Hostname: page-component-7b7567c8c-7pwfn Total loading time: 0 Render date: 2026-11-16T02:19:06.613Z Has data issue: false hasContentIssue false

Deterministic and stochastic bifurcations in two-dimensional electroconvective flows

Published online by Cambridge University Press:  12 July 2021

Zhe Feng
Affiliation:
Department of Mechanical Engineering, National University of Singapore, 9 Engineering Drive 1, 117575 Republic of Singapore
Mengqi Zhang*
Affiliation:
Department of Mechanical Engineering, National University of Singapore, 9 Engineering Drive 1, 117575 Republic of Singapore
Pedro A. Vazquez
Affiliation:
Departamento de Física Aplicada III, Universidad de Sevilla, ESI, Camino de los Descubrimientos s/n, 41092 Sevilla, Spain
Chang Shu
Affiliation:
Department of Mechanical Engineering, National University of Singapore, 9 Engineering Drive 1, 117575 Republic of Singapore
*
†Email address for correspondence: mpezmq@nus.edu.sg

Abstract

We investigate deterministic and stochastic bifurcations in electroconvecitve flows of a dielectric liquid confined between two parallel plates subjected to a strong unipolar injection by direct numerical simulations. A long-standing discrepancy of linear instability criteria between the experiment and theory exists in this flow. We here test the hypothesis that the discrepancy may be related to the inhomogeneity in ion-exchange membranes used in experiments, contrasted by the homogeneous ion injection assumed in theoretical and numerical analyses. To study this effect, we consider stochastic boundary conditions around linear criticality and first bifurcations in this flow. For a complete presentation of flow bifurcations, deterministic bifurcation analysis (without stochasticity) is first performed to investigate primary bifurcations in this flow by progressively increasing the strength of electric field. Lyapunov spectrum and dimension are calculated and probed to characterise the chaotic motion therein. Our results confirm the high dimensionality of chaos in electroconvective flows and reveal for the first time that its chaos is extensive in a range of finite-sized systems. We then conduct stochastic bifurcation analyses by considering random perturbations in the boundary conditions of charge density and electric potential. Owing to the subcritical nature of electroconvective flows, the linear instability criteria under stochastic boundaries are closer to the experimental values than former theoretical and numerical results (assuming the homogeneous charge injection) for different levels of stochasticity, which confirms the hypothesis aforementioned. Furthermore, stochasticity can also enhance the efficiency of ionic transport.

Information

Type
JFM Papers
Copyright
© The Author(s), 2021. Published by Cambridge University Press
Figure 0

Figure 1.Schematic of stochastic perturbations $g_i(x)$ on bottom and top plates with perturbation amplitude $\sigma =10\,\%$ and two different correlation lengths. Solid line, $l_c=1.0$; dash–dotted line, $l_c=0.25$. Smaller correlation length can induce more wavy oscillations. The two dashed ellipses refer to the electroconvective flow within the dielectric liquid. Stochastic boundaries may induce asymmetric motions.

Figure 1

Figure 2.Hydrostatic solutions of charge density $\bar {Q}$ for electroconvective flows in deterministic and stochastic cases with $C=10$ and $Fe=10^4$: (a) base field $\bar {Q}$ under deterministic boundaries; (b) base field $\bar {Q}$ under stochastic boundaries with $\sigma =2\,\%$ and $l_c=0.5$. Only part of the field $0< y<0.2$ is presented for a clear visualisation (the charge density in the range of $0.2< y<1$ is relatively small); (c) comparison of profiles for $\bar {Q}$ between deterministic (black line) and stochastic cases (grey lines, at 12 different $x$ stations, separated equally).

Figure 2

Table 1.Grid independency test for electroconvective flows by time-averaged electric Nusselt numbers $Ne_{avg}$ at different electric Rayleigh numbers $T$ within the domain range $[0,1.228]$ in the $x$ direction and $[0,1]$ in the $y$ direction.

Figure 3

Figure 3.Validation of hydrostatic solutions of charge density $Q$ and electric field $E_y$ for electroconvective flows at $M=10, Fe=10^4$ and two different injection strengths: (a) $C=10$; (b) $C=20$. Circle and square symbols refer to results of charge density and normal electric field from the current numerical simulations, and the solid line is obtained by an ODE solver solving the electric equations.

Figure 4

Figure 4.Subcritical bifurcation of electroconvective flows represented by the electric Nusselt number $Ne$ at $C=10$, $M=10$ and $Fe=10^4$: upward-pointing triangles show the continuous increasing of $T$, while the downward-pointing triangles present the case of continuous decreasing. The two instability criteria are numerically predicted as $T_c=162.5$ and $T_f=110.4$.

Figure 5

Figure 5.Growth rate $\omega _r$ of the least stable mode in electroconvective flows at $M=10$. The circle, diamond and square symbols refer to results from the current time-stepping Arnoldi method for the $C=40$ and $Fe=10^6$, $C=10$ and $Fe=10^4$, and $C=10$ and $Fe=10^7$ cases, respectively. The solid lines are the corresponding linear fitted results and the black solid dots are the $x$-intercepts of the linear fitted lines where $\omega _r=0$. The critical $T_c$ values for the three cases are: $T_c=160.80$; $T_c=162.56$; $T_c=164.04$.

Figure 6

Figure 6.Eigenfunctions of least stable mode for electroconvective flows at $Fe=10^4$ and $T=165$, $C=10$, $M=10$ with periodic lateral boundaries: (a) current result of eigenfunction $u$; (b) comparison of profiles for $u$ between the current result at $x=0.1$ (circles) and linear stability analysis (solid line) from the work of Zhang et al. (2015); (c) current result of eigenfunction $v$; (d) comparison of profiles for $v$ between the current result at $x=1.05$ (circles) and linear stability analysis (solid line).

Figure 7

Figure 7.Temporal evolution of electric Nusselt number $Ne$ and corresponding flow pattern for electroconvective flows under periodic lateral boundaries: (a) time series of $Ne$ at $T=170$. The inset shows the value of $Ne-1$ on a semi-logarithmic scale. The growth rate $\omega (Ne)=0.419$ is very close to the linear stability analysis ($=0.409$); (b) contour of charge density $Q$ (i.e. the colour background) and two steady rolls (i.e. black lines) at $t=100$; (c) time series of $Ne$ at $T=270$. The growth rate $\omega (Ne)=4.504$ in the inset is very close to the linear stability analysis (${=}4.519$); (d) contour of $Q$ and two oscillating rolls at $t=230$.

Figure 8

Figure 8.Temporal evolution of electric Nusselt number $Ne$ for electroconvective flows under periodic lateral boundaries: (a) time series of $Ne$ at $T=325$. The growth rate $\omega (Ne)=6.050$ in the inset is very close to the linear stability analysis (${=}6.070$); (b) zoom-in of the time series in panel (a).

Figure 9

Figure 9.Temporal development of streamline patterns within the electroconvective flow of $T=325$ at the six instants shown in figure 8(b). Note that the periodic boundary condition is applied on the lateral walls. The background colour contour refers to the distribution of charge density $Q$.

Figure 10

Figure 10.Deterministic bifurcation diagrams for electroconvective flows with periodic lateral boundaries under different grid resolutions. The steady states are denoted by the solid line, while unsteady states are represented by solid circles which show all the possible local maxima and minima amplitudes of $Ne$ at a specific $T$: (a) element size $12\times 16$, the Hopf bifurcation takes place at approximately $T_{c1}=258$ and chaos of EC flow emerges at approximately $T_{c2}=296$; (b) element size $18\times 24$, the Hopf bifurcation takes place at approximately $T_{c1}=274$ and chaos of EC flow emerges at approximately $T_{c2}=300$.

Figure 11

Table 2.Largest Lyapunov exponents $\lambda _1$ and Lyapunov dimensions $D_{\lambda }$ of electroconvective flows at different electric Rayleigh numbers $T$ with periodic lateral boundaries and element size $12\times 16$.

Figure 12

Figure 11.Computation of Lyapunov exponents for electroconvective flows at $\varGamma =2$ and $T=900$: (a) temporal history of instantaneous largest Lyapunov exponent (solid line) and its finite-time largest Lyapunov exponent (dashed line); (b) spectrum of Lyapunov exponents $\lambda _k$ and their cumulative summation $S_k$, where $k=1,2,\ldots ,m$ refers to the index and $m=75$ is the number of Lyapunov exponents computed; (c) instantaneous (solid line) and averaged (dashed line) Lyapunov dimensions.

Figure 13

Table 3.Largest Lyapunov exponent $\lambda _1$ and Lyapunov dimensions $D_{\lambda }$ of electroconvective flows at different electric Rayleigh numbers with no-slip lateral boundaries and aspect ratio $\varGamma =0.614$.

Figure 14

Figure 12.Contour of charge density $Q$ for electroconvective flows at $T=900$ with no-slip lateral boundaries and different aspect ratios $\varGamma$: (a) $\varGamma =1$; (b) $\varGamma =2$; (c) $\varGamma =3$; (d) $\varGamma =4$.

Figure 15

Figure 13.Relation between the Lyapunov dimension $D_{\lambda }$ and aspect ratio $\varGamma$ for electroconvective flows. Circles refer to the result of the current numerical simulations and the solid line is the linear fitted result.

Figure 16

Figure 14.Temporal evolution of electric Nusselt number for electroconvective flows under deterministic (dashed line) and stochastic boundaries (solid line) with $\sigma =1\,\%$ and $l_c=0.5$: (a) time series at $T=150$ for both cases (there is no electroconvection in the deterministic case); (b) time series at $T=170$ for the stochastic and deterministic cases. All the simulations use $\boldsymbol {U}=\phi =Q=0$ as the initial condition.

Figure 17

Figure 15.Temporal development of streamline patterns within the electroconvective flow of $T=150$ under stochastic boundaries at the four instants shown in figure 14(a). The background contour refers to the distribution of charge density $Q$.

Figure 18

Figure 16.Stochastic bifurcation diagrams of electroconvective flows around the linear and nonlinear stability criteria with periodic lateral boundaries at $C=10$, $M=10$ and $Fe=10^4$. Presented are samples of $Ne$ versus $T$ corresponding to different choices of $\sigma$ and $l_c$. The solid line represents the deterministic bifurcation diagram around $T_c=162.5$ and $T_f=110.4$, and the red and blue dots refer to the $Ne$ obtained under stochastic boundary conditions. Red dots are converged $Ne$ computed by gradually increasing $T$ from the hydrostatic state and blue dots are obtained by gradually decreasing $T$ from the steady state of $T=200$. The stochastic boundary varies at different $T$ for obtaining a more general stochastic bifurcation diagram.

Figure 19

Figure 17.Probability density function $p(Ne)$ of the electric Nusselt number for electroconvective flows under stochastic boundaries with $\sigma =1.0\,\%$ and $l_c=0.5$ at six different electric Rayleigh numbers. For comparison, the dashed line represents the value of $Ne$ under deterministic boundary conditions. At each $T$, 2000 different stochastic boundaries are performed for statistical calculation.

Figure 20

Figure 18.Growth rate of the least stable mode for electroconvective flows under stochastic boundary conditions with different perturbation amplitudes $\sigma$ and correlation lengths $l_c$. Circle, diamond and square symbols refer to the result from the current time-stepping Arnoldi method, the solid lines are the corresponding linear fitted results and the black solid dots are the $x$ intercept of the linear fitted line where $\omega _r=0$. (a) $\sigma =0.1\,\%$: $T_c=160.36$ at $l_c=0.25$, $T_c=162.48$ at $l_c=0.5$, $T_c=162.55$ at $l_c=1.0$; (b) $\sigma =2.0\,\%$: $T_c=117.46$ at $l_c=0.25$, $T_c=144.73$ at $l_c=0.5$, $T_c=157.06$ at $l_c=1.0$; (c) $\sigma =5.0\,\%$: $T_c=90.25$ at $l_c=0.25$, $T_c=122.16$ at $l_c=0.5$, $T_c=147.20$ at $l_c=1.0$.

Figure 21

Figure 19.Deterministic bifurcation diagram for electroconvective flows with symmetric lateral boundaries. The Hopf bifurcation for the one-roll pattern takes place at approximately $T_{c1}=295$, and the one roll is split into two rolls at approximately $T_{c2}=311$. The Hopf bifurcation for two-roll pattern ensues at approximately $T_{c3}=660$.

Figure 22

Figure 20.Contour of charge density and streamline patterns of electroconvective flows with symmetric lateral boundaries. To clearly show at least one complete charge-void region, the flow domain is unfolded with respect to $x=0$ owing to the symmetric boundary in the lateral direction: (a) $T=190$, one-roll flow structure emerges within $[0,0.614]$ of the $x$ axis; (b) $T=340$, two-roll flow structure takes place within $[0,0.614]$ of the $x$ axis.

Figure 23

Figure 21.Power spectral density for the time series of electric Nusselt number at different electric Rayleigh numbers with symmetric lateral boundaries. The spectrum starts to become broadband around $T=920$.

Figure 24

Table 4.Largest Lyapunov exponents $\lambda _1$ and Lyapunov dimensions $D_{\lambda }$ of electroconvective flows at different electric Rayleigh numbers $T$ with symmetric lateral boundaries.

Figure 25

Figure 22.Distribution of the impulse disturbance added into the hydrostatic field of electroconvective flows: (a) streamwise velocity $U(x,y,t=0)$; (b) normal velocity $V(x,y,t=0)$. Their initial perturbation amplitude is very small. Only part of the flow field is presented here for a clear visualisation as the impulse is distributed around the central location $x_0=0.5$ and $y_0=60$.

Figure 26

Figure 23.Linear response of the impulse disturbance: (a) variation of temporal growth rate $\omega _r$ with the streamwise wavenumber $\alpha$ at $T=190$. The circles refer to the current result, and the solid line is the result from the local linear stability analysis (Zhang et al.2015); (b) relation between the wavenumber of the least stable mode $\alpha _l$ and the electric Rayleigh number $T$. Note that the step-shape result for small values of $\alpha _l$ is related to the smallest increment of wavenumber. A longer streamwise length $L$ can mitigate this problem. However, the comparison of results between DNS (circles) and the linear stability analysis (the solid line) clearly indicates the trend that the wavenumber of the least stable mode $\alpha _l$ increases with $T$.