Optimization Models and Linear Programming

Linear Programming (L.P.) is a mathematical modeling technique that typically deals with the problem of allocating limited resources
among competing activities in the best possible (i.e. optimal) way.
The variety of situations to which this description applies is diverse: it ranges,
- from allocation of production facilities to products to allocation of national resources to domestic needs;
- from allocation of aero plane fuel to bomber runs to allocation of farmland to different crops;
- from investment portfolio selection to selection of shipping patterns.
However, one common ingredient in each of these situations is the necessity for allocating resources to activities.

In the term “Linear Programming”, the adjective linear means that all mathematical functions involved in linear programming models
are required to be linear functions.
The word “programming” does not refer to computer programming; rather, it is essentially a synonym for planning. In other words,
linear programming involves the planning of activities to determine a decision policy that will produce the best results (in terms of
the specified goal, as defined in the mathematical model) among all possible decision policy alternatives.
Because of its basic simplicity in modeling and fast, efficient solution procedures, linear programming is one of the most powerful
and widely used operations research techniques.

Problem Description

Identification of the Decision Variables

Development of the Matematical Model

Optimization Tools 2.1 Intro to L.P.

Example Problem 1
A farmer has 100 acres of land, some or all of which he can use to raise only potatoes and/or wheat. He has 160 man-days of labor
and 11,000 TL capital available. He may spend only up to these amount to raise crops on his land.
He knows that wheat brings in a revenue of 1200 TL/acre and potatoes bring in a revenue of 400 TL/acre. The input costs (for
seeding, cultivating etc.) are 200 TL/acre for wheat and 100 TL/acre for potatoes. The amount of labor required to raise the crops is, 4
man-days/acre for wheat and 1 man-day/acre for potatoes.

Potatoes Wheat Capacity

Acreage 1 1 100
Labor 1 4 160
Capital 100 200 11,000

The farmer wishes to know how many acres of wheat and/or potatoes he should plant to maximize his revenue (i.e. what is the
production (crop-raising) policy that will maximize revenue, while satisflying the stated requirements?

Model Formulation
Decision Variables:
x1 : Number of acres of potatoes planted
x2 : Number of acres of wheat planted.
Resource constraints (as a function of the decision variables):
1) 1.0 x1 + 1.0 x2 ≤ 100 (Acreage Constraint)
2) 1.0 x1 + 4.0 x2 ≤ 160 (Labor Constraint)
3) 100x1 + 200 x2 ≤ 11000 (Capital Constraint)

1st constraint: The total acreage planted should not exceed 100 acres.
2nd constraint: The total labor required to raise x1 acres of potatoes and x2 acres of wheat should not exceed 160 man-days.
3rd constraint: The total expenditure required to raise x1 acres of potatoes & x2 acres of wheat should not exceed 11,000 TL.

Optimization Tools 2.2 Intro to L.P.

Nonnegativity constraints (i.e. acreage planted cannot be less than zero):
4) x1 ≥ 0
5) x2 ≥ 0

Expected revenue as a function of potatoe and wheat planted and sale prices:
Z = 400 x1 + 1200 x2

Accordingly, the resulting linear programming model is as follows:

Objective : Max Z = 400 x1+ 1200 x2 [1]

Constraints : 1 x1 + 1 x2 ≤ 100 [2]
1 x1 + 4 x2 ≤ 160 [3]
100 x1 + 200 x2 ≤ 11000 [4]
x1 ≥ 0 [5]
x2 ≥ 0 [6]

Optimization Tools 2.3 Intro to L.P.

Example Problem 2
A heavy vehicle tire manufacturer needs to make a monthly production plan.
The plant manufactures two types of tires: the first type, T9, is used in trucks, while the second one, B9, is used in buses.
Both products go through similar processes in the plant.
The general economic outlook for the coming month indicates that the company can market and sell all they can produce during the
month.
The market conditions are not suitable for changes in product sale prices, while no changes are expected in the production costs in the
coming month.
Accordingly, specialists have forecasted a profit level of 100TL /unit for T9 model tires and 80TL/unit for B9 model tires.
Both products go through the rubber mixing (M) and pressing (P) departments, and then to the Quality Control/Testing (Q)
department.
Fixed unit processing times in these departments, and the monthly capacity of these departments, as determined by companly
engineers are as follows.

