Topic 4: Sensitivity analysis and dual problem

• This topic covers sensitivity analysis, the definition

of the dual problem, the relationship between the
optimal primal and dual solution, the economic
interpretation of duality, the dual Simplex method,
and primal-dual
primal dual computations.
• The topic is covered in the textbook
under Section 3.6 and Chapter 4.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • A resource is
i ddesignated
i t d as scarce if th
the activities
ti iti
(variables) of the model use the resource
p y Otherwise, the resource is abundant.
completely.
• This information can be read from the optimum
tableau by checking the value of the slack variable
associated with the constraint representing the
resource. If the slack value is zero , the resource is
used completely and , hence, is classified as
scarce. Otherwise, a positive slack indicates that
the resource is abundant.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 SENSITIVITY ANALYSIS

• IIn LP,
LP the
th input
i t data
d t (parameters)
( t ) off the
th model
d l can
change within certain limits without causing the
p
optimum solution to change.
g

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 GRAPHICAL SENSITIVITY ANALYSIS

• T
Two cases willill b
be considered:
id d
1. Sensitivity of the optimum solution to changes
in the availability of the resources (right-hand
(right hand
side of the constraints)
2. Sensitivity of the optimum solution to changes
i unit
in i profit
fi or unit
i cost ((coefficients
ffi i off the
h
objective function)

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 CHANGES IN THE RIGHT-HAND SIDE

• Consider
C id a company producing
d i ttwo products:
d t
Time (hour)
Product 1
Product 1 Product 2
Product 2 Available time
Available time
Machine 1 2 1 8
Machine 2 1 3 8
Profit SR 30 SR 20

• The LP model that will maximize the company’s

profit is
maximize z = 30 x1 + 20 x2
subject to 2 x1 + x2 ≤ 8
x1 + 3 x2 ≤ 8
x1 , x2 ≥ 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • C
Consider
id iincreasing
i ththe capacity
it off machine
hi 1
from 8 hours and 9 hours.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th rate
t off change
h in
i optimum
ti z iis

⎛ rate of revenue change resulting ⎞

⎜ ⎟ 142 − 128
⎜ from increasing machine 1 ⎟= = 14 SR/hour
⎜ capacity by 1 hour ⎟ 9−8
⎝ ⎠

• This rate represents the unit worth of the resource

(hours).
• The rate is usually referred to as dual or shadow
price.
• The dual price for machine 2 is 2 SR/hour (verify).
(verify)

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • C
Consider
id th
the ffollowing
ll i questions:
ti
– If the company can increase the capacity of one
machine, which machine should be
considered?

• For one additional hour, machine 1 will

increase the profit by SR 14 and machine 2
by only SR 2,
2 the capacity of machine 1
should be increased.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 – If the cost of increasing the capacity of either
machine 1 or 2 is 10 SR/hour, which machine
should be considered?

• For machine 1, the additional net revenue

per hour is 14-10=4.
14 10=4
• For machine 2, the net revenue per hour is
2-10=-8.
• Machine 1 should be considered.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 – If the
th capacity
it off machine
hi 1 is
i iincreased
d ffrom 8
to 13 hours, what would be the revenue?

• The increased revenue is

14(13 - 8) = SR 70.
• The total revenue = 128 + 70 = 198 riyals.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 CHANGES IN THE OBJECTIVE COEFFICIENTS

• Ch
Changes iin per-unitit profit
fit ((coefficients
ffi i t ini th
the
objective function) will change the slope of z.
• The optimum solution will remain at point C as
long as the objective function lies between lines
BF and DE.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Write
W it the
th objective
bj ti ffunction
ti ini th
the generall fformatt

maximize z = c1 x1 + c2 x2

• The optimum solution will not change as long as z

lies between the two lines x1 + 3x2 = 8 and
2x1 + x2 = 8.
• The ratio c1/c2 (slope) can vary between 1/3 and
2/1: 1 c1
≤ ≤2
3 c2

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 – If the
th unitit revenues off product
d t 1 and
d 2 are
changed to 35 and 25 riyals, will the current
p
optimal solution remain unchanged?g

• Since

1 35
≤ ≤2
3 25

the current optimal solution will remain

unchanged.
unchanged
• The new optimal z = 152 riyals.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 – If the
th coefficient
ffi i t off x2 is
i fifixed
d att c2 = 20,
20 what
h t
will the range of c1 be that will keep the current
p
optimal solution unchanged?g

• Substitute in the ratio

1 c1
≤ ≤2
3 20
or
20
≤ c1 ≤ 40
3

• This range
g is called the optimality
p y range.
g

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 ALGEBRIC SENSITIVITY ANALYSIS – CHANGES IN THE
RIGHT-HAND SIDE
• A ttoy company produces
d 3 types
t off toys:
t
Operation times (min)
1 2 3 Revenue
Train 1 3 1 3
Truck 2 0 4 2
Car 1 2 0 5
Available 
430 460 420
time (min)

• The LP model for this company is

maximize z = 3 x1 + 2 x2 + 5 x3
subject to x1 + 2 x2 + x3 ≤ 430
3 x1 + 2 x3 ≤ 460
x1 + 4 x2 ≤ 420
x≥0
Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th optimum
ti ttableau
bl iis
Basic  x1  x2  x3  x4  x5  x6  Solution 
z
z  4 0 0 1 2 0 1350
x2 ‐ 1/4 1 0 0.5 ‐ 1/4 0 100
x3 1 1/2 0 1 0  1/2 0 230
x6 2 0 0 ‐2
2 1 1 20

• Change the available time for operation 1 from

430 to 431 minutes:
maximize z = 3 x1 + 2 x2 + 5 x3
subject to x1 + 2 x2 + x3 ≤ 431
3 x1 + 2 x3 ≤ 460
x1 + 4 x2 ≤ 420
x≥0
Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th optimal
ti l ttableau
bl iis
Basic  x1  x2  x3  x4  x5  x6  Solution 
z  4 0 0 1 2 0 1351
x2 ‐ 1/4 1 0 0.5 ‐ 1/4 0 100.5
x3 1 1/2 0 1 0  1/2 0 230
x6 2 0 0 ‐2
2 1 1 18

• Hence, for one minute increase in operation 1

capacity, the profit increased by 1 riyal.
• So, the dual price for operation 1 is 1 riyal/minute.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Ch
Change th
the available
il bl titime ffor operation
ti 2 from
f
460 to 461 minutes:
maximize
i i z = 3 x1 + 2 x2 + 5 x3
subject to x1 + 2 x2 + x3 ≤ 430
3 x1 + 2 x3 ≤ 461
x1 + 4 x2 ≤ 420
x≥0
• The optimal tableau is
Basic  x1  x2  x3  x4  x5  x6  Solution 
z
z  4 0 0 1 2 0 1352
x2 ‐ 1/4 1 0 0.5 ‐ 1/4 0 99.75
x3 1 1/2 0 1 0  1/2 0 230.5
x6 2 0 0 ‐2
2 1 1 21

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • H
Hence, a one minute
i t iincrease iin operation
ti 2
capacity resulted in 2 riyals increase in the value
of z.
• So, the dual price of operation 2 is 2 riyals/minute.

• These dual prices can be read easily from the z-

row in the optimal tableau:
Basic  x1  x2  x3  x4  x5  x6  Solution 
z  4 0 0 1 2 0 1350

Dual price of operation 3
Dual price of operation 1

Dual price of operation 2

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Thi
This information
i f ti isi summarized
i d in
i the
th following
f ll i
table:

Resource Slack variable Optimum z‐row  Dual price

coefficient of the 
slack variable
Operation 1 x4 1 1 riyal / minute
Operation 2 x5 2 2 riyals / minutes
Operation 3 x6 0 0 riyal / minute

• Operation 3 is called abundant because x6 > 0

(there is excess supply from it),
it) hence
hence, increasing
the capacity of operation 3 is not going to make z
increase. That’s why the dual price is equal to
zero.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 DETERMINATION OF THE FEASIBILITY RANGES

• L
Lett D1, D2, and
d D3 be
b th
the changes
h iin operation
ti
capacities:
maximize z = 3 x1 + 2 x2 + 5 x3
subject to x1 + 2 x2 + x3 ≤ 430 + D1
3 x1 + 2 x3 ≤ 460 + D2
x1 + 4 x2 ≤ 420 + D3
x≥0
• The starting tableau with the new right-hand sides:
Solution 
Basic 
Basic x1 1 x2 2 x3 3 x4 4 x5 5 x6 6 RHS D1 D2 D3
z  ‐3 ‐2 ‐5 0 0 0 0 0 0 0
x5  1 2 1 1 0 0 430 1 0 0
x6 6 3 0 2 0 1 0 460 0 1 0
x6 1 4 0 0 0 1 420 0 0 1

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th optimal
ti l ttableau
bl iis
Solution 
Basic  x1  x2  x3  x4  x5  x6  RHS D1 D2 D3
z  4 0 0 1 2 0 1350 1 2 0
x2 ‐ 1/4 1     0      1/2 ‐ 1/4 0 100  1/2 ‐ 1/4 0
x3 1 1/2 0     1     0      1/2 0 230 0      1/2 0
x6 2 0 0 ‐2 1 1 20 ‐2 1 1

• The optimum solution is

z = 1350 + D1 + 2 D2
x2 = 100 + 12 D1 − 14 D2
x3 = 230 + 12 D2
x6 = 20 − 2 D1 + D2 + D3

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • S
Suppose D1 = 50,
50 D2 = -20,
20 D3 = -10,
10 the
th solution
l ti
becomes:
x2 = 100 + 12 (50) − 14 (−20) = 130 > 0
x3 = 230 + 12 (−20) = 220 > 0
x6 = 20 − 2(50) + (−20) + (−10) = −110 < 0
• These changes in the time capacities do not lead
t a feasible
to f ibl solution.
l ti

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • N
Now sett D1 = -30,
30 D2 = -12,
12 D3 = 10,
10 the
th solution
l ti
becomes:
x2 = 100 + 12 (−30) − 14 (−12) = 88 > 0
x3 = 230 + 12 (−12) = 224 > 0
x6 = 20 − 2(−30) + (−12) + (10) = 78 > 0
• These changes in the time capacities lead to a
f
feasible
ibl solution.
l ti

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • T
To find
fi d the
th feasibility
f ibilit range ffor th
the capacity
it off
operation 1, we consider changing D1 only:

x2 = 100 + 12 D1 ≥ 0 ⇒ D1 ≥ −200⎫
⎬ − 200 ≤ D1 ≤ 10
x6 = 20 − 2 D1 ≥ 0 ⇒ D1 ≤ 10 ⎭

• So as long as D1 is within this range, the solution

remains
i ffeasible.
ibl

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 ALGEBRAIC SENSITIVITY ANALYSIS-OBJECTIVE
FUNCTION
• W
We wantt to
t determine
d t i the
th ranges off changes
h in
i
the coefficients in the objective function that will
keepp the current optimal
p solution unchanged.
g
• The optimal tableau for the LP model that will
maximize the profit of the toy company is
Basic  x1  x2  x3  x4  x5  x6  Solution 
z  4 0 0 1 2 0 1350
x2 ‐ 1/4 1 0 0.5 ‐ 1/4 0 100
x3 1 1/2 0 1 0  1/2 0 230
x6 2 0 0 ‐2 1 1 20
• The optimal z-row
z row can be written as
z + 4 x1 + x4 + 2 x5 = 1350
or z = 1350 − 4 x1 − x4 − 2 x5

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • At th
the optimal
ti l solution,
l ti x1 = 0.
0 Increasing
I i x1 by b
one unit will reduce the optimal z by 4. Hence, it is
not economical to p produce any y train toys
y ((i.e. x1 =
0).
• We call the coefficient of x1 in the optimal z-row
the reduced cost of product 1. 1
• The reduced cost of a product is defined as

⎛ reduced cost ⎞ ⎛ cost of consumed ⎞ ⎛ revenue ⎞

⎜⎜ ⎟⎟ = ⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ per unit ⎠ ⎝ resources per unit ⎠ ⎝ per unit ⎠

This cost can only be determined 
from the LP model

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • A
An unprofitable
fit bl product
d t can b be made
d profitable
fit bl bby:
1. increasing the unit revenue
2 decreasing the unit cost of consumed
2.
resources

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • L
Lett di be
b th
the change
h iin th
the unit
it revenue off product
d t
i.
• The objective function becomes
maximize z = (3 + d1 ) x1 + ( 2 + d 2 ) x2 + (5 + d 3 ) x3

• The z-row in the starting tableau is

Basic  x1  x2  x3  x4  x5  x6  Solution 
z  ‐3‐d1  ‐2‐d2  ‐5‐d3  0 0 0 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th optimal
ti l ttableau
bl iis
Basic  x1  x2  x3  x4  x5  x6  Solution 

zz  4 1/4 d2 + 3/2 d

4‐1/4 d + 3/2 d3 ‐ d
d1 0 0 1‐1/4 d
1 1/4 d2 + 1/2 d
+ 1/2 d3 2 0 1350+100 d
1350+100 d2 + 230 d
+ 230 d3
x2 ‐ 1/4 1 0 0.5 ‐ 1/4 0 100
x3 1 1/2 0 1 0  1/2 0 230
x6 2 0 0 ‐2 1 1 20

• Hence, changes in the objective function

coefficients can affect the optimality of the solution
only.
l

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • O
One convenient
i t way for
f computingti the
th new
reduced costs is to do the following:
add
dd d1 d2 d3 0 0 0 add
Basic  x1  x2  x3  x4  x5  x6  Solution 
1 z  4 0 0 1 2 0 1350
d2 x2 ‐ 1/4
/ 1 0 0.5 ‐ 1/4
/ 0 100
d3 x3 1 1/2 0 1 0  1/2 0 230
0 x6 2 0 0 ‐2 1 1 20

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th new reduced
d d costt ffor x1 is
i computed
t d as
d1
L f C l
Left Column x1  x1 column x left column
1 l l f l
1 4 4x1
d2 ‐ 1/4 ‐1/4 d2
d3 1 1/2
1 1/2 3/2 d3
3/2 d
0 2 2 x 0

reduced cost = 4 − 14 d 2 + 23 d 3 − d1

• Applying the same computations to z-row

prod ces
produces
z = 1350 + 100d 2 + 230d 3

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • H
Hence, th
the currentt optimal
ti l solution
l ti willill remain
i
optimal as long as the coefficients in the optimal z-
row are ggreater than or equal
q to zero:
4 − 14 d 2 + 23 d 3 − d1 ≥ 0
1 + 12 d 2 ≥ 0
2 − 14 d 2 + 12 d 3 ≥ 0

• If d1 = -1,
1 d2 = -1,
1 d3 = 1,
1 will
ill th
the solution
l ti remain
i
optimal?
4 − 14 ( −1) + 23 (1) − ( −1) = 6.75 > 0
1 + 12 ( −1) = 0.5 > 0 ⇒ Current solution 
remains optimal
2 − 14 ( −1) + 12 (1) = 2.75 > 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th optimal
ti l z = 1350 +100(-1)
100( 1) + 230(1) = 1480.
1480

• If d2 = d3 = 0, the range of d1 is
4 – d1 ≥ 0 → -∞ < d1 ≤ 4
• If d1 = d3 = 0, the range of d2 is
-2 ≤ d2 ≤ 8
• Hence, the unit revenue for product 2 is between 2
– 2 = 0 and 2 + 8 = 10 riyals.
• This means that as long as the unitnit re
revenue
en e of
truck toys is between 0 and 10 riyals, the optimal
solution will not change.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 DUAL PROBLEM

