8
$\begingroup$

I've been trying to prove Fresnel integrals by real methods and encountered an interesting problem.

Let's start with the known result:

$$\int_0^\infty \sin y^2 dy = \sqrt{\frac{\pi}{8}}$$

Can we prove it without complex methods?

I have tried to do the following:

$$\int_0^\infty \sin y^2 dy =\frac{1}{2} \int_0^\infty \sin x \frac{dx}{\sqrt{x}} =\frac{1}{2} \sum_{n=0}^\infty (-1)^n \int_0^\pi \sin x \frac{dx}{\sqrt{x+\pi n}}$$

This directly follows from the properties of the sine, except the series only converges conditionally, not absolutely, so there's a question of if we can bring it inside the integral.

I will do it without proper justification for now, but if anyone has a comment on this, I would appreciate it.

So we obtain, after a simple change of variables:

$$\int_0^\infty \sin y^2 dy = \frac{\sqrt{\pi}}{2} \int_0^1 \sin \pi t ~ dt \sum_{n=0}^\infty \frac{(-1)^n}{\sqrt{n+t}} $$

Wolfram gives for this series:

$$\sum_{n=0}^\infty \frac{(-1)^n}{\sqrt{n+t}}=\frac{1}{\sqrt{2}}\left( \zeta \left(\frac12, \frac{t}{2}\right)-\zeta \left(\frac12, \frac{t+1}{2}\right) \right)$$

Which is easy enough to show using the definition of Hurwitz zeta function.

Surprisingly enough, this brings us exactly the known value of the Fresnel integral as a coefficient:

$$\int_0^\infty \sin y^2 dy =\sqrt{\frac{\pi}{8}} \int_0^1 \sin \pi t ~ \left( \zeta \left(\frac12, \frac{t}{2}\right)-\zeta \left(\frac12, \frac{t+1}{2}\right) \right) dt$$

Which means that we need to prove the identity in the title of the question:

$$\int_0^1 \sin \pi t ~ \left( \zeta \left(\frac12, \frac{t}{2}\right)-\zeta \left(\frac12, \frac{t+1}{2}\right) \right) dt=1$$

Can we prove this by real methods, not using the Fresnel integral?

Mathematica confirms this identity numerically.

$\endgroup$
  • $\begingroup$ do you mean $$\sin(y^2)$$ or $$\sin^2(y)$$ $\endgroup$ – clathratus Oct 20 '18 at 21:31
  • $\begingroup$ @clathratus, the first one. Fresnel integral is well known, thus I didn't clarify $\endgroup$ – Yuriy S Oct 20 '18 at 21:33
  • $\begingroup$ Right. I wasn't thinking. Thanks. $\endgroup$ – clathratus Oct 20 '18 at 21:34
  • $\begingroup$ This related question has a lot of nice proofs for Fresnel integral math.stackexchange.com/q/2051273/269624 and this one too math.stackexchange.com/q/2739835/269624 $\endgroup$ – Yuriy S Oct 20 '18 at 21:49
  • $\begingroup$ Thank you, these are great. $\endgroup$ – clathratus Oct 20 '18 at 21:55
3
$\begingroup$

I usually don't answer my own questions, but I literally just derived the solution and I think it's beautiful.

So we know from the link in the OP that:

$$\zeta(s,a)=\frac{1}{\Gamma(s)} \int_0^\infty \frac{z^{s-1} dz}{e^{a z} (1-e^{-z})}$$

Which makes our expression:

$$\zeta \left(\frac12, \frac{t}{2}\right)-\zeta \left(\frac12, \frac{t+1}{2}\right)=\frac{1}{\sqrt{\pi}} \int_0^\infty \frac{e^{- t z/2}(1-e^{-z/2}) dz}{(1-e^{-z}) \sqrt{z}}=$$

$$=\frac{2}{\sqrt{\pi}} \int_0^\infty \frac{e^{- t u^2/2}(1-e^{-u^2/2}) du}{1-e^{-u^2}}$$

Now we can see that it's very easy to take integral over $t$ (integration by parts):

$$\int_0^1 \sin \pi t~ e^{- t u^2/2} dt=\frac{4 \pi}{4 \pi^2 +u^4} (1+e^{-u^2/2})$$

Now we substitute this into the second integral to get (this is the beautiful part):

$$8 \sqrt{\pi} \int_0^\infty \frac{(1+e^{-u^2/2})(1-e^{-u^2/2}) du}{(1-e^{-u^2})(4 \pi^2 +u^4)}=8 \sqrt{\pi}\int_0^\infty \frac{du}{4 \pi^2 +u^4}$$

After a change of variables we have:

$$\frac{2 \sqrt{2}}{\pi} \int_0^\infty \frac{dv}{1 +v^4}=\frac{2 \sqrt{2}}{\pi} \frac{\pi}{2 \sqrt{2}}=1$$

Just as it was supposed to be.

God, Mathematics is simply perfect sometimes.


Appendix

Trying to prove in a simple way that the integral formula for $\zeta(1/2,a)$ equals the series definition.

$$ \int_0^\infty \frac{z^{-1/2} dz}{e^{a z} (1-e^{-z})}=2 \int_0^\infty \frac{e^{-a u^2}du}{ 1-e^{-u^2}}=2 \sum_{n=0}^\infty \int_0^\infty e^{-(a+n) u^2} du=$$

$$=2 \sum_{n=0}^\infty \frac{1}{\sqrt{a+n}} \int_0^\infty e^{-w^2} dw= \sqrt{\pi} \sum_{n=0}^\infty \frac{1}{\sqrt{a+n}} $$

Formally, this exactly fits the series definition, but it doesn't converge (it's alright, as the function for $s=1/2$ is defined by analytic continuation).

On the other hand, for the particular function in my case, the proof works well, as the series inside the integral becomes alternating due to a factor $(1-e^{-u^2/2})$ in the numerator, so everyting converges.

I would say that all of this constitutes a nice real proof of the Fresnel integrals, especially since the Poisson integral also has a few real proofs.

Though I'm definitely not claiming this is a new result. It was new for me, but a two second google search found me this paper https://www.jstor.org/stable/2320230, and I'm sure there's plenty more.

$\endgroup$

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.