Constructing wavelets by welding segments of smooth functions (Q6875692)

The combination of refinement properties of B-splines [\textit{C. De Boor}, A practical guide to splines. Rev. ed. New York, NY: Springer (2001; Zbl 0987.65015)] with the lifting scheme [\textit{I. Daubechies} and \textit{W. Sweldens}, Lect. Notes Earth Sci. 90, 131--157 (2000; Zbl 0963.65154)] leads to B-spline wavelet transforms having desirable smoothness and approximation properties [\textit{M. Jansen}, Wavelets from a statistical perspective. Boca Raton, FL: CRC Press (2022; Zbl 1510.42001)]. In the present paper, the author generalizes this construction to wavelet bases/transforms built from blocks taken from infinitely supported, arbitrarily smooth functions, stitched together at nonequispaced knots with well-defined smoothness properties at these knots. The construction also allows for a flexible handling of boundary conditions when finite domains are involved. The wavelet transform described in the paper bears resemblance to classical multiscale algorithms associated with dyadic refinement equations and conjugate filter pairs of dyadic wavelet transforms.\N\NThe flexibility and optimality concerning smoothness, distribution of knot positions, and optimal approximation come at the cost of constructing wavelets, which are not shifted and scaled versions of a scaling function and a corresponding wavelet. The efficiency of the wavelet bases proposed in the paper is illustrated by approximation errors to discretely sampled data.\N\NThe adaptation of the procedure to Riesz bases or frames is still an open problem.