• Th
The dual
d l problem
bl iis an LP d
defined
fi d di
directly
tl and
d
systematically from the primal (original) LP and the
p
optimal solution of one pproblem directly
ypprovides
the optimal solution to the other.
• To construct the dual problem from the primal, the
primal problem has to be written in the equation
form:
– All constraints are equations
q
– Nonnegative right-hand side
– All variables ≥ 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Define
D fi ththe primal
i l problem
bl iin equation
ti fform as
n
e oor minimizee z = ∑ c j x j
maximize
a
j =1
n
subject to ∑a xj =1
ij j = bi , i = 1,2,..., m

x j ≥ 0, j = 1,2,..., n

• The variables xj include all surplus, slack, and

artificial variables.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • D
Duall iis constructed
t t d ffrom th
the d
duall iin th
the ffollowing
ll i
manner:
Primal variables
x1 x2 … xj … xn
Dual  Right 
variables c1 c2 … cj … cn hand
hand 
side
y1 a11 a12 … a1j … a1n b1
y2 a21 a22 … a2j … a2n b2
…

…
ym am1 am2 … amj … amn bm

Dual 
jth dual constraint objective 
coefficients

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The dual
d l iis constructed
t t d as ffollows:
ll
– A dual variable is defined for each primal
equation
– A dual constraint is defined for each primal
variable
– The constraint (column) coefficients of a primal
variable define the left hand side coefficients of
the dual constraints and its objective coefficient
define the right hand side
– The objective coefficients of the dual are the
right hand side of the primal constraint
equations

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th rules
l ffor constructing
t ti the th dual
d l are
Primal  Dual problem
problem
p Objective Constraint 
Constraint Variable sign
Variable sign
objective type
Max Min ≥ Unrestricted
Min Max ≤ Unrestricted

• Take the following LP

maximize z = 5 x1 + 12 x2 + 4 x3
subject t x1 + 2 x2 + x3 ≤ 10
bj t to
2 x1 − x2 + 3 x3 = 8
x1 , x2 , x3 ≥ 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th LP iin equation
ti fform
maximize z = 5 x1 + 12 x2 + 4 x3 + 0 x4
bj to x1 + 2 x2 + x3 + x4 = 10
subject
2 x1 − x2 + 3 x3 = 8
x1 , x2 , x3 , x4 ≥ 0

• There will be one dual variable for each primal

constraint, call them y1 and y2.
• The coefficients in the dual objective function are
the values of the right hand sides of the primal
constraints:
minimize w = 10 y1 + 8 y2

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The coefficients
ffi i t off x1 in
i the
th primal
i l objective
bj ti
function and the constraints define the first dual
constraint:
y1 + 2 y2 ≥ 5

• Carrying out the procedure in this manner, the

dual LP will be
minimize w = 10 y1 + 8 y2
subject to y1 + 2 y2 ≥ 5
2 y1 − y 2 ≥ 12
y1 + 3 y 2 ≥ 4
y1 ≥0
y 2 unrestricted

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 SIMPLEX TABLEAU LAYOUT

• Th
The starting
t ti and d generall Si
Simplex
l ttableaus
bl can b
be
represented as

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The optimal
ti l solution
l ti off th
the d
duall problem
bl can b
be
computed from the primal optimal solution.
• There are two methods of computing the optimal
dual solution:
– Method 1

⎛ optimal value ⎞ ⎛ optimal z - row coefficient ⎞ ⎛ unit profit ⎞

⎜⎜ ⎟⎟ = ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟
⎝ of yi ⎠ ⎝ of starting
g basic x ⎠ ⎝ of x ⎠
– Method 2
⎛ row vector of original ⎞
⎜ ⎟ ⎛ optimal
ti l ⎞
⎛ optimal values ⎞ ⎜ z - coefficients ⎟ ⎜ ⎟
⎜⎜ ⎟⎟ = ⎜ ⎟ × ⎜ primal ⎟
⎝ of dual variables ⎠ ⎜ of optimal primal ⎟ ⎜⎝ inverse ⎟⎠
⎜ basic variables ⎟
⎝ ⎠
Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Take
T k the
th following
f ll i LP
maximize z = 5 x1 + 12 x2 + 4 x3
subject to x1 + 2 x2 + x3 ≤ 10
2 x1 − x2 + 3 x3 = 8
x1 , x2 , x3 ≥ 0