T9 Model B9 Model Departmant

unit processing unit processing Capacity
time(hr) time(hr) (hr)
Rubber Mixing 0.10 0.15 150
Pressing 0.20 0.10 160
QC and Testing 3.00 1.00 -

The most labor intensive department in the facility is the QC and Testing Department. As a precaution against layoffs, union
agreements require that the montly workload in this department should not be below 1350 hours.
The Marketing Department has recommended the production of at least one B9 model tire for every 3 T9 models produced, to shore
up the market position.
The Costing Department, suggests the production of at least 600 units, so that idle time will be kept under control and fixed costs will
be evenly distributed.
Optimization Tools 2.4 Intro to L.P.
What is the monthly production policy that will maximize monthly profits, while satisfying all the conditions stated above?

Model Formulation
Decision Variables:
x1 : Number of T9 type tires to be produced
x2 : Number of B9 type tires to be produced.

Resource constraints (as a function of the decision variables):

1) 0.10 x1 + 0.15 x2 ≤ 150 (Mixing Department Capacity Constraint)
2) 0.20 x1 + 0.10 x2 ≤ 160 (Pressing Department Capacity Constraint)

1st constraint: Total mixing dept. processing time to produce x1 units of T9 & x2 units of B9 should not exceed 150 hours
2nd constraint: Total pressing dept. processing time to produce x1 units of T9 & x2 units of B9 should not exeed 160 hours.

Union agreement (as a function of the decision variables):

3) 3.0 x1 + 1.0 x2 ≥ 1350 (Labor Force Constraint)
Left hand side is the QC and Testing labor required to produce x1 units of T9 & x2 units of B9; the right hand side is the minimum
monthly workload.
Market position and production level constraints:
4) x1 ≤ 3x2 (Product Composition Constraint)
5) x1 + x2 ≥ 600 (Minimum Production Level)

Nonnegativity constraints (i.e. production levels being nonnegative):

6) x1 ≥ 0
7) x2 ≥ 0
Optimization Tools 2.5 Intro to L.P.
Monthly profit as a function of T9 and B9 production levels and unit profits:
Z = 100 x1 + 80 x2

The resulting L.P. model:

Objective : Max Z = 100 x1 + 80 x2 [7]

Constraints : 0.10 x1 + 0.15 x2 ≤ 150 [8]

0.20 x1 + 0.10 x2 ≤ 160 [9]
3.00 x1 + 1.00 x2 ≥ 1350 [10]
1.00 x1 - 3.00 x2 ≤ 0 [11]
1.00 x1 + 1.00 x2 ≥ 600 [12]
x1 ≥ 0 [13]
x2 ≥ 0 [14]

Optimization Tools 2.6 Intro to L.P.

Linear Programming General Formulation and Terminology

A general L.P. model, having "n" decision variables, "m" constraints, and a maximization objective:

Objective : (Max) Z = c1 x1 + .................... + cnxn

Constraints: a11x1 + .................... + a1nxn ≤ b1
a21x1 + .................... + a2nxn ≤ b2
.
.
am1x1 + .................... + amnxn ≤ bm
x1,..........., xn ≥ 0

The above general formulation is usually denoted as,

Objective:
n
max z = ∑ cj xj
j=1
Constraints:
n
∑ ai jxj ≤ bi i = 1, ..., m
j=1
xj ≥ 0 j = 1, ..., n

In the above general formulation,

(c1,..., cn) are called the profit (or cost) coefficients,
(b1,..., bm) are called the resource levels,
(aıı,..., amn) are called the technology coefficients,
n
z(x1 ,..., xn ) = ∑ cj xj
j=1 is called the objective function
Optimization Tools 2.7 Intro to L.P.
 n
∑ ai jxj ≤ bi ∀i
j=1
xj ≥ 0 ∀j (i.e. set of points (x1,..., xn) satisfying the stated conditions) is called the feasible region.

