Synchronization of Huygens' clocks and the Poincare method
SSynchronization of Huygens’ clocks and the Poincar´e method
Vojin JovanovicSystems, Implementation & IntegrationSmith Bits, A Schlumberger Co.1310 Rankin RoadHouston, TX 77073e-mail: [email protected] Sergiy KoshkinComputer and Mathematical SciencesUniversity of Houston-DowntownOne Main Street,
Abstract
We study two models of connected pendulum clocks synchronizing their oscillations, a phenomenonoriginally observed by Huygens. The oscillation angles are assumed to be small so that the pendulumsare modeled by harmonic oscillators, clock escapements are modeled by the van der Pol terms. Themass ratio of the pendulum bobs to their casings is taken as a small parameter. Analytic conditions forexistence and stability of synchronization regimes, and analytic expressions for their stable amplitudesand period corrections are derived using the Poincar´e theorem on existence of periodic solutions inautonomous quasi-linear systems. The anti-phase regime always exists and is stable under variationof the system parameters. The in-phase regime may exist and be stable, exist and be unstable, ornot exist at all depending on parameter values. As the damping in the frame connecting the clocksis increased the in-phase stable amplitude and period are decreasing until the regime first destabilizesand then disappears. The results are most complete for the traditional three degrees of freedom model,where the clock casings and the frame are consolidated into a single mass.
Keywords : Huygens’ clocks, synchronization, autonomous quasi-linear systems, periodic solutions a r X i v : . [ m a t h - ph ] D ec Introduction
In 1665 Huygens reported that two pendulum clocks attached to a common frame tended, after a while,to run in unison with their pendulums always pointing in mutually opposing directions. He eventuallyattributed this ”sympathy” to imperceptible motions of the frame that transfer motion from one clockto the other. With some modifications, Huygens’ experiments were repeated by Ellicott, Euler andPoisson, among others, who observed similar behavior, see descriptions in [1]. Original attempts to givea quantitative description of this sympathy were based on linear theory. Korteweg in a paper of 1905 [1]showed that the motion of the linearized system decomposes into three normal modes, two of which involvesizable oscillations of the frame. He then argued heuristically that damping in the frame siphons energyaway from these two modes, eventually leaving only the mode responsible for the anti-phase synchrony.While an improvement over Huygens’ intuitive explanation, Korteweg’s theory left all non-linear effectsof clock escapements and coupling out of his quantitative analysis. Nonetheless, upgraded variations ofmodal analysis appear in some recent approaches to the problem [2, 3].By the end of the 19th century coordinated behavior in systems of interacting self-excited oscillatorswas observed in many other acoustic, electric and mechanical systems, the phenomenon later came tobe known as synchronization. About the same time, motivated by applications to celestial mechanics,Poincar´e adapted the method of small parameter to finding periodic solutions to non-linear systems.Somewhat later, the Poincar´e method was applied to synchronization problems by Blekhman and others,see [4] and references therein, providing a first non-linear theory of the effect. Since the method requiresequations of motion to be analytic in phase variables Blekhman used the simplest analytic model of aself-excited oscillator, the van der Pol model, to describe clock escapements. His analysis predicted notonly the previously observed anti-phase synchronization, but also the in-phase synchronization, when thependulums always point in the same direction. According to Blekhman, both regimes always co-existin the same system being triggered by different initial conditions, he also reported observing both inexperiments.In the last decade interest in Huygens’ sympathy was renewed by the work of Bennett et al. [2], whotook great care to reproduce Huygens’ experiments and developed a new non-linear model to describethem. Their work builds on Korteweg’s modal analysis, but equations of motion include terms modelingclock escapements with impulsive kicks. Unlike Blekhman’s their model does not predict a stable in-phasesynchronization, but it does predict asymptotic vanishing of oscillations in one of the clocks, a phenomenonthey call beating death. This is a direct consequence of having a threshold angle for escapements to engage,if at some point the amplitude of oscillations falls below it energy loss forces oscillations to damp out. Theauthors even expressed puzzlement over Blekhman’s results since neither they nor Huygens ever observedthe in-phase synchronization. However, Czolczynski et al. [5] later confirmed co-existence of both regimesexperimentally using modern clocks with non-impulsive escapements. Numerical simulations in [6, 7] alsoshow co-existence for some parameter values. The common trait in all three cases is that the escapementsinvolved do not have an engagement threshold.We should also mention that the in-phase synchronization was observed by Pantaleone [8] and others[9, 10] after replacing pendulums with metronomes. Moreover, for small damping in the frame the in-phase synchronization was the only one observed, in contrast to Huygens’ setup. The reason is that2etronomes swing with amplitudes too large to be approximated by harmonic oscillators, changing thesituation entirely. Existence of the the in-phase synchronization for large amplitudes was proved in [11]using the van der Pol model for escapements.Our main purpose in this paper is to apply the Poincar´e method to the Huygens system and tocompare predictions it makes to known observations. Our application of the Poincar´e method builds onthe original work of Blekhman [4]. As already mentioned, one obvious limitation of the method is thatall terms in the equations have to be analytic, so realistic escapements with impulsive kicks can not beaccomodated. However, van der Pol approximation appears to be rather accurate as long as the pendulumamplitudes stay above the engagement threshold, see [8]. On the other hand, the method offers severalimportant benefits. First, there is a natural small parameter in the system, the mass ratio of pendulumbobs to the frame, which incorporates heavy clock casings. Second, one can take into account not onlyescapements bit also the non-linear coupling term, which remains after pendulum equations are partiallylinearized for small angles. In [2] and [6] this term had to be dropped because both approaches rely onthe system’s evolution being linear between escapement kicks. However, its magnitude is comparable tothe linear terms kept. Finally, the Poincar´e method identifies synchronization regimes rigorously withoutheuristic pre-simplifications required for modal analysis, and gives analytic expressions for their stabilityranges and amplitudes.One of the method’s key predictions, namely co-existence of the in-phase and the anti-phase regimes,appears to be at variance with some observations. We will show in Sections 3, 4 that this prediction isnot due to the method’s limitations, but arises from treating damping in the frame as a small parameter.In Section 3 we assume both coupling of the pendulums to the frame and damping in the frame to besmall. To avoid neglecting the second derivative in the frame’s position, we transform the system into afirst order form in a way different from that in [4]. However, our conclusions are the same in this setting:the in-phase and the anti-phase regimes co-exist for the same values of system parameters.On the other hand, once damping in the frame is treated as a regular parameter in Section 4 one seesthat the two regimes are not created equal. While the anti-phase synchronization remains robustly stableunder variation of its values, the in-phase regime is only stable when the damping is very small relativeto coupling. As the damping gets larger, it destabilizes and then disappears altogether. Although thePoincare theorem, as any asymptotic result, does not prescribe what ”small” is, simulations in Section 5suggest that in Huygens’ experiments values of damping were outside of the co-existence range. In otherwords, damping in the frame is a bifurcation parameter and experimental values for it can be both belowor above the bifurcation values. To illustrate the point, numerical simulations in Section 5 are comparedto experimental results in [2, 5].As in [2], additional clues are provided by a preliminary linear analysis in Section 2 on the system ob-tained by disregarding all non-linear terms. Although synchronization is often believed to be a non-linearphenomenon, it turns out not to be entirely so. Even the linear system has the sum of pendulum anglesconverge to zero, which is an anti-phase synchronization of sorts. Of course, due to linearity stable am-plitudes depend on initial conditions, it is their convergence to invariant values that is a non-linear effect.Also, in the linear model pendulums started with exactly equal phases have their oscillations asymptoti-cally damp out. Although van der Pol terms alter this asymptotic outcome, numerical simulations showthat the system may initially follow the linear pattern.3n addition to the traditional three degrees of freedom model, where the clock casings and the con-necting frame are modeled by a single mass, we also consider a four degrees of freedom system, whereeach casing is represented by a separate mass (Section 6). In the linear approximation this system reducesto just three degrees of freedom because the angle difference between the pendulums decouples from therest of the variables. Similar system but with a piecewise-linear escapement term was considered in [6],where existence of a stable anti-phase synchronization is proved for some parameter range.
