Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Let $f \in C^1_c(\Omega)$ where $\Omega \subset \mathbb{R}^d$ is a bounded domain. Let $\phi \in C^1_c(\mathbb{R}^d)$ be an approximation of the identity (i.e. $\int_{\mathbb{R}^d} \phi=1$, $\phi \geq 0$, $\phi_\epsilon := \frac{1}{\epsilon^d} \phi(\frac{x}{\epsilon})$.

How would you prove that

$$\int_\Omega |f(x) - f \ast \phi_\epsilon(x)| dx \leq \epsilon \int_\Omega |\nabla f| dx?$$

I'm trying to show that the family of $C^1_c$ functions convolved with a mollifier is uniformly close to the function in $L^1$ (which would be true after having this result if we assume something like the family of functions being bounded in $W^{1,1}(\Omega)$).

share|improve this question
    
Does Taylor's theorem (en.wikipedia.org/wiki/Taylor%27s_theorem) for the function $f$ around $x$ in the convolution help? –  Fabian Mar 9 '11 at 18:11
    
Well, you clearly have to be able to bound $|f-f*\phi_\epsilon|$ in terms of the derivatives of $f$. When $\epsilon$ is small, $|f-f*\phi_\epsilon|$ is a difference between the value of $f$ and the values nearby it averaged according to $\phi$... –  Mariano Suárez-Alvarez Mar 9 '11 at 18:32
add comment

1 Answer

up vote 7 down vote accepted

WLOG assume $\phi(x)$ supported in the unit ball (it has compact support, so it is supported in some ball). First look at

$$ |f(x) - f*\phi_\epsilon(x)| \leq \int_{|z|\leq 1} \phi(z) |f(x) - f(x-\epsilon z)| dz $$

replace the integrand

$$ f(x) - f(x-\epsilon z) = - \int_0^{\epsilon|z|} D_rf(x - r\omega) dr $$

with $\omega = z / |z|$ and using the fundamental theorem of calculus. So

Integrating the whole thing over $x$, and changing the order of integration, you have

$$ \int_{\Omega}|f(x) - f*\phi_\epsilon(x)|dx \leq \int_{|z|\leq 1} \phi(z) \int_0^{\epsilon|z|} \int_{\Omega} |D_rf(x - r\omega)| dx~ dr~ dz $$

The inside most integral for fixed $r\omega$ gives you $\int_\Omega |\nabla f| dx$. The integral over $r$ gives you the factor of $\epsilon$. And integrating $\phi(z)$ over the ball of radius 1 gives you 1.

share|improve this answer
    
An alternative approach, if you know about difference quotients, is to note $\int_\Omega|f(x) - f*\phi_\epsilon(x)|dx \leq \sup_{|z| < \epsilon} \int_\Omega |f(x - z) - f(x) | dx$ The integral on the right is bounded by $|z|$ times the $L^1$ norm of the $|z|$-difference quotient, which the standard theory of difference quotients tells you is bounded by the homogeneous ${W}^{1,1}$ norm. –  Willie Wong Mar 9 '11 at 18:45
    
@Willie: Don't you mean $f$ instead of $u$ in the first formula? Cool stuff nevertheless. –  Jonas Teuwen Mar 9 '11 at 19:40
    
@Jonas: thanks, corrected. –  Willie Wong Mar 9 '11 at 22:43
    
@Willie: Thanks a lot! Do you have any suggestions on where to read up on the "standard theory of difference quotients" (including the result about the $W^{1,1}$ norm that you mentioned)? Also, regarding the initial assumptions, we can always choose $\epsilon$ small enough to be supported in the unit ball eventually anyway right? If support is outside of the unit ball, is the bound of $\epsilon$ still guaranteed? –  user1736 Mar 10 '11 at 0:57
    
@user1736: If the support of $\phi$ is outside of the unit ball, then directly following the above argument you will lose a constant; that is, the RHS instead of being $\epsilon \|\nabla f\|_{1}$ will be $R\epsilon \|\nabla f\|_{1}$ where $R$ is the radius of the support set of $\phi$. Which as long as you fix your mollifier function is insignificant in the larger setting. –  Willie Wong Mar 10 '11 at 12:14
show 2 more comments

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.