Differential geometry is used to investigate the structure of neural-n
etwork-based control systems. The key aspect is relative order-an inva
riant property of dynamic systems. Finite relative order allows the sp
ecification of a minimal architecture for a recurrent network. Any sys
tem with finite relative order has a left inverse. It is shown that a
recurrent network with finite relative order has a local inverse that
is also a recurrent network with the same weights. The results have im
plications for the use of recurrent networks in the inverse-model-base
d control of nonlinear systems.