Professional Documents
Culture Documents
y Optimization
p
Methods
Z. P. Jiang
ECE Department
D t t Rm LC214
R
NYU-Poly. ext. 3646
eeweb.poly.edu/faculty/jiang
Spring 2008 Z.P. Jiang. No reproduction permitted without written permission of the author.
* calculus
l l off variations
i ti
* dynamic
y programming
p g g
• Balance
B l b
between
t th
theory and
d applications
li ti
• Numerical algorithms
! Linear Equations
$ Nonlinear Programming
1/25/2010 ECE Dept at Poly 4
!Linear Equations
Goal: a first pass to data fitting
Given three points: (2, 1), (3, 6) and (5, 4), find the
corresponding quadratic function
with unknown xi , 1 # i # 3
Objective:
j maximize the weekly
y revenue
Bookshelf 10 2 100
W. Drawers 25 8 200
Custom 20 12 400
Only 5000 units of wood and 1500 units of labor are available.
available
1/25/2010 ECE Dept at Poly 8
LP Model
Solve the following linear programming problem
with constraints:
%
ei ! mi ' x1 " x2ti " x3ti 2
&
1# i # N, N $ 3
1/25/2010 ECE Dept at Poly 11
Formulation of the Problem
Mathematically, we want to solve
N N
w1 w2
?
w3 w4
2 # x3 # 4, ' 3 # y3 # '1
6 # x4 # 8,
8 ' 2 # y4 # 2 .
1/25/2010 ECE Dept at Poly 15
Comment
Optimization problems may be
% Nonlinear with constraints
• Basic
B i N Notions
ti off Optimality
O ti lit
• Convexity
• Taylor Series
• Rates of Convergence
1/25/2010 ECE Dept at Poly 21
Basic Notions of Optimality
Consider an optimization problem
minimize f ( x) P ! f ( x) p
performance index
x-S
gi ( x) ! 0, i - I ./
subject
bj t tto 0 constraints
t i t
g j ( x) $ 0, j - J /1
It is worth noting that
maximize f ( x) ! ' minimize ' f ( x)
x-S x-S
1/25/2010 ECE Dept at Poly 22
Feasibility
• A point satisfying all the constraints is said to
be feasible.
• Feasible set S: all feasible points.
p
• Active constraints: when g j ( x ) ! 0.
• Inactive constraints: when g j ( x ) 2 0.
0
• Boundary: The set of feasible points at which
at least
east one
o e inequality
equa ty iss active.
act e
• Interior points: All other points than boundary
Points:
A = (0, 0, 2)
X1 B = (3, 0, 1)
C = (1, 1, 1).
1/25/2010 ECE Dept at Poly 24
Minimizers
• Global minimizer x3 if f (x3 ) # f (x), 4x - S.
• Strict global minimizer if, additionally,
• Convex function f, if
min f ( x)
x-S
For example,
example such a set S can be defined via concav
concavee functions gi ,
: ;
S ! x - "n | gi ( x) $ 0 .
1-variable:
1 2 pn (n)
f (x0 " p) ! f (x0 ) " pf '((x0 ) " p f "((x0 ) "#" f (x0 ) "#
2 n!
n-variables:
1 T 2
f (x0 " p) ! f (x0 ) " p <f (x0 ) " p < f (x0 ) p "#
T
2
1/25/2010 ECE Dept at Poly 30
Notation:
otat o
Gradient <f ( x) :
T
= 8f ( x) 8f ( x) 8f ( x) >
<f ( x ) ! ? , ,# , @ -"
n
= 8 2
f ( x ) > nC n
< f ( x) ! ?
2
@ -" .
? 8x 8x @
A i j B
1/25/2010 ECE Dept at Poly 31
Example
f ( x) ! x " 3 x1 x2 ' x
2
1
2
2
<f ( x0 ) ! 0.
0
< f ( x 0 ) ! 0,
0
< f ( x 0 ) 2 0.
2
• How
H ffastt does
d it converge?
?
• Linear convergence if r = 1.
1) 0 < C < 1, the sequence converges.
2) C > 1,
1 the sequence diverges.
diverges
• Quadratic convergence if r = 2.
1/25/2010 ECE Dept at Poly 38
A Brief Introduction to Lagrange
M l i li M
Multiplier Method
h d
Lagrange multiplier theory essentially targets
at constrained optimization problems. For example,
max P ! f 0 % x & , x - " n
j to the constraint c ! f1 ( x).
subject )
To this end
end, introduce the augmented performance index
Pa ! f 0 % x & " DT % c ' f1 ( x) &
ith D the
with th vector
t off Lagrange
L multipliers.
lti li
1/25/2010 ECE Dept at Poly 39
A Brief Introduction to Lagrange
M l i li M
Multiplier Method
h d
Then, necessary conditions for optimality are
8Pa
!0
8x
and
c ! f1 ( x).
For more details, wait for
the lectures on "nonlinear
nonlinear programming
programming".
1/25/2010 ECE Dept at Poly 40