separating functional

A separating functional is a linear functional that tells two things apart by assigning them different (or oppositely-signed) values. If x and y are distinct, some f has f(x) != f(y); if two convex sets do not overlap, some f and constant c put one set where f <= c and the other where f >= c. The functional draws a hyperplane between them.

The basic existence fact in finite dimensions is elementary: functionals separate points. For any nonzero vector v there is an f with f(v) != 0 — just extend v to a basis and take the matching dual functional. This is exactly the injectivity of the evaluation map, restated as 'no vector is invisible to every measurement'.

The convex version is the separating hyperplane theorem: two disjoint convex sets (with mild conditions, e.g. one compact and one closed) can be strictly separated by a functional. A supporting functional is the boundary case — a hyperplane that touches a convex body and keeps the whole body on one side, maximized exactly on the contact set. These are the geometric backbone of convex optimization, duality of linear programs, and Lagrange multipliers.

All of this is the finite-dimensional shadow of the Hahn-Banach theorem, which in infinite dimensions guarantees that functionals still separate points and sets and that bounded functionals extend from subspaces to the whole space. In finite dimensions you get these guarantees by hand; Hahn-Banach is the deep theorem that keeps them true when dimension counting no longer suffices.