10

There are $n$ equations.

I need answer for the case $n=3$.

$$ \frac{1}{x_1}(1-x_1)^2+\frac{1}{x_2}(1-x_2)^2+\cdots+\frac{1}{x_n}(1-x_n)^2=0, $$ and $$ \frac{1}{x_1}(1-x_1^2)^j+\frac{1}{x_2}(1-x_2^2)^j+\cdots+\frac{1}{x_n}(1-x_n^2)^j=0,\; 2\leq j\leq n,$$

where $x_j\geq 1$ or $x_j \leq -1$.

If $n=2,3,4$, computer told me that the only real solution is $(1,1,\cdots,1)$. I did not tell computer the condition $|x_j|\geq 1$.

I do not know if some inequality can be applied like chebyshev's inequality and rearrangement inequality. It seems also a little like the Vandermonde determinant.

Comments: : The following three answers perfectly solve the case $n=3$, two of them drop the condition $|x_j|\geq 1$. The case $n=3$ is enough for me, however, for more interested you can try with larger $n\geq 4$. I am also very curious about that.

cbi
  • 155
  • 3
    $x_j$ may be negative. – cbi Sep 28 '23 at 16:17
  • What is the source of the problem? some inequalities material? – amkpm90 Sep 28 '23 at 16:29
  • It is actually some of the coefficients in a taylor series (several variable), I need some non-vanish property. The first version of these equations is in the form of trigonometric function, $x_j=1/\cos(y_j)$ for $0\leq y_j\leq \pi$. – cbi Sep 28 '23 at 16:32

3 Answers3

7

I need answer for the case $n=3$.

We rename variables and obtain the following system of equations.

$$\cases{ \frac{(1-x)^2}{x}+\frac{(1-y)^2}{y}+\frac{(1-z)^2}{z} & =$\,$ 0\\ \frac{(1-x^2)^2}{x}+\frac{(1-y^2)^2}{y}+\frac{(1-z^2)^2}{z} & = $\,$0\\ \frac{(1-x^2)^3}{x}+\frac{(1-y^2)^3}{y}+\frac{(1-z^2)^3}{z} & = $\,$0} $$

Suppose for a contradiction that $(x,y,z)\ne (1,1,1)$. Then the system

$$\cases{ u+v+w & =$\,$ 0\\ (1+x)^2u+(1+y)^2v+(1+z)^2w & = $\,$0\\ (1+x)^2(1-x^2)u+(1+y)^2(1-y^2)v+(1+z)^2(1-z^2)w& = $\,$0\\} $$

of linear equations has a non-zero solution $(u,v,w)=\left(\frac{(1-x)^2}{x},\frac{(1-y)^2}{y},\frac{(1-z)^2}{z} \right)$.

Thus the system is degenerated, that is $$ 0=\left|\begin{matrix} 1 & 1 & 1\\ (1+x)^2 & (1+y)^2 & (1+z)^2\\ (1+x)^2(1-x^2) & (1+y)^2(1-y^2) & (1+z)^2(1-z^2) \end{matrix}\right|=$$ $$(x-y)(x-z)(y-z)((x^2y+x^2z+xy^2+xz^2+y^2z+yz^2)+2xyz+$$ $$2(x^2+y^2+z^2)+4(xy+xz+yz)+4(x+y+z)+2)=$$ $$(x-y)(x-z)(y-z)((x^2y+x^2z+xy^2+xz^2+y^2z+yz^2)+2xyz+$$ $$2(x+y+z)^2+4(x+y+z)+2).$$

Let $\sigma_1=x+y+z$, $\sigma_2=xy+xz+yz$, and $\sigma_3=xyz$ be the symmetric polynomials on variables $x$, $y$, and $z$. Then $x^2y+x^2z+xy^2+xz^2+y^2z+yz^2=\sigma_1\sigma_2-3\sigma_3$, $x^3+y^3+z^3=\sigma_1^3-3(\sigma_1\sigma_2-3\sigma_3)-6\sigma_3=$ $\sigma_1^3-3\sigma_1\sigma_2+3\sigma_3$, and $\frac 1x+\frac 1y+\frac 1z=\frac{\sigma_2}{\sigma_3}.$

