Vector fields and the gradient

A scalar field assigns one number to each point of space — temperature, pressure, altitude. A vector field assigns a vector — wind velocity, the gravitational pull, the electric field. They are different kinds of object and the distinction matters immediately: you cannot add a temperature to a wind velocity, and the two are drawn differently, one as a shaded map of values, the other as arrows.

The gradient is the bridge between them: it takes a scalar field and produces a vector field. At each point, ∇f is built from the partial derivatives (∂f/∂x, ∂f/∂y, ∂f/∂z), and it points in the direction that f increases fastest, with magnitude equal to that steepest rate of increase. Walking in the gradient direction climbs a hill as quickly as possible; walking opposite to it descends fastest — that is the basis of gradient-descent methods used throughout applied mathematics.

The gradient's direction is always perpendicular to the level surfaces of f — the contours along which f is constant, like the isotherms on a weather map. This is not a coincidence to memorise but a direct consequence of the definition: moving along a level surface produces no change in f, so the direction of fastest change must be orthogonal to every direction that produces no change. Contour maps make this visible directly — the steepest path uphill always crosses the contour lines at right angles, never running along one.