If $\int_{\mathbb R^d}K=1$ then $K_{\delta}(x)=\delta^{-d}K(\delta^{-1}x)$ is an approximate identity

111 Views Asked by At

Let $K\in L^1(\mathbb R^d)$ be bounded and supported on a set $E$ of finite measure and let $\int_{\mathbb R^d}K=1.$ Define $K_{\delta}(x)=\delta^{-d}K(\delta^{-1}x)$ for every $\delta>0$. Show that $(K_{\delta})_{\delta}$ is an approximation of the identity, that is, it satisfies: $$\text{i) }\int_{\mathbb R^d}K_\delta(x)dx=1\quad\quad\text{ii) }\vert K_\delta(x)\vert\leq A\delta^{-d}\quad\quad\text{iii) }\vert K_\delta(x)\vert\leq A\delta\vert x\vert^{-(d+1)}$$ for all $\delta>0$ and $x\in\mathbb R^d$.

$\textbf{My Attempt & Concerns:}$ To verify (i), I use the dilation property of the Lebesgue integral in $\mathbb R^d,$ which can be found here How to prove Dilation property of Lebesgue integral. Mainly, we have $$\int_{\mathbb R^d}K_\delta(x)dx=\int_{\mathbb R^d}\delta^{-d}K(\delta^{-1}x)dx=\delta^{-d}\delta^{d}\int_{\mathbb R^d}K(x)dx=1.$$

To verify (ii), we use the hypothesis that $K$ is bounded and supported on a set of finite measure, so that if $\delta^{-1}x\notin E$ then we have that $\vert K_{\delta}(x)\vert=0,$ so it is surely bounded by the claimed quatity, with say $A$ being the bound for $K$. If $\delta^{-1}x\in E$ then we use that $\vert K\vert\leq A$ in $\mathbb R^d,$ to obtain that $$\vert K_{\delta}(x)\vert=\vert\delta^{-d}K(\delta^{-1}x)\vert\leq A\delta^{-d}.$$

The third property is the one I do not see a way to prove. If $\delta^{-1}x\notin E$ then it is fine, as $K$ is supported on $E$ so we certainly have the bound. However, if $\delta^{-1}x\in E$, then we have that $$\vert K_\delta(x)\vert=\delta^{-d}\vert K(\delta^{-1}x)\vert=\frac{\delta\vert K(\delta^{-1}x)\vert}{\delta^{d+1}}\leq\frac{\delta A}{\delta^{d+1}},$$ but I do not see a way to get rid of $\delta^{d+1}$ at the bottom and end up with $\vert x\vert^{d+1}.$


Any hints on how to proceed on the third part? Please, only hints. Any feedback on the other parts are also welcomed.

Thank you for time.

1

There are 1 best solutions below

0
On

Prove it for the case when $K$ is compactly supported, say $supp(K)\subset [-n,n]$ . In general case for any $\epsilon$ you can find $n$ big enough such that the integral of $K$ outside $[-n,n]$ is smaller than $\epsilon$, so $K$ is essentially supported on $[-n,n]$ choosing a good $\epsilon$ and big $n$ should give you the result.