Graph Theory

and
Applications
-6pt-6pt Graph Theory and
Applications
-6pt-6pt
1 / 112
Graph Theory and Applications
Paul Van Dooren
Universit catholique de Louvain
Louvain-la-Neuve, Belgium
Dublin, August 2009
Inspired from the course notes of V. Blondel and L. Wolsey (UCL)
Appetizer -6pt-6pt Appetizer
-6pt-6pt
2 / 112
Graph theory started with Euler who was asked to nd a
nice path across the seven Kningsberg bridges
The (Eulerian) path
should cross over
each of the seven
bridges exactly once
Appetizer -6pt-6pt Appetizer
-6pt-6pt
3 / 112
Another early bird was Sir William Rowan Hamilton (1805-1865)
In 1859 he developed a toy based on nding a path visiting all
cities in a graph exactly once and sold it to a toy maker in Dublin.
It never was a big success.
Appetizer -6pt-6pt Appetizer
-6pt-6pt
4 / 112
But now graph theory is used for nding communities in networks
where we want to detect hierarchies of substructures
Appetizer -6pt-6pt Appetizer
-6pt-6pt
5 / 112
and their sizes can become quite big ...
Appetizer -6pt-6pt Appetizer
-6pt-6pt
6 / 112
It is also used for ranking (ordering) hyperlinks
Appetizer -6pt-6pt Appetizer
-6pt-6pt
7 / 112
or by your GPS to nd the shortest path home ...
Appetizer -6pt-6pt Appetizer
-6pt-6pt
8 / 112
or by your GPS to nd the shortest path home ...
Contents -6pt-6pt Contents
-6pt-6pt
9 / 112
What we will cover in this course

Basic theory about graphs

Connectivity

Paths

Trees

Networks and ows

Eulerian and Hamiltonian graphs

Coloring problems

Complexity issues

A number of applications (in large graphs)

Large scale problems in graphs

Similarity of nodes in large graphs

Telephony problems and graphs

Ranking in large graphs

Clustering of large graphs

What are
graphs
-6pt-6pt What are graphs
-6pt-6pt
10 / 112
A graph G = (V, E) is a pair of vertices (or nodes) V and
a set of edges E, assumed nite i.e. [V[ = n and [E[ = m.
Here V(G) = v
1
, v
2
, . . . , v
5
and E(G) = e
1
, e
2
, . . . , e
6
.
An edge e
k
= (v
i
, v
j
) is incident with the vertices v
i
and v
j
.
A simple graph has no self-loops or multiple edges like below
What are
graphs
-6pt-6pt What are graphs
-6pt-6pt
11 / 112
Some properties
The degree d(v) of a vertex V is its number of incident edges
A self-loop counts for 2 in the degree function.
An isolated vertex has degree 0.
Proposition The sum of the degrees of a graph G = (V, E)
equals 2[E[ = 2m (trivial)
Corollary The number of vertices of odd degree is even (trivial)
What are
graphs
-6pt-6pt What are graphs
-6pt-6pt
12 / 112
Special graphs
A complete graph K
n
is a simple graph with all B(n, 2) :=
n(n1)
2
possible edges, like the matrices below for n = 2, 3, 4, 5.
A k-regular graph is a simple graph with vertices of equal degree k
Corollary The complete graph K
n
is (n 1)-regular
What are
graphs
-6pt-6pt What are graphs
-6pt-6pt
13 / 112
A bipartite graph is one where V = V
1
V
2
such that there are no
edges between V
1
and V
2
(the black and white nodes below)
A complete bipartite graph is one where all edges between V
1
and
V
2
are present (i.e. [E[ = [V
1
[.[V
2
[). It is noted as K
n
1
,n
2
.
When is complete bipartite graph regular ?
What are
graphs
-6pt-6pt What are graphs
-6pt-6pt
14 / 112
When is G bipartite ?
Which graph is bipartite ?
It sufces to nd 2 colors that separate the edges as below
What are
graphs
-6pt-6pt What are graphs
-6pt-6pt
15 / 112
When is G bipartite ?
Which graph is bipartite ?
It sufces to nd 2 colors that separate the edges as below
The second example is not bipartite because it has a triangle
(to be continued)
Cycles -6pt-6pt Cycles
-6pt-6pt
16 / 112
Walking in a graph
A walk of length k from node v
0
to node v
k
is a non-empty graph
P = (V, E) of the form
V = v
0
, v
1
, . . . , v
k
E = (v
0
, v
1
), . . . , (v
k1
, v
k
)
where edge j connects nodes j 1 and j (i.e. [V[ = [E[ + 1).
A trail is a walk with all different edges.
A path is a walk with all different nodes (and hence edges).
A walk or trail is closed when v
0
= v
k
.
A cycle is a walk with different nodes except for v
0
= v
k
.
Cycles -6pt-6pt Cycles
-6pt-6pt
17 / 112
Try to prove the following wo (useful) lemmas
Proposition A walk from u to v ,= u contains a path from u to v
Hint : eliminate subcycles
Proposition A closed walk of odd length contains a cycle of odd
length
Hint : decompose recursively into distinct subgraphs and use
induction
Question Is this only for simple graphs ?
Cycles -6pt-6pt Cycles
-6pt-6pt
18 / 112
Directed graphs
In a directed graph or digraph, each edge has a direction.
For e = (v
s
, v
t
), v
s
is the source node and v
t
is the terminal node.
Each node v has an in-degree d
in
(v) and an out-degree d
out
(v).
A graph is balanced if d
in
(v) = d
out
(v) for all nodes.
Cycles -6pt-6pt Cycles
-6pt-6pt
19 / 112
Topological order
Let us now try to order the nodes in a digraph.
Dene a bijection f
ord
: V 1, 2, . . . , n, then f
ord
() is a
topological order for the graph G = (V, E) iff
f
ord
(i ) < f
ord
(j ), (i , j ) E
This is apparently possible for the above graph.
It is easy to see that such a graph should have no cycles.
But is this also sufcient ?
Cycles -6pt-6pt Cycles
-6pt-6pt
20 / 112
An acyclic graph is a graph without cycles.
Proposition
Every acyclic graph contains at least one node with zero in-degree
Proof By contradiction.
Assume d
in
(v) > 0 for all nodes, then each node i has a
predecessor p(i ) such that (v
p(i )
, v
i
) E.
Start from an arbitrary v
0
to form a list of predecessors as below
Since [V[ is bounded, one must eventually return to a node that
was already visited; hence there is a cycle.
Cycles -6pt-6pt Cycles
-6pt-6pt
21 / 112
Let us use this to nd a topological order
Algorithm FindTopOrd(G)
t := 0; G
0
:= G;
while v G
t
: d
in
(v) = 0 do
G
t +1
:= G
t
/v; order (v) := t + 1; t := t + 1;
end while
if t = n then G is acyclic;
else if t < n then G has a cycle; end if
end if
Let us verify this algorithm on the above example.
Cycles -6pt-6pt Cycles
-6pt-6pt
22 / 112
The only node of in-degree 0 is v
4
. So for t = 1 we have
After removing v
4
there are two nodes of in-degree 0, v
1
and v
3
.
If we pick v
3
then we have for t = 2
Further reductions yield the nal order v
4
, v
3
, v
1
, v
2
, v
5
, v
6
.
What is the complexity of this algorithm ?
Isomorphism -6pt-6pt Isomorphism
-6pt-6pt
23 / 112
Isomorphic graphs
Two graphs G
1
and G
2
are isomorphic iff there is a bijection
between their respective nodes which make each edge of G
1
correspond to exactly one edge of G
2
, and vice versa.
One must nd a label numbering that makes the graphs identical
This problem is still believed to be NP hard
Isomorphism -6pt-6pt Isomorphism
-6pt-6pt
24 / 112
Counting graphs
How many different simple graphs are there with n nodes ?
A graph with n nodes can have B(n, 2) := n(n 1)/2 different
edges and each of them can be present or not.
Hence there can be at most 2
n(n1)/2
graphs with n nodes.
For n = 3 only 4 of the graphs are different
(omitting the isomorphic ones)
With n = 4 one nds eventually 11
different graphs after collapsing the
isomorphic ones
Isomorphism -6pt-6pt Isomorphism
-6pt-6pt
25 / 112
Let there be T
n
non-isomorphic (simple) graphs with n nodes.
Then
L
n
:=
2
n(n1)/2
n!
T
n
2
n(n1)/2
Exercise Explain the lower bound
Taking logarithms and using n! < n
n
yields the bounds
B(n, 2) n logn logT
n
B(n, 2)
which gives an idea of the growth of T
n
n 2 3 4 5 6 7 8
T
n
2 4 11 34 156 1044 12346
L
n
| 2 2 3 9 46 417 6658
Isomorphism -6pt-6pt Isomorphism
-6pt-6pt
26 / 112
Bipartite revisited
Let us look again at bipartite graphs
Proposition A graph is bipartite iff it has no cycles of odd length
Necessity Trivial : color the nodes of the cycle black and white.
Sufciency Pick u V and let f (v) be the length of a shortest
path from u to v ( if there is no such path)
A = v V[f (v) = odd B = v V[f (v) = even
Then A and B form a partition of the nodes of V connected to u.
One then needs to show that there
can be no links between any two
nodes of A or any two nodes of B.
If this would be the case, one could
construct a cycle of odd length.
Repeat on each subgraph.
Representing
graphs
-6pt-6pt Representing graphs
-6pt-6pt
27 / 112
Representing graphs
A graph G = (V, E) is often represented by its adjacency matrix.
It is an n n matrix A with A(i , j ) = 1 iff (i , j ) E. For the graphs
the adjacency matrices are
A
1
=

