1

A sequence of mollifiers $(\rho_n)_{n \geq 1}$ is any sequence of functions on $\mathbb{R}^d$ such that $$ \rho_n \in C_c^{\infty} (\mathbb{R}^d), \quad \operatorname{supp} \rho_n \subset \overline{B(0, r_n)}, \quad \sup_n \|\rho_n\|_1 < \infty, $$ where $r_n \to 0^+$ and $B(x, r)$ denotes an open ball centered at $x$ with radius $r$.

In order to prove a result mentioned in this thread, I need to prove below approximation, i.e.,

Theorem 1 Let $f \in \mathcal C (\mathbb R^d)$. Then $\rho_n * f \xrightarrow{n \to \infty} f$ uniformly on compact sets.

Theorem 2 Let $f \in L^p (\mathbb R^d)$ with $p \in [1, \infty)$. Then $\rho_n * f \xrightarrow{n \to \infty} f$ in $L^p$.

Could you have a check on my below attempt?


Proof of Theorem 1 WLOG, we assume $\sup_n r_n \le 1$. Fix a compact subset $K$ of $\mathbb{R}^d$. Fix $\varepsilon>0$. Then $f$ is uniformly continuous on $K' :=K - \overline{B(0, 1)}$. There is $\delta>0$ such that if $x,y \in K'$ with $|x-y| < \delta$ then $|f(x)-f(y) |< \varepsilon$. Let $\alpha := \sup_n \|\rho_n\|_1 < \infty$. If $x \in K$ and $r_n < \delta$ then $$ \begin{align} |\rho_n * f (x)-f(x)| &\le \int_{\overline{B(0, r_n)}} |f(x-y)-f(x)| |\rho_n(y)| \, \mathrm d y \\ &\le \varepsilon \int_{\overline{B(0, r_n)}} |\rho_n(y)| \, \mathrm d y \\ &\le \varepsilon \alpha. \end{align} $$

The claim then follows by taking the limit $\varepsilon \downarrow 0$. QED.

Proof of Theorem 2 We have $\mathcal C_c (\mathbb R^d)$ is dense in $L^p (\mathbb R^d)$. Fix $\varepsilon>0$. There is $\hat f \in \mathcal C_c (\mathbb R^d)$ such that $\|f-\hat f\|_p < \varepsilon$. Then $\hat f * \rho_n \in \mathcal C_c (\mathbb R^d)$ for all $n$. WLOG, we assume $\sup_n r_n \le 1$. Then $$ \operatorname{supp} (\hat f * \rho_n) \subset \overline{\operatorname{supp} \hat f + \operatorname{supp} \rho_n} \subset K:=\operatorname{supp} \hat f + \overline{B(0, 1)}. $$

Clearly, $K$ is compact. By Theorem 1, $\hat f * \rho_n \xrightarrow{n \to \infty} \hat f$ uniformly. By triangle inequality, $$ \begin{align} \| \rho_n * f - f\|_p &\le \| \rho_n * (f - \hat f)\|_p + \| \rho_n * \hat f - \hat f\|_p + \|f-\hat f\|_p. \end{align} $$

Let $\alpha := \sup_n \|\rho_n\|_1 < \infty$. By Young's inequality, $$ \| \rho_n * (f - \hat f)\|_p \le \| \rho_n \|_1 \| f - \hat f\|_p \le \alpha \varepsilon. $$

The claim then follows from taking the limit $n \uparrow \infty$ and then $\varepsilon \downarrow 0$. QED.


Update As mentioned in comments, the condition $\int \rho_n =1$ for all $n$ is necessary.

Analyst
  • 6,351
  • The conditions $\rho\ge0$ and $\int_{\mathbb R^d} \rho=1$ are missing. The results are false as stated (i.e., if $\rho$ satisfies the conditions, then $-\rho$ as well). – daw Apr 11 '23 at 16:59
  • How is this true? $$|\rho_n * f (x)-f(x)| \le \int_{\overline{B(0, r_n)}} |f(x-y)-f(x)| |\rho_n(y) , \mathrm d y$$ – Brian Moehring Apr 11 '23 at 17:06
  • @BrianMoehring You are right! I will check it out. – Analyst Apr 11 '23 at 17:07
  • @daw from Brian's comment, I got that the condition $\int \rho = 1$ is essential. I could not see how $\rho \ge 0$ is important. Could you elaborate more? – Analyst Apr 11 '23 at 17:11
  • @daw $\rho\geq 0$ (and radial) is nice in applications but not necessary for the theorem. Also $c:=\int \rho=1$ is not necessary as long as we modify the conclusion to the convolutions converging to $cf$ rather than just $f$. – peek-a-boo Apr 11 '23 at 17:22
  • 1
    For theorem 2, you can prove it directly using Minkowski's inequality for integrals, no need to approximate and invoke theorem 1. – peek-a-boo Apr 11 '23 at 17:27
  • maybe the condition $\int_{\mathbb R^d } \rho_n \ dx \to 1$ is enough – daw Apr 12 '23 at 06:38
  • @peek-a-boo It seems to me we need at least $\sup_n |\rho_n|_1 < \infty$ or $\rho_n \ge 0$ to bound the term $\int |\rho_n|$. Please have a look at my question here. – Analyst Apr 12 '23 at 07:11
  • @daw Besides the essential condition $\int \rho_n \ dx \to 1$, It seems to me we need at least $\sup_n |\rho_n|_1 < \infty$ or $\rho_n \ge 0$ to bound the term $\int |\rho_n|$. Please have a look at my question here. – Analyst Apr 12 '23 at 07:15
  • 2
    here’s what I don’t understand. (Almost) every question you’ve asked can be found in a textbook. Of course textbooks occasionally skip a few steps here and there, and may sometimes state things with slightly varying levels of generality (otherwise they’d be thousands of pages long), but I mean is there something specific in the textbooks which is unclear or..? My honest advice is you need to be able to judge for yourself whether or not your proof is right, especially when 90-95% of the proof is already given in the books. Everything I can tell you is just stuff from Brezis/Folland/Rudin… – peek-a-boo Apr 12 '23 at 07:18

0 Answers0