5

Given $a,b,c>0$, $$a+b+c=ab+bc+ca$$, prove $$\sqrt{ab}+\sqrt{bc}+\sqrt{ca}-\sqrt{abc}\ge 2$$

I tried with derivatives but haven't solved it yet. Is there a more natural and elementary proof?

My progress with derivatives:

First $a+b+c=ab+bc+ca\le \frac{(a+b+c)^2}{3}$, so $a+b+c\ge 3$.

Then $a+b+c=ab+bc+ca \Leftrightarrow \frac1a+\frac1b+\frac1c=\frac1{ab}+\frac1{bc}+\frac1{ca}$, so similarly $\frac1a+\frac1b+\frac1c\ge 3$. Hence $ab+bc+ca\ge 3abc$.

WOLOG, we assume $a>b>c$. Obviously $a>1>c$.

We have $b+c > 1$, otherwise $ab+bc+ca=(b+c)a+bc<a+bc<a+b+c$: contradiction. Now we have $(\sqrt b +\sqrt c)^2 > b+c >1$, so $\sqrt b+ \sqrt c>1$.

With notation $s:=\sqrt a + \sqrt b + \sqrt c$ and $t:=\sqrt{abc}$, we have $\sqrt{ab}+\sqrt{bc}+\sqrt{ca}-\sqrt{abc}\ge 2 \Leftrightarrow (\sqrt{ab}+\sqrt{bc}+\sqrt{ca})^2 \ge (2+\sqrt t)^2 $, which is equivalent to $$ab+bc+ca + 2s \sqrt{t} \ge 4 + 4\sqrt t + t$$, call it (*), now we discuss by case a) $abc \ge 1$ and case b) $abc<1$.

Case a) $abc \ge 1$. Inequation (*) holds if $$3t+6\sqrt t \sqrt[6] t\ge 4 + 4\sqrt t + t$$ holds, which again is equivalent to $$-2t+4\sqrt t -6t^{\frac23}+4\le 0$$ , but this can be proved by computing the derivative.

Case b) $abc<1$. I haven't solved this yet.

athos
  • 5,177
  • Please edit your question to show the proof that you found. With respect to the (somewhat arbitrary) MathSE standards discussed here, this will allow mathSE reviewers to post alternative solutions. – user2661923 Feb 09 '22 at 22:52
  • @user2661923 sorry just realized the proof I found actually does not work for case b, as updated in the question. – athos Feb 09 '22 at 23:58
  • +1 to your question, after the edit. – user2661923 Feb 10 '22 at 00:30
  • one can do it with Lagrange multipliers in the case $abc\le 1$ by showing that under $a \ge b \ge c$ we have $(\sqrt {ab}+\sqrt {ac})^2 \ge (\sqrt a + \sqrt {abc} -\sqrt {bc})^2$ by fairly straightforward simplifications; so the inequality is true for $a >4$ and then we work in a box $1 \le a \le4, 0\le b \le 4, 0 \le c \le 1$ and find the critical points inside for the required expression under the constraint and then analyze the boundary, while the case $abc \ge 1$ can be proven from $abc \le 1$ using the substitutions $a \to 1/a$ etc and some considerations, but it is kind of ugly overall – Conrad Feb 10 '22 at 19:02
  • applying Lagrange multipliers is actually quite easy, the problem is in finding a good boundary and the main issue imho is that there is a minimum for $\sqrt{ab}+\sqrt{bc}+\sqrt{ca}-\sqrt{abc}$ at $(1,1,1)$ but also one in the limit at $(2, 2, 0)$ and that makes the case $c$ very small (and $abc$ small) tricky – Conrad Feb 10 '22 at 19:06
  • @Conrad I’m not sure if I understand your idea, do you mind expand it to an answer pls? – athos Feb 10 '22 at 19:27
  • for simplicity with $a=x^2, b=y^2, c=z^2, 0 \le z \le y \le x, x \ge 1$, you need to minimize $f(x,y,z)=xy+yz+zx-xyz$ under the constraint $x^2+y^2+z^2-(xy)^2-(yz)^2-(zx)^2=0$ and that is a non singular surface in our case ($x \ge1$) so you can use Lagrange multipliers assuming you have a good boundary – Conrad Feb 10 '22 at 19:44
  • 1
    There are solutions here. – Calvin Lin Feb 10 '22 at 23:27
  • 1
    @CalvinLin which reply on the thread is the proof? I really would like to digest but the thread is quite messy – athos Feb 12 '22 at 07:37

1 Answers1

1

Remark: @Calvin Lin's comment reminds me that, in 2019, I proved some similar inequalities under the condition $a + b + c = ab + bc + ca$.


WLOG, assume that $c = \min(a, b, c)$.

Using AM-GM, we have \begin{align*} &\sqrt{ab} + \sqrt{bc} + \sqrt{ca} - \sqrt{abc} - 2\\ \ge\,& \sqrt{ab} + 2\sqrt{\sqrt{bc}\sqrt{ca}} - \sqrt{abc} - 2\\ =\,& (1 - \sqrt{c})\sqrt{ab} + 2\sqrt{c}\sqrt[4]{ab} - 2. \tag{1} \end{align*}

From $ab + bc + ca = a + b + c$, we have $(1 - c)(a + b) = ab - c$. Thus, $c \le 1$ (easy). Thus, we have $(1 - c)\cdot 2\sqrt{ab} \le ab - c$ or $(\sqrt{ab} - 1 + c)^2 \ge c^2 - c + 1$ which results in $$\sqrt{ab} \ge \sqrt{c^2 - c + 1} + 1 - c. \tag{2}$$ Note: $\sqrt{ab} - 1 + c \le -\sqrt{c^2 - c + 1}$ is impossible, since $1 - c - \sqrt{c^2 - c + 1} < 0$ for all $c > 0$.

We split into two cases:

  1. $c = 1$:

From (2), we have $\sqrt{ab} \ge 1$. From (1), the desired result follows.

  1. $0< c < 1$:

From (1), it suffices to prove that $$(1 - \sqrt{c})\sqrt{ab} + 2\sqrt{c}\sqrt[4]{ab} - 2 \ge 0$$ which is written as $$(1 - \sqrt c)\left(\sqrt[4]{ab} + \frac{\sqrt c}{1 - \sqrt c}\right)^2 \ge \frac{2 + c - 2\sqrt c}{1 - \sqrt c}.$$ It suffices to prove that $$\sqrt[4]{ab} + \frac{\sqrt c}{1 - \sqrt c} \ge \frac{\sqrt{2 + c - 2\sqrt c}}{1 - \sqrt c}$$ or $$\sqrt[4]{ab} \ge \frac{\sqrt{2 + c - 2\sqrt c} - \sqrt c}{1 - \sqrt c} = \frac{2}{\sqrt{2 + c - 2\sqrt c} + \sqrt c}$$ or $$\sqrt{ab} \ge \frac{4}{(\sqrt{2 + c - 2\sqrt c} + \sqrt c)^2} = \frac{4}{2 + 2c - 2\sqrt c + 2\sqrt c \sqrt{1 + (1 - \sqrt c)^2}}.$$ It suffices to prove that $$\sqrt{ab} \ge \frac{4}{2 + 2c - 2\sqrt c + 2\sqrt c } = \frac{2}{1 + c}.$$

Using (2), it suffices to prove that $$\sqrt{c^2 - c + 1} + 1 - c \ge \frac{2}{1 + c}$$ or $$\sqrt{c^2 - c + 1} \ge \frac{1 + c^2}{1 + c}$$ or $$c^2 - c + 1 \ge \left(\frac{1 + c^2}{1 + c}\right)^2$$ or $$\frac{c(1 - c)^2}{(1 + c)^2}\ge 0.$$

We are done.

River Li
  • 37,323
  • amazing. thank you very much. may I ask, how did you figure out this? there are several steps not obvious to me at all. is it that i just have to work on all such inequalities such as https://artofproblemsolving.com/community/c6h1183059p5735328 to get familiar with the results? or there are actually some tricks? – athos Feb 13 '22 at 17:16
  • 1
    @athos I think the way here (finding the bounds of $ab$) to deal with the condition $a + b + c = ab + bc + ca$ works for some of such inequalities. In detail, for a symmetric $f(a,b,c)$, first find a bound $f(a,b,c)\ge g(ab, c)$ for some $g$ (with some assumption such as $c = \min(a,b,c)$), then use the bound $ab \ge \cdots$. – River Li Feb 14 '22 at 01:02
  • @athos By the way, letting $a = x^2, b = y^2, c = z^2$, we can use pqr method ($p = x + y + z, q = xy + yz + zx, r = xyz$). But perhaps it is quite complicated. – River Li Feb 14 '22 at 01:04
  • @athos The main part of the proof is (1) and (2). I gave one way to prove (1) based (2). Alternatively, one may directly insert (2) into (1) at the beginning to get an inequality of one variable $F(c) \ge 0$ for all $0 < c \le 1$. Then prove it. – River Li Feb 14 '22 at 01:08
  • thank you for sharing the insight! – athos Feb 15 '22 at 08:01
  • @athos You are welcome. – River Li Feb 15 '22 at 09:07