fullSlideDecisionMaking PDF

ECE 307 – Techniques for
Engineering Decisions
Engineering Decision Making Techniques
Topic 1: Course Overview
George Gross
Department of Electrical and Computer Engineering
University of Illinois at Urbana-Champaign
ECE 307 © 2005, 2006 George Gross, University of Illinois at Urbana-Champaign, All Rights Reserved.
SCOPE OF COURSE
The course covers techniques that are useful

when combined with the appropriate technical
knowledge, for making engineering/economic
decisions
Such decisions are typical of those made by
business firms, government-owned enterprises
and agencies and individuals
We focus on the systematic evaluation of
alternatives before a decision is made
regarding a particular problem
1
EXAMPLES OF DECISION MAKING
PROBLEMS
Introduction of a new product
Expansion of production facilities/warehousing
Adoption of new technology
Implementation of a new production schedule
Changes in the production mix
Risk management in purchase/sale activities
Optimal scheduling of processes/projects

BASIC THRUSTS
Development of the analytical framework for

decision making on a sound and systematic basis
with the goal to enable the decision maker to
undertake an appropriate analysis and systematic
evaluation of various alternatives
Provide training for engineers to play an
increasingly more prominent role in the decision
making processes in their work environment
2
THE UNDERLYING BASIS
Decisions are made by selecting from possible

alternatives with reference to the future which is
inherently uncertain
A common basis is set up by formulating the
decisions in economic terms
A key aspect is the assumptions introduced to
enable the undertaking of the analysis and the
evaluation of alternatives
PRODUCT MIX OPTIMIZATION

PROBLEM
A factory manufactures three different products
requiring various levels of resources and
providing different benefits (profits)
The constraints on resources are given
Problem: determine the optimal daily mix, i.e.,
the production schedule that maximizes profits
without violating any constraints

3
PROBLEM
product A B C limit
resources required
labor (h) 1 1 1 100

unit of product
material (lb) 10 4 5 600

per
A&G (h) 2 2 6 300
profits per unit of product ($) 10 6 4

PROBLEM
We formulate the decision problem by
introducing the decision variables:
xi = daily production level of product i, i = A, B, C
We construct a programming problem for the
schedule by expressing
the objective function
the constraints
the common sense requirements
in mathematical terms
4
PROBLEM
max Z = 10 x A + 6xB + 4xC objective
xA + xB + xC ≤ 100 labor
⎫
≤ 600 material ⎪
constraints
10 x A + 4 xB + 5 xC
⎪
⎬
2 xA + 2 xB + 6 xC ≤ 300 A & G ⎪
x A , x B , xC
⎪
≥ 0 reality check ⎭

PROBLEM
The optimal solution is
x A* = 33.33 x B* = 66.67 xC* = 0
corresponding to maximum profits

Z * = $ 733.33
The shadow prices corresponding to the
constraints give the change in profits for
additional resources:
labor : $ 3.33 material : $ 0.67 A&G : $ 0
5
PROBLEM
We next examine a sensitivity case correspon-
ding to the use of overtime labor
We are interested in determining how many
overtime hours would be profitable to schedule
without impacting on the optimal product mix
20 hours of labor overtime increases profits
by (20) (3.33) = $ 66.6
as long as the cost of overtime labor does
not exceed $ 66.6 it is worthwhile to use it
the optimal product mix remains unchanged:
since we only produce products A and B and
no product C
OPTIMAL TRAJECTORY PLANNING

Mr. Jones has to travel from a fixed starting
point to a fixed destination with a choice of
which intermediate points to go through
2 5
8
1 3 6 10
9
4 7
stage 1 stage 2 stage 3 stage 4

6
The relative “costs” for the various possible

paths are given by
5 6 7 8 9
10
2 3 4 2 7 4 6 5 1 4
8 3
1 2 4 3 3 3 2 4 6 6 3
9 4
4 4 1 5 7 3 3
The problem is to select the route that

minimizes the total costs of the trip
Solution approaches:
enumerate all possibilities: this is, in general,
too time consuming since we need to
consider
3 × 3 × 2 = 18
different routes for this simple case
select the best for each successive stage:
myopic decision making solution leads to the
path
1 2 6 9 10
with costs of 13
7
use some heuristic approach which allows to

sacrifice a little at one stage in the hope of
attaining savings thereafter: for example, the
path 1 4 6 has costs of 4 which are less
than those of the path 1 2 6, which are 6
The optimal route is
1 3 5 8 10
with costs of 11
There are two additional routes whose costs are
11:
1 4 5 8 10
1 4 6 9 10
Thus, this problem does not result in a unique
optimum
8
BUSING PROBLEM
Three school districts in Busville have a
distribution of Caucasians (C ) and African
Americans ( A) as shown in the table
number of students
district
C A
1 210 120
2 210 30
3 180 150
BUSING PROBLEM
Implementation of the Supreme Court ruling on

racial balance requires that each of the three
districts have exactly 300 students with identical
racial make-up, i.e., that
⎛ A⎞ ⎛ A⎞ ⎛ A⎞
⎜C⎟ = ⎜C⎟ = ⎜C⎟
⎝ ⎠1 ⎝ ⎠2 ⎝ ⎠3
and the only means of attaining the racial
balance goal is through busing
9
BUSING PROBLEM
Given the distances between the districts,

determine the total minimum distance that
students must be bussed to satisfy the social
balance requirements
number of students distance to district

district
C A 2 3
1 210 120 3 5
2 210 30 — 4
3 180 150 4 —
THE ENVELOPE QUESTION
On a television game show, the host subjects

contestants to unusual tests of mental skill
On one, a contestant may choose one of two
identical envelopes – labeled A and B – each of
which contains an unknown amount of money
The host reveals, though, that one envelope
contains twice as much money as the other
After choosing A, the host suggests that the
contestant might want to switch; the host states:
10
“Switching is clearly advantageous. Suppose

you have amount x in your envelope A. Then B
must contain either x/2 (with probability 0.5). In
fact, now that I think about it, I’ll only let you
switch if you give me a 10% cut of your
winnings. What do you say? You’ll still be
ahead.”
The contestant replies,
“No deal. But I’ll be happy to switch for free. In
fact, I’ll even let you choose which envelope I
get. I won’t even charge you anything!”
Who is right?
The host is proposing a decision tree that looks
like this:
keep x
0.5 x
switch
2
0.5 2x
which does not correctly represent the situation

11
Rather, we have for the two envelopes A and B

A has x (0.5)
x
keep A B has x/2 (0.5) x
contestant
2
payoff
A has x/2 (0.5) x
switch to B
2
B has x (0.5)
x
The two decision branches are identical from the
view of the decision maker
DECISION ANALYSIS PROTOTYPE

EXAMPLE
The Greazy Company owns a tract of land that
may contain oil; the report of a consulting
geologist indicates that there is one chance in
four that oil exists
Because of this prospect, another oil company
has offered to purchase the land for $ 90,000 but
Greazy is considering holding the land in order
to drill for oil itself: if oil is found, the profits are
expected to be $ 700,000 but if land is dry, the
losses are expected to be $ 100,000
12
EXAMPLE
payoff ($)
decision
alternative
land has oil land is dry
drill for oil 700,000 (100,000)
sell the land 90,000 90,000
probability 0.25 0.75

EXAMPLE
Evaluation of the two alternative actions
action expected payoff (k$)
1 0.25 (700) + 0.75 (-100) = 100
2 0.25 (90) + 0.75 (90) = 90
and the better choice is to drill for oil

The decision is strongly dependent on how good
is the knowledge of the probabilities
13
EXAMPLE
Sometimes it is possible to undertake further
work before a decision is taken; for example, an
available option before making a decision is to
conduct a detailed seismic survey with costs of
$ 30,000 to obtain a better estimate of the
probability of oil
We construct a decision tree to
visually display the problem
organize systematically the computation

EXAMPLE
oil 670
ill
dr
f
dry -130
le
ab c 60
or sel l
payoff in k$
a v
f
un oil 670
rve ic
l
su eism
b
il
dr
g
-130
y
fa v dry
ora
s
ble d
sel l 60
do
a oil 700
no il l h
s
su eism dr dry -100
rv
e y ic
e
sel l 90
14
A GENERAL FRAMEWORK FOR DECISION
MAKING UNDER UNCERTAINTY
The decision maker must select an action from a

set of possible actions – the set of feasible
alternatives
The underlying premise is that the choice of
action is made under uncertainty because the
outcome will be affected by random factors
outside the control of the decision maker; this
necessitates a classification of the possible
states of nature

For each contribution of an action and state of
nature, the value to the decision maker of the
consequences of an outcome is established and
quantified in terms of the payoff
The payoff is defined as the quantity measure of
the value to the decision maker of the
consequences of an outcome
15
The payoffs are used to select the optimal action

for the decision maker according to some
selected criterion
Example: Bayes’ decision rule involves the use
of the best available estimates of the
probabilities of the states of nature to calculate
the expected value of the payoff for each
possible action and then to choose the action
with the maximum expected payoff
DECISION ANALYSIS
We study decision analysis since its application

can lead to better decisions
We need to differentiate between good decisions
and lucky outcomes
Every decision may have a lucky outcome or an
unlucky outcome
A good decision is one that gives the best
outcome
The goal is to make effective decisions more
consistently
16
ECE 307
Prerequisite : ECE 210

Corequisite : ECE 413
Texts
A. Ravindran, D. T. Phillips and J. J. Solberg,
"Operations Research: Principles and
Practice," J. Wiley, New York, 1992
R. T. Clemen, “Making Hard Decision: An
Introduction to Decision Analysis,” Duxbury
Press/Wadsworth Publishing Company, 1995
Course Website:
http://energy.ece.uiuc.edu/gross/index_files/Page328.htm
GRADING POLICY
Two exams: midterm and final

Two team project presentations
Grade Breakdown
component percentage
Homework 15
Projects 10
Midterm 25
Final 50
Total 100
17
ECE 307 TOPICAL OUTLINE
Introduction: nature of engineering decisions;
structuring of decisions; role of models;
interplay of economics and technical/engine-
ering considerations; decision making under
certainty and uncertainty; good decisions vs.
good outcomes; tools
Resource allocation decision making using the
linear programming framework: problem
formulation; basic approach; duality; economic
interpretation; sensitivity analysis; interpretation
of results

Scheduling and assignment decisions using
network flow concepts: transshipment problem
formulation and solution; application to
matching decisions; network optimization;
scheduling applications
Sequential decision making in a dynamic
programming framework: nature of dynamic
programming approach; problem formulation;
solution procedures; key limitations
Probability theory: random variables; probability
distributions; expectation; conditional
probability; moments; convolution
18
Statistical concepts: data analysis; statistical

measures; estimation
Application of probabilistic concepts to the
modeling of uncertainty in decision making:
modeling of the impacts of uncertainty;
applications to siting, investment and price
volatility problems
Decision making under uncertainty: decision
trees; value of information; uses of data;
sensitivity analysis and statistics
Case studies and presentations
19
Topic 2: Introduction to Linear Programming
George Gross
OUTLINE
The nature of a programming or optimization
problem
The salient characteristics of a linear
programming (LP) problem
The LP problem formulation
The LP problem solution
Extensive illustrations with numerical examples

1
EXAMPLE 1: HIGH/LOW HEEL SHOE
CHOICE PROBLEM
You are headed to a party and are trying to find a
pair of shoes to wear; you choice is narrowed
down to two candidates:
a high heel pair; and
a low heel pair
The high heel shoes look more beautiful but are
not as comfortable as the competing pair
Which pair should you choose?
MODEL FORMULATION
You first quantify your assessment along the two

dimensions of looks and comfort and construct
assessment
maximum weight
aspect high low
value (%)
heels heels
esthetics 5.0 4.2 3.6 70
comfort 5.0 3.5 4.8 30
Next you represent your decision in terms of two
decision variables:
2
MODEL FORMULATION
⎧1 choose high ⎧1 choose low

xH = ⎨ xL = ⎨
⎩ 0 otherwise ⎩0 otherwise
Formulate your objectives to maximize the
weighted assessment as
max {70% * esthetics + 30% * comfort}
Use the defined variables to state the objective
max Z = x H ( 4.2 × 0.7 + 3.5 × 0.3 ) + x L ( 3.6 × 0.7 + 4.8 × 0.3 )

MODEL FORMULATION
Next consider the problem constraints:
only one pair of shoes can be selected
the decision variables are nonnegative
State the constraints in terms of x H and x L :
xH + xL = 1
xH ≥ 0 , xL ≥ 0
3
PROBLEM STATEMENT SUMMARY
Decision variables:
⎧1 choose high ⎧1 choose low

xH = ⎨ xL = ⎨
⎩0 otherwise ⎩0 otherwise
Objective function:
max Z = 3.99 x H + 3.96 x L
Constraints:
xH + xL = 1
xH ≥ 0, xL ≥ 0
OPTIMAL SOLUTION
We determine the values x *H and x *L which

result on the value of Z * such that
Z * = Z ( x *H , x *L ) ≥ Z ( x H , x L )
for all feasible ( x H , x L )

We call such a solution an optimal solution
A feasible solution is one that satisfies all the
constraints
*
The optimal solution, denoted by , is chosen
from all the feasible solutions
4
SOLUTION APPROACH: EXHAUSTIVE
SEARCH
We enumerate all the possible solutions: in this
problem there are only two choices:
⎧ xH = 1 ⎧ xH = 0
A ⎨ B ⎨
⎩ xL = 0 ⎩ xL = 1
We evaluate Z for A and B and compare
Z A = 3.99 Z B = 3.96
so that Z A > Z B and so A is the optimal choice
The optimal solution is
x *
H = 1 , x *
L = 0 and Z * = 3.99
CHARACTERISTICS OF A PROGRAMMING/
OPTIMIZATION PROBLEM
Objective is to make a decision among various
alternatives and therefore requires the definition
of the decision variables
The solution of the “best” decision is made
according to some objective and requires the
formulation of the objective function
The decision must satisfy certain specified
constraints and so requires the mathematical
statement of the problem constraints
5
CLASSIFICATION OF PROGRAMMING
PROBLEMS
The problem statement is characterized by :
continuous valued
decision variables
integer valued
linear
objective function
non linear
linear
constraints
non linear
PROGRAMMING PROBLEM CLASSES
Linear/nonlinear programming
Static/dynamic programming
Integer programming
Mixed programming
6
EXAMPLE 2: CONDUCTOR PROBLEM
A company is producing two types of conductors
for EHV transmission lines
production
metal needed profits
type conductor capacity
(tons/unit) ($/unit)
(unit/day)
1 ACSR 84/19 4 1/6 3
2 ACSR 18/7 6 1/9 5
The supply department can provide daily up to 1

ton of metal
We schedule the production so as to maximize
the profits of the company
PROBLEM ANALYSIS
Determination of the objective: to maximize the

profits of the company
Means of attaining this objective: decision of
how many units of product 1 and/or product 2 to
produce each day
Consideration of the constraints: the daily
production capacity limits, the daily metal supply
limit and common sense requirements
7
MODEL CONSTRUCTION
We define the decision variables to be
x1 = number of type 1 units produced per day

x2 = number of type 2 units produced per day
We define the objective to be
Z = profits ($/day)
= 3 x1 + 5 x2
Sanity check for units of the objective function

($/day) = ($/unit) • (unit/day)
PROBLEM STATEMENT
Objective function:
max Z = 3 x 1 + 5 x 2
Constraints:
capacity limits:
x1 ≤ 4 x2 ≤ 6
metal supply limit:
x1 x2
+ ≤ 1
6 9
common sense requirements:
x 1 ≥ 0, x 2 ≥ 0
8
PROBLEM STATEMENT
max Z = 3x1 + 5x 2
s .t .
x1 ≤ 4
x2 ≤ 6
x1 x
+ 2 ≤ 1
6 9
x1 ≥ 0 , x2 ≥ 0
CONSTRUCTION OF THE FEASIBLE

REGION
x2 x1 = 4
( 4,0 )
( 0, 0 ) x1
x 1 ≤ 4, x 1 ≥ 0 , x 2 ≥ 0
9
REGION
x2
x2 = 6
( 0 , 6)
x 2 ≤ 6, x 2 ≥ 0 , x 1 ≥ 0
( 0, 0 ) x1

REGION
x2
( 0,9) x1 x2
+ ≤1
6 9
( 6,0)
( 0,0)
x1
10
THE FEASIBLE REGION
x1 = 4
x2
( 2, 6) x2 = 6
( 0, 6)
feasible
region
( 4, 3)
x1 x2
+ = 1
6 9
( 0, 0 ) x1
( 4,0 )
FEASIBLE SOLUTIONS
x2 x1 = 4
( 2, 6)
x2 = 6
( 0, 6)
D max Z = 3 x 1 + 5 x 2
(1.5, 4.5)
Z = 27 ( 4, 3)
° ( 2, 2 ) x1 x2
Z = 16 + = 1
6 9
( 0, 0 ) ( 4,0 ) x1
11
CONTOURS OF CONSTANT Z
x2 x1 = 4
( 0 , 6) ( 2, 6)
x2 = 6
Z = 25
Z = 20 max Z = 3 x1 + 5 x2
Z = 15
Z = 36
Z = 10 ( 4, 3)
x1 x2
( 0 , 2) + = 1
6 9
( 0, 0 ) ( 4,0 ) x1
OPTIMAL SOLUTION
We can graphically determine the optimal
solution
The optimal solution of this problem is:
x *1 = 2 and x *2 = 6
The objective value at the optimal solution is
Z * = 3 x *1 + 5 x *2 = 36
12
LINEAR PROGRAMMING (LP)
PROBLEM
A linear programming problem is an
optimization problem with a linear
objective function and linear constraints.
EXAMPLE 3: ONE-POTATO, TWO-

POTATO PROBLEM
Mr. Spud manages the Potatoes-R-Us Co. which
processes potatoes into packages of freedom
fries ( F ), hash browns ( H ) and chips ( C )
Mr. Spud can buy potatoes from two sources;
each source has distinct characteristics
The problem is to determine the respective
quantities Mr. Spud should buy from source 1
and from source 2 to maximize profits
13
EXAMPLE 3: ONE-POTATO, TWO-
POTATO PROBLEM
The known data are summarized in the table
source 1 source 2
product sales limit (tons)
uses uses
F 20% 30% 1.8
H 20% 10% 1.2
C 30% 30% 2.4
profits ($/ton) 5 6
The following assumptions hold:

30% waste for each source
production may not exceed sales limit
ANALYSIS
Decision variables:
x1 = quantity purchased from source 1
x2 = quantity purchased from source 2
Objective function:
max Z = 5 x1 + 6 x2
Constraints:
0.2 x1 + 0.3 x2 ≤ 1.8 (F)
0.2 x1 + 0.1 x2 ≤ 1.2 (H) x1 ≥ 0 , x2 ≥ 0
0.3 x1 + 0.3 x2 ≤ 2.4 (C)
14
FEASIBLE REGION DETERMINATION
x2 x2
6 freedom fries 12
F 10
4
8
hash browns
2
H
x1 6
0 2 4 6 8
x2
4
8
chips 2
6
C 0 2 4 6 8 x1
4
2
0 2 4 6 8 x1
THE FEASIBLE REGION
x2
12
10
8 feasible
region
6
4
2
x1
2 4 6 8 10 12
15
EXAMPLE 3: CONTOURS OF
CONSTANT Z
Z = 36 x2
max Z = 5 x 1 + 6 x 2
Z = 30
6
Z = 24 Z = 40.5
5
Z = 18
4
3 ( 4.5, 3)
2
1
x1
1 2 3 4 5 6
THE OPTIMAL SOLUTION
The optimal solution of this problem is:
x *1 = 4.5 x *2 = 3
The objective value at the optimal solution is:
Z * = 5 x *1 + 6 x *2 = 40.5
16
OBSERVATIONS
Constant Z lines are parallel and change

monotonically along the normal direction to the
contours of constant values of Z
An optimal solution must be at one of the corner
points of the feasible region: fortuitously, there
are only a finite number of corner points
If a particular corner point gives a better solution
(in terms of the objective function) than the ones
at all its adjacent corner points, then, it is an
optimal solution
SOLUTION PROCEDURE: SIMPLEX

METHOD
Initialization step: start at a corner point
Iteration step: move to a better adjacent corner
point and repeat this step as many times as
needed
Stopping rule: stop when the corner point
solution is better than those at all its adjacent
corner points
17
EXAMPLE 3: THE SIMPLEX
APPROACH
x2
Z = 36 max Z = 5 x1 + 6 x2
( 0 , 6) 6
5
Z = 40.5
4
3
( 4.5, 3)
2
1 Z = 30
Z = 0 ( 6,0 )
( 0, 0 ) 0
1 2 3 4 5 6
x1
EXAMPLE 3 : APPLICATION OF THE

SIMPLEX METHOD
step x1 x2 Z
0 0 0 0
1 0 6 36
2 4.5 3 40.5
3 6 0 30
18
EXAMPLE 3 : THE SIMPLEX METHOD
Z = 36 x
2
(0,6) max Z = 5 x1 + 6 x 2
6
5
4
Z = 40.5
3 (4.5,3)
1 Z = 30
Z=0 (6,0)
(0,0) 0 x1
1 2 3 4 5 6
EXAMPLE 3: THE SIMPLEX METHOD
1. Start at (0,0) with Z(0,0) = 0

2. (i) Move from (0,0) to (0,6), Z(0,6) = 36
(ii) Move from (0,6) to (4.5,3) and evaluate
Z(4.5,3) = 40.5
3. Compare the objective at (4.5,3) to values at (6,0)
and at (6,0):
Z(4.5,3) ≥ Z(6,0)
Z(4.5,3) ≥ Z(6,0)
therefore, (4.5,3) is optimal
19
REVIEW
Key requirements of a programming problem:
to make a decision and so to define decision
variables
to achieve some objective and so to formulate
an objective function
to ensure that the decision satisfies certain
constraints which are mathematically stated

REVIEW
Key attributes of an LP
objective function is linear
constraints are linear
Basic steps in formulating a programming
problem
definition of decision variables
statement of objective function
formulation of constraints
20
REVIEW
Words of caution: care with units and attention to

not ignoring the implicit constraints such as
nonnegativity and common sense requirements
to an LP
Graphical solution approach
feasible region determination
contours of constant Z
identification of the vertex with optimal Z *
EXAMPLE 4: INSPECTION OF GOODS

PRODUCED
There are 8 grade 1 and 10 grade 2 inspectors
available for QC inspection; at least 1800
pieces must be inspected in each 8-hour day
Problem data are summarized below:
grade speed accuracy wages

level (unit/hr) (%) ($/h)
1 25 98 4
2 15 95 3
21
EXAMPLE 4: INSPECTION OF GOODS
PRODUCED
Each error costs $ 2
The problem is to determine the optimal
assignment of inspectors, i.e., the number of
inspectors of grade 1 and that of grade 2, respe-
ctively, to result in the least-cost inspection effort

EXAMPLE 4: FORMULATION
Definition of decision variables:
x1 = number of grade 1 inspectors assigned
Objective function
optimal assignment ⇒ minimum costs
costs = wages + errors

22
• each grade 1 inspector costs:
4 + 2 (25)(0.02) = 5 $/hr
• each grade 2 inspector costs:
3 + 2 (15)(0.05) = 4.5 $/hr
• total daily inspection costs in $ are
Z = 8 [5 x1 + 4.5 x2] = 40 x1 + 36 x2 ($)

Constraints:
job completion:
8(25) x1 + 8(15) x2 ≥ 1800
⇔ 200 x1 + 120 x2 ≥ 1800
⇔ 5 x1 + 3 x2 ≥ 45
availability limit:
x1 ≤ 8
x2 ≤ 10
nonnegativity:
x1 ≥ 0 , x2 ≥ 0
23
EXAMPLE 4: PROBLEM STATEMENT
SUMMARY
Decision variables:
Objective function:
min Z = 40 x1 + 36 x2
Constraints:
5 x1 + 3 x2 ≥ 45
x1 ≤ 8
x2 ≤ 10
x1 ≥ 0 , x2 ≥ 0
MULTI – PERIOD SCHEDULING
More than one period is involved
The result of each period affects the initial
conditions for the next period and therefore the
solution
We need to define variables to take into account
the initial conditions in addition to the decision
variables of the problem

24
EXAMPLE 5: HYDROELECTRIC
POWER SYSTEM OPERATIONS
We consider a single operator of a system
consisting of two water reservoirs with a
hydroelectric plant attached to each reservoir
We schedule the two power plant operations over
a two–period horizon
We are interested in a plan to maximize the total
revenues of the system operator

EXAMPLE 5: HYDROELECTRIC
POWER SYSTEM OPERATIONS
res A wA wA wB wB
res A plant res B plant
inflow A B
sA sB
res B
inflow
25
EXAMPLE 5: RESERVOIR DATA IN kAf
parameter reservoir A reservoir B

maximum capacity 2000 1500
predicted inflow in
200 40
period 1
predicted inflow in
130 15
period 2
minimum allowable
1200 800
level
level at start of period 1 1900 850
EXAMPLE 5: SYSTEM
CHARACTERISTICS
plant A 1 kAf 400 MWh

plant A
plant B 1 kAf 200 MWh

plant B
reservoir max kAf for generation per period
A 150
B 87.5
26
EXAMPLE 5: SYSTEM
CHARACTERISTICS
Demand in MWh ( for each period )
up to 50,000 MWh can be sold @ $ 20/MWh
all additional MWh are sold @ $ 14/MWh
$/MWh
non-linear objective
function
14 $/MWh
20
$/MWh
xH xL MWh
50,000
EXAMPLE 5: DECISION VARIABLES
variable quantity denoted units

x Hi energy sold at 20 $/MWh MWh
x Li energy sold at 14 $/MWh MWh
w iA plant A water supply for generation kAf
w Bi plant B water supply for generation kAf
s Ai reservoir A spill kAf
i
s B reservoir B spill kAf
i
rA
reservoir A end of period i level kAf
i
rB reservoir B end of period i level kAf
superscript i denotes period i, i = 1, 2

27
EXAMPLE 5: OBJECTIVE FUNCTION
maximize total revenues from sales
max Z = 20( x 1H + x H2 ) + 14( x 1L + x L2 )
4 of the 16 decision variables

2 for each period
units are in $
EXAMPLE 5 : CONSTRAINTS
Period 1
energy conservation in a lossless system
• total generation 400 w 1A + 200 w 1B ( MWh )
• total sales x1H + x 1L ( MWh )
• losses are negelected and so
x 1H + x 1L = 400 w 1A + 200 w 1B
maximum available capacity limit
w 1A ≤ 150
wB1 ≤ 87.5
28
conservation of flow relations for each
reservoir
• reservoir A:
w A1 + s A1 + rA1 = 1900 + 200 = 2100 ( kAf )
res. Res.level
level at at res. Res.level
level at at predicted
Res.level at
e.o.p.
e.o.p.10
e.o.p.
e.o.p.00 inflow
e.o.p. 0
• reservoir B:
w B1 + sB1 + rB1 = 850 + 400 + w A1 + s A1 ( kAf )
limitation on reservoir variables
• reservoir A:
1200 ≤ r A1 ≤ 2000 ( kAf )

• reservoir B:
800 ≤ r B1 ≤ 1500 ( kAf )

sales constraint
x H1 ≤ 50,000 ( kAf )
29
Period 2
energy conservation in a lossless system
• total generation 400 w A2 + 200 w B2 ( MWh )
• total sales x H2 + x L2 ( MWh )
• losses are negelected and so
x H2 + x L2 = 400 w A2 + 200 w B2
maximum available capacity limit
w A2 ≤ 150
wB2 ≤ 87.5
conservation of flow relations for each

reservoir
• reservoir A:
w A2 + s A2 + rA2 = rA1 + 130 ( kAf )
res. Res.level
level at at res. level atat
Res.level Res.level at
predicted
e.o.p.20 e.o.p. 0 e.o.p. 0
e.o.p. e.o.p. 1 inflow
• reservoir B:
w B2 + sB2 + r B2 = r B1 + 15 + w A2 + s A2 ( kAf )
30
limitation on reservoir variables
• reservoir A:
1200 ≤ rA2 ≤ 2000 ( kAf )

• reservoir B:
800 ≤ rB2 ≤ 1500 ( kAf )

sales constraint
x H2 ≤ 50,000 ( kAf )
EXAMPLE 5 : PROBLEM STATEMENT
16 decision variables:
x Hi , x Li , w Ai , w Bi , s Ai , sBi , r Ai , r Bi , i = 1, 2
Objective function:
max Z = 20( x 1H + x H2 ) + 14( x 1L + x L2 )
Constraints:
20 constraints for the periods 1 and 2
nonnegativity constraints on all variables

31
EXAMPLE 6 : DISHWASHER AND
WASHING MACHINE PROBLEM
The Appliance Co. manufactures dishwashers
and washing machines
The sales targets for next four quarters are:
quarter t
product variable
1 2 3 4
dishwasher Dt 2000 1300 3000 1000
washing
Wt 1200 1500 1000 1400
machine
EXAMPLE 6: QUARTERLY COST

COMPONENTS
quarter t
cost component parameter unit costs ($)
1 2 3 4
dishwasher ct 125 130 125 126
manufacturing
($/unit) washing
vt 90 100 95 95
machine
dishwasher jt 5.0 4.5 4.5 4.0
storage
($/unit) washing
kt 4.3 3.8 3.8 3.3
machine
hourly labor ($ /hour) pt 6.0 6.0 6.8 6.8
32
EXAMPLE 6: CONSTRAINTS
Each dishwasher uses 1.5 hours and each

washing machine uses 2 hours of labor
The labor hours in each quarter cannot grow or
decrease by more than 10%; there were 5000
labor hours in the quarter preceding the first
quarter
At the start of the first quarter, there are 750 dish-
washers and 50 washing machines in storage
EXAMPLE 6: PROBLEM AIM
How to schedule the production in each of the
four quarters so as to minimize the costs while
meeting the sales targets?
33
EXAMPLE 6: QUARTER t DECISION
VARIABLES
symbol variable
dt number of dishwashers produced
wt number of washing machines produced
rt final inventory of dishwashers
st final inventory of washing machines
ht available labor hours during Qt
t = 1, 2, 3, 4
EXAMPLE 6: OBJECTIVE FUNCTION
minimize the total costs for the four quarters
manufacturing storage labor

costs costs costs
min Z = c1d 1 + v1 w1 + j1r1 + k1 s1 + p1h1 quarter 1

+ c2 d 2 + v 2 w 2 + j2 r2 + k 2 s2 + p2 h2 quarter 2
+ c3d 3 + v 3 w3 + j3 r3 + k 3 s3 + p3 h3 quarter 3
+ c4 d 4 + v4 w4 + j4 r4 + k4 s4 + p4 h4 quarter 4
34
Quarterly flow balance relations:

d 1 , w1 d 2 , w2 d 3 , w3 d 4 , w4
r0 , s0 r1 , s1 r2 , s2 r3 , s3 r4 , s4
t=1 t=2 t=3 t=4
D1 ,W1 D2 ,W2 D3 ,W3 D4 ,W4
⎧ rt −1 + d t − rt = Dt
⎨ t = 1, 2, 3, 4
⎩ s t − 1 + w t − s t = Wt
Quarterly labor constraints
⎧1.5d t + 2 wt − h t ≤ 0
⎪
⎨ t = 1, 2, 3, 4
⎪ 0.9h ≤ h ≤ 1.1h
⎩ t −1 t t −1
h0 = 5000
35
EXAMPLE 6: PROBLEM STATEMENT
d1 w1 r1 s1 h1 d2 w2 r2 s2 h2 d3 w3 r3 s3 h3 d4 w4 r4 s4 h4
1 -1 = 1250
1 -1 = 1150
1.5 2 -1 ≤0
1 ≥ 4500
1 ≤ 5500
1 1 -1 = 1300
1 1 -1 = 1500
1.5 2 -1 ≤0
-0.9 1 ≥0
-1.1 1 ≤0
1 1 -1 = 3000
1 1 -1 = 1000
1.5 2 -1 ≤ 0
-0.9 1 ≥0
-1.1 1 ≤0
1 1 -1 = 1000
1 1 -1 = 1400
1.5 2 -1 ≤0
-0.9 1 ≥0
-1.1 1 ≤0
125 90 5.0 4.3 6.0 130 100 4.5 3.8 6.0 125 95 4.5 3.8 6.8 126 95 4.0 3.3 6.8 minimize
LINEAR PROGRAMMING PROBLEM
max (min) Z = c1x1 + ... + cnxn

s.t.
a11 x1 + a12 x2 + ... + a1n xn = b1
a21 x1 + a22 x2 + ... + a2n xn = b2
# #
am1 x1 + am2 x2 + ... + amn xn = bm
x1 ≥ 0, x2 ≥ 0, ... , xn ≥ 0
b1 ≥ 0, b2 ≥ 0, ... , bm≥ 0
36
STANDARD FORM OF LP (SFLP)
coefficient matrix
T
max (min) Z = c x m× n
A ∈ℜ
x ∈ ℜn
Ax= b decision
vector b ∈ ℜm
n
c ∈ℜ
x ≥ 0
requirement
vector profits
(costs)
vector
CONVERSION OF LP INTO SFLP
An inequality may be converted into an equality

by defining an additional nonnegative slack
variable
xslack ≥ 0
replace the given inequality ≤ b by
inequality + xslack = b
replace the given inequality ≥ b by
inequality – xslack = b
37
CONVERSION OF LP INTO SFLP
An unsigned variable xu is one whose sign is
unspecified
xu is converted into two signed variables x+ and x-
with
⎧ xu xu ≥ 0 ⎧ 0 xu ≥ 0
x+ = ⎨ x− = ⎨
⎩0 xu < 0 ⎩ - xu xu < 0
and with xu replaced by
xu = x+ − x−
SFLP CHARACTERISTICS
x is feasible if and only if x ≥ 0 and A x = b
S = { x | A x = b , x ≥ 0} is the feasible region
If S = Ø ⇒ LP is infeasible
x* is optimal ⇒ cT x * ≥ cTx , ∀ x ∈ S
x* may be unique, or may have multiple values
x * may be unbounded
38
ECE 307 – Techniques for Engineering
Decisions
Topic 3: Introduction to the Simplex Algorithm
George Gross
SOLUTION OF SYSTEMS OF LINEAR

EQUATIONS
We examine the solution of
Ax = b
using Gauss-Jordan elimination
We first use a simple example and then
generalize to cases of general interest
Consider the system of two equations in five
unknowns:
1
EQUATIONS
⎧⎪ x1 − 2 x 2 + x 3 − 4 x 4 + 2 x 5 = 2 (i )
S1 ⎨
⎪⎩ x1 − x 2 − x 3 − 3 x 4 − x 5 = 4 ( ii )
For this simple example, the number of
unknowns exceeds the number of equations
and so the system has multiple solutions;
this is the principal reason that the LP solution
is nontrivial

EQUATIONS
The Gauss-Jordan elimination uses elementary
row operations:
multiplication of any equation by a nonzero
constant
addition to any equation of a constant
multiple of any other equation
We transform S1 into the set S2 by multiplying
equation (i) by -1 and adding it to equation (ii)
⎧⎪ x1 − 2 x2 + x3 − 4 x4 + 2 x5 = 2
S2 ⎨
⎪⎩ x2 − 2 x3 + x4 − 3 x5 = 2
2
DEFINITIONS
A basic variable is a variable x i that appears
with the coefficient 1 in an equation and with the
coefficient 0 in all the other equations
The variables x j that are not basic are called
nonbasic variables
In the system S 2 , x 1 appears as a basic variable;
x 2 , x 3 , x 4 and x 5 are nonbasic variables
Basic variables may be generated through the
use of elementary row operations
DEFINITIONS
A pivot operation is the sequence of elementary

row operations that reduces a system of linear
equations into the form in which a specified
variable becomes basic
A canonical system is a set of linear equations
obtained through pivot operations with the
property that the system has the same number of
basic variables as the number of equations in the
set
3
CANONICAL SYSTEM FORM
We transform the system S2 into the canonical
form of system S3 :
⎧ x1 − 3 x3 − 2 x4 − 4 x5 = 6
S3 ⎪⎨
⎪⎩ x2 − 2 x3 + x4 − 3 x5 = 2
The basic solution is obtained from a canonical

system with all the nonbasic variables set to 0
For the example, we set x 3 = x 4 = x 5 = 0 and so

x 1 = 6 and x 2 = 2
BASIC FEASIBLE SOLUTION

A basic feasible solution is a basic solution in
which the values of all the basic variables are
nonnegative
In the example of system S 2 , we may choose

any two variables to be basic
In general for a system of m equations in n

⎛n⎞
unknowns there are ⎜ ⎟ combinations of
⎝m⎠
possible basic variables
4
BASIC FEASIBLE SOLUTION
As n increases the number of combinations
becomes large (although finite)
For the example, we have
⎛ 5⎞ 5!
⎜ ⎟ = = 10
2
⎝ ⎠ 3! 2!
combinations of possible choices

THE SIMPLEX SOLUTION METHOD
We next use a simple example to construct the

simplex solution method
The simplex method is a systematic and

efficient way of examining a subset of the basic
feasible solutions of the LP to hone in on an
optimal solution
We apply the notions introduced in the

definitions above
5
SIMPLEX METHODOLOGY EXAMPLE
max Z = 5 x 1 + 2 x 2 + 3 x 3 − x 4 + x 5
s .t .
⎧
canonical ⎪ x 1 + 2 x 2 + 2 x 3 + x 4 = 8 (*)
⎪
⎨
form ⎪
⎪⎩3 x 1 + 4x2 + x3 + x5 = 7 (**)
xi ≥ 0 i = 1, … ,5
The canonical form of the example allows the

determination of a basic feasible solution
x1 = x 2 = x 3 = 0 x4 = 8, x5 = 7
The corresponding value of the objective is
Z = − 8 + 7 = −1
The next step is to improve the basic feasible

solution by finding an adjacent basic feasible
solution
6
ADJACENT FEASIBLE SOLUTION
An adjacent feasible solution is one which
differs from the current basic feasible solution
in exactly one basic variable
Note, we characterize a basic feasible solution
by the following traits

basic variable ≥ 0
nonbasic variable = 0

The search of an adjacent basic feasible
solution uses the idea of making a nonbasic
variable into a basic variable by increasing its
value from 0 to the largest positive value
without violating any constraints
To make the search efficient, we choose the

nonbasic variable that can improve the value of
Z by the largest amount
7
In the example, consider the nonbasic variable
x 1 , we leave x 2 = x 3 = 0 and examine the
possibility of making x 1 into a basic variable
The variable x 1 enters in both constraints
x1 + x 4 = 8
3 x1 + x5 = 7
The largest value x 1 can assume without

making x 4 or x 5 negative is
⎧ 7⎫ 7
min ⎨8, ⎬ =
⎩ 3⎭ 3
We have the new basic variable with the value

7
x1 = ,
3
and the previous basic variable
17
x4 = ,
3
8
and the three nonbasic variables are set to 0:
x 2 = x 3 = 0 and x 5 = 0
Note that we have an improvement in Z since its

value becomes
7 17 18
Z =5 i − = = 6 > −1
3 3 3
We next need to put the system of equations
into canonical form:
max Z = 5 x 1 + 2 x 2 + 3 x 3 − x 4 + x 5
s .t .
⎧ x + 2x + 2 x3 + x4 = 8 (*)
canonical ⎪⎪
1 2
⎨
form ⎪
⎪⎩ 3 x 1 + 4 x 2 + x3 + x5 = 7 (**)
xi ≥ 0 i = 1, … ,5
9
multiply equation (**) by −

1 and add to
3
equation (*)
2 5 1 17
x 2 + x 3 + x4 − x 5 =
3 3 3 3
multiply equation (**) by 1

3
4 1 1 7
x1+ x2 + x + x5 =
3 3 3 3 3

We continue this process until the condition of
optimality is satisfied:
in a maximization problem, a basic feasible

solution is optimal if and only if the relative
profits of each nonbasic variable is ≤ 0
in a minimization problem, a basic feasible

solution is optimal if and only if the relative
costs of each nonbasic variable is ≥ 0
10
The relative profits (costs) are given by the
change in Z corresponding to a unit change in a
nonbasic variable
SIMPLEX ALGORITHM FOR

MAXIMIZATION
Step 1: start with an initial basic feasible
solution with all constraint equations in
canonical form
Step 2: check for optimality condition: if the
relative profits are < 0 for each
nonbasic variable, then the basic
feasible solution is optimal and stop;
else, go to Step 3
11
SIMPLEX ALGORITHM FOR
MAXIMIZATION
Step 3: select a nonbasic variable to become
the new basic variable; check the limits
on the nonbasic variable – the limiting
constraint determines the basic variable
that is being replaced by the selected
nonbasic variable
Step 4: determine the canonical form for the
new set of basic variables through
elementary row operations; compute the
basic feasible solution, Z and return to
Step 2
THE SIMPLEX TABLEAU
We use an efficient way to represent visually the
steps in the simplex method through a sequence
of so-called tableaus
We illustrate the tableau for the simple example
for the initial basic feasible solution

12
THE SIMPLEX TABLEAU
coefficients of the coefficient of xj in Z

basic variables in Z
cj 5 2 3 –1 1
constraint
cB basic constants
x1 x2 x3 x4 x5
variables
–1 x4 1 2 2 1 8
1 x5 3 4 1 1 7
THE SIMPLEX TABLEAU

The optimality check requires the evaluation of
⎛ column corresponding ⎞
c j = c j − ⎜⎜ c TB i to x in canonical form ⎟⎟
⎝ j ⎠
For each nonbasic variable j, for our example,
we have
⎡1 ⎤
c1 = 5 − ⎡⎣−1, 1⎤⎦ i ⎢ ⎥ = 3
⎣ 3⎦
⎡ 2⎤
c 2 = 2 − ⎡⎣−1, 1⎤⎦ i ⎢ ⎥ = 0
⎣4 ⎦
⎡ 2⎤
c 3 = 3 − ⎡⎣−1, 1⎤⎦ i ⎢ ⎥ = 4
⎣1 ⎦
13
THE SIMPLEX TABLEAU
We interpret as the change in Z corresponding
to a unit increase in x j
cj 5 2 3 –1 1 constraint
basic constants
cB x1 x2 x3 x4 x5
variables
–1 x4 1 2 2 1 8
1 x5 3 4 1 1 7
cT 3 0 4 0 0 Z = –1
SIMPLEX TABLEAU
Note that the optimality test indicates that
c1 = 3 > 0 and c3 = 4 > 0

and so the initial basic feasible solution is non-
optimal
Since c 3 > c1 , we pick x 3 as the nonbasic

variable to become a basic variable
We examine the limiting solution for x 3 in the

two constraint equations:
14
THE SIMPLEX TABLEAU
equation limiting basic upper limit on

variable x3
1 x4 (8/2) = 4
2 x5 (7/1) = 7
and so the limiting value is
min { 4, 7 } = 4
We replace the basic variable x 4 by x 3

max Z = 5 x 1 + 2 x 2 + 3 x 3 − x 4 + x 5
s .t .
⎧ x + 2x + 2 x3 + x4 = 8 (*)
canonical ⎪⎪
1 2
⎨
form ⎪
⎪⎩ 3 x 1 + 4 x 2 + x3 + x5 = 7 (**)
xi ≥0 i = 1, … , 5
15
THE SIMPLEX TABLEAU
For the new basic feasible solution, we put the
equations into canonical form
multiplying (*) by 1 to produce (*†)

2
subtract (*†) from (**) to produce (**†)
1 1
x1 + x 2 + x 3 + x = 4 (* †)
2 2 4
5 1
x1 + 3 x 2 − x + x 5 = 3 (* * †)
2 2 4
The adjacent basic feasible solution is
x1 = x2 = x4 = 0 x3 = 4, x5 = 3
THE SIMPLEX TABLEAU
We prepare the corresponding tableau
cj 5 2 3 –1 1
constraint
cB basic
x1 x2 x3 x4 x5 constants
variables
3 x3 1/2 1 1 1/2 4
1 x5 5/2 3 –1/2 1 3
cT 1 –4 0 –2 0 Z = 15
x 3 = 4, x 5 = 3
16
THE SIMPLEX TABLEAU
Since c 1 > 0 , the basic feasible solution is non-

optimal
We examine how to bring x1 into the basis
equation limiting basic upper limit on

variable x1
(*†) x3 4/(1/2) = 8
(**†) x5 3/(5/2) = 6/5
THE SIMPLEX TABLEAU

The variable x1 enters the basis with the value
⎧ 6⎫ 6
min ⎨ 8, ⎬ =
⎩ 5⎭ 5
and x5 is replaced as a basic variable by x 1
We need to put the equations
1 1
x1 + x 2 + x 3 + x = 4 (*†)
2 2 4
5 1
x1 + 3 x 2 − x + x 5 = 3 (**†)
2 2 4
into canonical form for the basic variables
x 3 and x 1
17
THE SIMPLEX TABLEAU
The following elementary row operations are
used
multiply (**†) by – 1/5 and add to (*)

2 3 1 17
x + x3 + x − x =
5 2 5 4 5 5 5
multiply (**†) by 2/5
6 1 2 6
x1 + x − x + x =
5 2 5 4 5 5 5
and construct the corresponding tableau

