Let $F: R^n \to R^m $ be a vector valued function with components $F = (f_1 ,f_2 ,..., f_m).$ What is the most proper and universal accepted definition for the function $F$ be twice differentiable at the point $x$? Does that simply mean $f_i$ be twice-differentiable at point $x$ for all $i = 1,2,...,m$. I know this is the case when we are talking about a function being $C^2$ (see What does it mean for a vector function to be twice differentiable? ).

Or does it mean $F$ be differentiable around $x$ and the derivative of $F$ which is a matrix valued function, denoted by $\nabla F : U \to R^m \times R^n $ be differentiable (in some sense) at the point $x$.

Any reference about that would be appreciated .

  • 1
    $\begingroup$ I think you can define it as just each coordinate being twice continuously differentiable. But this is equivalent to thinking about it as one object, i.e. differentiating a 1-form. Take a look at this: math.stackexchange.com/questions/925322/… $\endgroup$ – zoidberg Dec 11 '18 at 2:41
  • $\begingroup$ Yes, I'm pretty sure that the dimension of the codomain doesn't complicate things - the two definitions ($f_i$ being twice-differentiable vs $\nabla F : U \to \mathbb R^{m\times n}$ being Fréchet differentiable) should coincide. $\endgroup$ – Anthony Carapetis Dec 11 '18 at 3:28

The derivative of $F$ at $x \in \mathbb{R}^n$ is a linear map $dF(x) : \mathbb{R}^n \to \mathbb{R}^m$. It is characterized as the best linear approximation of $F$ of $x$. See for example my answer to Higher dimensional total derivative.

If $F$ is differentiable in an open neighborhood $U$ if $x$, we obtain a map $dF : U \to \mathcal{L} = \mathcal{L}(\mathbb{R}^n, \mathbb{R}^m)$ = vector space of all linear maps $\mathbb{R}^n \to \mathbb{R}^m$. This is a finite-dimensional vector space which can be identified with $\mathbb{R}^{n\cdot m}$. We call $F$ twice differentiable at $x$ is $dF$ is differentiable at $x$.

Note that it essential that $dF(x')$ exists in a neighborhood of $x$.


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.