Understanding k-th derivative formula

analysisderivativesreal-analysis

Let $\Omega \subset \mathbb{R^n} $ an open set, $f \in C^{k}(\Omega)$ and $D^k f(x)$ the k-th derivative of f in $x \in \Omega$

Then it holds:

$$D^k f(x)(v_1,….,v_k)= \sum_{i_1,….,i_k=1}^n (\partial_{i_1}….\partial_{i_k} f)(x)(v_1)_{i_1}….(v_k)_{i_k}$$

For what do I need the $v_k$ on the left side?
How can I understand $(v_1)_{i_1}$ for example?

Best Answer

I am not sure if this answer will be satisfying for you but since noone else provided an answer, I will give it a try.

The reason that the $v_i$ appear on the left hand side is that the k-th derivative of $f$ evaluated at a point $x_0$ is in fact a (symmetric) k-linear (i.e. linear in each entry) map from $(\mathbb{R}^n)^k\to \mathbb{R}$. Each $v_i$ is an element of $\mathbb{R}^n$ and the notation $(v_1)_{i_1}$ simply refers to the $i_1$-th component of the vector $v_1$.

To be more explicit, let's focus on the case $n=2$ and fix some point $x_0\in \Omega\subset \mathbb{R}^2$. Then we can form the derivative $Df(x_0)$, which is a linear map from $\mathbb{R}^2$ to $\mathbb{R}$. In particular we can evaluate this map for some $v_1\in \mathbb{R}^2$ and we get \begin{align} Df(x_0)(v_1)=\sum_{i_1=1}^2\partial_{i_1}f(x_0) (v_1)_{i_1}. \end{align} If we define now $g(x)=Df(x)(v_1)$ then this will again be a smooth function (assuming $f$ is smooth) from $\mathbb{R}^2$ to $\mathbb{R}$. Therefore we can evaluate it's derivative at the point $x_0$ and get again a linear map from $\mathbb{R}^2$ to $\mathbb{R}$. If we evaluate this linear map for some $v_2\in \mathbb{R}^2$, we get

\begin{align} Dg(x_0)(v_2)&=\sum_{i_2=1}^2 \partial_{i_2}g(x_0)(v_2)_{i_2}\\ &=\sum_{i_2=1}^2 \partial_{i_2}\bigg(\sum_{i_1=1}^2\partial_{i_1}f(x_0)(v_1)_{i_1} \bigg)(v_2)_{i_2}\\ &=\sum_{i_2=1}^2\sum_{i_1=1}^2 \partial_{i_2}\partial_{i_1}f(x_0)(v_1)_{i_1}(v_2)_{i_2} \end{align} On the other hand \begin{align} Dg(x_0)(v_2)=D(Df(x_0)(v_1))(v_2)=:D^2f(x_0)(v_1,v_2) \end{align}

Of course this is all not completely rigorous but I tried to show where the multilinearity comes from. Hope this helps!

Related Question