Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $f$ be a real function of a single real variable. Suppose that $f$ is $n$ times differentiable at some $x$, for some integer $n\geq 1$. Making no further assumptions, we have

$$ f(x+h) = f(x) + f'(x)h + \frac{f''(x)h^2}{2} + \ldots + \frac{ f^{(n)}(x)h^n}{n!} + o(h^n) $$

although this tells us nothing about the size of the error on any fixed neighbourhood of $x$ (further regularity assumptions are needed to get the traditional forms for the remainder). Wikipedia has a proof of this which relies on, of all things, L'Hôpital's rule for which a proof is provided as well.

Recently I've been working through the basic theorems of calculus in the Banach space setting. It's not terribly surprising and I have seen it written in a few places that the expansion

$$ f(x+h) = f(x) + f'(x)h + \frac{f''(x)h^2}{2} + \ldots + \frac{ f^{(n)}(x)h^n}{n!} + o(\|h\|^n) $$

is still valid when $f$ is a map of Banach spaces $X \to Y$. In this setting, the $k$th derivative $f^{(k)}(x)$ of $f$ at $x$ is a continuous, symmetric, $k$-linear map $X^k \to Y$ and $h^k$ is short-hand for $(h,\ldots,h) \in X^k$. I would like to know:

How do I prove this last expansion is valid?

I'm having trouble adapting the proof from the single variable case because I can't seem to prove an appropriate analogue of L'Hôpital's rule. An industry standard proof of L'Hôpital's rule depends on Cauchy's mean value theorem and I am not sure whether this admits a Banach space analogue either. In fact, I think I would also be curious to know:

Do Cauchy's mean value theorem or L'Hôpital's rule have natural generalizations in the context of maps between Banach spaces?

Thanks in advance for any replies.

share|improve this question
    
Not sure if there's a way to apply this in the given situation, but (I think) I succeeded in proving $F(x+h+k) - F(x+h) - F(x+k) + F(x) = F''(x)(h,k) + o(\|h\|\|k\|)$ as means to proving derivatives are symmetric maps. –  Mike F Sep 10 '11 at 3:48
    
For given $x, h \in X$ apply the one-dimensional case to $t \mapsto f(x+th)$ and use the chain rule. –  martini Sep 10 '11 at 6:43
    
@martini That would be insufficient because the error term $o$ would depend on the direction $h$, not just its magnitude. –  scineram Sep 10 '11 at 11:28

1 Answer 1

up vote 2 down vote accepted

The classical mean value theorems, like Cauchy, fail when the range is a higher dimensional space. The key result in multidimensional analysis is the Mean Value Inequality: if $u,v\in X$, $f\colon X\rightarrowtail Y$ is differentiable on the open line segment $]u,v[$ and continuous on the closed segment $[u,v]$, then $$\|f(u)-f(v)\|\le\sup_{0<t<1}\|f'(u+tv)(u-v)\|.$$ Now the Taylor theorem is obviously true for $n=1$ by definition, and then induction is used. The key is that the Taylor polynomial of $f'$ is the derivative of the Taylor polynomial of $f$. Thus by the above inequality and the induction hypothesis, and that $f(x)=T(x)$, $$\frac{\|f(x+h)-T(x+h)\|}{\|h\|^{n+1}}\le\sup_{0<t<1}\frac{\|(f-T)'(x+th)(h)\|}{\|h\|^{n+1}}<\varepsilon,$$ when $\|h\|$ is sufficiently small.

share|improve this answer
    
Wow I see it now. Thanks a lot - for some reason I'd convinced myself it was harder than this. –  Mike F Sep 11 '11 at 1:19

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.