2
$\begingroup$

The following occurred while working on some research project. Since the methods of proof I used were lengthy, I wish to see a skillful or insightful approach (perhaps even conceptual). Anyhow, here it is. Let $$f(x)=\left(\frac{x}{e^x-1}\right)^2 + \left(\frac{x+1}{e^{x+1}+1}\right)^2.$$ Can one give a short and elegant proof of these statements?

(1) $f(x)$ is a strictly decreasing function of $x$ over $\mathbb{R}$.

(2) In fact, the statement holds true if $e$ is replaced by any real number $t>1$.

$\endgroup$
6
  • 3
    $\begingroup$ I do not understand. Is this a problem or a challenge? In the second case, please note that MO is devoted to questions, and that challenges are not well-received here. $\endgroup$ Commented Jul 8, 2016 at 10:56
  • 2
    $\begingroup$ The last sentence makes it sound like you already know this to be true. If so, then it's not really a question for this site. It certainly may still be interesting and worth sharing, but in a paper / preprint / blog / problems column instead. $\endgroup$ Commented Jul 8, 2016 at 11:54
  • 1
    $\begingroup$ Perhaps you have confused us with that other place math.stackexchange.com , where problems with known answer are sometimes allowed. $\endgroup$ Commented Jul 8, 2016 at 12:20
  • 1
    $\begingroup$ In order to find a "short and elegant proof", it could be useful to know the OP's strategy of proof. $\endgroup$ Commented Jul 8, 2016 at 15:00
  • 4
    $\begingroup$ Okay, it seems clear after the last comment and edit this is not meant as a "challenge", any more than any other question. At least not in the sense of "I have a wonderful proof; can you match it?" Rather, OP is dissatisfied with his/her solution and wants to know if one can improve on it. So I will vote to reopen, but of course we don't know if we can do better without some idea of that solution. So I'd like to ask the OP to furnish more details of the solution at hand. $\endgroup$ Commented Jul 8, 2016 at 20:07

1 Answer 1

3
$\begingroup$

I do not know wether this helps you or not, but you may do as follows.

Denote $f(x)=(e^x-1)/x$. Note that $f'(x)=\frac{fe^x(x-1+e^{-x})}{x(e^x-1)}>0$, so $f$ is a positive increasing function.

Lemma. The function $1/f=x/(e^x-1)$ is a (positive decreasing) convex function.

Proof. $$ (1/f)''=\frac{e^x(2+x-(2-x)e^x)}{(e^x-1)^3}, $$ we have to check that it is non-negative. If $x\geqslant 2$, this is clear. If $0<x<2$, this reduces to $$ e^x\leqslant \frac{2+x}{2-x}=1+x+x^2/2+x^3/2^2+x^4/2^3+\dots,\,\,(1) $$ that holds coefficient-wise: $n!\geqslant 2^{n-1}$ for $n\geqslant 1$. Finally, the case $x<0$ reduces to $x>0$, since $2+x-(2-x)e^x$ and $2-x-(2+x)e^{-x}$ always have opposite signs.

Corollary. $g:=1/f^2=x^2/(e^x-1)^2$ is convex.

Proof. $-(1/f^2)'=2(-1/f)'(1/f)$, both multiples are positive decreasing functions, thus their product also decreases.

Now we claim that $g(x-a)+x^2/(e^x+1)^2$ decreases for each $a\geqslant 0$, for $a=1$ we get your statement (and for other $a$ something equivalent to your remark). Since $g'$ increases, we have $g'(x-a)\leqslant g'(x)$, so it suffices to check this for $a=0$. We have $g(x)+x^2/(e^x+1)^2=2x^2(e^{2x}+1)/(e^{2x}-1)^2$. Denote $2x=y$, we need to check that $y^2(e^y+1)/(e^y-1)^2$ decreases. Taking logarithmic derivative, this is equivalent to $$\frac{2}x+\frac{e^x}{e^x+1}-\frac{2e^x}{e^x-1}\leqslant 0.$$ For $x=-y>0$ we have $$\frac2{y}-\frac2{e^y-1}\geqslant \frac2y-\frac2{y+y^2/2}=\frac1{1+y/2}>\frac1{1+e^y}$$ as desired. For $x>0$ we have $$ \frac{2}x+\frac{e^x}{e^x+1}-\frac{2e^x}{e^x-1}\leqslant \frac{2+x}x-\frac{2e^x}{e^x-1}= \frac{(2-x)e^x-(2+x)}{x(e^x-1)}\leqslant 0 $$ by (1).

$\endgroup$
7
  • 3
    $\begingroup$ The second summand is not decreasing. $\endgroup$ Commented Jul 8, 2016 at 8:39
  • $\begingroup$ Oh, it is plus in denominator, not minus. Sorry. $\endgroup$ Commented Jul 8, 2016 at 9:06
  • 1
    $\begingroup$ As a rule of thumb, one should not answer off-topic questions (for instance, in order not to encourage them). $\endgroup$ Commented Jul 8, 2016 at 13:09
  • 2
    $\begingroup$ @FrancescoPolizzi sorry, this text was written before the question became officially considered an off-topic, or simultaneously. If you find it reasonable, I may remove it and send to OP elsewhere (I know him personally). I think, if the question is supposed to mean something like 'please explain this fact arising in my research with a short and elegant argument', it is suitable on MO, there are many such questions. $\endgroup$ Commented Jul 8, 2016 at 13:20
  • 1
    $\begingroup$ If you replace $e$ to $t=e^a$, then multiply by $a^2$ and denote $ax+a=y$, you get $g(y-a)+y^2/(e^y+1)^2$. $\endgroup$ Commented Jul 10, 2016 at 14:55

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .