6

Evaluating $$\lim\limits_{n\rightarrow \infty} \frac1{n^2}\ln \left( \frac{(n!)^n}{(0!1!2!...n!)^2} \right)$$

I'm not quite sure where to start in evaluating this. Some pointers, or a solution, would greatly be appreciated.

Did
  • 285,232
Trogdor
  • 10,619

3 Answers3

8

Since $$ \int_1^n\log(x)\,\mathrm{d}x\le\log(n!)\le\int_1^{n+1}\log(x)\,\mathrm{d}x\tag{1} $$ we have that $$ \log(n!)=n\log(n)-n+O(\log(n))\tag{2} $$ therefore, $$ \log\left(\frac{n^{n-1}}{n!}\right)=n+O(\log(n))\tag{3} $$


Note that $$ \left.\frac{(n!)^n}{(0!1!2!\cdots n!)^2}\middle/\frac{((n-1)!)^{n-1}}{(0!1!2!\cdots(n-1)!)^2}\right. =\frac{n^{n-1}}{n!}\tag{4} $$ Applying $(3)$, we have $$ \begin{align} \log\left(\frac{(n!)^n}{(0!1!2!\cdots n!)^2}\right) &=\sum_{k=1}^n\log\left(\frac{k^{k-1}}{k!}\right)\\ &=\sum_{k=1}^n(k+O(\log(k)))\\ &=\frac{n^2}2+O(n\log(n))\tag{5} \end{align} $$ Therefore, $$ \lim_{n\to\infty}\frac1{n^2}\log\left(\frac{(n!)^n}{(0!1!2!\cdots n!)^2}\right)=\frac12\tag{6} $$

robjohn
  • 355,927
7

No need to reference Barnes G-functions. These are integers after all.

$$\log{\left ( \frac{(n!)^n}{(0!1!2!...n!)^2} \right )} = n \log{n!} - 2 \sum_{k=0}^n \log{k!}$$

Now

$$\sum_{k=0}^n \log{k!} = \sum_{k=0}^n \log{\Gamma(k+1)} = \sum_{k=1}^{n+1} \log{\Gamma(k)}$$

Now use Euler-Maclurin:

$$\begin{align}\sum_{k=1}^{n+1} \log{\Gamma(k)}&= \int_0^{n+1} dx \, \log{\Gamma(x)} + \frac12 \left [\log{\Gamma(n+1)} - \log{\Gamma(1)} \right ] + o(n)\end{align}$$

Now, we use Raabe's integral, which is

$$ \int_{k}^{k+1} dx \, \log \Gamma(x) = \frac12 \log{(2 \pi)} + k \log{k}-k$$

so that

$$\begin{align} \int_{0}^{n+1} dx \, \log{\Gamma(x)} &= \sum_{k=0}^n \int_{k}^{k+1} dx \, \log{\Gamma(x)}\\ &= \frac{n+1}{2} \log (2 \pi) - \frac{n (n+1)}{2} + \sum_{k=1}^n k \log{k} \\ &= \frac{n+1}{2} \log (2 \pi) - \frac{n (n+1)}{2} + \log{(1^1 2^2 \cdots n^n)} \\ &= \frac{n+1}{2} \log (2 \pi) - \frac{n (n+1)}{2} + \log{\left [\frac{n!^n}{(n-1)!(n-2)!^2 \cdots 2^{n-2} 1^{n-1}} \right ]} \\ &= \frac{n+1}{2} \log (2 \pi) - \frac{n (n+1)}{2} + n \log{n!} - \sum_{k=1}^{n-1} \log{k!}\end{align}$$

Putting this all together, we find that

$$2 \sum_{k=1}^{n} \log{k!} = \frac{n+1}{2} \log (2 \pi) - \frac{n (n+1)}{2} + n \log{n!} + \frac{3}{2} \log{n!} +o(n) $$

so that, finally, using the fact that Stirling's approximation is equivalent to $\log{n!} = n \log{n} - n + o(n) $,

$$\lim_{n \to \infty} \frac1{n^2} \left ( n \log{n!} - 2 \sum_{k=1}^{n} \log{k!} \right ) = \frac12 $$

ADDENDUM

Raabe's integral is actually not all that hard to evaluate. Rewrite as

$$\int_k^{k+1} dx \, \log{\Gamma(x)} = \int_0^1 dx \, \log{\Gamma(x+k)} $$

$$\log{\Gamma(x+k)} = \log{\Gamma(x)} + \sum_{m=0}^{k-1} \log{(x+m)} $$

Thus,

$$\begin{align}\int_k^{k+1} dx \, \log{\Gamma(x)} &= \int_0^1 dx \, \log{\Gamma(x)} + \sum_{m=0}^{k-1} \int_0^1 dx \, \log{(x+m)}\\ &= \int_0^1 dx \, \log{\Gamma(x)} + \sum_{m=0}^{k-1} [(m+1) \log{(m+1)} - m \log{m} - (m+1)+m] \\ &= \int_0^1 dx \, \log{\Gamma(x)} + k \log{k} - k \end{align}$$