• Writing the LP in equation form, we get

maximize z = 5 x1 + 12 x2 + 4 x3 + 0 x4 − MR
j to x1 + 2 x2 + x3 + x4 = 10
subject
2 x1 − x2 + 3 x3 +R =8
x1 , x2 , x3 , x4 , R ≥ 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th dual
d l iis
minimize w = 10 y1 + 8 y2
subject to y1 + 2 y2 ≥ 5
2 y1 − y2 ≥ 12
y1 + 3 y2 ≥ 4
y1 ≥0
y2 ≥ − M → y2 unrestricted
• The optimal solution of the primal is

Basic  x1 1 x2 2 x3 3 x4 R Solution 

z  0     0      3/5 5 4/5 ‐2/5+M 54  4/5 
x2 0     1     ‐ 1/5  2/5 ‐ 1/5 2 2/5
x1 1     0     1 2/5 1/5  2/5 5 1/5

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The starting
t ti b basic
i variables
i bl are x4 anddR R, which
hi h
will be used to compute the values of the optimal
y1 and y2.
• Using Method 1,
⎛ optimal
p z - row coefficient ⎞ ⎛ unit pprofit ⎞
y1 = ⎜⎜ ⎟ + ⎜⎜
⎟ ⎟⎟
⎝ of starting x4 ⎠ ⎝ of x4 ⎠
29
= +0
5
29
=
5

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 and
d
⎛ optimal z - row coefficient ⎞ ⎛ unit profit ⎞
y2 = ⎜⎜ ⎟⎟ + ⎜⎜ ⎟⎟
⎝ off starting
i R ⎠ ⎝ off R ⎠
2
=− +M −M
5
2
=−
5

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The optimal
ti l iinverse matrix,
t i which
hi h iis d
defined
fi d bby th
the
columns under the starting basic variables x4 and
R, is

⎡ 2
5
−1
5 ⎤ Row of x2
⎢ 1 2 ⎥
⎣ ⎦
Row of x1
5 5

Column 
Column Column 
Column
of x4 of R
x2 first, x1 second
• Hence
Hence,
⎛ row vector of original ⎞
⎜ ⎟
⎜ z - coefficients ⎟ = ( x2 , x1 ) = (12,5)
⎜ of primal basic variables ⎟
⎝ ⎠
Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th optimal
ti l dduall variables
i bl are
⎡ 52 −1
⎤
( y1 , y2 ) = (12,5)⎢ 1 5
2 ⎥
⎣5 5 ⎦
= ( 295 ,− 52 )

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • For
F any pair
i off feasible
f ibl primal
i l and
ddduall solutions:
l ti

⎛ z in maximization ⎞ ⎛ w in minimization ⎞
⎜⎜ ⎟⎟ ≤ ⎜⎜ ⎟⎟
⎝ problem ⎠ ⎝ problem ⎠

• When the primal solution and the dual solution are

optimal,
z=w

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 SIMPLEX TABLEAU COMPUTATIONS

• A
Any iteration
it ti off the
th entire
ti Simplex
Si l ttableau
bl can bbe
generated from the original data of the problem,
the inverse associated with the iteration, and the
dual problem.
• The computations of the constraint columns (left
and right hand) are as:
⎛ Constraint column ⎞ ⎛ inverse in ⎞ ⎛ original ⎞
⎜⎜ ⎟⎟ = ⎜⎜ ⎟⎟ × ⎜⎜ ⎟⎟
⎝ in
i iteration
it ti i ⎠ ⎝ iteration
it ti i ⎠ ⎝ constraint
t i t column
l ⎠

• The computations
p of the z-row coefficient of xj are
as
⎛ primal z - row ⎞ ⎛ left hand side ⎞ ⎛ right hand side of ⎞
⎜⎜ ⎟⎟ = ⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ coefficien
ffi i t off x j ⎠ ⎝ off jth dual
d l constriant
t i t ⎠ ⎝ jth dual
d l constraint
t i t⎠

