25.11.2015 Views

The MOSEK command line tool Version 7.0 (Revision 141)

The MOSEK command line tool. Version 7.0 ... - Documentation

The MOSEK command line tool. Version 7.0 ... - Documentation

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

26 CHAPTER 4. PROBLEM FORMULATION AND SOLUTIONS<br />

4.3 Semidefinite optimization<br />

Semidefinite optimization is an extension of conic quadratic optimization (see Section 4.2) allowing<br />

positive semidefinite matrix variables to be used in addition to the usual scalar variables. A semidefinite<br />

optimization problem can be written as<br />

minimize<br />

subject to l c i ≤<br />

n−1<br />

∑ ∑p−1<br />

〈 〉<br />

c j x j + Cj , X j + c<br />

f<br />

j=0<br />

n−1<br />

j=0<br />

p−1<br />

∑ ∑ 〈 〉<br />

a ij x j + Aij , X j<br />

j=0<br />

j=0<br />

≤ u c i, i = 0, . . . , m − 1<br />

lj x ≤ x j ≤ u x j , j = 0, . . . , n − 1<br />

x ∈ C, X j ∈ S r + j<br />

, j = 0, . . . , p − 1<br />

(4.10)<br />

where the problem has p symmetric positive semidefinite variables X j ∈ S r + j<br />

of dimension r j with<br />

symmetric coefficient matrices C j ∈ S rj and A i,j ∈ S rj . We use standard notation for the matrix inner<br />

product, i.e., for U, V ∈ R m×n we have<br />

〈U, V 〉 :=<br />

m−1<br />

∑<br />

i=0<br />

n−1<br />

∑<br />

U ij V ij .<br />

With semidefinite optimization we can model a wide range of problems as demonstrated in [2].<br />

j=0<br />

4.3.1 Duality for semidefinite optimization<br />

<strong>The</strong> dual problem corresponding to the semidefinite optimization problem (4.10) is given by<br />

maximize<br />

subject to<br />

(l c ) T s c l − (u c ) T s c u + (l x ) T s x l − (u x ) T s x u + c f<br />

c − A T y + s x u − s x l = s x n,<br />

m∑<br />

C j − y i A ij = S j , j = 0, . . . , p − 1<br />

i=0<br />

s c l − s c u = y,<br />

s c l , s c u, s x l , s x u ≥ 0,<br />

s x n ∈ C ∗ , S j ∈ S r + j<br />

, j = 0, . . . , p − 1<br />

(4.11)<br />

where A ∈ R m×n , A ij = a ij , which is similar to the dual problem for conic quadratic optimization (see<br />

Section 4.7), except for the addition of dual constraints<br />

m∑<br />

(C j − y i A ij ) ∈ S r + j<br />

.<br />

Note that the dual of the dual problem is identical to the original primal problem.<br />

i=0

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!