For $u\in\mathcal E'(\mathbb R^n)$ and $v\in\mathcal S'(\mathbb R^n)$, we defined $u\ast v$ by $\langle u\ast v, \phi\rangle = \langle v, \check u \ast \phi \rangle$ for all $\phi\in\mathcal S(\mathbb R^n)$ (we showed before that this definition makes sense).

I shall now show, that $\langle u\ast v, \phi\rangle = \langle u, \phi\ast \check v \rangle$ for all $\phi\in\mathcal S(\mathbb R^n)$ holds aswell

I reduced that to the case where $v$ is a smooth function which additionally is of polynomial growth but I don't know how to proceed, any hints?

  • $\begingroup$ My standard idea would be to show that both expressions are continuous in $v$ for the topology of $\mathcal{S}'$. Did you try this? $\endgroup$ – Vobo Jul 8 '15 at 16:43

Notation: For a fixed function $f$ defined on $\mathbb{R}^{n}$ and $y\in\mathbb{R}^{n}$, we write $\tilde{f}(x)=f(-x)$ and $(\tau^{y}f)(x)=f(x-y)$.

I am assuming that you are familiar with the following result.

Theorem. For $\phi\in\mathcal{S}(\mathbb{R}^{n})$, the tempered distribution $u\ast\phi$ coincides with a $C^{\infty}(\mathbb{R}^{n})$ function $\langle{u,\tau^{x}\tilde{\phi}}\rangle$ satisfying the growth estimates $$\left|\partial^{\alpha}\langle{u,\tau^{x}\tilde{\phi}}\rangle\right|\leq C_{\alpha}(1+\left|x\right|)^{N_{\alpha}},\qquad\forall x\in\mathbb{R}^{n}$$ for all multi-indices $\alpha$. Moreover, if $u$ has compact support, then $u\ast\phi$ coincides with a Schwartz function.

And I leave it to you to prove the following lemma.

Lemma. Let $\phi\in\mathcal{S}(\mathbb{R}^{n})$ be such that $\int\phi=1$, and set $\phi_{\epsilon}:=\epsilon^{-n}\phi(\epsilon^{-1}\cdot)$. Then $$f\ast\phi_{\epsilon}\rightarrow f \text{ in }\mathcal{S}(\mathbb{R}^{n}), \qquad\forall f\in\mathcal{S}(\mathbb{R}^{n})$$ as $\epsilon\downarrow 0$.

By duality, we see that for any tempered distribution $u$, $$\langle{u\ast\phi_{\epsilon},f}\rangle=\langle{u,f\ast\widetilde{\phi_{\epsilon}}}\rangle=\langle{u,f\ast(\tilde{\phi})_{\epsilon}}\rangle\rightarrow \langle{u,f}\rangle$$ as $\epsilon\downarrow 0$. Whence, $u\ast\phi_{\epsilon}\rightarrow u$ in $\mathcal{S}'(\mathbb{R}^{n})$ as $\epsilon\downarrow 0$.

Define linear functionals $w_{1}$ and $w_{2}$ on $\mathcal{S}(\mathbb{R}^{n})$ respectively by $$\langle{w_{1},\psi}\rangle:=\langle{u,\tilde{v}\ast\psi}\rangle, \quad \langle{w_{2},\psi}\rangle:=\langle{v,\tilde{u}\ast\psi}\rangle \qquad\forall \psi\in\mathcal{S}(\mathbb{R}^{n})$$ If I read your post correctly, you know how to verify that $w_{1}$ and $w_{2}$ are tempered distributions, so I will omit that detail. If we can show that $w_{1}\ast\phi=w_{2}\ast\phi$ in $\mathcal{S}'(\mathbb{R}^{n})$, then by replacing $\phi$ with $\phi_{\epsilon}$ and letting $\epsilon\downarrow 0$, we conclude that $w_{1}=w_{2}$ in the sense of tempered distributions.

Fix $\phi\in\mathcal{S}(\mathbb{R}^{n})$ and observe that for any $\psi\in\mathcal{S}(\mathbb{R}^{n})$, \begin{align*} \langle{w_{1}\ast\phi,\psi}\rangle=\langle{w_{1},\tilde{\phi}\ast\psi}\rangle&=\langle{u,\tilde{v}\ast(\tilde{\phi}\ast\psi)}\rangle\\ &=\langle{u,(\tilde{v}\ast\psi)\ast\tilde{\phi}}\rangle\\ &=\int_{\mathbb{R}^{n}}(u\ast\phi)(x)(\tilde{v}\ast\psi)(x)\mathrm{d}x\\ \end{align*} where we use the commutativity of the convolution of ordinary functions, and the associativity of the convolution of a function and a distribution.

Similarly, \begin{align*} \langle{w_{2}\ast\phi,\psi}\rangle=\langle{v,\tilde{u}\ast(\tilde{\phi}\ast\psi)}\rangle=\langle{v,(\tilde{u}\ast\tilde{\phi})\ast\psi}\rangle&=\langle{v\ast\tilde{\psi},\tilde{u}\ast\tilde{\phi}}\rangle\\ \end{align*} But \begin{align*} \int_{\mathbb{R}^{n}}(u\ast\phi)(x)(\tilde{v}\ast\psi)(x)\mathrm{d}x=\int_{\mathbb{R}^{n}}\langle{u,\tau^{x}\tilde{\phi}}\rangle\langle{v,\widetilde{\tau^{x}\tilde{\psi}}}\rangle\mathrm{d}x&=\int_{\mathbb{R}^{n}}\langle{u,\tau^{x}\tilde{\phi}}\rangle\langle{v,\tau^{-x}\psi}\rangle\mathrm{d}x\\ &=\int_{\mathbb{R}^{n}}\langle{u,\tau^{-x}\tilde{\phi}}\rangle\langle{v,\tau^{x}\psi}\rangle\mathrm{d}x\\ &=\int_{\mathbb{R}^{n}}\langle{u,\widetilde{\tau^{x}\phi}}\rangle\langle{v,\tau^{x}\psi}\rangle\mathrm{d}x\\ &=\langle{v\ast\tilde{\phi},\tilde{u}\ast\tilde{\psi}}\rangle \end{align*} This completes the proof.

  • $\begingroup$ You need the lemma for $\phi\in\mathcal{S}'$ and the convergence $\phi\ast f_\varepsilon \to \phi$ in $\mathcal{S}'$. $\endgroup$ – Vobo Aug 7 '15 at 22:29
  • $\begingroup$ @Vobo: I don't follow your comment. What exactly is the issue in the solution? After the statement of the lemma, there is a remark that for any tempered distribution $u$ fixed, we have that $\forall f\in\mathcal{S}$, $\langle{u\ast \phi_{\epsilon},f}\rangle=\langle{u,\tilde{\phi}_{\epsilon}\ast f}\rangle\rightarrow\langle{u,f}\rangle$ as $\epsilon\rightarrow 0$. $\endgroup$ – Matt Rosenzweig Aug 7 '15 at 22:46
  • $\begingroup$ Ahh, sorry. I didn't read that section. $\endgroup$ – Vobo Aug 7 '15 at 22:52
  • $\begingroup$ where can I find the proof of the theorem of your answer? @MattRosenzweig $\endgroup$ – Andrew Mar 16 '16 at 19:20
  • 1
    $\begingroup$ @JohnMartin : L. Grafakos, Classical Fourier Analysis, somewhere in Chapter 2. $\endgroup$ – Matt Rosenzweig Mar 16 '16 at 19:22

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.