2

What are some different methods to evaluate

$$ \int_{-\infty}^{\infty} x^4 e^{-ax^2} dx$$

for $a > 0$.

This integral arises in a number of contexts in Physics and was the original motivation for my asking. It also arises naturally in statistics as a higher moment of the normal distribution.

I have given a few methods of evaluation below. Anyone know of others?

Simon S
  • 26,524
  • Why not just write up a $\LaTeX$ file for it instead of making a post here...? – Cameron Williams Feb 16 '15 at 01:57
  • If it's not a post here, it's worth creating I think. I'm surprised it hasn't been asked. It comes up quite a bit in Physics. – Simon S Feb 16 '15 at 01:58
  • 1
    You can use known results from the normal distribution ($\mu=0$ and $1/(2\sigma^2)=a$) to answer your question: http://en.wikipedia.org/wiki/Normal_distribution#Moments – Jason Feb 16 '15 at 01:58
  • Right ... that's a bit more theoretical baggage than I want, although my preferred solution uses that idea. – Simon S Feb 16 '15 at 01:59
  • 4
    It most definitely has. The standard trick people use is parametric differentiation. Another way is to evaluate it is via the gamma function. The least sexy way to evaluate it is of course just successive integration by parts. – Cameron Williams Feb 16 '15 at 02:00
  • I'm marking this as a duplicate since it is very closely related to the linked post. – Cameron Williams Feb 16 '15 at 02:03
  • Sorry, but this is quite different! Look again. – Simon S Feb 16 '15 at 02:04
  • It's really not. The technique you even used is in that thread. – Cameron Williams Feb 16 '15 at 02:10
  • @CameronWilliams Setting up a recurrence with parts is pretty nice. – Ian Feb 16 '15 at 02:51
  • Again, asking people not to vote to close this. The post being pointed to as the duplicate does not work through this example in any detail, or give as many methods, or give anything close to the value of the integral. – Simon S Feb 16 '15 at 04:51
  • @Simon S: this seems like a PSQ to me, so I have voted to close for lack of context. There are an infinite number of integral questions, so we can't expect every one to be answered explicitly. The question does not indicate why this one is special. Also, if you are planning to immediately answer your own question, then you probably do not need to ask it in the first place - this site is for questions to which the asker does not know the answer. – Carl Mummert Feb 20 '15 at 02:34
  • This integral comes up a lot in physics and this post is being pointed to by an MITx forum. (No idea what PSQ means.) – Simon S Feb 20 '15 at 16:20
  • A PSQ (Problem Statement Question) is a question that just states a problem with no context, motivation, or displayed progress. Such questions are generally discouraged on math.SE. http://meta.math.stackexchange.com/questions/9201/proposal-discourage-questions-that-are-nothing-besides-a-problem-statement @Simon S – Carl Mummert Feb 20 '15 at 16:41
  • I've edited the question slightly and hope it is now clear from my comments what the context of the question is. It should also be clear that there is motivation and from my answer there is considerable displayed progress! So whatever else this post is, it is not a PSQ. It has garnered serious responses.

    I am not a new user to Math SE; I have answered literally hundreds of questions and posed only 13. I am disappointed users are voting to close this post.

    – Simon S Feb 20 '15 at 18:15
  • FYI, there is now (at least) one other post on the identical question with a link to this one: http://math.stackexchange.com/questions/1376575/show-that-the-standard-integral-int-0-infty-x4-mathrme-alpha-x2 – Simon S Jul 28 '15 at 22:21

4 Answers4

11

1 Here's a relatively elegant method.

Notice that $\frac{\partial \ }{\partial a} e^{-a x^2} = - x^2 e^{-a x^2}$ and hence $\frac{\partial^2 \ }{\partial a^2} e^{-a x^2} = + x^4 e^{-a x^2}$

Thus, as the integrand is bounded and $C^\infty$ in both variables,

$$I = \int_{-\infty}^{\infty} x^4 e^{-ax^2} dx = \int_{-\infty}^{\infty} \frac{\partial^2 \ }{\partial a^2} e^{-ax^2} dx = \frac{d^2 \ }{da^2}\int_{-\infty}^{\infty} e^{-ax^2} dx $$

Since $\displaystyle \int_{-\infty}^{\infty} e^{-ax^2} dx = \frac{\sqrt \pi}{\sqrt a}$,

$$I = \frac{d^2 \ }{da^2} \frac{\sqrt \pi}{\sqrt a} = \frac{3\sqrt \pi}{4a^{5/2}}$$


2 Another method:

$$I^2 = \int_{-\infty}^{\infty} x^4 e^{-ax^2} dx \ \cdot \int_{-\infty}^{\infty} y^4 e^{-ay^2} dx = \int\int_{\mathbb R^2} (xy)^4 e^{-a(x^2 + y^2)} dx \ dy$$

Moving to polar coordinates,

$$I^2 = \int_0^{2\pi} \int_0^\infty r^8 \cos^4\theta\sin^4\theta e^{-ar^2} r \ dr \ d\theta = \int_0^\infty r^9e^{-ar^2} \ dr \ \cdot \ \int_0^{2\pi} \left(\frac{1}{2}\sin2\theta\right)^4 \ d\theta$$

With substitution $u = r^2$, the first integral is $\frac{4!}{2a^5}$. As $\sin^4 2\theta = \frac{1}{8} ( -4\cos4\theta + \cos 8\theta + 3)$, in the second integral the first two terms vanish over the domain of integration $[0,2\pi]$ and

$$I^2 = \frac{4!}{2a^5} \cdot \frac{1}{2^4} \frac{3}{8} 2\pi = \frac{9\pi}{16a^5}$$

Hence, as $I$ is positive,

$$I = \frac{3\sqrt \pi}{4a^{5/2}}$$


3 High school method:

Integrating by parts,

$$I = {-1 \over 2a} \int_{-\infty}^{\infty} x^3 (-2ax)e^{-ax^2} dx = {3\over 2a} \int_{-\infty}^{\infty} x^2 e^{-ax^2} dx $$

$$= {-3\over (2a)^2} \int_{-\infty}^{\infty} x (-2ax) e^{-ax^2} dx = {3 \over 4a^2} \int_{-\infty}^{\infty} e^{-ax^2} dx $$

and hence

$$I = {3 \over 4}{\sqrt\pi \over a^{5/2}}$$

Simon S
  • 26,524
  • I think you should probably justify switching the order of integration and differentiation. – Jason Feb 16 '15 at 02:11
6

Assuming $a>0$, we have: $$ I = \frac{1}{a^{5/2}}\int_{0}^{+\infty}x^{3/2}e^{-x}\,dx = \frac{\Gamma\left(5/2\right)}{a^{5/2}}=\color{red}{\frac{3\sqrt{\pi}}{4\, a^{5/2}}}.$$

Jack D'Aurizio
  • 353,855
4

Let $b>0$ be any positive, even integer, and let $a>0$. Then you have $$ \int_{-\infty}^\infty dx \, x^b e^{-ax^2} = 2 \int_0^\infty dx \, x^b e^{-ax^2} = a^{- (b+1)/2} \int_0^\infty e^{-t} t^{\frac{b-1}{2}} = \color{red}{a^{-(b+1)/2} \Gamma \left(\frac{b+1}{2} \right) }. $$ The particular case $b=4$ gives your result.

glS
  • 6,818
4

Generating functions, as often, provide a nice way to compute these integrals.

Let $I_n := \int_\mathbb{R} x^n e^{-ax^2} \ dx$, and for all complex $\lambda$:

$$G(\lambda) := \sum_{n=0}^\infty \frac{I_n}{n!}\lambda^n.$$

Then:

$$G(\lambda) = \int_\mathbb{R} \sum_{n=0}^\infty \frac{(\lambda x)^n}{n!} e^{-ax^2} \ dx = \int_\mathbb{R} e^{\lambda x -ax^2} \ dx = e^{\frac{\lambda^2}{4a}} \int_\mathbb{R} e^{-a\left(x-\frac{\lambda}{2a}\right)^2} \ dx = e^{\frac{\lambda^2}{4a}} \int_\mathbb{R} e^{-ax^2} \ dx.$$

Now, since $\int_\mathbb{R} e^{-ax^2} \ dx = \sqrt{\pi}/\sqrt{a}$, we finally get:

$$\sum_{n=0}^\infty \frac{I_n}{n!}\lambda^n = \frac{\sqrt{\pi} e^{\frac{\lambda^2}{4a}}}{\sqrt{a}} = \sum_{n=0}^\infty \frac{\sqrt{\pi}}{\sqrt{a}} \frac{1}{(4a)^n n!}\lambda^{2n}.$$

Identification of the coefficients yields $I_{2n+1} = 0$ for all $n$, and:

$$I_{2n} = \frac{\sqrt{\pi}}{\sqrt{a}} \frac{(2n)!}{(4a)^n n!}.$$

For $n=2$, this yields:

$$I_4 = \frac{\sqrt{\pi}}{\sqrt{a}} \frac{24}{32a^2} = \frac{3\sqrt{\pi}}{4a^{5/2}}.$$

D. Thomine
  • 10,870