Banded matrices with banded inverses. II: Locally finite decomposition of spline spaces (Q1801490): Difference between revisions

[For part I see J. Comput. Appl. Math. 41, No. 3, 281--300 (1992; Zbl 0757.15012).] The paper contains the following results which are interesting especially in connection with multiscale methods. Given a two-slanted matrix \(A:=(a_{ij})_{i,j\in Z}\), i.e., \(a_{ik}=0\) if \(k\notin\{2i,\ldots,2i+k+1\}\), with \[ \text{det}\left(\begin{matrix} a_{i,2i+r+1} & \cdots & a_{i,2i+2r+1} \\ \vdots & & \vdots \\ a_{i+r,2i+r+1} & \cdots & a_{i+r,2i+2r+1} \end{matrix} \right)\neq 0\quad (i\in\mathbb{Z},\;0\leq r\leq k), \] the authors prove the existence of two-slanted matrices \(B,C,D\) satisfying \[ AB^ T=I,\;AC^ T=0,\;DC^ T=I,\;DB^ T=0\quad\text{and }B^ TA+C^ TD=I.\tag{1} \] They use these general results for locally finite decompositions of spline spaces: Let \(x=\{x_ i\}_{i\in Z}\), \(y=\{y_ i\}_{i\in Z}\) be strictly increasing knot sequences such that \(x_ i<y_ i<x_{i+1}\) and let \(\{t_ i\}_{i\in Z}:=X\cup Y\). Define the spline spaces \(V_ 0\), \(V_ 1\) as the linear span of the corresponding \(B\)-splines \[ C_ i(x):=(x_{i+k+1}-x_ i)^{1-1/p}[x_ i,\ldots,x_{i+k+1}](\cdot- x)^ k_ +, \] \[ F_ i(x):=(t_{i+k+1}-t_ i)^{1-1/p}[t_ i,\ldots,t_{i+k+1}](\cdot-x)^ k_ +\;(1\leq p\leq\infty). \] Since \(V_ 0\subset V_ 1\), we have \(C_ i(x)=\sum_{j\in Z}a_{ij}F_ j(x)\). The corresponding matrix \(A=(a_{ij})_{i,j\in Z}\) meets the above conditions such that there exist matrices \(B,C,D\) fulfilling the matrix equations (1). Define \(\psi_ j(x):=\sum_{j\in Z}d_{ij}F_ j(x)\), \(W:=\text{span}\{\psi_ j:j\in\mathbb{Z}\}\). Then \(F_ l(x)=\sum_{j\in Z}b_{jl}C_ j(x)+\sum_{j\in Z}c_{jl}\psi_ j(x)\) and \(V_ 1=V_ 0+W\).