12.07.2015 Views

v2010.10.26 - Convex Optimization

v2010.10.26 - Convex Optimization

v2010.10.26 - 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.

490 CHAPTER 5. EUCLIDEAN DISTANCE MATRIXBecause point labelling is arbitrary, this fifth Euclidean metricrequirement must apply to each of the N points as though each were in turnlabelled x 1 ; hence the new index k in (1162). Just as the triangle inequalityis the ultimate test for realizability of only three points, the relative-angleinequality is the ultimate test for only four. For four distinct points, thetriangle inequality remains a necessary although penultimate test; (5.4.3)Four Euclidean metric properties (5.2).Angle θ inequality (885) or (1162).⇔ −V T N DV N ≽ 0D ∈ S 4 h⇔ D = D(Θ) ∈ EDM 4The relative-angle inequality, for this case, is illustrated in Figure 136.5.14.2.2 Beyond the fifth metric property(1163)When cardinality N exceeds 4, the first four properties of the Euclideanmetric and the relative-angle inequality together become insufficientconditions for realizability. In other words, the four Euclidean metricproperties and relative-angle inequality remain necessary but become asufficient test only for positive semidefiniteness of all the principal 3 × 3submatrices [sic] in −VN TDV N . Relative-angle inequality can be consideredthe ultimate test only for realizability at each vertex x k of each and everypurported tetrahedron constituting a hyperdimensional body.When N = 5 in particular, relative-angle inequality becomes thepenultimate Euclidean metric requirement while nonnegativity of thenunwieldy det(Θ T Θ) corresponds (by the positive (semi)definite principalsubmatrices theorem inA.3.1.0.4) to the sixth and last Euclidean metricrequirement. Together these six tests become necessary and sufficient, andso on.Yet for all values of N , only assuming nonnegative d ij , relative-anglematrix inequality in (1075) is necessary and sufficient to certify realizability;(5.4.3.1)Euclidean metric property 1 (5.2).Angle matrix inequality Ω ≽ 0 (963).⇔ −V T N DV N ≽ 0D ∈ S N h⇔ D = D(Ω,d) ∈ EDM N(1164)Like matrix criteria (886), (910), and (1075), the relative-angle matrixinequality and nonnegativity property subsume all the Euclidean metricproperties and further requirements.

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

Saved successfully!

Ooh no, something went wrong!