10.03.2015 Views

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

v2009.01.01 - Convex Optimization

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

310 CHAPTER 4. SEMIDEFINITE PROGRAMMING<br />

4.6.0.0.2 Example. Procrustes problem. [45]<br />

Example 4.6.0.0.1 is extensible. An orthonormal matrix Q∈ R n×p is<br />

completely characterized by Q T Q = I . Consider the particular case<br />

Q = [x y ]∈ R n×2 as variable to a Procrustes problem (C.3): given<br />

A∈ R m×n and B ∈ R m×2<br />

minimize ‖AQ − B‖ F<br />

Q∈R n×2<br />

subject to Q T Q = I<br />

(704)<br />

which is nonconvex. By vectorizing matrix Q we can make the assignment:<br />

⎡ ⎤ ⎡ ⎤ ⎡ ⎤<br />

x [ x T y T 1] X Z x xx T xy T x<br />

G = ⎣ y ⎦ = ⎣ Z T Y y ⎦=<br />

∆ ⎣ yx T yy T y ⎦∈ S 2n+1 (705)<br />

1<br />

x T y T 1 x T y T 1<br />

Now orthonormal Procrustes problem (704) can be equivalently restated:<br />

minimize<br />

X , Y , Z , x , y<br />

‖A[x y ] − B‖ F<br />

⎡<br />

X Z<br />

⎤<br />

x<br />

subject to G = ⎣ Z T Y y ⎦<br />

x T y T 1<br />

(G ≽ 0)<br />

rankG = 1<br />

trX = 1<br />

trY = 1<br />

trZ = 0<br />

(706)<br />

To solve this, we form the convex problem sequence:<br />

minimize<br />

X , Y , Z , x , y<br />

‖A[x y ] −B‖ F + 〈G, W 〉<br />

⎡<br />

X Z<br />

⎤<br />

x<br />

subject to G = ⎣ Z T Y y ⎦ ≽ 0<br />

x T y T 1<br />

trX = 1<br />

trY = 1<br />

trZ = 0<br />

(707)

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

Saved successfully!

Ooh no, something went wrong!