Scientific Computing Notes 10-17

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Scientific Computing Notes: 10-19

Quadratic Programming:
Recall the basic quadratic problem:
min{x~T G~x + g~T ~x
such that : AT ~x = ~b AT ~x ~b = ~0
We can apply a Lagrangian in order to simplify this problem into something
more manageable:
L(~x, ~) = 12 ~xT G~x + ~g T ~x T (AT ~x ~b)
We next take the gradient for the Lagrangian in terms of x and in terms of
:
x L(~x, ~) = G~x + ~g A~
L(~x, ~) = AT ~x ~b
We now transform this into one system:
G~x + ~g A~ = 0
AT ~x ~b = 0
Which can be phrased as the system:


G
~
A ~x g

~ = ~
AT

0

b
We
the transpose
matrix above to get:

take
of the left-most
T T
G

GT
A
(A
)


=
(A)T
0
0T AT
GT = G - G must be symetric. Why?




1 T 1 2
= xT Gx
Consider min{ 2 x
1 3
G is not necessaarily symetric:
xT Gx = (xT Gx)T = xT GT x
Let H = .5 (G + GT )

Then xT Hx = xT (.5 (G + GT ))x = .5 xT Gx + .5xT GT x


= .5xT Gx + .5xT Gx
xT Gx
If we check this with the example that he mentioned a minute ago, w see that
we end up with a result which is symetric.
1

When solving these sorts of problems:


1. Acknowledge all constraints, variables.
2. Phrase the difference constraints in terms of = 0

x1
x2
x1
x2

x
x2
3. Construct system of equations in the form:
1

L1 (x, )
L2 (x, )

0
0

0

4
2
4. Solve the system of linear equations.

x3 ...
x3 ...

x3 ...
=

You might also like