Traveling Salesman Problem (TSP)

Traveling Salesman Problem (TSP) deals with finding a minimum cost (or distance
or time) tour for a salesman visiting n cities where
 each city is visited exactly once
 and the salesman returns to the starting city

Tour: A tour is a complete route through the n cities where no city is visited more
than once
Subtour: A subtour is a round trip that does not pass through all cities

Network representation
of the TSP

1
 Traveling Salesman Problem (TSP) (Contd…)
Integer programming formulation of TSP
Suppose TSP consists of n cities i = j = 1,2,…, n.
cij = distance (time or cost) from city i to city j, for i  j
cij = M (a large positive number), for i  j
1 if the salesman goes from city i to city j
xij  
n n 0 otherwise
Min Z   cij xij
i 1 j 1

n
s.t. x
i 1
ij 1 for j  1,2, , n (the salesman must arrive at each city once)

x
j 1
ij 1 for i  1,2,, n (the salesman must leave each city once)

ui  u j  nxij  n  1 for i  j; i  2,3,, n; j  2,3,, n (subtour elimination constraint)

xij {0,1} for i  1,2,, n; j  1,2,, n

uj  0 for j  1, 2, ,n 2
 Traveling Salesman Problem (TSP) (Contd…)
Cycle Total Distance (in miles)
1. 1-2-3-4-5-1 79
3
2. 1-2-3-5-4-1 108
29
3. 1-2-4-3-5-1 80 11
4. 1-2-4-5-3-1 105 30
20
2
5. 1-2-5-3-4-1 78
6. 1-2-5-4-3-1 74 8 4
7. 1-3-4-2-5-1 67 => Shortest Tour
13
8. 1-3-2-4-5-1 97 22
9. 1-3-5-2-4-1 96 20

10. 1-3-2-5-4-1 95 16

11. 1-4-2-3-5-1 101 1

12. 1-4-3-2-5-1 70 6 5

A problem with n nodes (locations) has (n-1)! possible tours.

For symmetric problems, this number is divided by 2. 3
 Traveling Salesman Problem (TSP) (Contd…)

• Here, there are 5 nodes, n = 5, therefore there are (5-1)! / 2= 4! / 2

= 12 possible tour for this problem.

• The optimal tour is 1-3-4-2-5-1 of length 67 Miles.

• Of course, This trial approach can only be used for small problems.

• For example, the number of possible tours is 181440 for a problem

with 10 nodes.

• For a problem with 15 nodes, it is 43589145600.

4
 Branch and Bound approach for TSP

Observations for developing the solution procedure:

 Deletion of subtour elimination constraint from TSP results

into a Assignment Problem (Subproblem)

 The subproblem (assignment problem) can be solved using

Hungarian Algorithm.

 If the optimal solution to the assignment problem is feasible

to the TSP (i.e., if the optimal solution of the assignment
problem yields a tour), then it is also optimal for the TSP.

5
 Branch and Bound approach for TSP
Solution Procedure
Step 0: Ignore the subtour elimination constraint from TSP. The resulting problem is an
assignment problem.
Step 1: Solve the associated assignment problem using Hungarian Algorithm.
Step 2: If the optimal solution to the assignment problem provides a tour (i.e. no subtour), then it
is the optimal solution to the TSP and STOP. OTHERWISE (the optimal solution to the
assignment problem yields subtours), go to Step 3 (restrictions are imposed to remove the
subtours).
Step 3: Select a subtour with the smallest number of cities (because it creates the smallest number
of subproblems). Let k be the number of cities in the selected subtour (i1 - i2 -…- ik - i1).
Step 4: Branch into k subproblems (branches). Then for the subproblems P1, P2,…,Pk , set
c i1i2  M
c i2i3  M

c ik i1  M, respectively
Step 5: Solve one of the subproblem. If the solution of the resulting assignment problem provides
a tour it provides an upper bound (minimum tour length). Otherwise, Go to Step 3.
Step 6: Continue the process until all the unexplored subproblems have been fathomed (better
upper bound or subproblem cannot produce a better solution).
Step 7: The optimum tour is the one associated with the best upper bound.
6
 Branch and Bound approach for TSP

Example Problem: 5 City with the following distance matrix

City 1 City 2 City 3 City 4 City 5

City 1 0 10 3 6 9

City 2 5 0 5 4 2

City 3 4 9 0 7 8

City 4 7 1 3 0 4

City 5 3 2 6 5 0

7
 Ignoring the subtour elimination constraint from TSP: Solve the
resulting Assignment Problem using Hungarian Algorithm

Subproblem P0

City 1 City 2 City 3 City 4 City 5

City 1 M 10 3 6 9
City 2 5 M 5 4 2
City 3 4 9 M 7 8
City 4 7 1 3 M 4
City 5 3 2 6 5 M

Z* = Bound= M

8
 Row Reduction
City 1 City 2 City 3 City 4 City 5
City 1 M 7 0 3 6
City 2 3 M 3 2 0
City 3 0 5 M 3 4
City 4 6 0 2 M 3
City 5 1 0 4 3 M

Column Reduction
City 1 City 2 City 3 City 4 City 5
City 1 M 7 0 1 6
City 2 3 M 3 0 0
City 3 0 5 M 1 4
City 4 6 0 2 M 3
City 5 1 0 4 1 M
9
Optimal Solution: x13  x25  x31  x42  x54  1, all other xij = 0, ZA = 15.

This solution has two subtours: 1 – 3 – 1 and 2 – 5 – 4–2. So, this is not optimal
solution to the TSP. 10
Z* = Bound = M P0: ZA = 15
(1-3-1) (2-5-4-2)
x13 = 0 x31 = 0

P1:

Branch on subproblem P0:

Subproblem P1: Subproblem P0 with constraint x13 = 0, i.e. Set c13 = M
Subproblem P1
City 1 City 2 City 3 City 4 City 5
City 1 M 10 M 6 9
City 2 5 M 5 4 2
City 3 4 9 M 7 8
City 4 7 1 3 M 4
City 5 3 2 6 5 M
11
Optimal Solution: x14  x25  x31  x43  x52  1, all other xij = 0, ZA = 17.
This solution contains two subtours: 1 – 4 – 3 – 1 and 2 – 5 – 2. This is not optimal
solution to the TSP.
Z* = Bound= M
12
 P0: ZA = 15
(1-3-1) (2-5-4-2)
x13 = 0 x31 = 0

P1: ZA = 17
(2-5-2)(1-4-3-1)
x25 = 0 x52= 0

P2:
Branch on Subproblem P1:
Subproblem P2: Subproblem P1 with constraint x25 =0, i.e. Set c25 = M)
Subproblem P2 City 1 City 2 City 3 City 4 City 5
City 1 M 10 M 6 9
City 2 5 M 5 4 M
City 3 4 9 M 7 8
City 4 7 1 3 M 4
City 5 3 2 6 5 M
13
Optimal Solution: x14  x23  x31  x45  x52  1, all other xij = 0, ZA = 21
This solution gives a tour: 1 – 4 – 5 – 2 – 3 – 1. So, this is a feasible solution to the
TSP. Fathomed (feasible solution and update the bound)
Z* = Bound= 21
14
Z* = Bound= 21 P0: ZA = 15
(1-3-1) (2-5-4-2)
x13 = 0 x31 = 0

P1: ZA = 17
(2-5-2)(1-4-3-1)
x25 = 0 x52= 0

P2: ZA = 21 P3:
(1-4-5-2-3-1)
Fathomed
Subproblem P3: Subproblem P1 with constraint x52 =0, i.e. Set c52 = M)
City 1 City 2 City 3 City 4 City 5
City 1 M 10 M 6 9
Subproblem P3
City 2 5 M 5 4 2
City 3 4 9 M 7 8
City 4 7 1 3 M 4
City 5 3 M 6 5 M
15
 Optimal Solution: x14  x25  x31  x42  x53  1, all other xij = 0, ZA = 19
This solution gives a tour: 1 – 4 – 2 – 5 –3 – 1. So, this is a feasible solution to the TSP.
Fathomed (Update bound since this is better solution) Z* = Bound= 19 16
 P0: ZA = 15
Z* = Bound= 19
(1-3-1) (2-5-4-2)
x13 = 0 x31 = 0

P1: ZA = 17 P4:
(2-5-2)(1-4-3-1)
x25 = 0 x52= 0

P2: ZA = 21 P3: ZA = 19
(1-4-5-2-3-1) (1-4-2-5-3-1)
Fathomed Fathomed

Subproblem P4: Subproblem P0 with constraint x31 = 0 i.e. Set c31 = M

Subproblem P4:

17
Optimal Solution: x13  x25  x34  x42  x51  1, all other xij = 0, Z = 16.
This solution gives a tour: 1 – 3 – 4 – 2 –5 – 1. So, this is a feasible solution to the TSP.
Fathomed (Update bound since this is better solution)
Z* = Bound= 16
18
 Optimal
Z* = 16 solution to
P0: ZA = 15
(1-3-1) (2-5-4-2) the TSP

x13 = 0 x31 = 0

P1: ZA = 17 P4: ZA = 16
(2-5-2)(1-4-3-1) (1-3-4-2-5-1)
Fathomed
x25 = 0 x52= 0

P2: ZA = 21 P3: ZA = 19
(1-4-5-2-3-1) (1-4-2-5-3-1)
Fathomed Fathomed

Optimal Sequence of visit by Salesman: 1-3-4-2-5-1,

and Total distance: 16
19