Coherence incoherence patterns in a ring of non-locally coupled phase oscillators

Weierstraß-Institut
für Angewandte Analysis und Stochastik
Leibniz-Institut im Forschungsverbund Berlin e. V.
Preprint ISSN 0946 – 8633
Coherence-incoherence patterns in a ring of non-locally
coupled phase oscillators
Oleh Omel’chenko
submitted: April 15, 2013
Weierstrass Institute
Mohrenstr. 39
10117 Berlin
Germany
E-Mail: Oleh.Omelchenko@wias-berlin.de
No. 1777
Berlin 2013
2010 Mathematics Subject Classiﬁcation. 34C15, 35B36, 35B32, 35B42, 35Q83.
2008 Physics and Astronomy Classiﬁcation Scheme. 05.45.Xt, 89.75.Kd.
Key words and phrases. Coupled phase oscillators, non-local coupling, coherence-incoherence patterns, chimera
states, bifurcation analysis, pattern formation.

Edited by
Weierstraß-Institut für Angewandte Analysis und Stochastik (WIAS)
Leibniz-Institut im Forschungsverbund Berlin e. V.
Mohrenstraße 39
10117 Berlin
Germany
Fax: +49 30 20372-303
E-Mail: preprint@wias-berlin.de
World Wide Web: http://www.wias-berlin.de/

Abstract
We consider a paradigmatic spatially extended model of non-locally coupled phase
oscillators which are uniformly distributed within a one-dimensional interval and interact
depending on the distance between their sites modulo periodic boundary conditions. This
model can display peculiar spatio-temporal patterns consisting of alternating patches with
synchronized (coherent) or irregular (incoherent) oscillator dynamics, hence the name
coherence-incoherence pattern, or chimera state. For such patterns we formulate a gen-
eral bifurcation analysis scheme based on a hierarchy of continuum limit equations. This
gives us possibility to classify known coherence-incoherence patterns and to suggest di-
rections for searching new ones.
1 Introduction
Emergence of collective behavior (synchrony) in large groups of oscillators plays an important
role in a wide variety of significant applications. Examples are synchronization of neuronal os-
cillations in brain [1, 2, 3], entrainment in coupled Belousov-Zhabotinsky chemically reacting
cells [4, 5], Josephson junction circuits [6] and mode-locked lasers [7], other references also
can be found in [8]. It is well-known that in the context of N weakly coupled oscillators [9, 10],
main features of synchronization phenomenon can be adequately reproduced by a correspond-
ing phase reduced model
dθk
dt
= ωk −
1
N
N
j=1
GkjF(θk(t) − θj(t)), k = 1, . . . , N, (1)
where the state of the k-th oscillator is represented solely by its scalar phase θk ∈ R. Further,
natural frequencies ωk ∈ R reflect the individual differences between oscillators, and a coupling
matrix Gkj ∈ RN×N
together with a 2π-periodic function F : R → R determine the details
of interaction between oscillators. To get a qualitative insight into model (1) it is often enough
to replace a particular function F with its leading Fourier harmonics of the form sin(θ + α)
where α ∈ R is referred to as phase lag parameter. In the case of global coupling (when
all Gkj are equal) and natural frequencies ωk drawn from a certain probability distribution,
this simplification gives a paradigmatic Kuramoto-Sakaguchi model [11] that describes synchro-
nization transition in a homogeneous bulk. However, bulk approximation is not always suitable.
For many spatially extended systems, it appears more natural to assume that coupling coeffi-
cients Gkj vary depending on real physical distance between oscillators. To mimic this feature
Kuramoto and Battogtokh suggested in [12] another kind of model (1) that reads
dθk
dt
= ω −
2π
N
N
j=1
Gkj sin(θk(t) − θj(t) + α), k = 1, . . ., N. (2)
1

Here all oscillators have the same natural frequency ω ∈ R and are equipped with additional
spatial labels xk denoting their positions within the one-dimensional reference interval [−π, π],
hence the normalization prefactor 2π at the sum. Assuming that xk are uniformly distributed
over [−π, π] (typical choice is xk = −π + 2πk/N), one defines
Gkj = G(xk − xj), (3)
where G : R → R is a continuous even and 2π-periodic coupling function introducing a sym-
metric ring coupling topology between oscillators. If G is constant and non-zero then Eq. (3)
determines global coupling. On the other hand, any non-constant function G produces a qual-
itatively different coupling topology which according to established tradition is called non-local
coupling. The non-local coupling is defined on a macroscopic level, hence using a particular
function G one obtains a sequence of systems (2)–(3) with different sizes N. Nevertheless for
sufficiently large N all these systems demonstrate macroscopically similar long-term dynam-
ics. In particular, starting from randomly chosen initial conditions and simulating system (2)–(3)
one typically obtains either a completely synchronized solution, Fig. 1 (a), or a twisted phase-
locked solution, Fig. 1 (b). However, it was an exciting discovery that in spite of apparent struc-
tural symmetry, for some parameter values system (2)–(3) can also support unexpected spatio-
temporal patterns shown in Fig. 1 (c) where synchronized (coherent) and irregular (incoher-
ent) oscillator dynamics is observed simultaneously but at different sites xk. Such coherence-
incoherence patterns attracted great interest and have been the subject of numerous studies in
last decade [12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30]. In most
of above publications they are referred to as chimera states. This name sounds attractive but
is very loosely defined. For example, it was used to denote asymmetric partially synchronized
states in two or three interacting groups of globally coupled oscillators [31, 32], or to denote
non-trivial fixed points in networks of non-locally coupled maps [33, 34]. Therefore, in order to
identify more precisely the main object of our study we intentionally use the name coherence-
incoherence pattern instead of chimera state.
To the best of our knowledge, all previous studies of coherence-incoherence patterns were
concerned with a particular choice of coupling function G. Kuramoto and Battogtokh considered
an exponential coupling [12] of the form
G(x) ∼ e−κ|x|
with κ > 0
motivated by adiabatic elimination of fast diffusive variable in a multi-component oscillatory
chemical system. A few years later, the cosine coupling
G(x) =
1
2π
(1 + A cos x) with A ∈ (0, 1) (4)
was suggested [14, 15] in order to simplify the mathematical complexity of the problem. Finally,
the piecewise-constant coupling with a radius r ∈ (0, 1)
G(x) =
(2πr)−1
for |x| ≤ πr,
0 for |x| > πr
(5)
was used in [23, 25, 26] to study the dynamical nature of coherence-incoherence patterns in
the context of finite size systems.
2

(c)
-π
0
π
(b)(a)
t
0
50
100
-π 0 π
xk
-π 0 π
xk
-π 0 π
xk
-π
0
π
θk
Figure 1: Solutions observed in numerical simulations of Eqs. (2)–(3): (a) completely synchro-
nized state, (b) twisted state, and (c) coherence-incoherence pattern. Top and bottom panels
display time dynamics of oscillators and corresponding instantaneous snapshots, respectively.
Translational symmetry of Eqs. (2)–(3) was used to center the resulting plots. Parameters:
N = 200, ω = 0, α = π/2−0.1, and piecewise-constant coupling (5) with r = 0.8. Each so-
lution can be approached with a nonvanishing probability starting from randomly chosen initial
data.
In contrast, this work focuses on the systematic description of coherence-incoherence patterns
and other dynamical regimes observed in system (2)–(3) after an initial transient for arbitrary
choice of coupling function G. For this we employ the large N limit formalism explained in Sec-
tion 2. There, applying the ideas of the Ott-Antonsen invariant manifold theory we show that
on a macroscopic level the effective dynamics of system (2)–(3) can be described in terms
of a complex-valued function — local order parameter, that evolves according to some ex-
plicitly known equation in Banach space of continuous functions. It turns out that coherence-
incoherence patterns as well as completely synchronized and twisted solutions can be repre-
sented as standing wave solutions of the latter equation. Moreover, spatially dependent ampli-
tudes of these waves and their rotational frequencies satisfy a nonlinear integral equation with
O(2) × S1
-symmetry, which has a form of the inﬁnite-dimensional nonlinear eigenvalue prob-
lem (NEVP). In Section 3 we perform a detailed analysis of the solution set to this NEVP. In
particular, we show that every non-zero harmonics in the Fourier series of coupling function G
gives rise to several solution curves (modulo the symmetry group action) organized as primary
an secondary branches bifurcating from the trivial solution. All the branches can be interpreted
either as coherence-incoherence patterns or as phase-locked solutions of system (2)–(3). Re-
markably, only a few of them are stable and can be observed in numerical simulations. The
underlying stability analysis with respect to the evolution equation for local order parameter is
addressed in Section 4. Discussion and concluding remarks are presented in Section 5.
3

Notations. Throughout this paper we assume the following notations. For any integer k ≥ 0,
by Ck
per([−π, π]; R) and Ck
per([−π, π]; C) we denote the Banach spaces of 2π-periodic, real-
and complex-valued functions, respectively, which are continuous together with their derivatives
up to the k-th order. These spaces are equipped with their usual supremum norms
u k := u Ck
per([−π,π];R) = u Ck
per([−π,π];C) = max
x∈[−π,π]
k
j=0
|u(j)
(x)|.
Instead of C0
per([−π, π]; R) and C0
per([−π, π]; C) we use notations Cper([−π, π]; R) and
Cper([−π, π]; C), respectively.
In some cases we identify function u ∈ Ck
per([−π, π]; C) with a vector-function of its real and
imaginary parts
(Re u, Im u)T
∈ Ck
per([−π, π]; R2
)
that simultaneously provides a natural definition of the latter Banach space. Then for a mapping
F : R2
(u1, u2)T
→ (f1(u1, u2), f2(u1, u2))T
∈ R2
where f1 and f2 are given smooth functions, we use an equivalent complex notation
u = u1 + iu2 → f(u) = f1(u1, u2) + if2(u1, u2) with f : C → C.
This notation implies a simple formula for derivative of F (in the R2
-topology) that reads
F (u)v =
d
dt
f(u + tv)
t=0
. (6)
Finally, for any u, v ∈ Cper([−π, π]; C) we define the inner product
u, v :=
π
−π
u(x)v(x)dx, (7)
where u denotes the complex conjugate of u.
2 Continuum limit approach
Completely synchronized and twisted solutions of Eqs. (2)–(3) are known explicitly, therefore
their dynamical properties can be analyzed in the framework of this finite-dimensional sys-
tem [35, 36]. In contrary, numerical simulations demonstrate [26] that coherence-incoherence
patterns are hyper-chaotic trajectories with the number of positive Lyapunov exponents being
proportional to the system size N. For increasing N such patterns seem to become tremen-
dously complicated objects. However, this is not the case. In fact their macroscopic dynamics
for N → ∞ turns out to be in a certain sense equivalent to the dynamics of a periodic orbit of
some deterministic infinite-dimensional evolution equation.
4

2.1 Continuum limit equation
If N → ∞ then instead of looking for individual phases θk(t) we may look for their probability
distribution f(θ, x, t), which for fixed time t gives a relative number of oscillators with θk(t) ≈ θ
and xk ≈ x. Evolution equation for distribution f can be derived phenomenologically as we
explain below. Its rigorous justification is more involved and can be found somewhere else, see
for example recent work of Luçon and Stannat [37].
First, we rewrite Eq. (2) in a local form
dθk
dt
= ω + Im Zk(t)e−i(θk(t)+α)
, k = 1, . . ., N, (8)
where each phase θk is coupled solely with the averaged driving force
Zk(t) =
2π
N
N
j=1
Gkjeiθj (t)
.
For N → ∞, the latter sum can be formally replaced with a value of integral
Z(x, t) =
π
−π
G(x − y)
2π
0
f(θ, y, t)eiθ
dθ dy (9)
evaluated at the position of the k-th oscillator x = xk. Suppose for a moment that all Zk(t) =
Z(xk, t) are known, then dynamics of f corresponding to Eq. (8) has to obey a continuity
equation
∂f
∂t
+
∂
∂θ
(fJ) = 0, (10)
where
J(θ, x, t) = ω + Im Z(x, t)e−i(θ+α)
(11)
is the right-hand side of Eq. (8). However in our case Z(x, t) is not given rather depends
on f, see (9). Hence self-consistent dynamics of probability distribution f is governed by the
nonlinear integro-differential equation (10)–(11) where Z(x, t) is expressed by formula (9). Note
the integral term of Eq. (10) is hidden in formula (9) whereas quadratic nonlinearity appears
from the product fJ. Following established tradition we call Eqs. (10)–(11) the continuum limit
equation.
2.2 Ott-Antonsen invariant manifold and local order parameter
Direct study of continuum limit equation is quite a difficult task, in particular because its solutions
typically are generalized functions, or measures. Already in the case of bulk synchronization,
bifurcation analysis of continuum limit equation is so complicated [38, 39] that most of interesting
facts about the qualitative behaviour of its solutions turn out to be hidden behind mathematical
details. Fortunately, the fact that phase interaction in Eq. (2) is given by a pure sinusoidal term
allows us to simplify the analysis of Eqs. (10)–(11) significantly. To this end, we use the Ott-
Antonsen method suggested in [40, 41]. Roughly speaking it says that Eqs. (10)–(11) have an
explicitly known attracting invariant manifold that contains all relevant dynamics of Eqs. (2)–(3)
for large N. Exact form of this manifold is described in the following lemma.
5

Lemma 1 Suppose that:
(i) z(x, ·) : [0, ∞) → Cper([−π, π]; C) satisfies the inequality |z(x, t)| < 1,
(ii) z(x, t) is a solution of the equation
dz
dt
= iωz(x, t) +
1
2
e−iα
Gz −
1
2
eiα
z2
(x, t)Gz, (12)
where G : Cper([−π, π]; C) → Cper([−π, π]; C) is an integral convolution operator defined
by
(Gϕ)(x) :=
π
−π
G(x − y)ϕ(y)dy. (13)
Then, probability distribution f given by
f(θ, x, t) =
1
2π
1 +
∞
n=1
zn
(x, t)einθ
+ zn
(x, t)e−inθ
, (14)
is a solution of the continuum limit equation (10)–(11).
Proof: For |z| < 1 the series in the right-hand side of (14) converges absolutely. Substituting
this into Eqs. (10)–(11) and reordering terms appropriately we obtain
∂f
∂t
+
∂
∂θ
(fJ) =
1
2π
dz
dt
− iωz −
1
2
e−iα
Gz +
1
2
eiα
z2
Gz
∞
n=1
nzn−1
e−inθ
+
1
2π
dz
dt
+ iωz −
1
2
eiα
Gz +
1
2
e−iα
z2
Gz
∞
n=1
nzn−1
einθ
.
Clearly, if z satisfies Eq. (12) then each expression in parenthesis vanishes and hence Eq. (10)
is fulfilled. •
Remark 1 Eq. (12) was first obtained by Laing in [20]. Later on, we pointed out [26] an al-
ternative way to derive it using the Watanabe-Strogatz ansatz in the context of hierarchical
populations of coupled oscillators, see [42, 43] for detail.
Distribution f defined by formula (14) has a clear interpretation with respect to the complex-
valued function z. Indeed for any |z(x, t)| < 1, summation in (14) yields
Pz(θ) =
1
2π
1 − |z|2
1 − 2|z| cos(θ − arg z) + |z|2
that is a Poissonian distribution in the phase θ (see Fig. 2). In particular, arg z indicates the
location of its center, whereas |z| characterizes the degree of non-uniformity of this distribution.
For |z| = 1, distribution (14) degenerates into a delta function
Pz(θ) = δ(θ − arg z),
6

θ
Pz(θ)
arg z-π+arg z π+arg z
(a)
θ
Pz(θ)
(b)
θ
Pz(θ)
(c)
Figure 2: Poissonian distributions Pz(θ) for (a): |z| = 0, (b): 0 < |z| < 1, and (c): |z| = 1.
therefore we call oscillators with sites xk ≈ x coherent. In contrary, oscillators with posi-
tions xk ≈ x and |z(x, t)| < 1 are referred to as incoherent.
For any f given by (14) it holds
z(x, t) =
2π
0
f(θ, y, t)eiθ
dθ.
This formula resembles the well-known Kuramoto’s order parameter [44] with the only difference
that |z(x, t)| measures the synchronization of oscillators around point x rather than the global
synchronization of all oscillators, therefore we call function z the local order parameter. Note
the concept of local mean filed is consistent with Eq. (12). Indeed, Eq. (12) defines a smooth
dynamical system on the Banach space Cper([−π, π]; C). Moreover next lemma shows if we
take an initial data z0 such that z0 0 ≤ 1 and trace its evolution according to Eq. (12) then at
every time moment t we get |z(x, t)| ≤ 1.
Lemma 2 Eq. (12) has an invariant set of the form
U := {z ∈ Cper([−π, π]; C) : z 0 ≤ 1} .
Proof: The complex conjugate of Eq. (12) reads
dz
dt
= −iωz(x, t) +
1
2
eiα
Gz −
1
2
e−iα
z2
(x, t)Gz,
hence
d|z|2
dt
= z
dz
dt
+ z
dz
dt
=
1 − |z|2
2
eiα
zGz + e−iα
zGz
= 1 − |z|2
Re eiα
zGz .
Taking into account that z(x, t) is a continuous function of its arguments, for any fixed x ∈
[−π, π] we get either |z(x, t)| = 1 for all t ∈ R, or |z(x, t)| = 1 for all t ∈ R. This implies the
invariance of set U. •
7

2.3 Standing wave solutions and self-consistency equation
For particular examples of coherence-incoherence patterns considered in [19, 20] it was shown
that they have periodic orbit counterparts lying in the Ott-Antonsen manifold. Assuming this
correspondence as general conjecture, we develop a strategy for analysis of periodic solutions
of Eq. (12).
We remark that Eq. (12) has two continuous symmetries. It is equivariant with respect to:
spatial translations z(x, t) → z(x + s, t), s ∈ R,
and complex phase shifts z(x, t) → eiϕ
z(x, t), ϕ ∈ R.
Moreover, since coupling function G is even, the equation is also symmetric with respect to
spatial reflections z(x, t) → z(−x, t).
These symmetries suggest us to seek periodic solutions of Eq. (12) in the following form
z(x, t) = a(x)eiΩt
, (15)
where Ω ∈ R is a collective frequency, and a ∈ Cper([−π, π]; C) is a spatial profile. In
accordance with the probabilistic interpretation of local order parameter z, we have to ensure
that |a(x)| ≤ 1 for all x ∈ [−π, π]. Then, we refer to points x ∈ [−π, π] with |a(x)| = 1 and
|a(x)| < 1 as to the coherent and incoherent regions, respectively. In such a way we divide all
the solutions described by ansatz (15) into three groups:
Name of solution Description
Coherent (or phase-locked) state |a(x)| = 1 for all x ∈ [−π, π]
Incoherent state |a(x)| < 1 for all x ∈ [−π, π]
Coherence-incoherence pattern Both coherent and incoherent regions
are non-empty
For coherent and incoherent states, we also distinguish spatially uniform (if a(x) is constant)
and spatially modulated (otherwise) forms of these states. Finally, we call trivial solution z ≡ 0
the completely incoherent state.
Substituting ansatz (15) into Eq. (12) we obtain a self-consistency equation
eiα
a2
(x)Ga − 2i(ω − Ω)a(x) − e−iα
Ga = 0. (16)
To clarify its nature we perform a series of transformations. Assuming that Ω = ω we define a
new unknown function
w(x) := (ω − Ω)−1
(Ga) (x), (17)
8

which is an analog of the local mean field in [12]. Then we rewrite Eq. (16) in the following form
e−iβ
w(x)a2
(x) − 2a(x) + eiβ
w(x) = 0, (18)
where
β :=
π
2
− α. (19)
For fixed w(x), quadratic Eq. (18) has two solution branches
a±(x) =



1 ± 1 − |w(x)|2
e−iβ
w(x)
for |w(x)| < 1,
1 ± i |w(x)|2 − 1
e−iβ
w(x)
for |w(x)| ≥ 1,
(20)
which satisfy |a−(x)||a+(x)| = 1. If |w(x)| < 1 then it holds |a−(x)| < 1 and |a+(x)| >
1. Therefore in order to be consistent with the requirement |a(x)| ≤ 1 we have to assume
a(x) = a−(x). On the other hand, if |w(x)| ≥ 1 we get |a−(x)| = |a+(x)| = 1, and
both branches a−(x) and a+(x) are appropriate candidates for a(x). Taking into account the
above definition of coherent and incoherent regions, we characterize them with the help of
function w(x) as follows
Scoh(w) := {x ∈ [−π, π] : |w(x)| ≥ 1} ,
Sincoh(w) := {x ∈ [−π, π] : |w(x)| < 1} . (21)
Now, to make definition (17) consistent with Eq. (18) we have to assume that function w(x)
satisfies
w(x) = (ω − Ω)−1



Sincoh(w)
G(x − y)
1 − 1 − |w(y)|2
e−iβw(y)
dy
+
Scoh(w)
G(x − y)
1 ± i |w(y)|2 − 1
e−iβw(y)
dy


 . (22)
Note that depending on the choice a(x) = a−(x) or a(x) = a+(x) in the coherent re-
gion Scoh(w), we obtain two different equations (22)− and (22)+. Defining a complex parameter
µ := (ω − Ω)e−iβ
= (ω − Ω)e−i(π/2−α)
(23)
and a complex-valued function
h(s) :=



1 −
√
1 − s
s = 1
1 +
√
1 − s
for 0 ≤ s < 1,
1 − i
√
s − 1
s = 1
1 + i
√
s − 1
for s ≥ 1,
(24)
9

we rewrite Eqs. (22)− and (22)+ as
µw(x) =
π
−π
G(x − y)h |w(y)|2
w(y)dy, (25)
and
µw(x) =
π
−π
G(x − y)h |w(y)|2
w(y)dy, (26)
respectively. These Eqs. (25) and (26) are nonlinear eigenvalue problems (NEVP) to be solved
with respect to µ ∈ C and w ∈ Cper([−π, π]; C) simultaneously. Obviously they are not
independent: Every solution (µ, w) of Eq. (25) gives a solution (µ, w) of Eq. (26) and vice
versa.
The way we derived Eqs. (25) and (26) implies that every solution (µ, w) of Eq. (25) or Eq. (26)
corresponds to a solution of Eq. (12), which is given by ansatz (15) with collective frequency Ω
and spatial profile a(x) determined by formulas (20) and (23). Note that for any µ = 0 iden-
tity (23) has two equivalent forms
(ω − Ω)e−iβ
= µ = |µ|ei arg µ
= −|µ|ei(π+arg µ)
,
and hence two possible choices of (Ω, β). This fact together with the relationship between
solutions of Eqs. (25) and (26) results in the following
Proposition 1 Suppose that pair (µ, w) ∈ C×Cper([−π, π]; C) solves Eq. (25), then Eq. (12)
has four solutions of the form (15):
(i) Ω = ω − |µ|, a(x) = h (|w(x)|2
) w(x) for β = − arg µ,
(ii) Ω = ω + |µ|, a(x) = h (|w(x)|2
) w(x) for β = −π − arg µ,
(iii) Ω = ω − |µ|, a(x) = h (|w(x)|2
) w(x) for β = arg µ,
(iv) Ω = ω + |µ|, a(x) = h (|w(x)|2
) w(x) for β = π + arg µ.
Remark 2 If pair (µ, w) is real, e.g. µ ∈ R and w ∈ Cper([−π, π]; R), then solutions defined
in lines (iii) and (iv) are identical to those from lines (i) and (ii), respectively.
Proposition 1 claims that every solution of Eq. (25) or (26) corresponds to some standing wave
solution of Eq. (12). But the opposite, in general, is not true. Indeed, deriving Eq. (22) we as-
sumed that either a(x) = a−(x) or a(x) = a+(x) uniformly for all x ∈ Scoh(w). Now suppose
that the set Scoh(w) consists of several disjoint intervals. Then, in each of those intervals we
may select between branches a−(x) and a+(x) independently, without violating the continuity
of the resulting profile a(x). Every such choice produces a different, more complicated form of
Eq. (22). Solving it we obtain standing waves, which are described neither by Eq. (25) nor by
Eq. (26). Fortunately, these ’exotic’ standing wave solutions turn out to be unstable, see Propo-
sition 2. Therefore, we may expect that solving Eq. (25) and applying Proposition 1 we find all
solutions relevant to observable states of system (2)–(3).
10

0
-π
π
θk
(a) (b) (c)
0
0
0.5
1.0
-π
π
|a(x)|arga(x)
0
0
0.5
1.0
-π 0 π
x
-π
π
|w(x)|argw(x)
-π 0 π
x
-π 0 π
x
Figure 3: Continuum limit representation of two coherence-incoherence patterns (a) and (b), and
completely synchronized state (c). Top panels display snapshots of the states found in numerical
simulation of system (2)–(3) with N = 500, ω = 0, α = π/2 − 0.1, and cosine coupling (4)
where A = 0.9. Other panels beneath display the amplitude a(x) of corresponding standing
wave, and the corresponding complex proﬁle w(x) = (ω − Ω)−1
(Gw)(x).
11

Now we can outline our program for bifurcation analysis of coherence-incoherence patterns in
system (2)–(3). Fig. 3 illustrates our main tools. Instead of looking for generalized solutions of
continuum limit equation (10)–(11) we look for continuous functions z(x, t) solving Eq. (12). We
focus on standing wave solutions described by ansatz (15) with a ∈ Cper([−π, π]; R) and Ω ∈
R. The coherent and incoherent regions of such waves are fixed by conditions |a(x)| = 1
and |a(x)| < 1, respectively. Therefore spatial amplitude a is usually continuous but not
smooth. Replacing Eq. (16) by Eq. (25) we typically get a smooth unknown function w such that
coherent and incoherent regions are determined by conditions |w(x)| ≥ 1 and |w(x)| < 1, re-
spectively. In the next section we will try to characterize complete solution set of Eq. (25). Then
applying Proposition 1 we transform each of found solutions (µ, w) into standing waves (15).
Finally performing stability analysis of these waves with respect to Eq. (12) we will select really
observable coherence-incoherence patterns.
3 Nonlinear eigenvalue problem
Defining a nonlinear substitution operator
H : Cper([−π, π]; C) → Cper([−π, π]; C), H(w) := h |w|2
w, (27)
where h is given by (24), and recalling that G denotes the integral convolution operator (13), we
rewrite Eq. (25) in the abstract form
F(µ, w) := µw − GH(w) = 0. (28)
This is a nonlinear eigenvalue problem to be solved with respect to pair of unknowns (µ, w) ∈
C × Cper([−π, π]; C). Its infinite-dimensional nonlinearity GH(w) has comparatively sim-
ple mathematical nature since for coupling function G relevant to present study, correspond-
ing operator G is usually compact in Cper([−π, π]; C) or even in C1
per([−π, π]; C). However,
NEVP (28) inherits all continuous and discrete symmetries of Eq. (12), therefore its analysis re-
quires new approaches going beyond the classical paradigm of NEVP formulated by M. A. Kras-
noselskii [45] and P. H. Rabinowitz [46].
Above we have seen that typical dynamical regimes observed in large N system (2)–(3) can
be described as standing wave solutions of Eq. (12). In order to find such periodic solutions we
use NEVP (28) and the correspondence between its solutions and standing waves of Eq. (12)
given by Proposition 1. In Section 3.1, we address some auxiliary questions concerned with
Fourier representation of integral operator G and differentiability of substitution operator H.
Then, in Section 3.2 we show that in general case solution set of NEVP (28) is locally organized
as a curve (modulo the symmetry group action) in the Banach space C × Cper([−π, π]; C).
Some explicitly known solutions of NEVP (28) are described in Section 3.3. These solutions
are counterparts of completely synchronized and twisted solutions in original system (2)–(3). In
Section 3.4 we analyze primary solution branches bifurcating from the trivial solution w = 0.
Then, in Section 3.5 we describe a sequence of secondary solution branches which appear
from the primary branch of spatially uniform solutions. At the end we present two illustrative
examples for piecewise-constant and cosine coupling.
12

3.1 Auxiliary facts about operators G and H
Generalizing examples from [12, 14, 23], let us suppose that coupling function G satisfies the
following conditions
it is piecewise-smooth, that means it is C1
-smooth on a finite number of disjoint intervals,
obtained as a partition of the interval [−π, π],
it is non-constant,
it is even, i.e. G(−x) = G(x),
it is 2π-periodic.
These conditions in particular imply that G has an absolutely converging Fourier series
G(x) =
∞
k=−∞
gkeikx
, with coefficients gk =
1
2π
π
−π
G(x)e−ikx
dx. (29)
Since G is even, all gk’s are real and it holds gk = g−k. Formula (29) also implies that opera-
tor G is diagonizable in the Fourier space and can be represented in the form
(Gw) (x) =
∞
k=−∞
gkeikx
π
−π
e−iky
w(y)dy.
Applying this we easily get
Geikx
(x) = 2πgkeikx
for any k ∈ Z. (30)
Moreover, using identity gk = g−k it is easy to verify that for any integer k ≥ 0 and any function
u ∈ Cper([−π, π]; C) it holds
cos(kx), Gu = 2πgk cos(kx), u ,
sin(kx), Gu = 2πgk sin(kx), u ,
where inner product ·, · is defined by (7).
Presence of integral operator G usually makes Eq. (28) infinite-dimensional. However, perform-
ing its numerical evaluation we often can replace this operator with a finite-rank approximation
obtained via truncating the Fourier series (29). For example, if we define
(GKw)(x) :=
π
−π
GK(x − y)w(y)dy, K = 1, 2, . . .,
where
GK(x) =
K
k=−K
gkeikx
= g0 +
K
k=1
2gk cos(kx), (31)
13

then using the Hölder inequality and the L2
-convergence of Fourier series (29) we easily obtain
lim
K→∞
G − GK = 0, (32)
where convergence takes place in the strong operator norm of Banach space Cper([−π, π]; C).
Infinite-dimensional range of operator G is not the only difficulty concerned with NEVP (28). In
principle, we could expect that local structure of its solution set can be revealed via the Implicit
Function Theorem. But, for this we must know the derivative
∂wF(µ, w) = µI − GH (w), (33)
which is not correctly defined on Cper([−π, π]; C) because definition of operator H contains
the absolute value |w|. To avoid this problem let us identify Banach space Cper([−π, π]; C)
with its real counterpart Cper([−π, π]; R2
). In other words, let us consider Eq. (28) as a system
of two equations for real and imaginary parts separately. After such a reformulation, we easily
calculate
H (w)v =
d
ds
h(|w + sv|2
)(w + sv)
s=0
= h(|w|2
)v + 2wh (|w|2
) Re(wv), (34)
where
h (s) =



