The Third Basic Science International Conference - 2013 M03-1

On The Structural Properties of Latin Square in Max-Plus

Algebra
Muhammad Syifa’ul Mufid, Subiono

 denoted by  and denoted by a  b  a  b for every a, b

Abstract— Matrices L of size n × n are called Latin square if in R  . For example, 3  2  max{ 3, 2}  3 and
every column and every row of L contain n different numbers.
And, Max-Plus Algebra is algebraic system using two operations,  2  6  2  6  4 .
max and plus. In this paper, we derive some properties of a Latin It is easy to show that both operations , are
square in Max-Plus Algebra and their eigenvalues and commutative in max-plus algebra. Because all x  R  satisfy
corresponding eigenvector.
x      x  x and x  0  0  x  x , then the zero and
Index Terms— Latin square, Max-Plus Algebra, Eigenvalue, unit element in max-plus algebra is  and 0 , respectively.
Eigenvector. The set of all n  m matrices in max-plus algebra is denoted
by R n  m , and for m  1 we denoted the set of all n  1 vectors
I. INTRODUCTION by R n . Let A  R n  m , the entry of A in ith row and jth column
A Latin square of order n is square matrix of size n × n such is denoted by a i , j and sometimes we write [ A ]i , j . The ith row
that every row and every column has n distinct numbers. For and jth column of A is denoted by [ A ] i ,  and [ A ]  , j
convenience, we use n = {1,2,…,n}. The notion of Latin
respectively. For A , B  R n  m A  B is defined by
square is firstly introduced by Leonhard Euler. A Latin square
is in reduced form if first row is [1, 2, 3, …, n] and first [ A  B ] i , j  a i , j  b i , j  max{ a i , j , b i , j }
column is [1, 2, 3, …, n]T. If numbers in both diagonals also and for A  R n  p , B  R m  p , A  B is defined by
distinct then we called it by Latin square-X. An example of
[ A  B ] i , j  max{( a i ,1  b1, j ), ( a i , 2  b 2 , j ),..., ( a 1, p  b p , j )}
Latin square and reduced Latin square is given below
For example,
2 3 1 1 2 3
3 1 1 2  3 1  
1 2  2 3 1 

 1 2 3 

 3 1 2  A    1 4 , B   0  5 1 
 
0 2 3  1  2
The study of Latin square is mainly about discrete We get
mathematics aspect especially enumeration of Latin square.
 3 1 2  4 0 4
Until now, the exact number of Latin square is known only for
1 ≤ n ≤ 11. The result of enumeration Latin square-X is can be A  B  0  1 4 , A  B  3  6 6
 
found in [1]. The number of Latin square of order 5 and 6 is 0 2 3 3  1 5
960 and 92160 respectively, and for order 7 the number of
Latin square is increasing sharply, that is 862848000. In max-plus algebra, we defined A 2  A  A or
 k 1 k
Accordance with its name, Max-Plus Algebra is algebra that generally A  A  A for k  1, 2 ,...
using two operations, max and plus. In Max-Plus algebra Let A  R n  n , a digraph (directed graph) of A is denoted
defined algebraic structure ( R  ,  , ) where is set of ε
as G(A). Graph G(A) has n vertices and there is an edge from
extended real numbers, i.e. R   R  { } . In this paper, we vertex i to vertex j if a j , i   and this edge is denoted by (i, j).
denoted infinite element, i.e.    Operation max denoted
Weight of edge (i, j) is denoted by w ( i , j ) and equal to a j , i .
by  and defined by a  b  max{ a , b} , and operation plus
Sequence of edges ( j1 , j 2 ), ( j 2 , j 3 ),..., ( j k 1 , j k ) is called by
a path and if all vertices j1 , j 2 , j 3 ,..., j k 1 are different then
Manuscript received April 7, 2013.
Muhammad Syifa’ul Mufid, Graduate Student Department of called by elementary path. Circuit is an elementary close path,
Mathematics, Institut Teknologi Sepuluh Nopember Surabaya (e-mail: i.e. ( j1 , j 2 ), ( j 2 , j 3 ),..., ( j k 1 , j1 ). Circuit that consisting of
syifaul.mufid11@mhs.matematika.its.ac.id)
Subiono, Department of Mathematics, Institut Teknologi Sepuluh single edge, from a vertex to itself, is called by looping.
Nopember Surabaya (e-mail: subiono2008@matematika.its.ac.id)
.
The Third Basic Science International Conference - 2013 M03-2

