1069

Progress of Theoretical Physics, Vol. 79, No.5, May 1988

Mutual Entrainment in Oscillator Lattices

with N onvariational Type Interaction

Hidetsugu SAKAGUCHI, Shigeru SHINOMOTO and Y oshiki KURAMOTO

Department of Physics, Kyoto University, Kyoto 606

(Received December 16, 1987)

A model system for limit cycle oscillators distributed on a d ·dimensional cubic lattice is studied.
This model has the form qii=Wi-K~jENI{sin(<pi-<pj+a)-sina}, .i=l, 2"'Ld, where <Pi is the phase
of the oscillator on the i-th site, the natural frequencies may change from site to site and Ni represents
the set of the nearest neighbor sites of the i-th site. In a previous work we studied the case of

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

vanishing a and discussed the possibility of phase transition. In the present work we deal with the
case of nonvanishing a, and show by computer simulation that the parameter a has a strong effect
on global entrainment. Two types of effects of a are discussed. One is that it suppresses large scale
phase fluctuations, thus facilitating global entrainment. The other is associated with its effect on
mutual entrainment in the presence of vortices.

§ 1. Introduction
Systems of coupled limit cycle oscillators have been the subject matter of exten-
sive experimental and theoretical studies. Many of them have been motivated by the
fact that self-entrainment of oscillators are functionally much relevant to living
organism, such as the heart and intestine. A most remarkable feature of many-
oscillator systems is that they exhibit phase transition-like phenomena when the
natural frequencies are distributed and this problem was studied by using various
model equation.I)-IS) Inspired by Winfree's earlier idea of phase description,I)
Kuramoto proposed a well-defined model and obtained analytical results in the case of
mean field coupling.3 ) The model employed there is given by
. K N •
rpi={J)i- N#Ism(rpi-rpJ, (1)

where rpi represents the phase of the i-th oscillator, and N the total number of the
oscillators. The natural frequencies (J)i generally change randomly; the normalized
number density of the oscillators having natural frequency (J) is denoted as g({J)~. The
above system shows a phase transition such that collective oscillations appear only
above a critical coupling strength. For this special model the mean field approxima-
tion holds exactly and analytical expressions for various quantities can be obtained.
The mean field coupling is not very realistic, however, and we considered the case of
finite-range coupling in· a previous paper. IS) . There the oscillators were regularly
arranged on ad-dimensional cubic lattice, and they were assumed to interact only
with their nearest neighbors. The model system is then given by a generalization of
Eq. (1) in the form
(2)

where Ni is a set of the nearest neighbor sites of the i-th site. As the coupling

_---.J
1070 H. Sakaguchi, S. Shinomoto and Y. Kuramoto

strength is increased, the oscillators first exhibit local entrainment, and the entrained
clusters thus formed become larger and larger. It was revealed in the previous
work 18) that there exists a critical coupling strength above which a macroscopic
number of oscillators are mutually entrained into a unique frequency provided the
system dimension equals three or higher.
The interaction assumed in systems described by Eq. (1) or Eq. (2) is very special,
while the theory of the phase description states nothing more than that the general
form of coupling should be a 2Jr-periodic function of the phase difference cPi- cPj
between the interacting pair.4 ),13) Moreover, the model in Eq. (2) cannot consistently
explain the following fact. That is, in real living systems such as the mammalian .i
intestine and heart, the cells forming a tissue oscillate with frequencies larger than
when they are isolated from each other. 2),lO) These facts naturally lead us to the
study of the second simplest model:

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

(3)

The interaction here is of attractive type and tendsto increase the frequencies when
a is positive. Note that the interactioD- cannot be derived from any kinetic
potential,S) and we call this type of interaction nonvariational. The nonvariational
type interaction seems to be characteristic to general nonequilibrium systems. We
discuss in this paper some remarkable effects of a on local and global entrainments.

