1: %%
2: %% rc.tex -- LaTeX2e
3: %% Boundary effects and the onset of Taylor vortices
4: %% Alastair Rucklidge and Alan Champneys
5: %%
6: %% started 30/7/2002
7: %% lastest aditions ARC 16/05/2003
8: %% some revision/addition AMR 30/05/2003
9: %% further additions AMR 6/8/2003, including rcm -> rc
10: %% some minor revisions ARC 13/8/2003
11: %% references and tidying up AMR 18/8/03
12: %% submitted to Physica D (editor: Fauve) 22/8/03
13:
14: % cover letter to the editor:
15: % Dear Professor Fauve,
16: %
17: % Please accept this manuscript, 'Boundary effects and the onset of Taylor
18: % vortices' by myself and Alan Champneys, for consideration for publication in
19: % Physica D.
20: %
21: % In the paper, we consider a discrepancy that lies at the heart of the
22: % Taylor-Couette problem: the onset of vortices occurs sharply at the critical
23: % value of Reynold's number predicted by linear stability theory (using periodic
24: % boundary conditions along the axis), and yet the boundary effects from the top
25: % and bottom boundaries are strong: anomalous modes occur only for Reynold's
26: % numbers considerably greater than the critical value. We show how to resolve
27: % this apparent contradiction in the context of a Swift-Hohenberg model: the
28: % onset of vortices occurs via a wall mode whose exponential tail penetrates
29: % further into the bulk of the domain as the driving parameter increases. In a
30: % large domain, the amplitude as measured in the centre is only significant once
31: % the Reynold's number is close to the ideal critical value. The disconnected
32: % anomalous branch does not arise as a bifurcation from the wall mode and remains
33: % order one away the onset parameter value.
34: %
35: % With best wishes,
36: %
37: % Alastair Rucklidge
38:
39: % Upper-case A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
40: % Lower-case a b c d e f g h i j k l m n o p q r s t u v w x y z
41: % Digits 0 1 2 3 4 5 6 7 8 9
42: % Exclamation ! Double quote " Hash (number) #
43: % Dollar $ Percent % Ampersand &
44: % Acute accent ' Left paren ( Right paren )
45: % Asterisk * Plus + Comma ,
46: % Minus - Point . Solidus /
47: % Colon : Semicolon ; Less than <
48: % Equals = Greater than > Question mark ?
49: % At @ Left bracket [ Backslash \
50: % Right bracket ] Circumflex ^ Underscore _
51: % Grave accent ` Left brace { Vertical bar |
52: % Right brace } Tilde ~
53:
54: \documentclass[12pt]{elsart}
55:
56: \renewcommand{\baselinestretch}{1.2}
57:
58: % packages
59:
60: % this is for the mathbb to get Rset below
61: \usepackage{amsfonts}
62: \usepackage{xspace}
63:
64: % this is for dfrac
65: \usepackage{amsmath}
66:
67: % for included graphics
68: \usepackage{psfig,color,graphicx}
69:
70: % locally defined commands
71: \newcommand{\smallpf}{{\mbox{\footnotesize pf}} }
72:
73: % these from the LaTeX companion p.50
74: \renewcommand{\etal}{{\it et al.}\@\xspace}
75:
76: \begin{document}
77:
78: %\bibliographystyle{plain}
79: %\bibliographystyle{elsart-num}
80:
81: \begin{frontmatter}
82: %\centerline{Physica D {\bf ???} (19??) ???--???\hfill}
83: \centerline{Submitted to Physica D, August 2003\hfill}
84:
85: \title{Boundary effects and the onset of Taylor vortices}
86:
87: \author{A.M.~Rucklidge}
88: \address{Department of Applied Mathematics,
89: University of Leeds, Leeds LS2 9JT, UK}
90: \and
91: \author{A.R.~Champneys}
92: \address{Department of Engineering Mathematics, University of Bristol,
93: Bristol BS8 1TR, UK}
94:
95: \begin{abstract}
96: It is well established that the onset of spatially periodic vortex states in
97: the Taylor--Couette flow between rotating cylinders occurs at the value of
98: Reynold's number predicted by local bifurcation theory. However, the symmetry
99: breaking induced by the top and bottom plates means that the true situation
100: should be a disconnected pitchfork. Indeed, experiments have shown that the
101: fold of the disconnected branch can occur at more than double the Reynold's
102: number of onset. This leads to an apparent contradiction: why should Taylor
103: vortices set in so sharply at the value Reynold's number predicted by the
104: symmetric theory, given such large symmetry-breaking effects caused by the
105: boundary conditions? This paper offers a generic explanation. The details are
106: worked out using a Swift--Hohenberg pattern formation model that shares the
107: same qualitative features as the Taylor--Couette flow. Onset occurs via a wall
108: mode whose exponential tail penetrates further into the bulk of the domain as
109: the driving parameter increases. In a large domain of length~$L$, we show that
110: the wall mode creates significant amplitude in the centre at parameter values
111: that are $O(L^{-2})$ away from the value of onset in the problem with ideal
112: boundary conditions. We explain this as being due to a Hamiltonian Hopf
113: bifurcation in space, which occurs at the same parameter value as the pitchfork
114: bifurcation of the temporal dynamics. The disconnected anomalous branch remains
115: $O(1)$ away from the onset parameter since it does not arise as a bifurcation
116: from the wall mode.
117: \end{abstract}
118:
119: \begin{keyword}
120: Pattern formation; Boundary effects; Taylor--Couette experiment; Anomalous
121: modes.\newline
122: 47.20.Ky; 47.54.+r.
123: \end{keyword}
124:
125: \end{frontmatter}
126:
127: \section{Introduction}
128:
129: The Taylor--Couette experiment provided one of the first quantitative
130: verifications of the correctness of the Navier--Stokes partial differential
131: equations (PDEs) describing the dynamics of fluid flows. The experiment, in its
132: simplest form, consists of a pair of concentric cylinders with a fluid-filled
133: gap in between; as the inner cylinder is rotated, a shearing flow (the Couette
134: flow) is established between the cylinders, and this becomes unstable to
135: axisymmetric vortices (Taylor vortices) at a critical value of the rotation
136: rate (as measured by a dimensionless Reynold's number~$R$).
137: See~\cite{refD55,Ko:93} for reviews. One notable achievement of Taylor's
138: work~\cite{refT61} in 1923 was the theoretical prediction and the experimental
139: measurement of the critical Reynold's number~$R_c$ for the onset of vortices,
140: with remarkably good agreement between the two. In doing the stability
141: calculation, Taylor assumed that the vortices would be periodic in the
142: direction along the axis, and neglected the effects of the top and the bottom
143: plates of the experiment. With this assumption, the bifurcation leading to
144: Taylor vortices is a pitchfork (figure~\ref{fig:brokenpitchfork}a), and the
145: characteristic sharp transition of this bifurcation, with the strength of the
146: vortices going as the square root of the degree of supercriticality $R-R_c$,
147: has been confirmed experimentally~\cite{refG97}. The symmetry that is broken in
148: this pitchfork is a translation symmetry along the axes of the cylinders.
