6634 CH04 UG

8/23/02

1:53 PM

Page 20

C H A P T E R

Linear Programming Sensitivity Analysis

SOLUTIONS TO DISCUSSION QUESTIONS AND PROBLEMS

4-1. In most real world situations that are modeled using LP, conditions are dynamic and changing. Hence, input data such as resource availabilities, prices, and costs used in the LP model are estimated, rather than known with certainty. In such environments, sensitivity analysis can be used to identify the ranges of values of these input data for which the current LP solution remains optimal. This is done without solving the problem again each time we need to examine a change in an input datas value. When all model values are deterministic, that is, known with certainty, sensitivity analysis may not be needed from the perspective of evaluating data accuracy. This may be the case in a portfolio selection model in which we select from among a series of bonds whose returns and cash-in values are set for long periods of time. 4-2. Sensitivity analysis is important in all decision modeling techniques. For example, it is important in breakeven analysis to test the models sensitivity to selling price, xed cost, and variable cost. Likewise, it is important in inventory models in which we tests the results sensitivity to changes in demand, lead time, costs, and so on. 4-3. A change in a resources availability (right-hand-side) changes the size of a feasible region. An increase means more units of that resource are available, causing the feasible region to increase in size. A decrease means fewer units are available. Obviously, if more units of a binding resource are available, it may be possible for the optimal objective value to improve. In contrast, if more units of a non-binding resource are available, the additional units would just contribute to more slack and there would be no improvement in the optimal objective value. 4-4. A change in an objective function coefcient changes the slope of the objective function, with respect to that variable. The change in the slope may be sufcient to make a different corner point become the new optimal solution to the LP model. 4-5. Simultaneous changes in input data values are extremely logical in many contexts. For example, it may be possible to trade one type of resource for another, causing a decrease in the availability of one resource and an increase in the others availability. Likewise, market conditions may cause us to simultaneously reduce the price of all our products, not just in a single product. 4-6. We use the 100% rule to verify if the shadow prices in the current sensitivity report are still valid to analyze the impact of a

proposed simultaneous change in input data values. To analyze a simultaneous change, we compute the ratio of each proposed change in a parameters value to the maximum allowable change in its value, as given in the sensitivity report. The sum of these ratios must not exceed 1 (or 100%) in order for the information given in the current sensitivity report to be valid. If the sum of the ratios does exceed 1, the current information may still be valid; we just cannot guarantee its validity. However, if the ratio does not exceed 1, the information is denitely valid. 4-7. The pricing-out procedure allows a rm to analyze if a new product will be protable enough to be included in the optimal solution, without having to reformulate and solve the model after including this additional variable. Alternatively, we can use the pricing-out procedure to determine the minimum prot contribution required from a new product in order for it to be included in the optimal production mix. 4-8. In most cases, when the Allowable Increase or Allowable Decrease column for the objective function coefcient of a variable has a value of zero in the Adjustable Cells table, this indicates the presence of alternate optimal solutions. 4-9. The shadow price of a resource indicates the marginal value of each additional unit of that resource to the rm. In an LP model with several resources, this information helps the rm to prioritize its resources in terms of their marginal value. 4-10.
100

The Current Solution from Problem 2-14 is:

75 Isoprofit Line for X +Y = $66.67 a ( X = 331/3, Y = 33 1/3 ) b 25 optimal solution

Y 50

0 0 25

c 50 X 75 100

20

6634 CH04 UG

8/23/02

1:53 PM

Page 21

CHAPTER 4

LINEAR PROGRAMMING SENSITIVITY ANALYSIS

21

a.
150

If the prot equation is $3X tion switches to point c.

$1Y, the optimal solu-

If the prot changes to $4.50 per unit of X, the optimal solution shifts to point c. If the objective function becomes $3X remains optimal.
150

$3Y, the corner point b

Isoprofit Line for 3X + 1Y = $150 100 a

Profit Line for 3X + 3Y

100

Profit Line for 4.50X + 2Y

50

X + 2Y 100 2X + Y 100 Optimal Solution is Now Here

b 50 Profit Line for 3X + 2Y

0 0 50 X 100 150 0 0 50 X c 100 150

b.

