DFS(v) – Recursive version

Global Initialization:
for all nodes v, v.dfs# = -1 // mark v "undiscovered"
dfscounter = 0
for v = 1 to n do
if state(v) != fully-explored then
DFS(v):

DFS(v)
v.dfs# = dfscounter++ // v "discovered", number it
Mark v ``discovered”.
for each edge (v,x)
if (x.dfs# == -1) // (x previously undiscovered)
DFS(x)
else …
Mark v “fully-explored”
 Kinds of edges – DFS on
Edge (u,v) directed graphs
Tree
[u [v v] u]

Forward
[u [v v] u]

Cross
[v v] [u u]

Back
134
[v [u u] v]
 Topological Sort using DFS
Global Initialization:
for all nodes v, v.dfs# = -1 // mark v "undiscovered"
dfscounter = 0
current_label = n
for v = 1 to n do
if state(v) != fully-explored then
DFS(v):

DFS(v)
v.dfs# = dfscounter++ // v "discovered", number it
Mark v ``discovered”.
for each edge (v,x)
if (x.dfs# == -1) // (x previously undiscovered)
DFS(x)
else // add check for cycle if needed
Mark v “fully-explored”
f(v) = current_label // f(v) values give the topological order
current_label --;
 Analysis
Running time O(n+m)

Correctness: Need to show that:

if (u,v) is an edge then f(u) < f(v)
Case 1: DFS(u) called before DFS(v), so DFS(v)
finishes first, which means f(v) > f(u).
Case 2: DFS(v) called before DFS(u). But there
cannot be a directed path from v to u, so
recursive call to DFS(v) will finish before
recursive call to DFS(u) starts, so f(v) > f(u) 136
 A simple problem on trees
Given: tree T, a value L(v) defined for every
vertex v in T
Goal: find M(v), the min value of L(v)
anywhere in the subtree rooted at v
(including v itself).
How? Depth first search, using:
" L(v) if v is a leaf %
M(v) = # &
$min(L(v), min w a child of v M(w)) otherwise '

137

€
 DFS(v) – Recursive version
Global Initialization:
for all nodes v, v.dfs# = -1 // mark v "undiscovered"
dfscounter = 0 // (global variable)
DFS(s); // start DFS at node s;

DFS(v)
v.dfs# = dfscounter++ // v "discovered", number it
for each edge (v,x)
if (x.dfs# = -1) // tree edge (x previously undiscovered)
DFS(x)

138
 Application: Articulation Points
A node in an undirected graph is an
articulation point iff removing it
disconnects the graph

articulation points represent vulnerabilities in

a network – single points whose failure would
split the network into 2 or more
disconnected components

139
 Identifying key proteins on the anthrax predicted network

140
Articulation point proteins Ram Samudrala/Jason McDermott
 Articulation Points
1 articulation point

iff its removal

disconnects

2 the graph
10
3
11
12
7 8
13
6 9
4
141
5
 Articulation Points
1

2
10
3
11
12
7 8
13
6 9
4
142
5
 Simple Case: Artic. Pts in a tree
Which nodes in a rooted tree are articulation
points?

143
 Simple Case: Artic. Pts in a tree
Leaves – never articulation points
Internal nodes – always articulation points
Root – articulation point if and only if two or
more children

Non-tree: extra edges remove some

articulation points (which ones?)
144
Recall: all edges either tree edges or back
edges in DFS on undirected graph
Consider edge (u,v).
If u discovered first, then edge (u,v) will be
explored before DFS(u) completes.
If at the time it is explored v is undiscovered, the
edge will become a tree edge.
If v is already discovered, then since DFS(v) was
called after DFS(u), it completes before DFS(u)
completes,
So v is a descendent of u. 145
Recall: all edges either tree edges or back
edges in DFS on undirected graph
If u is an ancestor of v, then
dfs# of u is lower than dfs# of v

146
 Simple Case: Artic. Pts in a tree
Leaves – never articulation points
Internal nodes – always articulation points
Root – articulation point if and only if two or
more children

Non-tree: extra edges remove some

articulation points (which ones?)
147
 Articulation Points from DFS
Root node is an articulation point
iff ….
Leaf is never an articulation point
non-leaf, non-root u
node u is an y x
articulation point
⇔

148
 Articulation Points from DFS
Root node is an articulation point
iff it has more than one child
Leaf is never an articulation point
non-leaf, non-root u
node u is an y x
articulation point
⇔

If removal of u does NOT

∃ some child y of u s.t. separate x, there must be an
no non-tree edge goes exit from x's subtree. How?
above u from y or below Via back edge. 149
 Articulation Points:
the "LOW" function
Definition: LOW(v) is the lowest dfs# of any
vertex that is either in the dfs subtree rooted at v
(including v itself) or connected to a vertex in that
subtree by a back edge.

150
 LOW(v) is the lowest dfs# of any vertex that is either in
the dfs subtree rooted at v (including v itself) or connected
to a vertex in that subtree by a back edge.
A 1
2 B Vertex DFS # Low
A 1
B 2
3
C C 3
D 4
4 8 E 8
D E F 5
9 G 9
5 H 10
F G H 10
11 I 6
J J 11
6
I L 12 K 7
L 12
7 M 13
K M 13 151
 Articulation Points
A 1
2 B Vertex DFS # Low
A 1 1
B 2 1
3
C C 3 1
D 4 3
4 8 E 8 1
D E F 5 3
9 G 9 9
5 H 10 1
F G H 10
11 I 6 3
J J 11 10
6
I L 12 K 7 3
L 12 10
7 M 13 13
K M 13 152
 Articulation Points
A 1
2 B Vertex DFS # Low
A 1 1
B 2 1
3
C C 3 1
D 4 3
4 8 E 8 1
D E F 5 3
9 G 9 9
5 H 10 1
F G H 10
11 I 6 3
J J 11 10
6
I L 12 K 7 3
L 12 10
7 M 13 13
K M 13 153
 Articulation Points:
the "LOW" function
Definition: LOW(v) is the lowest dfs# of any
vertex that is either in the dfs subtree rooted at v
(including v itself) or connected to a vertex in that
subtree by a back edge.

v articulation point iff…

154
 Articulation Points:
the "LOW" function
Definition: LOW(v) is the lowest dfs# of any
vertex that is either in the dfs subtree rooted at v
(including v itself) or connected to a vertex in that
subtree by a back edge.
v (non-root) articulation point iff some
child x of v has LOW(x) ≥ dfs#(v)

155
 Articulation Points:
the "LOW" function
Definition: LOW(v) is the lowest dfs# of any
vertex that is either in the dfs subtree rooted at v
(including v itself) or connected to a vertex in that
subtree by a back edge.
v (nonroot) articulation point iff some child x of v
has LOW(x) ) ≥ dfs#(v)

LOW(v) =
min ( {dfs#(v)} ∪ {LOW(w) | w a child of v } ∪
{ dfs#(x) | {v,x} is a back edge from v } ) 156
 DFS(v) for
Finding Articulation Points
Global initialization: v.dfs# = -1 for all v.

Except for root. Why?

DFS(v)
v.dfs# = dfscounter++
v.low = v.dfs# // initialization
for each edge {v,x}
if (x.dfs# == -1) // x is undiscovered
DFS(x)
v.low = min(v.low, x.low)
if (x.low >= v.dfs#)
print "v is art. pt., separating x" Equiv: "if( {v,x}
else if (x is not v's parent) is a back edge)"

157
v.low = min(v.low, x.dfs#) Why?
 Summary
Graphs –abstract relationships among pairs of objects
Terminology – node/vertex/vertices, edges, paths, multi-
edges, self-loops, connected
Representation – edge list, adjacency matrix
Nodes vs Edges – m = O(n2), often less
BFS – Layers, queue, shortest paths, all edges go to same or
adjacent layer
DFS – recursion/stack; all edges ancestor/descendant
Algorithms – connected components, bipartiteness,
topological sort, articulation points
158