You are on page 1of 5

Maximal and minimal solutions for two-sided problem of

type (1, n) in max-min algebra


Alena Pozdílkova, Pavel Krbálek
University of Hradec Králové, Faculty of Informatics and Management,
pavel.krbalek@uhk.cz, alena.pozdilkova@uhk.cz

Abstract
Basic characteristics and the solvability of linear systems in max-min algebra are presented. It
is shown how to compute maximal and minimal solutions for two-sided problem A ⊗ x =
B ⊗ x for matrices of type (1, n). The basic problem for matrices of type (1, n) is extended by
adding an upper bound and a lower bound for the solution. Concrete examples are provided
for better understanding of computation principles in max-min algebra. Reader can try the
presented algorithms on the web site. Algorithms are implemented in programming language
Java and are offered as an open library.
Key words: max-min algebra, two-sided problem

1. What is max-min algebra


Max–min algebra (R,⊕,⊗) is an algebraic structure with two binary operations ⊕, ⊗ and a
set R = R ∪ {− ∞, ∞}, which is an extension of real numbers. The operations minimum and
maximum in max-min algebra are derived from the linear ordering in the set of real numbers.
Max-min algebra belongs to the family of so-called extremal algebras. Interval systems in
extremal algebras were investigated by Cechlárová, Cuninghame-Green (2002), Myšková
(2007), two-sided systems in max-plus algebras were studied by Butkovič, Zimmermann
(2006), and two-sided systems in max-min algebra by Gavalec, Zimmermann.

The operations in max-min algebra are defined as follows:


for x, y ∈ R : x ⊕ y = max (x, y), x ⊗ y = min (x, y).

For matrices A, B over R we can define operations ⊕, ⊗ analogously as in linear algebra over
R with addition and multiplication. We assume matrices A, B of suitable types.

Example 1
 5
 
(3 7 2) ⊗  4  = (3 ⊗ 5) ⊕ (7 ⊗ 4) ⊕ (2 ⊗ 9) = 3 ⊕ 4 ⊕ 2 = 4
9
 

Two types of matrix equation are of interest in extremal algebras. The first type is the problem
of the form A ⊗ x = b (one-sided problem), with matrix A on the left-hand side and with the
right-hand side vector b. Such a matrix equation corresponds to a system of linear equations,
as we know it from linear algebra.

The second type of matrix equation is of the form A ⊗ x = B ⊗ x (two-sided problem) , with
matrices on both sides of the equation. In the extremal algebras there is no inverse operation
to ⊕, and hence the matrix B on the right side cannot be transferred to the left side of the
equation.

In this paper we solve the two-sided problem for matrices of special type (1, n). Let us denote
two matrices A, B of type (1, n) and vector of unknown values x of type (n,1), sets of indexes
M = {1, 2, …, m}, N = {1, 2, …, n}. We are looking for the maximal or minimal solution
x = ( x1 , x 2 ,..., x n ) T of linear equations of the system

A ⊗ x=B ⊗ x (1)

For given matrices A, B, the set of all solutions to the system (1) is denoted by S (A, B).

2. Maximal solution for matrices A, B of type (1, n)


In this chapter we are looking for maximal solution of the system (1) for given matrices A, B
of type (1, n) without restriction, and also with lower-bound constraint for the solution (see
the remark at the end of the chapter). In more detail, the equation can be written in the form
 x11   x11 
   
( ) x12 
(
a11 a12 ... a1n ⊗   = b11 b12 ... b1n ⊗  12  , or
...
x 
)
...
   
x  x 
 1n   1n 
(a11 ⊗ x1) ⊕ (a12 ⊗ x2) ⊕ … ⊕ (a1n ⊗ Xn) = (b11 ⊗ x1) ⊕ (b12 ⊗ x2) ⊕ … ⊕ (b1n ⊗ xn).

Equivalently, we are looking for maximal values of the unknown vector x satisfying the
equation
max j =1..n (min( a1 j , x j )) = max j =1..n (min(b1 j , x j )) , j = 1, …, n.
n n
Let us denote a = max j =1..n (a11 , L , a1n ) = ⊕ a1 j , b = max j =1..n (b11 , L , b1n ) = ⊕ b1 j . There are 3
j =1 j =1

possibilities: a = b, a < b, or a > b. The following Lemmas describe the first two cases; the
third case can be described analogously as the second one.

Lemma 1 If a = b, then x = (∞, ∞,...∞) T is the greatest solution of linear system A ⊗ x =B ⊗ x.


a if b j > a
Lemma 2 If a < b, then denote x = ( x1 , x 2 ,..., x n ) T , x j :=  for j = 1,…n
∞ otherwise
Example 2
 x1   x1 
   
(3 7 2) ⊗  x 2  = (4 2 5) ⊗  x 2  ,
   
x
 3
 x3 
n n
a = ⊕ (a1 j ) = 7 , b = ⊕ (b1 j ) = 5 , a > b, so according to Lemma 2:
j =1 j =1

x1 = ∞ , x 2 =5 because a2 > b, x 3 = ∞ , so x = ( ∞ , 5, ∞ )

Remark 1
Maximal solution with given lower-bound constraint (l1 ,...l n ) T ≤ ( x1 ,...x n ) T - is defined
analogously as the maximal solution without restriction. If maximal solution without
restriction is not greater or equal than the given lower bound l, than system A ⊗ x=B ⊗ x,
l ≤ x has no maximal solution. In other cases solution is the same, as the solution of the
system without any restriction.

3. Maximal solution of linear systems equation with defined upper


bound for matrices A, B of type (1, n)
We will try to find maximal existing solution x = ( x 1 , x 2 ,..., x n ) T of linear system
A ⊗ x =B ⊗ x , matrices A, B of type (n, 1) with constraint condition ( x 1 ,...x n ) T ≤ (u1 ,...u n ) T .
Denote x = (u1 ,...u n ) T = u, a = A ⊗ x, b =B ⊗ x. For j = 1, …n there exist 3 possibilities:
a = b, a < b, or a > b. The following Lemmas describe the first two cases; the third case can
be described analogously as the second case.

Lemma 3 If a = b then u is maximal solution.

Lemma 4 If a > b, then denote x = ( x1 , x 2 ,..., x n ) T .


b if min(ai , u i ) > b
If xi :=  , i = 1, …, n, then x is maximal solution.
u i otherwise
Example 3
 x1   x1  11
     
(10 8 10) ⊗  x 2  = (9 9 9) ⊗  x 2  , u= 11 .
    11
 x3   x3   
Set x = u, a(x) = A ⊗ x = 10, b(x) = B ⊗ u = 9, a(x) > b(x), so according to Lemma 4:
x1 = 9, because min(a1,x1) = min(10,11) > 9, x 3 = 9, because min(a3,x3) = min(10,11)>9. There
exists maximal solution with values: x = (9, 11, 9).

Figure 1 Graph of matrices from example 3

4. Minimal solution for matrices A, B of type (1, n)


Minimal solution of the system (1) without restriction or only with upper bound is always
x = (−∞,−∞,... − ∞) T , because (−∞,−∞,... − ∞) T is a solution of the system (1) and it is clearly
the lowest of all solutions.

5. Minimal solution of linear systems equation with defined lower


bound for matrices A, B of type (1, n)
We will try to find minimal existing solution x = ( x1 , x 2 ,..., x n )T of linear system
A ⊗ x =B ⊗ x , matrices A, B of type (1, n) with constraint condition ( x1 ,...x n )T ≥ (l1 ,...ln )T .
n n
Denote x = l = (l1 , l 2 ,..., l n ) T , a = A ⊗ x = ⊕( ⊗ (a j , x j )) , b = B ⊗ x = ⊕( ⊗ (b j , x j )) . There
j =1 j =1

exist 3 possibilities: a = b, a < b, or a > b. The following Lemmas describe the first two cases;
the third case can be described analogously as the second case.

Lemma 5 If a = b then l is minimal solution.

Lemma 6 If a > b, then denote x = ( x1 , x 2 ,..., x n )T , N 1 = {j ∈ N : b j ≥ a}. If j ∈ N1 and


x j = a and xk = lk for k ≠ j , then x is minimal solution.
i. If N1 = { j} then there is exactly one minimal solution.
ii. If N1 = Ø, then there is no solution.
iii. If N1 > 1 , then there are more minimal solutions. Namely, there exist as many
minimal solutions as there are elements in set N1.

Example 4
Example of exactly one minimal solution, a > b:
A ⊗ x =B ⊗ x , set equation with values:
 x1   x1 
   
(5 7 3) ⊗  x 2  = (3 5 4) ⊗  x 2  .
x  x 
 3  3
 6
  n n
l =  4  , a = ⊕ (a j ⊗ l j ) = ⊕ (5 ⊗ 6, 7 ⊗ 4, 3 ⊗ 6) = 5,
j =1 j =1
 6
 
n n
b = ⊕ (b j ⊗ l j ) = ⊕ (3 ⊗ 6, 5 ⊗ 4, 4 ⊗ 6) = 4,
j =1 j =1

x = (6, 5, 6).

6. Implementation of presented solutions in programming language


The algorithms described in this article are implemented in programming language Java. The
implementation can be used in two different ways. The one is the web on address [4] where
the demonstrations of basic principles are accessible. On this page there are few sections
where reader can interact with elementary computes. There are examples of multiply of two
matrices, finding the maximum value, finding the maximum with given upper bound and
others. The second way of using this Java program is to connect this java implementation to
another program. The implementation offers an interface that can be used to solve concrete
problems. Program is presented as an open library and it is free to use or extend.

7. Conclusions
This paper describes the notation of max-min algebra and shows how to compute maximal
and minimal solution for two sided problem A ⊗ x = B ⊗ x for matrices of type (1, n), which is
further extended by adding upper bound and lower bound and provided with examples.
Algorithm is implemented in programming language Java and is free available on the web
site.
Alena Pozdílkova, Pavel Krbálek, Department of Information Technologies, Fakuty of
Informatics and Management, University of Hradec Králové, Rokitanského 62, 500 03
Hradec Králové, Pavel.Krbalek@uhk.cz, Alena.Pozdilkova@uhk.cz

Literature
1. Butkovič, P., Zimmermann, K.: Strongly polynomial algorithm for solving two-sided
systems of (max, plus)-linear equations, Discrete Applied Math. 154 (2006), pp. 437-446.

2. Cechlárová, K., Cuninghame-Green, R.A.: Interval systems of max-separable linear


equations, Linear Algebra and Appl. 340 (2002), pp. 215-224.

3. Gavalec, M., Zimmermann, K.: Solving systems of two-sided (max,min)-linear equations,


Linear Algebra and Appl. (submitted)

4. Krbálek, P.: MAX-MIN ALGEBRA How to solve max-min (demonstration) [online]


http://lide.uhk.cz/krbalpa1/max-algebra/

5. Myšková, H.: Solvability of interval systems of fuzzy linear equations., Proc. of the
International Conference Mathematical Methods in Economics and Industry 2007,
Herlany (Slovakia), pp. 153-157.

You might also like