A generalization of Floater-Hormann interpolants (Q6145206)

The authors propose a family of rational interpolants \(r(x)\) of type \((\gamma n-(\gamma-1)d,\gamma (n-d))\), depending on an integer parameter \(\gamma\ge1\), for functions prescribed on a set of points \(a=x_0<x_1<\cdots<x_{n-1}<x_n=b\). The generalization in the title refers to the introduction of the integer parameter \(\gamma\ge1\) since the original Floater-Hormann (FH) interpolants appear for \(\gamma=1\). (See [\textit{M.S. Floater} and \textit{K. Hormann}, Numer.\ Math. 107, No.~2, 315--331 (2007; Zbl 1221.41002)].) In particular for integer \(d\), \(0\le d\le n\), \[ r(x)=\frac{\sum_{i=0}^{n-d} \lambda_i(x)p_i(x)}{\sum_{i=0}^{n-d} \lambda_i(x)},\quad \lambda_i(x)=\frac{(-1)^{i\gamma}}{\prod_{j=i}^{i+d}(x-x_j)^{\gamma}},\quad i=0,\ldots,n-d \] where the \(p_i\) are the interpolating polynomials in the nodes \(\{x_i,\ldots,x_{i+d}\}\). Classical properties are kept: \(r\) has no real poles and it reproduces polynomials of degree at most \(d\). A barycentric expression and estimates for the Lebesgue constants are derived. These estimates remain bounded as \(n\to\infty\), even for (almost) equidistant interpolation points. Several results about converge and rate of convergence are proved depending on the smoothness of the function being interpolated. Numerical examples illustrate the results and show how \(\gamma=2\) outperforms \(\gamma=1\).