0 0 0 1 0
0 0 1 0 1
0 1 0 1 1
1 0 1 0 1
0 1 1 1 0

A
2
=

0 0 1 0
1 0 0 0
0 1 0 0
0 1 1 0

Representing
graphs
-6pt-6pt Representing graphs
-6pt-6pt
28 / 112
A graph can also be represented by its n m incidence matrix T.
For an undirected graph T(i , k) = T(j , k) = 1 iff e
k
= (v
i
, v
j
).
For a directed graph T(i , k) = 1; T(j , k) = 1 iff e
k
= (v
i
, v
j
).
For the graphs
the incidence matrices are
T
1
=

1 0 0 0 0 0
0 1 1 0 0 0
0 0 1 1 1 0
1 0 0 0 1 1
0 1 0 1 0 1

T
2
=

1 0 0 0 1
0 1 0 1 1
1 0 1 1 0
0 1 1 0 0

Representing
graphs
-6pt-6pt Representing graphs
-6pt-6pt
29 / 112
One can also use a sparse matrix representation of A and T.
This is in fact nothing but a list of edges, organized e.g. by nodes.
Notice that the size of the representation of a graph is thus linear
in the number of edges in the graph (i.e. in m = [E[).
To be more precise, one should count the number of bits needed
to represent all entries :
L = (n + m) logn
since one needs logn bits to represent the vertex pointers.
Representing
graphs
-6pt-6pt Representing graphs
-6pt-6pt
30 / 112
Counting degrees
Let 1 be the vector of all ones, then d
in
= A
T
1 and d
out
= A1
are the vectors of in-degrees and out-degrees of the nodes of A
and d
out
= d
in
= d for undirected graphs.
How should we then take self-loops into account ?
In an adjacency matrix of an undirected graph A(i , i ) = 2
In an adjacency matrix of a directed graph A(i , i ) = 1
For an undirected graph, we have d = T1.
For a directed graph one can dene T
t
and T
s
as the matrices
containing the terminal and source nodes : T = T
t
T
s
with
T
t
:=

0 0 0 0 1
0 1 0 1 0
1 0 1 0 0
0 0 0 0 0

, T
s
:=

1 0 0 0 0
0 0 0 0 1
0 0 0 1 0
0 1 1 0 0

Then also we have d

in
= T
t
1 and d
out
= T
s
1.
Representing
graphs
-6pt-6pt Representing graphs
-6pt-6pt
31 / 112
Powers of A
Proposition (A
k
)
ij
is the number of walks of length k from i to j
Proof Trivial for k=1; by induction for larger k.
The element (i , j ) of A
k+1
= A
k
A is the sum of the walks of length
k to nodes that are linked to node j via the adjacency matrix A.
One veries this in the following little example
A =

0 1 0
0 0 1
1 1 0

, A
2
=

0 0 1
1 1 0
0 1 1

Corollary In a simple undirected graph one has the identities

tr (A) = 0, tr (A
2
)/2 = [E[ and
tr (A
3
)/6 equals the number of triangles in G.
Connectivity -6pt-6pt Connectivity
-6pt-6pt
32 / 112
Connected components
In a directed graph G = (V, E), u and v are strongly connected
if there exists a walk from u to v and from v to u.
This is an equivalence relation and hence leads to equivalence
classes, which are called th connected components of the graph
G.
The graph reduced to its connected components is acyclic (why ?)
This shows up in many applications, e.g. in the dictionary graph.
The connected components are the groups of words that use each
other in their denition (see later).
Connectivity -6pt-6pt Connectivity
-6pt-6pt
33 / 112
After the reduction one has an acyclic graph, which can be
ordered topologically.
What do you obtain then ? Class orderings
An initial class has d
in
(c) = 0. A nal class has d
out
(c) = 0.
The other ones are intermediate.
Connectivity -6pt-6pt Connectivity
-6pt-6pt
34 / 112
Verify (strong) connectivity of a graph based on its adjacency list
Idea : start from node s, explore the graph, mark what you visit
Algorithm GenericSearch(G,s)
mark(s); L := s
while L ,= do
choose u L;
if (u, v) such that v is unmarked then
mark(v); L := L v;
else
L := Lu;
end if
end while
Connectivity -6pt-6pt Connectivity
-6pt-6pt
35 / 112
Below we marked the chosen nodes and the discovered nodes
L mark
{2} 2
{2, 1} 1
{2, 1, 5} 5
{2, 1, 5, 6} 6
{1, 5, 6}
{1, 5, 6, 4} 4
{5, 6, 4}
{5, 4}
{5, 4, 3} 3
{5, 3}
{5, 3, 7} 7
{5, 3}
{3}
{3, 8} 8
{3}
{}
This algorithm has 2n steps : each node is added once and
removed once. Its complexity is therefore linear in n.
Connectivity -6pt-6pt Connectivity
-6pt-6pt
36 / 112
Because of the choices, this algorithm allows for different versions
Let us use a LIFO list for L (Last In First Out) and choose for u the
last element added to L. This is a depth rst search (DFS).
Algorithm DeptFirstSearch(G,s)
mark(s); L := s;
while L ,= do
u := last (L)
if (u, v) such that v is unmarked then
choose (u, v) with v of smallest index;
mark(v); L := L v;
else
L := Lu
end if
end while
Connectivity -6pt-6pt Connectivity
-6pt-6pt
37 / 112
Below we marked the chosen nodes and the discovered nodes
L mark
{2} 2
{2, 1} 1
{2, 1, 4} 4
{2, 1, 4, 3} 3
{2, 1, 4, 3, 7} 7
{2, 1, 4, 3}
{2, 1, 4, 3, 8} 8
{2, 1, 4, 3}
{2, 1, 4}
{2, 1, 4, 6} 6
{2, 1, 4, 6, 5} 5
{2, 1, 4, 6}
{2, 1, 4}
{2, 1}
{2}
{}
This algorithm builds longer paths than the generic one (depth
rst).
Connectivity -6pt-6pt Connectivity
-6pt-6pt
38 / 112
We now use a FIFO list for L (First In First Out) and choose for u
the rst element added to L. This is a breadth rst search (BFS).
Algorithm BreadthFirstSearch(G,s)
mark(s); L := s;
while L ,= do
u := rst (L)
if (u, v) such that v is unmarked then
choose (u, v) with v of smallest index;
mark(v); L := L v;
else
L := Lu
end if
end while
Connectivity -6pt-6pt Connectivity
-6pt-6pt
39 / 112
Below we marked the chosen nodes and the discovered nodes
L mark
{2} 2
{2, 1} 1
{2, 1, 5} 5
{1, 5}
{1, 5, 4} 4
{1, 5, 4, 6} 6
{5, 4, 6}
{4, 6}
{4, 6, 3} 3
{6, 3}
{3}
{3, 7}
{3, 7, 8} 8
{7, 8}
{8}
{}
This algorithm builds a wider tree (breadth rst).
Connectivity -6pt-6pt Connectivity
-6pt-6pt
40 / 112
Testing connectivity
The exploration algorithm nds the set of all nodes that can be
reached by a path from a given node u V.
If the graph is undirected, each node in that set can follow a path
back to u. They thus form the connected component C(u) of u.
To nd all connected components, repeat this exploration on a
node of VC(u), etc.
Connectivity -6pt-6pt Connectivity
-6pt-6pt
41 / 112
Testing strong connectivity
Proposition Let G = (V, E) be a digraph and let u V.
If v V there exists a path from u to v and a path from v to u,
then G is strongly connected.
The exploration algorithm nds the set of all nodes that can be
reached by a path from a given node u V.
How can one nd the nodes from which u can be reached ?
Construct for that the inverse graph by reversing all arrows
Show that the adjacency matrix of this graph is just A
T
.
Connectivity -6pt-6pt Connectivity
-6pt-6pt
42 / 112
Proposition Let G = (V, E) be a digraph and let u V.
Let R
+
(u) be the nodes that can be reached from u
and let R

(u) be the nodes that can reach u,

then the strongly connected component of u is
C(u) = R
+
(u) R

(u)
The exploration algorithm applied to the inverse graph, starting
from u nds the set R

(u)
Here R
+
(v
6
) = 4, 6 while R

(v
6
) = V hence C(v
6
) = 4, 6
Find the other connected components.
Shortest path -6pt-6pt Shortest path
-6pt-6pt
43 / 112
Shortest path problems
Find the shortest total length of a path between two nodes of a
directed graph with lengths associated with each edge.
E.g. Find the best piecewise linear approximation of a function
A cost c
ij
= +

j
k=i
(f (x
k
) g(x
k
))
2
is associated with each
linear section. This amounts to nding the shortest path in
Shortest path -6pt-6pt Shortest path
-6pt-6pt
44 / 112
Other example : Find the best production policy for a plant with a
monthly demand d
i
, a launching cost f
i
, a storage cost h
i
and a
unit price p
i
, for each period i = 1, . . . , n.
In the path below, we are e.g. producing in stages 1, 4 and 5.
A cost is associated with each section. For the path (1,4) it is e.g.
c
14
= f
1
+ p
1
(d
1
+ d
2
+ d
3
) + h
1
(d
2
+ d
3
) + h
2
(d
3
) which is the
xed cost + the production cost in periods 1, 2 and 3 + storage
costs at the end of periods 1 and 2.
The minimization of the total cost amounts to a shortest path
problem in a graph combining paths as above.
Shortest path -6pt-6pt Shortest path
-6pt-6pt
45 / 112
Proposition If there is a shortest walk from s to t , there is also a
shortest path from s to t
Proof
Assume the walk is not a path;
hence there is a recurring node.
Eliminate the cycle between the
rst and last occurrence of this
node. Repeat this procedure.
In the above graph the path (7, 8, 6, 3, 1, 5, 6, 10, 4, 6, 9) has a
cycle (6, 3, 1, 5, 6, 10, 4, 6). After its elimination we have a path
(7, 8, 6, 9).
Corollary If G does not contain cycles of negative length, the
resulting path is one of lower cost.
Proof Trivial
Shortest path -6pt-6pt Shortest path
-6pt-6pt
46 / 112
Dijkstras algorithm
This method is for a digraph G that has positive edge lengths.
For undirected graphs one can duplicate each edge as follows
Below, V
+
(u) denotes the set of children of u.
Algorithm Dijkstra(G,u)
S := u; d(u) := 0; d(v) := c(u, v) v ,= u;
while S ,= V do
choose v

