1: \documentclass[pre,twocolumn]{revtex4}
2:
3: \usepackage{graphicx}
4: \usepackage{dcolumn}
5: \usepackage{amsmath}
6: \usepackage{amssymb}
7: \usepackage{amsfonts}
8:
9: %a better dot product
10: \newcommand{\dotprod}{{\scriptscriptstyle \stackrel{\bullet}{{}}}}
11: \def\eqdef{\mathrel{\mathop=^{\rm def}}}
12:
13: \begin{document}
14:
15: \title{Pattern Formation and Dynamics in Rayleigh-B\'{e}nard Convection:
16: Numerical Simulations of Experimentally Realistic Geometries}
17:
18: \author{M.R. Paul}
19: \email{mpaul@caltech.edu}
20: \homepage{http://www.cmp.caltech.edu/~stchaos}
21: \author{K.-H. Chiam}
22: \author{M.C. Cross}
23: \affiliation{Department of Physics, California Institute of
24: Technology 114-36, Pasadena, California 91125}
25:
26: \author{P.F. Fischer}
27: \affiliation{Mathematics and Computer Science Division, Argonne
28: National Laboratory, Argonne, Illinois 60439}
29:
30: \author{H. S. Greenside}
31: \affiliation{Department of Physics, Duke University, Durham, North
32: Carolina 27708-0305}
33:
34: \date{\today}
35:
36: \begin{abstract}
37: Rayleigh-B\'{e}nard convection is studied and quantitative
38: comparisons are made, where possible, between theory and
39: experiment by performing numerical simulations of the Boussinesq
40: equations for a variety of experimentally realistic situations.
41: Rectangular and cylindrical geometries of varying aspect ratios
42: for experimental boundary conditions, including fins and spatial
43: ramps in plate separation, are examined with particular attention
44: paid to the role of the mean flow. A small cylindrical convection
45: layer bounded laterally either by a rigid wall, fin, or a ramp is
46: investigated and our results suggest that the mean flow plays an
47: important role in the observed wavenumber. Analytical results are
48: developed quantifying the mean flow sources, generated by
49: amplitude gradients, and its effect on the pattern wavenumber for
50: a large-aspect-ratio cylinder with a ramped boundary. Numerical
51: results are found to agree well with these analytical predictions.
52: We gain further insight into the role of mean flow in pattern
53: dynamics by employing a novel method of quenching the mean flow
54: numerically. Simulations of a spiral defect chaos state where the
55: mean flow is suddenly quenched is found to remove the time
56: dependence, increase the wavenumber and make the pattern more
57: angular in nature.
58: \end{abstract}
59:
60: \pacs{47.54.+r,47.52.+j,47.20.Bp,47.27.Te}
61:
62: \maketitle
63:
64: \section{Introduction}
65: \label{section:introduction} Rayleigh-B\'{e}nard convection has
66: played a crucial role in guiding both theory and experiment
67: towards an understanding of the emergence of complex dynamics from
68: nonequilibrium systems~\cite{cross:1993}. However, an important
69: missing link has been the ability to make quantitative and
70: reliable comparisons between theory and experiment.
71:
72: Nearly all previous three-dimensional convection calculations have
73: been subject to a variety of limitations. Many simulations have
74: been for small aspect ratios where the lateral boundaries dominate
75: the dynamics, and as a result, complicate the analysis. When
76: larger aspect ratios are considered, it is often with the
77: assumption of periodic boundaries, which is convenient numerically
78: yet does not correspond to any laboratory experiment. As a result
79: of algorithmic inefficiencies, or the lack of computer resources,
80: simulations have frequently not been carried out for long times.
81: This presents the difficulty in determining whether the observed
82: behavior represents the asymptotic non-transient state, which is
83: usually the state that is most easily understood theoretically.
84:
85: Fortunately, advances in parallel computers, numerical algorithms
86: and data storage are such that direct numerical simulations of the
87: full three-dimensional time dependent equations are possible for
88: experimentally realistic situations. We have performed simulations
89: with experimentally correct boundary conditions, in geometries of
90: varying shapes and aspect ratios over long enough times so as to
91: allow a detailed quantitative comparison between theory and
92: experiment.
93:
94: Alan Newell has made numerous important contributions to the
95: discussion of pattern formation in non-equilibrium systems. In
96: this paper, presented in this special issue in his honor, we give
97: a survey of our recent results that touch on many of the issues he
98: has raised, and in turn make use of some of the tools that he has
99: helped develop to understand our simulations.
100:
101: \section{Simulation of Realistic Geometries}
102: \label{section:numerical simulation} We have performed full
103: numerical simulations of the fluid and heat equations using a
104: parallel spectral element algorithm (described in detail elsewhere
105: \cite{fischer:1997}). The velocity $\vec{u}$, temperature $T$, and
106: pressure $p$, evolve according to the Boussinesq equations,
107: \begin{eqnarray}
108: {\sigma}^{-1} \left(
109: {\partial}_t + \vec{u} \dotprod \vec{\nabla} \right) \vec{u}
110: &=& -\vec{\nabla} p + RT \hat{z} + \nabla^2 \vec{u} , \label{eq:mom}\\
111: \left( {\partial}_t + \vec{u} \dotprod \vec{\nabla} \right) T
112: &=& \nabla^2 T , \label{eq:energy}\\
113: \vec{\nabla} \dotprod \vec{u} &=& 0,
114: \label{eq:mass}
115: \end{eqnarray}
116: where $\partial_t$ indicates time differentiation, $\hat{z}$ is a
117: unit vector in the vertical direction opposite of gravity, $R$ is
118: the Rayleigh number, and $\sigma$ is the Prandtl number. The
119: equations are nondimensionalized in the standard manner using the
120: layer height $h$, the vertical diffusion time for heat ${\tau}_v
121: \equiv h^2/\kappa$ where $\kappa$ is the thermal diffusivity, and
122: the temperature difference across the layer $\Delta T$, as the
123: length, time, and temperature scales, respectively.
124:
125: We have investigated a wide range of geometries including
126: cylindrical and rectangular domains, which are the most common
127: experimentally, in addition to elliptical and annular domains.
128: Rotation about the vertical axis of the convection layer for any
129: of these situations is also possible but will not be presented
130: here. All bounding surfaces are no-slip, $\vec{u}=0$, and the
131: lower and upper surfaces and are held at constant temperature,
132: $T(z=0)=1$ and $T(z=1)=0$.
133:
134: A variety of sidewall boundary conditions are shown in
135: Fig.~\ref{fig:sidewalls}. Common thermal boundary conditions on
136: the lateral sidewalls are insulating, $\hat{n} \dotprod
137: \vec{\nabla}T = 0$ where $\hat{n}$ is a unit vector normal to the
138: boundary at a given point, and conducting, $T=1-z$. In the future
139: we will have the flexibility of imposing a more experimentally
140: accurate thermal boundary condition by coupling the fluid to a
141: lateral wall of finite thickness and known finite thermal
142: conductivity that is bounded on the outside by a vacuum.
