Open access peer-reviewed chapter

Invariants for a Dynamical System with Strong Random Perturbations

By Elena Karachanskaya

Submitted: November 2nd 2020Reviewed: January 27th 2021Published: February 23rd 2021

DOI: 10.5772/intechopen.96235

Downloaded: 132


In this chapter we consider the invariant method for stochastic system with strong perturbations, and its application to many different tasks related to dynamical systems with invariants. This theory allows constructing the mathematical model (deterministic and stochastic) of actual process if it has invariant functions. These models have a kind of jump-diffusion equations system (stochastic differential Itô equations with a Wiener and a Poisson paths). We show that an invariant function (with probability 1) for stochastic dynamical system under strong perturbations exists. We consider a programmed control with Prob. 1 for stochastic dynamical systems – PSP1. We study the construction of stochastic models with invariant function based on deterministic model with invariant one and show the results of numerical simulation. The concept of a first integral for stochastic differential equation Itô introduce by V. Doobko, and the generalized Itô – Wentzell formula for jump-diffusion function proved us, play the key role for this research.


  • Itô equation
  • Poisson jump
  • invariant function
  • differential equations system construction
  • stochastic system with invariants
  • programmed control with probability 1

1. Introduction

Models for actual dynamical processes are based on some restrictions. These restrictions are represented as a conservation law.

The conservation law states that a particular measurable property of an isolated dynamical system does not change as the system evolves over time.

Actual dynamical systems are open, and they are subject to strong external disturbances that violate the laws of conservation for the given system.

Conventionally, deterministic dynamical systems have an invariant function. Doobko1 V. in [1] proved that stochastic dynamical systems have an invariant function as well. For dynamical system which are described using a system of stochastic differential Itô equations, a first integral – or an invariant function, exists with probability 1 [2, 3, 4, 5, 6, 7, 8, 9, 10].

When we know only a conservation law for a dynamical system, and equations which describing this system are unknown, the invariant functions are a good tool for determination of these equations.

Our method differs for other (see, for example, [11]) preliminary in the fact that we construct a system of differential equation with the given first integral under arbitrary initial conditions. Besides, this algorithm is realized as software and it allows us to choose a set of functions for simulation. Moreover, we can construct both a system of stochastic differential equations and a system of deterministic ones.

The goal of this chapter is representation of modern approach to describe of dynamical systems having a set of invariant functions.

This chapter is structured as follows. Firstly, we show that the invariant functions for stochastic systems exist. Then, the generalized Itô – Wentzell formula is represented. It is a differentiated rule for Jump-diffusion function under variables which solves the Jump-diffusion equations system. This rule is basic for the necessary and sufficient conditions for the stochastic first integral (or invariant function with probability 1) for the Jump-diffusion equations system. The next step is the construction of the differential equations system using the given invariant functions. It can be applied for stochastic and nonstochastic cases. The concept of PCP1 (Programmed control with Prob. 1) for stochastic dynamical systems is introduced. Finally, we show an application of the stochastic invariant theory for a transit from deterministic model with invariant to the same stochastic model. Several examples of application of this theory are given and confirmed by results of numerical calculations.


2. Notation and preliminaries

Now we introduce the main concepts which we will use below.

Let wt, t0be a Wiener process or a (standard) Brownian motion, i. e.

  • w0=0,

  • it has stationary, independent increments,

  • for every t>0, wthas a normal N0tdistribution,

  • it has continuous sample paths,

  • every trajectory of wtis not differentiated for all t0.

A νtAis called a Poisson random measure or standard Poisson measure (PM) if it is non-negative integer random variable with the Poisson distribution νtAPoitΠA, and it has the properties of measure:

  • νtAis a random variable for every t0T, ARn,

  • νtAN0, νt=0,

  • if AB=, then νtAB=νtA+νtB,

  • EνtA=tΠA,

  • if #Ais a number of random events from set Aduring t, then


ν˜tA=νtAEνtAis called a centered Poisson measure (CPM).

Let wt=w1twmtbe an m-dimensional Wiener process, such that the one-dimensional Wiener processes wktfor k=1,,mis mutually independent.

