25
$\begingroup$

What type of singularity is $z=0$ for $\log z$ (any branch)? What is the Laurent series for $\log z$ centered at 0, if exist? If the Laurent series has the form $\sum_{k=-\infty}^{\infty} a_kx^k$, then certainly among $a_{-1},a_{-2},...,a_{-j},...$, at least one is nonzero (or otherwise $\log z$ would be analytic at $0$). Since $\lim_{z\to 0}z\log z=0$, we must have $a_{-j}=0$ for all $j>0$, a contradiction. Hence the Laurent series centered at $0$ cannot exist.

Is the singularity at $0$ a pole? If so, what is its order? Thanks.

$\endgroup$
1
  • $\begingroup$ I've heard it called a, well, logarithmic singularity, but I've never actually seen the phrase defined. $\endgroup$ – user14972 Aug 3 '13 at 8:11
23
$\begingroup$

The singularity of $\log z$ is not an isolated singularity, so the usual classification into, pole, essential, or removable does not apply. In particular, there is no Laurent expansion about 0 and you cannot apply residue theory.

In this case the singularity is known as a branch point, and it is the typical example.

$\endgroup$
7
  • $\begingroup$ Then how about the integral in the link (please refer to my comments above)? Thanks. $\endgroup$ – user70520 Aug 3 '13 at 5:07
  • $\begingroup$ If you're referring to the integral of 1/(z-a) , in the link above then you use the fact that once you select a specific branch, then, within that branch, you have that d/dz(Log(z-a))=1/(z-a), i.e., within a branch, d/dzLogz=1/z . Then, integral 1/(z-a) from a to -a =Log(z-a) from a=1 to a=-1 =Log(z+1)-Log(z-1)=Log(z+1)/Log(z-1). Is that your question; did I understand you correctly? $\endgroup$ – DBFdalwayse Aug 3 '13 at 5:18
  • $\begingroup$ @user70520: The answer given there is incorrect (even if he gets the correct answer for the integral). $z=0$ is not an isolated singularity so the residue is not defined. You can still use residues to calculate the integral but you have to use a contour which can be found here, i.stack.imgur.com/ZjMpl.jpg. $\endgroup$ – Owen Sizemore Aug 3 '13 at 5:24
  • $\begingroup$ But we need the real axis... so I think the contour still need to be semi-circular, but we could exclude the boundary from the original by replacing the interval $[-r,r]$ by the arc $re^{i\theta}$, $\theta\in[0,\pi]$. Then let $r\to0$. Would this fix the solution in the link? $\endgroup$ – user70520 Aug 3 '13 at 5:27
  • $\begingroup$ Aha. I guess your new link is pretty much what I described (except you only considered half of it) after you replaced en.wikipedia.org/wiki/File:Keyhole_contour.svg :) $\endgroup$ – user70520 Aug 3 '13 at 5:30
3
$\begingroup$

The singularity is not a pole, since Logz (or at least its real part) does not blow up to $\infty$ as you approach 0. Maybe the best name is that $0$ is a branch point for Logz; and it is a branch point of infinite index. The branch point tells you that if you wind around the unit circle a point once, i.e., if you wind around by a value of $2\pi$, you do not ever return to your initial value , i.e., if you consider the values of {$Log(z+2n\pi)$} for $n=1,2,3,...$, these are all different values. Compare and contrast this with the case for the n-th root $z^{1/n}$ . Here $0$ is also a branch point for $z^{1/n}$ , but this time it is of index n, because $e^{i\theta/n}=e^{i\theta+2n\pi/n}$ , i.e., you return to the original value of your function after looping n times.

EDIT: As pointed out in the comments, I was wrong in my claim that |Logz| does not go to $\infty$ as $z\rightarrow 0$ ; it does, since lnx does blow up near $0$

$\endgroup$
9
  • 1
    $\begingroup$ But $\lim_{x\to0^+}\log x=-\infty$ $\endgroup$ – user70520 Aug 3 '13 at 4:47
  • $\begingroup$ You're right, but I think you need the function to blow up to $\infty$ in all directions, as in the case of $1/z$. Let me double-check $\endgroup$ – DBFdalwayse Aug 3 '13 at 4:49
  • $\begingroup$ Regarding math.stackexchange.com/questions/434289/… does the residue theory still apply for $z=0$ of $\log z$? $\endgroup$ – user70520 Aug 3 '13 at 4:49
  • $\begingroup$ You also need, in order for f(z) to have a pole at w, for f(z) to be analytic in $\mathbb C \{w}$, which does not happen here. $\endgroup$ – DBFdalwayse Aug 3 '13 at 4:53
  • $\begingroup$ But usually when you apply residue theory, you have a real-valued expression which restricts to a complex expression in the x-axis, and so that the integral vanishes in some contour containing part of the x-axis. In the example you linked, you want to evaluate lnx/(x^4+1), so you consider this as the restriction to the real axis of Logz/(z^4+1).Do you have some expression like that in mind, or is your question more general? $\endgroup$ – DBFdalwayse Aug 3 '13 at 5:00
1
$\begingroup$

$\log{z}$ is viewed as a branch point due to its multivaluedness. That is, $\log{z}$ is only determined to within an integer multiple of $i 2 \pi$. $\log{z}$ is unique within a single branch; that is, as long as a contour along which $\log{z}$ is uniquely defined does not cross a branch cut that has been defined. That $\log{z}$ blows up as $z \to 0$ is beside the point; you simply do not include branch points such as those for which the argument of the log function is zero because of the nonuniqueness of the log near there.

$\endgroup$
2
  • $\begingroup$ So even though $|\log z|\to\infty$ as $z\to0$, $z=0$ is not considered as a pole, but part of the branch cut. Am I right? So is $z=0$ a special kind of singularity? (and is it called "branch cut" singularity"?) $\endgroup$ – user70520 Aug 3 '13 at 19:56
  • $\begingroup$ Yes, it is a branch point singularity, similar in nature to, say $z^{\alpha}$ for $\alpha$ noninteger. $\endgroup$ – Ron Gordon Aug 3 '13 at 20:06

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.