JOVANA
Explore Library Glossary Getting Started Three Levels Fields How it works Mission
Join the mission
All guides

Cauchy's Theorem and Goursat's Idea

The theorem that makes complex analysis miraculous: a holomorphic function integrated around any closed loop in a simply connected region gives exactly zero. Goursat's clever triangle argument earns that conclusion with almost no extra assumptions — and quietly powers everything that follows.

The claim, and why it should astonish you

In the last guide you learned that an integral becomes painless the moment the integrand has a primitive: the fundamental theorem for contour integrals says the value depends only on the endpoints, so every closed loop integrates to zero. That was a conditional gift — it asked you to first find an antiderivative F with F' = f. Cauchy's integral theorem removes the condition. It declares: if f is holomorphic throughout a nice region, then the integral of f around any closed contour in that region is zero — no antiderivative required, no formula to find first.

Pause on how strange this is. In ordinary real calculus, the line integral of a vector field around a loop is usually NOT zero — it measures circulation, and a swirling field has plenty of it. The integral vanishes only for special, conservative fields. Cauchy's theorem says that for a holomorphic f, the field built from f is automatically conservative — always, everywhere it is holomorphic. Holomorphy is not a mild smoothness assumption; recall it forces the Cauchy-Riemann equations, and those equations are exactly the rigidity that makes the loop integral collapse.

Simply connected: regions with no holes

The note above tells you which regions are safe. The clean hypothesis is that the domain be simply connected. A simply connected domain is, intuitively, a region with no holes: any closed loop inside it can be shrunk continuously to a point without ever leaving the region or snagging on a missing spot. A disk is simply connected. The whole plane is simply connected. A half-plane and a square are too.

By contrast, a multiply connected domain has at least one hole. The punctured plane — the plane with the origin removed — is the standard example: a loop encircling the origin cannot be shrunk to a point without crossing the missing center. This is exactly the situation of 1/z, whose only singular point is the puncture. So Cauchy's theorem and the geometry of the region are two sides of one coin: the theorem holds precisely when the loop has no forbidden point trapped in its interior.

Goursat's idea: the triangle that earns everything

Cauchy's original proof leaned on Green's theorem, which secretly assumes the derivative f' is continuous. That feels like cheating: we do not yet know a holomorphic function has a continuous derivative — that is a conclusion we want to earn, not a hypothesis to borrow. Goursat's triumph was to prove the theorem for a triangle using ONLY that f is complex-differentiable, with no continuity of f' assumed. This is the Cauchy-Goursat theorem, and the engine inside it is the triangle argument.

  1. Call the integral of f around a triangle T its value I, and suppose for contradiction that |I| is not zero. We will trap I and show it must vanish after all.
  2. Bisect the three sides to cut T into four congruent sub-triangles. The integrals over the shared inner edges cancel in pairs, so the sum of the four sub-integrals equals I exactly.
  3. At least one sub-triangle, call it T_1, carries an integral of size at least |I|/4. Keep that one and discard the rest.
  4. Repeat forever. You get nested triangles T_1, T_2, T_3, ... whose integrals are at least |I|/4, |I|/16, |I|/64, ... while their perimeters and diameters halve at each step.
  5. The nested triangles shrink to a single point z_0. There f is differentiable, so f(z) = f(z_0) + f'(z_0)(z - z_0) + (error), where the error is far smaller than |z - z_0| near z_0.
  6. The linear part f(z_0) + f'(z_0)(z - z_0) is a polynomial — it HAS a primitive — so its loop integral is zero. Only the tiny error term survives.
  7. Estimate the leftover with the ML inequality from earlier in this rung. The result is that |I| must be smaller than any positive number — so |I| = 0. The contradiction is resolved, and the integral was zero all along.

The beauty is what does the heavy lifting: only the LINEAR approximation of f matters, because a linear function trivially has a primitive and so kills its own loop integral. Everything nonlinear is pushed into an error so small that the ML estimate crushes it to nothing. Differentiability at a single point — the one fact we are allowed to assume — is exactly enough, repeated across a shrinking net of triangles, to force the global integral to zero.

From a triangle to any loop

A triangle seems like a narrow prize, but it is the whole game. From the triangle case you bootstrap upward in three honest moves. First, glue triangles together: any polygon splits into triangles, so the integral around any closed polygon is zero. Second, a polygon with arbitrarily many short sides approximates any smooth contour as closely as you like, and the integrals converge — so the Cauchy integral theorem holds for any closed contour in a convex region.

Third, you globalize from convex pieces to any simply connected region. Because closed-loop integrals vanish locally, the integral from a fixed base point becomes well-defined independent of the path — which is precisely path independence. That path-independent integral IS a primitive F of f, so we have recovered, as a CONCLUSION, the antiderivative that the last guide had to assume. The circle closes beautifully: holomorphy gives Cauchy-Goursat, Cauchy-Goursat gives path independence, and path independence builds the primitive.

F(z) = integral from z_base to z of f(w) dw     (any path inside the region)

Cauchy-Goursat  ==>  this is independent of the path
                ==>  F is well-defined
                ==>  F'(z) = f(z),  so F is a primitive of f
The path-independent integral manufactures the very primitive the previous guide had to assume.

Holes, deformation, and what comes next

What if the region is NOT simply connected — what if a singularity sits inside the loop after all? Then the loop integral need not be zero, but Cauchy's theorem still controls the situation through deformation of contours. As long as you move the contour through a region where f stays holomorphic, the integral does not change. So you may slide and bend a big ugly loop into a tiny circle hugging the singular point, and the two give the same value. The integral becomes a measurement of what is trapped inside, not of the curve's particular shape.

This is the doorway to the rest of the subject, and the next guide walks through it. Deformation invariance is really a statement about homotopy — loops that can be deformed into each other carry equal integrals — and counting HOW MANY times a loop wraps a point is the job of the winding number. Together they upgrade Cauchy's theorem from "zero in nice regions" to a precise bookkeeping of holes that, a couple of guides from now, blossoms into the residue calculus that evaluates real integrals no elementary antiderivative can touch.