3315

OPERATIONS
RESEARCH I

Dr. Bill Corley

STANDARD FORM FOR A LPP

Maximize z = c1x1 + c2x2 + …+ cnxn
( x 1 , ..., x n )
s u b je c t to

a11x1 + a12x2 +….+ a1nxn ≤ , = , ≥ b1

------------------------------------------------------------------
am1x1+ am2x2 +….+ amnxn ≤ , = , ≥ bm

x1,…, xn ≥ 0

1
 DIET EXAMPLE
Consider an impecunious student who
decides to eat meals consisting of
only milk, tuna fish, bread, and
spinach. Formulate a linear program
to minimize cost while meeting the
required daily allowance (RDA) of
various nutrients.

DATA
Gallons of Pounds of Loaves of Pounds of RDA
milk Tuna Bread Spinach
x1 x2 x3 x4
Vit. A 6400 237 0 34000 5000 IU

Vit. C 40 0 0 71 75 mg

Vit. D 540 0 0 0 400 IU

Iron 28 7 13 8 12 mg

Cost $3.00 $2.70 $1.80 $2.16

IU = International Unit; mg= milligram

2
Minimize z = 3.00x1 + 2.70x2 +1.80x3 + 2.16x4

s.t.

6400x1 +237x2 + 0x3 + 34000x4 > 5000 vit. A

40x1 + 0x2 + 0x3 + 71x4 > 75 vit. C
540x1 + 0x2 + 0x3 + 0x4 > 400 vit. D
28x1 + 7x2 + 13x3+ 8x4 > 12 iron

x1, x2, x3, x4 > 0

EXAMPLE
Greenthumb.com, a fertilizer company, wants
to make two types of fertilizers: high-nitrogen
and all-purpose fertilizers. There are two types
of components needed to make these
fertilizers. Component 1 consists of 60%
nitrogen and 10% phosphorous and costs 20
cents per pound. Component 2 consists of 10%
nitrogen and 40% phosphorous and costs 30
cents per pound. The company wants to
produce 5000 25-pound bags of high-nitrogen
fertilizer and 7000 25-pound bags of all-
purpose fertilizer.

3
Let x1 denote the amount of component 1 required to
make high-nitrogen fertilizer and x2 denote the amount
of component 2 required to make high-nitrogen fertilizer.
Similarly let y1 and y2 denote the amounts of
component 1 and component 2, respectively, required to
make all-purpose fertilizer. High-nitrogen fertilizer must
contain 40 – 50 % nitrogen by weight and all-purpose
fertilizer must contain at most 20 % phosphorous by weight.
Greenthumb.com wants to minimize the total cost of
producing these fertilizers.

Minimize z = 0.2( x1 +y1) + 0.3(x2 + y2)

s.t.
x1 + x2 > 125,000
y1 + y2 > 175,000
-0.20x1 + 0.30x2 < 0
0.10x1 – 0.40x2 < 0
-0.10y1 + 0.20y2 < 0
x1, x2, y1, y2 > 0.

4
 EXAMPLE
ChemLabs uses raw materials I and II to produce
two domestic cleaning solutions A and B. The daily
availabilities of raw materials I and II are 160 and
145 units, respectively. One unit of solution A
consumes 0.5 unit of raw material I and 0.6 unit of
raw material II, and one unit of solution B uses 0.4
unit of raw material I and 0.3 unit of raw material II.
The profits per unit of solutions A and B are $8 and
$10, respectively. The daily demands for solutions A
and B are exactly 175 and 200 units, respectively, all
of which may not be able to be fulfilled. Find the
optimal daily production amounts of A and B.

SOLUTION
Let x be the units of solution A to be made, and let
y be the units of solution B to be made.

Maximize z = 8x + 10y
s.t.
0.5x + 0.4y ≤ 160 I
0.6x + 0.3y ≤ 145 II
x ≤ 175 A
y ≤ 200 B
x,y ≥ 0

5
 HOMEWORK 1
An industrial recycling center uses two scrap
aluminum metals, A and B, to produce a special
alloy. Scrap A contains 6% aluminum, 3% silicon,
and 4% carbon. Scrap B has 3% aluminum, 6%
silicon and 3% carbon. The costs per ton for
scraps A and B are $100 and $80, respectively.
The specifications of the special alloy are as
follows:

1. The aluminum content must be at least 3% and

at most 6%
2. The silicon content must lie between 3% and
5%
3. The carbon content must be between 3% and
7%
Formulate a linear program that can be used to
determine the amounts of scrap A and B that
should be used to minimize the cost of creating
1000 tons of the special alloy.

6
GRAPHICAL SOLUTION OF
LINEAR PROGRAMMING
Maximize z = 3x1+5x2
subject to
3x1 + 2x2 < 18
x1 <4
x2 < 6
x 1, x 2 > 0

GRAPHICAL SOLUTION
x2 x2 = 6

(2,6)
6
x1 = 4

4 3x1 + 2x2 = 18

(0,0) 2 4 6 x1 3x1 + 5x2 =36

7
 DEFINITIONS

FEASIBLE SOLUTION: an (x1,…xn) satisfying

all the constraints

OPTIMAL SOLUTION: a best feasible

solution

EXTREME POINT: a corner of the convex set

of feasible solutions

PROPERTIES

PROPERTY 1:
The set of feasible solutions to a linear
programming problem is a convex set.
PROPERTY 2:
If there exists an optimal solution, then at
least one extreme point is optimal.
PROPERTY 3:
There are only a finite number of extreme
points.

8
Before giving properties 4 and 5, rewrite the
problem by adding slack variables. A slack
variable is added to make an inequality into
an equation. By adding slack variables, the
problem becomes

Maximize z = 3x1+5x2
subject to

3x1 + 2x2 + x3 = 18
x1 + x4 =4
x2 + x5 = 6
x1, x2, x3, x4, x5 > 0.

9
GRAPHICAL INTERPRETATION
x1

x1+x2 < 1
x1 , x2  > 0
(0,1)

(0,0) (1,0) x2

x1+x2 + x3 = 1
x1, x2, x3 > 0
x3
not an extreme 
point, not even 
feasible. feasible region 
(0,0,1)
in this 2‐dimensional surface 
embedded in 3 dimensions
(0,1,0) x2
(1,0,0)

x1

10
 BASIC SOLUTIONS
For m equations and n unknown variables (m<n),
solve for m variables in terms of (n-m) set to 0.
Consider
x1+x2 + x3 = 2
2x1 ‐x2 + x3 = 1.

Pick x3, set it to zero, and solve for the others.

x1+x2 = 2
2x1 – x2 = 1

x1 = 1,  x2 = 1 are basic variables. 
x3 = 0 is a non‐basic variable. 

EXAMPLE
x1+x2 + x3 = 2 
2 x1 ‐x2 + x3 = 1  

Pick x2, set it to zero, and solve for the others.
x1+ x3 = 2
2x1+ x3 = 1        
x1 = – 1, x3 = 3 are basic variables. 
x2 = 0  is a non‐basic variable.

11
 BASIC FEASIBLE
SOLUTION
A basic solution to the constraints (omitting non‐
negativity) of LPP with slacks and surplus variables 
with all basic variables > 0.

In the previous examples, Example 1 yielded a BFS 
while Example 2 did not.

DEGENERATE BFS
A  basic feasible solution where at least one basic 
variable is 0.
Example 1:
x1+x2 + x3 = 1
2x1+x2 + x3 = 2
Pick x3 = 0 as nonbasic and x1, x2  as basic. Solve
x1+x2 = 1
2x1+x2 = 2       
to obtain x1 = 1,  x2 = 0  for the basic variables.

12
 Example 2:
x1+x2 + x3 = 1
x1+x2 + 3x3 = 2

Arbitrarily chose x3 as the nonbasic variable to give 
x1+x2 = 1
x1+x2 = 2.

Note the contradiction. Thus there is no basic solution 
with x3 = 0.

Property 4:

With slacks as needed to make all constraints 
equality, there is a one‐to‐one correspondence 
between the BFS of these equality constraints and 
the extreme points of the original constraints.

13
Property 5:
Changing one BV in a BFS moves to an 
adjacent extreme point.
x2

one 
iteration

x1

EXAMPLE
Maximize z = 3x1 + 5x2 
s.t.
3x1 + 2x2   < 18
x1                 < 4
x2    < 6
x1, x2 > 0

14
 Rewrite to get

Maximize z = 3x1+5x2
subject to
3x1 + 2x2 + x3 = 18,
x1 + x4 = 4,
x2 + x5 = 6,
x1,x2,x3,x4,x5 >0

Rewrite to give
z – 3x1 – 5x2                                     = 0
3x1 + 2x2 + x3                     = 18
x1                           + x4           = 4
x2                        + x5 = 6.

Note: m=3 , n=5  => 10 possible BFS’s
A BFS is found using the last 3 equations 
while keeping track of  z using the first equation.

15
Initial BFS:
x3 = 18, x4 = 4, x5 = 6     (BV) 
x1 =   0, x2 = 0              (NBV)
and     z =   0

Note that making x2 positive increases z the 
most per unit.
Hence, make x2 a BV and keep x1=0 a NBV.
Next, rewrite the last 3 equations to give
x3 = 18 – 3x1 – 2x2
x4 =   4 – x1
x5 =   6          – x2.

Now
x3 = 18 – 2x2  > 0    =>  x2 < 9
x4 =   4
x5 =   6 – x2         > 0     =>  x2 < 6

So  x2  enters and  x5  leaves.

The new BFS is adjacent to the earlier one.

16
 z – 3x1 – 5x2                                  =   0
3x1 + 2x2 + x3                     = 18
x1                           + x4           =  4
x2                        + x5 =  6
Multiply 4th equation by 5 and add to z row and
multiply 4th equation by ‐2 and add to 2nd equation.
z – 3x1                                  + 5 x5 = 30           
3x1              + x3      – 2x5 =  6
x1                           + x4           =  4
x2                        + x5 =  6

It follows that 
z                + x3      +3x5= 36
x1      +1/3x3              ‐2/3x5= 2
– 1/3x3   +x4  +2/3x5= 2 

x2                               + x5= 6.