If Xs prot coefcient was overestimated, but should only have been $1.25, it is easy to see graphically that the solution at point b remains optimal.

c.
100

4-12. The optimal solution of $26 prot lies at the point X Y 3, in Problem 2-16.
8

2,

75

2X +Y 100 6

6X + 4Y 24

Y 50

( X = 426/7, Y = 142/7; Profit = $571/7)

Y 4 Profit = 4X + 6Y = $26

25

X + 4Y 100

b 0

Optimal Solution Remains at Point b

X + 2Y 8

25

50 X

75

100

The optimal solution is at point b, but prot has decreased from $662 to $571, and the solution has changed considerably. 3 7 4-11. Using the isoprot line or corner point method, we saw in Problem 2-15 that point b (where X 37.5 and Y 75) is optimal if the prot $3X $2Y.

4 X

a. If the rst constraint is altered to 1X 3Y 8, the feasible region and optimal solution shift considerably, as shown in the next gure.

6634 CH04 UG

8/23/02

1:53 PM

Page 22

22

CHAPTER 4

LINEAR PROGRAMMING SENSITIVITY ANALYSIS

6X + 4Y 24 6 Profit = 4X + 6Y = $21.71 Y 4 Optimal solution at X = 26/7, Y = 15/7 2 1X + 3Y 8

long the number of contacts is between 0 and 6,620.69. Obviously, if the objective function coefcient is 0, it is not worthwhile to use any TV spots. The reasons why TV spots are part of the optimal solution even if the exposure is as low as 0.001 are the limits on the number of newspaper ads and the total radio budget. See if you can recognize this fact by altering the LP model in le 3-1.XLS. 4-14. We use the sensitivity report given in Program 4.7 to answer the following questions. a. Persons 30 or younger who live in a border state currently cost $7.50 each. This cost must decrease to $6.90 or less ( $7.50 $0.60) before it is worthwhile to include these individuals in the survey. b. Based on the shadow price for the total households constraint, each additional person who needs to be included in the survey will cause the cost to increase by $5.98. Hence, if the sample size is increased to 3,000, the total cost will increase to $19,352 ( $15,166 700 x $5.98). c. Based on the reduced cost, each person 3150 not living in a border state who is included in the survey will increase the total cost of the survey by $0.45. d. First, we check the 100% rule. (100/1000) (50/700) 1. Hence, the shadow prices are valid. The revised total cost is $15,166 100 x $0.92 50 x $0.82 $15,115. Note that the reduction in the 30 and younger requirement will cause the total cost to go down, while the increase in the 3150 requirement will cause the total cost to go up. 4-15. We use the sensitivity report given in Program 4.8 to answer the following questions. a. If the daily allowance of protein is reduced to 2.9 units, the total cost will decrease by 0.1 x $0.038 $0.0038. b. Revised cost of grain A $0.33/1.05 $0.314. Revised cost of grain B $0.47/0.9 $0.522. First, we check the 100% rule. (0.016/Innity) (0.052/Innity) 1. Therefore, the current optimal solution remains optimal. The new total cost is $0.0529. c. Check the 100% rule. (0.1/0.25) (0.2/0.35) 0.97 1. The revised total cost changes by (0.2 $0) (0.1 $0.038), or a decrease of $0.0038. 4-16. See le P4-16.XLS for the Excel solution and Solver sensitivity report. a. The decrease of $0.01 per pound is beyond the allowable decrease limit of $0.005 per pound. Therefore, the optimal solution will change. b. The constraints prescribing the minimum daily requirements for ingredient A and ingredient D are binding. For each additional unit of ingredient A required in the mix, the cost will increase by $0.003. For each additional unit of ingredient D required in the mix, the cost will increase by $0.083. c. A 20% decrease in the cost of mineral implies that the cost is now 0.8 x $0.17 $0.136, a decrease of $0.034. Since this is within the allowable decrease limit, the current solution remains optimal. The revised total cost is $0.57. d. The price of oats can uctuate between $0.085 and $0.093 per pound for the current solution to remain optimal. 4-17. See le P4-17.XLS for the Excel solution and Solver sensitivity report.