/ S : d(v

) d(v) v / S;
S := S v

;
for each v V
+
(v

) do
d(v) = mind(v), d(v

) + c(v

, v)
end for
end while
Shortest path -6pt-6pt Shortest path
-6pt-6pt
47 / 112
Idea : Update a set S for which we know all shortest paths from u
Let us see the behavior of this
algorithm on an example.
The table below indicates the
steps and the distances
computed for each node
Iter S d(u) d(1) d(2) d(3) d(4)
0 u 0 1 3 6
1 u, 1 0 1 2 4 6
2 u, 1, 2 0 1 2 3 6
3 u, 1, 2, 3 0 1 2 3 5
4 u, 1, 2, 3, 4 0 1 2 3 5
We indicate in more detail the exploration of the graph
Shortest path -6pt-6pt Shortest path
-6pt-6pt
48 / 112
S d(u) d(1) d(2) d(3) d(4)
{u} 0 1 3 6
{u, 1} 0 1 2 4 6
{u, 1, 2} 0 1 2 3 6
{u, 1, 2, 3} 0 1 2 3 5
{u, 1, 2, 3, 4} 0 1 2 3 5
Below, the node u is blue and the explored nodes are red
Shortest path -6pt-6pt Shortest path
-6pt-6pt
49 / 112
Proposition Dijkstras algorithm nds in O(n
2
) time the shortest
path from u to all other nodes of V.
Proof By induction on the size of S, we show that
1. v S, d(v) is the length of the shortest path from u to v
2. v S
+
(children of nodes of S), d(v) is the length of the
shortest path from u to v not passing exclusively via nodes of S
Trivial for S = u, d(u) = 0, d(v) = c(u, v).
Let v

/ S : d(v

) = min
v / S
d(v) then the
shortest path to v

must lie completely in S.

If not, v

outside S at a shorter distance.

We can update S := S v

and compute the shortest path

from u to children of v

as d(v) = mind(v), d(v

) + c(v

, v)).
This gives the length of the shortest path to all v S
+
.
The other distances are unknown as yet and hence set to .
Shortest path -6pt-6pt Shortest path
-6pt-6pt
50 / 112
Variants
For a graph with edge lengths 1 it sufces to do a BFSearch and
to keep track of the path lengths by incrementing them with 1
during the exploration phase. This is thus an O(m) time algorithm.
Algorithm ShortestPathBFS(G,v)
mark(v); S := v; d(v) = 0;
while not S = do
v := rst (S)
if (v, v

) such that v

is unmarked then
choose (v, v

) with v

of smallest index;
mark(v

); S := S v

; d(v

) = d(v) + 1;
else
S := Sv
end if
end while
Shortest path -6pt-6pt Shortest path
-6pt-6pt
51 / 112
Proposition All nodes at distance exactly k are correctly
identied before proceeding further.
Proof For k = 0 this is trivial (S is the original node u).
Induction step : suppose the statement is correct up to k.
After all nodes at distance k have been found, one nds nodes
that are at a distance larger than k but since they are all
neighboring nodes, they must be at distance exactly k + 1.
Shortest path -6pt-6pt Shortest path
-6pt-6pt
52 / 112
For an acyclic graph, one can just compute the topological order in
O(m) time (see earlier).
To solve the shortest path problem one then uses the algorithm
Algorithm ShortestPathAcyclic(G,v)
d(1) = 0; d(i ) := for i = 2, . . . , n;
for i = 1 : n 1 do
for j V
+
(i ) do
d(j ) := min
j
d(j ), d(i ) + c(i , j );
end for
end for
What is the complexity of this second step ?
One can also see the shortest path problem as a ow problem or
as a linear programming problem.
This leads to other algorithms like the Bellman-Ford Algorithm.
Trees -6pt-6pt Trees
-6pt-6pt
53 / 112
Trees and forests
A tree is an acyclic and
connected graph
A forest is an acyclic
graph (and hence a union
of trees)
Proposition For a graph G = (V, E) of order n = [V[, the
following are equivalent
1. G is connected and has n 1 edges
2. G is acyclic and has n 1 edges
3. G is connected and acyclic
4. u, v V there is one and only one path from u to v
5. G is acyclic and adding an edge creates one and only one cycle
6. G is connected and removing an arbitrary edge disconnects it
Proofs ?
Trees -6pt-6pt Trees
-6pt-6pt
54 / 112
The following denitions are especially relevant for trees.
The eccentricity (u) = max
vV
d(u, v) of a node is the maximum
distance to any node v V. The eccentricity of each node is
indicated in the graph below
The radius rad(G) = min
uV
(u) of a graph G is the minimal
eccentricity of all nodes in V
The diameter diam(G) = max
uV
(u) of a graph G is the maximal
eccentricity of all nodes in V. It is also the maximal distance
between any two nodes in V
The center of a graph G is the set of nodes in V of minimal
eccentricity (the black node)
Trees -6pt-6pt Trees
-6pt-6pt
55 / 112
A leaf of a tree T is a node of degree 1
Proposition Let T be a tree and let T

be the tree obtained by

removing all its leafs, then (T

) = (T) 1 for all nodes of T

.
Proof ?
Proposition The center of a tree is a single node or a pair of
adjacent nodes.
Proof By induction using the previous proposition.
Show that the center does not change.
Trees -6pt-6pt Trees
-6pt-6pt
56 / 112
Counting trees
How many different (labeled) trees are there with n nodes ?
The following table gives the count for small n
The following theorem of Cayley gives the exact formula.
Proposition
The number of distinct labeled trees of order n equals n
n2
Trees -6pt-6pt Trees
-6pt-6pt
57 / 112
We construct a bijection of T
n
with a sequence via the algorithm
Algorithm PrferSequence(T)
s := (); t := ();
while [E[ > 1 do
choose the leaf of smallest index i ;
T := Ti ; s := (s, i ); t := (t , neighbour (i ));
end while
On the graph below, it yields the table next to it
One shows that the graph can be reconstructed from the
sequence t
i
which are n 2 numbers from 1, . . . , n
and there are exactly n
n2
such sequences.
Trees -6pt-6pt Trees
-6pt-6pt
58 / 112
Spanning tree
Remove from a connected graph as many edges as possible while
remainig connected; this should yield a tree with n 1 edges.
This is the minimal spanning tree problem solved by the following
algorithm, of time complexity O(mlogm)
Algorithm KruskalMST(G)
E
ord
:= sort (E); E

:= ; E
rest
:= E
ord
;
while [E

[ > n 1 do
:= rst (E
rest
); E
rest
:= E
rest
;
if (V, E

) is acyclic then
E

:= E

;
end if
end while
The sorting is done efciently
in O(mlogm) time as well.
Let us look at an example
Trees -6pt-6pt Trees
-6pt-6pt
59 / 112
The different steps of the algorithm are
This constructs a tree which is a subgraph with n 1 edges.
Trees -6pt-6pt Trees
-6pt-6pt
60 / 112
Now we look at an alternative algorithm of time complexity
O((m + n) logn)
The idea is to pick a random node and then grow a minimal tree
from there
Algorithm PrimMST(G)
Choose u V; V

:= u; E

:= ;
for i = 1 : n 1 do
E

:= edges linking V to V

;
choose e = (u, v) E

of minimal weight and such that

(V

v, E

e is acyclic;
V

:= V

v; E

:= E

e;
end for
Let us look at the same example
Trees -6pt-6pt Trees
-6pt-6pt
61 / 112
The different steps of the algorithm are
The graph (V, E

) is a minimal spanning tree with n 1 edges

Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
62 / 112
Planar graphs
When drawing connected graphs one is naturally lead to the
question of crossing edges. One says that a graph is planar if it
can be drawn (or represented) without crossing edges
The above graphs represent K
3,3
(not planar) and K
4
(planar)
Proposition (Fary, 1948)
Every planar graph can be
represented in the plane using
straight edges only
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
63 / 112
For such graphs, one can now dene
faces. These are the regions encircled
by edges that form a cycle. One has to
identify also an exterior face as shown
in this gure with 6 faces
Proposition A planar representation of a graph can be
transformed to another one where any face becomes the exterior
face (a proof comes later)
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
64 / 112
Proposition A graph can be
represented in a plane if and
only if it can be represented on
a sphere (immersion)
Proof
Use a stereographic projection
Every face of the plane is mapped to a sector on the sphere. No
point on the sphere can therefore belong to two different sectors.
The external face is mapped to a sector containing the north pole.
For the external face result, notice that by rotating the sphere, one
can move any point (and hence sector) to the north pole
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
65 / 112
Characterisation
Proposition (Euler formula) Let G be planar, and let n(G) be its
number of vertices, e(G) its number of edges, and f (G) its number
of faces. Then f = e n + 2.
In the example shown here
n = 8; e = 10; f = 4
Proof Use induction on the number of faces f .
For f = 1 there are no cycles and hence the connected graph is a
three, for which we know e = n 1 and hence f = e n + 2.
For f 2, remove an edge (u, v) between two faces to construct
G

:= G(u, v). Then f (G

) = f (G) 1; e(G

) = e(G) 1 and
n(G

) = n(G). Use the result for smaller f to prove it for f .

Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
66 / 112
Some exercices
Proposition Let G be planar with f > 1, then 3f 2e
Proposition Let G be planar with f > 1 and G have no triangles,
then 2f e
Proposition Let G be a planar (connected) graph.
If n 3 then e 2n 6
Proposition Let G be a planar (connected) graph.
If G has no triangles or is bipartite, then e 2n 4
These help to prove the following lemma
Proposition K
5
and K
3,3
are not planar.
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
67 / 112
Corollary The average degree of the vertices of a planar
connected graph G is smaller than 6
12
n
Corollary In a planar (connected) graph there always exists a
vertex such that d(v) 5
Corollary A planar graph can be colored with 6 colors (see later)
Proposition (Platonic solid) There are only 5 regular polyhedra
These so-called Platonic solids are shown below
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
68 / 112
The Platonic solids are characterized by three equations
nk = 2e, = 2e for k, l integers, and n + f = e + 2
Explain why
It then follows that 2e/k + 2e/l e = 2 hence 2/k + 2/l > 1
or (k 2)(l 2) < 4. The integer solutions are given by
Name k l e n f
Tetraeder 3 3 6 4 4
Cube 3 4 12 8 6
Dodecaeder 3 5 30 20 12
Octaeder 4 3 12 6 8
Icosaeder 5 3 30 12 20
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
69 / 112
Test for planar graphs
We rst need to introduce subdivisions and subgraphs.
Let us expand a graph G = (V, E) by a subdivision of one of its
edges e = (u, v) E. We put a new node w on e and replace it
by two new edges e
1
= (u, w) and e
2
= (w, v). The new graph is
thus given by G

= (V w, E e
1
, e
2
e).
Two graphs are said to be homeomorphic to each other iff one can
be derived from the other via a sequence of subdivisions.
Corollary Homeomorphism is an equivalence relation.
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
70 / 112
A graph G

= (V

, E

) is a subgraph of a graph G = (V, E) if

V

V and E

E (edges must disappear along with nodes)

Proposition (Kuratowsky, 1930) A graph is planar iff it does not
contain a subgraph homeomorphic to K
3,3
or K
5
.
Example : the Petersen graph (subgraph + homeomorphism)
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
71 / 112
Minors
Let e = (u, v) be an edge of a graph G = (V, E). A contraction of
the edge e consists of eliminating e and merging the nodes u and
v into a new node w. The new graph G

is thus
G

= (Vu, v w, Ee)
Proposition (Wagner, 1937) A graph is planar iff it does not have
K
3,3
or K
5
as a minor.
Example :
the Petersen
graph again
Planar graphs -6pt-6pt Planar graphs
-6pt-6pt
72 / 112
Proposition (Robertson-Seymour)
For a graph G, determining if a given graph H is a minor of H, can
be solved in polynomial time (with respect to n(G) and m(G)).
A dual graph G

of a planar graph is obtained as follows

1. G

has a vertex in each face of G

2. G

has an edge between two vertices if G has an edge

between the corresponding faces
This is again a planar graph
but it might be a multigraph
(with more than one edge
betwee two vertices)
Exercise Show that Eulers formula is preserved
Exercise Show that G = (G

Flows -6pt-6pt Flows

-6pt-6pt
73 / 112
Networks and ows
A network is a directed graph N = (V, E) with a source node s
(with d
out
(s) > 0) and a terminal node t (with d
in
(t ) > 0).
Moreover each edge has a strictly positive capacity c(e) > 0.
A ow f : V
2
R
+
is associated with each edge e = (u, v) s.t.
1. for each edge e E we have 0 f (e) c(e)
2. for each intermediate node v Vs, t the in- and out-ow at
that node

uV

(v)
f (u, v) =

uV
+
(v)
f (v, u) match
The total ow F of the network is then what leaves s or reaches t
F(N) :=

uV
f (s, u)

uV
f (u, s) =

uV
f (u, t )

uV
f (t , u)
Flows -6pt-6pt Flows
-6pt-6pt
74 / 112
Here is an example of a ow
It has a value of F(N) = 7 and the conservation law is veried
inside.
But the ow is not maximal, while the next one is (F(N) = 9) as
we will show later. Notice that one edge is not being used (f = 0)
Flows -6pt-6pt Flows
-6pt-6pt
75 / 112
Cut of a network
A cut of a network is a partition of the vertex set V = P P into
two disjoint sets P (containing s ) and P (containing t )
The capacity of a cut is the sum of the capacities of the edges
(u, v) between P and P
(P, P) =

uP;vP
c(u, v)
which in the above example equals 5 + 3 + 3 + 1 = 9.
We now derive important properties of this capacity.
Flows -6pt-6pt Flows
-6pt-6pt
76 / 112
Proposition Let (P, P) be any cut of a network N = (V, E)
then the associated ow is given by
F(N) =

uP;vP
f (u, v)

uP;vP
f (v, u)
Proof First show that F(N) =

uP
(

v
f (u, v)

v
f (v, u))
by summing all contributions in P and using conservation.
For all v P the term between brackets is zero (conservation).
Hence we only need to keep the edges across the partition.
Corollary A ow is bounded by the capacity of any cut
F(N) (P, P)
A minimal cut (with minimal capacity) also bounds F(N)
(we will construct one and will see it is in fact equal to F(N))
Flows -6pt-6pt Flows
-6pt-6pt
77 / 112
Applications
The dining problem
Can we seat 4 families with number of members (3,4,3,2) at 4
tables with number of seats (5,2,3,2) so that no two members of a
same family sit at the same table ?
The central edges are the table assignments (a capacity of 1).
The cut shown has a capacity 11 which upper bounds F(N).
We can therefore not seat all 12 members of the four families.
Flows -6pt-6pt Flows
-6pt-6pt
78 / 112
The marriage problem
One wants to nd a maximimum number of couplings between
men and women where each couple has expressed whether or
not this coupling was acceptable (central edges that exist or not)
One wants to nd a maximum number of disjoint paths in this
directed graph. All the capacities of the existing edges are 1.
Flows -6pt-6pt Flows
-6pt-6pt
79 / 112
Given a network N(V, E) and a ow f then its residual network N
f
is a network with the same nodes V but with new capacities
c
f
(u, v) =

c(u, v) f (u, v) if (u, v) E;

f (v, u) if (v, u) E
0 otherwise.
An augmenting path is a directed path v
0
, . . . , v
k
from S = v
0
to
t = v
k
for which

i
= c(v
i
, v
i +1
) f (v
i
, v
i +1
) > 0 (v
i
, v
i +1
) E or

i
= c(v
i
, v
i +1
) f (v
i +1
, v
i
) > 0 (v
i +1
, v
i
) E
This path is not optimal since the original ow can be increased.
Flows -6pt-6pt Flows
-6pt-6pt
80 / 112
Max-ow Min-cut
Proposition
The ow is optimal if there exists no augmentation path from s to t
Proof Construct a cut (P, P) where u P if there is an
augmentation path from s to u and u P otherwise.
Show that (P, P) is a valid cut for which F(N) = (P, P).
Proposition In a network N the following are equivalent
1. A ow is optimal
2. The residual graph does not contain an augmenting path
3. F(N) = (P, P) for some cut (P, P)
The value of the optimal ow thus equals F(N) = min(P, P)
Proof Left to the reader (combine earlier results)
This becomes an LP problem in the ows x
ij
on the edges (i , j )
max

i :(s,i )
x
si
=

i :(i ,s)
x
is

subject to

i
x
ij
=

i
x
ji
and 0 x
ij
c
ij
Flows -6pt-6pt Flows
-6pt-6pt
81 / 112
The Ford-Fulkerson algorithm (1956) calculates this optimal ow
using augmentation paths.
Algorithm MaxFlowFF(N,s,t)
f (u, v) := 0 (u, v) E;
while N
f
contains a path from s to t do
choose an augmentation path A
p
from s to t
:= min
(u,v)A
p

i
Augment the ow by along A
p
Update N
f
end while
Finding a path in the residual graph can be implemented with a
BFS or DFS exploration as shown below
At each step we show the graph (left) and the residual graph (right)
Augmentation paths are in red. In 5 steps we nd F(N) = 14
Flows -6pt-6pt Flows
-6pt-6pt
82 / 112
Euler -6pt-6pt Euler
-6pt-6pt
83 / 112
Eulerian tour (1756)
An Eulerian cycle (path) is a subgraph G
e
= (V, E
e
) of G = (V, E)
which passes exactly once through each edge of G.
G must thus be connected and all vertices V are visited
(perhaps more than once). One then says that G is Eulerian
Proposition A graph G has an Eulerian cycle iff it is connected
and has no vertices of odd degree
A graph G has an Eulerian path (i.e. not closed) iff it is connected
and has 2 or no vertices of odd degree
This would prove that the above graph is not Eulerian.
Euler -6pt-6pt Euler
-6pt-6pt
84 / 112
Proof (of the rst part regarding cycles)
Necessity Since G is Eulerian there is a cycle visiting all nodes.
Each time we visit v V, we leave it again, hence d(v) is even.
Sufciency For a single isolated node, it is trivial. For [V[ > 1
there must be a cycle in the graph. Consider the subgraph H
with the same nodes but with the edges of removed.
Each of its components H
i
satisfy the even degree condition and
again have an Eulerian cycle
i
. By recurrence we then reduce G
to its isolated vertices.
To reconstruct the Eulerian cycle, start from a basic cycle .
Each time a node of another cycle
i
is encountered, substitute
that cycle to the node (and do this recursively).
Euler -6pt-6pt Euler
-6pt-6pt
85 / 112
Proof (of the second part regarding paths) Left as an exercise
The path problem says if you can draw a graph without lifting your
pen. Apply this to the following examples.
Proposition A directed graph G = V, E
d
has an Eulerian tour G
e
iff it is connected and balanced, i.e. all its nodes have
d
in
(v) = d
out
(v).
Proof Left as an exercise
Euler -6pt-6pt Euler
-6pt-6pt
86 / 112
The following algorithm of Fleury (1883) reconstructs a cycle C if it
exists. E

is the set of edges already visited by the algorithm.

Algorithm FindEulerianCycle(G)
Choose v
0
V; E

:= ; C := );
for i = 1 : m do
choose e = (v
i 1
, v
i
) s.t. G

= (V, EE

) has 1 conn. comp.;

E

:= E

e; C := C, e); v
i
:= v
i 1
;
end for
Exercise
Propose a modication addressing the Eulerian Path Problem
But what if the graph is not Eulerian ? Can we nd a mininimum
cost modication of the problem ?
Euler -6pt-6pt Euler
-6pt-6pt
87 / 112
Chinese postman (1962)
We consider a minimum cost modication of the Eulerian cycle
problem. A chinese postman needs to nd a tour passing along all
edges of a graph and minimize the length of the path.
The edges have a cost and we need to make the graph Eulerian
Exercise
1. Give a simple lower bound. 2. When can this bound be met ?
3. Is there another solution (or a better one) ?
Euler -6pt-6pt Euler
-6pt-6pt
88 / 112
Solution : nd all odd degree vertices and nd the shortest paths
between them
Now nd a perfect matching of the nodes in this graph.
A perfect matching in a graph is a set of disjoint edges of a graph
to which all vertices are incident.
This can be solved in 0(n
3
) time with the Hungarian algorithm.
Hamilton -6pt-6pt Hamilton
-6pt-6pt
89 / 112
Hamiltonian cycle (1859)
Was a game sold by Hamilton in 1859 to a toy maker in Dublin.
A Hamiltonian cycle is a cyclic subgraph G
h
= (V, E
h
) of
G = (V, E) which passes exactly once through all nodes
It is a so-called hard problem and there is no general condition
for its existence (in contrast with the Eulerian path problem).
It exists for Platonic solids and complete graphs,
but not for the Petersen graph
Hamilton -6pt-6pt Hamilton
-6pt-6pt
90 / 112
Proposition (Dirac, 1951) A graph G with n 3 nodes and
d(v) n/2, v V, is Hamiltonian
Proof
G is connected, otherwise its smallest component would have all
edges with d(v) < n/2
Then consider a longest path v
1
v
2
...v
n
(with maybe n < [V[)
Because d(v
1
), d(v
n
) n/2, it must also be covered by a cycle
(because all the neigbors of v
1
and v
n
are on that path)
Because of connectedness n = [V[ and it is a Hamiltonian cycle.
Hamilton -6pt-6pt Hamilton
-6pt-6pt
91 / 112
Exercise Construct a graph with d(v) < n/2 and yet has a
Hamiltonian cycle
Proposition
If G = (V, E) has a Hamiltonian cycle, then G V

has at most
[V

[ connected components for any subset of vertices V

V.
Proof Let H be a Hamiltonian subgraph of G, then H V

has
less than [V

[ connected components. But G V

has the same

vertices as H V

and it has additional edges.

Exercise Does this graph have a Hamiltonian cycle ?
Exercise Prove that a complete bipartite graph K
m,n
is
Hamiltonian iff m = n
Hamilton -6pt-6pt Hamilton
-6pt-6pt
92 / 112
Traveling Salemen Problem
A traveling salesman is supposed to visit a number of cities
(nodes in a graph) and minimize the travel time (or total length)
This is NP-hard but can often be solved approximately in
reasonable time. Consider a distance graphs with triangle
inequality d(u, v) + d(v, w) d(u, w) u, v, w V
Construct a minimal weight spanning
tree T and visit the nodes using BFS.
For this example we would have a
cycle (a,b,c,b,h,b,a,d,e,f,e,g,e,a)
Notice that all edges are visited twice.
The optimal path P

satises the inequalities

cost (T) < cost (P

) 2.cost (T)
Exercise Explain why
Hamilton -6pt-6pt Hamilton
-6pt-6pt
93 / 112
Test for planar graph
There is a simple way to test if a Hamiltonian graph is planar
1. Draw G with the Hamiltonian graph H at the outside
The following graph is already drawn
with H = (a, b, c, d, e, f , a) outside
2. Dene K as the graph whose nodes are the edges e
1
, . . . , e
r
not in H and with an edge between e
i
and e
j
if they cross in G.
The following graph has the vertices
(a, d), (b, f ), b, e), c, e), (d, f ) and
ve edges, corresponding to the
crossings in G
Then G is planar iff K is bipartite
Exercise Explain why
Coloring -6pt-6pt Coloring
-6pt-6pt
94 / 112
Four color problem
In 1852 it was conjectured that a country map (like the USA map)
could always be colored with only four colors. There is an
underlying assumption for point borders.
This was proven in 1976 by K. Appel and W. Haken but their proof
used a computer search over 1200 so-called critical cases.
Exercise What property does the underlying graph have ?
Coloring -6pt-6pt Coloring
-6pt-6pt
95 / 112
Coloring nodes
A k-coloring of a graph G = (V, E) is a mapping f : V 1, . . . , k
such that f (v
i
) ,= f (v
j
) if (v
i
, v
j
) E.
The chromatic number of a graph is the smallest number k for
which there exists a k-coloring.
Some examples of known chromatic numbers are :
Bipartite graph
(G) = 2
Even cycle
(G) = 2
Odd cycle
(G) = 3
Clique
(K
n
) = n
Petersen Graph
(G) = 3
Planar graph
(G) = 4
Coloring -6pt-6pt Coloring
-6pt-6pt
96 / 112
The coloring problem for general graphs is NP-complete but
such problems often lead to more interesting applications
Exam scheduling problem
The table on the left gives the exams each student takes
The chromatic number (G) of the corresponding graph
gives the minimum numbers of time slots for the exams
Exercise Can you formulate such a slot problem with students
choosing out of k pre-set programs ?
Coloring -6pt-6pt Coloring
-6pt-6pt
97 / 112
Bounds
Proposition
Let G be connected and m = [E[, then (G)
1
2
+

2m +
1
4
Proof Let C = C
1
, . . . , C
k
be the partition of V according to
colors. There is at least one edge between two colors, which
implies m > B(k, 2) and hence k
2
k 2m 0.
Proposition
Let (G) = maxd(v)[v V, then (G) (G) + 1 (trivial)
Proposition (Brooks, 1941)
(G) (G) for any graph different from K
n
or an odd cycle
Proposition
(G) 1 + max
i
min(d
i
, i 1) when ordering d
1
. . . d
n
.
Proof Order the nodes like the d
i
s and use the greedy algorithm
Coloring -6pt-6pt Coloring
-6pt-6pt
98 / 112
Greedy algorithm
Algorithm GreedyColor(G)
L := sort (V); c := sort (colors)
for v V do
choose smallest c
i
not used by colored neigbors
end for
On a bipartite graph this greedy algorithm is optimal when
numbering the nodes per part but it can be bad for other
numberings, such as u
1
, v
1
, u
2
, v
2
, u
3
, v
3
, u
4
, v
4

