The space $\mathcal{S}$ of rapidly decaying functions is endowed with its usual Fréchet topology. Its topological dual $\mathcal{S}'$ is endowed with the strong topology.

I assume I have a norm $\lVert \cdot \rVert$ on $\mathcal{S}$ that is continuous over $\mathcal{S}$ and denote by $\mathcal{X}$ the Banach space obtained as the completion of $(\mathcal{S},\lVert \cdot \rVert)$.

Is it correct to say that we have the embeddings $$\mathcal{S}\subseteq \mathcal{X} \subseteq \mathcal{S}'?$$ What if $\lVert \cdot \rVert$ is only a semi-norm?


No, this isn't correct. To make things easier consider the "discrete" case of the rapidly decreasing sequences $s$ so that $s'$ is the space of slowly increasing sequences. As a norm on $s$ take $\|x\|= \sum_{n=1}^\infty |x_n|/e^n$. The completion of $(s,\|\cdot\|)$ is then the weighted $\ell^1$-space of all sequences with $\sum_{n=1}^\infty |x_n|/e^n$ finite which is not contained in $s'$.

The point is, that the inclusion $(s,\|\cdot\|) \hookrightarrow s'$ is not continuous. If $\|\cdot\|$ is only a seminorm, $\cal X$ should be the Hausdorff completion of $(s,\|\cdot\|)$ and the canonocal map $s\to \cal X$ will no longer be injective.

  • $\begingroup$ Nice simple example. Thank you. If we write $s' = \cup s_n$ with $(s_n, \lVert \cdot \rVert_n)$ an adequate family of weighted $\ell_2$-spaces, I guess the inclusion $(s,\lVert \cdot \rVert) \hookrightarrow s'$ is continuous if there exists $n$ such that $\lVert \cdot \rVert_n \leq C \lVert \cdot \rVert$? $\endgroup$ – Goulifet Mar 15 '17 at 16:50
  • $\begingroup$ For the seminorm, I am mostly interested by the following case. Let's say that $\mathrm{L}$ is an operator that continuously map $s$ to $\mathcal{X}$ (still the completion of $s$ for the norm $\lVert \cdot \rVert$), and let's assume that $s \subseteq \mathcal{X} \subseteq s'$, then can we identify the completion of $(s ,\lVert \mathrm{L} \cdot \rVert)$ with a subspace embedded in $s'$? $\endgroup$ – Goulifet Mar 15 '17 at 16:55

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.