10.06.2016 Views

eldo_user

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Optimization<br />

Eldo Optimizer SQP Algorithm<br />

Role of Tolerances in the Eldo Optimizer SQP Method<br />

This topic describes the role of the parameters TOL_GRAD, TOL_FEAS and TOL_OPT.<br />

For more information on these parameters, see Eldo Optimizer/SQP Parameters in the Eldo<br />

Reference Manual.<br />

Suppose a value of x that is a local minimizer of f(x) in the interval a ≤ x ≤ b is to be obtained:<br />

Minimize: f(x)<br />

Subject to: a ≤ x ≤ b<br />

Or using a SPICE formulation:<br />

.OPTIMIZE<br />

* Minimize Statement<br />

.OBJECTIVE EXTRACT_INFO LABEL=F<br />

+ {$MACRO|FUNCTION}<br />

+ GOAL=MINIMIZE<br />

* Design variable specification<br />

.PARAMOPT X=(X0, A, B)<br />

The optimality conditions OPTIM(x) in the definition of the TOL_GRAD parameter are listed<br />

in Table 13-2:<br />

Table 13-2. Optimality Conditions<br />

OPTIM(x) x<br />

f’(x)<br />

a < x < b<br />

min(f’(x), 0) x = a<br />

max(f’(x),0) x = b<br />

Consider the first case in Figure 13-3. The minimum is the point x = b. The blue vector<br />

represents the derivative f’(x) at point b. The derivative f’(x) is negative, so max(f’(x), 0) = 0. A<br />

zero or very small OPTIM(x) indicates that the point x is an optimum. The number TOL_GRAD<br />

is used in the stopping test of the Eldo optimizer SQP method. The point x is optimal when the<br />

absolute value of OPTIM(x) is less than TOL_GRAD.<br />

Consider the second case in Figure 13-3. Point y is an unconstrained minimizer of f, since it lies<br />

strictly in the interval [a, b]. The optimality condition is then OPTIM(y) = f’(y) which is the<br />

slope of f at point y. Here OPTIM(y) equals zero.<br />

624<br />

Eldo® User's Manual, 15.3

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!