0% found this document useful (0 votes)
4 views

Pages Used in Last Problem 1

The document discusses inequality constraints in optimization problems, focusing on necessary and sufficient conditions for minimization with both equality and inequality constraints. It introduces the Karush-Kuhn-Tucker conditions and explores first-order and second-order conditions for regular points of the constraints. Additionally, it addresses sensitivity results related to variations in constraints and their impact on solutions and Lagrange multipliers.
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
4 views

Pages Used in Last Problem 1

The document discusses inequality constraints in optimization problems, focusing on necessary and sufficient conditions for minimization with both equality and inequality constraints. It introduces the Karush-Kuhn-Tucker conditions and explores first-order and second-order conditions for regular points of the constraints. Additionally, it addresses sensitivity results related to variations in constraints and their impact on solutions and Lagrange multipliers.
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

11.

8 Inequality Constraints 341

Because by assumption x∗ is a regular point and Lx∗  is positive definite on M,


it follows that this matrix is nonsingular (see Exercise 11). Thus, by the Implicit
Function Theorem, there is a solution xc c to the system which is in fact
continuously differentiable.
By the chain rule we have

 c fxc =  x fx∗  c x0
c=0

and

 c hxc =  x hx∗  c x0
c=0

In view of (31), the second of these is equal to the identity I on E m , while this, in
view of (30), implies that the first can be written

 c fxc = −T 
c=0

11.8 INEQUALITY CONSTRAINTS


We consider now problems of the form

minimize fx
subject to hx = 0 (32)
gx  0

We assume that f and h are as before and that g is a p-dimensional function.


Initially, we assume f h g ∈ C 1 .
There are a number of distinct theories concerning this problem, based on
various regularity conditions or constraint qualifications, which are directed toward
obtaining definitive general statements of necessary and sufficient conditions. One
can by no means pretend that all such results can be obtained as minor extensions
of the theory for problems having equality constraints only. To date, however, these
alternative results concerning necessary conditions have been of isolated theoretical
interest only—for they have not had an influence on the development of algorithms,
and have not contributed to the theory of algorithms. Their use has been limited to
small-scale programming problems of two or three variables. We therefore choose
to emphasize the simplicity of incorporating inequalities rather than the possible
complexities, not only for ease of presentation and insight, but also because it is
this viewpoint that forms the basis for work beyond that of obtaining necessary
conditions.
342 Chapter 11 Constrained Minimization Conditions

First-Order Necessary Conditions


With the following generalization of our previous definition it is possible to parallel
the development of necessary conditions for equality constraints.
Definition. Let x∗ be a point satisfying the constraints

hx∗  = 0 gx∗   0 (33)

and let J be the set of indices j for which gj x∗  = 0. Then x∗ is said to be a
regular point of the constraints (33) if the gradient vectors hi x∗ , gj x∗ ,
1  i  m j ∈ J are linearly independent.
We note that, following the definition of active constraints given in
Section 11.1, a point x∗ is a regular point if the gradients of the active constraints
are linearly independent. Or, equivalently, x∗ is regular for the constraints if it is
regular in the sense of the earlier definition for equality constraints applied to the
active constraints.

Karush–Kuhn–Tucker Conditions. Let x∗ be a relative minimum point for the


problem

minimize fx
(34)
subject to hx = 0 gx  0

and suppose x∗ is a regular point for the constraints. Then there is a vector
 ∈ E m and a vector  ∈ E p with   0 such that

fx∗  + T hx∗  + T gx∗  = 0 (35)



 gx  = 0
T
(36)

Proof. We note first, since   0 and gx∗   0, (36) is equivalent to the statement
that a component of  may be nonzero only if the corresponding constraint is
active. This a complementary slackness condition, stating that gx∗ i < 0 implies

i = 0 and i > 0 implies gx i = 0.

Since x is a relative minimum point over the constraint set, it is also a relative
minimum over the subset of that set defined by setting the active constraints to zero.
Thus, for the resulting equality constrained problem defined in a neighborhood of
x∗ , there are Lagrange multipliers. Therefore, we conclude that (35) holds with

j = 0 if gj x  = 0 (and hence (36) also holds).
It remains to be shown that   0. Suppose k < 0 for some k ∈ J . Let S
and M be the surface and tangent plane, respectively, defined by all other active
constraints at x∗ . By the regularity assumption, there is a y such that y ∈ M and
11.8 Inequality Constraints 343

gk x∗ y < 0. Let xt be a curve on S passing through x∗ (at t = 0) with ẋ0 = y.
Then for small t  0, xt is feasible, and

df
xt = fx∗ y < 0
dt t=0

by (35), which contradicts the minimality of x∗ .

