Say I have a Schwartz function $\varphi\in\mathcal{S}(\mathbb{R}^{2n})$. Consider a map $\psi(u)(x)=\langle u,\varphi(x,\cdot)\rangle$ for $u\in\mathcal{S}'(\mathbb{R}^{n}).$ I'd like to know how to show (1) that if $\psi:\mathcal{S}'(\mathbb{R}^n)\rightarrow \mathcal{S}(\mathbb{R}^n)$ and (2) that the map $\psi$ is continuous.

For (1), I initially wrote $$|x^\alpha D^\beta \psi (u)(x)|=|\langle x^\alpha D^\beta u,\varphi(x,\cdot)\rangle|=|\langle u,x^\alpha D^\beta\varphi(x,\cdot)\rangle|$$ for any $x$, but I'm not sure if moving the $x^\alpha D^\beta$ over actually makes sense here. Typically, we do this when $x$ is the variable that the action is with respect to, which is not the case here. If this were allowed, then the fact that $\varphi$ is Schwartz would guarantee that $x^\alpha D^\beta \varphi(x,\cdot)$ is Schwartz, and this would be finite for any $x$ since $u$ is tempered. So, it comes down to simply understanding how the $x^\alpha D^\beta$ action is defined.

For (2), I tried to use the sequential criterion, but I could not show that if $u_n\rightarrow u$ in $\mathcal{S}'$ (i.e. weak$^*$/pointwise), then $\psi (u_n)\rightarrow \psi (u)$ in $\mathcal{S}$ (i.e. in every Schwartz seminorm). I think it would come from the first part, as if I could move over the $x^\alpha D^\beta,$ then I would have $$x^\alpha D^\beta(\psi(u_n)-\psi(u))=\langle (u_n-u), x^\alpha D^\beta \varphi(x,\cdot)\rangle\rightarrow 0$$ since $x^\alpha D^\beta \varphi(x,\cdot)$ is Schwartz and $\langle u_n, f\rangle \rightarrow \langle u, f\rangle$ for any $f\in\mathcal{S}.$

Alternatively, since $\mathcal{S}$ is dense in $\mathcal{S}'$ in the weak$^*$ topology, might it suffice to check these claims for $u\in\mathcal{S}$ and extend by density? It's not clear to me how that extension process would work.

EDIT: Perhaps, I can make sense of the derivative claim by showing that difference quotients converge? Also, I believe that moving that $x^\alpha$ from one side to the other is justified since, for a fixed $x$, this is as good as a constant as far as $u$ is concerned, so I can use linearity. Is that correct?

  • 2
    $\begingroup$ Asking whether $\psi:\mathcal{S}'(\mathbb{R}^n)\rightarrow \mathcal{S}(\mathbb{R}^n)$ doesn't make much sense, because in fact $\psi:\Bbb C\to\Bbb C$. Maybe what you actually had in mind was $\psi(u)(x)=\langle u,\varphi(x,\cdot)\rangle$? Or, with the notation you used, the obvious question seems to be whether $\psi\in\mathcal S$ $\endgroup$ Aug 2, 2019 at 14:09
  • $\begingroup$ @DavidC.Ullrich Yes, I meant to write $\psi (u)(x),$ my mistake! $\endgroup$
    – user269711
    Aug 2, 2019 at 19:55
  • 1
    $\begingroup$ Now I'm not, did you show that continuous on the Schwartz space implies $|<u,\phi>| \le C_u \sum_{|r|,|s| \le N_u} \|D_y^r y^s \phi\|_\infty$, the argument is quite the same as in mathoverflow.net/a/348398/84768 $\endgroup$
    – reuns
    Dec 23, 2019 at 20:29
  • $\begingroup$ @reuns Do you mean have I seen the definition of continuity for elements of $\mathcal{S}'$ in that manner. If so, I have seen that argument before awhile ago. I think what was tripping me up on your answer was that I wasn't sure of the definition of a continuous function between two locally convex topological vector spaces in terms of seminorms. I looked it up yesterday and saw that was what your answer was showing. Looking back at that argument, I see exactly what you mean. Thanks! $\endgroup$
    – user269711
    Dec 23, 2019 at 20:33

1 Answer 1


For $\varphi \in S(\mathbb{R}^n \times \mathbb{R}^n), u \in S'(\mathbb{R}^n)$ let $$T[u](x) = <u,\varphi(x,.)> \ \in R, \qquad (D_x^a x^b) T[u](x_0)=<u,(D_x^a x^b \varphi)(x_0,.)>$$

Since $u$ is a tempered distribution you know that for some $N_u,C_u$, for all $\phi \in S(\mathbb{R}^n)$ $$|<u,\phi>| \le C_u \sum_{|r|,|s| \le N_u} \|D_y^r y^s \phi\|_\infty$$ whence

$$\sup_{x_0} |(D_x^a x^b) T[u](x_0)|=\sup_{x_0} |<u,(D_x^a x^b \varphi)(x_0,.)>|\\ \le \sup_{x_0} C_u\sum_{|r|,|s| \le N_u} \|D_y^r y^s (D_x^a x^b \varphi)(x_0,.)\|_\infty \le C_u \sum_{|r|,|s| \le N_u} \|D_y^r y^s D_x^a x^b \varphi \|_\infty = C_u \kappa(N_u,a,b)$$

Which is quite the definition of that $u \mapsto T[u]$ is a continuous linear map $S'(\mathbb{R}^n) \to S(\mathbb{R}^n)$.

  • $\begingroup$ I don't think this is correct. I haven't checked but I also think the map is not continuous if $S'$ is given its usual (wrong) topology, i.e., the weak-star one. If $S'$ has the strong topology then everything works fine. Not only $\psi$ is continuous but the map $\phi\mapsto\psi$ is a TVS isomorphism (all topologies being strong,i.e., given by uniform convergence on bounded sets). $\endgroup$ Dec 30, 2019 at 13:33
  • $\begingroup$ Also, this result is just a part of Fubini's Theorem for temperate distributions. The best way to prove it is to rewrite it in terms of sequence spaces, thanks to Hermite functions. $\endgroup$ Dec 30, 2019 at 13:35
  • $\begingroup$ @AbdelmalekAbdesselam I didn't notice, but I think you're right. The RHS of the inequality would need to be in terms of the seminorms coming from the weak* topology, I think. Also, the result should use the weak* topology on $\mathcal{S}',$ not the strong one. For reference, I was trying to fill in the details of the proof of theorem 4.1 on page 40 of math.berkeley.edu/~evans/semiclassical.pdf $\endgroup$
    – user269711
    Jan 3, 2020 at 20:41
  • $\begingroup$ Why do you care, continuous for the strong topology implies continuous for the weak topology. @user269711 $\endgroup$
    – reuns
    Jan 3, 2020 at 20:49
  • $\begingroup$ @reuns: Could you state a bit more completely the thm about strong continuity implies weak continuity. For what maps and spaces? $\endgroup$ Jan 6, 2020 at 17:57

You must log in to answer this question.

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