6

I've been wrestling with this problem (available here ):

Evaluate the integral of $f(z) = \sqrt{z^2-1}$ around the circle $\{z: |z| = 2\}$, where the branch of the square root function is chosen so that $\sqrt{2^2-1} > 0$.

I came up with the answer $$\int\limits_{|z| = 2} \sqrt{z^2-1} dz = \pi i$$ using the following argument, which seems a little like voodoo to me.

You notice that there are two points, at $z \pm 1$, where you might have to pick a branch of $log()$ to define the square root. Then, I convinced myself that as long as you exclude from your domain the real interval $-1 \le x \le 1$, the function $f(z)$ can be continuously defined (maybe the two branches somehow cancel?), and so is analytic on its domain.

Therefore you can shrink the contour of integration continuously using the Cauchy Closed Curve Integral Theorem, so that you wind up reducing it to the real integral $$2\int_{-1}^1 \sqrt{x^2-1} dx = \pi i$$

I'm not really sure the answer is correct, and I'm even less certain about the way I got it. I had a hard time convincing myself that $\sqrt{z^2-1}$ was not multi-valued on the domain. To try to get at that I tried $\sqrt{z^2-1} = \sqrt{z+1} \sqrt{z-1}$ and reasoned that the product makes the discontinuity disappear, using $$e^{\frac{\pi}{2}i}\cdot e^{\frac{\pi}{2}i} = -1 = e^{-\frac{\pi}{2}i}\cdot e^{-\frac{\pi}{2}i} $$ for the different branches of $log()$ you can choose.

I'd appreciate any input that would help me get a a better way to look at this.

Thanks!


EDIT

An attempt to use the Residue at infinity:

By substituting $z = 1/w$ you get \begin{eqnarray} \int\limits_{|z| = 2} \sqrt{z^2-1} dz &=& \int\limits_{|w| = 1/2} \frac{1}{w^2}\sqrt{\frac{1}{w^2}-1} dw \\ &=& \int\limits_{|w| = 1/2} \frac{1}{w^2}\sqrt{\frac{1-w^2}{w^2}} dw \\ &=& \int\limits_{|w| = 1/2} \frac{1}{w^3}\sqrt{1-w^2} dw \end{eqnarray} The integrand is analytic in $|w| \le 1/2$ except at $w = 0$, so we can apply the Residue Theorem. Near $w = 0$, $\sqrt{1-w^2} = 1 - \frac{1}{2}w^2 + \dots$, so the residue of $\frac{1}{w^3}\sqrt{1-w^2} $ at $w = 0$ is $-\frac{1}{2}$. Then the integral is $$\int\limits_{|z| = 2} \sqrt{z^2-1} dz = 2\pi i \cdot (-1/2) = -\pi i$$

bryanj
  • 3,938

2 Answers2

5

You are correct that by making the branch cut along $[−1,1]$ you get a well-defined analytic function. Each square root picks up a factor of $−1$ when you go around its branch point, so the product picks up $(−1)(−1)=1$ and is thus well-defined.

The nicest way to compute the integral is to apply the Residue Theorem, computing the residue at infinity. Yes, although they don't tell you this in a beginning complex variables class, you're actually computing residues of the meromorphic $1$-form $f(z)\,dz$, so when you substitute $z=1/w$, this becomes $-f(1/w)\,dw/w^2$; but, as you noted, the orientation of the circle as the boundary of a neighborhood of $\infty$ reverses, so we get $-2\pi i \text{ res}_\infty(f(z)\,dz)=2\pi i\text{ res}_0(f(1/w)\,dw/w^2)$.

Conceptually, the Riemann surface of $\sqrt{z^2-1}$ is made by removing the interval $[-1,1]\subset\mathbb R\subset\mathbb C$ and attaching two sheets, as usual with the square root function. If we compactify, we get two points at infinity, one on each branch; the compact Riemann surface is analytically equivalent to $\mathbb CP^1 \cong S^2$.

