29.12.2013 Views

Chapter 3 Quadratic Programming

Chapter 3 Quadratic Programming

Chapter 3 Quadratic Programming

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.

Optimization I; <strong>Chapter</strong> 3 62<br />

where<br />

A −1 =<br />

( 0 A<br />

−1<br />

11<br />

A −T<br />

11 −A −T<br />

11 B 11 A −1<br />

11<br />

)<br />

. (3.23)<br />

We assume that Ã11 ∈ lR m 1×m 1<br />

is an easily invertible approximation of A 11 and<br />

define<br />

( )<br />

B11 Ã<br />

à =<br />

T 11<br />

. (3.24)<br />

à 11 0<br />

We remark that the inverse of à is given as in (3.23) with A −1<br />

11 , A −T<br />

11 replaced<br />

by Ã−1 11 and Ã−T 11 , respectively.<br />

We introduce the right transform<br />

( )<br />

I − Ã<br />

K R =<br />

−1 B T<br />

, (3.25)<br />

0 I<br />

which gives rise to the regular splitting<br />

( ) ( Ã 0 (I − A<br />

KK R =<br />

B<br />

} {{ ˜S<br />

Ã<br />

−<br />

−1 )Ã (−I + )<br />

AÃ−1 )B T<br />

, (3.26)<br />

0 0<br />

} } {{ }<br />

=: M 1 =: M 2 ∼ 0<br />

where<br />

˜S := D − BÃ−1 B T . (3.27)<br />

Given a start iterate ψ (0) = (ψ (0)<br />

1 , ψ (0)<br />

2 ) T , we solve the KKT system (3.20) by<br />

the transforming null-space iterations<br />

ψ (k+1) = ψ (k) + K R M1 −1 (α − Kψ (k) ) = (3.28)<br />

= (I − K R M1 −1 K)ψ (k) + K R M1 −1 α , k ≥ 0 .<br />

3.5 Active set strategies for convex QP problems<br />

3.5.1 Primal active set strategies<br />

We consider the constrained QP problem<br />

minimize f(x) := 1 2 xT Bx − x T b (3.29a)<br />

over x ∈ lR n<br />

subject to Cx = c (3.29b)<br />

Ax ≤ d ,<br />

(3.29c)

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

Saved successfully!

Ooh no, something went wrong!