Chapter 3 Geometry of convex functions - Meboo Publishing ...
Chapter 3 Geometry of convex functions - Meboo Publishing ...
Chapter 3 Geometry of convex functions - Meboo Publishing ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
252 CHAPTER 3. GEOMETRY OF CONVEX FUNCTIONS<br />
Necessary and sufficient discretization (476) allows relaxation <strong>of</strong> the<br />
semiinfinite number <strong>of</strong> conditions {w ≽ 0} instead to {w ∈ {e i , i=1... M }}<br />
the extreme directions <strong>of</strong> the selfdual nonnegative orthant. Each extreme<br />
→Y −X<br />
direction picks out a real entry f i and df(X) i<br />
from the vector-valued<br />
function and its directional derivative, then Theorem 3.7.2.0.1 applies.<br />
The vector-valued function case (592) is therefore a straightforward<br />
application <strong>of</strong> the first-order <strong>convex</strong>ity condition for real <strong>functions</strong> to each<br />
entry <strong>of</strong> the vector-valued function.<br />
3.7.4 second-order <strong>convex</strong>ity condition<br />
Again, by discretization (476), we are obliged only to consider each individual<br />
entry f i <strong>of</strong> a vector-valued function f ; id est, the real <strong>functions</strong> {f i }.<br />
For f(X) : R p →R M , a twice differentiable vector-valued function with<br />
vector argument on open <strong>convex</strong> domain,<br />
∇ 2 f i (X) ≽<br />
0 ∀X ∈ domf , i=1... M (595)<br />
S p +<br />
is a necessary and sufficient condition for <strong>convex</strong>ity <strong>of</strong> f . Obviously,<br />
when M = 1, this <strong>convex</strong>ity condition also serves for a real function.<br />
Condition (595) demands nonnegative curvature, intuitively, hence<br />
precluding points <strong>of</strong> inflection as in Figure 77 (p.260).<br />
Strict inequality in (595) provides only a sufficient condition for strict<br />
<strong>convex</strong>ity, but that is nothing new; videlicet, the strictly <strong>convex</strong> real function<br />
f i (x)=x 4 does not have positive second derivative at each and every x∈ R .<br />
Quadratic forms constitute a notable exception where the strict-case converse<br />
holds reliably.<br />
3.7.4.0.1 Exercise. Real fractional function. (confer3.4,3.6.0.0.4)<br />
Prove that real function f(x,y) = x/y is not <strong>convex</strong> on the first quadrant.<br />
Also exhibit this in a plot <strong>of</strong> the function. (f is quasilinear (p.261) on<br />
{y > 0} , in fact, and nonmonotonic even on the first quadrant.) <br />
3.7.4.0.2 Exercise. Stress function.<br />
Define |x −y| √ (x −y) 2 and<br />
X = [x 1 · · · x N ] ∈ R 1×N (76)