In Wavelet theory, one constructs wavelet bases via translations a dialations of an $L^2$ function... Is it possible for some set of translations alone to form an Orthonormal Basis? That is:

  1. Does there exist $w \in L^2 (\mathbb{R})$ and $S$ a subset of real numbers so that the family of translates $\{w(t-s)\}_{s\in S}$ is an ONB for $ L^2$?

  2. If so, can we pick $ S $ to be only integers? If not, can we pick $ S $ to be only integers if we relax the requirement to only be a complete system rather than an ONB?

UPDATE: For the case $ S =\mathbb{R} $, we can see it is not complete by noticing the fourier transforms are all $\hat{w} $ times $1-$periodic functions... and the result is easy from there. Part 1 is still open!


ONB? Surely not, not that I have a proof. Otoh if $S$ is dense then there does exist $f$ such that the closed span of the translates of $f$ by elements of $S$ is all of $L^2$. I suspect that if $S$ is not dense there cannot exist such an $f$; see below for why I suspect this.

Say $\tau_x f(t)=f(t-x)$.

Say $f\in L^2$ satisfies $\hat f\ne0$ almost everywhere. If $g$ is orthogonal to $\tau_x f$ for every $x\in S$ then the convolution $f*h$ vanishes on $S$, where $h(t)=g(-t)$. Suppose $S$ is dense. Since $f*h$ is continuous it follows that $f*h=0$. Hence $\hat f\hat h=0$, hence $\hat h=0$, so $g=0$.

Now if $S$ is not dense there does exist $\phi\in L^1$ with $\phi\ne0$ such that $\hat\phi$ vanishes on $S$. This doesn't quite show that $\tau_x f$ for $x\in S$ cannot span $L^2$, but it seems like evidence to that effect; saying those translates of $f$ do not span $L^2$ is the same as saying that we can take $\phi=\hat f\hat g$ for some $g\in L^2$.

  • $\begingroup$ Thanks for your input! I can prove that a (almost everywhere) nonzero Fourier transform is necessary and sufficient for completeness in the case that $S=\mathbb{R}$. I also suspect the answer to be no, mostly because this is a fairly strict condition, and changing $S=\mathbb{R}$ to $S=\mathbb{Z}$ is a huge leap. To proceed by contradiction, we know the nonzero Fourier tranform is still necessary... but I'm not sure where to go from there. $\endgroup$ – Spanky Dec 9 '15 at 19:06

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.