Phase reduction analysis of traveling breathers in reaction–diffusion systems

Takahiro Arai∗ and Yoji Kawamura

Center for Mathematical Science and Advanced Technology,
Japan Agency for Marine-Earth Science and Technology, Yokohama 236-0001, Japan
(Dated: July 22, 2025)
We formulate a theory for phase reduction analysis of traveling breathers in reaction–diffusion
systems with spatial translational symmetry. In this formulation, the spatial and temporal phases
represent the position and oscillation of a traveling breather, respectively. We perform phase re-
duction analysis on a pair of FitzHugh–Nagumo models exhibiting standing breathers and a pair
of Gray–Scott models exhibiting traveling breathers. The derived phase equations for the spatial
and temporal phases indicate nontrivial spatiotemporal dynamics, where both phases are mutually
coupled. Using the phase equations, we obtain the time evolution of the phase differences, which is
consistent with that obtained from direct numerical simulations.
arXiv:2507.14942v1 [nlin.AO] 20 Jul 2025

I. INTRODUCTION solution, and oscillating solution. Developing a phase re-

duction theory for spatiotemporal patterns is essential for
Synchronization phenomena are usually important in understanding and controlling various experimental sys-
various fields, including physics, chemistry, and life sci- tems, such as the Belousov–Zhabotinsky reaction [46–49].
ences [1–5]. Rhythmic systems are typically described For reaction–diffusion systems, the synchronization
as self-sustained oscillators, where ordinary differential properties of spatiotemporal dynamics can be investi-
equations have stable limit-cycle solutions. In such sys- gated using phase reduction theory, which can be applied
tems, phase reduction theory provides a useful framework to partial differential equations with limit-cycle solutions;
for elucidating the synchronization properties of coupled however, its applicability is limited to specific types of
oscillators [1–12]. On the basis of this framework, the col- spatiotemporal patterns. For example, the theory devel-
lective synchronization of phase oscillators has been ex- oped in Ref. [44] can be applied to circulating pulses, ro-
tensively investigated in various network configurations tating spirals, oscillating spots, and target waves. These
including globally coupled systems, nonlocally coupled patterns typically rely on a single symmetry, such as spa-
systems, and complex networks [1–19]. Furthermore, tial translational, spatial rotational, or temporal trans-
data-driven approaches based on the phase reduction the- lational symmetry. In contrast, patterns involving both
ory have enabled us to uncover synchronization proper- spatial and temporal translational symmetries are out-
ties in various phenomena. For example, such approaches side the scope of the theory.
have been employed in the fields of neurophysiology and Traveling breather and standing breather [50, 51],
biology, for the locomotor coordination of humans and which are typical examples of traveling and oscillating
millipedes [20–23], brain waves [17, 24], the human car- spatiotemporal patterns, cannot be analyzed using the
diorespiratory system [25], and selective attention mech- existing theory for the limit-cycle solution. A traveling
anisms in frog choruses [26]. breather spontaneously breaks spatial reflection symme-
The theory for phase reduction analysis has been gen- try of the system; as a result, its traveling velocity is
eralized to spatiotemporal dynamics. Recent studies nonzero. A standing breather can be regarded as a spe-
have developed phase reduction theory for limit-cycle so- cial case, where spatial reflection symmetry is not broken.
lutions to partial differential equations describing oscil- For these breathers, the formulation of a theory for the
latory thermal convection [27, 28], periodic flows [29– phase reduction analysis describing the mutual influences
34], and beating flagella [35]. Furthermore, considering between translation and oscillation is necessary.
that the traveling solution [36, 37] and oscillating solu- Recent developments in phase reduction theory for os-
tion [38, 39] of the nonlinear Fokker–Planck equations are cillatory thermal convection systems with spatial transla-
within the scope of the phase reduction theory, collective tional symmetry [52, 53] provide a foundation for analyz-
oscillations of globally coupled elements can be analyzed. ing the synchronization properties of breathers. In these
In reaction–diffusion systems, which are particularly im- studies, traveling and oscillating spatiotemporal patterns
portant in this study, theoretical frameworks have also of oscillatory thermal convection were investigated in a
been developed for traveling pulses [40, 41] and rotating cylindrical Hele–Shaw cell and a two-dimensional incom-
spirals [42, 43]. Notably, in Refs. [44, 45], phase equations pressible Navier–Stokes flow system. Both systems ex-
can be derived without assuming symmetry with respect hibit lateral periodicity and are governed by partial dif-
to continuous spatial translation or rotation, making the ferential equations with limit-torus solutions. The the-
theory applicable to each of traveling solution, rotating ory proposed in Refs. [52, 53] are a generalization of the
theory from partial differential equations with limit-cycle
solutions [27, 38, 44] to those with limit-torus solutions.
In this study, we formulate a phase reduction theory for
∗ [email protected] traveling and oscillating solutions in reaction–diffusion
 2

systems with spatial translational symmetry. The formu- framework developed in Refs. [52, 53].
lation of the theory is based on Refs. [52, 53]. Then, we
apply the theory to a pair of FitzHugh–Nagumo (FHN)
models exhibiting standing breathers [51] and a pair of
A. Reaction–diffusion system with spatial
Gray–Scott models exhibiting traveling breathers [50], translational symmetry
and investigate their synchronization properties. The
theory formulated in this study is applicable to both trav-
eling breathers and standing breathers; the latter corre- We consider one-dimensional reaction–diffusion sys-
spond to the case of zero traveling velocity. tems, which are described by the following partial dif-
An overview of the present study is shown in Fig. 1. ferential equation:
We consider reaction–diffusion systems exhibiting trav-
eling and oscillating spatiotemporal patterns, i.e., trav- ∂ ∂2
X(x, t) = F (X(x, t)) + D 2 X, (1)
eling breathers (top panel in Fig. 1). The system satis- ∂t ∂x
fies medium homogeneity and periodic boundary condi-
tions and therefore exhibits spatial translational symme- where the vector field X(x, t) ∈ RN denotes a state of the
try. The coupling function of a coupled system affects medium at point x at time t, F (X(x, t)) ∈ RN denotes
its breather dynamics, leading to variations in the trav- the local reaction dynamics, and D(∂ 2 X/∂x2 ) denotes
eling velocity and oscillation frequency of the breather. the diffusion of X(x, t) over the medium with a diffusion
On the basis of the phase reduction theory formulated in coefficient matrix D ∈ RN ×N . The system size is 2L,
this study, we derive the spatiotemporal phase dynamics i.e., x ∈ [0, 2L). We consider the 2L-periodic boundary
of the system as phase equations (middle panel in Fig. 1). conditions as follows:
The position and oscillation of a breather are represented
by the spatial phase and temporal phase, respectively. X(x + 2L, t) = X(x, t). (2)
The temporal phase is a concept similar to that used in
conventional theory for limit-cycle solutions, while the Because of the homogeneity of the medium in Eq. (1)
spatial phase is an additional concept arising from the and the periodic boundary condition on x, the system
spatial translational symmetry. The phase coupling func- exhibits continuous spatial translational symmetry with
tions in the phase equations describe the effect of cou- respect to x.
pling between a pair of reaction–diffusion systems on the Furthermore, the system exhibits spatial reflection
spatial and temporal phases. Using the phase equations, symmetry, meaning that Eq. (1) is invariant under the
we obtain the theoretical values of spatiotemporal phase transformation x → −x. However, the proposed theory
dynamics (bottom panel in Fig. 1). On the basis of the does not require this reflection symmetry.
phase equations, we derive the dynamics of spatial and
temporal phase differences. The analysis of these dynam-
ics enables us to investigate the time evolution of spatial
B. Limit-torus solutions and their Floquet-type
and temporal phase differences as well as the stability systems
of the synchronized states between breathers. Further-
more, we validate the theoretical values by comparing the
time evolution of the phase differences obtained from the We assume that the reaction–diffusion system (Eq. (1))
phase equations with those obtained from direct numer- has a stable limit-torus solution, representing a traveling
ical simulations. and oscillating spatiotemporal pattern, i.e., a traveling
This paper is organized as follows. In Sec. II, we for- breather. The limit-torus solution is described as follows:
mulate a theory for the phase reduction analysis of a
reaction–diffusion system exhibiting a traveling breather. X(x, t) = X0 (x − Φ(t), Θ(t)), Φ̇(t) = c, Θ̇(t) = ω. (3)
In Sec. III, we analyze the spatiotemporal phase dynam-
ics of a pair of FHN models exhibiting standing breathers The position and oscillation of the spatiotemporal pat-
(Sec. III A) and a pair of Gray–Scott models exhibiting tern are represented by the spatial phase Φ ∈ [0, 2L) and
traveling breathers (Sec. III B). Concluding remarks are temporal phase Θ ∈ [0, 2π), respectively. The traveling
given in Sec. IV. Supplementary information on phase velocity c and oscillation frequency ω are constant. A
reduction analysis is provided in Appendices A–C. breather with c 6= 0 is referred to as a traveling breather,
whereas a breather with c = 0 is referred to as a stand-
ing breather [50]. The standing breather is regarded as a
II. THEORY FOR PHASE REDUCTION
special case of the traveling breather. The limit-torus so-
ANALYSIS OF TRAVELING BREATHERS lution X0 (x − Φ, Θ) is 2L-periodic and 2π-periodic with
respect to (x − Φ) and Θ, respectively, i.e.,
In this section, we formulate a theory for phase re-
X0 (x − Φ + 2L, Θ) = X0 (x − Φ, Θ), (4)
duction analysis of reaction–diffusion systems exhibiting
traveling breathers. The proposed theory is based on the X0 (x − Φ, Θ + 2π) = X0 (x − Φ, Θ). (5)
 3

By substituting Eq. (3) into Eq. (1), we obtain that By partially integrating Eq. (12), the adjoint operator
X0 (x − Φ, Θ) satisfies the following equation: L̂∗ (x − Φ, Θ) is explicitly described as
 
∂ ∂ L̂∗ (x − Φ, Θ)y ∗ (x − Φ, Θ)
−c +ω X0 (x − Φ, Θ)
∂x ∂Θ 
∂ ∂

∂2 = L̂∗ (x − Φ, Θ) − c +ω y ∗ (x − Φ, Θ), (13)
= F (X0 (x − Φ, Θ)) + D 2 X0 . (6) ∂x ∂Θ
∂x
where
Let us consider a small deviation y(x − Φ, Θ, t) ∈ RN
from the limit-torus solution X0 (x − Φ, Θ). A perturbed L̂∗ (x − Φ, Θ)y ∗ (x − Φ, Θ)
solution can be expressed as 2
 
T T ∂
:= J (X0 (x − Φ, Θ)) + D y ∗ (x − Φ, Θ). (14)
X(x, t) = X0 (x − Φ, Θ) + y (x − Φ, Θ, t) . (7) ∂x2

By linearizing of Eq. (1) with respect to y(x − Φ, Θ, t), In the derivation of Eqs. (13) and (14), we assumed that
we obtain the following equation: y ∗ (x − Φ, Θ) satisfies the periodic boundary conditions
with a period 2L in (x − Φ) and 2π in Θ; these serve
∂ as adjoint boundary conditions. As a result, the bilinear
y(x − Φ, Θ, t) = L̂(x − Φ, Θ)y(x − Φ, Θ, t). (8)
∂t concomitant in Eq. (12) vanishes.
The linear operator in Eq. (8) is defined as follows:
L̂(x − Φ, Θ)y (x − Φ, Θ) C. Floquet zero eigenfunctions and their

∂ ∂
 orthonormalization
= L̂(x − Φ, Θ) + c −ω y (x − Φ, Θ) , (9)
∂x ∂Θ
We use the Floquet and adjoint eigenfunctions asso-
where ciated with the two zero eigenvalues of L̂(x − Φ, Θ) and
L̂(x − Φ, Θ)y(x − Φ, Θ) L̂∗ (x−Φ, Θ). According to Eqs. (9) and (13), these eigen-

∂2
 functions satisfy the following equations:
:= J(X0 (x − Φ, Θ)) + D 2 y (x − Φ, Θ) . (10)
∂x L̂(x − Φ, Θ)Us (x − Φ, Θ)
 
The Jacobi matrix of F (X(x, t)) at X(x, t) = X0 (x − ∂ ∂
= L̂(x − Φ, Θ) + c −ω Us (x − Φ, Θ) = 0,
Φ, Θ) is denoted by J(X0 (x−Φ, Θ)) ∈ RN ×N . In Eqs. (9) ∂x ∂Θ
and (10), we omit the explicit t-dependence of y(x − (15)
Φ, Θ, t) and simply denote it as y(x − Φ, Θ). Note that L̂(x − Φ, Θ)Ut (x − Φ, Θ)
the t-dependence of function y(x− Φ, Θ) does not appear  
hereafter since we consider only the eigenvalue problem ∂ ∂
= L̂(x − Φ, Θ) + c −ω Ut (x − Φ, Θ) = 0,
of the linear operator L̂(x − Φ, Θ). As in the limit-torus ∂x ∂Θ
solution X0 (x − Φ, Θ), function y(x − Φ, Θ) satisfies the (16)
2L-periodicity and 2π-periodicity with respect to (x − L̂∗ (x − Φ, Θ)Us∗ (x − Φ, Θ)
Φ) and Θ, respectively. Note that the linear operator  
L̂(x − Φ, Θ) is periodic with respect to both (x − Φ) and ∗ ∂ ∂
= L̂ (x − Φ, Θ) − c +ω Us∗ (x − Φ, Θ) = 0,
Θ. Therefore, Eq. (8) represents a Floquet-type system ∂x ∂Θ
with two zero eigenvalues. In such systems, the two zero (17)
eigenvalues are associated with breaking of spatial and L̂∗ (x − Φ, Θ)Ut∗ (x − Φ, Θ)
temporal translational symmetries.  
We introduce the adjoint operator of the linear oper- ∂ ∂
= L̂∗ (x − Φ, Θ) − c +ω Ut∗ (x − Φ, Θ) = 0,
ator L̂(x − Φ, Θ) by defining the inner product of two ∂x ∂Θ
functions as follows: (18)
Jy ∗ (x − Φ, Θ), y(x − Φ, Θ)K where Up (x − Φ, Θ) ∈ RN and Up∗ (x − Φ, Θ) ∈ RN
Z 2π Z 2L (p ∈ {s, t}) denote the Floquet zero eigenfunction and its
1
:= dΘ dx y ∗ (x − Φ, Θ) · y(x − Φ, Θ). corresponding adjoint zero eigenfunctions, respectively.
2π 0 0 The spatial phase is associated with Us (x − Φ, Θ) and
(11) Us∗ (x − Φ, Θ), whereas the temporal phase is associated
with Ut (x − Φ, Θ) and Ut∗ (x − Φ, Θ). The two Floquet
The definition of the adjoint operator L̂∗ (x − Φ, Θ) is
based on the following relation: zero eigenfunctions can be selected as
r z ∂
y ∗ (x − Φ, Θ), L̂(x − Φ, Θ)y(x − Φ, Θ) Us (x − Φ, Θ) = X0 (x − Φ, Θ), (19)
∂x
r z ∂
= L̂∗ (x − Φ, Θ)y ∗ (x − Φ, Θ), y(x − Φ, Θ) . (12) Ut (x − Φ, Θ) = X0 (x − Φ, Θ). (20)
∂Θ
 4

The selected eigenfunctions can be verified by differenti- details of the adjoint method and the orthonormaliza-
ating Eq. (6) with respect to x and Θ, respectively. For tion method, which is based on Eq. (21), are described
the inner product defined in Eq. (11) and the two Floquet in Appendix A. The adjoint method has also been used
zero eigenfunctions given in Eqs. (19) and (20), the corre- in previous studies for limit-cycle solutions to ordinary
sponding adjoint zero eigenfunctions satisfy the following differential equations [6–12] and partial differential equa-
orthonormalization condition: tions [27, 28, 35, 38, 39, 44, 45]. The adjoint methods
presented in this study and in Refs. [52, 53] can be consid-
JUp∗ (x − Φ, Θ), Uq (x − Φ, Θ)K ered as a generalization of the existing methods, extend-
Z 2π Z 2L
1 ing the framework from ordinary differential equations to
= dΘ dx Up∗ (x − Φ, Θ) · Uq (x − Φ, Θ) = δpq , partial differential equations, and from limit-cycle solu-
2π 0 0
(21) tions to limit-torus solutions.

where p, q ∈ {s, t} and δpq denotes the Kronecker delta.

Additionally, the following condition is satisfied:
Z 2L
∂ D. Traveling breather with weak perturbations
dx Up∗ (x − Φ, Θ) · Uq (x − Φ, Θ)
∂Θ 0
Z 2L (  
∂ We consider a reaction–diffusion system with a weak
= dx Up∗ (x − Φ, Θ) · Uq (x − Φ, Θ)
0 ∂Θ perturbation, which is described as follows:
  )
∂ ∗
+ U (x − Φ, Θ) · Uq (x − Φ, Θ) ∂ ∂2
∂Θ p X(x, t) = F (X(x, t)) + D 2 X + εp(x, t), (26)
( ∂t ∂x
1 2L
Z
= dx Up∗ (x − Φ, Θ)
ω 0 where εp(x, t) ∈ RN , with the intensity ε ≪ 1, represents
a weak perturbation. We assume that the perturbed so-
 
∂
· L̂(x − Φ, Θ) + c Uq (x − Φ, Θ) lution X(x, t) remains in the vicinity of the orbit of the
∂x
  limit-torus solution X0 (x − Φ, Θ). Considering that
∗ ∂
− L̂ (x − Φ, Θ) − c Up∗ (x − Φ, Θ)
∂x
) ∂
X(x, t)
· Uq (x − Φ, Θ) = 0, (22) ∂t
∂ ∂
≃ −Φ̇(t) X0 (x − Φ, Θ) + Θ̇(t) X0 (x − Φ, Θ),
∂x ∂Θ
where Eqs. (15)–(18) are employed in the derivation. (27)
From Eq. (22), the following orthonormalization condi-
tion is satisfied independently for each Θ:
Z 2L the dynamics of the perturbed system (Eq. (26)) can be
dx Up∗ (x − Φ, Θ) · Uq (x − Φ, Θ) = δpq . (23) projected onto the unperturbed limit-torus solution with
0 respect to the spatial and temporal phases as follows:
According to Eqs. (17) and (18), we obtain the follow-
ing adjoint equation: Z 2L
−Φ̇(t) ≃ dx Us∗ (x − Φ, Θ)
∂ ∗ 0
ω U (x − Φ, Θ)
∂Θ  p ∂2
 
 · F (X0 (x − Φ, Θ)) + D 2 X0 + εp(x, t)
∂ ∂x
= − L̂∗ (x − Φ, Θ) − c Up∗ (x − Φ, Θ), (24)
∂x Z 2L
= dx Us∗ (x − Φ, Θ)
for p ∈ {s, t}. By substituting Θ = −ωs, the above 0
equation is transformed into the following equation:
 
∂ ∂X0
· −c X0 (x − Φ, Θ) + ω + εp(x, t)
∂ ∗ ∂x ∂Θ
U (x − Φ, −ωs)
∂s  p
Z 2L
∂
 = dx Us∗ (x − Φ, Θ)
∗
= L̂ (x − Φ, −ωs) − c Up∗ (x − Φ, −ωs). (25) 0
∂x
h i
· −cUs (x − Φ, Θ) + ωUt (x − Φ, Θ) + εp(x, t)
A relaxation method based on the adjoint equation (25), Z 2L
which is also referred to as the adjoint method, is em- =−c+ε dx Us∗ (x − Φ, Θ) · p(x, t), (28)
ployed to obtain the adjoint zero eigenfunctions. The 0
 5

and a sufficiently weak coupling intensity (ε ≪ 1) to ensure

Z 2L
that a solution Xσ (x, t) remains in the vicinity of the or-
Θ̇(t) ≃ dx Ut∗ (x − Φ, Θ) bit of the limit-torus solution X0 (x − Φ, Θ). Under this
0 assumption, the phase equations are obtained as follows:
∂2
 
· F (X0 (x − Φ, Θ)) + D 2 X0 + εp(x, t) Φ̇σ (t) = c + εΓ̃s (Φσ − Φτ , Θσ , Θτ ) , (35)
∂x
Z 2L Θ̇σ (t) = ω + εΓ̃t (Φσ − Φτ , Θσ , Θτ ) , (36)
= dx Ut∗ (x − Φ, Θ)
0 where
 
∂ ∂X0
· −c X0 (x − Φ, Θ) + ω + εp(x, t)
∂x ∂Θ Γ̃s (Φσ − Φτ , Θσ , Θτ )
Z 2L Z 2L
= dx Ut∗ (x − Φ, Θ) = dx Zs (x − Φσ , Θσ )
h0 i 0

· −cUs (x − Φ, Θ) + ωUt (x − Φ, Θ) + εp(x, t) · G(X0 (x − Φσ , Θσ ), X0 (x − Φτ , Θτ )), (37)

Z 2L Γ̃t (Φσ − Φτ , Θσ , Θτ )
=ω + ε dx Ut∗ (x − Φ, Θ) · p(x, t). (29) Z 2L
0 = dx Zt (x − Φσ , Θσ )
0
In the derivation of Eqs. (28) and (29), we approxi- · G(X0 (x − Φσ , Θσ ), X0 (x − Φτ , Θτ )). (38)
mated X(x, t) by the unperturbed limit-torus solution
X0 (x − Φ, Θ) and used Eqs. (6), (19), (20), and (23). By As seen in Eqs. (35)–(38), the phase coupling functions,
rewriting Eqs. (28) and (29) in the following form, we Γ̃s (Φσ − Φτ , Θσ , Θτ ) and Γ̃t (Φσ − Φτ , Θσ , Θτ ), depend on
obtain the phase equations: the spatial phase difference between the systems and the
Z 2L temporal phases of both systems.
Φ̇(t) = c + ε dx Zs (x − Φ, Θ) · p(x, t), (30) Below we introduce the slow phase variables as
0
2L
φσ (t) = Φσ (t) − ct, (39)
Z
Θ̇(t) = ω + ε dx Zt (x − Φ, Θ) · p(x, t). (31)
0 θσ (t) = Θσ (t) − ωt, (40)

The phase sensitivity functions of the spatial and tempo- Then, Eqs. (35) and (36) can be rewritten as follows:
ral phases, which quantify linear response characteristics,
are given as follows: φ̇σ (t) = εΓ̃s (φσ − φτ , ωt + θσ , ωt + θτ ), (41)
Zs (x − Φ, Θ) = −Us∗ (x − Φ, Θ), (32) θ̇σ (t) = εΓ̃t (φσ − φτ , ωt + θσ , ωt + θτ ). (42)
Zt (x − Φ, Θ) = +Ut∗ (x − Φ, Θ). (33)
Applying the averaging method with respect to the tem-
The phase equations, Eqs. (30) and (31), indicate that poral phases, we obtain the following equations:
the spatial and temporal phases are mutually coupled,
resulting in nontrivial spatiotemporal phase dynamics. φ̇σ (t) = εΓs (φσ − φτ , θσ − θτ ), (43)
θ̇σ (t) = εΓt (φσ − φτ , θσ − θτ ), (44)

E. Traveling breathers with weak couplings where the phase coupling functions for the spatial and
temporal phases are given by
In this subsection, we consider weakly coupled
reaction–diffusion systems, which are described by the Γs (φσ − φτ , θσ − θτ )
following equation: 1
Z 2π
= dλ Γ̃s (φσ − φτ , λ + θσ , λ + θτ ), (45)
∂ 2π 0
Xσ (x, t) Γt (φσ − φτ , θσ − θτ )
∂t
∂2
Z 2π
1
= F (Xσ (x, t)) + D Xσ + εG(Xσ (x, t), Xτ (x, t)), = dλ Γ̃t (φσ − φτ , λ + θσ , λ + θτ ). (46)
∂x2 2π 0
(34)
Therefore, we obtain the following phase equations:
where subscripts σ and τ denote indices of the
systems ((σ, τ ) = (1, 2) or (2, 1)), and the term Φ̇σ (t) = c + εΓs (Φσ − Φτ , Θσ − Θτ ), (47)
εG(Xσ (x, t), Xτ (x, t)) ∈ RN represents the coupling be-
tween the two systems at a spatial point x. We assume Θ̇σ (t) = ω + εΓt (Φσ − Φτ , Θσ − Θτ ), (48)
 6

where the phase coupling functions are explicitly de- where Γp (∆Φ, ∆Θ) is simply denoted as Γp for p ∈ {s, t}.
scribed as This equation describes the stability of the fixed point
(∆Φ, ∆Θ) = (0, 0). We consider the following coupling
Γs (Φσ − Φτ , Θσ − Θτ ) function as a typical case:
Z 2π Z 2L
1
= dλ dx Zs (x − Φσ , λ + Θσ ) εG(Xσ , Xτ ) = ε [Xτ (x, t) − Xσ (x, t)] , (60)
2π 0 0
· G(X0 (x − Φσ , λ + Θσ ), X0 (x − Φτ , λ + Θτ )), (49) which implies that each component is linearly coupled to
Γt (Φσ − Φτ , Θσ − Θτ ) its counterpart with equal weight. Under the linear cou-
Z 2π Z 2L pling described in Eq. (60), the linearized equation (59)
1 can be reduced to
= dλ dx Zt (x − Φσ , λ + Θσ )
2π 0 0     
· G(X0 (x − Φσ , λ + Θσ ), X0 (x − Φτ , λ + Θτ )). (50) d ∆Φ 1 0 ∆Φ
= −2ε . (61)
dt ∆Θ 0 1 ∆Θ
The phase coupling functions, Γs (Φσ − Φτ , Θσ − Θτ ) and
Γt (Φσ − Φτ , Θσ − Θτ ), depend only on the spatial and This indicates that the in-phase synchronization with
temporal phase differences. respect to both spatial and temporal phases, i.e.,
We define the spatial and temporal phase differences (∆Φ, ∆Θ) = (0, 0), is linearly stable for systems with
as follows: the coupling described in Eq. (60). The orthonormal-
ization condition (Eq. (21)) is used in deriving Eq. (61).
∆Φ(t) := Φ1 (t) − Φ2 (t), (51) A similar stability analysis of the in-phase synchronized
∆Θ(t) := Θ1 (t) − Θ2 (t). (52) state for limit-cycle solutions can be found in Ref. [45].
Finally, we note that the forms of Eqs. (47) and (48)
From Eqs. (47) and (48), we obtain the following equa- are the same as those of the phase equations derived
tions that describe the time evolution of the spatial and for weakly coupled limit-torus oscillators, which are de-
temporal phase differences: scribed by finite-dimensional dynamical systems [54, 55].
Therefore, as with ordinary limit-torus oscillators, cou-
d pled reaction–diffusion systems with limit-torus solutions
∆Φ(t) = εΓ(a)
s (∆Φ, ∆Θ), (53)
dt can be reduced to a set of phase equations. We also note
d (a) that similar phase equations for oscillatory convection
∆Θ(t) = εΓt (∆Φ, ∆Θ). (54)
dt systems with spatial translational symmetry have been
derived in previous studies [52, 53].
The antisymmetric components of the phase cou-
(a)
pling functions are denoted by Γs (∆Φ, ∆Θ) and
(a)
Γt (∆Φ, ∆Θ). They are given by III. PHASE REDUCTION ANALYSIS OF
TRAVELING BREATHERS
Γ(a)
s (∆Φ, ∆Θ) = Γs (∆Φ, ∆Θ) − Γs (−∆Φ, −∆Θ), (55)
(a)
Γt (∆Φ, ∆Θ) = Γt (∆Φ, ∆Θ) − Γt (−∆Φ, −∆Θ). (56) In this section, we perform phase reduction analysis
of coupled reaction–diffusion systems. Specifically, we
These two functions satisfy the following properties: analyze the FHN models exhibiting standing breathers
(Sec. III A) [51] and the Gray–Scott models exhibit-
Γ(a) (a)
s (−∆Φ, −∆Θ) = −Γs (∆Φ, ∆Θ), (57) ing traveling breathers (Sec. III B) [50]. Additionally,
(a) (a) we perform direct numerical simulations of the coupled
Γt (−∆Φ, −∆Θ) = −Γt (∆Φ, ∆Θ), (58)
reaction–diffusion systems and compare the time evolu-
which represent antisymmetry with respect to the origin tion of the spatial and temporal phase differences ob-
(∆Φ, ∆Θ) = (0, 0). It can be readily seen from Eqs. (53)– tained from the phase equations with those obtained from
(56) that the system has a fixed point at (∆Φ, ∆Θ) = the simulations.
(0, 0). The details of the phase reduction analysis presented in
Notably, we can analyze the stability of the in-phase this section can be found in Appendices A–C as supple-
synchronized state, i.e., (∆Φ, ∆Θ) = (0, 0). By lineariz- mentary information. Appendix A describes the proce-
ing Eqs. (53) and (54) around (∆Φ, ∆Θ) = (0, 0), we dure for calculating the adjoint zero eigenfunctions. Ap-
obtain the following equations: pendix B describes the method for calculating the trav-
eling velocity c 6= 0, which is employed in Sec. III B. Ap-
pendix C describes how the spatial and temporal phases
 
d ∆Φ
dt ∆Θ are determined from the spatiotemporal dynamics ob-
    tained from the direct numerical simulations.
∂∆Φ Γs ∂∆Θ Γs ∆Φ In the calculation of the limit-torus solutions and the
= 2ε , (59)
∂∆Φ Γt ∂∆Θ Γt (∆Φ,∆Θ) ∆Θ solution of the adjoint equation, the system was dis-
=(0,0) cretized using spatial grid sizes of ∆x = 2L/210 and
 7

∆x = 2L/213 in the FHN and Gray–Scott models, re- the governing equations, the latter of which is mentioned
spectively. Time integration was performed using the in Sec. II A.
explicit Euler scheme with time steps of ∆t = TΘ /216 The Floquet zero eigenfunctions Us (x − Φ, Θ) and
and ∆t = TΘ /220 in the FHN and Gray–Scott mod- Ut (x − Φ, Θ) are determined using Eqs. (19) and (20).
els, respectively. The constant TΘ := 2π/ω denotes the Because of the reflection symmetry of the limit-torus so-
period of the temporal phase. The traveling velocity lution, the Floquet zero eigenfunctions exhibit the fol-
and oscillation frequency of the unperturbed solution, lowing reflection properties (see Fig. 3):
X(x, t) = X0 (x − Φ(t), Θ(t)), in the FHN model were
found to be c = 0 and ω ≃ 5.59023 × 10−2 , respectively. Us (−(x − Φ), Θ) = −Us (x − Φ, Θ), (66)
In the Gray–Scott model, the corresponding values were Ut (−(x − Φ), Θ) = +Ut (x − Φ, Θ). (67)
c ≃ 4.31310×10−2 and ω ≃ 2.78994×10−2. These values Consequently, the phase sensitivity functions (adjoint
of c and ω were used to solve the adjoint equation, which zero eigenfunctions) exhibit the following reflection prop-
was integrated over total durations of 100TΘ and 2700TΘ, erties (see Fig. 4):
in the FHN and Gray–Scott models, respectively.
In the direct numerical simulations, the system was Zs (−(x − Φ), Θ) = −Zs (x − Φ, Θ), (68)
discretized using spatial grid sizes of ∆x = 2L/210 and Zt (−(x − Φ), Θ) = +Zt (x − Φ, Θ). (69)
∆x = 2L/212 in the FHN and Gray–Scott models, re-
spectively. Time integration was performed using the The Floquet zero eigenfunctions and phase sensitivity
explicit Euler scheme with time steps of ∆t = 0.002 and functions are localized around the core region of the
∆t = 0.0008 in the FHN and Gray–Scott models, respec- breather. Regarding the uσ component, the correspond-
tively. ing components of the Floquet zero eigenfunctions and
phase sensitivity functions are localized near the edges of
the core region. Although this localization appears only
A. Standing breathers in the FHN models at the edges, we still consider it to be one manifestation
of the localization around the core region.
Because of the spatial reflection properties of the limit-
In this subsection, we examine the FHN models ex- torus solutions and the phase sensitivity functions, the
hibiting standing breathers, described by the following antisymmetric components of the phase coupling func-
equations: tions exhibit additional spatial reflection properties. Us-
ing the phase differences ∆Φ and ∆Θ, the phase coupling
uσ − u3σ − vσ
   
uσ
Xσ (x, t) = , F (Xσ (x, t)) = , functions can be rewritten as
vσ µ(uσ − a1 vσ − a0 )
(62) Γs (∆Φ, ∆Θ)
Z 2π Z 2L
εG(Xσ (x, t), Xτ (x, t)) = εK(Xτ (x, t) − Xσ (x, t)), 1
(63) = dλ dx Zs (x − ∆Φ, λ + ∆Θ)
2π 0 0
· K(X0 (x, λ) − X0 (x − ∆Φ, λ + ∆Θ)), (70)
where uσ := uσ (x, t) and vσ := vσ (x, t), and the co-
efficient matrix K is diagonal. The system has a size Γt (∆Φ, ∆Θ)
Z 2π Z 2L
of 2L = 120, which is sufficiently large compared to 1
that of the breather. The diffusion coefficient is D = = dλ dx Zt (x − ∆Φ, λ + ∆Θ)
2π 0 0
diag(1.0, 2.5). The remaining parameters are µ = 0.03,
a0 = −0.1, and a1 = 2.0. The solution Xσ (x, t) exhibits · K(X0 (x, λ) − X0 (x − ∆Φ, λ + ∆Θ)). (71)
standing breathers. Figure 2 shows the limit-torus solu- From Eqs. (65), (68), and (69), the phase coupling func-
tion X0 (x − Φ, Θ) with Φ = 0. In this study, the spatial tions satisfy the following reflection properties:
region where the values of u0 (x − Φ, Θ) and v0 (x − Φ, Θ)
significantly differ from those in the surrounding region Γs (−∆Φ, ∆Θ) = −Γs (∆Φ, ∆Θ), (72)
is referred to as the core region of the breather. In the Γt (−∆Φ, ∆Θ) = +Γt (∆Φ, ∆Θ). (73)
absence of coupling (ε = 0), the traveling velocity of the
Accordingly, the antisymmetric components of the phase
standing breather is zero, i.e.,
coupling functions satisfy similar reflection properties:
c = 0. (64) Γ(a) (a)
s (−∆Φ, ∆Θ) = −Γs (∆Φ, ∆Θ), (74)
(a) (a)
Furthermore, the limit-torus solution satisfies the follow- Γt (−∆Φ, ∆Θ) = +Γt (∆Φ, ∆Θ). (75)
ing spatial reflection symmetry (see Fig. 2): Furthermore, by combining Eqs. (57), (58), (74), and
(75), we obtain the following additional properties:
X0 (−(x − Φ), Θ) = X0 (x − Φ, Θ). (65)
Γ(a) (a)
s (∆Φ, −∆Θ) = +Γs (∆Φ, ∆Θ), (76)
The zero traveling velocity is expected from the spatial (a) (a)
reflection symmetry of both the limit-torus solution and Γt (∆Φ, −∆Θ) = −Γt (∆Φ, ∆Θ). (77)
 8

(a)
From Eq. (74), the nullclines of Γs (∆Φ, ∆Θ) are lo- (∆Φ/L, ∆Θ/π) space. The convergence to (1, 0) is signif-
cated at ∆Φ = 0 and L. Similarly, from Eq. (77), (a)
icantly slower than that to (0, 1) because |Γs (∆Φ, ∆Θ)|
(a) (a)
the nullclines of Γt (∆Φ, ∆Θ) are located at ∆Θ = 0 and |Γt (∆Φ, ∆Θ)| are small near ∆Φ/L = 1. It is worth
and π. Therefore, the rectangular region defined by noting that Eq. (61) cannot be derived under the condi-
0 ≤ ∆Φ/L ≤ 1 and 0 ≤ ∆Θ/π ≤ 1 is surrounded by tion K = diag(1, 0) with the coupling function given in
four nullclines and contains four fixed points located at Eq. (63). Therefore, it cannot be ensured that the fixed
(∆Φ/L, ∆Θ/π) = (0, 0), (1, 0), (0, 1), and (1, 1). Tak- point (0, 0) is linearly stable. Indeed, Fig. 8(c) indicates
ing Eqs. (57), (58), and (74)–(77) into consideration, the that it is a saddle point.
other three rectangular regions, which are symmetric to In both cases, the antisymmetric components of the
the specified one with respect to the origin and the axes, phase coupling functions exhibit a characteristic struc-
also exhibit similar configurations of nullclines and fixed ture centered at ∆Φ/L = 0 (see Figs. 5 and 7). The
points. width of this characteristic region, measured in terms of
Here, we consider two types of coupling for the FHN ∆Φ, is independent of the system size. This indepen-
model: (i) coupling via both uσ and vσ , with K = dence is justified by the fact that the size of the breather
diag(1, 1); (ii) coupling via uσ alone, with K = diag(1, 0). is independent of the system size. Consequently, the Flo-
Although both types of coupling are examined in this quet zero eigenfunctions and phase sensitivity functions
study, it is more common in FHN models to consider are independent of the system size because they are lo-
coupling only via the membrane potential variable uσ . calized around the core region (see the fifth paragraph of
Case (i). The antisymmetric components of the phase Sec. IV for details).
coupling functions are shown in Fig. 5. They exhibit
antisymmetry with respect to the origin (see Eqs. (57)
B. Traveling breathers in the Gray–Scott models
and (58)). Additionally, they satisfy the properties de-
scribed in Eqs. (74)–(77). Figure 6 presents comparisons
between the time evolution of phase differences obtained In this subsection, we consider the Gray–Scott model
from simulations of the FHN models with coupling inten- exhibiting traveling breathers (c 6= 0), which is described
sity ε and the corresponding theoretical predictions. Fig- by
ures 6(a) and 6(b) show representative trajectories con-  
uσ
 2
uσ vσ − (f + k)uσ

verging to the stable fixed points (∆Φ/L, ∆Θ/π) = (0, 0) Xσ (x, t) = , F (Xσ (x, t)) = ,
vσ −u2σ vσ + f (1 − vσ )
and (1, 1), respectively. Figure 6(c) shows these trajec-
(78)
tories in the (∆Φ/L, ∆Θ/π) space along with nullclines.
The convergence to (0, 0) is observed only for a relatively εG(Xσ (x, t), Xτ (x, t)) = ε(Xτ (x, t) − Xσ (x, t)), (79)
narrow range of the initial phase differences, whereas
where uσ := uσ (x, t) and vσ := vσ (x, t). The system has
the convergence to (1, 1) is observed for a much broader
a size of 2L = 250, which is sufficiently large compared
range. The convergence to (1, 1) is significantly slower
(a) to that of the breather. The diffusion coefficient is D =
than that to (0, 0) because |Γt (∆Φ, ∆Θ)| is small near diag(1.0, 1.9). The other parameters are set to f = 0.018
∆Φ/L = 1. Note that the fixed point (0, 0) must be lin- and k = 0.052. The time series X1 (x, t) in the absence of
early stable because Eq. (61) was derived under the con- coupling (ε = 0) is presented in Fig. 9. We observe that
dition K = diag(1, 1) with the coupling function given in the traveling breather exhibits the spatial and temporal
Eq. (63). phases Φ1 = ct and Θ1 = ωt, respectively. Figure 10
Case (ii). Figure 7 shows the antisymmetric compo- shows the limit-torus solution X0 (x − Φ, Θ) with Φ = 0.
nents of the phase coupling functions. As in case (i), both The spatiotemporal pattern of X0 (x − Φ, Θ) breaks the
the antisymmetry with respect to the origin (see Eqs. (57) spatial reflection symmetry; this can be clearly observed
and (58)) and the properties described in Eqs. (74)–(77) in the spatial asymmetry of the core region (see Fig. 10).
are satisfied. As in Fig. 6, Fig. 8 presents comparisons Thus, the limit-torus solution does not satisfy Eq. (65).
between the theoretical values and the results obtained Figure 11 shows the Floquet zero eigenfunctions Us (x−
from direct numerical simulations. Figures 8(a) and 8(b) Φ, Θ) and Ut (x − Φ, Θ). Figure 12 shows the corre-
show representative trajectories converging to the fixed sponding phase sensitivity functions Zs (x − Φ, Θ) and
points (∆Φ/L, ∆Θ/π) = (0, 1) and (1, 0), respectively, Zt (x − Φ, Θ). As in the FHN model, the Floquet zero
both of which are linearly stable. Figure 8(c) shows eigenfunctions and phase sensitivity functions are local-
these trajectories in the (∆Φ/L, ∆Θ/π) space along with ized around the core region of the breather. Since the size
nullclines. We observe that the convergence to (0, 1) of the breather is independent of the system size, the
and (1, 0) is approximately separated by the nullcline of Floquet zero eigenfunctions and phase sensitivity func-
(a) tions are also independent of the system size. The anti-
Γs (∆Φ, ∆Θ). When the trajectory converges to (0, 1),
∆Φ decreases monotonically, while ∆Θ initially oscillates symmetric components of the phase coupling functions,
(a) (a)
and then converges to the fixed point. The trajectories Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ), are shown in Fig. 13.
converging to (1, 0) approach the fixed point monoton- They exhibit antisymmetry with respect to the origin
ically for both ∆Φ and ∆Θ over a large part of the (see Eqs. (57) and (58)). Since Eq. (65) does not hold,
 9

