View source for Approximate derivative

{{MSC|28A33|49Q15}}

[[Category:Classical measure theory]]


{{TEX|done}}

A generalization of the concept of a derivative, where the ordinary  limit is replaced by an [[Approximate limit|approximate limit]].

====Definition====
Consider a (Lebesgure) measurable set $E\subset \mathbb R$, a  measurable map $f:E\to \mathbb R^k$ and a point $x_0\in E$ where $E$ has  [[Density of a set|Lebesgue density]] $1$. If the approximate limit
\[
{\rm ap}\, \lim_{x\to x_0} \frac{f(x)-f(x_0)}{x-x_0}
\]
exists at it is finite, then it is called approximate derivative of the function $f$ at $x_0$ and the function is called approximately differentiable at $x_0$ (when $k=1$ some authors include also the case in which the limit is $\pm\infty$). As a consequence of the definition, $f$ is [[Approximate continuity|approximately continuous]] at every $x_0$ where it is approximately differentiable. The concept can be extended further to functions of several variables: see [[Approximate differentiability]]. Some authors denote the approximate derivative by $f'_{ap} (x_0)$, whereas some authors keep the notation $f' (x_0)$. Indeed if the classical derivative exists, then it coincides with the approximate derivative, whereas the opposite is false (note, for instance, that if $g$ coincides almost everywhere with $f$, then $g$ is as well approximately differentiable at $x_0$).

====Properties====
The following useful proposition relates further the two concepts.

'''Proposition 1'''
Let $E$, $f$ and $x_0$ be as above. The approximate derivative at $x_0$ exists if and only if there is a measurable set $F\subset E$ which has density $1$ at $x_0$ and such that the classical derivative exists for $f|_F$ at $x_0$. Moreover, the approximate derivative of $f$ at $x_0$ equals the classical derivative of $f|_F$ at the same point.

As a corollary of the previous proposition, the classical rules for the differentiation of a sum, a difference, a  product, and a quotient of functions apply to (finite) approximate  derivatives as well. The theorem on the differentiation of a compositition of approximately differentiable functions  does not apply in general. However it applies to $\varphi\circ f$ if $f$ is approximately differentiable at $x_0$ and $\varphi$ is '''classically differentiable''' at $f(x_0)$.

====Approximate Dini derivatives and Denjoy-Khinchin theorem====
Approximate Dini derivatives are defined by analogy with ordinary Dini derivatives (cf. [[Dini derivative|Dini derivative]]): $D^+ f (x_0)$, $D^- f (x_0)$
and $d^+ f (x_0)$, $d^- f (x_0)$ are, respectively, the right and left approximate upper limits and the right and left approximate lower limits of the quotient
\[
\frac{f(x)-f(x_0)}{x-x_0}\, .
\]
The following theorem applies.

'''Theorem 2 (Denjoy-Khinchin)'''
If $E\subset \R$ and $f:E\to \R$ are (Lebesgue) measurable, at least one of the following three alternatives holds at almost every point $x_0\in E$: 
* either $f$ has a finite approximate derivative, 
* or $D^+ f (x_0)= D^- f (x_0)= \infty$,
* or $d^+ f (x_0)= d^- f (x_0)=-\infty$.

(Observe that the two last alternatives are not exclusive!)

====Relation with the integral====
As it happens for the ordinary derivative, however, the existence almost everywhere of the approximate derivative does not imply that the fundamental theorem
of calculus applies. More precisely, there are examples of functions $f$ which are classically differentiable almost everywhere on the interval $[0,1]$ but such that the identity
\[
f (b)-f(a) =\int_a^b f' (t)\, dt
\]
fails on a set of positive measure of extrema $(a,b)\in [0,1]^2$ (cp. with the example in [[Absolute continuity]]).

If, on the other hand, $F$ and $f$ are measurable functions on $[0,1]$ taking values in $\mathbb R^k$ and the identity
\begin{equation}\label{e:fundamental}
F(a)-F(0) =\int_0^a f (t)\, dt
\end{equation}
holds for a.e. $a\in [0,1]$, then the function $F$ can be redefined on a set of measure zero so that
* $F$ is absolutely continuous
* the identity \ref{e:fundamental} holds for ''every'' $a$
* the function $F$ is almost everywhere differentiable and $F'=f$ almost everywhere.

====Approximate partial derivatives====
If $E\subset \mathbb R^n$ and $f: E\to\mathbb R^k$ are measurable, approximate partial derivatives can be defined as follows. Consider a system of coordinates 
$x_1, \dots , x_n$, a point $p=(p_1, \ldots , p_n)\in E$ and a coordinate $i$. If
* $E':=\{ h\in\mathbb R: (p_1, \ldots, p_{i-1}, h, p_{i+1}, \ldots , p_n)\in E\}$ is measurable,
* the map $h\mapsto g(h):= f ((p_1, \ldots, p_{i-1}, h, p_{i+1}, \ldots , p_n)$ is measurable
and the approximate limit
\[
L := {\rm ap} \lim_{h\to p_i} \frac{g(h)- g (p_i)}{h-p_i}
\]
exists and it is finite, then $L$ is the ''approximate partial derivative of $f$ at $p$ in the $i$-th direction''.

====Comments====
There exist continuous functions without an ordinary or an approximate derivative at any point of a given interval.
The concept of approximate derivative was introduced by A.Ya. Khinchin in 1916.

====References====
{|
|-
|valign="top"|{{Ref|AFP}}||   L. Ambrosio, N.   Fusco, D.  Pallara, "Functions of bounded variations  and  free   discontinuity  problems". Oxford Mathematical Monographs. The    Clarendon  Press,  Oxford University Press, New York, 2000.      {{MR|1857292}}{{ZBL|0957.49001}} 
|-
|valign="top"|{{Ref|Br}}|| A.M. Bruckner,   "Differentiation of real functions" , Springer  (1978)
|-
|valign="top"|{{Ref|Fe}}||    H. Federer, "Geometric measure theory". Volume 153 of Die Grundlehren    der mathematischen Wissenschaften. Springer-Verlag New York Inc., New    York, 1969.
|-
|valign="top"|{{Ref|Mu}}||  M.E. Munroe,   "Introduction to measure and integration" , Addison-Wesley  (1953)
|-
|valign="top"|{{Ref|Ro}}|| H.L. Royden,   "Real analysis" , Macmillan  (1968)
|-
|valign="top"|{{Ref|Ru}}||  W. Rudin, "Principles of  mathematical analysis", Third edition,  McGraw-Hill (1976) {{MR|038502}}  {{ZBL|0346.2600}}  
|-
|valign="top"|{{Ref|Sa}}|| S. Saks,   "Theory of the integral" , Hafner  (1952)
|-
|valign="top"|{{Ref|Th}}|| B.S. Thomson,   "Real functions" , Springer  (1985)
|-
|}