Example. Consider the problem

minimize 2x12 + 2x1 x2 + x22 − 10x1 − 10x2


subject to x12 + x22  5
3x1 + x2  6

The first-order necessary conditions, in addition to the constraints, are

4x1 + 2x2 − 10 + 2 1 x1 + 3 2 =0
2x1 + 2x2 − 10 + 2 1 x2 + 2 =0

1  0 2 0

1 x1 + x2 − 5 =0
2 2

2 3x1 + x2 − 6 = 0

To find a solution we define various combinations of active constraints and check


the signs of the resulting Lagrange multipliers. In this problem we can try setting
none, one, or two constraints active. Assuming the first constraint is active and the
second is inactive yields the equations

4x1 + 2x2 − 10 + 2 1 x1 =0
2x1 + 2x2 − 10 + 2 1 x2 =0
x12 + x22 = 5

which has the solution

x1 = 1 x2 = 2 1 = 1

This yields 3x1 + x2 = 5 and hence the second constraint is satisfied. Thus, since
1 > 0, we conclude that this solution satisfies the first-order necessary conditions.
344 Chapter 11 Constrained Minimization Conditions

Second-Order Conditions
The second-order conditions, both necessary and sufficient, for problems with
inequality constraints, are derived essentially by consideration only of the equality
constrained problem that is implied by the active constraints. The appropriate
tangent plane for these problems is the plane tangent to the active constraints.

Second-Order Necessary Conditions. Suppose the functions f g h ∈ C 2 and


that x∗ is a regular point of the constraints (33). If x∗ is a relative minimum
point for problem (32), then there is a  ∈ E m ,  ∈ E p ,   0 such that (35)
and (36) hold and such that

Lx∗  = Fx∗  + T Hx∗  + T Gx∗  (37)

is positive semidefinite on the tangent subspace of the active constraints at x∗ .


Proof. If x∗ is a relative minimum point over the constraints (33), it is also a
relative minimum point for the problem with the active constraints taken as equality
constraints.
Just as in the theory of unconstrained minimization, it is possible to formulate
a converse to the Second-Order Necessary Condition Theorem and thereby obtain a
Second-Order Sufficiency Condition Theorem. By analogy with the unconstrained
situation, one can guess that the required hypothesis is that Lx∗  be positive definite
on the tangent plane M. This is indeed sufficient in most situations. However, if
there are degenerate inequality constraints (that is, active inequality constraints
having zero as associated Lagrange multiplier), we must require Lx∗  to be positive
definite on a subspace that is larger than M.

Second-Order Sufficiency Conditions. Let f g h ∈ C 2 . Sufficient conditions


that a point x∗ satisfying (33) be a strict relative minimum point of problem
(32) is that there exist  ∈ E m ,  ∈ E p , such that

0 (38)

 gx  = 0
T
(39)
fx∗  + T hx∗  + T1 gx∗  = 0 (40)

and the Hessian matrix

Lx∗  = Fx∗  + T Hx∗  + T Gx∗  (41)

is positive definite on the subspace

M = y  hx∗ y = 0 gj x∗ y = 0 for all j ∈ J 

where

J = j  gj x∗  = 0 j >0 
11.8 Inequality Constraints 345

Proof. As in the proof of the corresponding theorem for equality constraints in


Section 11.5, assume that x∗ is not a strict relative minimum point; let yk be a
sequence of feasible points converging to x∗ such that fyk   fx∗ , and write each
yk in the form yk = x∗ + k sk with sk  = 1 k > 0. We may assume that k → 0
and sk → s∗ . We have 0  fx∗ s∗ , and for each i = 1     m we have

hi x∗ s∗ = 0

Also for each active constraint gj we have gj yk  − gj x∗   0, and hence

gj x∗ s∗  0

If gj x∗ s∗ = 0 for all j ∈ J , then the proof goes through just as in Section 11.5.
If gj x∗ s∗ < 0 for at least one j ∈ J , then

0  fx∗ s∗ = −T hx∗ s∗ − T gx∗ s∗ > 0

which is a contradiction.
We note in particular that if all active inequality constraints have strictly
positive corresponding Lagrange multipliers (no degenerate inequalities), then the
set J includes all of the active inequalities. In this case the sufficient condition is that
the Lagrangian be positive definite on M, the tangent plane of active constraints.

Sensitivity
The sensitivity result for problems with inequalities is a simple restatement of the
result for equalities. In this case, a nondegeneracy assumption is introduced so
that the small variations produced in Lagrange multipliers when the constraints are
varied will not violate the positivity requirement.
Sensitivity Theorem. Let f g h ∈ C 2 and consider the family of problems

minimize fx
subject to hx = c (42)
gx  d

Suppose that for c = 0, d = 0, there is a local solution x∗ that is a regular


point and that, together with the associated Lagrange multipliers,    0,
satisfies the second-order sufficiency conditions for a strict local minimum.
Assume further that no active inequality constraint is degenerate. Then for
every c d ∈ E m+p in a region containing 0 0 there is a solution xc d,
depending continuously on c d, such that x0 0 = x∗ , and such that xc d
is a relative minimum point of (42). Furthermore,

 c fxc d = −T (43)
00

 d fxc d = −T  (44)
00

You might also like