See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/299969587

Describing synchronization and topological excitations in arrays of magnetic

spin torque oscillators through the Kuramoto model

Article in Scientific Reports · September 2016

DOI: 10.1038/srep32528

CITATIONS READS

54 781

3 authors:

Vegard Flovik Ferran Macià

Aize New York University
19 PUBLICATIONS 220 CITATIONS 80 PUBLICATIONS 1,516 CITATIONS

SEE PROFILE SEE PROFILE

Erik Wahlstrom
Norwegian University of Science and Technology
61 PUBLICATIONS 2,543 CITATIONS

SEE PROFILE

All content following this page was uploaded by Vegard Flovik on 01 September 2016.

The user has requested enhancement of the downloaded file.

 www.nature.com/scientificreports

OPEN Describing synchronization and

topological excitations in arrays of
magnetic spin torque oscillators
received: 26 April 2016
accepted: 10 August 2016 through the Kuramoto model
Published: 01 September 2016
Vegard Flovik1, Ferran Macià2 & Erik Wahlström1

The collective dynamics in populations of magnetic spin torque oscillators (STO) is an intensely studied
topic in modern magnetism. Here, we show that arrays of STO coupled via dipolar fields can be modeled
using a variant of the Kuramoto model, a well-known mathematical model in non-linear dynamics. By
investigating the collective dynamics in arrays of STO we find that the synchronization in such systems
is a finite size effect and show that the critical coupling—for a complete synchronized state—scales
with the number of oscillators. Using realistic values of the dipolar coupling strength between STO we
show that this imposes an upper limit for the maximum number of oscillators that can be synchronized.
Further, we show that the lack of long range order is associated with the formation of topological
defects in the phase field similar to the two-dimensional XY model of ferromagnetism. Our results
shed new light on the synchronization of STO, where controlling the mutual synchronization of several
oscillators is considered crucial for applications.

The emergence of coherent phases of interacting oscillators is at the foundation of the cooperative functioning
of a wealth of different systems in nature1. Examples of collective behavior can be chosen within a wide range
of systems such as laser arrays2, Josephson junctions3, chemical reactions4, synchronously flashing firefly popu-
lations1, disease spreading5, or cortical oscillations in the brain6,7. Science has sought mathematical models for
understanding collective phenomena in large populations of oscillators that were tractable both analytically and
numerically.
The Kuramoto model is a well known mathematical model in non-linear dynamics that describes large
systems of coupled phase oscillators8. The model, with a remarkable simplicity, has been used to describe the
essential features of collective excitations in a vast set of biological and physical phenomena8–16. Although the
Kuramoto model originally described oscillators interacting all-to-all with the same strength, variations of the
model have been used to describe systems with phase offset and time delays in the couplings, other topologies like
one-dimensional structures with local couplings etc. (see e.g. ref. 9 for an overview of extensions of the Kuramoto
model). In particular, two-dimensional Kuramoto networks with diffusive local coupling accept solutions con-
sisting in waves, spirals and many other patterns17.
Understanding the collective behavior in oscillator networks is also an intensely studied topic in modern
magnetism: the synchronization of spin torque oscillators (STO). STO are strongly non-linear magnetic oscil-
lators that can be implemented into nanoscale devices working at microwave frequencies, and can be frequency
and phase locked to external oscillatory signals or other STO18–32. They are envisaged to be useful for a variety of
advanced magnetic nanodevices, as microwave sources and for signal processing in telecommunication technol-
ogies (see e.g. refs 33–35 and references therein). STO have also been proposed as possible candidates for a full
spintronic implementations of neural networks, based on nano-devices emulating both neurons and synapses35,36.
Building artificial neural networks for computation is an emerging field of research within bio-inspired comput-
ing33–40, where controlling the collective behavior in oscillator networks is crucial.
In both experimental and theoretical studies, most of the work has been performed for limited number of
oscillators. Experimentally, the synchronization of STO has proven to be difficult, and the synchronization of

1
Department of Physics, NTNU, Norwegian University of Science and Technology, N-7491 Trondheim, Norway.
2
Institut de Ciència de Materials de Barcelona (ICMAB-CSIC), Campus UAB, 08193 Bellaterra, Spain. Correspondence
and requests for materials should be addressed to V.F. (email: [email protected])

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 1

www.nature.com/scientificreports/

Figure 1. (a) The gyrotropic motion of the vortex core around the center of the disc can be described by the
radius r and phase, θ. (b) Two vortex based STO separated by a distance dij, showing the curling magnetization
in the disc plane and the location of the vortex core indicated in black. (c) Network model for an array of STO.
The interaction strength is determined by the spacing dij and we include interactions within a coupling radius
R, indicated by the blue circle in the figure. The local correlation function βi at position i (blue) is given by the
degree of synchronization with its neighbors (red).

only a few oscillators has been demonstrated30,32. Theoretically, the magnetization dynamics of STO is modeled
with the Landau-Lifshitz-Gilbert-Slonzewski (LLGS) equation41,42, but large number of STO lead to challenging
computations caused by the non-local dipolar fields. It is important to consider that in these non-linear systems
“more is different”, and that the collective behavior can not be derived simply from the behavior of its individual
elements. Thus, a theoretical framework capable to capture the essential dynamics would be ideal to explore those
systems.
Here, we show that two-dimensional arrays of STO coupled via dipolar fields can be modeled by a variant
of the Kuramoto model. We begin with describing two coupled STO with the Thiele equation43 and show that
for small-amplitude oscillations the system can be described as a simple phase oscillator model. Next, we model
the interactions for the case of a two-dimensional array of oscillators based on the dipolar coupling and obtain a
modified Kuramoto model. Finally we compare the results from our model to the micromagnetic solution of the
LLGS equation.
We find that the synchronization in two-dimensional arrays of dipolar coupled STO is purely a finite size effect
and the critical coupling strength for obtaining a globally synchronized state scales with the number of oscillators
N as λcrit ∝​  log(N). Using realistic values of the dipolar coupling strength between STO we show that this imposes
an upper limit for the maximum number of STO that can be synchronized. Further, we study the synchronization
transition between the initial formation of locally synchronized clusters and the globally synchronized and phase
coherent state and correlate it with a transition in the local order of the system. We also observe the emergence
of topological defects and the formation of patterns in the phase field similar to the two-dimensional XY-model
of magnetism—suggesting a connection between arrays of STO, systems described by a 2d Kuramoto model and
the 2d XY model of statistical mechanics.

Results
From the Thiele equation to the Kuramoto model. We are considering STO whose free layer ground
state configuration is a magnetic vortex. The vortex state is characterized by in-plane curling magnetization, and
a small (~10 nm) region of the vortex core with out-of-plane magnetization44. The gyrotropic motion of the vortex
core is driven by the injection of a DC spin polarized current through the STO stack, and can be described by
a gyration radius r and phase θ, as illustrated in Fig. 1a. We first model the interaction of two vortices with the
Thiele equation with an extra term that accounts for the vortex interaction. These equations describe the vortices
motion given by their coordinates X1,2 in their self induced gyrotropic mode, and include the spin-transfer-torque
(STT) as well as a coupling term43,45.

G (ez × X 1,2) − k (X1,2) X1,2 − D1,2 X 1,2 − FSTT1,2 − Fint (X2,1) = 0. (1)

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 2

www.nature.com/scientificreports/

Here, G is the gyroconstant, k(X1,2) the confining force, D1,2 the damping coefficient and FSTT the STT. The inter-
action between the neighboring STO illustrated in Fig. 1b is summarized by a dipolar coupling term given by
Fint =​  −​μ(d)X2,1, where μ(d) describes the interaction strength as a function of the separation d between the two
STO.
Assuming a small difference in the nominal frequencies of two coupled STO described by Eq. (1), one can lin-
earize the set of equations following the approach by Belanovsky et al.27, showing that the dynamics of the phase
difference between the STO can be described by Adler’s equation46. Following these approximations, the set of
equations reduce to that of two coupled phase oscillators θ1 and θ2: (see ‘Supplementary information’ for details).

θ 1 = ω1 + λ sin(θ2 − θ1), (2)

θ 2 = ω2 + λ sin(θ1 − θ2 ), (3)
where ω1,2 are eigenfrequencies of oscillators θ1,2 respectively, and λ describes the interaction strength trying to
synchronize them.
To check the validity of the approximations, one can compare the results obtained using a simplified phase
oscillator model to a numerical solution of Eq. (1) as well as a micromagnetic solution of the full system using
the LLGS equation. This was done by Belanovsky et al.27, where they found that for a small difference in nominal
frequencies, in their case given by a difference in STO disc diameter Δ​D/D0 ≤​ 5%, the synchronization can be
qualitatively described using the simplified model. Assuming the error in state-of-the-art fabrication processes is
below this limit, the simplified equations are a valid description of the system.
The functional form of Eqs (2–3) is the same as that of the well known Kuramoto model8,9, which is a gener-
alization for the case of an ensemble of weakly coupled phase oscillators. Considering the interaction between
several STO, we obtain a Kuramoto model where the single oscillator state is described through the dynamic
equation of its phase θi due to the interaction with its surrounding oscillators θj:
d θi
dt
= ωi + ∑λij sin(θ j − θi ).
j ≠i (4)
The coupling term is here generalized to include the interaction between several oscillators, determined by the
interaction strength λij between oscillators θi and θj. This determines the nature of the interaction, ranging from
a global all-to-all coupling where λij =​  λ for all oscillators, to a local interaction where λij =​ 0 for all but the nearest
neighbors. Here, we are considering the intermediate case of a non-local coupling to mimic the dipolar interac-
tion between neighboring STO. Starting from a macrodipole approximation for the dipolar energy between two
magnetic dipoles μ1 and μ2, the average interaction strength is found to decay as µ (d ) ∝ d ij−3 47, where dij is the
distance between oscillators θi and θj. We thus set the coupling strength to

λ /d ij3 d ij < R

λij = 
 0 d ij > R ,
 (5)
where we include interactions within a coupling radius R, indicated by the blue circle in Fig. 1c. The network
model for the STO array is implemented with bi-periodic boundary conditions and the time evolution of the
oscillator phases given by Eq. (4) is solved numerically. A small random disorder in the oscillator eigenfrequen-
cies is included by setting ωi =​  ω0 ±​  δωi. Here, ω0 =​ 1 GHz and δωi represents a uniformly distributed random
disorder where δωi/ω0 ≤​ 2.5%. The interaction strength is determined by the STO spacing and size, as well as the
magnetic material properties31. Here, the interaction strength has been varied in the range λ =​ 1–20 MHz, and is
in the same range as the interaction strength extracted from micromagnetic simulations for similar STO31.
In order to evaluate the Kuramoto model as a valid description for arrays of STO, we compare it to a micro-
magnetic solution of the complete system, accounting for all dynamic dipolar terms (see ‘Methods’ section). To
compare the Kuramoto model and the micromagnetic solution, we define a suitable order parameter to distin-
guish disordered and synchronized states. The phase of the individual oscillators θi is used to define the order
parameter ρ, describing the phase coherence in a system of N oscillators:

1
ρ=
N
∑e i ⋅ θ j .
j (6)
The case ρ =​ 0 corresponds to the maximally disordered state, whereas ρ =​ 1 represents the state where all
oscillators are perfectly synchronized and phase coherent. In addition to the global order parameter ρ, we define
a local correlation function β:

1
βi =
n
∑e i ⋅ θ j ,
j (7)
where the brackets indicate a summation over neighboring oscillators, and n is the number of neighbors. βi is
a measure of the phase correlation of oscillator θi and its neighbors, indicated by the blue and red oscillators in
Fig. 1c respectively. If oscillator θi is located within a synchronized cluster, βi →​  1. Calculating β thus allows for

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 3

www.nature.com/scientificreports/

Figure 2. Snapshots of the phase map θi and local correlation function βi for the Kuramoto model and the
micromagnetic solution at time t1 and t2 > t1 for a network of 45 × 45 oscillators.

investigating the formation of locally synchronized clusters and the emergence of patterns of synchronized states,
which can not be obtained simply from the global order parameter ρ.

Kuramoto model versus micromagnetic simulations. We now compare the Kuramoto model given
by Eq. (4) and the full micromagnetic solution of the LLGS equation (see ‘Methods’ section for details). Starting
from a disordered initial state, we investigate the synchronization dynamics by calculating the time evolution of
the phase distribution θi and local correlation βi. As an example, we show in Fig. 2 snapshots of θi and βi for the
Kuramoto model and the micromagnetic solution for a system of 45 ×​ 45 oscillators at times t1 and t2 >​  t1, with
random initial phases. At time t1, one notices the initial formation of small locally synchronized clusters, as seen
through both the phase distribution and the bright areas in the correlation maps, where β →​ 1. As time progress
to t2, these clusters grow in size and merge with neighboring clusters.
Comparing the two models, we find that they both show the same behavior. For weak interaction strengths,
the system tends to be in a disordered state with no correlation between neighboring oscillators. By increasing the
interaction strength above a certain threshold, synchronized clusters begin to form. The oscillators within each
cluster are synchronized, but might not be phase coherent with other clusters. This can be seen in the phase maps
in Fig. 2, where the individual clusters have different phases. As time progresses, the transition from a disordered
to a synchronized state is governed by the growth and merging of neighboring clusters, reaching a globally syn-
chronized and phase coherent state for sufficiently strong interactions.
Depending on the interaction strength, the system ends up in either a disordered, partially synchronized
or globally synchronized state. Controlling the interaction strength is thus the key parameter to determine the
system behavior. The interaction strength needed to obtain synchronization will depend on the differences in
the nominal frequencies of the oscillators31. Another important consideration, is whether the critical interaction
strength also depends on the number of oscillators. Lee et al.48 have studied the synchronization in a 2d Kuramoto
model with a nearest neighbor interaction. They showed that the transition to a synchronized state depends
strongly on system size, and that the critical coupling strength needed to synchronize scales with the number of
oscillators N as λcrit ∝​  log(N). This raises the question if such a scaling law can also be observed in our models:
observing the same scaling laws in both the Kuramoto model and the micromagnetic solution would strengthen
the suggestion of the Kuramoto model as a valid description of arrays of STO.
We first consider our Kuramoto model, which has a non-local interaction to mimic the dipolar interaction in
arrays of STO. Due to the increased complexity compared to the nearest neighbor model studied by Lee et al.48, an
analytical derivation of the scaling behavior with system size is to our knowledge still an open question. To inves-
tigate the scaling behavior we thus resort to a numerical solution. We performed simulations with the number of
oscillators ranging from N =​  9 to N =​ 2500, gradually increasing the interaction strength between each simulation
until the system reaches a synchronized state at a critical coupling strength, λcrit. 100 simulations were performed
for each system size, with different initial oscillator phases and eigenfrequencies. In Fig. 3a we show a plot of
λcrit vs. number of oscillators, N. The results indicate that the critical coupling strength scales as λcrit ∝​  log(N),
same as the nearest neighbor Kuramoto model investigated by Lee et al.48. The main difference compared with
the nearest neighbor model is that we include interactions within a coupling radius R, as indicated by the blue
circle in Fig. 1c. The imposed cutoff radius has a physical justification when considering a realistic system, which
would inevitably include thermal noise. As the STO we consider are weakly coupled, and the dipolar interaction
decay with distance, there will be a limiting spacing where the thermal noise level is comparable to the coupling

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 4

www.nature.com/scientificreports/

Figure 3. (a) Critical coupling strength λcrit vs. number of oscillators

∼ N in the Kuramoto model, normalized to
the case where N =​  3 ×​ 3. Red dotted line: For a system size N , there is a corresponding minimum coupling
strength λ to obtain a synchronized state, and vice versa. (b) Blue solid line: Expected scaling between critical
oscillator spacing, dcrit, and number of oscillators when assuming a dipolar interaction decaying as 1/d3. Red
datapoints: Results from micromagnetic simulations. (c) Black solid line: Power output assuming an ideal
scaling, P ∝​  N2. Blue dotted line: Calculated power output for an interaction strength λ= 7.5 MHz, normalized
to the case where N =​  3 ×​ 3. Inset: Order parameter ρ for 3 ×​ 3 and 13 ×​ 13 STO respectively, showing the
transition from a synchronized state to chaotic behavior as the number of STO is increased.

strength. In our model we observe that the results do not depend qualitatively on the range of the cutoff radius,
and that the large scale behavior is dominated by the diffusive coupling. This suggests that insight from studying
the analytically tractable nearest neighbor model of Lee et al.48 might provide valuable insight into the behavior
of STO arrays.
The coupling in the Kuramoto model is defined simply as an interaction strength given by λ in Eq. (4). In the
micromagnetic model the coupling comes from dipolar interactions, determined by the magnetic material prop-
erties and the STO spacing. In order to investigate the scaling behavior in the micromagnetic solution, one thus
needs to relate the critical interaction strength to a critical STO spacing. Following the aforementioned macrodi-
pole approximation that the effective dipolar interaction decay as 1/d3, one can relate the coupling strength λcrit
to a critical spacing dcrit between neighboring STO as dcrit ∝​  [log(N)]−1/3. Micromagnetic simulations were then
performed to obtain dcrit vs. number of oscillators in the range N =​  3 ×​  3 to N =​  15 ×​ 15. The results are shown
as the red datapoints in Fig. 3b. For comparison, we show as a solid line the expected scaling from the Kuramoto
model: dcrit ∝​  [log(N)]−1/3. The good agreement between the micromagnetics result and the Kuramoto model
indicate that they both follow the same scaling law, strengthening the suggestion of the Kuramoto model as a valid
description for arrays of STO.

Scaling of output power in arrays of STO. To investigate the implications of the scaling with system size,
we consider a model calculation of the output power as we increase the number of STO. For applications of STO
as e.g. nanoscale microwave generators, the power output of a single STO is not competitive. Decisive improve-
ment is expected from the synchronization and phase locking of several STO, as this would result in a quadratic
increase of the output power, P ∝​  N2 for N synchronized oscillators.
The output of a single STO can be described by its amplitude and phase, a j e i θ j. In our model we have assumed
a constant amplitude for all STO and the total amplitude A for an array of STO is given by: A = ∑ j ae i θ j. The
2
power output is proportional to |A|2, and for N oscillators we obtain P ∝ aN 1 ∑ j e i θ j ∝ (ρN )2, from the defi-
N
nition of the order parameter ρ in Eq. (6). A quadratic scaling in the power output, P ∝​  N2, implies a perfectly
synchronized and phase coherent state, given by ρ =​ 1. However, as the coupling strength to obtain a synchro-
nized state scales with the number of STO, this will affect the power output when scaling up to large arrays.
As an example, we consider a system of STO composed of 200 nm diameter spin valve nanopillars with 15 nm
thick Permalloy as the ferromagnetic layer. The average interaction energy can be extracted from micromagnetic

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 5

www.nature.com/scientificreports/

simulations, and for an edge-edge spacing of 150 nm the interaction strength is found to be λ≈  7–8 MHz31. To
increase the output power, we are now interested in scaling up to a large number of STO. We keep the same STO
spacing when scaling up, e.g. keeping λ fixed. From Fig. 3a, we see that for an interaction strength λ  , there is a
∼ 
corresponding number of STO, N , where λcrit > λ . This can be illustrated by calculating the total power output
as the number of STO is increased, as shown in Fig. 3c. For a small number of STO, we see that the power output
follows close to the ideal N2 scaling. However, as the number of STO is increased, the scaling with system size
∼
becomes increasingly important. For a certain number of STO, indicated by N in Fig. 3a, the interaction is no
longer strong enough to obtain a synchronized state. This is also illustrated in the inset of Fig. 3c, where we plot
the order parameter ρ for arrays of 3 ×​ 3 and 13 ×​ 13 STO respectively, showing the transition from a synchro-
∼
nized state to chaotic behavior as the number of STO is increased above N .
This illustrates the importance of our findings that synchronization in such 2d arrays is purely a finite size
effect. The interaction strength is limited by the material properties and STO spacing, and using realistic values of
the coupling strength we start to see significant deviations from the ideal P ∝​  N2 scaling for array sizes larger than
10 ×​ 10 STO (see Fig. 3c). This means that in a physical realizable system, the scaling with system size imposes an
upper limit for the maximum number of STO that can be synchronized.

Topological defects. That the synchronized and phase coherent state is purely a finite size effect, is similar
to that of the classical 2d XY model of magnetism. The Kuramoto model is indeed similar to the 2d XY model49,
where the direction of spin in the XY model corresponds to the oscillator phase in the Kuramoto model. In the
2d XY model a long range ordered phase is absent due to the presence of spin wave fluctuations and topological
defects. The lack of long range order is a specific case of the Mermin-Wagner theorem in spin systems50, stating
that continuous symmetries cannot be spontaneously broken in systems with sufficiently short range interactions
in dimensions d ≤​ 2. The fluctuations preventing long range order in the 2d XY model diverge logarithmically
with system size49, in agreement with the logarithmic scaling observed in our system of STO.
Similar topological defects at the boundaries between locally synchronized clusters have previously been
observed in the nearest neighbor Kuramoto model48 as well as in other two-dimensional oscillator network
models17,51–53. In oscillator networks this is associated with the appearance of topological defects in the oscillator
phase field, θi. In the continuum limit this is expressed as:
1
2π
∮ ∇θ ( r , t ) ⋅ d l = ± n , (8)
where dl is an integration path enclosing the defect, and n is the topological charge.
Such topological features are observed also in our Kuramoto model for arrays of STO. The presence of vortices
in the phase field is more pronounced as the system size increases. As an example we here consider an array of
50 ×​ 50 oscillators. The disorder in the system is kept constant (given by the difference in the nominal frequen-
cies of the oscillators) and the interaction strength λ is varied, acting as the inverse temperature: as coupling
increases, the system becomes more ordered. Starting from a disordered initial state and varying the interaction
strength between each simulation, we observe 4 different regimes: For weak interaction strengths we observe the
formation of locally synchronized clusters, where cluster sizes increase with interaction strength. Apart from the
localized clusters there is no long range order in the system (indicated as regime 1 in Fig. 4). Increasing λ above a
certain threshold, we enter regime 2. Here we observe the formation of vortices in the phase field, and as an exam-
ple we show in Fig. 4c a state with 4 vortices. The topological charge is conserved in the system, and two vortices
of charge ±​1 respectively is present.
In both regime 1 and 2 long range order in the system is absent and ρ ≈​ 0, as indicated in Fig. 4a. (that ρ >​  0
here is a result of fluctuations due to the finite array size). Increasing λ further we enter regime 3, where the tran-
sition from regime 2 →​ 3 is governed by vortex annihilation processes (see ‘Supplementary information’). Here
there are no topological defects in the phase field, and the lack of global phase coherence is due to spin waves in
the phase field where the oscillator phases change smoothly across the array (regime 3 in Fig. 4b). For sufficiently
strong interactions we enter regime 4. Increasing the interaction strength is analogous to increasing the exchange
coupling in a Ferromagnetic system, resulting in a more ordered state. The result here is a gradual suppression
of the spin waves in the phase field observed in regime 3 as the interaction strength is increased. Regime 4 is
thus characterized as the globally phase coherent state where all oscillators are synchronized and phase coherent
(regime 4 in Fig. 4b).
The growth of the order parameter ρ with increasing coupling strength λ in Fig. 4a resembles that of a phase
transition. Previous work have shown that the synchronization transition in the globally coupled Kuramoto
model can be described as a phase transition, where the nature of the transition can be of first or second-order
depending on the frequency distribution and coupling topology8,54. The Kuramoto model with finite range cou-
plings is less studied, as these systems are difficult to analyze and solve analytically. A study of the locally coupled
Kuramoto model on a d-dimensional lattice have shown that the synchronization transition depends strongly on
the lattice dimensionality, and indicates d =​ 4 as the lower critical dimension for phase synchronization55. This is
in agreement with the observed scaling with system size in our model, which indicates that the synchronization
transition is purely a finite size effect.
In order to investigate the synchronization transition in our model further, we calculate the spatial correlation
function for the array of oscillators. The correlations decay with distance, and asymptotically the correlation
function is given by: 〈​θ(r) ⋅​  θ(R)〉​  ∝​  e−|r−R|/ξ/|r −​  R|η. This describes the correlation between oscillators at positions
r and R respectively, and the correlation length ξ is obtained by averaging over all positions r and R in the array
(an example is shown in ‘Supplementary information’ Fig. S2). From the decay of the correlation function, we

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 6

www.nature.com/scientificreports/

Figure 4. (a) Order parameter ρ vs. interaction strength λ in the Kuramoto model for a system of N =​  50 ×​  50
oscillators, showing the transition from a disordered (ρ ≈​ 0) to a globally synchronized and phase coherent
state (ρ →​ 1). Inset: Calculated correlation length ξ vs. interaction strength λ, where the correlation length is
normalized to the system size, ξ/L (N =​  L ×​  L) (b) The corresponding phase maps, showing the transition from
a disordered state via the formation of locally synchronized clusters (1), vortices (2), spin waves (3) and the
globally phase coherent state (4). (c) Zoom in of phase map for regime 2, showing the appearance of 4 vortices
of charges ±​1 respectively, as defined in Eq. (8).

obtain the correlation length ξ as a function of the interaction strength λ. Conventional phase transitions are
accompanied by a diverging correlation length close to the transition. Here, we do not observe a diverging ξ going
from the disordered to the phase coherent state (1 →​ 4 in Fig. 4), and the correlation length remains finite. As
inset in Fig. 4a we show a log-log plot of the correlation length normalized to the system size, ξ/L, (N =​  L ×​  L).
The results indicate a power law relating the correlation length and the interaction strength as ξ ∝​  λν, where the
exponent for this case was found to be ν =​  2.1 ±​ 0.1. This means that the correlation length simply scales with the
coupling strength, and the transition between regimes 1–4 in Fig. 4 correspond to structures of ever increasing
length scales. The transition to the phase coherent state (ρ →​ 1) occurs when the correlation length approaches
the system size L, underlining the finite size effects on the synchronization transition and that the system is not
undergoing a conventional phase transition. Further investigations of finite size effects, the lack of long ranger
order in the Kuramoto model and the connection to the 2d XY model will be the subject of future work.

