 \sqrt b + k x_3 \sqrt {ab}$ $\endgroup$ – Will Jagy Dec 17 '15 at 18:04

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.