15.05.2015 Views

Introduction to Unconstrained Optimization - Scilab

Introduction to Unconstrained Optimization - Scilab

Introduction to Unconstrained Optimization - Scilab

SHOW MORE
SHOW LESS

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

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

f(x)<br />

strong local<br />

optimum<br />

weak local<br />

optimum<br />

strong global<br />

optimum<br />

x<br />

Figure 5: Different types of optimum – strong local, weak local and strong global.<br />

These types of optimum are presented in figure 5.<br />

Definition 1.1. (Strong local minimum) The point x ⋆ is a strong local minimum<br />

of the constrained optimization problem 2–4 if there exists δ > 0 such that the two<br />

following conditions are satisfied :<br />

• f(x) is defined on N(x ⋆ , δ), and<br />

• f(x ⋆ ) < f(y), ∀y ∈ N(x ⋆ , δ), y ≠ x ⋆ .<br />

Definition 1.2. (Weak local minimum) The point x ⋆ is a weak local minimum of<br />

the constrained optimization problem 2–4 if there exists δ > 0 such that the three<br />

following conditions are satisfied<br />

• f(x) is defined on N(x ⋆ , δ),<br />

• f(x ⋆ ) ≤ f(y), ∀y ∈ N(x ⋆ , δ),<br />

• x ⋆ is not a strong local minimum.<br />

Definition 1.3. (Strong global minimum) The point x ⋆ is a strong global minimum<br />

of the constrained optimization problem 2–4 if there exists δ > 0 such that the two<br />

following conditions are satisfied :<br />

• f(x) is defined on the set of feasible points,<br />

• f(x ⋆ ) < f(y), for all y ∈ R n feasible and y ≠ x ⋆ .<br />

Most algorithms presented in this document are searching for a strong local<br />

minimum. The global minimum may be found in particular situations, for example<br />

when the cost function is convex. The difference between weak and strong local<br />

minimum is also of very little practical use, since it is difficult <strong>to</strong> determine what<br />

are the values of the function for all points except the computed point x ⋆ .<br />

In practical situations, the previous definitions does not allow <strong>to</strong> get some insight<br />

about a specific point x ⋆ . This is why we will derive later in this document first<br />

order and second order necessary conditions, which are computable characteristics<br />

of the optimum.<br />

13

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

Saved successfully!

Ooh no, something went wrong!