5

Fix $w=re^{i\theta}\neq 0$ and let $\gamma$ be a rectifiable path in $\mathbb{C}\setminus\{0\}$ from $1$ to $w$. Show that there is a $k\in\mathbb{Z}$ such that $\displaystyle\int_{\gamma}\dfrac 1z=\log r+i\theta+2\pi i k$.

If we could find some primitive $F$ of $\dfrac 1z$, we would have $\displaystyle\int_{\gamma}\dfrac 1z=F(w)-F(1)$. If we could take $F(z)=\log z$, the principal branch, then $F(w)=\log r + i\theta$ and $F(1)=0$. But where does that $k\in\mathbb{Z}$ come from? It seems we can't take the principal branch so lightly.

How can we find a good primitive?

Tanius
  • 1,195
  • 1
    the $2 \pi k i $ part is related to number of times you encircle the origin in clockwise (-1) or counterclockwise (+) direction. – tired Nov 04 '15 at 16:07
  • 1
    Don't forget the $\mathrm{d}z$. Without the $\mathrm{d}z$ your integrals will not transform correctly under changes of coordinate. (When you integrate a "function", you're actually integrating something called a differential one-form, and without a $\mathrm{d}z$, you don't have a one-form.) – Fly by Night Nov 04 '15 at 16:13
  • Thank you very much, I see it now :) – Tanius Nov 04 '15 at 20:19

1 Answers1

13

Since the contour $\gamma$ is rectifiable, we can divide it into a finite number of segments. We divide, therefore, $\gamma$ into $N$ segments such we can parameterize each segment with a single-valued, differentiable function.

On the $n$'th segment, $1\le n \le N$, we write $z$ as the single-valued differentiable function of the parameter $t$, $z_n(t)$, where $z_n(t)$ is given in polar form by

$$z=\rho_n(t)e^{i\phi_n(t)}$$

for $t_{n-1}\le t \le t_n$, with $0\le \phi_n(t)<2\pi$.

On segment $n$, we have $dz=(\rho_n'(t)+i\rho_n(t)\phi'_n(t))e^{i\phi_n(t)}\,dt$ with $t$ starting at $t_{n-1}$ and ending at $t_{n}$. Then, we can write

$$\begin{align} \int_{\gamma}\dfrac{1}{z}\,dz&=\sum_{n=1}^{N}\int_{t_{n-1}}^{t_n} \dfrac{\rho'_n(t)+i\rho_n(t)\phi'_n(t)}{\rho_n(t)e^{i\phi_n(t)}}e^{i\phi_n(t)}\,dt\\\\ &=\sum_{n=1}^{N}\int_{t_{n-1}}^{t_n}\left(\dfrac{\rho'_n(t)}{\rho_n(t)}+i\phi'_n(t)\right)\,dt\\\\ &=\log(\rho(t_N)/\rho(t_0))+i\sum_{n=1}^{N}\left(\phi_n(t_{n})-\phi_n(t_{n-1})\right)\\\\ &=\log r + i \left(\phi_N(t_{N})-\phi_1(t_{0})\right)\\\\ &+i\sum_{i=1}^{N-1}\left(\phi_{n}(t_{n})-\phi_{n+1}(t_{n})\right) \tag 1 \end{align}$$

First, we note that $\phi_N(t_{N})=\theta$ and $\phi_1(t_{0})=0$ so that

$$ \phi_N(t_{N})-\phi_1(t_{0})=\theta $$

Next we note that, unless segments $n$ and $n+1$ together cross the real axis, we have

$$\phi_{n}(t_{n})-\phi_{n+1}(t_{n})=0 $$

Now, if the segments do constitute a crossing of the positive real axis, then either $\phi_{n}(t_{n})=2\pi$ and $\phi_{n+1}(t_{n})=0$ or $\phi_{n}(t_{n})=0$ and $\phi_{n+1}(t_{n})=2\pi$.

Therefore, the sum of the right-hand side of $(1)$ is equal to $2\pi k$ where $k$ is the net number of times $\gamma$ crosses the positive real axis from the fourth quadrant to first quadrant.

Putting everything together, we obtain

$$\bbox[5px,border:2px solid #C0A000]{\int_{\gamma}\dfrac{1}{z}\,dz=\log r+i(\theta+2k\pi)}$$

as was to be shown!

Mark Viola
  • 179,405
  • 1
    i don't understand the voting behaviour on this site. this is a clear (+1) and the answer should have many more! – tired Nov 06 '15 at 17:46
  • @tired I too don't understand the surplus of "trigger-happy" down voters and the deficit of up voters. And thank you for the up vote!! Very much appreciative. - Mark – Mark Viola Nov 06 '15 at 17:52