Weight of a
path p  ( j1 , j 2 ), ( j 2 , j 3 ),..., ( j k 1 , j k ) is If L  LS n we consider matrix L 2  L  L . Because there
denoted by | p | w and equal to the sum of weight each edge. is only one  in every row and every column of L then
Length of path p is denoted by | p | l and equal to the number [ L 2 ] i , j is finite for all i , j  n . Therefore in graph G(L) there
are some paths with length at least 2 from vertex i to vertex j
of edges in p. Average weight of p is defined by | p | w .
| p |l for all i , j  n . Then we can conclude that G(L) is strongly
Any circuit with maximum average weight is called by connected and consequently L is irreducible.
critical circuit. Graph G(A) is called strongly connected if 
there is path for any vertex i to any vertex j in G(A). If graph B. Property of closed under operation 
G(A) is strongly connected, then matrix A is irreducible. From
matrix A, [ A  k ] i , j is equal to the maximal weight of a path We say that Latin squares are closed under operation  if
with length k from vertex i to vertex j. for all Latin squares A and B, A  B is Latin square.
Lemma 2. Let both A and B are in LS n or in LS n . A  B is
II. LATIN SQUARE IN MAX-PLUS ALGEBRA Latin square if and only if A  B .
Because the discussion is in max-plus algebra, it is allowed
to use infinite element    as number/element of Latin Proof.
square. In this paper we define two types of Latin square: Let A, B  LS n and C  A  B . Because [ A ] i , j , [ B ] i , j  n
a. Latin square without infinite element, the numbers that then [C ] i , j  [ A ] i , j  [ B ] i , j  n . If C is Latin square then
used are in n  {1, 2 ..., n}
b. Latin square with infinite element, the numbers that C  LS n . To prove A = B we only need considering first
used are in n   { ,1, 2 ..., n  1} column. See the illustration below
The set of all Latin squares of order n without infinite element  b1,1 
 c 1,1   a 1,1
is denoted by LS n and the set of all Latin squares of order n c   a
 2 ,1   2 ,1  b 2 ,1 
with infinite element is denoted by LS n . Example of two types
 .  . 
of Latin square is given below.    
We can infer that L1  LS 4 and L 2  LS 4 .  .   . 
 c n ,1   a n ,1  b n ,1 
  
2 3 1 4  1 3 2
Let a 1,1  x  n , if b1,1  x then x can appear more than one
3 1 4 2  3 2  1 
L1   , L2   or not appear in left side matrix. Therefore we get
4 2 3 1 2  1 3
    a 1,1  b1,1 and by same way we get a 2 ,1  b 2 ,1 ,..., a n1  b n ,1 or
1 4 2 3 1 3 2 
generally a i ,1  b i ,1 for all i  n . Consequently, the first column
of A and B is equal or generally [ A ]  , i  [ B ]  , i for all i  n ,
III. PROPERTIES OF LATIN SQUARE IN MAX-PLUS in other word A  B .
ALGEBRA
Conversely, if A  B then
Properties of Latin square in max-plus algebra that will be
derived are:
a. Irreducible. Are all Latin squares in max-plus algebra c i , j  a i , j  b i , j  a i , j  a i , j  max{ a i , j , a i , j }  a i , j
irreducible?
b. Close under operation  . Are all Latin squares closed Consequently, C  A  B  A  A  A and C is Latin square.
under operation  ? For A, B  LS n it can be proved by similar way.
c. Close under operation  . Are all Latin squares closed 
under operation  ? By Lemma 2 we can conclude that Latin square is not closed
under operation 
A. Property of Irreducibility
Lemma 1. All Latin squares are irreducible matrix.

Proof.
Let L be Latin square. If L  LS n then all numbers of L are
finite. Therefore, in graph G(L) there is a path with length 1
from vertex i to vertex j for all i , j  n . Then we can conclude
that G(L) is strongly connected and consequently L is Example:
irreducible.
The Third Basic Science International Conference - 2013 M03-3

1 2 3   2 1 3 1 2 3  2 3 1

A  2 3 1, B  1 3 2
 
A  2 3 1 , B  3 1 2

3 1 2 3 2 1 3 1 2 1 2 3
We get We get
2 2 3 5 5 6 
A  B  2 3 2 A  B  5 6 5
3 2 2 6 5 5
and A  B is not Latin square. and A  B is not Latin square.

IV. EIGENPROBLEM OF LATIN SQUARE IN MAX-PLUS

C. Property of closed under operation 
ALGEBRA
We say that Latin squares are closed under operation  if Matrix L of order n has eigenvalue   R and
for all Latin square A and B, A  B is Latin square. In other corresponding eigenvector v  R n  n if both of them satisfy
word, there is Latin square C and natural number p such
that A  B  p  C . Lvv
Lemma 3. If three Latin squares A , B , C  LS satisfy n

In this paper we denoted  ( A ) be eigenvalue of matrix A.

A  B  p  C then p  n .
From [2,3], there is algorithm to find eigenvalue
corresponding eigenvector that called by Power Algorithm. If
Proof. Let A  B  D , and D is Latin square, then
L is irreducible matrix, then eigenproblem is equivalent to
[ D ] i , j  max{( a i ,1  b1, j ), ( a i , 2  b 2 , j ),..., ( a 1, n  b n , j )} problem to find critical circuit in G(L), where eigenvalue is
Because maximum value both a i ,k and a k , j for all equal to weight of that critical circuit.
We define L   (   )  L and
k  n are n , then maximum value of [ D ] i , j is 2 n . Next, we
L   L  L  2  L 3  ...  L  n
determine the minimum value of [ D ] i , j .
It can be proved that [ L  ]  , j is eigenvector of L if
Let a i , k  b k , j  d k then [ D ] i , j  max{ d 1 , d 2 ,..., d n } and
[ L  ] j , j  0 [3].
n
we know that d
k 1
k  n(n  1) . It is easy to find that the A. Eigenvalue of Latin square in Max-Plus Algebra
From Lemma 1, all Latin squares are irreducible matrix.
minimum value of [ D ] i , j occur when Therefore, to find eigenvalue of L we need to find the weight
d 1  d 2  ...  d n  n  1 , of critical circuit in G(L).
then [ D ] i , j  max{ n  1, n  1,..., n  1}  n  1 . If there are If L  LS n then [ L ] j , j  {1, 2 ,..., n} and it is clear that

some k such that d k  n  1 then there are some l such max{[ L ] i , j }  n . Let
that d l  n  1 and consequently [ D ] i , j  n  1 . So, it is clear p  ( j1 , j 2 ), ( j 2 , j 3 ),..., ( j k 1 , j k ), ( j k , j1 )
that minimum value of [ D ] i , j is n  1 . be critical circuit with length k with k  n in G(L), then
| p | w  w ( j1 , j 2 )  w ( j 2 , j 3 )  ...  w ( j k 1 , j k )  kn
Because D is Latin square of order n
and average weight of p is equal to
and n  1  [ D ] i , j  2 n , then we can conclude that
| p | w kn
[ D ] i , j  {n  1, n  2 ,..., 2 n  1, 2 n}  { n  k |  k  n}  n
| p |l k
So, if D  p  C , we get p n. Because n occurs exactly one in every row and column of L,
 we can ensure that average weight of critical circuit p is equal
From Lemma 3, one of requirement for A  B producing
to n. Therefore, eigenvalue of L is equal to n, in other word
Latin square is for all i  n there is j  n such that   n.
[ A ] i ,   [ B ]  , j  [ n  1 n  1 . . n  1 n  1] By the same method, we get eigenvalue of L  LS n , that is
So we can conclude that Latin square is not closed under   n  1.
operation  .
B. Eigenvector of Latin square in Max-Plus Algebra
Let A  LS n and B  LS n . From the definition we get
A   (  n )  A and B   (  n  1)  A . It is clear that
Example:
The Third Basic Science International Conference - 2013 M03-4

average weight of critical circuit both in G(A) and and G(B) is   1  3   1
0.  0  4 0 
From [3], if p is critical circuit of G(L) then for all vertices    
A   4 
 in p satisfy [ L  ] ,  0 . But in this case,   1  3   1
for A  LS n and B  LS n , the average weight of A and B is      
  2 2  2
equal to maximum value of matrix A and B, i.e
  1  3   1
 ( A )  max{[ A ] i , j }  n   2 2  2
A       4 
and   1  3   1
 ( B )  max{[ B] i , j }  n  1 .      
Moreover, n occurs exactly one in every row and column of  0  4 0 
A and n  1 occurs exactly one in every row and column
of B . Consequently, there is always critical circuit in A such
that the initial vertex is i. The same condition is for matrix B . V. CONCLUSION
Therefore, for all   n in satisfy In this paper we can conclude that Latin squares have
[ A ] ,  0 and [ B ] ,  0 properties in Max-Plus Algebra, that is
1. All Latin squares are irreducible matrix
or in other word, all diagonal entries of both A and B are equal
2. Latin squares are not closed under operation 
to 0. So we can conclude that all columns of A  is
3. Latin squares are not closed under operation 
eigenvector of A and all columns of B  is eigenvector of B
Moreover, eigenvalue of Latin squares L are equal to
maximum value of L and all columns of L are eigenvector of
Example.
L.
1 3 4 2
2 4 3 1  REFERENCES
A [1] Mufid, M.S., 2012, Kajian Teori Permutasi and Its Application to
4 2 1 3 Enumeration of Latin Square-X and Sudoku-X (In Indonesian),
  Bachelor Thesis, Mathematics and Natural Science Faculty, Institut
3 1 2 4 Teknologi Sepuluh Nopember, Surabaya.
We get  ( A)  4 and
[2] Subiono and van der Woude, J.W., Power Algorithm for (max, +)- and
Bipartite (Min, Max,+)-system, Discrete Event Dynamic Systems:
 3 1 0  2 Theory and Application 10, pp. 369-389, 2000.
[3] Subiono, 2012, Aljabar Max-Plus dan Terapannya, Buku Ajar Mata
 2 0 1  3 Kuliah Pilihan Pasca Sarjana Matematika, Institut Teknologi Sepuluh
A   Nopember, Surabaya.
 0 2 3  1
 
1 3 2 0 
0 1 0  1
 1 0 1  2
A  

0 1 0  1
 
 1 2 1 0 


from A we get three different columns and we can check that
all of them are eigenvector of A

 0  4 0
 1 3  1
A       4 ,
 0  4 0
     
 1 3  1