THE SIMPLEX TABLEAU
cj 5 2 3 –1 1
constraint
cB basic x1 x2 x3 x4 x5 constants
variables
3 x3 2/5 1 3/5 – 1/5 17/5
5 x1 1 6/5 – 1/5 2/5 6/5
cT 0 – 26/5 0 – 9/5 – 2/5 Z = 81/5
16.2 > 15
c j ≤ 0 implies optimality
18
SIMPLEX TABLEAU EXAMPLE
max Z = 3 x1 + 2x2
s .t .
− x1 + 2x2 ≤ 4
x1 + 2x2 ≤ 14
x1 − x2 ≤ 3
x1 ≥ 0 x 2 ≥ 0
We put this problem into standard form:

max Z = 3 x 1 + 2 x 2 canonical form
s .t .
⎫
− x1 + 2 x 2 + x 3 = 4 ⎪
⎪⎪
x1 + 2 x 2 + x4 = 14 ⎬
⎪
x1 − x 2 + x5 = 3 ⎪
⎪⎭
x1 , …, x5 ≥ 0
x 3 , x 4 , x 5 are fictitious variables
19
cj 3 2 0 0 0 constraint
cB basic
variables
0 x3 –1 2 1 4
0 x4 3 2 1 14
0 x5 1 –1 1 3
cT 3 2 0 0 0 Z = 0
c j = c j − ( c BT • column corresponding to x j )

The data in c T indicates that the highest relative
profits correspond to x 1 so want to make x 1 a
basic variable
To bring x 1 into the basis requires to evaluate
⎧ 14 ⎫
min ⎨∞ , , 3⎬ = 3
⎩ 3 ⎭
and so x 1 replaces x 5 with the value 3
We evaluate the basic variable at the adjacent
basic feasible solution and convert into
canonical form; the new tableau becomes
20
cB basic
variables
0 x3 1 1 1 7
0 x4 5 1 –3 5
3 x1 1 –1 1 3
cT 0 5 0 0 –3 Z = 9
We reproduce here the calculation of the c T
components
c j = c j −( c T
B i column corresponding to x j )
for each nonbasic variable x j
Note that ci = 0 for each basic variable x i by
definition
21
The calculations give

c1 = 0 by definition since x 1 is in the basis
⎡1⎤
⎢ ⎥ indicates possible
c 2 = 2 − ⎡⎣0 0 3⎤⎦ ⎢ 5 ⎥ = 5 improvement
⎢ ⎥
⎣ − 1⎦
c 3 = 0 by definition since x 3 is in the basis
c 4 = 0 by definition since x 4 is in the basis
⎡ 1 ⎤
⎢ ⎥
c 5 = 0 − ⎡⎣0 0 3⎤⎦ ⎢ − 3⎥ = − 3
⎢ ⎥
⎣ 1 ⎦
Clearly, the only choice is to get x 2 into the

basis and so we need to establish the limiting
condition from the three equations by evaluating
min {7 , 1, ∞} = 1
and so x 2 replaces x 4 , which becomes a

nonbasic variable
We need to rewrite the equations into canonical

form and construct the new tableau
22
cB basic
variables
0 x3 1 –1/5 8/5 6
2 x2 1 1/5 –3/5 1
3 x1 1 1/5 2/5 4
cT 0 0 0 –1 Z = 14
cj ≤ 0∀ j ⇒ optimum
An optimum is at the solution of

canonical form
1 8
x3 − x + x = 6 ⎫
5 4 5 5
x2 + 1
x − 2
x = 1 ⎬
⎪
5 4 5 5
x1 + 1
x + 2
x = 4 ⎭
⎪
5 4 5 5
given by
x4 = x5 = 0
x3 = 6
x2 = 1
x1 = 4
23
LINEAR PROGRAMMING EXAMPLE
Consider the following LP

max Z = 3 x1 + 2 x 2
s .t .
− x1 + 2x2 ≤ 4
3 x1 + 2 x 2 ≤ 14
x1 − x 2 ≤ 3
x1 ≥ 0 x 2 ≥ 0
The graphical representation corresponds to
x2 = 4
x2
x + 2
-
D
3
3
=
3x 1
2
- x
E
+
3x 1
2x 2
2
1
-x
3x 1
=1
tableau 3
2x 2
4
+
2x 2
1
C
Z
=
=9
Z
tableau 1
=3
B
A x1
1 2 3 tableau 2
24
The tableau approach leads to C which is an

optimal solution with
x 1 = 4, x 2 = 1, x 3 = 6, x 4 = 0, x 5 = 0
Note that any point along CD has Z = 14 and as
such D is another optimal solution correspon-
ding to an adjacent basic feasible solution
We may obtain D from C by bringing into the
basis the nonbasic variable x 5 in Tableau 3;
note that c5 = 0
We may choose x 5 as a basic variable without
effecting Z since the relative profits are 0 ; we
compute the limiting value of x 5
The limit is imposed by x 3 which, consequent-
ly, leaves the basis
The corresponding tableau is:

25
cj 3 2 0 0 0
constraint
cB basic x1 x2 x3 x4 x5 constants
variables
0 x3 5/8 –1/8 1 15/4
2 x2 1 3/8 1/8 13/4
3 x1 1 –1/4 1/4 5/2
cT 0 0 0 –1 0 Z = 14
cj ≤ 0 ∀ j
The adjacent feasible solution is given by
5 13 15
x1 = , x2 = , x3 = x4 = 0 , x5 =
2 4 4
Note that at this basic feasible solution,
cj ≤ 0 ∀ j
and so this is also an optimal solution
26
ALTERNATE OPTIMAL SOLUTION
In general, an alternate optimal solutions is
indicated whenever there exists a nonbasic
variable x j with c j = 0 in an optimal tableau
MINIMIZATION LP
Consider a minimization problem

n
min Z = ∑ ci xi
i =1
s.t .
Ax = b
x≥0
In the simplex scheme, replace the optimality
check by the following :
if each coefficient c j is ≥ 0 , stop; else, select
the nonbasic variable with the most negative
value in c to become the new basic variable
27
MINIMIZATION LP
Every minimization LP may be solved as a

maximization LP because of equivalence
min Z = cT x max Z ′ = (− c T ) x
s.t . s.t .
Ax = b Ax = b
x ≥ 0 x ≥ 0
with the solutions of Z and Z ′ related by
min{ Z } = − max { Z ′}
COMPLICATIONS IN THE SIMPLEX

METHODOLOGY
Two variables x j and x k are tied in the selection

of the nonbasic variables to replace a current
basic variable when c j = c k ; the choice of the
new nonbasic variable to enter the basis is
arbitrary
Two or more constraints may give rise to the
same minimum ratio value in selecting the basic
variable to be replaced
We consider the example of the following tableau
28
METHODOLOGY
cj 0 0 0 2 0 3/2 constraint
cB basic
x1 x2 x3 x4 x5 x6 constants
variables
0 x1 1 1 –1 0 2
0 x2 1 2 0 1 4
0 x3 1 1 1 1 3
cT 0 0 0 2 0 3/2 Z = 0
candidate for basic variable


METHODOLOGY
in selecting the nonbasic variable x 4 to
enter the basis, we observe that the first
two constraints give the same minimum
ratio: this means that when x 4 is first
increased to 2 , both the basic variables x 1
and x 2 will reduce to zero even though only
one of them can be made a nonbasic
variable
we arbitrarily decide to remove x 1 from the
basis to get the new basic feasible solution:
29
METHODOLOGY
3
cj 0 0 0 2 0
2 constraint
cB basic constants
variables x1 x2 x3 x4 x5 x6
2 x4 1 1 –1 2
0 x2 –2 1 2 1 0
0 x3 –1 1 1 1 1
cT –2 0 0 0 2 3/2 Z = 4

METHODOLOGY
in the new basic feasible solution
x1 = 0 x 2 = 0 x 3 = 1 x 4 = 2 x 5 = 0 x 6 = 0 ,
we treat x 2 as a basic variable whose value is
0, the same as if it were a nonbasic variable

30
DEGENERACY
A degenerate basic feasible solution is one

where one or more basic variables is 0
Degeneracy may lead to a number of
complications in the simplex approach: an
important implication is a minimum ratio of 0 , so
that no new nonbasic variable maybe included in
the basis and therefore the basis remains
unchanged
We consider the following example tableau
DEGENERACY
cj 0 0 0 2 0 3
2 constraint
cB basic
x1 x2 x3 x4 x5 x6 constants
variables
2 x4 1/2 1 1/2 2
0 x5 –1 1/2 1 1/2 0
0 x3 1 –1 1 0 1
cT 0 –1 0 0 0 1/2 Z = 4
31
DEGENERACY
the logical choice being the nonbasic variable

x 6 to enter the basis; this leads to finding the
limiting constraint from two equations
1
x = 2 − x4
2 6
1
x = 0 − x5
2 6
and no constraint in the third equation; thus
x 6 = min {4, 0 , ∞ }
DEGENERACY
Degeneracy may result in the construction of

new tableaus without improvement in the
objective function value, thereby reducing the
efficiency of the computations: theoretically, an
infinite loop, the so-called cycling, is possible
Whenever ties occur in the minimum ratio rule,
an arbitrary decision is made regarding which
basic variable is replaced, ignoring the
theoretical consequences of degeneracy and
cycling
32
MINIMUM RATIO RULE COMPLICATIONS
The minimum ratio rule may not be able to

determine the basic variable to be replaced:
this is the case when all equations lead to ∞
as the limit
Consider the example and corresponding tableau
max Z = 2 x1 + 3 x 2
s.t .
x1 − x 2 + x 3 = 2
− 3 x1 + x 2 + x4 = 4
x i ≥ 0, i = 1,...,4
cj 2 3 0 0 constraint
cB basic
x1 x2 x3 x4 constants
variables
0 x3 1 –1 1 2
0 x4 –3 1 1 4
2 3 0 0 Z = 0
cT
The nonbasic variable x 2 enters the basis to
replacing x 4 and the new tableau is

33
cj 2 3 0 0 constraint
cB basic constants
variables x1 x2 x3 x4
0 x3 –2 1 1 6
3 x2 –3 1 1 4
cT 11 0 0 –3 Z = 12
We select x 1 to enter the basis but we are

unable to get limiting constraints from the two
equations
1
−2 x 1 + x 3 = 6 x1 = x − 3
2 3
1 4
−3 x 1 + x 2 = 4 x1 = x2 −
3 3
In fact, as x 1 increases so do x 2 and x 3 and Z
and therefore, the solution is unbounded
The failure of the minimum ratio rule to result in
a bound at any simplex tableau implies that the
problem has an unbounded solution

34
ECE-307
Duality Concepts in Linear Programming
George Gross
© 2006 University of Illinois Board of Trustees, All Rights Reserved

ECE 307 © 2006 George Gross, University of Illinois at Urbana-Champaign, All Rights Reserved.
DUALITY
Definition: A LP is in symmetric form if all the
variables are restricted to be nonnegative and
all the constraints are inequalities:

corresponding
objective type
inequality type
max ≤
min ≥
1
DUALITY DEFINITIONS
We define the primal problem as
max Z = cT x
s.t.
Ax ≤ b (P)
x ≥ 0
The dual problem is therefore
min W = bTy
s.t.
AT y ≥ c (D)
y ≥ 0
DUALITY DEFINITIONS
The problems (P ) and (D) are called the
symmetric dual LP
max Z = c1 x1 + c2 x 2 + ... + cn xn ⎫
⎪
s.t. ⎪
a11 x1 + a12 x 2 + ... + a1 n xn ≤ b1 ⎪
⎪⎪
a21 x1 + a22 x2 + ... + a2 n xn ≤ b2 ⎬ (P)
# ⎪
⎪
am 1 x1 + am 2 x2 + ... + amn xn ≤ bm ⎪
⎪
x1 ≥ 0, x2 ≥ 0, ..., xn ≥ 0 ⎪⎭
2
DUALITY DEFINITIONS
min W = b1 y1 + b2 y2 + ... + bm ym ⎫
⎪
s.t. ⎪
⎪
a11 y1 + a21 y2 + ... + am 1 ym ≥ c1 ⎪
⎪⎪
a12 y1 + a 22 y2 + ... + am 2 ym ≥ c2 ⎬ ( D)
⎪
# ⎪
⎪
a1 n y1 + a2 n y2 + ... + amn xm ≥ cn ⎪
⎪
y1 ≥ 0, y2 ≥ 0, ..., ym ≥ 0 ⎪⎭
EXAMPLE 1: MANUFACTURING
TRANSPORTATION PROBLEM
discount / cost coefficients R1

retail stores
warehouses
R1 R2 R3 W1
W1 2 4 3 R2
W2 5 3 4 W2
R3
3
EXAMPLE 1: MANUFACTURING
supplies @ W1 ≤ 300
W2 ≤ 600
demands @ R1 ≥ 200
R2 ≥ 300
R3 ≥ 400
The problem is to determine the least-cost
shipping schedule
xij = quantity shipped from Wi to Rj
i = 1, 2 j = 1, 2, 3
cij = elements of transportation cost matrix
FORMULATION STATEMENT
2 3
min Z = ∑∑ c
i =1 j =1
ij xij = 2 x11 + 4 x12 + 3 x13 + 5 x21 + 3 x22 + 4 x23
s.t.
x11 + x12 + x13 ≤ 300
x21 + x22 + x23 ≤ 600
x11 + x21 ≥ 200
x12 + x22 ≥ 300
x13 + x23 ≥ 400
xij ≥ 0 i = 1, 2 j = 1, 2, 3
4
DUAL PROBLEM SETUP
2 3
min Z = ∑∑ c
i =1 j =1
ij xij
s.t.
y1 ↔ − x11 − x12 − x13 ≥ −300
y2 ↔ − x21 − x22 − x23 ≥ −600
y3 ↔ x11 + x21 ≥ 200
y4 ↔ x12 + x22 ≥ 300
y5 ↔ x13 + x23 ≥ 400
xij ≥ 0 i = 1, 2 j = 1, 2, 3
DUAL PROBLEM SETUP

maxW = − 300 y1 − 600 y2 + 200 y3 + 300 y4 + 400 y5
s.t.
− y1 + y3 ≤ c11 = 2
− y1 + y4 ≤ c12 = 4
− y1 + y5 ≤ c13 = 3
− y2 + y 3 ≤ c21 = 5
− y2 + y4 ≤ c22 = 3
− y2 + y5 ≤ c23 = 4
yi ≥ 0 i = 1, 2, ..., 5
5
INTERPRETATION OF DUAL PROBLEM
The moving company approaches the manufac-

turer with the following proposition to:
buy all the 300 units at W1 for y1 /unit
buy all the 600 units at W2 for y2 /unit
sell all the 200 units at R1 at y3 /unit
sell all the 300 units at R 2 at y4 /unit
sell all the 400 units at R 3 at y5 /unit
To convince the manufacturer to get his
business, the mover ensures that he can deliver
for less than the transportation costs the
manufacturer would incur (the dual constraints)
− y1 + y3 ≤ c11 = 2
− y1 + y4 ≤ c12 = 4
− y1 + y5 ≤ c13 = 3
− y2 + y3 ≤ c21 = 5
− y2 + y4 ≤ c22 = 3
− y2 + y5 ≤ c23 = 4
The mover wishes to maximize profits

revenues − costs ⇒ dual cost objective function
maxW = − 300 y1 − 600 y2 + 200 y3 + 300 y4 + 400 y5
6
EXAMPLE 2: MANUFACTURING OF
FURNITURE PRODUCTS
Resource requirements
item sales price ($)
desks 60
tables 30
chairs 20
lumber board
requirements
finishing
labor
carpentry
EXAMPLE 2: MANUFACTURING OF
FURNITURE PRODUCTS
The Dakota Furniture Company manufactures:
resource desk table chair available
lumber board (ft) 8 6 1 48
finishing (h) 4 2 1.5 20
carpentry (h) 2 1.5 0.5 8
We assume that the demand for desks, tables

and chairs is unlimited and the available resour-
ces are already purchased
The decision problem is to maximize total
revenues
7
PRIMAL AND DUAL PROBLEM
FORMULATION
We define decision variables
x1 = number of desks produced
x2 = number of tables produced
x3 = number of chairs produced
The Dakota problem is
max Z = 60 x1 + 30 x2 + 20 x3
y1 ↔ 8 x1 + 6 x2 + x3 ≤ 48 lumber
y2 ↔ 4 x1 + 2 x2 + 1.5 x3 ≤ 20 finishing
y3 ↔ 2 x1 + 1.5 x2 + 0.5 x3 ≤ 8 carpentry
x1 , x2 , x3 ≥ 0

FORMULATION
The dual problem is
min W = 48 y1 + 20 y2 + 8 y3
8 y1 + 4 y2 + 2 y3 ≥ 60 desk
6 y1 + 2 y2 + 1.5 y3 ≥ 30 table
y1 + 1.5 y2 + 0.5 y3 ≥ 20 chair
y1 , y2 , y3 ≥ 0
8
FORMULATION
max Z = 60 x1 + 30 x2 + 20 x3
y1 ↔ 8 x1 + 6 x2 + x3 ≤ 48 lumber
y2 ↔ 4 x1 + 2 x2 + 1.5 x3 ≤ 20 finishing
y3 ↔ 2 x1 + 1.5 x2 + 0.5 x3 ≤ 8 carpentry
x1 , x2 , x3 ≥ 0
min W = 48 y1 + 20 y2 + 8 y3
8 y1 + 4 y2 + 2 y3 ≥ 60 desk
6 y1 + 2 y2 + 1.5 y3 ≥ 30 table
y1 + 1.5 y2 + 0.5 y3 ≥ 20 chair
y1 , y2 , y3 ≥ 0
An entrepreneur wishes to purchase all of

Dakota’s resources
He thus needs to determine the price to pay for
each unit of the resources
y1 = price paid for 1 lumber board ft
y2 = price paid for 1 h of finishing
y3 = price paid for 1 h of carpentry
We solve the Dakota dual problem to determine

y1, y2, y3
9
To induce Dakota to sell the raw resources, the
resource prices must be set sufficiently high
For example, the entrepreneur must offer Dakota
at least $60 for a combination of resources that
includes 8ft of lumber board, 4h of finishing and
2h of carpentry since Dakota could use this
combination to sell a desk for $60
This implies the construction of the dual
constraint
8 y1 + 4 y2 + 2 y3 ≥ 60
In the same way we obtain the two additional
constraints for a table and for a chair
In the same way, the ith primal variable corresp-
onds to the dual variable of the ith constraint in
the dual problem statement
The ith dual variable corresponds to the ith primal
constraint
10
EXAMPLE 3: DIET PROBLEM
A new diet requires that all food eaten come from
one of the four “basic food groups”: chocolate
cake, ice cream, soda and cheesecake
The four foods available for consumption are
given in the table
Requirements for each day are intakes of:
At least 500 cal
At least 6 oz chocolate
At least 10 oz sugar
At least 8 oz fat

chocolate costs
food calories sugar (oz) fat (oz)
(oz) (cents)
brownie 400 3 2 2 50
chocolate
ice cream 200 2 2 4 20
(scoop)
cola
150 0 4 1 30
(bottle)
pineapple
cheeseca- 500 0 4 5 80
ke (piece)
11
PROBLEM FORMULATION
Objective of the problem is to minimize the costs

of the diet
Decision variables are defined for each day’s
purchases
x1 = number of brownies
x2 = number of chocolate ice cream scoops
x3 = number of bottles of soda
x4 = number of pineapple cheesecake pieces
PROBLEM FORMULATION
The problem statement is
min Z = 50 x1 + 20 x2 + 30 x3 + 80 x4
s.t.
400 x1 + 200 x2 + 150 x3 + 500 x4 ≥ 500 cal
3 x1 + 2 x2 ≥ 6oz
2 x1 + 2 x 2 + 4 x 3 + 4 x4 ≥ 10oz
2 x1 + 4 x 2 + x 3 + 5 x4 ≥ 8oz
xi ≥ 0 i = 1, 4
12
The dual problem is
max W = 500 y1 + 6 y2 + 10 y3 + 8 y4
s.t.
400 y1 + 3 y2 + 2 y3 + 2 y4 ≤ 50 brownie
200 y1 + 2 y2 + 2 y3 + 4 y4 ≤ 20 ice - cream
150 y1 + 4 y3 + y4 ≤ 30 soda
500 y1 + 4 y3 + 5 y4 ≤ 80 cheesecake
y1 , y2 , y3 , y4 ≥ 0
INTERPRETATION OF DUAL
Consider we have a sales person of “nutrients”

who is interested in assuming that each dieter
meets daily requirements by purchasing
calories, sugar, fat and chocolate
The key decision is to determine the prices
yi = price per unit charged to dieter
Objective of sales person is to set the prices yi
so as to maximize revenues from selling to the
dieter the daily ration of required nutrients
13
INTERPRETATION OF DUAL
Now, the dieter can purchase a brownie for 50¢
and have 400cal, 30oz of chocolate, 2oz of sugar
and 2oz of fat
Salesperson must set yi sufficiently low to
entice the buyer to get the required nutrients
from the brownie:
brownie
400 y1 + 3 y2 + 2 y3 + 2 y4 ≤ 50
constraint
We derive similar constraints for the ice cream,
soda and cheesecake
DUAL PROBLEMS
max Z = cT x
s.t.
Ax ≤ b (P)
x ≥ 0
min W = bTy
s.t.
AT y ≥ c (D)
y ≥ 0
14
WEAK DUALITY THEOREM
Far any x feasible for ( P ) and any y feasible for
(D)
c T x ≤ bT y
Proof:
AT y ≥ c ⇒ c T ≤ yT A ⇒ c T x ≤ yT Ax
c T x ≤ yT Ax ≤ yT b = b T y
COROLLARIES OF THE WEAK

DUALITY THEOREM
Corollary 1:
x feasible for ( P ) ⇒ c T x ≤ yT b
for any feasible y for ( D )
c T x ≤ y*T b = min W
For any feasible x for ( P ) ,
cT x ≤ min W
15
DUALITY THEOREM
Corollary 2:
y feasible for ( D ) ⇒ c T x ≤ yT b
for every feasible x for ( P )
max Z = max c T x = c T x* < y*T b
For any feasible y of ( D ) ,
yT b ≥ maz Z

DUALITY THEOREM: 3
Corollary 3:
( P ) is feasible and max Z is unbounded, i.e.,
Z → +∞ .
Then, ( D ) has no feasible solution.
Corollary 4:
( D ) is feasible and min Z is unbounded, i.e.,
Z → −∞ .
Then, ( P ) is infeasible
16
DUALITY THEOREM APPLICATION
Consider the maximization problem
max Z = x1 + 2 x2 + 3 x3 + 4 x4 = [1, 2, 3, 4] x

c T
s.t.
(P )
⎡1 2 2 3 ⎤ ⎡ 20 ⎤
⎢ ⎥ x ≤ ⎢ ⎥
⎣
2 1 3 2⎦

⎣ 20 ⎦
N
A b
x ≥ 0
The corresponding dual is given by
min W = bT y
s.t.
(D )
AT y ≥ c
y ≥ 0
With the approximate substitutions, we have

17
min W = 20 y1 + 20 y2
s.t.
y1 + 2 y2 ≥ 1
2 y1 + y2 ≥ 2
2 y1 + 3 y2 ≥ 3
3 y1 + 2 y2 ≥ 4
y1 ≥ 0, y2 ≥ 0
GENERALIZED FORM OF THE DUAL
Consider the primal decision

x i = 1, i = 1, 2, 3, 4
which is feasible for (P) with

Z = c T x = 10
and the dual decision

y i = 1, i = 1, 2,
which is feasible for (D) with

W = b T y = 40
18
Clearly,
Z ( x1 , x2 , x3 , x4 ) = 10 < 40 = W ( y1 , y2 )
and so clearly, (P) and (D) satisfy the Weak
Duality Theorem
Moreover, we have
Corollary 1 ⇒ 10 ≤ min W = W ( y *1 , y 2* )
Corollary 2 ⇒ max Z = Z ( x *1 , x 2* , x 3* , x 4* ) ≤ b T y = 40
ADDITIONAL CORROLARIES
Corollary 5:
(P) is feasible and (D) is infeasible. Then,
(P) is unbounded
Corollary 6:
(D) is feasible and (P) is infeasible. Then,
(D) is unbounded
19
EXAMPLE
Consider the primal dual problems:

max Z = x1 + x2 min W = 2 y1 + y2
s.t . s.t .
− x1 + x2 + x3 ≤ 2 ( P ) − y1 − 2 y2 ≥ 1 ( D)
−2 x1 + x2 − x3 ≤ 1 y1 + y2 ≥ 1
x1 , x2 , x3 ≥ 0 y1 − y2 ≥ 0
y1 , y2 ≥ 0
Now
x = 0 is feasible for ( P )
but
EXAMPLE
− y1 − 2 y2 ≥ 1
is impossible for (D) since it is inconsistent with
y1 , y2 ≥ 0
It follows from Corollary 5 that (D) is infeasible
Z →∞
You should be able to show this result by
solving (P) using the simplex

20
OPTIMALITY CRITERION THEOREM
We consider the primal-dual problems (P) and (D)

with
x 0 is feasible for ( P ) x 0 is optimal for ( P )
y 0 is feasible for ( D ) ⇒ and
c T x 0 = bT x 0 y 0 is optimal for ( D )
We next provide the proof:
x 0 is feasible for ( P ) Weak Duality

⇒ c T x 0 ≤ bT y 0
y is feasible for ( D )
0
Theorem
OPTIMALITY CRITERION THEOREM
but we are given that

cT x 0 = b t y 0
and so it follows that
∀ feasible x , y 0 feasible c T x ≤ bT y 0 = c T x 0
and therefore
x 0 is optimal
Similarly
∀ feasible y , x 0 feasible bT y ≥ c T x 0 = bT y 0
so it follows that
y 0 is optimal
21
MAIN DUALITY THEOREM
Given that (P) is feasible and (D) is feasible;
∃ x * feasible for ( P ) which is optimal and

then,
∃ y* feasible for ( D ) which is optimal such that
c T x * = b T y*
COMPLEMENTARY SLACKNESS
CONDITIONS
x* and y * are optimal for (P) and (D) respectively,
if and only if
( ) (b− A x )
*T
0 = y *T A − c T x * + y *
= y*T b − c T x*
To prove this equivalence result, we define the
slack variables u ∈ \ m and v ∈ \ n such that x
and y are feasible; at the optimum,
A x * + u* = b x * , u* ≥ 0
AT y* − v * = c y* , v * ≥ 0
22
CONDITIONS
where the optimal values of the slack variables
u* and v * correspond to x * , y *
Now,
y *T Ax* + y *T u* = y *T b = b T y *
x *T A T y * − x *T v * = x *T c = c T x *

y Ax *
*T
CONDITIONS
This implies that
y *T u* + v *T x* = b T y* − c T x *
We need to prove optimality which is true if and
only if
y *T u* + v *T x* = 0
23
CONDITIONS
However,
x * , y* are optimal
Main
⇒
Duality Theorem
c T x* = b T y ⇒ y u + v *T x* = 0
* *T *
also,
y *T u* + v *T x* = 0 ⇒ b T y* = c T x *
Optimality
⇒
Criterion Theorem
x* is optimal for ( P )
y* is optimal for ( D )
CONDITIONS
Note that
x * , y * , u * , v * > 0 ⇒ component - wise each element ≥ 0
y *T u* + v * x* = 0 ⇒ y *i u *i = 0 ∀i = 1, ..., m and
v *j x *j = 0 ∀j = 1, ..., n
At the optimum,
⎛ n ⎞
y ⎜ bi −
*
i ∑ a ij x j⎟ = 0
*
i = 1, ..., m
⎝ j =1 ⎠
and
⎛ m ⎞
x ⎜ ∑ a ji y i* − c j ⎟ = 0
*
j j = 1, ..., n
⎝ i =1 ⎠
24
CONDITIONS
Hence, for i = 1, 2, … , m
n
y > 0 ⇒ bi =
*
i ∑a
j =1
ij x *j
and
n
bi − ∑a
j =1
ij x *j > 0 ⇒ y i* = 0
Similarly for j = 1, 2, … , n
m
x *j > 0 ⇒ ∑a
i =1
ji y i* = c j
and
m
∑a
i =1
ji y i* − c j > 0 ⇒ x *j = 0
EXAMPLE
max Z = x1 + 2 x2 + 3 x3 + 4 x4
s.t .
x1 + 2 x2 + 2 x3 + 3 x4 ≤ 20 ( P)
2 x1 + x2 + 3 x3 + 2 x4 ≤ 20
xi ≥ 0 i = 1,...,4
min W = 20 y1 + 20 y2
s.t .
y1 + 2 y2 ≥ 1
2 y1 + y2 ≥ 2
2 y1 + 3 y2 ≥ 3 ( D)
3 y1 + 2 y2 ≥ 4
y1 ≥ 0
25
EXAMPLE
x * , y* optimal ⇒
y *1 ( 20 − x *1 − 2 x 2* − 2 x 3* − 3 x 4* ) = 0
y 2* ( 20 − 2 x *1 − x 2* − 3 x 3* − 2 x 4* ) = 0
⎛ 1.2 ⎞
y* = ⎜ ⎟ is given as an optimal solution with
⎝ 0.2 ⎠
min W = 28
Then,
EXAMPLE
x *1 + 2 x 2* + 2 x 3* + 3 x 4* = 20
2 x *1 + x 2* + 3 x 3* + 2 x 4* = 20
Also,
y *1 + 2 y 2* = 1.2 + 0.4 > 1 ⇒ x *1 = 0
2 y *1 + y 2* = 2.4 + 0.2 > 2 ⇒ x 2* = 0
2 y *1 + 3 y 2* = 2.4 + 0.6 = 3
3 y *1 + 2 y 2* = 3.6 + 0.4 = 4
therefore
2 x 3* + 3 x 4* = 20 ⇒ x 3* = 4
max Z = 28
3 x 3* + 2 x 4* = 20 ⇒ x 4* = 4
26
USES OF THE COMPLEMENTARY
SLACKNESS CONDITION
Key applications
finding optimal (P) solution given optimal (D)
solution and vice versa
verification of optimality of solution (whether
a feasible solution is optimal)
We can start with a feasible solution and attempt
to construct an optimal dual solution; if we
succeed, then the feasible prime solution is
optimal
DUALITY
max Z = cT x
s.t.
Ax ≤ b (P)
x ≥ 0
min W = bTy
s.t.
AT y ≥ c (D)
y ≥ 0
27
DUALITY
Suppose the primal problem is minimization, then,
min Z = cT x
s.t.
Ax ≥ b (P)
x ≥ 0
max W = bTy
s.t.
AT y ≤ c (D)
y ≥ 0
INTERPRETATION
The economic interpretation is
Z * = max Z = c T x* = b T y* = W * = minW
b i − constrained quantities of the resources i = 1, 2, ..., m
y *i − optimal dual variables
Suppose,
bi → bi + Δbi ⇒ Δ Z = y *i Δbi
In words, the optimal dual variable for each
primal constraint gives the net change in the
optimal value of the objective function Z for a
one unit change in the constraint on resources
28
INTERPRETATION
Economists refer to this as a shadow price on
the constraint resource
The shadow price determines the value/worth of
having an additional quantity of a resource
In the previous example, the optimal dual
variables indicate that the worth of another unit
of resource 1 is 1.2 and that of another unit of
resource 2 is 0.2
We start out with

max Z = cT x
s.t.
Ax = b (P)
x ≥ 0
To find (D), we first put (P ) in symmetric form
y+ ↔ Ax ≤ b
⎡ A⎤ x ⎡ b⎤ symmetric
y− ↔ − Ax ≤ − b ⎢− A ⎥ ≤ ⎢− b ⎥
⎣ ⎦ ⎣ ⎦ form
x ≥ 0
29
Let
y = y + − y−
We rewrite the problem as

min W = b T y
s.t .
AT y ≥ c
y is unsigned
The c.s. conditions apply
(
x *T AT y* − c ) = 0
EXAMPLE 5
max Z = x1 − x2 + x3 − x4
s .t .
y1 ↔ x1 + x2 + x3 + x4 = 8
y2 ↔ x1 ≤ 8
y3 ↔ x2 ≤ 4
y4 ↔ − x2 ≤ 4 (P)
y5 ↔ x3 ≤ 4
y6 ↔ − x3 ≤ 2
y7 ↔ x4 ≤ 10
x1 , x4 ≥ 0
x2 , x3 unsigned
30
EXAMPLE 5
min W = 8 y1 + 8 y2 + 4 y3 + 4 y4 + 4 y5 + 2 y6 + 10 y7
s .t .
x1 ↔ y1 + y2 ≥ 1
x2 ↔ y1 + y3 − y4 = −1
x3 ↔ + y5 − y6 = 1
x4 ↔ + y7 ≥ − 1
y2 , ..., y7 ≥ 0
y1 unsigned
( D)
EXAMPLE 5
We are given
⎡ 8 ⎤
⎢− 4 ⎥
x* = ⎢ ⎥
⎢ 4 ⎥
⎢⎣ 0 ⎥⎦
is optimal for (P)

Then the c.s. conditions obtain
x *1 ( y *1 + y 2* − 1) = 0
x *1 = 8 > 0 ⇒ y *1 + y 2* = 1
31
EXAMPLE 5
The other c.s. conditions obtain
⎛ 4 ⎞
y = ⎜ ∑ aij x *j − b i ⎟ = 0
*
i
⎝ j =1 ⎠
Now, x 4 = 0 implies x 4 − 10 < 0 and so
* *
y 7* = 0
Also, x 3* = 4 implies
y 6* = 0
Similarly, the c.s. conditions
⎛ 7 ⎞
x *j = ⎜ ∑ a ji y i* − c j ⎟ = 0
⎝ i =1 ⎠
EXAMPLE 5
have implications on the y i* variable
Since, x 2* = − 4 then, we have
y 3* = 0
Now, with y 7* = 0 we have
y 1* > − 1
Since, x 1* = 8 we have
y 2* = 1 − y 1*
32
EXAMPLE 5
Also
y 1* + y 5* − y 6* = 1
implies
1 + y 5* = 1
and so
y 5* = 0
EXAMPLE 5
Suppose
y 1* = 1
and so,
y 2* = 0
Furthermore,
y 1* + y 3* − y 4* = 1 − y 4* = − 1 ⇒ y 4* = 2
implies
y 4* = 2
33
EXAMPLE 5
Therefore
( )
W y * = ( 8 )( 1) + ( 8 )( 0 ) + ( 4 )( 0 ) + ( 4 )( 2 ) +
( 4 )( 0 ) + ( 2 )( 0 ) + ( 10 )( 0 )
= 16
and so
W * = 16 = Z * ⇔ optimality of ( P ) and ( D )
PRIMAL – DUAL TABLE
primal (maximize) dual (minimize)

A (coefficient matrix) A T (transpose of the coefficient matrix)
b (right-hand side vector) b (cost vector)
c (price vector) c (right hand side vector)
i th constraint is = type the dual variable y i is unrestricted in sign
i th constraint is ≤ type the dual variable y i ≥ 0
i th constraint is ≥ type the dual variable y i ≤ 0
x j is unrestricted j th dual constraint is = type
xj ≥ 0 j th dual constraint is ≥ type
xj ≤ 0 j th dual constraint is ≤ type
34
ECE 307GG – Techniques for
Networks and Flows
George Gross
ECE 307GG © 2005 George Gross, University of Illinois at Urbana-Champaign, All Rights Reserved.
NETWORKS AND FLOWS
A network is a system of lines or channels

connecting different points
Examples abound in nearly all aspects of life:
electrical systems
communication networks
airline web
local area networks
distribution systems
1
NETWORKS AND FLOWS
The network structure is also common to many

other systems that at first glance are not
necessarily viewed as networks
distribution system consisting of
manufacturing plants, warehouses and retail
outlets
matching problems such as work to people,
assignments to machines and computer
dating
NETWORKS AND FLOWS
river systems with pondage for electric

generation
mail delivery networks
project management of multiple tasks in a

large undertaking such as construction or a
space flight
We consider a broad range of network and

network flow problems
2
THE TRANSPORTATION PROBLEM
The basic idea of the transportation problem is
illustrated with the problem of distribution of a
specified homogenous product from several
sources to a number of localities at least cost
We consider a system with m warehouses, n
markets and links between them with specified
costs

supply demand
⎧ a1 b1 ⎫
⎪1 transportation 1 ⎪⎪
⎪
⎪ links with ⎪
⎪ ⎪
a2 b2 ⎪
warehouses
⎪2 costs 2⎪
⎪
markets
⎪
⎪ .. cij .. ⎪
⎬
⎨
⎪
. . ⎪
⎪ ai bj ⎪
⎪i cij = ∞ whenever
j⎪
⎪
⎪
⎪
.. warehouse i cannot
.. ⎪
⎪
. . ⎪
⎪
⎪m am ship to market j bm
⎩ m⎪⎭
3
all the supply comes from the m ware-

houses; we associate the index i with a
warehouse i = 1, 2, …, m
all the demand is at the n markets: we

associate the index j with a market
j = 1, 2, …, n
shipping costs of cij for each unit from

warehouse i to market j
The transportation problem is to determine the
optimal shipping schedule that minimizes
shipping costs for the set of m warehouses to
the set of n markets

4
LP FORMULATION OF THE
The decision variables are
quantity shipped from

xij =
warehouse i to market j
The objective function is
m n
min ∑ ∑ cij xij
i=1 j =1
The constraints are:
n
∑
j =1
xij ≤ ai i = 1, 2,..., m
∑
i =1
xij ≥ bj j = 1, 2,..., n
i = 1, 2,..., m
xij ≥ 0
j = 1, 2,..., n
5
Note that feasibility requires
m n
∑
i =1
ai ≥ ∑
j =1
bj
When
m n
∑ ai =
i =1
∑
j =1
bj
every available unit of supply at the m ware-

houses is shipped to meet all the demands of
the n markets; this problem is known as the
standard transportation problem
STANDARD TRANSPORTATION
PROBLEM
m n
min ∑
i =1
∑
j =1
cij xij
∑
j =1
xij = ai
m i = 1,..., m
∑
i =1
xij = bj
j = 1,..., n
xij ≥ 0
6
EXAMPLE
market j M1 M2 M3 M4 supplies
w/h i
W1 x11 x12 x13 x14 a1

c11 c12 c13 c14
W2 x21 x22 x23 x24 a2

c21 c22 c23 c24
W3 x31 x32 x33 x34 a3

c31 c32 c33 c34
demands b1 b2 b3 b4
PROBLEM
The standard transportation problem has
m n variables xij
m + n equality constraints
Since
m n m n
∑ ∑ xij
i =1 j = 1
= ∑
i =1
ai = ∑
j =1
bj
there are at most m + n – 1 independent

constraints and consequently at most
m + n – 1 independent variables xij
7
EXAMPLE
market j M1 M2 M3 M4 ai
w/h i
W1 3
2 2 2 1
W2 7
10 8 5 4
W3 5
7 6 6 8
bj 4 3 4 4
THE LEAST – COST RULE

PROCEDURE
This scheme is used to generate an initial basic
feasible solution which has no more than
m + n – 1 positive valued basic variables
The key idea of the scheme is to select, at each
step, the variable xij with the lowest shipping
costs cij as the next basic variable

8
APPLICATION OF THE LEAST – COST
RULE
c14 is the lowest cij and we select x14 as a basic
variable
We choose x14 as large as possible without
violating any constraints:
min { a1 , b4 } = min { 3 , 4 } = 3
We set x14 = 3 and
x11 = x12 = x13 = 0
We delete row 1 from further consideration
since all supplies from W1 are exhausted
APPLICATION OF THE LEAST COST

RULE
w/h i
W1 3 3
2 2 2 1
W2 7
10 8 5 4
W3 5
7 6 6 8
bj 4 3 4 4
9
RULE
The remaining demand at M4 is
4–3 =1
which is the value for the modified demand at
M4
We again apply the criterion selection for the
reduced tableau: c24 is the lowest cij is for
i = 2, j = 4 and we select x24 as a basic variable


RULE
We choose x24 as large as possible without
violating any constraints:
min { a2 , b4 } = min { 7 , 1 } = 1
and we set x24 = 1 and
x34 = 0
We delete column 4 from further consideration
since no more demand remains at M4

10
RULE
The remaining supply at W2 is
7–1 =6
which is the value for the modified supply at W2
We repeat these steps until we find the nonzero
basic variables and obtain a basic feasible
solution
For the reduced tableau


RULE
market j M1 M2 M3 ai
w/h i
W2 4 6
10 8 5
W3 0 5
7 6 6
bj 4 3 4
11
RULE
pick x23 to enter the basis
set
x23 = min { 6, 4 } = 4
and set x33 = 0
eliminate column 3 and reduce the supply
at W2 to
6–4 = 2
For the reduced tableau

RULE
market j M1 M2 ai
w/h i
W2 0 2
10 8
W3 3 5
7 6
bj 4 3
12
RULE
set
x32 = min { 3, 5 } = 3
and set x22 = 0
eliminate column 2 in the reduced tableau
and reduce the supply at W3 to
5–3 = 2
The last reduced tableau is

RULE
market j M1 ai
w/h i
W2 2
10
W3 2 2
bj 4
13
RULE
set
x31 = min { 3, 5 } = 3
reduce the demand at M1 to
4–2 = 2
the value of
x21 = 2
is obtained by default
INITIAL BASIC FEASIBLE SOLUTION

w/h i
W1 3 3
2 2 2 1
W2 2 4 1 7
10 8 5 4
W3 2 3 5
7 6 6 8
bj 4 3 4 4
14
RULE
The feasible solution involves only the basic
variables and results in total costs of
3 4
∑ ∑
i =1 j =1
c x
ij ij = 1 ⋅ 3 + 4 ⋅ 1 + 5 ⋅ 4 + 6 ⋅ 3 + 7 ⋅ 2 + 10 ⋅ 2 = 79
THE STANDARD TRANSPORTATION

PROBLEM
The primal problem is
m n
⎫
⎪
min Z = ∑∑
i =1 j =1
cij xij ⎪
⎪
s .t . ⎪
⎪
n
⎪
ui ↔ ∑
j =1
xij = ai ⎬
⎪
(P)
m i = 1, ... , m ⎪
vj ↔ ∑
i =1
xij = bj ⎪
⎪
j = 1, ... , n ⎪
xij ≥ 0 ⎪⎭
15
PROBLEM
The dual problem is
m n ⎫
max W = ∑ ai ui +
i =1
∑ bj v j
j =1
⎪
⎪
⎪
⎪
s.t . ⎪
⎪
⎬ ( D)
xij ↔ ui + v j ≤ cij i = 1, ... , m ⎪
⎪
⎪
j = 1, ... , n ⎪
⎪
ui , v j are unrestricted in sign ⎪⎭

PROBLEM
The complementary slackness conditions are
xij* [ ui* + v *j − cij* ] = 0
And due to the equalities in ( P ) the other
complementary slackness conditions fail to
provide additional information

16
The complementary slackness conditions obtain
xij* > 0 ⇒ ui* + v*j = cij
ui* + v *j < cij ⇒ xij* = 0
We make use of the duality characteristics to
develop the u – v method for solving the
standard transportation problem

THE u-v METHOD
The u-v method starts with a basic feasible
solution for the primal problem, obtains the
corresponding dual variables (as if the solution
were optimal) and uses the duals to determine
the adjacent basic feasible solution; the process
continues until the optimality condition is
satisfied
17
THE u-v METHOD
For a basic feasible solution, we find the dual
variable ui and vj using the complementary
slackness conditions
ui + v j = cij ∀ basic xij
with ui and vj being unrestricted in sign

THE u-v METHOD

We compute using
cij = cij − ( ui + v j ) ∀ nonbasic xij
the step is the analogue of computing c T in the

simplex tableau approach
The complementary-slackness-based optimality

test is performed :
if cij ≥ 0 ∀ nonbasic xij ⎡⎣ xij = 0 ⎤⎦ ,
then the basic feasible solution is optimal
18
THE u-v METHOD
Otherwise, there exists some nonbasic variable
x pq such that
c pq = c pq − ( u p + vq ) < 0
and we determine
c pq =
min

pq ∋ x pq
{c }pq
is nonbasic
We, then, select x pq to become a basic variable
and repeat the process for this new basic
feasible solution
PROBLEM EXAMPLE
We apply the u – v scheme to the example
previously discussed
The basic step from the dual formulation is to
require
ui + v j = cij ∀ basic xij
19
PROBLEM EXAMPLE
We start with the basic feasible solution and
apply the complementary slackness conditions
u1 + v4 = 1 = c14
u2 + v4 = 4 = c24
u2 + v3 = 5 = c23
u3 + v2 = 6 = c32
u3 + v1 = 7 = c31
u2 + v1 = 10 = c21
We have 6 equations in 7 unknowns and so there

is an infinite number of solutions
PROBLEM EXAMPLE
Arbitrarily, we set
v4 = 0
and solve the equations above to obtain
u1 = 1
u2 = 4
v3 = 1
v1 = 6
u3 = 1
v2 = 5
20
PROBLEM EXAMPLE
The cij for the nonbasic variables are
x11 : c11 = c11 − ( u1 + v1 ) = 2 − (1 + 6) = − 5
x12 : c12 = c12 − ( u1 + v2 ) = 2 − (1 + 5) = − 4
x13 : c13 = c13 − ( u1 + v3 ) = 2 − (1 + 1) = 0
x34 : c34 = c34 − ( u3 + v4 ) = 8 − (1 + 0 ) = 7
x33 : c33 = c33 − ( u3 + v3 ) = 6 − (1 + 1) = 4

PROBLEM EXAMPLE
We determine
c pq = min

= c11 = − 5
pq ∋ x pq
is nonbasic
and consequently the nonbasic variable x11 is
introduced into the basis
We determine the maximal value of x11 by
setting x11 = θ from the tableau

