SAW16, dated September 28, 2018

Synchronization of organ pipes

1, 2 1 Jakub Sawicki, ∗ Markus Abel, and Eckehard Schöll 1Institut für Theoretische Physik, Technische Universität Berlin, Hardenbergstraße 36, 10623 Berlin, Germany 2Department of Physics and Astronomy, Potsdam University, Karl-Liebknecht-Straße 24, 14476 Potsdam Germany We investigate synchronization of coupled organ pipes. Synchronization and reflection in the organ lead to undesired weakening of the sound in special cases. Recent experiments have shown that sound interaction is highly complex and nonlinear, however, we show that two delay-coupled Van-der-Pol oscillators appear to be a good model for the occurring dynamical phenomena. Here the coupling is realized as distance-dependent, or time-delayed, equivalently. Analytically, we investigate the synchronization frequency and bifurcation scenarios which occur at the boundaries of the Arnold tongues. We successfully compare our results to experimental data.

PACS numbers: 05.45.Xt, 05.45.-a

Introduction - The physics of organ pipes is an in- we investigate the bifurcation scenarios in the context of terdisciplinary topic where many fields of science meet. two delay-coupled Van-der-Pol oscillators as a represen- It is highly interesting as it includes elements of non- tation of the system of two coupled organ pipes, such linear dynamical system theory [4, 6, 11], aeroacoustic as in the experimental setup of Bergweiler et al. [5]. modeling [13] and synchronization theory [17]. The fo- In extension of previous work, we study the dependence cus of these different research areas is the “queen of in- of Arnold tongues under variation of the time delay τ struments” which captivates through the grandeur of her and the coupling strength κ, to explore how undesired sight and majesty of her sound. Here, we investigate synchronization or chaotic behavior can be avoided. We an interesting nonlinear effect: organ pipes close to each compare our results to experimental measurements of the other synchronize. Recent studies have been of exper- synchronization under variation of the pipe distance and imental nature as well as theoretical [1, 2, 7, 8]. For find a qualitative coincidence of the nonmonotonic mod- musical purposes, synchronization of sound might be de- ulation of the shape of the Arnold tongue which is in sired or not: it might stabilize the pitch of special organ contrast to the linear boundaries of the Arnold tongues pipes as a favorable effect, whereas sound weakening, as for systems with undelayed coupling. observed in the prospect of an organ, is highly undesired In Section II we introduce two delay-coupled Van der [9, 18, 20]. This weakening occurs as an amplitude mini- Pol oscillators as a simple model of coupled organ pipes. mum due to destructive interaction between pipes during In section II we apply two analytical methods to get a the actuating of the swell box, where the pipes stand close better understanding of the synchronization phenomena. to each other. The central part of the paper is Sec. III, where we present A qualitative understanding of the nonlinear mecha- the analytical results. In Sec. IV we compare these results nisms is obtained following the arguments of [2, 7]: a with acoustic experiments. We conclude with Sec. V. single organ pipe can be described as a self-sustained os- cillator, where the oscillating unit consists of the jet, or “air sheet” which exits at the pipe mouth. The resonator I. A MODEL OF COUPLED ORGAN PIPES is of course the pipe body; there, sound waves emitted at the labium (i.e., the sharp edge in the upper part of To obtain a deeper insight into the synchronization the pipe’s opening) travel up and down and can trigger a phenomena of two coupled organ pipes we model the regular oscillation of the air sheet. Energy is supplied by pipes by Van der Pol oscillators with delayed cross- arXiv:1708.06571v1 [nlin.AO] 22 Aug 2017 the generating unit, which is the pressure reservoir be- coupling: neath the pipe at a basically constant rate. Experimen- 2 tal and numerical investigations by Abel et al. [2] yield x¨i + ωi xi µ x˙ i f˙(xi) + κxj(t τ) = 0, (1) the conclusion that an organ pipe can be approximated − − − h i satisfactorily by a Van der Pol oscillator. Whereas the where i, j = 1, 2. These equations represent a harmonic work of Fischer [7] focuses on the nonlinearities in sound oscillator with an intrinsic angular frequency ωi, supple- generation and their effect on the synchronization prop- mented with linear and nonlinear damping of strength erties, in this paper we investigate the effect of the finite µ > 0. The nonlinear damping can be described by the distance of two coupled pipes which in turn is reflected nonlinear function by a delay in the coupling function. More specifically, γ f(x ) = x3, (2) i 3 i

where γ is the anisochronicity parameter and f˙(xi) = ∗ corresponding author: [email protected] 2 γxi x˙i. The coupling strength in Eq. (1) is κ, and the 2

 II. ANALYTIC APPROACHES width of sync-region A. Method of averaging 

The method of averaging (quasiharmonic reduction)  describes weakly nonlinear oscillations in terms of slowly varying amplitude and phase. For µ = 0 the uncoupled system reduces to the harmonic 2  oscillator x¨i + ωi xi = 0 with solution

͙ xi = Ri sin(ωit + φi), (4)  with constant amplitude R and phase φ . For 0 < µ 1 i i  ω-sync we look for a solution in the form Eq. (4) but assume that the amplitude R 0 and the phase φ are time-  i ≥ i dependent functions:

 xi = Ri(t) sin(ωit + φi(t)), (5) x˙ i = Ri(t)ωi cos(ωit + φi(t)).        where terms involving the slowly varying functions R˙i, Φ˙i are neglected. Without loss of generality, we choose ω2 = FIG. 1: Angular frequency Ω of oscillator x1 (dark 1. For small µ we use the method of averaging, assuming green circles), and oscillator x2 (light yellow circles) that the product µτ is small, and Taylor expand R (t τ) i − versus the detuning ∆ of the oscillators. Full (empty) and φi(t τ) in the following way: circles correspond to symmetric (non-symmetric) initial − conditions. Parameters: ω = 1, µ = 0.1, γ = 1, κ = 0.4, τ 2 2 R (t τ) = R (t) τR˙ (t) + R¨ (t) + ..., (6) τ = 0.1π. i − i − i 2 i

We introduce the phase difference ψ(t) = φ1(t) φ2(t). ˜ 2t − Defining a new time scale t = µ we find the equations coupling delay is τ. Since for synchronization the fre- which describe the system (1) on a slow time scale: quency difference of the two oscillators is important, we ˜ 2 introduce the detuning parameter ∆ R by R˙ (t˜) = R (t˜) 1 γR1/2(t) ∈ 1/2 1/2 − 4   κR2/1(t˜) sin(ψ(t˜) + τ), (7) 2 2 ∓ ω1 = ω2 + µ∆. (3) ˜ ˙ ˜ R1(t) ˜ ψ(t) = ∆ + κ ˜ cos(ψ(t) τ) − R2(t) − ˜h R2(t) ˜ In Fig. 1 we display frequency locking as obtained by nu- ˜ cos(ψ(t) + τ) . (8) − R1(t) merical simulation of Eq. (1) with symmetric initial con- i ditions . The observed angular frequencies Ω are plotted For simplicity, in the following we omit the tilda . The ∼ versus the detuning ∆ of two Van der Pol oscillators. A method of averaging together with truncation of the pronounced synchronization region and a sharp transi- Taylor expansion in τ reduces the infinite-dimensional tion to synchronization is observed. Within the synchro- problem to a finite-dimensional problem by assuming nization region, only for small ∆ the in-phase synchro- that the product µτ is small. This key step enables us | | nized solution is observed (lower branch), while for larger to handle the original delay differential equation as a ∆ the anti-phase synchronized solution (upper branch) system of ordinary differential equations [19, 22]. We | | is obtained even for symmetric initial conditions (full cir- now have two dynamical equations (7) for the amplitudes cles). Note that for small ∆ the anti-phase synchronized R1 and R2, and one equation (8) for the phase difference | | solution is also observed for any non-symmetric initial ψ(t) which is also called slow phase. The latter equation conditions (empty circles). Our purpose is to analyze is a generalized Adler equation [3] and contains the main the synchronization frequency, the width of the synchro- features of synchronization. nization region, the phase difference in the synchronized state, its stability, and the bifurcation scenarios which Generalized Adler equation - The equilibria of the occur at the boundaries. Adler equation correspond to the locking of phase and 3 3 frequency, since the difference between the phases is con- B ×10-3 stant. To investigate the stability and bifurcation sce- E nario of such fixed points we take a closer look at the 6 generalized Adler equation(8), written in general form: width of 4 sync-region ψ˙(t) = ∆ + κq(ψ(t)) (9) N(A)- G(iωs) − 2 where the averaged forcing term q(ψ(t)) is the 2π- NONLINEAR LINEAR periodic function -2 A q(ψ(t)) = R1(t) cos [ψ(t) τ] R2(t) cos [ψ(t) + τ] . R2(t) − − R1(t) -4 (10) -6 FIG. 3: Block diagram of a nonlinear feedback control The generalized Adler equation (8) is necessary for the loop consisting of a linear transfer function G(i!s) and calculation of the Arnold tongue (see below), which is one 0 1 2 3 4 5 6 anonlineardescribingfunctionN(A). A stands for the of the main characteristics of synchronization in nonlin- C feedback signal and E for the reference input, which is ear systems. For further analysis it is useful to eliminate ×10-2 added at the summation point (circle). BOBMZUJD
SFTVMUT the amplitudes Ri(t) from Eq. (10). Therefore we express OVNFSJD
SFTVMUT R2(t) by R1(t) in the case of a relative equilibrium of the 4 th amplitudes (R˙ i(t) = 0). We achieve two relevant solu- Gi(i!s) is called the transfer function of the i node and tions for the stationary amplitude. Inserting the values N(Ai) the describing function of the nonlinear control of Ri into Eq. (10) – according to the numerical results – 3 loop, where !s is the frequency of the limit cycle and we can plot ψ˙ versus ψ in Fig. 2a, which gives a graph of Ai its amplitude. i is the phase difference between the right-hand side of the generalized Adler equation(9), the oscillation of the limit cycle and the reference input, i.e., the function q(ψ) in the case ∆ = 0. Since in the syn- which has the amplitude Ei.Equation(12)holdsfor

XJEUI
PG
TZODSFHJPO 2 chronization region ψ˙ = 0 and hence ∆ = q(ψ), the maxi- a static, odd non-linearity and a transfer function which mum and minimum in Fig. 2a correspond to the border of behaves like a low pass filter. For the method of harmonic the synchronization tongue when varying ∆ as we can see 1 balance we use Eqs. (1) and (2): in Fig. 1. A change of ∆ shifts, according to Eq. (9), the 0 1 2 3 4 5 6 7 8 2 ˙ curve in the y-direction but does not change its shape. In x¨i + !i xi µ(˙xi f(xi))=µxj(t ⌧). (13) this way we can calculate the width of the synchroniza- FIG.FIG. 2: 2: (a) (a) The The right-hand right-hand side side of of the the Adler Adler equation LINEAR NONLINEAR tion tongue as a function of τ and compare these ana- (10) versus the slow phase . The difference between equation (9) ψ˙ = q(ψ) for zero frequency detuning lytic results to our numerical ones from the simulation of the maxima of the Adler equation gives us the border of The| Laplace{z transform} | {z (X}i(i!s)= xi(t) = ∆ = 0 versus the slow phase ψ. The difference between 1 i!st L{ } Eqs.(1) (see Fig. 2b). The agreement between the results the synchronization region. The frequency detuning is 0 xi(t)e dt)ofEq.(13)leadsto the maximum=0 ⌧ =0 and.1⇡ minimum = 1 =1 gives! the=1 width ofµ the=0.1 is remarkable, even though there is an unavoidable small , , 5 , , 1/2 ,and . R 2 2 synchronization tongue. Parameters: ω1 = ω2 = 1, ! i!sµ !  deviation because of the limited numerical accuracy of (b) Analytic and numeric results of the dependence of i s i!s⌧ µ = 0.1, γ = 1, κ = 0.2, τ = 0.1π. (b) Analytic (line) Xi + F (Xi)= ie Xj, (14) our simulation. The transient times are very large at bi- the width of the synchronization region on the delay i!sµ !s and numeric (dots) results2 for the width of the 1 furcation points. Furthermore we gain information about time ⌧ for  = 5 , =1, !2 =1,andµ =0.1. G (i! ) Ei synchronization tongue as a function of the delay time τ i s the stability of the synchronization state from Fig. 2a: for ω2 = 1, µ = 0.1, γ = 1, κ = 0.4. 3 | {z } | {z } For ψ = 0 (or, equivalently, ψ = 2π) we have an unstable where F (Xi) is the Laplace transform of the nonlinear equilibrium since, and for ψ = π (anti-phase oscillation) function f(xi). The transfer function G(i!s) is the char- ˙ ˙ B. Describing function method a stable B equilibrium since ψ < 0 for ψ > 0 and ψ > 0 E acteristic of the linear part of our system and can be read for ψ < 0, in accordance with experimental results [7], from Eq. (14): width of as discussed in Fig. 7a below. The experimentally ob- The describing function method (also called the sync-region i!sµ served decrease of the amplitude at ∆ = 0 indicates an method of harmonic balance) uses frequency domain G (i! )= . (15) N(A)- G(iω ) i s !2 i! µ !2 anti-phase oscillation [1]. techniques to investigate limit cycle behaviors in non- i s s linear systems, which typically is represented by a so NONLINEAR LINEAR The !2-term in the denominator guarantees the low pass called block diagram (Fig. 3). This engineering method s behavior. The describing function N(A) is the amplifi- B. Describing function method requires an approximation, but nevertheless often gives a A cation of the fundamental harmonics of the periodic sig- reliable prediction about the frequency !s and amplitude nal xi(t)=Ai sin(!st + i) by the nonlinear function The describing function method (also called the A of the limit cycle in a non-linear system. f(x )= x3 of our system: method of harmonic balance) uses frequency domain The condition for the harmonic balance, i.e. the limit i 3 i techniques to investigate limit cycle behavior in nonlin- FIG. 3: Block diagram of a nonlinear feedback control FIG.cycle, 3: of Block our system diagram can of a be nonlinear formulate feedback in following control way F (X ) f(x ) loop consisting of a linear transfer function G(i!s) and i i 2 ear systems, which is typically represented by a block di- loop[6, 7]: consisting of a linear transfer function G(iωs) and N(Ai)= = = Ai . (16) anonlineardescribingfunctionN(A). A stands for the Xi xi 4 agram (Fig. 3). This engineering method requires an ap- a nonlinear describing function N(A). A stands for the E proximation, but nevertheless often gives a reliable pre- feedback signal andE for the reference input, which is C feedback signal and for the referenceE input, which is The approximation of the describing function method is diction of the frequency ωs and amplitude A of the limit 1 added at the summationi pointii (circle). BOBMZUJD
SFTVMUT addedGi at(i! thes) + summationN(Ai) point= (circle).e ,i=1, 2. (12) based on the exclusive consideration of the fundamen- cycle in a nonlinear system. OVNFSJD
SFTVMUT Ai LINEAR NONLINEAR tal harmonics in Eq. (16). For its calculation we have G (i! ) is called the transfer function of the ith node and |i {zs } | {z } N(Ai) the describing function of the nonlinear control loop, where !s is the frequency of the limit cycle and Ai its amplitude. i is the phase difference between the oscillation of the limit cycle and the reference input, which has the amplitude Ei.Equation(12)holdsfor XJEUI
PG
TZODSFHJPO a static, odd non-linearity and a transfer function which behaves like a low pass filter. For the method of harmonic balance we use Eqs. (1) and (2): EFMBZ
UJNF x¨ + !2x µ(˙x f˙(x ))=µx (t ⌧). (13) i i i i i j FIG. 2: (a) The right-hand side of the Adler equation LINEAR NONLINEAR (10) versus the slow phase . The difference between the maxima of the Adler equation gives us the border of The| Laplace{z transform} | {z (X}i(i!s)= xi(t) = 1 i!st L{ } the synchronization region. The frequency detuning is 0 xi(t)e dt)ofEq.(13)leadsto =0 ⌧ =0.1⇡  = 1 =1 ! =1 µ =0.1 , , 5 , , 1/2 ,and . R 2 2 !i i!sµ !s i!s⌧  (b) Analytic and numeric results of the dependence of Xi + F (Xi)= ie Xj, (14) the width of the synchronization region on the delay i!sµ !s 2 time ⌧ for  = , =1, !2 =1,andµ =0.1. 1 Ei 5 Gi (i!s) | {z } | {z } where F (Xi) is the Laplace transform of the nonlinear function f(xi). The transfer function G(i!s) is the char- B. Describing function method acteristic of the linear part of our system and can be read from Eq. (14): The describing function method (also called the i!sµ method of harmonic balance) uses frequency domain G (i! )= . (15) i s !2 i! µ !2 techniques to investigate limit cycle behavior in non- i s s linear systems, which typically is represented by a so The !2-term in the denominator guarantees the low pass called block diagram (Fig. 3). This engineering method s behavior. The describing function N(A) is the amplifi- requires an approximation, but nevertheless often gives a cation of the fundamental harmonics of the periodic sig- reliable prediction about the frequency !s and amplitude nal xi(t)=Ai sin(!st + i) by the nonlinear function A of the limit cycle in a non-linear system. f(x )= x3 of our system: The condition for the harmonic balance, i.e. the limit i 3 i cycle, of our system can be formulate in following way F (Xi) f(xi) 2 [6, 7]: N(Ai)= = = Ai . (16) Xi xi 4 The approximation of the describing function method is 1 Ei ii Gi (i!s) + N(Ai) = e ,i=1, 2. (12) based on the exclusive consideration of the fundamen- Ai LINEAR NONLINEAR tal harmonics in Eq. (16). For its calculation we have | {z } | {z } 4 4 The condition for the harmonic balance, i.e., the limit where we introduced the phase difference = . cycle, of our system can be formulated in the following 6 1 2 way [10, 12]: 5 1 Ei i∆φi Gi− (iωs) + N(Ai) = e− , i = 1, 2. (11) Ai 4 LINEAR NONLINEAR

