24

I want to prove that

$$\int_0^{\pi/2} \frac{\log(\cos x)}{x^2+\log^2(\cos x)}dx = \frac{\pi}{2}\left(1-\frac{1}{\log 2}\right)$$

Anthony
  • 1,553
  • 4
    What have you tried? In what context did you come across the problem? (Also, try to not to formulate your question as a command, i.e. "Prove that". Some people tend to be offended by this.) – mrf Mar 09 '13 at 16:51
  • This is more polite. It's still not a question though, but a statement. The bigger problem is that you offer no context and no hint of your own thoughts, possible failed attempts etc. – mrf Mar 09 '13 at 16:58
  • Well then, go ahead and prove it. – Zarrax Mar 09 '13 at 18:34

3 Answers3

26

Note that for $|x| < \frac{\pi}{2}$, we have

$$ \frac{\log\cos x}{\log^2 \cos x + x^2} = \Re \frac{1}{\log\left( \frac{1+e^{2ix}}{2} \right)}. $$

Thus if $I$ denotes the given integral, we have

\begin{align*} I &= \frac{1}{2}\int_{-\frac{\pi}{2}}^{\frac{\pi}{2}} \Re \frac{1}{\log\left( \frac{1+e^{2ix}}{2} \right)} \, dx \\ &= \frac{1}{4} \mathrm{PV}\int_{-\pi}^{\pi} \Re \frac{1}{\log\left( \frac{1+e^{ix}}{2} \right)} \, dx \\ &= \frac{1}{4} \Re \mathrm{PV}\int_{-\pi}^{\pi} \frac{1}{\log\left( \frac{1+e^{ix}}{2} \right)} \, dx. \end{align*}

Now let $C_{\epsilon}$ be the counter-clockwised contour consisting of the circle of radius 1 centered at the origin, with two semicircular indents $\gamma_{1,\epsilon}$ around $1$ and $\gamma_{2,\epsilon}$ around $-1$ as follows:

enter image description here

By writing

\begin{align*} I = \frac{1}{4} \Re \lim_{\delta\to0^{+}}\left( \int_{-\pi+\delta}^{-\delta} + \int_{\delta}^{\pi-\delta} \right) \frac{1}{\log\left( \frac{1+e^{ix}}{2} \right)} \, dx \end{align*}

and plugging the substitution $z = e^{ix}$, we observe that

\begin{align*} I = \frac{1}{4} \Im \lim_{\epsilon\to 0^{+}}\left(\oint_{C_{\epsilon}} - \int_{\gamma_{1,\epsilon}} - \int_{\gamma_{2,\epsilon}} \right) \frac{dz}{z \log\left(\frac{1+z}{2}\right)} \end{align*}

Let

\begin{align*} f(z) = \frac{1}{z \log\left(\frac{1+z}{2}\right)}. \end{align*}

It is plain from the logarithmic singularity that

\begin{align*} \lim_{\epsilon \to 0^{+}} \int_{\gamma_{2,\epsilon}} f(z) \, dz = 0. \end{align*}

Also it follows that

\begin{align*} \lim_{\epsilon\to 0^{+}} \oint_{C_{\epsilon}} f(z) \, dz &= 2\pi i \operatorname{Res}_{z=0} f(z) = -\frac{2\pi i}{\log 2}, \\ \lim_{\epsilon\to 0^{+}} \int_{\gamma_{1,\epsilon}} f(z) \, dz &= -\pi i \operatorname{Res}_{z=1} f(z) = -2\pi i. \end{align*}

Therefore we have

\begin{align*} I &= \frac{1}{4} \Im \left( 2\pi i - \frac{2\pi i}{\log 2} \right) = \frac{\pi}{2} \left( 1 - \frac{1}{\log 2} \right). \end{align*}

