4

Can somebody prove that this inequality is true for $0<x<n$?

$$ \frac{1-e^{-x^2}}{x^2}e^{-(x-n)^2}<\frac{2}{n^2}$$

I'm pretty much stuck.

Jack D'Aurizio
  • 353,855
  • 1
    Don't hesitate to try and explain what you've already tried to do, what failed, what seemed almost correct, etc. The more you show your investment, the better the community will respond to your post. – Azami May 25 '16 at 15:16

3 Answers3

2

We may notice that $\frac{1-e^{-x^2}}{x^2}\leq \frac{2}{2+x^2}$ for any $x\in\mathbb{R}$ and $e^{-(x-n)^2}\leq \frac{1}{1+(x-n)^2}$ for any $x\in[0,n]$, hence it is enough to prove that: $$ \forall x\in(0,n),\qquad \frac{2}{2+x^2}\cdot\frac{1}{1+(n-x)^2}<\frac{2}{n^2} \tag{1}$$ or: $$ \forall x\in(0,n),\qquad (2+x^2)\cdot(1+(n-x)^2) > n^2\tag{2} $$ But due to the Cauchy-Schwarz inequality we have:

$$ \forall x\in(0,n),\qquad (\sqrt{x^2+1}^2+1)(1+(n-x)^2)\geq \left(\sqrt{x^2+1}+n-x\right)^2 \tag{3}$$

so $(2)$ is trivial, $(1)$ follows and the claim is proved.

Jack D'Aurizio
  • 353,855
0

The first thing to try with this sort of inequality is to put $y=\frac{x}{n}$ and try to prove the resulting equality for $0<y<1$.

This seems a good idea in this case because multiplying both sides by $x^2$ will give a power of $\frac{x}{n}$ on the right-hand side, and on the left-hand side $(x-n)$ is $n$ times $(\frac{x}{n}-1)$.

So:

  1. Multiply both sides by $x^2$.
  2. Substitute $x=ny$.
  3. Simplify.
0

I wanted to find another answer different from the elegant one given by Jack D’Aurizio. Here is what I got. $$\frac{1-e^{-x^2}}{x^2}e^{-(x-n)^2}<\frac{2}{n^2}\iff \frac{e^{x^2}-1}{x^2}\lt\frac{2e^{x^2+(x-n)^2}}{n^2}$$ Let $$f(x)= \frac{e^{x^2}-1}{x^2}\\g(x)= \frac{2e^{x^2+(x-n)^2}}{n^2}$$ Notice first that $f$ and $g$ are both convex functions and that $g(0)=g(n)=\frac {2e^{n^2}}{n^2}$. Besides $$f(0)=\lim_{x\to 0}f(x)=1\lt \frac {2e^{n^2}}{n^2}=g(0)$$ furthermore $g$ takes its minimum at $x=\frac n2$ and $$f(\frac n2)=\frac{4(e^{\frac{n^2}{4}}-1)}{n^2}\lt \frac {2e^{\frac{n^2}{2}}}{n^2}=g(\frac n2)$$ (it is immediately verified that $2(e^{\frac x2}-1)\lt e^x$ for all x).

It follows by convexity of both functions and by $$f(0)\lt g(0) \text { and}\space f(\frac n2)\lt g(\frac n2)$$ that $$\color{red}{f(x)\lt g(x)\text { for}\space 0\lt x\le \frac n2}\qquad (*)$$ Besides $$f(n)= \frac{e^{n^2}-1}{n^2}\lt\frac {2e^{n^2}}{n^2}=g(n)$$ and similarly, by convexity, $f(\frac n2)\lt g(\frac n2)$ and $f(n)\lt g(n)$, we have

$$\color{red}{ f(x)\lt g(x)\text { for}\space \frac n2 \le x\lt n}\qquad (**)$$

Thus, by $(*)$ and $(**)$, $$ f(x)\lt g(x)\text { for}\space 0 \lt x\lt n$$

NOTE.-However that is not necessary in this case where the functions $f$ and $g$ are the same gender, it is better to add that $f(x)= g(x)\iff x=0$ to have fully valid argument used convexity.

Piquito
  • 29,594