
Nonlinear systems 87
3.5.1 Local optimisation methods
Local algorithms typically make use of gradient information (calculated either
analytically of numerically) o f the cost function to find the search direction
while determining the optimum. Global algorithms, in contrast, typically use
randomisation and/ or heuristic search techniques which require only the cal-
culation of the objective function value. The search space, or design space, for
the s e t of optimisation parameters being used may b e convex or non-convex.
Fig. 3 .11 shows a two-dimensional convex search space in the parameters x
and y, with a corr esp onding cost function z. Clearly,