Discussion
To summarize, we have shown that the Kuramoto model provides a good description of arrays of STO. It pro-
vides a simple theoretical model to study large populations of coupled STO, which were previously unaccessible
due to the long computation time for a full micromagnetic solution. By investigating the collective dynamics
in large arrays of STO, we observed a scaling with system size indicating that the synchronization in arrays of
dipolar-coupled STO is purely a finite size effect. The critical coupling strength to obtain a globally synchro-
nized state scales with the number of oscillators, as λcrit ∝​  log(N), preventing global synchronization for large
system sizes. As a consequence of the scaling with system size, we showed that for realistic values of the dipolar
coupling strength between STO this imposes an upper limit for the maximum number of oscillators that can be
synchronized. Further, we showed that the lack of long range order and scaling with system size is associated with
the emergence of topological defects and the formation of patterns in the phase field, similar to that of the 2d
XY-model of magnetism.
In the present study we considered dipolar-coupled STO, where the short time delay in the coupling between
neighboring oscillators compared to the oscillator frequency means that phase delay in the couplings can be
neglected. However, for other coupling mechanisms, time delay can become significant. Interaction mediated by
spin waves provide a different mechanism to obtain synchronization of STO32, where the finite propagation speed
of the spin waves results in a phase offset in the couplings. Another recent proposal includes the use of non-local
electrical couplings, where the coupling phase can be externally tuned through an electrical delay line56.
From a dynamical systems point of view, the study of time delay induced modifications to the couplings is of
fundamental interest, as well as of practical relevance for modeling of physical, biological and chemical systems.
In such systems, time delay is associated with finite propagation velocity of the couplings via e.g latency times of

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 7

www.nature.com/scientificreports/

neuronal excitations, reaction times in chemical systems etc. (see e.g. ref. 9 and references therein). The possibil-
ity of designing STO arrays with a defined phase offset in the couplings suggests a real world analog to the more
general Sakaguchi-Kuramoto model on a 2d lattice, which allows for a phase lag in the couplings57.
Our study suggests, on the one hand, that the use of models from non-linear dynamics can be useful for
describing synchronization of magnetic oscillators and, on the other hand, arrays of STO as a a physical realizable
model system for the Kuramoto model on a 2d lattice.

Methods
Micromagnetic model. The micromagnetic model is defined as arrays of discs, where the system is divided
into a grid with a mesh size of 5 nm, limited by the exchange length of the ferromagnetic material (here Py). The
volumetric quantities such as the magnetization M and effective field Heff are treated at the center of each cell,
whereas coupling quantities such as exchange strength are considered at the faces between cells. The numerical
solution was obtained using the micromagnetic solver Mumax3 58, which uses a RKF 45 method to solve the
Landau-Lifshitz-Gilbert-Slonzewski (LLGS) equation41,42 given by:

dM α  dM  χ  
γ M×
= − Heff
 + M M × dt  − d JP (θ) M × (M × m f  .
dt 

s 
 
Gyration
Damping Spin − TransferTorque (9)
Here, γ is the gyromagnetic ratio, α the damping parameter and M s the saturation magnetization. The
spin-transfer torque term is given by χ = g µb /(2Ms2 e), the charge current density J and the free layer thickness d.
P(θ) is a polarization function assumed to increase with the relative angle θ between the magnetization of the free
layer and the fixed layer and mf is a unit vector in the direction of the magnetization of the free layer.
In the model, each disc is composed by a magnetic free layer and a fixed polarizer which generates a perpen-
dicular spin polarization pz. The free layer in the STO is 30 nm thick Py with a disc diameter of 150 nm, and the
damping parameter α was set to 0.01. A small disorder in the eigenfrequencies of the individual STO is included
through a random distribution in the saturation magnetization in the range [865, 885] ⋅​  103 A/m, resulting in a
slight variation of STO eigenfrequencies. All discs were initialized with a vortex of same polarity and chirality,
and the center-center spacing of the discs was varied to change the interaction strength. The polarizing layers are
not included in the model as these layers, being uniformly magnetized in z direction have almost no influence
on the vortices motion. The vortex gyration was driven by a DC spin current with a polarization pz =​  0.3, and
current density J ≈​  4.3 ⋅​  107 A/m2. During the simulations, a static magnetic field of 150 mT was applied along the
z direction to set the vortex core polarity.

References
1. Strogatz, S. H. & Stewart, I. Coupled oscillators and biological synchronization. Sci. Am. 269(6), 102–109 (1993).
2. Kourtchatov, S. Yu., Likhanski, V. V., Napartovich, A. P., Arecchi, F. T. & Lapucci, A. Theory of phase locking of globally coupled laser
arrays. Phys. Rev. A 52, 4089 (1995).
3. Wiesenfeld, K., Colet, P. & Strogatz, S. H. Synchronization Transitions in a Disordered Josephson Series Array. Phys. Rev. Lett. 76,
404 (1996).
4. Singh, R. & Sinha, S. Spatiotemporal order, disorder, and propagating defects in homogeneous system of relaxation oscillators. Phys.
Rev. E. 87, 012907 (2013).
5. Kuperman, M. & Abramson, G. Small World Effect in an Epidemiological Mode. Phys. Rev. Lett. 86, 2909 (2001).
6. Breakspear, M., Heitmann, S. & Daffertshofer, A. Generative models of cortical oscillations: neurobiological implications of the
Kuramoto model. Front. Hum. Neurosci. (2010).
7. Izhikevich, E. M. Which Model to Use for Cortical Spiking Neurons? IEEE transactions on neural networks. 15.5, 1063–1070 (2004).
8. Kuramoto, Y. & Nishikawa, I. Statistical macrodynamics of large dynamical systems. Case of a phase transition in oscillator
communities. J. Stat. Phys. 49, 569 (1987).
9. Acebrón, J. A. et al. The Kuramoto model: A simple paradigm for synchronization phenomena. Rev. Mod. Phys. 77, 137 (2005).
10. Filatrella, G., Nielsen, A. H. & Pedersen, N. F. Analysis of a power grid using a Kuramoto-like model. The European Physical Journal
B. 61(4), 485–491 (2008).
11. Daniels, B. C., Dissanayake, S. T. M. & Trees, B. R. Synchronization of coupled rotators: Josephson junction ladders and the locally
coupled Kuramoto model. Phys. Rev. E 67, 026216 (2003).
12. Wiesenfeld, K., Colet, P. & Strogatz, S. H. Frequency locking in Josephson arrays: Connection with the Kuramoto model. Phys. Rev.
E 57, 1563 (1998).
13. Néda, Z., Ravasz, E., Vicsek, T., Brechet, T. & Barabási, A. L. Physics of the rhythmic applause. Phys. Rev. E 61, 6987 (2000).
14. Heinrich, G., Ludwig, M., Qian, J., Kubala, B. & Marquardt, F. Collective Dynamics in Optomechanical Arrays. Phys. Rev. Lett. 107,
043603 (2011).
15. Cumin, D. & Unsworth, C. P. Generalising the Kuramoto model for the study of neuronal synchronisation in the brain. Physica D:
Nonlinear Phenomena, 226(2), 181–196 (2007).
16. Timms, L. & English, L. Q. Synchronization in phase-coupled Kuramoto oscillator networks with axonal delay and synaptic
plasticity. Phys. Rev. E. 89, 032906 (2014).
17. Lauter, R., Brendel, C., Habraken, S. J. M. & Marquardt, F. Pattern phase diagram for two-dimensional arrays of coupled limit-cycle
oscillators. Phys. Rev. E. 012902 (2015).
18. Rippard, W. H. et al. Injection Locking and Phase Control of Spin Transfer Nano-oscillators. Phys. Rev. Lett. 95 067203 (2005).
19. Kaka, S. et al. Mutual phase-locking of microwave spin torque nano-oscillators. Nature, 437 389–392 (2005).
20. Mancoff, F. B., Rizzo, N. D., Engel, B. N. & Tehrani, S. Phase-locking in double-point-contact spin-transfer devices. Nature 437
393–395 (2005).
21. Pufall, M. R., Rippard, W. H., Russek, S. E., Kaka, S. & Katine, J. A. Electrical Measurement of Spin-Wave Interactions of Proximate
Spin Transfer Nanooscillators. Phys. Rev. Lett. 97, 087206 (2006).
22. Bonin, R., Bertotti, G., Serpico, C., Mayergoyz, I. D. & d’Aquino, M. Analytical treatment of synchronization of spin-torque
oscillators by microwave magnetic fields. Eur. Phys. J. B. 68, 221 (2009).
23. Zhou, Y., Person, J. & Åkerman, J. Intrinsic phase shift between a spin torque oscillator and an alternating current. J. Appl. Phys. 101,
09A510 (2007).

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 8

www.nature.com/scientificreports/

24. Recende, S. M., de Aguiar, F. M. & Azevedo, A. Spin-Wave Theory for the Dynamics Induced by Direct Currents in Magnetic
Multilayers. Phys. Rev. Lett. 94, 037202. (2005).
25. Rezende, S. M., de Aguiar, F. M. & Azevedo, A. Spin-Wave Theory for the Dynamics Induced by Direct Currents in Magnetic
Multilayers. Phys. Rev. B. 73, 094402 (2006).
26. Slavin, A. N. & Tiberkevich, V. S. Theory of mutual phase locking of spin-torque nanosized oscillators. Phys. Rev. B. 74, 104401
(2006).
27. Belanovsky, A. D. et al. Numerical and analytical investigation of the synchronization of dipolarly coupled vortex spin-torque nano-
oscillators. Appl. Phys. Lett. 103, 122405 (2013).
28. Belanovsky, A. D. et al. Phase locking dynamics of dipolarly coupled vortex-based spin transfer oscillators. Phys. Rev. B 100409(R)
(2012).
29. Georges, B., Grollier, J., Cros, V. & Fert, A. Impact of the electrical connection of spin transfer nano-oscillators on their
synchronization: an analytical study. Appl. Phys. Lett 92, 232504 (2008).
30. Ruotolo, A. et al. Phase-locking of magnetic vortices mediated by antivortices. Nature Nanotechnology 4, 528–532 (2009).
31. Locatelli, N. et al. Efficient Synchronization of Dipolarly Coupled Vortex-Based Spin Transfer Nano-Oscillators. Sci. Rep. 5, 17039
(2015).
32. Housang, A. et al. Spin-wave-beam driven synchronization of nanocontact spin-torque oscillators. Nature Nanotechnology 11,
280–286 (2016).
33. Bonetti, S. & Åkerman, J. Magnonics: From Fundamentals to Applications (2013).
34. Macià, F., Kent, A. D. & Hoppensteadt, F. C. Spin-wave interference patterns created by spin-torque nano-oscillators for memory
and computation. Nanotechnology 22(9), 095301 (2011).
35. Locatelli, N., Cros, V. & Grollier, J. Spin-torque building blocks. Nature Materials, 13, 11 (2014).
36. Csaba, G. & Porod, W. Computational Study of Spin-Torque Oscillator Interactions for Non-Boolean Computing Applications. IEEE
Trans. Magn. 49(7), 4447–4451 (2013).
37. Hoppensteadt, F. C. & Izhikevich, E. M. Oscillatory Neurocomputers with Dynamic Connectivity. Phys. Rev. Lett. 82, 2983 (1999).
38. Maffezzon, P., Bahr, B., Zheng, Z. & Daniel, L. Oscillator Array Models for Associative Memory and Pattern Recognition. IEEE
Trans. Circuits Syst. I, Reg. Papers, 62(6), 1591–1598 (2015).
39. Locatelli, N. et al. Spintronic devices as key elements for energy-efficient neuroinspired architectures. Proceedings of the 2015
Design, Automation & Test in Europe Conference & Exhibition.
40. Vincent, A. F. et al. Spin-transfer torque magnetic memory as a stochastic memristive synapse for neuromorphic systems. IEEE
Transactions on Biomedical Circuits and Systems. 9(2) (2015).
41. Landau, L. & Lifshitz, E. On the Theory of the Dispersion of Magnetic Permeability in Ferromagnetic Bodies. Phys. Z. Sowjetunion
8, 153 (1935).
42. Slonczewski, J. C. Current-driven excitation of magnetic multilayers. J. Magn. Mater, 159, 1–2, L1–L7 (1996).
43. Thiele, A. A. Steady-State Motion of Magnetic Domains. Phys. Rev. Lett. 30, 230 (1973).
44. Shinjo, T., Okuno, T., Hassdorf, R., Shigeto, K. & Ono, T. Magnetic Vortex Core Observation in Circular Dots of Permalloy. Science,
289, 930–932 (2000).
45. Khvalkovskiy, A. V., Grollier, J., Dussaux, A., Zvezdin, K. A. & Cros, V. Vortex oscillations induced by spin-polarized current in a
magnetic nanopillar: Analytical versus micromagnetic calculations. Phys. Rev. B 80, 140401(R) (2009).
46. Adler, R. A Study of Locking Phenomena in Oscillators. Proc. IRE 34, 351 (1946).
47. Araujo, F. A. et al. Optimizing magnetodipolar interactions for synchronizing vortex based spin-torque nano-oscillators. Phys. Rev.
B 92, 045419 (2015).
48. Lee, T. E., Tam, H., Refael, G., Rogers, J. L. & Cross, M. C. Vortices and the entrainment transition in the two-dimensional Kuramoto
model. Phys. Rev. E 82, 036202 (2010).
49. Kosterlitz, J. M. & Touless, D. J. Ordering, metastability and phase transitions in two-dimensional systems. J. Phys. C. 6. 1181 (1973).
50. Mermin, N. D. & Wagner, H. Absence of Ferromagnetism or Antiferromagnetism in One- or Two-Dimensional Isotropic
Heisenberg Models. Phys. Rev. Lett. 17, 1133 (1966).
51. Davidsen, J. & Kapral, R. Phase synchronization and topological defects in inhomogeneous media. Phys. Rev. E. 66, 055202(R)
(2002).
52. Singh, R., Xu, J., Garnier, N. G., Pumir, A. & Sinha, S. Self-Organized Transition to Coherent Activity in Disordered Media. Phys.
Rev. Lett. 108, 068102 (2012).
53. Groβ​mann, G., Peruani, F. & Bär, M. Superdiffusion, large-scale synchronization, and topological defects. Phys. Rev. E. 93,
040102(R) (2016).
54. Basnarkov, L. & Urumov, V. Phase transitions in the Kuramoto model. Phys. Rev. E. 76, 057201 (2007).
55. Hong, H., Park, H. & Choi, M. Y. Collective synchronization in spatially extended systems of coupled oscillators with random
frequencies. Phys. Rev. E. 72, 036217 (2005).
56. Lebrun, R. et al. Mutual synchronization of spin torque nano-oscillators through a non-local and tunable electrical coupling.
arXiv:1601.01247.
57. Sakaguch, H. & Kuramoto, Y. A Soluble Active Rotater Model Showing Phase Transitions via Mutual Entertainment. Prog. Theor.
Phys. 76, 576 (1986).
58. Vansteenkiste, A. et al. The design and verification of MuMax3. AIP Advances 4, 107133 (2014).

Acknowledgements
This work was supported by the Norwegian Research Council (NFR), project number 216700. V.F. acknowledge
partial funding obtained from the Norwegian PhD Network on Nanotechnology for Microsystems, which
is sponsored by the Research Council of Norway, Division for Science, under contract no. 221860/F40. F.M.
acknowledges financial support from the Ramón y Cajal program through RYC-2014-16515 and from the
MINECO through the Severo Ochoa Program for Centers of Excellence in R&D (SEV- 2015-0496).

Author Contributions
V.F. initiated the project, developed the model, performed the calculations/simulations and wrote the manuscript.
F.M. and E.W. supervised the project and provided valuable input during the analysis/discussion of results and
writing of the manuscript.

Additional Information
Supplementary information accompanies this paper at http://www.nature.com/srep
Competing financial interests: The authors declare no competing financial interests.

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 9

www.nature.com/scientificreports/

How to cite this article: Flovik, V. et al. Describing synchronization and topological excitations in arrays of
magnetic spin torque oscillators through the Kuramoto model. Sci. Rep. 6, 32528; doi: 10.1038/srep32528
(2016).
This work is licensed under a Creative Commons Attribution 4.0 International License. The images
or other third party material in this article are included in the article’s Creative Commons license,
unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license,
users will need to obtain permission from the license holder to reproduce the material. To view a copy of this
license, visit http://creativecommons.org/licenses/by/4.0/

© The Author(s) 2016

Scientific Reports | 6:32528 | DOI: 10.1038/srep32528 10

Supplementary information: Describing synchronization and
topological excitations in arrays of magnetic spin torque
oscillators through the Kuramoto model

Vegard Flovik,1, ∗ Ferran Macià,2 and Erik Wahlström1

1
Department of Physics, NTNU, Norwegian University
of Science and Technology, N-7491 Trondheim, Norway
2
Institut de Ciència de Materials de Barcelona (ICMAB-CSIC),
Campus UAB, 08193 Bellaterra, Spain

i
From the Thiele equation to the phase oscillator model

We here provide some more details on the derivation of the Kuramoto model starting from
the coupled Thiele equation:

G(ez × Ẋ1,2 ) − k(X1,2 )X1,2 − D1,2 Ẋ1,2 − FSTT1,2 − Fint (X2,1 ) = 0. (S1)

Here, G = −2πpMs h/γ is the gyroconstant, p is the core polarity, γ is the gyromagnetic ratio,
Ms is the saturation magnetization and h is the thickness of the ferromagnetic layer. The
 
X21,2
confining force is given by k(X1,2 ) = ω01,2 G 1 + a R1,2 [1, 2], where R1,2 are the disc radii
20
and the gyrotropic frequency for disc 1, 2 is ω01,2 = 9
γMs h/R1,2 . The damping coeficcient
  q
R1,2 A
−D1,2 = αη1,2 G, where η1,2 = 21 ln 2le
+ 38 . Here, le = 2πMs
is the exchange length given
by the exchange stiffness A and the saturation magnetization Ms . Assuming a uniform
perpendicularly magnetized polarizer layer, FSTT = πγaJ Ms h(X1,2 × ez ) = κ(X1,2 × ez )
[3], where the spin torque coefficient is given by aJ = ~pz J/(2|e|hMs ), ~ is the Planck‘s
constant, J is the current density and e is the elementary charge. The interaction between the
neighboring vortices is summarized by a dipolar coupling term given by Fint = −µ(d)X2,1 ,
where µ(d) describes the interaction strength as a function of the separation d between the
STO. A study of the dipolar interaction between neighboring vortices has been performed
by Araujo et al. [4]. Starting from a macrodipole approximation for the dipolar energy
between two magnetic dipoles µ1 and µ2 , they show that the average interaction energy can
be written as hWint i = µeff C1 C2 X1 X2 . Here, Ci and Xi are the chirality and gyration radius
respectively and µeff is given by:

