v2009.01.01 - Convex Optimization
v2009.01.01 - Convex Optimization v2009.01.01 - Convex Optimization
482 CHAPTER 6. CONE OF DISTANCE MATRICES From the results inE.7.2.0.2, we know matrix product V H V is the orthogonal projection of H ∈ S N on the geometric center subspace S N c . Thus the projection product P K2 H = H − P S N + P S N c H (1168) 6.8.1.1.1 Lemma. Projection on PSD cone ∩ geometric center subspace. P S N + ∩ S N c = P S N + P S N c (1169) ⋄ Proof. For each and every H ∈ S N , projection of P S N c H on the positive semidefinite cone remains in the geometric center subspace S N c = {G∈ S N | G1 = 0} (1874) = {G∈ S N | N(G) ⊇ 1} = {G∈ S N | R(G) ⊆ N(1 T )} = {V Y V | Y ∈ S N } ⊂ S N (1875) (895) That is because: eigenvectors of P S N c H corresponding to 0 eigenvalues span its nullspace N(P S N c H). (A.7.3.0.1) To project P S N c H on the positive semidefinite cone, its negative eigenvalues are zeroed. (7.1.2) The nullspace is thereby expanded while eigenvectors originally spanning N(P S N c H) remain intact. Because the geometric center subspace is invariant to projection on the PSD cone, then the rule for projection on a convex set in a subspace governs (E.9.5, projectors do not commute) and statement (1169) follows directly. From the lemma it follows Then from (1901) {P S N + P S N c H | H ∈ S N } = {P S N + ∩ S N c H | H ∈ SN } (1170) − ( S N c ∩ S N +) ∗ = {H − PS N + P S N c H | H ∈ S N } (1171) From (284) we get closure of a vector sum K 2 = − ( ) S N c ∩ S N ∗ + = S N⊥ c − S N + (1172) therefore the equality EDM N = K 1 ∩ K 2 = S N h ∩ ( ) S N⊥ c − S N + (1173)
6.8. DUAL EDM CONE 483 whose veracity is intuitively evident, in hindsight, [78, p.109] from the most fundamental EDM definition (798). Formula (1173) is not a matrix criterion for membership to the EDM cone, it is not an EDM definition, and it is not an equivalence between EDM operators nor an isomorphism. Rather, it is a recipe for constructing the EDM cone whole from large Euclidean bodies: the positive semidefinite cone, orthogonal complement of the geometric center subspace, and symmetric hollow subspace. A realization of this construction in low dimension is illustrated in Figure 124 and Figure 125. The dual EDM cone follows directly from (1173) by standard properties of cones (2.13.1.1): EDM N∗ = K ∗ 1 + K ∗ 2 = S N⊥ h − S N c ∩ S N + (1174) which bears strong resemblance to (1153). 6.8.1.2 nonnegative orthant That EDM N is a proper subset of the nonnegative orthant is not obvious from (1173). We wish to verify EDM N = S N h ∩ ( ) S N⊥ c − S N + ⊂ R N×N + (1175) While there are many ways to prove this, it is sufficient to show that all entries of the extreme directions of EDM N must be nonnegative; id est, for any particular nonzero vector z = [z i , i=1... N]∈ N(1 T ) (6.5.3.2), δ(zz T )1 T + 1δ(zz T ) T − 2zz T ≥ 0 (1176) where the inequality denotes entrywise comparison. The inequality holds because the i,j th entry of an extreme direction is squared: (z i − z j ) 2 . We observe that the dyad 2zz T ∈ S N + belongs to the positive semidefinite cone, the doublet δ(zz T )1 T + 1δ(zz T ) T ∈ S N⊥ c (1177) to the orthogonal complement (1876) of the geometric center subspace, while their difference is a member of the symmetric hollow subspace S N h . Here is an algebraic method provided by Trosset to prove nonnegativity: Suppose we are given A∈ S N⊥ c and B = [B ij ]∈ S N + and A −B ∈ S N h . Then we have, for some vector u , A = u1 T + 1u T = [A ij ] = [u i + u j ] and δ(B)= δ(A)= 2u . Positive semidefiniteness of B requires nonnegativity A −B ≥ 0 because (e i −e j ) T B(e i −e j ) = (B ii −B ij )−(B ji −B jj ) = 2(u i +u j )−2B ij ≥ 0 (1178)
- Page 431 and 432: 5.13. RECONSTRUCTION EXAMPLES 431 T
- Page 433 and 434: 5.13. RECONSTRUCTION EXAMPLES 433 w
- Page 435 and 436: 5.14. FIFTH PROPERTY OF EUCLIDEAN M
- Page 437 and 438: 5.14. FIFTH PROPERTY OF EUCLIDEAN M
- Page 439 and 440: 5.14. FIFTH PROPERTY OF EUCLIDEAN M
- Page 441 and 442: 5.14. FIFTH PROPERTY OF EUCLIDEAN M
- Page 443 and 444: 5.14. FIFTH PROPERTY OF EUCLIDEAN M
- Page 445 and 446: Chapter 6 Cone of distance matrices
- Page 447 and 448: 6.1. DEFINING EDM CONE 447 6.1 Defi
- Page 449 and 450: 6.2. POLYHEDRAL BOUNDS 449 This con
- Page 451 and 452: 6.3. √ EDM CONE IS NOT CONVEX 451
- Page 453 and 454: 6.4. A GEOMETRY OF COMPLETION 453 (
- Page 455 and 456: 6.4. A GEOMETRY OF COMPLETION 455 (
- Page 457 and 458: 6.4. A GEOMETRY OF COMPLETION 457 F
- Page 459 and 460: 6.5. EDM DEFINITION IN 11 T 459 by
- Page 461 and 462: 6.5. EDM DEFINITION IN 11 T 461 6.5
- Page 463 and 464: 6.5. EDM DEFINITION IN 11 T 463 1 0
- Page 465 and 466: 6.5. EDM DEFINITION IN 11 T 465 6.5
- Page 467 and 468: 6.6. CORRESPONDENCE TO PSD CONE S N
- Page 469 and 470: 6.6. CORRESPONDENCE TO PSD CONE S N
- Page 471 and 472: 6.6. CORRESPONDENCE TO PSD CONE S N
- Page 473 and 474: 6.7. VECTORIZATION & PROJECTION INT
- Page 475 and 476: 6.7. VECTORIZATION & PROJECTION INT
- Page 477 and 478: 6.8. DUAL EDM CONE 477 When the Fin
- Page 479 and 480: 6.8. DUAL EDM CONE 479 Proof. First
- Page 481: 6.8. DUAL EDM CONE 481 EDM 2 = S 2
- Page 485 and 486: 6.8. DUAL EDM CONE 485 6.8.1.3.1 Ex
- Page 487 and 488: 6.8. DUAL EDM CONE 487 has dual aff
- Page 489 and 490: 6.8. DUAL EDM CONE 489 6.8.1.7 Scho
- Page 491 and 492: 6.9. THEOREM OF THE ALTERNATIVE 491
- Page 493 and 494: 6.10. POSTSCRIPT 493 When D is an E
- Page 495 and 496: Chapter 7 Proximity problems In sum
- Page 497 and 498: 497 project on the subspace, then p
- Page 499 and 500: 499 H S N h 0 EDM N K = S N h ∩ R
- Page 501 and 502: 501 7.0.3 Problem approach Problems
- Page 503 and 504: 7.1. FIRST PREVALENT PROBLEM: 503 f
- Page 505 and 506: 7.1. FIRST PREVALENT PROBLEM: 505 7
- Page 507 and 508: 7.1. FIRST PREVALENT PROBLEM: 507 d
- Page 509 and 510: 7.1. FIRST PREVALENT PROBLEM: 509 7
- Page 511 and 512: 7.1. FIRST PREVALENT PROBLEM: 511 w
- Page 513 and 514: 7.1. FIRST PREVALENT PROBLEM: 513 T
- Page 515 and 516: 7.2. SECOND PREVALENT PROBLEM: 515
- Page 517 and 518: 7.2. SECOND PREVALENT PROBLEM: 517
- Page 519 and 520: 7.2. SECOND PREVALENT PROBLEM: 519
- Page 521 and 522: 7.2. SECOND PREVALENT PROBLEM: 521
- Page 523 and 524: 7.2. SECOND PREVALENT PROBLEM: 523
- Page 525 and 526: 7.3. THIRD PREVALENT PROBLEM: 525 g
- Page 527 and 528: 7.3. THIRD PREVALENT PROBLEM: 527 w
- Page 529 and 530: 7.3. THIRD PREVALENT PROBLEM: 529 7
- Page 531 and 532: 7.3. THIRD PREVALENT PROBLEM: 531 7
482 CHAPTER 6. CONE OF DISTANCE MATRICES<br />
From the results inE.7.2.0.2, we know matrix product V H V is the<br />
orthogonal projection of H ∈ S N on the geometric center subspace S N c . Thus<br />
the projection product<br />
P K2 H = H − P S N<br />
+<br />
P S N c<br />
H (1168)<br />
6.8.1.1.1 Lemma. Projection on PSD cone ∩ geometric center subspace.<br />
P S N<br />
+ ∩ S N c<br />
= P S N<br />
+<br />
P S N c<br />
(1169)<br />
⋄<br />
Proof. For each and every H ∈ S N , projection of P S N c<br />
H on the positive<br />
semidefinite cone remains in the geometric center subspace<br />
S N c = {G∈ S N | G1 = 0} (1874)<br />
= {G∈ S N | N(G) ⊇ 1} = {G∈ S N | R(G) ⊆ N(1 T )}<br />
= {V Y V | Y ∈ S N } ⊂ S N (1875)<br />
(895)<br />
That is because: eigenvectors of P S N c<br />
H corresponding to 0 eigenvalues<br />
span its nullspace N(P S N c<br />
H). (A.7.3.0.1) To project P S N c<br />
H on the positive<br />
semidefinite cone, its negative eigenvalues are zeroed. (7.1.2) The nullspace<br />
is thereby expanded while eigenvectors originally spanning N(P S N c<br />
H)<br />
remain intact. Because the geometric center subspace is invariant to<br />
projection on the PSD cone, then the rule for projection on a convex set<br />
in a subspace governs (E.9.5, projectors do not commute) and statement<br />
(1169) follows directly. <br />
From the lemma it follows<br />
Then from (1901)<br />
{P S N<br />
+<br />
P S N c<br />
H | H ∈ S N } = {P S N<br />
+ ∩ S N c H | H ∈ SN } (1170)<br />
− ( S N c ∩ S N +) ∗<br />
= {H − PS N<br />
+<br />
P S N c<br />
H | H ∈ S N } (1171)<br />
From (284) we get closure of a vector sum<br />
K 2 = − ( )<br />
S N c ∩ S N ∗<br />
+ = S<br />
N⊥<br />
c − S N + (1172)<br />
therefore the equality<br />
EDM N = K 1 ∩ K 2 = S N h ∩ ( )<br />
S N⊥<br />
c − S N +<br />
(1173)