In $\mathbb{R}^n$ if the derivative of a function exists at a point $x$, then the directional derivative of that function at $x$ in the direction $v$ is an inner product between the derivative at $x$ and the direction $v$.

Does the same idea apply to Gateaux/Frechet derivatives? If the Frechet derivative exists is the Gateaux differential an inner product?


1 Answer 1


Let's concentrate on the Frechet derivative for now. From Wikipedia:

Let $V$ and $W$ be normed vector spaces, and $U \subseteq V$ be an open subset of $V$. A function $f : U \to W$ is called Fréchet differentiable at $x \in U$ if there exists a bounded linear operator $A : V \to W$ such that $$\lim_{\|h\| \to 0} \frac{ \| f(x + h) - f(x) - Ah \|_{W} }{ \|h\|_{V} } = 0.$$

What I want to highlight here is that $A$, which is the Frechet derivative, is defined to be a bounded linear operator.

In the particular case when $V$ is a Hilbert space, and $W$ is the scalar field (real or complex), then the derivative becomes a bounded linear functional. The Riesz representation theorem implies that such functionals take the form $\langle \cdot, u \rangle$, where $u$ is some fixed vector in $V$. So, in the case of maps from Hilbert spaces to their scalar fields, yes, the derivative is indeed expressed via an inner product. Although it's an abuse of notation, we sometimes call $u$ the Frechet derivative, even though it's really the functional $\langle \cdot, u \rangle$.

If $W$ is not the scalar field, then inner products are clearly not appropriate; the derivative is a linearisation and should map into $W$, not the scalar field, as an inner product does. This is true in finite-dimensions too.

Also, if $V$ is not an inner product space, then there's no inner product with which to express the derivative!

Now, the Gateaux derivative is a little bit more general. I've seen a couple of inequivalent definitions of the Gateaux derivative. The version I'm used to seeing looks quite similar to the above definition:

Let $V$ and $W$ be normed vector spaces, $x \in V$, and $U \subseteq V$ such that $x \in \operatorname{core} U$. A function $f : U \to W$ is called Gateaux differentiable at $x \in U$ if there exists a bounded linear operator $A : V \to W$ such that, for every $h \in V$, $$\lim_{t \to 0} \frac{ \| f(x + th) - f(x) - A(th) \|_{W} }{ t } = 0.$$

That is, it's much the same as the above definition, except that the difference quotient has to tend to $0$ invidividually along each line, rather than uniformly along all lines.

However, note that $A$ is still required to be bounded linear operator. By the same logic, you can express this as an inner product whenever $V$ is a Hilbert space and $W$ is the scalar field, otherwise no.

The other definition for Gateaux derivative drops the requirement that $A$ be linear or bounded, and defines it in terms of the directional derivative. The Gateaux derivative of $f$ at $x$ is defined to be the function $$h \mapsto \lim_{t \to 0} \frac{f(x + th) - f(x)}{t},$$ which may not be linear or continuous (even in finite dimensions). In this case, even when $V$ is an inner product space and $W$ is the scalar field, the Gateaux derivative may not be expressible as an inner product.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .