4
$\begingroup$

For context, this gives one way to evaluate the Fresnel sine integral at infinity. The problem I'm running into is

$$ \int_0^\infty \left[ \int_{-\infty}^\infty \vert\sin(x^2)x e^{-t^2 x^2}\vert dt \right] dx$$ $$= \sqrt{\pi} \int_0^\infty \vert \sin(x^2) \vert dx$$ $$ = \infty,$$

so Fubini does not apply. However, naively switching the order does give the right result. Is there a nice way to justify exchanging the order?

$\endgroup$
  • $\begingroup$ I think it might help to put this in a bit more context. If this is from a physics course, it might be a simple lack of rigor. If however it is from a math course, there might be something else going on underneath. $\endgroup$ – Cameron Williams Jul 20 '15 at 2:07
  • $\begingroup$ @Winther There is a pretty steep (exponential) singularity at $t=0$ if you do the integral in the opposite order. I'm not sure Fubini-Tonelli applies here. $\endgroup$ – Cameron Williams Jul 20 '15 at 2:12
  • $\begingroup$ @Winther I believe I've demonstrated that the premises of Fubini-Tonelli fail. $\endgroup$ – Philip Hoskins Jul 20 '15 at 2:13
  • $\begingroup$ @CameronWilliams Yeah. I just noticed, I only focused on the $\infty$-end of the integral. $\endgroup$ – Winther Jul 20 '15 at 2:13
  • $\begingroup$ @Winther It happens. This is a very easily doable Riemann integral but horrific with Lebesgue integration. OP: you may need to interpret this simply as a Riemann integral. The Lebesgue integral fails sometimes with traditional integrals over the whole real line - this is one such case. $\endgroup$ – Cameron Williams Jul 20 '15 at 2:14
4
$\begingroup$

By Fubini, we have $$ \int_0^R \int_{-\infty}^\infty \sin(x^2)x e^{-t^2 x^2} \,dt\, dx = \int_{-\infty}^\infty \int_0^R \sin(x^2)x e^{-t^2 x^2} \,dx \, dt = \int_{-\infty}^\infty f(R,t) \, dt$$ where $$ f(R,t) = \int_0^R \sin(x^2)x e^{-t^2 x^2} \,dx = \frac{1-(\cos(R^2) + t^2 \sin(R^2))e^{-t^2 R^2}}{2(1+t^4)} .$$ (This follows from integrating by parts twice, and solving the resulting equation.) Hence $$ |f(R,t)| \le \frac{2+t^2}{2(1+t^4)} ,$$ which is integrable. Hence by the Lebesgue dominated convergence theorem, we have $$ \lim_{R\to \infty} \int_{-\infty}^\infty f(R,t) \, dt = \int_{-\infty}^\infty \lim_{R\to \infty} f(R,t) \, dt .$$ Hence $$ \int_0^\infty \int_{-\infty}^\infty \sin(x^2)x e^{-t^2 x^2} \,dt\, dx = \int_{-\infty}^\infty \int_0^\infty \sin(x^2)x e^{-t^2 x^2} \, dx \, dt$$

$\endgroup$
  • $\begingroup$ Should it be $2+t^2$ in the numerator? $\endgroup$ – Philip Hoskins Jul 20 '15 at 2:36
  • $\begingroup$ @PhilipHoskins Thanks. I fixed it. $\endgroup$ – Stephen Montgomery-Smith Jul 20 '15 at 2:56
  • $\begingroup$ Cool. Thanks for taking the time to answer my question! $\endgroup$ – Philip Hoskins Jul 20 '15 at 3:14
  • $\begingroup$ Unless I'm wrong, the only reason we can apply Fubini's here in the first step is by knowing that the integrand $g(x,t)=\sin{(x^2)}xe^{-t^2x^2}$ is $L^1$ on $[0,R] \times (-\infty, \infty)$, correct? $\endgroup$ – DaveNine Jul 20 '15 at 3:15
  • $\begingroup$ @DaveNine Yes, and I used Phil Hoskin's argument in showing that. $\endgroup$ – Stephen Montgomery-Smith Jul 20 '15 at 3:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.