149:
150: \begin{figure}
151: \mbox{\psfig{file=brokenpitchfork.eps,width=\hsize}}
152: \caption{Sketches of the amplitude of Taylor vortices measured in the centre
153: of the apparatus as a function of Reynold's number~$R$, under various
154: assumptions.
155: (a)~With ideal (reflecting or periodic) boundary conditions, there is a sharp
156: transition to Taylor vortices at a pitchfork bifurcation at $R=R_c$.
157: (b)~With the assumption of a weakly broken pitchfork bifurcation, there is
158: still a relatively sharp onset of Taylor vortices close to $R=R_c$
159: (after~\cite{refB127a});
160: (c)~The experiments of Benjamin and Mullin~\cite{refB128} suggest that the
161: upper half is still a weakly broken pitchfork, whereas the lower half is a
162: strongly broken pitchfork, with anomalous modes only appearing at Reynold's
163: numbers at least twice the critical value. Solid (dashed) lines indicate stable
164: (unstable) solutions.}
165: \label{fig:brokenpitchfork}
166: \end{figure}
167:
168: Subsequent theoretical developments explored the role of the top and bottom
169: plates in the experiment, which spoil the idealisation of spatial periodicity
170: in the direction parallel to the axis of rotation, and which break the
171: translation symmetry assumed in the original theoretical work. Ekman boundary
172: layers cause the fluid near the boundary to spiral preferentially inwards for
173: any non-zero rotation rate, and the Taylor vortices to develop first in the
174: boundary layer, moving smoothly into the bulk of the fluid as $R$
175: approaches~$R_c$ -- this has been observed in experiments~\cite{refC124} and in
176: calculations~\cite{refA71}. In the words of Benjamin~\cite{refB127a}, `no
177: precise critical value of~$R$ exists for the onset of cellular motion'.
178: Benjamin~\cite{refB127a} interpreted the formation of Taylor vortices as a
179: broken pitchfork (figure~\ref{fig:brokenpitchfork}b), with the end plates
180: driving a flow near the boundary for all non-zero~$R$, and this flow exciting a
181: cellular flow that penetrates the central region with increasing Reynold's
182: number -- see figure~\ref{fig:abshagen}.
183:
184: \begin{figure}
185: \begin{center}
186: \mbox{\psfig{file=abshagen_original_fig4_onset1.eps,width=0.45\hsize}}
187: \mbox{\psfig{file=abshagen_original_fig4_onset8.eps,width=0.45\hsize}}
188: \end{center}
189: \caption{Finite-element results reproduced with permission from~\cite{refA70}
190: showing two radial velocity profiles below and above the critical value of~$R$
191: at which the onset of periodic vortices would occur with periodic boundary
192: conditions.
193: (a)~$R/R_c=63.74/68.189=0.935$;
194: (b)~$R/R_c=69.02/68.189=1.013$.}
195: \label{fig:abshagen}
196: \end{figure}
197:
198: The qualitative role of the end boundaries was explored further by
199: Schaeffer~\cite{refS111}, who introduced a homotopy parameter~$\tau$
200: ($0\leq\tau\leq1$), with $\tau=0$ corresponding to ideal end boundary
201: conditions, where a state of pure Couette flow exists for all values of~$R$,
202: $\tau=1$ corresponding to physically realistic end boundary conditions, and
203: intermediate values of~$\tau$ interpolating between these two extremes. The
204: results for $\tau$~close to zero are indeed consistent with a weakly broken
205: pitchfork bifurcation (figure~\ref{fig:brokenpitchfork}b), and are in
206: qualitative agreement with experimental results of
207: Benjamin~\cite{refB127a,refB127b}: as the Reynold's number is increased slowly,
208: vortices grow smoothly, with the most rapid growth occurring for $R$~close to
209: the critical value. This is illustrated in figure~\ref{fig:abshagen}: note how
210: the radial velocity profiles are not zero for $R$ below~$R_c$; instead we see a
211: pair of `wall modes' connecting the non-parallel flow at the walls to Couette
212: flow in the main body of the cylinder. Similarly, the steady solution for $R$
213: greater than~$R_c$ is not a pure periodic solution but has modulation near the
214: two walls in order to satisfy the inhomogeneous boundary conditions.
215:
216: However, by starting the experiment impulsively, it is possible to find another
217: branch of vortices ({\it anomalous modes}) that resemble ordinary vortices, but
218: have the opposite sign -- close to the ends, these anomalous vortices have an
219: outwards radial velocity, opposite to the normal vortices~\cite{refB127b}.
220: Other experiments and calculations indicate that anomalous modes may also have
221: stagnation regions or narrow counter-cells close to the top and bottom
222: boundaries~\cite{refB129,refC123}.
223:
224: However, anomalous modes cannot be found close to the critical Reynold's
225: number, as would be suggested by figure~\ref{fig:brokenpitchfork}b. Instead,
226: $R$~needs to be at least twice its critical value (and the cylinder started
227: impulsively) before anomalous modes can be found~\cite{refB128}. Once they are
228: established, the anomalous modes persist as $R$ is decreased to a lower
229: stability bound, the exact value of which depend on experimental parameters
230: such as the gap width between the inner and outer cylinders, or the aspect
231: ratio~$L$, a dimensionless measure of the length of the column. The lower
232: existence boundary appears always to be at least twice the critical
233: value~$R_c$~\cite{refB128}, and can be much higher if the gap between the
234: cylinders is narrow~\cite{refC122}. Thus the experimental situation is depicted
235: in figure~\ref{fig:brokenpitchfork}c. Interestingly, the lower stability
236: boundary of the anomalous modes seems to be independent of the aspect ratio of
237: the apparatus (for large aspect ratios), and remains at an appreciable multiple
238: of the critical Reynold's number for ordinary modes even as $L$ becomes
239: large~\cite{refC122,refL66}. This emphasises the fact that the large (but
240: finite) aspect ratio limit is very different from the idealisation of
241: periodicity in the axial direction.
242:
243: It is worth emphasising the two surprising and apparently contradictory aspects
244: that have emerged. First, the distance ($d$~in
245: figure~\ref{fig:brokenpitchfork}(c)) between the fold on the disconnected
246: (anomalous) branch is such that the Reynold's number at the fold is at least a
247: factor of two larger than the Reynold's number of the onset of ordinary Taylor
248: vortices. Making the cylinder longer (so that the boundary effects are moved
249: `towards infinity') does not make $d$ tend to zero, and many authors have
250: concluded that the onset of Taylor vortices can in no way be regarded as a
251: weakly broken pitchfork. Second, the onset of vortices is sharp when viewed in
252: terms of measures such as the radial velocity of the midpoint of the apparatus,
253: and it occurs at almost exactly the value of $R$ that is predicted for the
254: problem without end effects -- so the onset of Taylor vortices apparently can
255: be described as a weakly broken pitchfork bifurcation. This discrepancy would
256: not be expected for a generic unfolding of a pitchfork bifurcation. The
257: purpose of the present paper is to explain this apparent contradiction.
258:
259: The rest of the paper is outlined as follows. In section~2, we introduce the
260: Swift--Hohenberg equations as a model for the Taylor--Couette flow. We note
261: that the modification of pattern formation due to the presence of weak forcing
262: at lateral boundaries in Swift--Hohenberg equations has been addressed in the
263: work of Daniels and co-workers~\cite{refD56,DaHoSk:03} with application to
264: Rayleigh--B\'enard convection in mind. In contrast with their work, we include
265: an order one boundary condition that forces the flow strongly. This approach is
266: also distinguished from that of~\cite{refP58}, which focused on a
267: Ginzberg--Landau equation for the envelope of the vortex amplitude.
268:
269: Section~3 then includes an analysis of the linearised version of this model.
270: Despite being purely linear, it is found that mode shapes like those in
271: figure~\ref{fig:abshagen} emerge under static increase of the bifurcation
272: parameter through the value at which the symmetric problem bifurcates. We
273: explain this in terms of the relation between the temporal pitchfork
274: bifurcation and a spatial Hamiltonian Hopf bifurcation.
275:
276: Section~4 goes on to consider a nonlinear bifurcation analysis. It transpires
277: that the unfolded pitchfork resembles figure~\ref{fig:brokenpitchfork}(c).
278: There are many anomalous branches that emerge from primary and secondary
279: symmetry breaking bifurcations in the symmetric problem. In the
280: Swift--Hohenberg example studied in detail, the stable anomalous branch does
281: not always emerge from the primary pitchfork bifurcation when the
282: symmetry-breaking terms tend to zero. We explore this issue in some detail.
283: Finally, section~5 draws conclusions and discusses wider implications of the
284: results.
285:
286: \section{Swift--Hohenberg model}
287:
288: Rather than consider the nonlinear axisymmetric hydrodynamic partial
289: differential equations (PDEs) that describe the flow between two rotating
290: cylinders, we focus on the simpler Swift--Hohenberg~\cite{refS109} PDE, which
291: shares many of the same pattern-forming features. In fact,
292: Melbourne~\cite{refM99} has demonstrated that bifurcation problems of the
293: Taylor--Couette type (steady state bifurcations with nonzero critical
294: wave\-number in systems with Euclidean symmetry) reduce to equations of
295: Swift--Hohenberg form (though with more general nonlinear terms). The model
296: equation we use is:
297: \begin{equation}
298: U_t=\mu U - (1+2U_{xx}+U_{xxxx}) - U^3 - U\,U_x,
299: \label{eq:SH}
300: \end{equation}
301: where the subscripts denote partial derivatives with respect to space~$x$ and
302: time~$t$. The dependent variable $U(x,t)\in\Rset$ is defined on
303: $x\in[-L/2,L/2]$, where $L$ represents the length of the Taylor column. The
304: parameter~$\mu$ represents the forcing~$R-R_c$. The usual form of the
305: Swift--Hohenberg equation has only a cubic nonlinearity, but we include a
306: quadratic term to ensure there is no $U\rightarrow-U$ symmetry
307: (see, e.g., \cite[eq.~(7.21)]{Ko:93}. This term also
308: makes the model equation non-variational and so allows unsteady behaviour as
309: an asymptotic state (though we focus entirely on steady states).
310:
311: In order to relate the order parameter~$U$ to the fluid flow, we interpret~$U$
312: as a stream function, and so $U_x$ represents the radial velocity in the
313: column. The effect of the non-slip boundary conditions on the top and bottom
314: plates is to induce an inwards flow
315: near the boundaries, although the radial velocity is zero on the boundaries
316: themselves. We model this
317: strong forcing at the end walls with the inhomogeneous boundary conditions:
318: \begin{equation}
319: U(-L/2)=U(L/2)=0,\qquad
320: U_x(-L/2)=U_x(L/2)=-1.
321: \label{eq:BCreal}
322: \end{equation}
323: The only symmetry in the problem is the reflection in the equatorial mid-plane
324: of the apparatus:
325: \begin{equation}
326: (U,x)\rightarrow(-U,-x).
327: \label{eq:symmetry}
328: \end{equation}
329:
330:
331: We also consider idealised reflecting boundary conditions:
332: \begin{equation}
333: U(-L/2)=U(L/2)=0,\qquad
334: U_{xx}(-L/2)=U_{xx}(L/2)=0,
335: \label{eq:BCreflecting}
336: \end{equation}
337: which have an additional hidden symmetry~\cite{refG98}: the problem can be
338: extended by reflection onto the domain $x\in[-L,L]$ with periodic boundary
339: conditions and so acquires a continuous translation symmetry. It is this
340: translation symmetry that is broken in the pitchfork bifurcation in the
341: idealised version of this problem (figure~\ref{fig:brokenpitchfork}a). This
342: symmetry is strongly broken by the inhomogeneous boundary
343: conditions~(\ref{eq:BCreal}).
344:
345: Experiments find states where the mid-plane reflection symmetry is preserved,
346: so we focus on this case by using reflecting boundary conditions at $x=0$. We
347: are also interested in examining the transition from idealised to realistic
348: boundary conditions, and so we use:
349: \begin{equation}
350: U(0)=U_{xx}(0)=U(L/2)=
351: \tau\left(U_x(L/2)+1\right) + (1-\tau)U_{xx}(L/2)=0.
352: \label{eq:BCboth}
353: \end{equation}
354: Here $\tau$ ($0\leq\tau\leq1$) is a homotopy parameter~\cite{refS111}, such
355: that $\tau=0$ corresponds to the ideal problem (with hidden translation
356: symmetry) and $\tau=1$ to the realistic non-slip boundary conditions.
