10.07.2015 Views

v2007.09.13 - Convex Optimization

v2007.09.13 - Convex Optimization

v2007.09.13 - 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.

5.4. EDM DEFINITION 325because (A.3.1.0.5)Ω ≽ 0 ⇒ Θ T Θ ≽ 0 (778)Decomposition (775) and the relative-angle matrix inequality Ω ≽ 0 lead toa different expression of an inner-product form EDM definition (770)D(Ω,d) ∆ ==[ ] 01dT + 1 [ 0 d ] √ ([ ]) [ ] √ ([ ])0 0 0 T T 0− 2 δδd 0 Ω d[ ]0 dTd d1 T + 1d T − 2 √ δ(d) Ω √ ∈ EDM Nδ(d)(779)and another expression of the EDM cone:EDM N ={D(Ω,d) | Ω ≽ 0, √ }δ(d) ≽ 0(780)In the particular circumstance x 1 = 0, we can relate interpoint anglematrix Ψ from the Gram decomposition in (714) to relative-angle matrixΩ in (775). Thus,[ ] 1 0TΨ ≡ , x0 Ω 1 = 0 (781)5.4.3.2 Inner-product form −V T N D(Θ)V N convexityOn[√page 323]we saw that each EDM entry d ij is a convex quadratic functiondikof √dkj and a quasiconvex function of θ ikj . Here the situation forinner-product form EDM operator D(Θ) (770) is identical to that in5.4.1for list-form D(X) ; −D(Θ) is not a quasiconvex function of Θ by the samereasoning, and from (774)−V T N D(Θ)V N = Θ T Θ (782)is a convex quadratic function of Θ on domain R n×N−1 achieving its minimumat Θ = 0.

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

Saved successfully!

Ooh no, something went wrong!