Spring 2011 - Problem 7

Goursat's theorem

Prove Goursat's theorem: If f ⁣:CC\func{f}{\C}{\C} is complex differentiable (and so continuous), then for every triangle C\triangle \subseteq \C

f(z)dz=0\oint_{\partial\triangle} f\p{z} \,\diff{z} = 0

where the line integral is over the three sides of the triangle.

Solution.

Fix a triangle \triangle with vertices {z1,z2,z3}\set{z_1, z_2, z_3} oriented counter-clockwise. Suppose otherwise, so that there exists ε>0\epsilon > 0 such that

f(z)dzε>0.\abs{\oint_{\partial\triangle} f\p{z} \,\diff{z}} \geq \epsilon > 0.

Decompose 1\triangle_1 \coloneqq \triangle into four further triangles: 1(1),1(2),1(3),1(4)\triangle_1^{\p{1}}, \triangle_1^{\p{2}}, \triangle_1^{\p{3}}, \triangle_1^{\p{4}} so that

1(1)has vertices{z1,z1+z22,z1+z32}1(2)has vertices{z1+z22,z2+z32,z1+z3z2}1(3)has vertices{z1+z22,z2,z2+z32}1(4)has vertices{z1+z32,z2+z32,z3},\begin{aligned} \triangle_1^{\p{1}} & \quad\text{has vertices}\quad \set{z_1, \frac{z_1 + z_2}{2}, \frac{z_1 + z_3}{2}} \\ \triangle_1^{\p{2}} & \quad\text{has vertices}\quad \set{\frac{z_1 + z_2}{2}, \frac{z_2 + z_3}{2}, \frac{z_1 + z_3}{z_2}} \\ \triangle_1^{\p{3}} & \quad\text{has vertices}\quad \set{\frac{z_1 + z_2}{2}, z_2, \frac{z_2 + z_3}{2}} \\ \triangle_1^{\p{4}} & \quad\text{has vertices}\quad \set{\frac{z_1 + z_3}{2}, \frac{z_2 + z_3}{2}, z_3}, \end{aligned}

which have disjoint interiors and oriented counter-clockwise. By cancellation,

ε1f(z)dz=j=141(j)f(z)dzj=141(j)f(z)dz,\begin{aligned} \epsilon &\leq \abs{\int_{\partial\triangle_1} f\p{z} \,\diff{z}} \\ &= \abs{\sum_{j=1}^4 \int_{\partial\triangle_1^{\p{j}}} f\p{z} \,\diff{z}} \\ &\leq \sum_{j=1}^4 \abs{\int_{\partial\triangle_1^{\p{j}}} f\p{z} \,\diff{z}}, \end{aligned}

by the triangle inequality. Thus, there exists j0j_0 so that 1(j0)f(z)dzε4\abs{\int_{\partial\triangle_1^{\p{j_0}}} f\p{z} \,\diff{z}} \geq \frac{\epsilon}{4} or else the triangle inequality implies

εj=141(j)f(z)dz<j=14ε4=ε,\epsilon \leq \sum_{j=1}^4 \abs{\int_{\partial\triangle_1^{\p{j}}} f\p{z} \,\diff{z}} < \sum_{j=1}^4 \frac{\epsilon}{4} = \epsilon,

which is impossible. Set 2=1(j0)\triangle_2 = \triangle_1^{\p{j_0}}. By induction (e.g., by replacing \triangle above with n\triangle_n), we get a sequence of triangles {n}n\set{\triangle_n}_n such that nn+1\triangle_n \supseteq \triangle_{n+1} and nf(z)dzε4n1\abs{\int_{\partial\triangle_n} f\p{z} \,\diff{z}} \geq \frac{\epsilon}{4^{n-1}}. Notice also that diamn=diam12n10\diam{\triangle_n} = \frac{\diam{\triangle_1}}{2^{n-1}} \to 0 as nn \to \infty, so by completeness of C\C,

n=11={z}.\bigcap_{n=1}^\infty \triangle_1 = \set{z^*}.

Since ff is complex differentiable, f(z)f'\p{z^*} exists and if we set h(z)=f(z)f(z)zzf(z)h\p{z} = \frac{f\p{z} - f\p{z^*}}{z - z^*} - f'\p{z^*}, then h(z)0h\p{z} \to 0 as zzz \to z^* and

f(z)=f(z)+f(z)(zz)+h(z)(zz).f\p{z} = f\p{z^*} + f'\p{z^*}\p{z - z^*} + h\p{z}\p{z - z^*}.

Moreover, there exists δ>0\delta > 0 so that h(z)<η\abs{h\p{z}} < \eta, so because diamn0\diam\triangle_n \to 0, there exists n1n \geq 1 such that nB(z,δ)\triangle_n \subseteq B\p{z^*, \delta}. Observe that by the fundamental theorem of calculus, γaz+bdz=0\int_\gamma az + b \,\diff{z} = 0 for any closed curve γ\gamma. Hence,

ε4n1nf(z)dz=nf(z)+f(z)(zz)+h(z)(zz)dz=nh(z)(zz)dzηnsupznzzdzη(diamn)2η(diam1)24n1    εη(diam1)2.\begin{aligned} \frac{\epsilon}{4^{n-1}} &\leq \abs{\int_{\partial\triangle_n} f\p{z} \,\diff{z}} \\ &= \abs{\int_{\partial\triangle_n} f\p{z^*} + f'\p{z^*}\p{z - z^*} + h\p{z}\p{z - z^*} \,\diff{z}} \\ &= \abs{\int_{\partial\triangle_n} h\p{z}\p{z - z^*} \,\diff{z}} \\ &\leq \eta \int_{\partial\triangle_n} \sup_{z \in \partial\triangle_n} \abs{z - z^*} \,\diff\abs{z} \\ &\leq \eta \p{\diam\triangle_n}^2 \\ &\leq \frac{\eta\p{\diam{\triangle_1}}^2}{4^{n-1}} \\ \implies \epsilon &\leq \eta\p{\diam{\triangle_1}}^2. \end{aligned}

Letting η0\eta \to 0, however, we see that ε=0\epsilon = 0, which is impossible. Thus, ε\epsilon must have been 00 to begin with, which completes the proof.