Considering the definition of convolution:

If $T,S \in \mathcal{D'(\mathbb{R})} $ and at least one of them has compact support, $\forall \varphi \in \mathcal{D}(\mathbb{R})$ $$\langle T*S, \varphi\rangle= \langle T(y), \langle S(x), \varphi(x+y)\rangle \rangle $$

and the translation property: $$\langle T(y-a), \varphi\rangle = \langle T(y),\varphi(y+a) \rangle $$

I am trying to prove the following property:

$$(T * S )(x-a)=T * S (x-a)= T (x-a)* S $$ where $S, T$ are distributions: $( S,T \in \mathcal{D'}(\mathbb{R}))$

My try: I take a test function $\varphi \in \mathcal{D}$

Using the property of translation:

$$\langle (T*S)(y-a), \varphi\rangle =\langle (T*S)(y), \varphi(y+a) \rangle .....(1) $$ calling $\varphi(y+a)=\psi(y)$:

$$=\langle (T*S)(y), \psi(y) \rangle$$

and by definition of convolution of distributions:

$$= \langle T(y), \langle S(x), \psi(x+y)\rangle \rangle $$ $$= \langle T(y), \langle S(x), \varphi(x+y+a)\rangle \rangle $$ using translation again: $$= \langle T(y), \langle S(x-a), \varphi(x+y)\rangle \rangle .....(2)$$

and now I wish I could call this $T * S (x-a) $, but for that I guess I would instead need to have $= \langle T(y), \langle S(x-a), \varphi(x+y-a)\rangle \rangle $, because according to the definition of convolution, I need $\varphi$ to have as argument the sum of the arguments of the $T$ and $S$.

Is this correct? I have the following doubts :

(1) I am not very sure if $(T*S)(y-a)$ means $T(y-a)*S(y-a)$ or if the $(y-a)$ applies just to S , so that $(T*S)(y-a)= T(x)*S(y-a)$. I used the second one, because according to the definition of convolution, T and S take different variables: $$\langle T*S, \varphi\rangle= \langle T(y), \langle S(x), \psi(x+y)\rangle \rangle $$

I am also starting with a $y$ variable, because otherwise I don't get the $x-a$ in the final step The fact that using variables with T and S is an abuse of notation also is making it confusing, because I don't know if I am abusing correctly

(2) As I mentioned at the end of the proof can I really conclude :$T * S (x-a) $ ? For that I guess I would instead need to have $= \langle T(y), \langle S(x-a), \varphi(x+y-a)\rangle \rangle $


1 Answer 1


Define the operator $\tau_a$ on $\mathcal{D}(\mathbb R)$ by $(\tau_a \varphi)(x) = \varphi(x-a)$ and on $\mathcal{D}'(\mathbb R)$ by $\langle \tau_a T, \varphi \rangle := \langle T, \tau_{-a}\varphi \rangle$.

We can then write $$ \langle \tau_a(T*S), \varphi \rangle = \langle T*S, \tau_{-a}\varphi \rangle = \langle T(t), \langle S(s), (\tau_{-a}\varphi)(t+s) \rangle \rangle = \langle T(t), \langle S, \tau_{-a-t}\varphi \rangle \rangle = \langle T(t), \langle \tau_a S, \tau_{-t}\varphi \rangle \rangle = \langle T(t), \langle (\tau_a S)(s), \varphi(s+t) \rangle \rangle = \langle T*\tau_a S, \varphi \rangle, $$ i.e. $\tau_a(T*S) = T*\tau_a S.$

  • $\begingroup$ Thanks, very clear, I am trying to translate back with the notation they used in my lectures, Before you applied the definition of convolution, how should $T*S$ be understood :$ T(t)*S(t),$ $T(t)*S(s) $ or $T(s)*S(s)$ ? $\endgroup$
    – mathlover
    Sep 1, 2020 at 19:23
  • $\begingroup$ $T*S$ is a function of only one variable, so possibly $T(t)*S(t)$ or $T(s)*S(s),$ not $T(t)*S(s).$ The best certainly is to write $(T*S)(x)$ without setting variables on the individual factors. $\endgroup$
    – md2perpe
    Sep 1, 2020 at 19:53

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.