All of the negative variables have been
removed from the objective function meaning that
it has been optimized. Thus z = 36, x1 = 2, x2 = 6.

17
 TABLEAU APPROACH

BV z x1 x2 x3 x4 x5 RHS

z 1 -3 -5 0 0 0 0
x3 0 3 2 1 0 0 18
x4 0 1 0 0 1 0 4
x5 0 0 1 0 0 1 6

TABLEAU ONE

GAUSS – JORDAN
ELIMINATION
The pivot element is the intersection of the leaving and 
entering variables in the tableau.

a                b

c d

Let a = pivot element.
Then the new element dnew = (ad – bc) / a .

18
 SIMPLEX ALGORITHM

1. Must have all RHS (not z ) > 0.

2. Entering variable is most negative element in top
row.
3. Leaving variable is found by finding the smallest
ratio of the RHS to positive elements of pivot
column. (If no positive elements, the problem is
unbounded.)
4. Form new tableau using Gauss-Jordan elimination.
5. If elements in top row > 0, stop. Otherwise, go to 2.

BV z x1 x2 x3 x4 x5 RHS

z 1 -3 0 0 0 5 30
x3 0 3 0 1 0 -2 6
x4 0 1 0 0 1 0 4
x2 0 0 1 0 0 1 6

TABLEAU TWO

19
BV z x1 x2 x3 x4 x5 RHS

z 1 0 0 1 0 3 36
x1 0 1 0 1/3 0 -2/3 2
x4 0 0 0 -1/3 1 2/3 2
x2 0 0 1 0 0 1 6

FINAL TABLEAU

SOLUTION

z =36
x1=2
x2=6

20
 EXAMPLE

Maximize z = 2x1+3x2
subject to
x1 + x2 < 3
x1 - x2 < 1
x1,x2 > 0

Adding slack variables, we get

z – 2x1 – 3x2 =0
x1    +x2 +x3          = 3
x1    ‐x2          +x4 = 1
x1, x2, x3, x4 > 0.

21
 TABLEAU

BV z x1 x2 x3 x4 RHS

z 1 -2 -3 0 0 0
x3 0 1 1 1 0 3
x4 0 1 -1 0 1 1
z 1 1 0 3 0 9
x2 0 1 1 1 0 3
x4 0 2 0 1 1 4

SOLUTION

z =9
x1=0
x2=3

22
 HOMEWORK 2
Maximize z = 2x1+5x2
subject to
x1 + x2 < 12
3x1+ x2 < 18
x1,x2 > 0
Solve using the simplex method by the tableau
approach.

HOMEWORK 3-8
Six linear programming problems solved by Lingo are due
February 20 at the beginning of class. Some form of computer
printout of the solution is required. In all cases, simply apply
Tora to the mathematical problem and ignore the directions. In
the tenth edition of Taha, these problems are:

page 122 - #3-32

page 126 - # 3-50
page 126 - # 3-52
page 129 - #3-59
page 129 - #3-62
page 363 - #9-60.

23
 COMPLICATIONS
1. Minimization

f(x)

‐ f(x)

Thus Min f(x) = Max -f(x), so just maximize -z

Alternately, note that in maximization problem

the most negative value is chosen from the top
row of the tableau. For a minimization problem
the most positive value is selected from the top
row of the table. All other steps are the same for
both maximization and minimization problems.

24
 Minimize z = – 2x1+5x2
subject to
x1 +x2 < 12
3x1+x2 < 18
x1, x2 > 0.

Add slack variables to get

Minimize z = – 2x1+5x2
subject to
x1+x2+x3 = 12
3x1+x2 +x4 = 18
x1,x2,x3,x4 > 0.

BV z x1 x2 x3 x4 RHS

z 1 2 -5 0 0 0
x3 0 1 1 1 0 12
x4 0 3 1 0 1 18

z 1 0 -17/3 0 -2/3 -12

x3 0 0 2/3 1 -1/3 6
x1 0 1 1/3 0 1/3 6

25
Since there is no other positive value in the top row of
the tableau, the objective function has been
minimized with
z = -12
x1 = 6
x2 = 0 .

HOMEWORK 9
Minimize z = 2x1 – 4x2
s.t.
x1+x2 < 10
2x1+x2 < 16
x1,x2 > 0

26
2. Unrestricted variables

Maximize z = – 2x1 + x2
subject to
x1 + x2 < 10
2x1 + x2 < 16
x1 > 0
x2 UR

Let x2 = x2+ – x2 – , where x2+, x2 – > 0. These

new variables cannot be basic variables at
the same time. At least one of them will be
zero, or both of them will be zero.

Rewrite the problem as

Maximize z = –2x1 + (x2+- x2 – )
subject to
x1 + (x2+ – x2 – ) < 10
2x1 + (x2+ – x2 – ) < 16
x1, x2+, x2 – > 0 .
Add slack variables to give
Maximize z = –2x1 + x2+- x2 –
subject to
x1 + x2+ – x2 – +x3 = 10
2x1+ x2+ – x2 – +x4 = 16
x1, x2 , x2 , x3 , x4 > 0 .
+ –

Solve as usual and get x2 = (x2+ – x2 – ) afterwards.

27
 HOMEWORK 10
Maximize z = – 2x1 – 4x2
s.t.
x1 + 3x2 < 10
2x1 + x2 < 16
x1 > 0, x2 UR

3. Tie for Entering Variable

The top row of the maximization problem is

shown below. It is seen that coefficients of x1
and x2 are equal. Break the tie arbitrarily. It
doesn’t make difference which variable x1 or
x2 is chosen to enter.

BV z x1 x2 x3 x4 RHS
z 1 -1 -1 0 0 0

28
4. Tie for leaving variable
In the maximization tableau below the values are
same for both x3 and x4. Break the tie arbitrarily.

BV z x1 x2 x3 x4 RHS
z 1 0 -2 0 0 12
x4 0 1 1 0 1 12
x3 0 0 1 1 0 12
When there is a tie for leaving variable, the next
tableau has a BFS with a 0. Usually this has no
effect. Cycling is possible, but there are ways to
overcome it. We will discuss this later.

5. Alternate optimal solutions

The final iteration of the maximization
problem is shown below in the tableau. If a
non-basic variable has zero coefficient in the
top row, it means there are alternate optimal
solutions. Here x4 could be entered without
changing z.
BV z x1 x2 x3 x4 RHS
z 1 0 0 4 0 36
x1 0 1 0 3 2 12
x2 0 0 1 2 3 12

29
6. No feasible Solution - related to 7 below.

7. Initial basic feasible solution

To begin a problem, make every constraint
into an equality (hopefully with a
nonnegative rhs) by adding slack or
subtracting surplus variables. If it has <
sign, add a slack variable. If it has a > sign,
subtract a surplus variable. For every such
constraint without an obvious basic variable,
add an artificial variable.

Maximize z = x1+ 2x2 + 3x3

subject to,
x1 + x2 + x3 = 12
2x1 + x2 – x3 > 10
3x1 – x2 + 2x3 < 18
x1 , x2 , x3 > 0 .
Adding the slack variables to the constraints
x1 + x2 + x3 = 12
2x1 + x2 – x3 – x4 = 10
3x1 – x2 + 2x3 +x5 = 18
x1 , x2 , x3 , x4 , x5 > 0 .

30
Constraint 1 doesn’t have an obvious variable
that can be selected as basic variable. The
basic variable in constraint 2 is negative.
Therefore for these 2 constraints we add
artificial variables.
x1 + x2 + x3 + x6 = 12
2x1 + x2 – x3 – x4 + x7 = 10
3x1 – x2 + 2x3 + x5 = 18
x1 , x2 , x3 , x4 , x5, x6 , x7 > 0,
where x6 , x7 are artificial variables.

BIG-M METHOD
In the Big M method modify the objective
function penalizing artificial variable with M,
where positive M is a very big number.
Maximize z = x1+ 2x2 + 3x3 – Mx6 – Mx7
subject to
x1 + x2 + x3 + x6 = 12
2x1 + x2 – x3 – x4 + x7 = 10
3x1 - x2 + 2x3 + x5 = 18
x 1 , x 2 , x 3 , x 4 , x 5, x 6, x 7 > 0

31
 BIG-M METHOD
• Solve problem with artificial variables

• Two cases for solution:

– All artificial variables become non-basic
(zero) and z has no “M penality” if and only
if the original problem is feasible.
– At least one artificial variable is basic
(positive) and z has an “M penality” in
every optimal solution if and only if the
original problem is infeasible.

EXAMPLE OF CASE 1
Maximize z = 2x1 + x2
s.t.
x1 + x2 ≤10
x1 + 2x2 ≥ 16
x1, x2 ≥ 0.

Maximize z = 2x1 + x2 – Mx5

s.t.
x1 + x2 + x3 = 10
x1 + 2x2 – x4 + x5 = 16
x1, x , x , x , x ≥ 0,
2 3 4 5

where x5 is an artificial variable.

32
 BV z x1 x2 x3 x4 x5 RHS
z 1 -2 -1 0 0 M 0
x3 0 1 1 1 0 0 10
x5 0 1 2 0 -1 1 16
Making the coefficient of the basic variable 0 in
the first row.

BV z x1 x2 x3 x4 x5 RHS
z 1 -2-M -1-2M 0 M 0 -16M
x3 0 1 1 1 0 0 10
x5 0 1 2 0 -1 1 16

BV z x1 x2 X3 x4 x5 RHS
z 1 - 3/2 0 0 -1/2 1/2+ M 8
x3 0 1/2 0 1 1/2 - 1/2 2
x2 0 1/2 1 0 - 1/2 1/2 8

x5 left basis so original problem is feasible. Yay!

