CS213/293 Data Structure and Algorithms 2023

Lecture 17: Graphs - Depth-first search

Instructor: Ashutosh Gupta

IITB India

Compile date: 2023-10-23

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 1
 Topic 17.1

Depth-first search (DFS)

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 2
Let us solve the maze again

Breadth-first search is about considering all available options before exploring further.

We can have another strategy of search: explore a choice fully before considering another choice.

entry d g

f a b e exit

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 3
 Algorithm: DFS for search

Algorithm 17.1: DFS( Graph G = (V , E ), vertex r , Value x )

1 Stack S;
2 set visited;
3 S.push(r );
4 while not S.empty () do
5 v := S.pop();
6 if v .label == x then
7 return v
8 if v ∈
/ visited then
9 visited := visited ∪ {v };
10 for w ∈ G .adjacent(v ) do
11 S.push(w )

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 4
Example: DFS
Initially: S = [entry ]

After visiting entry: S = [a]

After visiting a: S = [f , b] entry d g

After visiting f: S = [b]

After visiting b: S = [d, e]

a b e exit
After visiting d: S = [g , e]

After visiting g: S = [e]

After visiting e: S = [exit, b] f

After visiting exit: Node is found

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 5
Algorithm: Recursive DFS
Recursive description of DFS is easier to follow.
Algorithm 17.2: DFS( graph G = (V , E ), vertex v )
1 for v ∈ V do
2 v .visited := False
3 DFSRec(G , v )
v can be in three possible states
Algorithm 17.3: DFSRec( Graph G , vertex v ) ▶ v is not visited
1 v .visited := True; ▶ v is on call stack
2 for w ∈ G .adjacent(v ) do
▶ v is visited and not on call stack
3 if w .visited == False then
4 DFSRec(w )

Exercise 17.1
Why there is no stack in the recursive DFS?
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 6
DFS Tree
Algorithm 17.4: DFS( graph G = (V , E ), vertex v )
1 global time := 0;
2 for v ∈ V do
3 v .visited := False
4 DFSRec(G , v )

Algorithm 17.5: DFSRec( Graph G , vertex v )

1 v .visited := True;
2 v .arrival := time + +;
3 for w ∈ G .adjacent(v ) do
4 if w .visited == False then
5 w .parent := v ;
6 DFSRec(w )

7 v .departure := time + +;
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 7
Example: recursive execution
Green numbers are arrival times and blue numbers are the departure times.

DFSRec(G ,entry ) 0
entry 5
g 6
d 7
DFSRec(G ,a)
DFSRec(G ,f )
Exit DFSRec(G ,f )
1 4 9
a b e exit
DFSRec(G ,b) 8

DFSRec(G ,d) In the extra edge,

DFSRec(G ,g ) b is ancestor of e
2
Exit DFSRec(G ,g ) 3 f
DFSRec(G ,e)
DFSRec(G ,exit)
Exercise 17.2
What are the exit timings of the remaining nodes?
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 8
Non-tree edges
A run of DFS induces a tree. There are two kinds of possible extra edges.

Exercise 17.3
a. Is blue edge possible? yes.
b. Is red edge possible? no.
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 9
Reachable coverage
Theorem 17.1
Let G = (V , E ) be a connected graph. For each {v1 , v2 } ∈ E , v1 is ancestor of v2 in DFS tree or
vice versa.

Proof.
Without loss of generality, we assume v1 .arrival < v2 .arrival at the end of DFS.

During the run of DFSRec(G , v1 ), v2 will be visited in one of the following two ways.
1. DFSRec(G , v2 ) is called by DFSRec(G , v1 ). v1 is parent of v2 .
2. DFSRec(G , v2 ) has been called already, when the loop in DFSRec(G , v1 ) reaches to v2 .
In either case, v2 .arrival < v1 .departure.

v1 is ancestor of v2 in the DFS tree. (Why?)

In case 1, we call {v1 , v2 } a tree edge. In case 2, we call {v1 , v2 } a back edge.
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 10
Running time of DFS

Theorem 17.2
The running time of DFS is O(|E | + |V |).

Proof.
The total number of recursive calls and iterations of initializations is O(|V |).

In call DFSRec(G , v ), the loop iteration is bounded by degree(v ).

Therefore, the total number of iterations are O(|E |).

Therefore, the running time is O(|E | + |V |).

Exercise 17.4
Prove that the running time besides initialization is O(|E |).
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 11
 Topic 17.2

Depth-first search on directed graph

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 12
Example : DFS on the directed graph
Consider the following directed graph

0 5 6
15
entry 12 d g 7

2 9
3 f a b e exit 10
1 14 4 13 8 11

Now we have three kind of extra-edges.

▶ Forward edge:(b, e), where b.departure > e.departure > e.arrival > b.arrival
▶ Back edge:(f , entry ), where entry .departure > f .departure > f .arrival > entry .arrival
▶ Cross edge:(e, g ), where e.departure > e.arrival > g .departure > g .arrival
There is no other possibility.
cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 13
 Topic 17.3

Topological sort

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 14
Algorithm: Topological sort using DFS

Algorithm 17.6: TopologicalSort( Directed graph G = (V , E ) )

1 Stack Sorted;
2 while ∃v such that v .visited == False do visit(v ) ;

Algorithm 17.7: Visit( Graph G = (V , E ), vertex v )

if v .visited then return;
if v .onPath then throw Cycle found ;
v .onPath = True;
for w ∈ G .adjacent(v ) do
Visit(w )
v .onPath = False;
v .visited = True;
Sorted.push(v );

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 15
 Topic 17.4

Problems

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 16
 Algorithm: Topological sort using DFS

Exercise 17.5
Is this variation of DFS correct? If yes, will it be more efficient or less?
Algorithm 17.8: DFSRec( Graph G , vertex v )
v .visited = True;
for w ∈ G .adjacent(v ) do
if v .visited then
4 DFSRec(w )

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 17
 End of Lecture 17

cbna CS213/293 Data Structure and Algorithms 2023 Instructor: Ashutosh Gupta IITB India 18