13.04.2014 Views

Mathematical Optimization in Graphics and Vision - Luiz Velho - Impa

Mathematical Optimization in Graphics and Vision - Luiz Velho - Impa

Mathematical Optimization in Graphics and Vision - Luiz Velho - Impa

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.

2.2. CONTINUOUS OPTIMIZATION PROBLEMS 23<br />

• the description (def<strong>in</strong>ition) of the solution set S;<br />

• the properties of the objective function f.<br />

The most important classification is the one based on the nature of the solution<br />

set S, that leads us to classify<strong>in</strong>g optimization problems <strong>in</strong>to four classes:<br />

cont<strong>in</strong>uous, discrete, comb<strong>in</strong>atorial <strong>and</strong> variational.<br />

A po<strong>in</strong>t x of a topological space S is an accumulation po<strong>in</strong>t, if, <strong>and</strong> only if, for<br />

any open ball B x , with x ∈ B x , there exists an element y ∈ S such that y ∈ B x .<br />

A po<strong>in</strong>t x ∈ S which is not an accumulation po<strong>in</strong>t is called an isolated po<strong>in</strong>t of S.<br />

Thus, x is isolated if there exists an open ball B x , such that x ∈ B x <strong>and</strong> no other<br />

po<strong>in</strong>t of S belongs to B x .<br />

A topological space is discrete if it conta<strong>in</strong>s no accumulation po<strong>in</strong>ts. It is cont<strong>in</strong>uous<br />

when all of its po<strong>in</strong>ts are accumulation po<strong>in</strong>ts.<br />

Example 8 (Discrete <strong>and</strong> cont<strong>in</strong>uous sets). Any f<strong>in</strong>ite subset of an euclidean<br />

space is obviously discrete. The set R n itself is cont<strong>in</strong>uous. The set Z n =<br />

{(i 1 , . . . , i n ); i j ∈ Z} is discrete. The set {0} ∪ {1/n; n ∈ Z} is not discrete because<br />

0 is an accumulation po<strong>in</strong>t (no matter how small we take a ball with center<br />

at 0, it will conta<strong>in</strong> some element 1/n for some large n); this set is not cont<strong>in</strong>uous<br />

either, because with the exception of 0 all of the po<strong>in</strong>ts are isolated. On the other<br />

h<strong>and</strong>, the set {1/n; n ∈ Z} (without the 0) is discrete.<br />

2.2 Cont<strong>in</strong>uous optimization problems<br />

An optimization problem is called cont<strong>in</strong>uous when the solution set S is a cont<strong>in</strong>uous<br />

subset of R n .<br />

The most common cases occur when S is a differentiable m-dimensional surface<br />

of R n (with or without boundary), or S ⊆ R n is a region of R n :<br />

• Maximize the function f(x, y) = 3x + 4y + 1 on the set S = {(x, y) ∈<br />

R 2 ; x 2 + y 2 = 1}. The solution set S is the unit circle S 1 , which is a curve<br />

of R 2 (i.e. a one-dimensional “surface”, see Figure 2.1(a)).<br />

• Maximize the function f(x, y) = 3x + 4y + 1 on the set S = {(x, y) ∈<br />

R 2 ; x 2 + y 2 ≤ 1}. The solution set is the disk of R 2 , a two-dimensional<br />

surface with boundary (see Figure 2.1(b)).

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

Saved successfully!

Ooh no, something went wrong!