Suppose $F_1$ and $F_2$ are given distributions with $F_2$ having compact support , then we define the convolution $F_1*F_2$ as the distributions $(F_1*F_2)(\varphi)=F_1(F_2^{a}*\varphi)$ where $F_2^a(\varphi)=F_2(\varphi^a)$ and $\varphi^a(x)=\varphi(-x)$ ...... Then we can show that $F_1*F_2$ is also a distribution and if $F_1$ is compact , we have $F_1*F_2(\varphi)=F_2*F_1(\varphi)$

The discussion above was in Stein's functional analysis Page$_{105}$ and the author state that to show $F_1*F_2$ is a distribution is quite straightforward so is left to the reader but I can not show it .

My attempt :
Since $F_2$ and $\varphi$ has compact support , then so is $(F_2^a*\varphi)(x)=F_2(\varphi(x+y))$ and it belongs to $C_0^{\infty}$. We can see $F_1*F_2$ is well defined for every $\varphi \in C_0^{\infty}$ .
To show $F_1*F_2$ is a distribution , we need to show whenever $\varphi_n \to \varphi $ in $D$ , we have $$\lim_{n\to \infty}(F_1*F_2)(\varphi_n)=(F_1*F_2)(\varphi)$$ note that $F_1*F_2(\varphi)=F_1(F_2(\varphi(x+y)))$ , it suffice to show $F_2(\varphi_n(x+y))\to F_2(\varphi(x+y))$
in $D$ . For each $x_0$ , we have $\varphi_n(x_0+y)\to \varphi(x_0 +y)$ , so $$\lim_{n\to \infty}LHS=RHS$$ for each $x$ , however , I can not show this limitation is also uniform convergence , nor does its derivatives.

My question:
$(1)$ I need some hint to show $F_2(\varphi_n(x+y)) $ converges to $F_2(\varphi(x+y))$ uniformly .
$(2)$ If $F$ is a distribution with respect to $f$ , I mean $F(\varphi)=\int f\varphi \, d\mu$ , then by fubini theorem I can show $F_1*F_2=F_2*F_1$ , but how to do this in the general case ?

  • 1
    $\begingroup$ (1) If $\varphi_n \to \varphi$ in $C_0^\infty$ then $\varphi_n(x+y) \to \varphi(x+y)$ in $C_0^\infty$. $\endgroup$ – md2perpe Apr 17 '19 at 21:16
  • $\begingroup$ For each $x$ , we have $\varphi_n(x+y)\to \varphi(x+y)$ , but can we prove $F(\varphi_n(x+y))\to F(\varphi(x+y))$ ? $\endgroup$ – J.Guo Apr 17 '19 at 22:48
  • $\begingroup$ What we here write as $\varphi_n(x+y)$ and $\varphi(x+y)$ are just $\varphi_n$ and $\varphi$ translated. You seem to already accept that $\varphi_n(x+y) \to \varphi(x+y)$ in $C_0^\infty$, so since $F$ is a distribution you do have $F(\varphi_n(x+y)) \to F(\varphi(x+y))$ by definition of a distribution. What you should show is that $F(\varphi(x+y))$ is a $C_c^\infty$ function. $\endgroup$ – md2perpe Apr 18 '19 at 4:29
  • $\begingroup$ @ md2perpe In Stein's book , we say $\varphi \to \varphi_n$ whenever $\varphi$ is a complexed valued function on$R^d$ with just one variable $x\in R^d$ . So we can not say $\varphi_n(x+y) \to \varphi(x+y)$ . If we define $g_n(x)=F(\varphi_n (x+y))$ and $g(x)=F(\varphi(x+y))$ and we need to prove $g_n(x)\to g(x)$ in $D$ . $\endgroup$ – J.Guo Apr 18 '19 at 11:11
  • $\begingroup$ My last comment was an answer to your question in the comment before: "but can we prove $F(\varphi_n(x+y))\to F(\varphi(x+y))$"? $\endgroup$ – md2perpe Apr 18 '19 at 12:14

By the help of @md2perpe , I find that we can prove a lemma for this problem first .

Let $F$ denote a distribution on $C_c^{\infty}$ , then we can find some positive integer $N$ such that $$|F(\varphi)| \leq C \|\varphi\|_N = C \sup_{x \in K ,{|\alpha|<N}} |\partial^\alpha \varphi(x)|$$
Indeed , assume otherwise . Then for each $n$ we can find $\varphi_n$ with $\|\varphi_n\|_n=1$ , while $|F(\varphi_n)|\ge n$ . Then let $\phi_n=\frac{\varphi_n}{n^{\frac12} }$ , we can see $\phi_n \to 0$ in $C_c^{\infty}$ while $F(\phi_n)\to \infty$ , contradicting that continuity of $F$ .

With the lemma above , we can see $F(\varphi_n(x+y))\to F(\varphi(x+y))$ in $C_c^{\infty}$ since $\varphi_n \to \varphi$ , so we complete the proof of the first problem .

For the second problem , Let $\phi \in C_c^{\infty}$ with $\int \, \phi(x) \, dx=1$ . Then we note that $\phi_n(x)=n^d\phi(nx)$ is an approximation to the identity . Let $f_n(x)=\phi * F(x)$ , we can see that $$\lim_{n\to \infty} \int f_n(x)\varphi(x) \,dx=F(\varphi(x))$$ For every $\varphi \in C_c^{\infty}$ and we can write $f_n \to F$. The proof of this part was in Stein's functional analysis Page$_{103}$ Corollary $1.2$ .

Now suppose $F$ and $G$ are two compact distributions with $f_n \to F$ and $g_m \to G$ , Then we have $$F(G(\varphi(x+y)))=F(G(\varphi(x+y)))-\int f_n(x)G(\varphi(x+y)) \,dx+\int f_n(x)G(\varphi(x+y)) \,dx$$ and $$\int f_n(x)G(\varphi(x+y)) \,dx=\int f_n(x)(G(\varphi(x+y))-\int g_m(y)\varphi(x+y)+\int g_m(y)\varphi(x+y) \, dy) \,dx$$ Note that for suffice large $N$ , $\int f_n(x) \, dx$ is bounded . Indeed $\{f_n \}$ are supported in some compact set which is contained in an open set $O$ , so we can find a function $\varphi_0 \in C_c^{\infty}$ , when $x\in O$ we have $\varphi_0(x)=1$ . Then we have $$\lim_{n\to \infty} \int f_n(x)\,dx=\lim_{n\to \infty} \int f_n(x)\varphi_0(x) \,dx=F(\varphi_0(x))=A$$ Next ,Since $$|G(\varphi(x+y))-\int g_m(y)\varphi(x+y)|=|G(\varphi(x+y)-\phi_m*\varphi(x+y))|$$
By the lemma above and a limit argument we can get the desired conclusion since $$\int\int f_n(x)g_m(y) \varphi(x+y) \,dx \,dy=\int\int f_n(x)g_m(y) \varphi(x+y) \,dy \,dx$$

  • $\begingroup$ You had missed $| \bullet |$ around $F(\varphi)$ so I added it. I also replaced || with \| that looks a little bit better. $\endgroup$ – md2perpe May 4 '19 at 21:37
  • $\begingroup$ The inequality is sometimes useful. Many authors define distributions to be linear functionals that satisfy the equality instead of defining them to be continuous w.r.t. limit of test functions. $\endgroup$ – md2perpe May 4 '19 at 21:39

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.