10.03.2015 Views

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

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.

2.13. DUAL CONE & GENERALIZED INEQUALITY 147<br />

whose optimal objective always has the saddle value 0 (regardless of the<br />

particular convex cone K and other problem parameters). [312,3.2] Thus<br />

determination of convergence for either primal or dual problem is facilitated.<br />

Were convex cone K polyhedral (2.12.1), then problems (p) and (d)<br />

would be linear programs. 2.54 Were K a positive semidefinite cone, then<br />

problem (p) has the form of prototypical semidefinite program 2.55 (584) with<br />

(d) its dual. It is sometimes possible to solve a primal problem by way of its<br />

dual; advantageous when the dual problem is easier to solve than the primal<br />

problem, for example, because it can be solved analytically, or has some special<br />

structure that can be exploited. [53,5.5.5] (4.2.3.1)<br />

<br />

2.13.1.1 Key properties of dual cone<br />

For any cone, (−K) ∗ = −K ∗<br />

For any cones K 1 and K 2 , K 1 ⊆ K 2 ⇒ K ∗ 1 ⊇ K ∗ 2 [286,2.7]<br />

(Cartesian product) For closed convex cones K 1 and K 2 , their<br />

Cartesian product K = K 1 × K 2 is a closed convex cone, and<br />

K ∗ = K ∗ 1 × K ∗ 2 (278)<br />

(conjugation) [266,14] [92,4.5] When K is any convex cone, the dual<br />

of the dual cone is the closure of the original cone; K ∗∗ = K . Because<br />

K ∗∗∗ = K ∗ K ∗ = (K) ∗ (279)<br />

When K is closed and convex, then the dual of the dual cone is the<br />

original cone; K ∗∗ = K .<br />

If any cone K has nonempty interior, then K ∗ is pointed;<br />

K nonempty interior ⇒ K ∗ pointed (280)<br />

2.54 The self-dual nonnegative orthant yields the primal prototypical linear program and<br />

its dual.<br />

2.55 A semidefinite program is any convex program constraining a variable to any subset of<br />

a positive semidefinite cone. The qualifier prototypical conventionally means: a program<br />

having linear objective, affine equality constraints, but no inequality constraints except<br />

for cone membership.

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

Saved successfully!

Ooh no, something went wrong!