This action might not be possible to undo. Are you sure you want to continue?
in a Fluctuating Environment
Yi Ming Lai,
1
Jay Newby,
1
and Paul C. Bressloff
1,2
1
Mathematical Institute, University of Oxford, 2426 St Giles, Oxford OX1 3LB, United Kingdom
2
Department of Mathematics, University of Utah, Salt Lake City, Utah 84112, USA
(Received 7 June 2011; published 8 September 2011)
We use the theory of noiseinduced phase synchronization to analyze the effects of demographic noise
on the synchronization of a metapopulation of predatorprey systems within a ﬂuctuating environment
(Moran effect). Treating each local predatorprey population as a stochastic urn model, we derive a
Langevin equation for the stochastic dynamics of the metapopulation. Assuming each local population
acts as a limit cycle oscillator in the deterministic limit, we use phase reduction and averaging methods to
derive the steadystate probability density for pairwise phase differences between oscillators, which is
then used to determine the degree of synchronization of the metapopulation.
DOI: 10.1103/PhysRevLett.107.118102 PACS numbers: 87.23.Cc, 02.50.Ey, 05.40.Àa
A major problem in ecology is understanding the
mechanisms for synchronizing spatially separated popula
tions or patches (metapopulations) [1]. It is important in
conservation because synchrony is strongly correlated with
the chances of global extinction [2]. On the other hand,
synchrony may be desirable in the case of pest or pathogen
control, since it can help to eliminate an outbreak [3].
There are two basic mechanisms for synchronizing patches
within a metapopulation: isolated patches can be driven by
the same environmental ﬂuctuations (the socalled Moran
effect [4]) or patches can interact with each other through
dispersal in a constant environment [5–7]. The dominant
mechanism will depend on the spatial scale of the meta
population and the nature of the local patch dynamics.
In this Letter, we use the theory of noiseinduced phase
synchronization to analyze the effects of demographic
noise on the synchronization of a metapopulation of
predatorprey systems within a ﬂuctuating environment.
Noiseinduced phase synchronization concerns the general
dynamical theory of how a global extrinsic noise source
can synchronize an ensemble of uncoupled limit cycle
oscillators [8,9]. Although the theory has applications to
a wide range of systems in the physical and life sciences
(most recently neuroscience), it has not previously been
applied within ecology, even though the Moran effect has
been known for a long time. Moreover, in most previous
applications, all sources of noise are extrinsic to the oscil
lator rather than partially arising from ﬁnite size effects
(see [10] for an analysis of demographic noise in neural
population oscillators). We start by incorporating demo
graphic noise into a single predatorprey system using a
stochastic urn model [11]. Approximating the associated
master equation using a KramersMoyal expansion [12],
we derive a Langevin equation for an ensemble of
predatorprey systems. We show that the multiplicative
Gaussian noise terms can be decomposed into a set of
independent white noise processes that are uncorrelated
across populations (demographic noise) and an additional
white noise term that is common to all populations (envi
ronmental noise). Assuming that each predatorprey sys
tem acts as a limit cycle oscillator in the deterministic
limit, we use phase reduction and averaging methods
to derive the steadystate probability density for pairwise
phase differences between oscillators, which is then used
to determine the degree of synchronization of the metapo
pulation. We illustrate this using the HollingTanner
model [13,14].
Suppose that at time i there are m individuals of species
C
1
(prey) and n individuals of species C
2
(predators). Let
the total number of individuals within a patch have an
upper bound N and use the symbol E to denote the set of
N ÀmÀn available slots for reproduction and intraspe
cies competition. Following Ref. [11], we consider a gen
eral stochastic urn model of a predatorprey system
consisting of the following processes: birth processes
C
]
E!
l
]
C
]
C
]
, death processes C
]
!
J
]
E, intraspecies com
petition C
]
C
]
!
c
]
C
]
E, and predatorprey interactions
C
1
C
2
!
¡
1
C
2
E, C
1
C
2
!
¡
2
C
2
C
2
. Here l
]
, J
]
, c
]
are the birth,
death, and competition rates, ¡
1
is the rate of prey con
sumption due to predation, and ¡
2
is the growth rate of
predators at the expense of prey. Suppose that a pair of
constituents is drawn with probability µ and a single
constituent is drawn with probability 1 Àµ. We can then
write down the following transition rates T
r,s
ðm, nÞ from
state (m, n) to (mþr, n þs) [11]:
T
À1,0
ðm, nÞ ¼ð1ÀµÞJ
1
m,Nþ2µc
1
,
m,mÀ1
þ2µ¡
1
,
m,n
,
T
0,À1
ðm, nÞ ¼ð1ÀµÞJ
2
n,Nþ2µc
2
,
n,nÀ1
,
T
1,0
ðm, nÞ ¼2µl
1
,
m,NÀmÀn
,
T
0,1
ðm, nÞ ¼2µl
2
,
n,NÀmÀn
,
T
À1,1
ðm, nÞ ¼2µ¡
2
,
m,n
,
PRL 107, 118102 (2011)
P HYS I CAL RE VI E W L E T T E RS
week ending
9 SEPTEMBER 2011
00319007,11,107(11),118102(5) 1181021 Ó 2011 American Physical Society
where ,
n,m
¼ nm,ðNðN À 1ÞÞ. The master equation for
the probability P
m,n
ðiÞ that the system consists of m prey
and n predators at time i then takes the form
JP
m,n
Ji
¼
X
r,s
½T
r,s
ðm
0
, n
0
ÞP
m
0
,n
0 ÀT
r,s
ðm, nÞP
m,n
, (1)
with m
0
¼ mÀr, n
0
¼ n Às.
An exact solution of the master equation (1) can rarely
be found, so it is necessary to consider some form of
approximation. In particular, for large N we can carry out
a KramersMoyal expansion of the master equation (1)
[12]. Introduce new variables x
1
¼ m,N, x
2
¼ n,N, and
set P
m,n
ðiÞ ¼ Pðx
1
, x
2
, iÞ. Treating x
1
, x
2
as continuous
variables and Taylor expanding each term on the right
hand side of Eq. (1) to second order in 1,N yields the
FokkerPlanck (FP) equation
·P
·i
¼ À
X
¼1,2
·ðA

ðxÞPÞ
·x

þ
e
2
2
X
,]¼1,2
·
2
ðG
]
ðxÞPÞ
·x

·x
]
, (2)
where x ¼ ðx
1
, x
2
Þ, e ¼ 1,
ﬃﬃﬃﬃ
N
p
( 1,
A
1
ðxÞ ¼2l
1
x
1
ð1Àx
1
Àx
2
Þ À2c
1
x
2
1
ÀJ
1
x
1
À2ð¡
1
þ¡
2
Þx
1
x
2
,
A
2
ðxÞ ¼2l
2
x
2
ð1Àx
1
Àx
2
Þ À2c
2
x
2
2
ÀJ
2
x
2
þ2¡
2
x
1
x
2
, (3)
and G
]]
ðxÞ is obtained from A
]
ðxÞ by reversing the minus
sign on the last three terms, whereas G
12
ðxÞ ¼ G
21
ðxÞ ¼
À2¡
2
x
1
x
2
. We have rescaled the various rates according
to J
k
! ð1 ÀµÞJ
k
,N, l
k
! µl
k
,ðN À 1Þ, c
k
! µc
k
,
ðN À 1Þ, ¡
k
! µ¡
k
,ðN À 1Þ.
Next we introduce the three stoichiometric vectors
v
1
¼ ð1, 0Þ
T
, v
2
¼ ð0, 1Þ
T
, v
3
¼ ðÀ1, 1Þ
T
, which (up to a
sign) represent all possible increments in the number of
predator and prey induced by single interactions. For ex
ample, predation C
1
C
2
! C
2
C
2
is represented by the stoi
chiometric vector ðÀ1, 1Þ
T
. It can then be shown that the
solution to the FP equation (2) determines the probability
density function for a corresponding stochastic process
X ¼ ðX
1
, X
2
Þ that evolves according to an Ito type
Langevin equation JX ¼ AðXÞJi þe
P
3
¼1
B
ðÞ
ðXÞJW

,
where B
ð]Þ
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
G
]]
ðxÞ À 2¡
2
x
1
x
2
q
v
]
, ] ¼ 1, 2, and B
ð3Þ
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2¡
2
x
1
x
2
p
v
3
. Here W

denotes an independent Wiener
process such that hJW

ðiÞi ¼ 0 and hJW

ðiÞJW
]
ðiÞi ¼
u
,]
Ji. The Langevin equation approximates the effects of
demographic noise arising from the ﬁnite size N of the
local predatorprey population. It is also possible to include
a common extrinsic noise source due to environmental
ﬂuctuations. For concreteness, suppose that we add a sto
chastic term to the birth rates l
k
according to l
k
! l
k
þ
a(ðiÞ,2, where (ðiÞ is a white noise term and a is the
strength of environmental noise. Substituting for l
k
in
Eq. (3) and expanding to OðaÞ leads to an additional
multiplicative term in the Langevin equation of the form
aHðXÞJW, where H ¼ ðH
1
, H
2
Þ with H
1
ðXÞ ¼ r
1
x
1
ð1 Àx
1
Àx
2
Þ, H
2
ðXÞ ¼ r
2
x
2
ð1 Àx
1
Àx
2
Þ, and JWðiÞ ¼
(ðiÞJi is an additional independent Wiener process, which
is treated in the sense of Stratonovich.
In the deterministic limit (e, a ! 0) the Langevin equa
tion reduces to the deterministic planar dynamical system
_ x
1
¼ A
1
ðxÞ, _ x
2
¼ A
2
ðxÞ with A
]
given by Eq. (3). Suppose
that the deterministic systemsupports a limit cycle solution
x ¼ x
Ã
ðiÞ with x
Ã
ði þkTÞ ¼ x
Ã
ðiÞ for all integers k, where
T is the period of oscillations. Now consider an ensemble
of N identical, spatially separated predatorprey patches,
each of which acts as a limit cycle oscillator in the deter
ministic limit. If we ignore spatial interactions (dispersal,
migration) between the isolated patches, then we have a
system of uncoupled population oscillators driven by a
common environmental noise source. Introducing the label
µ, µ ¼ 1, . . . , N, the Langevin equation for the ensemble
of patches is
JX
ðµÞ
¼ A
ðµÞ
Ji þe
X
3
¼1
B
ð,µÞ
JW
ðµÞ

þaH
ðµÞ
JW, (4)
where A
ðµÞ
¼ AðX
ðµÞ
Þ, etc. We associate an independent
set of Wiener processes W
ðµÞ
]
with each population oscil
lator (demographic noise) but take the environmental noise
to be common to all the oscillators: hJW
ðµÞ

ðiÞJW
ð:Þ
]
ðiÞi ¼
u
,]
u
µ,:
Ji, hJW
ðµÞ

ðiÞJWðiÞi ¼ 0, hJWðiÞJWðiÞi ¼ Ji.
Following previous studies of noiseinduced phase syn
chronization [8–10], we now carry out a stochastic phase
reduction of the Langevin equation (4). First, we introduce
the phase variable 0 2 ðÀr, r such that the dynamics of
a single oscillator in the absence of noise reduces to the
simple phase equation
_
0 ¼ w
0
, where w
0
¼ 2r,T. We
then extend the notion of phase into some neighborhood
M& R
2
of the limit cycle using an isochronal mapping
Å: M! ðÀr, r, with 0 ¼ ÅðxÞ, assuming that the
limit cycle is sufﬁciently attracting so that for small e
and a, the dynamics can be restricted to the neighborhood
Mwith high probability [15]. This allows us to deﬁne a
stochastic phase variable for each oscillator according to
Â
ðµÞ
¼ ÅðX
ðµÞ
Þ with X
ðµÞ
evolving according to the
Langevin equation (4). After converting the demographic
noise terms in Eq. (4) to Stratonovich form [which adds an
Oðe
2
Þ correction to the drift terms [10] ], we perform a
change of variables X
ðµÞ
! ðÂ
ðµÞ
, r
ðµÞ
Þ and then project
out the phase dynamics along the lines of Ref. [16]. This
ultimately leads to an Ito Langevin equation for the sto
chastic phase variables Â
ðµÞ
of the form
JÂ
ðµÞ
¼ w
ðµÞ
Ji þe
X
3
¼1
¡
ðµÞ

