4
$\begingroup$

For all $x \in \Bbb R_+^*$, we put: $$f(x)=\frac{1}{\Gamma(x)}\int_x^{+\infty}t^{x-1}e^{-t}dt.$$ Can we compute the limit : $\displaystyle\lim_{x \to +\infty} f(x) $?

$\endgroup$
9
  • $\begingroup$ Consider the limit of a product of two functions of $x$. $\endgroup$ – Mhenni Benghorbal Dec 29 '14 at 3:55
  • 2
    $\begingroup$ This should be $1/2$, and the intuition is that $f(x) = \Bbb{P}(X \geq \Bbb{E}X)$ for $X \sim \mathrm{Gamma}(x, 1)$. Since the gamma distribution arises when summing i.i.d. exponential distribution, exploiting CLT allows us to guess that the limit should be $1/2$. $\endgroup$ – Sangchul Lee Dec 29 '14 at 4:01
  • $\begingroup$ @Mhenni: if you mean the two fuctions $F=\frac 1{\Gamma}$ and $G:x \mapsto \int_x^{+\infty} t^{x-1}e^{-t} dt$ , it is known that $\displaystyle \lim_{+\infty} F = 0$ and i think we can prove that: $\displaystyle \lim_{+\infty} G=+\infty $ ... $\endgroup$ – Mohamed Dec 29 '14 at 4:17
  • $\begingroup$ @Mohamed: the limit of G should be $0$. $\endgroup$ – Mhenni Benghorbal Dec 29 '14 at 4:31
  • 1
    $\begingroup$ @Mhenni:Thank you, i well verify this. $\endgroup$ – Mohamed Dec 29 '14 at 4:58
3
$\begingroup$

$\newcommand{\angles}[1]{\left\langle\, #1 \,\right\rangle} \newcommand{\braces}[1]{\left\lbrace\, #1 \,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\, #1 \,\right\rbrack} \newcommand{\ceil}[1]{\,\left\lceil\, #1 \,\right\rceil\,} \newcommand{\dd}{{\rm d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\dsc}[1]{\displaystyle{\color{red}{#1}}} \newcommand{\expo}[1]{\,{\rm e}^{#1}\,} \newcommand{\fermi}{\,{\rm f}} \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,} \newcommand{\half}{{1 \over 2}} \newcommand{\ic}{{\rm i}} \newcommand{\iff}{\Longleftrightarrow} \newcommand{\imp}{\Longrightarrow} \newcommand{\Li}[1]{\,{\rm Li}_{#1}} \newcommand{\norm}[1]{\left\vert\left\vert\, #1\,\right\vert\right\vert} \newcommand{\pars}[1]{\left(\, #1 \,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}} \newcommand{\root}[2][]{\,\sqrt[#1]{\vphantom{\large A}\,#2\,}\,} \newcommand{\sech}{\,{\rm sech}} \newcommand{\sgn}{\,{\rm sgn}} \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\ul}[1]{\underline{#1}} \newcommand{\verts}[1]{\left\vert\, #1 \,\right\vert}$ $\ds{\fermi\pars{x}\equiv{1 \over \Gamma\pars{x}} \int_{x}^{\infty}t^{x - 1}\expo{-t}\,\dd t:\ {\large ?}}$


First, we have to derive an asymptotic behavior of the integral when $\ds{x \gg 1}$. The derivation is similar to the one which leads to the Gamma function asymptotic behavior ( namely, the Stirling approximation ):

\begin{align}&\left.\dsc{\int_{x}^{\infty}t^{x - 1}\expo{-t}\,\dd t}\, \right\vert_{\,x\ \gg\ 1} =\int_{x}^{\infty}\exp\pars{\bracks{x - 1}\ln\pars{t} - t}\,\dd t \\[5mm]&\sim\int_{x}^{\infty} \exp\pars{\bracks{x - 1}\ln\pars{x - 1} - \bracks{x - 1} - {\bracks{t - x + 1}^{2} \over 2\bracks{x - 1}}}\,\dd t \\[5mm]&=\pars{x - 1}^{x - 1}\expo{-\pars{x - 1}} \int_{1}^{\infty}\exp\pars{-t^{2} \over 2\bracks{x - 1}}\,\dd t \\[5mm]&=\bracks{\pars{x - 1}^{x - 1}\expo{-\pars{x - 1}}}\root{2\pars{x - 1}} \int_{1/\root{2\pars{x - 1}}}^{\infty}\exp\pars{-t^{2}}\,\dd t \\[5mm]&\sim\pars{x - 1}^{x - 1/2}\expo{-\pars{x - 1}}\root{2}\ \overbrace{\int_{0}^{\infty}\exp\pars{-t^{2}}\,\dd t}^{\dsc{\root{\pi} \over 2}} \\[5mm]&=\dsc{\half}\, \bracks{\root{2\pi}\pars{x - 1}^{x - 1/2}\expo{-\pars{x - 1}}} \end{align}

Then

\begin{align}&\color{#66f}{\large% \left.{\int_{x}^{\infty}t^{x - 1}\expo{-t}\,\dd t \over \Gamma\pars{x}}\, \right\vert_{\,x\ \gg\ 1}} \sim{\pars{\dsc{1/2}}\bracks{% \root{2\pi}\pars{x - 1}^{x - 1/2}\expo{-\pars{x - 1}}}\over \root{2\pi}\pars{x - 1}^{x - 1/2}\expo{-\pars{x - 1}}} \color{#66f}{\large\to\ \half} \end{align}

The $\ds{\Gamma\pars{s,z}}$ asymptotic behavior as given in this link is misleading because $\ds{\tt\mbox{it should be valid for fixed}}$ $\ds{s}$. In another words, $\ds{\lim_{x\ \to\ \infty}\lim_{s\ \to\ x}\Gamma\pars{s,x}}$ can not be evaluated with such expansion which was the one I used blindly in my previous answer.

Thanks to @Mohamed who call my attention to this point and to this paper and thanks to @anorton who was worried about the whole procedure. Thanks to both of them.

$\endgroup$
9
  • $\begingroup$ Isn't computing $\lim_{x\to\infty}\Gamma(x, x)$ still pretty difficult, even with that nice asymptotic form? (Or am I just not seeing something?) $\endgroup$ – apnorton Dec 29 '14 at 4:12
  • $\begingroup$ I find this method pretty easy. We do not have to 'reinvent the wheel'. Otherwise we are forced to rederive a well known result. Thanks. $\endgroup$ – Felix Marin Dec 29 '14 at 4:24
  • 1
    $\begingroup$ Hmm... Ok. I guess my limit-taking abilities aren't that great with sums like that. I certainly agree that reducing to a known problem is much easier, I just can't see how the sum reduces to the fraction you've listed... I'll work on it more in the morning, I guess. $\endgroup$ – apnorton Dec 29 '14 at 4:42
  • $\begingroup$ Why this paper gives $\frac 12$? See pages 468 and 469. cs.purdue.edu/homes/wxg/selected_works/section_02/068.pdf $\endgroup$ – Mohamed Dec 29 '14 at 5:14
  • $\begingroup$ @Mohamed I'll check it. Thanks. $\endgroup$ – Felix Marin Dec 29 '14 at 5:51
1
$\begingroup$

Does this help?

$\int_{0}^{\infty}t^{x-1}e^{-t}dt=\int_{0}^{x}t^{x-1}e^{-t}dt+\int_{x}^{\infty}t^{x-1}e^{-t}dt$

so

$\int_{x}^{\infty}t^{x-1}e^{-t}dt=\Gamma\left(x\right)-\int_{0}^{x}t^{x-1}e^{-t}dt$

In taking the limit we get zero: $\lim_{x\rightarrow\infty}\frac{\Gamma\left(x\right)-\int_{0}^{x}t^{x-1}e^{-t}dt}{\Gamma\left(x\right)}=1-1=0$

$\endgroup$
1
  • $\begingroup$ You've only asserted that the limit in the last line goes to zero. How do we know that? $\endgroup$ – apnorton Dec 29 '14 at 14:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.