I'm working on the following problem from Conway V.4.

Let $X$ be compact and supppose there is a norm on $C(X)$ that is given by an inner product making $C(X)$ into a Hilbert space such that for every $x \in X$ the functional $\Lambda_x : f\mapsto f(x)$ is continuous with respect to the Hilbert space norm. Show $X$ is finite.

Idea on approach: I think we need to assume $X$ is infinite and come to some sort of contradiction.

Facts noted:

  • $\text{ball}{(C(X))}$ is weakly compact.

  • $X^* \subset C(X)$, $X^*$ is definitely a subspace, and if it is closed under the norm induced by the inner product we have $X^*$ is Hilbert and hence reflexive. (Since I am not sure of the closedness of $X^*$ under the norm, I am not sure if this could be useful.)

  • The hypothesis is allowing us to extend the notion of a weak-$*$ topology on $X^*$ to $C(X)$.

  • The continuity of the $\Lambda_x$'s and Riesz Rep theorem imply that there is a $g \in C(X)$ such that $\Lambda_x(f) = \langle f,g \rangle = |f(x)| \leq M \langle f,f\rangle^{1/2}$.

I'm not entirely sure how to put these pieces together. I was hoping to take a $X \supset \{x_n\}$ with $x_n \to x$, and by weak compactness we have for any sequence $\{f_j\} \subset \text{ball}(C(X))$ there is a subsequence such that $f_{j_k}$ converges weakly to some $f \in \text{ball}(C(X))$, which by our hypotheses implies that $f_{j_k}(x_n) \to f(x_n)$ as $k \to\infty$. We also know that $f_{j_k}(x_n) \to f_{j_k}(x)$ as $n\to \infty$. But there is nothing contradictory I can see coming from this observation. Any hints would be appreciated. Thanks.

Edit: Also not entirely sure how $X$ compact fits in exactly, as I don't think I've taken much advantage of that fact in my observations except for claiming there is a convergent sequence.

  • $\begingroup$ Is $X$ a general compact topological space? $\endgroup$ Jul 23, 2018 at 0:36
  • $\begingroup$ Is $X$ Hausdorff? $\endgroup$
    – copper.hat
    Jul 26, 2018 at 14:49
  • $\begingroup$ Conway gives no clarification of what topology is associated with $X$ in the problem set, section, or chapter. The section is titled Weak Topologies, and the weakest topological space we seem to be working with in this section is a locally convex space. $\endgroup$ Jul 26, 2018 at 17:57

1 Answer 1


Hint: Use the uniform boundedness principle to show that the Hilbert space norm on $C(X)$ must be equivalent to the sup norm. There are probably many approaches you can take after that, but one approach is that it would suffice to show that $C(X)$ (with the sup norm) is not reflexive if $X$ is infinite.

A full solution is hidden below.

The functionals $\Lambda_x$ are pointwise bounded (since every $f\in C(X)$ is a bounded function), and therefore must be uniformly bounded. This means exactly that the identity map $C(X)\to C(X)$ is bounded where the domain has our Hilbert space norm and the codomain has the sup norm. By the open mapping theorem, this implies the Hilbert space norm is equivalent to the sup norm on $C(X)$.

Since any Hilbert space is reflexive, this means $C(X)$ is reflexive in the sup norm. By the Riesz-Markov theorem, the dual of $C(X)$ is the space $M(X)$ of regular complex Borel measures on $X$, acting on $C(X)$ by integration. If $X$ is infinite, there exists a non-isolated point $x\in X$, and then the functional $\mu\mapsto\mu(\{x\})$ is a bounded functional on $M(X)$ which is not given by integration of any continuous function on $X$. So in that case $C(X)$ is not reflexive, and so $X$ must be finite.

  • $\begingroup$ This was a great answer, thank you. You mentioned there were various other approaches to this problem once the equivalence of norms was proved. Do any of them involve not knowing about the relationship between $C(X)$ and $M(X)$? $\endgroup$ Jul 23, 2018 at 15:51
  • 1
    $\begingroup$ I haven't actually worked it through, but I bet you could explicitly show that no norm equivalent to the sup norm can satisfy the parallelogram law, by finding examples of functions where the parallelogram law fails "arbitrarily badly" (such that no constant factor error in the norm could fix all the examples). $\endgroup$ Jul 23, 2018 at 16:09
  • $\begingroup$ +1: This is a nice answer, albeit it took me a while to get my head around it! $\endgroup$
    – copper.hat
    Jul 27, 2018 at 3:46
  • $\begingroup$ I would add that you don't actually need the full strength of Riesz-Markov: you just need the more elementary claim that $M(X)$ embeds isometrically in the dual of $C(X)$ by integration. Then a functional on $M(X)$ which does not come from any element of $C(X)$ extends to the full dual of $C(X)$ by Hahn-Banach. $\endgroup$ Aug 23, 2018 at 4:52

You must log in to answer this question.

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