2

My first way is to put the inequality under the same degree, so I multipled both sides to $(a+b+c)$, however it leaded to a hard to solve result. Can anyone help me with this problem?

Tmt
  • 97
  • WLOG, we may assume that $a \geq b \geq c$, and we may use $3 = (a + b + c)^2 / 3$, then RHS becomes $\frac{1}{3} \times ((a + b + c)^2 + (c - a)^2)$. – Vezen BU Feb 10 '22 at 02:28
  • Actually if you assume that way the problem is 4/3x(c-a)^2, not 1/3x(c-a)^2 – Tmt Feb 10 '22 at 02:48
  • @VezenBU I don't think you can assume $a\ge b\ge c$ directly since LHS is only cyclic, not symmetric... Oh actually you can since exchanging $b,c$ makes it larger. – JetfiRex Feb 10 '22 at 02:58
  • I’m truly sorry for the mistake in the problem. The original one is a^2/b – Tmt Feb 10 '22 at 03:29
  • @Tmt never mind, still can be proved. – JetfiRex Feb 10 '22 at 04:19

2 Answers2

4

For the current inequality of $ \sum \frac{a^2}{b} \geq 3 + \frac{4}{3} \max ( a-b)^2 $:

  1. Replace 3 with $ a+b+c$. The reason for this choice is the "well-known" inequality: $ \sum \frac{ a^2}{b} \geq \sum a $. We're then asking how much more leeway there is in the difference of these terms.

  2. We can write the difference as a as Sum of Squares (Figure out how to do this before looking at the hint.). Thus, we want to show that

$$ \sum \frac{ (a-b)^2 } { b} \geq \frac{4}{3} \max ( a-b)^2.$$

This is just Cauchy Schwarz:

  • Numerator: $[ \sum | a - b | ] ^2= [2 ( \max (a,b,c) - \min (a, b, c) )]^2 = 4 \max((a-b)^2)$.
  • Denominator: $a + b + c = 3$.

Equality holds in the Cauchy Schwarz iff:

  • For $ a \geq b \geq c$: $\frac{ a-b}{b} = \frac{b-c}{c} = \frac{a-c}{a} , a+b+c = 3$ $\Rightarrow (a, b, c) = (1, 1, 1) , (\frac{3}{2} , \frac{ 3 \sqrt{5} - 3 } { 4} , \frac{ 9 - 3 \sqrt{5}}{ 4} ). $
  • For $ a \geq c \geq b$: $\frac{ a-b}{b} = \frac{c-b}{c} = \frac{a-c}{a} , a+b+c =3 $, which leads to the same cases.
Calvin Lin
  • 68,864
  • Your solution is fantastic and I want to ask a question. I tried to expand the problem and wondered if there is another number k>4/3 and replace for 4/3 so the inequality still true. Do you whether that k exists? – Tmt Feb 11 '22 at 02:27
  • @Tmt No, because of the non-constant equality case which makes the RHS non-zero. Thus 4/3 is the maximum possible constant. $\quad$ Please accept answers that you think are great. – Calvin Lin Feb 11 '22 at 04:25
  • Very nice answer. Btw, in your spoiler, you should have $\frac{(a-b)^2}{b}$, no? – V.S.e.H. Feb 14 '22 at 19:26
  • @V.S.e.H. Ah yes, fixed. – Calvin Lin Feb 14 '22 at 22:25
1

We still can WLOG $a\ge b\ge c$, let $a=c+s+t$ and $b=c+s$, multiply both sides by $3abc$ and write it as

$$(a+b+c)(a^3c+b^3a+c^3b) \ge abc(a+b+c)^2 + 4abc(a-c)^2$$

Left hand side is

$$ct^4+5cst^3+6c^2t^3+s^3t^2+12cs^2t^2+24c^2st^2+13c^3t^2+3s^4t+20cs^3t+45c^2s^2t+43c^3st+15c^4t+2s^5+13cs^4+33c^2s^3+43c^3s^2+30c^4s+9c^5$$

Right hand side is

$$5cst^3+5c^2t^3+17cs^2t^2+28c^2st^2+11c^3t^2+20cs^3t+50c^2s^2t+45c^3st+15c^4t+8cs^4+28c^2s^3+41c^3s^2+30c^4s+9c^5$$

So left minus right is

$$2t^2c^3-2stc^3+2s^2c^3+t^3c^2-4st^2c^2-5s^2tc^2+5s^3c^2+t^4c-5s^2t^2c+5s^4c+s^3t^2+3s^4t+2s^5$$

We know this equals to

$$(c+t+s)(ct-cs-s^2)^2+(c+s)(ct-ts-s^2)^2+c(t^2-cs-s^2)^2\ge 0$$

Or, hiding all the things above, we have

$$(a+b+c)(a^3c+b^3a+c^3b) \ge abc(a+b+c)^2 + 4abc(a-c)^2$$

equals to

$$a(ac-b^2)^2+b(ab-2ca+c^2)^2+c(a^2-2ab+bc)^2\ge 0$$

So, the proof is done.

The equality case is, besides $a=b=c$, we have $c=t=\frac{1+\sqrt 5}2 s$, which means $a=\frac{3}{2}$, $b=\frac{3\sqrt 5-3}{4}$ and $c=\frac{9 -3\sqrt 5}{4}$.


Below are the proof of $$\frac{a^3}{b}+\frac{b^3}{c} +\frac{c^3}{a} \ge3 + \frac{4}{3}\max\{(a-b)^2;(b-c)^2;(c-a)^2\}$$

And the person who asked this changed the problem.

WLOG $a\ge b\ge c$ (convince yourself that if you exchange $b,c$, the LHS gets larger). Now, brute force.

Let $a=c+s+t$ and $b=c+s$, multiply $abc$ at both sides and making $3$ into $(a+b+c)^2/3$, we have

Left hand side is $a^4c+b^4a+c^4b$, plugging $a=c+s+t$ and $b=c+s$ we have

$$ct^4+4cst^3+4c^2t^3+6cs^2t^2+12c^2st^2+6c^3t^2+s^4t+8cs^3t+18c^2s^2t+16c^3st+5c^4t+s^5+6cs^4+14c^2s^3+16c^3s^2+10c^4s+3c^5.$$

Right hand side is $$abc(\frac{(a+b+c)^2}3+4/3(s+t)^2),$$

which is

$$(5cst^3+5c^2t^3+17cs^2t^2+28c^2st^2+11c^3t^2+20cs^3t+50c^2s^2t+45c^3st+15c^4t+8cs^4+28c^2s^3+41c^3s^2+30c^4s+9c^5)/3$$

You can convince yourself again that LHS minus RHS is all non-negative.

Remark: Playing with Geogebra, $4/3$ can be raised to around $2.445$. The case $4/3$ change to $2$ is also easy.

Calvin Lin
  • 68,864
JetfiRex
  • 2,451
  • I haven’t check the calculation yet, but it seem like when you multiply abc on the LHS, its degree will be 4, not 5 so did you multiply the LHS with another value, such as (a+b+c) – Tmt Feb 10 '22 at 03:23
  • @Tmt Why the degree is four? $a^3/b\times abc=a^4c$, it is degree 5. – JetfiRex Feb 10 '22 at 03:24
  • Oh I’m so sorry, I don’t know why the LHS turn into a^3/b because the original problem was a^2/b – Tmt Feb 10 '22 at 03:27
  • @Tmt never mind, still can be proved. – JetfiRex Feb 10 '22 at 04:19