19

Question:

Find the integral $$I=\int_{-\frac{\pi}{2}}^{\frac{\pi}{2}}\dfrac{x\sin{x}}{1+\cos^4{x}}dx$$

my try: since $$I=2\int_{0}^{\frac{\pi}{2}}\dfrac{x\sin{x}}{1+\cos^4{x}}dx$$

then I can't.

I know this follow integral $$\int_{0}^{\pi}\dfrac{x\cos{x}}{1+\sin^2{x}}dx=(arcsinh{1})^2-(\arcsin1)^2$$ (this nice integral is sos440 solve it) $$\int_{0}^{\pi}\dfrac{x\sin{x}}{1+\cos^2{x}}dx=(\arcsin{1})^2-0^2$$ (this is very easy integral),because use $$\int_{0}^{\pi}xf(\sin{x})dx=\pi\int_{0}^{\frac{\pi}{2}}f(\sin{x})dx$$

But I can't my problem,Thank you very much!

Harry Peter
  • 7,819
math110
  • 93,304

6 Answers6

7

We have $$I = 2 \int_0^{\pi/2} \dfrac{x \sin(x)}{1+\cos^4(x)} dx = 2 \sum_{k=0}^{\infty} (-1)^k\int_0^{\pi/2} x \sin(x) \cos^{4k}(x)dx$$ Now let $\cos(x) =t$. We then get $$I_k = \int_0^{\pi/2} x \sin(x) \cos^{4k}(x)dx = \int_0^1 t^{4k} \arccos(t) dt$$ We now have $$\int t^{4k} \arccos(t) dt = \dfrac{t^{4k+1}}{2(8k^2+6k+1)}(t _2F_1(1/2,2k+1;2k+2;t^2) + 2(2k+1) \arccos(t)) + c$$ Hence, \begin{align} \int_0^1 t^{4k} \arccos(t) dt & = \dfrac1{2(8k^2+6k+1)}(_2F_1(1/2,2k+1;2k+2;1) + 2(2k+1) \arccos(1))\\ & = \dfrac{\sqrt{\pi}}{(4k+1)^2} \dfrac{\Gamma(2k+1)}{\Gamma(2k+1/2)} \end{align} Hence, \begin{align} I & = 2 \sum_{k=0}^{\infty} (-1)^k \dfrac{\sqrt{\pi}}{(4k+1)^2} \dfrac{\Gamma(2k+1)}{\Gamma(2k+1/2)}\\ & = 2 _4F_3(1/4,1/2,1,1;3/4,5/4,5/4;-1)\\ & \approx 1.845096\ldots \end{align} where the last step is nothing but the definition of the appropriate generalized hypergeometric series, i.e., $$_4F_3(1/4,1/2,1,1;3/4,5/4,5/4;z) = \sum_{k=0}^{\infty} z^k \dfrac{\sqrt{\pi}}{(4k+1)^2} \dfrac{\Gamma(2k+1)}{\Gamma(2k+1/2)}$$

2

$\newcommand{\+}{^{\dagger}}% \newcommand{\angles}[1]{\left\langle #1 \right\rangle}% \newcommand{\braces}[1]{\left\lbrace #1 \right\rbrace}% \newcommand{\bracks}[1]{\left\lbrack #1 \right\rbrack}% \newcommand{\ceil}[1]{\,\left\lceil #1 \right\rceil\,}% \newcommand{\dd}{{\rm d}}% \newcommand{\ds}[1]{\displaystyle{#1}}% \newcommand{\equalby}[1]{{#1 \atop {= \atop \vphantom{\huge A}}}}% \newcommand{\expo}[1]{\,{\rm e}^{#1}\,}% \newcommand{\fermi}{\,{\rm f}}% \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,}% \newcommand{\half}{{1 \over 2}}% \newcommand{\ic}{{\rm i}}% \newcommand{\iff}{\Longleftrightarrow} \newcommand{\imp}{\Longrightarrow}% \newcommand{\isdiv}{\,\left.\right\vert\,}% \newcommand{\ket}[1]{\left\vert #1\right\rangle}% \newcommand{\ol}[1]{\overline{#1}}% \newcommand{\pars}[1]{\left( #1 \right)}% \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}}% \newcommand{\root}[2][]{\,\sqrt[#1]{\,#2\,}\,}% \newcommand{\sech}{\,{\rm sech}}% \newcommand{\sgn}{\,{\rm sgn}}% \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\ul}[1]{\underline{#1}}% \newcommand{\verts}[1]{\left\vert\, #1 \,\right\vert}$ $\ds{% I \equiv \int_{-\pi/2}^{\pi/2}{x\sin\pars{x} \over 1 + \cos^{4}\pars{x}}\,\dd x:\ {\large ?}}$