357:
358: \section{Linear analysis: pitchfork and Hamiltonian Hopf}
359:
360: Solutions of the linearised problem can be written in terms of exponentials in
361: time and space, in the form:
362: \begin{equation}
363: U(x,t)=e^{st + (\sigma+ik)x},
364: \end{equation}
365: where $s$ is the temporal growth rate, $k$ is a spatial wave\-number, and
366: $\sigma$ is a spatial growth rate. In order to satisfy the linearised model
367: equation, $s$, $\sigma$ and $k$ must satisfy
368: \begin{equation}
369: 0=k\sigma(k^2-\sigma^2-1),\qquad
370: s=\mu-(k^2-\sigma^2-1)^2+4\sigma^2k^2,
371: \end{equation}
372: which can be rearranged to give three possibilities:
373: \begin{eqnarray}
374: A:& \quad k=0, \qquad &s=\mu-(1+\sigma^2)^2,\\
375: B:& \quad \sigma=0, \qquad &s=\mu-(1-k^2)^2,\\
376: C:& \quad k^2=\sigma^2+1, \qquad &s=\mu+4\sigma^2(1+\sigma^2).
377: \end{eqnarray}
378: Since the linearisation of the PDE~(\ref{eq:SH}) is first order in time and
379: fourth order in space, there is a unique temporal growth rate~$s$ and a total
380: of four complex spatial growth rates, which are roots of the equations above,
381: namely in case~A: $\pm\sigma_1$ and $\pm\sigma_2$; in case~B: $\pm ik_1$ and
382: $\pm ik_2$; in case~C: $\pm\sigma\pm ik$. Corresponding to these, the linear
383: solutions are of the form:
384: \begin{eqnarray}
385: A:& \quad U(x,t)=e^{st}\left(A\sinh(\sigma_1x)+B\sinh(\sigma_2x)\right),\\
386: B:& \quad U(x,t)=e^{st}\left(A\sin(k_1x)+B\sin(k_2x)\right),\\
387: C:& \quad U(x,t)=e^{st}\left(A\cos(kx)\sinh(\sigma x) +
388: B\sin(kx)\cosh(\sigma x)\right),
389: \end{eqnarray}
390: where $A$ and $B$ are constants that will be determined by the boundary
391: conditions at $x=L/2$. The odd boundary conditions at $x=0$ have already
392: been enforced by the choice of trigonometric functions.
393:
394: The next stage of the calculation depends on whether ideal ($\tau=0$) or
395: realistic ($\tau=1$) boundary conditions are being used. With ideal boundary
396: conditions, $U=0$ is always a solution of~(\ref{eq:SH}), and bifurcations from
397: this state occur when there are marginally stable ($s=0$) linear solutions.
398: Setting $s=0$ and $U(L/2)=U_{xx}(L/2)=0$ results in no solution in cases~A
399: and~C, and an eigenvalue problem in case~B, where $k$ can take on discrete
400: values: $k=2\pi n/L$, where $n$ is the number of vortices in the half-domain.
401: There are thus pitchfork bifurcations at
402: \begin{equation}
403: \mu_{\smallpf}=\left(1-\left(\frac{2\pi n}{L}\right)^2\right)^2,
404: \label{eq:mupitchfork}
405: \end{equation}
406: where $n$ is an integer. In particular, there is a pitchfork bifurcation at
407: $\mu=0$ whenever the domain is chosen to fit an exact number of vortices:
408: $n=L/2\pi$. Note that the condition for the onset of the vortices is $s=0$ and
409: $\sigma=0$, with a purely imaginary spatial wave\-number $\pm ik$ corresponding
410: to a spatially periodic pattern. On either side of this bifurcation point, the
411: temporal growth rate~$s$, indicating the stability of the Couette flow, changes
412: from negative to positive at the pitchfork bifurcation.
413:
414: \begin{figure}
415: \begin{center}
416: \mbox{\psfig{file=lin_nonlin_bifn.eps,width=0.9\hsize}}
417: \end{center} \caption{Dependence of the radial velocity at the midpoint
418: ($U_x(0)$) as a function of bifurcation parameter~$\mu$ for the nonlinear
419: (solid) and linearised (dashed) PDE~(\ref{eq:SH}), with $L=22\pi$ and realistic
420: boundary conditions ($\tau=1$). Also shown in (b) as a dotted line is an
421: integrated average of the solution and its first three derivatives, showing how
422: other measures of amplitude are not so sharp as the radial velocity at the
423: midpoint.}
424: \label{fig:lin_nonlin_bifn}
425: \end{figure}
426:
427: On the other hand, with inhomogeneous boundary conditions ($\tau=1$), there are
428: nonzero steady ($s=0$) linear solutions for all values of~$\mu$. Case~A can
429: only arise when $\mu\geq1$ (and in fact only a single solution is possible); in
430: case~B, there are two possible solutions for $0\geq\mu\geq1$ and one for
431: $\mu>1$, and case~C is possible only when~$\mu\leq0$. The solutions are thus,
432: for $\mu<0$: $U(x)=A\cos(kx)\sinh(\sigma x) + B\sin(kx)\cosh(\sigma x)$, where
433: $\sigma>0$ and $k>0$ are determined from $\mu$ by $4\sigma^4+4\sigma^2+\mu=0$
434: and $k^2=1+\sigma^2$; and for $0<\mu<1$: $U(x)=A\sin(k_1x)+B\sin(k_2x)$, where
435: $k_1>0$, $k_2>0$ and $k_1^2=1+\sqrt{\mu}$, $k_2^2=1-\sqrt{\mu}$. In these two
436: expressions for $U(x)$, the constants $A$ and $B$ are determined by a pair of
437: linear equations from the boundary conditions $U(L/2)=0$ and $U_x(L/2)=-1$. The
438: dependence of the solution, as measured by the radial velocity at the midpoint
439: ($U_x(0)=A\sigma + Bk$ for $\mu<0$, and $U_x(0)=Ak_1+Bk_2$ for $0<\mu<1$) is
440: shown in figure~\ref{fig:lin_nonlin_bifn} (dashed line).
441:
442: The amplitude of the linear solution goes to infinity for $\mu$~about $0.00826$
443: (with~$L=22\pi$). Since there is no sharp onset with realistic boundaries, one
444: cannot define a precise value of~$\mu$ at which pattern will be first observed
445: in a domain of finite length, but the value of $\mu=\mu_{\infty}$ for which the
446: linear solution goes to infinity is a suitable proxy. This is defined
447: implicitly by the condition
448: \begin{equation}
449: (k_1+k_2)\sin\left((k_1-k_2)\frac{L}{2}\right)=
450: (k_1-k_2)\sin\left((k_1+k_2)\frac{L}{2}\right),
451: \label{eq:muinfinity}
452: \end{equation}
453: where $k_{1,2}$ are defined in terms of~$\mu$ above.
454: When $L$ is large, the smallest positive solution $\mu_{\infty}$ occurs
455: when $(k_1-k_2)L/2 = \pi$. This yields
456: \begin{equation}
457: \mu_{\infty} = \frac{4\pi^2}{L^2} + O(L^{-3}).
