9
$\begingroup$

I was wondering if anyone knows a closed form for

$$\mathrm{I} = \int_0^{\infty}\sin(x^n)\mathbb{d}x$$

Preliminary evaluations on Wolfram Alpha seem to yield something like this:

$$\mathrm{I} = k\sin\left(\frac{\pi}{2n}\right)\Gamma\left(\frac an \right)$$

where $k$ is a proportionality constant, normally $1$, and $a$ is a constant, normally $1$ as well.

$\endgroup$
  • $\begingroup$ Mathematica gives $$\int_0^\infty \sin x^n dx = \Gamma\left(1 + \frac{1}{n}\right) \sin \frac{\pi}{2n}$$ $\endgroup$ – Henricus V. Nov 30 '15 at 4:39
  • $\begingroup$ I wonder if it could be done as $Im\int_0^{\infty} \exp(i x^n) dx$. $\endgroup$ – marty cohen Nov 30 '15 at 4:45
  • $\begingroup$ This question reminds me that I keep meaning to look up the details of the history of this integral. This is just my own speculation here, but I think the convergence of this integral is such a striking and counter-intuitive result that if a closed form value had not existed, they would have assigned it a Greek letter and called that a "closed form". :) $\endgroup$ – David H Nov 30 '15 at 5:52
  • $\begingroup$ @DavidH: the convergence of this integral is such a striking and counter-intuitive result - See Riemann-Lebesgue lemma. $\endgroup$ – Lucian Nov 30 '15 at 6:44
  • $\begingroup$ Hint: Use Euler's formula in conjunction with the integral expression for the $\Gamma$ function. $\endgroup$ – Lucian Nov 30 '15 at 6:47
8
$\begingroup$

It has a closed form

$$ \int_{0}^{\infty} \exp(i x^{1/\alpha}) \, \Bbb{d}x = \Gamma(1+\alpha)e^{i\pi\alpha/2}, \quad \alpha \in (0, 1). $$

Applying the substitution $x \mapsto x^{\alpha}$, we can write

$$ I(\alpha) := \int_{0}^{\infty} \exp(ix^{1/\alpha}) \, \Bbb{d}x = \int_{0}^{\infty} \alpha x^{\alpha-1} e^{ix} \, \Bbb{d}x, \tag{1} $$

which is much easier to work with.


Complex-analytic method. Rotate the contour of $\text{(1)}$ by $90^{\circ}$ degrees to have

$$ I(\alpha) = \alpha \int_{0}^{i\infty} z^{\alpha-1} e^{iz} \, dz. $$

This is a standard technique, and we can justify this by applying the Cauchy integration formula to a quadrant contour and then utilizing the Jordan's lemma.

Then by the substitution $z \mapsto ix$, we have

$$ I(\alpha) = \alpha i^{\alpha} \int_{0}^{\infty} x^{\alpha-1}e^{-x} \, dx = \Gamma(1+\alpha)i^{\alpha}. $$


Real-analytic method. Since the right-hand side of $\text{(1)}$ exists as improper-integral sense, it follows from the integral version of the Abel's theorem that

$$ I(\alpha) = \int_{0}^{\infty} \alpha x^{\alpha-1} e^{ix} \, \Bbb{d}x = \lim_{\epsilon \downarrow 0} \int_{0}^{\infty} \alpha x^{\alpha-1} e^{ix}e^{-\epsilon x} \, \Bbb{d}x. \tag{2} $$

Using the formula

$$ \frac{\Gamma(1-\alpha)}{x^{1-\alpha}} = \int_{0}^{\infty} u^{-\alpha}e^{-xu} \, \Bbb{d}u $$

and the Fubini's theorem, we can write

\begin{align*} \int_{0}^{\infty} \alpha x^{\alpha-1} e^{ix}e^{-\epsilon x} \, \Bbb{d}x &= \frac{\alpha}{\Gamma(1-\alpha)} \int_{0}^{\infty} \left(\int_{0}^{\infty} u^{-\alpha}e^{-xu} \, \Bbb{d}u\right) e^{-(\epsilon-i) x} \, \Bbb{d}x \\ &= \frac{\alpha}{\Gamma(1-\alpha)} \int_{0}^{\infty} \left(\int_{0}^{\infty} e^{-(u+\epsilon-i) x} \, \Bbb{d}x\right) u^{-\alpha} \, \Bbb{d}u \\ &= \frac{\alpha}{\Gamma(1-\alpha)} \int_{0}^{\infty} \frac{u^{-\alpha}}{u+\epsilon-i} \, \Bbb{d}u. \end{align*}

(Notice that we cannot apply the Fubini's theorem directly when $\epsilon = 0$ due to integrability issue. This explains why we are considering a regularized version $\text{(2)}$ of the original integral.)

Taking $\epsilon \downarrow 0$, we have

$$ I(\alpha) = \frac{\alpha}{\Gamma(1-\alpha)} \int_{0}^{\infty} \frac{u^{1-\alpha} + iu^{-\alpha}}{u^2+1} \, \Bbb{d}u. \tag{3} $$

Using the beta function identity, the last integral in $\text{(3)}$ can be computed explicitly as

$$ I(\alpha) = \frac{\alpha}{\Gamma(1-\alpha)} \cdot \frac{\pi}{2}\left( \frac{1}{\sin(\pi\alpha/2)} + \frac{i}{\cos(\pi\alpha/2)} \right). $$

Finally, from the Euler's reflection formula and the sine double-angle formula we have

$$ I(\alpha) = \Gamma(1+\alpha) e^{i\pi\alpha/2}. $$

$\endgroup$
  • $\begingroup$ Excellent Solutions! +1 $\endgroup$ – Mark Viola Nov 30 '15 at 5:25

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.