th 3 Gi|(iωs{z) is called} the| {z transfer} function of the i node and III. SYNCHRONIZATION PHENOMENA N(Ai) the describing function of the nonlinear control 2 loop, where ωs is the angular frequency of the limit cycle and A its amplitude. ∆φ is the phase difference between Arnold tongue - One important theoretical question is i i 1 the transition to synchronization, usually characterized the oscillation of the limit cycle and the reference input, in the parameter plane of frequency detuning and E 0 which has the amplitude i. Equation (11) holds for -0.02 -0.01 0.01 0.02 coupling strength . In the case of delayed coupling the a static, odd nonlinearity and a transfer function which time delay ⌧ has the same importance as the coupling behaves like a low-pass filter. For the method of harmonic  ( FIG. 4: The synchronization region in the parameter strength . The synchronization region in the , )or balance we use Eqs. (1) and (2): plane of delay time ⌧ and frequency detuning . The (⌧, )planeisgenerallycalledArnoldtongue,anditis plane of delay time τ and frequency detuning ∆. The plot shows an analytically computed Arnold tongue for one of the main characteristics of synchronizing nonlinear 2 ˙ plot showsthe coupling an analytically strength computed = 1 , =1 Arnold, ! =1 tongue,and for x¨i + ωi xi µ(x ˙ i f(xi)) = µκxj(t τ). (12) 5 2 systems. We analytically calculate the synchronization − − − the couplingµ =0.1 strength.Onecanseethesymmetric,κ = 0.2, ω2 = 1, µ =⇡-periodic 0.1, γ = 1.  LINEAR NONLINEAR region in the plane of the coupling strength and the One can see the symmetric,boundaries ofπ-periodic the tongue. boundaries of detuning . The delay time ⌧ is kept constant but does The Laplace| {z transform} | {z (X}i(iωs) = xi(t) = the tongue. not vanish. The nonlinear boundaries of the Arnold iωst L{ } tongue like in Fig. 7b is a remarkable result. Already ∞ xi(t)e− dt) of Eq. (12) leads to 3 0 used following trigonometric addition theorem: sin (⇣)= for a small delay time ⌧ this behavior can be observed. 1 R 2 2 4 [3 sin(⇣) sin(3⇣)].Accordingtotheright-handside Also the analytical calculation of the synchronization ωi iωsµ ωs iωsτ κ By applying Eqs. (14)-(16) to Eq. (11) we obtain − − Xi + F (Xi) = ie− Xj, (13) of Eq. (14) the reference input of one system is given region in the plane of the delay time ⌧ and the detuning iω µ ω s − s by the delayed2 output of the other system. The fac- π gives excellent agreement with numeric results. The i ωii!s⌧ γ 2 κAj⇡ i φ φ +ω τ+ 1 ie + ![ ⌧i j s 2 ] G− (iω ) Ei tor(ωs ) means1 + aA negative= 2 e− s −-shift in the time. (17) coupling strength  is kept constant but does not vanish. i s i µdomain,− ω sos that− the4 referenceωsA inputi in case of synchro- Instead of a monotonic nonlinear increase we have now A | {z } | {z } nization is given by j sin(! (t ⌧)+ ⇡ ). In the ⇡ where F (X ) is the Laplace transform of the nonlinear !s s j 2 a -periodic behavior at the boundaries of the tongue. i The imaginary part of Eq. (17) gives us information function f(x ). The transfer function G(iω ) is the char- case of the ith oscillator the right-hand side of Eq. (12) i s about the synchronization frequency ω versus the time acteristic of the linear part of our system and can be read yields s delay. Multiplying it by the imaginary part of the analog In- and anti-phase mode - Special solutions give the from Eq. (13): th E A ⇡ advantage to reduce the complexity of a problem. We equation for thei ji oscillator,j i[ wei obtain:j +!s⌧+ ] e = e 2 . (17) first investigate the case for a vanishing detuning =0, iωsµ Ai !sAi Gi(iωs) = . (14) 2 2 2 2 i.e., the central point of the synchronization region in 2 2 ωi ωj µ κ ωi iωsµ ωs Fig. 1. For =0the slow-phase Eqs. (8-9) possess the − − ωBys applyingω Eqs.s (15-17)= to Eq.2 (12)cos( weψ+ obtainωsτ) cos(ψ ωsτ), − ωs − ωs ! ωs − following equilibrium points which corresponds to the in- 2   2 The ωs -term in the denominator guarantees the low-pass i ! Aj ⇡ i 2 i[i j +!s⌧+ 2 ] (18) phase and anti-phase mode of Eq. (1): N(A) (!s ) 1+ Ai = e . (18) behavior. The describing function is the amplifi- whereµ we !s have introduced4 !sAi the phase difference cation of the fundamental harmonics of the periodic sig- ψ = φ1 φ2. nal xi(t) = Ai sin(ωst + φi) by the nonlinear function The− imaginary part of Eq. (18) gives us information γ 2 f(x ) = x3 about the synchronization frequency versus the time de- in =0 R1 = R2 = p1  sin ⌧, i 3 i of our system: The two analytic approaches in this section yield ! p lay. Multiplying it by the imaginary part of the analog (20) Eqs.equation (9) and for (18), the j respectively,th oscillator, we whereby obtain: we can get in- 2 F (Xi) f(xi) γ 2 anti = ⇡ R1 = R2 = p1+ sin ⌧. N(Ai) = = Ai . (15) formation about the phase difference and the frequency ! p Xi xi ≈ 4 2 2 2 2 of the coupled!i oscillators,!j respectively.µ  We use these re- !s !s = 2 cos( +!s⌧) cos( !s⌧), The approximation of the describing function method sults in the!s following ! section.s ! ✓ ◆ ! s is based on the exclusive consideration of the funda- (19) The Jacobian matrix J of Eqs. (8-9) is given by: mental harmonics in Eq. (15). For its calculation we have used the following trigonometric addition theorem: III. SYNCHRONIZATION PHENOMENA 3 1 3 2 sin (ζ) = 4 [3 sin(ζ) sin(3ζ)]. According to the right- 1 4 R1  sin( + ⌧) R2 cos( + ⌧) − 3 2 hand side of Eq. (13) the reference input of one system Arnold tongue - One important sin( theoretical⌧)1 question is R R1 cos( ⌧) J = 0 4 2 1 . (21) R2 cos( +⌧) cos( ⌧) cos( +⌧) R1 cos( ⌧) R2 sin( +⌧) R1 sin( ⌧) is given by the delayed output of the other system. The the transition to synchronization, 2 + usually characterized + 2  R R2 R1 R R1 R2 iωsτ π 1 2 factor ie− means a negative phase shift of + ω τ in the parameterB plane of frequency detuning ∆ and C − 2 s @ h i h i h i A in the time domain, so that the reference input in case of coupling strength κ. In the case of delayed coupling κAj sin(ω (t τ) + φ π ) the time delay τ has an importance comparable to the synchronization is given by ωs s j 2 . th − − couplingIn order strength to determineκ. The the synchronization stability of the in-phase region in and the points and calculate the eigenvalues, which satisfy the In the case of the i oscillator the right-hand side of anti-phase mode, we can evaluated J at the equilibrium Eq. (11) yields (κ, ∆) or (τ, ∆) plane is generally called Arnold tongue, and it is one of the main characteristics of synchronizing Ei i∆φ κAj i φ φ +ω τ+ π nonlinear systems. First, we keep the coupling strength e i = e− [ i− j s 2 ]. − (16) κ = 0 constant and calculate the synchronization region Ai ωsAi 6 5 analytically using the methods from the previous Section a qualitative and qualitative analysis of our problem. We in the plane of the delay time τ and the detuning ∆ find the following equilibrium points for Eqs. (7) and (8) (Fig. 4), in excellent agreement with the numeric results which in turn correspond to the in-phase and anti-phase from Eq. (1). The boundaries of the Arnold tongue are mode of Eq. (1): modulated periodically with a period π as τ is varied. Note that for our choice of ω2 = 1 the period of the 2π 2 uncoupled harmonic oscillator is . ψin = 0 R1 = R2 = √1 κ sin τ, ⇔ √γ − (19) In- and anti-phase mode - Eqs. (7) and (8) possess 2 ψanti = π R1 = R2 = √1 + κ sin τ. two equilibrium solutions, an in-phase and an anti-phase ⇔ √γ mode, as we will demonstrate below. As we recognize from Fig. 1 the center of the synchronization region plays a special role. This motivates a first investigation of In order to determine the stability of the in-phase and the solutions and their stability for vanishing detuning anti-phase mode, we linearize Eqs. (7),(8) around the ∆ = 0. Such a special parameter setting reduces the equilibrium points, which gives the Jacobian matrix J technical difficulties and nevertheless allows us to make of the system (7),(8):

3 2 1 4 γR1 κ sin(ψ + τ) κR2 cos(ψ + τ) − − 3 2 − J = κ sin(ψ τ) 1 4 γR2 κR1 cos(ψ τ) . (20)  R cos(ψ+τ) − cos(ψ τ) cos(ψ+−τ) R cos(ψ τ) R sin(ψ+τ) R− sin(ψ τ)  2 − 1 − 2 1 − κ R2 + R κ R + R2 κ R R  1 2 − 1 2 1 − 2   h i h i h i 

The eigenvalues λi, i = 1, 2, 3 of the Jacobian matrix The generic saddle-node bifurcation curves in the (κ, τ) evaluated at these equilibrium points determine their lin- plane is given by: ear stability. They are calculated from the characteristic equation κ1 = 0, 1 det(J λ I) = 0. (21) κ2 = , (26) − i −sin τ sin τ in dependence on the system parameters γ, κ, and τ. Let κ3 = . −1 + sin2 τ us first consider the in-phase mode ψin = 0 in Eq. (19): Note that κ1 = 0 represents an uncoupled system. The 2 (λ + 2 2κ sin τ)[6κ + λ(λ + 2) bifurcation curves κ in Eqs.(24) and (26) separate the − − (22) 2/3 2κ (κ cos 2τ + (3λ + 2) sin τ)] = 0. regions of stable and unstable equilibrium in the (κ, τ) plane for in-phase and anti-phase mode, respectively; This equation depends upon the two parameters κ and they are represented in Fig. 5(a), (b), respectively. By τ. The boundary of stability with respect to saddle-node fixing a value of κ, e.g., κ = 0.4 (horizontal dash-dotted bifurcations is given by the condition λ1 = 0, which de- line), one can trace the change of stability as τ is changed. fines the generic saddle-node bifurcation curves in the (κ, For τ = 0 Eqs. (22), (25) reduce to τ) plane: λ3 + 4λ2 + 4λ(κ2 + 1) + 8κ2 = 0 (27) κ κ sin τ 1 2κ sin τ + κ2(1 + sin2 τ) = 0. (23) − − which has no solution λ with positive real part, hence We obtain three solution branches for κ fulfilling
 Eq. (23): both equilibria are stable for τ = 0. The horizontal dash- dotted line κ = 0.4 intersects with the bifurcation curve κ = 0, sin τ 1 1 κ3 = 1+sin2 τ as shown in Fig. 5(a), and hence for 6 π < 5± 1 τ < π the in-phase mode becomes unstable (Re λi > κ2 = , (24) 6 sin τ 0), whereas the anti-phase mode becomes unstable for sin τ 7 11 6 π < τ < 6 π, see Fig. 5(b). In the remaining ranges κ3 = . 1 + sin2 τ of τ the in-phase and anti-phase modes, respectively, are stable. Note that bistability of in-phase and anti-phase In the case of the anti-phase mode ψin = π in Eq. (19) mode occur around τ = 0 and τ = π, as also visible in the characteristic equation (21) reads: Fig. 1 for τ = 0.1π. In Fig. 2(a) a smaller value κ = 0.2 is chosen, and hence for τ = 0.1π the in-phase mode is (λ + 2 + 2κ sin τ)[6κ2 + λ(λ + 2) − (25) unstable and no bistability exists, in full agreement with 2κ (κ cos 2τ (3λ + 2) sin τ)] = 0. Fig. 5. − 6

(a) ψ = 0 1.02 κ1 = 0 1.5 1 κ = 2 sin τ 1.01 1.0 sin τ κ κ3 = 1 + sin2 τ 0.5 s 1.00 ω 0.0 (b) ψ = π 0.99 κ1 = 0 1.5 1 κ = 2 −sin τ 1.0 sin τ κ κ3 = 0.98 −1 + sin2 τ 0 1 2 3 4 5 6 7 0.5 τ

0.0 FIG. 6: Comparison of the analytic (line) and numeric 0 1 2 3 4 5 6 (dots) results of the synchronization frequency ω τ s versus the delay time τ for ∆ = 0, ω1 = ω2 = 1, µ = 0.1, γ = 1, κ = 0.4. The analytic solution gives an in-phase FIG. 5: The saddle-node bifurcation curves (λ1 = 0) of (dark blue line) and an anti-phase mode (light green the characteristic equation (21)for ∆ = 0 in the plane of line), while the numeric solution (dots) with symmetric coupling strength κ and delay time τ given by Eq. (24) initial conditions only reproduces the in-phase mode. for the in-phase mode (a) and Eq. (26) for the Solid line means stable solution, whereas dashed line anti-phase mode (b). The horizontal dashed-dotted line stands for an unstable one as shown in Fig. 5. represents κ = 0.4 and its grey shaded part is the unstable region with λi > 0. The other parameters are ω1 = ω2 = 1, µ = 0.1, γ = 1. the single oscillator frequencies ωi = 1 in dependence upon the delay time. For small delay time, for instance, the in-phase oscillation frequency is lowered, while the Synchronization frequency - The describing function anti-phase oscillation frequency (light green line) is in- method yields Eq. (18) which determines the synchro- creased. The stability of the two branches changes as τ nization frequency ω , if the phase difference ψ is known. s is varied, as discussed above, see Fig. 5: At the extrema The method of averaging yields the generalized Adler of the frequency curve in Fig. 6 we can find bistability. equation (9) determining the dynamics of ψ. In Fig. 2a Note that for symmetric initial conditions the in-phase we have found numerically with the help of the Adler mode is found as numerical solution for all delay times equation a stable (ψ = π) and an unstable (ψ = 0) equi- although it is unstable in part of the τ range (but there librium point of ψ. In order to compare the numerical for any non-symmetric initial conditions the anti-phase simulations with the results of the describing function mode would be found). In Fig. 1 the upper frequency method, we set ω = ω = 1, i.e., ∆ = 0, in which case 1 2 branch in the synchronization region stays in the stable the equilibrium solutions of ψ are given by Eq. (19), and anti-phase mode for non-zero detuning ∆ (in congruence hence Eq. (18) can be simplified to with experimental data [1]), whereas the lower branch, 2 2 2 2 2 i.e., the stable in-phase mode, which is close to its insta- (ω 1) = µ κ cos (ω τ) (28) s − s bility point, is only observable in a small range of ∆. or