B.V. z x1 x2 x3 x4 x5 RHS
z 1 0 0 3 1 M-1 14
x1 0 1 0 2 1 -1 4
x2 0 0 1 -1 -1 1 6
All the coefficients of the x’s in the top row are non‐negative, 
so we have the optimal solution z =14, x1 = 4 and x2 = 6.

33
 EXAMPLE OF CASE 2
Maximize z = 2x1 + x2
s.t.
x1 + 2x2 ≤10
x1 + 2x2 ≥ 16
x1, x2 ≥ 0.

Maximize z = 2x1 + x2 – Mx5

s.t.
x1 + 2x2 + x3 = 10
x1 + 2x2 – x4 + x5 = 16
x1, x , x , x , x ≥ 0,
2 3 4 5

where x5 is an artificial variable.

BV z x1 x2 x3 x4 x5 RHS
z 1 -2 -1 0 0 M 0
x3 0 1 2 1 0 0 10
x5 0 1 2 0 -1 1 16
Making the coefficient of the basic variable 0 in
the first row.

BV z x1 x2 x3 x4 x5 RHS
z 1 -2-M -1-2M 0 M 0 -16M
x3 0 1 2 1 0 0 10
x5 0 1 2 0 -1 1 16

34
 BV z x1 x2 x3 x4 x5 RHS
z 1 - 3/2 0 1/2+ M M 0 5-6M
x2 0 1/2 1 1/2 0 0 5
x5 0 0 0 -1 -1 1 6

B.V. z x1 X2 x3 x4 x5 RHS
z 1 0 3 M+2 M 0 20-6M
x1 0 1 2 1 0 0 10
x5 0 0 0 -1 -1 1 6
Since all the coefficients of the x’s in the top row are non‐
negative, the solution is z =20‐6M, x1 = 10 and x5 = 6. But 
since x5 is artificial and positive z has a ‐6M penalty, the 
original problem is infeasible. 

HOMEWORK 11
Maximize z = x1 + 3x2
s.t.
x1 + x2 ≤ 8
4x1 + x2 ≥ 16
x1, x2 ≥ 0.

35
 8. Unbounded problem
Consider
Max z=x1
s.t.
x1 ≥ 0.
This problem is considered unbounded. Any LP
has one of the following:
• No feasible solution (LP is considered
infeasible)
• An unbounded objective (LP is considered
unbounded)
• An optimal solution

EXAMPLE
Maximize z = 2x1
s.t.
-x1 ≤10
x1 ≥ 0.

BV z x1 x2 RHS
z 1 -2 0 0
x2 0 -1 1 10
Since all the coefficients in the x1 column are negative, there
is no leaving variable. Hence, x1 and z can be increased
indefinitely, and the problem is unbounded.

36
 DUALITY
Primal Problem P:
n

Maximize z =  cjxj

j1

s.t.
a11x1 + a12x2 +…+ a1nxn < b1
---------------------------------
am1x1 + am2x2 +…+ amnxn < bm

x1,…,xn > 0

Dual Problem D:
m

Minimize w =     bi 1
iyi
s.t.
a11y1 + a21y2 +…+ am1ym > c1
---------------------------------
a1ny1 + a2ny2 +…+ amnym > cn

y1,…,ym > 0

37
DUALITY RELATIONSHIPS
1. The dual of the dual is the primal.

2. For any feasible (x1,…,xn) for P and any

feasible (y1,…,ym) for D ,then z < w.

3. For any optimal (x1*,…,xn*) for P and any

optimal (y1*,…,ym*) for D, then z* = w*.
Also, if z* = w*, then (x1*,…,xn*) and
(y1*,…,ym*) are optimal.

4. Primal unbounded => dual infeasible

Primal infeasible => dual unbounded or
dual infeasible
(See table)

Primal Dual States

Primal
Dual Infeasible Unbounded Optimal
Infeasible Rare X

Unbounded X

Optimal ☺

38
 ANY DUAL
Max Min
< Constraint > 0 Variable
> Constraint < 0 Variable
= Constraint UR Variable

> 0 Variable > Constraint

< 0 Variable < Constraint
UR Variable = Constraint

EXAMPLE

P. Minimize z = 2x1 + x2
s.t.
– x1 + 3x2 < -2
x1 – x2 = 6

x1 UR, x2 < 0

39
 D. Maximize w = – 2y1 + 6y2
s.t.
– y 1 + y2 = 2
3y1 – y2 > 1

y1 < 0, y2 UR

DUAL SIMPLEX
The dual simplex involves solving the dual on
the primal tableau with appropriate – signs.
1. Must have optimality criterion (nonnegative
coefficients for max problem) satisfied in top row.
2. Leaving variable is most negative RHS excluding
the z row.
3. Entering variable is found by finding the smallest
ratio of the absolute value of elements in top row
(excluding RHS) to negative elements of pivot
row. (If no negative elements, problem is
infeasible.)
4. Form a new tableau as before.
5. If all RHS (not z ) > 0, stop. Otherwise, go to 2.

40
 EXAMPLE
Minimize z = 2x1 + x2
s.t.
x1 + x2 > 4
5x1 + 3x2 > 15
x1 , x2 > 0.
Rewriting as equality constraints,

Minimize z = 2x1 + x2
s.t.
-x1 - x2 + x3 = -4
-5x1 - 3x2 + x4 = -15
x1 , x2 , x3 , x4 > 0.

DUAL SIMPLEX

BV z x1 x2 x3 x4 RHS
z 1 -2 -1 0 0 0
x3 0 -1 -1 1 0 -4
x4 0 -5 -3 0 1 -15
z 1 -1/3 0 0 -1/3 5
x3 0 2/3 0 1 -1/3 1
x2 0 5/3 1 0 -1/3 5

41
 HOMEWORK 12
Maximize z = – 3x1 – 2x2
s.t
x1 + x2 > 4
3x1 + 5x2 > 15
x1, x2 > 0.

APPLICATION

The dual simplex may be used to solve problems

in which one or more constraints are added after
an optimal solution is obtained to the original
problem.

Maximize z = 3x1 + 2x2

s.t
x1 + x2 < 4
x1, x2 > 0.

42
With a slack variable, the problem becomes

Maximize z = 3x1 + 2x2

s.t
x1 + x2 + x3 = 4,
x1, x2, x3 > 0.

BV z x1 x2 x3 RHS
z 1 -3 -2 0 0
x3 0 1 1 1 4

BV z x1 x2 x3 RHS
z 1 0 1 3 12
x1 0 1 1 1 4

43
Now add the constraint
x1 < 2
to the original problem. With a slack variable
x4 > 0, this constraint becomes
x1 + x4 = 2.
Add this equation to the tableau and proceed
using the dual simplex.

BV z x1 x2 x3 x4 RHS

z 1 0 1 3 0 12

x1 0 1 1 1 0 4

x4 0 1 0 0 1 2

44
BV z x1 x2 x3 x4 RHS

z 1 0 1 3 0 12 Dual
Simplex
x1 0 1 1 1 0 4

x4 0 0 -1 -1 1 -2

BV z x1 x2 x3 x4 RHS

z 1 0 0 2 1 10

x1 0 1 0 0 1 2

x2 0 0 1 1 -1 2

This tableau yields the new optimal solution.

45
 HOMEWORK 13

Minimize z = -2x1 + 5x2

s.t
3x1 + 4x2 < 12
x1,x2 > 0.

After solving this problem, add the constraint

x1 + 2x2 > 6.

INTEGER PROGRAMMING
Maximize z = 3x1 + x2
s.t.
x 1 + x2 < 4
3x1 + 5x2 < 15
x 1, x 2 > 0
x1, x2 integers.

If x1, x2 are integers, then the problem is a pure

integer programming problem. If only x1 or x2 is an
integer, then the problem is a mixed integer
programming problem.

46
The feasible region is given by the dots below.
x2

3 Feasible region for a LPP without

any integer constraints (relaxed LPP)
2

x1
1 2 3 4 5
x1 + x2 < 4 3x1 + 5x2 < 15

Integer programming problems are usually

solved by branch and bound algorithms.
There are other methods.

47
 EXAMPLE
Max z = 3x1 + x2
s.t.
4x1 + 3x2 < 10
x1, x2 > 0
x1, x2 integer.
The constraints become
4x1 + 3x2 + x3 =10
x 1, x 2 , x 3 > 0
x1, x2 integer.

Step 1: Solve relaxed LPP (without integer restrictions)

BV       z        x1               x2              x3                                         RHS
z          1       ‐3        ‐1         0                               0
x3  0        4         3           1                             10
5 3 15
4 4
z          1        0                                               2
3 1 5
x1 0         1                                               
4 4 2

48
Step 2: Use branch and bound

15
z =  2
x1 =  5 2
x2 = 0

x1 < 2 x1 > 3

z  =  20 3 Infeasible
x1 =  2 solution
x2 =  2 3

Use relaxed LP after adding x4 for x1 > 3 branch.

x1 – x4 = 3
- x1 + x4 = -3

Add x4 to the previous table

BV z x1 x2 x3 x4 RHS
5 15
z 1 0 4 ¾ 0 2
5
x1 0 1 ¾ ¼ 0 2
x4 0 -1 0 0 1 -3

Adding row constraints x1 and x4

x4 0 0 ¾ ¼ 1 -1/2

Dual is unbounded => Primal is infeasible

49
After adding x4 for x1 < 2, we obtain
x1 + x4 = 2
Add this equality with bv x4 to the optimal table.
BV       z         x1             x2            x3               x4                   RHS
5 15
z          1         0                      ¾          0              
4 2
x1 0         1          ¾         ¼           0                5 2
x4 0         1           0         0           1                2
Subtract the x1 row from the x4 row to get a new x4 row.
x4 3 1
0        0                                      1             1
4 4 2
Now x2 enters and x4 leaves.
z         1         0          0                    1 5 20
3 3 3
x1 0         1         0             0           1               2
1
x2           0          0         1                    
3 4
3
2
3

Fathoming Rules:

1. Infeasible
2. All xi integer that supposed to be
3. A value of z no better than some z for
feasible xi to original integer restrictions.

50
 Using branch and bound,
z   =  20 3
x1 = 2
x2 =  2 3
x2 < 0 x2 > 1

z = 6 z   =  25 4
x1 = 2 x1 =  7 4
x2 = 0 x2 = 1
x1 < 1 x1 > 2

z = 5
Infeasible
x1 = 1
x2 = 2

HOMEWORK 14
Max z = 3x1 + 2x2
s.t.
6x1 + 5x2 < 21
x 1, x 2 > 0
x1, x2 integer

51
 MIXED IPP EXAMPLE
Max z = 4 x1 – 2x2 + 7x3 – x4
s.t.
x1 +5 x3   < 10
x1  +   x2 – x3 < 1
6x1 – 5x2                < 0
– x1 + 2 x3 – 2 x4 < 3

xj > 0, for j = 1,2,3,4

xj integer for j = 1,2,3

Solution tree after first iteration:

z =  14 ¼
5 3
x1 =       , x
4 2 =  2
7
x3 =       , x
4 4 = 0

x1 > 2 x1 < 1

Infeasible z = 14 15
x1 = 1   , x2 =  6 5
9
x3 =      , x
5 4 = 0

52
 z = 14  15
x1 = 1  , x2 =  6 5
9
x3 =     , x
5 4 =  0

x2 > 2 x2 < 1

z = 12  16 z = 14  16
5
5
x1 =        , x
6 2 = 2
x1 =        , x
6 2 = 1
11
x3 =        , x
11
x3 =        , x
6 4 = 0 6 4 = 0

z  = 14 16
5
x1 =      , x
6 2 = 1
11
x3 =      ; x
6 4 = 0

x1 = 1 x1 = 0

Infeasible z = 13 ½
x1 = 0 , x2 = 0
x3 = 2 , x4 =   1 2
Optimal solution.

53
 HOMEWORK 15

Solve the following integer programming

problem.
Maximize z = x1 + x2
s.t.
x 1 + 5x2  11
3x1 + x2  8
x 1, x 2  0
x1 integer.

54
z   =  14
29
x1 =       14
x2 =  1425

x1 > 3 x1 < 2

Infeasible z   =  19 5
x1 =  2 
x2 =  9 5

54
 19
z   =  5
x1 =     2 
x2 =  9
5
x2 < 1 x2 > 2

z   =  3 z   =  3 
x1 =  2  x1 =  1 
x2 =  1 x2 =  2

BINARY IPP
Max z = 3x1 + x2
s.t.
4x1 + 3x2 < 10
x 1, x 2 > 0
x1, x2  { 0,1}
To solve, simply add
x1 < 1
x2 < 1
x1, x2 integer,
eliminate the binary constraint, and solve as a pure IPP.

55
 TRANSPORTATION
PROBLEM
1 2 m source
c12

1 2 n destination

The constant cij is the cost per unit from

source i to destination j, while xij is the
amount (integer number of units) shipped
from source i to destination j.

The aim of the model is to minimize the

shipping cost, i.e.,
m n

Min z =   cij xij

i 1 j 1

s.t.
n

 x = si , i = 1,…,m
j 1
ij

 x = dj , j = 1,…,n
i 1
ij

xij non-negative integer.

The transportation problem can be solved by

the simplex algorithm.

56
 ASSIGNMENT MODEL

1 2 n people
c12

1 2 n jobs

Here, cij=cost for assigning person i to do

the job j.

n n

Min z =   cij xij

i 1 j 1

s.t.
n

 xij = 1, i =1,…,n
j 1

 xij = 1, j = 1,…,n
n

i 1

xij  {0,1}.

The assignment model is a binary integer

programming problem that can be solved as
a transportation model.

57
 PRACTICE TEST 1
1. Apply the simplex method to the following
linear programming problem.

Maximize z = 6x1 – 3x2

s.t.
x1 + 2x2  8
x1 – x2  4
x1, x2 > 0.

Answer: z = 28, x1 = 16/3, x2 = 4/3.

2. Apply the simplex method to the following linear

programming problem.

Minimize z = – 5x1 + 6x2

s.t.
– 2x1 + x2  8
5x1 + 2x2 = 20
x1, x2  0.

Answer: z = 460/9, x1 = 4/9, x2 = 80/9.

58
3. The Crazy Nut Company wishes to market two special nut mixes during the
holiday season. Mix 1 contains 2/3 pound of peanuts and 1/3 pound of
cashews; mix 2 contains 3/5 pound of peanuts, 1/4 pound of cashews, and
3/20 pound of almonds. Mix 1 sells for $1.49 per pound; mix 2 sells for $1.69
per pound. The data pertinent to the raw ingredients appear in the table.
Assuming that Crazy Nut can sell all cans of either mix that it produces,
formulate an LP model to determine how much of mixes 1 and 2 the company
should produce.

Ingredient Amount Available Cost per

(lb) lb
Peanuts 30,000 $.35

Cashews 12,000 $.50

Almonds 10,000 $.70

Answer:
x1 = number # mix 1

x2 = number # mix 2

Max z = 1.09x1 + 1.25x2

s.t.
2/3x1 + 3/5x2  30000

1/3x1 + 1/4x2  12000

3/20x2  10000

x1, x2  0.

59
 4. Consider the linear programming problem
max z = 4x1 + x2
s.t.
2x1 + 3x2  6
x1, x2  0.
The optimal tableau is given below, where x3 is the slack variable added to
the constraint.
BV z x1 x2 x3 RH
S

Z 1 0 5 2 12

x1 0 1 3/2 1/2 3

Add the constraint x2  1 to the original problem and solve it

beginning with the above tableau.

Answer: z = 7, x1 = 3/2, x2 = 1.

5. Solve the following integer programming problem.

Maximize z = x1 + x2
s.t.

x1 + 5x2  11

3x1 + x2  8

x1, x2  0

x1, x2 integer.

Solution on next page:

60
 54
z = 14
29
x1 = 14
x2 = 25
14

x1 > 3 x1 < 2

Infeasible z = 19 5
x1 = 2
x2 = 9 5

19
z = 5
x1 = 2
x2 = 9
5

x2 < 1 x2 > 2

z = 3 z = 3
x1 = 2 x1 = 1
x2 = 1 x2 = 2

61
 REVIEW FOR QUIZ 1

NONLINEAR
PROGRAMMING
max f ( x1 , x2 , x3 )  x12  x1 x2  x1 x3
s.t.
x1
 x32  100
x2
x1  2 x2 x3  250
x1 , x2 , x3  0

62
 Applications of NLP
• Data networks – routing
• Production planning
• Resource allocation
• Modeling human or organizational
behavior

Example 1
Given a fixed area of cardboard A unit ,
formulate a nonlinear program to find the
dimension of a six-sided rectangular box
with maximum volume.

. . 2 2 2
, , ≥0

63
 Example 2
Consider the problem of determining locations for two new
high schools in a set of subdivisions . Let be the
number of students going to school A and be the
number of students going to school B from subdivision .
Assume that the student capacity of school A is and the
capacity of school B is and that the total number of
students in each subdivision is . We would like to
minimize the total distance traveled by all the students
given that they may attend either school A or B. Construct
a nonlinear program to determine the locations , and
, of high schools A and B, respectively assuming the
location of each subdivision is modeled as a single point
denoted , .

Answer

. .
∑ , 1, 2; 1, … ,
, 1, … ,

64
 Notation
• f :R1 R1 means f(x) = y
• f :Rn R1 means f(x1,…xn) = y
• In general we want to

max f ( x1 ,...xn )
s.t.
( x1 ,...xn )  A  R n
where A is the feasible region.

Classical Optimization in R1
• Let f :R1 R1. We first wish to
(P1) maximize f (x)
s.t.
x  R1
where f is differentiable. Here, there are no
constraints and the feasible region is R1.

65
 Definitions
1. The point x* is a local maximum if
f (x)  f (x*) for all x in some
neighborhood of x*.

2. The point x* is a global maximum if

f (x)  f (x*) for all x.

3. The term maximum means global

maximum and is a point in the domain.

SOME TERMS IN LOGIC

• Let p and q be propositions - statements
that can be judged true or false.

• Necessary condition
Consider the proposition: “If p, then q.”
Then q is said to be a necessary
condition for p.

66
• Sufficient condition
Consider the proposition: “If p, then q.”
Then p is said to be a sufficient condition for q.

• Contrapositive
Consider the following two propostions.
(a) “If p, then q.”
(b) “If not q, then not p.”
Then (b) is called the contrapositive of (a).

Definitions

f (x)

x1 x2 x3 x

67
 Results

• Necessary condition from calculus

– If x* solves P1, then f '(x*)=0.
• Sufficient condition 1
– Let x* be an stationary point (with f '(x*)=0).
Then if f ''(x*) < 0, x* is a local maximum.
• Sufficient condition 2
– If the order of the first non-vanishing
derivative is even and the value at
x* is < 0, then x* is a local max.

Critical Points
Any extremum of f (x) over [a,b] must be one
of the following critical points:

1. A stationary point in [a,b] where f '(x)=0

2. a,b
3. Points in [a,b] for which f '(x) is not defined.

68
 Critical Point Problem
• Optimize f (x)
s.t.
x[a,b]

a b

Example
Optimize f (x) = (x-2)3 + |x|
s.t.
x  [-1,3]

To find the stationary points we write

( x  2) 3  x, x  0
f ( x)  
( x  2) 3  x, x  0

69
 It follows that
 3( x  2) 2  1, x  0
f ( x)  
 3( x  2) 1, x  0.
2

Thus
3 x 2 -12 x +13=0, x  0
3 x 2 -12 x +11=0, x  0.

