residue theorem
Suppose your contour loops around several points where the function blows up. Each singularity leaves behind one number — its residue — that records how much it contributes to a loop integral, and nothing else about the function matters. The residue theorem says the whole contour integral is just the sum of these residues, scaled by 2 pi i. It collapses an integral into a bit of bookkeeping: find the trapped singularities, total their residues, multiply.
Stated precisely: if f is analytic on and inside a positively oriented closed contour C except for isolated singularities inside, then the integral of f(z) dz around C equals 2 pi i times the sum of the residues of f at those interior singularities. The residue at a point is the coefficient c_{-1} of 1/(z - a) in the Laurent series there — the one term whose loop integral survives. For a simple pole there is a shortcut: the residue equals the limit as z approaches a of (z - a) f(z). For a pole of order m, you differentiate (z - a)^m f(z) a total of m - 1 times and take a limit. The theorem unifies Cauchy's theorem (no singularities, so zero) and Cauchy's formula (one simple pole) under one roof.
This is the engine room of applied complex analysis. It evaluates definite real integrals that have no elementary antiderivative, sums infinite series by turning them into residue sums of cot(pi z) or csc(pi z), inverts Laplace and Fourier transforms, and computes the asymptotics of integrals. The reason it feels like magic is the economy: an entire integral over a curve is decided by a handful of local numbers, one per singularity, each computable from a short Laurent expansion.
Integrate f(z) = 1/(z^2 + 1) around a large circle enclosing both poles z = i and z = -i. At z = i the residue is the limit of (z - i)/((z - i)(z + i)) = 1/(z + i) = 1/(2i); at z = -i it is 1/(-2i). The two residues sum to zero, so the loop integral is 2 pi i times 0 = 0 — the contributions cancel exactly.
Sum the residues of the enclosed poles and multiply by 2 pi i — the whole integral in three small steps.
Only singularities strictly inside the contour count, and only with the correct orientation (counterclockwise positive). A simple-pole shortcut applied to a higher-order pole gives a wrong residue, and the theorem says nothing about a singularity sitting exactly on the contour — that case needs a principal-value or indented-contour treatment.