Operations Research

ENGR 3157
Lecture 09 – Integer Programming

Eugene Ben-Awuah, PhD, PEng

Bharti School of Engineering
Laurentian University

Office: F220
Tel: 705-675-1151 ext. 2195
Email: [email protected]

2
Recall: Types of Mathematical Programming

• Linear Programs (LP): the objective and constraint

functions are linear and the decision variables are
continuous.

• Integer Linear Programs (ILP): one or more of the decision

variables are restricted to integer values only and the
functions are linear.
– Pure IP: all decision variables are integer.
– Mixed IP (MIP): some decision variables are integer, others are
continuous.
– 1/0 MIP: some or all decision variables are further restricted to be
valued either “1” or “0”.

1
 3
MILP – Mine Production Scheduling
• Objective function: Maximize Net Present Value
T N
Max   v nt × x nt - q nt × y nt 
t=1 n=1

Block value (includes Cost of mining all

extra cost of mining ore) material as waste

• Constraints N
– Mining Tl     on  wn   ynt  Tut
t

n 1
N
– Processing Tl    on  xnt   Tut
t

n=1
– Blocks extraction precedence

4
A Simple LP Problem1

– A steel company must decide how to allocate production time on a

rolling mill. The mill takes unfinished slabs of steel as input and can
produce either of two products: bands and coils. The products come
off the mill at different rates and have different profitability:
Tons/ Profit/
hour ton
Bands 200 $25
Coils 140 $30
– The weekly production that can be justified based on current and
forecast orders are:
Maximum tons: Bands 6,000
Coils 4,000
– If 40 hours of production time are available, how many tons of bands
and coils should be produced to bring the greatest profit?

2 from, R. Fourer, D. Gay, B. Kernighan, AMPL, Boyd & Fraser, 1993, pp. 2-10.

2
 5
A Simple LP Problem – Solution

• Define the decision variables:

– xB number of tons of bands produced.
– xC number of tons of coils produced.

• Formulation:

Maximize: 25 xB  30 xC

Subject to: 1 200  xB  1 140  xC  40

0  xB  6000

0  xC  4000

6
A Simple LP Problem – Solution (2)

Coils
Coils

Constraints Profit
6000 6000
Hours
192K

4000 4000
Optimal solution

2000 Feasible region 2000

Bands 0 Bands
0
0 2000 4000 6000 8000 0 2000 4000 6000 8000

Optimal Solution:
Bands = 6000
Coils = 1400

3
 7
Expressed as an ILP Problem

• Assume that orders for bands and coils are placed (and
filled) in truckloads of 1,000s of tons only

• New symbolic model

– Let the new decision variables be the number of 1000 ton “units” or
orders of bands and coils.

Maximize: 25000 xB  30000 xC

Subject to: 1000 200  xB  1000 140  xC  40

0  xB  6 integer

0  xC  4 integer

8
Expressed as an ILP Problem (2)

• Solving graphically:

Coils

6 $185K

4 Optimal integer solution ($185K)

2
Feasible integer solutions
0 Bands
0 2 4 6 8

4
 9
Fixed-Charge ILP Problems

• A fixed-charge ILP problem is one in which a 0/1 variable, y,

represents a cost that is incurred if some other activity, x, is
undertaken at a non-zero level.
– e.g., if you produce any widgets, you also need to incur the cost of
installing the equipment to do so
– in other words: if x > 0 then y = 1, and if x = 0 then y = 0
– commonly used to formulate production and/or location problems

• It is implemented by adding a new constraint:

x ≤ M·y

• Can you see why this works?

10
In-Class Problem – Fixed-Charge ILP Problem

• Your factory can manufacture shirts, shorts, and pants for

profits of $6, $4, and $7, respectively. The machinery to
produce shirts costs $200 per week, for shorts is $150 per
week, and for pants is $100 per week. You have 150 hours
of labour and 160 m2 of cloth available per week. Labour
and cloth requirements for each product are as follows:
Labour (hrs) Cloth (m2)
shirt 3 4
shorts 2 3
pants 6 4

• Formulate an ILP that will maximize your profit.

5
 11
1/0 ILP Problems with Logical Constraints

• Many ILP problems will include economic indivisibilities,

where a project is either selected or not, and there is no
possibility of selecting a fraction of a project.
– These types of problems are called 1/0 ILP problems, and include at
least one variable whose feasible value is only 0 or 1.

• 1/0 ILP problems often have logical constraints, including:

– Either one particular project or another particular project can be
selected, but not both.
– The number of projects selected must be ≤, ≥, or = to some number
n.
– If a particular project is selected, then some other particular project
must also be selected.
– If a particular project is selected, then some other particular project
cannot also be selected.

12
Formulating Logical Constraints

• Either one particular project or another particular project

must be selected, but not both.
– If exactly one of two projects represented by variables x1 and x2 must
be selected, we add the following constraint:
x 1 + x2 = 1

• The number of projects selected must be ≤, ≥, or = to some

number n.
– If exactly n out of m projects must be selected, we add:
x1 + x2 + x3 + … + xm-1 + xm = n
– If no more than n out of m projects must be selected, we add:
x1 + x2 + x3 + … + xm-1 + xm ≤ n
– If at least n out of m projects must be selected, we add:
x1 + x2 + x3 + … + xm-1 + xm ≥ n

6
 13
Formulating Logical Constraints (2)

• If a particular project is selected, then some other particular

project must also be selected.
– If a project represented by variable x1 can only be selected if a
project represented by x2 is also selected, we add:
x1 ≤ x2
– If two projects represented by variables x1 and x2 must be selected
together or not at all, we add:
x 1 = x2

• If a particular project is selected, then some other particular

project cannot also be selected.
– If two projects represented by variables x1 and x2 are mutually
exclusive (we select one or neither, but not both), we add:
x 1 + x2 ≤ 1

14
In-Class Problem – Logical Constraints

• You have $18 000 available for investing in 6 different

projects. Each project has a cost and profit as follows:
1 2 3 4 5 6
Cost $5 $7 $4 $3 $4 $3
Profit $16 $22 $12 $8 $10 $15

• A project is either invested in at its full cost or not at all.

Formulate the ILP model that will maximize your profit if:
1. You must invest in either project 2 or 4, but not both.
2. You can only invest in either project 2 or 4, but not both.
3. You must invest in at least 3 different projects.
4. You can invest in no more than 3 different projects.
5. You must invest in exactly 3 different projects.
6. If you invest in project 2, you also need to invest in project 5.
7. If you invest in either project 1 or 2, you must invest in them both.

7
 15
Formulating Either-Or Constraints

• In some ILP problems, we might have a pair of constraints

where only one or the other needs to be satisfied.
– For example, either the total number of widgets and gadgets
produced weekly in factory A must be at least 5000 or the total
number produced weekly in factory B must be at least 7000.

• If only one of the following needs to be satisfied:

f(x1,x2,…,xn) ≤ 0
g(x1,x2,…,xn) ≤ 0

– we replace those constraints with:

f(x1,x2,…,xn) ≤ M·y
g(x1,x2,…,xn) ≤ M·(1-y)

– where y is a 1/0 variable (it either equals 1 or it equals 0)

16
Formulating If-Then Constraints

• Many applications require the use of an if-then type of

constraint system where:
– If f(x1, x2, …, xn) > 0 is satisfied, then g(x1, x2, …, xn) ≥ 0 must also be
satisfied.
– Or said another way, if f(x1, x2, …, xn) ≤ 0 is not satisfied, then g(x1,
x2, …, xn) ≥ 0 must be satisfied.

• To implement that system of constraints, we use:

-g(x1,x2,…,xn) ≤ M·y
f(x1,x2,…,xn) ≤ M·(1-y)

– where y is a 1/0 variable

8
 17
In-Class Problem – Either-Or and If-Then

• Referring back to the previous in-class problem (dealing with

logical constraints), formulate the model if:
– You must either invest in at least one of projects 1 and 2 or no more
than two of projects 3, 4, 5, and 6.
– If you invest in more than one of projects 1 and 2, then you can
invest in no more than one of projects 3, 4, 5, and 6.

18
Formulating Piecewise Linear Costs

• A piecewise linear function consists of several linear

segments, where the slope of the overall function changes
at break points throughout its range.
– For example, you will often get quantity discounts on goods or
supplies you purchase.

0
0 2 4 6 8 10 12

9
 19
Formulating Piecewise Linear Costs (2)

• Consider that quantities of x ≤ b1 cost c1 per unit, quantities

of b1 < x ≤ b2 cost c2 each, quantities of b2 < x ≤ b3 cost c3
each, and quantities of b3 < x ≤ b4 cost c4 each.

• These costs can be modelled by defining new variables:

0 ≤ x 1 ≤ b1
where each variable xi represents the
0 ≤ x2 ≤ w1 (b2 – b1)
number of units purchased at cost ci, and
0 ≤ x3 ≤ w2 (b3 – b2) wi is a 1/0 variable that represents whether
0 ≤ x4 ≤ w3 (b4 – b3) or not you purchase the full quantity at ci.

– and using an if-then type of constraints systems so that:

0 ≤ w1 (b1) ≤ x1
0 ≤ w2 (b2 – b1) ≤ x2
0 ≤ w3 (b3 – b2) ≤ x3

20
In-Class Problem – Piecewise Linear Costs

• Suppose you need to purchase computers whose costs are

as follows:
– $2,000 each if you buy 1 to 10
– $1,800 each if you buy 11 to 25
– $1,700 each if you buy 26 to 50
– Note that no more than 50 are available.

• Formulate the relevant part of the model to minimize the

total cost of purchasing the computers.

10
 21
Solving ILP Problems: Branch-and-Bound

• The LP relaxation of an ILP problem is where all integer

variables are relaxed (i.e., allowed to be real or fractional).

• The LP relaxation of an ILP problem will give one of two

results
– an integer solution (i.e., all decision variables that need to be
integer do in fact take on inter values)
– a fractional solution (i.e., some decision variables that need to be
integer take on non-integer values)

• The branch-and-bound method of solving an ILP starts by

solving its LP relaxation (say, via the simplex method).
– If the result is an integer solution, stop (it’s an optimal solution).
– If the result is a fractional solution, select one of the real-valued
integer decision variables for branching.

22
Solving ILP Problems: Branch-and-Bound (2)

• Branching generates two new LP sub-problems by:

– selecting one of the integer decisions variables, xi, whose value, c, is
non-integer, and
– adding constraints that will restrict that particular value from re-
occurring in subsequent sub-problems.

• The two new sub-problems take the form of:

– the current sub-problem plus the constraint xi ≤ floor(c)
– the current sub-problem plus the constraint xi ≥ ceiling(c)

• Solving each of the new sub-problems (say, with simplex)

will result in an integer solution, an infeasibility, or another
fractional solution.
– If the solution is fractional, the process is repeated.

11
 23
Solving ILP Problems: Branch-and-Bound (3)

• We continue branching on each new sub-problem until all

branches are fathomed, which means:
– the sub-problem is infeasible,
– the optimal solution of the sub-problem is worse than the current best
integer solution, or
– the sub-problem gives an integer solution.

• This creates a branch-and-bound solution space that

consists of a tree of LP sub-problems, where:
– the root node of the tree is the LP relaxation of the original ILP
problem (i.e., all integer variables are relaxed),
– integer variables become progressively more restricted in value
(potentially to the point where their values are fixed), and
– all integrality constraints of subsequent sub-problems are relaxed.

24
Example – Branch and Bound
Solution Bounds
LP relaxation of 1
Obj. = 151 0  x1  10
original problem x1 = 3.4 0  x2  10
x2 = 6.8
1
Bounds x1
0  x1  3 Bounds
0  x2  10 4  x1  10 10
2 3
0  x2  10 9
8
7
6
5
4
3
2
1 x2
0
0 1 2 3 4 5 6 7 8 9 10

Source: J. Doucette, Ph.D. Thesis, UofA, 2004

12
 25
Example – Branch and Bound (2)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
2
Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
7
Bounds 6
Bounds 0  x1  3 5
0  x1  3 8  x2  10
0  x2  7 4 5 4
3
2
1 x2
0
0 1 2 3 4 5 6 7 8 9 10

Source: J. Doucette, Ph.D. Thesis, UofA, 2004

26
Example – Branch and Bound (3)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
4
Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
2 7
Bounds 6
Solution Bounds 0  x1  3 5
Obj. = 169 0  x1  3 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 6.0 3
Integer Solution 2
1 x2
0
0 1 2 3 4 5 6 7 8 9 10

Source: J. Doucette, Ph.D. Thesis, UofA, 2004

13
 27
Example – Branch and Bound (4)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
5
Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
2 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution 2
1 x2
0
0 1 2 3 4 5 6 7 8 9 10

Source: J. Doucette, Ph.D. Thesis, UofA, 2004

28
Example – Branch and Bound (5)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
6
Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
2 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution
5
2
1 x2
Solution Bounds Bounds 0
Obj. = 164 0  x1  2 3  x1  3 0 1 2 3 4 5 6 7 8 9 10
x1 = 1.9 8  x2  10 6 7 8  x2  10
x2 = 8.4

Source: J. Doucette, Ph.D. Thesis, UofA, 2004

14
 29
Example – Branch and Bound (6)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
8
Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
2 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution
5
2
1 x2
Solution Bounds Bounds 0
Obj. = 164 0  x1  2 3  x1  3 0 1 2 3 4 5 6 7 8 9 10
x1 = 1.9 8  x2  10 6 7 8  x2  10
x2 = 8.4

6
Solution Bounds
Obj. = 166 0  x1  2
x1 = 2.0 8  x2  8 Bounds
x2 = 8.0 0  x1  2
8 9
Integer Solution 9  x2  10 Source: J. Doucette, Ph.D. Thesis, UofA, 2004

30
Example – Branch and Bound (7)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8

Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
2 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution
5
2
1 x2
Solution Bounds 0
Obj. = 164 0  x1  2 0 1 2 3 4 5 6 7 8 9 10
x1 = 1.9 8  x2  10 6 7
x2 = 8.4

6
Solution Bounds
Obj. = 166 0  x1  2 7 8
x1 = 2.0 8  x2  8 Solution Bounds
x2 = 8.0 infeasible 0  x1  2
8 9
Integer Solution 9  x2  10 Source: J. Doucette, Ph.D. Thesis, UofA, 2004

15
 31
Example – Branch and Bound (8)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
7
Solution Bounds x1
Obj. = 156 0  x1  3 Bounds
x1 = 2.4 0  x2  10 4  x1  10 10
2 3
x2 = 7.6 0  x2  10 9
8
2 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution
5
2
1 x2
Solution Bounds 10 11
Solution Bounds 0
Obj. = 164 0  x1  2 Obj. = 168 3  x1  3 0 1 2 3 4 5 6 7 8 9 10
x1 = 1.9 8  x2  10 6 7 x1 = 3.0 8  x2  10
x2 = 8.4 x2 = 9.0
Integer Solution
6 9
Solution Bounds
Obj. = 166 0  x1  2 7 8
x1 = 2.0 8  x2  8 Solution Bounds
x2 = 8.0 infeasible 0  x1  2
8 9
Integer Solution 9  x2  10 Source: J. Doucette, Ph.D. Thesis, UofA, 2004

32
Example – Branch and Bound (9)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8
3
Solution Bounds 14 15 x1
Obj. = 156 0  x1  3 Solution Bounds
x1 = 2.4 0  x2  10 Obj. = 167 4  x1  10 10
2 3
x2 = 7.6 x1 = 4.0 0  x2  10 9
x2 = 7.4 8
2 13 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution
5 12
2
1 x2
Solution Bounds 10 11
Solution Bounds 0
Obj. = 164 0  x1  2 Obj. = 168 3  x1  3 0 1 2 3 4 5 6 7 8 9 10
x1 = 1.9 8  x2  10 6 7 x1 = 3.0 8  x2  10
x2 = 8.4 x2 = 9.0
Integer Solution
6 9
Solution Bounds
Obj. = 166 0  x1  2 7 8
x1 = 2.0 8  x2  8 Solution Bounds
x2 = 8.0 infeasible 0  x1  2
8 9
Integer Solution 9  x2  10 Source: J. Doucette, Ph.D. Thesis, UofA, 2004

16
 33
Example – Branch and Bound (10)
Solution Bounds
1
Obj. = 151 0  x1  10
x1 = 3.4 0  x2  10
1 x2 = 6.8

Solution Bounds 14 15 x1
Obj. = 156 0  x1  3 Solution Bounds
x1 = 2.4 0  x2  10 Obj. = 167 4  x1  10 10
2 3
x2 = 7.6 x1 = 4.0 0  x2  10 9
x2 = 7.4 8
2 13 7
3 4 Solution Bounds 6
Solution Bounds Obj. = 162 0  x1  3 5
Obj. = 169 0  x1  3 x1 = 2.9 8  x2  10
x1 = 3.0 0  x2  7 4 5 4
x2 = 8.0
x2 = 6.0 3
Integer Solution
5 12
2
1 x2
Solution Bounds 10 11
Solution Bounds 0
Obj. = 164 0  x1  2 Obj. = 168 3  x1  3 0 1 2 3 4 5 6 7 8 9 10
x1 = 1.9 8  x2  10 6 7 x1 = 3.0 8  x2  10
x2 = 8.4 x2 = 9.0
Integer Solution
Solution Bounds
6 9 Overall best solution:
Obj. = 166 0  x1  2 7 8
x1 = 2.0 8  x2  8 Solution Bounds
x1 = 2 and x2 = 8
x2 = 8.0 infeasible 0  x1  2
8 9
Integer Solution 9  x2  10 Source: J. Doucette, Ph.D. Thesis, UofA, 2004

34
Branch and Bound Take Home

Max Z = 2x1  x2
Subject to: 5 x1  2 x2  8
x1  x2  3
x1, x2  0; x1 integer

17