CHAPTER 2

An Overview of
Optimization Methods

After the completion of this chapter, the student should be able to:
1. Solve optimization problems limited to one or two control variables by graphical
means.
2. Recognize that linear mathematical models have the special property that the
optimum solution will always occur at an extreme point.
3. Utilize the methods of calculus to find the critical-point solution of nonlinear
optimization problems consisting of one control variable and classify the critical
point as a minimum, maximum, or point of inflection.
The essential concepts for finding the optimum solution of resource allocation
models can be illustrated with graphical means and models limited to two control
variables, called bivariate models. That is, the control vector x consists of Xl and Xl
only. When models consist of more than two control variables, graphical methods
cannot be used and mathematical methods will be utilized. First, our discussion will
be limited to models with linear objective and linear constraint equations. This is an
extremely important class of problems, both from a mathematical and an engineering
point of view.
Next. the graphical method will be used to solve bivariate nonlinear problems.
Finally, calculus will be used to solve univariate mathematical models, models with
one control variable. The section on calculus is intended to be a review.
The graphical method is an extremely important method because we can visualize
the overall problem more clearly. We not only find the solution to a problem in a
straightforward manner, we usually obtain insights that are not apparent by studying
the mathematical model alone or by solving with mathematical methods. It is truly
unfortunate that the graphical method of solution is not applicable to models with
more than two control variables. The conclusions that we draw from solving uni-
variate and bivariate models, however, are incorporated in solving multivariate
linear models and the nonlinear programming found in later chapters of this book.
 Graphical Solution to linear Models 25

2.1 GRAPHICAL SOLUTION TO LINEAR MODELS

A bivariate linear mathematical model may be written as

Z = CIX1 + C2X2
QllX1 + a12x2 {=, s,~}b1

The number of constraint equations m will depend upon the problem statement.
The graphical procedure consists of the following steps.
1. Establish the feasible region from the set of constraint equations.
2. Assume a solution of ZO and establish the slope of the line C1X1 + C2X2 = zoo
3. Determine the optimum solution z* by establishing a line that is parallel to ZO
and lies on the boundary of the feasible region.
Consider the mathematical model of Section 1.1. for maximizing total revenue for
the building shown in Figure 1.2. The first step is to establish the feasible region.
Let us establish the feasible region by investigating each constraint equation
separately, and then we shall combine these results to construct the feasible region
for the entire constraint set of equations.
The constraints x 1 ~ 0 and X2 ~ 0 restrict the feasible region to be all points in
the positive quadrant or quadrant I. In Figure 2.la, we see that Xl > 0 restricts the
solution to be in quadrants I and IV. Similarly, in Figure 2.1b, we see that X2 ~ 0
restricts the solution to be in quadrants I and II. The intersection of both constraints,
x I 2:: 0 and X2 ~ 0, will always limit the solution to lie within the positive quadrant,
quadrant I. as shown in Figure 2.1c.
The constraints Xl + X2 ~ 5000, XI S 5000, and Xl S 3000 are shown in Figures
2.la, 2.2b, and 2.2c~respectively. The arrows show the half-plane where the feasible
region must lie. The intersection of these planes results in the feasible region as shown

X
2

II II

x,

III III IV III IV

(el) (b) (c)

Figure 2.1 Nonnegative constraints x, ~ 0 and x2 ~ O.

26 An Overview of Optimization Methods

X
2

x
5000 1
5000
(a) (b)

x2

o Extreme points

(c) (d)

Figure 2.2 Establishing a feasible region.

by the shaded portion of the graph in Figure 2.U. All points that lie within or on U:c
boundary of the feasible region are calledfeasible solutions. All points that lie outside
the feasible region are called infeasible solutions. Points Xl = (X!, and x~ = xn
(xi, x~) are examples of feasible and infeasible solutions, respectively.
The second step requires that an objective function be drawn for an assumed
value of zoo Since the objective function is a linear function, we can easily establish a
contour line or locus of points satisfying SOx1 + 6OX2 = zoo Let us arbitrarily choose
ZO to be equal to SI50.000. Thus, SOx1 + 60X2 = $150,000. For convenience ..... e
determine the points (x.. 0) and (0, X2). These points lie on the Xl and X2 axes,
respectively. Thus,
SOx1 + 60 . 0 = 150,000 or Xl = 3000 or (3000, 0)
and 50 . 0 + 60x 1 = 150,000 or Xl = 2500 or (0, 2500)
The ZO line must pass through points (3000, 0) and (0, 2500). All points that lie on tc
line ZO = $150,000 and do not intersect as shown in Figure 2.3 are called infeasible
solutions. None of these points can be the optimum solution.
 Graphical Solution to Linear Models 27

X
2

Extreme points

forect;on for
increasing value
otz Extreme
point

Figure 2.3 Searching for an optimum solution.

Ltilizing the third step, the optimum solution may be found by drawing a contour
line parallel to =0 that lies at an extreme point of the feasible region. The direction for
increasing the values of z is shown in Fig. 2.3. Any new contour line of z must intersect
the feasible solution to be a candidate for an optimum solution. For linear models,
an extreme point is defined to be the intersection of two or more constraint equations.
An extreme point will lie on the boundary of the feasible region; therefore, it is a
feasible solution and a candidate for the optimum solution. The contour line marked
t:" is parallel to =0 and it passes through the extreme point x*, the location of the
optimum solution for this maximization problem as shown in Figure 2.4. The point
x" is unique because = is a maximum and satisfies the conditions established in the
constraint set of the problem. Thus, the optimum solution to this problem is
xi = 5000 ft2 and xi = 3000 ft 2
with optimum or maximum total revenue equal to z* = $50·5000 + $60·3000 =
S430.000.
This is a straightforward approach to solving linear mathematical models. In
step 3. we are able to establish the optimum point z* because the slope of the objective
function is always parallel to the contour line ZO regardless of the assumed value of zoo
This is a property of linear functions. This approach can be used for any linear mathe-
matical model restricted to two control variables with a minimum or maximum
objective function.
28 An Overview of Optimization Me"thods

z * = $430,000

Active constraint:
Xl = 5000

Figure 2.4 The optimum solution.

Active and Inactive Constraints

When identifying whether or not a solution is located on the boundary or at an extreme
point of the feasible region, the terms active and inactive constraints are used. By
definition, an actice constraint will occur on the boundary line or at an extreme point
of thefeasible region. The constraint equation where g(x) is either a linear or non-
linear function of x,

g(x) ::; b

is said to be active constraint when the values of x satisfy the strict equality as:

g(x) = b

On the other hand, when the solution x is such that the inequality is satisfied,

g(x) < b

the constraint equation is said to be an inactive constraint. In Figure 2.4 the active
and inactive constraints are indicated
 Graphical Solution to linear Models 29

EXAMPLE 2.1 Minimum-Weight Truss

In Example 1.1, the following mathematical model was derived.
Minimize z = 612AI + 442A2
A2 ~ 4.17

Al ~ 6.67
At 2: 8.33

A2 2: 6.25

At 2: 16.7

where A 1 and A2 represent the cross-sectional areas of the compression and tension members,
respectively. Find the optimum member sizes using a graphical method of solution.

Solution
The first step is to establish the feasible region. Each equation has been plotted in Figure 2.5.Arrows
have been placed upon each constraint equation to show the location of the feasible region for each
one. The intersection of these constraints specify the feasible region. The feasible region is shown as
the shaded region. It is bounded by the equations A I = 16.7 and A 2 = 6.25 as shown in Figure 2.6.

30
'"
E
~= _ A1=6.67 A1=8.33

fi 20
"0

L_----~~-10~-----L---2~0----------3~0--------~40A1
Area of compression members (in.2)

Figure 2.5
30 An Overview of Optimization Methods

Al = 16.7

Infeasibfe region

V>

<l>
s:
E
<l>
E 20
c
a
.;;;
c
~
'0
C10
<l>

A2= 6.25

A
40 1

Figure 2.6

The next step is to determine the slope of the objective equation for an arbitrary value of zoo Let
=0 = 5000. We shall find the intercepts of612A1 + 442A2 = 5000 on the Al and A2 axes.

612AI + 442·0 = 5000; Al = 8.16 or (8.16, 0)

612·0 + 442A2 = 5000; A2 = 1131 or (0, 11.31)

The line for zO = 5000 is shown in Figure 26.

Since all the points of ZO lie in the infeasible region, no solution exists along the =0 line. By in-
creasing z, we can find the extreme point where z is a minimum. The arrow on =0 = 5000 show the
direction in which the optimum solution lies. The optimum solution z* = 13,000 is shown in the
figure. Note that lines z* and ZO are parallel.
The optimum solution is

AT = 16.7in.2, Ai = 6.25 in.2

Thus, all compression members, AC Be,and CD, have cross sectional areas of 16.7 in.2 and all
tension members, AB and BD, have cross sectional areas of 6.25 in,", The total weight of the mini-
mum truss is z* = 162·16.7 + 442·6.25 or

z" = 13,000lb
 Graphical Solution to linear Models 31

EXAMPLE 2.2 Minimum-Cost Aggregate Mix Model

A contractor is considering two gravel pits from which he may purchase material to supply a
project. The unit cost to load and deliver the material to the project site is S5.00/yd3 from pit I and
$7.00jyd3 from pit 2. He must deliver a minimum of 10,000yd ' to the site.
The mix that he delivers must consist of at least 50 percent sand. no more than 60 percent gravel,
nor more than 8 percent silt. The material at pit 1 consists of 30 percent sand and 70 percent gravel.
The material at pit 2 consists of 60 percent sand, 30 percent gravel, and 10 percent silt.
(a) Formulate a minimum-cost model.
(b) Determine the optimum solution by the graphical method.
(c) Determine the active and inactive constraint equations for the optimum solution.
(d) Determine the proportions of sand, gravel, and silt in the optimum solution.
Solution
(a) Formulation Since the gravel from pit 1 does not contain the minimum amcunt of sand to
meet project requirements, the contractor may not utilize the cheaper material exclusively. He
must mix the material from pits 1 and 2 to produce the required proportions.
We define the control variables to be
xI = amount of material taken from pit 1 (i.11 cubic yards)
x: = amount of material taken from pit 2 (in cubic yards)
The cost function is
minimize c = $5.oox1 + S7.00x.:
Let Xl + X2 equal the total amount of standard gravel mix delivered to the project site. The
contractor must deliver at least 10,000 yd ', thus the delivery constraint is
Xl + X2 ~ 10,000

The mixture must contain at least 50 percent sand. The contractor may obtain the desired
amount of sand by combining the materials from each pit.
0.3x1 + 0.6x.: ~ 0.5(XI + x:)
The products 0.3x 1 and 0.6X2are the amounts of sand taken from pits I and 2, respectively. The
term 0.5(x 1 + x 1) is the amount of sand in the mix. Sirnilarily. the constraint on the amount of
gravel to be delivered is
0.7Xl + 0.3X1 :::;;0.6(X1 + x:'
Finally, the constraint equation for silt is
0.1X2 :::;;0.08(XI + X2)
The minimum cost model may be written as
Minimize c = 5x 1 + Tx :
Xl + x2 ~ 10,000 (delivery)
0.3x1 + 0.6X2~ 0.5(XI + Xl} (sand)
0.7Xl + 0.3X2:::;;0.6(XI + Xl} (gravel)
0.IX2 :::;;0.08(XI + xJ (silt)
32 An Overview of Optimization Methods

or in standard form:

Minimize c = 5x1 + 7X2

x, + X2 ~ 10,000 (delivery)

(sand)

(gravel)

4Xl - X2 ~ ° (silt)

The second cost model will be utilized for plotting. The first cost model will be used to answer part
d and for checking the results.

(b) Graphical Solution

Step 1. Establish thefeasible region. Sincex, ~ Oand x2 ~ 0, the feasible region is restricted to be
in a positive quadrant. Since only two points are needed to determine a line, each constraint
equation is assumed to be a strict equality, and then the boundary of the constraint equation is
found. For instance, for the equation:

Xl + X2 = 10,000

when Xl = 0, X2 must be equal to 10,000 or X2 = 10,000 and when X2 = 0, Xl must be equal to

10,000 or Xl = 10,000. Thus, the points (0, 10,(00) and (10,000,0) are sufficient to determine the
boundary of the constraint, Xl + X2 2:: lO~OOO.Next, the boundary of the feasible region for the
inequality constraints of sand, gravel, and silt is determined. In Figure 2.7a the arrows show the
direction of the feasible region.
Since the gravel constraint - X 1 + 3Xl 2:: 0 lies outside the feasible region. it .....
-ill not be con-
sidered in the search for the optimum solution.
Steps 2 and 3. Estimate an optimum solution, and test the condition of optimality. The minimum
cost was assumed to be equal to S80,()(X) or c? = S80,000. Figure 2.7b shows this estimate to be too
high. This estimate does not satisfy the condition of optimality.
The optimum-cost line will be parallel and less than the initial estimate of CO = SSD.OOO. Further-
more, for linear mathematical models, we should search for the optimum solution at an extreme
point. For this problem, the optimum point is equal to

xi = 3300ft3, xi = 6700 ft3

with minimum cost c* = $63,400.

Note that the optimum-cost line is parallel to the initially estimated cost line, and it satisfies the
conditions of optimality. In addition, the optimum solution occurs at an extreme point.

(c) Active and Inactive Constraints Since the optimum solution passes through the inter-
section of the lines marked delivery and sand, the equations X2 - 2x1 = 0 and Xl + X2 = 10,000
are both active constraints. The remaining equations ar; inactive constraints. These are the lines
labeled silt and gravel. Since the constraint equation for gravel lies outside the feasible region, it
will always be an inactive constraint.
 ~ 10,000
c=

20,000 Xl

20,000

X; = 3,300 10,000 20,000 Xl

Pit 1 (yd3)

(h)
Figure 2.7 (a) Feasible region. (b) Optimum solution.