§ 2. Mutual entrainment between two oscillators

We begin with an elementary consideration of a system of two oscillators:

¢1=alI-K{sin(cPl-cP2+a)-sina} ,
(4)

The mean phase ¢=(cPl+cP2)/2; and the phase difference cP12=cPl-cP2 obey the equa-
tions
rj =w+ K(1-coscP12)sina,
(5)

where W=(WI + (fh)/2 and W12= Wl- W2. If IW12/2Kcosal > 1, cP12 changes monotonical-
ly with t implying that the oscillators undergo quasiperiodic oscillations with
independent frequencies (iiI and &2, where &i are defined and given by
- _}" cPi(T)-cPi(O) (6)
Wi-l£! T '
- - + -W12
WI-W
2
/1 - (2Kcosa)2 +K·SIna,
W12
.

- W12
W2=W---
2
/1 - (2Kcosa)2 +Ksma.
W12
. (7)
 Mutual Entrainment in Oscillator Lattices 1071

o o
N N

o o
CD CD

...
D·

o o
O+-~r--.__. - - .__- '__' - - '__' :~-.,--,--.--,---,--,--.--,
o
0.00 0.40 0.80 l. 20 l. 60 0.00 0.40. 0.80 l.20 1.60
K K
(a) (b)

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

Fig, 1. Effective frequencies iii versus coupling strength K in a two·oscillator system.
(a) a=O, (b) a=7r/4.

If IUh2/2Kcosal<l, <P12 finds its stable equilibrium at

,/.. -sin- 1( Uh2 ) Kcosacos<P12>0. (8)

'1'12- 2Kcosa'

The two frequencies are then identical and given by

- - IUh21-tana/( 2Kcosa)2
Uh=C02=w-- C012 - 1 +Ksma.
. (9)
2
Thus we have a critical coupling strength for mutual entrainment. The effective
frequencies ah and iih change with the coupling strength as shown in Fig. 1. When
0< a< JC/2 and the mutual entrainment has been established, the effective frequencies
become larger than the simple mean of the natural frequencies. This is because the
phase difference <P12 increases the instantaneous frequencies through the term
(l-cos<P12)sina in Eq. (5). 'The increase in frequency through the phase difference
causes many interesting phenomena as we see in later sections.

§ 3. One-dimensional oscillator lattices

In equilibrium systems, the appearance of long range order becomes increasingly

difficult as the spatial dimension is lowered. This is due to long wavelength fluctua-
tions of the order parameter. Also in our oscillator lattices with vanishing a, there
are no phase transitions. In fact, as we showed in the previous paper, the linear
analysis .of Eq. (2) gives expression for the mean square fluctuation of wavenumber k
as

(10)

where L]co is the root mean square of the frequency distribution g(co). The "infra-red"
divergence of phase fluctuations destroys the macroscopic order in lower dimensional
systems. In contrast, nonvanishing a seems to have the effect of suppressing the long

--~
 1072 H. Sakaguchi, S. Shinomoto and Y. Kuramoto

g (I-a) COUPLI NG
I. 00
>- ALPHA
u 0.0
z

~~1 ,~------------~------------~-------------r------------~-------------.
100 200 300 400 500

g (I-b) COUPLING
.; 2.00
>- ALPHA
u 0.0
z
Wo
::>0
o o·
W
a:
lL.
o
o
~~------------~-------------r-------------r------------~------------~