Take a vector γΘwith values in Rn. Denote by νΔtΔγthe PM on 0T×Rnmodeling independent random variables on disjoint intervals and sets. The Wiener processes wkt, k=1,,m, and the Poisson measure ν0TAdefined on the specified space are Ft-measurable and independent of one another.

Consider a random process xtwith values in Rn, n2, defined by the Equation [12]:


where At=a1tant, Bt=bj,ktis n×k- matrix, and gtγ=g1tγgntγRn, and γRnRγ, while wtis an m-dimensional Wiener process. In general the coefficients At, Bt, and gtγare random functions depending also on xt. Since the restrictions on these coefficients relate explicitly only to the variables tand γ, we use precisely this notation for the coefficients of (1) instead of writing Atxt, txt, and gtxtγ.

A system (1) is the stochastic differential Itô equation with Wiener and Poisson perturbations, which named below as a Jump-diffusion Itô equations system (GSDES).

We will consider the dynamical system described using ordinary deterministic differential equations (ODE) system and ordinary stochastic differential Itô equations (SDE) system of different types, taking into account the fact that xRn, n2.

3. An existence of an invariant function (with Prob.1) for stochastic dynamical system under strong perturbations

Consider the diffusion Itô equation in R3with orthogonal random action with respect to the vector of the solution


where vR3, wR3, and wit, i=1,2,3are independent Wiener processes. This equation is a specific form of the Langevin equation.

V. Doobko in [1] showed that the system (2) have an invariant function called a first integral of this system:


This, in particular, implies that


i.e. process vtis a nonrandom function and the random process vtitself is generated in a sphere of constant radius bμ.

In [4, 5, 10] it is shown that invariant function exists for other stochastic equations of Langevin type. To obtain this result, it is necessary to use the Itô’s formula.

4. The generalized Itô – Wentzell formula for jump-diffusion function

The rules for constructing stochastic differentials, e.g., the change rule, are very important in the theory of stochastic random processes. These are Itô’s formula [13, 14] for the differential of a nonrandom function of a random process and the Itô – Wentzell2 formula [15] enabling us to construct the differential of a function which per se is a solution to a stochastic equation. Many articles address the derivation of these formulas for various classes of processes by extending Itô’s formula and the Itô – Wentzell formula to a larger class of functions.

The next level is to obtain a new formula for the generalized Itô Equation [14] which involves Wiener and Poisson components. In 2002, V. Doobko presented [7] a generalization of stochastic differentials of random functions satisfying GSDES with CPM based on expressions for the kernels of integral invariants (only the ideas of a possible proof) were sketched in [7]. The result is called” the generalized Itô – Wentzell formula”.

In contrast to [7], the generalized Itô – Wentzell formula for the noncentered Poisson measure was represented in [9, 16, 17]. The proof [9] of the generalized Itô – Wentzell formula uses the method of stochastic integral invariants and equations for their kernels. In this case the requirement on the character of the Poisson distribution is only a general restriction, as the knowledge of its explicit form is unnecessary. Other proofs in [16, 17] are based on traditional stochastic analysis and the use of approximations to random functions related to stochastic differential equations by averaging their values at each point.

The generalized Itô – Wentzell formula relying on the kernels of integral invariants [9] requires stricter conditions on the coefficients of all equations under consideration: the existence of second derivatives. The reason is that the kernels of invariants for differential equations exist under certain restrictions on the coefficients.

Since the random function Ftxthas representation as stochastic diffusion Itô equation with jumps, we can use the generalized Itô – Wentzell formula, proved by us by several methods in accordance with different conditions for the equations coefficients. Now we consider only one case.

We will use the following notation: Cysis the space of functions having continuous derivatives of order swith respect to y, C0syis the space of bounded functions having bounded continuous derivatives of order swith respect to y.