JW
ðµÞ

þan
ðµÞ
JW, (5)
with ¡
ðµÞ

¼¡

ðÂ
ðµÞ
Þ ZðÂ
ðµÞ
Þ Á B
ðÞ
ðÂ
ðµÞ
Þ, n
ðµÞ
¼
nðÂ
ðµÞ
Þ ZðÂ
ðµÞ
Þ Á HðÂ
ðµÞ
Þ and w
ðµÞ
¼ wðÂ
ðµÞ
Þ
w
0
þe
2
1
ðÂ
ðµÞ
Þ þa
2
2
ðÂ
ðµÞ
Þ. Here Z ¼ ðZ
1
, Z
2
Þ
PRL 107, 118102 (2011)
P HYS I CAL RE VI E W L E T T E RS
week ending
9 SEPTEMBER 2011
1181022
where Z
k
, k ¼ 1, 2, is the kth component of the inﬁnitesi
mal phase resetting curve (PRC), which is deﬁned as
Z
k
ð0Þ ·
x
k
Åj
x¼x
Ã
ð0Þ
with
P
k¼1,2
Z
k
ð0ÞA
k
ðx
Ã
ð0ÞÞ ¼ w.
The PRC is the unique, normalized 2rperiodic solution
of the adjoint equation [17]
_
Z
k
¼ À
P
]¼1,2
A
]k
ðx
Ã
ðiÞÞZ
]
ðiÞ,
where A
]k
¼ ·A
]
,·x
k
. All terms in Eq. (5) are evaluated
on the limit cycle so that, for example, B
ðÞ
ð0Þ ¼
B
ðÞ
ðx
Ã
ð0ÞÞ. As ﬁrst shown in Ref. [16], and further devel
oped in Refs. [18], considerable care must be taken in
carrying out the phase reduction procedure in the presence
of Gaussian white noise in order to obtain the correct form
of the drift terms
1
,
2
. However, the latter will not be
important in the following analysis.
Following Ref. [9], we now introduce slow phase
variables c
ðµÞ
¼ Â
ðµÞ
Àw
0
i and set Qðfc
ðµÞ
g, iÞ ¼
Pðfc
ðµÞ
Àw
0
ig, iÞ where P is the probability density of
the phases Â
ðµÞ
. For small a and e, Q is a slowly varying
function of time, so we can average the FP equation for
Q over one cycle of length T ¼ 2r,w
0
to get ·
i
Q¼
À
"
P
N
µ¼1
·
c
ðµÞ Qþ
1
2
P
N
µ¼1
P
N
:¼1
·
c
ðµÞ
c
ð:Þ ½
"
C
ðµ:Þ
Q, where
"
¼
1
2r
R
r
Àr
½e
2
1
ð0Þ þa
2
2
ð0J0,
"
C
ðµ:Þ
¼a
2
gðc
ðµÞ
À
c
ð:Þ
Þ þe
2
hð0Þu
µ,:
with gðcÞ ¼
1
2r
R
r
Àr
nð0
0
Þnð0
0
þcÞJ0
0
and hðcÞ ¼
1
2r
R
r
Àr
P
3
¼1
¡