the properties described in Eqs. (66)–(69) and (72)–(77) models exhibiting standing breathers and Gray–Scott
are not satisfied. Note that the origin must be a linearly models exhibiting traveling breathers (Sec. III). Fig-
stable fixed point because Eq. (61) was derived under the ures 6, 8, and 14 show our main results. We observe
coupling function described in Eq. (79). that the phase reduction analysis successfully captures
As in Figs. 6 and 8, Fig. 14 shows comparisons between the nontrivial spatiotemporal phase dynamics. The the-
the theoretical values and the results obtained from direct oretical values obtained using phase reduction analysis
numerical simulations. Figures 14(a)–14(c) show repre- are consistent with the results obtained from the direct
sentative trajectories converging to the stable fixed point numerical simulations.
(∆Φ/L, ∆Θ/π) = (0, 0). Figures 14(d)–14(f) show these
trajectories in the (∆Φ/L, ∆Θ/π) space along with the For a pair of FHN models with K = diag(1, 1), using
nullclines. These figures show that the spatial and tem- Eq. (61), we found that the fixed point corresponding to
poral in-phase state is globally stable. In addition, its the in-phase synchronized state with respect to both the
linear stability is ensured by Eq. (61). Figures 14(a) and spatial and temporal phases is linearly stable. However,
14(d) show that the temporal and spatial phase differ- this synchronized state is reached only from a relatively
ences evolve monotonically in most parts of the space. narrow range of initial conditions for the phase differ-
Both phase differences exhibit characteristic behavior in ences (Fig. 6). For FHN models with K = diag(1, 0),
the region of |∆Φ/L| . 0.5. They evolve rapidly through- the system converges to one of the two distinct states:
out this region, except near ∆Φ/L ≃ 0, where the veloci- (i) in-phase synchronization with respect to the spatial
ties of the spatial and temporal phase differences decrease phase and anti-phase synchronization with respect to the
significantly. We refer to this region as the characteristic temporal phase, or (ii) synchronization with the opposite
region, following the terminology used in Sec. III A. The configuration (Fig. 8). During convergence to the former,
width of the characteristic region in the antisymmetric the temporal phase difference exhibits transient oscilla-
components of the phase coupling functions, measured tions before settling into a steady state, whereas the con-
in terms of ∆Φ, is independent of the system size. As vergence to the latter proceeds monotonically over most
mentioned in Sec. III A, this independence is justified by of the (∆Φ, ∆Θ) space (Fig. 8(c)). Figures 6(b) and 8(b)
the fact that the size of the breather is independent of the show that convergence to the anti-phase state with re-
system size. Consequently, the Floquet zero eigenfunc- spect to the spatial phase is slow. The underlying reason
tions and phase sensitivity functions are independent of is discussed in the fourth paragraph of this section.
the system size (see the fifth paragraph of Sec. IV for
details). Furthermore, we performed phase reduction analysis
The time evolution of the phase differences at the cen- of a pair of Gray–Scott models. The results showed that
ter of the characteristic region exhibits nontrivial behav- the in-phase synchronized state with respect to both the
ior. For example, Figs. 14(a) and 14(d) show nearly spatial and temporal phases is globally stable. Although
monotonic convergence with a slight curvature near the the models exhibit global stability, the dynamics of the
origin (∆Φ/L, ∆Θ/π) = (0, 0). In contrast, Figs. 14(b) phase differences near ∆Φ/L = 0 exhibit nontrivial be-
and 14(e) display a large swing of ∆Θ during the conver- havior, ranging from a nearly monotonic convergence to
gence to the fixed point. Finally, Figs. 14(c) and 14(f) large transient swings (Figs. 14(d) and 14(e)). Notably,
show a representative trajectory starting near the un- during convergence to the in-phase synchronized state
stable fixed point. Notably, while ∆Θ decreases mono- at (∆Φ/L, ∆Θ/π) = (0, 0), the spatial phase difference
tonically, ∆Φ becomes negative during the convergence, becomes negative, despite starting from a positive value
despite starting from a positive value. The transition of (Fig. 14(f)). This result indicates that during the con-
∆Φ from a positive to a negative value highlights the non- vergence process, the relative positions of the breathers
trivial nature of phase dynamics. Furthermore, the insets can be reversed. We emphasize that the phase reduction
in Figs. 14(c) and 14(f) show that the trajectory inter- analysis successfully captured the time evolution of both
(a) the spatial and temporal phase differences observed in di-
sects the nullcline of Γs (∆Φ, ∆Θ) within the time range
rect numerical simulations, even in the case of traveling
εt < 1. This early-time behavior, where ∆Φ initially in-
breathers.
creases and then decreases, reflects the configurations of
the nullclines. In this study, we investigated standing and traveling
breathers and found that their localized structure sig-
nificantly affects their spatiotemporal phase dynamics.
IV. CONCLUDING REMARKS In particular, for the FHN models, slow convergence
toward the anti-phase synchronization with respect to
In this paper, we formulated a theory for the phase the spatial phase was observed (Figs. 6(b) and 8(b)).
reduction analysis of reaction–diffusion systems exhibit- This is attributed to the localization of the phase sen-
ing traveling and oscillating spatiotemporal patterns, sitivity functions around the core region of the breather
namely, limit-torus solutions to partial differential equa- (Fig. 4). Specifically, when the coupling function is
tions with spatial translational symmetry (Sec. II). G(Xσ (x, t), Xτ (x, t)) = K(Xτ (x, t) − Xσ (x, t)), the an-
Then, we performed phase reduction analysis of the FHN tisymmetric component of the phase coupling function
 10

can be written as spatial reflection symmetry is broken or not, the phase

Z 2π Z 2L reduction analysis can successfully calculate the Floquet
1 zero eigenfunctions and phase sensitivity functions. Con-
Γ(a)
p (∆Φ, ∆Θ) = dλ dx Zp (x, λ)
2π 0 0 sidering its generality, the theory formulated in this study
· K[X0 (x + ∆Φ, λ − ∆Θ) − X0 (x − ∆Φ, λ + ∆Θ)], can be used in a wide range of applications, including sys-
(80) tems with different model parameters and various phys-
ical settings. For example, the theory formulated in this
where p ∈ {s, t}. For ∆Φ/L ≃ 1, the regions where study can be extended to reaction–diffusion systems with
|Zp (x, λ)| is localized do not overlap with those where the chemotaxis [56], as well as to classic models such as the
difference |X0 (x + ∆Φ, λ − ∆Θ) − X0 (x − ∆Φ, λ + ∆Θ)| Oregonator [57–59]. Analyzing a wide range of models
is large, resulting in a small magnitude of the integrand. can provide insights into the effect of the spatial distance
Consequently, the magnitudes of the antisymmetric com- between spatiotemporal patterns on their synchroniza-
(a) tion.
ponents of the phase coupling functions, |Γs (∆Φ, ∆Θ)| The importance of phase reduction theory for travel-
(a)
and |Γt (∆Φ, ∆Θ)|, decrease when ∆Φ/L ≃ 1 (Figs. 5 ing and oscillating spatiotemporal patterns is not limited
and 7). A similar tendency was observed in the case of to reaction–diffusion systems. Phase reduction frame-
the coupled Gray–Scott models, where K = diag(1, 1) works for limit-torus solutions have been developed for
was employed (for simplicity, K was omitted in Eqs. (78) the analysis of fluid dynamics [52, 53]. Building upon
and (79)). As shown in Fig. 12, the phase sensitivity these frameworks, future studies are expected to explore
functions are spatially localized around the core region their application to rotating and oscillating convection
of the breather, which leads to a decrease in the mag- (i.e., amplitude vacillation [60–62]) in systems of rotating
nitudes of the antisymmetric components of the phase fluid annuli [63–66]. These systems, which have been ex-
coupling functions near ∆Φ/L = 1 (Fig. 13). perimentally investigated, can be regarded as analogs to
Another important finding is that the Floquet zero atmospheric circulation systems. Therefore, phase reduc-
eigenfunctions and phase sensitivity functions are local- tion analysis of the traveling and oscillating spatiotempo-
ized around the core region, reflecting the localized struc- ral patterns that emerge in rotating fluid annuli can pro-
tures of both traveling and standing breathers. Because vide novel insights into meteorological phenomena, such
the size of the breather is independent of the system as synchronized atmospheric blocking events between the
size, the Floquet zero eigenfunctions and phase sensi- Northern and Southern hemispheres [67].
tivity functions are also independent of the system size. The development of the theory for phase reduc-
On the basis of Eq. (80) and the localized nature of the tion analysis is essential for understanding the synchro-
breather and phase sensitivity functions, the width of nization properties inherent in the governing equations
the characteristic region in the antisymmetric compo- and contributes to the development of data-driven ap-
nents of the phase coupling functions does not depend proaches. Data-driven approaches based on the theory
on the system size. In this study, the region centered for phase reduction analysis can reveal the synchroniza-
(a) (a)
at ∆Φ/L = 0, where Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ) tion properties of various phenomena without requiring
exhibit distinct structures, is referred to as the charac- detailed assumptions or prior knowledge of the underly-
teristic region. As the system size increases, only the re- ing governing equations [3, 17, 20, 22, 24–26, 68–70]. Al-
gion outside the characteristic region expands, whereas though several data-driven methods for identifying the
the width of the characteristic region remains unchanged. spatiotemporal phase dynamics of perturbed limit-cycle
The magnitudes of the antisymmetric components of the solutions have been recently proposed [71–73], no such
phase coupling functions are small outside this region. It method has yet been proposed for limit-torus solutions.
is worth noting that the structure of the characteristic Our study, along with Refs. [52, 53], provides a foun-
region, which reveals nontrivial behavior during conver- dation for the further development of data-driven ap-
gence to the in-phase synchronized state with respect to proaches to investigate the spatiotemporal phase dynam-
the spatial phase, reflects the detailed structure of the ics of traveling and oscillating patterns. The method de-
limit-torus solutions and the corresponding phase sensi- scribed in Appendix C, which successfully determines the
tivity functions. spatial and temporal phases from the spatiotemporal dy-
Although previous studies [52, 53] performed phase re- namics obtained from direct numerical simulations, can
duction analysis only on systems without spatial reflec- also underpin such developments.
tion symmetry breaking, the present study demonstrates
that the same framework is applicable to systems both
with and without such symmetry breaking. The limit- ACKNOWLEDGMENTS
torus solution of a traveling breather breaks spatial re-
flection symmetry (Fig. 10). Accordingly, the Floquet This work was supported by JSPS KAKENHI Grant
zero eigenfunctions and the phase sensitivity functions Numbers JP24K23908, JP24K06910, JP25K01160. Nu-
exhibit spatial asymmetry that reflects the structure of merical simulations were conducted using Earth Simula-
the breather (Figs. 11 and 12). Regardless of whether tor at JAMSTEC.
 11