458: \end{equation}
459: In other words, the divergence of the linear solution occurs for $\mu$~closer
460: to zero as $L$ increases. Moreover, we have that for large~$L$ the first
461: pitchfork bifurcation occurs at
462: \begin{equation}
463: \mu_{\smallpf} = \dfrac{4\delta^2}{L^2} + O(L^{-3}), \quad
464: \mbox{where} \quad \delta = L- 2 \pi \left [ L/(2\pi) \right],
465: \end{equation}
466: so $|\delta|<2\pi$. Hence $\mu_{\infty}$, $\mu_{\smallpf}$ and the difference
467: between them are all of the same order:~$L^{-2}$.
468:
469: Also shown in figure~\ref{fig:lin_nonlin_bifn} are nonlinear solutions
470: of~(\ref{eq:SH}). The linear and nonlinear solutions are close to each other
471: for $\mu$~negative: the discrepancy at low amplitude arises because the
472: linearised solution is not small close to the boundaries.
473:
474: \begin{figure}
475: \begin{center}
476: \mbox{\psfig{file=lin_solns.eps,width=0.9\hsize}}
477: \end{center}
478: \caption{Linear solutions of~(\ref{eq:SH}), with $L=40$ and~$\tau=1$, with
479: (a)~$\mu=-0.1$ and (b)~$\mu=0$, showing the growth of the solution in from the
480: edges ($x=\pm L/2$). Compare with figure~\ref{fig:abshagen}.}
481: \label{fig:lin_soln}
482: \end{figure}
483:
484: Linear solutions of the PDE~(\ref{eq:SH}) are shown in
485: figure~\ref{fig:lin_soln} for two values of~$\mu$. As $\mu$ is increased from
486: negative values to $\mu=0$, the exponentially decaying linear solution extends
487: further into the bulk of the fluid. In an arbitrarily long cylinder, the radial
488: velocity at the midpoint of the apparatus would remain almost zero until the
489: spatial decay rate (as measured by~$\sigma$) became zero. Therefore, the
490: condition for onset of steady vortices, as measured at the midpoint of the
491: apparatus, is $\sigma=0$ and $s=0$ -- which is the same condition as for the
492: onset of vortices with the idealised boundary conditions. In this case,
493: however, on either side of onset, it is the spatial eigenvalues $\pm\sigma\pm
494: ik$ and $\pm ik_1$, $\pm ik_2$ that change in nature, at a Hamiltonian Hopf
495: bifurcation \cite{vdeMe:85} (see figure~\ref{fig:HamiltonianHopf}).
496:
497: More accurately, we should describe the spatial bifurcation as a reversible 1:1
498: resonance, since the ODE obtained by setting $U_t=0$ in (\ref{eq:SH}) does not
499: conserve a first integral and so cannot correspond to a Hamiltonian system, but
500: is nevertheless reversible in the sense analysed by Iooss and Peroueme
501: \cite{IoPe:93} owing to the symmetry (\ref{eq:symmetry}). Here the spatial
502: bifurcation is supercritical and for $\mu>0$ there exist spatially periodic
503: solutions the maximum amplitude of which grows as the square root of~$\mu$ for
504: the nonlinear problem. The implications of this bifurcation are discussed in
505: more detail in Section~5.
506:
507: \begin{figure}
508: \begin{center}
509: \mbox{\psfig{file=hamiltonian_hopf.eps,width=0.9\hsize}}
510: \end{center}
511: \caption{Hamiltonian Hopf bifurcation:
512: (a)~$\mu<0$, the spatial eigenvalues are $\pm\sigma\pm ik$;
513: (b)~$\mu=0$, the eigenvalues $\pm i$;
514: (c)~$\mu>0$, the eigenvalues are $\pm ik_1$, $\pm ik_2$.}
515: \label{fig:HamiltonianHopf}
516: \end{figure}
517:
518: This explanation carries over from the Swift--Hohenberg model to the
519: Taylor--Couette problem and more general pattern forming situations
520: with strong forcing at the boundaries. If one assumes periodic
521: boundary conditions, with pure imaginary spatial wave\-numbers, the
522: criterion for onset in a general pattern forming problem is that the
523: temporal growth rate is zero. If, one the other hand, one takes the
524: end walls into account but the domain is very large, and the pattern
525: is measured only far away from the boundaries, then the steady
526: inhomogeneous solution will penetrate into the bulk of the fluid and
527: reach the centre when the real part of the spatial wave\-number is
528: zero. Thus the two perspectives will yield the same condition for the
529: onset of pattern formation: $\mu=0$ in the case of the model PDE, or
530: $R=R_c$ in the case of Taylor--Couette flow. This explains the sharp
531: transition seen in large domain Taylor--Couette experiments at the
532: Reynold's number predicted using idealised boundary conditions, even
533: though the boundaries are forcing the flow strongly.
534:
535: \section{Nonlinear steady-state bifurcation analysis}
536:
537: The remaining issues to be addressed are the effect of the length
538: of the domain on the nonlinear solutions, and the location of the
539: saddle-node bifurcation on the anomalous branch
540: (figure~\ref{fig:brokenpitchfork}c).
541:
542:
543: \begin{figure}
544: \begin{center}
545: \mbox{\psfig{file=nonlin_bifn_length.eps,width=0.9\hsize}}
546: \end{center}
547: \caption{As the length of the domain increases, the pattern, as
548: measured in the centre, sets in more sharply: (a)~detail near $\mu=0$,
549: for $L=22\pi$, $44\pi$ and $88\pi$; (b)~the larger picture, showing
550: the anomalous modes.}
551: \label{fig:bifn_length}
552: \end{figure}
553:
554: \begin{figure}
555: \begin{center}
556: \mbox{\psfig{file=nonlin_length.eps,width=0.9\hsize}}
557: \end{center}
558: \caption{As $\mu$ increases through zero, the wall mode penetrates
559: further into the bulk, but the amplitude at the centre of the domain remains
560: small for larger values of~$L$.
561: (a,b)~$L=50$, $\mu=-0.1$,~$0$;
562: (c,d)~$L=200$, $\mu=-0.1$,~$0$;
563: (e,f)~$L=1000$, $\mu=-0.1$,~$0$.
564: These are nonlinear solutions, but linear solutions look very similar.}
565: \label{fig:nonlin_length}
566: \end{figure}
567:
568: We focus on the steady state problem given by setting $U_t=0$
569: in~(\ref{eq:SH}).