ð0
0
Þ¡

ð0
0
þ cÞJ0
0
. In order
to characterize the level of synchrony, we focus on a
pair of population oscillators (N ¼ 2). Performing
the change of variables c ¼ ðc
ð1Þ
þ c
ð2Þ
Þ,2, ç ¼
c
ð1Þ
À c
ð2Þ
, and looking for separable steadystate solu
tions Qðc
ð1Þ
, c
ð2Þ
Þ ¼ ÉðcÞÈðçÞ, it can be shown that [9]
ÈðçÞ ¼ À
0
½a
2
ðgð0Þ ÀgðçÞÞ þe
2
hð0Þ
À1
(6)
with À
0
a normalization constant, and ÉðcÞ ¼ 1,2r. We
see that ÈðçÞ is independent of the drift terms
]
ð0Þ.
In the thermodynamic limit N ! 1 we have e ¼
N
À1,2
! 0 so that the independent noise source vanishes.
The distribution ÈðçÞ then diverges at ç ¼ 0 while keep
ing positive since it can be shown that gð0Þ ! gð0Þ [9].
Hence, the phase difference between any pair of oscillators
accumulates at zero, resulting in complete noiseinduced
synchronization. Thus, a randomly ﬂuctuating global en
vironmental signal can synchronize a metacommunity of
deterministic predatorprey systems in the absence of dis
persal, consistent with the Moran effect [4]. However, for
ﬁnite N, demographic noise broadens the distribution of
phase differences, resulting in the desynchronization of the
metacommunity. As we now show, the width of ÈðçÞ is a
good measure of the desynchronizing effects of demo
graphic noise.
As an illustrative example of an oscillatory predator
prey system, consider the HollingTanner model [13,14]
_ x
1
¼ rx
1
1 À
x
1
K
À
mx
1
x
2
A
0
þx
1
, _ x
2
¼ sx
2
1 À
hx
2
x
1
.
(7)
In this model, the prey exhibit logistic growth up to a
carrying capacity K in the absence of predation. The
predator consumes prey according to a MichaelisMenten
or Holling typeII functional response and grows logisti
cally up to a carrying capacity x,h proportional to the
current level of prey in the system. It is straightforward
to ﬁnd a set of parameters for which Eq. (7) supports a limit
cycle [14] and to numerically evaluate the corresponding
PRCs, see Fig. 1(a). Comparing the righthand side of
Eqs. (7) with the drift functions of Eq. (3), we set l
1
¼
r,2K, l
2
¼ ðs þJ
2
Þ,2, J
1
¼ rð1,K À 1Þ, c
1
¼ 0, c
2
¼
sh,2x
1
À ðs þJ
2
Þ,2, ¡
1
¼ m,½2ðA
0
þx
1
Þ À ðs þJ
2
Þ,
2 Àr,2K, ¡
2
¼ ðs þJ
2
Þ,2, where J
2
is a free parameter.
For concreteness, we set J
2
¼ s. We can now determine
the functions g, h in Eq. (6) for an ensemble of Holling
Tanner oscillators using the given PRC, see Fig. 1(b).
In Fig. 2 we relate our analytical results for the distri
bution ÈðçÞ to direct numerical simulations of the
Langevin Eq. (4) for the HollingTanner model. The simu
lations are carried out using an EulerMaruyama scheme,
with the Stratonovich environmental noise treated as a
solution of a zeromean OrnsteinUhlenbeck process with
a small but ﬁnite correlation time. In order to quantify the
effects of demographic noise on the metapopulation, we
introduce ensemble averaged concentrations of prey and
predator according to
"
X
]
ðiÞ ¼ N
À1
P
N
µ¼1
X
ðµÞ
]
ðiÞ, ] ¼ 1,
2. For a high ratio of extrinsic to intrinsic noise, there is
signiﬁcant synchronization of the metapopulation, as in
dicated by the sharp peak of the steadystate distribution
ÈðçÞ at ç ¼ 0 in Fig. 2(a) and the corresponding cluster
ing of the individual oscillators on the deterministic limit
cycle [Fig. 2(d)]. It can also be seen that
"
X
]
ðiÞ traces out a
trajectory in the phaseplane that remains close to the
deterministic limit cycle. For a moderate ratio of extrinsic
to intrinsic noise, the metapopulation is only partially
synchronized. The distribution È is now much broader
[Fig. 2(b)] and, although the oscillators tend to remain
close to the deterministic limit cycle, they are no longer
tightly clustered. Thus,
"
X
]
ðiÞ still exhibits some oscillatory
behavior but the amplitude of oscillations is reduced
0 2 4 6
−200
0
100
200
300
400
phase θ
Z
(
θ
)
Z
1
Z
2
(a)
0 1 2 3
1
0
1
2
φ
g
(
φ
)
(b)
X2
X1
0 0.1 0.2 0.3
0.04
0.06
0.08
500
0
500
1000
1500
h
(
φ
)
−100
FIG. 1. (a) Plot of components Z
1
, Z
2
of PRC for a limit cycle
solution of the HollingTanner equations with r ¼ 1, A
0
¼ 0.1,
s ¼ 0.1, h ¼ 2, m ¼ 3, and K ¼ 0.5. (b) Corresponding func
tions gðçÞ and hðçÞ. Inset shows limit cycle.
PRL 107, 118102 (2011)
P HYS I CAL RE VI E W L E T T E RS
week ending
9 SEPTEMBER 2011
1181023
[Fig. 2(e)]. Finally, for a low ratio of extrinsic to intrinsic
noise, the metapopulation is completely desynchronized.
The density È is almost ﬂat [Fig. 2(c)], and individual
oscillators are much more scattered in the phaseplane so
that
"
X
]
ðiÞ exhibits small ﬂuctuations about the mean of the
limit cycle [Fig. 2(f)]. In order to further quantify how
demographic noise affects synchrony, we introduce the
order parameter RðiÞ ¼ N
À1
j
P
N
µ¼1
e
Â
ðµÞ
ðiÞ
j and the
halfwidth Á with
R
Á,2
ÀÁ,2
ÈðçÞJç ¼ 0.5. In Fig. 3 we
show how the halfwidth Á provides an excellent predictor
of the level of synchrony by comparing its dependence on
system size N with the time average of RðiÞ.
In conclusion, we have applied stochastic phase reduc
tion and averaging methods to analzye an important
