3

Problem

Given a mollifier: $\varphi\in\mathcal{L}(\mathbb{R})$

Then it acts as an approximate identity: $$f\in\mathcal{C}(\mathbb{R}):\quad\int_{-\infty}^\infty n\varphi(nx)f(x)dx\to f(0)\cdot\int_{-\infty}^\infty\varphi(x)dx$$ How to prove this under reasonable assumptions?

Example

As an example regard the Gaussian: $$f\in\mathcal{C}(\mathbb{R}):\quad\frac{n}{\sqrt{\pi}}\int_{-\infty}^\infty e^{-(nx)^2}f(x)\mathrm{d}x\to f(0)$$ (This is a useful technique when studying operator semigroups.)

freishahiri
  • 17,045

2 Answers2

1

Now, I got a proof for the case...

Theorem

Given a mollifier: $\varphi\in\mathcal{L}(\mathbb{R})$

For bounded functions one has: $$f\in\mathcal{C}(\mathbb{R}):\quad\int_{-\infty}^{\infty}n\varphi(nx)f(x)\mathrm{d}x\to f(0)\cdot\int_{-\infty}^{\infty}\varphi(x)\mathrm{d}x\quad(\|f\|_\infty<\infty)$$ (In fact, this result extends to the Bochner integral!)

Proof

Split the integral into outer vanishing parts and the inner convergent part: $$\left|\int_{-\infty}^{\infty}\varphi(\hat{x})f(\tfrac{1}{n}\hat{x})\mathrm{d}\hat{x}-f(0)\cdot\int_{-\infty}^{\infty}\varphi(\hat{x})\mathrm{d}\hat{x}\right|\\\leq\int_{-\infty}^{-R}|\varphi(\hat{x})|\cdot2\|f\|_\infty\mathrm{d}\hat{x}+\int_{-R}^{R}\|\varphi\|_\infty\cdot\left|f\left(\tfrac{1}{n}\hat{x}\right)-f(0)\right|\mathrm{d}\hat{x}+\int_{R}^{\infty}|\varphi(\hat{x})|\cdot2\|f\|_\infty\mathrm{d}\hat{x}\\\leq2\|f\|_\infty\delta_R+\left(\int_{-\infty}^{\infty}\varphi(x)\mathrm{d}x\right)\delta_N+2\|f\|_\infty\delta_R=\varepsilon\quad(n\geq N_R(\varepsilon))$$ That proves the assertion.

Outview

Does it apply to unbounded functions too: $$f\in\mathcal{C}(\mathbb{R}):\quad\int_{-\infty}^{\infty}n\varphi(nx)f(x)\mathrm{d}x\to f(0)\cdot\int_{-\infty}^{\infty}\varphi(x)\mathrm{d}x$$ (Intuition and examples suggest it does.)

freishahiri
  • 17,045
0

You can always approximate $j(x)$ by bounded continuous functions, for which the result holds by dominated convergence and the usual Riemann substitution rule.

Alex R.
  • 33,289
  • Can you explain that more precisely, please? – freishahiri Nov 04 '14 at 16:51
  • Can you prove the result when $j(x)$ is a bounded, continuous, integrable function? – Alex R. Nov 04 '14 at 16:51
  • I guess I can prove it for simple functions... – freishahiri Nov 04 '14 at 16:52
  • No not really, I'm missing the case when the characteristic function has unbounded support. – freishahiri Nov 04 '14 at 16:54
  • 1
    Also it requires interchanging limits - not only with an integral... – freishahiri Nov 04 '14 at 16:56
  • You can always split the function into a bounded part plus a part where it's unbounded but has finite measure (it's integrable afterall). – Alex R. Nov 04 '14 at 16:58
  • Hmm I'm not getting it... – freishahiri Nov 04 '14 at 17:07
  • I freed the problem from assumptions and set the question from hint-required to proo-required. You can leave your answer here anyway. I won't bother, just wanted to let you know. Thanks anyway for your help!! :) – freishahiri Nov 05 '14 at 12:57
  • 1
    @AlexR. That won't necessarily work. Take $j=1/\sqrt{x}=f$ on $(0,1)$. – abnry Dec 07 '14 at 23:26
  • The crucial fact is that $j \geq 0$ is positive. I would take a positive, increasing sequence of simple functions approximating $j$ and then pass the limit through with the monotone convergence theorem. – abnry Dec 07 '14 at 23:27
  • @AlexR. Sorry, you can get things to work with the MCT so I guess I'm just adding you really need $j \geq 0$. – abnry Dec 07 '14 at 23:28
  • @nayrb: Can you explain how, please? I mean don't you miss monotony here? (Don't wonder I may free the setting but you can of course assume positivity.) – freishahiri Jan 09 '15 at 07:27