I want to prove that convolution of a function with itself smooths the function in some sense.

Given a function $f:\mathbb{R}^n \to \mathbb{R}$, the roughness $R = \int \Delta f $ where we integrate the Laplacian over all values of $f$.

Can we show that $R(f * f) < R(f)$ where $*$ is convolution. If not, is there another measure of smoothness we could use.


I'm not sure the "roughness" you define is a good measure of smoothness. For instance, if $n=1$, and $f$ is twice differentiable, then $$\int_{\mathbb R} \Delta f=\lim_{+\infty}f^\prime-\lim_{-\infty}f^\prime$$ when those limits exist. It only depends on the behavior of $f^\prime$ at the infinities.

Instead, you can look at something like a Sobolev norm $$\|f\|_S^2=\int |f|^2 + \int |\Delta f|^2$$ You see, if $f$ is not super smooth, then the Laplacian will have a lot of energy, and the norm will be high. So small Sobolev norm means smooth function.

Take $f \in L^1(\mathbb R^n)\cap L^2(\mathbb R^n)$. Then $f$ has a Fourier transform. Thanks to Parseval's theorem, the Sobolev norm can be expressed in the Fourier domain: $$\|f\|_S^2=\int_{\mathbb R^n}(1+\|\omega\|^2)|\hat{f}(\omega)|^2d\omega$$ Also, remember that the Fourier transform maps convolutions to products, so $$\|f*f\|_S^2=\int_{\mathbb R^n}(1+\|\omega\|^2)|\hat{f}(\omega)|^4d\omega$$ Now because $f$ is integrable, its Fourier transform is bounded by its $L^1$ norm $\|f\|_1$ (easy for you to check). So $$\|f*f\|_S^2\leq\|f\|_{1}^2\int_{\mathbb R^n}(1+\|\omega\|^2)|\hat{f}(\omega)|^2d\omega$$ In other words $$\|f*f\|_S\leq \|f\|_1\cdot\|f\|_S$$ Here $\|f\|_1$ acts as a normalization factor. That's necessary because $\|f*f\|_S$ is homogeneous of degree 2, while $\|f\|_S$ is of degree 1. So to make it easier to see the effect of the convolution, just focus on functions that are normalized such that $\|f\|_1=1$, then $$\boxed{\|f*f\|_S\leq \|f\|_S}$$ Convolving decreases the Sobolev norm, that is, the resulting function is smoother.

| cite | improve this answer | |
  • 1
    $\begingroup$ Thank you. The sobolev norm seems to work for what I am doing. $\endgroup$ – Halbort Jul 30 '19 at 13:52
  • $\begingroup$ Does the triangle inequality hold for the Sobolev norm. $\endgroup$ – Halbort Jul 30 '19 at 14:02
  • $\begingroup$ Yes, it is a true norm. $\endgroup$ – Stefan Lafon Jul 30 '19 at 14:21
  • $\begingroup$ You should try to prove it. It's a good exercise. $\endgroup$ – Stefan Lafon Jul 30 '19 at 14:22

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.