ENGINEERING OPTIMIZATION

ME F320/MF F320

BITS Pilani
Pilani Campus
 3.4 Artificial
Starting Solution

2
Artificial Starting Solution
• LPs in which all constraints are with
nonnegative right-hand sides offer a convenient
“all-slack” starting basic feasible solution.
• “Ill-behaved” LPs: models with (=) and/or (≥)
constraints.
– use artificial variables as slacks in 1st iteration 
artificial variables disposed of later.
– M Method
– 2-phase Method
3
 3.4.1 Big
M-Method

4
M-Method
• Example:
Minimize:
z = 4x1 + x2
subject to
3x1 + x2 = 3
4x1 + 3x2 ≥ 6
x1 + 2x2 ≤ 4
x1 , x2 ≥ 0

5
M-Method
Step 1: Convert to standard form
Minimize:
z = 4x1 + x2
subject to
3x1 + x2 = 3
4x1 + 3x2 − S1 = 6
x1 + 2x2 + s2 = 4
x1 , x2 , S1 , s2 ≥ 0

No all-slack basic solution exits.

6
M-Method
Step 2: Add artificial variables R1 and R2
Minimize: −M in
z = 4x1 + x2 + M(R1 + R 2 ) maximization,
and +M for
subject to minimization
3x1 + x2 + R1 = 3
4x1 + 3x2 − S1 + R 2 = 6
x1 + 2x2 + s2 = 4
x1 , x2 , S1 , s2 , R1 , R 2 ≥ 0
• Artificial variables forced to zero by the time optimum is
reached.
• M is a large positive value (M → ∞).
7
M-Method
Step 2: Add artificial variables R1 and R2
Minimize:
z = 4x1 + x2 + M(R1 + R 2 )
subject to
3x1 + x2 + R1 = 3
4x1 + 3x2 − S1 + R 2 = 6
x1 + 2x2 + s2 = 4
x1 , x2 , S1 , s2 , R1 , R 2 ≥ 0
• n = 6, m = 3
• Starting basic solution (R1 = 3, R 2 = 6, s2 = 4)

8
M-Method
Step 3: Choose M
• M should be sufficiently large to force the artificial
variable to zero.
• If artificial variables are non-zero in the optimum
solution  no feasible solution exits for the original
LPP.
• M shouldn’t be too large  causes round-off errors in
computations.
• Choose M = 100 for the current problem.

9
Simplex Tableau
• Starting Simplex Tableau (starts at origin)
z − 4x1 − x2 − 100 R1 + R 2 = 0
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 −4 −1 0 −100 −100 0 0
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

Is this the correct starting table?

10
Simplex Tableau
• Starting Simplex Tableau (starts at origin)
z − 4x1 − x2 − 100 R1 + R 2 = 0
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 −4 −1 0 −100 −100 0 0
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

Z-row is inconsistent!
11
Gauss-Jordan row operations
• Substitute R1 and R2 in z (in terms of non-basic
variables)
– New z row = Current z row + M R1 row + M(R2 row)

Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 696 399 −100 0 0 0 900
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

Can you spot the identity matrix?

12
Gauss-Jordan row operations
• Substitute R1 and R2 in z (in terms of non-basic
variables)
– New z row = Current z row + M R1 row + M(R2 row)

Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 696 399 −100 0 0 0 900
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

Can you spot the identity matrix?

13
New Simplex Tableau
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 696 399 −100 0 0 0 900
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

1. Optimality condition for minimization: x1 (with

most positive z-row coefficient) enters.
2. Feasibility condition: Perform MRT  R1 exits.

