6

I want to prove $$\int\limits_{-\infty}^\infty\int\limits_{-\infty}^\infty\frac{\sin(x^2+y^2)}{x^2+y^2}dxdy=\frac{\pi^2}{2}.$$ Since the function $(x,y)\mapsto\sin(x^2+y^2)/(x^2+y^2)$ is not integrable, I can't use the Theorem of Change of Variable. So, I'm trying to use residue formulae for some suitable holomorphic function to compute the inner integral, but I can't continue. Can someone suggest me a hint to solve this problem?

Addendum: I may be wrong, but I suspect Theorem of Change of Variable (TCV) is not the answer. The reason is the following: the number $\pi^2/2$ is gotten if we apply polar coordinates, but TCV guarantees that if we apply any other change of variable we can get the same number, $\pi^2/2$. If this function were integrable, this invariance property would be guaranteed, but it is not the case. Thus we may have strange solutions to this integral.

  • Note that the inner integral exists in the sense of the Lebesgue integral, but the outer integral is improper. – Ángel Valencia Feb 03 '16 at 18:52
  • Converting to polar co-ordinates we get $\pi \int_0^{\infty}u^{-1}\sin u ; du$ but I dk what's next. – DanielWainfleet Feb 03 '16 at 19:09
  • @user254665 there are plenty of methods listed here – user170231 Feb 03 '16 at 19:39
  • @user254665 you can't use the Theorem of Change of Variable because this function is not integrable. Converting to polar co-ordinates is a change of variable. – Ángel Valencia Feb 03 '16 at 19:42
  • 1.If it's not integrable how can you integrate it? It has a removable discontinuity at the origin and is continuously differentiable outside any disc centered at the origin.Integrate outside this disc and let the disc shrink. Then the change of variables is valid. – DanielWainfleet Feb 03 '16 at 20:00
  • Your suggestions are very useful. I integrated this function in an open disc centered in the origin and I take its radius $r\to\infty$. But we have a problem: how can we guarantee that the obtained value with any of these methods (which is $\pi^2/2$) coincides with the iterated integral? We may decide to define the value of this iterated integral as $\pi^2/2$ or we may compute this iterated integral to verify that its value is $\pi^2/2$. – Ángel Valencia Feb 03 '16 at 20:04
  • @user254665 It's badly phrased, but I suspect OP is referring to Fubini's theorem which says that we cannot change the order we integrate over if the function is not absolutely convergent, i.e. we require $\int |f(x,y)|{\rm d}x{\rm d}y$ to exist which is not the case here. – Winther Feb 03 '16 at 20:05
  • The inner integral does exists in the sense of Lebesgue-integral; the outer one is improper and it can be calculated as a limit $\lim_{a\to-\infty,b\to\infty}\int_a^b...$ – Ángel Valencia Feb 03 '16 at 20:06
  • Has anybody an idea how to do this by complex analysis? i would be interested – tired Feb 04 '16 at 03:46

2 Answers2

6

Let $D$ be any Jordan domain in $\mathbb{R}^2$, containing origin in its interior, whose boundary $\partial D$ has the form $r = f(\theta)$ in polar coordinates where $f \in C[0,2\pi]$.

Consider following integral as a functional of $D$: $$\mathcal{I}_D \stackrel{def}{=} \int_D \phi(x,y) dx dy \quad\text{ where }\quad\phi(x,y) = \frac{\sin(x^2+y^2)}{x^2+y^2} $$

Since the origin is a removable singularity for $\phi(x,y)$, as long as $D$ is of finite extent, there isn't any issue about integrability or change of variable. We have $$\mathcal{I}_D = \int_0^{2\pi} \int_0^{f(\theta)}\frac{\sin(r^2)}{r^2} rdr d\theta = \frac12\int_0^{2\pi} \left[\int_0^{f(\theta)^2}\frac{\sin t}{t} dt \right] d\theta $$ For any non-increasing, non-negative function $g$ on $(0,\infty)$. Using integration by part (the RS version), one can show that

$$\left|\int_a^b g(x) \sin(x) dx \right| \le 2 g(a)\quad\text{ for }\quad 0 < a < b < \infty$$

For any $R > 0$ where $B(0,R) \subset D$. By setting $g(x)$ to $1/x$, above inequality leads to following estimate for $\mathcal{I}_D$.

$$\left| \mathcal{I}_D - \mathcal{I}_{B(0,R)} \right| = \frac12 \left| \int_0^{2\pi} \left[\int_{R^2}^{f(\theta)^2}\frac{\sin t}{t} dt \right] d\theta \right| \le \frac12 \int_0^{2\pi} \left|\int_{R^2}^{f(\theta)^2}\frac{\sin t}{t} dt\right| d\theta \le \frac{2\pi}{R^2} $$

For any fixed $Y$, the integrand $\phi(x,y)$ is Lebesgue integrable over $(-\infty,\infty)\times [-Y,Y]$. Double integral of the form below is well defined. With help of DCT, one can evaluate it as a limit

$$\int_{-Y}^Y \int_{-\infty}^{\infty}\phi(x,y) dxdy = \lim_{X\to\infty}\int_{-Y}^Y \int_{-X}^X \phi(x,y) dxdy = \lim_{X\to\infty}\mathcal{I}_{[-X,X]\times[-Y,Y]}$$

