Appetizer

Graph theory started with Euler who was asked to find a

nice path across the seven Köningsberg bridges

The (Eulerian) path

should cross over
each of the seven
bridges exactly once
Appetizer

Another early bird was Sir William Rowan Hamilton (1805-1865)

In 1859 he developed a toy based on finding 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

But now graph theory is used for finding communities in networks

where we want to detect hierarchies of substructures

Appetizer

and their sizes can become quite big ...

Appetizer

It is also used for ranking (ordering) hyperlinks

Appetizer

or by your GPS to find the shortest path home ...

Appetizer

or by your GPS to find the shortest path home ...

Contents

What we will cover in this course

I Basic theory about graphs

I Connectivity
I Paths
I Trees
I Networks and flows
I Eulerian and Hamiltonian graphs
I Coloring problems
I Complexity issues

I A number of applications (in large graphs)

I Large scale problems in graphs
I Similarity of nodes in large graphs
I Telephony problems and graphs
I Ranking in large graphs
I Clustering of large graphs
What are
graphs
A graph G = (V , E) is a pair of vertices (or nodes) V and
a set of edges E, assumed finite i.e. |V | = n and |E| = m.

Here V (G) = {v1 , v2 , . . . , v5 } and E(G) = {e1 , e2 , . . . , e6 }.

An edge ek = (vi , vj ) is incident with the vertices vi and vj .

A simple graph has no self-loops or multiple edges like below

What are
graphs 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 Special graphs

n(n−1)
A complete graph Kn is a simple graph with all B(n, 2) := 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 Kn is (n − 1)-regular

What are
graphs

A bipartite graph is one where V = V1 ∪ V2 such that there are no

edges between V1 and V2 (the black and white nodes below)

A complete bipartite graph is one where all edges between V1 and

V2 are present (i.e. |E| = |V1 |.|V2 |). It is noted as Kn1 ,n2 .

When is complete bipartite graph regular ?

What are
graphs When is G bipartite ?

Which graph is bipartite ?

It suffices to find 2 colors that separate the edges as below

What are
graphs When is G bipartite ?

Which graph is bipartite ?

It suffices to find 2 colors that separate the edges as below

The second example is not bipartite because it has a triangle

(to be continued)
Cycles
Walking in a graph

A walk of length k from node v0 to node vk is a non-empty graph

P = (V , E) of the form

V = {v0 , v1 , . . . , vk } E = {(v0 , v1 ), . . . , (vk −1 , vk )}

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 v0 = vk .

A cycle is a walk with different nodes except for v0 = vk .
Cycles

Try to prove the following wo (useful) lemmas

Proposition A walk from u to v 6= 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
Directed graphs

In a directed graph or digraph, each edge has a direction.

For e = (vs , vt ), vs is the source node and vt is the terminal node.

Each node v has an in-degree din (v ) and an out-degree dout (v ).

A graph is balanced if din (v ) = dout (v ) for all nodes.
Cycles
Topological order

Let us now try to order the nodes in a digraph.

Define a bijection ford : V → {1, 2, . . . , n}, then ford (·) is a

topological order for the graph G = (V , E) iff

ford (i) < ford (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 sufficient ?

Cycles

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 din (v ) > 0 for all nodes, then each node i has a
predecessor p(i) such that (vp(i) , vi ) ∈ E.
Start from an arbitrary v0 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

Let us use this to find a topological order

Algorithm FindTopOrd(G)
t := 0; G0 := G;
while ∃v ∈ Gt : din (v ) = 0 do
Gt+1 := Gt /{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

The only node of in-degree 0 is v4 . So for t = 1 we have

After removing v4 there are two nodes of in-degree 0, v1 and v3 .

If we pick v3 then we have for t = 2

Further reductions yield the final order {v4 , v3 , v1 , v2 , v5 , v6 }.

What is the complexity of this algorithm ?

Isomorphism
Isomorphic graphs

Two graphs G1 and G2 are isomorphic iff there is a bijection

between their respective nodes which make each edge of G1
correspond to exactly one edge of G2 , and vice versa.

One must find a label numbering that makes the graphs identical
This problem is still believed to be NP hard
Isomorphism
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 2n(n−1)/2 graphs with n nodes.

For n = 3 only 4 of the graphs are different
(omitting the isomorphic ones)

With n = 4 one finds eventually 11

different graphs after collapsing the
isomorphic ones
Isomorphism

Let there be Tn non-isomorphic (simple) graphs with n nodes.

Then

2n(n−1)/2
Ln := ≤ Tn ≤ 2n(n−1)/2
n!
Exercise Explain the lower bound

Taking logarithms and using n! < nn yields the bounds

B(n, 2) − n log n ≤ log Tn ≤ B(n, 2)

which gives an idea of the growth of Tn

n 2 3 4 5 6 7 8
Tn 2 4 11 34 156 1044 12346
dLn e 2 2 3 9 46 417 6658
Isomorphism
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.
Sufficiency 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 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

 
0 0 0 1 0  
0 0 1 0
 0 0 1 0 1 
 1
  0 0 0 
A1 =  0 1 0 1 1  A2 = 
 0

  1 0 0 
 1 0 1 0 1 
0 1 1 0
0 1 1 1 0
Representing
graphs

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 ek = (vi , vj ).
For a directed graph T (i, k ) = −1; T (j, k ) = 1 iff ek = (vi , vj ).
For the graphs

the incidence matrices are

 
1 0 0 0 0 0  
 0 1 1 0 0 0  −1 0 0 0 1
   0 1 0 1 −1 
T1 =  0 0 1 1 1 0
 T2 =  
 1 0 0 0 1 1


 1 0 1 −1 0 
0 −1 −1 0 0
0 1 0 1 0 1
Representing
graphs

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) log n

since one needs log n bits to represent the vertex pointers.

Representing
graphs Counting degrees

Let 1 be the vector of all ones, then din = AT 1 and dout = A1

are the vectors of in-degrees and out-degrees of the nodes of A
and dout = din = 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 = T 1.
For a directed graph one can define Tt and Ts as the matrices
containing the terminal and source nodes : T = Tt − Ts with
   
0 0 0 0 1 1 0 0 0 0
 0 1 0 1 0   0 0 0 0 1 
Tt := 
 1 0 1 0 0  , Ts :=  0 0 0 1 0 
  

0 0 0 0 0 0 1 1 0 0

Then also we have din = Tt 1 and dout = Ts 1.

Representing
graphs Powers of A

Proposition (Ak )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 Ak +1 = Ak · A is the sum of the walks of length
k to nodes that are linked to node j via the adjacency matrix A.

One verifies this in the following little example

   
0 1 0 0 0 1
A =  0 0 1  , A2 =  1 1 0 
1 1 0 0 1 1

Corollary In a simple undirected graph one has the identities

tr (A) = 0, tr (A2 )/2 = |E| and
tr (A3 )/6 equals the number of triangles in G.
Connectivity
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 definition (see later).
Connectivity

After the reduction one has an acyclic graph, which can be

ordered topologically.
What do you obtain then ? Class orderings

An initial class has din (c) = 0. A final class has dout (c) = 0.
The other ones are intermediate.
Connectivity

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 6= ∅ do
choose u ∈ L;
if ∃(u, v ) such that v is unmarked then
mark(v ); L := L ∪ {v };
else
L := L\{u};
end if
end while
Connectivity

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

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 first search (DFS).
Algorithm DeptFirstSearch(G,s)
mark(s); L := {s};
while L 6= ∅ 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 := L\{u}
end if
end while
Connectivity

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
first).
Connectivity

We now use a FIFO list for L (First In First Out) and choose for u
the first element added to L. This is a breadth first search (BFS).
Algorithm BreadthFirstSearch(G,s)
mark(s); L := {s};
while L 6= ∅ do
u := first(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 := L\{u}
end if
end while
Connectivity

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 first).

Connectivity
Testing connectivity

The exploration algorithm finds 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 find all connected components, repeat this exploration on a

node of V \C(u), etc.
Connectivity
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 finds the set of all nodes that can be
reached by a path from a given node u ∈ V .

How can one find 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 AT .

Connectivity

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 finds the set R− (u)

Here R+ (v6 ) = {4, 6} while R− (v6 ) = V hence C(v6 ) = {4, 6}

Find the other connected components.
Shortest path
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

Pj
A cost cij = α + β k =i (f (xk ) − g(xk ))2 is associated with each
linear section. This amounts to finding the shortest path in
Shortest path

Other example : Find the best production policy for a plant with a
monthly demand di , a launching cost fi , a storage cost hi and a
unit price pi , 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.
c14 = f1 + p1 (d1 + d2 + d3 ) + h1 (d2 + d3 ) + h2 (d3 ) which is the
fixed 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

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
first 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
Dijkstra’s 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 6= u;
while S 6= V do
choose v 0 ∈
/ S : d(v 0 ) ≤ d(v ) ∀v ∈ / S;
S := S ∪ {v 0 };
for each v ∈ V + (v 0 ) do
d(v ) = min{d(v ), d(v 0 ) + c(v 0 , v )}
end for
end while
Shortest path

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
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

Proposition Dijkstra’s algorithm finds in O(n2 ) 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 0 ∈
/ S : d(v 0 ) = minv ∈S
/ d(v ) then the
shortest path to v 0 must lie completely in S.
If not, ∃v 00 outside S at a shorter distance.

We can update S := S ∪ {v 0 } and compute the shortest path

from u to children of v 0 as d(v ) = min{d(v ), d(v 0 ) + c(v 0 , 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
Variants

For a graph with edge lengths 1 it suffices 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 := first(S)
if ∃(v , v 0 ) such that v 0 is unmarked then
choose (v , v 0 ) with v 0 of smallest index;
mark(v 0 ); S := S ∪ {v 0 }; d(v 0 ) = d(v ) + 1;
else
S := S\{v }
end if
end while
Shortest path

Proposition All nodes at distance exactly k are correctly

identified 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 finds 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

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) := minj {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 flow problem or
as a linear programming problem.
This leads to other algorithms like the Bellman-Ford Algorithm.
Trees
Trees and forests

A tree is an acyclic and A forest is an acyclic

connected graph 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

The following definitions are especially relevant for trees.

The eccentricity ε(u) = maxv ∈V 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) = minu∈V ε(u) of a graph G is the minimal

eccentricity of all nodes in V
The diameter diam(G) = maxu∈V ε(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

A leaf of a tree T is a node of degree 1

Proposition Let T be a tree and let T 0 be the tree obtained by

removing all its leafs, then ε(T 0 ) = ε(T ) − 1 for all nodes of T 0 .
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
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 nn−2
Trees

We construct a bijection of Tn with a sequence via the algorithm

Algorithm PrüferSequence(T)
s := (); t := ();
while |E| > 1 do
choose the leaf of smallest index i;
T := T \{i}; 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 ti which are n − 2 numbers from {1, . . . , n}
and there are exactly nn−2 such sequences.
Trees
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(m log m)
Algorithm KruskalMST(G)
Eord := sort(E); E 0 := ∅; Erest := Eord ;
while |E 0 | > n − 1 do
α := first(Erest ); Erest := Erest \{α};
if (V , E 0 ∪ {α}) is acyclic then
E 0 := E 0 ∪ {α};
end if
end while

The sorting is done efficiently

in O(m log m) time as well.
Let us look at an example
Trees

The different steps of the algorithm are

This constructs a tree which is a subgraph with n − 1 edges.

Trees

Now we look at an alternative algorithm of time complexity

O((m + n) log n)
The idea is to pick a random node and then grow a minimal tree
from there
Algorithm PrimMST(G)
Choose u ∈ V ; V 0 := {u}; E 0 := ∅;
for i = 1 : n − 1 do
E 00 := edges linking V to V 0 ;
choose e = (u, v ) ∈ E 00 of minimal weight and such that
(V 0 ∪ {v }, E 0 {e} is acyclic;
V 0 := V 0 ∪ {v }; E 0 := E 0 ∪ {e};
end for

Let us look at the same example

Trees

The different steps of the algorithm are

The graph (V , E 0 ) is a minimal spanning tree with n − 1 edges

Planar graphs
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 K3,3 (not planar) and K4 (planar)

Proposition (Fary, 1948)

Every planar graph can be
represented in the plane using
straight edges only
Planar graphs

For such graphs, one can now define

faces. These are the regions encircled
by edges that form a cycle. One has to
identify also an exterior face as shown
in this figure 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

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
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
G0 := G\(u, v ). Then f (G0 ) = f (G) − 1; e(G0 ) = e(G) − 1 and
n(G0 ) = n(G). Use the result for smaller f to prove it for f .
Planar graphs
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 ≤ 3n − 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 K5 and K3,3 are not planar.

Planar graphs

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

The Platonic solids are characterized by three equations

nk = 2e, fl = 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
Tetrahedron 3 3 6 4 4
Cube 3 4 12 8 6
Dodecahedron 3 5 30 20 12
Octahedron 4 3 12 6 8
Icosahedron 5 3 30 12 20
Planar graphs
Test for planar graphs

We first 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 e1 = (u, w) and e2 = (w, v ). The new graph is
thus given by G0 = (V ∪ {w}, E ∪ {e1 , e2 }\{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

A graph G0 = (V 0 , E 0 ) is a subgraph of a graph G = (V , E) if

V 0 ⊆ V and E 0 ⊆ E (edges must disappear along with nodes)

Proposition (Kuratowsky, 1930) A graph is planar iff it does not

contain a subgraph homeomorphic to K3,3 or K5 .

Example : the Petersen graph (subgraph + homeomorphism)

Planar graphs
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 G0 is thus
G0 = (V \{u, v } ∪ {w}, E\{e})

Proposition (Wagner, 1937) A graph is planar iff it does not have

K3,3 or K5 as a minor.

Example :
the Petersen
graph again
Planar graphs

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 Euler’s formula is preserved

Exercise Show that G = (G∗ )∗
Flows
Networks and flows
A network is a directed graph N = (V , E) with a source node s
(with dout (s) > 0) and a terminal node t (with din (t) > 0).
Moreover each edge has a strictly positive capacity c(e) > 0.

A flow 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 eachPintermediate nodeP v ∈ V \{s, t} the in- and out-flow at
that node u∈V − (v ) f (u, v ) = u∈V + (v ) f (v , u) match

The total flow F of the network is then what leaves s or reaches t

X X X X
F (N) := f (s, u) − f (u, s) = f (u, t) − f (t, u)
u∈V u∈V u∈V u∈V
Flows

Here is an example of a flow

It has a value of F (N) = 7 and the conservation law is verified

inside.

But the flow 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
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
X
κ(P, P) = c(u, v )
u∈P;v∈P

which in the above example equals 5 + 3 + 1 = 9.

We now derive important properties of this capacity.
Flows

Proposition Let (P, P) be any cut of a network N = (V , E)

then the associated flow is given by
X X
F (N) = f (u, v ) − f (v , u)
u∈P;v ∈P u∈P;v ∈P
P P P
Proof First show that F (N) = u∈P ( 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 flow is bounded by the capacity of any cut

F (N) ≤ κ(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
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

The marriage problem

One wants to find 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 find a maximum number of disjoint paths in this

directed graph. All the capacities of the existing edges are 1.
Flows

Given a network N(V , E) and a flow f then its residual network Nf

is a network with the same nodes V but with new capacities

 c(u, v ) − f (u, v ) if (u, v ) ∈ E;
cf (u, v ) = f (v , u) if (v , u) ∈ E
0 otherwise.


An augmenting path is a directed path v0 , . . . , vk from S = v0 to

t = vk for which
∆i = c(vi , vi+1 ) − f (vi , vi+1 ) > 0 ∀(vi , vi+1 ) ∈ E or
∆i = c(vi , vi+1 ) − f (vi+1 , vi ) > 0 ∀(vi+1 , vi ) ∈ E
This path is not optimal since the original flow can be increased.
Flows
Max-flow Min-cut

Proposition
The flow 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) = κ(P, P).

Proposition In a network N the following are equivalent

1. A flow is optimal
2. The residual graph does not contain an augmenting path
3. F (N) = κ(P, P) for some cut (P, P)
The value of the optimal flow thus equals F (N) = min κ(P, P)
Proof Left to the reader (combine earlier results)

Thisbecomes an LP problem
 in the flows xij on the edges (i, j)
P P P P
max i:(s,i) xsi = i:(i,s) xis subject to i xij = i xji and 0 ≤ xij ≤ cij
Flows

The Ford-Fulkerson algorithm (1956) calculates this optimal flow

using augmentation paths.

Algorithm MaxFlowFF(N,s,t)
f (u, v ) := 0 ∀(u, v ) ∈ E;
while Nf contains a path from s to t do
choose an augmentation path Ap from s to t
∆ := min(u,v )∈Ap ∆i
Augment the flow by ∆ along Ap
Update Nf
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 find F (N) = 14
Flows
Euler
Eulerian tour (1756)

An Eulerian cycle (path) is a subgraph Ge = (V , Ee ) 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

Proof (of the first 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.
Sufficiency For a single isolated node, it is trivial. For |V | > 1
there must be a cycle φ in the graph. Consider the subgraph H
with the same nodes but with the edges of φ removed.

Each of its components Hi satisfy the even degree condition and

again have an Eulerian cycle φi . By recurrence we then reduce G
to its isolated vertices.

To reconstruct the Eulerian cycle, start from a basic cycle φ.

Each time a node of another cycle φi is encountered, substitute
that cycle to the node (and do this recursively).
Euler

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 , Ed has an Eulerian tour Ge

iff it is connected and balanced, i.e. all its nodes have
din (v ) = dout (v ).
Proof Left as an exercise
Euler

The following algorithm of Fleury (1883) reconstructs a cycle C if it

exists. E 0 is the set of edges already visited by the algorithm.
Algorithm FindEulerianCycle(G)
Choose v0 ∈ V ; E 0 := ∅; C := hi;
for i = 1 : m do
choose e = (vi−1 , vi ) s.t. G0 = (V , E\E 0 ) has 1 conn. comp.;
E 0 := E 0 ∪ {e}; C := hC, ei; vi := vi−1 ;
end for

Exercise
Propose a modification addressing the Eulerian Path Problem

But what if the graph is not Eulerian ? Can we find a mininimum

cost modification of the problem ?
Euler
Chinese postman (1962)

We consider a minimum cost modification of the Eulerian cycle

problem. A chinese postman needs to find 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

Solution : find all odd degree vertices and find the shortest paths
between them

Now find 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(n3 ) time with the Hungarian algorithm.

Hamilton
Hamiltonian cycle (1859)

Was a game sold by Hamilton in 1859 to a toy maker in Dublin.

A Hamiltonian cycle is a cyclic subgraph Gh = (V , Eh ) 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

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 v1 v2 ...vn (with maybe n < |V |)

Because d(v1 ), d(vn ) ≥ n/2, it must also be covered by a cycle

(because all the neigbors of v1 and vn are on that path)

Because of connectedness n = |V | and it is a Hamiltonian cycle.

Hamilton

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 0 has at most
|V 0 | connected components for any subset of vertices V 0 ⊂ V .
Proof Let H be a Hamiltonian subgraph of G, then H − V 0 has
less than |V 0 | connected components. But G − V 0 has the same
vertices as H − V 0 and it has additional edges.

Exercise Does this graph have a Hamiltonian cycle ?

Exercise Prove that a complete bipartite graph Km,n is

Hamiltonian iff m = n
Hamilton
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 ∗ satisfies the inequalities

cost(T ) < cost(P ∗ ) ≤ 2.cost(T )
Exercise Explain why
Hamilton
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. Define K as the graph whose nodes are the edges e1 , . . . , er

not in H and with an edge between ei and ej if they cross in G.
The following graph has the vertices
(a, d), (b, f ), b, e), c, e), (d, f ) and
five edges, corresponding to the
crossings in G

Then G is planar iff K is bipartite

Exercise Explain why
Coloring
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
Coloring nodes

A k -coloring of a graph G = (V , E) is a mapping f : V → 1, . . . , k

such that f (vi ) 6= f (vj ) if (vi , vj ) ∈ 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 Clique
χ(G) = 2 χ(Kn ) = n

Petersen Graph
Even cycle
χ(G) = 3
χ(G) = 2

Odd cycle Planar graph

χ(G) = 3 χ(G) = 4
Coloring

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 χ(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
Bounds

Proposition q
1 1
Let G be connected and m = |E|, then χ(G) ≤ 2 + 2m + 4
Proof Let C = {C1 , . . . , Ck } 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) = max{d(v )|v ∈ V }, then χ(G) ≤ ∆(G) + 1 (trivial)

Proposition (Brooks, 1941)

χ(G) ≤ ∆(G) for any graph different from Kn or an odd cycle

Proposition
χ(G) ≤ 1 + maxi {min(di , i − 1)} when ordering d1 ≥ . . . ≥ dn .
Proof Order the nodes like the di ’s and use the greedy algorithm
Coloring
Greedy algorithm

Algorithm GreedyColor(G)
L := sort(V ); c := sort(colors)
for v ∈ V do
choose smallest ci 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 {u1 , v1 , u2 , v2 , u3 , v3 , u4 , v4 }

Exercise
Does each graph have a good numbering for the greedy algorithm
Coloring

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
Chromatic polynomial (Birkhoff-Lewis 1918)

The chromatic polynomial of a graph pG (k ) indicates how many

different ways a graph can be colored with k colors. E.g.

Exercise Prove the above formulas

Notice that χ(G) = min{p(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 pG (k ) = pG−(u,v ) (k ) − pG◦(u,v ) (k )
Proof u and v have different colors in G and the same in G ◦ (u, v )
Coloring

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
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 α(G) is the size of the largest possible
stable set.
Proposition One has χ(G) · α(G) ≥ n (trivial)
Coloring

The following example requires finding 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 finding a maximal clique
(or complete subgraph) in
the complementary graph
Gc = (V , Ec ), where Ec is
the complement of E
Complexity
Algorithm complexity

We distinguish problems from algorithms used to solve them.

There is also the issue of time complexity and space complexity.
The function CA (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
CA (s) = O(p(n)) for some polynomial p(·), meaning

∃n0 : CA (s) ≤ p(n) ∀n ≥ n0 .

The relative times needed to solve problems of different complexity

This shows the importance of having a polynomial problem

Complexity

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
Comparing problems

A problem Y is reducible (in polynomial time) to a problem X if

X is at least as difficult to solve as Y , denoted as X ≥p Y . Then
X ≥p Y and X ∈ P implies Y ∈ P
X ≥p Y and Y ∈
/ P implies X ∈
/P

Define the problem [LongestPath(u, v , w, N)] of finding 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 G0 = G\e,
then G is Hamiltonian. Try out all m < n2 /2 edges.

Since we know that the Hamiltonian cycle problem in not in P

the longest path problem is also not in P.
Complexity

A Boolean clause is a disjunction of Boolean terms Xi ∈ {0, 1} and

their negation X i ∈ {0, 1}, e.g. X1 ∨ X 2 ∨ X4 ∨ X 7 is a 4-term.
Define the problem [SAT] as checking if a set of Boolean clauses
can be simultaneously satisfied ([3SAT] involves only 3-terms).
E.g. {X 2 ∨ X2 , X 2 ∨ X3 ∨ X 4 , X 1 ∨ X4 } can be satisfied by choosing
X1 = 1, X2 = 0, X3 = 1, X4 = 0.

Proposition [SAT ] ≤p [3SAT ] and [3SAT ] ≤p [StableSet]

Proof We do not prove the first 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 satisfiable.
Complexity
N P and N P-complete

A problem is Non-deterministic Polynomial (N P) 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 finding it is difficult.
In P the problem can be solved in polynomial time,
in N P a solution can be checked in polynomial time.
It is still an open question of P = N P (Cray prize = 1 million dollar)
A problem X is N P-complete if X ∈ N P and ∀Y ∈ N P, Y ≤p X .
Corollary If one N P-complete problem is in P then P = N P
Corollary If one N P-complete problem is not in P then P =
6 NP
Complexity

[3SAT] is known to be N P-complete.

We now prove that also the [CLIQUE] problem is N P-complete
The [CLIQUE] problem is checking if there exists a clique
(complete subgraph) of size k in a graph G = (V , E)
Proof Consider {X1 ∨ X 2 ∨ X 3 , X 1 ∨ X2 ∨ X3 , X1 ∨ X2 ∨ X3 }.
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 satisfiable.

References
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.