A computer program to minimize a function with many variables using computer evaluated exact higher-order derivatives (Q789874)

Let \(F:{\mathbb{R}}^ n\to {\mathbb{R}}\) have derivatives up to order r, \(r\geq 2\), and let A be the matrix obtained by the Hessian of F enlarged by a row and a column such that the first column is the first unit vector and the first row is given by \((1,F_{x_ 1},...,F_{x_ n}).\) Then the Newton method to minimize F may be defined by \(x^{i+1}=x^ i+d^ i,\) where \(d^ i\) is the second up to the \((n+1)\)-st element of the first row of \(A^{-1}\). Defining a ''super-Hessian'' matrix A including also higher derivatives and derivatives of powers of the \(F_{x_ i}\), higher order methods may be defined similarly. These methods have been developed by the authors in a parallel paper. In the paper at hand a Fortran program is given which uses the so-called table-method developed earlier by the authors to supply the higher derivatives needed. From the examples included it becomes not clear if the resulting algorithm is efficient in comparison with algorithms usually used to solve nonlinear least squares problems.