Simplex Method

RK Jana

1
 Lecture Outline

 Mathematical fundamentals of LPP

 Slack, surplus and artificial variables
 Computational procedure of Simplex Method
 Justification & significance of elements in Simplex
tableau
 Solving maximization and minimization LPP
 Technical issues involved in Simplex Method
 Software for solving LPP
Mathematical Fundamentals of LPP
 Corner-point Solution

 A corner-point is the point defined by the

intersection of constraint boundaries
 In two dimensions – intersection of two lines
 In higher dimensions – intersection of n (n-1)
dimensional “hyperplanes”
 The corner-point solution
 Feasible
 Infeasible

4
Example

Maximize Z = 3x1 + 5x2

Subject to
x1 4
2x2 12
3x1 + 2x2 18
x1 , x2 0

5
Corner-points Highlighted
x2 3x1+2x2=18 Both corner-point feasible and
(0, 9)
infeasible solutions are shown.
x1=4

(2, 6) (4, 6)
(0, 6) 2x2=12

(4, 3)
(4, 0) and (2, 6) are adjacent to (4, 3)

(0, 0)
(4, 0) (6, 0) x1

6
 Properties of Corner-Point
Solutions
1. If there is exactly one optimal solution, then
it must be a corner-point solution
2. If there are multiple optimal solutions, then
at least two solutions are adjacent corner-
point solutions
3. There are a finite number of corner-point
feasible solutions
4. If a corner-point feasible solution has no
better adjacent corner-point feasible
solution, then it is optimal
7
 The Simplex Method

 When decision variables are more than 2, it is always advisable

to use Simplex Method to avoid lengthy graphical procedure.

 The simplex method is not used to examine all the feasible

solutions.

 It deals only with a small and unique set of feasible solutions, the
set of vertex points (i.e., extreme points) of the convex feasible
space that contains the optimal solution.
The simplex method is a method for searching corner point
feasible solutions
8
 Continued…

 Locate an extreme point of the feasible region

 Examine each boundary edge intersecting at this point to
see whether movement along any edge increases the value
of the objective function
 If the value of the objective function increases along any
edge, move along this edge to the adjacent extreme point. If
several edges indicate improvement, the edge providing the
greatest rate of increase is selected
 Repeat second & third steps above until movement along
any edge no longer increases the value of objective function

9
 Geometric Interpretation

The Simplex method starts at a corner-point feasible

solution (usually the origin) and moves to adjacent CP
solutions along edges until an optimal solution is found.

 A CPF solution is connected to adjacent CPF solutions by

edges

 In a problem with n decision variables, each CPF solution

will have n edges emanating from it

10
 Continued…

 A CPF solution lies at the intersection of n constraint

boundaries.

 By relaxing a single constraint boundary equation (i.e.

equality for the constraint is relaxed) an edge is defined
which leads to an adjacent CPF solution.

 The simplex method defines

 How to choose which edge to traverse.
 When to stop or how to determine that a CPF solution is optimal

11
 Identifying Solution Type
Using Simplex Method
 Bounded feasible region exists
 If there is exactly one optimal solution, then it can
find the solution efficiently
 If there are multiple optimal solutions, then it can
identify the case & can find all solutions
 No bounded feasible region exists
 Unbounded feasible region can be identified
 No feasible region can be identified

12
 Slack, surplus and artificial
variables

 Slack Variable
 Surplus Variable
 Artificial Variable

13
Computational Procedure of
Simplex Method
LPP in Vector Notations

Maximize Z = cx

subject to
Ax = b, x 0

where c = (c1, c2, …, cj, …, cn)

x = (x1, x2, …, xj, …, xn)T
A = (a1, a2, …, aj, …, an)
b = (b1, b2, …, br…, bm)T
15
Notations

aj = a column vector whose elements are

coefficients of xj in A

B = initial basis containing m-columns of A

xB = initial basic feasible solution
b = requirement vector

16
 Initial Simplex Table

cj c1 c2 … cn

cB xB b x1 x2 … xn

cB1 xB1 b1 a11 a12 … a1n

cB2 xB2 b2 a21 a22 … a2n

… … … … … … …

cBm xBm bm am1 am2 … amn

z j – cj z 1 – c1 z2 –c2 … z n – cn

zj – cj = cB aj - cj= cB1 a1j + cB2 a2j + … + cBm amj - cj 17

 Characteristics

 Objective value
Z = cB xB = cB1 xB1 + cB2 xB2 + … + cBm xBm

 The Quantity Zj
zj = cB aj = cB1 a1j + cB2 a2j + … + cBm amj

18
 Characteristics

 Net evaluation
zj – cj = cB aj - cj= cB1 a1j + cB2 a2j + … + cBm amj – cj
 Optimality condition: zj - cj 0
 Unbounded solution: zj - cj 0 and aij 0

19
 Characteristics

 Alternate optimal solution

zj - cj =0for some non-basic variable & aij 0
for at least one j
 No feasible solution
If at any stage of the simplex method the optimality
condition is satisfied and still at least one artificial
variable remains in the basis at the positive level, then
the LPP has no feasible solution.

20
 Entering & Departing Vector Rules
(maximization problem)

 Entering Vector Rule (1):

xk will enter the basis based on the condition:
zk - ck = Minj { zj - cj | zj - cj 0 }

 Departing Vector Rule (2):

xr will leave the basis based on the condition:
Mini { xBi / aij | aij 0 }

21
 Simplex Algorithm
(for a maximization problem)
 Step 1: If the given LPP is of minimization type, then
convert it to a maximization type problem.
 Step 2: If any of the components of the requirement
vector is negative, multiply the corresponding
constraint by (-1) and adjust the direction of inequality.
If the constraint is an equality then multiply it by (-1)
only.
 Step 3: Add slack, surplus & artificial variables to the
constraints according to their requirement.

22
 Continued…

 Step 4: Assign coefficients of slack & surplus variable

as 0, and (-M) for artificial variables in the changed
objective function.
 Step 5: Construct the simplex table by choosing the
initial basic feasible solution.
 Step 6: Check the optimality of solution. If all
(zj - cj 0 ), then the present solution is optimal.
 Step 7: If for at least one aj , zj - cj 0 and aij 0 then
the problem has unbounded solution and stop there.

23
 Continued…

 Step 8: Choose the entering vector according to Rule

1. In case of tie, chose any one arbitrarily.
 Step 9: Choose the departing vector according to Rule
2. In case of tie, chose any one arbitrarily.
 Step 10: Form the new basis by dropping the departing
variable and introducing the new variable. Convert the
key element to unity and all other elements in its
column to zero.
 Step 11: Go to Step 6 and repeat the procedure until
either an optimum solution is obtained or there is an
indication of unbounded solution.
24
 Example

Max Z = x1 + x2 + 3x3
subject to
3x1 + 3x2 + x3 3
2x1 + x2 + 2x3 2
x1 , x2, x3 0

25
Continued…

Max Z = x1 + x2 + 3x3 + 0.s1 + 0.s2

subject to
3x1 + 3x2 + x3 + 1.s1 + 0.s2 =3
2x1 + x2 + 2x3 + 0.s1 + 1.s2 =2
x1, x2, x3, s1, s2 0

26
Continued…

Table 1
cj 1 1 3 0 0

cB xB b x1 x2 x3 s1 s2

0 s1 3 3 3 1 1 0

0 s2 2 2 1 2 0 1

zj – c j -1 -1 -3 0 0

27
 Continued…

Table 2
cj 1 1 3 0 0

cB xB b x1 x2 x3 s1 s2

0 s1 2 2 5/2 0 1 -1/2

3 x3 1 1 1/2 1 0 1/2

zj – cj 2 1/2 0 0 3/2

Optimum solution: x1 = x2 = 0, x3 = 1, max Z = 3

28
 Example

Max Z = 3x1 + 2x2 + x3

subject to
3x1 + 4x2 + 3x3 42
5x1 + 3x3 45
3x1 + 6x2 + 2x3 41
x1 , x2, x3 0
29
Continued…

Table 1
cj 3 2 1 0 0 0

cB xB b x1 x2 x3 s1 s2 s3

0 s1 42 3 4 3 1 0 0

0 s2 45 5 0 3 0 1 0

0 s3 41 3 6 2 0 0 1

zj – cj -3 -2 -1 0 0 0
30
Continued…

Table 2
cj 3 2 1 0 0 0

cB xB b x1 x2 x3 s1 s2 s3

0 s1 15 0 4 6/5 1 -3/5 0

3 x1 9 1 0 3/5 0 1/5 0

0 s3 14 0 6 1/5 0 -3/5 1

zj – cj 0 -2 4/5 0 3/5 0
31
Continued…
Table 3
cj 3 2 1 0 0 0

cB xB b x1 x2 x3 s1 s2 s3

0 s1 17/3 0 0 16/15 1 -1/5 -2/3

3 x1 9 1 0 3/5 0 1/5 0

2 x2 7/3 0 1 1/30 0 -1/10 1/6

zj – cj 0 0 13/15 0 2/5 1/3

Optimum solution: x1 =9; x2 = 7/3, x3 = 0, max Z = 95/3 32

Example (C.W.)

Min Z = x1 + x2

Subject to
5x1 + 9x2 45
x1 + x2 2
x2  4
x1 , x2 0

33
 Example

Max Z = 107x1 + x2 + 2x3

subject to
14x1 + x2 - 6x3 +3x4 =7
16x1 + x2/2 - 6x3 5
3x1 - x2 - x3 0
x1 , x2, x3 , x4 0
34
 Continued…

Max Z = 107x1 + x2 + 2x3 + 0.x4 +0.s1

+0.s2
subject to
14x1/3+ x2/3 - 2x3 + x4 = 7/3
16x1 + x2/2 - 6x3 + s1 = 5
3x1 - x2 - x3 + s2 = 0
x1 , x2, x3 , x4 , s1, s2 0
35
Continued…

Table 1
cj 107 1 2 0 0 0

cB xB b x1 x2 x3 x4 s1 s2

0 x4 7/3 14/3 1/3 -2 1 0 0

0 s1 5 16 1/2 -6 0 1 0

0 s2 0 3 -1 -1 0 0 1

zj – cj -107 -1 -2 0 0 0
36
 Continued…

Table 2
cj 107 1 2 0 0 0

cB xB b x1 x2 x3 x4 s1 s2

0 x4 7/3 0 17/9 -4/9 1 0 -114/9

0 s1 5 0 35/6 -2/3 0 1 -16/3

107 x1 0 1 -1/3 -1/3 0 0 1/3 Unbounded

solution.
zj – cj 0 -101/3 -113/3 0 0 -107/3
37
 Additional Problems
Max Z = 5x1 + 3x2 Max Z = x1 - x2 + 3x3
s. t. 3x1 + 5x2 15 s. t. x1 + x2 + x3 10
5x1 + 2x2 10 x1 - x 3 2
x1 , x2 0 2x1 -2x2 + 3x3 0
x1 , x2 , x3 0
( x1 = 20/19, x2 = 45/19 ( x1 = 0, x2 = 6, x3 = 4
Objective value = 235/19) Objective value = 6)

Max Z = 4x1 + x2 + 3x3 + 5x4 Min Z = x1 - 3x2 + 2x3

s. t. 4x1 - 6x2 - 5x3 - 4x4 -20 s. t. 3x1 - x2 + 2x3 7
3x1 - 2x2 - 4x3 + x4 10 -2x1 + 4x2   12
8x1 - 3x2 + 3x3 + 2x4 20 -4x1 + 3x2 + 8x3  10
x1 , x2 , x3, x4 0 x1 , x2 , x3 0
( x1 = 4, x2 = 5, x3 = 0
(Unbounded solution) Objective value = -11)

38
Big M-Method of A. Charnes
 Example

Max Z = 2x1 + 3x2

subject to
x1 + x2 8
x1 + 2x2 =5
2x1 + x2 8
x1 , x2 0
40
 Continued…

Max Z = 2x1 + 3x2 + 0.s1 - M.s2 + 0.s3

subject to
x1 + x2 + 1.s1 + 0.s2 + 0.s3 =8
x1 + 2x2 + 0.s1 + 1.s2 + 0.s3 =5
2x1 + x2 + 0.s1 + 0.s2 + 1.s3 =8
x1, x2, s1, s2, s3 0

41
Continued…

Table 1
cj 2 3 0 -M 0

cB xB b x1 x2 s1 s2 s3

0 s1 8 1 1 1 0 0

-M s2 5 1 2 0 1 0

0 s3 8 2 1 0 0 1

zj – cj -M-2 -2M-3 0 0 0
42
Continued…

Table 2
cj 2 3 0 -M 0

cB xB b x1 x2 s1 s2 s3

0 s1 11/2 1/2 0 1 … 0

3 x2 5/2 1/2 1 0 … 0

0 s3 11/2 3/2 0 0 … 1

zj – cj -1/2 0 0 … 0

43
Continued…

Table 3
cj 2 3 0 -M 0

cB xB b x1 x2 s1 s2 s3

0 s1 11/3 0 0 1 … -1/3

3 x2 2/3 0 1 0 … -1/3

2 x1 11/3 1 0 0 … 2/3

zj – cj 0 0 0 … 1/3

Optimum solution: x1 = 11/3, x2 = 2/3, max Z = 28/3

44
 Example

Min Z = 4x1 + 3x2

subject to
2x1 + x2 10
-3x1 + 2x2 6
x1 + x2 6
x1 , x2 0
45
 Continued…

Max Z* = -4x1 - 3x2 + 0.s1 + 0.s2 + 0.s3

subject to
2x1 + x2 - 1.s1 = 10
-3x1 + 2x2 + 1.s2 =6
x1 + x2 -1.s3 = 6
x1, x2, s1, s2, s3 0

