Transcript PowerPoint
Introduction to Multivariate
Optimality
Lecture VI
Introduction to Multivariate
Optimality (I)
Development
of the Unconstrained
optimum
• The vector form of the Taylor Series Expansion
is
f ( x) f ( x ) x f ( x )dx dx' f ( x )dx
*
*
1
2
2
xx
*
Introduction to Multivariate
Optimality (II)
• By similar arguments as discussed in the
univariate case we can then define
f ( x) f ( x ) x f ( x)dx dx' f ( x ) dx 0
*
1
2
2
xx
*
Introduction to Multivariate
Optimality (III)
Constrained
Multivariate Optimum
• The general problem of the constrained
multivariate optimum can be defined as
max f ( x )
x
st G ( x ) b
Introduction to Multivariate
Optimality (IV)
• Most of the problem in the constrained
optimum comes in defining a feasible
perturbation
– We start from a feasible point and use a Taylor
expansion of the multivariate constraint
G( x) G( x ) x G( x ) dx
*
*
Introduction to Multivariate
Optimality (V)
– A critical part of this discussion is the Jacobian
G1 ( x )
x1
G2 ( x )
x G ( x ) x1
G ( x)
m
x1
G1 ( x )
G1 ( x )
x2
xn
G2 ( x )
G2 ( x )
x2
xn
Gm ( x )
Gm ( x )
x2
xn
Introduction to Multivariate
Optimality (VI)
• Given the Taylor series expansion of the vector
equation, we can see that starting from a
feasible and stepping to another feasible point
involves solving the equation
G( x ) G( x ) x G( x ) dx 0
Introduction to Multivariate
Optimality (VI)
• Using this information to solve for a
perturbation which maintains feasibility
requires first splitting the dx vector up into an
m*m portion and a m*(n-m) portion
G
1
d x1
G2
0
d x2
G1d x1 G2 d x2 0
Introduction to Multivariate
Optimality (VII)
• Solving for the feasible change
1
1
d x1 G G2 d x2
Introduction to Multivariate
Optimality (VIII)
• Returning to the original unconstrained
objective function and using our familiar Taylor
series expansion
f ( x) f x * x f ( x * ) d x 12 d x' 2x f ( x * ) d x 0
f
1
d x1 1
f2
2 d x1 d x2
d x2
F11
F
21
F12 d x1
0
F22 d x2
Introduction to Multivariate
Optimality (IX)
• Substituting for the feasible changes in x from
the expansion of the constraint matrix, we have
f1
G11 G2 d x2 1
1
f2
2 d x2G2G1
d x2
F11
d x2
F21
F12 G11 G2 d x2
0
F22
d x2
f1 G11 G2 d x2 f 2 d x2 0
f 2 f1 G11G2 d x2 0
f 2 f1 G11G2 0
The
second-order necessary conditions are
then defined by
d x G G
2
2
1
1
F11
d x2
F21
dx G G 1 F dx F
2 21
2 2 1 11
F12 G11G2 dx2
F22 dx2
1
G
1
1 G2 dx2
dx2 G2 G1 F12 dx2 F22
dx2
dx G G 1 F G 1G dx dx F G 1G dx dx G G 1 F dx dx F dx
2 21 1
2
2
2
2 1
12
2
2 22
2
2 2 1 11 1 2 2
dx2 G2G11 F11G11G2 F21G11G2 G2G11F12 F22 dx2