14
New Simplex Tableau
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 696 399 −100 0 0 0 900
x1 row 0 3/3 1/3 0 1/3 0 0 3/3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

1. Optimality condition: x1 (with most positive z-row

coefficient) enters.
2. Feasibility condition: Perform MRT, R1 exits.
3. Gauss Jordan row-operation: On x1 row
15
New Simplex Tableau
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 696 399 −100 0 0 0 900
x1 row 0 1 1/3 0 1/3 0 0 1
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

1. Optimality condition: x1 (with most positive z-row

coefficient) enters.
2. Feasibility condition: Perform MRT, R1 exits.
3. Gauss Jordan row-operations: On other rows…
16
New Simplex Tableau
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 0 167 −100 −232 0 0 204
x1 row 0 1 1/3 0 1/3 0 0 1
R2 row 0 0 5/3 −1 −4/3 1 0 2
s2 row 0 0 5/3 0 −1/3 0 1 3

• R1 and R2 leave the basic solution (become zero) in the

1st and 2nd iterations.
• Use of penalty M will not force an artificial variable to
zero in final simplex iteration if LPP doesn’t have a
feasible solution.
17
New Simplex Tableau
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 0 167 −100 −232 0 0 204
x1 row 0 1 1/3 0 1/3 0 0 1
R2 row 0 0 5/3 −1 −4/3 1 0 2
s2 row 0 0 5/3 0 −1/3 0 1 3

• Complete the table…

• Optimum @ ???

18
New Simplex Tableau
Basic z x1 x2 S1 R1 R2 s2 Solution
z row 1 0 167 −100 −232 0 0 204
x1 row 0 1 1/3 0 1/3 0 0 1
R2 row 0 0 5/3 −1 −4/3 1 0 2
s2 row 0 0 5/3 0 −1/3 0 1 3

• Complete the table…

Optimum @ (x1 = 2/5, x2 = 9/5, z = 17/5)

19
3.4.2 Two-Phase
Method

20
Two-Phase Method
• The two-phase method eliminates use of
constant M.
• Solves LPP in two phases:
– Phase I: find a starting basic feasible solution.
– Phase II: solve original problem.

21
Two-Phase Method: Phase I
• Phase I:
i. Put problem in equation form, and add artificial
variables to constraints (exactly as in the M-
method) to secure a starting basic solution.
ii. Find basic solution of resulting equations that
minimizes sum of the artificial variables (r =
R1 + R2 +…).
iii. r > 0  no feasible solution, otherwise (r = 0),
proceed to Phase II.

22
Two-Phase Method: Phase I
• Example:
Minimize:
z = 4x1 + x2
subject to
3x1 + x2 = 3
4x1 + 3x2 ≥ 6
x1 + 2x2 ≤ 4
x1 , x2 ≥ 0

23
Two-Phase Method: Phase I
Step 1: Convert to standard form
Minimize:
z = 4x1 + x2
subject to
3x1 + x2 = 3
4x1 + 3x2 − S1 = 6
x1 + 2x2 + s2 = 4
x1 , x2 , S1 , s2 ≥ 0

No all-slack basic solution exits.

24
Two-Phase Method: Phase I
Step 2: Add artificial variables R1 and R2
Minimize:
z = 4x1 + x2
subject to
3x1 + x2 + R1 = 3
4x1 + 3x2 − S1 + R 2 = 6
x1 + 2x2 + s2 = 4
x1 , x2 , S1 , s2 , R1 , R 2 ≥ 0

25
Two-Phase Method: Phase I
Step 3: Solve the Phase I problem
Minimize:
r = R1 + R 2
subject to
3x1 + x2 + R1 = 3
4x1 + 3x2 − S1 + R 2 = 6
x1 + 2x2 + s2 = 4
x1 , x2 , S1 , s2 , R1 , R 2 ≥ 0
• Solution of the Phase I problem (r = 0; R1 = 0 and R2 = 0) will
generate a feasible basic solution for the original LPP.

26
Phase I: Simplex Tableau
• Starting Simplex Tableau (starts at origin)
r = R1 + R 2
Basic r x1 x2 S1 R1 R2 s2 Solution
r row 1 0 0 0 −1 −1 0 0
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

Is this the correct starting table?

27
Phase I: Simplex Tableau
• Starting Simplex Tableau (starts at origin)
r = R1 + R 2
Basic r x1 x2 S1 R1 R2 s2 Solution
r row 1 0 0 0 −1 −1 0 0
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

r-row is inconsistent!
28
Phase I: Row Operations
• Substitute R1 and R2 in r (r in terms of non-basic
variables)
– New r row = Current r row + R1 row + (R2 row)

Basic r x1 x2 S1 R1 R2 s2 Solution
r row 1 7 4 −1 0 0 0 9
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

Can you spot the identity matrix?

