A Method for Minimizing a Sum of Squares of Non-Linear Functions Without Calculating Derivatives

Abstract

The minimum of a sum of squares can often be found very efficiently by applying a generalization of the least squares method for solving overdetermined linear simultaneous equations. An original method that has comparable convergence but, unlike the classical procedure, does not require any derivatives is described and discussed in this paper. The number of times the individual terms of the sum of squares have to be calculated is approximately proportional to the number of variables. Finding a solution to a set of fifty non-linear equations in fifty unknowns required the left-hand sides of the equations to be worked out fewer than two hundred times.