We start with a general setup of n identical pendulums (’clocks’) of mass m and length l attached to arigid frame of mass M , which is restricted to move horizontally. In its turn, the frame is attached toa fixed support by a weightless spring with stiffness k and damping c , Figure 1. Pendulum angles withthe vertical are denoted by θ i and position of the frame is given by x . Horizontal restriction reflects theoriginal setup of Huygens, where clocks were hanging off a wooden beam resting on the backs of twochairs. Experiments show that behavior of the system changes substantially if non-horizontal motion isallowed, and synchronization may not occur at all [5]. Equations of motion can be derived using the .... Figure 1: Rigid frame with pendulums attached to a fixed support.Lagrangian approach, see e.g. [9], (cid:40) ml ¨ θ i + mgl sin( θ i ) + ml ¨ x cos( θ i ) = D i , i = 1 , ..., n ( M + nm )¨ x + c ˙ x + kx + ml (cid:80) ni =1 ¨ θ i cos( θ i ) = ml (cid:80) ni =1 ˙ θ i sin( θ i )Here D i := D ( θ i , ˙ θ i ) denote driving torques provided by the clock escapement mechanisms. The Poincar´emethod, which is our method of choice, requires all terms to be analytic in their variables. FollowingBlekhman [4], we choose the simplest analytic term that admits a limit cycle, the van der Pol term D ( θ, ˙ θ ) := e ( γ − θ ) ˙ θ . Here e measures the strength of the escapement and γ is the critical angle, atwhich the escapement switches from boosting to damping. This model does not always work at smallamplitudes due to engagement threshold and impulsive kicks in some escapements, but on average gives4 good approximation when the pendulums are close to their limit cycles [8]. We also assume thatoscillation angles remain sufficiently small so that the pendulums can be approximated by harmonicoscillators. Taking sin θ i ≈ θ i and cos θ i ≈ (cid:40) ml ¨ θ i + mglθ i + ml ¨ x = D i , i = 1 , ..., n ( M + nm )¨ x + c ˙ x + kx + ml (cid:80) ni =1 ¨ θ i = ml (cid:80) ni =1 ˙ θ i θ i . Rescaling the time as τ = (cid:114) gl t and introducing a dimensionless frame position variable y = xl we get adimensionless system (cid:40) ¨ θ i + θ i + ¨ y = F i , i = 1 , ..., n ¨ y + σ ˙ y + Ω y + β (cid:80) ni =1 ¨ θ i = β (cid:80) ni =1 ˙ θ i θ i , (1)where F i = D i mgl , σ = c ( M + nm ) (cid:112) g/l , Ω = kl ( M + nm ) g and β = mM + nm . Note that the only remain-ing non-linearities are on the right hand side of (1), namely the van der Pol terms and the non-linearcoupling term β n (cid:88) i =1 ˙ θ i θ i . Before proceeding with analysis of this system it will be instructive to investigatea linear version obtained by simply dropping the right hand sides in (1). The resulting linear system (cid:40) ¨ θ i + θ i + ¨ y = 0 , i = 1 , ..., n ¨ y + σ ˙ y + Ω y + β (cid:80) ni =1 ¨ θ i = 0 (2)is used in [2] to attempt a simple explanation of Huygens’ observations. Asymptotic behavior of thissystem can be analyzed using a Lyapunov function. It turns out that the sum of pendulum anglesasymptotically vanishes. Theorem 1.
Suppose that all coefficients in (2) are non-negative and γ > . Let θ i = θ i ( t ) be solutionsto it with arbitrary initial conditions, then θ + · · · + θ n −−−→ t →∞ .Proof. Set θ := θ + · · · + θ n , then adding the first n equations in (2) we get (cid:40) ¨ θ + θ + n ¨ y = 0¨ y + σ ˙ y + Ω y + β ¨ θ = 0 . (3)Define the Lyapunov function by E := β ( θ + ˙ θ ) + n (Ω y + ˙ y ) + 2 βn ˙ θ ˙ y = βθ + ( β − nβ ) ˙ θ + n Ω y + n ( β ˙ θ + ˙ y ) . Then E ≥ β − nβ ≥ ≤ β ≤ n . Given the definition of β above this holds even withstrict inequalities. Moreover, if 0 < β < n then E = 0 if and only if ˙ θ = θ = ˙ y = y = 0, i.e. E is positivedefinite. On the other hand, by (3)˙ E := 2 β ˙ θ (¨ θ + θ ) + 2 n ˙ y (¨ y + Ω y ) + 2 βn (¨ θ ˙ y + ˙ θ ¨ y ) = − βn ˙ θ ¨ y − n ( β ¨ θ ˙ y + 2 γ ˙ y ) + 2 βn (¨ θ ˙ y + ˙ θ ¨ y ) = − nγ ˙ y .
5y the LaSalle invariance principle any trajectory of (3) asymptotically converges to a trajectory containedin ˙ E − (0) since ˙ E ≤
0. But for γ > y = 0, hence ¨ y = 0 and y = const.The system then implies ¨ θ = − θ = const and ˙ θ = θ = 0, so any trajectory within ˙ E − (0) does not leavethe origin. Therefore, y −−−→ t →∞ θ = θ + · · · + θ n −−−→ t →∞ Corollary.
Let n = 2 in Theorem 1 and set δ ( t ) := θ ( t ) − θ ( t ) . Then asymptotically (cid:40) θ ( t ) ∼ A sin( t + φ ) θ ( t ) ∼ A sin( t + φ + π ) with A = (cid:113) δ (0) + ˙ δ (0) and sin φ = δ (0) /A .Proof. Subtracting the equation for θ in (2) from the one for θ we have ¨ δ + δ = 0. Therefore, δ ( t ) = A sin( t + φ ) with A and φ as in the statement of the corollary. Since asymptotically θ ( t ) ∼ − θ ( t ) byTheorem 1 it follows that θ ( t ) ∼ δ ( t ) ∼ − θ ( t ).(a) (b)Figure 2: Simulation of (1) with σ = 1 . β = 0 . = 0 . γ = 0 .
122 and e/mgl = 0 . θ ( t ) and θ ( t ) (indistinguishable at this resolution) superimposed, (b) θ ( t ) + θ ( t ). Linear pattern isfollowed for about 150 cycles.Thus, pendulum angles asymptotically align in anti-phase for any initial values. The difference withtrue synchronization is that the stable amplitude A depends on initial values. In particular, if thependulums are started at exactly in-phase, δ (0) = ˙ δ (0) = 0, then A = 0 and they will both asymptoticallydamp out. Even if initial phases are not exactly the same the asymptotic amplitude can be very small.Numerical simulations show that for large σ and small coupling β the non-linear system may follow this6inear pattern for many cycles before the van der Pol terms break the trend, see Figure 2. If an impulsiveescapement is used instead, and the engagement threshold for it is above this asymptotic amplitude, onecan expect that one or both pendulums will stop beating in finite time. Poincar´e method relies on non-linear terms entering equations being multiplied by a small parameter. Inpractice, this can be interpreted as some dimensionless parameter of the system being ’sufficiently’ small.In our case, it is natural to assume that β = mM + nm is small since masses of the pendulum bobs m aremuch smaller than the consolidated mass M of the clock casings and the frame. We will also assume thatthe escapement strength ε := emgl of the van der Pol term F i = ε ( γ − θ i ) ˙ θ i is small for the purposes ofthe Poincare theorem (which may in fact be large in practical terms). Eliminating ¨ y from the pendulumequations in (1) we get (cid:40) ¨ θ i + θ i − σ ˙ y − Ω y = F i − β − nβ F, i = 1 , ..., n ;¨ y + σ ˙ y + Ω y = β − nβ F , (4)where F := − n ( σ ˙ y + Ω y ) + n (cid:88) i =1 (cid:16) θ i + ˙ θ i θ i − F i (cid:17) . This transformation is needed to reduce the system to a first order form. In system (4) µ := β − nβ = mM + nm − nmM + nm = mM is a convenient choice for the small parameter. Since there can only be one small parameter we shallassume that ε = µa and set F i := a ( γ − θ i ) ˙ θ i , so that F i = µ F i . The system transforms into (cid:40) ¨ θ i + θ i − σ ˙ y − Ω y = µ ( F i − F ) , i = 1 , ..., n ;¨ y + σ ˙ y + Ω y = µF , (5)and its Poincar´e analysis for n = 2 will be our main goal in the next section.In this section however, we wish to consider a simplified case when σ is also considered small. Wewill often refer to σ = c ( M + nm ) (cid:112) g/l as ”damping in the frame”, although it is also obviously affectedby lenghts and masses of the pendulums and mass of the frame. Aside from serving as a stepping stoneto the general case, the case of small σ compares directly to the treatment in Chapter 5 of [4]. Similarextra assumption is made there, but system (1) is not transformed into (4), making it necessary to neglectthe backreaction of the frame on the pendulums given by ¨ y . In line with the above, we set σ = µb and7ubsume the ˙ y term in (5) into the right hand side. After neglecting the terms of order µ we finallyarrive at (cid:40) ¨ θ i + θ i − Ω y = µ ( F i − F ) , i = 1 , ..., n ;¨ y + Ω y = µ F , (6)with F i as above and F := − b ˙ y − y + n (cid:88) i =1 (cid:16) θ i (cid:17) θ i . System (6) can be rewritten in the first order form as ˙ x = Ax + µ Φ ( x ), where x = ( θ , ˙ θ , . . . , θ n , ˙ θ n , y, ˙ y ) T .From now on we only consider the case of two pendulums, n = 2. Then explicitly A = −
00 0 0 1 0 00 0 −
00 0 0 0 0 10 0 0 0 − Ω and Φ = F − F F − F F . The corresponding linear generating system (see Appendix for the terminology) is obtained by setting µ =0 in (6). Poincar´e theorem gives sufficient conditions for existence and stability of periodic solutions to theoriginal non-linear system that converge to periodic solutions of the generating system when µ →
0. Thelatter can be found most easily by diagonalizing A = VΛV − . In our case, Λ = diag ( i, i, − i, − i, i Ω , − i Ω)and x = V ξ with ξ = ( ϕ , ϕ , ψ , ψ , u, v ) T given by ϕ k = (cid:16) iθ k + ˙ θ k (cid:17) +
12 Ω Ω − ( iy + ˙ y ) ψ k = (cid:16) − iθ k + ˙ θ k (cid:17) +
12 Ω Ω − ( − iy + ˙ y ) u = ( i Ω y + ˙ y ) , v = ( − i Ω y + ˙ y ) . (7)Let us further assume that the system is non-resonant, i.e. the natural frequency of the frame Ω is not aninteger or a half-integer. In fact, it is realistic to expect | Ω | (cid:28)
1, meaning that the clocks oscillate muchfaster than the frame. Making the corresponding substitution in non-linear terms, namely f ( ξ ) = V − Φ( V ξ ) , (8)we get the diagonalized system ˙ ξ = Λ ξ + µ f ( ξ ) . (9)The general solution to the generating system is ˙ ξ = Λ ξ is ϕ k = α k e it , ψ k = β k e − it , u = C e i Ω t , v = D e − i Ω t .It is not periodic since Ω may be irrational, and, in any case, the periodic solutions we are interestedin should reduce to independent oscillations of the pendulums when the coupling is removed, i.e. when8 = 0. To put it differently, we are looking for periodic solutions with the amplitude of frame motionof order at most µ . This means that we are looking for non-linear counterparts of the solutions with C = D = 0.Physically, parameter µ measures the strength of the non-linear coupling of the pendulums to theframe. The generating system describes two pendulums attached to an immovable frame, so they aredecoupled. Once the frame is allowed to move the coupling is turned on and the behavior changes. It turnsout that the pendulums eventually settle into oscillating synchronously with limit amplitudes determinedby their escapement mechanisms. Depending on the initial phases this settled motion can be in-phaseor anti-phase. In the anti-phase regime the frame remains almost motionless once the synchronizationoccured, the pendulums behave as if they were nearly decoupled. More sizable motions are required tosustain the anti-phase regime, speeding up the oscillations compared to decoupled pendulums. The nexttheorem makes this description more precise providing explicit formulas for limit amplitudes and periodsto the first order in µ . It also implies that both regimes reemerge after small disturbances. Theorem 2.
For small µ > , a, b > and a non-integer system (6) with n = 2 admits both in-phaseand anti-phase synchronized periodic solutions with limit amplitude γ , where γ (cid:54) = 0 is the critical anglein the van der Pol escapement. Aside from the trivial one, these are the only solutions that converge when µ → to pendulums independently oscillating with equal amplitudes, and motionless frame. The period ofthe anti-phase solution is T ( µ ) = 2 π + o ( µ ) , and that of the in-phase solution is T ( µ ) = 2 π − γ − Ω µ + o ( µ ) .Both solutions are stable.Proof. In this proof we use notation and terminology associated with the Poincar´e theorem, see Appendix.Recall that the diagonalized matrix of the generating system is Λ = diag ( i, i, − i, − i, i Ω , − i Ω). We wantour periodic solutions to reduce to oscillations of decoupled pendulums for µ = 0 meaning that only α , . . . , α in (A.1) should be non-zero. Therefore, we must choose the first four eigenvalues to form theleading special group. The remaining two then belong to non-special groups, and the secondary specialgroup is empty by assumption about Ω.Our next task is to simplify constraint (A.3) from the Poincar´e theorem for our case. By definition, ξ = ( ϕ , ϕ , ψ , ψ , u, v ) T and the relevant components of ξ are the first four. Since x = V ξ must be a realsolution one can see from (7) that φ k and ψ k must be complex conjugates of each other for k = 1 ,
2. Inview of (A.1) we have α k +2 = α k , and we represent α k = r k e iφ k . Here r k is the half-amplitude of the k thpendulum and φ k is its phase. One can set φ = 0 without loss of generality since time can be shifted bya constant, then synchronization is characterized by a single phase φ = φ . By assumption about equalamplitudes set r = r = r . After some computations one can reduce (A.3) to a system of equations for r and φ , namely (cid:0) γ − r (cid:1) r e iφ = 0 γ (cid:0) γ (cid:1) sin( φ ) = 0 r (cid:0) r (cid:1) (cid:0) e iφ − (cid:1) = 0 . (10)Discarding the trivial solution r = 0, the only remaining ones are r = γ , φ = 0 and r = γ , φ = π ,corresponding to the in-phase and the anti-phase synchronization respectively. Substituting the values of α s corresponding to each solution into (A.5) we get the first order period corrections.9tability condition (A.6) reduces to the following equations having only roots with negative real parts φ = 0 : ( κ + aγ ) (cid:16) (1 − Ω ) κ + aγ (1 − Ω ) κ + (1 + γ ) (cid:17) = 0 ; φ = π : ( κ + aγ ) (cid:16) (1 − Ω ) κ + aγ (1 − Ω ) κ + 3 γ + 4 γ + 1 (cid:17) = 0 . This condition is equivalent to all the coefficients of the above polynomials being positive, which happensif a > γ (cid:54) = 0 and Ω (cid:54) = ±
1. Stability condition (A.8) for non-special eigenvalues reduces to b > µ is small. We believe that these are the only solutions to the Poincar´e system(A.3) even without the assumption of equal amplitudes. However, we were unable to solve it for r , r and φ without assuming that r = r . We can only show that if φ = 0 or φ = π is assumed in advancethen r = r follows from the equations. Based on simulations, we also believe that for small µ these arethe only attractors, i.e. the trivial solution is spurious (no periodic solutions with amplitudes of order µ ),and no quasiperiodic or chaotic behavior occurs.We see that for small σ there holds a counter-intuitive symmetry: both synchronization regimesalways coexist and have the same stable amplitudes. This conclusion coincides with Blekhman’s in [4]even though he neglects the second derivative of the frame’s position and applies the Poincar´e theoremto system (1) directly. However, such coexistence of regimes was not observed in some experiments, see[2]. We shall see in the next section that this symmetry is a deceptive consequence of σ being small,increasing it sufficiently eliminates the in-phase regime altogether. In this section we no longer assume that damping in the frame is small and investigate how its magnitudeinfluences synchronization. It is this assumption, it turns out, that puts the two synchronization regimeson an equal footing in Theorem 2. Recall that (6) was obtained from (5) after assuming that σ is small.Without this assumption, after neglecting the terms of order µ , we get instead (cid:40) ¨ θ i + θ i − σ ˙ y − Ω y = µ ( F i − F )¨ y + σ ˙ y + Ω y = µ F (11)with F i := a ( γ − θ i ) ˙ θ i and F := − n ( σ ˙ y + Ω y ) + n (cid:88) i =1 (1 + ˙ θ i ) θ i .
10s in the previous section, we rewrite this as a first order system ˙ x = Ax + µ Φ ( x ), where x =( θ , ˙ θ , . . . , θ n , ˙ θ n , y, ˙ y ) T . For two pendulums one has A = − σ − σ − Ω − σ and Φ = F − F F − F F The diagonalized matrix is Λ = diag ( i, i, − i, − i, − (cid:16) σ − √ σ − (cid:17) , − (cid:16) σ + √ σ − (cid:17) ), we omitthe diagonalizing transformation V .Behavior of the system is most conveniently described in terms of a modified damping parameter (cid:101) σ := σa ( (1 − Ω ) + σ ) . For small values of Ω and σ it is close to the ratio of damping σ to the escapementstrength a . According to the theorem below the system undergoes two transitions as σ and hence (cid:101) σ increase. For values smaller than (cid:101) σ := γ γ ) both synchronization regimes coexist and are stable. Thisagrees with the conclusions of the previous section. The settled frequency of the in-phase oscillationsdecreases as the damping in the frame grows, but still remains higher than for the anti-phase oscillations.Once the above value is exceeded the in-phase regime destabilizes, but remains in the picture untilthe value (cid:101) σ := γ is attained. At this point the in-phase regime disappears altogether. Parameter valuesin the experiments of Bennett et al. [2] (and presumably Huygens) seem to fall within this last range.This means that if clocks are started with nearly equal phases they will eventually settle into anti-phaseoscillations with amplitudes determined by the escapement mechanism. This agrees with observationsand indicates that σ should not be treated as a small parameter when the Poincare method is applied. Theorem 3.
