3
$\begingroup$

Let $0<\alpha<\beta$. I am looking for an equivalent for $\displaystyle I_n=\int_{\alpha n}^{\beta n}s^n\mathrm{e}^{-s}\mathrm{d}s$ when $n\longrightarrow+\infty$ : $I_n\underset{{n}\rightarrow{+\infty}}\sim ?$. I have tried integration by parts or use of asymptotics expansion for the Gamma function, I have changed variable, developed $\exp$... without success. Thank you.

$\endgroup$
4
  • $\begingroup$ What do you mean by $an$ and $bn$? The product of $a$ and $b$ with $n$? $\endgroup$ Commented Jun 2, 2021 at 9:34
  • $\begingroup$ Yes, $\alpha\times n$ and $\beta\times n$, product of $\alpha,\beta\in\mathbb{R}$ and $n\in\mathbb{N}^*$. $\endgroup$
    – P.Fazioli
    Commented Jun 2, 2021 at 9:43
  • $\begingroup$ @P.Fazioli Do we have any relations between $\alpha$, $\beta$ and $1$? $\endgroup$
    – Gary
    Commented Jun 2, 2021 at 10:40
  • $\begingroup$ The most interesting case for me is $\alpha\in]0,1[$ and $\beta=2\alpha$. $\endgroup$
    – P.Fazioli
    Commented Jun 2, 2021 at 10:48

1 Answer 1

4
$\begingroup$

If you make the change of integration variables $s = ne^t$, you find $$ \int_{\alpha n}^{\beta n} {s^n e^{ - s} ds} = n^{n + 1} e^{ - n} \int_{\log \alpha }^{\log \beta } {\exp ( - n(e^t - t - 1))e^t dt} . $$ Now $t=0$ is the sole saddle point of $e^t-t-1$ on the real line. The asymptotic form of the integral depends on whether this saddle is inside the interval of integration or not. See https://dlmf.nist.gov/2.3#iii and https://dlmf.nist.gov/2.4#iv

Alternatively, \begin{align*} & \int_{\alpha n}^{ + \infty } {s^n e^{ - s} ds} - \int_{\beta n}^{ + \infty } {s^n e^{ - s} ds} = \Gamma (n + 1,\alpha n) - \Gamma (n + 1,\beta n) \\ & = n\Gamma (n,\alpha n) - n\Gamma (n,\beta n) + (\alpha n)^n e^{ - \alpha n} - (\beta n)^n e^{ - \beta n} . \end{align*} The asymptotics of the incomplete gamma functions will depend on the sizes of $\alpha$ and $\beta$. See http://dlmf.nist.gov/8.11.iii You can also obtain a uniform asymptotic approximation where you do not need to distinguish any cases, see http://dlmf.nist.gov/8.12 with $Q(a,z)=\Gamma(a,z)/\Gamma(a)$.

$\endgroup$
4
  • $\begingroup$ Thank you very much for those ideas! $\endgroup$
    – P.Fazioli
    Commented Jun 2, 2021 at 11:28
  • $\begingroup$ You may accept it as the answer. It seems that you have never accepted any answers to your questions on this site before. $\endgroup$
    – Gary
    Commented Jun 2, 2021 at 11:32
  • $\begingroup$ Sorry, I had not understood. Thank you. $\endgroup$
    – P.Fazioli
    Commented Jun 2, 2021 at 11:46
  • $\begingroup$ No problem, now you see how the site works. $\endgroup$
    – Gary
    Commented Jun 2, 2021 at 11:55

You must log in to answer this question.

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