To evaluate the integral on the RHS, use the duplication formula:

$$\Gamma \left ( \frac{x}{2} \right ) \Gamma \left ( \frac{x+1}{2} \right ) = 2^{1-x} \sqrt{\pi} \Gamma(x) $$

so that

$$\log{\Gamma(x)} = \log{\left [\frac{\Gamma \left ( \frac{x}{2} \right ) \Gamma \left ( \frac{x+1}{2} \right )}{\sqrt{\pi} 2^{1-x}} \right ]} = -\frac12 \log{\pi} - \log{2} + \log{\Gamma \left ( \frac{x}{2} \right )} + \log{\Gamma \left ( \frac{x+1}{2} \right )} + x \log{2}$$

Thus,

$$\begin{align}\int_0^1 dx \, \log{\Gamma(x)} &= -\frac12 \log{(2 \pi)} + \int_0^1 dx \, \log{\Gamma \left ( \frac{x}{2} \right )} + \int_0^1 dx \, \log{\Gamma \left ( \frac{x+1}{2} \right )}\\ &= -\frac12 \log{(2 \pi)} + 2 \int_0^{1/2} dx \, \log{\Gamma(x)} + 2 \int_{1/2}^1 dx \, \log{\Gamma(x)} \\ &= -\frac12 \log{(2 \pi)} + 2 \int_0^{1} dx \, \log{\Gamma(x)} \end{align}$$

The result follows.

Ron Gordon
  • 142,110
  • 2
    Not sure why you said "No need to reference Barnes G-functions. These are integers after all." when you used a lot of machinery to get the answer. I hope you didn't refer to my answer. :-) – user 1591719 May 01 '15 at 10:32
  • 1
    @Chris'ssis: I was partially referring to your answer. I too got initially caught up with the G-function because of the integral of the log gamma. Yes I ended up with a lot of machinery, but I'd rather that than refer to a function that is likely unfamiliar to the OP. I didn't end up referring to anything worse than the digamma function (which was not really necessary to the analysis anyway, so expect a simplification later). The G-function would have been necessary if we needed a continuous version of it (i.e., integral). – Ron Gordon May 01 '15 at 10:40
  • 1
    OK. I've always appreciated all the solutions given to a problem since looking at a problem and tackling it from many points of views is one of the best learning ways I've found so far. That one I initially mentioned was the first tool that came to mind. – user 1591719 May 01 '15 at 10:46
5

1). Stirling's approximation and Stirling-like asymptotic series for Barnes G- function here at $(14)$ http://mathworld.wolfram.com/BarnesG-Function.html

Therefore,

$$\lim_{n\rightarrow \infty} \dfrac{\displaystyle \ln \left ( \frac{(n!)^n}{(0!1!2!…n!)^2} \right )}{n^2}=\frac{1}{2}.$$

2). Apply Stolz–Cesàro theorem for an elementary solution.

3). Combine the squeeze theorem, Stirling's approximation and the Riemann sums.

4). Make use of the Euler–Maclaurin formula.

user 1591719
  • 45,325
  • This is not an answer. – Apurv May 01 '15 at 06:59
  • @Apurv I assure you that it's all you need to compute the limit in fast and elegant manner. – user 1591719 May 01 '15 at 07:00
  • 2
    what did the person asking the question, or anyone else interested in an answer, learn? – ivancho May 01 '15 at 07:02
  • 2
    I think those comments are a bit unfair. Speaking for myself, I learned about the BarnesG function. Moreover, the person asking the question now have some more tools on how to work a limit like this out. I like it that the solution is not written out completely. – mickep May 01 '15 at 07:14
  • 1
    @mickep The original comments are from when the answer was only its first sentence. Which does not provide you with any tools, or pointers. It also doesn't tell you how to actually derive the result, just that some person on MathWorld and some person on math.stackexchange.com know about an asymptotic formula. – ivancho May 01 '15 at 07:18
  • 4
    Frankly, this is now a good answer. I like both 2. and 3., they don't have to be expanded into full solutions to be useful. – ivancho May 01 '15 at 07:23
  • I've been gone since Friday. I see that you haven't been in chat for 3 days and r9m has not been there in 5. Is everything okay? – robjohn May 27 '15 at 09:29