3

So recently a friend asked me to compute this limit:

$$\lim_{n\to \infty }\left(\sqrt[\leftroot{-2}\uproot{2}n+1]{(n+1)!}-\sqrt[\leftroot{-2}\uproot{2}n]{n!}\right)$$

Question : Does the limit exist? If yes is it finite and if yes what is its value?

How do we solve this?

Edit:

Note: I am only familiar with only basics of limit solving(upto L'Hôpital's rule) and have reasons to believe that this limit can be solved using these methods. If you could keep your answer simple that should help.

Update

Here is where I have gotten so far

$$ \lim_{n\to \infty} (n+1)! ^{1\over n+1} - (n)! ^{1\over n}$$ Can be written as $$ \lim_{n\to \infty}[1*2*3*...(n+1)]^{1\over n+1} - [1*2*3*...n] ^{1\over n}$$

$$\implies \lim_{n\to \infty} [(n+1)[{1 \over n+1}* {2 \over n+1} * {3\over n+1}...* {n+1 \over n+1}]^{1 \over n+1} - (n)[{1 \over n}* {2 \over n} * {3\over n}...* {n\over n}]^{1 \over n} ]$$

(Factoring n+1 out of first expression and n from second.)

$$\implies \lim_{n\to \infty} [(n+1) e^{{1 \over n+1} (\sum_{r=1}^{n+1}ln({r\over n+1}))} - (n) e^{{1 \over n} (\sum_{r=1}^{n}ln({r\over n}))} ]$$

From here I think second limit can be solved as a integral(limit of a sum) but I cannot solve first. How can I proceed further?

Thanks!

4 Answers4

7

We can provide a more elementary approach, assuming the limit exists.

By Stolz-Cesaro, the discrete version of L'Hôpital's rule, we have

$$\lim_{n\to\infty}\frac{\sqrt[n+1]{(n+1)!}-\sqrt[n]{n!}}1=\lim_{n\to\infty}\frac{\sqrt[n]{n!}}n=\lim_{n\to\infty}\sqrt[n]{\frac{n!}{n^n}}$$

and by the root-to-ratio limit, we also have

$$\lim_{n\to\infty}\sqrt[n]{\frac{n!}{n^n}}=\lim_{n\to\infty}\frac{(n+1)!}{(n+1)^{n+1}}\times\frac{n^n}{n!}=\lim_{n\to\infty}\frac1{\left(1+\frac1n\right)^n}$$

which, by the limit definition of $e$, is given by

$$\lim_{n\to\infty}\sqrt[n+1]{(n+1)!}-\sqrt[n]{n!}=\frac1e$$

5

Brute force, but from the Stirling formula $$ n! = \left( {\frac{n}{e}} \right)^n \sqrt {2\pi n} \left( {1 + \mathcal{O}\!\left( {\frac{1}{n}} \right)} \right), $$ one has $$ \sqrt[n]{{n!}} = \frac{n}{e}\exp \left( {\frac{1}{2n}\log (2\pi n)} \right)\left( {1 + \mathcal{O}\!\left( {\frac{1}{{n^2 }}} \right)} \right) = \frac{n}{e} + \frac{1}{2e}\log (2\pi n) + \mathcal{O}\!\left( {\frac{{\log ^2 n}}{n}} \right). $$ This gives $$ \sqrt[{n + 1}]{{(n + 1)!}} - \sqrt[n]{{n!}} = \frac{1}{e} + \frac{1}{2e}\log \left( {\frac{{n + 1}}{n}} \right) + \mathcal{O}\!\left( {\frac{{\log ^2 n}}{n}} \right) = \frac{1}{e} + \mathcal{O}\!\left( {\frac{{\log ^2 n}}{n}} \right). $$ Thus the limit is $1/e$.