143: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
144: \begin{figure}[tbh]
145: \begin{center}
146: \includegraphics[width=2.5in]{./fig1.eps}
147: \end{center}
148: \caption{Four lateral sidewall boundary conditions utilized in the
149: numerical simulations. The two thermal boundary conditions are
150: conducting and insulating whereas the fin and ramp represent
151: geometric conditions employed in experiments.}
152: \label{fig:sidewalls}
153: \end{figure}
154: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
155:
156: In experiment, however, small sidewall thermal forcing can have a
157: significant effect upon the resulting patterns and, as a result,
158: finned boundaries have been
159: employed~\cite{daviaud:1989,debruyn:1996,pocheau:1997}. These are
160: formed by inserting a very thin piece of paper or cardboard
161: between the top and bottom plates near the sidewalls. This
162: suppresses convection over the finned region ($R \sim h^3$ and the
163: layer height has effectively been reduced) whereas in the bulk of
164: the domain, i.e. the un-finned region, supercritical conditions
165: prevail. This is accomplished numerically by extending a no-slip
166: surface into the domain from the lateral sidewall. In all of our
167: simulations we have chosen the vertical position of the fin to be
168: $z=0.5$ but this is not necessary. The result is that the
169: supercritical portion of the convection layer is bounded by a
170: subcritical region of the same fluid and hence with the same
171: material properties. An additional effect is that the mean flow
172: may extend into the finned region which presents an interesting
173: scenario for exploring the effect of mean flows upon pattern
174: dynamics~that has been investigated both experimentally and
175: theoretically by Pocheau and Daviaud~
176: \cite{daviaud:1989,{pocheau:1997}} and is discussed further below.
177:
178: The sidewalls can also have an orienting effect and ramped
179: boundaries have been used as a ``soft boundary"~\cite{kramer:1982}
180: in an effort to minimize this. By gradually decreasing the plate
181: separation as the lateral sidewall is approached the convection
182: layer eventually becomes critical and then increasingly
183: subcritical. Using the spectral element algorithm we are able to
184: investigate arbitrary ramp shapes: we have chosen to investigate
185: the precise radial ramp utilized in recent
186: experiments~\cite{bajaj:1999,{ahlers:2001}} on a cylindrical
187: convection layer. Again the mean flow is able to extend into the
188: subcritical region.
189:
190: Perhaps the most common method employed experimentally to reduce
191: the influence of sidewalls is to use a large aspect ratio
192: $\Gamma$, where $\Gamma=r/h$ in a cylindrical domain where $r$ is
193: the radius and $\Gamma=L/h$ in a square domain where $L$ is the
194: length of side. Experiments can attain aspect ratios as large as
195: $\sim 500$. However, the majority of large aspect ratio
196: experiments are for $\Gamma \lesssim 100$. We have performed
197: numerical simulations using the spectral element algorithm for
198: $\Gamma \sim 60$ as shown in Fig.~\ref{fig:large_gamma}.
199:
200: The top panel in Fig.~\ref{fig:large_gamma} illustrates the
201: convection pattern present for the parameters of the classic
202: paper~\cite{ahlers:1974} where flow visualization was not
203: possible. Although the simulation has only been performed for a
204: short time $t_f \sim 100 \tau_v$ it appears that a slow process of
205: domain coarsening~\cite{cross:1995:physrevlett} is occurring. The
206: bottom of Fig.~\ref{fig:large_gamma} illustrates the time
207: dependent spatiotemporal chaotic state of spiral defect
208: chaos~\cite{morris:1993}. These, and other, interesting large
209: aspect ratio problems can now be addressed through the use of
210: numerical simulation.
211:
212: Heuristically, using the spectral element algorithm on an IBM SP
213: parallel supercomputer, it is our experience that it is practical
214: to perform full numerical simulations for aspect ratios $\Gamma
215: \sim 30$ for simulation times of $t_f \sim \tau_h$ (36 hours on 64
216: processors), where $\tau_h$ is the horizontal diffusion time for
217: heat ${\tau}_h={\Gamma}^2 {\tau}_v$, and $\Gamma \sim 60$ for $t_f
218: \sim 300 \tau_v$ (36 hours on 256 processors) for $\epsilon
219: \lesssim 1$, $0.5 \lesssim \sigma \lesssim 10$, $\Delta t \approx
220: 0.01$, and approximately cubic shaped spectral elements with an
221: edge length of unity and $11^{th}$ order polynomial expansions
222: (where $\epsilon=(R-R_c)/R_c$ and $R_c$ is the critical value of
223: the Rayleigh number). Of course for smaller domains the
224: computational requirements significantly decrease.
225:
226: A major benefit of numerical simulations is that a complete
227: knowledge of the flow field is produced. For example, we have
228: first used this to address a long standing open question
229: concerning chaos in small cylindrical domains. The existence of a
230: power-law behavior in the fall-off of the power spectral density
231: derived from a time series of the Nusselt number was not
232: understood~\cite{ahlers:1974}. The Nusselt number, $N(t)$, is a
233: global measurement of the temperature difference across the fluid
234: layer. In cryogenic experiments very precise measurements of
235: $N(t)$ are
236: possible~\cite{ahlers:1974,{ahlers:1978},{ahlers:1980},{libchaber:1978}},
237: however the flow field can not be visualized easily. Subsequent
238: room temperature experiments using compressed gasses allowed flow
239: visualization at the expense of precise measurements of the
240: Nusselt
241: number~\cite{pocheau:1985,{croquette:1986},{pocheau:1989}}.
242: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
243: \begin{figure}[tbh]
244: \begin{center}
245: \includegraphics[width=2.5in]{./fig2_top.eps}
246: \includegraphics[width=1.32in]{./fig2_bottom.eps}
247: \end{center}
248: \caption{Numerical simulations of two large-aspect-ratio
249: cylindrical convection layers. The pattern is illustrated by
250: contours of the thermal perturbation, dark regions represent cool
251: descending fluid and light regions warm ascending fluid. Both
252: simulations are initiated from random thermal perturbations and
253: the lateral sidewalls are insulating. (Top) $\Gamma=57$,
254: $\sigma=2.94$, $R=2169.2$ and $t=74 \tau_v$. (Bottom) A spiral
255: defect chaos state, $\Gamma=30$, $\sigma=1.0$, $R=2950$ and $t=254
256: \tau_v$. } \label{fig:large_gamma}
257: \end{figure}
258: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
259:
260: By performing long-time simulations, on the order of many
261: horizontal diffusion times, for the same parameters in cylindrical
262: domains with $\sigma=0.78$ and for a range of $\epsilon$, with
263: realistic boundary conditions, we had access to both precise
264: measurements of the Nusselt number, Fig.~\ref{fig:nu_all_cond},
265: and flow visualization, Fig.~\ref{fig:snapshots_mf}, allowing us
266: to resolve the issue~\cite{paul:2001}. Conducting sidewalls were
267: used and all simulations were initiated from small, $\delta T
268: \approx 0.01$, random thermal perturbations. Flow visualization of
269: the simulations represented in Fig.~\ref{fig:nu_all_cond} display
270: a rich variety of dynamics similar to what was observed in the
271: room temperature experiments. Using simulation results, the
272: particular dynamical events responsible for the $N(t)$ signature
273: were identified. The power-law behavior was found to be caused by
274: the nucleation of dislocation pairs and roll pinch-off events.
275: Additionally, the power spectral density was found to decay
276: exponentially for large frequencies as expected for
277: time-continuous deterministic dynamics. The large frequency regime
278: was not accessible to experiment because of the presence of the
279: noise floor.
280: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
281: \begin{figure}[tbh]
282: \begin{center}
283: \includegraphics[width=2.5in]{./fig3.eps}
284: \end{center}
285: \caption{Plots of the dimensionless heat transport N(t) for
286: reduced Rayleigh number $\epsilon = 0.557,0.614,0.8,1.0,1.5$, and
287: $3.0$, labelled (i-vi) respectively ($\Gamma = 4.72$). For cases
288: (i-v), $\Delta t =0.01$, and for case~(vi), $\Delta t = 0.005$
289: ($\Delta t$ is the time step).} \label{fig:nu_all_cond}
290: \end{figure}
291: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
292: \section{Role of mean flow} \label{section:mean_flow} The mean flow
293: present in these flow fields, and in general for $\sigma \lesssim
294: 1$, plays an important role in
295: theory~\cite{newell:1990:jfm,{cross:1984}} yet it is not possible
296: to measure or visualize the mean flows in the current generation
297: of experiments. In our simulations, however, we can quantify and
298: visualize the mean flow.
