3
$\begingroup$

I am trying to solve the following question, but I did not reach to any answer, I would be sol glad if anyone could help me on that.

If a, b are positive numbers such that a + b = 1 prove that for all x,

$ae^{\frac{x}{a}} + be^{-\frac{x}{b}}\leq e^{\frac{x^2}{8a^2b^2}}$

Thank you everyone !!!

$\endgroup$
  • $\begingroup$ @Michael would you please devote some time to prove this inequality. I would really appreciate it if you mind helping me or giving me some hints on how to deal with inequalities like the one that I posted above. $\endgroup$ – alfred noble Oct 25 '16 at 6:15
  • $\begingroup$ @Michael, if we consider the case where $a = b = \frac{1}{2}$ then we have: $cosh(2x) \le e^{2x^2}$. I would work on proving the latter inequality, this method may help to prove the main inequality $\endgroup$ – alfred noble Oct 25 '16 at 6:25
  • $\begingroup$ @πr8, I read your answer which proves this above inequality when $a = b = \frac{1}{2}$, now I would deeply appreciate it if you could help prove the above inequality in general form $\endgroup$ – alfred noble Oct 25 '16 at 9:26
  • $\begingroup$ Where did you come across this interesting inequality? $\endgroup$ – Thomas Ahle Aug 17 '18 at 16:19
5
$\begingroup$

It looks to be neat but true. Denote $x=abt$, we rewrite inequality as $ae^{bt}+be^{-at}\leqslant e^{t^2/8}$. We may suppose $t>0$, else replace $t$ to $-t$ and $b$ to $a$. Multiply by $e^{at}$ and rewrite as $ae^t+1-a\leqslant e^{at+t^2/8}$. Fix $t$ and vary $a\in [0,1]$. We should consider the minimal value of $H(a)=e^{at+t^2/8}-ae^t-1+a$ and prove that it is non-negative. This minimal value is attained either for $a=0$, or $a=1$, or for such $a$ that $H'(a)=0$. Obviously $H(0)\geqslant 0$, $H(1)\geqslant 0$, thus it remains to consider such $a$ that $H'(a)=0$. That is, $e^t-1=te^{at+t^2/8}$, $at=\log((e^t-1)/t)-t^2/8$. Now the inequality may be rewritten (multiply it by $t$ and substitute the values for $at$ and for $te^{at+t^2/8}$) as $$(e^t-1)\left(\log\frac{e^t-1}t-\frac{t^2}8\right)+t\leqslant e^t-1,$$ Divide by $e^t-1$ and note that for $t=0$ the equality takes place, so it suffices to prove that $$\left(\log\frac{e^t-1}t-\frac{t^2}8+\frac{t}{e^t-1}\right)'\leqslant 0,\,\forall t\geqslant 0.$$ Taking derivative and multiplying by $4t(e^t-1)^2$ we get (miracle!) $$ -(e^t (t-2)+t+2)^2\leqslant 0. $$

$\endgroup$
  • $\begingroup$ thanks a lot. But I have two questions about your answer. I) In the middle of your answer you took the derivatives of both sides of the inequality with respect to $a$ and you calculated the value of $a$ for which the derivations of both sides of the inequality is the same. I want to know why did you do that ? II) Are you saying that since you reached to an ever true inequality ($-(e^t(t-2) + t + 2)^2 \le 0$) the main inequality that I posted above is also true ? $\endgroup$ – alfred noble Oct 26 '16 at 6:41
  • $\begingroup$ If we need to prove that $H(a)\geqslant 0$ for some function $H$ (here $H=$RHS-LHS of the inequality) and $a\in [0,1]$, we look at a point $a\in [0,1]$, for which $H$ takes minimal value. It is either an endpoint of the segment, or a point where derivative $H'(a)$ equals 0. $\endgroup$ – Fedor Petrov Oct 26 '16 at 6:55
  • $\begingroup$ Yes, I think that this last miraculous identity for the derivative finishes the proof. $\endgroup$ – Fedor Petrov Oct 26 '16 at 6:58
  • $\begingroup$ This is the link to the prove of the above inequality in special case when $a = b = \frac{1}{2}$ I think it might be useful to prove the above inequality in the general form. LINK:math.stackexchange.com/questions/331367/cosh-x-inequality $\endgroup$ – alfred noble Oct 26 '16 at 16:01
  • 1
    $\begingroup$ Additional requirement is $f(0)=0$. For $f(x)=1/x$ it is not true, but for $-1+\log\frac{e^x-1}x-x^2/8+\frac{x}{e^x-1}$ it is true. $\endgroup$ – Fedor Petrov Oct 30 '16 at 11:26

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.