ecological problem, namely, the effects of demographic
noise on the synchronization of a metapopulation in a
ﬂuctuating environment (Moran effect). We have shown
that the degree of synchronization of the metapopulation
can be characterized in terms of the probability density for
pairwise phase differences. One major implication of our
work is that the stabilizing effects of demographic noise
[Fig. 2(f)] could provide an explanation for why oscilla
tions are often not observed in real ecological systems,
contrary to predictions from deterministic models [19].
There are numerous ecologically motivated extensions of
this work, including dispersal between local patches, spa
tial heterogeneity [20], and the decay of synchrony with
spatial separation of the patches [1,2].
This work was partially supported by King Abdullah
University of Science and Technology Grant No. KUKC1
01304.
[1] P. J. Hudson and I. M. Cattadori, Trends Ecol. Evol. 14, 1
(1999).
[2] I. Hanski, Nature (London) 396, 41 (1998).
[3] D. J. D. Earn, P. Rohani, and B. T. Grenfell, Proc. R. Soc.
B 265, 7 (1998); P. Rohani, D. J. D. Earn, and B. T.
Grenfell, Science 286, 968 (1999).
[4] P. A. P. Moran, Aust. J. Zool. 1, 291 (1953).
[5] L. M. Pecora and T. L. Carroll, Phys. Rev. Lett. 80, 2109
(1998).
[6] V. A. A. Jansen and A. L. Lloyd, J. Math. Biol. 41, 232
(2000).
[7] D. A. Vasseur and J. W. Fox, Nature (London) 460, 1007
(2009).
[8] A. S. Pikovsky, Radiophys. Quantum Electron. 27, 576
(1984); J. N. Teramae and D. Tanaka, Phys. Rev. Lett. 93,
204103 (2004); D. S. Goldobin and A. Pikovsky, Phys.
Rev. E 71, 045201 (2005); C. Ly and G. B. Ermentrout, J.
Comput. Neurosci. 26, 425 (2008).
[9] H. Nakao, K. Arai, and Y. Kawamura, Phys. Rev. Lett. 98,
184101 (2007).
[10] P. C. Bressloff and Y. M. Lai, J. Math. Neurosci. 1, 2
(2011).
[11] A. J. McKane and T. J. Newman, Phys. Rev. E 70, 041902
(2004); Phys. Rev. Lett. 94, 218102 (2005).
[12] C. W. Gardiner, Stochastic Methods: A Handbook for the
Natural and Social Sciences (Springer, New York, 2009),
4th ed.
[13] J. T. Tanner, Ecology 56, 855 (1975).
[14] A. Gasull, R. E. Kooij, and J. Torregrosa, Publ. Math. 41,
149 (1997).
[15] In the case of Gaussian noise there is a small but nonzero
probability for large deviations under which the phase
reduction method could break down. However, this does
not affect the theoretical arguments presented here.
[16] K. Yoshimura and K. Arai, Phys. Rev. Lett. 101, 154101
(2008).
0 0.5 1 1.5 2 2.5 3
0
4
8
12
Φ
0 0.1 0.2 0.3
0.04
0.06
0.08
X2
(a) (d)
Φ X2
(b)
(e)
0 0.5 1 1.5 2 2.5 3
0.15
0.16
0.17
0.18
0 0.1 0.2 0.3
0.04
0.06
0.08
Φ X2
(c)
(f)
φ
0 0.5 1 1.5 2 2.5 3
0.15
0.16
0 0.1 0.2 0.3
0.04
0.06
0.08
X1
FIG. 2 (color online). Comparison of numerical simulations
and analytical results for N ¼ 200 HollingTanner oscillators
using the same parameters as Fig. 1. (a)–(c) Steadystate proba
bility density ÈðçÞ of pairwise phase differences for various
levels of environmental and demographic noise: (a) a ¼ 10
À3
,
e ¼ 10
À6
; (b) a ¼ 10
À2
and e ¼ 10
À3
; (c) a ¼ 10
À6
and
e ¼ 10
À3
. (d)–(f) Corresponding ensemble dynamics in state
space showing a snapshot of all the oscillators (black dots) and a
scatterplot of the averaged trajectory
"
X
]
ðiÞ of the metapopulation
over multiple realizations (blue or gray dots).
4 4.5 5 5.5 6 6.5 7 7.5 8
0
0.2
0.4
0.6
0.8
1
R
0.5
0.6
0.7
0.8
0.9
log10(N)
1
−
∆
/
(
2
π
)
_
FIG. 3. Plots of timeaveraged order parameter
"
R (black curve)
and halfwidth 1 À Á,ð2rÞ (gray curve) as a function of system
size N. Variance of RðiÞ with respect to time is shown by error
bars. Parameters as in Fig. 2 with a ¼ 0.01.
PRL 107, 118102 (2011)
P HYS I CAL RE VI E W L E T T E RS
week ending
9 SEPTEMBER 2011
1181024
[17] G. B. Ermentrout and N. Kopell, J. Math. Biol. 29, 195
(1991).
[18] J. N. Teramae, H. Nakao, and G. B. Ermentrout, Phys. Rev.
Lett. 102, 194102 (2009); D. S. Goldobin et al., Phys. Rev.
Lett. 105, 154101 (2010); K. Yoshimura, in Reviews in
Nonlinear Dynamics and Complexity, edited by H. G.
Schuster (Wiley, New York, 2010), Vol. 3, Chap. 3.
[19] W. W. Murdoch, R. M. Nisbet, E. McCauley, A. M.
deRoos, and W. S. C. Gurney, Ecology 79, 1339 (1998).
[20] K. Yoshimura et al., Prog. Theor. Phys. 120, 621 (2008).
PRL 107, 118102 (2011)
P HYS I CAL RE VI E W L E T T E RS
week ending
9 SEPTEMBER 2011
1181025
