If We have $x+y=4 $, $ x\cdot y=-1$ and $x>y$ than $9 x^2+15 y^2=\;\;? $

  • $\begingroup$ We had a similar problem a few days ago and Bill Dubuque suggested using Gauss' algorithm (see math.stackexchange.com/questions/652252/…). I tried and for me it seems to fail for this example, probably I did something wrong. It breaks after second step when only $6y^2$ is left with lex-order $(0,2)$ which means one would have to caculate $6y^2-6(x+y)^{0-2}(xy)^2$. I don't think this leads to the solution here. Otherwise this would make a very good answer. $\endgroup$
    – Piwi
    Feb 1, 2014 at 15:03
  • $\begingroup$ This problem can't have a unique solution. Suppose $x=x_0,y=y_0$ is a solution, then so is $x=y_0,y=x_0$. So, unless $x_0=y_0$, there will be two solutions. $\endgroup$
    – dexter04
    Feb 1, 2014 at 15:09
  • 2
    $\begingroup$ @dexter04 $x > y$, $xy = -1$ ensures a unique solution. $x$ must necessarily be positive, $y$ negative. See my answer for details. $\endgroup$
    – amWhy
    Feb 1, 2014 at 15:11
  • $\begingroup$ oops. sorry i overlooked the fact. $\endgroup$
    – dexter04
    Feb 1, 2014 at 15:12
  • 4
    $\begingroup$ @Piwi: This problem differs from the problem you are refering to. $9 x^2+15 y^2$ is not a symmetric polynomial. $\endgroup$
    – miracle173
    Feb 1, 2014 at 15:53

2 Answers 2


Consider the fact that $9x^2+15y^2 = 12(x^2+y^2)+3(y^2-x^2)$. Now, $x^2+y^2 = (x+y)^2-2xy$ and $y-x = -\sqrt{(x+y)^2-4xy}$. The $-$sign is taken as $y<x$. So, $y^2-x^2 = (y-x)(y+x) $ can be calculated.

  • $\begingroup$ Good answer of an odd Question: $+1$ $\endgroup$ Feb 1, 2014 at 15:26
  • $\begingroup$ @lab This is a special case of an algorithm that works generally - see my answer. $\endgroup$ Feb 1, 2014 at 19:24

It's doable algorithmically by symmetric-antisymmetric decomposition and Gauss's algorithm.

$$\begin{eqnarray} f(x,y) &\,=\,& \overbrace{\frac{1}2 (f(x,y)+f(y,x))}^{\large \rm symmetric\ part} \,+\, \overbrace{\frac{1}2(f(x,y)-f(y,x))}^{\large \rm \color{#c00}{anti}symmetric\ part}\\ &=& 12(x^2+y^2)\, +\, 3(y^2-x^2)\\ &=& 12(\color{#0a0}{\color{#c0a0}{x^2+y^2}})\, -\, 3(\color{#c00}{x-y})(\color{#0a0}{x+y})\end{eqnarray}$$

Then, applying Gauss's algorithm (or inspection) to the $\rm\color{#0a0}{symmetric}$ polynomials and also to the (symmetric!) square of the $\rm\color{#c00}{anti}$symmetric} polynomial, we quickly rewrite these polynomials as polynomials in the elementary symmetric polynomials $\,x+y\,$ and $\,xy,\,$ namely

$$\begin{eqnarray} \color{#0a0}{x^2+y^2} &\,=\,& (x+y)^2 - 2(xy)\\ (\color{#c00}{x-y})^2 &=& (x+y)^2 - 4(xy)\end{eqnarray}$$

Since $\,x-y > 0\,$ we know $\, x-y = \sqrt{(x-y)^2},\,$ so computing these values in terms of the known values of the elementary symmetric polynomials $\,x+y\,$ and $\,xy\,$ then substituting these values into the first equation immediately yields the value of $\,f(x,y).\,$ This method works generally.

  • $\begingroup$ Thanks! I tried to "symmetrize" inutitively but failed gloriously. Small questions considering the generality: (1) What to do if you don't have $y<x$? Start tracking the different cases? (2) Does taking the square of the resulting antisymmetric always make it symmetric? (Sorry, probably easy to see. I don't.) $\endgroup$
    – Piwi
    Feb 1, 2014 at 23:52
  • 2
    $\begingroup$ @Piwi If you know the sign of the antisymmetric part $\,y-x\,$ then you can choose the correct square root to take after computing the value of its symmetric square (else the answer involves an unknown sign). Secondly, if $f$ is antisymmetric, i.e. $\,f(y,x) = -f(x,y),\,$ then squaring that equality shows that $f^2$ is symmetric. $\endgroup$ Feb 2, 2014 at 0:03

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.