So
3
x  2  ,x0
3
3
x  2   1433
. , 2.577, x  0.
3
In both cases, the solutions are not negative and
hence there are no stationary points for x  0.
The critical points are thus
1. 
2. x  1,3
3. x  0.

70
To obtain the optima, we evaluate the objective
function at the critical points.

x f(x)

min -1 -26

0 -8

max 3 4

HOMEWORK 16

Minimize f (x) = |x+3| + x3

s.t.
x[-2,6]

71
 Unconstrained Problems in Rn
• Let f : Rn  R1
• Let x  Rn

• n n
Q( x)  x Ax   aij xi x j
T

i 1 j 1
• Q(x) is called a quadratic form, where the
matrix A = [aij] is symmetric.

Q(x) is called:
1. Positive definite if Q(x)>0,  x0
2. Positive semidefinite if Q(x)0,  x and
x0 such that Q(x)=0
3. Negative definite if Q(x)<0,  x0
4. Negative semidefinite if Q(x)0,  x and
x0 such that Q(x)=0
5. Indefinite otherwise

72
 Positive/Negative Definite
• Q(x) = xTAx is positive definite if the
values of the leading principal minor
determinants of A are all positive. A is
called positive definite in this case.
• Q(x) = xTAx is negative definite if the
value of the kth leading principal minor
determinant has the sign [-1]k, k=1,2,…,n.
A is called negative definite in this case.

Hessian Matrices
• In particular, we are interested in
determining whether the Hessian matrix
H(x) of a function f at point x = (x1,…xn) is
positive definite, etc.
• The Hessian is defined as the n x n matrix

  2 f ( x) 
H ( x)   .
 
 i j 
x x

73
 Example
• Show that the Hessian matrix of the
function f(x) below is negative definite for
all x.

f( x1, x2 )  x1  3x2  4x1x2  2x12  3x22  7

• H= -4 -4 for all x
-4 -6

• Then the principal minor determinants are

-4 and 8

• Thus the Hessian of f(x) is negative definite for

all x.

74
 Necessary and Sufficient
Optimality Conditions
• Necessary condition
– A necessary condition for X0 to be an extreme
point of f(X) is that f(X0) = 0

• Sufficient condition
– A sufficient condition for a stationary point X0
to be an extremum is for the Hessian matrix H
evaluated at X0 to be
• Positive definite when X0 is a local minimum point
• Negative definite when X0 is a local maximum
point.

Example

• Consider the function

f ( x1, x2 , x3 )  x1  2x3  x2 x3  x12  x22  x32.

• The necessary condition f(X0) = 0 gives

X0 = (1/2, 2/3, 4/3).

75
 Example

• To establish sufficiency compute

H|X0 = -2 0 0
0 -2 1
0 1 -2
• The principle minor determinants have the values
– 2,4, and – 6.

• Thus H|X0 is a negative definite and X0 = (1/2, 2/3,

4/3) represents a local maximum point.

Convex and Concave Functions

• A function of f : Rn  R1 is convex on a convex
set S if for any two points x1,x2  S and any
[0,1]
f ( x1  (1   ) x2 )   f ( x1 )  (1   ) f ( x2 )

• A function of f : Rn  R1 is concave on a
convex set S if for any two points x1,x2  S and
any [0,1]
f ( x1  (1   ) x2 )   f ( x1 )  (1   ) f ( x2 )

76
 Convex and Concave
Functions
f(x) f(x)

x x

Convex and Concave Functions

• A function f is convex (concave) on a
convex set S if and only if its Hessian
matrix H(x) is positive (negative)
semidefinite for all xS.

• A convex function achieves its global

minimum at a stationary point.

• A concave function achieves its global

maximum at a stationary point.

77
 • For example we know that the function f
below is concave on R2 because we have
already shown that its Hessian matrix is
negative semidefinite for all (x1,x2).

f (x1, x2 )  x1 3x2  4x1x2 2x12 3x22 7

HOMEWORK 17
Check if the following function is convex or
concave:
, 3 7.

78
 Constrained Problems in Rn

(P) max f ( x1 ,..., xn )

s.t.
gi ( x1 ,..., xn )  0, i  1,..., m
hk ( x1 ,..., xn )  0, i  1,..., p

Checking for Short Cuts

• Check to see if you can find a solution to P by
ignoring some or all of the constraints. If this
solution also satisfies the neglected
constraints, then it is optimal to P.

• Check to see if you can reduce the number of

variables in a problem by solving some of the
equality constraints for variables and
substituting into the objective function and
remaining constraints.

79
 EXAMPLE 1
minimize (x1  1)2  ( x2  3) 2
s.t.
x13  x2.25  4
3
| sin (x1 x2 )|  .
2

By neglecting both constraints,the answer

by inspection is x1*  1, x2*  3.

EXAMPLE 2
1
maximize x 3

s.t.
|x|  1
x1* ,... x*n

Sin -1 (x )  1.5 .

By neglecting the second constraint, the

answer by inspection is x *  1.

80
 EXAMPLE 3
(I) maximize x1 (1  x 2 )
s.t.
x1  x 2  1
x1 2  x 2 2  1.
From the first constraint, x 2  1  x1 .
Substitution gives the problem
(II) maximize x1 2
s. t.
x1 2  (1  x1 ) 2  1.

The constraint to (II) reduces to

2 x12  2 x1  0
or
x1 ( x1  1)  0.
By taking cases this gives
0  x1  1.
The answer to (II) is thus x1*  1 by
inspection and to (I) is x1*  1, x2*  0.

81
 HOMEWORK 18

minimize x12  x22

s.t.
x1  x2  1

 x 12  cos( x )  1
 1  2 2
 
x1 , x2  0.

TO FIND A GLOBAL MAXIMUM FOR PROBLEM P IN PRACTICE:

• Find all interior points ( x1 ,..., xn ) for which the partials are 0
(stationary points).
• If f is not differentiable everywhere, include also points where the
f
partials x do not exist together with stationary points to give
i

critical points.
• Find all boundary points ( x1 ,..., xn ) satisfying
d   f ( x1 ,..., xn )  0 for all feasible directions d.
• Compare values at all these points and choose the largest value.
The points giving this largest value are the global maxima if a
maximum exists.
The following conditions provide a method for doing this.

82
 FRITZ JOHN NECESSARY
CONDITIONS FOR PROBLEM P
Suppose that f , g1 ,, gm , and h1 ,, hp
are all differentiable at the maximum
(x1* ,,x*n ) for P. Then there exist real
numbers (Lagrange multipliers)
 , 1 ,,  m , and 1 ,,  p such that
(x1* ,,x*n ) and these constants satisfy
the following conditions:

Fritz John
Necessary Conditions
f m
gi p
h
1.    i   k k  0, j  1,..n
x j i 1 x j k 1 x j
( x1* ,... xn* )

2. i gi ( x1* ,..., xn* )  0, i  1,..., m

3.  , i  0, i  1,..., m
4.  , 1 ,..., m , 1 ,...,  p are not all 0

5. gi ( x1* ,...xn* )  0, i  1,..., m

hk ( x1* ,...xn* )  0, i  1,..., p

83
 Karush-Kuhn-Tucker
Necessary Conditions
• Very frequently the Fritz John conditions
can be simplified somewhat because the
constraints are “nice” at (x1*,…xn*). In that
case =1 and condition 4 is automatically
satisfied.
• For our purposes, we take =1 only when
all constraints are linear (or Nota Bene
applies) although the optimum usually
occurs when =1. Otherwise the
necessary conditions do not involve the
objective function.

Procedure
• Solve the Fritz John conditions to get the
candidates for the maximum P.

• If the solution exists, it must be among the

candidates.

• Evaluate all candidates.

• The best one is maximum if a maximum exists.

84
 Nota Bene
• If you’re lucky, then
– The objective function f and all of the
inequality constraints g are concave.
– The equality constraints h are linear.

• In this case, the KKT conditions are both

necessary and sufficient.

• α= 1.

• No need to check all candidate solutions.

Weierstrass (Extreme Value)

Theorem
• A continuous function on a closed
bounded region achieves both its
maximum and minimum on the region.

• It follows that if all the functions in P are

continuous and the feasible region is
bounded, then P has a solution.

85
 HOMEWORK 19-23

Five nonlinear programming problems solved by

Lingo, Matlab, or some other standard software are
due on March 20 at the beginning of class. These
problems must be formulated by the student, so each
student will have different problems. A trial version of
Lingo can be obtained at www.lindo.com. A copy of
your problems and some form of computer printout of
the solution are required.

EXAMPLE 1
2 2
min x1  x2
s.t.
x1  2 x2  3
2 x1  x2  3

86
Example 1 in Standard Form:

max f ( x1 , x2 )   x12  x2 2
s.t.
g1 ( x1 , x2 )  3  x1  2 x2  0
g 2 ( x1 , x2 )  3  2 x1  x2  0

Example 1 - Questions

• Are all constraints linear?

– Yes, so =1
• Is the feasible region bounded?
– No
– We must worry about existence.

87
Example 1-Fritz John Conditions
 2 x1  1  2 2  0
(1) 
 2 x2  2 1  2  0
  ( x  2 x2  3)  0
(2)  1 1
 2 (2 x1  x2  3)  0
(3) 1 , 2  0
(4)okay
3  x1  2 x2  0
(5) 
3  2 x1  x2  0

Example 1 – Cases from condition (3)

• Case (a) 1=0, 2=0

(1)  x1  x2  0
• Case (b):1=0, 2>0
 2 x1  2 2  0  x1    2

(1)   2 x2   2  0  x2    2
2

 x1  2 x2
(2)  2 x1  x2  3  x1  6 / 5, x2  3 / 5
(5)okay

88
Example 1 – Cases from condition (3)

• Case (c) 1>0, 2=0

 2 x  2   0  x   1
 1 1 1 2

(1)   2 x2  2 1  0  x2   1
 x  2 x
 2 1

(2)  x1  2 x2  3  x1  3/ 5, x2  6 / 5
(5)okay

Example 1 – Cases from condition (3)

• Case (d) 1>0, 2>0

 x1  2 x2  3

(2)  2 x1  x2  3
 x  x  1
 1 2

(5)okay

89
Example 1 – Evaluation of candidates

(x1,x2) f (x1,x2)
max (0, 0) 0

(6/5, 3/5) -45/25

(3/5, 6/5) -45/25

(1, 1) -2

If there is a max, it occurs at the point (0,0)

But wait ….
• We didn’t ask a most important question -
is there a short cut or easier way ?

