1

We consider the following two families of seminorms on the Schwartz space and want to show that they induce the same topology.

First family: $(\| \cdot \|_N) $ for $N \in \mathbb{N}$ defined by $$ \| f \|_N=\sup_{\vert \alpha \vert \leq N} \sup_{x\in \mathbb{R}^n} (1+\vert x \vert ^2)^N \vert (\partial^\alpha f)(x) \vert. $$

Second family: $(\| \cdot \|_{(K,\alpha)})$ for $K \in \mathbb{N}$ and $\alpha$ a multi-index defined by $$ \| f \|_{(K,\alpha)}=\sup_{x \in \mathbb{R}^n} (1+\vert x \vert)^K \vert (\partial^\alpha f)(x)\vert. $$

I am not very familiar with Frechet spaces but from my topological knowledge I suspect that it is sufficient to show $$ \| f \|_N \leq C \sum_{i=1}^l \| f \|_{(K_i,\alpha_i)} $$ for some for each $N \in \mathbb{N}$, $C>0$ and $l,K_i\in \mathbb{N}$ and some multi-indices $\alpha_i$ where $C,l,K_i,\alpha_i$ may depend on $N$ and

$$ \| f \|_{(K,\alpha)} \leq C \sum_{i=1}^l \| f \|_{N_i} $$

for each $K \in \mathbb{N}$ and each multi-index $\alpha$ for some $C>0$, some $l \in \mathbb{N}$ and $N_i \in \mathbb{N}$ where $C,l,N_i$ may depend on $K$ and $\alpha$.

Is this approach correct or do I need to change something?

Benjamin
  • 1,220
Jonas Lenz
  • 1,435
  • Thanks a lot. :) – Jonas Lenz Oct 01 '17 at 19:16
  • In general you need to show the collection of neighborhoods $U,V$ induced by the semi-norms are equivalent, ie. $U_\epsilon \subset V_\delta \subset U_{C \epsilon}$ – reuns Oct 01 '17 at 19:24
  • If I have a subbase for both topology and can show the inclusion where all three sets are an element of the "correct" subbase then I should have shown that the topologies are equivalent. As this would show that "both" identity mappings are continuous.

    All this is based on the hope that I remember something from my topology class...

    – Jonas Lenz Oct 01 '17 at 19:40
  • Yes. So you need to remember how semi-norms induce a collection of neighborhoods (around $0$ it is enough since we are in a vector space). But your solution (that the Schwartz space is the intersection of some Banach spaces $B_N \supset B_{n+1}$ given by the norms you stated) is correct. – reuns Oct 01 '17 at 19:45

1 Answers1

3

It seems that the problem has already been solved but it happened that I have typed down the answer for myself so let's share.

A slightly more general statement can be found in Reed and Simon, vol. 1, p.126 (proof left as an exercise...) and I will stick to their framework and just proove one crucial ingredient.

Let $X$ be a vector space with a topology induced by a family $(\rho_{\alpha})_{\alpha \in A}$ of semi-norms. Let $p: X \mapsto [0,+\infty )$ be another semi-norm. It is continuous at $0$ iff $$\exists\, C >0,\ (\alpha_1,\ \cdots, \alpha_n) \in A^n\, \big/\ \forall\, x\in X,\quad p(x) \leq C \big( \rho_{\alpha_1}(x)+ \cdots + \rho_{\alpha_n}(x)\big)$$ A variant of this could also be $$\exists\, C >0,\ (\alpha_1,\ \cdots, \alpha_n) \in A^n\, \big/\ \forall\, x\in X,\quad p(x) \leq C \sup_{1\leq i \leq n} \rho_{\alpha_i}(x)$$

$(\Rightarrow):$ Let's write continuity of $p$ at $0$ with $\epsilon = C > 0$: there exists a neighborhood V of $0$ such that $p(V) \subset [0,M [$ (open or closed, not important, just a neighborhood of $p(0)=0$ in $\mathbb{R}_+$). Now since the topology of $X$ is generated by $(\rho_{\alpha})_{\alpha \in A},\ V$ contains an open subset/can be chosen of the form $$ V:= \big\lbrace y \in X,\ \rho_{\alpha_i}(y-0) =\rho_{\alpha_i}(y) < \delta,\ \forall\, i\in [\![1,n]\!]\big\rbrace $$ (implicitly, "there exists $\delta >0,\ n\in N$ and a subset $\{\alpha_1,\ \cdots, \alpha_n\}\subset A$" such that ...).

Now for an arbitrary $x\in X$,

  • either every $\rho_{\alpha_i}(x)=0$ so that $x\in V$ and $p(x) < C$ (then by homogeneity of $p$, one must have $p(x)=0$),
  • or $\rho_{\alpha_1}(x)+ \cdots + \rho_{\alpha_n}(x) > 0,\ $ resp. $ \sup_{1\leq i \leq n} \rho_{\alpha_i}(x) > 0$ and $$ \frac{x}{\rho_{\alpha_1}(x)+ \cdots + \rho_{\alpha_n}(x)} \in V,\quad \frac{x}{\sup_{1\leq i \leq n} \rho_{\alpha_i}(x) } \in V $$ (again, homogeneity of the $\rho_{\alpha_i}$) so that $$ p\bigg( \frac{x}{\rho_{\alpha_1}(x)+ \cdots + \rho_{\alpha_n}(x)}\bigg) = \frac{p(x)}{\rho_{\alpha_1}(x)+ \cdots + \rho_{\alpha_n}(x)} < C,\quad \frac{p(x)}{\sup_{1\leq i \leq n} \rho_{\alpha_i}(x) } < C $$

$(\Leftarrow):$ If the inequality holds, then for any $\epsilon > 0$, consider the following neighborhood of $0$: $$ V := \Big\lbrace y\in X,\ \rho_{\alpha_i}(\mathbf{y}) < \frac{\epsilon}{n\, C},\ \forall\, i\in [\![1,n]\!] \Big\rbrace $$ (or for the variant $$ V := \Big\lbrace y\in X,\ \rho_{\alpha_i}(\mathbf{y}) < \frac{\epsilon}{1+ C},\ \forall\, i\in [\![1,n]\!] \Big\rbrace $$ )

Then $p(V) \subset [0,\epsilon[$.

Remarks: This little proposition justifies that one can choose a directed family of semi-norms and it will play a role in the proof that the Schwartz space is Fréchet (metrizable complete). Here we do not talk about completeness.

Noix07
  • 3,839