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.

432 CHAPTER 5. EUCLIDEAN DISTANCE MATRIX<br />

That sort constraint demands: any optimal solution D ⋆ must possess the<br />

known comparative distance relationship that produces the original ordinal<br />

distance data O (1049). Ignoring the sort constraint, apparently, violates it.<br />

Yet even more remarkable is how much the map reconstructed using only<br />

ordinal data still resembles the original map of the USA after suffering the<br />

many violations produced by solving relaxed problem (1053). This suggests<br />

the simple reconstruction techniques of5.12 are robust to a significant<br />

amount of noise.<br />

5.13.2.2 Isotonic solution with sort constraint<br />

Because problems involving rank are generally difficult, we will partition<br />

(1052) into two problems we know how to solve and then alternate their<br />

solution until convergence:<br />

minimize ‖−VN T(D − O)V N ‖ F<br />

D<br />

subject to rankVN TDV N ≤ 3<br />

D ∈ EDM N<br />

minimize ‖σ − Πd‖<br />

σ<br />

subject to σ ∈ K M+<br />

(a)<br />

(b)<br />

(1054)<br />

where the sort-index matrix O (a given constant in (a)) becomes an implicit<br />

vectorized variable o i solving the i th instance of (1054b)<br />

o i ∆ = Π T σ ⋆ = 1 √<br />

2<br />

dvec O i ∈ R N(N−1)/2 , i∈{1, 2, 3...} (1055)<br />

As mentioned in discussion of relaxed problem (1053), a closed-form<br />

solution to problem (1054a) exists. Only the first iteration of (1054a) sees the<br />

original sort-index matrix O whose entries are nonnegative whole numbers;<br />

id est, O 0 =O∈ S N h ∩R N×N<br />

+ (1049). Subsequent iterations i take the previous<br />

solution of (1054b) as input<br />

O i = dvec −1 ( √ 2o i ) ∈ S N (1056)<br />

real successors to the sort-index matrix O .<br />

New problem (1054b) finds the unique minimum-distance projection of<br />

Πd on the monotone nonnegative cone K M+ . By defining<br />

Y †T ∆ = [e 1 − e 2 e 2 −e 3 e 3 −e 4 · · · e m ] ∈ R m×m (384)

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

Saved successfully!

Ooh no, something went wrong!