Graph Traversal
Depth-First Search
Md. Bakhtiar Hasan
Lecturer
Department of Computer Science and Engineering
Islamic University of Technology
Problem At Hand
“Exploring” a graph
MBH (CSE, IUT) Graph Traversal 2 / 13
Problem At Hand
“Exploring” a graph
s t
MBH (CSE, IUT) Graph Traversal 2 / 13
Problem At Hand
“Exploring” a graph
s t
MBH (CSE, IUT) Graph Traversal 2 / 13
Problem At Hand
“Exploring” a graph
s t
Possible Problems
Finding a way out
Exploring all possible paths
MBH (CSE, IUT) Graph Traversal 2 / 13
Solution Setup
Convert each junction to a node
c d k t
b e i m
a g h l
s f j n
MBH (CSE, IUT) Graph Traversal 3 / 13
Solution Setup
Convert each junction to a node
Create edge for connected junctions
c d k t
b e i m
a g h l
s f j n
MBH (CSE, IUT) Graph Traversal 3 / 13
Solution Idea
Start from source/start point
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
Backtrack along breadcrumbs until we reach unexplored neighbor
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
Backtrack along breadcrumbs until we reach unexplored neighbor
Explore those neighbors
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
Backtrack along breadcrumbs until we reach unexplored neighbor
Explore those neighbors
Avoid repeating junction
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
Backtrack along breadcrumbs until we reach unexplored neighbor
Explore those neighbors
Avoid repeating junction
Until we find the target/finish point
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
Backtrack along breadcrumbs until we reach unexplored neighbor
Explore those neighbors
Avoid repeating junction
Until we find the target/finish point
Depth-First Search
MBH (CSE, IUT) Graph Traversal 4 / 13
Solution Idea
Start from source/start point
Follow junctions, leaving breadcrumbs, till we reach dead-end
Backtrack along breadcrumbs until we reach unexplored neighbor
Explore those neighbors
Avoid repeating junction
Until we find the target/finish point
(Optional) Check and explore if any other junction is not explored
Depth-First Search
MBH (CSE, IUT) Graph Traversal 4 / 13
Example
Assume, adjL is sorted in alphabetical order
s a b
c d e
parent
s a b c d e
stack
MBH (CSE, IUT) Graph Traversal 5 / 13
Complexity
Traverse every vertex once → V
MBH (CSE, IUT) Graph Traversal 6 / 13
Complexity
Traverse every vertex once → V
From each vertex v , traverse adjL[v]
From the Handshaking
Lemma,
P E (Directed Graph)
adjL[v ] =
v ∈V 2E (Undirected Graph)
MBH (CSE, IUT) Graph Traversal 6 / 13
Complexity
Traverse every vertex once → V
From each vertex v , traverse adjL[v]
From the Handshaking
Lemma,
P E (Directed Graph)
adjL[v ] =
v ∈V 2E (Undirected Graph)
Time complexity: O(V + E )
MBH (CSE, IUT) Graph Traversal 6 / 13
Edge Classification
Tree Edge: Created by traversing child-parent. Used to visit new
vertex
s a b
c d e
MBH (CSE, IUT) Graph Traversal 7 / 13
Edge Classification
Tree Edge: Created by traversing child-parent. Used to visit new
vertex
Non-Tree Edge:
s a b
c d e
MBH (CSE, IUT) Graph Traversal 7 / 13
Edge Classification
Tree Edge: Created by traversing child-parent. Used to visit new
vertex
Non-Tree Edge:
• Forward Edge: Ancestor → Descendant
s a b
c d e
MBH (CSE, IUT) Graph Traversal 7 / 13
Edge Classification
Tree Edge: Created by traversing child-parent. Used to visit new
vertex
Non-Tree Edge:
• Forward Edge: Ancestor → Descendant
• Back Edge: Descendant → Ancestor
s a b
c d e
MBH (CSE, IUT) Graph Traversal 7 / 13
Edge Classification
Tree Edge: Created by traversing child-parent. Used to visit new
vertex
Non-Tree Edge:
• Forward Edge: Ancestor → Descendant
• Back Edge: Descendant → Ancestor
• Cross Edge: Between two non-ancestor related subtree
s a b
c d e
MBH (CSE, IUT) Graph Traversal 7 / 13
Edge Classification
Tree Edge: Created by traversing child-parent. Used to visit new
vertex
Non-Tree Edge:
• Forward Edge: Ancestor → Descendant
• Back Edge: Descendant → Ancestor
• Cross Edge: Between two non-ancestor related subtree
Undirected graph → Tree Edge + Back Edge
s a b
c d e
MBH (CSE, IUT) Graph Traversal 7 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
MBH (CSE, IUT) Graph Traversal 8 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
MBH (CSE, IUT) Graph Traversal 8 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a back-edge (ca).
a b c
MBH (CSE, IUT) Graph Traversal 8 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a back-edge (ca).