Theorem 1.1 (generalized Itô – Wentzell formula). Consider the real function FtxCt,x1,2, tx0T×Rnwith generalized stochastic differential of the form


whose coefficients satisfy the conditions:


If a random process xtobeys (1) and its coefficients satisfy the conditions


then the stochastic differential exists and


By analogy with the terminology proposed earlier, let us call formula (5) “the generalized Itô – Wentzell formula for the GSDES with PM” (GIWF).

By analogy with the classical Itô and Itô – Wentzell formulas, the generalized Itô – Wentzell formula is promising for various applications. In particular, it helped to obtain equations for the first and stochastic first integrals of the stochastic Itô system [9], equations for the density of stochastic dynamical invariants, Kolmogorov equations for the density of transition probabilities of random processes described by the generalized stochastic Itô differential Equation [8], as well as the construction of program controls with probability 1 for stochastic systems [18, 19].

5. A first integral for GSDES

In the theory of ODE, there are constructed equations to find deterministic functions, first integrals which preserve a constant value with any solutions to the equation. The concept of a first integral plays an important role in theoretical mechanics, for example, to solve inverse problems of mechanics or in constructing controls of dynamical systems.

It turned out that the first integral exists in the theory of stochastic differential equations (SDE) as well. However, there appears an additional classification connected with different interpretations. This gives a first integral for a system of SDE (see [1]), a first direct integral, and a first inverse integral for a system of Itô SDE (see [20]).

Definition 1.1 [1, 3]. Let xtbe an n-dimensional random process satisfying a system of Itô SDE


whose coefficients satisfy the conditions of the existence and uniqueness of a solution [12]. A nonrandom function utxCt,x1,2is called a first integral of the system of SDE if it takes a constant value depending only on x0on any trajectory solution to (6) with probability 1:


or, in other words, its stochastic differential is equal to zero: dtutxt=0.

Another important notion in the theory of deterministic dynamical systems is given by the notion of an integral invariant introduced by Poincaré [21].

As it turned out, there also exist integral invariants for stochastic dynamical systems [2, 3]. In [7] V. Doobko give the concept of a kernel (=density) of a stochastic integral invariant and, based on it, formulate the notion of a stochastic first integral and a first integral as a deterministic function for GSDES with the centered Poisson measure, which makes it possible to compose a list of first integrals for stochastic differential equations.

Consider a random process xt, xRn, which is a solution to GSDES


whose coefficients (in general, random functions) satisfy the conditions of the existence and uniqueness of a solution [12] and the following smoothness conditions:


Suppose that ρtxωis a random function connected with any deterministic function ftxSC01,2txby the relations


where yx0, and xtyis a solution to (7), and ωis a random event.

In the particular case when ftx=1, conditions (9) and (10) imply that


i.e., for the random function ρtxω, there exists a nonrandom functional preserving a constant value:


Then, with conditions (10) and (11), Eq. (9) can be regarded as a stochastic integral invariant, and the function (t, x) can be viewed as its density.

Definition 1.2 [3]. A nonnegative random function ρtxωis referred to as a stochastic kernel or the stochastic density of a stochastic integral invariant (of nth order) if conditions (9), (10), and (11) are held.

Note that a substantial difference which made it possible to consider the invariance of the random volume on the basis of a kernel of an integral operator in [3, 7], is that (9) contains a functional factor. Thus, the notion of a kernel of an integral invariant [3] for a system of ordinary differential equations can be regarded as a particular case by taking ftx=1and excluding from (7) the randomness determined by the Wiener and Poisson processes.

Using the GIWF (5), we obtain equation for the stochastic kernel function [9].


under restrictions


This result plays a major role in obtaining of equation for the stochastic first integral.


6. Necessary and sufficient conditions for the stochastic first integral

Lemma 1.1. If ρtxωis a stochastic kernel of an integral invariant of nth order of a stochastic process xtstarting from a point x0then, for every t, it satisfies the equality


where Jtx0ωis the Jacobian of transition from xtto x0.

Definition 1.3 A set of kernels of integral invariants of nth order is called complete if any other function that is the kernel of this integral invariant can be presented as a function of the elements of this set.

In [9] it is shown that a system of GSDE (7) whose coefficients satisfy the conditions in (8), has a complete set of kernels consisting of n+1functions.

Suppose that ρltxω0, l=1,,m, mn+1are kernels of the integral invariant (9). Lemma 1.1 implies that, for any ln+1, the ratio ρltxtyωρn+1txtyωis a constant depending only on the initial condition x0=yfor every solution xtto the GSDE (7) because


Since for some realization ω1we have


and it means, that dtutxt=0.

Definition 1.4 A random function utxωdefined on the same probability space as a solution to (7) is referred to as a stochastic first integral of the system (7) of Itô ˆ GSDE with NCM if the following condition holds with probability 1:


for every solution xtx0ωto (7).

For practical purposes, for example, to construct program controls for a dynamical system under strong random perturbations, the presence of a concrete realization is important, i.e., the parameter ωis absent in what follows. In this connection, we introduce one more notion.

Definition 1.5 A nonrandom function utxis called a first integral of the system of GSDE (7) if it preserves a constant value with probability 1 for every realization of a random process xtthat is a solution to this system:


Thus, a stochastic first integral includes all trajectories (or realizations) of the random process while the first integral is related to one realization.

Construct an equation for utxωusing the relation


as a result of assertion (15). Let us differentiate lnρtx(omit ω) using generalized Itô – Wentzell formula:


where d˜tρtxis the right side of Eq.(14) without the integral expression. Having written down the equations for lnρstxand lnρltx, and taking into account this result and Eq.(16), we obtain:


which means that a stochastic first integral utxωof the Itô generalized Eq. (7) is a solution to the GSDE (18).

For a first integral which is a nonrandom function of one realization, the differential is also defined by an equation of the form of (18).

Theorem 1.2 Let xtbe a solution to the GSDES (7) with conditions (8). A nonrandom function utxCt,x1,2is a first integral of system (7) if and only if it satisfies the conditions:

  1. utxt+utxxiaitx12bjk(tx)biktxxj=0,

  2. biktxutxxi=0, for all k=1,m¯,

  3. utxutx+gtxγ=0for any γRγin the entire domain of definition of the process.

Theorem (6) allows us to obtain a method for construction of differential equations systems on the basis of the given set of invariant functions.

7. Construction of the differential equations system using the given invariant functions

The concept of a first integral for a system of stochastic differential equations plays a key role in our theory. In this section, we will use a set of first integrals for the construction of a system of differential equations.

Let us write Eq. (7) in matrix form:


Theorem 1.3 [22]. Let Xtbe a solution of the Eq. (19) and let a nonrandom function stxbe continuous together with its first-order partial derivatives with respect to all its variables. Assume the set eoe1endefines an orthogonal basis in R+×Rn. If function stxis a first integral for the system (19), then the coefficients of Eq. (19) and the function stxtogether are related by the conditions:

1.Functions Bktx=i=1nbiktxeik=1m, which determine columns of the matrix Btx, belong to a set


where qootxis an arbitrary nonvanishing function,

2.Coefficient Atxbelongs to a set of functions defined by


where a column matrix Rtxwith components ritx, i=1n, is defined as follows:


Ctxis an algebraic adjunct of the element eoof a matrix Htxand detCtx0, a matrix Htxis defined as


and Bktxxis a Jacobi matrix for function Bktx,

3.Coefficient ΘtXγ=i=1nγitxγei, related to Poisson measure, is defined by the representation Θtxγ=ytxγx, where ytxγis a solution of the differential equations system


This solution satisfies the initial condition: ytxγγ=0=x.

The arbitrary functions fij=fijtx, hij=hijtx, and φij=φijtyγare defined by the equalities fijtx=fitxxj, hijtx=hitxxj, and φijtyγ=φityγyj. Sets of functions φityγand the function gtxtogether form a class of independent functions.

Using this theorem, we can to construct SDE system of different types and ODE system. Choice of arbitrary functions allows us to construct a set of differential equations systems with the given invariant functions. Theorem (7) allows us to introduce a concept of Programmed control with probability 1 for stochastic dynamical system.

8. Programmed control with Prob. 1 for stochastic dynamical systems

Definition 1.6 [18, 19]. A PCP1 is called a control of stochastic system which allows the preservation with probability 1 of a constant value for the same function which depends on this systems position for time periods of any length T.

Let us consider the stochastic nonlinear jump of diffusion equations system:


where P, Zare given matrix functions and B, Lare the functions that may either be known or not. For such systems we construct a unit of programmed control utXtKtXtMtXtwhich allows the system (24) to be on the given manifold u(tXt)=u0x0with Prob. 1 (PCP1) for each t0T, T.

Suppose that the nonrandom function stXtis the first integral for the same stochastic dynamical system. The PCP1 utXtKtXtMtXtis the solution for the algebraic system of linear equations.

Theorem 1.4 Let a controlled dynamical system be subjected to Brownian perturbations and Poisson jumps. The unit of PCP1 utXtKtXtMtXt, allowing this system to remain with probability 1 on the dynamically structured integral mfd stXtxoω=s0xo, is a solution of the linear equations system (with respect to functions utxt), KtXt, MtXtwhich consists of Eq. (19) and Eq. (24). The coefficients of the Eq. (19), (and the coefficients of the Eq. (24) respectively) are determined by the theorem 7. The response to the random action is defined completely.

We show how the stochastic invariants theory can be applied to solve different tasks.

9. Stochastic models with invariant function which are based on deterministic model with invariant one

In this section we consider a few examples for application of the theory above to modeling actual random processes with invariants [23]. Firstly, we consider an example of construction of a differential equation system with the given invariant. Secondly, we study a general scheme for the PCP1 determination. And finally, we show the possibility of construction of stochastic analogues for classical models described by a differential equations system with an invariant function. The suggested method of stochastization is based on both the concept of the first integral for a stochastic differentialItô equations system (SDE) and the theorem for construction of the SDE system using its first integral.

9.1 Construction of a differential equations system

It is necessary to construct a differential equations system for XR3, t0such that the equality


is satisfied with Prob.1. The equality (25) means that the differential equations system has a first integral stXtY1tY2t=XtY12t+Y2t+etwith initial condition 0,1,0:


We have




Thus the new drift coefficients are


According to term 3 of Theorem 1.3, we will determine a coefficient for Poisson measure. Now we rename variables: ZZ1Z2Z3XY1Y2. Then, we have:


where a function VtZγsolves the a differential equations system:


and satisfies the initial condition V0=Z. Then, we determine functions g1γ,g2γ,g3γ.

Assume, that φ1γ=1γ+1, φ2γ=2γ, φ3γ=1. Then, we get:


Finally, we have constructed three variants of differential equations system:

1. deterministic differential equations system:


2. stochastic differential equations system (Itó diffusion equations):


3. stochastic differential equations system (jump-diffusion Itó equations):


We choose the functions q00, fiand hi, i=1,2,3, in accordance with the restriction of the task and taking into account the utility for modeling.

9.2 Transit from deterministic model with invariant to the same stochastic model

Now we describe a general scheme for application of the theory above.

The suggested method of stochastization is based on both the concept of the first integral for a stochastic differentialItô equations system (SDE) and the theorem for construction of the SDE system using its first integral.

Let us consider a classical model


with an invariant uty.

Then we construct the GSDE system, taking into account the equality utxt=u0x0=C:


Hence, the stochastic model has a representation


Further, we determine complementary function which is unit of control functions for PCP1:


Finally, we have constructed stochastic analogue for classical model described by a differential equations system and having an invariant function.

9.3 The SIR (susceptible-infected-recovered) model

The SIR is a simple mathematical model of epidemic [24], which divides the (fixed) population of Nindividuals into three” compartments” which may vary as a function of time t.

Stare those susceptible but not yet infected with the disease,

Itis the number of infectious individuals,

Rtare those individuals who have recovered from the disease and now have immunity to it,

the parameter λdescribes the effective contact rate of the disease,

the parameter μis the mean recovery rate.

The SIR model describes the change in the population of each of these compartments in terms of two parameters:


and its restrictsion is


Let the model with strong perturbation be




Suppose that the function utxyz=x+y+zNis a first integral, vtxyz=2et+xand htxyz=yare complementary functions, and qtxyz=xis arbitrary function. The initial condition is: x0=1, y0=0, z0=0. Then constructed differential equations system has the form


Let us simulate a numerical solution of Eg.(36), where N=1(for example). Figure 1 shows simulation for system without jumps, the Figure 2 shows the processes with jumps.

Figure 1.

Numerical solution forEq.(36)without jumps.

Figure 2.

Numerical solution forEq.(36)with jumps.

In such a way we could use the system of differential equations


as initial step for construction of stochastic SIR-model. A good choice of complementary functions vtxyzand htxyzallows us to obtain such coefficients that ensure that the solution xtytztof the differential equations system satisfy some reasonable limitations.

9.4 The predator–prey model

The Lotka - Volterra equations or the predator–prey equations used to describe the dynamics of biological systems in which two species interact, one as a predator and the other as prey.

The Lotka - Volterra model makes a number of assumptions, not necessarily realizable in nature, about the environment and evolution of the predator and prey populations:

  • The prey population finds ample food at all times.

  • The food supply of the predator population depends entirely on the size of the prey population.

  • The rate of change of population is proportional to its size.

  • During the process, the environment does not change in favor of one species, and genetic adaptation is inconsequential.

  • Predators have limitless appetite.

Let us note: N1tis the number of prey, and N2tis the number of some predator, ε1, ε2, η1and η2are positive real parameters describing the interaction of the two species.

The populations change through time according to the pair of equations:


Eq. (38) has the invariant function


where C=const.

We can introduce the stochastic model as a form


with condition


Let us assume that ε1=2, ε2=1, η1=η2=1, and C=1, and initial condition is x0=y0=1. The function utxy=x1exy2e2yis a first integral, htxy=yx+etand qtxy=xare complementary functions.

We cannot find an analytical solution of the differential equations system


Then, the constructed SDE system includes only Wiener perturbation:




Finally, we have the stochastic Lotka Volterra model associated to (38) (Nt=N1tN2t):


where AtNt, BtNt, CtNt, DtNt, EtNtare determined by Eq.(43).

Figures 3 and 4 show two realizations for numerical solution of Eq. (44).

Figure 3.

Numerical simulation 1 for solution ofEq.(44).

Figure 4.

Numerical simulation 2 for solution ofEq.(44).

Another examples of a differential equation system construction and models see in [25, 26, 27, 28, 29].


10. Conclusion

The invariant method widens horizons for constructing and researching into mathematical models of real systems with the invariants that hold out under any strong random disturbances.


  • Different variant of transliteration of the name: Dubko
  • Different variants of transliteration of this formula name: Itô – Wentcell, Itô – Venttcel’, Itô – Ventzell.

© 2021 The Author(s). Licensee IntechOpen. This chapter is distributed under the terms of the Creative Commons Attribution 3.0 License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

How to cite and reference

Link to this chapter Copy to clipboard

Cite this chapter Copy to clipboard

Elena Karachanskaya (February 23rd 2021). Invariants for a Dynamical System with Strong Random Perturbations, Advances in Dynamical Systems Theory, Models, Algorithms and Applications, Bruno Carpentieri, IntechOpen, DOI: 10.5772/intechopen.96235. Available from:

chapter statistics

132total chapter downloads

More statistics for editors and authors

Login to your personal dashboard for more detailed statistics on your publications.

Access personal reporting

Related Content

This Book

Next chapter

Stochastic Theory of Coarse-Grained Deterministic Systems: Martingales and Markov Approximations

By Michel Moreau and Bernard Gaveau

Related Book

First chapter

Some Unconstrained Optimization Methods

By Snezana S. Djordjevic

We are IntechOpen, the world's leading publisher of Open Access books. Built by scientists, for scientists. Our readership spans scientists, professors, researchers, librarians, and students, as well as business professionals. We share our knowledge and peer-reveiwed research papers with libraries, scientific and engineering societies, and also work with corporate R&D departments and government entities.

More About Us