Separable Cubic Stochastic Operators
Separable Cubic Stochastic Operators
Separable Cubic Stochastic Operators
Introduction
There are many systems which are described by nonlinear operators. A quadratic stochastic operator
(QSO) is one of the simplest nonlinear cases. A QSO has meaning of a population evolution operator
and it was first introduced by Bernstein in [1]. For more than 80 years, the theory of QSOs has been
developed and many papers were published (see e.g. [4]-[7]). In recent years it has again become of
interest in connection with its numerous applications in many branches of mathematics, biology and
physics.
Let E = 1,2,..., m be a finite set and the set of all probability distribution on E
m
S m −1
= x = ( x1 , x2 ,..., xm ) m
: xi 0, for any i and xi = 1 (1)
i =1
be the ( m − 1) -dimensional simplex. A QSO is a mapping defined as V : S
m −1
→ S m−1 of the
simplex into itself, of the form V ( x ) = x S m−1 , where
m
xk = Pij ,k xi x j , k E, (2)
i , j =1
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
( )
x( n+1) = V x( n) = V ( n+1) x(0) , ( ) n = 0,1,2,... .
One of the main problems in mathematical biology is to study the asymptotic behavior of the
trajectories. This problem was solved completely for the Volterra QSO.
The operator V is called Volterra QSO, if Pij ,k = 0 for any k i, j , i, j , k E . For the Volterra
QSO the general formula was given in [4],
m
xk = xk 1 + aki xi , (4)
i =1
where aki = 2 Pik ,k − 1 for i k and akk = 0 . Moreover, aki = −aik and aki 1 for all i, k E .
In [4], the theory of Volterra QSO was developed using theory of the Lyapunov functions and
tournaments. But non-Volterra QSOs were not completely studied. Because, there is no general theory
that can be applied for study of non-Volterra operators.
In [7] Separable Quadratic Stochastic Operators (SQSOs) where introduced. Volterra QSO (4) has a
form as SQSO, but in [8] it was proved that it coincides with a SQSO if and only if it is a linear
operator.
In recent years, Cubic Stochastic Operators (CSOs) have begun to be studied, which different from
quadratic operators.
In this paper we consider another class of cubic operators which we call separable cubic stochastic
operators.
In Section 2, we recall the definition of CSOs and definitions and known results. In Section 3 for a
SCSO defined on the two-dimensional simplex, we prove that it has three fixed points and we find
conditions on parameter under which a fixed point is a repelling, attracting, or saddle point, the
boundary S is an invariant set.
2
xk = (V ( x ) )k = ( A ( x ) )k ( B ( x ) )k , (6)
where ( A ( x ) ) = aik xi , ( B ( x ) ) = b jk x j .
m m
k k
i =1 j =1
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
Definition 1. [7] The QSO (6) is called separable quadratic stochastic operator (SQSO).
m
From the conditions Pij ,k 0 and P k =1
ij , k
= 1 for all i, j it follows that the condition on matrices A
entries 1’s. If a (i ) = ( ai1 ,..., aim ) is the i -th row of the matrix A and b( j ) = ( b j1 ,..., b jm ) is the
( ) = (1,1,...,1) . (7)
T
A b( j )
If det ( A ) 0 , then (7) gives b( j ) = b( k ) for all k, j E , i.e., all rows of B are the same,
therefore det ( B ) = 0 . Similarly, if det ( B ) 0 , then all the rows of A must be the same, so
det ( A ) = 0 .
Cubic stochastic operator. The CSO is a mapping W : S m−1 → S m−1 of the form
m
xl = P
i , j , k =1
ijk ,l
xi x j xk , l E , (8)
and we suppose that the coefficients Pijk ,l do not change for any permutation of i, j , k .
Note that W is a non-linear operator, and its dimension increases with m . Higher-dimensional
dynamical systems are essential, but only relatively few dynamical systems have yet been analysed.
follows that ( x ) . If ( x ) consists of a single point, they the trajectory converges and
( 0) ( 0)
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
W ( x ) = x . Denote by Fix (W ) the set of all fixed points of the operator W , i.e.
Fix (W ) = x S m−1 :W ( x ) = x .
(i) attracting if all the eigenvalues of the Jacobian DW ( x ) are in the unit disk;
(ii) repelling if all the eigenvalues of the Jacobian DW ( x ) are outside the closed unit disk;
xl = (W ( x ) )l = ( A ( x ) )l ( B ( x ) )l ( C ( x ) )l (12)
( A( x )) = a x , ( B ( x)) = b (C ( x )) = c
m m m
where
l il i l jl
xj and
l kl
xk .
i =1 j =1 k =1
Definition 5. The CSO (12) is called separable cubic stochastic operator (SCSO).
Lemma 1. The condition (11) is sufficient for a CSO to be product of three linear operators, but the
condition is not necessary.
The CSO with coefficients Pijk ,l can be written as the product of three matrices A = ( ail ) ,
B = ( b jl ) and C = ( ckl ) if and only if, for any i, j , k , l E it holds
Pijk ,l + Pkij ,l + Pikj ,l + Pkji ,l + Pjik ,l + Pjki ,l = ail b jl ckl + akl bil c jl + ail bkl c jl +
Thus the
akl b jl cil + a jl bil ckl + a jl bkl cil (13)
condition (11) is a particular case of (13).
