v2009.01.01 - Convex Optimization
v2009.01.01 - Convex Optimization v2009.01.01 - Convex Optimization
700 APPENDIX E. PROJECTION b H 1 x 22 x 21 H 2 x 12 H 1 ∩ H 2 x 11 Figure 148: H 1 and H 2 are the same halfspaces as in Figure 143. Dykstra’s alternating projection algorithm generates the alternations b, x 21 , x 11 , x 22 , x 12 , x 12 ... x 12 . The path illustrated from b to x 12 in R 2 terminates at the desired result, Pb in Figure 143. The {y ki } correspond to the first two difference vectors drawn (in the first iteration i=1), then oscillate between zero and a negative vector thereafter. These alternations are not so robust in presence of noise as for the example in Figure 142. E.10.3 Optimization and projection Unique projection on the nonempty intersection of arbitrary convex sets to find the closest point therein is a convex optimization problem. The first successful application of alternating projection to this problem is attributed to Dykstra [105] [54] who in 1983 provided an elegant algorithm that prevails today. In 1988, Han [152] rediscovered the algorithm and provided a primal−dual convergence proof. A synopsis of the history of alternating projection E.21 can be found in [56] where it becomes apparent that Dykstra’s work is seminal. E.21 For a synopsis of alternating projection applied to distance geometry, see [306,3.1].
E.10. ALTERNATING PROJECTION 701 E.10.3.1 Dykstra’s algorithm Assume we are given some point b ∈ R n and closed convex sets {C k ⊂ R n | k=1... L}. Let x ki ∈ R n and y ki ∈ R n respectively denote a primal and dual vector (whose meaning can be deduced from Figure 148 and Figure 149) associated with set k at iteration i . Initialize y k0 = 0 ∀k=1... L and x 1,0 = b (1961) Denoting by P k t the unique minimum-distance projection of t on C k , and for convenience x L+1,i ∆ = x 1,i−1 , calculation of the iterates x 1i proceeds: E.22 for i=1, 2,...until convergence { for k=L... 1 { t = x k+1,i − y k,i−1 x ki = P k t y ki = P k t − t } } (1962) Assuming a nonempty intersection, then the iterates converge to the unique minimum-distance projection of point b on that intersection; [92,9.24] Pb = lim i→∞ x 1i (1963) In the case all the C k are affine, then calculation of y ki is superfluous and the algorithm becomes identical to alternating projection. [92,9.26] [123,1] Dykstra’s algorithm is so simple, elegant, and represents such a tiny increment in computational intensity over alternating projection, it is nearly always arguably cost effective. E.10.3.2 Normal cone Glunt [130,4] observes that the overall effect of Dykstra’s iterative procedure is to drive t toward the translated normal cone to ⋂ C k at the solution Pb (translated to Pb). The normal cone gets its name from its graphical construction; which is, loosely speaking, to draw the outward-normals at Pb (Definition E.9.1.0.1) to all the convex sets C k touching Pb . The relative interior of the normal cone subtends these normal vectors. E.22 We reverse order of projection (k=L...1) in the algorithm for continuity of exposition.
- Page 649 and 650: E.3. SYMMETRIC IDEMPOTENT MATRICES
- Page 651 and 652: E.3. SYMMETRIC IDEMPOTENT MATRICES
- Page 653 and 654: E.3. SYMMETRIC IDEMPOTENT MATRICES
- Page 655 and 656: E.5. PROJECTION EXAMPLES 655 E.4.0.
- Page 657 and 658: E.5. PROJECTION EXAMPLES 657 a ∗
- Page 659 and 660: E.5. PROJECTION EXAMPLES 659 E.5.0.
- Page 661 and 662: E.6. VECTORIZATION INTERPRETATION,
- Page 663 and 664: E.6. VECTORIZATION INTERPRETATION,
- Page 665 and 666: E.6. VECTORIZATION INTERPRETATION,
- Page 667 and 668: E.6. VECTORIZATION INTERPRETATION,
- Page 669 and 670: E.7. ON VECTORIZED MATRICES OF HIGH
- Page 671 and 672: E.7. ON VECTORIZED MATRICES OF HIGH
- Page 673 and 674: E.8. RANGE/ROWSPACE INTERPRETATION
- Page 675 and 676: E.9. PROJECTION ON CONVEX SET 675 A
- Page 677 and 678: E.9. PROJECTION ON CONVEX SET 677 W
- Page 679 and 680: E.9. PROJECTION ON CONVEX SET 679 P
- Page 681 and 682: E.9. PROJECTION ON CONVEX SET 681 E
- Page 683 and 684: E.9. PROJECTION ON CONVEX SET 683 T
- Page 685 and 686: E.9. PROJECTION ON CONVEX SET 685
- Page 687 and 688: E.10. ALTERNATING PROJECTION 687 E.
