A distribution is an element of the continuous dual space of some function space. Let us take the Schwartz space $\mathcal{S} := \mathcal{S}(\mathbb{R}^n)$ just as an example. A distribution $\phi \in \mathcal{S}'$ is then a map $$ \phi: \mathcal{S} \rightarrow \mathbb{C}.$$

My question is this: how do I interpret $\phi(x)$? I see this written a lot, but I don't understand how to work with it. What does for example $\phi(x) = \phi(-x)$ mean? The only thing I can think of is that $\phi(f) = \phi(\hat{f})$, where $\hat{f}(x) = f(-x)$.

And more specifically for the problem I'm working on: I have a distribution $$ \mathcal{W} : \mathcal{S}(\underbrace{\mathbb{R}^4 \times \dots \times \mathbb{R}^4}_{n \text{ times}}) \rightarrow \mathbb{C} $$

and then they say that $\mathcal{W}$ is translation invariant, i.e. $$\mathcal{W}(x_1 +a,\dots, x_n + a) = \mathcal{W}(x_1,\dots,x_n)$$ so it can be writthen as a distribution $\mathfrak{W}$ that only depends on the differences $x_1-x_2,\dots,x_{n-1} - x_n$: $$\mathcal{W}(x_1,\dots,x_n) = \mathfrak{W}(x_1-x_2,\dots,x_{n-1}-x_n).$$

How do I interpret this last line?


When you have a $C^{\infty}$ map $f : \mathbb{R}^n \to \mathbb{R}^m$ which is proper (it is automaticaly the case with a $C^{\infty}$ bijection like translations), you can define the pushforward of a distribution by $f$. If $u$ is a distribution on $\mathbb{R}^n$ then, $f_{!}u$ is a distribution on $\mathbb{R}^m$ defined by $f_!u(\varphi)=u(\varphi \circ f).$

So for example the (dangerous) notation $u(x) = u(-x)$ means that $f_!u = u$ with $f : \mathbb{R}^n \to \mathbb{R}^n, x \mapsto -x.$

The same occurs when considering your translation by $a$. Let us call it $\mu_a.$ The condition $$\mathcal{W}(x_1 +a,\dots, x_n + a) = \mathcal{W}(x_1,\dots,x_n)$$ means that $\mathcal{W}=\mu_{a!}\mathcal{W}$

Now define the map $g : (\mathbb{R}^4)^n \to (\mathbb{R}^4)^{n-1}, (x_1, ... ,x_n) \mapsto (x_1-x_2, ... , x_{n-1}-x_n).$ The last equality means that the condition $\mathcal{W}=\mu_{a!}\mathcal{W}$ implies that $\mathcal{W}=g^*\mathfrak{W}$ for a certain distribution $\mathfrak{W}$ on $(\mathbb{R}^4)^{n-1}$.

  • $\begingroup$ Oh, I was just trying to answer my own question with what you told me! I have that, with your map $g$, $\mathfrak{W} = g_! \mathcal{W}$, is that correct? I don't know what you mean by $g^* \mathfrak{W}$. $\endgroup$ – user353840 Jun 1 '18 at 10:08
  • $\begingroup$ You can always define the pullback of a distribution by a submersion $g$. (By duality, with integration over the fibers of $g$) Here $g$ is a submersion so the pullback has a meaning. If you want, if will edit my post in details. $\endgroup$ – C. Dubussy Jun 1 '18 at 10:54
  • $\begingroup$ I have no idea what submersion and fibers are, so if you could post details, it would be much appreciated, but I don't want to take too much of your time. :) $\endgroup$ – user353840 Jun 1 '18 at 10:56

The trouble is that given a map $f : X \rightarrow Y$, the notation $\phi(f(x))$ should really mean "$\phi$ pulled back across $f$." But of course, distributions don't pull back in a natural way. Therefore I think the notation $\phi(f(x))$ only makes sense if $f$ is an invertible function, in which case it really means the $(f^{-1})_*(\phi),$ that is, the pushforward of $\phi$ across the inverse of $f$.

  • $\begingroup$ Is that push forward defined as in the other answer? $\endgroup$ – user353840 Jun 1 '18 at 10:29
  • $\begingroup$ @user353840, probably. This question defines it. $\endgroup$ – goblin Jun 1 '18 at 10:36
  • $\begingroup$ @user353840, ah yes, we're defining pushforwards the same. But I disagree with that answer. $\phi(x-1)$, for example, should move $\phi$ to the right one unit. To do that, we take its pushforward across $x \mapsto x+1$. I'm pretty sure this contradicts what C. Dubbusy wrote. $\endgroup$ – goblin Jun 1 '18 at 10:39
  • $\begingroup$ How would you then do the second part? Where we use the function $g$ as in C. Dubussy's answer? Because that's not a bijection, so it doesn't have an inverse. $\endgroup$ – user353840 Jun 1 '18 at 10:49
  • $\begingroup$ Goblin is right, I should have considered the translation $\mu_{-a}$. But notice that the condition $\mathcal{W}=\mu_{a!}\mathcal{W}$ is the same as $\mathcal{W}=\mu_{-a!}\mathcal{W}$ so actually what I wrote is also correct. $\endgroup$ – C. Dubussy Jun 1 '18 at 10:59

If the motivation is Wightman functions, it's better to use the differences $x_1-x_n,\ldots,x_{n-1}-x_n$ with a single point of reference $x_n$ rather than the consecutive differences $x_1-x_2, \ldots,x_{n-1}-x_n$. So let me consider the statement that as elements in $S'(\mathbb{R}^{4n})$ $$ \mathcal{W}(x_1,\ldots,x_n)=V(x_1-x_n,\ldots,x_{n-1}-x_n) $$ for some $V$ in $S'(\mathbb{R}^{4(n-1)})$.

First do some nonrigorous calculations as follows. Let $f(x_1,\ldots,x_n)$ be a test function in $S'(\mathbb{R}^{4n})$.

$$ \int_{\mathbb{R}^{4n}} V(x_1-x_n,\ldots,x_{n-1}-x_n)f(x_1,\ldots,x_n) \ dx_1\cdots dx_n $$ $$ =\int_{\mathbb{R}^{4n}} V(z_1,\ldots,z_{n-1})f(z_1+x_n,\ldots,z_{n-1}+x_n,x_n) \ dz_1\cdots dz_{n-1}dx_n $$ by the change of variables $z_i=x_i-x_n$ for $1\le i\le n-1$. By Fubini the last integral becomes $$ \int_{\mathbb{R}^{4(n-1)}} V(z_1,\ldots,z_{n-1})g(z_1,\ldots,z_{n-1}) \ dz_1\cdots dz_{n-1} $$ where $g=\Gamma (f)$ is defined by $$ g(z_1,\ldots,z_{n-1})=\int_{\mathbb{R}^4} f(z_1+x_n,\ldots,z_{n-1}+x_n,x_n) \ dx_n\ . $$ Now you can make rigorous sense of your RHS as the composition $V\circ \Gamma$ of the distribution $V:S(\mathbb{R}^{4(n-1)})\longrightarrow\mathbb{C}$ with the continuous linear map $$ \Gamma:S(\mathbb{R}^{4n})\longrightarrow S(\mathbb{R}^{4(n-1)})\ . $$


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.