Appendix A: CALCULATION OF THE ADJOINT ZERO EIGENFUNCTIONS

In this appendix, we describe the procedure for calculating the adjoint zero eigenfunctions. In the phase reduction
analysis of the FHN model and Gray–Scott model (see Sec. III), we perform the following steps:
1. Calculate frequency ω using the following time series:
Z 2L
X(Θ(t)) = dx X(x, t)
0
Z 2L
= dx X0 (x − Φ(t), Θ(t)), (A1)
0

where X(x, t) = X0 (x − Φ(t), Θ(t)) is obtained by solving Eq. (1). The time series X(Θ(t)) depends only on
the temporal phase Θ(t) since the integration over the spatial domain of size 2L eliminates the dependence on
the spatial phase Φ(t).
2. If c 6= 0, determine the traveling velocity c using the time series X(x, t) = X0 (x − Φ(t), Θ(t)), which is obtained
by solving Eq. (1) (see Appendix B).
3. If c 6= 0, update ω using the time series X(Θ(t)), which is obtained by solving the following equation:
∂ ∂2 ∂
X(χ, t) = F (X(χ, t)) + D 2 X + c X. (A2)
∂t ∂χ ∂χ
The above equation is obtained by introducing the comoving coordinate χ := x − ct into Eq. (1). When the
traveling velocity is zero (c = 0), Eqs. (1) and (A2) coincide. This step is performed to correct for the slight
discrepancy in ω that arises from numerical errors in the time integration of Eqs. (1) and (A2).
4. Compute the limit-torus solution X0 (x − Φ, Θ) with Φ = 0 and Θ ∈ [0, 2π) by solving Eq. (A2) over one period
of the temporal phase, TΘ := 2π/ω. The state corresponding to Φ = 0 satisfies the condition described in
Eq. (C5) for c = 0 and the conditions described in Eqs. (C9) and (C10) for c 6= 0. (See Appendix C for the
details of these equations.)
5. Set the initial condition of Ũp∗ (x − Φ, 2π) with Φ = 0 to the corresponding Floquet zero eigenfunction Up (x −
Φ, 2π) with Φ = 0 (p ∈ {s, t}). Here, Ũp∗ (x − Φ, 2π) denotes the solution to the adjoint equation (25), which is
obtained by time integration (see step 6-A).
6. Iterate steps 6-A–6-C described below until the solution to the adjoint equation (25) converges. Throughout
these steps, we set Φ = 0.
6-A. Update Ũs∗ (x − Φ, Θ) and Ũt∗ (x − Φ, Θ) by solving Eq. (25) over TΘ .
6-B. Perform orthonormalization to obtain Us∗ (x − Φ, Θ) and Ut∗ (x − Φ, Θ) (see Eqs. (A3) and (A4)).
6-C. Replace Ũs∗ (x − Φ, Θ) and Ũt∗ (x − Φ, Θ) with Us∗ (x − Φ, Θ) and Ut∗ (x − Φ, Θ), respectively.
Any linear combination of the two adjoint zero eigenfunctions, each associated with a zero eigenvalue and satisfying
Eq. (25), still constitutes a solution to the same equation. Therefore, orthonormalization between the Floquet zero
eigenfunctions and the corresponding adjoint zero eigenfunctions is required to uniquely determine Us∗ (x − Φ, Θ) and
Ut∗ (x − Φ, Θ). In step 6-B, we calculate functions Us∗ (x − Φ, Θ) and Ut∗ (x − Φ, Θ) as follows:


Us∗ (x − Φ, Θ) = c Ũs∗ (x − Φ, Θ) − aŨt∗ (x − Φ, Θ) , (A3)
∗ ∗ ∗


Ut (x − Φ, Θ) = d Ũt (x − Φ, Θ) − bŨs (x − Φ, Θ) . (A4)
To perform orthonormalization, these functions are constructed so as to satisfy the following condition:
q ∗ y
Up (x − Φ, Θ), Uq (x − Φ, Θ) = δpq (p, q ∈ {s, t}), (A5)
where δpq denotes the Kronecker delta. The condition described in Eq. (A5) is satisfied by selecting a, b, c, and d as
follows:
σst σts
a= , b= , (A6)
σtt σss
σtt σss
c= , d= . (A7)
σss σtt − σst σts σss σtt − σst σts
 12

The inner product σpq is defined as:

r z
σpq := Ũp∗ (x − Φ, Θ), Uq (x − Φ, Θ) . (A8)

Appendix B: CALCULATION OF THE series Xσ (x, t), which is obtained from the direct numer-
TRAVELING VELOCITY ical simulations performed in Secs. III A and III B. For
convenience, we omit subscript σ and denote Xσ (x, t),
In this appendix, we describe the method for cal- Φσ (t), and Θσ (t) simply as X(x, t), Φ(t), and Θ(t), re-
culating the nonzero traveling velocity c of a traveling spectively.
breather. We assume that the system size is sufficiently Note that we distinguish between the perturbed solu-
large and that the period of Θ is shorter than that of Φ, tion of Eq. (34), denoted by X(x, t), and the unperturbed
i.e., 2π/ω < 2L/c. We consider the unperturbed solu- solution of Eq. (A2), denoted by X0 (x − Φ, Θ). We as-
tion X(x, t) = X0 (x − Φ(t), Θ(t)), which is obtained by sume that the system size is sufficiently large compared
solving Eq. (1). For notational simplicity, we consider to the size of the standing or traveling breather. For no-
the case of two components (N = 2), as in the FHN and tational simplicity, we consider two components (N = 2)
Gray–Scott models. That is, similar to the case of the FHN and Gray–Scott models.
  That is,
u0 (x − Φ, Θ)
X0 (x − Φ, Θ) = . (B1)
v0 (x − Φ, Θ)
 
u(x, t)
X(x, t) = , (C1)
v(x, t)
Let us define the following periodic function:  
u0 (x − Φ, Θ)
Z 2L h xi X0 (x − Φ, Θ) = . (C2)
A (Φ, Θ) := dx u0 (x − Φ, Θ) exp iπ ∈ C. (B2) v0 (x − Φ, Θ)
0 L
This assumption is made only for notational simplicity,
From the definition of A(Φ, Θ), we obtain the following and the method is applicable to systems with any number
expression: of components (i.e., any N ). Furthermore, we assume
π that the coupling intensity is sufficiently weak (ε ≪ 1)
arg A(Φ, Θ) = (Φ + B(Θ)) , (B3)
L so that a solution X(x, t) remains in the vicinity of the
orbit of the limit-torus solution X0 (x − Φ, Θ).
where B(Θ) ∈ R is a 2π-periodic function. Note that
Initially, we calculate the following time series:
it is not necessary to compute the waveform of B(Θ) in
the evaluation of the traveling velocity. By substituting Z 2L
tn−1 = (n − 1)TΘ and tn = nTΘ into Eq. (B3) and taking X(t) = dx X(x, t). (C3)
the difference between the two resulting expressions, we 0
obtain the following identity, which holds for any integer
We define the 2π-periodic function based on the unper-
n:
turbed solution
L
[arg A(Φ(tn ), Θ(tn )) − arg A(Φ(tn−1 ), Θ(tn−1 ))] = cTΘ , Z 2L
π X0 (Θ) = dx X0 (x − Φ, Θ). (C4)
(B4) 0

where Φ(tn ) = ct and Θ(tn ) = 2πn. Therefore, the This function depends only on the temporal phase Θ(t)
traveling velocity c can be evaluated using the values of because the integration over 2L eliminates the depen-
A(Φ(t), Θ(t)) at t = tn and tn−1 . These values are calcu- dence on Φ(t). To determine the temporal phase Θ(t)
lated using the time series X(x, t) = X0 (x − Φ(t), Θ(t)), corresponding to X(x, t), we identify the value of Θ that
which is obtained by solving Eq. (1). Although in this minimizes the distance between X0 (Θ) and X(t). Fig-
study, we use the variable u0 (x − Φ, Θ) to compute ures 15(a) and 15(b) show the waveforms of X0 (Θ) of
A(Φ, Θ) (see Eq. (B2)), other variables can also be used. the FHN model and Gray–Scott model, respectively.
In the following, we describe methods for calculating
the spatial phase in both the c = 0 and c 6= 0 cases. For
Appendix C: CALCULATION OF THE SPATIAL
AND TEMPORAL PHASES FROM TIME SERIES
c = 0, we consider the limit-torus solution X0 (x − Φ, Θ)
OBTAINED FROM DIRECT NUMERICAL with Φ = 0, which satisfies
SIMULATIONS Z 2L h xi
H(Θ) := dx u0 (x − Φ, Θ) exp iπ ∈ R+ .
In this appendix, we describe the method for calcu- 0 L Φ=0

lating the spatial and temporal phases using the time (C5)
 13

Furthermore, we consider the function A(Φ, Θ), which is Eq. (B3), which satisfies the following condition:
defined in Eq. (B2). For convenience, we restate it below:
2π
1
Z
dΘ B(Θ) = 0. (C9)
Z 2L h xi 2π 0
A (Φ, Θ) := dx u0 (x − Φ, Θ) exp iπ ∈ C.
0 L The above equation means that the constant term in
B(Θ) is zero. By setting Φ = 0 in Eq. (B3), the peri-
Note that the expression for A(Φ, Θ) is identical to that odic function B(Θ) is determined as follows:
of H(Θ), except that the spatial phase Φ is allowed to
take arbitrary values. From the definitions of H(Θ) and L
B(Θ) = arg A(0, Θ). (C10)
A(Φ, Θ), we obtain the following relation: π
  The waveform of B(Θ), computed using the limit-torus
Φ solution of the Gray–Scott model, is shown in Fig. 15(c).
A(Φ, Θ) = H(Θ) exp iπ . (C6)
L Given Eq. (C10), the condition described in Eq. (C9) en-
sures that the state corresponding to Φ = 0 in the limit-
We approximate A(Φ, Θ) as torus solution X0 (x − Φ, Θ) is uniquely determined. To
obtain the time series of the spatial phase, we approx-
Z 2L h xi imate A(Φ, Θ) in Eq. (B3) by Ã(t). Thus, the spatial
Ã(t) := dx u(x, t) exp iπ ∈ C. (C7) phase Φ(t) corresponding to X(x, t) is determined by
0 L
L
Φ(t) = arg Ã(t) − B(Θ(t)). (C11)
Thus, the spatial phase is determined as follows: π
It can be readily seen that Eq. (C8) is a special case
L of Eq. (C11); when c = 0, B(Θ) is zero because of the
Φ(t) = arg Ã(t). (C8)
π spatial reflection symmetry (see Eq. (65) and Fig. 2).
Although we use the variables u(x, t) and u0 (x − Φ, Θ) to
Next, we describe the method for calculating the spa- compute Ã(t) and A(Φ, Θ), respectively, other variables
tial phase for c 6= 0. We consider function B(Θ) in can also be used.

[1] Y. Kuramoto, Chemical Oscillations, Waves, and Turbu- [12] E. M. Izhikevich, Dynamical Systems in Neuroscience:
lence (Springer, New York, 1984). The Geometry of Excitability and Bursting (MIT Press,
[2] A. T. Winfree, The Geometry of Biological Time Cambridge, 2007).
(Springer, New York, 1980). [13] J. A. Acebrón, L. L. Bonilla, C. J. Pérez Vicente, F. Ri-
[3] A. Pikovsky, M. Rosenblum, and J. Kurths, Synchroniza- tort, and R. Spigler, The Kuramoto model: A simple
tion: A Universal Concept in Nonlinear Sciences (Cam- paradigm for synchronization phenomena, Rev. Mod.
bridge University Press, Cambridge, 2001). Phys. 77, 137 (2005).
[4] S. C. Manrubia, A. S. Mikhailov, and D. H. Zanette, [14] A. Arenas, A. Dı́az-Guilera, J. Kurths, Y. Moreno, and
Emergence of Dynamical Order: Synchronization Phe- C. Zhou, Synchronization in complex networks, Phys.
nomena in Complex Systems (World Scientific, Singa- Rep. 469, 93 (2008).
pore, 2004). [15] P. Ashwin, S. Coombes, and R. Nicks, Mathematical
[5] G. V. Osipov, J. Kurths, and C. Zhou, Synchronization frameworks for oscillatory network dynamics in neuro-
in Oscillatory Networks (Springer, New York, 2007). science, J. Math. Neurosci. 6, 2 (2016).
[6] E. Brown, J. Moehlis, and P. Holmes, On the phase re- [16] A. Pikovsky and M. Rosenblum, Dynamics of globally
duction and response dynamics of neural oscillator pop- coupled oscillators: Progress and perspectives, Chaos 25,
ulations, Neural Comput. 16, 673 (2004). 097616 (2015).
[7] B. Ermentrout, Type I membranes, phase resetting [17] T. Stankovski, T. Pereira, P. V. E. McClintock, and
curves, and synchrony, Neural Comput. 8, 979 (1996). A. Stefanovska, Coupling functions: Universal insights
[8] H. Nakao, Phase reduction approach to synchronization into dynamical interaction mechanisms, Rev. Mod. Phys.
of nonlinear oscillators, Contemp. Phys. 57, 188 (2016). 89, 045001 (2017).
[9] B. Pietras and A. Daffertshofer, Network dynamics of [18] S. H. Strogatz, From Kuramoto to Crawford: Exploring
coupled oscillators and phase reduction techniques, Phys. the onset of synchronization in populations of coupled
Rep. 819, 1 (2019). oscillators, Physica D 143, 1 (2000).
[10] B. Ermentrout and D. H. Terman, Mathematical Foun- [19] F. A. Rodrigues, T. K. DM. Peron, P. Ji, and J. Kurths,
dations of Neuroscience (Springer, New York, 2010). The Kuramoto model in complex networks, Phys. Rep.
[11] F. C. Hoppensteadt and E. M. Izhikevich, Weakly Con- 610, 1 (2016).
nected Neural Networks (Springer, New York, 1997). [20] T. Arai, K. Ota, T. Funato, K. Tsuchiya, T. Aoyagi, and
 14

S. Aoi, Interlimb coordination is not strictly controlled [40] J. Löber and H. Engel, Controlling the position of travel-
during walking, Commun. Biol. 7, 1152 (2024). ing waves in reaction–diffusion systems, Phys. Rev. Lett.
[21] H. Furukawa, T. Arai, T. Funato, S. Aoi, and T. Aoy- 112, 148305 (2014).
agi, Bayesian estimation of trunk-leg coordination dur- [41] J. Löber, M. Bär, and H. Engel, Front propagation in
ing walking using phase oscillator models, Neurosci. Res. one-dimensional spatially periodic bistable media, Phys.
215, 47 (2025). Rev. E 86, 066210 (2012).
[22] T. Funato, Y. Yamamoto, S. Aoi, T. Imai, T. Aoy- [42] V. N. Biktashev, D. Barkley, and I. V. Biktasheva, Or-
agi, N. Tomita, and K. Tsuchiya, Evaluation of the bital motion of spiral waves in excitable media, Phys.
phase-dependent rhythm control of human walking using Rev. Lett. 104, 058302 (2010).
phase response curves, PLoS Comput. Biol. 12, e1004950 [43] B. Sandstede, A. Scheel, and C. Wulff, Dynamics of spi-
(2016). ral waves on unbounded domains using center-manifold
[23] M. Yoshikawa and I. Aihara, Wavy synchronization in reductions, J. Differ. Equ. 141, 122 (1997).
locomotion of train millipedes, arXiv:2504.04173. [44] H. Nakao, T. Yanagita, and Y. Kawamura, Phase-
[24] T. Onojima, T. Goto, H. Mizuhara, and T. Aoyagi, A reduction approach to synchronization of spatiotemporal
dynamical systems approach for estimating phase inter- rhythms in reaction–diffusion systems, Phys. Rev. X 4,
actions between rhythms of different frequencies from 021032 (2014).
experimental data, PLoS Comput. Biol. 14, e1005928 [45] Y. Kawamura, S. Shirasaka, T. Yanagita, and H. Nakao,
(2018). Optimizing mutual synchronization of rhythmic spa-
[25] B. Kralemann, M. Frühwirth, A. Pikovsky, M. Rosen- tiotemporal patterns in reaction–diffusion systems, Phys.
blum, T. Kenner, J. Schaefer, and M. Moser, In vivo car- Rev. E 96, 012224 (2017).
diac phase response curve elucidates human respiratory [46] M. Hildebrand, J. Cui, E. Mihaliuk, J. Wang, and
heart rate variability, Nat. Commun. 4, 2418 (2013). K. Showalter, Synchronization of spatiotemporal pat-
[26] K. Ota, I. Aihara, and T. Aoyagi, Interaction mechanisms terns in locally coupled excitable media, Phys. Rev. E
quantified from dynamical features of frog choruses, R. 68, 026205 (2003).
Soc. Open Sci. 7, 191693 (2020). [47] P. Khan and S. Dutta, Complete synchronization of two
[27] Y. Kawamura and H. Nakao, Collective phase description spirals by a messenger wave in a reaction diffusion sys-
of oscillatory convection, Chaos 23, 043129 (2013). tem, Chaos 34, 091103 (2024).
[28] Y. Kawamura and H. Nakao, Noise-induced synchroniza- [48] H. Kalita, P. Khan, and S. Dutta, Rotational synchro-
tion of oscillatory convection and its optimization, Phys. nization of pinned spiral waves, Phys. Rev. E 106, 034201
Rev. E 89, 012912 (2014). (2022).
[29] V. Godavarthi, Y. Kawamura, and K. Taira, Optimal [49] P. Khan and S. Dutta, Synchronization of corotating spi-
waveform for fast synchronization of airfoil wakes, J. rals anchored to heterogeneities in a chemical system,
Fluid Mech. 976, R1 (2023). Eur. Phys. J. Spec. Top. 233, 787 (2024).
[30] Y. Kawamura, V. Godavarthi, and K. Taira, Adjoint- [50] M. Yadome, K.-I. Ueda, and M. Nagayama, Chaotic mo-
based phase reduction analysis of incompressible periodic tion of propagating pulses in the Gray-Scott model, Phys.
flows, Phys. Rev. Fluids 7, 104401 (2022). Rev. E 83, 056207 (2011).
[31] K. Taira and H. Nakao, Phase-response analysis of syn- [51] A. Hagberg and E. Meron, Pattern formation in non-
chronization for periodic flows, J. Fluid Mech. 846, R2 gradient reaction–diffusion systems: the effects of front
(2018). bifurcations, Nonlinearity 7, 805 (1994).
[32] M. Iima, Optimal external forces of the lock-in phenom- [52] Y. Kawamura and H. Nakao, Phase description of os-
ena for flow past an inclined plate in uniform flow, Phys. cillatory convection with a spatially translational mode,
Rev. E 109, 045102 (2024). Physica D 295-296, 11 (2015).
[33] M. Iima, Phase reduction technique on a target region, [53] Y. Kawamura, Phase reduction of limit-torus solutions to
Phys. Rev. E 103, 053303 (2021). partial differential algebraic equations, Phys. Rev. Res.
[34] M. Iima, Jacobian-free algorithm to calculate the phase 1, 033130 (2019).
sensitivity function in the phase reduction theory and its [54] E. M. Izhikevich, Weakly connected quasi-periodic oscil-
applications to Kármán’s vortex street, Phys. Rev. E 99, lators, FM interactions, and multiplexing in the brain,
062203 (2019). SIAM J. Appl. Math. 59, 2193 (1999).
[35] Y. Kawamura and R. Tsubaki, Phase reduction approach [55] A. Demir, C. Gu, and J. Roychowdhury, Phase equations
to elastohydrodynamic synchronization of beating flag- for quasi-periodic oscillators, IEEE/ACM International
ella, Phys. Rev. E 97, 022212 (2018). Conference on Computer-Aided Design (ICCAD) (2010)
[36] Y. Kawamura, H. Nakao, K. Arai, H. Kori, and Y. Ku- pp. 292–297.
ramoto, Collective phase sensitivity, Phys. Rev. Lett. [56] S. Kawaguchi, Pulse dynamics in a bistable reaction–
101, 024101 (2008). diffusion system with chemotaxis, Adv. Math. Phys.
[37] Y. Kawamura, H. Nakao, K. Arai, H. Kori, and 2022, 1637071 (2022).
Y. Kuramoto, Phase synchronization between collective [57] J. D. Dockery and R. J. Field, Numerical evidence of
rhythms of globally coupled oscillator groups: Noisy stationary and breathing concentration patterns in the
identical case, Chaos 20, 043109 (2010). Oregonator with equal diffusivities, Phys. Rev. E 58, 823
[38] Y. Kawamura, H. Nakao, and Y. Kuramoto, Collective (1998).
phase description of globally coupled excitable elements, [58] I. Berenstein and C. Beta, Cross-diffusion in the two-
Phys. Rev. E 84, 046211 (2011). variable Oregonator model, Chaos 23, 033119 (2013).
[39] Y. Kawamura, Collective phase reduction of globally cou- [59] E. M. Nicola, M. Bär, and H. Engel, Wave instability in-
pled noisy dynamical elements, Phys. Rev. E 95, 032225 duced by nonlocal spatial coupling in a model of the light-
(2017). sensitive Belousov-Zhabotinsky reaction, Phys. Rev. E
 15

73, 066225 (2006). 35, 043116 (2025).

