Multi-node higher order expansions of a function. (Q1421506)

The author starts with the linear operator \(L:C[a,b]\to C[a,b]\) \[ L(f;x)=\sum\limits^n_{i=0}\varphi_i(x) f(x_i) \] where \(a=x_0\leq x_1\leq\dots\leq x_n=b\) and \(\varphi_i\in C[a,b]\) are chosen so that \(L\) reproduces the polynomials of degree \(\leq m\). Replacing the values \(f(x_i)\) by \(\sum^r_{j=0} a_j(x-x_j)^j f^{(j)}(x_i)/j!\) with \[ a_j:=\frac{r!(m+r-j)!}{(m+r)!(r-j)!} \] \(j=0,1,\dots,r\) the following multi-node higher expansion, reproducing the polynomials of degree \(\leq m+r\) is obtained \[ H_{n,r}(f;x):=\sum^n_{i=0}\varphi_i(x) \sum^r_{j=0}\frac{a_j}{j!}\, (x-x_i)^j f^{(j)}(x_i) \] The error terms of the above expansion are given. As applications, the extensions of Taylor polynomials, Bernstein polynomials and Lagrange interpolation polynomials are presented. The comparisons of the errors given cubic approximation polynomials and the cubic Hermite interpolation polynomial are detailed.