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

100 200 300 400 500

g (I-c) COUPLING
3.00
>- ALPHA
u 0.0
z

1:1-,----,---- ~=~:_=~~~=
100 200 300 400 500

COUPLING
I. 00
ALPHA
45.0

100 200 300 400 500

o (2-b) COUPLING

~ "~
2.00
ALPHA
45.0

~~~
w
a:

lL.~ I
,~------------~-------------r------------~------------~--------------
100 200 300 400 500

~ ~l (2·c)
COUPLING
3.00

~~----~--~-----
ALPHA
45.0

::>0

~;.~j--------------r-------------~------------~------------~------------~
100 200 300 400 500

Fig. 2. Frequency patterns in a one-dimensional oscillator lattice.

(I-a) a=O K=l.O, (I-b) a=O K=2.O, (I-c) a=O K=3.0, (2-a) a=7r/4 K=l.O, (2-b) a=7r/4 K=2.O,
(2·c) a=7r/4 K=3.0.
 Mutual Entrainment in Oscillator Lattices 1073

wavelength fluctuations through the nonlinear term, sina{l-cos(¢i- ¢j)} in Eq. (3).
In order to see the situation in further detail, a computer simulation was carried out
in a one-dimensional lattice of 10000 oscillators. The distribution gem) is assumed to
be Gaussian with variance 1. As the method of numerical integration we used the
Euler method where the elementary time increment was taken to be 0.1. This may
look to be too crude an approximation to the ordinary differential equation. Actu-
ally, however, the errors produced then are unimportant for our purposes as we stated
in the previous paper for similar situations. 13 ),18) The uniform initial distribution and
the 'periodic boundary condition are chosen. The effective frequency of the i-th
oscillator is calculated from the approximate formula

(11)

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

where T is the total time of the numerical integration subtracted by the equilibration
time, and equals 600 in the present simulation.
How the spatial distribution of Wi in the same part of an oscillator chain changes
with the coupling strength and a is shown in Fig. 2. The flat regions in the frequency
patterns imply that the oscillators are locally entrained. The mean size of these
entrained clusters becomes larger as the coupling strength is increased. For a=O the
size distribution has a peak about the mean size, while for a= 7r/4 and relatively large
K· the frequency pattern comes to be composed of a small number of very large
clusters and many small ones. In the calculation we adopted the following criterion
for the mutual entrainment between the i-th and the neighboring j-th oscillators:

Figure 3 shows the decrease of the total number of the entrained clusters as the
coupling strength is increased. For a=O it decreases roughly as I/K 2 , while for
o o
o
'"o o

o
~
~o "
z' z'
o o
c..o
Do
--'':
,

:;'
,
o
o
~+--'--.-'--.--.--'--'--r-'r-,
'-1.00 -0.60 -0.20 0.20 0.60 1. 00 '-1.00 -0.60 -0.20 0.20 0.60 1.00
LDG( K) LDG( K)

(a) (b)
Fig. 3. Log-log plot of number density of entrained clusters versus coupling strength K in a one-
dimensional oscillator lattice. (a) a=O, (b) a=7r/4.
1074 H. Sakaguchi, S. Shinomoto and Y. Kuramoto

a= 7r/4 it exhibits a more rapid decrease. These results suggest that the local entrain- ,
inent is facilitated by nonvanishing a.
In our simulation the distribution g(w) is Gaussian as we stated. Then, due to the
presence of oscillators with arbitrary high frequencies, complete entrainment is no
longer possible as far as the system size is infinite. To learn the effect of a we apply
a linear analysis to the case that g(w) has high- and low-frequency cutoffs. Let the
system size be L and the coupling strength be strong enough so that ·the condition
iK/(w;- wJi~ 1 may be satisfied for any nearest neighbor pair (i, j). Because the
phase differences between the nearest neighbor oscillators are expected to be so small,
a power series expansion of Eq. (3) in the phase difference may be permitted. Up to
the second order in the phase difference, we obtain

(12)

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

When a is zero, the third term on the r.h.s. vanishes and the mean square of the phase·
difference can be easily calculated by using the Fourier transform. We have

(13)

Equation (13) implies that the phase difference increases indefinitely with the system
size L, so that the complete entrainment is collapsed when L becomes as large as