π 2 χ2 R2 h2
µeff = 3 , (S2)
2d3
where χ = 2/3, R is the disc radius, h the thickness and d is the inter-disc spacing. In polar
coordinates (X1,2 cos θ1,2 , X1,2 sin θ1,2 ), the coupled equations for two neighboring vortices
from Eq. (S1) can be written as:

Ẋ1 κ µX2
= αη1 θ˙1 − + sin(θ1 − θ2 ) (S3)
X1 G GX1

i
 k(X1 ) Ẋ1 µX2
θ˙1 = − − αη1 − cos(θ1 − θ2 ) (S4)
G X1 GX1

Ẋ2 κ µX1
= αη2 θ˙2 − − sin(θ1 − θ2 ) (S5)
X2 G GX2

k(X2 ) Ẋ2 µX1

θ˙2 = − − αη2 − cos(θ1 − θ2 ) (S6)
G X2 GX2
One can then show that after a few approximations, the set of equations reduce to that
of two coupled phase oscillators. We assume the same gyration radius for both vortices,
X2 = X1 , and that the steady state vortex gyrotropic radius is close to its mean value, X0 .
This means that Eq. (S3) can be set to zero, as Ẋ1 = 0, and we obtain:

κ µ
θ˙1 = − sin(θ1 − θ2 ) (S7)
αη1 G αη1 G

Setting Ẋ1 = 0 and X2 = X1 also in Eq. (S4):

k(X1 ) µ
θ˙1 = − − cos(θ1 − θ2 ) (S8)
G G
We then add Eqs. (S7) and (S8) to obtain:

κ − αη1 k(X1 ) µ
θ˙1 = − [sin(θ1 − θ2 ) + αη1 cos(θ1 − θ2 )] . (S9)
2αη1 G 2αη1 G
Following the same procedure for vortex nr. 2 and assuming low damping, αη << 1, we
obtain the equations for two coupled phase oscillators θ1 and θ2 :

θ˙1 = ω1 + λ sin(θ2 − θ1 ), (S10)

θ˙2 = ω2 + λ sin(θ1 − θ2 ), (S11)

κ−αη1,2 k(X1,2 ) µ
Where ω1,2 = 2αη1,2 G
and λ = 2αη1,2 G
. The functional form of Eqs. (S10)-(S11) is the
same as that of the well known Kuramoto model, which is a generalization for the case of an
ensemble of weakly coupled phase oscillators. Considering the interaction between several
STO, determined by the interaction strength λij between oscillators θi and θj , we obtain a

ii
Kuramoto model for a population of N interacting oscillators:

dθi X
= ωi + λij sin(θj − θi ). (S12)
dt j6=i

Vortex annihilation processes

Starting from a disordered initial condition, a number of vortices with n = ±1 is created

initially, depending on the array size. Thermal fluctuations of sufficient amplitude could
give rise to vortex unbinding, where free vortices proliferate due to thermal fluctuations. As
we do not consider thermal effects, such vortex unbinding is not observed this in our model.
Since a vortex is topological, it exists until it meets and annihilates with a vortex of opposite
polarity, and the transition from disordered to a synchronized state is governed by vortex
annihilation processes.

a) 1
0.8

0.6
ρ

0.4

0.2

0
0 500 1000 1500 2000 2500
b) Time [ns]
time [ns]
Phase
5 5 5 5

10 10 10 10

15 15 15 15

20 20 20 20

25 25 25 25

30 30 30 30

35 35 35 35

40 40 40 40

45 45 45 45

50 50 50 50
5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50 5 10 15 20 25 30 35 40 45 50

Correlation
5 5 5 5

10 10 10 10

15 15 15 15

20 20 20 20

25 25 25 25

30 30 30 30

35 35 35 35

40 40 40 40

45 45 45 45

5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45 5 10 15 20 25 30 35 40 45

FIG. S1. a) Order parameter ρ vs. time for an interaction strength of λ = 25 MHz for a system of 50 × 50
oscillators, starting from a disordered initial state. b) Snapshots of phase and correlation maps at various
timesteps (increasing time from left to right), showing the vortex annihilation processes.

iii
In Fig. S1a we show the order parameter ρ vs. time, starting from a disordered initial state
for a system of 50 × 50 oscillators using the Kuramoto model. The observed jumps in the
order parameter correspond to the annihilation of vortices of charge ±1. This process is also
illustrated in the panels of Fig. S1b, where we show snapshots of the phase map θi and local
correlation βi at various timesteps (with time increasing from left to right). The location
and polarity (n = ±1) of the vortices can be seen in the phase maps in the upper panels.
The position of the vortex core is identified by areas of low correlation (β → 0) between
neighboring oscillators, seen as the black spots in the lower panels. As time progress the
vortices annihilate, resulting in a globally synchronized and phase coherent state.

Correlation function and correlation length

The spatial correlation function is given asymptotically by: hθ(r)·θ(R)i ∝ e−|r−R|/ξ /|r −R|η .
The brackets indicate the correlation between oscillators at positions r and R, and the
correlation length ξ is obtained by averaging over all positions r and R in the array. An
example of the decay of spatial correlations is shown in Fig. S2 for a system of 50 ×
50 oscillators using the Kuramoto model, showing a dominating exponential decay in the
correlations for increasing distances between the oscillators. The spacing |r − R| is here
expressed in terms of the number of lattice spacings between the oscillators. From the decay
of the correlation function, we can then extract the correlation length ξ.

0.8
< θ (r)⋅θ (R) >

0.6

0.4

0.2

0
5 10 15
|r−R|
FIG. S2. Correlation as a function of oscillator spacing, |r − R| for a system of 50 × 50 oscillators using the
Kuramoto model.

iv
 ∗ [email protected]
[1] Guslienko, K. Yu., Han, X. F., Keavney, D. J., Divan, R., Bader, S. D. Magnetic Vortex Core
Dynamics in Cylindrical Ferromagnetic Dots. Phys. Rev. Lett. 96, 067205 (2006).
[2] Ivanov, B. A., Zaspel, C. E. Excitation of Spin Dynamics by Spin-Polarized Current in Vortex
State Magnetic Disks. Phys. Rev. Lett. 99, 247208, (2007).
[3] Khvalkovskiy, A. V., Grollier, J., Dussaux, A., Zvezdin, A., Cros, V. Vortex oscillations in-
duced by spin-polarized current in a magnetic nanopillar: Analytical versus micromagnetic
calculations. Phys. Rev. B 80, 140401(R) (2009).
[4] Araujo, F. A. et. al. Optimizing magnetodipolar interactions for synchronizing vortex based
spin-torque nano-oscillators. Phys. Rev. B 92, 045419 (2015).

v
View publication stats