MathOverflow is a question and answer site for professional mathematicians. 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

This came up when I did a brand-new (or maybe it's just "birdtracks" in disguise :-) graph-based construction of the E8 family. x,y,z are dimensions and thus integer. x<0 actually doesn't hurt. (For $y=x,z=(3 (-2 + o) o)/(10 + o)$ the standard E8 setup results, o must now divide 360 etc. pp.) So here is the equation:

$3 x (2 + x) (-2 + x + x^2 - 2 y) y (-x + x^2 - 2 z) z=q^2$

With rational y or z I wouldn't pester MO - I'd solve it on the spot. But here some nasty division properties are involved, and I'm lousy in number theory. Is there a finite, easy describable solution list?

share|cite|improve this question
Numerical search suggests there's lots of solutions: any of the LHS factors can be set to zero, or $x=1$ and $q = \pm 6yz$, and quite a bit besides that. – Noam D. Elkies Feb 6 '13 at 20:56
@Peter: No, not necessary. @Noam: Obviously (although: zero dimension...not sure if want)...the "besides" is the interesting part. (I hoped for a general applicable method.) Note that negative y or z is fishy - I'm still pondering if negative dimensions on the RHS of a Clebsch-Gordon expansion make any sense at all. – Hauke Reddmann Feb 7 '13 at 10:00
Well the $x=1$ family wasn't that obvious... (Though maybe it's obvious in the "standard E8 setup".) The usual heuristics suggest that there should be a sparse but infinite set of "random" solutions (if $\max(x,y,z)\in[H,2H)$ there are about $H^3$ numbers of size $H^6$ so we expect about $H^0$ of them to be squares), and a list of solutions up to say $10^3$ might also turn up new some parametric families. – Noam D. Elkies Feb 7 '13 at 16:37
There are still quite a lot besides those. For example, take $x=2$, and any $z$. Then the equation simplifies to $n_z((y-1)^2 -1)=q^2$, which is Pell's equation if $n_z$ is not a square. Choose $z$ so that $6(z^2-z)$ is not a square (another Pell equation, but this time we want non-solutions) and you'll get infinitely many more solutions. I think the same tricks will work with any $x$ value: complete squares, get a Pell-type equation for $z$ in terms of $x$ and choose a non-solution $n_z$, then write the Pell equation for $y$ in terms of that, and solve. – Zack Wolske Feb 7 '13 at 16:52

Here is an algorithm which will generate many families of solutions, provided you are willing to compute solutions to $a^2 - nb^2 = 1$, and will generate all of them if you instead solve $a^2 - nb^2 = k^2$.

Define the functions $F=\frac{x^2-x}{2}$, $G=\frac{x^2+x-2}{2}$ and $H=3x^2+6x$. Then your equation becomes $$ H(2y^2 - 2Fy)(2z^2-2Gz)=q^2 $$ Multiplying both sides by 4, completing the square, and substituting $Y=2y-F$ and $Z=2z-G$, we get
$$ H(Y^2 - F^2)(Z^2 - G^2) = (2q)^2 $$ Choose $n=n_{x,z}=H(Z^2 - G^2)$ so that it is not a square. Then $n$ divides $2q$, so factoring it out and writing $Q=\frac{2q}{n}$, you now want to solve $Y^2 - nQ^2 = F^2$. If you're willing to take just the solutions where $F$ divides $y$ - or the ones where $G$ divides $z$, since their roles are interchangeable - then you are just solving the classical Pell's equation. You can choose $n_{x,z}$ either by listing values for $x$ and $z$, or by taking the squarefree part of $H$ and solving another Pell type equation, but now you want $n$ to not be a solution to the equation (which will happen most of the time).

Noam Elkies's solutions have $F=G=0$, so you don't need to worry about making $n$ non-square, because there are just two terms in the final equation, and you only need to equate the squares.

share|cite|improve this answer

Let $P(x,y,z)=3 x (2 + x) (-2 + x + x^2 - 2 y) y (-x + x^2 - 2 z) z$ be the product which we wish to be a square. As noted,

  • $P(1,y,z)=\left(6yz\right)^2.$

A nice parametric solution is

  • $P(2r,3r,r+1)=\left(12r(r+1)(2r^2-2r-1)\right)^2$

Also, for fixed $x,y$ or $x,z$ we are left with a Pell Equation.

The parametric solution arises from first setting $y=x+z-1$ to make the two quadratic factors equal $P(x,x+z-1,z)=(x^2-x-2z)^23x(x+2)z(x+z-1).$ There are many other parametric solutions such as

  • $P(r^2,r^2+2,3)=\left(3r(r^2+2)^2(r^2-3)\right)^2$ and
  • $P(3r^2-2,3r^2-2,1)=\left(3r(3r^2-4)(3r^2-2)(3r^2-1)\right)^2.$

Although none that I know of with the right-hand side the square of a fourth degree polynomial in $r$.

Most solutions to $P=\Box$ do not have $y=x+z-1$. There are $1442$ choices of $2 \le,x \le 2000,1 \le z \le 2000$ which make $P(x,x+z-1,z)$ a square. Of them $1000$ have $x=2(z-1).$

For fixed $x$ and $z \gt \frac{x^2-x}2$ we have $P(x,y,z)=Ay(y-B)$ for some constants. If $A \gt 0$ has square-free part $\alpha \gt 1$ then we are left with a potentially solvable Pell equations. We see that $\gcd(y,y-B)$ divides $B$ so we seek solutions $$\{y,y-B\}=\{m\alpha_1u^2,m\alpha_2v^2\}$$ where $m\cdot \gcd(\alpha_1,\alpha_2)$ divides $B$ and $\alpha_1\alpha_2=\alpha.$

For example $x,z=3,5$ reduces to having $P(3,y,5)=2\cdot30^2\cdot y\cdot(y-5) $ a square.

There are no solutions to $\{y-5,y\}=\{u^2,2v^2\}$ (in either order) but $\{y-5,y\}=\{5u^2,10v^2\}$ means considering $P(3,5w,5)=2\cdot150^2 \cdot (w-1)\cdot w.$ We need $(w-1)w=2\Box$ with familiar solutions $(w-1)w=1\cdot2,8\cdot 9,49 \cdot 50,288 \cdot 289\cdots.$

share|cite|improve this answer
$Y^2-5$ may never be twice a square, but $y(y-5)$ certainly can be. Which one do you mean? – Gerry Myerson Feb 8 '13 at 4:21
Badly said, but I simply meant that $y,y-5=Y^2,2T^2$ (in one order or the other) has no solutions however $y,y-5=5Y^2,10T^2$ does. – Aaron Meyerowitz Feb 8 '13 at 4:32

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.