No label defined (Q4499545)

In [Approximation and interpolation methods, Novosibirsk, 98-107 (1981; Zbl 0565.41001)], the author proposed a generalization of the notion of rational splines. The spline \(S(x)\) that interpolates values \(\{f_i\}_{i=0}^N\) at \(\{x_i\}_{i=0}^N\) is as follows: \[ S(x)=(1-t)f_i+tf_{i+1}+C_i\left[\frac{t^3}{1+\phi_i(t)}-t\right]+ D_i\left[\frac{(1-t)^3}{1+\psi_i(t)}-(1-t)\right],\quad x\in[x_i,x_{i+1}], \] where \(t=(x-x_i)/(x_{i+1}-x_i)\), \(\phi_i(t),\psi_i(t)\in C^2[0,1]\). The problem of finding the coefficients \(C_i\) and \(D_i\) is reduced to solving a tridiagonal system of equations with respect to the moments \(S''(x_i)\). In most available constructions of such splines (which differ in the choice of \(\phi_i(t)\) and \(\psi_i(t)\)), the matrices of systems do not possess diagonal dominance. This circumstance considerably reduces the area of application of such splines, in particular, complicates the development of algorithms of convex interpolation by rational splines. The author suggests several ways of choosing the functions \(\phi_i(t)\) and \(\psi_i(t)\) which lead to systems with diagonal-dominant matrices.