I'm trying to show that a map $f$ between Banach spaces $X$ and $Y$ is Frechet differentiable at a point $u$. To do this, it is enough to calculate its Gateaux derivative at $u$ (call it $df(u)$) and show that $df(u)$ is continuous: so for every $\epsilon$, there exists a $\delta$ such that if $$\lVert h_1 - h_2 \rVert_X \leq \delta,$$ then $$\lVert df(u)h_1 - df(u)h_2 \rVert_Y < \epsilon.$$ Is that correct to show Frechet differentiability?

Secondly, my Banach spaces $X$ and $Y$ are Hölder spaces. Specifically, let $X = C^{k, \alpha}(S)$, where $S$ is the closure of a set. Let $g$ be a smooth function in its arguments (so it's continuous). Since $S$ is closed and the norm is over a space of that set, can I say that $g$ is bounded above and hence $$\lVert gh_1 - gh_2 \rVert_X \leq C\rVert h_1 - h_2 \rVert_X$$ for some constant?


  • $\begingroup$ Is $g$ a function defined over $S$? And is $gh_1$ pointwise multiplication of functions? $\endgroup$ – Willie Wong Jul 10 '12 at 15:20
  • $\begingroup$ @WillieWong Yes to both questions $\endgroup$ – soup Jul 10 '12 at 15:53

Re: 1st question.

You may be confusing two kinds of continuity. The Gateaux derivative is a linear functional from $X$ to $Y$, so one can talk about the continuity of this linear functional (which is usually a part of the definition of the Gateaux derivative). But the fact that it is a continuous linear functional does not make it a Frechet derivative. The additional requirement in the definition of the Frechet derivative is that the functional $df(u)h$ approximates $f(u+h)-f(u)$ in a uniform way: specifically, $\|df(u)h-(f(u+h)-f(u))\|/\|h\|\to 0$ as $\|h\|\to 0$.

It is true that a continuous Gateaux derivative is a Frechet derivative. But the continuity here is understood with respect to the point $u$ at which the derivative is taken: that it, $\|df(u_1)-df(u)\|\to 0$ when $u_1\to u$ in the norm.

Re: 2nd question: $g$ being bounded above is not a sufficient reason for your conclusion. You should get your hands dirty with the definition of the Hölder norm, and use the smoothness of $g$ in the process.

  • $\begingroup$ Thanks. So in the continuity you give above I guess you refer to the operator norm. Regarding the second question, I tried to get my hands dirty but the Holder norm is very ugly (since I'm using a parabolic Holder norm) with lots of sums and semi-norms. I guess I'll try again.. $\endgroup$ – soup Jul 10 '12 at 18:04

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.