4 X

b.
12

10

6X + 4Y 36

Y 6
1 ( X = 5, Y = 1 /2 ; Profit = $29 )

1X + 2Y 8

0 0 2 4 6 X 8 10 12

Using the corner point method, we determine that the optimal solution mix under the new constraint yields a $29 prot, or an increase of $3 over the $26 prot calculated. 4-13. We use the sensitivity report given in Program 4.6 to answer the following questions. a. Each additional $ in radio advertising (up to $1,575) will increase the audience by 2.03. Hence, if management approves an increase of $200, audience coverage will increase by 406 to 67,646. b. No. Since we are already placing 6.21 radio spots, this contractual agreement is not a binding constraint. c. The audience reached for each afternoon radio spot would have to increase to at least 3,144.83 ( 2,800 344.83) in order for these spots to become attractive. Hence, the proposed strategy will not change the current optimal solution. d. TV spots currently reach 5,000 contacts. Interestingly, the optimal number of TV spots to use does not change as

6634 CH04 UG

8/23/02

1:53 PM

Page 23

CHAPTER 4

LINEAR PROGRAMMING SENSITIVITY ANALYSIS

23

a. Each additional gram of carbohydrates allowed in the meal will reduce the meal cost by $0.007. Each additional mg of iron required in the diet will cause the meal cost to increase by $0.074. b. Each pound of milk used in the diet will cause the meal cost to increase by $0.148 (reduced cost). c. Beans currently cost $0.58 per pound. Looking at the reduced cost, the price of beans would have to decrease by at least $0.261 (to $0.319) before Kathy can consider including it in the meal. d. None of the allowable increase and allowable decrease values for the objective function coefcients is zero. Further, all items that are currently not in the meal (milk, sh, and beans) have non-zero reduced costs. The current solution is, therefore, a unique optimal solution. 4-18. See le P4-18.XLS for the Excel solution and Solver sensitivity report. a. Four products (circuit boards, oppy drives, hard drives, and memory boards) are not included in the optimal production plan. The reduced costs indicate the minimum amounts by which the prot contributions of these products must increase before they would be included in the production mix. For example, the prot contribution of a circuit board should increase by at least $138.64 (from the current $135.50 to at least $274.14) before it becomes a viable product. b. The current production plan required only 62.07 hours of the available 100 hours on test device 3. Therefore, if the sister concern takes 35 hours of time on this test device, Quitmeyers optimal solution will not be affected. c. Each additional hour of time (up to 47.83 hours) on test device 1 increases Quitmeyers prot by $1,284.52. This shadow price value, however, assumes that time on test device 1 costs only $15 per hour. The 20 additional hours at a cost of $25 per hour will therefore increase Quitmeyers prot by ($1,284.52 x 20), less the premium of ($10 x 20), or by $25,490.4. The new prot will be $220,995.23. The deal is worthwhile. d. First, we check the 100% rule. (20/48) (40/80) 0.916 1. The current shadow prices are valid. By making this trade, Quitmeyers new prot will be $195,504.83 20 $1,284.52 40 x $344.69 $183,602.03. The deal is not worthwhile. 4-19. See le P4-19.XLS for the Excel solution and Solver sensitivity report. a. Several of the allowable increase and allowable decrease values for the objective function coefcients are zero. For example, see these values for the number of acres of wheat in the SE parcel. Also, several variables (crops) that are currently not in the crop plan (for example, wheat in the NW parcel) have zero reduced costs. The current solution is, therefore, not a unique optimal solution and there are alternate optimal solutions. b. There are several ways in which we can get Solver to identify an alternate optimal solution. Perhaps the easiest approach is to rearrange the order in which the variables and/or constraints are presented in the model. For example, see sheet named Alternate in le P4-19.XLS, in which the columns corresponding to Alfalfa SE and Alfalfa N have been moved from columns G and H to columns J and K, respectively.

When we solve the problem now using Solver, we get a solution with the same prot ($337,862.07), but with a different crop plan. Other optimal crop plans are also possible. c. Increasing Barley sales by 10% would make the new limit 2,420 tons, or 1,100 acres. Based on the shadow price for this constraint, each additional acre of Barley (up to 108.57 acres) will increase Margarets prot by $37.59. The 100 additional acres will therefore increase prot by $3,758.60, to $341,620.67. d. Each additional acre-feet of water will allow Margaret to increase total prot by $20.69. 4-20. We use the answer and sensitivity reports given in Table 4.3 to answer the following questions. a. Each additional pound of material will increase prot by $0.5. The 2 pounds will therefore cause prot to increase to $29. b. Each additional hour of labor will increase prot by $1. The 1.5 hours will therefore cause prot to increase to $29.5. c. First, we check the 100% rule. (1/2) (1.5/4) 0.875 1. The shadow prices are valid. New prot $28 1.5 x $0.5 1 x $1 $27.75. The deal is not worthwhile. d. First, we check the 100% rule. (0.75/1) (0.25/1) 1. The solution remains optimal. New prot 2 x $4.75 4 x $4.75 $28.50. e. Decrease in current prot if 1 unit of the new product is produced 1 x $1 1 x $0.5 2 x $0 $1.5. Prot contribution of new product $2. Hence, the net prot will increase by $0.50 for each unit produced of the new product. 4-21. See le P4-21.XLS for the Excel solution and Solver sensitivity report. a. As we saw in Problem 3-14, there are approximately 2,791 medical patients and 2,105 surgical patients per year in the optimal solution. This translates to 61 medical beds and 29 surgical beds in the 90-bed addition. b. There are no empty beds with this optimal solution. Each additional patient day (over the current 32,850) will permit Mt. Sinai to increase revenue by $276.82. That is, by acquiring another bed (or 365 patient days), the revenue can be increased by $101,039. c. Labs have an unused capacity of 876.364. Acquiring more lab space is therefore not worthwhile. d. X-ray capacity is being utilized to its fullest extent. Each additional x-ray that can be handled will increase revenue by $65.45. e. The operating room has an unused capacity of 695.45. Acquiring more operating room is therefore not worthwhile. 4-22. We use the information given in Programs 4.9A and 4.9B to answer the following questions. a. The optimal production plan is to produce 540 Standard suitcases and 252 Deluxe suitcases, for a total prot of $7,668. At this point, no Luxury suitcases are scheduled for production. b. Given the optimal production plan, we can determine whether or not the new polishing process will have sufcient capacity to support that plan. The constraint (in hours) is: 1/6 (Standard) 1/6 (540) 1/4 (Deluxe) 1/3 (0) 1/3 (Luxury) 153 170 170 Now we substitute the previous optimal production decision: 1/4 (252)

6634 CH04 UG

8/23/02

1:53 PM

Page 24

24

CHAPTER 4

LINEAR PROGRAMMING SENSITIVITY ANALYSIS

Therefore, there is sufcient capacity in the proposed polishing operation to sustain the optimal production plan. c. The constraint for the water proong process (in hours) would be 1 (Standard) 1 (540) 1.5 (Deluxe) 1.75 (0) 1.75 (Luxury) 918 900 900 Substitution of the optimal production plan yields: 1.5 (252) Therefore, the proposed water proong process does not have the capacity to support the production plan. In order to determine the impact, it is necessary to go back to the original formulation and add the constraint. 4-23. are: The prot contributions per unit for the two new products Compact: $30 $5 0.2($8) $7.15 0.5($10) 0.75($6) 0.75($9)

production mix. Assuming labor costs do not change, this implies that material costs can vary between $1 ( $4 $5, rounded to $0) and $7.33 ( $4 $3.33) without affecting the current optimal production mix. Note that a decrease in material cost translates to an increase in the prot contribution, and vice versa. b. Given the optimal production plan, we can determine whether or not the new polishing process will have sufcient capacity to support that plan. The constraint (in hours) is: 1/6 (TiniTote) 1/6 (100) 1/4 (TubbyTote) 1/5 (90) 1/5 (ToddleTote) 43.42 48 48 Now we substitute the previous optimal production decision: 1/4 (35) Therefore, there is sufcient capacity in the proposed polishing operation to sustain the optimal production plan. 4-26 a. Each additional hour of fabrication time (up to 110.50 hours) will allow Strollers-to-Go to increase prot by $3.60. Hence, the rm would be willing to pay a premium of up to $3.60 for each additional hour of fabrication time, or up to $11.85 ( $8.25 $3.60) per hour. If the rm can get additional time for $10.50 per hour, it should take it. Each hour obtained this way (up to 110.50 hours) will increase prot by $1.35 [ $3.60 ($10.50 $8.25)]. b. The rst two bundles (80 hours total) should denitely be purchased at $10.50 per hour since the prot would increase by 80 x $1.35 $108. For the third bundle, we know that the rst 30.5 hours will cause prot to increase by 30.5 $1.35 $41.175. However, the shadow price of the last 9.5 hours of this bundle will be less than $3.60 (can you see why this is so?). In the worst case, if Strollers-to-Go has to pay $10.50 per hour for these hours too, and they remain unused, purchasing the third bundle becomes an unattractive option. The rm should therefore purchase only 2 bundles of 40 hours each. 4-27. The prot contribution per unit for the new products is: $72 $5.75 $9.375 3.5($8.25) 1.75($8.5) 1.5($8.75)

Kiddo: $37.50 $4.50 1.2($10) 0.5($9) 0.2($8) $10.40

0.75($6)

However, we need to remember that a positive prot contribution is not a sufcient condition for making the new product. That is because resources allocated to make the new product will have to be reallocated from existing products. We need to convert the value of these resources using the shadow prices, as follows: Compact: 0.5($4.38) 0.2($0) $7.395 Kiddo: 1.2($4.38) $8.726 0.75($0) 0.75($0) 0.75($6.94) 0.5($6.94) 0.2($0)

Since the prot contribution of $7.15 is smaller than the $7.395 value of the resources required, the Compact model is not attractive to make. However, the Kiddo model will more efciently convert resources to revenue (since the prot contribution of $10.40 exceeds the $8.726 value of the resources). We must include the Kiddo product in the formulation and solve the problem again using Solver to determine the new production plan. 4-24. We use the information given in Programs 4.10A and 4.10B to answer the questions problems 4-24 to 4-27. a. The optimal production plan is to make 100 TiniTote, 35 TubbyTote, and 90 ToddleTote strollers. The resulting prot is $2,086.25. The following constraints are binding: fabrication time, minimum production level for the TubbyTote model, and ratio of the ToddleTote model to the total production. b. All 620 hours are being used fabrication. Only 415 of the 500 hours are being used in Sewing, while only 385 of the 480 hours are being used in Assembly. c. Each additional hour of fabrication time (up to 110.50 hours) will allow Strollers-to-Go to increase prot by $3.60. Hence, the rm would be willing to pay a premium of up to $3.60 for each additional hour of fabrication time. In contrast, since sewing is a non-binding constraint, the rm would not be interested in obtaining any additional sewing time. d. None of the products are being produced at their maximum level (demand). However, the TubbyTote model is being produced at its minimum level. 4-25 a. From the sensitivity report, the prot contribution of TiniTote can vary between $5.92 ( $9.25 $3.33) and $14.25 ( $9.25 $5) without affecting the current optimal

However, we need to remember that a positive prot contribution is not a sufcient condition for making the new product. That is because resources allocated to make the new product will have to be reallocated from existing products. We need to convert the value of these resources using the shadow prices. If we assume that TwinTotes are not included in the 40% overall production limitation on ToddleTotes, the computation is as follows: 3.5($3.6) 1.75($0) 1.5($0) $12.60 Since the prot contribution of $9.375 is smaller than the $12.60 value of the resources required, the TwinTote model is not attractive to manufacture. Each TwinTote made will decrease prot by $3.225 (subject to round off). In contrast, if we assume that TwinTotes are also inclued in the 40% overall production limitation on ToddleTotes, the computation is now as follows: 3.5($3.6) 1.75($0) 1.5 ($0) 0.4($3.85) $11.06 In this case also, since the prot contribution of $9.375 is smaller than the $11.06 value of the resources required, the TwinTote model

6634 CH04 UG

8/23/02

1:53 PM

Page 25

CHAPTER 4

LINEAR PROGRAMMING SENSITIVITY ANALYSIS

25

is not attractive to manufacture. However, now each TwinTote made will decrease prot by only $1.685 (subject to round off).