$\large\tt\mbox{Hint:}$ \begin{align} I &= 2\int_{0}^{\pi/2}x\sin\pars{x}\,{1 \over 2\expo{\ic\pi/2}}\bracks{% {1 \over \cos^{2}\pars{x} - \expo{\ic\pi/2}} - {1 \over \cos^{2}\pars{x} + \expo{\ic\pi/2}}}\,\dd x \\[3mm]&= 2\Im\int_{0}^{\pi/2}{x\sin\pars{x} \over \cos^{2}\pars{x} - \expo{\ic\pi/2}}\,\dd x = 2\Im\int_{0}^{\pi/2}{x\sin\pars{x} \over 2\expo{\ic\pi/4}}\bracks{% {1 \over \cos\pars{x} - \expo{\ic\pi/4}} - {1 \over \cos\pars{x} + \expo{\ic\pi/4}}}\,\dd x \\[3mm]&= \Im\int_{0}^{\pi/2}x\sin\pars{x}\,{\root{2} \over 2}\pars{1 - \ic} \braces{2\ic\,\Im\bracks{1 \over \cos\pars{x} - \expo{\ic\pi/4}}}\,\dd x \\[3mm]&= \root{2}\Im\int_{0}^{\pi/2}{x\sin\pars{x} \over \cos\pars{x} - \expo{\ic\pi/4}} \,\dd x \end{align}

G&R-$7^{\ul{\rm a}}$ ed. has an identity $\pars{~{\bf 2.647}.2,\ \mbox{pag.}\ 224~}$ which seems close to this integral but unfortunately it's only valid for $\color{#0000ff}{\large m \not= 1}$: $$ \int{x^{n}\sin\pars{x}\,\dd x \over \bracks{a + b\cos\pars{x}}^{m}} = {x^{n} \over \pars{m - 1}\bracks{a + b\cos\pars{x}}^{m - 1}} - {n \over \pars{m - 1}b}\int{x^{n - 1}\,\dd x \over \bracks{a + b\cos\pars{x}}^{m - 1}} $$

Felix Marin
  • 89,464
1

We can use residue calculus here, although my attempt trades one nasty integral for a slightly simpler one that can be evaluated in terms of hypergeometric functions with a CAS.


Evaluation of $I$

$$\begin{align*} I &= \int_{-\tfrac\pi2}^{\tfrac\pi2} \frac{x \sin x}{1+\cos^4x} \, dx \\ &= \int_{-\infty}^\infty \frac{y \sqrt{y^2+1}}{\left(y^2+1\right)^2+1} \arctan y \, dy \tag1 \\ &= i2\pi \sum_{\operatorname{poles}\in\Gamma} \operatorname{Res} \left[\frac{z \sqrt{z^2+1}}{\left(z^2+1\right)^2+1} \arctan z\right] - \underbrace{\int_1^\infty \frac{r \sqrt{r^2-1}}{\left(r^2-1\right)^2+1} \log\frac{r-1}{r+1} \, dr}_J \tag2 \\ &= i2\pi \left(\frac{e^{-\tfrac{i\pi}4}}2 \arctan\sqrt{-1+i} - \frac{e^{\tfrac{i\pi}4}}2 \arctan\sqrt{-1-i}\right) - J \tag3 \\ &= \frac\pi{2\sqrt2} \left(\pi - \arctan\sqrt{2+2\sqrt2} + \log \sqrt{\frac{1+\sqrt2-\sqrt{2+2\sqrt2}}{1+\sqrt2+\sqrt{2+2\sqrt2}}}\right) - J \end{align*}$$

In the output from $(\star)$ linked below, recall $\arctan x+\arctan\dfrac1x=\dfrac\pi2$ if $x>0$, and observe $$\frac1{\sqrt2}-\frac12 = \frac12 \left(-1+\sqrt2\right) = \frac1{2\left(1+\sqrt2\right)}$$

to simplify the non-hypergeometric term in our result to

$$I = \frac\pi{\sqrt2} \ln\left(2 \sqrt[4]{\frac{1+\sqrt2-\sqrt{2+2\sqrt2}}{1+\sqrt2+\sqrt{2+2\sqrt2}}}\right) + 2 \, {}_4F_3 \left(\left.\begin{array}{c}\frac14,\frac12,1,1 \\ \frac34,\frac54,\frac54\end{array}\right\rvert {-1}\right) + \frac{3\pi}{32\sqrt2} \, {}_4F_3 \left(\left.\begin{array}{c}1,1,\frac54,\frac74 \\ \frac32,2,2\end{array}\right\rvert {-1}\right)$$

