- Advanced
- About 17 min
- 5 practice questions
Sensitivity: how far can a profit change before the optimum moves?
Find the range of a profit or cost coefficient over which a corner stays optimal, prove optimality with non-negative multipliers, and predict the new optimum when a coefficient leaves its range.
If something looks wrong, please tell us.
Profits and costs are estimates, and they change. A manager wants to know whether the plan must change if the profit on a product rises by ₹10, not just what the best plan is today. Answering this from a single picture, without re-solving, is called sensitivity analysis. It is also the basis of the hardest parameter questions in competitive exams.
The optimality condition at a corner
The practice course used the slope ladder: on a region of ≤ constraints, with p, q > 0, a corner is optimal for maximising px + qy when the slope −p/q lies between the slopes of the two edges meeting there. An equivalent form avoids slopes and works for any signs: write the objective's coefficient pair (p, q) as a combination of the coefficient pairs of the two constraints that meet at the corner. The corner is optimal exactly when both multipliers are non-negative. The multipliers also give a one-line proof of the maximum.
Max of px + qy at a corner where a₁x + b₁y ≤ c₁ and a₂x + b₂y ≤ c₂ meet: if (p, q) = u(a₁, b₁) + v(a₂, b₂) with u, v ≥ 0, then px + qy ≤ uc₁ + vc₂ everywhere in the region, with equality at the corner. For a minimum over ≥ constraints, the same rule gives a lower bound.
Worked example 1: the range of one profit
Maximise Z = 30x + 20y subject to x + y ≤ 8, 3x + y ≤ 18, x, y ≥ 0. The corners are (0, 0), (6, 0), (5, 3) and (0, 8), with Z = 0, 180, 210 and 160, so (5, 3) is optimal. Proof by multipliers: (30, 20) = u(1, 1) + v(3, 1) gives u + 3v = 30 and u + v = 20, so v = 5 and u = 15. Then 30x + 20y ≤ 15 × 8 + 5 × 18 = 210 on the whole region. Now let the profit on x be c instead of 30. Solving (c, 20) = u(1, 1) + v(3, 1) gives v = (c − 20)/2 and u = (60 − c)/2. Both are non-negative exactly when 20 ≤ c ≤ 60. Within this range the plan (5, 3) does not change, although the profit 5c + 60 does.
Worked example 2: what happens at and beyond the ends
At c = 20, Z = 20x + 20y is parallel to the edge x + y = 8, and (0, 8) and (5, 3) tie at 160, so the whole edge is optimal. At c = 60, (5, 3) and (6, 0) tie at 360. For c below 20 the optimum jumps to (0, 8), and for c above 60 to (6, 0). Similarly, keeping 30 for x and letting the profit on y be d, (5, 3) stays optimal for 10 ≤ d ≤ 30.
| Ratio p/q of the profits | Optimal corner | Maximum |
|---|---|---|
| p/q < 1 | (0, 8) | 8q |
| p/q = 1 | every point of the edge from (0, 8) to (5, 3) | 8q |
| 1 < p/q < 3 | (5, 3) | 5p + 3q |
| p/q = 3 | every point of the edge from (5, 3) to (6, 0) | 6p |
| p/q > 3 | (6, 0) | 6p |
Worked example 3: sensitivity of a minimum
Minimise Z = 4x + 5y subject to 2x + y ≥ 6, x + 2y ≥ 6, x, y ≥ 0. Corners: (0, 6), (2, 2), (6, 0), with Z = 30, 18, 24. Multipliers at (2, 2): (4, 5) = u(2, 1) + v(1, 2) gives 2u + v = 4, u + 2v = 5, so u = 1, v = 2, both non-negative: 4x + 5y ≥ 6 + 12 = 18 everywhere, and the minimum is 18. How far can the cost of x rise? With (p, 5): u = (2p − 5)/3 and v = (10 − p)/3, so (2, 2) stays optimal for 5/2 ≤ p ≤ 10. Above 10 the cheapest plan is (0, 6), using no x at all.
Key takeaways: a corner is optimal when the objective's coefficient pair is a non-negative combination of the coefficient pairs of the constraints meeting there. Setting each multiplier ≥ 0 gives the range of a coefficient directly; at the ends of the range two corners tie, and beyond them the optimum jumps to the neighbouring corner.
Check your understanding
-
1. On the region x + y ≤ 8, 3x + y ≤ 18, x, y ≥ 0, the profit is Z = cx + 20y. What is the largest value of c for which (5, 3) is still an optimal corner?
- 20
- 40
- 60
- 54
Show answer
Answer: C. (5, 3) is optimal while −c/20 lies between the edge slopes −1 (x + y = 8) and −3 (3x + y = 18), that is 20 ≤ c ≤ 60. At c = 60, (5, 3) and (6, 0) tie at 360; for c > 60 the optimum moves to (6, 0).
-
2. On the same region with Z = 30x + dy, for which values of d is (5, 3) optimal?
- 10 ≤ d ≤ 30
- 20 ≤ d ≤ 60
- 10 ≤ d ≤ 90
- 0 ≤ d ≤ 30
Show answer
Answer: A. Need −30/d between −1 and −3: 1 ≤ 30/d ≤ 3, so 10 ≤ d ≤ 30. At d = 30 the tie is with (0, 8) (both 240); at d = 10 with (6, 0) (both 180).
-
3. The region x + 2y ≤ 14, 3x + 2y ≤ 18, x, y ≥ 0 has corners (0, 0), (6, 0), (2, 6) and (0, 7). For which p is (2, 6) an optimal corner of Z = px + 3y?
- 1/2 ≤ p ≤ 3/2
- 2 ≤ p ≤ 6
- 1 ≤ p ≤ 3
- 3/2 ≤ p ≤ 9/2
Show answer
Answer: D. The edges at (2, 6) have slopes −1/2 (x + 2y = 14) and −3/2 (3x + 2y = 18). Need −1/2 ≥ −p/3 ≥ −3/2, so 3/2 ≤ p ≤ 9/2. Equivalently, (p, 3) = a(1, 2) + b(3, 2) with a, b ≥ 0: a = (9 − 2p)/4 and b = (2p − 3)/4.
-
4. On the region x + y ≤ 8, 3x + y ≤ 18, x, y ≥ 0, the profit on x rises from ₹30 to ₹70 while y stays at ₹20. What is the new optimum?
- (5, 3), with profit ₹410
- (6, 0), with profit ₹420
- (5, 3), with profit ₹210
- (0, 8), with profit ₹160
Show answer
Answer: B. 70 is above the upper limit 60 found in question 1, so (5, 3) is no longer optimal and the optimum moves to (6, 0): 70 × 6 = 420, against 350 + 60 = 410 at (5, 3).
-
5. On the unbounded region 2x + y ≥ 6, x + 2y ≥ 6, x, y ≥ 0 (corners (0, 6), (2, 2), (6, 0)), the cost is Z = 4x + qy with q > 0. For which q is the minimum at (2, 2)?
- 2 ≤ q ≤ 8
- 1/2 ≤ q ≤ 2
- 4 ≤ q ≤ 8
- 2 ≤ q ≤ 4
Show answer
Answer: A. For a minimum over ≥ constraints, (4, q) must be a non-negative combination of (2, 1) and (1, 2): 4/q must lie between 1/2 and 2, so 2 ≤ q ≤ 8. Check q = 8: (2, 2) gives 24 and (6, 0) gives 24 (a tie); q = 2: (2, 2) and (0, 6) both give 12.
Sign in to keep track of the lessons you finish.