Any point in the feasible region (i.e. any point (x1,..., xn) satisfying the stated conditions) is called a feasible solution of the L.P.
problem. Any feasible solution, x* = (x*1,..., x*n), which has,
z (x*1,..., x*n) ≥ z (x1,..., xn) for all feasible solutions (x1,..., xn)
is called an optimal solution of the L.P. problem.

The value, z (x*1,..., x*n), is called the optimal value of the L.P. problem. Note that an L.P. problem may have many (infinitely
many) optimal solutions but has a unique optimal value.

An L.P. problem can also be stated in in vector/matrix notation as,

max z = c xT
s.t. A xT ≤ b T
x≥0
where, c = [c1,..., cn] is called the profit (or cost) vector,
b = [b1,..., bm] is called the resource vector,
x = [x1,..., xn] is called the decision vector.
a1 1 a1 n
A=
am1 amn is called the technology matrix.

For example, in the farming problem,

1 1
A= 1 4
100 200 c = [400, 1200]; b = [100, 160, 11000]
Optimization Tools 2.8 Intro to L.P.
In order to solve the model and suggest an effective solution to the related decision problem, the values of these parameters have to
be available. The main work to be done in an L.P. study is to find "optimal values" for the decision variables (x1,..., xn) under the
stated (or assumed) parameter values.

Main Assumptions of a Linear Programming Model

a) Proportionality: In all functions involved, costs, profits, resource usages are all proportional to levels of the decision variables
(e.g., if one unit of product brings in 100 TL profit, then 5 units will bring in 500 TL profit).
b) Additivity: In all functions involved, effects of the decision variables are seperate from one another and are additive. For
example,
Z(x1, x2, x3) = f1(x1) + f2(x2) + f3(x3)
where, "f1(x1)= 500 x1"; "f2(x2)=400 x2"
c) Continuity: Decision variables may take any real value as long as all the stated constraint functions are satisfied.
d) Deterministic Nature: It is assumed that all parameter values are deterministic, static and known with certainty at the solution
stage.

The following type of questions may help in determining how closely the undertaken assumptions fit the real being invistigated. In
the example problems,
a) How realistic it is to assume constant unit costs and sale prices?
b) Is it at all possibe to know such costs and sale prices at planning time?
c) Is it realistic to have resource usage proportional to production levels?
d) Is a one month planning horizon sufficient and/or realistic? How are daily schedules compatable with the monthly plans to be
determined?
e) Is the dynamic nature and randomness of the demand an important issue? Is it possible to stock the goods produced for usage in
later periods?

It is possible to provide satisfactory answers to some of these by expanding the L.P. model. But, some cannot be satisfied at all
through the L.P. approach.

Optimization Tools 2.9 Intro to L.P.

The Graphical Solution Procedure
In this graphical procedure, each dimension (axis) of the graph represents a different variable, and accordingly the constraint
functions are drawn.
Regarding inequality constraints, depending on the constraint sign, the area in the graph above or below the line representing the
constraint equation, corresponds to the set of points satisfying that constraint.
Regarding equality constraints, the set of points exactly on the line representing the constraint equation, corresponds to the set of
points satisfying that constraint.
The set of points on the graph satisfying all constraints represents the feasible region of the problem.

100

X 1+X 2 = 100
80

60
100 x 1 + 200 x2 = 11000

40

20

x1 + 4 x2 = 160

20 40 60 80 100 120 140 160

Optimization Tools 2.10 Intro to L.P.

The objective function family is superimposed on this graph, and the maximum value it can take (the individual line in this family,
corresponding the the maximum possible objective function value while touching the feasible region) is visually determined.

100

X 1+X 2 = 100
80

400 x1 + 1200 x2 = k

60 x1 = 60
100 x1 + 200 x2 = 11000 x2 = 25
z = 54000
40

20

x1 + 4 x2 = 160

20 40 60 80 100 120 140 160

k = 24000 k = 36000 k = 48000

Optimization Tools 2.11 Intro to L.P.

The following figure displays the implementation of this procedure to the Tire Manufacture problem. The optimal solution and the
optimal value are
x1 = 450; x2 = 700; z = 101,000
In other words, the monthly production of 450 units of T9 type tires and 700 units of B9 type tires is being suggested. The expected
profit from this production policy is 101,000 TL.

0.2 X 1 + 0.1 X 2 = 160

1400

1200

100 X 1 + 80 X 2 = Pr ofit
1000

800

3X 1+X 2 =1350
600

400 X1 = 3 X2

X 1+ X 2 = 600
200 0.1 X 1 + 0.15 X 2 = 150

X1

200 400 600 800 1000 1200 1400

Optimization Tools 2.12 Intro to L.P.

Erroneous Formulations
In large and complex problems having many variables and constraints, it is easy to make formulation and/or data errors and
meaningless results may follow.
Two frequently faced such situations are as follows. In both cases, although the mathematical conclusions derived are logical, they
have almost no practical meaning, other than pointing to some formulation and/or data errors.
Unbounded Problems
This is the case of being able to increase the objective value (equivaletly, moving to ever higher lines in the objective function
family) to infinity, without bound. There is no optimal solution and optimal value is infinity.
In real life problems such a situation is almost never possible; but if one or more constraints are omitted or erroneously inputed
during the model development phase, then unboundedness may result.
For example, if constraints [8], [9] are omitted in the tire manufacture problem, then an unbounded optimal value will result.
X2

1400

1200

1000 100 X 1+ 80 X 2

800

600

400
X1 = 3 X2
X 1 + X 2 = 600

200

X1

200 400 600 800 1000 1200

Optimization Tools 2.13 Intro to L.P.

Infeasible (Inconsistent) Problem
This is the case of being able to find no feasible solution satisfying all the stated constraints.
There is no optimal solution and optimal value is undefined. In real life problems such a situation is again almost never possible
(especially if the model is of a working system, presumably having feasible solutions).
But if one or more constraints are erroneously inputed or non existing constraints are defined during the model development phase,
then inconsistancy may result.
For example, if constraint [11] is erroneously inputted as
7.00 x1 - 1.00 x2≤ 0 [24]
in the tire manufacture problem, then an inconsistent LP model will result.
X2

1400

1200 7 X1 = X2

1000

800

600

400

X1+X2 = 600
200

X1

200 400 600 800 1000 1200

Optimization Tools 2.14 Intro to L.P.
Insights Gained Through The Graphical Solution Procedure
Unfortunately, the graphical approach cannot be extended to handle problems having more that 3 variables:
The maximum number of dimensions possible in a visual graph is 3. Since the representation of each variable by a different axis
(dimension) is the defining characteristic of the graphical approach, it cannot be applied to problems having more than 3 variables.
Nevertheless, the graphical approach provides some critical insights about the feasible region of an LP problem and the optimal
solution. These insights give an intuitive explalation about the functioning and effectiveness of the algebraic solution procedure of LP
Problems, namely the Simplex Method.
i) The feasible regions displayed in the examples, have a particular geometry: Its boundaries are made up by straight lines (the
constraint equations) and it has a finite number of “corner points”.
In mathematical terms, this is called a “Convex Polytope”, and it turns out that the feasible region of an L.P. problem is always
a Convex Polytope, regardless of problem size and problem parameters.
ii) In the example problems, the optimal solutions obtained are at the corner points of the fesible region.
In mathematical terms, these are called “Extreme Points”, and it turns out that, if an L.P. problem has an optimal solution, it
has at least one Extreme Point Optimal Solution, regardless of problem size and parameters.
iii) Regarding changes in cost coefficients: It is the “slope” of the objective function family determining which corner point is
optimal, accordingly:
Changes in these coefficients that do not change the “family slope” (i.e. cost coefficients increasing of decreasing in proportion
to one another), will not effect the optimal solution(s) (optimal value will be changing, since the “k” value corresponding to
each line in the family will change).
“Small” changes in the “family slope” will neither effect the optimal corner(s), which implies that the optimal solution(s) will
not change, when one or more objective function coefficients are “slightly” changed.
“Small change” is not well defined; but, as slope is changed, at some point it matches the slope of a border line (and both
corner points of that border line become optimal), beyond which previous optimal corner point ceases to be optimal and the
adjacent corner point (i.e. the corner point at the other end of the border line) becomes optimal.
Such behavior of optimal value and solutions, is common to all LP problems. The concept of “adjacency” of extreme points
can also be defined mathematically, and is important in algebraic solution procedures (Simplex moves from one extreme point
to an adjacent one in any iteration).