Gary
  • 31,845
2

Calculation of the limit expressing the difference in the statement by strings whose limit is known. $$\sqrt[n+1]{(n+1)!}-\sqrt[n]{n!} =\sqrt[n]{n!}\left(\frac{\sqrt[n+1]{(n+1)!}}{\sqrt[n]{n!}}-1\right) =\sqrt[n]{n!}\cdot\frac{x_{n}-1}{\ln{x_{n}}}\cdot\ln\sqrt[n(n+1)]{\frac{(n+1)^n}{n!}}= $$ $$=\frac{\sqrt[n]{n!}}{n+1}\cdot\frac{x_{n}-1}{\ln{x_{n}}} \cdot\ln\frac{n+1}{\sqrt[n]{n!}}\rightarrow \frac{1}{e}\cdot1\cdot\ln{e}=\frac{1}{e}$$

PinkyWay
  • 4,565
medicu
  • 4,482
  • This is the string of Traian Lalescu, a Romanian mathematician.The problem may be solved on this site – medicu Mar 28 '20 at 18:14
  • Use \left(\frac{sth}{sth}\right) for the output $\left(\frac{sth}{sth}\right)$. (: – PinkyWay Mar 28 '20 at 18:19
  • 1
    ms._VerkhovtsevaKatya. Thank you! – medicu Mar 28 '20 at 18:27
  • 1
    Medicu, you're welcome! – PinkyWay Mar 28 '20 at 18:28
  • I understand the first line but lost it when you introduced logarithms(probably because I do not how strings work). Sorry for that. Meanwhile I changed my approach a bit(as updated in question) can you extend on those lines? Thanks for the answer btw. :) – Hrishabh Nayal Mar 29 '20 at 02:04
  • 1
    @HrishabhNayal: you should note that $$x_n=\dfrac{\sqrt[n+1]{(n+1)!}}{\sqrt[n]{n!}}=\left(\frac{((n+1)!)^n}{(n!)^{n+1}}\right)^{1/n(n+1)}=\left(\frac{(n+1)^n}{n!}\right)^{1/n(n+1)}=\sqrt[n+1]{\frac{n+1}{\sqrt[n] {n!} }}$$ and further that $x_n\to 1$ and $x_n^{n+1}\to e$. These limits are proved using root to ratio theorem used in another answer which itself is based on Cesaro-Stolz. – Paramanand Singh Mar 29 '20 at 02:37
  • @Paramanand Singh Why does $x_n \rightarrow 1$ and $x_n^{n+1} \rightarrow e$. I cannot see it. – Hrishabh Nayal Mar 29 '20 at 07:40
  • @medicu why does $=\frac{\sqrt[n]{n!}}{n+1}$ tend to $1\over e$? – Hrishabh Nayal Mar 29 '20 at 07:42
  • 2
    @HrishabhNayal: here is one simple approach to $$x_n^{n+1}=\dfrac{n+1}{\sqrt [n] {n!}} =\dfrac{n+1}{n}\cdot\dfrac{n}{\sqrt[n] {n!}} $$ The limit is same as that of $n/\sqrt[n] {n!} $ and taking logs this expression becomes $$-\frac{1}{n}\sum_{k=1}^{n}\log\frac{k}{n}$$ which tends to $-\int_{0}^{1}\log x, dx=1$ and therefore original limit is $e$. – Paramanand Singh Mar 29 '20 at 10:00
  • @Paramanand Singh: Your comments are quick and easy to understand. Thank you! – medicu Mar 29 '20 at 10:37
  • @Hrishabh Nayal :The limit in the last one comment is explained in the second answer to this question using the root criteria (Cauchy-D'Alembert). – medicu Mar 29 '20 at 10:48
  • By the way I have given my +1 already. – Paramanand Singh Mar 29 '20 at 11:23
1

So I'm providing a first-principles approach, after seeing that you're unable to understand most of the answers that were posted.