299: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
300: \begin{figure}[tbh]
301: \begin{center}
302: \includegraphics[width=2.5in]{./fig4.eps}
303: \end{center}
304: \caption{Flow visualization showing the pattern (solid dark lines)
305: and shaded contours of the vorticity potential, $\zeta$, for
306: $\epsilon=0.614$ labelled ii) in Fig~\ref{fig:nu_all_cond}
307: ($\Gamma = 4.72$). Dark regions corresponding to negative
308: vorticity generate clockwise mean flow and light regions to a
309: positive vorticity generating a counter clockwise mean flow. The
310: dark solid lines are zeros of the thermal perturbation at
311: mid-depth illustrating the outline of the convection rolls. From
312: top to bottom and left to right the panels are for $t =
313: 600,605,630,650,735,785$. The dislocations glide toward the right
314: wall focus (shown here); during the next half period, the
315: dislocations glide to the left wall focus. This left and right
316: alternation continues for the entire simulation.}
317: \label{fig:snapshots_mf}
318: \end{figure}
319: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
320:
321: The mean flow field, $\vec{U}(x,y)$, is the horizontal velocity
322: integrated over the depth and originates from the Reynolds stress
323: induced by pattern distortions. Recalling the fluid equations,
324: Eqs.~(\ref{eq:mom}) and~(\ref{eq:mass}), it is evident that the
325: pressure is not an independent dynamic variable. The pressure is
326: determined implicitly to enforce incompressibility,
327: \begin{equation} \label{eq:pressure}
328: \nabla^2 p = -\sigma^{-1} \vec{\nabla} \dotprod \left[ \left(
329: \vec{u} \dotprod \vec{\nabla} \right) \vec{u} \right] + R
330: \partial_z T.
331: \end{equation}
332: Focussing on the nonlinear Reynolds stress term and rewriting the
333: pressure as $p = p_o(x,y) + \bar{p}(x,y,z)$ yields,
334: \begin{equation} \label{eq:pressure_p0}
335: p_o(x,y) \sim \sigma^{-1} \int dx' dy' \ln \left( 1 / \left| r-r'
336: \right| \right) \left< \vec{\nabla}' \dotprod \left[ \left(
337: \vec{u} \dotprod \vec{\nabla} \right) \vec{u} \right] \right>_z .
338: \end{equation}
339: In Eq.~(\ref{eq:pressure_p0}) the $\ln(1/|r-r'|)$ is not exact, in
340: order to be more precise the finite system Green's function would
341: be required. However, the long range behavior persists. This gives
342: a contribution to the pressure that depends on distant parts of
343: the convection pattern. The Poiseuille-like flow driven by this
344: pressure field subtracts from the Reynolds stress induced flow
345: leading to a divergence free horizontal flow that can be described
346: in terms of a vertical vorticity.
347:
348: The mean flow is important not because of its strength; under most
349: conditions the mean flow is substantially smaller than the
350: magnitude of the roll flow making it extremely difficult to
351: quantify experimentally. The mean flow is important because it is
352: a nonlocal effect acting over large distances (many roll widths)
353: and changes important general predictions of the phase
354: equation~\cite{cross:1984}. The mean flow is driven by roll
355: curvature, roll compression and gradients in the convection
356: amplitude. The resulting mean flow advects the pattern, giving an
357: additional slow time dependence.
358:
359: The mean flow present in the simulation flow fields,
360: $\vec{U}_s(x,y)$, is formed by calculating the depth averaged
361: horizontal velocity,
362: \begin{equation}
363: \vec{U}_s(x,y)=\int^1_0 \vec{u}_{\perp}(x,y,z)dz
364: \label{eq:mean_flow_sim}
365: \end{equation}
366: where $\vec{u}_{\perp}$ is the horizontal velocity field.
367: Furthermore it will be convenient to work with the vorticity
368: potential, $\zeta$, defined as
369: \begin{equation} \label{eq:vorticity_potential}
370: \nabla^2_\perp \zeta = -\hat{z} \dotprod \left(
371: \vec{\nabla}_{\perp} \times \vec{U}_s \right) = - \omega_z
372: \end{equation}
373: where $\omega_z$ is the vertical vorticity and $\nabla^2_\perp$ is
374: the horizontal Laplacian.
375:
376: Six consecutive snapshots in time for the periodic dynamics shown
377: in Fig.~\ref{fig:nu_all_cond} case ii) are illustrated in
378: Fig.~\ref{fig:snapshots_mf}. One half period is displayed
379: illustrating the nucleation of a dislocation pair and its
380: subsequent annihilation in the opposing wall foci. The vorticity
381: potential, $\zeta$, is shown on a grey scale: dark regions
382: represent negative vorticity and light regions represent positive
383: vorticity which will generate a clockwise and a counter clockwise
384: rotating mean flow, respectively. The quadrupole spatial structure
385: of $\zeta$ in the first panel, i.e. four lobes of alternating
386: positive and negative vorticity with one lobe per quadrant,
387: generates a roll compressing mean flow that pushes the system
388: closer to a dislocation pair nucleation event. During dislocation
389: climb and glide the spatial structure of the vorticity potential
390: is more complicated until the pan-am pattern is reestablished in
391: final panel and a quadrupole structure of vorticity is again
392: formed and the process repeats. The dislocations alternate gliding
393: left and right resulting is a slight rocking back and forth of the
394: entire pattern with each half period which is visible in the
395: different pattern orientations in the first last panels. This
396: alternation persists for the entire simulation.
397:
398: A numerical investigation of the importance of the mean flow for
399: this small cylindrical domain was performed by implementing the
400: ramped and finned boundary conditions. In all of these simulations
401: the bulk region of constant $R$ extended out to a radius
402: $r_0=4.72$. In the finned case a fin at half height occupied the
403: region $4.72 \le r \le 7.66$. In the ramped case a radial ramp in
404: plate separation was given by,
405: \begin{equation}
406: h(r) = \left\{ \begin{array}{ll}
407: 1, & \mbox{$r < r_0$} \\
408: 1 - {\delta_r} \left[ 1- \cos \left( \frac{r-r_0} {r_1-r_0} \pi \right) \right], & \mbox{$r \ge
409: r_0$}
410: \end{array}\right.
411: \label{eq:platesep}
412: \end{equation}
413: where $r_0=4.72$, $r_1=10.0$, and $\delta=0.15$.
414:
415: The different mean wavenumber behavior (using the Fourier methods
416: discussed in \cite{morris:1993}) exhibited in these three
417: different cases is shown in Fig.~\ref{fig:wnall_smallgamma}. As
418: illustrated in Fig.~\ref{fig:rigid_fin_ramp} the behavior of the
419: vorticity potential suggests an explanation. In the simulations
420: with a rigid sidewall, not ramped or finned, the vorticity
421: potential generates a mean flow that enhances roll compression, as
422: described above. In the case of the finned and ramped boundaries
423: the vorticity potential and the resulting mean flow are being
424: generated by gradients in the convection amplitude and are largely
425: situated away from the bulk of the domain. Furthermore, the mean
426: flow generated is strongest in the subcritical finned or ramped
427: region away from the convection rolls. This is demonstrated by
428: comparing the average value of the mean flow over a fraction of
429: the bulk of the domain, $r \le 1$, where it was found that
430: $\bar{U}_s = $ 0.23, 0.09, and 0.02 for the rigid, finned and
431: ramped domains, respectively, and that the maximum flow field
432: velocity is $|\vec{u}| \approx 10$.
433: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
434: \begin{figure}[tbh]
435: \begin{center}
436: \includegraphics[width=3.0in]{./fig5.eps}
437: \end{center}
438: \caption{Mean pattern wavenumber measurements for a cylindrical
439: convection layer, $r_0=4.72$, with rigid sidewalls ($\Box$), fin
440: ($\circ, r_1=7.66$) and a spatial ramp in plate separation
441: ($\diamond, r_1=10.0$, $r_c=7.34$, and $\delta=0.15$). In all
442: three cases the sidewalls are perfectly conducting and
443: $\sigma=0.78$. For reference, solid lines labelled E, N, and SV
444: indicate the approximate location of the Eckhaus, Neutral and
445: Skewed Varicose stability boundaries for an infinite layer
446: straight parallel convection rolls. All patterns represented are
447: time independent.} \label{fig:wnall_smallgamma}
448: \end{figure}
449: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
450: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
451: \begin{figure}[tbh]
452: \begin{center}
453: \includegraphics[width=2.0in]{./fig6.eps}
454: \end{center}
455: \caption{Convection pattern and shaded contours of the vorticity
456: potential, $\zeta$, for a cylindrical convection layer,
457: $r_0=4.72$, with rigid sidewalls, fin ($r_1=7.66$) and a spatial
458: ramp in plate separation ($r_1=10.0$, $r_c=7.34$ and marked with a
459: dashed line, and $\delta=0.15$) shown top, middle and bottom,
460: respectively. The convection pattern is illustrated by plotting
461: zero contours of the thermal perturbation. In all three cases the
462: sidewalls are perfectly conducting, $\sigma=0.78$ and $R=2804$.}
463: \label{fig:rigid_fin_ramp}
464: \end{figure}
465: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
466:
467: It is attractive to pursue the case of a radial ramp in plate
468: separation because the variation in the convective amplitude
469: caused by the ramp can be determined analytically and the
470: influence of a mean flow upon nearly straight rolls can be
471: quantified~\cite{paul:2002:pre}. Usually the mean flow can only be
472: determined once the texture is known and it is hard to calculate
473: because of defects acting as sources, in addition to the regions
474: of smooth distortions.
475:
476: Near threshold an explicit expression for the mean flow,
477: $\vec{U}$, that advects the convection pattern
478: is~\cite{cross:1984}
479: \begin{equation}
480: \vec{U}(x,y) = - \gamma \vec{k} \vec{\nabla}_{\perp} \dotprod
481: \left( \vec{k} |A|^2 \right) - \vec{\nabla}_{\perp}p_o(x,y)
482: \label{eq:meanflow}
483: \end{equation}
484: where $\gamma$ is a coupling constant given by $\gamma = 0.42
485: {\sigma}^{-1} (\sigma+0.34)(\sigma+0.51)^{-1}$, $|A|^2$ is the
486: convection amplitude normalized so that the convective heat flow
487: per unit area relative to the conducted heat flow at $R_c$ is
488: $|A|^2R/R_c$, $p_o$ is the slowly varying pressure (see
489: Eq.~(\ref{eq:pressure_p0})) and $\vec{\nabla}_{\perp}$ is the
490: horizontal gradient operator. The vertical vorticity is then given
491: by the vertical component of the curl of Eq.~(\ref{eq:meanflow}),
492: \begin{equation}
493: {\omega}_z = \hat{z} \dotprod \left( \vec{\nabla}_{\perp} \times
494: \vec{U} \right) = - \gamma \hat{z} \dotprod \vec{\nabla}_{\perp}
495: \times \left[ \vec{k} \vec{\nabla}_{\perp} \dotprod \left( \vec{k}
496: |A|^2 \right) \right]. \label{eq:wz_general}
497: \end{equation}
498: Consider a cylindrical convection layer with a radial ramp in
499: plate separation containing a field of x-rolls given by
500: $\vec{k}=k_o \hat{x}$. The amplitude can be represented for large
501: $\epsilon_o$, using an adiabatic approximation, as
502: $|A|^2=\epsilon(r)/g_o$ for $\epsilon>0$ and $|A|^2=0$ for
503: $\epsilon(r)<0$ as shown in
504: Fig.~\ref{fig:ramp_vorticity_schematic}, making the amplitude a
505: function of radius only $|A|^2=f(r)$.
506: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
507: \begin{figure}[tbh]
508: \begin{center}
509: \includegraphics[width=3.5in]{./fig7.eps}
510: \end{center}
511: \caption{A schematic illustrating the radial variation, for purely
512: adiabatic conditions, of $\epsilon$ (dashed line), $|A|^2$ (solid
513: line), and $\omega_z$ (solid line with arrow) for a cylindrical
514: convection layer with a radial ramp in plate separation. Labelled
515: $r_0$ and $r_c$ are the radial values where the ramp begins and
516: where the ramp yields critical conditions, respectively. Note that
517: for $r>r_c$, $|A|^2=0$ in the adiabatic approximation.}
518: \label{fig:ramp_vorticity_schematic}
519: \end{figure}
520: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
521: Inserting $|A|^2=f(r)$ into Eq.~(\ref{eq:wz_general}) yields,
522: after some manipulation, the following expression for the vertical
523: vorticity,
524: \begin{equation}
525: {\omega}_z = \frac{\gamma {k_o}^2}{2} \left[ \frac{d^2|A|^2}{dr^2}
526: - \frac{1}{r} \frac{d|A|^2}{dr} \right] \sin 2 \theta.
527: \label{eq:wz}
528: \end{equation}
529: The vorticity generated by the amplitude variation caused by the
530: ramp is also shown in Fig.~\ref{fig:ramp_vorticity_schematic}:
531: there is a negative vorticity for $r_0<r<r_c$ and then a delta
532: function spike of positive vorticity at $r_c$. To correct for
533: nonadiabaticity and to smooth $|A(r)|^2$ near $r_c$, the
534: one-dimensional time independent amplitude
535: equation~\cite{newell:1969} is solved,
536: \begin{equation}
537: 0 = \epsilon (r) A + {{\xi}_o}^2 {\cos}^2 {\theta}
538: \frac{\partial^2A}{\partial r^2} - g_o |A|^2A, \label{eq:amp_dim}
539: \end{equation}
540: where ${{\xi}_o}^2 = 0.148$,
541: $g_o=(0.6995-0.0047\sigma^{-1}+0.0083\sigma^{-2})$ and
542: $\epsilon(r)$ is determined by
543: \begin{equation}
544: \epsilon(r) = \left\{ \begin{array}{ll}
545: \epsilon_o, & \mbox{$r < r_0$} \\
546: \epsilon_o (h^3-h_c^3)/(1-h_c^3), & \mbox{$r \ge
547: r_0$}
548: \end{array}\right.
549: \label{eq:eps}
550: \end{equation}
551: where $h_c=h(r_c)$. Equation~(\ref{eq:amp_dim}) is solved
552: numerically using the boundary conditions $\partial_r A=0$ at $r =
553: 0$, and $A=0$ at $r = r_1$.