We will combine this with above estimation. By setting $R = Y$ and $[-X,X] \times [-Y,Y]$ taking the role of $D$, one get

$$\left|\int_{-Y}^{Y} \int_{-\infty}^{\infty}\phi(x,y) dxdy - \mathcal{I}_{B(0,Y)}\right| \le \limsup_{X\to\infty}\left|\int_{-Y}^{Y} \int_{-X}^{X}\phi(x,y) dxdy - \mathcal{I}_{B(0,Y)}\right| \le \frac{2\pi}{Y^2}$$

Since following two limits exist, $$\lim_{Y\to\infty} \mathcal{I}_{B(0,Y)} = \lim_{Y\to\infty} \pi\int_0^{Y^2}\frac{\sin t}{t}dt = \pi\int_0^\infty \frac{\sin t}{t} dt = \frac{\pi^2}{2} \quad\text{ and }\quad \lim_{Y\to\infty}\frac{2\pi}{Y^2} = 0$$ By squeezing, the double integral at hand exists as an improper integral!

$$\int_{-\infty}^\infty \int_{-\infty}^\infty \phi(x,y) dxdy \stackrel{def}{=} \lim_{Y\to\infty} \int_{-Y}^Y \int_{-\infty}^\infty \phi(x,y) dxdy = \lim_{Y\to\infty} \mathcal{I}_{B(0,Y)} = \frac{\pi^2}{2}$$

achille hui
  • 125,323
4

using polar coordinates we have:

$$ \lim_{R\rightarrow\infty}\int_{x^2+y^2 \le R^2} \frac{\sin(x^2+y^2)}{x^2+y^2}dxdy=\lim_{R\rightarrow\infty}\int_0^{R}\int_0^{2\pi}\frac{\sin (r^2)}{r^2}r d\theta dr= $$ $$ =2\pi\int_0^{+\infty} \frac{\sin (r^2)}{r}dr=2\pi \frac{\pi}{4}=\frac{\pi^2}{2} $$

Disintegrating By Parts
  • 91,908
  • 6
  • 76
  • 168
Emilio Novati
  • 64,377
  • I would like to know why the downvote. I want to learn from my errors ! – Emilio Novati Feb 03 '16 at 20:18
  • it seems correct, the question now is "are there other ways to compute that"? – reuns Feb 03 '16 at 20:20
  • I can only guess, but a likely reason is that it looks like you did not read the question. OP says he/she thinks one "can't use the Theorem of Change of Variable" i.e. he/she thinks he cannot use polar coordinates here since the function is not absolutely integrable. – Winther Feb 03 '16 at 20:21
  • I was thinking to give the same answer! +1 – Mhenni Benghorbal Feb 03 '16 at 20:24
  • @Winther : the OP is wrong when he thinks that the change of variable can't be applied, so... – reuns Feb 03 '16 at 20:25
  • I think I'm not being clear. We would use the Theorem of Change of Variable if the function $\sin(x^2+y^2)/(x^2+y^2)$ were Lebesgue-integrable. But it is not the case: after the change to polar coordinates, you have to take the limit $\lim_{t\to\infty}\int_0^t(\sin{r^2}/r)dr$ beacuse this function is not integrable: $\int_0^\infty|\sin{r^2}/r|dr=\infty$. Now, the problem refers to an iterated integral: I want to integrate it at first wrt one variable and then wrt the other variable. – Ángel Valencia Feb 03 '16 at 20:26
  • Maybe that, since I've a physics formation, I'm not sensitive to subtle math problems, but it seems to me that $\sin(r^2)/r$ has a removable discontinuity at $r=0$ so I don't see problems on integrate it. I'm wrong? – Emilio Novati Feb 03 '16 at 20:29
  • What I mean is that, even if the function is not integrable, you can apply the Fubini's Theorem to state that the iterated integral is $\pi^2/2$. I think TCV can't apply for this problem because the function is not integrable; I may be wrong, but we may get weird results with other changes of variables. – Ángel Valencia Feb 03 '16 at 20:33
  • @user1952009 If OP is wrong is not the point; this is what OP is asking about. Also it's not 'obvious' that one can just change variables. There are many examples of similar integrals where it fails. – Winther Feb 03 '16 at 22:31
  • @Winther : here $\sin(x^2+y^2)/(x^2+y^2)$ is not absolutely integrable on any region of $\mathbb{R}^2$, but I think that after grouping it by connected regions where it has a constant sign,it will yield an alternated series which is absolutely summable after grouping its terms by $2$. if this was true, then changing the order of summation / exchanging the limits would be justified. – reuns Feb 03 '16 at 22:37
  • @user1952009 I should say that you are right that we can change variables. A proof can be found in achille's nice answer above. – Winther Feb 03 '16 at 22:40
  • (+1 from me) I modified your answer to include a limit. There's no way to interpret the integral except as some improper integral, and this interpretation is the only sensible one; and it makes your solution the best one. – Disintegrating By Parts Feb 04 '16 at 22:30