2 Sensitivity Analysis With Solutions
2 Sensitivity Analysis With Solutions
Sensitivity Analysis
• We can calculate an optimal solution x* of the primal and p* of the dual via the simplex method
• Usually when an optimal solution of a linear problem is found, the work is not done.
• Problem in practice: the problem parameters A,b,c are often just estimates based on predictions
• We have seen examples for this in the first chapter.
Sensitivity Analysis: what is the effect on the optimal solution and its costs if
parameters of the problem change
1. Detect sensitive parameters, i.e. parameters whose values cannot be changed without changing the
optimal basis.
2. Determine ranges for the parameters over which the optimal basis will remain unchanged.
3. Determine ranges for the parameters over which the solution remains feasible.
Example: Production Planning
Original Problem Revised Problem
Basic solution:
Optimality condition:
Preparation
Consider the primal problem again:
Let x* be an optimal basic solution of the problem and B the corresponding optimal basis.
Goal: determine the range of values of under which the current basis
remains optimal.
Note: the optimality condition is not affected by b
Hence we only have to check the feasibility condition
For in the latter range the original optimal basis remains feasible and optimal.
The solution, and therefore the optimal value depends on the new b:
If is outside of the calculated range, then the original optimal basis B is not feasible anymore (and
therefore not optimal)
We have
1. In which range can we change the parameter b1=10 such that the optimal basis does not change?
2. How much do the costs of the solution change in this range?
Solution 1. Assignment
• Optimal basis is B={1,2}
• We have:
• We get:
• Conditions on :
Note that the feasibility condition does not depend on the costs c. Optimal solution remains the same as long as the
basis does not change!
Case 1: j is a non-basic index, then appears only in the optimality condition for variable j:
Note, if we consider
or equivalent maximization problems,
then optimality is received
when the reduced costs are
non-positive. Hence in this
case the >= must be <=.
Changes in Cost Vector c
Case 2: j is a basic index (let‘s say j is the l-th basic index).
Then for all the optimality condition must hold (for i=j the reduced costs are 0 anyway):
Note, if we consider
where are the reduced costs of variable i and is the l-th entry of maximization problems,
then optimality is received
That means the range is given by the inequality system: when the reduced costs are
non-positive. Hence in this
case the <= must be >=.
Note that the reduced costs and the value can be directly read from the optimal simplex tableau.
In the determined range the basis B remains optimal and the solution x* remains optimal.
The parameter corresponds to the first basic variable (l=1). Hence we have to apply Case 2.
The optimal simplex tableau for the problem with original parameters is:
An optimal solution of the problem is x*=(2,2,0,0) and corresponding optimal simplex tableau is
Calculate the range of each cost parameter for which x* remains an optimal solution.
Solution
Changes in Non-Basic Column of A
We consider now the case where a parameter of A is changed which corresponds to a non-basic column Aj:
Again remember: if we
have a maximization
Since j is not in B the feasibility condition does not change. problem in the optimality
Also the optimal solution does not change as long as the optimal basis remains the same. condition the inequality
sign has to be reversed.
In the optimality condition only the reduced costs of the j-th column are affected:
We obtain, that if the optimal basis does not change and the optimal solution remains the same.
Changes in Basic Column of A
An optimal solution of the problem is x*=(2,2,0,0) and corresponding optimal simplex tableau is
Calculate the range of all constraint coefficients of variables x3 and x4 for which the
optimal solution does not change.
Solution
We can calculate the dual solution first: