Abstract
In this paper, we show that the oscillations’ amplitude increases and then saturates during the startup in the single-branch pulsating heat pipe (SBPHP) due to the interplay between an instability mechanism and a limiting mechanism produced by nonlinearities. The SBPHP is a small tube closed at one end (which is heated) and open at the other, in which sits a single vapor bubble followed by a liquid plug. Increasing the heater temperature above a critical value leads to oscillations of the liquid plug. The oscillations’ amplitude initially increases over time and eventually saturates (the system reaches a steady state). To describe this dynamics, we start from a known theoretical SBPHP model. The model is derived from the momentum balance on the liquid plug which includes both the pressure difference between the vapor and the environment and the viscous friction between the liquid and the wall. Pressure in the vapor varies due to compression and expansion of the vapor bubble, but also due to evaporation and condensation as the liquid plug oscillates between the heat source and the heat sink. Both the pressure (described by perfect gas law) and the evaporation and condensation (limited in both the heat source and the heat sink) make the differential equations in the model nonlinear. By applying the center manifold reduction technique followed by the normal form reduction technique, we obtain an approximate analytical solution (describing the position and velocity of the liquid plug, and the mass of vapor, as a function of time). The analytical solution describes well the startup and the subsequent steady state. Using this solution, we show that the initial increase in oscillations’ amplitude is due to the instability mechanism (produced by the interplay of phase change and viscous friction), while the following amplitude saturation is due to the pressure and phase-change nonlinearities, as the system reaches a steady state corresponding to a limit cycle. The limit cycle is created through a Poincaré–Andronov–Hopf bifurcation. We find that one can greatly increase the oscillations’ amplitude in the steady state by increasing the equilibrium instability and by decreasing these nonlinearities. This can be done by increasing the phase change and by decreasing the friction. We conclude that the SBPHP dynamics should be viewed as a combination of an instability mechanism which tends to increase the oscillations’ amplitude and a limiting mechanism produced by nonlinearities, which limits the amplitude.
Similar content being viewed by others
Data Availability Statement
The dataset analyzed during the current study is made available in the supplementary materials.
Notes
Experimental measurements over a longer time span show slight fluctuations of the limit cycle, but those occur on a longer time scale and are likely due to undesirable fluctuations of the external parameters such as the heat source temperature.. The dynamics on the limit cycle is periodic and corresponds to the oscillating steady-state regime. The geometry of the limit cycle dictates the oscillating amplitude.
We may omit to specify the dimensionless phase-space variables and parameters as dimensionless in the following, for conciseness.
We assumed \(T_g\) constant, see [49] for the general case.
Tessier-Poirier et al. [49] used a parameter \({\widetilde{C_{th}}}\). Here, we use \(\psi \) instead, based on the relation \(\psi =2{\widetilde{C_{th}}}/T_{HL}\). We use \(\psi \) because \({\widetilde{C_{th}}}\) only appears in the ratio \(2{\widetilde{C_{th}}}/T_{HL}\) in the normal form solution. Also, \(\psi \) has some nice properties such as being bounded between \(-\pi \) and \(+\pi \) and being a function of only \(T_H\), \(T_L\) and \(T_{g,sat,0}\). See Appendix B for details.
We used the ODE45 numerical solver from MATLAB unless specified otherwise.
See De Regt and Dieks [7, p.150,151] for a description of what understanding is. A key point is that understanding of a phenomenon is achieved when the scientist can “recognize qualitatively characteristic consequences of [the theory] without performing exact calculations.”
Note that we already obtained the linear solution in Sect. 2.2. The normal form’s results presented here are equivalent, within the approximation of small \({\delta }\) and from a perturbation of the equilibrium. Looking at the linear case again, but from the normal form’s perspective, allows to better understand the normal form’s differential Eq. (3.3a) and serve as a reference for the nonlinear case.
This bifurcation is commonly called a Hopf bifurcation but we use the longer denomination, Poincaré–Andronov–Hopf bifurcation, to acknowledge the contributions of Poincaré and Andronov, as recommended by Wiggins [55, p.385].
The \(1+\cos \left[ \psi \right] \) in the denominator does not change the sign of \(a_{0,T}\) because \(\cos \left[ \psi \right] >-1\), since \(T_{g,sat,0}<T_H\).
the solution \({\zeta _{f}}= - \sqrt{-1/\cos [\psi ]}\) is discarded, since \({\zeta _{f}}>0\).
Abbreviations
- A :
-
Inner section area of the tube
- \(a_0\) :
-
Nonlinear coefficient in the normal form
- \({c_P}\) :
-
Pressure nonlinearity coefficient
- \({c_T}\) :
-
Phase-change nonlinearity coefficient
- d :
-
Coefficient in the normal form
- F :
-
Force
- g :
-
Gravitational acceleration
- \(H_v\) :
-
Enthalpy of vaporization
- L :
-
Length
- m :
-
Mass
- P :
-
Pressure
- \(\hat{P}_{e}\) :
-
Effective external pressure (augmented by gravity)
- \(\dot{Q}_g\) :
-
Heat transfer between the walls and the vapor
- \(q_1\) :
-
Dimensionless position
- \(q_2\) :
-
Dimensionless velocity
- \(q_3\) :
-
Dimensionless mass of vapor
- R :
-
Inner radius of the tube
- r :
-
Radius in phase space
- \(R_g\) :
-
Gas constant
- \( { r_{LC} } \) :
-
Radius on the limit cycle
- t :
-
Time
- \(T_g\) :
-
Temperature of the vapor (spatially averaged)
- \(|T'_{w,0} |\) :
-
Absolute value of the axial wall thermal gradient
- \(T_H\) :
-
Temperature of the heat source
- \(T_{HL}\) :
-
Dimensionless phase-change limits
- \(T_L\) :
-
Temperature of the heat sink
- \(x_i\) :
-
Position of the meniscus relative to the equilibrium
- \(z_i\) :
-
Phase-space variables in the normal form
- \({\delta }\) :
-
Bifurcation parameter, \({\delta }={\sigma }-{\zeta _{f}}\)
- \(\lambda _1\) :
-
Dimensionless growth rate of the first exponential
- \(\mu \) :
-
Dynamic viscosity
- \(\nu \) :
-
Kinematic viscosity
- \(\omega \) :
-
Angular frequency
- \(\omega _n\) :
-
Natural angular frequency
- \(\Pi \) :
-
Instability number, \(\Pi ={\sigma }/{\zeta _{f}}\)
- \(\varphi \) :
-
Constant of the linearized solution
- \(\psi \) :
-
Measure of the asymmetry of the arctangent
- \(\rho \) :
-
Density
- \({\sigma }\) :
-
Dimensionless phase-change coefficient
- \(\tau \) :
-
Dimensionless time
- \(\theta \) :
-
Angle of the tube relative to the horizontal
- \({\zeta _{f}}\) :
-
Dimensionless friction coefficient
- 0:
-
Equilibrium value
- e :
-
External (on the air side)
- f :
-
Viscous friction
- g :
-
Gravity or gas (vapor)
- \(\ell \) :
-
Liquid
- m :
-
Produced by change of vapor mass
- sat :
-
At saturation condition
- v :
-
Produced by change of vapor volume
- w :
-
Wall
- \(\,\bar{}\) :
-
Averaged quantity
- \(\Delta \) :
-
Difference between a property and the equilibrium value
- \(\,\dot{}\) :
-
Time derivative (dimensional or dimensionless, according to the variable)
- \(\,\,\widetilde{}\) :
-
Dimensionless quantity
References
Abdelkefi, A., Nayfeh, A.H., Hajj, M.R.: Design of piezoaeroelastic energy harvesters. Nonlinear Dyn. 68(4), 519–530 (2012). https://doi.org/10.1007/s11071-011-0233-x
Akachi, H.: Structure of micro-heat pipe. Patent 5 219 020. (1993)
Aubin, P., D’Entremont, B., Cataldo, F., Marcinichen, J.B., Amalfi, R.L., Thome, J.R.: Numerical simulations of pulsating heat pipes, part 1: modeling. In: 2019 18th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (ITherm), pp. 232–242 (2019). https://doi.org/10.1109/ITHERM.2019.8757388
Bénard, H.: Les tourbillons cellulaires dans une nappe liquide. Revue Générale des Sciences Pures et Appliquées 11(1261–1271), 1309–1328 (1900)
Bénard, H.: Les tourbillons cellulaires dans une nappe liquide propageant de la chaleur par convection: en régime permanent. Gauthier-Villars (1901)
Das, S.P., Nikolayev, V.S., Lefevre, F., Pottier, B., Khandekar, S., Bonjour, J.: Thermally induced two-phase oscillating flow inside a capillary tube. Int. J. Heat Mass Transf. 53(19–20), 3905–3913 (2010). https://doi.org/10.1016/j.ijheatmasstransfer.2010.05.009
De Regt, H.W., Dieks, D.: A Contextual Approach to Scientific Understanding. Synthese 144(1), 137–170 (2005). https://doi.org/10.1007/s11229-005-5000-4
Dhooge, A., Govaerts, W., Kuznetsov, Y.A., Meijer, H.G., Sautois, B.: New features of the software MatCont for bifurcation analysis of dynamical systems. Math. Comput. Modell. Dynam. Syst. 14(2), 147–175 (2008). https://doi.org/10.1080/13873950701742754. MatCont available on https://sourceforge.net/projects/matcont/
Dobson, R.T.: An open oscillatory heat pipe water pump. Appl. Therm. Eng. 25(4), 603–621 (2005). https://doi.org/10.1016/j.applthermaleng.2004.07.005
Drazin, P.G.: Introduction to Hydrodynamic Stability. Cambridge University Press, Cambridge (2002)
d’Entremont, B.P., Thome, J.R.: A numerical study of pulsating heat pipe performance. In: Proceedings of the ASME 2015 International Technical Conference and Exhibition on Packaging and Integration of Electronic and Photonic Microsystems (InterPACK2015). American Society of Mechanical Engineers Digital Collection (2015). https://doi.org/10.1115/IPACK2015-48350
Faghri, A.: Heat pipes: review, opportunities and challenges. Front. Heat Pipes 5(1), 1–48 (2014)
Fourgeaud, L., Ercolani, E., Duplat, J., Gully, P., Nikolayev, V.S.: Evaporation-driven dewetting of a liquid film. Phys. Rev. Fluids 1(4), 041901 (2016). https://doi.org/10.1103/PhysRevFluids.1.041901
Ghanta, N., Pattamatta, A.: Modeling of compressible phase-change heat transfer in a Taylor–Bubble with application to pulsating heat pipe (PHP). Numer. Heat Transf. Part A: Appl. 69(12), 1355–1375 (2016). https://doi.org/10.1080/10407782.2016.1139980
Guckenheimer, J., Holmes, P.: Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer, New York (2013)
Hao, T., Ma, X., Lan, Z., Li, N., Zhao, Y.: Effects of superhydrophobic and superhydrophilic surfaces on heat transfer and oscillating motion of an oscillating heat pipe. J. Heat Transf. (2014). https://doi.org/10.1115/1.4027390
Hao, T., Ma, X., Lan, Z., Li, N., Zhao, Y., Ma, H.: Effects of hydrophilic surface on heat transfer performance and oscillating motion for an oscillating heat pipe. Int. J. Heat Mass Transf. 72, 50–65 (2014). https://doi.org/10.1016/j.ijheatmasstransfer.2014.01.007
Homsy, G.M.: Viscous fingering in porous media. Annu. Rev. Fluid Mech. 19(1), 271–311 (1987). https://doi.org/10.1146/annurev.fl.19.010187.001415
Jenkins, A.: Self-oscillation. Phys. Rep. 525(2), 167–222 (2013). https://doi.org/10.1016/j.physrep.2012.10.007
Jun, S., Kim, S.J.: Experimental study on a criterion for normal operation of pulsating heat pipes in a horizontal orientation. Int. J. Heat Mass Transf. 137, 1064–1075 (2019). https://doi.org/10.1016/j.ijheatmasstransfer.2019.03.163
Jung, C., Kim, S.J.: Effects of oscillation amplitudes on heat transfer mechanisms of pulsating heat pipes. Int. J. Heat Mass Transf. 165, 120642 (2021). https://doi.org/10.1016/j.ijheatmasstransfer.2020.120642
Kim, W., Kim, S.J.: Effect of reentrant cavities on the thermal performance of a pulsating heat pipe. Appl. Therm. Eng. 133, 61–69 (2018). https://doi.org/10.1016/j.applthermaleng.2018.01.027
Kuznetsov, Y.: Elements of Applied Bifurcation Theory. Springer, Berlin (1998)
Lin, Z., Wang, S., Shirakashi, R., Winston Zhang, L.: Simulation of a miniature oscillating heat pipe in bottom heating mode using CFD with unsteady modeling. Int. J. Heat Mass Transf. 57(2), 642–656 (2013). https://doi.org/10.1016/j.ijheatmasstransfer.2012.09.007
Monin, T., Tessier-Poirier, A., Léveillé, É., Juneau-Fecteau, A., Skotnicki, T., Formosa, F., Monfray, S., Fréchette, L.G.: First experimental demonstration of a self-oscillating fluidic heat engine (SOFHE) with piezoelectric power generation. J. Phys. Conf. Ser. (2016). https://doi.org/10.1088/1742-6596/773/1/012039
Moon, F., Rand, R.H.: Parametric stiffness control of flexible structures. In: G. Rodriguez (ed.) JPL Proceedings of the Workshop on Identification and Control of Flexible Space Structures, vol. 2. JPL Publication, California Institute of Technology, Pasadena, California (1985)
Nemati, R., Shafii, M.B.: Advanced heat transfer analysis of a U-shaped pulsating heat pipe considering evaporative liquid film trailing from its liquid slug. Appl. Therm. Eng. 138, 475–489 (2018). https://doi.org/10.1016/j.applthermaleng.2018.04.064
Nikolayev, V.S.: Effect of tube heat conduction on the single branch pulsating heat pipe start-up. Int. J. Heat Mass Transf. 95, 477–487 (2016). https://doi.org/10.1016/j.ijheatmasstransfer.2015.12.016
Nikolayev, V.S.: Physical principles and state-of-the-art of modeling of the pulsating heat pipe: a review. Appl. Thermal Eng. (2021). https://doi.org/10.1016/j.applthermaleng.2021.117111
Perko, L.: Differential Equations and Dynamical Systems. Springer, Berlin (2008)
Qu, J., Li, X., Wang, Q., Liu, F., Guo, H.: Heat transfer characteristics of micro-grooved oscillating heat pipes. Exp. Thermal Fluid Sci. 85, 75–84 (2017). https://doi.org/10.1016/j.expthermflusci.2017.02.022
Rand, R.H.: Lecture Notes on Nonlinear Vibrations. The Internet-First University Press, USA (2012)
Rao, M., Lefèvre, F., Khandekar, S., Bonjour, J.: Understanding transport mechanism of a self-sustained thermally driven oscillating two-phase system in a capillary tube. Int. J. Heat Mass Transf. 65, 451–459 (2013). https://doi.org/10.1016/j.ijheatmasstransfer.2013.05.067
Rao, M., Lefèvre, F., Khandekar, S., Bonjour, J.: Heat and mass transfer mechanisms of a self-sustained thermally driven oscillating liquid-vapour meniscus. Int. J. Heat Mass Transf. 86, 519–530 (2015). https://doi.org/10.1016/j.ijheatmasstransfer.2015.03.015
Rayleigh, L.: The Explanation of Certain Acoustical Phenomena. Nature (1878). https://doi.org/10.1038/018319a0
Rayleigh, J.W.S.B.: The Theory of Sound. Macmillan, New York (1896)
Rayleigh, L.: On convection currents in a horizontal layer of fluid, when the higher temperature is on the under side. The London, Edinburgh, and Dublin Philos. Magaz. J. Sci. (1916). https://doi.org/10.1080/14786441608635602
Reynolds, O.: An experimental investigation of the circumstances which determine whether the motion of water shall be direct or sinuous, and of the law of resistance in parallel channels. Proc. Royal Soc. London 35(224—-226), 84–99 (1883). https://doi.org/10.1098/rspl.1883.0018
Rijke, P.L.: Notice of a new method of causing a vibration of the air contained in a tube open at both ends. The London, Edinburgh, Dublin Philos. Magaz. J. Sci. (1859). https://doi.org/10.1080/14786445908642701
Rouaze, G., Marcinichen, J.B., Cataldo, F., Aubin, P., Thome, J.R.: Simulation and experimental validation of pulsating heat pipes. Appl. Therm. Eng. (2021). https://doi.org/10.1016/j.applthermaleng.2021.117271
Saffman, P.G., Taylor, G.I.: The penetration of a fluid into a porous medium or Hele–Shaw cell containing a more viscous liquid. Proc. Royal Soc. London. Ser. A. Math. Phys. Sci. 245(1242), 312–329 (1958). https://doi.org/10.1098/rspa.1958.0085
Shafii, M.B., Faghri, A., Zhang, Y.: Thermal modeling of unlooped and looped pulsating heat pipes. J. Heat Transf. 123(6), 1159–1172 (2001). https://doi.org/10.1115/1.1409266
Spinato, G., Borhani, N., Thome, J.R.: Understanding the self-sustained oscillating two-phase flow motion in a closed loop pulsating heat pipe. Energy (2015). https://doi.org/10.1016/j.energy.2015.07.119
Strogatz, S.H.: Nonlinear Dynamics and Chaos: With Applications to Physics, Chemistry, Biology and Engineering. Westview Press, USA (1994)
Sudevalayam, S., Kulkarni, P.: Energy harvesting sensor nodes: survey and implications. IEEE Commun. Surv. Tutorials 13(3), 443–461 (2011). https://doi.org/10.1109/SURV.2011.060710.00094
Sundmaeker, H., Guillemin, P., Friess, P., Woelfflé, S.: Vision and Challenges for Realising the Internet of Things. Cluster of European Research Projects on the Internet of Things (CERP-IoT), Brussels (2010)
Taylor, G.I.: Stability of a viscous liquid contained between two rotating cylinders. Philos. Trans. Royal Soc. London. Ser. A, Contain. Papers Math. Phys. Character 223, 289–343 (1923)
Tessier-Poirier, A., Monin, T., Léveillé, E., Formos, F., Monfray, S., Fréchette, L.G.: Influence of nonlinearities on the power output of the Self-Oscillating Fluidic Heat Engine (SOFHE). J. Phys.: Conf. Ser. 773(1), 012113 (2016). https://doi.org/10.1088/1742-6596/773/1/012113
Tessier-Poirier, A., Monin, T., Léveillé, É., Monfray, S., Formosa, F., Fréchette, L.G.: How evaporation and condensation lead to self-oscillations in the single-branch pulsating heat pipe. Phys. Rev. Fluids 4(10), 103901 (2019). https://doi.org/10.1103/PhysRevFluids.4.103901
Tessier-Poirier, A., Rand, R.H., Fréchette, L.G.: What limits the oscillations amplitude of the single-branch pulsating heat pipe. In: ISOPHP 2019. Daejeon, Korea (2019)
Theodorsen, T.: General Theory of Aerodynamic Instability and the Mechanism of Flutter. Technical Report. NACA-TR-496, National Advisory Committee for Aeronautics (NACA) (1935)
Theodorsen, T., Garrick, I.E.: Mechanism of Flutter A Theoretical and Experimental Investigation of the Flutter Problem. Technical Report. NACA-TR-685, National Advisory Committee for Aeronautics (NACA) (1940)
Wang, J., Ma, H., Zhu, Q.: Effects of the evaporator and condenser length on the performance of pulsating heat pipes. Appl. Therm. Eng. 91, 1018–1025 (2015). https://doi.org/10.1016/j.applthermaleng.2015.08.106
Wang, J., Ma, H., Zhu, Q., Dong, Y., Yue, K.: Numerical and experimental investigation of pulsating heat pipes with corrugated configuration. Appl. Therm. Eng. 102, 158–166 (2016). https://doi.org/10.1016/j.applthermaleng.2016.03.163
Wiggins, S.: Introduction to Applied Nonlinear Dynamical Systems and Chaos. Springer, Berlin (2003)
Zhang, Y., Faghri, A.: Advances and Unsolved Issues in Pulsating Heat Pipes. Heat Transfer Eng. 29(1), 20–44 (2008). https://doi.org/10.1080/01457630701677114
Zhang, Y., Faghri, A., Shafii, M.B.: Analysis of liquid-vapor pulsating flow in a U-shaped miniature tube. Int. J. Heat Mass Transf. 45(12), 2501–2508 (2002). https://doi.org/10.1016/S0017-9310(01)00348-9
Zuo, Z.J., North, M.T., Ray, L.: Combined pulsating and capillary heat pipe mechanism for cooling of high heat flux electronics. ASME 364, 237–244 (1999)
Acknowledgements
We are grateful to our colleagues Nooshin Karami, Étienne Léveillé and Ali Hossein Nikkhah for discussions on the subject. A special thanks to Nooshin Karami who also provided helpful comments on the manuscript. We also acknowledge financial support from the NSERC through the Scholarship and Discovery Programs (Canada).
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Appendices
Detailed procedure to derive the normal form
In this appendix, we show how to derive the normal form vector field for r and \(\theta \) (Eq. (3.3)) from the vector field for \(q_1\), \(q_2\) and \(q_3\) (Eq. (2.14)). We first use the center manifold reduction to reduce the three-dimensional system to a two-dimensional one (Sect. A.1). We then reduce the two-dimensional system to its normal form, which we can solve analytically (Sect. A.2). The solution is provided for new phase-space variables \(z_1\) and \(z_2\), related to the original one (\(q_1\), \(q_2\) and \(q_3\)) through a series of transformations. In Sect. A.3, we make the inverse transformation and provide the analytical solution for \(q_1\), \(q_2\) and \(q_3\). Our procedure is mainly based on Wiggins [55]. See Rand [32] for an introduction and Moon and Rand [26] for an example of the technique. Because the procedure involves lengthy manipulations, it is best handled using analytical software. The MATLAB code used for this purpose is made available in supplementary materials. In the following, we will not include the full explicit expressions of some of the quantities in the intermediary steps, please refer to the code for those.
1.1 Center manifold reduction
We start from Eq. (2.14), but replace \({\sigma }\) by the bifurcation parameter \({\delta }\), using \({\sigma }={\delta }+{\zeta _{f}}\) and resulting in
To apply the center manifold reduction technique, we need the real part of the complex conjugate eigenvalues to be 0. Here, however, the real part of the complex conjugate eigenvalues can be positive or negative, if \({\delta }\ne 0\) (see Sect. 2.2). Fortunately, we can transform the system to satisfy the center manifold requirement by a clever trick [55, p.251]: We define \({\delta }\) as a new phase-space variable and add the following differential equation to the system:
which defines \({\delta }\) as a constant. Because \({\delta }\) is a phase-space variable, the quantities \({\delta }q_1\) are now nonlinear. Thus, \({\delta }\) will not appear in the linear part and will not contribute to the eigenvalues anymore. The real part of the complex-conjugate eigenvalues will be 0. (We will show that in the following.) The new phase space given by Eq. (A.1) is four-dimensional. As a compact notation, we define the vector of phase-space variables \(\varvec{x}\) as \(\varvec{x}\equiv [q_1,q_2,q_3,{\delta }]^T\) and Eq. (A.1) becomes \(\dot{\varvec{x}}=\varvec{f(\varvec{x})}\), where \(\varvec{f}\) is the vector field.
1.1.1 Standard form
Before making the reduction, we must transform the system into a standard form. To do so, we first split the vector field as
where \(A \varvec{x}\) is the linear part and \(\varvec{F}(\varvec{x})\) is the nonlinear part. The matrix A is given by
where \(\mathrm {D}\varvec{f}(\varvec{0})\) is the Jacobian of the vector field evaluated at the equilibrium \(\varvec{x}=\varvec{0}\). (This is the linearization of the vector field.) The nonlinear part is given by \(\varvec{F}(\varvec{x}) = \varvec{f}(\varvec{x}) - A\varvec{x}\). We have that \(\varvec{F}(\varvec{0})=\varvec{0}\) and \(\mathrm {D}\varvec{F}(\varvec{0})=\varvec{0}\). For small \(\varvec{x}\), \(\varvec{F}(\varvec{x})\) is small and the system is weakly nonlinear.
The eigenvalues of A are: \(\lambda _1=0\), \(\lambda _2=-i\), \(\lambda _3=i\) and \(\lambda _4=-2{\zeta _{f}}\). We see here that we indeed have the real part of complex-conjugate eigenvalues \(\lambda _{2,3}\) being 0. The corresponding eigenvectors are the columns of the matrix
The standard form we seek requires the linear part to be in a block diagonal form [55, p.37], in the Jordan canonical form more specifically. To reach that form, we use a linear transformation given by a matrix \({T_a}\). For A having distinct real eigenvalues \(\lambda _j\) with corresponding eigenvectors \(\varvec{v_j}\) and distinct complex eigenvalues \(\lambda _j = a_j + i \, b_j\) and \(\bar{\lambda }_j = a_j - i \, b_j\) with corresponding complex eigenvectors \(\varvec{w_j} = \varvec{u_j} +i\, \varvec{v}_j\) and \(\varvec{\bar{w}_j} = \varvec{u_j} -i\, \varvec{v}_j\), with \(j = k+1,...,n\), the matrix \({T_a}\) is given by \({T_a}=\left[ \varvec{v_1},\varvec{u_1}, \dots , \varvec{v_n}, \varvec{u_n}\right] \) (see Perko [30, p.39,28,154,108] for details). Given the eigenvectors Eq. (A.4), we obtain, for \({T_a}\) and its inverse \({T_a}^{-1}\) (with \(\kappa \equiv 1/(1+ 4{\zeta _{f}}^2)\)),
We apply this transformation to the vector of phase-space variable \(\varvec{x}\) and obtain a transformed vector of phase-space variables, we have
and
Substitution of \(\varvec{x} = {T_a}\varvec{y}\) into the system \(\varvec{\dot{x}} = A \varvec{x} + \varvec{F}(\varvec{x})\) (Eq. (A.2) ) leads to
where J is the matrix \({T_a}^{-1} A \, {T_a}\) and \(\varvec{G}(\varvec{y}) = {T_a}^{-1} \varvec{F}({T_a}\varvec{y})\). The matrix J is given by
We can split Eq. (A.8) into three rows as
where
In Eq. (A.10), the first row having variable \(\varvec{y}_p={\delta }\) corresponds to a parameter subsystem, the second row having variables \(\varvec{y}_c=[y_2,y_3]^T\) corresponds to a center subsystem since its eigenvalues are complex conjugate eigenvalues with 0 real parts and the third row having variable \(\varvec{y}_s=y_4\) corresponds to a stable subsystem since its eigenvalue is real and negative. The parameter, center and stable directions are linearly uncoupled from one another. We have that \(\varvec{G}_c=\varvec{0}\), \(\varvec{G}_s=\varvec{0}\), \(\mathrm {D}\varvec{G}_c=\varvec{0}\) and \(\mathrm {D}\varvec{G}_s=\varvec{0}\). Equation (A.10) is the standard form required to apply the center manifold theorem.
1.1.2 Center manifold reduction
There exists a center manifold for the system Eq. (A.10) defined locally by
for \({\gamma }_c\) and \({\gamma }_p\) sufficiently small [55, Sec. 18.2, p.251]. The center manifold is a three-dimensional surface in the four-dimensional phase space. Sufficiently close to the origin \(\varvec{y}=\varvec{0}\), “small amplitude periodic orbits are contained in the center manifold” [55, p.247]. Consequently, we will be able to consider only the dynamics in the center manifold and thus reduce the dimensionality of the system. In Eq. (A.12), we have that \(\varvec{y}_s=y_4=h\left( \varvec{y}_c, \varvec{y}_p\right) \), the stable direction is reduced to a function h of \(\varvec{y}_c=[y_2,y_3]^T\) and \(\varvec{y}_p=y_1={\delta }\).
It is common to approximate h by a power expansion, we have
We now must find the coefficients \(c_j\). For \(W^c_{loc}(0)\) to be a center manifold, h must satisfy the differential equations of the system. By substitution of \(y_4=h({\delta },y_2,y_3)\) in Eq. (A.8), we find that h must satisfy the following quasilinear partial differential equation [55, p.252]
Collecting terms of like powers, we obtain 6 algebraic equations which we solve for the 6 c’s, leading to
By substitution of \(\varvec{y}_s=y_4=h({\delta },y_2,y_3)\) in system Eq. (A.10), the center subsystem becomes
where \(\varvec{y}_c = \left[ y_2 , y_3\right] ^T\) and \(\varvec{G}_c\) is a complicated function of \(y_2\) \(y_3\) and \({\delta }\). This system is now completely decoupled from the stable subsystem (does not depend on \(\varvec{y_s}\) at all). The parameter subsystem equation is again
Equation (A.16) is a three-dimensional phase space (with variables \(y_2\), \(y_3\) and \({\delta }\)). However, since \(\dot{{\delta }} = 0\), the motion on the plane \(y_2-y_3\) is invariant (remains on the plane), so knowing the value \({\delta }\) is enough to predict the motion on the plane \(y_2-y_3\). The dynamics is thus effectively two-dimensional and is given by Eq. (A.16a). Properties of the dynamics in the plane \(y_2-y_3\) can change as the height \({\delta }\) changes. In fact, we will find that a Hopf bifurcation occurs at \({\delta }=0\).
1.2 Normal form
Now that we have a two-dimensional system Eq. (A.16a), we can use the normal form reduction technique to obtain an approximative solution for it. In normal form, we want the instability to be captured in the linear part, we want \({\delta }\) to appear in the linear part [55, Eq.(19.2.1,19.2.3), p.278,279]. Unfortunately, we purposefully removed \({\delta }\) from the linear part to apply the center manifold reduction. We can bring \({\delta }\) back by splitting the vector field \(\varvec{g}_c\) in Eq. (A.16) again, but without taking \({\delta }\) as a phase-space variable. Doing so, we get
where \(B=\mathrm {D}_{\varvec{y}_c} \, \varvec{g}_c(\varvec{0})\) is the Jacobian of \(\varvec{g}_c\) evaluated at the fixed point \(\varvec{y}_c=\varvec{0}\). Also, \(\varvec{H} = \varvec{g}_c - B\varvec{y}_c\), \(\varvec{H}\in C^1(E)\), \(\varvec{H}(\varvec{0})=\varvec{0}\) and \(\mathrm {D}\varvec{H}(\varvec{0})=\varvec{0}\). As desired, the eigenvalues of B now depend on \({\delta }\). However, B is not in the Jordan canonical form, so we must perform a linear transformation again, similarly to what was done in Sect. A.1.1.
1.2.1 Standard form for complex conjugate eigenvalues
Here, we transform Eq. (A.17) in its standard form, such that the linear part is in a Jordan canonical form. To do so, we must perform a linear transformation through the multiplication of a matrix \({T_b}\). To find such matrix, we first Taylor expand B for small \({\delta }\). We find
Considering the truncated expression of B above, we compute the trace \({{\,\mathrm{tr}\,}}(B)\) and the determinant \(\det (B)\). Assuming the discriminant \({{\,\mathrm{tr}\,}}(B)^2-4\det (B)<0\) (true for either small \(|{\delta } |\) or \({\delta }>0\)), the eigenvalues are complex conjugates given by \(\lambda =\alpha + i\omega \) and \(\bar{\lambda }=\alpha - i\omega \) with the real part \({\alpha =\tfrac{1}{2} {{\,\mathrm{tr}\,}}(B)}\) and the imaginary part \({\omega =\tfrac{1}{2} \sqrt{{{\,\mathrm{tr}\,}}(B)^2-4\det (B)}}\). Taylor expansion of \(\alpha \) and \(\omega \) in terms of \({\delta }\) leads to
The eigenvectors corresponding to the eigenvalues \(\lambda \) and \(\bar{\lambda }\) are \(\varvec{w} = \varvec{u} +i\, \varvec{v}\), and \(\varvec{\bar{w}} = \varvec{u} -i\, \varvec{v}\), respectively, with
where the \(b_{ij}\) are the elements of the truncated matrix B (Eq. (A.18)).
From those, we build the transformation matrix as
Introducing a new phase-space vector of variables \(\varvec{z}\) with \(\varvec{y}_c={T_b}\varvec{z}\), Eq. (A.17) is transformed into
where \(\varvec{k}(\varvec{z})\) is a new vector field, \(J_B={T_b}^{-1} B \, {T_b}\) and \(\varvec{K}(\varvec{z}) = {T_b}^{-1} \varvec{H}({T_b}\varvec{z})\). Equation (A.22) is the required form to apply normal form reduction (requirement specified in [55, 197/,148/]). Indeed, we have that the linear part \(J_B\) satisfies
so \(J_B\) is in the Jordan canonical form at first order approximation. Also, the vector field \(\varvec{k}\) is \(C^r\) with \(r \geqslant 5\), centered at the fixed point of interest such that \(\varvec{k}\left( \varvec{z}=\varvec{0},{\delta }=0\right) =\varvec{0}\) and finally, \(\alpha ({\delta }=0)=0\) and \(\omega ({\delta }=0)\ne 0\).
1.2.2 Normal form reduction
One can reduce Eq. (A.22) to its normal form by third order Taylor expansion of the vector field and successive transformations (see Wiggins [55, Eq.(20.2.7) p.379, p.279] for details), leading to
where a and b are given by the normal form procedure. Equation (A.24) corresponds to Eq. (3.1) in the main text. We care especially about the amplitude of the oscillations, so it makes sense to introduce polar coordinates with \(z_1=r\cos (\theta )\) and \(z_2=r\sin (\theta )\), where r is the amplitude and \(\theta \) is the phase. Introducing those into Eq. (A.24) leads to
which are differential equations for r and \(\theta \). Because we study this system close to the bifurcation (close to \({\delta }=0\)), it makes sense to simplify Eq. (A.25) by Taylor expanding its coefficients in terms of \({\delta }\). We get
which is the desired normal form (corresponds to Eq. (3.3) in the main text).
The parameters d, \(\omega _0\) and c are given by (see supplementary material and Wiggins [55])
The expression of \(b_0\) will not be needed but can be found through the normal form procedure [15]. Finally, \(a_0\) is given by
with
which corresponds to Eq. (3.5) in the main text.
1.3 Solution in the q variables (back transformation)
We now would like to know what is the analytical solution for the original variables \(q_1\), \(q_2\) and \(q_3\). To obtain it, we must make the inverse transformations, going from \(\varvec{z}\) to \(\varvec{y_c}\) (normal form) to \(\varvec{y}\) (center manifold reduction) to \(\varvec{x}\) (standard form) to \(\varvec{q}\) (which is easy). Refer to previous sections to reproduce the following results.
First, we have that \(\varvec{y}_c = {T_b}\varvec{z}\). Then, we can obtain \(\varvec{y}\) from \(\varvec{y}_c\) based on the following, where we used row vectors [1, 0] and [0, 1] to select \(y_{c,1}\) and \(y_{c,2}\), respectively:
We then obtain \(\varvec{x}\) from \(\varvec{y}\) (based on Eq. (A.7))
\(\varvec{q}\) simply corresponds to the first three elements of \(\varvec{x}\), we have
where h is given by the following (based on Eq. (A.13) with \(y_1={\delta }\))
with \(y_2(\varvec{z})=[1,0] \cdot {T_b}\varvec{z}\) and \(y_3(\varvec{z})=[0,1] \cdot {T_b}\varvec{z}\).
Now, because the solutions are \(z_1=r \cos (\theta )\) and \(z_2=r \sin (\theta )\), the quantities \([n,m] \cdot {T_b}\varvec{z}\) are linear combinations of sines and cosines. Thus, we can write each \(q_i\) (with \(i=1,2,3\)) as Fourier series with terms up to second-order harmonics (these being produced by the terms \(c_4 {y_2}^2\), \(c_5 y_2 y_3\) and \(c_6 {y_3}^2\)). We have
with the coefficients computed as usual when dealing with Fourier series:
Because we care mostly about the amplitude of each terms, it makes sense to rewrite this Fourier series in an amplitude-phase form as
with the following coefficients:
Using Taylor expansions in terms of \({\delta }\) and \({\zeta _{f}}\), we obtain the following:
which led to the approximations given by Eq. (3.6), where we also considered \(\psi =0\).
Effect of deviation of the equilibrium from the inflexion point (\(\psi \))
In the main text, we discussed the dynamics based on the vector field \(\dot{r} = d {\delta }r + a_0 r^3\) (Eq. (3.3a)) for \(a_0<0\). In Sect. 3.3.2, we showed that we indeed have \(a_0<0\), if we assume \(\psi =0\). In this appendix, we investigate the general case, where \(\psi \ne 0\). We first described how we obtained the evaporation rate equation and how we introduced \(\psi \) (see Sect. B.1). We then explain what \(\psi \) means (a measure of the deviation of the equilibrium relative to the inflexion point of the wall temperature profile) and evaluate what values it can take, depending on \(T_{g,sat,0}\), \(T_H\) and \(T_L\) (Sect. B.2). We then evaluate the sign of \(a_0\), as a function of \(\psi \) and other parameters (Sect. B.3). Although we expect to have \(a_0<0\) in typical conditions, we find that we can indeed have \(a_0>0\) for large enough \(\psi \). Finally, we study what the dynamics looks like when \(a_0>0\) (Sect. B.4).
1.1 Derivation of the evaporation rate expression, with the deviation parameter \(\psi \)
Here, we describe the original expression from the evaporation rate from Tessier-Poirier et al. [49] (Sect. B.1.1) and show how we obtain the new expression used here, which introduce \(\psi \) (Sect. B.1.2).
1.1.1 Expression of the evaporation rate
The dimensionless evaporation rate \(\dot{q}_3\) was derived in Tessier-Poirier et al. [49, Eq.(10b) p.10] as (we adapted the nomenclature to match the one used in the main text)
In Eq. (B.1), the parameters are the phase-change limit \({T_{HL}}\), the dimensionless wall axial temperature gradient at the inflexion point \(| \widetilde{T'_{wc}} |\), the dimensionless position of the inflexion point relative to the equilibrium \(\widetilde{x_{c}}\) and \({\widetilde{C_{th}}}\), a constant which represents the evaporation rate when the meniscus is at the inflexion point, when \(q_1=\widetilde{x_{c}}\). Those parameters are defined as
The dimensionless axial temperature gradient at the equilibrium \(|T'_{w,0} |\) is an important parameter, which can be expressed as a function of the gradient at the inflexion point as
Finally, note that linearization of Eq. (B.1) leads to \(\dot{q}_3 = -2 \, {\sigma }\, q_1\) (Eq. (2.9c)) where
the last equality involving dimensional quantities, obtained by substitution.
1.1.2 New form for the evaporation rate
Here, we will show how to transform Eq. (B.1) into the form given by Eq. (2.7c). We prefer this latter form because it involves less parameters and those are easier to study. Setting \(\dot{q}_3=0\), we find that
Substitution of this into the equation for \(\dot{q}_3\) eliminates the parameter \(\widetilde{x_{c}}\), we have
In \(a_0\) (Eq. (A.28)), the quantity \({\widetilde{C_{th}}}\) only appears in the ratio \(2{\widetilde{C_{th}}}/T_{HL}\), as argument of cosine and sine functions. It therefore makes sense to define the quantity \(\psi \equiv 2{\widetilde{C_{th}}}/T_{HL}\), which we use to eliminate \({\widetilde{C_{th}}}\), leading to
Knowing that \(| \widetilde{T'_{wc}} |\) is ultimately expressed in phase-change coefficient \({\sigma }\) when the equation is linearized to \(\dot{q}_3=-2 \, {\sigma }\, q_1\), it makes sense to replace it right away with \({\sigma }\). Substitution of \(| \widetilde{T'_{wc}} |\, \widetilde{x_{c}}=-\tan [\psi /2]\) in Eq. (B.4) gives
where we use the trigonometric identity \(1+\tan ^2 \theta = \sec ^2 \theta = 1/\cos ^2 \theta \).
Solving Eq. (B.6) for \(| \widetilde{T'_{wc}} |\), we get \(| \widetilde{T'_{wc}} |= 2{\sigma }/{T_{HL}}\, (\cos ^2\left[ \psi /2\right] )\). Substitution of the results in the \(\dot{q}_3\) equation above, we obtain
which is Eq. (2.7c) in the main text.
1.2 Studying the \(\psi \) expression
In Sect. B.2.1, we explain what \(\psi \) means physically. We also derive an expression of \(\psi \) in terms of \(T_{g,sat,0}\), \(T_H\) and \(T_L\). In Sect. B.2.2, we evaluate what values \(\psi \) can take.
1.2.1 \(\psi \) as a measure of the deviation of the equilibrium from the inflexion point
Both \(\widetilde{x_{c}}\) and \({\widetilde{C_{th}}}\) are dimensionless measures of the deviation of the equilibrium relative to the inflexion point. This is obvious for \(\widetilde{x_{c}}\), since \(x_{c}\) is directly the distance between the equilibrium and the inflexion point. The quantity \({\widetilde{C_{th}}}\) is also a measure of the deviation since it is proportional to \(\tfrac{1}{2} \left( T_H+T_L\right) - T_{g,sat,0}\), where \(\tfrac{1}{2} \left( T_H+T_L\right) \) is the temperature at the inflexion point and where \(T_{g,sat,0}\) is the saturation temperature, which is the temperature at the equilibrium (where \(\dot{q}_3=0\)). Thus, \({\widetilde{C_{th}}}\ne 0\) only when the equilibrium is away from the inflexion point, when \(T_{g,sat,0}\ne \tfrac{1}{2} \left( T_H+T_L\right) \). In fact, we see from Eq. (B.2c) that \(x_{c}\) is also expressed as a function of \(\tfrac{1}{2} \left( T_H+T_L\right) - T_{g,sat,0}\). Now, the new quantity \(\psi \), which we use to eliminate both \(\widetilde{x_{c}}\) and \({\widetilde{C_{th}}}\), also expresses the deviation of the equilibrium relative to the inflexion point. Given \(\psi =2{\widetilde{C_{th}}}/T_{HL}\), \(\psi \) is also proportional to \(\tfrac{1}{2} \left( T_H+T_L\right) - T_{g,sat,0}\) and the conclusions for \({\widetilde{C_{th}}}\) above also hold for \(\psi \).
Let’s now express \(\psi \) as a function of the temperature \(T_H\), \(T_L\) and \(T_{g,sat,0}\). Starting from \(\psi \equiv 2{\widetilde{C_{th}}}/T_{HL}\), after substitution of \({\widetilde{C_{th}}}\) (Eq. (B.2d)) and \(T_{HL}\) (Eq. (B.2a)), we get
Additional manipulations leads to
where \(\psi \) depends on three parameters, \(T_H\), \(T_L\) and \(T_{g,sat,0}\). Dividing numerator and denominator of each terms by \(T_H\), we obtain
which now depends only on two groups of parameters, the ratios \(T_L/T_H\) and \(T_{g,sat,0}/T_H\). The inequalities above are obtained by considering that \(0<T_L<T_{g,sat,0}<T_H\) (and temperature expressed in kelvin so always positive). Equation (B.8) for \(\psi \) corresponds to Eq. (2.8c) in the main text.
1.2.2 Values \(\psi \) takes, as a function of \(T_{g,sat,0}\), \(T_H\) and \(T_L\)
We now want to investigate what values can \(\psi \) take. We will find that \(\psi \) ranges between \(-\pi \) (when \(T_{g,sat,0}=T_H\), the equilibrium is as far as possible in the heat source region) and \(+\pi \) (when \(T_{g,sat,0}=T_L\), the equilibrium is as far as possible in the heat sink region).
Figure 10 shows \(\psi /\pi \) values as a surface along the ratios \(T_L/T_H\) and \(T_{g,sat,0}/T_H\). Given the inequalities \(0<T_L<T_{g,sat,0}<T_H\), we have boundaries in Fig. 10. First, the condition \(T_{g,sat,0}=T_H\) (the equilibrium is far into the heat source) gives the boundary \(T_{g,sat,0}/T_H=1\) which is a straight horizontal line in Fig. 10, and we find that \(\psi =-1\) (Eq. (B.8)) there. Second, the condition \(T_{g,sat,0}=T_L\) (the equilibrium is far into the heat sink) gives the boundary \(T_{g,sat,0}/T_H=T_L/T_H\) which gives the diagonal line in Fig. 10, and we find that \(\psi =+1\) (Eq. (B.8)) there. The inequalities \(0<T_L/T_H<1\) and \(0<T_{g,sat,0}/T_H<1\) (Eq. (B.8)) close the surface in Fig. 10.
It is useful to define an additional line corresponding to the case where the equilibrium is perfectly centered on the inflexion point (\(x_{c}=0\)). We then have \(T_{g,sat,0}=(T_H+T_L)/2\) and one can easily verify that \(\psi =0\) as expected, from Eq. (B.8). Also, dividing \(T_{g,sat,0}=(T_H+T_L)/2\) by \(T_H\), we find that
which corresponds to a function of the form \(y=mx+b\) in the graph Fig. 10. Additional isolines (for constant \(\psi \) values) can be added to the graph by solving Eq. (2.8c) for \(T_{g,sat,0}/T_H\), we obtain
which clearly produce isolines of the form \(y=mx+b\) in Fig. 10.
For \(\psi =\pm \pi /2\), we have
which are also shown in Fig. 10.
1.3 Sign of \(a_0\)
Here, we evaluate the sign of \(a_0\) in the general case where \(\psi \ne 0\). First, consider the phase-change nonlinearity only (Appendix. B.3.1). For the pressure nonlinearity only, \(a_0\) does not depend on \(\psi \) and we have \(a_0<0\) always. We then study the sign of \(a_0\) for both nonlinearities (Appendix B.3.2).
1.3.1 Phase-change nonlinearity
For the phase-change nonlinearity only (\({c_P}=0\), \({c_T}=1\)), we have \(a_0=a_{0,T}\) (Eq. (A.28)). We clearly have from Eq. (A.28) that \(a_{0,T}> 0\) only if \(1+{\zeta _{f}}^2 \cos \left[ \psi \right] <0\)Footnote 10. Two necessary conditions for having \(a_{0,T}> 0\) are having \({\zeta _{f}}>1\) (however, \({\zeta _{f}}\) is usually much smaller than one) and having either \(\psi <-\pi /2\) or \(\psi >+\pi /2\), so that the equilibrium is far enough from the inflexion point. Solving \(1+{\zeta _{f}}^2 \cos \left[ \psi \right] =0\) gives the curvesFootnote 11\({\zeta _{f}}= + \sqrt{-1/\cos [\psi ]}\) where \(a_{0,T}=0\) and which delimits regions of different signs for \(a_{0,T}\). Those boundaries are the same whatever the value of \(T_{HL}\) is.
