11

While working on a differential equation I stumbled on this integral: $$\int_0^x\frac{e^{-t^2}}{a^2+t^2}dt,$$ where $a\in \mathbb{R_{>0}}$.

At first glance it looks so simple that it's ridiculous that I can't find closed form expression for it.

Can someone help me with this beast?

$Edit:$From the comments, it doesn't seem to exist a closed form. I tried a series expansion by writing the integrand as $e^{-t^2}g(t)$ and take the Taylor expansion of the function $g(t)$. It's a good approximation and it's possible to integrate the expression found but it's only valid for small values of $x$ since the series diverge.

PML
  • 1,006
  • I'm not sure it has a closed expression. Try to exchange the integrals, hopefully the other direction gets you some $t$ as multiplier. – Berci Jun 13 '13 at 23:25
  • Or, perhaps, do you want its limit as $x\to\infty$? – Berci Jun 13 '13 at 23:27
  • @Berci Well, there's no other integration to be made. I have a ODE whose solution is $f(x)=\int_0^x \frac{e^{-t^2}}{a^2+t^2}dt+...$. So I can't take the limit.

    I'm sorry if I misled you with the way I wrote the question.

    – PML Jun 13 '13 at 23:32
  • I would guess that this does not have an expression in terms of elementary functions. – parsiad Jun 13 '13 at 23:33
  • @PML: Just because it is an ODE does not mean its solution can be expressed in terms of elementary functions. – parsiad Jun 13 '13 at 23:34
  • 1
    @par I wasn't implying that it does. I was just answering the first two comments. – PML Jun 13 '13 at 23:36
  • This question may be relevant. Trying to solve the differential equation mentioned in the accepted answer leads to an integral similar to this one, so I'm not sure how useful it is. – Javier Jun 14 '13 at 00:15
  • @JavierBadia Very nice reference. Well my equation is second order but indeed it can be put in the same form as that equation. The difference is in the non-homogeneous part (their $f$ function). That theorem is very useful but it says nothing about special functions. For example the integral of $e^{x^2}$ is itself a function. – PML Jun 14 '13 at 00:53
  • I just ran it through Mathematica (which has quite advance formulas and algorithms for integration)... and nothing. In all likelihood, there is no closed form. – carlosayam Jun 14 '13 at 04:34
  • Why this question not tagged [tag:definite-integral]? – paul2357paul Jun 16 '13 at 18:35
  • @paul2357paul I forgot to add it. I just edited the question. – PML Jun 17 '13 at 21:14

2 Answers2

12

$\int_0^x\dfrac{e^{-t^2}}{a^2+t^2}dt=\int_0^x\dfrac{1}{a^2+t^2}\sum\limits_{n=0}^\infty\dfrac{(-t^2)^n}{n!}dt=\int_0^x\sum\limits_{n=0}^\infty\dfrac{(-1)^nt^{2n}}{n!(a^2+t^2)}dt$

Consider $\int\dfrac{t^{2n}}{a^2+t^2}dt$ ,

Let $t=a\tan\theta$ ,

Then $dt=a\sec^2\theta~d\theta$

$\therefore\int\dfrac{t^{2n}}{a^2+t^2}dt$

$=\int\dfrac{(a\tan\theta)^{2n}}{a^2+(a\tan\theta)^2}a\sec^2\theta~d\theta$

$=\int a^{2n-1}\tan^{2n}\theta~d\theta$

$=\sum\limits_{k=0}^{n-1}\dfrac{(-1)^ka^{2n-1}\tan^{2n-2k-1}\theta}{2n-2k-1}+(-1)^na^{2n-1}\theta+C$

$=\sum\limits_{k=0}^{n-1}\dfrac{(-1)^ka^{2k}t^{2n-2k-1}}{2n-2k-1}+(-1)^na^{2n-1}\tan^{-1}\dfrac{t}{a}+C$

Hence $\int_0^x\sum\limits_{n=0}^\infty\dfrac{(-1)^nt^{2n}}{n!(a^2+t^2)}dt$

$=\left[\sum\limits_{n=1}^\infty\sum\limits_{k=0}^{n-1}\dfrac{(-1)^{n+k}a^{2k}t^{2n-2k-1}}{n!(2n-2k-1)}+\sum\limits_{n=0}^\infty\dfrac{a^{2n-1}}{n!}\tan^{-1}\dfrac{t}{a}\right]_0^x$

$=\sum\limits_{n=1}^\infty\sum\limits_{k=0}^{n-1}\dfrac{(-1)^{n+k}a^{2k}x^{2n-2k-1}}{n!(2n-2k-1)}+\dfrac{e^{a^2}}{a}\tan^{-1}\dfrac{x}{a}$

  • This is outstanding!Thank you very much, really. I hope you get many upvotes for the work you've put into this answer. Thank you once again. – PML Jun 15 '13 at 00:50
5

I'll tackle the case of the large values of $x$. Let's express the integral as follows:

$$\int_0^x\frac{e^{-t^2}}{a^2+t^2}dt=\int_0^{\infty}\frac{e^{-t^2}}{a^2+t^2}dt-\int_x^{\infty}\frac{e^{-t^2}}{a^2+t^2}dt=$$

$$=I(a)-\int_x^{\infty}\frac{e^{-t^2}}{a^2+t^2}dt$$

The next step is evaluate the last integral by parts:

$$-\int_x^{\infty}\frac{e^{-t^2}}{a^2+t^2}dt=\frac{1}{2}\int_x^{\infty}\frac{d(e^{-t^2})}{t(a^2+t^2)}dt=$$

$$=-\frac{e^{-x^2}}{2x(a^2+x^2)}+\frac{1}{2}\int_x^{\infty}\frac{(3t^2+a^2)e^{-t^2}}{t^2(a^2+t^2)^2}dt$$

In the last expression, the first term dominates for large $x$ and we have approximately:

$$\int_0^x\frac{e^{-t^2}}{a^2+t^2}dt\approx I(a)-\frac{e^{-x^2}}{2x(a^2+x^2)}$$

We'll get a better result if we continue the integration by parts in the penultimate expression.

Now, consider

$$I(a)=\int_0^{\infty}\frac{e^{-t^2}}{a^2+t^2}dt$$

Let's introduce the parameter $b$ so that

$$J(b)=\int_0^{\infty}\frac{e^{-bt^2}}{a^2+t^2}dt$$

and $I(a)=J(1)$

By differentiating with respect to $b$ we will see that $J$ satisfies the following differential equation:

$$\frac{dJ}{db}-a^2J+\frac{\sqrt{\pi}}{2\sqrt{b}}=0$$

Taking into account that

$$J(0)=\int_0^{\infty}\frac{dt}{a^2+t^2}=\frac{\pi}{2a}$$ we get a solution of this diff-eq:

$$J(b)=\sqrt{\pi}\frac{e^{a^2b}}{a}\int_{a\sqrt{b}}^{\infty}e^{-y^2}dy$$

Finally

$$I(a)=J(1)=\sqrt{\pi}\frac{e^{a^2}}{a}\int_{a}^{\infty}e^{-y^2}dy=\frac{\pi}{2}\frac{e^{a^2}}{a}\text{erfc}(a)$$ where $\text{erfc}(x)=\frac{2}{\sqrt{\pi}}\int_{x}^{\infty}e^{-y^2}dy$ is so called "the complementary error function"

Martin Gales
  • 6,878
  • I'm learning a lot with your answers. I thank you very much with the time you took on your answer. Your approach is very interesting and pedagogical. Thank you once again – PML Jun 17 '13 at 21:12