Sangchul Lee
  • 186,861
  • I have no idea why you would bother typing this up for an OP who doesn't care enough about the question to tell us with one single word what (s)he tried. – mrf Mar 09 '13 at 18:10
  • 5
    @mrf, That's because I believe that this posting would also help others who share similar questions someday :) – Sangchul Lee Mar 09 '13 at 18:21
  • 1
    @mrf Bleh, don't make this such a big deal. If you want to answer the question, do so, but don't go around telling people if they should answer or not based on your moral. I think this question is OK as it stands, for example. If the guy shows work, well, then that is great, but maybe he has no clue, or maybe he's lazy. In either case, it is up to you that you care to answer the question. Ultimately, the OP will realize whether showing work or not is to his/her benefit. – Pedro Mar 09 '13 at 18:25
  • 1
    @PeterTamaroff I did not make it a big deal, and I certainly didn't mean to imply that the question shouldn't have been answered. I'm just genuinely curious what drives people to give very good answers that take a lot of effort to type when the OP shows no effort themselves. – mrf Mar 09 '13 at 18:31
  • 1
    @mrf I think sos440 enjoys solving such integrals, for example. Thus, he didn't mind about the lack of effort the OP showed. I also think this is quite a tough integral, so maybe there wasn't much the OP could do. In any case, why should the OP show effort? If it is a homework question, say, then I understand. But maybe the guy is just curious about the integral. – Pedro Mar 09 '13 at 18:38
  • Now that I think about it, the first line has me a bit confused. – Random Variable Mar 09 '13 at 18:38
  • It is a difficult integral, but the OP could have told us where he found it, how he knew the answer, why he wanted to compute it. What kind of techniques he knows, etc, etc etc, – mrf Mar 09 '13 at 18:41
  • 4
    @RandomVariable, It would become clearer if you notice that $$ \log \left( \frac{1+e^{2ix}}{2} \right) = \log (e^{ix} \cos x) = \log \cos x + ix $$ for $|x| < \pi / 2$. Of course, the branch cut is chosen as the negative real axis as usual. – Sangchul Lee Mar 09 '13 at 18:46
  • 1
    Thanks. I should have realized that. Two more quick questions. Why did you have the $\Re$ inside of the integral for two lines? Why wasn't it brought outside initially? And why it is immediately obvious that the branch point at $z=-1$ doesn't cause the integral to blow up when you take the limit? – Random Variable Mar 09 '13 at 19:02
  • 2
    @RandomVariable, For the first question, that's because of the singularity of the imaginary part. That's also why I introduced the principal value notation. And for the second question, I assumed that the reader is somewhat familiar to various contour integrals... but if you want to confirm, you can plug $z = -1 + \epsilon e^{i\theta}$ and take $\epsilon \to 0$. – Sangchul Lee Mar 09 '13 at 19:11
9

Consider $$f(z) = \frac{1}{\log(1-iz)} \frac{1}{1+z^{2}}$$ where the branch cut for $\log (1-iz)$ runs down the imaginary axis from $z=-i$.

Then integrate around a contour that consists of the line segment $[-R,R]$ and the upper half of the circle $|z|=R$, and is indented around the simple pole at the origin.

Since $ z f(z) ->0 $ uniformly as $R \to \infty$, $\displaystyle \int f(z) \ dz$ vanishes along the upper half of $|z|=R$ as $ R \to \infty$.

So we have

$$ \begin{align} \text{PV} \int_{-\infty}^{\infty} \frac{1}{\log(1-ix)} \frac{dx}{1+x^{2}} &= \text{PV} \int_{-\infty}^{\infty} \frac{1}{\frac{1}{2}\log(1+x^{2}) - i\arctan x} \frac{dx}{1+x^{2}} \\ &= \text{PV} \int_{-\infty}^{\infty}\frac{\frac{1}{2} \log(1+x^{2})+ i \arctan x }{\frac{1}{4} \log^{2}(1+x^{2})+ \arctan^{2} x} \frac{dx}{1+x^{2}} \\ &= 2 \pi i \ \text{Res}[f(z), i] + \pi i \ \text{Res}[f(z),0] \\ &= 2 \pi i \left(\frac{1}{2i \log 2} \right) + \pi i \left( -\frac{1}{i} \right) \\ &= \pi \left( \frac{1}{\log 2} - 1\right). \end{align}$$

Equating the real parts on both sides of the equation,

$$ \int_{-\infty}^{\infty} \frac{\frac{1}{2} \log(1+x^{2})}{\frac{1}{4} \log^{2}(1+x^{2}) + \arctan^{2} x} \frac{dx}{1+x^{2}} = \pi \left(\frac{1}{\log 2} -1 \right). $$

Now let $x= \tan u$.

Then

$$ \begin{align} \int_{-\pi /2}^{\pi /2} \frac{\frac{1}{2} \log(\sec^{2}u)}{\frac{1}{4} \log^{2}(\sec^{2}u) + u^{2}} du &= \int_{-\pi/2}^{\pi /2} \frac{\log (\sec u)}{\log^{2}(\sec u)+u^{2}} \ du \\ &= - \int_{-\pi/2}^{\pi/2} \frac{\log(\cos u)}{\log^{2}(\sec u) + u^{2}} \ du \\ &= \pi \left(\frac{1}{\log 2} -1 \right) \end{align}$$

which implies

$$ \int_{0}^{\pi/2} \frac{\log(\cos u)}{u^{2} + \log^{2}(\cos u)} \ du = \frac{\pi}{2} \left(1- \frac{1}{\log 2} \right).$$

6

$\newcommand{\+}{^{\dagger}} \newcommand{\angles}[1]{\left\langle\, #1 \,\right\rangle} \newcommand{\braces}[1]{\left\lbrace\, #1 \,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\, #1 \,\right\rbrack} \newcommand{\ceil}[1]{\,\left\lceil\, #1 \,\right\rceil\,} \newcommand{\dd}{{\rm d}} \newcommand{\down}{\downarrow} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,{\rm e}^{#1}\,} \newcommand{\fermi}{\,{\rm f}} \newcommand{\floor}[1]{\,\left\lfloor #1 \right\rfloor\,} \newcommand{\half}{{1 \over 2}} \newcommand{\ic}{{\rm i}} \newcommand{\iff}{\Longleftrightarrow} \newcommand{\imp}{\Longrightarrow} \newcommand{\isdiv}{\,\left.\right\vert\,} \newcommand{\ket}[1]{\left\vert #1\right\rangle} \newcommand{\ol}[1]{\overline{#1}} \newcommand{\pars}[1]{\left(\, #1 \,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\pp}{{\cal P}} \newcommand{\root}[2][]{\,\sqrt[#1]{\vphantom{\large A}\,#2\,}\,} \newcommand{\sech}{\,{\rm sech}} \newcommand{\sgn}{\,{\rm sgn}} \newcommand{\totald}[3][]{\frac{{\rm d}^{#1} #2}{{\rm d} #3^{#1}}} \newcommand{\ul}[1]{\underline{#1}} \newcommand{\verts}[1]{\left\vert\, #1 \,\right\vert} \newcommand{\wt}[1]{\widetilde{#1}}$ $\ds{\int_{0}^{\pi/2}{\ln\pars{\cos x} \over x^{2} + \ln^{2}\pars{\cos\pars{x}}} \,\dd x = {\pi \over 2}\,\bracks{1 - {1 \over \ln\pars{2}}}}$

${\sf\mbox{The integration can be performed without "leaving" the first quadrant !!!}}$.

We close the arc $\ds{\braces{z = \expo{\ic\theta}\ \mid\ 0 < \theta < {\pi \over 2}}}$ with the 'segments' $\ds{\braces{z=y\ic\ \mid\ y \in \pars{0,1}}}$ and $\ds{\braces{z=x \mid\ x \in \pars{0,1}}}$. The contour is properly indented as explained below.

\begin{align} &\color{#c00000}{% \int_{0}^{\pi/2}{\ln\pars{\cos x} \over x^{2} + \ln^{2}\pars{\cos\pars{x}}}\,\dd x} =\Re\int_{0}^{\pi/2}{\dd x \over \ln\pars{\cos\pars{x}} + x\ic} \\[3mm]&=\Re \int_{\verts{z}\ =\ 1 \atop {\vphantom{\Huge A}0\ <\ {\rm Arg}\pars{z}\ <\ \pi/2}} {1 \over \ln\pars{\bracks{z^{2} + 1}/\bracks{2z}} + \ln\pars{z}} \,{\dd z \over \ic z} \\[3mm]&=\Im \int_{\verts{z}\ =\ 1 \atop {\vphantom{\Huge A}0\ <\ {\rm Arg}\pars{z}\ <\ \pi/2}} {1 \over \ln\pars{\bracks{z^{2} + 1}/2}}\,{\dd z \over z} \\[3mm]&=\Im\left\lbrack\left.% -\int_{0}^{-\pi/2}{1 \over \ln\pars{\bracks{z^{2} + 1}/2}}\,{\dd z \over z} \right\vert_{z\ =\ \ic\ +\ \epsilon\expo{\ic\theta}} -\int_{1 - \epsilon}^{\epsilon}{1 \over \ln\pars{\bracks{-y^{2} + 1}/2}} \,{\ic\,\dd y \over \ic y} \right. \\[3mm]&\phantom{=\Im\left\lbrack a\right.}\color{#00f}{\left. -\int_{\pi/2}^{0}{1 \over \ln\pars{\bracks{z^{2} + 1}/2}}\,{\dd z \over z} \right\vert_{z\ =\ \epsilon\expo{\ic\theta}}} -\int_{\epsilon}^{1 - \epsilon}{1 \over \ln\pars{\bracks{x^{2} + 1}/2}} \,{\dd x \over x} \\[3mm]&\phantom{=\left\lbrack a\right.}\left.\color{#00f}{\left.-\int_{\pi}^{\pi/2} {1 \over \ln\pars{\bracks{z^{2} + 1}/2}}\,{\dd z \over z}\right\vert_{z\ =\ 1\ +\ \epsilon\expo{\ic\theta}}}\right\rbrack \end{align} The above integration is evaluated along the first quadrant, as explained above, and it's $\ds{"\epsilon\mbox{-indented}"}$ around $\ds{z = \ic, 0\ \mbox{and}\ 1}$. In the $\ds{\epsilon \to 0^{+}}$ limit, the contribution to the final result arises from the $\ds{\color{#00f}{\mbox{above blue terms}}}$

\begin{align} &\color{#c00000}{% \int_{0}^{\pi/2}{\ln\pars{\cos x} \over x^{2} + \ln^{2}\pars{\cos\pars{x}}}\,\dd x} \\[3mm]&=\lim_{\epsilon \to 0^{+}}\bracks{\left.% -\pars{-\,{\pi \over 2}}\,{1 \over \ln\pars{\bracks{z^{2} + 1}/2}} \right\vert_{z\ =\ \epsilon\expo{\ic\theta}} -\int_{\pi}^{\pi/2}{\epsilon\expo{\ic\theta}\,\dd\theta \over \ln\pars{1 + \epsilon\expo{\ic\theta} + \epsilon^{2}\expo{2\ic\theta}/2}}} \\[3mm]&=-\,{\pi \over 2}\,{1 \over \ln\pars{2}} + {\pi \over 2} \end{align}

$$\color{#66f}{\large% \int_{0}^{\pi/2}{\ln\pars{\cos x} \over x^{2} + \ln^{2}\pars{\cos\pars{x}}} \,\dd x = {\pi \over 2}\,\bracks{1 - {1 \over \ln\pars{2}}}} \approx {\tt -0.6953} $$

Felix Marin
  • 94,792