2 IV. COMPARISON WITH ACOUSTIC ωs = 1 µκ cos(ωsτ) (29) ± EXPERIMENTS where + and correspond to anti-phase and in-phase − oscillations, respectively. In Fig. 6 we plot the synchro- A comparison of a complex experiment with a sim- nization frequency ωs versus the time delay τ for ∆ = 0. ple oscillator model is an ambitious endeavor: On one The congruence between the numerical (from Eq. (1)) hand there is an organ pipe with a whole spectrum of and analytical result for the in-phase mode (dark blue overtones and a complicated aeroacoustic behavior, on line, from Eq. (29)) is excellent. It is remarkable that the other hand we consider a simple Van der Pol oscilla- the synchronization frequency ωs is modulated around tor. Nevertheless, such a simple model can already ex- 5 characteristic equation. For eigenvalue 0 =0,e.g.at the in-phase mode in Eq. (20), we have necessary condi- tions for generic saddle-node bifurcation curves: BOBMZUJD
SFTVMUT OVNFSJD
SFTVMUT   sin ⌧ 1 2 sin ⌧ + 2(1 + sin2 ⌧) =0(22) We obtain⇥ three values for  fulfilling Eq. (22):⇤

1 =0, 1  = , (23) 2 sin ⌧ sin ⌧ TZODISPOJ[BUJPO
GSFRVFODZ 3 = . 1+sin2 ⌧

1 =0represent a uncoupled system. Considering the other values of ,wecancalculatetheothereigenvalues. EFMBZ
UJNF

ϰ For 2: FIG. 5: Comparison of the analytic and numeric results  2 = 2(1 i cot ⌧), (24) of e synchron f ency versus the delay 1/2 ± and 3: 7 2 3 = , 1 1+sin2 ⌧ B C B C (25) 100 750 100 0 750 100 2(1 2sin2 ⌧) 750 90 90 3 = . 80 1.04 80 0.3 80 2 2 735 70 735 70 1+sin ⌧ 60 60 60 Ω 20 50 50 40 40 0.2 40 1.0 720 720 Numerical simulations prove these results. They attest 720 30 30

20 20 SPL (dB) 20 SPL (dB) SPL (dB) the range of validity for our analytical approach. 10 40 0.1 10 0 705 0 705 0 0.96 square-root −10 −10

Observed frequency ν (Hz) Observed frequency ν (Hz) behavior Synchronization frequency - The describing function −20 −20 Pipes distance d ( c m) -10 -5 5 10 -0.02−20 -0.01 0.01 0.02 690 −30 690 $ f (Hz) −30 690 −10 0 −10 0 method yields Eq. (19) where the information about the Observed frequency ν (Hz) Detuning ∆ f (Hz) Detuning ∆ f (Hz) -10Δf (Hz)10 -0.04 0.04 synchronization frequency !s still is kept. But we have (b) FIG. 8: Comparison(c) of experiment and theory: Arnold no information about the phase difference within this FIG. 7: Comparison of synchronization region in tongue in the plane of coupling strength κ vs. detuning method. The method of averaging yields to the Adler experiment750 and theory: (a)100 experimentally750 observed 100∆: (a)750 experiment [7], where100 the coupling strength is equation (10) determining the dynamic of . In Fig. 2a 90 90 90 sound pressure level (SPL) in80 the plane of observed 80 given by the distance d of the80 organ pipes, and (b) we have a pair of a stable and an unstable fixed point: 735 70 735 70 735 70 frequency ν vs. frequency detuning60 ∆f (in Hz) [7] and 60 analytic result for ω2 = 1, µ =60 0.1, γ = 1, and delay For =0we see – according to the numerical results – 50 50 τ = 0.1π 50 (b) numerically calculated angular40 frequency Ω vs. 40 time . 40 the unstable one and for = ⇡ the stable one. In order to 720 IV. COMPARISON WITH720 ACOUSTIC 720 dimensionless detuning ∆ for30 ω2 = 1, µ = 0.1, γ = 1, 30 30 compare the numerical simulations with the results of the EXPERIMENTS20 SPL (dB) 20 SPL (dB) 20 SPL (dB) κ = 0.4, τ = 1.1π. 10 10 10 describing function method, we set !1 = !2 =1where 705 0 705 0 705 0 −10 −10 −10 we know already two solutions for the phase difference Observed frequency ν (Hz) Observed frequency ν (Hz) Observed frequency ν (Hz) A comparison of a complex−20 experiment with a sim- −20layed coupling. To understand−20 theoretically the dynam- 690 −30 690 −30 690 −30 from Eq. (20), such that Eq. (19) can be simplified to −10 0 10 20 −10 0 −10 0 10 20 ple oscillatorDetuning model ∆ f (Hz) is an ambitious endeavor:Detuning ∆ f On (Hz) the ics of the system,Detuning we∆ f (Hz) have analyzed the locking scenarios. hibit complicated dynamical scenarios, as demonstrated one side there is a pipe with a whole spectrum of over- Further, we have numerically integrated the system, and 2 2 2 2 above. Qualitatively,(d) these scenarios agree well(e) with the (f) (!s 1) =  cos (!s⌧) (26) have found that the solution agrees well with existing ex- tones and a complicated aeroacoustic behavior, and on experiment,the other side as we a Van-der-Pol show by comparing oscillator. the Nevertheless graphs below. a periments. On this basis, we have systematically varied Consequently,750 our model provides100 a profound750 comprehen- 100 750 100 and compared to numerical results. In Fig. 5, we plot the simple model can provide a90 deeper comprehension of a 90 the coupling parameters, namely90 the coupling strength κ sion of the dynamical behavior80 observed in organ pipes. 80 80 the synchronization frequency !s against the time delay complex735 system. We compare70 the synchronization735 region 70 and the735 coupling delay τ. 70 This supports our point of60 view that complex behavior 60 60 ⌧. The congruence between the numerical and analytical in Fig. 6. There are many similarities50 between the plots. 50 For a deeper understanding of50 the various bifurcation sce- 40 40 40 can emerge720 from simple, low-dimensional720 systems, as it 720 results is excellent. It is remarkable that the synchroniza- Especially the behavior at the30 transition region between 30 narios we have developed and30 extended two complemen-

is generally accepted in dynamical20 SPL (dB) systems theory. 20 SPL (dB) 20 SPL (dB) tion frequency !s switches to a high or low frequency for frequency detuning and locking is remarkable, as well as tary analytical approaches: By the method of averaging 10 10 10 For visual705 comparison of experiment705 and theory, we show 705 aspecialdelaytime⌧, which is higher or lower than each the convex curvature of the synchronization0 regions itself. 0 we obtain a generalized Adler0 equation for the phase dy- −10 −10 −10 theObserved frequency ν (Hz) synchronization region versus theObserved frequency ν (Hz) detuning frequency Observed frequency ν (Hz) of the single oscillator frequencies !i. In Fig. 1 the upper In Fig. 7 we compare an experimental−20 and an analytically −20namics, which allows us to study−20 the stability of the equi- 690 −30 690 −30 690 −30 frequency branch in the synchronization region stays in ofcalculated the two−10 Arnold oscillators0 tongue.10 in Fig. Experimentally, 7. Both−20 plots−10 the show0 coupling10 very −20 −10 0 10 Detuning ∆ f (Hz) Detuning ∆ f (Hz) libria correspondingDetuning ∆ f (Hz) to frequency locking of the oscilla- the stable anti-phase mode (in congruence with experi- similarstrength features. is determined Especially by the the distance behavior of the of the two transi- organ tors. However, the averaging method does not provide in- mental data [1]), whereas the lower branch ( 0)stays tionpipes. regions The mostat the(g) considerable two boundaries similarity of the is locking the(h) nonlinear interval (i) ⇡ formation about the frequency in the locking region. The in the unstable in-phase mode. isbehavior remarkable, of their as boundaries. well as the We concave are able curvature to change of the the frequency, in contrast, can be found by the describing synchronizationAbbildung 2.7.: regionKarten itself. der In Entwicklung Fig. 8 we compareder Frequenz the und des Pegels des gekoppelten Zwei- Pfeifen Systems für die Pfeifenabstände a) 1 mm, b) 10 mm,function c) 30 mm, method d) 50 mm, which e) allows75 mm, us to determine the syn- experimentally observed and the analytically calculated f) 100 mm, g) 200 mm, h) 300 mm, i) 400 mm. Auf der y-Achsechronization ist die beobachtete frequency Frequenz and hence explain the curvature κ Arnoldaufgetragen. tongue in Auf the der plane x-Achse of the ist couplingdie Verstimmung strength des autonomenand of the Systems frequency aufgetragen. vs. detuning Die Pegel which is found in the numer- ∆ thesind detuning farbkodiert.. Experimentally, the coupling strength ical simulations as well as in the experiments (see Fig. 7). is determined by the distance of the two organ pipes. Altogether these approximations provide a detailed and The analytical calculation proceeds as described in Sect. complete analytic picture of both relative phase and fre- II.ne As Frequenzkopplung a result we obtain (Modenkopplung) an Arnold tongue with der 1. nonlin- Harmonischenquency. bei In kleinen general und we obtain mittleren excellent agreement of our ear,Abständen curved boundaries, zu sehen. Die see FrequenzenFig. 8b. This der is Orgelpfeifen a remark- synchronisierenanalytic results with in einem the numerical gewis- simulations and with able result which occurs already for a small delay time experiments. τ senand Bereich coincides kleiner well with Verstimmung. experiments Die [7, 8], Ausdehnung see Fig. 8a. dieses Bereiches ist abhängig vom Abstand der Orgelpfeifen zueinander. Stehen die PfeifenA sehr detailed dicht bifurcation nebeneinander analysis (Ab- has affirmed the existence The curvature of the boundaries may be further adjusted of in- and anti-phase synchronization. In each case the instand the calculationsd =1mm), by erstreckt replacing sich the der constant Bereichκ derby aSynchronisationτ- synchronization über etwa frequency6 Hz. Dabei has a different value which is dependentwird die Frequenz coupling strength der Pfeifeκ(τ P1), see um Appendix. etwa 10 Hz mitgenommen.in perfect Bei accordance größerenAbständen with our analytic calculations. The nimmt die Breite des Synchronisationsbereichs ab. Bei einembehavior Abstand of the der boundaries Orgelpfeifen of the Arnold tongue in the von 400 mm ist die Synchronisation nur noch marginal. Dieserplane of Abstand coupling liegt strength im Bereichκ and detuning ∆ depends on der WellenlängeV.0 CONCLUSION=476mm der autonom betriebenenthe Orgelpfeife interplay of mit the der coupling Frequenz strength and the coupling von 720 Hz. Aus den gewonnenen Ergebnissen lässt sichdelay schließen, time κ dass. In es general, sich bei the den nonlinear interdependence In this paper we have investigated the synchroniza- of κ and τ leads to curved boundaries in the (κ, ∆) plane, tion of organ pipes using the tools of nonlinear dynam- which is also clearly confirmed by experimental data. ics. Particular attention has been paid to the delay in It is interesting to note that there is some similarity the coupling which naturally occurs due to a finite dis- of our delayed coupling with15 the viscoelastic coupling, tance of pipes. We have used a simplified nonlinear os- which has been used in a recent study of two modi- cillator model for the pipes, i.e., two coupled Van der fied Van der Pol oscillators with the aim to describe Pol oscillators which interact by a dissipative, direct, de- cardiac synchronization [21]. This viscoelastic coupling τ ~ � ~ � � � � �+� � �−�−�� � � �= + � �−� � � • ��

� = ,, �, � � • � � τ

•

� =.�=.�=.

