v2010.10.26 - Convex Optimization

v2010.10.26 - Convex Optimization v2010.10.26 - Convex Optimization

convexoptimization.com
from convexoptimization.com More from this publisher
12.07.2015 Views

568 CHAPTER 7. PROXIMITY PROBLEMS7.2.1.1 Equivalent semidefinite program, Problem 2, convex caseConvex problem (1352) is numerically solvable for its global minimum usingan interior-point method [391] [289] [277] [382] [11] [145]. We translate (1352)to an equivalent semidefinite program (SDP) for a pedagogical reason madeclear in7.2.2.2 and because there exist readily available computer programsfor numerical solution [167] [384] [385] [359] [34] [383] [348] [336].Substituting a new matrix variable Y [y ij ]∈ R N×N+h ij√dij ← y ij (1354)Boyd proposes: problem (1352) is equivalent to the semidefinite program∑minimize d ij − 2y ij + h 2 ijD , Yi,j[ ]dij y ij(1355)subject to≽ 0 , i,j =1... Ny ij h 2 ijD ∈ EDM NTo see that, recall d ij ≥ 0 is implicit to D ∈ EDM N (5.8.1, (910)). Sowhen H ∈ R N×N+ is nonnegative as assumed,[ ]dij y √ij≽ 0 ⇔ h ij√d ij ≥ yij 2 (1356)y ijh 2 ijMinimization of the objective function implies maximization of y ij that isbounded above. √Hence nonnegativity of y ij is implicit to (1355) and, asdesired, y ij →h ij dij as optimization proceeds.If the given matrix H is now assumed symmetric and nonnegative,H = [h ij ] ∈ S N ∩ R N×N+ (1357)then Y = H ◦ ◦√ D must belong to K= S N h ∩ R N×N+ (1301). Because Y ∈ S N h(B.4.2 no.20), then‖ ◦√ D − H‖ 2 F = ∑ i,jd ij − 2y ij + h 2 ij = −N tr(V (D − 2Y )V ) + ‖H‖ 2 F (1358)

7.2. SECOND PREVALENT PROBLEM: 569So convex problem (1355) is equivalent to the semidefinite programminimize − tr(V (D − 2Y )V )D , Y[ ]dij y ijsubject to≽ 0 ,N ≥ j > i = 1... N −1y ij h 2 ij(1359)Y ∈ S N hD ∈ EDM Nwhere the constants h 2 ij and N have been dropped arbitrarily from theobjective.7.2.1.2 Gram-form semidefinite program, Problem 2, convex caseThere is great advantage to expressing problem statement (1359) inGram-form because Gram matrix G is a bidirectional bridge between pointlist X and distance matrix D ; e.g., Example 5.4.2.3.5, Example 6.7.0.0.1.This way, problem convexity can be maintained while simultaneouslyconstraining point list X , Gram matrix G , and distance matrix D at ourdiscretion.Convex problem (1359) may be equivalently written via linear bijective(5.6.1) EDM operator D(G) (903);minimizeG∈S N c , Y ∈ S N hsubject to− tr(V (D(G) − 2Y )V )[ ]〈Φij , G〉 y ij≽ 0 ,y ijh 2 ijN ≥ j > i = 1... N −1(1360)G ≽ 0where distance-square D = [d ij ] ∈ S N h (887) is related to Gram matrix entriesG = [g ij ] ∈ S N c ∩ S N + bywhered ij = g ii + g jj − 2g ij= 〈Φ ij , G〉(902)Φ ij = (e i − e j )(e i − e j ) T ∈ S N + (889)

568 CHAPTER 7. PROXIMITY PROBLEMS7.2.1.1 Equivalent semidefinite program, Problem 2, convex case<strong>Convex</strong> problem (1352) is numerically solvable for its global minimum usingan interior-point method [391] [289] [277] [382] [11] [145]. We translate (1352)to an equivalent semidefinite program (SDP) for a pedagogical reason madeclear in7.2.2.2 and because there exist readily available computer programsfor numerical solution [167] [384] [385] [359] [34] [383] [348] [336].Substituting a new matrix variable Y [y ij ]∈ R N×N+h ij√dij ← y ij (1354)Boyd proposes: problem (1352) is equivalent to the semidefinite program∑minimize d ij − 2y ij + h 2 ijD , Yi,j[ ]dij y ij(1355)subject to≽ 0 , i,j =1... Ny ij h 2 ijD ∈ EDM NTo see that, recall d ij ≥ 0 is implicit to D ∈ EDM N (5.8.1, (910)). Sowhen H ∈ R N×N+ is nonnegative as assumed,[ ]dij y √ij≽ 0 ⇔ h ij√d ij ≥ yij 2 (1356)y ijh 2 ijMinimization of the objective function implies maximization of y ij that isbounded above. √Hence nonnegativity of y ij is implicit to (1355) and, asdesired, y ij →h ij dij as optimization proceeds.If the given matrix H is now assumed symmetric and nonnegative,H = [h ij ] ∈ S N ∩ R N×N+ (1357)then Y = H ◦ ◦√ D must belong to K= S N h ∩ R N×N+ (1301). Because Y ∈ S N h(B.4.2 no.20), then‖ ◦√ D − H‖ 2 F = ∑ i,jd ij − 2y ij + h 2 ij = −N tr(V (D − 2Y )V ) + ‖H‖ 2 F (1358)

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

Saved successfully!

Ooh no, something went wrong!