Lecture 10

Kinds of graphs. Data structures for graph representation.

Connectivity: The naı́ve algorithm and Warshall algorithm

Isabela Drămnesc UVT

Computer Science Department,

West University of Timişoara,
Romania

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 1 / 25

Remember that:
Graph = mathematical structure G = (V , E ) where
V : set of nodes (or vertices)
E : set of edges incident to 2 nodes, or 1 node
Depending on the kind of edges e ∈ E , graphs are of two kinds:
I Undirected: Every edge has one or two endpoints
e
e
a b or a (loop)
Graphical representation:
I Directed (or digraphs): Every edge e ∈ E has a source (or start)
end a destination (or end)
e

a e b or a (loop)
Graphical representation:
The directed edges are called arcs.
Simple graph: graph (directed or not) with at most one arc between any
pair of nodes, and no loop.
Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 2 / 25
Common types of graphs
Taxonomy

G = (V , E )
e1 , e2 ∈ E are parallel edges if they are incident to the same nodes,
and
If G is directed, then start(e1 ) = start(e2 ) and
end(e1 ) = end(e2 )

I Multigraph (directed or not):no loops, and if the graph is

undirected: it can have parallel edges
directed: it can have parallel arcs
I Pseudograph: undirected graph that can have loops and
parallel edges.
I Weighted graph: every edge e ∈ E has a weight w (E );
usually w (e) ∈ R.

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 3 / 25

Graphical representations of graphs
Illustrative examples

I Simple graphs: draw lines or arcs between the connected

nodes
b c
Simple undirected graph: e
a d
b c
Simple directed graph: e
a d
I Simple weighted graphs: we indicate the weights along the
corresponding connections
2
9
b c
12
3

a 5 1 f
4
7

1
e d
Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 4 / 25
Graphical representations of graphs
Example (continued)

Multigraphs or pseudographs: if we want to distinguish parallel

edges, we can label them:
e2 e4
a
e3
e1 c d
multigraph:
e5
b e6
e2 e4 e9
a
e3
e1 c d
pseudograph:
e5 e7
b e6
e8

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 5 / 25

Concrete representations of graphs

1 List of nodes + list of edges

2 Adjacency lists
3 Adjacency matrix
4 Incidence matrix
5 Weight matrix

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 6 / 25

Simple graphs
The representation by list of nodes + list of edges

Example
b c List of nodes V = [a, b, c, d, e]
List of edges E = [{a, b}, {a, c}, {a, d}, {b, c}, {c, e}, {d, e}]
e Remarks: {a, b} = {b, a}, {a, c} = {c, a}, a.s.o.
a edge ↔ set of nodes adjacent to the edge
d
b c List of nodes V = [a, b, c, d, e]
List of arcs E = [(a, b), (c, a), (c, b), (d, a), (e, c), (e, d)]
e Remarks: (a, b) 6= (b, a), (a, c) 6= (c, a), a.s.o.
a edge ↔ pair (start,end)
d

Remark
If the graph has no isolated nodes (with 0 neighbors), there is no
need to store the list of nodes V :
I V can be computed from E

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 7 / 25

Simple graphs
The representation with adjacency lists

For every node u ∈ V we keep the list of nodes that are adjacent
to u
I If G is undirected, v is adjacent to u of there is an edge with
endpoints u and v .
In undirected graphs, the relation of adjacency is symmetric.
I If G is directed, v is adjacent to u if there is an arc e ∈ E
from u to v , i.e. start(e) = u and end(e) = v .

Example
b c a 7→ [b, c, d] d 7→ [a, e]
b 7→ [a, c] e 7→ [c, d]
e c 7→ [a, b, e]
a d
b c a 7→ [b] d 7→ [a]
b 7→ [] e 7→ [c, d]
e c 7→ [a, b]
a d
Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 8 / 25
Graphs
The representation with adjacency matrix AG

If G has n nodes, AG = (mij ) has dimension n × n and

mij := number of edges from the i-th node to the j-th node.

Remarks
1 Before computing M
G from G , be must fix an enumeration of
its nodes: [v1 , v2 , . . . , vn ]
2 If G is undirected, AG is a symmetric matrix
3 If G is a simple graph, AG contains just 0s and 1s

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 9 / 25

Undirected graphs
The adjacency matrix AG of an undirected graph G

The adjacency matrix the undirected graph

b c
G: e
a d
for the node enumeration [a, b, c, d, e] is
 
0 1 1 1 0
1 0 1 0 0
 
AG = 1 1 0 0 1
1 0 0 0 1
0 0 1 1 0

Remark: the matrix AG is symmetric.

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 10 / 25

Graphs
The digraph corresponding to a symmetric adjacency matrix

If A is a symmetric matrix of size n × n with aij ∈ N for all i, j, an

undirected graph G whose adjacency matrix is A can be built as
follows:
1 Draw n points v1 , . . . , vn in plane
2 For every i, j ∈ {1, . . . , n}, draw aij distinct edges between vi
and vj

Example
 
0 1 0 2 v2 v3
1 0 1 0

A= 0 ⇒G :
1 0 1
2 0 1 0 v1 v4

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 11 / 25

Digraphs
The adjacency matrix AG of a digraph G

The adjacency matrix of the digraph

b c
G: e
a d
for the enumeration [a, b, c, d, e] of nodes is
 
0 1 0 0 0
0 0 0 0 0
 
AG =  1 1 0 0 0 
1 0 0 0 0
0 0 1 1 0

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 12 / 25

Digraphs
The digraph corresponding to an adjacency matrix

If A is an n × n matrix with aij ∈ N for all i, j, a digraph G whose

adjacency matrix is A can be constructed as follows:
1 Draw n points v1 , . . . , vn in the plane
2 For every i, j ∈ {1, . . . , n}, draw aij distinct arcs from vi to vj

Example

 
0 1 0 2 v2 v3
0 0 1 0

A=
0 ⇒G :
1 1 1
1 0 0 0 v1 v4

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 13 / 25

Digraphs with labeled edges
The representation with incidence matrices

We assume given two lists (or enumerations):

V = [v1 , . . . , vn ] of the nodes of G
L = [e1 , . . . , ep ] of the labels of edges of G
The incidence
 matrix MG = (mij ) has dimension n × p and
 −1 if start(ej ) = vi
mij = 1 if end(ej ) = vi
0 otherwise.


Example
If V = [a, b, c, d, e], L = [e1 , e2 , e3 , e4 , e5 , e6 , e7 , e8 ] and
−1 0 0 −1 1 0 0 −1
 b
e2
1 −1 0 0 0 1 0 0 e1
MG =  −1 −1 ⇒ e4
 0 1 1 0 0 0

a c

e6
e5
0 0 1 0 0 0 1 1 e8
−1 −1

e3
0 0 0 0 0 0
e e7 d

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 14 / 25

Simple weighted graphs
The representation with weight matrix

The weight matrix WG = (wij ) of a simple weighted graph G with

n nodes [v1 , . . . , vn ] has dimension n × n and
B wii = 0 for all 1 ≤ i ≤ n.
B wij = w ({vi , vj }) for every edge {vi , vj } ∈ E , if G is
undirected.
B wij = w ((vi , vj )) for every arc (vi , vj ) ∈ E , if G is directed.
B wij = ∞ otherwise.

Example (Weight digraph with node enumeration [a, b, c, d, e, f ])

2
9 0 3 ∞ ∞ 4 ∞
b c
3 ∞ 0 2 ∞ ∞ ∞
12

∞ 9 0 ∞ 1 12
G :a f ⇒ WG = 
1

5 
4 ∞ ∞ ∞ 0 ∞ 7
7

1 ∞ 5 ∞ 1 0 ∞
e d ∞ ∞ ∞ ∞ ∞ 0

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 15 / 25

The representation of graphs
Comparative study

I The representation with list of edges

Adequate for the representation of simple graphs without
isolated nodes, and with |E |  |V |
Space complexity: O(|E |)
I The representation with lists of adjacencies
Allows the fast enumeration of the neighbors of a node
Space complexity: O(|V | + |E |)
I The representation with adjacency matrix AG = (aij ) or with
weight matrix WG = (wij )
Fast test to check direct connection between 2 nodes: O(1)
@(vi , vj ) ∈ E if aij = 0 or if wij = ∞
Space complexity: O(|V |2 )
- inadequate representation when |E |  |V |2
The representation with incidence matrix MG
I Time complexity: O(|V | · |E |)

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 16 / 25

Simple digraphs
Properties of the adjacency matrix AG

We assume that G is a simple (di)graph with n nodes and with the

adjacency matrix AG = (aij )
aij = 0 (or false) if @ arc vi → vj
aij = 1 (or true) if ∃ arc vi → vj
We define:
In : the identity matrix of size n × n
The Boolean operations  (conjunction) and ⊕ (disjunction):
 0 1 ⊕ 0 1 Remarks:
0 0 0 0 0 1 a  b = min(a, b)
1 0 1 1 1 1 a ⊕ b = max(a, b)
If U = (uij ), V = (vij ) are n × n matrices with elements 0 or
1, we define
U ⊕ V = (cij ) if cij = uij ⊕ vij for all i, j
U  V = (dij ) if dij = (ui1  v1j ) ⊕ . . . ⊕ (uin  vnj )
U k = U  . . .  U for every k > 0
| {z }
k times
Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 17 / 25
Simple digraphs
Properties of the adjacency matrix AG (contd.)

Properties
(k) (k)
1 If AkG = (aij ) for k ≥ 1 then aij = 1 if and only if there is a
path with length k from node vi to vj .
2 Let A∗G = In ⊕ AG ⊕ A2G ⊕ . . . ⊕ An−1
G = (aij ). Then
aij = 1 if and only if there is a path with length
j ∈ {1, . . . , n − 1} from node vi to vj .
A∗G can be computed in O(n4 ).
A∗G is called reflexive and transitive closure of AG .
3 vi and vj are connected ⇔ ∃ a simple path vi vj ⇔ aij = 1.
4 In ⊕ AG ⊕ A2G ⊕ . . . ⊕ Ak+1
G = In ⊕ (In ⊕ AG ⊕ A2G ⊕ . . . ⊕ AkG )  AG
⇒ A∗ can be computed in O(n4 ).

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 18 / 25

Simple digraphs
Properties of the adjacency matrix AG (contd.)

Properties
(k) (k)
1 If AkG = (aij ) for k ≥ 1 then aij = 1 if and only if there is a
path with length k from node vi to vj .
2 Let A∗G = In ⊕ AG ⊕ A2G ⊕ . . . ⊕ An−1
G = (aij ). Then
aij = 1 if and only if there is a path with length
j ∈ {1, . . . , n − 1} from node vi to vj .
A∗G can be computed in O(n4 ).
A∗G is called reflexive and transitive closure of AG .
3 vi and vj are connected ⇔ ∃ a simple path vi vj ⇔ aij = 1.
4 In ⊕ AG ⊕ A2G ⊕ . . . ⊕ Ak+1
G = In ⊕ (In ⊕ AG ⊕ A2G ⊕ . . . ⊕ AkG )  AG
⇒ A∗ can be computed in O(n4 ).

Corollary
The connectivity between all pairs of nodes in a simple digraph can
be checked in O(n4 ).
Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 18 / 25
Simple digraphs
Better ways to compute A∗G

Warshall algorithm computes A∗G in O(n3 ).

Warshall’s main idea

If V = [v1 , . . . , vn ] is an enumeration of the nodes of G and
vk ∈ V , then every simple path π : vi vj is of one of the
following two kinds:
1 vk does not appear along π between vi and vj
vk
vi vj
2 vk appears exactly once in π, between vi and vj
vk
vi vj

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 19 / 25

Simple digraphs
Warshall algorithm to compute A∗G

Assumption: AG = (aij ) has size n × n

[n]
I Compute recursively C [n] = (cij ) where
(
[k] aij if k = 0
cij := [k−1] [k−1] [k−1]
cij ⊕ (cik  ckj ) if k ≥ 1

Properties
1 C [0] = AG
[k]
2 cij = 1 if and only if there is a path π : vi vj where all
intermediary nodes are from the subset {v1 , . . . , vk }
3 C [n] = A∗G
4 C [n] can be computed in O(n3 ).

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 20 / 25

Simple weighted digraphs
The lightest paths

Consider the simple digraph G = (V , E ) with V = {1, . . . , n} and

the weight function w : E → R+
I In G , the weight of a path π : v1 → v2 → . . . → vp is
Pp−1
w (π) = i=1 w ((vi , vi+1 ))
(we add up the weights of all arcs of π)
I For every pair of nodes (i, j) from V , we wish to find
a lightest path from node i to node j
(there can be more than one lightest path)
the weight of a lightest path
Remember that the weight matrix of G is WG = (wij ) where

 0 if j = i,
wij = w ((i, j)) if (i, j) ∈ E ,
∞ if (i, j) 6∈ E


Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 21 / 25

Finding the lightest paths in a simple weighted digraph
Warshall algorithm: main idea

Let k ∈ V = {1, . . . , n}, and π : i j be a lightest path from i to

j. We distinguish two cases:
1 k is not an intermediary node of π
π1 π2
2 k is an intermediary node of π. Then π = i k j and
w (π) = w (π1 ) + w (π2 ).

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 22 / 25

Finding the lightest paths in a simple weighted digraph
Warshall algorithm: main idea

Let k ∈ V = {1, . . . , n}, and π : i j be a lightest path from i to

j. We distinguish two cases:
1 k is not an intermediary node of π
π1 π2
2 k is an intermediary node of π. Then π = i k j and
w (π) = w (π1 ) + w (π2 ).
An useful auxiliary data structure:
[k]
A matrix of paths P [k] = (pij ) where


 • special value: @ path i j through
intermediary nodes from {1, . . . , k}

[k]
pij :=

 π a lightest path from i to j through
intermediary nodes from {1, . . . , k}


Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 22 / 25

Finding the lightest paths in a simple weighted digraph
Warshall algorithm: the recursive formula of computation

We assume that WG = (wij )


 •
 if k = 0 and wij = ∞
[k]
pij := [i, j] if k = 0 and wij < ∞
 p-min(p [k−1] , p [k−1]  p [k−1] ) if k ≥ 1

ij ik kj

[k−1] [k−1]
where pik  pkj denotes the concatenation of the paths
[k−1] [k−1]
pik and pkj . If both exist (i.e., they are not •), and •
otherwise.
Remark
[n] [n]
If pij = •, there is no path from node i to j. Otherwise, pij is a
lightest path from node i to j.

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 23 / 25

Finding the lightest paths in a simple weighted digraph
Warshall algorithm: pseudocode

Compute recursively and simultaneously two matrices:

   
[k] [k]
W [k] = wij and P [k] = pij for 0 ≤ k ≤ n
(
[k]
wij  if k = 0,
wij =

[k−1] [k−1] [k−1]
max wij , min{wik , wkj } otherwise.

[0]


 • if k = 0 and wij = ∞,
 [0]
[k]
 [i, j] if k = 0 and wij 6= ∞,
pij = [k−1] [k−1] [k−1] [k−1]


 wij if k > 0 and wij ≤ wik + wkj ,
 w [k−1]  w [k−1]

otherwise.
ik kj

Remark
W [n] and P [n] can be computed in O(n3 ).

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 24 / 25

A special case of Warshall algorithm
Finding the shortest paths in a simple digraph

All arcs are assumed to have weight 1

v2 • • [v1 , v3 ] • •
 
[v2 , v1 ] • • • • 
v1 v3 [0]
PG =  • • • • [v3 , v5 ] ,
 
 • • [v4 , v3 ] • • 
v5 v4 [v5 , v1 ] • • [v5 , v4 ] •
• • [v1 , v3 ] • •
 
[v2 , v1 ] • [v2 , v1 , v3 ] • • 
[1]
PG =  • • • • [v3 , v5 ],
 
 • • [v4 , v3 ] • • 
[v5 , v1 ] • [v5 , v1 , v3 ] [v5 , v4 ] •
• • [v1 , v3 ] • [v1 , v3 , v5 ]
 
[v2 , v1 ] • [v2 , v1 , v3 ] • [v2 , v1 , v3 , v5 ]
[2] [1] [3]  [4] (3)
PG = PG , PG =  • • • • [v3 , v5 ] , PG = PG ,

 • • [v4 , v3 ] • [v4 , v3 , v5 ] 
[v5 , v1 ] • [v5 , v1 , v3 ] [v5 , v4 ] •
[v1 , v3 , v5 , v1 ] • [v1 , v3 ] [v1 , v3 , v5 , v4 ] [v1 , v3 , v5 ]
 
 [v2 , v1 ] • [v2 , v1 , v3 ] [v2 , v1 , v3 , v5 , v4 ] [v2 , v1 , v3 , v5 ]
[5] 
PG = [v3 , v5 , v1 ] [v3 , v5 , v1 , v3 ] [v3 , v5 , v1 , v3 ] • [v3 , v5 ] 

[v , v , v , v ]
4 3 5 1 • [v ,
4 3 v ] [v , v
4 3 5 4, v , v ] [v , v
4 3 5 , v ] 
[v5 , v1 ] • [v5 , v1 , v3 ] [v5 , v4 ] [v5 , v1 , v3 , v5 ]

Isabela Drămnesc UVT Graph Theory and Combinatorics – Lecture 10 25 / 25