EE365: The Bellman-Ford Algorithm

1
Shortest path problems

I given weighted graph and a destination vertex

I find lowest cost path from every vertex to destination

18 4

2 12 6
10

10
5 3 5
3
1 8
1 4
5
7 7

2
Dynamic programming principle

I let gij = cost of edge i → j (∞ if no edge)

I let vi = cost of shortest path from i to destination; it must satisfy

vi = min(gij + vj )
j

18 4

2 12 6
10

10
5 3 5
3
1 8
1 4
5
7 7

3
Dynamic programming principle
6
18 4

19 2 12 6
10
vi = min(gij + vj ) 10
j 5 9 3 5 0
3
1 8
12 1 4
5
7 7 8
I starting at vertex i 6
13
I gij is cost of next step

I shortest path minimizes sum of

I cost for next step

I shortest path from where you land

4
Dynamic programming principle
6
18 4

19 2 12 6
10
vi = min(gij + vj ) 10
j 5 9 3 5 0
3
1 8
12 1 4
5
7 7 8

6
13

I once we know v, we also know the optimal path from all initial vertices

I from vertex i, move to the minimizer j

5
Bellman-Ford algorithm

(
0 if i = destination
I let vi0 =
∞ otherwise

I for k = 0, . . . , n − 1

k+1
I vi = min{vik , min(gij + vjk )}
j

I vik is lowest cost path from i to destination in k steps or fewer

I if v n 6= v n−1 then graph has negative cycle, and cost may be made −∞

I stop early if v k+1 = v k

I n vertices, m edges, runs in O(mn) time

6
Bellman-Ford algorithm
6
18 4

19 2 12 6
10

10
5 9 3 5 0
vik+1 = min{vik , min(gij + vjk )}
j 3
1 8
12 1 4
5
7 7 8

6
13

      
∞ ∞ 13 12
∞ ∞ 20 19
       
∞ 10 9 9
v0 =  v1 =  v2 =  v3 = 
       
∞ 6 6 6
   
0 0 0 0
       
∞ ∞ 14 13
∞ 8 8 8
7
Dynamic programming

I breaks up large problem into nested subproblems

I works backward in time (for deterministic problems, can also work forwards)

I stores the solution of subproblems in the value function, to allow reuse at

many states

8
Shortest path problems

I Dijkstra’s algorithm is similar but faster (O(m + n log n)), and requires non-
negative weights

I both BF and Dijkstra give shortest path from every vertex to destination

I other algorithms, such as A? , find shortest path between two vertices