We happen to have the identity,

$$\begin{align*} {}_4F_3 \left(\left.\begin{array}{c}1,1,\frac54,\frac74 \\ \frac32,2,2\end{array}\right\rvert {-t}\right) &= \frac{16}{3t}\ln\left(\frac14\sqrt{1+\sqrt{1+t}}\left(\sqrt{1+\sqrt{1+t}}+\sqrt{2}\right)\right) \\[1ex] \implies {}_4F_3 \left(\left.\begin{array}{c}1,1,\frac54,\frac74 \\ \frac32,2,2\end{array}\right\rvert {-1}\right) &= \frac{32}3 \ln\left(\frac12 \sqrt[4]{\frac{1+\sqrt2+\sqrt{2+2\sqrt2}}{1+\sqrt2-\sqrt{2+2\sqrt2}}}\right) \end{align*}$$

so we reduce the result further to a single hypergeometric expression,

$$I = \boxed{2 \, {}_4F_3 \left(\left.\begin{array}{c}\frac14,\frac12,1,1 \\ \frac34,\frac54,\frac54\end{array}\right\rvert {-1}\right)}$$


Massaging $J$

Substituting $t^2=\dfrac{r-1}{r+1}$ then $2u=\dfrac1t-t$ yields a "simpler" integral,

$$\begin{align*} J &= \int_1^\infty \frac{r \sqrt{r^2-1}}{\left(r^2-1\right)^2+1} \log\frac{r-1}{r+1} \, dr \\ &= \int_0^1 \frac{16t^2(t^2+1)}{t^8-4t^6+22t^4-4t^2+1}\log t\,dt \\ &= \int_0^1 \frac{16\left(1+\frac1{t^2}\right)}{t^4+\frac1{t^4} - 4\left(t^2+\frac1{t^2}\right) + 22} \log t \, dt \\ &= 2 \int_0^\infty \frac{\log\left(\sqrt{u^2+1}-u\right)}{u^4+1} \, du \\ &= -2 \int_0^\infty \frac{\operatorname{arsinh} u}{u^4+1} \, du \end{align*}$$

WolframAlpha $(\star)$ can evaluate $J$ and even finds an antiderivative.


Explanation

  • $(1)$ : substitute $y=\tan x$
  • $(2)$ : residue theorem; see below for more details
  • $(3)$ : compute the residues at $-\sqrt{-1-i}=\sqrt[4]{2}\,e^{i\tfrac{5\pi}8}$ and $\sqrt{-1+i}=\sqrt[4]{2}\,e^{i\tfrac{3\pi}8}$ (using the convention $\sqrt z=\sqrt{\left|z\right|}\,e^{\tfrac i2\arg z}$ and $-\pi<\arg z\le \pi$); to simplify, we have

$$\begin{align*} \arctan\sqrt{-1+i} &= \frac i2 \log \frac{1-i\sqrt{-1+i}}{1+i\sqrt{-1+i}} \\ &= \frac i2 \log\left(-2i - 1 - 2\sqrt{-1+i}\right) \\ &= \frac{\pi - \arctan \sqrt{2+2\sqrt2}}2 + i \log\sqrt{1+\sqrt2+\sqrt{2+2\sqrt2}} \\[2ex] \arctan\sqrt{-1\color{red}{-}i} &= \frac{\pi - \arctan \sqrt{2+2\sqrt2}}2 + i \log\sqrt{1+\sqrt2\color{red}{-}\sqrt{2+2\sqrt2}} \end{align*}$$


Applying the residue theorem

Let $f(z)$ be the integrand, swapping $y$ for the complex variable $z$, and let $\Gamma$ denote an indented, semicircular contour. We make a cut along the imaginary axis at $[i,i\infty)$ (as well as along $[-i,-i\infty)$, but that's not totally relevant here). $\Gamma$ is composed of a line segment of length $2R$; a large, broken semicircle centered at the origin of radius $R$; a smaller, broken circle centered at $+i$ of radius $\varepsilon$; and two opposing line segments joining the arcs. Here's a rough sketch:

enter image description here

I claim without proof that the integrals along the circular contours vanish as $R\to\infty$ and as $\varepsilon\to0$.