We know that $$n! = n \cdot (n-1) \cdot (n-2) \cdot \cdot \cdot \cdot 2 \cdot 1$$ Taking a logarithm both sides, we get $$\ln n! = \sum_{r=1}^n\ln r$$ Looking at this summation and comparing it with the $\ln x$ function, we see that they are pretty similar, especially as $n \to \infty$ graph Since they are so similar, we can approximately replace the summation with a definite integral as follows: $$\ln n! \approx \int_1^n \ln x dx = n\ln n - n + 1$$

This is the principle behind the Stirling approximation, the formula that Gary used in his answer. It's a rough approximation: a more precise one uses the Euler-MacLaurin series but since our limit involves the difference of two of these terms, we can use this approximation as the error would be subtracted.

Taking an exponent and the nth root both sides gives us a very usable form of the term $(n!)^\frac 1n$ $$n! = \left( \frac ne \right)^ne \implies (n!)^\frac 1n = \frac ne \cdot e^\frac 1n$$

Now we can simply substitute the values of $n+1$ and $n$ in the obtained function and find their difference to obtain the limit. This is a fairly simple task and makes use of the taylor series expansion of $e^x$, which you should know.

$$L = \underset{n \to \infty}{\lim} \space \sqrt[\leftroot{-2}\uproot{2}n+1]{(n+1)!}-\sqrt[\leftroot{-2}\uproot{2}n]{n!} = \frac {n+1}e\cdot\sqrt[\leftroot{-2}\uproot{2}n+1]{e} - \frac {n}e\cdot\sqrt[\leftroot{-2}\uproot{2}n]{e}$$

$$ = \lim_{n \to \infty} \frac ne \left( \sqrt[\leftroot{-2}\uproot{2}n+1]{e} - \sqrt[\leftroot{-2}\uproot{2}n]{e}\right) + \frac {\sqrt[\leftroot{-2}\uproot{2}n]{e}}e$$

let $t \to 0 \implies n = \frac 1t$.

$$L = \lim_{t \to 0} \frac 1{et}\left( e^\frac{t}{t+1} - e^t \right) + \frac{e^\frac{t}{t+1}}{e}$$

Applying a taylor series expansion for $e^x$, we get

$$L = \lim_{t \to 0} \frac 1{et}\left( 1 + \frac t{t+1} + \frac {t^2}{(t+1)^2 \cdot 2!} + ... - 1 - t - \frac {t^2}{2!} - ...\right) + \frac {1 + \frac t{t+1} + \frac {t^2}{(t+1)^2 \cdot 2!} + ... }e$$ $$ = \lim_{t \to 0} \frac 1{e}\left( \frac {-t}{t+1} + O(t^2) \right) + \frac 1e$$ $$L = \frac 1e$$

which is the final answer.

Aniruddha Deb
  • 4,345
  • 11
  • 30
  • I understand how this approximation $$\ln n! \approx \int_1^n \ln x dx = n\ln n - n + 1$$ is valid graphically but I think I could do with a more formal proof. Is there a simpler way than Euler-MacLaurin series? – Hrishabh Nayal Mar 29 '20 at 06:23
  • you could use the Gamma function as shown in wikipedia: https://en.m.wikipedia.org/wiki/Stirling%27s_approximation – Aniruddha Deb Mar 29 '20 at 06:39
  • Actually the person I have to explain this answer is a IIT JEE aspirant(I assume you know what that explains about the knowledge of that person) so I am not sure he would understand Gamma function (let alone its integration). So if you could help me considering the knowledge of that person It would be very helpful . Btw +1 from me as you have really dumbed it down enough for me to understand :) – Hrishabh Nayal Mar 29 '20 at 06:46
  • @HrishabhNayal I am an IIT JEE Aspirant myself and while I do understand the Gamma function, I solved it without the Gamma Function/Euler-MacLaurin series and provided a simple enough version in my answer. I don't think a more formal proof is required for Stirling's approximation. – Aniruddha Deb Mar 29 '20 at 08:08
  • Well I might run into trouble explaining him why that approximation works. They rarely use these type approximation in their preparation right? – Hrishabh Nayal Mar 29 '20 at 08:22