Linear Programming: Graphical Analysis

Page 1

GRAPHICAL METHOD OF SOLUTION

For LP problems of 2 variables, and only a few constraints, the graphical method is the
easiest solution method.
X2
Product-Mix Problem 40

Maximize z = 300x1 + 250x2 30

s.t. 2x1 + x2 < 40 (labor constraint)
20
x1 + 3x2 < 45 (machine time constraint)
15
x1 < 12 (marketing constraint) 10
H
x1, x2 > 0 (nonnegativity constraints)
X1
The shaded area is the area where all the constraint 10 20 30 40 45
arrows intersect. The points within this area satisfy all x1 + 3x2 < 45
x1 < 12 2x1 + x2 < 40
the constraints of the problem. These points are called
feasible points, and the shaded area is called the feasible area or feasible region.
Any point outside this area is called an infeasible point (i.e., these infeasible points
will violate one or more of the constraints).
The intersections of the constraints are called corner (or extreme) points. There are
feasible corner points and infeasible corner points.
Corner point solutions are adjacent to each other when they are connected by a single
line segment on the boundary of the feasible region.
Identifying the optimal solution:
Because there are an infinite number of points in the shaded region, this implies that
there are an infinite number of possible solutions to the problem. How can we cut down on the
points we need to evaluate? Consider point H(10, 10). It is feasible, with Z = 300(10) + 250(10)
= 5500. If we move to a boundary say (12,10), Z increases to 6100. Notice that such an increase
may continue only until we are limited by a constraint. It can therefore be shown that no matter
what interior feasible point you choose as a solution, there will always be a point somewhere in
the boundary superior to that interior point. Furthermore, an optimal solution must be at a point
where 2 (or more) constraints intersect (feasible corner points). Note: if only one optimal
solution exists - then it must be a corner point. If more than one optimal solution exists - then at
least 2 optimal solutions must be on two adjacent corner points, and the other solutions will be
on the boundary connecting these 2 points.
 Linear Programming: Graphical Analysis
Page 2

Two methods can be used to search for the optimal solution:

a) You can enumerate all the corner points, and get their Z-value.
b) Use isoprofit (or isocost) lines. An isoprofit line is a line where all the points in it
have the same objective function (z) value. Isocost line is the term when we are
minimizing.
The dashed lines in the graph denote the isoprofit X2
40
lines for this problem. If you compute the z-value
of each line, you will notice that the profit 30
increases as you go to the right. 20
This implies that the optimal (or best) 15
solution lies in the intersection of constraints #2 10
(x1 + 3x2 < 45) and #3 (x1 < 12). This intersection X1
is computed as: (12, 11). 10 20 30 40 45
Therefore z = 300(12) + 250(11) = 6350. x1 + 3x2 < 45
x1 < 12 2x1 + x2 < 40

Blending Problem X2

Minimize z = 45x1 + 12x2

300
s.t. 1x1 + x2 > 300 (brightness specification)
3x1 > 250 (hue specification) 200
x1, x2 > 0
100
The feasible area, in this case, is on the upper right side of
X1
the graph. The Z-lines (called isocost lines here) have a
100 200 300
better value as it moves to the left. Therefore, the optimal
x1 + x2 > 300
solution is at the intersection of constraints 1 and 2. Through 3x1 > 250
algebraic manipulation, this point is (x1 = 83.33, x2 =
216.67) with a Z-value (cost) of 6350 cents.
GRAPHING AN EQUALITY CONSTRAINT
If (at least) one constraint appears as an equality, then the feasible region of solutions is
not an area anymore, but a line segment, directly on the line describing the equality constraint.
Generally speaking, the use of equations should be minimized in LP because it introduces
inflexibility.
 Linear Programming: Graphical Analysis
Page 3

