My question concerns the convolution theorem from harmonic analysis. In particular, I would like to know whether one can extend the result to a particular case.

We will define the Fourier transform of a Schwartz function $\varphi \in \mathcal{S}(\mathbb{R})$ by \begin{equation} \mathcal{F}(\varphi)(\xi):=\int_\mathbb{R} \varphi(x) \mathrm{e}^{-2 \pi \mathrm{i}x\xi} \, \mathrm{d}x. \end{equation} Moreover we define the convolution of two Schwartz functions $\varphi, \psi \in \mathcal{S}(\mathbb{R})$ by \begin{equation} \varphi \ast \psi (x) := \int_\mathbb{R} \varphi(y)\psi(x-y) \, \mathrm{d}y. \end{equation} With these definitions one can prove the convolution theorem:

Let $f,g \in \mathcal{S}(\mathbb{R})$. Then, \begin{equation} \mathcal{F}(f \cdot g) = \mathcal{F}(f) \ast \mathcal{F}(g). \end{equation}

One can prove the same for $f,g \in L^1(\mathbb{R})$ with $\mathcal{F}(f),\mathcal{F}(g) \in L^1(\mathbb{R})$. Using the notion of tempered distributions we can additionally extend this theorem quite easily:

We first set $\varphi^\#(x):=\varphi(-x)$, for $\varphi \in \mathcal{S}(\mathbb{R})$. Now, let $T \in \mathcal{S}^\prime(\mathbb{R})$ and $\varphi \in \mathcal{S}(\mathbb{R})$, we can then define $T \ast \varphi \in \mathcal{S}^\prime(\mathbb{R})$ and $\varphi \cdot T \in \mathcal{S}^\prime(\mathbb{R})$ by \begin{equation} \langle T \ast \varphi, \psi \rangle := \langle T, \varphi^\# \ast \psi \rangle, \qquad \langle \varphi \cdot T , \psi \rangle := \langle T, \varphi \cdot \psi \rangle, \end{equation} where $\psi \in \mathcal{S}(\mathbb{R})$ and $\langle \cdot , \cdot \rangle$ denotes the duality pairing in the Schwartz space. We also define the Fourier transform of a tempered distribution $T \in \mathcal{S}^\prime(\mathbb{R})$ by \begin{equation} \langle \mathcal{F}(T),\psi \rangle := \langle T, \mathcal{F}(\psi) \rangle, \end{equation} which gives us another tempered distribution $\mathcal{F}(T) \in \mathcal{S}(\mathbb{R})$. The convolution theorem now becomes:

Let $T \in \mathcal{S}^\prime(\mathbb{R})$ and $\varphi \in \mathcal{S}(\mathbb{R})$. Then, \begin{equation} \mathcal{F}(\varphi \cdot T) = \mathcal{F}(T) \ast \mathcal{F}(\varphi). \end{equation}

The Main Question:

Let us denote by $\mathcal{C}_0(\mathbb{R})$ the continuous functions on $\mathbb{R}$ vanishing at infinity, let $\sigma > 0$ and $p \in [1,\infty)$. Let us take $f,g \in \mathcal{C}_0(\mathbb{R}) \subset L^\infty(\mathbb{R}) \subset \mathcal{S}^\prime(\mathbb{R})$ and $\mathfrak{f}, \mathfrak{g} \in L^p([-\sigma,\sigma]) \subset L^1([-\sigma,\sigma])$, such that \begin{equation} \mathcal{F}(f) = \mathfrak{f}, \qquad \mathcal{F}(g) = \mathfrak{g}. \end{equation}

Does \begin{equation} \mathcal{F}(f \cdot g) = \mathfrak{f} \ast \mathfrak{g} \tag{1} \end{equation} hold?

I tried to use density of the Schwartz functions in the $L^p$ spaces, but did not manage to get very far. In summary my question is:

  1. Does $(1)$ hold?
  2. How would one proceed to prove $(1)$, in the case it holds?

1 Answer 1


Your identity is equivalent (since the Fourier transform is bijective on the space of tempered distributions) to $$ \mathcal{F}^{-1}(\mathfrak{f} \ast \mathfrak{g}) = f \cdot g. $$

But since $\mathfrak{f}, \mathfrak{g} \in L^1 ([-\sigma, \sigma])\subset L^1 (\Bbb{R})$, you have (by the usual convolution theorem for $L^1$, but applied to $\mathcal{F}^{-1}$ instead of $\mathcal{F}$ (the proof is the same)): $$ \mathcal{F}^{-1}(\mathfrak{f} \ast \mathfrak{g}) = \mathcal{F}^{-1}(\mathfrak{f}) \cdot \mathcal{F}^{-1}(\mathfrak{g}) = f \cdot g, $$ as desired.

  • $\begingroup$ Really nice, thank you!! $\endgroup$ Oct 24, 2016 at 8:27

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.