3

Let $x,y,z>0$ and $xyz=1$. Prove that $\dfrac{x^3}{(1+y)(1+z)}+\dfrac{y^3}{(1+x)(1+z)}+\dfrac{z^3}{(1+x)(1+y)}\ge \dfrac34$

My attempt:

Since it is given that $xyz=1$, I tried substituting $x=\dfrac{a}{b},y=\dfrac{b}{c},z=\dfrac{c}{a}$. But the expansion looked really ugly and I didn't think I could make out anything out of it.

So, I made another attempt, if each element was greater than $\dfrac{1}{4}$, we could have a solution, so, treating that way, I get $4x^3\ge 1+x+y+xy, 4y^3\ge 1+x+z+xz, 4z^3\ge 1+x+y+xy$. Using AM-GM I get an equality.

So, please help. Thank you.

dustin
  • 8,241
Swadhin
  • 1,381
  • Oh! These kind of inequalities remind me of my first courses in Analysis and I remember in many of them we had to use these classical theorems, like Bolzano's, Rolle's, mean value Theorems and Taylor expansions.. but I do not know the context :) I will like to see the solution too – Martingalo Feb 05 '15 at 08:24
  • Your approach with splitting up an inequality into multiple inequalities term-by-term is very unlikely to be the right way; it is a much stronger statement. I would recommend you to look up basic inequalities like QM-AM-GM-HM, Cauchy-Schwartz, Hölder, Maclaurin, Jensen etc., these are always helpful. – Redundant Aunt Feb 05 '15 at 09:10

3 Answers3

5

As another approach, you could prove it as follows.

Due to Hölder's inequality, we have: $$ \left(\sum_{cyc} \frac{x^3}{(1+y)(1+z)}\right)\cdot\left(\sum_{cyc} (1+y)\right)\cdot\left(\sum_{cyc} (1+z)\right)\ge(x+y+z)^3\iff \sum_{cyc} \frac{x^3}{(1+y)(1+z)}≥\frac{(x+y+z)^3}{\left(\sum_{cyc} (1+y)\right)\cdot\left(\sum_{cyc} (1+z)\right)}=\frac{(x+y+z)^3}{\left(3+x+y+z\right)^2} $$ So it remains to prove that: $$ \frac{(x+y+z)^3}{\left(3+x+y+z\right)^2}≥\frac34 $$ Setting $x+y+z=a$, this is equivalent to: $$ \frac{a^3}{\left(3+a\right)^2}≥\frac34\iff 4a^3≥27+18a+3a^2\iff 4a^3-3a^2-18a-27≥0 $$ We have equality at $a=3$, so we can factor out $a-3$: $$ (a-3)\left(4a^2+9a+9\right)≥0 $$ But since $$ a=x+y+z≥3(xyz)^{\frac{1}{3}}=3\iff a-3≥0 $$ due to AM-GM, this is always true, so we're done.

Redundant Aunt
  • 12,030
  • 2
  • 22
  • 66
1

Edit1:

another approach is to use $\dfrac{x^2}{a}+\dfrac{y^2}{b} \ge \dfrac{(x+y)^2}{a+b},x+y+z\ge3$

LHS $\ge \dfrac{(x^2+y^2+z^2)^2}{\sum_{cyc} {(1+y)(1+z)}{x}}\ge\dfrac{(x+y+z)^4}{9(x+y+z+2(xy+yz+xz)+3)}\ge \dfrac{(x+y+z)^4}{9(x+x+z)+6(x+y+z)^2+27}\ge\dfrac{(x+y+z)^4}{9(x+y+z)+6(x+y+z)^2+9(x+y+z)}=\dfrac{(x+y+z)^3}{18+6(x+y+z)}\ge\dfrac{(x+y+z)^3}{6(x+x+z)+6(x+y+z)}=\dfrac{(x+y+z)^2}{12}\ge \dfrac{3^2}{12}=\dfrac{3}{4} $

all "=" hold when $x=y=z=1$

chenbai
  • 7,581
0

I think you can do this using Lagrange multipliers. Define the function:

$$f(x,y,z)= \frac{x^3}{(1+y)(1+z)} +\frac{y^3}{(1+x)(1+z)} + \frac{z^3}{(1+x)(1+y)} - \frac{3}{4}.$$

We want to show that $f(x,y,z)\geq 0$ for all $x,y,z\in \mathbb R^3$ such that $xyz=1$. Define the Lagrangian:

$$L(x,y,z,\lambda) = f(x,y,z)- \lambda g(x,y,z)$$ where $g(x,y,z)=0$ is the function defining the set where you look at: that is $xyz=1$ and hence $g(x,y,z)=xyz-1$. As a result, $$L(x,y,z,\lambda) = f(x,y,z)- \lambda (xyz-1).$$

Differentiating $L$ w.r.t. all variables we get \begin{align*} \partial_x L(x,y,z,\lambda) =& \frac{3x^2}{(1+y)(1+z)} -\frac{y^3}{(1+x)^2(1+z)} - \frac{z^3}{(1+x)^2(1+y)} -\lambda y z = 0\\ \partial_y L(x,y,z,\lambda) =& -\frac{x^3}{(1+y)^2(1+z)} +\frac{3y^2}{(1+x)(1+z)} -\frac{z^3}{(1+x)(1+y)^2} - \lambda xz=0\\ \partial_z L(x,y,z,\lambda) =& -\frac{x^3}{(1+y)(1+z)^2} -\frac{y^3}{(1+x)(1+z)^2} + \frac{3z^2}{(1+x)(1+y)} - \lambda x y = 0\\ \partial_{\lambda} L(x,y,z,\lambda) =& 1-xyz=0 \end{align*}

Now try to exploit the symmetries or similarities in order to solve this system. The critical points can be maxima, minima or saddle points. If you happen to find only minima and such points satisfy $f(x,y,z)\geq 0$ then you will have proven the inequality. Since you are proving that the function $f$ has minima in the set $\{(x,y,z)\in \mathbb{R}^3 \, xyz=1\}$ and all of them are positive.

Be careful that the function is continuous and the set you are looking at is closed.

Another possibility is to subsitute the vaue for $z=1/(xy)$ (it is ok since $x,y,z\neq 0$) in the inequality and define a two-variable function $f(x,y)$ and do exactly the same without using Lagrange Method. Maybe that is easier.

Martingalo
  • 1,857
  • Unfortunately I am not familiar with Lagrange Multipliers. – Swadhin Feb 05 '15 at 08:41
  • What about the second option? Can you use two-variable calculus? – Martingalo Feb 05 '15 at 08:42
  • Depends on your usage, I don't understand properly, but knowledge is upto undergrad level, the book I am solving is of standard 9 olympiad level. – Swadhin Feb 05 '15 at 08:43
  • Aha! Then it is definetly a smart idea you should use here... I suspect one need "no knowedlge" to solve these kind of olympiad problems, in the sense that.. you need to recall results. Then let's see if someone has a good idea :) Good luck – Martingalo Feb 05 '15 at 08:45
  • and I have the feeling that your idea of proving that one single term is greater than 1/4 is the way to start.. – Martingalo Feb 05 '15 at 08:46
  • Can you tell me which book is it? –  Feb 05 '15 at 09:06