the canonical divisor

Functions are not the only meromorphic objects on a surface; meromorphic DIFFERENTIALS — things like f(z) dz — also have zeros and poles. The remarkable fact is that although a surface has many such differentials, the DIVISOR of any one of them is determined up to harmless adjustment. That distinguished class of divisors, intrinsic to the surface itself, is the canonical divisor — the surface's built-in signature.

Precisely, take any nonzero meromorphic 1-form omega on a compact Riemann surface X. Locally omega = f(z) dz, and its divisor div(omega) = sum over p of ord_p(omega) [p] is well-defined: under a holomorphic coordinate change the dz transforms with a nonvanishing factor, so the orders don't move. Any two nonzero meromorphic 1-forms differ by a meromorphic function (the ratio is a function), so their divisors are LINEARLY EQUIVALENT. That single linear-equivalence class is the canonical divisor K (or canonical class K_X). Its degree is forced by topology: deg(K) = 2g - 2, where g is the genus. So on the sphere deg K = -2, on a torus deg K = 0 (a holomorphic 1-form with no zeros or poles, like dz), and for g >= 2 deg K = 2g - 2 > 0.

Why it matters: K is the invisible counterweight all over Riemann-Roch, whose symmetric form pairs L(D) against L(K - D). The dimension of the space of holomorphic 1-forms equals g, and these correspond to effective divisors in the canonical class. The canonical MAP, built from a basis of holomorphic 1-forms, embeds every non-hyperelliptic curve of genus >= 3 into projective space, realizing the abstract surface as an explicit algebraic curve. Honesty points: K is a CLASS, defined only up to linear equivalence — there is no single canonical divisor, only a canonical class, and naming a specific representative requires choosing a differential. Also the degree formula deg K = 2g - 2 is the precise link between the analytic object (a differential) and the topological invariant (the genus).