-1

If $a$, $b$, $c$ are positive real number and abc=1. Then prove that $$\frac ab+\frac bc+\frac ca\geqslant a+b+c$$

Using AM-GM both sides will be greater than $3$, But I don't know how to proceed further. Which inequality should I use?

Ѕᴀᴀᴅ
  • 34,263
Forrest
  • 17

4 Answers4

1

An answer with my thought process.

We want to use $abc=1$, so we clear the denominator on the left:

$a^2c+ab^2+c^2b \ge a+b+c$

Now we are scared because we have something cyclic on the left, something symmetric on the right, and still inhomogenous. We decide to homogenize, so that we can use inequalities like AM-GM:

$a^2c+ab^2+c^2b \ge (a+b+c)(abc)^{2/3}=a^{5/3}b^{2/3}c^{2/3}+a^{2/3}b^{5/3}c^{2/3}+a^{2/3}b^{2/3}c^{5/3}$ [0]

Now we remember than in the past we saw some inequalities solved proving some "smaller" inequalities that, after being summed up, returned the wanted result.

Let's try now weighted AM-GM on the first two addends of the left member of [0]:

$p_1a^2c+p_2 ab^2 \ge a^{2p_1+p_2}b^{2p_2}c^{p_1}$ [1]

For this to be useful to prove [0], on the exponent of the right member we would like to have $p_1=2p_2$, so that $p_1=2/3,p_2=1/3$. So we recover $b^{2/3}c^{2/3}$, which is what we like. what about the exponent of $a$ ? We have $2p_1+p_2=5/3$. Very good it is going in the right direction. Maybe this is useful ? Let's write [1] with this choice of $p_1$ and $p_2$ :

$\frac{2}{3}a^2c+\frac{1}{3} ab^2 \ge a^{5/3}b^{2/3}c^{2/3}$ [1a]

So now this could be the first piece. Let's try to write other small inequalities for the remaining terms:

$\frac{2}{3}ab^2+\frac{1}{3} bc^2 \ge a^{2/3}b^{5/3}c^{2/3}$ [1b]

$\frac{2}{3}bc^2+\frac{1}{3} a^2c \ge a^{2/3}b^{2/3}c^{5/3}$ [1c]

Summing up [1a],[1b] and [1c] we recover [0] and we are done.

Please write any comments if you find mistakes, especially on the thought process :) .

I am also curious if Muirhead's inequality could help in proving [0] more directly, even if the some on the left is cyclic and not symmetric.

Thomas
  • 4,029
  • 1
    According to the meta post Enforcement of Quality Standards, which you've likely been notified about, before answering, you should attempt a search for a question like this, which has been asked frequently on this site. Then, if you find no dupes, you should assist the asker to improve their question. – amWhy Jun 01 '21 at 16:47
0

Let $a=e^u$,$b=e^v$,$c=e^w$

The inequality becomes :

Let $u+v+w=0$ then we have :

$$e^{u-v}+e^{v-w}+e^{w-u}\geq e^u+e^v+e^w$$

Or :

$$\sum_{cyc}e^u(e^{-v}-1)\geq 0$$

Or

$$\sum_{cyc}e^u(e^{w+u}-1)\geq 0$$

Now we use weighted jensen's inequality to get :

$$\sum_{cyc}e^u(e^{u+w}-1)\geq (e^u+e^v+e^w)(e^{\frac{(w+u)e^u+(u+v)e^v+(v+w)e^w}{e^u+e^v+e^w}}-1)$$

Remains to show that :

$$\frac{(w+u)e^u+(u+v)e^v+(v+w)e^w}{e^u+e^v+e^w}\geq 0$$

Or :

$$(w+u)e^u+(u+v)e^v+(v+w)e^w\geq 0$$

Wich is not hard and true using Tchebythev's inequality.

-1

We write $a=A^3$, $b=B^3$, $c=C^3$ with $A,B,C>0$ and $ABC=1$. Then $a/b=a^2c=A^6C^3$, and we have to show the homogeneous inequality: $$ A^6C^3 + B^6A^3 + C^6B^3\overset !\ge (A^3+B^3+C^3)A^2B^2C^2\ . $$ For this, use the AM-GM-inequaltity in the form: $$ \frac 13(A^6C^3 + A^6C^3 + B^6A^3) \ge (A^6C^3 \cdot A^6C^3 \cdot B^6A^3) ^{1/3} = A^5B^2C^2 $$ and its cyclic versions.

dan_fulea
  • 32,856
  • 1
    According to the meta post Enforcement of Quality Standards, which you've likely been notified about, before answering, you should attempt a search for a question like this, which has been asked frequently on this site. Then, if you find no dupes, you should assist the asker to improve their question. – amWhy Jun 01 '21 at 16:48
  • 1
    There is indeed a downvote for this answer. OK, here we have an inequality to show, and edits have been done, i know that the problem should be also in the text, not (only) in the title. It appears that the question has no context. However, the user posting the question adds the information "Using AM-GM both sides will be greater than 3, But I don't know how to proceed further." It is obvious, that he did something, and this is more than in the case of many questions around. I needed some seconds to decide if i answer or not, did answer as a quora link was not my favourite. @amWhy – dan_fulea Jun 01 '21 at 17:10
  • 1
    Just to have a comparison, which are the standards respected by the question here: https://math.stackexchange.com/questions/4129543/boolean-logic-to-arithmetic/4129569#4129569 ?! @amWhy – dan_fulea Jun 01 '21 at 17:13
  • 1
    Sometimes, the didactic method in mathematics means to offer a solution when (and only when) the first step is done, if this solution is following the path wanted by the person doing the first step. We definitively do not have here an expert in calculus, Lagrange multiplicators, and Muirhead or domination... We have a new user, who deserves to be encouraged. We have a compact hint about the fact that both sides are $\ge 3$, which is a first step. The above solution is simple, and follows the path of thinking in the OP. Just try to understand the situation from all sides now. @amWhy – dan_fulea Jun 01 '21 at 17:20
  • My post was answered before the updated policy. Please don't try to change the subject playing the "whataboutism" game, Dan. Thanks! – amWhy Jun 01 '21 at 17:22
  • 1
    @amWhy So an answer a few weaks ago is a valid answer. In between some policy is changed, this is the occasion for me to know about this. OK, my situation while answering was as follows. The question was edited, but i did not like the edit, in the comments there was a hint to a solution of some Quora forum, i did not like that one, one answer was almost immediately posted, i did not like that answer. Got 2 downvotes, the question is closed. The reason for closing is that there is an other solution i did not know, and it is a solution not on my taste. What should i do now? Delete? – dan_fulea Jun 01 '21 at 17:30
-1

Because the left sum is cyclic, we can assume $a \le b \le c$ or $c \le b \le a$.

First assume $a \le b \le c$. From $abc = 1$, we can also conclude $a \le 1 \le c$. Then there are $x,y\ge 0$ such that $a = 1/(x + 1)$, $b = (x + 1)/(y + 1)$, $c = y + 1$. It's a bit of algebra, but if you substitute these into the inequality, collect terms, and clear denominators, you'll find it's equivalent to $$ 2(x-y)^2 + P(x, y)\ge 0, $$ where $P$ is a polynomial with only positive coefficients. This is clearly true for all $x, y\ge 0$.

Now apply the same logic in the case $c\le b\le a$. In this case, the polynomial it reduces to is simpler: $$ (x-y)^2(x + y + 2) + x(x+2)y \ge 0 $$ This is also clearly true for all $x,y\ge 0$.

Since the statement is true in both cases, it is true in general.

eyeballfrog
  • 22,485