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • In
I the
th previous
i example,
l ththe optimal
ti l iinverse iis

⎡52 −1
5 ⎤
⎢1 2 ⎥
⎣5 5 ⎦
• The
e co
column
u o
of x1 in tthe
e opt
optimal
a tab
tableau
eau ca
can be
computed as
⎛ column ⎞ ⎛ original ⎞
⎜⎜ ⎟⎟ = (inverse)× ⎜⎜ ⎟⎟
⎝ of x1 ⎠ ⎝ column of x1 ⎠
⎡ 52 −51 ⎤ ⎡1 ⎤ ⎡0⎤
= ⎢1 2 ⎥ × ⎢ ⎥ = ⎢ ⎥
⎣ 5 5 ⎦ ⎣2⎦ ⎣1 ⎦

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • T
To compute
t the
th optimal
ti l z-row coefficients,
ffi i t we
need the dual solutions, which were computed as
( y1 , y 2 ) = ( 295 ,− 52 )

• The z-row coefficient of x1 is computed as

z − row coefficient of x1 = y1 + 2 y 2 − 5 = 29
5 + 2(− 52 ) − 5 = 0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 ECONOMIC INTERPRETATION OF DUALITY

• Th
The LP problem
bl can b
be viewed
i d as a resource
allocation model in which the objective is to
j
maximize revenue subject to the availability
y of
limited resources.
• Consider the correspondence between the primal
and dual LPs:
n m
maximize z = ∑ c j x j minimize w = ∑ bi yi
j =1 i =1

⇔
m

∑a
n
subject to ∑a
j =1
ij x j ≤ bi , i = 1,2,..., m subject to
i =1
ij yi ≥ c j , j = 1,2,..., n

x j ≥ 0, j = 1,2,..., n yi ≥ 0, i = 1,2,..., m

Primal Dual

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The primal
i l LP hash n economic i activities
ti iti and
dm
resources. The coefficient cj represents the unit
revenue of activityy jj. Resource j has a maximum
availability bj and is consumed at the rate of aij
units per unit of activity j.
• The primal and dual solutions must satisfy
n m
z = ∑ c j x j ≤ ∑ bi yi = w
j =1 i =1

• For the primal, z represents the total revenue from

all activities, hence
revenue = ∑ (units of resource i )× (unit revenue of resource i )
i

= ∑ (bi )× ( yi )
i

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • H
Hence, th the d
duall variable
i bl yi represents
t the
th unit
it
worth of resource i (dual or shadow price).
• For any two feasible primal and dual solutions, the
inequality z < w is interpreted as
(revenue) < (worth of resources)
• This means that as long as the total revenue from
all the activities is less than the worth of the
resources the primal and dual solutions are not
resources,
optimal.
• Optimality is reached when the resources have
been utilized completely.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • At any primal
i l ititeration:
ti
⎛ left hand side of dual ⎞ ⎛ right hand side ⎞
objective coefficient of x j = ⎜⎜ ⎟⎟ − ⎜⎜ ⎟⎟
⎝ constraint j ⎠ ⎝ of dual constraint j⎠
m
= ∑ aij yi − c j
i =1

• cj is riyals per unit (unit revenue)

m
• The quantity ∑ aij yi is the cost of resources for
i =1

activity jj, which is given by

m m
⎛ usage of resource i ⎞⎛ cost per unit ⎞
∑
i =1
aijj yi = ∑ ⎜⎜
i =1 ⎝ per unit
it off activity
ti it
⎟⎟⎜⎜ ⎟⎟
j ⎠⎝ off resource i ⎠

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The resources costt per unit it off resource i is
i given
i
by yi. m
• The quantity ∑ aij yi − c j is the reduced cost of
i =1

activity j.
• In
I maximization,
i i i an iincrease iin the
h llevell off an
unused (nonbasic) activity j can improve revenue
onlyy if its reduced cost is negative.
g
• This condition is stated as

