3

Evaluate:$$I=\int_{-\infty}^\infty \int_{-\infty}^\infty \mathrm e^{-(2x^2+2xy+2y^2)}dxdy$$

I am just clueless to solve it.your help will be appreciated.

MatheMagic
  • 2,386

4 Answers4

7

Note that $$\int_{-\infty}^{\infty} \frac{1}{\sigma \sqrt{2\pi}} e^{-\frac{1}{2}{\big(\frac{x-\mu}{\sigma}\big)}^2}=1; ~~\mu \in \mathbb{R}, ~\sigma>0 ~\cdots (\star)$$ Here the integrand is the probability density function (pdf) of the Normal distribution (as it is called by the statisticians), aka the Gaussian distribution (as it is called by the mathematicians). And by definition, for any pdf $f$, we have $\int_{\Omega}f(x)dx=1$, where $\Omega$ is the domain of $f$. Here $\Omega=\mathbb{R}$.

Now, \begin{align} I&=\int_{-\infty}^{\infty} \int_{-\infty}^{\infty} e^{-(2x^2+2xy+2y^2)} dx dy\\ &=\int_{-\infty}^{\infty} \int_{-\infty}^{\infty} e^{-2\big\{(x+\frac{y}{2})^2+\frac{3y^2}{4}\big\}} dx dy\\ &=\int_{-\infty}^{\infty} e^{-2\frac{3y^2}{4}}\Big\{ \int_{-\infty}^{\infty} e^{-2(x+\frac{y}{2})^2} dx\Big\}dy\\ &=\frac{1}{2}\sqrt{2\pi}\int_{-\infty}^{\infty} e^{-\frac{3y^2}{2}}\bigg\{ \int_{-\infty}^{\infty} \frac{1}{\frac{1}{2}\sqrt{2\pi}}e^{-\frac{1}{2}\bigg(\frac{x-(-\frac{y}{2})}{\frac{1}{2}}\bigg)^2} dx\bigg\}dy\\ &=\frac{\sqrt{\pi}}{2}\int_{-\infty}^{\infty} \big( e^{-\frac{3y^2}{2}} \times 1 \big)dy ~~\mbox{[Using $(\star)$]}\\ &=\sqrt{\frac{\pi}{2}} \sqrt{\frac{2\pi}{3}} \int_{-\infty}^{\infty} \frac{1}{\frac{1}{\sqrt{3}}\sqrt{2\pi}}e^{-\frac{1}{2}\bigg( \frac{y-0}{\frac{1}{\sqrt{3}}} \bigg)^2} dy\\ &=\sqrt{\frac{\pi}{2}} \sqrt{\frac{2\pi}{3}} \times 1 ~~\mbox{[Again, using $(\star)$]}=\frac{\pi}{\sqrt{3}} \end{align}

  • Do you actually think the normal distribution is called 'Gaussian distrinution' by mathematicians or you just didn't like me editing your answer? –  Dec 14 '16 at 08:05
  • I apologize for not noticing the edit. I'm new to this website, so still do not fully know how it works. Probably you edited the solution while I was editing it myself. So after saving it, your editing got overridden. It's absolutely fine to edit my answer :) And it'll be great if you could explain the process too.. –  Dec 14 '16 at 08:16
  • 1
    Oh... Sorry, I should have imagined that my edit just got overridden by yours. I am also relatively new to the site but I don't think there is so much to say about the editing progress; you can edit questions and answers when you have 2k+ reputation points, if you don't have that much rep. then your edit is saved as a suggested edit (for a high rep. user or the owner to validate), and finally the owner of the question/answer can always edit them (even if they were edited before by anyone or if it overrides someone's else edit ;). Welcome to MSE!! =) –  Dec 14 '16 at 08:28
  • Dear @aditi_ray: Your solution is neat and complete so +1 but since you're a new user i'd like to tell you that your solution is along the same lines of hint given by victorbarg so its better to not write the solution if your idea is similar to the one given already.Also,on this site most people prefer to give hint instead of writing full solution (since the problem might be from assignment/homework).Anyway,Welcome to MSE and good luck! – Arpit Kansal Dec 14 '16 at 10:34
  • @ArpitKansal: I'll keep it in mind. Thanks. –  Dec 14 '16 at 15:08
6

Hint: Gaussian Integral and the observation $x^2+xy+y^2=(x+ \frac {y}{2})^2+(\frac {\sqrt{3}}{2}y)^2$ is precisely what you need.

Math Lover
  • 3,602
6

$$ \begin{align} \int_0^{2\pi}\int_0^\infty e^{-(2+\sin(2\theta))\,r^2}r\,\mathrm{d}r\,\mathrm{d}\theta &=\frac12\int_0^{2\pi}\frac{\mathrm{d}\theta}{2+\sin(2\theta)}\\ &=\frac14\int_0^{4\pi}\frac{\mathrm{d}\theta}{2+\cos(\theta)}\\ &=\frac12\oint_{|z|=1}\frac1{2+\frac{z+1/z}2}\frac{\mathrm{d}z}{iz}\\ &=\frac1i\oint_{|z|=1}\frac{\mathrm{d}z}{z^2+4z+1}\\ &=2\pi\operatorname*{Res}_{z=-2+\sqrt3}\left(\frac1{z^2+4z+1}\right)\\ &=2\pi\left.\left(\frac1{2z+4}\right)\right|_{z=-2+\sqrt3}\\ &=\frac\pi{\sqrt3} \end{align} $$

robjohn
  • 345,667
  • how is $\frac{1}{2}\int_{0}^{2\pi}\frac{d\theta}{2+\sin(2\theta)}=\frac{1}{4}\int_{0}^{4\pi}\frac{d\theta}{2+\cos(\theta)}$? – vidyarthi Dec 14 '16 at 08:10
  • 1
    It's really three steps. The first is $\theta\mapsto\theta/2$, which would give $$\frac14\int_0^{4\pi}\frac{\mathrm{d}\theta}{2+\sin(\theta)}$$ The second is $\theta\mapsto\theta+\pi/2$ which gives $$\frac14\int_{-\pi/2}^{4\pi-\pi/2}\frac{\mathrm{d}\theta}{2+\cos(\theta)}$$ The third is to note that the period of $\cos(\theta)$ is $2\pi$ and we can move the integral from $[-\pi/2,0]$ to $[4\pi-\pi/2,4\pi]$. – robjohn Dec 14 '16 at 08:26
6

I find it strange how other answers don't keep the symmetry. The symmetry between $x$ and $y$ openly suggests substitution $u=\frac{1}{\sqrt2}(x+y)$, $v=\frac{1}{\sqrt2}(x-y)$.

$$u^2=\frac12x^2+xy+\frac12y^2$$ $$v^2=\frac12x^2-xy+\frac12y^2$$ $$3u^2+v^2=2x^2+2xy+2y^2$$ As this substitution is just a $45^\circ$ rotation of the $xy$ plane, the Jacobian is $1$. The integral splits into a product of two Gaussian integrals, which have a well known value.

$$I=\int_{-\infty}^\infty e^{-3u^2}{\rm d}u \int_{-\infty}^\infty e^{-v^2}{\rm d}v=(\sqrt{\pi/3})(\sqrt{\pi})=\frac{\pi}{\sqrt3}$$


For those who enjoy neat tricks, a probably not so helpful tour-de-force solution, stimulated by @mathbeing:

Write the exponent as a quadratic form $$I=\iint_{\mathbb{R}^2} e^{-{\bf r}g{\bf r}}d^2 {\bf r}$$ where $g=\begin{bmatrix}2 & 1\\1 & 2\end{bmatrix}$ is the metric tensor. Write $g=L^T L$, and use ${\bf p}=L{\bf r}$ as substitution. Any valid $L$ will do, we don't need to compute it. We know $\det J =(\det L)^{-1} = \frac{1}{\sqrt{\det g}}=\frac{1}{\sqrt3}$. $$I=\frac{1}{\sqrt3}\iint_{\mathbb{R}^2} e^{-|p|^2}d^2 {\bf p}\underbrace{=}_{\text{polar}}\frac{1}{\sqrt3}2\pi\int_0^\infty p e^{-p^2}dp=\frac{\pi}{\sqrt3}$$ To completely generalize, in any number of dimensions, for any possible nondegenerate positive definite quadratic form in the exponent, we have $$\iiint_{\mathbb{R}^N}e^{-{\bf r}g{\bf r}}d^N{\bf r}=\pi^{N/2} (\det g)^{-1/2}$$

orion
  • 15,781
  • Great answer!! I would just add one more line to make the solution explicit to the non-Lord-Kelvin-type-of-mathematicians ;). –  Dec 14 '16 at 09:12
  • @mathbeing excellent find! Actually, It would be directly applicable here (to avoid referring to well-known black box values), by $\sqrt{3}u=t$, and going into polar coordinates, but that would be too Kelvin :) – orion Dec 14 '16 at 09:33
  • I mostly posted that link for future readers to avoid black-boxing such an easy identity ;) Lol, too Kelvin indeed... –  Dec 14 '16 at 10:18
  • If you had a (symmetric) quadratic form with more variables, what would you substitute to exploit the symmetry? E.g. what would the substitution be for $n=3$? – user159517 Dec 19 '16 at 13:17
  • This was nothing more than guessing the eigenvectors of the form by using symmetry arguments. In higher dimensions, if it's symmetric to permutation of indices, it always has an eigenvector (1,1,1,...), so u=x+y+z would be a safe bet. If N>2, the rest of the eigenspace is degenerate (N-1 equal eigenvalues), so any basis of orthogonal complement would do the trick. In N=2, there is only one dimension left, so x-y is the only choice. In more dimensions, whatever you choose that's orthogonal to (1,1,1), is ok, and it actually doesn't matter. I'd take (1,-1,0) and (-1,-1,2), pretty standard. – orion Dec 19 '16 at 14:07
  • In fact, you can then write $g=a \delta_{ij}+b v_i v_j$ where $v=(1,1,1,1...)/\sqrt{N}$. So it's obvious that v is the eigenvector with eigenvalue a+b. The rest of the eigenvalues are a. So by taking $u=(x+y+z+\cdots)/\sqrt{N}$, you get $(a+b)u^2+a(u_1^2+u_2^2+u_3^2+\cdots)$ - whatever the choice for the rest of the substitution would be, you'd always get just a sum of squares in the second term, so you don't even have to decide on how to express them. – orion Dec 19 '16 at 14:13