6

Is there a general solution or pattern in the following series of integrals?

$$S(n)=\idotsint_{-\infty}^{\infty} \frac{\sin(x_1^2+x_2^2+...+x_n^2)}{x_1^2+x_2^2+...+x_n^2}\, dx_1 \dots dx_n$$

I can solve the first integral $$S(1)=\int_{-\infty}^\infty\frac{\sin(x^2)}{x^2} dx$$

$$=2\int_{0}^\infty\frac{\sin(x^2)}{x^2} dx$$ Subbing $t$ for $x^2$: $$=\int_{0}^\infty\frac{\sin(t)}{t^{3/2}} dt$$

Using the Laplace Transform functions, $\mathcal{L}(\sin t)=\frac{1}{1+p^2}$ and $\mathcal{L}^{-1}(t^{-3/2})=\frac{2}{\sqrt\pi}\sqrt{p}$:

$$=\frac{2}{\sqrt{\pi}}\int_{0}^{\infty}\frac{\sqrt{p}}{1+p^2} dp =\frac{4}{\sqrt{\pi}}\int_{0}^{\infty}\frac{x^2}{1+x^4} dx$$

And the last integral has been solved here through various methods to be $\frac{\pi}{2\sqrt{2}}$, meaning $S(1)=\sqrt{2\pi}$.

As another quick example, the second integral

$$S(2)=\iint_{-\infty}^\infty\frac{\sin(x^2+y^2)}{x^2+y^2} dx\, dy$$

can be solved by converting to polar coordinates where $r^2=x^2+y^2$.

$$=\int_0^{\infty}\int_0^{2\pi}\frac{\sin(r^2)}{r} d\theta\, dr =2\pi\int_0^{\infty}\frac{\sin(r^2)}{r} dr =\pi\int_0^{\infty}\frac{\sin(x)}{x} dx =\frac{\pi^2}{2}$$

Obviously, these techniques don't work as more variables of integration are added. What are the higher values of $S(n)$ and is there a general formula?

Ricky
  • 3,148
  • 1
    @whpowell96 Close, but the integral does converge for $n=3$. And from some quick calculations, one can "assign" values to the expression for $n>3$ by treating the integral as the action of a tempered distribution, that was unexpected. – Ninad Munshi Sep 30 '23 at 03:02
  • Oh my bad, I completely blanked on the $r^2$ inside the sine. In that case you should get a classical Fresnel integral for $n=3$ and likely some stationary phase arguments can lead to arguments for some weak convergence for $n>3$. That is indeed surprising. – whpowell96 Sep 30 '23 at 03:24

1 Answers1

8

You have already computed $S(1)$ and $S(2)$. So let's assume that $n\geq 3$. Define: $$f_n(a, b) = \int_{\mathbb R}\dots \int_{\mathbb R} \frac{e^{-(a-ib)(x_1^2+\dots+x_n^2)}}{x_1^2+\dots+x_n^2}dx_1\dots dx_n$$ Clearly, by evaluating at $a=0$ and $b=1$ and taking the imaginary part: $$S(n)=\Im f_n(0, 1)$$ Now $$\begin{split} \frac{\partial f_n}{\partial a}(a,0) &= -\int_{\mathbb R}\dots \int_{\mathbb R} e^{-a(x_1^2+\dots+x_n^2)}dx_1\dots dx_n\\ &=-\left( \int_{\mathbb R}e^{-ax^2}dx\right)^n\\ &=-\left(\frac \pi a\right)^{\frac n 2} \end{split}$$ Thus, noting that $f(a,0)\rightarrow 0$ as $a\rightarrow +\infty$, $$f(a,0)=\int_a^{+\infty}\left(\frac \pi u\right)^{\frac n 2}du=-\frac{\pi^{\frac n 2}}{1-\frac n 2}\cdot\frac {1}{a^{\frac n 2 -1}}\tag{1}$$ Also $$\begin{split} \frac{\partial f_n}{\partial b}(a,b) &= i\int_{\mathbb R}\dots \int_{\mathbb R} e^{-(a-ib)(x_1^2+\dots+x_n^2)}dx_1\dots dx_n\\ &=i\left( \int_{\mathbb R}e^{-(a-ib)x^2}dx\right)^n\\ &=i\left(\frac{\pi}{a-ib}\right)^{\frac n 2} \end{split}$$ Thus $$ f(a,b)-f(a,0)=\int_0^b i\left(\frac{\pi}{a-iv}\right)^{\frac n 2}dv\\ =\frac {\pi^{\frac n 2}} {\frac n 2-1}\cdot \left(\frac 1 {(a-ib)^{\frac n 2 -1}} - \frac 1 {a^{\frac n 2 -1}}\right) $$ Combining with $(1)$ gives $$f(a,b)=\frac {\pi^{\frac n 2}} {\frac n 2-1}\cdot \frac 1 {(a-ib)^{\frac n 2 -1}} $$ Thus $$f(0,1) = \frac {\pi^{\frac n 2}} {\frac n 2-1}\cdot \frac 1 {(-i)^{\frac n 2 -1}} = -\frac {i\pi^{\frac n 2}} {\frac n 2-1}e^{\frac{i\pi n}{4}}$$ and $$\boxed{S(n)=-\frac{\pi^{\frac n 2}}{\frac n 2 -1}\cos\left(\frac {n\pi}4\right)}$$

Edit: As noted by @Ninad Munshi, this formula works for all $n\geq 1$ (not just $n\geq 3$). For $n=2$, you need to take the limit $n\rightarrow 2$.

Stefan Lafon
  • 12,256
  • 11
  • 29
  • 1
    Your function becomes undefined at $n=2$. – Shivansh Jaiswal Oct 02 '23 at 06:50
  • 1
    Please read my answer again: I assumed $n\geq 3$ – Stefan Lafon Oct 02 '23 at 14:36
  • 1
    @ShivanshJaiswal had you done some bare minimum scratch work instead of complaining, you would've discovered that this formula is even more incredible than it appears on first glance. It is surprisingly valid even for $n=1,2$ as it returns the correct values found by OP. The $n=2$ can be verified by taking the limit. – Ninad Munshi Oct 04 '23 at 06:27
  • 1
    @NinadMunshi I had upvoted the answer, I didn't read the first line and was confused about $n=2$ case since I was running into the same problem while attempting the problem myself, which was my motivation to comment this. Complaining wasn't my intent. – Shivansh Jaiswal Oct 04 '23 at 06:47