4

I realise this question was asked here, but I'm not able to work with any of the answers. The hint given by my professor is

Integrate around the boundary of an indented sector of aperture $\frac{2 \pi}{3}$

but when I try that I can't figure out how to deal with the (divergent) integral along the radial line at angle $2 \pi / 3$. My issue with the accepted answer is that it uses the residue theorem where it doesn't apply, at least as we've learned it, since $$z \mapsto \frac{\log^2z}{z^3 - 1}$$ has non-isolated singularities on the closed region bounded by the proposed contour (due to branch cuts), and I am not sure how to relate the integral along the real axis to one over a contour modified to avoid the branch cut.

For a fixed $\varepsilon > 0$, and for any $\delta \in (0, 1 - \varepsilon)$, we could let $\log_{-\delta / 2}$ be the branch of the logarithmic function with a cut along the ray $\operatorname{arg}z = -\delta / 2$ and define a contour which goes along the positive real axis from $\varepsilon$ to $1 - \delta$, a semicircle in the upper half plane of radius $\delta$ around $1$, the positive real axis from $1 + \delta$ to $2$, an arc of radius $2$ around $0$ with central angle $2 \pi - \delta$, the ray $\operatorname{arg}z = 2 \pi - \delta$ from $r = 2$ to $r = \varepsilon$, and finally an arc of radius $\varepsilon$ around $0$ back to $\varepsilon$. But then, for example, I don't know how to calculate the limit of integral along the arc of radius $\varepsilon$ $$\lim_{\delta \to 0}\int_0^{2 \pi - \delta}\frac{\log_{-\delta / 2}^2(\varepsilon e^{i \theta})}{\varepsilon^3 e^{3 i \theta} - 1} \varepsilon i e^{i \theta} \, d\theta.$$

If I instead try to first use the substitution $x = e^u$ on the real integral and then compute a contour integral, I still get a divergent integral that I don't know how to handle, this time along the top of an indented rectangle.

  • 1
    A so called "keyhole contour" is the standard way to deal with branch cuts. But if you want, you can also use a change of variables on the real integral to get rid of the log so that a keyhole contour is not necessary. Your professor, however, suggests doing a sector. Since it is not a full circle around the origin, you don't have to worry about branch cuts here, either. – Ninad Munshi Dec 06 '20 at 21:38
  • If you mean what I think you do, that's what I was referring to when I said "if we modify the contour to avoid the branch cut". Would the change of variables be $u = e^x$? – themathandlanguagetutor Dec 06 '20 at 21:39
  • There is no non isolated singularity "inside" the contour because you take the limit as the arc around the origin gets closer and closer to that singularity, but from the outside. – Ninad Munshi Dec 06 '20 at 21:41
  • I realised I didn't word my last sentence well. My issue with the contour you suggest, which misses the branch cut, is not one concerning singularities, but how to relate it to the integral over the real axis. As I've been thinking about it, though, I think I'm starting to build an idea of how to do so – themathandlanguagetutor Dec 06 '20 at 21:49
  • Notice that when you parametrize the other leg of the sector, the denominator stays the same. So you will get two of the same integral along the real line that adds up to a constant (as well as other terms from the log). – Ninad Munshi Dec 07 '20 at 00:11
  • That's not true as far as I can tell, because $\operatorname{Log}(r e^{2 \pi i / 3}) = $\ln r + 2 \pi i / 3$, and the integral with this second term diverges. – themathandlanguagetutor Dec 07 '20 at 00:18

2 Answers2

2

Here we discuss two ways of tackling the integral using contour integral. I added Solution 1 to help you implement the hint suggested by the professor. However, I personally recommend you to jump directly to Solution 2.


Solution 1. Consider

$$ f(z) = \frac{\log z}{z^3 - 1}, $$

where $\log(\cdot)$ is the principal complex logarithm. Now we integrate $f$ along the boundary of the indented sector of opening $2\pi/3$:

Contour

Here, the radius of the larger arc $\Gamma_2$ (resp. smaller arc $\Gamma_1$) is $R$ (resp. $\epsilon$) and $0 < \epsilon < 1 < R$. Then it is easy to check that

$$ \left| \int_{\Gamma_1} f(z) \, \mathrm{d}z \right| \leq C \epsilon ( C + \log (1/\epsilon) ) \qquad\text{and}\qquad \left| \int_{\Gamma_2} f(z) \, \mathrm{d}z \right| \leq \frac{C(C + \log R)}{R^2} $$

for $C = 2\pi/3$, and so, the integrals along these curves vanish as $\epsilon \to 0^+$ and $R \to \infty$. So by the residue theorem,

\begin{align*} \int_{L_1} f(z) \, \mathrm{d}z + \int_{L_2} f(z) \, \mathrm{d}z + \int_{L_3} f(z) \, \mathrm{d}z + \int_{\gamma} f(z) \, \mathrm{d}z = o(1) \end{align*}

