In linear programming, we only consider two types of sensitivity analyses using Excel
solver in this module.
A. There is a change in the objective coefficient.
You should decide the sensitivity range from the sensitivity report first. If the new
objective coefficient is in the sensitivity range, the optimal solution will not change,
but you need to compute the new optimal value. If the new objective coefficient is
beyond the sensitivity range, we need to re-solve the optimization problem using
Excel solver.
The change of the objective coefficient will not affect the feasible region of the
problem since the feasible region is the intersection of all the constraints of the
problem. This also means that the binding constraints will remain the same, if the
new objective coefficient is still in the sensitivity range.
B. There is a change in the right-hand-side constraint coefficient of a constraint.
Two cases will be considered:
Case 1. If the change is made on the non-binding constraint.
- The shadow price is zero.
- If the new right-hand-side constraint coefficient is within the sensitivity range,
thus, there is no change in the optimal value and optimal solution. Moreover, the
set of binding constraints remain unchanged.
Case 2. If the change is made on the binding constraint.
- The shadow price is non-zero, but it can be positive or negative.
- If the new right-hand-side constraint coefficient is within the sensitivity range,
you can use the shadow price to predict the optimal value.
- For a resource constraint (<=), an increase in the right-hand-side constraint
coefficient (a larger feasible region) will lead to an improvement of the
optimization. That is, the objective value will increase for a maximization
problem, but the objective value will decrease for a minimization problem.
- For a requirement constraint (>=), an increase in the right-hand-side constraint
coefficient will lead to a deterioration of the optimization. That is, the objective
value will decrease for a maximization problem, but the objective will increase for
a minimization problem.
- To predict the new solution, you need to re-solve the binding constraints to get
the new solution.
