1

Do we have : $$\frac{x!}{\sqrt{2\pi x}\left(\frac{x}{e}\right)^{x}e^{\frac{1}{12x}}}\left(1+\frac{1}{6}\ln\left(8x^{3}+4x^{2}+x+\frac{1}{30}\right)-\ln\left(\sqrt{2x}e^{\frac{1}{12x-\frac{1}{\pi^{2}x^{2}}}}\right)\right)^{-1}<1$$ for $x>4$

I have combining various improvement of the Stirling's approximation .

The starting point is $$\ln(ex)\leq x$$ for $x>0$

Next the idea is to divide by the lower bound to get a better upper bound .

For reference see the Wiki page https://en.wikipedia.org/wiki/Stirling%27s_approximation

I haven't any clue to show it in the case it would be true .

How to (dis)prove it ?

Angelo
  • 12,328

1 Answers1

2

Do not be so modest !

It is true for any $x > x_0$

$$x_0=3.6956333556853889395248451976513568330892774031006\cdots$$

Using series for the logarithm of the lhs gives $$\log(\text{lhs})=\frac{80-11 \pi ^2}{11520 \pi ^2 x^4}+\frac{13}{13440 x^5}+\frac{11}{2073600 x^6}+O\left(\frac{1}{x^7}\right)$$

The positive root of the above quadratic equation is $$\frac{\pi \left(234 \pi +\sqrt{\frac{1}{5} \left(279709 \pi ^2-43120\right)}\right)}{42 \left(11 \pi ^2-80\right)}=3.85540$$

Using the series expansion to $O\left(\frac{1}{x^9}\right)$ gives $3.67853$.

Using the series expansion to $O\left(\frac{1}{x^{20}}\right)$ gives $3.69561$.

Edit

If you use the extended Ramanujan approximation as given in the edit of my answer here, the difference with Stirling approximation is $$\frac{8521}{34406400 x^8}\left(1-\frac{1}{2 x}+\frac{1}{8 x^2}+O\left(\frac{1}{x^3}\right) \right)$$

This does not affect the above results in any manner.

Update

In order to be fully consistent and not use at any point Stirling approximation, I repeated the calculation using the extended Ramanujan approximation

$$n!\sim \sqrt{\pi}\left(\frac ne\right)^n\sqrt [6]{8 n^3+4 n^2+n+\frac{1}{30} \left(1+\sum _{i=0}^m \frac{a_i} {n^{i}}\right)}$$ the first $a_i$ being $$\left\{0,-\frac{11}{8},\frac{79}{112},\frac{3539}{6720},-\frac{9511}{13440},-\frac{30 153}{71680},\frac{233934691}{212889600},\frac{3595113569}{5960908800},\cdots\right\}$$

For consistency, the argument of the first logarithm has been replaced by $$8 n^3+4 n^2+n+\frac{1}{30} \left(1+\sum _{i=0}^m \frac{a_i} {n^{i}}\right)$$

Now, the lhs is grater than $1$ for any $n$ ! In fact, the expansion of the lhs stabilizes very quickly to $$1+\frac{1}{144 \pi ^2 n^4} \large A$$ where $${\large A}=1+\frac{\pi ^2}{1800 n^2}+\frac{30+\pi ^2}{360 \pi ^2 n^3}+\frac{\left(\frac{175}{\pi ^4}-8\right) \pi ^2}{25200 n^4}+\frac{\left(7-\frac{5400}{\pi ^2}\right) \pi ^2}{6804000 n^5}+\frac{29400+980 \pi ^2+49 \pi ^4+1200 \pi ^6}{4233600 \pi ^4 n^6}+O\left(\frac{1}{n^7}\right)$$

Why is that ?

$$\Delta=(n!)_{\text{approximate}}-\log(\Gamma(n+1))=\frac{8521}{34406400 n^8}\large B$$ where $${\large B}=1-\frac{1}{2 n}+\frac{1}{8 n^2}+\frac{201648777301}{27635648040 n^3}+\frac{52815059449}{255098289600 n^4}+O\left(\frac{1}{n^5}\right)$$

$\forall n >0$, $\Delta >0$ and the inequality should not hold while with expansion of $\Gamma(n+1)$ is does !