Graph Traversals
Slides by Carl Kingsford
Feb. 1, 2013
Based on/Reading: Chapter 3 of Kleinberg & Tardos
Breadth-First Search
Breadth-first search explores the nodes of a graph in increasing
distance away from some starting vertex s.
It decomposes the component into layers Li such that the shortest
path from s to each of nodes in Li is of length i.
Breadth-First Search:
1. L0 is the set {s}.
2. Given layers L0 , L1 , . . . , Lj , then Lj+1 is the set of nodes that
are not in a previous layer and that have an edge to some
node in layer Lj .
BFS Tree
1 2 3 4 5
6 7 8 9
10 11 12 13 14 15 16
BFS Tree Example
A BFS traversal of a graph results in a breadth-first search tree:
2
1 3
s 2 3
1
3
BFS Tree Example
A BFS traversal of a graph results in a breadth-first search tree:
2
1 3
s 2 3
1
3
Can we say anything about the non-tree edges?
Example BFS
8:-1 3:-1
12:-1 6:-1
14:-1 9:-1
2:-1 7:-1 1:-1
4:-1
0:0
5:-1
10:-1 13:-1 11:-1
Example BFS
8:-1 3:-1
12:-1 6:-1
14:-1 9:-1
2:-1 7:1 1:1
4:-1
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:-1 6:-1
14:-1 9:-1
2:2 7:1 1:1
4:-1
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:-1 6:-1
14:-1 9:-1
2:2 7:1 1:1
4:-1
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:-1 6:2
14:-1 9:2
2:2 7:1 1:1
4:-1
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:-1 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:-1 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:-1 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:-1 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:3 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:3 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:3 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:3 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Example BFS
8:3 3:2
12:3 6:2
14:2 9:2
2:2 7:1 1:1
4:2
0:0
5:1
10:1 13:1 11:1
Depth-First Search
DFS keeps walking down a path
until it is forced to backtrack. s
It backtracks until it finds a
new path to go down.
1
Think: Solving a maze. 2
It results in a search tree, called
the depth-first search tree. 3
4
In general, the DFS tree will be
very different than the BFS
tree.
5
7 6
Depth-First Search
2
1 3
s 5 4
7
6
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:-1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:-1:-1
8:-1:-1
7:-1:-1 12:-1:-1
1:-1:-1 2:-1:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:-1:-1
8:-1:-1
7:-1:-1 12:-1:-1
1:-1:-1 2:-1:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:2:-1
8:-1:-1
7:-1:-1 12:-1:-1
1:-1:-1 2:-1:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:2:-1
8:-1:-1
7:-1:-1 12:3:-1
1:-1:-1 2:-1:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:2:-1
8:-1:-1
7:-1:-1 12:3:-1
1:-1:-1 2:4:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:2:-1
8:-1:-1
7:-1:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:-1:-1 13:-1:-1
5:2:-1
8:-1:-1
7:6:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:7:-1 13:-1:-1
5:2:-1
8:-1:-1
7:6:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:-1:-1 14:1:-1 9:-1:-1
6:7:-1 13:8:-1
5:2:-1
8:-1:-1
7:6:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:-1:-1 3:-1:-1 4:-1:-1
11:9:-1 14:1:-1 9:-1:-1
6:7:-1 13:8:-1
5:2:-1
8:-1:-1
7:6:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:10:-1 3:-1:-1 4:-1:-1
11:9:-1 14:1:-1 9:-1:-1
6:7:-1 13:8:-1
5:2:-1
8:-1:-1
7:6:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:10:11 3:-1:-1 4:-1:-1
11:9:12 14:1:-1 9:-1:-1
6:7:14 13:8:13
5:2:-1
8:15:-1
7:6:-1 12:3:-1
1:5:-1 2:4:-1
Example DFS
0:0:-1
10:10:11 3:-1:-1 4:23:-1
11:9:12 14:1:22 9:-1:-1
6:7:14 13:8:13
5:2:21
8:15:16
7:6:17 12:3:20
1:5:18 2:4:19
Example DFS
0:0:-1
10:10:11 3:24:-1 4:23:-1
11:9:12 14:1:22 9:-1:-1
6:7:14 13:8:13
5:2:21
8:15:16
7:6:17 12:3:20
1:5:18 2:4:19
Example DFS
0:0:-1
10:10:11 3:24:-1 4:23:-1
11:9:12 14:1:22 9:25:-1
6:7:14 13:8:13
5:2:21
8:15:16
7:6:17 12:3:20
1:5:18 2:4:19
Example DFS
0:0:29
10:10:11 3:24:27 4:23:28
11:9:12 14:1:22 9:25:26
6:7:14 13:8:13
5:2:21
8:15:16
7:6:17 12:3:20
1:5:18 2:4:19
General Tree Growing (following Gross & Yellen)
We can think of BFS and DFS (and several other algorithms) as
special cases of tree growing:
I Let T be the current tree T , and
I Maintain a list of frontier edges: the set of edges of G that
have one endpoint in T and one endpoint not in T :
I Repeatedly choose a frontier edge (somehow) and add it to T .
Tree Growing
TreeGrowing(graph G, vertex v, func nextEdge):
T = (v,)
S = set of edges incident to v
While S is not empty:
e = nextEdge(G, S)
T = T + e // add edge e to T
S = updateFrontier(G, S, e)
return T
I The function nextEdge(G, S) returns a frontier edge from S.
I updateFrontier(G, S, e) returns the new frontier after we
add edge e to T.
Tree Growing
These algorithms are all special cases / variants of Tree Growing,
with different versions of nextEdge:
1. Depth-first search
2. Breadth-first search
3. Prims minimum spanning tree algorithm
4. Dijkstras shortest path
5. A*
BFS & DFS as Tree Growing
Whats nextEdge for DFS?
Whats nextEdge for BFS?
BFS & DFS as Tree Growing
Whats nextEdge for DFS?
Select a frontier edge whose tree endpoint was discovered
most recently.
Why? We can use a stack to implement DFS.
Runtime: O(|Edges|)
Whats nextEdge for BFS?
BFS & DFS as Tree Growing
Whats nextEdge for DFS?
Select a frontier edge whose tree endpoint was discovered
most recently.
Why? We can use a stack to implement DFS.
Runtime: O(|Edges|)
Whats nextEdge for BFS?
Select a frontier edge whose tree endpoint was discovered earliest.
Why? We can use a queue to implement BFS.
Runtime: O(|Edges|)
Prims Algorithm
Prims Algorithm: Run TreeGrowing starting with any root
node, adding the frontier edge with the smallest weight.
Theorem. Prims algorithm produces a minimum spanning tree.
v
e
r u
S = set of nodes already in
the tree when e is added
Implementations of BFS and DFS
BFS implementation
procedure bfs(G, s):
Q := queue containing only s
while Q not empty
v := Q.front(); Q.remove front()
for w G.neighbors(v):
if w not seen:
mark w seen
Q.enqueue(w)
Recursive implementation of DFS
procedure dfs(G, u):
while u has an unvisited neighbor in G
v := an unvisited neighbor of u
mark v visited
dfs(G, v)
Stack-based implementation of DFS
procedure dfs(G, s):
S := stack containing only s
while S not empty
v := S.pop()
if v not visited:
mark v visited
for w G.neighbors(v): S.push(w)
Properties of BFS and DFS
Property of Non-BFS-Tree Edges
Theorem. Choose x Li and y Lj such that {x, y } is an edge
in undirected graph G . Then i and j differ by at most 1.
In other words, edges of G that do not appear in the tree connect
nodes either in the same layer or adjacent layer.
Proof. Suppose not, and that i < j 1.
All the neighbors of x will be found by layer i + 1.
Therefore, the layer of y is less than i + 1, so j i + 1, which
contradicts i < j 1.
A property of Non-DFS-Tree Edges
Theorem. Let x and y be nodes in the DFS tree TG such that
{x, y } is an edge in undirected graph G . Then one of x or y is an
ancestor of the other in TG .
Proof. Suppose, wlog, x is reached first in the DFS.
All the nodes that are marked explored between first encountering
x and leaving x for the last time are descendants of x in TG .
When we reach x, node y must not yet have been explored.
It must become explored before leaving x for the last time
(otherwise, we should add {x, y } to TG ). Hence, y is a descendent
of x in TG .
