Residues, and the theorem that sums them
The [[residue|residue]] of f at a pole a, written Res(f, a), is the coefficient c_{−1} in the Laurent series around a. The [[residue-theorem|residue theorem]] then says: for f holomorphic inside a positively oriented simple closed contour γ except at finitely many isolated singularities a_k inside, ∫_γ f(z) dz = 2πi · Σ_k Res(f, a_k). It generalizes both Cauchy's theorem (no singularities ⇒ integral 0) and the integral formula (one simple pole). All the local 2πi “seeds” from a single 1/(z − a) loop add up.
Evaluating a real integral by going complex
Here is the payoff. The real integral ∫_{−∞}^{∞} 1/(1 + x²) dx is doable by arctan, but the same method handles integrals that have no elementary antiderivative. The trick: close the real line with a big semicircle in the upper half-plane, apply the residue theorem, and show the semicircle's contribution vanishes by the ML estimate.
Evaluate I = integral_{-inf}^{inf} 1/(1 + x^2) dx by residues.
Let f(z) = 1/(1 + z^2) = 1/((z - i)(z + i)). Poles at z = i and z = -i (simple).
Contour: segment [-R, R] on the real axis + upper semicircle C_R of radius R.
Only the pole z = i lies inside (upper half-plane), so by the residue theorem:
integral_{-R}^{R} f dx + integral_{C_R} f dz = 2pi i * Res(f, i).
Residue at the simple pole z = i:
Res(f, i) = lim_{z->i} (z - i) f(z) = lim_{z->i} 1/(z + i) = 1/(2i).
So the right side = 2pi i * 1/(2i) = pi.
Semicircle vanishes: on C_R, |1 + z^2| >= R^2 - 1, so |f| <= 1/(R^2 - 1) = M,
length L = pi R, hence | integral_{C_R} | <= pi R / (R^2 - 1) -> 0 as R -> inf.
Let R -> infinity: integral_{-inf}^{inf} 1/(1+x^2) dx = pi.
(Cross-check: [arctan x] from -inf to inf = pi/2 - (-pi/2) = pi. Agrees.)Conformal maps: holomorphy as geometry
Finally, geometry. A [[conformal-map|conformal map]] preserves angles between curves. The clean theorem: if f is holomorphic and f'(a) ≠ 0, then near a, f is conformal — it rotates and scales infinitesimally, by the argument and modulus of f'(a). This is the geometric face of the Cauchy–Riemann equations: the local linear map has the special form of a rotation-scaling, never a shear or reflection, which is exactly what CR encodes. Conformal maps let you transplant a hard problem on an ugly region to an easy one on a disk.