Sensitivity Analysis and Post

Optimality

JKJHA@IITKGP
 Introduction
• In most practical applications, the value of LP
parameters is not known exactly and is often estimated
based on past data.
• It is therefore important to study how changes in an LP
parameter (objective function coefficients, right hand
sides, and technological coefficients) affect the optimal
solution.
• Algebraic approach enables to determine the revised
optimal solution from the original solution due to
change in an LP parameter, without resolving the
problem from scratch.
JKJHA@IITKGP
Following changes are considered
(1) Variations in objective function coefficients (cj)
(2) Variations in RHS constants (resource availability) (bi)
(3) Variations in constraint matrix (A)
- Adding new variables or activities
- Variation in the resource requirements (aij) of existing activities
- Adding new constraints

JKJHA@IITKGP
 How feasibility and optimality are affected?
Simplex Tableau at any iteration in matrix form

Coefficients of
Original variables (xT) Slack variables (xsT) RHS

xB B-1A B-1 B-1b

Z cBB-1A - c cBB-1 cBB-1b

The coefficient of xj in Z-row is represented as zj-cj = cBB-1aj – cj, where cBB-1aj is the cost of consumed
resources for the unit production of product j.

Feasibility Condition:
The current basis xB remains feasible if the value of all basic variables is non-negative, i.e.
xB = B-1b ≥ 0

Thus, the current basis may not be feasible due to the following:
- Change in resource availability of a resource (bi)
- Change in B due to change in resource requirement aij corresponding to a basic variable
- Adding one or more new constraints
JKJHA@IITKGP
 Optimality Condition:

The current basis xB remains optimal if the objective function coefficients are non-
negative, i.e.
cBB-1A – c ≥ 0 and cBB-1 ≥ 0

Thus, the current basis may not be optimal due to the following:
- Change in objective function coefficients (cj)
- Change in B due to change in resource requirement aij corresponding to a
basic
- Adding one or more new variables

JKJHA@IITKGP
 Example: Product Mix Problem
• Products: A, B, and C
• Resources: Labor and Material
• Objective: To maximize Profit
• Decision: To find optimal product mix (x1, x2, and x3)
• LP formulation
Maximize Z = 2x1 + 3x2 + x3
1 1 1
Subject to 𝑥 + 𝑥2 + 𝑥3 ≤ 1 Labor
3 1 3 3
1 4 7
𝑥 + 𝑥2 + 𝑥3 ≤ 3 Material
3 1 3 3
𝑥1 , 𝑥2 , 𝑥3 ≥ 0

• Parameters in matrix/vector form

1 1 1
1 3 3 3
𝐜= 2 3 1 𝐛= 𝐀= 1 4 7
3
3 3 3
JKJHA@IITKGP
Optimal Simplex Table
• Let x4 and x5 be slack variables corresponding to labor and
material constraints, respectively.

Basis x1 x2 x3 x4 x5 RHS
x1 1 0 -1 4 -1 1
x2 0 1 2 -1 1 2
zj - cj 0 0 3 5 1 8

• Optimal Solution:
x1 = 1, x2 = 2, x3 = 0, Z = 8

JKJHA@IITKGP
 Variations in objective function coefficients (cj)
(i) Changing objective function coefficient (cj) of a Non-Basic Variable
Product C is not produced in the optimal product mix because of low value of c3.
However, the optimal product mix may change if the value of c3 increases.

• Finding the range of c3 for which the current basis (solution) remains optimal:

If the value of c3 changes, only the value of z3 – c3 changes in the optimal table, then

−1
Calculate z3 – c3 = cBB-1a3-c3 = 2 3 − 𝑐3 = 4 − 𝑐3
2

Thus, the current solution remains optimal as long as z3 – c3 ≥ 0, i.e. 4 – c3 ≥ 0

=> c3 ≤ 4

• Suppose if c3 is increased from 1 to 6, then find the optimal solution.

=> z3 – c3 = 4 − 6 = -2 < 0, so the current solution is not optimal

JKJHA@IITKGP
 Finding the new optimal solution
• Update z3 – c3 entry as -2 in the current optimal simplex table
Basis x1 x2 x3 x4 x5 RHS
x1 1 0 -1 4 -1 1
x2 0 1 2 -1 1 2
zj - cj 0 0 -2 5 1 8

• Perform a primal simplex iteration to find the new optimal solution

Basis x1 x2 x3 x4 x5 RHS Ratio
x1 1 0 -1 4 -1 1 -
x2 0 1 2 -1 1 2 2/2 =1
zj - cj 0 0 -2 5 1 8
x1 1 1/2 0 7/2 -1/2 2
x3 0 1/2 1 -1/2 1/2 1
zj - cj 0 1 0 4 2 10
• New optimal solution: x1 = 2, x2 = 0, x3 = 1, Z = 10
JKJHA@IITKGP
(ii) Changing objective function coefficient (cj) of a Basic Variable
• To study the effect of change in c1 and c2
 Effect of change in the objective function coefficient (c1) of basic variable x1
- If c1 decreases below a certain limit, it may not be profitable to include product A in the
product mix.
- Even if c1 increases, the current product mix may change.
• Finding the range of c1 for the current basis (solution) to remain optimal:
Change in c1 changes the profit vector of the basic variables (cB)
Calculate zj – cj for all j = 1, 2, …, 5

1
z1 – c1 = cBB-1a1 − c1 = 𝑐1 3 − 𝑐1 = 0 Updated profit coefficient
0 of basic variables
0
z2 – c2 = cBB-1a2 − c2 = 𝑐1 3 −3=0 remains at zero
1

However, the updated profit coefficient of Non-basic variables will change

−1
z3 – c3 = cBB-1a3 − c3 = 𝑐1 3 − 1 = −𝑐1 + 5
2
4 −1
(z4 – c4 z5 – c5) = 𝐜𝐁 𝐁 −𝟏 = 𝑐1 3 = 4𝑐1 − 3 −𝑐1 + 3
−1 1
JKJHA@IITKGP
The current basis (solution) will remain optimal as long as
zj – cj ≥ 0 for all j = 1, 2, …, 5
Thus,
The range of c1 for
−𝑐1 + 5 ≥ 0 ⇒ 𝑐1 ≤ 5 which the current basis
3 (solution) will remain
4𝑐1 − 3 ≥ 0 ⇒ 𝑐1 ≥ optimal:
4 3
≤ 𝑐1 ≤ 3
−𝑐1 + 3 ≥ 0 ⇒ 𝑐1 ≤ 3 4

• If the value of c1 goes beyond these limits, one of the non-basic

variable will enter the basis, and apply a primal simplex iteration
to find the new optimal solution.

• Similar analysis for c2 can be done to find the range of c2 for

which the current basis (solution) will remain optimal, and the
range can be obtained as 2 ≤ 𝑐2 ≤ 8
JKJHA@IITKGP
(iii) Changing objective function coefficient (cj) of both Basic and Non-
basic Variables
• Suppose the profits on all three products are changed and let
objective function becomes Z = x1 + 4x2 + 2x3
• As a result, z3 – c3, z4 – c4, z4 – c4 will be affected, and z1 – c1, z2 – c2
remain unchanged (zero) with the current basis.
−1
• z3 – c3 = cBB-1a3 − c3 = 1 4 −2=5>0
2
4
z4 – c4 = cBB-1a4 − c4 = 1 4 −0=0
−1
−1
z5 – c5 = cBB-1a5 − c5 = 1 4 −0=3>0
1

• Thus, the optimal solution does not change.

• However, the problem has multiple optimal solutions, as z4 – c4 = 0

JKJHA@IITKGP
 Variations in RHS constants (bi)
• The current basis may not be feasible due to change in bi.
• Finding the range of b1 for the current basis to remain feasible
(optimal)
′ −𝟏 ′ 4 −1 𝑏1 4𝑏1 − 3
𝒙𝐁 = 𝐁 𝐛 = =
−1 1 3 −𝑏1 + 3

4𝑏1 − 3
For the current basis to remain feasible, 𝒙′𝐁 ≥𝟎⇒ ≥0
−𝑏1 + 3

3
Thus, the range of b1 for current basis to remain feasible: ≤ 𝑏1 ≤ 3
4

• Similarly, the range of b2 for current basis to remain feasible can be

found as 1 ≤ 𝑏2 ≤ 4
JKJHA@IITKGP
Suppose if b1 is increased from 1 to 4, then find the optimal
solution.
• For b1 = 4, in the current optimal simplex table, update RHS column as
4𝑏1 − 3 4𝑏1 − 3
= 13 and 𝑍 = 𝐂𝐁 𝐁 −𝟏 𝐛 = 2 3 = 23.
−𝑏1 + 3 −1 −𝑏1 + 3
• Apply an iteration of dual simplex method to find the new optimal solution
Basis x1 x2 x3 x4 x5 RHS
x1 1 0 -1 4 -1 13
x2 0 1 2 -1 1 -1
zj - cj 0 0 3 5 1 23
Ratio -5
x1 1 4 7 0 3 9
x4 0 -1 -2 1 -1 1
zj - cj 0 5 13 0 6 18

• New optimal solution: x1 = 9, x2 = 0, x3 = 0, Z = 18

JKJHA@IITKGP
 Variations in constraint matrix (A)
(i) Adding a new variable or an activity
• Suppose the company introduces a new product D which requires 1 unit of labor
and 1 unit of material and can be sold at a unit profit of 3.
• Inclusion of a new product is equivalent to adding a new variable x6, and a
1
column 𝐚6 = in the initial simplex table.
1
1
• Calculate z6 – c6 = cBB-1a6 − c6 = 5 1 − 3 = 3 > 0, therefore the current
1
solution is optimal and producing product D will not improve the present
maximum profit.
1
• If c6 = 8, then z6 – c6 = 5 1 − 8 = −2 < 0. The current solution is not
1
optimal and find the new optimal solution by applying a primal simplex iteration.
• Now, the column B-1a6 in the modified simplex table corresponding to the
4 −1 1 3
variable x6 can be obtained as B-1a6 = = .
−1 1 1 0

JKJHA@IITKGP
Modified Simplex table
Basis x1 x2 x3 x4 x5 x6 RHS Ratio
x1 1 0 -1 4 -1 3 1 1/3
x2 0 1 2 -1 1 0 2 -
zj - cj 0 0 3 5 1 -2 8
x6 1/3 0 -1/3 4/3 -1/3 1 1/3
x2 0 1 2 -1 1 0 2
zj - cj 2/3 0 7/3 23/3 1/3 0 26/3

New optimal solution: x1 = 0, x2 = 2, x3 = 0, x6 = 1/3, Z = 26/3

JKJHA@IITKGP
(ii) Variation in the resource requirements (aij) of
existing activities
• If the variation in aij corresponds to a non-basic variable xj (e.g., x3),
then it is required to recalculate B-1aj and cBB-1aj − cj. If cBB-1aj
− cj ≥ 0, the current solution remains optimal, otherwise a primal
simplex iteration is performed similar to the previous case to find
the new optimal solution.

• If the variation in aij corresponds to a basic variable xj (e.g., x1 or

x2), then basis matrix B (therefore, B-1 ) itself is affected. As a result,
all the quantities in the optimal simplex table is affected. In such
case, if the current basis is
- primal feasible but not optimal => Apply primal simplex
- primal infeasible but optimal (i.e. dual feasible) => Apply dual simplex
- neither primal feasible nor dual feasible=> Apply simplex method from the
start with new value of aij.

JKJHA@IITKGP
 (iii) Adding new constraints
• Suppose a carbon emission constraint due to carbon cap is included
in the original problem as x1 + 2x2 + x3 ≤ 10
• If the new constraint is satisfied with the present optimal solution,
then the present solution remains optimal.
• As 1 + 2 × 2 + 0 = 5 ≤ 10, so the present solution does not change.
• Suppose available carbon cap is only 4, then the new constraint
becomes x1 + 2x2 + x3 ≤ 4, and the present optimal solution violates
this constraint.

• Finding the new optimal solution

To find the new optimal solution, add the augmented form of
this new constraint as the third row using x6 as the slack variable.

JKJHA@IITKGP
Modified simplex table
Basis x1 x2 x3 x4 x5 x6 RHS
x1 1 0 -1 4 -1 0 1
x2 0 1 2 -1 1 0 2
x6 1 2 1 0 0 1 4
zj - cj 0 0 3 5 1 0 8
• Performing elementary row operations for the third row to bring the system of equations
in proper form (basic variables x1 and x2 are not forming identity matrix)
R3 = R3 – R1 – 2R2
Basis x1 x2 x3 x4 x5 x6 RHS
x1 1 0 -1 4 -1 0 1 Apply Dual Simplex
x2 0 1 2 -1 1 0 2
x6 0 0 -2 -2 -1 1 -1
zj - cj 0 0 3 5 1 0 8
Ratio -3/2 -5/2 -1
New optimal
x1 1 0 1 6 0 -1 2
solution: x1 = 2,
x2 0 1 0 -3 0 1 1 x2 = 1, x3 = 0,
x5 0 0 2 2 1 -1 1 Z=7
zj - cj 0 0 1 3 0 1 7
JKJHA@IITKGP