13.07.2015 Views

v2006.03.09 - Convex Optimization

v2006.03.09 - Convex Optimization

v2006.03.09 - 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.

4.4. EDM DEFINITION 223The collection of all Euclidean distance matrices EDM N is a convex subsetof R N×N+ called the EDM cone (5, Figure 89, p.390);0 ∈ EDM N ⊆ S N h ∩ R N×N+ ⊂ S N (459)An EDM D must be expressible as a function of some list X ; id est, it musthave the formD(X) ∆ = δ(X T X)1 T + 1δ(X T X) T − 2X T X ∈ EDM N (460)= [vec(X) T (Φ ij ⊗ I) vecX , i,j=1... N] (461)Function D(X) will make an EDM given any X ∈ R n×N , conversely, butD(X) is not a convex function of X (4.4.1). Now the EDM cone may bedescribed:EDM N = { D(X) | X ∈ R N−1×N} (462)Expression D(X) is a matrix definition of EDM and so conforms to theEuclidean metric properties:Nonnegativity of EDM entries (property 1,4.2) is obvious from thedistance-square definition (456), so holds for any D expressible in the formD(X) in (460).When we say D is an EDM, reading from (460), it implicitly meansthe main diagonal must be 0 (property 2, self-distance) and D must besymmetric (property 3); δ(D) = 0 and D T = D or, equivalently, D ∈ S N hare necessary matrix criteria.4.4.0.1 homogeneityFunction D(X) is homogeneous in the sense, for ζ ∈ R√ √◦D(ζX) = |ζ|◦D(X) (463)where the positive square root is entrywise.Any nonnegatively scaled EDM remains an EDM; id est, the matrix classEDM is invariant to nonnegative scaling (αD(X) for α≥0) because allEDMs of dimension N constitute a convex cone EDM N (5, Figure 75).

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

Saved successfully!

Ooh no, something went wrong!