Differentiability criterion via remainder estimate

Differentiability at a point is equivalent to a linear approximation with an o(||h||) error
Differentiability criterion via remainder estimate

Let URnU\subseteq\mathbb{R}^n be , let f:URmf:U\to\mathbb{R}^m, and fix aUa\in U.

Proposition (remainder estimate form of differentiability): The following are equivalent:

  • ff is at aa.
  • There exists a A:RnRmA:\mathbb{R}^n\to\mathbb{R}^m such that limh0f(a+h)f(a)Ahh=0. \lim_{h\to 0}\frac{\|f(a+h)-f(a)-Ah\|}{\|h\|}=0.
  • Equivalently: there exists a linear map AA such that for every ε>0\varepsilon>0 there exists δ>0\delta>0 with 0<h<δ    f(a+h)f(a)Ahεh. 0<\|h\|<\delta \implies \|f(a+h)-f(a)-Ah\|\le \varepsilon\,\|h\|.

In this formulation, AhAh is the best linear approximation to ff near aa and the remainder is “small compared to h\|h\|.”

Proof sketch: The second statement is the definition of differentiability. The third is exactly the ε\varepsilonδ\delta rewriting of the in the second statement: a limit equals 00 iff it can be made <ε<\varepsilon for all sufficiently small hh.