[60] M. Lappa, Rotating thermal flows in natural and indus- [67] G. S. Duane, P. J. Webster, and J. B. Weiss, Co-
trial processes (Wiley, Chichester, 2012). occurrence of northern and southern hemisphere blocks
[61] M. Ghil, P. Read, and L. Smith, Geophysical flows as as partially synchronized chaos, J. Atmos. Sci. 56, 4183
dynamical systems: the influence of Hide’s experiments, (1999).
Astron. Geophys. 51, 4.28 (2010). [68] K. Ota and T. Aoyagi, Direct extraction of phase dynam-
[62] P. L. Read, M. J. Bel, D. W. Johnson, and R. M. Small, ics from fluctuating rhythmic data based on a Bayesian
Quasi-periodic and chaotic flow regimes in a thermally approach, arXiv:1405.4126.
driven, rotating fluid annulus, J. Fluid Mech. 238, 599 [69] T. Stankovski, A. Duggento, P. V. E. McClintock, and
(1992). A. Stefanovska, Inference of time-evolving coupled dy-
[63] P. L. Read, X. Morice-Atkinson, E. J. Allen, and namical systems in the presence of noise, Phys. Rev. Lett.
A. A. Castrejón-Pita, Phase synchronization of baroclinic 109, 024101 (2012).
waves in a differentially heated rotating annulus exper- [70] T. Arai, Y. Kawamura, and T. Aoyagi, Extracting phase
iment subject to periodic forcing with a variable duty coupling functions between collectively oscillating net-
cycle, Chaos 27, 127001 (2017). works from time-series data, J. Phys. Soc. Jpn. 91,
[64] A. A. Castrejón-Pita and P. L. Read, Synchronization 124001 (2022).
in a pair of thermally coupled rotating baroclinic annuli: [71] T. Arai, Y. Kawamura, and T. Aoyagi, Setting of the
Understanding atmospheric teleconnections in the labo- Poincaré section for accurately calculating the phase of
ratory, Phys. Rev. Lett. 104, 204501 (2010). rhythmic spatiotemporal dynamics, Phys. Rev. E 111,
[65] F. J. R. Eccles, P. L. Read, A. A. Castrejón-Pita, and 014205 (2025).
T. W. N. Haine, Synchronization of modulated traveling [72] K. Fukami, H. Nakao, and K. Taira, Data-driven tran-
baroclinic waves in a periodically forced, rotating fluid sient lift attenuation for extreme vortex gust–airfoil in-
annulus, Phys. Rev. E 79, 015202 (2009). teractions, J. Fluid Mech. 992, A17 (2024).
[66] I. Oshima and Y. Kawamura, Synchronization phe- [73] K. Yawata, R. Sakuma, K. Fukami, K. Taira, and
nomenon of temperature oscillation in rotating fluid an- H. Nakao, Phase autoencoder for rapid data-driven
nulus and optimal waveforms of external forcing, Chaos synchronization of rhythmic spatiotemporal patterns,
arXiv:2506.12777.
 16

FIG. 1. Overview of this study. (top) A pair of reaction–diffusion systems exhibiting traveling breathers. The systems satisfy
the homogeneity of the medium and periodic boundary conditions. The coupling function between the systems affects the
dynamics of the breather, leading to variations in its traveling velocity and oscillation frequency. (middle) Phase equations
representing the spatiotemporal phase dynamics of the traveling breathers. The limit-torus solution is represented by a spatial
phase and a temporal phase, corresponding to the position and oscillation of the breather, respectively. The phase equations
for both phases are derived from the governing equations of the reaction–diffusion systems using phase reduction analysis.
The phase coupling functions in the phase equations describe the effect of coupling between the systems on the phases.
(bottom) Analysis of synchronization properties. The dynamics of the spatial and temporal phase differences are derived from
the phase equations shown in the middle panel. By analyzing these dynamics, the time evolution of the phase differences and
the stability of the synchronized states between breathers can be investigated. The figure in the bottom panel shows the result,
including a comparison between theoretical values and direct numerical simulations. Additionally, it shows the nullclines and
fixed points corresponding to the synchronized states as well as their linear stability. Details of this figure are provided in
Sec. III B and Fig. 14(e).
 17

FIG. 2. Limit-torus solution, X0 (x − Φ, Θ) = (u0 (x − Φ, Θ), v0 (x − Φ, Θ)) with Φ = 0, for the FHN model.
 18

FIG. 3. Floquet zero eigenfunctions, Us (x−Φ, Θ) = (Uu,s (x−Φ, Θ), Uv,s (x−Φ, Θ)) and Ut (x−Φ, Θ) = (Uu,t (x−Φ, Θ), Uv,t (x−
Φ, Θ)) with Φ = 0, for the FHN model.
 19

FIG. 4. Phase sensitivity functions, Zs (x − Φ, Θ) = (Zu,s (x − Φ, Θ), Zv,s (x − Φ, Θ)) and Zt (x − Φ, Θ) = (Zu,t (x − Φ, Θ), Zv,t (x −
Φ, Θ)) with Φ = 0, for the FHN model.
 20

(a) (a)
FIG. 5. Antisymmetric components of the phase coupling functions, Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ), for the coupled FHN
models with K=diag(1, 1).

FIG. 6. Comparison between direct numerical simulations [ε = 5.0 × 10−5 (gray) and 2.5 × 10−5 (blue)] and theoretical values
(red) for the coupled FHN models with K = diag(1, 1). (a) Time evolution of the phase differences, ∆Φ/L and ∆Θ/π,
as functions of εt, starting from (∆Φ/L, ∆Θ/π) ≃ (0.025, 0.4). (b) Same as panel (a) but starting from (∆Φ/L, ∆Θ/π) ≃
(0.2, 0.975). (c) Trajectories in the (∆Φ/L, ∆Θ/π) space corresponding to the time evolutions shown in panels (a) and (b).
(a) (a)
The green and orange lines indicate the nullclines of Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ), respectively. Stable and unstable fixed
points are indicated by closed and open circles, respectively; saddle points are indicated by gray circles. Considering Eqs. (57),
(58), and (74)–(77), the numbers of stable fixed points, unstable fixed points, and saddle points within the rectangular region
defined by −1 < ∆Φ/L ≤ 1 and −1 < ∆Θ/π ≤ 1 are 2, 1, and 3, respectively. The initial conditions are indicated by open
squares.
 21

(a) (a)
FIG. 7. Antisymmetric components of the phase coupling functions, Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ), for the coupled FHN
models with K=diag(1, 0).

FIG. 8. Comparison between direct numerical simulations [ε = 5.0 × 10−5 (gray) and 2.5 × 10−5 (blue)] and theoretical values
(red) for the coupled FHN models with K = diag(1, 0). (a) Time evolution of the phase differences, ∆Φ/L and ∆Θ/π, as
functions of εt, starting from (∆Φ/L, ∆Θ/π) ≃ (0.25, 0.25), (0.25, 0.50), and (0.25, 0.75). (b) Same as panel (a) but starting
from (∆Φ/L, ∆Θ/π) ≃ (0.4, 0.8). (c) Trajectories in the (∆Φ/L, ∆Θ/π) space corresponding to the time evolutions shown in
(a) (a)
panels (a) and (b). The green and orange lines indicate the nullclines of Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ), respectively. Stable
and unstable fixed points are indicated by closed and open circles, respectively; saddle points are indicated by gray circles.
Considering Eqs. (57), (58), and (74)–(77), the numbers of stable fixed points, unstable fixed points, and saddle points within
the rectangular region defined by −1 < ∆Φ/L ≤ 1 and −1 < ∆Θ/π ≤ 1 are 2, 2, and 4, respectively. The initial conditions
are indicated by open squares. One of the trajectories shown in panel (c), corresponding to the trajectory in panel (b), was
terminated before reaching the fixed point because of the limitations in computational time. The theoretical value (red) for
this trajectory was computed up to εt = 9 × 104 . The range of εt shown in panel (b) is limited to 5000.
 22

FIG. 9. Example of time series X1 (x, t) = (u1 (x, t), v1 (x, t))
of the Gray–Scott model in the absence of coupling (ε = 0).

FIG. 10. Limit-torus solution, X0 (x − Φ, Θ) = (u0 (x − Φ, Θ), v0 (x − Φ, Θ)) with Φ = 0, for the Gray–Scott model.
 23

FIG. 11. Floquet zero eigenfunctions, Us (x−Φ, Θ) = (Uu,s (x−Φ, Θ), Uv,s (x−Φ, Θ)) and Ut (x−Φ, Θ) = (Uu,t (x−Φ, Θ), Uv,t (x−
Φ, Θ)) with Φ = 0, for the Gray–Scott model.
 24

FIG. 12. Phase sensitivity functions, Zs (x−Φ, Θ) = (Zu,s (x−Φ, Θ), Zv,s (x−Φ, Θ)) and Zt (x−Φ, Θ) = (Zu,t (x−Φ, Θ), Zv,t (x−
Φ, Θ)) with Φ = 0, for the Gray–Scott model.
 25

(a) (a)
FIG. 13. Antisymmetric components of the phase coupling functions, Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ), for the coupled Gray–
Scott models.
 26

FIG. 14. Comparison between direct numerical simulations [ε = 2.0×10−6 (gray), 1.5×10−6 (green), and 1.0×10−6 (blue)] and
theoretical values (red) for the coupled Gray–Scott models. (a)–(c) Time evolution of the phase differences, ∆Φ/L and ∆Θ/π,
as functions of εt, starting from (∆Φ/L, ∆Θ/π) ≃ (0.84, −0.95), (0.28, −0.95), and (0.008, −0.96629). (d)–(f) Trajectories in
(a)
the (∆Φ/L, ∆Θ/π) space corresponding to panels (a)–(c). The green and orange lines indicate the nullclines of Γs (∆Φ, ∆Θ)
(a)
and Γt (∆Φ, ∆Θ), respectively. Stable and unstable fixed points are indicated by closed and open circles, respectively; saddle
points are indicated by gray circles. The numbers of stable fixed points, unstable fixed points, and saddle points within the
rectangular region defined by −1 < ∆Φ/L ≤ 1 and −1 < ∆Θ/π ≤ 1 are 1, 1, and 2, respectively. Furthermore, the nullclines
(a) (a)
of Γs (∆Φ, ∆Θ) and Γt (∆Φ, ∆Θ) overlap at ∆Φ/L = ±1, where the green line is depicted as a broken line. The initial
conditions are indicated by open squares. The insets in panels (c) and (f) show the early-time behavior, where the spatial
phase difference initially increases and then decreases.
 27

FIG. 15. Waveforms of the periodic functions X0 (Θ) and

B(Θ). (a) X0 (Θ) for the FHN model. (b) X0 (Θ) for the
Gray–Scott model. (c) B(Θ) for the Gray–Scott model.