0
$\begingroup$

To begin with, how to show series $a_n=\dfrac{n^{n+\frac{1}{2}}e^{-n}}{n!}, n\in\mathbb{N}^*$ is monotonic and, specifically, increasing?


Now to make it general, replace $\frac{1}{2}$ above with an arbitrary positive parameter $p$. Then when does the series $a_n=\dfrac{n^{n+p}e^{-n}}{n!}$ shows monoticity and if it is monotonic, how to determine if it is increasing or decreasing?

$\endgroup$
1
  • $\begingroup$ One way is to check the derivative of $x^{x+1/2} e^{-x}/\Gamma(x+1)$. $\endgroup$
    – Math Lover
    Nov 24, 2017 at 0:46

3 Answers 3

2
$\begingroup$

\begin{equation} a_{n+1}\geq a_n \iff \frac{(n+1)^{n+p+1}}{e^{n+1}(n+1)!}\geq \frac{n^{n+p}}{e^nn!} \iff \frac{(n+1)^{n+p}}{n^{n+p}}\geq e \end{equation} Note that; \begin{equation} \left(1+\frac{1}{n}\right)^{n+p} \to e \end{equation} Take log; \begin{equation} p \geq \frac{1}{\log(1+1/n)} - n \to \frac{1}{2} \end{equation} Therefore if $p > 1/2$, then $a_n$'s are increasing, and else they are decreasing, at least eventually.
If you are interested in $p=1/2$, $\log^{-1}(1+1/n)-n$ increases to $1/2$ hence $a_n$'s are increasing.


\begin{equation}\log^{-1}(1+1/n) - n \leq 1/2\end{equation}


Rearrange terms such as; \begin{equation} \log(1+1/n) \geq \frac{2}{2n+1} = \frac{1}{n}\left(\frac{1}{1+\frac{1}{2n}}\right) \end{equation} Expand both side as $n\to\infty$; \begin{equation} \sum_{k=1}^{\infty}\frac{(-1)^{k+1}}{kn^k} \geq \sum_{k=1}^{\infty}\frac{(-1)^{k+1}}{2^{k-1}n^k} \end{equation} Now, you can compare coefficients and note that $k=1,2$ are equal but then $2^{k-1}$ increases exponentially.

$\endgroup$
4
  • $\begingroup$ I assume you get the limit of 1/2 by using Taylor's series of $log(1+x)$? But how did you know that $log^{-1}(1+1/n)-n$ increases to $\frac{1}{2}$? $\endgroup$
    – Nicholas
    Nov 25, 2017 at 3:00
  • $\begingroup$ That's right, when I wrote down Taylor Series I got: $(1/2)\frac{1-2/3x+\cdots}{1-1/2x+\cdots}$ and argued that it increases to $1/2$ since the denominator is slightly larger than the numerator. (I spot a careless mistake in my calculations). Better way to argue is to directly show $\log^{-1}(1+1/n)-n \leq 1/2$, which is equivalent to show $\log(1+1/n) \geq \frac{2}{1+2n}$. Expand both sides and you notice that left has terms $1/k$ and right has $1/2^{k-1}$. $\endgroup$
    – Atbey
    Nov 25, 2017 at 4:24
  • $\begingroup$ Why is the series on the right-hand side in your last step smaller? Although $2^{k-1}$ is exponentially growing, the series are both alternating series, that is, you add a smaller term, but then you also subtract a smaller term afterward. Can you explain in your last step how you compare the series based on the coefficients? $\endgroup$
    – Nicholas
    Nov 27, 2017 at 1:03
  • $\begingroup$ Since we are interested in $n\to\infty$, only term that actually matters is $k=3$. As you pointed out, they are alternating series. Hence you can approximate error: $\frac{1}{3n^3} - \frac{1}{4n^4}\geq \frac{1}{4n^3} + \frac{1}{8n^4}$, which is true for $n>5$. $\endgroup$
    – Atbey
    Nov 27, 2017 at 3:34
0
$\begingroup$

When you face problems with factorials, Stirling approximation is very useful. $$a_n=\dfrac{n^{n+\frac{1}{2}}e^{-n}}{n!}\implies \log(a_n)=\left(n+\frac 12\right)\log(n)-n-\log(n!)$$ Using Stirling approximation, we then get $$\log(a_n)=-\frac 12 \log(2\pi)-\frac 1{12n}+O\left(\frac{1}{n^3}\right)$$ which is increasing (have a look at the derivative).

Now, for large values of $n$, use $$a_n=e^{\log(a_n)}=\frac 1 {\sqrt{2\pi}}\left(1-\frac 1{12n}\right)+O\left(\frac{1}{n^2}\right)$$

If you make it more general $$b_n=\dfrac{n^{n+p}e^{-n}}{n!}\implies \log(b_n)=\left(n+p\right)\log(n)-n-\log(n!)$$ and applying the same, we should arrive to $$\log(b_n)=-\frac 12 \log(2\pi)+\left(p-\frac{1}{2}\right) \log (n)-\frac 1{12n}+O\left(\frac{1}{n^3}\right)$$ and then, if $p>\frac 12$, $\log(b_n)$ and then $b_n$ would be increasing.

$\endgroup$
0
$\begingroup$

Since $\log(x)$ is a concave function on $\mathbb{R}^+$, by the Hermite-Hadamard inequality

$$ \sum_{k=1}^{n}\log(n)\leq \tfrac{1}{2}\log(n)+\int_{1}^{n}\log(x)\,dx = 1-n+\left(n+\tfrac{1}{2}\right)\log n$$ and by exponentiating both sides $n!\leq n^{n+\frac{1}{2}}e^{1-n}$ follows.
Let $a_n=\log\left(n^{n+\frac{1}{2}}e^{-n}\right)=(n+\frac{1}{2})\log(n)-n $. We have $$ a_{n+1}-a_n = \log(n+1)+\left(n+\tfrac{1}{2}\right)\log\left(1+\tfrac{1}{n}\right)-1$$ $$ \frac{d}{dx}\left[\log(x+1)+\left(x+\tfrac{1}{2}\right)\log\left(1+\tfrac{1}{x}\right)-1\right]=\log\left(1+\tfrac{1}{x}\right)-\tfrac{1}{2x(x+1)}\geq 0 $$ hence $\{a_n\}_{n\geq 1}$ is increasing and log-convex. Similarly, by letting $b_n=\log\left(n^{n+\frac{1}{2}}e^{-n}/n!\right)$ we have $$ b_{n+1}-b_n = \left(n+\tfrac{1}{2}\right)\log\left(1+\tfrac{1}{n}\right)-1\stackrel{\text{HHI}}{\geq} 0$$ $$ \frac{d}{dx}\left[\left(x+\tfrac{1}{2}\right)\log\left(1+\tfrac{1}{x}\right)-1\right]=\log\left(1+\tfrac{1}{x}\right)-\tfrac{x+\frac{1}{2}}{x(x+1)}\stackrel{\text{HHI}}{\leq} 0 $$ hence $\{b_n\}_{n\geq 1}$ is increasing and log-concave, since $$ \log\left(1+\tfrac{1}{x}\right)=\int_{x}^{x+1}\frac{dt}{t}\leq \tfrac{1}{2}\left(\tfrac{1}{x}+\tfrac{1}{x+1}\right).$$

$\endgroup$

You must log in to answer this question.

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