6

I wish to find a closed form for the integral :

$$\int_0^\infty \frac{\sin{\pi x}}{e^{2 \pi \sqrt{x}}-1} dx$$

which is approximately $0.017621752204741$

I found this integral in a French research paper on (I think) finding symbolic representations of numbers. The integral's value was provided by the paper itself.

I tried using differentiating under the integral with $e^{-ax}$ and substituting $x = u^2$ Writing $\sin(\pi x)$ as the imaginary part of $e^{i \pi x}$ but that didn't help since $x$ grows faster than $\sqrt{x}$.

I think that I need to use contour integration to solve this problem.

uggupuggu
  • 736

2 Answers2

8

Define

$$ I = \int_{-\infty}^{\infty} f(x) \, \mathrm{d}x, \qquad f(z) = \frac{1 - e^{i\pi z^2}}{(e^{\pi z} - e^{-\pi z})^2}. $$

This is related to OP's integral via the following computation:

\begin{align*} \int_{0}^{\infty} \frac{\sin(\pi x)}{e^{2\pi\sqrt{x}} - 1} \, \mathrm{d}x &= \int_{0}^{\infty} \frac{2x \sin(\pi x^2)}{e^{2\pi x} - 1} \, \mathrm{d}x \\ &= \left[ \frac{1 - \cos(\pi x^2)}{\pi (e^{2\pi x} - 1)} \right]_{0}^{\infty} + 2 \int_{0}^{\infty} \frac{e^{2\pi x}(1 - \cos(\pi x^2))}{(e^{2\pi x} - 1)^2} \, \mathrm{d}x \\ &= \operatorname{Re}(I). \end{align*}

By noting that $f(z)$ is meromorphic with poles $\pm i, \pm 2i, \ldots$ and using some estimates, it is not hard to show that for any sufficiently small $\varepsilon > 0$, we can shift the contour of integration of $I$ (red line) to the blue contour as follows:

This shift of contour leads

$$ I = I_1 + I_2, $$

where

\begin{align*} I_1 = \int_{\text{circular arc}} f(z) \, \mathrm{d}z \qquad\text{and}\qquad I_2 = \int_{\text{rays}} f(z) \, \mathrm{d}z. \end{align*}

Now by expanding $f(z)$ about $z = i$, we get

$$ f(z) = \frac{1}{2\pi^2 (z-i)^2} - \frac{1}{2\pi (z-i)} + \mathcal{O}(1), $$

and this immediately yields the asymptotic expansion

\begin{align*} I_1 &= \int_{-\pi}^{0} f(i + \varepsilon e^{i\theta}) (\varepsilon i e^{i\theta}) \, \mathrm{d}\theta \\ &= -\frac{1}{\pi^2 \varepsilon} - \frac{i}{2} + \mathcal{O}(\varepsilon). \end{align*}

On the other hand, noting that $e^{i\pi(x+i)^2} = -e^{i\pi x^2}e^{-2\pi x}$, we get

\begin{align*} I_2 &= \int_{\varepsilon}^{\infty} [f(i+x) + f(i-x)] \, \mathrm{d}x \\ &= \int_{\varepsilon}^{\infty} \frac{2 + e^{i\pi x^2}(e^{2\pi x} + e^{-2\pi x})}{(e^{\pi x} - e^{-\pi x})^2} \, \mathrm{d}x \\ &= \int_{\varepsilon}^{\infty} \frac{4}{(e^{\pi x} - e^{-\pi x})^2} \, \mathrm{d}x - 2 \int_{\varepsilon}^{\infty} \frac{1 - e^{i\pi x^2}}{(e^{\pi x} - e^{-\pi x})^2} \, \mathrm{d}x + \int_{\varepsilon}^{\infty} e^{i\pi x^2} \, \mathrm{d}x \\ &= \frac{2}{\pi(e^{2\pi \varepsilon} - 1)} - I + \int_{0}^{\infty} e^{i\pi x^2} \, \mathrm{d}x + \mathcal{O}(\varepsilon). \end{align*}

Combining altogether and simplifying, it follows that

\begin{align*} I &= \frac{1}{\pi(e^{2\pi \varepsilon} - 1)} - \frac{1}{2\pi^2 \varepsilon} - \frac{i}{4} + \frac{1}{2}\int_{0}^{\infty} e^{i\pi x^2} \, \mathrm{d}x + \mathcal{O}(\varepsilon) \\ &= -\frac{1}{2\pi} - \frac{i}{4} + \frac{1 + i}{4\sqrt{2}} + \mathcal{O}(\varepsilon). \end{align*}

Since the left-hand side is independent of $\varepsilon$, letting $\varepsilon \to 0^+$ and taking real parts only yields

$$ \operatorname{Re}(I) = \frac{1}{4\sqrt{2}} - \frac{1}{2\pi}, $$

confirming @JJacquelin's conjecture.

Sangchul Lee
  • 181,930
3

Not a closed form yet.

$$\frac{\sin(\pi x)}{e^{2 \pi \sqrt{x}}-1}=\sum_{n=0}^\infty e^{-2 \pi (n+1) \sqrt{x}} \sin (\pi x)$$ $$I_n=\int_{0}^\infty e^{-2 \pi (n+1) \sqrt{x}} \sin (\pi x)\,dx$$ $$I_n=\frac{1}{\pi }-(-1)^n\frac{ (n+1) }{\sqrt{2}}\Big(2 S\left((n+1)\sqrt{2}\right)-1\Big)$$ where $S$ is the sine Fresnel integral.

The partial sum $$S_p=\sum_{n=0}^p I_n$$ converge quite fast since, for large $n$

$$I_n=\frac{3}{4 \pi ^3 n^4}\Bigg(1-\frac{4}{n}+\frac{10}{n^2}-\frac{20}{n^3}+O\left(\frac{1}{n^4}\right) \Bigg)$$

Without any approximation $S_{1000}=\color{red}{0.017621752}20$