On polynomial symbols for subdivision schemes (Q2642362)

The author presents a method to define polynomial symbols associated to compactly supported refinable functions. Given a \(d\times d\) dilation matrix \(A\), a compatible compactly supported refinable function \(\varphi\), with mask a finite sequence \(a_k\subseteq \mathbb{Z}^d\), is any function in \(L_2(\mathbb{R}^d)\) such that \(\varphi(x)=| \det(A)| \sum_{k\in \mathbb{Z}^d} a_k\varphi(Ax-k)\). To any compatible refinable function associated to \(A\), it is associated a symbol \(M(\xi)=\sum_{k\in \mathbb{Z}^d}a_ke^{i\langle k,\xi\rangle}\). Conversely, if we have a symbol associated to a dilation matrix \(A\) satisfying certain conditions, then this symbol generates an interpolating refinable function. The article presents a way to compute a symbol that indeed is a trigonometrical polynomial. Such a polynomial symbol can be defined as the convolution product of a periodic function \(K\) and a trigonometric polynomial \(P\). The polynomial symbols that are defined as these convolution products never satisfy the condition \(M(0)=1\) which is necessary to ensure that \(M\) will generate a convergent interpolatory subdivision scheme. But this problem is solved by the author by a transformation that transforms the symbol \(M\) into a valid one. It is noticeable that this method is general. That is, given any trigonometric polynomial that might be valid as a symbol for \(A\), it can be obtained by the method presented in the paper. First, an alternative symbol is constructed by an appropriate convolution product. Then, this symbol is corrected to attain a value of 1 at the origin. The article ends showing how this approach can be adapted in some cases to compute symbols that satisfy higher Strang-Fix conditions.