Chapter 3 Geometry of convex functions - Meboo Publishing ...
Chapter 3 Geometry of convex functions - Meboo Publishing ...
Chapter 3 Geometry of convex functions - Meboo Publishing ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
218 CHAPTER 3. GEOMETRY OF CONVEX FUNCTIONS<br />
Over some <strong>convex</strong> set C given vector constant y or matrix constant Y<br />
arg inf<br />
x∈C ‖x − y‖ 2 = arg inf<br />
x∈C ‖x − y‖2 2 (490)<br />
arg inf<br />
X∈ C ‖X − Y ‖ F = arg inf<br />
X∈ C ‖X − Y ‖2 F (491)<br />
are unconstrained <strong>convex</strong> quadratic problems. Equality does not hold for a<br />
sum <strong>of</strong> norms. (5.4.2.3.2) Optimal solution is norm dependent: [59, p.297]<br />
minimize<br />
x∈R n ‖x‖ 1<br />
subject to x ∈ C<br />
≡<br />
minimize 1 T t<br />
x∈R n , t∈R n<br />
subject to −t ≼ x ≼ t<br />
x ∈ C<br />
(492)<br />
minimize<br />
x∈R n ‖x‖ 2<br />
subject to x ∈ C<br />
≡<br />
minimize<br />
x∈R n , t∈R<br />
subject to<br />
t<br />
[ tI x<br />
x T t<br />
x ∈ C<br />
]<br />
≽<br />
S n+1<br />
+<br />
0<br />
(493)<br />
minimize ‖x‖ ∞<br />
x∈R n<br />
subject to x ∈ C<br />
≡<br />
minimize t<br />
x∈R n , t∈R<br />
subject to −t1 ≼ x ≼ t1<br />
x ∈ C<br />
(494)<br />
In R n these norms represent: ‖x‖ 1 is length measured along a grid<br />
(taxicab distance), ‖x‖ 2 is Euclidean length, ‖x‖ ∞ is maximum |coordinate|.<br />
minimize<br />
x∈R n ‖x‖ 1<br />
subject to x ∈ C<br />
≡<br />
minimize 1 T (α + β)<br />
α∈R n , β∈R n<br />
subject to α,β ≽ 0<br />
x = α − β<br />
x ∈ C<br />
(495)<br />
These foregoing problems (486)-(495) are <strong>convex</strong> whenever set C is. Their<br />
equivalence transformations make objectives smooth.