10.03.2015 Views

Chapter 3 Geometry of convex functions - Meboo Publishing ...

Chapter 3 Geometry of convex functions - Meboo Publishing ...

Chapter 3 Geometry of convex functions - Meboo Publishing ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

254 CHAPTER 3. GEOMETRY OF CONVEX FUNCTIONS<br />

3.8.0.0.1 Definition. Convex matrix-valued function:<br />

1) Matrix-definition.<br />

A function g(X) : R p×k →S M is <strong>convex</strong> in X iff domg is a <strong>convex</strong> set and,<br />

for each and every Y,Z ∈domg and all 0≤µ≤1 [210,2.3.7]<br />

g(µY + (1 − µ)Z) ≼<br />

µg(Y ) + (1 − µ)g(Z) (597)<br />

S M +<br />

Reversing sense <strong>of</strong> the inequality flips this definition to concavity. Strict<br />

<strong>convex</strong>ity is defined less a stroke <strong>of</strong> the pen in (597) similarly to (478).<br />

2) Scalar-definition.<br />

It follows that g(X) : R p×k →S M is <strong>convex</strong> in X iff w T g(X)w : R p×k →R is<br />

<strong>convex</strong> in X for each and every ‖w‖= 1; shown by substituting the defining<br />

inequality (597). By dual generalized inequalities we have the equivalent but<br />

more broad criterion, (2.13.5)<br />

g <strong>convex</strong> w.r.t S M + ⇔ 〈W , g〉 <strong>convex</strong><br />

for each and every W ≽<br />

S M∗<br />

+<br />

0 (598)<br />

Strict <strong>convex</strong>ity on both sides requires caveat W ≠ 0. Because the<br />

set <strong>of</strong> all extreme directions for the selfdual positive semidefinite cone<br />

(2.9.2.4) comprises a minimal set <strong>of</strong> generators for that cone, discretization<br />

(2.13.4.2.1) allows replacement <strong>of</strong> matrix W with symmetric dyad ww T as<br />

proposed.<br />

△<br />

3.8.0.0.2 Example. Taxicab distance matrix.<br />

Consider an n-dimensional vector space R n with metric induced by the<br />

1-norm. Then distance between points x 1 and x 2 is the norm <strong>of</strong> their<br />

difference: ‖x 1 −x 2 ‖ 1 . Given a list <strong>of</strong> points arranged columnar in a matrix<br />

X = [x 1 · · · x N ] ∈ R n×N (76)<br />

then we could define a taxicab distance matrix<br />

D 1 (X) (I ⊗ 1 T n) | vec(X)1 T − 1 ⊗X | ∈ S N h ∩ R N×N<br />

+<br />

⎡<br />

⎤<br />

0 ‖x 1 − x 2 ‖ 1 ‖x 1 − x 3 ‖ 1 · · · ‖x 1 − x N ‖ 1<br />

‖x 1 − x 2 ‖ 1 0 ‖x 2 − x 3 ‖ 1 · · · ‖x 2 − x N ‖ 1<br />

=<br />

‖x<br />

⎢ 1 − x 3 ‖ 1 ‖x 2 − x 3 ‖ 1 0 ‖x 3 − x N ‖ 1<br />

⎥<br />

⎣ . .<br />

... . ⎦<br />

‖x 1 − x N ‖ 1 ‖x 2 − x N ‖ 1 ‖x 3 − x N ‖ 1 · · · 0<br />

(599)

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

Saved successfully!

Ooh no, something went wrong!