46
 Continued…

Max Z* = - 4x1 - 3x2 + 0.s1 + 0.s2 + 0.s3 - M.s4 - M.s5

subject to
2x1 + x2 - 1.s1 + 0.s2 + 0.s3 + 1.s4 + 0.s5 = 10
-3x1 + 2x2 + 0.s1 + 1.s2 + 0.s3 + 0.s4 + 0.s5 =6
x1 + x2 + 0.s1 + 0.s2 - 1.s3 + 0.s4 + 1.s5 = 6
x1, x2, s1, s2, s3, s4, s5 0

47
Continued…

Table 1
cj -4 -3 0 0 0 -M -M

cB xB b x1 x2 s1 s2 s3 s4 s5

-M s4 10 2 1 -1 0 0 1 0

0 S2 6 -3 2 0 1 0 0 0

-M s5 6 1 1 0 0 -1 0 1

zj – cj -3M+4 -2M+3 M 0 M 0 0
48
Continued…

Table 2
cj -4 -3 0 0 0 -M -M

cB xB b x1 x2 s1 s2 s3 s4 s5

-4 x1 5 1 1/2 -1/2 0 0 -- 0

0 S2 21 0 7/2 -3/2 1 0 -- 0

-M s5 0 0 1/2 1/2 0 -1 -- 1

-M/2 -M/2
zj – cj 0 0 M -- 0
+1 +2 49
Continued…
Table 3
cj -4 -3 0 0 0 -M -M

cB xB b x1 x2 s1 s2 s3 s4 s5

-4 x1 4 1 0 -1 0 1 -- --

0 S2 14 0 0 -5 1 7 -- --

-3 x2 2 0 1 1 0 2 -- --

zj – cj 0 0 1 0 2 -- --

Optimum solution: x1 = 4, x2 = 2, min Z = 22

50
 Example

Max Z = 3x1 + 2x2

subject to
2x1 + x2 2
3x1 + 4x2 12
x1 , x2 0

51
Continued…

Table 1
cj 3 2 0 0 -M

cB xB b x1 x2 s1 s2 s5

0 s1 2 2 1 1 0 0

-M s3 12 3 4 0 -1 1

zj – cj -3M-3 -4M-2 0 M 0

52
Continued…
Table 2
cj 3 2 0 0 -M

cB xB b x1 x2 s1 s2 s5

2 x2 2 2 1 1 0 0

-M s3 4 -5 0 -4 -1 1

zj – cj 5M+1 0 4M+2 M 0

All zj – cj  0 and an artificial variable s3 appears in the basis.

So, the LPP does not possess any feasible solution.
53
 Additional Problems
Min Z = 12x1 + 20x2 Max Z = 5x1 - 2x2 + 3x3
s. t. 6x1 + 8x2 100 s. t. 2x1 + 2x2 - x3 2
7x1 + 12x2 120 3x1 - 4x2 3
x1 , x2 0 x2 + 3x3 5
x1 , x2 , x3 0
( x1 = 15, x2 = 5/4 ( x1 = 23/3, x2 = 5, x3 = 0
Objective value = 205) Objective value = 85/3)

Min Z = x1 + x2 + x3 Min Z = 2x1 + x2

s. t. x1 - x2 + 2x3 =2 s. t. 3x1 + x2 =3
-x1 + 2x2 - x3 =1 4x1 + 3x2  6
x1 , x2 , x3 0 x1 + 2x2  3
( x1 = 0, x2 = 4/3, x3 = 5/3 x1 , x2 0
Objective value = 3) ( x1 = 3/5, x2 = 6/5
Objective value = 12/5)

54
 Continued…

Max Z = x1 - x2 + 2x3 Min Z = x1 - 2x2 - 3x3

s. t. -x1 + x2 + 2x3 5 s. t. -2x1 + x2 + 3x3 =2
-2x1 + 5x2 - x3 10 2x1 + 3x2 + 4x3 =1
2x1 - x2 + x3 4 x1 , x2 , x3 0
x1 , x2 , x3 0
( No feasible solution)
( Unbounded solution)

55
Case Study
 Bus Scheduling

A city is studying the feasibility of introducing a mass-

transit bus system that alleviate the smog problem by
reducing in-city driving. The study seeks the minimum
number of buses that cab handle the transportation
needs. After gathering necessary information, the city
engineer noticed that the minimum number of buses
needed fluctuated with the time of the day and that the
required number of buses could be approximated by
constant values over successive 4-hour intervals. The
summary of the engineer’s finding is shown in the next
slide. To carry out the required daily maintenance,
each bus can operate 8 successive hours a day only.
 Portfolio Selection

Stock fund portfolio selection is an important and

challenging problem. Using linear programming,
construct an efficient stock portfolio for an investor for
the following problem:
Objective: Maximize annual return on investment
Constraints:
Portfolio’s price earnings ratio
Investment
Investment diversification