UNIT 1 – Non Linear Programming

Dr. Santosh Kumar, Assistant Professor

Faculty of Management Studies

University of Delhi
 Introduction: Non-Linear
Programming
• Non-Linear Programming Overview

• Non-Linear Programming with Equality Constraints-Lagrange

Multipliers Method

• Necessary and sufficient conditions when concavity

(convexity) of objective function is not known, with single
equality constraint

• Non-linear programming with inequality constraints-Karush

Kuhn Tucker Conditions.
 Non-Linear Programming
• The objective of topic is to help learn the formulation of nonlinear programming models
and of presenting some of their applications in the industrial engineering and
management domain.

• In this unit we are formulating different prototype nonlinear programming problems,

and of modelling multivariant and multimodel functions with inequality constraints by
the Kuhn-Tucker conditions.

• Specifically, multi-modal and multi-variable problems with inequality constraints are

modelled. The solution is done by applying the Kuhn-Tucker conditions. It sets out
different non-linear programming problems with their solutions in relation to Industrial
Organisation Engineering and the management setting.
Non-Linear Programming
• A non-linear programming problem occurs when decision variables are continuous and
objective function, or any of the constraints, is not linear. In such problems, the feasible
solution region is not convex and the solution may be found from within. Besides, the
direction that the objective function shifts in cannot always take increasing or decreasing
values.

• In the solution for non-linear programming models, the fundamental results originate from
the development of mathematical calculation dating back to the eighteenth century, and the
basic concept is Lagrangian method .
• The characterisation of the optimality conditions required in restricted problems is
generalised from the Lagrange results in the well-known Kuhn and Tucker Theorem (1951),
which compiles and structures a set of investigations carried out by many authors in the
1940s, among which John (1948) is cited.
• Non-linear programming progressed in the 1960s and 1970s, and it was possible to tackle
medium-sized problems with several tens of constraints and a few hundred variables. Yet
research into the search for efficient algorithms actively continued because those that
existed werenot altogether satisfactory. Regarding interior algorithms, Karmarkar (1984)
stands out. These algorithms are an alternative to the traditional, well-known Simplex;
notwithstanding, it is quite proper to frame it in the non-linear optimization line where it may
be considered a particular case of penalty algorithms.
TYPE 1: Non-Linear Programming with Equality
Constraints-Lagrange Multipliers Method
Consider the problem of optimizing a continuous and differentiable function subject to equality constraints. That
is:

There are various methods for solving the above-defined problem. But in this section, we shall discuss
Only two methods: the Lagrange Multipliers Method.
Non-Linear Programming with Equality
Constraints-Lagrange Multipliers Method
In this method an additional variable in each of the given constraints is added. Thus, if the problem has n
variables and m equality constraints, then m additional variables are to be added so that the problem would have
n + m variables.
Hence the necessary conditions given for the solution of Eqs (1) and (2) can be obtained very easily by forming
a function L, called the Lagrange function, as:
Example 1: Obtain the optimum solution to the following problem:
Then the sufficient condition for an extreme point x to be a local minimum (or local
maximum) of f (x) subject to the constraints gi (x) = 0 (i = 1, 2, . . ., m)
Example 2. Solve the following problem by using the
method of Lagrangian multipliers.
Continued: To see that this solution corresponds to the minimum of Z, apply
the sufficient condition with the help of a matrix:
TYPE 2 : Necessary and sufficient conditions when concavity (convexity) of objective
function is not known, with single equality constraint
CONTINUED….
Example 3. Use the method of Lagrangian multipliers to solve the following NLP problem.
Does the solution maximize or minimize the objective function?
CONTINUED…..
Type 3: Non-linear programming with
inequality constraints-KKT Conditions
In this section the necessary and sufficient conditions for a local optimum of the general non-linear programming
problem, with both equality and inequality constraints will be derived. The Kuhn-Tucker conditions (necessary as well
as sufficient) will be used to derive optimality conditions. Consider the following general non-linear LP problem:
The new problem is the constrained multivariable optimization problem with equality constraints with
n + m variables. Thus, it can be solved by using the Lagrangian multiplier method. For this, let us form
the Lagrangian function as:

The necessary conditions for an extreme point to be local optimum (max or min) can be obtained by: solving
the following equations:
The applied solution method
is as follows:
Step 1. Verify that f and gi have partial continuous first-order derivatives.

Step 2. Construct the Lagrangian function.

Step 3. Form the system of algebraic equations and in equations (Kuhn-Tucker

conditions). Establish the solutions (x*, k*) of the system (Kuhn-Tucker points).

Step 4. If f is convex (concave) and gi are convex, then points x* are global minimums
(maximums). In another case, each solution must be examined individually.
Example 4. Find the optimum value of the objective function when separately
subject to the following three sets of constraints:
Solution(a):
Continued…..
Solution(b):
Solution(c):
Case 1: Selecting Investments
The manager of the firm TRONIC must decide how to
distribute an investment of 100 million dollars for the
following year among four concepts: R&D, Advertising,
Technological Resources, and Human Resources.
For all four concepts, his consultant has established that
the profit after 1 year of investing X million dollars is
obtained according to the formulation below (See
following Table ).
Profits are to be obtained according to the
amounts invested
Case 1: Selecting Investments
continued…
In accordance with the collective bargaining agreement
signed 1 month ago, the minimum amount to be invested
in Human Resources (promotions, bonuses, etc.) over the
next year is 20 million dollars.

When analyzing the formulae, it may be deduced that,

although important investments in Technological
Resources offer good investment revenue, when
investments are small or moderate, the investment
revenue offered by R&D or advertising is significantly
higher. However, due to pressure from competitors, the
investment made in Technological Resources should be
such that the contribution of this concept to the
Case 1: Selecting Investments
continued…
(a)Model the problem

(b) Consider the Kuhn-Tucker conditions for this problem

(c) Solve the problem

(d) Interpret the result.

Solution:-
(a)Model the problem
Decision variables:
: million monetary units invested in R&D, Advertising, Technological
Resources and Human Resources, respectively, where
Objective function:

Constraints:
(b) Consider the Kuhn-Tucker
conditions for this problem
Lagrangian function:

The Kuhn-Tucker conditions:

Kuhn-Tucker conditions for this
problem continued…
(c) Solve the problem
By assuming that the Human Resources condition is saturated, we obtain
=20
By assuming that the Technological Resources condition is saturated, we
obtain
With these values and based on the described equations, we obtain:
To check whether the solution found is an optimum solution, we
obtain the Hessian Matrix of f in the solution:
Second derivative: