6
$\begingroup$

Question: For $a,b,c,d > 0$ and $abcd = 1$, show that $$\frac{1}{a} + \frac{1}{b} + \frac{1}{c} + \frac{1}{d} + \frac{12}{a + b + c + d} \geq 7$$

My Attempts: Trivial to see that equality occurs for $a = b= c = d$ - so this indicates to use some inequality wherin the equality condition holds when all quantities are equal - I've tried AM-GM-HM, Power means, Rearrangement, Cauchy-Schwarz, Newton, Maclaurin, Weighted AM-GM - it doesn't help, and the reason is that these inequalities obtain a function of $a,b,c,d$ on the RHS whose minimum is $<7$ - and that minimum and the inequality can never hold simultaneously - giving a loose bound.

I've tried rearranging the inequality to the form $(\sum_{cyc} a )(\sum_{cyc} \frac{1}{a}) - 7 (\sum_{cyc} a) + 12 \geq 0 $ - This doesn't help. I've tried the substitution $(a,b,c,d) \rightarrow (\frac{1}{x}, \frac{1}{y}, \frac{1}{z}, \frac{1}{w} )$ - this is of no help either. Even tried the substitution $(a,b,c,d) \rightarrow (\frac{x}{y}, \frac{y}{z}, \frac{z}{w}, \frac{w}{x} )$ - of no help either. Writing it as $\frac{4}{M(-1)} + \frac{3}{M(1)} \geq 7$ where $M(x)$ is the power means function, and hoping to use some property of $M(x)$ doesn't work either.

My Analysis: Suppose $S_i$ denotes the sum taken $i$ at a time. It is trivial to see that $S_3 = \sum_{cyc} \frac{1}{a}$ is lower bounded by $S_4$ by Maclaurin or GM-HM. However $S_1$ is unbounded above even if $S_4 = 1$. However when $S_1$ reaches a very high value, so does $S_3$, so a minimum has to involve considering both terms simultaneously i.e. you can't minimize each separately and then hope to minimize the function.

Brute Force Solution: The brute force solution of considering it as a function of 4 variables and then optimizing obviously works, I'm hoping for something more smarter or elegant.

$\endgroup$
  • 2
    $\begingroup$ Great job telling us what you've already tried in the question statement! $\endgroup$ – Greg Martin Jun 5 '19 at 18:14
3
$\begingroup$

Mixing Variables and $uvw$ help.

I'll post my solution, which I found eight years ago.

Let $f(a,b,c,d)=\frac{1}{a}+\frac{1}{b}+\frac{1}{c}+\frac{1}{d}+\frac{12}{a+b+c+d}- 7$ and $a=\max\{a,b,c,d\}$.

Thus, $$f(a,b,c,d)-f\left(a,\sqrt[3]{bcd},\sqrt[3]{bcd},\sqrt[3]{bcd}\right)=$$ $$=\frac{bc+bd+cd-3\sqrt[3]{b^2c^2d^2}}{bcd}-\frac{12\left(b+c+d-3\sqrt[3]{bcd}\right)}{(a+b+c+d)\left(a+3\sqrt[3]{bcd}\right)}\geq$$ $$\geq\frac{bc+bd+cd-3\sqrt[3]{b^2c^2d^2}}{bcd}-\frac{12\left(b+c+d-3\sqrt[3]{bcd}\right)}{(\frac{b+c+d}{3}+b+c+d)\left(\frac{b+c+d}{3}+3\sqrt[3]{bcd}\right)}=$$ $$=\frac{bc+bd+cd-3\sqrt[3]{b^2c^2d^2}}{bcd}-\frac{27\left(b+c+d-3\sqrt[3]{bcd}\right)}{(b+c+d)\left(b+c+d+9\sqrt[3]{bcd}\right)}.$$ We'll prove that $$\frac{bc+bd+cd-3\sqrt[3]{b^2c^2d^2}}{bcd}-\frac{27\left(b+c+d-3\sqrt[3]{bcd}\right)}{(b+c+d)\left(b+c+d+9\sqrt[3]{bcd}\right)}\geq0.$$ Indeed, let $b+c+d=3u$, $bc+bd+cd=3v^2$ and $bcd=w^3$.

Thus, $$\frac{bc+bd+cd-3\sqrt[3]{b^2c^2d^2}}{bcd}-\frac{27\left(b+c+d-3\sqrt[3]{bcd}\right)}{(b+c+d)\left(b+c+d+9\sqrt[3]{bcd}\right)}\geq0$$ it's $ g(v^2)\geq0,$ where $g$ is a linear increasing function.

Id est, $g$ gets a minimal value, when $v^2$ gets a minimal value, which happens for equality case of two variables.

Since $g(v^2)\geq0$ is homogeneous inequality, it's enough to check one case only: $c=d=1$, which after substitution $b=x^3$ gives $$(x-1)^2(2x^7+x^6+18x^5-10x^4-50x^3+36x^2+26x+4)\geq0,$$ which is true.

Id est, $$f(a,b,c,d)\geq f\left(a,\sqrt[3]{bcd},\sqrt[3]{bcd},\sqrt[3]{bcd}\right)$$ and it's enough to prove that $f(a,b,b,b)\geq0$, where $a=\frac{1}{b^3}$, which gives $$(b-1)^2(3b^6+6b^5+9b^4-9b^3-5b^2-b+3)\geq0,$$ which is true.

$\endgroup$
  • $\begingroup$ I'll take me a while to go through this - but could you say give some insight as to how you formed your approach - was it a lucky idea or is there some theory behind it (cause I'm still a beginner, I wouldn't know) - i.e what ideas could I take back from this question? No offence it's a great method, but tbh optimizing over 4 variables would be more concise. $\endgroup$ – Kaind Jun 5 '19 at 18:37
  • 1
    $\begingroup$ @Kaind It was optimizing over 4 variables. There are very many methods of inequalities proofs, but non of those methods are working in the general. In our case the Mixing Variables method helps. $\endgroup$ – Michael Rozenberg Jun 5 '19 at 18:43
  • $\begingroup$ There's something called a 'mixing variables method' - could you point me to some link or pdf which explains this? (By optimizing 4 variables, I meant finding saddle points and critical points, and then getting answer). $\endgroup$ – Kaind Jun 5 '19 at 18:45
  • 1
    $\begingroup$ For those questions visit the AOPS website. $\endgroup$ – Dr. Sonnhard Graubner Jun 5 '19 at 18:52
  • 1
    $\begingroup$ @Kaind About MV you can read here: math.stackexchange.com/questions/tagged/mixing-variables Now I see that Lagrange multipliers helps, but for the proof we need to consider two cases: 1. $b=c=a$; 2. $b=a$ and $c=d$. $\endgroup$ – Michael Rozenberg Jun 5 '19 at 18:52

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.