In particular. ! 0) the Langevin equation reduces to the deterministic planar dynamical system _ _ x1 ¼ A1 ðxÞ. 1ÞT . 2.j dt. where H ¼ ðH1 . pk ! pk =ðN À 1Þ. which is treated in the sense of Stratonovich.PRL 107. [16]. then we have a system of uncoupled population oscillators driven by a common environmental noise source. We associate an independent ðÞ set of Wiener processes Wj with each population oscillator (demographic noise) but take the environmental noise ðÞ to be common to all the oscillators: hdWiðÞ ðtÞdWj ðtÞi ¼ ðÞ i. suppose that we add a stochastic term to the birth rates bk according to bk ! bk þ ðtÞ=2. 1ÞT .m ¼ nm=ðNðN À 1ÞÞ. . bk ! bk =ðN À 1Þ. such that the dynamics of a single oscillator in the absence of noise reduces to the _ simple phase equation ¼ !0 . Treating x1 . (1) to second order in 1=N yields the FokkerPlanck (FP) equation X @ðAi ðxÞPÞ 2 X @2 ðGij ðxÞPÞ @P ¼À þ . dt r. .n ðtÞ ¼ Pðx1 . (3) 2 and Gjj ðxÞ is obtained from Aj ðxÞ by reversing the minus sign on the last three terms.2 @xi @xj @t i¼1. we perform a change of variables XðÞ ! ðÂðÞ . i¼1 qﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ where BðjÞ ¼ Gjj ðxÞ À 2p2 x1 x2 vj . ¼ 1. with ¼ ÅðxÞ. Following previous studies of noiseinduced phase synchronization [8–10]. (3).2 pﬃﬃﬃﬃ where x ¼ ðx1 . First. hdWðtÞdWðtÞi ¼ dt. The master equation for the probability Pm. After converting the demographic noise terms in Eq. A2 ðxÞ ¼ 2b2 x2 ð1 À x1 À x2 Þ À 2c2 x2 À d2 x2 þ 2p2 x1 x2 .n ¼ ½Tr. predation C1 C2 ! C2 C2 is represented by the stoichiometric vector ðÀ1. and set Pm. ¼ 1= N ( 1. migration) between the isolated patches. In the deterministic limit (. the Langevin equation for the ensemble of patches is dXðÞ ¼ AðÞ dt þ 3 X i¼1 Bði. hdWi ðtÞdWðtÞi ¼ 0. dt.n0 À Tr.s (1) with m0 ¼ m À r. Introducing the label . x2 . For concreteness. (3) and expanding to OðÞ leads to an additional multiplicative term in the Langevin equation of the form (2) HðXÞdW. which (up to a sign) represent all possible increments in the number of predator and prey induced by single interactions. 0ÞT . X2 Þ that evolves according to an Ito type P Langevin equation dX ¼ AðXÞdt þ 3 BðiÞ ðXÞdWi . v3 ¼ ðÀ1. v2 ¼ ð0. x2 Þ. (4) to Stratonovich form [which adds an Oð2 Þ correction to the drift terms [10] ]. ck ! ck = ðN À 1Þ. Here Wi denotes an independent Wiener process such that hdWi ðtÞi ¼ 0 and hdWi ðtÞdWj ðtÞi ¼ i. and dWðtÞ ¼ ðtÞdt is an additional independent Wiener process. and Bð3Þ ¼ pﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ 2p2 x1 x2 v3 . We have rescaled the various rates according to dk ! ð1 À Þdk =N. rðÞ Þ and then project out the phase dynamics along the lines of Ref. x2 ¼ A2 ðxÞ with Aj given by Eq. N . An exact solution of the master equation (1) can rarely be found. It is also possible to include a common extrinsic noise source due to environmental ﬂuctuations. A1 ðxÞ ¼ 2b1 x1 ð1 À x1 À x2 Þ À 2c1 x2 À d1 x1 1 À 2ðp1 þ p2 Þx1 x2 . x2 as continuous variables and Taylor expanding each term on the righthand side of Eq. . we now carry out a stochastic phase reduction of the Langevin equation (4). j ¼ 1. (4) where AðÞ ¼ AðXðÞ Þ. Introduce new variables x1 ¼ m=N. This ultimately leads to an Ito Langevin equation for the stochastic phase variables ÂðÞ of the form dÂðÞ ¼ !ðÞ dt þ 3 X i¼1 . spatially separated predatorprey patches. @xi 2 i. where ðtÞ is a white noise term and is the strength of environmental noise.j¼1.s ðm. whereas G12 ðxÞ ¼ G21 ðxÞ ¼ À2p2 x1 x2 . 1ÞT . This allows us to deﬁne a stochastic phase variable for each oscillator according to ÂðÞ ¼ ÅðXðÞ Þ with XðÞ evolving according to the Langevin equation (4). where T is the period of oscillations. Suppose that the deterministic system supports a limit cycle solution x ¼ xÃ ðtÞ with xÃ ðt þ kTÞ ¼ xÃ ðtÞ for all integers k. Now consider an ensemble of N identical.Þ dWiðÞ þ HðÞ dW. x2 ¼ n=N. Next we introduce the three stoichiometric vectors v1 ¼ ð1. n0 ÞPm0 . For example. 118102 (2011) PHYSICAL REVIEW LETTERS week ending 9 SEPTEMBER 2011 where n. The Langevin equation approximates the effects of demographic noise arising from the ﬁnite size N of the local predatorprey population. nÞPm. tÞ.s ðm0 . It can then be shown that the solution to the FP equation (2) determines the probability density function for a corresponding stochastic process X ¼ ðX1 . assuming that the limit cycle is sufﬁciently attracting so that for small and . etc.j .n . H2 Þ with H1 ðXÞ ¼ r1 x1 ð1 À x1 À x2 Þ. If we ignore spatial interactions (dispersal. . we introduce the phase variable 2 ðÀ. H2 ðXÞ ¼ r2 x2 ð1 À x1 À x2 Þ. We then extend the notion of phase into some neighborhood M & R2 of the limit cycle using an isochronal mapping Å: M ! ðÀ. so it is necessary to consider some form of approximation. . where !0 ¼ 2=T. each of which acts as a limit cycle oscillator in the deterministic limit. n0 ¼ n À s. the dynamics can be restricted to the neighborhood M with high probability [15]. for large N we can carry out a KramersMoyal expansion of the master equation (1) [12].n ðtÞ that the system consists of m prey and n predators at time t then takes the form X dPm. Substituting for bk in Eq.
ðÞ dWiðÞ þ ðÞ dW. i (5) ðÞ ¼ with .
ðÞ ¼ .
Z2 Þ 1181022 . Here Z ¼ ðZ1 .i ðÂðÞ Þ ZðÂðÞ Þ Á BðiÞ ðÂðÞ Þ. i ðÞ ðÞ ðÞ ðÞ ðÂ Þ ZðÂ Þ Á HðÂ Þ and ! ¼ !ðÂðÞ Þ !0 þ 2 1 ðÂðÞ Þ þ 2 2 ðÂðÞ Þ.
is the kth component of the inﬁnitesimal phase resetting curve (PRC). where À N @ c ðÞ Q þ 1 N ¼1 2 ¼1 ¼1 @ c ðÞ c ðÞ ½C R 2 1 " " ¼ 2 À ½ 1 ðÞ þ 2 2 ðd.PRL 107. for example. CðÞ ¼ 2 gð c ðÞ À R 1 ðÞ 2 c Þ þ hð0Þ. 2. Q is a slowly varying function of time. 118102 (2011) PHYSICAL REVIEW LETTERS week ending 9 SEPTEMBER 2011 where Zk . we now introduce slow phase variables c ðÞ ¼ ÂðÞ À !0 t and set Qðf c ðÞ g. [18]. k ¼ 1. [9]. normalizedP 2periodic solution _ of the adjoint equation [17] Zk ¼ À j¼1. For small and .2 Ajk ðxÃ ðtÞÞZj ðtÞ. where Ajk ¼ @Aj =@xk . [16]. BðiÞ ðÞ ¼ BðiÞ ðxÃ ðÞÞ. All terms in Eq. the latter will not be important in the following analysis. with gð c Þ ¼ 2 À ð0 Þð0 þ c Þd0 R P3 1 0 0 0 and hð c Þ ¼ 2 À i¼1 . which is deﬁned as P Ã Zk ðÞ @xk Åjx¼xÃ ðÞ with k¼1.2 Zk ðÞAk ðx ðÞÞ ¼ !. Following Ref. tÞ ¼ Pðf c ðÞ À !0 tg. 2 . The PRC is the unique. so we can average the FP equation for Q over one cycle of length T ¼ 2=!0 to get @t Q ¼ P PN "P " ðÞ Q. and further developed in Refs. considerable care must be taken in carrying out the phase reduction procedure in the presence of Gaussian white noise in order to obtain the correct form of the drift terms 1 . However. As ﬁrst shown in Ref. (5) are evaluated on the limit cycle so that. tÞ where P is the probability density of the phases ÂðÞ .
i ð Þ.
2 we relate our analytical results for the distribution ÈðÞ to direct numerical simulations of the Langevin Eq. In Fig. 2(d)]. A0 ¼ 0:1. p2 ¼ ðs þ d2 Þ=2. b2 ¼ ðs þ d2 Þ=2. 1181023 . (4) for the HollingTanner model. see Fig. In the thermodynamic limit N ! 1 we have ¼ N À1=2 ! 0 so that the independent noise source vanishes. It is straightforward to ﬁnd a set of parameters for which Eq. For a moderate ratio of extrinsic to intrinsic noise.3 −100 Z2 (a) −200 0 2 4 phase θ 6 1 φ 2 3 500 FIG. with the Stratonovich environmental noise treated as a solution of a zeromean OrnsteinUhlenbeck process with a small but ﬁnite correlation time. c ð2Þ Þ ¼ Éð c ÞÈðÞ. and looking for separable steadystate solutions Qð c ð1Þ . (6) for an ensemble of HollingTanner oscillators using the given PRC. we introduce ensemble averaged concentrations of prey and P ðÞ " predator according to X j ðtÞ ¼ N À1 N Xj ðtÞ. It can also be seen that X j ðtÞ traces out a trajectory in the phaseplane that remains close to the deterministic limit cycle. In order to characterize the level of synchrony. For a high ratio of extrinsic to intrinsic noise. Z2 of PRC for a limit cycle solution of the HollingTanner equations with r ¼ 1. and K ¼ 0:5. for ﬁnite N. In order to quantify the effects of demographic noise on the metapopulation. However. the phase difference between any pair of oscillators accumulates at zero. where d2 is a free parameter. ¼ c ð1Þ À c ð2Þ . The distribution È is now much broader [Fig. (7) with the drift functions of Eq.08 0. (7) supports a limit cycle [14] and to numerically evaluate the corresponding PRCs. X j ðtÞ still exhibits some oscillatory behavior but the amplitude of oscillations is reduced 400 300 200 Z(θ) 2 g(φ) X2 0. As we now show.04 X1 0 0. p1 ¼ m=½2ðA0 þ x1 Þ À ðs þ d2 Þ= 2 À r=2K.i ð þ c Þd . resulting in the desynchronization of the metacommunity. Thus. d1 ¼ rð1=K À 1Þ. they are no longer " tightly clustered. c2 ¼ sh=2x1 À ðs þ d2 Þ=2. 2(a) and the corresponding clustering of the individual oscillators on the deterministic limit " cycle [Fig. the metapopulation is only partially synchronized. consider the HollingTanner model [13. For concreteness.06 1500 1000 500 0 h(φ) 100 0 Z1 1 0 1 (b) 0 0. j ¼ 1. 1. see Fig. x2 ¼ sx2 1 À : K A0 þ x1 x1 (7) In this model. as indicated by the sharp peak of the steadystate distribution ÈðÞ at ¼ 0 in Fig. (a) Plot of components Z1 . Inset shows limit cycle. the prey exhibit logistic growth up to a carrying capacity K in the absence of predation. (3). it can be shown that [9] ÈðÞ ¼ À0 ½2 ðgð0Þ À gðÞÞ þ 2 hð0ÞÀ1 (6) with À0 a normalization constant. and Éð c Þ ¼ 1=2. The predator consumes prey according to a MichaelisMenten or Holling typeII functional response and grows logistically up to a carrying capacity x=h proportional to the current level of prey in the system. The distribution ÈðÞ then diverges at ¼ 0 while keeping positive since it can be shown that gð0Þ ! gðÞ [9].1 0. 1(a). c1 ¼ 0. although the oscillators tend to remain close to the deterministic limit cycle. the width of ÈðÞ is a good measure of the desynchronizing effects of demographic noise. s ¼ 0:1. As an illustrative example of an oscillatory predatorprey system. resulting in complete noiseinduced synchronization. h in Eq. We see that ÈðÞ is independent of the drift terms j ðÞ. 1(b). demographic noise broadens the distribution of phase differences. m ¼ 3. a randomly ﬂuctuating global environmental signal can synchronize a metacommunity of deterministic predatorprey systems in the absence of dispersal. we set b1 ¼ r=2K. there is signiﬁcant synchronization of the metapopulation. Hence. The simulations are carried out using an EulerMaruyama scheme.14] x1 mx1 x2 hx2 _ _ x 1 ¼ rx1 1 À À .2 0. h ¼ 2. Comparing the righthand side of Eqs. Thus. Performing the change of variables c ¼ ð c ð1Þ þ c ð2Þ Þ=2. We can now determine the functions g. consistent with the Moran effect [4]. we set d2 ¼ s. ¼1 2. we focus on a pair of population oscillators (N ¼ 2). (b) Corresponding functions gðÞ and hðÞ. 2(b)] and.
S.2 0. and B. [16] K. Comput. 7 (1998). Rev. Math. 3 we ÀÁ=2 show how the halfwidth Á provides an excellent predictor of the level of synchrony by comparing its dependence on system size N with the time average of RðtÞ. Lett. 425 (2008).8 0. Rohani.08 (d) X2 0. Finally. (a)–(c) Steadystate probability density ÈðÞ of pairwise phase differences for various levels of environmental and demographic noise: (a) ¼ 10À3 . 93.5 log10(N) 7 7. 1 0. E 71. Lett. J. In Fig. S. 2(c)]. Aust. Variance of RðtÞ with respect to time is shown by error bars.2 0. M. J. Rev. C. the metapopulation is completely desynchronized. W. However. Pikovsky. Quantum Electron. including dispersal between local patches. (c) ¼ 10À6 and ¼ 10À3 . 855 (1975). K. N. 2(f)].5 1 1. [4] P. Moran.4 0. Stochastic Methods: A Handbook for the Natural and Social Sciences (Springer. The density È is almost ﬂat [Fig. A. 14. Soc. Rev. [13] J. Pecora and T. 94. 154101 (2008). Nature (London) 396. W.9 1−∆/(2π) 0.5 1 1. 3. Math.17 0. Parameters as in Fig. Neurosci. Tanaka. Earn. contrary to predictions from deterministic models [19]. Torregrosa. P. A. Phys. J.5 3 (c) 0. 2009). McKane and T. Trends Ecol. Hudson and I. 184101 (2007). 1.04 0 0. namely. and the decay of synchrony with spatial separation of the patches [1. Carroll. Tanner. 2 (2011).6 R 0. [11] A. Kooij. Phys.6 0.2 0 4 4. Bressloff and Y.2]. Lett. D.3 0. A. [7] D. C.7 0. New York. D.3 (f) 0.PRL 107. [6] V. L. 41. J.5 1 1. and B. 204103 (2004). M. [15] In the case of Gaussian noise there is a small but nonzero probability for large deviations under which the phase reduction method could break down. 118102 (2011) 12 8 Φ 4 0 0. Evol. Jansen and A. Pikovsky. Phys. L. 2(f)] could provide an explanation for why oscillations are often not observed in real ecological systems. Comparison of numerical simulations and analytical results for N ¼ 200 HollingTanner oscillators using the same parameters as Fig. This work was partially supported by King Abdullah University of Science and Technology Grant No. 2109 (1998). J. D. Grenfell. M. Phys. Lett. 232 (2000). Lett. Lai. J. B 265. Grenfell.1 0. Publ. J. R. Phys. [14] A.5 8 " FIG. 1. E. Ly and G. 101. 41. 045201 (2005). A. [3] D. Zool. 1181024 . Rohani.18 Φ 0. Science 286.06 0. T. One major implication of our work is that the stabilizing effects of demographic noise [Fig. and individual oscillators are much more scattered in the phaseplane so " that X j ðtÞ exhibits small ﬂuctuations about the mean of the limit cycle [Fig. 2 (color online).06 0. In conclusion. P. we have applied stochastic phase reduction and averaging methods to analzye an important ecological problem. Neurosci.04 0 0. Nature (London) 460.1 0. for a low ratio of extrinsic to intrinsic noise.5 0. T. Cattadori. Proc. 41 (1998). Math. 041902 (2004). 1 (1999). 2(e)]. Rev. Arai.5 6 6. P. Kawamura. spatial heterogeneity [20]. 1007 (2009).5 2 2. 2 with ¼ 0:01. [1] P. Phys. ¼ 10À6 . (d)–(f) Corresponding ensemble dynamics in state space showing a snapshot of all the oscillators (black dots) and a " scatterplot of the averaged trajectory X j ðtÞ of the metapopulation over multiple realizations (blue or gray dots). Arai. J.15 0 0. Plots of timeaveraged order parameter R (black curve) and halfwidth 1 À Á=ð2Þ (gray curve) as a function of system size N. 291 (1953).5 2 2. [10] P. 27. J. Earn. we introduce the P ðÞ order parameter RðtÞ ¼ N À1 j N eiÂ ðtÞ j and the ¼1 R halfwidth Á with Á=2 ÈðÞd ¼ 0:5. 968 (1999). R. 576 (1984). Newman.08 0.3 (e) week ending 9 SEPTEMBER 2011 FIG. Nakao. Gasull.5 3 φ X2 X2 0 0. Radiophys. this does not affect the theoretical arguments presented here. We have shown that the degree of synchronization of the metapopulation can be characterized in terms of the probability density for pairwise phase differences. Rev.8 _ 0. Lloyd. and Y. Gardiner. Hanski. [2] I. 4th ed. Fox. 98. B. Biol.06 0. 80. E 70. Yoshimura and K.5 5 5.04 0 0. [9] H.2 X1 0. the effects of demographic noise on the synchronization of a metapopulation in a ﬂuctuating environment (Moran effect).16 0. [5] L. Goldobin and A. [Fig. In order to further quantify how demographic noise affects synchrony. and J. 1. Ecology 56.15 0 0.5 2 2. 218102 (2005). Vasseur and J. J. Teramae and D. Phys. [12] C. 26.08 0. Rev. KUKC101304. There are numerous ecologically motivated extensions of this work. D.1 0. T. Ermentrout.16 Φ 0. (b) ¼ 10À2 and ¼ 10À3 .5 3 (b) (a) PHYSICAL REVIEW LETTERS 0. Rev. [8] A. 149 (1997).
3. K. E. in Reviews in Nonlinear Dynamics and Complexity. Kopell. Ermentrout and N. McCauley. Rev. B. Yoshimura et al. and G. Nakao. B. 194102 (2009). Biol. Rev. Murdoch. 120. Yoshimura.. Schuster (Wiley. M.. 3. Phys. D. Goldobin et al. N. 1339 (1998). Theor. 2010). A. 102. Vol. [18] J. 1181025 . edited by H.PRL 107. Phys. 118102 (2011) PHYSICAL REVIEW LETTERS week ending 9 SEPTEMBER 2011 [17] G. Lett. Ermentrout. 29. W. Gurney. Phys. 621 (2008). Math. M. and W. Prog. New York. J. S. H. [20] K. 105. 154101 (2010). Ecology 79. [19] W. Lett. deRoos. C. G. R. Chap. Nisbet. S. 195 (1991). Teramae.