2
$\begingroup$

So I was reading a paper which made the claim "It is easy to see that $\frac{1-e^{-\alpha}}{\alpha} > 1-\frac{\alpha}{2} > \frac{1}{1+\alpha}$ when $0 < \alpha < 1$."

Verifying that $1-\frac{\alpha}{2} > \frac{1}{1+\alpha}$ only involves some simple algebra, but in order to prove that $\frac{1-e^{-\alpha}}{\alpha} > 1-\frac{\alpha}{2}$ I had to use what I have only seen referred to as the "racetrack theorem\principle" from elementary calculus twice.

Namely, if $f(0) = g(0)$ and $f'(x) \geq g'(x)$ for all $x \geq 0$, then $f(x) \geq g(x)$ for all $x \geq 0$.

Taking derivatives involving the quotient rule is somewhat of a pain in the ass, so I was wondering if there were any slick ways of proving this inequality? Or maybe I missed a far simpler way of proving this inequality?

$\endgroup$
1
  • 2
    $\begingroup$ This follows from $e^x < 1+x+\frac{x^2}2$ for all $x<0$. $\endgroup$
    – Théophile
    Commented Dec 22, 2022 at 23:26

2 Answers 2

1
$\begingroup$

First, let's rearrange $\frac{1-e^{-\alpha}}{\alpha} > 1-\frac{\alpha}{2}$ to $e^{-\alpha} < 1 - \alpha + \frac{\alpha^2}{2}$ and try proving this second inequality.

We have an infinite series for $e^{-\alpha}$: it is $1 - \alpha + \frac{\alpha^2}{2!} - \frac{\alpha^3}{3!} + \frac{\alpha^4}{4!} - \cdots$ where the $k^{\text{th}}$ term is $(-1)^k \frac{\alpha^k}{k!}$. (This is one of the definitions of $e^x$.) When $0 < \alpha < 1$, the terms of this sum are decreasing in absolute value: we raise $\alpha$ to a higher power, and on top of that we increase the denominator. Therefore:

  • Stopping the sum after a positive term is an overestimate: we can group all the tail terms we dropped into consecutive pairs that are less than $0$.
  • Stopping the sum after a negative term is an underestimate: we can group all the tail terms we dropped into consecutive pairs that are greater than $0$.

In particular, stopping the sum after $\frac{\alpha^2}{2}$ is an overestimate: $e^{-\alpha} = 1 - \alpha + \frac{\alpha^2}{2}$.

$\endgroup$
0
$\begingroup$

Let $f(\alpha) = 1 - e^{-\alpha}, g(\alpha) = \alpha - \dfrac{\alpha^2}{2}, 0 \le \alpha < 1$. Check that $f(0) = g(0) = 0$, and also $f'(\alpha) = e^{-\alpha}, g'(\alpha) = 1-\alpha$, and $f''(\alpha) = -e^{-\alpha}, g''(\alpha) = -1$. Obverse that $e^{\alpha} \ge 1 \implies e^{-\alpha} \le 1\implies -e^{-\alpha} \ge -1\implies f''(\alpha) \ge g''(\alpha)$, and $f''(0) = g''(0)=1$. Thus apply the racetrack theorem twice to conclude: $f'(\alpha) \ge g'(\alpha)\implies f(\alpha) \ge g(\alpha)\implies \dfrac{1-e^{-\alpha}}{\alpha} > 1 - \dfrac{\alpha}{2}, \alpha > 0$. Done.

$\endgroup$
4
  • 1
    $\begingroup$ So this was the method that I used to prove it, I was asking for alternative methods of proof. $\endgroup$ Commented Dec 22, 2022 at 23:13
  • $\begingroup$ Note that it is the same racetrack method but the function is different. It does not involve quotient which you complained about. $\endgroup$
    – Wang YeFei
    Commented Dec 22, 2022 at 23:14
  • $\begingroup$ While I appreciate your response, I was looking for an answer which did not rely on the theorem that I already used to prove it. $\endgroup$ Commented Dec 22, 2022 at 23:16
  • $\begingroup$ You might want to consider the difference of the two but first multiply the alpha to get the quadratic function first and use traditional derivative to take care of it perhaps the fastest route now. $\endgroup$
    – Wang YeFei
    Commented Dec 22, 2022 at 23:22

You must log in to answer this question.

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