Optimization Tools 2.15 Intro to L.P.

 iv) If the slope of the objective function family is equal to the slope of a boundary line (i.e. a constraint function), the two corner
points at end of that boundary line become optimal solutions.
Actually, all feasible points on that boundary line become optimal solutions.
This situation is called having multiple optimal solutions (at least two extreme point optimal solutions and infinitely many non-
extreme point optimal solutions), and may happen in LP problems of any size.

100

80

60

40

20

20 40 60 80 100 120 140 160

Optimization Tools 2.16 Intro to L.P.

 100

80
600 x1 + 1200 x2 = k3
x1 = 90
450 x1 + 1220 x2 = k2 x2 = 10
60 x1 = 60
x2 = 25 z1 = 48000
z2 = 52700
z1 = 54000
z3 = 66000
40 z2 = 57500
z3 = 66000

400 x1 + 1200 x2 = k1
20

20 40 60 80 100 120 140 160

Optimization Tools 2.17 Intro to L.P.

 The Algebraic Solution Procedure (The Simplex Method)

L.P. Formulation Form Changes

- The objective of an L.P. problem could be maximization or minimization,
- The constraints could be any combination of "≥", “≤” or "=" constraints,
- Some or all variables could be restricted (to be nonnegative/non positive) or free (to take any real value).
Such variations are not important, since one form can easily be converted to another through simple algebraic
manipulations

Converting a “minimization” objective to a “maximization” objective:

The feasible and the optimal solution sets of the following problems are identical

n n
min z = ∑ c j x j max z = ∑ -c j x j
j= 1 j=1

P1 s.t. P2 s.t.
n n
∑ a ij x j ≤ b i i = 1,..., m ∑ a ij x j ≤ b i i = 1,..., m
j=1 j=1

xj ≥ 0 j = 1,…, n xj ≥ 0 j = 1 ,…, n

Optimization Tools 2.18 Intro to L.P.

A vector x* is an optimal (feasible) solution of P1, if and only if it is an optimal (feasible) solution of P2.

The optimal values differ only by a multiplication by -1. That is, if z1* and z2* are optimal values of P1 and P2
respectively, then necessarily z1* = - z2*.

To convert a minimization problem into an equivalent maximization problem (or vica versa), all objective function
coefficients should be multiplied by (-1).

The optimal solutions will be the same, the optimal values will be differing by a multiplication by -1.

Converting a “≥” constraint into a “≤” constaint:

The following relationships are fully equivalent.
⎧⎪ n ⎫⎪ ⎧⎪ n ⎫⎪
⎨∑ a ij x j ≥ b i ⎬ ⎨∑ -a ij x j ≤ −b i ⎬
⎪⎩ j=1 ⎪⎭ ⎪⎩ j=1 ⎪⎭

So, the first can be replaced by the following without losing any generality.
⎪⎧ n ⎪⎫
⎨∑ a ij x j ≤ bi ⎬ where a ij = -a ij ; b i = -b i
⎪⎩ j=1 ⎪⎭

Optimization Tools 2.19 Intro to L.P.

Converting a “≤” constraint into an “equality” (“=”) constaint:
The following relationships are fully equivalent.
⎧⎪ n ⎫⎪ ⎧⎪ n ⎫⎪
⎨∑ a ij x j ≤ b i ⎬ ⎨∑ a ij x j + si = b i ⎬
⎪⎩ j=1 ⎪⎭ ⎪⎩ j= 1 ⎪⎭
While the left hand side is required to be less or equal to the right hand side in the original relationship, a new
variable (si) is defined in the new relationship to represent the difference between the two sides. In other words,
n
si = b i − ∑ a ij x j ≥ 0
j= 1

“bi” is usually the availability level of a certain resource.