554:
555: To compare these analytical results with simulation we have chosen
556: to investigate a large-aspect-ratio cylinder with a gradual radial
557: ramp, defined by Eq.~(\ref{eq:platesep}), given by the parameters:
558: $r_0=11.31$, $r_1=20.0$, $\delta_r=0.036$, and $\sigma=0.87$. For
559: small $\epsilon$ the amplitude $A^2(r)$ is unable to adiabatically
560: follow the ramp, this nonadiabaticity results in a deviation from
561: $\epsilon(r)/g_o$ as shown in Fig.~\ref{fig:amp_vort_mf_exp1750}a.
562: However, as $\epsilon_0$ increases the amplitude $A^2(r)$ follows
563: $\epsilon(r)/g_o$ adiabatically almost over the entire ramp except
564: for a small kink at $r_c$. The structure of $\omega_z$ depends
565: upon this adiabaticity and is shown in
566: Fig.~\ref{fig:amp_vort_mf_exp1750}b where we have used the
567: solution to Eq.~(\ref{eq:amp_dim}) at $\theta=\pi/4$ in
568: Eq.~(\ref{eq:wz}). This is not strictly correct since the
569: non-adiabaticity of the amplitude is $\theta$ dependent which will
570: induce higher angular modes of the vorticity not given by
571: Eq.~(\ref{eq:wz}). However, the calculation should give a good
572: approximation to the main $\sin 2 \theta$ component of the
573: vorticity. It is evident from Fig.~\ref{fig:amp_vort_mf_exp1750}b
574: that the vertical vorticity, calculated from the simulation
575: results as an angular average weighted by $\sin 2 \theta$ has an
576: octupole angular dependence (octupole in the sense of an inner and
577: outer quadrupole) and is well approximated by theory without any
578: adjustable parameters.
579:
580: The mean flow generated by these vorticity distributions is
581: determined by solving Eq.~(\ref{eq:wz_general}) with the boundary
582: condition $\zeta(r_1)=0$. The vorticity potential is related to
583: the mean flow in polar coordinates by $(U_r,U_\theta)=(r^{-1}
584: \partial_\theta \zeta, -\partial_r \zeta)$. The vorticity
585: potential is expanded radially in second order Bessel functions
586: while maintaining the $\sin 2 \theta$ angular dependence. Of
587: particular interest is the mean flow perpendicular to the
588: convection rolls, $U_r(\theta=0)$ or equivalently $U_x(y=0)$,
589: which is shown in Fig.~\ref{fig:amp_vort_mf_exp1750}c. Again the
590: simulation results compare well with theory even in the absence of
591: adjustable parameters.
592: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
593: \begin{figure}[tbh]
594: \begin{center}
595: \includegraphics[width=2.0in]{./fig8.eps}
596: \end{center}
597: \caption{Panel~(a) shows the solution of Eq.~(\ref{eq:amp_dim})
598: plotted as $A^2(r)$, shown for comparison is $\epsilon(r)/g_o$.
599: Panel~(b) compares the vertical vorticity found analytically from
600: Eq.~(\ref{eq:amp_dim}) with an angular average, weighted by $\sin
601: 2 \theta$, of the vertical vorticity from simulation. Panel~(c)
602: compares the mean flow found analytically from
603: Eq.~(\ref{eq:vorticity_potential}) with the mean flow from
604: simulation. Parameters are $r_0=11.31$, $r_c=13.20$, $r_1=20.0$,
605: $\delta_r=0.036$, $\sigma=0.87$ and $\epsilon_o = 0.025$.}
606: \label{fig:amp_vort_mf_exp1750}
607: \end{figure}
608: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
609:
610: To make the connection between mean flow and wavenumber
611: quantitative it is noted that the wavenumber variation resulting
612: from a mean flow across a field of x-rolls can be determined from
613: the one-dimensional phase equation,
614: \begin{equation} \label{eq:phase}
615: U \partial_x \phi = D_\parallel \partial_{xx}\phi
616: \end{equation}
617: where the wavenumber is the gradient of the phase,
618: $k=\partial_x\phi$, $D_\parallel=\xi_o^2 \tau_o^{-1}$, and
619: $\tau_o^{-1}=19.65 \sigma (\sigma+0.5117)^{-1}$~\cite{cross:1993}.
620: Figure~\ref{fig:wn_mf_compare2000}a illustrates the wavenumber
621: variation for a large-aspect-ratio simulation, $k(r)$ for $r \le
622: r_0$, and makes evident the roll compression, $k(r=0)>k(r_0)$.
623: Figure~\ref{fig:wn_mf_compare2000}b compares the mean flow
624: calculated from simulation to the predicted value of the mean flow
625: required to produce the wavenumber variation shown in
626: Fig.~\ref{fig:wn_mf_compare2000}a. The agreement is good and the
627: discrepancy near $r_0$, which is contained within one roll
628: wavelength from where the ramp begins, is expected because the
629: influence of the ramp was not included in Eq.~(\ref{eq:phase}).
630: This illustrates quantitatively that is in indeed the mean flow
631: that compresses the rolls in the bulk of the domain.
632: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
633: \begin{figure}[tbh]
634: \begin{center}
635: \includegraphics[width=2.5in]{./fig9.eps}
636: \end{center}
637: \caption{Panel~(a), the variation in the local wavenumber along
638: the positive x-axis, or equivalently $k(r)$ at $\theta=0$.
639: Panel~(b), a comparison of the mean flow from simulation (solid
640: line) with the predicted value (dashed line) calculated from
641: Eq.~(\ref{eq:phase}) using the wavenumber variation from
642: panel~(a). Simulation parameters, $r_0=11.31$, $r_1=20$,
643: $\delta_r=0.036$, $\sigma=0.87$ and $\epsilon=0.171$.}
644: \label{fig:wn_mf_compare2000}
645: \end{figure}
646: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
647:
648: Finally, to better understand the connection between mean flow and
649: pattern dynamics, especially that of spatiotemporal chaotic states
650: exhibiting both temporal chaos as well as spatial disorder, we
651: apply a novel numerical procedure to eliminate mean flow from the
652: fluid equation, Eq.~(\ref{eq:mom}), thereby evolving the dynamics
653: of an artificial fluid with no explicit contributions from mean
654: flow. In this way, we can then obtain quantitative comparisons
655: between the patterns generated by this artificial fluid with mean
656: flow quenched and by the original fluid equation.
657:
658: We have applied this procedure to study spiral defect chaos (see
659: bottom of Fig.~\ref{fig:large_gamma}) \cite{morris:1993}. Numerous
660: attempts have been made to understand how a spiral defect chaos
661: state is formed and how it is sustained. For example, experiments
662: \cite{assenheimer:1993:prl,assenheimer:1994} have found that
663: spirals transition to targets when the Prandtl number is
664: increased. Owing to the fact that the magnitude of mean flow is
665: inversely proportional to the Prandtl number, c.f.
666: Eq.~(\ref{eq:meanflow}), it was believed that spiral defect chaos
667: is a low Prandtl number phenomenon for which mean flow is
668: essential to their dynamics. This is supported by studies of
669: convection models based on the generalized Swift-Hohenberg
670: equation \cite{xi:1993:prl,xi:1993:pre,xi:1995}, where spiral
671: defect chaos is not observed unless a term corresponding to mean
672: flow is explicitly coupled to the equation. However, these
673: observations are by themselves insufficient. For example, there
674: are many other effects in the fluid equations that grow towards
675: low Prandtl numbers, and there could be limitations in the
676: Swift-Hohenberg modelling. We have applied our numerical
677: procedure to this case to explicitly confirm the role of mean flow
678: in the dynamics of spiral defect chaos.
679: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
680: \begin{figure}[tbh]
681: \begin{center}
682: \includegraphics[width=1.5in]{./fig10_left.eps}
683: \includegraphics[width=1.5in]{./fig10_right.eps}
684: \end{center}
685: \caption{Spiral defect chaos (left) and angular textures (right)
686: obtained by quenching mean flow. The left panel is at
687: $t=152\tau_v$ and displays the pattern upon which the mean flow is
688: quenched, the right panel is at $t=320\tau_v$. In both cases,
689: $R=2950, \sigma=1.0$ and the lateral sidewalls are insulating. We
690: see that the spiral arms transition to angular textures when mean
691: flow is quenched. Also, the quenched state is stationary.}
692: \label{fig:quench}
693: \end{figure}
694: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
695:
696: Recalling that we can approximate mean flow to be the
697: depth-averaged horizontal velocity, c.f.
698: Eq.~(\ref{eq:mean_flow_sim}), we can first depth-average the
699: horizontal components of the fluid equation, Eq.~(\ref{eq:mom}),
700: to obtain a dynamical equation for the mean flow $\vec{U}_s$:
701: \begin{eqnarray}
702: \lefteqn{\sigma^{-1} \partial_t \vec{U}_s + \sigma^{-1} \int_0^1 dz
703: (\vec{u}\dotprod\vec{\nabla}) \vec{u}_\perp =} \nonumber \\
704: & & -\vec{\nabla}_\perp \int_0^1 dz p +
705: \nabla_\perp^2 \vec{U}_s + \int_0^1 dz \partial_{zz} \vec{u}_\perp.
706: \end{eqnarray}
707: In this equation, the term $-\nabla_\perp \int_0^1 dz p$ can be
708: absorbed into the nonlinear Reynolds stress term via
709: Eq.~(\ref{eq:pressure_p0}) and so will be ignored henceforth. The
710: resulting equation is then a diffusion equation in $\vec{U}_s$
711: with a source term $\vec{F}_s \equiv \int_0^1 dz
712: (\vec{u}\dotprod\nabla) \vec{u}_\perp -\sigma \int_0^1 dz
713: \partial_{zz} \vec{u}_\perp$. If this source term were not
714: present, then $\vec{U}_s$, being the solution to a diffusion
715: equation, evolves to zero with an effective diffusivity $\sigma$,
716: the Prandtl number. Thus, the role of $\vec{F}_s$ is to act as a
717: generating source for the mean flow $\vec{U}_s$. Subtracting it
718: from the fluid equation, Eq.~(\ref{eq:mom}), then results in the
719: mean flow being eliminated.
720:
721: In practice, we found that it is necessary to actually subtract
722: $\vec{F}_s$ multiplied by a constant to ensure that the magnitude
723: of mean flow becomes zero. This can be understood in terms of the
724: necessity to correct for the fact that
725: Eq.~(\ref{eq:mean_flow_sim}) is only an approximation to the flow
726: field that advects the rolls given by
727: \begin{equation}
728: \vec{U} = \int_0^1 dz g(z) \vec{u}_\perp
729: \end{equation}
730: where $g(z)$ is a weighting function depending on the full
731: nonlinear structure of the rolls. This is discussed further
732: elsewhere \cite{chiam:2002}.
733:
734: We have carried out this procedure by introducing the term
735: $\vec{F}_s$ to the right-hand-side of the fluid equation after a
736: spiral defect chaotic state becomes fully developed, typically
737: after about one horizontal diffusion time starting from random
738: thermal perturbations as initial condition. We see that the
739: spirals immediately, on the order of a vertical diffusion time,
740: ``straighten out'' to form angular chevron-like textures; see
741: Fig.~\ref{fig:quench}. Unlike spiral defect chaos, these angular
742: textures are stationary (with the exception of the slow motion of
743: defects such as the gliding of dislocation pairs). Thus, we have
744: shown that when mean flow is quenched via the subtraction of the
745: term $\vec{F}_s$ from the fluid equation, spiral defect chaos
746: ceases to exist.
747:
748: We have further quantified the differences between spiral defect
749: chaos and the angular textures. We mention here briefly one of
750: the results: by comparing the wavenumber distribution for both
751: sets of states, we have observed that the mean wavenumber
752: approaches the unique wavenumber possessed by axisymmetric
753: patterns asymptotically far away from the center
754: \cite{buell:1986:axi}. (The axisymmetric pattern, by symmetry,
755: does not have mean flow components.) We discuss this as well as
756: other results in a separate article \cite{chiam:2002}.
757: \section{Conclusion}
758: \label{section:conclusion} Full numerical simulations of
759: Rayleigh-B\'{e}nard convection in cylindrical and rectangular
760: shaped domains for a range of aspect ratios, $5 \lesssim \Gamma
761: \lesssim 60$, with experimentally realistic boundary conditions,
762: including rigid, finned and spatially ramped sidewalls, have been
763: performed. These simulations provide us with a complete knowledge
764: of the flow field allowing us to quantitatively address some
765: interesting open questions.
766:
767: In this paper we have emphasized the exploration of the mean flow.
768: The mean flow is important in a theoretical understanding of the
769: pattern dynamics, yet is very difficult to measure in experiment,
770: making numerical simulations attractive to close this gap.
771:
772: The mean flow is found to be important in small cylindrical
773: domains by investigating the result of imposing different sidewall
774: boundary conditions. Analytical results are developed for a
775: large-aspect-ratio cylinder with a radial ramp in plate
776: separation. Numerical results of the vertical vorticity and the
777: mean flow agree with these predictions. Furthermore, the
778: wavenumber behavior predicted using the mean flow in a
779: one-dimensional phase equation also agrees with the results of
780: simulation. This allows extrapolation of the analysis to larger
781: aspect ratios.
782:
783: Lastly we utilize the control and flexibility offered by numerical
784: simulation to investigate a novel method of quenching numerically
785: the mean flow. We apply this to a spiral defect chaos state and
786: find that the time dependent pattern becomes time independent,
787: angular in nature, and that the pattern wavenumber becomes larger.
788:
789: These quantitative comparisons illustrate the benefit of
790: performing numerical simulations for realistic geometries and
791: boundary conditions as a means to create quantitative links
792: between experiment and theory.
793:
794: We are grateful to G. Ahlers for helpful discussions. This
795: research was supported by the U.S.~Department of Energy, Grant
796: DE-FT02-98ER14892, and the Mathematical, Information, and
797: Computational Sciences Division subprogram of the Office of
798: Advanced Scientific Computing Research, U.S.~Department of Energy,
799: under Contract W-31-109-Eng-38. We also acknowledge the Caltech
800: Center for Advanced Computing Research and the North Carolina
801: Supercomputing Center.
802:
803: \begin{thebibliography}{29}
804: \expandafter\ifx\csname
805: natexlab\endcsname\relax\def\natexlab#1{#1}\fi
806: \expandafter\ifx\csname bibnamefont\endcsname\relax
807: \def\bibnamefont#1{#1}\fi
808: \expandafter\ifx\csname bibfnamefont\endcsname\relax
809: \def\bibfnamefont#1{#1}\fi
810: \expandafter\ifx\csname citenamefont\endcsname\relax
811: \def\citenamefont#1{#1}\fi
812: \expandafter\ifx\csname url\endcsname\relax
813: \def\url#1{\texttt{#1}}\fi
814: \expandafter\ifx\csname
815: urlprefix\endcsname\relax\def\urlprefix{URL }\fi
816: \providecommand{\bibinfo}[2]{#2}
817: \providecommand{\eprint}[2][]{\url{#2}}
818:
819: \bibitem[{\citenamefont{Cross and Hohenberg}(1993)}]{cross:1993}
820: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}} \bibnamefont{and}
821: \bibinfo{author}{\bibfnamefont{P.~C.} \bibnamefont{Hohenberg}},
822: \bibinfo{journal}{Rev. of Mod. Phys.}
823: \textbf{\bibinfo{volume}{65}}(\bibinfo{number}{3 II}), \bibinfo{pages}{851}
824: (\bibinfo{year}{1993}).