O((K/L1w)2). In a system of infinite size, typical phase difference between the nearest
neighbor oscillators in a cluster of size L may also be expected to be O((L1w/K)IL),
because no serious influences from the neighboring clusters are expected. By
definition the phase difference between any nearest neighbor pair in an entrained
cluster should be smaller than 7r, so that the above argument implies that a typical
cluster size should be proportional to (K/L1w)2. Figure 3 implies thatthe mean cluster
size is proportional to K2 for a=O and seems to support the argument above .
. When a is nonzero, the third term of Eq. (12) is important. If we go over to the
continuum approximation, Eq. (12) becomes
¢(x)=w(x)+ K cosa [72¢+ K sina([7 ¢)2,
w(x) = w; for i-::::::x-::::::i+ 1, (14)
where the lattice constant is taken to be 1. By the Cole-Hopf transform 19) ¢(x)
=cota InQ(x), Eq. (l4) becomes

dQ tana ( »).
1
- cosa . dt = ( - cosa 2
w x Q - K[7 Q . (15)

This is a Schrodinger equation with a random potential. For this type of systems the
Anderson localization is known to occur for one-dimensional systems. The n-th
eigenfunction Qn(X) with eigenvalue (-An/Cosa) will be localized about Xn with some
localization length ~n:

(16)
 Mutual Entrainment in Oscillator Lattices 1075

a TIME 1000.0
a
o,-----------------------------~
ro
"
W O
({)":
CIa
I"
,,-"
a
a
a
o~------------------ __________ ~

" 0 100 200

Fig. 4. Phase pattern in a one-dimensional oscillator lattice with K = 1.0 and a= 7(/4 at time 1000.
This value of K is large enough for complete entrainment to be established.

Then the solution of Eq. (13) is written as

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

(17)

where an(x) is the amplitude of the n-th eigenfunction Qn(X) including its. deviation
from the exponential behavior as in Eq. (16). In the summation in Eq_ (17), the term
corresponding to the largest An, which we denote as Am, becomes increasingly
dominant with t. Thus the approximate solution of Eq. (14) is given by
¢(X, t)=cota lnQ(x, t)

~cota{ln am(x) Ix~:ml +Amt}. (18)

Note that Am should be smaller than tana'maXWi so that it remains finite provided
g(w) has cutoffs. The solution in Eq. (18) shows that the phase pattern is centered at
Xm and ¢ decreases linearly with the distance from Xm. A simulation result is shown
in Fig. 4, where g(w) is assumed to be constant over ~he interval [0,1] and vanishing
outside. The length scale t;m is related to the coupling strength K, and is roughly
scaled as t;m ~ IK f m. The phase difference is roughly cota/t;m and is proportional
to 1/ IK so that 'complete entrainment is possible even if L goes to infinity. Thus,
nonvanishing a facilitate large scale entrainment )¥hen g( w) has frequency cutoffs.
Without cutoffs, complete entrainment is impossible, still nonvanishing a facilitates
the formation of large entrained clusters as was confirmed by our computer simula-
tion.

§ 4. Two-dimensional oscillator lattices

Two-dimensional systems admit topological defects called vortices. In equili-

brium X- Y spin systems, vortices play an important role in the phase transition as we
know by the term Kosterlitz-Thouless transition. One may ask how about the role
played by vortices in oscillator lattices.
In what follows (Figs. 5~8) we show some results of computer simulations on our
two-dimensional system with two types of patterns. The first one is called the phase
pattern which shows a snap shot of ¢i's, and the second is the frequency pattern which
1076 H. Sakaguchi, S. Shinomoto and Y. Kuramoto

(a) (b) (c)

Fig. 5. Phase patterns in a two·dimensional system without frequency distribution. The initial
condition is so chosen that a vortex is formed at the center.
(a) a= 7(/6, (b) a= 7(/4, (c) a= 7(/3.

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

clearly shows how mutually entrained oscillators are distributed. The frequency
pattern tells us whether the oscillators are mutually entrained and the phase pattern
tells us how the oscillators are mutually entrained. Wherever oscillators are strong-
ly syncronized, their phase changes smoothly in space, while if the oscillators are
desynchronized, the phase pattern becomes nonsmooth in general.
Even without distribution in natural frequency, vortices may cause nonuniformity