570: \begin{equation}
571: U_{xxxx} +2 U_{xx} + (1-\mu) U + U^3 + U\,U_x =0,
572: \label{steady}
573: \end{equation}
574: subject to boundary conditions~(\ref{eq:BCboth}). Nonlinear solutions are
575: computed using AUTO \cite{Auto} as a boundary value solver. The effect of the
576: size of the domain is illustrated in figures~\ref{fig:bifn_length}
577: and~\ref{fig:nonlin_length}. As the length~$L$ increases, the amplitude, as
578: measured in the centre, sets in more sharply as $\mu$~is increased through
579: zero, and the curve resembles half a pitchfork as $L\rightarrow\infty$. With
580: very large values of~$L$ (figure~\ref{fig:nonlin_length}), the exponential
581: decay into the bulk ensures that the inhomogeneous pattern has very small
582: amplitude for~$\mu\leq0$. This is consistent with our understanding from the
583: linear theory. The anomalous mode branches for $L=22\pi$, $44\pi$ and $88\pi$
584: are also shown in figure~\ref{fig:bifn_length}. Note how the anomalous modes do
585: not approach $\mu=0$ for larger values of~$L$, even though the
586: symmetry-breaking effects are being pushed further away.
587:
588: With a fixed value of~$L$, only one sign of $U_x(0)$ is possible near the
589: transition at $\mu=0$, but which one is observed depends whether the number of
590: vortices between $x=0$ and $x=L/2$ is even or odd.
591:
592: \begin{figure}
593: \begin{center}
594: \mbox{\psfig{file=bifurcations.eps,width=0.9\hsize}}
595: \end{center}
596: %\input{baseline.pstex_t}
597: \caption{(a,b)~The idealised bifurcation diagram with $\tau=0$ and
598: $L=22\pi$: $U_x(0)$ as a function of bifurcation parameter~$\mu$. Diamonds
599: represent pitchfork bifurcation points. Only the first five bifurcating
600: branches from the trivial solution are depicted. In increasing order of~$\mu$,
601: these have 11, 10, 12, 9 and 13~pairs of
602: vortices in the full domain. (c)~Bifurcation diagram with $\tau=1$, showing the
603: smooth onset of the 11~vortex solution, and several disconnected branches.
604: The plus signs indicate solutions that are depicted in
605: figure~\ref{fig:lp_examples}(a--f), at saddle-node
606: bifurcation points.
607: The thick lines represent branches that are known to be stable.}
608: \label{fig:baseline}
609: \end{figure}
610:
611: \begin{figure}
612: \begin{center}
613: \mbox{\psfig{file=ideal_real_examples.eps,width=0.9\hsize}}
614: \end{center}
615: \caption{Examples of $n=11$~vortex solutions with (a) ideal ($\tau=0$) and
616: (b)~realistic ($\tau=1$) boundary conditions, with $\mu=2$ (largest amplitude),
617: $\mu=1$, $\mu=0.25$, $\mu=0.1$, $\mu=0$, $\mu=-0.1$.}
618: \label{fig:ideal_real_examples}
619: \end{figure}
620:
621: Next, we consider the connection between pitchfork bifurcations (in the case of
622: ideal boundary conditions) and saddle-node bifurcations (in the case of
623: realistic boundary conditions). With ideal boundary conditions, pitchforks
624: occur both from the trivial solution and as secondary bifurcations from the
625: various primary branches. We concentrate on the case $L=22\pi$.
626: Figure~\ref{fig:baseline} shows the bifurcation diagram computed with ideal
627: (a,b: $\tau=0$) and realistic (c: $\tau=1$) boundary conditions. Bifurcation
628: points from the trivial solution (and secondary bifurcations from the primary
629: branches) are marked in the figure. The primary bifurcation points occur at
630: $\mu=\mu_{\smallpf}$ (\ref{eq:mupitchfork}), with the bifurcating branch being
631: locally proportional to $\sqrt{\mu-\mu_{\smallpf}}\sin(n\pi/L)$. With
632: $L=22\pi$, the solution with $n=11$ bifurcates precisely from $\mu=0$ and
633: corresponds to a pattern with 11~vortices in the half domain. The next four
634: bifurcating branches for positive $\mu$ are also shown in the figure. These
635: bifurcate at $\mu=0.0301$, $0.0361$, $0.1093$ and $0.1574$ and correspond to
636: $n=10$, 12, 9, and 13 vortices respectively. Note that these solutions are
637: invariant under reflections in the midpoint ($x=0$), and are spatially periodic
638: with period~$2L/n$.
639:
640: Figure~\ref{fig:baseline}(c) shows a few of the many branches that exist for
641: realistic boundary conditions ($\tau=1$), in the same parameter range as
642: figure~\ref{fig:baseline}(b). These were obtained by taking all the points on
643: the branches in figure~\ref{fig:baseline}(a) with $\mu=2$ and continuing these
644: to $\tau=1$, and then continuing in~$\mu$ once more. Note that all of the
645: pitchfork bifurcations have been destroyed, and have been replaced by a series
646: of saddle-node bifurcations. The fundamental pitchfork bifurcation at $\mu=0$
647: has been replaced by a smooth transition, though the remnant of the pitchfork
648: shape can clearly be seen, and there is a sharp rise in amplitude close to
649: $\mu=0$ as predicted by the linear theory of the preceding section. Nonlinear
650: solution profiles on this fundamental branch
651: (figure~\ref{fig:ideal_real_examples}b) are qualitatively similar to those
652: with ideal boundary condition (figure~\ref{fig:ideal_real_examples}a), and so
653: might be said to correspond to $n=11$ vortices.
654:
655: \begin{figure}
656: \begin{center}
657: \mbox{\psfig{file=lp_examples.eps,width=0.9\hsize}}
658: \end{center}
659: \caption{Examples of anomalous solutions at the labelled saddle-node
660: bifurcation points from figure~\ref{fig:baseline}(c). (g,h) are outside the
661: range shown in figure~\ref{fig:baseline}. Solid lines depict the solution at
662: the saddle node bifurcation with $\tau=1$, and dashed lines show the solution
663: that has been continued to $\tau\rightarrow0$, ending up at one of the
664: bifurcation points in figure~\ref{fig:baseline}(a).
665: (a)~$\mu=0.1367\rightarrow0.0901$;
666: (b)~$\mu=0.1498\rightarrow0.0770$;
667: (c)~$\mu=0.5193\rightarrow0.4452$;
668: (d)~$\mu=0.5581\rightarrow0.3787$;
669: (e)~$\mu=1.6794\rightarrow0.0425$;
670: (f)~$\mu=1.8227\rightarrow0.4452$;
671: (g)~$\mu=2.1318\rightarrow0.0770$;
672: (h)~$\mu=4.6220\rightarrow0.0000$.}
673: \label{fig:lp_examples}
674: \end{figure}
675:
676: \begin{figure}
677: \begin{center}
678: \mbox{\psfig{file=lp_to_zero.eps,width=0.9\hsize}}
679: \end{center}
680: \caption{Continuation of the saddle-node bifurcation:
681: (a)~$(\mu,\tau)=(4.622,1.000)$,
682: (b)~$(\mu,\tau)=(2.000,0.597)$,
683: (c)~$(\mu,\tau)=(0.100,0.076)$,
684: (d)~$(\mu,\tau)\approx(0.000,0.000)$.}
685: \label{fig:lp_to_zero}
686: \end{figure}
687:
688: This now brings us to the question of the supposed disconnected part of the
689: unfolded pitchfork. Examples of solutions at the labelled saddle-node
690: bifurcation points are shown in figure~\ref{fig:lp_examples}(a--h), with the
691: solution $U_x(x)$ at the saddle-node bifurcation points drawn as solid lines.
692: These saddle-node bifurcations were then continued in $(\mu,\tau)$ back to
693: $\tau=0$ to discover where they originate. Most connect to secondary
694: bifurcations in figure~\ref{fig:baseline}(a) and therefore represent unstable
695: solutions (and are shown as dashed lines in figure~\ref{fig:lp_examples}).
696: However, the saddle-node bifurcation on the stable anomalous 11~vortex branch,
697: shown in figure~\ref{fig:lp_examples}(e) is found to create a stable branch.
698: (The stability of a typical solution on this branch was checked by solving, in
699: addition to the ODEs (\ref{steady}), the linear variational equations
700: governing by a temporal eigenmode with eigenvalue~$s$. AUTO was then used to
701: continue solutions in~$s$ to establish that there are no nontrivial solutions
702: for~$s>0$.) Unexpected behaviour was found for this branch upon varying the
703: homotopy parameter~$\tau$. One might imagine that under homotopy to $\tau=0$
704: this saddle-node bifurcation point should approach the fundamental pitchfork
705: bifurcation at $\mu=0$. This is not the case for these parameter values: when
706: $\tau$ is reduced from 1 to 0, the saddle-node bifurcation itself undergoes a
707: pair of fold bifurcations, and ends up (at $\tau=0$) in the unfolding of the
708: $n=12$ bifurcation point at $\mu=0.0425$ (the dashed profile in
709: figure~\ref{fig:lp_examples}(e) corresponds to a 12~vortex pattern).
710:
711: Alternatively, one could try following the saddle-node bifurcation point that
712: occurs as one unfolds the pitchfork at $\mu=0$ under infinitesimal increase of
713: $\tau$ from zero. When this is done, the saddle-node bifurcation can be
714: continued up to $(\mu,\tau)=(4.6220,1)$ -- see figure~\ref{fig:lp_examples}(h)
715: and figure~\ref{fig:lp_to_zero}.
716:
717: The details of which saddle-node bifurcation (with $\tau=1$) connects to which
718: pitchfork bifurcation (with $\tau=0$) was found to depend sensitively on the
719: value of~$L$. For example, with $L=88\pi$, the saddle-node bifurcation on the
720: 44~vortex anomalous branch does continue down to the primary pitchfork to
721: 44~vortices at $\mu=0$, though in this case the anomalous branch is not stable.
722: The details of how the branches connect also depends on the particular choice
723: how the ideal and realistic boundary conditions are combined via homotopy. For
724: example replacing the final `$+$' in (\ref{eq:BCboth}) with a `$-$' lead to
725: significantly different results.
726:
727: \section{Conclusion}
728:
729: The apparent contradiction described in the introduction is resolved, at least
730: in the context of the Swift--Hohenberg model considered here. The onset of
731: Taylor vortices is not a weakly broken pitchfork bifurcation, owing to strong
732: inhomogeneous boundary forcing. Anomalous modes stay bounded away from the
733: critical value of the bifurcation parameter as they must overcome the strong
734: preference set by the boundary. When the amplitude of the pattern is measured
735: far away from the boundaries, the pattern appears to set in sharply, in half a
736: pitchfork bifurcation, as the decaying wall mode penetrates the bulk of the
737: domain. The parameter value at which pattern, as measured in the centre of a
738: large domain, become significantly different from zero is the same as the value
739: predicted assuming idealised boundary conditions, because the requirements for
740: both situations are the same: steady ($s=0$) and zero spatial growth rate
741: ($\sigma=0$). Moreover we have shown that for a long but finite domain of
742: length~$L$, that the parameter value corresponding to this large central growth
743: in pattern occurs according to linear theory at a value that is within
744: $O(L^{-2})$ of the idealised pitchfork.
745:
746: The ideas were developed for the model equation, but they apply equally well to
747: the Taylor--Couette case, and resolve the difficulties raised by Benjamin and
748: Mullin~\cite{refB128}.
749:
750: We have also observed that the saddle-node bifurcation on the anomalous branch
751: does not necessarily connect to the primary bifurcation as boundary conditions
752: vary from real to ideal. The specific results described here apply only to the
753: Swift--Hohenberg model, though the general conclusion that we can make is that,
754: under small perturbations from the ideal boundary conditions, we expect the
755: pitchfork to be perturbed in the generic way, as in
756: figure~\ref{fig:brokenpitchfork}(b). However, going all the way to $\tau=1$ is
757: not a small perturbation, and in general nothing can be said about whether the
758: the saddle-node bifurcation created in the unfolding of the primary pitchfork
759: is the same (or not the same) as the saddle-node at the end of the stable
760: anomalous branch, or even if there is an anomalous branch that is stable.
761:
762: This approach yields results that are applicable to other pattern formation
763: problems (for instance, Rayleigh--B\'enard convection). Earlier work on
764: convection~\cite{refD56,DaHoSk:03} has focused on weak forcing at the side
765: walls, primarily using Swift--Hohenberg theory. We have shown here how the
766: ideas can be extended to strong forcing.
767:
768: An interesting aspect of our work has been to link the mode selection problem
769: to the existence of a spatial Hamiltonian Hopf bifurcation for the infinite
770: length problem. This gives the possibility of the existence of branches of
771: spatially periodic solutions beyond the critical parameter value for the onset
772: of rolls, whose period for small amplitude is that given by the wave number of
773: the neutral mode of the temporal problem. The normal form of the Hamiltonian
774: Hopf bifurcation (even for reversible systems) is completely integrable up to
775: any order (see e.g., \cite{IoPe:93}) and instead of a unique spatially periodic
776: solution, there is a one parameter band of spatially periodic solutions, whose
777: envelope grows as the square root of the bifurcation parameter~$\mu$. There
778: also exists a two-parameter family of spatially quasi-periodic solutions whose
779: existence is bounded by the periodic solutions and homoclinic connections to
780: them. However, in the reversible case, not all these solutions will necessarily
781: exist in a full unfolding of the normal form that breaks its Hamiltonian
782: structure. We note from our numerical results for the nonlinear problem with
783: realistic boundary conditions and fixed~$L$, that the main branch seems to
784: develop a connection from the boundary to a pure periodic state in the middle
785: of the domain. It is not clear {\em a priori} why this solution and not others
786: are selected from the unfolding of the Hamiltonian Hopf bifurcation.
787:
788: Finally, we mention another outstanding issue in the problem of the onset of
789: Taylor vortices~\cite{refA70}: the timescale for the onset and decay of the
790: pattern are different. In particular, these authors computed steady solutions
791: at Reynold's numbers just above and just below critical, and in each case
792: altered the Reynold's number to an intermediate value and examined the
793: transient. In the case of onset, the pattern invaded the bulk as a front
794: travelling in from the boundary, while in the case of decay, there was uniform
795: decay throughout bulk. The timescales for these two processes were different,
796: and it may be possible to explain this using a Swift--Hohenberg based model, as
797: considered here.
798:
799: \begin{ack}
800: AMR is grateful for support from the EPSRC while this work was carried out. We
801: are very grateful to Tom Mullin for many coments and advice, and we thank Edgar
802: Knobloch for useful discussions.
803: \end{ack}
804:
805: % \bibliography{rc}
806:
807: \begin{thebibliography}{10}
808: \expandafter\ifx\csname url\endcsname\relax
809: \def\url#1{\texttt{#1}}\fi
810: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi
811:
812: \bibitem{refD55}
813: R.C. Di~Prima, H.L. Swinney, Instabilities and transition in flow between
814: concentric rotating cylinders, in: H.L. Swinney, J.P. Gollub (Eds.),
815: Hydrodynamic Instabilities and the Transition to Turbulence, Springer,
816: Berlin, 1981, pp. 139--180.
817:
818: \bibitem{Ko:93}
819: E.L. Koschmieder, B\'{e}rnard Cells and Taylor Vortices, CUP, Cambridge, 1993.
820:
821: \bibitem{refT61}
822: G.I. Taylor, Stability of a viscous liquid contained between two rotating
823: cylinders, Phil. Trans. R.~Soc. Lond.~A 223 (1923) 289--343.
824:
825: \bibitem{refG97}
826: J.P. Gollub, M.H. Freilich, Optical heterodyne test of perturbation expansions
827: for the Taylor instability, Phys. Fluids 19 (1976) 618--626.
828:
829: \bibitem{refB127a}
830: T.B. Benjamin, Bifurcation phenomena in steady flows of a viscous liquid.
831: I.~Theory, Proc. R.~Soc. Lond.~{\rm A} 359 (1978) 1--26.
832:
833: \bibitem{refB128}
834: T.B. Benjamin, T.~Mullin, Anomalous modes in the Taylor experiment, Proc.
835: R.~Soc. Lond.~{\rm A} 377 (1981) 221--249.
836:
837: \bibitem{refC124}
838: D.~Coles, Transition in circular Couette flow, J.~Fluid Mech. 21 (1965)
839: 385--425.
840:
841: \bibitem{refA71}
842: T.~Alizary~de Roquefort, G.~Grillaud, Computation of Taylor vortex flow by a
843: transient implicit method, Computers and Fluids 6 (1978) 259--269.
844:
845: \bibitem{refA70}
846: J.~Abshagen, O.~Meincke, G.~Pfister, K.A. Cliffe, T.~Mullin, Transient dynamics
847: at the onset of Taylor vortices, J.~Fluid Mech. 476 (2003) 335--343.
848:
849: \bibitem{refS111}
850: D.~Schaeffer, Qualitative analysis of a model for boundary effects in the
851: Taylor problem, Math. Proc. Camb. Phil. Soc. 87 (1980) 307--337.
852:
853: \bibitem{refB127b}
854: T.B. Benjamin, Bifurcation phenomena in steady flows of a viscous liquid.
855: II.~Experiment, Proc. R.~Soc. Lond.~{\rm A} 359 (1978) 27--43.
856:
857: \bibitem{refB129}
858: J.H. Bolstad, H.B. Keller, Computation of anomalous modes in the Taylor
859: experiment, J.~Comput. Phys. 69 (1987) 230--251.
860:
861: \bibitem{refC123}
862: K.A. Cliffe, T.~Mullin, A numerical and experimental study of anomalous modes
863: in the Taylor experiment, J.~Fluid Mech. 153 (1985) 243--258.
864:
865: \bibitem{refC122}
866: K.A. Cliffe, J.J. Kobine, T.~Mullin, The role of anomalous modes in
867: Taylor--Couette flow, Proc. R.~Soc. Lond.~{\rm A} 439 (1992) 341--357.
868:
869: \bibitem{refL66}
870: A.~Lorenzen, T.~Mullin, Anomalous modes and finite-length effects in
871: Taylor--Couette flow, Phys. Rev.~A 31 (1985) 3463--3465.
872:
873: \bibitem{refD56}
874: P.G. Daniels, The effect of distant sidewalls on the transition to finite
875: amplitude B{\accent 19 e}nard convection, Proc. R.~Soc. Lond.~{\rm A} 358
876: (1977) 173--197.
877:
878: \bibitem{DaHoSk:03}
879: P.G. Daniels, D.~Ho, A.C. Skeldon, Solutions for nonlinear convection in the
880: presence of a lateral boundary, Physica D 178 (2003) 83--102.
881:
882: \bibitem{refP58}
883: G.~Pfister, I.~Rehberg, Space-dependent order parameter in circular Couette
884: flow transitions, Phys. Lett.~A 83 (1981) 19--22.
885:
886: \bibitem{refS109}
887: J.~Swift, P.C. Hohenberg, Hydrodynamic fluctuations at the convective
888: instability, Phys. Rev.~A 15 (1977) 319--328.
889:
890: \bibitem{refM99}
891: I.~Melbourne, Steady-state bifurcation with Euclidean symmetry, Trans. Am.
892: Math. Soc. 351 (1999) 1575--1603.
893:
894: \bibitem{refG98}
895: M.~Golubitsky, J.~Marsden, D.~Schaeffer, Bifurcation problems with hidden
896: symmetries, in: W.E. Fitzgibbon~III (Ed.), Partial Differential Equations and
897: Dynamical Systems, Pitman, Boston, 1984, pp. 181--210.
898:
899: \bibitem{vdeMe:85}
900: J~van~der Meer, The Hamiltonian Hopf bifurcation, Lecture Notes in Mathematics
901: 1160, Springer-Verlag, Berlin, 1985.
902:
903: \bibitem{IoPe:93}
904: G.~Iooss, M.C. Peroueme, Perturbed homoclinic solutions in reversible 1:1
905: resonance vector fields, J. Diff. Eq. 102 (1993) 62--88.
906:
907: \bibitem{Auto}
908: E.J Doedel, A.R Champneys, T.R. Fairgrieve, Yu.A. Kuznetsov, B.~Sandstede, X.J.
909: Wang, {AUTO97} Continuation and bifurcation software for ordinary
910: differential equations (1997).
911:
912: \end{thebibliography}
913:
914:
915: \end{document}
916:
917:
918:
919:
920:
921:
922: