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.

6.10. POSTSCRIPT 493<br />

When D is an EDM [223,2]<br />

N(D) ⊂ N(1 T ) = {z | 1 T z = 0} (1203)<br />

Because [137,2] (E.0.1)<br />

then<br />

DD † 1 = 1<br />

1 T D † D = 1 T (1204)<br />

R(1) ⊂ R(D) (1205)<br />

6.10 postscript<br />

We provided an equality (1173) relating the convex cone of Euclidean distance<br />

matrices to the convex cone of positive semidefinite matrices. Projection on<br />

a positive semidefinite cone, constrained by an upper bound on rank, is easy<br />

and well known; [107] simply, a matter of truncating a list of eigenvalues.<br />

Projection on a positive semidefinite cone with such a rank constraint is, in<br />

fact, a convex optimization problem. (7.1.4)<br />

In the past, it was difficult to project on the EDM cone under a<br />

constraint on rank or affine dimension. A surrogate method was to invoke<br />

the Schoenberg criterion (817) and then project on a positive semidefinite<br />

cone under a rank constraint bounding affine dimension from above. But a<br />

solution acquired that way is necessarily suboptimal.<br />

In7.3.3 we present a method for projecting directly on the EDM cone<br />

under a constraint on rank or affine dimension.

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

Saved successfully!

Ooh no, something went wrong!