“∑ aij xj” is the amount of that resource being used by the decision policy x.
The difference is to the idle (unused) amount in that resource.
Such variables (si) are called “slack variables” and are assigned zero coefficients in the objective (if there is no
cost/profit of having idle resources).

Optimization Tools 2.20 Intro to L.P.

Converting a “≥” constraint into an “equality” (“=”) constaint:
The following relationships are fully equivalent.
⎧⎪ n ⎫⎪ ⎧⎪ n ⎫⎪
⎨∑ a ij x j ≥ b i ⎬ ⎨∑ a ij x j − si = b i ⎬
⎪⎩ j=1 ⎪⎭ ⎪⎩ j= 1 ⎪⎭
While the left hand side is required to be greater or or equal to the right hand side in the original relationship, a new
variable (si) is defined in the new relationship to represent the difference between the two sides. In other words,
n
si = ∑ a ij x j − b i ≥ 0
j= 1

“bi” is the minimum required level of a certain activity.

“∑ aij xj” is the amount of that activity being deployed by the decision policy x.
The difference corresponds to the surplus amount in that activity.
Such variables (si) are called “surplus variables” and are assigned zero coefficients in objective (if there is no
cost/profit of having surplus activities).

Optimization Tools 2.21 Intro to L.P.

Converting an “=” constraint into “≤” constaints:
The following relationships are fully equivalent.

⎧⎪ n ⎫⎪ ⎧⎪ n n ⎫⎪
⎨∑ a ij x j = b i ⎬ ⎨∑ a ij x j ≤ b i ; ∑ a ij x j ≥ b i ⎬⎪
⎪⎩ j=1 ⎪⎭ ⎪⎩ j=1 j=1 ⎭

which is further equivalent to

⎧n ⎫
⎪∑ a ij x j ≤ b i ⎪
⎪ j=1 ⎪
⎨n ⎬
⎪ ⎪
⎪∑ -a ij x j ≤ −b i ⎪
⎩ j=1 ⎭

Optimization Tools 2.22 Intro to L.P.

Transforming Free Variables into Restricted Variables
In the following L.P. problem one variable is free to take negative or positive values, while the others are restricted
to be non-negative.

Max Z= c1 x1 + c2 x2 + c3 x3 + c4 x4 + c5x5
s.t.
a11x1 + a12x2 + a13x3 + a14x4 + a15x5 ≤ b1
a21x1 + a22x2 + a23x3 + a24x4 + a25x5 ≤ b2
a31x1 + a32x2 + a33x3 + a34x4 + a35x5 ≤ b3
x1, x2, x3, x4 ≥ 0; x5 free

Set up two new, non-negative variables and define x5 to be their difference:

x6, x7 ≥ 0; x5 = x6 - x7

Replace x5 with “x6 - x7” to obtain,

Max Z= c1 x1 + c2 x2 + c3 x3 + c4 x4 + c5(x6 - x7)

s.t.
a11x1 + a12x2 + a13x3 + a14x4 + a15(x6 - x7) ≤ b1
a21x1 + a22x2 + a23x3 + a24x4 + a25(x6 - x7) ≤ b2
a31x1 + a32x2 + a33x3 + a34x4 + a35(x6 - x7) ≤ b3
x1, x2, x3, x4, x6, x7 ≥ 0
This formulation does not impose any new restrictions on the free variable x5 (the difference of two non-negative
variables could be negative or positive);
But since it is removed from the formulation, the new formulation does not contain any free variables.
Optimization Tools 2.23 Intro to L.P.
The Standard Form
An L.P. problem, where,
- the objective is maximization,
- all functional constraints are represented as equality constraints,
- all variables are restricted to be nonnegative,
- the constraint vector is nonnegative,
is said to be in “Standard Form”.

Objective:
n
max z = ∑ c j x j
j=1

Constraints:
n
∑ a ij x j = b i i = 1,..., m
j=1

xj ≥ 0 j = 1,…, n

Any L.P. problem must first be brought into this standard form for the implementation of the Simplex solution
procedure.

Fortunately, through the discussed form changes, convertion into standard form is quite simple.

The standard form of the problem previously discussed will be as follows:

Optimization Tools 2.24 Intro to L.P.

 Objective :

Max Z = 550x1+ 600x2+ 350x3+ 400x4 + 200 x5 [25]

s.t.
12 x1 + 20 x2 + 25 x4 + 15 x5 + s1 = 17280 [26]
10 x1 + 8 x2 + 16 x3 + s2 = 11520 [27]
20 x1 + 20 x2 + 20 x3 + 20 x4 + 20 x5 + s3 = 23040 [28]

x1 ≥ 0 [29]
x2 ≥ 0 [30]
x3 ≥ 0 [31]
x4 ≥ 0 [32]
x5 ≥ 0 [33]
s1 ≥ 0 [34]
s2 ≥ 0 [35]
s3 ≥ 0 [36]

Optimization Tools 2.25 Intro to L.P.

The Simplex Tableau
The Simplex Method is a “tabular representation mechanism” and a set of “elementary row operations” (ERO)
applied to the constraint equations and objective function of any given L.P. problem in standard form.

The “Tabular Format” known as the Simplex Tableau is as follows:

- there is a row for the objective function (row [0]),
- there is a row for every constraint equation (rows [1] to [m]),
- there is a column for every variable (columns [1] to [n]),
- there is a column for the constraint vector (column [n+1]),
- there is a column associated with the parameters of “z” in the objective function and and in the constraints
(column 0). (The objective function value “z” is also treated like a variable).

ERO are following algebraic operations applied to systems of linear equations:

- Multiplication of a linear equation by a scalar,
- Addition of any scalar multiple of an equation to another equation.
Important property: ERO do not alter the feasible solution set of the given system of linear equations.

“Initial” simplex tableau contains the parameters of the L.P. problem. During the solution procedure, the Simplex
ERO are applied to rows of the simplex tableau.

The typical initial simplex tableau of an “n” variable, “m” constraint L.P. problem in standard form is displayed
below.

Optimization Tools 2.26 Intro to L.P.

 z x1 x2 x3 xn-1 xn b
Row [0] 1 - c1 - c2 - c3 .......... - cn-1 - cn 0
Row [1] 0 a11 a12 a13 .......... a1,n-1 a1,n b1
Row [2] 0 a21 a22 a23 .......... a2,n-1 a2,n b2
. . . . . . .
. . . . . . .
Row [m] 0 am1 am2 am3 .......... am,n-1 am,n bm
. . . . . . .

An intuitive interpretation of the information in an initial simplex tableau:

- Equating inner product of the first (n+1) columns of row [0] and vector (z, x1, x2, ..., xn-1, xn) to the last entry of
row [0] generates the objective fuction
z - c1 x1 - c2 x2 - ....... - cn-1 xn-1 - cn xn = 0
- Equating the inner product of the first (n+1) columns of row [i] and the vector (z, x1, x2, ..., xn-1, xn) to the last
entry of row [i] generates the ith constraint equation as,
ai1 x1 + ai2 x2 + ....... + ai, n-1 xn-1 + ain xn = bi
The initial simplex tableau of the previously described production problem:
z x1 x2 x3 x4 x5 s1 s2 s3
1 - 550 - 600 - 350 - 400 - 200 0 0 0 0
0 12 20 0 25 15 1 0 0 17280
0 10 8 16 0 0 0 1 0 11520
0 20 20 20 20 20 0 0 1 23040
Optimization Tools 2.27 Intro to L.P.
A key characteristic of optimal solutions (when the problem is stated in “standard form”):
The number of variables having positive values will be at most m (i.e. number of equality constraints).
This is mathematicall equivalent to “the associated optimal solution being a corner point/extreme point of the
feasible region”.

Optimization Tools 2.28 Intro to L.P.