Week 4 - Simplex method, Sensitivity analysis in LP Flashcards
In the simplex method, should we leave only basic or nonbasic variables in the objective function?
Nonbasic variables
Bland’s rule
- prevents simplex from cycling
- choose the variable with the SMALLEST INDEX to ENTER, then the variable with the smallest index to exit
For the starting solution in simplex, what to do if we cannot set the original variables to 0 (nonbasic)?
- Add another variable per equation
eg. s1 for constraint 1, s2 for constraint 2, s3 for constraint 3 etc. - Ignore the original objective function
eg. instead of maximise 3x1 + 2x2, now MINIMISE s1 + s2 + s3 - The original problem has a feasible solution where all s=0
» so if we solve the modified LP to get a feasible solution where all s=0, we solve also the original LP - To solve the modified LP, minimise the sum of the s’s (as detailed in step 2)
*let all “s>=0” when we first write out
In sensitivity analysis, what does the size & sign of a dual value tell us?
yi = 𝛿(optimum value) / 𝛿(RHS of effective constraint)
yi = the RATE OF CHANGE of the OBJECTIVE FUNCTION VALUE
Sensitivity analysis - What happens to the optimal dual solution y* if the RHS of an INEFFECTIVE constraint changes?
y* remains feasible.
As long as the EFFECTIVE CONSTRAINTS of the optimal primal solution don’t change, the optimality conditions {y*} remain satisfied.
Ineffective constraints
2x1 + x2 <= b1
How small can b1 be?
For effective constraints: sub in coordinates into the constraint to calculate the RHS values, eg. -6 <= b5 <= 4
- Sub in x* = (1.2, 3.6) into the objective function to get =6
- Optimal solution x* does not change as long as b1 is within the range 6 <= b1 < +∞
Changes in an objective function coefficient
Ranges for Non-basic variables - eg. for a MAXimisation LP problem, what to do? (see notes for workings)
Ranges for Basic variables: optimal contour rotates counter-clockwise/clockwise around x* until PARALLEL to…
eg. -24 <= c1 <= 8/3
x* remains optimal as long as y* remains feasible for the dual
1. Sub in the y* solution into the dual constraint, then
eg. -∞ <= c2 <= 4
Sensitivity analysis - 100% rule
You can COMBINE several changes of the SAME TYPE,
ie. objective coefficients or RHS coefficients but not both, if the CHANGES to each - as a FRACTION of what is ALLOWED - sum to < 1
Thus, the objective value change = the sum of the changes