The first two equations of the initial system imply $\frac{\sigma_2}{\sigma_3}-6+\sigma_1=0$ and $$\frac{\sigma_2}{\sigma_3}-2\sigma_1+\sigma_1^3-3\sigma_1\sigma_2+3\sigma_3=0.$$ It follows $$3\sigma_1\sigma_2-3\sigma_3=\frac{\sigma_2}{\sigma_3}-2\sigma_1+\sigma_1^3=6-\sigma_1-2\sigma_1+\sigma_1^3.$$

Suppose first that $x$, $y$, and $z$ are pairwise distinct. Then $$0=(x^2y+x^2z+xy^2+xz^2+y^2z+yz^2)+2xyz+2(x+y+z)^2+4(x+y+z)+2=$$ $$\sigma_1\sigma_2-3\sigma_3+2\sigma_3+2\sigma_1^2+4\sigma_1+2=$$ $$\sigma_1\sigma_2-\sigma_3+2\sigma_1^2+4\sigma_1+2=$$ $$2-\sigma_1+\frac 13\sigma_1^3+2\sigma_1^2+4\sigma_1+2=$$ $$\frac 13\sigma_1^3+2\sigma_1^2+3\sigma_1+4.$$

The only real root of the latter equation is $\sigma_1=-A-A^{-1}-2\approx -4.613$, where $A=\sqrt[3]{5+2\sqrt{6}}$. We have $\sigma_2=(\sigma_1-6)\sigma_3$ and so $$3\sigma_1(\sigma_1-6)\sigma_3-3\sigma_3=3\sigma_1\sigma_2-3\sigma_3=6-3\sigma_1+\sigma_1^3.$$ Thus $$\sigma_3=\frac{6-3\sigma_1+\sigma_1^3}{3\sigma_1(\sigma_1-6)-3}\mbox{ and }\sigma_2=(\sigma_1-6)\frac{6-3\sigma_1+\sigma_1^3}{3\sigma_1(\sigma_1-6)-3}.$$

Put $p(t)=t^3-\sigma_1t^2+\sigma_2t-\sigma_3$. By Vieta's formulae, $x$, $y$, and $z$ are roots of the equation $p(t)=0$. But the discriminant of the polynomial $p(t)$ equals $$\sigma_1^2\sigma_2^2-4\sigma_2^3-4\sigma_1^3\sigma_3-27\sigma_3^2+18\sigma_1\sigma_2\sigma_3\approx -21.702<0,$$ so the polynomial $p(t)$ has only one real root, a contradiction.

Suppose now that $x$, $y$, and $z$ are not pairwise distinct. Permuting $x$, $y$, and $z$, if needed, we can assume that $y=z$. Then $$\cases{ \frac{(1-x)^2}{x}+2\frac{(1-y)^2}{y}& =$\,$ 0\\ \frac{(1-x^2)^2}{x}+2\frac{(1-y^2)^2}{y} & = $\,$0\\ \frac{(1-x^2)^3}{x}+2\frac{(1-y^2)^3}{y} & = $\,$0\\ }.$$

From the first and the second equation we obtain $$(1+x)^2\cdot (-2)\frac{(1-y)^2}{y}+2\frac{(1-y^2)^2}{y}=0$$ $$-(1+x)^2+(1+y)^2=0$$ $$1+x=\pm (1+y)$$

If $x=y$ then $x=y=z$ and so $3\frac{(1-x)^2}{x}=0$ and $(x,y,z)=(1,1,1)$.

If $x\ne y$ then $y=-2-x$.

From the second and the third equation we obtain $$(1-x^2)\cdot (-2)\frac{(1-y^2)^2}{y}+2\frac{(1-y^2)^3}{y}=0.$$

Since $y\ne \pm 1$, we can divide the equality by $2\frac{(1-y^2)^2}{y}$ and obtain $-(1-x^2)+1-y^2=0$, so $x^2=y^2$. Since $x\ne y$, we have $y=-x=-2-x$, a contradiction.

