5
$\begingroup$

Let $x,y,z>0$. Prove that $$\frac{x+y+z}{3}+\frac{3}{\frac1x+\frac1y+\frac1z}\geq5\sqrt[3]{\frac{xyz}{16}}$$

On the left-hand side we have arithmetic and harmonic means, while on the right geometric mean. The AM-GM-HM inequality states that AM $\geq$ GM $\geq$ HM, but here HM is on the wrong side of the inequality.

$\endgroup$
5
$\begingroup$

Since we have a homogeneous inequality, we may assume $xyz=1$ without loss of generality, then find the infimum of $AM(x,y,z)+HM(x,y,z)$ over the set $$D=\left\{(x,y,z)\in\mathbb{R}_{+}^3: xyz=1\right\}.$$ So we want: $$ \inf_{\mathbb{R}^+\times\mathbb{R}^+}\left(\frac{x+y+\frac{1}{xy}}{3}+\frac{3}{xy+\frac{1}{x}+\frac{1}{y}}\right)=\inf_{\mathbb{R}^+\times\mathbb{R}^+}\left(\frac{x^2 y+xy^2+1}{3xy}+\frac{3xy}{x^2y^2+x+y}\right).$$ Let $f(x,y)$ be the last function. It is a symmetric function, hence it is enough to study its behaviour over the set $0<y\leq x$. So, let $y=kx$ with $k\in(0,1]$. We have: $$ g_k(x)=f(x,kx) = \frac{1}{3kx^2}+\frac{(k+1)x}{3}+\frac{3kx}{1+k+k^2 x^3} $$ and disregarding the last (positive) term, by the AM-GM inequality: $$ g_k(x)\geq \frac{1}{3kx^2}+\frac{(k+1)x}{6}+\frac{(k+1)x}{6} \geq 3\sqrt[3]{\frac{(k+1)^2}{108k}}$$ where the last term is a decreasing function over $(0,1]$. We just need a little improvement of the last inequality to prove the original claim. So, let us study $f(x,y)$ under another perspective. Assuming $x+y=k$ we have: $$ f(x,y) = \frac{k+\frac{1}{xy}}{3}+\frac{3}{\frac{k}{xy}+xy}=\frac{k+\frac{1}{x(k-x)}}{3}+\frac{3}{\frac{k}{x(k-x)}+x(k-x)} $$ and by putting $x=\lambda k, y=(1-\lambda)k$ we are left with: $$ \frac{k+\frac{1}{k^2\lambda(1-\lambda)}}{3}+\frac{3}{\frac{1}{k\lambda(1-\lambda)}+k^2\lambda(1-\lambda)}=\color{red}{\frac{k+\frac{1}{k^2 L}}{3}+\frac{3}{\frac{1}{kL}+k^2 L}=h(L,k)}$$ where $L$ ranges over $\left(0,\frac{1}{4}\right)$ and $k$ ranges over $\mathbb{R}_+$. The stationary points of $h(L,k)$ for a fixed $k$ occur at $$ L \in\left\{\pm\frac{1}{\sqrt{2} k^{3/2}},\pm\frac{1}{\sqrt{5} k^{3/2}}\right\} $$ and $$ h\left(\frac{1}{\sqrt{2} k^{3/2}},k\right) = \frac{4}{3}\sqrt{\frac{2}{k}}+\frac{k}{3},\qquad h\left(\frac{1}{\sqrt{5} k^{3/2}},k\right) = \frac{5}{6}\sqrt{\frac{5}{k}}+\frac{k}{3},$$ so the AM-GM inequality proves the claim.

In facts, the original inequality is optimal and our approach shows that equality is attained by $$ \color{red}{(x,y,z) = \lambda\cdot (1,4,4)} $$ and cyclic shifts.

$\endgroup$
  • $\begingroup$ @Lucian: where? After defining $D$, I have eliminated $z$ through the constraint $xyz=1$. $\endgroup$ – Jack D'Aurizio May 30 '16 at 21:27
  • $\begingroup$ Have you also tried tackling this question ? $\endgroup$ – Lucian May 30 '16 at 21:38
  • $\begingroup$ @Lucian: I will think about that, but non-homogeneous inequalities usually requires different tricks than a suitable change of variables plus partial derivatives. $\endgroup$ – Jack D'Aurizio May 30 '16 at 22:21
  • $\begingroup$ Hello Jack. Is there any book you reccomend where to learn about inequalities (homogenous or non-homogenous)? I heard "homogenous inequality" before but I never knew why "homogenous" is important; now thanks to your answer I see some of the tricks you can use, and I would like to learn more $\endgroup$ – Ovi May 25 '18 at 4:36
  • $\begingroup$ @Ovi: I recommend the books from Micheal Steele (The Cauchy-Schwarz master class) and Pham Kim Hung (I do not recall the exact title at the moment, but it should not be too difficult to find). The former is a bit more argumentative, the latter a bit more concise and technical, but they are both good sources of knowledge about the topic. $\endgroup$ – Jack D'Aurizio May 25 '18 at 12:56
0
$\begingroup$

Let $x+y+z=3u$, $xy+xz+yz=3v^2$ and $xyz=w^3$.

Thus, we need to prove that $f(u)\geq0,$ where $$f(u)=u+\frac{w^3}{v^2}-\frac{5w}{\sqrt[3]{16}}.$$ Id est, it's enough to prove our inequality for the minimal value of $u$, which happens for equality case of two variables.

Since our inequality is symmetric and homogeneous, we can assume $y=z=1$ and $z=16a^3$

and we need to prove that: $$\frac{16a^3+2}{3}+\frac{48a^3}{32a^3+1}\geq5a$$ or $$(4a-1)^2(32a^4+16a^3-24a^2+a+2)\geq0$$ or $$(4a-1)^2(2(4a^2-1)^2+a(4a-1)^2)\geq0,$$ which is obvious.

$\endgroup$

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.