Helmholtz decomposition
/ HELM-holts /
Every vector field, however tangled, can be split into two clean pieces: one that only spreads (has divergence but no curl) and one that only swirls (has curl but no divergence). Think of separating a wind map into a part that flows outward from sources and into sinks, plus a part that purely circulates in eddies. The Helmholtz decomposition is the theorem that this split always exists and pins down exactly what each piece is.
Precisely, under suitable decay conditions, a vector field F can be written F = -grad phi + curl A, the sum of an irrotational (curl-free) part -grad phi and a solenoidal (divergence-free) part curl A. The first piece is built from a scalar potential phi and carries all of the divergence of F (since curl of a gradient is zero); the second piece is built from a vector potential A and carries all of the curl (since divergence of a curl is zero). In effect, the divergence of F and the curl of F together — the sources and the swirls — determine F almost completely, with only boundary or far-field data left to fix.
This decomposition is why divergence and curl are the right two quantities to study: they are exactly the data that reconstruct a field. In fluid dynamics it splits a flow into a potential (irrotational) flow plus a vorticity-carrying part, central to aerodynamics. In electromagnetism it underlies writing the magnetic field as B = curl A and the electrostatic field as E = -grad phi, and it is the structural reason Maxwell's equations — which prescribe the divergence and curl of E and B — suffice to determine the fields. The two great theorems (divergence and Stokes') govern the two halves separately.
The radial field (x, y, z) is purely irrotational: its curl is zero, so its entire content lives in the -grad phi piece (with phi = -(x^2 + y^2 + z^2)/2). The swirl field (-y, x, 0) is purely solenoidal: its divergence is zero, so it lives entirely in the curl A piece. Most real fields are a genuine mixture of the two.
Two extreme fields, one all-spread and one all-swirl; a general field blends both pieces.
The decomposition is unique only once you fix boundary or decay conditions: on unbounded space you need the field to fall off fast enough at infinity, and on a bounded region you can always add a harmonic field (one that is both curl-free and divergence-free) without changing div F or curl F, so the split is not unique without extra conditions.