I've got a question regarding the MVT for vector valued functions which states

$$ f(x)-f(y)= \int_0^1 Df(tx+(1-t)y) dt (x-y) $$ where $ f: \mathbb{R}^n \to \mathbb{R}^n $ differentiable with Jacobian $Df()$ and $x,y \in \mathbb{R}^n$, or in a convex compact set.

The MVT for $f: \mathbb{R}^n \to \mathbb{R} $ states $$ f(x+h)-f(x)=\nabla f(\psi)^Th $$ for some $\psi$ one the line x to x+h.

So if I derivate this term with respect to x(if f is twice differentiable) I got

$$\nabla f(x+h)-\nabla f(x)=\nabla^2 f(\psi)h $$

where $\nabla f:\mathbb{R}^n \to \mathbb{R}^n $

so am I right to assume there exisits an $\psi \in \mathbb{R}^n $ so that this equation holds? Why is that (not) the case? I dont quite understand why we need to integrate the whole term?

Thank you for your help. Greetings.


Consider $f(x,y)=x^4+y^3.$ It is $\nabla f(0,0)=(0,0),\nabla f(1,1)=(4,3)$ and $$\nabla^2 f(x,y)=\begin{pmatrix} 12x^2&0\\0& 6y\end{pmatrix}.$$ Now, $$\nabla f(1,1)-\nabla f(0,0)=(4,3)$$ and $$\nabla^2 f (\theta,\theta)\cdot (1,1)^T=\begin{pmatrix}12\theta^2&0\\0& 6\theta\end{pmatrix}\begin{pmatrix}1\\ 1\end{pmatrix}=\begin{pmatrix}12\theta^2\\ 6\theta\end{pmatrix}.$$

But $$\begin{cases}12\theta^2&=4\\6\theta&=3\end{cases}\iff \begin{cases}\theta&=\pm\frac{1}{\sqrt{3}}\\\theta&=\frac{1}{2}\end{cases}$$ which is impossible.


I add some comments. The vector phi in the mean-value theorem depent on x. Therefore, the differentiation is not correct, because phi become a function.


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.