825:
826: \bibitem[{\citenamefont{Fischer}(1997)}]{fischer:1997}
827: \bibinfo{author}{\bibfnamefont{P.~F.} \bibnamefont{Fischer}},
828: \bibinfo{journal}{J. Comp. Phys.} \textbf{\bibinfo{volume}{133}},
829: \bibinfo{pages}{84} (\bibinfo{year}{1997}).
830:
831: \bibitem[{\citenamefont{Daviaud and Pocheau}(1989)}]{daviaud:1989}
832: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Daviaud}} \bibnamefont{and}
833: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Pocheau}},
834: \bibinfo{journal}{Europhysics Letters}
835: \textbf{\bibinfo{volume}{9}}(\bibinfo{number}{7}), \bibinfo{pages}{675}
836: (\bibinfo{year}{1989}).
837:
838: \bibitem[{\citenamefont{de~Bruyn et~al.}(1996)\citenamefont{de~Bruyn,
839: Bodenschatz, Morris, Cannell, and Ahlers}}]{debruyn:1996}
840: \bibinfo{author}{\bibfnamefont{J.~R.} \bibnamefont{de~Bruyn}},
841: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Bodenschatz}},
842: \bibinfo{author}{\bibfnamefont{S.~W.} \bibnamefont{Morris}},
843: \bibinfo{author}{\bibfnamefont{D.~S.} \bibnamefont{Cannell}},
844: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}},
845: \bibinfo{journal}{Rev. Sci. Instrum.}
846: \textbf{\bibinfo{volume}{67}}(\bibinfo{number}{6}), \bibinfo{pages}{2043}
847: (\bibinfo{year}{1996}).
848:
849: \bibitem[{\citenamefont{Pocheau and Daviaud}(1997)}]{pocheau:1997}
850: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Pocheau}} \bibnamefont{and}
851: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Daviaud}},
852: \bibinfo{journal}{Phys. Rev. E}
853: \textbf{\bibinfo{volume}{55}}(\bibinfo{number}{1}), \bibinfo{pages}{353}
854: (\bibinfo{year}{1997}).
855:
856: \bibitem[{\citenamefont{Kramer et~al.}(1982)\citenamefont{Kramer, Ben-Jacob,
857: Brand, and Cross}}]{kramer:1982}
858: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Kramer}},
859: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Ben-Jacob}},
860: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Brand}}, \bibnamefont{and}
861: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}},
862: \bibinfo{journal}{Phys. Rev. Lett.}
863: \textbf{\bibinfo{volume}{49}}(\bibinfo{number}{26}), \bibinfo{pages}{1891}
864: (\bibinfo{year}{1982}).
865:
866: \bibitem[{\citenamefont{Bajaj et~al.}(1999)\citenamefont{Bajaj, Mukolobwiez,
867: Currier, and Ahlers}}]{bajaj:1999}
868: \bibinfo{author}{\bibfnamefont{K.~M.~S.} \bibnamefont{Bajaj}},
869: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Mukolobwiez}},
870: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Currier}}, \bibnamefont{and}
871: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}},
872: \bibinfo{journal}{Phys. Rev. Lett.}
873: \textbf{\bibinfo{volume}{83}}(\bibinfo{number}{25}), \bibinfo{pages}{5282}
874: (\bibinfo{year}{1999}).
875:
876: \bibitem[{\citenamefont{Ahlers et~al.}(2001)\citenamefont{Ahlers, Bajaj,
877: Mukolobwicz, and Oh}}]{ahlers:2001}
878: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}},
879: \bibinfo{author}{\bibfnamefont{K.~M.~S.} \bibnamefont{Bajaj}},
880: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Mukolobwicz}},
881: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Oh}},
882: \emph{\bibinfo{title}{unpublished}} (\bibinfo{year}{2001}).
883:
884: \bibitem[{\citenamefont{Ahlers}(1974)}]{ahlers:1974}
885: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}},
886: \bibinfo{journal}{Phys. Rev. Lett.}
887: \textbf{\bibinfo{volume}{33}}(\bibinfo{number}{20}), \bibinfo{pages}{1185}
888: (\bibinfo{year}{1974}).
889:
890: \bibitem[{\citenamefont{Cross and Tu}(1995)}]{cross:1995:physrevlett}
891: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}} \bibnamefont{and}
892: \bibinfo{author}{\bibfnamefont{Y.}~\bibnamefont{Tu}}, \bibinfo{journal}{Phys.
893: Rev. Lett.} \textbf{\bibinfo{volume}{75}}(\bibinfo{number}{5}),
894: \bibinfo{pages}{834} (\bibinfo{year}{1995}).
895:
896: \bibitem[{\citenamefont{Morris et~al.}(1993)\citenamefont{Morris, Bodenschatz,
897: Cannell, and Ahlers}}]{morris:1993}
898: \bibinfo{author}{\bibfnamefont{S.~W.} \bibnamefont{Morris}},
899: \bibinfo{author}{\bibfnamefont{E.}~\bibnamefont{Bodenschatz}},
900: \bibinfo{author}{\bibfnamefont{D.~S.} \bibnamefont{Cannell}},
901: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}},
902: \bibinfo{journal}{Phys. Rev. Lett.}
903: \textbf{\bibinfo{volume}{71}}(\bibinfo{number}{13}), \bibinfo{pages}{2026}
904: (\bibinfo{year}{1993}).
905:
906: \bibitem[{\citenamefont{Ahlers and Behringer}(1978)}]{ahlers:1978}
907: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}} \bibnamefont{and}
908: \bibinfo{author}{\bibfnamefont{R.~P.} \bibnamefont{Behringer}},
909: \bibinfo{journal}{Phys. Rev. Lett.} \textbf{\bibinfo{volume}{40}},
910: \bibinfo{pages}{712} (\bibinfo{year}{1978}).
911:
912: \bibitem[{\citenamefont{Ahlers and Walden}(1980)}]{ahlers:1980}
913: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Ahlers}} \bibnamefont{and}
914: \bibinfo{author}{\bibfnamefont{R.~W.} \bibnamefont{Walden}},
915: \bibinfo{journal}{Phys. Rev. Lett.}
916: \textbf{\bibinfo{volume}{44}}(\bibinfo{number}{7}), \bibinfo{pages}{445}
917: (\bibinfo{year}{1980}).
918:
919: \bibitem[{\citenamefont{Libchaber and Maurer}(1978)}]{libchaber:1978}
920: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Libchaber}} \bibnamefont{and}
921: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Maurer}},
922: \bibinfo{journal}{J. Physique Lett.} \textbf{\bibinfo{volume}{39}},
923: \bibinfo{pages}{369} (\bibinfo{year}{1978}).
924:
925: \bibitem[{\citenamefont{Pocheau et~al.}(1985)\citenamefont{Pocheau, Croquette,
926: and Le~Gal}}]{pocheau:1985}
927: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Pocheau}},
928: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Croquette}},
929: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Le~Gal}},
930: \bibinfo{journal}{Phys. Rev. Lett.}
931: \textbf{\bibinfo{volume}{55}}(\bibinfo{number}{10}), \bibinfo{pages}{1094}
932: (\bibinfo{year}{1985}).
