7. Planar Graphs 6.1.

Planar graphs

Definitions: planar and plane graphs. A graph G(n,m) said to be realizable or embeddable on surface S if

The Euler'
Euler's formula and its consequences. it is possible to distinguish a collection of n distinct points of S
that correspond to the vertices of G and
Characterization of plane graphs.
a collection of m curves,
curves, pairwise disjoint except possibly for
Connection between the planar graphs and polyhedra.
polyhedra. endpoints,
endpoints, on S that correspond to the edges of G
Sufficient condition for planarity. The Kuratowsky theorem. such that if a curve A corresponds to the edge e = uv,
uv, then only the
Hamiltonian planar graphs. endpoints of A correspond to vertices of G, namely u and v.
Crossing number
Intuitively,
Intuitively, G is embeddable on S if G can be drawn on S so that
Electrical networks edges (i.e. the curves corresponding to edges) intersect only at a
Squaring the square. vertex (a point corresponding to a vertex) mutually incident with
them.
them.
Graph Theory 6 1 Graph Theory 6 2

Theorem 6.1.: Embedding a graph in the plane is equivalent to

A graph G is a planar graph if it can be drawn in the plane in such a embedding it on the sphere.
sphere.
way that no two edges intersect,
intersect, i.e.
i.e. if it can be embedded in the
plane. Proof
plane.

A graph G is planar if it is possible to represent it in the plane in We prove the theorem by using a stereographic projection:
such a way that the vertices correspond to distinct points and the Let S be a sphere tangent to a plane , where P is the point of S
edges to simple Jordan curves connecting the points of its diametrically opposite to the point of tangency.
tangency.
endvertices in such a way that every two curves are either disjoint or
meet only at a common endpoint.
endpoint. If a graph G is embedded on S in such a way that no vertex of G is P
and no edge of G passes through P, then G may be projected onto
to produce an embedding of G on .
The invers of this projection shows that any graph that can be
embedded in the plane can also be embedded on the sphere.
sphere.

Graph Theory 6 3 Graph Theory 6 4

 If a planar graph is embedded in the plane,
plane, then it is called a plane
graph.
P S:
G1 G2 G3

G1 is planar graph, but as drawn it is not plane.

plane.
G2 is planar and plane.
plane.
G3 is not planar.
planar.

Graph Theory 6 5 Graph Theory 6 6

A region of G is a maximal portion of the plane for which any two Statement:
Statement: A plane graph G can alway
always be embedded in the plane so
points may be connected by a curve A such that each points of A that a given region of G becomes the exterior region.
neither corresponds to a vertex of G nor lies on any curve
corresponding to an edge of G. Consequence:
Consequence: A plane graph G can always be realized in the plane so
that any vertex or edge lies on the boundary of its exterior region.
For a plane graph, the boundary of a region R consists of all those
points x corresponding to vertices and edges of G having the A plane graph together with the set of regions it determines is called
property that x can be connected to a point of R by a curve,
curve, all of a plane map.
map.
whose points different from x belong to R.
The faces of a plane map are usually called countries.
countries.
Every plane graph G has an unbounded region called the exterior
region of G. Two countries are neighbouring if their boundaries have an edge in
common.
common.
Since a cycle separates the points of the plane into two components,
components,
each edge of a cycle is in the boundary of two region.

Graph Theory 6 7 Graph Theory 6 8

unbounded region Theorem 6.2. (Euler's
(Euler's formula)
formula) : If a connected plane graph G has n
vertices, m edges and r regions,
regions, then
bounded regions n m + r= 2.
Proof.
By induction on the number of regions.
regions.

If r = 1 then G does not contain a cycle so it is a tree and the

result holds.
neighbouring countries boundaries Suppose that r >1 and the result holds for any r ' < r. Since r >1,
so G contains at least one cycle.

Let ab be an edge in a cycle, and suppose that ab is in the bound-

bound-
ary of two regions,
regions, say S and T.

Graph Theory 6 9 Graph Theory 6 10

G For the planar graphs in general,

general, we have the following result:
T Corollary 6.3.: If G is a plane graph with n vertices, m edges and r
regions,
regions, then
S a n m + r = 1 + k(G).
k(G).

b A planar graph G is called maximal planar if, for every pair u, v of

non-
non-adjacent vertices of G, the graph G + uv is nonplanar.
nonplanar.

Ommiting ab, Statement:

Statement: Any embedding of a maximal planar graph G having
ab, in the new plane graph G'
n' = n, m'= m 1 and r'= r order n 3, the boundary of every region of G is a triangle.
triangle.
r 1.
A maximal planar graph is also referred to as triangulated planar
graph.
n m + r= n m + r = 2
Graph Theory 6 11 Graph Theory 6 12
Theorem 6.4.: If G is maximal planar graph of order n 3and size m,
then m = 3n 6. Corollary 6.5: If G is a planar graph of order n 3 and size m, then
m 3n 6.
Proof.
Proof.
Embed G in the plane,
plane, resulting in a plane graph with r regions.
regions.
The boundary of every regions
regions is a triangle,
triangle, and each edge is on the Add to G sufficiently many edges so that the resulting graph of order
boundary of two regions.
regions. n' and size m' is maximal planar.
planar.
Therefore,
Therefore, if the number of edges on the boundary of a region is Clearly,
Clearly, n = n'
n' and m m, so if we apply the Theorem 6.4,
6.4, we get
summed over all regions,
regions, the result is 3r. the statement of the Corollary.
Corollary.
On the other hand,
hand, such a sum counts each edge twice,
twice, so 3r = 2m.
Applying the Euler'
Euler's formula, we obtain the desired result:
2m
nm+ =2
and so 3
3n m = 6.
Graph Theory 6 13 Graph Theory 6 14

Corollary 6.6: Every planar graph contains a vertex of degree at Corollary 6.7: In a maximal planar graph of order at least 4 the
most 5. minimum degree is at least 3.
Proof. Proof.

Let G be a planar graph of order n and size m with V(G) ={

={v1,,vn}. Let G be the corresponding graph with size m. Then, by Theorem
If n 6 then the result is obvious.
obvious. 6.4. m = 3n 6.
Suppose n > 6.
6. The m 3n 6 implies that trivial, that G does not contain vertices with degree 0 or 1.
It is trivial,
n We will show that G contains no vertex of degree 2.
deg vi = 2m 6 n 12.
i =1 contrary, and let v
Suppose the contrary, V(G) for which deg v = 2.
Assume that all vertices have degree at least 6. Then 2m 6n which Then G v is a planar graph of order n 1 and size m 2.
contradicts the above formel.
formel. Since m = 3n 6 we get
So,
So, G contains a vertex of degree 5 or less. m 2 = 3n 8 = 3(n 1) 5 > 3(n 1) 6,
which contradicts Corollary 6.5.

Graph Theory 6 15 Graph Theory 6 16

Corollary 6.8: Let G be a maximal planar graph of order n 4, and Is there any connection between planar graphs and polyhedra?
polyhedra?
let ni denote the number of vertices of degree i in G for i = 3,4,
3,4,,k =
(G). Then Every polyhedron P is associated a connected planar graph G(P),
3n3 + 2n4 + n5 = n7 + 2n8 + + (k 6)nk + 12. whose vertices and edges are the vertices and edges of P.

Proof. If G(P) is a plane graph, then the faces of P are the regions of G(P)
Let G have a size m. Then, by Theorem 6.4, m = 3n 6. Since and every edge of G(P) is on the boundary of two regions.
k k
n= ni and in i = 2 m
i=3 i=3
it follows that k k
in i = 6 n i 12
i=3 i=3

and, consequently,
consequently,

3n3 + 2n4 + n5 = n7 + 2n8 + + (k 6)nk + 12.

A polyhedron and its associated planar graph
Graph Theory 6 17 Graph Theory 6 18

Theorem 6.10.: At least one face of every polyhedron is bounded by

Theorem 6.9 (Euler's Polyhedron formula): If V, E and F denote the
6.9. (Euler'
a k-cycle for some k = 3,4,5.
number of vertices, edges and faces of a polyhedron, respectively,
respectively,
then Proof.
V E + F = 2. contrary, that F3= F4= F5 = 0. By equation (*)
Assume, to the contrary,
2E = kFk 6 Fk = 6 Fk = 6 F .
For a given polyhedron P we will denote by Vk the number of k 6 k 6 k 6

vertices of degree k, and the number of faces bounded by a k-cycle Hence E 3F. Similarly
by Fk. Then
2 E = kVk = kFk . (*) 2E = kV k 3V k = 3 V k = 3V .
k 3 k 3 k 3
k 3 k 3

So 2E 3V. By Theorem 6.9,

By the Corollary 6.6,
6.6, every polyhedron has at least one vertex of
6 = 3V 3E + 3F 2E 3E + E = 0,
degree 3, 4 or 5. As an analogue to this result, we have the following which is a contradiction.
theorem:
theorem:

Graph Theory 6 19 Graph Theory 6 20

A regular polyhedron is a polyhedron whose faces are bounded by
congruent regular polygons and whose polyhedral angles are
congruent.
congruent. 8 = 4 E 4V 4 F

For a regular polyhedron F = Fs for some s and V = Vt for some t. = 2 E + 2 E 4V 4 F

= k Fk + kV k 4 Vk 4 Fk
Theorem 6.11.: There are exactly five regular polyhedra.
polyhedra. They are k 3 k 3 k 3 k 3
called platonic solids.
solids.
= ( k 4 )F k + ( k 4 )V k
Proof. k 3 k 3

Let P be a regular polyhedron and let G(P) be an associated planar Since P is regular, there exist integers s ( 3) and t ( 3) such that F
graph. = Fs, and V = Vt. Hence
Then V E + F = 2, where V, E and F denote the number of 8 = (s 4)Fs +(t 4)Vt.
vertices, edges and faces of P and G(P). Therefore Moreover, 3 s 5, 3 t 5, and sFs = 2E = tVt.
This gives us nine cases to consider:

Graph Theory 6 21 Graph Theory 6 22

We will denote the case s = i and t = j by Case (i, j).

4. Case (4,3): Here we have
1. Case (3,3): Here we have 8 = V3 and 4F4 = 3V3
8 = F3 V3 and 3F3 = 3V3 so V3 = 8 and F4 = 6 . Thus P is the cube.
so F3 = V3 = 4. Thus P is the tetrahedron.

5. Case (4,4): This is impossible since 8 0.

2. Case (3,4): Now

6. Case (4,5): This is also impossible since 8 V5.

8 = F3 and 3F3 = 4V3
so F3 = 8 and V4 = 6, implying that P is the octahedron. 7. Case (5,3): Here we have
3. Case (3,5): In this case, 8 = F5 V3 and 5F4 = 3V3.
8 = F3 + F5 and 3F3 = 5V3 so V3 = 20 and F5 = 12 . Thus P is the dodecahedron.
so F3 = 20 and V5= 12, and P is the icosahedron.

Graph Theory 6 23 Graph Theory 6 24



7. Case (5,4): This is impossible since 8 F5.

8. Case (5,5): This is also impossible since 8 F5 + V5.

tetrahedron cube octahedron
This completes the proof.

dodecahedron icosahedron
Graph Theory 6 25 Graph Theory 6 26

Exercises.
Exercises. (G. Chartrand and L. Lesniak page 135.)
135.) When will be a graph G a planar-
planar-graph?
graph?
Euler's formula is only necessary but not sufficient!
The Euler' sufficient!

The Euler's
Euler's formula gives an upper bound for the number of edges of
a planar graph of order n:
m + 1 n + F.
So, a planar graph may not contains too many
many edges.

Let us denote by Fi the number of faces having exactly i edges in

their boundaries.
boundaries.

Fi = F (1)
i

and if G has no bridge then

iF i = 2 m (2)
Graph Theory 6 27 Graph Theory 6 i 28
 If we introduce the definition of the girth,
girth, then we can improve the Let G be a planar graph of order n, size m and girth at least g.
upper bound.
bound. If n g 1 then G is acyclic so m n 1.
Assume now that n g and the assertion holds for n' < n.
The girth is the number of edges in a shortest cycle. (The girth in an
acyclic graph is . ) A. First we suppose that G contains a bridge. If ab is a bridge then G
ab is the union of two vertex disjoint subgraphs,
subgraphs, say G1 and
Theorem 6.12.: A planar graph of order n and girth at least g, 3 g G2. Let ni and mi, i=1,2. Because of the induction hypothesis for
< , has size at most the two disconnected subgraph the assertion is valid,
valid, so
g
max
g
( n 2 ), n 1 . m = m1 + m2 + 1 max ( n1 2 ), n1 1
g 2 g 2
g
+ max ( n2 2 ), n2 1 + 1
Proof. g 2
The Theorem 6.4 is the special case of this statement (g = 3). g
= max ( n 2 ), n 1 .
g2
Graph Theory 6 29 Graph Theory 6 30

B. If G is bridgless,
bridgless, then by (1) and (2) we get Is the complete graph of order 5 (K5) planar?
planar?
2m = iF i = iF i gF i = gF
i i g i
e( K 5 ) = 10 > 3( 5 2 )

so, the graph is not planar.

planar.
Hence, by the Euler's
Euler's formula,
2
m+ 2 = n + F n+ m
g
and so Is the complete 3 by 3 bipartite (K3,3) planar?
planar?
g
m (n2)
g2 e ( K 3 ,3 ) = 9 > 4
42 (6 2 ) = 8

so, this graph is also not planar.

planar.
Graph Theory 6 31 Graph Theory 6 32
 A graph H is the subdivision of the graph G if we substitute an edge
with a path which joins to the edge only in its endvertices.
endvertices.

Theorem 6.13. (Kuratowski

(Kuratowski 1930): A graph if planar iff it does not
contain a subdivision of K5 or K3,3.

Graph Theory 6 Three houses and three wheels 33 Graph Theory 6 34

6.2. Hamiltonian Planar Graphs v1

Let G be a hamiltonian plane graph of order n and let C be a fixed v6

hamiltonian cycle in G. With respect to this cycle, a chord is an edge v5 v2
v10 v7
of G that does not lie on C.
v9 v8
Let ri (i=3,4,,n) denote the number of regions of G in the interior
of C whose boundary contains exactly i edges. v4 v3

Let r'i (i=3,4,,n) denote the number of regions of G in the exterior

of C whose boundary contains exactly i edges.

ri = 0 if i 4, and r4 = 4.

r'i = 0 if i 4, 5 and r'4 = 1, r'
r'5= 2.

Graph Theory 6 35 Graph Theory 6 36

Theorem 6.14. (E. J. Grinberg,
Grinberg, 1968) : Let G be a plane graph of
order n with hamiltonian cycle C. Then with respect to this cycle C, Let the number of edges bounding a region interior to C be summed
n over all d + 1 such regions, and let us denote the result by N. Hence
( i 2 )( ri ri ) = 0 . n
i=3 N = iri .
i=3
Proof.
However, N counts each interior chord twice and each edge of G
We first consider the interior of C. If d denotes the number of once, so that N = 2d + n. Thus,
chords of G in the interior of C, then exactly d+1 regions of G lie n

inside C. Therefore, iri = 2 d + n . (2)

Therefore, i=3
n
ri = d + 1, Substituting (1) into this equation, we get
i =3 n n
iri = 2 ri 2 + n ,
implying that i=3 i=3

n so n
d= ri 1. (1) ( i 2 ) ri = n 2 . (3)
i =3 i=3

Graph Theory 6 37 Graph Theory 6 38

By considering the exterior of C, we conclude in a similar fashion Let G be a plane graph with hamiltonian cycle C (see the picture.)
picture.)
that Suppose that the edge e of G is on the boundary of two regions R1
n
and R2 of G.
( i 2 )ri = n 2. (4)
i=3
If e is an edge of
of C, then one of R1 and R2 is in the interior of C
It follows from (3) and (4) that and the other is in the exterior of C.
n
If e is not an edge of
of C, then R1 and R2 are either both in in the
( i 2 )(ri ri) = 0.
i=3 interior of C or both in the exterior of C.

P.G. Tait (1880) conjectured that every 3-connected cubic planar

graph is hamiltonian.
hamiltonian.

Tutte (1946) disproved this conjecture by producing a graph as a

counterexample:
counterexample:

Graph Theory 6 39 Graph Theory 6 40

 v1 v2 Theorem 6.15.: The Tutte graph is not hamiltonian.
hamiltonian.
G:
Proof.
f1 f2 Assume, to the contrary, that the Tutte graph G, which has order 46,
w1 w2 contains a hamiltonian cycle C.
R1 R2
w e Observe that C must contain exactly two of the edges e, f1 and f2.
Consider the regions R1, R2 and R3 of G. Suppose that two of them,
say R1 and R2, lie in the exterior of C.
Then the edges f1 and f2 do not belong to C since the unbounded
R3 region of G also lies in the exterior of C.
This is impossible, thus at most one of the regions R1, R2 and R3 lies
in the exterior of C.
We conclude that at least two of these regions, say R1 and R2, lie in
the interior of C.
The Tutte graph
Graph Theory 6 41 Graph Theory 6 42

This implies that their common boundary edge e does not belong to Since v1 and v2 is an edge of C2 and since the unbounded region of
C. G2 lies in the exterior of C2, we have that

Therefore, f1 and f2 are edges of C. r3 r3 = 1 0 = 1 and r8 r8 = 0 1 = 1 .

Now, let G1 denote the component of G {e, f1, f2} containing w. Therefore from (1) we obtain
2( r4 r4 ) + 3( r5 r5 ) = 5
Then the cycle C contains a v1 v2 subpath P that is a hamiltonian
path of G1. Since degG2 w = 2 both ww1 and ww2 are edges of C2. This implies that
Consider the graph G2 = G1 + v1v2. r4 1, so
r4 r4 = 1 1 = 0 or r4 r4 = 2 0 = 2 .
Then G2 has hamiltonian cycle C2 consisting of P together with the
edge v1v2. If r4 r4 = 0 then 3( r5 r5 ) = 5 which is impossible.
Let us apply Theorem 6.14 to G2 and cycle C. We will get
If r4 r4 = 2 then 3( r5 r5 ) = 1 again impossible.
impossible.
1( r3 r3 ) + 2( r4 r4 ) + 3( r5 r5 ) + 6 ( r8 r8 ) = 0 (1)

So we conclude that the Tutte graph is not hamiltonian.

hamiltonian.
Graph Theory 6 43 Graph Theory 6 44
For many years, Tutte's
Tutte's graph was the only known example of a 3-
3- 6.3.Crossing Number
connected cubic planar graph.
Later, other graphs have been found: How can we measure the non-
non-planarity of a graph?
The crossing number (G) of a graph G is the minimum number of
crossing of its edges among the drawings of G in the plane.

In any drawing we assume that

adjacent edges never cross
two nonadjacent edges cross at most once
no edge crosses itself
no more than two edges cross at a point of the plane
the (open) arc in the plane corresponding to an edge of the graph
contains no vertex of the graph.

The Grinberg graph The Herschel graph

Graph Theory 6 45 Graph Theory 6 46

Some easy observations : Turn'

Turn's Brick-
Brick-Factory Problem:
There were
a graph G is planar iff (G) = 0.
some kilns where the bricks were made, and
If G H then iff (G) (H).

some open storage yards were the bricks were stored

If H is a subdivision of G then (G) = (H)
All the kilns were connected by rail with all the storage yards, and
Theorem 6.16. (Guy, 1960): For complete graphs, the bricks were carried on small wheeled trucks to the storage yards.
yards.
1 n n 1 n 2 n 3 The trouble was at the crossing: the trucks generally jumped the rails
(G )
4 2 2 2 2 there, and the bricks fell out of them.
The loss of time could have been minimized if the number of
crossing of the rails had been minimized.
Theorem 6.17. (Guy, 1960): For 1 n 10,
10,
1 n n 1 n 2 n3 Turn realized that the actual situation could have been improved,
( Kn ) = but the general problem with s kilns and t storage yards seemed to be
4 2 2 2 2
very difficult.
Graph Theory 6 47 Graph Theory 6 48
Conjecture 6.18. (Zarankiewicz
(Zarankiewicz--conjecture): The situation regarding crossing numbers of complete k-partite
s s 1 t t 1 graphs, k 3, is even more complecated.
complecated.
( K s ,t ) = .
2 2 2 2 Theorem 6.20.: The crossing number K2,2,3 is (K2,2,3) = 2.
Proof.
Theorem 6.19. (Kleitman
(Kleitman,, 1970): If s and t are integers (s t) and
Let (K2,2,3) = c. Since K3,3 is nonplanar and K3,3 K2,2,3, it follows

either s 6 or s = 7 and t 10, then
that K2,2,3 is nonplanar so that c 1.
s s 1 t t 1
( K s ,t ) = . Let there be given a drawing of K2,2,3 in the plane with c crossing. At
2 2 2 2
each crossing we introduce a new vertex, producing a connected
plane graph G of order n = 7+c and size m = 16 + 2c.
Consequences:
(K3,3) = 1 (K4,4) = 4 (K5,5) = 16 By Corollary 6.5, m 3n 6.
Let u1u2 and v1v2 be two (nonadjacent) edges of K2,2,3 that cross in the
(K6,6) = 36 (K7,7) = 81 given drawing, giving rise to a new vertex.

Graph Theory 6 49 Graph Theory 6 50

If G is a triangulation, then C: u1,v1,u2,v2,u1 is a cycle of G, imply-

imply- v2
ing that the induced subgraph {u1,u2,v1,v2} in K2,2,3 is isomorphic to
K4.
However, K2,2,3 contains no such subgraphs;
However, subgraphs; thus G, is not a
triangulation so that m < 3n 6. So we have u2
16 + 2c < 3(7 + c) 6,
from which it follows that c > 1, and so, c 2.
v1
Otherwise, c 2 follows from the fact that there exists a drawing of
Otherwise,
K2,2,3 with two crossings.
crossings.

u1

Graph Theory 6 51 Graph Theory 6 52

Exercises.
Exercises. (G. Chartrand and L. Lesniak.)
Lesniak.) 7.1. Electrical networks
A simple electrical network can be regarded as a graph in which each
1. Show the
the equality of Theorem 6.17 for n = 5 and n = 6.
6.
edge ei has been assigned a real number ri called its resistance.
resistance.
2. Draw K7 in the plane with 9 crossing.
crossing. If there is a potential difference pi between the endvertices of ei say a
and b, then an electrical current will flow in the edge ei from a to b
3. Determine (K2,2,2). according to the Ohm's
Ohm's law:
p
wi = i
ri
4. Determine (K1,2,3).
We will correspond to an electrical network a directed multigraph
where we orient each edge arbitrary from one endvertex to the other
so that we may use pi to denote the potential difference in the edge
ei, meaning the difference between the initial vertex and the
endvertex.

Graph Theory 6 53 Graph Theory 6 54

To solve the problem we need use the Kirchoff'

Kirchoff's laws:
We will denote by wi the current in the edge ei, meaning the current
in ei in the direction of the edge. Kirchoff'
Kirchoff's potential (or voltage) law state that the potential differen-
differen-
We regard a negative current wi as a positive current wi in the other ces round any cycle x1x2xk sum to 0:
direction. p x1 x2 + p x2 x3 + ... + p xk 1 xk + p xk x1 = 0
Kirchoff'
Kirchoff's current law postulates that the current outflow from any
In this section we will use ab to denote an edge from a to b. point is 0:
wab + wac + ... + wau + wa = 0
In many practical problems electrical currents are made to enter the Here ab,
ab, ac, ,au are the edges incident with a, and wa denotes the
network at some points and leave it at others, and we interested in the amount of current that leaves the network at a.
conseqent currents and potential differences in the edges.
In keeping with our convention, wa = - wa is the amount of current
entering the network at a.
So, for vertices not connected to external points we have
wab + wac + ... + wau = 0

Graph Theory 6 55 Graph Theory 6 56

If we know the resistances, then the potential law can be rewritten
rewritten as In the most fundamental problems current is only allowed to enter
enter
a restriction on the currents in the edges. So, the currents are the network at a single point s, called source,
source, and only leave it at
governed by the Kirchoff laws only; the phisical characteristics of a another point t, called sink.
sink.
network (the resistances) affect only the parameters in these laws.
laws.
If the size of the current from s to t is w and the potential difference
between s and t is p, then by Ohm'
Ohm's law r = p/w is the total resistance
The potential law is equivalent to saying that one can assign absolute
absolute of the network between s and t.
potentials Va, Vb, to the vertices a,b,
a,b, so that the potential differ-
differ-
s 1 s Vs
ence between a and b is Va Vb = pab.

Consequence: If the network is connected and the potential 5 4 e+f 1-e-f

differnces pab are given for the edges, then we are free to choose a 3 f
b a b Va=e Vb=2(1-
=2(1-e)
arbitrarily the potential of one of these vertices, say Va, but then all
the other potentials are determined. 1 2 e 1-e
1
t t Vt=0
resistances currents potentials
Graph Theory 6 57 Graph Theory 6 58

This network has 5 resistors, of values 1,2,3,4 and 5 ohms, as shown

So we get
in the first picture.
e = 2 2e + 3f
If we suppose that a unit current flow flows into the system at s and
leaves it at t, then the consequent edge currents must be as in the and
second picture, for suitable values of e and f. 6e + 5f = 6 6e 4f,
Finally, the potentials Vt=0, Va, Vb, Vs assigned to the vertices must giving
satisfy Ohm'
Ohm's law, so
e = 4/7, f = -2/21
Va = 1
1e = e, Vb = 2(1-
2(1-e) and Vs = Va + 5(e + f) = 6e + 5f.
and
Ohm's law has to be satisfied in more two edges, ab and bs, giving
Ohm'
us Vs = 6e + 5f = 62/21.
Va = e = Vb + 3f = 2(1-
2(1-e) + 3f In particular, the total resistance from s to t is
and (Vs Vt)/w = 62/21.
Vs = 6e + 5f = Vb + 4(1-
4(1-e-f) = 2(1-
2(1-e) + 4(1-
4(1-e-f).

Graph Theory 6 59 Graph Theory 6 60

 Theorem 7.1.: The principle of superposition implies that in case
case of
multiple sources and sinks there is at most one solution, no matter
matter
The Kirchoff'
Kirchoff's equations are linear and homogenous in all currents
how the sources and sinks are distributed.
and potential differences.
Proof
Consequence 1: The principle of superposition is valid for solutions.
solutions. Suppose we have two different solutions.
Consequence 2: Any current resulting from multiple sources and Because of Consequence 1, the difference of these solutions is also
also a
sinks can be obtained by superposing flows belonging to one source
source solution, and this flow no current enters or leaves the network at any
and one sink. point.
Suppose that in this flow there is a positive current in some edge
edge
from a to b.
Then, by the current law, a positive current must go from b to c,
then from c to d, etc., giving a trail abcd.
abcd.
Since the network is finite, this trail has to return to a point
previously visited.
Graph Theory 6 61 Graph Theory 6 62

So we obtain a circuit in whose edges positive currents flow in one

direction. s s
It implies that the potential of each vertex is strictly greater than that r2
of the next one round the circuit, and this is impossible.
a r1 r2
r1
t t
Before proving the existence of a solution (which is obvious if we
believe the physical interpretation) we shall calculate the total
total Vt = 0 Vt = 0
resistance of two networks.
Unless the networks are very small, the calculations can get very
very Resistors connected in series and in parallel.
heavy, and electrical engineers have a number standard tricks to
make them easier.
Let us put a current of size 1 through the networks from s to t. We
will calculate the total resistances in both cases.

Graph Theory 6 63 Graph Theory 6 64

In case of series connection: The conductance is the reciprocal of resistance.
Va = r1 and Vs = Va + r2 = r1 + r2
Theorem 7.2.: For series connection the resistances add and for
and so, the total resistance is parallel connection the conductances add.
r = r1 + r 2
How can we use convenient the conductances?
conductances?
In case of parallel connection we suppose that a current of size e
goes through the first resistorand so, a current of size 1 e through Let us examine certain limiting cases of Ohm'
Ohm's law:
the second. So If the resistance of an edge ab is 0, then we necessarily have Va=
Vs = r1e = r2(1 e) Vb, and from an electrical point of view the vertices can be
and so, regarded as identical. (We say sometimes that a has been
r2
e= shorted to b.) Of course, a may be shorted to b if there is some
r1 + r2 other reason why Va = Vb.
The total resistance is given by
If an edge has a conductance 0, then it does not have any effect to
rr 1 1 1
r= 12 or = + the currents and the potential. (We can produce such a situation if
r1 + r2 r r1 r2 we cut an edge.)
Graph Theory 6 65 Graph Theory 6 66

a 1 1
b 1 1
Now, we will show an example how can we use a possiblye shorting 1 s s 2 2
s c
of vertices to determine the total resistance of an electrical network.
network. 1 1
1
1 1 1 1 1 1
We take the network formed by the edges of a cube,cube, in which each d 2
edge has 1 ohm resistance. What is the total resistance across an edge e 1 1

st? t t 1 1 t 1 1
f 2 2
Considering the first picture we see that Va = Vc and Vd = Vf, so a
can be shirted to c and f can be shorted to d. So we get the second
1 1
picture. s 2
s 2 s s
Now, for all the parallel resistors we can use the conductance
1 5 7 7
addition rule,
rule, so we get the network at the next picture.
picture. 1 2 1 1
2 2 5 12

From now on we can simplify resistors connected parallel and in

t 1 t 1
series, until we find the total resistance is 7/12. t t
2 2

Graph Theory 6 67 Graph Theory 6 68

Another important device in practical calculations is the so-
so-called Of course, we may apply the transformation in reverse, replacing A',
star-
star-delta transformation: B', C'
C' by
A = B'
B'C'/T, B = C' C'A'/T and C = A' A'B'/T,
If vertex v is joined to just three vertices, say a, b and c, by edges of where
resistances A, B and C, the we call v the center of a star. T = A'
A' + B'
B' + C'
C'.
If no current is allowed to enter or leave at v, then we are allow to
replace this star by the delta configuration shown in the next picture: The formulae become symmetrical if we use resistances in the first
transformation and conductances in the second:
a a

A' = B + C BC/A and = '+ '+ ' '/





A
C'=S/C B'=S/B

where , , are the conductances.
conductances.


v
B C
b c A'=S/A The other formulae are symmetrical.
symmetrical.
b c

Graph Theory 6
S = AB + BC + CA 69 Graph Theory 6 70

I. Delta-
Delta-star transformation:
1
2 T = 1+1+2 = 4. A = 1.1/4 = B = 1.2 /4 = C = 1.2/4 =
5 1 5 3
5 2
1 1
5
4 5 5
4
II. Star-
Star-delta transformation:
4
S = AB + BC + CA = 5/4
A' = S/A = 5/4 2/3 = 5/6
B' = S/B = 5/4 4 =5
5
C' = S/C = 5/4 2 = 5/2
5
6 6

5 5
5 5 5
5
2 2 3 1 III. Use the Conductances Addition Rule:
2
5
5 6 5 5 25
4

rr 55 5 25 4 5
r= 12 = = r= 2 4 = 8 = =
r1 + r2 10 2 5 5 15
+ 8 15 6
2 4 4
Application of the star-
star-delta transformation
Graph Theory 6 71 Graph Theory 6 72
We are in doubt with the existence of a solution.
solution. Theorem 7.2.: Given an edge ab, denote by N(s, N(s, a, b, t) the number
of spanning trees of G in which the (unique) path from s to t contains
We will prove that if a current of size 1 is put throug a network then a and b, in this order. Define N(s,
N(s, b, a, t) analogously and write N
the current in an edge can be expressed in terms of the numbers of for the total number of spanning trees. Finally, let
certain spanning trees. wab = {N(s,
N(s, a, b, t) N(s,
N(s, b, a, t)}/N.
During the proof we assume that the graph G of the network is Distribute currents in the edges of G by sending a current of size wab
coneected, from a to b for every edge ab. Then there is a total current size 1
coneected, each edge has unit resistance, and a current of size enters
at a vertex s and leaves at t. from s to t satisfying the Kirchoff laws.
Proof
For each spanning tree T there is exactly one neighbour xT of s that is
on the st path PT contained in T. (otherwise, T would be a cycle!)
Hence
b ( s )
N ( s , s ,b ,t ) = N ,

where (s) is the set of neighbours of s.

Graph Theory 6 73 Graph Theory 6 74

If y is not the st path PT then T contributes nothing to N(s,x,y,t)

N(s,x,y,t) or

Since N(s,b,s,t)=0
N(s,b,s,t)=0 for every vertex b (s), we find that
b (s) sb=1.
w N(s,y,x,t).
N(s,y,x,t).





If y is on the st path PT, say PT = sxyzt, then T contributes 1

By symmetry wat = 1.


a (t)


to N(s,y,z,t)
N(s,y,z,t) and it also contributes 1 to N(s,x,y,t).
N(s,x,y,t). In other words T
Kirchoff's current law is satisfied at s and t provided a current
So, the Kirchoff'
contributes a current of size 1 from y and a current of size 1 into y.
of size 1 enters the network at s and leaves it at t.
To prove the theorem we have to show that if no current is allowed to So, the current law is satisfied.
enter or leave the network at any other point then the current and
potential laws are satisfied.
satisfied. Now we consider the potential law.
As all edges have the resistance, the potential law claims that the total
First, we consider the current law. current in a cycle with some orientation is zero.
To simplify the situation we multiply all currents by N. To show this, we first reformulate slightly the definition of N(s,a,bt).
N(s,a,bt).

What is the contribution of a spanning tree T to the current antering We call a forest F a thicket if it has exactly two components, say Fs,
and leaving a vertex y distinct from s and t? and Ft, such that s is in Fs and t is Ft.

Graph Theory 6 75 Graph Theory 6 76

Then N(s,a,b,t)
N(s,a,b,t) is the number of thickets F=F
F=Fs Ft for which a Fs The consequence is the following theorem:

and b Ft, and N(s,b,a,t)
N(s,b,a,t) is defined analogously.
Then the number of cycle edges from Fs to Ft minus the number of Theorem 7.3.: There is a distribution of currents satisfying Kirchoff'
Kirchoff's
cycle edges from Ft to Fs is the contribution of a thicket F=F
F=Fs Ft to laws in which a current of size 1 enters at s and leaves at t. The value

the current in a cycle. And this is zero. of the current in and edge ab is given by
{N*(s,a,b,t
N*(s,a,b,t)) N*(s,b,a,t
N*(s,b,a,t))} / N.
N.
It is easy to see that the above can be rewritten word for word to
give a solution in the case when the edges have arbitrary conduct
conduct- Corollary 7.4.: If the conductances of the edges are rational and a
ances:
ances: current of size 1 goes through the network then the current in each
For a spanning tree T define the weight w(T)
w(T) of T as the product of edge has rational value.
the conductances of its edges.
Let N* be the sum of the weights of all spanning trees, let N*(s,a,b,t)
s,a,b,t)
be the sum of the weights of all the spanning trees in which b
follows a on the st path in the tree, and let N*(s,b,a,t)
s,b,a,t) = N*(t,a,b,s).
t,a,b,s).

Graph Theory 6 77 Graph Theory 6 78

We cut this rectangle out of a sheet of nichrom (or any other material
7.2. Squaring the Square with low conductivity), and the rods are made of silver (or any other
high conductivity material).
The problem arising from the recreational mathematics: Is it possible
possible What happens if we ensure that the silver rod at the top is at 7 volts
subdivide a closed square into finetely many but at least two while the rod at the bottom is kept at 0?
square regions of district sizes that intersect only at their boundaries?
boundaries? In fact the potential at a point of the rectangle will depend on the
s s height of the point: the potential at height x will be x volts. So, there
Vs = 7
is no potential difference any two points at the same height.
a2 Therefore, there will be no current across the rectangle, only from from
4 a Va = 5
b 2 top to bottom.
b Vb = 3 Thus the current will not change at all if
3 3 we place silver rods on the horizontal sides of the squares and
cut narrow slits along the vertical sides, as we will show
show at the
Vt = 0
t t picture.
Graph Theory 6 79 Graph Theory 6 80
Since silver is a very good conductor, the points of each silver rod What is the potential drop in an edge?
have been shortened, so can be identified.
Thus as an electric conductor the whole rectangle behaves like the
the It is the side length of the corresponding square.
plane network shown in the second picture of the slide, in which the
conductance of an edge is equal to the conductance of the
corresponding square from top to bottom. What is the resistance of the whole system?

Clearly, the conductance of a rectangle from top to bottom is The ratio of the horizontal side of the original big rectangle to
to the
proportional to the length of a horizontal side and the resistance
resistance is vertical side, that is 7/6.
proportional to a vertical side.
Consequently, all squares have the same resistance, say unit Why have we done this procedure?
resistance, so all edges have unit resistance in the network model.
model.

Graph Theory 6 81 Graph Theory 6 82

The process is reversible: every squared rectangle can be obtained

obtained Many squared squares have been found:
from some network and so we have an effective tool to help us in our Sprague 1939.
search for squuared squares: Brooks, Smith, Stone and Tutte 1940.
Take a connected planar graph G and turn it into an electrical Duijvestijn 1975 with computer.
network by giving each edge resistance 1.
The connection between squaring a rectangle and electrical networks
networks
Calculate the total resistance from vertex s to a vertex t.
gives a nice result first proved by Dehni in 1903:
If this is also 1, the network may correspond to a suitably squared
square. Theorem 7.5 (Dehni
(Dehni,, 1903): If a rectangle can be tiled with squares
then the ration of two neighbouring sides of the rectangle is rational.
rational.
If the potential differences in the edges are distinct, all square
square have
different sizes, so we have a perfect squared square. Proof
The statement follows immediately from the Corollary 7.4 which
At this stage our problem is far from being solved. Why? states that if each edge has resistance 1 and a unit current flows
flows
through the system then in each edge the value of the current is
rational.
Graph Theory 6 83 Graph Theory 6 84
 7.3. Vector Spaces and Matrices Associated with Graphs
35 27
50
8
11
19 Ezt majd mg utlag,
utlag, ha lesz id r
15 17
2 6
9 7
18 24
29 25 16
4

33 37 42

A perfect squared square:

a tiling of a square with 21 incongruent squares
Graph Theory 6 85 Graph Theory 6 86

Exercises.
Exercises.

1. Consider a simple network,

network, and we give the potential differences
pab for all edges. Give the other potentials if Va = 0 and if Va = 3.

Graph Theory 6 87 Graph Theory 6 88