Lecture 14

Edmonds-Karp Algorithm
 Edmonds-Karp Algorithm

The augmenting path is a shortest path from s to t in the

4
2 5
3 1 1 1

2 2
s 4 t
3 2
1

3
This is the original network.
Ford-Fulkerson Max Flow

4
2 5
3 1 1 1

2 2
s 4 t
3 2
1

Choose a shortest path from

4
2 5
3 1 1 1

2 2
s 4 t
3 2
1

3
This is residual graph after
the 1st augmentation.
Ford-Fulkerson Max Flow

4
2 5
3 1 1 1

2 2
s 4 t
3 2
1

3
Choose a shortest path from
s to t.
Ford-Fulkerson Max Flow

4
2 5
3 1 1 1

2 2
s 4 t
1
1 2
2
3
The residual graph after the
2nd augmentation.
Ford-Fulkerson Max Flow

4
2 5
3 1 1 1

2 2
s 4 t
1
1 2
2
3
Choose a shortest path from
s to t.
Ford-Fulkerson Max Flow

3
2 5
1
2 1 1
1 1
2 2
s 4 t
1
1 2
2
3
The residual graph after the
3rd augmentation.
 Let  f ( s, x) the shortest path distance from s to x in the residual
network G f , where each edge has unit distance.
Lemma
When Edmonds - Karp algorithm runs,  f ( s, x) increases
monotonically with each flow augmentation.
Proof Denote  f ( x)   f ( s, x).
For contradiction, suppose flow f ' is obtained from flow f
through an augmentation and  f ' (v)   f (v) for some node v.
W.l.g., assume  f ' (v) is the smallest among such v, i.e.,
 f ' (u )   f ' (v)   f ' (u )   f (u ).
 Suppose (u, v ) is on the shortest path from s to v in G f ' .
Case 1. (u , v)  G f .
 f (v)   f (u )  1   f ' (u )  1   f ' (v), ().
Case 2. (u , v)  G f . Then (v, u ) must be on augmenting path in G f .
 f (v)   f (u )  1   f ' (u )  1   f ' (v)  2   f ' (v), ().

u v
s in G f '
u v
s in G f
(u , v) is critical in G f if (u , v) has the smallest capacity in
augmenting path p in G f .
Lemma
Each (u , v) can be critical at most (n  1) / 2 times.
Proof
Suppose (u , v) is critical in G f . Then (u, v) will dispear in next
residual graph. Before (u , v) appears again, (v, u ) has to appear
in augmenting path of a residual graph G f ' . Thus, we have
 f ' (u )   f ' (v)  1.
Since  f (v)   f ' (v), we have
 f ' (u )   f ' (v)  1   f (v)  1   f (u )  2.
Theorem
Edmonds - Karp algorithm runs in time O(| V |  | E |2 ).
Proof
In every augmentati on, there exists an arc critical.
Since each arc can be critical at most (n  1) / 2 times,
there are at most O (| V |  | E |) augmentati ons.
In each augmentati on, finding the shortest path
takes O (| E |) time.
 Matching in Bipartite Graph

Given a bipartite graph G  (V ,U , E ), a matching is a subset

Maximum Matching
Given a bipartite graph G  (V , E ), find a matching with
maximum cardinality.
1

1
Note: Every edge has capacity 1.
1. Can we do augmentation directly
in bipartite graph?

2. Can we do those augmentation