Solution of nonlinear ordinary differential equations by feedforward neural networks (Q1344634)

This paper is concerned with the construction of feedforward neural networks to approximate given nonlinear ordinary differential equations without the need for training. The weights of these networks can be determined by solving systems of both linear and nonlinear algebraic equations. The resulting nets are architecturally identical to those that use conventional training techniques, yet they possess monotonically decreasing errors for increasing number of processing elements. Numerical examples are discussed.