33
34 An Overview of Optimization Methods

(d) The Optimum Mix The amount of sand, gravel, and silt that is the optimum mix is most
conveniently determined from the first cost model. The amount of sand is

0.3xi' + 0.6x1 = 0.3·3300 + 0.6·6700 = 5010 yd '

The minimum amount required is 0.5(x1' + xi) = 5000, where xi + xi is the total amount
delivered.
The amount of gravel is

0.7xi' + 0.3xi = 0.7·3300 + 0.3·6700 = 4320 yd '

This is less than 0.6(x1' + xi) = 6000, the maximum amount of gravel permitted in the mix.
The amount of silt is
O.lx! = 0.1(6700) = 670
This is less than 0.08(x1' + xi) = 800, the maximum amount of silt permitted in the mix.
The constraint conditions are satisfied. The minimum cost to deliver the mix is $63,400.

PROBLEMS

Problem 1
The constraint set is
(1)

(2)

(3)

(a) Clearly show the feasible region on a graph for the constraint set.
(b) Label all extreme points.
(c) If the lessthan or equal to constraint ofEq. (1) is changed to a strict equality,show the feasible
region and label the extreme points.

Problem 2
Consider the following constraint sets:

XI - 2x2 ~ 2
2xI + X2 ~ 9
-3xI + 2x2 ~ 3
Xl is unrestricted in sign

(a) Clearly show the feasible region on a graph for the following constraint set.
(b) Label all extreme points.
(c) If XI ~ 0 is imposed, show the feasible region and label the extreme points.
 Graphical Solution to linear Models 35
Problem 3
By graphical means, determine the location of the optimum solution.

Maximize z = 3x1 + 2X2

(a) Clearly show the feasible region.

(b) Determine the magnitude and location of the optimum point.
(c) Label the active and inactive constraints.

Problem 4
By the graphical approach, determine the optimum solution to

Maximize z = 5x1 - 2x2

XI is unrestricted in sign

(a) Clearly show the feasible region.

(b) Determine the magnitude and location of the optimum point.
(c) Label the active and inactive constraints.

Problem 5
Minimize z = 3x1 + Xl

2xI - 4X2 = 6

XI 2. 0,

(a) Clearly show the feasible region.

(b) By graphical means determine the optimum solution.
(c) If the nonnegative restriction on X2 (i,e., X2 2. 0) were removed, determine the effect on the
optimal value of z,
36 An Overview of Optimization Methods

Problem 6
Maximize z = 2x2

(1)

-XI + 2X2 ~ 2 (2)

2xI + 2x2 = 3 (3)

XI is unrestricted in sign

(a) Clearly show the feasibility region.

(b) Determine the magnitude of the optimum point.
(c) Label the active and inactive constraints.
(d) How is the optimum point affected if constraint (2) is removed. Why?

Problems 7, 8. 9, and 10 were formulated in Section 1.1 as problems 1. 2, 3, and 4, respectively. The
question here deals with the solution of the problems.

Problem 7
A contractor has two sand and gravel pits where he may purchase material. The unit cost including
delivery from pits 1 and 2 is S5 and 57 per cubic yard, respectively. The contractor requires 100 yd '
of mix. The mix must contain a minimum of 30 percent sand. Pit 1 contains 25 percent sand, and pit
2 contains 50 percent sand.
The object is to minimize the cost of material.
(a) Draw the feasible region.
(b) Determine the optimum solution by the graphical approach.
(c) Label the active and inactive constraints.

Problem 8
An aggregate mix of sand and gra vel must contain no less than 20 percent nor more than 30 percent
gravel. The in situ soil contains 40 percent gravel and 60 percent sand. Pure sand may be purchased
and shipped to site at S5.00/yd3• A total mix of 1000 yd ' is needed. There is no charge to use in situ
material.
The goal is to minimize cost subject to the mix constraints.
(a) Draw the feasible region on the graph.
(b) Determine the optimum solution.
(c) Label the active and inactive constraints.

Problem 9
There are two suppliers of pipe:

UNIT COST SUPPLY

SOURCE (S/lINEAR FOOT) (LINEAR FOOT)

1 $100 100 ft maximum

2 $125 Unlimited
 Graphical Solution to linear Models 37

Nine hundred feet of pipe is required. The goal is to minimize the total cost of pipe.
(a) Find the optimum solution.
(b) Formulate a mathematical model with the supply of pipe from source No. 2 limited to 700
linear feet.
(c) Does a solution to part b exist? Use the graph to prove your answer.

Problem 10
A company requires at least 4.0 MgaIjday more water than it is currently using. A water-supply
facility can supply up to 10 Mgaljday of extra supply. A local stream can supply an additional
2 Mgal.day. The concentration of pollution must be less than 100 mg/l BOD, the biological
oxygen demand. The water from the water-supply facility and from the stream has a BOD
concentration of 50 mg/l and 200 mg/l, respectively. The cost of water from the water supply is
SIOO/Mgal and from the local stream is S50jMgal. The goal is to minimize cost of supplying
extra water that meets water quality standards.
By graphical means, show the feasible region, and determine the location of the optimum
solution.

Problem 11
A treatment plant has a capacity of 8 Mgal/day and an operating efficiency of 80 percent. The
operating efficiency is defined as the amount of BODs, the 5-day biological oxygen demand,
removed by the treatment plant facility. For example, if 200 mg/l of BODs enters the plant, the
amount of BOD~ leaving it is
200(1 - 0.8) = 40 mg/l of BODs

Owing to regional population growth, the rate of flow of wastewater has increased to 20 Mgalf
day. The BOD~ of the wastewater is 200 mg/l. Since the plant can only treat Mgal/day, 12 Mgalj
day is being diverted from the plant and is entering the river without treatment (Figure 28a).

20 Mgal/day

(treated) 12 Mgal/day (untreated)

(a) (b)

Figure 2.8
 38 An Overview of Optimization Methods

(a) Determine the BO 05 of the combined flow of treated and untreated water entering the river
(BODs in mgjI).
(b) The cost of building a new treatment plant (Figure 2.8b) is

(Q = flow)

Define all control variables and formulate a mathematical model to treat 20 Mgalfday of
wastewater which has 200 mgjI of BOD5• The combined flow of treated water entering the
river from the old and new plants must not exceed 25 mgjI of BODs. The efficiency of the
existing plant is 80 percent, and of the new plant is 90 percent
(c) Determine the minimum-cost solution by graphical means.

Problem 12
A 30-mile stretch of roadway (Figure 2.90) is considered to have a poor level of service. Twenty-five
million dollars (S25M) has been allocated for the project. The cost to improve the roadway is
S I M'rnile ; therefore, the entire roadway cannot be improved.
The existing and proposed upgraded roadways are assumed to havea speed-density relation-
ship as shown in Figure 29b.
(a) Derive speed-density functions for the existing and upgraded roads. The speed u is assumed
to be a linear function of density k.

k = density (vehicles/mile)

u = speed (miles/hr)

(b) Derive a flow-density function for the existing and upgraded roads, q = ku, and draw the
function for the existing and upgraded roadways, where flow is defined as q = flow (vehicles/
hr).
(c) What are the maximum capacities of the existing and upgraded roads? Call them qrnax'
(d) Determine the density at capacity kmax for the existing and upgraded roads.
(e) The flow on the existing roadway is 800 vehiclesjhr between point A and intersection C,
and 1200vehicles/hr between C and B. Show that the vehicular speeds for uncongested flow,
q $ qrnax' are as follows:

Existing road
Segment A-C IiAC = 31.5 rniles/hr
Segment B-C IIllc = 20 miles/hr
Upgraded road
Segment A-C UAC = 52.4 milesjhr
Segment B-C UHC = 47.4 milesjhr
(f) Determine the total travel time speed between A and B for existing road or liAC = 31.5 miles/
hr and Usc = 20 milesjhr.
(g) Formulate a mathematical model to minimize total travel time between A and B. Assume a
budgetary constraint of $25M. Clearly define control variables for mileage of roadway to be
constructed.
(h) Solve for the optimum solution in part g by graphical means.
 Graphical Solution to Nonlinear Models 39

Intersectingroad

Roadway
q = 800vehicles/hr q = 1200 vehicles/hr
A ---+- c ~ 8

L 10 m;le,--~-----20
(a)
mileS-----;J~/

.....

--·E'" 40
s:

!E.

-v
c:J
~ 30
Cli

Density (vehicles/mile)
(b)

Figure 2.9

2.2 GRAPHICAL SOLUTION TO NONLINEAR MODELS

We shall extend our method for finding the optimum solution to nonlinear design
problems limited to two control variables. For optimization problems of the general
form, the mathematical model form is
z = f(x1, X2)

gj(x1, X2){ =, ~, ~ }bi i = 1,2, ... , m

The graphical procedure consists of the following steps.
1. Establish the feasible region from the set of constraint equations.
2. Estimate an optimum solution, Z = .z", and draw the function, Zo = f(x1, Xl)'
40 An Overview of Optimization Methods

3. Test to determine if the conditions of optimality are satisfied. If not, repeat steps
2 and 3 for a new estimate of z.

These steps are the same ones used to find the solution of linear models. The major
difference between the search for an optimum solution of linear and nonlinear models
is that the search for an optimum solution of a linear model is restricted to extreme
points, whereas the search for an optimum solution of a nonlinear model must
consider the whole feasible region. The optimum solution may occur at an interior,
boundary or extreme point.

Conditions of Optimality
Compare the following functions

The functions ZI and Zl are monotonic functions and Z3 is a nonmonotonic function.

The nonmonotonic function has one or more locations where local minima or
maxima exist. The function Z3 has a local minimum at Xl = 3 and x2 = 4. Functions
Z I and Zl do not possess this property. If Xl and Xl are restricted to a certain feasible
region, it is possible that the nonmonotonic function Z3 will possess the property of
a monotonic function. For instance, Z3 is a monotonic function for the region,
o ~ x 1 ~ 1 and 0 < x 1 ~ 2.
The conditions of optimality are satisfied when a solution (xi, xi) satisfies all
constraint equations and, at the same time, minimizes or maximizes the objective
function of z. For a monotonically objective function of z, the condition of optimality
is satisfied when the contour line of Z is a boundary or extreme point of the feasible
region. For nonmonotonically functions of z, the condition of optimality is satisfied
when the contour line of z is a boundary point, extreme point, or an interior point
of the feasible region.

Monotonic Objective Functions

A mathematical model is defined as being a nonlinear model when either the objective
function is a nonlinear function or when one or more of the constraint equations are
nonlinear functions. For example, the following nonlinear cost model consists of a
linear objective function with a nonlinear constraint equation.
Minimize Z = C1X1 + C2X2
g(XI' x2) = b
Xl ~ 0
Xl ~ 0
The linear objective function is a monotonically increasing function of x.
 Graphical Solution to Nonlinear Models 41

The first step to solve this problem by graphical means is to draw the contour of the
constraint equation on the graph. Since the constraint equation is a strict equality,
any combination of Xl and X2 that lies on the constraint line is a feasible solution and a
candidate for the optimum solution. The next step is to estimate the total cost ZO
and draw the cost function.
After steps 1 and 2 of the graphical procedure are performed, three possible out-
comes may occur. They are illustrated in Figure 2.10.
Let us investigate each possible outcome with regard to step 3, that is, the testing
of the condition of optimality. The solid lines marked ZO are the initial estimates.
The dashed lines marked z* are the optimal solutions.
In Figure 2.10a the initial estimate ZO is the optimum solution. This point is unique.
No other solution will meet the constraint condition at smaller cost. At point I,
the tanqents of the cost and constraint lines are equaL This is an important finding.
For nonlinear models, it is possible that an optimum solution will occur at a tangency
(Joint.
Figure 2. lOb shows the initial estimate ZO to be below the constraint line g(x 1, x2)
= h. No feasible solution exists for this estimate. Since the feasible solution must lie
on the constraint line, any combination of Xl and Xl that we select on the cost line
will not fall in the feasibility region, therefore, it cannot satisfy the constraint require-
ment. Another interpretation of this result is that insufficient funds ZO have been
provided to achieve the minimum required output b. More money is needed. The new
estimate of the cost line must be made as indicated by the arrows in Figure 2.1Ob. In
lieu of repeating the estimating and testing for the conditions of optimality, we may
utilize the properties of the linear cost function and conditions of optimality to solve
this problem. The initial estimate of the cost line ZO has established the slope of the cost
function. All other cost lines must be parallel to this initial estimate. Since the con-
ditions of optimality state that the objective function must be tangent to the constraint
line of the feasible region, the optimum solution for this example must occur-at a point
where the cost line is tangent to the boundary g(xl, Xl) = b, shown as point 1 in
Figure 2.1Ob. The optimum is located at (xi, xi).
In Figure 2.10c the initial guess is too high. Feasible solutions occur at point 2
and 3. The amount output at points 2 and 3 is the same q(xi, xD= g(xi, x~) = h. We

(a) (b) (c)

Figure 2.10 A linear objective and nonlinearconstraintequation.

