A strong notion of differentiability for functions between normed vector spaces: at a point x the function f admits a bounded linear operator L (the Fréchet derivative) such that f(x+h)-f(x)-L(h)=o(‖h‖) as h→0, i.e. L uniformly approximates the first-order increment of f.

Principle

The organizing rule is existence of a single bounded linear map that provides a uniform first-order approximation: the remainder divided by the norm of the increment tends to zero. This combines linearity, boundedness, and a uniform smallness condition on the error.

Demonstration

Concrete example in finite dimensions: f: R^2→R defined by f(x,y)=x^2+y^2 is Fréchet differentiable at (x0,y0) with L(h1,h2)=2x0·h1+2y0·h2. One checks (f(x0+h)-f(x0)-L(h))/‖h‖→0 as h→0.

Misapplication

Assuming that existence of directional derivatives along every vector (Gâteaux derivatives) automatically gives a Fréchet derivative; in infinite-dimensional spaces this fails — directional derivatives can exist without a single bounded linear map approximating f uniformly.

Consequence

When present, the Fréchet derivative is unique, linear and bounded; it implies continuity of f and enables standard calculus rules (sum, product/composition where defined) and error estimates used in Newton-type methods and stability analyses.

Reversal

The opposite situation is a function that has directional derivatives in many or all directions but no bounded linear map approximating increments—there is no uniform linear first-order approximation, so Fréchet differentiability fails.

Boundary

Applies only in normed (or metric-compatible) vector spaces and is a pointwise property; excludes mere existence of directional derivatives without uniform control, and excludes approximation by nonlinear maps. It also presumes small-norm asymptotics and a linear target space for L.

Semantic Tension

Often contrasted with Gâteaux differentiability: both use directional information, but Fréchet requires a single bounded linear approximant and a stronger uniform error condition. In finite dimensions they coincide under mild hypotheses, but not in general.

Synthesis

Fréchet differentiability is the requirement that a function admit a unique bounded linear map that gives the best uniform linear approximation to increments at a point, yielding the familiar calculus structure in normed spaces.