Integer programming

1.1 introduction
Optimization consists linear and non linear programming problems. Under linear
programming problem , the decision variables may take continuous or discrete values.
Integer programming problem is when the decision variables take discrete value.
Integer programming (IP) is an LP in which some or all values of decision variables are
integer ( discrete).
There are three types of integer programming problems.
1) Pure integer programming
An IP in which all the values of decision variable must be integer.
Optimize Z = Cx
s.t Ax = B, x  
Ex maximize Z= 3x + 4y
S.t 2x + y  12
x- y  8 x ,y  
 Integer programming
2. Mixed integer programming problem
Optimize Z= C xi
s.t A xi = B, i = 1,2,3,……n xi   for i = 1,2,3,….m (m < n)
Ex maximize Z= 3x + 4y
S.t 2x + y  12
x- y  8 x  
3. Binary integer programming problem
Optimize Z= Cx
s.t Ax = B x ,y  {0,1}
Ex maximize Z= 3x + 4y
S.t 2x + 10y  12
x- y  8 x ,y  {0,1}
Note that 1 and 2 can also have binary solution.
Challenge: compare and contrast the three types of IP above.
 Integer programming

• Identify the following as pure, mixed, or binary

integer programming problem.
1.A lorry is giving transportation service for goods
and goats from Addis Ababa to Dilla. A total weight
that a lorry can be loaded is 600kg a fare for a unit
of good is birr 5 and birr 15 for a goat. A space
available on the lorry is 200m2. A unit of package of
good needs 5m2 and a goat needs 3m2. A unit of
package of goods is 12 kg and each goat weigh at
least 13kg.What number of goods and goats can be
loaded in order to get max transportation income?
 Integer programming
2. A truck company requires the following number of drivers for its
trucks during 24 hours
Tine interval Number of drivers required

00-04 5

04-08 10

08-12 20

12-16 12

16-20 22

20-24 8

According to the shift schedule a driver may join for duty at mid
night, 04, 08, 12, 16, and 20hrs and work continuously for 8hrs.
Formulate the problem as LP model for optimal shift plan.
 Integer programming
3. Five projects are launched by Du staff for being evaluated over
3 years planning horizon. The table below is with expected
returns for each project and the associated yearly
expenditures. Which projects should be selected over the 3
year horizon?
Expenditure in thousands of $ per year Returns in
Year 1 2 3 thousands of $

Project

1 5 1 8 20

2 4 7 10 40

3 3 9 2 20

4 7 4 1 15

5 8 6 10 30

Available funds 25 25 25 -
thousands of $
 METHODS OF SOLVING IP
BRANCH AND BOUND METHOD
Procedure
• Solve LPPR
• Find the upper and lower bound of objective
function
• Branch the value of a basic variable if the value
is not integer
• Check the feasibility of each branch
• Solve each branch by simplex for feasible
branches by LPPR approach
 Some terms in branch and bound method

Relaxation: Assuming the values of all the decision variables are not integer.
Branch: If the optimal solution of LPPR is not integer, we partition the original
problem in to some other smaller IP problems which is known as branch.

Bound: For each branch, we compute the maximum and minimum values of the
objective function which is the lower or upper bound of the optimal value of the
original problem. If the problem is maximization the optimal value of IP is less than
or equal to the upper bound whereas the optimal value of IP is greater than or
equal to the lower bound, if the problem is minimization.

Fathoming: Stopping further branching either because of infeasibility or bound.

Incumbent solution: An integer solution found from a certain branch with
in the calculated bound of LPPR until we finish or fathom other branches.
 Branch and Bound method
• Construct new LPP for each branch
• Solve each LPPR separately
• If you arrive at integer solution, no need of branching.
Example Max Z=5x1 + 2x2
Subject to 3x1 + x2  12
x1 + x2  5 , x1, x2  0 x1, x2 integer
LPPR of this problem is max Z=5x1 + 2x2
Subject to 3x1 + x2  12
x1 + x2  5 , x1, x2  0
Solving this by simplex method
Cj 5 2 0 0
CBV BV x1 x2 S1 S2 RHS R

0 S1 3 1 0 0 12 4
0 S2 1 1 0 1 5 5
Zj 0 0 0 0 0
Cj-Zj 5 2 0 0
 Branch and bound method

Cj 5 2 0 0
cbv bv X1 X2 S1 S2 RHS R
5 X1 1 1/3 1/3 0 4 12
0 S2 0 2/3 -1/3 1 1 3/2
Zj 5 5/3 5/3 0 20
Cj-Zj 0 1/3 -5/3 0
5 X1 1 0 1/2 -1/2 7/2
2 X2 0 1 -1/2 3/2 3/2
Zj 5 2 3/2 1/2 41/2
Cj-Zj 0 0 -3/2 -1/2
 Branch and bound

• Since Cj-Zj  0, it satisfies the optimality condition.

• The solution is x1= 7/2 and x2= 3/2 Z max = 41/2
• The upper bound of Z = 41/2 and its lower bound is Z=17 at x1= 3 and x2 = 1
Now the branches from x1 are x1   and x1   while
Branches from x2 are x2   and x2  2 now we proceed to construct sub LPP
Sub LPP of x1 max Z= 5 x1 + 2 x2 and max Z= 5x1 + 2x2
subject to 3 x1 + x2  12 subject to 3 x1 + x2  12
x1 + x 2  5 x 1 + x2  5
x1   x1 
Sub LPP of x2 max Z= 5 x1 + 2 x2 and max Z= 5 x1 + 2 x2
subject to 3 x1 + x2  12 subject to 3 x1 + x2  12
x1 + x 2  5 x 1 + x2  5
x2  1 x2  2
Solve each of these branches by checking infeasibility, feasibility, optimality and the bounded
cases.
the solution is x1 = 4 x2= 0 and Z max =20
 Cutting Plane method
Let us use the above example to discus about this method.
Example Max Z=5x1 + 2x2
Subject to 3x1 + x2  12
x1 + x2  5 , x1, x2  0 x1, x2 integer
BY using dual simplex method

bv x1 x2 S1 S2 RHS
S1 3 1 1 0 12
S2 1 1 0 1 5
Z -5 -2 0 0 0
x1 1 1/3 1/3 0 4
S2 0 2/3 -1/3 1 1
Z 0 -1/3 5/3 0 20
x1 1 0 1/2 -1/2 7/2
x2 0 1 -1/2 3/2 3/2
Z 0 0 3/2 1/2 41/2
 Cutting plane method

Solution is x1 = 7/2 = 3+ 1/2 x2 = 3/2 = 1+1/2 and Z= 41/2 = 20+1/2

We choose the equation of a row with max fractional value of x1 and x2 .
max { ½, ½ } = ½ we choose x1 row randomly since both fractions are equal.
The equation is x1 + ½ S1 -1/2 S2 = 7/2
by rewriting x1 + ½ S1 +(-1+1/2 S2) = 3 + ½ we use this equation to get Gomory cut constraint.
1/2 S1 +½ S2  ½ (by the Gomory cut algorithm)
-1/2 S1 – ½ S2  -1/2
the equation form is -1/2 S1 – ½ S2 + S3 = -1/2
Now we have new LPP with additional constraint.

Max Z= 5x1 + 2x2

Subject to 3x1 + x2  12
x1 + x2  5 ,
-1/2 S1 -1/2 S2  -1/2 x1, x2 S1, S2,  0 x1, x2 integer
 Cutting Plane method

basis x1 x2 S1 S2 S3 RHS
x1 1 0 ½ -1/2 0 7/2
x2 0 1 -1/2 3/2 0 3/2
S3 0 0 -1/2 -1/2 1 -1/2
Z 0 0 3/2 ½ 0 41/2
x1 1 0 1 0 -1 4
x2 0 1 -2 0 3 0
S2 0 0 1 1 -2 1
Z 0 0 1 0 1 20

We select the min of {3/21/2, ½  1/2 } =min {3, 1}=1

i.e S2 is key column and S3 is key row in the above table.