2 2
min x1  x2
s.t.
x1  2 x2  3
2 x1  x2  3

90
 Questions we should ask:
 Is there an easier way?
 Yes
 Then do it and save yourself lots of time.

 Are all constraints linear?

 Yes, so =1’

 Is the feasible region bounded?

 No
 We must worry about existence.

EXAMPLE 2

min x1  3 x 2
s.t .
2 x12  5 x 2 2  230

91
 Put in standard form.

max f (x1 , x2 )   x1  3x2

s.t.
h1 ( x1 , x2 )  2 x12  5 x2 2  230  0

Questions we should ask:

Is there an easier way?
 No

Are all constraints linear?

 No

Is the feasible region bounded?

 Yes
 The best candidate is the solution.

92
 Fritz John Conditions
   4 x11  0
(1) 
 3  10 x21  0
(2) NA
(3)  0
(4) , 1 not both zero
(5)2 x12  5 x22  230  0

Case I:   0
(5)  x1 and x2 not both 0.
(1)  4 x11  0  10 x2 1.
 1  0  (4).

93
Case II :   1
 4 x  1
(1)   1 1  1  0.
101 x2  3
 x2  6 5 x1.
(5)  2 x12  5  6 5 x1   230.
2

 ( x1 , x2 )  (5,6), (5,  6).

Obviously both satisfy (5).

 x1 , x2  f ( x1 , x2 )
(5,6) 23
max ( 5,6) 23

Therefore the minimum to the original

problem is (5,6).

94
 EXAMPLE 3

maximize x1  x2
s.t.
x12  x22  1
x1 , x2  0.

We rewrite in the standard form of P.

maximize f (x1 , x2 )  x1  x2
s.t.
g1 (x1 , x2 )  x1 0

g 2 (x1 , x2 )  x2 0

h1 (x1 , x2 )  x12  x22  1  0.

95
 Questions we should ask:
Is there an easier way?
 No

Are all constraints linear?

 No

Is the feasible region bounded?

 Yes
 The best candidate is the solution.

Write the Fritz John Conditions.

  1  2 x11  0
(1) 
  2  2 x21  0
 x  0
(2)  1 1
  2 x2  0
(3) , 1 , 2  0
(4) , 1 , 2 , 1 not all zero
 x1  0

(5)  x2  0
 x2  x2  1  0
 1 2

96
Case I:   0
subcase (a): 1  0,  2  0
(4)  1  0.
(1)  x1  x2  0  (5).

subcase (b): 1  0,  2  0
(2)  x2  0.
(1)   2  0   2  0.

subcase (c): 1  0,  2  0
(2)  x1  0.
(1)  1  0  1  0.

subcase (d): 1  0,  2  0
(2)  x1  x2  0  (5).

97
Case II :   1
subcase (a) : 1  0, 2  0
1  21 x1  0
(1)  
1  21 x2  0.
 1  0.
Hence x1  x2   12 .
1

(5)  x12  x12  1.

  1
2,
1
 is a candidate
2

but not   12 ,  12  .

subcase (b) : 1  0,  2  0
(2)  x2  0.
(5)  x1  1.
 (1,0) but not (-1,0) is a candidate.

subcase (c) : 1  0,  2  0
 a candidate is (0,1) as before.

subcase (d) : 1  0,  2  0
(2)  x1  x2  0  (5) as before.

98
  x1 , x2  f (x1 , x2 )

(1, 0) 1

 0,1 1

max  1
2,
1
2  2

EXAMPLE 4

maximize x1  x 2
s.t .
x12  x 2 2  4
2 x1  x 2  2

99
 Put in standard form.

max f ( x1 , x2 )  x1  x2
s.t.
g1 ( x1 , x2 )   x12  x2 2  4  0
h1 ( x1 , x2 ) 2 x1  x2  2  0

Questions we should ask:

 Is there an easier way?
 No

 Are all constraints linear?

 No

 Is the feasible region bounded?

 Yes
 The best candidate is the solution.

100
 Fritz John Conditions
 x  2 1 x1  21  0
(1)  2
 x1  2 1 x2  1  0
(2) 1 ( x12  x22  4)  0
(3)  , 1  0
(4)  , 1 , 1 not all zero
 x12  x22  4  0
(5) 
2 x1  x2  2  0

Case I :   0
subcase (a) : 1  0
(1)  1  0  (4).
subcase (b) : 1  0
(2)  x12  x2 2  4
(5)  2 x1  x2  2  x2  2  2 x1
 x12   2  2 x1   4.
2

So  x1 , x2    0, 2  ,  8 5 ,  6 5  .
Both these satisfy (5) and are candidates.

101
Case II :   1
subcase (a) : 1  0
 x2  21  0
(1)  
 x1  1  0
 x2  2 x1.
(5)  2 x1  x2  2
 x1  1
2 ,x2  1.
 1 2 ,1 satisfies (5).

subcase (b) : 1  0
(2)  x12  x2 2  4
(5)  2 x1  x2  2
These have already been solved.

( x1 , x2 ) f ( x1 , x2 )

(0, 2) 0
 85 ,  65   48 25
max  12 ,1 1
2

102
HOMEWORK 24 - 27
(24) maximize 2 x1  5 x2
s.t.
2 x12  5 x2 2  13
(25) minimize x1  2 x2
s.t.
x12  x2 2  4
3 x1 +x2  4
(26) maximize x1  x2
s.t.
x1  x2  4
x1 , x2  0
(27) maximize x1  x2
s.t.
x1  x2  4

REVIEW FOR QUIZ 2

103
 NETWORK ANALYSIS
arc / branch

distance / time node

SHORTEST PATH
7
A D
2 2 4 5

O B 1
5 T
3
4 1
7
C E
4

Note that the arcs / branches have no direction,

ie., are undirected.

104
 Seek a path from origin node O to terminal node T that
minimizes the total distance, which is positive.

• Find all 1st closest nodes to O; then label them with the
total distance to O and the preceding node in the path.

• Find the second closest nodes adjacent to the first closest

nodes or origin.

• Proceeding, we eventually find T as the nth closest node

for some n.

• Working backwards gives the shortest paths.

This is essentially Dijkstra’s method. If arcs are directed,

we can only do this in one direction.

HOMEWORK 28

4
A D
7 6 4 3

O 8
B E
6 6 T

5 5 3
5
C F
4

105
 EXAMPLE
John drives to UTA every day. Having just
completed a course in network analysis, John is
able to determine the shortest route to work.
Unfortunately, the selected route is heavily
patrolled by police, and with all the fines paid for
speeding, the shortest route may not be the best
choice. John has thus decided to choose a route
that maximizes the probability of not being
stopped by police. The next figure shows the
possible route between home and work, and the
associated probabilities of not being stopped on
each segment.

Shortest path: 1-3-5-7 = 0.0675

106
 FLOW IN NETWORKS

A B
4 1

represents capacity of A to B

represents capacity of B to A

MAXIMAL FLOW

3 0
A D 9
0 0
1 1
5 1 4
Flow
O 0
B Flow
7 0 5 T
4 2 0
1
0 0
0
C E 6
4 0

O = Source T = Sink

107
 PROCEDURE
1. Find a path with positive capacity from source to sink.
Put that much flow in it.

2. Reduce the forward capacity by the flow and increase the

backward capacity by the same amount to allow the
possibility of undoing what you did.

3. Repeat until this cannot be done. At that point you can find
a cut with 0 remaining capacity from source side of cut to
sink side.

4. For an example, see the link http://optlab-

server.sce.carleton.ca/POAnimations2007/MaxFlow.html.

• Definition: A cut is a minimum set of branches

whose breaking will separate the source from
sink. Its cut value is the sum of the capacities of
its branches from the source side of the cut to
the sink side.

• Max. Flow - Min. Cut Theorem: The maximum

flow through the network equals the minimum
cut value, where the cut value of a cut is the
sum of capacities from source to sink direction.

108
 Homework 29

6 3
A D 9
2 0
5 5
8 0 5
O 0
B
0 3 T
9
0
5
4 0
4
C E 10
2 2

DYNAMIC
PROGRAMMING

109
 SURFING EXAMPLE
States
5 6
N Se Ch Wa
6 4 5 7

6
M SF 6 KC 6 At
6 5 6 7 5

5 7 5
S LA EP NO Miami

1 2 3 Stages

The goal of the above problem is to reach

west coast from Miami in a minimum
possible time.

Stage – where a decision is made,

numbered here as the number of
decisions left to make.

State - where you are in a stage.

110
 SOLUTION
stages: cities
states: north, middle, and south
decision variables: destination city at
next stage
returns: time to reach next destination at
next stage

d3 d2 d1
s3 s2 s1
3 2 1

r3 r2 r1

Stage State time to best dstn. rem. time

1 N 4 SF 4
M 5 LA 5
S 5 LA 5
_____________________________________
2 N 6+4 10
7+5
M 5+4 9
6+5
7+5

111
 Stage State time to dstn. rem. time

S 6+5 11
7+5
_________________________________
3 S 6+10
5+9 14
5+11

ANSWER

Miami At Ch SF

minimum time 5+5+4 = 14 hours

112
 KEY POINTS OF DYNAMIC
PROGRAMMING
1) Work backwards.
2) Break a problem into subproblems.
3) Bellman’s principle of optimality: If a
decision is to be made, make the best
decision from here forward. Forget the
past. Optimal subpolicies so obtained
yield an optimal overall policy.

PROPERTIES OF SOLUTION
PROCEDURE

1) A problem is divided into stages. A stage

is where a decision is made.
2) DP transforms higher-dimensional
problems into multiple lower-dimensional
ones.
3) Each stage has a number of states
associated with it. A state is “where
you’re at” in the stage.

113
 4) One proceeds from a state in one
stage to another state in the next stage.
The optimal decision is found for each
possible state of a stage, where optimal
refers to future stages.
The “curse of dimensionality” refers to
the fact that one could be required to
solve large numbers of problem with
large numbers of states.
5) Given the current stage, the optimal
policy for the remaining stages is
independent of the policy adopted in
previous stages.

PROBLEM FORMULATION
AND NOTATION

• dn= decision variable at stage n

• sn = state variable at stage n
• rn= rn(sn, dn) = return at stage n in state sn for
decision dn
• tn= stage transformation sn-1= tn(sn, dn) at
stage n

114
• fn (sn) = optimal n stage return objective
function that is only a function of
state

fn (sn) = optimum [rn(sn, dn)  fn-1 (sn-1)]

dn

s.t
sn-1 = tn(sn, dn), n=2,3,……,N.

f1 (s1) = optimum [r1(s1, d1)], n=1.

d1

dN dN-1 dn dn-1 d1
sN sN-1 sN-2 sn sn-1 sn-2 s1
N N-1 n n-1 1

rN rN-1 rn rn-1 r1

fn-1(sn-1)

115
STEPS IN SOLVING A PROBLEM

1) Define the mathematical optimization

problem to be solved. This step may involve
modeling a real-world situation.

2) Define the stages, the states, and the

entities
d n , s n , r n , tn , fn .

3) Give DP formulation to (1) by specifying the

recursive equations of dynamic programming

fn (sn) = optimum [rn(sn, dn)  fn -1 (sn -1)]

dn

s.t
sn-1 = tn(sn, dn), n = 2,3,…,N.

f1 (s1) = optimum [r1(s1, d1)], n = 1.

d1

4) Perform the computations.

116
Example 1: Allocation Problem
David Goodfellow has 3 children whose ages are
2, 3, 4. One day after work he decides to treat
his children by buying them some candy.
Unfortunately he has only 3¢, so he buys them 4
jellybeans. He desires, however, to utilize these
4 jellybeans to the fullest. He knows from
experience that none of his children will eat more
than 2 jellybeans. Moreover, being an observant
father and knowing his children’s characteristics,
he estimates the units of pleasure that 0, 1, 2
jellybeans will give each child. The following
table gives these estimates :

Units of Pleasure
jellybeans Roosevelt David Jr. Meridith
(Davy)
0 0 0 1

1 1 2 3

2 3 4 0

117
 Suppose David, Sr. wishes to maximize the
total pleasure that the jellybeans will give his
children. Using dynamic programming,
allocate the jellybeans optimally among the
3 children.

Solution

stages: children
states: # of jellybeans left
decision variables: how many jellybeans to
give the child at stage n
returns: pleasure units (ahs!) at stage n

118
 Stage 1
s1 d1 r1 f1(s1) = max r1
4 0 0
1 1
2* 3 3
3 0 0
1 1
2* 3 3
2 0 0
1 1
2* 3 3
1 0 0
1* 1 1
0 0* 0 0

Stage 2
s2 d2 r2 s1 = s2 - d2 r2 + f1(s1) f2(s2)
4 0 0 4 0+3=3
1 2 3 2+3=5
2* 4 2 4+3=7 7

3 0 0 3 0+3=3
1* 2 2 2+3=5 5
2* 4 1 4+1=5 5

2 0 0 2 0+3=3
1 2 1 2+1=3
2* 4 0 4+0=4 4

119
 Stage 3

s3 d3 r3 s2 = s3 -d3 r3 + f2(s2) f3(s3)

4 0* 1 4 1+7=8 8
1* 3 3 3+5=8 8
2 0 2 0+4=4

Answer:

M D R
0 2 2
1 1 2
1 2 1

Example 2: Allocation Problem

A small plane is being loaded with cargo

consisting of 3 types of items. These
items must be loaded in integer quantities
, and the plane can carry only 1000
pounds of cargo. The profit realized from
one unit of each type of item, as well as
its weight per unit, is given below.
Maximize the value of the cargo for the
plane subject to the weight restriction.

120
 Data
item i Weight per Profit per
unit unit
A 250 $ 3000

B 300 $ 4000

C 400 $ 5000

Solution
• stages: items
• states: remaining weight available to
allocate
• decision variables: units of item n
• return: profit for item n
d3 d2 d1
s2 s1
s3 =1000  3  2  1
r3 r2 r1

121
 Stage 1 - C

s1 d1* f1(s1)

0 ≤ s1< 400 0 0
400 ≤ s1< 800 1 5000
800 ≤ s1 ≤1000 2 10,000

Stage 2 - B
s2 d2 r2 s1 r2+f1(s1) f2(s2)=max [r2+f1(s1)]

1000 0 0 1000 0 + 10,000

1 4000 700 4000 + 5000
2 * 8000 400 8000 + 5000 13,000
3 12,000 100 12000 + 0

750 0 0 750 0 + 5000

1* 4000 450 4000 + 5000 9000
2 8000 150 8000 + 0

500 0* 0 500 0 + 5000 5000

1 4000 200 4000 + 0

250 0* 0 250 0+0 0

0 0* 0 0 0+0 0

122
 Stage 3 - A
s3 d3 r3 s2 r3+f2(s2) f3(s3)=max [r3+f2(s2)]

1000 0* 0 1000 0 + 13,000 13,000

1 3000 750 3000 + 9000
2 6000 500 6000 + 5000
3 9000 250 9000 + 0
4 12,000 0 12,000 + 0

Answer : item optimal #

A 0
B 2
C 1

HOMEWORK 30
A student has final examinations in 3
courses X,Y, Z, each of which is a 3 credit-
hour course. He has 12 hours available
for study period. He feels that it would be
best to break the 12 hours up into 3 blocks
of 4 hours each and to devote each 4-hour
block to one particular course. His
estimates of his grades based on various
numbers of hours devoted to studying
each course are as follows. Using dynamic
programming, allocate his study time
optimally.

123
 Data
Number of hours

0 4 8 12

F D D B
X

D D B A
Course Y

F D B B
Z

Example 3: Allocation Problem

Consider a system with 3 components in
series of types A, B, and C respectively.
If one component fails, the system fails.
The reliability of the system (that is, the
probability that all types of components
work properly) can be improved by
installing redundant components in
parallel. Suppose that the unit cost and
probability of failure of each type of
component is given below.

124
 Type Failure Probability Cost
A 0.6 $1
B 0.4 $2
C 0.5 $3

Assume that you have $10 to spend on

components. This answer includes money for
at least one of each type. Use dynamic
programming to determine the optimum
number of components of each type to buy so
as to maximize the reliability of the system.

Answer
Maximize [1  0 . 6 d 1  1 ][1  0 . 4 d 2  1 ][1  0 . 5 d 3  1 ]
d1 , d 2 , d 3

s.t.
1d1  2 d 2  3d 3  4
d1 , d 2 , d 3  0,1, 2,3,

• stages: components
• states: si = $ left to spend at stage i
• decision variables: d n = # redundancies of type n
• returns: rn = prob. of success at stage n
dn 1
rn (d n , sn )  1  qn
• stage transformations
s n -1  s n - c n d n

125
 Stage 1: type A
*
s1 d1 r1  (1  0.6 d1 1 ) f1 ( s1 )  max(1  0.6 d1 1 )

4 4 1- 0.65 =0.92 0.92

3 3 1- 0.64 =0.87 0.87

2 2 1- 0.63 =0.78 0.78

1 1 1- 0.62 =0.64 0.64

0 0 1- 0.61 =0.40 0.40

Stage 2: type B
s2 d2 r2  1  0.4d2 1 s1  s2  2d2 r2  f1 (s1 ) f 2 (s2 )  max[r2  f1 (s1 )]

4 0 0.6 4 0.60  .92=0.55

1* 0.84 2 0.84  0.78=0.65 0.65
2 0.94 0 0.94  0.40=0.37

1 0* 0.6 1 0.6  0.64=0.38 0.38

Stage 3: type C

s3 d3 r3  1  0.5d3 1 s2  s3  3d3 r3  f 2 (s2 ) f 3 (s3 )  max[r3  f 2 (s2 )]

4 0* 0.5 4 0.5  0.65=0.325 0.325
1 0.75 1 0.75  0.38=0.29

126
 Maximum reliability = 0.325

type redundancies total number

C 0 1

B 1 2

A 2 3

Example 4: Optimal Stopping Rule

An oral examination is designed as follows.
There are 4 questions of which everyone is
required to attempt the first question. If this first
question is answered correctly, a student
receives a numerical grade of 50 in the course.
Otherwise, he receives a 0. For the remaining 3
questions , a student has the option of keeping
his numerical grade from the previous question
or attempting the next question for a higher
grade. At each question, if the question is not
answered correctly, the student receives a
consolation grade.

127
The probabilities of correctly answering the three
optional questions, the grade received for a
correct response, and the consolation grade are
given below. Formulate an optimal policy for
taking this oral examination to maximize the
expected numerical grade you receive.

