Chain Rule for multivariable functions to one variable functions

63 Views Asked by At

I'm trying to understand some questions about the chain rule of multivariable and how it relates with one variable functions.

Ok, so, imagine that I have two functions $f: U \subset E \rightarrow \mathbb{R}$ and $\alpha: (-\epsilon, \epsilon) \subset \mathbb{R} \rightarrow E$, where $E$ is a normed vector space.

Then, by the chain rule, the derivative of $f \circ \alpha$ in some point $a \in U$ is:

$(f \circ \alpha)'(a) = Df_{\alpha(a)} \cdot D\alpha_a$.

The point that I don't get is that $(f \circ \alpha)'(a)$ is a number, but $Df_{\alpha(a)} \cdot D\alpha_a$. is a linear application. So, I'm a little bit confused, can the number $(f \circ \alpha)'(a)$ be interpreted in some way by a linear application?

Thanks!

1

There are 1 best solutions below

0
On

The derivative of a function $f:\mathbb R\to\mathbb R$ at $x\in\mathbb R$ is a number $c\in\mathbb R$ characterized by the fact that $$f(x+h)=f(x)+ch+o(h^2).$$ The term $ch$ is essentially a linear map $h\mapsto ch$, which is entirely characterized by the number $c$, which is the derivative of $f$. That's how the derivative of a 1d function should be interpreted as a linear map. More conceptually, the tangent function $t(x+h)=f(x)+f'(x)h$ is a linear map (affine linear, to be exact) as a function of $h$, determined by the derivative. The actually linear part $f'(x)h$ is the linear map we're talking about.

For higher dimensions, this works exactly the same: If $\mathrm Df_x$ is the derivative of $f$ in $x$, then $$f(x+h)=f(x)+\mathrm Df_x(h)+o(\vert h\vert^2).$$ Just like in the 1d case.