v2007.09.13 - Convex Optimization
v2007.09.13 - Convex Optimization v2007.09.13 - Convex Optimization
There is a great race under way to determine which important problemscan be posed in a convex setting. Yet, that skill acquired by understandingthe geometry and application of Convex Optimization will remain more anart for some time to come; the reason being, there is generally no uniquetransformation of a given problem to its convex equivalent. This means, tworesearchers pondering the same problem are likely to formulate the convexequivalent differently; hence, one solution is likely different from the otherfor the same problem. Any presumption of only one right or correct solutionbecomes nebulous. Study of equivalence, sameness, and uniqueness thereforepervade study of Optimization.Tremendous benefit accrues when an optimization problem can betransformed to its convex equivalent, primarily because any locally optimalsolution is then guaranteed globally optimal. Solving a nonlinear system,for example, by instead solving an equivalent convex optimization problemis therefore highly preferable. 0.1 Yet it can be difficult for the engineer toapply theory without an understanding of Analysis.These pages comprise my journal over a seven year period bridginggaps between engineer and mathematician; they constitute a translation,unification, and cohering of about two hundred papers, books, and reportsfrom several different fields of mathematics and engineering. Beaconsof historical accomplishment are cited throughout. Much of what iswritten here will not be found elsewhere. Care to detail, clarity, accuracy,consistency, and typography accompanies removal of ambiguity andverbosity. Consequently there is much cross-referencing and backgroundmaterial provided in the text, footnotes, and appendices so as to beself-contained and to provide understanding of fundamental concepts.−Jon DattorroStanford, California20070.1 That is what motivates a convex optimization known as geometric programming[46, p.188] [45] which has driven great advances in the electronic circuit design industry.[27,4.7] [180] [288] [291] [67] [129] [137] [138] [139] [140] [141] [142] [192] [193] [201]8
Convex Optimization&Euclidean Distance Geometry1 Overview 192 Convex geometry 332.1 Convex set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.2 Vectorized-matrix inner product . . . . . . . . . . . . . . . . . 452.3 Hulls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532.4 Halfspace, Hyperplane . . . . . . . . . . . . . . . . . . . . . . 592.5 Subspace representations . . . . . . . . . . . . . . . . . . . . . 732.6 Extreme, Exposed . . . . . . . . . . . . . . . . . . . . . . . . . 762.7 Cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 812.8 Cone boundary . . . . . . . . . . . . . . . . . . . . . . . . . . 902.9 Positive semidefinite (PSD) cone . . . . . . . . . . . . . . . . . 972.10 Conic independence (c.i.) . . . . . . . . . . . . . . . . . . . . . 1202.11 When extreme means exposed . . . . . . . . . . . . . . . . . . 1262.12 Convex polyhedra . . . . . . . . . . . . . . . . . . . . . . . . . 1262.13 Dual cone & generalized inequality . . . . . . . . . . . . . . . 1343 Geometry of convex functions 1833.1 Convex function . . . . . . . . . . . . . . . . . . . . . . . . . . 1843.2 Matrix-valued convex function . . . . . . . . . . . . . . . . . . 2143.3 Quasiconvex . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2203.4 Salient properties . . . . . . . . . . . . . . . . . . . . . . . . . 2229
- Page 1 and 2: DATTORROCONVEXOPTIMIZATION&EUCLIDEA
- Page 3 and 4: Convex Optimization&Euclidean Dista
- Page 5 and 6: for Jennie Columba♦Antonio♦♦&
- Page 7: PreludeThe constant demands of my d
- Page 11 and 12: CONVEX OPTIMIZATION & EUCLIDEAN DIS
- Page 13 and 14: List of Figures1 Overview 191 Orion
- Page 15 and 16: LIST OF FIGURES 1559 Quadratic func
- Page 17 and 18: LIST OF FIGURES 17E Projection 5791
- Page 19 and 20: Chapter 1OverviewConvex Optimizatio
- Page 21 and 22: ˇx 4ˇx 3ˇx 2Figure 2: Applicatio
- Page 23 and 24: 23Figure 4: This coarsely discretiz
- Page 25 and 26: ases (biorthogonal expansion). We e
- Page 27 and 28: 27Figure 7: These bees construct a
- Page 29 and 30: its membership to the EDM cone. The
- Page 31 and 32: 31appendicesProvided so as to be mo
- Page 33 and 34: Chapter 2Convex geometryConvexity h
- Page 35 and 36: 2.1. CONVEX SET 35Figure 9: A slab
- Page 37 and 38: 2.1. CONVEX SET 372.1.6 empty set v
- Page 39 and 40: 2.1. CONVEX SET 392.1.7.1 Line inte
- Page 41 and 42: 2.1. CONVEX SET 41(a)R 2(b)R 3(c)(d
- Page 43 and 44: 2.1. CONVEX SET 43This theorem in c
- Page 45 and 46: 2.2. VECTORIZED-MATRIX INNER PRODUC
- Page 47 and 48: 2.2. VECTORIZED-MATRIX INNER PRODUC
- Page 49 and 50: 2.2. VECTORIZED-MATRIX INNER PRODUC
- Page 51 and 52: 2.2. VECTORIZED-MATRIX INNER PRODUC
- Page 53 and 54: 2.3. HULLS 53Figure 12: Convex hull
- Page 55 and 56: 2.3. HULLS 55Aaffine hull (drawn tr
- Page 57 and 58: 2.3. HULLS 57The union of relative
There is a great race under way to determine which important problemscan be posed in a convex setting. Yet, that skill acquired by understandingthe geometry and application of <strong>Convex</strong> <strong>Optimization</strong> will remain more anart for some time to come; the reason being, there is generally no uniquetransformation of a given problem to its convex equivalent. This means, tworesearchers pondering the same problem are likely to formulate the convexequivalent differently; hence, one solution is likely different from the otherfor the same problem. Any presumption of only one right or correct solutionbecomes nebulous. Study of equivalence, sameness, and uniqueness thereforepervade study of <strong>Optimization</strong>.Tremendous benefit accrues when an optimization problem can betransformed to its convex equivalent, primarily because any locally optimalsolution is then guaranteed globally optimal. Solving a nonlinear system,for example, by instead solving an equivalent convex optimization problemis therefore highly preferable. 0.1 Yet it can be difficult for the engineer toapply theory without an understanding of Analysis.These pages comprise my journal over a seven year period bridginggaps between engineer and mathematician; they constitute a translation,unification, and cohering of about two hundred papers, books, and reportsfrom several different fields of mathematics and engineering. Beaconsof historical accomplishment are cited throughout. Much of what iswritten here will not be found elsewhere. Care to detail, clarity, accuracy,consistency, and typography accompanies removal of ambiguity andverbosity. Consequently there is much cross-referencing and backgroundmaterial provided in the text, footnotes, and appendices so as to beself-contained and to provide understanding of fundamental concepts.−Jon DattorroStanford, California20070.1 That is what motivates a convex optimization known as geometric programming[46, p.188] [45] which has driven great advances in the electronic circuit design industry.[27,4.7] [180] [288] [291] [67] [129] [137] [138] [139] [140] [141] [142] [192] [193] [201]8