We plot those curves in Fig. 11, with \({\zeta _{f}}\) and \(\psi \) as the axis. The region in the center corresponds to \(a_{0,T}<0\). There are two regions where \(a_{0,T}>0\) in the top corners, delimited by the curves \({\zeta _{f}}=+\sqrt{-1/\cos [\psi ]}\).
1.3.2 Pressure and phase-change nonlinearities
We now compute the boundaries \(a_0=0\) for both pressure and phase-change nonlinearities (\({c_P}=1\), \({c_T}=1\)) numerically. We find that they now depend on the value of \(T_{HL}\) (see Fig. 12). From Eq. (A.28), we see that \(a_{0,T}\propto 1/{T_{HL}}^2\), \(a_{P,T}\propto 1/{T_{HL}}\) and \(a_{0,P}\) is not a function of \(T_{HL}\). Thus, for very small \(T_{HL}\), the term \(a_{0,T}\) dominates and the boundaries (Fig. 12a) look very much like the one for the phase change only (Fig. 11). As we increase \(T_{HL}\), the boundaries move, the region for \(a_0>0\) in the left corner moves to the right, the one on the right corner moves to the left and a new one appears in the bottom right (see Fig. 12a–c). The two regions for \(a_0>0\) in the right eventually merge (Fig. 12c and d). As \(T_{HL}\) is further increase, the \(a_{0,P}\) terms starts to dominate, so the regions where \(a_0>0\) must eventually disappear. This is what we observe as both regions for \(a_0>0\) move out of the graph (Fig. 12d–f). The region for \(a_0<0\) eventually takes all the available space for large \(T_{HL}\).
1.4 Dynamics for \(a_0>0\)
Let’s now discuss how the dynamics is affected by having a positive term \(a_0\) in the vector field \(\dot{r}=d{\delta }r + a_0 r^3\) (Eq. (3.3a)). For \(a_0<0\), the function \(\dot{r}(r)\) was bending downwards. Now, for \(a_0>0\), the function \(\dot{r}(r)\) curves upwards (see Fig. 13a), the nonlinearities are not saturating, but instead, tends to increase \(\dot{r}\). The stability of the equilibrium point at \(r=0\) is still controlled by \({\delta }\). Although the slope at \(r=0\) is given by \(d{\delta }\), the sign of the slope only depends on \({\delta }\) because \(d>0\) always (so \(d{\delta }>0\) and \({\delta }>0\) means the same thing as well as \(d{\delta }<0\) and \({\delta }<0\)). For \(d{\delta }>0\), the amplitude grows faster than the linear case (\(\dot{r}\) is larger for \(a_0>0\)) and without bounds, the system never reaches a limit cycle. For \(d{\delta }<0\), the equilibrium is stable, but an unstable fixed point (corresponding to a unstable limit cycle in either the \(z_i\) or the \(q_i\) variables) exists at \(r= { r_{LC} } \). For \(r< { r_{LC} } \), the amplitude decreases toward \(r=0\) and for \(r> { r_{LC} } \), increases without bounds. As \({\delta }\) goes from positive to negative, the unstable limit cycle is created, a process called a subcritical Poincaré-Andronov-Hopf bifurcation.
Now, we have to keep in mind that the full vector field also includes higher order terms (\(\mathcal {O}(r^5)\), see Eq. (A.26)). The equilibrium stability is still controlled by \({\delta }\). The truncated form \(\dot{r}=d{\delta }r + a_0 r^3\) is sufficient to predict the limit cycle bifurcation but higher order terms are, however, required to predict the dynamics further. In Fig. 13b, we show how the vector fields \(\dot{r}\) can look like when a negative quintic term is included. At small r, the vector fields looks like Fig. 13a, but differs qualitatively for larger r. At \(d{\delta }\) negative and large enough, there is only a stable equilibrium (see \(\dot{r}=-3r+r^3-0.2r^5\) curve). At \(d{\delta }\) negative but small enough, there is one unstable limit cycle but also one larger stable limit cycle, due to the quintic term (see curve \(\dot{r}=-r+r^3-0.2r^5\)). At positive \(d{\delta }\), there is now only one large stable limit cycle (see curve \(\dot{r}=r+r^3-0.2r^5\)). Thus, adding a quintic term may prevent the amplitude from growing without bounds, by creating a stable limit cycle surrounding the unstable limit cycle. Let’s note that there is still the creation of an unstable limit cycle as \({\delta }\) becomes negative, through a subcritical Poincaré–Andronov–Hopf bifurcation. However, there is an additional, global, bifurcation that can occur. For \(d{\delta }\) negative and sufficiently small (curve \(\dot{r}=-3r+r^3-0.2r^5\)), increasing \(d{\delta }\) leads to the creation of an unstable limit cycle and a larger stable limit cycle (curve \(\dot{r}=-r+r^3-0.2r^5\)). This process is called a a saddle-node bifurcation of cycles (see Strogatz [44, Subcritical Hopf Bifurcation p.251, Saddle-node Bifurcation of Cycles p.261]).
The behavior we observe numerically for \(a_0>0\) qualitatively corresponds to the behavior with a negative quintic term described above. Figure 14 shows numerical simulations of Eq. (2.14) for various initial conditions. In Fig. 14a, \(d{\delta }\) is small and negative, the dynamics suggest the existence of a stable equilibrium, an unstable limit cycle and a larger stable limit cycle, just like the curve \(\dot{r}=-r+r^3-0.2r^5\) in Fig. 13b. In Fig. 14b, \(d{\delta }\) is positive, the dynamics suggest the existence of a unstable equilibrium, and one large and stable limit cycle, just like the curve \(\dot{r}=r+r^3-0.2r^5\) in Fig. 13b.
Let’s now provide a few final remarks on the vector field with a negative quintic term as shown in Fig. 13b. If the system behaves as the curve \(\dot{r}=-r+r^3-0.2r^5\), the unstable limit cycle acts as a barrier: Self-oscillations are possible even though the equilibrium is stable, but an initial push beyond the unstable limit cycle is needed. Also, the system described by the vector field with a negative quintic term has an hysteresis: As \(d{\delta }\) is increased above 0, large oscillations starts (given a small perturbation) but can only be turned off by bringing \(d{\delta }\) below some negative value. The described behavior with a negative quintic term could be made possible by other nonlinearities than the one considered in this paper. For example, having a phase change which suddenly increases when the amplitude is large enough could produce such behavior.
Rights and permissions
About this article
Cite this article
Tessier-Poirier, A., Rand, R.H. & Fréchette, L.G. What limits the oscillations’ amplitude in the single-branch pulsating heat pipe. Nonlinear Dyn 108, 27–59 (2022). https://doi.org/10.1007/s11071-021-07188-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11071-021-07188-3