�=. �=.�

Δ

� � •

•

� �

� �

τ 8

Δ� = � − � We are grateful to Jost Fischer and Natalia Spitha for B C 0 τ Δ 

m) fruitful discussions. c 2  Δω • � � τ � = .6; � = . 4 • „observed“  • 6 τ

Pipes distance d ( APPENDIX • Δω     -0.03 0.03 $ f (Hz)Δ




ͅ � � � values .4; Δω � = .; � = .6 FIG. 9: Comparison of experiment and theory: Arnold For a more refined modeling of the non-monotonic be- tongue in the plane of coupling strength κ vs. detuning havior of the Arnold tongue as observed in experiment, • ∆: (a) experiment [7], where the coupling strength is see Fig. 9a for d = 5cm, instead of a constant coupling given by the distance d of the organ pipes, and (b) factor κ as in Eq. (1), a coupling strength κ(τ) which de- • τ τ ω = 1 µ = 0.1 γ = 1 � � numerical result of Eq. (31) with 2 , , , pends on the delay time τ, should be used. The coupling � � κ1 = κ2 = 0.04. The red arrows in (a) indicate the is delayed, because the sound travels a certain distance • non-monotonic behavior of the Arnold tongue. d between the pipes. The coupling strength depends on � that distance, since the sound wave is attenuated accord- • ing to the radiation of a spherical wave emitted from the Master‘s Thesis, Technische Universität Berlin (2014). is modeled within the Maxwell model of viscous creep pipe mouth. Within the coupling strength κ(τ) we have 1 1 by an additional differential equation describing a har- a near-field term ( 2 ) and a far-field term ( ) with ∝ τ ∝ τ monic spring in series with a linear damper of damping coefficients κ1, κ2 > 0: rate (viscosity) γ. This linear inhomogeneous differen- tial equation can be eliminated using a Green’s function approach, thereby introducing a distributed delay in the κ1 κ2 κ(τ) = + . (30) coupling of the two oscillators with an exponential de- τ 2 τ lay kernel with decay rate γ corresponding to a temporal By replacing κ in Eq. (1) by Eq. (30) memory [14–16]. In this viscoelastic model also in-phase and anti-phase synchronization scenarios were found. x¨ + ω 2x µ x˙ f˙(x ) + κ(τ)x (t τ) = 0, (31) i i i − i − i j − ACKNOWLEDGMENTS h i

This work was partially supported by Deutsche we are able to model the boundaries of the Arnold tongue Forschungsgemeinschaft in the framework of SFB 910. more realistically (see Fig. 9b).

[1] M. Abel, S. Bergweiler, and R. Gerhard-Multhaupt: Syn- [7] J. Fischer: Nichtlineare Kopplungsmechanismen akus- chronization of organ pipes: experimental observations tischer Oszillatoren am Beispiel der Synchronisation von and modeling, J. Acoustical Society of America 119, Orgelpfeifen, Ph.D. thesis, Universität Potsdam (2014). 2467–2475 (2006). [8] J. Fischer: Aeroacoustical coupling and synchronization [2] M. Abel, K. Ahnert, and S. Bergweiler: Synchronization of organ pipes, J. Acoust. Soc. Am. 140, 2344–2351 of sound sources, Phys. Rev. Lett. 103, 114301 (2009). (2016). [3] R. Adler: A study of locking phenomena in oscillators, [9] N. H. Fletcher: Mode locking in nonlinearly excited in- Proc. IEEE 61, 1380 (1973). harmonic musical oscillators, J. Acoust. Soc. Am. 64, [4] R. Bader: Nonlinearities and Synchronization in Musical 1566–1569 (1978). Acoustics and Music Psychology, (Springer, 2013). [10] O. Föllinger: Nichtlineare Regelungen 2: Harmonische [5] S. Bergweiler: Körperoszillation und Schallabstrah- Balance, Popow- und Kreiskriterium, Hyperstabilität, lung akustischer Wellenleiter unter Berücksichtigung von Synthese im Zustandsraum: mit 18 Übungsaufgaben mit Wandungseinflüssen und Kopplungseffekten: Verändern Lösungen (De Gruyter, 1993). Metalllegierung und Wandungsprofil des Rohrresonators [11] V. Flunkert, I. Fischer, and E. Schöll: Dynamics, control den Klang der labialen Orgelpfeife?, Ph.D. thesis, Uni- and information in delay-coupled systems: An overview, versität Potsdam (2006). Philos. Trans. R. Soc. A 371, 20120465 (2013). [6] B. Fabre, A. Hirschberg: Physical modeling of flue in- [12] H. Gholizade-Narm, A. Azemi, and M. Khademi: Phase struments: A review of lumped models, Acustica - Acta synchronization and synchronization frequency of two- Acustica 86, 599–610 (2000). coupled van der Pol oscillators with delayed coupling, Chinese Physics B 22, 070502 (2013). 9

[13] M. S. Howe: Theory of vortex sound, vol. 33 (Cambridge [19] V. Semenov, A. Feoktistov, T. Vadivasova, E. Schöll, and University Press, 2003). A. Zakharova: Time-delayed feedback control of coher- [14] Y. N. Kyrychko, K. B. Blyuss, and E. Schöll: Amplitude ence resonance near subcritical Hopf bifurcation: theory death in systems of coupled oscillators with distributed- versus experiment, Chaos 25, 033111 (2015). delay coupling, Eur. Phys. J. B 84, 307 (2011). [20] D. Stanzial, D. Bonsi, and D. Gonzales: Nonlinear mod- [15] Y. N. Kyrychko, K. B. Blyuss, and E. Schöll: Amplitude elling of the Mitnahme-Effekt in coupled organ pipes, In- and phase dynamics in oscillators with distributed-delay ternational symposium on musical acoustics (ISMA), Pe- coupling, Phil. Trans. R. Soc. A 371, 20120466 (2013). rugia, Italy, 333–337 (2001). [16] Y. N. Kyrychko, K. B. Blyuss, and E. Schöll: Synchro- [21] S. Stein, S. Luther, and U. Parlitz: Impact of viscoelastic nization of networks of oscillators with distributed-delay coupling on the synchronization of symmetric and asym- coupling, Chaos 24, 043117 (2014). metric self-sustained oscillators, New J. Phys. 19, 063040 [17] A. Pikovsky, M. Rosenblum, and J. Kurths: Synchroniza- (2017). tion: a universal concept in nonlinear sciences, vol. 12 [22] S. Wirkus and R. Rand: The dynamics of two coupled (Cambridge University Press, 2001). Van der Pol oscillators with delay coupling, Nonlinear [18] J. W. S. Rayleigh: On the pitch of organ-pipes, Phil. Mag. Dynamics 30, 205–221 (2002). XIII, 340–347 (1882).