It is shown that the familiar least-squares/Taylor technique may easily be generalized into a least

th/Taylor method. The process is only slightly more complicated if p > 2; then the maximum error is smaller than for

and the convergence often improves.