⎛ costt ⎞
⎜ ⎟ ⎛ revenue per unit ⎞
⎜ of resources used by one unit ⎟ < ⎜⎜ ⎟⎟
⎜ of activity j ⎟ ⎝ oof act
activity
v ty j ⎠
⎝ ⎠
Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
Thus, it is
i economically
i ll advantageous
d t tto iincrease
an activity to a positive level if its unit revenue
p
exceeds its unit imputed cost.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 DUAL SIMPLEX ALGORITHM

• Th
The dual
d l simplex
i l method th d starts
t t with
ith a b
better
tt ththan
optimal and infeasible basic solution.
• Dual feasibility condition:
The leaving variable xr is the basic variable
having the most negative value. If all the basic
variables
i bl are nonnegative, i the
h algorithm
l i h ends. d

• Dual optimality condition:

Given that xr is the leaving variable, let c j be the
reduced cost of nonbasic variable xj and αrj the
constraint coefficient in the xr-row and xj-column
of the tableau. The entering variable is the
nonbasic variable with αrjj < 0 that

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 corresponds
d to
t
min
nonbasic x j
{ cj
α rj }
, α rj < 0

if αrj ≥ 0 for
o aall nonbasic
o bas c xj, the
t e problem
p ob e has
as no
o
feasible solution.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Th
The two
t requirements
i t off the
th dual
d l Simplex
Si l are
1. The objective function must satisfy the
optimality condition of the regular Simplex
method.
2. All the constraints are of type ≤.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • Take
T k this
thi LP
minimize z = 3 x1 + 2 x2 + x3
subject to 3 x1 + x2 + x3 ≥ 3
− 3 x1 + 3 x2 + x3 ≥ 6
x1 + x2 + x3 ≤ 3
x≥0

• Convert all constraints to ≤:

minimize z = 3 x1 + 2 x2 + x3
subject to − 3 x1 − x2 − x3 ≤ −3
3 x1 − 3 x2 − x3 ≤ −6
x1 + x2 + x3 ≤ 3
x≥0

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th starting
t ti ttableau
bl i
is
Basic  x1  x2  x3  x4  x5  x6  Solution 
z  ‐3 ‐2 ‐1 0 0 0 0
x4 ‐3 ‐1 ‐1 1 0 0 ‐3
x5 3 ‐3 ‐1 0 1 0 ‐6
x6 1 1 1 0 0 1 3

• The tableau is optimal but not feasible.

• The leaving variable is x5, because it has the most
negati e value.
negative al e

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • To
T determine
d t i ththe entering
t i variable,
i bl apply
l th
the ratio:
ti
Non basic  x1 x2 x3
variable
z‐row ‐3 ‐2 ‐1
x5‐row 3 ‐3 ‐1
Ratio ‐ 2/3 1

• Having the smallest ratio, x2 enters the basic

solution.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th tableau
t bl b
becomes
Basic  x1  x2  x3  x4  x5  x6  Solution 
z  ‐5     0     ‐ 1/3
/ 0     ‐ 2/3
/ 0     4    
x4 ‐4     0     ‐ 2/3 1     ‐ 1/3 0     ‐1    
x2 ‐1     1      1/3 0     ‐ 1/3 0     2    
x6 2     0     2/3 0     1/3 1     1    

• The tableau is still not feasible.

• By taking the ratio, x3 must enter the basic
solution.

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM
 • The
Th tableau
t bl b
becomes
Basic  x1  x2  x3  x4  x5  x6  Solution 
zz  ‐3    
3 0    
0 0    
0 ‐ 1/2
1/2 ‐ 1/2
1/2 0    
0 4 1/2
4 1/2
x3 6     0     1     ‐1 1/2  1/2 0     1 1/2
x2 ‐3     1     0      1/2 ‐ 1/2 0     1 1/2
x6 ‐2    
2 0
0     0
0     1
1     0
0     1
1     0
0    

• The tableau is now feasible.

• This is the end of the dual Simplex and the optimal
solution is
x1 = 0
x2 = 3/2
x3 = 3/2
z = 9/2

Dr Muhammad Al‐Salamah, Industrial Engineering, KFUPM