Chapter 3 Quadratic Programming
Chapter 3 Quadratic Programming
Chapter 3 Quadratic Programming
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 75<br />
We will have a closer look at the relationship between a minimizer of B (β) (x)<br />
and a point (x, µ) satisfying the KKT conditions for (3.62a)-(3.62b).<br />
If x(β) is a minimizer of B (β) (x), we obviously have<br />
∇ x B (β) (x(β)) = ∇ x Q(x(β)) +<br />
p∑<br />
i=1<br />
β<br />
d i − a T i x(β) a i = 0 . (3.72)<br />
Definition 3.9<br />
Perturbed complementarity<br />
The vector z (β) ∈ lR p with components<br />
z (β)<br />
i :=<br />
β<br />
d i − a T i x(β) , 1 ≤ i ≤ p (3.73)<br />
is called perturbed or approximate complementarity.<br />
The reason for the above definition will become obvious shortly.<br />
Indeed, in terms of the perturbed complementarity, (3.72) can be equivalently<br />
stated as<br />
∇ x Q(x(β)) +<br />
p∑<br />
i=1<br />
z (β)<br />
i a i = 0 . (3.74)<br />
We have to compare (3.74) with the first of the KKT conditions for (3.62a)-<br />
(3.62b) which is given by<br />
∇ x L(x, µ) = ∇ x Q(x) +<br />
p∑<br />
µ i a i = 0 . (3.75)<br />
i=1<br />
Obviously, (3.75) looks very much the same as (3.74).<br />
The other KKT conditions are as follows:<br />
a T i x − d i ≤ 0 , 1 ≤ i ≤ p , (3.76a)<br />
µ i ≥ 0 , 1 ≤ i ≤ p , (3.76b)<br />
µ i (a T i x − d i ) = 0 , 1 ≤ i ≤ p . (3.76c)<br />
Apparently, (3.76a) and (3.76b) are satisfied by x = x(β) and µ = z (β) .<br />
However, (3.76c) does not hold true, since it follows readily from (3.73) that<br />
z (β)<br />
i (d i − a T i x(β)) = β > 0 , 1 ≤ i ≤ p . (3.77)<br />
On the other hand, as β → 0 a minimizer x(β) and the associated z (β) come<br />
closer and closer to satisfying (3.76c). This is reason why z (β) is called perturbed<br />
(approximate) complementarity.