Let $G$ be an open subset of $\mathbb R ^n$ and $D(G)$ denotes the set of smooth functions with compact support in $G$.

Consider following families of seminorms,

  1. For $f\in D(G)$

$||f||_N = \sup \{ | D^{\alpha}f(x)|: x\in G,|\alpha| \leq N \}$ for $N\in \mathbb N$.

  1. Let $K_n$ be a nested sequence of compact sets which exhaust $G$ then for $f \in D(G)$ define

$\nu_N (f) = \sup \{ | D^{\alpha}f(x)|: x\in K_N,|\alpha| \leq N \}$ for $N\in\mathbb N$.

where $\alpha$ is a multi-index.

Are topologies induced on $D(G)$ by above two families same?

My guess is NO but I am unable to prove it.

EDIT- I am interested in this question because on $C^\infty(G)$ we give topology induced by family 2. In chapter 6 of Rudin’s Functional Analysis, it says family 1 doesn’t give good topology on $D(G)$ (it’s not complete). But Rudin didn’t talk about family 2 even though it’s subspace topology on $D(G)$ as a subset of $C^\infty(G)$.

  • $\begingroup$ In the definition of $\nu_N$, you wrote $x\in K_N$. Did you mean $x\in K_m$? Since $N$ is already used as the size of the maximal multi index? $\endgroup$ Dec 26 '18 at 7:33
  • $\begingroup$ It’s $x\in K_N$. In fact it’s a topology which we give on $C^\infty (G)$. $\endgroup$
    – Mayuresh L
    Dec 26 '18 at 8:07
  • $\begingroup$ Can you define the natural basis of neighborhoods of $0$ in both topology ? (the translates of those neighborhoods are a basis of the topology). Will a neighborhood in the 1st topology be open in the 2nd topology, or is it containing some open set of the 2nd topology ? $\endgroup$
    – reuns
    Dec 27 '18 at 6:13
  • $\begingroup$ @reuns I tried that but couldn’t figure out! $\endgroup$
    – Mayuresh L
    Dec 27 '18 at 6:39
  • $\begingroup$ Let $\|f\|_{K_n,n}= \sup_{x \in K_n, |\alpha| \le n}|D^\alpha f(x)|$ and $U_1(r,N) = \{ f \in C^\infty_c(G), \forall n \le N, \|f\|_{G,n}<r_n \}$. Is it open in the 1st topology ? Does it contain an open set in the 2nd topology ? (let $T(C^\infty_c(G), \|.\|_{K_n,n})$ be the set of open sets of the topological vector space $C^\infty_c(G), \|.\|_{K_n,n}$, then $T_2 = \bigcap_n T(C^\infty_c(G), \|.\|_{K_n,n})$) $\endgroup$
    – reuns
    Dec 27 '18 at 12:19

With the topology from 2, for functions in $C^\infty(\Omega)$, $\lim_n f_n = f$ can be thought as of $D^\alpha f_n \rightarrow D^\alpha f$ uniformly on any compact subsets, and for each $|\alpha| < \infty$.

One can construct $f_n\in \mathcal{D}(\Omega)$ and yet the limit $f\in C^\infty(\Omega)$ does not have compact support.

2 gives a weaker topology than 1 on $\mathcal{D}(\Omega)$, since 1 is "converge uniformly" and 2 is "converge locally uniformly". For example, let $\phi$ have support in $[0,1]$, define $$\psi_m(x) = \phi(x) + \phi(x-1) + \cdots + \phi(x-m)$$ this sequence is Cauchy in 2, but not Cauchy in 1. This also shows that $\mathcal{D}(\Omega)$ with the topology induced by 2 is not complete. (Similar to Rudin's example on page 151.)

  • $\begingroup$ So we are not interested in topology 2 because it also is not complate. But how to justify the line “2 gives a weaker topology than 1 on $D(\Omega)$”? $\endgroup$
    – Mayuresh L
    Dec 28 '18 at 15:37
  • $\begingroup$ Two families of semi-norms generate the same topology if and only if any semi-norm from one family can be bounded by some semi-norm of the other family. In this case $ \nu_N \leq C\|\cdot \|_N$ is clear, and I don't see a way to have the bound the other way around. $\endgroup$
    – Xiao
    Dec 28 '18 at 16:32
  • $\begingroup$ Maybe the more precise statement is given two families of semi-norms $\mathcal P$ and $\mathcal Q$, $\tau_P$ is finer, i.e. $\tau_Q \subset \tau_p$ if for each $q\in \mathcal {Q}$, there exist $n$ semi-norms $p_1, \cdots, p_n \in \mathcal {P}$ and a constant $C$ such that $$q\leq C \max \{ p_1, \cdots p_n\}.$$ $\endgroup$
    – Xiao
    Dec 28 '18 at 16:43
  • $\begingroup$ Could you give reference for statement in your last comment?. Thanks a lot for your help. $\endgroup$
    – Mayuresh L
    Dec 28 '18 at 17:08
  • 1
    $\begingroup$ @MayureshL It should be easy to see with the metric explicitly. For $k,l\geq M$, $|\psi_k -\psi_l|$ will vanish on $[0, M]$. This means $\nu_N (|\psi_k -\psi_j|)$ will all be zero for $N=1, 2,\cdots, M'$. $\endgroup$
    – Xiao
    Dec 31 '18 at 17:58

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.