Conclusions

After all this mathematic theory, some words must be spend about the choice of which algorithm it is feasible to use.

Table 4.1: Optimization algorithms

Algorithm	Pro	Con
Mono-dimensional
Section search	Simple implementation.	Converges to local minima.
Parabolic interpolation	Fast convergence.	Has some pre-requirements.
SLOP	The simplest implementation.	Converges to local minima. Very slow.
Multi-dimensional
Conjugate directions	Good convergence.	Requires gradient knowledge.
Powell scheme	Fast convergence. Does not require gradient knowledge. Is not trapped by local minima.	Difficult implementation.
Simulated annealing	Simple implementation. Does not require gradient knowledge. Is not trapped by local minima.	Very slow. Fragile with respect to some critical parameters.

The characteristics of each algorithm are summarized in the table 4.1: this table should be indicate several characteristics that can be useful for the real implementation of circuit optimizer.
In the same manner the previous sections illustrate all the basic theory, useful to justify some choices made in the implementation of the optimizer.