Linear Optimization

http://home.ubalt.edu/ntsbarsh/opre640a/partVIII.htm#rplp
LP Problem Formulation Process: To formulate an LP problem, I recommend using the
following guidelines after reading the problem statement carefully a few times.
Any linear program consists of four parts: a set of decision variables, the parameters, the
objective function, and a set of constraints. In formulating a given decision problem in
mathematical form, you should practice understanding the problem (i.e., formulating a
mental model) by carefully reading and re-reading the problem statement. While trying to
understand the problem, ask yourself the following general questions:
1. What are the decision variables? That is, what are controllable inputs? Define the
decision variables precisely, using descriptive names. Remember that the
controllable inputs are also known as controllable activities, decision variables,
and decision activities.

2. What are the parameters? That is, what are the uncontrollable inputs? These are
usually the given constant numerical values. Define the parameters precisely,
using descriptive names.

3. What is the objective? What is the objective function? Also, what does the owner
of the problem want? How the objective is related to his decision variables? Is it a
maximization or minimization problem? The objective represents the goal of the
decision-maker.

4. What are the constraints? That is, what requirements must be met? Should I use
inequality or equality type of constraint? What are the connections among
variables? Write them out in words before putting them in mathematical form.
Learn that the feasible region has nothing or little to do with the objective function (min
or max). These two parts in any LP formulation come mostly from two distinct and
different sources. The objective function is set up to fulfill the decision-maker's desire
(objective), whereas the constraints which shape the feasible region usually comes from
the decision-maker's environment putting some restrictions/conditions on achieving
his/her objective.
Page1/9

The Carpenter's Problem: Allocating Scarce Resources Among Competitive

Means
Maximize 9X1 + 6X2 [(1.5X1 + X2) + (2.5X1 + 2X2) + F1 + F2],
Maximize 5 X1 + 3 X2
Subject to:
2 X1 + X2 40 labor constraint
X1 + 2 X2 50 material constraint
and both X1, X2 are non-negative.
Graphical Solution Method: Procedure for Graphical Method of Solving LP Problems:
1. Is the problem an LP? Yes, if and only if:
All variables have power of 1, and they are added or subtracted (not divided or
multiplied). The constraint must be of the following forms ( , , or =, that is, the LPconstraints are always closed), and the objective must be either maximization or
minimization.
For example, the following problem is not an LP: Max X, subject to X 1. This very
simple problem has no solution.
2.Can I use the graphical method? Yes, if the number of decision variables is either 1 or 2
3.Use Graph Paper. Graph each constraint one by one, by pretending that they are
equalities (pretend all and , are = ) and then plot the line. Graph the straight line on a
system of coordinates on a graph paper. A system of coordinate has two axes: a horizontal
axis called the x-axis (abscissa), and a vertical axis, called the y-axis (ordinate). The axes
are numbered, usually from zero to the largest value expected for each variable.
4. As each line is created, divide the region into 3 parts with respect to each line. To
identify the feasible region for this particular constraint, pick a point in either side of the
line and plug its coordinates into the constraint. If it satisfies the condition, this side is
feasible; otherwise the other side is feasible. For equality constraints, only the points on
the line are feasible.
5. Throw away the sides that are not feasible.
After all the constraints are graphed, you should have a non-empty (convex) feasible
region, unless the problem is infeasible.
Page2/9

Unfortunately, some of the boundaries of the feasible regions described in your textbook
are wrong. See, e.g., the figures depicted on page 56. Almost all inequalities must be
changed to equality. Right?
6. Create (at least) two iso-value lines from the objective function, by setting the
objective function to any two distinct numbers. Graph the resulting lines. By
moving these lines parallel, you will find the optimal corner (extreme point), if it
does exist.
Know that, the LP constraints provide the vertices and the corner-points. A vertex is the
intersection of 2-lines, or in general n-hyperplanes in LP problems with n-decision
variables. A corner-point is a vertex that is also feasible.

Note: There is an alternative to the Iso-value objective function approach with problems
that have few constraints and a bounded feasible region. First, find all the corner points,
which are called extreme points. Then, evaluate the objective function at the extreme
points to find the optimal value and the optimal solution. Clearly, the carpenter has many
alternative sets of actions to take. However, the four "extreme" options are:
Page 3/9

Objective Function Value at Each Corner (i.e., Extreme) Point

Decision-Maker's Choices Corner Point Coordinates Net Income Function
Number of Tables or Chairs

X1, X2

5 X1 + 3 X2

Make No Table nor Chair

0, 0

Make All Tables You Can

20, 0

100

Make All Chairs You Can

0, 25

75

Make Mixed Products

10, 20

110

Since the objective is to maximize, from the above table we read off the optimal value to
be 110, which is obtainable if the carpenter follows the optimal strategy of X1 = 10, and
X2 = 20.
Notice that in the carpenter problem, the convex feasible region provides the corner
points with coordinates shown in the above table.
The main deficiency of the graphical method is that it is limited to solving LP problems
with 1 or 2 decision variables only. However, the main and useful conclusion we easily
observe from the graphical methods, is as follow:
If a linear program has a non-empty, bounded feasible region, then
the optimal solution is always one of the corner points.
The proof of this claim follows from the results of the following two facts:
Fact No. 1: The feasible region of any linear program is always a convex set.

Fact No. 2: The Iso-value of a linear program objective function is always a linear
function.
Page 4/9

Combining the above two facts, it follows that, if a linear program has a non-empty,
bounded feasible region, then the optimal solution is always one of the corner points.
To overcome the deficiency of the graphical method, we will utilize this useful and
practical conclusion in developing an algebraic method that is applicable to multidimensional LP problems.
The convexity of the feasible region for linear programs makes the LP problems easy to
solve. Because of this property and linearity of the objective function, the solution is
always one of the vertices. Moreover, since number of vertices is limited, one has to find
all feasible vertices, and then evaluate the objective function at these vertices to seek the
optimal point.
For nonlinear programs, the problem is much harder to solve, because the solution could
be anywhere inside the feasible region on the boundary of the feasible region, or at a
vertex.
Fortunately, most of the Business optimization problems have linear constraints,
which is why LP is so popular. There are well over 400 computer packages in the market
today solving LP problems. Most of them are based on vertex searching, that is, jumping
from one vertex to the neighboring one in search of an optimal point.
You have already noticed that, a graph of a system of inequalities and/or equalities is
called the feasible region. These two representations, graphical, and algebraic are
equivalent to each other, which means the coordinate of any point satisfying the
constraints is located in the feasible region, and the coordinate of any point in the feasible
region satisfies all the constraints.
Algebraic Method: As George Dantzig pointed out, linear programming is strictly "the
theory and solution of linear inequality systems." The basic solutions to a linear program
are the solutions to the systems of equations consisting of constraints at binding position.
Not all basic solutions satisfy all the problem constraints. Those that do meet all the
restrictions are called basic feasible solutions. The basic feasible solutions correspond
precisely to the extreme points of the feasible region.
Page5/9

For example, for Carpenter's Problem, one can compute all the basic solutions, by taking
any two of the equations and solving them simultaneously and then, using the constraints
of other equations to check for feasibility of this solution. If feasible, then this solution is
a basic feasible solution that provides the coordinates of a corner point of the feasible
region. To illustrate the procedure, consider the Carpenter's constraints at binding (i.e., all
with = sign) position:
2X1 + X2 = 40
X1 + 2X2 = 50
X1 = 0
X2 = 0
Here we have 4 equations with 2 unknowns. In terms of a "binomial coefficient", there
are at most C42 = 4! / [2! (4-2)!] = 6 basic solutions. Solving the six resultant systems of
equations, we have:
Six Basic Solutions with Four Basic Feasible Solutions
X1
10
0
20
0
50
0

X2 5X1 + 3X2
20
110*
40 infeasible
0
100
25
75
0 infeasible
0
0

Four of the above basic solutions are basic feasible solutions satisfying all the constraints,
belonging to the coordinates of the vertices of the bounded feasible region. By plugging
in the basic feasible solution in the objective function, we compute the optimal value.
Therefore, from the above table, we see that, the optimal solution is X1 = 10, X2 = 20,
with optimal value of $110. The above approach can be applied in solving higher
dimension LP problems provided the optimal solution is bounded.
Numerical Example: The Transportation Problem
Transportation models play an important role in logistics and supply chain management
for reducing cost and improving service. Therefore, the goal is to find the most cost
effective way to transport the goods.
Consider a model with 2 origins and 2 destinations. The supply and demand at each
origin (e.g; warehouse) O1, O2 and destination (e.g.; market) D1 and D2, together with
the unit transportation cost are summarized in the following table.
Page 6/9

The Unit Transportation

Cost Matrix
D1
D2
O1 20
30
O2 10
40
Demand 150
150

Supply
200
100
300

Let Xij's denote the amount of shipment from source i to destination j. The LP
formulation of the problem minimizing the total transportation cost is:
Min 20X11 + 30X12 + 10X21 + 40X22
subject to:
X11 + X12 = 200
X21 + X22 = 100
X11 + X21 = 150
X12 + X22 = 150
all Xij 0
Notice that the feasible region is bounded, therefore one may use the algebraic method.
Because this transportation problem is a balanced one (total supply = total demand) all
constraints are in equality form. Moreover, any one of the constraints is redundant
(adding any two constraints and subtracting another one, we obtain the remaining one).
Let us delete the last constraint. Therefore, the problem reduces to:
Min 20X11 + 30X12 + 10X21 + 40X22
subject to:
X11 + X12 = 200
X21 + X22 = 100
X11 + X21 = 150
all Xij 0
This LP problem cannot be solved by the graphical method. However, the algebraic
method has no limitation on the LP dimension. The constraints are already at binding
(equality) position.
Notice that we have m=3 equality constraints with (four implied non-negative) decision
variables. Therefore, out of these four variables, there is at most m=3 variables with
positive value and the rest must be at zero level. For example, by setting any one of the
variables in turn to zero, we get:
Page7/9

X11
0
200
150
50

X12
200
0
50
150

X21
150
-50
0
100

X22 Total Transportation Cost

-50
infeasible
150
infeasible
100
8500
0
6500*

Now by setting any one (or more) variables to zero, it is easy to see, by inspecting the
constraints that all other solutions are infeasible. Thus, from the above table, we obtain
the optimal strategy to be: X11 = 50, X12 = 150, X21 = 100, and X22 = 0, with the least
total transportation cost of $6,500.
You may like to run this problem using Module Net.Exe in your WinQSB Package to
check these results for yourself.
Notice that in the above example, there are m=3 constraints (excluding the non-negativity
conditions), and n=4 decision variables. The optimal shipment indicates that, the manager
should not send any shipment from one source to one destination. The optimal solution
consists of at most 3 positive decision variables which is equal to the number of
constrains. If the manager is shipping goods from every source to every destination, then
the result is not optimal.
The above results can be generalize, in the following main economic result:
Given an LP with m constraints (excluding the non-negativity conditions) constituting a
bounded feasible region, and n decision variables, if n m then at most m decision
variables have positive value at the optimal solution and the rest of (i.e., n-m) decision
variables must be set at zero level.
The above result follows from the fact that, using the shadow prices indicates that the
opportunity cost for the decision variable at zero level exceed its contribution.
Numerical Example: Find the optimal solution for the following production problem
with n=3 products and m=1 (resource) constraint:
Maximize 3X1 + 2X2 + X3
Subject to: 4X1 + 2X2 + 3X3 12
all variables Xi's 0
Since the feasible region is bounded, following the Algebraic Method by setting all the
constraints at the binding position, we have the following system of equations:
Page 8/9

4X1 + 2X2 + 3X3 = 12

X1 = 0
X2 = 0
X3 = 0
The (basic) solutions obtained, from this system of equations are summarized in the
following table.
X1
0
0
3
0

X2
0
6
0
0

X3 Total Net Profit

4
4
0
12*
0
9
0
0

Thus, the optimal strategy is X1 = 0, X2 = 6, X3 = 0, with the maximum net profit of

$12.
The result in the above table is consistent with the application of the above main
economic result. In other words, the optimal solution can be found by setting at least n m = 3 - 1 = 2 decision variables to zero:
For large-scale LP problems with many constraints, the Algebraic Method involves
solving many linear systems of equations. When the LP problem has many variables and
constraints, solving many systems of equations by hand can become very tedious. Even
for very large-scale problems it is an impossible task. Therefore, we need the computer to
do the computations for us. One of the algorithmic and computerized approaches is The
Simplex Method, which is an efficient and effective implementation of the Algebraic
Method. There are well over 400 LP solvers, all of which using the Simplex method,
including your software. Upon solving the LP problem by computer packages, the
optimal solution provides valuable information, such as sensitivity analysis ranges.
Page9/9