as $\epsilon \to 0^+$ and $R\to\infty$. However, using the fact that $\omega = e^{2\pi i/3}$ is a simple pole of $f(z)$, the function $(z - \omega)f(z)$ is analytic at $z = \omega$. So

\begin{align*} \lim_{\epsilon \to 0^+} \int_{\gamma} f(z) \, \mathrm{d}z &= \lim_{\epsilon \to 0^+} i\int_{2\pi/3}^{-\pi/3} \epsilon e^{i\theta} f(\omega + \epsilon e^{i\theta}) \, \mathrm{d}\theta \tag{$z=\omega+\epsilon e^{i\theta}$} \\ &= -i \pi \lim_{z \to \omega} (z - \omega)f(z) = -i \pi \mathop{\mathrm{Res}}_{z = \omega} f(z) \\ &= \frac{2\pi^2}{9} \omega. \end{align*}

Moreover,

\begin{align*} \int_{L_2} f(z) \, \mathrm{d}z &= -\omega \int_{1+\epsilon}^{R} f(\omega x) \, \mathrm{d}x = -\omega \int_{1+\epsilon}^{R} \left( f(x) + \frac{2\pi i}{3} \cdot \frac{1}{x^3 - 1} \right) \, \mathrm{d}x \end{align*}

and likewise

\begin{align*} \int_{L_3} f(z) \, \mathrm{d}z = -\omega \int_{\epsilon}^{1-\epsilon} \left( f(x) + \frac{2\pi i}{3} \cdot \frac{1}{x^3 - 1} \right) \, \mathrm{d}x. \end{align*}

Combining altogether and using that $f(z)$ is analytic at $z = 1$,

\begin{align*} (1 - \omega) \int_{\epsilon}^{R} f(x) \, \mathrm{d}x - \frac{2\pi i \omega}{3} \left( \int_{\epsilon}^{1-\epsilon} \frac{1}{x^3 - 1} \, \mathrm{d}x + \int_{1+\epsilon}^{R} \frac{1}{x^3 - 1} \, \mathrm{d}x \right) + \frac{2\pi^2}{9} \omega = o(1). \end{align*}

Letting $\epsilon \to 0^+$ and $R \to \infty$,

\begin{align*} (1 - \omega) \int_{0}^{\infty} f(x) \, \mathrm{d}x - \frac{2\pi i \omega}{3} \left( \mathop{\mathrm{PV}}\! \int_{0}^{\infty} \frac{1}{x^3 - 1} \, \mathrm{d}x \right) + \frac{2\pi^2}{9} \omega = 0. \end{align*}

By noting that

\begin{align*} \mathop{\mathrm{PV}}\! \int_{0}^{\infty} \frac{1}{x^3 - 1} \, \mathrm{d}x &=-\frac{\pi}{3\sqrt{3}}, \end{align*}

we end up with

\begin{align*} \int_{0}^{\infty} f(x) \, \mathrm{d}x = \frac{\omega}{1 - \omega} \left( -\frac{\pi}{3\sqrt{3}} \cdot \frac{2\pi i}{3} - \frac{2\pi^2}{9} \right) = \frac{4\pi^2}{27}. \end{align*}


Solution 2. Here is a more elegant solution. Let $\operatorname{Log}$ denote the complex logarithm chosen so that its argument lies between $0$ and $2\pi$. (Note: Using the principal complex logarithm, this can be realized by $\operatorname{Log}(z) = i\pi + \log(-z)$.) Then consider

$$ g(z) = \frac{(\operatorname{Log}(z) - 2\pi i)\operatorname{Log}(z)}{z^3 - 1}. $$

Then it is not hard to see that, for $x > 0$,

\begin{align*} g(x + i0^+) := \lim_{\epsilon \to 0^+} g(x + i\epsilon) &= \frac{(\log x - 2\pi i)\log x}{x^3 - 1} \\ g(x - i0^+) := \lim_{\epsilon \to 0^+} g(x - i\epsilon) &= \frac{(\log x + 2\pi i)\log x}{x^3 - 1}. \end{align*}

So by using the keyhole contour,

$$ \int_{0}^{\infty} \bigl( g(x + i0^+) - g(x - i0^+) \bigr) \, \mathrm{d}x = 2\pi i \biggl( \mathop{\mathrm{Res}}_{z=e^{2\pi i/3}} g(z) + \mathop{\mathrm{Res}}_{z=e^{4\pi i/3}} g(z) \biggr) $$

Now the left-hand side is

$$ (-4\pi i) \int_{0}^{\infty} \frac{\log x}{x^3 - 1} \, \mathrm{d}x $$

and the right-hand side is

$$ 2\pi i \biggl( \frac{\bigl(\frac{2\pi i}{3} - 2\pi i \bigr)\bigl( \frac{2\pi i}{3} \bigr)}{3 e^{4\pi i}} + \frac{\bigl(\frac{4\pi i}{3} - 2\pi i \bigr)\bigl( \frac{4\pi i}{3} \bigr)}{3 e^{8\pi i}} \biggr) = 2\pi i \left( -\frac{8\pi^2}{27} \right).$$

Therefore the answer is again $\frac{4\pi^2}{27}$.

Sangchul Lee
  • 167,468
  • Thanks for your solution! I eventually realised essentially that I could use the principle value of the portion that was giving me trouble once I stopped trying to have the larger arc be radius 1. It was still an ugly solution. Unfortunately, I think I simply don't have the tools to justify Solution 2. For example, I don't see how I would be able to justify $\int_0^\infty(g(x + i 0^+) - g(x + i 0^-)) , dx = 2 \pi i (\operatorname{Res}(g, e^{2 \pi i / 3}) + (\operatorname{Res}(g, e^{2 \pi i / 3}))$. – themathandlanguagetutor Dec 08 '20 at 11:20
0

$\newcommand{\bbx}[1]{\,\bbox[15px,border:1px groove navy]{\displaystyle{#1}}\,} \newcommand{\braces}[1]{\left\lbrace\,{#1}\,\right\rbrace} \newcommand{\bracks}[1]{\left\lbrack\,{#1}\,\right\rbrack} \newcommand{\dd}{\mathrm{d}} \newcommand{\ds}[1]{\displaystyle{#1}} \newcommand{\expo}[1]{\,\mathrm{e}^{#1}\,} \newcommand{\ic}{\mathrm{i}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\mrm}[1]{\mathrm{#1}} \newcommand{\on}[1]{\operatorname{#1}} \newcommand{\pars}[1]{\left(\,{#1}\,\right)} \newcommand{\partiald}[3][]{\frac{\partial^{#1} #2}{\partial #3^{#1}}} \newcommand{\root}[2][]{\,\sqrt[#1]{\,{#2}\,}\,} \newcommand{\totald}[3][]{\frac{\mathrm{d}^{#1} #2}{\mathrm{d} #3^{#1}}} \newcommand{\verts}[1]{\left\vert\,{#1}\,\right\vert}$ $\ds{\cal C}$ is a key-hole contour which "takes care" of the $\ds{\ln}$-branch cut with $\ds{0 < \arg\pars{z} < 2\pi}$. Integrand poles are $\ds{\expo{2\pi\ic/3}}$ and $\ds{\expo{4\pi\ic/3}}$.


Residue Theorem yields: \begin{align} \oint_{\cal C}{\ln^{2}\pars{z} \over z^{3} - 1}\,\dd z& = 2\pi\ic\bracks{% \pars{{2\pi \over 3}\,\ic}^{2}\,{\expo{2\pi\ic/3} \over 3} + \pars{{4\pi \over 3}\,\ic}^{2}\,{\expo{4\pi\ic/3} \over 3}} \\[5mm] & = -\,{4 \over 27}\pars{3\root{3} - 5\ic}\pi^{3} \end{align}
Note that the singularity at $\ds{z = 1}$ lies on the $\ds{\ln}$-branch cut. Hereafter we properly deals with this fact ( see below ).

Namely, \begin{align} &\bbox[5px,#ffd]{\oint_{\cal C}{\ln^{2}\pars{z} \over z^{3} - 1}\,\dd z} \\[5mm] = &\ \int_{0}^{\infty}{\ln^{2}\pars{x} \over \pars{x - 1 + \ic 0^{+}}\pars{x^{2} + x + 1}}\,\dd x \\[2mm] + &\ \int_{\infty}^{0}{\bracks{\ln\pars{x} + 2\pi\ic}^{2} \over \pars{x - 1 - \ic 0^{+}}\pars{x^{2} + x + 1}}\,\dd x \\[5mm] = &\ \mrm{P.V.}\int_{0}^{\infty}{\ln^{2}\pars{x} \over \pars{x - 1}\pars{x^{2} + x + 1}}\,\dd x \\[2mm] - &\ \mrm{P.V.}\int_{0}^{\infty}{\ln^{2}\pars{x} + 4\pi\ic\ln\pars{x} - 4\pi^{2} \over \pars{x - 1}\pars{x^{2} + x + 1}}\,\dd x + {4\pi^{3} \over 3}\,\ic \\[5mm] = &\ -4\pi\ic\int_{0}^{\infty}{\ln\pars{x} \over x^{3} - 1}\,\dd x + 4\pi^{2}\ \underbrace{\mrm{P.V.}\int_{0}^{\infty}{\dd x \over x^{3} - 1} \,\dd x}_{\ds{-\,{\root{3} \over 9}\,\pi}}\ +\ {4\pi^{3} \over 3}\,\ic \end{align}


\begin{align} &\int_{0}^{\infty}{\ln\pars{x} \over x^{3} - 1}\,\dd x \\[2mm] = &\ {-\,\pars{4/27}\pars{3\root{3} - 5\ic}\pi^{3} - 4\pi^{2}\pars{-\root{3}\pi/9} - 4\pi^{3}\ic/3 \over -4\pi\ic} \\[2mm] & = \bbx{4\pi^{2} \over 27} \approx 1.4622 \\ & \end{align}
Felix Marin
  • 89,464