A weaker, directional notion of differentiability: at a point x the directional derivative along each vector v exists, i.e. the limit lim_{t→0} (f(x+tv)-f(x))/t exists for every v. It records first-order directional rates but does not a priori require a single bounded linear approximant.

Principle

The organizing idea is pointwise existence of directional limits in all directions; linearity or boundedness of the resulting mapping v↦D_f(x)(v) may be an extra property rather than automatic. Gâteaux differentiability captures directional sensitivity rather than uniform linear approximation.

Demonstration

Example: consider f:R→R, f(x)=|x| at x=0. The directional derivative along v exists for every v (right and left one-sided slopes), so directional rates exist, but there is no single linear map that approximates f at 0, showing failure of Fréchet while directional derivatives exist.

Misapplication

Treating existence of directional derivatives as sufficient for applying calculus rules that require Fréchet differentiability (e.g. using a Jacobian matrix and continuity estimates); in infinite-dimensional settings this can lead to incorrect conclusions about stability or linearization.

Consequence

When combined with linearity (the mapping v↦D_f(x)(v) is linear) and appropriate continuity, Gâteaux differentiability promotes to Fréchet differentiability. It is often the first step in functional analysis to test differentiability directionwise.

Reversal

The converse is a function that is Fréchet differentiable (hence directionally differentiable) but whose directional derivatives alone might hide the uniform linear approximation property; knowing all directional rates is weaker information than knowing a bounded linear derivative.

Boundary

Defined for maps on vector spaces (often normed); it requires limits along rays but does not require uniformity in v or boundedness of the directional derivative operator. It excludes claims about error control unless extra continuity is assumed.

Semantic Tension

Tension arises with Fréchet differentiability and with weaker pointwise notions: Gâteaux captures directional information and can exist without linear or continuous structure, whereas Fréchet demands a single linear bounded approximant; confusion between them is a common source of error.

Synthesis

Gâteaux differentiability asserts existence of directional first-order limits in every direction, giving directional rates of change; with linearity and continuity of the direction map it upgrades to the full Fréchet derivative.