Is my application of chain rule on Banach spaces correct?

63 Views Asked by At

I'm reviewing my understanding on derivative on Banach spaces. Could you please verify if it looks fine or contains logical gaps/errors? Thank you so much for your help!

Suppose that $E,F$ are Banach spaces and that $f:E \to F$ is differentiable on $E$. For $a, v \in E$, we define a map $$\begin {array}{l|rcl} g & \mathbb R & \longrightarrow & F \\ & t & \longmapsto & f(a+tv) \end{array}$$

Compute $\partial g (t)$.


My understanding:

By the definition of total derivative, we have $\partial f (x) \in \mathcal L(E,F)$, i.e. $$\begin {array}{l|rcl} \partial f (x) & E & \longrightarrow & F \\ & y & \longmapsto & \partial f (x) (y) \end{array}$$

Here $\partial f (x) (y)$ denotes the value of the map $\partial f (x)$ when evaluated at $y$. Hence $\partial f (x) (y)$ is not the arithmetic product of $\partial f (x)$ and $y$.

Consider the map $$\begin {array}{l|rcl} h & \mathbb R & \longrightarrow & E \\ & t & \longmapsto & a+tv \end{array}$$

It follows that $$\begin {array}{l|rcl} \partial h(t) & \mathbb R & \longrightarrow & E \\ & l & \longmapsto & lv \end{array}$$

It follows from $g = f \circ h$ and the chain rule that $\partial g (t) = \partial f (h(t)) \circ \partial h(t)$. This means $$\begin {array}{l|rcl} \partial g(t) & \mathbb R & \longrightarrow & F \\ & l & \longmapsto & \partial f(a+tv)(lv) \end{array}$$ and consequently $$\forall t,l \in \mathbb R:\partial g(t) (l) = \partial f(a+tv)(lv)$$

It follows from $\partial f(a+tv)$ is a linear map that $$\forall t,l \in \mathbb R:\partial g(t) (l) = \Big(\partial f(a+tv)(v)\Big) \cdot l$$

Here $\Big(\partial f(a+tv)(v)\Big) \cdot l$ denotes the arithmetic product of $\partial f(a+tv)(v)$ and $l$.

By the the definition of directional derivative, we have $$\forall t,l \in \mathbb R:\partial g(t) (l) = \Big(D_v f(a+tv)\Big) \cdot l$$

Here $\Big(D_v f(a+tv)\Big) \cdot l$ denotes the arithmetic product of $D_v f(a+tv)$ and $l$.

1

There are 1 best solutions below

7
On BEST ANSWER

I don't think you've told us the question you're trying to solve. But it seems that you have written some map $g:\mathbb R \to F$ as a composition $$ g = f \circ h$$ and are trying to apply chain rule. Note that the functions are related to the spaces like

$$\mathbb R \xrightarrow{h} E \xrightarrow{f} F$$ in particular, $\partial h$ has no reason to care about $F$. It should be $$\partial h : \mathbb R\to L(\mathbb R,E)$$ or $$ \partial h(t) : \mathbb R \to E.$$

The chain rule is then correctly applied, if I understand your yet undefined symbols correctly, but I would advise against using the symbol $l$ for a parameter and then writing it as if its a linear map $l(\dots)$.

PS: following the discussion on Meta here: The problem with proof verification I think it would be better if in the future, you posted your solution as an answer to your own question.