Home | Assessment | Notes | Index | Worksheets | Blackboard
There are many versions of Cauchy's theorem in complex analysis. They all give conditions under which the contour integral of a holomorphic function over a closed contour will be zero. Next week we will cover a sophisticated version whose condition is in terms of winding numbers. Here, we cover the most basic version: Cauchy's theorem in a rectangle.
Fix a domain $D \subset \C$ and $f : D \to \C$ holomorphic. For any rectangle $R$ which, together with its interior, is entirely contained within $D$ we have \[\int\limits_\gamma f(z) \intd z = 0\] where $\gamma$ is the contour parameterizing the edges of $R$ in turn.
Before we prove this theorem, there are two comments to make about its hypotheses.
We will also need the following lemma.
Fix $\gamma$ a smooth path in a domain $D$ and $f : D \to \C$ continuous. Then \[\left| \int\limits_\gamma f \right| \le \ell(\gamma) M\] where $M$ is any upper bound for $\{ |f(\gamma(t))| : a \le t \le b \}$.
Recall that the contour integral is a limit of Riemann sums \[\int\limits_\gamma f = \lim_{N \to \infty} \frac{b-a}{N} \sum_{n=1}^N f \left( \gamma \left( a + n \frac{b-a}{N} \right) \right) \gamma'\left( a + n \frac{b-a}{N} \right)\] so \[ \begin{aligned} \left| \int\limits_\gamma f \right| &= \lim_{N \to \infty} \frac{b-a}{N} \left| \sum_{n=1}^N f \left( \gamma \left( a + n \frac{b-a}{N} \right) \right) \gamma'\left( a + n \frac{b-a}{N} \right) \right| \\ & \le \lim_{N \to \infty} \frac{b-a}{N} \sum_{n=1}^N \left| f \left( \gamma \left( a + n \frac{b-a}{N} \right) \right) \gamma'\left( a + n \frac{b-a}{N} \right) \right| \\ & \le M \lim_{N \to \infty} \frac{b-a}{N} \sum_{n=1}^N \left| \gamma' \left( a + n \frac{b-a}{N} \right) \right| \\ & = M \int\limits_a^b |\gamma'(t)| \intd t = M \ell(\gamma) \end{aligned}\] as claimed. $\square$
Suppose that the integral is not zero. Write $\rho$ for the perimiter of $R$. Subdivide $R$ into four equally sized rectangles $R(1),R(2),R(3),R(4)$. We have \[\int\limits_{R(1)} f(z) \intd z + \int\limits_{R(2)} f(z) \intd z + \int\limits_{R(3)} f(z) \intd z + \int\limits_{R(4)} f(z) \intd z = \int\limits_R f(z) \intd z \] so there is a rectangle $S(1) \in \{R(1), R(2), R(3), R(4) \}$ with \[ \left| \int\limits_{S(1)} f(z) \intd z \right| \ge \frac{1}{4} \left| \int\limits_R f(z) \intd z \right| \] as otherwise the triangle inequality would give a contradiction. Repeating this argument, now subdividing $S(1)$ into four equal sub-rectangles, one of the sub-rectangles $S(2)$ will satisfy \[ \left| \int\limits_{S(2)} f(z) \intd z \right| \ge \frac{1}{4} \left| \int\limits_{S(1)} f(z) \intd z \right| \ge \frac{1}{4^2} \left| \int\limits_R f(z) \intd z \right| \] via the same argument as before. Continuing by induction, we get a sequence $S(n)$ of rectangles with
There is a unique point $\zeta$ contained in all of the rectangles $S(n)$. Fix \[\epsilon = \frac{1}{2\rho^2} \left| \int\limits_\gamma f(z) \intd z \right|\] and choose $r > 0$ such that \[\left| f(\zeta + h) - \left( f(\zeta) + h f'(\zeta) \right) \right| < |h| \epsilon\] holds whenever $|h| < r$. Fix $n$ so large that $S(n) \subset \ball(\zeta,r)$. Thus for every $z \in S(n)$ we have \[ \left| f(z) - \Big( f(\zeta) + (z - \zeta) f'(\zeta) \Big) \right| \le |z - \zeta| \epsilon \] and can estimate that \[ \begin{align*} & \left| \int\limits_{S(n)} \!\! f(z) \intd z \right| \\ ={} & \bigg| \int\limits_{S(n)} f(z) - \left( f(\zeta) + (z - \zeta) f'(\zeta) \right) \intd z \\ & \qquad + \int\limits_{S(n)} f(\zeta) + (z - \zeta) f'(\zeta) \intd z \bigg| \\ \le{} & \epsilon \int\limits_{S(n)} |z-\zeta| \intd z \le \frac{\epsilon \rho^2}{4^n} \end{align*} \] using the estimation lemma and the fact that \[g(z) = f(\zeta) + (z - \zeta) f'(\zeta)\] has an anti-derivative on $D$. Relating this to the original rectangle $R$ gives \[ \frac{\epsilon \rho^2}{4^n} \ge \left| \int\limits_{S(n)} f(z) \intd z \right| \ge \frac{1}{4^n} \left| \int\limits_R f(z) \intd z \right| \] and allows us to conclude that \[ \epsilon \rho^2 \ge \left| \int\limits_R f(z) \intd z \right| \] contradicting our choice of $\epsilon$. The original integral must therefore be zero. $\square$