42 An Overview of Optimization Methods

x·
2

g(X1, x2) = /11

x·1 Xl L-------X
.• L ------------x1
1

(al (b)

Figure 2.11 Conditions of optimality for a typical minimum-cost

model.

are wasting money by producing at either one of these combinations of inputs of Xl

and Xl' If we change the values of x 1 and Xl in such a manner as to remain on the
constraint line, we can reduce the cost and remain in the feasible region. Since the
slope of all cost lines must be parallel to the cost line z", we determine the optimum
solution to be at point L Clearly, the optimum solution is located at (xi, xi). Further-
more, note the output x* is the same at Xl or x3, g(xi, xi) = g(xi, x~) = g(xt, x~) = b.
Consider the optimum solutions to another nonlinear mathematical model con-
sisting of two control variables and two constraint equations.
Minimizez = !(xl. x2)

gl(X1.X2) ~ b,
g2(XI, X2) ~ b2
If the mathematical model has a nonlinear, monotonic, increasing objective function
and two nonlinear constraint equations. the optimum solution may occur at a tangent
point or at an extreme point as shown in Figures 2.I 1a and 2.11b, respectively.
If the objective function is nonlinear and the two constraints are linear, we may
obtain the solution at either a point tangent to the boundary or an extreme point as
shown in Figure 2.12a and 2.l2b.
X
2

x·
2

(a) (b)
Figure 2.12 A nonlinearobjective and linear constraint equation.
 Graphical Solution to Nonlinear Models 43

Nonmonotonic Objective Functions

Figures 2.10 through 2.12 show the optimum solutions to nonlinear mathematical
models with various combinations of linear and monotonic nonlinear objective and
constraint equations. We can see from these figures that the optimum solution will
be located at a point that is either a tangent point or an extreme point. If the objective
function is a non monotonic function as illustrated in Figure 2.13a and 2.14a the
optimum solution of
Maximize z = l(xl, x2)
gl(xI, x2) < b,
g2(X!> x2) ~ b2

X X
2 Contour of a concavefunction 2

z = [(xl' x2) where

:;:;0>:;:;1 >:;:;2

(.2) (bl

Optimum at a boundary point

(c) (d)

Figure 2.13 Nonmonotomic objective functions.

44 An Overview of Optimization Methods

~ ~
Contour of z = f(x1, x2)
where :;;0 <:: 1 <:;;2 < :;;3
Optimum x" at an interior point
and 20 == [(xo) is a maximum.

Unique maximum x" at x=

L-----------------------x1 ~--X1

(a) (b)

X X
2 2

Optimum x : at a bounoary point Optimum x " at a boundary point

(c) (d)

Figure 2.14 An objective function with a unique minimum at x",

may occur at an interior, boundary, or extreme point. Various linear constraint

equations are shown in Figures 2.13 and 2.14.
In Figure 2.14a. a contour mapping of the function z = f(x1, Xl) is shown. It is a
concave function with an unique maximum point at XV. Functions that have a single
minimum or maximum point have special mathematical properties that will be used
to help locate the solution of unconstrained and constrained optimization problems.
Those features will be discussed in greater detail in the following section and in
Chapter 8. For now, we use the graphical method of solution. For the above mathe-
matical modeL the optimum solution x* will not necessarily occur at xC. In Figures
2.14b and 2.14c the optimum solution is x* = x", but in Figure 2.14d it is a boundary
point where x* =F x=.

EXAMPLE 2.3 Maximum Output Model

The output of a given production is a nonlinear function of two control variables x 1 and .x
«= x1xj·5
 Graphical Solution to Nonlinear Models 45

The cost of production is

(a) Formulate a maximum output model given that the production cost is equal to $100.
(b) Determine the optimum solution utilizing a graphical method of solution.

Solution

(a) Formulation The mathematical model is

Maximize q = Xl x1·5
X~ + x1·5 = 100

(b) Graphical Solution The maximum output will occur when the strict equality XI + X~·5 =
100 is imposed. Since the objective function is a monotonically increasing function, the solution
will occur at a tangency point. The feasible solution is drawn and is shown as C* = 100. Several
different estimates of output were made. The level of output q* is 375.5 units. This is shown as the
tangency point on the graph. The optimum level of production is

xi = 5 units
x2 = 17.8 units

30

20
Optimum solution
17.8

10

The feasible region is the line, CO = 100

°0~--------~--------~------------------~20~-----------------730

Figure 2~15
46 An Overview of Optimization Methods

PROBLEMS

Problem I
Minimize z = 4x J + Xl

(a) On a graph show the feasible region.

(b) Show the location of optimum solution (xt, xi) and determine z*.
(c) Classify each constraint as an active or inactive constraint.

Problem 2
(a) By graphical means, find the optimum solution to the minimum-cost model.
Minimize z = 20x 1 + lOxl

Xl + Xl = 100

Xl ~ 0

X2 ~ 0
Label the optimum point as being an extreme, boundary, or interior point.
(b) Repeat part a with the objecture function equal to

:: = 20x?'s + lOx2
(c) Repeat part a with the objective function equal to

z = 20x:·1 +- 10x2

Problem 3
The output q is a function of the inputs Xl and X2' The cost production is
q=XIX~·5

C= X?·5 + X~·8
(a) Formulate a minimum-cost model given that the level of production q is 375 units and the
resources x J and x~ are restricted to a maximum of 10 and 25 units, respectively.
(b) Show the feasible region.
(c) By graphical means determine the minimum-cost solution.

Problem 4
Design a chemical reactor of minimum cost (Figure 216). The flow of liquid is steady, Q =
2 Mgal.day. The tank volume will be a function of the flow rate Q and retention time t. The reten-
tion time is equal to
V
t =-
Q
 Graphical Solution to Nonlinear Models 47

Cylindrical tank

Figure 2.16

where Q is the flow rate and V is the volume. The retention time t is 90 mins. The cost to construct
the walls and flooring is $2.60/ft2 of wall surface area and S 1.20/ft2 of floor surface area, respectively.
(1 gallon = 231 cubic inches.)
(a) Define the control variables.
(b) Formulate a minimum-cost model.
(c) Determine the optimum solution by graphical means.

Problem 5
A structural engineering consulting firm requires a large number of wide flange members for a
particular structure. Each wide flange member is to be designed as a simply supported beam sub-
ject to a maximum load of 100 kips (l kip = 1000 lb). In lieu of purchasing standard shapes, they
will fabricate the wide flange sections by cutting!' in. steel plate and welding it together. The cross
section of the member will be btl. where band d are design variables. The sheet steel will be cut in
a single pass with two torches.
(a) Formulate a mathematical model to minimize cost. or equivalently, the amount of steel to
construct a wide flange member from flat plate. All calculations should be done in kips and
inches. Consider the following AISC Building Code provisions.
A. Bending (AISC Section 1.5.1.5.1)

Flo = 0.66 F).

Fb = allowable bending stress

B. Shear (AISC Section 1.5.1.2)

r, = 0.4 F,.

The cross section may be taken as the product of the overall depth d and the web thick-
ness t.

F, = allowable shear stress

48 An Overview of Optimization Methods

P = 100 kips

r
d
L

..!.. in. steel plate passed

2 under torches

I
Movement in this direction

Figure 2.17

C Lateral Stability (AISC Section 1.5.1.4.1e)

I 76b
-<--
b - rr
...;F.,
The compressionflangeshall be laterally supported at intervals not to exceed76b/.JF;
Assumeno lateral support.
D. Material strength
F; = 36 ksi
E. For bendingsimplifyby assuming the moment of inertiaabout the weband center lineof
the flangesis zero. That is, assume the moments of inertia about the centerline of the
member is simply

(b) Determine the feasibleregion.

(c) By graphical means,determine the optimum solution.
 Graphical Solution to Nonlinear Models 47

Q = 2 Mgal/day

Cylindrical tank

Figure 2.16

where Q is the flow rate and V is the volume. The retention time t is 90 mins. The cost to construct
the walls and flooring is $2.60/ft2 of wall surface area and S 1.20/ft2 of floor surface area, respectively.
(1 gallon = 231 cubic inches.)
(a) Define the control variables.
(b) Formulate a minimum-cost model.
(c) Determine the optimum solution by graphical means.

Problem 5
A structural engineering consulting firm requires a large number of wide flange members for a
particular structure. Each wide flange member is to be designed as a simply supported beam sub-
ject to a maximum load of 100 kips (1 kip = 1000 lb). In lieu of purchasing standard shapes, they
will fabricate the wide flange sections by cutting !. in. steel plate and welding it together. The cross
section of the member will be: bd. where band d are design variables. The sheet steel will be cut in
a single pass with two torches.
(a) Formulate a mathematical model to minimize cost. or equivalently, the amount of steel to
construct a wide flange member from flat plate. All calculations should be done in kips and
inches. Consider the following AISC Building Code provisions.
A. Bending (AISC Section 1.5.1.5.1)

F,. = 0.66 F\_

Fb = allowable bending stress

B. Shear (AISC Section 1.5.1.2)

The cross section may be taken as the product of the overall depth d and the web thick-
ness t.

F, = allowable shear stress

48 An Overview of Optimization Methods

P = 100 kips

r
d
L

..!.. in. steel plate passed

2 under torches

I
Movement in this direction
Figure 2.17

C. Lateral Stability (AISC Section 1.5.1.4.le)

I 76b
-<--rt:
s: yFy

The compression flange shall be laterally supported at intervals not to exceed 76b/~
Assume no lateral support.
D. Material strength
Fy = 36 ksi

E. For bending simplify by assuming the moment of inertia about the web and center line of
the flanges is zero. That is, assume the moments of inertia about the centerline of the
member is simply

(b) Determine the feasible region.

(c) By graphical means, determine the optimum solution.
CHAPTER 6

Optimization of Linear
Mathematical Models

After completion of this chapter, the student should:

1. Beable to find the optimum solution of a linear programming problem with the
simplex method.
2. Be able to find the optimum solution of a linear problem requiring artificial
variables by the two-phase simplex method.
3. Be able to formulate linear mathematical models classified as resource allocation
and net analysis problems.
In Chapter 2 linear mathematical models were formulated and the optimum
solution for two-dimensional resource allocation problems were determined by
graphical procedures. We concluded that the optimum solution oflinear mathematical
model will always occur at an extreme point. We utilize these concepts to establish
mathematical methods to analyze the feasible region represented by a linear equation
of the form:

Minimize z = c'x

ajx {=, <, ~}bi i = 1,2, ... , m

x~O
Here we introduce surplus and slack variables to transform inequality constraint
equations to strict equality constraint equations:

Minimize z = c'x
Ax = b
x~O
The new mathematical model is in the standard form of linear programming. With
the use of basic and nonbasic variables, an extreme point of Ax = b may be evaluated
and tested as being the minimum z, An algorithm, called the simplex method, will be
introduced to efficientlysearch and test extreme points as the location ofthe minimum
value of z.
Matrix algebra is used in this chapter. An introduction of essential properties and
methods of solving problems of the form Ax = b is given in Appendix A.
286 Optimization of Linear Mathematical Models

6.1 THE STANDARD FORM

The most important conclusion that was drawn from the graphical determination of
the optimum solution of a bivariate linear mathematical model is that the optimum
solution will occur at an extreme point. Our search for the optimum will always be
restricted to the evaluation of extreme points. In this chapter mathematical methods
are used to solve linear models having n control variables of x. A linear mathematical
model is formulated and placed in the so-called standard form. New control variables
called slack and surplus variables will be used to transform a set of linear equations
with equality less-than-or-equal-to and greater-than-or-equal-to constraints to strict
equality constraints only. Furthermore, procedures to search the extreme points are
established. This discussion lays the groundwork for finding the optimum solution by
the simplex method to be discussed in the following section.

Slack and Surplus Variables

The standard form of Q linear mathematical model utilizing matrix notation is
z = c'.x
Ax = b
x~O
where x is an n vector and A is an m x n matrix. The standard form requires that (1)
all members of the b vector be positive calues and (2) each constraint equation be a
strict equality.
After the model is formulated utilizing inequality constraints, it can be placed in
standard form. The first requirement that all elements of b be positive may be easily
satisfied by multiplying a given equation with a negative value of b by -1. If the
constraint equation has an inequality constraint it must be transformed from a less-
than-or-equal-to constraint ~ to a greater-than-or-equal-to constraint ~ and vice
versa. For example, a model contains the following set:
-Xl + 2xl Z -3
2XI - 6x1 ~ -2
Xl ~ 0, Xl ~ 0
It may be transformed to positive values by multiplying each equation by -1 and
changing the sense of the inequality constraints. The transformed equation set
becomes
Xl - 2xl ~ 3
-2Xl + 6Xl ~ 2
Xl ~ 0, Xl ~ 0
The second requirement for standard form is that the constraint set contain only
equality constraints. Thus, the equation set does not satisfy this requirement and must
 The Standard Form 287

be modified. New control variables called slack and surplus variables will be introduced
into the model.
A slack variable is a positive, independent, control variable that is introduced into
a constraint equation containing a less-than-or-equal-to constraint

a'x ::;b
or
To transform this equation to one with a strict equality requires the introduction of a
slack variable xp. The transformed equation is
a'x + xp = b
or
A surplus variable serves a function that is similar to a slack variable. It is a positive,
independent, control variable that is introduced into the constraint equation containing
a greater-than-or-equal-to constraint:
a'x 2: b

or
Let the surplus variable be designated as the positive control variable x, 2: 0. The
transformed equation is
a'x - Xr =b
or a1x1 + Q2X2 + ... + anxn - Xr = b

Since surplus and slack variables are introduced into constraint equations with
less-than-or-equal-to (::;) and greater-than-or-equal-to (2:::) constraints, respectively,
both a surplus and a slack variable will never appear in the same equation.
Furthermore, slack and surplus variables may be used to transform both linear
and nonlinear equations. Thus, if g(x) represents a nonlinear function. the following
two inequality constraints may be transferred to equality constraints by utilizing a
slack and surplus variable, respectively.
g(x) ::; b => g(x) + xp = b
g(x) ~ b => g(x) - Xr =b
Consider the constraint set in the numerical example. In this chapter we are con-
cerned with linear equations only. The first equation of the equation set requires a
surplus variable, and the second equation requires a slack variable. Thus, the con-
straint set
Xl - 2X2 S 3 => Xl - 2X2 + X3 = 3
-2XI + 6X2 2: 2 => -2Xl + 6X2 - X4 =·2
Xl 2::: 0, X2 2::: ° => Xl 2: 0, X2 2: 0, X3 2::: 0, X4 ~ °
288 Optimization of linear Mathematical Models

where X3 is the slack variable and X4 is a surplus variable. The requirement that all
slack and surplus variable are positive values is explicitly expressed as X3 ~ 0 and
X4 2: O.
Slack and surplus variables have physical meaning and may be defined just like the
control variables used in the original formulation problem. Furthermore, since they
are control variables, all mathematical rules that apply to the original set of control
variables also apply to the slack and surplus variables. As a result, it is unnecessary to
distinguish among the original, slack, and surplus control variables when performing
mathematical manipulations on the mathematical model.

