Problem Set – Integer Programming (Solution)

Problem 1: The Northshore Bank is working to develop an efficient work schedule for full time and
part time tellers. The schedule must provide for efficient operation of the bank including adequate
customer service, employee breaks, and so on. On Friday, the bank is open from 9:00 A.M to 7:00
P.M. The number of tellers necessary to provide adequate customer service during each hour of
operation is summarized here.

Time Number of Tellers Time Number of Tellers

9:00 A.M -10:00 A.M. 6 2:00 P.M.-3:00 P.M. 6
10:A.M.-11:00 A.M. 4 3:00 P.M.-4:00 P.M. 4
11:00 A.M.-Noon 8 4:00 P.M.-5:00 P.M. 7
Noon-1:00 P.M. 10 5:00 P.M.-6:00 P.M. 6
1:00 P.M.-2:00 P.M. 9 6:00P.M.-7:00P.M. 6
Each full time employee starts on the hour and works a 4-hour shift, followed by 1 hour for lunch
and then a 3-hour shift. Part time employee works one 4-hour shift beginning on the hour.
Considering salary and fringe benefits, full time employee cost the bank $15 per hour ($105 a day)
and part time employees cost the bank $8 per hour ($32 per day).

a) Formulate an integer-programming model that can be used to develop a schedule that will
satisfy customer service needs at a minimum employee cost.
b) Solve the LP Relaxation of your model in part (a)
c) Solve for the optimal schedule of tellers. Comment on the solution.
d) After reviewing the solution to part (c), the bank manager realized that some additional
requirement to specified. Specifically, she wants to ensure that one full time employee is on
duty at all times and that there is a staff of at least five full time employees. Revise your
model to incorporate these additional requirements and solve for the optimal solution.

It is a practical, complex scheduling problem that can be easily modelled and solved as IP. Note that
part time and full time employees are paid at different rates. So, it is important that the bank
operates to provide required service level (indicated by the number of tellers per hour) while
minimizing total cost due to staff salary.

We use x as full time and y as part time staff.

It would help if you construct a table as shown below to better analyze the problem before
attempting to model it. Red shows work hrs of full time staff and green is for part timers. Just follow
their work hour rules. The table lists all valid alternatives.
9:00- 10:00- 11:00- 12:00- 13:00- 14:00- 15:00- 16:00- 17:00- 18:00-
10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00

So, full timers have 3 options:

(i) start at 9 – work till 1pm, take 1 hr break, then work 2 – 5 pm

(ii) start at 10 – work till 2pm, take 1 hr break, then work 3 – 6 pm
(iii) start at 11 – work till 3pm, take 1 hr break, then work 4 – 7 pm
that’s all… bank closes at 7 pm.

However, part timers don’t have 1 hr break. They work straight 4 hrs and leave. See green cells
above to understand their alternatives.

Now, decision variables should include this work hour information. Here goes their definition. You
must clearly define decision variables in exam, otherwise, the model is meaningless.

X9 = number of full time staff who start at 9 and follow the hour-related rule (no need to specify
more details; those are already undestood)

X10 = number of full time staff who start at 10 and follow the hour-related rule

X11 = number of full time staff who start at 11 and follow the hour-related rule

Y9 = number of part time staff who start at 9 and follow the hour-related rule

Y10 = number of part time staff who start at 10 and follow the hour-related rule

…. This way you define up to Y15.

You must also specify that:

Xi, Yi >=0, integers.

Objective Function:

Minimize Z = 105*(x9 + x10 + x11) + 32* (y9 + y10 + … + y 15)

You will need one “covering” constraint for each working hour. Both full and part-timers can “cover”
the duty of each hour. Look at the color-coded table (above) column-wise. The 9:00-10:00 hour can
be covered only by x9 and y9 staff; next hour by x9, x10, y9, y10 only,… and so on. So, the
constraints are:

