3

First I tried to use integration: $$y=\lim_{n\to\infty}\frac{a^n}{n!}=\lim_{n\to\infty}\frac{a}{1}\cdot\frac{a}{2}\cdot\frac{a}{3}\cdots\frac{a}{n}$$ $$\log y=\lim_{n\to\infty}\sum_{r=1}^n\log\frac{a}{r}$$ But I could not express it as a riemann integral. Now I am thinking about sandwich theorem.

$$\frac{a}{n!}=\frac{a}{1}\cdot\frac{a}{2}\cdot\frac{a}{3}\cdots\frac{a}{t} \cdot\frac{a}{t+1}\cdot\frac{a}{t+2}\cdots\frac{a}{n}=\frac{a}{t!}\cdot\frac{a}{t+1}\cdot\frac{a}{t+2}\cdots\frac{a}{n}$$ Since $\frac{a}{t+1}>\frac{a}{t+2}>\frac{a}{t+1}>\cdots>\frac{a}{n}$ $$\frac{a^n}{n!}<\frac{a^t}{t!}\cdot\big(\frac{a}{t+1}\big)^{n-t}$$ since $\frac{a}{t+1}<1$, $$\lim_{n\to\infty}\big(\frac{a}{t+1}\big)^{n-t}=0$$ Hence, $$\lim_{n\to\infty}\frac{a^t}{t!}\big(\frac{a}{t+1}\big)^{n-t}=0$$ And by using sandwich theorem, $y=0$. Is this correct?

Aditya Dev
  • 5,004

6 Answers6

11

You missed the lower slice of bread in your sandiwch :)

Assuming that $a>0$, you should make clear that all terms are positive, no matter how obvious seems to be.

You should also write explicitly that there is some $t\in\Bbb N$ such that $t>a$. This is called "Archimedean property" of real numbers.

This would be my proof (I insist, assuming that $a>0$):

There exists some natural $t$ such that $t>a$. Then, for $n> t$ $$0<\frac{a^n}{n!}=\frac{a^t}{t!}\frac{a^{n-t}}{(t+1)\cdots n}<\frac{a^t}{t!}\left(\frac at\right)^{n-t}$$

Since $a/t<1$, the rightmost expression tends to $0$, and hence, by the sandwich theorem $$\frac{a^n}{n!}\to 0$$

Remark: If $a\le 0$, the limit is still $0$, but in this case you should use this fact:

If $a_n$ is a sequence of real numbers such that $\lim |a_n|=0$ then $\lim a_n=0$.

ajotatxe
  • 66,849
7

This is the shortest proof

$\displaystyle \sum _{n=1} ^{\infty} \frac{a^{n}}{n!}$ converges by ratio test.

Let $x_{n}=\frac{a^{n}}{n!} $

Then the convergence of $\displaystyle \sum _{n=1} ^{\infty }x_{n} $ implies $\lbrace x_{n} \rbrace $ converges to zero

6

What is,

$$\lim_{n \to {\infty}} {{a^n} \over {n!}}$$

Well, for large values of $n$, $n!$ can be evaluated with Stirling's Approximation.

$$\lim_{n \to {\infty}} {{a^n} \over {n!}}=\lim_{n \to {\infty}} {{a^n} \over {\sqrt{2 \pi n} \cdot (n/e)^n}}$$

$${{a^n} \over {\sqrt{2 \pi n} \cdot (n/e)^n}} = {1 \over {\sqrt{2 \pi n}}} \cdot a^n \cdot (e/n)^n = {1 \over {\sqrt{2 \pi n}}} \cdot \left({{a \cdot e } \over {n}} \right)^n$$

The value of the limit is clearly $0$ for any finite $a$, $$\lim_{n \to {\infty}} {{a^n} \over {\sqrt{2 \pi n} \cdot (n/e)^n}} = 0$$ This is because the square root term goes to zero, and because the term inside the parentheses must be less than 1 since $a$ is finite.

Thus,

$$\lim_{n \to {\infty}} {{a^n} \over {n!}}=0$$

Zach466920
  • 8,419
  • Thanks for mentioning "Stirling's Approximation". But I dont understand why you neglected $\sqrt{2\pi n}$. – Aditya Dev Aug 18 '15 at 17:53
  • 1
    @AdityaDev I already added it back in, it was definitely needed for completeness :) – Zach466920 Aug 18 '15 at 17:53
  • for a finite $a$, $\big(\frac{a\cdot e}{n}\big)^n$ is like $0^{\infty}$ which is an indeterminate form. – Aditya Dev Aug 18 '15 at 17:54
  • @AdityaDev The limit is not indeterminate. $\lim_{n \to {\infty}} 0^n=0$ See wolfram – Zach466920 Aug 18 '15 at 17:56
  • I am sorry. I confused it with $\infty^0$. – Aditya Dev Aug 18 '15 at 18:00
  • 2
    I learned something new from this answer :) – Aditya Dev Aug 18 '15 at 18:00
  • @Zach466920, Aditya is right. The limit is not clearly 0. It is an indeterminate form, since the basis approaches 0 and the exponent approaches $\infty$. Your justification that $\lim_{n\to\infty} 0^n=0$ implies that you were allowed to take the limit of the basis first and then take the limit as the exponent approaches $\infty$. This is not what happens in reality, as both basis and exponent vary together. Furthermore, there's a square root in the denominator..

    Although your claim is right, for the sake of those who will look at this in the future, I would add a few more steps...

    – bartgol Aug 18 '15 at 18:45
  • @bartgol Actually no. The term ${{a \cdot e} \over n}$ becomes less than $1$ long before $n$ is infinite, so you're justified in taking the limit as above. The square root term in the denominator clearly goes to zero as well. Since you're multiplying a term that must be less than 1 by a term that goes to 0, it's obvious the limit goes to 0. In other words, I will not be whipping the epsilon-delta tools out... – Zach466920 Aug 18 '15 at 18:54
  • Forget about the denominator, I don't know what I was thinking about there.

    As for the other issue, you are then using the squeeze theorem! That is, you are comparing the sequence $(a\cdot e/n)^n$ with some $c^n$ (for $n$ large enough and $c<1$). Sure, that's correct. I just believe that people who read this thread in search for an answer may not see that intermediate step... ;-)

    – bartgol Aug 18 '15 at 19:14
  • @bartgol I don't know about that, I just note that the term inside the parentheses, by the assumption $a$ is finite, must be less than 1. Zero times a number less than 1 is 0, you don't have to use the squeeze theorem. Btw, I edited anyway. – Zach466920 Aug 18 '15 at 19:19
  • Wait, where do you see "0 times a number less than 1"? There's never "0 times anything" there. – bartgol Aug 18 '15 at 19:57
  • @bartgol actually, it's explicitly in my answer, look again, I separate according to the terms. – Zach466920 Aug 18 '15 at 20:03
  • Ok, I think I went too long with this (stackexchange is scolding me). I still believe that the limit you claim to be zero (and it is 0), is not immediate for the readers that read this thread looking for an answer. After all, that limit is indeterminate ($\infty/\infty$). But as long as people understand why that is 0, it's fine with me. – bartgol Aug 18 '15 at 21:12
  • You don't need anything anywhere near as onerous to prove as Stirling's formula in order to find this limit. – Michael Hardy Dec 24 '17 at 01:09
3

You can prove it as follows:

for every $\varepsilon >0$ and $m+1>\left| a \right| $ and if $n$ is big enough then $$0<\left| \frac {a^n}{ n! } \right| =\frac { \left| a \right| }{ 1 } \cdot \frac { \left| a \right| }{ 2 } \cdots \frac { \left| a \right| }{ m } \cdot \frac { \left| a \right| }{ m+1 } \cdots \frac { \left| a \right| }{ n } <\frac { { \left| a \right| }^m }{ m! } { \left( \frac { \left| a \right| }{ m+1 } \right) }^{ n-m }<\varepsilon $$

haqnatural
  • 22,026
3

Let $x_n=\frac{a^n}{n!}$.

$$\left|\frac{x_{n+1}}{x_n}\right|=\frac{\frac{a^{n+1}}{(n+1)!}}{\frac{a^n}{n!}}=\frac{a^{n+1}n!}{a^n (n+1)!}= \frac{a}{n+1}\underset{n\to \infty }{\longrightarrow }0$$

and thus, by $x_n\to 0$ by Ratio test.

idm
  • 12,049
1

Here's an easy way that I'm surprised is not yet here:

$$ \frac{a^n}{n!} = \frac{\overbrace{a\cdots\cdots\cdots\cdots a}}{\underbrace{1\cdot2\cdot3\cdots\,\cdots n}} $$ When $n$ reaches the point of being twice as big as $a,$ then every time you increment $n$ by $1$ after that, you multiply the numerator by $a$ and the denominator by more than $2a,$ so the fraction gets multiplied by something whose absolute value is less than $1/2.$ Multiplying by something less than $1/2$ over and over again will give you a product approaching $0.$