v2009.01.01 - Convex Optimization
v2009.01.01 - Convex Optimization v2009.01.01 - Convex Optimization
254 CHAPTER 4. SEMIDEFINITE PROGRAMMING with an equal number of twos and zeros along the main diagonal. Indeed, optimal solution (595) is a terminal solution along the central path taken by the interior-point method as implemented in [342,2.5.3]; it is also a solution of highest rank among all optimal solutions to (594). Clearly, rank of this primal optimal solution exceeds by far a rank-1 solution predicted by upper bound (245). 4.1.1.5 Later developments This rational example (594) indicates the need for a more generally applicable and simple algorithm to identify an optimal solution X ⋆ satisfying Barvinok’s Proposition 2.9.3.0.1. We will review such an algorithm in4.3, but first we provide more background. 4.2 Framework 4.2.1 Feasible sets Denote by C and C ∗ the convex sets of primal and dual points respectively satisfying the primal and dual constraints in (584), each assumed nonempty; ⎧ ⎡ ⎤ ⎫ ⎨ 〈A 1 , X〉 ⎬ C = ⎩ X ∈ Sn + | ⎣ . ⎦= b ⎭ = A ∩ Sn + 〈A m , X〉 (596) { } m∑ C ∗ = S ∈ S n + , y = [y i ]∈ R m | y i A i + S = C These are the primal feasible set and dual feasible set. Geometrically, primal feasible A ∩ S n + represents an intersection of the positive semidefinite cone S n + with an affine subset A of the subspace of symmetric matrices S n in isometrically isomorphic R n(n+1)/2 . The affine subset has dimension n(n+1)/2 −m when the A i are linearly independent. Dual feasible set C ∗ is the Cartesian product of the positive semidefinite cone with its inverse image (2.1.9.0.1) under affine transformation C − ∑ y i A i . 4.6 Both feasible 4.6 The inequality C − ∑ y i A i ≽0 follows directly from (584D) (2.9.0.1.1) and is known as a linear matrix inequality. (2.13.5.1.1) Because ∑ y i A i ≼C , matrix S is known as a slack variable (a term borrowed from linear programming [82]) since its inclusion raises this inequality to equality. i=1
4.2. FRAMEWORK 255 sets are closed and convex, and the objective functions on a Euclidean vector space are linear. Hence, convex optimization problems (584P) and (584D). 4.2.1.1 A ∩ S n + emptiness determination via Farkas’ lemma 4.2.1.1.1 Lemma. Semidefinite Farkas’ lemma. Given an arbitrary set {A i ∈ S n , i=1... m} and a vector b = [b i ]∈ R m , define the affine subset A = {X ∈ S n | 〈A i , X〉 = b i , i=1... m} (587) Primal feasible set A ∩ S n + is nonempty if and only if y T b ≥ 0 holds for each and every vector y = [y i ]∈ R m ∑ such that m y i A i ≽ 0. Equivalently, primal feasible set A ∩ S n + is nonempty if and only if y T b ≥ 0 holds for each and every norm-1 vector ‖y‖= 1 such that m∑ y i A i ≽ 0. ⋄ i=1 Semidefinite Farkas’ lemma follows directly from a membership relation (2.13.2.0.1) and the closed convex cones from linear matrix inequality example 2.13.5.1.1; given convex cone K and its dual where i=1 K = {A svec X | X ≽ 0} (337) m∑ K ∗ = {y | y j A j ≽ 0} (343) ⎡ A = ⎣ j=1 then we have membership relation and equivalents ⎤ svec(A 1 ) T . ⎦ ∈ R m×n(n+1)/2 (585) svec(A m ) T b ∈ K ⇔ 〈y,b〉 ≥ 0 ∀y ∈ K ∗ (288) b ∈ K ⇔ ∃X ≽ 0 A svec X = b ⇔ A ∩ S n + ≠ ∅ (597) b ∈ K ⇔ 〈y,b〉 ≥ 0 ∀y ∈ K ∗ ⇔ A ∩ S n + ≠ ∅ (598)
- Page 203 and 204: 3.1. CONVEX FUNCTION 203 k/m 1 0.9
- Page 205 and 206: k∑ i=1 3.1. CONVEX FUNCTION 205 S
- Page 207 and 208: 3.1. CONVEX FUNCTION 207 rather x >
- Page 209 and 210: 3.1. CONVEX FUNCTION 209 rather ] x
- Page 211 and 212: 3.1. CONVEX FUNCTION 211 3.1.6.0.2
- Page 213 and 214: 3.1. CONVEX FUNCTION 213 q(x) f(x)
- Page 215 and 216: 3.1. CONVEX FUNCTION 215 3.1.7.0.2
- Page 217 and 218: 3.1. CONVEX FUNCTION 217 We learned
- Page 219 and 220: 3.1. CONVEX FUNCTION 219 Since opti
- Page 221 and 222: 3.1. CONVEX FUNCTION 221 2 1.5 1 0.
- Page 223 and 224: 3.1. CONVEX FUNCTION 223 Setting th
- Page 225 and 226: 3.1. CONVEX FUNCTION 225 Similarly,
- Page 227 and 228: 3.1. CONVEX FUNCTION 227 For vector
- Page 229 and 230: 3.1. CONVEX FUNCTION 229 This means
- Page 231 and 232: 3.1. CONVEX FUNCTION 231 f(Y ) −
- Page 233 and 234: 3.2. MATRIX-VALUED CONVEX FUNCTION
- Page 235 and 236: 3.2. MATRIX-VALUED CONVEX FUNCTION
- Page 237 and 238: 3.2. MATRIX-VALUED CONVEX FUNCTION
- Page 239 and 240: 3.3. QUASICONVEX 239 exponential al
- Page 241 and 242: 3.3. QUASICONVEX 241 Unlike convex
- Page 243 and 244: 3.4. SALIENT PROPERTIES 243 6. (af
- Page 245 and 246: Chapter 4 Semidefinite programming
- Page 247 and 248: 4.1. CONIC PROBLEM 247 where K is a
- Page 249 and 250: 4.1. CONIC PROBLEM 249 4.1.1.2 Redu
- Page 251 and 252: 4.1. CONIC PROBLEM 251 In any SDP f
- Page 253: 4.1. CONIC PROBLEM 253 Proposition
- Page 257 and 258: 4.2. FRAMEWORK 257 4.2.1.1.3 Exampl
- Page 259 and 260: 4.2. FRAMEWORK 259 4.2.2 Duals The
- Page 261 and 262: 4.2. FRAMEWORK 261 When equality is
- Page 263 and 264: 4.2. FRAMEWORK 263 The pseudoinvers
- Page 265 and 266: 4.2. FRAMEWORK 265 For the data giv
- Page 267 and 268: 4.2. FRAMEWORK 267 minimizes an aff
- Page 269 and 270: 4.3. RANK REDUCTION 269 whose rank
- Page 271 and 272: 4.3. RANK REDUCTION 271 and where m
- Page 273 and 274: 4.3. RANK REDUCTION 273 4.3.3 Optim
- Page 275 and 276: 4.3. RANK REDUCTION 275 Initialize:
- Page 277 and 278: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 279 and 280: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 281 and 282: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 283 and 284: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 285 and 286: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 287 and 288: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 289 and 290: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 291 and 292: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 293 and 294: 4.4. RANK-CONSTRAINED SEMIDEFINITE
- Page 295 and 296: 4.5. CONSTRAINING CARDINALITY 295 m
- Page 297 and 298: 4.5. CONSTRAINING CARDINALITY 297 m
- Page 299 and 300: 4.5. CONSTRAINING CARDINALITY 299 a
- Page 301 and 302: 4.5. CONSTRAINING CARDINALITY 301 f
- Page 303 and 304: 4.5. CONSTRAINING CARDINALITY 303 n
4.2. FRAMEWORK 255<br />
sets are closed and convex, and the objective functions on a Euclidean vector<br />
space are linear. Hence, convex optimization problems (584P) and (584D).<br />
4.2.1.1 A ∩ S n + emptiness determination via Farkas’ lemma<br />
4.2.1.1.1 Lemma. Semidefinite Farkas’ lemma.<br />
Given an arbitrary set {A i ∈ S n , i=1... m} and a vector b = [b i ]∈ R m ,<br />
define the affine subset<br />
A = {X ∈ S n | 〈A i , X〉 = b i , i=1... m} (587)<br />
Primal feasible set A ∩ S n + is nonempty if and only if y T b ≥ 0 holds for<br />
each and every vector y = [y i ]∈ R m ∑<br />
such that m y i A i ≽ 0.<br />
Equivalently, primal feasible set A ∩ S n + is nonempty if and only<br />
if y T b ≥ 0 holds for each and every norm-1 vector ‖y‖= 1 such that<br />
m∑<br />
y i A i ≽ 0.<br />
⋄<br />
i=1<br />
Semidefinite Farkas’ lemma follows directly from a membership relation<br />
(2.13.2.0.1) and the closed convex cones from linear matrix inequality<br />
example 2.13.5.1.1; given convex cone K and its dual<br />
where<br />
i=1<br />
K = {A svec X | X ≽ 0} (337)<br />
m∑<br />
K ∗ = {y | y j A j ≽ 0} (343)<br />
⎡<br />
A = ⎣<br />
j=1<br />
then we have membership relation<br />
and equivalents<br />
⎤<br />
svec(A 1 ) T<br />
. ⎦ ∈ R m×n(n+1)/2 (585)<br />
svec(A m ) T<br />
b ∈ K ⇔ 〈y,b〉 ≥ 0 ∀y ∈ K ∗ (288)<br />
b ∈ K ⇔ ∃X ≽ 0 A svec X = b ⇔ A ∩ S n + ≠ ∅ (597)<br />
b ∈ K ⇔ 〈y,b〉 ≥ 0 ∀y ∈ K ∗ ⇔ A ∩ S n + ≠ ∅ (598)