Ted Shifrin
  • 115,160
  • Ted - Thank you very much! That's a great trick! Soooo much better. I edited the question to try my hand at it - it seemed to me like using the Residue at infinity is sort of change of variables, so hopefully I executed it right. – bryanj Jun 13 '13 at 13:24
  • I am not actually taking a course, but rather have been working through the Bak/Newman book. I don't see "forms" or "residue at infinity" in there. Do you have another book to suggest? – bryanj Jun 13 '13 at 13:40
  • So, Bryan, what is your math background? And your long-range goals? – Ted Shifrin Jun 13 '13 at 13:41
  • I'm sort of embarrassed to admit that I am computer programmer who has recently sort of (re)discovered math. I wound up working on a project which needed some queuing theory which I didn't really get, so I decided to learn some math. The goal was to learn a few applied probability topics, but I got side-tracked because the math is more fun! In the last year I've worked through (meaning have done most of the exercises ) Spivak's Calculus, Spivak's Calculus on Manifolds, some of Baby Rudin, Curtis' Linear Algebra, Bak/Newman's book. Stein & Shakarchi's "Real Analysis" is next. – bryanj Jun 13 '13 at 13:52
  • Absolutely no need to be embarrassed :) If you did Spivak's Calculus on Manifolds, you certainly saw differential forms. If you want a bit more accessible treatment of multivariable calculus (with differential forms), still at the level of Spivak's Calculus, I'll toot my own horn and recommend my own Multivariable Mathematics: Multivariable Calculus, Linear Algebra, and Manifolds :) For a more mature treatment of complex, you might look at Henri Cartan's little book or Gamelin's. For stuff on differential forms, Cartan also has a little book, as does Flanders (with physics apps). – Ted Shifrin Jun 13 '13 at 14:49
  • OK great. I'll check out Cartan's stuff. Would you recommend Ahlfors' book? I've heard good things about it. I will also check out your book. – bryanj Jun 13 '13 at 15:08
  • I have taught out of Ahlfors several times, but it's a bit old-fashioned, although carefully written. Serge Lang has a good book on complex variables, as well. – Ted Shifrin Jun 13 '13 at 15:10
  • @TedShifrin Hi Ted. Just curious … How would one evaluate the integral $\oint_{C}\log(z+\sqrt{z^2-1}),dz$ where $C$ is any rectifiable curve that encloses (one time) the branch points at $z=-1$ and $z=1$? – Mark Viola Apr 23 '20 at 16:27
2

An idea:You could also try the following

$$\int\limits_\gamma \sqrt{z^2-1}\;dz=i\int\limits_\gamma \sqrt{1-z^2}\;dz$$

Now substitution

$$z=\sin t\;,\;u'=\cos t\;,\;\;0\le t\le 2\pi\;,\;\;\text{so that we "shrink" the circle as you did ,}$$

$$i\int\limits_0^{2\pi}\cos^2t\,dt=\left.i\frac{t+\cos t\sin t}{2}\right|_0^{2\pi}=\pi i$$

DonAntonio
  • 211,718
  • 17
  • 136
  • 287
  • 1
    Thanks DonAntonio. That helps clarify my first attempt. – bryanj Jun 13 '13 at 14:02
  • This isn't giving the right answer, is it? You can't deform to a path that isn't in the domain of the function. And if you try to make the other obvious branch cut, $\gamma$ is not contained in the resulting surface. – Ted Shifrin Jun 13 '13 at 14:46
  • No, this doesn't make sense. You're trying to do the integral precisely where the branch cut is. – Ted Shifrin Jun 13 '13 at 14:50
  • Well, this is trying to mimic what the OP did by "shrinking" the circle path (read the OP). Whether it is correct or not I can't tell off the top of my head now. – DonAntonio Jun 13 '13 at 14:50
  • Whether or not it's correct I don't know. I figured you could deform the circle into a little rectangle with corners at $\pm 1 \pm \epsilon i$. When I started looking at the parametrization of the contour integral it looked like it converged to the real integral on $[-1, 1]$. – bryanj Jun 13 '13 at 14:57
  • Thanks you both for all your help. One more thing: When trying to take the limit of the contour integral as $\epsilon \to 0$, I thought that maybe I could break the integral into an upper half and a lower half. Then, because one branch is continuous on the closed upper half-disk and the other is continuous on the lower closed half disk, you can use uniform continuity to interchange the limit and the integral. Although the branches converge to different values on $[-1,1]$, it seems like they differ by a factor of -1, which cancels the change in direction on the real axis. – bryanj Jun 13 '13 at 15:34