If $x,y,z$ are real and $x^2+y^2+z^2=9$, how can we prove that $2(x+y+z)-xyz\leq 10$?

Please provide a solution without the use of calculus. I know the solution in that way.

  • 1
    $\begingroup$ Does the solution you know uses Lagrange multipliers ? $\endgroup$ – Thibaut Dumont Dec 6 '12 at 9:43
  • 1
    $\begingroup$ @ThibautDumont: I think that's what he means by "without the use of calculus". $\endgroup$ – Mhenni Benghorbal Dec 6 '12 at 9:48
  • $\begingroup$ I might investigate expansions of $(x-1)(y-1)(z-1)$ and $(x+y+z)(x+y+z)$. All the terms that you want will show up, and you might be able to put bounds on the extraneous terms. $\endgroup$ – orlandpm Dec 7 '12 at 2:00
  • $\begingroup$ @orlandpm I already tried it that way.Doesn't seem to produce anything fruitful. $\endgroup$ – Sai Krishna Deep Dec 7 '12 at 6:57

Case 1: At least one of $x,y,z < 0$. Let's say $z < 0$. If we let $f(x,y,z) = 2(x+y+z) - xyz$, it's then clear that $f(x,y,z) \leq f(t,t,z)$, where $t = \sqrt{\frac{x^2+y^2}{2}}$. It then suffices to check $f(t,t,z) \leq 10$. Let $z = -u$, so that $u$ is positive. Then we need to show $$f(t,t,-u) = 2(2t - u) + t^2 u \leq 10$$ on the condition $2t^2 + u^2 = 9$. We can try to express everything in terms of $u$: $$ 2(2t - u) + t^2 u \leq 10 \\ \Leftrightarrow 4(2t - u) + (9 - u^2)u \leq 20 \\ \Leftrightarrow u^3 - 5u + 20 \ge 8t = 8 \sqrt{\frac{9 - u^2}{2}} \\ \Leftrightarrow (u^3 - 5u + 20)^2 - 32 (9-u^2) \ge 0 \\ \Leftrightarrow (u-1)^2(u^4 + 2u^3 - 7u^2 + 24u + 112) \ge 0$$

The last inequality is clearly true, since $u^4 + 49 \ge 14u^2 > 7u^2$. So this case holds, with equality $z = -u = -1$, $x = y = t = 2$.

Case 2: $x,y,z \ge 0$. WLOG assume $x \ge y \ge z$.

Case 2a: $x^2 \leq \frac{9 + \sqrt{65}}{2}$. ($x$ is "small")

We check that $f(x,y,z) \leq f(x, \sqrt{y^2+z^2}, 0)$. $$2(x+y+z) - xyz \leq 2(x+\sqrt{y^2+z^2}) \\ \Leftrightarrow 2(y+z - \sqrt{y^2+z^2}) \leq xyz \\ \Leftrightarrow \frac{4}{y+z+\sqrt{y^2+z^2}} \leq x \\ \Leftrightarrow 4 \leq x(y+z+\sqrt{y^2+z^2} $$ The last inequality follows from $$x(y+z+\sqrt{y^2+z^2}) \ge 2x\sqrt{y^2+z^2} = 2 x \sqrt{9-x^2} \ge 4$$ and the condition on $x$.

Thus we may assume $z = 0$. Then $f(x,y,z) = 2(x+y) \leq 2\sqrt{2(x^2+y^2)} = 2\sqrt{18} < 10$.

Case 2b: $x^2 \ge \frac{9 + \sqrt{65}}{2}$. ($x$ is "large")

We get a trivial bound on $y+z$ by Cauchy-Schwarz: $$y + z \leq \sqrt{2(y^2+z^2)} = \sqrt{2(9-x^2)} \leq \sqrt{9 - \sqrt{65}} < 1$$ Then $$f(x,y,z) = 2(x+y+z) - xyz < 2(x+y+z) \leq 2 (3+1) < 10.$$

All in all, it's possible to do this without calculus, although I can't understand why one would not use a tool like Lagrange multiplier that is especially well suited for this situation. It's exactly the power of those tools that render brute-force solutions like this one unnecessary.


without loss of generality we can assume that $|x|\le |y|\le |z|$,and Note that $3z^2\ge x^2+y^2+z^2$, which implies that $z^2\ge 3$

we have \begin{align} [2(x+y+z)-xyz]^2&=[2(x+y)+z(2-xy)]^2\\ &\le [(x+y)^2+z^2][4+(2-xy)^2]\\ &=(9+2xy)(8-4xy+(xy)^2))\\ &=2(xy)^3+(xy)^2-20(xy)+72\\ &=(xy+2)^2(2xy-7)+100 \end{align} from $z^2\ge 3$ it follows that $2xy\le x^2+y^2=9-z^2\le 6$,then $$[2(x+y+z)-xyz]^2\le 100$$ or equivalently $$2(x+y+z)-xyz\le 10$$


Here is an attempt. I believe it has to be reconsidered for negative numbers, though.

$0\leq (x+y+z-1)^2$

$0\leq x^2+y^2+z^2+1-2(x+y+z)+2xyz(\frac{1}{x}+\frac{1}{y}+\frac{1}{z})$

$2(x+y+z)-2xyz(\frac{1}{x}+\frac{1}{y}+\frac{1}{z})\leq x^2+y^2+z^2+1$

EDIT: As pointed out my bound is the wrong way :( But I hope the first steps gives others some ideas.

That's almost what we want. We only need to bound the harmonic mean by the quadratic mean now.

$\frac{3}{\frac{1}{x}+\frac{1}{y}+\frac{1}{z}}\leq \sqrt{\frac{x^2+y^2+z^2}{3}}$

In our case: $\frac{1}{x}+\frac{1}{y}+\frac{1}{z}\geq \sqrt{3}$

which is enough.

Hope there is no mistake :)

  • $\begingroup$ You need an upper bound for $1/x + 1/y + 1/z$ to get conclusion, not the lower bound. $\endgroup$ – user27126 Dec 8 '12 at 9:37
  • $\begingroup$ Oh true. And that means this way cannot work? :( $\endgroup$ – Gerenuk Dec 8 '12 at 9:45

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.