49
$\begingroup$

Prove the following integral $$I=\int\limits_{0}^{\frac{\pi}{2}}\dfrac{dx}{1+\sin^2{(\tan{x})}}=\dfrac{\pi}{2\sqrt{2}}\left(\dfrac{e^2+3-2\sqrt{2}}{e^2-3+2\sqrt{2}}\right)$$

This integral result was calculated using Mathematica and I like this integral. But I can't solve it.

My idea:

Let $$\tan{x}=t\Longrightarrow dx=\dfrac{1}{1+t^2}dt$$ so $$I=\int\limits_{0}^{\infty}\dfrac{dt}{1+\sin^2{t}}\cdot \dfrac{1}{1+t^2}$$

then I can't proceed. Can you help me? Thank you.

$\endgroup$
  • $\begingroup$ I tried integrating $f(z) = 1 / ((1 + \sin^2 z)(1 + z^2))$ on a contour including an interval and a semicircle, but computing the residues and summing them will be quite ugly I think. There's symmetry in the residues between the poles at $n\pi \pm i \sinh^{-1}(1)$, but it doesn't really simplify matters. $\endgroup$ – user61527 Mar 29 '14 at 2:50
  • $\begingroup$ Nice identity! Esthetically very pleasing! $\endgroup$ – Markus Scheuer Oct 30 '14 at 10:54
  • 2
    $\begingroup$ This is a particular case of a more general setting see my paper "Exact Evaluation of Some Highly Oscillatory Integrals. Journal of Classical Analysis, 3, Issue 1. (2013). pp. 45-57." files.ele-math.com/articles/jca-03-04.pdf $\endgroup$ – Omran Kouba Oct 31 '14 at 14:50
61
+50
$\begingroup$

First note that

$$ \begin{align} \int_{0}^{\pi /2} \frac{1}{1+ \sin^{2} ( \tan x)} \ dx &= \int_{0}^{\infty} \frac{1}{(1+\sin^{2} t)(1+t^{2})} \ dt \\ &= 2 \int_{0}^{\infty} \frac{1}{(3 - \cos 2t)(1+t^{2})} \ dt \end{align}$$

Then using the identity $$\sum_{k=0}^{\infty} a^{k} \cos(kx) = \frac{1- a \cos x}{1-2a \cos x + a^{2}} , \ \ |a|<1$$

we have

$$1 + 2 \sum_{k=1}^{\infty} a^{k} \cos(kx) = 1 + 2 \left(\frac{1-a \cos x}{1-2a \cos x +a^{2}} -1\right) = \frac{1-a^{2}}{1-2 a \cos x + a^{2}}$$

Therefore,

$$ \begin{align} \int_{0}^{\infty} \frac{1}{(1-2a \cos 2x +a^{2})(1+x^{2})} \ dx &= \frac{1}{1-a^{2}} \int_{0}^{\infty} \Big(1+2 \sum_{k=1}^{\infty} a^{k} \cos(2kx) \Big) \ \frac{1}{1+x^{2}} \ dx \\ &= \frac{\pi}{2} \frac{1}{1-a^{2}} + \frac{2}{1-a^{2}} \sum_{k=1}^{\infty} a^{k} \int_{0}^{\infty} \frac{\cos(2kx)}{1+x^{2}} \ dx \\ &= \frac{\pi}{2} \frac{1}{1-a^{2}} + \frac{\pi}{1-a^{2}} \sum_{k=1}^{\infty} \Big(\frac{a}{e^{2}} \Big)^{k} \\ &= \frac{\pi}{2} \frac{1}{1-a^{2}} + \frac{\pi}{1-a^{2}} \frac{a/e^{2}}{1-a^/e^{2}} \\ &= \frac{\pi}{2} \frac{1}{1-a^{2}} \Big(1+ \frac{2a}{e^{2}-a} \Big) = \frac{\pi}{2} \frac{1}{1-a^{2}} \frac{e^{2}+a}{e^{2}-a} \end{align}$$

Now rewrite the integral as

$$ \frac{1}{1+a^{2}} \int_{0}^{\infty} \frac{1}{(1 - \frac{2a}{1+a^{2}} \cos 2x)(1+x^{2})} \ dx$$

and let $a= 3 - 2 \sqrt{2}$.

Then

$$ \begin{align} \frac{1}{1+a^{2}} \int_{0}^{\infty} \frac{1}{(1- \frac{1}{3} \cos 2x)(1+x^{2})} \ dx &= \frac{3}{1+a^{2}} \int_{0}^{\infty} \frac{1}{(3 - \cos 2x)(1+x^{2})} \ dx \\ &= \frac{\pi}{2} \frac{1}{1-a^{2}} \frac{e^{2} + 3 - 2 \sqrt{2}}{e^{2} - 3 + 2 \sqrt{2}} \end{align}$$

which implies

$$ \begin{align} \int_{0}^{\pi /2} \frac{1}{1+\sin^{2} (\tan x)} \ dx &= \frac{\pi}{3} \frac{1+a^{2}}{1-a^{2}}\frac{e^{2} + 3 - 2 \sqrt{2}}{e^{2} - 3 + 2 \sqrt{2}} \\ &= \frac{\pi}{3} \frac{3}{2 \sqrt{2}}\frac{e^{2} + 3 - 2 \sqrt{2}}{e^{2} - 3 + 2 \sqrt{2}} \\ &= \frac{\pi}{2 \sqrt{2}}\frac{e^{2} + 3 - 2 \sqrt{2}}{e^{2} - 3 + 2 \sqrt{2}} \end{align}$$

$\endgroup$
  • 1
    $\begingroup$ In the fourth display line, the factor $1/(1+x^2)$ seems to have been omitted from the integrand on the RHS. The deduction of the next two lines assumes the result $\int_0^\infty\dfrac{\cos2kx}{1+x^2}\mathrm{d}x=\dfrac{\pi}{2\mathrm{e}^{2k}}.$ To me, this looks nontrivial; and perhaps a proof is called for. $\endgroup$ – John Bentin Mar 29 '14 at 9:24
  • $\begingroup$ @JohnBentin Thanks for pointing out the omission. And there are several threads on here about evaluating that integral. math.stackexchange.com/questions/272622/… $\endgroup$ – Random Variable Mar 29 '14 at 11:38
  • $\begingroup$ Aren't the upper limits different from OP? $\endgroup$ – orion Mar 29 '14 at 11:49
  • 2
    $\begingroup$ Oh, nevermind. Nice manipulation of the cosine series. $\endgroup$ – orion Mar 29 '14 at 11:53
  • 1
    $\begingroup$ Very Nice method RV ! $\endgroup$ – Zaid Alyafeai May 4 '14 at 20:56
14
$\begingroup$

Let $f : [-1,1] \to \mathbb{R}$ be any continuous even function on $[-1,1]$. Consider following integral

$$I_f \stackrel{def}{=} \int_0^{\pi/2} f(\sin(\tan x)) dx = \frac12 \int_{-\pi/2}^{\pi/2} f(\sin(\tan x)) dx = \frac12 \int_{-\infty}^{\infty} f(\sin y)\frac{dy}{1+y^2} $$ where $y = \tan x$. Since $f(\cdot)$ is even, $f(\sin y)$ is periodic in $y$ with period $\pi$. We can rewrite $I_f$ as $$ I_f = \frac12 \int_{-\pi/2}^{\pi/2} f(\sin y)g(y) dy $$ where $$g(y) = \sum_{n=-\infty}^\infty \frac{1}{1+(y+n\pi)^2} = \frac{1}{2i}\sum_{n=-\infty}^\infty \left(\frac{1}{n\pi + y - i} - \frac{1}{n\pi+y + i}\right)$$ Let $z = \tan y$ and $T = -i\tan i = \tanh 1 = \frac{e^2-1}{e^2+1}$. Recall following expansion of $\cot y$,

$$\cot y = \frac{1}{y} + \sum_{n \in \mathbb{Z} \setminus \{0\}}\left( \frac{1}{y+n\pi} - \frac{1}{n\pi}\right)$$

We find $$g(y) = \frac{1}{2i}(\cot(y-i) - \cot(y+i)) = \frac{1}{2i} \left(\frac{1 + izT}{z - iT} - \frac{1 - izT}{z + iT}\right) = \frac{(z^2+1)T}{z^2+T^2} $$ As a result, we can get rid of the explicit trigonometric dependence in $I_f$: $$ I_f = \frac12 \int_{-\infty}^\infty f\left(\frac{z}{\sqrt{1+z^2}}\right) \frac{(z^2+1)T}{z^2+T^2} \frac{dz}{1+z^2} = \frac{T}{2}\int_{-\infty}^\infty f\left(\frac{z}{\sqrt{1+z^2}}\right) \frac{dz}{z^2+T^2} $$ For any $a > 0$, let $b = \sqrt{a^2+1}$ and apply above formula to $f_{a}(z) \stackrel{def}{=} \frac{1}{a^2+z^2}$, we have $$\begin{align} I_{f_a} &= \frac{T}{2}\int_{-\infty}^\infty \frac{z^2+1}{a^2+b^2z^2}\frac{dz}{z^2+T^2}\\ &= \frac{T}{2(a^2-b^2T^2)}\int_{-\infty}^\infty \left(\frac{1-T^2}{z^2+T^2} - \frac{1}{a^2+b^2z^2}\right) dz\\ &= \frac{T}{2(a^2-b^2T^2)}\left((1-T^2)\frac{\pi}{T} - \frac{\pi}{ab}\right) = \frac{\pi}{2ab}\frac{b + aT}{a + bT} \end{align} $$ When $a = 1$, $b$ becomes $\sqrt{2}$ and $I_{f_1}$ reduces to the integral $I$ we want to compute, i.e. $$I = I_{f_1} = \frac{\pi}{2\sqrt{2}}\frac{\sqrt{2}+\frac{e^2-1}{e^2+1}}{1+\sqrt{2}\frac{e^2-1}{e^2+1}} = \frac{\pi}{2\sqrt{2}}\frac{(\sqrt{2}+1)e^2 + (\sqrt{2}-1)}{(\sqrt{2}+1)e^2 - (\sqrt{2}-1)} = \frac{\pi}{2\sqrt{2}}\frac{e^2 + (\sqrt{2}-1)^2}{e^2 - (\sqrt{2}-1)^2} $$

$\endgroup$
  • $\begingroup$ Somehow our methods have something in common, but i don't really how your answer formally connects to by more straightforward approach. But nice work (+1) $\endgroup$ – tired Mar 31 '15 at 1:01
5
$\begingroup$

I hope nobody cares that i exhume this question, but i found it interesting that it is possible to obtain this integral by a relativly straightforward contour integration method.

Observe that,following the question opener and using parity, that we can rewrite the integral as

$$ \frac{1}{2}\int^{\infty}_{-\infty}\frac{1}{1+t^2}\frac{1}{1+\sin^2(t)} $$

It is now easy to show that the poles are

$$ t_{\pm}=\pm i\\ t_{n\pm}=\pi n\pm i \text{arcsinh(1)} $$

so we have two isolated poles and the rest lies on two straight lines paralell to the real axis.

Because the integrand interpreted as a complex function converges as $|z|\rightarrow\infty$ we can choose a semicircle closed in the upper half plane as an integration contour. We find

$$ I=\pi i\sum_{n=-\infty}^{\infty}\text{res}(t_{n+})+\pi i \text{res}(t_{+}) $$

Where the residues are given by $$ \text{res}(t_{+})=\frac{i}{2}\frac{1}{2 \sinh^2(1)-1}\\ \text{res}(t_{n+})=\frac{-i}{2\sqrt{2}}\frac{1}{1+(n \pi+i \text{arcsinh(1)} )^2} $$

Therefore the integral reduces to the following sum

$$ I=\frac{\pi}{2\sqrt{2}} \sum_{n=-\infty}^{\infty} \frac{1}{1+(n \pi+i \text{arcsinh(1)})^2} -\frac{\pi}{2}\frac{1}{2 \sinh^2(1)-1} $$

Using a partial fraction decomposition together with the Mittag-Leffler expansion of $\coth(x)$, this can be rewritten as

$$ I=\frac{\pi}{4\sqrt{2}} \sum_{n=-\infty}^{\infty} \frac{-i}{-i+n \pi+ \text{arcsinh(1)}}+ \frac{i}{i+n \pi+ i\text{arcsinh(1)}}-\frac{\pi}{2}\frac{1}{2 \sinh^2(1)-1}=\\ \frac{\sqrt{2} \pi}{8} \left( \coth \left(1-\text{arcsinh(1)}\right)+ \coth \left(1+\text{arcsinh(1)}\right)\right)-\frac{\pi}{2}\frac{1}{2 \sinh^2(1)-1}\\ $$

Or $$ I\approx 1.16353 $$

Which matches the claimed result.

One can also compute this explicitly noting that $\text{arcsinh(1)}=\log(1+\sqrt{2})$ (*). But this is rather tedious so i just leave this step to the reader and conclude that $$ I=\frac{e^2+3-2\sqrt{2}}{e^2-3+2\sqrt{2}} $$

Appendix

Just to give some details of the last part of the calculations:

Using (*) the part stemming from the sum is $$ \frac{\pi}{4\sqrt{2}}\left(\frac{ \frac{1+\sqrt{2}}{e}+\frac{e}{1+\sqrt{2}}}{ \frac{e}{1+\sqrt{2}}-\frac{1+\sqrt{2}}{e}}+\frac{e \left(1+\sqrt{2}\right)+\frac{1}{1+\sqrt{2} e} }{\left(1+\sqrt{2}\right) e-\frac{1}{\left(1+\sqrt{2}\right) e}}\right)=\\ \frac{\left(e^4-1\right) \pi }{2 \sqrt{2} \left(1-6 e^2+e^4\right)} $$

The part of the single pole gives

$$ \frac{\pi }{2 \left(\left(\frac{e}{2}-\frac{1}{2 e}\right)^2-1\right)}=\frac{2 e^2 \pi }{1-6 e^2+e^4} $$

Adding both terms and factorizing then yields the desired result

$\endgroup$
  • $\begingroup$ Nice solution. +1. Is there a way to choose another contour than the semicircle to arrive at finite number of residues without the current infinite number of residues and the expansion of $\coth x$ as bridge? $\endgroup$ – Hans Sep 24 '16 at 8:26

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.