Error formulas for divided difference expansions and numerical differentiation (Q1810589)

The author derives an expression for the remainder in divided difference expansions of the form \[ [x_0,\dots,x_n]f=\sum_{k=n}^{p-1}c_kf^{(k)}(x)/k!+R_p,\quad x_0\leqslant x_1\leqslant\dots\leqslant x_n. \] As the canonical example, the finite difference expansion is taken \[ \frac{f(x_0)-2f(x_1)+f(x_2)}{2h^2}=\frac{f''(x_1)}{2!}+ h^2\frac{f^{(4)}(x_1)}{4!}=\dots+h^{p-2}\frac{f^{(p)}(\zeta)}{p!},\tag{1} \] in which \(n = 2\), \(x_1-x_0 =x_2-x_1 = h\), \(x = x_1\), \(p\) is even, and \(\zeta\) is some number in \([x_0,x_2]\). Assume that the function \(f\) belongs to \(C^p[x_0,x_n]\) and denote by \(\|.\|\) the max norm over \([x_0,x_n]\). By deriving a new formula for the remainder \(R_p\), which is better than (1) because it only involves divided differences of \(f\) of the same order \(p\), a simple proof of the following theorem is given: There exists a constant \(C\), depending only on \(n\) and \(p\), such that \[ |R_p|\leqslant Ch^{p-n}\|f^{(p)}\|. \] When \(p- n\) is even, the remainder is expressed in a form similar to that of (1). The important case \(p = n + 2\), which occurs frequently in finite difference schemes, is studied. The paper is completed with some examples.