SOLUTION TO RED BRAND CANNERS CASE

1. The main issue in this case is how to allocate 3 million pounds of tomatoes. The overall objective is to maximize total sales less variable costs. These costs include production and selling expenses. Twenty percent of the crop was grade A and the rest was grade B. In setting up the constraints, the amount of grade A tomatoes cannot exceed 20% of 3 million pounds. Thus not more than 600,000 pounds of grade A tomatoes can be used. Similarly, not more than 2,400,000 pounds of grade B tomatoes can be used. Furthermore, the demand for 50,000 cases of tomato juice and 80,000 cases of tomato paste should be met. The demand for whole tomatoes is not a constraint in this problem. Finally, minimum quality requirements should be met. This includes an average of 8 points per pound for whole tomatoes and 6 points per pound for tomato juice. There is no constraint for tomato paste. Another issue is whether or not to buy 80,000 additional pounds of grade A tomatoes. This would increase the amount of available grade A tomatoes from 600,000 pounds to 680,000 pounds. 2. The problem can be formulated using LP as follows. WA WB JA JB PA PB pounds of whole A tomatoes pounds of whole B tomatoes pounds of juice A tomatoes pounds of juice B tomatoes pounds of paste A tomatoes pounds of paste B tomatoes 0.0822WB 0.066JA 0.066JB

variable cost (less the tomatoes) is $2.52. Since the tomatoes are already on hand (and no salvage appears to be possible), they represent a sunk cost and are not part of the decision process. Since there are 18 pounds per case, the unit prot is (4.00 2.52)/18 0.0822. Similar analyses hold for the other terms in the objective function. The rst constraint refers to the 14.4 million pounds of whole tomatoes800,000 cases at 18 pounds per casethat constitutes maximum demand. Similarly, the maximum demand for tomato juice is 50,000 cases at 20 pounds per case or 1 million pounds, and the maximum demand for tomato paste is 80,000 cases at 25 pounds per case or 2 million pounds, and these are constraints 2 and 3. Constraints 4 and 5 reect the availability of grade A and grade B tomatoes, respectively, and the last two constraints are the quality constraints. The requirements that canned tomatoes must average at least 8 points means that 9WA 5WB 8(WA WB) 1WA 3WB 0 Similarly, the requirements that tomato juice must average at least 6 points is the last constraint. 3. See le P4-Red Brand.XLS for the Excel implementation of this model and its solution. The optimal solutions is: WA 525,000, WB 175,000, JA 75,000, JB 225,000, PA 0, and PB 2,000,000. The maximum prot is $225,355.60. 4. From the answer report (see le P4-Red Brand.XLS), we see that the following constraints are binding: grade A tomato availability, grade B tomato availability, paste demand, canned tomato quality, and juice quality. 5. From the sensitivity report (see le P4-Red Brand.XLS), the shadow price of grade A tomatoes is $0.0903 per pound, valid up to 600,000 additional pounds. The 80,000 pounds being offered at $0.085 per pound will therefore increase prot by $0.0053 per pound, or by a total of $424. 6. The demands for whole tomatoes and juice are not binding constraints. Hence, their underestimation will not affect the optimal solution. However, each additional pound of paste (up to 200,000 pounds) demanded will increase prot by $0.0161. If demand is 5% more (that is, 4,000 cases more at 25 pounds per case), the additional demand for 100,000 pounds will increase prot by $1,610. 7. First, we check the 100% rule. (100,000)/(466,666.67) (200,000)/(466,666.67) 0.64 1. The shadow prices are valid. The new prot is: $225,355.56 (200,000)($0.0579) (100, 000)($0.0903) $227,905.60. Gordon should accept this trade.

Maximize: 0.0822WA 0.074PA 0.074PB subject to 1WA 1WB 1JA 1WA 1WB 1WA 3WB 3JA All Variables 0 1JA

14,400,000 1JB 1PA 1PA 1JB 1JB 1PB 1PB 1,000,000 2,000,000 600,000 2,400,000 0 0

The coefcients in the objective function are the unit prots. A case of whole tomatoes (grade A and grade B) sells for $4. The