On our selected branch we have

$$\begin{align*} \sqrt{z^2+1} &= \sqrt{\left|z^2+1\right|} e^{\tfrac i2 \arg(z^2+1)} \\[1ex] \arctan z &= \frac i2 \log \frac{1-iz}{1+iz} \\ &= \frac i2 \left(\log\left|\frac{1-iz}{1+iz}\right| + i \arg\frac{1-iz}{1+iz}\right) \end{align*}$$

Both $\sqrt{z^2+1}$ and $\arctan z$ behave similarly to either side of the cut in the sense that the argument term for both component functions approaches $-\pi$ from the left/outward bank and $+\pi$ from the right/inward bank. In summary,

$$\begin{array}{|c|c|c|} \hline \rm bank & \rm path & f(z) \\ \hline \rm outward & \begin{cases}z = ir e^{i\phi} \\ dz = i e^{i\phi} \, dr \\ r\in[1+\varepsilon,R]\end{cases} & \displaystyle \frac{z\,e^{-\tfrac{i\pi}2}\,\sqrt{\left|z^2+1\right|}}{\left(z^2+1\right)^2+1} \cdot \frac i2 \left(\log\left|\frac{1-iz}{1+iz}\right| - \frac{i\pi}2\right) \\ \hline \rm inward & \begin{cases}z = ir e^{-i\phi} \\ dz = ie^{-i\phi} \, dr \\ r\in[R,1+\varepsilon]\end{cases} & \displaystyle \frac{z \, e^{\tfrac{i\pi}2} \, \sqrt{\left|z^2+1\right|}}{\left(z^2+1\right)^2+1} \cdot \frac i2 \left(\log\left|\frac{1-iz}{1+iz}\right| + \frac{i\pi}2\right) \\ \hline \end{array}$$

As $\phi\to0$, the integrals along the banks converge to, and provide a total contribution of,

$$\begin{align*} \int_{\rm out} f(z) \, dz &= -\frac12 \int_1^\infty \frac{r \sqrt{r^2-1}}{\left(r^2-1\right)^2+1} \left(\log\frac{r+1}{r-1} - i\pi\right) \, dr \\[2ex] \int_{\rm in} f(z) \, dz &= -\frac12 \int_1^\infty \frac{r \sqrt{r^2-1}}{\left(r^2-1\right)^2+1} \left(\log\frac{r+1}{r-1} + i\pi\right) \, dr \\[2ex] \implies \int_{\rm net} f(z) \, dz &= \int_1^\infty \frac{r \sqrt{r^2-1}}{\left(r^2-1\right)^2+1} \log\frac{r-1}{r+1} \, dr \\ \end{align*}$$

user170231
  • 19,334
1

I might be confused, but since $\int \frac{\sin x}{1+\cos^4 x} dx$ is quite easy to integrate (call the integral $F(x),$ it has an arctan and some logs), your integral is easily done by parts, where the answer is $x F(x)\left|_{-\pi/2}^{\pi/2}\right. - \int_{-\pi/2}^{\pi/2} F(x) d x.$

Igor Rivin
  • 25,994
  • 1
  • 19
  • 40
  • You forgot the $x$ before $\sin(x)$. – 0912 Dec 03 '13 at 19:19
  • @0912 No, I did not. – Igor Rivin Dec 03 '13 at 19:19
  • And how do you integrate $F(x)$? –  Dec 03 '13 at 19:19
  • @Sanchez $F(x)$ is a sum of arctans and logs of linear functions of $x.$ All of them can be integrated by parts. – Igor Rivin Dec 03 '13 at 19:21
  • It's not linear functions of $x$, it's a function of $\cos x$ that involves squares and square root. See http://www.wolframalpha.com/input/?i=integrate+1%2F%281%2Bx^4%29 –  Dec 03 '13 at 19:23
  • @Sanchez I am a little dazed, so you are right (forgot to backsubstitute). However, (1) square roots are of numbers, not of cosines (2) squares (of cosines) appear under logs, and by factoring the quadratics you get sums of logs of linear functions of cos. All those can be integrated, admittedly in terms of dilogarithms. – Igor Rivin Dec 03 '13 at 19:26
  • You are right, sorry. 2. I'm not familiar with polylogarithms, but as in comments above 0912 claimed that it has a solution in terms of hypergeometric series, what you said is possibly right then.
  • –  Dec 03 '13 at 19:29
  • @Sanchez Yes, I am pretty sure (despite being half-asleep and forgetting to back-substitute cosines :)) – Igor Rivin Dec 03 '13 at 19:35