1
2(1 +
√
1 − s)2
√
1 − s
for 0 ≤ s < 1,
− i
2(1 + i
√
s − 1)2
√
s − 1
for s ≥ 1.
(35)
Note that derivative h (s) has a singularity at s = 1, therefore operator H still fails to be
differentable on the whole space Cper([−π, π]; R2
). On the other hand, within the ball
w ∈ Cper([−π, π]; R2
) : |w(x)| < 1 for all x ∈ [−π, π]
it has derivatives of arbitrary order. In particular,
H (w)v2
= 4vh (|w|2
) Re(wv) + 2wh (|w|2
)|v|2
+ 4wh (|w|2
) Re2
(wv), (36)
H (w)v3
= 6h (|w|2
)|v|2
v + 12wh (|w|2
)|v|2
Re(wv)
+ 12vh (|w|2
) Re2
(wv) + 8wh (|w|2
) Re3
(wv). (37)
Above we have seen that for coherence-incoherence patterns both inequalities |w(x)| < 1
and |w(x)| ≥ 1 occur at different positions x. Therefore nondifferentiability of H is unavoidable.
Fortunately this makes no problem if we take into account the smoothening effect of integral op-
erator G. Indeed, suppose that (µ, w) is a solution of NEVP (28), then for any piecewise-smooth
coupling function G it holds w ∈ C1
per([−π, π]; R2
). Hence, without loss of generality we may
restrict NEVP (28) to the subspace of smooth functions w. After we did this, we can prove the
local differentiability of product operator GH, what in most cases is enough for application of
the Implicit Function Theorem.
14

Lemma 3 Suppose that G is a piecewise-smooth function, then for any w0 ∈ C1
per([−π, π]; R2
)
satisfying
κ(w0) := inf
x∈[−π,π]
|w0(x)| − 1 +
d|w0(x)|2
dx
> 0, (38)
there exists δ > 0 such that operator GH(w) is differentiable at any w ∈ C1
per([−π, π]; R2
)
within the ball w − w0 1 < δ.
Proof: Due to assumption (38) equation |w0(x)| = 1 has at most a ﬁnite number, say L, of
simple roots ξ0
1, . . . , ξ0
L ∈ [−π, π]. Moreover, for each ξ0
k there exists δk > 0 such that
|w0(x)| − 1 ≥
κ
2
|x − ξ0
k| for all |x − ξ0
k| < δk.
Hence, choosing δ > 0 small enough, we ensure that for all w ∈ C1
per([−π, π]; R2
) within the
ball w − w0 1 < δ the following properties are fulﬁlled:
(i) The equation |w(x)| = 1 has exactly L simple roots ξ1(w), . . ., ξL(w), which continuously
depend on w, and ξk(w0) = ξ0
k, k = 1, . . . , L.
(ii) For all |x − ξk(w)| < δk/2 it holds
|w(x)| − 1 ≥
κ
4
|x − ξk(w)| .
The latter inequality together with formula (35) imply
f |w(x)|2
≤ C0 |w(x)| − 1
−1/2
≤
4C0
κ
|x − ξk(w)|−1/2
for all |x − ξk(w)| < δk/2, where C0(δ) > 0 is a constant independent of w. Now, from
formula (34) we easily see that the substitution operator
H : C1
per([−π, π]; R2
) → L1
per([−π, π]; C)
is differentiable at any w ∈ C1
per([−π, π]; R2
) within the ball w − w0 1 < δ. On the
other hand, for any smooth coupling function G the corresponding convolution operator G is
a bounded linear operator from L1
per([−π, π]; C) into C1
per([−π, π]; R2
). Hence, the assertion
of lemma is proved. •
Roughly speaking, Lemma 3 claims that product operator GH is differentiable everywhere
in C1
per([−π, π]; R2
) except of the zero measure set
Σ0 := w ∈ C1
per([−π, π]; R2
) : κ(w) = 0 . (39)
Thus we may expect that local behavior of the solution set of NEVP (28) is generically de-
termined by properties of the linearized operator ∂wF(µ, w). However, a particular singular
behaviour may happen if w passes through the set Σ0.
15

3.2 Local structure of the solution set of NEVP
Suppose that we know a particular solution (µ0, w0) of NEVP (28). Then, what can we say
about the local structure of its solution set close to this point? Below we provide a constructive
answer on this question.
We assume that µ0 = 0 and w0 = 0 (the case of trivial solution w0 = 0 is considered in
Section 3.4), then we choose a nonvanishing function ψ ∈ C1
per([−π, π]; R) such that
p0 =
ψ, w0
ψ, ψ
= 0 (40)
(such ψ can always be found for example among the Fourier harmonics cos(kx) and sin(kx),
k = 0, 1, . . .). Let us deﬁne a Fourier projection of the form
Pψw :=
ψ, w
ψ, ψ
ψ,
with the kernel
Πψ := w ∈ C1
per([−π, π]; R2
) : Pψw = 0 .
Applying operators Pψ and I − Pψ to Eq. (28) and omitting the non-zero factor ψ/ ψ, ψ in
the former result, we rewrite Eq. (28) as an equivalent system
µ ψ, w = ψ, GH(w) , (41)
µ(I − Pψ)w = (I − Pψ)GH(w). (42)
Then taking into account assumption (40), we divide Eq. (42) by Eq. (41) and obtain
(I − Pψ)w
ψ, w
=
(I − Pψ)GH(w)
ψ, GH(w)
. (43)
System of two equations (41) and (43) is equivalent (at least in a vicinity of point (µ0, w0)) to
original Eq. (28). On the other hand, Eq. (43) is decoupled from Eq. (41) and does not contain
complex unknown µ. In order to solve Eq. (43) we represent w in the form
w = v + pψ, where v ∈ Πψ and p ∈ R+. (44)
Such a decomposition is uniquely determined due to the phase-shift symmetry of original Eq. (28).
Substituting ansatz (44) into Eq. (43) we get
v =
(I − Pψ)GH(v + pψ)
ψ, GH(v + pψ)
ψ, ψ p =: Eψ(v, p). (45)
This is an equation to be solved with respect to v ∈ Πψ, for a given p ∈ R+ close to p0. As
soon as Eq. (45) is solved, formulas (41) and (44) determine the corresponding pair (µ, w).
Hence, we have established a one-to-one correspondence between solutions of Eq. (45) and
solutions of Eq. (28) which obey the phase-pinning condition ψ, w = | ψ, w |.
16

Remark if ψ is not identically constant, then Eq. (45) generically does not inherit continuous
symmetries of Eq. (28). That means, Eq. (45) is not equivariant either with respect to spatial
translations, or with respect to phase-shifts (the latter symmetry has gone after we had fixed p
to be real). This observation clarifies the true nature of Eq. (28) and opens the way for computing
its approximate solutions with the help of classical numerical schemes. Indeed, suppose that for
a given w0 = 0 there exists bounded derivative GH (w0), then for appropriate choice of ψ
satisfying (40), we can linearize reduced Eq. (45) around v0, where v0 is uniquely determined
from the decomposition w0 = v0 + p0ψ with v0 ∈ Πψ and p0 ∈ R+. Corresponding linearized
operator I − ∂vEψ(v0, p0) is Fredholm of index zero (as a sum of invertible I plus compact
operator), and hence it has at most a finite dimensional kernel. Therefore solving Eq. (45) we
generically meet one of the following three situations:
the derivative GH (w0) is well-defined and ker(I − ∂vEψ(v0, p0)) = {0}, then there
exists a unique, locally determined solution curve v = v(p) such that v(p0) = v0;
the derivative GH (w0) is well-defined but ker(I − ∂vEψ(v0, p0)) = {0}, then point
(v0, p0) is a bifurcation point of Eq. (45) with finite codimension;
operator GH is not differentiable at w0, then analysis gets more involved but it is possbile
to demonstrate that close to point w0 we typically observe a particular transformation of
coherence-incoherence pattern when the pattern gets/loses a coherence region.
If operator GH(w) is differentiable at w0, then reduced Eq. (45) can be solved locally by apply-
ing a variant of Galerkin’s method. For the sake of completeness we recall general framework
of this approach. Suppose that we chose ψ for decomposition (44) and look for solution v of
Eq. (45) in a Banach space X ⊂ Πψ which has an orthogonal basis {vk}. This solution can be
found approximately as a sum of the form
v ≈
K
k=1
ˆvkvk, (46)
where coefficients ˆvk ∈ C are determined from the projected system
vn, vn ˆvn = vn, Eψ
K
k=1
ˆvkvk, p , n = 1, . . ., K. (47)
In all cases relevant to present study, operator Eψ(v, p) is compact on X. Therefore for suf-
ficiently large integer K, sum (46) approximates all exact solutions of Eq. (45) that exist in a
small vicinity of v0 = w0 − Pψw0. More precisely, solving system (47) we distinguish all local
bifurcations occurring in the non-reduced Eq. (28).
But what shall we do with singular points (µ0, w0) where operator GH is not differentiable? In
principle, we may look for appropriate substitution revealing bifurcation scenario there. However,
we choose another, simpler way. Let us define H (w) = h (|w|)w where
h (s) =
1
1 +
√
1 − s + i
and > 0.
17

For any fixed > 0 operator H (w) is differentiable in Cper([−π, π]; R2
) and satisfies
lim
→+0
H (w) = H(w).
Hence, if in Eqs. (46) we replace operator GH by its regularized analog GH and solve the
resulting system, then we obtain an -dependent solution curve which passes through the ’dan-
gerous’ vicinity of point (µ0, w0). Now restoring operator GH in Eqs. (46) and taking the -
dependent solution as initial guess for the Newton’s scheme, we can calculate two parts of true
solution curve (for → +0) approaching singular point from two different directions. Typically
we obtain a solution curve that bends at the singular point, see Figs. 5 and 6, therefore we refer
to this singularity as to the bending bifurcation point. Such behaviour may look strange, but if
we recall that solving Eq. (18) we obtained two complex-conjugate equations (25) and (26) it
becomes clear that bending bifurcation is nothing else but a reminiscence of complex fold.
3.3 Explicit solutions of NEVP
Some solutions of Eq. (28) can be found explicitly. They have a form
w(x) = peikx
where k ∈ Z and p ∈ (0, ∞). (48)
Indeed, substituting ansatz (48) into Eq. (28) and taking into account identity (30) we get
µ − 2πgkh(p2
) peikx
= 0
where function h was defined above, see (24). Hence for every k ∈ Z formula (48) determines
a nontrivial solution of Eq. (28), provided
µ = 2πgkh(p2
). (49)
Applying definition of function h we rewrite this equation more explicitly distinguishing two par-
ticular cases:
p < 1 and µ =
2πgk
1 + 1 − p2
, (50)
p ≥ 1 and µ =
2πgk
1 − i p2 − 1
. (51)
In the former case, µ is real, see Fig. 4. Therefore definition (23) implies that all solutions (48)
with p ∈ (0, 1) correspond to two particular phase lags only, α = π/2 and α = −π/2. In
contrary, for p ≥ 1 formulas (48) and (51) connect different values p with different values α.
Moreover according to the local order parameter interpretation, see Section 2, we may consider
solutions (48) with p ≥ 1 as counterparts of the completely synchronized state (for k = 0) or
twisted states (for k = 0) observed in original system (2)–(3).
18

0
π
2π
|µ|
0
π/2
0 1 5
argµ
p
Figure 4: Explicit solution of NEVP (28) determined by Eq. (50) (dashed line) and Eq. (51) (solid
line) for gk = 1.
3.4 Primary solution branches
We start off with a more systematic approach for describing the solution set of NEVP (28). To
this end, we linearize this equation around its trivial solution w = 0 and look for such values µ
where conditions of the Implicit Function Theorem are violated. Applying formulas (33) and (34)
we obtain
∂wF(µ, 0) = µI − GH (0) = µI −
1
2
G. (52)
Spectrum of operator G is known explicitly, see Section 3.1, therefore we easily demonstrate
that derivative ∂wF(µ, 0) is not invertible if
µ = µk := πgk, k = 0, 1, 2, . . .. (53)
Moreover, in generic case for each non-zero µk it holds
ker ∂wF(µk, 0) = span eikx
, e−ikx
.
Remark that Eq. (49) implies 2πgkh(0) = µk, hence all solutions of NEVP (28) described by
formula (48) bifurcate from zero exactly at critical values of operator ∂wF(µ, 0). However, there
is also another sequence of primary branches bifurcating at the same values µk.
Lemma 4 Suppose that gk, k ≥ 1, is a non-zero Fourier coefﬁcient of G.
Then, for all sufﬁciently small µ − µk ∈ R such that (µ − µk)gk > 0, generically there exists
a non-trivial solution of Eq. (28) with an asymptotic
w(x) = 4
1
3πgk
(µ − µk) sin(kx) + O(|µ − µk|) for µ → µk. (54)
19

Proof: We restrict Eq. (28) to its invariant subspace
(µ, w) ∈ R × Cper,odd([−π, π]; R)
where
Cper,odd([−π, π]; R) := w ∈ Cper([−π, π]; R) : w(−x) = −w(x) .
Then derivative ∂wF(µ, 0) is degenerate at µ = µk with one-dimensional kernel spanned
by sin(kx). Note the latter claim is not true if some other Fourier coefficient of G coincides
with gk, but such situation is non-generic. We proceed further with a standard Lyapunov-Schmidt
reduction. For this, we define ε2
= µ − µk > 0 and substitute an ansatz
w = εw1 + ε2
w2 + ε3
w3 + . . . (55)
into Eq. (28). Then expanding nonlinearity GF(w) with the help of asymptotic formula
GF(w) =
1
2
Gw +
1
8
Gw3
+ O( w 5
0) for w 0 → 0,
and collecting the same order terms, we obtain a sequence of equations determining wk. In
particular,
µkI −
1
2
G w1 = 0 and µkI −
1
2
G w2 = 0.
This implies w1(x) = p1 sin(kx) and w2(x) = p2 sin(kx) where real constants p1 and p2 are
not known at this step. For the next order we get
µkI −
1
2
G w3 = −w1 +
1
8
Gw3
1.
According to the Fredholm alternative, this equation has a solution if and only if its right-hand
side satisfies the solvability condition
π
−π
sin(kx) −w1(x) +
1
8
Gw3
1 (x) dx = πp1
3
16
πgkp2
1 − 1 = 0.
For gk > 0, this algebraic equation has a pair of opposite sign solutions p1 resulting in the
leading term of asymptotics (54) (in the case gk < 0 we have to assume ε2
= −(µ − µk)
and then to repeat all above arguments). As soon as we know the leading term, we prove the
existence of corresponding solution applying the Implicit Function Theorem. •
In general case, all Fourier coefficients gk with k ≥ 0 are non-zero and different. Hence,
Lemma 4 defines infinite number of primary branches bifurcating from the trivial solution. All
these branches can be numerically continuated as explained in Section 3.2. More precisely, for
each k ∈ N we choose ψ = sin(kx) and solve Galerkin’s system (47) in a Banach subspace
X ⊂ Cper([−π, π]; C) corresponding to the isotropy group of sin(kx). This Fourier harmonics
is invariant with respect to transformations
w(x) → w x +
2π
k
, w(x) → −w(−x), w x +
π
2k
→ w −x +
π
2k
,
20

therefore we fix X to be a subspace of Cper([−π, π]; C) with orthogonal basis
sin((2m − 1)kx), m = 2, 3, . . ..
Particular examples considered below reveal that all primary branches have similar qualitative
behaviour. Each of them initially lies in the ball Σ0, see definition (39), but later approaches its
boundary, undergoes bending bifurcation and gives rise to a curve of coherence-incoherence
patterns with equidistant coherent regions, e.g. |w(x)| ≥ 1, emerging in the incoherent back-
ground, e.g. |w(x)| < 1.
3.5 Secondary solution branches
Performing numerical continuation of primary branches we may encounter other (secondary)
bifurcation points. For example, let us consider the simplest of explicitly known solutions (48)
with k = 0. For p ∈ [0, 1] these solutions are determined by Eq. (50) which can be rewritten in
the form
w(x) ≡ p0(µ) :=
2πg0
µ
µ
πg0
− 1 where
1
2
≤
µ
2πg0
< 1. (56)
Applying formulas (34)–(35) we easily calculate
∂wF(µ, p0(µ)) = µI − GH (p0(µ)) = µI −
µ2
|2πg0(µ − 2πg0)|
G.
Hence critical values of ∂wF(µ, p0(µ)) again can be found from the point spectrum of opera-
tor G. Indeed operator ∂wF(µ, p0(µ)) is not invertible for every µ ∈ R which satisfies at least
one of the following equations
|2πg0(µ − 2πg0)|
µ2 = 2πgk, k = 1, 2, . . .,
which are equivalent to equations
2πg0
µ
µ
2πg0
− 1 =
gk
g0
, k = 1, 2, . . .. (57)
It is easy to verify that each Eq. (57) has a unique solution µ/(2πg0) ∈ [1/2, 1) provided
gk/g0 ∈ (0, 1], and has no solution from this interval otherwise. If this solution exists, it is given
by
νk :=
2πg2
0
g0 + gk
. (58)
Remark that ν0 = πg0 is identical to the bifurcation value µ0 where primary branch of spatially
uniform solutions (48), e.g. branch with k = 0, appears from the trivial solution. Next lemma
shows that all other νk with k ≥ 1 are points of secondary bifurcation.
21

Lemma 5 Suppose that g0 = 0 and gk/g0 ∈ (0, 1] for some k ≥ 1.
Then for sufﬁciently small µ − νk ∈ R such that (µ − νk)C0/C1 > 0, generically there exists
a non-trivial solution of Eq. (28) with an asymptotic
w(y) = p0(νk) +
C0
C1
(µ − νk) cos(kx) + O(|µ − νk|) for µ → νk, (59)
where
C0 = 1 − 2πgkc2p0(νk),
C1 =
1
4
πgk c3 +
4πg0c2
2
νk − 2πg0c1
+
2πg2kc2
2
νk − 2πg2kc1
,
and
cm := H(m)
(p0(νk)) =
dk
duk
u
1 +
√
1 − u2
u=p0(νk)
.
Proof: The proof is a more technical version of that in Lemma 4. We restrict Eq. (28) to its
invariant subspace
(µ, w) ∈ R × Cper,even([−π, π]; R)
where
Cper,even([−π, π]; R) := w ∈ Cper([−π, π]; R) : w(−x) = w(x) .
Then we denote ε2
= µ − νk > 0 and look for solutions of the form
w = p0(νk + ε2
) + εw1 + ε2
w2 + ε3
w3 + . . . .
This ansatz implies
µw = (νk + ε2
)w = νkp0(νk) + ενkw1
+ ε2
νkw2 + p0(νk) + νkp0(νk) + ε3
νkw3 + w1 + O(ε4
)
and
H(p0(µ) + w) = H(p0(νk + ε2
) + w) = H(p0(νk)) + εH (p0(νk))w1
+ε2
H (p0(νk))(w2 + p0(νk)) +
1
2
H (p0(νk))w2
1
+ε3
H (p0(νk))w3 + H (p0(νk))w1(w2 + p0(νk)) +
1
6
H (p0(νk))w3
1 + O(ε4
).
Thus expanding Eq. (28) with respect to small parameter ε and collecting the same order terms,
we obtain equations
(νkI − c1G) w1 = 0, (60)
(νkI − c1G) w2 =
1
2
c2Gw2
1, (61)
(νkI − c1G) w3 = −w1 + c2Gw1(w2 + p0(νk)) +
1
6
a3Gw3
1. (62)
22

Note that to simplify the right-hand sides of Eqs. (61)–(62) we used identity F(µ, p0(µ)) = 0
and its full derivative with respect to µ.
Eqs. (60) and (61) can be solved explicitly. Thus we obtain
w1(x) = p1 cos(kx),
w2(x) = p2 cos(kx) +
1
4
c2p2
1
2πg0
νk − 2πg0c1
+
2πc2k
νk − 2πg2kc1
cos(2kx) ,
where p1, p2 ∈ R are not known at this step. Remark that in generic case denominators in
formula for w2 do not vanish. Now writing solvability condition for Eq. (62) and inserting there
expressions for w1 and w2, we get
π
−π
cos(kx) −w1(x) + c2 Gw1(w2 + p0(νk)) (x) +
1
6
c3 Gw3
1 (x) dx
= πp1(C1p2
1 − C0) = 0.
For C0/C1 > 0 this algebraic equation has a pair of nontrivial solutions p1 (for C0/C1 < 0
one has to repeat all arguments with ε2
= −(µ − νk)). Hence applying the Implicit Function
Theorem we can justify asymptotics (59). •
Similar to Section 3.4, we may suggest that in general case Lemma 5 defines infinite number
of secondary solution branches. However, this is only a part of the story. In fact, appearance of
secondary branches is more sensitive to the signs and absolute values of Fourier coefficients gk.
In particular, bifurcation corresponding to gk occurs only if gk/g0 ∈ (0, 1] and does not take
place otherwise. This is the main difference between primary branches and secondary branches
described here.
In order to continuate a secondary branch given by asymptotics (59) we again apply approx-
imate method of Section 3.2. Taking into account that solutions w(x) corresponding to gk
with k ≥ 1 are even and 2π/k-periodic, we choose ψ = cos(kx) and solve Galerkin’s sys-
tem (47) in a Banach subspace X ⊂ Cper([−π, π]; C) with orthogonal basis
1 and cos(mkx), m = 2, 3, . . ..
Particular examples considered below reveal that each secondary branch is born in the ball Σ0,
see definition (39), then extends to the ball’s boundary where it undergoes bending bifurcation
and gives rise to a curve of coherence-incoherence patterns with equidistant coherent regions,
e.g. |w(x)| ≥ 1, in the incoherent background, e.g. |w(x)| < 1.
We want to emphasize that our analysis of secondary branches is far from being complete.
For example, we did not consider k-twisted solutions described by formula (48) with k = 0.
Some secondary (or even higher order) bifurcation points may also occur along numerically
continuated curves of coherence-incoherence patterns, see for example recent work [30]. We
plan to address these questions in our future studies.
23

3.6 Example: Piecewise-constant coupling
Let us consider a particular example of NEVP (28) with piecewise-constant coupling function (5).
This function can be written in the form of complex Fourier series (29) with coefficients
g0 =
1
2π
and gk = g−k =
sin(πkr)
2π2kr
, k = 1, 2, . . ..
First we fix coupling radius to be r = 0.6 and apply analytical results and numerical algorithms
of Sections 3.2–3.5 to obtain a part of the NEVP’s solution set that corresponds to leading
Fourier coefficients g0, g1, g2 and g3. The result is shown in Fig. 5. For each solution (µ, w) ∈
||w||
Re µ
Im µ
µ0
µ1
µ2
µ3
ν1 1
0
4
-0.5
-π 0 π
x
0
1
-π 0 π
x
-π 0 π
x
0
1
-π 0 π
x
0
1
0
1
-π 0 π
x
0
1
-π 0 π
x
Figure 5: Solution of NEVP (28) for piecewise-constant coupling function (5) with coupling ra-
dius r = 0.6. Primary branches (black,blue,purple and orange thick lines) bifurcate from the
trivial solution w = 0 at µ = µk := πgk where gk are Fourier coefficients of coupling func-
tion G. The only secondary branch (red thick line) bifurcates from the primary branch of spa-
tially uniform solutions (black thick line) at µ = ν1, see (58). Each branch first lies in the plane
Im µ = 0 (dashed lines), but after approaching a bending bifurcation point (empty circle) it
turns and proceeds with complex values µ (solid lines). Thin lines show projections on the bot-
tom plane w = 0 of the corresponding thick solid lines with the same color. Additional panels,
using color-code of the main plot, demonstrate |w(x)|-profiles typical for each solution branch
before (left panel) and after (right panel) the bending bifurcation point. Inequalities |w(x)| ≥ 1
and |w(x)| < 1 correspond to coherent and incoherent regions, respectively. Thus all thick
solid lines in the main plot (except black one) denote coherence-incoherence patterns.
C × Cper([−π, π]; C) we show its 3D-projection where function w is replaced by its L2
-norm.
24

Thick black line (dashed and solid) displays primary branch of spatially uniform solutions (49)
which bifurcates at µ = µ0, see (53), and is described by Eqs. (50)–(51) with k = 0. Dashed
part of the line lies within the plane Im µ = 0. When the line approaches bending bifurcation
point (empty circle), it abruptly turns and leaves the plane Im µ = 0 becoming a solid line of
the same color. For the convenience of 3D-perception, we plot the projection of solid black line
on the plane w = 0 as a thin black line. Other k-twisted solutions defined by Eqs. (50)–(51)
are not shown.
In a similar way, we represent other primary (blue, purple and orange curves) and secondary
(red curve) solution branches discussed in Sections 3.4 and 3.5, respectively. The primary
branches bifurcate at values µ = µk, see (53), and the secondary branches bifurcate at val-
ues µ = νk, see (58). Remark an important role of the Fourier coefficients’ signs. Branches ap-
pearing at µ2, µ3 < 0 have different location compared to those appearing at µ0, µ1 > 0. This
means that they are observed for different values of phase lag parameter α, see formula (23).
Moreover, only positive Fourier coefficient g1 gives rise to a secondary solution branch bifur-
cating from the branch of spatially uniform solutions. Negative coefficients g2, g3 do not satisfy
condition gk ∈ (0, g0), therefore due to Lemma 5 they do not define bifurcation points νk.
Taking into account that Fourier coefficients of piecewise-constant coupling function (5) change
their signs for varying coupling radius r, we conclude that each secondary branch can be found
for particular interval (or intervals) of r only, whereas primary branches exist for all values of
coupling radius r.
Insert panels around the main plot are arranged in pairs. Using the same color-code they display
typical |w(x)|-profiles observed along each solution branch before (left panel) and after (right
panel) the corresponding bending bifurcation point marked as empty circle in the main plot.
Along dashed lines (which lie in the plane Im µ = 0) we obtain |w(x)|-profiles with |w(x)| < 1,
hence incoherent states. In contrary, every thick solid line of the main plot (except black one!)
denotes coherence-incoherence patterns since corresponding |w(x)|-profiles comprise both
patches of coherence |w(x)| ≥ 1 and incoherence |w(x)| < 1.
3.7 Example: Cosine coupling
Let us consider another example of NEVP (28) with cosine coupling function (4). A particular
feature of this function is its complex Fourier series (29) which has only three nonvanishing
coefficients
g0 =
1
2π
and g1 = g−1 =
A
4π
.
This means that applying results of Sections 3.4 and 3.5 we obtain only two primary solution
branches bifurcating at µ = µ0 and µ = µ1, and at most one secondary solution branch
bifurcating at µ = ν1, see formulas (53) and (58), respectively. We recall that due to Lemma 5
secondary branch exists for g1 ∈ (0, g0) only, that is equivalent to condition A ∈ (0, 2).
Remarkably, simple form of cosine coupling (4) allows us to write more explicitly primary branch
bifurcating at µ = µ1. Indeed, simple check demonstrates that function
w(x) = p sin x where p ∈ (0, ∞)
25

satisﬁes NEVP (28), provided
µ =
A
2π
π
−π
h p2
sin2
y sin2
y dy.
Hence this branch is completely known without any numerical continuation.
The solution set of NEVP (28) for cosine coupling function with A = 0.9 is shown in Fig. 6.
Here we used the same color-code representation as in Fig. 5. Again thick black line denotes a
||w||
Re µ
Im µ
µ0
µ1 ν1 1
0
4
-0.5
0
1
0
1
-π 0 π
x
0
1
-π 0 π
x
Figure 6: Solution of NEVP (28) for cosine coupling function (5) with parameter A = 0.9.
Notations and color-code are the same as in Fig. 5.
branch of spatially uniform solutions, whereas thick blue and red lines are another primary and
secondary branches, respectively. Solid parts of the curves represent dynamical regimes from
Fig. 3 in accordance with the color-code of their snapshots. Thus solid blue and red lines de-
note two different types of coherence-incoherence patterns. Remark the anti-phase alternation
between phases of neighbouring coherent regions in the coherence-incoherence pattern corre-
sponding to primary branch, see panel with arg a(x) in Fig. 3(a). This feature is not observed
in the secondary coherence-incoherence pattern, see Fig. 3(b), and apparently can be used to
distinguish primary and secondary patterns in numerical simulations of system (2)–(3).
4 Stability analysis
Suppose that z(x, t) = a(x)eiΩt
is one of the standing waves corresponding to a solu-
tion (µ, w) of NEVP (28), see Proposition 1. Below we consider stability properties of this
standing wave within the Ott-Antonsen manifold. To this end, we use an ansatz
z(x, t) = (a(x) + v(x, t))eiΩt
with a small perturbation v. Substituting this into Eq. (12) and linearizing the result with respect
to v, we obtain
dv
dt
= η(x)v −
i
2
eiβ
Gv + e−iβ
a(x)
2
Gv (63)
26

where
η(x) = i ω − Ω − e−iβ
a(x)Ga . (64)
Applying formulas (17), (19) and (20), we rewrite coefﬁcient η as follows
η(x) =
i(ω − Ω) 1 − |w(x)|2 for |w(x)| < 1,
±(ω − Ω) |w(x)|2 − 1 for |w(x)| ≥ 1,
(65)
where in the second line we use sign ’+’ or ’−’ depending on the choice a(x) = a+(x) or
a(x) = a−(x) in the coherent region. Eq. (12) contains complex conjugated terms therefore
one has to consider it not as a single complex equation, but as a short form of system of two real
equations. Respectively, we have to rewrite complex Eq. (63) as a system for a two component
vector-function
v(t) =
Re v(x, t)
Im v(x, t)
∈ Cper([−π, π]; R2
).
Then equivalent form of Eq. (63) reads
dv
dt
= Lv where L := M + K, (66)
and M : Cper([−π, π]; R2
) → Cper([−π, π]; R2
) is a multiplication operator
(Mv)(x) := M(x)v with M(x) :=
Re η(x) −Im η(x)
Im η(x) Re η(x)
, (67)
whereas K : Cper([−π, π]; R2
) → Cper([−π, π]; R2
) is an integral operator
(Kv)(x) :=
π
−π
K(x, y)v(y)dy
with a matrix kernel given by
K(x, y) =
1
2
QP(x)G(x − y) (68)
where for the sake of shortage we used notations
P(x) =



Re 1 + e−iβ
a(x)
2
−Im 1 − e−iβ
a(x)
2
Im 1 + e−iβ
a(x)
2
Re 1 − e−iβ
a(x)
2


 (69)
and
Q =
sin β cos β
− cos β sin β
.
Linear operator L is time-independent, hence based on the location of its spectrum σ(L) we
can make a conclusion about the stability of corresponding standing wave a(x)eiΩt
. It is easy to
27

see that for piecewise-smooth coupling function G and continuous wave amplitude a both op-
erators M and K are bounded linear operators in Cper([−π, π]; R2
). Moreover, deﬁnition (68)
implies that integral operator K is compact in Cper([−π, π]; R2
). Following general spectral
theory of linear operators in Banach space [47], we distinguish two parts in the spectrum σ(L).
The point spectrum σpt(L) ⊂ σ(L) consists of all complex numbers λ such that λI − L is a
Fredholm operator of index zero. In other words, each λ ∈ σpt(L) is an isolated eigenvalue of
ﬁnite multiplicity. The complement of discrete spectrum we call essential spectrum and denote
it as σess(L).
For multiplication operator M, it is known [48] that it has only essential spectrum consisting of
all λ ∈ C such that
det
λ − Re η(x) Im η(x)
−Im η(x) λ − Re η(x)
= 0 for some x ∈ [−π, π], (70)
therefore
σ(M) = σess(M) = η(x) : −π ≤ x ≤ π ∪ {c.c.} , (71)
where {c.c.} denotes the complex conjugate of the previous set. Essential spectrum is in-
variant under compact perturbations [49], hence σess(L) = σess(M). The remaining point
spectrum σpt(L) has no simple representation but can be calculated numerically as we explain
below. Anyhow for bounded compact operator K it is always known [48] that
σpt(L) ⊂ λ ∈ C : dist(λ, σess(M)) ≤ K . (72)
For every coherence-incoherence pattern found as a solution of NEVP (28), formula (65) gives
either real or pure imaginary values of η(x). Hence according to formula (71) such coherence-
incoherence pattern has a T-shaped essential spectrum σess(L) with a symmetric interval of
neutral eigenvalues corresponding to incoherent region. Another interval corresponds to coher-
ent region and consists solely of real eigenvalues, which all lie either in the left or in the right
complex half-plane, see Fig. 7(a) and (b). On the other hand, sign of Re η(x) is determined by
the choice of a = a+ or a = a− in coherent region. Therefore we easily verify that only two of
four standing waves given by Proposition 1 have stable essential spectrum.
Proposition 2 Suppose that (µ, w) is a solution of NEVP (28) which has coherent |w(x)| ≥ 1
as well as incoherent |w(x)| < 1 regions. Then standing waves (15) given by
Ω = ω − |µ|, a(x) = h |w(x)|2
w(x) with β = − arg µ
and
Ω = ω + |µ|, a(x) = h |w(x)|2
w(x) with β = π + arg µ
have stable essential spectrum, whereas the other two standing waves mentioned in Proposi-
tion 1 are always unstable (at least because of the location of their essential spectrum).
28

0
0
Imλ
Re λ
coherent part
incoherent part
(a)
0
0
Imλ
Re λ
coherent part
incoherent part
(b)
Figure 7: Two forms of essential spectrum calculated by formulas (65) and (71) for a coherence-
incoherence pattern. The interval along real axis and the other one along imaginary axis are
due to coherent |w(x)| ≥ 1 and incoherent |w(x)| < 1 parts, respectively. The real part of the
spectrum is stable (a) or unstable (b) depending on the choice of sign ’−’ or ’+’ in the second
line of Eq. (65).
Remark 3 Most of classical bifurcations can be inferred from the spectrum of corresponding
linearized operator. For example, Hopf bifurcation occurs when a pair of complex-conjugate
eigenvalues crosses the imaginary axis. Similar criterion exists for Turing bifurcation when one
looks for an intersection between the imaginary axis and continuous spectrum. In this context,
formulas (65) and (71) clarify spectral behaviour typical for the bending bifurcation deﬁned at
the end of Section 3.2. At this bifurcation point a new coherent region, e.g. |w(x)| ≥ 1, ap-
pears in the incoherent surrounding, e.g. |w(x)| < 1. Respectively, due to Eq. (65) essential
spectrum σess(L) gets a new portion of stable/unstable spectrum protruding at the zero from its
neutral part.
Stability analysis of standing wave (15) also relies on the analysis of corresponding point spec-
trum σpt(L). The latter consists of all λ ∈ C such that equation
(λI − M − K) v = 0 (73)
has a non-trivial solution v ∈ Cper([−π, π]; R2
). Note for λ /∈ σess(L), spectral equation (73)
can be rewritten in a more convenient form
v = (λI − M(x))−1
(Kv)(x) (74)
where I is the two-dimensional identity matrix and M is given by (67). In order to solve Eq. (74)
we apply the following approximate scheme. Taking into account operator convergence (32) we
replace compact operator K with its ﬁnite-rank approximation of the form
(Kv)(x) ≈
π
−π
1
2
QP(x)GK(x − y)v(y)dy
=
1
2
QP(x)
K
k=−K
gkeikx
eikx
, v (75)
29

where GK is given by a truncated Fourier sum (31) with sufficiently large K, and for every
v = (v1, v2)T
we use a notation
eikx
, v =
eikx
, v1
eikx
, v2
.
Then substituting (75) into Eq. (74) and calculating 2K + 1 non-vanishing Fourier projections
we obtain a system
eijx
, v =
K
k=−K
Bjk(λ) eikx
, v , j = −K, . . . , K, (76)
where
Bjk(λ) :=
1
2
gk eijx
, eikx
(λI − M(x))−1
QP(x) .
Obviously, Eqs. (76) have a nontrivial solution eikx
, v , k = −K, . . . , K, if and only if λ
satisfies a characteristic equation
χ(λ) := det (I − B(λ)) = 0 (77)
with a (4K + 2)-dimensional square matrix given by
B(λ) :=






B(−K)(−K)(λ) B(−K)(−K+1)(λ) . . . B(−K)K(λ)
B(−K+1)(−K)(λ) B(−K+1)(−K+1)(λ) . . . B(−K+1)K(λ)
...
...
...
...
BK(−K)(λ) BK(−K+1)(λ) . . . BKK(λ)






.
It is easy to see that each matrix Bjk(λ) is an analytic function of λ in Cσess(L). Hence
the same is true for function χ(λ) from Eq. (77). Thus taking into account spectral radius es-
timate (72) and the fact that the zeros of an analytic function are always isolated, we may
expect to find at most a finite number of solutions to Eq. (77). These solutions determine only
approximate point spectrum. However, operator convergence (32) insures that using in (75)
functions GK with larger K we obtain increasingly better approximations of σpt(L).
4.1 Example: Cosine coupling
Let us consider two coherence-incoherence patterns, see Fig. 6, which were obtained as pri-
mary and secondary solution branches of NEVP (28) with cosine coupling function (4). Then ap-
plying Proposition 2 for each (µ, w) we construct standing waves (15) corresponding to these
patterns. Such waves have stable essential spectrum, Fig. 7(a), but may have stable or un-
stable point spectrum. Note that Fourier series of cosine coupling function contains only three
nonvanishing harmonics therefore in order to calculate the point spectrum we don’t need to
approximate G. Indeed already for K = 1 characteristic equation (77) is exact. Solving this
equation numerically we found only a finite number of eigenvalues in point spectrum. Based on
30

0
0.1
0.2
0.3
0 1 2
A
π/2-α
(a)
0
-π
π
θk
0
-π
π
θk
0
0
(b)
λ
0
0
(c)
λ
Figure 8: Bifurcation diagram (a) for two coherence-incoherence patterns corresponding to pri-
mary (inset panel with blue dots) and secondary (inset panel with red dots) solution branches
of NEVP (28) with cosine coupling function (4), see Fig. 6. The former solution is stable in blue
region, and the latter is stable in hatched region. Black line denotes fold bifurcation. Red and
blue stability boundaries correspond to the cases when a single real (b) or a pair of complex
conjugate (c) eigenvalues appear from essential spectrum (T-shaped purple line).
their analysis we found stability regions of primary (blue) and secondary (hatched) coherence-
incoherence patterns. As expected (recall Fig. 3(a) and (b)) there is a considerable overlap of the
regions indicating parameter values where multistability is favoured. The boundaries of stability
regions are of two qualitatively different types. Red line corresponds to the case of single real
eigenvalue appearing from essential spectrum, whereas blue lines denote Hopf-like boundaries
where a pair of complex conjugate eigenvalues bifurcates from essential spectrum.
5 Discussion
In present work we formulated a general approach for bifurcation analysis of coherence-incoherence
patterns observed in system (2)–(3). We rely on the assumption that in the large N limit such
patterns are described by standing wave solutions of Eq. (12). The assertion is based on the
continuum limit representation and the Ott-Antonsen invariant manifold reduction available in
the case of sinusoidally coupled phase oscillators. Importantly, each standing wave solution of
Eq. (12) has clear interpretation in terms of local order parameter which demarcates coherent
and incoherent regions of corresponding pattern.
Our approach includes two main steps. First, given a coupling function G we solve NEVP (28),
which plays the role of self-consistency equation for spatial amplitude of unknown standing
wave and its collective frequency. Second, applying Proposition 2 we transform every solution of
NEVP (28) into two standing waves and perform their stability analysis with respect to Eq. (12).
The waves are always neutrally stable due to the location of their essential spectrum, but may
31

be stable or not with respect to the location of their point spectrum. Both spectral components
are determined by time-independent operator L appearing in the right-hand side of linearized
Eq. (66). More precisely, essential spectrum of L is known explicitly, whereas point spectrum
of L can be calculated using approximate algorithm of Section 4.
Already in the first step we obtain a plethora of useful information about standing wave solu-
tions of Eq. (12) from the Fourier series of coupling function G. In particular, we showed that
every non-zero Fourier coefficient gk, k ≥ 1, is responsible for appearance of two qualitatively
different primary branches in the solution set of NEVP (28). The first branch describes so-
called k-twisted solutions of original system (2)–(3), while the second branch includes standing
waves corresponding to (2π/k)-periodic coherence-incoherence patterns. Note that moving
along the latter branch we find coherence-incoherence patterns only beyond a particular point
which we refer to as bending bifurcation point. Another interesting fact is, that sign of gk in-
dicates whether corresponding coherence-incoherence pattern exists for α ∈ (−π/2, π/2),
if gk > 0, or for α ∈ (π/2, 3π/2), if gk < 0. However, the pattern’s stability is still determined
by point spectrum of corresponding linearized operator L.
The primary branch corresponding to the leading Fourier coefficient g0 = 0 describes spatially
uniform solutions of Eq. (12). Its larger part represents the completely coherent state of orig-
inal system (2)–(3). If for some gk, k ≥ 1, we get gk/g0 ∈ (0, 1], then along this primary
branch there exists a bifurcation point giving rise to a secondary branch in the solution set of
NEVP (28). Moving along the secondary branch beyond its bending bifurcation point we ob-
tain another (2π/k)-periodic coherence-incoherence pattern. Particular examples considered
above illustrate a series of qualitative differences between primary and secondary coherence-
incoherence patterns. The solution curve of NEVP (28) corresponding to secondary patterns
looks as a half-loop starting and ending at the primary branch of spatially uniform solutions,
therefore one usually finds on it a fold bifurcation point, see red lines in Figs. 5 and 6. In contrast,
the solution curve of NEVP (28) corresponding to primary patterns simply extends to infinity, see
all lines except red in Figs. 5 and 6. Another difference refers to the fact that |w(x)|-profile of
primary pattern approaches zero for some x ∈ [−π, π], see insert panels in Figs. 5 and 6. As
a result, neighbouring coherent regions turn out to be in anti-phase, see panel with arg a(x) in
Fig. 3(a).
It is a surprising fact that almost all previous studies related to the topic reported only on the
secondary coherence-incoherence pattern corresponding to the first Fourier coefficient g1, see
for example [12, 14, 15, 19, 20]. The primary pattern corresponding to g1 was found much
later in [28]. Therefore we conjecture that most of the patterns described above still wait for
their observation. A good illustration for them could be coherence-incoherence patterns found
numerically by Y. Maistrenko [50] in system (2)–(3) with piecewise-constant coupling function (5)
and α ≈ π (so-called repulsive coupling case). However their classification goes beyond the
scope of this paper.
For effective numerical search of new coherence-incoherence patterns one also has to know
their stability boundaries, e.g. the second step of our bifurcation analysis scheme. In this paper
we calculated such boundaries in the simplest case of cosine coupling only, see Fig. 8. Even
this particular result turns out to be instructive, since we encounter two new types of stability
boundaries, see Fig. 8(b) and (c), which to the best of our knowledge were not reported before.
32

This and other facts clearly demonstrate how far is our knowledge of coherence-incoherence
patterns from being complete. In this context, we believe that our work helps to understand
better this complex phenomenon and shows promising directions for future research.
Acknowledgments
I would like to thank M. Wolfrum, A. Pikovsky and Y. Maistrenko for many helpful discussions.
References
[1] W. Singer. Synchronization of cortical activity and its putative role in information processing
and learning. Ann. Rev. Physiol., 55:349–374, 1993.
[2] C. M. Gray. Synchronous oscillations in neuronal systems: mechanisms and functions. J.
Comput. Neurosci., 1:11–38, 1994.
[3] S. Farmer. Neural rhythms in Parkinson’s disease. Brain, 125:1175–1176, 2002.
[4] A. F. Taylor, M. R. Tinsley, F. Wang, Z. Huang, and K. Showalter. Dynamical quorum sens-
ing and synchronization in large populations of chemical oscillators. Science, 323:614–
617, 2009.
[5] A. F. Taylor, S. Nkomo, and K. Showalter. Chimera and phase-cluster states in populations
of coupled chemical oscillators. Nature Physics, 8:662–665, 2012.
[6] K. Wiesenfeld and J. W. Swift. Averaged equations for Josephson junction series arrays.
Phys. Rev. E, 51:1020–1025, 1995.
[7] H. Haus. Modelocking of lasers. IEEE J. Sel. Top. Quantum Electron., 6:1173–1185, 2000.
[8] A. Pikovsky, M. Rosenblum, and J. Kurths. Synchronization, a universal concept in nonlin-
ear sciences. Cambridge University Press, Cambridge, 2001.
[9] Y. Kuramoto. Chemical Oscillations, Waves, and Turbulence. Springer, Berlin Heidelberg
New York, 1984.
[10] F. C. Hoppensteadt and E. M. Izhikevich. Weakly Connected Neural Networks. Springer,
Berlin Heidelberg New York, 1997.
[11] H. Sakaguchi and Y. Kuramoto. A soluble active rotater model showing phase transitions
via mutual entertainment. Prog. Theor. Phys., 76:576, 1986.
[12] Y. Kuramoto and D. Battogtokh. Coexistence of coherence and incoherence in nonlocally
coupled phase oscillators. Nonlinear Phenom. Complex Syst., 5:380–385, 2002.
33

[13] S. I. Shima and Y. Kuramoto. Rotating spiral waves with phase-randomized core in nonlo-
cally coupled oscillators. Phys. Rev. E, 69:036213, 2004.
[14] D. M. Abrams and S. H. Strogatz. Chimera states for coupled oscillators. Phys. Rev. Lett.,
93:174102, 2004.
[15] D. M. Abrams and S. H. Strogatz. Chimera states in a ring of nonlocally coupled oscillators.
Int. J. Bif. and Chaos, 16:21–37, 2006.
[16] H. Sakaguchi. Instability of synchronized motion in nonlocally coupled neural oscillators.
Phys. Rev. E, 73:031907, 2006.
[17] G. C. Sethia, A. Sen, and F. M. Atay. Clustered chimera states in delay-coupled oscillator
systems. Phys. Rev. Lett., 100:144102, 2008.
[18] O. E. Omel’chenko, Y. L. Maistrenko, and P. A. Tass. Chimera states: the natural link
between coherence and incoherence. Phys. Rev. Lett., 100:044105, 2008.
[19] C. R. Laing. Chimera states in heterogeneous networks. Chaos, 19:013113, 2009.
[20] C. R. Laing. The dynamics of chimera states in heterogeneous Kuramoto networks. Phys-
ica D, 238:1569–1588, 2009.
[21] E. A. Martens, C. R. Laing, and S. H. Strogatz. Solvable model of spiral wave chimeras.
Phys. Rev. Lett., 104:044101, 2010.
[22] G. Bordyugov, A. Pikovsky, and M. Rosenblum. Self-emerging and turbulent chimeras in
oscillator chains. Phys. Rev. E, 82:035205, 2010.
[23] O. E. Omel’chenko, M. Wolfrum, and Y. L. Maistrenko. Chimera states as chaotic spa-
tiotemporal patterns. Phys. Rev. E, 81:065201, 2010.
[24] O. E. Omel’chenko, Y. L. Maistrenko, and P. A. Tass. Chimera states induced by spatially
modulated delayed feedback. Phys. Rev. E, 82:066201, 2010.
[25] M. Wolfrum and O. E. Omel’chenko. Chimera states are chaotic transients. Phys. Rev. E,
84:015201, 2011.
[26] M. Wolfrum, O. E. Omel’chenko, S. Yanchuk, and Y. L. Maistrenko. Spectral properties of
chimera states. Chaos, 21:013112, 2011.
[27] C. R. Laing. Fronts and bumps in spatially extended Kuramoto networks. Physica D,
240:1960–1971, 2011.
[28] Y. Zhu, Y. Li, M. Zhang, and J. Yang. The oscillating two-cluster chimera state in non-locally
coupled phase oscillators. EPL, 97:10009, 2012.
[29] O. E. Omel’chenko, M. Wolfrum, S. Yanchuk, Y. Maistrenko, and O. Sudakov. Stationary
patterns of coherence and incoherence in two-dimensional arrays of non-locally-coupled
phase oscillators. Phys. Rev. E, 85:036210, 2012.
34

[30] M. J. Panaggio and D. M. Abrams. Chimera states on a ﬂat torus. Phys. Rev. Lett.,
110:094102, 2013.
[31] D. M. Abrams, R. Mirollo, S. H. Strogatz, and D. A. Wiley. Solvable model for chimera states
of coupled oscillators. Phys. Rev. Lett., 101:084103, 2008.
[32] E. A. Martens. Bistable chimera attractors on a triangular network of oscillator populations.
Phys. Rev. E, 82:016216, 2010.
[33] I. Omelchenko, Y. Maistrenko, P. Hövel, and E. Schöll. Loss of coherence in dynamical
networks: Spatial chaos and chimera states. Phys. Rev. Lett., 106:234102, 2011.
[34] A. M. Hagerstrom, T. E. Murphy, R. Roy, P. Hövel, I. Omelchenko, and E. Schöll. Experi-
mental observation of chimeras in coupled-map lattices. Nature Physics, 8:658–661, 2012.
[35] D. A. Wiley, S. H. Strogatz, and M. Girvan. The size of the sync basin. Chaos, 16:015103,
2006.
[36] T. Girnyk, M. Hasler, and Y. Maistrenko. Multistability of twisted states in non-locally cou-
pled Kuramoto-type models. Chaos, 22:013114, 2012.
[37] E. Luçon and W. Stannat. Mean ﬁeld limit for disordered diffusions with singular interac-
tions. arXiv:1301.6521, 2013.
[38] J. D. Crawford and K. T. R. Davies. Synchronization of globally coupled phase oscillators:
singularities and scaling for general couplings. Physica D, 125:1–46, 1999.
[39] R. Mirollo and S. H. Strogatz. The spectrum of the partially locked state for the Kuramoto
model. J. Nonlinear Sci., 17:309–347, 2007.
[40] E. Ott and T. M. Antonsen. Low dimensional behavior of large systems of globally coupled
oscillators. Chaos, 18:037113, 2008.
[41] E. Ott and T. M. Antonsen. Long time evolution of phase oscillator systems. Chaos,
19:023117, 2009.
[42] S. Watanabe and S. H. Strogatz. Constants of motion for superconducting Josephson
arrays. Physica D, 74:197–253, 1994.
[43] A. Pikovsky and M. Rosenblum. Partially integrable dynamics of hierarchical populations
of coupled oscillators. Phys. Rev. Lett., 101:264103, 2008.
[44] S. H. Strogatz. From Kuramoto to Crawford: exploring the onset of synchronization in
populations of coupled oscillators. Physica D, 143:1–20, 2000.
[45] M. A. Krasnoselski. Topological Methods in the Theory of Nonlinear Integral Equations.
Macmillan, New Jork, 1965.
[46] P. H. Rabinowitz. Some global results for nonlinear eigenvalue problems. J. Functional
Analysis, 7:487–513, 1971.
35

[47] D. E. Edmunds and W. Evans. Spectral theory and differential operators. Oxford University
Press, Oxford, 1987.
[48] E. A. Biberdorf and M. Vath. On the spectrum of orthomorphisms and Barbashin operators.
Journal for Analysis and its Applications, 18:859–873, 1999.
[49] T. Kato. Perturbation theory for linear operators. Springer, Berlin Heidelberg New York,
1966.
[50] Y. Maistrenko. Chimera states for repulsively coupled phase oscillators. In B. Mehlig,
O. Ghavami, S. Östlund, and D. Hanstorp, editors, XXXII Dynamics Days Europe: Book of
Abstracts, page 35, Gothenburg, Sweden, 2012. University of Gothenburg.
36

Coherence incoherence patterns in a ring of non-locally coupled phase oscillators

More Related Content

What's hot (19)

Similar to Coherence incoherence patterns in a ring of non-locally coupled phase oscillators (20)

Recently uploaded (20)

Coherence incoherence patterns in a ring of non-locally coupled phase oscillators