0
$\begingroup$

I try to understand the behavior of the following integral as $a\rightarrow\infty$ $$I(\delta)=\int_{-\delta}^\delta{\rm erfc}\Big(-\frac t{\sqrt a}\Big){\rm e}^{bt-t^2/a}{\rm d} t,$$ where ${\rm erfc(x)}$ is the complementary error function and $b$ can be written as a power of $a$. According to Mathematica, I already know that $$J(\delta)=\int_{-\delta}^{\delta}{\rm e}^{bt-t^2/a}{\rm d} t=\frac12\sqrt{\pi a}{\rm e}^{ab^2/4}\Big\{{\rm erf}\Big(\frac{ab+2\delta}{2\sqrt a}\Big)-{\rm erf}\Big(\frac{ab-2\delta}{2\sqrt a}\Big)\Big\},$$ but Mathematica doesn't compute $I(\delta)$. It appears from the above estimation that, with $\delta=ab$, we have $$J(\delta)=_{a\to\infty}\sqrt{\pi a}{\rm e}^{ab^2/4}\{1+O(a^{-m})\},\ m\in\mathbb{N}.$$ Do we have a similar approximation for $I(\delta)$? Is it possible to find a closed formula for $J(\delta)$ similar to the one for $I(\delta)$?

$\endgroup$
2
  • $\begingroup$ You have three parameters, $\delta, a, b$, could you please tell us explicitly what is the relationship between them in the situations you are interested in? $\endgroup$ Commented Sep 2 at 16:28
  • $\begingroup$ We can assume that $b=O(a^{-\alpha})$ with $0<\alpha<1/2$ and I need to find $\delta$ such that $I(\delta)=I(\infty)\{1+O(a^{-m})\},\quad(m\in\mathbb{N})$. $\endgroup$
    – Jean DELI
    Commented Sep 3 at 7:39

1 Answer 1

1
$\begingroup$

$\newcommand\de\delta\newcommand\erfc{\operatorname{erfc}}$If $0<\de=o(\sqrt a)$, then $\erfc(-\frac t{\sqrt a})\to1$ and $e^{-t^2/a}\to1$ uniformly in $t\in[-\de,\de]$, whence $$\begin{aligned} I(\de)&\sim \int_{-\de}^{\de}e^{bt}\,dt =\frac2b\,\sinh(b\de) \end{aligned}$$ if $b\ne0$, and $I(\de)\sim2\de$ if $b=0$.


Now consider the following limit conditions, which seem to be close to what the OP had in mind: $$a\to\infty,\quad b\ge1,\quad(\de-ab/2)/\sqrt a\to\infty. $$ Then we can find a varying $C$ such that $$C\to\infty,\quad C-b\sqrt a/4\to-\infty. $$ Write $$I(\de)=I_1+I_2,$$ where $$I_1:=\int_{-\de}^{C\sqrt a}\erfc(-t/\sqrt a)e^{bt-t^2/a}dt,\quad I_2:=\int_{C\sqrt a}^\de\erfc(-t/\sqrt a)e^{bt-t^2/a}dt,$$ Next, \begin{equation} I_1\le2\int_{-\de}^{C\sqrt a}e^{bt}dt=O(e^{Cb\sqrt a})=o(e^{ab^2/4}), \end{equation} \begin{equation} I_2\sim2\int_{C\sqrt a}^\de e^{bt-t^2/a}dt =2e^{ab^2/4}\sqrt a\,\int_{C-b\sqrt a/2}^{(\de-ab/2)/\sqrt a} e^{-z^2}dz \sim2e^{ab^2/4}\sqrt a\,\int_{-\infty}^\infty e^{-z^2}dz =2\sqrt{\pi a}\, e^{ab^2/4}. \end{equation} So, \begin{equation} I(\de)\sim2\sqrt{\pi a}\, e^{ab^2/4}. \end{equation}

$\endgroup$
6
  • $\begingroup$ Thank you, it helped. Going further, one can derive an error term for $I(\delta)$. Indeed, with the condition $b=O(a^{-\alpha})$, $0<\alpha<1/2$ and the choices $C:=\log(a)$, $\delta:=a^{-\alpha+1}$, we can see, using $$\rm{erfc}(x)=2+O\Big(\frac{{\rm e}^{-x^2/a}\sqrt{a}}{x\sqrt\pi}\Big),$$ that $$I(\delta)=2\sqrt{\pi a}{\rm e}^{ab^2/4}\{1+O(a^{-m}\},\quad(m\in\mathbb{N}).$$ $\endgroup$
    – Jean DELI
    Commented Sep 3 at 7:24
  • $\begingroup$ Can we imagine doing something similar with higher degree polynomial inside the exponential? Dealing with something like $$\int_{-\delta}^{\delta}{\rm erfc}\Big(-\frac{t}{\sqrt a}\Big){\rm e}^{bt-t^2/a+t^3/a^2-t^4/a^3}$$ and so on? $\endgroup$
    – Jean DELI
    Commented Sep 3 at 9:50
  • 1
    $\begingroup$ @JeanDELI : I think this should be possible, under appropriate conditions on $\delta, a, b$. You may want to consider posting such a question separately, after some preparations, including stating the conditions on the parameters. $\endgroup$ Commented Sep 3 at 12:52
  • $\begingroup$ Alright, will do. Last question, what if $b\leq -1$ this time? I guess we can derive a similar result, cutting the integral at $-C\sqrt a$? $\endgroup$
    – Jean DELI
    Commented Sep 3 at 12:59
  • $\begingroup$ @losifPinelis When $b<0$, it seems that the final expression is quite different from the one in the case $b>0$. Using this I found that $$\int_{-\infty}^{\infty}{\rm erfc}\Big(-\frac{t}{\sqrt a}\Big){\rm e}^{-bt-t^2/a} dt\sim\frac{2\sqrt 2}{|b|}{\rm e}^{ab^2/8}$$ I would like to prove this stands for $I(\delta)$ with a similar value for $\delta$ than the case $b>0$ but we can't use exactly the same method due to the change of sign of ${\rm erfc}$ after the substitution $u=-t$. $\endgroup$
    – Jean DELI
    Commented Sep 3 at 15:01

You must log in to answer this question.

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