Alex Ravsky
  • 106,166
  • Thanks! I will check it today or tomorrow. – cbi Oct 02 '23 at 16:19
  • 1
    Good, I also tried to compute the degenarate determinant but fail. However, it seems that in this method we have to consider several cases about the variables equal or not separately, which makes it a little intricate as dimension increases (here $n=3$ is enough). – cbi Oct 03 '23 at 03:56
  • It seems that in your answer $|x_j|\geq 1$ is not needed? – cbi Oct 03 '23 at 05:01
  • @cbi Right, the solution highly depends on the particular case of $n$, so I expect there can be problems to generalize it to bigger values of $n$. Also it is right that I never used conditions $|x_i|\ge 1$ in the answer. – Alex Ravsky Oct 03 '23 at 05:53
6

Solution for $n=3$

For $|x|,|y|,|z|\ge1$, we aim to solve the system \begin{align}\frac{(1-x)^2}x+\frac{(1-y)^2}y+\frac{(1-z)^2}z&=0\\\frac{(1-x^2)^2}x+\frac{(1-y^2)^2}y+\frac{(1-z^2)^2}z&=0\\\frac{(1-x^2)^3}x+\frac{(1-y^2)^3}y+\frac{(1-z^2)^3}z&=0.\end{align} After expanding each equation and subtracting appropriately, we find an interesting equivalence \begin{align}x+y+z+\frac1x+\frac1y+\frac1z&=6\\x^3+y^3+z^3+3\left(\frac1x+\frac1y+\frac1z\right)&=12\\x^5+y^5+z^5+5\left(\frac1x+\frac1y+\frac1z\right)&=18\end{align} which we can attack using the elementary symmetric polynomials $e_1=\sum_{\rm cyc}x,e_2=\sum_{\rm cyc}xy,e_3=xyz$. Using Newton's identities, this is equivalent to \begin{align}e_1+\frac{e_2}{e_3}&=6\\e_1^3-3e_1e_2+3e_3+\frac{3e_2}{e_3}&=12\\e_1^5-5e_1^3e_2+5e_1^2e_3+5e_1e_2^2-5e_2e_3+\frac{5e_2}{e_3}&=18.\end{align}

The rest of my original solution which is quite tedious can be found here, and I will illustrate a much simpler solution by @MykolaPochekai:

Letting $e_2=re_3$, we have $e_1=6-r$ and $e_3=\dfrac{(12-3r)-(6-r)^3}{3(r^2-6r+1)}$. Substituting into the third equation yields $$\frac{(r-3)^4(r^3-24r^2+189r-498)}{9(r^2-6r+1)}=0.$$ Note that $|r|\le3$ from $|x|,|y|,|z|\ge1$, and it is easy to show the cubic has no real roots over this interval. Therefore, $e_1=3$, $e_2=3$ and $e_3=1$ and Vieta tell us each of $x,y,z$ solves $t^3-3t^2+3t-1=0$ so $x=y=z=1$ is the unique solution. Note that for completeness, the denominator is zero when $r=3-2\sqrt2$ but this is irrelevant since $12-3r-(6-r)^3\ne0$ and $e_3$ is finite. $\square$

