APPLICATIONS OF NON-NEGATIVE MATRICES

A Project submitted to
A Veeriya Vandayar Memorial Sri Pushpam College (Autonomous), Poondi
(Affiliated to Bharathidasan University, Thiruchirapalli)
In partial fulfilment of the requirements for the award of the Degree of

MASTER OF SCIENCE
In
MATHEMATICS

Submitted by
K.BHARATHI
(Reg.no: 18MAB402)

Under the Supervision of

Dr.R.SANGEETHA, M.Sc., M.Phil., PhD.,

P.G. & RESEARCH DEPARTMENT OF MATHEMATICS

A.V.V.M SRI PUSHPAM COLLEGE (AUTONOMOUS)
POONDI-613 503,
THANJAVUR DIST.

SEPTEMBER-2020

i
 PG & Research Department of Mathematics
A.V.V.M SRI PUSHPAM COLLEGE
(Autonomous)
Poondi – 613 503, Thanjavur district,
Tamilnadu.

Dr.R.SANGEETHA, M.Sc., M.Phil., PhD., Place: Poondi

Date:

CERTIFICATE

This is to certify that the project entitled “APPLICATIONS OF NON-

NEGATIVE MATRICES” is a bonafide record of research work done by
K.BHARATHI (Reg. No:18MAB402) submitted to A.V.V.M Sri
Pushpam College (Autonomous), Poondi in partial fulfilment of the
requirements for the award of the degree of Master of Science in
Mathematics during the year 2018-2020 in the department of
Mathematics under my guidance.

If further certify that the project represents independent work in the part
of the candidate.

Signature of the Coordinator Signature of the Guide

(G. MARIMUTHU) (R.SANGEETHA)

ii
 A.V.V.M SRI PUSHPAM COLLEGE (AUTONOMOUS),
POONDI-613 503, THANJAVUR DISTRICT
Affiliated to BHARATHIDASAN UNIVERSITY, Thiruchirapalli-24

MASTER OF SCIENCE
In
MATHEMATICS

Viva-Voce Exam conducted on

Internal Examiner External Examiner

iii
 DECLARATION

I do here by declare that this project work has been originally

carried out by me at the Department of Mathematics, A.V.V.M Sri

Pushpam College (Autonomous), Poondi, Thanjavur during the

academic year 2018-2020 under the guidance of

Dr.R.SANGEETHA, M.Sc., M.Phil., PhD., and this work has

not been submitted elsewhere for any other degree.

Place: Poondi Signature of the Candidate

Date: (K.BHARATHI)

iv
 ACKNOWLEDGEMENT

First and foremost, I thank to the ALMIGHTY GOD for his presences in me to do
this work with fulfil manner.

I am grateful to the MANAGEMENT of A.V.V.M Sri Pushpam College

(Autonomous), Poondi, Thanjavur, for having given me an opportunity to do my PG
course.

My sincere thanks to Dr. R. SIVAKUMAR, Principal of

A.V.V.M Sri Pushpam College (Autonomous), Poondi, Thanjavur District, for
providing me an opportunity to do this course.

I wish to express my sincere and heartfelt thanks to my guide

Dr.R.SANGEETHA, and the coordinator Dr. G. MARIMUTHU, Associate Professor
of Mathematics in A.V.V.M Sri Pushpam College (Autonomous), Poondi, for having
suggested the topic of my project and taken dedicated and consistent interest in the
preparation of the project.

I express my sincere thanks to the staff members of the Department of

Mathematics A.V.V.M Sri Pushpam College (Autonomous), Poondi, for their help
throughout the M.Sc., course. I wish to express my thanks to my beloved PARENTS,
FAMILY MEMBERS and FRIENDS.

K.BHARATHI

v
 CONTENT
APPLICATIONS OF NON - NEGATIVE MATRICES

Chapters Title Page no.

Introduction 1

I Basic definition 3

II Non negative matrices 13

III Adjacency matrix of graphs 22

IV Directed graph of a matrix 32

Conclusion 48

Bibliography 49

VI
 APPLICATIONS OF NON - NEGATIVE MATRICES

INTRODUCTION

Graph theory is one of the most important and efficient branches in

Mathematics; since it is applied in all areas of science it may be regarded as an
applied area of Mathematics.

Graph theory as a mathematical discipline was created by Leonard Euler in the

year 1736 in his famous discussion of Konigsberg Bridge problem. Thus he
becomes the father of Graph theory.

There are applications of Graph theory to some areas of Physics, Chemistry,

Computer Science, Communication Network, Operations Research, Genetics and
Economics. The theory is intimately related to many branches of Mathematics,
including Group theory, Matrix theory, Numerical Analysis, Probability and
Topology.

In Chapter-I we confine our attention to basic definition of Graphs such as

Sub graph, Super Graph, Bipartite graph, Complete graph, Labelled graph
Directed, Regular.

In Chapter – II we contains the thrust of a non negative matrices with of

some important well know Theorem, Proposition, Lemma and some illustration
which are related to the Graph spectra.

1
 In Chapter – III, we present the Adjacency matix of graph with the property
of irreducible, symmetric and characteristics polynomial, perfect matching.

In Chapter – IV, it deals with Matrices over the 2-element Boolean Algebra
with Semirings, Directed graph of matices, The pattern Homomorphism of
Irreducibility and Primitivity, graphical interpretation of Irreducibility, Matrices
over antinegative, Zero- divisor semirings etc..

2
 –

BASIC DEFINITIONS OF GRAPHS

Definition: 1.1

A graph G consists of pair (V(G), X(G)) where V(G) is a non-empty finite

set whose elements are called points or vertices and X(G) is set of unordered pair
of distinct elements of V(G).

Example:

Definition: 1.2

A graph is infinite if it has infinitely many vertices or edges or both;

otherwise the graph is finite. An infinite graph where every vertex has finite degree
is called locally finite.

3
Definition: 1.3

A graph G is called labelled if its n points are distinguished from one

another by names such as v1 ,v2 ,...vn.

Example:

Definition: 1.4

A graph G is said to be connected if every pair of its points are connected.

Example:

4
Definition:1.5

A graph G in which any two distinct points are adjacent is called complete
graph. The complete graph with p points is denoted by Kp.

Example:

Definition: 1.6

A loop is an edge whose end points are equal.

Example:

5
Definition: 1.7

A simple graph having no loops or multiple edges.

Example:

Definition: 1.8

The degree of a point v in a graph G is the number of edges incident with v .

It is denoted by dg(v) or simply d(v).

Example:

6
Definition: 1.9

A regular graph is a graph where each vertex has the same number of
neighbours, every vertex has the same degree or valence. A regular graph with
vertices of degree k is called a k-regular graph or regular graph of degree k.

Example:

Definition: 1.10

A graph G is called a bigraph or bipartite graph if V can be partitioned into

two disjoint subsets V1 and V2 such that every line of G joins a point V1 to a point
of V2. (V1, V2) is called bipartition of G.

Example:

7
Definition: 1.11

A graph G is said to be complete bipartite graph if the graph contains every

line joining a point of V1 to a points of V2.

Example:

If V1 contains m points and V2 contains n points then the complete graph G

is denoted by Km,n.

Definition: 1.12

A path is a simple graph whose vertices can be ordered so that two vertices
are adjacent if and only if they are consecutive in the list.

8
Example:

Definition: 1.13

A cycle is a graph whose vertices can be ordered so that two vertices can be
placed around a circle so that two vertices are adjacent if and only if they appear
consecutively along the circle.

Example: C3

9
Definition: 1.14

A simple undirected graph with three vertices and three edges. Each vertex
has degree two, so this is a regular graph. An undirected graph is one in which
edges have no orientation. The edge (a,b) is identical to the edge (b,a), they are not
ordered pairs, but sets {u,v} of vertices.

Definition: 1.15

A graph that contains no cycles is called an acyclic graph.

Definition 1.16

A connected acyclic graph is called a tree.

Example:

10
Definition: 1.17

Any graph without cycles is called a forest.

Definition 1.18

A wheel graph Wn is a graph with n vertices (n≥4), formed by connecting a

single vertex to all vertices of an (n-1) cycle.

Example: W5

Definition: 1.19

The graph K1,m is called star graph for m ≥ 1.

11
Example:K1, 3

Definition: 1.20

If the vertices of the graph are assigned values subject to certain condition(s)
then it is known as graph labeling.

12
 CHAPTER-II

NONNEGATIVE MATRICES

Anxn= [aij] is called nonnegative (positive) if all aij ≥ 0 ( a> 0 ). We use the
notation A ≥ 0 (A > 0) to denote a nonnegative (positive) matrix
Let Anxn ≥ 0. The directed graph Γ (A) = ( V,E ) associated with A is the
graph with V = { 1,2,…,n } and E = { (u,v) : auv> 0 }.
A directed graph is called strongly connected if there is a directed path
between every pair of (distinct) vertices.
A nonnegative matrix Anxn, n ≥ 2 is said to be irreducible if the associated
graph Γ (A) is strongly connected, otherwise it is called reducible. The matrix 01x1
is defined as reducible.

Below we have a matrix A and its associated directed graph.

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

Observe that Γ (A) being a directed cycle on 4 vertices is strongly

connected. Thus A is irreducible. Below we show the matrix A2 and its associated
directed graph. Notice that A2 is reducible

13
 0 0 1 0
0 0 0 1
1 0 0 0
0 1 0 0

Theorem: 2.1

Let A ≥ 0 be reducible. Then there is a permutation matrix P such that

P’AP = where B and D are square matrices.

Proof:

Suppose that u is not reducible (via a directed path) from v in Γ (A).

Let [u] = { w : u is reachable from w }.

Notice that A( [u]c, [u] ) is a zero matrix.

If [u] = { i1,…,ik } and [u]c = { ik+1,…,in }, then define P( ij,j) = 1.

Then

(Pt A P)k+1,r = Pt(k+l, : ) AP(:,r) = A(ik+1, :) P(:,r) = A(ik+l , ir) = 0,

if l ≥ 1 and r ≤ k.

Proposition: 2.2

Let Anxn≥ 0 and k . Then the (I,j) – entry (Ak)ij of Ak is positive ⇔there is a
directed walk of length k from I to j in Γ(A).

14
Proof:

Note that, for k ≥ 2,

a ij …….
,…,

So (Ak)ij> 0 ⇔ ……. 0 for some i1,i2, … ik-1, that is,

[ i, i1, …. Ik-1, j ] is a directed walk of length k from i to j.

Proposition: 2.3

Let Anxn≥ 0, n ≥ 0. Then A is irreducible ⇔ (I+A)n-1> 0.

Proof :

Let A be irreducible.

Then Γ (A) is strongly connected and Γ( I + A ) is obtained from Γ(A) by

adding a loop at each vertex.

Let P( I,j ) = [ i, i1,… ik-1, j ] be a directed path of length k in Γ(A).

Then

[ i, i1, …, ik-1, j,…,j ]

Is a i-j directed walk of length n – 1 in Γ( I + A ).

Hence ( I + A )n-1 > 0.

Conversely, if { ( I + A )n-1 }ij> 0, then there is a i – j directed walk of length

n – 1 in Γ( I + A ).

By dropping the cycles we obtain a i – j directed path in Γ(A). So Γ(A) is

strongly connected, that is, A is irreducible.

15
Lemma: 2.4

Let Anxn ≥ 0 be irreducible. Then there is a positive eigenvalue of A with

the corresponding eigenvector positive (entrywise).

Proof:

Let S = { x n
: x ≥ 0, ∥x∥1 : |xi| = 1 } be the set of probability vectors.

Let f(x) = Ax / ∥ Ax ∥1.

Then f is a continuous function S to S. Notice that S is convex, compact.

Hence by Schauder fixed point theorem, there is a fixed point y of f.

Put =∥ ∥1.

Then Ay = y and ( I + A)n-1y = ( 1 + )n-1y.

Since A is irreducible,

( I+A )n-1 is positive and so y is positive.

Lemma: 2.5

In the eigenvalue is geometrically simple.

Proof:

Let y > 0 be an eigenvalue of A corresponding to .

If possible let there be another eigenvector,

Az = z, z = z1 + iz2.

Since A and are real we see that Az1 = z1 and Az2 = z2.

16
 If z1 is not a scalar multiple of y, then we can take a suitable linear
combination of y and z1 to produce a nonnegative eigenvector x of A
corresponding to , with one entry zero.

Since ( I+A)n-1 x = (1+ )n-1x we see that x must be positive, a

contradiction. Hence z1 is a scalar multiple of y.

Similarly, z2 is a scalar multiple of y.

Lemma: 2.6

Let Anxn ≥ 0 be irreducible. Then there is a positive eigenvalue of A with

the corresponding eigen vector positive (entrywise), is the only eigenvalue of A
with a positive eigenvector.

Proof:

Let (A) have a positive eigenvector z.

Using lemma, we get a eigenvalue > 0 of At with a positive eigenvector x. So

xty = xtAy = xty and xtz = xtAz = xtz.

We get .

The positive eigenvalue and a positive vector y as described in lemma 5 are

called the perron value and a perron vector of A, respectively. A left – parron
vector is a perron vector of At.

Lemma 2.7

Let A ≥ 0 be irreducible. Then (A) is a perron value of A.

17
Proof :

Let |s| = (A), Ax = sx, x ≠ 0, the perron value of A and z a left-perron

vector, then

. .
| | | | | | . .
. .

| | ……… 1

Hence (A)zt|x| ≤ zt|x| and so ≤ , an eigenvalue of A. Thus

(A) = .

Recall that for any square matrix B we have

det B = ∑ [ …… ……………. 2

where Sn is the group of all permutations of 1,…,n.

Theorem 2.8

Let A ≥ 0 be irreducible. Then (A) is simple.

Proof:

Consider the characteristics polynomial

P( = det( I – A) = det B = ∑ [ ……

Then the derivative

18
 ∑ …… + … … …
…… …+ ∑ …… .

1|1 ⋯….. |

Where B (i/i) is the principle sub matrix of B after deleting row i and
column i.

Hence,

det 1|1 ⋯ . . det | … 3

Assume that is not simple. Then tr(adj( I – A)= P'( ) = 0.

As we know that has geometric multiplecity one, it follows that the Jordan
matrix for A has only one Jordan block corresponding to .

Hence ( I – A) has rank n-1. LetC =adj( I-A). Then,

det 0

It follows that column of C are either 0 or nonzero multiples of the perron

vector y. As the rank of, I – A) is n-1, it has a nonsingular principal submatrix of
order n-1, that is, at least one element of C is non-zero.

Hence C has a column of the form ry, r ≠ 0, as C is real, r is real. Consider

At. Proceeding as before for ( I – At), we see that C has a row sz, s ≠ 0, s ,
where z is the perron vector for At. It follows that r and s have the same sign and
thus C > 0. So tr(C) ≠ 0, a contradiction.

19
Theorem: 2.9

Let A ≥ 0 be irreducible. Let ∈ , ∈ 0,2 with a

corresponding eigenvector x. then is simple and that is
remains unchanged with a rotation by an angle .

Proof:

Notice that | | | || | | | | | | | | || | | |, so that

| | | | 0.

Let z be a left perron vector of A. then | | | | 0. So that

| | | |.

As simple, we may assume that | | , a perron vector.

Let and put D = diag( ……… so that .

Then,

; ≔ | | .

Note that for each i , ∑ ∑ , so that , j=1,………n have the

same angular part(lie on the same ray starting from origin).

Notice that ∑ also lies on this ray. As 0 , the ray must

be the positive x-axis. Hence 0 and so

| | ,
……. 4

Note that multiplicity of in is the same as that of in , which is

the same as the multiplicity of in Which is 1.

20
Corollary: 2.10

Let 0 be irreducible and suppose that A has k>1 eigenvalues of

absolute value . then these are , 0, 1 , … … 1. further,

remains unchanged under a rotation of .

Lemma: 2.11

Let A > 0. Then the perron value is the only eigenvalue such that | |= .

Proof:

Let y be a perron vector.

Suppose that Ax = x and | | = .

As |x| = |Ax|=≤ A|x|, we have A|x|- x ≥ 0.

Choosing a left perron vector z of A we have zt(A|x|- |x|) = 0 and hence

| | | |

As is simple, we may assume that y = |x|.

Further, as | aijxj| = aij|xj|, we conclude that for each i the numbers aijxj,

j =1,…,n, have the same angular part(lie on one ray from the origin).

Hence x ye , 0.

Now ,

21
 CHAPTER - III

ADJACENCY MATRIX OF GRAPHS

Let G = (V, E) be a simple (undirected) graph on vertices 1,…,n. the

adjacency matrix A (G) (or simply A) is described by

, ∈ ,

The adjacency matrices of the 4- cycle C4 and the path P4 on 4 vertices are
given below.

0 1 0 1 0 1 0 0
1 0 1 0 1 0 1 0
,
0 1 0 1 0 1 0 1
1 0 1 0 0 0 1 0

Notice that G is connected ⟺ A (G) is irreducible. As A (G) is symmetric,

the eigenvalues of A(G) (we sometimes call them ‘eigenvalues of G’ ) are real
numbers.

The characteristics polynomial (G) of G is the polynomial det (xI-A(G)), in

a variable x.

Lemma: 3.1

Let G be a graph ∑ . Then a0 = 1, a1 = 0, and a2 = -

|E(G)|. In general,

ak = (-1)k ∑

where the sum is taken over all principal sub matrices of A(G) of size k.

22
Proof:

Let B = xI-A. it easy to see that in the expansion of det B the coefficient of
xn is 1.

The coefficient of xn-1 is determined by permutation of for which ,

at exactly n-1 places.

That is, i = (i) at n-1 places. In this case is the identity permutation,
which cannot give xn-1.

Thus coefficient of xn-1 = 0.

Proceeding in a similar way, let be a permutation at n-2 places, say

at 1,…,n-2 then makes a nonzero contribution to the coefficient of xn-2iff the
vertices n-1 and n are adjacent.

In this case = (n-1, n) and sgn( ) = -1. Each such corresponds to an edge
E uniquely and either contribution to a2 is -1.

Thus a2 = - |E|.

With a similar argument we have that the coefficient of

xn-k is ak = (-1)k∑ det where the sum is taken over all principal
submatrices of A of size k.

Corollary: 3.2

Let T be a tree. Then A = A(T) is invertible iff T has a perfect matching.

Proof:

If t is invertible, that T has a linear sub graph of order n. But then it has to be
a perfect matching.

23
 Conversely, if T has a perfect matching, then that matching a linear sub
graph of order n and the contribution to det A is (-1)n/2.

So det A = (-1)n/2b, where b is the number of perfect matching of T. Thus

A(T) is invertible.

Corollary: 3.3

Let G be a bipartite graph and ∑ . Then a2i+1=0, for

i=0, 1,… furthermore, if G =T, a tree, for i ≥ 1, a2i = (-1)i x the number of i –
matching in T.

Proof:

Notice that G does not have a linear subgraph of order 2i+1.

Using Sach’s formula we see that a2i+1 = 0.

For T, again using Sach’s formula, we have

∑ 1 2 ∑ 1
| |

Theorem: 3.4

Let G be a bipartite graph with parts P and Q. Then (A) = - (A). Thus is
an eigenvector iff – is an eigenvalue with the same multiplicity

Proof:

Let x be an eigenvector of A corresponding to .

24
With permutation similarity we have

0
0

Where the partition is made with respect to P and Q.

It follows that

0
0

Thus – is an eigenvalue of A. If x1,…,xk are linearly independent

eigenvector of A corresponding to , then ,……. , are linearly independent
eigenvector of A corresponding to – . Thus the multiplicity of and – are the
same.

Lemma: 3.5

Let G be a non-bipartite and (G) = ∑ i x n-i. Then a2i+1 ≠ 0, for some i.

Proof:

Let C = [1,2,…,2i+1,1] be an odd cycle of shortest length.

Then by Sach’s formula a2i+1 = -2 x the number of such cycles.

Theorem: 3.6

Let G be connected with the negative property: ⇔ .

(Here we do not assume that the have the same multiplicity). Then G is bipartite.

Proof:

Since A is irreducible, we know that and – are simple eigenvalue.

Furthermore, (A) is unchanged under rotation by 180,
25
 (i.e), - (A) = (A).

Thus the multiplicity of any eigenvalues is the same as the multiplicity of – .

Hence

(G) ≃xk∏ x2 – 2
)≃∑ Ix
n-i

has each odd coefficient zero.

Hence G is bipartite.

Theorem: 3.7

A graph G is bipartite iff it has strict negative property:

(A) ⟺ - (A) with multiplicity.

Proof:

If G is bipartite, then by theorem, it has strict negative property.

Conversely, suppose G has strict negative property.

Then (G) has odd coefficients zero.

By corollary, G is bipartite.

Theorem: 3.8

Consider the graph G = H K1. Then (A(G)) ⇔ -1/ (A(G) with

multiplicity. Thus if H is bipartite, then G has (SR).

Proof:

Put n = |H|. Notice that

26
 Let 1,… k be the eigenvalues of A(H) corresponding to the eigenvectors
x1,…,xn respectively, where the set { x1,…,xn} is orthonormal.

Then the vectors

2 , 2 ,….. 2 , 2
4 4 4 4

Are all eigenvectors of A(G) corresponding to the eigenvalues respectively.

4 4 4 4
, ,….. ,
2 2 2 2

We observe that the first conclusion follows.

Note that if H is bipartite, then G is bipartite. Thus if (G), then by the

above -1/ (G) and as G is bipartite 1/ (G).

Lemma: 3.9

Let G be a graph with (SR) of order n and P(x) = (G,x) be the

characteristics polynomial of A(G). Then |ar| = |an-r|.

Proof:

Since G has a property (SR), G is nonsingular.

Moreover P(x) and xnP(1/x) have the same roots.

Since P(x) is monic and the leading coefficient of xnP(1/x) is 1, it follows that
P(x) = xnP(1/x) and the conclusion follows.

27
Example

It is not true that every graph with (SR) is a corona. Consider the following
graph G

We have 7 7 1. so G has (SR)

Example

It is not even true that every graph with (SR) is bipartite. Consider the
following graph G

28
 It can be seen that 11 2 24 2 11 1
Thus G has (SR)

Theorem: 3.10

Let T be a tree on n vertices. Then T has a property (SR) ⟺ T = T1 K1, for

some treeT1.

Proof:

Let T have property (SR).

Then n = 2k, for some k.

If k = 1,2, then the only nonsingular trees of size 2k are the paths which are
K1 K1, K2 K1.

Assume that k ≥ 3. Further, T has a prefect matching, say, = { f i = u i v i, i

= 1,…,k }.

It is enough to show that for each i at least one of ui, vi is a pendant vertex.
Suppose this is not the case.

Thus there is an edge, say f2, such that u2,v2both have a degrees greater than
1, say u1 u2, v2 v3 are present.

29
 We should have that 2k -1 = |a2| = |a2k-2| = number of ( k-1) matching
in T.

Observe that - { fi }, i = 1,2,…,k are k many (k-1) matchings.

Also each edge e ∉ is adjacent to exactly two fi, fj . Thus - { fi,fj } +

{ e } is a (k-1) matching. There are k-1 many like this.

Further, - { f1,f2,f3} + { u1 u2, v2 v3 } is also a ( k-1 ) matching. We have

contradiction.

Lemma: 3.11

Let G be bipartite and connected of order n with a unique perfect matching

. Define B = [ bi,j ] as

1 ,

, ∈ , , ∈ ,
0,

Then B = A-1

Proof:

We use K ~ i to mean that k is adjacent to i.

Note that ( AB )ii = ∑ ikbki = 1, as there is exactly one matching edge

incident with i.

So let i≠j. If no alternating path exists from each k~i to j, then obviously
(AB)ii = ∑ ~ =0

30
 Suppose that i has a neighbour k such that there is an alternating path

P = [ k, i1,i2,…,j ] from k to j. If [k,i] , then i1 = i, and q = [ i2,i3,…, j] is also an

alternating path.

These are only vertices from which there is an alternating path going to j.
Together they contribute zero to

If [k,i] ∉ , ∃ a r such that [r,i] ∉ .

Then Q = [ r,i,i1,i2,…,j ] is also an alternating path. These are the only

vertices from which there is an alternating path going to j. Together the contribute
zero to

If T is a tree with perfect matching, then B = A(T)-1 is similar to |B| and it is

the adjacency matrix of a graph, called the inverse T-1 of T.

31
 CHAPTER- IV

Directed Graphs of A Matrix

SEMIRINGS:

A semi-ring is a binary system (S, +, x) such that (S, +x) is an abelian

monoid (identity 0), (S, x) is a monoid (identity 1), x distributes over +,

0xs = sx0 = 0 for all s in S, and 1 0.

Usually S denotes the system and x is denoted by juxtaposition. if (s,x) is

abelian then s is anti-negative. All rings with unity are semi-rings but no such ring
is anti-negative.

Algebraic terms such as unit and zero divisor are defined for semi-rings as
for rings. Algebraic operations on matrices over a semi ring and such notions as
linearity and invertibility are also defined as if the underlying scalars were in a
field.

If S is not commutative, unless otherwise noted, we will take the operation

of multiplication by scalars to be left multiplication :(s, A) → sA.

Here are some examples of semi rings which occur in combinatorics. They
are all commutative and anti-negative. Let H be a Boolean algebra, then (H ∪,∩ ) is
a semi ring.

Let C be any chain with lower bound 0 and upper bound 1, then
(C, max, min) is a semi ring (a chain semi ring). In particular, if F the real interval
[0,1] , then (F, max, min) is a semi ring, the fuzzy semi ring. The nonnegative
members of any sub ring of the reals ( under real addition and multiplication) is a

32
semi ring. In particular, the non-negative reals and the non-negative integers are
semi rings.

Let (s) denote the n matrices over S. The n identity matrix, ,

and the m zero matrix, 0 , , are defined as if S were a field. The n matrix
all of whose entries are zero except its (i, j)th which is 1, is denoted . We call
a cell; it is a link if i and a loop if i = j. We denote the m matrix all of
those entries are 1 by , .

Linearity of transformation is defined as for vector spaces over fields.

A linear transformation on M is completely determined by its behavior on the set
of cells in M.

The directed graph of a matrix

Suppose A∈ the directed graph D(A) associated with A has n vertices

, …, with being an arc if and only if 0.

A family F of matrices in M is graphical if for all A ∈ , all the matrices

whose directed graphs are isomorphic to D(A) are also in F. Thus, membership in
F depends upon the properties of the di (rected) graph of a matrix. For example, an
n n non negative matrix is irreducible if and only if its digraph is strongly
connected. So the irreducible if and only if its digraphs is strongly connected. So
the irreducible n n non negative matrices are a graphical family. The index
imprimitivity of such a matrix is the greatest common divisor of the cycle of its
digraph, an isomorphism-invarient. Therefore the family of all irreducible n n non
negative matrices with index of imprimitivity k, is graphical for each k.

33
 In describing the digraph of a matrix A, we will often disregard its isolated
vertices and the limit our description to the digraph induced by the arcs of D(A).
For example, if E is a link, then D (E + Et) really consists of a digon (i.e. an
oriented 2-cycle) and n-2 isolated vertices, but we will refer to it as a “digon”. We
also refer E + Et as digon-matrix. An s-star matrix is the sum of s links all in the
same line ( row or column). For example, ∑ is a 3-star matrix. Its digraph
is (except for isolated vertices) an orientation of the complete bipartite graph K1,3
with v1 as a source v2,v3, and v4 as sinks.

As another example, we refer to D(E12 + E32) and D(E12 + E23) as

orientations of a 2-path, ignoring the n-3 isolated vertices. The 2-path in question
is the (undirected) graph with vertex-set {v1, v2, v3} and two edges v1v2, v2v3; a
replica of K1,2.

The Pattern Homomorphism; Irreducibility And Primitivity

The pattern ̅, of a matrix A in Mn(S) is the (0,1) matrix whose (i, j)th entry
is 0 if and only if ai,j = 0. We will also assume that ̅ is in Mn(B) where B is the
Boolean algebra of two elements 0 and 1. Here, the rules of arithmetic are the
same as if B were a ring except 1 + 1 = 1.

The mapping accomplished by associating each matrix, A, in M(S) with its

pattern, ̅, in M(B) is a semiring homomorphism when S is anti-negative and zero-
divisor-free.

34
 f T is a linear operator on M(S), let its pattern, be the operator on M(B)
defined by = for all (i, j). Then ̅ for all A in M(S).

Equality holds in above inequality. If S is an anti-negative semi ring having

no zero divisors. Therefore we have the following lemma.

Lemma: 4.1

The mapping T → is a homomorphism of the semi ring of a linear

operators on M(S) onto the semi ring of linear operators on M(B) if S is anti
negative and free of zero-divisors.

A matrix M in Mn(S) is reducible if n 1 and for some permutation matrix

P, P-1 MP = where X is m m and 1 m n. Otherwise M is

0
irreducible.

It is clear that if M is the Mn(B) and n 1, then M is irreducible if and only

if there exists a permutation matrix P and a positive integer d = d(A) such that

0 0
P AP = 0
-1 0
0 0

Where for 1 i , the i-th zero block on the diagonal is , and for
some integer q, P-1 AdqP is the direct sum ⋃ .

(i.e) A is primitive.

We denote the family of all irreducible matrices in Mn(B) of index of

imprimitivity k by Fk(B).

If k 1, then no member of Fk(B) has a diagonal entry equal to 1, by the

matix. Therefore, Fk(B) ⊆ K when k 1.

35
Graphical interpretation of irreducibility.

Suppose D = D(A) is the directed graph of the (0, 1) matrix of A. Then D is

strongly connected if and only if A is irreducible. Now suppose A is irreducible
and d = d(A), the index of imprimitivity of A. Then d is the greatest common
divisor of the lengths of the (directed) cycles in D.

Moreover the vertex-set of D is the union of d mutually disjoint non-empty

sets V1, V2,…,Vd with the following properties. If uv is an arbitrary arc of D from
vertex u to vertex v, then some index i with 1 i d, u is a vertex Vi and v is a
vertex of Vi+1 (here, Vd+1 = V1). The sets Vi are called cyclically transferring
classes.

Terminology

The number of non zero entries in a matrix is denoted by |A|. The number
elements in a set is denoted |Ѱ| too. We trust this will cause no confusion.

We denote the Schur product of A and B in M by A•B. That is C = A•B if

and only if Cij = aijbij for all i and j.

When X and Y are in Mn,n(B) we define X\Y to be the matrix Z such that
Zi j = 1 if and only if Xi j = 1 and Yi j = 0. For example, the matrix in Mn,n(S) having
all off diagonal entries 1 and all diagonal entries 0 is denoted Kn, or just K when n
is understood. When S = B, K = J\I. We also denote {X ∈ Mn, n(S): X K} by K.
Thus K is the set of all matrices in M with all diagonal entries equal to zero.

If A and B are in M, we say B covers A (written B A or A B) if bi j = 0

implies ai j = 0 for all I,j. This provides a reflexive, transitive relation on M.

36
 Suppose F is a family of matrices in M. We say F covers a matrix A in M, if
A is covered by some member of F. A member X of F is critical if X covers no
other member of F. We say that F covers a family of matrices G selectively if
whenever a member of G is covered by F, it is covered by a critical member of F.

Theorem: 4.2

Any linear operator on K that is bijective on the links and preserves 2-star
matrices is in the group of operators on K generated by transposition and the
permutation similarly operators.

Theorem: 4.3

Suppose F is a finally of matrices in K that covers the set of links and

selectively covers the set of matrices having at most 2 noon zero entries, then any
linear operator on K that strongly preserves F is bijective on the links.

We will show that Fk satisfies the hypothesis when 2 k n, and that

unless k = 2and n = 3, any linear operator strongly preserving Fk preserves 2-star
matrices.

Proof:

Let L be a linear operator on K that strongly preserves F. There is a positive

integer p for which Lp is idempotent.

Since K is a finite set: let Q = Lp. Then Q, as a power L, also strongly

preserves F.

First, we will show that Q is nonsingular,

(i.e.) that Q(X) = 0 only if X = 0.

Suppose Q(X) = 0 and there is a link E .

37
 Since F covers the links, selectivity implies that some critical M in F covers
E. Then Q(M) = Q(M\E) because q is linear and Q(E) = 0.

But that contradicts the fact that Q strongly preserves F, because M\E ∄ F.

Since M is a critical member of F.

Next we will show that Q fixes each link. Let E be any link, then for some
link, Q(E) F because Q is non singular.

Suppose F E and let H = E + F. We have E M for some member and

hence some critical member M of F.

Then, Q(M) = Q(M + E) = Q(M) + Q2(E) Q(M + F) Q(M) because Q is

idempotent and Q(E) F.

But then Q(M) = Q(M + F), so M + F ∈ F because Q is preserves F strongly.

Therefore H is covered by F. By selectivity, H C for some critical member
C ∈ F.

Let N = C\F, then N ∄ F and

Q(C) Q(N) = Q(N + E) = Q(N) + Q2(E) Q(N +F)

Therefore Q(N) = Q(C), contradicting the fact that Q preserves F strongly.

Therefore F = E and hence Q(E) = E.

Finally we will show that L is bijective on the links. Let E be any link, then
L(E) 0 because Q is linear and non singular.

Therefore L(E) F for some link F and hence E = Lp(E) Lp-1(F). But
Lp-1(F) 0 because Q is non singular.

38
 Consequently E = Lp-1(F) and hence L(E) = F, a link. If L(G) = L(E) for
some link G, then Q(G) = Q(E) and hence G = E.

Lemma: 4.4

For all 2 k n, Fk covers the links and selectively covers K2, the set of
matrices in K having at most two nonzero entries. Specifically, F2 covers each A
in K2 and, when k 2, Fk covers each A in K2 except for the digon-matrices.

Proof:

We will define a family G(n, k) of strong digraphs on n vertices with k

cyclically transferring classes; each digaph has the property that the removal of any
of its arcs produces a digraph that is not strong.

Thus their matrices will be critical members of Fk.

Case 1

If n k 2, the diagram in the figure below describes a specific member

D1(n, k) of G(n, k); the parameter p = n-k+1. The k cyclically transferring classes
for the digraph D1(n, k) are:

V1 = {1, 2, …, p}, V2 = {p+1},… V1 = {p + i -1},…Vk-1 = {n-1}, Vk = {n}.

The rest of the members of G(n, k) with k 2 are obtained by relabeling the
vertices of D1(n, k).

Every arc, every orientation of a 2-path, and every independent pair of arcs
can be embedded in a member of G(n, k) by appropriately relabeling the vertices of
the digraph of this figure.

Since Fk covers no digon-matrix when k 2, and the matrix of each member

G(n, k) is a critical member of Fk, the lemma follows.

39
 Notice that when k = 2, the digraph D1(n, 2) of this figure is a union of n-1
digons with common vertex {n}. If n 4, F2 covers matrices whose digraphs
contain independent pairs of arcs.

For those n, D1(n, 2) and hence each of its vertex-relabeling, contains no

independent pair of arcs and so they are unsuitable for use in proving this lemma.
For that reason we use a different class of digraphs to define G (n, k) when k = 2.

Case 2

If k = 2, For odd n 3, it describes a specific member Da(n, 2) of G(n, 2).

The two cyclically transferring classes for the digraph Da(n, 2) are :

W1 = {2, 4, 6,…,n-3,n-1} and W2 = { 1,3,5,…n-2,n}.

40
 For even n 4, the two cyclically transferring classes for the digraph
Db(n, 2) are,

U1 = {2, 4, 6,…, n-2, n} and U2= {1, 3, 5, …,n-3, n-1}.

When n 2, the rest of the members of G(n, 2) are obtained by relabeling

the vertices of Da(n, 2) if n is odd and Db(n, 2) if n is even.

Let G(2, 2) consist of a single digon. Every arc, and the digraph of any
matrix in K2 can be embedded in some members of F2, we have completed the
lemma.

Operators on K strongly preserving Fk(B) for k 2.

We noted in (2, 3) that Fk(B) ⊆ K. Let T denote an arbitrary linear operator

on K strongly preserving Fk = Fk(B); we will assume 2 k .

It is clear that T is bijective on the links; so satisfy the hypothesis of

theorem 1, we must show that T preserves 2- star matrices. We will do that in the
next sequence of lemma.

Note that T-1 strongly preserves Fk and that for all X, X is covered by Fk if
and only if T(X) is covered by Fk.

Lemma: 4.5

Suppose S ∈ K.

a) If D(S) is an orientation of a 2-path, then so is D(T(S)).

b) If D(S) is a digon, so is D(T(S)).
c) If D(S) is an independent pair of arcs, so is D(T(S)).

41
Proof: (a)

Suppose D(S) is an orientation of a 2-path.

Then there is a link G such that D(G + S) is an acyclic orientation of a

triangle. It follows that G+S is not covered by Fk.

But S is covered by Fk, so T(S) is covered by Fk. Therefore, when k 2,

T(S) cannot be a digon-matrix.

By applying the argument of the previous paragraph to the other two

orientations of 2-paths contained in D(G + S), it can be shown that T(G+ S) covers
no digon-matrices.

So D(T(G+ S)) has no digon-matrices and it has three arcs. It is either an

orientation of a triangle or an orientation of an acyclic undirected graph with three
edges.

The lattar type of digraphs can be embedded in some member of G(n, k), so
their matrices are covered by Fk, since T(G + S) is not covered by Fk, its digraph
must be an orientation of a triangle. Consequently, D(T(S)) is an orientation of a
2-path.

b) Suppose D(S) is a digon.

Since a) applies to T-1, as well as to T, D(T(S)) is either a digon or an independent

pair of arcs. If the latter, then n 3 and D(T(S)) can be embedded in one of the
digraphs in G(n, k). Therefore T(S), and hence S is covered by Fk.

If k 2, then that is impossible, so we can assume that k = 2. The

orientations of triangles are the only digraphs with three arcs that are subgraphs of
no strong digraph of index two. That follows from the fact that the index of
imprimitivity is the greatest common divisor of the directed cycle lengths.
42
 Let E be a link covered by S and choose links F and G so that the digraph of
X = E + F + G is an orientation of a triangle. Otherwise it would be a subgraph of
some strong digraph of index 2 and so T(X), and hence X, would be covered by F2.

Similarly, Y = Et + F + G and hence D(T(Y)) is an orientation of a triangle.

Thus the oriented triangles D(T(X)) and D(T(Y)) both contain the oriented 2-path,
T(F + G). Therefore, D(T(E)) and D(T(Et) are orientations of the same edge, so
D(T(S)) is a digon.

Since digons, orientations of 2-paths and independent pairs of arcs are the
only types of digraphs that D(T(S)) can be, a) and b) imply c).

Operators on Mn(B) strongly preserving Fk(B), k 2.

Let M = Mn(B) and 2 k n. If s is a linear transformation mapping the

diagonal members D of M into M, let

Ls(X) = X K + s(X I), for all X in M.

If is any linear operator on M, let

L*(X) = L(X K) + X I, for all X in M.

Lemma: 4.6

L strongly preserves Fk. Then

a) L(K) ⊆ K and
b) For all nonzero A in D, L(A) is not covered by Fk.

43
Proof:

Part(a).

Suppose L(E) F for some link E and some cell F.

The link E, and hence L(E), is covered by Fk. Therefore F is covered by Fk.
It follows that F is a link since k 2.

Part(b)

Suppose L(A) M ∈ Fk. Then L(A) + X = M, for some X K

Since X = L(Y) for some Y K.

Therefore A ∈ K. We say that Ls is compatible (with Fk) if s(A) is not

covered by Fk for all diagonal nonzero A.

Lemma: 4.7

Ls strongly preserves Fk if and only if Ls is compatible.

Proof:

If D is diagonal, then Ls(D) = s(D). Suppose Ls preserves Fk. Then by the

previous lemma (b) part implies that D = 0 if s(D) is covered by Fk.

Conversely, suppose Ls is compatible.

If X ∈ K. So Ls(X) = X ∈ Fk, then I X = 0 by compatibility, because

s(I X) Ls(X).

Thus X = X K = Ls(X) ∈ Fk

44
Lemma: 4.8

If L strongly preserves Fk, define a linear operator s mapping D into M by

s(D) = L(D), all D in D. Then Ls and L* strongly preserve Fk and L = LsL*.

If a linear operator L on a M strongly preserves Fk, then so does L* and L*

maps K into K. This implies that, unless k = 2 and n = 3,

L*(X K) = P-1( P for some permutation (pi, ) is either the identity

map or transposition. But similarity and transposition are linear and fix the
diagonal of each matrix, so L*(x) = P-1( (X))P.

Therefore L* is in the group of operators on M generated by transposition

and similarity.

Theorem: 4.9

If 2 k n, then the semigroup of linear operators on Mn(B) strongly

preserving Fk(B) is generated by transposition, the permutation similarity operators
and the compatible operators, unless k = 2, and n = 3. In that case it is generated
by those operators and the special operator defined by

0 0
0 →→→ 0 .
0 0

Matrices over ant negative, zero-divisor semi rings.

As we noted in semi rings, the non-negative reals, the non-negative integers

and the fuzzy scalars are all examples of anti-negative semi rings with no zero
divisors.

45
 When S is an anti-negative semi-ring without zero-divisors, then the pattern
homomorphism ensures that A in M(S) is irreducible if and only if its pattern ̅ in
M(B) is irreducible, it holds for A and that d(A) = d( ̅) for all irreducible A in
M(S)

Theorem: 4.10

Suppose S is an anti-negative semi ring with no zero divisors. If 2 k n,

unless k = 2 and n = 3, the semi group of linear operators on M = Mn(S) strongly
preserving Fk = Fk(S) is generated by; transposition, the similarity operators (X →
PXPt, for fixed invertible P), the compatible diagonal replacement operators (X →
X K + s(X I) for some fixed linear mapping s of the diagonal matrices of M into
M such that s(A) is not covered by Fk for any diagonal A 0), and the non zero
scalling operators, X → A X I, for some A, B in M with no zero entries ( we can
take X → A X when S is commutative). IAF n = 3 and k = 2, it is generated by the
above operators and the special operator defined by

0 0
0 →→→ 0 .
0 0

Note that “similarity operator” in the statement of the above theorem could be
replaced by “permutation operator” because each invertible matrix P in M(S) is a
scalar product Q (JP) where Q is the permutation matrix in M(S) whose pattern in
M(B) is and J is the matrix of ones. And so each similarity operator is a
composition of a permutation similarity with a non-zero scalling operator.

46
Matrices Over Cancellation Semi Rings

In this section, S is a cancellation semi ring, that is, for all a,b in S if x 0
and either ax = bx or xa = xb then a = b.

For example, any field or any division ring. Note that cancellation semi rings are
zero-dvisior-free.

Theorem: 4.11

Suppose S is a cancellation semi ring with at least n2 + 1 elements. Unless

n = 3 and k = 2, the semi group of linear operators on Mn(S) strongly
p[reserving Fk(S) is generated by the operators of the following types:

a) Transposition,
b) Similaritry,
c) Nonzero entrywise scaling,
d) Compatible diagonal replacement.

If n = 3 and k = 2, then the semigroup is generated by the above types of operators

and the operator defined by

0 0
0 →→→ 0
0 0

Note that as the above theorem, that “similarity” can be replaced by

“permutation similarity” ion the statement of theorem because each similarity
operators is a composition of a permutation similarity and a nonzero entrywise
scalling operator.

47
 Conclusion

The literature review involves a qualitative content analysis of available

information already published in the some form. It can be a study of the research
objects alone, with the aim of collecting information about its structure, process
and relationships., increasing of familiarity of the researcher with the research
object And establishing and credibility of the project. In addition it can consider
previous research, attempting to link it with the student currently planned. It may
also be geared towards a historical or comparative analysis of the issue in question
so the current study can be placed in a historical context.

Finally, it may review a theory of methods and techniques most suitable for
the study, simply by looking at the topic and by evaluating their suitability and
effectiveness.

48
 BIBLIOGRAPHY

1. S. Barik, M. Nath, S. Pati and B.K. Sarma. Unicyclic graphs with

strong reciprocal eigenvalue property. Electronic Journal of
Linear algebra, 7(march 2008). 139 – 153.
2. S. Barik, S. Pati and B.K. Sarma. The spectrum of corona of two
graphs. SIAM J. On Discrete Mathematics. 21(1) (jan 2007), 47-56.
3. S. Barik, M. Neumannand S. Pati. On non singular trees and a
reciprocal eigenvalue property. Linear and Multilinear algebra, 54(6)
(Dec 2006) 453-465
4. F. Buckly, L. L. Doty and F. Harary, On graph with signed inverses.
Network’s 18: 151-157, 1988
5. D. M Cvetkovic, M. Doob and H. Sachs, Spectra of Graphs,
Academic Press. 1979
6. R. Frucht and F. Harary, On the corona of two graphs. Aequations
Math., 4:322-325, 1970.
7. Ho R.A. Horn and C.R. Johnson. Matrix analysis. Cambridge
University Press, New York. (1987)
8. S. Pavlikova and J. Krc – jediny. On the inverse and dual index of a
tree. Linear and Multilinear Algebra, 28:93-109, 1990

49