1
$\begingroup$

I am reading an article on wavelet connection coefficients (G. Beylkin, "On the representation of operators in bases of compactly supported wavelets", 1992 (MSN)) and I came across Equation (3.31): \begin{equation} \sum_{l=-\infty}^\infty l^m\phi(x-l) = x^m + \sum_{l=1}^m (-1)^l \begin{pmatrix} m\\l \end{pmatrix} M_l^\phi x^{m-l} \end{equation} where $\phi(x)$ is the scaling function and \begin{equation} M_l^\phi = \int_{-\infty}^\infty x^l\phi(x)\,dx \end{equation} is the $l$-th momentum of $\phi$.

The author claims that the equation is well-known if $M_l^\phi = 0$ for $l=0,\dotsc,m$, and the general case follows from taking Fourier transforms. However, I could not find it, and trying to prove it myself is not working.

I recognize that both sides are kinds of convolutions, but when taking the Fourier transform the expressions (apparently) lead nowhere. Is there some trick I need to be aware of, or is it simply lack of practice/knowledge?


I do not know if this information should help, but the ultimate goal is to prove that \begin{equation} \sum_{l=-\infty}^\infty lr_l = -1 \end{equation} where \begin{equation} r_l = \int_{-\infty}^\infty \phi(x-l)\phi'(x)\,dx. \end{equation}


EDIT: Using the Poisson summation as @Nemo suggested in a comment, I was able to find that \begin{equation} \sum_{l=-\infty}^\infty l^m\phi(x-l) = \sum_{k=0}^m \sum_{l=-\infty}^\infty (-1)^k \begin{pmatrix} m\\k \end{pmatrix} e^{-ilx} i^k \frac{d^k\hat{\phi}}{d\xi^k}(-l) x^{m-k}. \end{equation}

Now, I know that $i^k\frac{d^k\hat{\phi}}{d\xi^k}(0) = M_l^\phi$ but I'm still stuck with the terms $i^k e^{-ilx} \frac{d^k\hat{\phi}}{d\xi^k}(-l)$ for $l\ne0$. Is there any identity I am not aware of?

$\endgroup$
  • 1
    $\begingroup$ mathworld.wolfram.com/PoissonSumFormula.html $\endgroup$ – user82588 Mar 11 at 21:44
  • $\begingroup$ I used it by defining $f(y) = y^m\phi(x-y)$ and almost found the result. However, instead of $M_\phi^l$, I got $i^k e^{i l x} \frac{d^k\hat{\phi}}{d\xi^k} (-l)$, which (I think) I cannot turn into $M_\phi^l$... $\endgroup$ – AspiringMathematician Mar 13 at 13:28
3
+50
$\begingroup$

The identity in the OP does not hold for any $m$, but only for $m< N$ where $N$ is the number of vanishing moments of the wavelet function. To complete the Poisson-summation derivation, one needs the socalled Strang-Fix condition, which says that $\frac{d^k\hat{\phi}}{d\xi^k}(-l)=0$ for integer $l$ unequal to 0 and $k< N$.

The identity says that integer shifts of the scaling function can reproduce polynomials of order $N$. For a proof, see theorems 4.26 and 4.27 in this book. (The identity is equivalent to the recursion relation 4.51.)

| cite | improve this answer | |
$\endgroup$
  • 1
    $\begingroup$ I couldn't have a look at it during the weekend, but this does look like what I was needing. I'll check the theorems and award the bounty later. $\endgroup$ – AspiringMathematician Mar 16 at 13:59

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.