Utilization of Resources – SLACK AND SURPLUS VARIABLES

SLACK VARIABLES for < constraints

Product-Mix Problem Maximize Z = 300x1 + 250x2
s.t. 2x1 + x2 < 40 (labor constraint )
The optimal solution for this problem has x1 =
x1 + 3x2 < 45 (machine time constraint)
12 and x2 = 11. Let us now examine what
x1 < 12 (marketing constraint )
will happen to the constraints when we
x1 > 0, x2 > 0(nonnegativity constraints)
substitute the optimal solution.
Labor Constraint: 2x1 + x2 < 40: 2(12) + (11) = 35 hours < 40.
That is, the demand for labor is 35 hours, whereas the supply is 40. Therefore there
would be 40 – 35 = 5 hours of unused labor potential. This unused supply is called slack, si,
where i = constraint i. The slack must take only nonnegative values (si > 0). Designating this
slack variable by s1 (where 1 = first constraint), we have:
2x1 + x2 + s1 = 40. Therefore s1 = 40 – (2x1 + x2) = 45 – 24 – 11 = 5.
Machine Constraint: x1 + 3x2 < 45 => x1 + 3x2 + s2 = 45
Substituting optimal solution: s2 = 45 – (x1 + 3x2) = 45 – 12 – 3(11) = 0
Marketing Constraint: x1 < 12 => x1 + s3 = 12
Substitute optimal solution: s3 = 12 – x1 = 12 – 12 = 0
The last 2 constraints had zero slack. This means that both constraints are being exactly met.
 When slack = 0, you have what is called a binding constraint.
 When slack > 0, you have a non-binding constraint.
Looking at the graph, note that the optimal solution lies on the intersection of binding
constraints; and a non-binding constraint does not intersect the point of the optimal solution.
Note that binding and non-binding constraints can only be evaluated at the optimal solution. If
not yet optimal, the binding and non-binding constraints have no useful function.
SURPLUS VARIABLES for > constraints
Minimize z = 45x1 + 12x2
Recall the Blending Problem. The first constraint: s.t. 1x1 + x2 > 300 (brightness)
x1 + x2 > 300 => x1 + x2 – s1 = 300 3x1 > 250 (hue)
s1 = x1 + x2 – 300 = 83.33 + 216.67 – 300 = 0. x1, x2 > 0

For the second constraint: 3x1 – s2 = 250 => s2 = 3x1 – 250 = 3(83.33) – 250 = 0
Note that, because of its definition, a surplus variable is always > 0.
Slack and surplus variables have a zero coefficient in the objective function.
 Linear Programming: Graphical Analysis
Page 4

SPECIAL CASES OF LP PROBLEMS

The two examples we had had a unique optimal solution. But there are other types of LP’s that
do not have unique optimal solutions.
1. Some LP’s have an infinite number of solutions (alternative or multiple optimal
solutions).
2. Some LP’s have no feasible solution (infeasible LP).
3. Some LP’s are unbounded, where there is no limit to how much Z can improve.
EXAMPLE OF CASE 1. EXAMPLE OF CASE 2. EXAMPLE OF CASE 3.
Maximize Z = 3x1 + 2x2 Maximize Z = 3x1 + 2x2 Maximize Z = 2x1 – x2
s.t. 6x1 + 4x2 < 24 s.t. 6x1 + 4x2 < 24 s.t. x1 – x2 < 1
x1 + x2 < 5 x1 + x2 < 5 2x1 + x2 > 6
x1, x2 > 0 x1 > 3 x1, x2 > 0
x2 > 2
x1, x2 > 0
 Linear Programming: Graphical Analysis
Page 5

Case 1.
Maximize Z = 3x1 + 2x2
s.t. 6x1 + 4x2 < 24
x1 + x2 < 5
x1, x2 > 0

X2

1
X1
1 2 3 4 5 x1 + x2 < 5
6x1 + 4x2 < 24

LP with alternative or multiple optimal solutions.

 Linear Programming: Graphical Analysis
Page 6

Case 2.
Maximize Z = 3x1 + 2x2
s.t. 6x1 + 4x2 < 24
x1 + x2 < 5
x1 > 3
x2 > 2
x1, x2 > 0

X2

5 x1 > 3

3
x2 > 2
2

X1
1 2 3 4 5 x1 + x2 < 5

6x1 + 4x2 < 24

Infeasible LP – no feasible region

 Linear Programming: Graphical Analysis
Page 7

Case 3.
Maximize Z = 2x1 – x2
s.t. x1 – x2 < 1
2x1 + x2 > 6
x1, x2 > 0

X2
6

4 x1 – x2 < 1

0 X1

–1 1 2 3 4 5

2x1 + x2 > 6

LP with an Unbounded Solution