21
EXAMPLE
w/h i
W1 θ 3-θ 3
W2 2-θ 4 1+θ 7
W3 2 3 5
bj 4 3 4 4
EXAMPLE
Therefore,
max θ = min { 2, 3 } = 2
Consequently, x21 = 0 and leaves the basis
We obtain the basic feasible solution
x14 = 1, x11 = 2, x31 = 2, x32 = 3, x23 = 4, x24 = 3
and repeat to solve the u – v problem for this
new basic feasible solution

22
EXAMPLE
market j v1 = 2 v2 = 1 v3 = 2 v4 = 1 ai
w/h i
u1 = 0 2 1 3
2 2 2 1
u2 = 3 4 3 7
10 8 5 4
u3 = 5 2 3 5
7 6 6 8
bj 4 3 4 4
EXAMPLE
The complementary slackness conditions of the
nonzero valued basic variables obtain
u1 + v1 = c11 = 2
u1 + v4 = c14 = 1
u2 + v3 = c23 = 5
u2 + v4 = c24 = 4
u3 + v1 = c31 = 7
u3 + v2 = c32 = 6
23
EXAMPLE
We set
u1 = 0
and therefore
v3 = 2 v1 = 2
u3 = 5 u3 = 5
v2 = 1 v2 = 0
We compute cij for each nonbasic variable xij

EXAMPLE
c12 = c12 − ( u1 + v2 ) = 2 − (0 + 1) = 1
c13 = c13 − ( u1 + v3 ) = 2 − (0 + 2) = 0
c21 = c21 − ( u2 + v1 ) = 10 − (3 + 2) = 5
c22 = c22 − ( u2 + v2 ) = 8 − (3 + 1) = 4
c33 = c33 − ( u3 + v3 ) = 6 − (5 + 2) = −1
c34 = c34 − ( u3 + v4 ) = 8 − (5 + 1) = 2
only possible improvement

We introduce x33 as a basic variable and
determine its nonnegative value θ from the
tableau
24
EXAMPLE
w/h i
W1 3
2+θ 1–θ
W2 7
4–θ 3+θ
W3 5
2–θ 3 θ
bj 4 3 4 4
EXAMPLE
The limiting value of θ is
θ = min { 2, 4, 1 } = 1
Consequently, x14 leaves the basis and x33
enters the basis with the value 1
We obtain the adjacent basic feasible solution in

25
EXAMPLE
market j v1 = 2 v2 = 1 v3 = 1 v4 = 0 ai
w/h i
u1 = 0 3 3
2 2 2 1
u2 = 4 3 4 7
10 8 5 4
u3 = 5 1 3 1 5
7 6 6 8
bj 4 3 4 4
EXAMPLE
We evaluate cij for each nonbasic variable;
cij < 0 and so we have an optimal solution with
shipping 3 from W1 to M 1 with costs 6
shipping 1 from W 3 to M 1 with costs 7
and resulting in the least total costs of 68
26
ELECTRICITY DISTRIBUTION
EXAMPLE
We consider in an electric utility system in which
3 power plants are used to supply the demand of
4 cities
The supplies available from the 3 plants are given
The demands of the 4 cities are specified
The costs of supplying each 10 6 kWh are given

ELECTRICITY COSTS
to city supplies
from (10 6 kWh)
1 2 3 4
1 8 6 10 9 35
balanced
2 9 12 13 7 50
plant transportation
3 problem
14 9 16 5 40
demands 45 20 30 30 125
(10 6 kWh)
27
ELECTRICITY ALLOCATION EXAMPLE
We note that
3 4
∑
i =1
ai = ∑
j =1
bj
and so we have a balanced transportation
problem
We find a basic feasible solution using the
least-cost rule
ELECTRICITY ALLOCATION EXAMPLE:

SOLUTION
to city supplies
from (10 6 kWh)
1 2 3 4
1 0 35
8 6 10 9
2 0 50
plant
9 12 13 7
3 30 10
40
14 9 16 5
demands 45 20 30 30 125
(10 6 kWh)
28
SOLUTION
And we set
x34 = 30 , x14 = x24 = 0
and remove column 4 from further consideration
We continue with the reduced system

SOLUTION
to city supplies
from (10 6 kWh)
1 2 3
1 20 15
35
8 6 10
2 0 50
plant
9 12 13
3 0 10
14 9 16
demands 45 20 30
(10 6 kWh)
29
SOLUTION
and so we set
x12 = 20 , x22 = x32 = 0
The new reduced system obtains

SOLUTION
to city supplies
from (10 6 kWh)
1 3
1 15 0 15
8 10
2 50
plant
9 13
3 10
14 16
demands 30
45 30
(10 6 kWh)
30
SOLUTION
and therefore we set
x11 = 15 , x13 = 0
and remove row 1 from further consideration
The operation is repeated for the further reduced
system

SOLUTION
to city supplies
from (10 6 kWh)
1 3
2 30 20
50
9 13
plant 3 10
0
14 16
demands 30 30
(10 6 kWh)
31
SOLUTION
and therefore we set
x21 = 30 , x31 = 0
We are finally left with


SOLUTION
to city supplies
(10 6 kWh)
from 3
2 20 20
13
plant 3 10
10
16
demands 30
(10 6 kWh)
32
SOLUTION
which allows us to set
x23 = 20 , x33 = 10
The basic feasible solution has the costs
Z = 30 · 5 + 20 · 6 + 15 · 8 + 30 · 9 + 20 · 13 + 10 · 16 = 1080
We improve this solution by using the u - v

scheme
The first tableau corresponding to the initial

basic feasible solution is:

SOLUTION
to city supplies
from
3 4 (10 6 kWh)
1 2
1 15 20 35
8 6
plant
2 30 20 50
9 13
3 10 30 40
16 5
demands 45 20 30 30
(10 6 kWh)
33
EXAMPLE
We compute, the possible improvements at
each nonbasic variable:
c31 = c31 − ( u3 + v1 ) = 14 − (4 + 8) = 2
c22 = c22 − ( u2 + v2 ) = 12 − (1 + 6) = 5
c32 = c32 − ( u3 + v2 ) = 9 − (4 + 6) = −1
c13 = c13 − ( u1 + v3 ) = 10 − (0 + 12) = −2
c14 = c14 − ( u1 + v4 ) = 9 − (0 + 1) = 8
c24 = c24 − ( u2 + v4 ) = 7 − (1 + 1) = 5
improvement possible
better possibility
EXAMPLE
We bring x13 into the basis and determine the
value of θ using the tableau structure
From the tableau we conclude that
θ = min { 15, 20 } = 15
and therefore x11 leaves the basis

34
EXAMPLE
cities 1 2 3 4 ai
plants
1 15 − θ 20 θ 35
2 30 + θ 20 – θ 50
3 10 30 40
bj 45 20 30 30
EXAMPLE
x21 = 45, x12 = 20, x13 = 15, x23 = 5, x33 = 10, x34 = 30
and the new value of Z is
Z = 20 · 6 + 15 · 10 + 45 · 9 + 5 · 13 + 10 · 16 + 30 · 5
= 1050 < 1080
We again pursue u-v improvement and start with

the tableau
35
EXAMPLE
cities v1 = 6 v2 = 6 v3 = 10 v4 = -1 supplies
plants
20 15 35
u1 = 0
6 10
u2 = 3 45 5 50
9 13
u3 = 6 10 30 40
16 5
demands 45 20 30 30
EXAMPLE
The complementary slackness conditions
obtain the possible improvements
c11 = c11 − ( u1 + v1 ) = 8 − (0 + 6) = 2
c31 = c31 − ( u3 + v1 ) = 14 − (6 + 6) = 2
c22 = c22 − ( u2 + v2 ) = 12 − (3 + 6) = 3
c32 = c32 − ( u3 + v2 ) = 9 − (6 + 6) = −3
c14 = c14 − ( u1 + v4 ) = 9 − (0 − 1) = 10
c24 = c24 − ( u2 + v4 ) = 7 − (3 − 1) = 5
only possible improvement
We bring x32 into the basis and determine its
value θ using
36
EXAMPLE
plants 1 2 3 4 ai
cities
1 20 – θ 15 + θ 35
2 45 5 50
3 θ 10 – θ 30 40
bj 45 20 30 30
EXAMPLE
and so
θ = min { 10, 20 } = 10
x21 = 45 x12 = 10 x32 = 10
x13 = 25 x23 = 5 x34 = 30
and the value of Z becomes
Z = 45 · 9 + 10 · 6 + 10 · 9 + 25 · 10 + 5 · 13 30 · 5 =1020
You are asked to prove, using complementary
slackness conditions, that this is the optimal
37
NONSTANDARD TRANSPORTATION
PROBLEM
The nonstandard transportation problem arises
when supply and demand are unbalanced: either
supply exceeds demand or vice versa
The approach is to create a fictitious entity and
thereby restore the problem to balanced status

PROBLEM
For the case
m n
∑
i =1
ai > ∑
j =1
bj

supply demand
we create a fictitious market Mn+1 to absorb all
⎛ m n ⎞
the excess supply ⎜ ∑ ai − ∑ b j ⎟⎟ ;we set ci n+1 = 0
⎜
⎝ i =1 j =1 ⎠
∀ i = 1, 2,..., m since M does not exist in reality
n+1
and the problem is then in standard form with

j = 1,ECE… , n+1
307GG © 2005 George Gross, University of Illinois at Urbana-Champaign, All Rights Reserved.
38
PROBLEM
For the case
n m
∑
j =1
bj > ∑
i =1
ai

demand supply
the problem is not , in effect, feasible since all
the demands cannot be met and therefore the
least-cost shipping schedule is that which will
supply as much as possible of the demands of
the markets
PROBLEM
For the overdemand case, we introduce the
fictitious warehouse Wm+1 to supply the
⎡ n m ⎤
shortage ⎢ ∑ b j − ∑ ai ⎥ ; we set cm+1, j = 0
⎢⎣ j =1 i =1 ⎥ ⎦
for j = 1, 2, …, n and the problem is in standard
form with i = 1,…, m + 1

39
PROBLEM
Note that the variable xm+1, j is the shortage at
market j and is the shortfall in the demand bj
experienced by the market Mj due to inadequate
supplies
For each market j , xm+1, j is a measure of the
infeasibility of the problem

EXAMPLE: CANNING OPERATIONS

SCHEDULING
This problem is concerned with the schedule of
2 plants A and B in the purchase of the raw
supplies from 3 growers
grower availability (ton) price ( $ / ton )
Smith 200 10
Jones 300 9
Richard 400 8
40
SCHEDULING
and shipping costs in $ / ton given by
to plant
from A B
Smith 2 2.5
Jones 1 1.5
Richard 5 3

SCHEDULING
The plants’ labor costs and capacity limits are
plant A B
capacity 450 550

( ton )
labor costs 25 20
( $ / ton )
41
SCHEDULING
The selling price for canned goods is 50 $ / ton
and the company can sell all it produces
The problem is to determine the maximum profit
schedule
Note that this is an unbalanced problem since
supply = 200 + 300 + 400 = 900 tons
demand = 450 + 550 = 1000 tons > 900 tons
Clearly, the decision variables are the amounts
purchased from each grower and shipped to
each plant
EXAMPLE: CANING OPERATIONS

SCHEDULING
The objectives is formulated as
⎡ ⎤ ⎡ ⎤
max Z = ⎢ 50 − 25 − 10 − 2 ⎥ x ⎢ − 9 −
1 ⎥ xJA
⎢
⎥ SA + ⎢
50 − 25
⎥
⎣ 13 ⎦ ⎣ 15 ⎦
⎡ ⎤ ⎡ ⎤
+ ⎢
50 − 25 5 ⎥ x RA
− 8 −
+ ⎢
50 − 20 − 10 − 2.5

⎥

⎥ x SB
⎢ ⎥ ⎢
⎣ 12 ⎦ ⎣ 17.5 ⎦
⎡ ⎤ ⎡ ⎤
⎢
+
50 − 20 − 9 − 1.5 ⎥ ⎢ − 8 −
3 ⎥ x RB
⎢
⎥ xJB + ⎢
50 − 20
⎥
⎣ 19.5 ⎦ ⎣ 19 ⎦
42
SCHEDULING
The supply constraints are
x SA + x SB ≤ 200
xJA + xJB ≤ 300
x RA + x RB ≤ 400
The demand constraints are
x SA + xJA + x RA ≤ 450
x SB + xJA + x RB ≤ 550

SCHEDULING
Clearly, all decision variables are nonnegative
The unbalanced nature of the problem requires
the introduction of a fictitious grower F with the
supply 100 corresponding to the supply shortage
We set up the standard transportation problem

43
SCHEDULING
plant j A B supply
grower i
S 200
13 17.5
J 300
15 19.5
R 400
12 19
F 100
0 0
demand 450 550

SCHEDULING
Please note that the objective is a maximization
rather than a minimization
As a homework exercise, show that the duality
complementary slackness conditions allow us to
change the u-v algorithm in the following way:

44
SCHEDULING
the selection of the nonbasic variable xij to
enter the basis is from those xij where the
corresponding
cij > ui + v j
and we evaluate and focus on all cij > 0 so
that xij is a candidate to enter the basis
we pick xpq
c pq = max

pq ∋ x pq
{c }pq
is nonbasic
EXAMPLE SOLUTION
plant j A B supply
grower i
S 200 0 200
13
J 250 50 300
15 19.5
R 0 400 400
19
F 0 100 100
0 0
demand 450 550
45
EXAMPLE SOLUTION
We construct the u - v relations for this solution
u1 + v1 = 13 u2 + v2 = 19.5
u2 + v1 = 15 u3 + v2 = 19
u4 + v2 = 0
We arbitrarily set u1 = 0 and compute
v1 = 13, u2 = 2, v2 = 17.5, u3 = 1.5, u4 = – 17.5

EXAMPLE SOLUTION
We evaluate the cij corresponding to the
nonbasic variables
c31 = c31 − ( u3 + v1 ) = 12 − (1.5 + 13) = − 2.5
c41 = c41 − ( u4 + v1 ) = 0 − ( − 17.5 + 13) = 4.5
c12 = c12 − ( u1 + v2 ) = 17.5 − (0 + 17.5) = 0

single possible improvement
Thus, x41 enters the basis and we determine θ
46
EXAMPLE SOLUTION
plant j A B supply
grower i
S 200 200
13
J 250 – θ 50 + θ 300
15 19.5
R 400 400
19
F θ 100 – θ 100
0 0
demand 450 550
EXAMPLE SOLUTION
It follows that
θ = min { 250, 100 } = 100
and so the adjacent basic feasible solution is
x11 = 200, x21 = 150, x41 = 100, x22 = 150, x32 = 400
We repeat the u - v procedure to obtain

47
EXAMPLE SOLUTION
u1 + v1 = 13 u2 + v2 = 19.5
u2 + v1 = 15 u3 + v2 = 19
u4 + v1 = 0
We solve by arbitrarily setting u1 = 0 and obtain
v1 = 13 , u2 = 2 , v2 = 17.5 , u3 = 1.5 , u4 = – 13
EXAMPLE SOLUTION
We compute the cij for the nonbasic variables
c12 = 17.5 − ( 0 + 17.5) = 0
c31 = 12 − (1.5 + 13) = − 2.5
c42 = 0 − ( − 13 + 17.5) = − 4.5
48
EXAMPLE SOLUTION
Since each cij is ≤ 0 no more improvement in
the maximization is possible and so the
maximum profits are
Z = (200)13 + (150)15 + (100)0 +(150)19.5 + (400)19
= 15,375 $
SCHEDULING PROBLEM AS A STANDARD

The problem is concerned with the weekly
production scheduling over a 4-week period
production costs from each item
first two weeks $ 10
last two weeks $ 15
demands that need to be met are

week 1 2 3 4
demand 300 700 900 800
49
weekly plant capacity is 700
overtime is possible for weeks 2 and 3 with
- the production of additional 200 units
- additional cost per unit of $ 5
$ 3 for weekly storage of excess production
the objective is to minimize the total costs for
the 4 week schedule
The decision variables are
xij = production in week i for use in week j market

mkt. 1 2 3 4 F supply
production
M is a very large number
1 700
10 13 16 19 0
normal 700
M 10 13 16 0
2 o/t 200
M 15 18 21
3200
normal 700
M M 15 18 0
3 o/t 200
M M 203200 - 23 0
2700 2700
4 700
M M M 15 0
demand 300 700 900 800 500
50
ASSIGNMENT PROBLEM
We are given
n machines M1 , M2 , … , Mn ↔ i
n jobs J 1 , J2 , … , Jn ↔ j
cij = cost of doing job j on machine i
cij = M if job j cannot be done on machine i
each machine can only do one job and we wish
to determine the optimal match, i.e., the
assignment with the lowest total costs of doing
all the jobs j on the n machines
ASSIGNMENT PROBLEM
The brute force approach is simply enumeration:
consider n = 10 and there are 3, 628, 800 possible
choices
We can, however, introduce categorical decision
variables
⎧1 job j is assigned to machine i
xij = ⎪⎨
⎪⎩0 otherwise
51
ASSIGNMENT PROBLEM
and the problem constraints can be stated as
n
∑
j =1
xij = 1 ∀ i each machine does exactly 1 job
n
∑
i =1
xij = 1 ∀ j each job is assigned
to only 1 machine
The objective, then, is

n n
min Z = ∑∑ cij xij
i =1 j =1
ASSIGNMENT PROBLEM
This assignment problem is a standard
transportation problem with
ai = 1 ∀ i
bj = 1 ∀ j
n n
∑ ai =
i =1
∑
j =1
bj
52
NONSTANDARD ASSIGNMENT
PROBLEM
Suppose we have m machines and n jobs with
m ≠ n
We may convert this into an equivalent standard
assignment problem with equal number of
machines and jobs
The conversion requires the introduction of
either fictitious jobs or fictitious machines

PROBLEM
In the case m > n :
we create ( m – n ) fictitious jobs and we have

m machines and n + m – n = m jobs; we
assign the machinery costs for the fictitious
goods to be 0 : note that there is no change
in the objective function since a fictitious job
assigned to a machine is, in effect, a
machine that is idle
53
PROBLEM
For the case n > m :
we create ( n – m ) fictitious machines with
machine costs of 0 and the solution
obtained has the ( n – m ) jobs that cannot be
done due to lack of machines

PROBLEM
In principle, any assignment problem may be
solved using the transportation problem
technique; in practice , this is not good since
there exists degeneracy in every basic feasible
solution
We note that in the standard assignment
problem for m machines with m = n , there are
exactly m xij that are 1(nonzero) but every basic
feasible solution of the transportation problem
has (2m – 1) basic variables and therefore
contains (m – 1) zero valued basic variables
54
Transshipment and Shortest Path Problems
George Gross
ECE 307© 2006 George Gross, University of Illinois at Urbana-Champaign, All Rights Reserved.
TRANSSHIPMENT PROBLEMS
We consider the shipment of a certain homo-

geneous commodity from a specified point or
source to a particular destination or sink
In general, the source and the sink need not be
directly connected; rather, the flow goes through
the transshipment points or intermediate nodes
The objective is to determine the maximal flow
from the source to the sink
1
FLOW NETWORK EXAMPLE
1
4
s 3 t
2 5
TRANSSHIPMENT PROBLEMS
nodes 1, 2, 3, 4 and 5 are the transshipment

points
arcs of the network are ( s, 1 ), ( s, 2 ), ( 1, 2 ),
( 1, 3 ), ( 2, 5 ), ( 3, 4 ), ( 3, 5 ),( 4, 5 ), ( 5, 4 ) ( 4, t ),
( 5, t ) ; the existence of an arc from 4 to 5 and
from 5 to 4 allows bidirectional flows between
the two nodes
each arc may be constrained in terms of a
limit on the flow through the arc
2
MAX FLOW PROBLEM
We denote by fij the flow from i to j and this

equals the commodity amount shipped from i
to j on an arc (i , j) that directly connects the
nodes i and j
The problem is to determine the maximal flow f
from s to t taking into account the flow limits kij
on each arc (i , j)
The mathematical statement of the problem is
MAX FLOW PROBLEM
max Z = f
s.t .
0 ≤ f ij ≤ kij ∀ arc ( i , j ) that connects
nodes i and j
f = ∑fi
si at source s conservation
of flow
∑i f it = f at sink t relations
⎫⎪
∑i f ij = ∑k f jk ⎬ at each transshipment node j
⎪⎭
3
MAX FLOW PROBLEM
While the simplex approach can solve the max
flow problem, it is possible to construct a highly
efficient network method to find f directly
NETWORK DEFINITIONS
Each arc is directed and so for an arc (i, j), fij ≥ 0

A forward arc at a node i is one that leaves
node i to some node j and is denoted by (i, j)
A backward arc at node i is one that enters
node i from some node j and is denoted by (j, i)
A path connecting node i to node j is a
sequence of arcs starting at node i and
terminating at node j
4
NETWORK DEFINITION
we denote a path by
P = { ( i, k ), ( k, l ), …, ( m, j ) }
in the example network
( 1, 2 ), ( 2, 5 ), ( 5, 4 ) is a path from 1 to 4
( 1, 3 ), ( 3, 4 ) is another path from 1 to 4
A cycle is a path with i = j , i.e.,
P = { ( i, k ), ( k, l ), …, ( m, i ) }
NETWORK DEFINITION
We denote the set of nodes of the network by N
the definition is
N = { i : i is a node of the network }
In the example network
N = { s , 1, 2, 3, 4, 5, t }
5
NETWORK CUT
A cut is a partitioning of nodes into two distinct

subsets S and T with
N = S ∪ T and S ∩ T = φ
We are interested in cuts with the property that
s ∈ S and t ∈ T
the sets S and T are said to provide an s - t
cut
in the example network,
S = { s, 1, 2 } and T = { 3, 4, 5, t }
provide an s - t cut
NETWORK CUT
The capacity of a cut is

Κ (S ,T ) = ∑ kij
i ∈S
j∈T
In the example network with
S = { s, 1, 2 } and T = { 3, 4, 5, t }
we have
Κ (S , T ) = k13 + k25
but for the cut with
S = {s, 1, 2, 3, 4 } and T = { 5, t }
6
NETWORK CUT
Κ (S ,T ) = k4, t + k4, 5 + k3, 5 + k2, 5
Note : arc (5, 4) is directed from a node in T to a
node in S and is not included in the summation
An important characteristic of the s - t cuts of
interest is that if all the arcs in the cut are

removed then no path exists from s to t
Consequently no flow is possible since any flow

from s to t must go through the arcs in a cut and
thus the flow is limited by the capacity of the cut
NETWORK CUT LEMMA
Lemma: for any directed network the flow f
from s to t is constrained by an s to t cut
f ≤ Κ (S , T ) for any cut S , T
Clearly, this lemma implies that
max flow ≤ Κ (S , T ) ∀ S , T
and consequently
max flow ≤ min Κ (S , T )

S ,T
7
MAX FLOW MIN CUT THEOREM
For any network, the value of the maximal flow
from s to t is equal to the minimal cut, i.e., the

cut S , T with the smallest capacity
The max-flow min-cut theorem allows us, in
principle, to find the maximal flow in a network
by finding the capacities of all the cuts and
choosing the minimum capacity cut

MAX FLOW
The maximal flow algorithm is based on finding

a path through which a positive flow from s to t
can be sent, the so called flow augmenting
path; the procedure is continued until no such
flow augmenting path can be found and
therefore we have the maximal flow
The labeling procedure is used to find a flow

augmenting path from s to t
8
LABELING PROCEDURE
Step 0 : start with node s

Step 1 : label node j : node j can be labeled if
and only if flow can be sent from s to t ;
we label node j given that node i is
labeled only if
(i) either there exists an arc (i, j) and
fij < kij
(ii) or there exists an arc ( j, i ) and
fji < 0
Step 2 : if j = t , stop; else, go to Step 1
THE MAX FLOW ALGORITHM

Step 0 : start with a feasible flow
Step 1 : use the labeling procedure to find a flow
augmenting path
Step 2 : determine the maximum value δ for the
max increase (decrease) of flow on all
forward (backward) arcs
Step 3 : use the labeling procedure to find a flow
augmenting path; if no such path exists,
stop; else, go to Step 2
9
EXAMPLE APPLICATION
Consider the simple network with the flow

capacities on each arc indicated
7 1 9
s 3 t
9 8
2
EXAMPLE APPLICATION
We initialize the network with a flow 1

f = 1
(1,7) 1 (0, 9)
s (1, 3) t
(0, 9) (1,8)
2
10
EXAMPLE APPLICATION
We apply the labeling procedure

f = min { 6, 2, 7 } = 2
6 1
s 2 t
7
2
EXAMPLE APPLICATION
Consider the simple network with the flow

capacities on each arc indicated f = 3
(3,7) 1 (0, 9)
s (3, 3) t
(0, 9) (3, 8)
2
11
EXAMPLE APPLICATION
We repeat application of the labeling procedure
f = min { 5, 9 } = 5
s t
9 5
2
EXAMPLE APPLICATION
We increase the flow by 5

f = 8
(3,7) 1 (0, 9)
s t
(3, 3)
(5, 9) (8, 8)
2
12
EXAMPLE APPLICATION
f = min { 4, 9 } = 4
4 1 9
s t
EXAMPLE APPLICATION

f = 12
(7,7) 1 (4, 9)
s t
(3, 3)
(5, 9) (8, 8)
2
13
EXAMPLE APPLICATION
f = min { 4, 3, 5 } = 3
1 5
s t
3
4
2
EXAMPLE APPLICATION

f = 15
(7,7) 1 (7, 9)
s (0, 3)
t
(8, 9) (8, 8)
2
no possibility of flow augmenting path

14
UNDIRECTED NETWORKS
A network with undirected arcs is called an
undirected network: the flows on each arc ( i, j )
with the limit kij cannot violate the capacity
constraints in either direction
Mathematically, we require
f ij ≤ kij ⎫ interpretation of
⎪
f ji ≤ k ji ⎪⎬ unidirectional flow below
⎪
f ij f ji = 0 ⎪⎭ capacity limit
EXAMPLES OF AN UNDIRECTED
NETWORK WITH ARCS
Consider the network with the three unrestricted
arcs
30
1 3
40 50
s 15 20 25 t
30 30
2 4
50
15
EXAMPLE OF A NETWORK WITH
UNDIRECTED ARCS
To make the problems realistic, we assume that
the numbers represent traffic flow capacities:
the directed arcs correspond to unidirectional
streets and the problem is to place one-way
signs on each street ( i, j ) not already directed
so as to maximize traffic flow from s to t
The procedure is to replace each undirected arc

by two directed arcs ( i, j ) and ( j, i ) to deter-
mine the maximal s – t flow

UNDIRECTED ARCS
Consider the network with the three unrestricted
arcs
1 30 3
40 50
s 15 15 20
25 25 t
20
30 30
2 4
50
16
UNDIRECTED ARCS
We apply the max flow scheme to the directed
network and give the following interpretations to
the flows on the max flow bidirectional arcs that
are the initially undirected arcs ( i, j ) : if
fij > 0 , fji > 0 and fij > fji
set up the flow from i to j with value fij – fji
and remove the arc ( j, i )
The computation of the max flow f for this
example is left as a homework exercise

UNDIRECTED ARCS
(30 ,30)
30
1 3 (30
(40 ,50)
50
(40
(30 ,40)
40
20
s 15 15 25 (10 ,25)
25 t
20
(10 ,20)
30
(30 ,30) (30 ,30)
30
2 4
50
(30 ,50)
17
UNDIRECTED ARCS
flow: s 1 3 n = 30
flow: s 2 4 n = 30
flow: s 1 4 3 t = 10
and so the maximum flow is 30 + 30 + 10 = 70

one way signs should be put from 1 4 and 4 3
an alternative routing of a flow of 10 is the path
s 1 2 4 3 t which would require one
way signs from 1 2 and 4 3
NETWORKS WITH MULTIPLE

SOURCES AND MULTIPLE SINKS
We next consider a network with several supply
and several demand points
We introduce a super source ŝ linking to all the
sources and a super sink t̂ linking all the sinks
We can consequently apply the max flow
algorithm to the modified network

18
NETWORKS WITH MULTIPLE
SOURCES AND MULTIPLE SINKS
s1 t1
n
e
t
ŝ s2 w t2 t̂
o
. . .
. . .
r
sm k
tn
MULTIPLE SOURCE / MULTIPLE SINK

NETWORK EXAMPLE
15 sink with
2 5 demand 15
10 sink with
5
10 5 demand 20
1 3 5 6 10 8
20
5 5
sources 10 10
with
supply 20 4 5 7
19
NETWORK EXAMPLE
2 15 5
super sink
node t̂
10 15
5 t̂
10 5
20 20
ŝ 1 3 5 6 10 8
20
20 5
5 10 10
super source
node ŝ 5
4 7

NETWORK EXAMPLE
The transshipment problem is feasible if and
only if the maximal sˆ − t̂ flow f satisfies
f = ∑
sinks
demands
Show that the transshipment problem is infeas-
ible since the network cannot accommodate the
total demand of 35 ; also determine the small-
est shortage for this problem

20
NETWORK EXAMPLE
2 15 5 f
15
10 t̂
5
10 5
20
1 3 5 6 10 8
20
20 5 5
10 10
ŝ
20 4 5 7
f

NETWORK EXAMPLE
The minimum cut is shown and has capacity
15 + 5 + 5 + 5 = 30
and the maximum flow is, therefore, 30
Since the maximum flow fails to meet the total
demand of 35 units, the problem is infeasible;
the minimum shortage is 5

21
APPLICATION TO MANPOWER
SCHEDULING
Consider the case of a company that must comp-
lete 4 projects in 6 months
manpower
earliest start latest finish
project requirements
month month
( person/month )
A 1 4 6
B 1 6 8
C 2 5 3
D 1 6 4
SCHEDULING
There are the following additional constraints:
the company has only 4 engineers
at most 2 engineers may be assigned to any

one project in a given month
no engineer maybe assigned to more than

one project at any time
The question is whether there is a feasible

assignment
22
SCHEDULING
The solution approach is to setup the problem
as a transshipment network
the sources are the 6 months
the sinks are the 4 projects
arcs ( i, j ) are introduced whenever a
feasible assignment of the engineers in
month i can be made to project j ; each arc
has capacity
kij = 2 i = 1, 2, …, 6 , j = A, B, C, D
there is no arc ( 1, C ) since project C cannot
start before month 2
SCHEDULING
4 1
A 6
4 2
8
man / months
B
man / months
demands
4 3
supply
4 4 C 3
4 5
D 4
4 6 each arc has capacity 2
23
SCHEDULING
The transshipment problem is feasible if the total
demand 6 + 8 + 3 + 4 = 21 can be met
As a homework problem, determine whether a
feasible schedule exists and if so, find it

SCHEDULING
1 (2, 2)
(4, 4) (2, 2) (2, 2) A (6, 6)
(4, 4)
2 (2, 2)
(2, 2)
(4, 4)
(2, 2) B (8, 8)
3
ŝ (2, 2) t̂
(4, 4) 4 (2, 2) (1, 2) C (3, 3)
(3, 4)
5 (2, 2)
(2, 4)
D (4, 4)
6 (2, 2)
24
SCHEDULING
(2, 2)
1 (6, 6)
(4, 4) (2, 2) (2, 2) A
21
(4, 4)
2 (2, 2)
(2, 2) (8, 8)
(4, 4)
(2, 2) B
3
ŝ (2, 2) t̂
(3, 3)
(4, 4) 4 (2, 2) C
(1, 2) 21
(3, 4)
5 (2, 2)
(4, 4)
(2, 4)
D
6 (2, 2)
SHORTEST ROUTE PROBLEM

The problem is to determine the shortest path
from s to t for a network with a set of nodes
N = { s = 1, 2, …, n = t }
and arcs ( i, j ) where for each arc ( i, j ) of the
network
dij = distance or transit time
The determination of the shortest path from 1 to
n requires the specification of the path
{ ( 1, i1 ) , ( i1, i2 ) , …, ( iq , n ) }
25
SHORTEST ROUTE PROBLEM
We can write an LP formulation of this problem in

the form of a transshipment problem:
ship 1 unit from node 1 to node n by

minimizing the shipping costs given the
parameters
dij = shipping costs for 1 unit from i to j ;

for i and j are not directly connected,
dij = ∞
DIJKSTRA’S ALGORITHM
The solution is very efficiently performed using
Dijkstra’s algorithm
The assumptions are
dij is given for each pair of nodes
dij ≥ 0
The scheme is basically a label assignment

procedure, which assigns nodes with either a
permanent or a temporary label
26
The temporary label of node i is the upper
bound on the shortest distance from node 1 to
node i
The permanent label is the actual shortest
distance from node 1 to node i

Step 0 : assign the permanent label 0 to node 1
Step 1 : assign the temporary labels to all the other
nodes
d1j if node j is directly connected to
node 1
∞ if node j is not directly connected
to node 1
and select the minimum of the temporary
labels and declare it permanent ; in case
of ties, the choice is arbitrary
27
Step 2 : let A be the node most recently assigned
a permanent label and consider each node
j with a temporary label; recompute the
label to be
⎪⎧ temporary label permanent label ⎪⎫
min ⎨ , + d Aj ⎬
⎪⎩ of node j of node A ⎪⎭
Step 3 : select the smallest of the temporary labels

and declare it permanent ; in case of ties,
the choice is arbitrary
Step 4 : if this is node n , stop; else, go to Step 2
The shortest path is obtained by retracing the
sequence of permanently labeled nodes back
from n to the node 1
The path is then given in the forward direction
28
EXAMPLE : SHORTEST PATH
Consider the undirected network

9
2 6
3 6
2
1 7 3 3
1 3
4
4 5
3
The problem is to
find the shortest path from 1 to 6
compute the length of the shortest path
We apply the Dijkstra algorithm and assign
successively a permanent label to each node

29
0
3,7,4,∞ ,∞ ⎤⎦
Steps 0 and 1 : L (0 ) = ⎡⎣0 ,,3,7,4,
1
Steps 2 : L (1) = 3, 5, 4, ∞ ,12 ⎤⎦

⎡0 ,,3,5,4,
⎣
2
Steps 2,3 and 4 : L (2) = ⎡0 ,3,5,4,7,12

⎣
, 3, 5, 4,7,12⎤⎦
3
Steps 2,3 and 4 : L (3) = ⎡0 ,,3,5,4,7,11

⎣
3, 5, 4, 7,11⎤⎦
4
Steps 2,3 and 4 : L (4) = ⎡0 ,,3,5,4,7,10

⎣
3, 5, 4, 7,10⎤⎦
5
L (4) = ⎡0,3,5,4,7,10 ⎤
⎣ ⎦
6
30
9
2 6 5
3 1 6
2
1 7 3 3 3
0
1 3
4 2
4 5 4
3
The shortest distance is 10 and we get the path
{ ( 1, 4 ) , ( 4, 5 ) , ( 5, 6 ) }
PATH RETRACING
We retrace the path from n back to 0 using the
scheme:
pick node j preceding node n as the node

with the property
permanent label of shortest

+ djn =
node j distance
In the retracing scheme, certain nodes may be

skipped
31
SHORTEST PATH BETWEEN ANY TWO
NODES
Dijkstra’s algorithm may be applied to
determine the shortest distance between any
pair of nodes i , j by taking i as the source
and j as the sink
Consider as an example the following five node
network
EXAMPLE : PATH RETRACING
4 8 10
0 1 2 3 4
3 3 2 3
4 4 8
32
L (0 ) = 3,4,8,10,∞ ⎤⎦
⎡0 ,,3,4,8,10,
⎣
0
L (1) = ⎡0 ,,3,4,7,10
⎣
3, 4,7,10 ⎤⎦
1
L (2) = ⎡0 ,3,4,6,8
⎣
, 3, 4, 6,8⎤⎦
2
L (3) = ⎡0 ,,3,4,6,8
⎣
3, 4, 6, 8 ⎤⎦
3
Retracing the path we get
8 = 4
O + d 24

node 2 4
and so the path is

0 →2→4
gives the shortest distance from 0 to every other
node
33
EXAMPLE : PATH RETRACING
1 4 8 3 10
2
0 1 2 3 4 4
3 3 2 3
4 4 8
0
APPPLICATION : EQUIPMENT
REPLACEMENT PROBLEM
We consider the problem of replacing old
equipment or continuing its maintenance
As equipment ages, the level of maintenance

required increases and typically, this results in
increased operating costs
O&M costs may be reduced by replacing aging

equipment; however this increases capital
investment costs
34
REPLACEMENT PROBLEM
The problem is how often to replace equipment
in order to minimize
total capital O&M

= +
costs costs costs
fixed variable
EQUIPMENT REPLACEMENT
PROBLEM
Equipment replacement is planned during the
next 5 years
The cost elements are
pj = purchase costs in year j
sj = salvage value of original
equipment after j years of use
cj = O&M costs in year j of operation
of equipment with the property that
… cj < cj+1 < cj+2 < …
We formulate the equipment replacement
problem as a shortest route problem on a
directed network
35
EQUIPMENT REPLACEMENT PROBLEM
end of
planning
d16
d15 period
d14 d36
d13 d35
1 d12
2 d23
3 4 5 6
d34 d45 d56
d24 d46
d25
beginning of d26
planning period
REPLACEMENT PROBLEM
where, the “distances” dij are defined to be

finite if i < j , i.e., year i precedes the year j ,
with
j−i
d ij = pi − s j−i + ∑ cτ
τ =1
j > i

purchase salvage value O&M costs

price in after j – i for j - i years
year i years of use of operation
36
REPLACEMENT PROBLEM
For example, if the purchase is made in year 1
5
d16 = p1 − s5 + ∑c
τ =1
τ
The solution is the shortest distance path from

year 1 to year 6 ; if for example the path is
{ ( 1, 2) , (2, 3) , (3, 4) , (4, 5) , (5, 6) }
then the solution is interpreted as the replace-
ment of the equipment each year with
5
total costs = ∑
i =1
pi − 5 s1 + 5c1
COMPACT BOOK STORAGE IN A

LIBRARY
This problem concerns the storage of books in a
limited size library
Books are stored according to their size, in

terms of height and thickness, with books
placed in groups of same or higher height; the
set of book heights { Hi } is arranged in
ascending order
H1 < H2 < … < Hn

37
LIBRARY
Any book of height Hi may be shelved on a
shelf of height ≥ Hi
The length Li of shelving required for height

class i is computed given the thickness of each
book; the total shelf area required is ∑i H i Li
if only 1 height class [ corresponding to the

tallest book ] exists, total shelf area required
is the total length of the thickness of all
books times the height of the tallest book

LIBRARY
if 2 or more height classes are considered,
the total area required is less than the total
area required for a single class
The costs of construction of shelf areas for each

height class Hi have the components
si fixed costs [ independent of shelf area ]
ci variable costs / unit area

38
LIBRARY
For example, if we consider the problem with 2
height classes Hm and Hn with Hm < Hn
all books of height ≤ Hm are shelved in shelf
with height Hm
all other books are shelved on the shelf with
height Hn
The corresponding total costs are
⎡ m ⎤ ⎡ n ⎤
⎢ sm + c m H m
⎢⎣
∑ L j ⎥ + ⎢ sn + c n H n
⎥⎦ ⎢⎣
∑ Lj ⎥
⎥
j =1 j = m +1 ⎦

LIBRARY
The problem is to find the set of shelf heights and
lengths to minimize the total shelving costs
The solution approach is to use a network flow

model for a network with the
set of n + 1 nodes
N = { 1, 2, . . . , n }
corresponding to the n + 1 book heights with
1 ↔ H1 < H2 < … < Hn ↔ n

39
LIBRARY
directed arcs ( i , j ) only if j > i resulting in
a total of n ( n + 1) arcs
2
“distance” dij on each arc given by
⎧ j
⎪ s + cj H j
d ij = ⎨ j ∑
k = i +1
Lk if j > i
⎪ ∞ otherwise
⎩

LIBRARY
For this network, we solve the shortest route
problem for the specified “distances” dij
Suppose that for a problem with n = 17 , we
determine the optimal trajectory to be
{ ( 0, 7 ) , ( 7, 9 ) , ( 9, 15 ) , (1 5, 17 ) }
the interpretation of this solution is :
40
LIBRARY
store all the books of height ≤ H7 on the
shelf of height H7
store all the books of height ≤ H9 but > H7
on the shelf of height H9
store all the books of height ≤ H15 but > H9
on the shelf of height H15
store all the books of height ≤ H17 but >
H15 on the shelf of height H17
41
Dynamic Programming
George Gross
© 2006 University of Illinois Board of Trustees, All Rights Reserved

DYNAMIC PROGRAMMING
Multi-step problem solving technique
Systematic approach to sequential decision
making
Key characteristic: ability to to separate
problem into stages

1
STAGES AND STATES
We consider the problem to be composed of

multiple stages
A stage is the point in time, space, geographic
location or structural element at which we make
a decision; this point is associated with a
number of states
A state of the system describes a possible
configuration of the system at a given stage
STAGES AND STATES
decision
dn
variable
sn stage n
state
(input)
2
RETURN FUNCTION
A decision dn at the stage n transforms the state
sn at the current stage n into the state sn+1 at
the stage n + 1
The state sn and the decision dn are associated
with the value of the objective; the effect is
measured by the return function denoted by
rn ( s n , d n )
The optimal decision at stage n is the decision
d *n that optimizes the return function for state sn
RETURN FUNCTION
dn decision
variable
sn stage n
state return
rn ( d n, sn)
(input) function
3
ROAD TRIP EXAMPLE
A poor student is traveling from NY to LA
To minimize costs, the student plans to sleep
at friends’ houses each night in cities along the
trip
Based on past experience he can reach
Columbus, Nashville or Louisville after 1
day
Kansas City, Omaha or Dallas after 2 days
San Antonio or Denver after 3 days
LA after 4 days
ROAD TRIP EXAMPLE
680 610
2 Columbus 5 K. City 8 Denver
580
550 515 1030
790
900 6 Omaha
NY 1 3 Nashville 760 10 LA
700 790
770 510 1050 1390
940
660 790
830
4 Louisville 7 Dallas 9 S. Antonio
270
stage 1 stage 2 stage 3 stage 4 stage 5

4
ROAD TRIP
The student wishes to minimize the number of

miles driven and so he wishes to determine the
shortest path between NY and LA
To solve the problem, he works backwards
We adopt the following notation
cij = distance between states i and j
fk( i ) = distance of the shortest path to
LA from state i in stage k
ROAD TRIP EXAMPLE CALCULATIONS
stage 4 : f 4 (8) = 1030 f 4 (9) = 1390
⎧ ⎫
⎪ ⎪
stage 3 : f 3 (5) = min ⎨(610 + 1030),(790 + 1390)⎬ = 1640

⎪⎩ 1640 2180 ⎪⎭
⎧ ⎫
⎪ ⎪
f 3 (6) = min ⎨(540 + 1030),(940 + 1390)⎬ = 1570

⎪⎩ 1570 2330 ⎪⎭
⎧ ⎫
⎪ ⎪
f 3 (7) = min ⎨(790 + 1030),(270 + 1390)⎬ = 1660

⎪⎩ 1820 1660 ⎪⎭
5
ROAD TRIP EXAMPLE CALCULATIONS
stage 2 :
⎧⎪ ⎫⎪
f 2 (2) = min ⎨(680 + 1640) , (790 + 1570) , (1050 + 1660) ⎬ = 2320

⎩⎪ 2320 2360 2710 ⎭⎪
⎧⎪ ⎫⎪
f 2 (3) = min ⎨(580 + 1640) , (760 + 1570) , (660 + 1660)⎬ = 2220

⎩⎪ 2220 2330 2320 ⎭⎪
⎧⎪ ⎫⎪
f 2 (4) = min ⎨(510 + 1640) , (700 + 1570) , (830 + 1660)⎬ = 2150

⎩⎪ 2150 2270 2490 ⎭⎪
stage 1 :
⎧⎪ ⎫⎪
f1 (1) = min ⎨(1550 + 2320) , (900 + 2220) , (770 + 2150) ⎬ = 2870

⎩⎪ *2870* 3120 292 ⎭⎪
ROAD TRIP EXAMPLE
The shortest path is 2,870 miles and corresponds

to the trajectory { (1, 2) , ( 2, 5 ) , ( 5, 8 ) , ( 8, 10 ) } ,
i.e., from New York, the student reaches
Columbus on the first day, Kansas city on the
second day, Denver the third day and then Los
Angeles
Any other trajectory to LA leads to higher costs
and so is, by definition, suboptimal
6
WORKING BACKWARDS: PICK UP
MATCHES GAME
There are 30 matches on a table and 2 players
Each player can pick up 1, 2, or 3 matches and
continue until the last match is picked up
The loser is the person who picks up the last
match
How can the player P1 who goes first ensure a
win?