For small µ > and a, σ > system (11) with n = 2 admits anti-phase synchronized periodicsolutions with the limit amplitude γ , where γ (cid:54) = 0 is the critical angle in the van der Pol escapement.In-phase synchronized periodic solutions can only exist if (cid:101) σ < γ , where (cid:101) σ := σa (cid:16)(cid:0) − Ω (cid:1) + σ (cid:17) − , andhave amplitude (cid:113) γ − (cid:101) σ (cid:101) σ . Aside from the trivial one, these are the only solutions that converge when µ → to pendulums independently oscillating with equal amplitudes, and motionless frame. In-phasesolutions exist and are stable at least if the following inequalities hold: either aσ < and (cid:101) σ < γ γ ) , or aσ > and aσ − aσ γ < (cid:101) σ < γ γ ) . The period of the anti-phase solutions is T ( µ ) = 2 π + o ( µ ) , and thatof the in-phase solutions is T ( µ ) = 2 π − (1 − Ω )(1 + γ )(1 − Ω ) + 2 σa + σ µ + o ( µ ) . Proof.
The general outline of the proof is the same as in Theorem 2. The leading special group is againformed by the first four eigenvalues and the remaining two are non-critical for σ >
0. We wish to simplify11onstraint (A.3) from the Poincar´e theorem. After setting α k = r k e iφ k and taking into account that α k +2 = α k , r = r = r , φ = 0 and φ = φ one can obtain: e iφ (cid:0) γ − r (cid:1) − (cid:101) σ e iφ (cid:0) r (cid:1) − (cid:101) σ (cid:0) r (cid:1) = 0 (cid:2) i (cid:0) Ω − (cid:1) − σ (cid:3) (cid:0) r (cid:1) e − iφ − (cid:2) i (cid:0) Ω − (cid:1) + σ (cid:3) (cid:0) r (cid:1) e iφ +2 (cid:104)(cid:0) − Ω (cid:1) + σ (cid:105) (cid:0) γ − r (cid:1) a − σ (cid:0) r (cid:1) = 0 (cid:0) σ + i (cid:0) − Ω (cid:1)(cid:1) (cid:0) − e − iφ (cid:1) (cid:0) r (cid:1) = 0 . Last equation immediately implies that φ = 0 or φ = π . Substituting these values into the first equationwe get the desired half-amplitudes r for the in-phase and the anti-phase regimes. The second equationis then satisfied automatically. Expressions for periods are obtained by substituting the values of α s corresponding to each solution into (A.5).Stability condition (A.6) for φ = π reduces to the equation( κ + aγ ) (cid:16) κ + a (cid:0) (1 + 4 (cid:101) σ ) γ + 2 (cid:101) σ (cid:1) κ + a (cid:101) σσ (1 + γ ) (cid:0) γ ( aσ + 3) (cid:1)(cid:17) = 0having only roots with negative real parts. For this to hold, all the coefficients must be positive. One cansee by inspection that they are as long as a, σ > γ (cid:54) = 0. The same stability condition for φ = 0 issomewhat more involved: (cid:0) κ + a ( γ − (cid:101) σ ) (cid:1) (cid:16) κ + a γ − (cid:101) σ (2 + γ )1 + 2 a (cid:101) σ κ + a (cid:101) σ (1 + γ ) σ (1 + 2 (cid:101) σ ) (cid:0) aσ (cid:101) σ + γ (1 − aσ ) (cid:1)(cid:17) = 0 . If (cid:101) σ < γ γ ) and aσ <
1, or aσ > aσ − aσ γ < (cid:101) σ < γ γ ) then the coefficients in the secondparentheses are positive. In both cases (cid:101) σ < γ is automatically satisfied, so that the coefficient in the firstparentheses is also positive. Other stability conditions (A.7,A.8) do not apply here since the remainingeigenvalues are non-critical.Note that the no resonance condition on Ω is no longer necessary because σ > Numerical simulations
In this section we use numerical simulations to illustrate the existence and stability conditions of Theorem3. We also wish to demonstrate that the theory based on the Poincar´e method can explain, at least(a) (b)Figure 3: In-phase synchronization for parameters of Czolczynski et al. with θ (0) = 0 .
25 and θ (0) = 0 . θ ( t ) (lighter) and θ ( t ) (darker) superimposed, (b) θ ( t ) − θ ( t ).qualitatively, conflicting reports on the type of synchronization observed in experiments. Correspondingly,we tried to use the values of parameters from the experiments of Bennett et al. [2] and Czolczynski et al.[5]. Unfortunately, in the case of the former the information on these values is insufficient.First three pairs of graphs correspond to the values g = 9 . m = 0 . c = M = 11 . k = 1 . l = 0 .
269 from [5], we picked γ = 0 .
122 for the critical angle and ε = 5 .
047 for the escapement strength.Time is measured in the number of cycles of a decoupled harmonic oscillator with the same parameters.One easily checks that aσ < (cid:101) σ < γ γ ) in the notation from Theorem 3, hence both synchronizationregimes exist and are locally stable. When the initial values are close (Figure 3) the system synchronizesin-phase. As they are pulled apart, the asymptotic regime switches to anti-phase, but it is preceded by thedepicted period of beats (Figure 4). Finally, when the initial values are sufficiently far apart (Figure 5),the anti-phase synchronization is achieved faster and without beats. This is in line with the observationsof Czolczynski et al. In all three cases the stable amplitudes are nearly equal to the limit values given byTheorem 3.Next, we tried to simulate the experiments of [2] with g = 9 . m = 0 . l = 0 .
14. Unfortunately,there is no information on the values of stiffness k or damping c . The mass M was adjustable in [2] andwe show results for two different choices of its value. We kept γ = 0 .
122 and assigned the rest of theparameters as indicated in the captions. In the case of Figure 6 the second inequality from Theorem 3fails and the in-phase regime is unstable. The system synchronizes in anti-phase but only after ratherviolent beating (individual oscillations are not visible on the figure because of a large time interval). The13a) (b)Figure 4: Anti-phase synchronization with beats for parameters of Czolczynski et al. with θ (0) = 0 . θ (0) = 0 .
5. (a) θ ( t ) (lighter) and θ ( t ) (darker) superimposed, (b) θ ( t ) + θ ( t ).(a) (b)Figure 5: Anti-phase synchronization for parameters of Czolczynski et al. with θ (0) = 0 . θ (0) = − .
3. (a) θ ( t ) (lighter) and θ ( t ) (darker) superimposed, (b) θ ( t ) + θ ( t ).14a) (b)Figure 6: Anti-phase synchronization for parameters of Bennett et al., M = 9 . c = k = M , with θ (0) = 0 .
25 and θ (0) = 0 .
3. (a) θ ( t ) (lighter) and θ ( t ) (darker) superimposed, (b) θ ( t ) + θ ( t ).(a) (b)Figure 7: In-phase synchronization for parameters of Bennett et al., M = 6 . c = k = 0 . θ (0) = 0 . θ (0) = 0 .
3. (a) θ ( t ) (lighter) and θ ( t ) (darker) superimposed, (b) θ ( t ) − θ ( t ).15nset of synchronization takes much longer than in the previous cases in line with the observations ofBennett et al. In fact, we adjusted the missing parameters to match the onset times from [2]. Dependingon the escapement threshold one may observe either the anti-phase synchronization or the vanishing ofoscillations in one of the pendulums in this case. To stabilize the in-phase regime we had to decreasethe damping by more than an order of magnitude (Figure 7). We do not believe that such a value of c is realistic for the experiments of [2] (or of Huygens) explaining why they never observed the in-phasesynchronization. One can see that the onset of the in-phase synchronization can also be preceded bybeats.As one can see, Theorem 3 provides some theoretical guidance to observing coexistence of synchro-nization regimes, or lack thereof, in experimental setups. First, the van der Pol parameters a and γ shouldbe adjusted to match the system as closely as possible, e.g. along the lines of [8]. Then the inequalities aσ < (cid:101) σ < γ γ ) give the parameter range, where both types of synchronization can be expectedto manifest (the other pair of inequalities from Theorem 3 calls for unrealistically large values of damping σ ). When γ γ ) < (cid:101) σ < γ one can expect beats to precede the onset of anti-phase synchrionization dueto the presence of unstable in-phase regime. Starting with Korteweg most work on the Huygens’ problem was based on variations of the three degreesof freedom model we considered above. In it the casings of the clocks and the connecting body (beam,platform) are consolidated into a single massive frame. However, it is the casings that make the framemassive, the connecting body itself is actually quite light [2, 5, 8]. In this section we consider a refinementof this model suggested by Dil˜ao [6], where casings are modeled by two separate masses M , and theconnecting body is modeled by a massles spring of stiffness k and damping c , see Fig.8. The rigid frameof the previous model (with mass 2 M and detached from a fixed support) is recovered when k → ∞ .Each mass carries its own pendulum of length l and mass m , and, as before, the system is restricted tomove horizontally. The positions of the casings are given by x , x and the pendulum angles are θ , θ .Unlike Dil˜ao, who uses a piecewise-linear model for escapements, we continue to use the van der Pol terms D ( θ, ˙ θ ) := e ( γ − θ ) ˙ θ so that the Poincare method can be applied. Assuming small pendulum angles asbefore, the partially linearized equations of motion are ml ¨ θ + mgl θ + ml ¨ x = D ml ¨ θ + mgl θ + ml ¨ x = D ( M + m )¨ x + c ˙ x − k ( x − x ) + ml ¨ θ = ml ˙ θ θ ( M + m )¨ x + c ˙ x + k ( x − x ) + ml ¨ θ = ml ˙ θ θ with D i := D ( θ i , ˙ θ i ). Note, that Dil˜ao drops the non-linear terms on the right in the last two equations.The reason is the same as for his choice of a piecewise-linear escapement. Dil˜ao’s approach requiresevolution of the system to be linear as long as θ i stay below a critical angle, at which the escapmentswitches from boosting to damping. In contrast, the Poicare method only requires that the non-linear16igure 8: Pendulums with clock casings connected by elastic frame.terms be analytic. Rescaling time as in Section 2, and introducing dimensionless parameters x = ly and x = ly we obtain a dimensionless system (compare to (1)): ¨ θ + θ + ¨ y = F ¨ θ + θ + ¨ y = F ¨ y + σ ˙ y − κ ( y − y ) + β ¨ θ = ˙ θ θ ¨ y + σ ˙ y + κ ( y − y ) + β ¨ θ = ˙ θ θ , where F i = D i mgl = emgl ( γ − θ i ) ˙ θ i , σ = c ( M + m ) √ g/l , κ = kl ( M + m ) g and β = mM + m . We choose our smallparameter to be µ := β − β = mM and set emgl = µa as before. Separating terms multiplied by µ anddiscarding higher order terms we transform the system into the form ¨ θ + θ − σ ˙ y + κ ( y − y ) = µ (cid:16) a ( γ − θ ) ˙ θ − F (cid:17) ¨ θ + θ − σ ˙ y − κ ( y − y ) = µ (cid:16) a ( γ − θ ) ˙ θ − F (cid:17) ¨ y + σ ˙ y − κ ( y − y ) = µ F ¨ y + σ ˙ y + κ ( y − y ) = µ F , (12)where F = − σ ˙ y + κ ( y − y ) + θ (cid:16) θ (cid:17) and F = − σ ˙ y − κ ( y − y ) + θ (cid:16) θ (cid:17) .17ystem (12) can be rewritten in the first order form ˙ ξ = A ξ + µ Φ ( ξ ), where ξ = ( θ , ˙ θ , θ , ˙ θ , y , ˙ y , y , ˙ y ) T and A = − κ σ − κ
00 0 0 1 0 0 0 00 0 − − κ κ σ − κ − σ κ
00 0 0 0 0 0 0 10 0 0 0 κ − κ − σ and Φ = a ( γ − θ ) ˙ θ − F a ( γ − θ ) ˙ θ − F F F The diagonalized matrix is Λ = diag (cid:16) i, i, − i, − i, − (cid:16) σ − √ σ − κ (cid:17) , − (cid:16) σ + √ σ − κ (cid:17) , − σ, (cid:17) , weomit the diagonalizing transformation. Note that one of the eigenvalues is zero, which indicates thepresence of a rigid body mode. Indeed, one can see directly from Fig.8 that the system as a whole is freeto move rigidly. In the original model this was prevented by attaching the frame to a fixed support. Butsince this attachment, represented by Ω in (11), is a minor influence on synchronization we dropped ithere for simplicity. As a compensation, we will only consider solutions that do not excite the rigid bodymode.As the next theorem shows, qualitatively the behavior does not differ much from the case with therigid frame. The in-phase regime eventually disappears, albeit at somewhat larger values of σ . The effectof the frame stiffness κ is similar in nature to that of the attachment stiffness Ω in Theorem 3. Oneshould also keep in mind that M in the one mass frame model is twice its value in the two mass modelhere. This leads to an extra coefficient of 2 in amplitude formulas and µ/ µ . With this inmind, the case of Theorem 3 for Ω = 0 is obtained in the limit κ → ∞ , i.e. the limit of rigid frame. Theorem 4.
For small µ > and a, σ > system (12) admits anti-phase synchronized periodic solutionsonly if γ > σ an , where σ an := σa (cid:16) (2 κ − + σ (cid:17) − , with amplitude (cid:113) γ − σ an σ an and period T ( µ ) = 2 π − (2 κ − γ )(2 κ − + σa + σ µ o ( µ ) . It admits in-phase synchronized periodic solutions only if γ > σ in , where σ in := σa (cid:0) σ (cid:1) − , withamplitude (cid:113) γ − σ in σ in and period T ( µ ) = 2 π − γ σa + σ µ o ( µ ) . As before, γ (cid:54) = 0 is the critical anglein the van der Pol escapement. The proof is similar to that of Theorems 2 and 3 and we omit it. The only structural difference is thatthe 0 eigenvalue has to be included into the special leading group, but the corresponding equation from(A.3) is vacuously satisfied. Note the correspondence between σ in and (cid:101) σ from Theorem 3. Note also that σ an < σ in for large κ , meaning that the range of existence of the anti-phase regime is wider than of the18n-phase one, with disparity increasing as κ → ∞ . Unfortunatley, we were unable to simplify the stabilityconditions to a tractable form in this case. But, by a continuity argument from Theorem 3, at least forlarge κ the in-phase regime should destabilize at smaller values of σ than the anti-phase regime. We used the Poincar´e method to study models of two connected clocks synchronizing their oscillations,a phenomenon originally observed by Huygens. The oscillation angles are assumed to be small so thatthe pendulums are modeled by harmonic oscillators, and the mass ratio µ of the pendulum bobs to theircasings is taken as the small parameter of the Poincar´e method. Clock escapements are modeled by thevan der Pol terms. The results are most complete for the three degrees of freedom model, where theconnecting frame and the clock casings are consolidated into a single mass. The partial results on thefour degrees of freedom model with casings represented by separate masses do not alter the qualitativepicture.The anti-phase synchronization regime is dominant in the sense that it is universally stable undervariation of the system parameters. In this regime (to the first order in µ ) the frame is motionless, and thestable amplitudes and periods of the pendulums are the same as for the decoupled van der Pol oscillators.The in-phase synchronization regime may exist and be stable, exist and be unstable, or not exist atall depending on parameter values. As the damping in the frame, or more precisely the dimensionlessparameter σ = c ( M +2 m ) √ g/l , is increased the in-phase stable amplitude and period are decreasing untilthis regime first destabilizes and then disappears. Analytic conditions for existence and stability ofsynchronization regimes, and analytic expressions for their stable amplitudes and period corrections arederived based on the Poincar´e theorem. In particular, they provide explicit relations between escapement’sstrength and critical angle, and damping in the frame, that distinguish experimental setups supportingand not supporting the in-phase synchronization.Our results vindicate the intuitive explanation of Huygens’ sympathy due to Korteweg, that the in-phase linear mode damps out faster than the anti-phase mode, because it requires more sizable motionsof the frame, unless the damping is kept small on purpose. Interaction between the two modes explainsthe beats observed by Ellicott and others [1]. Beats are most likely to be observed when the in-phaseregime is present but unstable.The following remarks are based on numerical simulations and are therefore somewhat speculative.When both regimes coexist, even if the in-phase one is unstable, the system may undergo a long periodof beats before settling asymptotically. This is most likely to happen when the initial values are near theboundary between the basins of attraction of the regimes. In the course of beating the pendulum anglesmay fall significantly below their initial and asymptotic values. If escapement models with engagementthreshold are used instead of the van der Pol terms the beats may cause one of the clocks to stop infinite time. When the damping is particularly heavy the system goes through a linear stage of anti-phaseoscillations with asymptotic amplitude dependent on the initial values. If the engagement threshold isabsent however, the clocks eventually synchronize in anti-phase with stable amplitudes. Thus, despite thelimitations of the van der Pol model for escapements, we can account for all types of behavior observed19n experiments [2, 5] and, to some extent, predict quantitatively when they occur. Appendix: Poincar´e method
For convenience of the reader we briefly describe a method of studying periodic solutions to sytems ofquasi-linear equations that we use in this paper. More detailed accounts can be found in [12], Ch.IX and[4], Ch. 10, proofs are given in [13], Ch.V.Consider a first order autonomous quasi-linear system ˙ x = Ax + µ Φ ( x ) and assume that the l × l matrix A can be diagonalized as A = VΛV − by a nonsingular linear transformation V . Set F ( ξ ) = V − Φ ( V ξ ),then a change of variables x = Vy transforms the system into the form ˙ y = Λy + µ F ( y ). The linearsystem ˙ x = Ax , as well as ˙ y = Λy , is called the generating system . We are interested in the periodicsolutions to the original system that converge to periodic solutions to the generating system when µ → A , the diagonal elements of Λ , are purely imaginary, andthey are stable only if A has no eigenvalues with positive real parts. From now on we assume thatall eigenvalues have non-positive real parts, and call purely imaginary ones critical . In synchronizationproblems one is typically interested in particular solutions to the generating system, which correspond tosynchronous motion of decoupled oscillators.Let λ = iω be a critical eigenvalue. Following Blekhman [13, 4], we collect all eigenvalues of the form λ s = in s ω with integer n s into the leading special group , and assume, without loss of generality, that theyoccur for s = 1 , . . . , k . Then the most general periodic solution of the generating system with period T = πω is given by y ◦ s = (cid:40) α s e in s ωt , s = 1 , . . . , k , s = k + 1 , . . . , l , (A.1)where α s are complex amplitudes. According to the usual method of small parameter, one now looks forsolutions y = y ( t, µ, α, β ), with y s = y ◦ s + β s for t = 0, to the original system as multivariable powerseries in µ, α, β . If such solutions are periodic the period will be of the form T ( µ ) = T (1 − δ ( µ )), where δ ( µ ) = δ µ + o ( µ ), since T (0) = T by assumption.Existence of periodic solutions to the original system now reduces to whether one can find power seriesthat also satisfy y ( T ( µ ) , µ, α, β ) − y (0 , µ, α, β ) = 0 . (A.2)Indeed, by autonomy this implies that y is T ( µ ) periodic. This is a system of l analytic equations with l + 1 unknowns µ, β , . . . , β l , so one extra condition on β ’s can be imposed. The system is solvable onlyif certain consistency constraints are satisfied. It turns out that µ factors out of the first k equationsand by setting their free terms to 0 one obtains consistency constraints on δ and the amplitudes α s .Eliminating δ we finally end up with k − k amplitudes. Again, one extra conditioncan be imposed on α ’s, e.g. α k − = α k . This freedom reflects the possibility of shifting time in solutionsto autonomous sytems. Thus, although the linear system has periodic solutions for arbitrary values of α s , non-linear periodic solutions only exist for some special collections of their values. Once the values of α s are found from (A.3) the first order period correction δ can also be determined, see (A.5).20f one is interested, as we are, in real-valued systems, then A is real and its eigenvalues come incomplex conjugate pairs. The leading special group will contain both elements of a pair or none, so k = 2 m and without loss of generality, λ m + s = λ s for s = 1 , . . . , m . Generating periodic solutions willalso be real if and only if α m + s = α s .Stability of a periodic trajectory can be investigated by looking at the solution to the matrix equation˙ Z = ( A + µ ∇ Φ ( x )) Z with Z (0) = I , where ∇ Φ ( x ) is the derivative of the non-linear term computed alongthe trajectory. This is the system of variations. Stability is determined by eigenvalues ρ ( µ ) of Z ( T ( µ )),e.g. it suffices that they were less than one by absolute value. For µ = 0 we obviously have Z ( t ) = e t A ,so ρ s (0) = e λ s T and one can look for these eigenvalues in the form ρ s ( µ ) = e λ s T (1 + κ µ + o ( µ )). Fornon-critical eigenvalues | ρ s (0) | < | ρ s ( µ ) | < µ by continuity. Forcritical eigenvalues we will have | ρ s ( µ ) | < µ > κ ) < inω with n an integer. The point is that stability conditions decouple intoseparate conditions for each group. Recall that λ s = in s ω with integer n s form the leading special group.The secondary special group is λ s = i ( n s + ) ω , and the remaining non-special groups have eigenvaluesof the form λ s = iν ( r ) s , where r enumerates non-special groups. After heavy duty computations theconditions we described can be brought into a convenient form below, which is due to Blekhman [4]. Theorem ( Poincar´e ) . Let ˙ y = Λy + µ F ( y ) be an autonomous quasi-linear system with diagonal Λ .Assume that all eigenvalues of Λ have non-positive real parts, and there is at least one purely imaginaryeigenvalue λ = iω . For convenience, suppose that all eigenvalues of the form λ s = in s ω with integer n s are listed as first k . Then periodic solutions to the system becoming at µ = 0 periodic solutionsto the generating system ˙ y = Λy with period T = πω can only exist for such values of amplitudes α , ..., α k − , α k − = α k , that satisfy equations Q s ( α , ..., α k ) = α k n k P s − α s n s P k = 0 , s = 1 , ..., k − , (A.3) where P s ( α , ..., α k ) := (cid:10) F s e − in s ωt (cid:11) := 1 T (cid:90) T F s (cid:0) α e in ωt , ..., α k e in k ωt , , ..., (cid:1) e − in s ωt dt , s = 1 , ..., k . (A.4) The first order period correction for these solutions is δ = P k ( α ∗ , ..., α ∗ k ) iα ∗ k n k T , (A.5) where α ∗ s are the values found from (A.3) .These solutions will indeed exist and be stable (asymptotically orbitally stable) for small positive µ ifall the roots κ of the following algebraic equations have negative real parts: (cid:12)(cid:12)(cid:12)(cid:12) ∂Q s ∂α j − α k n k δ sj κ (cid:12)(cid:12)(cid:12)(cid:12) = 0 , s, j = 1 , ..., k − (cid:12)(cid:12)(cid:12)(cid:12)(cid:28) ∂F s ∂y j e i ( n j − n s ) ωt (cid:29) − δ sj (cid:32) (cid:0) n s + (cid:1) P k n s α k + κ (cid:33)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0 , s, j in the secondary special group; (A.7) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:28) ∂F s ∂y j e i (cid:16) ν ( r ) j − ν ( r ) s (cid:17) ωt (cid:29) − δ sj (cid:32) ν ( r ) s P k n s ωα k + κ (cid:33)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = 0 , s, j in the r th non-special group. (A.8) If the real part of at least one root of these equations is positive then the corresponding solution is unstable.
References [1] D. Korteweg. Huygens’ sympathic clocks and related phenomena in connection with the principal andthe compound oscillations presenting themselves when two pendulums are suspended to a mechanismwith one degree of freedom.
Proceedings of the Royal Academy of Amsterdam , 8:436–455, 1905.[2] M. Bennett, M. Schatz, H. Rockwood, and K. Wiesenfeld. Huygens’s clocks.
The Royal Society ofLondon. Proceedings. Series A. Mathematical, Physical and Engineering Sciences , 458(2019):563–579,2002.[3] M. Senator. Synchronization of two coupled escapement-driven pendulum clocks.
Journal of Soundand Vibration , 291:566–603, 2006.[4] I. Blekhman.
Synchronization in science and technology . ASME Press, New York, 1988.[5] K. Czolczynski, P. Perlikowski, A. Stefanski, and T. Kapitaniak. Huygens’ odd sympathy ex-periment revisited.
International Journal of Bifurcation and Chaos . To appear, DOI No:10.1142/S0218127411029628.[6] R. Dil˜ao. On the problem of synchronization of identical dynamical systems: the Huygens’s clocks.
Springer Optimization and Its Applications , 33:163–181, 2009.[7] A. Fradkov and B. Andrievsky. Synchronization and phase relations in the motion of two-pendulumsystem.
International Journal of Non-Linear Mechanics , 42(6):895–901, 2007.[8] J. Pantaleone. Synchronization of metronomes.
American Journal of Physics , 70(10):992–1000, 2002.[9] K. Czolczynski, P. Perlikowski, A. Stefanski, and T. Kapitaniak. Clustering of Huygens’ clocks.
Progress of Theoretical Physics , 122(4):1027–1033, 2009.[10] W. Oud, H. Nijmeijer, and A. Pogromsky. A study of Huijgens’ synchronization: experimentalresults.
Lecture Notes in Control and Information Sciences , 336:191–203, 2006.[11] N. Kuznetsov, G. Leonov, H. Nijmeijer, and A Pogromsky. Synchronization of two metronomes.
Proceedings of the 3rd IFAC Workshop on Periodic Control Systems , 3, 2007.2212] A. Andronov, A. Vitt, and S. Khaikin.
Theory of oscillators . Dover, New York, 1987.[13] I. Blekhman.