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
Let us consider the following matrices:
a
1 1 1 +
1 0 0 2 1 1+ a 1
A = 0 1 0 , B = 1 − a 1 1 , C = 1 1 1 + a , (18)
0 0 1 1 1− a 1 a
1 − 1 1
2
where a [ −1,1] .
a
x1 = x1 ( x1 + (1 − a ) x2 + x3 ) x1 + x2 + 1 − x3 ,
2
W : x2 = x2 ( x1 + x2 + (1 − a ) x3 ) ( (1 + a ) x1 + x2 + x3 ) , (19)
x = x 1 + a x + x + x ( x + (1 + a ) x + x ) .
3 1 2 3
3 2
1 2 3
W : x2 = x2 (1 − ax3 )(1 + ax1 ) , (20)
x3 = x3 1 + a x1 (1 + ax2 ) .
2
Evidently, that if a =0 the SCSO (20) is the identity map. For this in the below, we consider the
case when a 0 .
Let the set int S 2 = x S 2 : x1 x2 x3 0 and let the set S 2 = S 2 \ int S 2 be the interior
and the boundary of the simplex S2, respectively. Let e1 = (1,0,0 ) , e 2 = ( 0,1,0 ) ,
e3 = ( 0,0,1) be the vertexes of the two-dimensional simplex.
Theorem 1. For the SCSO W (20), the following assertions true:
(i) The face 1,2 , 1,3 , 2,3 of the simplex S2 are invariant sets;
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
(ii) To find the fixed points we consider the equation W ( x ) = x , that is the following system of
equations
a
1 x = x (1 − ax )
2
1 − x3 ,
2
1
x2 = x2 (1 − ax3 )(1 + ax1 ) , (21)
x3 = x3 1 + a x1 (1 + ax2 ) .
2
(a) Let x1 = 0 . Then from the second equation of (21) it follows
x2 = x2 − ax2 x3 x2 = 0 and 1 = 1 − ax3 .
It is clear that if x2 = 0 then we have x3 = 1 . If x2 0 then from the last equation one has
x3 = 0 x2 = 1.
We take solution x2 = 1, then it follows that x3 = 0 . Consequently, if x1 = 0 , we obtain the fixed
points e 2 , e3 .
Similarly, in the case x2 = 0 and x3 = 0 , we have the fixed points e1 , e 3 and e1 , e 2 , respectively.
(b) Suppose that x1 x2 x3 0 . Then from the system (21), one has
a
1 = (1 − ax )
2
1 − x3 ,
2
1 = (1 − ax3 )(1 + ax1 ) , (22)
1 = 1 + a x1 (1 + ax2 ) .
2
If a −1,0 ) , then from the first equation of (22) we get
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
a a 1
1 = (1 − ax2 ) 1 − x3 x2 x3 = x2 + x3 .
2 2 2
But from a −1,0 ) and from the last equation it follows that x2 = 0 and x3 = 0 , this
contradicts to x1 x2 x3 0 .
a
x1 = x1 (1 − ax2 ) 1 − 2 (1 − x1 − x2 ) , (23)
x = x (1 + ax ) (1 − a (1 − x − x ) ) .
2 2 1 1 2
where ( x , x ) ( x, y ) : x, y 0, 0 x + y 1 and
1 2
x1 , x2 are the first two coordinates of
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/
CAJMNS Volume: 04 Issue: 06 | Nov-Dec 2023
References
1. S. Bernstein, Solution of a mathematical problem connected with the theory of heredity}, Ann.
Math. Stat. 13(1) (1942), pp. 53-61. DOI:10.1214/aoms/1177731642.
2. R. R. Davronov, U. U. Jamilov(Zhamilov), and M. Ladra, Conditional cubic stochastic operator, J.
Differ. Equ. Appl. 21(12) (2015), pp. 1163-1170. DOI:10.1080/10236198.2015.1062481.
3. R. L. Devaney, An Introduction to Chaotic Dynamical Systems, in Studies in Nonlinearity,
Westview Press, Boulder, CO, 2003.
4. R. N. Ganikhodzhaev, Quadratic stochastic operators, Lyapunov functions and tournaments, Sb.
Math.76(2) (1993), pp. 489-506. DOI:10.1070/SM1993v076n02ABEH003423.
5. R. N. Ganikhodzhaev and D. B. Eshmamatova, Quadratic automorphisms of a simplex and the
asymptotic behavior of their trajectories, Vladikavkaz. Mat. Zh. 8(2) (2006), pp. 12-28.
6. A. Yu. Khamraev, On cubic operators of Volterra type, Uzbek. Mat. Zh. 2004(2) (2004), pp. 79-
84. (in Russian).
7. U. A. Rozikov, S. Nazir, Separable Quadratic Stochastic Operators, Lobschevskii J. Math. (3),
215(2010), pp. 215-221. DOI: 10.1134/S1995080210030030.
8. U. A. Rozikov, A. Zada, On a Class of Separable Quadratic Stochastic Operators, Lobschevskii J.
Math. (4), 32(2011), pp. 385-394. DOI: 10.1134/S1995080211040196.
Copyright (c) 2023 Author (s). This is an open-access article distributed under the terms of Creative Commons
Attribution License (CC BY).To view a copy of this license, visit https://creativecommons.org/licenses/by/4.0/