Evaluating Extreme Points

If a unique solution of Ax = b exists, then the number of unknowns must equal the
number of independent equations, an inverse of A must exist, and x = A-I b.
Consequently, the A matrix must be a square n x n matrix.
For linear mathematical models in standard form, generally, the A matrix is not a
square matrix; therefore, a unique solution does not exist.
z = c'.x

Ax = b
x~O
-For optimization problems, the number of control variables n will be greater the
number of constraint equations 114 or n > m. The discussion focuses upon the solution
of the constraint set Ax = b, where A is an m x 11 matrix. The vector x contains II
elements, x' = [Xl X2'" xn].
The concepts of basic and non basic variables, partitioned matrices, and matrix
algebra will be used to determine the solution to the constraint set. Remember, the
purpose of this work is to be able to search the extreme points of the constraint set for
possible solutions to the optimization problem. The understanding of the concept of
basic and non basic variables is an essential prerequisite for understanding the
principles of the simplex method, which is discussed in the next section.

The Basis
A vector space is a set of vectors that has the property that if two vectors of the set are
added or if a vector of the set is multiplied by a scalar, the sum or product is a member
of the set. A basis is a set of independent vectors in a vector space that spans a vector
space; that is, every vector in the vector space is a sum of scalar multiples of vectors
from the basis. For example, el' = [1 00], e2' = [0 1 0], and e3' = [001] are a basis
for the vector space of three-dimensional vectors. For the constraint equation
Ax = b, where A is an In x 11 matrix and n > m, A may be represented as a set of
column vectors, or
 The Standard Form 289

where the set of all aj vectors is assumed to be independent. According to the definition
of a basis it is possible linearly to combine m column vectors of A in the following
manner.

r Xl]
r-. a2... amJ·l]~ = b

Here, the first 111 column vectors of A were selected in this illustration. Other com-
binations of a, may be chosen. The matrix [al a2 ••• am] or some other combination
is called the basic matrix AB and [Xl X2 ... xmJ is called the basic vector XB. Thus

ABxB = b
The elements of XB are

Since AB is assumed to consist of a set of linearly independent a vectors, the inverse

Ai 1 exists and a solution to XB is obtained. Thus, the elements of XB form a unique
linear combination of the basic matrix vectors a, that span the vector space.
Matrix partitioning allows the constraint set to be written as

The vector x has been partitioned into two subvectors XB and XN' the basic and non-
basic rectors; respectively. The number of control variables assigned to XB is equal to
the number of constraint equations m. The remaining control variables will be assigned
to XJ\', thus it will contain (n - m) variables. The matrix AB is a square m x m matrix
and AN is an m x (11 - m) matrix. A solution of XB may be determined in terms of the
nonbasic variables XN:

When evaluating an extreme point, all non basic vector elements are assigned values
of zero, X,\, = O.
Consider the following set of inequality constraints:
Xl + 2X2 :::; 2
X2 :::; 1-

Since there are only two control variables, this constraint set can be interpreted with
the use of Figure 6.1.
The goal is to be able to determine each of the extreme points (1, 0), (2, 0), (1-, 1-),
and (1, !-) with basic and nonbasic variables.
290 Optimization of linear Mathematical Models

1
2

Figure 6.1 Evaluation of extreme points.

Consider the extreme point (1, 1). This point occurs at the intersection of the
equations Xl + Xl = 1 and .xl = 1. In other words, these equations are active con-
straints. Now consider the constraint set in standard form. It is
=2
-1.
-1

- Xs = 1
where X3 and X4 are slack variables and Xs is a surplus variable.
In order for these equations to be active constraints, the following ruust hold.
_1.
- 1

- Xs = 1

1+ X4 = t
t + 1- x , = 1
or X4 = 0 and Xs = O. For Xl + 2Xl + X3 = 2 to be an inactive constraint, the
following equation must be satisfied.
Xl + 2X1 + X3 = 2
1 + 2(1) + x 3 = 2
or x 3 = 1. The solution in vector form is
I
Xl "2
1
Xl "2
I
X= X3 "2
X4 0
Xs 0
 The Standard Form 291

Now, consider obtaining this solution using matrix partitioning, basic and non-
basic variables, and matrix algebra. The constraint set is written in matrix form
Ax = b as

2 1 o
o 1
o o
Xs

Since there are three constraint equations and five control variables, the basic
vector XB will consist of three control variables, and the remaining control variables
will be assigned to Xs. Consider the following assignment:

The partitioned matrix is:

[~~~].[::]+[~~].[::]=[~]
1 0 X3 0 -1 1
Solving for XB gives a solution of XB as a function of XN:
XB = A;l[b - ANxN]

or

[
Xl]
X2

X3
= ~ +[!] [1-1 0.::
"2
1] [ ]
1-1
Expanding. it becomes:
Xl = t + X4 + Xs

Xl = t - X4
X3 =! + X4 - Xs
 \

292 Optimization of linear Mathematical Models

Since the nonbasic variables X4 and Xs equal zero, the solution is

1
"2

I
1

X = [::J =

o
This is the same result as obtained previously.
Note, if X4 and xs, the nonbasic vector XN, are assigned values other than zero, the
solution vector x does not contain the extreme point (!, t).
Obviously, the choice of variables that enter the XB and XN will determine the
extreme point under investigation. The basic and non basic variables were carefully
chosen to obtain the desired result. 'With the procedures discussed thus far, there is no
guarantee that all basic variables are nonnegative, XB 2:: o. Procedures to ensure that
only nonzero variables enter the basis will be established in the following section.

EXAMPLE 6.1 A ConstructionManagementModel

A construction company is contracted to excavate 6-ft and 18-ft-widetrenches. It may remove no
more 10,000 ydJ>jdayof excavation material from a site because of a limited supply of dump trucks.
To meet schedule requirements it must excavate at least 1600yd3jday from the 6-ft trench and at
least 3000 yd 3/day from the 18-ft trench. The contractor has 12 backhoe operators that he may
assign either to the -t-yd3 tractor backhoe or the 2-t-yd3 hydraulic backhoe.
The performance and cost data for the backhoes are

BACKHOE OPERATING COST DAILY PERFORMANCE

~-yd3 tractor $394/machine day 200 yd3/day for 6-ft trench

2~-yd 3 hydraulic $1.110/machine day 1.000 yd3/day for 18-ft trench

(a) Formulate the problem to minimize daily operating cost.

(b) Formulate the mathematical model in standard form.
(c) Define the slack and surplus control variables in terms of the original formulation.
(d) Assume that eight and four workers are assigned to the tractor and hydraulic backhoes,
respectively. Interpret the meaning of the slack and surplus variables.
Solution
(a) Define the control variables as
Xl = number of machine operators assigned to operate the -t-yd3 tractor backhoe.
x2 = number of machine operators assigned to operate the 2-!-yd3 hydraulic backhoe.
The total daily cost to operate is
C = $394Xt + $11lOx2 Operating cost
The total amount that he may remove from the site each day because of the limited num-
ber of trucks will beequal to the cubic yards of material removed from each trench per day; or
200xt + 1OOOx2 :s; 10,000 Trucks
 The Standard Form 293
In order to meet schedule requirements, an appropriate number of operators must be
assigned to each trench:
200x I 1600
;:::: 6-ft trench
1000x2 ;::::3000 18-fttrench
Since the contractor has only 12 workers, the contractor's assignment of operators must
satisfy the following condition:
Workers
Owing to the limited number of trucks, it may not be necessary to assign all 12 operators;
hence the inequality constraint is specified. The mathematical model is
Minimize C = $394xI + Sl110x2 Operating cost
200x I + 1OOOx
2 ~ 10,000 Trucks
200xI ;::::1600 6-ft trench
1OOOx
2 ;::::3000 18-fttrench
x~ ~ 12 Workers
XI ~ 0, X2 ~ 0
(b) To represent this constraint set in standard form, slack and surplus variables must be
introduced. Since the first and fourth constraint equations have less-than-or-equal-to con-
straints. slack. variables X3 and X6 will be introduced. The second and third constraint
equations have greater-than-or-equal to constraints; thus, surplus variables X4 and X5 are
introduced. The model in standard form is
Minimize C = 394xI + 1110x2

200xI + 1OOOx2+ XJ = 10,000 Trucks

200xl = 1600 6-ft trench
- Xs =3000 18-fttrench
Workers
Xl ~ 0, X2 ~ 0, XJ ~ 0, X4 ~ 0, Xs ~ 0, X6 ;::::0

In matrix form, the objection function and constraint set may be represented as
Minimize C = c' x

Ax = b

x;::::O
with
1000 0
°
A=

x'
roo200

= [Xl X2
°1
0
1000
1
X3 X4
°0
0 °
0
Xs
-1

X6]
0
.,-1

° !l
b' = [10,000 1600 3000 12]
294 Optimization of linear Mathematical Models

(c) Let the slack and surplus variables be defined as:

X3 = amount of excavation material below the maximum amount of 10.000 yd ' that is
removed from the site per day, in cubic yards per day.
x~ = amount of excavation material above the minimum scheduled amount of 1600 yd ' that
is removed from theo-It trench per day, in cubic yards per day.
X5 = amount of excavation material above the minimum scheduled amount of 3000 yd ' that
is removed from the 18-fttrench per day, in cubic yards per day.
X6 = number of operators that are not assigned to work on this project
(d) Let x I = 8 and x 2 = 4. For the truck constraint equation, solve for x 3 in terms of x 1 and x 2:

X3 = 10,000 - 200x1 - IOooX2

X3 = 10,000 - 200(8) - 1000(4) = 4400

Since X3 -# 0, the constraint equation is an inactive constraint. The total trucking capacity is
not being used. An additional 4400 yd' of material may be removed from the site if more
operators are assigned to backhoes.
From the 6 ft trench constraint equation, solve for X4 in terms of Xl:

X4 = 200x I - 1600
X4 = 200(8) - 1600 = 0
Since X4 = 0, the constraint equation is active. This may be interpreted as meaning that the
minimum amount of material is being removed from the 6-ft trench to meet sched~~-
ments. ~
From the 18 ft trench constraint equation, solve for x 5 in terms of Xl:
X5 = lOOOX1- 3000

X5 = 1000(4) - 3000 = 1000

Since X5 > O. the constraint is inactive. More than enough material is being excavated each
day to meet the minimum schedule requirement. Excavating with four hydraulic backhoe
operators produces an output 1000yd3/day greater than the minimum schedule requirement
From the worker constraint equation, solve for X6 in terms of x 1 and Xl:
X6 = 12 - Xl - Xl

X6 = 12 - 8 - 4 = 0
Since X6 = O. al112 operators are assigned to the project, and the constraint is active.

EXAMPLE 6.2 Formulating a Basis Model

(a) Use matrix partitioning and matrix algebra to solve for a basic vector consisting of Xl and X3
in terms of Xl and X4•
2x 1 + 3x 1 + 2x 3 + X4 = 14

3x1 + 2~3 + X4 = 6
(b) Use matrix partitioning and matrix inversion to solve for a basic vector consisting of X 2 and
X4 in terms of Xl and X3'
 The Standard Form 295
Solution

(a) Write the equation set in the matrix form Ax = b:

[o2 3 2 1].
3 2 1
[;~l [14]
X3

X4
=
6

The solution XB = Ai; I[b - A,\xNl The partition matrix is

where

Express the partition matrix as

ABxB + Asxs = b

Rearrange the columns of A and the rows of x as

[~~~:l[jil = [I:]
[~ ~]-[::] + G :] -[::] = [I:]

[:::J = [~ ~r{[I:] - [~ ~l[::]}

I•

[::} H~-~H[I:]- [! :]t:]}

[:::J G]- [~ ~l [::]
=

Expanding. the solution becomes

7::'. ~. D
(b) The partitioned matrix where x~ = [X2 X4] and x~. = [Xl X3] is: ( Cl$ tk "',;? )".1.
tk~, v,,!l,d! i&·~
.\l~ )
[~ :]. [::] + [~ ~]. [:;] = [':] c..I.; j~r"l'q
Since the inverse of AB does not exist, no solution of XB exists when Xl = X3 = O,
 296 Optimization of linear Mathematical Models

EXAMPLE 6.3 Minimum-Cost Aggregate Mix Model

Aconstruction site requires a minimum of 10,000 yd ' of sand and gravel mixture. The mixture must
contain no less than 5000 yd ' of sand and no more than 6000 yd ' of gravel, Materials may be
obtained from two sites:

DELIVERY PERCENT PERCENT

SITE COST (SJyd3) SAND GRAVEL

1 $5 30 70
2 $7 60 40

(a) Formulate and solve the problem by graphical means.

(b) Rewrite the mathematical model formulated in standard form.
(c) Define the slack and surplus variables.
(d) Identify the basic and non basic variables found in part a, and show that the same solution
obtained graphically may be found utilizing XB = Ai l[h - ANxNl
(e) Determine the minimum cost by use of matrix methods, where C* = cx*.
Solutions
(a) The linear mathematical model is
Minimum C = 5xI + 7X2
Xl + x2 ~ 10,000 Delivery
0.3Xl + O.6X2 ~ 5000 Sand
O.7Xl + OAx2 ~ 6000 Gravel
Xl ~ 0
X2 ~ 0
where the control variables Xl and X2 are defined as
Xl = amount of material taken from site 1 in cubic yards.

X.2 = amount of material taken from site 2 in cubic yards.

The feasible region is shown in Figure 6.2. The graphical solution gives
xj = 3300 yd3
xi = 6700 yd '
(b), Since the delivery and sand constraint equations have greater-than-or-equal-to constraints,
surplus variables must be introduced. Sincethe gravel constraint equation has a less-than-or-
equal-to constraint, a slack variable must be introduced. Thus, the linear mathematical
modelin standard form is
Minimum C = 5x1 + 7X2
= 10,000

0.3x1 + O.6X2 = 5000

O.7XI + OAx2 + Xs = 6000

 The Standard Form 297

20,000

0.7x, + OAx2 = 6,000

co x, + x2 = 10,000
ci: 10,000
x· = (3,300, 6,700)
0.3x, + 0.6x2 = 5,000

Pit A (ft3)

Figure 6.2

In matrix notation the model may be stated as

Minimum C = cx
Ax = b
x~O
where
c' = [5 7 0 0 0]
x' = [Xl X2 X3 X4 XsJ
1.0 -1 0
A =
[LO
0.3 0.6 0 -1
0.7 0.4 0 0
b' = [10.000 5000 6000J

(c) The slack and surplus are defined as

X3 = excess quantity of mix in cubic yards delivered from sites 1 and 2 to the construction
site.
298 Optimization of linear Mathematical Models.

X4 = amount of excess in cubic yards that is contained In the mix above the rrururnum
50 percent sand requirement.
X5 = amount of gravel in cubic yards that is contained in the mix below the maximum 60
percent gravel requirement.

(d) From Figure 6.2, the optimum solution occurs at the intersection of x I + X2 = 10,000and
O.3xl + 0.6X2 = 5000. These equations thus being active constraints. the control variables
X3 and X4, must therefore be members of the nonbasic vector Xs. The remaining variables

Xl'x2, and X5 are basic variables XB. The constraint set Ax = b may be expressed as

The solution is

Since x", = 0, the solution of XB or optimum solution x* is

Xl] [1.0 1.0

x* = = 0.3 0.6
[ X2

X5 0.7 0.4

(e) The solution vector x* from part d is

. xi]
xi 6700
[3300
x* =
[ x!
;; =

1000
~

The definition of the control variables may be used to interpret the solution x". Since x; = 0,
a minimum amount of 10,000yd ' of material is delivered to the site. The amount of material
taken from pits 1 and 2 is 3300 yd3 and 6700 yd ', respectively. Since x! = 0, the minimum
amount of sand (50 percent of the mix is sand) is delivered. Since x; = 1000 yd ', the mix
contains less than the maximum amount of gravel permitted.
The minimum total cost may be obtained by using the cost function:

C* = c'x*

3300
6700
C* = [5 7 0 0 0]· ~
[
1000
C* = $63,400
 The Standard Form 299

PROBLEMS

Problem 1
Consider the followingconstraint set:

::;4

(a) Remove the inequality constraints by adding slack and surplus variables X4• Xs. and X6.
Identify X4• Xs. and X6 as being slack or surplus variables.
(b) Use matrix partitioning and matrix inversion methods to solve for a basis consisting of
X2,X3.XS•
(c) Solve for a basis consisting of X2, X3• X4•

Problem 2
2~1+X2::;9 (1)

Xl - 2X2 ::; 2 (2)

-3XI + 2X2 s 3 (3)

Xl 2:: 0, x2 2:: 0

(a) Draw the feasibleregion on a graph.

(b) Indicate all the extreme points 00 the graph.
(c) Rewrite the constraint set with slack variables.
(d) (i) If Eqs (1) and (2) are active constraints, what is the extreme point on the graph? Call it
XB·
(ii) Define the nonbasic and basic variable vectors XB and XN•
(iii) Solve for the basic variable vector by matrix methods.
(iv) Compare (iii) with (i). The two solutions should be the same.

Problem 3
Maximize z = -2X2 + Xl
X2 2:: 1 + Xl

Xl unrestricted, X2 2:: 0

Since Xl is unrestricted in sign, let Xl = X: - Xl, where x:

and Xl are two new control variables
that replace x r-They are nonnegative values satisfying the followingconstraints:

Xl 2:: 0

Substitute (xi - Xl) for Xl in the given problem.

(a) Formulate the problem using xi, Xl' and X2 as control variables.
(b) Write in the standard form of linear programming.
300 Optimization of linear Mathematical Models

Problem 4
An aggregate mixture contains a minimum of 30 percent sand and no more than 60 percent gravel
nor 10 percent silt. Three sources of aggregate are available.

PIT 2 3

% Sand 5 30 100
% Gravel 60 70
% Silt 35
Cost/vd? $2 $10 $8

Determine the mix of minimum unit cost, $/yd3

(a) Formulate a minimum cost modeL Define all control variables with proper unit measures.
(b) Write in the standard form ofIinear programming. Define slack and surplus control variables
and their units.

Problem 5
A sand and gravel mixture is to consist of exactly 50 percent sand and 50 percent gravel, Three pits
are available. The percentage of sand and gravel and the cost of each mixture are:

PIT 2 3

% Sand 35 40 45
% Gravel 65 60 55
Unit cost/vd? $8 $9 $11

Since it is impossible to obtain an exact 50-50 mixture of sand and gravel, gravel will have to be
removed at a unit cost of S4jyd3, and sand will have to be added to the mixture at a unit cost
of S20/yd 3 of sand. A total mix of 10,000yd 3 is required.
(a) Formulate a minimum cost model. Define the control variables and their units.
(b) Write in the standard form of linear programming. Define slack and surplus variables with
proper unit measures.

Problem 6
Towns A and B produce 3 and 2 Mgaljday of wastewater, respectively, The BOD level of the
wastewater is 200 mg/I. (See Figure 6.3_)Towns A and B belong to a regional water-treatment
cooperative that operates water-treatment plants I and 11_Plants I and II have a capacity of 3
and 4 Mgaljday and operate at a BOD removal efficiencyof 90 and 75 percent, respectively.
The cooperative has an authority to assign the wastewater flow from towns A and B to either of
plants I and II, as shown in the schematic diagram. Formulate a system-analysis model to mini-
mize total annual treatment and pumping costs subject to the constraint that the total flow dis-
charged into the river does not exceed 30 mgJI of BOD. The unit treatment and pumping costs
 The Standard Form 301

Effluent
Figure 6.3

for each pipeline are given in the table. Let Xl. Xz. X3. and X. be the wastewater flow in million
gallons per day assigned to pipelines 1.2, 3, and 4. respectively.

UNIT ANNUAL TREATMENT

AND PUMPING COSTS
(51000/Mgal/dav-yr)
PIPELINE

1 $46
2 $50
3 $55
4 $40

(a) Formulate a minimum cost model Define the control variables with unit measures,
(b) Write in the standard form of linear programming. Define slack and surplus variables with
proper unit measures,

Problem 7
A contractor will bid on a building construction project-that consists of the following tasks:

PAYMCNT SCHEDULE
TASK MONTH

1. Excavatesite. remove 30.000 yd3 of soil 3rd

2. Pour concrete foundation. 300 yd3 of concrete 6th
3. Erectsteel frame. 14Aoo linear ft of steel 12th
 302 Optimization of Linear Mathematical Models

A payment will be made after each task is completed. An estimate of the completion date of each
task is indicated in the payment schedule.
The contractor has submitted a successful bid of $400,000 on the entire project.
The contractor is a shrewd businessman, and he knows that it is better to charge larger sums
of money for the earlier completed tasks and lesser amounts for later tasks. In other words, he
would like to maximize the present worth of the money received. This type of bidding is called
unbalanced bidding. The contractor will be able to accomplish his objective by adjusting the unit
costs he charges for each task. His past experience shows he must charge between $ L50/yd 3 and
$3.00 for soil excavation, at least $1OO/yd3 for cast-in-place concrete, and, but no more than $1OO/ft
for steel erection. Assume an interest rate of 12percent per year.
(a) Formulate a mathematical model to meet the contractor's objective. Define control variables
and their units.
(b) Write in the standard form of linear programming. Define slack and surplus variables with
proper units.

6.2 THE SIMPLEX METHOD

The method will be used to determine the optimum solution of the following form:

Minimize z = c' x

Ax =b
x ~"O
where all elements of the vector b are positive, b ~ O.
Our development of the simplex method will be established to minimize the objec-
tive function, z = c'x. A maximization problem is solved by transforming the maxi-
mization problem to a minimization problem, and then, the minimization simplex
algorithm is used to obtain the location of the optimum. Thus, we are"able to solve
both minimization and maximization problems with the same method. The simplex
method is an iterative method to find the optimum solution of a linear mathematical
model. It consists of the following steps:
1. Establish an initial candidate" solution xo.
2. T~ the initial candidate solution XO to determine ifit isthe optimum solution x*.
3. If the test for the optimum solution fails, establish anew candidate solution xl,
and repeat steps 2 and 3 until the optimality test is satisfied.
The superscripts 0 and 1of XO and Xl denote the initial estimate of x· and the estimate
of x· for iteration 1. For an intermediate iteration k, the candidate solution of x* is
denoted as ~. If the optimality fails, ~ =1= x· (step 2), then "anew candidate solution
~+ 1 is determined (step 3). " .
The simplex method involves a hill-climbing approach of moving from one extreme
point to another one. New candidate solutions are established. by comparing the
relative values of the slopes of the objective function "and"selecting the point that not
only minimizes the objective function. the greatest amount, but also ensures "that the
new candidate is a feasible solution satisfying the constraint set.
 The Simplex Method 303

.fatrix algebra is used to explain the basic properties of the algorithm. Since the
hod is an iterative procedure, it is particularly suitable to computer solution.

ndamerrtal Relationships
. candidate solution x satisfies the constraint set, Ax = b and x ~ 0, it is a basic
.sible solution and a candidate for optimum solution of z = c'x. The superscript k
.H be dropped in this section because it does not add to the understanding of
:Se procedure. A candidate solution will be designated as the partitioned vector,
_.. = [XB xsl The basic candidate solution vector XB may be written as a function of
-:"{tenonbasic variables X, ...;, or

.xpanding this equation gives

XB = Aii1b - AiiIAJ\'XN
The value of z, written as a function of the nonbasic variables XN, is determined by
bstituting the candidate solution in the objective function:
z = c~[(Aii 1 b - A; I ANxN)] + C~XN
'" here c' = [c~ c~]. Collecting the coefficients of xs, we may write z as
z = c~A; 1b + [«,. ~ c~A; 1AN)]x",

"n a latter part of our discussion we refer to [(c~ - c~A~AN)] as c~, the cost indicator,
Since Xs are non basic variables and all elements of XN are equal to zero, the values
of the candidate solution XB and candidate optimum solution z may be easily deter-
mined. Substitute Xs = 0 and simply.
xB=A;lb

This analysis shows that a candidate solution XB and the value of the objective
function z are obtained by establishing a basic matrix AB•
In the simplex method, the choice of columnvectors of A, a, that enter iB will be
'Selected to ensure that XB is always a basic feasible solution. Furthermore, in each
teration the basic matrix is selected to ensure that new values of XB will reduce the
-'alue of z or. at least, be equal to the value of z in the previous iteration. Thus, our
.iscussion will begin by assuming that a basis feasible candidate solution ~ exists,
~khere k = 0, 1,2,. .... Recall that all basic feasible solutions are nonnegative vectors,
x
-:~~ O. 1 ~ 0, x; ~ 0, ....

eThe Test for an Optimum Solution

A method for establishing if Z(Xk) = -J< is the optimum minimum solution z* must be
developed. In other words, we must answer the question, is ~ = x* the location of
the minimum z*? We introduce the basic concepts by example.
304 Optimization of linear Mathematical Models

Suppose the objective function is

z = 2Xl + 3x1 + SX3 - 2X4

and the candidate solution Xk = [x~ x~J, is

~ = [1J = GJ x~ = [;.J [~J =

The value of the candidate solution is
Zk = 2·2 + 3·1 + s·o - 2·0 = 7
The candidate solution Xk mayor may not be the location of the optimum minimum
solution. We may determine if the current basic feasible solution x~ is the optimum
solution by evaluating the elements of the cost indicators c~.
(z - Zk) = C~'XN

Substitute ~ = 7 and assume the objective function is a function of the nonbasic

variables:
(z - 7) = SX3 - 2X4

or
z = 7 + SX3 - 2X4

Thus c~ = [S - 2]. If x~ is the location of the minimum solution, all other values of x
in the feasible region will be greater than the candidate solution Zk or z > Zk. If x~ is
not the location of the optimum solution, then a new basic vector x~+ 1 must be
established. A non basic variable X3 or X4 must be entered into x~+ 1, and a basic
variable Xl or Xl must be removed. In order for either x 3 or X4 to enter ~ + 1, there
must be positive values to satisfy the condition of a basic feasible solution XB ~ O.
This is an important restriction that will be included in our test for optimality.
First, assume X3 is entered into the basis and X4 remains a member of the nonbasis,
X4 = O.Thus

zk+ 1 = 7(+-5)3
......_~"""'._.(..t<

Since X3 is restricted to be a positive number, X3 ~ 0, the value of zk+ 1 = 7 + SX3 will

always be greater than Zk = 7. Thus, we can conclude that entering X3 into X~+ 1 will
not reduce Zk. The present basis x~' = [x~ x~], remains a candidate for the optimum
minimum solution.
Next, X4 is assumed to be entered into the basis ~ + 1 and x 3 remains in the nonbasis,
x 3 = O.The objective function is

Since X4 must be a positive number, the value of Zk+ 1 will be less than ~ for X4 > O.
Thus, the candidate solution Zk = 7 may be reduced by entering X4 into the basis.
The current candidate solution ~'. = [~ ~]. is not the minimum solution and a new
 The Simplex Method 305

candidate solution xk+ 1 = [x1/ 1 x~+ 1] must be determined. We discuss the removal
of Xl or X2 from Xk+ 1 in the subsection entitled "Establishing a New Basis."
This example illustrates the essential steps utilized in testing a candidate solution
as the optimum solution. We shall establish a more general test with the c~vector.
Since anyone of the control variables of x~ may enter Xk + 1 and form a new basic
feasible solution, the sign of coefficients of c~ will indicate if the solution oTZk will be
increased or not. If all indicators c1 are positive, introducing any member of x,,· into
the basis will increase the value of Zk. If this condition occurs, then the location of the
optimum point has been determined, z* = :1'. We state this as a rule.
RULE 1 If the sign of all indicators of c~are positive, then the current basic feasible
solution Xk is the location of the optimal minimum solution, x* and Zk = z*.