a b c
=⇒ c must be descendant of a
MBH (CSE, IUT) Graph Traversal 8 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a back-edge (ca).
a b c
=⇒ c must be descendant of a
=⇒ There’s a path consisting of tree-edge(s) from a to c
MBH (CSE, IUT) Graph Traversal 8 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a back-edge (ca).
a b c
=⇒ c must be descendant of a
=⇒ There’s a path consisting of tree-edge(s) from a to c
=⇒ Cycle
MBH (CSE, IUT) Graph Traversal 8 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
Let, v0 be the first vertex visited in the cycle
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
Let, v0 be the first vertex visited in the cycle
v1 will finish before finishing v0
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
Let, v0 be the first vertex visited in the cycle
v1 will finish before finishing v0
v2 will finish before finishing v1 and v0
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
Let, v0 be the first vertex visited in the cycle
v1 will finish before finishing v0
v2 will finish before finishing v1 and v0
...
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
Let, v0 be the first vertex visited in the cycle
v1 will finish before finishing v0
v2 will finish before finishing v1 and v0
...
vn will finish before finishing vn−1 , . . . , v0
MBH (CSE, IUT) Graph Traversal 9 / 13
Cycle Detection
Claim
A graph G has a cycle ⇐⇒ DFS has a back edge
Proof:
Consider that, our graph has a cycle (< v0 , v1 , . . . , vn , v0 >).
v2 v3
v1 v4
v0 vn
Let, v0 be the first vertex visited in the cycle
v1 will finish before finishing v0
v2 will finish before finishing v1 and v0
...
vn will finish before finishing vn−1 , . . . , v0
v0 is an ancestor of vn → vn , v0 is a back-edge
MBH (CSE, IUT) Graph Traversal 9 / 13
Job Scheduling
Directed Acyclic Graph (DAG)
MBH (CSE, IUT) Graph Traversal 10 / 13
Job Scheduling
Directed Acyclic Graph (DAG)
Vertices → Task
Edge → Dependency
MBH (CSE, IUT) Graph Traversal 10 / 13
Job Scheduling
Directed Acyclic Graph (DAG)
Vertices → Task
Edge → Dependency
Order task without violating the dependencies
MBH (CSE, IUT) Graph Traversal 10 / 13
Job Scheduling
Directed Acyclic Graph (DAG)
Vertices → Task
Edge → Dependency
Order task without violating the dependencies
SP-I SP-II DBMS
Algo
DS
SA DM
P&S
MBH (CSE, IUT) Graph Traversal 10 / 13
Job Scheduling
Run DFS on any vertex
MBH (CSE, IUT) Graph Traversal 11 / 13
Job Scheduling
Run DFS on any vertex
Push vertex in stack when finished processing
MBH (CSE, IUT) Graph Traversal 11 / 13
Job Scheduling
Run DFS on any vertex
Push vertex in stack when finished processing
Avoid repeating vertices
MBH (CSE, IUT) Graph Traversal 11 / 13
Job Scheduling
Run DFS on any vertex
Push vertex in stack when finished processing
Avoid repeating vertices
Continue running DFS until all vertices are visited
MBH (CSE, IUT) Graph Traversal 11 / 13
Job Scheduling
Run DFS on any vertex
Push vertex in stack when finished processing
Avoid repeating vertices
Continue running DFS until all vertices are visited
Pop elements from the stack and print → Task Ordering
MBH (CSE, IUT) Graph Traversal 11 / 13
Job Scheduling
Run DFS on any vertex
Push vertex in stack when finished processing
Avoid repeating vertices
Continue running DFS until all vertices are visited
Pop elements from the stack and print → Task Ordering
Topological Sort
MBH (CSE, IUT) Graph Traversal 11 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
• v finishes before u
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
• v finishes before u
Case 2: v starts before u
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
• v finishes before u
Case 2: v starts before u
• Problem if we visit u before finishing v
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
• v finishes before u
Case 2: v starts before u
• Problem if we visit u before finishing v
• Can only happen when (v , u) exists
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
• v finishes before u
Case 2: v starts before u
• Problem if we visit u before finishing v
• Can only happen when (v , u) exists
• Will create a back-edge → can’t exists in DAG
MBH (CSE, IUT) Graph Traversal 12 / 13
Job Scheduling
Claim
For any edge, e = (u, v ), v finishes before u in DFS.
u v
Proof:
Case 1: u starts before v
• DFS will visit v before finishing u
• v finishes before u
Case 2: v starts before u
• Problem if we visit u before finishing v
• Can only happen when (v , u) exists
• Will create a back-edge → can’t exists in DAG
• v finishes before u
MBH (CSE, IUT) Graph Traversal 12 / 13
Reading
CLRS 22.3, 22.4
Dasgupta 3.2, 3.3
Jeff Erickson 5.6, 5.7, 6.1-6.3, 6.5, 8.5
MBH (CSE, IUT) Graph Traversal 13 / 13
