A Proposed Hybrid-Computer Method for Functional Optimization

To obtain successively improved control-variable functions
^k
u(t), we propose to add a random perturbation δu(t) to
^k
u(t). A fast analog-computer run then solves the dynamical system equations and computes the given criterion function. If the latter has been improved, then
^k+1
u(t)=u(t)+δu(t) is stored digitally for the next iteration; otherwise 5u( t) is discarded and a new control pertubation is tried.