- Page 689 and 690: E.10. ALTERNATING PROJECTION 689 b
- Page 691 and 692: E.10. ALTERNATING PROJECTION 691 a
- Page 693 and 694: E.10. ALTERNATING PROJECTION 693 (a
- Page 695 and 696: E.10. ALTERNATING PROJECTION 695 wh
- Page 697 and 698: E.10. ALTERNATING PROJECTION 697 E.
- Page 699: E.10. ALTERNATING PROJECTION 699 10
- Page 703 and 704: E.10. ALTERNATING PROJECTION 703 E
- Page 705 and 706: Appendix F Notation and a few defin
- Page 707 and 708: 707 a.i. c.i. l.i. w.r.t affinely i
- Page 709 and 710: 709 is or ← → t → 0 + as in
- Page 711 and 712: 711 ∑ π(γ) Ξ Π ∏ ψ(Z) D D
- Page 713 and 714: 713 R m×n Euclidean vector space o
- Page 715 and 716: 715 H − H + ∂H ∂H ∂H −
- Page 717 and 718: 717 O O sort-index matrix order of
- Page 719 and 720: (x,y) angle between vectors x and y
- Page 721 and 722: Bibliography [1] Suliman Al-Homidan
- Page 723 and 724: BIBLIOGRAPHY 723 [24] Alexander I.
- Page 725 and 726: BIBLIOGRAPHY 725 [52] Stephen Boyd,
- Page 727 and 728: BIBLIOGRAPHY 727 [78] Frank Critchl
- Page 729 and 730: BIBLIOGRAPHY 729 [105] Richard L. D
- Page 731 and 732: BIBLIOGRAPHY 731 [132] Michel X. Go
- Page 733 and 734: BIBLIOGRAPHY 733 [162] T. Herrmann,
- Page 735 and 736: BIBLIOGRAPHY 735 [191] Mark Kahrs a
- Page 737 and 738: BIBLIOGRAPHY 737 [220] K. V. Mardia
- Page 739 and 740: BIBLIOGRAPHY 739 [250] Pythagoras P
- Page 741 and 742: BIBLIOGRAPHY 741 [277] Anthony Man-
- Page 743 and 744: BIBLIOGRAPHY 743 [306] Michael W. T
- Page 745 and 746: [333] Margaret H. Wright. The inter
- Page 747 and 748: Index 0-norm, 203, 261, 294, 296, 2
- Page 749 and 750: INDEX 749 product, 43, 92, 147, 254
700 APPENDIX E. PROJECTION<br />
b<br />
H 1<br />
x 22<br />
x 21<br />
H 2<br />
x 12<br />
H 1 ∩ H 2<br />
x 11<br />
Figure 148: H 1 and H 2 are the same halfspaces as in Figure 143.<br />
Dykstra’s alternating projection algorithm generates the alternations<br />
b, x 21 , x 11 , x 22 , x 12 , x 12 ... x 12 . The path illustrated from b to x 12 in R 2<br />
terminates at the desired result, Pb in Figure 143. The {y ki } correspond<br />
to the first two difference vectors drawn (in the first iteration i=1), then<br />
oscillate between zero and a negative vector thereafter. These alternations<br />
are not so robust in presence of noise as for the example in Figure 142.<br />
E.10.3<br />
<strong>Optimization</strong> and projection<br />
Unique projection on the nonempty intersection of arbitrary convex sets to<br />
find the closest point therein is a convex optimization problem. The first<br />
successful application of alternating projection to this problem is attributed<br />
to Dykstra [105] [54] who in 1983 provided an elegant algorithm that prevails<br />
today. In 1988, Han [152] rediscovered the algorithm and provided a<br />
primal−dual convergence proof. A synopsis of the history of alternating<br />
projection E.21 can be found in [56] where it becomes apparent that Dykstra’s<br />
work is seminal.<br />
E.21 For a synopsis of alternating projection applied to distance geometry, see [306,3.1].