Non-Linear

Programming
Constrained optimization (Equality and Inequality constraints)
Constrained Non-linear Programming – General case

• The general non-linear programming problem (NLP) considering both equality and inequality constraints:

Where,

• First we will consider non-linear programming problem (NLP) considering only inequality constraints:

Where,
Optimality conditions (Inequality constraints)
Necessary conditions of optimality (Specific case - inequality) Eqn (3) implies that either
• If and means that the jth constraint is inactive and
hence can be ignored. In other words, the point X is
The inequality constraints can be transformed to
within the feasible region of , but not on the boundary
equality constraints by adding non-negative slack
of the constraints (i.e. )
variables, .
The problem becomes • If and , means that the jth constraint is active. In other
words, the point X is on the boundary of the constraint
, (i.e. )
Consider the division of constraints into two subsets, J 1
This problem can be easily solved by method of
Where, ,
and J2.
Lagrange multipliers. We can construct the Lagrange - Set J1 indicates the indices of those constraints that are
function as
active.
)=where =
- Set J2 indicates the indices of those constraints that are
inactive.
is a vector of Lagrange multipliers. The optimality conditions can

..(1) - J1 and J2 represents the total set of m constraints.

written as

..(2)
the Eqn (1) can be written as
..(3)
Or )
Or )
Assuming the first p constraints are active:

Above Eqn indicates that the gradient of the objective

function can be expressed as a linear combination of the
gradients of the active constraints at the optimum point.
Optimality conditions (Inequality constraints)

• , (
Optimality conditions (Inequality constraints)
𝜵𝐟

c1>c2>c3

• ,(
Optimality conditions (Inequality constraints)

Kuhn-Tucker necessary conditions of optimality:

=+
At local minimum point (X*):

(minimization case)

If . then is local minima

Kuhn-Tucker sufficient conditions of optimality:

= , where
Optimality conditions (Equality & Inequality constraints)

Where,
Kuhn-Tucker necessary conditions of optimality:
At local minimum point (X*): =++

(minimization case)

If . then is local minima

Kuhn-Tucker sufficient conditions of optimality:

= , where
Illustrative example-1
Case 1: =
Solving equation (5) and (6) we get , )=(, )
= doesn’t satisfy constraint (2), not a feasible point

Solution: Case 2: =0,

Due to constraint (4) and , we get = Solving equations
=, = (5) and (6) we get, , )=(, ) and . Since all the KT conditions
== are satisfied, , )=(, ) is a local minimum.
= is positive definite
==0
==1
, )=(, ) is a local minimum

++ = Case 3: ,
Due to constraint (3) and , we get = Solving equations
++ = (5) and (6) we get, , )=(, ) and , which violates the case-3
Kuhn-Tucker (KT) conditions: condition ( ).
…(1) Case 4: 0,
…(2) Due to constraints (3) and (4), 0 and , we get, , )=(, ).
Solving equations (5) and (6) we get, and which violates
…(3)
the case-4 conditions ( , )
…(4)
…(5)
…(6)
0, 0 …(7)
 Illustrative example-2
Case 1: =
Solving equations (7) (8) and (9) we get , , )=(0, 0, 0)
Subject to ; ; which doesn’t satisfy constraints (1) (2),& not a feasible
Solution: point
=, =, =
== =
Case 2:0, 0,
Solving the equations , and , we get , , )=(, , 50) .
== 0 =
Substituting in equations (7) (8)(9) and solving we get , , , which
== = satisfies all the KT conditions. , , )=(, , 50) is a candidate for local
== =
= is positive definite
minimum
++ =
++ = ., , )=(, , 50) is a local minimum
++ = Case 3: =0, =0,
Kuhn-Tucker (KT) conditions: Solving the equations (7) (8) (9) and (from eqn (6), since
…(1) ), we get , , )=(, , 60) and constraint (1) is not
satisfied, , , )=(, , 60) is a infeasible point
…(2)
…(3) Case 4: =0, 0,
…(4) Solving the equations (7) (8) (9) and (from eqn (5), since
…(5) ), we get , , )=(, , 0) and hich doesn’t satisfy constraints (1)
…(6) & not a feasible point
…(7)
Case 5: 0, 0,
…(8) Solving the equations (7) (8) (9) and (from eqn (4), since
), we get , , )=(, , 0) and hich doesn’t satisfy constraints (2)
& not a feasible point
…(9)
0, 0 , 0 …(10)
 Illustrative example-2
Case 6: =0, 0,
Solving the equations and , we get =50. Solving
Subject to ; ; equations (7) (8) and (9) we get , , )=(, , 50), , , which
Solution: doesn’t satisfy constraints (1) not a feasible point
=, =, =
== =
Case 7:0, 0,
Solving the equations and , we get =100. Solving
== 0 =
equations (7) (8) and (9) we get , , )=(, , 55), , , which
== = doesn’t satisfy constraints (2) not a feasible point
== =
++ = Case 8: 0, 0,
Solving the equations and , we get =50. From equation (9) we
++ = get =0, which doesn’t satisfy constraints (3) not a feasible
++ = point.
Kuhn-Tucker (KT) conditions: Only feasible local minimum for the problem is , , )=(, , 50)
…(1) with lagrange multipliers , , .
…(2)
…(3)
…(4)
…(5)
…(6)
…(7)
…(8)
…(9)
0, 0 , 0 …(10)
Illustrative example-3
Case 1: μ=0
From equation (6) we get λ=
Solving equations (5) and (6) we get
Substituting in equation (1) we get
Solution:
But substituting and in constraint (2), satisfies it.
== 1 μ=0 satisfies KT conditions and results in a local
==1 minimum feasible point.
==1
+λ +μ = = , which is positive definite
+λ +μ =
Kuhn-Tucker (KT) conditions:
…(1) , )=(, ) is a local minimizer
…(2)
Case 2: μ0
…(3)
Due to constraint (3) and μ0, + , which when considered
…(4) and solved along with eqn (1), gives , , which provides
…(5) μ=0, which violates the case-2 condition (μ0)
…(6)
Next session…
• Optimality conditions can solve only simple problems
• Algorithms for NLP-constrained optimization involving equality and
inequality constraints will be discussed