Exercise
Does each graph have a good numbering for the greedy algorithm
Coloring -6pt-6pt Coloring
-6pt-6pt
99 / 112
Let us come back to the map coloring problem
and try to prove the following (simpler) result
Exercise Every planar graph can be colored with 6 colors
Show that e 3n 6
Show then that for planar graphs average(d(v)) 6 12/n
Finally prove that there exists a v such that d(v) 5
Now use induction to prove the proposition (remove nodes)
Coloring -6pt-6pt Coloring
-6pt-6pt
100 / 112
Chromatic polynomial (Birkhoff-Lewis 1918)
The chromatic polynomial of a graph p
G
(k) indicates how many
different ways a graph can be colored with k colors. E.g.
Exercise Prove the above formulas
Notice that (G) = minp(G)(k) > 0. Does this help ?
There is a powerful induction theorem using the simpler graphs
G (u, v) (remove an edge) and G (u, v) (contract an edge)
Proposition If (u, v) E then p
G
(k) = p
G(u,v)
(k) p
G(u,v)
(k)
Proof u and v have different colors in G and the same in G (u, v)
Coloring -6pt-6pt Coloring
-6pt-6pt
101 / 112
This can be used to compute the chromatic polynomial of more
complex networks
but the problem remains combinatorial and thus hard
Exercise Derive this quicker using the result for a tree
Coloring -6pt-6pt Coloring
-6pt-6pt
102 / 112
Stable sets
An independent or stable set S in a graph G = (V, E) is a
subgraph of G without any edges, i.e. u, v S : (u, v) E
The two sets of black nodes are stable sets of the left graph
Such sets can clearly be colored with only one color, which proves
Proposition If a graph is k-colorable then V can be partitioned as
k stable sets
The independence number (G) is the size of the largest possible
stable set.
Proposition One has (G) (G) n (trivial)
Coloring -6pt-6pt Coloring
-6pt-6pt
103 / 112
The following example requires nding a maximal stable set.
Find the maximum number of projects one can realize when
the table indicates which students are needed for each project.
Notice that it is equivalent
to nding a maximal clique
(or complete subgraph) in
the complementary graph
G
c
= (V, E
c
), where E
c
is
the complement of E
Complexity -6pt-6pt Complexity
-6pt-6pt
104 / 112
Algorithm complexity
We distinguish problems from algorithms used to solve them.
There is also the issue of time complexity and space complexity.
The function C
A
(s) of an algorithm is the number of time steps
needed to solve a problem of size s with that algorithm.
A problem is called polynomial if there exists an algorithm with
C
A
(s) = O(p(n)) for some polynomial p(), meaning
n
0
: C
A
(s) p(n) n n
0
.
The relative times needed to solve problems of different complexity
This shows the importance of having a polynomial problem
Complexity -6pt-6pt Complexity
-6pt-6pt
105 / 112
Better is to look at the size of
the problems one can solve
when the machines speed up
100 or 1000 times
Here are a number of polynomial time problems
Finding the shortest path between 2 vertices
Testing if a graph is planar
Testing if a graph is Eulerian
Finding a spanning tree
Solving the perfect marriage problem
Here are a number of problems that are not polynomial
Finding the chromatic number of a graph
Finding a Hamiltonian cycle in a graph
Finding the largest stable set in a graph
Solving the travelling salesman problem
Testing if two graphs are isomorphic (not known)
Complexity -6pt-6pt Complexity
-6pt-6pt
106 / 112
Comparing problems
A problem Y is reducible (in polynomial time) to a problem X if
X is at least as difcult to solve as Y, denoted as X
p
Y. Then
X
p
Y and X T implies Y T
X
p
Y and Y / T implies X / T
Dene the problem [LongestPath(u, v, w, N)] of nding a path
of length any N from u to v in a graph with integer weights w
Proposition [HamiltonianCycle]
p
[LongestPath(u, v, w, N)]
Proof Choose unit weights w. Pick an edge e = (u, v).
If there is a longest path of length N = n 1 in G

= Ge,
then G is Hamiltonian. Try out all m < n
2
/2 edges.
Since we know that the Hamiltonian cycle problem in not in T
the longest path problem is also not in T.
Complexity -6pt-6pt Complexity
-6pt-6pt
107 / 112
A Boolean clause is a disjunction of Boolean terms X
i
0, 1 and
their negation X
i
0, 1, e.g. X
1
X
2
X
4
X
7
is a 4-term.
Dene the problem [SAT] as checking if a set of Boolean clauses
can be simultaneously satised ([3SAT] involves only 3-terms).
E.g. X
2
X
2
, X
2
X
3
X
4
, X
1
X
4
can be satised by choosing
X
1
= 1, X
2
= 0, X
3
= 1, X
4
= 0.
Proposition [SAT]
p
[3SAT] and [3SAT]
p
[StableSet ]
Proof We do not prove the rst part involving only 3-terms.
Construct a triangle for
each 3-term and then
connect the negations
across triangles
For a stable set, I can choose only one node in each triangle.
Then there is a stable set of size n/3 iff [3SAT] is satisable.
Complexity -6pt-6pt Complexity
-6pt-6pt
108 / 112
^T and ^T-complete
A problem is Non-deterministic Polynomial (^T) if the validity of a
solution can be checked in polynomial time.
Checking if a given cycle is Hamiltonian can be solved in
polynomial time, but nding it is difcult.
In T the problem can be solved in polynomial time,
in ^T a solution can be checked in polynomial time.
It is still an open question of T = ^T (Cray prize = 1 million dollar)
A problem X is ^T-complete if X ^T and Y ^T, Y
p
X.
Corollary If one ^T-complete problem is in T then T = ^T
Corollary If one ^T-complete problem is not in T then T , = ^T
Complexity -6pt-6pt Complexity
-6pt-6pt
109 / 112
[3SAT] is known to be ^T-complete.
We now prove that also the [CLIQUE] problem is ^T-complete
The [CLIQUE] problem is checking if there exists a clique
(complete subgraph) of size k in a graph G = (V, E)
Proof Consider X
1
X
2
X
3
, X
1
X
2
X
3
, X
1
X
2
X
3
.
Construct a graph with the terms of each clause as nodes.
Then connect all pairs of variable except their negation (partially
done below)
If this graph contains a clique of size 3, the clause is satisable.
References -6pt-6pt References
-6pt-6pt
110 / 112
Some useful literature
J.A. Bondy and U.S.R. Murty, Graph Theory with Applications,
(2nd Edition), North Holland, 1976.
Reinhard Diestel, Graph Theory, Graduate Texts in Mathematics,
Vol. 173, Springer Verlag, Berlin, 1991.
Douglas West, Introduction to Graph Theory, (2nd Edition),
Prentice Hall, 2000.
B. Bollobas, Modern Graph Theory, Springer-Verlag.
Fan Cheung and Linyuan Lu, Complex Graphs and Networks,
Regional Conference Series in Mathematics, Vol. 107, AMS, 2004
Dieter Jungnickel, Graphs, Networks and Algorithms, Algorithms
and Computation in Mathematics, Vol. 5, Springer Verlag, Berlin,
2005.