in local frequency and as a result the feature of mutual entrainment may drastically
be changed. In the vicinity of a vortex the phase difference between nearest neighbor
oscillators is expected to be as large as 7r/2. Because phase difference induces the
increase in local frequency if a is positive, the vortex acts as a pacemaker entraining
the surrounding medium. This is why a spiral pattern arises round the vortex. A
spiral pattern with a phase singularity at the center has a phase gradient which is
nearly constant along the radial direction. Figure 5 shows phase patterns for a= 7r/6,
7r/4,7r/3 and (Oi=O for all i. We demonstrate the phase patterns by assigning the
mark + on lattice point i if the condition sincPi(T)>0 is satisfied. For a=7r/6, the
spiral pattern rotates steadily. As a is increased, the spiral winds up more and more
tightly, so that the phase gradient in the radial direction is steepened. The global
entrainment collapses above a certain critical value ac and then the center of the
spiral oscillates at a frequency different from the surrounding medium. Thus the

(a) (b)
Fig. 6. Phase patterns in a two-dimensional system with a=7(/6 and K=l.O, where the central
oscillators have natural frequencies higher than the surrounding medium by O.S. Vortex is absent
in (a) and present in (b).
 Mutual Entrainment in Oscillator Lattices 1077

vortex breaks the mutual entrainment even in homogeneous systems. The computer
simulation for a= 7[/4 shows that the center of the spiral moves around as a result of
the desynchronlzation. For even larger a many vortex-antivortex pairs are created
and the phase patterns become very chaotic.21 ) When the initial phase distribution is
relatively uniform, a perfectly uniform state is finally established as far as lal < 7[/2,
as the uniform state is linearly stable then. It is important to realize that the final
pattern of entrainment depends on the initial condition in a two-dimensional oscillator
lattice.
When the system is not homogeneous in natural frequency, the local entrainment
extends with the coupling strength. However, the pattern of entrainment is more
complicated than in one-dimensionarIattices because of the vortices involved. We
show some simulation results which help understand the effect of the vortices on
mutual entrainment. Figure 6 shows phase patterns with and without vortex. The

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

natural frequencies assumed are w;=0.8 in a circular domain of radius 12 and w;=O
outside it, and K=1.0 and a=7[/6. When there is no vortex, all oscillators are
entrained to the central region and a target-like phase pattern emerges then. When
a vortex is introduced, it increases the frequencies of the central oscillators, and the
entrainment over the entire system is no longer possible. A spiral pattern is then
formed near the center, although it cannot extend outward. As a result of desyn-
chronization between the inner and outer regions additional disturbances are
produced. This gives a typical case that a vortex combined with the distribut.ion of
natural frequencies breaks the global entrainment.
Figure 7 shows the situation in which the complete entrainment can be established
even in the presence of vortex and frequency distribution for a= 7[/4. In contrast
there is a situation that a vortex breaks global entrainment in the system without
frequency distribution for the same value of a as is shown in Fig. 5. The origin of the
desynchronization is that the central part oscillates so fast that the outer region is
unable to follow it. It is therefore expected that the global entrainment is recovered
if the natural frequencies in the central part are made suitably lower than the
surrounding. In Fig. 7 the central four oscillators have natural frequencies lower by
(3/4)/2 than the surrounding. Then a steadily rotating spiral pattern is clearly seen
implying global entrainment.
We have seen that the mutual
entrainment can sensitively be depend-
ent on the distribution of natural fre-
quencies and also on vortex distribution.
We show itl' Fig. 8 another simulation
result with frequency distribution show-
ing that mutual entrainment can strong-
ly be dependent on initial condition.
The distribution g(w) is taken to be
Fig. 7. Phase patterns in a two·dimensional sys- Gaussian with variance 1 and the other
tem with a= 7[/4 and K = 1.0, where the central parameters are fixed as K=2.6 and a
oscillators have frequencies lower thap the = 7[/4, and periodic boundary condition
surrounding medium by (3/4)/2.
1078 H. Sakaguchi, S. Shinomoto and Y. Kuramoto

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

(a) (b) (c)
Fig. 8. Frequency patterns and phase patterns in a two-dimensional random system_ The distribution
g(w) is taken to be a Gaussian with variance 1 and a=7r/4 and K=2_6_ Patterns in (a), (b) and
(c) correspond to different initial conditions.

is chosen.
Three different initial conditions are chosen, and the resulting frequency patterns
and phase patterns are shown in Fig. 8. In the frequency pattern, a bond is assigned
between the nearest neighbor sites if their frequencies are regarded identical. An
entrained cluster linked mutually by bonds represents itself as a giant oscillator. IS)
Quite different cluster distributions are obtained from different initial conditions for
the same system. A spiral or target pattern, though imperfect, can be seen where the
oscillators form a large entrained cluster. In contrast, where the oscillators are not
mutually entrained, the phase pattern is chaptico . These two characteristic regions
are clearly separated from each other.
We have shown some result:s of our computer simulations in two-dimensional
oscillator lattice. We found that the nonvariational type coupling makes some
topological defects involved act as pacemaker. This effect introduces an additional
origin of complexity into pattern formation associated with mutual entrainment.

§ 5. Summary

We have shown some remarkable system behaviors characteristic to the nonvar-

iational type interaction in oscillator lattices. This type of interaction increases the
effective frequency through the phase gradient produced by the distribution of natural
frequencies or by vortices. The system has a possible ability of adjusting local phase
gradient so that the difference in natural frequency is compensated. Thus the mutual
entrainment on a large scale can be facilitated. Conversely, the effective frequency
 Mutual Entrainment in' Oscillator Lattices 1079

increase by the phase difference can be the cause of desynchronization in the presence
of vortices. This is because the effective frequency would be too much increased in
the vicinity of phase singularity. Thus it may be said the nonvariational type
interaction has ambivalent effects on mutual entrainment. Finally, we have to say
that nothing is clear yet about the effect of a on phase transitions; this is completely
open to future investigation.

References
1) A. T. Winfree, J. Theor. BioI. 16 (1967)," 15.
2) A. T. Winfree, The Geometry of Biological Time (Springer, New York, 1980).
3) Y. Kuramoto, Lecture Notes Phys. Vol. 39 (Springer, New York, 1975).
4) Y. Kuramoto, Chemical Oscillation, Waves and Turbulence (Springer, Heidelberg, New York,
Tokyo, 1984).
5) Y. Kuramoto, Prog. Theor. Phys. Suppl. No. 79 (1984), 223.
6) Y. Aizawa, Prog. Theor. Phys. 56 (1976), 703.

Downloaded from http://ptp.oxfordjournals.org/ by guest on December 21, 2015

7) J. C. Neu, SIAM, J. Appl. Math. 37 (1979), 307.
8) Y. Yamaguchi, K. Kometani and H. Shimizu, J. Stat. Phys. 26 (1981), 719.
9) Y. Yamaguchi and H. Shimizil, Ph.ysica lID (1984), 212.'
10) G. B. Ermentrout and N. Kopel!, SIAM]. Math. Anal. 15 (1984), 21.
11) G. B. Ermentrout, J. Math. BioI. 22 (1985), 1.
12) H. Daido, Prog. Theor. Phys. 75 (1986), 1460.
13) . H. Sakaguchi and Y. Kuramoto, Prog. Theor. Phys. 76 (1986), 576.
14) Y. Kuramoto, Physica 106A (1981), 128.
15) S. Shinomoto and Y. Kuramoto, Prog. Theor. Phys. 75 (1986), 1105.
16) S. Shinomoto and Y. Kuramoto, Prog. Theor. Phys. 75 (1986), 1319.
17) Y. Kuramoto and I. Nishikawa, J. Stat. Phys. 49 (1987), 569.
18) H. Sakaguchi, S. Shinomoto and Y. Kuramoto, Prog. Theor. Phys. 77 (1987), 1005.
19) Y. Kuramoto and T. Yamada, Prog. Theor. Phys. 56 (1976), 724.
20) T. Yamada and Y. Kuramoto, Prog. Theor. Phys. 55 (1976), 2034.
21) Y. Kuramoto and S. Koga, Prog. Theor. Phys. 66 (1981), 1081.