I read the definition of differentiability as given in Tom M. Apostol's Mathematical Analysis as : The function $f$ is said to be differentiable at $c$ if there exists a linear function $T_c : \mathbb R^n \to \mathbb R^m$ such that $f(c + v) = f(c) + T_c (v) + ||v|| E_c (v)$, where $E_c (v) \to 0$ as $v \to 0$. There $T_c()$ is called a linear function and called the total derivative of $f$ at $c$, $c$ is a point in $\mathbb R^n$. I can relate to the "error term" $E_c$ as I am familiar with Taylor's formula and this is a first order approximation of $f$. But I can't make a connection with functions in one variable. Is there a total derivative equivalent in one dimension? And also, everywhere on the internet this definition seems to be obscure and I only find total derivative defined as $df/dt$, like here. What connection am I missing?

  • 1
    $\begingroup$ A $1 \times 1$ matrix is identified with a single real entry. $\endgroup$ – Randall Sep 13 '17 at 19:14
  • $\begingroup$ Tom has failed. Cash out and limit our losses before something worse happens. $\endgroup$ – mathreadler Sep 13 '17 at 19:15
  • $\begingroup$ @Randall can you elaborate please, I don't completely understand $\endgroup$ – john doe Sep 13 '17 at 19:15
  • $\begingroup$ Given $f: \mathbb{R}^n \rightarrow \mathbb{R}^m$ and a point $\mathbf{v}$ in the domain, $df_\mathbf{v}$ is the linear map you mention, which is represented by an $m \times n$ matrix. That's the total derivative. In single-variable calculus you say $f'(2)=4$ instead of discussing the matrix $df_2 = (4)$. $\endgroup$ – Randall Sep 13 '17 at 19:17
  • $\begingroup$ How is $df_\mathbf{v}$ linear? And how is $f'(x)$ linear? Eg. when $f'(x)$ = $x^2$, $f'(cx)$ = $c^2f'(x)$? $\endgroup$ – john doe Sep 13 '17 at 19:24

Take $f: \mathbb{R}^n \rightarrow \mathbb{R}^m$ (domain can also be an open subset of $\mathbb{R}^n$). We say $f$ is differentiable at $c$ if there exists a linear transformation $T_\mathbf{c}: \mathbb{R}^n \rightarrow \mathbb{R}^m$ such that $$ \lim_{\mathbf{h} \to 0} \ \frac{|f(\mathbf{c}+\mathbf{h})-f(\mathbf{c})-T_\mathbf{c}(\mathbf{h})|}{|\mathbf{h}|} = 0. $$ This is equivalent to what you wrote.

At any rate, we write $df_\mathbf{c} = T_\mathbf{c}$ when it exists. By definition, $df_\mathbf{c}$ is a linear transformation (input variable $\mathbf{h}$) and so is representable by an $m \times n$ matrix $A$. In the general case this is the matrix of partials of the component functions.

In the 1-variable case, this is a $1 \times 1$ matrix, which is just a scalar and the ordinary derivative you're used to.

Example: let $f(x)=x^2$ and let's examine $f'(3)$ (which will be 6, trust me). The linear transformation $T: \mathbb{R} \to \mathbb{R}$ via $T(h)=6h$ satisfies my condition: $$ \lim_{h \to 0} \ \frac{f(3+h)-f(3)-6h}{h} = 0. $$ (I don't need norms anymore since my $h$ is now a scalar, not a vector: single variable calculus!)

Hence $T(h)=6h$ is the magic transformation, and as $T(1)=6$ we write the derivative as $f'(3)=6$, which is what you've always known from Calc I.

  • $\begingroup$ Oh yes, it all makes sense now! $\endgroup$ – john doe Sep 13 '17 at 19:49
  • $\begingroup$ Wow, really? Great! It took a looooong time for this to hit me when I first encountered it. $\endgroup$ – Randall Sep 13 '17 at 19:49
  • $\begingroup$ It took me quite long trying to understand it by myself before I posted the question here. $\endgroup$ – john doe Sep 13 '17 at 19:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.