Stochastic Dynamic Modeling of Damped

Lotka-Volterra System
Sandip Banerjee and C. G. Chakrabarti
Department of Applied mathematics
University of Calcutta
Calcutta- 700009, India
(SAMS 1998. Vol.30, pp. 1-10)

The object of the paper is to develop a stochastic dynamic model of damped

Lotka Volterra ecosystem and to examine the role of the technique of statistical
linearization in the stochastic analysis of stability and extinction of the ecosys-
tem.

Keywords: Damped Lotka-Volterra system; statistical linearization; mo-

ment equations; non-equilibrium fluctuation; stability

1 INTRODUCTION
Stochasticity plays a vital role in the structure and function of an ecosystem.
A central obstacle in the stochastic modeling of an ecosystem is the lack of
mathematical machinery available to analyze non-linear multi-dimensional sto-
chastic process [I]. A quantum leap in the mathematical sophistication of eco-
logical modeling occurred when MAY [2] introduced stochastic differential equa-
tions to investigate limits to niche overlap in randomly fluctuating environment.
Since then most stochastic multi-species models are derived by generalizing well-
known deterministic models, such as Lotka-Volterra equations, to include de-
mographic or environmental stochasticity. This involves non-linear stochastic
differential equations. The solutions of such non-linear stochastic differential
equations pose great difficulties. The finding of the probability density for the
determining the state of the system by solving non-linear Fokker-Planck equa-
tion resulting from the non-linear stochastic differential equation although is
desirable, but is not always possible analytically. For this different techniques
of linearization of non-linear stochastic differential equations giving rise to a set
of deterministic moment equations has been receiving a great deal of attention
in different fields of Science and Technology [3 - 10]. These approaches although
limited in their validity compared to the original non-linear Fokker-Planck equa-
tions have, however, great advantages in different branches of Applied Sciences.

1
In particular, it was pointed out by Jumarie [11] that the above moment tech-
nique can be used to solve a large class of problems of stochastic optimization
in stochastic optimal control.
In this paper we have used the techniques of statistical linearization which
consists of the replacement of a non-linear stochastic differential equation with
a linear one so as to make the ensemble average of the square of the error due to
replacement a minimum [6]. The advantage of this method over other lineariza-
tion techniques lies in its simplicity involving a few coupled linear equations
(two in the case of a single variable and five in the case of two variables). The
method is very useful even if we are concerned with the first two moments only.
This method also throws light on situation where there are multiple steady
states. Another advantageous point of this technique is its uniform validity for
all times. This technique has been successfully applied earlier to some ecolog-
ical systems [4, 6]. Lotka- Volterra (generalized) equation plays a significant
role in the theoretical development of ecosystems. A non-equilibrium thermo-
dynamic theory of generalized Lotka-Volterra system based on a generalized
measure of entropy was developed earlier to study the stability, evolution and
succession of an ecosystem [12,13]. The aim of the present paper is to carry out
the programme further to the stochastic regime to include the effect of random
environmental noise on the stability of the ecosystem. The method has provided
us with an expression of non-equilibrium fluctuation and the associated criteria
of stability or instability of a predator-prey system. The biological relevance of
this stochastic model and the results obtained has been investigated.

2 DAMPED LOTKA-VOLTERRA SYSTEM:BASIC

STOCHASTIC DIFFERENTIAL EQUATIONS
We consider a population of prey. population X1 (t) and a predator population
X2 (t) at time t. We assume that in absence of predator, the prey would obey the
logistic equation and the rate of cropping the prey is proportional to the product
of the prey and the predator population. This results in Damped Lotka-Volterra
equations (after reducing the number of parameters):

dX1
= X1 (1 − X1 ) − aX1 X2 (1)
dt
dX2
= aX1 X2 − bX2 (a, b > 0)
dt

The system of equations admits a single non-zero steady state solution

b a−b
X1∗ = , X∗ = (2)
a 2 a2

2
 The constants a and b are both positive by definition and we also assume
a > b which guarantees the stability of the deterministic system [14], To study
the behavior of the system about the steady state (2) we put

X1 = x + X1∗ , Y1 = y + Y1∗ (3)

and the system of equation (1) reduces to the form (in terms of the deviation
variables (x, y)):

dx b
= − x − x2 − by − axy (4)
dt a
dy a−b
= axy + x (5)
dt a

These are basic deterministic ordinary differential equations governing the

system behavior about the steady state. The solutions {x(t), y(t)} of (4) and (5)
subject to known initial values {x(t0 ), y(t0 )} represent the state of the system
at time t > t0 .
If the random aspects of the system are to be concerned. a number of mod-
ifications can be made in the formation of the initial value problem (4) and (5).
We shall however, follow the case when the r. h. s. of the equations (4) and (5)
are extended by imposing random input or perturbation. More preciously we are
concerned with stochastic differential equations which are driven by’ Gaussian
white-noises and interpreted mathematically as Ito equations. Gaussian white
noise which is a delta-correlated random process is very irregular and as such
it is to be treated carefully. In spite of this it is an immensely useful concept
to model rapidly fluctuating phenomenon. Many phenomena of physical and
natural systems such as electrical resistance, Brownian particle, climate fluc-
tuation etc. are white noises to a very good approximations and support the
usefulness of the white noises idealization in application to ecological systems
[15]. To study the fluctuation about the steady state, the equations (4) and (5)
are extended to Ito type of stochastic differential equations (non-linear coupled
bivariate Langevin equations):

dx b
= − x − x2 − by − axy + η1 (t) (6)
dt a
dy a−b
= x + axy + η2 (t) (7)
dt a

where η1 (t) and η2 (t) are assumed to be Gaussian white-noises satisfying the
conditions:

hηi (t)i = 0 (8)

0 0
hηi (t)ηj (t ) = 2i δij (t − t )(i, j = 1, 2)

3
 where i is the intensity or strength of the random perturbation, δij are Dirac
delta functions and the bracket h·i represents the ensemble average. Equations
(6) and (7) are the basic non-linear stochastic differential equations determining
the statistical behavior of the system near the steady state.

3 STATISTICAL LINEARIZATION: MOMENT

EQUATION
The statistical linearization of the equations (6) and (7) consists of replacing
the equations by the system of linear equations [6]:

dx
= α1 x + β1 y + C1 + η1 (t) (9)
dt
dy
= α2 x + β2 y + C2 + η2 (t) (10)
dt

where the errors in the above linearization are given by

b
e1 = − x − x2 − by − axy − α1 x − β1 y − C1 (11)
a
a−b
e2 = x + axy − α2 x − β2 y − C2 (12)
a

Minimization of the average of the squares of the errors (11) and (12) and
the assumption of ergodicity, helps us to determine the unknown coefficients
α1 , α2 , β1 , β2 of (9) and (10). Simple calculations lead to the equation of the
first two moments:

d<x> b
= − < x > − < x2 > −b < y > −a < xy > (13)
dt a
d<y> a−b
= < x > +a < xy > (14)
dt a
d < x2 > b
= 2[− < x2 > − < x3 > −b < xy > −a < x2 y > +1 ] (15)
dt a
d < y2 > a−b
= 2[ < xy > +a < xy 2 > +2 ] (16)
dt a
d < xy > b
= − < xy > − < x2 y > −b < y 2 > −a < xy 2 > (17)
dt a
b
+ (1 − ) < x2 > +a < x2 y >
a

4
 where we have used the relation

< xη1 >= 1 , < xη2 >= 0, < yη2 >= 2 , < yη1 >= 0 (18)

which can be obtained in the linear approximation.

Let us now assume that the system size expansion is valid such that all correla-
tion and variance are of the order of N1 compared to the averages [16,17],
<x> <y>
< xy > ∝ o[ ]oro[ ] (19)
N N
<x>
< x2 > ∝ o[ ]
N
<x>
and < x2 > ∝ o[ ]
N

where N is the population of the system.

We also assume that the correlations 1 and 2 given by (18) decrease with
the increase of the population size and they are assumed to be of the order of
the inverse of the population of size N [6]
1
i ∝ o( )(i = 1, 2) (20)
N

Then using the expressions [6]

< x2 y > = 2 < x > {< xy > − < x >< y >}+ < x2 >< y > (21)
< xy 2 > = 2 < y > {< xy > − < x >< y >}+ < x >< y 2 >
< x3 > = 3 < x >< x2 > −2 < x >3

and keeping the lowest order terms and replacing the averages < x > and
y > by their steady state values < x >= 0 and < y >= 0 we get the following
reduced equations:
2b
[D + ] < x2 > = −2b < xy > (22)
a
a−b
D < y2 > = 2 < xy > (23)
a
b b a−b
(D + ) < xy > = − < xy > − < x2 > −b < y 2 > (24)
a a a
d
where D ≡ dt .
Equations (22)-(24) are the required ordinary coupled linear equations of mo-
ments.

5
4 NON-EQUILIBRIUM FLUCTUATION AND
STABILITY
We multiply (24) by [D + 2b a ] and using (22) and (23); again multiplying the
result by D and using (23) and (24), we get

b b2 b b2
[D3 + 3 D2 + (2 2 + 4 (a − b))D + 4 2 (a − b)] < xy > = 0 (25)
a a a a

Let < xy >= Aemt be the trial solution so that

b b2 b b2
m3 + 3 m2 + (2 2 + 4 (a − b))m + 4 2 (a − b) = 0 (26)
a a a a

is the auxiliary equation.

4b3
9a3 [4(a − b) − ab ]3 is the discriminant of the cubic.

Case I If 4(a − b) − ab > 0, then (26) have one real and two imaginary roots
and the required solution is

√ √
< xy > = A1 + A2 cos( 3Ht) + A3 sin( 3Ht) (27)
√ √
< y 2 > = B1 t + B2 cos( 3Ht) + B3 sin( 3Ht) + C0 (28)
√ √
< x2 > = c1 t + c2 cos( 3Ht) + c3 sin( 3Ht) + D0 (29)

where A0i s, Bi (a, b, Ai )0 s and Ci (a, b, Ai , Bi )0 s are constants (i = 1, 2, 3). C0

and D0 are constants of integration.

Case II If 4(a − b) − ab < 0, then the roots of (26) are all real and in this
case the required solution is

√ √
< xy > = L1 + L2 e H0 t
+ L3 e− H0 t
(H0 = −H) (30)
√ √
2 H0 t − H0 t
< y > = M 1 t + M2 e + L3 e + K1 (31)
√ √
2 H0 t − H0 t
< x > = N1 t + N2 e + N3 e + K2 (32)

where L0i s, Mi (a, b, Li )0 s and Ni (a, b, Li , Mi )0 s are constants (i = 1, 2, 3). K1

and K2 are constants of integration.

Let us now study the effects of random perturbation on the stability of

damped Lotka-Volterra system. Two cases arise according as H > 0 or H < 0.

6
For H > 0, the moments < y 2 > and < x2 > given by (28) and (29) respectively
increase linearly with time with a periodic background noise but the co-variance
< xy > given by (27) remains finite, being a periodic function of time except an
additive constant. For H < 0, each of < xy >, < y 2 > and < x2 > given by (30),
(31) and (32) respectively diverges with increasing time. Both the cases H > 0
or H < 0 are statistically unstable; for both the cases, the prey and the predator
population will have fluctuations so severe which lead to complete shattering of
the system, implying the extinction of the population [18]. There is, however
a difference between the two cases. In the case H > 0, the approach to the
extinction is somewhat in an organized manner, as the covariance < xy > or the
interrelation between the prey and the predator remains finite throughout. We
thus observe that the damped Lotka-Volterra system which in the deterministic
case .is stable under the condition a > b becomes statistically unstable under the
random perturbation subject to some conditions. Not only that, the predator
and the prey population go to extinction as time evolves.

5 CONCLUSION
In conclusion wish to add a few points on the utility of the present stochastic
model and analysis for the biological problem under consideration. We first note
that the present stochastic analysis is capable in explaining the observational re-
sults of some biological experiments conducted on predator-prey systems [19,20].
For example we consider the experiment by Gause [19] conducted on a preda-
tor, Didinium Nasutum and a prey species, Paramecium Candatum. When the
two species were cultured together in an infusion of oats with no place for P.
Candatum to hide, the prey population became extinct in all cases, the preda-
tor dying of starvation shortly afterwards. These results can be interpreted in
two ways: (i) either the predator have exploited their ways or (ii) the para-
mecium became extinct because of chance fluctuation [20]. We also note that
both the deterministic Lotka-Volterra and Damped Lotka-Volterra models fail
to mimic the results of Gause’s experiment. A possible explanation of their
failure opens the possibility of chance extinction of predator and prey popula-
tions. The stochastic theory presented here thus provides within the analysis
of non-equilibrium fluctuation and stability, an explanation of the extinction of
predator-prey population of a biological community in confirmation with the
well-known experimental results.

Acknowledgement The first author (S.B) wishes to thank C.S.I.R. (India)

for a Senior Research Fellowship.

References
[1] Turelli, M. (1986). Stochastic Community Theory: A Partially Guided
Tour. In Iarhemarical ecology ed. by T. G. Hallam and S. Levin. Springer-
Verlag: Berlin.

7
 [2] May, R. M. (1974). Stability and Complexity in Model Ecosystem.
Princeton Univ. Press. Princeton.

[3] Socha, L. and Soong, T. T. (1991). Linearization in analysis of a non-

linear Stochastic System. Appl. Jtech. Rev., 44. 399-422.

[4] Sandip Banerjee, and Chakrabarti, C. G. (1996). Stochastic Model of

Symmetric Lotka-Volterra Competition System: Non-Equilibrium Fluctuation
and Stability, Bull/. Cal. Marh. Soc.. 88, 235-244.

[5] West. B. 1. er al. (1978). StUdies in non-linear stochastic process IV, J.

Slat. Physics, 18,217-233.

[6] VaIsakumar, N. C. et al. (1983). On the linearization of Non-linear

Langevin type Stochastic Differential Equations, J. Statist. Phys., 30, 617-631.

[7] Haken, H. (1983). Advanced Synergetics, Springer-Verlag, Berlin.

[8] Baishya, M. C. and Chakrabarti, C. G. (1987). Non-equilibrium Fluctu-

ation in Volterra-Lotka systems, Bull. Math. Bioi., 49, 125-131.

[9] Samanta, G. P. and Chakrabarti, C. G. (1991). Stochastic Behavior of

a damped Volterra-Lotka system with time-dependent parameter, Jour. Math.
Phy. Sci., 25, 4, 399-422.

[10] Samanta, G. P. and Chakrabarti, C. G. (1989). Stochastically perturbed

Hopf Bifurcation in an extended Volterra-Lotka system, Appl. Math. Lel/s., 2,
163 -166.

[11] Jumarie, G. (1995). A Practical Variational Approach to Stochastic

Optimal Control via state moment equations, Journal of Franklin Institute:
Engineering and Applied Mathematics, 332, 761- 772.

[12] Chakrabarti, C. G. Sutapa Ghosh and Syamali Bhadra, (1995). Non-

Equilibrium Thermodynamics of Lotka-Volterra Ecosystems: Stability and Evo-
lution, J. BioI. Phys., 21, 273-284.

[13] Chakrabarti, C. G. and Sutapa Ghosh. (1995). Entropic Model and

Analysis of Complex Systems, SAMS, 23, 103-112.

[14] Nisbet, R. M. and Gurney, S. C. (1982). Mode1ling Fluctuating Popu-

lation. Wiley and Sons, New-York.

[15] Horsthemke, W. and Lefever, R. (1984). Noise Induced Transitions,

Springer Verlag, Berlin.

8
 [16] VanKampen, N. G. (1981). Stochastic Processes in Physics and Chem-
istry, North Holland. Pub!. Co. Amsterdam.

[17] Nicolis, G. and Prigogine, I. (1977). Self-organization in Non-Equilibrium

System. Wiley and Sons. New-York.

[18] Svirezhev, Yu, M. and Logofet, D. O. (1983). Stability of Biological

Community, Mir Pub!. Moscow.

[19] Gauss, G. F. (1934). The struggle for existence, Wi1liams and Wilkins,
Baltimore.

[20] Poole, R. W. (1974). Introduction to Quantitative Ecology, Mc-Graw

Hill, New York.