Establishing a New Candidate Solution

If one or more of the indicators of c~are negative, the candidate solution x~ is not an
optimum, and the solution of Zk can be reduced. If there is only one value of c1 that is
negative, then the corresponding nonbasic variable is entered into a new basis. Only
-one variable of~. is entered into x~+ 1 during this step. Iftwo or more of the indicators
of ~ are negative, then the variable associated with the smallest negative value of c~
is entered into x~+ 1. For example, suppose that z - Zk = ck x~. is
(z - 8) = 3X3 - 4X4 - 8xs
Since the coefficients of X4 and x 5 are negative, entering either one of them into the
basis will reduce z, Clearly, introducing Xs into ~+ 1 will reduce z a larger amount;
therefore, it is chosen as the variable to enter x~.
Entering the variable with smallest negative ~ coefficient does not ensure that the
value of =1; is reduced the greatest amount, because the constraints may limit the
range of the new basic variable. For example, consider the situation where Xs can
only increase to x 5 = 1 while X4 may be increased to X4 = 3; X4 is thus the better
choice. However, the gain is generally not considered with the extra computer pro-
gramming effort, hence Rule 2.
RULE 2 If one or more of the indicators of ~ are negative, the current present
solution can be reduced. The variable of x~ that enters the new basis, x~ + 1, denoted as
xp' is the one associated with smallest negative value of c~.

We can interpret Rule 2 by tracing the possible paths from the current candidate
solution at point a = xo. (x~, x~), to the optimum solution at point c = x*, (x1, xi) as
shown in Figure 6.4. For the hypothetical situation, we assume that there are two
negative indicators of~. With two negative indicators, there are two possible paths
to follow, from extreme point a to extreme points b or c, respectively. In following
Rule 2. we find the smallest negative indicator and introduce the corresponding
control variable into the basis. In Figure 6.4 this will be equivalent to moving from
extreme point a to extreme point c. In this case, we find the minimum or optimum
solution x* in one iteration. Ifwe choose the other path, extreme points b and d must
be evaluated. These points are not the location of the optimum solution; therefore,
306 Optimization of linear Mathematical Models

Z1

~-----------------------------------------X1
Figure 6.4 Hill-climbing approach and the search of
extreme points.

two other iterations are required. Choosing the smallest negative value of c~ may
eliminate extra iterations.
By choosing the greatest negative valueof~, we proceed to move down a path with
the steepest descent. If we were maximizing, we would follow a path of steepest
ascent; for this reason the simplex method is often called a hill-climbing procedure.
By following Rule 2, we attempt to proceed to the minimum solution in the least
number of iterations.

Establishing a New Basis

By Rule 2, we establish which control variable enters into a new candidate basis. We
must determine which variable of the current basis x~ must be removed and placed
in the nonbasis x~+ 1. In essence, we are trading places. One control variable in the
basis is being removed, and a control variable of the nonbasis is taking its place. The
control variable that is removed from the basis is placed in the nonbasis. This approach
satisfies the requirement that the number of control variables in the basis must be
equal to the number of constraint equations. If there are m constraint equations, there
must be In control variables in ~+ 1.
 The Simplex Method 307

For example, suppose we have a linear mathematical model consisting of three

constraint equations m = 3, and five control variables n = 5. Furthermore, let us
assume that the current basis contains

After completing the operations indicated by Rule 2, we find that x, is to be entered

into the basis. If x s is entered into x~ + I, one of the basic variables Xl' X3, or X4 must
be removed and placed in x~+ 1. Suppose the constraint set is written as follows:

[::1 m- r-iJ-X5
=
or
Xl = 1 + x,
X3 = 5 - x,
x4 = 3 - 2xs
Since our goal is to remove one of the current control variables from ~, one of the
control variables x I, X3, or X4 must be entered in X~/ 1. This goal is accomplished by
setting one of the variables of x~ equal to zero and determining if the remaining
variables of ~ are basic feasible solutions.
Assume x 1 becomes a member of ~.+ 1, or Xl = O.Solving for X s, we find
Xl =0= 1 +xs
Xs = -1
Since X5 is negative, the basis ~ + I' = [xs X3 x4] cannot be a basic feasible solution.
Assume X3 becomes a member of ~.+ 1, or X3 = O.
X3 = 0= 5- Xs

Xs =5
Since x , > 0, it is possible that X3 can be a member of xt+ 1 if the remaining variables of
XB, Xl- and X4 are positive variables.
x I = 1 + X s = 1 + (5) =6

X4 = 3 - 2xs = 3 - 2(5) = - 7
Since X~ < 0, the basis x~+ r: = [Xl Xs x4], is not a basic feasible solution.
Assume X4 becomes a member of XN, X4 = 0:
X4 = 0 = 3 - 2xs
308 Optimization of linear Mathematical Models

Determine x 1 and x 3 :
Xl = 1 + Xs = 1 + (3) = 4
X3 = 5- Xs = 5 - (3) = 2
Since all values Xl, X 3, and X 5 are positive, a new basic feasible solution has been found.
The basic feasible and nonbasic vectors are

The same result may be obtained by observing that the variable associated with the
minimum nonnegative ratio of bUa~,p is the variable that is removed from x~ and
placed in ~,+ 1. From the foregoing example x~+ 1 = bk - x~a~, where xp = Xs may
B 1 + -:I
be written as xk+ p akp = , or e.

For simplicity in notation, the superscript k is removed. The ratios of bJai,p may be
written as
bl
-1
a1• s -1
b3 5
a3, s

b4 3
-
a4.5 2
Since bt/al. 5 < 0, Xl cannot enter the new
XB' Since b4/a3. 5 is the smallest positive
rat io, Xq =
is removed from the basis.
X4
This example gives the essential steps used in determining which control variable is
removed from the current basis ~ and is placed in the new nonbasis xt+l. A general
procedure for performing these steps may be establishing with the use of the equation:
XkB + xka
p k p = bk

The control variable xI' of x~ is defined to be the variable to be removed from 4 and
x;
placed in .l{+ 1_ The set of equations x~ + ak = bk may be stated without the
superscript k as
t = 1,2, ... , m

The determination of Xq proceeds by selecting only those equations where the sign
> O.The control variable.x, , associated with the minimum value
aIat,p is pasitive.at.p
 The Simplex Method 309

of b,/a,. P' will be the control variable that is removed from XB and placed in XN' Since
b, must be positive by definition, b ~ 0, all ratios must be positive. This procedure is
stated as Rule 3 of the simplex method.

RULE 3 Calculate the ratio of bfa., p for all at. p > 0 in the current basis,

The control variable Xq that corresponds to the smallest positive value of b,jat,p is
removed from the basis and placed in the nonbasis. The control variable xp replaces
Xq in the basis k + 1. The superscript k is not shown for simplicity in notation.

In case of a tie among minimum value ratios of briar. p' anyone of the control
variables associated with the minimum ratio may enter the basis. If no variable
satisfies Rule 3, the domain is unbounded, and no minimum value exists. When this
result occurs, no practical real-world solution exists, because, most likely, the problem
was incorrectly formulated. In this case, the mathematical model must be reformulated
and solved again.

The Simplex Tableau

For hand calculations, the three rules of the simplex method and the method of succes-
sive elimination are generally used. The linear mathematical model may be written as
Minimize (z - z") = cr 4 + (r~

A~xt + lx~ = bk
4=0
x~ ~ 0
where ~ and x~. represent a basic feasible candidate solution of the minimum of z.
Special note should be made that the unit costs associated with ~ must be zero,
CB = 0, and that all members of bk must be positive. The candidate solution ~ may
be written in an augmented matrix as

X~ A~ I bk b/aJ
[ - cr 0' (z - Zk) -

The column vector ~ contains a list of the control variables that are current members
of basis vector. It is understood that the variables that are not listed in·~ are members
of the non basic vector ~. and, by definition, are equal to zero, xt
= O. The numerical
values of the basic variables x~ and the objective function Zk are recorded in the tableau
as bk and (z - Zk), respectively. The current values of the A~ matrix and the indicators
of ~ are placed in the appropriate location of the augmented matrix. The column
bu is reserved for performing the calculation required by Rule 3.
In Section 6.3, the selection of the initial basic feasible solution x~, k = 0, will be
discussed in greater detaiL In this section we consider a special form ofthe linear model
310 Optimization of linear Mathematical Models

where the linear mathematical models consists of a set of constraint equations with
less-than-or-equal-to constraints only.
Minimize z = c' x
i = 1,2, ... , m
xzO
Slack variables are introduced, and the model is written as
Minimize z = c' x

Ax = b
xzO
For this special case the easiest method to establish an initial basic feasible solution
x~ is to assign all slack variables to x~. No calculations are needed for this initial guess.

Maximization Problems
The optimum solution of maximization problems of the form:
Maximize z = c'x
Ax = b

xzO
may be transformed to a minimization problem by multiplying the objective function
by -1. Define the objective function as .
a = (-I)z = -c'x
Finding the location of the minimum value of a is equivalent to determining the
maximum value of z, This operation does not alter the shape ofthe objective function;
it only changes the sign of the function. The transformed mathematical model is
Minimize a = -c'x
Ax = b
xzO
The minimum value of a may be determined with the simplex method. The optimum
solution is designated at x* with a*.The maximum value of z is located at x* with
maximum z equal to
z* = -a*
We illustrate the basic concept of solving a maximization problem with a one-
dimensional model (see Figure 6.5.):
Maximize z = 2x I
Xl ~ 4
 The Simplex Method 311
z a x*
2 4

z* =- a" = - (- 8) = 8

(a) (b)

( c) ( d)

Figure 6.5 Solving a maximization problem as a minimization problem.

The optimum solution as shown in Figure 6.5a is xi = 4 and z* = 8. Solving this

problem as a minimization problem with a = - z, the model is

Minimize a = - 2x I
Xl ~ 4
The optimum solution as shown in Figure 6.5b is x* = 4 and a* = - 8. Since z = - a,
the maximum value of z is equal to z* = 8. The location of the optimum solution xi
for the two problems is the same extreme point, xi = 4.
 312 Optimization of linear Mathematical Models

The optimum solution of the following two-dimensional maximization model is

shown in Figure 65c.
Maximize z = XI + Xl
Xl + 2Xl ::::;;4

Xl ~ 0

Xl ~ 0

The optimum solution is x* = (4, 0) with z* = 4. Transforming this model to a

minimization model by multiplying z by - 1 gives the same result as shown in Figure
6.5d:
Minimize a = -Xl - Xl

Xl + 2X2 ::::;;4

Xl ~ 0

Xl ~ 0
The optimum solution x* is (4,0) with a* = -4. Comparing the two solutions shows
that the location of optimum solution is the same. Since z = - a,the optimum solution
is z* = -a* = 4 and the same maximum value of z is obtained. The transformation
a = ( -1)z will be used to solve multidimensional maximization problems with the
simplex algorithm for minimization.

EXAMPLE 6.4 A Minimization Problem

(a) Use the simplex method to determine the solution of the following linear mathematical
model:
Minimize z = -llxl - 4x].

3x1 + 5x1 s 15
5x1 + 2X2 ::; 10

Xl ~ O'X2 ~ °
(b) Trace the path to the optimum solution on a graph.

Solution
(a) Slack variables X3 and X4 are added to the constraint equations to remove the inequality
constraints:
Minimizez = -llxl - 4X2
3x1 + 5x2 + X3 = 15
5x1 + 2X2 + X4 = 10
 The Simplex Method 313

There are two constraint equations, m = 2; therefore, the initial basic feasible solution x~
must contain two variables. The easiest way to establish the initial basis is to assign all slack
variables to x%,: ~

Initial Solution
XB Xl Xl X3 X4 b b/a
3 5 0 15
[XJ ~'~ 5J
X4 G) 2 0 10 ~ =2 ~
Indicator row c~ -11 -4 0 0 (z - 0)
t

The simplex method may now proceed:

RULE 1 Since the C1 indicators of the variables of Xl and X2 are negative, the current solution is
not a minimum.
RU LE 2 The control variable that enters the basis is Xl because Cl = -1 I is the smallest
negative value of the indicator vector. An arrow is drawn under the Xl column to indicate that it
will enter the basis.
RULE 3 The control variable that leaves the basis is X4 because 15° = 2 is the smallest positive
bja ratio. An arrow is drawn next to row 2 to indicate that X4 is to be removed from the basis.

Iteration 1 Successive elimination is used to determine the next candidate solution: that is,
XB' = [X3 x a.
A circle has been drawn around the pivot element of the A matrix. The pivot
element will be reduced to 1 by dividing each element of the row by 5. The new row is written in
a new simplex tableau. The remaining elements in column Xl' all and Ct, must be set equal to zero
by making the appropriate row operations. In this case all elements of the X I row, row 2 in the
newly established simplex tableau below, are multiplied by - 3 and added to each element of
the X3 row, row 1 of the old tableau. In similar fashion, multiple each element of Xl' row: below,
by 11and add them to indicator row of the old tableau. The resulting simplex tableau is

First Iteration
Xb XI Xl X3 X4 b bja
0 19
"""5 -i 9
[XJ 1-
Xl 5 0 ! 2
c; 0 5" 0 5
II
Z + 22
The solution is an optimum solution because all indicator row values are positive:
1
xt = 2, x! = 9 and z* = - 22
(b) The graphical solution is shown in Figure 6.6. The equation 5Xl + 2Xl = to is an active
constraint, and the equation 3xI + 5Xl ~ 15, is an inactive constraint as shown In the
diagram.
The path of the simplex method shows the initial solution at point (0, 0). The eaedidate
solution after one iteration and the optimum solution is (2, 0). The path of the soletion is
shown by the arrow.
 314 Optimization of linear Mathematical Models

Inactive constraint
3x1 + 5x2 = 15

Active constraint
5x, + 2x2 = 10

;I' x·, Optimum solution

2

Initial solution,
Path

Figure 6.6

EXAM PLE 6.5 Maximization Problem

Determine the solution of the followinglinear mathematical model:

Maximizez = + 3x1 + x2
Xl - 2X2 ~ 10

2Xl + x2 ~ 24

(a) Transform the model from a maximizationto a minimization problem.

(b) Solve part a by use of the simplexmethod.
(c) Trace the path to the optimum solution on a graph.
(d) On a graph show that the locationsof optimum solutions x· for the minimization and maxi-
mization problems are at the same extremepoint.
 The Simplex Method 31 5

Solution
(a) The simplex method utilized in this text requires that the objective function be a minimum.
By multiplying the objective by - 1, the objective function is inverted without changing the
shape of the function. Define the transformed objective function as a.

a = ( - l)z = - 3x I - X2

Thus, the objective function is:

Minimize a = -3Xl - X2

(b) Slack variables X3, X4, and X5 are introduced to each constraint equation. The mathematical
model is
Minimize a = -3Xl - X2

=10
= 24

+ X5 = 5
Xl ~ 0, X2 ~ 0, X3 ~ 0, X4 ~ 0, X5 ~ 0

The slack variables X3, X4, and Xs are assigned to the initial basis x~, X3 = 10, X4 = 24, and
X 5 = 5. The simplex tableau is

Initial Solution
XB Xl X2 X3 X4 X5 b bJa

[ x,
X4 21
-2 1
0
0
1
0
0
10
24
~ =~ 1210]
224

X5 CD -1 0 0 1 5 t = 5 .-
c1 -3 -1 0 0 0 (a - 0)
i
The simplex method for minimization of a proceeds as follows.
RULE 1 Since the C1 indicators of X 1 and x 2 are negative, the initial solution is not a minimum.
RULE 2 The control variable X 1 enters the basis because C I = - 3 is the smallestnegative
value of CI. An arrow is drawn under the X I column to indicate that X I enters the basis.
RULE 3 The control variable X5 leaves the basis because bla = 5 is the smallest positive
value of the bJa vector. An arrow indicates that X s leaves the basis.
Iteration 1 Successive elimination is used to determine the next candidate solution, that is,
x~ = [X3 X4 Xl]. The result is

XB Xl ,Xb X3 X4 Xs b b]a
0 0 5

[X' X4

Xl
0
1
-1
Q)
...:..1
1
0
0
1
0
-1
-2 14
5
¥]~
Cr 0 -4 0 0 3 (a + 15)
i

'~
316 Optimization of linear Mathematical Models

RULE 1 Since the c, indicator of Xl is negative, the candidate solution is not an optimum mini-
mum solution.
RULE 2 The control variable Xl enters the basis because it is the only negative indicator of C/.
RULE 3 The control variable X4 leaves the basis because bla = !~is the only positive ratio
of the bla vector. Negative values of the element of Qi.p are not considered in the b]a elevation.
A circle has been drawn around the pivot element of the A matrix. Successive elimination is
performed. The new simplex tableau after the second iteration is
Second Iteration
XB Xl Xl X) x4 X5 b b[a
I 5 29
0 0 1 3 -3 "3
[X3Xl 0 1 0 I
3
2
3
14
"3
I 1 29
Xl 0 0 3 3 "3

cJ 0 0 0 3
4 1
3 (a + l~l) 1
x2

~tionof
decreasing
values of a

20
x·1

Figure 6.7
 The Simplex Method 317

RULE 1 Since all indicators of CI are positive, the candidate solution is the optimum minimum
solution.
The optimum location is

xt = 11 = 91 x! = 134 =4t

The optimum minimum of a is

-lOl
a* =--= -33t
3
Since z = -la, the maximum of z is
z* = (-1)( -331) = 33~
(c) The path to the minimum solution is shown in Figure 6.7. The path to the optimum solution
initiates at (0, 0), proceeds to (5, 0), and then to optimum solution at (9t, 4t).
(d) The feasibility region for the maximization of z or for the minimization of a is the same. The
feasibility region is shown in Figure 6.8.

Aectionof
increasing z

15 20

Figure 6.8
318 Optimization of linear Mathematical Models

The maximum value of z is 33t as shown in the figure. As Xl and X2 are increased, the
value of z increases because the unit costs C1 and C2 of the objective function are positive
values; that is, C1 = 3 and Cz = 1. The minimum value a" will occur at the same location as
z* because the value of a decreases with increasing values of Xl and x2• Since a = -lz, the
unit costs of C1 and Xz of a are negative values; that is, CI = -3 and C2 = -1. MUltiplying
the objective function of z by -1 does not change the shape of the linear function; therefore,
the location of x* will not change.
Remarks
The simplex method may be used for any multidimensional linear mathematical model. Since the
original formulation of this problem contains only two control variables, we are able to point out a
shortcoming of the mathematical approach as compared to the graphical approach. The constraint
equation Xl - 2xz ~ 10 will never be an active constraint, because it is not part of the feasible
region. See Figure 6.7.

PROBLEMS
For each of the following problems:
(a) Solve by graphical means.
(b) Use the simplex method to solve the following linear mathematical models.
(c) On the graph trace the path to the optimum solution as followed by the simplex method
solved above.

Problem 1
Maximize z = l Ix , + 4xz
3xI + 5xz :;; 15
5xI + 2X2 ~ 10

Problem 2
Minimize z = Xl - 2xz

XI.- x2 ~ 8
Xl + Xl ~ 15
X2 ::; 9
Xl ~ 0, Xl ~ 0

Problem 3
Maximize z = X2

-Xl + X2 :$; 0

2x1 + X2 :$; 8
 The Two-Phase Simplex Method 319

Problem 4
Minimize z = - 5x I + 2X2
2xI + x2::S; 9
XI - 2X2 ::s; 2
-3xI + 2X2 ::s; 3
XI ~ 0, X2 ;?: °
Use the simplex method to solve the following problems:

Problem 5

Problem 6

6.3 THE TWO-PHASE SIMPLEX METHOD

This discussion focuses upon the selection of the initial basic feasible solution x~ for
linear mathematical models with =, ~, and ~ constraints. In the previous section,
only z; constraints were considered. Slack variables were added to each equation,
and the initial basic feasible solution could be easily determined by assigning all slack
variables to x~. Unfortunately, if a constraint set of linear mathematical model
contains an = constraint or a 2: constraint, the initial assignment of variables to x~
does give a basic feasible solution. For equations having greater-than constraints (2:)
surplus variables are added to the equation. This variable cannot be assigned to the
set of basic variables XB, because it violates the condition that a basic variable mustbe
a positive value, XB 2: O.Equations with equality constraints offer a different problem
in establishing an initial basic feasible solution.
The two-phase simplex method is straightforward because it a voids the need to
make calculations in selecting the initial basis. The primary reason for utilizing this
approach may appear to be solely for convenience. Not so, this approach is used
320 Optimization of linear Mathematical Models

because it eliminates the possibility of selecting an initial basis and, upon subsequent
calculation, finding that it does not satisfy the requirement for a basic feasible solution.
Let us investigate the following linear programming problem and attempt to
assign an initial basic feasible solution without performing any calculations:
Minimize z = 3x1 + 4X2 + 5X3
Xl + X2 + X3 ~ 5
2X2 + X3 Z 2

Xl + 3x2 =4
X1>X2,X3 Z0
Add slack and surplus variables to the first and second constraint equations, respec-
tively. The standard linear programming format is
Minimize z = 3x1 + 4X2 + 5X3

+ + X4 <
x1 X2 + X3 =
....,
2Xl + X3 - X5 =
,
Xl + 3.\2 =

Since the goal is to assign an initial basis without performing any calculations,
Xl' Xl, and X3 will be assigned to x,~_ The control variables X4 and X5 are assigned to
x~. Since there are three constraint equations, m = 3, the basis x~ must contain three
variables in order to satisfy the requirement that
x~ = Ai l[b - Axx~J
Clearly, another variable Xl' X2, or X3 must be assigned to x~. Unfortunately, placing
anyone of these control variables in XB will require calculations, In addition, it is
possible that this assignment will not give a feasible solution, x z O.If this happens,
another variable will have to be assigned to x~ and the calculations will have to be
repeated. This method is unacceptable for models consisting of a large number of
control variables. We shall investigate this example in greater detail and then show
how artificial variables may be used to establish an initial basic feasible solution.
The control variables Xl' x2, and X3 will be assigned to xx-
Each constraint equation will be investigated independently, Since Xl> X2, and X3
are members of x~. and equal to zero, the control variable X4 from Xl + Xl + X3 +
X4 = 5 is equal to
X4 = 5

Since X4 is positive and the associated unit cost of X4 is equal to zero, X4 is a viable
choice for the initial basis. Thus, X4 can be assigned to x~_
For 2X2 + X3 - X5 = 2, substituting zero for Xl, X2, and X3 results in Xs being
equal to
Xs = -2
 The Two-Phase Simpiex Method 321

The surplus variable X5 violates the requirement that the basic variable be positive.
Therefore, X5 cannot be assigned to a member of the initial basis x~.
Equation Xl + 3X2 = 4 presents a different type of problem. Since Xl and X2 are
assumed to be members of x~- and equal to zero, the constraint equation cannot be
satisfied.
(0) + 3(0) ¥= 4
Obviously" both x 1 and X2 cannot be members of the non basis x~.
This analysis shows that slack variables can be readily assigned to an x~. Equations
with surplus variable or equality constraints present problems.

Artificial Variables
A new approach, called the two-phase simplex method, makes the initial assignment of
control variables to x~ an easy task. The method makes use of a new control variable
called the artificial cariable. The use of artificial variables will be illustrated by intro-
ducing them to the constraint set where needed. Assign Xl> X2, and X3 to x~, and X4;
the slack variable from Xl + Xl + X3 + X4 = 5 is assigned to x~.
Add an artificial variable X6 to 2X3 + X3 - X5 = 2.
2\:2 + X3 - X5 + X6 = 2
The control variable X5 is assigned to x~.; thus, the value of the artificial variable is
X6 = 2
It is a basic feasible solution that can be assigned to XB•
For Xl + 3x2 = 4 another artificial variable Xi is introduced.
Xl + 3x 2 + X7 = 4
Since Xl and X2 are nonbasic variables, X7 must be an artificial variable.

Since X-:- is positive, it may be assigned to XB•

As a result of adding artificial variables to the basis, the initial basic and non basic
vectors are
x~'= [X4 X6 Xi] = [5 2 4J
x~: = [x 1 X2 X3 X 5J = [0 0 0 0]
Since all members of x~ are positive values and all members of x~- are zero, we have
established an initial basis and nonbasis that satisfies the requirement that x~ ~ 0
and x~- = O. The constraint set becomes
=5
2X2 + X3 =2
Xl + 3X2
 322 ·
Opti~ization of linear Mathematical Models

The Artificial Objective Function

The original model is still intact, and two extra variables, artificial variables, have
been added to the constraint set. When X6 and X7 are members of the basis, the con-
straint set has no physical meaning. It is impossible to interpret these variables
in terms of the original formulation. They offer only a mathematical convenience.
When X6 = 0 and X7 = 0, the model reduces to its original form. Hence, our goal is to
find some procedure to remove the artificial variables from XB and place them in XN.
The two-phase simplex method systematically removes artificial variables from the
basis and assigns them to the nonbasis and, concurrently, establishes a candidate
solution comprised of variables from the original formulation. Finding an optimum
minimum solution of the original mathematical model is achieved by introducing a
second objective function to the model. The purpose of this second function is to
remove the artificial variables from the model and, in the process, establish candidate
solutions that satisfy the requirements for basic feasible solutions. This can be
achieved by utilizing the rules of the simplex method discussed in Section 6.2.
The second objective function will be defined to be equal to the sum of the artificial
variables:
W=LX;
i

where the control variable Xi represents an artificial variable. From the foregoing
example, the value ofthe artificial objective function w is expressed as

The cost coefficients of X6 and X7 are both equal to one. The cost coefficients associated
with basic tariables must be zero for the condition that (z - zo) = c~x~, be satisfied.
The second objective function will be rewritten in terms of the elements of XN' By
utilizing the equations from the constraint set, X6 and X7 may be written as

X6 = 2 - 2X2 - X3 + Xs
X7 =4- Xl - 3x2

Substituting these equations into w = X6 + X7 gives

or, equivalently,

Since all artificial variables are initially members of XB and, by definition, positive
values, the value of W must always be positive. Our goal will be to minimize w until it is
reduced to zero. When w = 0, all artificial variables will be equal to zero. When all
artificial variables are members of XN' the model is a function of the originally for-
mulated control variables. At this point, the mathematical model has been restored
to its original physical meaning. The artificial objective function is introduced into
 The Two-Phase Simplex Method 323

the mathematical model as

Minimize (w - 6) = -Xl - 5xl - X3 + Xs

Minimize z 3xI + 4Xl + 5X3

Xl + Xl + X3 + X4 =5
2Xl + X3 - Xs + X6 =2
Xl + 3xl + X7 = 4
Xl,Xl""~O

The initial basis x~ for a two-phase linear mathematical model may be written in
the following general form:
Minimize (w - we) = ,ld~x~.
Minimize (z - 0) = c~x~,
A~.x~. + Ix~ = b''
where d, = hI is defined as the unit costs of the artificial objective function. The
simplex tableau for this model is

[ x~
A~.

c~
I
0
b
(z - 0)
bja 1
d~ 0 (w - wo)

The Algorithm
The two-phase simplex method uses the same rules established in Section 6.2 for
minimizing linear mathematical models. In the two-phase method, we must satisfy
the requirements of two objective functions. During the first phase, the goal is to
minimize the numerical value of 'W to zero. The cost coefficients of the w function dl
are used as the indicators during this stage. When w = 0, all artificial variables have
been removed from the basis and the model is in terms of the original control variables
only. At this point, the second phase of the method is begun. Since the goal is to
minimize the z function, the cost coefficients of CI are used as the indicators. During
the second phase, the artificial variables are necer considered as possible entries to the
basis. Once they are removed, they are never reentered into the basis again.

EXAMPLE 6.6 Two-Phase Minimization Problem

(a) Use the two-phase simplex method to solve for the following linear mathematical model:
Minirnize z = Xz

3x1 + 4X2 ~ 9
5x1 + 2xz :;;;8
3x1 - x2:;;; 0
Xl ~ 0, Xl ~ 0
(b) Trace the path to the optimum solution of a graph.
324 Optimization of linear Mathematical Models

Solution
(a) Slack and surplus variables are introduced into the model to remove the inequality con-
straints. The model becomes
Minimize z = X2

3x1+ 4Xl - X3 =9 (1)

5x1 + 2Xl + X4 =8 (2)

3x1 - Xl +X5 =0 (3)

We may establish the basic feasible solution by introducing an artificial variable to the first
constraint equation:
3x1 + 4Xl - X3 + X6 = 9
The variable X6 is the only variable necessary to establish an initial basic feasible solution.
The artificial objective function is simply equal to

The value of w must be written in terms of XN in order to qualify for the basis. From the con-
straint equation, we determine X6 to be equal to
X6 =9- 3Xl - 4X2 + X3
The artificial objective function in terms of non basic variables XN is
w = 9- 3Xl - 4X2 + X3
The mathematical model may be written as a two-phase linear mathematical model:
Minimize (w - 9) = -3x} - 4X2 + X3
Minimize (z - 0) =

=8
3Xl - X2 + Xs =0
The initial basis and non basis vectors are
X~' = [X4 Xs X6]

x~' = [Xl X2 X3J

The simplex tableau for this model is

Initial Solution
X8 Xl Xl X3 X4 X5 X6 h bJa
3 ® -1 0 0 1 9
1 ~ 2.25] +-
X4 5 2 0 1 0 0 8 ! =4
[X'
Xs 3 -1 0 0 1 0 0
C1 0 0 0 0 0 (z - 0)
dr -3 -4 0 0 0 (w - 9)
i
 The Two-Phase Simplex Method 325
The rules of the simplex method are applied to the w indicator row until X6, the artificial variable,
is removed from the basis XB•
RULE 1 Since not all values of the indicator row d, are positive, the initial solution is not a
minimum.
RULE 2 The control variable Xl enters the basis because dl = -4 is the smallest negative
value of the d, indicator vector. An arrow indicates that Xl enters the basis.
RULE 3 The artificial variable X6 leaves the basis because bla = 2.25 is the smallest positive
value of the bla vector. An arrow indicates that bla leaves the basis. Note that Xs is not considered
because the" a" coefficient is negative. a = -1.
Iteration 1 Successive elimination is used to determine the next candidate solution,
x1' = [x , X4 xs]. A circle has been drawn around the pivot element of the A matrix. The results of
successive elimination are as follows:
XB XI X2 X3 X4 Xs X6 h b/a
X2 0.75 1.00 -0.25 0.00 0.00 0.25 225 2.25/0.75 = 3
X4 3.50 0.00 0.50 1.00 O,()() -0.50 3.50 3.5/3.5 =1
Xs @ 0.00 -0.25 0.00 1.00 025 225 225/3.75 = 0.6 +-

CI -0.75 0.00 0.25 0.00 0.00 0.00 (z - 225)

d/ 0.00 0.00 0.00 0.00 0.00 1.0 (w - 0.0)

~

Since the artificial variable is no longer a member of the basis, the indicators of the z row are used
as indicators.
RULE 1 Since not all values of CI are positive, the candidate solution is not an optimum.
RULE 2 The control variable Xl will be introduced to the basis because Ct = -0.75 is the
only negative indicator of C/.
RULE 3 The control variable Xs is removed from the basis because bja = 0.6 is the smallest
positive value of the b a vector.
Iteration 2 Since \\. = O. the second indicator row of wand the column for X6, the artificial
variable is removed from further consideration. They are written in the following tableau.
Successive elimination is performed. The new candidate solution is

XB XI Xl X3 X4 Xs b
0.00 1.00 -0.20 0.00 -0.20 1.80
[X'X
4

Xl
0.00
1.00
0.00 0.73
0.00 -0.07
1.00 -0.93
0.00 0.27 1.~
0.60 1
CI 0.00 0.00 0.20 0.00 0.20 (z - 1.8)

RULE 1 Since all Clare positive, the optimum solution has been obtained.
The minimum is
xi = 1.80 X! = 1.40
The minimum value of z is
z* = 1.8
326 Optimization of linear Mathematical Models

3x,-x2 = 0
I-------~-----zi = 1.8
Second and final iteration, x"

Figure 6.9

(b) The path to the optimum minimum solution is shown in Figure 6.9. The initial solution is
(0, 0). After one iteration, we proceed to (0, 2.25). From the first-iteration simplex tableau, we
see that the artificial variable has been removed. The next iteration proceeds to the optimum
solution (0.6, 1.8).

EXAMPLE 6.7 A Two Phase Maximization Problem

(a) Use the two-phase simplex method to solve the following problem:

Maximize z = 2XI + X2

2XI + 3X2 ~ 6
2Xl + X2 ~ 6

x, ~ 0,X2 ~ 0
(b) Trace the path to the optimum solution on a graph.
 The Two-Phase Simplex Method 327
Solution

(a) Add slack and surplus variables to the linear model. The negative transpose of the objective
function, a = -lz, is required. The model becomes
Minimize a = -2xl - Xl

2xI + 3xl - X3 =6
=6
- Xs = 0

Xl ~ 0, X2 2:: 0, X4 2:: 0, X5 ~ 0
The basic feasible solution is established by introducing artificial variables X6 and X7 to the
first and third constraint equations:

-Xl + X2 - Xs + X7 = 0
The artificial objective function is

or, in terms of x,- variables

The two-phase linear mathematical model becomes

Minimize (\\. - 6) = - Xl - 4X2 + X3 + X5
Minimize (a - 0) = -2xI - Xl

2x, + 3Xl - X3 =6

2Xl + Xl + X4 =6
- XI + Xl - X5 + X7 = 0
Xb""X7 2:: 0
The initial basis and non basis are
x~' = [X6 X4 X7]

x~' = [Xl Xl X3 xs]

Initial Solution

The simplex tableau is

XB XI X:! X3 X4 Xs X6 X7 b bla

X4
[X'
X7
2
2
-1
3
1
Q)
-1
0
0
0
1
0
0
0
-1
1
0
0
0
0
1
6
6
0
~~21
t
¥
=6
=0 -
CI -2 0 0 0 0 0 . (a - 0)
d[ -1 -4 0 0 0 (w - 6)
 328 Optimization of linear Mathematical Models

The simplex method proceeds as follows:

RULE 1 Since not all values of d, are positive, the initial solution is not a minimum.
RULE 2 The control variable X2 enters the basis because d2 = -4 is the smallest negative
value of d/.
RULE 3 The artificial variable X7 is removed from the basis because bla = 0 is the smallest
value of bja.
Iteration 1 The next candidate solution is
XB Xl X2 X3 X4 X5 X6 X7 b bla
X6 @ 0.00 -1.00 0.00 3.00 1.00 -3.00 6.00 ~ = 1.2

X4 3.00 0.00 0.00 tOO 1.00 0.00 -1.00 6.00 !=2

X2 -1.00 1.00 0.00 0.00 -1.00 0.00 1.00 0.00
C[ -1.00 0.00 0.00 0.00 1.00 0.00 1.00 (a - 0.00)
dI -5.00 0.00 1.00 0.00 -3.00 0.00 4.00 (w - 6.CX))
j
RULE 1 Since dr contain negative values, the candidate solution is not an optimum.
RULE 2 The control variable Xl is entered since d, = -5 is the smaller negative value of dr.
RULE 3 The control variable X6 is removed because bla = 1.2 is the smaller positive value of
the bla vector.
Iteration 2 The next candidate solution is
xB Xl Xl x] x4 X5 Xc. x7 b b]a
Xl 1.00 0.00 -0.20 0.00 0.60 0.20 -0.60 1.20
X4 0.00 0.00 ® 1.00 -0.80 -0.60 0.80 2.40 2.4/0.60 =4 +-

Xl 0.00 1.00 -0.20 0.00 -0.40 0.20 0.40 1.20

<s 0.00 0.00 -ofo 0.00 0.80 0.60 -0.80 (a + 3.60)
dI 0.00 0.00 0.00 0.00 0.00 1.00 1.00 (w - 0)
j

RULE 1 Since all artificial variables have been removed from the basis. the C1 indicators are
used to determine if this candidate solution is an optimum. Since all values of C1 are not positive,
an optimum has not been found. 1.-
RULE 2 The control variable X3 is entered in the basis because C3 = -O.~ is the only negative
value of cs. The cost coefficients c.; and C7 are not considered, because they are cost coefficients
associated with the artificial variables X6 and X7•
RULE 3 The control variable X4 is removed from the basis because bla = 2.4 is the only
positive value of the bfa vector.
Iteration 3 The next candidate solution is
XB Xl X2 X3 X4 X5 b bla
1.00 0.00 0.00 033 0.33 200

r'
X]
X2

c1
0.00
0.00
0.00
0.00
1.00
0.00
1.00
0.00
0.00
1.67 -1.33
0.33 -0.67
0.00 1.00
4.00
2.00
(a + 6) 1
 The Two-Phase Simplex Method 329

RU LE 1 Since all values of CT are positive, an optimum solution is obtained.

Since a* = - 6. the maximum value of z is
z* = -1(-6) = 6

The optimum solution occurs at

xj =2 xi = 2 x! = 4
(b) The graph of the feasible region and the optimum solution is shown in Figure 6.10.
Remarks

When a member of the basis is equal to zero, the solution isdefined to bedegenerate.lfadegenerate
solution occurs. it is possible that an optimum solution will not be obtained. In this unlikely
situation. the solution of ; and \t' will not be improved upon further iterations. When this occurs a
control variable will be removed and reentered upon the subsequent iterations without reducing

x2

this boundary.

Third and final iteration, x "

Initial and
first iteration, xO and Xl
\

Figure 6.10
330 Optimization of linear Mathematical Models

w or z. This situation is called circling. Since X7 = 0 is in the initial basis and is being replaced by
X6 = 0, we have the possibility of circling, but it does not occur in this problem.
In the initial tableau, X7 was removed from the basis XB because bja.= Ojl is the smallest value.
Compare this tableau with tableau of iteration 1.In this case, X6 was removed from X8 instead of Xl-
The bja for X2 is equal to zero, bja = OJ-1 = 0; however, the "a" coefficient is a negative value,
thus violating rule 3. Consequently, Xl is not considered for entry into X8·
The slope of objective function z is equal to the slope of the constraint equation, lx, + X2 = 6.
The optimum solution x* consists of all points between points 1 and 2 as shown in the figure.
Since the simplex method only considers extreme points as candidate solutions, points I and 2 are
the only solutions that can be considered with the simplex algorithm. For the analysis conducted
in part a, point (0. 6) was never considered as a candidate solution, because once an optimum
solution is obtained. the procedure ends.

PROBLEMS
For Problem 1:
(a) Solve with the two-phase simplex method.
(b) Solve by graphical means.
(c) Trace the path to the optimum solution as followed by the simplex method of part a.

Problem I
Minimize r = 4x, + 5x1

Problem 2
'\1aximize z = 2.x 1 + 3x 2 - 2x 3

Xl + 3x 1 + X3 .2:: 3

6x 1 - X2 - X3 ~ 6

2x 1 - 3x 2 + Zx 3 ~ 12

Xl + 6x 1 - 4x.~~ 5

Problem 3