Optional Question Probability Grade Consolation Grade

 ( first ) 0.6 65 40
 ( second ) 0.5 80 55
 ( third ) 0.3 100 70

Answer
• stages: optional questions
• decision variable: stop or go
• returns: expected grade
• states: grade in hand

128
Stage 1: Question γ

Max{80, 0.3(100)+0.7(70) = 79}= 80 Stop

Stage 2: Question 

Max{65, 0.5(80)+0.5(55) = 67.5}= 67.5 Go

Stage 3: Question 

Max{50, 0.6(67.5)+0.4(40) = 56.5}= 56.5 Go

Question Decision

α ( first ) Go

β ( second ) Go

γ ( third ) Stop

129
 Example 5: Traveling
Salesman Problem
A businessman must travel to each of the
following cities B, C, D, E, starting from A
and ending in A. He can go through each
city only once except that, of course, he
ends in A after starting there. The “map”
below indicates the possible routes that he
could take, where the numbers represent
distances. Find the optimal route that he
should follow to minimize the total distance
that he travels.

B
3 3
3
4
5
4 C E
A

6 2 4

130
 Solution

stages: cities (first three stops)

states: visited cities in order
decision variable: next city
returns: distance to next destination

Stage 1
s1 d1 r1 f1(s1)

A, B, C E* 15 15
A, B, D E* 13 13
A, B, E D* 10 10
C 13 13
A, C, B E* 13 13
A, C, D E* 10 10
A, C, E B 13 13
D* 11 11
A, D, B E* 12 12
A, D, C E* 11 11
A, D, E B* 10 10
C 11 11

131
 Stage 2

s2 d2 s1 = s2  d2 r2 + f1(s1) f2(s2)

A, B C A,B,C 3+15 =18

D A,B,D 4+13 =17
E* A,B,E 3+10 =13 13

A, C B A,C,B 3+13 =16

D* A,C,D 2+10 =12 12
E A,C,E 5+11 =16

A, D B A,D,B 4+12 =16

C* A,D,C 2+11 =13 13
E A,D,E 4+10 =14

Stage 3

s3 d3 s2= s3  d3 r3 + f2(s2) f3(s3)

A B* A,B 3 + 13 =16 16
C* A,C 4 + 12 =16 16
D A,D 6 + 13 =19

Answer:

ABEDCA
ACDEBA

132
 MULTIPLE OBJECTIVE
DECISION MAKING

GOAL PROGRAMMING
• Involves satisficing.
• Goals are ranked by order of importance.
• In preemptive goal programming, higher
priority goal is assumed to be infinitely more
important than a lower priority goal.
• Goal programming achieves as many higher-
priority goals as possible, then attempts to get
as close as possible to satisfying the
remaining goals.

133
PRODUCT-MIX EXAMPLE
Faze Linear Company is a small manufacturer of
high-fidelity components. It has facilities to
produce only power amps, only preamps, or a
combination of both. Due to limited resources, it
is critical to produce appropriate quantities of
power amps and/or preamps to maximize profit.

selling price profit

per unit per unit
power amps $799.95 $200

preamps $1000 $500

134
 microchips assembly time inspection time

resources 40 units 240 hours 81 hours

transformation process

final products x1 x2

power amps($200) preamps($500)

• Because of a shortage of high quality

microchips, at most 40 items can be
manufactured on a daily basis.
• All other electronic components are in
adequate supply.
• Only 240 hours of assembly worker time is
available each day.
• Each power amp requires 1.2 hours of
assembly time and preamp requires 4 hours.
• 81 hours are available for testing and
inspection each day.
• Power amps require 0.5 hours and preamps
require 1 hour of inspection.

135
• Since power amps do not require a microchip,
the limited availability of microchips will directly
affect only the number of preamps produced
each day. So we have x2 ≤ 40.
• Both components require assembly time.
• The combined assembly time must not exceed
240 hours. So we have 1.2x1+4x2 ≤ 240.
• For inspection and testing time, the constraint
is 0.5x1+x2 ≤ 81. Also, there are the
nonnegativity conditions x1, x2 ≥ 0. (We ignore
integer requirements here.)

LP TO MAXIMIZE PROFIT

Maximize 200x1 + 500x2

s.t.
x2 ≤ 40
1.2x1 + 4x2 ≤ 240
0.5x1 + x2 ≤ 81
x 1, x 2 ≥ 0

136
 TWO-GOAL MODEL
• Assume that the Faze Linear management would
like to achieve two goals:
priority 1: achieve profit of $40,000/day.
priority 2: limit overtime of inspectors.

• For the higher-priority goal, introduce two

deviational variables:
d1- = amount by which the target profit of
$40,000 is underachieved.
d1+ = amount by which the target profit of
$40,000 is overachieved.

• The profit goal as a constraint can now be

written as 200x1+500x2+ d1- - d1+ = 40,000.

• Notice that the $40,000 profit goal need not be

met exactly. If d1- has a positive value, we will
fall short of 40,000 and if d1+ has a positive
value, we will exceed 40,000.

• By the simplex method, d1- and d1+ cannot both

be positive at the same time.

137
• Assuming that the 81 hours of inspection time is
regular time (not any overtime), we formulate the
second goal by defining the deviational variables.
d2- = amount by which inspection time is
underutilized (ie., short of 81 hours)
d2+ = amount by which inspection time is
overutilized (ie., more than 81 hours)
• Formulate the overtime goal as
0.5x1 + x2 + d2- - d2+ = 81.
• Incorporating these goals as constraints in the LP
formulation, we obtain the goal programming
formulation.

Minimize P1 d1- + P2 d2+

s.t.
 x 2  40 microchips 
  resource constraints
1.2 x1  4 x 2  240 assembly 

200 x1  500 x 2  d 1  d 1  40,000

 

  goal constraints
0 .5 x1  x 2  d 2

 d 2

 81 

x1, x 2, d 1 , d 1 , d 2  , d 2   0

138
 • The P1 and P2 symbols in the objective
function reflect the fact that d1- and d2+
represent priority 1 and 2 goals,
respectively.
• Priorities of higher rank are infinitely
more important than are those of a lower
rank, i.e. P1>>P2
• In giving values to P1 and P2, take into
account the relative size of the units of
the variables.

• The deviational variables are to be minimized

as they represent the amount by which the
goals are not satisfied.

• The result of the ordinal priority rankings is

that the profit goal will be achieved to the
greatest extent possible before the overtime
goal is considered.

• If satisfying any part of the overtime goal

causes any reduction in the higher-ranking
profit goal, the overtime goal will not be
satisfied at all.

139
 HOMEWORK 31
Acme Appliance must determine how many
washers and dryers should stocked. It costs Acme
$350 to purchase a washer and $250 to purchase
a dryer. A washer requires 3 sq. yd of storage
space, and a dryer requires 3.5 sq. yd. The sale of
a washer earns Highland a profit of $200, and the
sale of a dryer $150. Acme has set the following
goals (listed in order of importance):

• Goal 1: Highland should earn at least

$30,000 in profits from the sale of washers and
dryers.
• Goal 2: Washers and dryers should not use
up more than 400 sq. yd. of storage space.
Formulate a preemptive goal programming
model for Acme to determine how many
washers and dryers to order. Use D for the
number of dryers and W for the number of
washers as your variables. There should be
only goal constraints in your formulation for
this particular problem.

140
 Pareto Optimality
• Example 1. You have two objective functions given as
elements in a two-tuple and a decision involving two
possible actions. For each action, consider the following
returns.
Return for actions 1 and 2
Action 1 (1, 7)
Action 2 (2, 8)

Obviously, action 2 is better because the return for both

objectives is better than those for action 1. Action 1 is
said to dominate action 2, so you choose action 1 as
your decision.

Pareto Optimality
• Example 2: Again you have two objective functions and
a decision involving two possible actions. For each
action, consider the following returns.
Return for actions 1 and 2
Action 1 (8, 2)
Action 2 (9, 1)

This time action 1 is not obviously better than action 2

and vice versa. Neither dominates the other in both
returns, so both are said to be Pareto optimal.

141
 Pareto Optimality
Optimization Formulation

Pareto Optimality – Mathematical

Definition

142
 HOMEWORK 32
Consider the Pareto optimization problem

Vmax (2x2 - 3y2, 25-5y)

x,y
s.t.
0 < x < 10
0 < y < 5,

where the first objective function f1(x,y) = 2x2 - 3y2 represents

profit in hundreds of dollars employee per day and the second
objective function f2(x,y) = 25-5y represents happiness in the
average number of smiles per employee per work day. Is (10,5)
an efficient point for this problem? Is (0,0)? Show your work.

Pareto Optimality – Example 1

This point is not Pareto
optimal so it is not on
the efficient frontier..

The efficient frontier of two objective functions can be seen graphically, where
the efficient frontier is the set of the objective function values over the feasible
region that are not dominated and hence are Pareto optimal two-tuples.

143
 Pareto Optimality – Example 2
Example: Profit Pollution Trade-Off Curve

Problem Statement
Chemco is considering producing three products. The per-unit contribution to
profit, labor requirements, raw material used per unit produced, and pollution
produced per unit of product are given below. Currently, 1300 labor hours and
1000 units of raw material are available. Chemco’s two objectives are to
maximize profit and minimize pollution produced. Graph the trade-off curve for
this problem.
Product
1 2 3
Profit ($) 10 9 8
Labor (hours) 4 3 2
Raw material (units) 3 2 2
Pollution (units) 10 6 3

Pareto Optimality – Example 2

Example: Profit Pollution Trade-Off Curve

Problem Statement
Chemco is considering producing three products. The per-unit contribution to
profit, labor requirements, raw material used per unit produced, and pollution
produced per unit of product are given below. Currently, 1300 labor hours and
1000 units of raw material are available. Chemco’s two objectives are to
maximize profit and minimize pollution produced. Graph the trade-off curve for
this problem.
Product
1 2 3
Profit ($) 10 9 8
Labor (hours) 4 3 2
Raw material (units) 3 2 2
Pollution (units) 10 6 3

144
Pareto Optimality – Example 2

Pareto Optimality – Example 2

145
Pareto Optimality – Example 2

Pareto Optimality – Example 2

146
REVIEW FOR QUIZ 3

147