TheSimpliFire
  • 28,020
  • Thanks! I will check it today or tomorrow. – cbi Oct 02 '23 at 16:19
  • Yesterday I used pqr method to prove it. I found it is nothing but $p = e_1, q = e_2, r = e_3$ (already +1 yesterday). Your original solution is quite tedious (using sturm theorem etc.) Actually, these equations are quite simple to eliminate variables (although I do polynomial manipulations by computer), without using MykolaPochekai's idea (seems equivalent). – River Li Oct 03 '23 at 01:17
  • @RiverLi Great, can you share your pqr method? – cbi Oct 03 '23 at 02:14
  • 1
    (Sorry I don't know how to @) I think this answer is right, however, if we know $r=3$ is the only solution in that (most complicated) equation, we get $e_2=3e_3$, which implies $3|xyz|=|xy+yz+xz|\leq |xy|+|yz|+|xz|\leq 3|xyz|$ and thus $|x|=|y|=|z|=1$. – cbi Oct 03 '23 at 02:18
  • By the way, can I give bounty for $\geq 2$ poster? – cbi Oct 03 '23 at 02:23
  • @cbi The pqr method is similar to this solution. – River Li Oct 03 '23 at 02:24
  • +1 TheSimplFire:, the equivalent form was very nice. – Tito Piezas III May 01 '25 at 14:53
  • I mentioned your approach to the system in this new post. – Tito Piezas III May 05 '25 at 15:05
1

Let $x_1+x_2+x_3=3u$, $x_1x_2+x_1x_3+x_2x_3=3v^2,$ where $v^2$ may be negative and $x_1x_2x_3=w^3$.

Thus, $$\sum_{cyc}\frac{(x_1-1)^2}{x_1}=0$$ gives $$u-2+\frac{v^2}{w^3}=0,$$ $$\sum_{cyc}\frac{(x_1^2-1)^2}{x_1}=0$$ gives $$9u^3-9uv^2+w^3-2u+2-u=0$$ and we obtain: $$v^2=\frac{(9u^3-3u+2)(2-u)}{18u-9u^2-1}$$ and $$w^3=\frac{9u^3-3u+2}{18u-9u^2-1}.$$ About the case $u=2$ we'll say later.

Also, $$\sum_{cyc}\frac{(x_1^2-1)^3}{x_1}=0$$ gives $$81u^5-135u^3v^2+45uv^4+15u^2w^3-5v^2w^3-(27u^3-27uv^2+3w^3)+3u+u-2=0$$ and we obtain: $$(u-1)^4(9u^3+18u^2+9u+4)(18u-9u^2-1)=0.$$ If $u=2$, so $v^2=0$ and $w^3=-68$, but the equation $$x^3-6x^2+68=0$$ has two complex roots.

Thus, $u\neq2$ and from here $18u-9u^2-1\neq0.$

Also, since $$\prod_{cyc}(x_1-x_2)^2=27(3u^2v^4-4v^6-4u^3w^3+6uv^2w^3-w^6)\geq0,$$ we obtain: $$\frac{(u-1)^6(9u^3-27u-14)(9u^3-3u+2)}{(18u-9u^2-1)^3}\geq0.$$ Also, $u^2\geq v^2$ gives $$u^2\geq \frac{(9u^3-3u+2)(2-u)}{18u-9u^2-1}$$ or $$\frac{(u-1)^2}{18u-1-9u^2}\leq0,$$ which gives $18u-1-9u^2<0$ or $u=1$.

Now, let $u\neq1$.

Thus, $18u-1-9u^2<0$, $(9u^3-27u-14)(9u^3-3u+2)\leq0$ and $9u^3+18u^2+9u+4=0,$ which is impossible.

Thus, $u=1$, $v^2=w^3=1$, which gives $x_1=x_2=x_3=1.$

We see that the condition $x_j\geq1$ or $x_j\leq-1$ is not necessary.

  • 1
    Thanks! I will check it today or tomorrow. – cbi Oct 02 '23 at 16:19
  • Great, by the observation $u^2\geq v^2$, your computation (I haven't completely checked) eliminates the restriction on $|x_j|$. – cbi Oct 03 '23 at 03:16
  • However, I am not clear why you consider the case $u=2$ individually. – cbi Oct 03 '23 at 03:49
  • 1
    @cbi Because $(u-2)w^3+v^2=0$ and if $u\neq2$, so $w^3=\frac{v^2}{2-u}.$ Also, the case $u\neq2$ gives a possibility to prove that $18u-1-9u^2<0$ for $u\neq1$. – Michael Rozenberg Oct 03 '23 at 04:24