8
$\begingroup$

Let $x>0$. I've encountered the following integral: $$ I(x) \ = \ \int_{-\frac{\pi}{4}}^{\frac{\pi}{4}} \frac{\log\left( x^2 + 2 x \sin(\theta) + 1 \right)}{x + 2 \sin(\theta)} d\theta $$

Is there a way to integrate this? I've been looking through Gradshteyn and Ryzhik, and I have found some similar integrals but nothing that looks exactly like this. Maybe through a contour integration?

If there is no way to integrate this, is it possible to derive asymptotics for the above $I(x)$ in the limit $x \to \infty$?

$\endgroup$
  • 3
    $\begingroup$ Try $\displaystyle\mathrm{I}'\left(x\right)$. $\endgroup$ – Felix Marin May 7 '18 at 0:07
  • $\begingroup$ Interesting suggestion, but I get the following mess which is even worse to deal with: $$I'(x) = \int_{-\frac{\pi}{4}}^{\frac{\pi}{4}} \left[ \frac{2 x + 2 \sin(\theta)}{\left(x + 2 \sin(\theta) \right)\left( x^2 +2 x \sin(\theta) + 1 \right)} - \frac{\log\left( x^2 + 2 x \sin(\theta) + 1 \right)}{\left(x + 2 \sin(\theta)\right)^2}\right] d\theta$$ $\endgroup$ – Greg.Paul May 7 '18 at 4:20
  • $\begingroup$ Surprising or not, a CAS is able to produce the antiderivative which is just .... a monster ! $\endgroup$ – Claude Leibovici May 7 '18 at 4:40
  • $\begingroup$ @ClaudeLeibovici What is CAS ? $\endgroup$ – onurcanbektas May 7 '18 at 5:56
  • $\begingroup$ @onurcanbektas. Computer Algebra System. Have a look at en.wikipedia.org/wiki/Computer_algebra_system $\endgroup$ – Claude Leibovici May 7 '18 at 6:08
3
$\begingroup$

As I wrote in a comment, a CAS was able to find the antiderivative (the resulting formula is a few pages long !).

The only thing I was able to do is to expand the integrand as a Taylor series built at $t=0$ and then to integrate between the given bounds. Using the expansion to $O(t^{5})$ the result is $$3840 x^5 \left(x^2+1\right)^4\,I(x)=2 \left(960 \pi x^4+80 \pi ^3 x^2-\pi ^5 \left(x^2-12\right)\right) \left(x^2+1\right)^4 \log \left(x^2+1\right)+\pi ^3 x^2 \left(\pi ^2 \left(3 x^8-42 x^6-97 x^4-82 x^2-24\right)-80 \left(3 x^2+2\right) \left(x^3+x\right)^2\right)$$

Using $x=10^k$ and the above formula, we can compare with the numerical integration. Below are some results for ten significant figures. $$\left( \begin{array}{ccc} k & \text{approximation} & \text{integration} & \color{blue} {\text{Maxim's asymptotics}}\\ 1 & 0.7285409898 & 0.7285572863 & \color{blue} {0.7284940770}\\ 2 & 0.1446859930 & 0.1446860624 & \color{blue} {0.1446860611}\\ 3 & 0.02170136876 & 0.02170136887 &\color{blue} { 0.02170136887}\\ 4 & 0.002893513786 & 0.002893513786 & \color{blue} {0.002893513786}\\ 5 & 0.0003616892206 & 0.0003616892206 & \color{blue} {0.0003616892206}\\ 6 & 0.00004340270647 & 0.00004340270647 & \color{blue} {0.00004340270647} \end{array} \right)$$ Using the result and expanding again as a Taylor series for infinitely large values of $x$ $$I(x)=\pi \frac {\log(x)} x+O\left(\frac{\log(x)}{x^3}\right)$$

Edit

For the evaluation fo the integral, it could be simpler to build a $[1,1]$ Padé approximant at $t=0$. This will give for the integrand $$\frac{a +b\, t}{1+c\, t}=\frac b c+\frac{a-\frac b c}{1+c\,t}$$ (easy to integrate) where $$a=\frac{\log \left(x^2+1\right)}{x}$$ $$b=\frac{\left(x^2+2\right) \log \left(x^2+1\right)-2 x^2}{\left(x^2+1\right) \left(\left(x^2+1\right) \log \left(x^2+1\right)-x^2\right)}$$ $$c=\frac{2 \left(x^2+1\right)^2 \log \left(x^2+1\right)-x^2 \left(3 x^2+2\right)}{x \left(x^2+1\right) \left(\left(x^2+1\right) \log \left(x^2+1\right)-x^2\right)}$$ This would give $$\left( \begin{array}{ccc} k & \text{approximation} & \text{integration} \\ 1 & 0.7290369069 & 0.7285572863 \\ 2 & 0.1446872219 & 0.1446860624 \\ 3 & 0.02170137072 & 0.02170136887 \\ 4 & 0.002893513788 & 0.002893513786 \\ 5 & 0.0003616892207 & 0.0003616892206 \\ 6 & 0.00004340270647 & 0.00004340270647 \end{array} \right)$$

Update

Concerning the initial Taylor expansion, the problem is a little simpler considering $$\frac{\log (a+2 b \sin (t))}{b+2 \sin (t)}=\sum_{n=0}^4 c_i \, t^i + O\left(t^5\right)$$ where $a=x^2+1$ and $b=x$. This leads to $$c_0=\frac{\log (a)}{b}\qquad c_1=\frac{2 \left(b^2-a \log (a)\right)}{a b^2}$$ $$c_2=\frac{4 a^2 \log (a)-2 b^2\left(2 a +b^2\right)}{a^2 b^3}$$ $$c_3=\frac{a^3 \left(b^2-24\right) \log (a)+24 a^2 b^2-(a-12) a b^4+8 b^6}{3 a^3 b^4}$$ $$c_4=\frac{-4 a^4 \left(b^2-12\right) \log (a)-48 a^3 b^2+4 (a-6) a^2 b^4+2 (a-8) a b^6-12 b^8}{3 a^4 b^5}$$ Now $$\int_{-\frac \pi 4}^{\frac \pi 4} \frac{\log (a+2 b \sin (t))}{b+2 \sin (t)}\,dt\approx\frac{\pi}{2} c_0+\frac{\pi ^3}{96} c_2+\frac{\pi ^5 }{2560}c_4$$ Going to asymptotics $$I(x)=\pi\frac{ \log (x)}{x}\left(1+\frac{\pi ^2 \left(80-\pi ^2\right)}{960\, x^2} +\cdots\right)$$

Update

After Maxim's answer, I added in blue the value obtained for the integral using jis/her asymptotics. Much better than mine.

$\endgroup$
  • $\begingroup$ The remainder in the formula for $I(x)$ should be $O(x^{-3} \log x)$. $\endgroup$ – Maxim May 7 '18 at 18:14
  • $\begingroup$ @Maxim. You are indeed very correct ! Thanks for pointing out. By the way, have you been able to do something with this integral ? Cheers. $\endgroup$ – Claude Leibovici May 8 '18 at 4:48
  • 1
    $\begingroup$ I've added an answer. The coefficient at $x^{-3} \ln x$ should be $2 (\pi - 2)$. The problem is that when you expand the integrand to higher orders around $t = 0$, you'll get $x^{-3} \ln x$ contributions from higher order terms as well. $\endgroup$ – Maxim May 8 '18 at 14:39
3
$\begingroup$

If you make the standard substitution $\theta = 2 \arctan u$, the integrand becomes $$\frac {\ln \frac {P(u)} {Q(u)}} {R(u)},$$ where $P, Q, R$ are quadratic polynomials. Factoring the polynomials and formally expanding the logarithm reduces the indefinite integral to a sum of integrals of the form $$\int \frac {\ln u} {u + a} du = \ln u \ln \left( 1 + \frac u a \right) + \operatorname{Li}_2 \left( - \frac u a \right).$$ With the natural choice of the factors, the antiderivative will be continuous in $u$ for large positive $x$, yielding a closed form for the definite integral.

To obtain the asymptotic, you can simply expand the integrand around $x = \infty$ and integrate term by term: $$\frac {\ln (x^2 + 2 x \sin \theta + 1)} {x + 2 \sin \theta} = \\ \frac {2 \ln x} x - \frac {4 \ln x \sin t - 2 \sin t } {x^2} + \frac {8 \ln x \sin^2 t + \cos 2 t - 4 \sin^2 t} {x^3} + o(x^{-3}), \\ I(x) = \frac {\pi \ln x} x + \frac {2 (\pi - 2) \ln x - \pi + 3} {x^3} + o(x^{-3}).$$

$\endgroup$
  • $\begingroup$ Thanks for posting this answer. You are totally right for the asymptotics; your approach is much better than mine. $\endgroup$ – Claude Leibovici May 8 '18 at 14:43
  • $\begingroup$ What is intresting is that $2(\pi-2)\approx 2.28319$ while $\frac{\pi ^3 \left(80-\pi ^2\right)}{960}\approx 2.26509 $ and that $\pi-3= 0.14159$. $\endgroup$ – Claude Leibovici May 8 '18 at 15:06
  • $\begingroup$ Hoping that you don't mind, I added to my answer the value of the integral using the asymptotics you provided. Impressive ! $\endgroup$ – Claude Leibovici May 8 '18 at 15:20
2
$\begingroup$

For $x$ sufficiently large, and putting all in the form

$$2\int_{-\pi/4}^{\pi/4} \frac{\log(x)}{x}\cdot \frac{\textrm{d}\theta}{1+2\sin(\theta)/x}+\int_{-\pi/4}^{\pi/4} \frac{\log\left(1 + 1/x (2\sin(\theta) + 1/x) \right)}{x }\cdot \frac{\textrm{d}\theta}{1+2\sin(\theta)/x} $$ combined with the power series of $\log(1+x)$ and $\displaystyle \frac{1}{1+x}$, and integrating termwise, we arrive, for example, at the asymptotical behaviour

$$I(x)\approx\pi\frac{\log(x)}{x}+\frac{2(\pi-2)\log(x)+3-\pi}{x^3}+\frac{12(3\pi-8)\log(x)-21\pi+64}{6x^5}.$$

$\endgroup$
  • 1
    $\begingroup$ Using this,we almost get the exact value of the integral. For $x=10$, we get $0.7285564014$ and for $x=100$, we get $0.1446860624$. $\endgroup$ – Claude Leibovici May 9 '18 at 3:12

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.