MATCHES GAME
We solve this problem by reasoning in a back-
wards fashion so as to ensure that when a
single match remains, P2 has the turn
Consider the situation where 5 matches remain
and it is P2’s turn; for P1 to win we, consider all
possible situations:
7
MATCHES GAME
1 ⇒ 4 left P1 removes 3
P2’s move is to pick 2 ⇒ 3 left P1 removes 2
3 ⇒ 2 left P1 removes 1
We can reason similarly for the cases of 9, 13,
17, 21, 25, and 29 matches
Therefore, P1 wins if P1 picks 30 – 29 = 1 match
in the first move
In this manner, we can assure a win for any
arbitrary number of matches in the game
OIL TRANSPORT TECHNOLOGY
We consider the development of a transport

network from north slope of Alaska to one of 6
possible shipping points in the U.S
The network must meet the problem feasibility
requirements
7 pumping stations from a north slope
ground storage plant to a shipping port
use of only those paths that are physically
and environmentally feasible
8
intermediate
region
oil
substations
storage final
destinations
Objective: determine a feasible pumping
configuration that minimizes the
construction costs of branches

total
= of network of the feasible
costs
pumping configuration
9
Possible approaches to solving such a problem
include:
enumeration: exhaustive evaluation of all
possible paths; too costly since there are
more than 100 possible paths
myopic decision rule: at each node, pick
as the next mode the one reachable by the
cheapest path (in case of ties the pick is
arbitrary) ; for example,
0 3 11 15 19 26 29 31 36
oil I-E II-E III-D IV-E V-D VI-D VII-C B
storage
but such a path is not unique and not
guaranteed to be optimal
serial dynamic programming (DP) : we
need to construct the problem solution by
defining the stages, states and decisions
10
DP SOLUTION
We define a stage to represent each pumping
region and so each stage corresponds to the
set of vertical nodes I, II, . . . , VII
We use backwards recursion: start from a final
destination and work backwards to the oil stage
We define a state to denote a particular
pumping station ↔ sk
A decision refers to the selection of the branch
from each state sk , so there are at most three
choices for a decision dk :
L ↔ left F ↔ forward R ↔ right
OIL TRANSPORT TECHNOLOGY :

DP SOLUTION
The return function is defined as the costs
associated with the decision at the state
dk ↔ rk ( sk , d k )
The transition function is the total costs in
proceeding from one stage to another
We proceed to solve the problem by moving
backwards from each final state to the states in
the previous stage
11
DP SOLUTION:
STAGE 1 REGION VII
optimal decision return
d1
s1 d *1 f *1 ( s1 )
R L F
A 7 R 7
B 6 3 F 3
C 7 5 6 L 5
D 6 5 3 F 3
E 7 8 5 F 5
F 4 2 6 L 2
DP SOLUTION
STAGE 2 REGION VI
optimal decision
cumulative cost in proceeding
from s2 to a final destination
d2
s2 d 2* f 2* ( s2 )
R L F
A 10 12 R 10
B 9 12 7 F 7
C 5 6 7 R 5
D 8 7 6 F 6
E 7 6 11 L 6
12
OIL TRANSPORT : STAGE 2
⎛ ⎞
f ( s2 ) = min r2 ( s2 , d 2 ) + f 1 ( s1 ) ⎟
* ⎜ *
d2 ⎜
⎟
2
⎝ ⎠
a function of only s1
⇓
for given d 2 , s1 is set
DP SOLUTION:
STAGE 3 REGION V
{
f 3* ( s3 ) = min r3 ( s3 , d 3 ) + f 2* ( s2 )
d3
}
d3
s3 d *3 f *3 ( s3 )
R L F
A 14 16 R 14
B 14 17 15 R 14
C 10 5 6 R 10
D 9 12 9 R, F 9
E 12 15 L 12
13
DP SOLUTION:
STAGE 4 REGION IV
{
f 4* ( s4 ) = min r4 ( s4 , d 4 ) + f 3* ( s3 )
d4
}
d4
s4 d 4* f 4* ( s4 )
R L F
B 17 18 23 R 17
C 15 22 16 R 15
D 18 17 16 F 16
E 16 21 L 16
DP SOLUTION:
STAGE 5 REGION III
{
f 5* ( s5 ) = min r5 ( s5 , d 5 ) + f 4* ( s4 )
d5
}
d5
s5 d *5 f *5 ( s5 )
R L F
A 19 R 19
B 18 18 R,F 18
C 24 23 17 F 17
D 20 19 25 L 19
E 21 17 F 17
F 20 L 20
14
DP SOLUTION:
STAGE 6 REGION II
{
f 6* ( s6 ) = min r6 ( s6 , d 6 ) + f 5* ( s5 )
d6
}
d6
s6 d *6 f 6* ( s6 )
R L F
A 25 R 19
B 21 25 18 R,F 18
C 28 21 23 F 17
D 27 26 29 L 19
E 26 23 22 F 17
F 18 23 L 20
DP SOLUTION:
STAGE 7 REGION I
{
f 7* ( s7 ) = min r7 ( s7 , d 7 ) + f 6* ( s6 )
d7
}
d7
s7 d *7 f 7* ( s7 )
R L F
A 27 32 R 27
B 26 33 26 R,F 26
C 34 25 27 L 25
D 25 27 33 R 25
E 27 35 30 R 27
15
THE OPTIMAL TRAJECTORY
For the last stage

I-A I-B I-C I-D I-E
f 8* ( s8 ) = min{27 + 6, 26 + 4, 25 + 7, 25 + 8, 27 + 3 }
= 30
To find trajectory we retrace forwards through
the stages 7, 6, . . . , 1 and obtain

THE OPTIMAL TRAJECTORY
d8=L d7=R d6=F d5=R d4=R d3=R d2=F
oil
storage
I II III VI V VI VII
B A A B C D D
In addition to this optimal solution, others are
possible since the path need not be unique

16
OIL TRANSPORT PROBLEM
SOLUTION
We obtain the diagram shown on the next slide
by retracing the steps at each stage
The solution
provides all the optimal solutions
is based on logically breaking up the problem
into stages with computation in each stage
being a function of the number of states in the
stage
provides all the suboptimal paths

SOLUTION final destinations
3 3
6 3
9 6
oil substations
storage
17
SOLUTION
For example, we may calculate the least cost
optimal path to any sub optimal shipping point
different than D
From the solution, we can also determine the
sub optimal path if the construction of a feasible
path is not undertaken

OIL TRANSPORT : SENSITIVITY CASE

Consider the case where we got to stage VI but
the arc VI – D to VII – D cannot be built due to
some environmental intervention
We determine, then, the least-cost path from VI –
D to find the final destination D whose value is 9
instead of 6
2 7 final
VI - D VII - C destination
D
and so the sub optimal cost solution costs are 33
18
FACILITIES SELECTION
A company is expanding to meet a wider market
and considers:
3 location alternatives
4 different building types (sizes) at each site
Revenues and costs vary with each location and
building type
Revenues R increase monotonically with
building size; these are net revenues or profits
Costs C increase monotonically with building
size
The data for building sizes, revenues and costs
are given in the table

19
building size
B1 B2 B3 B4 none
site
R1 C1 R2 C2 R3 C3 R4 C4 R0 C0
I 0.50 1 0.65 2 0.8 3 1.4 5 0 0
II 0.62 2 0.8 5 0.96 6 1.8 8 0 0
III 0.71 4 1.2 7 1.6 9 2 11 0 0
THE FACILITIES SELECTION PROBLEM
The company can afford to invest at most 21
million $ in the total expansion project
Determine the optimal expansion policy, i.e., the
buildings to be built at each site

20
DP SOLUTION APPROACH
We use the DP approach to solve this problem;
first, however, we need to define the DP
parameters
For the facilities siting problem we realize that
without the choice of a site, the building type is
irrelevant and so the elements that control the
entire decision process are building sites

stage ↔ site
amount of funds available
state ↔
for construction
decision ↔ building type
profits attained, i.e., the
return function ↔
revenues
impact of a decision on the
transition function ↔
availablity of resources
21
As we move from stage n backwards to stage n-1
after a decision dn is taken the funds available
for construction are
sn−1 = sn − cn ← costs of decision d n
The recursion relation is
{
f n* ( sn ) = max f n ( sn , d n ) + f n−*1 ( sn−1 )
dn
} n = 1,2, 3
with
s n − 1 = sn − c n
and
f n ( sn , d n ) = rn ( sn , d n ) = Rn← profits of decision d n
We use backwards DP and start with site I, a
purely arbitrary choice, as stage 1, where this
stage 1 represents the last decision in the 3 -
stage sequence
The amount of funds available is unknown since

the decision at sites II and III are assumed to
have been previously made, but, clearly,
0 ≤ s1 ≤ 21
22
DP SOLUTION : STAGE 1 SITE I
f1 ( s1 ) = max { r1 ( s1 , d1 )}

0 ≤ d1 ≤ 4 R1
s1 0 1 2 3 4 d 1* f 1* ( s1 )
21 ≥ s1 ≥ 5 0 .5 .65 .80 1.4 4 1.4
4 ≥ s1 ≥ 3 0 .5 .65 .80 3 .80
2 0 .5 .65 2 .65
1 0 .5 1 .50
0 0 0 0 0 0 0 0
DP SOLUTION : STAGE 2 SITE II
The amount of funds s2 available is unknown
since the decision at site III is assumed to have
been already made
The value of d2 is a function of s2 and we
construct a decision table using
f 2* ( s2 ) = max { r2 ( s2 , d 2 ) + f 1* ( s1 )}

0 ≤ d2 ≤ 4 R2
23
DP SOLUTION : STAGE 2 SITE II
s2 0 d2 1 2 3 4 dd **22 ff**22(S
( s 22))
21 ≥ s ≥ 13 1.40
2 2.02 2.18 2.36 3.20 4 3.20
12 1.40 2.02 2.18 2.36 2.60 4 2.60
11 1.40 2.02 2.18 2.36 2.60 4 2.60
10 1.40 2.02 2.18 1.76 2.45 4 2.45
9 1.40 2.02 1.58 1.61 2.30 4 2.30
8 1.40 2.02 1.58 1.61 1.80 1 2.02
7 1.40 2.02 1.43 1.61 1 2.02
6 1.40 1.42 1.28 1.46 3 1.46
5 1.40 1.42 0.78 0.96 1 1.42
4 0.80 1.27 1 1.27
3 0.80 1.12 1 1.12
2 0.65 0.62 0 0.65
1 0.50 0 0.50
0 0.00 0 0.00
SAMPLE CALCULATIONS
Consider the case s2 = 10 and d2 = 0 :
C2 = 0 and R2 = 0
then, s1 = 10 and d 1* = 4 so that
f 1* ( s1 ) = 1.4
consequently, f2(s2) = 1.4
Consider the case s2 = 10 and d2 = 4 :
C2 = 8 and R2 = 1.8
then, s1 = 2 and d 1* = 2 so that
f 1* ( s1 ) = .65
consequently, f2(s2) = 2.45
We can, then, show that
f 2* ( s2 ) = 2.45
24
DP SOLUTION : STAGE 3 SITE III
At stage 3 , the first actual decision is made and
so exactly 21 million is available and s3 = 21
We compute the elements in the table using
f 3* ( s3 ) = max{ r3 ( s3 , d 3 ) + f *2 ( s2 )}
d3 N
where R3
s2 = s 3 − C 3
d3
d 3* f *3 ( s3 )
s3 0 1 2 3 4
21 3.2 3.91 4.4 4.1 4.45 4 4.45
OPTIMAL SOLUTION
Optimal profits are 4.45 million
The optimal decision is obtained by retracing
steps from stage 3 to stage 1:
d *3 = 4 ↔ construct B2 at site III
s2 = s3 – C3 = 21 – 11 = 10
d *2 = 4 ↔ construct B4 at site II
s1 = s2 – C2 = 10 – 8 = 2
d 1* = 2 ↔ construct B2 at site 1
C1 = 5 and C1+C2+C3 = 21
25
SENSITIVITY CASE
We next consider the case where the maximum
investment available is 15 million
By inspection, the results in stages 1 and 2
remain unchanged; however, we must recom-
pute stage 3 results with the 15 million limit
d3
s3 d 3* f *3 ( s3 )
0 1 2 3 4
15 3.2 3.31 3.22 3.06 3.27 1 3.31
SENSITIVITY CASE
The optimal solution for this sensitivity case
obtains maximum profits of 3.31 million and the
decision is
d *3 = 1 ↔ construct B1 at site III
s 2 = s3 − C 3 = 15 − 4 = 11
d *2 = 4 ↔ construct B4 at site II
s1 = s2 − C 2 = 11 − 8 = 3
d 1* = 3 ↔ construct B3 at site I
C1 = 3 and C1 + C 2 + C 3 = 15
26
OPTIMAL CUTTING STOCK PROBLEM
A paper company gets an order for:
8 rolls of 2 ft paper at 2.50 $/roll
6 rolls of 2.5 ft paper at 3.10 $/roll
The company only has 13 ft of paper to fill these

orders; partial orders can be filled
Determine how to fill orders to maximize profits

A stage is an order and since there are 4 orders
we construct a 4 – stage DP
d4 d3 d2 d1
s4 stage s3 stage s2 stage s1 stage

4 3 2 1
r4 r2 r3
r1
A state in stage n is the remaining ft of paper
left for the order being processed at stage n and
all the remaining stages
27
A decision in stage n is the amount of rolls to
produce in stage n :
⎡F ⎤ F
d n = ⎢ 0 ⎥ , the largest integer in 0
⎣ Ln ⎦ Ln
where
Ln = length of order n ( ft )
F0 = length of available paper ( ft )

The return function at stage n is the additional
revenues gained from producing dn rolls
The transition function measures amount of
paper remaining at stage n
sn − 1 = sn − d n Ln n = 2,3,4
s0 = s1 − d1 L1
and s0 should be as close as possible to 0
⎡s ⎤
Clearly, d1 = ⎢ 1 ⎥
⎣ L1 ⎦
28
The recursion relation is
f n* ( sn ) = max {r ( s , d
n n n ) + f n−*1 ( sn−1 ) }
⎡ ⎤
0 ≤ dn ≤ ⎢ sn ⎥
L
⎣ n⎦
n = 1, … , 4
where
sn−1 = sn − d n Ln
and
f 0* ( s0 ) = 0
f n ( sn , d n ) = rnd n + f n−*1 ( sn − d n Ln )
Arbitrary order of the stages and we pick
stage n 1 2 3 4
length of
2.5 4 3 2
order ( ft )
We proceed backwards from stage 1 to stage 4
and we know that

29
DP SOLUTION : STAGE 1
f 1* ( s1 ) = max
0 ≤ d1 ≤ 5
{r1 ( s1 , d1 )} = max {3.10 d1 }
0 ≤ d1 ≤ 5
⎡ 13 ⎤
d1 ≤ ⎢ = 5
⎣ 2.5 ⎥⎦
s1
f *1 ( s 1 )
DP : SOLUTION STAGE 2
f *2 ( s2 ) = max
0 ≤ d2 ≤ 3
{5.25 d 2 + f 1* ( s2 − 4 d 2 ) }
⎡ 13 ⎤
d2 ≤ ⎢ ⎥ = 3
⎣4⎦
s2
f *2 ( s 2 )
30
f *3 ( s3 ) = max
0 ≤ d3 ≤ 4
{4.40 d 3 + f *2 ( s3 − 3 d 3 ) }
⎡ 13 ⎤
d3 ≤ ⎢ ⎥ = 4
⎣3⎦
s2
f *3 ( s 3 )
f *4 ( s4 ) = max
0 ≤ d4 ≤ 6
{ 2.5 d 4 + f *3 ( s4 − 2 d 4 ) }
⎡ 13 ⎤
d4 ≤ ⎢ ⎥ = 6
⎣2⎦
*
d4 0 1 2 3 4 5 6 d 4* f 4 ( s4 )
s4 = 13 18.45 17.5 18.2 17.15 16.2 16.9 15 0 18.45
31
DP OPTIMAL SOLUTION
The maximum profits are $18.45
The optimal solution is obtained by retracing
f 4* ( s4 = 13) = 18.45 with d 4* = 0 ↔ no rolls of 2 ft
f 3* ( s3 = 13) = 18.45 with d 3* = 3 ↔ 3 rolls of 3 ft
f 2* ( s2 = 4) = 5.25 with d 2* = 1 ↔ 1 roll of 4 ft
f 1* ( s1 = 0 ) = 0 with d 1* = 0 ↔ no rolls of 2.5 ft
SENSITIVITY CASE
Consider the case that due to an incorrect
measurement, in truth, there are only 11 ft
available for the rolls
We note that the solution for the original 13 ft
covers this possibility in the stages 1, 2 and 3 but
we need to recompute the results of stage 4,
which we now call stage 4′

32
SENSITIVITY CASE : STAGE 4′
The stage 4′ computations become
⎡ 11 ⎤
d 4′ ≤ ⎢ ⎥ = 5
⎣2⎦
d 4′ 0 1 2 3 4 5 d 4*′ f 4*′ ( s4 )
s4 = 11 15 15.7 14.65 13.7 14.4 12.5 1 15.7
SENSITIVITY CASE OPTIMUM
The optimal profits in this sensitivity case are
$15.7
d 4*′ = 1, ↔ 1 roll of 2 ft
d 3*′ = 3 ↔ 3 rolls of 3 ft
d 2′ = d 1′ = 0 ↔ no rolls of 4 ft
* *
↔ no rolls of 2.5 ft
33
ANOTHER SENSITIVITY CASE
We consider the case with the initial 13 ft, but in
addition we get the constraint that at least 1 roll
of 2 ft must be produced:
d4 ≥ 1
Note that no additional work is needed since the

computations in the first tables have all the
necessary data
This sensitivity case optimum profits are $18.2
The optimum solution is :
OPTIMAL CUTTING STOCK PROBLEM
f 4*′′ ( s4 = 13) = 18.2 with d 4*′′ = 2 ↔ 2 rolls of 2 ft

f 3*′′ ( s3 = 9) = 13.2 with d 3*′′ = 3 ↔ 3 rolls of 3 ft
and since s2 = s1 = 0
d 2*′′ = 0 ↔ 0 rolls of 4 ft
d 1*′′ = 0 ↔ 0 rolls of 2.5 ft
The constraint reduces optimum from $ 18.45 to
$18.2 and so it costs $ .25
34
INVENTORY CONTROL PROBLEM
This problem is concerned with the development
of an optimal ordering policy for a retailer
The sales of a seasonal item has the demands
month Oct Nov Dec Jan Feb Mar
demand 40 60 30 40 30 20
All units sold are purchased from a vendor at
4 $/unit `; units are sold in lots of 10, 20, 30, 40 or
50 with the corresponding discount
lot size 10 20 30 40 50
discount
4 5 10 20 25
%
35
There are ordering costs: each order incurs
fixed costs of $ 2 and $ 8 for shipping, handling
and insurance
The storage limitations of the retailer require
that no more than 40 units be in inventory at the
end of the month and the storage charges are
0.2 $/unit; inventory at time 0 and inventory at the
e.o.p. 6 are 0
Assumption: demand occurs at a constant rate
throughout the month
We formulate the problem as a DP and use a
backward process
Each stage corresponds to a month
month Oct Nov Dec Jan Feb Mar
stage
6 5 4 3 2 1
n
36
d6 d5 d4 d3 d2 d1
s6 stage s5 stage s4 stage s3 stage s2 stage s1 stage s0

6 5 4 3 2 1
r6 r5 r4 r3 r2 r1
The state variable at stage n is defined as the
amount entering inventory given that there are
n months remaining – the present month n
plus months n – 1 , n – 2 , . . . , 1
The decision variable at stage n is the amount of
units ordered to satisfy demands for the n
months
The transition function is defined by
sn-1 = sn + dn – Dn n = 1, 2, . . . , 6
s0 = 0 s6 = 0 demand in month n
37
The return function at stage n is given by
rn ( d n , sn ) = φ ( d n ) + hn ( sn + d n − Dn )

0.2( sn + dn − Dn )
with storage costs
φ ( dn ) = 10
+ [1 − ρ ( d n ) ] d n

fixed discount
costs factor
d n = 10,20,30,40,50
dn 0 10 20 30 40 50
φ (d n ) 0 48 86 118 138 160
In the DP approach, at each stage we minimize
{
f n* ( sn ) = min φ (d n ) + hn ⎡⎣ sn + d n − Dn ⎤⎦ + f n−*1 ( sn−1 )
dn
}
n = 1,...,6
s0 = 0 and so f 0* ( s0 ) = 0
38
s0 = 0 ⎫
⎬ ⇒ s1 = 20, 10 or 0 ⇒ d *1 = 0 , 10 or 20
D1 = 20 ⎭
f 1* ( s1 ) = min{ φ (d1 ) + 0 } = φ (d1* )

d1
s1 20 10 0
d *1 0 10 20
φ ( d *1 ) 0 48 86
D2 = 30 and s1 = s2 + d 2 − 30
{
f *2 ( s2 ) = min φ (d 2 ) + 0.2 ⎡⎣ s2 + d 2 − 30⎤⎦ + φ (d *1 )
d2
}
d2
d *2 f 2* ( s2 )
s2 0 10 20 30 40 50
0 204 188 164 50 164
10 172 168 142 40 142
20 134 136 122 122 30 122
30 86 98 90 0 86
40 50 52 0 50
39
D3 = 40 and s2 = s3 + d 3 − 40
*
3
d3
{
f ( s3 ) = min φ (d 3 ) + 0.2 ⎡⎣ s3 + d 3 − 40⎤⎦ + f *2 ( s2 ) }
d3
d *3 f 3* ( s3 )
s3 0 10 20 30 40 50
0 302 304 40 302
10 282 282 286 30, 40 282
20 250 262 264 252 20 250
30 212 230 244 230 218 10 218
40 164 192 212 210 196 0 164
D4 = 30 and s3 = s4 + d 4 − 30
{
f *4 ( s4 ) = min φ (d 4 ) + 0.2 ⎡⎣ s4 + d 4 − 30⎤⎦ + f *3 ( s3 )
d4
}
d4
d 4* f 4* ( s4 )
s4 0 10 20 30 40 50
0 420 422 414 50 414
10 388 402 392 384 50 384
20 350 370 372 362 332 50 332
30 302 332 340 342 210 0 302
40 284 302 310 290 0 284
40
D5 = 20 and s4 = s5 + d 5 − 20
{
f 5* ( s5 ) = min φ (d 5 ) + 0.2 ⎡⎣ s5 + d 5 − 20⎤⎦ + f *5 ( s5 )
d5
}
d5
d 5* f *5 ( s5 )
s5 0 10 20 30 40 50
0 500 504 474 468 50 468
10 462 472 454 446 452 40 446
20 414 434 422 426 430 0 414
30 286 384 394 410 10 384
40 336 356 378 0 336
n=6 D6 = 40 and s6 = 0
s5 = s6 + d 6 − 40 = d 6 − 40
{
f *6 ( s6 ) = min φ (d 6 ) + 0.2 ⎡⎣ s6 + d 6 − 40⎤⎦ + f *5 ( s5 )
d
}
6
d6 0 10 20 30 40 50 d *6 f 6* ( s6 )
f6(s6) 606 608 40 606
d *6 = 40 ⇒ d *5 = 50 ⇒ d *4 = 0 ⇒ d *3 = 40 ⇒ d *2 = 50 ⇒ d 1* = 0
41
OPTIMAL SOLUTION
d *6 = 40 which corresponds to s5 = 0 and costs 606
d 1* = 0 with costs 0
OPTIMAL SOLUTION
d *6 = 40 d *5 = 50 d *4 = 0 d *3 = 40 d *2 = 50 d *1 = 0
s6 = 0 s5 = 0 s4= 30 s3= 0 s2= 0 s1 = 20 s0=0
stage stage stage stage stage stage
6 5 4 3 2 1
r6 (40,0) r4 (0,30) r2 (50,0)

= 138 r5 (50,0) =0 r3 (40,0) = 164 r1 (0,20)
= 166 = 138 =0
42
OPTIMAL SOLUTION
optimal trajectory is
s0 = 0 s1 = 20 s2 = 0 s3 = 0 s4 = 30 s5 = 0
with total costs
0 + 164 + 138 + 0 + 166 + 138 = 606

MUTUAL FUND INVESTMENT

STRATEGIES
We consider a 5-year investment of
10 k$ invested in year 1
1 k$ invested in each year 2, 3, 4 and 5
into 2 mutual funds with different yields for

both the short-term (1 year) and the long-term
(up to 5 years)
A decision at the beginning of each year is the

allocation of investment in each fund
43
STRATEGIES
We operate under the protocol that
once invested, the money cannot be
withdrawn until the end of the 5-year horizon
all short term gains may be reinvested in
either of the two funds or withdrawn in which
case the withdrawn funds earns no more
further interest
The objective is to maximize the total returns at
the end of 5 years

STRATEGIES
The earnings on the investment are
LTD : the long-term dividend specified as

% return / year on the accumulated capital
STD : the short-term interest dividend is the

cash returned to the investor at the end of
the period; cash may be reinvested and any
money not invested in one of the funds earns
nothing
44
STRATEGIES
STD rate in for year n

LTD
fund
1 2 3 4 5 rate I
A 0.02 0.0225 0.0225 0.025 0.025 0.04
B 0.06 0.0475 0.05 0.04 0.04 0.03
We use backwards DP to solve the problem
The stages are the 5 investment periods
stage n Δ year 6 − n n = 1,2,3,4,5

r5 r4 r3 r2 r1
s5 stage s4 stage s3 stage s2 stage s1 stage s0

5 4 3 2 1
d5 d4 d3 d2 d1
45
DP SOLUTION METHOD
For stage n , the state sn is the amount of capital
available for investment in the year 6 – n
The decision dn is the amount of capital

invested in fund A in year 6 – n ; the amount of
capital invested in fund B in the year 6 – n is
therefore sn – dn
In each year, we need to determine the amount

to invest in each fund
DP SOLUTION METHOD
The use of backward recursion considers year 5
first and each of the previous years sequentially
Basic considerations:
for each year 6 – n
dn is invested in fund A returns d n i A ( SDT )
( sn − d n )is invested in fund B returns ( sn − d n )i B ( SDT )
for the year 6 – n + 1
sn−1 = d n i A + ( sn − d n ) iB + 1000 n = 2, 3,4,5
s5 = 10,000
46
THE OBJECTIVE
The objective is to maximize the total returns
5
max R = ∑r
n =1
n
We express all returns in end of the year 5

dollars : rn is the future value of long term
earnings in the years 1, 2, 3 and 4
rn = (1 + I A )n d n + (1 + I B )n ( sn − d n ) n = 1, ... ,5
But for n = 1, r1 is the present value of all earn-
ings in stage 1
r1 = (1 + I A ) d1 + (1 + I B ) ( s1 − d1 ) + i A d1 + i B ( s1 − d1 )
For stage 1
r1
s1
d1
where
r1 = (1 + I A )d1 + (1 + I B )( s1 − d1 ) + i Ad1 + i B( s1 − d1 )
= ( I A + i A − I B − iB ) d1 + (1 + I B + i B ) s1
47
r1 = earnings in stage 1 (returns realized at
the end of 5 years)
⎧ d1 ( I A + i A − I B − i B ) + ⎫
f 1* ( s1 ) = max { r1 } = max ⎨ ⎬
d1 d1 ⎩ s1 (1 + I B + i B ) ⎭
⎧ d ( 0.04 + 0.025 − 0.03 − 0.04) + ⎫
= max ⎨ 1 ⎬
0 ≤ d1 ≤ s1 ⎩ s1 (1 + 0.03 + 0.04) ⎭
= max {d1 ( − 0.005) + s1 (1.07)} maximum
optimal d1
return in
decision stage 1
d *1 = 0 f 1* ( s1 ) = 1.07 s1
r2 = returns realized at the end of 5 years due to
the decision in stage 2
= d 2 (1 + I A )2 + ( s2 − d 2 )(1 + I B )2
= d 2 ⎡⎣(1 + I A )2 − (1 + I B )2 ⎤⎦ + s2 (1 + I B )2
s1 = s2 iB + d 2 ( i A − iB ) + 1,000
48
We select d 2* to maximize
f 2* ( s2 ) = max
d2
{r 2 + f 1* ( s1 ) }
= max
d2
{d 2 (1.042 − 1.032 ) + s2 (1.03)2 + f 1* ( s1 ) }
⎧ d 2 ( .0207) + 1.0609 s2 + ⎫
= max ⎨ ⎬
0 ≤ d2 ≤ s2 ⎩1.07[ .04 s2 + d 2 ( − .015) + 1,000]⎭
= max
d2
{d 2 (.0046) + 1.1037 s2 + 1070}
d 2* = s2 f 2* ( s2 ) = 1.108 s2 + 1070
the decision d3
= d 3 (1 + I A )3 + ( s3 − d 3 )(1 + I B )3
= d 3 ⎡⎣(1 + I A )3 − (1 + I B )3 ⎤⎦ + s3 (1 + I B )3
s2 = s3 iB + d 3 ( i A − iB ) + 1,000
49
f 3* ( s3 ) = max
d3
{r 3 + f 2* ( s2 ) }
⎧ d 3 (1.043 − 1.033 ) + s3 (1.03)3 + ⎫
= max ⎨ ⎬
d3
⎩1.108 s2 + 1,070 ⎭
= max {2,178 + 1.1481s3 + .0018d 3 }
0 ≤ d3 ≤ s3
d 3* = s3 f 3* ( s3 ) = 1.15 s3 + 2,178

the decision d4
= d 4 (1 + I A )4 + ( s4 − d 4 )(1 + I B )4
= d 4 ⎡⎣(1 + I A )4 − (1 + I B )4 ⎤⎦ + s4 (1 + I B )4
s3 = s4 i B + d 4 ( i A − iB ) + 1,000
50
f 4* ( s4 ) = max
d4
{r + f
4
*
3 ( s3 )}
= max
d4
{d 4 (1.044 − 1.034 ) + s4 (1.03)4 + 1.15 s3 + 2,178 }
= max
0 ≤ d4 ≤ s4
{3328 + 1.1772 s4 + .0156d 4}
d 4* = s4 f 4* ( s4 ) = 1.193 s4 + 3,328
the decision d5
= d 5 (1 + I A )5 + ( s5 − d 5 )(1 + I B )5
= d 5 ⎡⎣1.045 − 1.035 ⎤⎦ + s5 (1.03)5
s5 = 10,000 ← capital available for investment
s4 = s5 i B + d 5 ( i A − iB ) + 1,000
= 10,000 iB + d 5 ( i A − iB ) + 1,000
51
⎧ ⎫
⎪ ⎪
f 5 ( s5 ) = max ⎨10,000(1.03) + d 5 (1.04 − 1.03 ) + f 4 ( s4 ) ⎬
* 5 5 5 *
0 ≤ d5 ≤ s4 ⎪
⎪⎭
⎩ 11,593 0.0574
⎡⎣1,000 + 600 + d 5 ( −.04)⎤⎦ 1.193 + 3,328
⎧ (.0574 − 0.048) ⎫
= max ⎨16,830 + d 5 ⎬
0 ≤ d5 ≤ s5 ⎩ 0.097 ⎭
= 16,830 + 0.097(10,000)
d 5* = 10,000 f 5* ( s5 ) = 16,927
OPTIMAL SOLUTION
optimal return at end of 5 years is 16,927 using
the following strategy
beginning investment in
of year fund A fund B
1 10,000 0
2 STD returns + 1,000 0
5 0 STD returns + 1,000
52
Combinatorial Analysis
George Gross
COMBINATORIAL ANALYSIS
Many problems in probability theory can be
solved by simply counting the number of ways a
certain event may occur
We review some basic aspects of combinatorial
analysis
combinations
permutations
1
BASIC PRINCIPLE OF COUNTING
Suppose that two experiments are to be

performed:
experiment 1 may result in any one of the m
possible outcomes
for each outcome of experiment 1, there exist
n possible outcomes of experiment 2
Therefore, there are mn possible outcomes of the
two experiments
BASIC PRINCIPLE OF COUNTING

The basic principle is easy to prove by
exhaustive enumeration
(1, 1), (1, 2), (1, 3), ... (1, n)
(2, 1), (2, 2), (2, 3), ... (2, n)
.
.
.
(m, 1), (m, 2), (m, 3), . . . (m, n)
where, (i , j) is the notation we use to refer to
outcome i in experiment 1 and outcome j in
experiment 2
2
EXAMPLE 1 : PAIR FORMATION
Pairs need to be formed consisting of 1 boy and
1 girl by choosing from a group of 7 boys and 9
girls
There exist (7)(9)= 63 possible pairs since there
are 7 ways to pick a boy and 9 ways to pick a girl

GENERALIZED VERSION OF THE

BASIC PRINCIPLE
For r experiments with the first experiment
having n1 possible outcomes; for every outcome
of the first experiment, there are n2 possible
outcomes for the second experiment, and so on
.......
1
2
2
. .......
n1 . 2
1 .
.......
2
3
GENERALIZED VERSION OF THE
BASIC PRINCIPLE
There are
r
Π ni = n1 ⋅ n2 ⋅ n3 ...... ⋅ nr
i =1
possible outcomes for the r experiments, i.e.,
r
there are Π ni possible branches in the
i =1
illustration
EXAMPLE 2 : SUBCOMMITTEE
CHOICES
The executive committee of an Engineering
Open House function consists of:
3 first year students
4 sophomores
5 juniors
2 seniors
We need to form a subcommittee of 4 with each
year represented:
There are 3 ⋅ 4 ⋅ 5 ⋅ 2 = 120 different subcommittees
4
EXAMPLE 3 : LICENSE PLATE
We consider possible combinations for a six-
place license plate with the first three places
consisting of letters and the last three places of
numbers
Number of combinations with repeats allowed is
(26) (26) (26) (10) (10) (10) = 17,576,000
Combination number if no repetition allowed is
(26) (25) (24) (10) (9) (8) = 11,232,000

EXAMPLE 4 : n POINTS
Consider n points at which we evaluate the
function
f ( i ) ∈ {0 ,1} i = 1,2,...., n
Therefore, there are 2n possible function values

5
PERMUTATIONS
A set of 3 objects{ A, B, C } may be arranged in 6
different ways:
BCA ABC CBA
BAC ACB CAB
Each arrangement is called a permutation
The total number of permutations is derived from
the Basic Principle:
there are 3 ways to pick the first element
there are 2 ways to pick the second element
there is 1 way to pick the third element
PERMUTATIONS
Therefore, there are 3 ⋅ 2 ⋅ 1 = 6 ways to arrange
the 3 elements
In general, a set of n objects can be arranged
into
n! = n ( n – 1 ) ( n – 2 ) . . . 1
different permutations
6
EXAMPLE 5 : BASEBALL
Number of possible batting orders for a baseball
team with nine members is
9! = 362,880
Suppose that the team, however, has altogether

12 members; the number of possible batting
orders is the product of the number of team
formations and the number of permutations is
12! 12!
i 9! = = 2(11!) = 79,833,600
3! 9! 3!
EXAMPLE 6 : CLASSROOM
A class with 6 boy and 4 girl students is ranked
in terms of weight ; assume that no two students
have the same weight
There are
10! = 3,628,800
possible rankings
If the boys (girls) are ranked among themselves,
the number of different possible rankings is
6!4! = 17,280
7
EXAMPLE 7 : BOOKS
A student has 10 books to put on the shelf:
4 EE, 3 Math, 2 Econ, and 1 Decision
Student arranges books so that all books in
each category are together
There are 4!3!2!1! arrangements so that all EE
books are first in line, then the Math books, Econ
books, and Decision book

EXAMPLE 8 : BOOKS
But, there are 4! possible orderings of the
subjects
Therefore, there are
4!4!3!2!1! = 6912
permutations of arranging the 10 books

8
EXAMPLE 9 : PEPPER
We wish to determine the number of different
letter arrangements in the word PEPPER
Consider first the letters P1 E1 P2 P3 E2 R where we
distinguish the repeated letters among
themselves: there are 6! permutations of the 6
distinct letters
EXAMPLE 9 : PEPPER
However, if we consider any single permutation
of the 6 letters – for example, P1 P2 E1 P3 E2 R –
provides the same word PPEPER as 11 other
permutations if we fail to distinguish between
the same letters
Therefore, there are 6! permutations for distinct
letters but only
6!
= 60
3!2!
permutations when repeated letters are not
distinct
9
GENERAL STATEMENT
Consider a set of n objects in which
n1 are alike ( category 1 )
n2 are alike ( category 2 )
.
.
.
nr are alike ( category r )
There are
n!
n1 ! n2 !.....nr !
different permutations
EXAMPLE 9 : COLORED BALLS
We have 3 white, 4 red, and 4 black balls which
we arrange in a row; similarly colored balls are
indistinguishable from each other
There are
11!
= 11,550
3!4!4!
possible arrangements
10
COMBINATIONS
Given n objects, we form groups of r objects

and establish the number of different groups
we can form
For example, consider 5 objects denoted as

A,B,C,D and E and form groups of 3 objects;
we can pick the first item in exactly 5 ways
we can pick the second item in exactly 4 ways
we can pick the third item in exactly 3 ways

COMBINATIONS
and, therefore, we can select
5 ⋅ 4 ⋅ 3 = 60
possible groups in which the order of the groups
is taken into account
But, if the order of the objects is not of interest,
i.e., we ignore that each group of three objects
can be arranged in 6 different permutations, the
total number of distinct groups is
5! 60
= = 10
2!3! 6
11
GENERAL STATEMENT ON
COMBINATIONS
The objective is to arrange n elements into
groups of r elements
We can select groups of r
n!
( n − r )!
different ways
But, each group of r has r ! permutations
The number of different combinations is
n!
( n − r )! r !
BINOMIAL COEFFICIENTS
We define the term
⎛ n⎞ n!
⎜r ⎟ ( n − r )! r !
⎝ ⎠
as the binomial coefficient of n and r
A binomial coefficient gives the number of
possible combinations of n elements taken r at
a time
12
EXAMPLE 10 : COMMITTEE
SELECTION
We wish to select three persons to represent a
class of forty
There are
40! 40 ⋅ 39 ⋅ 38
= = 20 ⋅ 13 ⋅ 38 = 9880
37!3! 3⋅ 2⋅1
possible committee selections

EXAMPLE 11 : GROUP FORMATION
Given a group of 5 boys and 7 girls, form sets
consisting of 2 boys and 3 girls
There are
⎛ 5⎞⎛ 7⎞ 5! 7 ! 5⋅4 7⋅6⋅5

⎜ 2 ⎟ ⎜ 3 ⎟ = 3!2! 4!3! = 2 ⋅
= 350
⎝ ⎠⎝ ⎠ 3 2
possible ways to form such groups

13
GENERAL COMBINATORIAL IDENTITY
⎛ n⎞ ⎛ n − 1⎞ ⎛ n − 1⎞
⎜ ⎟ = ⎜ ⎟ + ⎜ ⎟
⎝r ⎠ ⎝r −1⎠ ⎝ r ⎠
number of number of number of

ways of ways of ways of
selecting selecting selecting
groups of r groups of r – 1 groups of r
from n from n from n – 1
MULTINOMIAL COEFFICIENTS
Given a set of n distinct items, form r distinct
groups of respective sizes n1, n2, . . . , and nr with

r
∑n
i =1
i = n
There are
⎛n ⎞
⎜n ⎟
⎝ 1⎠
possible choices for the first group

14
For each choice of the first group, there are

⎛ n − n1 ⎞
⎜ ⎟
⎝ n2 ⎠
possible choices for the second group
We continue with this reasoning and we
conclude that there are

n!
n1 ! n2 ! ... nr !
possible groups
The previous conclusion was gained by realizing

that
⎛ n ⎞ ⎛ n − n1 ⎞ ⎛ n − n1 − n2 ⎞ ⎛ n − n1 − n2 − . . . nr −1 ⎞
⎜ n ⎟⎜ n ⎟⎜ n ⎟ . . .⎜ ⎟=
⎝ 1 ⎠⎝ 2 ⎠ ⎝ 3 ⎠ ⎝ n r ⎠
n! (n − n1 )! n − n1 − n2 − ...nr −1
... =
(n − n1 )!n1! (n − n1 − n2 )!n2 ! 0 !nr !
n!
n1 !n2 ! ...nr !
15
Let
n = n1 + n2 + n3 + . . . + nr
we define the multinomial coefficient
⎛ n ⎞ n!
⎜ n ,n , . . . ,n ⎟
⎝ 1 2 r ⎠ n1 ! n2 ! n3 ! . . . nr !
A multinomial coefficient represents the number
of possible divisions of n distinct objects into r
distinct groups of respective sizes n, n2, . . . , nr

EXAMPLE 12 : POLICE
A police department of a small town has 10
officers
The department policy is to have 5 members on
street patrol, 2 members at the station and 3 on
reserve
The number of possible divisions is

10!
= 2,520
5!3!2!
16
EXAMPLE 13 : TEAM FORMATION
We need to form two teams, the A team and the
B team, with each team having 5 boys from a
group of 10 boys
There are
10!
= 252
5!5!
possible divisions
EXAMPLE 13 : TEAM FORMATION
Suppose that these two teams are to play

against one another
In this case, the order of the two teams is
irrelevant since there is no team A and team B
per se but simply a division of 10 boys into 2
groups of 5 each
The number of ways to form the two teams is
1 ⎛ 10! ⎞
⎜ ⎟ = 126
2! ⎝ 5!5! ⎠
17
EXAMPLE 14 : TEA PARTY
A woman has 8 friends of whom she will invite 5
to a tea party
How many choices does she have if 2 of the
friends are feuding and refuse to attend
together?
How many choices does she have if 2 of her
friends will only attend together?

18

fullSlideDecisionMaking PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

fullSlideDecisionMaking PDF

Загружено:

Авторское право:

Доступные форматы

ECE 307 – Techniques for

 The course covers techniques that are useful

 Expansion of production facilities/warehousing

 Adoption of new technology

 Implementation of a new production schedule

 Changes in the production mix

 Risk management in purchase/sale activities

 Optimal scheduling of processes/projects

 Development of the analytical framework for

 Decisions are made by selecting from possible

PRODUCT MIX OPTIMIZATION

requiring various levels of resources and

providing different benefits (profits)

 The constraints on resources are given

 Problem: determine the optimal daily mix, i.e.,

the production schedule that maximizes profits

without violating any constraints

labor (h) 1 1 1 100

material (lb) 10 4 5 600

A&G (h) 2 2 6 300

profits per unit of product ($) 10 6 4

PRODUCT MIX OPTIMIZATION

PRODUCT MIX OPTIMIZATION

corresponding to maximum profits

OPTIMAL TRAJECTORY PLANNING

stage 1 stage 2 stage 3 stage 4

 The relative “costs” for the various possible

 The problem is to select the route that

OPTIMAL TRAJECTORY PLANNING

 use some heuristic approach which allows to

OPTIMAL TRAJECTORY PLANNING

 There are two additional routes whose costs are

 Thus, this problem does not result in a unique

 Three school districts in Busville have a

distribution of Caucasians (C ) and African

Americans ( A) as shown in the table

 Implementation of the Supreme Court ruling on

 Given the distances between the districts,

number of students distance to district

THE ENVELOPE QUESTION

 On a television game show, the host subjects

“Switching is clearly advantageous. Suppose

THE ENVELOPE QUESTION

 The host is proposing a decision tree that looks

which does not correctly represent the situation

 Rather, we have for the two envelopes A and B

DECISION ANALYSIS PROTOTYPE

drill for oil 700,000 (100,000)

sell the land 90,000 90,000

probability 0.25 0.75

DECISION ANALYSIS PROTOTYPE

action expected payoff (k$)

1 0.25 (700) + 0.75 (-100) = 100

2 0.25 (90) + 0.75 (90) = 90

and the better choice is to drill for oil

DECISION ANALYSIS PROTOTYPE

 The decision maker must select an action from a

A GENERAL FRAMEWORK FOR DECISION

 For each contribution of an action and state of

nature, the value to the decision maker of the

consequences of an outcome is established and

quantified in terms of the payoff

 The payoff is defined as the quantity measure of

The course covers techniques that are useful

Expansion of production facilities/warehousing

Adoption of new technology

Implementation of a new production schedule

Changes in the production mix

Risk management in purchase/sale activities

Optimal scheduling of processes/projects

Development of the analytical framework for

Decisions are made by selecting from possible

The constraints on resources are given

Problem: determine the optimal daily mix, i.e.,

The relative “costs” for the various possible

The problem is to select the route that

use some heuristic approach which allows to

There are two additional routes whose costs are

Thus, this problem does not result in a unique

Three school districts in Busville have a

Implementation of the Supreme Court ruling on

Given the distances between the districts,

On a television game show, the host subjects

The host is proposing a decision tree that looks

Rather, we have for the two envelopes A and B

The decision maker must select an action from a

For each contribution of an action and state of

The payoff is defined as the quantity measure of

The payoffs are used to select the optimal action

We study decision analysis since its application

Prerequisite : ECE 210

Two exams: midterm and final

Statistical concepts: data analysis; statistical

The nature of a programming or optimization

The salient characteristics of a linear

The LP problem formulation

The LP problem solution

Extensive illustrations with numerical examples

You first quantify your assessment along the two

Use the defined variables to state the objective

Next consider the problem constraints:

only one pair of shoes can be selected

the decision variables are nonnegative

State the constraints in terms of x H and x L :

We determine the values x H and x L which

The supply department can provide daily up to 1

Determination of the objective: to maximize the

We define the decision variables to be

Sanity check for units of the objective function

We can graphically determine the optimal

The optimal solution of this problem is:

The objective value at the optimal solution is

The following assumptions hold:

The optimal solution of this problem is:

The objective value at the optimal solution is:

Constant Z lines are parallel and change

Iteration step: move to a better adjacent corner

Stopping rule: stop when the corner point

Key requirements of a programming problem:

to make a decision and so to define decision

to achieve some objective and so to formulate

to ensure that the decision satisfies certain

Words of caution: care with units and attention to

The problem is to determine the optimal

Definition of decision variables:

optimal assignment ⇒ minimum costs