29
Phase I: Row Operations
• Substitute R1 and R2 in r (r in terms of non-basic
variables)
– New r row = Current r row + R1 row + (R2 row)

Basic r x1 x2 S1 R1 R2 s2 Solution
r row 1 7 4 −1 0 0 0 9
R1 row 0 3 1 0 1 0 0 3
R2 row 0 4 3 −1 0 1 0 6
s2 row 0 1 2 0 0 0 1 4

30
Phase I: Optimum
• Minimum value is r = 0
– Phase I produced a feasible basic solution.
– (x1 = 3/5, x2 = 6/5, s2 = 1)
– Eliminate artificial variables.

Basic r x1 x2 S1 R1 R2 s2 Solution
r row 1 0 0 0 −1 −1 0 0
x1 row 0 1 0 1/5 3/5 −1/5 0 3/5
x2 row 0 0 1 −3/5 −4/5 3/5 0 6/5
s2 row 0 0 0 1 1 −1 1 1

31
Phase II
• Delete artificial columns, original problem becomes:
Minimize:
z = 4x1 + x2
subject to
1 3
x1 + S1 =
5 5
3 6
x2 − S1 =
5 5
S1 + s2 = 1
x1 , x2 , S1 , s2 ≥ 0
32
Phase II: Simplex Tableau
• Starting Simplex Tableau
z = 4x1 + x2

Basic z x1 x2 S1 s2 Solution
z row 1 4 1 0 0 0
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

Is this the correct starting tableau?

33
Phase II: Simplex Tableau
• Starting Simplex Tableau
z = 4x1 + x2

Basic z x1 x2 S1 s2 Solution
z row 1 −4 −1 0 0 0
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

z-row signs are wrong!

34
Phase II: Simplex Tableau
• Starting Simplex Tableau
z = 4x1 + x2

Basic z x1 x2 S1 s2 Solution
z row 1 −4 −1 0 0 0
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

Is this the correct starting tableau?

35
Phase II: Simplex Tableau
• Starting Simplex Tableau
z = 4x1 + x2

Basic z x1 x2 S1 s2 Solution
z row 1 −4 −1 0 0 0
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

z-row is inconsistent!

36
Phase II: Row Operations
• Basic variables (x1 and x2) must be substituted out:
New z row = Current z row + 4 × x1 row + (1 × x2 row)

Basic z x1 x2 S1 s2 Solution
z row 1 −4 −1 0 0 0
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

37
Phase II: Starting Simplex Tableau
• Starting Simplex Tableau
1 18
Minimize: z = − S1 +
5 5
Basic z x1 x2 S1 s2 Solution
z row 1 0 0 1/5 0 18/5
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

Is this the correct starting tableau?

38
Phase II: Starting Simplex Tableau
• Starting Simplex Tableau
1 18
Minimize: z = − S1 +
5 5
Basic z x1 x2 S1 s2 Solution
z row 1 0 0 1/5 0 18/5
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

Is this the correct starting tableau? Yes!!!

39
Phase II: Starting Simplex Tableau
• Starting Simplex Tableau
1 18
Minimize: z = − S1 +
5 5
Basic z x1 x2 S1 s2 Solution
z row 1 0 0 1/5 0 18/5
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

Compute the optimum value.

40
Phase II: Final Simplex Tableau
• Entering and leaving variables
1 18
Minimize: z = − S1 +
5 5
Basic z x1 x2 S1 s2 Solution
z row 1 0 0 1/5 0 18/5
x1 row 0 1 0 1/5 0 3/5
x2 row 0 0 1 −3/5 0 6/5
s2 row 0 0 0 1 1 1

41
Phase II: Final Simplex Tableau
• Final Simplex Tableau:
1 17
Minimize: z = s2 +
5 5

Basic z x1 x2 S1 s2 Solution
z row 1 0 0 0 −1/5 17/5
x1 row 0 1 0 0 −1/5 2/5
x2 row 0 0 1 0 3/5 9/5
S1 row 0 0 0 1 1 1

42
Thank you!

43