Operations Research

Chapter 2: Solving Integer Programming

Sonia Rebai- M. Naceur Azaiez

Tunis Business School

University of Tunis
Relaxation
Definition

The LP obtained by omitting all integer or 0-1 constraints on variables is called

the LP relaxation of the IP.

• Any IP may be viewed as the corresponding LP relaxation plus some

additional constraints.
• The LP relaxation is a less constrained version of the IP; which means that:
the feasible region for any IP must be contained in the feasible region for
the corresponding LP relaxation.
Exemple 1
Max Z= 3 X1 + 2 X2

3 X1 + X2 £ 9

X1 + 3 X2 £ 7

- X1 + X2 £ 1

X1, X2 integers
Exemple 1 - continued
A naive approach for
solving an IP: First solve
B (1)

C the LP relaxation; then

(3)

A
(2)
round off each variable
that required to be an
O D
integer.

The optimal solution of the relaxed LP is determined by the point C:

X1=2.5; X2=1.5 et Z=10.5
• If the feasible region for the LP relaxation is bounded, the feasible

region for the IP will consist of a finite number of points.

• Such an IP could be solved by enumerating the z-values for each

feasible point and determining the feasible point having the

largest z-value.
• The feasible region for this problem is the

following set of points:

B (1)
S={(0,0),(1,0),(2,0),(3,0),(0,1),(1,1),(2,1)}
(3) C

• The optimal solution is D: x1=3, x2 = 0, A

(2)

z = 9. O D
The Branch-and-Bound method
ü Most IPs are solved by using the technique of branch-and-bound.

ü Branch-and bound methods find the optimal solution by efficiently

enumerating the points in a subproblem’s feasible region.

ü The branch-and-bound method begins by solving the LP relaxation of the IP.

ü If all the decision variables assume integer values in the optimal solution to
the LP relaxation, then the optimal solution to the LP relaxation will be the
optimal solution to the IP.
The Branch-and-Bound method - continued
ü We call the LP relaxation subproblem 1.

ü Unfortunately, the optimal solution to the LP relaxation is non integers

X1=2.5; X2=1.5 et Z=10.5

ü We know that for any IP that is a max problem:

Optimal z-value for LP relaxation ≥ Optimal z-value for IP

This implies that the optimal z-value for the IP cannot exceed 10.5. Thus,
the optimal z-value for the LP relaxation is an upper bound for the IP.
The Branch-and-Bound method - continued
ü To find out more about the location of the IP’s optimal solution, the feasible

region for the LP relaxation is partitioned as follows.

ü We arbitrarily choose a variable that is fractional in the optimal solution to

the LP relaxation-say, X1.

ü Note that every point in the feasible region for the IP must have either X1£2

or X1³3.
The Branch-and-Bound method - continued
ü Then, we ‘branch’ on the variable X1 and create the following two
additional subproblems.
LP2 = LP1 + X1 £ 2
LP3 = LP1 + X1 ³ 3 LP1
X1 = 2,5 X2 = 1,5
ü We say that subproblem 2 and 3 Z = 10,5
were created by branching on X1. X1 £ 2 X1 ³ 3

LP2 LP3
The Branch-and-Bound method - continued
ü The feasible regions for subproblem 2 and subproblem 3 have no points in
common.
ü We choose any subproblem
that has not yet been solved
as an LP. We arbitrarily choose
B (1)
E
to solve subproblem 2. (3)
C
(2)
ü We get the optimal solution A

to subproblem 2 (point E) O F D
X1= 2, X2= 5/3 and Z = 28/3.
The Branch-and-Bound method - continued
ü Since the optimal solution to subproblem 2 did not yield an all-integer

solution, we choose to use subproblem 2 to create two new subproblems.

ü We choose a fractional-valued variable in the optimal solution to

subproblem 2 and then branch on that variable.

ü Since X2 is the only fractional variable in the optimal solution to

subproblem 2, we branch on X2.

The Branch-and-Bound method - continued
ü We partition the feasible region for subproblem 2 into those points having
LP1
X2 £ 1 and X2 ³ 2. This creates X1 = 2,5 X2 = 1,5
Z = 10,5
the following two subproblems: X1 £ 2 X1 ³ 3

LP4 = LP2 + X2 £ 1 LP2

X1 = 2, X2 = 5/3 LP3
LP5 = LP2 + X2 ³ 2 Z = 28/3

X2 £ 1 X2 ³ 2

LP4 LP5
The Branch-and-Bound method - continued
ü LIFO (last-in-first-out) rule: we choose to solve the most recently created
subproblem.
ü The LIFO rule implies that
subproblems 4 or 5 should be
solved next. B (1)
E
(3)
ü The optimal solution to C
(2)
subproblem 4 is point G: A G

X1 = 2, X2 = 1 and Z = 8. O F D
The Branch-and-Bound method - continued
ü Since both X1 and X2 assume integer values, this solution is also feasible
LP1
for the original IP and is a candidate X1 = 2.5 X2 = 1.5
Z = 10.5
solution. Also, Z=8 constitutes a X1 £ 2 X1 ³ 3

LP2
lower bound for the optimal X1 = 2, X2 = 5/3 LP3
Z = 28/3
value of the objective X2 £ 1 X2 ³ 2
X
LP4
function. X1 = 2, X2 = 1 LP5
Z=8
The Branch-and-Bound method - continued
ü It would be useless to continue the branching from any subproblem having

an objective function value less than 8.

ü The LIFO rule implies that

subproblem 5 should be solved next.

B (1)
ü The optimal solution to (3) E C

(2)
subproblem 5 is point B: A G

X1 = 1, X2 = 3 and Z = 7. O F D
The Branch-and-Bound method - continued
ü The solution of LP5 is an integer solution but it cannot be retained as a
LP1
candidate solution because X1 = 2,5 X2 = 1,5
Z = 10,5
the lower bound 8 is not respected . X1 £ 2 X1 ³ 3

LP2 LP3
X1 = 2, X2 = 5/3
Z = 28/3
X2 £ 1 X2 ³ 2

LP4 LP5
X X1 = 2, X2 = 1 X1 = 1, X2 = 2 X
Z=8 Z=7<8
The Branch-and-Bound method - continued
ü The optimal solution to subproblem 3 is LP1
X1 = 2,5 X2 = 1,5
point D: X1=3, X2=0 and Z=9. Z = 10,5

The lower bound 8 is respected X1 £ 2 X1 ³ 3

and it constitutes a better LP2

LP3
X1 = 2, X2 = 5/3
candidate solution and Z = 28/3
X1 = 3, X2 = 0 X
Z=9
The lower bound should be X2 £ 1 X2 ³ 2

updated to 9.
LP4 LP5
X X1 = 2, X2 = 1 X1 = 1, X2 = 2
ü Since all subproblems Z=8 Z=7<8 X
have been solved the last candidate solution constitute the IP optimal solution.
The Branch-and-Bound method - continued
ü It is unnecessary to branch on a subproblem in the following three
situations :

• The subproblem is infeasible;

• The subproblem yields a solution in which all variables have integer

values;

• The optimal z-value for the subproblem does not exceed (in a max
problem) the current LB
The Branch-and-Bound method - continued
ü To solve a mixed IP by branch-and-bound, we branch only on variables that
are required to be integers.

ü Also, for a solution to a subproblem to be a candidate solution, it needs only

assign integer values to those variables that are required to be integers.
Other Exact Methods for Solving IP
ü Cutting Plane • Advantage of exact methods is to

ü Branch and Cut identify the optimal solution

ü Branch and Price • The drawback of exact methods is

the computational inefficiency
ü Dynamic Programming
Non-Exact Methods for Solving IP
• Tradeoff between computational
ü Heuristics Methods efficiency and quality of the solution
ü Meta-Heuristics • The performance of an approximate
ü Hyper-Heuristics solution method is assessed through
the computational time and the
neighborhood to the optimal solution.
Solvers for IP
ü CPLEX

ü Lingo

ü Gams

ü Gurobi