IE417: Nonlinear Programming: Lecture 15
Jeﬀ Linderoth
Department of Industrial and Systems Engineering
Lehigh University
28th March 2006
Jeﬀ Linderoth
IE417:Lecture 15
Variational Inequality?
Do you remember Nicholas Steir’s Talk?
He used
First Order Conditions
!
He also mentioned
variational inequalities
.
Let
Ω
⊂
R
n
be a closed convex set, and consider the
function/mapping
f
: Ω
→
R
n
. The
variational inequality
problem
involves ﬁnding a vector
x
*
∈
R
n
such that
F
(
x
*
)
T
(
x

x
*
)
≥
0
∀
x
∈
Ω
What does this have to do with optimization? If
x
*
is a local
minimizer of
f
in
Ω
, then
x
*
solves the following VIP:
∇
f
(
x
*
)
T
(
x

x
*
)
≥
0
∀
x
∈
Ω
Recall normal cone, for
x
∈
Ω
N
Ω
(
x
) =
{
v

v
T
(
w

x
)
≤
0
∀
w
∈
Ω
}
Geometrically, you want that
∇
f
(
x
)
∈
N
(
x
)
Jeﬀ Linderoth
IE417:Lecture 15
Loose Categories
min
x
∈
R
n
f
(
x
)
subject to
c
i
(
x
) = 0
∀
i
∈ E
,
c
i
(
x
)
≥
0
∀
i
∈ I
1
Quadratic Programming:
f
(
x
)
def
=
x
T
Qx
+
d
T
i
x
+
b
c
i
(
x
)
def
=
a
T
i
x
+
b
i
∀
i
∈ E ∪ I
2
Bound Constraint Optimization:
c
i
(
x
)
def
=
x
i
+
b
i
∀
i
∈ I
,
(
E
=
∅
)
3
Convex Programming:
f
(
x
)
is convex,
c
i
(
x
)
is linear for all
i
∈ E
, and
c
i
(
x
)
is concave for all
i
∈ I
Jeﬀ Linderoth
IE417:Lecture 15
Loose Categories: Solution Methods
1
Penalty Function: (Say for
I
=
∅
)
f
(
x
) +
1
2
μ
±
i
∈E
c
2
i
(
x
)
2
Barrier Method: (Say for
E
=
∅
)
f
(
x
) +
μ
±
i
∈I
log
c
i
(
x
)
3
Augmented Lagrangian: (Say for
I
=
∅
)
L
A
(
x, λ, μ
) =
f
(
x
)

±
i
∈E
λ
i
c
i
(
x
) +
1
2
μ
±
i
∈E
c
2
i
(
x
)
4
Sequential Quadratic Programming. Iterate via solving a
sequence of
quadratic
approximations to the Lagrangian that
satisfy linear approximation to constraints
Jeﬀ Linderoth
IE417:Lecture 15
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
This is the end of the preview.
Sign up
to
access the rest of the document.
 Spring '08
 Linderoth
 Optimization, Quadratic programming, Jeff Linderoth, Jeff Linderoth IE417

Click to edit the document details