SDP_Duality

SDP_Duality - EE 227A: Convex Optimization and Applications...

Info iconThis preview shows pages 1–3. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: EE 227A: Convex Optimization and Applications October 14, 2008 Lecture 13: SDP Duality Lecturer: Laurent El Ghaoui Reading assignment: Chapter 5 of BV. 13.1 Direct approach 13.1.1 Primal problem Consider the SDP in standard form: p * := max X C,X : A i ,X = b i , i = 1 ,...,m, X , (13.1) where C,A i are given symmetric matrices, A,B = Tr AB denotes the scalar product between two symmetric matrices, and b ∈ R m is given. 13.1.2 Dual problem At first glance, the problem (13.1) is not amenable to the duality theory developed so far, since the constraint X 0 is not a scalar one. Minimum eigenvalue representation. We develop a dual based on a representation of the problem via the minimum eigenvalue, as p * = max X C,X : A i ,X = b i , i = 1 ,...,m, λ min ( X ) ≥ , (13.2) where we have used the minimum eigenvalue function of a symmetric matrix A , given by λ min ( A ) := min Y Y,A : Y , Tr Y = 1 (13.3) to represent the positive semi-definiteness condition X 0 in the SDP. The proof of the above representation of the minimum eigenvalue can be obtained by first showing that we can without loss of generality assume that A is diagonal, and noticing that we can then restrict Y to be diagonal as well. Note that the above representation proves that λ min is concave, so problem (13.2) is indeed convex as written. 13-1 EE 227A Lecture 13 — October 14, 2008 Fall 2008 Lagrangian and dual function. The Lagrangian for the maximization problem (13.2) is L ( X,λ,ν ) = C,X + m i =1 ν i ( b i- A i ,X ) + λ · λ min ( X ) = ν T b + C- m i =1 ν i A i ,X + λ · λ min ( X ) , where nu ∈ R m and λ ≥ 0 are the dual variables. The corresponding dual function g ( λ, ν ) := max X L ( X,λ,ν ) . involves the following subproblem, in which Z = C- ∑ m i =1 ν i A i and λ ≥ 0 are given: G ( λ, Z ) := max X Z,X + λλ min ( X ) . (13.4) For fixed λ ≥ 0, the function G ( · ,λ ) is the conjugate of the convex function- λλ min ....
View Full Document

This note was uploaded on 09/03/2011 for the course EE 227A taught by Professor Staff during the Spring '11 term at Berkeley.

Page1 / 6

SDP_Duality - EE 227A: Convex Optimization and Applications...

This preview shows document pages 1 - 3. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online