9:00- X9 + Y9 >= 6
10:00
10:00- X9 + X10 + Y9 + Y10 >=4
11:00
11:00- X9 + X10 + X11 + Y9 + Y10 + Y11 >=8
12:00
12:00- X9 + X10 + X11 + Y9 + Y10 + Y11 + Y12 >=10
13:00
13:00- X10 + X11 + Y10 + Y11 + Y12 +Y13 >=9
14:00
14:00- X9 + X11 + Y11 + Y12 +Y13 + Y14 >=6
15:00
15:00- X9 + X10 + Y12 +Y13 + Y14 + Y15 >=4
16:00
16:00- X9 + X10 +X11 +Y13 + Y14 + Y15 >=7
17:00
17:00- X10 +X11 + Y14 + Y15 >=6
18:00
18:00- X11 + Y15 >=6
19:00

b. Solution to LP Relaxation (verify for correctness): Cost: $672.

y9 = 6, y12 = 6, y3 = 6, y11 = 2, y13 = 1

All other variables = 0.

c. The solution to the LP Relaxation is integral, so, it is the optimal to the IP as well.

Managerial Insight: A difficulty with this solution is that only part-time employees are used; this may
cause problems with supervision, etc. The large surpluses from 5, 12-1 (4 employees), and 3-4 (9
employees) indicate times when the tellers are not needed for customer service and may be
reassigned to other tasks.

d. Add following constraints:

x9 ≥ 1

x11 ≥ 1 (why I didn’t add x10>=1? Think.)

x9 + x 10 + x 11 ≥ 5

The new optimal solution, which has a daily cost of $909 is

x9 = 1, y9 = 5, x11 = 4, y12 = 5, y3 = 2

There is now much less reliance on part-time employees. The new solution uses 5 full-time
employees and 12 part-time employees; the previous solution used no full-time employees and 21
part-time employees.
Problem 2: Burnside Marketing Research conducted to study for Barker Foods on some designs for
a new dry cereal. Three attributes were found to be most influential in determining which cereal had
the best taste: ratio of wheat to corn in the cereal flake, type of sweetener (Sugar, honey, or
artificial), and the presence or absence of flavour bits. Seven children participated in taste tests and
provide the following part worths for attributes.

Wheat/Corn Sweetener Flavor Bits

Child Low High Sugar Honey Artificial Present Absent
1 15 35 30 40 25 15 9
2 30 20 40 35 35 8 11
3 40 25 20 40 10 7 14
4 35 30 25 20 30 15 18
5 25 40 40 20 35 18 14
6 20 25 20 35 30 9 16
7 30 15 25 40 40 20 11

a) Suppose the overall utility (sum of part worths) of the current favourite cereal is 75 for each
child. What is the product design that will maximize the share of choices for the seven
children in sample?
b) Assume the overall utility of the current favourite cereal for the first four children in the
group is 70, and the overall utility of the current favourite cereal for the last three children in
the group is 80. What is the product design that will maximize the share of choices for the
seven children in sample?

a.

Decision variables

Let L_ij = 1, if level i is chosen for attribute j, 0 otherwise

y_k = 1, if child k prefers the new cereal design, 0 otherwise

Math model:

Max y_1 + y_2 + y_3 + y_4 + y_5 + y_6

s.t.
15L_11 + 35L_21 + 30L_12 + 40L_22 + 25L_32 + 15L_13 + 9L_23 - 75y_1 ≥ 1
30L_11 + 20L_21 + 40L_12 + 35L_22 + 25L_32 + 8L_13 + 11L_23 - 75y_2 ≥ 1
40L_11 + 25L_21 + 20L_12 + 40L_22 + 10L_32 + 7L_13 + 14L_23 - 75y_3 ≥ 1
35L_11 + 30L_21 + 25L_12 + 20L_22 + 30L_32 + 15L_13 + 18L_23 - 75y_4 ≥ 1
25L_11 + 40L_21 + 40L_12 + 20L_22 + 35L_32 + 18L_13 + 14L_23 - 75y_5 ≥ 1
20L_11 + 25L_21 + 20L_12 + 35L_22 + 30L_32 + 9L_13 + 16L_23 - 75y_6 ≥ 1
30L_11 + 15L_21 + 25L_12 + 40L_22 + 40L_32 + 20L_13 + 11L_23 - 75y_7 ≥ 1

(only one level to choose for each attribute)

L_11 + L_21 = 1
 L_12 + L_22 + L_32 = 1
L_13 + L_23 = 1
The optimal solution obtained on Excel shows L_11 = L_32 = L_13 = 1. This indicates that a
cereal with a low wheat/corn ratio, artificial sweetener, and no flavor bits will maximize the
share of choices. The optimal solution also has y_4 = y_5 = y_7 = 1 which indicates that
children 4, 5, and 7 will prefer this cereal.

b. The coefficients for the y_i variable must be changed to (-70) in constraints 1-4 and to
(-80) in constraints 5-7. The new optimal solution has L_21 = L_12 = L_23 = 1. This is a
cereal with a high wheat/corn ratio, a sugar sweetener, and no flavor bits. Four children
will prefer this design: 1, 2, 4, and 5.

Diagram of display rooms for Bayside art gallery (For Problem 3)

Problem 3: The Bayside Art Gallery is considering installing a video camera security system to
reduce its insurance premium. A diagram of the eight display rooms that Bayside uses for exhibition
is shown in figure below; the openings between the rooms are numbered 1-13. A security firm
proposed that two-way camera be installed at some room opening. Each camera has ability to
monitor the two rooms between which the camera is located. For example, if a camera were located
at opening number 4, room 1 and 4 would be covered; if a camera were located at opening 11, room
7 and 8 would be covered; and so on. Management decided not to locate a camera system at the
entrance to the display rooms. The objective is to provide security coverage for all eight rooms using
the minimum number of two-way cameras.

a) Formulate a 0-1 integer linear programming model that will enable Bayside’s management
to determine the location for the camera systems.
b) Solve the model formulated in part (a) to determine how many two-way cameras to
purchase and where they should be located.
c) Suppose that management wants to provide additional security coverage for room 7.
Specifically, management want room 7 to be covered by two cameras. How your model
would formulate in part (a) have to change to accommodate this policy restriction?
d) With the policy, restriction specified in part (c) determines how many two-way camera
systems will need to be purchased and where they will be located.

a.

X_i = 1, if camera is located at opening I, 0 otherwise

min x_1 + x_2 + x_3 + x_4 + x_5 + x_6 + x_7 + x_8 + x_9 + x_10 + x_11 + x_12 + x_13
s.t.
x_1 + x_4 + x_6 ≥ 1 Room 1
x_6 + x_8 + x_12 ≥ 1 Room 2
x_1 + x_2 + x_3 ≥ 1 Room 3
x_3 + x_4 + x_5 + x_7 ≥ 1 Room 4
x_7 + x_8 + x_9 + x_10 ≥ 1 Room 5
x_10 + x_12 + x_13 ≥ 1 Room 6
x_2 + x_5 + x_9 + x_11 ≥ 1 Room 7
x_11 + x_13 ≥ 1 Room 8

b.
x 1 = x 5 = x 8 = x 13 = 1. Thus, cameras should be located at 4 openings: 1, 5, 8, and 13.
An alternative optimal solution is x 1 = x 7 = x 11 = x 12 = 1.

c.

Change the constraint for room 7 to x 2 + x 5 + x 9 + x 11 ≥ 2

d.

x 3 = x 6 = x 9 = x 11 = x 12 = 1. Thus, cameras should be located at openings 3, 6, 9, 11, and 12. An

alternate optimal solution is x 2 = x 4 = x 6 = x 10 = x 11 = 1. Optimal Value = 5
Problem 4: Roedel Electronics produces a variety of electrical components, including a remote
controller for televisions and a remote controller for VCR. Each controller consists of three
subassemblies that are manufactured by Roedel: a base, a cartridge and a keypad. Both controllers
use the same base subassembly, but different cartridge and keypad subassemblies.

Roedel’s sale forecast indicates that 7000 TV controllers and 5000 VCR controllers will be needed to
satisfy demand during the upcoming Christmas season. Because only 500 hours of in-house
manufacturing time are available, Roedel is considering purchasing some or all of the subassemblies
from outside suppliers. If Roedel manufactures a subassembly in-house, it incurs a fixed setup costs
as well as a variable manufacturing cost. The following table shows the setup cost, the
manufacturing time per subassembly, the manufacturing cost per subassembly, and the cost to
purchase each of the subassemblies from an outside supplier.

Subassembly Setup Cost($) Manufacturing Manufacturing Purchase Cost

Time per Cost per Unit($) per Unit($)
Unit(min)
Base 1000 0.9 0.40 0.65
TV cartridge 1200 2.2 2.90 3.45
VCR cartridge 1900 3.0 3.15 3.70
TV keypad 1500 0.8 0.30 0.50
VCR keypad 1500 1.0 0.55 0.70

a) Determine how many units of each subassembly Roedel should manufacture and how many
units Roedel should purchase. What is the total manufacturing and purchase cost associated
with your recommendation?
b) Suppose Roedel is considering purchasing new machinery to produce VCR cartridges, for the
new machinery, the setup cost is $3000; the manufacturing time is 2.5 minutes per
cartridge, and the manufacturing cost is $2.60 per cartridge. Assuming that the new
machinery is purchased, determine how many units of each subassembly Roedel should
manufacture and how many units of each subassembly Roedel should purchase. What is the
total manufacturing and purchase cost associated with your recommendation? Do you think
the new machinery should be purchased? Explain.

a.
A mixed integer linear program can be set up to solve this make-buy decision problem. Binary
variables are used to indicate whether or not we setup to produce the subassemblies.

Let SB = 1 if bases are produced; 0 if not

STVC = 1 if the subassembly TV cartridges are produced; 0 if not
SVCRC = 1 if the subassembly VCR cartridges are produced; 0 if not
STVP = 1 if the subassembly TV keypads are produced; 0 if not
SVCRP = 1 if the subassembly VCR keypads are produced; 0 if not
BM >=0, No. of bases made
BP >=0, No. of bases purchased
TVCM >=0, No. of TV cartridges made
…
…
VCRPP >=0, No. of VCR keypads purchased
A mixed integer linear programming model for solving this problem follows. There are 11
constraints. Constraints (1) to (5) are to satisfy demand. Constraint (6) reflects the limitation on
manufacturing time. Finally, constraints (7) - (11) are constraints not allowing production unless the
setup variable equals 1 (FORCING CONSTRAINTS). Variables SB, STVC, SVCRC, STVP, and SVCRP must
be specified as 0/1.

MIN Z = (0.4 BM + 2.9 TVCM + 3.15 VCRCM + 0.3 TVPM + 0.55 VCRPM)
+(0.65 BP + 3.45 TVCP + 3.7VCRCP + 0.5 TVPP + 0.7 VCRPP)
+(1000 SB + 1200 STVC + 1900 SVCRC + 1500 STVP + 1500 SVCRP)

S.T.
1) 1BM+1BP=12000 //Base’s demand
2) +1TVCM+1TVCP=7000 //TV controller’s demand
3) +1VCRCM+1VCRCP=5000 //VCR controller’s demand
4) +1TVPM+1TVPP=7000 //TV keypad’s demand
5) +1VCRPM+1VCRPP=5000 //VCR keypad’s demand
6) 0.9 BM + 2.2 TVCM + 3 VCRCM + 0.8 TVPM + 1VCRPM <= 30000 min (=500 hr) //time constraint
7) BM <= 12000 * SB
8) TVCM <= 7000STVC
9) VCRCM <= 5000SVCRC
10) TVPM <= 7000STVP
11) VCRPM <= 5000SVCRP

OPTIMAL SOLUTION
Objective Function Value = 52800.00
Variable Value
-------------- ---------------
BM 12000.000
TVCM 7000.000
VCRCM 0.000
TVPM 0.000
VCRPM 0.000
BP 0.000
TVCP 0.000
VCRCP 5000.000
TVPP 7000.000
VCRPP 5000.000
SB 1.000
STVC 1.000
SVCRC 0.000
STVP 0.000
SVCRP 0.000

b. Do it yourself.
 Problem 5: East Coast Trucking provide service from Boston to Miami using regional offices located
in Boston, New York, Philadelphia, Baltimore, Washington, Richmond, Raleigh, Florence, Savannah,
Jacksonville and Tampa. The number of miles between each of the regional offices is provided in the
following table.

New Philadelphia Baltimore Washington Richmond Raleigh Florence Savannah Jacksonville Tampa Miami
York

Boston 211 320 424 459 565 713 884 1056 1196 1399 1669

New York 109 213 248 354 502 673 845 985 1188 1458
Philadelphia 104 139 245 393 564 736 876 1079 1349
Baltimore 35 141 289 460 632 772 975 1245
Washington 106 254 425 597 737 940 1210
Richmond 148 319 491 631 834 1104
Raleigh 171 343 483 686 956

Florence 172 312 515 785

Savannah 140 343 613
Jacksonville 203 473
Tampa 270

The company’s expansion plans involve constructing service facilities in some of the cities where a
regional office is located. Each regional office must be within 400 miles of a service facility. For
instance, if a service facility is constructed in Richmond, it can provide service to regional office
located in New York, Philadelphia, Baltimore, Washington, Richmond, Raleigh, and Florence.
Management would like to determine the minimum number of service facilities needed and where
they should be located.

a) Formulate an integer linear program that can be used to determine the minimum number of
services facilities needed and their location.
b) Solve the linear program formulated in part (a). How many service facilities are required,
and where should they be located?
c) Suppose that each service facility can only provide service to regional office within 300
miles. How many service facilities are required and where should they be located?

Candidate facility locations: Boston, NY, …, Tampa (column-wise).

Customer (service receiver) locations: NY, …, Miami (top row of table)
The distance table is a from-to matrix – whose diagonal elements are zeros. Also a_ij = a_ji
(not explicitly given). So, New York to Baltimore = Baltimore to NY = 213. You need to
complete the table before further proceeding.
Now, if dist between cities I and j = d_{ij} <= 400 miles, then a_{ij} = 1, else 0. This way,
construct a coverage matrix [a_{ij}].
Then, you have a covering constraint of “>=” type for each customer i.
 x_i = 1, if Facility is located at I, 0 otherwise; where i=1 ,…, 11. Note that Boston is only a
supply node and Miami is only a demand node. All other nodes are both demand nodes as
well as candidate facility locations.

min x_1 + x_2 + x_3 + x_4 + x_5 + x_6 + x_7 + x_8 + x_9 + x_10 + x_11 + x_12
s.t.
(New York) x_1 + x_2 + x_3 + x_4 + x_5 + x_6 ≥ 1
(Philadelphia) x_1 + x_2 + x_3 + x_4 + x_5 + x_6 + x_7 ≥ 1
(Baltimore) x_2 + x_3 + x_4 + x_5 + x_6 + x_7 ≥ 1
(Washington) x_2 + x_3 + x_4 + x_5 + x_6 + x_7 ≥ 1
(Richmond) x_2 + x_3 + x_4 + x_5 + x_6 + x_7 + x_8 ≥ 1
(Raleigh) x_3 + x_4 + x_5 + x_6 + x_7 + x_8 + x_9 ≥ 1
(Florence) x_6 + x_7 + x_8 + x_9 + x_10 ≥ 1
(Savannah) x_7 + x_8 + x_9 + x_10 + x_11 ≥ 1
(Jacksonville) x_8 + x_9 + x_10 + x_11 ≥ 1
(Tampa) x_9 + x_10 + x_11 ≥ 1
(Miami) x_11 ≥ 1

a. 2 service facilities: Richmond and Tampa.

b. This needs re-work of the constraints. Baltimore, Savannah, Tampa are now open.
Here is a summary of further analysis:

Coverage Dist No. Of Facilities

400 2
350 3
300 3
250 infeasible

Problem 6. Solve the followings by branch and bound. (Z+ means set of all positive integers)

a.

b.
Note the following is a Mixed IP (MIP) problem.

c.