933:
934: \bibitem[{\citenamefont{Croquette et~al.}(1986)\citenamefont{Croquette, Le~Gal,
935: and Pocheau}}]{croquette:1986}
936: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Croquette}},
937: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Le~Gal}}, \bibnamefont{and}
938: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Pocheau}},
939: \bibinfo{journal}{Phys. Scr.} \textbf{\bibinfo{volume}{T13}},
940: \bibinfo{pages}{135} (\bibinfo{year}{1986}).
941:
942: \bibitem[{\citenamefont{Pocheau}(1989)}]{pocheau:1989}
943: \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Pocheau}}, \bibinfo{journal}{J.
944: Phys. France} \textbf{\bibinfo{volume}{50}}, \bibinfo{pages}{2059}
945: (\bibinfo{year}{1989}).
946:
947: \bibitem[{\citenamefont{Paul et~al.}(2001)\citenamefont{Paul, Cross, Fischer,
948: and Greenside}}]{paul:2001}
949: \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{Paul}},
950: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}},
951: \bibinfo{author}{\bibfnamefont{P.~F.} \bibnamefont{Fischer}},
952: \bibnamefont{and} \bibinfo{author}{\bibfnamefont{H.~S.}
953: \bibnamefont{Greenside}}, \bibinfo{journal}{Phys. Rev. Lett.}
954: \textbf{\bibinfo{volume}{87}}(\bibinfo{number}{15}), \bibinfo{pages}{154501}
955: (\bibinfo{year}{2001}).
956:
957: \bibitem[{\citenamefont{Newell et~al.}(1990)\citenamefont{Newell, Passot, and
958: Souli}}]{newell:1990:jfm}
959: \bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Newell}},
960: \bibinfo{author}{\bibfnamefont{T.}~\bibnamefont{Passot}}, \bibnamefont{and}
961: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Souli}}, \bibinfo{journal}{J.
962: Fluid Mech.} \textbf{\bibinfo{volume}{220}}, \bibinfo{pages}{187}
963: (\bibinfo{year}{1990}).
964:
965: \bibitem[{\citenamefont{Cross and Newell}(1984)}]{cross:1984}
966: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}} \bibnamefont{and}
967: \bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Newell}},
968: \bibinfo{journal}{Physica D} \textbf{\bibinfo{volume}{10}},
969: \bibinfo{pages}{299} (\bibinfo{year}{1984}).
970:
971: \bibitem[{\citenamefont{Paul et~al.}(2002)\citenamefont{Paul, Cross, and
972: Fischer}}]{paul:2002:pre}
973: \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{Paul}},
974: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}}, \bibnamefont{and}
975: \bibinfo{author}{\bibfnamefont{P.~F.} \bibnamefont{Fischer}},
976: \bibinfo{journal}{Phys. Rev. E} \textbf{\bibinfo{volume}{66}},
977: \bibinfo{pages}{046210} (\bibinfo{year}{2002}).
978:
979: \bibitem[{\citenamefont{Newell and Whitehead}(1969)}]{newell:1969}
980: \bibinfo{author}{\bibfnamefont{A.~C.} \bibnamefont{Newell}} \bibnamefont{and}
981: \bibinfo{author}{\bibfnamefont{J.~A.} \bibnamefont{Whitehead}},
982: \bibinfo{journal}{J. Fluid Mech.}
983: \textbf{\bibinfo{volume}{38}}(\bibinfo{number}{2}), \bibinfo{pages}{279}
984: (\bibinfo{year}{1969}).
985:
986: \bibitem[{\citenamefont{Assenheimer and
987: Steinberg}(1993)}]{assenheimer:1993:prl}
988: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Assenheimer}} \bibnamefont{and}
989: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Steinberg}},
990: \bibinfo{journal}{Phys. Rev. Lett.}
991: \textbf{\bibinfo{volume}{70}}(\bibinfo{number}{25}), \bibinfo{pages}{3888}
992: (\bibinfo{year}{1993}).
993:
994: \bibitem[{\citenamefont{Assenheimer and Steinberg}(1994)}]{assenheimer:1994}
995: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Assenheimer}} \bibnamefont{and}
996: \bibinfo{author}{\bibfnamefont{V.}~\bibnamefont{Steinberg}},
997: \bibinfo{journal}{Nature} \textbf{\bibinfo{volume}{367}}
998: (\bibinfo{year}{1994}).
999:
1000: \bibitem[{\citenamefont{Xi et~al.}(1993{\natexlab{a}})\citenamefont{Xi, Gunton,
1001: and Vi\~{n}als}}]{xi:1993:prl}
1002: \bibinfo{author}{\bibfnamefont{H.-w.} \bibnamefont{Xi}},
1003: \bibinfo{author}{\bibfnamefont{J.~D.} \bibnamefont{Gunton}},
1004: \bibnamefont{and}
1005: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Vi\~{n}als}},
1006: \bibinfo{journal}{Phys. Rev. Lett.}
1007: \textbf{\bibinfo{volume}{71}}(\bibinfo{number}{13}), \bibinfo{pages}{2030}
1008: (\bibinfo{year}{1993}{\natexlab{a}}).
1009:
1010: \bibitem[{\citenamefont{Xi et~al.}(1993{\natexlab{b}})\citenamefont{Xi, Gunton,
1011: and Vi\~{n}als}}]{xi:1993:pre}
1012: \bibinfo{author}{\bibfnamefont{H.-w.} \bibnamefont{Xi}},
1013: \bibinfo{author}{\bibfnamefont{J.~D.} \bibnamefont{Gunton}},
1014: \bibnamefont{and}
1015: \bibinfo{author}{\bibfnamefont{J.}~\bibnamefont{Vi\~{n}als}},
1016: \bibinfo{journal}{Phys. Rev. E}
1017: \textbf{\bibinfo{volume}{47}}(\bibinfo{number}{5}), \bibinfo{pages}{2987}
1018: (\bibinfo{year}{1993}{\natexlab{b}}).
1019:
1020: \bibitem[{\citenamefont{Xi and Gunton}(1995)}]{xi:1995}
1021: \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Xi}} \bibnamefont{and}
1022: \bibinfo{author}{\bibfnamefont{J.~D.} \bibnamefont{Gunton}},
1023: \bibinfo{journal}{Phys. Rev. E}
1024: \textbf{\bibinfo{volume}{52}}(\bibinfo{number}{5}), \bibinfo{pages}{4963}
1025: (\bibinfo{year}{1995}).
1026:
1027: \bibitem[{\citenamefont{Chiam et~al.}(2002)\citenamefont{Chiam, Paul, Cross,
1028: and Greenside}}]{chiam:2002}
1029: \bibinfo{author}{\bibfnamefont{K.-H.} \bibnamefont{Chiam}},
1030: \bibinfo{author}{\bibfnamefont{M.~R.} \bibnamefont{Paul}},
1031: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Cross}}, \bibnamefont{and}
1032: \bibinfo{author}{\bibfnamefont{H.~S.} \bibnamefont{Greenside}},
1033: \emph{\bibinfo{title}{unpublished}} (\bibinfo{year}{2002}).
1034:
1035: \bibitem[{\citenamefont{Buell and Catton}(1986)}]{buell:1986:axi}
1036: \bibinfo{author}{\bibfnamefont{J.~C.} \bibnamefont{Buell}} \bibnamefont{and}
1037: \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Catton}},
1038: \bibinfo{journal}{Phys. Fluids} \textbf{\bibinfo{volume}{29}},
1039: \bibinfo{pages}{23} (\bibinfo{year}{1986}).
1040:
1041: \end{thebibliography}
1042:
1043: \end{document}
1044: